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 "inline_invlist.c"
84 #include "unicode_constants.h"
87 /* At least one required character in the target string is expressible only in
89 static const char* const non_utf8_target_but_utf8_required
90 = "Can't match, because target string needs to be in UTF-8\n";
93 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
94 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
98 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
101 #define STATIC static
104 /* Valid only for non-utf8 strings: avoids the reginclass
105 * call if there are no complications: i.e., if everything matchable is
106 * straight forward in the bitmap */
107 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \
108 : ANYOF_BITMAP_TEST(p,*(c)))
114 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
115 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
117 #define HOPc(pos,off) \
118 (char *)(reginfo->is_utf8_target \
119 ? reghop3((U8*)pos, off, \
120 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
123 #define HOPBACKc(pos, off) \
124 (char*)(reginfo->is_utf8_target \
125 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
126 : (pos - off >= reginfo->strbeg) \
130 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
131 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
133 /* lim must be +ve. Returns NULL on overshoot */
134 #define HOPMAYBE3(pos,off,lim) \
135 (reginfo->is_utf8_target \
136 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
137 : ((U8*)pos + off <= lim) \
141 /* like HOP3, but limits the result to <= lim even for the non-utf8 case.
142 * off must be >=0; args should be vars rather than expressions */
143 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
144 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
145 : (U8*)((pos + off) > lim ? lim : (pos + off)))
147 #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
148 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
150 #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
152 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
153 #define NEXTCHR_IS_EOS (nextchr < 0)
155 #define SET_nextchr \
156 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
158 #define SET_locinput(p) \
163 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
165 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
166 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
167 1, 0, invlist, &flags); \
172 /* If in debug mode, we test that a known character properly matches */
174 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
177 utf8_char_in_property) \
178 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \
179 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
181 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
184 utf8_char_in_property) \
185 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
188 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
189 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
191 PL_XPosix_ptrs[_CC_WORDCHAR], \
192 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
194 #define LOAD_UTF8_CHARCLASS_GCB() /* Grapheme cluster boundaries */ \
196 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_regular_begin, \
197 "_X_regular_begin", \
199 LATIN_CAPITAL_LETTER_SHARP_S_UTF8); \
200 LOAD_UTF8_CHARCLASS_DEBUG_TEST(PL_utf8_X_extend, \
203 COMBINING_GRAVE_ACCENT_UTF8); \
206 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
207 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
209 /* for use after a quantifier and before an EXACT-like node -- japhy */
210 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
212 * NOTE that *nothing* that affects backtracking should be in here, specifically
213 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
214 * node that is in between two EXACT like nodes when ascertaining what the required
215 * "follow" character is. This should probably be moved to regex compile time
216 * although it may be done at run time beause of the REF possibility - more
217 * investigation required. -- demerphq
219 #define JUMPABLE(rn) ( \
221 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
223 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
224 OP(rn) == PLUS || OP(rn) == MINMOD || \
226 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
228 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
230 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
233 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
234 we don't need this definition. */
235 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
236 #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 )
237 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
240 /* ... so we use this as its faster. */
241 #define IS_TEXT(rn) ( OP(rn)==EXACT )
242 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
243 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
244 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
249 Search for mandatory following text node; for lookahead, the text must
250 follow but for lookbehind (rn->flags != 0) we skip to the next step.
252 #define FIND_NEXT_IMPT(rn) STMT_START { \
253 while (JUMPABLE(rn)) { \
254 const OPCODE type = OP(rn); \
255 if (type == SUSPEND || PL_regkind[type] == CURLY) \
256 rn = NEXTOPER(NEXTOPER(rn)); \
257 else if (type == PLUS) \
259 else if (type == IFMATCH) \
260 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
261 else rn += NEXT_OFF(rn); \
265 /* These constants are for finding GCB=LV and GCB=LVT in the CLUMP regnode.
266 * These are for the pre-composed Hangul syllables, which are all in a
267 * contiguous block and arranged there in such a way so as to facilitate
268 * alorithmic determination of their characteristics. As such, they don't need
269 * a swash, but can be determined by simple arithmetic. Almost all are
270 * GCB=LVT, but every 28th one is a GCB=LV */
271 #define SBASE 0xAC00 /* Start of block */
272 #define SCount 11172 /* Length of block */
275 #define SLAB_FIRST(s) (&(s)->states[0])
276 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
278 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
279 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
280 static regmatch_state * S_push_slab(pTHX);
282 #define REGCP_PAREN_ELEMS 3
283 #define REGCP_OTHER_ELEMS 3
284 #define REGCP_FRAME_ELEMS 1
285 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
286 * are needed for the regexp context stack bookkeeping. */
289 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
291 const int retval = PL_savestack_ix;
292 const int paren_elems_to_push =
293 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
294 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
295 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
297 GET_RE_DEBUG_FLAGS_DECL;
299 PERL_ARGS_ASSERT_REGCPPUSH;
301 if (paren_elems_to_push < 0)
302 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
303 (int)paren_elems_to_push, (int)maxopenparen,
304 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
306 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
307 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
308 " out of range (%lu-%ld)",
310 (unsigned long)maxopenparen,
313 SSGROW(total_elems + REGCP_FRAME_ELEMS);
316 if ((int)maxopenparen > (int)parenfloor)
317 PerlIO_printf(Perl_debug_log,
318 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
323 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
324 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
325 SSPUSHIV(rex->offs[p].end);
326 SSPUSHIV(rex->offs[p].start);
327 SSPUSHINT(rex->offs[p].start_tmp);
328 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
329 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
331 (IV)rex->offs[p].start,
332 (IV)rex->offs[p].start_tmp,
336 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
337 SSPUSHINT(maxopenparen);
338 SSPUSHINT(rex->lastparen);
339 SSPUSHINT(rex->lastcloseparen);
340 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
345 /* These are needed since we do not localize EVAL nodes: */
346 #define REGCP_SET(cp) \
348 PerlIO_printf(Perl_debug_log, \
349 " Setting an EVAL scope, savestack=%"IVdf"\n", \
350 (IV)PL_savestack_ix)); \
353 #define REGCP_UNWIND(cp) \
355 if (cp != PL_savestack_ix) \
356 PerlIO_printf(Perl_debug_log, \
357 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
358 (IV)(cp), (IV)PL_savestack_ix)); \
361 #define UNWIND_PAREN(lp, lcp) \
362 for (n = rex->lastparen; n > lp; n--) \
363 rex->offs[n].end = -1; \
364 rex->lastparen = n; \
365 rex->lastcloseparen = lcp;
369 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
373 GET_RE_DEBUG_FLAGS_DECL;
375 PERL_ARGS_ASSERT_REGCPPOP;
377 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
379 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
380 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
381 rex->lastcloseparen = SSPOPINT;
382 rex->lastparen = SSPOPINT;
383 *maxopenparen_p = SSPOPINT;
385 i -= REGCP_OTHER_ELEMS;
386 /* Now restore the parentheses context. */
388 if (i || rex->lastparen + 1 <= rex->nparens)
389 PerlIO_printf(Perl_debug_log,
390 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
395 paren = *maxopenparen_p;
396 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
398 rex->offs[paren].start_tmp = SSPOPINT;
399 rex->offs[paren].start = SSPOPIV;
401 if (paren <= rex->lastparen)
402 rex->offs[paren].end = tmps;
403 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
404 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
406 (IV)rex->offs[paren].start,
407 (IV)rex->offs[paren].start_tmp,
408 (IV)rex->offs[paren].end,
409 (paren > rex->lastparen ? "(skipped)" : ""));
414 /* It would seem that the similar code in regtry()
415 * already takes care of this, and in fact it is in
416 * a better location to since this code can #if 0-ed out
417 * but the code in regtry() is needed or otherwise tests
418 * requiring null fields (pat.t#187 and split.t#{13,14}
419 * (as of patchlevel 7877) will fail. Then again,
420 * this code seems to be necessary or otherwise
421 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
422 * --jhi updated by dapm */
423 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
424 if (i > *maxopenparen_p)
425 rex->offs[i].start = -1;
426 rex->offs[i].end = -1;
427 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
428 " \\%"UVuf": %s ..-1 undeffing\n",
430 (i > *maxopenparen_p) ? "-1" : " "
436 /* restore the parens and associated vars at savestack position ix,
437 * but without popping the stack */
440 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
442 I32 tmpix = PL_savestack_ix;
443 PL_savestack_ix = ix;
444 regcppop(rex, maxopenparen_p);
445 PL_savestack_ix = tmpix;
448 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
451 S_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_PSXSPC: return isPSXSPC_LC(character);
479 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
480 case _CC_ENUM_SPACE: return isSPACE_LC(character);
481 case _CC_ENUM_UPPER: return isUPPER_LC(character);
482 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
483 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
484 default: /* VERTSPACE should never occur in locales */
485 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
488 assert(0); /* NOTREACHED */
493 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
495 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
496 * 'character' is a member of the Posix character class given by 'classnum'
497 * that should be equivalent to a value in the typedef
498 * '_char_class_number'.
500 * This just calls isFOO_lc on the code point for the character if it is in
501 * the range 0-255. Outside that range, all characters avoid Unicode
502 * rules, ignoring any locale. So use the Unicode function if this class
503 * requires a swash, and use the Unicode macro otherwise. */
505 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
507 if (UTF8_IS_INVARIANT(*character)) {
508 return isFOO_lc(classnum, *character);
510 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
511 return isFOO_lc(classnum,
512 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
515 if (classnum < _FIRST_NON_SWASH_CC) {
517 /* Initialize the swash unless done already */
518 if (! PL_utf8_swash_ptrs[classnum]) {
519 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
520 PL_utf8_swash_ptrs[classnum] =
521 _core_swash_init("utf8",
524 PL_XPosix_ptrs[classnum], &flags);
527 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
529 TRUE /* is UTF */ ));
532 switch ((_char_class_number) classnum) {
534 case _CC_ENUM_PSXSPC: return is_XPERLSPACE_high(character);
536 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
537 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
538 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
539 default: return 0; /* Things like CNTRL are always
543 assert(0); /* NOTREACHED */
548 * pregexec and friends
551 #ifndef PERL_IN_XSUB_RE
553 - pregexec - match a regexp against a string
556 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
557 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
558 /* stringarg: the point in the string at which to begin matching */
559 /* strend: pointer to null at end of string */
560 /* strbeg: real beginning of string */
561 /* minend: end of match must be >= minend bytes after stringarg. */
562 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
563 * itself is accessed via the pointers above */
564 /* nosave: For optimizations. */
566 PERL_ARGS_ASSERT_PREGEXEC;
569 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
570 nosave ? 0 : REXEC_COPY_STR);
576 /* re_intuit_start():
578 * Based on some optimiser hints, try to find the earliest position in the
579 * string where the regex could match.
581 * rx: the regex to match against
582 * sv: the SV being matched: only used for utf8 flag; the string
583 * itself is accessed via the pointers below. Note that on
584 * something like an overloaded SV, SvPOK(sv) may be false
585 * and the string pointers may point to something unrelated to
587 * strbeg: real beginning of string
588 * strpos: the point in the string at which to begin matching
589 * strend: pointer to the byte following the last char of the string
590 * flags currently unused; set to 0
591 * data: currently unused; set to NULL
593 * The basic idea of re_intuit_start() is to use some known information
594 * about the pattern, namely:
596 * a) the longest known anchored substring (i.e. one that's at a
597 * constant offset from the beginning of the pattern; but not
598 * necessarily at a fixed offset from the beginning of the
600 * b) the longest floating substring (i.e. one that's not at a constant
601 * offset from the beginning of the pattern);
602 * c) Whether the pattern is anchored to the string; either
603 * an absolute anchor: /^../, or anchored to \n: /^.../m,
604 * or anchored to pos(): /\G/;
605 * d) A start class: a real or synthetic character class which
606 * represents which characters are legal at the start of the pattern;
608 * to either quickly reject the match, or to find the earliest position
609 * within the string at which the pattern might match, thus avoiding
610 * running the full NFA engine at those earlier locations, only to
611 * eventually fail and retry further along.
613 * Returns NULL if the pattern can't match, or returns the address within
614 * the string which is the earliest place the match could occur.
616 * The longest of the anchored and floating substrings is called 'check'
617 * and is checked first. The other is called 'other' and is checked
618 * second. The 'other' substring may not be present. For example,
620 * /(abc|xyz)ABC\d{0,3}DEFG/
624 * check substr (float) = "DEFG", offset 6..9 chars
625 * other substr (anchored) = "ABC", offset 3..3 chars
628 * Be aware that during the course of this function, sometimes 'anchored'
629 * refers to a substring being anchored relative to the start of the
630 * pattern, and sometimes to the pattern itself being anchored relative to
631 * the string. For example:
633 * /\dabc/: "abc" is anchored to the pattern;
634 * /^\dabc/: "abc" is anchored to the pattern and the string;
635 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
636 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
637 * but the pattern is anchored to the string.
641 Perl_re_intuit_start(pTHX_
644 const char * const strbeg,
648 re_scream_pos_data *data)
650 struct regexp *const prog = ReANY(rx);
651 SSize_t start_shift = prog->check_offset_min;
652 /* Should be nonnegative! */
653 SSize_t end_shift = 0;
654 /* current lowest pos in string where the regex can start matching */
655 char *rx_origin = strpos;
657 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
658 U8 other_ix = 1 - prog->substrs->check_ix;
660 char *other_last = strpos;/* latest pos 'other' substr already checked to */
661 char *check_at = NULL; /* check substr found at this pos */
662 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
663 RXi_GET_DECL(prog,progi);
664 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
665 regmatch_info *const reginfo = ®info_buf;
666 GET_RE_DEBUG_FLAGS_DECL;
668 PERL_ARGS_ASSERT_RE_INTUIT_START;
669 PERL_UNUSED_ARG(flags);
670 PERL_UNUSED_ARG(data);
672 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
673 "Intuit: trying to determine minimum start position...\n"));
675 /* for now, assume that all substr offsets are positive. If at some point
676 * in the future someone wants to do clever things with look-behind and
677 * -ve offsets, they'll need to fix up any code in this function
678 * which uses these offsets. See the thread beginning
679 * <20140113145929.GF27210@iabyn.com>
681 assert(prog->substrs->data[0].min_offset >= 0);
682 assert(prog->substrs->data[0].max_offset >= 0);
683 assert(prog->substrs->data[1].min_offset >= 0);
684 assert(prog->substrs->data[1].max_offset >= 0);
685 assert(prog->substrs->data[2].min_offset >= 0);
686 assert(prog->substrs->data[2].max_offset >= 0);
688 /* for now, assume that if both present, that the floating substring
689 * doesn't start before the anchored substring.
690 * If you break this assumption (e.g. doing better optimisations
691 * with lookahead/behind), then you'll need to audit the code in this
692 * function carefully first
695 ! ( (prog->anchored_utf8 || prog->anchored_substr)
696 && (prog->float_utf8 || prog->float_substr))
697 || (prog->float_min_offset >= prog->anchored_offset));
699 /* byte rather than char calculation for efficiency. It fails
700 * to quickly reject some cases that can't match, but will reject
701 * them later after doing full char arithmetic */
702 if (prog->minlen > strend - strpos) {
703 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
704 " String too short...\n"));
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->check_utf8 && prog->check_substr)
719 to_utf8_substr(prog);
720 check = prog->check_utf8;
722 if (!prog->check_substr && prog->check_utf8) {
723 if (! to_byte_substr(prog)) {
724 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
727 check = prog->check_substr;
730 /* dump the various substring data */
731 DEBUG_OPTIMISE_MORE_r({
733 for (i=0; i<=2; i++) {
734 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
735 : prog->substrs->data[i].substr);
739 PerlIO_printf(Perl_debug_log,
740 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
741 " useful=%"IVdf" utf8=%d [%s]\n",
743 (IV)prog->substrs->data[i].min_offset,
744 (IV)prog->substrs->data[i].max_offset,
745 (IV)prog->substrs->data[i].end_shift,
752 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
754 /* ml_anch: check after \n?
756 * A note about IMPLICIT: on an un-anchored pattern beginning
757 * with /.*.../, these flags will have been added by the
759 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
760 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
762 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
763 && !(prog->intflags & PREGf_IMPLICIT);
765 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
766 /* we are only allowed to match at BOS or \G */
768 /* trivially reject if there's a BOS anchor and we're not at BOS.
770 * Note that we don't try to do a similar quick reject for
771 * \G, since generally the caller will have calculated strpos
772 * based on pos() and gofs, so the string is already correctly
773 * anchored by definition; and handling the exceptions would
774 * be too fiddly (e.g. REXEC_IGNOREPOS).
776 if ( strpos != strbeg
777 && (prog->intflags & (PREGf_ANCH_BOL|PREGf_ANCH_SBOL)))
779 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
780 " Not at start...\n"));
784 /* in the presence of an anchor, the anchored (relative to the
785 * start of the regex) substr must also be anchored relative
786 * to strpos. So quickly reject if substr isn't found there.
787 * This works for \G too, because the caller will already have
788 * subtracted gofs from pos, and gofs is the offset from the
789 * \G to the start of the regex. For example, in /.abc\Gdef/,
790 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
791 * caller will have set strpos=pos()-4; we look for the substr
792 * at position pos()-4+1, which lines up with the "a" */
794 if (prog->check_offset_min == prog->check_offset_max
795 && !(prog->intflags & PREGf_CANY_SEEN))
797 /* Substring at constant offset from beg-of-str... */
798 SSize_t slen = SvCUR(check);
799 char *s = HOP3c(strpos, prog->check_offset_min, strend);
801 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
802 " Looking for check substr at fixed offset %"IVdf"...\n",
803 (IV)prog->check_offset_min));
806 /* In this case, the regex is anchored at the end too.
807 * Unless it's a multiline match, the lengths must match
808 * exactly, give or take a \n. NB: slen >= 1 since
809 * the last char of check is \n */
811 && ( strend - s > slen
812 || strend - s < slen - 1
813 || (strend - s == slen && strend[-1] != '\n')))
815 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
816 " String too long...\n"));
819 /* Now should match s[0..slen-2] */
822 if (slen && (*SvPVX_const(check) != *s
823 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
825 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
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 PerlIO_printf(Perl_debug_log,
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 - strpos),
881 (IV)prog->check_offset_min,
884 (IV)prog->check_end_shift);
887 if (prog->intflags & PREGf_CANY_SEEN) {
888 start_point= (U8*)(rx_origin + start_shift);
889 end_point= (U8*)(strend - end_shift);
890 if (start_point > end_point)
893 end_point = HOP3(strend, -end_shift, strbeg);
894 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
900 /* If the regex is absolutely anchored to either the start of the
901 * string (BOL,SBOL) or to pos() (ANCH_GPOS), then
902 * check_offset_max represents an upper bound on the string where
903 * the substr could start. For the ANCH_GPOS case, we assume that
904 * the caller of intuit will have already set strpos to
905 * pos()-gofs, so in this case strpos + offset_max will still be
906 * an upper bound on the substr.
909 && prog->intflags & PREGf_ANCH
910 && prog->check_offset_max != SSize_t_MAX)
912 SSize_t len = SvCUR(check) - !!SvTAIL(check);
913 const char * const anchor =
914 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
916 /* do a bytes rather than chars comparison. It's conservative;
917 * so it skips doing the HOP if the result can't possibly end
918 * up earlier than the old value of end_point.
920 if ((char*)end_point - anchor > prog->check_offset_max) {
921 end_point = HOP3lim((U8*)anchor,
922 prog->check_offset_max,
928 DEBUG_OPTIMISE_MORE_r({
929 PerlIO_printf(Perl_debug_log, " fbm_instr len=%d str=<%.*s>\n",
930 (int)(end_point - start_point),
931 (int)(end_point - start_point) > 20 ? 20 : (int)(end_point - start_point),
935 check_at = fbm_instr( start_point, end_point,
936 check, multiline ? FBMrf_MULTILINE : 0);
938 /* Update the count-of-usability, remove useless subpatterns,
942 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
943 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
944 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
945 (check_at ? "Found" : "Did not find"),
946 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
947 ? "anchored" : "floating"),
950 (check_at ? " at offset " : "...\n") );
955 /* Finish the diagnostic message */
956 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%ld...\n", (long)(check_at - strpos)) );
958 /* set rx_origin to the minimum position where the regex could start
959 * matching, given the constraint of the just-matched check substring.
960 * But don't set it lower than previously.
963 if (check_at - rx_origin > prog->check_offset_max)
964 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
968 /* now look for the 'other' substring if defined */
970 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
971 : prog->substrs->data[other_ix].substr)
973 /* Take into account the "other" substring. */
977 struct reg_substr_datum *other;
980 other = &prog->substrs->data[other_ix];
982 /* if "other" is anchored:
983 * we've previously found a floating substr starting at check_at.
984 * This means that the regex origin must lie somewhere
985 * between min (rx_origin): HOP3(check_at, -check_offset_max)
986 * and max: HOP3(check_at, -check_offset_min)
987 * (except that min will be >= strpos)
988 * So the fixed substr must lie somewhere between
989 * HOP3(min, anchored_offset)
990 * HOP3(max, anchored_offset) + SvCUR(substr)
993 /* if "other" is floating
994 * Calculate last1, the absolute latest point where the
995 * floating substr could start in the string, ignoring any
996 * constraints from the earlier fixed match. It is calculated
999 * strend - prog->minlen (in chars) is the absolute latest
1000 * position within the string where the origin of the regex
1001 * could appear. The latest start point for the floating
1002 * substr is float_min_offset(*) on from the start of the
1003 * regex. last1 simply combines thee two offsets.
1005 * (*) You might think the latest start point should be
1006 * float_max_offset from the regex origin, and technically
1007 * you'd be correct. However, consider
1009 * Here, float min, max are 3,5 and minlen is 7.
1010 * This can match either
1014 * In the first case, the regex matches minlen chars; in the
1015 * second, minlen+1, in the third, minlen+2.
1016 * In the first case, the floating offset is 3 (which equals
1017 * float_min), in the second, 4, and in the third, 5 (which
1018 * equals float_max). In all cases, the floating string bcd
1019 * can never start more than 4 chars from the end of the
1020 * string, which equals minlen - float_min. As the substring
1021 * starts to match more than float_min from the start of the
1022 * regex, it makes the regex match more than minlen chars,
1023 * and the two cancel each other out. So we can always use
1024 * float_min - minlen, rather than float_max - minlen for the
1025 * latest position in the string.
1027 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1028 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1031 assert(prog->minlen >= other->min_offset);
1032 last1 = HOP3c(strend,
1033 other->min_offset - prog->minlen, strbeg);
1035 if (other_ix) {/* i.e. if (other-is-float) */
1036 /* last is the latest point where the floating substr could
1037 * start, *given* any constraints from the earlier fixed
1038 * match. This constraint is that the floating string starts
1039 * <= float_max_offset chars from the regex origin (rx_origin).
1040 * If this value is less than last1, use it instead.
1042 assert(rx_origin <= last1);
1044 /* this condition handles the offset==infinity case, and
1045 * is a short-cut otherwise. Although it's comparing a
1046 * byte offset to a char length, it does so in a safe way,
1047 * since 1 char always occupies 1 or more bytes,
1048 * so if a string range is (last1 - rx_origin) bytes,
1049 * it will be less than or equal to (last1 - rx_origin)
1050 * chars; meaning it errs towards doing the accurate HOP3
1051 * rather than just using last1 as a short-cut */
1052 (last1 - rx_origin) < other->max_offset
1054 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1057 assert(strpos + start_shift <= check_at);
1058 last = HOP4c(check_at, other->min_offset - start_shift,
1062 s = HOP3c(rx_origin, other->min_offset, strend);
1063 if (s < other_last) /* These positions already checked */
1066 must = utf8_target ? other->utf8_substr : other->substr;
1067 assert(SvPOK(must));
1070 (unsigned char*)last + SvCUR(must) - (SvTAIL(must)!=0),
1072 multiline ? FBMrf_MULTILINE : 0
1075 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1076 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1077 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1078 s ? "Found" : "Contradicts",
1079 other_ix ? "floating" : "anchored",
1080 quoted, RE_SV_TAIL(must));
1085 /* last1 is latest possible substr location. If we didn't
1086 * find it before there, we never will */
1087 if (last >= last1) {
1088 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1089 ", giving up...\n"));
1093 /* try to find the check substr again at a later
1094 * position. Maybe next time we'll find the "other" substr
1096 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1097 ", trying %s at offset %ld...\n",
1098 (other_ix ? "floating" : "anchored"),
1099 (long)(HOP3c(check_at, 1, strend) - strpos)));
1101 other_last = HOP3c(last, 1, strend) /* highest failure */;
1103 other_ix /* i.e. if other-is-float */
1104 ? HOP3c(rx_origin, 1, strend)
1105 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1109 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " at offset %ld...\n",
1110 (long)(s - strpos)));
1112 if (other_ix) { /* if (other-is-float) */
1113 /* other_last is set to s, not s+1, since its possible for
1114 * a floating substr to fail first time, then succeed
1115 * second time at the same floating position; e.g.:
1116 * "-AB--AABZ" =~ /\wAB\d*Z/
1117 * The first time round, anchored and float match at
1118 * "-(AB)--AAB(Z)" then fail on the initial \w character
1119 * class. Second time round, they match at "-AB--A(AB)(Z)".
1124 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1125 other_last = HOP3c(s, 1, strend);
1130 DEBUG_OPTIMISE_MORE_r(
1131 PerlIO_printf(Perl_debug_log,
1132 " Check-only match: offset min:%"IVdf" max:%"IVdf
1133 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
1134 " strend-strpos:%"IVdf"\n",
1135 (IV)prog->check_offset_min,
1136 (IV)prog->check_offset_max,
1137 (IV)(check_at-strpos),
1138 (IV)(rx_origin-strpos),
1139 (IV)(rx_origin-check_at),
1145 postprocess_substr_matches:
1147 /* handle the extra constraint of /^.../m if present */
1149 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1152 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1153 " looking for /^/m anchor"));
1155 /* we have failed the constraint of a \n before rx_origin.
1156 * Find the next \n, if any, even if it's beyond the current
1157 * anchored and/or floating substrings. Whether we should be
1158 * scanning ahead for the next \n or the next substr is debatable.
1159 * On the one hand you'd expect rare substrings to appear less
1160 * often than \n's. On the other hand, searching for \n means
1161 * we're effectively flipping been check_substr and "\n" on each
1162 * iteration as the current "rarest" string candidate, which
1163 * means for example that we'll quickly reject the whole string if
1164 * hasn't got a \n, rather than trying every substr position
1168 s = HOP3c(strend, - prog->minlen, strpos);
1169 if (s <= rx_origin ||
1170 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1172 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1173 " Did not find /%s^%s/m...\n",
1174 PL_colors[0], PL_colors[1]));
1178 /* earliest possible origin is 1 char after the \n.
1179 * (since *rx_origin == '\n', it's safe to ++ here rather than
1180 * HOP(rx_origin, 1)) */
1183 if (prog->substrs->check_ix == 0 /* check is anchored */
1184 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1186 /* Position contradicts check-string; either because
1187 * check was anchored (and thus has no wiggle room),
1188 * or check was float and rx_origin is above the float range */
1189 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1190 " Found /%s^%s/m, restarting lookup for check-string at offset %ld...\n",
1191 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1195 /* if we get here, the check substr must have been float,
1196 * is in range, and we may or may not have had an anchored
1197 * "other" substr which still contradicts */
1198 assert(prog->substrs->check_ix); /* check is float */
1200 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1201 /* whoops, the anchored "other" substr exists, so we still
1202 * contradict. On the other hand, the float "check" substr
1203 * didn't contradict, so just retry the anchored "other"
1205 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1206 " Found /%s^%s/m at offset %ld, rescanning for anchored from offset %ld...\n",
1207 PL_colors[0], PL_colors[1],
1208 (long)(rx_origin - strpos),
1209 (long)(rx_origin - strpos + prog->anchored_offset)));
1210 goto do_other_substr;
1213 /* success: we don't contradict the found floating substring
1214 * (and there's no anchored substr). */
1215 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1216 " Found /%s^%s/m at offset %ld...\n",
1217 PL_colors[0], PL_colors[1], (long)(rx_origin - strpos)));
1220 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1221 " (multiline anchor test skipped)\n"));
1227 /* if we have a starting character class, then test that extra constraint.
1228 * (trie stclasses are too expensive to use here, we are better off to
1229 * leave it to regmatch itself) */
1231 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1232 const U8* const str = (U8*)STRING(progi->regstclass);
1234 /* XXX this value could be pre-computed */
1235 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1236 ? (reginfo->is_utf8_pat
1237 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1238 : STR_LEN(progi->regstclass))
1242 /* latest pos that a matching float substr constrains rx start to */
1243 char *rx_max_float = NULL;
1245 /* if the current rx_origin is anchored, either by satisfying an
1246 * anchored substring constraint, or a /^.../m constraint, then we
1247 * can reject the current origin if the start class isn't found
1248 * at the current position. If we have a float-only match, then
1249 * rx_origin is constrained to a range; so look for the start class
1250 * in that range. if neither, then look for the start class in the
1251 * whole rest of the string */
1253 /* XXX DAPM it's not clear what the minlen test is for, and why
1254 * it's not used in the floating case. Nothing in the test suite
1255 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1256 * Here are some old comments, which may or may not be correct:
1258 * minlen == 0 is possible if regstclass is \b or \B,
1259 * and the fixed substr is ''$.
1260 * Since minlen is already taken into account, rx_origin+1 is
1261 * before strend; accidentally, minlen >= 1 guaranties no false
1262 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1263 * 0) below assumes that regstclass does not come from lookahead...
1264 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1265 * This leaves EXACTF-ish only, which are dealt with in
1269 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1270 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1271 else if (prog->float_substr || prog->float_utf8) {
1272 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1273 endpos= HOP3c(rx_max_float, cl_l, strend);
1278 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1279 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
1280 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1281 (IV)start_shift, (IV)(check_at - strbeg),
1282 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1284 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1287 if (endpos == strend) {
1288 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1289 " Could not match STCLASS...\n") );
1292 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1293 " This position contradicts STCLASS...\n") );
1294 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1295 && !(prog->intflags & PREGf_IMPLICIT))
1298 /* Contradict one of substrings */
1299 if (prog->anchored_substr || prog->anchored_utf8) {
1300 if (prog->substrs->check_ix == 1) { /* check is float */
1301 /* Have both, check_string is floating */
1302 assert(rx_origin + start_shift <= check_at);
1303 if (rx_origin + start_shift != check_at) {
1304 /* not at latest position float substr could match:
1305 * Recheck anchored substring, but not floating.
1306 * The condition above is in bytes rather than
1307 * chars for efficiency. It's conservative, in
1308 * that it errs on the side of doing 'goto
1309 * do_other_substr', where a more accurate
1310 * char-based calculation will be done */
1311 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1312 " Looking for anchored substr starting at offset %ld...\n",
1313 (long)(other_last - strpos)) );
1314 goto do_other_substr;
1322 /* In the presence of ml_anch, we might be able to
1323 * find another \n without breaking the current float
1326 /* strictly speaking this should be HOP3c(..., 1, ...),
1327 * but since we goto a block of code that's going to
1328 * search for the next \n if any, its safe here */
1330 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1331 " Looking for /%s^%s/m starting at offset %ld...\n",
1332 PL_colors[0], PL_colors[1],
1333 (long)(rx_origin - strpos)) );
1334 goto postprocess_substr_matches;
1337 /* strictly speaking this can never be true; but might
1338 * be if we ever allow intuit without substrings */
1339 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1342 rx_origin = rx_max_float;
1345 /* at this point, any matching substrings have been
1346 * contradicted. Start again... */
1348 rx_origin = HOP3c(rx_origin, 1, strend);
1350 /* uses bytes rather than char calculations for efficiency.
1351 * It's conservative: it errs on the side of doing 'goto restart',
1352 * where there is code that does a proper char-based test */
1353 if (rx_origin + start_shift + end_shift > strend) {
1354 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1355 " Could not match STCLASS...\n") );
1358 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1359 " Looking for %s substr starting at offset %ld...\n",
1360 (prog->substrs->check_ix ? "floating" : "anchored"),
1361 (long)(rx_origin + start_shift - strpos)) );
1367 if (rx_origin != s) {
1368 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1369 " By STCLASS: moving %ld --> %ld\n",
1370 (long)(rx_origin - strpos), (long)(s - strpos))
1374 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1375 " Does not contradict STCLASS...\n");
1380 /* Decide whether using the substrings helped */
1382 if (rx_origin != strpos) {
1383 /* Fixed substring is found far enough so that the match
1384 cannot start at strpos. */
1386 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1387 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1390 /* The found rx_origin position does not prohibit matching at
1391 * strpos, so calling intuit didn't gain us anything. Decrement
1392 * the BmUSEFUL() count on the check substring, and if we reach
1394 if (!(prog->intflags & PREGf_NAUGHTY)
1396 prog->check_utf8 /* Could be deleted already */
1397 && --BmUSEFUL(prog->check_utf8) < 0
1398 && (prog->check_utf8 == prog->float_utf8)
1400 prog->check_substr /* Could be deleted already */
1401 && --BmUSEFUL(prog->check_substr) < 0
1402 && (prog->check_substr == prog->float_substr)
1405 /* If flags & SOMETHING - do not do it many times on the same match */
1406 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1407 /* XXX Does the destruction order has to change with utf8_target? */
1408 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1409 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1410 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1411 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1412 check = NULL; /* abort */
1413 /* XXXX This is a remnant of the old implementation. It
1414 looks wasteful, since now INTUIT can use many
1415 other heuristics. */
1416 prog->extflags &= ~RXf_USE_INTUIT;
1420 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1421 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1422 PL_colors[4], PL_colors[5], (long)(rx_origin - strpos)) );
1426 fail_finish: /* Substring not found */
1427 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1428 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1430 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1431 PL_colors[4], PL_colors[5]));
1436 #define DECL_TRIE_TYPE(scan) \
1437 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1438 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold } \
1439 trie_type = ((scan->flags == EXACT) \
1440 ? (utf8_target ? trie_utf8 : trie_plain) \
1441 : (scan->flags == EXACTFA) \
1442 ? (utf8_target ? trie_utf8_exactfa_fold : trie_latin_utf8_exactfa_fold) \
1443 : (utf8_target ? trie_utf8_fold : trie_latin_utf8_fold))
1445 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1448 U8 flags = FOLD_FLAGS_FULL; \
1449 switch (trie_type) { \
1450 case trie_utf8_exactfa_fold: \
1451 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1453 case trie_utf8_fold: \
1454 if ( foldlen>0 ) { \
1455 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1460 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1461 len = UTF8SKIP(uc); \
1462 skiplen = UNISKIP( uvc ); \
1463 foldlen -= skiplen; \
1464 uscan = foldbuf + skiplen; \
1467 case trie_latin_utf8_exactfa_fold: \
1468 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1470 case trie_latin_utf8_fold: \
1471 if ( foldlen>0 ) { \
1472 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1478 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1479 skiplen = UNISKIP( uvc ); \
1480 foldlen -= skiplen; \
1481 uscan = foldbuf + skiplen; \
1485 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1492 charid = trie->charmap[ uvc ]; \
1496 if (widecharmap) { \
1497 SV** const svpp = hv_fetch(widecharmap, \
1498 (char*)&uvc, sizeof(UV), 0); \
1500 charid = (U16)SvIV(*svpp); \
1505 #define DUMP_EXEC_POS(li,s,doutf8) \
1506 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1509 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1513 && (ln == 1 || folder(s, pat_string, ln)) \
1514 && (reginfo->intuit || regtry(reginfo, &s)) )\
1520 #define REXEC_FBC_UTF8_SCAN(CODE) \
1522 while (s < strend) { \
1528 #define REXEC_FBC_SCAN(CODE) \
1530 while (s < strend) { \
1536 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1537 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1539 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1548 #define REXEC_FBC_CLASS_SCAN(COND) \
1549 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1551 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1560 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1561 if (utf8_target) { \
1562 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1565 REXEC_FBC_CLASS_SCAN(COND); \
1568 /* The three macros below are slightly different versions of the same logic.
1570 * The first is for /a and /aa when the target string is UTF-8. This can only
1571 * match ascii, but it must advance based on UTF-8. The other two handle the
1572 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1573 * for the boundary (or non-boundary) between a word and non-word character.
1574 * The utf8 and non-utf8 cases have the same logic, but the details must be
1575 * different. Find the "wordness" of the character just prior to this one, and
1576 * compare it with the wordness of this one. If they differ, we have a
1577 * boundary. At the beginning of the string, pretend that the previous
1578 * character was a new-line.
1580 * All these macros uncleanly have side-effects with each other and outside
1581 * variables. So far it's been too much trouble to clean-up
1583 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1584 * a word character or not.
1585 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1587 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1589 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1590 * are looking for a boundary or for a non-boundary. If we are looking for a
1591 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1592 * see if this tentative match actually works, and if so, to quit the loop
1593 * here. And vice-versa if we are looking for a non-boundary.
1595 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1596 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1597 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1598 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1599 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1600 * complement. But in that branch we complement tmp, meaning that at the
1601 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1602 * which means at the top of the loop in the next iteration, it is
1603 * TEST_NON_UTF8(s-1) */
1604 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1605 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1606 tmp = TEST_NON_UTF8(tmp); \
1607 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1608 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1610 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1617 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1618 * TEST_UTF8 is a macro that for the same input code points returns identically
1619 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1620 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1621 if (s == reginfo->strbeg) { \
1624 else { /* Back-up to the start of the previous character */ \
1625 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1626 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1627 0, UTF8_ALLOW_DEFAULT); \
1629 tmp = TEST_UV(tmp); \
1630 LOAD_UTF8_CHARCLASS_ALNUM(); \
1631 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1632 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1641 /* Like the above two macros. UTF8_CODE is the complete code for handling
1642 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1644 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1645 if (utf8_target) { \
1648 else { /* Not utf8 */ \
1649 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1650 tmp = TEST_NON_UTF8(tmp); \
1651 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1652 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1661 /* Here, things have been set up by the previous code so that tmp is the \
1662 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1663 * utf8ness of the target). We also have to check if this matches against \
1664 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1665 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1667 if (tmp == ! TEST_NON_UTF8('\n')) { \
1674 /* This is the macro to use when we want to see if something that looks like it
1675 * could match, actually does, and if so exits the loop */
1676 #define REXEC_FBC_TRYIT \
1677 if ((reginfo->intuit || regtry(reginfo, &s))) \
1680 /* The only difference between the BOUND and NBOUND cases is that
1681 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1682 * NBOUND. This is accomplished by passing it as either the if or else clause,
1683 * with the other one being empty (PLACEHOLDER is defined as empty).
1685 * The TEST_FOO parameters are for operating on different forms of input, but
1686 * all should be ones that return identically for the same underlying code
1688 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1690 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1691 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1693 #define FBC_BOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1695 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1696 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1698 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1700 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1701 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1703 #define FBC_NBOUND_A(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1705 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1706 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1709 /* We know what class REx starts with. Try to find this position... */
1710 /* if reginfo->intuit, its a dryrun */
1711 /* annoyingly all the vars in this routine have different names from their counterparts
1712 in regmatch. /grrr */
1714 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1715 const char *strend, regmatch_info *reginfo)
1718 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1719 char *pat_string; /* The pattern's exactish string */
1720 char *pat_end; /* ptr to end char of pat_string */
1721 re_fold_t folder; /* Function for computing non-utf8 folds */
1722 const U8 *fold_array; /* array for folding ords < 256 */
1728 I32 tmp = 1; /* Scratch variable? */
1729 const bool utf8_target = reginfo->is_utf8_target;
1730 UV utf8_fold_flags = 0;
1731 const bool is_utf8_pat = reginfo->is_utf8_pat;
1732 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1733 with a result inverts that result, as 0^1 =
1735 _char_class_number classnum;
1737 RXi_GET_DECL(prog,progi);
1739 PERL_ARGS_ASSERT_FIND_BYCLASS;
1741 /* We know what class it must start with. */
1745 REXEC_FBC_UTF8_CLASS_SCAN(
1746 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1749 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1754 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1761 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1762 assert(! is_utf8_pat);
1765 if (is_utf8_pat || utf8_target) {
1766 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1767 goto do_exactf_utf8;
1769 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1770 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1771 goto do_exactf_non_utf8; /* isn't dealt with by these */
1773 case EXACTF: /* This node only generated for non-utf8 patterns */
1774 assert(! is_utf8_pat);
1776 utf8_fold_flags = 0;
1777 goto do_exactf_utf8;
1779 fold_array = PL_fold;
1781 goto do_exactf_non_utf8;
1784 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1785 utf8_fold_flags = FOLDEQ_LOCALE;
1786 goto do_exactf_utf8;
1788 fold_array = PL_fold_locale;
1789 folder = foldEQ_locale;
1790 goto do_exactf_non_utf8;
1794 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1796 goto do_exactf_utf8;
1799 if (is_utf8_pat || utf8_target) {
1800 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1801 goto do_exactf_utf8;
1804 /* Any 'ss' in the pattern should have been replaced by regcomp,
1805 * so we don't have to worry here about this single special case
1806 * in the Latin1 range */
1807 fold_array = PL_fold_latin1;
1808 folder = foldEQ_latin1;
1812 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1813 are no glitches with fold-length differences
1814 between the target string and pattern */
1816 /* The idea in the non-utf8 EXACTF* cases is to first find the
1817 * first character of the EXACTF* node and then, if necessary,
1818 * case-insensitively compare the full text of the node. c1 is the
1819 * first character. c2 is its fold. This logic will not work for
1820 * Unicode semantics and the german sharp ss, which hence should
1821 * not be compiled into a node that gets here. */
1822 pat_string = STRING(c);
1823 ln = STR_LEN(c); /* length to match in octets/bytes */
1825 /* We know that we have to match at least 'ln' bytes (which is the
1826 * same as characters, since not utf8). If we have to match 3
1827 * characters, and there are only 2 availabe, we know without
1828 * trying that it will fail; so don't start a match past the
1829 * required minimum number from the far end */
1830 e = HOP3c(strend, -((SSize_t)ln), s);
1832 if (reginfo->intuit && e < s) {
1833 e = s; /* Due to minlen logic of intuit() */
1837 c2 = fold_array[c1];
1838 if (c1 == c2) { /* If char and fold are the same */
1839 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1842 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1850 /* If one of the operands is in utf8, we can't use the simpler folding
1851 * above, due to the fact that many different characters can have the
1852 * same fold, or portion of a fold, or different- length fold */
1853 pat_string = STRING(c);
1854 ln = STR_LEN(c); /* length to match in octets/bytes */
1855 pat_end = pat_string + ln;
1856 lnc = is_utf8_pat /* length to match in characters */
1857 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1860 /* We have 'lnc' characters to match in the pattern, but because of
1861 * multi-character folding, each character in the target can match
1862 * up to 3 characters (Unicode guarantees it will never exceed
1863 * this) if it is utf8-encoded; and up to 2 if not (based on the
1864 * fact that the Latin 1 folds are already determined, and the
1865 * only multi-char fold in that range is the sharp-s folding to
1866 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1867 * string character. Adjust lnc accordingly, rounding up, so that
1868 * if we need to match at least 4+1/3 chars, that really is 5. */
1869 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1870 lnc = (lnc + expansion - 1) / expansion;
1872 /* As in the non-UTF8 case, if we have to match 3 characters, and
1873 * only 2 are left, it's guaranteed to fail, so don't start a
1874 * match that would require us to go beyond the end of the string
1876 e = HOP3c(strend, -((SSize_t)lnc), s);
1878 if (reginfo->intuit && e < s) {
1879 e = s; /* Due to minlen logic of intuit() */
1882 /* XXX Note that we could recalculate e to stop the loop earlier,
1883 * as the worst case expansion above will rarely be met, and as we
1884 * go along we would usually find that e moves further to the left.
1885 * This would happen only after we reached the point in the loop
1886 * where if there were no expansion we should fail. Unclear if
1887 * worth the expense */
1890 char *my_strend= (char *)strend;
1891 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1892 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1893 && (reginfo->intuit || regtry(reginfo, &s)) )
1897 s += (utf8_target) ? UTF8SKIP(s) : 1;
1903 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1906 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
1909 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1912 FBC_BOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1915 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
1918 FBC_NBOUND_A(isWORDCHAR_A, isWORDCHAR_A, isWORDCHAR_A);
1921 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1924 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
1927 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
1928 is_LNBREAK_latin1_safe(s, strend)
1932 /* The argument to all the POSIX node types is the class number to pass to
1933 * _generic_isCC() to build a mask for searching in PL_charclass[] */
1940 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
1941 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
1956 /* The complement of something that matches only ASCII matches all
1957 * non-ASCII, plus everything in ASCII that isn't in the class. */
1958 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
1959 || ! _generic_isCC_A(*s, FLAGS(c)));
1968 /* Don't need to worry about utf8, as it can match only a single
1969 * byte invariant character. */
1970 REXEC_FBC_CLASS_SCAN(
1971 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
1979 if (! utf8_target) {
1980 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
1986 classnum = (_char_class_number) FLAGS(c);
1987 if (classnum < _FIRST_NON_SWASH_CC) {
1988 while (s < strend) {
1990 /* We avoid loading in the swash as long as possible, but
1991 * should we have to, we jump to a separate loop. This
1992 * extra 'if' statement is what keeps this code from being
1993 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
1994 if (UTF8_IS_ABOVE_LATIN1(*s)) {
1995 goto found_above_latin1;
1997 if ((UTF8_IS_INVARIANT(*s)
1998 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2000 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2001 && to_complement ^ cBOOL(
2002 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2006 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2018 else switch (classnum) { /* These classes are implemented as
2020 case _CC_ENUM_SPACE: /* XXX would require separate code if we
2021 revert the change of \v matching this */
2024 case _CC_ENUM_PSXSPC:
2025 REXEC_FBC_UTF8_CLASS_SCAN(
2026 to_complement ^ cBOOL(isSPACE_utf8(s)));
2029 case _CC_ENUM_BLANK:
2030 REXEC_FBC_UTF8_CLASS_SCAN(
2031 to_complement ^ cBOOL(isBLANK_utf8(s)));
2034 case _CC_ENUM_XDIGIT:
2035 REXEC_FBC_UTF8_CLASS_SCAN(
2036 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2039 case _CC_ENUM_VERTSPACE:
2040 REXEC_FBC_UTF8_CLASS_SCAN(
2041 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2044 case _CC_ENUM_CNTRL:
2045 REXEC_FBC_UTF8_CLASS_SCAN(
2046 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2050 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2051 assert(0); /* NOTREACHED */
2056 found_above_latin1: /* Here we have to load a swash to get the result
2057 for the current code point */
2058 if (! PL_utf8_swash_ptrs[classnum]) {
2059 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2060 PL_utf8_swash_ptrs[classnum] =
2061 _core_swash_init("utf8",
2064 PL_XPosix_ptrs[classnum], &flags);
2067 /* This is a copy of the loop above for swash classes, though using the
2068 * FBC macro instead of being expanded out. Since we've loaded the
2069 * swash, we don't have to check for that each time through the loop */
2070 REXEC_FBC_UTF8_CLASS_SCAN(
2071 to_complement ^ cBOOL(_generic_utf8(
2074 swash_fetch(PL_utf8_swash_ptrs[classnum],
2082 /* what trie are we using right now */
2083 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2084 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2085 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2087 const char *last_start = strend - trie->minlen;
2089 const char *real_start = s;
2091 STRLEN maxlen = trie->maxlen;
2093 U8 **points; /* map of where we were in the input string
2094 when reading a given char. For ASCII this
2095 is unnecessary overhead as the relationship
2096 is always 1:1, but for Unicode, especially
2097 case folded Unicode this is not true. */
2098 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2102 GET_RE_DEBUG_FLAGS_DECL;
2104 /* We can't just allocate points here. We need to wrap it in
2105 * an SV so it gets freed properly if there is a croak while
2106 * running the match */
2109 sv_points=newSV(maxlen * sizeof(U8 *));
2110 SvCUR_set(sv_points,
2111 maxlen * sizeof(U8 *));
2112 SvPOK_on(sv_points);
2113 sv_2mortal(sv_points);
2114 points=(U8**)SvPV_nolen(sv_points );
2115 if ( trie_type != trie_utf8_fold
2116 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2119 bitmap=(U8*)trie->bitmap;
2121 bitmap=(U8*)ANYOF_BITMAP(c);
2123 /* this is the Aho-Corasick algorithm modified a touch
2124 to include special handling for long "unknown char" sequences.
2125 The basic idea being that we use AC as long as we are dealing
2126 with a possible matching char, when we encounter an unknown char
2127 (and we have not encountered an accepting state) we scan forward
2128 until we find a legal starting char.
2129 AC matching is basically that of trie matching, except that when
2130 we encounter a failing transition, we fall back to the current
2131 states "fail state", and try the current char again, a process
2132 we repeat until we reach the root state, state 1, or a legal
2133 transition. If we fail on the root state then we can either
2134 terminate if we have reached an accepting state previously, or
2135 restart the entire process from the beginning if we have not.
2138 while (s <= last_start) {
2139 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2147 U8 *uscan = (U8*)NULL;
2148 U8 *leftmost = NULL;
2150 U32 accepted_word= 0;
2154 while ( state && uc <= (U8*)strend ) {
2156 U32 word = aho->states[ state ].wordnum;
2160 DEBUG_TRIE_EXECUTE_r(
2161 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2162 dump_exec_pos( (char *)uc, c, strend, real_start,
2163 (char *)uc, utf8_target );
2164 PerlIO_printf( Perl_debug_log,
2165 " Scanning for legal start char...\n");
2169 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2173 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2179 if (uc >(U8*)last_start) break;
2183 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2184 if (!leftmost || lpos < leftmost) {
2185 DEBUG_r(accepted_word=word);
2191 points[pointpos++ % maxlen]= uc;
2192 if (foldlen || uc < (U8*)strend) {
2193 REXEC_TRIE_READ_CHAR(trie_type, trie,
2195 uscan, len, uvc, charid, foldlen,
2197 DEBUG_TRIE_EXECUTE_r({
2198 dump_exec_pos( (char *)uc, c, strend,
2199 real_start, s, utf8_target);
2200 PerlIO_printf(Perl_debug_log,
2201 " Charid:%3u CP:%4"UVxf" ",
2213 word = aho->states[ state ].wordnum;
2215 base = aho->states[ state ].trans.base;
2217 DEBUG_TRIE_EXECUTE_r({
2219 dump_exec_pos( (char *)uc, c, strend, real_start,
2221 PerlIO_printf( Perl_debug_log,
2222 "%sState: %4"UVxf", word=%"UVxf,
2223 failed ? " Fail transition to " : "",
2224 (UV)state, (UV)word);
2230 ( ((offset = base + charid
2231 - 1 - trie->uniquecharcount)) >= 0)
2232 && ((U32)offset < trie->lasttrans)
2233 && trie->trans[offset].check == state
2234 && (tmp=trie->trans[offset].next))
2236 DEBUG_TRIE_EXECUTE_r(
2237 PerlIO_printf( Perl_debug_log," - legal\n"));
2242 DEBUG_TRIE_EXECUTE_r(
2243 PerlIO_printf( Perl_debug_log," - fail\n"));
2245 state = aho->fail[state];
2249 /* we must be accepting here */
2250 DEBUG_TRIE_EXECUTE_r(
2251 PerlIO_printf( Perl_debug_log," - accepting\n"));
2260 if (!state) state = 1;
2263 if ( aho->states[ state ].wordnum ) {
2264 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2265 if (!leftmost || lpos < leftmost) {
2266 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2271 s = (char*)leftmost;
2272 DEBUG_TRIE_EXECUTE_r({
2274 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2275 (UV)accepted_word, (IV)(s - real_start)
2278 if (reginfo->intuit || regtry(reginfo, &s)) {
2284 DEBUG_TRIE_EXECUTE_r({
2285 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2288 DEBUG_TRIE_EXECUTE_r(
2289 PerlIO_printf( Perl_debug_log,"No match.\n"));
2298 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2305 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2306 * flags have same meanings as with regexec_flags() */
2309 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2316 struct regexp *const prog = ReANY(rx);
2318 if (flags & REXEC_COPY_STR) {
2322 PerlIO_printf(Perl_debug_log,
2323 "Copy on write: regexp capture, type %d\n",
2326 /* Create a new COW SV to share the match string and store
2327 * in saved_copy, unless the current COW SV in saved_copy
2328 * is valid and suitable for our purpose */
2329 if (( prog->saved_copy
2330 && SvIsCOW(prog->saved_copy)
2331 && SvPOKp(prog->saved_copy)
2334 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2336 /* just reuse saved_copy SV */
2337 if (RXp_MATCH_COPIED(prog)) {
2338 Safefree(prog->subbeg);
2339 RXp_MATCH_COPIED_off(prog);
2343 /* create new COW SV to share string */
2344 RX_MATCH_COPY_FREE(rx);
2345 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2347 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2348 assert (SvPOKp(prog->saved_copy));
2349 prog->sublen = strend - strbeg;
2350 prog->suboffset = 0;
2351 prog->subcoffset = 0;
2356 SSize_t max = strend - strbeg;
2359 if ( (flags & REXEC_COPY_SKIP_POST)
2360 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2361 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2362 ) { /* don't copy $' part of string */
2365 /* calculate the right-most part of the string covered
2366 * by a capture. Due to look-ahead, this may be to
2367 * the right of $&, so we have to scan all captures */
2368 while (n <= prog->lastparen) {
2369 if (prog->offs[n].end > max)
2370 max = prog->offs[n].end;
2374 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2375 ? prog->offs[0].start
2377 assert(max >= 0 && max <= strend - strbeg);
2380 if ( (flags & REXEC_COPY_SKIP_PRE)
2381 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2382 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2383 ) { /* don't copy $` part of string */
2386 /* calculate the left-most part of the string covered
2387 * by a capture. Due to look-behind, this may be to
2388 * the left of $&, so we have to scan all captures */
2389 while (min && n <= prog->lastparen) {
2390 if ( prog->offs[n].start != -1
2391 && prog->offs[n].start < min)
2393 min = prog->offs[n].start;
2397 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2398 && min > prog->offs[0].end
2400 min = prog->offs[0].end;
2404 assert(min >= 0 && min <= max && min <= strend - strbeg);
2407 if (RX_MATCH_COPIED(rx)) {
2408 if (sublen > prog->sublen)
2410 (char*)saferealloc(prog->subbeg, sublen+1);
2413 prog->subbeg = (char*)safemalloc(sublen+1);
2414 Copy(strbeg + min, prog->subbeg, sublen, char);
2415 prog->subbeg[sublen] = '\0';
2416 prog->suboffset = min;
2417 prog->sublen = sublen;
2418 RX_MATCH_COPIED_on(rx);
2420 prog->subcoffset = prog->suboffset;
2421 if (prog->suboffset && utf8_target) {
2422 /* Convert byte offset to chars.
2423 * XXX ideally should only compute this if @-/@+
2424 * has been seen, a la PL_sawampersand ??? */
2426 /* If there's a direct correspondence between the
2427 * string which we're matching and the original SV,
2428 * then we can use the utf8 len cache associated with
2429 * the SV. In particular, it means that under //g,
2430 * sv_pos_b2u() will use the previously cached
2431 * position to speed up working out the new length of
2432 * subcoffset, rather than counting from the start of
2433 * the string each time. This stops
2434 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2435 * from going quadratic */
2436 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2437 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2438 SV_GMAGIC|SV_CONST_RETURN);
2440 prog->subcoffset = utf8_length((U8*)strbeg,
2441 (U8*)(strbeg+prog->suboffset));
2445 RX_MATCH_COPY_FREE(rx);
2446 prog->subbeg = strbeg;
2447 prog->suboffset = 0;
2448 prog->subcoffset = 0;
2449 prog->sublen = strend - strbeg;
2457 - regexec_flags - match a regexp against a string
2460 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2461 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2462 /* stringarg: the point in the string at which to begin matching */
2463 /* strend: pointer to null at end of string */
2464 /* strbeg: real beginning of string */
2465 /* minend: end of match must be >= minend bytes after stringarg. */
2466 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2467 * itself is accessed via the pointers above */
2468 /* data: May be used for some additional optimizations.
2469 Currently unused. */
2470 /* flags: For optimizations. See REXEC_* in regexp.h */
2473 struct regexp *const prog = ReANY(rx);
2477 SSize_t minlen; /* must match at least this many chars */
2478 SSize_t dontbother = 0; /* how many characters not to try at end */
2479 const bool utf8_target = cBOOL(DO_UTF8(sv));
2481 RXi_GET_DECL(prog,progi);
2482 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2483 regmatch_info *const reginfo = ®info_buf;
2484 regexp_paren_pair *swap = NULL;
2486 GET_RE_DEBUG_FLAGS_DECL;
2488 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2489 PERL_UNUSED_ARG(data);
2491 /* Be paranoid... */
2492 if (prog == NULL || stringarg == NULL) {
2493 Perl_croak(aTHX_ "NULL regexp parameter");
2497 debug_start_match(rx, utf8_target, stringarg, strend,
2501 startpos = stringarg;
2503 if (prog->intflags & PREGf_GPOS_SEEN) {
2506 /* set reginfo->ganch, the position where \G can match */
2509 (flags & REXEC_IGNOREPOS)
2510 ? stringarg /* use start pos rather than pos() */
2511 : (sv && (mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2512 /* Defined pos(): */
2513 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2514 : strbeg; /* pos() not defined; use start of string */
2516 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2517 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
2519 /* in the presence of \G, we may need to start looking earlier in
2520 * the string than the suggested start point of stringarg:
2521 * if prog->gofs is set, then that's a known, fixed minimum
2524 * /ab|c\G/: gofs = 1
2525 * or if the minimum offset isn't known, then we have to go back
2526 * to the start of the string, e.g. /w+\G/
2529 if (prog->intflags & PREGf_ANCH_GPOS) {
2530 startpos = reginfo->ganch - prog->gofs;
2532 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2534 DEBUG_r(PerlIO_printf(Perl_debug_log,
2535 "fail: ganch-gofs before earliest possible start\n"));
2539 else if (prog->gofs) {
2540 if (startpos - prog->gofs < strbeg)
2543 startpos -= prog->gofs;
2545 else if (prog->intflags & PREGf_GPOS_FLOAT)
2549 minlen = prog->minlen;
2550 if ((startpos + minlen) > strend || startpos < strbeg) {
2551 DEBUG_r(PerlIO_printf(Perl_debug_log,
2552 "Regex match can't succeed, so not even tried\n"));
2556 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2557 * which will call destuctors to reset PL_regmatch_state, free higher
2558 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2559 * regmatch_info_aux_eval */
2561 oldsave = PL_savestack_ix;
2565 if ((prog->extflags & RXf_USE_INTUIT)
2566 && !(flags & REXEC_CHECKED))
2568 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2573 if (prog->extflags & RXf_CHECK_ALL) {
2574 /* we can match based purely on the result of INTUIT.
2575 * Set up captures etc just for $& and $-[0]
2576 * (an intuit-only match wont have $1,$2,..) */
2577 assert(!prog->nparens);
2579 /* s/// doesn't like it if $& is earlier than where we asked it to
2580 * start searching (which can happen on something like /.\G/) */
2581 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2584 /* this should only be possible under \G */
2585 assert(prog->intflags & PREGf_GPOS_SEEN);
2586 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2587 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2591 /* match via INTUIT shouldn't have any captures.
2592 * Let @-, @+, $^N know */
2593 prog->lastparen = prog->lastcloseparen = 0;
2594 RX_MATCH_UTF8_set(rx, utf8_target);
2595 prog->offs[0].start = s - strbeg;
2596 prog->offs[0].end = utf8_target
2597 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2598 : s - strbeg + prog->minlenret;
2599 if ( !(flags & REXEC_NOT_FIRST) )
2600 S_reg_set_capture_string(aTHX_ rx,
2602 sv, flags, utf8_target);
2608 multiline = prog->extflags & RXf_PMf_MULTILINE;
2610 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2611 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2612 "String too short [regexec_flags]...\n"));
2616 /* Check validity of program. */
2617 if (UCHARAT(progi->program) != REG_MAGIC) {
2618 Perl_croak(aTHX_ "corrupted regexp program");
2621 RX_MATCH_TAINTED_off(rx);
2623 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2624 reginfo->intuit = 0;
2625 reginfo->is_utf8_target = cBOOL(utf8_target);
2626 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2627 reginfo->warned = FALSE;
2628 reginfo->strbeg = strbeg;
2630 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2631 reginfo->strend = strend;
2632 /* see how far we have to get to not match where we matched before */
2633 reginfo->till = stringarg + minend;
2635 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
2636 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2637 S_cleanup_regmatch_info_aux has executed (registered by
2638 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2639 magic belonging to this SV.
2640 Not newSVsv, either, as it does not COW.
2642 assert(!IS_PADGV(sv));
2643 reginfo->sv = newSV(0);
2644 SvSetSV_nosteal(reginfo->sv, sv);
2645 SAVEFREESV(reginfo->sv);
2648 /* reserve next 2 or 3 slots in PL_regmatch_state:
2649 * slot N+0: may currently be in use: skip it
2650 * slot N+1: use for regmatch_info_aux struct
2651 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2652 * slot N+3: ready for use by regmatch()
2656 regmatch_state *old_regmatch_state;
2657 regmatch_slab *old_regmatch_slab;
2658 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2660 /* on first ever match, allocate first slab */
2661 if (!PL_regmatch_slab) {
2662 Newx(PL_regmatch_slab, 1, regmatch_slab);
2663 PL_regmatch_slab->prev = NULL;
2664 PL_regmatch_slab->next = NULL;
2665 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
2668 old_regmatch_state = PL_regmatch_state;
2669 old_regmatch_slab = PL_regmatch_slab;
2671 for (i=0; i <= max; i++) {
2673 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
2675 reginfo->info_aux_eval =
2676 reginfo->info_aux->info_aux_eval =
2677 &(PL_regmatch_state->u.info_aux_eval);
2679 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
2680 PL_regmatch_state = S_push_slab(aTHX);
2683 /* note initial PL_regmatch_state position; at end of match we'll
2684 * pop back to there and free any higher slabs */
2686 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
2687 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
2688 reginfo->info_aux->poscache = NULL;
2690 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
2692 if ((prog->extflags & RXf_EVAL_SEEN))
2693 S_setup_eval_state(aTHX_ reginfo);
2695 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
2698 /* If there is a "must appear" string, look for it. */
2700 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
2701 /* We have to be careful. If the previous successful match
2702 was from this regex we don't want a subsequent partially
2703 successful match to clobber the old results.
2704 So when we detect this possibility we add a swap buffer
2705 to the re, and switch the buffer each match. If we fail,
2706 we switch it back; otherwise we leave it swapped.
2709 /* do we need a save destructor here for eval dies? */
2710 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
2711 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
2712 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
2719 /* Simplest case: anchored match need be tried only once. */
2720 /* [unless only anchor is BOL and multiline is set] */
2721 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
2722 if (s == startpos && regtry(reginfo, &s))
2724 else if (multiline || (prog->intflags & (PREGf_IMPLICIT | PREGf_ANCH_MBOL))) /* XXXX SBOL? */
2729 dontbother = minlen - 1;
2730 end = HOP3c(strend, -dontbother, strbeg) - 1;
2731 /* for multiline we only have to try after newlines */
2732 if (prog->check_substr || prog->check_utf8) {
2733 /* because of the goto we can not easily reuse the macros for bifurcating the
2734 unicode/non-unicode match modes here like we do elsewhere - demerphq */
2737 goto after_try_utf8;
2739 if (regtry(reginfo, &s)) {
2746 if (prog->extflags & RXf_USE_INTUIT) {
2747 s = re_intuit_start(rx, sv, strbeg,
2748 s + UTF8SKIP(s), strend, flags, NULL);
2757 } /* end search for check string in unicode */
2759 if (s == startpos) {
2760 goto after_try_latin;
2763 if (regtry(reginfo, &s)) {
2770 if (prog->extflags & RXf_USE_INTUIT) {
2771 s = re_intuit_start(rx, sv, strbeg,
2772 s + 1, strend, flags, NULL);
2781 } /* end search for check string in latin*/
2782 } /* end search for check string */
2783 else { /* search for newline */
2785 /*XXX: The s-- is almost definitely wrong here under unicode - demeprhq*/
2788 /* We can use a more efficient search as newlines are the same in unicode as they are in latin */
2789 while (s <= end) { /* note it could be possible to match at the end of the string */
2790 if (*s++ == '\n') { /* don't need PL_utf8skip here */
2791 if (regtry(reginfo, &s))
2795 } /* end search for newline */
2796 } /* end anchored/multiline check string search */
2798 } else if (prog->intflags & PREGf_ANCH_GPOS)
2800 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
2801 assert(prog->intflags & PREGf_GPOS_SEEN);
2802 /* For anchored \G, the only position it can match from is
2803 * (ganch-gofs); we already set startpos to this above; if intuit
2804 * moved us on from there, we can't possibly succeed */
2805 assert(startpos == reginfo->ganch - prog->gofs);
2806 if (s == startpos && regtry(reginfo, &s))
2811 /* Messy cases: unanchored match. */
2812 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
2813 /* we have /x+whatever/ */
2814 /* it must be a one character string (XXXX Except is_utf8_pat?) */
2820 if (! prog->anchored_utf8) {
2821 to_utf8_substr(prog);
2823 ch = SvPVX_const(prog->anchored_utf8)[0];
2826 DEBUG_EXECUTE_r( did_match = 1 );
2827 if (regtry(reginfo, &s)) goto got_it;
2829 while (s < strend && *s == ch)
2836 if (! prog->anchored_substr) {
2837 if (! to_byte_substr(prog)) {
2838 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2841 ch = SvPVX_const(prog->anchored_substr)[0];
2844 DEBUG_EXECUTE_r( did_match = 1 );
2845 if (regtry(reginfo, &s)) goto got_it;
2847 while (s < strend && *s == ch)
2852 DEBUG_EXECUTE_r(if (!did_match)
2853 PerlIO_printf(Perl_debug_log,
2854 "Did not find anchored character...\n")
2857 else if (prog->anchored_substr != NULL
2858 || prog->anchored_utf8 != NULL
2859 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
2860 && prog->float_max_offset < strend - s)) {
2865 char *last1; /* Last position checked before */
2869 if (prog->anchored_substr || prog->anchored_utf8) {
2871 if (! prog->anchored_utf8) {
2872 to_utf8_substr(prog);
2874 must = prog->anchored_utf8;
2877 if (! prog->anchored_substr) {
2878 if (! to_byte_substr(prog)) {
2879 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2882 must = prog->anchored_substr;
2884 back_max = back_min = prog->anchored_offset;
2887 if (! prog->float_utf8) {
2888 to_utf8_substr(prog);
2890 must = prog->float_utf8;
2893 if (! prog->float_substr) {
2894 if (! to_byte_substr(prog)) {
2895 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
2898 must = prog->float_substr;
2900 back_max = prog->float_max_offset;
2901 back_min = prog->float_min_offset;
2907 last = HOP3c(strend, /* Cannot start after this */
2908 -(SSize_t)(CHR_SVLEN(must)
2909 - (SvTAIL(must) != 0) + back_min), strbeg);
2911 if (s > reginfo->strbeg)
2912 last1 = HOPc(s, -1);
2914 last1 = s - 1; /* bogus */
2916 /* XXXX check_substr already used to find "s", can optimize if
2917 check_substr==must. */
2919 strend = HOPc(strend, -dontbother);
2920 while ( (s <= last) &&
2921 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
2922 (unsigned char*)strend, must,
2923 multiline ? FBMrf_MULTILINE : 0)) ) {
2924 DEBUG_EXECUTE_r( did_match = 1 );
2925 if (HOPc(s, -back_max) > last1) {
2926 last1 = HOPc(s, -back_min);
2927 s = HOPc(s, -back_max);
2930 char * const t = (last1 >= reginfo->strbeg)
2931 ? HOPc(last1, 1) : last1 + 1;
2933 last1 = HOPc(s, -back_min);
2937 while (s <= last1) {
2938 if (regtry(reginfo, &s))
2941 s++; /* to break out of outer loop */
2948 while (s <= last1) {
2949 if (regtry(reginfo, &s))
2955 DEBUG_EXECUTE_r(if (!did_match) {
2956 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
2957 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
2958 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
2959 ((must == prog->anchored_substr || must == prog->anchored_utf8)
2960 ? "anchored" : "floating"),
2961 quoted, RE_SV_TAIL(must));
2965 else if ( (c = progi->regstclass) ) {
2967 const OPCODE op = OP(progi->regstclass);
2968 /* don't bother with what can't match */
2969 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
2970 strend = HOPc(strend, -(minlen - 1));
2973 SV * const prop = sv_newmortal();
2974 regprop(prog, prop, c, reginfo);
2976 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
2978 PerlIO_printf(Perl_debug_log,
2979 "Matching stclass %.*s against %s (%d bytes)\n",
2980 (int)SvCUR(prop), SvPVX_const(prop),
2981 quoted, (int)(strend - s));
2984 if (find_byclass(prog, c, s, strend, reginfo))
2986 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
2990 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
2998 if (! prog->float_utf8) {
2999 to_utf8_substr(prog);
3001 float_real = prog->float_utf8;
3004 if (! prog->float_substr) {
3005 if (! to_byte_substr(prog)) {
3006 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3009 float_real = prog->float_substr;
3012 little = SvPV_const(float_real, len);
3013 if (SvTAIL(float_real)) {
3014 /* This means that float_real contains an artificial \n on
3015 * the end due to the presence of something like this:
3016 * /foo$/ where we can match both "foo" and "foo\n" at the
3017 * end of the string. So we have to compare the end of the
3018 * string first against the float_real without the \n and
3019 * then against the full float_real with the string. We
3020 * have to watch out for cases where the string might be
3021 * smaller than the float_real or the float_real without
3023 char *checkpos= strend - len;
3025 PerlIO_printf(Perl_debug_log,
3026 "%sChecking for float_real.%s\n",
3027 PL_colors[4], PL_colors[5]));
3028 if (checkpos + 1 < strbeg) {
3029 /* can't match, even if we remove the trailing \n
3030 * string is too short to match */
3032 PerlIO_printf(Perl_debug_log,
3033 "%sString shorter than required trailing substring, cannot match.%s\n",
3034 PL_colors[4], PL_colors[5]));
3036 } else if (memEQ(checkpos + 1, little, len - 1)) {
3037 /* can match, the end of the string matches without the
3039 last = checkpos + 1;
3040 } else if (checkpos < strbeg) {
3041 /* cant match, string is too short when the "\n" is
3044 PerlIO_printf(Perl_debug_log,
3045 "%sString does not contain required trailing substring, cannot match.%s\n",
3046 PL_colors[4], PL_colors[5]));
3048 } else if (!multiline) {
3049 /* non multiline match, so compare with the "\n" at the
3050 * end of the string */
3051 if (memEQ(checkpos, little, len)) {
3055 PerlIO_printf(Perl_debug_log,
3056 "%sString does not contain required trailing substring, cannot match.%s\n",
3057 PL_colors[4], PL_colors[5]));
3061 /* multiline match, so we have to search for a place
3062 * where the full string is located */
3068 last = rninstr(s, strend, little, little + len);
3070 last = strend; /* matching "$" */
3073 /* at one point this block contained a comment which was
3074 * probably incorrect, which said that this was a "should not
3075 * happen" case. Even if it was true when it was written I am
3076 * pretty sure it is not anymore, so I have removed the comment
3077 * and replaced it with this one. Yves */
3079 PerlIO_printf(Perl_debug_log,
3080 "String does not contain required substring, cannot match.\n"
3084 dontbother = strend - last + prog->float_min_offset;
3086 if (minlen && (dontbother < minlen))
3087 dontbother = minlen - 1;
3088 strend -= dontbother; /* this one's always in bytes! */
3089 /* We don't know much -- general case. */
3092 if (regtry(reginfo, &s))
3101 if (regtry(reginfo, &s))
3103 } while (s++ < strend);
3111 /* s/// doesn't like it if $& is earlier than where we asked it to
3112 * start searching (which can happen on something like /.\G/) */
3113 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3114 && (prog->offs[0].start < stringarg - strbeg))
3116 /* this should only be possible under \G */
3117 assert(prog->intflags & PREGf_GPOS_SEEN);
3118 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3119 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3125 PerlIO_printf(Perl_debug_log,
3126 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3133 /* clean up; this will trigger destructors that will free all slabs
3134 * above the current one, and cleanup the regmatch_info_aux
3135 * and regmatch_info_aux_eval sructs */
3137 LEAVE_SCOPE(oldsave);
3139 if (RXp_PAREN_NAMES(prog))
3140 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3142 RX_MATCH_UTF8_set(rx, utf8_target);
3144 /* make sure $`, $&, $', and $digit will work later */
3145 if ( !(flags & REXEC_NOT_FIRST) )
3146 S_reg_set_capture_string(aTHX_ rx,
3147 strbeg, reginfo->strend,
3148 sv, flags, utf8_target);
3153 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
3154 PL_colors[4], PL_colors[5]));
3156 /* clean up; this will trigger destructors that will free all slabs
3157 * above the current one, and cleanup the regmatch_info_aux
3158 * and regmatch_info_aux_eval sructs */
3160 LEAVE_SCOPE(oldsave);
3163 /* we failed :-( roll it back */
3164 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3165 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3170 Safefree(prog->offs);
3177 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3178 * Do inc before dec, in case old and new rex are the same */
3179 #define SET_reg_curpm(Re2) \
3180 if (reginfo->info_aux_eval) { \
3181 (void)ReREFCNT_inc(Re2); \
3182 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3183 PM_SETRE((PL_reg_curpm), (Re2)); \
3188 - regtry - try match at specific point
3190 STATIC I32 /* 0 failure, 1 success */
3191 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3194 REGEXP *const rx = reginfo->prog;
3195 regexp *const prog = ReANY(rx);
3197 RXi_GET_DECL(prog,progi);
3198 GET_RE_DEBUG_FLAGS_DECL;
3200 PERL_ARGS_ASSERT_REGTRY;
3202 reginfo->cutpoint=NULL;
3204 prog->offs[0].start = *startposp - reginfo->strbeg;
3205 prog->lastparen = 0;
3206 prog->lastcloseparen = 0;
3208 /* XXXX What this code is doing here?!!! There should be no need
3209 to do this again and again, prog->lastparen should take care of
3212 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3213 * Actually, the code in regcppop() (which Ilya may be meaning by
3214 * prog->lastparen), is not needed at all by the test suite
3215 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3216 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3217 * Meanwhile, this code *is* needed for the
3218 * above-mentioned test suite tests to succeed. The common theme
3219 * on those tests seems to be returning null fields from matches.
3220 * --jhi updated by dapm */
3222 if (prog->nparens) {
3223 regexp_paren_pair *pp = prog->offs;
3225 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3233 result = regmatch(reginfo, *startposp, progi->program + 1);
3235 prog->offs[0].end = result;
3238 if (reginfo->cutpoint)
3239 *startposp= reginfo->cutpoint;
3240 REGCP_UNWIND(lastcp);
3245 #define sayYES goto yes
3246 #define sayNO goto no
3247 #define sayNO_SILENT goto no_silent
3249 /* we dont use STMT_START/END here because it leads to
3250 "unreachable code" warnings, which are bogus, but distracting. */
3251 #define CACHEsayNO \
3252 if (ST.cache_mask) \
3253 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3256 /* this is used to determine how far from the left messages like
3257 'failed...' are printed. It should be set such that messages
3258 are inline with the regop output that created them.
3260 #define REPORT_CODE_OFF 32
3263 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3264 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3265 #define CHRTEST_NOT_A_CP_1 -999
3266 #define CHRTEST_NOT_A_CP_2 -998
3268 /* grab a new slab and return the first slot in it */
3270 STATIC regmatch_state *
3273 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3276 regmatch_slab *s = PL_regmatch_slab->next;
3278 Newx(s, 1, regmatch_slab);
3279 s->prev = PL_regmatch_slab;
3281 PL_regmatch_slab->next = s;
3283 PL_regmatch_slab = s;
3284 return SLAB_FIRST(s);
3288 /* push a new state then goto it */
3290 #define PUSH_STATE_GOTO(state, node, input) \
3291 pushinput = input; \
3293 st->resume_state = state; \
3296 /* push a new state with success backtracking, then goto it */
3298 #define PUSH_YES_STATE_GOTO(state, node, input) \
3299 pushinput = input; \
3301 st->resume_state = state; \
3302 goto push_yes_state;
3309 regmatch() - main matching routine
3311 This is basically one big switch statement in a loop. We execute an op,
3312 set 'next' to point the next op, and continue. If we come to a point which
3313 we may need to backtrack to on failure such as (A|B|C), we push a
3314 backtrack state onto the backtrack stack. On failure, we pop the top
3315 state, and re-enter the loop at the state indicated. If there are no more
3316 states to pop, we return failure.
3318 Sometimes we also need to backtrack on success; for example /A+/, where
3319 after successfully matching one A, we need to go back and try to
3320 match another one; similarly for lookahead assertions: if the assertion
3321 completes successfully, we backtrack to the state just before the assertion
3322 and then carry on. In these cases, the pushed state is marked as
3323 'backtrack on success too'. This marking is in fact done by a chain of
3324 pointers, each pointing to the previous 'yes' state. On success, we pop to
3325 the nearest yes state, discarding any intermediate failure-only states.
3326 Sometimes a yes state is pushed just to force some cleanup code to be
3327 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3328 it to free the inner regex.
3330 Note that failure backtracking rewinds the cursor position, while
3331 success backtracking leaves it alone.
3333 A pattern is complete when the END op is executed, while a subpattern
3334 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3335 ops trigger the "pop to last yes state if any, otherwise return true"
3338 A common convention in this function is to use A and B to refer to the two
3339 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3340 the subpattern to be matched possibly multiple times, while B is the entire
3341 rest of the pattern. Variable and state names reflect this convention.
3343 The states in the main switch are the union of ops and failure/success of
3344 substates associated with with that op. For example, IFMATCH is the op
3345 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3346 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3347 successfully matched A and IFMATCH_A_fail is a state saying that we have
3348 just failed to match A. Resume states always come in pairs. The backtrack
3349 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3350 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3351 on success or failure.
3353 The struct that holds a backtracking state is actually a big union, with
3354 one variant for each major type of op. The variable st points to the
3355 top-most backtrack struct. To make the code clearer, within each
3356 block of code we #define ST to alias the relevant union.
3358 Here's a concrete example of a (vastly oversimplified) IFMATCH
3364 #define ST st->u.ifmatch
3366 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3367 ST.foo = ...; // some state we wish to save
3369 // push a yes backtrack state with a resume value of
3370 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3372 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3375 case IFMATCH_A: // we have successfully executed A; now continue with B
3377 bar = ST.foo; // do something with the preserved value
3380 case IFMATCH_A_fail: // A failed, so the assertion failed
3381 ...; // do some housekeeping, then ...
3382 sayNO; // propagate the failure
3389 For any old-timers reading this who are familiar with the old recursive
3390 approach, the code above is equivalent to:
3392 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3401 ...; // do some housekeeping, then ...
3402 sayNO; // propagate the failure
3405 The topmost backtrack state, pointed to by st, is usually free. If you
3406 want to claim it, populate any ST.foo fields in it with values you wish to
3407 save, then do one of
3409 PUSH_STATE_GOTO(resume_state, node, newinput);
3410 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3412 which sets that backtrack state's resume value to 'resume_state', pushes a
3413 new free entry to the top of the backtrack stack, then goes to 'node'.
3414 On backtracking, the free slot is popped, and the saved state becomes the
3415 new free state. An ST.foo field in this new top state can be temporarily
3416 accessed to retrieve values, but once the main loop is re-entered, it
3417 becomes available for reuse.
3419 Note that the depth of the backtrack stack constantly increases during the
3420 left-to-right execution of the pattern, rather than going up and down with
3421 the pattern nesting. For example the stack is at its maximum at Z at the
3422 end of the pattern, rather than at X in the following:
3424 /(((X)+)+)+....(Y)+....Z/
3426 The only exceptions to this are lookahead/behind assertions and the cut,
3427 (?>A), which pop all the backtrack states associated with A before
3430 Backtrack state structs are allocated in slabs of about 4K in size.
3431 PL_regmatch_state and st always point to the currently active state,
3432 and PL_regmatch_slab points to the slab currently containing
3433 PL_regmatch_state. The first time regmatch() is called, the first slab is
3434 allocated, and is never freed until interpreter destruction. When the slab
3435 is full, a new one is allocated and chained to the end. At exit from
3436 regmatch(), slabs allocated since entry are freed.
3441 #define DEBUG_STATE_pp(pp) \
3443 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3444 PerlIO_printf(Perl_debug_log, \
3445 " %*s"pp" %s%s%s%s%s\n", \
3447 PL_reg_name[st->resume_state], \
3448 ((st==yes_state||st==mark_state) ? "[" : ""), \
3449 ((st==yes_state) ? "Y" : ""), \
3450 ((st==mark_state) ? "M" : ""), \
3451 ((st==yes_state||st==mark_state) ? "]" : "") \
3456 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3461 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3462 const char *start, const char *end, const char *blurb)
3464 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3466 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3471 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3472 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3474 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3475 start, end - start, 60);
3477 PerlIO_printf(Perl_debug_log,
3478 "%s%s REx%s %s against %s\n",
3479 PL_colors[4], blurb, PL_colors[5], s0, s1);
3481 if (utf8_target||utf8_pat)
3482 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3483 utf8_pat ? "pattern" : "",
3484 utf8_pat && utf8_target ? " and " : "",
3485 utf8_target ? "string" : ""
3491 S_dump_exec_pos(pTHX_ const char *locinput,
3492 const regnode *scan,
3493 const char *loc_regeol,
3494 const char *loc_bostr,
3495 const char *loc_reg_starttry,
3496 const bool utf8_target)
3498 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3499 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3500 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3501 /* The part of the string before starttry has one color
3502 (pref0_len chars), between starttry and current
3503 position another one (pref_len - pref0_len chars),
3504 after the current position the third one.
3505 We assume that pref0_len <= pref_len, otherwise we
3506 decrease pref0_len. */
3507 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3508 ? (5 + taill) - l : locinput - loc_bostr;
3511 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3513 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3515 pref0_len = pref_len - (locinput - loc_reg_starttry);
3516 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3517 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3518 ? (5 + taill) - pref_len : loc_regeol - locinput);
3519 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3523 if (pref0_len > pref_len)
3524 pref0_len = pref_len;
3526 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3528 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3529 (locinput - pref_len),pref0_len, 60, 4, 5);
3531 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3532 (locinput - pref_len + pref0_len),
3533 pref_len - pref0_len, 60, 2, 3);
3535 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3536 locinput, loc_regeol - locinput, 10, 0, 1);
3538 const STRLEN tlen=len0+len1+len2;
3539 PerlIO_printf(Perl_debug_log,
3540 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3541 (IV)(locinput - loc_bostr),
3544 (docolor ? "" : "> <"),
3546 (int)(tlen > 19 ? 0 : 19 - tlen),
3553 /* reg_check_named_buff_matched()
3554 * Checks to see if a named buffer has matched. The data array of
3555 * buffer numbers corresponding to the buffer is expected to reside
3556 * in the regexp->data->data array in the slot stored in the ARG() of
3557 * node involved. Note that this routine doesn't actually care about the
3558 * name, that information is not preserved from compilation to execution.
3559 * Returns the index of the leftmost defined buffer with the given name
3560 * or 0 if non of the buffers matched.
3563 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3566 RXi_GET_DECL(rex,rexi);
3567 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3568 I32 *nums=(I32*)SvPVX(sv_dat);
3570 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3572 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3573 if ((I32)rex->lastparen >= nums[n] &&
3574 rex->offs[nums[n]].end != -1)
3584 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3585 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3587 /* This function determines if there are one or two characters that match
3588 * the first character of the passed-in EXACTish node <text_node>, and if
3589 * so, returns them in the passed-in pointers.
3591 * If it determines that no possible character in the target string can
3592 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3593 * the first character in <text_node> requires UTF-8 to represent, and the
3594 * target string isn't in UTF-8.)
3596 * If there are more than two characters that could match the beginning of
3597 * <text_node>, or if more context is required to determine a match or not,
3598 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3600 * The motiviation behind this function is to allow the caller to set up
3601 * tight loops for matching. If <text_node> is of type EXACT, there is
3602 * only one possible character that can match its first character, and so
3603 * the situation is quite simple. But things get much more complicated if
3604 * folding is involved. It may be that the first character of an EXACTFish
3605 * node doesn't participate in any possible fold, e.g., punctuation, so it
3606 * can be matched only by itself. The vast majority of characters that are
3607 * in folds match just two things, their lower and upper-case equivalents.
3608 * But not all are like that; some have multiple possible matches, or match
3609 * sequences of more than one character. This function sorts all that out.
3611 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3612 * loop of trying to match A*, we know we can't exit where the thing
3613 * following it isn't a B. And something can't be a B unless it is the
3614 * beginning of B. By putting a quick test for that beginning in a tight
3615 * loop, we can rule out things that can't possibly be B without having to
3616 * break out of the loop, thus avoiding work. Similarly, if A is a single
3617 * character, we can make a tight loop matching A*, using the outputs of
3620 * If the target string to match isn't in UTF-8, and there aren't
3621 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3622 * the one or two possible octets (which are characters in this situation)
3623 * that can match. In all cases, if there is only one character that can
3624 * match, *<c1p> and *<c2p> will be identical.
3626 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3627 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3628 * can match the beginning of <text_node>. They should be declared with at
3629 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3630 * undefined what these contain.) If one or both of the buffers are
3631 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3632 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3633 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3634 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3635 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3637 const bool utf8_target = reginfo->is_utf8_target;
3639 UV c1 = CHRTEST_NOT_A_CP_1;
3640 UV c2 = CHRTEST_NOT_A_CP_2;
3641 bool use_chrtest_void = FALSE;
3642 const bool is_utf8_pat = reginfo->is_utf8_pat;
3644 /* Used when we have both utf8 input and utf8 output, to avoid converting
3645 * to/from code points */
3646 bool utf8_has_been_setup = FALSE;
3650 U8 *pat = (U8*)STRING(text_node);
3651 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
3653 if (OP(text_node) == EXACT) {
3655 /* In an exact node, only one thing can be matched, that first
3656 * character. If both the pat and the target are UTF-8, we can just
3657 * copy the input to the output, avoiding finding the code point of
3662 else if (utf8_target) {
3663 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3664 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3665 utf8_has_been_setup = TRUE;
3668 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3671 else { /* an EXACTFish node */
3672 U8 *pat_end = pat + STR_LEN(text_node);
3674 /* An EXACTFL node has at least some characters unfolded, because what
3675 * they match is not known until now. So, now is the time to fold
3676 * the first few of them, as many as are needed to determine 'c1' and
3677 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
3678 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
3679 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
3680 * need to fold as many characters as a single character can fold to,
3681 * so that later we can check if the first ones are such a multi-char
3682 * fold. But, in such a pattern only locale-problematic characters
3683 * aren't folded, so we can skip this completely if the first character
3684 * in the node isn't one of the tricky ones */
3685 if (OP(text_node) == EXACTFL) {
3687 if (! is_utf8_pat) {
3688 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
3690 folded[0] = folded[1] = 's';
3692 pat_end = folded + 2;
3695 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
3700 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
3702 *(d++) = (U8) toFOLD_LC(*s);
3707 _to_utf8_fold_flags(s,
3710 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
3721 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
3722 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
3724 /* Multi-character folds require more context to sort out. Also
3725 * PL_utf8_foldclosures used below doesn't handle them, so have to
3726 * be handled outside this routine */
3727 use_chrtest_void = TRUE;
3729 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
3730 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
3732 /* Load the folds hash, if not already done */
3734 if (! PL_utf8_foldclosures) {
3735 _load_PL_utf8_foldclosures();
3738 /* The fold closures data structure is a hash with the keys
3739 * being the UTF-8 of every character that is folded to, like
3740 * 'k', and the values each an array of all code points that
3741 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
3742 * Multi-character folds are not included */
3743 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
3748 /* Not found in the hash, therefore there are no folds
3749 * containing it, so there is only a single character that
3753 else { /* Does participate in folds */
3754 AV* list = (AV*) *listp;
3755 if (av_tindex(list) != 1) {
3757 /* If there aren't exactly two folds to this, it is
3758 * outside the scope of this function */
3759 use_chrtest_void = TRUE;
3761 else { /* There are two. Get them */
3762 SV** c_p = av_fetch(list, 0, FALSE);
3764 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3768 c_p = av_fetch(list, 1, FALSE);
3770 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
3774 /* Folds that cross the 255/256 boundary are forbidden
3775 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
3776 * one is ASCIII. Since the pattern character is above
3777 * 255, and its only other match is below 256, the only
3778 * legal match will be to itself. We have thrown away
3779 * the original, so have to compute which is the one
3781 if ((c1 < 256) != (c2 < 256)) {
3782 if ((OP(text_node) == EXACTFL
3783 && ! IN_UTF8_CTYPE_LOCALE)
3784 || ((OP(text_node) == EXACTFA
3785 || OP(text_node) == EXACTFA_NO_TRIE)
3786 && (isASCII(c1) || isASCII(c2))))
3799 else /* Here, c1 is <= 255 */
3801 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
3802 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
3803 && ((OP(text_node) != EXACTFA
3804 && OP(text_node) != EXACTFA_NO_TRIE)
3807 /* Here, there could be something above Latin1 in the target
3808 * which folds to this character in the pattern. All such
3809 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
3810 * than two characters involved in their folds, so are outside
3811 * the scope of this function */
3812 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
3813 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
3816 use_chrtest_void = TRUE;
3819 else { /* Here nothing above Latin1 can fold to the pattern
3821 switch (OP(text_node)) {
3823 case EXACTFL: /* /l rules */
3824 c2 = PL_fold_locale[c1];
3827 case EXACTF: /* This node only generated for non-utf8
3829 assert(! is_utf8_pat);
3830 if (! utf8_target) { /* /d rules */
3835 /* /u rules for all these. This happens to work for
3836 * EXACTFA as nothing in Latin1 folds to ASCII */
3837 case EXACTFA_NO_TRIE: /* This node only generated for
3838 non-utf8 patterns */
3839 assert(! is_utf8_pat);
3844 c2 = PL_fold_latin1[c1];
3848 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
3849 assert(0); /* NOTREACHED */
3855 /* Here have figured things out. Set up the returns */
3856 if (use_chrtest_void) {
3857 *c2p = *c1p = CHRTEST_VOID;
3859 else if (utf8_target) {
3860 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
3861 uvchr_to_utf8(c1_utf8, c1);
3862 uvchr_to_utf8(c2_utf8, c2);
3865 /* Invariants are stored in both the utf8 and byte outputs; Use
3866 * negative numbers otherwise for the byte ones. Make sure that the
3867 * byte ones are the same iff the utf8 ones are the same */
3868 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
3869 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
3872 ? CHRTEST_NOT_A_CP_1
3873 : CHRTEST_NOT_A_CP_2;
3875 else if (c1 > 255) {
3876 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
3881 *c1p = *c2p = c2; /* c2 is the only representable value */
3883 else { /* c1 is representable; see about c2 */
3885 *c2p = (c2 < 256) ? c2 : c1;
3891 /* returns -1 on failure, $+[0] on success */
3893 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
3895 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3899 const bool utf8_target = reginfo->is_utf8_target;
3900 const U32 uniflags = UTF8_ALLOW_DEFAULT;
3901 REGEXP *rex_sv = reginfo->prog;
3902 regexp *rex = ReANY(rex_sv);
3903 RXi_GET_DECL(rex,rexi);
3904 /* the current state. This is a cached copy of PL_regmatch_state */
3906 /* cache heavy used fields of st in registers */
3909 U32 n = 0; /* general value; init to avoid compiler warning */
3910 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
3911 char *locinput = startpos;
3912 char *pushinput; /* where to continue after a PUSH */
3913 I32 nextchr; /* is always set to UCHARAT(locinput) */
3915 bool result = 0; /* return value of S_regmatch */
3916 int depth = 0; /* depth of backtrack stack */
3917 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
3918 const U32 max_nochange_depth =
3919 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
3920 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
3921 regmatch_state *yes_state = NULL; /* state to pop to on success of
3923 /* mark_state piggy backs on the yes_state logic so that when we unwind
3924 the stack on success we can update the mark_state as we go */
3925 regmatch_state *mark_state = NULL; /* last mark state we have seen */
3926 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
3927 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
3929 bool no_final = 0; /* prevent failure from backtracking? */
3930 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
3931 char *startpoint = locinput;
3932 SV *popmark = NULL; /* are we looking for a mark? */
3933 SV *sv_commit = NULL; /* last mark name seen in failure */
3934 SV *sv_yes_mark = NULL; /* last mark name we have seen
3935 during a successful match */
3936 U32 lastopen = 0; /* last open we saw */
3937 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
3938 SV* const oreplsv = GvSVn(PL_replgv);
3939 /* these three flags are set by various ops to signal information to
3940 * the very next op. They have a useful lifetime of exactly one loop
3941 * iteration, and are not preserved or restored by state pushes/pops
3943 bool sw = 0; /* the condition value in (?(cond)a|b) */
3944 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
3945 int logical = 0; /* the following EVAL is:
3949 or the following IFMATCH/UNLESSM is:
3950 false: plain (?=foo)
3951 true: used as a condition: (?(?=foo))
3953 PAD* last_pad = NULL;
3955 I32 gimme = G_SCALAR;
3956 CV *caller_cv = NULL; /* who called us */
3957 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
3958 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
3959 U32 maxopenparen = 0; /* max '(' index seen so far */
3960 int to_complement; /* Invert the result? */
3961 _char_class_number classnum;
3962 bool is_utf8_pat = reginfo->is_utf8_pat;
3965 GET_RE_DEBUG_FLAGS_DECL;
3968 /* protect against undef(*^R) */
3969 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
3971 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
3972 multicall_oldcatch = 0;
3973 multicall_cv = NULL;
3975 PERL_UNUSED_VAR(multicall_cop);
3976 PERL_UNUSED_VAR(newsp);
3979 PERL_ARGS_ASSERT_REGMATCH;
3981 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
3982 PerlIO_printf(Perl_debug_log,"regmatch start\n");
3985 st = PL_regmatch_state;
3987 /* Note that nextchr is a byte even in UTF */
3990 while (scan != NULL) {
3993 SV * const prop = sv_newmortal();
3994 regnode *rnext=regnext(scan);
3995 DUMP_EXEC_POS( locinput, scan, utf8_target );
3996 regprop(rex, prop, scan, reginfo);
3998 PerlIO_printf(Perl_debug_log,
3999 "%3"IVdf":%*s%s(%"IVdf")\n",
4000 (IV)(scan - rexi->program), depth*2, "",
4002 (PL_regkind[OP(scan)] == END || !rnext) ?
4003 0 : (IV)(rnext - rexi->program));
4006 next = scan + NEXT_OFF(scan);
4009 state_num = OP(scan);
4015 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
4017 switch (state_num) {
4018 case BOL: /* /^../ */
4019 case SBOL: /* /^../s */
4020 if (locinput == reginfo->strbeg)
4024 case MBOL: /* /^../m */
4025 if (locinput == reginfo->strbeg ||
4026 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
4033 if (locinput == reginfo->ganch)
4037 case KEEPS: /* \K */
4038 /* update the startpoint */
4039 st->u.keeper.val = rex->offs[0].start;
4040 rex->offs[0].start = locinput - reginfo->strbeg;
4041 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
4042 assert(0); /*NOTREACHED*/
4043 case KEEPS_next_fail:
4044 /* rollback the start point change */
4045 rex->offs[0].start = st->u.keeper.val;
4047 assert(0); /*NOTREACHED*/
4049 case MEOL: /* /..$/m */
4050 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4054 case EOL: /* /..$/ */
4056 case SEOL: /* /..$/s */
4057 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4059 if (reginfo->strend - locinput > 1)
4064 if (!NEXTCHR_IS_EOS)
4068 case SANY: /* /./s */
4071 goto increment_locinput;
4079 case REG_ANY: /* /./ */
4080 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
4082 goto increment_locinput;
4086 #define ST st->u.trie
4087 case TRIEC: /* (ab|cd) with known charclass */
4088 /* In this case the charclass data is available inline so
4089 we can fail fast without a lot of extra overhead.
4091 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
4093 PerlIO_printf(Perl_debug_log,
4094 "%*s %sfailed to match trie start class...%s\n",
4095 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4098 assert(0); /* NOTREACHED */
4101 case TRIE: /* (ab|cd) */
4102 /* the basic plan of execution of the trie is:
4103 * At the beginning, run though all the states, and
4104 * find the longest-matching word. Also remember the position
4105 * of the shortest matching word. For example, this pattern:
4108 * when matched against the string "abcde", will generate
4109 * accept states for all words except 3, with the longest
4110 * matching word being 4, and the shortest being 2 (with
4111 * the position being after char 1 of the string).
4113 * Then for each matching word, in word order (i.e. 1,2,4,5),
4114 * we run the remainder of the pattern; on each try setting
4115 * the current position to the character following the word,
4116 * returning to try the next word on failure.
4118 * We avoid having to build a list of words at runtime by
4119 * using a compile-time structure, wordinfo[].prev, which
4120 * gives, for each word, the previous accepting word (if any).
4121 * In the case above it would contain the mappings 1->2, 2->0,
4122 * 3->0, 4->5, 5->1. We can use this table to generate, from
4123 * the longest word (4 above), a list of all words, by
4124 * following the list of prev pointers; this gives us the
4125 * unordered list 4,5,1,2. Then given the current word we have
4126 * just tried, we can go through the list and find the
4127 * next-biggest word to try (so if we just failed on word 2,
4128 * the next in the list is 4).
4130 * Since at runtime we don't record the matching position in
4131 * the string for each word, we have to work that out for
4132 * each word we're about to process. The wordinfo table holds
4133 * the character length of each word; given that we recorded
4134 * at the start: the position of the shortest word and its
4135 * length in chars, we just need to move the pointer the
4136 * difference between the two char lengths. Depending on
4137 * Unicode status and folding, that's cheap or expensive.
4139 * This algorithm is optimised for the case where are only a
4140 * small number of accept states, i.e. 0,1, or maybe 2.
4141 * With lots of accepts states, and having to try all of them,
4142 * it becomes quadratic on number of accept states to find all
4147 /* what type of TRIE am I? (utf8 makes this contextual) */
4148 DECL_TRIE_TYPE(scan);
4150 /* what trie are we using right now */
4151 reg_trie_data * const trie
4152 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
4153 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
4154 U32 state = trie->startstate;
4157 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
4159 if (trie->states[ state ].wordnum) {
4161 PerlIO_printf(Perl_debug_log,
4162 "%*s %smatched empty string...%s\n",
4163 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4169 PerlIO_printf(Perl_debug_log,
4170 "%*s %sfailed to match trie start class...%s\n",
4171 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
4178 U8 *uc = ( U8* )locinput;
4182 U8 *uscan = (U8*)NULL;
4183 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
4184 U32 charcount = 0; /* how many input chars we have matched */
4185 U32 accepted = 0; /* have we seen any accepting states? */
4187 ST.jump = trie->jump;
4190 ST.longfold = FALSE; /* char longer if folded => it's harder */
4193 /* fully traverse the TRIE; note the position of the
4194 shortest accept state and the wordnum of the longest
4197 while ( state && uc <= (U8*)(reginfo->strend) ) {
4198 U32 base = trie->states[ state ].trans.base;
4202 wordnum = trie->states[ state ].wordnum;
4204 if (wordnum) { /* it's an accept state */
4207 /* record first match position */
4209 ST.firstpos = (U8*)locinput;
4214 ST.firstchars = charcount;
4217 if (!ST.nextword || wordnum < ST.nextword)
4218 ST.nextword = wordnum;
4219 ST.topword = wordnum;
4222 DEBUG_TRIE_EXECUTE_r({
4223 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
4224 PerlIO_printf( Perl_debug_log,
4225 "%*s %sState: %4"UVxf" Accepted: %c ",
4226 2+depth * 2, "", PL_colors[4],
4227 (UV)state, (accepted ? 'Y' : 'N'));
4230 /* read a char and goto next state */
4231 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
4233 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
4234 uscan, len, uvc, charid, foldlen,
4241 base + charid - 1 - trie->uniquecharcount)) >= 0)
4243 && ((U32)offset < trie->lasttrans)
4244 && trie->trans[offset].check == state)
4246 state = trie->trans[offset].next;
4257 DEBUG_TRIE_EXECUTE_r(
4258 PerlIO_printf( Perl_debug_log,
4259 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
4260 charid, uvc, (UV)state, PL_colors[5] );
4266 /* calculate total number of accept states */
4271 w = trie->wordinfo[w].prev;
4274 ST.accepted = accepted;
4278 PerlIO_printf( Perl_debug_log,
4279 "%*s %sgot %"IVdf" possible matches%s\n",
4280 REPORT_CODE_OFF + depth * 2, "",
4281 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
4283 goto trie_first_try; /* jump into the fail handler */
4285 assert(0); /* NOTREACHED */
4287 case TRIE_next_fail: /* we failed - try next alternative */
4291 REGCP_UNWIND(ST.cp);
4292 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
4294 if (!--ST.accepted) {
4296 PerlIO_printf( Perl_debug_log,
4297 "%*s %sTRIE failed...%s\n",
4298 REPORT_CODE_OFF+depth*2, "",
4305 /* Find next-highest word to process. Note that this code
4306 * is O(N^2) per trie run (O(N) per branch), so keep tight */
4309 U16 const nextword = ST.nextword;
4310 reg_trie_wordinfo * const wordinfo
4311 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
4312 for (word=ST.topword; word; word=wordinfo[word].prev) {
4313 if (word > nextword && (!min || word < min))
4326 ST.lastparen = rex->lastparen;
4327 ST.lastcloseparen = rex->lastcloseparen;
4331 /* find start char of end of current word */
4333 U32 chars; /* how many chars to skip */
4334 reg_trie_data * const trie
4335 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
4337 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
4339 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
4344 /* the hard option - fold each char in turn and find
4345 * its folded length (which may be different */
4346 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
4354 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
4362 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
4367 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
4383 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
4384 ? ST.jump[ST.nextword]
4388 PerlIO_printf( Perl_debug_log,
4389 "%*s %sTRIE matched word #%d, continuing%s\n",
4390 REPORT_CODE_OFF+depth*2, "",
4397 if (ST.accepted > 1 || has_cutgroup) {
4398 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
4399 assert(0); /* NOTREACHED */
4401 /* only one choice left - just continue */
4403 AV *const trie_words
4404 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
4405 SV ** const tmp = av_fetch( trie_words,
4407 SV *sv= tmp ? sv_newmortal() : NULL;
4409 PerlIO_printf( Perl_debug_log,
4410 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
4411 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
4413 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
4414 PL_colors[0], PL_colors[1],
4415 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
4417 : "not compiled under -Dr",
4421 locinput = (char*)uc;
4422 continue; /* execute rest of RE */
4423 assert(0); /* NOTREACHED */
4427 case EXACT: { /* /abc/ */
4428 char *s = STRING(scan);
4430 if (utf8_target != is_utf8_pat) {
4431 /* The target and the pattern have differing utf8ness. */
4433 const char * const e = s + ln;
4436 /* The target is utf8, the pattern is not utf8.
4437 * Above-Latin1 code points can't match the pattern;
4438 * invariants match exactly, and the other Latin1 ones need
4439 * to be downgraded to a single byte in order to do the
4440 * comparison. (If we could be confident that the target
4441 * is not malformed, this could be refactored to have fewer
4442 * tests by just assuming that if the first bytes match, it
4443 * is an invariant, but there are tests in the test suite
4444 * dealing with (??{...}) which violate this) */
4446 if (l >= reginfo->strend
4447 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
4451 if (UTF8_IS_INVARIANT(*(U8*)l)) {
4458 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
4468 /* The target is not utf8, the pattern is utf8. */
4470 if (l >= reginfo->strend
4471 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
4475 if (UTF8_IS_INVARIANT(*(U8*)s)) {
4482 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
4494 /* The target and the pattern have the same utf8ness. */
4495 /* Inline the first character, for speed. */
4496 if (reginfo->strend - locinput < ln
4497 || UCHARAT(s) != nextchr
4498 || (ln > 1 && memNE(s, locinput, ln)))
4507 case EXACTFL: { /* /abc/il */
4509 const U8 * fold_array;
4511 U32 fold_utf8_flags;
4513 folder = foldEQ_locale;
4514 fold_array = PL_fold_locale;
4515 fold_utf8_flags = FOLDEQ_LOCALE;
4518 case EXACTFU_SS: /* /\x{df}/iu */
4519 case EXACTFU: /* /abc/iu */
4520 folder = foldEQ_latin1;
4521 fold_array = PL_fold_latin1;
4522 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
4525 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
4527 assert(! is_utf8_pat);
4529 case EXACTFA: /* /abc/iaa */
4530 folder = foldEQ_latin1;
4531 fold_array = PL_fold_latin1;
4532 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
4535 case EXACTF: /* /abc/i This node only generated for
4536 non-utf8 patterns */
4537 assert(! is_utf8_pat);
4539 fold_array = PL_fold;
4540 fold_utf8_flags = 0;
4548 || state_num == EXACTFU_SS
4549 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
4551 /* Either target or the pattern are utf8, or has the issue where
4552 * the fold lengths may differ. */
4553 const char * const l = locinput;
4554 char *e = reginfo->strend;
4556 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
4557 l, &e, 0, utf8_target, fold_utf8_flags))
4565 /* Neither the target nor the pattern are utf8 */
4566 if (UCHARAT(s) != nextchr
4568 && UCHARAT(s) != fold_array[nextchr])
4572 if (reginfo->strend - locinput < ln)
4574 if (ln > 1 && ! folder(s, locinput, ln))
4580 /* XXX Could improve efficiency by separating these all out using a
4581 * macro or in-line function. At that point regcomp.c would no longer
4582 * have to set the FLAGS fields of these */
4583 case BOUNDL: /* /\b/l */
4584 case NBOUNDL: /* /\B/l */
4585 case BOUND: /* /\b/ */
4586 case BOUNDU: /* /\b/u */
4587 case BOUNDA: /* /\b/a */
4588 case NBOUND: /* /\B/ */
4589 case NBOUNDU: /* /\B/u */
4590 case NBOUNDA: /* /\B/a */
4591 /* was last char in word? */
4593 && FLAGS(scan) != REGEX_ASCII_RESTRICTED_CHARSET
4594 && FLAGS(scan) != REGEX_ASCII_MORE_RESTRICTED_CHARSET)
4596 if (locinput == reginfo->strbeg)
4599 const U8 * const r =
4600 reghop3((U8*)locinput, -1, (U8*)(reginfo->strbeg));
4602 ln = utf8n_to_uvchr(r, (U8*) reginfo->strend - r,
4605 if (FLAGS(scan) != REGEX_LOCALE_CHARSET) {
4606 ln = isWORDCHAR_uni(ln);
4610 LOAD_UTF8_CHARCLASS_ALNUM();
4611 n = swash_fetch(PL_utf8_swash_ptrs[_CC_WORDCHAR], (U8*)locinput,
4616 ln = isWORDCHAR_LC_uvchr(ln);
4617 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC_utf8((U8*)locinput);
4622 /* Here the string isn't utf8, or is utf8 and only ascii
4623 * characters are to match \w. In the latter case looking at
4624 * the byte just prior to the current one may be just the final
4625 * byte of a multi-byte character. This is ok. There are two
4627 * 1) it is a single byte character, and then the test is doing
4628 * just what it's supposed to.
4629 * 2) it is a multi-byte character, in which case the final
4630 * byte is never mistakable for ASCII, and so the test
4631 * will say it is not a word character, which is the
4632 * correct answer. */
4633 ln = (locinput != reginfo->strbeg) ?
4634 UCHARAT(locinput - 1) : '\n';
4635 switch (FLAGS(scan)) {
4636 case REGEX_UNICODE_CHARSET:
4637 ln = isWORDCHAR_L1(ln);
4638 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_L1(nextchr);
4640 case REGEX_LOCALE_CHARSET:
4641 ln = isWORDCHAR_LC(ln);
4642 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_LC(nextchr);
4644 case REGEX_DEPENDS_CHARSET:
4645 ln = isWORDCHAR(ln);
4646 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR(nextchr);
4648 case REGEX_ASCII_RESTRICTED_CHARSET:
4649 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
4650 ln = isWORDCHAR_A(ln);
4651 n = NEXTCHR_IS_EOS ? 0 : isWORDCHAR_A(nextchr);
4654 Perl_croak(aTHX_ "panic: Unexpected FLAGS %u in op %u", FLAGS(scan), OP(scan));
4657 /* Note requires that all BOUNDs be lower than all NBOUNDs in
4659 if (((!ln) == (!n)) == (OP(scan) < NBOUND))
4663 case ANYOF: /* /[abc]/ */
4667 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
4670 locinput += UTF8SKIP(locinput);
4673 if (!REGINCLASS(rex, scan, (U8*)locinput))
4679 /* The argument (FLAGS) to all the POSIX node types is the class number
4682 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
4686 case POSIXL: /* \w or [:punct:] etc. under /l */
4690 /* Use isFOO_lc() for characters within Latin1. (Note that
4691 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4692 * wouldn't be invariant) */
4693 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4694 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
4698 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4699 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
4700 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4701 *(locinput + 1))))))
4706 else { /* Here, must be an above Latin-1 code point */
4707 goto utf8_posix_not_eos;
4710 /* Here, must be utf8 */
4711 locinput += UTF8SKIP(locinput);
4714 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
4718 case POSIXD: /* \w or [:punct:] etc. under /d */
4724 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
4726 if (NEXTCHR_IS_EOS) {
4730 /* All UTF-8 variants match */
4731 if (! UTF8_IS_INVARIANT(nextchr)) {
4732 goto increment_locinput;
4738 case POSIXA: /* \w or [:punct:] etc. under /a */
4741 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
4742 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
4743 * character is a single byte */
4746 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
4752 /* Here we are either not in utf8, or we matched a utf8-invariant,
4753 * so the next char is the next byte */
4757 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
4761 case POSIXU: /* \w or [:punct:] etc. under /u */
4763 if (NEXTCHR_IS_EOS) {
4768 /* Use _generic_isCC() for characters within Latin1. (Note that
4769 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
4770 * wouldn't be invariant) */
4771 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
4772 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
4779 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
4780 if (! (to_complement
4781 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
4789 else { /* Handle above Latin-1 code points */
4790 classnum = (_char_class_number) FLAGS(scan);
4791 if (classnum < _FIRST_NON_SWASH_CC) {
4793 /* Here, uses a swash to find such code points. Load if if
4794 * not done already */
4795 if (! PL_utf8_swash_ptrs[classnum]) {
4796 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
4797 PL_utf8_swash_ptrs[classnum]
4798 = _core_swash_init("utf8",
4801 PL_XPosix_ptrs[classnum], &flags);
4803 if (! (to_complement
4804 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
4805 (U8 *) locinput, TRUE))))
4810 else { /* Here, uses macros to find above Latin-1 code points */
4812 case _CC_ENUM_SPACE: /* XXX would require separate
4813 code if we revert the change
4814 of \v matching this */
4815 case _CC_ENUM_PSXSPC:
4816 if (! (to_complement
4817 ^ cBOOL(is_XPERLSPACE_high(locinput))))
4822 case _CC_ENUM_BLANK:
4823 if (! (to_complement
4824 ^ cBOOL(is_HORIZWS_high(locinput))))
4829 case _CC_ENUM_XDIGIT:
4830 if (! (to_complement
4831 ^ cBOOL(is_XDIGIT_high(locinput))))
4836 case _CC_ENUM_VERTSPACE:
4837 if (! (to_complement
4838 ^ cBOOL(is_VERTWS_high(locinput))))
4843 default: /* The rest, e.g. [:cntrl:], can't match
4845 if (! to_complement) {
4851 locinput += UTF8SKIP(locinput);
4855 case CLUMP: /* Match \X: logical Unicode character. This is defined as
4856 a Unicode extended Grapheme Cluster */
4857 /* From http://www.unicode.org/reports/tr29 (5.2 version). An
4858 extended Grapheme Cluster is:
4861 | Prepend* Begin Extend*
4864 Begin is: ( Special_Begin | ! Control )
4865 Special_Begin is: ( Regional-Indicator+ | Hangul-syllable )
4866 Extend is: ( Grapheme_Extend | Spacing_Mark )
4867 Control is: [ GCB_Control | CR | LF ]
4868 Hangul-syllable is: ( T+ | ( L* ( L | ( LVT | ( V | LV ) V* ) T* ) ))
4870 If we create a 'Regular_Begin' = Begin - Special_Begin, then
4873 Begin is ( Regular_Begin + Special Begin )
4875 It turns out that 98.4% of all Unicode code points match
4876 Regular_Begin. Doing it this way eliminates a table match in
4877 the previous implementation for almost all Unicode code points.
4879 There is a subtlety with Prepend* which showed up in testing.
4880 Note that the Begin, and only the Begin is required in:
4881 | Prepend* Begin Extend*
4882 Also, Begin contains '! Control'. A Prepend must be a
4883 '! Control', which means it must also be a Begin. What it
4884 comes down to is that if we match Prepend* and then find no
4885 suitable Begin afterwards, that if we backtrack the last
4886 Prepend, that one will be a suitable Begin.
4891 if (! utf8_target) {
4893 /* Match either CR LF or '.', as all the other possibilities
4895 locinput++; /* Match the . or CR */
4896 if (nextchr == '\r' /* And if it was CR, and the next is LF,
4898 && locinput < reginfo->strend
4899 && UCHARAT(locinput) == '\n')
4906 /* Utf8: See if is ( CR LF ); already know that locinput <
4907 * reginfo->strend, so locinput+1 is in bounds */
4908 if ( nextchr == '\r' && locinput+1 < reginfo->strend
4909 && UCHARAT(locinput + 1) == '\n')
4916 /* In case have to backtrack to beginning, then match '.' */
4917 char *starting = locinput;
4919 /* In case have to backtrack the last prepend */
4920 char *previous_prepend = NULL;
4922 LOAD_UTF8_CHARCLASS_GCB();
4924 /* Match (prepend)* */
4925 while (locinput < reginfo->strend
4926 && (len = is_GCB_Prepend_utf8(locinput)))
4928 previous_prepend = locinput;
4932 /* As noted above, if we matched a prepend character, but
4933 * the next thing won't match, back off the last prepend we
4934 * matched, as it is guaranteed to match the begin */
4935 if (previous_prepend
4936 && (locinput >= reginfo->strend
4937 || (! swash_fetch(PL_utf8_X_regular_begin,
4938 (U8*)locinput, utf8_target)
4939 && ! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)))
4942 locinput = previous_prepend;
4945 /* Note that here we know reginfo->strend > locinput, as we
4946 * tested that upon input to this switch case, and if we
4947 * moved locinput forward, we tested the result just above
4948 * and it either passed, or we backed off so that it will
4950 if (swash_fetch(PL_utf8_X_regular_begin,
4951 (U8*)locinput, utf8_target)) {
4952 locinput += UTF8SKIP(locinput);
4954 else if (! is_GCB_SPECIAL_BEGIN_START_utf8(locinput)) {
4956 /* Here did not match the required 'Begin' in the
4957 * second term. So just match the very first
4958 * character, the '.' of the final term of the regex */
4959 locinput = starting + UTF8SKIP(starting);
4963 /* Here is a special begin. It can be composed of
4964 * several individual characters. One possibility is
4966 if ((len = is_GCB_RI_utf8(locinput))) {
4968 while (locinput < reginfo->strend
4969 && (len = is_GCB_RI_utf8(locinput)))
4973 } else if ((len = is_GCB_T_utf8(locinput))) {
4974 /* Another possibility is T+ */
4976 while (locinput < reginfo->strend
4977 && (len = is_GCB_T_utf8(locinput)))
4983 /* Here, neither RI+ nor T+; must be some other
4984 * Hangul. That means it is one of the others: L,
4985 * LV, LVT or V, and matches:
4986 * L* (L | LVT T* | V * V* T* | LV V* T*) */
4989 while (locinput < reginfo->strend
4990 && (len = is_GCB_L_utf8(locinput)))
4995 /* Here, have exhausted L*. If the next character
4996 * is not an LV, LVT nor V, it means we had to have
4997 * at least one L, so matches L+ in the original
4998 * equation, we have a complete hangul syllable.
5001 if (locinput < reginfo->strend
5002 && is_GCB_LV_LVT_V_utf8(locinput))
5004 /* Otherwise keep going. Must be LV, LVT or V.
5005 * See if LVT, by first ruling out V, then LV */
5006 if (! is_GCB_V_utf8(locinput)
5007 /* All but every TCount one is LV */
5008 && (valid_utf8_to_uvchr((U8 *) locinput,
5013 locinput += UTF8SKIP(locinput);
5016 /* Must be V or LV. Take it, then match
5018 locinput += UTF8SKIP(locinput);
5019 while (locinput < reginfo->strend
5020 && (len = is_GCB_V_utf8(locinput)))
5026 /* And any of LV, LVT, or V can be followed
5028 while (locinput < reginfo->strend
5029 && (len = is_GCB_T_utf8(locinput)))
5037 /* Match any extender */
5038 while (locinput < reginfo->strend
5039 && swash_fetch(PL_utf8_X_extend,
5040 (U8*)locinput, utf8_target))
5042 locinput += UTF8SKIP(locinput);
5046 if (locinput > reginfo->strend) sayNO;
5050 case NREFFL: /* /\g{name}/il */
5051 { /* The capture buffer cases. The ones beginning with N for the
5052 named buffers just convert to the equivalent numbered and
5053 pretend they were called as the corresponding numbered buffer
5055 /* don't initialize these in the declaration, it makes C++
5060 const U8 *fold_array;
5063 folder = foldEQ_locale;
5064 fold_array = PL_fold_locale;
5066 utf8_fold_flags = FOLDEQ_LOCALE;
5069 case NREFFA: /* /\g{name}/iaa */
5070 folder = foldEQ_latin1;
5071 fold_array = PL_fold_latin1;
5073 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5076 case NREFFU: /* /\g{name}/iu */
5077 folder = foldEQ_latin1;
5078 fold_array = PL_fold_latin1;
5080 utf8_fold_flags = 0;
5083 case NREFF: /* /\g{name}/i */
5085 fold_array = PL_fold;
5087 utf8_fold_flags = 0;
5090 case NREF: /* /\g{name}/ */
5094 utf8_fold_flags = 0;
5097 /* For the named back references, find the corresponding buffer
5099 n = reg_check_named_buff_matched(rex,scan);
5104 goto do_nref_ref_common;
5106 case REFFL: /* /\1/il */
5107 folder = foldEQ_locale;
5108 fold_array = PL_fold_locale;
5109 utf8_fold_flags = FOLDEQ_LOCALE;
5112 case REFFA: /* /\1/iaa */
5113 folder = foldEQ_latin1;
5114 fold_array = PL_fold_latin1;
5115 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5118 case REFFU: /* /\1/iu */
5119 folder = foldEQ_latin1;
5120 fold_array = PL_fold_latin1;
5121 utf8_fold_flags = 0;
5124 case REFF: /* /\1/i */
5126 fold_array = PL_fold;
5127 utf8_fold_flags = 0;
5130 case REF: /* /\1/ */
5133 utf8_fold_flags = 0;
5137 n = ARG(scan); /* which paren pair */
5140 ln = rex->offs[n].start;
5141 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5142 if (rex->lastparen < n || ln == -1)
5143 sayNO; /* Do not match unless seen CLOSEn. */
5144 if (ln == rex->offs[n].end)
5147 s = reginfo->strbeg + ln;
5148 if (type != REF /* REF can do byte comparison */
5149 && (utf8_target || type == REFFU || type == REFFL))
5151 char * limit = reginfo->strend;
5153 /* This call case insensitively compares the entire buffer
5154 * at s, with the current input starting at locinput, but
5155 * not going off the end given by reginfo->strend, and
5156 * returns in <limit> upon success, how much of the
5157 * current input was matched */
5158 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
5159 locinput, &limit, 0, utf8_target, utf8_fold_flags))
5167 /* Not utf8: Inline the first character, for speed. */
5168 if (!NEXTCHR_IS_EOS &&
5169 UCHARAT(s) != nextchr &&
5171 UCHARAT(s) != fold_array[nextchr]))
5173 ln = rex->offs[n].end - ln;
5174 if (locinput + ln > reginfo->strend)
5176 if (ln > 1 && (type == REF
5177 ? memNE(s, locinput, ln)
5178 : ! folder(s, locinput, ln)))
5184 case NOTHING: /* null op; e.g. the 'nothing' following
5185 * the '*' in m{(a+|b)*}' */
5187 case TAIL: /* placeholder while compiling (A|B|C) */
5190 case BACK: /* ??? doesn't appear to be used ??? */
5194 #define ST st->u.eval
5199 regexp_internal *rei;
5200 regnode *startpoint;
5202 case GOSTART: /* (?R) */
5203 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
5204 if (cur_eval && cur_eval->locinput==locinput) {
5205 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
5206 Perl_croak(aTHX_ "Infinite recursion in regex");
5207 if ( ++nochange_depth > max_nochange_depth )
5209 "Pattern subroutine nesting without pos change"
5210 " exceeded limit in regex");
5217 if (OP(scan)==GOSUB) {
5218 startpoint = scan + ARG2L(scan);
5219 ST.close_paren = ARG(scan);
5221 startpoint = rei->program+1;
5225 /* Save all the positions seen so far. */
5226 ST.cp = regcppush(rex, 0, maxopenparen);
5227 REGCP_SET(ST.lastcp);
5229 /* and then jump to the code we share with EVAL */
5230 goto eval_recurse_doit;
5232 assert(0); /* NOTREACHED */
5234 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
5235 if (cur_eval && cur_eval->locinput==locinput) {
5236 if ( ++nochange_depth > max_nochange_depth )
5237 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
5242 /* execute the code in the {...} */
5246 OP * const oop = PL_op;
5247 COP * const ocurcop = PL_curcop;
5251 /* save *all* paren positions */
5252 regcppush(rex, 0, maxopenparen);
5253 REGCP_SET(runops_cp);
5256 caller_cv = find_runcv(NULL);
5260 if (rexi->data->what[n] == 'r') { /* code from an external qr */
5262 (REGEXP*)(rexi->data->data[n])
5265 nop = (OP*)rexi->data->data[n+1];
5267 else if (rexi->data->what[n] == 'l') { /* literal code */
5269 nop = (OP*)rexi->data->data[n];
5270 assert(CvDEPTH(newcv));
5273 /* literal with own CV */
5274 assert(rexi->data->what[n] == 'L');
5275 newcv = rex->qr_anoncv;
5276 nop = (OP*)rexi->data->data[n];
5279 /* normally if we're about to execute code from the same
5280 * CV that we used previously, we just use the existing
5281 * CX stack entry. However, its possible that in the
5282 * meantime we may have backtracked, popped from the save
5283 * stack, and undone the SAVECOMPPAD(s) associated with
5284 * PUSH_MULTICALL; in which case PL_comppad no longer
5285 * points to newcv's pad. */
5286 if (newcv != last_pushed_cv || PL_comppad != last_pad)
5288 U8 flags = (CXp_SUB_RE |
5289 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
5290 if (last_pushed_cv) {
5291 CHANGE_MULTICALL_FLAGS(newcv, flags);
5294 PUSH_MULTICALL_FLAGS(newcv, flags);
5296 last_pushed_cv = newcv;
5299 /* these assignments are just to silence compiler
5301 multicall_cop = NULL;
5304 last_pad = PL_comppad;
5306 /* the initial nextstate you would normally execute
5307 * at the start of an eval (which would cause error
5308 * messages to come from the eval), may be optimised
5309 * away from the execution path in the regex code blocks;
5310 * so manually set PL_curcop to it initially */
5312 OP *o = cUNOPx(nop)->op_first;
5313 assert(o->op_type == OP_NULL);
5314 if (o->op_targ == OP_SCOPE) {
5315 o = cUNOPo->op_first;
5318 assert(o->op_targ == OP_LEAVE);
5319 o = cUNOPo->op_first;
5320 assert(o->op_type == OP_ENTER);
5324 if (o->op_type != OP_STUB) {
5325 assert( o->op_type == OP_NEXTSTATE
5326 || o->op_type == OP_DBSTATE
5327 || (o->op_type == OP_NULL
5328 && ( o->op_targ == OP_NEXTSTATE
5329 || o->op_targ == OP_DBSTATE
5333 PL_curcop = (COP*)o;
5338 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
5339 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
5341 rex->offs[0].end = locinput - reginfo->strbeg;
5342 if (reginfo->info_aux_eval->pos_magic)
5343 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
5344 reginfo->sv, reginfo->strbeg,
5345 locinput - reginfo->strbeg);
5348 SV *sv_mrk = get_sv("REGMARK", 1);
5349 sv_setsv(sv_mrk, sv_yes_mark);
5352 /* we don't use MULTICALL here as we want to call the
5353 * first op of the block of interest, rather than the
5354 * first op of the sub */
5355 before = (IV)(SP-PL_stack_base);
5357 CALLRUNOPS(aTHX); /* Scalar context. */
5359 if ((IV)(SP-PL_stack_base) == before)
5360 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
5366 /* before restoring everything, evaluate the returned
5367 * value, so that 'uninit' warnings don't use the wrong
5368 * PL_op or pad. Also need to process any magic vars
5369 * (e.g. $1) *before* parentheses are restored */
5374 if (logical == 0) /* (?{})/ */
5375 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
5376 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
5377 sw = cBOOL(SvTRUE(ret));
5380 else { /* /(??{}) */
5381 /* if its overloaded, let the regex compiler handle
5382 * it; otherwise extract regex, or stringify */
5383 if (SvGMAGICAL(ret))
5384 ret = sv_mortalcopy(ret);
5385 if (!SvAMAGIC(ret)) {
5389 if (SvTYPE(sv) == SVt_REGEXP)
5390 re_sv = (REGEXP*) sv;
5391 else if (SvSMAGICAL(ret)) {
5392 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
5394 re_sv = (REGEXP *) mg->mg_obj;
5397 /* force any undef warnings here */
5398 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
5399 ret = sv_mortalcopy(ret);
5400 (void) SvPV_force_nolen(ret);
5406 /* *** Note that at this point we don't restore
5407 * PL_comppad, (or pop the CxSUB) on the assumption it may
5408 * be used again soon. This is safe as long as nothing
5409 * in the regexp code uses the pad ! */
5411 PL_curcop = ocurcop;
5412 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
5413 PL_curpm = PL_reg_curpm;
5419 /* only /(??{})/ from now on */
5422 /* extract RE object from returned value; compiling if
5426 re_sv = reg_temp_copy(NULL, re_sv);
5431 if (SvUTF8(ret) && IN_BYTES) {
5432 /* In use 'bytes': make a copy of the octet
5433 * sequence, but without the flag on */
5435 const char *const p = SvPV(ret, len);
5436 ret = newSVpvn_flags(p, len, SVs_TEMP);
5438 if (rex->intflags & PREGf_USE_RE_EVAL)
5439 pm_flags |= PMf_USE_RE_EVAL;
5441 /* if we got here, it should be an engine which
5442 * supports compiling code blocks and stuff */
5443 assert(rex->engine && rex->engine->op_comp);
5444 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
5445 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
5446 rex->engine, NULL, NULL,
5447 /* copy /msix etc to inner pattern */
5452 & (SVs_TEMP | SVs_GMG | SVf_ROK))
5453 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
5454 /* This isn't a first class regexp. Instead, it's
5455 caching a regexp onto an existing, Perl visible
5457 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
5463 RXp_MATCH_COPIED_off(re);
5464 re->subbeg = rex->subbeg;
5465 re->sublen = rex->sublen;
5466 re->suboffset = rex->suboffset;
5467 re->subcoffset = rex->subcoffset;
5469 re->lastcloseparen = 0;
5472 debug_start_match(re_sv, utf8_target, locinput,
5473 reginfo->strend, "Matching embedded");
5475 startpoint = rei->program + 1;
5476 ST.close_paren = 0; /* only used for GOSUB */
5477 /* Save all the seen positions so far. */
5478 ST.cp = regcppush(rex, 0, maxopenparen);
5479 REGCP_SET(ST.lastcp);
5480 /* and set maxopenparen to 0, since we are starting a "fresh" match */
5482 /* run the pattern returned from (??{...}) */
5484 eval_recurse_doit: /* Share code with GOSUB below this line
5485 * At this point we expect the stack context to be
5486 * set up correctly */
5488 /* invalidate the S-L poscache. We're now executing a
5489 * different set of WHILEM ops (and their associated
5490 * indexes) against the same string, so the bits in the
5491 * cache are meaningless. Setting maxiter to zero forces
5492 * the cache to be invalidated and zeroed before reuse.
5493 * XXX This is too dramatic a measure. Ideally we should
5494 * save the old cache and restore when running the outer
5496 reginfo->poscache_maxiter = 0;
5498 /* the new regexp might have a different is_utf8_pat than we do */
5499 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
5501 ST.prev_rex = rex_sv;
5502 ST.prev_curlyx = cur_curlyx;
5504 SET_reg_curpm(rex_sv);
5509 ST.prev_eval = cur_eval;
5511 /* now continue from first node in postoned RE */
5512 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
5513 assert(0); /* NOTREACHED */
5516 case EVAL_AB: /* cleanup after a successful (??{A})B */
5517 /* note: this is called twice; first after popping B, then A */
5518 rex_sv = ST.prev_rex;
5519 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5520 SET_reg_curpm(rex_sv);
5521 rex = ReANY(rex_sv);
5522 rexi = RXi_GET(rex);
5524 /* preserve $^R across LEAVE's. See Bug 121070. */
5525 SV *save_sv= GvSV(PL_replgv);
5526 SvREFCNT_inc(save_sv);
5527 regcpblow(ST.cp); /* LEAVE in disguise */
5528 sv_setsv(GvSV(PL_replgv), save_sv);
5529 SvREFCNT_dec(save_sv);
5531 cur_eval = ST.prev_eval;
5532 cur_curlyx = ST.prev_curlyx;
5534 /* Invalidate cache. See "invalidate" comment above. */
5535 reginfo->poscache_maxiter = 0;
5536 if ( nochange_depth )
5541 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
5542 /* note: this is called twice; first after popping B, then A */
5543 rex_sv = ST.prev_rex;
5544 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
5545 SET_reg_curpm(rex_sv);
5546 rex = ReANY(rex_sv);
5547 rexi = RXi_GET(rex);
5549 REGCP_UNWIND(ST.lastcp);
5550 regcppop(rex, &maxopenparen);
5551 cur_eval = ST.prev_eval;
5552 cur_curlyx = ST.prev_curlyx;
5553 /* Invalidate cache. See "invalidate" comment above. */
5554 reginfo->poscache_maxiter = 0;
5555 if ( nochange_depth )
5561 n = ARG(scan); /* which paren pair */
5562 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
5563 if (n > maxopenparen)
5565 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
5566 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
5570 (IV)rex->offs[n].start_tmp,
5576 /* XXX really need to log other places start/end are set too */
5577 #define CLOSE_CAPTURE \
5578 rex->offs[n].start = rex->offs[n].start_tmp; \
5579 rex->offs[n].end = locinput - reginfo->strbeg; \
5580 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
5581 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
5583 PTR2UV(rex->offs), \
5585 (IV)rex->offs[n].start, \
5586 (IV)rex->offs[n].end \
5590 n = ARG(scan); /* which paren pair */
5592 if (n > rex->lastparen)
5594 rex->lastcloseparen = n;
5595 if (cur_eval && cur_eval->u.eval.close_paren == n) {
5600 case ACCEPT: /* (*ACCEPT) */
5604 cursor && OP(cursor)!=END;
5605 cursor=regnext(cursor))
5607 if ( OP(cursor)==CLOSE ){
5609 if ( n <= lastopen ) {
5611 if (n > rex->lastparen)
5613 rex->lastcloseparen = n;
5614 if ( n == ARG(scan) || (cur_eval &&
5615 cur_eval->u.eval.close_paren == n))
5624 case GROUPP: /* (?(1)) */
5625 n = ARG(scan); /* which paren pair */
5626 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
5629 case NGROUPP: /* (?(<name>)) */
5630 /* reg_check_named_buff_matched returns 0 for no match */
5631 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
5634 case INSUBP: /* (?(R)) */
5636 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
5639 case DEFINEP: /* (?(DEFINE)) */
5643 case IFTHEN: /* (?(cond)A|B) */
5644 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
5646 next = NEXTOPER(NEXTOPER(scan));
5648 next = scan + ARG(scan);
5649 if (OP(next) == IFTHEN) /* Fake one. */
5650 next = NEXTOPER(NEXTOPER(next));
5654 case LOGICAL: /* modifier for EVAL and IFMATCH */
5655 logical = scan->flags;
5658 /*******************************************************************
5660 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
5661 pattern, where A and B are subpatterns. (For simple A, CURLYM or
5662 STAR/PLUS/CURLY/CURLYN are used instead.)
5664 A*B is compiled as <CURLYX><A><WHILEM><B>
5666 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
5667 state, which contains the current count, initialised to -1. It also sets
5668 cur_curlyx to point to this state, with any previous value saved in the
5671 CURLYX then jumps straight to the WHILEM op, rather than executing A,
5672 since the pattern may possibly match zero times (i.e. it's a while {} loop
5673 rather than a do {} while loop).
5675 Each entry to WHILEM represents a successful match of A. The count in the
5676 CURLYX block is incremented, another WHILEM state is pushed, and execution
5677 passes to A or B depending on greediness and the current count.
5679 For example, if matching against the string a1a2a3b (where the aN are
5680 substrings that match /A/), then the match progresses as follows: (the
5681 pushed states are interspersed with the bits of strings matched so far):
5684 <CURLYX cnt=0><WHILEM>
5685 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
5686 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
5687 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
5688 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
5690 (Contrast this with something like CURLYM, which maintains only a single
5694 a1 <CURLYM cnt=1> a2
5695 a1 a2 <CURLYM cnt=2> a3
5696 a1 a2 a3 <CURLYM cnt=3> b
5699 Each WHILEM state block marks a point to backtrack to upon partial failure
5700 of A or B, and also contains some minor state data related to that
5701 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
5702 overall state, such as the count, and pointers to the A and B ops.
5704 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
5705 must always point to the *current* CURLYX block, the rules are:
5707 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
5708 and set cur_curlyx to point the new block.
5710 When popping the CURLYX block after a successful or unsuccessful match,
5711 restore the previous cur_curlyx.
5713 When WHILEM is about to execute B, save the current cur_curlyx, and set it
5714 to the outer one saved in the CURLYX block.
5716 When popping the WHILEM block after a successful or unsuccessful B match,
5717 restore the previous cur_curlyx.
5719 Here's an example for the pattern (AI* BI)*BO
5720 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
5723 curlyx backtrack stack
5724 ------ ---------------
5726 CO <CO prev=NULL> <WO>
5727 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5728 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5729 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
5731 At this point the pattern succeeds, and we work back down the stack to
5732 clean up, restoring as we go:
5734 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
5735 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
5736 CO <CO prev=NULL> <WO>
5739 *******************************************************************/
5741 #define ST st->u.curlyx
5743 case CURLYX: /* start of /A*B/ (for complex A) */
5745 /* No need to save/restore up to this paren */
5746 I32 parenfloor = scan->flags;
5748 assert(next); /* keep Coverity happy */
5749 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
5752 /* XXXX Probably it is better to teach regpush to support
5753 parenfloor > maxopenparen ... */
5754 if (parenfloor > (I32)rex->lastparen)
5755 parenfloor = rex->lastparen; /* Pessimization... */
5757 ST.prev_curlyx= cur_curlyx;
5759 ST.cp = PL_savestack_ix;
5761 /* these fields contain the state of the current curly.
5762 * they are accessed by subsequent WHILEMs */
5763 ST.parenfloor = parenfloor;
5768 ST.count = -1; /* this will be updated by WHILEM */
5769 ST.lastloc = NULL; /* this will be updated by WHILEM */
5771 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
5772 assert(0); /* NOTREACHED */
5775 case CURLYX_end: /* just finished matching all of A*B */
5776 cur_curlyx = ST.prev_curlyx;
5778 assert(0); /* NOTREACHED */
5780 case CURLYX_end_fail: /* just failed to match all of A*B */
5782 cur_curlyx = ST.prev_curlyx;
5784 assert(0); /* NOTREACHED */
5788 #define ST st->u.whilem
5790 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
5792 /* see the discussion above about CURLYX/WHILEM */
5797 assert(cur_curlyx); /* keep Coverity happy */
5799 min = ARG1(cur_curlyx->u.curlyx.me);
5800 max = ARG2(cur_curlyx->u.curlyx.me);
5801 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
5802 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
5803 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
5804 ST.cache_offset = 0;
5808 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5809 "%*s whilem: matched %ld out of %d..%d\n",
5810 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
5813 /* First just match a string of min A's. */
5816 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5818 cur_curlyx->u.curlyx.lastloc = locinput;
5819 REGCP_SET(ST.lastcp);
5821 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
5822 assert(0); /* NOTREACHED */
5825 /* If degenerate A matches "", assume A done. */
5827 if (locinput == cur_curlyx->u.curlyx.lastloc) {
5828 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5829 "%*s whilem: empty match detected, trying continuation...\n",
5830 REPORT_CODE_OFF+depth*2, "")
5832 goto do_whilem_B_max;
5835 /* super-linear cache processing.
5837 * The idea here is that for certain types of CURLYX/WHILEM -
5838 * principally those whose upper bound is infinity (and
5839 * excluding regexes that have things like \1 and other very
5840 * non-regular expresssiony things), then if a pattern like
5841 * /....A*.../ fails and we backtrack to the WHILEM, then we
5842 * make a note that this particular WHILEM op was at string
5843 * position 47 (say) when the rest of pattern failed. Then, if
5844 * we ever find ourselves back at that WHILEM, and at string
5845 * position 47 again, we can just fail immediately rather than
5846 * running the rest of the pattern again.
5848 * This is very handy when patterns start to go
5849 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
5850 * with a combinatorial explosion of backtracking.
5852 * The cache is implemented as a bit array, with one bit per
5853 * string byte position per WHILEM op (up to 16) - so its
5854 * between 0.25 and 2x the string size.
5856 * To avoid allocating a poscache buffer every time, we do an
5857 * initially countdown; only after we have executed a WHILEM
5858 * op (string-length x #WHILEMs) times do we allocate the
5861 * The top 4 bits of scan->flags byte say how many different
5862 * relevant CURLLYX/WHILEM op pairs there are, while the
5863 * bottom 4-bits is the identifying index number of this
5869 if (!reginfo->poscache_maxiter) {
5870 /* start the countdown: Postpone detection until we
5871 * know the match is not *that* much linear. */
5872 reginfo->poscache_maxiter
5873 = (reginfo->strend - reginfo->strbeg + 1)
5875 /* possible overflow for long strings and many CURLYX's */
5876 if (reginfo->poscache_maxiter < 0)
5877 reginfo->poscache_maxiter = I32_MAX;
5878 reginfo->poscache_iter = reginfo->poscache_maxiter;
5881 if (reginfo->poscache_iter-- == 0) {
5882 /* initialise cache */
5883 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
5884 regmatch_info_aux *const aux = reginfo->info_aux;
5885 if (aux->poscache) {
5886 if ((SSize_t)reginfo->poscache_size < size) {
5887 Renew(aux->poscache, size, char);
5888 reginfo->poscache_size = size;
5890 Zero(aux->poscache, size, char);
5893 reginfo->poscache_size = size;
5894 Newxz(aux->poscache, size, char);
5896 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5897 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
5898 PL_colors[4], PL_colors[5])
5902 if (reginfo->poscache_iter < 0) {
5903 /* have we already failed at this position? */
5904 SSize_t offset, mask;
5906 reginfo->poscache_iter = -1; /* stop eventual underflow */
5907 offset = (scan->flags & 0xf) - 1
5908 + (locinput - reginfo->strbeg)
5910 mask = 1 << (offset % 8);
5912 if (reginfo->info_aux->poscache[offset] & mask) {
5913 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
5914 "%*s whilem: (cache) already tried at this position...\n",
5915 REPORT_CODE_OFF+depth*2, "")
5917 sayNO; /* cache records failure */
5919 ST.cache_offset = offset;
5920 ST.cache_mask = mask;
5924 /* Prefer B over A for minimal matching. */
5926 if (cur_curlyx->u.curlyx.minmod) {
5927 ST.save_curlyx = cur_curlyx;
5928 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5929 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
5931 REGCP_SET(ST.lastcp);
5932 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
5934 assert(0); /* NOTREACHED */
5937 /* Prefer A over B for maximal matching. */
5939 if (n < max) { /* More greed allowed? */
5940 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
5942 cur_curlyx->u.curlyx.lastloc = locinput;
5943 REGCP_SET(ST.lastcp);
5944 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
5945 assert(0); /* NOTREACHED */
5947 goto do_whilem_B_max;
5949 assert(0); /* NOTREACHED */
5951 case WHILEM_B_min: /* just matched B in a minimal match */
5952 case WHILEM_B_max: /* just matched B in a maximal match */
5953 cur_curlyx = ST.save_curlyx;
5955 assert(0); /* NOTREACHED */
5957 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
5958 cur_curlyx = ST.save_curlyx;
5959 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5960 cur_curlyx->u.curlyx.count--;
5962 assert(0); /* NOTREACHED */
5964 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
5966 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
5967 REGCP_UNWIND(ST.lastcp);
5968 regcppop(rex, &maxopenparen);
5969 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
5970 cur_curlyx->u.curlyx.count--;
5972 assert(0); /* NOTREACHED */
5974 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
5975 REGCP_UNWIND(ST.lastcp);
5976 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
5977 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
5978 "%*s whilem: failed, trying continuation...\n",
5979 REPORT_CODE_OFF+depth*2, "")
5982 if (cur_curlyx->u.curlyx.count >= REG_INFTY
5983 && ckWARN(WARN_REGEXP)
5984 && !reginfo->warned)
5986 reginfo->warned = TRUE;
5987 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
5988 "Complex regular subexpression recursion limit (%d) "
5994 ST.save_curlyx = cur_curlyx;
5995 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
5996 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
5998 assert(0); /* NOTREACHED */
6000 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
6001 cur_curlyx = ST.save_curlyx;
6002 REGCP_UNWIND(ST.lastcp);
6003 regcppop(rex, &maxopenparen);
6005 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
6006 /* Maximum greed exceeded */
6007 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6008 && ckWARN(WARN_REGEXP)
6009 && !reginfo->warned)
6011 reginfo->warned = TRUE;
6012 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6013 "Complex regular subexpression recursion "
6014 "limit (%d) exceeded",
6017 cur_curlyx->u.curlyx.count--;
6021 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6022 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
6024 /* Try grabbing another A and see if it helps. */
6025 cur_curlyx->u.curlyx.lastloc = locinput;
6026 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6028 REGCP_SET(ST.lastcp);
6029 PUSH_STATE_GOTO(WHILEM_A_min,
6030 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
6032 assert(0); /* NOTREACHED */
6035 #define ST st->u.branch
6037 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
6038 next = scan + ARG(scan);
6041 scan = NEXTOPER(scan);
6044 case BRANCH: /* /(...|A|...)/ */
6045 scan = NEXTOPER(scan); /* scan now points to inner node */
6046 ST.lastparen = rex->lastparen;
6047 ST.lastcloseparen = rex->lastcloseparen;
6048 ST.next_branch = next;
6051 /* Now go into the branch */
6053 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
6055 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
6057 assert(0); /* NOTREACHED */
6059 case CUTGROUP: /* /(*THEN)/ */
6060 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
6061 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6062 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
6063 assert(0); /* NOTREACHED */
6065 case CUTGROUP_next_fail:
6068 if (st->u.mark.mark_name)
6069 sv_commit = st->u.mark.mark_name;
6071 assert(0); /* NOTREACHED */
6075 assert(0); /* NOTREACHED */
6077 case BRANCH_next_fail: /* that branch failed; try the next, if any */
6082 REGCP_UNWIND(ST.cp);
6083 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6084 scan = ST.next_branch;
6085 /* no more branches? */
6086 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
6088 PerlIO_printf( Perl_debug_log,
6089 "%*s %sBRANCH failed...%s\n",
6090 REPORT_CODE_OFF+depth*2, "",
6096 continue; /* execute next BRANCH[J] op */
6097 assert(0); /* NOTREACHED */
6099 case MINMOD: /* next op will be non-greedy, e.g. A*? */
6104 #define ST st->u.curlym
6106 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
6108 /* This is an optimisation of CURLYX that enables us to push
6109 * only a single backtracking state, no matter how many matches
6110 * there are in {m,n}. It relies on the pattern being constant
6111 * length, with no parens to influence future backrefs
6115 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6117 ST.lastparen = rex->lastparen;
6118 ST.lastcloseparen = rex->lastcloseparen;
6120 /* if paren positive, emulate an OPEN/CLOSE around A */
6122 U32 paren = ST.me->flags;
6123 if (paren > maxopenparen)
6124 maxopenparen = paren;
6125 scan += NEXT_OFF(scan); /* Skip former OPEN. */
6133 ST.c1 = CHRTEST_UNINIT;
6136 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
6139 curlym_do_A: /* execute the A in /A{m,n}B/ */
6140 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
6141 assert(0); /* NOTREACHED */
6143 case CURLYM_A: /* we've just matched an A */
6145 /* after first match, determine A's length: u.curlym.alen */
6146 if (ST.count == 1) {
6147 if (reginfo->is_utf8_target) {
6148 char *s = st->locinput;
6149 while (s < locinput) {
6155 ST.alen = locinput - st->locinput;
6158 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
6161 PerlIO_printf(Perl_debug_log,
6162 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
6163 (int)(REPORT_CODE_OFF+(depth*2)), "",
6164 (IV) ST.count, (IV)ST.alen)
6167 if (cur_eval && cur_eval->u.eval.close_paren &&
6168 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6172 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
6173 if ( max == REG_INFTY || ST.count < max )
6174 goto curlym_do_A; /* try to match another A */
6176 goto curlym_do_B; /* try to match B */
6178 case CURLYM_A_fail: /* just failed to match an A */
6179 REGCP_UNWIND(ST.cp);
6181 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
6182 || (cur_eval && cur_eval->u.eval.close_paren &&
6183 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
6186 curlym_do_B: /* execute the B in /A{m,n}B/ */
6187 if (ST.c1 == CHRTEST_UNINIT) {
6188 /* calculate c1 and c2 for possible match of 1st char
6189 * following curly */
6190 ST.c1 = ST.c2 = CHRTEST_VOID;
6192 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
6193 regnode *text_node = ST.B;
6194 if (! HAS_TEXT(text_node))
6195 FIND_NEXT_IMPT(text_node);
6198 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
6200 But the former is redundant in light of the latter.
6202 if this changes back then the macro for
6203 IS_TEXT and friends need to change.
6205 if (PL_regkind[OP(text_node)] == EXACT) {
6206 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6207 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6217 PerlIO_printf(Perl_debug_log,
6218 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
6219 (int)(REPORT_CODE_OFF+(depth*2)),
6222 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
6223 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
6224 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6225 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6227 /* simulate B failing */
6229 PerlIO_printf(Perl_debug_log,
6230 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
6231 (int)(REPORT_CODE_OFF+(depth*2)),"",
6232 valid_utf8_to_uvchr((U8 *) locinput, NULL),
6233 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
6234 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
6236 state_num = CURLYM_B_fail;
6237 goto reenter_switch;
6240 else if (nextchr != ST.c1 && nextchr != ST.c2) {
6241 /* simulate B failing */
6243 PerlIO_printf(Perl_debug_log,
6244 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
6245 (int)(REPORT_CODE_OFF+(depth*2)),"",
6246 (int) nextchr, ST.c1, ST.c2)
6248 state_num = CURLYM_B_fail;
6249 goto reenter_switch;
6254 /* emulate CLOSE: mark current A as captured */
6255 I32 paren = ST.me->flags;
6257 rex->offs[paren].start
6258 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
6259 rex->offs[paren].end = locinput - reginfo->strbeg;
6260 if ((U32)paren > rex->lastparen)
6261 rex->lastparen = paren;
6262 rex->lastcloseparen = paren;
6265 rex->offs[paren].end = -1;
6266 if (cur_eval && cur_eval->u.eval.close_paren &&
6267 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
6276 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
6277 assert(0); /* NOTREACHED */
6279 case CURLYM_B_fail: /* just failed to match a B */
6280 REGCP_UNWIND(ST.cp);
6281 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6283 I32 max = ARG2(ST.me);
6284 if (max != REG_INFTY && ST.count == max)
6286 goto curlym_do_A; /* try to match a further A */
6288 /* backtrack one A */
6289 if (ST.count == ARG1(ST.me) /* min */)
6292 SET_locinput(HOPc(locinput, -ST.alen));
6293 goto curlym_do_B; /* try to match B */
6296 #define ST st->u.curly
6298 #define CURLY_SETPAREN(paren, success) \
6301 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
6302 rex->offs[paren].end = locinput - reginfo->strbeg; \
6303 if (paren > rex->lastparen) \
6304 rex->lastparen = paren; \
6305 rex->lastcloseparen = paren; \
6308 rex->offs[paren].end = -1; \
6309 rex->lastparen = ST.lastparen; \
6310 rex->lastcloseparen = ST.lastcloseparen; \
6314 case STAR: /* /A*B/ where A is width 1 char */
6318 scan = NEXTOPER(scan);
6321 case PLUS: /* /A+B/ where A is width 1 char */
6325 scan = NEXTOPER(scan);
6328 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
6329 ST.paren = scan->flags; /* Which paren to set */
6330 ST.lastparen = rex->lastparen;
6331 ST.lastcloseparen = rex->lastcloseparen;
6332 if (ST.paren > maxopenparen)
6333 maxopenparen = ST.paren;
6334 ST.min = ARG1(scan); /* min to match */
6335 ST.max = ARG2(scan); /* max to match */
6336 if (cur_eval && cur_eval->u.eval.close_paren &&
6337 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6341 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
6344 case CURLY: /* /A{m,n}B/ where A is width 1 char */
6346 ST.min = ARG1(scan); /* min to match */
6347 ST.max = ARG2(scan); /* max to match */
6348 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
6351 * Lookahead to avoid useless match attempts
6352 * when we know what character comes next.
6354 * Used to only do .*x and .*?x, but now it allows
6355 * for )'s, ('s and (?{ ... })'s to be in the way
6356 * of the quantifier and the EXACT-like node. -- japhy
6359 assert(ST.min <= ST.max);
6360 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
6361 ST.c1 = ST.c2 = CHRTEST_VOID;
6364 regnode *text_node = next;
6366 if (! HAS_TEXT(text_node))
6367 FIND_NEXT_IMPT(text_node);
6369 if (! HAS_TEXT(text_node))
6370 ST.c1 = ST.c2 = CHRTEST_VOID;
6372 if ( PL_regkind[OP(text_node)] != EXACT ) {
6373 ST.c1 = ST.c2 = CHRTEST_VOID;
6377 /* Currently we only get here when
6379 PL_rekind[OP(text_node)] == EXACT
6381 if this changes back then the macro for IS_TEXT and
6382 friends need to change. */
6383 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
6384 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
6396 char *li = locinput;
6399 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
6405 if (ST.c1 == CHRTEST_VOID)
6406 goto curly_try_B_min;
6408 ST.oldloc = locinput;
6410 /* set ST.maxpos to the furthest point along the
6411 * string that could possibly match */
6412 if (ST.max == REG_INFTY) {
6413 ST.maxpos = reginfo->strend - 1;
6415 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
6418 else if (utf8_target) {
6419 int m = ST.max - ST.min;
6420 for (ST.maxpos = locinput;
6421 m >0 && ST.maxpos < reginfo->strend; m--)
6422 ST.maxpos += UTF8SKIP(ST.maxpos);
6425 ST.maxpos = locinput + ST.max - ST.min;
6426 if (ST.maxpos >= reginfo->strend)
6427 ST.maxpos = reginfo->strend - 1;
6429 goto curly_try_B_min_known;
6433 /* avoid taking address of locinput, so it can remain
6435 char *li = locinput;
6436 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
6437 if (ST.count < ST.min)
6440 if ((ST.count > ST.min)
6441 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
6443 /* A{m,n} must come at the end of the string, there's
6444 * no point in backing off ... */
6446 /* ...except that $ and \Z can match before *and* after
6447 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
6448 We may back off by one in this case. */
6449 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
6453 goto curly_try_B_max;
6455 assert(0); /* NOTREACHED */
6458 case CURLY_B_min_known_fail:
6459 /* failed to find B in a non-greedy match where c1,c2 valid */
6461 REGCP_UNWIND(ST.cp);
6463 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6465 /* Couldn't or didn't -- move forward. */
6466 ST.oldloc = locinput;
6468 locinput += UTF8SKIP(locinput);
6472 curly_try_B_min_known:
6473 /* find the next place where 'B' could work, then call B */
6477 n = (ST.oldloc == locinput) ? 0 : 1;
6478 if (ST.c1 == ST.c2) {
6479 /* set n to utf8_distance(oldloc, locinput) */
6480 while (locinput <= ST.maxpos
6481 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
6483 locinput += UTF8SKIP(locinput);
6488 /* set n to utf8_distance(oldloc, locinput) */
6489 while (locinput <= ST.maxpos
6490 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
6491 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
6493 locinput += UTF8SKIP(locinput);
6498 else { /* Not utf8_target */
6499 if (ST.c1 == ST.c2) {
6500 while (locinput <= ST.maxpos &&
6501 UCHARAT(locinput) != ST.c1)
6505 while (locinput <= ST.maxpos
6506 && UCHARAT(locinput) != ST.c1
6507 && UCHARAT(locinput) != ST.c2)
6510 n = locinput - ST.oldloc;
6512 if (locinput > ST.maxpos)
6515 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
6516 * at b; check that everything between oldloc and
6517 * locinput matches */
6518 char *li = ST.oldloc;
6520 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
6522 assert(n == REG_INFTY || locinput == li);
6524 CURLY_SETPAREN(ST.paren, ST.count);
6525 if (cur_eval && cur_eval->u.eval.close_paren &&
6526 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6529 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
6531 assert(0); /* NOTREACHED */
6534 case CURLY_B_min_fail:
6535 /* failed to find B in a non-greedy match where c1,c2 invalid */
6537 REGCP_UNWIND(ST.cp);
6539 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6541 /* failed -- move forward one */
6543 char *li = locinput;
6544 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
6551 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
6552 ST.count > 0)) /* count overflow ? */
6555 CURLY_SETPAREN(ST.paren, ST.count);
6556 if (cur_eval && cur_eval->u.eval.close_paren &&
6557 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6560 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
6564 assert(0); /* NOTREACHED */
6568 /* a successful greedy match: now try to match B */
6569 if (cur_eval && cur_eval->u.eval.close_paren &&
6570 cur_eval->u.eval.close_paren == (U32)ST.paren) {
6574 bool could_match = locinput < reginfo->strend;
6576 /* If it could work, try it. */
6577 if (ST.c1 != CHRTEST_VOID && could_match) {
6578 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
6580 could_match = memEQ(locinput,
6585 UTF8SKIP(locinput));
6588 could_match = UCHARAT(locinput) == ST.c1
6589 || UCHARAT(locinput) == ST.c2;
6592 if (ST.c1 == CHRTEST_VOID || could_match) {
6593 CURLY_SETPAREN(ST.paren, ST.count);
6594 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
6595 assert(0); /* NOTREACHED */
6600 case CURLY_B_max_fail:
6601 /* failed to find B in a greedy match */
6603 REGCP_UNWIND(ST.cp);
6605 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6608 if (--ST.count < ST.min)
6610 locinput = HOPc(locinput, -1);
6611 goto curly_try_B_max;
6615 case END: /* last op of main pattern */
6618 /* we've just finished A in /(??{A})B/; now continue with B */
6620 st->u.eval.prev_rex = rex_sv; /* inner */
6622 /* Save *all* the positions. */
6623 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
6624 rex_sv = cur_eval->u.eval.prev_rex;
6625 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6626 SET_reg_curpm(rex_sv);
6627 rex = ReANY(rex_sv);
6628 rexi = RXi_GET(rex);
6629 cur_curlyx = cur_eval->u.eval.prev_curlyx;
6631 REGCP_SET(st->u.eval.lastcp);
6633 /* Restore parens of the outer rex without popping the
6635 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
6638 st->u.eval.prev_eval = cur_eval;
6639 cur_eval = cur_eval->u.eval.prev_eval;
6641 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
6642 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
6643 if ( nochange_depth )
6646 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
6647 locinput); /* match B */
6650 if (locinput < reginfo->till) {
6651 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6652 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
6654 (long)(locinput - startpos),
6655 (long)(reginfo->till - startpos),
6658 sayNO_SILENT; /* Cannot match: too short. */
6660 sayYES; /* Success! */
6662 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
6664 PerlIO_printf(Perl_debug_log,
6665 "%*s %ssubpattern success...%s\n",
6666 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
6667 sayYES; /* Success! */
6670 #define ST st->u.ifmatch
6675 case SUSPEND: /* (?>A) */
6677 newstart = locinput;
6680 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
6682 goto ifmatch_trivial_fail_test;
6684 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
6686 ifmatch_trivial_fail_test:
6688 char * const s = HOPBACKc(locinput, scan->flags);
6693 sw = 1 - cBOOL(ST.wanted);
6697 next = scan + ARG(scan);
6705 newstart = locinput;
6709 ST.logical = logical;
6710 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
6712 /* execute body of (?...A) */
6713 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
6714 assert(0); /* NOTREACHED */
6717 case IFMATCH_A_fail: /* body of (?...A) failed */
6718 ST.wanted = !ST.wanted;
6721 case IFMATCH_A: /* body of (?...A) succeeded */
6723 sw = cBOOL(ST.wanted);
6725 else if (!ST.wanted)
6728 if (OP(ST.me) != SUSPEND) {
6729 /* restore old position except for (?>...) */
6730 locinput = st->locinput;
6732 scan = ST.me + ARG(ST.me);
6735 continue; /* execute B */
6739 case LONGJMP: /* alternative with many branches compiles to
6740 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
6741 next = scan + ARG(scan);
6746 case COMMIT: /* (*COMMIT) */
6747 reginfo->cutpoint = reginfo->strend;
6750 case PRUNE: /* (*PRUNE) */
6752 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6753 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
6754 assert(0); /* NOTREACHED */
6756 case COMMIT_next_fail:
6760 case OPFAIL: /* (*FAIL) */
6762 assert(0); /* NOTREACHED */
6764 #define ST st->u.mark
6765 case MARKPOINT: /* (*MARK:foo) */
6766 ST.prev_mark = mark_state;
6767 ST.mark_name = sv_commit = sv_yes_mark
6768 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6770 ST.mark_loc = locinput;
6771 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
6772 assert(0); /* NOTREACHED */
6774 case MARKPOINT_next:
6775 mark_state = ST.prev_mark;
6777 assert(0); /* NOTREACHED */
6779 case MARKPOINT_next_fail:
6780 if (popmark && sv_eq(ST.mark_name,popmark))
6782 if (ST.mark_loc > startpoint)
6783 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6784 popmark = NULL; /* we found our mark */
6785 sv_commit = ST.mark_name;
6788 PerlIO_printf(Perl_debug_log,
6789 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
6790 REPORT_CODE_OFF+depth*2, "",
6791 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
6794 mark_state = ST.prev_mark;
6795 sv_yes_mark = mark_state ?
6796 mark_state->u.mark.mark_name : NULL;
6798 assert(0); /* NOTREACHED */
6800 case SKIP: /* (*SKIP) */
6802 /* (*SKIP) : if we fail we cut here*/
6803 ST.mark_name = NULL;
6804 ST.mark_loc = locinput;
6805 PUSH_STATE_GOTO(SKIP_next,next, locinput);
6807 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
6808 otherwise do nothing. Meaning we need to scan
6810 regmatch_state *cur = mark_state;
6811 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
6814 if ( sv_eq( cur->u.mark.mark_name,
6817 ST.mark_name = find;
6818 PUSH_STATE_GOTO( SKIP_next, next, locinput);
6820 cur = cur->u.mark.prev_mark;
6823 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
6826 case SKIP_next_fail:
6828 /* (*CUT:NAME) - Set up to search for the name as we
6829 collapse the stack*/
6830 popmark = ST.mark_name;
6832 /* (*CUT) - No name, we cut here.*/
6833 if (ST.mark_loc > startpoint)
6834 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
6835 /* but we set sv_commit to latest mark_name if there
6836 is one so they can test to see how things lead to this
6839 sv_commit=mark_state->u.mark.mark_name;
6843 assert(0); /* NOTREACHED */
6846 case LNBREAK: /* \R */
6847 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
6854 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
6855 PTR2UV(scan), OP(scan));
6856 Perl_croak(aTHX_ "regexp memory corruption");
6858 /* this is a point to jump to in order to increment
6859 * locinput by one character */
6861 assert(!NEXTCHR_IS_EOS);
6863 locinput += PL_utf8skip[nextchr];
6864 /* locinput is allowed to go 1 char off the end, but not 2+ */
6865 if (locinput > reginfo->strend)
6874 /* switch break jumps here */
6875 scan = next; /* prepare to execute the next op and ... */
6876 continue; /* ... jump back to the top, reusing st */
6877 assert(0); /* NOTREACHED */
6880 /* push a state that backtracks on success */
6881 st->u.yes.prev_yes_state = yes_state;
6885 /* push a new regex state, then continue at scan */
6887 regmatch_state *newst;
6890 regmatch_state *cur = st;
6891 regmatch_state *curyes = yes_state;
6893 regmatch_slab *slab = PL_regmatch_slab;
6894 for (;curd > -1;cur--,curd--) {
6895 if (cur < SLAB_FIRST(slab)) {
6897 cur = SLAB_LAST(slab);
6899 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
6900 REPORT_CODE_OFF + 2 + depth * 2,"",
6901 curd, PL_reg_name[cur->resume_state],
6902 (curyes == cur) ? "yes" : ""
6905 curyes = cur->u.yes.prev_yes_state;
6908 DEBUG_STATE_pp("push")
6911 st->locinput = locinput;
6913 if (newst > SLAB_LAST(PL_regmatch_slab))
6914 newst = S_push_slab(aTHX);
6915 PL_regmatch_state = newst;
6917 locinput = pushinput;
6920 assert(0); /* NOTREACHED */
6925 * We get here only if there's trouble -- normally "case END" is
6926 * the terminating point.
6928 Perl_croak(aTHX_ "corrupted regexp pointers");
6934 /* we have successfully completed a subexpression, but we must now
6935 * pop to the state marked by yes_state and continue from there */
6936 assert(st != yes_state);
6938 while (st != yes_state) {
6940 if (st < SLAB_FIRST(PL_regmatch_slab)) {
6941 PL_regmatch_slab = PL_regmatch_slab->prev;
6942 st = SLAB_LAST(PL_regmatch_slab);
6946 DEBUG_STATE_pp("pop (no final)");
6948 DEBUG_STATE_pp("pop (yes)");
6954 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
6955 || yes_state > SLAB_LAST(PL_regmatch_slab))
6957 /* not in this slab, pop slab */
6958 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
6959 PL_regmatch_slab = PL_regmatch_slab->prev;
6960 st = SLAB_LAST(PL_regmatch_slab);
6962 depth -= (st - yes_state);
6965 yes_state = st->u.yes.prev_yes_state;
6966 PL_regmatch_state = st;
6969 locinput= st->locinput;
6970 state_num = st->resume_state + no_final;
6971 goto reenter_switch;
6974 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
6975 PL_colors[4], PL_colors[5]));
6977 if (reginfo->info_aux_eval) {
6978 /* each successfully executed (?{...}) block does the equivalent of
6979 * local $^R = do {...}
6980 * When popping the save stack, all these locals would be undone;
6981 * bypass this by setting the outermost saved $^R to the latest
6983 /* I dont know if this is needed or works properly now.
6984 * see code related to PL_replgv elsewhere in this file.
6987 if (oreplsv != GvSV(PL_replgv))
6988 sv_setsv(oreplsv, GvSV(PL_replgv));
6995 PerlIO_printf(Perl_debug_log,
6996 "%*s %sfailed...%s\n",
6997 REPORT_CODE_OFF+depth*2, "",
6998 PL_colors[4], PL_colors[5])
7010 /* there's a previous state to backtrack to */
7012 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7013 PL_regmatch_slab = PL_regmatch_slab->prev;
7014 st = SLAB_LAST(PL_regmatch_slab);
7016 PL_regmatch_state = st;
7017 locinput= st->locinput;
7019 DEBUG_STATE_pp("pop");
7021 if (yes_state == st)
7022 yes_state = st->u.yes.prev_yes_state;
7024 state_num = st->resume_state + 1; /* failure = success + 1 */
7025 goto reenter_switch;
7030 if (rex->intflags & PREGf_VERBARG_SEEN) {
7031 SV *sv_err = get_sv("REGERROR", 1);
7032 SV *sv_mrk = get_sv("REGMARK", 1);
7034 sv_commit = &PL_sv_no;
7036 sv_yes_mark = &PL_sv_yes;
7039 sv_commit = &PL_sv_yes;
7040 sv_yes_mark = &PL_sv_no;
7044 sv_setsv(sv_err, sv_commit);
7045 sv_setsv(sv_mrk, sv_yes_mark);
7049 if (last_pushed_cv) {
7052 PERL_UNUSED_VAR(SP);
7055 assert(!result || locinput - reginfo->strbeg >= 0);
7056 return result ? locinput - reginfo->strbeg : -1;
7060 - regrepeat - repeatedly match something simple, report how many
7062 * What 'simple' means is a node which can be the operand of a quantifier like
7065 * startposp - pointer a pointer to the start position. This is updated
7066 * to point to the byte following the highest successful
7068 * p - the regnode to be repeatedly matched against.
7069 * reginfo - struct holding match state, such as strend
7070 * max - maximum number of things to match.
7071 * depth - (for debugging) backtracking depth.
7074 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
7075 regmatch_info *const reginfo, I32 max, int depth)
7077 char *scan; /* Pointer to current position in target string */
7079 char *loceol = reginfo->strend; /* local version */
7080 I32 hardcount = 0; /* How many matches so far */
7081 bool utf8_target = reginfo->is_utf8_target;
7082 int to_complement = 0; /* Invert the result? */
7084 _char_class_number classnum;
7086 PERL_UNUSED_ARG(depth);
7089 PERL_ARGS_ASSERT_REGREPEAT;
7092 if (max == REG_INFTY)
7094 else if (! utf8_target && loceol - scan > max)
7095 loceol = scan + max;
7097 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
7098 * to the maximum of how far we should go in it (leaving it set to the real
7099 * end, if the maximum permissible would take us beyond that). This allows
7100 * us to make the loop exit condition that we haven't gone past <loceol> to
7101 * also mean that we haven't exceeded the max permissible count, saving a
7102 * test each time through the loop. But it assumes that the OP matches a
7103 * single byte, which is true for most of the OPs below when applied to a
7104 * non-UTF-8 target. Those relatively few OPs that don't have this
7105 * characteristic will have to compensate.
7107 * There is no adjustment for UTF-8 targets, as the number of bytes per
7108 * character varies. OPs will have to test both that the count is less
7109 * than the max permissible (using <hardcount> to keep track), and that we
7110 * are still within the bounds of the string (using <loceol>. A few OPs
7111 * match a single byte no matter what the encoding. They can omit the max
7112 * test if, for the UTF-8 case, they do the adjustment that was skipped
7115 * Thus, the code above sets things up for the common case; and exceptional
7116 * cases need extra work; the common case is to make sure <scan> doesn't
7117 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
7118 * count doesn't exceed the maximum permissible */
7123 while (scan < loceol && hardcount < max && *scan != '\n') {
7124 scan += UTF8SKIP(scan);
7128 while (scan < loceol && *scan != '\n')
7134 while (scan < loceol && hardcount < max) {
7135 scan += UTF8SKIP(scan);
7142 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
7143 if (utf8_target && loceol - scan > max) {
7145 /* <loceol> hadn't been adjusted in the UTF-8 case */
7153 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7157 /* Can use a simple loop if the pattern char to match on is invariant
7158 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
7159 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
7160 * true iff it doesn't matter if the argument is in UTF-8 or not */
7161 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
7162 if (utf8_target && loceol - scan > max) {
7163 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
7164 * since here, to match at all, 1 char == 1 byte */
7165 loceol = scan + max;
7167 while (scan < loceol && UCHARAT(scan) == c) {
7171 else if (reginfo->is_utf8_pat) {
7173 STRLEN scan_char_len;
7175 /* When both target and pattern are UTF-8, we have to do
7177 while (hardcount < max
7179 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
7180 && memEQ(scan, STRING(p), scan_char_len))
7182 scan += scan_char_len;
7186 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
7188 /* Target isn't utf8; convert the character in the UTF-8
7189 * pattern to non-UTF8, and do a simple loop */
7190 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
7191 while (scan < loceol && UCHARAT(scan) == c) {
7194 } /* else pattern char is above Latin1, can't possibly match the
7199 /* Here, the string must be utf8; pattern isn't, and <c> is
7200 * different in utf8 than not, so can't compare them directly.
7201 * Outside the loop, find the two utf8 bytes that represent c, and
7202 * then look for those in sequence in the utf8 string */
7203 U8 high = UTF8_TWO_BYTE_HI(c);
7204 U8 low = UTF8_TWO_BYTE_LO(c);
7206 while (hardcount < max
7207 && scan + 1 < loceol
7208 && UCHARAT(scan) == high
7209 && UCHARAT(scan + 1) == low)
7217 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
7218 assert(! reginfo->is_utf8_pat);
7221 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7225 utf8_flags = FOLDEQ_LOCALE;
7228 case EXACTF: /* This node only generated for non-utf8 patterns */
7229 assert(! reginfo->is_utf8_pat);
7235 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
7239 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
7241 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
7243 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
7246 if (c1 == CHRTEST_VOID) {
7247 /* Use full Unicode fold matching */
7248 char *tmpeol = reginfo->strend;
7249 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
7250 while (hardcount < max
7251 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
7252 STRING(p), NULL, pat_len,
7253 reginfo->is_utf8_pat, utf8_flags))
7256 tmpeol = reginfo->strend;
7260 else if (utf8_target) {
7262 while (scan < loceol
7264 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
7266 scan += UTF8SKIP(scan);
7271 while (scan < loceol
7273 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
7274 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
7276 scan += UTF8SKIP(scan);
7281 else if (c1 == c2) {
7282 while (scan < loceol && UCHARAT(scan) == c1) {
7287 while (scan < loceol &&
7288 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
7298 while (hardcount < max
7300 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
7302 scan += UTF8SKIP(scan);
7306 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
7311 /* The argument (FLAGS) to all the POSIX node types is the class number */
7318 if (! utf8_target) {
7319 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
7325 while (hardcount < max && scan < loceol
7326 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
7329 scan += UTF8SKIP(scan);
7342 if (utf8_target && loceol - scan > max) {
7344 /* We didn't adjust <loceol> at the beginning of this routine
7345 * because is UTF-8, but it is actually ok to do so, since here, to
7346 * match, 1 char == 1 byte. */
7347 loceol = scan + max;
7349 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
7362 if (! utf8_target) {
7363 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
7369 /* The complement of something that matches only ASCII matches all
7370 * non-ASCII, plus everything in ASCII that isn't in the class. */
7371 while (hardcount < max && scan < loceol
7372 && (! isASCII_utf8(scan)
7373 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
7375 scan += UTF8SKIP(scan);
7386 if (! utf8_target) {
7387 while (scan < loceol && to_complement
7388 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
7395 classnum = (_char_class_number) FLAGS(p);
7396 if (classnum < _FIRST_NON_SWASH_CC) {
7398 /* Here, a swash is needed for above-Latin1 code points.
7399 * Process as many Latin1 code points using the built-in rules.
7400 * Go to another loop to finish processing upon encountering
7401 * the first Latin1 code point. We could do that in this loop
7402 * as well, but the other way saves having to test if the swash
7403 * has been loaded every time through the loop: extra space to
7405 while (hardcount < max && scan < loceol) {
7406 if (UTF8_IS_INVARIANT(*scan)) {
7407 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
7414 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
7415 if (! (to_complement
7416 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
7425 goto found_above_latin1;
7432 /* For these character classes, the knowledge of how to handle
7433 * every code point is compiled in to Perl via a macro. This
7434 * code is written for making the loops as tight as possible.
7435 * It could be refactored to save space instead */
7437 case _CC_ENUM_SPACE: /* XXX would require separate code
7438 if we revert the change of \v
7441 case _CC_ENUM_PSXSPC:
7442 while (hardcount < max
7444 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
7446 scan += UTF8SKIP(scan);
7450 case _CC_ENUM_BLANK:
7451 while (hardcount < max
7453 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
7455 scan += UTF8SKIP(scan);
7459 case _CC_ENUM_XDIGIT:
7460 while (hardcount < max
7462 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
7464 scan += UTF8SKIP(scan);
7468 case _CC_ENUM_VERTSPACE:
7469 while (hardcount < max
7471 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
7473 scan += UTF8SKIP(scan);
7477 case _CC_ENUM_CNTRL:
7478 while (hardcount < max
7480 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
7482 scan += UTF8SKIP(scan);
7487 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
7493 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
7495 /* Load the swash if not already present */
7496 if (! PL_utf8_swash_ptrs[classnum]) {
7497 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7498 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
7502 PL_XPosix_ptrs[classnum], &flags);
7505 while (hardcount < max && scan < loceol
7506 && to_complement ^ cBOOL(_generic_utf8(
7509 swash_fetch(PL_utf8_swash_ptrs[classnum],
7513 scan += UTF8SKIP(scan);
7520 while (hardcount < max && scan < loceol &&
7521 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
7526 /* LNBREAK can match one or two latin chars, which is ok, but we
7527 * have to use hardcount in this situation, and throw away the
7528 * adjustment to <loceol> done before the switch statement */
7529 loceol = reginfo->strend;
7530 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
7551 /* These are all 0 width, so match right here or not at all. */
7555 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
7556 assert(0); /* NOTREACHED */
7563 c = scan - *startposp;
7567 GET_RE_DEBUG_FLAGS_DECL;
7569 SV * const prop = sv_newmortal();
7570 regprop(prog, prop, p, reginfo);
7571 PerlIO_printf(Perl_debug_log,
7572 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
7573 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
7581 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
7583 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
7584 create a copy so that changes the caller makes won't change the shared one.
7585 If <altsvp> is non-null, will return NULL in it, for back-compat.
7588 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
7590 PERL_ARGS_ASSERT_REGCLASS_SWASH;
7596 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL));
7600 Perl__get_regclass_nonbitmap_data(pTHX_ const regexp *prog,
7601 const regnode* node,
7604 SV** only_utf8_locale_ptr)
7606 /* For internal core use only.
7607 * Returns the swash for the input 'node' in the regex 'prog'.
7608 * If <doinit> is 'true', will attempt to create the swash if not already
7610 * If <listsvp> is non-null, will return the printable contents of the
7611 * swash. This can be used to get debugging information even before the
7612 * swash exists, by calling this function with 'doinit' set to false, in
7613 * which case the components that will be used to eventually create the
7614 * swash are returned (in a printable form).
7615 * Tied intimately to how regcomp.c sets up the data structure */
7618 SV *si = NULL; /* Input swash initialization string */
7621 RXi_GET_DECL(prog,progi);
7622 const struct reg_data * const data = prog ? progi->data : NULL;
7624 PERL_ARGS_ASSERT__GET_REGCLASS_NONBITMAP_DATA;
7626 assert(ANYOF_FLAGS(node)
7627 & (ANYOF_UTF8|ANYOF_NONBITMAP_NON_UTF8|ANYOF_LOC_FOLD));
7629 if (data && data->count) {
7630 const U32 n = ARG(node);
7632 if (data->what[n] == 's') {
7633 SV * const rv = MUTABLE_SV(data->data[n]);
7634 AV * const av = MUTABLE_AV(SvRV(rv));
7635 SV **const ary = AvARRAY(av);
7636 U8 swash_init_flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
7638 si = *ary; /* ary[0] = the string to initialize the swash with */
7640 /* Elements 3 and 4 are either both present or both absent. [3] is
7641 * any inversion list generated at compile time; [4] indicates if
7642 * that inversion list has any user-defined properties in it. */
7643 if (av_tindex(av) >= 2) {
7644 if (only_utf8_locale_ptr
7646 && ary[2] != &PL_sv_undef)
7648 *only_utf8_locale_ptr = ary[2];
7651 assert(only_utf8_locale_ptr);
7652 *only_utf8_locale_ptr = NULL;
7655 if (av_tindex(av) >= 3) {
7658 swash_init_flags |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
7666 /* Element [1] is reserved for the set-up swash. If already there,
7667 * return it; if not, create it and store it there */
7668 if (ary[1] && SvROK(ary[1])) {
7671 else if (doinit && ((si && si != &PL_sv_undef)
7672 || (invlist && invlist != &PL_sv_undef))) {
7674 sw = _core_swash_init("utf8", /* the utf8 package */
7678 0, /* not from tr/// */
7681 (void)av_store(av, 1, sw);
7686 /* If requested, return a printable version of what this swash matches */
7688 SV* matches_string = newSVpvs("");
7690 /* The swash should be used, if possible, to get the data, as it
7691 * contains the resolved data. But this function can be called at
7692 * compile-time, before everything gets resolved, in which case we
7693 * return the currently best available information, which is the string
7694 * that will eventually be used to do that resolving, 'si' */
7695 if ((! sw || (invlist = _get_swash_invlist(sw)) == NULL)
7696 && (si && si != &PL_sv_undef))
7698 sv_catsv(matches_string, si);
7701 /* Add the inversion list to whatever we have. This may have come from
7702 * the swash, or from an input parameter */
7704 sv_catsv(matches_string, _invlist_contents(invlist));
7706 *listsvp = matches_string;
7711 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
7714 - reginclass - determine if a character falls into a character class
7716 n is the ANYOF regnode
7717 p is the target string
7718 p_end points to one byte beyond the end of the target string
7719 utf8_target tells whether p is in UTF-8.
7721 Returns true if matched; false otherwise.
7723 Note that this can be a synthetic start class, a combination of various
7724 nodes, so things you think might be mutually exclusive, such as locale,
7725 aren't. It can match both locale and non-locale
7730 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
7733 const char flags = ANYOF_FLAGS(n);
7737 PERL_ARGS_ASSERT_REGINCLASS;
7739 /* If c is not already the code point, get it. Note that
7740 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
7741 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
7743 c = utf8n_to_uvchr(p, p_end - p, &c_len,
7744 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
7745 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
7746 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
7747 * UTF8_ALLOW_FFFF */
7748 if (c_len == (STRLEN)-1)
7749 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
7752 /* If this character is potentially in the bitmap, check it */
7754 if (ANYOF_BITMAP_TEST(n, c))
7756 else if (flags & ANYOF_NON_UTF8_NON_ASCII_ALL
7762 else if (flags & ANYOF_LOCALE_FLAGS) {
7763 if (flags & ANYOF_LOC_FOLD) {
7764 if (ANYOF_BITMAP_TEST(n, PL_fold_locale[c])) {
7768 if (! match && ANYOF_POSIXL_TEST_ANY_SET(n)) {
7770 /* The data structure is arranged so bits 0, 2, 4, ... are set
7771 * if the class includes the Posix character class given by
7772 * bit/2; and 1, 3, 5, ... are set if the class includes the
7773 * complemented Posix class given by int(bit/2). So we loop
7774 * through the bits, each time changing whether we complement
7775 * the result or not. Suppose for the sake of illustration
7776 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
7777 * is set, it means there is a match for this ANYOF node if the
7778 * character is in the class given by the expression (0 / 2 = 0
7779 * = \w). If it is in that class, isFOO_lc() will return 1,
7780 * and since 'to_complement' is 0, the result will stay TRUE,
7781 * and we exit the loop. Suppose instead that bit 0 is 0, but
7782 * bit 1 is 1. That means there is a match if the character
7783 * matches \W. We won't bother to call isFOO_lc() on bit 0,
7784 * but will on bit 1. On the second iteration 'to_complement'
7785 * will be 1, so the exclusive or will reverse things, so we
7786 * are testing for \W. On the third iteration, 'to_complement'
7787 * will be 0, and we would be testing for \s; the fourth
7788 * iteration would test for \S, etc.
7790 * Note that this code assumes that all the classes are closed
7791 * under folding. For example, if a character matches \w, then
7792 * its fold does too; and vice versa. This should be true for
7793 * any well-behaved locale for all the currently defined Posix
7794 * classes, except for :lower: and :upper:, which are handled
7795 * by the pseudo-class :cased: which matches if either of the
7796 * other two does. To get rid of this assumption, an outer
7797 * loop could be used below to iterate over both the source
7798 * character, and its fold (if different) */
7801 int to_complement = 0;
7803 while (count < ANYOF_MAX) {
7804 if (ANYOF_POSIXL_TEST(n, count)
7805 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
7818 /* If the bitmap didn't (or couldn't) match, and something outside the
7819 * bitmap could match, try that. */
7821 if (c >= 256 && (flags & ANYOF_ABOVE_LATIN1_ALL)) {
7822 match = TRUE; /* Everything above 255 matches */
7824 else if ((flags & ANYOF_NONBITMAP_NON_UTF8)
7825 || (utf8_target && (flags & ANYOF_UTF8))
7826 || ((flags & ANYOF_LOC_FOLD)
7827 && IN_UTF8_CTYPE_LOCALE
7828 && ARG(n) != ANYOF_NONBITMAP_EMPTY))
7830 SV* only_utf8_locale = NULL;
7831 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
7838 } else { /* Convert to utf8 */
7839 utf8_p = utf8_buffer;
7840 append_utf8_from_native_byte(*p, &utf8_p);
7841 utf8_p = utf8_buffer;
7844 if (swash_fetch(sw, utf8_p, TRUE)) {
7848 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
7849 match = _invlist_contains_cp(only_utf8_locale, c);
7853 if (UNICODE_IS_SUPER(c)
7854 && (flags & ANYOF_WARN_SUPER)
7855 && ckWARN_d(WARN_NON_UNICODE))
7857 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
7858 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c);
7862 #if ANYOF_INVERT != 1
7863 /* Depending on compiler optimization cBOOL takes time, so if don't have to
7865 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
7868 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
7869 return (flags & ANYOF_INVERT) ^ match;
7873 S_reghop3(U8 *s, SSize_t off, const U8* lim)
7875 /* return the position 'off' UTF-8 characters away from 's', forward if
7876 * 'off' >= 0, backwards if negative. But don't go outside of position
7877 * 'lim', which better be < s if off < 0 */
7879 PERL_ARGS_ASSERT_REGHOP3;
7882 while (off-- && s < lim) {
7883 /* XXX could check well-formedness here */
7888 while (off++ && s > lim) {
7890 if (UTF8_IS_CONTINUED(*s)) {
7891 while (s > lim && UTF8_IS_CONTINUATION(*s))
7894 /* XXX could check well-formedness here */
7901 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
7903 PERL_ARGS_ASSERT_REGHOP4;
7906 while (off-- && s < rlim) {
7907 /* XXX could check well-formedness here */
7912 while (off++ && s > llim) {
7914 if (UTF8_IS_CONTINUED(*s)) {
7915 while (s > llim && UTF8_IS_CONTINUATION(*s))
7918 /* XXX could check well-formedness here */
7924 /* like reghop3, but returns NULL on overrun, rather than returning last
7928 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
7930 PERL_ARGS_ASSERT_REGHOPMAYBE3;
7933 while (off-- && s < lim) {
7934 /* XXX could check well-formedness here */
7941 while (off++ && s > lim) {
7943 if (UTF8_IS_CONTINUED(*s)) {
7944 while (s > lim && UTF8_IS_CONTINUATION(*s))
7947 /* XXX could check well-formedness here */
7956 /* when executing a regex that may have (?{}), extra stuff needs setting
7957 up that will be visible to the called code, even before the current
7958 match has finished. In particular:
7960 * $_ is localised to the SV currently being matched;
7961 * pos($_) is created if necessary, ready to be updated on each call-out
7963 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
7964 isn't set until the current pattern is successfully finished), so that
7965 $1 etc of the match-so-far can be seen;
7966 * save the old values of subbeg etc of the current regex, and set then
7967 to the current string (again, this is normally only done at the end
7972 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
7975 regexp *const rex = ReANY(reginfo->prog);
7976 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
7978 eval_state->rex = rex;
7981 /* Make $_ available to executed code. */
7982 if (reginfo->sv != DEFSV) {
7984 DEFSV_set(reginfo->sv);
7987 if (!(mg = mg_find_mglob(reginfo->sv))) {
7988 /* prepare for quick setting of pos */
7989 mg = sv_magicext_mglob(reginfo->sv);
7992 eval_state->pos_magic = mg;
7993 eval_state->pos = mg->mg_len;
7994 eval_state->pos_flags = mg->mg_flags;
7997 eval_state->pos_magic = NULL;
7999 if (!PL_reg_curpm) {
8000 /* PL_reg_curpm is a fake PMOP that we can attach the current
8001 * regex to and point PL_curpm at, so that $1 et al are visible
8002 * within a /(?{})/. It's just allocated once per interpreter the
8003 * first time its needed */
8004 Newxz(PL_reg_curpm, 1, PMOP);
8007 SV* const repointer = &PL_sv_undef;
8008 /* this regexp is also owned by the new PL_reg_curpm, which
8009 will try to free it. */
8010 av_push(PL_regex_padav, repointer);
8011 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
8012 PL_regex_pad = AvARRAY(PL_regex_padav);
8016 SET_reg_curpm(reginfo->prog);
8017 eval_state->curpm = PL_curpm;
8018 PL_curpm = PL_reg_curpm;
8019 if (RXp_MATCH_COPIED(rex)) {
8020 /* Here is a serious problem: we cannot rewrite subbeg,
8021 since it may be needed if this match fails. Thus
8022 $` inside (?{}) could fail... */
8023 eval_state->subbeg = rex->subbeg;
8024 eval_state->sublen = rex->sublen;
8025 eval_state->suboffset = rex->suboffset;
8026 eval_state->subcoffset = rex->subcoffset;
8028 eval_state->saved_copy = rex->saved_copy;
8030 RXp_MATCH_COPIED_off(rex);
8033 eval_state->subbeg = NULL;
8034 rex->subbeg = (char *)reginfo->strbeg;
8036 rex->subcoffset = 0;
8037 rex->sublen = reginfo->strend - reginfo->strbeg;
8041 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
8044 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
8046 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
8047 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
8050 Safefree(aux->poscache);
8054 /* undo the effects of S_setup_eval_state() */
8056 if (eval_state->subbeg) {
8057 regexp * const rex = eval_state->rex;
8058 rex->subbeg = eval_state->subbeg;
8059 rex->sublen = eval_state->sublen;
8060 rex->suboffset = eval_state->suboffset;
8061 rex->subcoffset = eval_state->subcoffset;
8063 rex->saved_copy = eval_state->saved_copy;
8065 RXp_MATCH_COPIED_on(rex);
8067 if (eval_state->pos_magic)
8069 eval_state->pos_magic->mg_len = eval_state->pos;
8070 eval_state->pos_magic->mg_flags =
8071 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
8072 | (eval_state->pos_flags & MGf_BYTES);
8075 PL_curpm = eval_state->curpm;
8078 PL_regmatch_state = aux->old_regmatch_state;
8079 PL_regmatch_slab = aux->old_regmatch_slab;
8081 /* free all slabs above current one - this must be the last action
8082 * of this function, as aux and eval_state are allocated within
8083 * slabs and may be freed here */
8085 s = PL_regmatch_slab->next;
8087 PL_regmatch_slab->next = NULL;
8089 regmatch_slab * const osl = s;
8098 S_to_utf8_substr(pTHX_ regexp *prog)
8100 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
8101 * on the converted value */
8105 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
8108 if (prog->substrs->data[i].substr
8109 && !prog->substrs->data[i].utf8_substr) {
8110 SV* const sv = newSVsv(prog->substrs->data[i].substr);
8111 prog->substrs->data[i].utf8_substr = sv;
8112 sv_utf8_upgrade(sv);
8113 if (SvVALID(prog->substrs->data[i].substr)) {
8114 if (SvTAIL(prog->substrs->data[i].substr)) {
8115 /* Trim the trailing \n that fbm_compile added last
8117 SvCUR_set(sv, SvCUR(sv) - 1);
8118 /* Whilst this makes the SV technically "invalid" (as its
8119 buffer is no longer followed by "\0") when fbm_compile()
8120 adds the "\n" back, a "\0" is restored. */
8121 fbm_compile(sv, FBMcf_TAIL);
8125 if (prog->substrs->data[i].substr == prog->check_substr)
8126 prog->check_utf8 = sv;
8132 S_to_byte_substr(pTHX_ regexp *prog)
8134 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
8135 * on the converted value; returns FALSE if can't be converted. */
8139 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
8142 if (prog->substrs->data[i].utf8_substr
8143 && !prog->substrs->data[i].substr) {
8144 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
8145 if (! sv_utf8_downgrade(sv, TRUE)) {
8148 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
8149 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
8150 /* Trim the trailing \n that fbm_compile added last
8152 SvCUR_set(sv, SvCUR(sv) - 1);
8153 fbm_compile(sv, FBMcf_TAIL);
8157 prog->substrs->data[i].substr = sv;
8158 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
8159 prog->check_substr = sv;
8168 * c-indentation-style: bsd
8170 * indent-tabs-mode: nil
8173 * ex: set ts=8 sts=4 sw=4 et:
projects / perl / modules / re-engine-Hooks.git / blob