5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
40 #define B_ON_NON_UTF8_LOCALE_IS_WRONG \
41 "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale"
44 * pregcomp and pregexec -- regsub and regerror are not used in perl
46 * Copyright (c) 1986 by University of Toronto.
47 * Written by Henry Spencer. Not derived from licensed software.
49 * Permission is granted to anyone to use this software for any
50 * purpose on any computer system, and to redistribute it freely,
51 * subject to the following restrictions:
53 * 1. The author is not responsible for the consequences of use of
54 * this software, no matter how awful, even if they arise
57 * 2. The origin of this software must not be misrepresented, either
58 * by explicit claim or by omission.
60 * 3. Altered versions must be plainly marked as such, and must not
61 * be misrepresented as being the original software.
63 **** Alterations to Henry's code are...
65 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
66 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
67 **** by Larry Wall and others
69 **** You may distribute under the terms of either the GNU General Public
70 **** License or the Artistic License, as specified in the README file.
72 * Beware that some of this code is subtly aware of the way operator
73 * precedence is structured in regular expressions. Serious changes in
74 * regular-expression syntax might require a total rethink.
77 #define PERL_IN_REGEXEC_C
80 #ifdef PERL_IN_XSUB_RE
86 #include "inline_invlist.c"
87 #include "unicode_constants.h"
90 /* At least one required character in the target string is expressible only in
92 static const char* const non_utf8_target_but_utf8_required
93 = "Can't match, because target string needs to be in UTF-8\n";
96 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
97 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%s", non_utf8_target_but_utf8_required));\
101 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
104 #define STATIC static
107 /* Valid only for non-utf8 strings: avoids the reginclass
108 * call if there are no complications: i.e., if everything matchable is
109 * straight forward in the bitmap */
110 #define REGINCLASS(prog,p,c) (ANYOF_FLAGS(p) ? reginclass(prog,p,c,c+1,0) \
111 : ANYOF_BITMAP_TEST(p,*(c)))
117 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
118 #define CHR_DIST(a,b) (reginfo->is_utf8_target ? utf8_distance(a,b) : a - b)
120 #define HOPc(pos,off) \
121 (char *)(reginfo->is_utf8_target \
122 ? reghop3((U8*)pos, off, \
123 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
126 #define HOPBACKc(pos, off) \
127 (char*)(reginfo->is_utf8_target \
128 ? reghopmaybe3((U8*)pos, -off, (U8*)(reginfo->strbeg)) \
129 : (pos - off >= reginfo->strbeg) \
133 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
134 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
136 /* lim must be +ve. Returns NULL on overshoot */
137 #define HOPMAYBE3(pos,off,lim) \
138 (reginfo->is_utf8_target \
139 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
140 : ((U8*)pos + off <= lim) \
144 /* like HOP3, but limits the result to <= lim even for the non-utf8 case.
145 * off must be >=0; args should be vars rather than expressions */
146 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
147 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
148 : (U8*)((pos + off) > lim ? lim : (pos + off)))
150 #define HOP4(pos,off,llim, rlim) (reginfo->is_utf8_target \
151 ? reghop4((U8*)(pos), off, (U8*)(llim), (U8*)(rlim)) \
153 #define HOP4c(pos,off,llim, rlim) ((char*)HOP4(pos,off,llim, rlim))
155 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
156 #define NEXTCHR_IS_EOS (nextchr < 0)
158 #define SET_nextchr \
159 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
161 #define SET_locinput(p) \
166 #define LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist) STMT_START { \
168 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST; \
169 swash_ptr = _core_swash_init("utf8", property_name, &PL_sv_undef, \
170 1, 0, invlist, &flags); \
175 /* If in debug mode, we test that a known character properly matches */
177 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
180 utf8_char_in_property) \
181 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist); \
182 assert(swash_fetch(swash_ptr, (U8 *) utf8_char_in_property, TRUE));
184 # define LOAD_UTF8_CHARCLASS_DEBUG_TEST(swash_ptr, \
187 utf8_char_in_property) \
188 LOAD_UTF8_CHARCLASS(swash_ptr, property_name, invlist)
191 #define LOAD_UTF8_CHARCLASS_ALNUM() LOAD_UTF8_CHARCLASS_DEBUG_TEST( \
192 PL_utf8_swash_ptrs[_CC_WORDCHAR], \
194 PL_XPosix_ptrs[_CC_WORDCHAR], \
195 LATIN_CAPITAL_LETTER_SHARP_S_UTF8);
197 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
198 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
200 /* for use after a quantifier and before an EXACT-like node -- japhy */
201 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
203 * NOTE that *nothing* that affects backtracking should be in here, specifically
204 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
205 * node that is in between two EXACT like nodes when ascertaining what the required
206 * "follow" character is. This should probably be moved to regex compile time
207 * although it may be done at run time beause of the REF possibility - more
208 * investigation required. -- demerphq
210 #define JUMPABLE(rn) ( \
212 (OP(rn) == CLOSE && (!cur_eval || cur_eval->u.eval.close_paren != ARG(rn))) || \
214 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
215 OP(rn) == PLUS || OP(rn) == MINMOD || \
217 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
219 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
221 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
224 /* Currently these are only used when PL_regkind[OP(rn)] == EXACT so
225 we don't need this definition. XXX These are now out-of-sync*/
226 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==REF || OP(rn)==NREF )
227 #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 )
228 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL || OP(rn)==REFFL || OP(rn)==NREFFL )
231 /* ... so we use this as its faster. */
232 #define IS_TEXT(rn) ( OP(rn)==EXACT || OP(rn)==EXACTL )
233 #define IS_TEXTFU(rn) ( OP(rn)==EXACTFU || OP(rn)==EXACTFLU8 || OP(rn)==EXACTFU_SS || OP(rn) == EXACTFA || OP(rn) == EXACTFA_NO_TRIE)
234 #define IS_TEXTF(rn) ( OP(rn)==EXACTF )
235 #define IS_TEXTFL(rn) ( OP(rn)==EXACTFL )
240 Search for mandatory following text node; for lookahead, the text must
241 follow but for lookbehind (rn->flags != 0) we skip to the next step.
243 #define FIND_NEXT_IMPT(rn) STMT_START { \
244 while (JUMPABLE(rn)) { \
245 const OPCODE type = OP(rn); \
246 if (type == SUSPEND || PL_regkind[type] == CURLY) \
247 rn = NEXTOPER(NEXTOPER(rn)); \
248 else if (type == PLUS) \
250 else if (type == IFMATCH) \
251 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
252 else rn += NEXT_OFF(rn); \
256 #define SLAB_FIRST(s) (&(s)->states[0])
257 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
259 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
260 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
261 static regmatch_state * S_push_slab(pTHX);
263 #define REGCP_PAREN_ELEMS 3
264 #define REGCP_OTHER_ELEMS 3
265 #define REGCP_FRAME_ELEMS 1
266 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
267 * are needed for the regexp context stack bookkeeping. */
270 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen)
272 const int retval = PL_savestack_ix;
273 const int paren_elems_to_push =
274 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
275 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
276 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
278 GET_RE_DEBUG_FLAGS_DECL;
280 PERL_ARGS_ASSERT_REGCPPUSH;
282 if (paren_elems_to_push < 0)
283 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
284 (int)paren_elems_to_push, (int)maxopenparen,
285 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
287 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
288 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %"UVuf
289 " out of range (%lu-%ld)",
291 (unsigned long)maxopenparen,
294 SSGROW(total_elems + REGCP_FRAME_ELEMS);
297 if ((int)maxopenparen > (int)parenfloor)
298 PerlIO_printf(Perl_debug_log,
299 "rex=0x%"UVxf" offs=0x%"UVxf": saving capture indices:\n",
304 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
305 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
306 SSPUSHIV(rex->offs[p].end);
307 SSPUSHIV(rex->offs[p].start);
308 SSPUSHINT(rex->offs[p].start_tmp);
309 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
310 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"\n",
312 (IV)rex->offs[p].start,
313 (IV)rex->offs[p].start_tmp,
317 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
318 SSPUSHINT(maxopenparen);
319 SSPUSHINT(rex->lastparen);
320 SSPUSHINT(rex->lastcloseparen);
321 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
326 /* These are needed since we do not localize EVAL nodes: */
327 #define REGCP_SET(cp) \
329 PerlIO_printf(Perl_debug_log, \
330 " Setting an EVAL scope, savestack=%"IVdf"\n", \
331 (IV)PL_savestack_ix)); \
334 #define REGCP_UNWIND(cp) \
336 if (cp != PL_savestack_ix) \
337 PerlIO_printf(Perl_debug_log, \
338 " Clearing an EVAL scope, savestack=%"IVdf"..%"IVdf"\n", \
339 (IV)(cp), (IV)PL_savestack_ix)); \
342 #define UNWIND_PAREN(lp, lcp) \
343 for (n = rex->lastparen; n > lp; n--) \
344 rex->offs[n].end = -1; \
345 rex->lastparen = n; \
346 rex->lastcloseparen = lcp;
350 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p)
354 GET_RE_DEBUG_FLAGS_DECL;
356 PERL_ARGS_ASSERT_REGCPPOP;
358 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
360 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
361 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
362 rex->lastcloseparen = SSPOPINT;
363 rex->lastparen = SSPOPINT;
364 *maxopenparen_p = SSPOPINT;
366 i -= REGCP_OTHER_ELEMS;
367 /* Now restore the parentheses context. */
369 if (i || rex->lastparen + 1 <= rex->nparens)
370 PerlIO_printf(Perl_debug_log,
371 "rex=0x%"UVxf" offs=0x%"UVxf": restoring capture indices to:\n",
376 paren = *maxopenparen_p;
377 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
379 rex->offs[paren].start_tmp = SSPOPINT;
380 rex->offs[paren].start = SSPOPIV;
382 if (paren <= rex->lastparen)
383 rex->offs[paren].end = tmps;
384 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
385 " \\%"UVuf": %"IVdf"(%"IVdf")..%"IVdf"%s\n",
387 (IV)rex->offs[paren].start,
388 (IV)rex->offs[paren].start_tmp,
389 (IV)rex->offs[paren].end,
390 (paren > rex->lastparen ? "(skipped)" : ""));
395 /* It would seem that the similar code in regtry()
396 * already takes care of this, and in fact it is in
397 * a better location to since this code can #if 0-ed out
398 * but the code in regtry() is needed or otherwise tests
399 * requiring null fields (pat.t#187 and split.t#{13,14}
400 * (as of patchlevel 7877) will fail. Then again,
401 * this code seems to be necessary or otherwise
402 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
403 * --jhi updated by dapm */
404 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
405 if (i > *maxopenparen_p)
406 rex->offs[i].start = -1;
407 rex->offs[i].end = -1;
408 DEBUG_BUFFERS_r( PerlIO_printf(Perl_debug_log,
409 " \\%"UVuf": %s ..-1 undeffing\n",
411 (i > *maxopenparen_p) ? "-1" : " "
417 /* restore the parens and associated vars at savestack position ix,
418 * but without popping the stack */
421 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p)
423 I32 tmpix = PL_savestack_ix;
424 PL_savestack_ix = ix;
425 regcppop(rex, maxopenparen_p);
426 PL_savestack_ix = tmpix;
429 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
432 S_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
434 /* Returns a boolean as to whether or not 'character' is a member of the
435 * Posix character class given by 'classnum' that should be equivalent to a
436 * value in the typedef '_char_class_number'.
438 * Ideally this could be replaced by a just an array of function pointers
439 * to the C library functions that implement the macros this calls.
440 * However, to compile, the precise function signatures are required, and
441 * these may vary from platform to to platform. To avoid having to figure
442 * out what those all are on each platform, I (khw) am using this method,
443 * which adds an extra layer of function call overhead (unless the C
444 * optimizer strips it away). But we don't particularly care about
445 * performance with locales anyway. */
447 switch ((_char_class_number) classnum) {
448 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
449 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
450 case _CC_ENUM_ASCII: return isASCII_LC(character);
451 case _CC_ENUM_BLANK: return isBLANK_LC(character);
452 case _CC_ENUM_CASED: return isLOWER_LC(character)
453 || isUPPER_LC(character);
454 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
455 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
456 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
457 case _CC_ENUM_LOWER: return isLOWER_LC(character);
458 case _CC_ENUM_PRINT: return isPRINT_LC(character);
459 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
460 case _CC_ENUM_SPACE: return isSPACE_LC(character);
461 case _CC_ENUM_UPPER: return isUPPER_LC(character);
462 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
463 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
464 default: /* VERTSPACE should never occur in locales */
465 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
468 NOT_REACHED; /* NOTREACHED */
473 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character)
475 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
476 * 'character' is a member of the Posix character class given by 'classnum'
477 * that should be equivalent to a value in the typedef
478 * '_char_class_number'.
480 * This just calls isFOO_lc on the code point for the character if it is in
481 * the range 0-255. Outside that range, all characters use Unicode
482 * rules, ignoring any locale. So use the Unicode function if this class
483 * requires a swash, and use the Unicode macro otherwise. */
485 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
487 if (UTF8_IS_INVARIANT(*character)) {
488 return isFOO_lc(classnum, *character);
490 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
491 return isFOO_lc(classnum,
492 TWO_BYTE_UTF8_TO_NATIVE(*character, *(character + 1)));
495 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, character + UTF8SKIP(character));
497 if (classnum < _FIRST_NON_SWASH_CC) {
499 /* Initialize the swash unless done already */
500 if (! PL_utf8_swash_ptrs[classnum]) {
501 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
502 PL_utf8_swash_ptrs[classnum] =
503 _core_swash_init("utf8",
506 PL_XPosix_ptrs[classnum], &flags);
509 return cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum], (U8 *)
511 TRUE /* is UTF */ ));
514 switch ((_char_class_number) classnum) {
515 case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character);
516 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
517 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
518 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
522 return FALSE; /* Things like CNTRL are always below 256 */
526 * pregexec and friends
529 #ifndef PERL_IN_XSUB_RE
531 - pregexec - match a regexp against a string
534 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
535 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
536 /* stringarg: the point in the string at which to begin matching */
537 /* strend: pointer to null at end of string */
538 /* strbeg: real beginning of string */
539 /* minend: end of match must be >= minend bytes after stringarg. */
540 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
541 * itself is accessed via the pointers above */
542 /* nosave: For optimizations. */
544 PERL_ARGS_ASSERT_PREGEXEC;
547 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
548 nosave ? 0 : REXEC_COPY_STR);
554 /* re_intuit_start():
556 * Based on some optimiser hints, try to find the earliest position in the
557 * string where the regex could match.
559 * rx: the regex to match against
560 * sv: the SV being matched: only used for utf8 flag; the string
561 * itself is accessed via the pointers below. Note that on
562 * something like an overloaded SV, SvPOK(sv) may be false
563 * and the string pointers may point to something unrelated to
565 * strbeg: real beginning of string
566 * strpos: the point in the string at which to begin matching
567 * strend: pointer to the byte following the last char of the string
568 * flags currently unused; set to 0
569 * data: currently unused; set to NULL
571 * The basic idea of re_intuit_start() is to use some known information
572 * about the pattern, namely:
574 * a) the longest known anchored substring (i.e. one that's at a
575 * constant offset from the beginning of the pattern; but not
576 * necessarily at a fixed offset from the beginning of the
578 * b) the longest floating substring (i.e. one that's not at a constant
579 * offset from the beginning of the pattern);
580 * c) Whether the pattern is anchored to the string; either
581 * an absolute anchor: /^../, or anchored to \n: /^.../m,
582 * or anchored to pos(): /\G/;
583 * d) A start class: a real or synthetic character class which
584 * represents which characters are legal at the start of the pattern;
586 * to either quickly reject the match, or to find the earliest position
587 * within the string at which the pattern might match, thus avoiding
588 * running the full NFA engine at those earlier locations, only to
589 * eventually fail and retry further along.
591 * Returns NULL if the pattern can't match, or returns the address within
592 * the string which is the earliest place the match could occur.
594 * The longest of the anchored and floating substrings is called 'check'
595 * and is checked first. The other is called 'other' and is checked
596 * second. The 'other' substring may not be present. For example,
598 * /(abc|xyz)ABC\d{0,3}DEFG/
602 * check substr (float) = "DEFG", offset 6..9 chars
603 * other substr (anchored) = "ABC", offset 3..3 chars
606 * Be aware that during the course of this function, sometimes 'anchored'
607 * refers to a substring being anchored relative to the start of the
608 * pattern, and sometimes to the pattern itself being anchored relative to
609 * the string. For example:
611 * /\dabc/: "abc" is anchored to the pattern;
612 * /^\dabc/: "abc" is anchored to the pattern and the string;
613 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
614 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
615 * but the pattern is anchored to the string.
619 Perl_re_intuit_start(pTHX_
622 const char * const strbeg,
626 re_scream_pos_data *data)
628 struct regexp *const prog = ReANY(rx);
629 SSize_t start_shift = prog->check_offset_min;
630 /* Should be nonnegative! */
631 SSize_t end_shift = 0;
632 /* current lowest pos in string where the regex can start matching */
633 char *rx_origin = strpos;
635 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
636 U8 other_ix = 1 - prog->substrs->check_ix;
638 char *other_last = strpos;/* latest pos 'other' substr already checked to */
639 char *check_at = NULL; /* check substr found at this pos */
640 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
641 RXi_GET_DECL(prog,progi);
642 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
643 regmatch_info *const reginfo = ®info_buf;
644 GET_RE_DEBUG_FLAGS_DECL;
646 PERL_ARGS_ASSERT_RE_INTUIT_START;
647 PERL_UNUSED_ARG(flags);
648 PERL_UNUSED_ARG(data);
650 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
651 "Intuit: trying to determine minimum start position...\n"));
653 /* for now, assume that all substr offsets are positive. If at some point
654 * in the future someone wants to do clever things with look-behind and
655 * -ve offsets, they'll need to fix up any code in this function
656 * which uses these offsets. See the thread beginning
657 * <20140113145929.GF27210@iabyn.com>
659 assert(prog->substrs->data[0].min_offset >= 0);
660 assert(prog->substrs->data[0].max_offset >= 0);
661 assert(prog->substrs->data[1].min_offset >= 0);
662 assert(prog->substrs->data[1].max_offset >= 0);
663 assert(prog->substrs->data[2].min_offset >= 0);
664 assert(prog->substrs->data[2].max_offset >= 0);
666 /* for now, assume that if both present, that the floating substring
667 * doesn't start before the anchored substring.
668 * If you break this assumption (e.g. doing better optimisations
669 * with lookahead/behind), then you'll need to audit the code in this
670 * function carefully first
673 ! ( (prog->anchored_utf8 || prog->anchored_substr)
674 && (prog->float_utf8 || prog->float_substr))
675 || (prog->float_min_offset >= prog->anchored_offset));
677 /* byte rather than char calculation for efficiency. It fails
678 * to quickly reject some cases that can't match, but will reject
679 * them later after doing full char arithmetic */
680 if (prog->minlen > strend - strpos) {
681 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
682 " String too short...\n"));
686 RX_MATCH_UTF8_set(rx,utf8_target);
687 reginfo->is_utf8_target = cBOOL(utf8_target);
688 reginfo->info_aux = NULL;
689 reginfo->strbeg = strbeg;
690 reginfo->strend = strend;
691 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
693 /* not actually used within intuit, but zero for safety anyway */
694 reginfo->poscache_maxiter = 0;
697 if (!prog->check_utf8 && prog->check_substr)
698 to_utf8_substr(prog);
699 check = prog->check_utf8;
701 if (!prog->check_substr && prog->check_utf8) {
702 if (! to_byte_substr(prog)) {
703 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
706 check = prog->check_substr;
709 /* dump the various substring data */
710 DEBUG_OPTIMISE_MORE_r({
712 for (i=0; i<=2; i++) {
713 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
714 : prog->substrs->data[i].substr);
718 PerlIO_printf(Perl_debug_log,
719 " substrs[%d]: min=%"IVdf" max=%"IVdf" end shift=%"IVdf
720 " useful=%"IVdf" utf8=%d [%s]\n",
722 (IV)prog->substrs->data[i].min_offset,
723 (IV)prog->substrs->data[i].max_offset,
724 (IV)prog->substrs->data[i].end_shift,
731 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
733 /* ml_anch: check after \n?
735 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
736 * with /.*.../, these flags will have been added by the
738 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
739 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
741 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
742 && !(prog->intflags & PREGf_IMPLICIT);
744 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
745 /* we are only allowed to match at BOS or \G */
747 /* trivially reject if there's a BOS anchor and we're not at BOS.
749 * Note that we don't try to do a similar quick reject for
750 * \G, since generally the caller will have calculated strpos
751 * based on pos() and gofs, so the string is already correctly
752 * anchored by definition; and handling the exceptions would
753 * be too fiddly (e.g. REXEC_IGNOREPOS).
755 if ( strpos != strbeg
756 && (prog->intflags & PREGf_ANCH_SBOL))
758 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
759 " Not at start...\n"));
763 /* in the presence of an anchor, the anchored (relative to the
764 * start of the regex) substr must also be anchored relative
765 * to strpos. So quickly reject if substr isn't found there.
766 * This works for \G too, because the caller will already have
767 * subtracted gofs from pos, and gofs is the offset from the
768 * \G to the start of the regex. For example, in /.abc\Gdef/,
769 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
770 * caller will have set strpos=pos()-4; we look for the substr
771 * at position pos()-4+1, which lines up with the "a" */
773 if (prog->check_offset_min == prog->check_offset_max
774 && !(prog->intflags & PREGf_CANY_SEEN))
776 /* Substring at constant offset from beg-of-str... */
777 SSize_t slen = SvCUR(check);
778 char *s = HOP3c(strpos, prog->check_offset_min, strend);
780 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
781 " Looking for check substr at fixed offset %"IVdf"...\n",
782 (IV)prog->check_offset_min));
785 /* In this case, the regex is anchored at the end too.
786 * Unless it's a multiline match, the lengths must match
787 * exactly, give or take a \n. NB: slen >= 1 since
788 * the last char of check is \n */
790 && ( strend - s > slen
791 || strend - s < slen - 1
792 || (strend - s == slen && strend[-1] != '\n')))
794 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
795 " String too long...\n"));
798 /* Now should match s[0..slen-2] */
801 if (slen && (*SvPVX_const(check) != *s
802 || (slen > 1 && memNE(SvPVX_const(check), s, slen))))
804 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
805 " String not equal...\n"));
810 goto success_at_start;
815 end_shift = prog->check_end_shift;
817 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
819 Perl_croak(aTHX_ "panic: end_shift: %"IVdf" pattern:\n%s\n ",
820 (IV)end_shift, RX_PRECOMP(prog));
825 /* This is the (re)entry point of the main loop in this function.
826 * The goal of this loop is to:
827 * 1) find the "check" substring in the region rx_origin..strend
828 * (adjusted by start_shift / end_shift). If not found, reject
830 * 2) If it exists, look for the "other" substr too if defined; for
831 * example, if the check substr maps to the anchored substr, then
832 * check the floating substr, and vice-versa. If not found, go
833 * back to (1) with rx_origin suitably incremented.
834 * 3) If we find an rx_origin position that doesn't contradict
835 * either of the substrings, then check the possible additional
836 * constraints on rx_origin of /^.../m or a known start class.
837 * If these fail, then depending on which constraints fail, jump
838 * back to here, or to various other re-entry points further along
839 * that skip some of the first steps.
840 * 4) If we pass all those tests, update the BmUSEFUL() count on the
841 * substring. If the start position was determined to be at the
842 * beginning of the string - so, not rejected, but not optimised,
843 * since we have to run regmatch from position 0 - decrement the
844 * BmUSEFUL() count. Otherwise increment it.
848 /* first, look for the 'check' substring */
854 DEBUG_OPTIMISE_MORE_r({
855 PerlIO_printf(Perl_debug_log,
856 " At restart: rx_origin=%"IVdf" Check offset min: %"IVdf
857 " Start shift: %"IVdf" End shift %"IVdf
858 " Real end Shift: %"IVdf"\n",
859 (IV)(rx_origin - strbeg),
860 (IV)prog->check_offset_min,
863 (IV)prog->check_end_shift);
866 if (prog->intflags & PREGf_CANY_SEEN) {
867 start_point= (U8*)(rx_origin + start_shift);
868 end_point= (U8*)(strend - end_shift);
869 if (start_point > end_point)
872 end_point = HOP3(strend, -end_shift, strbeg);
873 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
879 /* If the regex is absolutely anchored to either the start of the
880 * string (SBOL) or to pos() (ANCH_GPOS), then
881 * check_offset_max represents an upper bound on the string where
882 * the substr could start. For the ANCH_GPOS case, we assume that
883 * the caller of intuit will have already set strpos to
884 * pos()-gofs, so in this case strpos + offset_max will still be
885 * an upper bound on the substr.
888 && prog->intflags & PREGf_ANCH
889 && prog->check_offset_max != SSize_t_MAX)
891 SSize_t len = SvCUR(check) - !!SvTAIL(check);
892 const char * const anchor =
893 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
895 /* do a bytes rather than chars comparison. It's conservative;
896 * so it skips doing the HOP if the result can't possibly end
897 * up earlier than the old value of end_point.
899 if ((char*)end_point - anchor > prog->check_offset_max) {
900 end_point = HOP3lim((U8*)anchor,
901 prog->check_offset_max,
907 check_at = fbm_instr( start_point, end_point,
908 check, multiline ? FBMrf_MULTILINE : 0);
910 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
911 " doing 'check' fbm scan, [%"IVdf"..%"IVdf"] gave %"IVdf"\n",
912 (IV)((char*)start_point - strbeg),
913 (IV)((char*)end_point - strbeg),
914 (IV)(check_at ? check_at - strbeg : -1)
917 /* Update the count-of-usability, remove useless subpatterns,
921 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
922 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
923 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s%s",
924 (check_at ? "Found" : "Did not find"),
925 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
926 ? "anchored" : "floating"),
929 (check_at ? " at offset " : "...\n") );
934 /* set rx_origin to the minimum position where the regex could start
935 * matching, given the constraint of the just-matched check substring.
936 * But don't set it lower than previously.
939 if (check_at - rx_origin > prog->check_offset_max)
940 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
941 /* Finish the diagnostic message */
942 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
943 "%ld (rx_origin now %"IVdf")...\n",
944 (long)(check_at - strbeg),
945 (IV)(rx_origin - strbeg)
950 /* now look for the 'other' substring if defined */
952 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
953 : prog->substrs->data[other_ix].substr)
955 /* Take into account the "other" substring. */
959 struct reg_substr_datum *other;
962 other = &prog->substrs->data[other_ix];
964 /* if "other" is anchored:
965 * we've previously found a floating substr starting at check_at.
966 * This means that the regex origin must lie somewhere
967 * between min (rx_origin): HOP3(check_at, -check_offset_max)
968 * and max: HOP3(check_at, -check_offset_min)
969 * (except that min will be >= strpos)
970 * So the fixed substr must lie somewhere between
971 * HOP3(min, anchored_offset)
972 * HOP3(max, anchored_offset) + SvCUR(substr)
975 /* if "other" is floating
976 * Calculate last1, the absolute latest point where the
977 * floating substr could start in the string, ignoring any
978 * constraints from the earlier fixed match. It is calculated
981 * strend - prog->minlen (in chars) is the absolute latest
982 * position within the string where the origin of the regex
983 * could appear. The latest start point for the floating
984 * substr is float_min_offset(*) on from the start of the
985 * regex. last1 simply combines thee two offsets.
987 * (*) You might think the latest start point should be
988 * float_max_offset from the regex origin, and technically
989 * you'd be correct. However, consider
991 * Here, float min, max are 3,5 and minlen is 7.
992 * This can match either
996 * In the first case, the regex matches minlen chars; in the
997 * second, minlen+1, in the third, minlen+2.
998 * In the first case, the floating offset is 3 (which equals
999 * float_min), in the second, 4, and in the third, 5 (which
1000 * equals float_max). In all cases, the floating string bcd
1001 * can never start more than 4 chars from the end of the
1002 * string, which equals minlen - float_min. As the substring
1003 * starts to match more than float_min from the start of the
1004 * regex, it makes the regex match more than minlen chars,
1005 * and the two cancel each other out. So we can always use
1006 * float_min - minlen, rather than float_max - minlen for the
1007 * latest position in the string.
1009 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1010 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1013 assert(prog->minlen >= other->min_offset);
1014 last1 = HOP3c(strend,
1015 other->min_offset - prog->minlen, strbeg);
1017 if (other_ix) {/* i.e. if (other-is-float) */
1018 /* last is the latest point where the floating substr could
1019 * start, *given* any constraints from the earlier fixed
1020 * match. This constraint is that the floating string starts
1021 * <= float_max_offset chars from the regex origin (rx_origin).
1022 * If this value is less than last1, use it instead.
1024 assert(rx_origin <= last1);
1026 /* this condition handles the offset==infinity case, and
1027 * is a short-cut otherwise. Although it's comparing a
1028 * byte offset to a char length, it does so in a safe way,
1029 * since 1 char always occupies 1 or more bytes,
1030 * so if a string range is (last1 - rx_origin) bytes,
1031 * it will be less than or equal to (last1 - rx_origin)
1032 * chars; meaning it errs towards doing the accurate HOP3
1033 * rather than just using last1 as a short-cut */
1034 (last1 - rx_origin) < other->max_offset
1036 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1039 assert(strpos + start_shift <= check_at);
1040 last = HOP4c(check_at, other->min_offset - start_shift,
1044 s = HOP3c(rx_origin, other->min_offset, strend);
1045 if (s < other_last) /* These positions already checked */
1048 must = utf8_target ? other->utf8_substr : other->substr;
1049 assert(SvPOK(must));
1052 char *to = last + SvCUR(must) - (SvTAIL(must)!=0);
1056 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1057 " skipping 'other' fbm scan: %"IVdf" > %"IVdf"\n",
1058 (IV)(from - strbeg),
1064 (unsigned char*)from,
1067 multiline ? FBMrf_MULTILINE : 0
1069 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1070 " doing 'other' fbm scan, [%"IVdf"..%"IVdf"] gave %"IVdf"\n",
1071 (IV)(from - strbeg),
1073 (IV)(s ? s - strbeg : -1)
1079 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1080 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1081 PerlIO_printf(Perl_debug_log, " %s %s substr %s%s",
1082 s ? "Found" : "Contradicts",
1083 other_ix ? "floating" : "anchored",
1084 quoted, RE_SV_TAIL(must));
1089 /* last1 is latest possible substr location. If we didn't
1090 * find it before there, we never will */
1091 if (last >= last1) {
1092 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1093 "; giving up...\n"));
1097 /* try to find the check substr again at a later
1098 * position. Maybe next time we'll find the "other" substr
1100 other_last = HOP3c(last, 1, strend) /* highest failure */;
1102 other_ix /* i.e. if other-is-float */
1103 ? HOP3c(rx_origin, 1, strend)
1104 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1105 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1106 "; about to retry %s at offset %ld (rx_origin now %"IVdf")...\n",
1107 (other_ix ? "floating" : "anchored"),
1108 (long)(HOP3c(check_at, 1, strend) - strbeg),
1109 (IV)(rx_origin - strbeg)
1114 if (other_ix) { /* if (other-is-float) */
1115 /* other_last is set to s, not s+1, since its possible for
1116 * a floating substr to fail first time, then succeed
1117 * second time at the same floating position; e.g.:
1118 * "-AB--AABZ" =~ /\wAB\d*Z/
1119 * The first time round, anchored and float match at
1120 * "-(AB)--AAB(Z)" then fail on the initial \w character
1121 * class. Second time round, they match at "-AB--A(AB)(Z)".
1126 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1127 other_last = HOP3c(s, 1, strend);
1129 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1130 " at offset %ld (rx_origin now %"IVdf")...\n",
1132 (IV)(rx_origin - strbeg)
1138 DEBUG_OPTIMISE_MORE_r(
1139 PerlIO_printf(Perl_debug_log,
1140 " Check-only match: offset min:%"IVdf" max:%"IVdf
1141 " check_at:%"IVdf" rx_origin:%"IVdf" rx_origin-check_at:%"IVdf
1142 " strend:%"IVdf"\n",
1143 (IV)prog->check_offset_min,
1144 (IV)prog->check_offset_max,
1145 (IV)(check_at-strbeg),
1146 (IV)(rx_origin-strbeg),
1147 (IV)(rx_origin-check_at),
1153 postprocess_substr_matches:
1155 /* handle the extra constraint of /^.../m if present */
1157 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1160 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1161 " looking for /^/m anchor"));
1163 /* we have failed the constraint of a \n before rx_origin.
1164 * Find the next \n, if any, even if it's beyond the current
1165 * anchored and/or floating substrings. Whether we should be
1166 * scanning ahead for the next \n or the next substr is debatable.
1167 * On the one hand you'd expect rare substrings to appear less
1168 * often than \n's. On the other hand, searching for \n means
1169 * we're effectively flipping between check_substr and "\n" on each
1170 * iteration as the current "rarest" string candidate, which
1171 * means for example that we'll quickly reject the whole string if
1172 * hasn't got a \n, rather than trying every substr position
1176 s = HOP3c(strend, - prog->minlen, strpos);
1177 if (s <= rx_origin ||
1178 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1180 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1181 " Did not find /%s^%s/m...\n",
1182 PL_colors[0], PL_colors[1]));
1186 /* earliest possible origin is 1 char after the \n.
1187 * (since *rx_origin == '\n', it's safe to ++ here rather than
1188 * HOP(rx_origin, 1)) */
1191 if (prog->substrs->check_ix == 0 /* check is anchored */
1192 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1194 /* Position contradicts check-string; either because
1195 * check was anchored (and thus has no wiggle room),
1196 * or check was float and rx_origin is above the float range */
1197 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1198 " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
1199 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1203 /* if we get here, the check substr must have been float,
1204 * is in range, and we may or may not have had an anchored
1205 * "other" substr which still contradicts */
1206 assert(prog->substrs->check_ix); /* check is float */
1208 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1209 /* whoops, the anchored "other" substr exists, so we still
1210 * contradict. On the other hand, the float "check" substr
1211 * didn't contradict, so just retry the anchored "other"
1213 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1214 " Found /%s^%s/m, rescanning for anchored from offset %ld (rx_origin now %"IVdf")...\n",
1215 PL_colors[0], PL_colors[1],
1216 (long)(rx_origin - strbeg + prog->anchored_offset),
1217 (long)(rx_origin - strbeg)
1219 goto do_other_substr;
1222 /* success: we don't contradict the found floating substring
1223 * (and there's no anchored substr). */
1224 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1225 " Found /%s^%s/m with rx_origin %ld...\n",
1226 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1229 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1230 " (multiline anchor test skipped)\n"));
1236 /* if we have a starting character class, then test that extra constraint.
1237 * (trie stclasses are too expensive to use here, we are better off to
1238 * leave it to regmatch itself) */
1240 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1241 const U8* const str = (U8*)STRING(progi->regstclass);
1243 /* XXX this value could be pre-computed */
1244 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1245 ? (reginfo->is_utf8_pat
1246 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1247 : STR_LEN(progi->regstclass))
1251 /* latest pos that a matching float substr constrains rx start to */
1252 char *rx_max_float = NULL;
1254 /* if the current rx_origin is anchored, either by satisfying an
1255 * anchored substring constraint, or a /^.../m constraint, then we
1256 * can reject the current origin if the start class isn't found
1257 * at the current position. If we have a float-only match, then
1258 * rx_origin is constrained to a range; so look for the start class
1259 * in that range. if neither, then look for the start class in the
1260 * whole rest of the string */
1262 /* XXX DAPM it's not clear what the minlen test is for, and why
1263 * it's not used in the floating case. Nothing in the test suite
1264 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1265 * Here are some old comments, which may or may not be correct:
1267 * minlen == 0 is possible if regstclass is \b or \B,
1268 * and the fixed substr is ''$.
1269 * Since minlen is already taken into account, rx_origin+1 is
1270 * before strend; accidentally, minlen >= 1 guaranties no false
1271 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1272 * 0) below assumes that regstclass does not come from lookahead...
1273 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1274 * This leaves EXACTF-ish only, which are dealt with in
1278 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1279 endpos= HOP3c(rx_origin, (prog->minlen ? cl_l : 0), strend);
1280 else if (prog->float_substr || prog->float_utf8) {
1281 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1282 endpos= HOP3c(rx_max_float, cl_l, strend);
1287 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1288 " looking for class: start_shift: %"IVdf" check_at: %"IVdf
1289 " rx_origin: %"IVdf" endpos: %"IVdf"\n",
1290 (IV)start_shift, (IV)(check_at - strbeg),
1291 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1293 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1296 if (endpos == strend) {
1297 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1298 " Could not match STCLASS...\n") );
1301 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1302 " This position contradicts STCLASS...\n") );
1303 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1304 && !(prog->intflags & PREGf_IMPLICIT))
1307 /* Contradict one of substrings */
1308 if (prog->anchored_substr || prog->anchored_utf8) {
1309 if (prog->substrs->check_ix == 1) { /* check is float */
1310 /* Have both, check_string is floating */
1311 assert(rx_origin + start_shift <= check_at);
1312 if (rx_origin + start_shift != check_at) {
1313 /* not at latest position float substr could match:
1314 * Recheck anchored substring, but not floating.
1315 * The condition above is in bytes rather than
1316 * chars for efficiency. It's conservative, in
1317 * that it errs on the side of doing 'goto
1318 * do_other_substr'. In this case, at worst,
1319 * an extra anchored search may get done, but in
1320 * practice the extra fbm_instr() is likely to
1321 * get skipped anyway. */
1322 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1323 " about to retry anchored at offset %ld (rx_origin now %"IVdf")...\n",
1324 (long)(other_last - strbeg),
1325 (IV)(rx_origin - strbeg)
1327 goto do_other_substr;
1335 /* In the presence of ml_anch, we might be able to
1336 * find another \n without breaking the current float
1339 /* strictly speaking this should be HOP3c(..., 1, ...),
1340 * but since we goto a block of code that's going to
1341 * search for the next \n if any, its safe here */
1343 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1344 " about to look for /%s^%s/m starting at rx_origin %ld...\n",
1345 PL_colors[0], PL_colors[1],
1346 (long)(rx_origin - strbeg)) );
1347 goto postprocess_substr_matches;
1350 /* strictly speaking this can never be true; but might
1351 * be if we ever allow intuit without substrings */
1352 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1355 rx_origin = rx_max_float;
1358 /* at this point, any matching substrings have been
1359 * contradicted. Start again... */
1361 rx_origin = HOP3c(rx_origin, 1, strend);
1363 /* uses bytes rather than char calculations for efficiency.
1364 * It's conservative: it errs on the side of doing 'goto restart',
1365 * where there is code that does a proper char-based test */
1366 if (rx_origin + start_shift + end_shift > strend) {
1367 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1368 " Could not match STCLASS...\n") );
1371 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
1372 " about to look for %s substr starting at offset %ld (rx_origin now %"IVdf")...\n",
1373 (prog->substrs->check_ix ? "floating" : "anchored"),
1374 (long)(rx_origin + start_shift - strbeg),
1375 (IV)(rx_origin - strbeg)
1382 if (rx_origin != s) {
1383 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1384 " By STCLASS: moving %ld --> %ld\n",
1385 (long)(rx_origin - strbeg), (long)(s - strbeg))
1389 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1390 " Does not contradict STCLASS...\n");
1395 /* Decide whether using the substrings helped */
1397 if (rx_origin != strpos) {
1398 /* Fixed substring is found far enough so that the match
1399 cannot start at strpos. */
1401 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " try at offset...\n"));
1402 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1405 /* The found rx_origin position does not prohibit matching at
1406 * strpos, so calling intuit didn't gain us anything. Decrement
1407 * the BmUSEFUL() count on the check substring, and if we reach
1409 if (!(prog->intflags & PREGf_NAUGHTY)
1411 prog->check_utf8 /* Could be deleted already */
1412 && --BmUSEFUL(prog->check_utf8) < 0
1413 && (prog->check_utf8 == prog->float_utf8)
1415 prog->check_substr /* Could be deleted already */
1416 && --BmUSEFUL(prog->check_substr) < 0
1417 && (prog->check_substr == prog->float_substr)
1420 /* If flags & SOMETHING - do not do it many times on the same match */
1421 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, " ... Disabling check substring...\n"));
1422 /* XXX Does the destruction order has to change with utf8_target? */
1423 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1424 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1425 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1426 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1427 check = NULL; /* abort */
1428 /* XXXX This is a remnant of the old implementation. It
1429 looks wasteful, since now INTUIT can use many
1430 other heuristics. */
1431 prog->extflags &= ~RXf_USE_INTUIT;
1435 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
1436 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1437 PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );
1441 fail_finish: /* Substring not found */
1442 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1443 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1445 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch rejected by optimizer%s\n",
1446 PL_colors[4], PL_colors[5]));
1451 #define DECL_TRIE_TYPE(scan) \
1452 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1453 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1454 trie_utf8l, trie_flu8 } \
1455 trie_type = ((scan->flags == EXACT) \
1456 ? (utf8_target ? trie_utf8 : trie_plain) \
1457 : (scan->flags == EXACTL) \
1458 ? (utf8_target ? trie_utf8l : trie_plain) \
1459 : (scan->flags == EXACTFA) \
1461 ? trie_utf8_exactfa_fold \
1462 : trie_latin_utf8_exactfa_fold) \
1463 : (scan->flags == EXACTFLU8 \
1467 : trie_latin_utf8_fold)))
1469 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1472 U8 flags = FOLD_FLAGS_FULL; \
1473 switch (trie_type) { \
1475 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1476 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1477 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1479 goto do_trie_utf8_fold; \
1480 case trie_utf8_exactfa_fold: \
1481 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1483 case trie_utf8_fold: \
1484 do_trie_utf8_fold: \
1485 if ( foldlen>0 ) { \
1486 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1491 uvc = _to_utf8_fold_flags( (const U8*) uc, foldbuf, &foldlen, flags); \
1492 len = UTF8SKIP(uc); \
1493 skiplen = UNISKIP( uvc ); \
1494 foldlen -= skiplen; \
1495 uscan = foldbuf + skiplen; \
1498 case trie_latin_utf8_exactfa_fold: \
1499 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1501 case trie_latin_utf8_fold: \
1502 if ( foldlen>0 ) { \
1503 uvc = utf8n_to_uvchr( (const U8*) uscan, UTF8_MAXLEN, &len, uniflags ); \
1509 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1510 skiplen = UNISKIP( uvc ); \
1511 foldlen -= skiplen; \
1512 uscan = foldbuf + skiplen; \
1516 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1517 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1518 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc + UTF8SKIP(uc)); \
1522 uvc = utf8n_to_uvchr( (const U8*) uc, UTF8_MAXLEN, &len, uniflags ); \
1529 charid = trie->charmap[ uvc ]; \
1533 if (widecharmap) { \
1534 SV** const svpp = hv_fetch(widecharmap, \
1535 (char*)&uvc, sizeof(UV), 0); \
1537 charid = (U16)SvIV(*svpp); \
1542 #define DUMP_EXEC_POS(li,s,doutf8) \
1543 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1546 #define REXEC_FBC_EXACTISH_SCAN(COND) \
1550 && (ln == 1 || folder(s, pat_string, ln)) \
1551 && (reginfo->intuit || regtry(reginfo, &s)) )\
1557 #define REXEC_FBC_UTF8_SCAN(CODE) \
1559 while (s < strend) { \
1565 #define REXEC_FBC_SCAN(CODE) \
1567 while (s < strend) { \
1573 #define REXEC_FBC_UTF8_CLASS_SCAN(COND) \
1574 REXEC_FBC_UTF8_SCAN( /* Loops while (s < strend) */ \
1576 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1585 #define REXEC_FBC_CLASS_SCAN(COND) \
1586 REXEC_FBC_SCAN( /* Loops while (s < strend) */ \
1588 if (tmp && (reginfo->intuit || regtry(reginfo, &s))) \
1597 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1598 if (utf8_target) { \
1599 REXEC_FBC_UTF8_CLASS_SCAN(CONDUTF8); \
1602 REXEC_FBC_CLASS_SCAN(COND); \
1605 /* The three macros below are slightly different versions of the same logic.
1607 * The first is for /a and /aa when the target string is UTF-8. This can only
1608 * match ascii, but it must advance based on UTF-8. The other two handle the
1609 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1610 * for the boundary (or non-boundary) between a word and non-word character.
1611 * The utf8 and non-utf8 cases have the same logic, but the details must be
1612 * different. Find the "wordness" of the character just prior to this one, and
1613 * compare it with the wordness of this one. If they differ, we have a
1614 * boundary. At the beginning of the string, pretend that the previous
1615 * character was a new-line.
1617 * All these macros uncleanly have side-effects with each other and outside
1618 * variables. So far it's been too much trouble to clean-up
1620 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1621 * a word character or not.
1622 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1624 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1626 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1627 * are looking for a boundary or for a non-boundary. If we are looking for a
1628 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1629 * see if this tentative match actually works, and if so, to quit the loop
1630 * here. And vice-versa if we are looking for a non-boundary.
1632 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1633 * REXEC_FBC_UTF8_SCAN loops is a loop invariant, a bool giving the return of
1634 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1635 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1636 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1637 * complement. But in that branch we complement tmp, meaning that at the
1638 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1639 * which means at the top of the loop in the next iteration, it is
1640 * TEST_NON_UTF8(s-1) */
1641 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1642 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1643 tmp = TEST_NON_UTF8(tmp); \
1644 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1645 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1647 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1654 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1655 * TEST_UTF8 is a macro that for the same input code points returns identically
1656 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1657 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1658 if (s == reginfo->strbeg) { \
1661 else { /* Back-up to the start of the previous character */ \
1662 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1663 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1664 0, UTF8_ALLOW_DEFAULT); \
1666 tmp = TEST_UV(tmp); \
1667 LOAD_UTF8_CHARCLASS_ALNUM(); \
1668 REXEC_FBC_UTF8_SCAN( /* advances s while s < strend */ \
1669 if (tmp == ! (TEST_UTF8((U8 *) s))) { \
1678 /* Like the above two macros. UTF8_CODE is the complete code for handling
1679 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1681 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1682 if (utf8_target) { \
1685 else { /* Not utf8 */ \
1686 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1687 tmp = TEST_NON_UTF8(tmp); \
1688 REXEC_FBC_SCAN( /* advances s while s < strend */ \
1689 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1698 /* Here, things have been set up by the previous code so that tmp is the \
1699 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1700 * utf8ness of the target). We also have to check if this matches against \
1701 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1702 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1704 if (tmp == ! TEST_NON_UTF8('\n')) { \
1711 /* This is the macro to use when we want to see if something that looks like it
1712 * could match, actually does, and if so exits the loop */
1713 #define REXEC_FBC_TRYIT \
1714 if ((reginfo->intuit || regtry(reginfo, &s))) \
1717 /* The only difference between the BOUND and NBOUND cases is that
1718 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1719 * NBOUND. This is accomplished by passing it as either the if or else clause,
1720 * with the other one being empty (PLACEHOLDER is defined as empty).
1722 * The TEST_FOO parameters are for operating on different forms of input, but
1723 * all should be ones that return identically for the same underlying code
1725 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1727 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1728 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1730 #define FBC_BOUND_A(TEST_NON_UTF8) \
1732 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1733 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1735 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1737 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1738 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1740 #define FBC_NBOUND_A(TEST_NON_UTF8) \
1742 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1743 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1745 /* Takes a pointer to an inversion list, a pointer to its corresponding
1746 * inversion map, and a code point, and returns the code point's value
1747 * according to the two arrays. It assumes that all code points have a value.
1748 * This is used as the base macro for macros for particular properties */
1749 #define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \
1750 invmap[_invlist_search(invlist, cp)]
1752 /* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
1753 * of a code point, returning the value for the first code point in the string.
1754 * And it takes the particular macro name that finds the desired value given a
1755 * code point. Merely convert the UTF-8 to code point and call the cp macro */
1756 #define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \
1757 (__ASSERT_(pos < strend) \
1758 /* Note assumes is valid UTF-8 */ \
1759 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))
1761 /* Returns the GCB value for the input code point */
1762 #define getGCB_VAL_CP(cp) \
1763 _generic_GET_BREAK_VAL_CP( \
1765 Grapheme_Cluster_Break_invmap, \
1768 /* Returns the GCB value for the first code point in the UTF-8 encoded string
1769 * bounded by pos and strend */
1770 #define getGCB_VAL_UTF8(pos, strend) \
1771 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)
1774 /* Returns the SB value for the input code point */
1775 #define getSB_VAL_CP(cp) \
1776 _generic_GET_BREAK_VAL_CP( \
1778 Sentence_Break_invmap, \
1781 /* Returns the SB value for the first code point in the UTF-8 encoded string
1782 * bounded by pos and strend */
1783 #define getSB_VAL_UTF8(pos, strend) \
1784 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)
1786 /* Returns the WB value for the input code point */
1787 #define getWB_VAL_CP(cp) \
1788 _generic_GET_BREAK_VAL_CP( \
1790 Word_Break_invmap, \
1793 /* Returns the WB value for the first code point in the UTF-8 encoded string
1794 * bounded by pos and strend */
1795 #define getWB_VAL_UTF8(pos, strend) \
1796 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)
1798 /* We know what class REx starts with. Try to find this position... */
1799 /* if reginfo->intuit, its a dryrun */
1800 /* annoyingly all the vars in this routine have different names from their counterparts
1801 in regmatch. /grrr */
1803 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
1804 const char *strend, regmatch_info *reginfo)
1807 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
1808 char *pat_string; /* The pattern's exactish string */
1809 char *pat_end; /* ptr to end char of pat_string */
1810 re_fold_t folder; /* Function for computing non-utf8 folds */
1811 const U8 *fold_array; /* array for folding ords < 256 */
1817 I32 tmp = 1; /* Scratch variable? */
1818 const bool utf8_target = reginfo->is_utf8_target;
1819 UV utf8_fold_flags = 0;
1820 const bool is_utf8_pat = reginfo->is_utf8_pat;
1821 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
1822 with a result inverts that result, as 0^1 =
1824 _char_class_number classnum;
1826 RXi_GET_DECL(prog,progi);
1828 PERL_ARGS_ASSERT_FIND_BYCLASS;
1830 /* We know what class it must start with. */
1833 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1837 REXEC_FBC_UTF8_CLASS_SCAN(
1838 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
1841 REXEC_FBC_CLASS_SCAN(REGINCLASS(prog, c, (U8*)s));
1846 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
1853 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
1854 assert(! is_utf8_pat);
1857 if (is_utf8_pat || utf8_target) {
1858 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
1859 goto do_exactf_utf8;
1861 fold_array = PL_fold_latin1; /* Latin1 folds are not affected by */
1862 folder = foldEQ_latin1; /* /a, except the sharp s one which */
1863 goto do_exactf_non_utf8; /* isn't dealt with by these */
1865 case EXACTF: /* This node only generated for non-utf8 patterns */
1866 assert(! is_utf8_pat);
1868 utf8_fold_flags = 0;
1869 goto do_exactf_utf8;
1871 fold_array = PL_fold;
1873 goto do_exactf_non_utf8;
1876 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1877 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
1878 utf8_fold_flags = FOLDEQ_LOCALE;
1879 goto do_exactf_utf8;
1881 fold_array = PL_fold_locale;
1882 folder = foldEQ_locale;
1883 goto do_exactf_non_utf8;
1887 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
1889 goto do_exactf_utf8;
1892 if (! utf8_target) { /* All code points in this node require
1893 UTF-8 to express. */
1896 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
1897 | FOLDEQ_S2_FOLDS_SANE;
1898 goto do_exactf_utf8;
1901 if (is_utf8_pat || utf8_target) {
1902 utf8_fold_flags = is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
1903 goto do_exactf_utf8;
1906 /* Any 'ss' in the pattern should have been replaced by regcomp,
1907 * so we don't have to worry here about this single special case
1908 * in the Latin1 range */
1909 fold_array = PL_fold_latin1;
1910 folder = foldEQ_latin1;
1914 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
1915 are no glitches with fold-length differences
1916 between the target string and pattern */
1918 /* The idea in the non-utf8 EXACTF* cases is to first find the
1919 * first character of the EXACTF* node and then, if necessary,
1920 * case-insensitively compare the full text of the node. c1 is the
1921 * first character. c2 is its fold. This logic will not work for
1922 * Unicode semantics and the german sharp ss, which hence should
1923 * not be compiled into a node that gets here. */
1924 pat_string = STRING(c);
1925 ln = STR_LEN(c); /* length to match in octets/bytes */
1927 /* We know that we have to match at least 'ln' bytes (which is the
1928 * same as characters, since not utf8). If we have to match 3
1929 * characters, and there are only 2 availabe, we know without
1930 * trying that it will fail; so don't start a match past the
1931 * required minimum number from the far end */
1932 e = HOP3c(strend, -((SSize_t)ln), s);
1934 if (reginfo->intuit && e < s) {
1935 e = s; /* Due to minlen logic of intuit() */
1939 c2 = fold_array[c1];
1940 if (c1 == c2) { /* If char and fold are the same */
1941 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1);
1944 REXEC_FBC_EXACTISH_SCAN(*(U8*)s == c1 || *(U8*)s == c2);
1952 /* If one of the operands is in utf8, we can't use the simpler folding
1953 * above, due to the fact that many different characters can have the
1954 * same fold, or portion of a fold, or different- length fold */
1955 pat_string = STRING(c);
1956 ln = STR_LEN(c); /* length to match in octets/bytes */
1957 pat_end = pat_string + ln;
1958 lnc = is_utf8_pat /* length to match in characters */
1959 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
1962 /* We have 'lnc' characters to match in the pattern, but because of
1963 * multi-character folding, each character in the target can match
1964 * up to 3 characters (Unicode guarantees it will never exceed
1965 * this) if it is utf8-encoded; and up to 2 if not (based on the
1966 * fact that the Latin 1 folds are already determined, and the
1967 * only multi-char fold in that range is the sharp-s folding to
1968 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
1969 * string character. Adjust lnc accordingly, rounding up, so that
1970 * if we need to match at least 4+1/3 chars, that really is 5. */
1971 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
1972 lnc = (lnc + expansion - 1) / expansion;
1974 /* As in the non-UTF8 case, if we have to match 3 characters, and
1975 * only 2 are left, it's guaranteed to fail, so don't start a
1976 * match that would require us to go beyond the end of the string
1978 e = HOP3c(strend, -((SSize_t)lnc), s);
1980 if (reginfo->intuit && e < s) {
1981 e = s; /* Due to minlen logic of intuit() */
1984 /* XXX Note that we could recalculate e to stop the loop earlier,
1985 * as the worst case expansion above will rarely be met, and as we
1986 * go along we would usually find that e moves further to the left.
1987 * This would happen only after we reached the point in the loop
1988 * where if there were no expansion we should fail. Unclear if
1989 * worth the expense */
1992 char *my_strend= (char *)strend;
1993 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
1994 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
1995 && (reginfo->intuit || regtry(reginfo, &s)) )
1999 s += (utf8_target) ? UTF8SKIP(s) : 1;
2005 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2006 if (FLAGS(c) != TRADITIONAL_BOUND) {
2007 if (! IN_UTF8_CTYPE_LOCALE) {
2008 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2009 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2014 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
2018 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2019 if (FLAGS(c) != TRADITIONAL_BOUND) {
2020 if (! IN_UTF8_CTYPE_LOCALE) {
2021 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2022 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2027 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8);
2030 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2032 assert(FLAGS(c) == TRADITIONAL_BOUND);
2034 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
2037 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2039 assert(FLAGS(c) == TRADITIONAL_BOUND);
2041 FBC_BOUND_A(isWORDCHAR_A);
2044 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2046 assert(FLAGS(c) == TRADITIONAL_BOUND);
2048 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8);
2051 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2053 assert(FLAGS(c) == TRADITIONAL_BOUND);
2055 FBC_NBOUND_A(isWORDCHAR_A);
2059 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
2060 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
2071 switch((bound_type) FLAGS(c)) {
2072 case TRADITIONAL_BOUND:
2073 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8);
2076 if (s == reginfo->strbeg) { /* GCB always matches at begin and
2078 if (to_complement ^ cBOOL(reginfo->intuit
2079 || regtry(reginfo, &s)))
2083 s += (utf8_target) ? UTF8SKIP(s) : 1;
2087 GCB_enum before = getGCB_VAL_UTF8(
2089 (U8*)(reginfo->strbeg)),
2090 (U8*) reginfo->strend);
2091 while (s < strend) {
2092 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
2093 (U8*) reginfo->strend);
2094 if (to_complement ^ isGCB(before, after)) {
2095 if (reginfo->intuit || regtry(reginfo, &s)) {
2103 else { /* Not utf8. Everything is a GCB except between CR and
2105 while (s < strend) {
2106 if (to_complement ^ (UCHARAT(s - 1) != '\r'
2107 || UCHARAT(s) != '\n'))
2109 if (reginfo->intuit || regtry(reginfo, &s)) {
2117 if (to_complement ^ cBOOL(reginfo->intuit || regtry(reginfo, &s))) {
2123 if (s == reginfo->strbeg) { /* SB always matches at beginning */
2125 ^ cBOOL(reginfo->intuit || regtry(reginfo, &s)))
2130 /* Didn't match. Go try at the next position */
2131 s += (utf8_target) ? UTF8SKIP(s) : 1;
2135 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
2137 (U8*)(reginfo->strbeg)),
2138 (U8*) reginfo->strend);
2139 while (s < strend) {
2140 SB_enum after = getSB_VAL_UTF8((U8*) s,
2141 (U8*) reginfo->strend);
2142 if (to_complement ^ isSB(before,
2144 (U8*) reginfo->strbeg,
2146 (U8*) reginfo->strend,
2149 if (reginfo->intuit || regtry(reginfo, &s)) {
2157 else { /* Not utf8. */
2158 SB_enum before = getSB_VAL_CP((U8) *(s -1));
2159 while (s < strend) {
2160 SB_enum after = getSB_VAL_CP((U8) *s);
2161 if (to_complement ^ isSB(before,
2163 (U8*) reginfo->strbeg,
2165 (U8*) reginfo->strend,
2168 if (reginfo->intuit || regtry(reginfo, &s)) {
2177 /* Here are at the final position in the target string. The SB
2178 * value is always true here, so matches, depending on other
2180 if (to_complement ^ cBOOL(reginfo->intuit
2181 || regtry(reginfo, &s)))
2189 if (s == reginfo->strbeg) {
2190 if (to_complement ^ cBOOL(reginfo->intuit
2191 || regtry(reginfo, &s)))
2195 s += (utf8_target) ? UTF8SKIP(s) : 1;
2199 /* We are at a boundary between char_sub_0 and char_sub_1.
2200 * We also keep track of the value for char_sub_-1 as we
2201 * loop through the line. Context may be needed to make a
2202 * determination, and if so, this can save having to
2204 WB_enum previous = WB_UNKNOWN;
2205 WB_enum before = getWB_VAL_UTF8(
2208 (U8*)(reginfo->strbeg)),
2209 (U8*) reginfo->strend);
2210 while (s < strend) {
2211 WB_enum after = getWB_VAL_UTF8((U8*) s,
2212 (U8*) reginfo->strend);
2213 if (to_complement ^ isWB(previous,
2216 (U8*) reginfo->strbeg,
2218 (U8*) reginfo->strend,
2221 if (reginfo->intuit || regtry(reginfo, &s)) {
2230 else { /* Not utf8. */
2231 WB_enum previous = WB_UNKNOWN;
2232 WB_enum before = getWB_VAL_CP((U8) *(s -1));
2233 while (s < strend) {
2234 WB_enum after = getWB_VAL_CP((U8) *s);
2235 if (to_complement ^ isWB(previous,
2238 (U8*) reginfo->strbeg,
2240 (U8*) reginfo->strend,
2243 if (reginfo->intuit || regtry(reginfo, &s)) {
2253 if (to_complement ^ cBOOL(reginfo->intuit
2254 || regtry(reginfo, &s)))
2264 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2265 is_LNBREAK_latin1_safe(s, strend)
2269 /* The argument to all the POSIX node types is the class number to pass to
2270 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2277 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2278 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s)),
2279 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
2294 /* The complement of something that matches only ASCII matches all
2295 * non-ASCII, plus everything in ASCII that isn't in the class. */
2296 REXEC_FBC_UTF8_CLASS_SCAN(! isASCII_utf8(s)
2297 || ! _generic_isCC_A(*s, FLAGS(c)));
2306 /* Don't need to worry about utf8, as it can match only a single
2307 * byte invariant character. */
2308 REXEC_FBC_CLASS_SCAN(
2309 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
2317 if (! utf8_target) {
2318 REXEC_FBC_CLASS_SCAN(to_complement ^ cBOOL(_generic_isCC(*s,
2324 classnum = (_char_class_number) FLAGS(c);
2325 if (classnum < _FIRST_NON_SWASH_CC) {
2326 while (s < strend) {
2328 /* We avoid loading in the swash as long as possible, but
2329 * should we have to, we jump to a separate loop. This
2330 * extra 'if' statement is what keeps this code from being
2331 * just a call to REXEC_FBC_UTF8_CLASS_SCAN() */
2332 if (UTF8_IS_ABOVE_LATIN1(*s)) {
2333 goto found_above_latin1;
2335 if ((UTF8_IS_INVARIANT(*s)
2336 && to_complement ^ cBOOL(_generic_isCC((U8) *s,
2338 || (UTF8_IS_DOWNGRADEABLE_START(*s)
2339 && to_complement ^ cBOOL(
2340 _generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*s,
2344 if (tmp && (reginfo->intuit || regtry(reginfo, &s)))
2356 else switch (classnum) { /* These classes are implemented as
2358 case _CC_ENUM_SPACE:
2359 REXEC_FBC_UTF8_CLASS_SCAN(
2360 to_complement ^ cBOOL(isSPACE_utf8(s)));
2363 case _CC_ENUM_BLANK:
2364 REXEC_FBC_UTF8_CLASS_SCAN(
2365 to_complement ^ cBOOL(isBLANK_utf8(s)));
2368 case _CC_ENUM_XDIGIT:
2369 REXEC_FBC_UTF8_CLASS_SCAN(
2370 to_complement ^ cBOOL(isXDIGIT_utf8(s)));
2373 case _CC_ENUM_VERTSPACE:
2374 REXEC_FBC_UTF8_CLASS_SCAN(
2375 to_complement ^ cBOOL(isVERTWS_utf8(s)));
2378 case _CC_ENUM_CNTRL:
2379 REXEC_FBC_UTF8_CLASS_SCAN(
2380 to_complement ^ cBOOL(isCNTRL_utf8(s)));
2384 Perl_croak(aTHX_ "panic: find_byclass() node %d='%s' has an unexpected character class '%d'", OP(c), PL_reg_name[OP(c)], classnum);
2385 NOT_REACHED; /* NOTREACHED */
2390 found_above_latin1: /* Here we have to load a swash to get the result
2391 for the current code point */
2392 if (! PL_utf8_swash_ptrs[classnum]) {
2393 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2394 PL_utf8_swash_ptrs[classnum] =
2395 _core_swash_init("utf8",
2398 PL_XPosix_ptrs[classnum], &flags);
2401 /* This is a copy of the loop above for swash classes, though using the
2402 * FBC macro instead of being expanded out. Since we've loaded the
2403 * swash, we don't have to check for that each time through the loop */
2404 REXEC_FBC_UTF8_CLASS_SCAN(
2405 to_complement ^ cBOOL(_generic_utf8(
2408 swash_fetch(PL_utf8_swash_ptrs[classnum],
2416 /* what trie are we using right now */
2417 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2418 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2419 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2421 const char *last_start = strend - trie->minlen;
2423 const char *real_start = s;
2425 STRLEN maxlen = trie->maxlen;
2427 U8 **points; /* map of where we were in the input string
2428 when reading a given char. For ASCII this
2429 is unnecessary overhead as the relationship
2430 is always 1:1, but for Unicode, especially
2431 case folded Unicode this is not true. */
2432 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2436 GET_RE_DEBUG_FLAGS_DECL;
2438 /* We can't just allocate points here. We need to wrap it in
2439 * an SV so it gets freed properly if there is a croak while
2440 * running the match */
2443 sv_points=newSV(maxlen * sizeof(U8 *));
2444 SvCUR_set(sv_points,
2445 maxlen * sizeof(U8 *));
2446 SvPOK_on(sv_points);
2447 sv_2mortal(sv_points);
2448 points=(U8**)SvPV_nolen(sv_points );
2449 if ( trie_type != trie_utf8_fold
2450 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2453 bitmap=(U8*)trie->bitmap;
2455 bitmap=(U8*)ANYOF_BITMAP(c);
2457 /* this is the Aho-Corasick algorithm modified a touch
2458 to include special handling for long "unknown char" sequences.
2459 The basic idea being that we use AC as long as we are dealing
2460 with a possible matching char, when we encounter an unknown char
2461 (and we have not encountered an accepting state) we scan forward
2462 until we find a legal starting char.
2463 AC matching is basically that of trie matching, except that when
2464 we encounter a failing transition, we fall back to the current
2465 states "fail state", and try the current char again, a process
2466 we repeat until we reach the root state, state 1, or a legal
2467 transition. If we fail on the root state then we can either
2468 terminate if we have reached an accepting state previously, or
2469 restart the entire process from the beginning if we have not.
2472 while (s <= last_start) {
2473 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2481 U8 *uscan = (U8*)NULL;
2482 U8 *leftmost = NULL;
2484 U32 accepted_word= 0;
2488 while ( state && uc <= (U8*)strend ) {
2490 U32 word = aho->states[ state ].wordnum;
2494 DEBUG_TRIE_EXECUTE_r(
2495 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2496 dump_exec_pos( (char *)uc, c, strend, real_start,
2497 (char *)uc, utf8_target );
2498 PerlIO_printf( Perl_debug_log,
2499 " Scanning for legal start char...\n");
2503 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2507 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2513 if (uc >(U8*)last_start) break;
2517 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2518 if (!leftmost || lpos < leftmost) {
2519 DEBUG_r(accepted_word=word);
2525 points[pointpos++ % maxlen]= uc;
2526 if (foldlen || uc < (U8*)strend) {
2527 REXEC_TRIE_READ_CHAR(trie_type, trie,
2529 uscan, len, uvc, charid, foldlen,
2531 DEBUG_TRIE_EXECUTE_r({
2532 dump_exec_pos( (char *)uc, c, strend,
2533 real_start, s, utf8_target);
2534 PerlIO_printf(Perl_debug_log,
2535 " Charid:%3u CP:%4"UVxf" ",
2547 word = aho->states[ state ].wordnum;
2549 base = aho->states[ state ].trans.base;
2551 DEBUG_TRIE_EXECUTE_r({
2553 dump_exec_pos( (char *)uc, c, strend, real_start,
2555 PerlIO_printf( Perl_debug_log,
2556 "%sState: %4"UVxf", word=%"UVxf,
2557 failed ? " Fail transition to " : "",
2558 (UV)state, (UV)word);
2564 ( ((offset = base + charid
2565 - 1 - trie->uniquecharcount)) >= 0)
2566 && ((U32)offset < trie->lasttrans)
2567 && trie->trans[offset].check == state
2568 && (tmp=trie->trans[offset].next))
2570 DEBUG_TRIE_EXECUTE_r(
2571 PerlIO_printf( Perl_debug_log," - legal\n"));
2576 DEBUG_TRIE_EXECUTE_r(
2577 PerlIO_printf( Perl_debug_log," - fail\n"));
2579 state = aho->fail[state];
2583 /* we must be accepting here */
2584 DEBUG_TRIE_EXECUTE_r(
2585 PerlIO_printf( Perl_debug_log," - accepting\n"));
2594 if (!state) state = 1;
2597 if ( aho->states[ state ].wordnum ) {
2598 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2599 if (!leftmost || lpos < leftmost) {
2600 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2605 s = (char*)leftmost;
2606 DEBUG_TRIE_EXECUTE_r({
2608 Perl_debug_log,"Matches word #%"UVxf" at position %"IVdf". Trying full pattern...\n",
2609 (UV)accepted_word, (IV)(s - real_start)
2612 if (reginfo->intuit || regtry(reginfo, &s)) {
2618 DEBUG_TRIE_EXECUTE_r({
2619 PerlIO_printf( Perl_debug_log,"Pattern failed. Looking for new start point...\n");
2622 DEBUG_TRIE_EXECUTE_r(
2623 PerlIO_printf( Perl_debug_log,"No match.\n"));
2632 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
2639 /* set RX_SAVED_COPY, RX_SUBBEG etc.
2640 * flags have same meanings as with regexec_flags() */
2643 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
2650 struct regexp *const prog = ReANY(rx);
2652 if (flags & REXEC_COPY_STR) {
2656 PerlIO_printf(Perl_debug_log,
2657 "Copy on write: regexp capture, type %d\n",
2660 /* Create a new COW SV to share the match string and store
2661 * in saved_copy, unless the current COW SV in saved_copy
2662 * is valid and suitable for our purpose */
2663 if (( prog->saved_copy
2664 && SvIsCOW(prog->saved_copy)
2665 && SvPOKp(prog->saved_copy)
2668 && SvPVX(sv) == SvPVX(prog->saved_copy)))
2670 /* just reuse saved_copy SV */
2671 if (RXp_MATCH_COPIED(prog)) {
2672 Safefree(prog->subbeg);
2673 RXp_MATCH_COPIED_off(prog);
2677 /* create new COW SV to share string */
2678 RX_MATCH_COPY_FREE(rx);
2679 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
2681 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
2682 assert (SvPOKp(prog->saved_copy));
2683 prog->sublen = strend - strbeg;
2684 prog->suboffset = 0;
2685 prog->subcoffset = 0;
2690 SSize_t max = strend - strbeg;
2693 if ( (flags & REXEC_COPY_SKIP_POST)
2694 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2695 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
2696 ) { /* don't copy $' part of string */
2699 /* calculate the right-most part of the string covered
2700 * by a capture. Due to look-ahead, this may be to
2701 * the right of $&, so we have to scan all captures */
2702 while (n <= prog->lastparen) {
2703 if (prog->offs[n].end > max)
2704 max = prog->offs[n].end;
2708 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
2709 ? prog->offs[0].start
2711 assert(max >= 0 && max <= strend - strbeg);
2714 if ( (flags & REXEC_COPY_SKIP_PRE)
2715 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
2716 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
2717 ) { /* don't copy $` part of string */
2720 /* calculate the left-most part of the string covered
2721 * by a capture. Due to look-behind, this may be to
2722 * the left of $&, so we have to scan all captures */
2723 while (min && n <= prog->lastparen) {
2724 if ( prog->offs[n].start != -1
2725 && prog->offs[n].start < min)
2727 min = prog->offs[n].start;
2731 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
2732 && min > prog->offs[0].end
2734 min = prog->offs[0].end;
2738 assert(min >= 0 && min <= max && min <= strend - strbeg);
2741 if (RX_MATCH_COPIED(rx)) {
2742 if (sublen > prog->sublen)
2744 (char*)saferealloc(prog->subbeg, sublen+1);
2747 prog->subbeg = (char*)safemalloc(sublen+1);
2748 Copy(strbeg + min, prog->subbeg, sublen, char);
2749 prog->subbeg[sublen] = '\0';
2750 prog->suboffset = min;
2751 prog->sublen = sublen;
2752 RX_MATCH_COPIED_on(rx);
2754 prog->subcoffset = prog->suboffset;
2755 if (prog->suboffset && utf8_target) {
2756 /* Convert byte offset to chars.
2757 * XXX ideally should only compute this if @-/@+
2758 * has been seen, a la PL_sawampersand ??? */
2760 /* If there's a direct correspondence between the
2761 * string which we're matching and the original SV,
2762 * then we can use the utf8 len cache associated with
2763 * the SV. In particular, it means that under //g,
2764 * sv_pos_b2u() will use the previously cached
2765 * position to speed up working out the new length of
2766 * subcoffset, rather than counting from the start of
2767 * the string each time. This stops
2768 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
2769 * from going quadratic */
2770 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
2771 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
2772 SV_GMAGIC|SV_CONST_RETURN);
2774 prog->subcoffset = utf8_length((U8*)strbeg,
2775 (U8*)(strbeg+prog->suboffset));
2779 RX_MATCH_COPY_FREE(rx);
2780 prog->subbeg = strbeg;
2781 prog->suboffset = 0;
2782 prog->subcoffset = 0;
2783 prog->sublen = strend - strbeg;
2791 - regexec_flags - match a regexp against a string
2794 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
2795 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
2796 /* stringarg: the point in the string at which to begin matching */
2797 /* strend: pointer to null at end of string */
2798 /* strbeg: real beginning of string */
2799 /* minend: end of match must be >= minend bytes after stringarg. */
2800 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
2801 * itself is accessed via the pointers above */
2802 /* data: May be used for some additional optimizations.
2803 Currently unused. */
2804 /* flags: For optimizations. See REXEC_* in regexp.h */
2807 struct regexp *const prog = ReANY(rx);
2811 SSize_t minlen; /* must match at least this many chars */
2812 SSize_t dontbother = 0; /* how many characters not to try at end */
2813 const bool utf8_target = cBOOL(DO_UTF8(sv));
2815 RXi_GET_DECL(prog,progi);
2816 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
2817 regmatch_info *const reginfo = ®info_buf;
2818 regexp_paren_pair *swap = NULL;
2820 GET_RE_DEBUG_FLAGS_DECL;
2822 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
2823 PERL_UNUSED_ARG(data);
2825 /* Be paranoid... */
2827 Perl_croak(aTHX_ "NULL regexp parameter");
2831 debug_start_match(rx, utf8_target, stringarg, strend,
2835 startpos = stringarg;
2837 if (prog->intflags & PREGf_GPOS_SEEN) {
2840 /* set reginfo->ganch, the position where \G can match */
2843 (flags & REXEC_IGNOREPOS)
2844 ? stringarg /* use start pos rather than pos() */
2845 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
2846 /* Defined pos(): */
2847 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
2848 : strbeg; /* pos() not defined; use start of string */
2850 DEBUG_GPOS_r(PerlIO_printf(Perl_debug_log,
2851 "GPOS ganch set to strbeg[%"IVdf"]\n", (IV)(reginfo->ganch - strbeg)));
2853 /* in the presence of \G, we may need to start looking earlier in
2854 * the string than the suggested start point of stringarg:
2855 * if prog->gofs is set, then that's a known, fixed minimum
2858 * /ab|c\G/: gofs = 1
2859 * or if the minimum offset isn't known, then we have to go back
2860 * to the start of the string, e.g. /w+\G/
2863 if (prog->intflags & PREGf_ANCH_GPOS) {
2864 startpos = reginfo->ganch - prog->gofs;
2866 ((flags & REXEC_FAIL_ON_UNDERFLOW) ? stringarg : strbeg))
2868 DEBUG_r(PerlIO_printf(Perl_debug_log,
2869 "fail: ganch-gofs before earliest possible start\n"));
2873 else if (prog->gofs) {
2874 if (startpos - prog->gofs < strbeg)
2877 startpos -= prog->gofs;
2879 else if (prog->intflags & PREGf_GPOS_FLOAT)
2883 minlen = prog->minlen;
2884 if ((startpos + minlen) > strend || startpos < strbeg) {
2885 DEBUG_r(PerlIO_printf(Perl_debug_log,
2886 "Regex match can't succeed, so not even tried\n"));
2890 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
2891 * which will call destuctors to reset PL_regmatch_state, free higher
2892 * PL_regmatch_slabs, and clean up regmatch_info_aux and
2893 * regmatch_info_aux_eval */
2895 oldsave = PL_savestack_ix;
2899 if ((prog->extflags & RXf_USE_INTUIT)
2900 && !(flags & REXEC_CHECKED))
2902 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
2907 if (prog->extflags & RXf_CHECK_ALL) {
2908 /* we can match based purely on the result of INTUIT.
2909 * Set up captures etc just for $& and $-[0]
2910 * (an intuit-only match wont have $1,$2,..) */
2911 assert(!prog->nparens);
2913 /* s/// doesn't like it if $& is earlier than where we asked it to
2914 * start searching (which can happen on something like /.\G/) */
2915 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
2918 /* this should only be possible under \G */
2919 assert(prog->intflags & PREGf_GPOS_SEEN);
2920 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2921 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
2925 /* match via INTUIT shouldn't have any captures.
2926 * Let @-, @+, $^N know */
2927 prog->lastparen = prog->lastcloseparen = 0;
2928 RX_MATCH_UTF8_set(rx, utf8_target);
2929 prog->offs[0].start = s - strbeg;
2930 prog->offs[0].end = utf8_target
2931 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
2932 : s - strbeg + prog->minlenret;
2933 if ( !(flags & REXEC_NOT_FIRST) )
2934 S_reg_set_capture_string(aTHX_ rx,
2936 sv, flags, utf8_target);
2942 multiline = prog->extflags & RXf_PMf_MULTILINE;
2944 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
2945 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
2946 "String too short [regexec_flags]...\n"));
2950 /* Check validity of program. */
2951 if (UCHARAT(progi->program) != REG_MAGIC) {
2952 Perl_croak(aTHX_ "corrupted regexp program");
2955 RX_MATCH_TAINTED_off(rx);
2956 RX_MATCH_UTF8_set(rx, utf8_target);
2958 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
2959 reginfo->intuit = 0;
2960 reginfo->is_utf8_target = cBOOL(utf8_target);
2961 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
2962 reginfo->warned = FALSE;
2963 reginfo->strbeg = strbeg;
2965 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
2966 reginfo->strend = strend;
2967 /* see how far we have to get to not match where we matched before */
2968 reginfo->till = stringarg + minend;
2970 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
2971 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
2972 S_cleanup_regmatch_info_aux has executed (registered by
2973 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
2974 magic belonging to this SV.
2975 Not newSVsv, either, as it does not COW.
2977 reginfo->sv = newSV(0);
2978 SvSetSV_nosteal(reginfo->sv, sv);
2979 SAVEFREESV(reginfo->sv);
2982 /* reserve next 2 or 3 slots in PL_regmatch_state:
2983 * slot N+0: may currently be in use: skip it
2984 * slot N+1: use for regmatch_info_aux struct
2985 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
2986 * slot N+3: ready for use by regmatch()
2990 regmatch_state *old_regmatch_state;
2991 regmatch_slab *old_regmatch_slab;
2992 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
2994 /* on first ever match, allocate first slab */
2995 if (!PL_regmatch_slab) {
2996 Newx(PL_regmatch_slab, 1, regmatch_slab);
2997 PL_regmatch_slab->prev = NULL;
2998 PL_regmatch_slab->next = NULL;
2999 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3002 old_regmatch_state = PL_regmatch_state;
3003 old_regmatch_slab = PL_regmatch_slab;
3005 for (i=0; i <= max; i++) {
3007 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3009 reginfo->info_aux_eval =
3010 reginfo->info_aux->info_aux_eval =
3011 &(PL_regmatch_state->u.info_aux_eval);
3013 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3014 PL_regmatch_state = S_push_slab(aTHX);
3017 /* note initial PL_regmatch_state position; at end of match we'll
3018 * pop back to there and free any higher slabs */
3020 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3021 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
3022 reginfo->info_aux->poscache = NULL;
3024 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
3026 if ((prog->extflags & RXf_EVAL_SEEN))
3027 S_setup_eval_state(aTHX_ reginfo);
3029 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
3032 /* If there is a "must appear" string, look for it. */
3034 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
3035 /* We have to be careful. If the previous successful match
3036 was from this regex we don't want a subsequent partially
3037 successful match to clobber the old results.
3038 So when we detect this possibility we add a swap buffer
3039 to the re, and switch the buffer each match. If we fail,
3040 we switch it back; otherwise we leave it swapped.
3043 /* do we need a save destructor here for eval dies? */
3044 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
3045 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3046 "rex=0x%"UVxf" saving offs: orig=0x%"UVxf" new=0x%"UVxf"\n",
3053 /* Simplest case: anchored match need be tried only once, or with
3054 * MBOL, only at the beginning of each line.
3056 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3057 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3058 * match at the start of the string then it won't match anywhere else
3059 * either; while with /.*.../, if it doesn't match at the beginning,
3060 * the earliest it could match is at the start of the next line */
3062 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3065 if (regtry(reginfo, &s))
3068 if (!(prog->intflags & PREGf_ANCH_MBOL))
3071 /* didn't match at start, try at other newline positions */
3074 dontbother = minlen - 1;
3075 end = HOP3c(strend, -dontbother, strbeg) - 1;
3077 /* skip to next newline */
3079 while (s <= end) { /* note it could be possible to match at the end of the string */
3080 /* NB: newlines are the same in unicode as they are in latin */
3083 if (prog->check_substr || prog->check_utf8) {
3084 /* note that with PREGf_IMPLICIT, intuit can only fail
3085 * or return the start position, so it's of limited utility.
3086 * Nevertheless, I made the decision that the potential for
3087 * quick fail was still worth it - DAPM */
3088 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3092 if (regtry(reginfo, &s))
3096 } /* end anchored search */
3098 if (prog->intflags & PREGf_ANCH_GPOS)
3100 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3101 assert(prog->intflags & PREGf_GPOS_SEEN);
3102 /* For anchored \G, the only position it can match from is
3103 * (ganch-gofs); we already set startpos to this above; if intuit
3104 * moved us on from there, we can't possibly succeed */
3105 assert(startpos == reginfo->ganch - prog->gofs);
3106 if (s == startpos && regtry(reginfo, &s))
3111 /* Messy cases: unanchored match. */
3112 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
3113 /* we have /x+whatever/ */
3114 /* it must be a one character string (XXXX Except is_utf8_pat?) */
3120 if (! prog->anchored_utf8) {
3121 to_utf8_substr(prog);
3123 ch = SvPVX_const(prog->anchored_utf8)[0];
3126 DEBUG_EXECUTE_r( did_match = 1 );
3127 if (regtry(reginfo, &s)) goto got_it;
3129 while (s < strend && *s == ch)
3136 if (! prog->anchored_substr) {
3137 if (! to_byte_substr(prog)) {
3138 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3141 ch = SvPVX_const(prog->anchored_substr)[0];
3144 DEBUG_EXECUTE_r( did_match = 1 );
3145 if (regtry(reginfo, &s)) goto got_it;
3147 while (s < strend && *s == ch)
3152 DEBUG_EXECUTE_r(if (!did_match)
3153 PerlIO_printf(Perl_debug_log,
3154 "Did not find anchored character...\n")
3157 else if (prog->anchored_substr != NULL
3158 || prog->anchored_utf8 != NULL
3159 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
3160 && prog->float_max_offset < strend - s)) {
3165 char *last1; /* Last position checked before */
3169 if (prog->anchored_substr || prog->anchored_utf8) {
3171 if (! prog->anchored_utf8) {
3172 to_utf8_substr(prog);
3174 must = prog->anchored_utf8;
3177 if (! prog->anchored_substr) {
3178 if (! to_byte_substr(prog)) {
3179 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3182 must = prog->anchored_substr;
3184 back_max = back_min = prog->anchored_offset;
3187 if (! prog->float_utf8) {
3188 to_utf8_substr(prog);
3190 must = prog->float_utf8;
3193 if (! prog->float_substr) {
3194 if (! to_byte_substr(prog)) {
3195 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3198 must = prog->float_substr;
3200 back_max = prog->float_max_offset;
3201 back_min = prog->float_min_offset;
3207 last = HOP3c(strend, /* Cannot start after this */
3208 -(SSize_t)(CHR_SVLEN(must)
3209 - (SvTAIL(must) != 0) + back_min), strbeg);
3211 if (s > reginfo->strbeg)
3212 last1 = HOPc(s, -1);
3214 last1 = s - 1; /* bogus */
3216 /* XXXX check_substr already used to find "s", can optimize if
3217 check_substr==must. */
3219 strend = HOPc(strend, -dontbother);
3220 while ( (s <= last) &&
3221 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
3222 (unsigned char*)strend, must,
3223 multiline ? FBMrf_MULTILINE : 0)) ) {
3224 DEBUG_EXECUTE_r( did_match = 1 );
3225 if (HOPc(s, -back_max) > last1) {
3226 last1 = HOPc(s, -back_min);
3227 s = HOPc(s, -back_max);
3230 char * const t = (last1 >= reginfo->strbeg)
3231 ? HOPc(last1, 1) : last1 + 1;
3233 last1 = HOPc(s, -back_min);
3237 while (s <= last1) {
3238 if (regtry(reginfo, &s))
3241 s++; /* to break out of outer loop */
3248 while (s <= last1) {
3249 if (regtry(reginfo, &s))
3255 DEBUG_EXECUTE_r(if (!did_match) {
3256 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
3257 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
3258 PerlIO_printf(Perl_debug_log, "Did not find %s substr %s%s...\n",
3259 ((must == prog->anchored_substr || must == prog->anchored_utf8)
3260 ? "anchored" : "floating"),
3261 quoted, RE_SV_TAIL(must));
3265 else if ( (c = progi->regstclass) ) {
3267 const OPCODE op = OP(progi->regstclass);
3268 /* don't bother with what can't match */
3269 if (PL_regkind[op] != EXACT && op != CANY && PL_regkind[op] != TRIE)
3270 strend = HOPc(strend, -(minlen - 1));
3273 SV * const prop = sv_newmortal();
3274 regprop(prog, prop, c, reginfo, NULL);
3276 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
3278 PerlIO_printf(Perl_debug_log,
3279 "Matching stclass %.*s against %s (%d bytes)\n",
3280 (int)SvCUR(prop), SvPVX_const(prop),
3281 quoted, (int)(strend - s));
3284 if (find_byclass(prog, c, s, strend, reginfo))
3286 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "Contradicts stclass... [regexec_flags]\n"));
3290 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3298 if (! prog->float_utf8) {
3299 to_utf8_substr(prog);
3301 float_real = prog->float_utf8;
3304 if (! prog->float_substr) {
3305 if (! to_byte_substr(prog)) {
3306 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3309 float_real = prog->float_substr;
3312 little = SvPV_const(float_real, len);
3313 if (SvTAIL(float_real)) {
3314 /* This means that float_real contains an artificial \n on
3315 * the end due to the presence of something like this:
3316 * /foo$/ where we can match both "foo" and "foo\n" at the
3317 * end of the string. So we have to compare the end of the
3318 * string first against the float_real without the \n and
3319 * then against the full float_real with the string. We
3320 * have to watch out for cases where the string might be
3321 * smaller than the float_real or the float_real without
3323 char *checkpos= strend - len;
3325 PerlIO_printf(Perl_debug_log,
3326 "%sChecking for float_real.%s\n",
3327 PL_colors[4], PL_colors[5]));
3328 if (checkpos + 1 < strbeg) {
3329 /* can't match, even if we remove the trailing \n
3330 * string is too short to match */
3332 PerlIO_printf(Perl_debug_log,
3333 "%sString shorter than required trailing substring, cannot match.%s\n",
3334 PL_colors[4], PL_colors[5]));
3336 } else if (memEQ(checkpos + 1, little, len - 1)) {
3337 /* can match, the end of the string matches without the
3339 last = checkpos + 1;
3340 } else if (checkpos < strbeg) {
3341 /* cant match, string is too short when the "\n" is
3344 PerlIO_printf(Perl_debug_log,
3345 "%sString does not contain required trailing substring, cannot match.%s\n",
3346 PL_colors[4], PL_colors[5]));
3348 } else if (!multiline) {
3349 /* non multiline match, so compare with the "\n" at the
3350 * end of the string */
3351 if (memEQ(checkpos, little, len)) {
3355 PerlIO_printf(Perl_debug_log,
3356 "%sString does not contain required trailing substring, cannot match.%s\n",
3357 PL_colors[4], PL_colors[5]));
3361 /* multiline match, so we have to search for a place
3362 * where the full string is located */
3368 last = rninstr(s, strend, little, little + len);
3370 last = strend; /* matching "$" */
3373 /* at one point this block contained a comment which was
3374 * probably incorrect, which said that this was a "should not
3375 * happen" case. Even if it was true when it was written I am
3376 * pretty sure it is not anymore, so I have removed the comment
3377 * and replaced it with this one. Yves */
3379 PerlIO_printf(Perl_debug_log,
3380 "%sString does not contain required substring, cannot match.%s\n",
3381 PL_colors[4], PL_colors[5]
3385 dontbother = strend - last + prog->float_min_offset;
3387 if (minlen && (dontbother < minlen))
3388 dontbother = minlen - 1;
3389 strend -= dontbother; /* this one's always in bytes! */
3390 /* We don't know much -- general case. */
3393 if (regtry(reginfo, &s))
3402 if (regtry(reginfo, &s))
3404 } while (s++ < strend);
3412 /* s/// doesn't like it if $& is earlier than where we asked it to
3413 * start searching (which can happen on something like /.\G/) */
3414 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3415 && (prog->offs[0].start < stringarg - strbeg))
3417 /* this should only be possible under \G */
3418 assert(prog->intflags & PREGf_GPOS_SEEN);
3419 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
3420 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3426 PerlIO_printf(Perl_debug_log,
3427 "rex=0x%"UVxf" freeing offs: 0x%"UVxf"\n",
3434 /* clean up; this will trigger destructors that will free all slabs
3435 * above the current one, and cleanup the regmatch_info_aux
3436 * and regmatch_info_aux_eval sructs */
3438 LEAVE_SCOPE(oldsave);
3440 if (RXp_PAREN_NAMES(prog))
3441 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3443 /* make sure $`, $&, $', and $digit will work later */
3444 if ( !(flags & REXEC_NOT_FIRST) )
3445 S_reg_set_capture_string(aTHX_ rx,
3446 strbeg, reginfo->strend,
3447 sv, flags, utf8_target);
3452 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch failed%s\n",
3453 PL_colors[4], PL_colors[5]));
3455 /* clean up; this will trigger destructors that will free all slabs
3456 * above the current one, and cleanup the regmatch_info_aux
3457 * and regmatch_info_aux_eval sructs */
3459 LEAVE_SCOPE(oldsave);
3462 /* we failed :-( roll it back */
3463 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
3464 "rex=0x%"UVxf" rolling back offs: freeing=0x%"UVxf" restoring=0x%"UVxf"\n",
3469 Safefree(prog->offs);
3476 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3477 * Do inc before dec, in case old and new rex are the same */
3478 #define SET_reg_curpm(Re2) \
3479 if (reginfo->info_aux_eval) { \
3480 (void)ReREFCNT_inc(Re2); \
3481 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3482 PM_SETRE((PL_reg_curpm), (Re2)); \
3487 - regtry - try match at specific point
3489 STATIC I32 /* 0 failure, 1 success */
3490 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3493 REGEXP *const rx = reginfo->prog;
3494 regexp *const prog = ReANY(rx);
3496 RXi_GET_DECL(prog,progi);
3497 GET_RE_DEBUG_FLAGS_DECL;
3499 PERL_ARGS_ASSERT_REGTRY;
3501 reginfo->cutpoint=NULL;
3503 prog->offs[0].start = *startposp - reginfo->strbeg;
3504 prog->lastparen = 0;
3505 prog->lastcloseparen = 0;
3507 /* XXXX What this code is doing here?!!! There should be no need
3508 to do this again and again, prog->lastparen should take care of
3511 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3512 * Actually, the code in regcppop() (which Ilya may be meaning by
3513 * prog->lastparen), is not needed at all by the test suite
3514 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3515 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3516 * Meanwhile, this code *is* needed for the
3517 * above-mentioned test suite tests to succeed. The common theme
3518 * on those tests seems to be returning null fields from matches.
3519 * --jhi updated by dapm */
3521 if (prog->nparens) {
3522 regexp_paren_pair *pp = prog->offs;
3524 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3532 result = regmatch(reginfo, *startposp, progi->program + 1);
3534 prog->offs[0].end = result;
3537 if (reginfo->cutpoint)
3538 *startposp= reginfo->cutpoint;
3539 REGCP_UNWIND(lastcp);
3544 #define sayYES goto yes
3545 #define sayNO goto no
3546 #define sayNO_SILENT goto no_silent
3548 /* we dont use STMT_START/END here because it leads to
3549 "unreachable code" warnings, which are bogus, but distracting. */
3550 #define CACHEsayNO \
3551 if (ST.cache_mask) \
3552 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3555 /* this is used to determine how far from the left messages like
3556 'failed...' are printed. It should be set such that messages
3557 are inline with the regop output that created them.
3559 #define REPORT_CODE_OFF 32
3562 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3563 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3564 #define CHRTEST_NOT_A_CP_1 -999
3565 #define CHRTEST_NOT_A_CP_2 -998
3567 /* grab a new slab and return the first slot in it */
3569 STATIC regmatch_state *
3572 #if PERL_VERSION < 9 && !defined(PERL_CORE)
3575 regmatch_slab *s = PL_regmatch_slab->next;
3577 Newx(s, 1, regmatch_slab);
3578 s->prev = PL_regmatch_slab;
3580 PL_regmatch_slab->next = s;
3582 PL_regmatch_slab = s;
3583 return SLAB_FIRST(s);
3587 /* push a new state then goto it */
3589 #define PUSH_STATE_GOTO(state, node, input) \
3590 pushinput = input; \
3592 st->resume_state = state; \
3595 /* push a new state with success backtracking, then goto it */
3597 #define PUSH_YES_STATE_GOTO(state, node, input) \
3598 pushinput = input; \
3600 st->resume_state = state; \
3601 goto push_yes_state;
3608 regmatch() - main matching routine
3610 This is basically one big switch statement in a loop. We execute an op,
3611 set 'next' to point the next op, and continue. If we come to a point which
3612 we may need to backtrack to on failure such as (A|B|C), we push a
3613 backtrack state onto the backtrack stack. On failure, we pop the top
3614 state, and re-enter the loop at the state indicated. If there are no more
3615 states to pop, we return failure.
3617 Sometimes we also need to backtrack on success; for example /A+/, where
3618 after successfully matching one A, we need to go back and try to
3619 match another one; similarly for lookahead assertions: if the assertion
3620 completes successfully, we backtrack to the state just before the assertion
3621 and then carry on. In these cases, the pushed state is marked as
3622 'backtrack on success too'. This marking is in fact done by a chain of
3623 pointers, each pointing to the previous 'yes' state. On success, we pop to
3624 the nearest yes state, discarding any intermediate failure-only states.
3625 Sometimes a yes state is pushed just to force some cleanup code to be
3626 called at the end of a successful match or submatch; e.g. (??{$re}) uses
3627 it to free the inner regex.
3629 Note that failure backtracking rewinds the cursor position, while
3630 success backtracking leaves it alone.
3632 A pattern is complete when the END op is executed, while a subpattern
3633 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
3634 ops trigger the "pop to last yes state if any, otherwise return true"
3637 A common convention in this function is to use A and B to refer to the two
3638 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
3639 the subpattern to be matched possibly multiple times, while B is the entire
3640 rest of the pattern. Variable and state names reflect this convention.
3642 The states in the main switch are the union of ops and failure/success of
3643 substates associated with with that op. For example, IFMATCH is the op
3644 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
3645 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
3646 successfully matched A and IFMATCH_A_fail is a state saying that we have
3647 just failed to match A. Resume states always come in pairs. The backtrack
3648 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
3649 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
3650 on success or failure.
3652 The struct that holds a backtracking state is actually a big union, with
3653 one variant for each major type of op. The variable st points to the
3654 top-most backtrack struct. To make the code clearer, within each
3655 block of code we #define ST to alias the relevant union.
3657 Here's a concrete example of a (vastly oversimplified) IFMATCH
3663 #define ST st->u.ifmatch
3665 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3666 ST.foo = ...; // some state we wish to save
3668 // push a yes backtrack state with a resume value of
3669 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
3671 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
3674 case IFMATCH_A: // we have successfully executed A; now continue with B
3676 bar = ST.foo; // do something with the preserved value
3679 case IFMATCH_A_fail: // A failed, so the assertion failed
3680 ...; // do some housekeeping, then ...
3681 sayNO; // propagate the failure
3688 For any old-timers reading this who are familiar with the old recursive
3689 approach, the code above is equivalent to:
3691 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
3700 ...; // do some housekeeping, then ...
3701 sayNO; // propagate the failure
3704 The topmost backtrack state, pointed to by st, is usually free. If you
3705 want to claim it, populate any ST.foo fields in it with values you wish to
3706 save, then do one of
3708 PUSH_STATE_GOTO(resume_state, node, newinput);
3709 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
3711 which sets that backtrack state's resume value to 'resume_state', pushes a
3712 new free entry to the top of the backtrack stack, then goes to 'node'.
3713 On backtracking, the free slot is popped, and the saved state becomes the
3714 new free state. An ST.foo field in this new top state can be temporarily
3715 accessed to retrieve values, but once the main loop is re-entered, it
3716 becomes available for reuse.
3718 Note that the depth of the backtrack stack constantly increases during the
3719 left-to-right execution of the pattern, rather than going up and down with
3720 the pattern nesting. For example the stack is at its maximum at Z at the
3721 end of the pattern, rather than at X in the following:
3723 /(((X)+)+)+....(Y)+....Z/
3725 The only exceptions to this are lookahead/behind assertions and the cut,
3726 (?>A), which pop all the backtrack states associated with A before
3729 Backtrack state structs are allocated in slabs of about 4K in size.
3730 PL_regmatch_state and st always point to the currently active state,
3731 and PL_regmatch_slab points to the slab currently containing
3732 PL_regmatch_state. The first time regmatch() is called, the first slab is
3733 allocated, and is never freed until interpreter destruction. When the slab
3734 is full, a new one is allocated and chained to the end. At exit from
3735 regmatch(), slabs allocated since entry are freed.
3740 #define DEBUG_STATE_pp(pp) \
3742 DUMP_EXEC_POS(locinput, scan, utf8_target); \
3743 PerlIO_printf(Perl_debug_log, \
3744 " %*s"pp" %s%s%s%s%s\n", \
3746 PL_reg_name[st->resume_state], \
3747 ((st==yes_state||st==mark_state) ? "[" : ""), \
3748 ((st==yes_state) ? "Y" : ""), \
3749 ((st==mark_state) ? "M" : ""), \
3750 ((st==yes_state||st==mark_state) ? "]" : "") \
3755 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
3760 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
3761 const char *start, const char *end, const char *blurb)
3763 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
3765 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
3770 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
3771 RX_PRECOMP_const(prog), RX_PRELEN(prog), 60);
3773 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
3774 start, end - start, 60);
3776 PerlIO_printf(Perl_debug_log,
3777 "%s%s REx%s %s against %s\n",
3778 PL_colors[4], blurb, PL_colors[5], s0, s1);
3780 if (utf8_target||utf8_pat)
3781 PerlIO_printf(Perl_debug_log, "UTF-8 %s%s%s...\n",
3782 utf8_pat ? "pattern" : "",
3783 utf8_pat && utf8_target ? " and " : "",
3784 utf8_target ? "string" : ""
3790 S_dump_exec_pos(pTHX_ const char *locinput,
3791 const regnode *scan,
3792 const char *loc_regeol,
3793 const char *loc_bostr,
3794 const char *loc_reg_starttry,
3795 const bool utf8_target)
3797 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
3798 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
3799 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
3800 /* The part of the string before starttry has one color
3801 (pref0_len chars), between starttry and current
3802 position another one (pref_len - pref0_len chars),
3803 after the current position the third one.
3804 We assume that pref0_len <= pref_len, otherwise we
3805 decrease pref0_len. */
3806 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
3807 ? (5 + taill) - l : locinput - loc_bostr;
3810 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
3812 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
3814 pref0_len = pref_len - (locinput - loc_reg_starttry);
3815 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
3816 l = ( loc_regeol - locinput > (5 + taill) - pref_len
3817 ? (5 + taill) - pref_len : loc_regeol - locinput);
3818 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
3822 if (pref0_len > pref_len)
3823 pref0_len = pref_len;
3825 const int is_uni = (utf8_target && OP(scan) != CANY) ? 1 : 0;
3827 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
3828 (locinput - pref_len),pref0_len, 60, 4, 5);
3830 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
3831 (locinput - pref_len + pref0_len),
3832 pref_len - pref0_len, 60, 2, 3);
3834 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
3835 locinput, loc_regeol - locinput, 10, 0, 1);
3837 const STRLEN tlen=len0+len1+len2;
3838 PerlIO_printf(Perl_debug_log,
3839 "%4"IVdf" <%.*s%.*s%s%.*s>%*s|",
3840 (IV)(locinput - loc_bostr),
3843 (docolor ? "" : "> <"),
3845 (int)(tlen > 19 ? 0 : 19 - tlen),
3852 /* reg_check_named_buff_matched()
3853 * Checks to see if a named buffer has matched. The data array of
3854 * buffer numbers corresponding to the buffer is expected to reside
3855 * in the regexp->data->data array in the slot stored in the ARG() of
3856 * node involved. Note that this routine doesn't actually care about the
3857 * name, that information is not preserved from compilation to execution.
3858 * Returns the index of the leftmost defined buffer with the given name
3859 * or 0 if non of the buffers matched.
3862 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
3865 RXi_GET_DECL(rex,rexi);
3866 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
3867 I32 *nums=(I32*)SvPVX(sv_dat);
3869 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
3871 for ( n=0; n<SvIVX(sv_dat); n++ ) {
3872 if ((I32)rex->lastparen >= nums[n] &&
3873 rex->offs[nums[n]].end != -1)
3883 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
3884 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
3886 /* This function determines if there are one or two characters that match
3887 * the first character of the passed-in EXACTish node <text_node>, and if
3888 * so, returns them in the passed-in pointers.
3890 * If it determines that no possible character in the target string can
3891 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
3892 * the first character in <text_node> requires UTF-8 to represent, and the
3893 * target string isn't in UTF-8.)
3895 * If there are more than two characters that could match the beginning of
3896 * <text_node>, or if more context is required to determine a match or not,
3897 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
3899 * The motiviation behind this function is to allow the caller to set up
3900 * tight loops for matching. If <text_node> is of type EXACT, there is
3901 * only one possible character that can match its first character, and so
3902 * the situation is quite simple. But things get much more complicated if
3903 * folding is involved. It may be that the first character of an EXACTFish
3904 * node doesn't participate in any possible fold, e.g., punctuation, so it
3905 * can be matched only by itself. The vast majority of characters that are
3906 * in folds match just two things, their lower and upper-case equivalents.
3907 * But not all are like that; some have multiple possible matches, or match
3908 * sequences of more than one character. This function sorts all that out.
3910 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
3911 * loop of trying to match A*, we know we can't exit where the thing
3912 * following it isn't a B. And something can't be a B unless it is the
3913 * beginning of B. By putting a quick test for that beginning in a tight
3914 * loop, we can rule out things that can't possibly be B without having to
3915 * break out of the loop, thus avoiding work. Similarly, if A is a single
3916 * character, we can make a tight loop matching A*, using the outputs of
3919 * If the target string to match isn't in UTF-8, and there aren't
3920 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
3921 * the one or two possible octets (which are characters in this situation)
3922 * that can match. In all cases, if there is only one character that can
3923 * match, *<c1p> and *<c2p> will be identical.
3925 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
3926 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
3927 * can match the beginning of <text_node>. They should be declared with at
3928 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
3929 * undefined what these contain.) If one or both of the buffers are
3930 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
3931 * corresponding invariant. If variant, the corresponding *<c1p> and/or
3932 * *<c2p> will be set to a negative number(s) that shouldn't match any code
3933 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
3934 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
3936 const bool utf8_target = reginfo->is_utf8_target;
3938 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
3939 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
3940 bool use_chrtest_void = FALSE;
3941 const bool is_utf8_pat = reginfo->is_utf8_pat;
3943 /* Used when we have both utf8 input and utf8 output, to avoid converting
3944 * to/from code points */
3945 bool utf8_has_been_setup = FALSE;
3949 U8 *pat = (U8*)STRING(text_node);
3950 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
3952 if (OP(text_node) == EXACT || OP(text_node) == EXACTL) {
3954 /* In an exact node, only one thing can be matched, that first
3955 * character. If both the pat and the target are UTF-8, we can just
3956 * copy the input to the output, avoiding finding the code point of
3961 else if (utf8_target) {
3962 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
3963 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
3964 utf8_has_been_setup = TRUE;
3967 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
3970 else { /* an EXACTFish node */
3971 U8 *pat_end = pat + STR_LEN(text_node);
3973 /* An EXACTFL node has at least some characters unfolded, because what
3974 * they match is not known until now. So, now is the time to fold
3975 * the first few of them, as many as are needed to determine 'c1' and
3976 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
3977 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
3978 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
3979 * need to fold as many characters as a single character can fold to,
3980 * so that later we can check if the first ones are such a multi-char
3981 * fold. But, in such a pattern only locale-problematic characters
3982 * aren't folded, so we can skip this completely if the first character
3983 * in the node isn't one of the tricky ones */
3984 if (OP(text_node) == EXACTFL) {
3986 if (! is_utf8_pat) {
3987 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
3989 folded[0] = folded[1] = 's';
3991 pat_end = folded + 2;
3994 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
3999 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
4001 *(d++) = (U8) toFOLD_LC(*s);
4006 _to_utf8_fold_flags(s,
4009 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
4020 if ((is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
4021 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
4023 /* Multi-character folds require more context to sort out. Also
4024 * PL_utf8_foldclosures used below doesn't handle them, so have to
4025 * be handled outside this routine */
4026 use_chrtest_void = TRUE;
4028 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
4029 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
4031 /* Load the folds hash, if not already done */
4033 if (! PL_utf8_foldclosures) {
4034 _load_PL_utf8_foldclosures();
4037 /* The fold closures data structure is a hash with the keys
4038 * being the UTF-8 of every character that is folded to, like
4039 * 'k', and the values each an array of all code points that
4040 * fold to its key. e.g. [ 'k', 'K', KELVIN_SIGN ].
4041 * Multi-character folds are not included */
4042 if ((! (listp = hv_fetch(PL_utf8_foldclosures,
4047 /* Not found in the hash, therefore there are no folds
4048 * containing it, so there is only a single character that
4052 else { /* Does participate in folds */
4053 AV* list = (AV*) *listp;
4054 if (av_tindex(list) != 1) {
4056 /* If there aren't exactly two folds to this, it is
4057 * outside the scope of this function */
4058 use_chrtest_void = TRUE;
4060 else { /* There are two. Get them */
4061 SV** c_p = av_fetch(list, 0, FALSE);
4063 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
4067 c_p = av_fetch(list, 1, FALSE);
4069 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
4073 /* Folds that cross the 255/256 boundary are forbidden
4074 * if EXACTFL (and isnt a UTF8 locale), or EXACTFA and
4075 * one is ASCIII. Since the pattern character is above
4076 * 255, and its only other match is below 256, the only
4077 * legal match will be to itself. We have thrown away
4078 * the original, so have to compute which is the one
4080 if ((c1 < 256) != (c2 < 256)) {
4081 if ((OP(text_node) == EXACTFL
4082 && ! IN_UTF8_CTYPE_LOCALE)
4083 || ((OP(text_node) == EXACTFA
4084 || OP(text_node) == EXACTFA_NO_TRIE)
4085 && (isASCII(c1) || isASCII(c2))))
4098 else /* Here, c1 is <= 255 */
4100 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
4101 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
4102 && ((OP(text_node) != EXACTFA
4103 && OP(text_node) != EXACTFA_NO_TRIE)
4106 /* Here, there could be something above Latin1 in the target
4107 * which folds to this character in the pattern. All such
4108 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
4109 * than two characters involved in their folds, so are outside
4110 * the scope of this function */
4111 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4112 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
4115 use_chrtest_void = TRUE;
4118 else { /* Here nothing above Latin1 can fold to the pattern
4120 switch (OP(text_node)) {
4122 case EXACTFL: /* /l rules */
4123 c2 = PL_fold_locale[c1];
4126 case EXACTF: /* This node only generated for non-utf8
4128 assert(! is_utf8_pat);
4129 if (! utf8_target) { /* /d rules */
4134 /* /u rules for all these. This happens to work for
4135 * EXACTFA as nothing in Latin1 folds to ASCII */
4136 case EXACTFA_NO_TRIE: /* This node only generated for
4137 non-utf8 patterns */
4138 assert(! is_utf8_pat);
4143 c2 = PL_fold_latin1[c1];
4147 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
4148 NOT_REACHED; /* NOTREACHED */
4154 /* Here have figured things out. Set up the returns */
4155 if (use_chrtest_void) {
4156 *c2p = *c1p = CHRTEST_VOID;
4158 else if (utf8_target) {
4159 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
4160 uvchr_to_utf8(c1_utf8, c1);
4161 uvchr_to_utf8(c2_utf8, c2);
4164 /* Invariants are stored in both the utf8 and byte outputs; Use
4165 * negative numbers otherwise for the byte ones. Make sure that the
4166 * byte ones are the same iff the utf8 ones are the same */
4167 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
4168 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
4171 ? CHRTEST_NOT_A_CP_1
4172 : CHRTEST_NOT_A_CP_2;
4174 else if (c1 > 255) {
4175 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
4180 *c1p = *c2p = c2; /* c2 is the only representable value */
4182 else { /* c1 is representable; see about c2 */
4184 *c2p = (c2 < 256) ? c2 : c1;
4190 /* This creates a single number by combining two, with 'before' being like the
4191 * 10's digit, but this isn't necessarily base 10; it is base however many
4192 * elements of the enum there are */
4193 #define GCBcase(before, after) ((GCB_ENUM_COUNT * before) + after)
4196 S_isGCB(const GCB_enum before, const GCB_enum after)
4198 /* returns a boolean indicating if there is a Grapheme Cluster Boundary
4199 * between the inputs. See http://www.unicode.org/reports/tr29/ */
4201 switch (GCBcase(before, after)) {
4203 /* Break at the start and end of text.
4207 Break before and after controls except between CR and LF
4208 GB4. ( Control | CR | LF ) ÷
4209 GB5. ÷ ( Control | CR | LF )
4211 Otherwise, break everywhere.
4216 /* Do not break between a CR and LF.
4218 case GCBcase(GCB_CR, GCB_LF):
4221 /* Do not break Hangul syllable sequences.
4222 GB6. L × ( L | V | LV | LVT ) */
4223 case GCBcase(GCB_L, GCB_L):
4224 case GCBcase(GCB_L, GCB_V):
4225 case GCBcase(GCB_L, GCB_LV):
4226 case GCBcase(GCB_L, GCB_LVT):
4229 /* GB7. ( LV | V ) × ( V | T ) */
4230 case GCBcase(GCB_LV, GCB_V):
4231 case GCBcase(GCB_LV, GCB_T):
4232 case GCBcase(GCB_V, GCB_V):
4233 case GCBcase(GCB_V, GCB_T):
4236 /* GB8. ( LVT | T) × T */
4237 case GCBcase(GCB_LVT, GCB_T):
4238 case GCBcase(GCB_T, GCB_T):
4241 /* Do not break between regional indicator symbols.
4242 GB8a. Regional_Indicator × Regional_Indicator */
4243 case GCBcase(GCB_Regional_Indicator, GCB_Regional_Indicator):
4246 /* Do not break before extending characters.
4248 case GCBcase(GCB_Other, GCB_Extend):
4249 case GCBcase(GCB_Extend, GCB_Extend):
4250 case GCBcase(GCB_L, GCB_Extend):
4251 case GCBcase(GCB_LV, GCB_Extend):
4252 case GCBcase(GCB_LVT, GCB_Extend):
4253 case GCBcase(GCB_Prepend, GCB_Extend):
4254 case GCBcase(GCB_Regional_Indicator, GCB_Extend):
4255 case GCBcase(GCB_SpacingMark, GCB_Extend):
4256 case GCBcase(GCB_T, GCB_Extend):
4257 case GCBcase(GCB_V, GCB_Extend):
4260 /* Do not break before SpacingMarks, or after Prepend characters.
4261 GB9a. × SpacingMark */
4262 case GCBcase(GCB_Other, GCB_SpacingMark):
4263 case GCBcase(GCB_Extend, GCB_SpacingMark):
4264 case GCBcase(GCB_L, GCB_SpacingMark):
4265 case GCBcase(GCB_LV, GCB_SpacingMark):
4266 case GCBcase(GCB_LVT, GCB_SpacingMark):
4267 case GCBcase(GCB_Prepend, GCB_SpacingMark):
4268 case GCBcase(GCB_Regional_Indicator, GCB_SpacingMark):
4269 case GCBcase(GCB_SpacingMark, GCB_SpacingMark):
4270 case GCBcase(GCB_T, GCB_SpacingMark):
4271 case GCBcase(GCB_V, GCB_SpacingMark):
4274 /* GB9b. Prepend × */
4275 case GCBcase(GCB_Prepend, GCB_Other):
4276 case GCBcase(GCB_Prepend, GCB_L):
4277 case GCBcase(GCB_Prepend, GCB_LV):
4278 case GCBcase(GCB_Prepend, GCB_LVT):
4279 case GCBcase(GCB_Prepend, GCB_Prepend):
4280 case GCBcase(GCB_Prepend, GCB_Regional_Indicator):
4281 case GCBcase(GCB_Prepend, GCB_T):
4282 case GCBcase(GCB_Prepend, GCB_V):
4286 NOT_REACHED; /* NOTREACHED */
4289 #define SBcase(before, after) ((SB_ENUM_COUNT * before) + after)
4292 S_isSB(pTHX_ SB_enum before,
4294 const U8 * const strbeg,
4295 const U8 * const curpos,
4296 const U8 * const strend,
4297 const bool utf8_target)
4299 /* returns a boolean indicating if there is a Sentence Boundary Break
4300 * between the inputs. See http://www.unicode.org/reports/tr29/ */
4302 U8 * lpos = (U8 *) curpos;
4306 PERL_ARGS_ASSERT_ISSB;
4308 /* Break at the start and end of text.
4311 if (before == SB_EDGE || after == SB_EDGE) {
4315 /* SB 3: Do not break within CRLF. */
4316 if (before == SB_CR && after == SB_LF) {
4320 /* Break after paragraph separators. (though why CR and LF are considered
4321 * so is beyond me (khw)
4322 SB4. Sep | CR | LF ÷ */
4323 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4327 /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF.
4328 * (See Section 6.2, Replacing Ignore Rules.)
4329 SB5. X (Extend | Format)* → X */
4330 if (after == SB_Extend || after == SB_Format) {
4334 if (before == SB_Extend || before == SB_Format) {
4335 before = backup_one_SB(strbeg, &lpos, utf8_target);
4338 /* Do not break after ambiguous terminators like period, if they are
4339 * immediately followed by a number or lowercase letter, if they are
4340 * between uppercase letters, if the first following letter (optionally
4341 * after certain punctuation) is lowercase, or if they are followed by
4342 * "continuation" punctuation such as comma, colon, or semicolon. For
4343 * example, a period may be an abbreviation or numeric period, and thus may
4344 * not mark the end of a sentence.
4346 * SB6. ATerm × Numeric */
4347 if (before == SB_ATerm && after == SB_Numeric) {
4351 /* SB7. (Upper | Lower) ATerm × Upper */
4352 if (before == SB_ATerm && after == SB_Upper) {
4354 backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4355 if (backup == SB_Upper || backup == SB_Lower) {
4360 /* SB8a. (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm)
4361 * SB10. (STerm | ATerm) Close* Sp* × ( Sp | Sep | CR | LF ) */
4364 while (backup == SB_Sp) {
4365 backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4367 while (backup == SB_Close) {
4368 backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4370 if ((backup == SB_STerm || backup == SB_ATerm)
4371 && ( after == SB_SContinue
4372 || after == SB_STerm
4373 || after == SB_ATerm
4382 /* SB8. ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR | LF |
4383 * STerm | ATerm) )* Lower */
4384 if (backup == SB_ATerm) {
4385 U8 * rpos = (U8 *) curpos;
4386 SB_enum later = after;
4388 while ( later != SB_OLetter
4389 && later != SB_Upper
4390 && later != SB_Lower
4394 && later != SB_STerm
4395 && later != SB_ATerm
4396 && later != SB_EDGE)
4398 later = advance_one_SB(&rpos, strend, utf8_target);
4400 if (later == SB_Lower) {
4405 /* Break after sentence terminators, but include closing punctuation,
4406 * trailing spaces, and a paragraph separator (if present). [See note
4408 * SB9. ( STerm | ATerm ) Close* × ( Close | Sp | Sep | CR | LF ) */
4411 while (backup == SB_Close) {
4412 backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4414 if ((backup == SB_STerm || backup == SB_ATerm)
4415 && ( after == SB_Close
4425 /* SB11. ( STerm | ATerm ) Close* Sp* ( Sep | CR | LF )? ÷ */
4427 backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4428 if ( backup == SB_Sep
4437 while (backup == SB_Sp) {
4438 backup = backup_one_SB(strbeg, &lpos, utf8_target);
4440 while (backup == SB_Close) {
4441 backup = backup_one_SB(strbeg, &lpos, utf8_target);
4443 if (backup == SB_STerm || backup == SB_ATerm) {
4447 /* Otherwise, do not break.
4454 S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4458 PERL_ARGS_ASSERT_ADVANCE_ONE_SB;
4460 if (*curpos >= strend) {
4466 *curpos += UTF8SKIP(*curpos);
4467 if (*curpos >= strend) {
4470 sb = getSB_VAL_UTF8(*curpos, strend);
4471 } while (sb == SB_Extend || sb == SB_Format);
4476 if (*curpos >= strend) {
4479 sb = getSB_VAL_CP(**curpos);
4480 } while (sb == SB_Extend || sb == SB_Format);
4487 S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4491 PERL_ARGS_ASSERT_BACKUP_ONE_SB;
4493 if (*curpos < strbeg) {
4498 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4499 if (! prev_char_pos) {
4503 /* Back up over Extend and Format. curpos is always just to the right
4504 * of the characater whose value we are getting */
4506 U8 * prev_prev_char_pos;
4507 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1,
4510 sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4511 *curpos = prev_char_pos;
4512 prev_char_pos = prev_prev_char_pos;
4515 *curpos = (U8 *) strbeg;
4518 } while (sb == SB_Extend || sb == SB_Format);
4522 if (*curpos - 2 < strbeg) {
4523 *curpos = (U8 *) strbeg;
4527 sb = getSB_VAL_CP(*(*curpos - 1));
4528 } while (sb == SB_Extend || sb == SB_Format);
4534 #define WBcase(before, after) ((WB_ENUM_COUNT * before) + after)
4537 S_isWB(pTHX_ WB_enum previous,
4540 const U8 * const strbeg,
4541 const U8 * const curpos,
4542 const U8 * const strend,
4543 const bool utf8_target)
4545 /* Return a boolean as to if the boundary between 'before' and 'after' is
4546 * a Unicode word break, using their published algorithm. Context may be
4547 * needed to make this determination. If the value for the character
4548 * before 'before' is known, it is passed as 'previous'; otherwise that
4549 * should be set to WB_UNKNOWN. The other input parameters give the
4550 * boundaries and current position in the matching of the string. That
4551 * is, 'curpos' marks the position where the character whose wb value is
4552 * 'after' begins. See http://www.unicode.org/reports/tr29/ */
4554 U8 * before_pos = (U8 *) curpos;
4555 U8 * after_pos = (U8 *) curpos;
4557 PERL_ARGS_ASSERT_ISWB;
4559 /* WB1 and WB2: Break at the start and end of text. */
4560 if (before == WB_EDGE || after == WB_EDGE) {
4564 /* WB 3: Do not break within CRLF. */
4565 if (before == WB_CR && after == WB_LF) {
4569 /* WB 3a and WB 3b: Otherwise break before and after Newlines (including CR
4571 if ( before == WB_CR || before == WB_LF || before == WB_Newline
4572 || after == WB_CR || after == WB_LF || after == WB_Newline)
4577 /* Ignore Format and Extend characters, except when they appear at the
4578 * beginning of a region of text.
4579 * WB4. X (Extend | Format)* → X. */
4581 if (after == WB_Extend || after == WB_Format) {
4585 if (before == WB_Extend || before == WB_Format) {
4586 before = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
4589 switch (WBcase(before, after)) {
4590 /* Otherwise, break everywhere (including around ideographs).
4595 /* Do not break between most letters.
4596 WB5. (ALetter | Hebrew_Letter) × (ALetter | Hebrew_Letter) */
4597 case WBcase(WB_ALetter, WB_ALetter):
4598 case WBcase(WB_ALetter, WB_Hebrew_Letter):
4599 case WBcase(WB_Hebrew_Letter, WB_ALetter):
4600 case WBcase(WB_Hebrew_Letter, WB_Hebrew_Letter):
4603 /* Do not break letters across certain punctuation.
4604 WB6. (ALetter | Hebrew_Letter)
4605 × (MidLetter | MidNumLet | Single_Quote) (ALetter
4607 case WBcase(WB_ALetter, WB_MidLetter):
4608 case WBcase(WB_ALetter, WB_MidNumLet):
4609 case WBcase(WB_ALetter, WB_Single_Quote):
4610 case WBcase(WB_Hebrew_Letter, WB_MidLetter):
4611 case WBcase(WB_Hebrew_Letter, WB_MidNumLet):
4612 /*case WBcase(WB_Hebrew_Letter, WB_Single_Quote):*/
4613 after = advance_one_WB(&after_pos, strend, utf8_target);
4614 return after != WB_ALetter && after != WB_Hebrew_Letter;
4616 /* WB7. (ALetter | Hebrew_Letter) (MidLetter | MidNumLet |
4617 * Single_Quote) × (ALetter | Hebrew_Letter) */
4618 case WBcase(WB_MidLetter, WB_ALetter):
4619 case WBcase(WB_MidLetter, WB_Hebrew_Letter):
4620 case WBcase(WB_MidNumLet, WB_ALetter):
4621 case WBcase(WB_MidNumLet, WB_Hebrew_Letter):
4622 case WBcase(WB_Single_Quote, WB_ALetter):
4623 case WBcase(WB_Single_Quote, WB_Hebrew_Letter):
4625 = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
4626 return before != WB_ALetter && before != WB_Hebrew_Letter;
4628 /* WB7a. Hebrew_Letter × Single_Quote */
4629 case WBcase(WB_Hebrew_Letter, WB_Single_Quote):
4632 /* WB7b. Hebrew_Letter × Double_Quote Hebrew_Letter */
4633 case WBcase(WB_Hebrew_Letter, WB_Double_Quote):
4634 return advance_one_WB(&after_pos, strend, utf8_target)
4635 != WB_Hebrew_Letter;
4637 /* WB7c. Hebrew_Letter Double_Quote × Hebrew_Letter */
4638 case WBcase(WB_Double_Quote, WB_Hebrew_Letter):
4639 return backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
4640 != WB_Hebrew_Letter;
4642 /* Do not break within sequences of digits, or digits adjacent to
4643 * letters (“3a”, or “A3”).
4644 WB8. Numeric × Numeric */
4645 case WBcase(WB_Numeric, WB_Numeric):
4648 /* WB9. (ALetter | Hebrew_Letter) × Numeric */
4649 case WBcase(WB_ALetter, WB_Numeric):
4650 case WBcase(WB_Hebrew_Letter, WB_Numeric):
4653 /* WB10. Numeric × (ALetter | Hebrew_Letter) */
4654 case WBcase(WB_Numeric, WB_ALetter):
4655 case WBcase(WB_Numeric, WB_Hebrew_Letter):
4658 /* Do not break within sequences, such as “3.2” or “3,456.789”.
4659 WB11. Numeric (MidNum | MidNumLet | Single_Quote) × Numeric
4661 case WBcase(WB_MidNum, WB_Numeric):
4662 case WBcase(WB_MidNumLet, WB_Numeric):
4663 case WBcase(WB_Single_Quote, WB_Numeric):
4664 return backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
4667 /* WB12. Numeric × (MidNum | MidNumLet | Single_Quote) Numeric
4669 case WBcase(WB_Numeric, WB_MidNum):
4670 case WBcase(WB_Numeric, WB_MidNumLet):
4671 case WBcase(WB_Numeric, WB_Single_Quote):
4672 return advance_one_WB(&after_pos, strend, utf8_target)
4675 /* Do not break between Katakana.
4676 WB13. Katakana × Katakana */
4677 case WBcase(WB_Katakana, WB_Katakana):
4680 /* Do not break from extenders.
4681 WB13a. (ALetter | Hebrew_Letter | Numeric | Katakana |
4682 ExtendNumLet) × ExtendNumLet */
4683 case WBcase(WB_ALetter, WB_ExtendNumLet):
4684 case WBcase(WB_Hebrew_Letter, WB_ExtendNumLet):
4685 case WBcase(WB_Numeric, WB_ExtendNumLet):
4686 case WBcase(WB_Katakana, WB_ExtendNumLet):
4687 case WBcase(WB_ExtendNumLet, WB_ExtendNumLet):
4690 /* WB13b. ExtendNumLet × (ALetter | Hebrew_Letter | Numeric
4692 case WBcase(WB_ExtendNumLet, WB_ALetter):
4693 case WBcase(WB_ExtendNumLet, WB_Hebrew_Letter):
4694 case WBcase(WB_ExtendNumLet, WB_Numeric):
4695 case WBcase(WB_ExtendNumLet, WB_Katakana):
4698 /* Do not break between regional indicator symbols.
4699 WB13c. Regional_Indicator × Regional_Indicator */
4700 case WBcase(WB_Regional_Indicator, WB_Regional_Indicator):
4705 NOT_REACHED; /* NOTREACHED */
4709 S_advance_one_WB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4713 PERL_ARGS_ASSERT_ADVANCE_ONE_WB;
4715 if (*curpos >= strend) {
4721 /* Advance over Extend and Format */
4723 *curpos += UTF8SKIP(*curpos);
4724 if (*curpos >= strend) {
4727 wb = getWB_VAL_UTF8(*curpos, strend);
4728 } while (wb == WB_Extend || wb == WB_Format);
4733 if (*curpos >= strend) {
4736 wb = getWB_VAL_CP(**curpos);
4737 } while (wb == WB_Extend || wb == WB_Format);
4744 S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4748 PERL_ARGS_ASSERT_BACKUP_ONE_WB;
4750 /* If we know what the previous character's break value is, don't have
4752 if (*previous != WB_UNKNOWN) {
4754 *previous = WB_UNKNOWN;
4755 /* XXX Note that doesn't change curpos, and maybe should */
4757 /* But we always back up over these two types */
4758 if (wb != WB_Extend && wb != WB_Format) {
4763 if (*curpos < strbeg) {
4768 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4769 if (! prev_char_pos) {
4773 /* Back up over Extend and Format. curpos is always just to the right
4774 * of the characater whose value we are getting */
4776 U8 * prev_prev_char_pos;
4777 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos,
4781 wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4782 *curpos = prev_char_pos;
4783 prev_char_pos = prev_prev_char_pos;
4786 *curpos = (U8 *) strbeg;
4789 } while (wb == WB_Extend || wb == WB_Format);
4793 if (*curpos - 2 < strbeg) {
4794 *curpos = (U8 *) strbeg;
4798 wb = getWB_VAL_CP(*(*curpos - 1));
4799 } while (wb == WB_Extend || wb == WB_Format);
4805 /* returns -1 on failure, $+[0] on success */
4807 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
4809 #if PERL_VERSION < 9 && !defined(PERL_CORE)
4813 const bool utf8_target = reginfo->is_utf8_target;
4814 const U32 uniflags = UTF8_ALLOW_DEFAULT;
4815 REGEXP *rex_sv = reginfo->prog;
4816 regexp *rex = ReANY(rex_sv);
4817 RXi_GET_DECL(rex,rexi);
4818 /* the current state. This is a cached copy of PL_regmatch_state */
4820 /* cache heavy used fields of st in registers */
4823 U32 n = 0; /* general value; init to avoid compiler warning */
4824 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
4825 char *locinput = startpos;
4826 char *pushinput; /* where to continue after a PUSH */
4827 I32 nextchr; /* is always set to UCHARAT(locinput) */
4829 bool result = 0; /* return value of S_regmatch */
4830 int depth = 0; /* depth of backtrack stack */
4831 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
4832 const U32 max_nochange_depth =
4833 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
4834 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
4835 regmatch_state *yes_state = NULL; /* state to pop to on success of
4837 /* mark_state piggy backs on the yes_state logic so that when we unwind
4838 the stack on success we can update the mark_state as we go */
4839 regmatch_state *mark_state = NULL; /* last mark state we have seen */
4840 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
4841 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
4843 bool no_final = 0; /* prevent failure from backtracking? */
4844 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
4845 char *startpoint = locinput;
4846 SV *popmark = NULL; /* are we looking for a mark? */
4847 SV *sv_commit = NULL; /* last mark name seen in failure */
4848 SV *sv_yes_mark = NULL; /* last mark name we have seen
4849 during a successful match */
4850 U32 lastopen = 0; /* last open we saw */
4851 bool has_cutgroup = RX_HAS_CUTGROUP(rex) ? 1 : 0;
4852 SV* const oreplsv = GvSVn(PL_replgv);
4853 /* these three flags are set by various ops to signal information to
4854 * the very next op. They have a useful lifetime of exactly one loop
4855 * iteration, and are not preserved or restored by state pushes/pops
4857 bool sw = 0; /* the condition value in (?(cond)a|b) */
4858 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
4859 int logical = 0; /* the following EVAL is:
4863 or the following IFMATCH/UNLESSM is:
4864 false: plain (?=foo)
4865 true: used as a condition: (?(?=foo))
4867 PAD* last_pad = NULL;
4869 I32 gimme = G_SCALAR;
4870 CV *caller_cv = NULL; /* who called us */
4871 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
4872 CHECKPOINT runops_cp; /* savestack position before executing EVAL */
4873 U32 maxopenparen = 0; /* max '(' index seen so far */
4874 int to_complement; /* Invert the result? */
4875 _char_class_number classnum;
4876 bool is_utf8_pat = reginfo->is_utf8_pat;
4881 GET_RE_DEBUG_FLAGS_DECL;
4884 /* protect against undef(*^R) */
4885 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
4887 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
4888 multicall_oldcatch = 0;
4889 multicall_cv = NULL;
4891 PERL_UNUSED_VAR(multicall_cop);
4892 PERL_UNUSED_VAR(newsp);
4895 PERL_ARGS_ASSERT_REGMATCH;
4897 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
4898 PerlIO_printf(Perl_debug_log,"regmatch start\n");
4901 st = PL_regmatch_state;
4903 /* Note that nextchr is a byte even in UTF */
4906 while (scan != NULL) {
4909 SV * const prop = sv_newmortal();
4910 regnode *rnext=regnext(scan);
4911 DUMP_EXEC_POS( locinput, scan, utf8_target );
4912 regprop(rex, prop, scan, reginfo, NULL);
4914 PerlIO_printf(Perl_debug_log,
4915 "%3"IVdf":%*s%s(%"IVdf")\n",
4916 (IV)(scan - rexi->program), depth*2, "",
4918 (PL_regkind[OP(scan)] == END || !rnext) ?
4919 0 : (IV)(rnext - rexi->program));
4922 next = scan + NEXT_OFF(scan);
4925 state_num = OP(scan);
4931 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
4933 switch (state_num) {
4934 case SBOL: /* /^../ and /\A../ */
4935 if (locinput == reginfo->strbeg)
4939 case MBOL: /* /^../m */
4940 if (locinput == reginfo->strbeg ||
4941 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
4948 if (locinput == reginfo->ganch)
4952 case KEEPS: /* \K */
4953 /* update the startpoint */
4954 st->u.keeper.val = rex->offs[0].start;
4955 rex->offs[0].start = locinput - reginfo->strbeg;
4956 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
4958 NOT_REACHED; /* NOTREACHED */
4960 case KEEPS_next_fail:
4961 /* rollback the start point change */
4962 rex->offs[0].start = st->u.keeper.val;
4965 NOT_REACHED; /* NOTREACHED */
4967 case MEOL: /* /..$/m */
4968 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4972 case SEOL: /* /..$/ */
4973 if (!NEXTCHR_IS_EOS && nextchr != '\n')
4975 if (reginfo->strend - locinput > 1)
4980 if (!NEXTCHR_IS_EOS)
4984 case SANY: /* /./s */
4987 goto increment_locinput;
4995 case REG_ANY: /* /./ */
4996 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
4998 goto increment_locinput;
5002 #define ST st->u.trie
5003 case TRIEC: /* (ab|cd) with known charclass */
5004 /* In this case the charclass data is available inline so
5005 we can fail fast without a lot of extra overhead.
5007 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
5009 PerlIO_printf(Perl_debug_log,
5010 "%*s %sfailed to match trie start class...%s\n",
5011 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
5015 NOT_REACHED; /* NOTREACHED */
5018 case TRIE: /* (ab|cd) */
5019 /* the basic plan of execution of the trie is:
5020 * At the beginning, run though all the states, and
5021 * find the longest-matching word. Also remember the position
5022 * of the shortest matching word. For example, this pattern:
5025 * when matched against the string "abcde", will generate
5026 * accept states for all words except 3, with the longest
5027 * matching word being 4, and the shortest being 2 (with
5028 * the position being after char 1 of the string).
5030 * Then for each matching word, in word order (i.e. 1,2,4,5),
5031 * we run the remainder of the pattern; on each try setting
5032 * the current position to the character following the word,
5033 * returning to try the next word on failure.
5035 * We avoid having to build a list of words at runtime by
5036 * using a compile-time structure, wordinfo[].prev, which
5037 * gives, for each word, the previous accepting word (if any).
5038 * In the case above it would contain the mappings 1->2, 2->0,
5039 * 3->0, 4->5, 5->1. We can use this table to generate, from
5040 * the longest word (4 above), a list of all words, by
5041 * following the list of prev pointers; this gives us the
5042 * unordered list 4,5,1,2. Then given the current word we have
5043 * just tried, we can go through the list and find the
5044 * next-biggest word to try (so if we just failed on word 2,
5045 * the next in the list is 4).
5047 * Since at runtime we don't record the matching position in
5048 * the string for each word, we have to work that out for
5049 * each word we're about to process. The wordinfo table holds
5050 * the character length of each word; given that we recorded
5051 * at the start: the position of the shortest word and its
5052 * length in chars, we just need to move the pointer the
5053 * difference between the two char lengths. Depending on
5054 * Unicode status and folding, that's cheap or expensive.
5056 * This algorithm is optimised for the case where are only a
5057 * small number of accept states, i.e. 0,1, or maybe 2.
5058 * With lots of accepts states, and having to try all of them,
5059 * it becomes quadratic on number of accept states to find all
5064 /* what type of TRIE am I? (utf8 makes this contextual) */
5065 DECL_TRIE_TYPE(scan);
5067 /* what trie are we using right now */
5068 reg_trie_data * const trie
5069 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
5070 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
5071 U32 state = trie->startstate;
5073 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
5074 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5076 && UTF8_IS_ABOVE_LATIN1(nextchr)
5077 && scan->flags == EXACTL)
5079 /* We only output for EXACTL, as we let the folder
5080 * output this message for EXACTFLU8 to avoid
5082 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
5087 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
5089 if (trie->states[ state ].wordnum) {
5091 PerlIO_printf(Perl_debug_log,
5092 "%*s %smatched empty string...%s\n",
5093 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
5099 PerlIO_printf(Perl_debug_log,
5100 "%*s %sfailed to match trie start class...%s\n",
5101 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5])
5108 U8 *uc = ( U8* )locinput;
5112 U8 *uscan = (U8*)NULL;
5113 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
5114 U32 charcount = 0; /* how many input chars we have matched */
5115 U32 accepted = 0; /* have we seen any accepting states? */
5117 ST.jump = trie->jump;
5120 ST.longfold = FALSE; /* char longer if folded => it's harder */
5123 /* fully traverse the TRIE; note the position of the
5124 shortest accept state and the wordnum of the longest
5127 while ( state && uc <= (U8*)(reginfo->strend) ) {
5128 U32 base = trie->states[ state ].trans.base;
5132 wordnum = trie->states[ state ].wordnum;
5134 if (wordnum) { /* it's an accept state */
5137 /* record first match position */
5139 ST.firstpos = (U8*)locinput;
5144 ST.firstchars = charcount;
5147 if (!ST.nextword || wordnum < ST.nextword)
5148 ST.nextword = wordnum;
5149 ST.topword = wordnum;
5152 DEBUG_TRIE_EXECUTE_r({
5153 DUMP_EXEC_POS( (char *)uc, scan, utf8_target );
5154 PerlIO_printf( Perl_debug_log,
5155 "%*s %sState: %4"UVxf" Accepted: %c ",
5156 2+depth * 2, "", PL_colors[4],
5157 (UV)state, (accepted ? 'Y' : 'N'));
5160 /* read a char and goto next state */
5161 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
5163 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
5164 uscan, len, uvc, charid, foldlen,
5171 base + charid - 1 - trie->uniquecharcount)) >= 0)
5173 && ((U32)offset < trie->lasttrans)
5174 && trie->trans[offset].check == state)
5176 state = trie->trans[offset].next;
5187 DEBUG_TRIE_EXECUTE_r(
5188 PerlIO_printf( Perl_debug_log,
5189 "Charid:%3x CP:%4"UVxf" After State: %4"UVxf"%s\n",
5190 charid, uvc, (UV)state, PL_colors[5] );
5196 /* calculate total number of accept states */
5201 w = trie->wordinfo[w].prev;
5204 ST.accepted = accepted;
5208 PerlIO_printf( Perl_debug_log,
5209 "%*s %sgot %"IVdf" possible matches%s\n",
5210 REPORT_CODE_OFF + depth * 2, "",
5211 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
5213 goto trie_first_try; /* jump into the fail handler */
5216 NOT_REACHED; /* NOTREACHED */
5218 case TRIE_next_fail: /* we failed - try next alternative */
5222 REGCP_UNWIND(ST.cp);
5223 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
5225 if (!--ST.accepted) {
5227 PerlIO_printf( Perl_debug_log,
5228 "%*s %sTRIE failed...%s\n",
5229 REPORT_CODE_OFF+depth*2, "",
5236 /* Find next-highest word to process. Note that this code
5237 * is O(N^2) per trie run (O(N) per branch), so keep tight */
5240 U16 const nextword = ST.nextword;
5241 reg_trie_wordinfo * const wordinfo
5242 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
5243 for (word=ST.topword; word; word=wordinfo[word].prev) {
5244 if (word > nextword && (!min || word < min))
5257 ST.lastparen = rex->lastparen;
5258 ST.lastcloseparen = rex->lastcloseparen;
5262 /* find start char of end of current word */
5264 U32 chars; /* how many chars to skip */
5265 reg_trie_data * const trie
5266 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
5268 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
5270 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
5275 /* the hard option - fold each char in turn and find
5276 * its folded length (which may be different */
5277 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
5285 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
5293 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
5298 uvc = utf8n_to_uvchr(uscan, UTF8_MAXLEN, &len,
5314 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
5315 ? ST.jump[ST.nextword]
5319 PerlIO_printf( Perl_debug_log,
5320 "%*s %sTRIE matched word #%d, continuing%s\n",
5321 REPORT_CODE_OFF+depth*2, "",
5328 if (ST.accepted > 1 || has_cutgroup) {
5329 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
5331 NOT_REACHED; /* NOTREACHED */
5333 /* only one choice left - just continue */
5335 AV *const trie_words
5336 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
5337 SV ** const tmp = trie_words
5338 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL;
5339 SV *sv= tmp ? sv_newmortal() : NULL;
5341 PerlIO_printf( Perl_debug_log,
5342 "%*s %sonly one match left, short-circuiting: #%d <%s>%s\n",
5343 REPORT_CODE_OFF+depth*2, "", PL_colors[4],
5345 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
5346 PL_colors[0], PL_colors[1],
5347 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
5349 : "not compiled under -Dr",
5353 locinput = (char*)uc;
5354 continue; /* execute rest of RE */
5359 case EXACTL: /* /abc/l */
5360 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5362 /* Complete checking would involve going through every character
5363 * matched by the string to see if any is above latin1. But the
5364 * comparision otherwise might very well be a fast assembly
5365 * language routine, and I (khw) don't think slowing things down
5366 * just to check for this warning is worth it. So this just checks
5367 * the first character */
5368 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
5369 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
5372 case EXACT: { /* /abc/ */
5373 char *s = STRING(scan);
5375 if (utf8_target != is_utf8_pat) {
5376 /* The target and the pattern have differing utf8ness. */
5378 const char * const e = s + ln;
5381 /* The target is utf8, the pattern is not utf8.
5382 * Above-Latin1 code points can't match the pattern;
5383 * invariants match exactly, and the other Latin1 ones need
5384 * to be downgraded to a single byte in order to do the
5385 * comparison. (If we could be confident that the target
5386 * is not malformed, this could be refactored to have fewer
5387 * tests by just assuming that if the first bytes match, it
5388 * is an invariant, but there are tests in the test suite
5389 * dealing with (??{...}) which violate this) */
5391 if (l >= reginfo->strend
5392 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
5396 if (UTF8_IS_INVARIANT(*(U8*)l)) {
5403 if (TWO_BYTE_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
5413 /* The target is not utf8, the pattern is utf8. */
5415 if (l >= reginfo->strend
5416 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
5420 if (UTF8_IS_INVARIANT(*(U8*)s)) {
5427 if (TWO_BYTE_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
5439 /* The target and the pattern have the same utf8ness. */
5440 /* Inline the first character, for speed. */
5441 if (reginfo->strend - locinput < ln
5442 || UCHARAT(s) != nextchr
5443 || (ln > 1 && memNE(s, locinput, ln)))
5452 case EXACTFL: { /* /abc/il */
5454 const U8 * fold_array;
5456 U32 fold_utf8_flags;
5458 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5459 folder = foldEQ_locale;
5460 fold_array = PL_fold_locale;
5461 fold_utf8_flags = FOLDEQ_LOCALE;
5464 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
5465 is effectively /u; hence to match, target
5467 if (! utf8_target) {
5470 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S1_ALREADY_FOLDED
5471 | FOLDEQ_S1_FOLDS_SANE;
5472 folder = foldEQ_latin1;
5473 fold_array = PL_fold_latin1;
5476 case EXACTFU_SS: /* /\x{df}/iu */
5477 case EXACTFU: /* /abc/iu */
5478 folder = foldEQ_latin1;
5479 fold_array = PL_fold_latin1;
5480 fold_utf8_flags = is_utf8_pat ? FOLDEQ_S1_ALREADY_FOLDED : 0;
5483 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8
5485 assert(! is_utf8_pat);
5487 case EXACTFA: /* /abc/iaa */
5488 folder = foldEQ_latin1;
5489 fold_array = PL_fold_latin1;
5490 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
5493 case EXACTF: /* /abc/i This node only generated for
5494 non-utf8 patterns */
5495 assert(! is_utf8_pat);
5497 fold_array = PL_fold;
5498 fold_utf8_flags = 0;
5506 || state_num == EXACTFU_SS
5507 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
5509 /* Either target or the pattern are utf8, or has the issue where
5510 * the fold lengths may differ. */
5511 const char * const l = locinput;
5512 char *e = reginfo->strend;
5514 if (! foldEQ_utf8_flags(s, 0, ln, is_utf8_pat,
5515 l, &e, 0, utf8_target, fold_utf8_flags))
5523 /* Neither the target nor the pattern are utf8 */
5524 if (UCHARAT(s) != nextchr
5526 && UCHARAT(s) != fold_array[nextchr])
5530 if (reginfo->strend - locinput < ln)
5532 if (ln > 1 && ! folder(s, locinput, ln))
5538 case NBOUNDL: /* /\B/l */
5542 case BOUNDL: /* /\b/l */
5543 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5545 if (FLAGS(scan) != TRADITIONAL_BOUND) {
5546 if (! IN_UTF8_CTYPE_LOCALE) {
5547 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
5548 B_ON_NON_UTF8_LOCALE_IS_WRONG);
5554 if (locinput == reginfo->strbeg)
5555 ln = isWORDCHAR_LC('\n');
5557 ln = isWORDCHAR_LC_utf8(reghop3((U8*)locinput, -1,
5558 (U8*)(reginfo->strbeg)));
5560 n = (NEXTCHR_IS_EOS)
5561 ? isWORDCHAR_LC('\n')
5562 : isWORDCHAR_LC_utf8((U8*)locinput);
5564 else { /* Here the string isn't utf8 */
5565 ln = (locinput == reginfo->strbeg)
5566 ? isWORDCHAR_LC('\n')
5567 : isWORDCHAR_LC(UCHARAT(locinput - 1));
5568 n = (NEXTCHR_IS_EOS)
5569 ? isWORDCHAR_LC('\n')
5570 : isWORDCHAR_LC(nextchr);
5572 if (to_complement ^ (ln == n)) {
5577 case NBOUND: /* /\B/ */
5581 case BOUND: /* /\b/ */
5585 goto bound_ascii_match_only;
5587 case NBOUNDA: /* /\B/a */
5591 case BOUNDA: /* /\b/a */
5593 bound_ascii_match_only:
5594 /* Here the string isn't utf8, or is utf8 and only ascii characters
5595 * are to match \w. In the latter case looking at the byte just
5596 * prior to the current one may be just the final byte of a
5597 * multi-byte character. This is ok. There are two cases:
5598 * 1) it is a single byte character, and then the test is doing
5599 * just what it's supposed to.
5600 * 2) it is a multi-byte character, in which case the final byte is
5601 * never mistakable for ASCII, and so the test will say it is
5602 * not a word character, which is the correct answer. */
5603 ln = (locinput == reginfo->strbeg)
5604 ? isWORDCHAR_A('\n')
5605 : isWORDCHAR_A(UCHARAT(locinput - 1));
5606 n = (NEXTCHR_IS_EOS)
5607 ? isWORDCHAR_A('\n')
5608 : isWORDCHAR_A(nextchr);
5609 if (to_complement ^ (ln == n)) {
5614 case NBOUNDU: /* /\B/u */
5618 case BOUNDU: /* /\b/u */
5624 switch((bound_type) FLAGS(scan)) {
5625 case TRADITIONAL_BOUND:
5626 ln = (locinput == reginfo->strbeg)
5627 ? 0 /* isWORDCHAR_L1('\n') */
5628 : isWORDCHAR_utf8(reghop3((U8*)locinput, -1,
5629 (U8*)(reginfo->strbeg)));
5630 n = (NEXTCHR_IS_EOS)
5631 ? 0 /* isWORDCHAR_L1('\n') */
5632 : isWORDCHAR_utf8((U8*)locinput);
5633 match = cBOOL(ln != n);
5636 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
5637 match = TRUE; /* GCB always matches at begin and
5641 /* Find the gcb values of previous and current
5642 * chars, then see if is a break point */
5643 match = isGCB(getGCB_VAL_UTF8(
5644 reghop3((U8*)locinput,
5646 (U8*)(reginfo->strbeg)),
5647 (U8*) reginfo->strend),
5648 getGCB_VAL_UTF8((U8*) locinput,
5649 (U8*) reginfo->strend));
5653 case SB_BOUND: /* Always matches at begin and end */
5654 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
5658 match = isSB(getSB_VAL_UTF8(
5659 reghop3((U8*)locinput,
5661 (U8*)(reginfo->strbeg)),
5662 (U8*) reginfo->strend),
5663 getSB_VAL_UTF8((U8*) locinput,
5664 (U8*) reginfo->strend),
5665 (U8*) reginfo->strbeg,
5667 (U8*) reginfo->strend,
5673 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
5677 match = isWB(WB_UNKNOWN,
5679 reghop3((U8*)locinput,
5681 (U8*)(reginfo->strbeg)),
5682 (U8*) reginfo->strend),
5683 getWB_VAL_UTF8((U8*) locinput,
5684 (U8*) reginfo->strend),
5685 (U8*) reginfo->strbeg,
5687 (U8*) reginfo->strend,
5693 else { /* Not utf8 target */
5694 switch((bound_type) FLAGS(scan)) {
5695 case TRADITIONAL_BOUND:
5696 ln = (locinput == reginfo->strbeg)
5697 ? 0 /* isWORDCHAR_L1('\n') */
5698 : isWORDCHAR_L1(UCHARAT(locinput - 1));
5699 n = (NEXTCHR_IS_EOS)
5700 ? 0 /* isWORDCHAR_L1('\n') */
5701 : isWORDCHAR_L1(nextchr);
5702 match = cBOOL(ln != n);
5706 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
5707 match = TRUE; /* GCB always matches at begin and
5710 else { /* Only CR-LF combo isn't a GCB in 0-255
5712 match = UCHARAT(locinput - 1) != '\r'
5713 || UCHARAT(locinput) != '\n';
5717 case SB_BOUND: /* Always matches at begin and end */
5718 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
5722 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)),
5723 getSB_VAL_CP(UCHARAT(locinput)),
5724 (U8*) reginfo->strbeg,
5726 (U8*) reginfo->strend,
5732 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
5736 match = isWB(WB_UNKNOWN,
5737 getWB_VAL_CP(UCHARAT(locinput -1)),
5738 getWB_VAL_CP(UCHARAT(locinput)),
5739 (U8*) reginfo->strbeg,
5741 (U8*) reginfo->strend,
5748 if (to_complement ^ ! match) {
5753 case ANYOFL: /* /[abc]/l */
5754 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5756 case ANYOF: /* /[abc]/ */
5760 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
5763 locinput += UTF8SKIP(locinput);
5766 if (!REGINCLASS(rex, scan, (U8*)locinput))
5772 /* The argument (FLAGS) to all the POSIX node types is the class number
5775 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
5779 case POSIXL: /* \w or [:punct:] etc. under /l */
5780 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5784 /* Use isFOO_lc() for characters within Latin1. (Note that
5785 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
5786 * wouldn't be invariant) */
5787 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
5788 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
5792 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
5793 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
5794 (U8) TWO_BYTE_UTF8_TO_NATIVE(nextchr,
5795 *(locinput + 1))))))
5800 else { /* Here, must be an above Latin-1 code point */
5801 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
5802 goto utf8_posix_above_latin1;
5805 /* Here, must be utf8 */
5806 locinput += UTF8SKIP(locinput);
5809 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
5813 case POSIXD: /* \w or [:punct:] etc. under /d */
5819 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
5821 if (NEXTCHR_IS_EOS) {
5825 /* All UTF-8 variants match */
5826 if (! UTF8_IS_INVARIANT(nextchr)) {
5827 goto increment_locinput;
5833 case POSIXA: /* \w or [:punct:] etc. under /a */
5836 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
5837 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
5838 * character is a single byte */
5841 || ! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
5847 /* Here we are either not in utf8, or we matched a utf8-invariant,
5848 * so the next char is the next byte */
5852 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
5856 case POSIXU: /* \w or [:punct:] etc. under /u */
5858 if (NEXTCHR_IS_EOS) {
5862 /* Use _generic_isCC() for characters within Latin1. (Note that
5863 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
5864 * wouldn't be invariant) */
5865 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
5866 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
5873 else if (UTF8_IS_DOWNGRADEABLE_START(nextchr)) {
5874 if (! (to_complement
5875 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(nextchr,
5883 else { /* Handle above Latin-1 code points */
5884 utf8_posix_above_latin1:
5885 classnum = (_char_class_number) FLAGS(scan);
5886 if (classnum < _FIRST_NON_SWASH_CC) {
5888 /* Here, uses a swash to find such code points. Load if if
5889 * not done already */
5890 if (! PL_utf8_swash_ptrs[classnum]) {
5891 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
5892 PL_utf8_swash_ptrs[classnum]
5893 = _core_swash_init("utf8",
5896 PL_XPosix_ptrs[classnum], &flags);
5898 if (! (to_complement
5899 ^ cBOOL(swash_fetch(PL_utf8_swash_ptrs[classnum],
5900 (U8 *) locinput, TRUE))))
5905 else { /* Here, uses macros to find above Latin-1 code points */
5907 case _CC_ENUM_SPACE:
5908 if (! (to_complement
5909 ^ cBOOL(is_XPERLSPACE_high(locinput))))
5914 case _CC_ENUM_BLANK:
5915 if (! (to_complement
5916 ^ cBOOL(is_HORIZWS_high(locinput))))
5921 case _CC_ENUM_XDIGIT:
5922 if (! (to_complement
5923 ^ cBOOL(is_XDIGIT_high(locinput))))
5928 case _CC_ENUM_VERTSPACE:
5929 if (! (to_complement
5930 ^ cBOOL(is_VERTWS_high(locinput))))
5935 default: /* The rest, e.g. [:cntrl:], can't match
5937 if (! to_complement) {
5943 locinput += UTF8SKIP(locinput);
5947 case CLUMP: /* Match \X: logical Unicode character. This is defined as
5948 a Unicode extended Grapheme Cluster */
5951 if (! utf8_target) {
5953 /* Match either CR LF or '.', as all the other possibilities
5955 locinput++; /* Match the . or CR */
5956 if (nextchr == '\r' /* And if it was CR, and the next is LF,
5958 && locinput < reginfo->strend
5959 && UCHARAT(locinput) == '\n')
5966 /* Get the gcb type for the current character */
5967 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput,
5968 (U8*) reginfo->strend);
5970 /* Then scan through the input until we get to the first
5971 * character whose type is supposed to be a gcb with the
5972 * current character. (There is always a break at the
5974 locinput += UTF8SKIP(locinput);
5975 while (locinput < reginfo->strend) {
5976 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput,
5977 (U8*) reginfo->strend);
5978 if (isGCB(prev_gcb, cur_gcb)) {
5983 locinput += UTF8SKIP(locinput);
5990 case NREFFL: /* /\g{name}/il */
5991 { /* The capture buffer cases. The ones beginning with N for the
5992 named buffers just convert to the equivalent numbered and
5993 pretend they were called as the corresponding numbered buffer
5995 /* don't initialize these in the declaration, it makes C++
6000 const U8 *fold_array;
6003 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6004 folder = foldEQ_locale;
6005 fold_array = PL_fold_locale;
6007 utf8_fold_flags = FOLDEQ_LOCALE;
6010 case NREFFA: /* /\g{name}/iaa */
6011 folder = foldEQ_latin1;
6012 fold_array = PL_fold_latin1;
6014 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6017 case NREFFU: /* /\g{name}/iu */
6018 folder = foldEQ_latin1;
6019 fold_array = PL_fold_latin1;
6021 utf8_fold_flags = 0;
6024 case NREFF: /* /\g{name}/i */
6026 fold_array = PL_fold;
6028 utf8_fold_flags = 0;
6031 case NREF: /* /\g{name}/ */
6035 utf8_fold_flags = 0;
6038 /* For the named back references, find the corresponding buffer
6040 n = reg_check_named_buff_matched(rex,scan);
6045 goto do_nref_ref_common;
6047 case REFFL: /* /\1/il */
6048 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6049 folder = foldEQ_locale;
6050 fold_array = PL_fold_locale;
6051 utf8_fold_flags = FOLDEQ_LOCALE;
6054 case REFFA: /* /\1/iaa */
6055 folder = foldEQ_latin1;
6056 fold_array = PL_fold_latin1;
6057 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6060 case REFFU: /* /\1/iu */
6061 folder = foldEQ_latin1;
6062 fold_array = PL_fold_latin1;
6063 utf8_fold_flags = 0;
6066 case REFF: /* /\1/i */
6068 fold_array = PL_fold;
6069 utf8_fold_flags = 0;
6072 case REF: /* /\1/ */
6075 utf8_fold_flags = 0;
6079 n = ARG(scan); /* which paren pair */
6082 ln = rex->offs[n].start;
6083 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
6084 if (rex->lastparen < n || ln == -1)
6085 sayNO; /* Do not match unless seen CLOSEn. */
6086 if (ln == rex->offs[n].end)
6089 s = reginfo->strbeg + ln;
6090 if (type != REF /* REF can do byte comparison */
6091 && (utf8_target || type == REFFU || type == REFFL))
6093 char * limit = reginfo->strend;
6095 /* This call case insensitively compares the entire buffer
6096 * at s, with the current input starting at locinput, but
6097 * not going off the end given by reginfo->strend, and
6098 * returns in <limit> upon success, how much of the
6099 * current input was matched */
6100 if (! foldEQ_utf8_flags(s, NULL, rex->offs[n].end - ln, utf8_target,
6101 locinput, &limit, 0, utf8_target, utf8_fold_flags))
6109 /* Not utf8: Inline the first character, for speed. */
6110 if (!NEXTCHR_IS_EOS &&
6111 UCHARAT(s) != nextchr &&
6113 UCHARAT(s) != fold_array[nextchr]))
6115 ln = rex->offs[n].end - ln;
6116 if (locinput + ln > reginfo->strend)
6118 if (ln > 1 && (type == REF
6119 ? memNE(s, locinput, ln)
6120 : ! folder(s, locinput, ln)))
6126 case NOTHING: /* null op; e.g. the 'nothing' following
6127 * the '*' in m{(a+|b)*}' */
6129 case TAIL: /* placeholder while compiling (A|B|C) */
6133 #define ST st->u.eval
6138 regexp_internal *rei;
6139 regnode *startpoint;
6141 case GOSTART: /* (?R) */
6142 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
6143 if (cur_eval && cur_eval->locinput==locinput) {
6144 if (cur_eval->u.eval.close_paren == (U32)ARG(scan))
6145 Perl_croak(aTHX_ "Infinite recursion in regex");
6146 if ( ++nochange_depth > max_nochange_depth )
6148 "Pattern subroutine nesting without pos change"
6149 " exceeded limit in regex");
6156 if (OP(scan)==GOSUB) {
6157 startpoint = scan + ARG2L(scan);
6158 ST.close_paren = ARG(scan);
6160 startpoint = rei->program+1;
6164 /* Save all the positions seen so far. */
6165 ST.cp = regcppush(rex, 0, maxopenparen);
6166 REGCP_SET(ST.lastcp);
6168 /* and then jump to the code we share with EVAL */
6169 goto eval_recurse_doit;
6172 case EVAL: /* /(?{A})B/ /(??{A})B/ and /(?(?{A})X|Y)B/ */
6173 if (cur_eval && cur_eval->locinput==locinput) {
6174 if ( ++nochange_depth > max_nochange_depth )
6175 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
6180 /* execute the code in the {...} */
6184 OP * const oop = PL_op;
6185 COP * const ocurcop = PL_curcop;
6189 /* save *all* paren positions */
6190 regcppush(rex, 0, maxopenparen);
6191 REGCP_SET(runops_cp);
6194 caller_cv = find_runcv(NULL);
6198 if (rexi->data->what[n] == 'r') { /* code from an external qr */
6200 (REGEXP*)(rexi->data->data[n])
6203 nop = (OP*)rexi->data->data[n+1];
6205 else if (rexi->data->what[n] == 'l') { /* literal code */
6207 nop = (OP*)rexi->data->data[n];
6208 assert(CvDEPTH(newcv));
6211 /* literal with own CV */
6212 assert(rexi->data->what[n] == 'L');
6213 newcv = rex->qr_anoncv;
6214 nop = (OP*)rexi->data->data[n];
6217 /* normally if we're about to execute code from the same
6218 * CV that we used previously, we just use the existing
6219 * CX stack entry. However, its possible that in the
6220 * meantime we may have backtracked, popped from the save
6221 * stack, and undone the SAVECOMPPAD(s) associated with
6222 * PUSH_MULTICALL; in which case PL_comppad no longer
6223 * points to newcv's pad. */
6224 if (newcv != last_pushed_cv || PL_comppad != last_pad)
6226 U8 flags = (CXp_SUB_RE |
6227 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
6228 if (last_pushed_cv) {
6229 CHANGE_MULTICALL_FLAGS(newcv, flags);
6232 PUSH_MULTICALL_FLAGS(newcv, flags);
6234 last_pushed_cv = newcv;
6237 /* these assignments are just to silence compiler
6239 multicall_cop = NULL;
6242 last_pad = PL_comppad;
6244 /* the initial nextstate you would normally execute
6245 * at the start of an eval (which would cause error
6246 * messages to come from the eval), may be optimised
6247 * away from the execution path in the regex code blocks;
6248 * so manually set PL_curcop to it initially */
6250 OP *o = cUNOPx(nop)->op_first;
6251 assert(o->op_type == OP_NULL);
6252 if (o->op_targ == OP_SCOPE) {
6253 o = cUNOPo->op_first;
6256 assert(o->op_targ == OP_LEAVE);
6257 o = cUNOPo->op_first;
6258 assert(o->op_type == OP_ENTER);
6262 if (o->op_type != OP_STUB) {
6263 assert( o->op_type == OP_NEXTSTATE
6264 || o->op_type == OP_DBSTATE
6265 || (o->op_type == OP_NULL
6266 && ( o->op_targ == OP_NEXTSTATE
6267 || o->op_targ == OP_DBSTATE
6271 PL_curcop = (COP*)o;
6276 DEBUG_STATE_r( PerlIO_printf(Perl_debug_log,
6277 " re EVAL PL_op=0x%"UVxf"\n", PTR2UV(nop)) );
6279 rex->offs[0].end = locinput - reginfo->strbeg;
6280 if (reginfo->info_aux_eval->pos_magic)
6281 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
6282 reginfo->sv, reginfo->strbeg,
6283 locinput - reginfo->strbeg);
6286 SV *sv_mrk = get_sv("REGMARK", 1);
6287 sv_setsv(sv_mrk, sv_yes_mark);
6290 /* we don't use MULTICALL here as we want to call the
6291 * first op of the block of interest, rather than the
6292 * first op of the sub */
6293 before = (IV)(SP-PL_stack_base);
6295 CALLRUNOPS(aTHX); /* Scalar context. */
6297 if ((IV)(SP-PL_stack_base) == before)
6298 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
6304 /* before restoring everything, evaluate the returned
6305 * value, so that 'uninit' warnings don't use the wrong
6306 * PL_op or pad. Also need to process any magic vars
6307 * (e.g. $1) *before* parentheses are restored */
6312 if (logical == 0) /* (?{})/ */
6313 sv_setsv(save_scalar(PL_replgv), ret); /* $^R */
6314 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
6315 sw = cBOOL(SvTRUE(ret));
6318 else { /* /(??{}) */
6319 /* if its overloaded, let the regex compiler handle
6320 * it; otherwise extract regex, or stringify */
6321 if (SvGMAGICAL(ret))
6322 ret = sv_mortalcopy(ret);
6323 if (!SvAMAGIC(ret)) {
6327 if (SvTYPE(sv) == SVt_REGEXP)
6328 re_sv = (REGEXP*) sv;
6329 else if (SvSMAGICAL(ret)) {
6330 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
6332 re_sv = (REGEXP *) mg->mg_obj;
6335 /* force any undef warnings here */
6336 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
6337 ret = sv_mortalcopy(ret);
6338 (void) SvPV_force_nolen(ret);
6344 /* *** Note that at this point we don't restore
6345 * PL_comppad, (or pop the CxSUB) on the assumption it may
6346 * be used again soon. This is safe as long as nothing
6347 * in the regexp code uses the pad ! */
6349 PL_curcop = ocurcop;
6350 S_regcp_restore(aTHX_ rex, runops_cp, &maxopenparen);
6351 PL_curpm = PL_reg_curpm;
6357 /* only /(??{})/ from now on */
6360 /* extract RE object from returned value; compiling if
6364 re_sv = reg_temp_copy(NULL, re_sv);
6369 if (SvUTF8(ret) && IN_BYTES) {
6370 /* In use 'bytes': make a copy of the octet
6371 * sequence, but without the flag on */
6373 const char *const p = SvPV(ret, len);
6374 ret = newSVpvn_flags(p, len, SVs_TEMP);
6376 if (rex->intflags & PREGf_USE_RE_EVAL)
6377 pm_flags |= PMf_USE_RE_EVAL;
6379 /* if we got here, it should be an engine which
6380 * supports compiling code blocks and stuff */
6381 assert(rex->engine && rex->engine->op_comp);
6382 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
6383 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
6384 rex->engine, NULL, NULL,
6385 /* copy /msixn etc to inner pattern */
6390 & (SVs_TEMP | SVs_GMG | SVf_ROK))
6391 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
6392 /* This isn't a first class regexp. Instead, it's
6393 caching a regexp onto an existing, Perl visible
6395 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
6401 RXp_MATCH_COPIED_off(re);
6402 re->subbeg = rex->subbeg;
6403 re->sublen = rex->sublen;
6404 re->suboffset = rex->suboffset;
6405 re->subcoffset = rex->subcoffset;
6407 re->lastcloseparen = 0;
6410 debug_start_match(re_sv, utf8_target, locinput,
6411 reginfo->strend, "Matching embedded");
6413 startpoint = rei->program + 1;
6414 ST.close_paren = 0; /* only used for GOSUB */
6415 /* Save all the seen positions so far. */
6416 ST.cp = regcppush(rex, 0, maxopenparen);
6417 REGCP_SET(ST.lastcp);
6418 /* and set maxopenparen to 0, since we are starting a "fresh" match */
6420 /* run the pattern returned from (??{...}) */
6422 eval_recurse_doit: /* Share code with GOSUB below this line
6423 * At this point we expect the stack context to be
6424 * set up correctly */
6426 /* invalidate the S-L poscache. We're now executing a
6427 * different set of WHILEM ops (and their associated
6428 * indexes) against the same string, so the bits in the
6429 * cache are meaningless. Setting maxiter to zero forces
6430 * the cache to be invalidated and zeroed before reuse.
6431 * XXX This is too dramatic a measure. Ideally we should
6432 * save the old cache and restore when running the outer
6434 reginfo->poscache_maxiter = 0;
6436 /* the new regexp might have a different is_utf8_pat than we do */
6437 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
6439 ST.prev_rex = rex_sv;
6440 ST.prev_curlyx = cur_curlyx;
6442 SET_reg_curpm(rex_sv);
6447 ST.prev_eval = cur_eval;
6449 /* now continue from first node in postoned RE */
6450 PUSH_YES_STATE_GOTO(EVAL_AB, startpoint, locinput);
6452 NOT_REACHED; /* NOTREACHED */
6455 case EVAL_AB: /* cleanup after a successful (??{A})B */
6456 /* note: this is called twice; first after popping B, then A */
6457 rex_sv = ST.prev_rex;
6458 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6459 SET_reg_curpm(rex_sv);
6460 rex = ReANY(rex_sv);
6461 rexi = RXi_GET(rex);
6463 /* preserve $^R across LEAVE's. See Bug 121070. */
6464 SV *save_sv= GvSV(PL_replgv);
6465 SvREFCNT_inc(save_sv);
6466 regcpblow(ST.cp); /* LEAVE in disguise */
6467 sv_setsv(GvSV(PL_replgv), save_sv);
6468 SvREFCNT_dec(save_sv);
6470 cur_eval = ST.prev_eval;
6471 cur_curlyx = ST.prev_curlyx;
6473 /* Invalidate cache. See "invalidate" comment above. */
6474 reginfo->poscache_maxiter = 0;
6475 if ( nochange_depth )
6480 case EVAL_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
6481 /* note: this is called twice; first after popping B, then A */
6482 rex_sv = ST.prev_rex;
6483 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
6484 SET_reg_curpm(rex_sv);
6485 rex = ReANY(rex_sv);
6486 rexi = RXi_GET(rex);
6488 REGCP_UNWIND(ST.lastcp);
6489 regcppop(rex, &maxopenparen);
6490 cur_eval = ST.prev_eval;
6491 cur_curlyx = ST.prev_curlyx;
6492 /* Invalidate cache. See "invalidate" comment above. */
6493 reginfo->poscache_maxiter = 0;
6494 if ( nochange_depth )
6500 n = ARG(scan); /* which paren pair */
6501 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
6502 if (n > maxopenparen)
6504 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log,
6505 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf" tmp; maxopenparen=%"UVuf"\n",
6509 (IV)rex->offs[n].start_tmp,
6515 /* XXX really need to log other places start/end are set too */
6516 #define CLOSE_CAPTURE \
6517 rex->offs[n].start = rex->offs[n].start_tmp; \
6518 rex->offs[n].end = locinput - reginfo->strbeg; \
6519 DEBUG_BUFFERS_r(PerlIO_printf(Perl_debug_log, \
6520 "rex=0x%"UVxf" offs=0x%"UVxf": \\%"UVuf": set %"IVdf"..%"IVdf"\n", \
6522 PTR2UV(rex->offs), \
6524 (IV)rex->offs[n].start, \
6525 (IV)rex->offs[n].end \
6529 n = ARG(scan); /* which paren pair */
6531 if (n > rex->lastparen)
6533 rex->lastcloseparen = n;
6534 if (cur_eval && cur_eval->u.eval.close_paren == n) {
6539 case ACCEPT: /* (*ACCEPT) */
6543 cursor && OP(cursor)!=END;
6544 cursor=regnext(cursor))
6546 if ( OP(cursor)==CLOSE ){
6548 if ( n <= lastopen ) {
6550 if (n > rex->lastparen)
6552 rex->lastcloseparen = n;
6553 if ( n == ARG(scan) || (cur_eval &&
6554 cur_eval->u.eval.close_paren == n))
6563 case GROUPP: /* (?(1)) */
6564 n = ARG(scan); /* which paren pair */
6565 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
6568 case NGROUPP: /* (?(<name>)) */
6569 /* reg_check_named_buff_matched returns 0 for no match */
6570 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
6573 case INSUBP: /* (?(R)) */
6575 sw = (cur_eval && (!n || cur_eval->u.eval.close_paren == n));
6578 case DEFINEP: /* (?(DEFINE)) */
6582 case IFTHEN: /* (?(cond)A|B) */
6583 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
6585 next = NEXTOPER(NEXTOPER(scan));
6587 next = scan + ARG(scan);
6588 if (OP(next) == IFTHEN) /* Fake one. */
6589 next = NEXTOPER(NEXTOPER(next));
6593 case LOGICAL: /* modifier for EVAL and IFMATCH */
6594 logical = scan->flags;
6597 /*******************************************************************
6599 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
6600 pattern, where A and B are subpatterns. (For simple A, CURLYM or
6601 STAR/PLUS/CURLY/CURLYN are used instead.)
6603 A*B is compiled as <CURLYX><A><WHILEM><B>
6605 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
6606 state, which contains the current count, initialised to -1. It also sets
6607 cur_curlyx to point to this state, with any previous value saved in the
6610 CURLYX then jumps straight to the WHILEM op, rather than executing A,
6611 since the pattern may possibly match zero times (i.e. it's a while {} loop
6612 rather than a do {} while loop).
6614 Each entry to WHILEM represents a successful match of A. The count in the
6615 CURLYX block is incremented, another WHILEM state is pushed, and execution
6616 passes to A or B depending on greediness and the current count.
6618 For example, if matching against the string a1a2a3b (where the aN are
6619 substrings that match /A/), then the match progresses as follows: (the
6620 pushed states are interspersed with the bits of strings matched so far):
6623 <CURLYX cnt=0><WHILEM>
6624 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
6625 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
6626 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
6627 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
6629 (Contrast this with something like CURLYM, which maintains only a single
6633 a1 <CURLYM cnt=1> a2
6634 a1 a2 <CURLYM cnt=2> a3
6635 a1 a2 a3 <CURLYM cnt=3> b
6638 Each WHILEM state block marks a point to backtrack to upon partial failure
6639 of A or B, and also contains some minor state data related to that
6640 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
6641 overall state, such as the count, and pointers to the A and B ops.
6643 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
6644 must always point to the *current* CURLYX block, the rules are:
6646 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
6647 and set cur_curlyx to point the new block.
6649 When popping the CURLYX block after a successful or unsuccessful match,
6650 restore the previous cur_curlyx.
6652 When WHILEM is about to execute B, save the current cur_curlyx, and set it
6653 to the outer one saved in the CURLYX block.
6655 When popping the WHILEM block after a successful or unsuccessful B match,
6656 restore the previous cur_curlyx.
6658 Here's an example for the pattern (AI* BI)*BO
6659 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
6662 curlyx backtrack stack
6663 ------ ---------------
6665 CO <CO prev=NULL> <WO>
6666 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
6667 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
6668 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
6670 At this point the pattern succeeds, and we work back down the stack to
6671 clean up, restoring as we go:
6673 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
6674 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
6675 CO <CO prev=NULL> <WO>
6678 *******************************************************************/
6680 #define ST st->u.curlyx
6682 case CURLYX: /* start of /A*B/ (for complex A) */
6684 /* No need to save/restore up to this paren */
6685 I32 parenfloor = scan->flags;
6687 assert(next); /* keep Coverity happy */
6688 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
6691 /* XXXX Probably it is better to teach regpush to support
6692 parenfloor > maxopenparen ... */
6693 if (parenfloor > (I32)rex->lastparen)
6694 parenfloor = rex->lastparen; /* Pessimization... */
6696 ST.prev_curlyx= cur_curlyx;
6698 ST.cp = PL_savestack_ix;
6700 /* these fields contain the state of the current curly.
6701 * they are accessed by subsequent WHILEMs */
6702 ST.parenfloor = parenfloor;
6707 ST.count = -1; /* this will be updated by WHILEM */
6708 ST.lastloc = NULL; /* this will be updated by WHILEM */
6710 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
6712 NOT_REACHED; /* NOTREACHED */
6715 case CURLYX_end: /* just finished matching all of A*B */
6716 cur_curlyx = ST.prev_curlyx;
6719 NOT_REACHED; /* NOTREACHED */
6721 case CURLYX_end_fail: /* just failed to match all of A*B */
6723 cur_curlyx = ST.prev_curlyx;
6726 NOT_REACHED; /* NOTREACHED */
6730 #define ST st->u.whilem
6732 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
6734 /* see the discussion above about CURLYX/WHILEM */
6739 assert(cur_curlyx); /* keep Coverity happy */
6741 min = ARG1(cur_curlyx->u.curlyx.me);
6742 max = ARG2(cur_curlyx->u.curlyx.me);
6743 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
6744 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
6745 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
6746 ST.cache_offset = 0;
6750 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
6751 "%*s whilem: matched %ld out of %d..%d\n",
6752 REPORT_CODE_OFF+depth*2, "", (long)n, min, max)
6755 /* First just match a string of min A's. */
6758 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6760 cur_curlyx->u.curlyx.lastloc = locinput;
6761 REGCP_SET(ST.lastcp);
6763 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
6765 NOT_REACHED; /* NOTREACHED */
6768 /* If degenerate A matches "", assume A done. */
6770 if (locinput == cur_curlyx->u.curlyx.lastloc) {
6771 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
6772 "%*s whilem: empty match detected, trying continuation...\n",
6773 REPORT_CODE_OFF+depth*2, "")
6775 goto do_whilem_B_max;
6778 /* super-linear cache processing.
6780 * The idea here is that for certain types of CURLYX/WHILEM -
6781 * principally those whose upper bound is infinity (and
6782 * excluding regexes that have things like \1 and other very
6783 * non-regular expresssiony things), then if a pattern like
6784 * /....A*.../ fails and we backtrack to the WHILEM, then we
6785 * make a note that this particular WHILEM op was at string
6786 * position 47 (say) when the rest of pattern failed. Then, if
6787 * we ever find ourselves back at that WHILEM, and at string
6788 * position 47 again, we can just fail immediately rather than
6789 * running the rest of the pattern again.
6791 * This is very handy when patterns start to go
6792 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
6793 * with a combinatorial explosion of backtracking.
6795 * The cache is implemented as a bit array, with one bit per
6796 * string byte position per WHILEM op (up to 16) - so its
6797 * between 0.25 and 2x the string size.
6799 * To avoid allocating a poscache buffer every time, we do an
6800 * initially countdown; only after we have executed a WHILEM
6801 * op (string-length x #WHILEMs) times do we allocate the
6804 * The top 4 bits of scan->flags byte say how many different
6805 * relevant CURLLYX/WHILEM op pairs there are, while the
6806 * bottom 4-bits is the identifying index number of this
6812 if (!reginfo->poscache_maxiter) {
6813 /* start the countdown: Postpone detection until we
6814 * know the match is not *that* much linear. */
6815 reginfo->poscache_maxiter
6816 = (reginfo->strend - reginfo->strbeg + 1)
6818 /* possible overflow for long strings and many CURLYX's */
6819 if (reginfo->poscache_maxiter < 0)
6820 reginfo->poscache_maxiter = I32_MAX;
6821 reginfo->poscache_iter = reginfo->poscache_maxiter;
6824 if (reginfo->poscache_iter-- == 0) {
6825 /* initialise cache */
6826 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
6827 regmatch_info_aux *const aux = reginfo->info_aux;
6828 if (aux->poscache) {
6829 if ((SSize_t)reginfo->poscache_size < size) {
6830 Renew(aux->poscache, size, char);
6831 reginfo->poscache_size = size;
6833 Zero(aux->poscache, size, char);
6836 reginfo->poscache_size = size;
6837 Newxz(aux->poscache, size, char);
6839 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
6840 "%swhilem: Detected a super-linear match, switching on caching%s...\n",
6841 PL_colors[4], PL_colors[5])
6845 if (reginfo->poscache_iter < 0) {
6846 /* have we already failed at this position? */
6847 SSize_t offset, mask;
6849 reginfo->poscache_iter = -1; /* stop eventual underflow */
6850 offset = (scan->flags & 0xf) - 1
6851 + (locinput - reginfo->strbeg)
6853 mask = 1 << (offset % 8);
6855 if (reginfo->info_aux->poscache[offset] & mask) {
6856 DEBUG_EXECUTE_r( PerlIO_printf(Perl_debug_log,
6857 "%*s whilem: (cache) already tried at this position...\n",
6858 REPORT_CODE_OFF+depth*2, "")
6860 sayNO; /* cache records failure */
6862 ST.cache_offset = offset;
6863 ST.cache_mask = mask;
6867 /* Prefer B over A for minimal matching. */
6869 if (cur_curlyx->u.curlyx.minmod) {
6870 ST.save_curlyx = cur_curlyx;
6871 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
6872 ST.cp = regcppush(rex, ST.save_curlyx->u.curlyx.parenfloor,
6874 REGCP_SET(ST.lastcp);
6875 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
6878 NOT_REACHED; /* NOTREACHED */
6881 /* Prefer A over B for maximal matching. */
6883 if (n < max) { /* More greed allowed? */
6884 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6886 cur_curlyx->u.curlyx.lastloc = locinput;
6887 REGCP_SET(ST.lastcp);
6888 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
6890 NOT_REACHED; /* NOTREACHED */
6892 goto do_whilem_B_max;
6895 NOT_REACHED; /* NOTREACHED */
6897 case WHILEM_B_min: /* just matched B in a minimal match */
6898 case WHILEM_B_max: /* just matched B in a maximal match */
6899 cur_curlyx = ST.save_curlyx;
6902 NOT_REACHED; /* NOTREACHED */
6904 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
6905 cur_curlyx = ST.save_curlyx;
6906 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
6907 cur_curlyx->u.curlyx.count--;
6910 NOT_REACHED; /* NOTREACHED */
6912 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
6914 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
6915 REGCP_UNWIND(ST.lastcp);
6916 regcppop(rex, &maxopenparen);
6917 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
6918 cur_curlyx->u.curlyx.count--;
6921 NOT_REACHED; /* NOTREACHED */
6923 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
6924 REGCP_UNWIND(ST.lastcp);
6925 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
6926 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6927 "%*s whilem: failed, trying continuation...\n",
6928 REPORT_CODE_OFF+depth*2, "")
6931 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6932 && ckWARN(WARN_REGEXP)
6933 && !reginfo->warned)
6935 reginfo->warned = TRUE;
6936 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6937 "Complex regular subexpression recursion limit (%d) "
6943 ST.save_curlyx = cur_curlyx;
6944 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
6945 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
6948 NOT_REACHED; /* NOTREACHED */
6950 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
6951 cur_curlyx = ST.save_curlyx;
6952 REGCP_UNWIND(ST.lastcp);
6953 regcppop(rex, &maxopenparen);
6955 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
6956 /* Maximum greed exceeded */
6957 if (cur_curlyx->u.curlyx.count >= REG_INFTY
6958 && ckWARN(WARN_REGEXP)
6959 && !reginfo->warned)
6961 reginfo->warned = TRUE;
6962 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
6963 "Complex regular subexpression recursion "
6964 "limit (%d) exceeded",
6967 cur_curlyx->u.curlyx.count--;
6971 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
6972 "%*s trying longer...\n", REPORT_CODE_OFF+depth*2, "")
6974 /* Try grabbing another A and see if it helps. */
6975 cur_curlyx->u.curlyx.lastloc = locinput;
6976 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
6978 REGCP_SET(ST.lastcp);
6979 PUSH_STATE_GOTO(WHILEM_A_min,
6980 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
6983 NOT_REACHED; /* NOTREACHED */
6986 #define ST st->u.branch
6988 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
6989 next = scan + ARG(scan);
6992 scan = NEXTOPER(scan);
6995 case BRANCH: /* /(...|A|...)/ */
6996 scan = NEXTOPER(scan); /* scan now points to inner node */
6997 ST.lastparen = rex->lastparen;
6998 ST.lastcloseparen = rex->lastcloseparen;
6999 ST.next_branch = next;
7002 /* Now go into the branch */
7004 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
7006 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
7009 NOT_REACHED; /* NOTREACHED */
7011 case CUTGROUP: /* /(*THEN)/ */
7012 sv_yes_mark = st->u.mark.mark_name = scan->flags ? NULL :
7013 MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7014 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
7016 NOT_REACHED; /* NOTREACHED */
7018 case CUTGROUP_next_fail:
7021 if (st->u.mark.mark_name)
7022 sv_commit = st->u.mark.mark_name;
7025 NOT_REACHED; /* NOTREACHED */
7030 NOT_REACHED; /* NOTREACHED */
7032 case BRANCH_next_fail: /* that branch failed; try the next, if any */
7037 REGCP_UNWIND(ST.cp);
7038 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7039 scan = ST.next_branch;
7040 /* no more branches? */
7041 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
7043 PerlIO_printf( Perl_debug_log,
7044 "%*s %sBRANCH failed...%s\n",
7045 REPORT_CODE_OFF+depth*2, "",
7051 continue; /* execute next BRANCH[J] op */
7054 case MINMOD: /* next op will be non-greedy, e.g. A*? */
7059 #define ST st->u.curlym
7061 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
7063 /* This is an optimisation of CURLYX that enables us to push
7064 * only a single backtracking state, no matter how many matches
7065 * there are in {m,n}. It relies on the pattern being constant
7066 * length, with no parens to influence future backrefs
7070 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7072 ST.lastparen = rex->lastparen;
7073 ST.lastcloseparen = rex->lastcloseparen;
7075 /* if paren positive, emulate an OPEN/CLOSE around A */
7077 U32 paren = ST.me->flags;
7078 if (paren > maxopenparen)
7079 maxopenparen = paren;
7080 scan += NEXT_OFF(scan); /* Skip former OPEN. */
7088 ST.c1 = CHRTEST_UNINIT;
7091 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
7094 curlym_do_A: /* execute the A in /A{m,n}B/ */
7095 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
7097 NOT_REACHED; /* NOTREACHED */
7099 case CURLYM_A: /* we've just matched an A */
7101 /* after first match, determine A's length: u.curlym.alen */
7102 if (ST.count == 1) {
7103 if (reginfo->is_utf8_target) {
7104 char *s = st->locinput;
7105 while (s < locinput) {
7111 ST.alen = locinput - st->locinput;
7114 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
7117 PerlIO_printf(Perl_debug_log,
7118 "%*s CURLYM now matched %"IVdf" times, len=%"IVdf"...\n",
7119 (int)(REPORT_CODE_OFF+(depth*2)), "",
7120 (IV) ST.count, (IV)ST.alen)
7123 if (cur_eval && cur_eval->u.eval.close_paren &&
7124 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
7128 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
7129 if ( max == REG_INFTY || ST.count < max )
7130 goto curlym_do_A; /* try to match another A */
7132 goto curlym_do_B; /* try to match B */
7134 case CURLYM_A_fail: /* just failed to match an A */
7135 REGCP_UNWIND(ST.cp);
7137 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
7138 || (cur_eval && cur_eval->u.eval.close_paren &&
7139 cur_eval->u.eval.close_paren == (U32)ST.me->flags))
7142 curlym_do_B: /* execute the B in /A{m,n}B/ */
7143 if (ST.c1 == CHRTEST_UNINIT) {
7144 /* calculate c1 and c2 for possible match of 1st char
7145 * following curly */
7146 ST.c1 = ST.c2 = CHRTEST_VOID;
7148 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
7149 regnode *text_node = ST.B;
7150 if (! HAS_TEXT(text_node))
7151 FIND_NEXT_IMPT(text_node);
7154 (HAS_TEXT(text_node) && PL_regkind[OP(text_node)] == EXACT)
7156 But the former is redundant in light of the latter.
7158 if this changes back then the macro for
7159 IS_TEXT and friends need to change.
7161 if (PL_regkind[OP(text_node)] == EXACT) {
7162 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
7163 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
7173 PerlIO_printf(Perl_debug_log,
7174 "%*s CURLYM trying tail with matches=%"IVdf"...\n",
7175 (int)(REPORT_CODE_OFF+(depth*2)),
7178 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
7179 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
7180 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
7181 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
7183 /* simulate B failing */
7185 PerlIO_printf(Perl_debug_log,
7186 "%*s CURLYM Fast bail next target=0x%"UVXf" c1=0x%"UVXf" c2=0x%"UVXf"\n",
7187 (int)(REPORT_CODE_OFF+(depth*2)),"",
7188 valid_utf8_to_uvchr((U8 *) locinput, NULL),
7189 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
7190 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
7192 state_num = CURLYM_B_fail;
7193 goto reenter_switch;
7196 else if (nextchr != ST.c1 && nextchr != ST.c2) {
7197 /* simulate B failing */
7199 PerlIO_printf(Perl_debug_log,
7200 "%*s CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
7201 (int)(REPORT_CODE_OFF+(depth*2)),"",
7202 (int) nextchr, ST.c1, ST.c2)
7204 state_num = CURLYM_B_fail;
7205 goto reenter_switch;
7210 /* emulate CLOSE: mark current A as captured */
7211 I32 paren = ST.me->flags;
7213 rex->offs[paren].start
7214 = HOPc(locinput, -ST.alen) - reginfo->strbeg;
7215 rex->offs[paren].end = locinput - reginfo->strbeg;
7216 if ((U32)paren > rex->lastparen)
7217 rex->lastparen = paren;
7218 rex->lastcloseparen = paren;
7221 rex->offs[paren].end = -1;
7222 if (cur_eval && cur_eval->u.eval.close_paren &&
7223 cur_eval->u.eval.close_paren == (U32)ST.me->flags)
7232 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
7234 NOT_REACHED; /* NOTREACHED */
7236 case CURLYM_B_fail: /* just failed to match a B */
7237 REGCP_UNWIND(ST.cp);
7238 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7240 I32 max = ARG2(ST.me);
7241 if (max != REG_INFTY && ST.count == max)
7243 goto curlym_do_A; /* try to match a further A */
7245 /* backtrack one A */
7246 if (ST.count == ARG1(ST.me) /* min */)
7249 SET_locinput(HOPc(locinput, -ST.alen));
7250 goto curlym_do_B; /* try to match B */
7253 #define ST st->u.curly
7255 #define CURLY_SETPAREN(paren, success) \
7258 rex->offs[paren].start = HOPc(locinput, -1) - reginfo->strbeg; \
7259 rex->offs[paren].end = locinput - reginfo->strbeg; \
7260 if (paren > rex->lastparen) \
7261 rex->lastparen = paren; \
7262 rex->lastcloseparen = paren; \
7265 rex->offs[paren].end = -1; \
7266 rex->lastparen = ST.lastparen; \
7267 rex->lastcloseparen = ST.lastcloseparen; \
7271 case STAR: /* /A*B/ where A is width 1 char */
7275 scan = NEXTOPER(scan);
7278 case PLUS: /* /A+B/ where A is width 1 char */
7282 scan = NEXTOPER(scan);
7285 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
7286 ST.paren = scan->flags; /* Which paren to set */
7287 ST.lastparen = rex->lastparen;
7288 ST.lastcloseparen = rex->lastcloseparen;
7289 if (ST.paren > maxopenparen)
7290 maxopenparen = ST.paren;
7291 ST.min = ARG1(scan); /* min to match */
7292 ST.max = ARG2(scan); /* max to match */
7293 if (cur_eval && cur_eval->u.eval.close_paren &&
7294 cur_eval->u.eval.close_paren == (U32)ST.paren) {
7298 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
7301 case CURLY: /* /A{m,n}B/ where A is width 1 char */
7303 ST.min = ARG1(scan); /* min to match */
7304 ST.max = ARG2(scan); /* max to match */
7305 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
7308 * Lookahead to avoid useless match attempts
7309 * when we know what character comes next.
7311 * Used to only do .*x and .*?x, but now it allows
7312 * for )'s, ('s and (?{ ... })'s to be in the way
7313 * of the quantifier and the EXACT-like node. -- japhy
7316 assert(ST.min <= ST.max);
7317 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
7318 ST.c1 = ST.c2 = CHRTEST_VOID;
7321 regnode *text_node = next;
7323 if (! HAS_TEXT(text_node))
7324 FIND_NEXT_IMPT(text_node);
7326 if (! HAS_TEXT(text_node))
7327 ST.c1 = ST.c2 = CHRTEST_VOID;
7329 if ( PL_regkind[OP(text_node)] != EXACT ) {
7330 ST.c1 = ST.c2 = CHRTEST_VOID;
7334 /* Currently we only get here when
7336 PL_rekind[OP(text_node)] == EXACT
7338 if this changes back then the macro for IS_TEXT and
7339 friends need to change. */
7340 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
7341 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
7353 char *li = locinput;
7356 regrepeat(rex, &li, ST.A, reginfo, ST.min, depth)
7362 if (ST.c1 == CHRTEST_VOID)
7363 goto curly_try_B_min;
7365 ST.oldloc = locinput;
7367 /* set ST.maxpos to the furthest point along the
7368 * string that could possibly match */
7369 if (ST.max == REG_INFTY) {
7370 ST.maxpos = reginfo->strend - 1;
7372 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
7375 else if (utf8_target) {
7376 int m = ST.max - ST.min;
7377 for (ST.maxpos = locinput;
7378 m >0 && ST.maxpos < reginfo->strend; m--)
7379 ST.maxpos += UTF8SKIP(ST.maxpos);
7382 ST.maxpos = locinput + ST.max - ST.min;
7383 if (ST.maxpos >= reginfo->strend)
7384 ST.maxpos = reginfo->strend - 1;
7386 goto curly_try_B_min_known;
7390 /* avoid taking address of locinput, so it can remain
7392 char *li = locinput;
7393 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max, depth);
7394 if (ST.count < ST.min)
7397 if ((ST.count > ST.min)
7398 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
7400 /* A{m,n} must come at the end of the string, there's
7401 * no point in backing off ... */
7403 /* ...except that $ and \Z can match before *and* after
7404 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
7405 We may back off by one in this case. */
7406 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
7410 goto curly_try_B_max;
7413 NOT_REACHED; /* NOTREACHED */
7415 case CURLY_B_min_known_fail:
7416 /* failed to find B in a non-greedy match where c1,c2 valid */
7418 REGCP_UNWIND(ST.cp);
7420 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7422 /* Couldn't or didn't -- move forward. */
7423 ST.oldloc = locinput;
7425 locinput += UTF8SKIP(locinput);
7429 curly_try_B_min_known:
7430 /* find the next place where 'B' could work, then call B */
7434 n = (ST.oldloc == locinput) ? 0 : 1;
7435 if (ST.c1 == ST.c2) {
7436 /* set n to utf8_distance(oldloc, locinput) */
7437 while (locinput <= ST.maxpos
7438 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
7440 locinput += UTF8SKIP(locinput);
7445 /* set n to utf8_distance(oldloc, locinput) */
7446 while (locinput <= ST.maxpos
7447 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
7448 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
7450 locinput += UTF8SKIP(locinput);
7455 else { /* Not utf8_target */
7456 if (ST.c1 == ST.c2) {
7457 while (locinput <= ST.maxpos &&
7458 UCHARAT(locinput) != ST.c1)
7462 while (locinput <= ST.maxpos
7463 && UCHARAT(locinput) != ST.c1
7464 && UCHARAT(locinput) != ST.c2)
7467 n = locinput - ST.oldloc;
7469 if (locinput > ST.maxpos)
7472 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
7473 * at b; check that everything between oldloc and
7474 * locinput matches */
7475 char *li = ST.oldloc;
7477 if (regrepeat(rex, &li, ST.A, reginfo, n, depth) < n)
7479 assert(n == REG_INFTY || locinput == li);
7481 CURLY_SETPAREN(ST.paren, ST.count);
7482 if (cur_eval && cur_eval->u.eval.close_paren &&
7483 cur_eval->u.eval.close_paren == (U32)ST.paren) {
7486 PUSH_STATE_GOTO(CURLY_B_min_known, ST.B, locinput);
7489 NOT_REACHED; /* NOTREACHED */
7491 case CURLY_B_min_fail:
7492 /* failed to find B in a non-greedy match where c1,c2 invalid */
7494 REGCP_UNWIND(ST.cp);
7496 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7498 /* failed -- move forward one */
7500 char *li = locinput;
7501 if (!regrepeat(rex, &li, ST.A, reginfo, 1, depth)) {
7508 if (ST.count <= ST.max || (ST.max == REG_INFTY &&
7509 ST.count > 0)) /* count overflow ? */
7512 CURLY_SETPAREN(ST.paren, ST.count);
7513 if (cur_eval && cur_eval->u.eval.close_paren &&
7514 cur_eval->u.eval.close_paren == (U32)ST.paren) {
7517 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
7522 NOT_REACHED; /* NOTREACHED */
7525 /* a successful greedy match: now try to match B */
7526 if (cur_eval && cur_eval->u.eval.close_paren &&
7527 cur_eval->u.eval.close_paren == (U32)ST.paren) {
7531 bool could_match = locinput < reginfo->strend;
7533 /* If it could work, try it. */
7534 if (ST.c1 != CHRTEST_VOID && could_match) {
7535 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
7537 could_match = memEQ(locinput,
7542 UTF8SKIP(locinput));
7545 could_match = UCHARAT(locinput) == ST.c1
7546 || UCHARAT(locinput) == ST.c2;
7549 if (ST.c1 == CHRTEST_VOID || could_match) {
7550 CURLY_SETPAREN(ST.paren, ST.count);
7551 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
7553 NOT_REACHED; /* NOTREACHED */
7558 case CURLY_B_max_fail:
7559 /* failed to find B in a greedy match */
7561 REGCP_UNWIND(ST.cp);
7563 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
7566 if (--ST.count < ST.min)
7568 locinput = HOPc(locinput, -1);
7569 goto curly_try_B_max;
7573 case END: /* last op of main pattern */
7576 /* we've just finished A in /(??{A})B/; now continue with B */
7578 st->u.eval.prev_rex = rex_sv; /* inner */
7580 /* Save *all* the positions. */
7581 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
7582 rex_sv = cur_eval->u.eval.prev_rex;
7583 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7584 SET_reg_curpm(rex_sv);
7585 rex = ReANY(rex_sv);
7586 rexi = RXi_GET(rex);
7587 cur_curlyx = cur_eval->u.eval.prev_curlyx;
7589 REGCP_SET(st->u.eval.lastcp);
7591 /* Restore parens of the outer rex without popping the
7593 S_regcp_restore(aTHX_ rex, cur_eval->u.eval.lastcp,
7596 st->u.eval.prev_eval = cur_eval;
7597 cur_eval = cur_eval->u.eval.prev_eval;
7599 PerlIO_printf(Perl_debug_log, "%*s EVAL trying tail ... %"UVxf"\n",
7600 REPORT_CODE_OFF+depth*2, "",PTR2UV(cur_eval)););
7601 if ( nochange_depth )
7604 PUSH_YES_STATE_GOTO(EVAL_AB, st->u.eval.prev_eval->u.eval.B,
7605 locinput); /* match B */
7608 if (locinput < reginfo->till) {
7609 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log,
7610 "%sMatch possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
7612 (long)(locinput - startpos),
7613 (long)(reginfo->till - startpos),
7616 sayNO_SILENT; /* Cannot match: too short. */
7618 sayYES; /* Success! */
7620 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
7622 PerlIO_printf(Perl_debug_log,
7623 "%*s %ssubpattern success...%s\n",
7624 REPORT_CODE_OFF+depth*2, "", PL_colors[4], PL_colors[5]));
7625 sayYES; /* Success! */
7628 #define ST st->u.ifmatch
7633 case SUSPEND: /* (?>A) */
7635 newstart = locinput;
7638 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
7640 goto ifmatch_trivial_fail_test;
7642 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
7644 ifmatch_trivial_fail_test:
7646 char * const s = HOPBACKc(locinput, scan->flags);
7651 sw = 1 - cBOOL(ST.wanted);
7655 next = scan + ARG(scan);
7663 newstart = locinput;
7667 ST.logical = logical;
7668 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
7670 /* execute body of (?...A) */
7671 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
7673 NOT_REACHED; /* NOTREACHED */
7676 case IFMATCH_A_fail: /* body of (?...A) failed */
7677 ST.wanted = !ST.wanted;
7680 case IFMATCH_A: /* body of (?...A) succeeded */
7682 sw = cBOOL(ST.wanted);
7684 else if (!ST.wanted)
7687 if (OP(ST.me) != SUSPEND) {
7688 /* restore old position except for (?>...) */
7689 locinput = st->locinput;
7691 scan = ST.me + ARG(ST.me);
7694 continue; /* execute B */
7698 case LONGJMP: /* alternative with many branches compiles to
7699 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
7700 next = scan + ARG(scan);
7705 case COMMIT: /* (*COMMIT) */
7706 reginfo->cutpoint = reginfo->strend;
7709 case PRUNE: /* (*PRUNE) */
7711 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7712 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
7714 NOT_REACHED; /* NOTREACHED */
7716 case COMMIT_next_fail:
7720 case OPFAIL: /* (*FAIL) */
7723 NOT_REACHED; /* NOTREACHED */
7725 #define ST st->u.mark
7726 case MARKPOINT: /* (*MARK:foo) */
7727 ST.prev_mark = mark_state;
7728 ST.mark_name = sv_commit = sv_yes_mark
7729 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7731 ST.mark_loc = locinput;
7732 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
7734 NOT_REACHED; /* NOTREACHED */
7736 case MARKPOINT_next:
7737 mark_state = ST.prev_mark;
7740 NOT_REACHED; /* NOTREACHED */
7742 case MARKPOINT_next_fail:
7743 if (popmark && sv_eq(ST.mark_name,popmark))
7745 if (ST.mark_loc > startpoint)
7746 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
7747 popmark = NULL; /* we found our mark */
7748 sv_commit = ST.mark_name;
7751 PerlIO_printf(Perl_debug_log,
7752 "%*s %ssetting cutpoint to mark:%"SVf"...%s\n",
7753 REPORT_CODE_OFF+depth*2, "",
7754 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
7757 mark_state = ST.prev_mark;
7758 sv_yes_mark = mark_state ?
7759 mark_state->u.mark.mark_name : NULL;
7762 NOT_REACHED; /* NOTREACHED */
7764 case SKIP: /* (*SKIP) */
7766 /* (*SKIP) : if we fail we cut here*/
7767 ST.mark_name = NULL;
7768 ST.mark_loc = locinput;
7769 PUSH_STATE_GOTO(SKIP_next,next, locinput);
7771 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
7772 otherwise do nothing. Meaning we need to scan
7774 regmatch_state *cur = mark_state;
7775 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7778 if ( sv_eq( cur->u.mark.mark_name,
7781 ST.mark_name = find;
7782 PUSH_STATE_GOTO( SKIP_next, next, locinput);
7784 cur = cur->u.mark.prev_mark;
7787 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
7790 case SKIP_next_fail:
7792 /* (*CUT:NAME) - Set up to search for the name as we
7793 collapse the stack*/
7794 popmark = ST.mark_name;
7796 /* (*CUT) - No name, we cut here.*/
7797 if (ST.mark_loc > startpoint)
7798 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
7799 /* but we set sv_commit to latest mark_name if there
7800 is one so they can test to see how things lead to this
7803 sv_commit=mark_state->u.mark.mark_name;
7808 NOT_REACHED; /* NOTREACHED */
7811 case LNBREAK: /* \R */
7812 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
7819 PerlIO_printf(Perl_error_log, "%"UVxf" %d\n",
7820 PTR2UV(scan), OP(scan));
7821 Perl_croak(aTHX_ "regexp memory corruption");
7823 /* this is a point to jump to in order to increment
7824 * locinput by one character */
7826 assert(!NEXTCHR_IS_EOS);
7828 locinput += PL_utf8skip[nextchr];
7829 /* locinput is allowed to go 1 char off the end, but not 2+ */
7830 if (locinput > reginfo->strend)
7839 /* switch break jumps here */
7840 scan = next; /* prepare to execute the next op and ... */
7841 continue; /* ... jump back to the top, reusing st */
7845 /* push a state that backtracks on success */
7846 st->u.yes.prev_yes_state = yes_state;
7850 /* push a new regex state, then continue at scan */
7852 regmatch_state *newst;
7855 regmatch_state *cur = st;
7856 regmatch_state *curyes = yes_state;
7858 regmatch_slab *slab = PL_regmatch_slab;
7859 for (;curd > -1;cur--,curd--) {
7860 if (cur < SLAB_FIRST(slab)) {
7862 cur = SLAB_LAST(slab);
7864 PerlIO_printf(Perl_error_log, "%*s#%-3d %-10s %s\n",
7865 REPORT_CODE_OFF + 2 + depth * 2,"",
7866 curd, PL_reg_name[cur->resume_state],
7867 (curyes == cur) ? "yes" : ""
7870 curyes = cur->u.yes.prev_yes_state;
7873 DEBUG_STATE_pp("push")
7876 st->locinput = locinput;
7878 if (newst > SLAB_LAST(PL_regmatch_slab))
7879 newst = S_push_slab(aTHX);
7880 PL_regmatch_state = newst;
7882 locinput = pushinput;
7890 * We get here only if there's trouble -- normally "case END" is
7891 * the terminating point.
7893 Perl_croak(aTHX_ "corrupted regexp pointers");
7896 NOT_REACHED; /* NOTREACHED */
7900 /* we have successfully completed a subexpression, but we must now
7901 * pop to the state marked by yes_state and continue from there */
7902 assert(st != yes_state);
7904 while (st != yes_state) {
7906 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7907 PL_regmatch_slab = PL_regmatch_slab->prev;
7908 st = SLAB_LAST(PL_regmatch_slab);
7912 DEBUG_STATE_pp("pop (no final)");
7914 DEBUG_STATE_pp("pop (yes)");
7920 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
7921 || yes_state > SLAB_LAST(PL_regmatch_slab))
7923 /* not in this slab, pop slab */
7924 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
7925 PL_regmatch_slab = PL_regmatch_slab->prev;
7926 st = SLAB_LAST(PL_regmatch_slab);
7928 depth -= (st - yes_state);
7931 yes_state = st->u.yes.prev_yes_state;
7932 PL_regmatch_state = st;
7935 locinput= st->locinput;
7936 state_num = st->resume_state + no_final;
7937 goto reenter_switch;
7940 DEBUG_EXECUTE_r(PerlIO_printf(Perl_debug_log, "%sMatch successful!%s\n",
7941 PL_colors[4], PL_colors[5]));
7943 if (reginfo->info_aux_eval) {
7944 /* each successfully executed (?{...}) block does the equivalent of
7945 * local $^R = do {...}
7946 * When popping the save stack, all these locals would be undone;
7947 * bypass this by setting the outermost saved $^R to the latest
7949 /* I dont know if this is needed or works properly now.
7950 * see code related to PL_replgv elsewhere in this file.
7953 if (oreplsv != GvSV(PL_replgv))
7954 sv_setsv(oreplsv, GvSV(PL_replgv));
7961 PerlIO_printf(Perl_debug_log,
7962 "%*s %sfailed...%s\n",
7963 REPORT_CODE_OFF+depth*2, "",
7964 PL_colors[4], PL_colors[5])
7976 /* there's a previous state to backtrack to */
7978 if (st < SLAB_FIRST(PL_regmatch_slab)) {
7979 PL_regmatch_slab = PL_regmatch_slab->prev;
7980 st = SLAB_LAST(PL_regmatch_slab);
7982 PL_regmatch_state = st;
7983 locinput= st->locinput;
7985 DEBUG_STATE_pp("pop");
7987 if (yes_state == st)
7988 yes_state = st->u.yes.prev_yes_state;
7990 state_num = st->resume_state + 1; /* failure = success + 1 */
7991 goto reenter_switch;
7996 if (rex->intflags & PREGf_VERBARG_SEEN) {
7997 SV *sv_err = get_sv("REGERROR", 1);
7998 SV *sv_mrk = get_sv("REGMARK", 1);
8000 sv_commit = &PL_sv_no;
8002 sv_yes_mark = &PL_sv_yes;
8005 sv_commit = &PL_sv_yes;
8006 sv_yes_mark = &PL_sv_no;
8010 sv_setsv(sv_err, sv_commit);
8011 sv_setsv(sv_mrk, sv_yes_mark);
8015 if (last_pushed_cv) {
8018 PERL_UNUSED_VAR(SP);
8021 assert(!result || locinput - reginfo->strbeg >= 0);
8022 return result ? locinput - reginfo->strbeg : -1;
8026 - regrepeat - repeatedly match something simple, report how many
8028 * What 'simple' means is a node which can be the operand of a quantifier like
8031 * startposp - pointer a pointer to the start position. This is updated
8032 * to point to the byte following the highest successful
8034 * p - the regnode to be repeatedly matched against.
8035 * reginfo - struct holding match state, such as strend
8036 * max - maximum number of things to match.
8037 * depth - (for debugging) backtracking depth.
8040 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
8041 regmatch_info *const reginfo, I32 max, int depth)
8043 char *scan; /* Pointer to current position in target string */
8045 char *loceol = reginfo->strend; /* local version */
8046 I32 hardcount = 0; /* How many matches so far */
8047 bool utf8_target = reginfo->is_utf8_target;
8048 unsigned int to_complement = 0; /* Invert the result? */
8050 _char_class_number classnum;
8052 PERL_UNUSED_ARG(depth);
8055 PERL_ARGS_ASSERT_REGREPEAT;
8058 if (max == REG_INFTY)
8060 else if (! utf8_target && loceol - scan > max)
8061 loceol = scan + max;
8063 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
8064 * to the maximum of how far we should go in it (leaving it set to the real
8065 * end, if the maximum permissible would take us beyond that). This allows
8066 * us to make the loop exit condition that we haven't gone past <loceol> to
8067 * also mean that we haven't exceeded the max permissible count, saving a
8068 * test each time through the loop. But it assumes that the OP matches a
8069 * single byte, which is true for most of the OPs below when applied to a
8070 * non-UTF-8 target. Those relatively few OPs that don't have this
8071 * characteristic will have to compensate.
8073 * There is no adjustment for UTF-8 targets, as the number of bytes per
8074 * character varies. OPs will have to test both that the count is less
8075 * than the max permissible (using <hardcount> to keep track), and that we
8076 * are still within the bounds of the string (using <loceol>. A few OPs
8077 * match a single byte no matter what the encoding. They can omit the max
8078 * test if, for the UTF-8 case, they do the adjustment that was skipped
8081 * Thus, the code above sets things up for the common case; and exceptional
8082 * cases need extra work; the common case is to make sure <scan> doesn't
8083 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
8084 * count doesn't exceed the maximum permissible */
8089 while (scan < loceol && hardcount < max && *scan != '\n') {
8090 scan += UTF8SKIP(scan);
8094 while (scan < loceol && *scan != '\n')
8100 while (scan < loceol && hardcount < max) {
8101 scan += UTF8SKIP(scan);
8108 case CANY: /* Move <scan> forward <max> bytes, unless goes off end */
8109 if (utf8_target && loceol - scan > max) {
8111 /* <loceol> hadn't been adjusted in the UTF-8 case */
8119 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8120 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
8121 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
8125 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8129 /* Can use a simple loop if the pattern char to match on is invariant
8130 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
8131 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
8132 * true iff it doesn't matter if the argument is in UTF-8 or not */
8133 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
8134 if (utf8_target && loceol - scan > max) {
8135 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
8136 * since here, to match at all, 1 char == 1 byte */
8137 loceol = scan + max;
8139 while (scan < loceol && UCHARAT(scan) == c) {
8143 else if (reginfo->is_utf8_pat) {
8145 STRLEN scan_char_len;
8147 /* When both target and pattern are UTF-8, we have to do
8149 while (hardcount < max
8151 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
8152 && memEQ(scan, STRING(p), scan_char_len))
8154 scan += scan_char_len;
8158 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
8160 /* Target isn't utf8; convert the character in the UTF-8
8161 * pattern to non-UTF8, and do a simple loop */
8162 c = TWO_BYTE_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
8163 while (scan < loceol && UCHARAT(scan) == c) {
8166 } /* else pattern char is above Latin1, can't possibly match the
8171 /* Here, the string must be utf8; pattern isn't, and <c> is
8172 * different in utf8 than not, so can't compare them directly.
8173 * Outside the loop, find the two utf8 bytes that represent c, and
8174 * then look for those in sequence in the utf8 string */
8175 U8 high = UTF8_TWO_BYTE_HI(c);
8176 U8 low = UTF8_TWO_BYTE_LO(c);
8178 while (hardcount < max
8179 && scan + 1 < loceol
8180 && UCHARAT(scan) == high
8181 && UCHARAT(scan + 1) == low)
8189 case EXACTFA_NO_TRIE: /* This node only generated for non-utf8 patterns */
8190 assert(! reginfo->is_utf8_pat);
8193 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
8197 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8198 utf8_flags = FOLDEQ_LOCALE;
8201 case EXACTF: /* This node only generated for non-utf8 patterns */
8202 assert(! reginfo->is_utf8_pat);
8207 if (! utf8_target) {
8210 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
8211 | FOLDEQ_S2_FOLDS_SANE;
8216 utf8_flags = reginfo->is_utf8_pat ? FOLDEQ_S2_ALREADY_FOLDED : 0;
8220 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
8222 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
8224 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
8227 if (c1 == CHRTEST_VOID) {
8228 /* Use full Unicode fold matching */
8229 char *tmpeol = reginfo->strend;
8230 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
8231 while (hardcount < max
8232 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
8233 STRING(p), NULL, pat_len,
8234 reginfo->is_utf8_pat, utf8_flags))
8237 tmpeol = reginfo->strend;
8241 else if (utf8_target) {
8243 while (scan < loceol
8245 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
8247 scan += UTF8SKIP(scan);
8252 while (scan < loceol
8254 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
8255 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
8257 scan += UTF8SKIP(scan);
8262 else if (c1 == c2) {
8263 while (scan < loceol && UCHARAT(scan) == c1) {
8268 while (scan < loceol &&
8269 (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
8278 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8282 while (hardcount < max
8284 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
8286 scan += UTF8SKIP(scan);
8290 while (scan < loceol && REGINCLASS(prog, p, (U8*)scan))
8295 /* The argument (FLAGS) to all the POSIX node types is the class number */
8302 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8303 if (! utf8_target) {
8304 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
8310 while (hardcount < max && scan < loceol
8311 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
8314 scan += UTF8SKIP(scan);
8327 if (utf8_target && loceol - scan > max) {
8329 /* We didn't adjust <loceol> at the beginning of this routine
8330 * because is UTF-8, but it is actually ok to do so, since here, to
8331 * match, 1 char == 1 byte. */
8332 loceol = scan + max;
8334 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
8347 if (! utf8_target) {
8348 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
8354 /* The complement of something that matches only ASCII matches all
8355 * non-ASCII, plus everything in ASCII that isn't in the class. */
8356 while (hardcount < max && scan < loceol
8357 && (! isASCII_utf8(scan)
8358 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
8360 scan += UTF8SKIP(scan);
8371 if (! utf8_target) {
8372 while (scan < loceol && to_complement
8373 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
8380 classnum = (_char_class_number) FLAGS(p);
8381 if (classnum < _FIRST_NON_SWASH_CC) {
8383 /* Here, a swash is needed for above-Latin1 code points.
8384 * Process as many Latin1 code points using the built-in rules.
8385 * Go to another loop to finish processing upon encountering
8386 * the first Latin1 code point. We could do that in this loop
8387 * as well, but the other way saves having to test if the swash
8388 * has been loaded every time through the loop: extra space to
8390 while (hardcount < max && scan < loceol) {
8391 if (UTF8_IS_INVARIANT(*scan)) {
8392 if (! (to_complement ^ cBOOL(_generic_isCC((U8) *scan,
8399 else if (UTF8_IS_DOWNGRADEABLE_START(*scan)) {
8400 if (! (to_complement
8401 ^ cBOOL(_generic_isCC(TWO_BYTE_UTF8_TO_NATIVE(*scan,
8410 goto found_above_latin1;
8417 /* For these character classes, the knowledge of how to handle
8418 * every code point is compiled in to Perl via a macro. This
8419 * code is written for making the loops as tight as possible.
8420 * It could be refactored to save space instead */
8422 case _CC_ENUM_SPACE:
8423 while (hardcount < max
8425 && (to_complement ^ cBOOL(isSPACE_utf8(scan))))
8427 scan += UTF8SKIP(scan);
8431 case _CC_ENUM_BLANK:
8432 while (hardcount < max
8434 && (to_complement ^ cBOOL(isBLANK_utf8(scan))))
8436 scan += UTF8SKIP(scan);
8440 case _CC_ENUM_XDIGIT:
8441 while (hardcount < max
8443 && (to_complement ^ cBOOL(isXDIGIT_utf8(scan))))
8445 scan += UTF8SKIP(scan);
8449 case _CC_ENUM_VERTSPACE:
8450 while (hardcount < max
8452 && (to_complement ^ cBOOL(isVERTWS_utf8(scan))))
8454 scan += UTF8SKIP(scan);
8458 case _CC_ENUM_CNTRL:
8459 while (hardcount < max
8461 && (to_complement ^ cBOOL(isCNTRL_utf8(scan))))
8463 scan += UTF8SKIP(scan);
8468 Perl_croak(aTHX_ "panic: regrepeat() node %d='%s' has an unexpected character class '%d'", OP(p), PL_reg_name[OP(p)], classnum);
8474 found_above_latin1: /* Continuation of POSIXU and NPOSIXU */
8476 /* Load the swash if not already present */
8477 if (! PL_utf8_swash_ptrs[classnum]) {
8478 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
8479 PL_utf8_swash_ptrs[classnum] = _core_swash_init(
8483 PL_XPosix_ptrs[classnum], &flags);
8486 while (hardcount < max && scan < loceol
8487 && to_complement ^ cBOOL(_generic_utf8(
8490 swash_fetch(PL_utf8_swash_ptrs[classnum],
8494 scan += UTF8SKIP(scan);
8501 while (hardcount < max && scan < loceol &&
8502 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
8507 /* LNBREAK can match one or two latin chars, which is ok, but we
8508 * have to use hardcount in this situation, and throw away the
8509 * adjustment to <loceol> done before the switch statement */
8510 loceol = reginfo->strend;
8511 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
8520 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
8534 /* These are all 0 width, so match right here or not at all. */
8538 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
8540 NOT_REACHED; /* NOTREACHED */
8547 c = scan - *startposp;
8551 GET_RE_DEBUG_FLAGS_DECL;
8553 SV * const prop = sv_newmortal();
8554 regprop(prog, prop, p, reginfo, NULL);
8555 PerlIO_printf(Perl_debug_log,
8556 "%*s %s can match %"IVdf" times out of %"IVdf"...\n",
8557 REPORT_CODE_OFF + depth*2, "", SvPVX_const(prop),(IV)c,(IV)max);
8565 #if !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION)
8567 - regclass_swash - prepare the utf8 swash. Wraps the shared core version to
8568 create a copy so that changes the caller makes won't change the shared one.
8569 If <altsvp> is non-null, will return NULL in it, for back-compat.
8572 Perl_regclass_swash(pTHX_ const regexp *prog, const regnode* node, bool doinit, SV** listsvp, SV **altsvp)
8574 PERL_ARGS_ASSERT_REGCLASS_SWASH;
8580 return newSVsv(_get_regclass_nonbitmap_data(prog, node, doinit, listsvp, NULL, NULL));
8583 #endif /* !defined(PERL_IN_XSUB_RE) || defined(PLUGGABLE_RE_EXTENSION) */
8586 - reginclass - determine if a character falls into a character class
8588 n is the ANYOF-type regnode
8589 p is the target string
8590 p_end points to one byte beyond the end of the target string
8591 utf8_target tells whether p is in UTF-8.
8593 Returns true if matched; false otherwise.
8595 Note that this can be a synthetic start class, a combination of various
8596 nodes, so things you think might be mutually exclusive, such as locale,
8597 aren't. It can match both locale and non-locale
8602 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
8605 const char flags = ANYOF_FLAGS(n);
8609 PERL_ARGS_ASSERT_REGINCLASS;
8611 /* If c is not already the code point, get it. Note that
8612 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
8613 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
8615 c = utf8n_to_uvchr(p, p_end - p, &c_len,
8616 (UTF8_ALLOW_DEFAULT & UTF8_ALLOW_ANYUV)
8617 | UTF8_ALLOW_FFFF | UTF8_CHECK_ONLY);
8618 /* see [perl #37836] for UTF8_ALLOW_ANYUV; [perl #38293] for
8619 * UTF8_ALLOW_FFFF */
8620 if (c_len == (STRLEN)-1)
8621 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
8622 if (c > 255 && OP(n) == ANYOFL && ! is_ANYOF_SYNTHETIC(n)) {
8623 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
8627 /* If this character is potentially in the bitmap, check it */
8628 if (c < NUM_ANYOF_CODE_POINTS) {
8629 if (ANYOF_BITMAP_TEST(n, c))
8631 else if ((flags & ANYOF_MATCHES_ALL_NON_UTF8_NON_ASCII)
8637 else if (flags & ANYOF_LOCALE_FLAGS) {
8638 if ((flags & ANYOF_LOC_FOLD)
8640 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
8644 else if (ANYOF_POSIXL_TEST_ANY_SET(n)
8648 /* The data structure is arranged so bits 0, 2, 4, ... are set
8649 * if the class includes the Posix character class given by
8650 * bit/2; and 1, 3, 5, ... are set if the class includes the
8651 * complemented Posix class given by int(bit/2). So we loop
8652 * through the bits, each time changing whether we complement
8653 * the result or not. Suppose for the sake of illustration
8654 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
8655 * is set, it means there is a match for this ANYOF node if the
8656 * character is in the class given by the expression (0 / 2 = 0
8657 * = \w). If it is in that class, isFOO_lc() will return 1,
8658 * and since 'to_complement' is 0, the result will stay TRUE,
8659 * and we exit the loop. Suppose instead that bit 0 is 0, but
8660 * bit 1 is 1. That means there is a match if the character
8661 * matches \W. We won't bother to call isFOO_lc() on bit 0,
8662 * but will on bit 1. On the second iteration 'to_complement'
8663 * will be 1, so the exclusive or will reverse things, so we
8664 * are testing for \W. On the third iteration, 'to_complement'
8665 * will be 0, and we would be testing for \s; the fourth
8666 * iteration would test for \S, etc.
8668 * Note that this code assumes that all the classes are closed
8669 * under folding. For example, if a character matches \w, then
8670 * its fold does too; and vice versa. This should be true for
8671 * any well-behaved locale for all the currently defined Posix
8672 * classes, except for :lower: and :upper:, which are handled
8673 * by the pseudo-class :cased: which matches if either of the
8674 * other two does. To get rid of this assumption, an outer
8675 * loop could be used below to iterate over both the source
8676 * character, and its fold (if different) */
8679 int to_complement = 0;
8681 while (count < ANYOF_MAX) {
8682 if (ANYOF_POSIXL_TEST(n, count)
8683 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
8696 /* If the bitmap didn't (or couldn't) match, and something outside the
8697 * bitmap could match, try that. */
8699 if (c >= NUM_ANYOF_CODE_POINTS
8700 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
8702 match = TRUE; /* Everything above the bitmap matches */
8704 else if ((flags & ANYOF_HAS_NONBITMAP_NON_UTF8_MATCHES)
8705 || (utf8_target && (flags & ANYOF_HAS_UTF8_NONBITMAP_MATCHES))
8706 || ((flags & ANYOF_LOC_FOLD)
8707 && IN_UTF8_CTYPE_LOCALE
8708 && ARG(n) != ANYOF_ONLY_HAS_BITMAP))
8710 SV* only_utf8_locale = NULL;
8711 SV * const sw = _get_regclass_nonbitmap_data(prog, n, TRUE, 0,
8712 &only_utf8_locale, NULL);
8718 } else { /* Convert to utf8 */
8719 utf8_p = utf8_buffer;
8720 append_utf8_from_native_byte(*p, &utf8_p);
8721 utf8_p = utf8_buffer;
8724 if (swash_fetch(sw, utf8_p, TRUE)) {
8728 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
8729 match = _invlist_contains_cp(only_utf8_locale, c);
8733 if (UNICODE_IS_SUPER(c)
8734 && (flags & ANYOF_WARN_SUPER)
8735 && ckWARN_d(WARN_NON_UNICODE))
8737 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
8738 "Matched non-Unicode code point 0x%04"UVXf" against Unicode property; may not be portable", c);
8742 #if ANYOF_INVERT != 1
8743 /* Depending on compiler optimization cBOOL takes time, so if don't have to
8745 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
8748 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
8749 return (flags & ANYOF_INVERT) ^ match;
8753 S_reghop3(U8 *s, SSize_t off, const U8* lim)
8755 /* return the position 'off' UTF-8 characters away from 's', forward if
8756 * 'off' >= 0, backwards if negative. But don't go outside of position
8757 * 'lim', which better be < s if off < 0 */
8759 PERL_ARGS_ASSERT_REGHOP3;
8762 while (off-- && s < lim) {
8763 /* XXX could check well-formedness here */
8768 while (off++ && s > lim) {
8770 if (UTF8_IS_CONTINUED(*s)) {
8771 while (s > lim && UTF8_IS_CONTINUATION(*s))
8774 /* XXX could check well-formedness here */
8781 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
8783 PERL_ARGS_ASSERT_REGHOP4;
8786 while (off-- && s < rlim) {
8787 /* XXX could check well-formedness here */
8792 while (off++ && s > llim) {
8794 if (UTF8_IS_CONTINUED(*s)) {
8795 while (s > llim && UTF8_IS_CONTINUATION(*s))
8798 /* XXX could check well-formedness here */
8804 /* like reghop3, but returns NULL on overrun, rather than returning last
8808 S_reghopmaybe3(U8* s, SSize_t off, const U8* lim)
8810 PERL_ARGS_ASSERT_REGHOPMAYBE3;
8813 while (off-- && s < lim) {
8814 /* XXX could check well-formedness here */
8821 while (off++ && s > lim) {
8823 if (UTF8_IS_CONTINUED(*s)) {
8824 while (s > lim && UTF8_IS_CONTINUATION(*s))
8827 /* XXX could check well-formedness here */
8836 /* when executing a regex that may have (?{}), extra stuff needs setting
8837 up that will be visible to the called code, even before the current
8838 match has finished. In particular:
8840 * $_ is localised to the SV currently being matched;
8841 * pos($_) is created if necessary, ready to be updated on each call-out
8843 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
8844 isn't set until the current pattern is successfully finished), so that
8845 $1 etc of the match-so-far can be seen;
8846 * save the old values of subbeg etc of the current regex, and set then
8847 to the current string (again, this is normally only done at the end
8852 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
8855 regexp *const rex = ReANY(reginfo->prog);
8856 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
8858 eval_state->rex = rex;
8861 /* Make $_ available to executed code. */
8862 if (reginfo->sv != DEFSV) {
8864 DEFSV_set(reginfo->sv);
8867 if (!(mg = mg_find_mglob(reginfo->sv))) {
8868 /* prepare for quick setting of pos */
8869 mg = sv_magicext_mglob(reginfo->sv);
8872 eval_state->pos_magic = mg;
8873 eval_state->pos = mg->mg_len;
8874 eval_state->pos_flags = mg->mg_flags;
8877 eval_state->pos_magic = NULL;
8879 if (!PL_reg_curpm) {
8880 /* PL_reg_curpm is a fake PMOP that we can attach the current
8881 * regex to and point PL_curpm at, so that $1 et al are visible
8882 * within a /(?{})/. It's just allocated once per interpreter the
8883 * first time its needed */
8884 Newxz(PL_reg_curpm, 1, PMOP);
8887 SV* const repointer = &PL_sv_undef;
8888 /* this regexp is also owned by the new PL_reg_curpm, which
8889 will try to free it. */
8890 av_push(PL_regex_padav, repointer);
8891 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
8892 PL_regex_pad = AvARRAY(PL_regex_padav);
8896 SET_reg_curpm(reginfo->prog);
8897 eval_state->curpm = PL_curpm;
8898 PL_curpm = PL_reg_curpm;
8899 if (RXp_MATCH_COPIED(rex)) {
8900 /* Here is a serious problem: we cannot rewrite subbeg,
8901 since it may be needed if this match fails. Thus
8902 $` inside (?{}) could fail... */
8903 eval_state->subbeg = rex->subbeg;
8904 eval_state->sublen = rex->sublen;
8905 eval_state->suboffset = rex->suboffset;
8906 eval_state->subcoffset = rex->subcoffset;
8908 eval_state->saved_copy = rex->saved_copy;
8910 RXp_MATCH_COPIED_off(rex);
8913 eval_state->subbeg = NULL;
8914 rex->subbeg = (char *)reginfo->strbeg;
8916 rex->subcoffset = 0;
8917 rex->sublen = reginfo->strend - reginfo->strbeg;
8921 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
8924 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
8926 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
8927 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
8930 Safefree(aux->poscache);
8934 /* undo the effects of S_setup_eval_state() */
8936 if (eval_state->subbeg) {
8937 regexp * const rex = eval_state->rex;
8938 rex->subbeg = eval_state->subbeg;
8939 rex->sublen = eval_state->sublen;
8940 rex->suboffset = eval_state->suboffset;
8941 rex->subcoffset = eval_state->subcoffset;
8943 rex->saved_copy = eval_state->saved_copy;
8945 RXp_MATCH_COPIED_on(rex);
8947 if (eval_state->pos_magic)
8949 eval_state->pos_magic->mg_len = eval_state->pos;
8950 eval_state->pos_magic->mg_flags =
8951 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
8952 | (eval_state->pos_flags & MGf_BYTES);
8955 PL_curpm = eval_state->curpm;
8958 PL_regmatch_state = aux->old_regmatch_state;
8959 PL_regmatch_slab = aux->old_regmatch_slab;
8961 /* free all slabs above current one - this must be the last action
8962 * of this function, as aux and eval_state are allocated within
8963 * slabs and may be freed here */
8965 s = PL_regmatch_slab->next;
8967 PL_regmatch_slab->next = NULL;
8969 regmatch_slab * const osl = s;
8978 S_to_utf8_substr(pTHX_ regexp *prog)
8980 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
8981 * on the converted value */
8985 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
8988 if (prog->substrs->data[i].substr
8989 && !prog->substrs->data[i].utf8_substr) {
8990 SV* const sv = newSVsv(prog->substrs->data[i].substr);
8991 prog->substrs->data[i].utf8_substr = sv;
8992 sv_utf8_upgrade(sv);
8993 if (SvVALID(prog->substrs->data[i].substr)) {
8994 if (SvTAIL(prog->substrs->data[i].substr)) {
8995 /* Trim the trailing \n that fbm_compile added last
8997 SvCUR_set(sv, SvCUR(sv) - 1);
8998 /* Whilst this makes the SV technically "invalid" (as its
8999 buffer is no longer followed by "\0") when fbm_compile()
9000 adds the "\n" back, a "\0" is restored. */
9001 fbm_compile(sv, FBMcf_TAIL);
9005 if (prog->substrs->data[i].substr == prog->check_substr)
9006 prog->check_utf8 = sv;
9012 S_to_byte_substr(pTHX_ regexp *prog)
9014 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
9015 * on the converted value; returns FALSE if can't be converted. */
9019 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
9022 if (prog->substrs->data[i].utf8_substr
9023 && !prog->substrs->data[i].substr) {
9024 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
9025 if (! sv_utf8_downgrade(sv, TRUE)) {
9028 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
9029 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
9030 /* Trim the trailing \n that fbm_compile added last
9032 SvCUR_set(sv, SvCUR(sv) - 1);
9033 fbm_compile(sv, FBMcf_TAIL);
9037 prog->substrs->data[i].substr = sv;
9038 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
9039 prog->check_substr = sv;
9047 * ex: set ts=8 sts=4 sw=4 et: