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 ( (U8) NATIVE_UTF8_TO_I8(*s) >= ANYOF_FLAGS(c)
2183 && reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target)));
2188 if (utf8_target) { /* Can't possibly match a non-UTF-8 target */
2190 /* We know what the first byte of any matched string should be */
2191 U8 first_byte = FLAGS(c);
2193 REXEC_FBC_FIND_NEXT_UTF8_BYTE_SCAN(first_byte,
2194 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
2199 if (utf8_target) { /* Can't possibly match a non-UTF-8 target */
2200 REXEC_FBC_CLASS_SCAN(TRUE,
2201 ( inRANGE((U8) NATIVE_UTF8_TO_I8(*s),
2202 LOWEST_ANYOF_HRx_BYTE(ANYOF_FLAGS(c)),
2203 HIGHEST_ANYOF_HRx_BYTE(ANYOF_FLAGS(c)))
2204 && reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target)));
2208 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
2209 assert(! is_utf8_pat);
2213 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII
2214 |FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
2215 goto do_exactf_utf8;
2217 else if (utf8_target) {
2219 /* Here, and elsewhere in this file, the reason we can't consider a
2220 * non-UTF-8 pattern already folded in the presence of a UTF-8
2221 * target is because any MICRO SIGN in the pattern won't be folded.
2222 * Since the fold of the MICRO SIGN requires UTF-8 to represent, we
2223 * can consider a non-UTF-8 pattern folded when matching a
2224 * non-UTF-8 target */
2225 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
2226 goto do_exactf_utf8;
2229 /* Latin1 folds are not affected by /a, except it excludes the sharp s,
2230 * which these functions don't handle anyway */
2231 fold_array = PL_fold_latin1;
2232 folder = foldEQ_latin1_s2_folded;
2233 goto do_exactf_non_utf8;
2235 case EXACTF: /* This node only generated for non-utf8 patterns */
2236 assert(! is_utf8_pat);
2238 goto do_exactf_utf8;
2240 fold_array = PL_fold;
2242 goto do_exactf_non_utf8;
2245 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2246 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
2247 utf8_fold_flags = FOLDEQ_LOCALE;
2248 goto do_exactf_utf8;
2250 fold_array = PL_fold_locale;
2251 folder = foldEQ_locale;
2252 goto do_exactf_non_utf8;
2254 case EXACTFUP: /* Problematic even though pattern isn't UTF-8. Use
2255 full functionality normally not done except for
2257 assert(! is_utf8_pat);
2258 goto do_exactf_utf8;
2261 if (! utf8_target) { /* All code points in this node require
2262 UTF-8 to express. */
2265 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
2266 | FOLDEQ_S2_FOLDS_SANE;
2267 goto do_exactf_utf8;
2270 if (! utf8_target) {
2273 assert(is_utf8_pat);
2274 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
2275 goto do_exactf_utf8;
2278 if (is_utf8_pat || utf8_target) {
2279 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
2280 goto do_exactf_utf8;
2283 /* Any 'ss' in the pattern should have been replaced by regcomp,
2284 * so we don't have to worry here about this single special case
2285 * in the Latin1 range */
2286 fold_array = PL_fold_latin1;
2287 folder = foldEQ_latin1_s2_folded;
2291 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
2292 are no glitches with fold-length differences
2293 between the target string and pattern */
2295 /* The idea in the non-utf8 EXACTF* cases is to first find the
2296 * first character of the EXACTF* node and then, if necessary,
2297 * case-insensitively compare the full text of the node. c1 is the
2298 * first character. c2 is its fold. This logic will not work for
2299 * Unicode semantics and the german sharp ss, which hence should
2300 * not be compiled into a node that gets here. */
2301 pat_string = STRING(c);
2302 ln = STR_LEN(c); /* length to match in octets/bytes */
2304 /* We know that we have to match at least 'ln' bytes (which is the
2305 * same as characters, since not utf8). If we have to match 3
2306 * characters, and there are only 2 availabe, we know without
2307 * trying that it will fail; so don't start a match past the
2308 * required minimum number from the far end */
2309 e = HOP3c(strend, -((SSize_t)ln), s);
2314 c2 = fold_array[c1];
2315 if (c1 == c2) { /* If char and fold are the same */
2317 s = (char *) memchr(s, c1, e + 1 - s);
2322 /* Check that the rest of the node matches */
2323 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2324 && (reginfo->intuit || regtry(reginfo, &s)) )
2332 U8 bits_differing = c1 ^ c2;
2334 /* If the folds differ in one bit position only, we can mask to
2335 * match either of them, and can use this faster find method. Both
2336 * ASCII and EBCDIC tend to have their case folds differ in only
2337 * one position, so this is very likely */
2338 if (LIKELY(PL_bitcount[bits_differing] == 1)) {
2339 bits_differing = ~ bits_differing;
2341 s = (char *) find_next_masked((U8 *) s, (U8 *) e + 1,
2342 (c1 & bits_differing), bits_differing);
2347 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2348 && (reginfo->intuit || regtry(reginfo, &s)) )
2355 else { /* Otherwise, stuck with looking byte-at-a-time. This
2356 should actually happen only in EXACTFL nodes */
2358 if ( (*(U8*)s == c1 || *(U8*)s == c2)
2359 && (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2360 && (reginfo->intuit || regtry(reginfo, &s)) )
2374 /* If one of the operands is in utf8, we can't use the simpler folding
2375 * above, due to the fact that many different characters can have the
2376 * same fold, or portion of a fold, or different- length fold */
2377 pat_string = STRING(c);
2378 ln = STR_LEN(c); /* length to match in octets/bytes */
2379 pat_end = pat_string + ln;
2380 lnc = is_utf8_pat /* length to match in characters */
2381 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
2384 /* We have 'lnc' characters to match in the pattern, but because of
2385 * multi-character folding, each character in the target can match
2386 * up to 3 characters (Unicode guarantees it will never exceed
2387 * this) if it is utf8-encoded; and up to 2 if not (based on the
2388 * fact that the Latin 1 folds are already determined, and the
2389 * only multi-char fold in that range is the sharp-s folding to
2390 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
2391 * string character. Adjust lnc accordingly, rounding up, so that
2392 * if we need to match at least 4+1/3 chars, that really is 5. */
2393 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
2394 lnc = (lnc + expansion - 1) / expansion;
2396 /* As in the non-UTF8 case, if we have to match 3 characters, and
2397 * only 2 are left, it's guaranteed to fail, so don't start a
2398 * match that would require us to go beyond the end of the string
2400 e = HOP3c(strend, -((SSize_t)lnc), s);
2402 /* XXX Note that we could recalculate e to stop the loop earlier,
2403 * as the worst case expansion above will rarely be met, and as we
2404 * go along we would usually find that e moves further to the left.
2405 * This would happen only after we reached the point in the loop
2406 * where if there were no expansion we should fail. Unclear if
2407 * worth the expense */
2410 char *my_strend= (char *)strend;
2411 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
2412 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
2413 && (reginfo->intuit || regtry(reginfo, &s)) )
2417 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2423 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2424 if (FLAGS(c) != TRADITIONAL_BOUND) {
2425 if (! IN_UTF8_CTYPE_LOCALE) {
2426 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2427 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2432 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2436 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2437 if (FLAGS(c) != TRADITIONAL_BOUND) {
2438 if (! IN_UTF8_CTYPE_LOCALE) {
2439 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2440 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2445 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2448 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2450 assert(FLAGS(c) == TRADITIONAL_BOUND);
2452 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2455 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2457 assert(FLAGS(c) == TRADITIONAL_BOUND);
2459 FBC_BOUND_A(isWORDCHAR_A);
2462 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2464 assert(FLAGS(c) == TRADITIONAL_BOUND);
2466 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2469 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2471 assert(FLAGS(c) == TRADITIONAL_BOUND);
2473 FBC_NBOUND_A(isWORDCHAR_A);
2477 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
2478 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2489 switch((bound_type) FLAGS(c)) {
2490 case TRADITIONAL_BOUND:
2491 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2494 if (s == reginfo->strbeg) {
2495 if (reginfo->intuit || regtry(reginfo, &s))
2500 /* Didn't match. Try at the next position (if there is one) */
2501 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2502 if (UNLIKELY(s >= reginfo->strend)) {
2508 GCB_enum before = getGCB_VAL_UTF8(
2510 (U8*)(reginfo->strbeg)),
2511 (U8*) reginfo->strend);
2512 while (s < strend) {
2513 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
2514 (U8*) reginfo->strend);
2515 if ( (to_complement ^ isGCB(before,
2517 (U8*) reginfo->strbeg,
2520 && (reginfo->intuit || regtry(reginfo, &s)))
2525 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2528 else { /* Not utf8. Everything is a GCB except between CR and
2530 while (s < strend) {
2531 if ((to_complement ^ ( UCHARAT(s - 1) != '\r'
2532 || UCHARAT(s) != '\n'))
2533 && (reginfo->intuit || regtry(reginfo, &s)))
2541 /* And, since this is a bound, it can match after the final
2542 * character in the string */
2543 if ( reginfo->intuit
2544 || (s <= reginfo->strend && regtry(reginfo, &s)))
2551 if (s == reginfo->strbeg) {
2552 if (reginfo->intuit || regtry(reginfo, &s)) {
2555 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2556 if (UNLIKELY(s >= reginfo->strend)) {
2562 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
2564 (U8*)(reginfo->strbeg)),
2565 (U8*) reginfo->strend);
2566 while (s < strend) {
2567 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
2568 if (to_complement ^ isLB(before,
2570 (U8*) reginfo->strbeg,
2572 (U8*) reginfo->strend,
2574 && (reginfo->intuit || regtry(reginfo, &s)))
2579 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2582 else { /* Not utf8. */
2583 LB_enum before = getLB_VAL_CP((U8) *(s -1));
2584 while (s < strend) {
2585 LB_enum after = getLB_VAL_CP((U8) *s);
2586 if (to_complement ^ isLB(before,
2588 (U8*) reginfo->strbeg,
2590 (U8*) reginfo->strend,
2592 && (reginfo->intuit || regtry(reginfo, &s)))
2601 if ( reginfo->intuit
2602 || (s <= reginfo->strend && regtry(reginfo, &s)))
2610 if (s == reginfo->strbeg) {
2611 if (reginfo->intuit || regtry(reginfo, &s)) {
2614 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2615 if (UNLIKELY(s >= reginfo->strend)) {
2621 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
2623 (U8*)(reginfo->strbeg)),
2624 (U8*) reginfo->strend);
2625 while (s < strend) {
2626 SB_enum after = getSB_VAL_UTF8((U8*) s,
2627 (U8*) reginfo->strend);
2628 if ((to_complement ^ isSB(before,
2630 (U8*) reginfo->strbeg,
2632 (U8*) reginfo->strend,
2634 && (reginfo->intuit || regtry(reginfo, &s)))
2639 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2642 else { /* Not utf8. */
2643 SB_enum before = getSB_VAL_CP((U8) *(s -1));
2644 while (s < strend) {
2645 SB_enum after = getSB_VAL_CP((U8) *s);
2646 if ((to_complement ^ isSB(before,
2648 (U8*) reginfo->strbeg,
2650 (U8*) reginfo->strend,
2652 && (reginfo->intuit || regtry(reginfo, &s)))
2661 /* Here are at the final position in the target string. The SB
2662 * value is always true here, so matches, depending on other
2664 if ( reginfo->intuit
2665 || (s <= reginfo->strend && regtry(reginfo, &s)))
2673 if (s == reginfo->strbeg) {
2674 if (reginfo->intuit || regtry(reginfo, &s)) {
2677 s += (utf8_target) ? UTF8_SAFE_SKIP(s, reginfo->strend) : 1;
2678 if (UNLIKELY(s >= reginfo->strend)) {
2684 /* We are at a boundary between char_sub_0 and char_sub_1.
2685 * We also keep track of the value for char_sub_-1 as we
2686 * loop through the line. Context may be needed to make a
2687 * determination, and if so, this can save having to
2689 WB_enum previous = WB_UNKNOWN;
2690 WB_enum before = getWB_VAL_UTF8(
2693 (U8*)(reginfo->strbeg)),
2694 (U8*) reginfo->strend);
2695 while (s < strend) {
2696 WB_enum after = getWB_VAL_UTF8((U8*) s,
2697 (U8*) reginfo->strend);
2698 if ((to_complement ^ isWB(previous,
2701 (U8*) reginfo->strbeg,
2703 (U8*) reginfo->strend,
2705 && (reginfo->intuit || regtry(reginfo, &s)))
2711 s += UTF8_SAFE_SKIP(s, reginfo->strend);
2714 else { /* Not utf8. */
2715 WB_enum previous = WB_UNKNOWN;
2716 WB_enum before = getWB_VAL_CP((U8) *(s -1));
2717 while (s < strend) {
2718 WB_enum after = getWB_VAL_CP((U8) *s);
2719 if ((to_complement ^ isWB(previous,
2722 (U8*) reginfo->strbeg,
2724 (U8*) reginfo->strend,
2726 && (reginfo->intuit || regtry(reginfo, &s)))
2736 if ( reginfo->intuit
2737 || (s <= reginfo->strend && regtry(reginfo, &s)))
2745 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2746 is_LNBREAK_latin1_safe(s, strend)
2750 /* The argument to all the POSIX node types is the class number to pass to
2751 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2758 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2759 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s, (U8 *) strend)),
2760 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
2775 /* The complement of something that matches only ASCII matches all
2776 * non-ASCII, plus everything in ASCII that isn't in the class. */
2777 REXEC_FBC_CLASS_SCAN(1, ! isASCII_utf8_safe(s, strend)
2778 || ! _generic_isCC_A(*s, FLAGS(c)));
2786 /* Don't need to worry about utf8, as it can match only a single
2787 * byte invariant character. But we do anyway for performance reasons,
2788 * as otherwise we would have to examine all the continuation
2791 REXEC_FBC_CLASS_SCAN(1, _generic_isCC_A(*s, FLAGS(c)));
2796 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
2797 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
2805 if (! utf8_target) {
2806 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
2807 to_complement ^ cBOOL(_generic_isCC(*s,
2813 classnum = (_char_class_number) FLAGS(c);
2816 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2817 to_complement ^ cBOOL(_invlist_contains_cp(
2818 PL_XPosix_ptrs[classnum],
2819 utf8_to_uvchr_buf((U8 *) s,
2823 case _CC_ENUM_SPACE:
2824 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2825 to_complement ^ cBOOL(isSPACE_utf8_safe(s, strend)));
2828 case _CC_ENUM_BLANK:
2829 REXEC_FBC_CLASS_SCAN(1,
2830 to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
2833 case _CC_ENUM_XDIGIT:
2834 REXEC_FBC_CLASS_SCAN(1,
2835 to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
2838 case _CC_ENUM_VERTSPACE:
2839 REXEC_FBC_CLASS_SCAN(1,
2840 to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
2843 case _CC_ENUM_CNTRL:
2844 REXEC_FBC_CLASS_SCAN(1,
2845 to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
2855 /* what trie are we using right now */
2856 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2857 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2858 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2860 const char *last_start = strend - trie->minlen;
2862 const char *real_start = s;
2864 STRLEN maxlen = trie->maxlen;
2866 U8 **points; /* map of where we were in the input string
2867 when reading a given char. For ASCII this
2868 is unnecessary overhead as the relationship
2869 is always 1:1, but for Unicode, especially
2870 case folded Unicode this is not true. */
2871 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2875 GET_RE_DEBUG_FLAGS_DECL;
2877 /* We can't just allocate points here. We need to wrap it in
2878 * an SV so it gets freed properly if there is a croak while
2879 * running the match */
2882 sv_points=newSV(maxlen * sizeof(U8 *));
2883 SvCUR_set(sv_points,
2884 maxlen * sizeof(U8 *));
2885 SvPOK_on(sv_points);
2886 sv_2mortal(sv_points);
2887 points=(U8**)SvPV_nolen(sv_points );
2888 if ( trie_type != trie_utf8_fold
2889 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2892 bitmap=(U8*)trie->bitmap;
2894 bitmap=(U8*)ANYOF_BITMAP(c);
2896 /* this is the Aho-Corasick algorithm modified a touch
2897 to include special handling for long "unknown char" sequences.
2898 The basic idea being that we use AC as long as we are dealing
2899 with a possible matching char, when we encounter an unknown char
2900 (and we have not encountered an accepting state) we scan forward
2901 until we find a legal starting char.
2902 AC matching is basically that of trie matching, except that when
2903 we encounter a failing transition, we fall back to the current
2904 states "fail state", and try the current char again, a process
2905 we repeat until we reach the root state, state 1, or a legal
2906 transition. If we fail on the root state then we can either
2907 terminate if we have reached an accepting state previously, or
2908 restart the entire process from the beginning if we have not.
2911 while (s <= last_start) {
2912 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2920 U8 *uscan = (U8*)NULL;
2921 U8 *leftmost = NULL;
2923 U32 accepted_word= 0;
2927 while ( state && uc <= (U8*)strend ) {
2929 U32 word = aho->states[ state ].wordnum;
2933 DEBUG_TRIE_EXECUTE_r(
2934 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2935 dump_exec_pos( (char *)uc, c, strend, real_start,
2936 (char *)uc, utf8_target, 0 );
2937 Perl_re_printf( aTHX_
2938 " Scanning for legal start char...\n");
2942 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2946 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2952 if (uc >(U8*)last_start) break;
2956 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2957 if (!leftmost || lpos < leftmost) {
2958 DEBUG_r(accepted_word=word);
2964 points[pointpos++ % maxlen]= uc;
2965 if (foldlen || uc < (U8*)strend) {
2966 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
2967 (U8 *) strend, uscan, len, uvc,
2968 charid, foldlen, foldbuf,
2970 DEBUG_TRIE_EXECUTE_r({
2971 dump_exec_pos( (char *)uc, c, strend,
2972 real_start, s, utf8_target, 0);
2973 Perl_re_printf( aTHX_
2974 " Charid:%3u CP:%4" UVxf " ",
2986 word = aho->states[ state ].wordnum;
2988 base = aho->states[ state ].trans.base;
2990 DEBUG_TRIE_EXECUTE_r({
2992 dump_exec_pos( (char *)uc, c, strend, real_start,
2993 s, utf8_target, 0 );
2994 Perl_re_printf( aTHX_
2995 "%sState: %4" UVxf ", word=%" UVxf,
2996 failed ? " Fail transition to " : "",
2997 (UV)state, (UV)word);
3003 ( ((offset = base + charid
3004 - 1 - trie->uniquecharcount)) >= 0)
3005 && ((U32)offset < trie->lasttrans)
3006 && trie->trans[offset].check == state
3007 && (tmp=trie->trans[offset].next))
3009 DEBUG_TRIE_EXECUTE_r(
3010 Perl_re_printf( aTHX_ " - legal\n"));
3015 DEBUG_TRIE_EXECUTE_r(
3016 Perl_re_printf( aTHX_ " - fail\n"));
3018 state = aho->fail[state];
3022 /* we must be accepting here */
3023 DEBUG_TRIE_EXECUTE_r(
3024 Perl_re_printf( aTHX_ " - accepting\n"));
3033 if (!state) state = 1;
3036 if ( aho->states[ state ].wordnum ) {
3037 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
3038 if (!leftmost || lpos < leftmost) {
3039 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
3044 s = (char*)leftmost;
3045 DEBUG_TRIE_EXECUTE_r({
3046 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
3047 (UV)accepted_word, (IV)(s - real_start)
3050 if (reginfo->intuit || regtry(reginfo, &s)) {
3055 if (s < reginfo->strend) {
3058 DEBUG_TRIE_EXECUTE_r({
3059 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
3062 DEBUG_TRIE_EXECUTE_r(
3063 Perl_re_printf( aTHX_ "No match.\n"));
3072 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
3079 /* set RX_SAVED_COPY, RX_SUBBEG etc.
3080 * flags have same meanings as with regexec_flags() */
3083 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
3090 struct regexp *const prog = ReANY(rx);
3092 if (flags & REXEC_COPY_STR) {
3095 DEBUG_C(Perl_re_printf( aTHX_
3096 "Copy on write: regexp capture, type %d\n",
3098 /* Create a new COW SV to share the match string and store
3099 * in saved_copy, unless the current COW SV in saved_copy
3100 * is valid and suitable for our purpose */
3101 if (( prog->saved_copy
3102 && SvIsCOW(prog->saved_copy)
3103 && SvPOKp(prog->saved_copy)
3106 && SvPVX(sv) == SvPVX(prog->saved_copy)))
3108 /* just reuse saved_copy SV */
3109 if (RXp_MATCH_COPIED(prog)) {
3110 Safefree(prog->subbeg);
3111 RXp_MATCH_COPIED_off(prog);
3115 /* create new COW SV to share string */
3116 RXp_MATCH_COPY_FREE(prog);
3117 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
3119 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
3120 assert (SvPOKp(prog->saved_copy));
3121 prog->sublen = strend - strbeg;
3122 prog->suboffset = 0;
3123 prog->subcoffset = 0;
3128 SSize_t max = strend - strbeg;
3131 if ( (flags & REXEC_COPY_SKIP_POST)
3132 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
3133 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
3134 ) { /* don't copy $' part of string */
3137 /* calculate the right-most part of the string covered
3138 * by a capture. Due to lookahead, this may be to
3139 * the right of $&, so we have to scan all captures */
3140 while (n <= prog->lastparen) {
3141 if (prog->offs[n].end > max)
3142 max = prog->offs[n].end;
3146 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
3147 ? prog->offs[0].start
3149 assert(max >= 0 && max <= strend - strbeg);
3152 if ( (flags & REXEC_COPY_SKIP_PRE)
3153 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
3154 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
3155 ) { /* don't copy $` part of string */
3158 /* calculate the left-most part of the string covered
3159 * by a capture. Due to lookbehind, this may be to
3160 * the left of $&, so we have to scan all captures */
3161 while (min && n <= prog->lastparen) {
3162 if ( prog->offs[n].start != -1
3163 && prog->offs[n].start < min)
3165 min = prog->offs[n].start;
3169 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
3170 && min > prog->offs[0].end
3172 min = prog->offs[0].end;
3176 assert(min >= 0 && min <= max && min <= strend - strbeg);
3179 if (RXp_MATCH_COPIED(prog)) {
3180 if (sublen > prog->sublen)
3182 (char*)saferealloc(prog->subbeg, sublen+1);
3185 prog->subbeg = (char*)safemalloc(sublen+1);
3186 Copy(strbeg + min, prog->subbeg, sublen, char);
3187 prog->subbeg[sublen] = '\0';
3188 prog->suboffset = min;
3189 prog->sublen = sublen;
3190 RXp_MATCH_COPIED_on(prog);
3192 prog->subcoffset = prog->suboffset;
3193 if (prog->suboffset && utf8_target) {
3194 /* Convert byte offset to chars.
3195 * XXX ideally should only compute this if @-/@+
3196 * has been seen, a la PL_sawampersand ??? */
3198 /* If there's a direct correspondence between the
3199 * string which we're matching and the original SV,
3200 * then we can use the utf8 len cache associated with
3201 * the SV. In particular, it means that under //g,
3202 * sv_pos_b2u() will use the previously cached
3203 * position to speed up working out the new length of
3204 * subcoffset, rather than counting from the start of
3205 * the string each time. This stops
3206 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
3207 * from going quadratic */
3208 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
3209 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
3210 SV_GMAGIC|SV_CONST_RETURN);
3212 prog->subcoffset = utf8_length((U8*)strbeg,
3213 (U8*)(strbeg+prog->suboffset));
3217 RXp_MATCH_COPY_FREE(prog);
3218 prog->subbeg = strbeg;
3219 prog->suboffset = 0;
3220 prog->subcoffset = 0;
3221 prog->sublen = strend - strbeg;
3229 - regexec_flags - match a regexp against a string
3232 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
3233 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
3234 /* stringarg: the point in the string at which to begin matching */
3235 /* strend: pointer to null at end of string */
3236 /* strbeg: real beginning of string */
3237 /* minend: end of match must be >= minend bytes after stringarg. */
3238 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
3239 * itself is accessed via the pointers above */
3240 /* data: May be used for some additional optimizations.
3241 Currently unused. */
3242 /* flags: For optimizations. See REXEC_* in regexp.h */
3245 struct regexp *const prog = ReANY(rx);
3249 SSize_t minlen; /* must match at least this many chars */
3250 SSize_t dontbother = 0; /* how many characters not to try at end */
3251 const bool utf8_target = cBOOL(DO_UTF8(sv));
3253 RXi_GET_DECL(prog,progi);
3254 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
3255 regmatch_info *const reginfo = ®info_buf;
3256 regexp_paren_pair *swap = NULL;
3258 GET_RE_DEBUG_FLAGS_DECL;
3260 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
3261 PERL_UNUSED_ARG(data);
3263 /* Be paranoid... */
3265 Perl_croak(aTHX_ "NULL regexp parameter");
3269 debug_start_match(rx, utf8_target, stringarg, strend,
3273 startpos = stringarg;
3275 /* set these early as they may be used by the HOP macros below */
3276 reginfo->strbeg = strbeg;
3277 reginfo->strend = strend;
3278 reginfo->is_utf8_target = cBOOL(utf8_target);
3280 if (prog->intflags & PREGf_GPOS_SEEN) {
3283 /* set reginfo->ganch, the position where \G can match */
3286 (flags & REXEC_IGNOREPOS)
3287 ? stringarg /* use start pos rather than pos() */
3288 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
3289 /* Defined pos(): */
3290 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
3291 : strbeg; /* pos() not defined; use start of string */
3293 DEBUG_GPOS_r(Perl_re_printf( aTHX_
3294 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
3296 /* in the presence of \G, we may need to start looking earlier in
3297 * the string than the suggested start point of stringarg:
3298 * if prog->gofs is set, then that's a known, fixed minimum
3301 * /ab|c\G/: gofs = 1
3302 * or if the minimum offset isn't known, then we have to go back
3303 * to the start of the string, e.g. /w+\G/
3306 if (prog->intflags & PREGf_ANCH_GPOS) {
3308 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
3310 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
3312 DEBUG_r(Perl_re_printf( aTHX_
3313 "fail: ganch-gofs before earliest possible start\n"));
3318 startpos = reginfo->ganch;
3320 else if (prog->gofs) {
3321 startpos = HOPBACKc(startpos, prog->gofs);
3325 else if (prog->intflags & PREGf_GPOS_FLOAT)
3329 minlen = prog->minlen;
3330 if ((startpos + minlen) > strend || startpos < strbeg) {
3331 DEBUG_r(Perl_re_printf( aTHX_
3332 "Regex match can't succeed, so not even tried\n"));
3336 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
3337 * which will call destuctors to reset PL_regmatch_state, free higher
3338 * PL_regmatch_slabs, and clean up regmatch_info_aux and
3339 * regmatch_info_aux_eval */
3341 oldsave = PL_savestack_ix;
3345 if ((prog->extflags & RXf_USE_INTUIT)
3346 && !(flags & REXEC_CHECKED))
3348 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
3353 if (prog->extflags & RXf_CHECK_ALL) {
3354 /* we can match based purely on the result of INTUIT.
3355 * Set up captures etc just for $& and $-[0]
3356 * (an intuit-only match wont have $1,$2,..) */
3357 assert(!prog->nparens);
3359 /* s/// doesn't like it if $& is earlier than where we asked it to
3360 * start searching (which can happen on something like /.\G/) */
3361 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3364 /* this should only be possible under \G */
3365 assert(prog->intflags & PREGf_GPOS_SEEN);
3366 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3367 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3371 /* match via INTUIT shouldn't have any captures.
3372 * Let @-, @+, $^N know */
3373 prog->lastparen = prog->lastcloseparen = 0;
3374 RXp_MATCH_UTF8_set(prog, utf8_target);
3375 prog->offs[0].start = s - strbeg;
3376 prog->offs[0].end = utf8_target
3377 ? (char*)utf8_hop_forward((U8*)s, prog->minlenret, (U8 *) strend) - strbeg
3378 : s - strbeg + prog->minlenret;
3379 if ( !(flags & REXEC_NOT_FIRST) )
3380 S_reg_set_capture_string(aTHX_ rx,
3382 sv, flags, utf8_target);
3388 multiline = prog->extflags & RXf_PMf_MULTILINE;
3390 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
3391 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3392 "String too short [regexec_flags]...\n"));
3396 /* Check validity of program. */
3397 if (UCHARAT(progi->program) != REG_MAGIC) {
3398 Perl_croak(aTHX_ "corrupted regexp program");
3401 RXp_MATCH_TAINTED_off(prog);
3402 RXp_MATCH_UTF8_set(prog, utf8_target);
3404 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3405 reginfo->intuit = 0;
3406 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3407 reginfo->warned = FALSE;
3409 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
3410 /* see how far we have to get to not match where we matched before */
3411 reginfo->till = stringarg + minend;
3413 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
3414 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3415 S_cleanup_regmatch_info_aux has executed (registered by
3416 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3417 magic belonging to this SV.
3418 Not newSVsv, either, as it does not COW.
3420 reginfo->sv = newSV(0);
3421 SvSetSV_nosteal(reginfo->sv, sv);
3422 SAVEFREESV(reginfo->sv);
3425 /* reserve next 2 or 3 slots in PL_regmatch_state:
3426 * slot N+0: may currently be in use: skip it
3427 * slot N+1: use for regmatch_info_aux struct
3428 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3429 * slot N+3: ready for use by regmatch()
3433 regmatch_state *old_regmatch_state;
3434 regmatch_slab *old_regmatch_slab;
3435 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3437 /* on first ever match, allocate first slab */
3438 if (!PL_regmatch_slab) {
3439 Newx(PL_regmatch_slab, 1, regmatch_slab);
3440 PL_regmatch_slab->prev = NULL;
3441 PL_regmatch_slab->next = NULL;
3442 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3445 old_regmatch_state = PL_regmatch_state;
3446 old_regmatch_slab = PL_regmatch_slab;
3448 for (i=0; i <= max; i++) {
3450 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3452 reginfo->info_aux_eval =
3453 reginfo->info_aux->info_aux_eval =
3454 &(PL_regmatch_state->u.info_aux_eval);
3456 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3457 PL_regmatch_state = S_push_slab(aTHX);
3460 /* note initial PL_regmatch_state position; at end of match we'll
3461 * pop back to there and free any higher slabs */
3463 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3464 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
3465 reginfo->info_aux->poscache = NULL;
3467 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
3469 if ((prog->extflags & RXf_EVAL_SEEN))
3470 S_setup_eval_state(aTHX_ reginfo);
3472 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
3475 /* If there is a "must appear" string, look for it. */
3477 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
3478 /* We have to be careful. If the previous successful match
3479 was from this regex we don't want a subsequent partially
3480 successful match to clobber the old results.
3481 So when we detect this possibility we add a swap buffer
3482 to the re, and switch the buffer each match. If we fail,
3483 we switch it back; otherwise we leave it swapped.
3486 /* avoid leak if we die, or clean up anyway if match completes */
3488 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
3489 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3490 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
3498 if (prog->recurse_locinput)
3499 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3501 /* Simplest case: anchored match need be tried only once, or with
3502 * MBOL, only at the beginning of each line.
3504 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3505 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3506 * match at the start of the string then it won't match anywhere else
3507 * either; while with /.*.../, if it doesn't match at the beginning,
3508 * the earliest it could match is at the start of the next line */
3510 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3513 if (regtry(reginfo, &s))
3516 if (!(prog->intflags & PREGf_ANCH_MBOL))
3519 /* didn't match at start, try at other newline positions */
3522 dontbother = minlen - 1;
3523 end = HOP3c(strend, -dontbother, strbeg) - 1;
3525 /* skip to next newline */
3527 while (s <= end) { /* note it could be possible to match at the end of the string */
3528 /* NB: newlines are the same in unicode as they are in latin */
3531 if (prog->check_substr || prog->check_utf8) {
3532 /* note that with PREGf_IMPLICIT, intuit can only fail
3533 * or return the start position, so it's of limited utility.
3534 * Nevertheless, I made the decision that the potential for
3535 * quick fail was still worth it - DAPM */
3536 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3540 if (regtry(reginfo, &s))
3544 } /* end anchored search */
3546 if (prog->intflags & PREGf_ANCH_GPOS)
3548 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3549 assert(prog->intflags & PREGf_GPOS_SEEN);
3550 /* For anchored \G, the only position it can match from is
3551 * (ganch-gofs); we already set startpos to this above; if intuit
3552 * moved us on from there, we can't possibly succeed */
3553 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
3554 if (s == startpos && regtry(reginfo, &s))
3559 /* Messy cases: unanchored match. */
3560 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
3561 /* we have /x+whatever/ */
3562 /* it must be a one character string (XXXX Except is_utf8_pat?) */
3568 if (! prog->anchored_utf8) {
3569 to_utf8_substr(prog);
3571 ch = SvPVX_const(prog->anchored_utf8)[0];
3572 REXEC_FBC_SCAN(0, /* 0=>not-utf8 */
3574 DEBUG_EXECUTE_r( did_match = 1 );
3575 if (regtry(reginfo, &s)) goto got_it;
3576 s += UTF8_SAFE_SKIP(s, strend);
3577 while (s < strend && *s == ch)
3584 if (! prog->anchored_substr) {
3585 if (! to_byte_substr(prog)) {
3586 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3589 ch = SvPVX_const(prog->anchored_substr)[0];
3590 REXEC_FBC_SCAN(0, /* 0=>not-utf8 */
3592 DEBUG_EXECUTE_r( did_match = 1 );
3593 if (regtry(reginfo, &s)) goto got_it;
3595 while (s < strend && *s == ch)
3600 DEBUG_EXECUTE_r(if (!did_match)
3601 Perl_re_printf( aTHX_
3602 "Did not find anchored character...\n")
3605 else if (prog->anchored_substr != NULL
3606 || prog->anchored_utf8 != NULL
3607 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
3608 && prog->float_max_offset < strend - s)) {
3613 char *last1; /* Last position checked before */
3617 if (prog->anchored_substr || prog->anchored_utf8) {
3619 if (! prog->anchored_utf8) {
3620 to_utf8_substr(prog);
3622 must = prog->anchored_utf8;
3625 if (! prog->anchored_substr) {
3626 if (! to_byte_substr(prog)) {
3627 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3630 must = prog->anchored_substr;
3632 back_max = back_min = prog->anchored_offset;
3635 if (! prog->float_utf8) {
3636 to_utf8_substr(prog);
3638 must = prog->float_utf8;
3641 if (! prog->float_substr) {
3642 if (! to_byte_substr(prog)) {
3643 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3646 must = prog->float_substr;
3648 back_max = prog->float_max_offset;
3649 back_min = prog->float_min_offset;
3655 last = HOP3c(strend, /* Cannot start after this */
3656 -(SSize_t)(CHR_SVLEN(must)
3657 - (SvTAIL(must) != 0) + back_min), strbeg);
3659 if (s > reginfo->strbeg)
3660 last1 = HOPc(s, -1);
3662 last1 = s - 1; /* bogus */
3664 /* XXXX check_substr already used to find "s", can optimize if
3665 check_substr==must. */
3667 strend = HOPc(strend, -dontbother);
3668 while ( (s <= last) &&
3669 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
3670 (unsigned char*)strend, must,
3671 multiline ? FBMrf_MULTILINE : 0)) ) {
3672 DEBUG_EXECUTE_r( did_match = 1 );
3673 if (HOPc(s, -back_max) > last1) {
3674 last1 = HOPc(s, -back_min);
3675 s = HOPc(s, -back_max);
3678 char * const t = (last1 >= reginfo->strbeg)
3679 ? HOPc(last1, 1) : last1 + 1;
3681 last1 = HOPc(s, -back_min);
3685 while (s <= last1) {
3686 if (regtry(reginfo, &s))
3689 s++; /* to break out of outer loop */
3696 while (s <= last1) {
3697 if (regtry(reginfo, &s))
3703 DEBUG_EXECUTE_r(if (!did_match) {
3704 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
3705 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
3706 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n",
3707 ((must == prog->anchored_substr || must == prog->anchored_utf8)
3708 ? "anchored" : "floating"),
3709 quoted, RE_SV_TAIL(must));
3713 else if ( (c = progi->regstclass) ) {
3715 const OPCODE op = OP(progi->regstclass);
3716 /* don't bother with what can't match */
3717 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
3718 strend = HOPc(strend, -(minlen - 1));
3721 SV * const prop = sv_newmortal();
3722 regprop(prog, prop, c, reginfo, NULL);
3724 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
3725 s,strend-s,PL_dump_re_max_len);
3726 Perl_re_printf( aTHX_
3727 "Matching stclass %.*s against %s (%d bytes)\n",
3728 (int)SvCUR(prop), SvPVX_const(prop),
3729 quoted, (int)(strend - s));
3732 if (find_byclass(prog, c, s, strend, reginfo))
3734 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n"));
3738 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3746 if (! prog->float_utf8) {
3747 to_utf8_substr(prog);
3749 float_real = prog->float_utf8;
3752 if (! prog->float_substr) {
3753 if (! to_byte_substr(prog)) {
3754 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3757 float_real = prog->float_substr;
3760 little = SvPV_const(float_real, len);
3761 if (SvTAIL(float_real)) {
3762 /* This means that float_real contains an artificial \n on
3763 * the end due to the presence of something like this:
3764 * /foo$/ where we can match both "foo" and "foo\n" at the
3765 * end of the string. So we have to compare the end of the
3766 * string first against the float_real without the \n and
3767 * then against the full float_real with the string. We
3768 * have to watch out for cases where the string might be
3769 * smaller than the float_real or the float_real without
3771 char *checkpos= strend - len;
3773 Perl_re_printf( aTHX_
3774 "%sChecking for float_real.%s\n",
3775 PL_colors[4], PL_colors[5]));
3776 if (checkpos + 1 < strbeg) {
3777 /* can't match, even if we remove the trailing \n
3778 * string is too short to match */
3780 Perl_re_printf( aTHX_
3781 "%sString shorter than required trailing substring, cannot match.%s\n",
3782 PL_colors[4], PL_colors[5]));
3784 } else if (memEQ(checkpos + 1, little, len - 1)) {
3785 /* can match, the end of the string matches without the
3787 last = checkpos + 1;
3788 } else if (checkpos < strbeg) {
3789 /* cant match, string is too short when the "\n" is
3792 Perl_re_printf( aTHX_
3793 "%sString does not contain required trailing substring, cannot match.%s\n",
3794 PL_colors[4], PL_colors[5]));
3796 } else if (!multiline) {
3797 /* non multiline match, so compare with the "\n" at the
3798 * end of the string */
3799 if (memEQ(checkpos, little, len)) {
3803 Perl_re_printf( aTHX_
3804 "%sString does not contain required trailing substring, cannot match.%s\n",
3805 PL_colors[4], PL_colors[5]));
3809 /* multiline match, so we have to search for a place
3810 * where the full string is located */
3816 last = rninstr(s, strend, little, little + len);
3818 last = strend; /* matching "$" */
3821 /* at one point this block contained a comment which was
3822 * probably incorrect, which said that this was a "should not
3823 * happen" case. Even if it was true when it was written I am
3824 * pretty sure it is not anymore, so I have removed the comment
3825 * and replaced it with this one. Yves */
3827 Perl_re_printf( aTHX_
3828 "%sString does not contain required substring, cannot match.%s\n",
3829 PL_colors[4], PL_colors[5]
3833 dontbother = strend - last + prog->float_min_offset;
3835 if (minlen && (dontbother < minlen))
3836 dontbother = minlen - 1;
3837 strend -= dontbother; /* this one's always in bytes! */
3838 /* We don't know much -- general case. */
3841 if (regtry(reginfo, &s))
3850 if (regtry(reginfo, &s))
3852 } while (s++ < strend);
3860 /* s/// doesn't like it if $& is earlier than where we asked it to
3861 * start searching (which can happen on something like /.\G/) */
3862 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3863 && (prog->offs[0].start < stringarg - strbeg))
3865 /* this should only be possible under \G */
3866 assert(prog->intflags & PREGf_GPOS_SEEN);
3867 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3868 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3872 /* clean up; this will trigger destructors that will free all slabs
3873 * above the current one, and cleanup the regmatch_info_aux
3874 * and regmatch_info_aux_eval sructs */
3876 LEAVE_SCOPE(oldsave);
3878 if (RXp_PAREN_NAMES(prog))
3879 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3881 /* make sure $`, $&, $', and $digit will work later */
3882 if ( !(flags & REXEC_NOT_FIRST) )
3883 S_reg_set_capture_string(aTHX_ rx,
3884 strbeg, reginfo->strend,
3885 sv, flags, utf8_target);
3890 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n",
3891 PL_colors[4], PL_colors[5]));
3894 /* we failed :-( roll it back.
3895 * Since the swap buffer will be freed on scope exit which follows
3896 * shortly, restore the old captures by copying 'swap's original
3897 * data to the new offs buffer
3899 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3900 "rex=0x%" UVxf " rolling back offs: 0x%" UVxf " will be freed; restoring data to =0x%" UVxf "\n",
3907 Copy(swap, prog->offs, prog->nparens + 1, regexp_paren_pair);
3910 /* clean up; this will trigger destructors that will free all slabs
3911 * above the current one, and cleanup the regmatch_info_aux
3912 * and regmatch_info_aux_eval sructs */
3914 LEAVE_SCOPE(oldsave);
3920 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3921 * Do inc before dec, in case old and new rex are the same */
3922 #define SET_reg_curpm(Re2) \
3923 if (reginfo->info_aux_eval) { \
3924 (void)ReREFCNT_inc(Re2); \
3925 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3926 PM_SETRE((PL_reg_curpm), (Re2)); \
3931 - regtry - try match at specific point
3933 STATIC bool /* 0 failure, 1 success */
3934 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3937 REGEXP *const rx = reginfo->prog;
3938 regexp *const prog = ReANY(rx);
3941 U32 depth = 0; /* used by REGCP_SET */
3943 RXi_GET_DECL(prog,progi);
3944 GET_RE_DEBUG_FLAGS_DECL;
3946 PERL_ARGS_ASSERT_REGTRY;
3948 reginfo->cutpoint=NULL;
3950 prog->offs[0].start = *startposp - reginfo->strbeg;
3951 prog->lastparen = 0;
3952 prog->lastcloseparen = 0;
3954 /* XXXX What this code is doing here?!!! There should be no need
3955 to do this again and again, prog->lastparen should take care of
3958 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3959 * Actually, the code in regcppop() (which Ilya may be meaning by
3960 * prog->lastparen), is not needed at all by the test suite
3961 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3962 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3963 * Meanwhile, this code *is* needed for the
3964 * above-mentioned test suite tests to succeed. The common theme
3965 * on those tests seems to be returning null fields from matches.
3966 * --jhi updated by dapm */
3968 /* After encountering a variant of the issue mentioned above I think
3969 * the point Ilya was making is that if we properly unwind whenever
3970 * we set lastparen to a smaller value then we should not need to do
3971 * this every time, only when needed. So if we have tests that fail if
3972 * we remove this, then it suggests somewhere else we are improperly
3973 * unwinding the lastparen/paren buffers. See UNWIND_PARENS() and
3974 * places it is called, and related regcp() routines. - Yves */
3976 if (prog->nparens) {
3977 regexp_paren_pair *pp = prog->offs;
3979 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3987 result = regmatch(reginfo, *startposp, progi->program + 1);
3989 prog->offs[0].end = result;
3992 if (reginfo->cutpoint)
3993 *startposp= reginfo->cutpoint;
3994 REGCP_UNWIND(lastcp);
3998 /* this is used to determine how far from the left messages like
3999 'failed...' are printed in regexec.c. It should be set such that
4000 messages are inline with the regop output that created them.
4002 #define REPORT_CODE_OFF 29
4003 #define INDENT_CHARS(depth) ((int)(depth) % 20)
4006 Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...)
4010 PerlIO *f= Perl_debug_log;
4011 PERL_ARGS_ASSERT_RE_EXEC_INDENTF;
4012 va_start(ap, depth);
4013 PerlIO_printf(f, "%*s|%4" UVuf "| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" );
4014 result = PerlIO_vprintf(f, fmt, ap);
4018 #endif /* DEBUGGING */
4020 /* grab a new slab and return the first slot in it */
4022 STATIC regmatch_state *
4025 regmatch_slab *s = PL_regmatch_slab->next;
4027 Newx(s, 1, regmatch_slab);
4028 s->prev = PL_regmatch_slab;
4030 PL_regmatch_slab->next = s;
4032 PL_regmatch_slab = s;
4033 return SLAB_FIRST(s);
4039 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
4040 const char *start, const char *end, const char *blurb)
4042 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
4044 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
4049 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
4050 RX_PRECOMP_const(prog), RX_PRELEN(prog), PL_dump_re_max_len);
4052 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
4053 start, end - start, PL_dump_re_max_len);
4055 Perl_re_printf( aTHX_
4056 "%s%s REx%s %s against %s\n",
4057 PL_colors[4], blurb, PL_colors[5], s0, s1);
4059 if (utf8_target||utf8_pat)
4060 Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n",
4061 utf8_pat ? "pattern" : "",
4062 utf8_pat && utf8_target ? " and " : "",
4063 utf8_target ? "string" : ""
4069 S_dump_exec_pos(pTHX_ const char *locinput,
4070 const regnode *scan,
4071 const char *loc_regeol,
4072 const char *loc_bostr,
4073 const char *loc_reg_starttry,
4074 const bool utf8_target,
4078 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
4079 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
4080 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
4081 /* The part of the string before starttry has one color
4082 (pref0_len chars), between starttry and current
4083 position another one (pref_len - pref0_len chars),
4084 after the current position the third one.
4085 We assume that pref0_len <= pref_len, otherwise we
4086 decrease pref0_len. */
4087 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
4088 ? (5 + taill) - l : locinput - loc_bostr;
4091 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
4093 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
4095 pref0_len = pref_len - (locinput - loc_reg_starttry);
4096 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
4097 l = ( loc_regeol - locinput > (5 + taill) - pref_len
4098 ? (5 + taill) - pref_len : loc_regeol - locinput);
4099 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
4103 if (pref0_len > pref_len)
4104 pref0_len = pref_len;
4106 const int is_uni = utf8_target ? 1 : 0;
4108 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
4109 (locinput - pref_len),pref0_len, PL_dump_re_max_len, 4, 5);
4111 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
4112 (locinput - pref_len + pref0_len),
4113 pref_len - pref0_len, PL_dump_re_max_len, 2, 3);
4115 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
4116 locinput, loc_regeol - locinput, 10, 0, 1);
4118 const STRLEN tlen=len0+len1+len2;
4119 Perl_re_printf( aTHX_
4120 "%4" IVdf " <%.*s%.*s%s%.*s>%*s|%4u| ",
4121 (IV)(locinput - loc_bostr),
4124 (docolor ? "" : "> <"),
4126 (int)(tlen > 19 ? 0 : 19 - tlen),
4134 /* reg_check_named_buff_matched()
4135 * Checks to see if a named buffer has matched. The data array of
4136 * buffer numbers corresponding to the buffer is expected to reside
4137 * in the regexp->data->data array in the slot stored in the ARG() of
4138 * node involved. Note that this routine doesn't actually care about the
4139 * name, that information is not preserved from compilation to execution.
4140 * Returns the index of the leftmost defined buffer with the given name
4141 * or 0 if non of the buffers matched.
4144 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
4147 RXi_GET_DECL(rex,rexi);
4148 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4149 I32 *nums=(I32*)SvPVX(sv_dat);
4151 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
4153 for ( n=0; n<SvIVX(sv_dat); n++ ) {
4154 if ((I32)rex->lastparen >= nums[n] &&
4155 rex->offs[nums[n]].end != -1)
4163 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
4164 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
4165 #define CHRTEST_NOT_A_CP_1 -999
4166 #define CHRTEST_NOT_A_CP_2 -998
4169 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
4170 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
4172 /* This function determines if there are zero, one, two, or more characters
4173 * that match the first character of the passed-in EXACTish node
4174 * <text_node>, and if there are one or two, it returns them in the
4175 * passed-in pointers.
4177 * If it determines that no possible character in the target string can
4178 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
4179 * the first character in <text_node> requires UTF-8 to represent, and the
4180 * target string isn't in UTF-8.)
4182 * If there are more than two characters that could match the beginning of
4183 * <text_node>, or if more context is required to determine a match or not,
4184 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
4186 * The motiviation behind this function is to allow the caller to set up
4187 * tight loops for matching. If <text_node> is of type EXACT, there is
4188 * only one possible character that can match its first character, and so
4189 * the situation is quite simple. But things get much more complicated if
4190 * folding is involved. It may be that the first character of an EXACTFish
4191 * node doesn't participate in any possible fold, e.g., punctuation, so it
4192 * can be matched only by itself. The vast majority of characters that are
4193 * in folds match just two things, their lower and upper-case equivalents.
4194 * But not all are like that; some have multiple possible matches, or match
4195 * sequences of more than one character. This function sorts all that out.
4197 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
4198 * loop of trying to match A*, we know we can't exit where the thing
4199 * following it isn't a B. And something can't be a B unless it is the
4200 * beginning of B. By putting a quick test for that beginning in a tight
4201 * loop, we can rule out things that can't possibly be B without having to
4202 * break out of the loop, thus avoiding work. Similarly, if A is a single
4203 * character, we can make a tight loop matching A*, using the outputs of
4206 * If the target string to match isn't in UTF-8, and there aren't
4207 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
4208 * the one or two possible octets (which are characters in this situation)
4209 * that can match. In all cases, if there is only one character that can
4210 * match, *<c1p> and *<c2p> will be identical.
4212 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
4213 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
4214 * can match the beginning of <text_node>. They should be declared with at
4215 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
4216 * undefined what these contain.) If one or both of the buffers are
4217 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
4218 * corresponding invariant. If variant, the corresponding *<c1p> and/or
4219 * *<c2p> will be set to a negative number(s) that shouldn't match any code
4220 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
4221 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
4223 const bool utf8_target = reginfo->is_utf8_target;
4225 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
4226 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
4227 bool use_chrtest_void = FALSE;
4228 const bool is_utf8_pat = reginfo->is_utf8_pat;
4230 /* Used when we have both utf8 input and utf8 output, to avoid converting
4231 * to/from code points */
4232 bool utf8_has_been_setup = FALSE;
4236 U8 *pat = (U8*)STRING(text_node);
4237 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
4239 if ( OP(text_node) == EXACT
4240 || OP(text_node) == EXACT_ONLY8
4241 || OP(text_node) == EXACTL)
4244 /* In an exact node, only one thing can be matched, that first
4245 * character. If both the pat and the target are UTF-8, we can just
4246 * copy the input to the output, avoiding finding the code point of
4249 assert(OP(text_node) != EXACT_ONLY8);
4252 else if (utf8_target) {
4253 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
4254 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
4255 utf8_has_been_setup = TRUE;
4257 else if (OP(text_node) == EXACT_ONLY8) {
4258 return FALSE; /* Can only match UTF-8 target */
4261 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
4264 else { /* an EXACTFish node */
4265 U8 *pat_end = pat + STR_LEN(text_node);
4267 /* An EXACTFL node has at least some characters unfolded, because what
4268 * they match is not known until now. So, now is the time to fold
4269 * the first few of them, as many as are needed to determine 'c1' and
4270 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
4271 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
4272 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
4273 * need to fold as many characters as a single character can fold to,
4274 * so that later we can check if the first ones are such a multi-char
4275 * fold. But, in such a pattern only locale-problematic characters
4276 * aren't folded, so we can skip this completely if the first character
4277 * in the node isn't one of the tricky ones */
4278 if (OP(text_node) == EXACTFL) {
4280 if (! is_utf8_pat) {
4281 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
4283 folded[0] = folded[1] = 's';
4285 pat_end = folded + 2;
4288 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
4293 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
4294 if (isASCII(*s) && LIKELY(! PL_in_utf8_turkic_locale)) {
4295 *(d++) = (U8) toFOLD_LC(*s);
4300 _toFOLD_utf8_flags(s,
4304 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
4315 if ( ( is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
4316 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
4318 /* Multi-character folds require more context to sort out. Also
4319 * PL_utf8_foldclosures used below doesn't handle them, so have to
4320 * be handled outside this routine */
4321 use_chrtest_void = TRUE;
4323 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
4324 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
4326 if ( UNLIKELY(PL_in_utf8_turkic_locale)
4327 && OP(text_node) == EXACTFL
4328 && UNLIKELY( c1 == 'i' || c1 == 'I'
4329 || c1 == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE
4330 || c1 == LATIN_SMALL_LETTER_DOTLESS_I))
4331 { /* Hard-coded Turkish locale rules for these 4 characters
4332 override normal rules */
4334 c2 = LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
4336 else if (c1 == 'I') {
4337 c2 = LATIN_SMALL_LETTER_DOTLESS_I;
4339 else if (c1 == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4342 else if (c1 == LATIN_SMALL_LETTER_DOTLESS_I) {
4346 else if (c1 > 255) {
4347 const unsigned int * remaining_folds;
4348 unsigned int first_fold;
4350 /* Look up what code points (besides c1) fold to c1; e.g.,
4351 * [ 'K', KELVIN_SIGN ] both fold to 'k'. */
4352 Size_t folds_count = _inverse_folds(c1, &first_fold,
4354 if (folds_count == 0) {
4355 c2 = c1; /* there is only a single character that could
4358 else if (folds_count != 1) {
4359 /* If there aren't exactly two folds to this (itself and
4360 * another), it is outside the scope of this function */
4361 use_chrtest_void = TRUE;
4363 else { /* There are two. We already have one, get the other */
4366 /* Folds that cross the 255/256 boundary are forbidden if
4367 * EXACTFL (and isnt a UTF8 locale), or EXACTFAA and one is
4368 * ASCIII. The only other match to c1 is c2, and since c1
4369 * is above 255, c2 better be as well under these
4370 * circumstances. If it isn't, it means the only legal
4371 * match of c1 is itself. */
4373 && ( ( OP(text_node) == EXACTFL
4374 && ! IN_UTF8_CTYPE_LOCALE)
4375 || (( OP(text_node) == EXACTFAA
4376 || OP(text_node) == EXACTFAA_NO_TRIE)
4377 && (isASCII(c1) || isASCII(c2)))))
4383 else /* Here, c1 is <= 255 */
4385 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
4386 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
4387 && ( ( OP(text_node) != EXACTFAA
4388 && OP(text_node) != EXACTFAA_NO_TRIE)
4391 /* Here, there could be something above Latin1 in the target
4392 * which folds to this character in the pattern. All such
4393 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
4394 * than two characters involved in their folds, so are outside
4395 * the scope of this function */
4396 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4397 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
4400 use_chrtest_void = TRUE;
4403 else { /* Here nothing above Latin1 can fold to the pattern
4405 switch (OP(text_node)) {
4407 case EXACTFL: /* /l rules */
4408 c2 = PL_fold_locale[c1];
4411 case EXACTF: /* This node only generated for non-utf8
4413 assert(! is_utf8_pat);
4414 if (! utf8_target) { /* /d rules */
4419 /* /u rules for all these. This happens to work for
4420 * EXACTFAA as nothing in Latin1 folds to ASCII */
4421 case EXACTFAA_NO_TRIE: /* This node only generated for
4422 non-utf8 patterns */
4423 assert(! is_utf8_pat);
4428 c2 = PL_fold_latin1[c1];
4432 NOT_REACHED; /* NOTREACHED */
4435 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
4436 NOT_REACHED; /* NOTREACHED */
4442 /* Here have figured things out. Set up the returns */
4443 if (use_chrtest_void) {
4444 *c2p = *c1p = CHRTEST_VOID;
4446 else if (utf8_target) {
4447 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
4448 uvchr_to_utf8(c1_utf8, c1);
4449 uvchr_to_utf8(c2_utf8, c2);
4452 /* Invariants are stored in both the utf8 and byte outputs; Use
4453 * negative numbers otherwise for the byte ones. Make sure that the
4454 * byte ones are the same iff the utf8 ones are the same */
4455 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
4456 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
4459 ? CHRTEST_NOT_A_CP_1
4460 : CHRTEST_NOT_A_CP_2;
4462 else if (c1 > 255) {
4463 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
4468 *c1p = *c2p = c2; /* c2 is the only representable value */
4470 else { /* c1 is representable; see about c2 */
4472 *c2p = (c2 < 256) ? c2 : c1;
4479 S_isGCB(pTHX_ const GCB_enum before, const GCB_enum after, const U8 * const strbeg, const U8 * const curpos, const bool utf8_target)
4481 /* returns a boolean indicating if there is a Grapheme Cluster Boundary
4482 * between the inputs. See http://www.unicode.org/reports/tr29/. */
4484 PERL_ARGS_ASSERT_ISGCB;
4486 switch (GCB_table[before][after]) {
4493 case GCB_RI_then_RI:
4496 U8 * temp_pos = (U8 *) curpos;
4498 /* Do not break within emoji flag sequences. That is, do not
4499 * break between regional indicator (RI) symbols if there is an
4500 * odd number of RI characters before the break point.
4501 * GB12 sot (RI RI)* RI × RI
4502 * GB13 [^RI] (RI RI)* RI × RI */
4504 while (backup_one_GCB(strbeg,
4506 utf8_target) == GCB_Regional_Indicator)
4511 return RI_count % 2 != 1;
4514 case GCB_EX_then_EM:
4516 /* GB10 ( E_Base | E_Base_GAZ ) Extend* × E_Modifier */
4518 U8 * temp_pos = (U8 *) curpos;
4522 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4524 while (prev == GCB_Extend);
4526 return prev != GCB_E_Base && prev != GCB_E_Base_GAZ;
4529 case GCB_Maybe_Emoji_NonBreak:
4533 /* Do not break within emoji modifier sequences or emoji zwj sequences.
4534 GB11 \p{Extended_Pictographic} Extend* ZWJ × \p{Extended_Pictographic}
4536 U8 * temp_pos = (U8 *) curpos;
4540 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4542 while (prev == GCB_Extend);
4544 return prev != GCB_XPG_XX;
4552 Perl_re_printf( aTHX_ "Unhandled GCB pair: GCB_table[%d, %d] = %d\n",
4553 before, after, GCB_table[before][after]);
4560 S_backup_one_GCB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4565 PERL_ARGS_ASSERT_BACKUP_ONE_GCB;
4567 if (*curpos < strbeg) {
4572 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4573 U8 * prev_prev_char_pos;
4575 if (! prev_char_pos) {
4579 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4580 gcb = getGCB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4581 *curpos = prev_char_pos;
4582 prev_char_pos = prev_prev_char_pos;
4585 *curpos = (U8 *) strbeg;
4590 if (*curpos - 2 < strbeg) {
4591 *curpos = (U8 *) strbeg;
4595 gcb = getGCB_VAL_CP(*(*curpos - 1));
4601 /* Combining marks attach to most classes that precede them, but this defines
4602 * the exceptions (from TR14) */
4603 #define LB_CM_ATTACHES_TO(prev) ( ! ( prev == LB_EDGE \
4604 || prev == LB_Mandatory_Break \
4605 || prev == LB_Carriage_Return \
4606 || prev == LB_Line_Feed \
4607 || prev == LB_Next_Line \
4608 || prev == LB_Space \
4609 || prev == LB_ZWSpace))
4612 S_isLB(pTHX_ LB_enum before,
4614 const U8 * const strbeg,
4615 const U8 * const curpos,
4616 const U8 * const strend,
4617 const bool utf8_target)
4619 U8 * temp_pos = (U8 *) curpos;
4620 LB_enum prev = before;
4622 /* Is the boundary between 'before' and 'after' line-breakable?
4623 * Most of this is just a table lookup of a generated table from Unicode
4624 * rules. But some rules require context to decide, and so have to be
4625 * implemented in code */
4627 PERL_ARGS_ASSERT_ISLB;
4629 /* Rule numbers in the comments below are as of Unicode 9.0 */
4633 switch (LB_table[before][after]) {
4638 case LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4641 case LB_SP_foo + LB_BREAKABLE:
4642 case LB_SP_foo + LB_NOBREAK:
4643 case LB_SP_foo + LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4645 /* When we have something following a SP, we have to look at the
4646 * context in order to know what to do.
4648 * SP SP should not reach here because LB7: Do not break before
4649 * spaces. (For two spaces in a row there is nothing that
4650 * overrides that) */
4651 assert(after != LB_Space);
4653 /* Here we have a space followed by a non-space. Mostly this is a
4654 * case of LB18: "Break after spaces". But there are complications
4655 * as the handling of spaces is somewhat tricky. They are in a
4656 * number of rules, which have to be applied in priority order, but
4657 * something earlier in the string can cause a rule to be skipped
4658 * and a lower priority rule invoked. A prime example is LB7 which
4659 * says don't break before a space. But rule LB8 (lower priority)
4660 * says that the first break opportunity after a ZW is after any
4661 * span of spaces immediately after it. If a ZW comes before a SP
4662 * in the input, rule LB8 applies, and not LB7. Other such rules
4663 * involve combining marks which are rules 9 and 10, but they may
4664 * override higher priority rules if they come earlier in the
4665 * string. Since we're doing random access into the middle of the
4666 * string, we have to look for rules that should get applied based
4667 * on both string position and priority. Combining marks do not
4668 * attach to either ZW nor SP, so we don't have to consider them
4671 * To check for LB8, we have to find the first non-space character
4672 * before this span of spaces */
4674 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4676 while (prev == LB_Space);
4678 /* LB8 Break before any character following a zero-width space,
4679 * even if one or more spaces intervene.
4681 * So if we have a ZW just before this span, and to get here this
4682 * is the final space in the span. */
4683 if (prev == LB_ZWSpace) {
4687 /* Here, not ZW SP+. There are several rules that have higher
4688 * priority than LB18 and can be resolved now, as they don't depend
4689 * on anything earlier in the string (except ZW, which we have
4690 * already handled). One of these rules is LB11 Do not break
4691 * before Word joiner, but we have specially encoded that in the
4692 * lookup table so it is caught by the single test below which
4693 * catches the other ones. */
4694 if (LB_table[LB_Space][after] - LB_SP_foo
4695 == LB_NOBREAK_EVEN_WITH_SP_BETWEEN)
4700 /* If we get here, we have to XXX consider combining marks. */
4701 if (prev == LB_Combining_Mark) {
4703 /* What happens with these depends on the character they
4706 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4708 while (prev == LB_Combining_Mark);
4710 /* Most times these attach to and inherit the characteristics
4711 * of that character, but not always, and when not, they are to
4712 * be treated as AL by rule LB10. */
4713 if (! LB_CM_ATTACHES_TO(prev)) {
4714 prev = LB_Alphabetic;
4718 /* Here, we have the character preceding the span of spaces all set
4719 * up. We follow LB18: "Break after spaces" unless the table shows
4720 * that is overriden */
4721 return LB_table[prev][after] != LB_NOBREAK_EVEN_WITH_SP_BETWEEN;
4725 /* We don't know how to treat the CM except by looking at the first
4726 * non-CM character preceding it. ZWJ is treated as CM */
4728 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4730 while (prev == LB_Combining_Mark || prev == LB_ZWJ);
4732 /* Here, 'prev' is that first earlier non-CM character. If the CM
4733 * attatches to it, then it inherits the behavior of 'prev'. If it
4734 * doesn't attach, it is to be treated as an AL */
4735 if (! LB_CM_ATTACHES_TO(prev)) {
4736 prev = LB_Alphabetic;
4741 case LB_HY_or_BA_then_foo + LB_BREAKABLE:
4742 case LB_HY_or_BA_then_foo + LB_NOBREAK:
4744 /* LB21a Don't break after Hebrew + Hyphen.
4745 * HL (HY | BA) × */
4747 if (backup_one_LB(strbeg, &temp_pos, utf8_target)
4748 == LB_Hebrew_Letter)
4753 return LB_table[prev][after] - LB_HY_or_BA_then_foo == LB_BREAKABLE;
4755 case LB_PR_or_PO_then_OP_or_HY + LB_BREAKABLE:
4756 case LB_PR_or_PO_then_OP_or_HY + LB_NOBREAK:
4758 /* LB25a (PR | PO) × ( OP | HY )? NU */
4759 if (advance_one_LB(&temp_pos, strend, utf8_target) == LB_Numeric) {
4763 return LB_table[prev][after] - LB_PR_or_PO_then_OP_or_HY
4766 case LB_SY_or_IS_then_various + LB_BREAKABLE:
4767 case LB_SY_or_IS_then_various + LB_NOBREAK:
4769 /* LB25d NU (SY | IS)* × (NU | SY | IS | CL | CP ) */
4771 LB_enum temp = prev;
4773 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4775 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric);
4776 if (temp == LB_Numeric) {
4780 return LB_table[prev][after] - LB_SY_or_IS_then_various
4784 case LB_various_then_PO_or_PR + LB_BREAKABLE:
4785 case LB_various_then_PO_or_PR + LB_NOBREAK:
4787 /* LB25e NU (SY | IS)* (CL | CP)? × (PO | PR) */
4789 LB_enum temp = prev;
4790 if (temp == LB_Close_Punctuation || temp == LB_Close_Parenthesis)
4792 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4794 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric) {
4795 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4797 if (temp == LB_Numeric) {
4800 return LB_various_then_PO_or_PR;
4803 case LB_RI_then_RI + LB_NOBREAK:
4804 case LB_RI_then_RI + LB_BREAKABLE:
4808 /* LB30a Break between two regional indicator symbols if and
4809 * only if there are an even number of regional indicators
4810 * preceding the position of the break.
4812 * sot (RI RI)* RI × RI
4813 * [^RI] (RI RI)* RI × RI */
4815 while (backup_one_LB(strbeg,
4817 utf8_target) == LB_Regional_Indicator)
4822 return RI_count % 2 == 0;
4830 Perl_re_printf( aTHX_ "Unhandled LB pair: LB_table[%d, %d] = %d\n",
4831 before, after, LB_table[before][after]);
4838 S_advance_one_LB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4844 PERL_ARGS_ASSERT_ADVANCE_ONE_LB;
4846 if (*curpos >= strend) {
4851 *curpos += UTF8SKIP(*curpos);
4852 if (*curpos >= strend) {
4855 lb = getLB_VAL_UTF8(*curpos, strend);
4859 if (*curpos >= strend) {
4862 lb = getLB_VAL_CP(**curpos);
4869 S_backup_one_LB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4874 PERL_ARGS_ASSERT_BACKUP_ONE_LB;
4876 if (*curpos < strbeg) {
4881 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4882 U8 * prev_prev_char_pos;
4884 if (! prev_char_pos) {
4888 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4889 lb = getLB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4890 *curpos = prev_char_pos;
4891 prev_char_pos = prev_prev_char_pos;
4894 *curpos = (U8 *) strbeg;
4899 if (*curpos - 2 < strbeg) {
4900 *curpos = (U8 *) strbeg;
4904 lb = getLB_VAL_CP(*(*curpos - 1));
4911 S_isSB(pTHX_ SB_enum before,
4913 const U8 * const strbeg,
4914 const U8 * const curpos,
4915 const U8 * const strend,
4916 const bool utf8_target)
4918 /* returns a boolean indicating if there is a Sentence Boundary Break
4919 * between the inputs. See http://www.unicode.org/reports/tr29/ */
4921 U8 * lpos = (U8 *) curpos;
4922 bool has_para_sep = FALSE;
4923 bool has_sp = FALSE;
4925 PERL_ARGS_ASSERT_ISSB;
4927 /* Break at the start and end of text.
4930 But unstated in Unicode is don't break if the text is empty */
4931 if (before == SB_EDGE || after == SB_EDGE) {
4932 return before != after;
4935 /* SB 3: Do not break within CRLF. */
4936 if (before == SB_CR && after == SB_LF) {
4940 /* Break after paragraph separators. CR and LF are considered
4941 * so because Unicode views text as like word processing text where there
4942 * are no newlines except between paragraphs, and the word processor takes
4943 * care of wrapping without there being hard line-breaks in the text *./
4944 SB4. Sep | CR | LF ÷ */
4945 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
4949 /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF.
4950 * (See Section 6.2, Replacing Ignore Rules.)
4951 SB5. X (Extend | Format)* → X */
4952 if (after == SB_Extend || after == SB_Format) {
4954 /* Implied is that the these characters attach to everything
4955 * immediately prior to them except for those separator-type
4956 * characters. And the rules earlier have already handled the case
4957 * when one of those immediately precedes the extend char */
4961 if (before == SB_Extend || before == SB_Format) {
4962 U8 * temp_pos = lpos;
4963 const SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4964 if ( backup != SB_EDGE
4973 /* Here, both 'before' and 'backup' are these types; implied is that we
4974 * don't break between them */
4975 if (backup == SB_Extend || backup == SB_Format) {
4980 /* Do not break after ambiguous terminators like period, if they are
4981 * immediately followed by a number or lowercase letter, if they are
4982 * between uppercase letters, if the first following letter (optionally
4983 * after certain punctuation) is lowercase, or if they are followed by
4984 * "continuation" punctuation such as comma, colon, or semicolon. For
4985 * example, a period may be an abbreviation or numeric period, and thus may
4986 * not mark the end of a sentence.
4988 * SB6. ATerm × Numeric */
4989 if (before == SB_ATerm && after == SB_Numeric) {
4993 /* SB7. (Upper | Lower) ATerm × Upper */
4994 if (before == SB_ATerm && after == SB_Upper) {
4995 U8 * temp_pos = lpos;
4996 SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
4997 if (backup == SB_Upper || backup == SB_Lower) {
5002 /* The remaining rules that aren't the final one, all require an STerm or
5003 * an ATerm after having backed up over some Close* Sp*, and in one case an
5004 * optional Paragraph separator, although one rule doesn't have any Sp's in it.
5005 * So do that backup now, setting flags if either Sp or a paragraph
5006 * separator are found */
5008 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
5009 has_para_sep = TRUE;
5010 before = backup_one_SB(strbeg, &lpos, utf8_target);
5013 if (before == SB_Sp) {
5016 before = backup_one_SB(strbeg, &lpos, utf8_target);
5018 while (before == SB_Sp);
5021 while (before == SB_Close) {
5022 before = backup_one_SB(strbeg, &lpos, utf8_target);
5025 /* The next few rules apply only when the backed-up-to is an ATerm, and in
5026 * most cases an STerm */
5027 if (before == SB_STerm || before == SB_ATerm) {
5029 /* So, here the lhs matches
5030 * (STerm | ATerm) Close* Sp* (Sep | CR | LF)?
5031 * and we have set flags if we found an Sp, or the optional Sep,CR,LF.
5032 * The rules that apply here are:
5034 * SB8 ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR
5035 | LF | STerm | ATerm) )* Lower
5036 SB8a (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm)
5037 SB9 (STerm | ATerm) Close* × (Close | Sp | Sep | CR | LF)
5038 SB10 (STerm | ATerm) Close* Sp* × (Sp | Sep | CR | LF)
5039 SB11 (STerm | ATerm) Close* Sp* (Sep | CR | LF)? ÷
5042 /* And all but SB11 forbid having seen a paragraph separator */
5043 if (! has_para_sep) {
5044 if (before == SB_ATerm) { /* SB8 */
5045 U8 * rpos = (U8 *) curpos;
5046 SB_enum later = after;
5048 while ( later != SB_OLetter
5049 && later != SB_Upper
5050 && later != SB_Lower
5054 && later != SB_STerm
5055 && later != SB_ATerm
5056 && later != SB_EDGE)
5058 later = advance_one_SB(&rpos, strend, utf8_target);
5060 if (later == SB_Lower) {
5065 if ( after == SB_SContinue /* SB8a */
5066 || after == SB_STerm
5067 || after == SB_ATerm)
5072 if (! has_sp) { /* SB9 applies only if there was no Sp* */
5073 if ( after == SB_Close
5083 /* SB10. This and SB9 could probably be combined some way, but khw
5084 * has decided to follow the Unicode rule book precisely for
5085 * simplified maintenance */
5099 /* Otherwise, do not break.
5106 S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
5111 PERL_ARGS_ASSERT_ADVANCE_ONE_SB;
5113 if (*curpos >= strend) {
5119 *curpos += UTF8SKIP(*curpos);
5120 if (*curpos >= strend) {
5123 sb = getSB_VAL_UTF8(*curpos, strend);
5124 } while (sb == SB_Extend || sb == SB_Format);
5129 if (*curpos >= strend) {
5132 sb = getSB_VAL_CP(**curpos);
5133 } while (sb == SB_Extend || sb == SB_Format);
5140 S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5145 PERL_ARGS_ASSERT_BACKUP_ONE_SB;
5147 if (*curpos < strbeg) {
5152 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5153 if (! prev_char_pos) {
5157 /* Back up over Extend and Format. curpos is always just to the right
5158 * of the characater whose value we are getting */
5160 U8 * prev_prev_char_pos;
5161 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1,
5164 sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5165 *curpos = prev_char_pos;
5166 prev_char_pos = prev_prev_char_pos;
5169 *curpos = (U8 *) strbeg;
5172 } while (sb == SB_Extend || sb == SB_Format);
5176 if (*curpos - 2 < strbeg) {
5177 *curpos = (U8 *) strbeg;
5181 sb = getSB_VAL_CP(*(*curpos - 1));
5182 } while (sb == SB_Extend || sb == SB_Format);
5189 S_isWB(pTHX_ WB_enum previous,
5192 const U8 * const strbeg,
5193 const U8 * const curpos,
5194 const U8 * const strend,
5195 const bool utf8_target)
5197 /* Return a boolean as to if the boundary between 'before' and 'after' is
5198 * a Unicode word break, using their published algorithm, but tailored for
5199 * Perl by treating spans of white space as one unit. Context may be
5200 * needed to make this determination. If the value for the character
5201 * before 'before' is known, it is passed as 'previous'; otherwise that
5202 * should be set to WB_UNKNOWN. The other input parameters give the
5203 * boundaries and current position in the matching of the string. That
5204 * is, 'curpos' marks the position where the character whose wb value is
5205 * 'after' begins. See http://www.unicode.org/reports/tr29/ */
5207 U8 * before_pos = (U8 *) curpos;
5208 U8 * after_pos = (U8 *) curpos;
5209 WB_enum prev = before;
5212 PERL_ARGS_ASSERT_ISWB;
5214 /* Rule numbers in the comments below are as of Unicode 9.0 */
5218 switch (WB_table[before][after]) {
5225 case WB_hs_then_hs: /* 2 horizontal spaces in a row */
5226 next = advance_one_WB(&after_pos, strend, utf8_target,
5227 FALSE /* Don't skip Extend nor Format */ );
5228 /* A space immediately preceeding an Extend or Format is attached
5229 * to by them, and hence gets separated from previous spaces.
5230 * Otherwise don't break between horizontal white space */
5231 return next == WB_Extend || next == WB_Format;
5233 /* WB4 Ignore Format and Extend characters, except when they appear at
5234 * the beginning of a region of text. This code currently isn't
5235 * general purpose, but it works as the rules are currently and likely
5236 * to be laid out. The reason it works is that when 'they appear at
5237 * the beginning of a region of text', the rule is to break before
5238 * them, just like any other character. Therefore, the default rule
5239 * applies and we don't have to look in more depth. Should this ever
5240 * change, we would have to have 2 'case' statements, like in the rules
5241 * below, and backup a single character (not spacing over the extend
5242 * ones) and then see if that is one of the region-end characters and
5244 case WB_Ex_or_FO_or_ZWJ_then_foo:
5245 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5248 case WB_DQ_then_HL + WB_BREAKABLE:
5249 case WB_DQ_then_HL + WB_NOBREAK:
5251 /* WB7c Hebrew_Letter Double_Quote × Hebrew_Letter */
5253 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5254 == WB_Hebrew_Letter)
5259 return WB_table[before][after] - WB_DQ_then_HL == WB_BREAKABLE;
5261 case WB_HL_then_DQ + WB_BREAKABLE:
5262 case WB_HL_then_DQ + WB_NOBREAK:
5264 /* WB7b Hebrew_Letter × Double_Quote Hebrew_Letter */
5266 if (advance_one_WB(&after_pos, strend, utf8_target,
5267 TRUE /* Do skip Extend and Format */ )
5268 == WB_Hebrew_Letter)
5273 return WB_table[before][after] - WB_HL_then_DQ == WB_BREAKABLE;
5275 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_NOBREAK:
5276 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_BREAKABLE:
5278 /* WB6 (ALetter | Hebrew_Letter) × (MidLetter | MidNumLet
5279 * | Single_Quote) (ALetter | Hebrew_Letter) */
5281 next = advance_one_WB(&after_pos, strend, utf8_target,
5282 TRUE /* Do skip Extend and Format */ );
5284 if (next == WB_ALetter || next == WB_Hebrew_Letter)
5289 return WB_table[before][after]
5290 - WB_LE_or_HL_then_MB_or_ML_or_SQ == WB_BREAKABLE;
5292 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_NOBREAK:
5293 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_BREAKABLE:
5295 /* WB7 (ALetter | Hebrew_Letter) (MidLetter | MidNumLet
5296 * | Single_Quote) × (ALetter | Hebrew_Letter) */
5298 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5299 if (prev == WB_ALetter || prev == WB_Hebrew_Letter)
5304 return WB_table[before][after]
5305 - WB_MB_or_ML_or_SQ_then_LE_or_HL == WB_BREAKABLE;
5307 case WB_MB_or_MN_or_SQ_then_NU + WB_NOBREAK:
5308 case WB_MB_or_MN_or_SQ_then_NU + WB_BREAKABLE:
5310 /* WB11 Numeric (MidNum | (MidNumLet | Single_Quote)) × Numeric
5313 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5319 return WB_table[before][after]
5320 - WB_MB_or_MN_or_SQ_then_NU == WB_BREAKABLE;
5322 case WB_NU_then_MB_or_MN_or_SQ + WB_NOBREAK:
5323 case WB_NU_then_MB_or_MN_or_SQ + WB_BREAKABLE:
5325 /* WB12 Numeric × (MidNum | MidNumLet | Single_Quote) Numeric */
5327 if (advance_one_WB(&after_pos, strend, utf8_target,
5328 TRUE /* Do skip Extend and Format */ )
5334 return WB_table[before][after]
5335 - WB_NU_then_MB_or_MN_or_SQ == WB_BREAKABLE;
5337 case WB_RI_then_RI + WB_NOBREAK:
5338 case WB_RI_then_RI + WB_BREAKABLE:
5342 /* Do not break within emoji flag sequences. That is, do not
5343 * break between regional indicator (RI) symbols if there is an
5344 * odd number of RI characters before the potential break
5347 * WB15 sot (RI RI)* RI × RI
5348 * WB16 [^RI] (RI RI)* RI × RI */
5350 while (backup_one_WB(&previous,
5353 utf8_target) == WB_Regional_Indicator)
5358 return RI_count % 2 != 1;
5366 Perl_re_printf( aTHX_ "Unhandled WB pair: WB_table[%d, %d] = %d\n",
5367 before, after, WB_table[before][after]);
5374 S_advance_one_WB(pTHX_ U8 ** curpos,
5375 const U8 * const strend,
5376 const bool utf8_target,
5377 const bool skip_Extend_Format)
5382 PERL_ARGS_ASSERT_ADVANCE_ONE_WB;
5384 if (*curpos >= strend) {
5390 /* Advance over Extend and Format */
5392 *curpos += UTF8SKIP(*curpos);
5393 if (*curpos >= strend) {
5396 wb = getWB_VAL_UTF8(*curpos, strend);
5397 } while ( skip_Extend_Format
5398 && (wb == WB_Extend || wb == WB_Format));
5403 if (*curpos >= strend) {
5406 wb = getWB_VAL_CP(**curpos);
5407 } while ( skip_Extend_Format
5408 && (wb == WB_Extend || wb == WB_Format));
5415 S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5420 PERL_ARGS_ASSERT_BACKUP_ONE_WB;
5422 /* If we know what the previous character's break value is, don't have
5424 if (*previous != WB_UNKNOWN) {
5427 /* But we need to move backwards by one */
5429 *curpos = reghopmaybe3(*curpos, -1, strbeg);
5431 *previous = WB_EDGE;
5432 *curpos = (U8 *) strbeg;
5435 *previous = WB_UNKNOWN;
5440 *previous = (*curpos <= strbeg) ? WB_EDGE : WB_UNKNOWN;
5443 /* And we always back up over these three types */
5444 if (wb != WB_Extend && wb != WB_Format && wb != WB_ZWJ) {
5449 if (*curpos < strbeg) {
5454 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5455 if (! prev_char_pos) {
5459 /* Back up over Extend and Format. curpos is always just to the right
5460 * of the characater whose value we are getting */
5462 U8 * prev_prev_char_pos;
5463 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos,
5467 wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5468 *curpos = prev_char_pos;
5469 prev_char_pos = prev_prev_char_pos;
5472 *curpos = (U8 *) strbeg;
5475 } while (wb == WB_Extend || wb == WB_Format || wb == WB_ZWJ);
5479 if (*curpos - 2 < strbeg) {
5480 *curpos = (U8 *) strbeg;
5484 wb = getWB_VAL_CP(*(*curpos - 1));
5485 } while (wb == WB_Extend || wb == WB_Format);
5491 /* Macros for regmatch(), using its internal variables */
5492 #define NEXTCHR_EOS -10 /* nextchr has fallen off the end */
5493 #define NEXTCHR_IS_EOS (nextchr < 0)
5495 #define SET_nextchr \
5496 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
5498 #define SET_locinput(p) \
5502 #define sayYES goto yes
5503 #define sayNO goto no
5504 #define sayNO_SILENT goto no_silent
5506 /* we dont use STMT_START/END here because it leads to
5507 "unreachable code" warnings, which are bogus, but distracting. */
5508 #define CACHEsayNO \
5509 if (ST.cache_mask) \
5510 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
5513 #define EVAL_CLOSE_PAREN_IS(st,expr) \
5516 ( ( st )->u.eval.close_paren ) && \
5517 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5520 #define EVAL_CLOSE_PAREN_IS_TRUE(st,expr) \
5523 ( ( st )->u.eval.close_paren ) && \
5525 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5529 #define EVAL_CLOSE_PAREN_SET(st,expr) \
5530 (st)->u.eval.close_paren = ( (expr) + 1 )
5532 #define EVAL_CLOSE_PAREN_CLEAR(st) \
5533 (st)->u.eval.close_paren = 0
5535 /* push a new state then goto it */
5537 #define PUSH_STATE_GOTO(state, node, input, eol, sr0) \
5538 pushinput = input; \
5542 st->resume_state = state; \
5545 /* push a new state with success backtracking, then goto it */
5547 #define PUSH_YES_STATE_GOTO(state, node, input, eol, sr0) \
5548 pushinput = input; \
5552 st->resume_state = state; \
5553 goto push_yes_state;
5555 #define DEBUG_STATE_pp(pp) \
5557 DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \
5558 Perl_re_printf( aTHX_ \
5559 "%*s" pp " %s%s%s%s%s\n", \
5560 INDENT_CHARS(depth), "", \
5561 PL_reg_name[st->resume_state], \
5562 ((st==yes_state||st==mark_state) ? "[" : ""), \
5563 ((st==yes_state) ? "Y" : ""), \
5564 ((st==mark_state) ? "M" : ""), \
5565 ((st==yes_state||st==mark_state) ? "]" : "") \
5571 regmatch() - main matching routine
5573 This is basically one big switch statement in a loop. We execute an op,
5574 set 'next' to point the next op, and continue. If we come to a point which
5575 we may need to backtrack to on failure such as (A|B|C), we push a
5576 backtrack state onto the backtrack stack. On failure, we pop the top
5577 state, and re-enter the loop at the state indicated. If there are no more
5578 states to pop, we return failure.
5580 Sometimes we also need to backtrack on success; for example /A+/, where
5581 after successfully matching one A, we need to go back and try to
5582 match another one; similarly for lookahead assertions: if the assertion
5583 completes successfully, we backtrack to the state just before the assertion
5584 and then carry on. In these cases, the pushed state is marked as
5585 'backtrack on success too'. This marking is in fact done by a chain of
5586 pointers, each pointing to the previous 'yes' state. On success, we pop to
5587 the nearest yes state, discarding any intermediate failure-only states.
5588 Sometimes a yes state is pushed just to force some cleanup code to be
5589 called at the end of a successful match or submatch; e.g. (??{$re}) uses
5590 it to free the inner regex.
5592 Note that failure backtracking rewinds the cursor position, while
5593 success backtracking leaves it alone.
5595 A pattern is complete when the END op is executed, while a subpattern
5596 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
5597 ops trigger the "pop to last yes state if any, otherwise return true"
5600 A common convention in this function is to use A and B to refer to the two
5601 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
5602 the subpattern to be matched possibly multiple times, while B is the entire
5603 rest of the pattern. Variable and state names reflect this convention.
5605 The states in the main switch are the union of ops and failure/success of
5606 substates associated with with that op. For example, IFMATCH is the op
5607 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
5608 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
5609 successfully matched A and IFMATCH_A_fail is a state saying that we have
5610 just failed to match A. Resume states always come in pairs. The backtrack
5611 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
5612 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
5613 on success or failure.
5615 The struct that holds a backtracking state is actually a big union, with
5616 one variant for each major type of op. The variable st points to the
5617 top-most backtrack struct. To make the code clearer, within each
5618 block of code we #define ST to alias the relevant union.
5620 Here's a concrete example of a (vastly oversimplified) IFMATCH
5626 #define ST st->u.ifmatch
5628 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
5629 ST.foo = ...; // some state we wish to save
5631 // push a yes backtrack state with a resume value of
5632 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
5634 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
5637 case IFMATCH_A: // we have successfully executed A; now continue with B
5639 bar = ST.foo; // do something with the preserved value
5642 case IFMATCH_A_fail: // A failed, so the assertion failed
5643 ...; // do some housekeeping, then ...
5644 sayNO; // propagate the failure
5651 For any old-timers reading this who are familiar with the old recursive
5652 approach, the code above is equivalent to:
5654 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
5663 ...; // do some housekeeping, then ...
5664 sayNO; // propagate the failure
5667 The topmost backtrack state, pointed to by st, is usually free. If you
5668 want to claim it, populate any ST.foo fields in it with values you wish to
5669 save, then do one of
5671 PUSH_STATE_GOTO(resume_state, node, newinput, new_eol);
5672 PUSH_YES_STATE_GOTO(resume_state, node, newinput, new_eol);
5674 which sets that backtrack state's resume value to 'resume_state', pushes a
5675 new free entry to the top of the backtrack stack, then goes to 'node'.
5676 On backtracking, the free slot is popped, and the saved state becomes the
5677 new free state. An ST.foo field in this new top state can be temporarily
5678 accessed to retrieve values, but once the main loop is re-entered, it
5679 becomes available for reuse.
5681 Note that the depth of the backtrack stack constantly increases during the
5682 left-to-right execution of the pattern, rather than going up and down with
5683 the pattern nesting. For example the stack is at its maximum at Z at the
5684 end of the pattern, rather than at X in the following:
5686 /(((X)+)+)+....(Y)+....Z/
5688 The only exceptions to this are lookahead/behind assertions and the cut,
5689 (?>A), which pop all the backtrack states associated with A before
5692 Backtrack state structs are allocated in slabs of about 4K in size.
5693 PL_regmatch_state and st always point to the currently active state,
5694 and PL_regmatch_slab points to the slab currently containing
5695 PL_regmatch_state. The first time regmatch() is called, the first slab is
5696 allocated, and is never freed until interpreter destruction. When the slab
5697 is full, a new one is allocated and chained to the end. At exit from
5698 regmatch(), slabs allocated since entry are freed.
5700 In order to work with variable length lookbehinds, an upper limit is placed on
5701 lookbehinds which is set to where the match position is at the end of where the
5702 lookbehind would get to. Nothing in the lookbehind should match above that,
5703 except we should be able to look beyond if for things like \b, which need the
5704 next character in the string to be able to determine if this is a boundary or
5705 not. We also can't match the end of string/line unless we are also at the end
5706 of the entire string, so NEXTCHR_IS_EOS remains the same, and for those OPs
5707 that match a width, we have to add a condition that they are within the legal
5708 bounds of our window into the string.
5712 /* returns -1 on failure, $+[0] on success */
5714 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
5717 const bool utf8_target = reginfo->is_utf8_target;
5718 const U32 uniflags = UTF8_ALLOW_DEFAULT;
5719 REGEXP *rex_sv = reginfo->prog;
5720 regexp *rex = ReANY(rex_sv);
5721 RXi_GET_DECL(rex,rexi);
5722 /* the current state. This is a cached copy of PL_regmatch_state */
5724 /* cache heavy used fields of st in registers */
5727 U32 n = 0; /* general value; init to avoid compiler warning */
5728 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
5729 SSize_t endref = 0; /* offset of end of backref when ln is start */
5730 char *locinput = startpos;
5731 char *loceol = reginfo->strend;
5732 char *pushinput; /* where to continue after a PUSH */
5733 char *pusheol; /* where to stop matching (loceol) after a PUSH */
5734 U8 *pushsr0; /* save starting pos of script run */
5735 I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */
5737 bool result = 0; /* return value of S_regmatch */
5738 U32 depth = 0; /* depth of backtrack stack */
5739 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
5740 const U32 max_nochange_depth =
5741 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
5742 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
5743 regmatch_state *yes_state = NULL; /* state to pop to on success of
5745 /* mark_state piggy backs on the yes_state logic so that when we unwind
5746 the stack on success we can update the mark_state as we go */
5747 regmatch_state *mark_state = NULL; /* last mark state we have seen */
5748 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
5749 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
5751 bool no_final = 0; /* prevent failure from backtracking? */
5752 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
5753 char *startpoint = locinput;
5754 SV *popmark = NULL; /* are we looking for a mark? */
5755 SV *sv_commit = NULL; /* last mark name seen in failure */
5756 SV *sv_yes_mark = NULL; /* last mark name we have seen
5757 during a successful match */
5758 U32 lastopen = 0; /* last open we saw */
5759 bool has_cutgroup = RXp_HAS_CUTGROUP(rex) ? 1 : 0;
5760 SV* const oreplsv = GvSVn(PL_replgv);
5761 /* these three flags are set by various ops to signal information to
5762 * the very next op. They have a useful lifetime of exactly one loop
5763 * iteration, and are not preserved or restored by state pushes/pops
5765 bool sw = 0; /* the condition value in (?(cond)a|b) */
5766 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
5767 int logical = 0; /* the following EVAL is:
5771 or the following IFMATCH/UNLESSM is:
5772 false: plain (?=foo)
5773 true: used as a condition: (?(?=foo))
5775 PAD* last_pad = NULL;
5777 U8 gimme = G_SCALAR;
5778 CV *caller_cv = NULL; /* who called us */
5779 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
5780 U32 maxopenparen = 0; /* max '(' index seen so far */
5781 int to_complement; /* Invert the result? */
5782 _char_class_number classnum;
5783 bool is_utf8_pat = reginfo->is_utf8_pat;
5785 I32 orig_savestack_ix = PL_savestack_ix;
5786 U8 * script_run_begin = NULL;
5788 /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */
5789 #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL))
5790 # define SOLARIS_BAD_OPTIMIZER
5791 const U32 *pl_charclass_dup = PL_charclass;
5792 # define PL_charclass pl_charclass_dup
5796 GET_RE_DEBUG_FLAGS_DECL;
5799 /* protect against undef(*^R) */
5800 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
5802 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
5803 multicall_oldcatch = 0;
5804 PERL_UNUSED_VAR(multicall_cop);
5806 PERL_ARGS_ASSERT_REGMATCH;
5808 st = PL_regmatch_state;
5810 /* Note that nextchr is a byte even in UTF */
5814 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
5815 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5816 Perl_re_printf( aTHX_ "regmatch start\n" );
5819 while (scan != NULL) {
5820 next = scan + NEXT_OFF(scan);
5823 state_num = OP(scan);
5827 if (state_num <= REGNODE_MAX) {
5828 SV * const prop = sv_newmortal();
5829 regnode *rnext = regnext(scan);
5831 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5832 regprop(rex, prop, scan, reginfo, NULL);
5833 Perl_re_printf( aTHX_
5834 "%*s%" IVdf ":%s(%" IVdf ")\n",
5835 INDENT_CHARS(depth), "",
5836 (IV)(scan - rexi->program),
5838 (PL_regkind[OP(scan)] == END || !rnext) ?
5839 0 : (IV)(rnext - rexi->program));
5846 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
5848 switch (state_num) {
5849 case SBOL: /* /^../ and /\A../ */
5850 if (locinput == reginfo->strbeg)
5854 case MBOL: /* /^../m */
5855 if (locinput == reginfo->strbeg ||
5856 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
5863 if (locinput == reginfo->ganch)
5867 case KEEPS: /* \K */
5868 /* update the startpoint */
5869 st->u.keeper.val = rex->offs[0].start;
5870 rex->offs[0].start = locinput - reginfo->strbeg;
5871 PUSH_STATE_GOTO(KEEPS_next, next, locinput, loceol,
5873 NOT_REACHED; /* NOTREACHED */
5875 case KEEPS_next_fail:
5876 /* rollback the start point change */
5877 rex->offs[0].start = st->u.keeper.val;
5879 NOT_REACHED; /* NOTREACHED */
5881 case MEOL: /* /..$/m */
5882 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5886 case SEOL: /* /..$/ */
5887 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5889 if (reginfo->strend - locinput > 1)
5894 if (!NEXTCHR_IS_EOS)
5898 case SANY: /* /./s */
5899 if (NEXTCHR_IS_EOS || locinput >= loceol)
5901 goto increment_locinput;
5903 case REG_ANY: /* /./ */
5905 || locinput >= loceol
5910 goto increment_locinput;
5914 #define ST st->u.trie
5915 case TRIEC: /* (ab|cd) with known charclass */
5916 /* In this case the charclass data is available inline so
5917 we can fail fast without a lot of extra overhead.
5919 if ( ! NEXTCHR_IS_EOS
5920 && locinput < loceol
5921 && ! ANYOF_BITMAP_TEST(scan, nextchr))
5924 Perl_re_exec_indentf( aTHX_ "%sTRIE: failed to match trie start class...%s\n",
5925 depth, PL_colors[4], PL_colors[5])
5928 NOT_REACHED; /* NOTREACHED */
5931 case TRIE: /* (ab|cd) */
5932 /* the basic plan of execution of the trie is:
5933 * At the beginning, run though all the states, and
5934 * find the longest-matching word. Also remember the position
5935 * of the shortest matching word. For example, this pattern:
5938 * when matched against the string "abcde", will generate
5939 * accept states for all words except 3, with the longest
5940 * matching word being 4, and the shortest being 2 (with
5941 * the position being after char 1 of the string).
5943 * Then for each matching word, in word order (i.e. 1,2,4,5),
5944 * we run the remainder of the pattern; on each try setting
5945 * the current position to the character following the word,
5946 * returning to try the next word on failure.
5948 * We avoid having to build a list of words at runtime by
5949 * using a compile-time structure, wordinfo[].prev, which
5950 * gives, for each word, the previous accepting word (if any).
5951 * In the case above it would contain the mappings 1->2, 2->0,
5952 * 3->0, 4->5, 5->1. We can use this table to generate, from
5953 * the longest word (4 above), a list of all words, by
5954 * following the list of prev pointers; this gives us the
5955 * unordered list 4,5,1,2. Then given the current word we have
5956 * just tried, we can go through the list and find the
5957 * next-biggest word to try (so if we just failed on word 2,
5958 * the next in the list is 4).
5960 * Since at runtime we don't record the matching position in
5961 * the string for each word, we have to work that out for
5962 * each word we're about to process. The wordinfo table holds
5963 * the character length of each word; given that we recorded
5964 * at the start: the position of the shortest word and its
5965 * length in chars, we just need to move the pointer the
5966 * difference between the two char lengths. Depending on
5967 * Unicode status and folding, that's cheap or expensive.
5969 * This algorithm is optimised for the case where are only a
5970 * small number of accept states, i.e. 0,1, or maybe 2.
5971 * With lots of accepts states, and having to try all of them,
5972 * it becomes quadratic on number of accept states to find all
5977 /* what type of TRIE am I? (utf8 makes this contextual) */
5978 DECL_TRIE_TYPE(scan);
5980 /* what trie are we using right now */
5981 reg_trie_data * const trie
5982 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
5983 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
5984 U32 state = trie->startstate;
5986 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
5987 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5990 && UTF8_IS_ABOVE_LATIN1(nextchr)
5991 && scan->flags == EXACTL)
5993 /* We only output for EXACTL, as we let the folder
5994 * output this message for EXACTFLU8 to avoid
5996 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
6002 || locinput >= loceol
6003 || ! TRIE_BITMAP_TEST(trie, nextchr)))
6005 if (trie->states[ state ].wordnum) {
6007 Perl_re_exec_indentf( aTHX_ "%sTRIE: matched empty string...%s\n",
6008 depth, PL_colors[4], PL_colors[5])
6014 Perl_re_exec_indentf( aTHX_ "%sTRIE: failed to match trie start class...%s\n",
6015 depth, PL_colors[4], PL_colors[5])
6022 U8 *uc = ( U8* )locinput;
6026 U8 *uscan = (U8*)NULL;
6027 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
6028 U32 charcount = 0; /* how many input chars we have matched */
6029 U32 accepted = 0; /* have we seen any accepting states? */
6031 ST.jump = trie->jump;
6034 ST.longfold = FALSE; /* char longer if folded => it's harder */
6037 /* fully traverse the TRIE; note the position of the
6038 shortest accept state and the wordnum of the longest
6041 while ( state && uc <= (U8*)(loceol) ) {
6042 U32 base = trie->states[ state ].trans.base;
6046 wordnum = trie->states[ state ].wordnum;
6048 if (wordnum) { /* it's an accept state */
6051 /* record first match position */
6053 ST.firstpos = (U8*)locinput;
6058 ST.firstchars = charcount;
6061 if (!ST.nextword || wordnum < ST.nextword)
6062 ST.nextword = wordnum;
6063 ST.topword = wordnum;
6066 DEBUG_TRIE_EXECUTE_r({
6067 DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth );
6069 PerlIO_printf( Perl_debug_log,
6070 "%*s%sTRIE: State: %4" UVxf " Accepted: %c ",
6071 INDENT_CHARS(depth), "", PL_colors[4],
6072 (UV)state, (accepted ? 'Y' : 'N'));
6075 /* read a char and goto next state */
6076 if ( base && (foldlen || uc < (U8*)(loceol))) {
6078 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
6079 (U8 *) loceol, uscan,
6080 len, uvc, charid, foldlen,
6087 base + charid - 1 - trie->uniquecharcount)) >= 0)
6089 && ((U32)offset < trie->lasttrans)
6090 && trie->trans[offset].check == state)
6092 state = trie->trans[offset].next;
6103 DEBUG_TRIE_EXECUTE_r(
6104 Perl_re_printf( aTHX_
6105 "TRIE: Charid:%3x CP:%4" UVxf " After State: %4" UVxf "%s\n",
6106 charid, uvc, (UV)state, PL_colors[5] );
6112 /* calculate total number of accept states */
6117 w = trie->wordinfo[w].prev;
6120 ST.accepted = accepted;
6124 Perl_re_exec_indentf( aTHX_ "%sTRIE: got %" IVdf " possible matches%s\n",
6126 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
6128 goto trie_first_try; /* jump into the fail handler */
6130 NOT_REACHED; /* NOTREACHED */
6132 case TRIE_next_fail: /* we failed - try next alternative */
6136 /* undo any captures done in the tail part of a branch,
6138 * /(?:X(.)(.)|Y(.)).../
6139 * where the trie just matches X then calls out to do the
6140 * rest of the branch */
6141 REGCP_UNWIND(ST.cp);
6142 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6144 if (!--ST.accepted) {
6146 Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n",
6154 /* Find next-highest word to process. Note that this code
6155 * is O(N^2) per trie run (O(N) per branch), so keep tight */
6158 U16 const nextword = ST.nextword;
6159 reg_trie_wordinfo * const wordinfo
6160 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
6161 for (word=ST.topword; word; word=wordinfo[word].prev) {
6162 if (word > nextword && (!min || word < min))
6175 ST.lastparen = rex->lastparen;
6176 ST.lastcloseparen = rex->lastcloseparen;
6180 /* find start char of end of current word */
6182 U32 chars; /* how many chars to skip */
6183 reg_trie_data * const trie
6184 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
6186 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
6188 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
6193 /* the hard option - fold each char in turn and find
6194 * its folded length (which may be different */
6195 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
6203 /* XXX This assumes the length is well-formed, as
6204 * does the UTF8SKIP below */
6205 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
6213 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
6218 uvc = utf8n_to_uvchr(uscan, foldlen, &len,
6234 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
6235 ? ST.jump[ST.nextword]
6239 Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n",
6247 if ( ST.accepted > 1 || has_cutgroup || ST.jump ) {
6248 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc, loceol,
6250 NOT_REACHED; /* NOTREACHED */
6252 /* only one choice left - just continue */
6254 AV *const trie_words
6255 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
6256 SV ** const tmp = trie_words
6257 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL;
6258 SV *sv= tmp ? sv_newmortal() : NULL;
6260 Perl_re_exec_indentf( aTHX_ "%sTRIE: only one match left, short-circuiting: #%d <%s>%s\n",
6261 depth, PL_colors[4],
6263 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
6264 PL_colors[0], PL_colors[1],
6265 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
6267 : "not compiled under -Dr",
6271 locinput = (char*)uc;
6272 continue; /* execute rest of RE */
6277 case EXACTL: /* /abc/l */
6278 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6280 /* Complete checking would involve going through every character
6281 * matched by the string to see if any is above latin1. But the
6282 * comparision otherwise might very well be a fast assembly
6283 * language routine, and I (khw) don't think slowing things down
6284 * just to check for this warning is worth it. So this just checks
6285 * the first character */
6286 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
6287 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
6291 if (! utf8_target) {
6295 case EXACT: { /* /abc/ */
6300 if (utf8_target != is_utf8_pat) {
6301 /* The target and the pattern have differing utf8ness. */
6303 const char * const e = s + ln;
6306 /* The target is utf8, the pattern is not utf8.
6307 * Above-Latin1 code points can't match the pattern;
6308 * invariants match exactly, and the other Latin1 ones need
6309 * to be downgraded to a single byte in order to do the
6310 * comparison. (If we could be confident that the target
6311 * is not malformed, this could be refactored to have fewer
6312 * tests by just assuming that if the first bytes match, it
6313 * is an invariant, but there are tests in the test suite
6314 * dealing with (??{...}) which violate this) */
6317 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
6321 if (UTF8_IS_INVARIANT(*(U8*)l)) {
6328 if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
6338 /* The target is not utf8, the pattern is utf8. */
6341 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
6345 if (UTF8_IS_INVARIANT(*(U8*)s)) {
6352 if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
6364 /* The target and the pattern have the same utf8ness. */
6365 /* Inline the first character, for speed. */
6366 if ( loceol - locinput < ln
6367 || UCHARAT(s) != nextchr
6368 || (ln > 1 && memNE(s, locinput, ln)))
6377 case EXACTFL: /* /abc/il */
6380 const U8 * fold_array;
6382 U32 fold_utf8_flags;
6384 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6385 folder = foldEQ_locale;
6386 fold_array = PL_fold_locale;
6387 fold_utf8_flags = FOLDEQ_LOCALE;
6390 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
6391 is effectively /u; hence to match, target
6393 if (! utf8_target) {
6396 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
6397 | FOLDEQ_S2_FOLDS_SANE;
6398 folder = foldEQ_latin1_s2_folded;
6399 fold_array = PL_fold_latin1;
6402 case EXACTFU_ONLY8: /* /abc/iu with something in /abc/ > 255 */
6403 if (! utf8_target) {
6406 assert(is_utf8_pat);
6407 fold_utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
6410 case EXACTFUP: /* /foo/iu, and something is problematic in
6411 'foo' so can't take shortcuts. */
6412 assert(! is_utf8_pat);
6413 folder = foldEQ_latin1;
6414 fold_array = PL_fold_latin1;
6415 fold_utf8_flags = 0;
6418 case EXACTFU: /* /abc/iu */
6419 folder = foldEQ_latin1_s2_folded;
6420 fold_array = PL_fold_latin1;
6421 fold_utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
6424 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8
6426 assert(! is_utf8_pat);
6428 case EXACTFAA: /* /abc/iaa */
6429 folder = foldEQ_latin1_s2_folded;
6430 fold_array = PL_fold_latin1;
6431 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6432 if (is_utf8_pat || ! utf8_target) {
6434 /* The possible presence of a MICRO SIGN in the pattern forbids
6435 * us to view a non-UTF-8 pattern as folded when there is a
6437 fold_utf8_flags |= FOLDEQ_S2_ALREADY_FOLDED
6438 |FOLDEQ_S2_FOLDS_SANE;
6443 case EXACTF: /* /abc/i This node only generated for
6444 non-utf8 patterns */
6445 assert(! is_utf8_pat);
6447 fold_array = PL_fold;
6448 fold_utf8_flags = 0;
6456 || state_num == EXACTFUP
6457 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
6459 /* Either target or the pattern are utf8, or has the issue where
6460 * the fold lengths may differ. */
6461 const char * const l = locinput;
6464 if (! foldEQ_utf8_flags(l, &e, 0, utf8_target,
6465 s, 0, ln, is_utf8_pat,fold_utf8_flags))
6473 /* Neither the target nor the pattern are utf8 */
6474 if (UCHARAT(s) != nextchr
6476 && UCHARAT(s) != fold_array[nextchr])
6480 if (loceol - locinput < ln)
6482 if (ln > 1 && ! folder(locinput, s, ln))
6488 case NBOUNDL: /* /\B/l */
6492 case BOUNDL: /* /\b/l */
6495 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6497 if (FLAGS(scan) != TRADITIONAL_BOUND) {
6498 if (! IN_UTF8_CTYPE_LOCALE) {
6499 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
6500 B_ON_NON_UTF8_LOCALE_IS_WRONG);
6506 if (locinput == reginfo->strbeg)
6507 b1 = isWORDCHAR_LC('\n');
6509 b1 = isWORDCHAR_LC_utf8_safe(reghop3((U8*)locinput, -1,
6510 (U8*)(reginfo->strbeg)),
6511 (U8*)(reginfo->strend));
6513 b2 = (NEXTCHR_IS_EOS)
6514 ? isWORDCHAR_LC('\n')
6515 : isWORDCHAR_LC_utf8_safe((U8*) locinput,
6516 (U8*) reginfo->strend);
6518 else { /* Here the string isn't utf8 */
6519 b1 = (locinput == reginfo->strbeg)
6520 ? isWORDCHAR_LC('\n')
6521 : isWORDCHAR_LC(UCHARAT(locinput - 1));
6522 b2 = (NEXTCHR_IS_EOS)
6523 ? isWORDCHAR_LC('\n')
6524 : isWORDCHAR_LC(nextchr);
6526 if (to_complement ^ (b1 == b2)) {
6532 case NBOUND: /* /\B/ */
6536 case BOUND: /* /\b/ */
6540 goto bound_ascii_match_only;
6542 case NBOUNDA: /* /\B/a */
6546 case BOUNDA: /* /\b/a */
6550 bound_ascii_match_only:
6551 /* Here the string isn't utf8, or is utf8 and only ascii characters
6552 * are to match \w. In the latter case looking at the byte just
6553 * prior to the current one may be just the final byte of a
6554 * multi-byte character. This is ok. There are two cases:
6555 * 1) it is a single byte character, and then the test is doing
6556 * just what it's supposed to.
6557 * 2) it is a multi-byte character, in which case the final byte is
6558 * never mistakable for ASCII, and so the test will say it is
6559 * not a word character, which is the correct answer. */
6560 b1 = (locinput == reginfo->strbeg)
6561 ? isWORDCHAR_A('\n')
6562 : isWORDCHAR_A(UCHARAT(locinput - 1));
6563 b2 = (NEXTCHR_IS_EOS)
6564 ? isWORDCHAR_A('\n')
6565 : isWORDCHAR_A(nextchr);
6566 if (to_complement ^ (b1 == b2)) {
6572 case NBOUNDU: /* /\B/u */
6576 case BOUNDU: /* /\b/u */
6579 if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) {
6582 else if (utf8_target) {
6584 switch((bound_type) FLAGS(scan)) {
6585 case TRADITIONAL_BOUND:
6588 b1 = (locinput == reginfo->strbeg)
6589 ? 0 /* isWORDCHAR_L1('\n') */
6590 : isWORDCHAR_utf8_safe(
6591 reghop3((U8*)locinput,
6593 (U8*)(reginfo->strbeg)),
6594 (U8*) reginfo->strend);
6595 b2 = (NEXTCHR_IS_EOS)
6596 ? 0 /* isWORDCHAR_L1('\n') */
6597 : isWORDCHAR_utf8_safe((U8*)locinput,
6598 (U8*) reginfo->strend);
6599 match = cBOOL(b1 != b2);
6603 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6604 match = TRUE; /* GCB always matches at begin and
6608 /* Find the gcb values of previous and current
6609 * chars, then see if is a break point */
6610 match = isGCB(getGCB_VAL_UTF8(
6611 reghop3((U8*)locinput,
6613 (U8*)(reginfo->strbeg)),
6614 (U8*) reginfo->strend),
6615 getGCB_VAL_UTF8((U8*) locinput,
6616 (U8*) reginfo->strend),
6617 (U8*) reginfo->strbeg,
6624 if (locinput == reginfo->strbeg) {
6627 else if (NEXTCHR_IS_EOS) {
6631 match = isLB(getLB_VAL_UTF8(
6632 reghop3((U8*)locinput,
6634 (U8*)(reginfo->strbeg)),
6635 (U8*) reginfo->strend),
6636 getLB_VAL_UTF8((U8*) locinput,
6637 (U8*) reginfo->strend),
6638 (U8*) reginfo->strbeg,
6640 (U8*) reginfo->strend,
6645 case SB_BOUND: /* Always matches at begin and end */
6646 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6650 match = isSB(getSB_VAL_UTF8(
6651 reghop3((U8*)locinput,
6653 (U8*)(reginfo->strbeg)),
6654 (U8*) reginfo->strend),
6655 getSB_VAL_UTF8((U8*) locinput,
6656 (U8*) reginfo->strend),
6657 (U8*) reginfo->strbeg,
6659 (U8*) reginfo->strend,
6665 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6669 match = isWB(WB_UNKNOWN,
6671 reghop3((U8*)locinput,
6673 (U8*)(reginfo->strbeg)),
6674 (U8*) reginfo->strend),
6675 getWB_VAL_UTF8((U8*) locinput,
6676 (U8*) reginfo->strend),
6677 (U8*) reginfo->strbeg,
6679 (U8*) reginfo->strend,
6685 else { /* Not utf8 target */
6686 switch((bound_type) FLAGS(scan)) {
6687 case TRADITIONAL_BOUND:
6690 b1 = (locinput == reginfo->strbeg)
6691 ? 0 /* isWORDCHAR_L1('\n') */
6692 : isWORDCHAR_L1(UCHARAT(locinput - 1));
6693 b2 = (NEXTCHR_IS_EOS)
6694 ? 0 /* isWORDCHAR_L1('\n') */
6695 : isWORDCHAR_L1(nextchr);
6696 match = cBOOL(b1 != b2);
6701 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6702 match = TRUE; /* GCB always matches at begin and
6705 else { /* Only CR-LF combo isn't a GCB in 0-255
6707 match = UCHARAT(locinput - 1) != '\r'
6708 || UCHARAT(locinput) != '\n';
6713 if (locinput == reginfo->strbeg) {
6716 else if (NEXTCHR_IS_EOS) {
6720 match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)),
6721 getLB_VAL_CP(UCHARAT(locinput)),
6722 (U8*) reginfo->strbeg,
6724 (U8*) reginfo->strend,
6729 case SB_BOUND: /* Always matches at begin and end */
6730 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6734 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)),
6735 getSB_VAL_CP(UCHARAT(locinput)),
6736 (U8*) reginfo->strbeg,
6738 (U8*) reginfo->strend,
6744 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6748 match = isWB(WB_UNKNOWN,
6749 getWB_VAL_CP(UCHARAT(locinput -1)),
6750 getWB_VAL_CP(UCHARAT(locinput)),
6751 (U8*) reginfo->strbeg,
6753 (U8*) reginfo->strend,
6760 if (to_complement ^ ! match) {
6766 case ANYOFL: /* /[abc]/l */
6767 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6769 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE)
6771 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
6774 case ANYOFD: /* /[abc]/d */
6775 case ANYOF: /* /[abc]/ */
6776 if (NEXTCHR_IS_EOS || locinput >= loceol)
6778 if ( (! utf8_target || UTF8_IS_INVARIANT(*locinput))
6779 && ! (ANYOF_FLAGS(scan) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP))
6781 if (! ANYOF_BITMAP_TEST(scan, * (U8 *) (locinput))) {
6787 if (!reginclass(rex, scan, (U8*)locinput, (U8*) loceol,
6792 goto increment_locinput;
6798 || (UCHARAT(locinput) & FLAGS(scan)) != ARG(scan)
6799 || locinput >= loceol)
6803 locinput++; /* ANYOFM is always single byte */
6808 || (UCHARAT(locinput) & FLAGS(scan)) == ARG(scan)
6809 || locinput >= loceol)
6813 goto increment_locinput;
6819 || ANYOF_FLAGS(scan) > NATIVE_UTF8_TO_I8((U8) *locinput)
6820 || ! reginclass(rex, scan, (U8*)locinput, (U8*) loceol,
6825 goto increment_locinput;
6831 || ANYOF_FLAGS(scan) != (U8) *locinput
6832 || ! reginclass(rex, scan, (U8*)locinput, (U8*) loceol,
6837 goto increment_locinput;
6843 || ! inRANGE((U8) NATIVE_UTF8_TO_I8(*locinput),
6844 LOWEST_ANYOF_HRx_BYTE(ANYOF_FLAGS(scan)),
6845 HIGHEST_ANYOF_HRx_BYTE(ANYOF_FLAGS(scan)))
6846 || ! reginclass(rex, scan, (U8*)locinput, (U8*) loceol,
6851 goto increment_locinput;
6854 /* The argument (FLAGS) to all the POSIX node types is the class number
6857 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
6861 case POSIXL: /* \w or [:punct:] etc. under /l */
6862 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6863 if (NEXTCHR_IS_EOS || locinput >= loceol)
6866 /* Use isFOO_lc() for characters within Latin1. (Note that
6867 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6868 * wouldn't be invariant) */
6869 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6870 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
6878 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6879 /* An above Latin-1 code point, or malformed */
6880 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
6882 goto utf8_posix_above_latin1;
6885 /* Here is a UTF-8 variant code point below 256 and the target is
6887 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
6888 EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6889 *(locinput + 1))))))
6894 goto increment_locinput;
6896 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
6900 case POSIXD: /* \w or [:punct:] etc. under /d */
6906 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
6908 if (NEXTCHR_IS_EOS || locinput >= loceol) {
6912 /* All UTF-8 variants match */
6913 if (! UTF8_IS_INVARIANT(nextchr)) {
6914 goto increment_locinput;
6920 case POSIXA: /* \w or [:punct:] etc. under /a */
6923 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
6924 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
6925 * character is a single byte */
6927 if (NEXTCHR_IS_EOS || locinput >= loceol) {
6933 if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
6939 /* Here we are either not in utf8, or we matched a utf8-invariant,
6940 * so the next char is the next byte */
6944 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
6948 case POSIXU: /* \w or [:punct:] etc. under /u */
6950 if (NEXTCHR_IS_EOS || locinput >= loceol) {
6954 /* Use _generic_isCC() for characters within Latin1. (Note that
6955 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6956 * wouldn't be invariant) */
6957 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6958 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
6965 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6966 if (! (to_complement
6967 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6975 else { /* Handle above Latin-1 code points */
6976 utf8_posix_above_latin1:
6977 classnum = (_char_class_number) FLAGS(scan);
6980 if (! (to_complement
6981 ^ cBOOL(_invlist_contains_cp(
6982 PL_XPosix_ptrs[classnum],
6983 utf8_to_uvchr_buf((U8 *) locinput,
6984 (U8 *) reginfo->strend,
6990 case _CC_ENUM_SPACE:
6991 if (! (to_complement
6992 ^ cBOOL(is_XPERLSPACE_high(locinput))))
6997 case _CC_ENUM_BLANK:
6998 if (! (to_complement
6999 ^ cBOOL(is_HORIZWS_high(locinput))))
7004 case _CC_ENUM_XDIGIT:
7005 if (! (to_complement
7006 ^ cBOOL(is_XDIGIT_high(locinput))))
7011 case _CC_ENUM_VERTSPACE:
7012 if (! (to_complement
7013 ^ cBOOL(is_VERTWS_high(locinput))))
7018 case _CC_ENUM_CNTRL: /* These can't match above Latin1 */
7019 case _CC_ENUM_ASCII:
7020 if (! to_complement) {
7025 locinput += UTF8_SAFE_SKIP(locinput, reginfo->strend);
7029 case CLUMP: /* Match \X: logical Unicode character. This is defined as
7030 a Unicode extended Grapheme Cluster */
7031 if (NEXTCHR_IS_EOS || locinput >= loceol)
7033 if (! utf8_target) {
7035 /* Match either CR LF or '.', as all the other possibilities
7037 locinput++; /* Match the . or CR */
7038 if (nextchr == '\r' /* And if it was CR, and the next is LF,
7040 && locinput < loceol
7041 && UCHARAT(locinput) == '\n')
7048 /* Get the gcb type for the current character */
7049 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput,
7050 (U8*) reginfo->strend);
7052 /* Then scan through the input until we get to the first
7053 * character whose type is supposed to be a gcb with the
7054 * current character. (There is always a break at the
7056 locinput += UTF8SKIP(locinput);
7057 while (locinput < loceol) {
7058 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput,
7059 (U8*) reginfo->strend);
7060 if (isGCB(prev_gcb, cur_gcb,
7061 (U8*) reginfo->strbeg, (U8*) locinput,
7068 locinput += UTF8SKIP(locinput);
7075 case REFFLN: /* /\g{name}/il */
7076 { /* The capture buffer cases. The ones beginning with N for the
7077 named buffers just convert to the equivalent numbered and
7078 pretend they were called as the corresponding numbered buffer
7080 /* don't initialize these in the declaration, it makes C++
7085 const U8 *fold_array;
7088 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
7089 folder = foldEQ_locale;
7090 fold_array = PL_fold_locale;
7092 utf8_fold_flags = FOLDEQ_LOCALE;
7095 case REFFAN: /* /\g{name}/iaa */
7096 folder = foldEQ_latin1;
7097 fold_array = PL_fold_latin1;
7099 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7102 case REFFUN: /* /\g{name}/iu */
7103 folder = foldEQ_latin1;
7104 fold_array = PL_fold_latin1;
7106 utf8_fold_flags = 0;
7109 case REFFN: /* /\g{name}/i */
7111 fold_array = PL_fold;
7113 utf8_fold_flags = 0;
7116 case REFN: /* /\g{name}/ */
7120 utf8_fold_flags = 0;
7123 /* For the named back references, find the corresponding buffer
7125 n = reg_check_named_buff_matched(rex,scan);
7130 goto do_nref_ref_common;
7132 case REFFL: /* /\1/il */
7133 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
7134 folder = foldEQ_locale;
7135 fold_array = PL_fold_locale;
7136 utf8_fold_flags = FOLDEQ_LOCALE;
7139 case REFFA: /* /\1/iaa */
7140 folder = foldEQ_latin1;
7141 fold_array = PL_fold_latin1;
7142 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7145 case REFFU: /* /\1/iu */
7146 folder = foldEQ_latin1;
7147 fold_array = PL_fold_latin1;
7148 utf8_fold_flags = 0;
7151 case REFF: /* /\1/i */
7153 fold_array = PL_fold;
7154 utf8_fold_flags = 0;
7157 case REF: /* /\1/ */
7160 utf8_fold_flags = 0;
7164 n = ARG(scan); /* which paren pair */
7167 ln = rex->offs[n].start;
7168 endref = rex->offs[n].end;
7169 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7170 if (rex->lastparen < n || ln == -1 || endref == -1)
7171 sayNO; /* Do not match unless seen CLOSEn. */
7175 s = reginfo->strbeg + ln;
7176 if (type != REF /* REF can do byte comparison */
7177 && (utf8_target || type == REFFU || type == REFFL))
7179 char * limit = loceol;
7181 /* This call case insensitively compares the entire buffer
7182 * at s, with the current input starting at locinput, but
7183 * not going off the end given by loceol, and
7184 * returns in <limit> upon success, how much of the
7185 * current input was matched */
7186 if (! foldEQ_utf8_flags(s, NULL, endref - ln, utf8_target,
7187 locinput, &limit, 0, utf8_target, utf8_fold_flags))
7195 /* Not utf8: Inline the first character, for speed. */
7196 if ( ! NEXTCHR_IS_EOS
7197 && locinput < loceol
7198 && UCHARAT(s) != nextchr
7200 || UCHARAT(s) != fold_array[nextchr]))
7205 if (locinput + ln > loceol)
7207 if (ln > 1 && (type == REF
7208 ? memNE(s, locinput, ln)
7209 : ! folder(locinput, s, ln)))
7215 case NOTHING: /* null op; e.g. the 'nothing' following
7216 * the '*' in m{(a+|b)*}' */
7218 case TAIL: /* placeholder while compiling (A|B|C) */
7222 #define ST st->u.eval
7223 #define CUR_EVAL cur_eval->u.eval
7229 regexp_internal *rei;
7230 regnode *startpoint;
7233 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
7234 arg= (U32)ARG(scan);
7235 if (cur_eval && cur_eval->locinput == locinput) {
7236 if ( ++nochange_depth > max_nochange_depth )
7238 "Pattern subroutine nesting without pos change"
7239 " exceeded limit in regex");
7246 startpoint = scan + ARG2L(scan);
7247 EVAL_CLOSE_PAREN_SET( st, arg );
7248 /* Detect infinite recursion
7250 * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/
7251 * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever.
7252 * So we track the position in the string we are at each time
7253 * we recurse and if we try to enter the same routine twice from
7254 * the same position we throw an error.
7256 if ( rex->recurse_locinput[arg] == locinput ) {
7257 /* FIXME: we should show the regop that is failing as part
7258 * of the error message. */
7259 Perl_croak(aTHX_ "Infinite recursion in regex");
7261 ST.prev_recurse_locinput= rex->recurse_locinput[arg];
7262 rex->recurse_locinput[arg]= locinput;
7265 GET_RE_DEBUG_FLAGS_DECL;
7267 Perl_re_exec_indentf( aTHX_
7268 "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n",
7269 depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg]
7275 /* Save all the positions seen so far. */
7276 ST.cp = regcppush(rex, 0, maxopenparen);
7277 REGCP_SET(ST.lastcp);
7279 /* and then jump to the code we share with EVAL */
7280 goto eval_recurse_doit;
7283 case EVAL: /* /(?{...})B/ /(??{A})B/ and /(?(?{...})X|Y)B/ */
7284 if (cur_eval && cur_eval->locinput==locinput) {
7285 if ( ++nochange_depth > max_nochange_depth )
7286 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
7291 /* execute the code in the {...} */
7295 OP * const oop = PL_op;
7296 COP * const ocurcop = PL_curcop;
7300 /* save *all* paren positions */
7301 regcppush(rex, 0, maxopenparen);
7302 REGCP_SET(ST.lastcp);
7305 caller_cv = find_runcv(NULL);
7309 if (rexi->data->what[n] == 'r') { /* code from an external qr */
7311 (REGEXP*)(rexi->data->data[n])
7313 nop = (OP*)rexi->data->data[n+1];
7315 else if (rexi->data->what[n] == 'l') { /* literal code */
7317 nop = (OP*)rexi->data->data[n];
7318 assert(CvDEPTH(newcv));
7321 /* literal with own CV */
7322 assert(rexi->data->what[n] == 'L');
7323 newcv = rex->qr_anoncv;
7324 nop = (OP*)rexi->data->data[n];
7327 /* Some notes about MULTICALL and the context and save stacks.
7330 * /...(?{ my $x)}...(?{ my $y)}...(?{ my $z)}.../
7331 * since codeblocks don't introduce a new scope (so that
7332 * local() etc accumulate), at the end of a successful
7333 * match there will be a SAVEt_CLEARSV on the savestack
7334 * for each of $x, $y, $z. If the three code blocks above
7335 * happen to have come from different CVs (e.g. via
7336 * embedded qr//s), then we must ensure that during any
7337 * savestack unwinding, PL_comppad always points to the
7338 * right pad at each moment. We achieve this by
7339 * interleaving SAVEt_COMPPAD's on the savestack whenever
7340 * there is a change of pad.
7341 * In theory whenever we call a code block, we should
7342 * push a CXt_SUB context, then pop it on return from
7343 * that code block. This causes a bit of an issue in that
7344 * normally popping a context also clears the savestack
7345 * back to cx->blk_oldsaveix, but here we specifically
7346 * don't want to clear the save stack on exit from the
7348 * Also for efficiency we don't want to keep pushing and
7349 * popping the single SUB context as we backtrack etc.
7350 * So instead, we push a single context the first time
7351 * we need, it, then hang onto it until the end of this
7352 * function. Whenever we encounter a new code block, we
7353 * update the CV etc if that's changed. During the times
7354 * in this function where we're not executing a code
7355 * block, having the SUB context still there is a bit
7356 * naughty - but we hope that no-one notices.
7357 * When the SUB context is initially pushed, we fake up
7358 * cx->blk_oldsaveix to be as if we'd pushed this context
7359 * on first entry to S_regmatch rather than at some random
7360 * point during the regexe execution. That way if we
7361 * croak, popping the context stack will ensure that
7362 * *everything* SAVEd by this function is undone and then
7363 * the context popped, rather than e.g., popping the
7364 * context (and restoring the original PL_comppad) then
7365 * popping more of the savestack and restoring a bad
7369 /* If this is the first EVAL, push a MULTICALL. On
7370 * subsequent calls, if we're executing a different CV, or
7371 * if PL_comppad has got messed up from backtracking
7372 * through SAVECOMPPADs, then refresh the context.
7374 if (newcv != last_pushed_cv || PL_comppad != last_pad)
7376 U8 flags = (CXp_SUB_RE |
7377 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
7379 if (last_pushed_cv) {
7380 CHANGE_MULTICALL_FLAGS(newcv, flags);
7383 PUSH_MULTICALL_FLAGS(newcv, flags);
7385 /* see notes above */
7386 CX_CUR()->blk_oldsaveix = orig_savestack_ix;
7388 last_pushed_cv = newcv;
7391 /* these assignments are just to silence compiler
7393 multicall_cop = NULL;
7395 last_pad = PL_comppad;
7397 /* the initial nextstate you would normally execute
7398 * at the start of an eval (which would cause error
7399 * messages to come from the eval), may be optimised
7400 * away from the execution path in the regex code blocks;
7401 * so manually set PL_curcop to it initially */
7403 OP *o = cUNOPx(nop)->op_first;
7404 assert(o->op_type == OP_NULL);
7405 if (o->op_targ == OP_SCOPE) {
7406 o = cUNOPo->op_first;
7409 assert(o->op_targ == OP_LEAVE);
7410 o = cUNOPo->op_first;
7411 assert(o->op_type == OP_ENTER);
7415 if (o->op_type != OP_STUB) {
7416 assert( o->op_type == OP_NEXTSTATE
7417 || o->op_type == OP_DBSTATE
7418 || (o->op_type == OP_NULL
7419 && ( o->op_targ == OP_NEXTSTATE
7420 || o->op_targ == OP_DBSTATE
7424 PL_curcop = (COP*)o;
7429 DEBUG_STATE_r( Perl_re_printf( aTHX_
7430 " re EVAL PL_op=0x%" UVxf "\n", PTR2UV(nop)) );
7432 rex->offs[0].end = locinput - reginfo->strbeg;
7433 if (reginfo->info_aux_eval->pos_magic)
7434 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
7435 reginfo->sv, reginfo->strbeg,
7436 locinput - reginfo->strbeg);
7439 SV *sv_mrk = get_sv("REGMARK", 1);
7440 sv_setsv(sv_mrk, sv_yes_mark);
7443 /* we don't use MULTICALL here as we want to call the
7444 * first op of the block of interest, rather than the
7445 * first op of the sub. Also, we don't want to free
7446 * the savestack frame */
7447 before = (IV)(SP-PL_stack_base);
7449 CALLRUNOPS(aTHX); /* Scalar context. */
7451 if ((IV)(SP-PL_stack_base) == before)
7452 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
7458 /* before restoring everything, evaluate the returned
7459 * value, so that 'uninit' warnings don't use the wrong
7460 * PL_op or pad. Also need to process any magic vars
7461 * (e.g. $1) *before* parentheses are restored */
7466 if (logical == 0) { /* (?{})/ */
7467 SV *replsv = save_scalar(PL_replgv);
7468 sv_setsv(replsv, ret); /* $^R */
7471 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
7472 sw = cBOOL(SvTRUE_NN(ret));
7475 else { /* /(??{}) */
7476 /* if its overloaded, let the regex compiler handle
7477 * it; otherwise extract regex, or stringify */
7478 if (SvGMAGICAL(ret))
7479 ret = sv_mortalcopy(ret);
7480 if (!SvAMAGIC(ret)) {
7484 if (SvTYPE(sv) == SVt_REGEXP)
7485 re_sv = (REGEXP*) sv;
7486 else if (SvSMAGICAL(ret)) {
7487 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
7489 re_sv = (REGEXP *) mg->mg_obj;
7492 /* force any undef warnings here */
7493 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
7494 ret = sv_mortalcopy(ret);
7495 (void) SvPV_force_nolen(ret);
7501 /* *** Note that at this point we don't restore
7502 * PL_comppad, (or pop the CxSUB) on the assumption it may
7503 * be used again soon. This is safe as long as nothing
7504 * in the regexp code uses the pad ! */
7506 PL_curcop = ocurcop;
7507 regcp_restore(rex, ST.lastcp, &maxopenparen);
7508 PL_curpm_under = PL_curpm;
7509 PL_curpm = PL_reg_curpm;
7512 PUSH_STATE_GOTO(EVAL_B, next, locinput, loceol,
7518 /* only /(??{})/ from now on */
7521 /* extract RE object from returned value; compiling if
7525 re_sv = reg_temp_copy(NULL, re_sv);
7530 if (SvUTF8(ret) && IN_BYTES) {
7531 /* In use 'bytes': make a copy of the octet
7532 * sequence, but without the flag on */
7534 const char *const p = SvPV(ret, len);
7535 ret = newSVpvn_flags(p, len, SVs_TEMP);
7537 if (rex->intflags & PREGf_USE_RE_EVAL)
7538 pm_flags |= PMf_USE_RE_EVAL;
7540 /* if we got here, it should be an engine which
7541 * supports compiling code blocks and stuff */
7542 assert(rex->engine && rex->engine->op_comp);
7543 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
7544 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
7545 rex->engine, NULL, NULL,
7546 /* copy /msixn etc to inner pattern */
7551 & (SVs_TEMP | SVs_GMG | SVf_ROK))
7552 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
7553 /* This isn't a first class regexp. Instead, it's
7554 caching a regexp onto an existing, Perl visible
7556 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
7562 RXp_MATCH_COPIED_off(re);
7563 re->subbeg = rex->subbeg;
7564 re->sublen = rex->sublen;
7565 re->suboffset = rex->suboffset;
7566 re->subcoffset = rex->subcoffset;
7568 re->lastcloseparen = 0;
7571 debug_start_match(re_sv, utf8_target, locinput,
7572 reginfo->strend, "EVAL/GOSUB: Matching embedded");
7574 startpoint = rei->program + 1;
7575 EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0;
7576 * close_paren only for GOSUB */
7577 ST.prev_recurse_locinput= NULL; /* only used for GOSUB */
7578 /* Save all the seen positions so far. */
7579 ST.cp = regcppush(rex, 0, maxopenparen);
7580 REGCP_SET(ST.lastcp);
7581 /* and set maxopenparen to 0, since we are starting a "fresh" match */
7583 /* run the pattern returned from (??{...}) */
7585 eval_recurse_doit: /* Share code with GOSUB below this line
7586 * At this point we expect the stack context to be
7587 * set up correctly */
7589 /* invalidate the S-L poscache. We're now executing a
7590 * different set of WHILEM ops (and their associated
7591 * indexes) against the same string, so the bits in the
7592 * cache are meaningless. Setting maxiter to zero forces
7593 * the cache to be invalidated and zeroed before reuse.
7594 * XXX This is too dramatic a measure. Ideally we should
7595 * save the old cache and restore when running the outer
7597 reginfo->poscache_maxiter = 0;
7599 /* the new regexp might have a different is_utf8_pat than we do */
7600 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
7602 ST.prev_rex = rex_sv;
7603 ST.prev_curlyx = cur_curlyx;
7605 SET_reg_curpm(rex_sv);
7610 ST.prev_eval = cur_eval;
7612 /* now continue from first node in postoned RE */
7613 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, startpoint, locinput,
7614 loceol, script_run_begin);
7615 NOT_REACHED; /* NOTREACHED */
7618 case EVAL_postponed_AB: /* cleanup after a successful (??{A})B */
7619 /* note: this is called twice; first after popping B, then A */
7621 Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n",
7622 depth, cur_eval, ST.prev_eval);
7625 #define SET_RECURSE_LOCINPUT(STR,VAL)\
7626 if ( cur_eval && CUR_EVAL.close_paren ) {\
7628 Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\
7630 CUR_EVAL.close_paren - 1,\
7634 rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\
7637 SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput);
7639 rex_sv = ST.prev_rex;
7640 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7641 SET_reg_curpm(rex_sv);
7642 rex = ReANY(rex_sv);
7643 rexi = RXi_GET(rex);
7645 /* preserve $^R across LEAVE's. See Bug 121070. */
7646 SV *save_sv= GvSV(PL_replgv);
7648 SvREFCNT_inc(save_sv);
7649 regcpblow(ST.cp); /* LEAVE in disguise */
7650 /* don't move this initialization up */
7651 replsv = GvSV(PL_replgv);
7652 sv_setsv(replsv, save_sv);
7654 SvREFCNT_dec(save_sv);
7656 cur_eval = ST.prev_eval;
7657 cur_curlyx = ST.prev_curlyx;
7659 /* Invalidate cache. See "invalidate" comment above. */
7660 reginfo->poscache_maxiter = 0;
7661 if ( nochange_depth )
7664 SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput);
7668 case EVAL_B_fail: /* unsuccessful B in (?{...})B */
7669 REGCP_UNWIND(ST.lastcp);
7672 case EVAL_postponed_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
7673 /* note: this is called twice; first after popping B, then A */
7675 Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n",
7676 depth, cur_eval, ST.prev_eval);
7679 SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput);
7681 rex_sv = ST.prev_rex;
7682 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7683 SET_reg_curpm(rex_sv);
7684 rex = ReANY(rex_sv);
7685 rexi = RXi_GET(rex);
7687 REGCP_UNWIND(ST.lastcp);
7688 regcppop(rex, &maxopenparen);
7689 cur_eval = ST.prev_eval;
7690 cur_curlyx = ST.prev_curlyx;
7692 /* Invalidate cache. See "invalidate" comment above. */
7693 reginfo->poscache_maxiter = 0;
7694 if ( nochange_depth )
7697 SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput);
7702 n = ARG(scan); /* which paren pair */
7703 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
7704 if (n > maxopenparen)
7706 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
7707 "OPEN: rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf " tmp; maxopenparen=%" UVuf "\n",
7712 (IV)rex->offs[n].start_tmp,
7718 case SROPEN: /* (*SCRIPT_RUN: */
7719 script_run_begin = (U8 *) locinput;
7724 n = ARG(scan); /* which paren pair */
7725 CLOSE_CAPTURE(n, rex->offs[n].start_tmp,
7726 locinput - reginfo->strbeg);
7727 if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) )
7732 case SRCLOSE: /* (*SCRIPT_RUN: ... ) */
7734 if (! isSCRIPT_RUN(script_run_begin, (U8 *) locinput, utf8_target))
7742 case ACCEPT: /* (*ACCEPT) */
7744 sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7748 cursor && OP(cursor)!=END;
7749 cursor=regnext(cursor))
7751 if ( OP(cursor)==CLOSE ){
7753 if ( n <= lastopen ) {
7754 CLOSE_CAPTURE(n, rex->offs[n].start_tmp,
7755 locinput - reginfo->strbeg);
7756 if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) )
7765 case GROUPP: /* (?(1)) */
7766 n = ARG(scan); /* which paren pair */
7767 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
7770 case GROUPPN: /* (?(<name>)) */
7771 /* reg_check_named_buff_matched returns 0 for no match */
7772 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
7775 case INSUBP: /* (?(R)) */
7777 /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg
7778 * of SCAN is already set up as matches a eval.close_paren */
7779 sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n);
7782 case DEFINEP: /* (?(DEFINE)) */
7786 case IFTHEN: /* (?(cond)A|B) */
7787 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7789 next = NEXTOPER(NEXTOPER(scan));
7791 next = scan + ARG(scan);
7792 if (OP(next) == IFTHEN) /* Fake one. */
7793 next = NEXTOPER(NEXTOPER(next));
7797 case LOGICAL: /* modifier for EVAL and IFMATCH */
7798 logical = scan->flags;
7801 /*******************************************************************
7803 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
7804 pattern, where A and B are subpatterns. (For simple A, CURLYM or
7805 STAR/PLUS/CURLY/CURLYN are used instead.)
7807 A*B is compiled as <CURLYX><A><WHILEM><B>
7809 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
7810 state, which contains the current count, initialised to -1. It also sets
7811 cur_curlyx to point to this state, with any previous value saved in the
7814 CURLYX then jumps straight to the WHILEM op, rather than executing A,
7815 since the pattern may possibly match zero times (i.e. it's a while {} loop
7816 rather than a do {} while loop).
7818 Each entry to WHILEM represents a successful match of A. The count in the
7819 CURLYX block is incremented, another WHILEM state is pushed, and execution
7820 passes to A or B depending on greediness and the current count.
7822 For example, if matching against the string a1a2a3b (where the aN are
7823 substrings that match /A/), then the match progresses as follows: (the
7824 pushed states are interspersed with the bits of strings matched so far):
7827 <CURLYX cnt=0><WHILEM>
7828 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
7829 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
7830 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
7831 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
7833 (Contrast this with something like CURLYM, which maintains only a single
7837 a1 <CURLYM cnt=1> a2
7838 a1 a2 <CURLYM cnt=2> a3
7839 a1 a2 a3 <CURLYM cnt=3> b
7842 Each WHILEM state block marks a point to backtrack to upon partial failure
7843 of A or B, and also contains some minor state data related to that
7844 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
7845 overall state, such as the count, and pointers to the A and B ops.
7847 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
7848 must always point to the *current* CURLYX block, the rules are:
7850 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
7851 and set cur_curlyx to point the new block.
7853 When popping the CURLYX block after a successful or unsuccessful match,
7854 restore the previous cur_curlyx.
7856 When WHILEM is about to execute B, save the current cur_curlyx, and set it
7857 to the outer one saved in the CURLYX block.
7859 When popping the WHILEM block after a successful or unsuccessful B match,
7860 restore the previous cur_curlyx.
7862 Here's an example for the pattern (AI* BI)*BO
7863 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
7866 curlyx backtrack stack
7867 ------ ---------------
7869 CO <CO prev=NULL> <WO>
7870 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7871 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7872 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
7874 At this point the pattern succeeds, and we work back down the stack to
7875 clean up, restoring as we go:
7877 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7878 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7879 CO <CO prev=NULL> <WO>
7882 *******************************************************************/
7884 #define ST st->u.curlyx
7886 case CURLYX: /* start of /A*B/ (for complex A) */
7888 /* No need to save/restore up to this paren */
7889 I32 parenfloor = scan->flags;
7891 assert(next); /* keep Coverity happy */
7892 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
7895 /* XXXX Probably it is better to teach regpush to support
7896 parenfloor > maxopenparen ... */
7897 if (parenfloor > (I32)rex->lastparen)
7898 parenfloor = rex->lastparen; /* Pessimization... */
7900 ST.prev_curlyx= cur_curlyx;
7902 ST.cp = PL_savestack_ix;
7904 /* these fields contain the state of the current curly.
7905 * they are accessed by subsequent WHILEMs */
7906 ST.parenfloor = parenfloor;
7911 ST.count = -1; /* this will be updated by WHILEM */
7912 ST.lastloc = NULL; /* this will be updated by WHILEM */
7914 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput, loceol,
7916 NOT_REACHED; /* NOTREACHED */
7919 case CURLYX_end: /* just finished matching all of A*B */
7920 cur_curlyx = ST.prev_curlyx;
7922 NOT_REACHED; /* NOTREACHED */
7924 case CURLYX_end_fail: /* just failed to match all of A*B */
7926 cur_curlyx = ST.prev_curlyx;
7928 NOT_REACHED; /* NOTREACHED */
7932 #define ST st->u.whilem
7934 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
7936 /* see the discussion above about CURLYX/WHILEM */
7941 assert(cur_curlyx); /* keep Coverity happy */
7943 min = ARG1(cur_curlyx->u.curlyx.me);
7944 max = ARG2(cur_curlyx->u.curlyx.me);
7945 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
7946 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
7947 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
7948 ST.cache_offset = 0;
7952 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: matched %ld out of %d..%d\n",
7953 depth, (long)n, min, max)
7956 /* First just match a string of min A's. */
7959 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, maxopenparen);
7960 cur_curlyx->u.curlyx.lastloc = locinput;
7961 REGCP_SET(ST.lastcp);
7963 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput, loceol,
7965 NOT_REACHED; /* NOTREACHED */
7968 /* If degenerate A matches "", assume A done. */
7970 if (locinput == cur_curlyx->u.curlyx.lastloc) {
7971 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: empty match detected, trying continuation...\n",
7974 goto do_whilem_B_max;
7977 /* super-linear cache processing.
7979 * The idea here is that for certain types of CURLYX/WHILEM -
7980 * principally those whose upper bound is infinity (and
7981 * excluding regexes that have things like \1 and other very
7982 * non-regular expresssiony things), then if a pattern like
7983 * /....A*.../ fails and we backtrack to the WHILEM, then we
7984 * make a note that this particular WHILEM op was at string
7985 * position 47 (say) when the rest of pattern failed. Then, if
7986 * we ever find ourselves back at that WHILEM, and at string
7987 * position 47 again, we can just fail immediately rather than
7988 * running the rest of the pattern again.
7990 * This is very handy when patterns start to go
7991 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
7992 * with a combinatorial explosion of backtracking.
7994 * The cache is implemented as a bit array, with one bit per
7995 * string byte position per WHILEM op (up to 16) - so its
7996 * between 0.25 and 2x the string size.
7998 * To avoid allocating a poscache buffer every time, we do an
7999 * initially countdown; only after we have executed a WHILEM
8000 * op (string-length x #WHILEMs) times do we allocate the
8003 * The top 4 bits of scan->flags byte say how many different
8004 * relevant CURLLYX/WHILEM op pairs there are, while the
8005 * bottom 4-bits is the identifying index number of this
8011 if (!reginfo->poscache_maxiter) {
8012 /* start the countdown: Postpone detection until we
8013 * know the match is not *that* much linear. */
8014 reginfo->poscache_maxiter
8015 = (reginfo->strend - reginfo->strbeg + 1)
8017 /* possible overflow for long strings and many CURLYX's */
8018 if (reginfo->poscache_maxiter < 0)
8019 reginfo->poscache_maxiter = I32_MAX;
8020 reginfo->poscache_iter = reginfo->poscache_maxiter;
8023 if (reginfo->poscache_iter-- == 0) {
8024 /* initialise cache */
8025 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
8026 regmatch_info_aux *const aux = reginfo->info_aux;
8027 if (aux->poscache) {
8028 if ((SSize_t)reginfo->poscache_size < size) {
8029 Renew(aux->poscache, size, char);
8030 reginfo->poscache_size = size;
8032 Zero(aux->poscache, size, char);
8035 reginfo->poscache_size = size;
8036 Newxz(aux->poscache, size, char);
8038 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
8039 "%sWHILEM: Detected a super-linear match, switching on caching%s...\n",
8040 PL_colors[4], PL_colors[5])
8044 if (reginfo->poscache_iter < 0) {
8045 /* have we already failed at this position? */
8046 SSize_t offset, mask;
8048 reginfo->poscache_iter = -1; /* stop eventual underflow */
8049 offset = (scan->flags & 0xf) - 1
8050 + (locinput - reginfo->strbeg)
8052 mask = 1 << (offset % 8);
8054 if (reginfo->info_aux->poscache[offset] & mask) {
8055 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: (cache) already tried at this position...\n",
8058 cur_curlyx->u.curlyx.count--;
8059 sayNO; /* cache records failure */
8061 ST.cache_offset = offset;
8062 ST.cache_mask = mask;
8066 /* Prefer B over A for minimal matching. */
8068 if (cur_curlyx->u.curlyx.minmod) {
8069 ST.save_curlyx = cur_curlyx;
8070 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
8071 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
8072 locinput, loceol, script_run_begin);
8073 NOT_REACHED; /* NOTREACHED */
8076 /* Prefer A over B for maximal matching. */
8078 if (n < max) { /* More greed allowed? */
8079 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
8081 cur_curlyx->u.curlyx.lastloc = locinput;
8082 REGCP_SET(ST.lastcp);
8083 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput, loceol,
8085 NOT_REACHED; /* NOTREACHED */
8087 goto do_whilem_B_max;
8089 NOT_REACHED; /* NOTREACHED */
8091 case WHILEM_B_min: /* just matched B in a minimal match */
8092 case WHILEM_B_max: /* just matched B in a maximal match */
8093 cur_curlyx = ST.save_curlyx;
8095 NOT_REACHED; /* NOTREACHED */
8097 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
8098 cur_curlyx = ST.save_curlyx;
8099 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
8100 cur_curlyx->u.curlyx.count--;
8102 NOT_REACHED; /* NOTREACHED */
8104 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
8106 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
8107 REGCP_UNWIND(ST.lastcp);
8108 regcppop(rex, &maxopenparen);
8109 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
8110 cur_curlyx->u.curlyx.count--;
8112 NOT_REACHED; /* NOTREACHED */
8114 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
8115 REGCP_UNWIND(ST.lastcp);
8116 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
8117 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "WHILEM: failed, trying continuation...\n",
8121 if (cur_curlyx->u.curlyx.count >= REG_INFTY
8122 && ckWARN(WARN_REGEXP)
8123 && !reginfo->warned)
8125 reginfo->warned = TRUE;
8126 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
8127 "Complex regular subexpression recursion limit (%d) "
8133 ST.save_curlyx = cur_curlyx;
8134 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
8135 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
8136 locinput, loceol, script_run_begin);
8137 NOT_REACHED; /* NOTREACHED */
8139 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
8140 cur_curlyx = ST.save_curlyx;
8142 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
8143 /* Maximum greed exceeded */
8144 if (cur_curlyx->u.curlyx.count >= REG_INFTY
8145 && ckWARN(WARN_REGEXP)
8146 && !reginfo->warned)
8148 reginfo->warned = TRUE;
8149 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
8150 "Complex regular subexpression recursion "
8151 "limit (%d) exceeded",
8154 cur_curlyx->u.curlyx.count--;
8158 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "WHILEM: B min fail: trying longer...\n", depth)
8160 /* Try grabbing another A and see if it helps. */
8161 cur_curlyx->u.curlyx.lastloc = locinput;
8162 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
8164 REGCP_SET(ST.lastcp);
8165 PUSH_STATE_GOTO(WHILEM_A_min,
8166 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
8167 locinput, loceol, script_run_begin);
8168 NOT_REACHED; /* NOTREACHED */
8171 #define ST st->u.branch
8173 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
8174 next = scan + ARG(scan);
8177 scan = NEXTOPER(scan);
8180 case BRANCH: /* /(...|A|...)/ */
8181 scan = NEXTOPER(scan); /* scan now points to inner node */
8182 ST.lastparen = rex->lastparen;
8183 ST.lastcloseparen = rex->lastcloseparen;
8184 ST.next_branch = next;
8187 /* Now go into the branch */
8189 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput, loceol,
8192 PUSH_STATE_GOTO(BRANCH_next, scan, locinput, loceol,
8195 NOT_REACHED; /* NOTREACHED */
8197 case CUTGROUP: /* /(*THEN)/ */
8198 sv_yes_mark = st->u.mark.mark_name = scan->flags
8199 ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ])
8201 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput, loceol,
8203 NOT_REACHED; /* NOTREACHED */
8205 case CUTGROUP_next_fail:
8208 if (st->u.mark.mark_name)
8209 sv_commit = st->u.mark.mark_name;
8211 NOT_REACHED; /* NOTREACHED */
8215 NOT_REACHED; /* NOTREACHED */
8217 case BRANCH_next_fail: /* that branch failed; try the next, if any */
8222 REGCP_UNWIND(ST.cp);
8223 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8224 scan = ST.next_branch;
8225 /* no more branches? */
8226 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
8228 Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n",
8235 continue; /* execute next BRANCH[J] op */
8238 case MINMOD: /* next op will be non-greedy, e.g. A*? */
8243 #define ST st->u.curlym
8245 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
8247 /* This is an optimisation of CURLYX that enables us to push
8248 * only a single backtracking state, no matter how many matches
8249 * there are in {m,n}. It relies on the pattern being constant
8250 * length, with no parens to influence future backrefs
8254 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
8256 ST.lastparen = rex->lastparen;
8257 ST.lastcloseparen = rex->lastcloseparen;
8259 /* if paren positive, emulate an OPEN/CLOSE around A */
8261 U32 paren = ST.me->flags;
8262 if (paren > maxopenparen)
8263 maxopenparen = paren;
8264 scan += NEXT_OFF(scan); /* Skip former OPEN. */
8272 ST.c1 = CHRTEST_UNINIT;
8275 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
8278 curlym_do_A: /* execute the A in /A{m,n}B/ */
8279 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput, loceol, /* match A */
8281 NOT_REACHED; /* NOTREACHED */
8283 case CURLYM_A: /* we've just matched an A */
8285 /* after first match, determine A's length: u.curlym.alen */
8286 if (ST.count == 1) {
8287 if (reginfo->is_utf8_target) {
8288 char *s = st->locinput;
8289 while (s < locinput) {
8295 ST.alen = locinput - st->locinput;
8298 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
8301 Perl_re_exec_indentf( aTHX_ "CURLYM now matched %" IVdf " times, len=%" IVdf "...\n",
8302 depth, (IV) ST.count, (IV)ST.alen)
8305 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8309 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
8310 if ( max == REG_INFTY || ST.count < max )
8311 goto curlym_do_A; /* try to match another A */
8313 goto curlym_do_B; /* try to match B */
8315 case CURLYM_A_fail: /* just failed to match an A */
8316 REGCP_UNWIND(ST.cp);
8319 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
8320 || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8323 curlym_do_B: /* execute the B in /A{m,n}B/ */
8324 if (ST.c1 == CHRTEST_UNINIT) {
8325 /* calculate c1 and c2 for possible match of 1st char
8326 * following curly */
8327 ST.c1 = ST.c2 = CHRTEST_VOID;
8329 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
8330 regnode *text_node = ST.B;
8331 if (! HAS_TEXT(text_node))
8332 FIND_NEXT_IMPT(text_node);
8333 if (PL_regkind[OP(text_node)] == EXACT) {
8334 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
8335 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
8345 Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%" IVdf "...\n",
8346 depth, (IV)ST.count)
8348 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
8349 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
8351 /* (We can use memEQ and memNE in this file without
8352 * having to worry about one being shorter than the
8353 * other, since the first byte of each gives the
8354 * length of the character) */
8355 if ( memNE(locinput, ST.c1_utf8, UTF8_SAFE_SKIP(locinput,
8357 && memNE(locinput, ST.c2_utf8, UTF8_SAFE_SKIP(locinput,
8360 /* simulate B failing */
8362 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%" UVXf " c1=0x%" UVXf " c2=0x%" UVXf "\n",
8364 valid_utf8_to_uvchr((U8 *) locinput, NULL),
8365 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
8366 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
8368 state_num = CURLYM_B_fail;
8369 goto reenter_switch;
8372 else if (nextchr != ST.c1 && nextchr != ST.c2) {
8373 /* simulate B failing */
8375 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
8377 (int) nextchr, ST.c1, ST.c2)
8379 state_num = CURLYM_B_fail;
8380 goto reenter_switch;
8385 /* emulate CLOSE: mark current A as captured */
8386 U32 paren = (U32)ST.me->flags;
8388 CLOSE_CAPTURE(paren,
8389 HOPc(locinput, -ST.alen) - reginfo->strbeg,
8390 locinput - reginfo->strbeg);
8393 rex->offs[paren].end = -1;
8395 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8404 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput, loceol, /* match B */
8406 NOT_REACHED; /* NOTREACHED */
8408 case CURLYM_B_fail: /* just failed to match a B */
8409 REGCP_UNWIND(ST.cp);
8410 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8412 I32 max = ARG2(ST.me);
8413 if (max != REG_INFTY && ST.count == max)
8415 goto curlym_do_A; /* try to match a further A */
8417 /* backtrack one A */
8418 if (ST.count == ARG1(ST.me) /* min */)
8421 SET_locinput(HOPc(locinput, -ST.alen));
8422 goto curlym_do_B; /* try to match B */
8425 #define ST st->u.curly
8427 #define CURLY_SETPAREN(paren, success) \
8430 CLOSE_CAPTURE(paren, HOPc(locinput, -1) - reginfo->strbeg, \
8431 locinput - reginfo->strbeg); \
8434 rex->offs[paren].end = -1; \
8435 rex->lastparen = ST.lastparen; \
8436 rex->lastcloseparen = ST.lastcloseparen; \
8440 case STAR: /* /A*B/ where A is width 1 char */
8444 scan = NEXTOPER(scan);
8447 case PLUS: /* /A+B/ where A is width 1 char */
8451 scan = NEXTOPER(scan);
8454 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
8455 ST.paren = scan->flags; /* Which paren to set */
8456 ST.lastparen = rex->lastparen;
8457 ST.lastcloseparen = rex->lastcloseparen;
8458 if (ST.paren > maxopenparen)
8459 maxopenparen = ST.paren;
8460 ST.min = ARG1(scan); /* min to match */
8461 ST.max = ARG2(scan); /* max to match */
8462 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
8464 /* handle the single-char capture called as a GOSUB etc */
8465 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8467 char *li = locinput;
8468 if (!regrepeat(rex, &li, scan, loceol, reginfo, 1))
8476 case CURLY: /* /A{m,n}B/ where A is width 1 char */
8478 ST.min = ARG1(scan); /* min to match */
8479 ST.max = ARG2(scan); /* max to match */
8480 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
8483 * Lookahead to avoid useless match attempts
8484 * when we know what character comes next.
8486 * Used to only do .*x and .*?x, but now it allows
8487 * for )'s, ('s and (?{ ... })'s to be in the way
8488 * of the quantifier and the EXACT-like node. -- japhy
8491 assert(ST.min <= ST.max);
8492 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
8493 ST.c1 = ST.c2 = CHRTEST_VOID;
8496 regnode *text_node = next;
8498 if (! HAS_TEXT(text_node))
8499 FIND_NEXT_IMPT(text_node);
8501 if (! HAS_TEXT(text_node))
8502 ST.c1 = ST.c2 = CHRTEST_VOID;
8504 if ( PL_regkind[OP(text_node)] != EXACT ) {
8505 ST.c1 = ST.c2 = CHRTEST_VOID;
8508 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
8509 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
8521 char *li = locinput;
8524 regrepeat(rex, &li, ST.A, loceol, reginfo, ST.min)
8530 if (ST.c1 == CHRTEST_VOID)
8531 goto curly_try_B_min;
8533 ST.oldloc = locinput;
8535 /* set ST.maxpos to the furthest point along the
8536 * string that could possibly match */
8537 if (ST.max == REG_INFTY) {
8538 ST.maxpos = loceol - 1;
8540 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
8543 else if (utf8_target) {
8544 int m = ST.max - ST.min;
8545 for (ST.maxpos = locinput;
8546 m >0 && ST.maxpos < loceol; m--)
8547 ST.maxpos += UTF8SKIP(ST.maxpos);
8550 ST.maxpos = locinput + ST.max - ST.min;
8551 if (ST.maxpos >= loceol)
8552 ST.maxpos = loceol - 1;
8554 goto curly_try_B_min_known;
8558 /* avoid taking address of locinput, so it can remain
8560 char *li = locinput;
8561 ST.count = regrepeat(rex, &li, ST.A, loceol, reginfo, ST.max);
8562 if (ST.count < ST.min)
8565 if ((ST.count > ST.min)
8566 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
8568 /* A{m,n} must come at the end of the string, there's
8569 * no point in backing off ... */
8571 /* ...except that $ and \Z can match before *and* after
8572 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
8573 We may back off by one in this case. */
8574 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
8578 goto curly_try_B_max;
8580 NOT_REACHED; /* NOTREACHED */
8582 case CURLY_B_min_fail:
8583 /* failed to find B in a non-greedy match.
8584 * Handles both cases where c1,c2 valid or not */
8586 REGCP_UNWIND(ST.cp);
8588 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8591 if (ST.c1 == CHRTEST_VOID) {
8592 /* failed -- move forward one */
8593 char *li = locinput;
8594 if (!regrepeat(rex, &li, ST.A, loceol, reginfo, 1)) {
8599 if (!( ST.count <= ST.max
8600 /* count overflow ? */
8601 || (ST.max == REG_INFTY && ST.count > 0))
8607 /* Couldn't or didn't -- move forward. */
8608 ST.oldloc = locinput;
8610 locinput += UTF8SKIP(locinput);
8615 curly_try_B_min_known:
8616 /* find the next place where 'B' could work, then call B */
8618 n = (ST.oldloc == locinput) ? 0 : 1;
8619 if (ST.c1 == ST.c2) {
8620 /* set n to utf8_distance(oldloc, locinput) */
8621 while ( locinput <= ST.maxpos
8622 && locinput < loceol
8623 && memNE(locinput, ST.c1_utf8,
8624 UTF8_SAFE_SKIP(locinput, reginfo->strend)))
8626 locinput += UTF8_SAFE_SKIP(locinput,
8632 /* set n to utf8_distance(oldloc, locinput) */
8633 while ( locinput <= ST.maxpos
8634 && locinput < loceol
8635 && memNE(locinput, ST.c1_utf8,
8636 UTF8_SAFE_SKIP(locinput, reginfo->strend))
8637 && memNE(locinput, ST.c2_utf8,
8638 UTF8_SAFE_SKIP(locinput, reginfo->strend)))
8640 locinput += UTF8_SAFE_SKIP(locinput, reginfo->strend);
8645 else { /* Not utf8_target */
8646 if (ST.c1 == ST.c2) {
8647 locinput = (char *) memchr(locinput,
8649 ST.maxpos + 1 - locinput);
8651 locinput = ST.maxpos + 1;
8655 U8 c1_c2_bits_differing = ST.c1 ^ ST.c2;
8657 if (! isPOWER_OF_2(c1_c2_bits_differing)) {
8658 while ( locinput <= ST.maxpos
8659 && UCHARAT(locinput) != ST.c1
8660 && UCHARAT(locinput) != ST.c2)
8666 /* If c1 and c2 only differ by a single bit, we can
8667 * avoid a conditional each time through the loop,
8668 * at the expense of a little preliminary setup and
8669 * an extra mask each iteration. By masking out
8670 * that bit, we match exactly two characters, c1
8671 * and c2, and so we don't have to test for both.
8672 * On both ASCII and EBCDIC platforms, most of the
8673 * ASCII-range and Latin1-range folded equivalents
8674 * differ only in a single bit, so this is actually
8675 * the most common case. (e.g. 'A' 0x41 vs 'a'
8677 U8 c1_masked = ST.c1 &~ c1_c2_bits_differing;
8678 U8 c1_c2_mask = ~ c1_c2_bits_differing;
8679 while ( locinput <= ST.maxpos
8680 && (UCHARAT(locinput) & c1_c2_mask)
8687 n = locinput - ST.oldloc;
8689 if (locinput > ST.maxpos)
8692 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
8693 * at b; check that everything between oldloc and
8694 * locinput matches */
8695 char *li = ST.oldloc;
8697 if (regrepeat(rex, &li, ST.A, loceol, reginfo, n) < n)
8699 assert(n == REG_INFTY || locinput == li);
8704 CURLY_SETPAREN(ST.paren, ST.count);
8705 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput, loceol,
8707 NOT_REACHED; /* NOTREACHED */
8711 /* a successful greedy match: now try to match B */
8713 bool could_match = locinput < loceol;
8715 /* If it could work, try it. */
8716 if (ST.c1 != CHRTEST_VOID && could_match) {
8717 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
8719 could_match = memEQ(locinput, ST.c1_utf8,
8720 UTF8_SAFE_SKIP(locinput,
8722 || memEQ(locinput, ST.c2_utf8,
8723 UTF8_SAFE_SKIP(locinput,
8727 could_match = UCHARAT(locinput) == ST.c1
8728 || UCHARAT(locinput) == ST.c2;
8731 if (ST.c1 == CHRTEST_VOID || could_match) {
8732 CURLY_SETPAREN(ST.paren, ST.count);
8733 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput, loceol,
8735 NOT_REACHED; /* NOTREACHED */
8740 case CURLY_B_max_fail:
8741 /* failed to find B in a greedy match */
8743 REGCP_UNWIND(ST.cp);
8745 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8748 if (--ST.count < ST.min)
8750 locinput = HOPc(locinput, -1);
8751 goto curly_try_B_max;
8755 case END: /* last op of main pattern */
8758 /* we've just finished A in /(??{A})B/; now continue with B */
8759 SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput);
8760 st->u.eval.prev_rex = rex_sv; /* inner */
8762 /* Save *all* the positions. */
8763 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
8764 rex_sv = CUR_EVAL.prev_rex;
8765 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
8766 SET_reg_curpm(rex_sv);
8767 rex = ReANY(rex_sv);
8768 rexi = RXi_GET(rex);
8770 st->u.eval.prev_curlyx = cur_curlyx;
8771 cur_curlyx = CUR_EVAL.prev_curlyx;
8773 REGCP_SET(st->u.eval.lastcp);
8775 /* Restore parens of the outer rex without popping the
8777 regcp_restore(rex, CUR_EVAL.lastcp, &maxopenparen);
8779 st->u.eval.prev_eval = cur_eval;
8780 cur_eval = CUR_EVAL.prev_eval;
8782 Perl_re_exec_indentf( aTHX_ "END: EVAL trying tail ... (cur_eval=%p)\n",
8784 if ( nochange_depth )
8787 SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput);
8789 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, /* match B */
8790 st->u.eval.prev_eval->u.eval.B,
8791 locinput, loceol, script_run_begin);
8794 if (locinput < reginfo->till) {
8795 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
8796 "%sEND: Match possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
8798 (long)(locinput - startpos),
8799 (long)(reginfo->till - startpos),
8802 sayNO_SILENT; /* Cannot match: too short. */
8804 sayYES; /* Success! */
8806 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
8808 Perl_re_exec_indentf( aTHX_ "%sSUCCEED: subpattern success...%s\n",
8809 depth, PL_colors[4], PL_colors[5]));
8810 sayYES; /* Success! */
8813 #define ST st->u.ifmatch
8815 case SUSPEND: /* (?>A) */
8817 ST.start = locinput;
8822 case UNLESSM: /* -ve lookaround: (?!A), or with 'flags', (?<!A) */
8824 goto ifmatch_trivial_fail_test;
8826 case IFMATCH: /* +ve lookaround: (?=A), or with 'flags', (?<=A) */
8828 ifmatch_trivial_fail_test:
8829 ST.count = scan->next_off + 1; /* next_off repurposed to be
8830 lookbehind count, requires
8832 if (! scan->flags) { /* 'flags' zero means lookahed */
8834 /* Lookahead starts here and ends at the normal place */
8835 ST.start = locinput;
8839 PERL_UINT_FAST8_T back_count = scan->flags;
8842 /* Lookbehind can look beyond the current position */
8845 /* ... and starts at the first place in the input that is in
8846 * the range of the possible start positions */
8847 for (; ST.count > 0; ST.count--, back_count--) {
8848 s = HOPBACKc(locinput, back_count);
8855 /* If the lookbehind doesn't start in the actual string, is a
8856 * trivial match failure */
8859 sw = 1 - cBOOL(ST.wanted);
8864 /* Here, we didn't want it to match, so is actually success */
8865 next = scan + ARG(scan);
8873 ST.logical = logical;
8874 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
8876 /* execute body of (?...A) */
8877 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), ST.start,
8878 ST.end, script_run_begin);
8879 NOT_REACHED; /* NOTREACHED */
8884 case IFMATCH_A_fail: /* body of (?...A) failed */
8885 if (! ST.logical && ST.count > 1) {
8887 /* It isn't a real failure until we've tried all starting
8888 * positions. Move to the next starting position and retry */
8890 ST.start = HOPc(ST.start, 1);
8892 logical = ST.logical;
8896 /* Here, all starting positions have been tried. */
8900 case IFMATCH_A: /* body of (?...A) succeeded */
8903 sw = matched == ST.wanted;
8904 if (! ST.logical && !sw) {
8908 if (OP(ST.me) != SUSPEND) {
8909 /* restore old position except for (?>...) */
8910 locinput = st->locinput;
8911 loceol = st->loceol;
8912 script_run_begin = st->sr0;
8914 scan = ST.me + ARG(ST.me);
8917 continue; /* execute B */
8922 case LONGJMP: /* alternative with many branches compiles to
8923 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
8924 next = scan + ARG(scan);
8929 case COMMIT: /* (*COMMIT) */
8930 reginfo->cutpoint = loceol;
8933 case PRUNE: /* (*PRUNE) */
8935 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8936 PUSH_STATE_GOTO(COMMIT_next, next, locinput, loceol,
8938 NOT_REACHED; /* NOTREACHED */
8940 case COMMIT_next_fail:
8944 NOT_REACHED; /* NOTREACHED */
8946 case OPFAIL: /* (*FAIL) */
8948 sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8950 /* deal with (?(?!)X|Y) properly,
8951 * make sure we trigger the no branch
8952 * of the trailing IFTHEN structure*/
8958 NOT_REACHED; /* NOTREACHED */
8960 #define ST st->u.mark
8961 case MARKPOINT: /* (*MARK:foo) */
8962 ST.prev_mark = mark_state;
8963 ST.mark_name = sv_commit = sv_yes_mark
8964 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8966 ST.mark_loc = locinput;
8967 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput, loceol,
8969 NOT_REACHED; /* NOTREACHED */
8971 case MARKPOINT_next:
8972 mark_state = ST.prev_mark;
8974 NOT_REACHED; /* NOTREACHED */
8976 case MARKPOINT_next_fail:
8977 if (popmark && sv_eq(ST.mark_name,popmark))
8979 if (ST.mark_loc > startpoint)
8980 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8981 popmark = NULL; /* we found our mark */
8982 sv_commit = ST.mark_name;
8985 Perl_re_exec_indentf( aTHX_ "%sMARKPOINT: next fail: setting cutpoint to mark:%" SVf "...%s\n",
8987 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
8990 mark_state = ST.prev_mark;
8991 sv_yes_mark = mark_state ?
8992 mark_state->u.mark.mark_name : NULL;
8994 NOT_REACHED; /* NOTREACHED */
8996 case SKIP: /* (*SKIP) */
8998 /* (*SKIP) : if we fail we cut here*/
8999 ST.mark_name = NULL;
9000 ST.mark_loc = locinput;
9001 PUSH_STATE_GOTO(SKIP_next,next, locinput, loceol,
9004 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
9005 otherwise do nothing. Meaning we need to scan
9007 regmatch_state *cur = mark_state;
9008 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
9011 if ( sv_eq( cur->u.mark.mark_name,
9014 ST.mark_name = find;
9015 PUSH_STATE_GOTO( SKIP_next, next, locinput, loceol,
9018 cur = cur->u.mark.prev_mark;
9021 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
9024 case SKIP_next_fail:
9026 /* (*CUT:NAME) - Set up to search for the name as we
9027 collapse the stack*/
9028 popmark = ST.mark_name;
9030 /* (*CUT) - No name, we cut here.*/
9031 if (ST.mark_loc > startpoint)
9032 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
9033 /* but we set sv_commit to latest mark_name if there
9034 is one so they can test to see how things lead to this
9037 sv_commit=mark_state->u.mark.mark_name;
9041 NOT_REACHED; /* NOTREACHED */
9044 case LNBREAK: /* \R */
9045 if ((n=is_LNBREAK_safe(locinput, loceol, utf8_target))) {
9052 PerlIO_printf(Perl_error_log, "%" UVxf " %d\n",
9053 PTR2UV(scan), OP(scan));
9054 Perl_croak(aTHX_ "regexp memory corruption");
9056 /* this is a point to jump to in order to increment
9057 * locinput by one character */
9059 assert(!NEXTCHR_IS_EOS);
9061 locinput += PL_utf8skip[nextchr];
9062 /* locinput is allowed to go 1 char off the end (signifying
9063 * EOS), but not 2+ */
9064 if (locinput > loceol)
9073 /* switch break jumps here */
9074 scan = next; /* prepare to execute the next op and ... */
9075 continue; /* ... jump back to the top, reusing st */
9079 /* push a state that backtracks on success */
9080 st->u.yes.prev_yes_state = yes_state;
9084 /* push a new regex state, then continue at scan */
9086 regmatch_state *newst;
9089 regmatch_state *cur = st;
9090 regmatch_state *curyes = yes_state;
9092 regmatch_slab *slab = PL_regmatch_slab;
9093 for (i = 0; i < 3 && i <= depth; cur--,i++) {
9094 if (cur < SLAB_FIRST(slab)) {
9096 cur = SLAB_LAST(slab);
9098 Perl_re_exec_indentf( aTHX_ "%4s #%-3d %-10s %s\n",
9101 depth - i, PL_reg_name[cur->resume_state],
9102 (curyes == cur) ? "yes" : ""
9105 curyes = cur->u.yes.prev_yes_state;
9108 DEBUG_STATE_pp("push")
9111 st->locinput = locinput;
9112 st->loceol = loceol;
9113 st->sr0 = script_run_begin;
9115 if (newst > SLAB_LAST(PL_regmatch_slab))
9116 newst = S_push_slab(aTHX);
9117 PL_regmatch_state = newst;
9119 locinput = pushinput;
9121 script_run_begin = pushsr0;
9127 #ifdef SOLARIS_BAD_OPTIMIZER
9128 # undef PL_charclass
9132 * We get here only if there's trouble -- normally "case END" is
9133 * the terminating point.
9135 Perl_croak(aTHX_ "corrupted regexp pointers");
9136 NOT_REACHED; /* NOTREACHED */
9140 /* we have successfully completed a subexpression, but we must now
9141 * pop to the state marked by yes_state and continue from there */
9142 assert(st != yes_state);
9144 while (st != yes_state) {
9146 if (st < SLAB_FIRST(PL_regmatch_slab)) {
9147 PL_regmatch_slab = PL_regmatch_slab->prev;
9148 st = SLAB_LAST(PL_regmatch_slab);
9152 DEBUG_STATE_pp("pop (no final)");
9154 DEBUG_STATE_pp("pop (yes)");
9160 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
9161 || yes_state > SLAB_LAST(PL_regmatch_slab))
9163 /* not in this slab, pop slab */
9164 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
9165 PL_regmatch_slab = PL_regmatch_slab->prev;
9166 st = SLAB_LAST(PL_regmatch_slab);
9168 depth -= (st - yes_state);
9171 yes_state = st->u.yes.prev_yes_state;
9172 PL_regmatch_state = st;
9175 locinput= st->locinput;
9177 script_run_begin = st->sr0;
9179 state_num = st->resume_state + no_final;
9180 goto reenter_switch;
9183 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n",
9184 PL_colors[4], PL_colors[5]));
9186 if (reginfo->info_aux_eval) {
9187 /* each successfully executed (?{...}) block does the equivalent of
9188 * local $^R = do {...}
9189 * When popping the save stack, all these locals would be undone;
9190 * bypass this by setting the outermost saved $^R to the latest
9192 /* I dont know if this is needed or works properly now.
9193 * see code related to PL_replgv elsewhere in this file.
9196 if (oreplsv != GvSV(PL_replgv)) {
9197 sv_setsv(oreplsv, GvSV(PL_replgv));
9198 SvSETMAGIC(oreplsv);
9206 Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n",
9208 PL_colors[4], PL_colors[5])
9220 /* there's a previous state to backtrack to */
9222 if (st < SLAB_FIRST(PL_regmatch_slab)) {
9223 PL_regmatch_slab = PL_regmatch_slab->prev;
9224 st = SLAB_LAST(PL_regmatch_slab);
9226 PL_regmatch_state = st;
9227 locinput= st->locinput;
9229 script_run_begin = st->sr0;
9231 DEBUG_STATE_pp("pop");
9233 if (yes_state == st)
9234 yes_state = st->u.yes.prev_yes_state;
9236 state_num = st->resume_state + 1; /* failure = success + 1 */
9238 goto reenter_switch;
9243 if (rex->intflags & PREGf_VERBARG_SEEN) {
9244 SV *sv_err = get_sv("REGERROR", 1);
9245 SV *sv_mrk = get_sv("REGMARK", 1);
9247 sv_commit = &PL_sv_no;
9249 sv_yes_mark = &PL_sv_yes;
9252 sv_commit = &PL_sv_yes;
9253 sv_yes_mark = &PL_sv_no;
9257 sv_setsv(sv_err, sv_commit);
9258 sv_setsv(sv_mrk, sv_yes_mark);
9262 if (last_pushed_cv) {
9264 /* see "Some notes about MULTICALL" above */
9266 PERL_UNUSED_VAR(SP);
9269 LEAVE_SCOPE(orig_savestack_ix);
9271 assert(!result || locinput - reginfo->strbeg >= 0);
9272 return result ? locinput - reginfo->strbeg : -1;
9276 - regrepeat - repeatedly match something simple, report how many
9278 * What 'simple' means is a node which can be the operand of a quantifier like
9281 * startposp - pointer to a pointer to the start position. This is updated
9282 * to point to the byte following the highest successful
9284 * p - the regnode to be repeatedly matched against.
9285 * loceol - pointer to the end position beyond which we aren't supposed to
9287 * reginfo - struct holding match state, such as utf8_target
9288 * max - maximum number of things to match.
9289 * depth - (for debugging) backtracking depth.
9292 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
9293 char * loceol, regmatch_info *const reginfo, I32 max _pDEPTH)
9296 char *scan; /* Pointer to current position in target string */
9298 char *this_eol = loceol; /* potentially adjusted version. */
9299 I32 hardcount = 0; /* How many matches so far */
9300 bool utf8_target = reginfo->is_utf8_target;
9301 unsigned int to_complement = 0; /* Invert the result? */
9303 _char_class_number classnum;
9305 PERL_ARGS_ASSERT_REGREPEAT;
9307 /* This routine is structured so that we switch on the input OP. Each OP
9308 * case: statement contains a loop to repeatedly apply the OP, advancing
9309 * the input until it fails, or reaches the end of the input, or until it
9310 * reaches the upper limit of matches. */
9313 if (max == REG_INFTY) /* This is a special marker to go to the platform's
9316 else if (! utf8_target && this_eol - scan > max)
9317 this_eol = scan + max;
9319 /* Here, for the case of a non-UTF-8 target we have adjusted <this_eol> down
9320 * to the maximum of how far we should go in it (leaving it set to the real
9321 * end, if the maximum permissible would take us beyond that). This allows
9322 * us to make the loop exit condition that we haven't gone past <this_eol> to
9323 * also mean that we haven't exceeded the max permissible count, saving a
9324 * test each time through the loops. But it assumes that the OP matches a
9325 * single byte, which is true for most of the OPs below when applied to a
9326 * non-UTF-8 target. Those relatively few OPs that don't have this
9327 * characteristic will have to compensate.
9329 * There is no adjustment for UTF-8 targets, as the number of bytes per
9330 * character varies. OPs will have to test both that the count is less
9331 * than the max permissible (using <hardcount> to keep track), and that we
9332 * are still within the bounds of the string (using <this_eol>. A few OPs
9333 * match a single byte no matter what the encoding. They can omit the max
9334 * test if, for the UTF-8 case, they do the adjustment that was skipped
9337 * Thus, the code above sets things up for the common case; and exceptional
9338 * cases need extra work; the common case is to make sure <scan> doesn't
9339 * go past <this_eol>, and for UTF-8 to also use <hardcount> to make sure the
9340 * count doesn't exceed the maximum permissible */
9345 while (scan < this_eol && hardcount < max && *scan != '\n') {
9346 scan += UTF8SKIP(scan);
9350 scan = (char *) memchr(scan, '\n', this_eol - scan);
9358 while (scan < this_eol && hardcount < max) {
9359 scan += UTF8SKIP(scan);
9367 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9368 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
9369 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
9374 if (! utf8_target) {
9380 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
9384 /* Can use a simple find if the pattern char to match on is invariant
9385 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
9386 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
9387 * true iff it doesn't matter if the argument is in UTF-8 or not */
9388 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
9389 if (utf8_target && this_eol - scan > max) {
9390 /* We didn't adjust <this_eol> because is UTF-8, but ok to do so,
9391 * since here, to match at all, 1 char == 1 byte */
9392 this_eol = scan + max;
9394 scan = (char *) find_span_end((U8 *) scan, (U8 *) this_eol, (U8) c);
9396 else if (reginfo->is_utf8_pat) {
9398 STRLEN scan_char_len;
9400 /* When both target and pattern are UTF-8, we have to do
9402 while (hardcount < max
9404 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
9405 && memEQ(scan, STRING(p), scan_char_len))
9407 scan += scan_char_len;
9411 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
9413 /* Target isn't utf8; convert the character in the UTF-8
9414 * pattern to non-UTF8, and do a simple find */
9415 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
9416 scan = (char *) find_span_end((U8 *) scan, (U8 *) this_eol, (U8) c);
9417 } /* else pattern char is above Latin1, can't possibly match the
9422 /* Here, the string must be utf8; pattern isn't, and <c> is
9423 * different in utf8 than not, so can't compare them directly.
9424 * Outside the loop, find the two utf8 bytes that represent c, and
9425 * then look for those in sequence in the utf8 string */
9426 U8 high = UTF8_TWO_BYTE_HI(c);
9427 U8 low = UTF8_TWO_BYTE_LO(c);
9429 while (hardcount < max
9430 && scan + 1 < this_eol
9431 && UCHARAT(scan) == high
9432 && UCHARAT(scan + 1) == low)
9440 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
9441 assert(! reginfo->is_utf8_pat);
9444 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
9445 if (reginfo->is_utf8_pat || ! utf8_target) {
9447 /* The possible presence of a MICRO SIGN in the pattern forbids us
9448 * to view a non-UTF-8 pattern as folded when there is a UTF-8
9450 utf8_flags |= FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
9455 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9456 utf8_flags = FOLDEQ_LOCALE;
9459 case EXACTF: /* This node only generated for non-utf8 patterns */
9460 assert(! reginfo->is_utf8_pat);
9464 if (! utf8_target) {
9467 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
9468 | FOLDEQ_S2_FOLDS_SANE;
9472 if (! utf8_target) {
9475 assert(reginfo->is_utf8_pat);
9476 utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
9480 utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
9487 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
9489 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
9491 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
9494 if (c1 == CHRTEST_VOID) {
9495 /* Use full Unicode fold matching */
9496 char *tmpeol = loceol;
9497 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
9498 while (hardcount < max
9499 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
9500 STRING(p), NULL, pat_len,
9501 reginfo->is_utf8_pat, utf8_flags))
9508 else if (utf8_target) {
9510 while (scan < this_eol
9512 && memEQ(scan, c1_utf8, UTF8_SAFE_SKIP(scan,
9515 scan += UTF8SKIP(c1_utf8);
9520 while (scan < this_eol
9522 && ( memEQ(scan, c1_utf8, UTF8_SAFE_SKIP(scan,
9524 || memEQ(scan, c2_utf8, UTF8_SAFE_SKIP(scan,
9527 scan += UTF8_SAFE_SKIP(scan, loceol);
9532 else if (c1 == c2) {
9533 scan = (char *) find_span_end((U8 *) scan, (U8 *) this_eol, (U8) c1);
9536 /* See comments in regmatch() CURLY_B_min_known_fail. We avoid
9537 * a conditional each time through the loop if the characters
9538 * differ only in a single bit, as is the usual situation */
9539 U8 c1_c2_bits_differing = c1 ^ c2;
9541 if (isPOWER_OF_2(c1_c2_bits_differing)) {
9542 U8 c1_c2_mask = ~ c1_c2_bits_differing;
9544 scan = (char *) find_span_end_mask((U8 *) scan,
9550 while ( scan < this_eol
9551 && (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
9562 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9564 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) {
9565 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
9571 while (hardcount < max
9573 && reginclass(prog, p, (U8*)scan, (U8*) this_eol, utf8_target))
9575 scan += UTF8SKIP(scan);
9579 else if (ANYOF_FLAGS(p) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
9580 while (scan < this_eol
9581 && reginclass(prog, p, (U8*)scan, (U8*)scan+1, 0))
9585 while (scan < this_eol && ANYOF_BITMAP_TEST(p, *((U8*)scan)))
9591 if (utf8_target && this_eol - scan > max) {
9593 /* We didn't adjust <this_eol> at the beginning of this routine
9594 * because is UTF-8, but it is actually ok to do so, since here, to
9595 * match, 1 char == 1 byte. */
9596 this_eol = scan + max;
9599 scan = (char *) find_span_end_mask((U8 *) scan, (U8 *) this_eol, (U8) ARG(p), FLAGS(p));
9604 while ( hardcount < max
9606 && (*scan & FLAGS(p)) != ARG(p))
9608 scan += UTF8SKIP(scan);
9613 scan = (char *) find_next_masked((U8 *) scan, (U8 *) this_eol, (U8) ARG(p), FLAGS(p));
9618 if (utf8_target) { /* ANYOFH only can match UTF-8 targets */
9619 while ( hardcount < max
9621 && NATIVE_UTF8_TO_I8((U8) *scan) >= ANYOF_FLAGS(p)
9622 && reginclass(prog, p, (U8*)scan, (U8*) this_eol, TRUE))
9624 scan += UTF8SKIP(scan);
9631 if (utf8_target) { /* ANYOFHb only can match UTF-8 targets */
9633 /* we know the first byte must be the FLAGS field */
9634 while ( hardcount < max
9636 && (U8) *scan == ANYOF_FLAGS(p)
9637 && reginclass(prog, p, (U8*)scan, (U8*) this_eol,
9640 scan += UTF8SKIP(scan);
9647 if (utf8_target) { /* ANYOFH only can match UTF-8 targets */
9648 while ( hardcount < max
9650 && inRANGE((U8) NATIVE_UTF8_TO_I8(*scan),
9651 LOWEST_ANYOF_HRx_BYTE(ANYOF_FLAGS(p)),
9652 HIGHEST_ANYOF_HRx_BYTE(ANYOF_FLAGS(p)))
9653 && NATIVE_UTF8_TO_I8((U8) *scan) >= ANYOF_FLAGS(p)
9654 && reginclass(prog, p, (U8*)scan, (U8*) this_eol, TRUE))
9656 scan += UTF8SKIP(scan);
9662 /* The argument (FLAGS) to all the POSIX node types is the class number */
9669 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9670 if (! utf8_target) {
9671 while (scan < this_eol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
9677 while (hardcount < max && scan < this_eol
9678 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
9682 scan += UTF8SKIP(scan);
9695 if (utf8_target && this_eol - scan > max) {
9697 /* We didn't adjust <this_eol> at the beginning of this routine
9698 * because is UTF-8, but it is actually ok to do so, since here, to
9699 * match, 1 char == 1 byte. */
9700 this_eol = scan + max;
9702 while (scan < this_eol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
9715 if (! utf8_target) {
9716 while (scan < this_eol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
9722 /* The complement of something that matches only ASCII matches all
9723 * non-ASCII, plus everything in ASCII that isn't in the class. */
9724 while (hardcount < max && scan < this_eol
9725 && ( ! isASCII_utf8_safe(scan, loceol)
9726 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
9728 scan += UTF8SKIP(scan);
9739 if (! utf8_target) {
9740 while (scan < this_eol && to_complement
9741 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
9748 classnum = (_char_class_number) FLAGS(p);
9751 while ( hardcount < max && scan < this_eol
9752 && to_complement ^ cBOOL(_invlist_contains_cp(
9753 PL_XPosix_ptrs[classnum],
9754 utf8_to_uvchr_buf((U8 *) scan,
9758 scan += UTF8SKIP(scan);
9763 /* For the classes below, the knowledge of how to handle
9764 * every code point is compiled in to Perl via a macro.
9765 * This code is written for making the loops as tight as
9766 * possible. It could be refactored to save space instead.
9769 case _CC_ENUM_SPACE:
9770 while (hardcount < max
9773 ^ cBOOL(isSPACE_utf8_safe(scan, this_eol))))
9775 scan += UTF8SKIP(scan);
9779 case _CC_ENUM_BLANK:
9780 while (hardcount < max
9783 ^ cBOOL(isBLANK_utf8_safe(scan, this_eol))))
9785 scan += UTF8SKIP(scan);
9789 case _CC_ENUM_XDIGIT:
9790 while (hardcount < max
9793 ^ cBOOL(isXDIGIT_utf8_safe(scan, this_eol))))
9795 scan += UTF8SKIP(scan);
9799 case _CC_ENUM_VERTSPACE:
9800 while (hardcount < max
9803 ^ cBOOL(isVERTWS_utf8_safe(scan, this_eol))))
9805 scan += UTF8SKIP(scan);
9809 case _CC_ENUM_CNTRL:
9810 while (hardcount < max
9813 ^ cBOOL(isCNTRL_utf8_safe(scan, this_eol))))
9815 scan += UTF8SKIP(scan);
9825 while (hardcount < max && scan < this_eol &&
9826 (c=is_LNBREAK_utf8_safe(scan, this_eol))) {
9831 /* LNBREAK can match one or two latin chars, which is ok, but we
9832 * have to use hardcount in this situation, and throw away the
9833 * adjustment to <this_eol> done before the switch statement */
9834 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
9843 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9857 /* These are all 0 width, so match right here or not at all. */
9861 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
9862 NOT_REACHED; /* NOTREACHED */
9869 c = scan - *startposp;
9873 GET_RE_DEBUG_FLAGS_DECL;
9875 SV * const prop = sv_newmortal();
9876 regprop(prog, prop, p, reginfo, NULL);
9877 Perl_re_exec_indentf( aTHX_ "%s can match %" IVdf " times out of %" IVdf "...\n",
9878 depth, SvPVX_const(prop),(IV)c,(IV)max);
9886 - reginclass - determine if a character falls into a character class
9888 n is the ANYOF-type regnode
9889 p is the target string
9890 p_end points to one byte beyond the end of the target string
9891 utf8_target tells whether p is in UTF-8.
9893 Returns true if matched; false otherwise.
9895 Note that this can be a synthetic start class, a combination of various
9896 nodes, so things you think might be mutually exclusive, such as locale,
9897 aren't. It can match both locale and non-locale
9902 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
9905 const char flags = (inRANGE(OP(n), ANYOFH, ANYOFHr))
9911 PERL_ARGS_ASSERT_REGINCLASS;
9913 /* If c is not already the code point, get it. Note that
9914 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
9915 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
9917 const U32 utf8n_flags = UTF8_ALLOW_DEFAULT;
9918 c = utf8n_to_uvchr(p, p_end - p, &c_len, utf8n_flags | UTF8_CHECK_ONLY);
9919 if (c_len == (STRLEN)-1) {
9920 _force_out_malformed_utf8_message(p, p_end,
9922 1 /* 1 means die */ );
9923 NOT_REACHED; /* NOTREACHED */
9926 && (OP(n) == ANYOFL || OP(n) == ANYOFPOSIXL)
9927 && ! ANYOFL_UTF8_LOCALE_REQD(flags))
9929 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
9933 /* If this character is potentially in the bitmap, check it */
9934 if (c < NUM_ANYOF_CODE_POINTS && ! inRANGE(OP(n), ANYOFH, ANYOFHb)) {
9935 if (ANYOF_BITMAP_TEST(n, c))
9938 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9945 else if (flags & ANYOF_LOCALE_FLAGS) {
9946 if ( (flags & ANYOFL_FOLD)
9947 && c < sizeof(PL_fold_locale)
9948 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
9952 else if ( ANYOF_POSIXL_TEST_ANY_SET(n)
9953 && c <= U8_MAX /* param to isFOO_lc() */
9956 /* The data structure is arranged so bits 0, 2, 4, ... are set
9957 * if the class includes the Posix character class given by
9958 * bit/2; and 1, 3, 5, ... are set if the class includes the
9959 * complemented Posix class given by int(bit/2). So we loop
9960 * through the bits, each time changing whether we complement
9961 * the result or not. Suppose for the sake of illustration
9962 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
9963 * is set, it means there is a match for this ANYOF node if the
9964 * character is in the class given by the expression (0 / 2 = 0
9965 * = \w). If it is in that class, isFOO_lc() will return 1,
9966 * and since 'to_complement' is 0, the result will stay TRUE,
9967 * and we exit the loop. Suppose instead that bit 0 is 0, but
9968 * bit 1 is 1. That means there is a match if the character
9969 * matches \W. We won't bother to call isFOO_lc() on bit 0,
9970 * but will on bit 1. On the second iteration 'to_complement'
9971 * will be 1, so the exclusive or will reverse things, so we
9972 * are testing for \W. On the third iteration, 'to_complement'
9973 * will be 0, and we would be testing for \s; the fourth
9974 * iteration would test for \S, etc.
9976 * Note that this code assumes that all the classes are closed
9977 * under folding. For example, if a character matches \w, then
9978 * its fold does too; and vice versa. This should be true for
9979 * any well-behaved locale for all the currently defined Posix
9980 * classes, except for :lower: and :upper:, which are handled
9981 * by the pseudo-class :cased: which matches if either of the
9982 * other two does. To get rid of this assumption, an outer
9983 * loop could be used below to iterate over both the source
9984 * character, and its fold (if different) */
9987 int to_complement = 0;
9989 while (count < ANYOF_MAX) {
9990 if (ANYOF_POSIXL_TEST(n, count)
9991 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
10004 /* If the bitmap didn't (or couldn't) match, and something outside the
10005 * bitmap could match, try that. */
10007 if (c >= NUM_ANYOF_CODE_POINTS
10008 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
10010 match = TRUE; /* Everything above the bitmap matches */
10012 /* Here doesn't match everything above the bitmap. If there is
10013 * some information available beyond the bitmap, we may find a
10014 * match in it. If so, this is most likely because the code point
10015 * is outside the bitmap range. But rarely, it could be because of
10016 * some other reason. If so, various flags are set to indicate
10017 * this possibility. On ANYOFD nodes, there may be matches that
10018 * happen only when the target string is UTF-8; or for other node
10019 * types, because runtime lookup is needed, regardless of the
10020 * UTF-8ness of the target string. Finally, under /il, there may
10021 * be some matches only possible if the locale is a UTF-8 one. */
10022 else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP
10023 && ( c >= NUM_ANYOF_CODE_POINTS
10024 || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)
10025 && ( UNLIKELY(OP(n) != ANYOFD)
10026 || (utf8_target && ! isASCII_uni(c)
10027 # if NUM_ANYOF_CODE_POINTS > 256
10031 || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags)
10032 && IN_UTF8_CTYPE_LOCALE)))
10034 SV* only_utf8_locale = NULL;
10035 SV * const definition = _get_regclass_nonbitmap_data(prog, n, TRUE,
10036 0, &only_utf8_locale, NULL);
10042 } else { /* Convert to utf8 */
10043 utf8_p = utf8_buffer;
10044 append_utf8_from_native_byte(*p, &utf8_p);
10045 utf8_p = utf8_buffer;
10048 /* Turkish locales have these hard-coded rules overriding
10050 if ( UNLIKELY(PL_in_utf8_turkic_locale)
10051 && isALPHA_FOLD_EQ(*p, 'i'))
10054 if (_invlist_contains_cp(definition,
10055 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE))
10060 else if (*p == 'I') {
10061 if (_invlist_contains_cp(definition,
10062 LATIN_SMALL_LETTER_DOTLESS_I))
10068 else if (_invlist_contains_cp(definition, c)) {
10072 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
10073 match = _invlist_contains_cp(only_utf8_locale, c);
10077 /* In a Turkic locale under folding, hard-code the I i case pair
10079 if ( UNLIKELY(PL_in_utf8_turkic_locale)
10081 && (flags & ANYOFL_FOLD)
10084 if (c == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
10085 if (ANYOF_BITMAP_TEST(n, 'i')) {
10089 else if (c == LATIN_SMALL_LETTER_DOTLESS_I) {
10090 if (ANYOF_BITMAP_TEST(n, 'I')) {
10096 if (UNICODE_IS_SUPER(c)
10098 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
10100 && ckWARN_d(WARN_NON_UNICODE))
10102 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
10103 "Matched non-Unicode code point 0x%04" UVXf " against Unicode property; may not be portable", c);
10107 #if ANYOF_INVERT != 1
10108 /* Depending on compiler optimization cBOOL takes time, so if don't have to
10110 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
10113 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
10114 return (flags & ANYOF_INVERT) ^ match;
10118 S_reghop3(U8 *s, SSize_t off, const U8* lim)
10120 /* return the position 'off' UTF-8 characters away from 's', forward if
10121 * 'off' >= 0, backwards if negative. But don't go outside of position
10122 * 'lim', which better be < s if off < 0 */
10124 PERL_ARGS_ASSERT_REGHOP3;
10127 while (off-- && s < lim) {
10128 /* XXX could check well-formedness here */
10129 U8 *new_s = s + UTF8SKIP(s);
10130 if (new_s > lim) /* lim may be in the middle of a long character */
10136 while (off++ && s > lim) {
10138 if (UTF8_IS_CONTINUED(*s)) {
10139 while (s > lim && UTF8_IS_CONTINUATION(*s))
10141 if (! UTF8_IS_START(*s)) {
10142 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
10145 /* XXX could check well-formedness here */
10152 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
10154 PERL_ARGS_ASSERT_REGHOP4;
10157 while (off-- && s < rlim) {
10158 /* XXX could check well-formedness here */
10163 while (off++ && s > llim) {
10165 if (UTF8_IS_CONTINUED(*s)) {
10166 while (s > llim && UTF8_IS_CONTINUATION(*s))
10168 if (! UTF8_IS_START(*s)) {
10169 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
10172 /* XXX could check well-formedness here */
10178 /* like reghop3, but returns NULL on overrun, rather than returning last
10182 S_reghopmaybe3(U8* s, SSize_t off, const U8* const lim)
10184 PERL_ARGS_ASSERT_REGHOPMAYBE3;
10187 while (off-- && s < lim) {
10188 /* XXX could check well-formedness here */
10195 while (off++ && s > lim) {
10197 if (UTF8_IS_CONTINUED(*s)) {
10198 while (s > lim && UTF8_IS_CONTINUATION(*s))
10200 if (! UTF8_IS_START(*s)) {
10201 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
10204 /* XXX could check well-formedness here */
10213 /* when executing a regex that may have (?{}), extra stuff needs setting
10214 up that will be visible to the called code, even before the current
10215 match has finished. In particular:
10217 * $_ is localised to the SV currently being matched;
10218 * pos($_) is created if necessary, ready to be updated on each call-out
10220 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
10221 isn't set until the current pattern is successfully finished), so that
10222 $1 etc of the match-so-far can be seen;
10223 * save the old values of subbeg etc of the current regex, and set then
10224 to the current string (again, this is normally only done at the end
10229 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
10232 regexp *const rex = ReANY(reginfo->prog);
10233 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
10235 eval_state->rex = rex;
10236 eval_state->sv = reginfo->sv;
10239 /* Make $_ available to executed code. */
10240 if (reginfo->sv != DEFSV) {
10242 DEFSV_set(reginfo->sv);
10244 /* will be dec'd by S_cleanup_regmatch_info_aux */
10245 SvREFCNT_inc_NN(reginfo->sv);
10247 if (!(mg = mg_find_mglob(reginfo->sv))) {
10248 /* prepare for quick setting of pos */
10249 mg = sv_magicext_mglob(reginfo->sv);
10252 eval_state->pos_magic = mg;
10253 eval_state->pos = mg->mg_len;
10254 eval_state->pos_flags = mg->mg_flags;
10257 eval_state->pos_magic = NULL;
10259 if (!PL_reg_curpm) {
10260 /* PL_reg_curpm is a fake PMOP that we can attach the current
10261 * regex to and point PL_curpm at, so that $1 et al are visible
10262 * within a /(?{})/. It's just allocated once per interpreter the
10263 * first time its needed */
10264 Newxz(PL_reg_curpm, 1, PMOP);
10265 #ifdef USE_ITHREADS
10267 SV* const repointer = &PL_sv_undef;
10268 /* this regexp is also owned by the new PL_reg_curpm, which
10269 will try to free it. */
10270 av_push(PL_regex_padav, repointer);
10271 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
10272 PL_regex_pad = AvARRAY(PL_regex_padav);
10276 SET_reg_curpm(reginfo->prog);
10277 eval_state->curpm = PL_curpm;
10278 PL_curpm_under = PL_curpm;
10279 PL_curpm = PL_reg_curpm;
10280 if (RXp_MATCH_COPIED(rex)) {
10281 /* Here is a serious problem: we cannot rewrite subbeg,
10282 since it may be needed if this match fails. Thus
10283 $` inside (?{}) could fail... */
10284 eval_state->subbeg = rex->subbeg;
10285 eval_state->sublen = rex->sublen;
10286 eval_state->suboffset = rex->suboffset;
10287 eval_state->subcoffset = rex->subcoffset;
10288 #ifdef PERL_ANY_COW
10289 eval_state->saved_copy = rex->saved_copy;
10291 RXp_MATCH_COPIED_off(rex);
10294 eval_state->subbeg = NULL;
10295 rex->subbeg = (char *)reginfo->strbeg;
10296 rex->suboffset = 0;
10297 rex->subcoffset = 0;
10298 rex->sublen = reginfo->strend - reginfo->strbeg;
10302 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
10305 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
10307 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
10308 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
10311 Safefree(aux->poscache);
10315 /* undo the effects of S_setup_eval_state() */
10317 if (eval_state->subbeg) {
10318 regexp * const rex = eval_state->rex;
10319 rex->subbeg = eval_state->subbeg;
10320 rex->sublen = eval_state->sublen;
10321 rex->suboffset = eval_state->suboffset;
10322 rex->subcoffset = eval_state->subcoffset;
10323 #ifdef PERL_ANY_COW
10324 rex->saved_copy = eval_state->saved_copy;
10326 RXp_MATCH_COPIED_on(rex);
10328 if (eval_state->pos_magic)
10330 eval_state->pos_magic->mg_len = eval_state->pos;
10331 eval_state->pos_magic->mg_flags =
10332 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
10333 | (eval_state->pos_flags & MGf_BYTES);
10336 PL_curpm = eval_state->curpm;
10337 SvREFCNT_dec(eval_state->sv);
10340 PL_regmatch_state = aux->old_regmatch_state;
10341 PL_regmatch_slab = aux->old_regmatch_slab;
10343 /* free all slabs above current one - this must be the last action
10344 * of this function, as aux and eval_state are allocated within
10345 * slabs and may be freed here */
10347 s = PL_regmatch_slab->next;
10349 PL_regmatch_slab->next = NULL;
10351 regmatch_slab * const osl = s;
10360 S_to_utf8_substr(pTHX_ regexp *prog)
10362 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
10363 * on the converted value */
10367 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
10370 if (prog->substrs->data[i].substr
10371 && !prog->substrs->data[i].utf8_substr) {
10372 SV* const sv = newSVsv(prog->substrs->data[i].substr);
10373 prog->substrs->data[i].utf8_substr = sv;
10374 sv_utf8_upgrade(sv);
10375 if (SvVALID(prog->substrs->data[i].substr)) {
10376 if (SvTAIL(prog->substrs->data[i].substr)) {
10377 /* Trim the trailing \n that fbm_compile added last
10379 SvCUR_set(sv, SvCUR(sv) - 1);
10380 /* Whilst this makes the SV technically "invalid" (as its
10381 buffer is no longer followed by "\0") when fbm_compile()
10382 adds the "\n" back, a "\0" is restored. */
10383 fbm_compile(sv, FBMcf_TAIL);
10385 fbm_compile(sv, 0);
10387 if (prog->substrs->data[i].substr == prog->check_substr)
10388 prog->check_utf8 = sv;
10394 S_to_byte_substr(pTHX_ regexp *prog)
10396 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
10397 * on the converted value; returns FALSE if can't be converted. */
10401 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
10404 if (prog->substrs->data[i].utf8_substr
10405 && !prog->substrs->data[i].substr) {
10406 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
10407 if (! sv_utf8_downgrade(sv, TRUE)) {
10408 SvREFCNT_dec_NN(sv);
10411 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
10412 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
10413 /* Trim the trailing \n that fbm_compile added last
10415 SvCUR_set(sv, SvCUR(sv) - 1);
10416 fbm_compile(sv, FBMcf_TAIL);
10418 fbm_compile(sv, 0);
10420 prog->substrs->data[i].substr = sv;
10421 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
10422 prog->check_substr = sv;
10429 #ifndef PERL_IN_XSUB_RE
10432 Perl__is_grapheme(pTHX_ const U8 * strbeg, const U8 * s, const U8 * strend, const UV cp)
10434 /* Temporary helper function for toke.c. Verify that the code point 'cp'
10435 * is a stand-alone grapheme. The UTF-8 for 'cp' begins at position 's' in
10436 * the larger string bounded by 'strbeg' and 'strend'.
10438 * 'cp' needs to be assigned (if not a future version of the Unicode
10439 * Standard could make it something that combines with adjacent characters,
10440 * so code using it would then break), and there has to be a GCB break
10441 * before and after the character. */
10445 GCB_enum cp_gcb_val, prev_cp_gcb_val, next_cp_gcb_val;
10446 const U8 * prev_cp_start;
10448 PERL_ARGS_ASSERT__IS_GRAPHEME;
10450 if ( UNLIKELY(UNICODE_IS_SUPER(cp))
10451 || UNLIKELY(UNICODE_IS_NONCHAR(cp)))
10453 /* These are considered graphemes */
10457 /* Otherwise, unassigned code points are forbidden */
10458 if (UNLIKELY(! ELEMENT_RANGE_MATCHES_INVLIST(
10459 _invlist_search(PL_Assigned_invlist, cp))))
10464 cp_gcb_val = getGCB_VAL_CP(cp);
10466 /* Find the GCB value of the previous code point in the input */
10467 prev_cp_start = utf8_hop_back(s, -1, strbeg);
10468 if (UNLIKELY(prev_cp_start == s)) {
10469 prev_cp_gcb_val = GCB_EDGE;
10472 prev_cp_gcb_val = getGCB_VAL_UTF8(prev_cp_start, strend);
10475 /* And check that is a grapheme boundary */
10476 if (! isGCB(prev_cp_gcb_val, cp_gcb_val, strbeg, s,
10477 TRUE /* is UTF-8 encoded */ ))
10482 /* Similarly verify there is a break between the current character and the
10486 next_cp_gcb_val = GCB_EDGE;
10489 next_cp_gcb_val = getGCB_VAL_UTF8(s, strend);
10492 return isGCB(cp_gcb_val, next_cp_gcb_val, strbeg, s, TRUE);
10496 =head1 Unicode Support
10498 =for apidoc isSCRIPT_RUN
10500 Returns a bool as to whether or not the sequence of bytes from C<s> up to but
10501 not including C<send> form a "script run". C<utf8_target> is TRUE iff the
10502 sequence starting at C<s> is to be treated as UTF-8. To be precise, except for
10503 two degenerate cases given below, this function returns TRUE iff all code
10504 points in it come from any combination of three "scripts" given by the Unicode
10505 "Script Extensions" property: Common, Inherited, and possibly one other.
10506 Additionally all decimal digits must come from the same consecutive sequence of
10509 For example, if all the characters in the sequence are Greek, or Common, or
10510 Inherited, this function will return TRUE, provided any decimal digits in it
10511 are from the same block of digits in Common. (These are the ASCII digits
10512 "0".."9" and additionally a block for full width forms of these, and several
10513 others used in mathematical notation.) For scripts (unlike Greek) that have
10514 their own digits defined this will accept either digits from that set or from
10515 one of the Common digit sets, but not a combination of the two. Some scripts,
10516 such as Arabic, have more than one set of digits. All digits must come from
10517 the same set for this function to return TRUE.
10519 C<*ret_script>, if C<ret_script> is not NULL, will on return of TRUE
10520 contain the script found, using the C<SCX_enum> typedef. Its value will be
10521 C<SCX_INVALID> if the function returns FALSE.
10523 If the sequence is empty, TRUE is returned, but C<*ret_script> (if asked for)
10524 will be C<SCX_INVALID>.
10526 If the sequence contains a single code point which is unassigned to a character
10527 in the version of Unicode being used, the function will return TRUE, and the
10528 script will be C<SCX_Unknown>. Any other combination of unassigned code points
10529 in the input sequence will result in the function treating the input as not
10530 being a script run.
10532 The returned script will be C<SCX_Inherited> iff all the code points in it are
10533 from the Inherited script.
10535 Otherwise, the returned script will be C<SCX_Common> iff all the code points in
10536 it are from the Inherited or Common scripts.
10543 Perl_isSCRIPT_RUN(pTHX_ const U8 * s, const U8 * send, const bool utf8_target)
10545 /* Basically, it looks at each character in the sequence to see if the
10546 * above conditions are met; if not it fails. It uses an inversion map to
10547 * find the enum corresponding to the script of each character. But this
10548 * is complicated by the fact that a few code points can be in any of
10549 * several scripts. The data has been constructed so that there are
10550 * additional enum values (all negative) for these situations. The
10551 * absolute value of those is an index into another table which contains
10552 * pointers to auxiliary tables for each such situation. Each aux array
10553 * lists all the scripts for the given situation. There is another,
10554 * parallel, table that gives the number of entries in each aux table.
10555 * These are all defined in charclass_invlists.h */
10557 /* XXX Here are the additional things UTS 39 says could be done:
10559 * Forbid sequences of the same nonspacing mark
10561 * Check to see that all the characters are in the sets of exemplar
10562 * characters for at least one language in the Unicode Common Locale Data
10563 * Repository [CLDR]. */
10567 /* Things that match /\d/u */
10568 SV * decimals_invlist = PL_XPosix_ptrs[_CC_DIGIT];
10569 UV * decimals_array = invlist_array(decimals_invlist);
10571 /* What code point is the digit '0' of the script run? (0 meaning FALSE if
10572 * not currently known) */
10573 UV zero_of_run = 0;
10575 SCX_enum script_of_run = SCX_INVALID; /* Illegal value */
10576 SCX_enum script_of_char = SCX_INVALID;
10578 /* If the script remains not fully determined from iteration to iteration,
10579 * this is the current intersection of the possiblities. */
10580 SCX_enum * intersection = NULL;
10581 PERL_UINT_FAST8_T intersection_len = 0;
10583 bool retval = TRUE;
10584 SCX_enum * ret_script = NULL;
10588 PERL_ARGS_ASSERT_ISSCRIPT_RUN;
10590 /* All code points in 0..255 are either Common or Latin, so must be a
10591 * script run. We can return immediately unless we need to know which
10593 if (! utf8_target && LIKELY(send > s)) {
10594 if (ret_script == NULL) {
10598 /* If any character is Latin, the run is Latin */
10600 if (isALPHA_L1(*s) && LIKELY(*s != MICRO_SIGN_NATIVE)) {
10601 *ret_script = SCX_Latin;
10606 /* Here, all are Common */
10607 *ret_script = SCX_Common;
10611 /* Look at each character in the sequence */
10613 /* If the current character being examined is a digit, this is the code
10614 * point of the zero for its sequence of 10 */
10619 /* The code allows all scripts to use the ASCII digits. This is
10620 * because they are in the Common script. Hence any ASCII ones found
10621 * are ok, unless and until a digit from another set has already been
10622 * encountered. digit ranges in Common are not similarly blessed) */
10623 if (UNLIKELY(isDIGIT(*s))) {
10624 if (UNLIKELY(script_of_run == SCX_Unknown)) {
10629 if (zero_of_run != '0') {
10641 /* Here, isn't an ASCII digit. Find the code point of the character */
10642 if (! UTF8_IS_INVARIANT(*s)) {
10644 cp = valid_utf8_to_uvchr((U8 *) s, &len);
10651 /* If is within the range [+0 .. +9] of the script's zero, it also is a
10652 * digit in that script. We can skip the rest of this code for this
10654 if (UNLIKELY( zero_of_run
10655 && cp >= zero_of_run
10656 && cp - zero_of_run <= 9))
10661 /* Find the character's script. The correct values are hard-coded here
10662 * for small-enough code points. */
10663 if (cp < 0x2B9) { /* From inspection of Unicode db; extremely
10664 unlikely to change */
10666 || ( isALPHA_L1(cp)
10667 && LIKELY(cp != MICRO_SIGN_NATIVE)))
10669 script_of_char = SCX_Latin;
10672 script_of_char = SCX_Common;
10676 script_of_char = _Perl_SCX_invmap[
10677 _invlist_search(PL_SCX_invlist, cp)];
10680 /* We arbitrarily accept a single unassigned character, but not in
10681 * combination with anything else, and not a run of them. */
10682 if ( UNLIKELY(script_of_run == SCX_Unknown)
10683 || UNLIKELY( script_of_run != SCX_INVALID
10684 && script_of_char == SCX_Unknown))
10690 /* For the first character, or the run is inherited, the run's script
10691 * is set to the char's */
10692 if ( UNLIKELY(script_of_run == SCX_INVALID)
10693 || UNLIKELY(script_of_run == SCX_Inherited))
10695 script_of_run = script_of_char;
10698 /* For the character's script to be Unknown, it must be the first
10699 * character in the sequence (for otherwise a test above would have
10700 * prevented us from reaching here), and we have set the run's script
10701 * to it. Nothing further to be done for this character */
10702 if (UNLIKELY(script_of_char == SCX_Unknown)) {
10706 /* We accept 'inherited' script characters currently even at the
10707 * beginning. (We know that no characters in Inherited are digits, or
10708 * we'd have to check for that) */
10709 if (UNLIKELY(script_of_char == SCX_Inherited)) {
10713 /* If the run so far is Common, and the new character isn't, change the
10714 * run's script to that of this character */
10715 if (script_of_run == SCX_Common && script_of_char != SCX_Common) {
10716 script_of_run = script_of_char;
10719 /* Now we can see if the script of the new character is the same as
10720 * that of the run */
10721 if (LIKELY(script_of_char == script_of_run)) {
10722 /* By far the most common case */
10723 goto scripts_match;
10726 /* Here, the script of the run isn't Common. But characters in Common
10727 * match any script */
10728 if (script_of_char == SCX_Common) {
10729 goto scripts_match;
10732 #ifndef HAS_SCX_AUX_TABLES
10734 /* Too early a Unicode version to have a code point belonging to more
10735 * than one script, so, if the scripts don't exactly match, fail */
10736 PERL_UNUSED_VAR(intersection_len);
10742 /* Here there is no exact match between the character's script and the
10743 * run's. And we've handled the special cases of scripts Unknown,
10744 * Inherited, and Common.
10746 * Negative script numbers signify that the value may be any of several
10747 * scripts, and we need to look at auxiliary information to make our
10748 * deterimination. But if both are non-negative, we can fail now */
10749 if (LIKELY(script_of_char >= 0)) {
10750 const SCX_enum * search_in;
10751 PERL_UINT_FAST8_T search_in_len;
10752 PERL_UINT_FAST8_T i;
10754 if (LIKELY(script_of_run >= 0)) {
10759 /* Use the previously constructed set of possible scripts, if any.
10761 if (intersection) {
10762 search_in = intersection;
10763 search_in_len = intersection_len;
10766 search_in = SCX_AUX_TABLE_ptrs[-script_of_run];
10767 search_in_len = SCX_AUX_TABLE_lengths[-script_of_run];
10770 for (i = 0; i < search_in_len; i++) {
10771 if (search_in[i] == script_of_char) {
10772 script_of_run = script_of_char;
10773 goto scripts_match;
10780 else if (LIKELY(script_of_run >= 0)) {
10781 /* script of character could be one of several, but run is a single
10783 const SCX_enum * search_in = SCX_AUX_TABLE_ptrs[-script_of_char];
10784 const PERL_UINT_FAST8_T search_in_len
10785 = SCX_AUX_TABLE_lengths[-script_of_char];
10786 PERL_UINT_FAST8_T i;
10788 for (i = 0; i < search_in_len; i++) {
10789 if (search_in[i] == script_of_run) {
10790 script_of_char = script_of_run;
10791 goto scripts_match;
10799 /* Both run and char could be in one of several scripts. If the
10800 * intersection is empty, then this character isn't in this script
10801 * run. Otherwise, we need to calculate the intersection to use
10802 * for future iterations of the loop, unless we are already at the
10803 * final character */
10804 const SCX_enum * search_char = SCX_AUX_TABLE_ptrs[-script_of_char];
10805 const PERL_UINT_FAST8_T char_len
10806 = SCX_AUX_TABLE_lengths[-script_of_char];
10807 const SCX_enum * search_run;
10808 PERL_UINT_FAST8_T run_len;
10810 SCX_enum * new_overlap = NULL;
10811 PERL_UINT_FAST8_T i, j;
10813 if (intersection) {
10814 search_run = intersection;
10815 run_len = intersection_len;
10818 search_run = SCX_AUX_TABLE_ptrs[-script_of_run];
10819 run_len = SCX_AUX_TABLE_lengths[-script_of_run];
10822 intersection_len = 0;
10824 for (i = 0; i < run_len; i++) {
10825 for (j = 0; j < char_len; j++) {
10826 if (search_run[i] == search_char[j]) {
10828 /* Here, the script at i,j matches. That means this
10829 * character is in the run. But continue on to find
10830 * the complete intersection, for the next loop
10831 * iteration, and for the digit check after it.
10833 * On the first found common script, we malloc space
10834 * for the intersection list for the worst case of the
10835 * intersection, which is the minimum of the number of
10836 * scripts remaining in each set. */
10837 if (intersection_len == 0) {
10839 MIN(run_len - i, char_len - j),
10842 new_overlap[intersection_len++] = search_run[i];
10847 /* Here we've looked through everything. If they have no scripts
10848 * in common, not a run */
10849 if (intersection_len == 0) {
10854 /* If there is only a single script in common, set to that.
10855 * Otherwise, use the intersection going forward */
10856 Safefree(intersection);
10857 intersection = NULL;
10858 if (intersection_len == 1) {
10859 script_of_run = script_of_char = new_overlap[0];
10860 Safefree(new_overlap);
10861 new_overlap = NULL;
10864 intersection = new_overlap;
10872 /* Here, the script of the character is compatible with that of the
10873 * run. That means that in most cases, it continues the script run.
10874 * Either it and the run match exactly, or one or both can be in any of
10875 * several scripts, and the intersection is not empty. However, if the
10876 * character is a decimal digit, it could still mean failure if it is
10877 * from the wrong sequence of 10. So, we need to look at if it's a
10878 * digit. We've already handled the 10 decimal digits, and the next
10879 * lowest one is this one: */
10880 if (cp < FIRST_NON_ASCII_DECIMAL_DIGIT) {
10881 continue; /* Not a digit; this character is part of the run */
10884 /* If we have a definitive '0' for the script of this character, we
10885 * know that for this to be a digit, it must be in the range of +0..+9
10887 if ( script_of_char >= 0
10888 && (zero_of_char = script_zeros[script_of_char]))
10890 if ( cp < zero_of_char
10891 || cp > zero_of_char + 9)
10893 continue; /* Not a digit; this character is part of the run
10898 else { /* Need to look up if this character is a digit or not */
10899 SSize_t index_of_zero_of_char;
10900 index_of_zero_of_char = _invlist_search(decimals_invlist, cp);
10901 if ( UNLIKELY(index_of_zero_of_char < 0)
10902 || ! ELEMENT_RANGE_MATCHES_INVLIST(index_of_zero_of_char))
10904 continue; /* Not a digit; this character is part of the run.
10908 zero_of_char = decimals_array[index_of_zero_of_char];
10911 /* Here, the character is a decimal digit, and the zero of its sequence
10912 * of 10 is in 'zero_of_char'. If we already have a zero for this run,
10913 * they better be the same. */
10915 if (zero_of_run != zero_of_char) {
10920 else { /* Otherwise we now have a zero for this run */
10921 zero_of_run = zero_of_char;
10923 } /* end of looping through CLOSESR text */
10925 Safefree(intersection);
10927 if (ret_script != NULL) {
10929 *ret_script = script_of_run;
10932 *ret_script = SCX_INVALID;
10939 #endif /* ifndef PERL_IN_XSUB_RE */
10942 * ex: set ts=8 sts=4 sw=4 et: