5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "invlist_inline.h"
84 #include "unicode_constants.h"
86 #define B_ON_NON_UTF8_LOCALE_IS_WRONG \
87 "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale"
89 static const char utf8_locale_required[] =
90 "Use of (?[ ]) for non-UTF-8 locale is wrong. Assuming a UTF-8 locale";
93 /* At least one required character in the target string is expressible only in
95 static const char non_utf8_target_but_utf8_required[]
96 = "Can't match, because target string needs to be in UTF-8\n";
99 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
100 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%s", non_utf8_target_but_utf8_required));\
104 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
107 #define STATIC static
114 #define CHR_SVLEN(sv) (utf8_target ? sv_len_utf8(sv) : SvCUR(sv))
116 #define HOPc(pos,off) \
117 (char *)(reginfo->is_utf8_target \
118 ? reghop3((U8*)pos, off, \
119 (U8*)(off >= 0 ? reginfo->strend : reginfo->strbeg)) \
122 /* like HOPMAYBE3 but backwards. lim must be +ve. Returns NULL on overshoot */
123 #define HOPBACK3(pos, off, lim) \
124 (reginfo->is_utf8_target \
125 ? reghopmaybe3((U8*)pos, (SSize_t)0-off, (U8*)(lim)) \
126 : (pos - off >= lim) \
130 #define HOPBACKc(pos, off) ((char*)HOPBACK3(pos, off, reginfo->strbeg))
132 #define HOP3(pos,off,lim) (reginfo->is_utf8_target ? reghop3((U8*)(pos), off, (U8*)(lim)) : (U8*)(pos + off))
133 #define HOP3c(pos,off,lim) ((char*)HOP3(pos,off,lim))
135 /* lim must be +ve. Returns NULL on overshoot */
136 #define HOPMAYBE3(pos,off,lim) \
137 (reginfo->is_utf8_target \
138 ? reghopmaybe3((U8*)pos, off, (U8*)(lim)) \
139 : ((U8*)pos + off <= lim) \
143 /* like HOP3, but limits the result to <= lim even for the non-utf8 case.
144 * off must be >=0; args should be vars rather than expressions */
145 #define HOP3lim(pos,off,lim) (reginfo->is_utf8_target \
146 ? reghop3((U8*)(pos), off, (U8*)(lim)) \
147 : (U8*)((pos + off) > lim ? lim : (pos + off)))
148 #define HOP3clim(pos,off,lim) ((char*)HOP3lim(pos,off,lim))
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 PLACEHOLDER /* Something for the preprocessor to grab onto */
156 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
158 /* for use after a quantifier and before an EXACT-like node -- japhy */
159 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
161 * NOTE that *nothing* that affects backtracking should be in here, specifically
162 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
163 * node that is in between two EXACT like nodes when ascertaining what the required
164 * "follow" character is. This should probably be moved to regex compile time
165 * although it may be done at run time beause of the REF possibility - more
166 * investigation required. -- demerphq
168 #define JUMPABLE(rn) ( \
170 (OP(rn) == CLOSE && \
171 !EVAL_CLOSE_PAREN_IS(cur_eval,ARG(rn)) ) || \
173 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
174 OP(rn) == PLUS || OP(rn) == MINMOD || \
176 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
178 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
180 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
183 Search for mandatory following text node; for lookahead, the text must
184 follow but for lookbehind (rn->flags != 0) we skip to the next step.
186 #define FIND_NEXT_IMPT(rn) STMT_START { \
187 while (JUMPABLE(rn)) { \
188 const OPCODE type = OP(rn); \
189 if (type == SUSPEND || PL_regkind[type] == CURLY) \
190 rn = NEXTOPER(NEXTOPER(rn)); \
191 else if (type == PLUS) \
193 else if (type == IFMATCH) \
194 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
195 else rn += NEXT_OFF(rn); \
199 #define SLAB_FIRST(s) (&(s)->states[0])
200 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
202 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
203 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
204 static regmatch_state * S_push_slab(pTHX);
206 #define REGCP_PAREN_ELEMS 3
207 #define REGCP_OTHER_ELEMS 3
208 #define REGCP_FRAME_ELEMS 1
209 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
210 * are needed for the regexp context stack bookkeeping. */
213 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen _pDEPTH)
215 const int retval = PL_savestack_ix;
216 const int paren_elems_to_push =
217 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
218 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
219 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
221 GET_RE_DEBUG_FLAGS_DECL;
223 PERL_ARGS_ASSERT_REGCPPUSH;
225 if (paren_elems_to_push < 0)
226 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
227 (int)paren_elems_to_push, (int)maxopenparen,
228 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
230 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
231 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %" UVuf
232 " out of range (%lu-%ld)",
234 (unsigned long)maxopenparen,
237 SSGROW(total_elems + REGCP_FRAME_ELEMS);
240 if ((int)maxopenparen > (int)parenfloor)
241 Perl_re_exec_indentf( aTHX_
242 "rex=0x%" UVxf " offs=0x%" UVxf ": saving capture indices:\n",
248 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
249 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
250 SSPUSHIV(rex->offs[p].end);
251 SSPUSHIV(rex->offs[p].start);
252 SSPUSHINT(rex->offs[p].start_tmp);
253 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
254 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "\n",
257 (IV)rex->offs[p].start,
258 (IV)rex->offs[p].start_tmp,
262 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
263 SSPUSHINT(maxopenparen);
264 SSPUSHINT(rex->lastparen);
265 SSPUSHINT(rex->lastcloseparen);
266 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
271 /* These are needed since we do not localize EVAL nodes: */
272 #define REGCP_SET(cp) \
274 Perl_re_exec_indentf( aTHX_ \
275 "Setting an EVAL scope, savestack=%" IVdf ",\n", \
276 depth, (IV)PL_savestack_ix \
281 #define REGCP_UNWIND(cp) \
283 if (cp != PL_savestack_ix) \
284 Perl_re_exec_indentf( aTHX_ \
285 "Clearing an EVAL scope, savestack=%" \
286 IVdf "..%" IVdf "\n", \
287 depth, (IV)(cp), (IV)PL_savestack_ix \
292 /* set the start and end positions of capture ix */
293 #define CLOSE_CAPTURE(ix, s, e) \
294 rex->offs[ix].start = s; \
295 rex->offs[ix].end = e; \
296 if (ix > rex->lastparen) \
297 rex->lastparen = ix; \
298 rex->lastcloseparen = ix; \
299 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
300 "CLOSE: rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf "..%" IVdf " max: %" UVuf "\n", \
305 (IV)rex->offs[ix].start, \
306 (IV)rex->offs[ix].end, \
310 #define UNWIND_PAREN(lp, lcp) \
311 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
312 "UNWIND_PAREN: rex=0x%" UVxf " offs=0x%" UVxf ": invalidate (%" UVuf "..%" UVuf "] set lcp: %" UVuf "\n", \
317 (UV)(rex->lastparen), \
320 for (n = rex->lastparen; n > lp; n--) \
321 rex->offs[n].end = -1; \
322 rex->lastparen = n; \
323 rex->lastcloseparen = lcp;
327 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p _pDEPTH)
331 GET_RE_DEBUG_FLAGS_DECL;
333 PERL_ARGS_ASSERT_REGCPPOP;
335 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
337 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
338 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
339 rex->lastcloseparen = SSPOPINT;
340 rex->lastparen = SSPOPINT;
341 *maxopenparen_p = SSPOPINT;
343 i -= REGCP_OTHER_ELEMS;
344 /* Now restore the parentheses context. */
346 if (i || rex->lastparen + 1 <= rex->nparens)
347 Perl_re_exec_indentf( aTHX_
348 "rex=0x%" UVxf " offs=0x%" UVxf ": restoring capture indices to:\n",
354 paren = *maxopenparen_p;
355 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
357 rex->offs[paren].start_tmp = SSPOPINT;
358 rex->offs[paren].start = SSPOPIV;
360 if (paren <= rex->lastparen)
361 rex->offs[paren].end = tmps;
362 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
363 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "%s\n",
366 (IV)rex->offs[paren].start,
367 (IV)rex->offs[paren].start_tmp,
368 (IV)rex->offs[paren].end,
369 (paren > rex->lastparen ? "(skipped)" : ""));
374 /* It would seem that the similar code in regtry()
375 * already takes care of this, and in fact it is in
376 * a better location to since this code can #if 0-ed out
377 * but the code in regtry() is needed or otherwise tests
378 * requiring null fields (pat.t#187 and split.t#{13,14}
379 * (as of patchlevel 7877) will fail. Then again,
380 * this code seems to be necessary or otherwise
381 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
382 * --jhi updated by dapm */
383 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
384 if (i > *maxopenparen_p)
385 rex->offs[i].start = -1;
386 rex->offs[i].end = -1;
387 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
388 " \\%" UVuf ": %s ..-1 undeffing\n",
391 (i > *maxopenparen_p) ? "-1" : " "
397 /* restore the parens and associated vars at savestack position ix,
398 * but without popping the stack */
401 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p _pDEPTH)
403 I32 tmpix = PL_savestack_ix;
404 PERL_ARGS_ASSERT_REGCP_RESTORE;
406 PL_savestack_ix = ix;
407 regcppop(rex, maxopenparen_p);
408 PL_savestack_ix = tmpix;
411 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
413 #ifndef PERL_IN_XSUB_RE
416 Perl_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
418 /* Returns a boolean as to whether or not 'character' is a member of the
419 * Posix character class given by 'classnum' that should be equivalent to a
420 * value in the typedef '_char_class_number'.
422 * Ideally this could be replaced by a just an array of function pointers
423 * to the C library functions that implement the macros this calls.
424 * However, to compile, the precise function signatures are required, and
425 * these may vary from platform to to platform. To avoid having to figure
426 * out what those all are on each platform, I (khw) am using this method,
427 * which adds an extra layer of function call overhead (unless the C
428 * optimizer strips it away). But we don't particularly care about
429 * performance with locales anyway. */
431 switch ((_char_class_number) classnum) {
432 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
433 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
434 case _CC_ENUM_ASCII: return isASCII_LC(character);
435 case _CC_ENUM_BLANK: return isBLANK_LC(character);
436 case _CC_ENUM_CASED: return isLOWER_LC(character)
437 || isUPPER_LC(character);
438 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
439 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
440 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
441 case _CC_ENUM_LOWER: return isLOWER_LC(character);
442 case _CC_ENUM_PRINT: return isPRINT_LC(character);
443 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
444 case _CC_ENUM_SPACE: return isSPACE_LC(character);
445 case _CC_ENUM_UPPER: return isUPPER_LC(character);
446 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
447 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
448 default: /* VERTSPACE should never occur in locales */
449 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
452 NOT_REACHED; /* NOTREACHED */
458 PERL_STATIC_INLINE I32
459 S_foldEQ_latin1_s2_folded(const char *s1, const char *s2, I32 len)
461 /* Compare non-UTF-8 using Unicode (Latin1) semantics. s2 must already be
462 * folded. Works on all folds representable without UTF-8, except for
463 * LATIN_SMALL_LETTER_SHARP_S, and does not check for this. Nor does it
464 * check that the strings each have at least 'len' characters.
466 * There is almost an identical API function where s2 need not be folded:
467 * Perl_foldEQ_latin1() */
469 const U8 *a = (const U8 *)s1;
470 const U8 *b = (const U8 *)s2;
472 PERL_ARGS_ASSERT_FOLDEQ_LATIN1_S2_FOLDED;
477 assert(! isUPPER_L1(*b));
478 if (toLOWER_L1(*a) != *b) {
487 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character, const U8* e)
489 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
490 * 'character' is a member of the Posix character class given by 'classnum'
491 * that should be equivalent to a value in the typedef
492 * '_char_class_number'.
494 * This just calls isFOO_lc on the code point for the character if it is in
495 * the range 0-255. Outside that range, all characters use Unicode
496 * rules, ignoring any locale. So use the Unicode function if this class
497 * requires an inversion list, and use the Unicode macro otherwise. */
501 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
503 if (UTF8_IS_INVARIANT(*character)) {
504 return isFOO_lc(classnum, *character);
506 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
507 return isFOO_lc(classnum,
508 EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1)));
511 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, e);
513 switch ((_char_class_number) classnum) {
514 case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character);
515 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
516 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
517 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
519 return _invlist_contains_cp(PL_XPosix_ptrs[classnum],
520 utf8_to_uvchr_buf(character, e, NULL));
523 return FALSE; /* Things like CNTRL are always below 256 */
527 S_find_span_end(U8 * s, const U8 * send, const U8 span_byte)
529 /* Returns the position of the first byte in the sequence between 's' and
530 * 'send-1' inclusive that isn't 'span_byte'; returns 'send' if none found.
533 PERL_ARGS_ASSERT_FIND_SPAN_END;
537 if ((STRLEN) (send - s) >= PERL_WORDSIZE
538 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
539 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
541 PERL_UINTMAX_T span_word;
543 /* Process per-byte until reach word boundary. XXX This loop could be
544 * eliminated if we knew that this platform had fast unaligned reads */
545 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
546 if (*s != span_byte) {
552 /* Create a word filled with the bytes we are spanning */
553 span_word = PERL_COUNT_MULTIPLIER * span_byte;
555 /* Process per-word as long as we have at least a full word left */
558 /* Keep going if the whole word is composed of 'span_byte's */
559 if ((* (PERL_UINTMAX_T *) s) == span_word) {
564 /* Here, at least one byte in the word isn't 'span_byte'. */
572 /* This xor leaves 1 bits only in those non-matching bytes */
573 span_word ^= * (PERL_UINTMAX_T *) s;
575 /* Make sure the upper bit of each non-matching byte is set. This
576 * makes each such byte look like an ASCII platform variant byte */
577 span_word |= span_word << 1;
578 span_word |= span_word << 2;
579 span_word |= span_word << 4;
581 /* That reduces the problem to what this function solves */
582 return s + _variant_byte_number(span_word);
586 } while (s + PERL_WORDSIZE <= send);
589 /* Process the straggler bytes beyond the final word boundary */
591 if (*s != span_byte) {
601 S_find_next_masked(U8 * s, const U8 * send, const U8 byte, const U8 mask)
603 /* Returns the position of the first byte in the sequence between 's'
604 * and 'send-1' inclusive that when ANDed with 'mask' yields 'byte';
605 * returns 'send' if none found. It uses word-level operations instead of
606 * byte to speed up the process */
608 PERL_ARGS_ASSERT_FIND_NEXT_MASKED;
611 assert((byte & mask) == byte);
615 if ((STRLEN) (send - s) >= PERL_WORDSIZE
616 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
617 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
619 PERL_UINTMAX_T word, mask_word;
621 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
622 if (((*s) & mask) == byte) {
628 word = PERL_COUNT_MULTIPLIER * byte;
629 mask_word = PERL_COUNT_MULTIPLIER * mask;
632 PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;
634 /* If 'masked' contains bytes with the bit pattern of 'byte' within
635 * it, xoring with 'word' will leave each of the 8 bits in such
636 * bytes be 0, and no byte containing any other bit pattern will be
640 /* This causes the most significant bit to be set to 1 for any
641 * bytes in the word that aren't completely 0 */
642 masked |= masked << 1;
643 masked |= masked << 2;
644 masked |= masked << 4;
646 /* The msbits are the same as what marks a byte as variant, so we
647 * can use this mask. If all msbits are 1, the word doesn't
649 if ((masked & PERL_VARIANTS_WORD_MASK) == PERL_VARIANTS_WORD_MASK) {
654 /* Here, the msbit of bytes in the word that aren't 'byte' are 1,
655 * and any that are, are 0. Complement and re-AND to swap that */
657 masked &= PERL_VARIANTS_WORD_MASK;
659 /* This reduces the problem to that solved by this function */
660 s += _variant_byte_number(masked);
663 } while (s + PERL_WORDSIZE <= send);
669 if (((*s) & mask) == byte) {
679 S_find_span_end_mask(U8 * s, const U8 * send, const U8 span_byte, const U8 mask)
681 /* Returns the position of the first byte in the sequence between 's' and
682 * 'send-1' inclusive that when ANDed with 'mask' isn't 'span_byte'.
683 * 'span_byte' should have been ANDed with 'mask' in the call of this
684 * function. Returns 'send' if none found. Works like find_span_end(),
685 * except for the AND */
687 PERL_ARGS_ASSERT_FIND_SPAN_END_MASK;
690 assert((span_byte & mask) == span_byte);
692 if ((STRLEN) (send - s) >= PERL_WORDSIZE
693 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
694 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
696 PERL_UINTMAX_T span_word, mask_word;
698 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
699 if (((*s) & mask) != span_byte) {
705 span_word = PERL_COUNT_MULTIPLIER * span_byte;
706 mask_word = PERL_COUNT_MULTIPLIER * mask;
709 PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;
711 if (masked == span_word) {
723 masked |= masked << 1;
724 masked |= masked << 2;
725 masked |= masked << 4;
726 return s + _variant_byte_number(masked);
730 } while (s + PERL_WORDSIZE <= send);
734 if (((*s) & mask) != span_byte) {
744 * pregexec and friends
747 #ifndef PERL_IN_XSUB_RE
749 - pregexec - match a regexp against a string
752 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
753 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
754 /* stringarg: the point in the string at which to begin matching */
755 /* strend: pointer to null at end of string */
756 /* strbeg: real beginning of string */
757 /* minend: end of match must be >= minend bytes after stringarg. */
758 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
759 * itself is accessed via the pointers above */
760 /* nosave: For optimizations. */
762 PERL_ARGS_ASSERT_PREGEXEC;
765 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
766 nosave ? 0 : REXEC_COPY_STR);
772 /* re_intuit_start():
774 * Based on some optimiser hints, try to find the earliest position in the
775 * string where the regex could match.
777 * rx: the regex to match against
778 * sv: the SV being matched: only used for utf8 flag; the string
779 * itself is accessed via the pointers below. Note that on
780 * something like an overloaded SV, SvPOK(sv) may be false
781 * and the string pointers may point to something unrelated to
783 * strbeg: real beginning of string
784 * strpos: the point in the string at which to begin matching
785 * strend: pointer to the byte following the last char of the string
786 * flags currently unused; set to 0
787 * data: currently unused; set to NULL
789 * The basic idea of re_intuit_start() is to use some known information
790 * about the pattern, namely:
792 * a) the longest known anchored substring (i.e. one that's at a
793 * constant offset from the beginning of the pattern; but not
794 * necessarily at a fixed offset from the beginning of the
796 * b) the longest floating substring (i.e. one that's not at a constant
797 * offset from the beginning of the pattern);
798 * c) Whether the pattern is anchored to the string; either
799 * an absolute anchor: /^../, or anchored to \n: /^.../m,
800 * or anchored to pos(): /\G/;
801 * d) A start class: a real or synthetic character class which
802 * represents which characters are legal at the start of the pattern;
804 * to either quickly reject the match, or to find the earliest position
805 * within the string at which the pattern might match, thus avoiding
806 * running the full NFA engine at those earlier locations, only to
807 * eventually fail and retry further along.
809 * Returns NULL if the pattern can't match, or returns the address within
810 * the string which is the earliest place the match could occur.
812 * The longest of the anchored and floating substrings is called 'check'
813 * and is checked first. The other is called 'other' and is checked
814 * second. The 'other' substring may not be present. For example,
816 * /(abc|xyz)ABC\d{0,3}DEFG/
820 * check substr (float) = "DEFG", offset 6..9 chars
821 * other substr (anchored) = "ABC", offset 3..3 chars
824 * Be aware that during the course of this function, sometimes 'anchored'
825 * refers to a substring being anchored relative to the start of the
826 * pattern, and sometimes to the pattern itself being anchored relative to
827 * the string. For example:
829 * /\dabc/: "abc" is anchored to the pattern;
830 * /^\dabc/: "abc" is anchored to the pattern and the string;
831 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
832 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
833 * but the pattern is anchored to the string.
837 Perl_re_intuit_start(pTHX_
840 const char * const strbeg,
844 re_scream_pos_data *data)
846 struct regexp *const prog = ReANY(rx);
847 SSize_t start_shift = prog->check_offset_min;
848 /* Should be nonnegative! */
849 SSize_t end_shift = 0;
850 /* current lowest pos in string where the regex can start matching */
851 char *rx_origin = strpos;
853 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
854 U8 other_ix = 1 - prog->substrs->check_ix;
856 char *other_last = strpos;/* latest pos 'other' substr already checked to */
857 char *check_at = NULL; /* check substr found at this pos */
858 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
859 RXi_GET_DECL(prog,progi);
860 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
861 regmatch_info *const reginfo = ®info_buf;
862 GET_RE_DEBUG_FLAGS_DECL;
864 PERL_ARGS_ASSERT_RE_INTUIT_START;
865 PERL_UNUSED_ARG(flags);
866 PERL_UNUSED_ARG(data);
868 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
869 "Intuit: trying to determine minimum start position...\n"));
871 /* for now, assume that all substr offsets are positive. If at some point
872 * in the future someone wants to do clever things with lookbehind and
873 * -ve offsets, they'll need to fix up any code in this function
874 * which uses these offsets. See the thread beginning
875 * <20140113145929.GF27210@iabyn.com>
877 assert(prog->substrs->data[0].min_offset >= 0);
878 assert(prog->substrs->data[0].max_offset >= 0);
879 assert(prog->substrs->data[1].min_offset >= 0);
880 assert(prog->substrs->data[1].max_offset >= 0);
881 assert(prog->substrs->data[2].min_offset >= 0);
882 assert(prog->substrs->data[2].max_offset >= 0);
884 /* for now, assume that if both present, that the floating substring
885 * doesn't start before the anchored substring.
886 * If you break this assumption (e.g. doing better optimisations
887 * with lookahead/behind), then you'll need to audit the code in this
888 * function carefully first
891 ! ( (prog->anchored_utf8 || prog->anchored_substr)
892 && (prog->float_utf8 || prog->float_substr))
893 || (prog->float_min_offset >= prog->anchored_offset));
895 /* byte rather than char calculation for efficiency. It fails
896 * to quickly reject some cases that can't match, but will reject
897 * them later after doing full char arithmetic */
898 if (prog->minlen > strend - strpos) {
899 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
900 " String too short...\n"));
904 RXp_MATCH_UTF8_set(prog, utf8_target);
905 reginfo->is_utf8_target = cBOOL(utf8_target);
906 reginfo->info_aux = NULL;
907 reginfo->strbeg = strbeg;
908 reginfo->strend = strend;
909 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
911 /* not actually used within intuit, but zero for safety anyway */
912 reginfo->poscache_maxiter = 0;
915 if ((!prog->anchored_utf8 && prog->anchored_substr)
916 || (!prog->float_utf8 && prog->float_substr))
917 to_utf8_substr(prog);
918 check = prog->check_utf8;
920 if (!prog->check_substr && prog->check_utf8) {
921 if (! to_byte_substr(prog)) {
922 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
925 check = prog->check_substr;
928 /* dump the various substring data */
929 DEBUG_OPTIMISE_MORE_r({
931 for (i=0; i<=2; i++) {
932 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
933 : prog->substrs->data[i].substr);
937 Perl_re_printf( aTHX_
938 " substrs[%d]: min=%" IVdf " max=%" IVdf " end shift=%" IVdf
939 " useful=%" IVdf " utf8=%d [%s]\n",
941 (IV)prog->substrs->data[i].min_offset,
942 (IV)prog->substrs->data[i].max_offset,
943 (IV)prog->substrs->data[i].end_shift,
950 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
952 /* ml_anch: check after \n?
954 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
955 * with /.*.../, these flags will have been added by the
957 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
958 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
960 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
961 && !(prog->intflags & PREGf_IMPLICIT);
963 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
964 /* we are only allowed to match at BOS or \G */
966 /* trivially reject if there's a BOS anchor and we're not at BOS.
968 * Note that we don't try to do a similar quick reject for
969 * \G, since generally the caller will have calculated strpos
970 * based on pos() and gofs, so the string is already correctly
971 * anchored by definition; and handling the exceptions would
972 * be too fiddly (e.g. REXEC_IGNOREPOS).
974 if ( strpos != strbeg
975 && (prog->intflags & PREGf_ANCH_SBOL))
977 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
978 " Not at start...\n"));
982 /* in the presence of an anchor, the anchored (relative to the
983 * start of the regex) substr must also be anchored relative
984 * to strpos. So quickly reject if substr isn't found there.
985 * This works for \G too, because the caller will already have
986 * subtracted gofs from pos, and gofs is the offset from the
987 * \G to the start of the regex. For example, in /.abc\Gdef/,
988 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
989 * caller will have set strpos=pos()-4; we look for the substr
990 * at position pos()-4+1, which lines up with the "a" */
992 if (prog->check_offset_min == prog->check_offset_max) {
993 /* Substring at constant offset from beg-of-str... */
994 SSize_t slen = SvCUR(check);
995 char *s = HOP3c(strpos, prog->check_offset_min, strend);
997 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
998 " Looking for check substr at fixed offset %" IVdf "...\n",
999 (IV)prog->check_offset_min));
1001 if (SvTAIL(check)) {
1002 /* In this case, the regex is anchored at the end too.
1003 * Unless it's a multiline match, the lengths must match
1004 * exactly, give or take a \n. NB: slen >= 1 since
1005 * the last char of check is \n */
1007 && ( strend - s > slen
1008 || strend - s < slen - 1
1009 || (strend - s == slen && strend[-1] != '\n')))
1011 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1012 " String too long...\n"));
1015 /* Now should match s[0..slen-2] */
1018 if (slen && (strend - s < slen
1019 || *SvPVX_const(check) != *s
1020 || (slen > 1 && (memNE(SvPVX_const(check), s, slen)))))
1022 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1023 " String not equal...\n"));
1028 goto success_at_start;
1033 end_shift = prog->check_end_shift;
1035 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
1037 Perl_croak(aTHX_ "panic: end_shift: %" IVdf " pattern:\n%s\n ",
1038 (IV)end_shift, RX_PRECOMP(rx));
1043 /* This is the (re)entry point of the main loop in this function.
1044 * The goal of this loop is to:
1045 * 1) find the "check" substring in the region rx_origin..strend
1046 * (adjusted by start_shift / end_shift). If not found, reject
1048 * 2) If it exists, look for the "other" substr too if defined; for
1049 * example, if the check substr maps to the anchored substr, then
1050 * check the floating substr, and vice-versa. If not found, go
1051 * back to (1) with rx_origin suitably incremented.
1052 * 3) If we find an rx_origin position that doesn't contradict
1053 * either of the substrings, then check the possible additional
1054 * constraints on rx_origin of /^.../m or a known start class.
1055 * If these fail, then depending on which constraints fail, jump
1056 * back to here, or to various other re-entry points further along
1057 * that skip some of the first steps.
1058 * 4) If we pass all those tests, update the BmUSEFUL() count on the
1059 * substring. If the start position was determined to be at the
1060 * beginning of the string - so, not rejected, but not optimised,
1061 * since we have to run regmatch from position 0 - decrement the
1062 * BmUSEFUL() count. Otherwise increment it.
1066 /* first, look for the 'check' substring */
1072 DEBUG_OPTIMISE_MORE_r({
1073 Perl_re_printf( aTHX_
1074 " At restart: rx_origin=%" IVdf " Check offset min: %" IVdf
1075 " Start shift: %" IVdf " End shift %" IVdf
1076 " Real end Shift: %" IVdf "\n",
1077 (IV)(rx_origin - strbeg),
1078 (IV)prog->check_offset_min,
1081 (IV)prog->check_end_shift);
1084 end_point = HOPBACK3(strend, end_shift, rx_origin);
1087 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
1092 /* If the regex is absolutely anchored to either the start of the
1093 * string (SBOL) or to pos() (ANCH_GPOS), then
1094 * check_offset_max represents an upper bound on the string where
1095 * the substr could start. For the ANCH_GPOS case, we assume that
1096 * the caller of intuit will have already set strpos to
1097 * pos()-gofs, so in this case strpos + offset_max will still be
1098 * an upper bound on the substr.
1101 && prog->intflags & PREGf_ANCH
1102 && prog->check_offset_max != SSize_t_MAX)
1104 SSize_t check_len = SvCUR(check) - !!SvTAIL(check);
1105 const char * const anchor =
1106 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
1107 SSize_t targ_len = (char*)end_point - anchor;
1109 if (check_len > targ_len) {
1110 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1111 "Target string too short to match required substring...\n"));
1115 /* do a bytes rather than chars comparison. It's conservative;
1116 * so it skips doing the HOP if the result can't possibly end
1117 * up earlier than the old value of end_point.
1119 assert(anchor + check_len <= (char *)end_point);
1120 if (prog->check_offset_max + check_len < targ_len) {
1121 end_point = HOP3lim((U8*)anchor,
1122 prog->check_offset_max,
1123 end_point - check_len
1126 if (end_point < start_point)
1131 check_at = fbm_instr( start_point, end_point,
1132 check, multiline ? FBMrf_MULTILINE : 0);
1134 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1135 " doing 'check' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
1136 (IV)((char*)start_point - strbeg),
1137 (IV)((char*)end_point - strbeg),
1138 (IV)(check_at ? check_at - strbeg : -1)
1141 /* Update the count-of-usability, remove useless subpatterns,
1145 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1146 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
1147 Perl_re_printf( aTHX_ " %s %s substr %s%s%s",
1148 (check_at ? "Found" : "Did not find"),
1149 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
1150 ? "anchored" : "floating"),
1153 (check_at ? " at offset " : "...\n") );
1158 /* set rx_origin to the minimum position where the regex could start
1159 * matching, given the constraint of the just-matched check substring.
1160 * But don't set it lower than previously.
1163 if (check_at - rx_origin > prog->check_offset_max)
1164 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
1165 /* Finish the diagnostic message */
1166 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1167 "%ld (rx_origin now %" IVdf ")...\n",
1168 (long)(check_at - strbeg),
1169 (IV)(rx_origin - strbeg)
1174 /* now look for the 'other' substring if defined */
1176 if (prog->substrs->data[other_ix].utf8_substr
1177 || prog->substrs->data[other_ix].substr)
1179 /* Take into account the "other" substring. */
1183 struct reg_substr_datum *other;
1186 other = &prog->substrs->data[other_ix];
1187 if (!utf8_target && !other->substr) {
1188 if (!to_byte_substr(prog)) {
1189 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
1193 /* if "other" is anchored:
1194 * we've previously found a floating substr starting at check_at.
1195 * This means that the regex origin must lie somewhere
1196 * between min (rx_origin): HOP3(check_at, -check_offset_max)
1197 * and max: HOP3(check_at, -check_offset_min)
1198 * (except that min will be >= strpos)
1199 * So the fixed substr must lie somewhere between
1200 * HOP3(min, anchored_offset)
1201 * HOP3(max, anchored_offset) + SvCUR(substr)
1204 /* if "other" is floating
1205 * Calculate last1, the absolute latest point where the
1206 * floating substr could start in the string, ignoring any
1207 * constraints from the earlier fixed match. It is calculated
1210 * strend - prog->minlen (in chars) is the absolute latest
1211 * position within the string where the origin of the regex
1212 * could appear. The latest start point for the floating
1213 * substr is float_min_offset(*) on from the start of the
1214 * regex. last1 simply combines thee two offsets.
1216 * (*) You might think the latest start point should be
1217 * float_max_offset from the regex origin, and technically
1218 * you'd be correct. However, consider
1220 * Here, float min, max are 3,5 and minlen is 7.
1221 * This can match either
1225 * In the first case, the regex matches minlen chars; in the
1226 * second, minlen+1, in the third, minlen+2.
1227 * In the first case, the floating offset is 3 (which equals
1228 * float_min), in the second, 4, and in the third, 5 (which
1229 * equals float_max). In all cases, the floating string bcd
1230 * can never start more than 4 chars from the end of the
1231 * string, which equals minlen - float_min. As the substring
1232 * starts to match more than float_min from the start of the
1233 * regex, it makes the regex match more than minlen chars,
1234 * and the two cancel each other out. So we can always use
1235 * float_min - minlen, rather than float_max - minlen for the
1236 * latest position in the string.
1238 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1239 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1242 assert(prog->minlen >= other->min_offset);
1243 last1 = HOP3c(strend,
1244 other->min_offset - prog->minlen, strbeg);
1246 if (other_ix) {/* i.e. if (other-is-float) */
1247 /* last is the latest point where the floating substr could
1248 * start, *given* any constraints from the earlier fixed
1249 * match. This constraint is that the floating string starts
1250 * <= float_max_offset chars from the regex origin (rx_origin).
1251 * If this value is less than last1, use it instead.
1253 assert(rx_origin <= last1);
1255 /* this condition handles the offset==infinity case, and
1256 * is a short-cut otherwise. Although it's comparing a
1257 * byte offset to a char length, it does so in a safe way,
1258 * since 1 char always occupies 1 or more bytes,
1259 * so if a string range is (last1 - rx_origin) bytes,
1260 * it will be less than or equal to (last1 - rx_origin)
1261 * chars; meaning it errs towards doing the accurate HOP3
1262 * rather than just using last1 as a short-cut */
1263 (last1 - rx_origin) < other->max_offset
1265 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1268 assert(strpos + start_shift <= check_at);
1269 last = HOP4c(check_at, other->min_offset - start_shift,
1273 s = HOP3c(rx_origin, other->min_offset, strend);
1274 if (s < other_last) /* These positions already checked */
1277 must = utf8_target ? other->utf8_substr : other->substr;
1278 assert(SvPOK(must));
1281 char *to = last + SvCUR(must) - (SvTAIL(must)!=0);
1287 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1288 " skipping 'other' fbm scan: %" IVdf " > %" IVdf "\n",
1289 (IV)(from - strbeg),
1295 (unsigned char*)from,
1298 multiline ? FBMrf_MULTILINE : 0
1300 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1301 " doing 'other' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
1302 (IV)(from - strbeg),
1304 (IV)(s ? s - strbeg : -1)
1310 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1311 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1312 Perl_re_printf( aTHX_ " %s %s substr %s%s",
1313 s ? "Found" : "Contradicts",
1314 other_ix ? "floating" : "anchored",
1315 quoted, RE_SV_TAIL(must));
1320 /* last1 is latest possible substr location. If we didn't
1321 * find it before there, we never will */
1322 if (last >= last1) {
1323 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1324 "; giving up...\n"));
1328 /* try to find the check substr again at a later
1329 * position. Maybe next time we'll find the "other" substr
1331 other_last = HOP3c(last, 1, strend) /* highest failure */;
1333 other_ix /* i.e. if other-is-float */
1334 ? HOP3c(rx_origin, 1, strend)
1335 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1336 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1337 "; about to retry %s at offset %ld (rx_origin now %" IVdf ")...\n",
1338 (other_ix ? "floating" : "anchored"),
1339 (long)(HOP3c(check_at, 1, strend) - strbeg),
1340 (IV)(rx_origin - strbeg)
1345 if (other_ix) { /* if (other-is-float) */
1346 /* other_last is set to s, not s+1, since its possible for
1347 * a floating substr to fail first time, then succeed
1348 * second time at the same floating position; e.g.:
1349 * "-AB--AABZ" =~ /\wAB\d*Z/
1350 * The first time round, anchored and float match at
1351 * "-(AB)--AAB(Z)" then fail on the initial \w character
1352 * class. Second time round, they match at "-AB--A(AB)(Z)".
1357 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1358 other_last = HOP3c(s, 1, strend);
1360 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1361 " at offset %ld (rx_origin now %" IVdf ")...\n",
1363 (IV)(rx_origin - strbeg)
1369 DEBUG_OPTIMISE_MORE_r(
1370 Perl_re_printf( aTHX_
1371 " Check-only match: offset min:%" IVdf " max:%" IVdf
1372 " check_at:%" IVdf " rx_origin:%" IVdf " rx_origin-check_at:%" IVdf
1373 " strend:%" IVdf "\n",
1374 (IV)prog->check_offset_min,
1375 (IV)prog->check_offset_max,
1376 (IV)(check_at-strbeg),
1377 (IV)(rx_origin-strbeg),
1378 (IV)(rx_origin-check_at),
1384 postprocess_substr_matches:
1386 /* handle the extra constraint of /^.../m if present */
1388 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1391 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1392 " looking for /^/m anchor"));
1394 /* we have failed the constraint of a \n before rx_origin.
1395 * Find the next \n, if any, even if it's beyond the current
1396 * anchored and/or floating substrings. Whether we should be
1397 * scanning ahead for the next \n or the next substr is debatable.
1398 * On the one hand you'd expect rare substrings to appear less
1399 * often than \n's. On the other hand, searching for \n means
1400 * we're effectively flipping between check_substr and "\n" on each
1401 * iteration as the current "rarest" string candidate, which
1402 * means for example that we'll quickly reject the whole string if
1403 * hasn't got a \n, rather than trying every substr position
1407 s = HOP3c(strend, - prog->minlen, strpos);
1408 if (s <= rx_origin ||
1409 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1411 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1412 " Did not find /%s^%s/m...\n",
1413 PL_colors[0], PL_colors[1]));
1417 /* earliest possible origin is 1 char after the \n.
1418 * (since *rx_origin == '\n', it's safe to ++ here rather than
1419 * HOP(rx_origin, 1)) */
1422 if (prog->substrs->check_ix == 0 /* check is anchored */
1423 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1425 /* Position contradicts check-string; either because
1426 * check was anchored (and thus has no wiggle room),
1427 * or check was float and rx_origin is above the float range */
1428 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1429 " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
1430 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1434 /* if we get here, the check substr must have been float,
1435 * is in range, and we may or may not have had an anchored
1436 * "other" substr which still contradicts */
1437 assert(prog->substrs->check_ix); /* check is float */
1439 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1440 /* whoops, the anchored "other" substr exists, so we still
1441 * contradict. On the other hand, the float "check" substr
1442 * didn't contradict, so just retry the anchored "other"
1444 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1445 " Found /%s^%s/m, rescanning for anchored from offset %" IVdf " (rx_origin now %" IVdf ")...\n",
1446 PL_colors[0], PL_colors[1],
1447 (IV)(rx_origin - strbeg + prog->anchored_offset),
1448 (IV)(rx_origin - strbeg)
1450 goto do_other_substr;
1453 /* success: we don't contradict the found floating substring
1454 * (and there's no anchored substr). */
1455 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1456 " Found /%s^%s/m with rx_origin %ld...\n",
1457 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1460 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1461 " (multiline anchor test skipped)\n"));
1467 /* if we have a starting character class, then test that extra constraint.
1468 * (trie stclasses are too expensive to use here, we are better off to
1469 * leave it to regmatch itself) */
1471 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1472 const U8* const str = (U8*)STRING(progi->regstclass);
1474 /* XXX this value could be pre-computed */
1475 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1476 ? (reginfo->is_utf8_pat
1477 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1478 : STR_LEN(progi->regstclass))
1482 /* latest pos that a matching float substr constrains rx start to */
1483 char *rx_max_float = NULL;
1485 /* if the current rx_origin is anchored, either by satisfying an
1486 * anchored substring constraint, or a /^.../m constraint, then we
1487 * can reject the current origin if the start class isn't found
1488 * at the current position. If we have a float-only match, then
1489 * rx_origin is constrained to a range; so look for the start class
1490 * in that range. if neither, then look for the start class in the
1491 * whole rest of the string */
1493 /* XXX DAPM it's not clear what the minlen test is for, and why
1494 * it's not used in the floating case. Nothing in the test suite
1495 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1496 * Here are some old comments, which may or may not be correct:
1498 * minlen == 0 is possible if regstclass is \b or \B,
1499 * and the fixed substr is ''$.
1500 * Since minlen is already taken into account, rx_origin+1 is
1501 * before strend; accidentally, minlen >= 1 guaranties no false
1502 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1503 * 0) below assumes that regstclass does not come from lookahead...
1504 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1505 * This leaves EXACTF-ish only, which are dealt with in
1509 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1510 endpos = HOP3clim(rx_origin, (prog->minlen ? cl_l : 0), strend);
1511 else if (prog->float_substr || prog->float_utf8) {
1512 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1513 endpos = HOP3clim(rx_max_float, cl_l, strend);
1518 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1519 " looking for class: start_shift: %" IVdf " check_at: %" IVdf
1520 " rx_origin: %" IVdf " endpos: %" IVdf "\n",
1521 (IV)start_shift, (IV)(check_at - strbeg),
1522 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1524 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1527 if (endpos == strend) {
1528 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1529 " Could not match STCLASS...\n") );
1532 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1533 " This position contradicts STCLASS...\n") );
1534 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1535 && !(prog->intflags & PREGf_IMPLICIT))
1538 /* Contradict one of substrings */
1539 if (prog->anchored_substr || prog->anchored_utf8) {
1540 if (prog->substrs->check_ix == 1) { /* check is float */
1541 /* Have both, check_string is floating */
1542 assert(rx_origin + start_shift <= check_at);
1543 if (rx_origin + start_shift != check_at) {
1544 /* not at latest position float substr could match:
1545 * Recheck anchored substring, but not floating.
1546 * The condition above is in bytes rather than
1547 * chars for efficiency. It's conservative, in
1548 * that it errs on the side of doing 'goto
1549 * do_other_substr'. In this case, at worst,
1550 * an extra anchored search may get done, but in
1551 * practice the extra fbm_instr() is likely to
1552 * get skipped anyway. */
1553 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1554 " about to retry anchored at offset %ld (rx_origin now %" IVdf ")...\n",
1555 (long)(other_last - strbeg),
1556 (IV)(rx_origin - strbeg)
1558 goto do_other_substr;
1566 /* In the presence of ml_anch, we might be able to
1567 * find another \n without breaking the current float
1570 /* strictly speaking this should be HOP3c(..., 1, ...),
1571 * but since we goto a block of code that's going to
1572 * search for the next \n if any, its safe here */
1574 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1575 " about to look for /%s^%s/m starting at rx_origin %ld...\n",
1576 PL_colors[0], PL_colors[1],
1577 (long)(rx_origin - strbeg)) );
1578 goto postprocess_substr_matches;
1581 /* strictly speaking this can never be true; but might
1582 * be if we ever allow intuit without substrings */
1583 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1586 rx_origin = rx_max_float;
1589 /* at this point, any matching substrings have been
1590 * contradicted. Start again... */
1592 rx_origin = HOP3c(rx_origin, 1, strend);
1594 /* uses bytes rather than char calculations for efficiency.
1595 * It's conservative: it errs on the side of doing 'goto restart',
1596 * where there is code that does a proper char-based test */
1597 if (rx_origin + start_shift + end_shift > strend) {
1598 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1599 " Could not match STCLASS...\n") );
1602 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1603 " about to look for %s substr starting at offset %ld (rx_origin now %" IVdf ")...\n",
1604 (prog->substrs->check_ix ? "floating" : "anchored"),
1605 (long)(rx_origin + start_shift - strbeg),
1606 (IV)(rx_origin - strbeg)
1613 if (rx_origin != s) {
1614 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1615 " By STCLASS: moving %ld --> %ld\n",
1616 (long)(rx_origin - strbeg), (long)(s - strbeg))
1620 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1621 " Does not contradict STCLASS...\n");
1626 /* Decide whether using the substrings helped */
1628 if (rx_origin != strpos) {
1629 /* Fixed substring is found far enough so that the match
1630 cannot start at strpos. */
1632 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " try at offset...\n"));
1633 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1636 /* The found rx_origin position does not prohibit matching at
1637 * strpos, so calling intuit didn't gain us anything. Decrement
1638 * the BmUSEFUL() count on the check substring, and if we reach
1640 if (!(prog->intflags & PREGf_NAUGHTY)
1642 prog->check_utf8 /* Could be deleted already */
1643 && --BmUSEFUL(prog->check_utf8) < 0
1644 && (prog->check_utf8 == prog->float_utf8)
1646 prog->check_substr /* Could be deleted already */
1647 && --BmUSEFUL(prog->check_substr) < 0
1648 && (prog->check_substr == prog->float_substr)
1651 /* If flags & SOMETHING - do not do it many times on the same match */
1652 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " ... Disabling check substring...\n"));
1653 /* XXX Does the destruction order has to change with utf8_target? */
1654 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1655 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1656 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1657 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1658 check = NULL; /* abort */
1659 /* XXXX This is a remnant of the old implementation. It
1660 looks wasteful, since now INTUIT can use many
1661 other heuristics. */
1662 prog->extflags &= ~RXf_USE_INTUIT;
1666 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1667 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1668 PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );
1672 fail_finish: /* Substring not found */
1673 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1674 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1676 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch rejected by optimizer%s\n",
1677 PL_colors[4], PL_colors[5]));
1682 #define DECL_TRIE_TYPE(scan) \
1683 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1684 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1685 trie_utf8l, trie_flu8, trie_flu8_latin } \
1686 trie_type = ((scan->flags == EXACT) \
1687 ? (utf8_target ? trie_utf8 : trie_plain) \
1688 : (scan->flags == EXACTL) \
1689 ? (utf8_target ? trie_utf8l : trie_plain) \
1690 : (scan->flags == EXACTFAA) \
1692 ? trie_utf8_exactfa_fold \
1693 : trie_latin_utf8_exactfa_fold) \
1694 : (scan->flags == EXACTFLU8 \
1697 : trie_flu8_latin) \
1700 : trie_latin_utf8_fold)))
1702 /* 'uscan' is set to foldbuf, and incremented, so below the end of uscan is
1703 * 'foldbuf+sizeof(foldbuf)' */
1704 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uc_end, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1707 U8 flags = FOLD_FLAGS_FULL; \
1708 switch (trie_type) { \
1710 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1711 if (UTF8_IS_ABOVE_LATIN1(*uc)) { \
1712 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end); \
1714 goto do_trie_utf8_fold; \
1715 case trie_utf8_exactfa_fold: \
1716 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1718 case trie_utf8_fold: \
1719 do_trie_utf8_fold: \
1720 if ( foldlen>0 ) { \
1721 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \
1726 uvc = _toFOLD_utf8_flags( (const U8*) uc, uc_end, foldbuf, &foldlen, \
1728 len = UTF8_SAFE_SKIP(uc, uc_end); \
1729 skiplen = UVCHR_SKIP( uvc ); \
1730 foldlen -= skiplen; \
1731 uscan = foldbuf + skiplen; \
1734 case trie_flu8_latin: \
1735 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1736 goto do_trie_latin_utf8_fold; \
1737 case trie_latin_utf8_exactfa_fold: \
1738 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1740 case trie_latin_utf8_fold: \
1741 do_trie_latin_utf8_fold: \
1742 if ( foldlen>0 ) { \
1743 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \
1749 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1750 skiplen = UVCHR_SKIP( uvc ); \
1751 foldlen -= skiplen; \
1752 uscan = foldbuf + skiplen; \
1756 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1757 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1758 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end); \
1762 uvc = utf8n_to_uvchr( (const U8*) uc, uc_end - uc, &len, uniflags ); \
1769 charid = trie->charmap[ uvc ]; \
1773 if (widecharmap) { \
1774 SV** const svpp = hv_fetch(widecharmap, \
1775 (char*)&uvc, sizeof(UV), 0); \
1777 charid = (U16)SvIV(*svpp); \
1782 #define DUMP_EXEC_POS(li,s,doutf8,depth) \
1783 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1784 startpos, doutf8, depth)
1786 #define REXEC_FBC_SCAN(UTF8, CODE) \
1788 while (s < strend) { \
1791 ? UTF8_SAFE_SKIP(s, reginfo->strend) \
1796 #define REXEC_FBC_CLASS_SCAN(UTF8, COND) \
1798 while (s < strend) { \
1799 REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND) \
1803 #define REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND) \
1806 s += ((UTF8) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1);\
1807 previous_occurrence_end = s; \
1810 s += ((UTF8) ? UTF8SKIP(s) : 1); \
1813 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1814 if (utf8_target) { \
1815 REXEC_FBC_CLASS_SCAN(1, CONDUTF8); \
1818 REXEC_FBC_CLASS_SCAN(0, COND); \
1821 /* We keep track of where the next character should start after an occurrence
1822 * of the one we're looking for. Knowing that, we can see right away if the
1823 * next occurrence is adjacent to the previous. When 'doevery' is FALSE, we
1824 * don't accept the 2nd and succeeding adjacent occurrences */
1825 #define FBC_CHECK_AND_TRY \
1827 || s != previous_occurrence_end) \
1828 && ( reginfo->intuit \
1829 || (s <= reginfo->strend && regtry(reginfo, &s)))) \
1835 /* This differs from the above macros in that it calls a function which returns
1836 * the next occurrence of the thing being looked for in 's'; and 'strend' if
1837 * there is no such occurrence. */
1838 #define REXEC_FBC_FIND_NEXT_SCAN(UTF8, f) \
1839 while (s < strend) { \
1841 if (s >= strend) { \
1846 s += (UTF8) ? UTF8SKIP(s) : 1; \
1847 previous_occurrence_end = s; \
1850 /* This differs from the above macros in that it is passed a single byte that
1851 * is known to begin the next occurrence of the thing being looked for in 's'.
1852 * It does a memchr to find the next occurrence of 'byte', before trying 'COND'
1853 * at that position. */
1854 #define REXEC_FBC_FIND_NEXT_UTF8_BYTE_SCAN(byte, COND) \
1855 while (s < strend) { \
1856 s = (char *) memchr(s, byte, strend -s); \
1858 s = (char *) strend; \
1864 s += UTF8_SAFE_SKIP(s, reginfo->strend); \
1865 previous_occurrence_end = s; \
1872 /* The three macros below are slightly different versions of the same logic.
1874 * The first is for /a and /aa when the target string is UTF-8. This can only
1875 * match ascii, but it must advance based on UTF-8. The other two handle the
1876 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1877 * for the boundary (or non-boundary) between a word and non-word character.
1878 * The utf8 and non-utf8 cases have the same logic, but the details must be
1879 * different. Find the "wordness" of the character just prior to this one, and
1880 * compare it with the wordness of this one. If they differ, we have a
1881 * boundary. At the beginning of the string, pretend that the previous
1882 * character was a new-line.
1884 * All these macros uncleanly have side-effects with each other and outside
1885 * variables. So far it's been too much trouble to clean-up
1887 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1888 * a word character or not.
1889 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1891 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1893 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1894 * are looking for a boundary or for a non-boundary. If we are looking for a
1895 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1896 * see if this tentative match actually works, and if so, to quit the loop
1897 * here. And vice-versa if we are looking for a non-boundary.
1899 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1900 * REXEC_FBC_SCAN loops is a loop invariant, a bool giving the return of
1901 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1902 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1903 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1904 * complement. But in that branch we complement tmp, meaning that at the
1905 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1906 * which means at the top of the loop in the next iteration, it is
1907 * TEST_NON_UTF8(s-1) */
1908 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1909 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1910 tmp = TEST_NON_UTF8(tmp); \
1911 REXEC_FBC_SCAN(1, /* 1=>is-utf8; advances s while s < strend */ \
1912 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1914 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1921 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1922 * TEST_UTF8 is a macro that for the same input code points returns identically
1923 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1924 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1925 if (s == reginfo->strbeg) { \
1928 else { /* Back-up to the start of the previous character */ \
1929 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1930 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1931 0, UTF8_ALLOW_DEFAULT); \
1933 tmp = TEST_UV(tmp); \
1934 REXEC_FBC_SCAN(1, /* 1=>is-utf8; advances s while s < strend */ \
1935 if (tmp == ! (TEST_UTF8((U8 *) s, (U8 *) reginfo->strend))) { \
1944 /* Like the above two macros. UTF8_CODE is the complete code for handling
1945 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1947 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1948 if (utf8_target) { \
1951 else { /* Not utf8 */ \
1952 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1953 tmp = TEST_NON_UTF8(tmp); \
1954 REXEC_FBC_SCAN(0, /* 0=>not-utf8; advances s while s < strend */ \
1955 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1964 /* Here, things have been set up by the previous code so that tmp is the \
1965 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1966 * utf8ness of the target). We also have to check if this matches against \
1967 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1968 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1970 if (tmp == ! TEST_NON_UTF8('\n')) { \
1977 /* This is the macro to use when we want to see if something that looks like it
1978 * could match, actually does, and if so exits the loop. It needs to be used
1979 * only for bounds checking macros, as it allows for matching beyond the end of
1980 * string (which should be zero length without having to look at the string
1982 #define REXEC_FBC_TRYIT \
1983 if (reginfo->intuit || (s <= reginfo->strend && regtry(reginfo, &s))) \
1986 /* The only difference between the BOUND and NBOUND cases is that
1987 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1988 * NBOUND. This is accomplished by passing it as either the if or else clause,
1989 * with the other one being empty (PLACEHOLDER is defined as empty).
1991 * The TEST_FOO parameters are for operating on different forms of input, but
1992 * all should be ones that return identically for the same underlying code
1994 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1996 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1997 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1999 #define FBC_BOUND_A(TEST_NON_UTF8) \
2001 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
2002 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
2004 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
2006 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
2007 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
2009 #define FBC_NBOUND_A(TEST_NON_UTF8) \
2011 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
2012 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
2016 S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) {
2017 IV cp_out = _invlist_search(invlist, cp_in);
2018 assert(cp_out >= 0);
2021 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
2022 invmap[S_get_break_val_cp_checked(invlist, cp)]
2024 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
2025 invmap[_invlist_search(invlist, cp)]
2028 /* Takes a pointer to an inversion list, a pointer to its corresponding
2029 * inversion map, and a code point, and returns the code point's value
2030 * according to the two arrays. It assumes that all code points have a value.
2031 * This is used as the base macro for macros for particular properties */
2032 #define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \
2033 _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp)
2035 /* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
2036 * of a code point, returning the value for the first code point in the string.
2037 * And it takes the particular macro name that finds the desired value given a
2038 * code point. Merely convert the UTF-8 to code point and call the cp macro */
2039 #define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \
2040 (__ASSERT_(pos < strend) \
2041 /* Note assumes is valid UTF-8 */ \
2042 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))
2044 /* Returns the GCB value for the input code point */
2045 #define getGCB_VAL_CP(cp) \
2046 _generic_GET_BREAK_VAL_CP( \
2051 /* Returns the GCB value for the first code point in the UTF-8 encoded string
2052 * bounded by pos and strend */
2053 #define getGCB_VAL_UTF8(pos, strend) \
2054 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)
2056 /* Returns the LB value for the input code point */
2057 #define getLB_VAL_CP(cp) \
2058 _generic_GET_BREAK_VAL_CP( \
2063 /* Returns the LB value for the first code point in the UTF-8 encoded string
2064 * bounded by pos and strend */
2065 #define getLB_VAL_UTF8(pos, strend) \
2066 _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend)
2069 /* Returns the SB value for the input code point */
2070 #define getSB_VAL_CP(cp) \
2071 _generic_GET_BREAK_VAL_CP( \
2076 /* Returns the SB value for the first code point in the UTF-8 encoded string
2077 * bounded by pos and strend */
2078 #define getSB_VAL_UTF8(pos, strend) \
2079 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)
2081 /* Returns the WB value for the input code point */
2082 #define getWB_VAL_CP(cp) \
2083 _generic_GET_BREAK_VAL_CP( \
2088 /* Returns the WB value for the first code point in the UTF-8 encoded string
2089 * bounded by pos and strend */
2090 #define getWB_VAL_UTF8(pos, strend) \
2091 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)
2093 /* We know what class REx starts with. Try to find this position... */
2094 /* if reginfo->intuit, its a dryrun */
2095 /* annoyingly all the vars in this routine have different names from their counterparts
2096 in regmatch. /grrr */
2098 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
2099 const char *strend, regmatch_info *reginfo)
2103 /* TRUE if x+ need not match at just the 1st pos of run of x's */
2104 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
2106 char *pat_string; /* The pattern's exactish string */
2107 char *pat_end; /* ptr to end char of pat_string */
2108 re_fold_t folder; /* Function for computing non-utf8 folds */
2109 const U8 *fold_array; /* array for folding ords < 256 */
2116 /* In some cases we accept only the first occurence of 'x' in a sequence of
2117 * them. This variable points to just beyond the end of the previous
2118 * occurrence of 'x', hence we can tell if we are in a sequence. (Having
2119 * it point to beyond the 'x' allows us to work for UTF-8 without having to
2121 char * previous_occurrence_end = 0;
2123 I32 tmp; /* Scratch variable */
2124 const bool utf8_target = reginfo->is_utf8_target;
2125 UV utf8_fold_flags = 0;
2126 const bool is_utf8_pat = reginfo->is_utf8_pat;
2127 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
2128 with a result inverts that result, as 0^1 =
2130 _char_class_number classnum;
2132 RXi_GET_DECL(prog,progi);
2134 PERL_ARGS_ASSERT_FIND_BYCLASS;
2136 /* We know what class it must start with. */
2140 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2142 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) {
2143 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
2150 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2151 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
2153 else if (ANYOF_FLAGS(c) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
2154 /* We know that s is in the bitmap range since the target isn't
2155 * UTF-8, so what happens for out-of-range values is not relevant,
2156 * so exclude that from the flags */
2157 REXEC_FBC_CLASS_SCAN(0, reginclass(prog,c, (U8*)s, (U8*)s+1, 0));
2160 REXEC_FBC_CLASS_SCAN(0, ANYOF_BITMAP_TEST(c, *((U8*)s)));
2164 case ANYOFM: /* ARG() is the base byte; FLAGS() the mask byte */
2165 /* UTF-8ness doesn't matter because only matches UTF-8 invariants, so
2167 REXEC_FBC_FIND_NEXT_SCAN(0,
2168 (char *) find_next_masked((U8 *) s, (U8 *) strend,
2169 (U8) ARG(c), FLAGS(c)));
2172 case NANYOFM: /* UTF-8ness does matter because can match UTF-8 variants.
2174 REXEC_FBC_FIND_NEXT_SCAN(utf8_target,
2175 (char *) find_span_end_mask((U8 *) s, (U8 *) strend,
2176 (U8) ARG(c), FLAGS(c)));
2180 if (utf8_target) { /* Can't possibly match a non-UTF-8 target */
2181 REXEC_FBC_CLASS_SCAN(TRUE,
2182 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
2187 if (utf8_target) { /* Can't possibly match a non-UTF-8 target */
2189 /* We know what the first byte of any matched string should be */
2190 U8 first_byte = FLAGS(c);
2192 REXEC_FBC_FIND_NEXT_UTF8_BYTE_SCAN(first_byte,
2193 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
2197 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
2198 assert(! is_utf8_pat);
2202 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII
2203 |FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
2204 goto do_exactf_utf8;
2206 else if (utf8_target) {
2208 /* Here, and elsewhere in this file, the reason we can't consider a
2209 * non-UTF-8 pattern already folded in the presence of a UTF-8
2210 * target is because any MICRO SIGN in the pattern won't be folded.
2211 * Since the fold of the MICRO SIGN requires UTF-8 to represent, we
2212 * can consider a non-UTF-8 pattern folded when matching a
2213 * non-UTF-8 target */
2214 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
2215 goto do_exactf_utf8;
2218 /* Latin1 folds are not affected by /a, except it excludes the sharp s,
2219 * which these functions don't handle anyway */
2220 fold_array = PL_fold_latin1;
2221 folder = foldEQ_latin1_s2_folded;
2222 goto do_exactf_non_utf8;
2224 case EXACTF: /* This node only generated for non-utf8 patterns */
2225 assert(! is_utf8_pat);
2227 goto do_exactf_utf8;
2229 fold_array = PL_fold;
2231 goto do_exactf_non_utf8;
2234 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2235 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
2236 utf8_fold_flags = FOLDEQ_LOCALE;
2237 goto do_exactf_utf8;
2239 fold_array = PL_fold_locale;
2240 folder = foldEQ_locale;
2241 goto do_exactf_non_utf8;
2243 case EXACTFUP: /* Problematic even though pattern isn't UTF-8. Use
2244 full functionality normally not done except for
2246 assert(! is_utf8_pat);
2247 goto do_exactf_utf8;
2250 if (! utf8_target) { /* All code points in this node require
2251 UTF-8 to express. */
2254 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
2255 | FOLDEQ_S2_FOLDS_SANE;
2256 goto do_exactf_utf8;
2259 if (! utf8_target) {
2262 assert(is_utf8_pat);
2263 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
2264 goto do_exactf_utf8;
2267 if (is_utf8_pat || utf8_target) {
2268 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
2269 goto do_exactf_utf8;
2272 /* Any 'ss' in the pattern should have been replaced by regcomp,
2273 * so we don't have to worry here about this single special case
2274 * in the Latin1 range */
2275 fold_array = PL_fold_latin1;
2276 folder = foldEQ_latin1_s2_folded;
2280 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
2281 are no glitches with fold-length differences
2282 between the target string and pattern */
2284 /* The idea in the non-utf8 EXACTF* cases is to first find the
2285 * first character of the EXACTF* node and then, if necessary,
2286 * case-insensitively compare the full text of the node. c1 is the
2287 * first character. c2 is its fold. This logic will not work for
2288 * Unicode semantics and the german sharp ss, which hence should
2289 * not be compiled into a node that gets here. */
2290 pat_string = STRING(c);
2291 ln = STR_LEN(c); /* length to match in octets/bytes */
2293 /* We know that we have to match at least 'ln' bytes (which is the
2294 * same as characters, since not utf8). If we have to match 3
2295 * characters, and there are only 2 availabe, we know without
2296 * trying that it will fail; so don't start a match past the
2297 * required minimum number from the far end */
2298 e = HOP3c(strend, -((SSize_t)ln), s);
2303 c2 = fold_array[c1];
2304 if (c1 == c2) { /* If char and fold are the same */
2306 s = (char *) memchr(s, c1, e + 1 - s);
2311 /* Check that the rest of the node matches */
2312 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2313 && (reginfo->intuit || regtry(reginfo, &s)) )
2321 U8 bits_differing = c1 ^ c2;
2323 /* If the folds differ in one bit position only, we can mask to
2324 * match either of them, and can use this faster find method. Both
2325 * ASCII and EBCDIC tend to have their case folds differ in only
2326 * one position, so this is very likely */
2327 if (LIKELY(PL_bitcount[bits_differing] == 1)) {
2328 bits_differing = ~ bits_differing;
2330 s = (char *) find_next_masked((U8 *) s, (U8 *) e + 1,
2331 (c1 & bits_differing), bits_differing);
2336 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2337 && (reginfo->intuit || regtry(reginfo, &s)) )
2344 else { /* Otherwise, stuck with looking byte-at-a-time. This
2345 should actually happen only in EXACTFL nodes */
2347 if ( (*(U8*)s == c1 || *(U8*)s == c2)
2348 && (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2349 && (reginfo->intuit || regtry(reginfo, &s)) )
2363 /* If one of the operands is in utf8, we can't use the simpler folding
2364 * above, due to the fact that many different characters can have the
2365 * same fold, or portion of a fold, or different- length fold */
2366 pat_string = STRING(c);
2367 ln = STR_LEN(c); /* length to match in octets/bytes */
2368 pat_end = pat_string + ln;
2369 lnc = is_utf8_pat /* length to match in characters */
2370 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
2373 /* We have 'lnc' characters to match in the pattern, but because of
2374 * multi-character folding, each character in the target can match
2375 * up to 3 characters (Unicode guarantees it will never exceed
2376 * this) if it is utf8-encoded; and up to 2 if not (based on the
2377 * fact that the Latin 1 folds are already determined, and the
2378 * only multi-char fold in that range is the sharp-s folding to
2379 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
2380 * string character. Adjust lnc accordingly, rounding up, so that
2381 * if we need to match at least 4+1/3 chars, that really is 5. */
2382 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
2383 lnc = (lnc + expansion - 1) / expansion;
2385 /* As in the non-UTF8 case, if we have to match 3 characters, and
2386 * only 2 are left, it's guaranteed to fail, so don't start a
2387 * match that would require us to go beyond the end of the string
2389 e = HOP3c(strend, -((SSize_t)lnc), s);
2391 /* XXX Note that we could recalculate e to stop the loop earlier,
2392 * as the worst case expansion above will rarely be met, and as we
2393 * go along we would usually find that e moves further to the left.
2394 * This would happen only after we reached the point in the loop
2395 * where if there were no expansion we should fail. Unclear if
2396 * worth the expense */
2399 char *my_strend= (char *)strend;
2400 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
2401 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
2402 && (reginfo->intuit || regtry(reginfo, &s)) )
2406 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2412 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2413 if (FLAGS(c) != TRADITIONAL_BOUND) {
2414 if (! IN_UTF8_CTYPE_LOCALE) {
2415 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2416 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2421 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2425 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2426 if (FLAGS(c) != TRADITIONAL_BOUND) {
2427 if (! IN_UTF8_CTYPE_LOCALE) {
2428 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2429 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2434 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2437 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2439 assert(FLAGS(c) == TRADITIONAL_BOUND);
2441 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2444 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2446 assert(FLAGS(c) == TRADITIONAL_BOUND);
2448 FBC_BOUND_A(isWORDCHAR_A);
2451 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2453 assert(FLAGS(c) == TRADITIONAL_BOUND);
2455 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2458 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2460 assert(FLAGS(c) == TRADITIONAL_BOUND);
2462 FBC_NBOUND_A(isWORDCHAR_A);
2466 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
2467 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2478 switch((bound_type) FLAGS(c)) {
2479 case TRADITIONAL_BOUND:
2480 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2483 if (s == reginfo->strbeg) {
2484 if (reginfo->intuit || regtry(reginfo, &s))
2489 /* Didn't match. Try at the next position (if there is one) */
2490 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2491 if (UNLIKELY(s >= reginfo->strend)) {
2497 GCB_enum before = getGCB_VAL_UTF8(
2499 (U8*)(reginfo->strbeg)),
2500 (U8*) reginfo->strend);
2501 while (s < strend) {
2502 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
2503 (U8*) reginfo->strend);
2504 if ( (to_complement ^ isGCB(before,
2506 (U8*) reginfo->strbeg,
2509 && (reginfo->intuit || regtry(reginfo, &s)))
2514 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2517 else { /* Not utf8. Everything is a GCB except between CR and
2519 while (s < strend) {
2520 if ((to_complement ^ ( UCHARAT(s - 1) != '\r'
2521 || UCHARAT(s) != '\n'))
2522 && (reginfo->intuit || regtry(reginfo, &s)))
2530 /* And, since this is a bound, it can match after the final
2531 * character in the string */
2532 if ( reginfo->intuit
2533 || (s <= reginfo->strend && regtry(reginfo, &s)))
2540 if (s == reginfo->strbeg) {
2541 if (reginfo->intuit || regtry(reginfo, &s)) {
2544 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2545 if (UNLIKELY(s >= reginfo->strend)) {
2551 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
2553 (U8*)(reginfo->strbeg)),
2554 (U8*) reginfo->strend);
2555 while (s < strend) {
2556 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
2557 if (to_complement ^ isLB(before,
2559 (U8*) reginfo->strbeg,
2561 (U8*) reginfo->strend,
2563 && (reginfo->intuit || regtry(reginfo, &s)))
2568 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2571 else { /* Not utf8. */
2572 LB_enum before = getLB_VAL_CP((U8) *(s -1));
2573 while (s < strend) {
2574 LB_enum after = getLB_VAL_CP((U8) *s);
2575 if (to_complement ^ isLB(before,
2577 (U8*) reginfo->strbeg,
2579 (U8*) reginfo->strend,
2581 && (reginfo->intuit || regtry(reginfo, &s)))
2590 if ( reginfo->intuit
2591 || (s <= reginfo->strend && regtry(reginfo, &s)))
2599 if (s == reginfo->strbeg) {
2600 if (reginfo->intuit || regtry(reginfo, &s)) {
2603 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2604 if (UNLIKELY(s >= reginfo->strend)) {
2610 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
2612 (U8*)(reginfo->strbeg)),
2613 (U8*) reginfo->strend);
2614 while (s < strend) {
2615 SB_enum after = getSB_VAL_UTF8((U8*) s,
2616 (U8*) reginfo->strend);
2617 if ((to_complement ^ isSB(before,
2619 (U8*) reginfo->strbeg,
2621 (U8*) reginfo->strend,
2623 && (reginfo->intuit || regtry(reginfo, &s)))
2628 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2631 else { /* Not utf8. */
2632 SB_enum before = getSB_VAL_CP((U8) *(s -1));
2633 while (s < strend) {
2634 SB_enum after = getSB_VAL_CP((U8) *s);
2635 if ((to_complement ^ isSB(before,
2637 (U8*) reginfo->strbeg,
2639 (U8*) reginfo->strend,
2641 && (reginfo->intuit || regtry(reginfo, &s)))
2650 /* Here are at the final position in the target string. The SB
2651 * value is always true here, so matches, depending on other
2653 if ( reginfo->intuit
2654 || (s <= reginfo->strend && regtry(reginfo, &s)))
2662 if (s == reginfo->strbeg) {
2663 if (reginfo->intuit || regtry(reginfo, &s)) {
2666 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2667 if (UNLIKELY(s >= reginfo->strend)) {
2673 /* We are at a boundary between char_sub_0 and char_sub_1.
2674 * We also keep track of the value for char_sub_-1 as we
2675 * loop through the line. Context may be needed to make a
2676 * determination, and if so, this can save having to
2678 WB_enum previous = WB_UNKNOWN;
2679 WB_enum before = getWB_VAL_UTF8(
2682 (U8*)(reginfo->strbeg)),
2683 (U8*) reginfo->strend);
2684 while (s < strend) {
2685 WB_enum after = getWB_VAL_UTF8((U8*) s,
2686 (U8*) reginfo->strend);
2687 if ((to_complement ^ isWB(previous,
2690 (U8*) reginfo->strbeg,
2692 (U8*) reginfo->strend,
2694 && (reginfo->intuit || regtry(reginfo, &s)))
2700 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2703 else { /* Not utf8. */
2704 WB_enum previous = WB_UNKNOWN;
2705 WB_enum before = getWB_VAL_CP((U8) *(s -1));
2706 while (s < strend) {
2707 WB_enum after = getWB_VAL_CP((U8) *s);
2708 if ((to_complement ^ isWB(previous,
2711 (U8*) reginfo->strbeg,
2713 (U8*) reginfo->strend,
2715 && (reginfo->intuit || regtry(reginfo, &s)))
2725 if ( reginfo->intuit
2726 || (s <= reginfo->strend && regtry(reginfo, &s)))
2734 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2735 is_LNBREAK_latin1_safe(s, strend)
2739 /* The argument to all the POSIX node types is the class number to pass to
2740 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2747 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2748 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s, (U8 *) strend)),
2749 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
2764 /* The complement of something that matches only ASCII matches all
2765 * non-ASCII, plus everything in ASCII that isn't in the class. */
2766 REXEC_FBC_CLASS_SCAN(1, ! isASCII_utf8_safe(s, strend)
2767 || ! _generic_isCC_A(*s, FLAGS(c)));
2775 /* Don't need to worry about utf8, as it can match only a single
2776 * byte invariant character. But we do anyway for performance reasons,
2777 * as otherwise we would have to examine all the continuation
2780 REXEC_FBC_CLASS_SCAN(1, _generic_isCC_A(*s, FLAGS(c)));
2785 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
2786 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
2794 if (! utf8_target) {
2795 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
2796 to_complement ^ cBOOL(_generic_isCC(*s,
2802 classnum = (_char_class_number) FLAGS(c);
2805 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2806 to_complement ^ cBOOL(_invlist_contains_cp(
2807 PL_XPosix_ptrs[classnum],
2808 utf8_to_uvchr_buf((U8 *) s,
2812 case _CC_ENUM_SPACE:
2813 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2814 to_complement ^ cBOOL(isSPACE_utf8_safe(s, strend)));
2817 case _CC_ENUM_BLANK:
2818 REXEC_FBC_CLASS_SCAN(1,
2819 to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
2822 case _CC_ENUM_XDIGIT:
2823 REXEC_FBC_CLASS_SCAN(1,
2824 to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
2827 case _CC_ENUM_VERTSPACE:
2828 REXEC_FBC_CLASS_SCAN(1,
2829 to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
2832 case _CC_ENUM_CNTRL:
2833 REXEC_FBC_CLASS_SCAN(1,
2834 to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
2844 /* what trie are we using right now */
2845 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2846 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2847 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2849 const char *last_start = strend - trie->minlen;
2851 const char *real_start = s;
2853 STRLEN maxlen = trie->maxlen;
2855 U8 **points; /* map of where we were in the input string
2856 when reading a given char. For ASCII this
2857 is unnecessary overhead as the relationship
2858 is always 1:1, but for Unicode, especially
2859 case folded Unicode this is not true. */
2860 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2864 GET_RE_DEBUG_FLAGS_DECL;
2866 /* We can't just allocate points here. We need to wrap it in
2867 * an SV so it gets freed properly if there is a croak while
2868 * running the match */
2871 sv_points=newSV(maxlen * sizeof(U8 *));
2872 SvCUR_set(sv_points,
2873 maxlen * sizeof(U8 *));
2874 SvPOK_on(sv_points);
2875 sv_2mortal(sv_points);
2876 points=(U8**)SvPV_nolen(sv_points );
2877 if ( trie_type != trie_utf8_fold
2878 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2881 bitmap=(U8*)trie->bitmap;
2883 bitmap=(U8*)ANYOF_BITMAP(c);
2885 /* this is the Aho-Corasick algorithm modified a touch
2886 to include special handling for long "unknown char" sequences.
2887 The basic idea being that we use AC as long as we are dealing
2888 with a possible matching char, when we encounter an unknown char
2889 (and we have not encountered an accepting state) we scan forward
2890 until we find a legal starting char.
2891 AC matching is basically that of trie matching, except that when
2892 we encounter a failing transition, we fall back to the current
2893 states "fail state", and try the current char again, a process
2894 we repeat until we reach the root state, state 1, or a legal
2895 transition. If we fail on the root state then we can either
2896 terminate if we have reached an accepting state previously, or
2897 restart the entire process from the beginning if we have not.
2900 while (s <= last_start) {
2901 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2909 U8 *uscan = (U8*)NULL;
2910 U8 *leftmost = NULL;
2912 U32 accepted_word= 0;
2916 while ( state && uc <= (U8*)strend ) {
2918 U32 word = aho->states[ state ].wordnum;
2922 DEBUG_TRIE_EXECUTE_r(
2923 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2924 dump_exec_pos( (char *)uc, c, strend, real_start,
2925 (char *)uc, utf8_target, 0 );
2926 Perl_re_printf( aTHX_
2927 " Scanning for legal start char...\n");
2931 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2935 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2941 if (uc >(U8*)last_start) break;
2945 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2946 if (!leftmost || lpos < leftmost) {
2947 DEBUG_r(accepted_word=word);
2953 points[pointpos++ % maxlen]= uc;
2954 if (foldlen || uc < (U8*)strend) {
2955 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
2956 (U8 *) strend, uscan, len, uvc,
2957 charid, foldlen, foldbuf,
2959 DEBUG_TRIE_EXECUTE_r({
2960 dump_exec_pos( (char *)uc, c, strend,
2961 real_start, s, utf8_target, 0);
2962 Perl_re_printf( aTHX_
2963 " Charid:%3u CP:%4" UVxf " ",
2975 word = aho->states[ state ].wordnum;
2977 base = aho->states[ state ].trans.base;
2979 DEBUG_TRIE_EXECUTE_r({
2981 dump_exec_pos( (char *)uc, c, strend, real_start,
2982 s, utf8_target, 0 );
2983 Perl_re_printf( aTHX_
2984 "%sState: %4" UVxf ", word=%" UVxf,
2985 failed ? " Fail transition to " : "",
2986 (UV)state, (UV)word);
2992 ( ((offset = base + charid
2993 - 1 - trie->uniquecharcount)) >= 0)
2994 && ((U32)offset < trie->lasttrans)
2995 && trie->trans[offset].check == state
2996 && (tmp=trie->trans[offset].next))
2998 DEBUG_TRIE_EXECUTE_r(
2999 Perl_re_printf( aTHX_ " - legal\n"));
3004 DEBUG_TRIE_EXECUTE_r(
3005 Perl_re_printf( aTHX_ " - fail\n"));
3007 state = aho->fail[state];
3011 /* we must be accepting here */
3012 DEBUG_TRIE_EXECUTE_r(
3013 Perl_re_printf( aTHX_ " - accepting\n"));
3022 if (!state) state = 1;
3025 if ( aho->states[ state ].wordnum ) {
3026 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
3027 if (!leftmost || lpos < leftmost) {
3028 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
3033 s = (char*)leftmost;
3034 DEBUG_TRIE_EXECUTE_r({
3035 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
3036 (UV)accepted_word, (IV)(s - real_start)
3039 if (reginfo->intuit || regtry(reginfo, &s)) {
3044 if (s < reginfo->strend) {
3047 DEBUG_TRIE_EXECUTE_r({
3048 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
3051 DEBUG_TRIE_EXECUTE_r(
3052 Perl_re_printf( aTHX_ "No match.\n"));
3061 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
3068 /* set RX_SAVED_COPY, RX_SUBBEG etc.
3069 * flags have same meanings as with regexec_flags() */
3072 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
3079 struct regexp *const prog = ReANY(rx);
3081 if (flags & REXEC_COPY_STR) {
3084 DEBUG_C(Perl_re_printf( aTHX_
3085 "Copy on write: regexp capture, type %d\n",
3087 /* Create a new COW SV to share the match string and store
3088 * in saved_copy, unless the current COW SV in saved_copy
3089 * is valid and suitable for our purpose */
3090 if (( prog->saved_copy
3091 && SvIsCOW(prog->saved_copy)
3092 && SvPOKp(prog->saved_copy)
3095 && SvPVX(sv) == SvPVX(prog->saved_copy)))
3097 /* just reuse saved_copy SV */
3098 if (RXp_MATCH_COPIED(prog)) {
3099 Safefree(prog->subbeg);
3100 RXp_MATCH_COPIED_off(prog);
3104 /* create new COW SV to share string */
3105 RXp_MATCH_COPY_FREE(prog);
3106 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
3108 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
3109 assert (SvPOKp(prog->saved_copy));
3110 prog->sublen = strend - strbeg;
3111 prog->suboffset = 0;
3112 prog->subcoffset = 0;
3117 SSize_t max = strend - strbeg;
3120 if ( (flags & REXEC_COPY_SKIP_POST)
3121 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
3122 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
3123 ) { /* don't copy $' part of string */
3126 /* calculate the right-most part of the string covered
3127 * by a capture. Due to lookahead, this may be to
3128 * the right of $&, so we have to scan all captures */
3129 while (n <= prog->lastparen) {
3130 if (prog->offs[n].end > max)
3131 max = prog->offs[n].end;
3135 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
3136 ? prog->offs[0].start
3138 assert(max >= 0 && max <= strend - strbeg);
3141 if ( (flags & REXEC_COPY_SKIP_PRE)
3142 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
3143 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
3144 ) { /* don't copy $` part of string */
3147 /* calculate the left-most part of the string covered
3148 * by a capture. Due to lookbehind, this may be to
3149 * the left of $&, so we have to scan all captures */
3150 while (min && n <= prog->lastparen) {
3151 if ( prog->offs[n].start != -1
3152 && prog->offs[n].start < min)
3154 min = prog->offs[n].start;
3158 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
3159 && min > prog->offs[0].end
3161 min = prog->offs[0].end;
3165 assert(min >= 0 && min <= max && min <= strend - strbeg);
3168 if (RXp_MATCH_COPIED(prog)) {
3169 if (sublen > prog->sublen)
3171 (char*)saferealloc(prog->subbeg, sublen+1);
3174 prog->subbeg = (char*)safemalloc(sublen+1);
3175 Copy(strbeg + min, prog->subbeg, sublen, char);
3176 prog->subbeg[sublen] = '\0';
3177 prog->suboffset = min;
3178 prog->sublen = sublen;
3179 RXp_MATCH_COPIED_on(prog);
3181 prog->subcoffset = prog->suboffset;
3182 if (prog->suboffset && utf8_target) {
3183 /* Convert byte offset to chars.
3184 * XXX ideally should only compute this if @-/@+
3185 * has been seen, a la PL_sawampersand ??? */
3187 /* If there's a direct correspondence between the
3188 * string which we're matching and the original SV,
3189 * then we can use the utf8 len cache associated with
3190 * the SV. In particular, it means that under //g,
3191 * sv_pos_b2u() will use the previously cached
3192 * position to speed up working out the new length of
3193 * subcoffset, rather than counting from the start of
3194 * the string each time. This stops
3195 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
3196 * from going quadratic */
3197 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
3198 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
3199 SV_GMAGIC|SV_CONST_RETURN);
3201 prog->subcoffset = utf8_length((U8*)strbeg,
3202 (U8*)(strbeg+prog->suboffset));
3206 RXp_MATCH_COPY_FREE(prog);
3207 prog->subbeg = strbeg;
3208 prog->suboffset = 0;
3209 prog->subcoffset = 0;
3210 prog->sublen = strend - strbeg;
3218 - regexec_flags - match a regexp against a string
3221 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
3222 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
3223 /* stringarg: the point in the string at which to begin matching */
3224 /* strend: pointer to null at end of string */
3225 /* strbeg: real beginning of string */
3226 /* minend: end of match must be >= minend bytes after stringarg. */
3227 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
3228 * itself is accessed via the pointers above */
3229 /* data: May be used for some additional optimizations.
3230 Currently unused. */
3231 /* flags: For optimizations. See REXEC_* in regexp.h */
3234 struct regexp *const prog = ReANY(rx);
3238 SSize_t minlen; /* must match at least this many chars */
3239 SSize_t dontbother = 0; /* how many characters not to try at end */
3240 const bool utf8_target = cBOOL(DO_UTF8(sv));
3242 RXi_GET_DECL(prog,progi);
3243 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
3244 regmatch_info *const reginfo = ®info_buf;
3245 regexp_paren_pair *swap = NULL;
3247 GET_RE_DEBUG_FLAGS_DECL;
3249 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
3250 PERL_UNUSED_ARG(data);
3252 /* Be paranoid... */
3254 Perl_croak(aTHX_ "NULL regexp parameter");
3258 debug_start_match(rx, utf8_target, stringarg, strend,
3262 startpos = stringarg;
3264 /* set these early as they may be used by the HOP macros below */
3265 reginfo->strbeg = strbeg;
3266 reginfo->strend = strend;
3267 reginfo->is_utf8_target = cBOOL(utf8_target);
3269 if (prog->intflags & PREGf_GPOS_SEEN) {
3272 /* set reginfo->ganch, the position where \G can match */
3275 (flags & REXEC_IGNOREPOS)
3276 ? stringarg /* use start pos rather than pos() */
3277 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
3278 /* Defined pos(): */
3279 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
3280 : strbeg; /* pos() not defined; use start of string */
3282 DEBUG_GPOS_r(Perl_re_printf( aTHX_
3283 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
3285 /* in the presence of \G, we may need to start looking earlier in
3286 * the string than the suggested start point of stringarg:
3287 * if prog->gofs is set, then that's a known, fixed minimum
3290 * /ab|c\G/: gofs = 1
3291 * or if the minimum offset isn't known, then we have to go back
3292 * to the start of the string, e.g. /w+\G/
3295 if (prog->intflags & PREGf_ANCH_GPOS) {
3297 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
3299 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
3301 DEBUG_r(Perl_re_printf( aTHX_
3302 "fail: ganch-gofs before earliest possible start\n"));
3307 startpos = reginfo->ganch;
3309 else if (prog->gofs) {
3310 startpos = HOPBACKc(startpos, prog->gofs);
3314 else if (prog->intflags & PREGf_GPOS_FLOAT)
3318 minlen = prog->minlen;
3319 if ((startpos + minlen) > strend || startpos < strbeg) {
3320 DEBUG_r(Perl_re_printf( aTHX_
3321 "Regex match can't succeed, so not even tried\n"));
3325 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
3326 * which will call destuctors to reset PL_regmatch_state, free higher
3327 * PL_regmatch_slabs, and clean up regmatch_info_aux and
3328 * regmatch_info_aux_eval */
3330 oldsave = PL_savestack_ix;
3334 if ((prog->extflags & RXf_USE_INTUIT)
3335 && !(flags & REXEC_CHECKED))
3337 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
3342 if (prog->extflags & RXf_CHECK_ALL) {
3343 /* we can match based purely on the result of INTUIT.
3344 * Set up captures etc just for $& and $-[0]
3345 * (an intuit-only match wont have $1,$2,..) */
3346 assert(!prog->nparens);
3348 /* s/// doesn't like it if $& is earlier than where we asked it to
3349 * start searching (which can happen on something like /.\G/) */
3350 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3353 /* this should only be possible under \G */
3354 assert(prog->intflags & PREGf_GPOS_SEEN);
3355 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3356 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3360 /* match via INTUIT shouldn't have any captures.
3361 * Let @-, @+, $^N know */
3362 prog->lastparen = prog->lastcloseparen = 0;
3363 RXp_MATCH_UTF8_set(prog, utf8_target);
3364 prog->offs[0].start = s - strbeg;
3365 prog->offs[0].end = utf8_target
3366 ? (char*)utf8_hop_forward((U8*)s, prog->minlenret, (U8 *) strend) - strbeg
3367 : s - strbeg + prog->minlenret;
3368 if ( !(flags & REXEC_NOT_FIRST) )
3369 S_reg_set_capture_string(aTHX_ rx,
3371 sv, flags, utf8_target);
3377 multiline = prog->extflags & RXf_PMf_MULTILINE;
3379 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
3380 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3381 "String too short [regexec_flags]...\n"));
3385 /* Check validity of program. */
3386 if (UCHARAT(progi->program) != REG_MAGIC) {
3387 Perl_croak(aTHX_ "corrupted regexp program");
3390 RXp_MATCH_TAINTED_off(prog);
3391 RXp_MATCH_UTF8_set(prog, utf8_target);
3393 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3394 reginfo->intuit = 0;
3395 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3396 reginfo->warned = FALSE;
3398 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
3399 /* see how far we have to get to not match where we matched before */
3400 reginfo->till = stringarg + minend;
3402 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
3403 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3404 S_cleanup_regmatch_info_aux has executed (registered by
3405 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3406 magic belonging to this SV.
3407 Not newSVsv, either, as it does not COW.
3409 reginfo->sv = newSV(0);
3410 SvSetSV_nosteal(reginfo->sv, sv);
3411 SAVEFREESV(reginfo->sv);
3414 /* reserve next 2 or 3 slots in PL_regmatch_state:
3415 * slot N+0: may currently be in use: skip it
3416 * slot N+1: use for regmatch_info_aux struct
3417 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3418 * slot N+3: ready for use by regmatch()
3422 regmatch_state *old_regmatch_state;
3423 regmatch_slab *old_regmatch_slab;
3424 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3426 /* on first ever match, allocate first slab */
3427 if (!PL_regmatch_slab) {
3428 Newx(PL_regmatch_slab, 1, regmatch_slab);
3429 PL_regmatch_slab->prev = NULL;
3430 PL_regmatch_slab->next = NULL;
3431 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3434 old_regmatch_state = PL_regmatch_state;
3435 old_regmatch_slab = PL_regmatch_slab;
3437 for (i=0; i <= max; i++) {
3439 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3441 reginfo->info_aux_eval =
3442 reginfo->info_aux->info_aux_eval =
3443 &(PL_regmatch_state->u.info_aux_eval);
3445 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3446 PL_regmatch_state = S_push_slab(aTHX);
3449 /* note initial PL_regmatch_state position; at end of match we'll
3450 * pop back to there and free any higher slabs */
3452 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3453 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
3454 reginfo->info_aux->poscache = NULL;
3456 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
3458 if ((prog->extflags & RXf_EVAL_SEEN))
3459 S_setup_eval_state(aTHX_ reginfo);
3461 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
3464 /* If there is a "must appear" string, look for it. */
3466 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
3467 /* We have to be careful. If the previous successful match
3468 was from this regex we don't want a subsequent partially
3469 successful match to clobber the old results.
3470 So when we detect this possibility we add a swap buffer
3471 to the re, and switch the buffer each match. If we fail,
3472 we switch it back; otherwise we leave it swapped.
3475 /* avoid leak if we die, or clean up anyway if match completes */
3477 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
3478 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3479 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
3487 if (prog->recurse_locinput)
3488 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3490 /* Simplest case: anchored match need be tried only once, or with
3491 * MBOL, only at the beginning of each line.
3493 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3494 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3495 * match at the start of the string then it won't match anywhere else
3496 * either; while with /.*.../, if it doesn't match at the beginning,
3497 * the earliest it could match is at the start of the next line */
3499 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3502 if (regtry(reginfo, &s))
3505 if (!(prog->intflags & PREGf_ANCH_MBOL))
3508 /* didn't match at start, try at other newline positions */
3511 dontbother = minlen - 1;
3512 end = HOP3c(strend, -dontbother, strbeg) - 1;
3514 /* skip to next newline */
3516 while (s <= end) { /* note it could be possible to match at the end of the string */
3517 /* NB: newlines are the same in unicode as they are in latin */
3520 if (prog->check_substr || prog->check_utf8) {
3521 /* note that with PREGf_IMPLICIT, intuit can only fail
3522 * or return the start position, so it's of limited utility.
3523 * Nevertheless, I made the decision that the potential for
3524 * quick fail was still worth it - DAPM */
3525 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3529 if (regtry(reginfo, &s))
3533 } /* end anchored search */
3535 if (prog->intflags & PREGf_ANCH_GPOS)
3537 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3538 assert(prog->intflags & PREGf_GPOS_SEEN);
3539 /* For anchored \G, the only position it can match from is
3540 * (ganch-gofs); we already set startpos to this above; if intuit
3541 * moved us on from there, we can't possibly succeed */
3542 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
3543 if (s == startpos && regtry(reginfo, &s))
3548 /* Messy cases: unanchored match. */
3549 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
3550 /* we have /x+whatever/ */
3551 /* it must be a one character string (XXXX Except is_utf8_pat?) */
3557 if (! prog->anchored_utf8) {
3558 to_utf8_substr(prog);
3560 ch = SvPVX_const(prog->anchored_utf8)[0];
3561 REXEC_FBC_SCAN(0, /* 0=>not-utf8 */
3563 DEBUG_EXECUTE_r( did_match = 1 );
3564 if (regtry(reginfo, &s)) goto got_it;
3565 s += UTF8_SAFE_SKIP(s, strend);
3566 while (s < strend && *s == ch)
3573 if (! prog->anchored_substr) {
3574 if (! to_byte_substr(prog)) {
3575 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3578 ch = SvPVX_const(prog->anchored_substr)[0];
3579 REXEC_FBC_SCAN(0, /* 0=>not-utf8 */
3581 DEBUG_EXECUTE_r( did_match = 1 );
3582 if (regtry(reginfo, &s)) goto got_it;
3584 while (s < strend && *s == ch)
3589 DEBUG_EXECUTE_r(if (!did_match)
3590 Perl_re_printf( aTHX_
3591 "Did not find anchored character...\n")
3594 else if (prog->anchored_substr != NULL
3595 || prog->anchored_utf8 != NULL
3596 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
3597 && prog->float_max_offset < strend - s)) {
3602 char *last1; /* Last position checked before */
3606 if (prog->anchored_substr || prog->anchored_utf8) {
3608 if (! prog->anchored_utf8) {
3609 to_utf8_substr(prog);
3611 must = prog->anchored_utf8;
3614 if (! prog->anchored_substr) {
3615 if (! to_byte_substr(prog)) {
3616 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3619 must = prog->anchored_substr;
3621 back_max = back_min = prog->anchored_offset;
3624 if (! prog->float_utf8) {
3625 to_utf8_substr(prog);
3627 must = prog->float_utf8;
3630 if (! prog->float_substr) {
3631 if (! to_byte_substr(prog)) {
3632 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3635 must = prog->float_substr;
3637 back_max = prog->float_max_offset;
3638 back_min = prog->float_min_offset;
3644 last = HOP3c(strend, /* Cannot start after this */
3645 -(SSize_t)(CHR_SVLEN(must)
3646 - (SvTAIL(must) != 0) + back_min), strbeg);
3648 if (s > reginfo->strbeg)
3649 last1 = HOPc(s, -1);
3651 last1 = s - 1; /* bogus */
3653 /* XXXX check_substr already used to find "s", can optimize if
3654 check_substr==must. */
3656 strend = HOPc(strend, -dontbother);
3657 while ( (s <= last) &&
3658 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
3659 (unsigned char*)strend, must,
3660 multiline ? FBMrf_MULTILINE : 0)) ) {
3661 DEBUG_EXECUTE_r( did_match = 1 );
3662 if (HOPc(s, -back_max) > last1) {
3663 last1 = HOPc(s, -back_min);
3664 s = HOPc(s, -back_max);
3667 char * const t = (last1 >= reginfo->strbeg)
3668 ? HOPc(last1, 1) : last1 + 1;
3670 last1 = HOPc(s, -back_min);
3674 while (s <= last1) {
3675 if (regtry(reginfo, &s))
3678 s++; /* to break out of outer loop */
3685 while (s <= last1) {
3686 if (regtry(reginfo, &s))
3692 DEBUG_EXECUTE_r(if (!did_match) {
3693 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
3694 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
3695 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n",
3696 ((must == prog->anchored_substr || must == prog->anchored_utf8)
3697 ? "anchored" : "floating"),
3698 quoted, RE_SV_TAIL(must));
3702 else if ( (c = progi->regstclass) ) {
3704 const OPCODE op = OP(progi->regstclass);
3705 /* don't bother with what can't match */
3706 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
3707 strend = HOPc(strend, -(minlen - 1));
3710 SV * const prop = sv_newmortal();
3711 regprop(prog, prop, c, reginfo, NULL);
3713 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
3714 s,strend-s,PL_dump_re_max_len);
3715 Perl_re_printf( aTHX_
3716 "Matching stclass %.*s against %s (%d bytes)\n",
3717 (int)SvCUR(prop), SvPVX_const(prop),
3718 quoted, (int)(strend - s));
3721 if (find_byclass(prog, c, s, strend, reginfo))
3723 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n"));
3727 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3735 if (! prog->float_utf8) {
3736 to_utf8_substr(prog);
3738 float_real = prog->float_utf8;
3741 if (! prog->float_substr) {
3742 if (! to_byte_substr(prog)) {
3743 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3746 float_real = prog->float_substr;
3749 little = SvPV_const(float_real, len);
3750 if (SvTAIL(float_real)) {
3751 /* This means that float_real contains an artificial \n on
3752 * the end due to the presence of something like this:
3753 * /foo$/ where we can match both "foo" and "foo\n" at the
3754 * end of the string. So we have to compare the end of the
3755 * string first against the float_real without the \n and
3756 * then against the full float_real with the string. We
3757 * have to watch out for cases where the string might be
3758 * smaller than the float_real or the float_real without
3760 char *checkpos= strend - len;
3762 Perl_re_printf( aTHX_
3763 "%sChecking for float_real.%s\n",
3764 PL_colors[4], PL_colors[5]));
3765 if (checkpos + 1 < strbeg) {
3766 /* can't match, even if we remove the trailing \n
3767 * string is too short to match */
3769 Perl_re_printf( aTHX_
3770 "%sString shorter than required trailing substring, cannot match.%s\n",
3771 PL_colors[4], PL_colors[5]));
3773 } else if (memEQ(checkpos + 1, little, len - 1)) {
3774 /* can match, the end of the string matches without the
3776 last = checkpos + 1;
3777 } else if (checkpos < strbeg) {
3778 /* cant match, string is too short when the "\n" is
3781 Perl_re_printf( aTHX_
3782 "%sString does not contain required trailing substring, cannot match.%s\n",
3783 PL_colors[4], PL_colors[5]));
3785 } else if (!multiline) {
3786 /* non multiline match, so compare with the "\n" at the
3787 * end of the string */
3788 if (memEQ(checkpos, little, len)) {
3792 Perl_re_printf( aTHX_
3793 "%sString does not contain required trailing substring, cannot match.%s\n",
3794 PL_colors[4], PL_colors[5]));
3798 /* multiline match, so we have to search for a place
3799 * where the full string is located */
3805 last = rninstr(s, strend, little, little + len);
3807 last = strend; /* matching "$" */
3810 /* at one point this block contained a comment which was
3811 * probably incorrect, which said that this was a "should not
3812 * happen" case. Even if it was true when it was written I am
3813 * pretty sure it is not anymore, so I have removed the comment
3814 * and replaced it with this one. Yves */
3816 Perl_re_printf( aTHX_
3817 "%sString does not contain required substring, cannot match.%s\n",
3818 PL_colors[4], PL_colors[5]
3822 dontbother = strend - last + prog->float_min_offset;
3824 if (minlen && (dontbother < minlen))
3825 dontbother = minlen - 1;
3826 strend -= dontbother; /* this one's always in bytes! */
3827 /* We don't know much -- general case. */
3830 if (regtry(reginfo, &s))
3839 if (regtry(reginfo, &s))
3841 } while (s++ < strend);
3849 /* s/// doesn't like it if $& is earlier than where we asked it to
3850 * start searching (which can happen on something like /.\G/) */
3851 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3852 && (prog->offs[0].start < stringarg - strbeg))
3854 /* this should only be possible under \G */
3855 assert(prog->intflags & PREGf_GPOS_SEEN);
3856 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3857 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3861 /* clean up; this will trigger destructors that will free all slabs
3862 * above the current one, and cleanup the regmatch_info_aux
3863 * and regmatch_info_aux_eval sructs */
3865 LEAVE_SCOPE(oldsave);
3867 if (RXp_PAREN_NAMES(prog))
3868 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3870 /* make sure $`, $&, $', and $digit will work later */
3871 if ( !(flags & REXEC_NOT_FIRST) )
3872 S_reg_set_capture_string(aTHX_ rx,
3873 strbeg, reginfo->strend,
3874 sv, flags, utf8_target);
3879 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n",
3880 PL_colors[4], PL_colors[5]));
3883 /* we failed :-( roll it back.
3884 * Since the swap buffer will be freed on scope exit which follows
3885 * shortly, restore the old captures by copying 'swap's original
3886 * data to the new offs buffer
3888 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3889 "rex=0x%" UVxf " rolling back offs: 0x%" UVxf " will be freed; restoring data to =0x%" UVxf "\n",
3896 Copy(swap, prog->offs, prog->nparens + 1, regexp_paren_pair);
3899 /* clean up; this will trigger destructors that will free all slabs
3900 * above the current one, and cleanup the regmatch_info_aux
3901 * and regmatch_info_aux_eval sructs */
3903 LEAVE_SCOPE(oldsave);
3909 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3910 * Do inc before dec, in case old and new rex are the same */
3911 #define SET_reg_curpm(Re2) \
3912 if (reginfo->info_aux_eval) { \
3913 (void)ReREFCNT_inc(Re2); \
3914 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3915 PM_SETRE((PL_reg_curpm), (Re2)); \
3920 - regtry - try match at specific point
3922 STATIC bool /* 0 failure, 1 success */
3923 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3926 REGEXP *const rx = reginfo->prog;
3927 regexp *const prog = ReANY(rx);
3930 U32 depth = 0; /* used by REGCP_SET */
3932 RXi_GET_DECL(prog,progi);
3933 GET_RE_DEBUG_FLAGS_DECL;
3935 PERL_ARGS_ASSERT_REGTRY;
3937 reginfo->cutpoint=NULL;
3939 prog->offs[0].start = *startposp - reginfo->strbeg;
3940 prog->lastparen = 0;
3941 prog->lastcloseparen = 0;
3943 /* XXXX What this code is doing here?!!! There should be no need
3944 to do this again and again, prog->lastparen should take care of
3947 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3948 * Actually, the code in regcppop() (which Ilya may be meaning by
3949 * prog->lastparen), is not needed at all by the test suite
3950 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3951 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3952 * Meanwhile, this code *is* needed for the
3953 * above-mentioned test suite tests to succeed. The common theme
3954 * on those tests seems to be returning null fields from matches.
3955 * --jhi updated by dapm */
3957 /* After encountering a variant of the issue mentioned above I think
3958 * the point Ilya was making is that if we properly unwind whenever
3959 * we set lastparen to a smaller value then we should not need to do
3960 * this every time, only when needed. So if we have tests that fail if
3961 * we remove this, then it suggests somewhere else we are improperly
3962 * unwinding the lastparen/paren buffers. See UNWIND_PARENS() and
3963 * places it is called, and related regcp() routines. - Yves */
3965 if (prog->nparens) {
3966 regexp_paren_pair *pp = prog->offs;
3968 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3976 result = regmatch(reginfo, *startposp, progi->program + 1);
3978 prog->offs[0].end = result;
3981 if (reginfo->cutpoint)
3982 *startposp= reginfo->cutpoint;
3983 REGCP_UNWIND(lastcp);
3987 /* this is used to determine how far from the left messages like
3988 'failed...' are printed in regexec.c. It should be set such that
3989 messages are inline with the regop output that created them.
3991 #define REPORT_CODE_OFF 29
3992 #define INDENT_CHARS(depth) ((int)(depth) % 20)
3995 Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...)
3999 PerlIO *f= Perl_debug_log;
4000 PERL_ARGS_ASSERT_RE_EXEC_INDENTF;
4001 va_start(ap, depth);
4002 PerlIO_printf(f, "%*s|%4" UVuf "| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" );
4003 result = PerlIO_vprintf(f, fmt, ap);
4007 #endif /* DEBUGGING */
4009 /* grab a new slab and return the first slot in it */
4011 STATIC regmatch_state *
4014 regmatch_slab *s = PL_regmatch_slab->next;
4016 Newx(s, 1, regmatch_slab);
4017 s->prev = PL_regmatch_slab;
4019 PL_regmatch_slab->next = s;
4021 PL_regmatch_slab = s;
4022 return SLAB_FIRST(s);
4028 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
4029 const char *start, const char *end, const char *blurb)
4031 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
4033 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
4038 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
4039 RX_PRECOMP_const(prog), RX_PRELEN(prog), PL_dump_re_max_len);
4041 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
4042 start, end - start, PL_dump_re_max_len);
4044 Perl_re_printf( aTHX_
4045 "%s%s REx%s %s against %s\n",
4046 PL_colors[4], blurb, PL_colors[5], s0, s1);
4048 if (utf8_target||utf8_pat)
4049 Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n",
4050 utf8_pat ? "pattern" : "",
4051 utf8_pat && utf8_target ? " and " : "",
4052 utf8_target ? "string" : ""
4058 S_dump_exec_pos(pTHX_ const char *locinput,
4059 const regnode *scan,
4060 const char *loc_regeol,
4061 const char *loc_bostr,
4062 const char *loc_reg_starttry,
4063 const bool utf8_target,
4067 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
4068 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
4069 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
4070 /* The part of the string before starttry has one color
4071 (pref0_len chars), between starttry and current
4072 position another one (pref_len - pref0_len chars),
4073 after the current position the third one.
4074 We assume that pref0_len <= pref_len, otherwise we
4075 decrease pref0_len. */
4076 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
4077 ? (5 + taill) - l : locinput - loc_bostr;
4080 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
4082 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
4084 pref0_len = pref_len - (locinput - loc_reg_starttry);
4085 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
4086 l = ( loc_regeol - locinput > (5 + taill) - pref_len
4087 ? (5 + taill) - pref_len : loc_regeol - locinput);
4088 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
4092 if (pref0_len > pref_len)
4093 pref0_len = pref_len;
4095 const int is_uni = utf8_target ? 1 : 0;
4097 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
4098 (locinput - pref_len),pref0_len, PL_dump_re_max_len, 4, 5);
4100 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
4101 (locinput - pref_len + pref0_len),
4102 pref_len - pref0_len, PL_dump_re_max_len, 2, 3);
4104 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
4105 locinput, loc_regeol - locinput, 10, 0, 1);
4107 const STRLEN tlen=len0+len1+len2;
4108 Perl_re_printf( aTHX_
4109 "%4" IVdf " <%.*s%.*s%s%.*s>%*s|%4u| ",
4110 (IV)(locinput - loc_bostr),
4113 (docolor ? "" : "> <"),
4115 (int)(tlen > 19 ? 0 : 19 - tlen),
4123 /* reg_check_named_buff_matched()
4124 * Checks to see if a named buffer has matched. The data array of
4125 * buffer numbers corresponding to the buffer is expected to reside
4126 * in the regexp->data->data array in the slot stored in the ARG() of
4127 * node involved. Note that this routine doesn't actually care about the
4128 * name, that information is not preserved from compilation to execution.
4129 * Returns the index of the leftmost defined buffer with the given name
4130 * or 0 if non of the buffers matched.
4133 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
4136 RXi_GET_DECL(rex,rexi);
4137 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4138 I32 *nums=(I32*)SvPVX(sv_dat);
4140 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
4142 for ( n=0; n<SvIVX(sv_dat); n++ ) {
4143 if ((I32)rex->lastparen >= nums[n] &&
4144 rex->offs[nums[n]].end != -1)
4152 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
4153 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
4154 #define CHRTEST_NOT_A_CP_1 -999
4155 #define CHRTEST_NOT_A_CP_2 -998
4158 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
4159 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
4161 /* This function determines if there are zero, one, two, or more characters
4162 * that match the first character of the passed-in EXACTish node
4163 * <text_node>, and if there are one or two, it returns them in the
4164 * passed-in pointers.
4166 * If it determines that no possible character in the target string can
4167 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
4168 * the first character in <text_node> requires UTF-8 to represent, and the
4169 * target string isn't in UTF-8.)
4171 * If there are more than two characters that could match the beginning of
4172 * <text_node>, or if more context is required to determine a match or not,
4173 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
4175 * The motiviation behind this function is to allow the caller to set up
4176 * tight loops for matching. If <text_node> is of type EXACT, there is
4177 * only one possible character that can match its first character, and so
4178 * the situation is quite simple. But things get much more complicated if
4179 * folding is involved. It may be that the first character of an EXACTFish
4180 * node doesn't participate in any possible fold, e.g., punctuation, so it
4181 * can be matched only by itself. The vast majority of characters that are
4182 * in folds match just two things, their lower and upper-case equivalents.
4183 * But not all are like that; some have multiple possible matches, or match
4184 * sequences of more than one character. This function sorts all that out.
4186 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
4187 * loop of trying to match A*, we know we can't exit where the thing
4188 * following it isn't a B. And something can't be a B unless it is the
4189 * beginning of B. By putting a quick test for that beginning in a tight
4190 * loop, we can rule out things that can't possibly be B without having to
4191 * break out of the loop, thus avoiding work. Similarly, if A is a single
4192 * character, we can make a tight loop matching A*, using the outputs of
4195 * If the target string to match isn't in UTF-8, and there aren't
4196 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
4197 * the one or two possible octets (which are characters in this situation)
4198 * that can match. In all cases, if there is only one character that can
4199 * match, *<c1p> and *<c2p> will be identical.
4201 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
4202 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
4203 * can match the beginning of <text_node>. They should be declared with at
4204 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
4205 * undefined what these contain.) If one or both of the buffers are
4206 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
4207 * corresponding invariant. If variant, the corresponding *<c1p> and/or
4208 * *<c2p> will be set to a negative number(s) that shouldn't match any code
4209 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
4210 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
4212 const bool utf8_target = reginfo->is_utf8_target;
4214 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
4215 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
4216 bool use_chrtest_void = FALSE;
4217 const bool is_utf8_pat = reginfo->is_utf8_pat;
4219 /* Used when we have both utf8 input and utf8 output, to avoid converting
4220 * to/from code points */
4221 bool utf8_has_been_setup = FALSE;
4225 U8 *pat = (U8*)STRING(text_node);
4226 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
4228 if ( OP(text_node) == EXACT
4229 || OP(text_node) == EXACT_ONLY8
4230 || OP(text_node) == EXACTL)
4233 /* In an exact node, only one thing can be matched, that first
4234 * character. If both the pat and the target are UTF-8, we can just
4235 * copy the input to the output, avoiding finding the code point of
4238 assert(OP(text_node) != EXACT_ONLY8);
4241 else if (utf8_target) {
4242 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
4243 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
4244 utf8_has_been_setup = TRUE;
4246 else if (OP(text_node) == EXACT_ONLY8) {
4247 return FALSE; /* Can only match UTF-8 target */
4250 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
4253 else { /* an EXACTFish node */
4254 U8 *pat_end = pat + STR_LEN(text_node);
4256 /* An EXACTFL node has at least some characters unfolded, because what
4257 * they match is not known until now. So, now is the time to fold
4258 * the first few of them, as many as are needed to determine 'c1' and
4259 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
4260 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
4261 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
4262 * need to fold as many characters as a single character can fold to,
4263 * so that later we can check if the first ones are such a multi-char
4264 * fold. But, in such a pattern only locale-problematic characters
4265 * aren't folded, so we can skip this completely if the first character
4266 * in the node isn't one of the tricky ones */
4267 if (OP(text_node) == EXACTFL) {
4269 if (! is_utf8_pat) {
4270 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
4272 folded[0] = folded[1] = 's';
4274 pat_end = folded + 2;
4277 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
4282 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
4283 if (isASCII(*s) && LIKELY(! PL_in_utf8_turkic_locale)) {
4284 *(d++) = (U8) toFOLD_LC(*s);
4289 _toFOLD_utf8_flags(s,
4293 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
4304 if ( ( is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
4305 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
4307 /* Multi-character folds require more context to sort out. Also
4308 * PL_utf8_foldclosures used below doesn't handle them, so have to
4309 * be handled outside this routine */
4310 use_chrtest_void = TRUE;
4312 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
4313 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
4315 if ( UNLIKELY(PL_in_utf8_turkic_locale)
4316 && OP(text_node) == EXACTFL
4317 && UNLIKELY( c1 == 'i' || c1 == 'I'
4318 || c1 == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE
4319 || c1 == LATIN_SMALL_LETTER_DOTLESS_I))
4320 { /* Hard-coded Turkish locale rules for these 4 characters
4321 override normal rules */
4323 c2 = LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
4325 else if (c1 == 'I') {
4326 c2 = LATIN_SMALL_LETTER_DOTLESS_I;
4328 else if (c1 == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4331 else if (c1 == LATIN_SMALL_LETTER_DOTLESS_I) {
4335 else if (c1 > 255) {
4336 const unsigned int * remaining_folds;
4337 unsigned int first_fold;
4339 /* Look up what code points (besides c1) fold to c1; e.g.,
4340 * [ 'K', KELVIN_SIGN ] both fold to 'k'. */
4341 Size_t folds_count = _inverse_folds(c1, &first_fold,
4343 if (folds_count == 0) {
4344 c2 = c1; /* there is only a single character that could
4347 else if (folds_count != 1) {
4348 /* If there aren't exactly two folds to this (itself and
4349 * another), it is outside the scope of this function */
4350 use_chrtest_void = TRUE;
4352 else { /* There are two. We already have one, get the other */
4355 /* Folds that cross the 255/256 boundary are forbidden if
4356 * EXACTFL (and isnt a UTF8 locale), or EXACTFAA and one is
4357 * ASCIII. The only other match to c1 is c2, and since c1
4358 * is above 255, c2 better be as well under these
4359 * circumstances. If it isn't, it means the only legal
4360 * match of c1 is itself. */
4362 && ( ( OP(text_node) == EXACTFL
4363 && ! IN_UTF8_CTYPE_LOCALE)
4364 || (( OP(text_node) == EXACTFAA
4365 || OP(text_node) == EXACTFAA_NO_TRIE)
4366 && (isASCII(c1) || isASCII(c2)))))
4372 else /* Here, c1 is <= 255 */
4374 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
4375 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
4376 && ( ( OP(text_node) != EXACTFAA
4377 && OP(text_node) != EXACTFAA_NO_TRIE)
4380 /* Here, there could be something above Latin1 in the target
4381 * which folds to this character in the pattern. All such
4382 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
4383 * than two characters involved in their folds, so are outside
4384 * the scope of this function */
4385 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4386 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
4389 use_chrtest_void = TRUE;
4392 else { /* Here nothing above Latin1 can fold to the pattern
4394 switch (OP(text_node)) {
4396 case EXACTFL: /* /l rules */
4397 c2 = PL_fold_locale[c1];
4400 case EXACTF: /* This node only generated for non-utf8
4402 assert(! is_utf8_pat);
4403 if (! utf8_target) { /* /d rules */
4408 /* /u rules for all these. This happens to work for
4409 * EXACTFAA as nothing in Latin1 folds to ASCII */
4410 case EXACTFAA_NO_TRIE: /* This node only generated for
4411 non-utf8 patterns */
4412 assert(! is_utf8_pat);
4417 c2 = PL_fold_latin1[c1];
4421 NOT_REACHED; /* NOTREACHED */
4424 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
4425 NOT_REACHED; /* NOTREACHED */
4431 /* Here have figured things out. Set up the returns */
4432 if (use_chrtest_void) {
4433 *c2p = *c1p = CHRTEST_VOID;
4435 else if (utf8_target) {
4436 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
4437 uvchr_to_utf8(c1_utf8, c1);
4438 uvchr_to_utf8(c2_utf8, c2);
4441 /* Invariants are stored in both the utf8 and byte outputs; Use
4442 * negative numbers otherwise for the byte ones. Make sure that the
4443 * byte ones are the same iff the utf8 ones are the same */
4444 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
4445 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
4448 ? CHRTEST_NOT_A_CP_1
4449 : CHRTEST_NOT_A_CP_2;
4451 else if (c1 > 255) {
4452 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
4457 *c1p = *c2p = c2; /* c2 is the only representable value */
4459 else { /* c1 is representable; see about c2 */
4461 *c2p = (c2 < 256) ? c2 : c1;
4468 S_isGCB(pTHX_ const GCB_enum before, const GCB_enum after, const U8 * const strbeg, const U8 * const curpos, const bool utf8_target)
4470 /* returns a boolean indicating if there is a Grapheme Cluster Boundary
4471 * between the inputs. See http://www.unicode.org/reports/tr29/. */
4473 PERL_ARGS_ASSERT_ISGCB;
4475 switch (GCB_table[before][after]) {
4482 case GCB_RI_then_RI:
4485 U8 * temp_pos = (U8 *) curpos;
4487 /* Do not break within emoji flag sequences. That is, do not
4488 * break between regional indicator (RI) symbols if there is an
4489 * odd number of RI characters before the break point.
4490 * GB12 sot (RI RI)* RI × RI
4491 * GB13 [^RI] (RI RI)* RI × RI */
4493 while (backup_one_GCB(strbeg,
4495 utf8_target) == GCB_Regional_Indicator)
4500 return RI_count % 2 != 1;
4503 case GCB_EX_then_EM:
4505 /* GB10 ( E_Base | E_Base_GAZ ) Extend* × E_Modifier */
4507 U8 * temp_pos = (U8 *) curpos;
4511 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4513 while (prev == GCB_Extend);
4515 return prev != GCB_E_Base && prev != GCB_E_Base_GAZ;
4518 case GCB_Maybe_Emoji_NonBreak:
4522 /* Do not break within emoji modifier sequences or emoji zwj sequences.
4523 GB11 \p{Extended_Pictographic} Extend* ZWJ × \p{Extended_Pictographic}
4525 U8 * temp_pos = (U8 *) curpos;
4529 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4531 while (prev == GCB_Extend);
4533 return prev != GCB_XPG_XX;
4541 Perl_re_printf( aTHX_ "Unhandled GCB pair: GCB_table[%d, %d] = %d\n",
4542 before, after, GCB_table[before][after]);
4549 S_backup_one_GCB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4554 PERL_ARGS_ASSERT_BACKUP_ONE_GCB;
4556 if (*curpos < strbeg) {
4561 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4562 U8 * prev_prev_char_pos;
4564 if (! prev_char_pos) {
4568 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4569 gcb = getGCB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4570 *curpos = prev_char_pos;
4571 prev_char_pos = prev_prev_char_pos;
4574 *curpos = (U8 *) strbeg;
4579 if (*curpos - 2 < strbeg) {
4580 *curpos = (U8 *) strbeg;
4584 gcb = getGCB_VAL_CP(*(*curpos - 1));
4590 /* Combining marks attach to most classes that precede them, but this defines
4591 * the exceptions (from TR14) */
4592 #define LB_CM_ATTACHES_TO(prev) ( ! ( prev == LB_EDGE \
4593 || prev == LB_Mandatory_Break \
4594 || prev == LB_Carriage_Return \
4595 || prev == LB_Line_Feed \
4596 || prev == LB_Next_Line \
4597 || prev == LB_Space \
4598 || prev == LB_ZWSpace))
4601 S_isLB(pTHX_ LB_enum before,
4603 const U8 * const strbeg,
4604 const U8 * const curpos,
4605 const U8 * const strend,
4606 const bool utf8_target)
4608 U8 * temp_pos = (U8 *) curpos;
4609 LB_enum prev = before;
4611 /* Is the boundary between 'before' and 'after' line-breakable?
4612 * Most of this is just a table lookup of a generated table from Unicode
4613 * rules. But some rules require context to decide, and so have to be
4614 * implemented in code */
4616 PERL_ARGS_ASSERT_ISLB;
4618 /* Rule numbers in the comments below are as of Unicode 9.0 */
4622 switch (LB_table[before][after]) {
4627 case LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4630 case LB_SP_foo + LB_BREAKABLE:
4631 case LB_SP_foo + LB_NOBREAK:
4632 case LB_SP_foo + LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4634 /* When we have something following a SP, we have to look at the
4635 * context in order to know what to do.
4637 * SP SP should not reach here because LB7: Do not break before
4638 * spaces. (For two spaces in a row there is nothing that
4639 * overrides that) */
4640 assert(after != LB_Space);
4642 /* Here we have a space followed by a non-space. Mostly this is a
4643 * case of LB18: "Break after spaces". But there are complications
4644 * as the handling of spaces is somewhat tricky. They are in a
4645 * number of rules, which have to be applied in priority order, but
4646 * something earlier in the string can cause a rule to be skipped
4647 * and a lower priority rule invoked. A prime example is LB7 which
4648 * says don't break before a space. But rule LB8 (lower priority)
4649 * says that the first break opportunity after a ZW is after any
4650 * span of spaces immediately after it. If a ZW comes before a SP
4651 * in the input, rule LB8 applies, and not LB7. Other such rules
4652 * involve combining marks which are rules 9 and 10, but they may
4653 * override higher priority rules if they come earlier in the
4654 * string. Since we're doing random access into the middle of the
4655 * string, we have to look for rules that should get applied based
4656 * on both string position and priority. Combining marks do not
4657 * attach to either ZW nor SP, so we don't have to consider them
4660 * To check for LB8, we have to find the first non-space character
4661 * before this span of spaces */
4663 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4665 while (prev == LB_Space);
4667 /* LB8 Break before any character following a zero-width space,
4668 * even if one or more spaces intervene.
4670 * So if we have a ZW just before this span, and to get here this
4671 * is the final space in the span. */
4672 if (prev == LB_ZWSpace) {
4676 /* Here, not ZW SP+. There are several rules that have higher
4677 * priority than LB18 and can be resolved now, as they don't depend
4678 * on anything earlier in the string (except ZW, which we have
4679 * already handled). One of these rules is LB11 Do not break
4680 * before Word joiner, but we have specially encoded that in the
4681 * lookup table so it is caught by the single test below which
4682 * catches the other ones. */
4683 if (LB_table[LB_Space][after] - LB_SP_foo
4684 == LB_NOBREAK_EVEN_WITH_SP_BETWEEN)
4689 /* If we get here, we have to XXX consider combining marks. */
4690 if (prev == LB_Combining_Mark) {
4692 /* What happens with these depends on the character they
4695 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4697 while (prev == LB_Combining_Mark);
4699 /* Most times these attach to and inherit the characteristics
4700 * of that character, but not always, and when not, they are to
4701 * be treated as AL by rule LB10. */
4702 if (! LB_CM_ATTACHES_TO(prev)) {
4703 prev = LB_Alphabetic;
4707 /* Here, we have the character preceding the span of spaces all set
4708 * up. We follow LB18: "Break after spaces" unless the table shows
4709 * that is overriden */
4710 return LB_table[prev][after] != LB_NOBREAK_EVEN_WITH_SP_BETWEEN;
4714 /* We don't know how to treat the CM except by looking at the first
4715 * non-CM character preceding it. ZWJ is treated as CM */
4717 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4719 while (prev == LB_Combining_Mark || prev == LB_ZWJ);
4721 /* Here, 'prev' is that first earlier non-CM character. If the CM
4722 * attatches to it, then it inherits the behavior of 'prev'. If it
4723 * doesn't attach, it is to be treated as an AL */
4724 if (! LB_CM_ATTACHES_TO(prev)) {
4725 prev = LB_Alphabetic;
4730 case LB_HY_or_BA_then_foo + LB_BREAKABLE:
4731 case LB_HY_or_BA_then_foo + LB_NOBREAK:
4733 /* LB21a Don't break after Hebrew + Hyphen.
4734 * HL (HY | BA) × */
4736 if (backup_one_LB(strbeg, &temp_pos, utf8_target)
4737 == LB_Hebrew_Letter)
4742 return LB_table[prev][after] - LB_HY_or_BA_then_foo == LB_BREAKABLE;
4744 case LB_PR_or_PO_then_OP_or_HY + LB_BREAKABLE:
4745 case LB_PR_or_PO_then_OP_or_HY + LB_NOBREAK:
4747 /* LB25a (PR | PO) × ( OP | HY )? NU */
4748 if (advance_one_LB(&temp_pos, strend, utf8_target) == LB_Numeric) {
4752 return LB_table[prev][after] - LB_PR_or_PO_then_OP_or_HY
4755 case LB_SY_or_IS_then_various + LB_BREAKABLE:
4756 case LB_SY_or_IS_then_various + LB_NOBREAK:
4758 /* LB25d NU (SY | IS)* × (NU | SY | IS | CL | CP ) */
4760 LB_enum temp = prev;
4762 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4764 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric);
4765 if (temp == LB_Numeric) {
4769 return LB_table[prev][after] - LB_SY_or_IS_then_various
4773 case LB_various_then_PO_or_PR + LB_BREAKABLE:
4774 case LB_various_then_PO_or_PR + LB_NOBREAK:
4776 /* LB25e NU (SY | IS)* (CL | CP)? × (PO | PR) */
4778 LB_enum temp = prev;
4779 if (temp == LB_Close_Punctuation || temp == LB_Close_Parenthesis)
4781 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4783 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric) {
4784 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4786 if (temp == LB_Numeric) {
4789 return LB_various_then_PO_or_PR;
4792 case LB_RI_then_RI + LB_NOBREAK:
4793 case LB_RI_then_RI + LB_BREAKABLE:
4797 /* LB30a Break between two regional indicator symbols if and
4798 * only if there are an even number of regional indicators
4799 * preceding the position of the break.
4801 * sot (RI RI)* RI × RI
4802 * [^RI] (RI RI)* RI × RI */
4804 while (backup_one_LB(strbeg,
4806 utf8_target) == LB_Regional_Indicator)
4811 return RI_count % 2 == 0;
4819 Perl_re_printf( aTHX_ "Unhandled LB pair: LB_table[%d, %d] = %d\n",
4820 before, after, LB_table[before][after]);
4827 S_advance_one_LB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4833 PERL_ARGS_ASSERT_ADVANCE_ONE_LB;
4835 if (*curpos >= strend) {
4840 *curpos += UTF8SKIP(*curpos);
4841 if (*curpos >= strend) {
4844 lb = getLB_VAL_UTF8(*curpos, strend);
4848 if (*curpos >= strend) {
4851 lb = getLB_VAL_CP(**curpos);
4858 S_backup_one_LB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4863 PERL_ARGS_ASSERT_BACKUP_ONE_LB;
4865 if (*curpos < strbeg) {
4870 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4871 U8 * prev_prev_char_pos;
4873 if (! prev_char_pos) {
4877 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4878 lb = getLB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4879 *curpos = prev_char_pos;
4880 prev_char_pos = prev_prev_char_pos;
4883 *curpos = (U8 *) strbeg;
4888 if (*curpos - 2 < strbeg) {
4889 *curpos = (U8 *) strbeg;
4893 lb = getLB_VAL_CP(*(*curpos - 1));
4900 S_isSB(pTHX_ SB_enum before,
4902 const U8 * const strbeg,
4903 const U8 * const curpos,
4904 const U8 * const strend,
4905 const bool utf8_target)
4907 /* returns a boolean indicating if there is a Sentence Boundary Break
4908 * between the inputs. See http://www.unicode.org/reports/tr29/ */
4910 U8 * lpos = (U8 *) curpos;
4911 bool has_para_sep = FALSE;
4912 bool has_sp = FALSE;
4914 PERL_ARGS_ASSERT_ISSB;
4916 /* Break at the start and end of text.
4919 But unstated in Unicode is don't break if the text is empty */
4920 if (before == SB_EDGE || after == SB_EDGE) {
4921 return before != after;
4924 /* SB 3: Do not break within CRLF. */
4925 if (before == SB_CR && after == SB_LF) {
4929 /* Break after paragraph separators. CR and LF are considered
4930 * so because Unicode views text as like word processing text where there
4931 * are no newlines except between paragraphs, and the word processor takes
4932 * care of wrapping without there being hard line-breaks in the text *./
4933 SB4. Sep | CR | LF ÷ */
4934 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4938 /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF.
4939 * (See Section 6.2, Replacing Ignore Rules.)
4940 SB5. X (Extend | Format)* → X */
4941 if (after == SB_Extend || after == SB_Format) {
4943 /* Implied is that the these characters attach to everything
4944 * immediately prior to them except for those separator-type
4945 * characters. And the rules earlier have already handled the case
4946 * when one of those immediately precedes the extend char */
4950 if (before == SB_Extend || before == SB_Format) {
4951 U8 * temp_pos = lpos;
4952 const SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4953 if ( backup != SB_EDGE
4962 /* Here, both 'before' and 'backup' are these types; implied is that we
4963 * don't break between them */
4964 if (backup == SB_Extend || backup == SB_Format) {
4969 /* Do not break after ambiguous terminators like period, if they are
4970 * immediately followed by a number or lowercase letter, if they are
4971 * between uppercase letters, if the first following letter (optionally
4972 * after certain punctuation) is lowercase, or if they are followed by
4973 * "continuation" punctuation such as comma, colon, or semicolon. For
4974 * example, a period may be an abbreviation or numeric period, and thus may
4975 * not mark the end of a sentence.
4977 * SB6. ATerm × Numeric */
4978 if (before == SB_ATerm && after == SB_Numeric) {
4982 /* SB7. (Upper | Lower) ATerm × Upper */
4983 if (before == SB_ATerm && after == SB_Upper) {
4984 U8 * temp_pos = lpos;
4985 SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4986 if (backup == SB_Upper || backup == SB_Lower) {
4991 /* The remaining rules that aren't the final one, all require an STerm or
4992 * an ATerm after having backed up over some Close* Sp*, and in one case an
4993 * optional Paragraph separator, although one rule doesn't have any Sp's in it.
4994 * So do that backup now, setting flags if either Sp or a paragraph
4995 * separator are found */
4997 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4998 has_para_sep = TRUE;
4999 before = backup_one_SB(strbeg, &lpos, utf8_target);
5002 if (before == SB_Sp) {
5005 before = backup_one_SB(strbeg, &lpos, utf8_target);
5007 while (before == SB_Sp);
5010 while (before == SB_Close) {
5011 before = backup_one_SB(strbeg, &lpos, utf8_target);
5014 /* The next few rules apply only when the backed-up-to is an ATerm, and in
5015 * most cases an STerm */
5016 if (before == SB_STerm || before == SB_ATerm) {
5018 /* So, here the lhs matches
5019 * (STerm | ATerm) Close* Sp* (Sep | CR | LF)?
5020 * and we have set flags if we found an Sp, or the optional Sep,CR,LF.
5021 * The rules that apply here are:
5023 * SB8 ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR
5024 | LF | STerm | ATerm) )* Lower
5025 SB8a (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm)
5026 SB9 (STerm | ATerm) Close* × (Close | Sp | Sep | CR | LF)
5027 SB10 (STerm | ATerm) Close* Sp* × (Sp | Sep | CR | LF)
5028 SB11 (STerm | ATerm) Close* Sp* (Sep | CR | LF)? ÷
5031 /* And all but SB11 forbid having seen a paragraph separator */
5032 if (! has_para_sep) {
5033 if (before == SB_ATerm) { /* SB8 */
5034 U8 * rpos = (U8 *) curpos;
5035 SB_enum later = after;
5037 while ( later != SB_OLetter
5038 && later != SB_Upper
5039 && later != SB_Lower
5043 && later != SB_STerm
5044 && later != SB_ATerm
5045 && later != SB_EDGE)
5047 later = advance_one_SB(&rpos, strend, utf8_target);
5049 if (later == SB_Lower) {
5054 if ( after == SB_SContinue /* SB8a */
5055 || after == SB_STerm
5056 || after == SB_ATerm)
5061 if (! has_sp) { /* SB9 applies only if there was no Sp* */
5062 if ( after == SB_Close
5072 /* SB10. This and SB9 could probably be combined some way, but khw
5073 * has decided to follow the Unicode rule book precisely for
5074 * simplified maintenance */
5088 /* Otherwise, do not break.
5095 S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
5100 PERL_ARGS_ASSERT_ADVANCE_ONE_SB;
5102 if (*curpos >= strend) {
5108 *curpos += UTF8SKIP(*curpos);
5109 if (*curpos >= strend) {
5112 sb = getSB_VAL_UTF8(*curpos, strend);
5113 } while (sb == SB_Extend || sb == SB_Format);
5118 if (*curpos >= strend) {
5121 sb = getSB_VAL_CP(**curpos);
5122 } while (sb == SB_Extend || sb == SB_Format);
5129 S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5134 PERL_ARGS_ASSERT_BACKUP_ONE_SB;
5136 if (*curpos < strbeg) {
5141 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5142 if (! prev_char_pos) {
5146 /* Back up over Extend and Format. curpos is always just to the right
5147 * of the characater whose value we are getting */
5149 U8 * prev_prev_char_pos;
5150 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1,
5153 sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5154 *curpos = prev_char_pos;
5155 prev_char_pos = prev_prev_char_pos;
5158 *curpos = (U8 *) strbeg;
5161 } while (sb == SB_Extend || sb == SB_Format);
5165 if (*curpos - 2 < strbeg) {
5166 *curpos = (U8 *) strbeg;
5170 sb = getSB_VAL_CP(*(*curpos - 1));
5171 } while (sb == SB_Extend || sb == SB_Format);
5178 S_isWB(pTHX_ WB_enum previous,
5181 const U8 * const strbeg,
5182 const U8 * const curpos,
5183 const U8 * const strend,
5184 const bool utf8_target)
5186 /* Return a boolean as to if the boundary between 'before' and 'after' is
5187 * a Unicode word break, using their published algorithm, but tailored for
5188 * Perl by treating spans of white space as one unit. Context may be
5189 * needed to make this determination. If the value for the character
5190 * before 'before' is known, it is passed as 'previous'; otherwise that
5191 * should be set to WB_UNKNOWN. The other input parameters give the
5192 * boundaries and current position in the matching of the string. That
5193 * is, 'curpos' marks the position where the character whose wb value is
5194 * 'after' begins. See http://www.unicode.org/reports/tr29/ */
5196 U8 * before_pos = (U8 *) curpos;
5197 U8 * after_pos = (U8 *) curpos;
5198 WB_enum prev = before;
5201 PERL_ARGS_ASSERT_ISWB;
5203 /* Rule numbers in the comments below are as of Unicode 9.0 */
5207 switch (WB_table[before][after]) {
5214 case WB_hs_then_hs: /* 2 horizontal spaces in a row */
5215 next = advance_one_WB(&after_pos, strend, utf8_target,
5216 FALSE /* Don't skip Extend nor Format */ );
5217 /* A space immediately preceeding an Extend or Format is attached
5218 * to by them, and hence gets separated from previous spaces.
5219 * Otherwise don't break between horizontal white space */
5220 return next == WB_Extend || next == WB_Format;
5222 /* WB4 Ignore Format and Extend characters, except when they appear at
5223 * the beginning of a region of text. This code currently isn't
5224 * general purpose, but it works as the rules are currently and likely
5225 * to be laid out. The reason it works is that when 'they appear at
5226 * the beginning of a region of text', the rule is to break before
5227 * them, just like any other character. Therefore, the default rule
5228 * applies and we don't have to look in more depth. Should this ever
5229 * change, we would have to have 2 'case' statements, like in the rules
5230 * below, and backup a single character (not spacing over the extend
5231 * ones) and then see if that is one of the region-end characters and
5233 case WB_Ex_or_FO_or_ZWJ_then_foo:
5234 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5237 case WB_DQ_then_HL + WB_BREAKABLE:
5238 case WB_DQ_then_HL + WB_NOBREAK:
5240 /* WB7c Hebrew_Letter Double_Quote × Hebrew_Letter */
5242 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5243 == WB_Hebrew_Letter)
5248 return WB_table[before][after] - WB_DQ_then_HL == WB_BREAKABLE;
5250 case WB_HL_then_DQ + WB_BREAKABLE:
5251 case WB_HL_then_DQ + WB_NOBREAK:
5253 /* WB7b Hebrew_Letter × Double_Quote Hebrew_Letter */
5255 if (advance_one_WB(&after_pos, strend, utf8_target,
5256 TRUE /* Do skip Extend and Format */ )
5257 == WB_Hebrew_Letter)
5262 return WB_table[before][after] - WB_HL_then_DQ == WB_BREAKABLE;
5264 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_NOBREAK:
5265 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_BREAKABLE:
5267 /* WB6 (ALetter | Hebrew_Letter) × (MidLetter | MidNumLet
5268 * | Single_Quote) (ALetter | Hebrew_Letter) */
5270 next = advance_one_WB(&after_pos, strend, utf8_target,
5271 TRUE /* Do skip Extend and Format */ );
5273 if (next == WB_ALetter || next == WB_Hebrew_Letter)
5278 return WB_table[before][after]
5279 - WB_LE_or_HL_then_MB_or_ML_or_SQ == WB_BREAKABLE;
5281 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_NOBREAK:
5282 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_BREAKABLE:
5284 /* WB7 (ALetter | Hebrew_Letter) (MidLetter | MidNumLet
5285 * | Single_Quote) × (ALetter | Hebrew_Letter) */
5287 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5288 if (prev == WB_ALetter || prev == WB_Hebrew_Letter)
5293 return WB_table[before][after]
5294 - WB_MB_or_ML_or_SQ_then_LE_or_HL == WB_BREAKABLE;
5296 case WB_MB_or_MN_or_SQ_then_NU + WB_NOBREAK:
5297 case WB_MB_or_MN_or_SQ_then_NU + WB_BREAKABLE:
5299 /* WB11 Numeric (MidNum | (MidNumLet | Single_Quote)) × Numeric
5302 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5308 return WB_table[before][after]
5309 - WB_MB_or_MN_or_SQ_then_NU == WB_BREAKABLE;
5311 case WB_NU_then_MB_or_MN_or_SQ + WB_NOBREAK:
5312 case WB_NU_then_MB_or_MN_or_SQ + WB_BREAKABLE:
5314 /* WB12 Numeric × (MidNum | MidNumLet | Single_Quote) Numeric */
5316 if (advance_one_WB(&after_pos, strend, utf8_target,
5317 TRUE /* Do skip Extend and Format */ )
5323 return WB_table[before][after]
5324 - WB_NU_then_MB_or_MN_or_SQ == WB_BREAKABLE;
5326 case WB_RI_then_RI + WB_NOBREAK:
5327 case WB_RI_then_RI + WB_BREAKABLE:
5331 /* Do not break within emoji flag sequences. That is, do not
5332 * break between regional indicator (RI) symbols if there is an
5333 * odd number of RI characters before the potential break
5336 * WB15 sot (RI RI)* RI × RI
5337 * WB16 [^RI] (RI RI)* RI × RI */
5339 while (backup_one_WB(&previous,
5342 utf8_target) == WB_Regional_Indicator)
5347 return RI_count % 2 != 1;
5355 Perl_re_printf( aTHX_ "Unhandled WB pair: WB_table[%d, %d] = %d\n",
5356 before, after, WB_table[before][after]);
5363 S_advance_one_WB(pTHX_ U8 ** curpos,
5364 const U8 * const strend,
5365 const bool utf8_target,
5366 const bool skip_Extend_Format)
5371 PERL_ARGS_ASSERT_ADVANCE_ONE_WB;
5373 if (*curpos >= strend) {
5379 /* Advance over Extend and Format */
5381 *curpos += UTF8SKIP(*curpos);
5382 if (*curpos >= strend) {
5385 wb = getWB_VAL_UTF8(*curpos, strend);
5386 } while ( skip_Extend_Format
5387 && (wb == WB_Extend || wb == WB_Format));
5392 if (*curpos >= strend) {
5395 wb = getWB_VAL_CP(**curpos);
5396 } while ( skip_Extend_Format
5397 && (wb == WB_Extend || wb == WB_Format));
5404 S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5409 PERL_ARGS_ASSERT_BACKUP_ONE_WB;
5411 /* If we know what the previous character's break value is, don't have
5413 if (*previous != WB_UNKNOWN) {
5416 /* But we need to move backwards by one */
5418 *curpos = reghopmaybe3(*curpos, -1, strbeg);
5420 *previous = WB_EDGE;
5421 *curpos = (U8 *) strbeg;
5424 *previous = WB_UNKNOWN;
5429 *previous = (*curpos <= strbeg) ? WB_EDGE : WB_UNKNOWN;
5432 /* And we always back up over these three types */
5433 if (wb != WB_Extend && wb != WB_Format && wb != WB_ZWJ) {
5438 if (*curpos < strbeg) {
5443 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5444 if (! prev_char_pos) {
5448 /* Back up over Extend and Format. curpos is always just to the right
5449 * of the characater whose value we are getting */
5451 U8 * prev_prev_char_pos;
5452 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos,
5456 wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5457 *curpos = prev_char_pos;
5458 prev_char_pos = prev_prev_char_pos;
5461 *curpos = (U8 *) strbeg;
5464 } while (wb == WB_Extend || wb == WB_Format || wb == WB_ZWJ);
5468 if (*curpos - 2 < strbeg) {
5469 *curpos = (U8 *) strbeg;
5473 wb = getWB_VAL_CP(*(*curpos - 1));
5474 } while (wb == WB_Extend || wb == WB_Format);
5480 /* Macros for regmatch(), using its internal variables */
5481 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
5482 #define NEXTCHR_IS_EOS (nextchr < 0)
5484 #define SET_nextchr \
5485 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
5487 #define SET_locinput(p) \
5491 #define sayYES goto yes
5492 #define sayNO goto no
5493 #define sayNO_SILENT goto no_silent
5495 /* we dont use STMT_START/END here because it leads to
5496 "unreachable code" warnings, which are bogus, but distracting. */
5497 #define CACHEsayNO \
5498 if (ST.cache_mask) \
5499 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
5502 #define EVAL_CLOSE_PAREN_IS(st,expr) \
5505 ( ( st )->u.eval.close_paren ) && \
5506 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5509 #define EVAL_CLOSE_PAREN_IS_TRUE(st,expr) \
5512 ( ( st )->u.eval.close_paren ) && \
5514 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5518 #define EVAL_CLOSE_PAREN_SET(st,expr) \
5519 (st)->u.eval.close_paren = ( (expr) + 1 )
5521 #define EVAL_CLOSE_PAREN_CLEAR(st) \
5522 (st)->u.eval.close_paren = 0
5524 /* push a new state then goto it */
5526 #define PUSH_STATE_GOTO(state, node, input, eol, sr0) \
5527 pushinput = input; \
5531 st->resume_state = state; \
5534 /* push a new state with success backtracking, then goto it */
5536 #define PUSH_YES_STATE_GOTO(state, node, input, eol, sr0) \
5537 pushinput = input; \
5541 st->resume_state = state; \
5542 goto push_yes_state;
5544 #define DEBUG_STATE_pp(pp) \
5546 DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \
5547 Perl_re_printf( aTHX_ \
5548 "%*s" pp " %s%s%s%s%s\n", \
5549 INDENT_CHARS(depth), "", \
5550 PL_reg_name[st->resume_state], \
5551 ((st==yes_state||st==mark_state) ? "[" : ""), \
5552 ((st==yes_state) ? "Y" : ""), \
5553 ((st==mark_state) ? "M" : ""), \
5554 ((st==yes_state||st==mark_state) ? "]" : "") \
5560 regmatch() - main matching routine
5562 This is basically one big switch statement in a loop. We execute an op,
5563 set 'next' to point the next op, and continue. If we come to a point which
5564 we may need to backtrack to on failure such as (A|B|C), we push a
5565 backtrack state onto the backtrack stack. On failure, we pop the top
5566 state, and re-enter the loop at the state indicated. If there are no more
5567 states to pop, we return failure.
5569 Sometimes we also need to backtrack on success; for example /A+/, where
5570 after successfully matching one A, we need to go back and try to
5571 match another one; similarly for lookahead assertions: if the assertion
5572 completes successfully, we backtrack to the state just before the assertion
5573 and then carry on. In these cases, the pushed state is marked as
5574 'backtrack on success too'. This marking is in fact done by a chain of
5575 pointers, each pointing to the previous 'yes' state. On success, we pop to
5576 the nearest yes state, discarding any intermediate failure-only states.
5577 Sometimes a yes state is pushed just to force some cleanup code to be
5578 called at the end of a successful match or submatch; e.g. (??{$re}) uses
5579 it to free the inner regex.
5581 Note that failure backtracking rewinds the cursor position, while
5582 success backtracking leaves it alone.
5584 A pattern is complete when the END op is executed, while a subpattern
5585 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
5586 ops trigger the "pop to last yes state if any, otherwise return true"
5589 A common convention in this function is to use A and B to refer to the two
5590 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
5591 the subpattern to be matched possibly multiple times, while B is the entire
5592 rest of the pattern. Variable and state names reflect this convention.
5594 The states in the main switch are the union of ops and failure/success of
5595 substates associated with with that op. For example, IFMATCH is the op
5596 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
5597 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
5598 successfully matched A and IFMATCH_A_fail is a state saying that we have
5599 just failed to match A. Resume states always come in pairs. The backtrack
5600 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
5601 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
5602 on success or failure.
5604 The struct that holds a backtracking state is actually a big union, with
5605 one variant for each major type of op. The variable st points to the
5606 top-most backtrack struct. To make the code clearer, within each
5607 block of code we #define ST to alias the relevant union.
5609 Here's a concrete example of a (vastly oversimplified) IFMATCH
5615 #define ST st->u.ifmatch
5617 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
5618 ST.foo = ...; // some state we wish to save
5620 // push a yes backtrack state with a resume value of
5621 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
5623 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
5626 case IFMATCH_A: // we have successfully executed A; now continue with B
5628 bar = ST.foo; // do something with the preserved value
5631 case IFMATCH_A_fail: // A failed, so the assertion failed
5632 ...; // do some housekeeping, then ...
5633 sayNO; // propagate the failure
5640 For any old-timers reading this who are familiar with the old recursive
5641 approach, the code above is equivalent to:
5643 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
5652 ...; // do some housekeeping, then ...
5653 sayNO; // propagate the failure
5656 The topmost backtrack state, pointed to by st, is usually free. If you
5657 want to claim it, populate any ST.foo fields in it with values you wish to
5658 save, then do one of
5660 PUSH_STATE_GOTO(resume_state, node, newinput, new_eol);
5661 PUSH_YES_STATE_GOTO(resume_state, node, newinput, new_eol);
5663 which sets that backtrack state's resume value to 'resume_state', pushes a
5664 new free entry to the top of the backtrack stack, then goes to 'node'.
5665 On backtracking, the free slot is popped, and the saved state becomes the
5666 new free state. An ST.foo field in this new top state can be temporarily
5667 accessed to retrieve values, but once the main loop is re-entered, it
5668 becomes available for reuse.
5670 Note that the depth of the backtrack stack constantly increases during the
5671 left-to-right execution of the pattern, rather than going up and down with
5672 the pattern nesting. For example the stack is at its maximum at Z at the
5673 end of the pattern, rather than at X in the following:
5675 /(((X)+)+)+....(Y)+....Z/
5677 The only exceptions to this are lookahead/behind assertions and the cut,
5678 (?>A), which pop all the backtrack states associated with A before
5681 Backtrack state structs are allocated in slabs of about 4K in size.
5682 PL_regmatch_state and st always point to the currently active state,
5683 and PL_regmatch_slab points to the slab currently containing
5684 PL_regmatch_state. The first time regmatch() is called, the first slab is
5685 allocated, and is never freed until interpreter destruction. When the slab
5686 is full, a new one is allocated and chained to the end. At exit from
5687 regmatch(), slabs allocated since entry are freed.
5689 In order to work with variable length lookbehinds, an upper limit is placed on
5690 lookbehinds which is set to where the match position is at the end of where the
5691 lookbehind would get to. Nothing in the lookbehind should match above that,
5692 except we should be able to look beyond if for things like \b, which need the
5693 next character in the string to be able to determine if this is a boundary or
5694 not. We also can't match the end of string/line unless we are also at the end
5695 of the entire string, so NEXTCHR_IS_EOS remains the same, and for those OPs
5696 that match a width, we have to add a condition that they are within the legal
5697 bounds of our window into the string.
5701 /* returns -1 on failure, $+[0] on success */
5703 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
5706 const bool utf8_target = reginfo->is_utf8_target;
5707 const U32 uniflags = UTF8_ALLOW_DEFAULT;
5708 REGEXP *rex_sv = reginfo->prog;
5709 regexp *rex = ReANY(rex_sv);
5710 RXi_GET_DECL(rex,rexi);
5711 /* the current state. This is a cached copy of PL_regmatch_state */
5713 /* cache heavy used fields of st in registers */
5716 U32 n = 0; /* general value; init to avoid compiler warning */
5717 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
5718 SSize_t endref = 0; /* offset of end of backref when ln is start */
5719 char *locinput = startpos;
5720 char *loceol = reginfo->strend;
5721 char *pushinput; /* where to continue after a PUSH */
5722 char *pusheol; /* where to stop matching (loceol) after a PUSH */
5723 U8 *pushsr0; /* save starting pos of script run */
5724 I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */
5726 bool result = 0; /* return value of S_regmatch */
5727 U32 depth = 0; /* depth of backtrack stack */
5728 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
5729 const U32 max_nochange_depth =
5730 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
5731 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
5732 regmatch_state *yes_state = NULL; /* state to pop to on success of
5734 /* mark_state piggy backs on the yes_state logic so that when we unwind
5735 the stack on success we can update the mark_state as we go */
5736 regmatch_state *mark_state = NULL; /* last mark state we have seen */
5737 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
5738 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
5740 bool no_final = 0; /* prevent failure from backtracking? */
5741 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
5742 char *startpoint = locinput;
5743 SV *popmark = NULL; /* are we looking for a mark? */
5744 SV *sv_commit = NULL; /* last mark name seen in failure */
5745 SV *sv_yes_mark = NULL; /* last mark name we have seen
5746 during a successful match */
5747 U32 lastopen = 0; /* last open we saw */
5748 bool has_cutgroup = RXp_HAS_CUTGROUP(rex) ? 1 : 0;
5749 SV* const oreplsv = GvSVn(PL_replgv);
5750 /* these three flags are set by various ops to signal information to
5751 * the very next op. They have a useful lifetime of exactly one loop
5752 * iteration, and are not preserved or restored by state pushes/pops
5754 bool sw = 0; /* the condition value in (?(cond)a|b) */
5755 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
5756 int logical = 0; /* the following EVAL is:
5760 or the following IFMATCH/UNLESSM is:
5761 false: plain (?=foo)
5762 true: used as a condition: (?(?=foo))
5764 PAD* last_pad = NULL;
5766 U8 gimme = G_SCALAR;
5767 CV *caller_cv = NULL; /* who called us */
5768 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
5769 U32 maxopenparen = 0; /* max '(' index seen so far */
5770 int to_complement; /* Invert the result? */
5771 _char_class_number classnum;
5772 bool is_utf8_pat = reginfo->is_utf8_pat;
5774 I32 orig_savestack_ix = PL_savestack_ix;
5775 U8 * script_run_begin = NULL;
5777 /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */
5778 #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL))
5779 # define SOLARIS_BAD_OPTIMIZER
5780 const U32 *pl_charclass_dup = PL_charclass;
5781 # define PL_charclass pl_charclass_dup
5785 GET_RE_DEBUG_FLAGS_DECL;
5788 /* protect against undef(*^R) */
5789 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
5791 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
5792 multicall_oldcatch = 0;
5793 PERL_UNUSED_VAR(multicall_cop);
5795 PERL_ARGS_ASSERT_REGMATCH;
5797 st = PL_regmatch_state;
5799 /* Note that nextchr is a byte even in UTF */
5803 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
5804 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5805 Perl_re_printf( aTHX_ "regmatch start\n" );
5808 while (scan != NULL) {
5809 next = scan + NEXT_OFF(scan);
5812 state_num = OP(scan);
5816 if (state_num <= REGNODE_MAX) {
5817 SV * const prop = sv_newmortal();
5818 regnode *rnext = regnext(scan);
5820 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5821 regprop(rex, prop, scan, reginfo, NULL);
5822 Perl_re_printf( aTHX_
5823 "%*s%" IVdf ":%s(%" IVdf ")\n",
5824 INDENT_CHARS(depth), "",
5825 (IV)(scan - rexi->program),
5827 (PL_regkind[OP(scan)] == END || !rnext) ?
5828 0 : (IV)(rnext - rexi->program));
5835 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
5837 switch (state_num) {
5838 case SBOL: /* /^../ and /\A../ */
5839 if (locinput == reginfo->strbeg)
5843 case MBOL: /* /^../m */
5844 if (locinput == reginfo->strbeg ||
5845 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
5852 if (locinput == reginfo->ganch)
5856 case KEEPS: /* \K */
5857 /* update the startpoint */
5858 st->u.keeper.val = rex->offs[0].start;
5859 rex->offs[0].start = locinput - reginfo->strbeg;
5860 PUSH_STATE_GOTO(KEEPS_next, next, locinput, loceol,
5862 NOT_REACHED; /* NOTREACHED */
5864 case KEEPS_next_fail:
5865 /* rollback the start point change */
5866 rex->offs[0].start = st->u.keeper.val;
5868 NOT_REACHED; /* NOTREACHED */
5870 case MEOL: /* /..$/m */
5871 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5875 case SEOL: /* /..$/ */
5876 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5878 if (reginfo->strend - locinput > 1)
5883 if (!NEXTCHR_IS_EOS)
5887 case SANY: /* /./s */
5888 if (NEXTCHR_IS_EOS || locinput >= loceol)
5890 goto increment_locinput;
5892 case REG_ANY: /* /./ */
5894 || locinput >= loceol
5899 goto increment_locinput;
5903 #define ST st->u.trie
5904 case TRIEC: /* (ab|cd) with known charclass */
5905 /* In this case the charclass data is available inline so
5906 we can fail fast without a lot of extra overhead.
5908 if ( ! NEXTCHR_IS_EOS
5909 && locinput < loceol
5910 && ! ANYOF_BITMAP_TEST(scan, nextchr))
5913 Perl_re_exec_indentf( aTHX_ "%sTRIE: failed to match trie start class...%s\n",
5914 depth, PL_colors[4], PL_colors[5])
5917 NOT_REACHED; /* NOTREACHED */
5920 case TRIE: /* (ab|cd) */
5921 /* the basic plan of execution of the trie is:
5922 * At the beginning, run though all the states, and
5923 * find the longest-matching word. Also remember the position
5924 * of the shortest matching word. For example, this pattern:
5927 * when matched against the string "abcde", will generate
5928 * accept states for all words except 3, with the longest
5929 * matching word being 4, and the shortest being 2 (with
5930 * the position being after char 1 of the string).
5932 * Then for each matching word, in word order (i.e. 1,2,4,5),
5933 * we run the remainder of the pattern; on each try setting
5934 * the current position to the character following the word,
5935 * returning to try the next word on failure.
5937 * We avoid having to build a list of words at runtime by
5938 * using a compile-time structure, wordinfo[].prev, which
5939 * gives, for each word, the previous accepting word (if any).
5940 * In the case above it would contain the mappings 1->2, 2->0,
5941 * 3->0, 4->5, 5->1. We can use this table to generate, from
5942 * the longest word (4 above), a list of all words, by
5943 * following the list of prev pointers; this gives us the
5944 * unordered list 4,5,1,2. Then given the current word we have
5945 * just tried, we can go through the list and find the
5946 * next-biggest word to try (so if we just failed on word 2,
5947 * the next in the list is 4).
5949 * Since at runtime we don't record the matching position in
5950 * the string for each word, we have to work that out for
5951 * each word we're about to process. The wordinfo table holds
5952 * the character length of each word; given that we recorded
5953 * at the start: the position of the shortest word and its
5954 * length in chars, we just need to move the pointer the
5955 * difference between the two char lengths. Depending on
5956 * Unicode status and folding, that's cheap or expensive.
5958 * This algorithm is optimised for the case where are only a
5959 * small number of accept states, i.e. 0,1, or maybe 2.
5960 * With lots of accepts states, and having to try all of them,
5961 * it becomes quadratic on number of accept states to find all
5966 /* what type of TRIE am I? (utf8 makes this contextual) */
5967 DECL_TRIE_TYPE(scan);
5969 /* what trie are we using right now */
5970 reg_trie_data * const trie
5971 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
5972 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
5973 U32 state = trie->startstate;
5975 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
5976 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5979 && UTF8_IS_ABOVE_LATIN1(nextchr)
5980 && scan->flags == EXACTL)
5982 /* We only output for EXACTL, as we let the folder
5983 * output this message for EXACTFLU8 to avoid
5985 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
5991 || locinput >= loceol
5992 || ! TRIE_BITMAP_TEST(trie, nextchr)))
5994 if (trie->states[ state ].wordnum) {
5996 Perl_re_exec_indentf( aTHX_ "%sTRIE: matched empty string...%s\n",
5997 depth, PL_colors[4], PL_colors[5])
6003 Perl_re_exec_indentf( aTHX_ "%sTRIE: failed to match trie start class...%s\n",
6004 depth, PL_colors[4], PL_colors[5])
6011 U8 *uc = ( U8* )locinput;
6015 U8 *uscan = (U8*)NULL;
6016 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
6017 U32 charcount = 0; /* how many input chars we have matched */
6018 U32 accepted = 0; /* have we seen any accepting states? */
6020 ST.jump = trie->jump;
6023 ST.longfold = FALSE; /* char longer if folded => it's harder */
6026 /* fully traverse the TRIE; note the position of the
6027 shortest accept state and the wordnum of the longest
6030 while ( state && uc <= (U8*)(loceol) ) {
6031 U32 base = trie->states[ state ].trans.base;
6035 wordnum = trie->states[ state ].wordnum;
6037 if (wordnum) { /* it's an accept state */
6040 /* record first match position */
6042 ST.firstpos = (U8*)locinput;
6047 ST.firstchars = charcount;
6050 if (!ST.nextword || wordnum < ST.nextword)
6051 ST.nextword = wordnum;
6052 ST.topword = wordnum;
6055 DEBUG_TRIE_EXECUTE_r({
6056 DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth );
6058 PerlIO_printf( Perl_debug_log,
6059 "%*s%sTRIE: State: %4" UVxf " Accepted: %c ",
6060 INDENT_CHARS(depth), "", PL_colors[4],
6061 (UV)state, (accepted ? 'Y' : 'N'));
6064 /* read a char and goto next state */
6065 if ( base && (foldlen || uc < (U8*)(loceol))) {
6067 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
6068 (U8 *) loceol, uscan,
6069 len, uvc, charid, foldlen,
6076 base + charid - 1 - trie->uniquecharcount)) >= 0)
6078 && ((U32)offset < trie->lasttrans)
6079 && trie->trans[offset].check == state)
6081 state = trie->trans[offset].next;
6092 DEBUG_TRIE_EXECUTE_r(
6093 Perl_re_printf( aTHX_
6094 "TRIE: Charid:%3x CP:%4" UVxf " After State: %4" UVxf "%s\n",
6095 charid, uvc, (UV)state, PL_colors[5] );
6101 /* calculate total number of accept states */
6106 w = trie->wordinfo[w].prev;
6109 ST.accepted = accepted;
6113 Perl_re_exec_indentf( aTHX_ "%sTRIE: got %" IVdf " possible matches%s\n",
6115 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
6117 goto trie_first_try; /* jump into the fail handler */
6119 NOT_REACHED; /* NOTREACHED */
6121 case TRIE_next_fail: /* we failed - try next alternative */
6125 /* undo any captures done in the tail part of a branch,
6127 * /(?:X(.)(.)|Y(.)).../
6128 * where the trie just matches X then calls out to do the
6129 * rest of the branch */
6130 REGCP_UNWIND(ST.cp);
6131 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6133 if (!--ST.accepted) {
6135 Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n",
6143 /* Find next-highest word to process. Note that this code
6144 * is O(N^2) per trie run (O(N) per branch), so keep tight */
6147 U16 const nextword = ST.nextword;
6148 reg_trie_wordinfo * const wordinfo
6149 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
6150 for (word=ST.topword; word; word=wordinfo[word].prev) {
6151 if (word > nextword && (!min || word < min))
6164 ST.lastparen = rex->lastparen;
6165 ST.lastcloseparen = rex->lastcloseparen;
6169 /* find start char of end of current word */
6171 U32 chars; /* how many chars to skip */
6172 reg_trie_data * const trie
6173 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
6175 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
6177 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
6182 /* the hard option - fold each char in turn and find
6183 * its folded length (which may be different */
6184 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
6192 /* XXX This assumes the length is well-formed, as
6193 * does the UTF8SKIP below */
6194 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
6202 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
6207 uvc = utf8n_to_uvchr(uscan, foldlen, &len,
6223 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
6224 ? ST.jump[ST.nextword]
6228 Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n",
6236 if ( ST.accepted > 1 || has_cutgroup || ST.jump ) {
6237 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc, loceol,
6239 NOT_REACHED; /* NOTREACHED */
6241 /* only one choice left - just continue */
6243 AV *const trie_words
6244 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
6245 SV ** const tmp = trie_words
6246 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL;
6247 SV *sv= tmp ? sv_newmortal() : NULL;
6249 Perl_re_exec_indentf( aTHX_ "%sTRIE: only one match left, short-circuiting: #%d <%s>%s\n",
6250 depth, PL_colors[4],
6252 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
6253 PL_colors[0], PL_colors[1],
6254 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
6256 : "not compiled under -Dr",
6260 locinput = (char*)uc;
6261 continue; /* execute rest of RE */
6266 case EXACTL: /* /abc/l */
6267 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6269 /* Complete checking would involve going through every character
6270 * matched by the string to see if any is above latin1. But the
6271 * comparision otherwise might very well be a fast assembly
6272 * language routine, and I (khw) don't think slowing things down
6273 * just to check for this warning is worth it. So this just checks
6274 * the first character */
6275 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
6276 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
6280 if (! utf8_target) {
6284 case EXACT: { /* /abc/ */
6289 if (utf8_target != is_utf8_pat) {
6290 /* The target and the pattern have differing utf8ness. */
6292 const char * const e = s + ln;
6295 /* The target is utf8, the pattern is not utf8.
6296 * Above-Latin1 code points can't match the pattern;
6297 * invariants match exactly, and the other Latin1 ones need
6298 * to be downgraded to a single byte in order to do the
6299 * comparison. (If we could be confident that the target
6300 * is not malformed, this could be refactored to have fewer
6301 * tests by just assuming that if the first bytes match, it
6302 * is an invariant, but there are tests in the test suite
6303 * dealing with (??{...}) which violate this) */
6306 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
6310 if (UTF8_IS_INVARIANT(*(U8*)l)) {
6317 if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
6327 /* The target is not utf8, the pattern is utf8. */
6330 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
6334 if (UTF8_IS_INVARIANT(*(U8*)s)) {
6341 if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
6353 /* The target and the pattern have the same utf8ness. */
6354 /* Inline the first character, for speed. */
6355 if ( loceol - locinput < ln
6356 || UCHARAT(s) != nextchr
6357 || (ln > 1 && memNE(s, locinput, ln)))
6366 case EXACTFL: /* /abc/il */
6369 const U8 * fold_array;
6371 U32 fold_utf8_flags;
6373 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6374 folder = foldEQ_locale;
6375 fold_array = PL_fold_locale;
6376 fold_utf8_flags = FOLDEQ_LOCALE;
6379 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
6380 is effectively /u; hence to match, target
6382 if (! utf8_target) {
6385 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
6386 | FOLDEQ_S2_FOLDS_SANE;
6387 folder = foldEQ_latin1_s2_folded;
6388 fold_array = PL_fold_latin1;
6391 case EXACTFU_ONLY8: /* /abc/iu with something in /abc/ > 255 */
6392 if (! utf8_target) {
6395 assert(is_utf8_pat);
6396 fold_utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
6399 case EXACTFUP: /* /foo/iu, and something is problematic in
6400 'foo' so can't take shortcuts. */
6401 assert(! is_utf8_pat);
6402 folder = foldEQ_latin1;
6403 fold_array = PL_fold_latin1;
6404 fold_utf8_flags = 0;
6407 case EXACTFU: /* /abc/iu */
6408 folder = foldEQ_latin1_s2_folded;
6409 fold_array = PL_fold_latin1;
6410 fold_utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
6413 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8
6415 assert(! is_utf8_pat);
6417 case EXACTFAA: /* /abc/iaa */
6418 folder = foldEQ_latin1_s2_folded;
6419 fold_array = PL_fold_latin1;
6420 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6421 if (is_utf8_pat || ! utf8_target) {
6423 /* The possible presence of a MICRO SIGN in the pattern forbids
6424 * us to view a non-UTF-8 pattern as folded when there is a
6426 fold_utf8_flags |= FOLDEQ_S2_ALREADY_FOLDED
6427 |FOLDEQ_S2_FOLDS_SANE;
6432 case EXACTF: /* /abc/i This node only generated for
6433 non-utf8 patterns */
6434 assert(! is_utf8_pat);
6436 fold_array = PL_fold;
6437 fold_utf8_flags = 0;
6445 || state_num == EXACTFUP
6446 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
6448 /* Either target or the pattern are utf8, or has the issue where
6449 * the fold lengths may differ. */
6450 const char * const l = locinput;
6453 if (! foldEQ_utf8_flags(l, &e, 0, utf8_target,
6454 s, 0, ln, is_utf8_pat,fold_utf8_flags))
6462 /* Neither the target nor the pattern are utf8 */
6463 if (UCHARAT(s) != nextchr
6465 && UCHARAT(s) != fold_array[nextchr])
6469 if (loceol - locinput < ln)
6471 if (ln > 1 && ! folder(locinput, s, ln))
6477 case NBOUNDL: /* /\B/l */
6481 case BOUNDL: /* /\b/l */
6484 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6486 if (FLAGS(scan) != TRADITIONAL_BOUND) {
6487 if (! IN_UTF8_CTYPE_LOCALE) {
6488 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
6489 B_ON_NON_UTF8_LOCALE_IS_WRONG);
6495 if (locinput == reginfo->strbeg)
6496 b1 = isWORDCHAR_LC('\n');
6498 b1 = isWORDCHAR_LC_utf8_safe(reghop3((U8*)locinput, -1,
6499 (U8*)(reginfo->strbeg)),
6500 (U8*)(reginfo->strend));
6502 b2 = (NEXTCHR_IS_EOS)
6503 ? isWORDCHAR_LC('\n')
6504 : isWORDCHAR_LC_utf8_safe((U8*) locinput,
6505 (U8*) reginfo->strend);
6507 else { /* Here the string isn't utf8 */
6508 b1 = (locinput == reginfo->strbeg)
6509 ? isWORDCHAR_LC('\n')
6510 : isWORDCHAR_LC(UCHARAT(locinput - 1));
6511 b2 = (NEXTCHR_IS_EOS)
6512 ? isWORDCHAR_LC('\n')
6513 : isWORDCHAR_LC(nextchr);
6515 if (to_complement ^ (b1 == b2)) {
6521 case NBOUND: /* /\B/ */
6525 case BOUND: /* /\b/ */
6529 goto bound_ascii_match_only;
6531 case NBOUNDA: /* /\B/a */
6535 case BOUNDA: /* /\b/a */
6539 bound_ascii_match_only:
6540 /* Here the string isn't utf8, or is utf8 and only ascii characters
6541 * are to match \w. In the latter case looking at the byte just
6542 * prior to the current one may be just the final byte of a
6543 * multi-byte character. This is ok. There are two cases:
6544 * 1) it is a single byte character, and then the test is doing
6545 * just what it's supposed to.
6546 * 2) it is a multi-byte character, in which case the final byte is
6547 * never mistakable for ASCII, and so the test will say it is
6548 * not a word character, which is the correct answer. */
6549 b1 = (locinput == reginfo->strbeg)
6550 ? isWORDCHAR_A('\n')
6551 : isWORDCHAR_A(UCHARAT(locinput - 1));
6552 b2 = (NEXTCHR_IS_EOS)
6553 ? isWORDCHAR_A('\n')
6554 : isWORDCHAR_A(nextchr);
6555 if (to_complement ^ (b1 == b2)) {
6561 case NBOUNDU: /* /\B/u */
6565 case BOUNDU: /* /\b/u */
6568 if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) {
6571 else if (utf8_target) {
6573 switch((bound_type) FLAGS(scan)) {
6574 case TRADITIONAL_BOUND:
6577 b1 = (locinput == reginfo->strbeg)
6578 ? 0 /* isWORDCHAR_L1('\n') */
6579 : isWORDCHAR_utf8_safe(
6580 reghop3((U8*)locinput,
6582 (U8*)(reginfo->strbeg)),
6583 (U8*) reginfo->strend);
6584 b2 = (NEXTCHR_IS_EOS)
6585 ? 0 /* isWORDCHAR_L1('\n') */
6586 : isWORDCHAR_utf8_safe((U8*)locinput,
6587 (U8*) reginfo->strend);
6588 match = cBOOL(b1 != b2);
6592 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6593 match = TRUE; /* GCB always matches at begin and
6597 /* Find the gcb values of previous and current
6598 * chars, then see if is a break point */
6599 match = isGCB(getGCB_VAL_UTF8(
6600 reghop3((U8*)locinput,
6602 (U8*)(reginfo->strbeg)),
6603 (U8*) reginfo->strend),
6604 getGCB_VAL_UTF8((U8*) locinput,
6605 (U8*) reginfo->strend),
6606 (U8*) reginfo->strbeg,
6613 if (locinput == reginfo->strbeg) {
6616 else if (NEXTCHR_IS_EOS) {
6620 match = isLB(getLB_VAL_UTF8(
6621 reghop3((U8*)locinput,
6623 (U8*)(reginfo->strbeg)),
6624 (U8*) reginfo->strend),
6625 getLB_VAL_UTF8((U8*) locinput,
6626 (U8*) reginfo->strend),
6627 (U8*) reginfo->strbeg,
6629 (U8*) reginfo->strend,
6634 case SB_BOUND: /* Always matches at begin and end */
6635 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6639 match = isSB(getSB_VAL_UTF8(
6640 reghop3((U8*)locinput,
6642 (U8*)(reginfo->strbeg)),
6643 (U8*) reginfo->strend),
6644 getSB_VAL_UTF8((U8*) locinput,
6645 (U8*) reginfo->strend),
6646 (U8*) reginfo->strbeg,
6648 (U8*) reginfo->strend,
6654 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6658 match = isWB(WB_UNKNOWN,
6660 reghop3((U8*)locinput,
6662 (U8*)(reginfo->strbeg)),
6663 (U8*) reginfo->strend),
6664 getWB_VAL_UTF8((U8*) locinput,
6665 (U8*) reginfo->strend),
6666 (U8*) reginfo->strbeg,
6668 (U8*) reginfo->strend,
6674 else { /* Not utf8 target */
6675 switch((bound_type) FLAGS(scan)) {
6676 case TRADITIONAL_BOUND:
6679 b1 = (locinput == reginfo->strbeg)
6680 ? 0 /* isWORDCHAR_L1('\n') */
6681 : isWORDCHAR_L1(UCHARAT(locinput - 1));
6682 b2 = (NEXTCHR_IS_EOS)
6683 ? 0 /* isWORDCHAR_L1('\n') */
6684 : isWORDCHAR_L1(nextchr);
6685 match = cBOOL(b1 != b2);
6690 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6691 match = TRUE; /* GCB always matches at begin and
6694 else { /* Only CR-LF combo isn't a GCB in 0-255
6696 match = UCHARAT(locinput - 1) != '\r'
6697 || UCHARAT(locinput) != '\n';
6702 if (locinput == reginfo->strbeg) {
6705 else if (NEXTCHR_IS_EOS) {
6709 match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)),
6710 getLB_VAL_CP(UCHARAT(locinput)),
6711 (U8*) reginfo->strbeg,
6713 (U8*) reginfo->strend,
6718 case SB_BOUND: /* Always matches at begin and end */
6719 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6723 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)),
6724 getSB_VAL_CP(UCHARAT(locinput)),
6725 (U8*) reginfo->strbeg,
6727 (U8*) reginfo->strend,
6733 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6737 match = isWB(WB_UNKNOWN,
6738 getWB_VAL_CP(UCHARAT(locinput -1)),
6739 getWB_VAL_CP(UCHARAT(locinput)),
6740 (U8*) reginfo->strbeg,
6742 (U8*) reginfo->strend,
6749 if (to_complement ^ ! match) {
6755 case ANYOFL: /* /[abc]/l */
6756 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6758 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE)
6760 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
6763 case ANYOFD: /* /[abc]/d */
6764 case ANYOF: /* /[abc]/ */
6765 if (NEXTCHR_IS_EOS || locinput >= loceol)
6767 if ( (! utf8_target || UTF8_IS_INVARIANT(*locinput))
6768 && ! (ANYOF_FLAGS(scan) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP))
6770 if (! ANYOF_BITMAP_TEST(scan, * (U8 *) (locinput))) {
6776 if (!reginclass(rex, scan, (U8*)locinput, (U8*) loceol,
6781 goto increment_locinput;
6787 || (UCHARAT(locinput) & FLAGS(scan)) != ARG(scan)
6788 || locinput >= loceol)
6792 locinput++; /* ANYOFM is always single byte */
6797 || (UCHARAT(locinput) & FLAGS(scan)) == ARG(scan)
6798 || locinput >= loceol)
6802 goto increment_locinput;
6808 || ! reginclass(rex, scan, (U8*)locinput, (U8*) loceol,
6813 goto increment_locinput;
6819 || ANYOF_FLAGS(scan) != (U8) *locinput
6820 || ! reginclass(rex, scan, (U8*)locinput, (U8*) loceol,
6825 goto increment_locinput;
6828 /* The argument (FLAGS) to all the POSIX node types is the class number
6831 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
6835 case POSIXL: /* \w or [:punct:] etc. under /l */
6836 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6837 if (NEXTCHR_IS_EOS || locinput >= loceol)
6840 /* Use isFOO_lc() for characters within Latin1. (Note that
6841 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6842 * wouldn't be invariant) */
6843 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6844 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
6852 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6853 /* An above Latin-1 code point, or malformed */
6854 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
6856 goto utf8_posix_above_latin1;
6859 /* Here is a UTF-8 variant code point below 256 and the target is
6861 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
6862 EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6863 *(locinput + 1))))))
6868 goto increment_locinput;
6870 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
6874 case POSIXD: /* \w or [:punct:] etc. under /d */
6880 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
6882 if (NEXTCHR_IS_EOS || locinput >= loceol) {
6886 /* All UTF-8 variants match */
6887 if (! UTF8_IS_INVARIANT(nextchr)) {
6888 goto increment_locinput;
6894 case POSIXA: /* \w or [:punct:] etc. under /a */
6897 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
6898 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
6899 * character is a single byte */
6901 if (NEXTCHR_IS_EOS || locinput >= loceol) {
6907 if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
6913 /* Here we are either not in utf8, or we matched a utf8-invariant,
6914 * so the next char is the next byte */
6918 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
6922 case POSIXU: /* \w or [:punct:] etc. under /u */
6924 if (NEXTCHR_IS_EOS || locinput >= loceol) {
6928 /* Use _generic_isCC() for characters within Latin1. (Note that
6929 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6930 * wouldn't be invariant) */
6931 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6932 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
6939 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6940 if (! (to_complement
6941 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6949 else { /* Handle above Latin-1 code points */
6950 utf8_posix_above_latin1:
6951 classnum = (_char_class_number) FLAGS(scan);
6954 if (! (to_complement
6955 ^ cBOOL(_invlist_contains_cp(
6956 PL_XPosix_ptrs[classnum],
6957 utf8_to_uvchr_buf((U8 *) locinput,
6958 (U8 *) reginfo->strend,
6964 case _CC_ENUM_SPACE:
6965 if (! (to_complement
6966 ^ cBOOL(is_XPERLSPACE_high(locinput))))
6971 case _CC_ENUM_BLANK:
6972 if (! (to_complement
6973 ^ cBOOL(is_HORIZWS_high(locinput))))
6978 case _CC_ENUM_XDIGIT:
6979 if (! (to_complement
6980 ^ cBOOL(is_XDIGIT_high(locinput))))
6985 case _CC_ENUM_VERTSPACE:
6986 if (! (to_complement
6987 ^ cBOOL(is_VERTWS_high(locinput))))
6992 case _CC_ENUM_CNTRL: /* These can't match above Latin1 */
6993 case _CC_ENUM_ASCII:
6994 if (! to_complement) {
6999 locinput += UTF8_SAFE_SKIP(locinput, reginfo->strend);
7003 case CLUMP: /* Match \X: logical Unicode character. This is defined as
7004 a Unicode extended Grapheme Cluster */
7005 if (NEXTCHR_IS_EOS || locinput >= loceol)
7007 if (! utf8_target) {
7009 /* Match either CR LF or '.', as all the other possibilities
7011 locinput++; /* Match the . or CR */
7012 if (nextchr == '\r' /* And if it was CR, and the next is LF,
7014 && locinput < loceol
7015 && UCHARAT(locinput) == '\n')
7022 /* Get the gcb type for the current character */
7023 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput,
7024 (U8*) reginfo->strend);
7026 /* Then scan through the input until we get to the first
7027 * character whose type is supposed to be a gcb with the
7028 * current character. (There is always a break at the
7030 locinput += UTF8SKIP(locinput);
7031 while (locinput < loceol) {
7032 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput,
7033 (U8*) reginfo->strend);
7034 if (isGCB(prev_gcb, cur_gcb,
7035 (U8*) reginfo->strbeg, (U8*) locinput,
7042 locinput += UTF8SKIP(locinput);
7049 case REFFLN: /* /\g{name}/il */
7050 { /* The capture buffer cases. The ones beginning with N for the
7051 named buffers just convert to the equivalent numbered and
7052 pretend they were called as the corresponding numbered buffer
7054 /* don't initialize these in the declaration, it makes C++
7059 const U8 *fold_array;
7062 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
7063 folder = foldEQ_locale;
7064 fold_array = PL_fold_locale;
7066 utf8_fold_flags = FOLDEQ_LOCALE;
7069 case REFFAN: /* /\g{name}/iaa */
7070 folder = foldEQ_latin1;
7071 fold_array = PL_fold_latin1;
7073 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7076 case REFFUN: /* /\g{name}/iu */
7077 folder = foldEQ_latin1;
7078 fold_array = PL_fold_latin1;
7080 utf8_fold_flags = 0;
7083 case REFFN: /* /\g{name}/i */
7085 fold_array = PL_fold;
7087 utf8_fold_flags = 0;
7090 case REFN: /* /\g{name}/ */
7094 utf8_fold_flags = 0;
7097 /* For the named back references, find the corresponding buffer
7099 n = reg_check_named_buff_matched(rex,scan);
7104 goto do_nref_ref_common;
7106 case REFFL: /* /\1/il */
7107 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
7108 folder = foldEQ_locale;
7109 fold_array = PL_fold_locale;
7110 utf8_fold_flags = FOLDEQ_LOCALE;
7113 case REFFA: /* /\1/iaa */
7114 folder = foldEQ_latin1;
7115 fold_array = PL_fold_latin1;
7116 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7119 case REFFU: /* /\1/iu */
7120 folder = foldEQ_latin1;
7121 fold_array = PL_fold_latin1;
7122 utf8_fold_flags = 0;
7125 case REFF: /* /\1/i */
7127 fold_array = PL_fold;
7128 utf8_fold_flags = 0;
7131 case REF: /* /\1/ */
7134 utf8_fold_flags = 0;
7138 n = ARG(scan); /* which paren pair */
7141 ln = rex->offs[n].start;
7142 endref = rex->offs[n].end;
7143 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7144 if (rex->lastparen < n || ln == -1 || endref == -1)
7145 sayNO; /* Do not match unless seen CLOSEn. */
7149 s = reginfo->strbeg + ln;
7150 if (type != REF /* REF can do byte comparison */
7151 && (utf8_target || type == REFFU || type == REFFL))
7153 char * limit = loceol;
7155 /* This call case insensitively compares the entire buffer
7156 * at s, with the current input starting at locinput, but
7157 * not going off the end given by loceol, and
7158 * returns in <limit> upon success, how much of the
7159 * current input was matched */
7160 if (! foldEQ_utf8_flags(s, NULL, endref - ln, utf8_target,
7161 locinput, &limit, 0, utf8_target, utf8_fold_flags))
7169 /* Not utf8: Inline the first character, for speed. */
7170 if ( ! NEXTCHR_IS_EOS
7171 && locinput < loceol
7172 && UCHARAT(s) != nextchr
7174 || UCHARAT(s) != fold_array[nextchr]))
7179 if (locinput + ln > loceol)
7181 if (ln > 1 && (type == REF
7182 ? memNE(s, locinput, ln)
7183 : ! folder(locinput, s, ln)))
7189 case NOTHING: /* null op; e.g. the 'nothing' following
7190 * the '*' in m{(a+|b)*}' */
7192 case TAIL: /* placeholder while compiling (A|B|C) */
7196 #define ST st->u.eval
7197 #define CUR_EVAL cur_eval->u.eval
7203 regexp_internal *rei;
7204 regnode *startpoint;
7207 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
7208 arg= (U32)ARG(scan);
7209 if (cur_eval && cur_eval->locinput == locinput) {
7210 if ( ++nochange_depth > max_nochange_depth )
7212 "Pattern subroutine nesting without pos change"
7213 " exceeded limit in regex");
7220 startpoint = scan + ARG2L(scan);
7221 EVAL_CLOSE_PAREN_SET( st, arg );
7222 /* Detect infinite recursion
7224 * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/
7225 * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever.
7226 * So we track the position in the string we are at each time
7227 * we recurse and if we try to enter the same routine twice from
7228 * the same position we throw an error.
7230 if ( rex->recurse_locinput[arg] == locinput ) {
7231 /* FIXME: we should show the regop that is failing as part
7232 * of the error message. */
7233 Perl_croak(aTHX_ "Infinite recursion in regex");
7235 ST.prev_recurse_locinput= rex->recurse_locinput[arg];
7236 rex->recurse_locinput[arg]= locinput;
7239 GET_RE_DEBUG_FLAGS_DECL;
7241 Perl_re_exec_indentf( aTHX_
7242 "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n",
7243 depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg]
7249 /* Save all the positions seen so far. */
7250 ST.cp = regcppush(rex, 0, maxopenparen);
7251 REGCP_SET(ST.lastcp);
7253 /* and then jump to the code we share with EVAL */
7254 goto eval_recurse_doit;
7257 case EVAL: /* /(?{...})B/ /(??{A})B/ and /(?(?{...})X|Y)B/ */
7258 if (cur_eval && cur_eval->locinput==locinput) {
7259 if ( ++nochange_depth > max_nochange_depth )
7260 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
7265 /* execute the code in the {...} */
7269 OP * const oop = PL_op;
7270 COP * const ocurcop = PL_curcop;
7274 /* save *all* paren positions */
7275 regcppush(rex, 0, maxopenparen);
7276 REGCP_SET(ST.lastcp);
7279 caller_cv = find_runcv(NULL);
7283 if (rexi->data->what[n] == 'r') { /* code from an external qr */
7285 (REGEXP*)(rexi->data->data[n])
7287 nop = (OP*)rexi->data->data[n+1];
7289 else if (rexi->data->what[n] == 'l') { /* literal code */
7291 nop = (OP*)rexi->data->data[n];
7292 assert(CvDEPTH(newcv));
7295 /* literal with own CV */
7296 assert(rexi->data->what[n] == 'L');
7297 newcv = rex->qr_anoncv;
7298 nop = (OP*)rexi->data->data[n];
7301 /* Some notes about MULTICALL and the context and save stacks.
7304 * /...(?{ my $x)}...(?{ my $y)}...(?{ my $z)}.../
7305 * since codeblocks don't introduce a new scope (so that
7306 * local() etc accumulate), at the end of a successful
7307 * match there will be a SAVEt_CLEARSV on the savestack
7308 * for each of $x, $y, $z. If the three code blocks above
7309 * happen to have come from different CVs (e.g. via
7310 * embedded qr//s), then we must ensure that during any
7311 * savestack unwinding, PL_comppad always points to the
7312 * right pad at each moment. We achieve this by
7313 * interleaving SAVEt_COMPPAD's on the savestack whenever
7314 * there is a change of pad.
7315 * In theory whenever we call a code block, we should
7316 * push a CXt_SUB context, then pop it on return from
7317 * that code block. This causes a bit of an issue in that
7318 * normally popping a context also clears the savestack
7319 * back to cx->blk_oldsaveix, but here we specifically
7320 * don't want to clear the save stack on exit from the
7322 * Also for efficiency we don't want to keep pushing and
7323 * popping the single SUB context as we backtrack etc.
7324 * So instead, we push a single context the first time
7325 * we need, it, then hang onto it until the end of this
7326 * function. Whenever we encounter a new code block, we
7327 * update the CV etc if that's changed. During the times
7328 * in this function where we're not executing a code
7329 * block, having the SUB context still there is a bit
7330 * naughty - but we hope that no-one notices.
7331 * When the SUB context is initially pushed, we fake up
7332 * cx->blk_oldsaveix to be as if we'd pushed this context
7333 * on first entry to S_regmatch rather than at some random
7334 * point during the regexe execution. That way if we
7335 * croak, popping the context stack will ensure that
7336 * *everything* SAVEd by this function is undone and then
7337 * the context popped, rather than e.g., popping the
7338 * context (and restoring the original PL_comppad) then
7339 * popping more of the savestack and restoring a bad
7343 /* If this is the first EVAL, push a MULTICALL. On
7344 * subsequent calls, if we're executing a different CV, or
7345 * if PL_comppad has got messed up from backtracking
7346 * through SAVECOMPPADs, then refresh the context.
7348 if (newcv != last_pushed_cv || PL_comppad != last_pad)
7350 U8 flags = (CXp_SUB_RE |
7351 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
7353 if (last_pushed_cv) {
7354 CHANGE_MULTICALL_FLAGS(newcv, flags);
7357 PUSH_MULTICALL_FLAGS(newcv, flags);
7359 /* see notes above */
7360 CX_CUR()->blk_oldsaveix = orig_savestack_ix;
7362 last_pushed_cv = newcv;
7365 /* these assignments are just to silence compiler
7367 multicall_cop = NULL;
7369 last_pad = PL_comppad;
7371 /* the initial nextstate you would normally execute
7372 * at the start of an eval (which would cause error
7373 * messages to come from the eval), may be optimised
7374 * away from the execution path in the regex code blocks;
7375 * so manually set PL_curcop to it initially */
7377 OP *o = cUNOPx(nop)->op_first;
7378 assert(o->op_type == OP_NULL);
7379 if (o->op_targ == OP_SCOPE) {
7380 o = cUNOPo->op_first;
7383 assert(o->op_targ == OP_LEAVE);
7384 o = cUNOPo->op_first;
7385 assert(o->op_type == OP_ENTER);
7389 if (o->op_type != OP_STUB) {
7390 assert( o->op_type == OP_NEXTSTATE
7391 || o->op_type == OP_DBSTATE
7392 || (o->op_type == OP_NULL
7393 && ( o->op_targ == OP_NEXTSTATE
7394 || o->op_targ == OP_DBSTATE
7398 PL_curcop = (COP*)o;
7403 DEBUG_STATE_r( Perl_re_printf( aTHX_
7404 " re EVAL PL_op=0x%" UVxf "\n", PTR2UV(nop)) );
7406 rex->offs[0].end = locinput - reginfo->strbeg;
7407 if (reginfo->info_aux_eval->pos_magic)
7408 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
7409 reginfo->sv, reginfo->strbeg,
7410 locinput - reginfo->strbeg);
7413 SV *sv_mrk = get_sv("REGMARK", 1);
7414 sv_setsv(sv_mrk, sv_yes_mark);
7417 /* we don't use MULTICALL here as we want to call the
7418 * first op of the block of interest, rather than the
7419 * first op of the sub. Also, we don't want to free
7420 * the savestack frame */
7421 before = (IV)(SP-PL_stack_base);
7423 CALLRUNOPS(aTHX); /* Scalar context. */
7425 if ((IV)(SP-PL_stack_base) == before)
7426 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
7432 /* before restoring everything, evaluate the returned
7433 * value, so that 'uninit' warnings don't use the wrong
7434 * PL_op or pad. Also need to process any magic vars
7435 * (e.g. $1) *before* parentheses are restored */
7440 if (logical == 0) { /* (?{})/ */
7441 SV *replsv = save_scalar(PL_replgv);
7442 sv_setsv(replsv, ret); /* $^R */
7445 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
7446 sw = cBOOL(SvTRUE_NN(ret));
7449 else { /* /(??{}) */
7450 /* if its overloaded, let the regex compiler handle
7451 * it; otherwise extract regex, or stringify */
7452 if (SvGMAGICAL(ret))
7453 ret = sv_mortalcopy(ret);
7454 if (!SvAMAGIC(ret)) {
7458 if (SvTYPE(sv) == SVt_REGEXP)
7459 re_sv = (REGEXP*) sv;
7460 else if (SvSMAGICAL(ret)) {
7461 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
7463 re_sv = (REGEXP *) mg->mg_obj;
7466 /* force any undef warnings here */
7467 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
7468 ret = sv_mortalcopy(ret);
7469 (void) SvPV_force_nolen(ret);
7475 /* *** Note that at this point we don't restore
7476 * PL_comppad, (or pop the CxSUB) on the assumption it may
7477 * be used again soon. This is safe as long as nothing
7478 * in the regexp code uses the pad ! */
7480 PL_curcop = ocurcop;
7481 regcp_restore(rex, ST.lastcp, &maxopenparen);
7482 PL_curpm_under = PL_curpm;
7483 PL_curpm = PL_reg_curpm;
7486 PUSH_STATE_GOTO(EVAL_B, next, locinput, loceol,
7492 /* only /(??{})/ from now on */
7495 /* extract RE object from returned value; compiling if
7499 re_sv = reg_temp_copy(NULL, re_sv);
7504 if (SvUTF8(ret) && IN_BYTES) {
7505 /* In use 'bytes': make a copy of the octet
7506 * sequence, but without the flag on */
7508 const char *const p = SvPV(ret, len);
7509 ret = newSVpvn_flags(p, len, SVs_TEMP);
7511 if (rex->intflags & PREGf_USE_RE_EVAL)
7512 pm_flags |= PMf_USE_RE_EVAL;
7514 /* if we got here, it should be an engine which
7515 * supports compiling code blocks and stuff */
7516 assert(rex->engine && rex->engine->op_comp);
7517 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
7518 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
7519 rex->engine, NULL, NULL,
7520 /* copy /msixn etc to inner pattern */
7525 & (SVs_TEMP | SVs_GMG | SVf_ROK))
7526 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
7527 /* This isn't a first class regexp. Instead, it's
7528 caching a regexp onto an existing, Perl visible
7530 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
7536 RXp_MATCH_COPIED_off(re);
7537 re->subbeg = rex->subbeg;
7538 re->sublen = rex->sublen;
7539 re->suboffset = rex->suboffset;
7540 re->subcoffset = rex->subcoffset;
7542 re->lastcloseparen = 0;
7545 debug_start_match(re_sv, utf8_target, locinput,
7546 reginfo->strend, "EVAL/GOSUB: Matching embedded");
7548 startpoint = rei->program + 1;
7549 EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0;
7550 * close_paren only for GOSUB */
7551 ST.prev_recurse_locinput= NULL; /* only used for GOSUB */
7552 /* Save all the seen positions so far. */
7553 ST.cp = regcppush(rex, 0, maxopenparen);
7554 REGCP_SET(ST.lastcp);
7555 /* and set maxopenparen to 0, since we are starting a "fresh" match */
7557 /* run the pattern returned from (??{...}) */
7559 eval_recurse_doit: /* Share code with GOSUB below this line
7560 * At this point we expect the stack context to be
7561 * set up correctly */
7563 /* invalidate the S-L poscache. We're now executing a
7564 * different set of WHILEM ops (and their associated
7565 * indexes) against the same string, so the bits in the
7566 * cache are meaningless. Setting maxiter to zero forces
7567 * the cache to be invalidated and zeroed before reuse.
7568 * XXX This is too dramatic a measure. Ideally we should
7569 * save the old cache and restore when running the outer
7571 reginfo->poscache_maxiter = 0;
7573 /* the new regexp might have a different is_utf8_pat than we do */
7574 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
7576 ST.prev_rex = rex_sv;
7577 ST.prev_curlyx = cur_curlyx;
7579 SET_reg_curpm(rex_sv);
7584 ST.prev_eval = cur_eval;
7586 /* now continue from first node in postoned RE */
7587 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, startpoint, locinput,
7588 loceol, script_run_begin);
7589 NOT_REACHED; /* NOTREACHED */
7592 case EVAL_postponed_AB: /* cleanup after a successful (??{A})B */
7593 /* note: this is called twice; first after popping B, then A */
7595 Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n",
7596 depth, cur_eval, ST.prev_eval);
7599 #define SET_RECURSE_LOCINPUT(STR,VAL)\
7600 if ( cur_eval && CUR_EVAL.close_paren ) {\
7602 Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\
7604 CUR_EVAL.close_paren - 1,\
7608 rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\
7611 SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput);
7613 rex_sv = ST.prev_rex;
7614 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7615 SET_reg_curpm(rex_sv);
7616 rex = ReANY(rex_sv);
7617 rexi = RXi_GET(rex);
7619 /* preserve $^R across LEAVE's. See Bug 121070. */
7620 SV *save_sv= GvSV(PL_replgv);
7622 SvREFCNT_inc(save_sv);
7623 regcpblow(ST.cp); /* LEAVE in disguise */
7624 /* don't move this initialization up */
7625 replsv = GvSV(PL_replgv);
7626 sv_setsv(replsv, save_sv);
7628 SvREFCNT_dec(save_sv);
7630 cur_eval = ST.prev_eval;
7631 cur_curlyx = ST.prev_curlyx;
7633 /* Invalidate cache. See "invalidate" comment above. */
7634 reginfo->poscache_maxiter = 0;
7635 if ( nochange_depth )
7638 SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput);
7642 case EVAL_B_fail: /* unsuccessful B in (?{...})B */
7643 REGCP_UNWIND(ST.lastcp);
7646 case EVAL_postponed_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
7647 /* note: this is called twice; first after popping B, then A */
7649 Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n",
7650 depth, cur_eval, ST.prev_eval);
7653 SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput);
7655 rex_sv = ST.prev_rex;
7656 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7657 SET_reg_curpm(rex_sv);
7658 rex = ReANY(rex_sv);
7659 rexi = RXi_GET(rex);
7661 REGCP_UNWIND(ST.lastcp);
7662 regcppop(rex, &maxopenparen);
7663 cur_eval = ST.prev_eval;
7664 cur_curlyx = ST.prev_curlyx;
7666 /* Invalidate cache. See "invalidate" comment above. */
7667 reginfo->poscache_maxiter = 0;
7668 if ( nochange_depth )
7671 SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput);
7676 n = ARG(scan); /* which paren pair */
7677 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
7678 if (n > maxopenparen)
7680 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
7681 "OPEN: rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf " tmp; maxopenparen=%" UVuf "\n",
7686 (IV)rex->offs[n].start_tmp,
7692 case SROPEN: /* (*SCRIPT_RUN: */
7693 script_run_begin = (U8 *) locinput;
7698 n = ARG(scan); /* which paren pair */
7699 CLOSE_CAPTURE(n, rex->offs[n].start_tmp,
7700 locinput - reginfo->strbeg);
7701 if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) )
7706 case SRCLOSE: /* (*SCRIPT_RUN: ... ) */
7708 if (! isSCRIPT_RUN(script_run_begin, (U8 *) locinput, utf8_target))
7716 case ACCEPT: /* (*ACCEPT) */
7718 sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7722 cursor && OP(cursor)!=END;
7723 cursor=regnext(cursor))
7725 if ( OP(cursor)==CLOSE ){
7727 if ( n <= lastopen ) {
7728 CLOSE_CAPTURE(n, rex->offs[n].start_tmp,
7729 locinput - reginfo->strbeg);
7730 if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) )
7739 case GROUPP: /* (?(1)) */
7740 n = ARG(scan); /* which paren pair */
7741 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
7744 case GROUPPN: /* (?(<name>)) */
7745 /* reg_check_named_buff_matched returns 0 for no match */
7746 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
7749 case INSUBP: /* (?(R)) */
7751 /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg
7752 * of SCAN is already set up as matches a eval.close_paren */
7753 sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n);
7756 case DEFINEP: /* (?(DEFINE)) */
7760 case IFTHEN: /* (?(cond)A|B) */
7761 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7763 next = NEXTOPER(NEXTOPER(scan));
7765 next = scan + ARG(scan);
7766 if (OP(next) == IFTHEN) /* Fake one. */
7767 next = NEXTOPER(NEXTOPER(next));
7771 case LOGICAL: /* modifier for EVAL and IFMATCH */
7772 logical = scan->flags;
7775 /*******************************************************************
7777 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
7778 pattern, where A and B are subpatterns. (For simple A, CURLYM or
7779 STAR/PLUS/CURLY/CURLYN are used instead.)
7781 A*B is compiled as <CURLYX><A><WHILEM><B>
7783 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
7784 state, which contains the current count, initialised to -1. It also sets
7785 cur_curlyx to point to this state, with any previous value saved in the
7788 CURLYX then jumps straight to the WHILEM op, rather than executing A,
7789 since the pattern may possibly match zero times (i.e. it's a while {} loop
7790 rather than a do {} while loop).
7792 Each entry to WHILEM represents a successful match of A. The count in the
7793 CURLYX block is incremented, another WHILEM state is pushed, and execution
7794 passes to A or B depending on greediness and the current count.
7796 For example, if matching against the string a1a2a3b (where the aN are
7797 substrings that match /A/), then the match progresses as follows: (the
7798 pushed states are interspersed with the bits of strings matched so far):
7801 <CURLYX cnt=0><WHILEM>
7802 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
7803 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
7804 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
7805 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
7807 (Contrast this with something like CURLYM, which maintains only a single
7811 a1 <CURLYM cnt=1> a2
7812 a1 a2 <CURLYM cnt=2> a3
7813 a1 a2 a3 <CURLYM cnt=3> b
7816 Each WHILEM state block marks a point to backtrack to upon partial failure
7817 of A or B, and also contains some minor state data related to that
7818 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
7819 overall state, such as the count, and pointers to the A and B ops.
7821 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
7822 must always point to the *current* CURLYX block, the rules are:
7824 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
7825 and set cur_curlyx to point the new block.
7827 When popping the CURLYX block after a successful or unsuccessful match,
7828 restore the previous cur_curlyx.
7830 When WHILEM is about to execute B, save the current cur_curlyx, and set it
7831 to the outer one saved in the CURLYX block.
7833 When popping the WHILEM block after a successful or unsuccessful B match,
7834 restore the previous cur_curlyx.
7836 Here's an example for the pattern (AI* BI)*BO
7837 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
7840 curlyx backtrack stack
7841 ------ ---------------
7843 CO <CO prev=NULL> <WO>
7844 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7845 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7846 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
7848 At this point the pattern succeeds, and we work back down the stack to
7849 clean up, restoring as we go:
7851 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7852 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7853 CO <CO prev=NULL> <WO>
7856 *******************************************************************/
7858 #define ST st->u.curlyx
7860 case CURLYX: /* start of /A*B/ (for complex A) */
7862 /* No need to save/restore up to this paren */
7863 I32 parenfloor = scan->flags;
7865 assert(next); /* keep Coverity happy */
7866 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
7869 /* XXXX Probably it is better to teach regpush to support
7870 parenfloor > maxopenparen ... */
7871 if (parenfloor > (I32)rex->lastparen)
7872 parenfloor = rex->lastparen; /* Pessimization... */
7874 ST.prev_curlyx= cur_curlyx;
7876 ST.cp = PL_savestack_ix;
7878 /* these fields contain the state of the current curly.
7879 * they are accessed by subsequent WHILEMs */
7880 ST.parenfloor = parenfloor;
7885 ST.count = -1; /* this will be updated by WHILEM */
7886 ST.lastloc = NULL; /* this will be updated by WHILEM */
7888 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput, loceol,
7890 NOT_REACHED; /* NOTREACHED */
7893 case CURLYX_end: /* just finished matching all of A*B */
7894 cur_curlyx = ST.prev_curlyx;
7896 NOT_REACHED; /* NOTREACHED */
7898 case CURLYX_end_fail: /* just failed to match all of A*B */
7900 cur_curlyx = ST.prev_curlyx;
7902 NOT_REACHED; /* NOTREACHED */
7906 #define ST st->u.whilem
7908 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
7910 /* see the discussion above about CURLYX/WHILEM */
7915 assert(cur_curlyx); /* keep Coverity happy */
7917 min = ARG1(cur_curlyx->u.curlyx.me);
7918 max = ARG2(cur_curlyx->u.curlyx.me);
7919 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
7920 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
7921 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
7922 ST.cache_offset = 0;
7926 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: matched %ld out of %d..%d\n",
7927 depth, (long)n, min, max)
7930 /* First just match a string of min A's. */
7933 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, maxopenparen);
7934 cur_curlyx->u.curlyx.lastloc = locinput;
7935 REGCP_SET(ST.lastcp);
7937 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput, loceol,
7939 NOT_REACHED; /* NOTREACHED */
7942 /* If degenerate A matches "", assume A done. */
7944 if (locinput == cur_curlyx->u.curlyx.lastloc) {
7945 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: empty match detected, trying continuation...\n",
7948 goto do_whilem_B_max;
7951 /* super-linear cache processing.
7953 * The idea here is that for certain types of CURLYX/WHILEM -
7954 * principally those whose upper bound is infinity (and
7955 * excluding regexes that have things like \1 and other very
7956 * non-regular expresssiony things), then if a pattern like
7957 * /....A*.../ fails and we backtrack to the WHILEM, then we
7958 * make a note that this particular WHILEM op was at string
7959 * position 47 (say) when the rest of pattern failed. Then, if
7960 * we ever find ourselves back at that WHILEM, and at string
7961 * position 47 again, we can just fail immediately rather than
7962 * running the rest of the pattern again.
7964 * This is very handy when patterns start to go
7965 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
7966 * with a combinatorial explosion of backtracking.
7968 * The cache is implemented as a bit array, with one bit per
7969 * string byte position per WHILEM op (up to 16) - so its
7970 * between 0.25 and 2x the string size.
7972 * To avoid allocating a poscache buffer every time, we do an
7973 * initially countdown; only after we have executed a WHILEM
7974 * op (string-length x #WHILEMs) times do we allocate the
7977 * The top 4 bits of scan->flags byte say how many different
7978 * relevant CURLLYX/WHILEM op pairs there are, while the
7979 * bottom 4-bits is the identifying index number of this
7985 if (!reginfo->poscache_maxiter) {
7986 /* start the countdown: Postpone detection until we
7987 * know the match is not *that* much linear. */
7988 reginfo->poscache_maxiter
7989 = (reginfo->strend - reginfo->strbeg + 1)
7991 /* possible overflow for long strings and many CURLYX's */
7992 if (reginfo->poscache_maxiter < 0)
7993 reginfo->poscache_maxiter = I32_MAX;
7994 reginfo->poscache_iter = reginfo->poscache_maxiter;
7997 if (reginfo->poscache_iter-- == 0) {
7998 /* initialise cache */
7999 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
8000 regmatch_info_aux *const aux = reginfo->info_aux;
8001 if (aux->poscache) {
8002 if ((SSize_t)reginfo->poscache_size < size) {
8003 Renew(aux->poscache, size, char);
8004 reginfo->poscache_size = size;
8006 Zero(aux->poscache, size, char);
8009 reginfo->poscache_size = size;
8010 Newxz(aux->poscache, size, char);
8012 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
8013 "%sWHILEM: Detected a super-linear match, switching on caching%s...\n",
8014 PL_colors[4], PL_colors[5])
8018 if (reginfo->poscache_iter < 0) {
8019 /* have we already failed at this position? */
8020 SSize_t offset, mask;
8022 reginfo->poscache_iter = -1; /* stop eventual underflow */
8023 offset = (scan->flags & 0xf) - 1
8024 + (locinput - reginfo->strbeg)
8026 mask = 1 << (offset % 8);
8028 if (reginfo->info_aux->poscache[offset] & mask) {
8029 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: (cache) already tried at this position...\n",
8032 cur_curlyx->u.curlyx.count--;
8033 sayNO; /* cache records failure */
8035 ST.cache_offset = offset;
8036 ST.cache_mask = mask;
8040 /* Prefer B over A for minimal matching. */
8042 if (cur_curlyx->u.curlyx.minmod) {
8043 ST.save_curlyx = cur_curlyx;
8044 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
8045 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
8046 locinput, loceol, script_run_begin);
8047 NOT_REACHED; /* NOTREACHED */
8050 /* Prefer A over B for maximal matching. */
8052 if (n < max) { /* More greed allowed? */
8053 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
8055 cur_curlyx->u.curlyx.lastloc = locinput;
8056 REGCP_SET(ST.lastcp);
8057 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput, loceol,
8059 NOT_REACHED; /* NOTREACHED */
8061 goto do_whilem_B_max;
8063 NOT_REACHED; /* NOTREACHED */
8065 case WHILEM_B_min: /* just matched B in a minimal match */
8066 case WHILEM_B_max: /* just matched B in a maximal match */
8067 cur_curlyx = ST.save_curlyx;
8069 NOT_REACHED; /* NOTREACHED */
8071 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
8072 cur_curlyx = ST.save_curlyx;
8073 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
8074 cur_curlyx->u.curlyx.count--;
8076 NOT_REACHED; /* NOTREACHED */
8078 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
8080 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
8081 REGCP_UNWIND(ST.lastcp);
8082 regcppop(rex, &maxopenparen);
8083 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
8084 cur_curlyx->u.curlyx.count--;
8086 NOT_REACHED; /* NOTREACHED */
8088 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
8089 REGCP_UNWIND(ST.lastcp);
8090 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
8091 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "WHILEM: failed, trying continuation...\n",
8095 if (cur_curlyx->u.curlyx.count >= REG_INFTY
8096 && ckWARN(WARN_REGEXP)
8097 && !reginfo->warned)
8099 reginfo->warned = TRUE;
8100 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
8101 "Complex regular subexpression recursion limit (%d) "
8107 ST.save_curlyx = cur_curlyx;
8108 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
8109 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
8110 locinput, loceol, script_run_begin);
8111 NOT_REACHED; /* NOTREACHED */
8113 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
8114 cur_curlyx = ST.save_curlyx;
8116 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
8117 /* Maximum greed exceeded */
8118 if (cur_curlyx->u.curlyx.count >= REG_INFTY
8119 && ckWARN(WARN_REGEXP)
8120 && !reginfo->warned)
8122 reginfo->warned = TRUE;
8123 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
8124 "Complex regular subexpression recursion "
8125 "limit (%d) exceeded",
8128 cur_curlyx->u.curlyx.count--;
8132 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "WHILEM: B min fail: trying longer...\n", depth)
8134 /* Try grabbing another A and see if it helps. */
8135 cur_curlyx->u.curlyx.lastloc = locinput;
8136 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
8138 REGCP_SET(ST.lastcp);
8139 PUSH_STATE_GOTO(WHILEM_A_min,
8140 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
8141 locinput, loceol, script_run_begin);
8142 NOT_REACHED; /* NOTREACHED */
8145 #define ST st->u.branch
8147 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
8148 next = scan + ARG(scan);
8151 scan = NEXTOPER(scan);
8154 case BRANCH: /* /(...|A|...)/ */
8155 scan = NEXTOPER(scan); /* scan now points to inner node */
8156 ST.lastparen = rex->lastparen;
8157 ST.lastcloseparen = rex->lastcloseparen;
8158 ST.next_branch = next;
8161 /* Now go into the branch */
8163 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput, loceol,
8166 PUSH_STATE_GOTO(BRANCH_next, scan, locinput, loceol,
8169 NOT_REACHED; /* NOTREACHED */
8171 case CUTGROUP: /* /(*THEN)/ */
8172 sv_yes_mark = st->u.mark.mark_name = scan->flags
8173 ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ])
8175 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput, loceol,
8177 NOT_REACHED; /* NOTREACHED */
8179 case CUTGROUP_next_fail:
8182 if (st->u.mark.mark_name)
8183 sv_commit = st->u.mark.mark_name;
8185 NOT_REACHED; /* NOTREACHED */
8189 NOT_REACHED; /* NOTREACHED */
8191 case BRANCH_next_fail: /* that branch failed; try the next, if any */
8196 REGCP_UNWIND(ST.cp);
8197 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8198 scan = ST.next_branch;
8199 /* no more branches? */
8200 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
8202 Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n",
8209 continue; /* execute next BRANCH[J] op */
8212 case MINMOD: /* next op will be non-greedy, e.g. A*? */
8217 #define ST st->u.curlym
8219 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
8221 /* This is an optimisation of CURLYX that enables us to push
8222 * only a single backtracking state, no matter how many matches
8223 * there are in {m,n}. It relies on the pattern being constant
8224 * length, with no parens to influence future backrefs
8228 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
8230 ST.lastparen = rex->lastparen;
8231 ST.lastcloseparen = rex->lastcloseparen;
8233 /* if paren positive, emulate an OPEN/CLOSE around A */
8235 U32 paren = ST.me->flags;
8236 if (paren > maxopenparen)
8237 maxopenparen = paren;
8238 scan += NEXT_OFF(scan); /* Skip former OPEN. */
8246 ST.c1 = CHRTEST_UNINIT;
8249 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
8252 curlym_do_A: /* execute the A in /A{m,n}B/ */
8253 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput, loceol, /* match A */
8255 NOT_REACHED; /* NOTREACHED */
8257 case CURLYM_A: /* we've just matched an A */
8259 /* after first match, determine A's length: u.curlym.alen */
8260 if (ST.count == 1) {
8261 if (reginfo->is_utf8_target) {
8262 char *s = st->locinput;
8263 while (s < locinput) {
8269 ST.alen = locinput - st->locinput;
8272 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
8275 Perl_re_exec_indentf( aTHX_ "CURLYM now matched %" IVdf " times, len=%" IVdf "...\n",
8276 depth, (IV) ST.count, (IV)ST.alen)
8279 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8283 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
8284 if ( max == REG_INFTY || ST.count < max )
8285 goto curlym_do_A; /* try to match another A */
8287 goto curlym_do_B; /* try to match B */
8289 case CURLYM_A_fail: /* just failed to match an A */
8290 REGCP_UNWIND(ST.cp);
8293 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
8294 || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8297 curlym_do_B: /* execute the B in /A{m,n}B/ */
8298 if (ST.c1 == CHRTEST_UNINIT) {
8299 /* calculate c1 and c2 for possible match of 1st char
8300 * following curly */
8301 ST.c1 = ST.c2 = CHRTEST_VOID;
8303 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
8304 regnode *text_node = ST.B;
8305 if (! HAS_TEXT(text_node))
8306 FIND_NEXT_IMPT(text_node);
8307 if (PL_regkind[OP(text_node)] == EXACT) {
8308 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
8309 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
8319 Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%" IVdf "...\n",
8320 depth, (IV)ST.count)
8322 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
8323 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
8325 /* (We can use memEQ and memNE in this file without
8326 * having to worry about one being shorter than the
8327 * other, since the first byte of each gives the
8328 * length of the character) */
8329 if ( memNE(locinput, ST.c1_utf8, UTF8_SAFE_SKIP(locinput,
8331 && memNE(locinput, ST.c2_utf8, UTF8_SAFE_SKIP(locinput,
8334 /* simulate B failing */
8336 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%" UVXf " c1=0x%" UVXf " c2=0x%" UVXf "\n",
8338 valid_utf8_to_uvchr((U8 *) locinput, NULL),
8339 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
8340 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
8342 state_num = CURLYM_B_fail;
8343 goto reenter_switch;
8346 else if (nextchr != ST.c1 && nextchr != ST.c2) {
8347 /* simulate B failing */
8349 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
8351 (int) nextchr, ST.c1, ST.c2)
8353 state_num = CURLYM_B_fail;
8354 goto reenter_switch;
8359 /* emulate CLOSE: mark current A as captured */
8360 U32 paren = (U32)ST.me->flags;
8362 CLOSE_CAPTURE(paren,
8363 HOPc(locinput, -ST.alen) - reginfo->strbeg,
8364 locinput - reginfo->strbeg);
8367 rex->offs[paren].end = -1;
8369 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8378 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput, loceol, /* match B */
8380 NOT_REACHED; /* NOTREACHED */
8382 case CURLYM_B_fail: /* just failed to match a B */
8383 REGCP_UNWIND(ST.cp);
8384 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8386 I32 max = ARG2(ST.me);
8387 if (max != REG_INFTY && ST.count == max)
8389 goto curlym_do_A; /* try to match a further A */
8391 /* backtrack one A */
8392 if (ST.count == ARG1(ST.me) /* min */)
8395 SET_locinput(HOPc(locinput, -ST.alen));
8396 goto curlym_do_B; /* try to match B */
8399 #define ST st->u.curly
8401 #define CURLY_SETPAREN(paren, success) \
8404 CLOSE_CAPTURE(paren, HOPc(locinput, -1) - reginfo->strbeg, \
8405 locinput - reginfo->strbeg); \
8408 rex->offs[paren].end = -1; \
8409 rex->lastparen = ST.lastparen; \
8410 rex->lastcloseparen = ST.lastcloseparen; \
8414 case STAR: /* /A*B/ where A is width 1 char */
8418 scan = NEXTOPER(scan);
8421 case PLUS: /* /A+B/ where A is width 1 char */
8425 scan = NEXTOPER(scan);
8428 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
8429 ST.paren = scan->flags; /* Which paren to set */
8430 ST.lastparen = rex->lastparen;
8431 ST.lastcloseparen = rex->lastcloseparen;
8432 if (ST.paren > maxopenparen)
8433 maxopenparen = ST.paren;
8434 ST.min = ARG1(scan); /* min to match */
8435 ST.max = ARG2(scan); /* max to match */
8436 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
8438 /* handle the single-char capture called as a GOSUB etc */
8439 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8441 char *li = locinput;
8442 if (!regrepeat(rex, &li, scan, loceol, reginfo, 1))
8450 case CURLY: /* /A{m,n}B/ where A is width 1 char */
8452 ST.min = ARG1(scan); /* min to match */
8453 ST.max = ARG2(scan); /* max to match */
8454 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
8457 * Lookahead to avoid useless match attempts
8458 * when we know what character comes next.
8460 * Used to only do .*x and .*?x, but now it allows
8461 * for )'s, ('s and (?{ ... })'s to be in the way
8462 * of the quantifier and the EXACT-like node. -- japhy
8465 assert(ST.min <= ST.max);
8466 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
8467 ST.c1 = ST.c2 = CHRTEST_VOID;
8470 regnode *text_node = next;
8472 if (! HAS_TEXT(text_node))
8473 FIND_NEXT_IMPT(text_node);
8475 if (! HAS_TEXT(text_node))
8476 ST.c1 = ST.c2 = CHRTEST_VOID;
8478 if ( PL_regkind[OP(text_node)] != EXACT ) {
8479 ST.c1 = ST.c2 = CHRTEST_VOID;
8482 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
8483 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
8495 char *li = locinput;
8498 regrepeat(rex, &li, ST.A, loceol, reginfo, ST.min)
8504 if (ST.c1 == CHRTEST_VOID)
8505 goto curly_try_B_min;
8507 ST.oldloc = locinput;
8509 /* set ST.maxpos to the furthest point along the
8510 * string that could possibly match */
8511 if (ST.max == REG_INFTY) {
8512 ST.maxpos = loceol - 1;
8514 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
8517 else if (utf8_target) {
8518 int m = ST.max - ST.min;
8519 for (ST.maxpos = locinput;
8520 m >0 && ST.maxpos < loceol; m--)
8521 ST.maxpos += UTF8SKIP(ST.maxpos);
8524 ST.maxpos = locinput + ST.max - ST.min;
8525 if (ST.maxpos >= loceol)
8526 ST.maxpos = loceol - 1;
8528 goto curly_try_B_min_known;
8532 /* avoid taking address of locinput, so it can remain
8534 char *li = locinput;
8535 ST.count = regrepeat(rex, &li, ST.A, loceol, reginfo, ST.max);
8536 if (ST.count < ST.min)
8539 if ((ST.count > ST.min)
8540 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
8542 /* A{m,n} must come at the end of the string, there's
8543 * no point in backing off ... */
8545 /* ...except that $ and \Z can match before *and* after
8546 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
8547 We may back off by one in this case. */
8548 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
8552 goto curly_try_B_max;
8554 NOT_REACHED; /* NOTREACHED */
8556 case CURLY_B_min_fail:
8557 /* failed to find B in a non-greedy match.
8558 * Handles both cases where c1,c2 valid or not */
8560 REGCP_UNWIND(ST.cp);
8562 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8565 if (ST.c1 == CHRTEST_VOID) {
8566 /* failed -- move forward one */
8567 char *li = locinput;
8568 if (!regrepeat(rex, &li, ST.A, loceol, reginfo, 1)) {
8573 if (!( ST.count <= ST.max
8574 /* count overflow ? */
8575 || (ST.max == REG_INFTY && ST.count > 0))
8581 /* Couldn't or didn't -- move forward. */
8582 ST.oldloc = locinput;
8584 locinput += UTF8SKIP(locinput);
8589 curly_try_B_min_known:
8590 /* find the next place where 'B' could work, then call B */
8592 n = (ST.oldloc == locinput) ? 0 : 1;
8593 if (ST.c1 == ST.c2) {
8594 /* set n to utf8_distance(oldloc, locinput) */
8595 while ( locinput <= ST.maxpos
8596 && locinput < loceol
8597 && memNE(locinput, ST.c1_utf8,
8598 UTF8_SAFE_SKIP(locinput, reginfo->strend)))
8600 locinput += UTF8_SAFE_SKIP(locinput,
8606 /* set n to utf8_distance(oldloc, locinput) */
8607 while ( locinput <= ST.maxpos
8608 && locinput < loceol
8609 && memNE(locinput, ST.c1_utf8,
8610 UTF8_SAFE_SKIP(locinput, reginfo->strend))
8611 && memNE(locinput, ST.c2_utf8,
8612 UTF8_SAFE_SKIP(locinput, reginfo->strend)))
8614 locinput += UTF8_SAFE_SKIP(locinput, reginfo->strend);
8619 else { /* Not utf8_target */
8620 if (ST.c1 == ST.c2) {
8621 locinput = (char *) memchr(locinput,
8623 ST.maxpos + 1 - locinput);
8625 locinput = ST.maxpos + 1;
8629 U8 c1_c2_bits_differing = ST.c1 ^ ST.c2;
8631 if (! isPOWER_OF_2(c1_c2_bits_differing)) {
8632 while ( locinput <= ST.maxpos
8633 && UCHARAT(locinput) != ST.c1
8634 && UCHARAT(locinput) != ST.c2)
8640 /* If c1 and c2 only differ by a single bit, we can
8641 * avoid a conditional each time through the loop,
8642 * at the expense of a little preliminary setup and
8643 * an extra mask each iteration. By masking out
8644 * that bit, we match exactly two characters, c1
8645 * and c2, and so we don't have to test for both.
8646 * On both ASCII and EBCDIC platforms, most of the
8647 * ASCII-range and Latin1-range folded equivalents
8648 * differ only in a single bit, so this is actually
8649 * the most common case. (e.g. 'A' 0x41 vs 'a'
8651 U8 c1_masked = ST.c1 &~ c1_c2_bits_differing;
8652 U8 c1_c2_mask = ~ c1_c2_bits_differing;
8653 while ( locinput <= ST.maxpos
8654 && (UCHARAT(locinput) & c1_c2_mask)
8661 n = locinput - ST.oldloc;
8663 if (locinput > ST.maxpos)
8666 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
8667 * at b; check that everything between oldloc and
8668 * locinput matches */
8669 char *li = ST.oldloc;
8671 if (regrepeat(rex, &li, ST.A, loceol, reginfo, n) < n)
8673 assert(n == REG_INFTY || locinput == li);
8678 CURLY_SETPAREN(ST.paren, ST.count);
8679 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput, loceol,
8681 NOT_REACHED; /* NOTREACHED */
8685 /* a successful greedy match: now try to match B */
8687 bool could_match = locinput < loceol;
8689 /* If it could work, try it. */
8690 if (ST.c1 != CHRTEST_VOID && could_match) {
8691 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
8693 could_match = memEQ(locinput, ST.c1_utf8,
8694 UTF8_SAFE_SKIP(locinput,
8696 || memEQ(locinput, ST.c2_utf8,
8697 UTF8_SAFE_SKIP(locinput,
8701 could_match = UCHARAT(locinput) == ST.c1
8702 || UCHARAT(locinput) == ST.c2;
8705 if (ST.c1 == CHRTEST_VOID || could_match) {
8706 CURLY_SETPAREN(ST.paren, ST.count);
8707 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput, loceol,
8709 NOT_REACHED; /* NOTREACHED */
8714 case CURLY_B_max_fail:
8715 /* failed to find B in a greedy match */
8717 REGCP_UNWIND(ST.cp);
8719 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8722 if (--ST.count < ST.min)
8724 locinput = HOPc(locinput, -1);
8725 goto curly_try_B_max;
8729 case END: /* last op of main pattern */
8732 /* we've just finished A in /(??{A})B/; now continue with B */
8733 SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput);
8734 st->u.eval.prev_rex = rex_sv; /* inner */
8736 /* Save *all* the positions. */
8737 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
8738 rex_sv = CUR_EVAL.prev_rex;
8739 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
8740 SET_reg_curpm(rex_sv);
8741 rex = ReANY(rex_sv);
8742 rexi = RXi_GET(rex);
8744 st->u.eval.prev_curlyx = cur_curlyx;
8745 cur_curlyx = CUR_EVAL.prev_curlyx;
8747 REGCP_SET(st->u.eval.lastcp);
8749 /* Restore parens of the outer rex without popping the
8751 regcp_restore(rex, CUR_EVAL.lastcp, &maxopenparen);
8753 st->u.eval.prev_eval = cur_eval;
8754 cur_eval = CUR_EVAL.prev_eval;
8756 Perl_re_exec_indentf( aTHX_ "END: EVAL trying tail ... (cur_eval=%p)\n",
8758 if ( nochange_depth )
8761 SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput);
8763 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, /* match B */
8764 st->u.eval.prev_eval->u.eval.B,
8765 locinput, loceol, script_run_begin);
8768 if (locinput < reginfo->till) {
8769 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
8770 "%sEND: Match possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
8772 (long)(locinput - startpos),
8773 (long)(reginfo->till - startpos),
8776 sayNO_SILENT; /* Cannot match: too short. */
8778 sayYES; /* Success! */
8780 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
8782 Perl_re_exec_indentf( aTHX_ "%sSUCCEED: subpattern success...%s\n",
8783 depth, PL_colors[4], PL_colors[5]));
8784 sayYES; /* Success! */
8787 #define ST st->u.ifmatch
8789 case SUSPEND: /* (?>A) */
8791 ST.start = locinput;
8796 case UNLESSM: /* -ve lookaround: (?!A), or with 'flags', (?<!A) */
8798 goto ifmatch_trivial_fail_test;
8800 case IFMATCH: /* +ve lookaround: (?=A), or with 'flags', (?<=A) */
8802 ifmatch_trivial_fail_test:
8803 ST.count = scan->next_off + 1; /* next_off repurposed to be
8804 lookbehind count, requires
8806 if (! scan->flags) { /* 'flags' zero means lookahed */
8808 /* Lookahead starts here and ends at the normal place */
8809 ST.start = locinput;
8813 PERL_UINT_FAST8_T back_count = scan->flags;
8816 /* Lookbehind can look beyond the current position */
8819 /* ... and starts at the first place in the input that is in
8820 * the range of the possible start positions */
8821 for (; ST.count > 0; ST.count--, back_count--) {
8822 s = HOPBACKc(locinput, back_count);
8829 /* If the lookbehind doesn't start in the actual string, is a
8830 * trivial match failure */
8833 sw = 1 - cBOOL(ST.wanted);
8838 /* Here, we didn't want it to match, so is actually success */
8839 next = scan + ARG(scan);
8847 ST.logical = logical;
8848 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
8850 /* execute body of (?...A) */
8851 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), ST.start,
8852 ST.end, script_run_begin);
8853 NOT_REACHED; /* NOTREACHED */
8858 case IFMATCH_A_fail: /* body of (?...A) failed */
8859 if (! ST.logical && ST.count > 1) {
8861 /* It isn't a real failure until we've tried all starting
8862 * positions. Move to the next starting position and retry */
8864 ST.start = HOPc(ST.start, 1);
8866 logical = ST.logical;
8870 /* Here, all starting positions have been tried. */
8874 case IFMATCH_A: /* body of (?...A) succeeded */
8877 sw = matched == ST.wanted;
8878 if (! ST.logical && !sw) {
8882 if (OP(ST.me) != SUSPEND) {
8883 /* restore old position except for (?>...) */
8884 locinput = st->locinput;
8885 loceol = st->loceol;
8886 script_run_begin = st->sr0;
8888 scan = ST.me + ARG(ST.me);
8891 continue; /* execute B */
8896 case LONGJMP: /* alternative with many branches compiles to
8897 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
8898 next = scan + ARG(scan);
8903 case COMMIT: /* (*COMMIT) */
8904 reginfo->cutpoint = loceol;
8907 case PRUNE: /* (*PRUNE) */
8909 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8910 PUSH_STATE_GOTO(COMMIT_next, next, locinput, loceol,
8912 NOT_REACHED; /* NOTREACHED */
8914 case COMMIT_next_fail:
8918 NOT_REACHED; /* NOTREACHED */
8920 case OPFAIL: /* (*FAIL) */
8922 sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8924 /* deal with (?(?!)X|Y) properly,
8925 * make sure we trigger the no branch
8926 * of the trailing IFTHEN structure*/
8932 NOT_REACHED; /* NOTREACHED */
8934 #define ST st->u.mark
8935 case MARKPOINT: /* (*MARK:foo) */
8936 ST.prev_mark = mark_state;
8937 ST.mark_name = sv_commit = sv_yes_mark
8938 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8940 ST.mark_loc = locinput;
8941 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput, loceol,
8943 NOT_REACHED; /* NOTREACHED */
8945 case MARKPOINT_next:
8946 mark_state = ST.prev_mark;
8948 NOT_REACHED; /* NOTREACHED */
8950 case MARKPOINT_next_fail:
8951 if (popmark && sv_eq(ST.mark_name,popmark))
8953 if (ST.mark_loc > startpoint)
8954 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8955 popmark = NULL; /* we found our mark */
8956 sv_commit = ST.mark_name;
8959 Perl_re_exec_indentf( aTHX_ "%sMARKPOINT: next fail: setting cutpoint to mark:%" SVf "...%s\n",
8961 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
8964 mark_state = ST.prev_mark;
8965 sv_yes_mark = mark_state ?
8966 mark_state->u.mark.mark_name : NULL;
8968 NOT_REACHED; /* NOTREACHED */
8970 case SKIP: /* (*SKIP) */
8972 /* (*SKIP) : if we fail we cut here*/
8973 ST.mark_name = NULL;
8974 ST.mark_loc = locinput;
8975 PUSH_STATE_GOTO(SKIP_next,next, locinput, loceol,
8978 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
8979 otherwise do nothing. Meaning we need to scan
8981 regmatch_state *cur = mark_state;
8982 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8985 if ( sv_eq( cur->u.mark.mark_name,
8988 ST.mark_name = find;
8989 PUSH_STATE_GOTO( SKIP_next, next, locinput, loceol,
8992 cur = cur->u.mark.prev_mark;
8995 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
8998 case SKIP_next_fail:
9000 /* (*CUT:NAME) - Set up to search for the name as we
9001 collapse the stack*/
9002 popmark = ST.mark_name;
9004 /* (*CUT) - No name, we cut here.*/
9005 if (ST.mark_loc > startpoint)
9006 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
9007 /* but we set sv_commit to latest mark_name if there
9008 is one so they can test to see how things lead to this
9011 sv_commit=mark_state->u.mark.mark_name;
9015 NOT_REACHED; /* NOTREACHED */
9018 case LNBREAK: /* \R */
9019 if ((n=is_LNBREAK_safe(locinput, loceol, utf8_target))) {
9026 PerlIO_printf(Perl_error_log, "%" UVxf " %d\n",
9027 PTR2UV(scan), OP(scan));
9028 Perl_croak(aTHX_ "regexp memory corruption");
9030 /* this is a point to jump to in order to increment
9031 * locinput by one character */
9033 assert(!NEXTCHR_IS_EOS);
9035 locinput += PL_utf8skip[nextchr];
9036 /* locinput is allowed to go 1 char off the end (signifying
9037 * EOS), but not 2+ */
9038 if (locinput > loceol)
9047 /* switch break jumps here */
9048 scan = next; /* prepare to execute the next op and ... */
9049 continue; /* ... jump back to the top, reusing st */
9053 /* push a state that backtracks on success */
9054 st->u.yes.prev_yes_state = yes_state;
9058 /* push a new regex state, then continue at scan */
9060 regmatch_state *newst;
9063 regmatch_state *cur = st;
9064 regmatch_state *curyes = yes_state;
9066 regmatch_slab *slab = PL_regmatch_slab;
9067 for (i = 0; i < 3 && i <= depth; cur--,i++) {
9068 if (cur < SLAB_FIRST(slab)) {
9070 cur = SLAB_LAST(slab);
9072 Perl_re_exec_indentf( aTHX_ "%4s #%-3d %-10s %s\n",
9075 depth - i, PL_reg_name[cur->resume_state],
9076 (curyes == cur) ? "yes" : ""
9079 curyes = cur->u.yes.prev_yes_state;
9082 DEBUG_STATE_pp("push")
9085 st->locinput = locinput;
9086 st->loceol = loceol;
9087 st->sr0 = script_run_begin;
9089 if (newst > SLAB_LAST(PL_regmatch_slab))
9090 newst = S_push_slab(aTHX);
9091 PL_regmatch_state = newst;
9093 locinput = pushinput;
9095 script_run_begin = pushsr0;
9101 #ifdef SOLARIS_BAD_OPTIMIZER
9102 # undef PL_charclass
9106 * We get here only if there's trouble -- normally "case END" is
9107 * the terminating point.
9109 Perl_croak(aTHX_ "corrupted regexp pointers");
9110 NOT_REACHED; /* NOTREACHED */
9114 /* we have successfully completed a subexpression, but we must now
9115 * pop to the state marked by yes_state and continue from there */
9116 assert(st != yes_state);
9118 while (st != yes_state) {
9120 if (st < SLAB_FIRST(PL_regmatch_slab)) {
9121 PL_regmatch_slab = PL_regmatch_slab->prev;
9122 st = SLAB_LAST(PL_regmatch_slab);
9126 DEBUG_STATE_pp("pop (no final)");
9128 DEBUG_STATE_pp("pop (yes)");
9134 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
9135 || yes_state > SLAB_LAST(PL_regmatch_slab))
9137 /* not in this slab, pop slab */
9138 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
9139 PL_regmatch_slab = PL_regmatch_slab->prev;
9140 st = SLAB_LAST(PL_regmatch_slab);
9142 depth -= (st - yes_state);
9145 yes_state = st->u.yes.prev_yes_state;
9146 PL_regmatch_state = st;
9149 locinput= st->locinput;
9151 script_run_begin = st->sr0;
9153 state_num = st->resume_state + no_final;
9154 goto reenter_switch;
9157 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n",
9158 PL_colors[4], PL_colors[5]));
9160 if (reginfo->info_aux_eval) {
9161 /* each successfully executed (?{...}) block does the equivalent of
9162 * local $^R = do {...}
9163 * When popping the save stack, all these locals would be undone;
9164 * bypass this by setting the outermost saved $^R to the latest
9166 /* I dont know if this is needed or works properly now.
9167 * see code related to PL_replgv elsewhere in this file.
9170 if (oreplsv != GvSV(PL_replgv)) {
9171 sv_setsv(oreplsv, GvSV(PL_replgv));
9172 SvSETMAGIC(oreplsv);
9180 Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n",
9182 PL_colors[4], PL_colors[5])
9194 /* there's a previous state to backtrack to */
9196 if (st < SLAB_FIRST(PL_regmatch_slab)) {
9197 PL_regmatch_slab = PL_regmatch_slab->prev;
9198 st = SLAB_LAST(PL_regmatch_slab);
9200 PL_regmatch_state = st;
9201 locinput= st->locinput;
9203 script_run_begin = st->sr0;
9205 DEBUG_STATE_pp("pop");
9207 if (yes_state == st)
9208 yes_state = st->u.yes.prev_yes_state;
9210 state_num = st->resume_state + 1; /* failure = success + 1 */
9212 goto reenter_switch;
9217 if (rex->intflags & PREGf_VERBARG_SEEN) {
9218 SV *sv_err = get_sv("REGERROR", 1);
9219 SV *sv_mrk = get_sv("REGMARK", 1);
9221 sv_commit = &PL_sv_no;
9223 sv_yes_mark = &PL_sv_yes;
9226 sv_commit = &PL_sv_yes;
9227 sv_yes_mark = &PL_sv_no;
9231 sv_setsv(sv_err, sv_commit);
9232 sv_setsv(sv_mrk, sv_yes_mark);
9236 if (last_pushed_cv) {
9238 /* see "Some notes about MULTICALL" above */
9240 PERL_UNUSED_VAR(SP);
9243 LEAVE_SCOPE(orig_savestack_ix);
9245 assert(!result || locinput - reginfo->strbeg >= 0);
9246 return result ? locinput - reginfo->strbeg : -1;
9250 - regrepeat - repeatedly match something simple, report how many
9252 * What 'simple' means is a node which can be the operand of a quantifier like
9255 * startposp - pointer to a pointer to the start position. This is updated
9256 * to point to the byte following the highest successful
9258 * p - the regnode to be repeatedly matched against.
9259 * loceol - pointer to the end position beyond which we aren't supposed to
9261 * reginfo - struct holding match state, such as utf8_target
9262 * max - maximum number of things to match.
9263 * depth - (for debugging) backtracking depth.
9266 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
9267 char * loceol, regmatch_info *const reginfo, I32 max _pDEPTH)
9270 char *scan; /* Pointer to current position in target string */
9272 char *this_eol = loceol; /* potentially adjusted version. */
9273 I32 hardcount = 0; /* How many matches so far */
9274 bool utf8_target = reginfo->is_utf8_target;
9275 unsigned int to_complement = 0; /* Invert the result? */
9277 _char_class_number classnum;
9279 PERL_ARGS_ASSERT_REGREPEAT;
9281 /* This routine is structured so that we switch on the input OP. Each OP
9282 * case: statement contains a loop to repeatedly apply the OP, advancing
9283 * the input until it fails, or reaches the end of the input, or until it
9284 * reaches the upper limit of matches. */
9287 if (max == REG_INFTY) /* This is a special marker to go to the platform's
9290 else if (! utf8_target && this_eol - scan > max)
9291 this_eol = scan + max;
9293 /* Here, for the case of a non-UTF-8 target we have adjusted <this_eol> down
9294 * to the maximum of how far we should go in it (leaving it set to the real
9295 * end, if the maximum permissible would take us beyond that). This allows
9296 * us to make the loop exit condition that we haven't gone past <this_eol> to
9297 * also mean that we haven't exceeded the max permissible count, saving a
9298 * test each time through the loops. But it assumes that the OP matches a
9299 * single byte, which is true for most of the OPs below when applied to a
9300 * non-UTF-8 target. Those relatively few OPs that don't have this
9301 * characteristic will have to compensate.
9303 * There is no adjustment for UTF-8 targets, as the number of bytes per
9304 * character varies. OPs will have to test both that the count is less
9305 * than the max permissible (using <hardcount> to keep track), and that we
9306 * are still within the bounds of the string (using <this_eol>. A few OPs
9307 * match a single byte no matter what the encoding. They can omit the max
9308 * test if, for the UTF-8 case, they do the adjustment that was skipped
9311 * Thus, the code above sets things up for the common case; and exceptional
9312 * cases need extra work; the common case is to make sure <scan> doesn't
9313 * go past <this_eol>, and for UTF-8 to also use <hardcount> to make sure the
9314 * count doesn't exceed the maximum permissible */
9319 while (scan < this_eol && hardcount < max && *scan != '\n') {
9320 scan += UTF8SKIP(scan);
9324 scan = (char *) memchr(scan, '\n', this_eol - scan);
9332 while (scan < this_eol && hardcount < max) {
9333 scan += UTF8SKIP(scan);
9341 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9342 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
9343 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
9348 if (! utf8_target) {
9354 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
9358 /* Can use a simple find if the pattern char to match on is invariant
9359 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
9360 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
9361 * true iff it doesn't matter if the argument is in UTF-8 or not */
9362 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
9363 if (utf8_target && this_eol - scan > max) {
9364 /* We didn't adjust <this_eol> because is UTF-8, but ok to do so,
9365 * since here, to match at all, 1 char == 1 byte */
9366 this_eol = scan + max;
9368 scan = (char *) find_span_end((U8 *) scan, (U8 *) this_eol, (U8) c);
9370 else if (reginfo->is_utf8_pat) {
9372 STRLEN scan_char_len;
9374 /* When both target and pattern are UTF-8, we have to do
9376 while (hardcount < max
9378 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
9379 && memEQ(scan, STRING(p), scan_char_len))
9381 scan += scan_char_len;
9385 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
9387 /* Target isn't utf8; convert the character in the UTF-8
9388 * pattern to non-UTF8, and do a simple find */
9389 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
9390 scan = (char *) find_span_end((U8 *) scan, (U8 *) this_eol, (U8) c);
9391 } /* else pattern char is above Latin1, can't possibly match the
9396 /* Here, the string must be utf8; pattern isn't, and <c> is
9397 * different in utf8 than not, so can't compare them directly.
9398 * Outside the loop, find the two utf8 bytes that represent c, and
9399 * then look for those in sequence in the utf8 string */
9400 U8 high = UTF8_TWO_BYTE_HI(c);
9401 U8 low = UTF8_TWO_BYTE_LO(c);
9403 while (hardcount < max
9404 && scan + 1 < this_eol
9405 && UCHARAT(scan) == high
9406 && UCHARAT(scan + 1) == low)
9414 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
9415 assert(! reginfo->is_utf8_pat);
9418 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
9419 if (reginfo->is_utf8_pat || ! utf8_target) {
9421 /* The possible presence of a MICRO SIGN in the pattern forbids us
9422 * to view a non-UTF-8 pattern as folded when there is a UTF-8
9424 utf8_flags |= FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
9429 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9430 utf8_flags = FOLDEQ_LOCALE;
9433 case EXACTF: /* This node only generated for non-utf8 patterns */
9434 assert(! reginfo->is_utf8_pat);
9438 if (! utf8_target) {
9441 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
9442 | FOLDEQ_S2_FOLDS_SANE;
9446 if (! utf8_target) {
9449 assert(reginfo->is_utf8_pat);
9450 utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
9454 utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
9461 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
9463 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
9465 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
9468 if (c1 == CHRTEST_VOID) {
9469 /* Use full Unicode fold matching */
9470 char *tmpeol = loceol;
9471 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
9472 while (hardcount < max
9473 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
9474 STRING(p), NULL, pat_len,
9475 reginfo->is_utf8_pat, utf8_flags))
9482 else if (utf8_target) {
9484 while (scan < this_eol
9486 && memEQ(scan, c1_utf8, UTF8_SAFE_SKIP(scan,
9489 scan += UTF8SKIP(c1_utf8);
9494 while (scan < this_eol
9496 && ( memEQ(scan, c1_utf8, UTF8_SAFE_SKIP(scan,
9498 || memEQ(scan, c2_utf8, UTF8_SAFE_SKIP(scan,
9501 scan += UTF8_SAFE_SKIP(scan, loceol);
9506 else if (c1 == c2) {
9507 scan = (char *) find_span_end((U8 *) scan, (U8 *) this_eol, (U8) c1);
9510 /* See comments in regmatch() CURLY_B_min_known_fail. We avoid
9511 * a conditional each time through the loop if the characters
9512 * differ only in a single bit, as is the usual situation */
9513 U8 c1_c2_bits_differing = c1 ^ c2;
9515 if (isPOWER_OF_2(c1_c2_bits_differing)) {
9516 U8 c1_c2_mask = ~ c1_c2_bits_differing;
9518 scan = (char *) find_span_end_mask((U8 *) scan,
9524 while ( scan < this_eol
9525 && (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
9536 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9538 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) {
9539 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
9545 while (hardcount < max
9547 && reginclass(prog, p, (U8*)scan, (U8*) this_eol, utf8_target))
9549 scan += UTF8SKIP(scan);
9553 else if (ANYOF_FLAGS(p) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
9554 while (scan < this_eol
9555 && reginclass(prog, p, (U8*)scan, (U8*)scan+1, 0))
9559 while (scan < this_eol && ANYOF_BITMAP_TEST(p, *((U8*)scan)))
9565 if (utf8_target && this_eol - scan > max) {
9567 /* We didn't adjust <this_eol> at the beginning of this routine
9568 * because is UTF-8, but it is actually ok to do so, since here, to
9569 * match, 1 char == 1 byte. */
9570 this_eol = scan + max;
9573 scan = (char *) find_span_end_mask((U8 *) scan, (U8 *) this_eol, (U8) ARG(p), FLAGS(p));
9578 while ( hardcount < max
9580 && (*scan & FLAGS(p)) != ARG(p))
9582 scan += UTF8SKIP(scan);
9587 scan = (char *) find_next_masked((U8 *) scan, (U8 *) this_eol, (U8) ARG(p), FLAGS(p));
9592 if (utf8_target) { /* ANYOFH only can match UTF-8 targets */
9593 while ( hardcount < max
9595 && reginclass(prog, p, (U8*)scan, (U8*) this_eol, TRUE))
9597 scan += UTF8SKIP(scan);
9604 if (utf8_target) { /* ANYOFHb only can match UTF-8 targets */
9606 /* we know the first byte must be the FLAGS field */
9607 while ( hardcount < max
9609 && (U8) *scan == ANYOF_FLAGS(p)
9610 && reginclass(prog, p, (U8*)scan, (U8*) this_eol,
9613 scan += UTF8SKIP(scan);
9619 /* The argument (FLAGS) to all the POSIX node types is the class number */
9626 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9627 if (! utf8_target) {
9628 while (scan < this_eol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
9634 while (hardcount < max && scan < this_eol
9635 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
9639 scan += UTF8SKIP(scan);
9652 if (utf8_target && this_eol - scan > max) {
9654 /* We didn't adjust <this_eol> at the beginning of this routine
9655 * because is UTF-8, but it is actually ok to do so, since here, to
9656 * match, 1 char == 1 byte. */
9657 this_eol = scan + max;
9659 while (scan < this_eol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
9672 if (! utf8_target) {
9673 while (scan < this_eol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
9679 /* The complement of something that matches only ASCII matches all
9680 * non-ASCII, plus everything in ASCII that isn't in the class. */
9681 while (hardcount < max && scan < this_eol
9682 && ( ! isASCII_utf8_safe(scan, loceol)
9683 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
9685 scan += UTF8SKIP(scan);
9696 if (! utf8_target) {
9697 while (scan < this_eol && to_complement
9698 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
9705 classnum = (_char_class_number) FLAGS(p);
9708 while ( hardcount < max && scan < this_eol
9709 && to_complement ^ cBOOL(_invlist_contains_cp(
9710 PL_XPosix_ptrs[classnum],
9711 utf8_to_uvchr_buf((U8 *) scan,
9715 scan += UTF8SKIP(scan);
9720 /* For the classes below, the knowledge of how to handle
9721 * every code point is compiled in to Perl via a macro.
9722 * This code is written for making the loops as tight as
9723 * possible. It could be refactored to save space instead.
9726 case _CC_ENUM_SPACE:
9727 while (hardcount < max
9730 ^ cBOOL(isSPACE_utf8_safe(scan, this_eol))))
9732 scan += UTF8SKIP(scan);
9736 case _CC_ENUM_BLANK:
9737 while (hardcount < max
9740 ^ cBOOL(isBLANK_utf8_safe(scan, this_eol))))
9742 scan += UTF8SKIP(scan);
9746 case _CC_ENUM_XDIGIT:
9747 while (hardcount < max
9750 ^ cBOOL(isXDIGIT_utf8_safe(scan, this_eol))))
9752 scan += UTF8SKIP(scan);
9756 case _CC_ENUM_VERTSPACE:
9757 while (hardcount < max
9760 ^ cBOOL(isVERTWS_utf8_safe(scan, this_eol))))
9762 scan += UTF8SKIP(scan);
9766 case _CC_ENUM_CNTRL:
9767 while (hardcount < max
9770 ^ cBOOL(isCNTRL_utf8_safe(scan, this_eol))))
9772 scan += UTF8SKIP(scan);
9782 while (hardcount < max && scan < this_eol &&
9783 (c=is_LNBREAK_utf8_safe(scan, this_eol))) {
9788 /* LNBREAK can match one or two latin chars, which is ok, but we
9789 * have to use hardcount in this situation, and throw away the
9790 * adjustment to <this_eol> done before the switch statement */
9791 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
9800 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9814 /* These are all 0 width, so match right here or not at all. */
9818 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
9819 NOT_REACHED; /* NOTREACHED */
9826 c = scan - *startposp;
9830 GET_RE_DEBUG_FLAGS_DECL;
9832 SV * const prop = sv_newmortal();
9833 regprop(prog, prop, p, reginfo, NULL);
9834 Perl_re_exec_indentf( aTHX_ "%s can match %" IVdf " times out of %" IVdf "...\n",
9835 depth, SvPVX_const(prop),(IV)c,(IV)max);
9843 - reginclass - determine if a character falls into a character class
9845 n is the ANYOF-type regnode
9846 p is the target string
9847 p_end points to one byte beyond the end of the target string
9848 utf8_target tells whether p is in UTF-8.
9850 Returns true if matched; false otherwise.
9852 Note that this can be a synthetic start class, a combination of various
9853 nodes, so things you think might be mutually exclusive, such as locale,
9854 aren't. It can match both locale and non-locale
9859 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
9862 const char flags = (OP(n) == ANYOFHb) ? 0 : ANYOF_FLAGS(n);
9866 PERL_ARGS_ASSERT_REGINCLASS;
9868 /* If c is not already the code point, get it. Note that
9869 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
9870 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
9872 const U32 utf8n_flags = UTF8_ALLOW_DEFAULT;
9873 c = utf8n_to_uvchr(p, p_end - p, &c_len, utf8n_flags | UTF8_CHECK_ONLY);
9874 if (c_len == (STRLEN)-1) {
9875 _force_out_malformed_utf8_message(p, p_end,
9877 1 /* 1 means die */ );
9878 NOT_REACHED; /* NOTREACHED */
9881 && (OP(n) == ANYOFL || OP(n) == ANYOFPOSIXL)
9882 && ! ANYOFL_UTF8_LOCALE_REQD(flags))
9884 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
9888 /* If this character is potentially in the bitmap, check it */
9889 if (c < NUM_ANYOF_CODE_POINTS && OP(n) != ANYOFH && OP(n) != ANYOFHb) {
9890 if (ANYOF_BITMAP_TEST(n, c))
9893 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9900 else if (flags & ANYOF_LOCALE_FLAGS) {
9901 if ( (flags & ANYOFL_FOLD)
9902 && c < sizeof(PL_fold_locale)
9903 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
9907 else if ( ANYOF_POSIXL_TEST_ANY_SET(n)
9908 && c <= U8_MAX /* param to isFOO_lc() */
9911 /* The data structure is arranged so bits 0, 2, 4, ... are set
9912 * if the class includes the Posix character class given by
9913 * bit/2; and 1, 3, 5, ... are set if the class includes the
9914 * complemented Posix class given by int(bit/2). So we loop
9915 * through the bits, each time changing whether we complement
9916 * the result or not. Suppose for the sake of illustration
9917 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
9918 * is set, it means there is a match for this ANYOF node if the
9919 * character is in the class given by the expression (0 / 2 = 0
9920 * = \w). If it is in that class, isFOO_lc() will return 1,
9921 * and since 'to_complement' is 0, the result will stay TRUE,
9922 * and we exit the loop. Suppose instead that bit 0 is 0, but
9923 * bit 1 is 1. That means there is a match if the character
9924 * matches \W. We won't bother to call isFOO_lc() on bit 0,
9925 * but will on bit 1. On the second iteration 'to_complement'
9926 * will be 1, so the exclusive or will reverse things, so we
9927 * are testing for \W. On the third iteration, 'to_complement'
9928 * will be 0, and we would be testing for \s; the fourth
9929 * iteration would test for \S, etc.
9931 * Note that this code assumes that all the classes are closed
9932 * under folding. For example, if a character matches \w, then
9933 * its fold does too; and vice versa. This should be true for
9934 * any well-behaved locale for all the currently defined Posix
9935 * classes, except for :lower: and :upper:, which are handled
9936 * by the pseudo-class :cased: which matches if either of the
9937 * other two does. To get rid of this assumption, an outer
9938 * loop could be used below to iterate over both the source
9939 * character, and its fold (if different) */
9942 int to_complement = 0;
9944 while (count < ANYOF_MAX) {
9945 if (ANYOF_POSIXL_TEST(n, count)
9946 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
9959 /* If the bitmap didn't (or couldn't) match, and something outside the
9960 * bitmap could match, try that. */
9962 if (c >= NUM_ANYOF_CODE_POINTS
9963 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
9965 match = TRUE; /* Everything above the bitmap matches */
9967 /* Here doesn't match everything above the bitmap. If there is
9968 * some information available beyond the bitmap, we may find a
9969 * match in it. If so, this is most likely because the code point
9970 * is outside the bitmap range. But rarely, it could be because of
9971 * some other reason. If so, various flags are set to indicate
9972 * this possibility. On ANYOFD nodes, there may be matches that
9973 * happen only when the target string is UTF-8; or for other node
9974 * types, because runtime lookup is needed, regardless of the
9975 * UTF-8ness of the target string. Finally, under /il, there may
9976 * be some matches only possible if the locale is a UTF-8 one. */
9977 else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP
9978 && ( c >= NUM_ANYOF_CODE_POINTS
9979 || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)
9980 && ( UNLIKELY(OP(n) != ANYOFD)
9981 || (utf8_target && ! isASCII_uni(c)
9982 # if NUM_ANYOF_CODE_POINTS > 256
9986 || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags)
9987 && IN_UTF8_CTYPE_LOCALE)))
9989 SV* only_utf8_locale = NULL;
9990 SV * const definition = _get_regclass_nonbitmap_data(prog, n, TRUE,
9991 0, &only_utf8_locale, NULL);
9997 } else { /* Convert to utf8 */
9998 utf8_p = utf8_buffer;
9999 append_utf8_from_native_byte(*p, &utf8_p);
10000 utf8_p = utf8_buffer;
10003 /* Turkish locales have these hard-coded rules overriding
10005 if ( UNLIKELY(PL_in_utf8_turkic_locale)
10006 && isALPHA_FOLD_EQ(*p, 'i'))
10009 if (_invlist_contains_cp(definition,
10010 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE))
10015 else if (*p == 'I') {
10016 if (_invlist_contains_cp(definition,
10017 LATIN_SMALL_LETTER_DOTLESS_I))
10023 else if (_invlist_contains_cp(definition, c)) {
10027 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
10028 match = _invlist_contains_cp(only_utf8_locale, c);
10032 /* In a Turkic locale under folding, hard-code the I i case pair
10034 if ( UNLIKELY(PL_in_utf8_turkic_locale)
10036 && (flags & ANYOFL_FOLD)
10039 if (c == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
10040 if (ANYOF_BITMAP_TEST(n, 'i')) {
10044 else if (c == LATIN_SMALL_LETTER_DOTLESS_I) {
10045 if (ANYOF_BITMAP_TEST(n, 'I')) {
10051 if (UNICODE_IS_SUPER(c)
10053 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
10055 && ckWARN_d(WARN_NON_UNICODE))
10057 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
10058 "Matched non-Unicode code point 0x%04" UVXf " against Unicode property; may not be portable", c);
10062 #if ANYOF_INVERT != 1
10063 /* Depending on compiler optimization cBOOL takes time, so if don't have to
10065 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
10068 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
10069 return (flags & ANYOF_INVERT) ^ match;
10073 S_reghop3(U8 *s, SSize_t off, const U8* lim)
10075 /* return the position 'off' UTF-8 characters away from 's', forward if
10076 * 'off' >= 0, backwards if negative. But don't go outside of position
10077 * 'lim', which better be < s if off < 0 */
10079 PERL_ARGS_ASSERT_REGHOP3;
10082 while (off-- && s < lim) {
10083 /* XXX could check well-formedness here */
10084 U8 *new_s = s + UTF8SKIP(s);
10085 if (new_s > lim) /* lim may be in the middle of a long character */
10091 while (off++ && s > lim) {
10093 if (UTF8_IS_CONTINUED(*s)) {
10094 while (s > lim && UTF8_IS_CONTINUATION(*s))
10096 if (! UTF8_IS_START(*s)) {
10097 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
10100 /* XXX could check well-formedness here */
10107 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
10109 PERL_ARGS_ASSERT_REGHOP4;
10112 while (off-- && s < rlim) {
10113 /* XXX could check well-formedness here */
10118 while (off++ && s > llim) {
10120 if (UTF8_IS_CONTINUED(*s)) {
10121 while (s > llim && UTF8_IS_CONTINUATION(*s))
10123 if (! UTF8_IS_START(*s)) {
10124 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
10127 /* XXX could check well-formedness here */
10133 /* like reghop3, but returns NULL on overrun, rather than returning last
10137 S_reghopmaybe3(U8* s, SSize_t off, const U8* const lim)
10139 PERL_ARGS_ASSERT_REGHOPMAYBE3;
10142 while (off-- && s < lim) {
10143 /* XXX could check well-formedness here */
10150 while (off++ && s > lim) {
10152 if (UTF8_IS_CONTINUED(*s)) {
10153 while (s > lim && UTF8_IS_CONTINUATION(*s))
10155 if (! UTF8_IS_START(*s)) {
10156 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
10159 /* XXX could check well-formedness here */
10168 /* when executing a regex that may have (?{}), extra stuff needs setting
10169 up that will be visible to the called code, even before the current
10170 match has finished. In particular:
10172 * $_ is localised to the SV currently being matched;
10173 * pos($_) is created if necessary, ready to be updated on each call-out
10175 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
10176 isn't set until the current pattern is successfully finished), so that
10177 $1 etc of the match-so-far can be seen;
10178 * save the old values of subbeg etc of the current regex, and set then
10179 to the current string (again, this is normally only done at the end
10184 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
10187 regexp *const rex = ReANY(reginfo->prog);
10188 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
10190 eval_state->rex = rex;
10193 /* Make $_ available to executed code. */
10194 if (reginfo->sv != DEFSV) {
10196 DEFSV_set(reginfo->sv);
10199 if (!(mg = mg_find_mglob(reginfo->sv))) {
10200 /* prepare for quick setting of pos */
10201 mg = sv_magicext_mglob(reginfo->sv);
10204 eval_state->pos_magic = mg;
10205 eval_state->pos = mg->mg_len;
10206 eval_state->pos_flags = mg->mg_flags;
10209 eval_state->pos_magic = NULL;
10211 if (!PL_reg_curpm) {
10212 /* PL_reg_curpm is a fake PMOP that we can attach the current
10213 * regex to and point PL_curpm at, so that $1 et al are visible
10214 * within a /(?{})/. It's just allocated once per interpreter the
10215 * first time its needed */
10216 Newxz(PL_reg_curpm, 1, PMOP);
10217 #ifdef USE_ITHREADS
10219 SV* const repointer = &PL_sv_undef;
10220 /* this regexp is also owned by the new PL_reg_curpm, which
10221 will try to free it. */
10222 av_push(PL_regex_padav, repointer);
10223 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
10224 PL_regex_pad = AvARRAY(PL_regex_padav);
10228 SET_reg_curpm(reginfo->prog);
10229 eval_state->curpm = PL_curpm;
10230 PL_curpm_under = PL_curpm;
10231 PL_curpm = PL_reg_curpm;
10232 if (RXp_MATCH_COPIED(rex)) {
10233 /* Here is a serious problem: we cannot rewrite subbeg,
10234 since it may be needed if this match fails. Thus
10235 $` inside (?{}) could fail... */
10236 eval_state->subbeg = rex->subbeg;
10237 eval_state->sublen = rex->sublen;
10238 eval_state->suboffset = rex->suboffset;
10239 eval_state->subcoffset = rex->subcoffset;
10240 #ifdef PERL_ANY_COW
10241 eval_state->saved_copy = rex->saved_copy;
10243 RXp_MATCH_COPIED_off(rex);
10246 eval_state->subbeg = NULL;
10247 rex->subbeg = (char *)reginfo->strbeg;
10248 rex->suboffset = 0;
10249 rex->subcoffset = 0;
10250 rex->sublen = reginfo->strend - reginfo->strbeg;
10254 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
10257 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
10259 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
10260 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
10263 Safefree(aux->poscache);
10267 /* undo the effects of S_setup_eval_state() */
10269 if (eval_state->subbeg) {
10270 regexp * const rex = eval_state->rex;
10271 rex->subbeg = eval_state->subbeg;
10272 rex->sublen = eval_state->sublen;
10273 rex->suboffset = eval_state->suboffset;
10274 rex->subcoffset = eval_state->subcoffset;
10275 #ifdef PERL_ANY_COW
10276 rex->saved_copy = eval_state->saved_copy;
10278 RXp_MATCH_COPIED_on(rex);
10280 if (eval_state->pos_magic)
10282 eval_state->pos_magic->mg_len = eval_state->pos;
10283 eval_state->pos_magic->mg_flags =
10284 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
10285 | (eval_state->pos_flags & MGf_BYTES);
10288 PL_curpm = eval_state->curpm;
10291 PL_regmatch_state = aux->old_regmatch_state;
10292 PL_regmatch_slab = aux->old_regmatch_slab;
10294 /* free all slabs above current one - this must be the last action
10295 * of this function, as aux and eval_state are allocated within
10296 * slabs and may be freed here */
10298 s = PL_regmatch_slab->next;
10300 PL_regmatch_slab->next = NULL;
10302 regmatch_slab * const osl = s;
10311 S_to_utf8_substr(pTHX_ regexp *prog)
10313 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
10314 * on the converted value */
10318 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
10321 if (prog->substrs->data[i].substr
10322 && !prog->substrs->data[i].utf8_substr) {
10323 SV* const sv = newSVsv(prog->substrs->data[i].substr);
10324 prog->substrs->data[i].utf8_substr = sv;
10325 sv_utf8_upgrade(sv);
10326 if (SvVALID(prog->substrs->data[i].substr)) {
10327 if (SvTAIL(prog->substrs->data[i].substr)) {
10328 /* Trim the trailing \n that fbm_compile added last
10330 SvCUR_set(sv, SvCUR(sv) - 1);
10331 /* Whilst this makes the SV technically "invalid" (as its
10332 buffer is no longer followed by "\0") when fbm_compile()
10333 adds the "\n" back, a "\0" is restored. */
10334 fbm_compile(sv, FBMcf_TAIL);
10336 fbm_compile(sv, 0);
10338 if (prog->substrs->data[i].substr == prog->check_substr)
10339 prog->check_utf8 = sv;
10345 S_to_byte_substr(pTHX_ regexp *prog)
10347 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
10348 * on the converted value; returns FALSE if can't be converted. */
10352 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
10355 if (prog->substrs->data[i].utf8_substr
10356 && !prog->substrs->data[i].substr) {
10357 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
10358 if (! sv_utf8_downgrade(sv, TRUE)) {
10361 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
10362 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
10363 /* Trim the trailing \n that fbm_compile added last
10365 SvCUR_set(sv, SvCUR(sv) - 1);
10366 fbm_compile(sv, FBMcf_TAIL);
10368 fbm_compile(sv, 0);
10370 prog->substrs->data[i].substr = sv;
10371 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
10372 prog->check_substr = sv;
10379 #ifndef PERL_IN_XSUB_RE
10382 Perl__is_grapheme(pTHX_ const U8 * strbeg, const U8 * s, const U8 * strend, const UV cp)
10384 /* Temporary helper function for toke.c. Verify that the code point 'cp'
10385 * is a stand-alone grapheme. The UTF-8 for 'cp' begins at position 's' in
10386 * the larger string bounded by 'strbeg' and 'strend'.
10388 * 'cp' needs to be assigned (if not a future version of the Unicode
10389 * Standard could make it something that combines with adjacent characters,
10390 * so code using it would then break), and there has to be a GCB break
10391 * before and after the character. */
10395 GCB_enum cp_gcb_val, prev_cp_gcb_val, next_cp_gcb_val;
10396 const U8 * prev_cp_start;
10398 PERL_ARGS_ASSERT__IS_GRAPHEME;
10400 if ( UNLIKELY(UNICODE_IS_SUPER(cp))
10401 || UNLIKELY(UNICODE_IS_NONCHAR(cp)))
10403 /* These are considered graphemes */
10407 /* Otherwise, unassigned code points are forbidden */
10408 if (UNLIKELY(! ELEMENT_RANGE_MATCHES_INVLIST(
10409 _invlist_search(PL_Assigned_invlist, cp))))
10414 cp_gcb_val = getGCB_VAL_CP(cp);
10416 /* Find the GCB value of the previous code point in the input */
10417 prev_cp_start = utf8_hop_back(s, -1, strbeg);
10418 if (UNLIKELY(prev_cp_start == s)) {
10419 prev_cp_gcb_val = GCB_EDGE;
10422 prev_cp_gcb_val = getGCB_VAL_UTF8(prev_cp_start, strend);
10425 /* And check that is a grapheme boundary */
10426 if (! isGCB(prev_cp_gcb_val, cp_gcb_val, strbeg, s,
10427 TRUE /* is UTF-8 encoded */ ))
10432 /* Similarly verify there is a break between the current character and the
10436 next_cp_gcb_val = GCB_EDGE;
10439 next_cp_gcb_val = getGCB_VAL_UTF8(s, strend);
10442 return isGCB(cp_gcb_val, next_cp_gcb_val, strbeg, s, TRUE);
10446 =head1 Unicode Support
10448 =for apidoc isSCRIPT_RUN
10450 Returns a bool as to whether or not the sequence of bytes from C<s> up to but
10451 not including C<send> form a "script run". C<utf8_target> is TRUE iff the
10452 sequence starting at C<s> is to be treated as UTF-8. To be precise, except for
10453 two degenerate cases given below, this function returns TRUE iff all code
10454 points in it come from any combination of three "scripts" given by the Unicode
10455 "Script Extensions" property: Common, Inherited, and possibly one other.
10456 Additionally all decimal digits must come from the same consecutive sequence of
10459 For example, if all the characters in the sequence are Greek, or Common, or
10460 Inherited, this function will return TRUE, provided any decimal digits in it
10461 are from the same block of digits in Common. (These are the ASCII digits
10462 "0".."9" and additionally a block for full width forms of these, and several
10463 others used in mathematical notation.) For scripts (unlike Greek) that have
10464 their own digits defined this will accept either digits from that set or from
10465 one of the Common digit sets, but not a combination of the two. Some scripts,
10466 such as Arabic, have more than one set of digits. All digits must come from
10467 the same set for this function to return TRUE.
10469 C<*ret_script>, if C<ret_script> is not NULL, will on return of TRUE
10470 contain the script found, using the C<SCX_enum> typedef. Its value will be
10471 C<SCX_INVALID> if the function returns FALSE.
10473 If the sequence is empty, TRUE is returned, but C<*ret_script> (if asked for)
10474 will be C<SCX_INVALID>.
10476 If the sequence contains a single code point which is unassigned to a character
10477 in the version of Unicode being used, the function will return TRUE, and the
10478 script will be C<SCX_Unknown>. Any other combination of unassigned code points
10479 in the input sequence will result in the function treating the input as not
10480 being a script run.
10482 The returned script will be C<SCX_Inherited> iff all the code points in it are
10483 from the Inherited script.
10485 Otherwise, the returned script will be C<SCX_Common> iff all the code points in
10486 it are from the Inherited or Common scripts.
10493 Perl_isSCRIPT_RUN(pTHX_ const U8 * s, const U8 * send, const bool utf8_target)
10495 /* Basically, it looks at each character in the sequence to see if the
10496 * above conditions are met; if not it fails. It uses an inversion map to
10497 * find the enum corresponding to the script of each character. But this
10498 * is complicated by the fact that a few code points can be in any of
10499 * several scripts. The data has been constructed so that there are
10500 * additional enum values (all negative) for these situations. The
10501 * absolute value of those is an index into another table which contains
10502 * pointers to auxiliary tables for each such situation. Each aux array
10503 * lists all the scripts for the given situation. There is another,
10504 * parallel, table that gives the number of entries in each aux table.
10505 * These are all defined in charclass_invlists.h */
10507 /* XXX Here are the additional things UTS 39 says could be done:
10509 * Forbid sequences of the same nonspacing mark
10511 * Check to see that all the characters are in the sets of exemplar
10512 * characters for at least one language in the Unicode Common Locale Data
10513 * Repository [CLDR]. */
10517 /* Things that match /\d/u */
10518 SV * decimals_invlist = PL_XPosix_ptrs[_CC_DIGIT];
10519 UV * decimals_array = invlist_array(decimals_invlist);
10521 /* What code point is the digit '0' of the script run? (0 meaning FALSE if
10522 * not currently known) */
10523 UV zero_of_run = 0;
10525 SCX_enum script_of_run = SCX_INVALID; /* Illegal value */
10526 SCX_enum script_of_char = SCX_INVALID;
10528 /* If the script remains not fully determined from iteration to iteration,
10529 * this is the current intersection of the possiblities. */
10530 SCX_enum * intersection = NULL;
10531 PERL_UINT_FAST8_T intersection_len = 0;
10533 bool retval = TRUE;
10534 SCX_enum * ret_script = NULL;
10538 PERL_ARGS_ASSERT_ISSCRIPT_RUN;
10540 /* All code points in 0..255 are either Common or Latin, so must be a
10541 * script run. We can return immediately unless we need to know which
10543 if (! utf8_target && LIKELY(send > s)) {
10544 if (ret_script == NULL) {
10548 /* If any character is Latin, the run is Latin */
10550 if (isALPHA_L1(*s) && LIKELY(*s != MICRO_SIGN_NATIVE)) {
10551 *ret_script = SCX_Latin;
10556 /* Here, all are Common */
10557 *ret_script = SCX_Common;
10561 /* Look at each character in the sequence */
10563 /* If the current character being examined is a digit, this is the code
10564 * point of the zero for its sequence of 10 */
10569 /* The code allows all scripts to use the ASCII digits. This is
10570 * because they are in the Common script. Hence any ASCII ones found
10571 * are ok, unless and until a digit from another set has already been
10572 * encountered. digit ranges in Common are not similarly blessed) */
10573 if (UNLIKELY(isDIGIT(*s))) {
10574 if (UNLIKELY(script_of_run == SCX_Unknown)) {
10579 if (zero_of_run != '0') {
10591 /* Here, isn't an ASCII digit. Find the code point of the character */
10592 if (! UTF8_IS_INVARIANT(*s)) {
10594 cp = valid_utf8_to_uvchr((U8 *) s, &len);
10601 /* If is within the range [+0 .. +9] of the script's zero, it also is a
10602 * digit in that script. We can skip the rest of this code for this
10604 if (UNLIKELY( zero_of_run
10605 && cp >= zero_of_run
10606 && cp - zero_of_run <= 9))
10611 /* Find the character's script. The correct values are hard-coded here
10612 * for small-enough code points. */
10613 if (cp < 0x2B9) { /* From inspection of Unicode db; extremely
10614 unlikely to change */
10616 || ( isALPHA_L1(cp)
10617 && LIKELY(cp != MICRO_SIGN_NATIVE)))
10619 script_of_char = SCX_Latin;
10622 script_of_char = SCX_Common;
10626 script_of_char = _Perl_SCX_invmap[
10627 _invlist_search(PL_SCX_invlist, cp)];
10630 /* We arbitrarily accept a single unassigned character, but not in
10631 * combination with anything else, and not a run of them. */
10632 if ( UNLIKELY(script_of_run == SCX_Unknown)
10633 || UNLIKELY( script_of_run != SCX_INVALID
10634 && script_of_char == SCX_Unknown))
10640 /* For the first character, or the run is inherited, the run's script
10641 * is set to the char's */
10642 if ( UNLIKELY(script_of_run == SCX_INVALID)
10643 || UNLIKELY(script_of_run == SCX_Inherited))
10645 script_of_run = script_of_char;
10648 /* For the character's script to be Unknown, it must be the first
10649 * character in the sequence (for otherwise a test above would have
10650 * prevented us from reaching here), and we have set the run's script
10651 * to it. Nothing further to be done for this character */
10652 if (UNLIKELY(script_of_char == SCX_Unknown)) {
10656 /* We accept 'inherited' script characters currently even at the
10657 * beginning. (We know that no characters in Inherited are digits, or
10658 * we'd have to check for that) */
10659 if (UNLIKELY(script_of_char == SCX_Inherited)) {
10663 /* If the run so far is Common, and the new character isn't, change the
10664 * run's script to that of this character */
10665 if (script_of_run == SCX_Common && script_of_char != SCX_Common) {
10666 script_of_run = script_of_char;
10669 /* Now we can see if the script of the new character is the same as
10670 * that of the run */
10671 if (LIKELY(script_of_char == script_of_run)) {
10672 /* By far the most common case */
10673 goto scripts_match;
10676 /* Here, the script of the run isn't Common. But characters in Common
10677 * match any script */
10678 if (script_of_char == SCX_Common) {
10679 goto scripts_match;
10682 #ifndef HAS_SCX_AUX_TABLES
10684 /* Too early a Unicode version to have a code point belonging to more
10685 * than one script, so, if the scripts don't exactly match, fail */
10686 PERL_UNUSED_VAR(intersection_len);
10692 /* Here there is no exact match between the character's script and the
10693 * run's. And we've handled the special cases of scripts Unknown,
10694 * Inherited, and Common.
10696 * Negative script numbers signify that the value may be any of several
10697 * scripts, and we need to look at auxiliary information to make our
10698 * deterimination. But if both are non-negative, we can fail now */
10699 if (LIKELY(script_of_char >= 0)) {
10700 const SCX_enum * search_in;
10701 PERL_UINT_FAST8_T search_in_len;
10702 PERL_UINT_FAST8_T i;
10704 if (LIKELY(script_of_run >= 0)) {
10709 /* Use the previously constructed set of possible scripts, if any.
10711 if (intersection) {
10712 search_in = intersection;
10713 search_in_len = intersection_len;
10716 search_in = SCX_AUX_TABLE_ptrs[-script_of_run];
10717 search_in_len = SCX_AUX_TABLE_lengths[-script_of_run];
10720 for (i = 0; i < search_in_len; i++) {
10721 if (search_in[i] == script_of_char) {
10722 script_of_run = script_of_char;
10723 goto scripts_match;
10730 else if (LIKELY(script_of_run >= 0)) {
10731 /* script of character could be one of several, but run is a single
10733 const SCX_enum * search_in = SCX_AUX_TABLE_ptrs[-script_of_char];
10734 const PERL_UINT_FAST8_T search_in_len
10735 = SCX_AUX_TABLE_lengths[-script_of_char];
10736 PERL_UINT_FAST8_T i;
10738 for (i = 0; i < search_in_len; i++) {
10739 if (search_in[i] == script_of_run) {
10740 script_of_char = script_of_run;
10741 goto scripts_match;
10749 /* Both run and char could be in one of several scripts. If the
10750 * intersection is empty, then this character isn't in this script
10751 * run. Otherwise, we need to calculate the intersection to use
10752 * for future iterations of the loop, unless we are already at the
10753 * final character */
10754 const SCX_enum * search_char = SCX_AUX_TABLE_ptrs[-script_of_char];
10755 const PERL_UINT_FAST8_T char_len
10756 = SCX_AUX_TABLE_lengths[-script_of_char];
10757 const SCX_enum * search_run;
10758 PERL_UINT_FAST8_T run_len;
10760 SCX_enum * new_overlap = NULL;
10761 PERL_UINT_FAST8_T i, j;
10763 if (intersection) {
10764 search_run = intersection;
10765 run_len = intersection_len;
10768 search_run = SCX_AUX_TABLE_ptrs[-script_of_run];
10769 run_len = SCX_AUX_TABLE_lengths[-script_of_run];
10772 intersection_len = 0;
10774 for (i = 0; i < run_len; i++) {
10775 for (j = 0; j < char_len; j++) {
10776 if (search_run[i] == search_char[j]) {
10778 /* Here, the script at i,j matches. That means this
10779 * character is in the run. But continue on to find
10780 * the complete intersection, for the next loop
10781 * iteration, and for the digit check after it.
10783 * On the first found common script, we malloc space
10784 * for the intersection list for the worst case of the
10785 * intersection, which is the minimum of the number of
10786 * scripts remaining in each set. */
10787 if (intersection_len == 0) {
10789 MIN(run_len - i, char_len - j),
10792 new_overlap[intersection_len++] = search_run[i];
10797 /* Here we've looked through everything. If they have no scripts
10798 * in common, not a run */
10799 if (intersection_len == 0) {
10804 /* If there is only a single script in common, set to that.
10805 * Otherwise, use the intersection going forward */
10806 Safefree(intersection);
10807 intersection = NULL;
10808 if (intersection_len == 1) {
10809 script_of_run = script_of_char = new_overlap[0];
10810 Safefree(new_overlap);
10811 new_overlap = NULL;
10814 intersection = new_overlap;
10822 /* Here, the script of the character is compatible with that of the
10823 * run. That means that in most cases, it continues the script run.
10824 * Either it and the run match exactly, or one or both can be in any of
10825 * several scripts, and the intersection is not empty. However, if the
10826 * character is a decimal digit, it could still mean failure if it is
10827 * from the wrong sequence of 10. So, we need to look at if it's a
10828 * digit. We've already handled the 10 decimal digits, and the next
10829 * lowest one is this one: */
10830 if (cp < FIRST_NON_ASCII_DECIMAL_DIGIT) {
10831 continue; /* Not a digit; this character is part of the run */
10834 /* If we have a definitive '0' for the script of this character, we
10835 * know that for this to be a digit, it must be in the range of +0..+9
10837 if ( script_of_char >= 0
10838 && (zero_of_char = script_zeros[script_of_char]))
10840 if ( cp < zero_of_char
10841 || cp > zero_of_char + 9)
10843 continue; /* Not a digit; this character is part of the run
10848 else { /* Need to look up if this character is a digit or not */
10849 SSize_t index_of_zero_of_char;
10850 index_of_zero_of_char = _invlist_search(decimals_invlist, cp);
10851 if ( UNLIKELY(index_of_zero_of_char < 0)
10852 || ! ELEMENT_RANGE_MATCHES_INVLIST(index_of_zero_of_char))
10854 continue; /* Not a digit; this character is part of the run.
10858 zero_of_char = decimals_array[index_of_zero_of_char];
10861 /* Here, the character is a decimal digit, and the zero of its sequence
10862 * of 10 is in 'zero_of_char'. If we already have a zero for this run,
10863 * they better be the same. */
10865 if (zero_of_run != zero_of_char) {
10870 else { /* Otherwise we now have a zero for this run */
10871 zero_of_run = zero_of_char;
10873 } /* end of looping through CLOSESR text */
10875 Safefree(intersection);
10877 if (ret_script != NULL) {
10879 *ret_script = script_of_run;
10882 *ret_script = SCX_INVALID;
10889 #endif /* ifndef PERL_IN_XSUB_RE */
10892 * ex: set ts=8 sts=4 sw=4 et: