5 * One Ring to rule them all, One Ring to find them
7 * [p.v of _The Lord of the Rings_, opening poem]
8 * [p.50 of _The Lord of the Rings_, I/iii: "The Shadow of the Past"]
9 * [p.254 of _The Lord of the Rings_, II/ii: "The Council of Elrond"]
12 /* This file contains functions for executing a regular expression. See
13 * also regcomp.c which funnily enough, contains functions for compiling
14 * a regular expression.
16 * This file is also copied at build time to ext/re/re_exec.c, where
17 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
18 * This causes the main functions to be compiled under new names and with
19 * debugging support added, which makes "use re 'debug'" work.
22 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
23 * confused with the original package (see point 3 below). Thanks, Henry!
26 /* Additional note: this code is very heavily munged from Henry's version
27 * in places. In some spots I've traded clarity for efficiency, so don't
28 * blame Henry for some of the lack of readability.
31 /* The names of the functions have been changed from regcomp and
32 * regexec to pregcomp and pregexec in order to avoid conflicts
33 * with the POSIX routines of the same names.
36 #ifdef PERL_EXT_RE_BUILD
41 * pregcomp and pregexec -- regsub and regerror are not used in perl
43 * Copyright (c) 1986 by University of Toronto.
44 * Written by Henry Spencer. Not derived from licensed software.
46 * Permission is granted to anyone to use this software for any
47 * purpose on any computer system, and to redistribute it freely,
48 * subject to the following restrictions:
50 * 1. The author is not responsible for the consequences of use of
51 * this software, no matter how awful, even if they arise
54 * 2. The origin of this software must not be misrepresented, either
55 * by explicit claim or by omission.
57 * 3. Altered versions must be plainly marked as such, and must not
58 * be misrepresented as being the original software.
60 **** Alterations to Henry's code are...
62 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
63 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
64 **** by Larry Wall and others
66 **** You may distribute under the terms of either the GNU General Public
67 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGEXEC_C
77 #ifdef PERL_IN_XSUB_RE
83 #include "invlist_inline.h"
84 #include "unicode_constants.h"
86 #define B_ON_NON_UTF8_LOCALE_IS_WRONG \
87 "Use of \\b{} or \\B{} for non-UTF-8 locale is wrong. Assuming a UTF-8 locale"
89 static const char utf8_locale_required[] =
90 "Use of (?[ ]) for non-UTF-8 locale is wrong. Assuming a UTF-8 locale";
93 /* At least one required character in the target string is expressible only in
95 static const char* const non_utf8_target_but_utf8_required
96 = "Can't match, because target string needs to be in UTF-8\n";
99 #define NON_UTF8_TARGET_BUT_UTF8_REQUIRED(target) STMT_START { \
100 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%s", non_utf8_target_but_utf8_required));\
104 #define HAS_NONLATIN1_FOLD_CLOSURE(i) _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
107 #define STATIC static
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 NEXTCHR_EOS -10 /* nextchr has fallen off the end */
156 #define NEXTCHR_IS_EOS (nextchr < 0)
158 #define SET_nextchr __ASSERT_(locinput <= reginfo->strend) \
159 nextchr = ((locinput < reginfo->strend) ? UCHARAT(locinput) : NEXTCHR_EOS)
161 #define SET_locinput(p) \
165 #define PLACEHOLDER /* Something for the preprocessor to grab onto */
166 /* TODO: Combine JUMPABLE and HAS_TEXT to cache OP(rn) */
168 /* for use after a quantifier and before an EXACT-like node -- japhy */
169 /* it would be nice to rework regcomp.sym to generate this stuff. sigh
171 * NOTE that *nothing* that affects backtracking should be in here, specifically
172 * VERBS must NOT be included. JUMPABLE is used to determine if we can ignore a
173 * node that is in between two EXACT like nodes when ascertaining what the required
174 * "follow" character is. This should probably be moved to regex compile time
175 * although it may be done at run time beause of the REF possibility - more
176 * investigation required. -- demerphq
178 #define JUMPABLE(rn) ( \
180 (OP(rn) == CLOSE && \
181 !EVAL_CLOSE_PAREN_IS(cur_eval,ARG(rn)) ) || \
183 OP(rn) == SUSPEND || OP(rn) == IFMATCH || \
184 OP(rn) == PLUS || OP(rn) == MINMOD || \
186 (PL_regkind[OP(rn)] == CURLY && ARG1(rn) > 0) \
188 #define IS_EXACT(rn) (PL_regkind[OP(rn)] == EXACT)
190 #define HAS_TEXT(rn) ( IS_EXACT(rn) || PL_regkind[OP(rn)] == REF )
193 Search for mandatory following text node; for lookahead, the text must
194 follow but for lookbehind (rn->flags != 0) we skip to the next step.
196 #define FIND_NEXT_IMPT(rn) STMT_START { \
197 while (JUMPABLE(rn)) { \
198 const OPCODE type = OP(rn); \
199 if (type == SUSPEND || PL_regkind[type] == CURLY) \
200 rn = NEXTOPER(NEXTOPER(rn)); \
201 else if (type == PLUS) \
203 else if (type == IFMATCH) \
204 rn = (rn->flags == 0) ? NEXTOPER(NEXTOPER(rn)) : rn + ARG(rn); \
205 else rn += NEXT_OFF(rn); \
209 #define SLAB_FIRST(s) (&(s)->states[0])
210 #define SLAB_LAST(s) (&(s)->states[PERL_REGMATCH_SLAB_SLOTS-1])
212 static void S_setup_eval_state(pTHX_ regmatch_info *const reginfo);
213 static void S_cleanup_regmatch_info_aux(pTHX_ void *arg);
214 static regmatch_state * S_push_slab(pTHX);
216 #define REGCP_PAREN_ELEMS 3
217 #define REGCP_OTHER_ELEMS 3
218 #define REGCP_FRAME_ELEMS 1
219 /* REGCP_FRAME_ELEMS are not part of the REGCP_OTHER_ELEMS and
220 * are needed for the regexp context stack bookkeeping. */
223 S_regcppush(pTHX_ const regexp *rex, I32 parenfloor, U32 maxopenparen _pDEPTH)
225 const int retval = PL_savestack_ix;
226 const int paren_elems_to_push =
227 (maxopenparen - parenfloor) * REGCP_PAREN_ELEMS;
228 const UV total_elems = paren_elems_to_push + REGCP_OTHER_ELEMS;
229 const UV elems_shifted = total_elems << SAVE_TIGHT_SHIFT;
231 GET_RE_DEBUG_FLAGS_DECL;
233 PERL_ARGS_ASSERT_REGCPPUSH;
235 if (paren_elems_to_push < 0)
236 Perl_croak(aTHX_ "panic: paren_elems_to_push, %i < 0, maxopenparen: %i parenfloor: %i REGCP_PAREN_ELEMS: %u",
237 (int)paren_elems_to_push, (int)maxopenparen,
238 (int)parenfloor, (unsigned)REGCP_PAREN_ELEMS);
240 if ((elems_shifted >> SAVE_TIGHT_SHIFT) != total_elems)
241 Perl_croak(aTHX_ "panic: paren_elems_to_push offset %" UVuf
242 " out of range (%lu-%ld)",
244 (unsigned long)maxopenparen,
247 SSGROW(total_elems + REGCP_FRAME_ELEMS);
250 if ((int)maxopenparen > (int)parenfloor)
251 Perl_re_exec_indentf( aTHX_
252 "rex=0x%" UVxf " offs=0x%" UVxf ": saving capture indices:\n",
258 for (p = parenfloor+1; p <= (I32)maxopenparen; p++) {
259 /* REGCP_PARENS_ELEMS are pushed per pairs of parentheses. */
260 SSPUSHIV(rex->offs[p].end);
261 SSPUSHIV(rex->offs[p].start);
262 SSPUSHINT(rex->offs[p].start_tmp);
263 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
264 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "\n",
267 (IV)rex->offs[p].start,
268 (IV)rex->offs[p].start_tmp,
272 /* REGCP_OTHER_ELEMS are pushed in any case, parentheses or no. */
273 SSPUSHINT(maxopenparen);
274 SSPUSHINT(rex->lastparen);
275 SSPUSHINT(rex->lastcloseparen);
276 SSPUSHUV(SAVEt_REGCONTEXT | elems_shifted); /* Magic cookie. */
281 /* These are needed since we do not localize EVAL nodes: */
282 #define REGCP_SET(cp) \
284 Perl_re_exec_indentf( aTHX_ \
285 "Setting an EVAL scope, savestack=%" IVdf ",\n", \
286 depth, (IV)PL_savestack_ix \
291 #define REGCP_UNWIND(cp) \
293 if (cp != PL_savestack_ix) \
294 Perl_re_exec_indentf( aTHX_ \
295 "Clearing an EVAL scope, savestack=%" \
296 IVdf "..%" IVdf "\n", \
297 depth, (IV)(cp), (IV)PL_savestack_ix \
302 /* set the start and end positions of capture ix */
303 #define CLOSE_CAPTURE(ix, s, e) \
304 rex->offs[ix].start = s; \
305 rex->offs[ix].end = e; \
306 if (ix > rex->lastparen) \
307 rex->lastparen = ix; \
308 rex->lastcloseparen = ix; \
309 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
310 "CLOSE: rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf "..%" IVdf " max: %" UVuf "\n", \
315 (IV)rex->offs[ix].start, \
316 (IV)rex->offs[ix].end, \
320 #define UNWIND_PAREN(lp, lcp) \
321 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_ \
322 "UNWIND_PAREN: rex=0x%" UVxf " offs=0x%" UVxf ": invalidate (%" UVuf "..%" UVuf "] set lcp: %" UVuf "\n", \
327 (UV)(rex->lastparen), \
330 for (n = rex->lastparen; n > lp; n--) \
331 rex->offs[n].end = -1; \
332 rex->lastparen = n; \
333 rex->lastcloseparen = lcp;
337 S_regcppop(pTHX_ regexp *rex, U32 *maxopenparen_p _pDEPTH)
341 GET_RE_DEBUG_FLAGS_DECL;
343 PERL_ARGS_ASSERT_REGCPPOP;
345 /* Pop REGCP_OTHER_ELEMS before the parentheses loop starts. */
347 assert((i & SAVE_MASK) == SAVEt_REGCONTEXT); /* Check that the magic cookie is there. */
348 i >>= SAVE_TIGHT_SHIFT; /* Parentheses elements to pop. */
349 rex->lastcloseparen = SSPOPINT;
350 rex->lastparen = SSPOPINT;
351 *maxopenparen_p = SSPOPINT;
353 i -= REGCP_OTHER_ELEMS;
354 /* Now restore the parentheses context. */
356 if (i || rex->lastparen + 1 <= rex->nparens)
357 Perl_re_exec_indentf( aTHX_
358 "rex=0x%" UVxf " offs=0x%" UVxf ": restoring capture indices to:\n",
364 paren = *maxopenparen_p;
365 for ( ; i > 0; i -= REGCP_PAREN_ELEMS) {
367 rex->offs[paren].start_tmp = SSPOPINT;
368 rex->offs[paren].start = SSPOPIV;
370 if (paren <= rex->lastparen)
371 rex->offs[paren].end = tmps;
372 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
373 " \\%" UVuf ": %" IVdf "(%" IVdf ")..%" IVdf "%s\n",
376 (IV)rex->offs[paren].start,
377 (IV)rex->offs[paren].start_tmp,
378 (IV)rex->offs[paren].end,
379 (paren > rex->lastparen ? "(skipped)" : ""));
384 /* It would seem that the similar code in regtry()
385 * already takes care of this, and in fact it is in
386 * a better location to since this code can #if 0-ed out
387 * but the code in regtry() is needed or otherwise tests
388 * requiring null fields (pat.t#187 and split.t#{13,14}
389 * (as of patchlevel 7877) will fail. Then again,
390 * this code seems to be necessary or otherwise
391 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
392 * --jhi updated by dapm */
393 for (i = rex->lastparen + 1; i <= rex->nparens; i++) {
394 if (i > *maxopenparen_p)
395 rex->offs[i].start = -1;
396 rex->offs[i].end = -1;
397 DEBUG_BUFFERS_r( Perl_re_exec_indentf( aTHX_
398 " \\%" UVuf ": %s ..-1 undeffing\n",
401 (i > *maxopenparen_p) ? "-1" : " "
407 /* restore the parens and associated vars at savestack position ix,
408 * but without popping the stack */
411 S_regcp_restore(pTHX_ regexp *rex, I32 ix, U32 *maxopenparen_p _pDEPTH)
413 I32 tmpix = PL_savestack_ix;
414 PERL_ARGS_ASSERT_REGCP_RESTORE;
416 PL_savestack_ix = ix;
417 regcppop(rex, maxopenparen_p);
418 PL_savestack_ix = tmpix;
421 #define regcpblow(cp) LEAVE_SCOPE(cp) /* Ignores regcppush()ed data. */
423 #ifndef PERL_IN_XSUB_RE
426 Perl_isFOO_lc(pTHX_ const U8 classnum, const U8 character)
428 /* Returns a boolean as to whether or not 'character' is a member of the
429 * Posix character class given by 'classnum' that should be equivalent to a
430 * value in the typedef '_char_class_number'.
432 * Ideally this could be replaced by a just an array of function pointers
433 * to the C library functions that implement the macros this calls.
434 * However, to compile, the precise function signatures are required, and
435 * these may vary from platform to to platform. To avoid having to figure
436 * out what those all are on each platform, I (khw) am using this method,
437 * which adds an extra layer of function call overhead (unless the C
438 * optimizer strips it away). But we don't particularly care about
439 * performance with locales anyway. */
441 switch ((_char_class_number) classnum) {
442 case _CC_ENUM_ALPHANUMERIC: return isALPHANUMERIC_LC(character);
443 case _CC_ENUM_ALPHA: return isALPHA_LC(character);
444 case _CC_ENUM_ASCII: return isASCII_LC(character);
445 case _CC_ENUM_BLANK: return isBLANK_LC(character);
446 case _CC_ENUM_CASED: return isLOWER_LC(character)
447 || isUPPER_LC(character);
448 case _CC_ENUM_CNTRL: return isCNTRL_LC(character);
449 case _CC_ENUM_DIGIT: return isDIGIT_LC(character);
450 case _CC_ENUM_GRAPH: return isGRAPH_LC(character);
451 case _CC_ENUM_LOWER: return isLOWER_LC(character);
452 case _CC_ENUM_PRINT: return isPRINT_LC(character);
453 case _CC_ENUM_PUNCT: return isPUNCT_LC(character);
454 case _CC_ENUM_SPACE: return isSPACE_LC(character);
455 case _CC_ENUM_UPPER: return isUPPER_LC(character);
456 case _CC_ENUM_WORDCHAR: return isWORDCHAR_LC(character);
457 case _CC_ENUM_XDIGIT: return isXDIGIT_LC(character);
458 default: /* VERTSPACE should never occur in locales */
459 Perl_croak(aTHX_ "panic: isFOO_lc() has an unexpected character class '%d'", classnum);
462 NOT_REACHED; /* NOTREACHED */
468 PERL_STATIC_INLINE I32
469 S_foldEQ_latin1_s2_folded(const char *s1, const char *s2, I32 len)
471 /* Compare non-UTF-8 using Unicode (Latin1) semantics. s2 must already be
472 * folded. Works on all folds representable without UTF-8, except for
473 * LATIN_SMALL_LETTER_SHARP_S, and does not check for this. Nor does it
474 * check that the strings each have at least 'len' characters.
476 * There is almost an identical API function where s2 need not be folded:
477 * Perl_foldEQ_latin1() */
479 const U8 *a = (const U8 *)s1;
480 const U8 *b = (const U8 *)s2;
482 PERL_ARGS_ASSERT_FOLDEQ_LATIN1_S2_FOLDED;
487 assert(! isUPPER_L1(*b));
488 if (toLOWER_L1(*a) != *b) {
497 S_isFOO_utf8_lc(pTHX_ const U8 classnum, const U8* character, const U8* e)
499 /* Returns a boolean as to whether or not the (well-formed) UTF-8-encoded
500 * 'character' is a member of the Posix character class given by 'classnum'
501 * that should be equivalent to a value in the typedef
502 * '_char_class_number'.
504 * This just calls isFOO_lc on the code point for the character if it is in
505 * the range 0-255. Outside that range, all characters use Unicode
506 * rules, ignoring any locale. So use the Unicode function if this class
507 * requires an inversion list, and use the Unicode macro otherwise. */
511 PERL_ARGS_ASSERT_ISFOO_UTF8_LC;
513 if (UTF8_IS_INVARIANT(*character)) {
514 return isFOO_lc(classnum, *character);
516 else if (UTF8_IS_DOWNGRADEABLE_START(*character)) {
517 return isFOO_lc(classnum,
518 EIGHT_BIT_UTF8_TO_NATIVE(*character, *(character + 1)));
521 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(character, e);
523 switch ((_char_class_number) classnum) {
524 case _CC_ENUM_SPACE: return is_XPERLSPACE_high(character);
525 case _CC_ENUM_BLANK: return is_HORIZWS_high(character);
526 case _CC_ENUM_XDIGIT: return is_XDIGIT_high(character);
527 case _CC_ENUM_VERTSPACE: return is_VERTWS_high(character);
529 return _invlist_contains_cp(PL_XPosix_ptrs[classnum],
530 utf8_to_uvchr_buf(character, e, NULL));
533 return FALSE; /* Things like CNTRL are always below 256 */
537 S_find_span_end(U8 * s, const U8 * send, const U8 span_byte)
539 /* Returns the position of the first byte in the sequence between 's' and
540 * 'send-1' inclusive that isn't 'span_byte'; returns 'send' if none found.
543 PERL_ARGS_ASSERT_FIND_SPAN_END;
547 if ((STRLEN) (send - s) >= PERL_WORDSIZE
548 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
549 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
551 PERL_UINTMAX_T span_word;
553 /* Process per-byte until reach word boundary. XXX This loop could be
554 * eliminated if we knew that this platform had fast unaligned reads */
555 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
556 if (*s != span_byte) {
562 /* Create a word filled with the bytes we are spanning */
563 span_word = PERL_COUNT_MULTIPLIER * span_byte;
565 /* Process per-word as long as we have at least a full word left */
568 /* Keep going if the whole word is composed of 'span_byte's */
569 if ((* (PERL_UINTMAX_T *) s) == span_word) {
574 /* Here, at least one byte in the word isn't 'span_byte'. */
582 /* This xor leaves 1 bits only in those non-matching bytes */
583 span_word ^= * (PERL_UINTMAX_T *) s;
585 /* Make sure the upper bit of each non-matching byte is set. This
586 * makes each such byte look like an ASCII platform variant byte */
587 span_word |= span_word << 1;
588 span_word |= span_word << 2;
589 span_word |= span_word << 4;
591 /* That reduces the problem to what this function solves */
592 return s + _variant_byte_number(span_word);
596 } while (s + PERL_WORDSIZE <= send);
599 /* Process the straggler bytes beyond the final word boundary */
601 if (*s != span_byte) {
611 S_find_next_masked(U8 * s, const U8 * send, const U8 byte, const U8 mask)
613 /* Returns the position of the first byte in the sequence between 's'
614 * and 'send-1' inclusive that when ANDed with 'mask' yields 'byte';
615 * returns 'send' if none found. It uses word-level operations instead of
616 * byte to speed up the process */
618 PERL_ARGS_ASSERT_FIND_NEXT_MASKED;
621 assert((byte & mask) == byte);
625 if ((STRLEN) (send - s) >= PERL_WORDSIZE
626 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
627 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
629 PERL_UINTMAX_T word, mask_word;
631 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
632 if (((*s) & mask) == byte) {
638 word = PERL_COUNT_MULTIPLIER * byte;
639 mask_word = PERL_COUNT_MULTIPLIER * mask;
642 PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;
644 /* If 'masked' contains bytes with the bit pattern of 'byte' within
645 * it, xoring with 'word' will leave each of the 8 bits in such
646 * bytes be 0, and no byte containing any other bit pattern will be
650 /* This causes the most significant bit to be set to 1 for any
651 * bytes in the word that aren't completely 0 */
652 masked |= masked << 1;
653 masked |= masked << 2;
654 masked |= masked << 4;
656 /* The msbits are the same as what marks a byte as variant, so we
657 * can use this mask. If all msbits are 1, the word doesn't
659 if ((masked & PERL_VARIANTS_WORD_MASK) == PERL_VARIANTS_WORD_MASK) {
664 /* Here, the msbit of bytes in the word that aren't 'byte' are 1,
665 * and any that are, are 0. Complement and re-AND to swap that */
667 masked &= PERL_VARIANTS_WORD_MASK;
669 /* This reduces the problem to that solved by this function */
670 s += _variant_byte_number(masked);
673 } while (s + PERL_WORDSIZE <= send);
679 if (((*s) & mask) == byte) {
689 S_find_span_end_mask(U8 * s, const U8 * send, const U8 span_byte, const U8 mask)
691 /* Returns the position of the first byte in the sequence between 's' and
692 * 'send-1' inclusive that when ANDed with 'mask' isn't 'span_byte'.
693 * 'span_byte' should have been ANDed with 'mask' in the call of this
694 * function. Returns 'send' if none found. Works like find_span_end(),
695 * except for the AND */
697 PERL_ARGS_ASSERT_FIND_SPAN_END_MASK;
700 assert((span_byte & mask) == span_byte);
702 if ((STRLEN) (send - s) >= PERL_WORDSIZE
703 + PERL_WORDSIZE * PERL_IS_SUBWORD_ADDR(s)
704 - (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK))
706 PERL_UINTMAX_T span_word, mask_word;
708 while (PTR2nat(s) & PERL_WORD_BOUNDARY_MASK) {
709 if (((*s) & mask) != span_byte) {
715 span_word = PERL_COUNT_MULTIPLIER * span_byte;
716 mask_word = PERL_COUNT_MULTIPLIER * mask;
719 PERL_UINTMAX_T masked = (* (PERL_UINTMAX_T *) s) & mask_word;
721 if (masked == span_word) {
733 masked |= masked << 1;
734 masked |= masked << 2;
735 masked |= masked << 4;
736 return s + _variant_byte_number(masked);
740 } while (s + PERL_WORDSIZE <= send);
744 if (((*s) & mask) != span_byte) {
754 * pregexec and friends
757 #ifndef PERL_IN_XSUB_RE
759 - pregexec - match a regexp against a string
762 Perl_pregexec(pTHX_ REGEXP * const prog, char* stringarg, char *strend,
763 char *strbeg, SSize_t minend, SV *screamer, U32 nosave)
764 /* stringarg: the point in the string at which to begin matching */
765 /* strend: pointer to null at end of string */
766 /* strbeg: real beginning of string */
767 /* minend: end of match must be >= minend bytes after stringarg. */
768 /* screamer: SV being matched: only used for utf8 flag, pos() etc; string
769 * itself is accessed via the pointers above */
770 /* nosave: For optimizations. */
772 PERL_ARGS_ASSERT_PREGEXEC;
775 regexec_flags(prog, stringarg, strend, strbeg, minend, screamer, NULL,
776 nosave ? 0 : REXEC_COPY_STR);
782 /* re_intuit_start():
784 * Based on some optimiser hints, try to find the earliest position in the
785 * string where the regex could match.
787 * rx: the regex to match against
788 * sv: the SV being matched: only used for utf8 flag; the string
789 * itself is accessed via the pointers below. Note that on
790 * something like an overloaded SV, SvPOK(sv) may be false
791 * and the string pointers may point to something unrelated to
793 * strbeg: real beginning of string
794 * strpos: the point in the string at which to begin matching
795 * strend: pointer to the byte following the last char of the string
796 * flags currently unused; set to 0
797 * data: currently unused; set to NULL
799 * The basic idea of re_intuit_start() is to use some known information
800 * about the pattern, namely:
802 * a) the longest known anchored substring (i.e. one that's at a
803 * constant offset from the beginning of the pattern; but not
804 * necessarily at a fixed offset from the beginning of the
806 * b) the longest floating substring (i.e. one that's not at a constant
807 * offset from the beginning of the pattern);
808 * c) Whether the pattern is anchored to the string; either
809 * an absolute anchor: /^../, or anchored to \n: /^.../m,
810 * or anchored to pos(): /\G/;
811 * d) A start class: a real or synthetic character class which
812 * represents which characters are legal at the start of the pattern;
814 * to either quickly reject the match, or to find the earliest position
815 * within the string at which the pattern might match, thus avoiding
816 * running the full NFA engine at those earlier locations, only to
817 * eventually fail and retry further along.
819 * Returns NULL if the pattern can't match, or returns the address within
820 * the string which is the earliest place the match could occur.
822 * The longest of the anchored and floating substrings is called 'check'
823 * and is checked first. The other is called 'other' and is checked
824 * second. The 'other' substring may not be present. For example,
826 * /(abc|xyz)ABC\d{0,3}DEFG/
830 * check substr (float) = "DEFG", offset 6..9 chars
831 * other substr (anchored) = "ABC", offset 3..3 chars
834 * Be aware that during the course of this function, sometimes 'anchored'
835 * refers to a substring being anchored relative to the start of the
836 * pattern, and sometimes to the pattern itself being anchored relative to
837 * the string. For example:
839 * /\dabc/: "abc" is anchored to the pattern;
840 * /^\dabc/: "abc" is anchored to the pattern and the string;
841 * /\d+abc/: "abc" is anchored to neither the pattern nor the string;
842 * /^\d+abc/: "abc" is anchored to neither the pattern nor the string,
843 * but the pattern is anchored to the string.
847 Perl_re_intuit_start(pTHX_
850 const char * const strbeg,
854 re_scream_pos_data *data)
856 struct regexp *const prog = ReANY(rx);
857 SSize_t start_shift = prog->check_offset_min;
858 /* Should be nonnegative! */
859 SSize_t end_shift = 0;
860 /* current lowest pos in string where the regex can start matching */
861 char *rx_origin = strpos;
863 const bool utf8_target = (sv && SvUTF8(sv)) ? 1 : 0; /* if no sv we have to assume bytes */
864 U8 other_ix = 1 - prog->substrs->check_ix;
866 char *other_last = strpos;/* latest pos 'other' substr already checked to */
867 char *check_at = NULL; /* check substr found at this pos */
868 const I32 multiline = prog->extflags & RXf_PMf_MULTILINE;
869 RXi_GET_DECL(prog,progi);
870 regmatch_info reginfo_buf; /* create some info to pass to find_byclass */
871 regmatch_info *const reginfo = ®info_buf;
872 GET_RE_DEBUG_FLAGS_DECL;
874 PERL_ARGS_ASSERT_RE_INTUIT_START;
875 PERL_UNUSED_ARG(flags);
876 PERL_UNUSED_ARG(data);
878 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
879 "Intuit: trying to determine minimum start position...\n"));
881 /* for now, assume that all substr offsets are positive. If at some point
882 * in the future someone wants to do clever things with lookbehind and
883 * -ve offsets, they'll need to fix up any code in this function
884 * which uses these offsets. See the thread beginning
885 * <20140113145929.GF27210@iabyn.com>
887 assert(prog->substrs->data[0].min_offset >= 0);
888 assert(prog->substrs->data[0].max_offset >= 0);
889 assert(prog->substrs->data[1].min_offset >= 0);
890 assert(prog->substrs->data[1].max_offset >= 0);
891 assert(prog->substrs->data[2].min_offset >= 0);
892 assert(prog->substrs->data[2].max_offset >= 0);
894 /* for now, assume that if both present, that the floating substring
895 * doesn't start before the anchored substring.
896 * If you break this assumption (e.g. doing better optimisations
897 * with lookahead/behind), then you'll need to audit the code in this
898 * function carefully first
901 ! ( (prog->anchored_utf8 || prog->anchored_substr)
902 && (prog->float_utf8 || prog->float_substr))
903 || (prog->float_min_offset >= prog->anchored_offset));
905 /* byte rather than char calculation for efficiency. It fails
906 * to quickly reject some cases that can't match, but will reject
907 * them later after doing full char arithmetic */
908 if (prog->minlen > strend - strpos) {
909 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
910 " String too short...\n"));
914 RXp_MATCH_UTF8_set(prog, utf8_target);
915 reginfo->is_utf8_target = cBOOL(utf8_target);
916 reginfo->info_aux = NULL;
917 reginfo->strbeg = strbeg;
918 reginfo->strend = strend;
919 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
921 /* not actually used within intuit, but zero for safety anyway */
922 reginfo->poscache_maxiter = 0;
925 if ((!prog->anchored_utf8 && prog->anchored_substr)
926 || (!prog->float_utf8 && prog->float_substr))
927 to_utf8_substr(prog);
928 check = prog->check_utf8;
930 if (!prog->check_substr && prog->check_utf8) {
931 if (! to_byte_substr(prog)) {
932 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(fail);
935 check = prog->check_substr;
938 /* dump the various substring data */
939 DEBUG_OPTIMISE_MORE_r({
941 for (i=0; i<=2; i++) {
942 SV *sv = (utf8_target ? prog->substrs->data[i].utf8_substr
943 : prog->substrs->data[i].substr);
947 Perl_re_printf( aTHX_
948 " substrs[%d]: min=%" IVdf " max=%" IVdf " end shift=%" IVdf
949 " useful=%" IVdf " utf8=%d [%s]\n",
951 (IV)prog->substrs->data[i].min_offset,
952 (IV)prog->substrs->data[i].max_offset,
953 (IV)prog->substrs->data[i].end_shift,
960 if (prog->intflags & PREGf_ANCH) { /* Match at \G, beg-of-str or after \n */
962 /* ml_anch: check after \n?
964 * A note about PREGf_IMPLICIT: on an un-anchored pattern beginning
965 * with /.*.../, these flags will have been added by the
967 * /.*abc/, /.*abc/m: PREGf_IMPLICIT | PREGf_ANCH_MBOL
968 * /.*abc/s: PREGf_IMPLICIT | PREGf_ANCH_SBOL
970 ml_anch = (prog->intflags & PREGf_ANCH_MBOL)
971 && !(prog->intflags & PREGf_IMPLICIT);
973 if (!ml_anch && !(prog->intflags & PREGf_IMPLICIT)) {
974 /* we are only allowed to match at BOS or \G */
976 /* trivially reject if there's a BOS anchor and we're not at BOS.
978 * Note that we don't try to do a similar quick reject for
979 * \G, since generally the caller will have calculated strpos
980 * based on pos() and gofs, so the string is already correctly
981 * anchored by definition; and handling the exceptions would
982 * be too fiddly (e.g. REXEC_IGNOREPOS).
984 if ( strpos != strbeg
985 && (prog->intflags & PREGf_ANCH_SBOL))
987 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
988 " Not at start...\n"));
992 /* in the presence of an anchor, the anchored (relative to the
993 * start of the regex) substr must also be anchored relative
994 * to strpos. So quickly reject if substr isn't found there.
995 * This works for \G too, because the caller will already have
996 * subtracted gofs from pos, and gofs is the offset from the
997 * \G to the start of the regex. For example, in /.abc\Gdef/,
998 * where substr="abcdef", pos()=3, gofs=4, offset_min=1:
999 * caller will have set strpos=pos()-4; we look for the substr
1000 * at position pos()-4+1, which lines up with the "a" */
1002 if (prog->check_offset_min == prog->check_offset_max) {
1003 /* Substring at constant offset from beg-of-str... */
1004 SSize_t slen = SvCUR(check);
1005 char *s = HOP3c(strpos, prog->check_offset_min, strend);
1007 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1008 " Looking for check substr at fixed offset %" IVdf "...\n",
1009 (IV)prog->check_offset_min));
1011 if (SvTAIL(check)) {
1012 /* In this case, the regex is anchored at the end too.
1013 * Unless it's a multiline match, the lengths must match
1014 * exactly, give or take a \n. NB: slen >= 1 since
1015 * the last char of check is \n */
1017 && ( strend - s > slen
1018 || strend - s < slen - 1
1019 || (strend - s == slen && strend[-1] != '\n')))
1021 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1022 " String too long...\n"));
1025 /* Now should match s[0..slen-2] */
1028 if (slen && (strend - s < slen
1029 || *SvPVX_const(check) != *s
1030 || (slen > 1 && (memNE(SvPVX_const(check), s, slen)))))
1032 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1033 " String not equal...\n"));
1038 goto success_at_start;
1043 end_shift = prog->check_end_shift;
1045 #ifdef DEBUGGING /* 7/99: reports of failure (with the older version) */
1047 Perl_croak(aTHX_ "panic: end_shift: %" IVdf " pattern:\n%s\n ",
1048 (IV)end_shift, RX_PRECOMP(rx));
1053 /* This is the (re)entry point of the main loop in this function.
1054 * The goal of this loop is to:
1055 * 1) find the "check" substring in the region rx_origin..strend
1056 * (adjusted by start_shift / end_shift). If not found, reject
1058 * 2) If it exists, look for the "other" substr too if defined; for
1059 * example, if the check substr maps to the anchored substr, then
1060 * check the floating substr, and vice-versa. If not found, go
1061 * back to (1) with rx_origin suitably incremented.
1062 * 3) If we find an rx_origin position that doesn't contradict
1063 * either of the substrings, then check the possible additional
1064 * constraints on rx_origin of /^.../m or a known start class.
1065 * If these fail, then depending on which constraints fail, jump
1066 * back to here, or to various other re-entry points further along
1067 * that skip some of the first steps.
1068 * 4) If we pass all those tests, update the BmUSEFUL() count on the
1069 * substring. If the start position was determined to be at the
1070 * beginning of the string - so, not rejected, but not optimised,
1071 * since we have to run regmatch from position 0 - decrement the
1072 * BmUSEFUL() count. Otherwise increment it.
1076 /* first, look for the 'check' substring */
1082 DEBUG_OPTIMISE_MORE_r({
1083 Perl_re_printf( aTHX_
1084 " At restart: rx_origin=%" IVdf " Check offset min: %" IVdf
1085 " Start shift: %" IVdf " End shift %" IVdf
1086 " Real end Shift: %" IVdf "\n",
1087 (IV)(rx_origin - strbeg),
1088 (IV)prog->check_offset_min,
1091 (IV)prog->check_end_shift);
1094 end_point = HOPBACK3(strend, end_shift, rx_origin);
1097 start_point = HOPMAYBE3(rx_origin, start_shift, end_point);
1102 /* If the regex is absolutely anchored to either the start of the
1103 * string (SBOL) or to pos() (ANCH_GPOS), then
1104 * check_offset_max represents an upper bound on the string where
1105 * the substr could start. For the ANCH_GPOS case, we assume that
1106 * the caller of intuit will have already set strpos to
1107 * pos()-gofs, so in this case strpos + offset_max will still be
1108 * an upper bound on the substr.
1111 && prog->intflags & PREGf_ANCH
1112 && prog->check_offset_max != SSize_t_MAX)
1114 SSize_t check_len = SvCUR(check) - !!SvTAIL(check);
1115 const char * const anchor =
1116 (prog->intflags & PREGf_ANCH_GPOS ? strpos : strbeg);
1117 SSize_t targ_len = (char*)end_point - anchor;
1119 if (check_len > targ_len) {
1120 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1121 "Target string too short to match required substring...\n"));
1125 /* do a bytes rather than chars comparison. It's conservative;
1126 * so it skips doing the HOP if the result can't possibly end
1127 * up earlier than the old value of end_point.
1129 assert(anchor + check_len <= (char *)end_point);
1130 if (prog->check_offset_max + check_len < targ_len) {
1131 end_point = HOP3lim((U8*)anchor,
1132 prog->check_offset_max,
1133 end_point - check_len
1136 if (end_point < start_point)
1141 check_at = fbm_instr( start_point, end_point,
1142 check, multiline ? FBMrf_MULTILINE : 0);
1144 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1145 " doing 'check' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
1146 (IV)((char*)start_point - strbeg),
1147 (IV)((char*)end_point - strbeg),
1148 (IV)(check_at ? check_at - strbeg : -1)
1151 /* Update the count-of-usability, remove useless subpatterns,
1155 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1156 SvPVX_const(check), RE_SV_DUMPLEN(check), 30);
1157 Perl_re_printf( aTHX_ " %s %s substr %s%s%s",
1158 (check_at ? "Found" : "Did not find"),
1159 (check == (utf8_target ? prog->anchored_utf8 : prog->anchored_substr)
1160 ? "anchored" : "floating"),
1163 (check_at ? " at offset " : "...\n") );
1168 /* set rx_origin to the minimum position where the regex could start
1169 * matching, given the constraint of the just-matched check substring.
1170 * But don't set it lower than previously.
1173 if (check_at - rx_origin > prog->check_offset_max)
1174 rx_origin = HOP3c(check_at, -prog->check_offset_max, rx_origin);
1175 /* Finish the diagnostic message */
1176 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1177 "%ld (rx_origin now %" IVdf ")...\n",
1178 (long)(check_at - strbeg),
1179 (IV)(rx_origin - strbeg)
1184 /* now look for the 'other' substring if defined */
1186 if (utf8_target ? prog->substrs->data[other_ix].utf8_substr
1187 : prog->substrs->data[other_ix].substr)
1189 /* Take into account the "other" substring. */
1193 struct reg_substr_datum *other;
1196 other = &prog->substrs->data[other_ix];
1198 /* if "other" is anchored:
1199 * we've previously found a floating substr starting at check_at.
1200 * This means that the regex origin must lie somewhere
1201 * between min (rx_origin): HOP3(check_at, -check_offset_max)
1202 * and max: HOP3(check_at, -check_offset_min)
1203 * (except that min will be >= strpos)
1204 * So the fixed substr must lie somewhere between
1205 * HOP3(min, anchored_offset)
1206 * HOP3(max, anchored_offset) + SvCUR(substr)
1209 /* if "other" is floating
1210 * Calculate last1, the absolute latest point where the
1211 * floating substr could start in the string, ignoring any
1212 * constraints from the earlier fixed match. It is calculated
1215 * strend - prog->minlen (in chars) is the absolute latest
1216 * position within the string where the origin of the regex
1217 * could appear. The latest start point for the floating
1218 * substr is float_min_offset(*) on from the start of the
1219 * regex. last1 simply combines thee two offsets.
1221 * (*) You might think the latest start point should be
1222 * float_max_offset from the regex origin, and technically
1223 * you'd be correct. However, consider
1225 * Here, float min, max are 3,5 and minlen is 7.
1226 * This can match either
1230 * In the first case, the regex matches minlen chars; in the
1231 * second, minlen+1, in the third, minlen+2.
1232 * In the first case, the floating offset is 3 (which equals
1233 * float_min), in the second, 4, and in the third, 5 (which
1234 * equals float_max). In all cases, the floating string bcd
1235 * can never start more than 4 chars from the end of the
1236 * string, which equals minlen - float_min. As the substring
1237 * starts to match more than float_min from the start of the
1238 * regex, it makes the regex match more than minlen chars,
1239 * and the two cancel each other out. So we can always use
1240 * float_min - minlen, rather than float_max - minlen for the
1241 * latest position in the string.
1243 * Note that -minlen + float_min_offset is equivalent (AFAIKT)
1244 * to CHR_SVLEN(must) - !!SvTAIL(must) + prog->float_end_shift
1247 assert(prog->minlen >= other->min_offset);
1248 last1 = HOP3c(strend,
1249 other->min_offset - prog->minlen, strbeg);
1251 if (other_ix) {/* i.e. if (other-is-float) */
1252 /* last is the latest point where the floating substr could
1253 * start, *given* any constraints from the earlier fixed
1254 * match. This constraint is that the floating string starts
1255 * <= float_max_offset chars from the regex origin (rx_origin).
1256 * If this value is less than last1, use it instead.
1258 assert(rx_origin <= last1);
1260 /* this condition handles the offset==infinity case, and
1261 * is a short-cut otherwise. Although it's comparing a
1262 * byte offset to a char length, it does so in a safe way,
1263 * since 1 char always occupies 1 or more bytes,
1264 * so if a string range is (last1 - rx_origin) bytes,
1265 * it will be less than or equal to (last1 - rx_origin)
1266 * chars; meaning it errs towards doing the accurate HOP3
1267 * rather than just using last1 as a short-cut */
1268 (last1 - rx_origin) < other->max_offset
1270 : (char*)HOP3lim(rx_origin, other->max_offset, last1);
1273 assert(strpos + start_shift <= check_at);
1274 last = HOP4c(check_at, other->min_offset - start_shift,
1278 s = HOP3c(rx_origin, other->min_offset, strend);
1279 if (s < other_last) /* These positions already checked */
1282 must = utf8_target ? other->utf8_substr : other->substr;
1283 assert(SvPOK(must));
1286 char *to = last + SvCUR(must) - (SvTAIL(must)!=0);
1292 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1293 " skipping 'other' fbm scan: %" IVdf " > %" IVdf "\n",
1294 (IV)(from - strbeg),
1300 (unsigned char*)from,
1303 multiline ? FBMrf_MULTILINE : 0
1305 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1306 " doing 'other' fbm scan, [%" IVdf "..%" IVdf "] gave %" IVdf "\n",
1307 (IV)(from - strbeg),
1309 (IV)(s ? s - strbeg : -1)
1315 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
1316 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
1317 Perl_re_printf( aTHX_ " %s %s substr %s%s",
1318 s ? "Found" : "Contradicts",
1319 other_ix ? "floating" : "anchored",
1320 quoted, RE_SV_TAIL(must));
1325 /* last1 is latest possible substr location. If we didn't
1326 * find it before there, we never will */
1327 if (last >= last1) {
1328 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1329 "; giving up...\n"));
1333 /* try to find the check substr again at a later
1334 * position. Maybe next time we'll find the "other" substr
1336 other_last = HOP3c(last, 1, strend) /* highest failure */;
1338 other_ix /* i.e. if other-is-float */
1339 ? HOP3c(rx_origin, 1, strend)
1340 : HOP4c(last, 1 - other->min_offset, strbeg, strend);
1341 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1342 "; about to retry %s at offset %ld (rx_origin now %" IVdf ")...\n",
1343 (other_ix ? "floating" : "anchored"),
1344 (long)(HOP3c(check_at, 1, strend) - strbeg),
1345 (IV)(rx_origin - strbeg)
1350 if (other_ix) { /* if (other-is-float) */
1351 /* other_last is set to s, not s+1, since its possible for
1352 * a floating substr to fail first time, then succeed
1353 * second time at the same floating position; e.g.:
1354 * "-AB--AABZ" =~ /\wAB\d*Z/
1355 * The first time round, anchored and float match at
1356 * "-(AB)--AAB(Z)" then fail on the initial \w character
1357 * class. Second time round, they match at "-AB--A(AB)(Z)".
1362 rx_origin = HOP3c(s, -other->min_offset, strbeg);
1363 other_last = HOP3c(s, 1, strend);
1365 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1366 " at offset %ld (rx_origin now %" IVdf ")...\n",
1368 (IV)(rx_origin - strbeg)
1374 DEBUG_OPTIMISE_MORE_r(
1375 Perl_re_printf( aTHX_
1376 " Check-only match: offset min:%" IVdf " max:%" IVdf
1377 " check_at:%" IVdf " rx_origin:%" IVdf " rx_origin-check_at:%" IVdf
1378 " strend:%" IVdf "\n",
1379 (IV)prog->check_offset_min,
1380 (IV)prog->check_offset_max,
1381 (IV)(check_at-strbeg),
1382 (IV)(rx_origin-strbeg),
1383 (IV)(rx_origin-check_at),
1389 postprocess_substr_matches:
1391 /* handle the extra constraint of /^.../m if present */
1393 if (ml_anch && rx_origin != strbeg && rx_origin[-1] != '\n') {
1396 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1397 " looking for /^/m anchor"));
1399 /* we have failed the constraint of a \n before rx_origin.
1400 * Find the next \n, if any, even if it's beyond the current
1401 * anchored and/or floating substrings. Whether we should be
1402 * scanning ahead for the next \n or the next substr is debatable.
1403 * On the one hand you'd expect rare substrings to appear less
1404 * often than \n's. On the other hand, searching for \n means
1405 * we're effectively flipping between check_substr and "\n" on each
1406 * iteration as the current "rarest" string candidate, which
1407 * means for example that we'll quickly reject the whole string if
1408 * hasn't got a \n, rather than trying every substr position
1412 s = HOP3c(strend, - prog->minlen, strpos);
1413 if (s <= rx_origin ||
1414 ! ( rx_origin = (char *)memchr(rx_origin, '\n', s - rx_origin)))
1416 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1417 " Did not find /%s^%s/m...\n",
1418 PL_colors[0], PL_colors[1]));
1422 /* earliest possible origin is 1 char after the \n.
1423 * (since *rx_origin == '\n', it's safe to ++ here rather than
1424 * HOP(rx_origin, 1)) */
1427 if (prog->substrs->check_ix == 0 /* check is anchored */
1428 || rx_origin >= HOP3c(check_at, - prog->check_offset_min, strpos))
1430 /* Position contradicts check-string; either because
1431 * check was anchored (and thus has no wiggle room),
1432 * or check was float and rx_origin is above the float range */
1433 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1434 " Found /%s^%s/m, about to restart lookup for check-string with rx_origin %ld...\n",
1435 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1439 /* if we get here, the check substr must have been float,
1440 * is in range, and we may or may not have had an anchored
1441 * "other" substr which still contradicts */
1442 assert(prog->substrs->check_ix); /* check is float */
1444 if (utf8_target ? prog->anchored_utf8 : prog->anchored_substr) {
1445 /* whoops, the anchored "other" substr exists, so we still
1446 * contradict. On the other hand, the float "check" substr
1447 * didn't contradict, so just retry the anchored "other"
1449 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1450 " Found /%s^%s/m, rescanning for anchored from offset %" IVdf " (rx_origin now %" IVdf ")...\n",
1451 PL_colors[0], PL_colors[1],
1452 (IV)(rx_origin - strbeg + prog->anchored_offset),
1453 (IV)(rx_origin - strbeg)
1455 goto do_other_substr;
1458 /* success: we don't contradict the found floating substring
1459 * (and there's no anchored substr). */
1460 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1461 " Found /%s^%s/m with rx_origin %ld...\n",
1462 PL_colors[0], PL_colors[1], (long)(rx_origin - strbeg)));
1465 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1466 " (multiline anchor test skipped)\n"));
1472 /* if we have a starting character class, then test that extra constraint.
1473 * (trie stclasses are too expensive to use here, we are better off to
1474 * leave it to regmatch itself) */
1476 if (progi->regstclass && PL_regkind[OP(progi->regstclass)]!=TRIE) {
1477 const U8* const str = (U8*)STRING(progi->regstclass);
1479 /* XXX this value could be pre-computed */
1480 const int cl_l = (PL_regkind[OP(progi->regstclass)] == EXACT
1481 ? (reginfo->is_utf8_pat
1482 ? utf8_distance(str + STR_LEN(progi->regstclass), str)
1483 : STR_LEN(progi->regstclass))
1487 /* latest pos that a matching float substr constrains rx start to */
1488 char *rx_max_float = NULL;
1490 /* if the current rx_origin is anchored, either by satisfying an
1491 * anchored substring constraint, or a /^.../m constraint, then we
1492 * can reject the current origin if the start class isn't found
1493 * at the current position. If we have a float-only match, then
1494 * rx_origin is constrained to a range; so look for the start class
1495 * in that range. if neither, then look for the start class in the
1496 * whole rest of the string */
1498 /* XXX DAPM it's not clear what the minlen test is for, and why
1499 * it's not used in the floating case. Nothing in the test suite
1500 * causes minlen == 0 here. See <20140313134639.GS12844@iabyn.com>.
1501 * Here are some old comments, which may or may not be correct:
1503 * minlen == 0 is possible if regstclass is \b or \B,
1504 * and the fixed substr is ''$.
1505 * Since minlen is already taken into account, rx_origin+1 is
1506 * before strend; accidentally, minlen >= 1 guaranties no false
1507 * positives at rx_origin + 1 even for \b or \B. But (minlen? 1 :
1508 * 0) below assumes that regstclass does not come from lookahead...
1509 * If regstclass takes bytelength more than 1: If charlength==1, OK.
1510 * This leaves EXACTF-ish only, which are dealt with in
1514 if (prog->anchored_substr || prog->anchored_utf8 || ml_anch)
1515 endpos = HOP3clim(rx_origin, (prog->minlen ? cl_l : 0), strend);
1516 else if (prog->float_substr || prog->float_utf8) {
1517 rx_max_float = HOP3c(check_at, -start_shift, strbeg);
1518 endpos = HOP3clim(rx_max_float, cl_l, strend);
1523 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1524 " looking for class: start_shift: %" IVdf " check_at: %" IVdf
1525 " rx_origin: %" IVdf " endpos: %" IVdf "\n",
1526 (IV)start_shift, (IV)(check_at - strbeg),
1527 (IV)(rx_origin - strbeg), (IV)(endpos - strbeg)));
1529 s = find_byclass(prog, progi->regstclass, rx_origin, endpos,
1532 if (endpos == strend) {
1533 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1534 " Could not match STCLASS...\n") );
1537 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1538 " This position contradicts STCLASS...\n") );
1539 if ((prog->intflags & PREGf_ANCH) && !ml_anch
1540 && !(prog->intflags & PREGf_IMPLICIT))
1543 /* Contradict one of substrings */
1544 if (prog->anchored_substr || prog->anchored_utf8) {
1545 if (prog->substrs->check_ix == 1) { /* check is float */
1546 /* Have both, check_string is floating */
1547 assert(rx_origin + start_shift <= check_at);
1548 if (rx_origin + start_shift != check_at) {
1549 /* not at latest position float substr could match:
1550 * Recheck anchored substring, but not floating.
1551 * The condition above is in bytes rather than
1552 * chars for efficiency. It's conservative, in
1553 * that it errs on the side of doing 'goto
1554 * do_other_substr'. In this case, at worst,
1555 * an extra anchored search may get done, but in
1556 * practice the extra fbm_instr() is likely to
1557 * get skipped anyway. */
1558 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1559 " about to retry anchored at offset %ld (rx_origin now %" IVdf ")...\n",
1560 (long)(other_last - strbeg),
1561 (IV)(rx_origin - strbeg)
1563 goto do_other_substr;
1571 /* In the presence of ml_anch, we might be able to
1572 * find another \n without breaking the current float
1575 /* strictly speaking this should be HOP3c(..., 1, ...),
1576 * but since we goto a block of code that's going to
1577 * search for the next \n if any, its safe here */
1579 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1580 " about to look for /%s^%s/m starting at rx_origin %ld...\n",
1581 PL_colors[0], PL_colors[1],
1582 (long)(rx_origin - strbeg)) );
1583 goto postprocess_substr_matches;
1586 /* strictly speaking this can never be true; but might
1587 * be if we ever allow intuit without substrings */
1588 if (!(utf8_target ? prog->float_utf8 : prog->float_substr))
1591 rx_origin = rx_max_float;
1594 /* at this point, any matching substrings have been
1595 * contradicted. Start again... */
1597 rx_origin = HOP3c(rx_origin, 1, strend);
1599 /* uses bytes rather than char calculations for efficiency.
1600 * It's conservative: it errs on the side of doing 'goto restart',
1601 * where there is code that does a proper char-based test */
1602 if (rx_origin + start_shift + end_shift > strend) {
1603 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1604 " Could not match STCLASS...\n") );
1607 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
1608 " about to look for %s substr starting at offset %ld (rx_origin now %" IVdf ")...\n",
1609 (prog->substrs->check_ix ? "floating" : "anchored"),
1610 (long)(rx_origin + start_shift - strbeg),
1611 (IV)(rx_origin - strbeg)
1618 if (rx_origin != s) {
1619 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1620 " By STCLASS: moving %ld --> %ld\n",
1621 (long)(rx_origin - strbeg), (long)(s - strbeg))
1625 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1626 " Does not contradict STCLASS...\n");
1631 /* Decide whether using the substrings helped */
1633 if (rx_origin != strpos) {
1634 /* Fixed substring is found far enough so that the match
1635 cannot start at strpos. */
1637 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " try at offset...\n"));
1638 ++BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr); /* hooray/5 */
1641 /* The found rx_origin position does not prohibit matching at
1642 * strpos, so calling intuit didn't gain us anything. Decrement
1643 * the BmUSEFUL() count on the check substring, and if we reach
1645 if (!(prog->intflags & PREGf_NAUGHTY)
1647 prog->check_utf8 /* Could be deleted already */
1648 && --BmUSEFUL(prog->check_utf8) < 0
1649 && (prog->check_utf8 == prog->float_utf8)
1651 prog->check_substr /* Could be deleted already */
1652 && --BmUSEFUL(prog->check_substr) < 0
1653 && (prog->check_substr == prog->float_substr)
1656 /* If flags & SOMETHING - do not do it many times on the same match */
1657 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ " ... Disabling check substring...\n"));
1658 /* XXX Does the destruction order has to change with utf8_target? */
1659 SvREFCNT_dec(utf8_target ? prog->check_utf8 : prog->check_substr);
1660 SvREFCNT_dec(utf8_target ? prog->check_substr : prog->check_utf8);
1661 prog->check_substr = prog->check_utf8 = NULL; /* disable */
1662 prog->float_substr = prog->float_utf8 = NULL; /* clear */
1663 check = NULL; /* abort */
1664 /* XXXX This is a remnant of the old implementation. It
1665 looks wasteful, since now INTUIT can use many
1666 other heuristics. */
1667 prog->extflags &= ~RXf_USE_INTUIT;
1671 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
1672 "Intuit: %sSuccessfully guessed:%s match at offset %ld\n",
1673 PL_colors[4], PL_colors[5], (long)(rx_origin - strbeg)) );
1677 fail_finish: /* Substring not found */
1678 if (prog->check_substr || prog->check_utf8) /* could be removed already */
1679 BmUSEFUL(utf8_target ? prog->check_utf8 : prog->check_substr) += 5; /* hooray */
1681 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch rejected by optimizer%s\n",
1682 PL_colors[4], PL_colors[5]));
1687 #define DECL_TRIE_TYPE(scan) \
1688 const enum { trie_plain, trie_utf8, trie_utf8_fold, trie_latin_utf8_fold, \
1689 trie_utf8_exactfa_fold, trie_latin_utf8_exactfa_fold, \
1690 trie_utf8l, trie_flu8, trie_flu8_latin } \
1691 trie_type = ((scan->flags == EXACT) \
1692 ? (utf8_target ? trie_utf8 : trie_plain) \
1693 : (scan->flags == EXACTL) \
1694 ? (utf8_target ? trie_utf8l : trie_plain) \
1695 : (scan->flags == EXACTFAA) \
1697 ? trie_utf8_exactfa_fold \
1698 : trie_latin_utf8_exactfa_fold) \
1699 : (scan->flags == EXACTFLU8 \
1702 : trie_flu8_latin) \
1705 : trie_latin_utf8_fold)))
1707 /* 'uscan' is set to foldbuf, and incremented, so below the end of uscan is
1708 * 'foldbuf+sizeof(foldbuf)' */
1709 #define REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc, uc_end, uscan, len, uvc, charid, foldlen, foldbuf, uniflags) \
1712 U8 flags = FOLD_FLAGS_FULL; \
1713 switch (trie_type) { \
1715 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1716 if (UTF8_IS_ABOVE_LATIN1(*uc)) { \
1717 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end); \
1719 goto do_trie_utf8_fold; \
1720 case trie_utf8_exactfa_fold: \
1721 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1723 case trie_utf8_fold: \
1724 do_trie_utf8_fold: \
1725 if ( foldlen>0 ) { \
1726 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \
1731 uvc = _toFOLD_utf8_flags( (const U8*) uc, uc_end, foldbuf, &foldlen, \
1733 len = UTF8SKIP(uc); \
1734 skiplen = UVCHR_SKIP( uvc ); \
1735 foldlen -= skiplen; \
1736 uscan = foldbuf + skiplen; \
1739 case trie_flu8_latin: \
1740 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1741 goto do_trie_latin_utf8_fold; \
1742 case trie_latin_utf8_exactfa_fold: \
1743 flags |= FOLD_FLAGS_NOMIX_ASCII; \
1745 case trie_latin_utf8_fold: \
1746 do_trie_latin_utf8_fold: \
1747 if ( foldlen>0 ) { \
1748 uvc = utf8n_to_uvchr( (const U8*) uscan, foldlen, &len, uniflags ); \
1754 uvc = _to_fold_latin1( (U8) *uc, foldbuf, &foldlen, flags); \
1755 skiplen = UVCHR_SKIP( uvc ); \
1756 foldlen -= skiplen; \
1757 uscan = foldbuf + skiplen; \
1761 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
1762 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*uc)) { \
1763 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(uc, uc_end); \
1767 uvc = utf8n_to_uvchr( (const U8*) uc, uc_end - uc, &len, uniflags ); \
1774 charid = trie->charmap[ uvc ]; \
1778 if (widecharmap) { \
1779 SV** const svpp = hv_fetch(widecharmap, \
1780 (char*)&uvc, sizeof(UV), 0); \
1782 charid = (U16)SvIV(*svpp); \
1787 #define DUMP_EXEC_POS(li,s,doutf8,depth) \
1788 dump_exec_pos(li,s,(reginfo->strend),(reginfo->strbeg), \
1789 startpos, doutf8, depth)
1791 #define REXEC_FBC_SCAN(UTF8, CODE) \
1793 while (s < strend) { \
1795 s += ((UTF8) ? UTF8SKIP(s) : 1); \
1799 #define REXEC_FBC_CLASS_SCAN(UTF8, COND) \
1801 while (s < strend) { \
1802 REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND) \
1806 #define REXEC_FBC_CLASS_SCAN_GUTS(UTF8, COND) \
1809 s += ((UTF8) ? UTF8SKIP(s) : 1); \
1810 previous_occurrence_end = s; \
1813 s += ((UTF8) ? UTF8SKIP(s) : 1); \
1816 #define REXEC_FBC_CSCAN(CONDUTF8,COND) \
1817 if (utf8_target) { \
1818 REXEC_FBC_CLASS_SCAN(1, CONDUTF8); \
1821 REXEC_FBC_CLASS_SCAN(0, COND); \
1824 /* We keep track of where the next character should start after an occurrence
1825 * of the one we're looking for. Knowing that, we can see right away if the
1826 * next occurrence is adjacent to the previous. When 'doevery' is FALSE, we
1827 * don't accept the 2nd and succeeding adjacent occurrences */
1828 #define FBC_CHECK_AND_TRY \
1830 || s != previous_occurrence_end) \
1831 && (reginfo->intuit || regtry(reginfo, &s))) \
1837 /* This differs from the above macros in that it calls a function which returns
1838 * the next occurrence of the thing being looked for in 's'; and 'strend' if
1839 * there is no such occurrence. */
1840 #define REXEC_FBC_FIND_NEXT_SCAN(UTF8, f) \
1841 while (s < strend) { \
1843 if (s >= strend) { \
1848 s += (UTF8) ? UTF8SKIP(s) : 1; \
1849 previous_occurrence_end = s; \
1852 /* The three macros below are slightly different versions of the same logic.
1854 * The first is for /a and /aa when the target string is UTF-8. This can only
1855 * match ascii, but it must advance based on UTF-8. The other two handle the
1856 * non-UTF-8 and the more generic UTF-8 cases. In all three, we are looking
1857 * for the boundary (or non-boundary) between a word and non-word character.
1858 * The utf8 and non-utf8 cases have the same logic, but the details must be
1859 * different. Find the "wordness" of the character just prior to this one, and
1860 * compare it with the wordness of this one. If they differ, we have a
1861 * boundary. At the beginning of the string, pretend that the previous
1862 * character was a new-line.
1864 * All these macros uncleanly have side-effects with each other and outside
1865 * variables. So far it's been too much trouble to clean-up
1867 * TEST_NON_UTF8 is the macro or function to call to test if its byte input is
1868 * a word character or not.
1869 * IF_SUCCESS is code to do if it finds that we are at a boundary between
1871 * IF_FAIL is code to do if we aren't at a boundary between word/non-word
1873 * Exactly one of the two IF_FOO parameters is a no-op, depending on whether we
1874 * are looking for a boundary or for a non-boundary. If we are looking for a
1875 * boundary, we want IF_FAIL to be the no-op, and for IF_SUCCESS to go out and
1876 * see if this tentative match actually works, and if so, to quit the loop
1877 * here. And vice-versa if we are looking for a non-boundary.
1879 * 'tmp' below in the next three macros in the REXEC_FBC_SCAN and
1880 * REXEC_FBC_SCAN loops is a loop invariant, a bool giving the return of
1881 * TEST_NON_UTF8(s-1). To see this, note that that's what it is defined to be
1882 * at entry to the loop, and to get to the IF_FAIL branch, tmp must equal
1883 * TEST_NON_UTF8(s), and in the opposite branch, IF_SUCCESS, tmp is that
1884 * complement. But in that branch we complement tmp, meaning that at the
1885 * bottom of the loop tmp is always going to be equal to TEST_NON_UTF8(s),
1886 * which means at the top of the loop in the next iteration, it is
1887 * TEST_NON_UTF8(s-1) */
1888 #define FBC_UTF8_A(TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1889 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1890 tmp = TEST_NON_UTF8(tmp); \
1891 REXEC_FBC_SCAN(1, /* 1=>is-utf8; advances s while s < strend */ \
1892 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1894 IF_SUCCESS; /* Is a boundary if values for s-1 and s differ */ \
1901 /* Like FBC_UTF8_A, but TEST_UV is a macro which takes a UV as its input, and
1902 * TEST_UTF8 is a macro that for the same input code points returns identically
1903 * to TEST_UV, but takes a pointer to a UTF-8 encoded string instead */
1904 #define FBC_UTF8(TEST_UV, TEST_UTF8, IF_SUCCESS, IF_FAIL) \
1905 if (s == reginfo->strbeg) { \
1908 else { /* Back-up to the start of the previous character */ \
1909 U8 * const r = reghop3((U8*)s, -1, (U8*)reginfo->strbeg); \
1910 tmp = utf8n_to_uvchr(r, (U8*) reginfo->strend - r, \
1911 0, UTF8_ALLOW_DEFAULT); \
1913 tmp = TEST_UV(tmp); \
1914 REXEC_FBC_SCAN(1, /* 1=>is-utf8; advances s while s < strend */ \
1915 if (tmp == ! (TEST_UTF8((U8 *) s, (U8 *) reginfo->strend))) { \
1924 /* Like the above two macros. UTF8_CODE is the complete code for handling
1925 * UTF-8. Common to the BOUND and NBOUND cases, set-up by the FBC_BOUND, etc
1927 #define FBC_BOUND_COMMON(UTF8_CODE, TEST_NON_UTF8, IF_SUCCESS, IF_FAIL) \
1928 if (utf8_target) { \
1931 else { /* Not utf8 */ \
1932 tmp = (s != reginfo->strbeg) ? UCHARAT(s - 1) : '\n'; \
1933 tmp = TEST_NON_UTF8(tmp); \
1934 REXEC_FBC_SCAN(0, /* 0=>not-utf8; advances s while s < strend */ \
1935 if (tmp == ! TEST_NON_UTF8((U8) *s)) { \
1944 /* Here, things have been set up by the previous code so that tmp is the \
1945 * return of TEST_NON_UTF(s-1) or TEST_UTF8(s-1) (depending on the \
1946 * utf8ness of the target). We also have to check if this matches against \
1947 * the EOS, which we treat as a \n (which is the same value in both UTF-8 \
1948 * or non-UTF8, so can use the non-utf8 test condition even for a UTF-8 \
1950 if (tmp == ! TEST_NON_UTF8('\n')) { \
1957 /* This is the macro to use when we want to see if something that looks like it
1958 * could match, actually does, and if so exits the loop */
1959 #define REXEC_FBC_TRYIT \
1960 if ((reginfo->intuit || regtry(reginfo, &s))) \
1963 /* The only difference between the BOUND and NBOUND cases is that
1964 * REXEC_FBC_TRYIT is called when matched in BOUND, and when non-matched in
1965 * NBOUND. This is accomplished by passing it as either the if or else clause,
1966 * with the other one being empty (PLACEHOLDER is defined as empty).
1968 * The TEST_FOO parameters are for operating on different forms of input, but
1969 * all should be ones that return identically for the same underlying code
1971 #define FBC_BOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1973 FBC_UTF8(TEST_UV, TEST_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1974 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1976 #define FBC_BOUND_A(TEST_NON_UTF8) \
1978 FBC_UTF8_A(TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER), \
1979 TEST_NON_UTF8, REXEC_FBC_TRYIT, PLACEHOLDER)
1981 #define FBC_NBOUND(TEST_NON_UTF8, TEST_UV, TEST_UTF8) \
1983 FBC_UTF8(TEST_UV, TEST_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1984 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1986 #define FBC_NBOUND_A(TEST_NON_UTF8) \
1988 FBC_UTF8_A(TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT), \
1989 TEST_NON_UTF8, PLACEHOLDER, REXEC_FBC_TRYIT)
1993 S_get_break_val_cp_checked(SV* const invlist, const UV cp_in) {
1994 IV cp_out = _invlist_search(invlist, cp_in);
1995 assert(cp_out >= 0);
1998 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
1999 invmap[S_get_break_val_cp_checked(invlist, cp)]
2001 # define _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp) \
2002 invmap[_invlist_search(invlist, cp)]
2005 /* Takes a pointer to an inversion list, a pointer to its corresponding
2006 * inversion map, and a code point, and returns the code point's value
2007 * according to the two arrays. It assumes that all code points have a value.
2008 * This is used as the base macro for macros for particular properties */
2009 #define _generic_GET_BREAK_VAL_CP(invlist, invmap, cp) \
2010 _generic_GET_BREAK_VAL_CP_CHECKED(invlist, invmap, cp)
2012 /* Same as above, but takes begin, end ptrs to a UTF-8 encoded string instead
2013 * of a code point, returning the value for the first code point in the string.
2014 * And it takes the particular macro name that finds the desired value given a
2015 * code point. Merely convert the UTF-8 to code point and call the cp macro */
2016 #define _generic_GET_BREAK_VAL_UTF8(cp_macro, pos, strend) \
2017 (__ASSERT_(pos < strend) \
2018 /* Note assumes is valid UTF-8 */ \
2019 (cp_macro(utf8_to_uvchr_buf((pos), (strend), NULL))))
2021 /* Returns the GCB value for the input code point */
2022 #define getGCB_VAL_CP(cp) \
2023 _generic_GET_BREAK_VAL_CP( \
2028 /* Returns the GCB value for the first code point in the UTF-8 encoded string
2029 * bounded by pos and strend */
2030 #define getGCB_VAL_UTF8(pos, strend) \
2031 _generic_GET_BREAK_VAL_UTF8(getGCB_VAL_CP, pos, strend)
2033 /* Returns the LB value for the input code point */
2034 #define getLB_VAL_CP(cp) \
2035 _generic_GET_BREAK_VAL_CP( \
2040 /* Returns the LB value for the first code point in the UTF-8 encoded string
2041 * bounded by pos and strend */
2042 #define getLB_VAL_UTF8(pos, strend) \
2043 _generic_GET_BREAK_VAL_UTF8(getLB_VAL_CP, pos, strend)
2046 /* Returns the SB value for the input code point */
2047 #define getSB_VAL_CP(cp) \
2048 _generic_GET_BREAK_VAL_CP( \
2053 /* Returns the SB value for the first code point in the UTF-8 encoded string
2054 * bounded by pos and strend */
2055 #define getSB_VAL_UTF8(pos, strend) \
2056 _generic_GET_BREAK_VAL_UTF8(getSB_VAL_CP, pos, strend)
2058 /* Returns the WB value for the input code point */
2059 #define getWB_VAL_CP(cp) \
2060 _generic_GET_BREAK_VAL_CP( \
2065 /* Returns the WB value for the first code point in the UTF-8 encoded string
2066 * bounded by pos and strend */
2067 #define getWB_VAL_UTF8(pos, strend) \
2068 _generic_GET_BREAK_VAL_UTF8(getWB_VAL_CP, pos, strend)
2070 /* We know what class REx starts with. Try to find this position... */
2071 /* if reginfo->intuit, its a dryrun */
2072 /* annoyingly all the vars in this routine have different names from their counterparts
2073 in regmatch. /grrr */
2075 S_find_byclass(pTHX_ regexp * prog, const regnode *c, char *s,
2076 const char *strend, regmatch_info *reginfo)
2080 /* TRUE if x+ need not match at just the 1st pos of run of x's */
2081 const I32 doevery = (prog->intflags & PREGf_SKIP) == 0;
2083 char *pat_string; /* The pattern's exactish string */
2084 char *pat_end; /* ptr to end char of pat_string */
2085 re_fold_t folder; /* Function for computing non-utf8 folds */
2086 const U8 *fold_array; /* array for folding ords < 256 */
2093 /* In some cases we accept only the first occurence of 'x' in a sequence of
2094 * them. This variable points to just beyond the end of the previous
2095 * occurrence of 'x', hence we can tell if we are in a sequence. (Having
2096 * it point to beyond the 'x' allows us to work for UTF-8 without having to
2098 char * previous_occurrence_end = 0;
2100 I32 tmp; /* Scratch variable */
2101 const bool utf8_target = reginfo->is_utf8_target;
2102 UV utf8_fold_flags = 0;
2103 const bool is_utf8_pat = reginfo->is_utf8_pat;
2104 bool to_complement = FALSE; /* Invert the result? Taking the xor of this
2105 with a result inverts that result, as 0^1 =
2107 _char_class_number classnum;
2109 RXi_GET_DECL(prog,progi);
2111 PERL_ARGS_ASSERT_FIND_BYCLASS;
2113 /* We know what class it must start with. */
2117 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2119 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(c)) && ! IN_UTF8_CTYPE_LOCALE) {
2120 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
2127 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2128 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
2130 else if (ANYOF_FLAGS(c) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
2131 /* We know that s is in the bitmap range since the target isn't
2132 * UTF-8, so what happens for out-of-range values is not relevant,
2133 * so exclude that from the flags */
2134 REXEC_FBC_CLASS_SCAN(0, reginclass(prog,c, (U8*)s, (U8*)s+1, 0));
2137 REXEC_FBC_CLASS_SCAN(0, ANYOF_BITMAP_TEST(c, *((U8*)s)));
2141 case ANYOFM: /* ARG() is the base byte; FLAGS() the mask byte */
2142 /* UTF-8ness doesn't matter, so use 0 */
2143 REXEC_FBC_FIND_NEXT_SCAN(0,
2144 (char *) find_next_masked((U8 *) s, (U8 *) strend,
2145 (U8) ARG(c), FLAGS(c)));
2149 REXEC_FBC_FIND_NEXT_SCAN(0,
2150 (char *) find_span_end_mask((U8 *) s, (U8 *) strend,
2151 (U8) ARG(c), FLAGS(c)));
2155 if (utf8_target) REXEC_FBC_CLASS_SCAN(TRUE,
2156 reginclass(prog, c, (U8*)s, (U8*) strend, utf8_target));
2159 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
2160 assert(! is_utf8_pat);
2164 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII
2165 |FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
2166 goto do_exactf_utf8;
2168 else if (utf8_target) {
2170 /* Here, and elsewhere in this file, the reason we can't consider a
2171 * non-UTF-8 pattern already folded in the presence of a UTF-8
2172 * target is because any MICRO SIGN in the pattern won't be folded.
2173 * Since the fold of the MICRO SIGN requires UTF-8 to represent, we
2174 * can consider a non-UTF-8 pattern folded when matching a
2175 * non-UTF-8 target */
2176 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
2177 goto do_exactf_utf8;
2180 /* Latin1 folds are not affected by /a, except it excludes the sharp s,
2181 * which these functions don't handle anyway */
2182 fold_array = PL_fold_latin1;
2183 folder = foldEQ_latin1_s2_folded;
2184 goto do_exactf_non_utf8;
2186 case EXACTF: /* This node only generated for non-utf8 patterns */
2187 assert(! is_utf8_pat);
2189 goto do_exactf_utf8;
2191 fold_array = PL_fold;
2193 goto do_exactf_non_utf8;
2196 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2197 if (is_utf8_pat || utf8_target || IN_UTF8_CTYPE_LOCALE) {
2198 utf8_fold_flags = FOLDEQ_LOCALE;
2199 goto do_exactf_utf8;
2201 fold_array = PL_fold_locale;
2202 folder = foldEQ_locale;
2203 goto do_exactf_non_utf8;
2205 case EXACTFUP: /* Problematic even though pattern isn't UTF-8. Use
2206 full functionality normally not done except for
2208 assert(! is_utf8_pat);
2209 goto do_exactf_utf8;
2212 if (! utf8_target) { /* All code points in this node require
2213 UTF-8 to express. */
2216 utf8_fold_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
2217 | FOLDEQ_S2_FOLDS_SANE;
2218 goto do_exactf_utf8;
2221 if (! utf8_target) {
2224 assert(is_utf8_pat);
2225 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
2226 goto do_exactf_utf8;
2229 if (is_utf8_pat || utf8_target) {
2230 utf8_fold_flags = FOLDEQ_S2_ALREADY_FOLDED;
2231 goto do_exactf_utf8;
2234 /* Any 'ss' in the pattern should have been replaced by regcomp,
2235 * so we don't have to worry here about this single special case
2236 * in the Latin1 range */
2237 fold_array = PL_fold_latin1;
2238 folder = foldEQ_latin1_s2_folded;
2242 do_exactf_non_utf8: /* Neither pattern nor string are UTF8, and there
2243 are no glitches with fold-length differences
2244 between the target string and pattern */
2246 /* The idea in the non-utf8 EXACTF* cases is to first find the
2247 * first character of the EXACTF* node and then, if necessary,
2248 * case-insensitively compare the full text of the node. c1 is the
2249 * first character. c2 is its fold. This logic will not work for
2250 * Unicode semantics and the german sharp ss, which hence should
2251 * not be compiled into a node that gets here. */
2252 pat_string = STRING(c);
2253 ln = STR_LEN(c); /* length to match in octets/bytes */
2255 /* We know that we have to match at least 'ln' bytes (which is the
2256 * same as characters, since not utf8). If we have to match 3
2257 * characters, and there are only 2 availabe, we know without
2258 * trying that it will fail; so don't start a match past the
2259 * required minimum number from the far end */
2260 e = HOP3c(strend, -((SSize_t)ln), s);
2265 c2 = fold_array[c1];
2266 if (c1 == c2) { /* If char and fold are the same */
2268 s = (char *) memchr(s, c1, e + 1 - s);
2273 /* Check that the rest of the node matches */
2274 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2275 && (reginfo->intuit || regtry(reginfo, &s)) )
2283 U8 bits_differing = c1 ^ c2;
2285 /* If the folds differ in one bit position only, we can mask to
2286 * match either of them, and can use this faster find method. Both
2287 * ASCII and EBCDIC tend to have their case folds differ in only
2288 * one position, so this is very likely */
2289 if (LIKELY(PL_bitcount[bits_differing] == 1)) {
2290 bits_differing = ~ bits_differing;
2292 s = (char *) find_next_masked((U8 *) s, (U8 *) e + 1,
2293 (c1 & bits_differing), bits_differing);
2298 if ( (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2299 && (reginfo->intuit || regtry(reginfo, &s)) )
2306 else { /* Otherwise, stuck with looking byte-at-a-time. This
2307 should actually happen only in EXACTFL nodes */
2309 if ( (*(U8*)s == c1 || *(U8*)s == c2)
2310 && (ln == 1 || folder(s + 1, pat_string + 1, ln - 1))
2311 && (reginfo->intuit || regtry(reginfo, &s)) )
2325 /* If one of the operands is in utf8, we can't use the simpler folding
2326 * above, due to the fact that many different characters can have the
2327 * same fold, or portion of a fold, or different- length fold */
2328 pat_string = STRING(c);
2329 ln = STR_LEN(c); /* length to match in octets/bytes */
2330 pat_end = pat_string + ln;
2331 lnc = is_utf8_pat /* length to match in characters */
2332 ? utf8_length((U8 *) pat_string, (U8 *) pat_end)
2335 /* We have 'lnc' characters to match in the pattern, but because of
2336 * multi-character folding, each character in the target can match
2337 * up to 3 characters (Unicode guarantees it will never exceed
2338 * this) if it is utf8-encoded; and up to 2 if not (based on the
2339 * fact that the Latin 1 folds are already determined, and the
2340 * only multi-char fold in that range is the sharp-s folding to
2341 * 'ss'. Thus, a pattern character can match as little as 1/3 of a
2342 * string character. Adjust lnc accordingly, rounding up, so that
2343 * if we need to match at least 4+1/3 chars, that really is 5. */
2344 expansion = (utf8_target) ? UTF8_MAX_FOLD_CHAR_EXPAND : 2;
2345 lnc = (lnc + expansion - 1) / expansion;
2347 /* As in the non-UTF8 case, if we have to match 3 characters, and
2348 * only 2 are left, it's guaranteed to fail, so don't start a
2349 * match that would require us to go beyond the end of the string
2351 e = HOP3c(strend, -((SSize_t)lnc), s);
2353 /* XXX Note that we could recalculate e to stop the loop earlier,
2354 * as the worst case expansion above will rarely be met, and as we
2355 * go along we would usually find that e moves further to the left.
2356 * This would happen only after we reached the point in the loop
2357 * where if there were no expansion we should fail. Unclear if
2358 * worth the expense */
2361 char *my_strend= (char *)strend;
2362 if (foldEQ_utf8_flags(s, &my_strend, 0, utf8_target,
2363 pat_string, NULL, ln, is_utf8_pat, utf8_fold_flags)
2364 && (reginfo->intuit || regtry(reginfo, &s)) )
2368 s += (utf8_target) ? UTF8SKIP(s) : 1;
2374 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2375 if (FLAGS(c) != TRADITIONAL_BOUND) {
2376 if (! IN_UTF8_CTYPE_LOCALE) {
2377 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2378 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2383 FBC_BOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2387 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2388 if (FLAGS(c) != TRADITIONAL_BOUND) {
2389 if (! IN_UTF8_CTYPE_LOCALE) {
2390 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2391 B_ON_NON_UTF8_LOCALE_IS_WRONG);
2396 FBC_NBOUND(isWORDCHAR_LC, isWORDCHAR_LC_uvchr, isWORDCHAR_LC_utf8_safe);
2399 case BOUND: /* regcomp.c makes sure that this only has the traditional \b
2401 assert(FLAGS(c) == TRADITIONAL_BOUND);
2403 FBC_BOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2406 case BOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2408 assert(FLAGS(c) == TRADITIONAL_BOUND);
2410 FBC_BOUND_A(isWORDCHAR_A);
2413 case NBOUND: /* regcomp.c makes sure that this only has the traditional \b
2415 assert(FLAGS(c) == TRADITIONAL_BOUND);
2417 FBC_NBOUND(isWORDCHAR, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2420 case NBOUNDA: /* regcomp.c makes sure that this only has the traditional \b
2422 assert(FLAGS(c) == TRADITIONAL_BOUND);
2424 FBC_NBOUND_A(isWORDCHAR_A);
2428 if ((bound_type) FLAGS(c) == TRADITIONAL_BOUND) {
2429 FBC_NBOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2440 switch((bound_type) FLAGS(c)) {
2441 case TRADITIONAL_BOUND:
2442 FBC_BOUND(isWORDCHAR_L1, isWORDCHAR_uni, isWORDCHAR_utf8_safe);
2445 if (s == reginfo->strbeg) {
2446 if (reginfo->intuit || regtry(reginfo, &s))
2451 /* Didn't match. Try at the next position (if there is one) */
2452 s += (utf8_target) ? UTF8SKIP(s) : 1;
2453 if (UNLIKELY(s >= reginfo->strend)) {
2459 GCB_enum before = getGCB_VAL_UTF8(
2461 (U8*)(reginfo->strbeg)),
2462 (U8*) reginfo->strend);
2463 while (s < strend) {
2464 GCB_enum after = getGCB_VAL_UTF8((U8*) s,
2465 (U8*) reginfo->strend);
2466 if ( (to_complement ^ isGCB(before,
2468 (U8*) reginfo->strbeg,
2471 && (reginfo->intuit || regtry(reginfo, &s)))
2479 else { /* Not utf8. Everything is a GCB except between CR and
2481 while (s < strend) {
2482 if ((to_complement ^ ( UCHARAT(s - 1) != '\r'
2483 || UCHARAT(s) != '\n'))
2484 && (reginfo->intuit || regtry(reginfo, &s)))
2492 /* And, since this is a bound, it can match after the final
2493 * character in the string */
2494 if ((reginfo->intuit || regtry(reginfo, &s))) {
2500 if (s == reginfo->strbeg) {
2501 if (reginfo->intuit || regtry(reginfo, &s)) {
2504 s += (utf8_target) ? UTF8SKIP(s) : 1;
2505 if (UNLIKELY(s >= reginfo->strend)) {
2511 LB_enum before = getLB_VAL_UTF8(reghop3((U8*)s,
2513 (U8*)(reginfo->strbeg)),
2514 (U8*) reginfo->strend);
2515 while (s < strend) {
2516 LB_enum after = getLB_VAL_UTF8((U8*) s, (U8*) reginfo->strend);
2517 if (to_complement ^ isLB(before,
2519 (U8*) reginfo->strbeg,
2521 (U8*) reginfo->strend,
2523 && (reginfo->intuit || regtry(reginfo, &s)))
2531 else { /* Not utf8. */
2532 LB_enum before = getLB_VAL_CP((U8) *(s -1));
2533 while (s < strend) {
2534 LB_enum after = getLB_VAL_CP((U8) *s);
2535 if (to_complement ^ isLB(before,
2537 (U8*) reginfo->strbeg,
2539 (U8*) reginfo->strend,
2541 && (reginfo->intuit || regtry(reginfo, &s)))
2550 if (reginfo->intuit || regtry(reginfo, &s)) {
2557 if (s == reginfo->strbeg) {
2558 if (reginfo->intuit || regtry(reginfo, &s)) {
2561 s += (utf8_target) ? UTF8SKIP(s) : 1;
2562 if (UNLIKELY(s >= reginfo->strend)) {
2568 SB_enum before = getSB_VAL_UTF8(reghop3((U8*)s,
2570 (U8*)(reginfo->strbeg)),
2571 (U8*) reginfo->strend);
2572 while (s < strend) {
2573 SB_enum after = getSB_VAL_UTF8((U8*) s,
2574 (U8*) reginfo->strend);
2575 if ((to_complement ^ isSB(before,
2577 (U8*) reginfo->strbeg,
2579 (U8*) reginfo->strend,
2581 && (reginfo->intuit || regtry(reginfo, &s)))
2589 else { /* Not utf8. */
2590 SB_enum before = getSB_VAL_CP((U8) *(s -1));
2591 while (s < strend) {
2592 SB_enum after = getSB_VAL_CP((U8) *s);
2593 if ((to_complement ^ isSB(before,
2595 (U8*) reginfo->strbeg,
2597 (U8*) reginfo->strend,
2599 && (reginfo->intuit || regtry(reginfo, &s)))
2608 /* Here are at the final position in the target string. The SB
2609 * value is always true here, so matches, depending on other
2611 if (reginfo->intuit || regtry(reginfo, &s)) {
2618 if (s == reginfo->strbeg) {
2619 if (reginfo->intuit || regtry(reginfo, &s)) {
2622 s += (utf8_target) ? UTF8SKIP(s) : 1;
2623 if (UNLIKELY(s >= reginfo->strend)) {
2629 /* We are at a boundary between char_sub_0 and char_sub_1.
2630 * We also keep track of the value for char_sub_-1 as we
2631 * loop through the line. Context may be needed to make a
2632 * determination, and if so, this can save having to
2634 WB_enum previous = WB_UNKNOWN;
2635 WB_enum before = getWB_VAL_UTF8(
2638 (U8*)(reginfo->strbeg)),
2639 (U8*) reginfo->strend);
2640 while (s < strend) {
2641 WB_enum after = getWB_VAL_UTF8((U8*) s,
2642 (U8*) reginfo->strend);
2643 if ((to_complement ^ isWB(previous,
2646 (U8*) reginfo->strbeg,
2648 (U8*) reginfo->strend,
2650 && (reginfo->intuit || regtry(reginfo, &s)))
2659 else { /* Not utf8. */
2660 WB_enum previous = WB_UNKNOWN;
2661 WB_enum before = getWB_VAL_CP((U8) *(s -1));
2662 while (s < strend) {
2663 WB_enum after = getWB_VAL_CP((U8) *s);
2664 if ((to_complement ^ isWB(previous,
2667 (U8*) reginfo->strbeg,
2669 (U8*) reginfo->strend,
2671 && (reginfo->intuit || regtry(reginfo, &s)))
2681 if (reginfo->intuit || regtry(reginfo, &s)) {
2688 REXEC_FBC_CSCAN(is_LNBREAK_utf8_safe(s, strend),
2689 is_LNBREAK_latin1_safe(s, strend)
2693 /* The argument to all the POSIX node types is the class number to pass to
2694 * _generic_isCC() to build a mask for searching in PL_charclass[] */
2701 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2702 REXEC_FBC_CSCAN(to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(c), (U8 *) s, (U8 *) strend)),
2703 to_complement ^ cBOOL(isFOO_lc(FLAGS(c), *s)));
2718 /* The complement of something that matches only ASCII matches all
2719 * non-ASCII, plus everything in ASCII that isn't in the class. */
2720 REXEC_FBC_CLASS_SCAN(1, ! isASCII_utf8_safe(s, strend)
2721 || ! _generic_isCC_A(*s, FLAGS(c)));
2729 /* Don't need to worry about utf8, as it can match only a single
2730 * byte invariant character. But we do anyway for performance reasons,
2731 * as otherwise we would have to examine all the continuation
2734 REXEC_FBC_CLASS_SCAN(1, _generic_isCC_A(*s, FLAGS(c)));
2739 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
2740 to_complement ^ cBOOL(_generic_isCC_A(*s, FLAGS(c))));
2748 if (! utf8_target) {
2749 REXEC_FBC_CLASS_SCAN(0, /* 0=>not-utf8 */
2750 to_complement ^ cBOOL(_generic_isCC(*s,
2756 classnum = (_char_class_number) FLAGS(c);
2759 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2760 to_complement ^ cBOOL(_invlist_contains_cp(
2761 PL_XPosix_ptrs[classnum],
2762 utf8_to_uvchr_buf((U8 *) s,
2766 case _CC_ENUM_SPACE:
2767 REXEC_FBC_CLASS_SCAN(1, /* 1=>is-utf8 */
2768 to_complement ^ cBOOL(isSPACE_utf8_safe(s, strend)));
2771 case _CC_ENUM_BLANK:
2772 REXEC_FBC_CLASS_SCAN(1,
2773 to_complement ^ cBOOL(isBLANK_utf8_safe(s, strend)));
2776 case _CC_ENUM_XDIGIT:
2777 REXEC_FBC_CLASS_SCAN(1,
2778 to_complement ^ cBOOL(isXDIGIT_utf8_safe(s, strend)));
2781 case _CC_ENUM_VERTSPACE:
2782 REXEC_FBC_CLASS_SCAN(1,
2783 to_complement ^ cBOOL(isVERTWS_utf8_safe(s, strend)));
2786 case _CC_ENUM_CNTRL:
2787 REXEC_FBC_CLASS_SCAN(1,
2788 to_complement ^ cBOOL(isCNTRL_utf8_safe(s, strend)));
2798 /* what trie are we using right now */
2799 reg_ac_data *aho = (reg_ac_data*)progi->data->data[ ARG( c ) ];
2800 reg_trie_data *trie = (reg_trie_data*)progi->data->data[ aho->trie ];
2801 HV *widecharmap = MUTABLE_HV(progi->data->data[ aho->trie + 1 ]);
2803 const char *last_start = strend - trie->minlen;
2805 const char *real_start = s;
2807 STRLEN maxlen = trie->maxlen;
2809 U8 **points; /* map of where we were in the input string
2810 when reading a given char. For ASCII this
2811 is unnecessary overhead as the relationship
2812 is always 1:1, but for Unicode, especially
2813 case folded Unicode this is not true. */
2814 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2818 GET_RE_DEBUG_FLAGS_DECL;
2820 /* We can't just allocate points here. We need to wrap it in
2821 * an SV so it gets freed properly if there is a croak while
2822 * running the match */
2825 sv_points=newSV(maxlen * sizeof(U8 *));
2826 SvCUR_set(sv_points,
2827 maxlen * sizeof(U8 *));
2828 SvPOK_on(sv_points);
2829 sv_2mortal(sv_points);
2830 points=(U8**)SvPV_nolen(sv_points );
2831 if ( trie_type != trie_utf8_fold
2832 && (trie->bitmap || OP(c)==AHOCORASICKC) )
2835 bitmap=(U8*)trie->bitmap;
2837 bitmap=(U8*)ANYOF_BITMAP(c);
2839 /* this is the Aho-Corasick algorithm modified a touch
2840 to include special handling for long "unknown char" sequences.
2841 The basic idea being that we use AC as long as we are dealing
2842 with a possible matching char, when we encounter an unknown char
2843 (and we have not encountered an accepting state) we scan forward
2844 until we find a legal starting char.
2845 AC matching is basically that of trie matching, except that when
2846 we encounter a failing transition, we fall back to the current
2847 states "fail state", and try the current char again, a process
2848 we repeat until we reach the root state, state 1, or a legal
2849 transition. If we fail on the root state then we can either
2850 terminate if we have reached an accepting state previously, or
2851 restart the entire process from the beginning if we have not.
2854 while (s <= last_start) {
2855 const U32 uniflags = UTF8_ALLOW_DEFAULT;
2863 U8 *uscan = (U8*)NULL;
2864 U8 *leftmost = NULL;
2866 U32 accepted_word= 0;
2870 while ( state && uc <= (U8*)strend ) {
2872 U32 word = aho->states[ state ].wordnum;
2876 DEBUG_TRIE_EXECUTE_r(
2877 if ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2878 dump_exec_pos( (char *)uc, c, strend, real_start,
2879 (char *)uc, utf8_target, 0 );
2880 Perl_re_printf( aTHX_
2881 " Scanning for legal start char...\n");
2885 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2889 while ( uc <= (U8*)last_start && !BITMAP_TEST(bitmap,*uc) ) {
2895 if (uc >(U8*)last_start) break;
2899 U8 *lpos= points[ (pointpos - trie->wordinfo[word].len) % maxlen ];
2900 if (!leftmost || lpos < leftmost) {
2901 DEBUG_r(accepted_word=word);
2907 points[pointpos++ % maxlen]= uc;
2908 if (foldlen || uc < (U8*)strend) {
2909 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
2910 (U8 *) strend, uscan, len, uvc,
2911 charid, foldlen, foldbuf,
2913 DEBUG_TRIE_EXECUTE_r({
2914 dump_exec_pos( (char *)uc, c, strend,
2915 real_start, s, utf8_target, 0);
2916 Perl_re_printf( aTHX_
2917 " Charid:%3u CP:%4" UVxf " ",
2929 word = aho->states[ state ].wordnum;
2931 base = aho->states[ state ].trans.base;
2933 DEBUG_TRIE_EXECUTE_r({
2935 dump_exec_pos( (char *)uc, c, strend, real_start,
2936 s, utf8_target, 0 );
2937 Perl_re_printf( aTHX_
2938 "%sState: %4" UVxf ", word=%" UVxf,
2939 failed ? " Fail transition to " : "",
2940 (UV)state, (UV)word);
2946 ( ((offset = base + charid
2947 - 1 - trie->uniquecharcount)) >= 0)
2948 && ((U32)offset < trie->lasttrans)
2949 && trie->trans[offset].check == state
2950 && (tmp=trie->trans[offset].next))
2952 DEBUG_TRIE_EXECUTE_r(
2953 Perl_re_printf( aTHX_ " - legal\n"));
2958 DEBUG_TRIE_EXECUTE_r(
2959 Perl_re_printf( aTHX_ " - fail\n"));
2961 state = aho->fail[state];
2965 /* we must be accepting here */
2966 DEBUG_TRIE_EXECUTE_r(
2967 Perl_re_printf( aTHX_ " - accepting\n"));
2976 if (!state) state = 1;
2979 if ( aho->states[ state ].wordnum ) {
2980 U8 *lpos = points[ (pointpos - trie->wordinfo[aho->states[ state ].wordnum].len) % maxlen ];
2981 if (!leftmost || lpos < leftmost) {
2982 DEBUG_r(accepted_word=aho->states[ state ].wordnum);
2987 s = (char*)leftmost;
2988 DEBUG_TRIE_EXECUTE_r({
2989 Perl_re_printf( aTHX_ "Matches word #%" UVxf " at position %" IVdf ". Trying full pattern...\n",
2990 (UV)accepted_word, (IV)(s - real_start)
2993 if (reginfo->intuit || regtry(reginfo, &s)) {
2999 DEBUG_TRIE_EXECUTE_r({
3000 Perl_re_printf( aTHX_ "Pattern failed. Looking for new start point...\n");
3003 DEBUG_TRIE_EXECUTE_r(
3004 Perl_re_printf( aTHX_ "No match.\n"));
3013 Perl_croak(aTHX_ "panic: unknown regstclass %d", (int)OP(c));
3020 /* set RX_SAVED_COPY, RX_SUBBEG etc.
3021 * flags have same meanings as with regexec_flags() */
3024 S_reg_set_capture_string(pTHX_ REGEXP * const rx,
3031 struct regexp *const prog = ReANY(rx);
3033 if (flags & REXEC_COPY_STR) {
3036 DEBUG_C(Perl_re_printf( aTHX_
3037 "Copy on write: regexp capture, type %d\n",
3039 /* Create a new COW SV to share the match string and store
3040 * in saved_copy, unless the current COW SV in saved_copy
3041 * is valid and suitable for our purpose */
3042 if (( prog->saved_copy
3043 && SvIsCOW(prog->saved_copy)
3044 && SvPOKp(prog->saved_copy)
3047 && SvPVX(sv) == SvPVX(prog->saved_copy)))
3049 /* just reuse saved_copy SV */
3050 if (RXp_MATCH_COPIED(prog)) {
3051 Safefree(prog->subbeg);
3052 RXp_MATCH_COPIED_off(prog);
3056 /* create new COW SV to share string */
3057 RXp_MATCH_COPY_FREE(prog);
3058 prog->saved_copy = sv_setsv_cow(prog->saved_copy, sv);
3060 prog->subbeg = (char *)SvPVX_const(prog->saved_copy);
3061 assert (SvPOKp(prog->saved_copy));
3062 prog->sublen = strend - strbeg;
3063 prog->suboffset = 0;
3064 prog->subcoffset = 0;
3069 SSize_t max = strend - strbeg;
3072 if ( (flags & REXEC_COPY_SKIP_POST)
3073 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
3074 && !(PL_sawampersand & SAWAMPERSAND_RIGHT)
3075 ) { /* don't copy $' part of string */
3078 /* calculate the right-most part of the string covered
3079 * by a capture. Due to lookahead, this may be to
3080 * the right of $&, so we have to scan all captures */
3081 while (n <= prog->lastparen) {
3082 if (prog->offs[n].end > max)
3083 max = prog->offs[n].end;
3087 max = (PL_sawampersand & SAWAMPERSAND_LEFT)
3088 ? prog->offs[0].start
3090 assert(max >= 0 && max <= strend - strbeg);
3093 if ( (flags & REXEC_COPY_SKIP_PRE)
3094 && !(prog->extflags & RXf_PMf_KEEPCOPY) /* //p */
3095 && !(PL_sawampersand & SAWAMPERSAND_LEFT)
3096 ) { /* don't copy $` part of string */
3099 /* calculate the left-most part of the string covered
3100 * by a capture. Due to lookbehind, this may be to
3101 * the left of $&, so we have to scan all captures */
3102 while (min && n <= prog->lastparen) {
3103 if ( prog->offs[n].start != -1
3104 && prog->offs[n].start < min)
3106 min = prog->offs[n].start;
3110 if ((PL_sawampersand & SAWAMPERSAND_RIGHT)
3111 && min > prog->offs[0].end
3113 min = prog->offs[0].end;
3117 assert(min >= 0 && min <= max && min <= strend - strbeg);
3120 if (RXp_MATCH_COPIED(prog)) {
3121 if (sublen > prog->sublen)
3123 (char*)saferealloc(prog->subbeg, sublen+1);
3126 prog->subbeg = (char*)safemalloc(sublen+1);
3127 Copy(strbeg + min, prog->subbeg, sublen, char);
3128 prog->subbeg[sublen] = '\0';
3129 prog->suboffset = min;
3130 prog->sublen = sublen;
3131 RXp_MATCH_COPIED_on(prog);
3133 prog->subcoffset = prog->suboffset;
3134 if (prog->suboffset && utf8_target) {
3135 /* Convert byte offset to chars.
3136 * XXX ideally should only compute this if @-/@+
3137 * has been seen, a la PL_sawampersand ??? */
3139 /* If there's a direct correspondence between the
3140 * string which we're matching and the original SV,
3141 * then we can use the utf8 len cache associated with
3142 * the SV. In particular, it means that under //g,
3143 * sv_pos_b2u() will use the previously cached
3144 * position to speed up working out the new length of
3145 * subcoffset, rather than counting from the start of
3146 * the string each time. This stops
3147 * $x = "\x{100}" x 1E6; 1 while $x =~ /(.)/g;
3148 * from going quadratic */
3149 if (SvPOKp(sv) && SvPVX(sv) == strbeg)
3150 prog->subcoffset = sv_pos_b2u_flags(sv, prog->subcoffset,
3151 SV_GMAGIC|SV_CONST_RETURN);
3153 prog->subcoffset = utf8_length((U8*)strbeg,
3154 (U8*)(strbeg+prog->suboffset));
3158 RXp_MATCH_COPY_FREE(prog);
3159 prog->subbeg = strbeg;
3160 prog->suboffset = 0;
3161 prog->subcoffset = 0;
3162 prog->sublen = strend - strbeg;
3170 - regexec_flags - match a regexp against a string
3173 Perl_regexec_flags(pTHX_ REGEXP * const rx, char *stringarg, char *strend,
3174 char *strbeg, SSize_t minend, SV *sv, void *data, U32 flags)
3175 /* stringarg: the point in the string at which to begin matching */
3176 /* strend: pointer to null at end of string */
3177 /* strbeg: real beginning of string */
3178 /* minend: end of match must be >= minend bytes after stringarg. */
3179 /* sv: SV being matched: only used for utf8 flag, pos() etc; string
3180 * itself is accessed via the pointers above */
3181 /* data: May be used for some additional optimizations.
3182 Currently unused. */
3183 /* flags: For optimizations. See REXEC_* in regexp.h */
3186 struct regexp *const prog = ReANY(rx);
3190 SSize_t minlen; /* must match at least this many chars */
3191 SSize_t dontbother = 0; /* how many characters not to try at end */
3192 const bool utf8_target = cBOOL(DO_UTF8(sv));
3194 RXi_GET_DECL(prog,progi);
3195 regmatch_info reginfo_buf; /* create some info to pass to regtry etc */
3196 regmatch_info *const reginfo = ®info_buf;
3197 regexp_paren_pair *swap = NULL;
3199 GET_RE_DEBUG_FLAGS_DECL;
3201 PERL_ARGS_ASSERT_REGEXEC_FLAGS;
3202 PERL_UNUSED_ARG(data);
3204 /* Be paranoid... */
3206 Perl_croak(aTHX_ "NULL regexp parameter");
3210 debug_start_match(rx, utf8_target, stringarg, strend,
3214 startpos = stringarg;
3216 /* set these early as they may be used by the HOP macros below */
3217 reginfo->strbeg = strbeg;
3218 reginfo->strend = strend;
3219 reginfo->is_utf8_target = cBOOL(utf8_target);
3221 if (prog->intflags & PREGf_GPOS_SEEN) {
3224 /* set reginfo->ganch, the position where \G can match */
3227 (flags & REXEC_IGNOREPOS)
3228 ? stringarg /* use start pos rather than pos() */
3229 : ((mg = mg_find_mglob(sv)) && mg->mg_len >= 0)
3230 /* Defined pos(): */
3231 ? strbeg + MgBYTEPOS(mg, sv, strbeg, strend-strbeg)
3232 : strbeg; /* pos() not defined; use start of string */
3234 DEBUG_GPOS_r(Perl_re_printf( aTHX_
3235 "GPOS ganch set to strbeg[%" IVdf "]\n", (IV)(reginfo->ganch - strbeg)));
3237 /* in the presence of \G, we may need to start looking earlier in
3238 * the string than the suggested start point of stringarg:
3239 * if prog->gofs is set, then that's a known, fixed minimum
3242 * /ab|c\G/: gofs = 1
3243 * or if the minimum offset isn't known, then we have to go back
3244 * to the start of the string, e.g. /w+\G/
3247 if (prog->intflags & PREGf_ANCH_GPOS) {
3249 startpos = HOPBACKc(reginfo->ganch, prog->gofs);
3251 ((flags & REXEC_FAIL_ON_UNDERFLOW) && startpos < stringarg))
3253 DEBUG_r(Perl_re_printf( aTHX_
3254 "fail: ganch-gofs before earliest possible start\n"));
3259 startpos = reginfo->ganch;
3261 else if (prog->gofs) {
3262 startpos = HOPBACKc(startpos, prog->gofs);
3266 else if (prog->intflags & PREGf_GPOS_FLOAT)
3270 minlen = prog->minlen;
3271 if ((startpos + minlen) > strend || startpos < strbeg) {
3272 DEBUG_r(Perl_re_printf( aTHX_
3273 "Regex match can't succeed, so not even tried\n"));
3277 /* at the end of this function, we'll do a LEAVE_SCOPE(oldsave),
3278 * which will call destuctors to reset PL_regmatch_state, free higher
3279 * PL_regmatch_slabs, and clean up regmatch_info_aux and
3280 * regmatch_info_aux_eval */
3282 oldsave = PL_savestack_ix;
3286 if ((prog->extflags & RXf_USE_INTUIT)
3287 && !(flags & REXEC_CHECKED))
3289 s = re_intuit_start(rx, sv, strbeg, startpos, strend,
3294 if (prog->extflags & RXf_CHECK_ALL) {
3295 /* we can match based purely on the result of INTUIT.
3296 * Set up captures etc just for $& and $-[0]
3297 * (an intuit-only match wont have $1,$2,..) */
3298 assert(!prog->nparens);
3300 /* s/// doesn't like it if $& is earlier than where we asked it to
3301 * start searching (which can happen on something like /.\G/) */
3302 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3305 /* this should only be possible under \G */
3306 assert(prog->intflags & PREGf_GPOS_SEEN);
3307 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3308 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3312 /* match via INTUIT shouldn't have any captures.
3313 * Let @-, @+, $^N know */
3314 prog->lastparen = prog->lastcloseparen = 0;
3315 RXp_MATCH_UTF8_set(prog, utf8_target);
3316 prog->offs[0].start = s - strbeg;
3317 prog->offs[0].end = utf8_target
3318 ? (char*)utf8_hop((U8*)s, prog->minlenret) - strbeg
3319 : s - strbeg + prog->minlenret;
3320 if ( !(flags & REXEC_NOT_FIRST) )
3321 S_reg_set_capture_string(aTHX_ rx,
3323 sv, flags, utf8_target);
3329 multiline = prog->extflags & RXf_PMf_MULTILINE;
3331 if (strend - s < (minlen+(prog->check_offset_min<0?prog->check_offset_min:0))) {
3332 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3333 "String too short [regexec_flags]...\n"));
3337 /* Check validity of program. */
3338 if (UCHARAT(progi->program) != REG_MAGIC) {
3339 Perl_croak(aTHX_ "corrupted regexp program");
3342 RXp_MATCH_TAINTED_off(prog);
3343 RXp_MATCH_UTF8_set(prog, utf8_target);
3345 reginfo->prog = rx; /* Yes, sorry that this is confusing. */
3346 reginfo->intuit = 0;
3347 reginfo->is_utf8_pat = cBOOL(RX_UTF8(rx));
3348 reginfo->warned = FALSE;
3350 reginfo->poscache_maxiter = 0; /* not yet started a countdown */
3351 /* see how far we have to get to not match where we matched before */
3352 reginfo->till = stringarg + minend;
3354 if (prog->extflags & RXf_EVAL_SEEN && SvPADTMP(sv)) {
3355 /* SAVEFREESV, not sv_mortalcopy, as this SV must last until after
3356 S_cleanup_regmatch_info_aux has executed (registered by
3357 SAVEDESTRUCTOR_X below). S_cleanup_regmatch_info_aux modifies
3358 magic belonging to this SV.
3359 Not newSVsv, either, as it does not COW.
3361 reginfo->sv = newSV(0);
3362 SvSetSV_nosteal(reginfo->sv, sv);
3363 SAVEFREESV(reginfo->sv);
3366 /* reserve next 2 or 3 slots in PL_regmatch_state:
3367 * slot N+0: may currently be in use: skip it
3368 * slot N+1: use for regmatch_info_aux struct
3369 * slot N+2: use for regmatch_info_aux_eval struct if we have (?{})'s
3370 * slot N+3: ready for use by regmatch()
3374 regmatch_state *old_regmatch_state;
3375 regmatch_slab *old_regmatch_slab;
3376 int i, max = (prog->extflags & RXf_EVAL_SEEN) ? 2 : 1;
3378 /* on first ever match, allocate first slab */
3379 if (!PL_regmatch_slab) {
3380 Newx(PL_regmatch_slab, 1, regmatch_slab);
3381 PL_regmatch_slab->prev = NULL;
3382 PL_regmatch_slab->next = NULL;
3383 PL_regmatch_state = SLAB_FIRST(PL_regmatch_slab);
3386 old_regmatch_state = PL_regmatch_state;
3387 old_regmatch_slab = PL_regmatch_slab;
3389 for (i=0; i <= max; i++) {
3391 reginfo->info_aux = &(PL_regmatch_state->u.info_aux);
3393 reginfo->info_aux_eval =
3394 reginfo->info_aux->info_aux_eval =
3395 &(PL_regmatch_state->u.info_aux_eval);
3397 if (++PL_regmatch_state > SLAB_LAST(PL_regmatch_slab))
3398 PL_regmatch_state = S_push_slab(aTHX);
3401 /* note initial PL_regmatch_state position; at end of match we'll
3402 * pop back to there and free any higher slabs */
3404 reginfo->info_aux->old_regmatch_state = old_regmatch_state;
3405 reginfo->info_aux->old_regmatch_slab = old_regmatch_slab;
3406 reginfo->info_aux->poscache = NULL;
3408 SAVEDESTRUCTOR_X(S_cleanup_regmatch_info_aux, reginfo->info_aux);
3410 if ((prog->extflags & RXf_EVAL_SEEN))
3411 S_setup_eval_state(aTHX_ reginfo);
3413 reginfo->info_aux_eval = reginfo->info_aux->info_aux_eval = NULL;
3416 /* If there is a "must appear" string, look for it. */
3418 if (PL_curpm && (PM_GETRE(PL_curpm) == rx)) {
3419 /* We have to be careful. If the previous successful match
3420 was from this regex we don't want a subsequent partially
3421 successful match to clobber the old results.
3422 So when we detect this possibility we add a swap buffer
3423 to the re, and switch the buffer each match. If we fail,
3424 we switch it back; otherwise we leave it swapped.
3427 /* avoid leak if we die, or clean up anyway if match completes */
3429 Newxz(prog->offs, (prog->nparens + 1), regexp_paren_pair);
3430 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3431 "rex=0x%" UVxf " saving offs: orig=0x%" UVxf " new=0x%" UVxf "\n",
3439 if (prog->recurse_locinput)
3440 Zero(prog->recurse_locinput,prog->nparens + 1, char *);
3442 /* Simplest case: anchored match need be tried only once, or with
3443 * MBOL, only at the beginning of each line.
3445 * Note that /.*.../ sets PREGf_IMPLICIT|MBOL, while /.*.../s sets
3446 * PREGf_IMPLICIT|SBOL. The idea is that with /.*.../s, if it doesn't
3447 * match at the start of the string then it won't match anywhere else
3448 * either; while with /.*.../, if it doesn't match at the beginning,
3449 * the earliest it could match is at the start of the next line */
3451 if (prog->intflags & (PREGf_ANCH & ~PREGf_ANCH_GPOS)) {
3454 if (regtry(reginfo, &s))
3457 if (!(prog->intflags & PREGf_ANCH_MBOL))
3460 /* didn't match at start, try at other newline positions */
3463 dontbother = minlen - 1;
3464 end = HOP3c(strend, -dontbother, strbeg) - 1;
3466 /* skip to next newline */
3468 while (s <= end) { /* note it could be possible to match at the end of the string */
3469 /* NB: newlines are the same in unicode as they are in latin */
3472 if (prog->check_substr || prog->check_utf8) {
3473 /* note that with PREGf_IMPLICIT, intuit can only fail
3474 * or return the start position, so it's of limited utility.
3475 * Nevertheless, I made the decision that the potential for
3476 * quick fail was still worth it - DAPM */
3477 s = re_intuit_start(rx, sv, strbeg, s, strend, flags, NULL);
3481 if (regtry(reginfo, &s))
3485 } /* end anchored search */
3487 if (prog->intflags & PREGf_ANCH_GPOS)
3489 /* PREGf_ANCH_GPOS should never be true if PREGf_GPOS_SEEN is not true */
3490 assert(prog->intflags & PREGf_GPOS_SEEN);
3491 /* For anchored \G, the only position it can match from is
3492 * (ganch-gofs); we already set startpos to this above; if intuit
3493 * moved us on from there, we can't possibly succeed */
3494 assert(startpos == HOPBACKc(reginfo->ganch, prog->gofs));
3495 if (s == startpos && regtry(reginfo, &s))
3500 /* Messy cases: unanchored match. */
3501 if ((prog->anchored_substr || prog->anchored_utf8) && prog->intflags & PREGf_SKIP) {
3502 /* we have /x+whatever/ */
3503 /* it must be a one character string (XXXX Except is_utf8_pat?) */
3509 if (! prog->anchored_utf8) {
3510 to_utf8_substr(prog);
3512 ch = SvPVX_const(prog->anchored_utf8)[0];
3513 REXEC_FBC_SCAN(0, /* 0=>not-utf8 */
3515 DEBUG_EXECUTE_r( did_match = 1 );
3516 if (regtry(reginfo, &s)) goto got_it;
3518 while (s < strend && *s == ch)
3525 if (! prog->anchored_substr) {
3526 if (! to_byte_substr(prog)) {
3527 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3530 ch = SvPVX_const(prog->anchored_substr)[0];
3531 REXEC_FBC_SCAN(0, /* 0=>not-utf8 */
3533 DEBUG_EXECUTE_r( did_match = 1 );
3534 if (regtry(reginfo, &s)) goto got_it;
3536 while (s < strend && *s == ch)
3541 DEBUG_EXECUTE_r(if (!did_match)
3542 Perl_re_printf( aTHX_
3543 "Did not find anchored character...\n")
3546 else if (prog->anchored_substr != NULL
3547 || prog->anchored_utf8 != NULL
3548 || ((prog->float_substr != NULL || prog->float_utf8 != NULL)
3549 && prog->float_max_offset < strend - s)) {
3554 char *last1; /* Last position checked before */
3558 if (prog->anchored_substr || prog->anchored_utf8) {
3560 if (! prog->anchored_utf8) {
3561 to_utf8_substr(prog);
3563 must = prog->anchored_utf8;
3566 if (! prog->anchored_substr) {
3567 if (! to_byte_substr(prog)) {
3568 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3571 must = prog->anchored_substr;
3573 back_max = back_min = prog->anchored_offset;
3576 if (! prog->float_utf8) {
3577 to_utf8_substr(prog);
3579 must = prog->float_utf8;
3582 if (! prog->float_substr) {
3583 if (! to_byte_substr(prog)) {
3584 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3587 must = prog->float_substr;
3589 back_max = prog->float_max_offset;
3590 back_min = prog->float_min_offset;
3596 last = HOP3c(strend, /* Cannot start after this */
3597 -(SSize_t)(CHR_SVLEN(must)
3598 - (SvTAIL(must) != 0) + back_min), strbeg);
3600 if (s > reginfo->strbeg)
3601 last1 = HOPc(s, -1);
3603 last1 = s - 1; /* bogus */
3605 /* XXXX check_substr already used to find "s", can optimize if
3606 check_substr==must. */
3608 strend = HOPc(strend, -dontbother);
3609 while ( (s <= last) &&
3610 (s = fbm_instr((unsigned char*)HOP4c(s, back_min, strbeg, strend),
3611 (unsigned char*)strend, must,
3612 multiline ? FBMrf_MULTILINE : 0)) ) {
3613 DEBUG_EXECUTE_r( did_match = 1 );
3614 if (HOPc(s, -back_max) > last1) {
3615 last1 = HOPc(s, -back_min);
3616 s = HOPc(s, -back_max);
3619 char * const t = (last1 >= reginfo->strbeg)
3620 ? HOPc(last1, 1) : last1 + 1;
3622 last1 = HOPc(s, -back_min);
3626 while (s <= last1) {
3627 if (regtry(reginfo, &s))
3630 s++; /* to break out of outer loop */
3637 while (s <= last1) {
3638 if (regtry(reginfo, &s))
3644 DEBUG_EXECUTE_r(if (!did_match) {
3645 RE_PV_QUOTED_DECL(quoted, utf8_target, PERL_DEBUG_PAD_ZERO(0),
3646 SvPVX_const(must), RE_SV_DUMPLEN(must), 30);
3647 Perl_re_printf( aTHX_ "Did not find %s substr %s%s...\n",
3648 ((must == prog->anchored_substr || must == prog->anchored_utf8)
3649 ? "anchored" : "floating"),
3650 quoted, RE_SV_TAIL(must));
3654 else if ( (c = progi->regstclass) ) {
3656 const OPCODE op = OP(progi->regstclass);
3657 /* don't bother with what can't match */
3658 if (PL_regkind[op] != EXACT && PL_regkind[op] != TRIE)
3659 strend = HOPc(strend, -(minlen - 1));
3662 SV * const prop = sv_newmortal();
3663 regprop(prog, prop, c, reginfo, NULL);
3665 RE_PV_QUOTED_DECL(quoted,utf8_target,PERL_DEBUG_PAD_ZERO(1),
3666 s,strend-s,PL_dump_re_max_len);
3667 Perl_re_printf( aTHX_
3668 "Matching stclass %.*s against %s (%d bytes)\n",
3669 (int)SvCUR(prop), SvPVX_const(prop),
3670 quoted, (int)(strend - s));
3673 if (find_byclass(prog, c, s, strend, reginfo))
3675 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "Contradicts stclass... [regexec_flags]\n"));
3679 if (prog->float_substr != NULL || prog->float_utf8 != NULL) {
3687 if (! prog->float_utf8) {
3688 to_utf8_substr(prog);
3690 float_real = prog->float_utf8;
3693 if (! prog->float_substr) {
3694 if (! to_byte_substr(prog)) {
3695 NON_UTF8_TARGET_BUT_UTF8_REQUIRED(phooey);
3698 float_real = prog->float_substr;
3701 little = SvPV_const(float_real, len);
3702 if (SvTAIL(float_real)) {
3703 /* This means that float_real contains an artificial \n on
3704 * the end due to the presence of something like this:
3705 * /foo$/ where we can match both "foo" and "foo\n" at the
3706 * end of the string. So we have to compare the end of the
3707 * string first against the float_real without the \n and
3708 * then against the full float_real with the string. We
3709 * have to watch out for cases where the string might be
3710 * smaller than the float_real or the float_real without
3712 char *checkpos= strend - len;
3714 Perl_re_printf( aTHX_
3715 "%sChecking for float_real.%s\n",
3716 PL_colors[4], PL_colors[5]));
3717 if (checkpos + 1 < strbeg) {
3718 /* can't match, even if we remove the trailing \n
3719 * string is too short to match */
3721 Perl_re_printf( aTHX_
3722 "%sString shorter than required trailing substring, cannot match.%s\n",
3723 PL_colors[4], PL_colors[5]));
3725 } else if (memEQ(checkpos + 1, little, len - 1)) {
3726 /* can match, the end of the string matches without the
3728 last = checkpos + 1;
3729 } else if (checkpos < strbeg) {
3730 /* cant match, string is too short when the "\n" is
3733 Perl_re_printf( aTHX_
3734 "%sString does not contain required trailing substring, cannot match.%s\n",
3735 PL_colors[4], PL_colors[5]));
3737 } else if (!multiline) {
3738 /* non multiline match, so compare with the "\n" at the
3739 * end of the string */
3740 if (memEQ(checkpos, little, len)) {
3744 Perl_re_printf( aTHX_
3745 "%sString does not contain required trailing substring, cannot match.%s\n",
3746 PL_colors[4], PL_colors[5]));
3750 /* multiline match, so we have to search for a place
3751 * where the full string is located */
3757 last = rninstr(s, strend, little, little + len);
3759 last = strend; /* matching "$" */
3762 /* at one point this block contained a comment which was
3763 * probably incorrect, which said that this was a "should not
3764 * happen" case. Even if it was true when it was written I am
3765 * pretty sure it is not anymore, so I have removed the comment
3766 * and replaced it with this one. Yves */
3768 Perl_re_printf( aTHX_
3769 "%sString does not contain required substring, cannot match.%s\n",
3770 PL_colors[4], PL_colors[5]
3774 dontbother = strend - last + prog->float_min_offset;
3776 if (minlen && (dontbother < minlen))
3777 dontbother = minlen - 1;
3778 strend -= dontbother; /* this one's always in bytes! */
3779 /* We don't know much -- general case. */
3782 if (regtry(reginfo, &s))
3791 if (regtry(reginfo, &s))
3793 } while (s++ < strend);
3801 /* s/// doesn't like it if $& is earlier than where we asked it to
3802 * start searching (which can happen on something like /.\G/) */
3803 if ( (flags & REXEC_FAIL_ON_UNDERFLOW)
3804 && (prog->offs[0].start < stringarg - strbeg))
3806 /* this should only be possible under \G */
3807 assert(prog->intflags & PREGf_GPOS_SEEN);
3808 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
3809 "matched, but failing for REXEC_FAIL_ON_UNDERFLOW\n"));
3813 /* clean up; this will trigger destructors that will free all slabs
3814 * above the current one, and cleanup the regmatch_info_aux
3815 * and regmatch_info_aux_eval sructs */
3817 LEAVE_SCOPE(oldsave);
3819 if (RXp_PAREN_NAMES(prog))
3820 (void)hv_iterinit(RXp_PAREN_NAMES(prog));
3822 /* make sure $`, $&, $', and $digit will work later */
3823 if ( !(flags & REXEC_NOT_FIRST) )
3824 S_reg_set_capture_string(aTHX_ rx,
3825 strbeg, reginfo->strend,
3826 sv, flags, utf8_target);
3831 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch failed%s\n",
3832 PL_colors[4], PL_colors[5]));
3835 /* we failed :-( roll it back.
3836 * Since the swap buffer will be freed on scope exit which follows
3837 * shortly, restore the old captures by copying 'swap's original
3838 * data to the new offs buffer
3840 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
3841 "rex=0x%" UVxf " rolling back offs: 0x%" UVxf " will be freed; restoring data to =0x%" UVxf "\n",
3848 Copy(swap, prog->offs, prog->nparens + 1, regexp_paren_pair);
3851 /* clean up; this will trigger destructors that will free all slabs
3852 * above the current one, and cleanup the regmatch_info_aux
3853 * and regmatch_info_aux_eval sructs */
3855 LEAVE_SCOPE(oldsave);
3861 /* Set which rex is pointed to by PL_reg_curpm, handling ref counting.
3862 * Do inc before dec, in case old and new rex are the same */
3863 #define SET_reg_curpm(Re2) \
3864 if (reginfo->info_aux_eval) { \
3865 (void)ReREFCNT_inc(Re2); \
3866 ReREFCNT_dec(PM_GETRE(PL_reg_curpm)); \
3867 PM_SETRE((PL_reg_curpm), (Re2)); \
3872 - regtry - try match at specific point
3874 STATIC bool /* 0 failure, 1 success */
3875 S_regtry(pTHX_ regmatch_info *reginfo, char **startposp)
3878 REGEXP *const rx = reginfo->prog;
3879 regexp *const prog = ReANY(rx);
3882 U32 depth = 0; /* used by REGCP_SET */
3884 RXi_GET_DECL(prog,progi);
3885 GET_RE_DEBUG_FLAGS_DECL;
3887 PERL_ARGS_ASSERT_REGTRY;
3889 reginfo->cutpoint=NULL;
3891 prog->offs[0].start = *startposp - reginfo->strbeg;
3892 prog->lastparen = 0;
3893 prog->lastcloseparen = 0;
3895 /* XXXX What this code is doing here?!!! There should be no need
3896 to do this again and again, prog->lastparen should take care of
3899 /* Tests pat.t#187 and split.t#{13,14} seem to depend on this code.
3900 * Actually, the code in regcppop() (which Ilya may be meaning by
3901 * prog->lastparen), is not needed at all by the test suite
3902 * (op/regexp, op/pat, op/split), but that code is needed otherwise
3903 * this erroneously leaves $1 defined: "1" =~ /^(?:(\d)x)?\d$/
3904 * Meanwhile, this code *is* needed for the
3905 * above-mentioned test suite tests to succeed. The common theme
3906 * on those tests seems to be returning null fields from matches.
3907 * --jhi updated by dapm */
3909 /* After encountering a variant of the issue mentioned above I think
3910 * the point Ilya was making is that if we properly unwind whenever
3911 * we set lastparen to a smaller value then we should not need to do
3912 * this every time, only when needed. So if we have tests that fail if
3913 * we remove this, then it suggests somewhere else we are improperly
3914 * unwinding the lastparen/paren buffers. See UNWIND_PARENS() and
3915 * places it is called, and related regcp() routines. - Yves */
3917 if (prog->nparens) {
3918 regexp_paren_pair *pp = prog->offs;
3920 for (i = prog->nparens; i > (I32)prog->lastparen; i--) {
3928 result = regmatch(reginfo, *startposp, progi->program + 1);
3930 prog->offs[0].end = result;
3933 if (reginfo->cutpoint)
3934 *startposp= reginfo->cutpoint;
3935 REGCP_UNWIND(lastcp);
3940 #define sayYES goto yes
3941 #define sayNO goto no
3942 #define sayNO_SILENT goto no_silent
3944 /* we dont use STMT_START/END here because it leads to
3945 "unreachable code" warnings, which are bogus, but distracting. */
3946 #define CACHEsayNO \
3947 if (ST.cache_mask) \
3948 reginfo->info_aux->poscache[ST.cache_offset] |= ST.cache_mask; \
3951 /* this is used to determine how far from the left messages like
3952 'failed...' are printed in regexec.c. It should be set such that
3953 messages are inline with the regop output that created them.
3955 #define REPORT_CODE_OFF 29
3956 #define INDENT_CHARS(depth) ((int)(depth) % 20)
3959 Perl_re_exec_indentf(pTHX_ const char *fmt, U32 depth, ...)
3963 PerlIO *f= Perl_debug_log;
3964 PERL_ARGS_ASSERT_RE_EXEC_INDENTF;
3965 va_start(ap, depth);
3966 PerlIO_printf(f, "%*s|%4" UVuf "| %*s", REPORT_CODE_OFF, "", (UV)depth, INDENT_CHARS(depth), "" );
3967 result = PerlIO_vprintf(f, fmt, ap);
3971 #endif /* DEBUGGING */
3974 #define CHRTEST_UNINIT -1001 /* c1/c2 haven't been calculated yet */
3975 #define CHRTEST_VOID -1000 /* the c1/c2 "next char" test should be skipped */
3976 #define CHRTEST_NOT_A_CP_1 -999
3977 #define CHRTEST_NOT_A_CP_2 -998
3979 /* grab a new slab and return the first slot in it */
3981 STATIC regmatch_state *
3984 regmatch_slab *s = PL_regmatch_slab->next;
3986 Newx(s, 1, regmatch_slab);
3987 s->prev = PL_regmatch_slab;
3989 PL_regmatch_slab->next = s;
3991 PL_regmatch_slab = s;
3992 return SLAB_FIRST(s);
3996 /* push a new state then goto it */
3998 #define PUSH_STATE_GOTO(state, node, input) \
3999 pushinput = input; \
4001 st->resume_state = state; \
4004 /* push a new state with success backtracking, then goto it */
4006 #define PUSH_YES_STATE_GOTO(state, node, input) \
4007 pushinput = input; \
4009 st->resume_state = state; \
4010 goto push_yes_state;
4017 regmatch() - main matching routine
4019 This is basically one big switch statement in a loop. We execute an op,
4020 set 'next' to point the next op, and continue. If we come to a point which
4021 we may need to backtrack to on failure such as (A|B|C), we push a
4022 backtrack state onto the backtrack stack. On failure, we pop the top
4023 state, and re-enter the loop at the state indicated. If there are no more
4024 states to pop, we return failure.
4026 Sometimes we also need to backtrack on success; for example /A+/, where
4027 after successfully matching one A, we need to go back and try to
4028 match another one; similarly for lookahead assertions: if the assertion
4029 completes successfully, we backtrack to the state just before the assertion
4030 and then carry on. In these cases, the pushed state is marked as
4031 'backtrack on success too'. This marking is in fact done by a chain of
4032 pointers, each pointing to the previous 'yes' state. On success, we pop to
4033 the nearest yes state, discarding any intermediate failure-only states.
4034 Sometimes a yes state is pushed just to force some cleanup code to be
4035 called at the end of a successful match or submatch; e.g. (??{$re}) uses
4036 it to free the inner regex.
4038 Note that failure backtracking rewinds the cursor position, while
4039 success backtracking leaves it alone.
4041 A pattern is complete when the END op is executed, while a subpattern
4042 such as (?=foo) is complete when the SUCCESS op is executed. Both of these
4043 ops trigger the "pop to last yes state if any, otherwise return true"
4046 A common convention in this function is to use A and B to refer to the two
4047 subpatterns (or to the first nodes thereof) in patterns like /A*B/: so A is
4048 the subpattern to be matched possibly multiple times, while B is the entire
4049 rest of the pattern. Variable and state names reflect this convention.
4051 The states in the main switch are the union of ops and failure/success of
4052 substates associated with with that op. For example, IFMATCH is the op
4053 that does lookahead assertions /(?=A)B/ and so the IFMATCH state means
4054 'execute IFMATCH'; while IFMATCH_A is a state saying that we have just
4055 successfully matched A and IFMATCH_A_fail is a state saying that we have
4056 just failed to match A. Resume states always come in pairs. The backtrack
4057 state we push is marked as 'IFMATCH_A', but when that is popped, we resume
4058 at IFMATCH_A or IFMATCH_A_fail, depending on whether we are backtracking
4059 on success or failure.
4061 The struct that holds a backtracking state is actually a big union, with
4062 one variant for each major type of op. The variable st points to the
4063 top-most backtrack struct. To make the code clearer, within each
4064 block of code we #define ST to alias the relevant union.
4066 Here's a concrete example of a (vastly oversimplified) IFMATCH
4072 #define ST st->u.ifmatch
4074 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
4075 ST.foo = ...; // some state we wish to save
4077 // push a yes backtrack state with a resume value of
4078 // IFMATCH_A/IFMATCH_A_fail, then continue execution at the
4080 PUSH_YES_STATE_GOTO(IFMATCH_A, A, newinput);
4083 case IFMATCH_A: // we have successfully executed A; now continue with B
4085 bar = ST.foo; // do something with the preserved value
4088 case IFMATCH_A_fail: // A failed, so the assertion failed
4089 ...; // do some housekeeping, then ...
4090 sayNO; // propagate the failure
4097 For any old-timers reading this who are familiar with the old recursive
4098 approach, the code above is equivalent to:
4100 case IFMATCH: // we are executing the IFMATCH op, (?=A)B
4109 ...; // do some housekeeping, then ...
4110 sayNO; // propagate the failure
4113 The topmost backtrack state, pointed to by st, is usually free. If you
4114 want to claim it, populate any ST.foo fields in it with values you wish to
4115 save, then do one of
4117 PUSH_STATE_GOTO(resume_state, node, newinput);
4118 PUSH_YES_STATE_GOTO(resume_state, node, newinput);
4120 which sets that backtrack state's resume value to 'resume_state', pushes a
4121 new free entry to the top of the backtrack stack, then goes to 'node'.
4122 On backtracking, the free slot is popped, and the saved state becomes the
4123 new free state. An ST.foo field in this new top state can be temporarily
4124 accessed to retrieve values, but once the main loop is re-entered, it
4125 becomes available for reuse.
4127 Note that the depth of the backtrack stack constantly increases during the
4128 left-to-right execution of the pattern, rather than going up and down with
4129 the pattern nesting. For example the stack is at its maximum at Z at the
4130 end of the pattern, rather than at X in the following:
4132 /(((X)+)+)+....(Y)+....Z/
4134 The only exceptions to this are lookahead/behind assertions and the cut,
4135 (?>A), which pop all the backtrack states associated with A before
4138 Backtrack state structs are allocated in slabs of about 4K in size.
4139 PL_regmatch_state and st always point to the currently active state,
4140 and PL_regmatch_slab points to the slab currently containing
4141 PL_regmatch_state. The first time regmatch() is called, the first slab is
4142 allocated, and is never freed until interpreter destruction. When the slab
4143 is full, a new one is allocated and chained to the end. At exit from
4144 regmatch(), slabs allocated since entry are freed.
4149 #define DEBUG_STATE_pp(pp) \
4151 DUMP_EXEC_POS(locinput, scan, utf8_target,depth); \
4152 Perl_re_printf( aTHX_ \
4153 "%*s" pp " %s%s%s%s%s\n", \
4154 INDENT_CHARS(depth), "", \
4155 PL_reg_name[st->resume_state], \
4156 ((st==yes_state||st==mark_state) ? "[" : ""), \
4157 ((st==yes_state) ? "Y" : ""), \
4158 ((st==mark_state) ? "M" : ""), \
4159 ((st==yes_state||st==mark_state) ? "]" : "") \
4164 #define REG_NODE_NUM(x) ((x) ? (int)((x)-prog) : -1)
4169 S_debug_start_match(pTHX_ const REGEXP *prog, const bool utf8_target,
4170 const char *start, const char *end, const char *blurb)
4172 const bool utf8_pat = RX_UTF8(prog) ? 1 : 0;
4174 PERL_ARGS_ASSERT_DEBUG_START_MATCH;
4179 RE_PV_QUOTED_DECL(s0, utf8_pat, PERL_DEBUG_PAD_ZERO(0),
4180 RX_PRECOMP_const(prog), RX_PRELEN(prog), PL_dump_re_max_len);
4182 RE_PV_QUOTED_DECL(s1, utf8_target, PERL_DEBUG_PAD_ZERO(1),
4183 start, end - start, PL_dump_re_max_len);
4185 Perl_re_printf( aTHX_
4186 "%s%s REx%s %s against %s\n",
4187 PL_colors[4], blurb, PL_colors[5], s0, s1);
4189 if (utf8_target||utf8_pat)
4190 Perl_re_printf( aTHX_ "UTF-8 %s%s%s...\n",
4191 utf8_pat ? "pattern" : "",
4192 utf8_pat && utf8_target ? " and " : "",
4193 utf8_target ? "string" : ""
4199 S_dump_exec_pos(pTHX_ const char *locinput,
4200 const regnode *scan,
4201 const char *loc_regeol,
4202 const char *loc_bostr,
4203 const char *loc_reg_starttry,
4204 const bool utf8_target,
4208 const int docolor = *PL_colors[0] || *PL_colors[2] || *PL_colors[4];
4209 const int taill = (docolor ? 10 : 7); /* 3 chars for "> <" */
4210 int l = (loc_regeol - locinput) > taill ? taill : (loc_regeol - locinput);
4211 /* The part of the string before starttry has one color
4212 (pref0_len chars), between starttry and current
4213 position another one (pref_len - pref0_len chars),
4214 after the current position the third one.
4215 We assume that pref0_len <= pref_len, otherwise we
4216 decrease pref0_len. */
4217 int pref_len = (locinput - loc_bostr) > (5 + taill) - l
4218 ? (5 + taill) - l : locinput - loc_bostr;
4221 PERL_ARGS_ASSERT_DUMP_EXEC_POS;
4223 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput - pref_len)))
4225 pref0_len = pref_len - (locinput - loc_reg_starttry);
4226 if (l + pref_len < (5 + taill) && l < loc_regeol - locinput)
4227 l = ( loc_regeol - locinput > (5 + taill) - pref_len
4228 ? (5 + taill) - pref_len : loc_regeol - locinput);
4229 while (utf8_target && UTF8_IS_CONTINUATION(*(U8*)(locinput + l)))
4233 if (pref0_len > pref_len)
4234 pref0_len = pref_len;
4236 const int is_uni = utf8_target ? 1 : 0;
4238 RE_PV_COLOR_DECL(s0,len0,is_uni,PERL_DEBUG_PAD(0),
4239 (locinput - pref_len),pref0_len, PL_dump_re_max_len, 4, 5);
4241 RE_PV_COLOR_DECL(s1,len1,is_uni,PERL_DEBUG_PAD(1),
4242 (locinput - pref_len + pref0_len),
4243 pref_len - pref0_len, PL_dump_re_max_len, 2, 3);
4245 RE_PV_COLOR_DECL(s2,len2,is_uni,PERL_DEBUG_PAD(2),
4246 locinput, loc_regeol - locinput, 10, 0, 1);
4248 const STRLEN tlen=len0+len1+len2;
4249 Perl_re_printf( aTHX_
4250 "%4" IVdf " <%.*s%.*s%s%.*s>%*s|%4u| ",
4251 (IV)(locinput - loc_bostr),
4254 (docolor ? "" : "> <"),
4256 (int)(tlen > 19 ? 0 : 19 - tlen),
4264 /* reg_check_named_buff_matched()
4265 * Checks to see if a named buffer has matched. The data array of
4266 * buffer numbers corresponding to the buffer is expected to reside
4267 * in the regexp->data->data array in the slot stored in the ARG() of
4268 * node involved. Note that this routine doesn't actually care about the
4269 * name, that information is not preserved from compilation to execution.
4270 * Returns the index of the leftmost defined buffer with the given name
4271 * or 0 if non of the buffers matched.
4274 S_reg_check_named_buff_matched(const regexp *rex, const regnode *scan)
4277 RXi_GET_DECL(rex,rexi);
4278 SV *sv_dat= MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
4279 I32 *nums=(I32*)SvPVX(sv_dat);
4281 PERL_ARGS_ASSERT_REG_CHECK_NAMED_BUFF_MATCHED;
4283 for ( n=0; n<SvIVX(sv_dat); n++ ) {
4284 if ((I32)rex->lastparen >= nums[n] &&
4285 rex->offs[nums[n]].end != -1)
4294 S_setup_EXACTISH_ST_c1_c2(pTHX_ const regnode * const text_node, int *c1p,
4295 U8* c1_utf8, int *c2p, U8* c2_utf8, regmatch_info *reginfo)
4297 /* This function determines if there are zero, one, two, or more characters
4298 * that match the first character of the passed-in EXACTish node
4299 * <text_node>, and if there are one or two, it returns them in the
4300 * passed-in pointers.
4302 * If it determines that no possible character in the target string can
4303 * match, it returns FALSE; otherwise TRUE. (The FALSE situation occurs if
4304 * the first character in <text_node> requires UTF-8 to represent, and the
4305 * target string isn't in UTF-8.)
4307 * If there are more than two characters that could match the beginning of
4308 * <text_node>, or if more context is required to determine a match or not,
4309 * it sets both *<c1p> and *<c2p> to CHRTEST_VOID.
4311 * The motiviation behind this function is to allow the caller to set up
4312 * tight loops for matching. If <text_node> is of type EXACT, there is
4313 * only one possible character that can match its first character, and so
4314 * the situation is quite simple. But things get much more complicated if
4315 * folding is involved. It may be that the first character of an EXACTFish
4316 * node doesn't participate in any possible fold, e.g., punctuation, so it
4317 * can be matched only by itself. The vast majority of characters that are
4318 * in folds match just two things, their lower and upper-case equivalents.
4319 * But not all are like that; some have multiple possible matches, or match
4320 * sequences of more than one character. This function sorts all that out.
4322 * Consider the patterns A*B or A*?B where A and B are arbitrary. In a
4323 * loop of trying to match A*, we know we can't exit where the thing
4324 * following it isn't a B. And something can't be a B unless it is the
4325 * beginning of B. By putting a quick test for that beginning in a tight
4326 * loop, we can rule out things that can't possibly be B without having to
4327 * break out of the loop, thus avoiding work. Similarly, if A is a single
4328 * character, we can make a tight loop matching A*, using the outputs of
4331 * If the target string to match isn't in UTF-8, and there aren't
4332 * complications which require CHRTEST_VOID, *<c1p> and *<c2p> are set to
4333 * the one or two possible octets (which are characters in this situation)
4334 * that can match. In all cases, if there is only one character that can
4335 * match, *<c1p> and *<c2p> will be identical.
4337 * If the target string is in UTF-8, the buffers pointed to by <c1_utf8>
4338 * and <c2_utf8> will contain the one or two UTF-8 sequences of bytes that
4339 * can match the beginning of <text_node>. They should be declared with at
4340 * least length UTF8_MAXBYTES+1. (If the target string isn't in UTF-8, it is
4341 * undefined what these contain.) If one or both of the buffers are
4342 * invariant under UTF-8, *<c1p>, and *<c2p> will also be set to the
4343 * corresponding invariant. If variant, the corresponding *<c1p> and/or
4344 * *<c2p> will be set to a negative number(s) that shouldn't match any code
4345 * point (unless inappropriately coerced to unsigned). *<c1p> will equal
4346 * *<c2p> if and only if <c1_utf8> and <c2_utf8> are the same. */
4348 const bool utf8_target = reginfo->is_utf8_target;
4350 UV c1 = (UV)CHRTEST_NOT_A_CP_1;
4351 UV c2 = (UV)CHRTEST_NOT_A_CP_2;
4352 bool use_chrtest_void = FALSE;
4353 const bool is_utf8_pat = reginfo->is_utf8_pat;
4355 /* Used when we have both utf8 input and utf8 output, to avoid converting
4356 * to/from code points */
4357 bool utf8_has_been_setup = FALSE;
4361 U8 *pat = (U8*)STRING(text_node);
4362 U8 folded[UTF8_MAX_FOLD_CHAR_EXPAND * UTF8_MAXBYTES_CASE + 1] = { '\0' };
4364 if ( OP(text_node) == EXACT
4365 || OP(text_node) == EXACT_ONLY8
4366 || OP(text_node) == EXACTL)
4369 /* In an exact node, only one thing can be matched, that first
4370 * character. If both the pat and the target are UTF-8, we can just
4371 * copy the input to the output, avoiding finding the code point of
4374 assert(OP(text_node) != EXACT_ONLY8);
4377 else if (utf8_target) {
4378 Copy(pat, c1_utf8, UTF8SKIP(pat), U8);
4379 Copy(pat, c2_utf8, UTF8SKIP(pat), U8);
4380 utf8_has_been_setup = TRUE;
4382 else if (OP(text_node) == EXACT_ONLY8) {
4383 return FALSE; /* Can only match UTF-8 target */
4386 c2 = c1 = valid_utf8_to_uvchr(pat, NULL);
4389 else { /* an EXACTFish node */
4390 U8 *pat_end = pat + STR_LEN(text_node);
4392 /* An EXACTFL node has at least some characters unfolded, because what
4393 * they match is not known until now. So, now is the time to fold
4394 * the first few of them, as many as are needed to determine 'c1' and
4395 * 'c2' later in the routine. If the pattern isn't UTF-8, we only need
4396 * to fold if in a UTF-8 locale, and then only the Sharp S; everything
4397 * else is 1-1 and isn't assumed to be folded. In a UTF-8 pattern, we
4398 * need to fold as many characters as a single character can fold to,
4399 * so that later we can check if the first ones are such a multi-char
4400 * fold. But, in such a pattern only locale-problematic characters
4401 * aren't folded, so we can skip this completely if the first character
4402 * in the node isn't one of the tricky ones */
4403 if (OP(text_node) == EXACTFL) {
4405 if (! is_utf8_pat) {
4406 if (IN_UTF8_CTYPE_LOCALE && *pat == LATIN_SMALL_LETTER_SHARP_S)
4408 folded[0] = folded[1] = 's';
4410 pat_end = folded + 2;
4413 else if (is_PROBLEMATIC_LOCALE_FOLDEDS_START_utf8(pat)) {
4418 for (i = 0; i < UTF8_MAX_FOLD_CHAR_EXPAND && s < pat_end; i++) {
4419 if (isASCII(*s) && LIKELY(! PL_in_utf8_turkic_locale)) {
4420 *(d++) = (U8) toFOLD_LC(*s);
4425 _toFOLD_utf8_flags(s,
4429 FOLD_FLAGS_FULL | FOLD_FLAGS_LOCALE);
4440 if ( ( is_utf8_pat && is_MULTI_CHAR_FOLD_utf8_safe(pat, pat_end))
4441 || (!is_utf8_pat && is_MULTI_CHAR_FOLD_latin1_safe(pat, pat_end)))
4443 /* Multi-character folds require more context to sort out. Also
4444 * PL_utf8_foldclosures used below doesn't handle them, so have to
4445 * be handled outside this routine */
4446 use_chrtest_void = TRUE;
4448 else { /* an EXACTFish node which doesn't begin with a multi-char fold */
4449 c1 = is_utf8_pat ? valid_utf8_to_uvchr(pat, NULL) : *pat;
4451 if ( UNLIKELY(PL_in_utf8_turkic_locale)
4452 && OP(text_node) == EXACTFL
4453 && UNLIKELY( c1 == 'i' || c1 == 'I'
4454 || c1 == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE
4455 || c1 == LATIN_SMALL_LETTER_DOTLESS_I))
4456 { /* Hard-coded Turkish locale rules for these 4 characters
4457 override normal rules */
4459 c2 = LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
4461 else if (c1 == 'I') {
4462 c2 = LATIN_SMALL_LETTER_DOTLESS_I;
4464 else if (c1 == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4467 else if (c1 == LATIN_SMALL_LETTER_DOTLESS_I) {
4471 else if (c1 > 255) {
4472 const unsigned int * remaining_folds;
4473 unsigned int first_fold;
4475 /* Look up what code points (besides c1) fold to c1; e.g.,
4476 * [ 'K', KELVIN_SIGN ] both fold to 'k'. */
4477 Size_t folds_count = _inverse_folds(c1, &first_fold,
4479 if (folds_count == 0) {
4480 c2 = c1; /* there is only a single character that could
4483 else if (folds_count != 1) {
4484 /* If there aren't exactly two folds to this (itself and
4485 * another), it is outside the scope of this function */
4486 use_chrtest_void = TRUE;
4488 else { /* There are two. We already have one, get the other */
4491 /* Folds that cross the 255/256 boundary are forbidden if
4492 * EXACTFL (and isnt a UTF8 locale), or EXACTFAA and one is
4493 * ASCIII. The only other match to c1 is c2, and since c1
4494 * is above 255, c2 better be as well under these
4495 * circumstances. If it isn't, it means the only legal
4496 * match of c1 is itself. */
4498 && ( ( OP(text_node) == EXACTFL
4499 && ! IN_UTF8_CTYPE_LOCALE)
4500 || (( OP(text_node) == EXACTFAA
4501 || OP(text_node) == EXACTFAA_NO_TRIE)
4502 && (isASCII(c1) || isASCII(c2)))))
4508 else /* Here, c1 is <= 255 */
4510 && HAS_NONLATIN1_FOLD_CLOSURE(c1)
4511 && ( ! (OP(text_node) == EXACTFL && ! IN_UTF8_CTYPE_LOCALE))
4512 && ( ( OP(text_node) != EXACTFAA
4513 && OP(text_node) != EXACTFAA_NO_TRIE)
4516 /* Here, there could be something above Latin1 in the target
4517 * which folds to this character in the pattern. All such
4518 * cases except LATIN SMALL LETTER Y WITH DIAERESIS have more
4519 * than two characters involved in their folds, so are outside
4520 * the scope of this function */
4521 if (UNLIKELY(c1 == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
4522 c2 = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
4525 use_chrtest_void = TRUE;
4528 else { /* Here nothing above Latin1 can fold to the pattern
4530 switch (OP(text_node)) {
4532 case EXACTFL: /* /l rules */
4533 c2 = PL_fold_locale[c1];
4536 case EXACTF: /* This node only generated for non-utf8
4538 assert(! is_utf8_pat);
4539 if (! utf8_target) { /* /d rules */
4544 /* /u rules for all these. This happens to work for
4545 * EXACTFAA as nothing in Latin1 folds to ASCII */
4546 case EXACTFAA_NO_TRIE: /* This node only generated for
4547 non-utf8 patterns */
4548 assert(! is_utf8_pat);
4553 c2 = PL_fold_latin1[c1];
4557 NOT_REACHED; /* NOTREACHED */
4560 Perl_croak(aTHX_ "panic: Unexpected op %u", OP(text_node));
4561 NOT_REACHED; /* NOTREACHED */
4567 /* Here have figured things out. Set up the returns */
4568 if (use_chrtest_void) {
4569 *c2p = *c1p = CHRTEST_VOID;
4571 else if (utf8_target) {
4572 if (! utf8_has_been_setup) { /* Don't have the utf8; must get it */
4573 uvchr_to_utf8(c1_utf8, c1);
4574 uvchr_to_utf8(c2_utf8, c2);
4577 /* Invariants are stored in both the utf8 and byte outputs; Use
4578 * negative numbers otherwise for the byte ones. Make sure that the
4579 * byte ones are the same iff the utf8 ones are the same */
4580 *c1p = (UTF8_IS_INVARIANT(*c1_utf8)) ? *c1_utf8 : CHRTEST_NOT_A_CP_1;
4581 *c2p = (UTF8_IS_INVARIANT(*c2_utf8))
4584 ? CHRTEST_NOT_A_CP_1
4585 : CHRTEST_NOT_A_CP_2;
4587 else if (c1 > 255) {
4588 if (c2 > 255) { /* both possibilities are above what a non-utf8 string
4593 *c1p = *c2p = c2; /* c2 is the only representable value */
4595 else { /* c1 is representable; see about c2 */
4597 *c2p = (c2 < 256) ? c2 : c1;
4604 S_isGCB(pTHX_ const GCB_enum before, const GCB_enum after, const U8 * const strbeg, const U8 * const curpos, const bool utf8_target)
4606 /* returns a boolean indicating if there is a Grapheme Cluster Boundary
4607 * between the inputs. See http://www.unicode.org/reports/tr29/. */
4609 PERL_ARGS_ASSERT_ISGCB;
4611 switch (GCB_table[before][after]) {
4618 case GCB_RI_then_RI:
4621 U8 * temp_pos = (U8 *) curpos;
4623 /* Do not break within emoji flag sequences. That is, do not
4624 * break between regional indicator (RI) symbols if there is an
4625 * odd number of RI characters before the break point.
4626 * GB12 sot (RI RI)* RI × RI
4627 * GB13 [^RI] (RI RI)* RI × RI */
4629 while (backup_one_GCB(strbeg,
4631 utf8_target) == GCB_Regional_Indicator)
4636 return RI_count % 2 != 1;
4639 case GCB_EX_then_EM:
4641 /* GB10 ( E_Base | E_Base_GAZ ) Extend* × E_Modifier */
4643 U8 * temp_pos = (U8 *) curpos;
4647 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4649 while (prev == GCB_Extend);
4651 return prev != GCB_E_Base && prev != GCB_E_Base_GAZ;
4654 case GCB_Maybe_Emoji_NonBreak:
4658 /* Do not break within emoji modifier sequences or emoji zwj sequences.
4659 GB11 \p{Extended_Pictographic} Extend* ZWJ × \p{Extended_Pictographic}
4661 U8 * temp_pos = (U8 *) curpos;
4665 prev = backup_one_GCB(strbeg, &temp_pos, utf8_target);
4667 while (prev == GCB_Extend);
4669 return prev != GCB_XPG_XX;
4677 Perl_re_printf( aTHX_ "Unhandled GCB pair: GCB_table[%d, %d] = %d\n",
4678 before, after, GCB_table[before][after]);
4685 S_backup_one_GCB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4690 PERL_ARGS_ASSERT_BACKUP_ONE_GCB;
4692 if (*curpos < strbeg) {
4697 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
4698 U8 * prev_prev_char_pos;
4700 if (! prev_char_pos) {
4704 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
4705 gcb = getGCB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
4706 *curpos = prev_char_pos;
4707 prev_char_pos = prev_prev_char_pos;
4710 *curpos = (U8 *) strbeg;
4715 if (*curpos - 2 < strbeg) {
4716 *curpos = (U8 *) strbeg;
4720 gcb = getGCB_VAL_CP(*(*curpos - 1));
4726 /* Combining marks attach to most classes that precede them, but this defines
4727 * the exceptions (from TR14) */
4728 #define LB_CM_ATTACHES_TO(prev) ( ! ( prev == LB_EDGE \
4729 || prev == LB_Mandatory_Break \
4730 || prev == LB_Carriage_Return \
4731 || prev == LB_Line_Feed \
4732 || prev == LB_Next_Line \
4733 || prev == LB_Space \
4734 || prev == LB_ZWSpace))
4737 S_isLB(pTHX_ LB_enum before,
4739 const U8 * const strbeg,
4740 const U8 * const curpos,
4741 const U8 * const strend,
4742 const bool utf8_target)
4744 U8 * temp_pos = (U8 *) curpos;
4745 LB_enum prev = before;
4747 /* Is the boundary between 'before' and 'after' line-breakable?
4748 * Most of this is just a table lookup of a generated table from Unicode
4749 * rules. But some rules require context to decide, and so have to be
4750 * implemented in code */
4752 PERL_ARGS_ASSERT_ISLB;
4754 /* Rule numbers in the comments below are as of Unicode 9.0 */
4758 switch (LB_table[before][after]) {
4763 case LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4766 case LB_SP_foo + LB_BREAKABLE:
4767 case LB_SP_foo + LB_NOBREAK:
4768 case LB_SP_foo + LB_NOBREAK_EVEN_WITH_SP_BETWEEN:
4770 /* When we have something following a SP, we have to look at the
4771 * context in order to know what to do.
4773 * SP SP should not reach here because LB7: Do not break before
4774 * spaces. (For two spaces in a row there is nothing that
4775 * overrides that) */
4776 assert(after != LB_Space);
4778 /* Here we have a space followed by a non-space. Mostly this is a
4779 * case of LB18: "Break after spaces". But there are complications
4780 * as the handling of spaces is somewhat tricky. They are in a
4781 * number of rules, which have to be applied in priority order, but
4782 * something earlier in the string can cause a rule to be skipped
4783 * and a lower priority rule invoked. A prime example is LB7 which
4784 * says don't break before a space. But rule LB8 (lower priority)
4785 * says that the first break opportunity after a ZW is after any
4786 * span of spaces immediately after it. If a ZW comes before a SP
4787 * in the input, rule LB8 applies, and not LB7. Other such rules
4788 * involve combining marks which are rules 9 and 10, but they may
4789 * override higher priority rules if they come earlier in the
4790 * string. Since we're doing random access into the middle of the
4791 * string, we have to look for rules that should get applied based
4792 * on both string position and priority. Combining marks do not
4793 * attach to either ZW nor SP, so we don't have to consider them
4796 * To check for LB8, we have to find the first non-space character
4797 * before this span of spaces */
4799 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4801 while (prev == LB_Space);
4803 /* LB8 Break before any character following a zero-width space,
4804 * even if one or more spaces intervene.
4806 * So if we have a ZW just before this span, and to get here this
4807 * is the final space in the span. */
4808 if (prev == LB_ZWSpace) {
4812 /* Here, not ZW SP+. There are several rules that have higher
4813 * priority than LB18 and can be resolved now, as they don't depend
4814 * on anything earlier in the string (except ZW, which we have
4815 * already handled). One of these rules is LB11 Do not break
4816 * before Word joiner, but we have specially encoded that in the
4817 * lookup table so it is caught by the single test below which
4818 * catches the other ones. */
4819 if (LB_table[LB_Space][after] - LB_SP_foo
4820 == LB_NOBREAK_EVEN_WITH_SP_BETWEEN)
4825 /* If we get here, we have to XXX consider combining marks. */
4826 if (prev == LB_Combining_Mark) {
4828 /* What happens with these depends on the character they
4831 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4833 while (prev == LB_Combining_Mark);
4835 /* Most times these attach to and inherit the characteristics
4836 * of that character, but not always, and when not, they are to
4837 * be treated as AL by rule LB10. */
4838 if (! LB_CM_ATTACHES_TO(prev)) {
4839 prev = LB_Alphabetic;
4843 /* Here, we have the character preceding the span of spaces all set
4844 * up. We follow LB18: "Break after spaces" unless the table shows
4845 * that is overriden */
4846 return LB_table[prev][after] != LB_NOBREAK_EVEN_WITH_SP_BETWEEN;
4850 /* We don't know how to treat the CM except by looking at the first
4851 * non-CM character preceding it. ZWJ is treated as CM */
4853 prev = backup_one_LB(strbeg, &temp_pos, utf8_target);
4855 while (prev == LB_Combining_Mark || prev == LB_ZWJ);
4857 /* Here, 'prev' is that first earlier non-CM character. If the CM
4858 * attatches to it, then it inherits the behavior of 'prev'. If it
4859 * doesn't attach, it is to be treated as an AL */
4860 if (! LB_CM_ATTACHES_TO(prev)) {
4861 prev = LB_Alphabetic;
4866 case LB_HY_or_BA_then_foo + LB_BREAKABLE:
4867 case LB_HY_or_BA_then_foo + LB_NOBREAK:
4869 /* LB21a Don't break after Hebrew + Hyphen.
4870 * HL (HY | BA) × */
4872 if (backup_one_LB(strbeg, &temp_pos, utf8_target)
4873 == LB_Hebrew_Letter)
4878 return LB_table[prev][after] - LB_HY_or_BA_then_foo == LB_BREAKABLE;
4880 case LB_PR_or_PO_then_OP_or_HY + LB_BREAKABLE:
4881 case LB_PR_or_PO_then_OP_or_HY + LB_NOBREAK:
4883 /* LB25a (PR | PO) × ( OP | HY )? NU */
4884 if (advance_one_LB(&temp_pos, strend, utf8_target) == LB_Numeric) {
4888 return LB_table[prev][after] - LB_PR_or_PO_then_OP_or_HY
4891 case LB_SY_or_IS_then_various + LB_BREAKABLE:
4892 case LB_SY_or_IS_then_various + LB_NOBREAK:
4894 /* LB25d NU (SY | IS)* × (NU | SY | IS | CL | CP ) */
4896 LB_enum temp = prev;
4898 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4900 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric);
4901 if (temp == LB_Numeric) {
4905 return LB_table[prev][after] - LB_SY_or_IS_then_various
4909 case LB_various_then_PO_or_PR + LB_BREAKABLE:
4910 case LB_various_then_PO_or_PR + LB_NOBREAK:
4912 /* LB25e NU (SY | IS)* (CL | CP)? × (PO | PR) */
4914 LB_enum temp = prev;
4915 if (temp == LB_Close_Punctuation || temp == LB_Close_Parenthesis)
4917 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4919 while (temp == LB_Break_Symbols || temp == LB_Infix_Numeric) {
4920 temp = backup_one_LB(strbeg, &temp_pos, utf8_target);
4922 if (temp == LB_Numeric) {
4925 return LB_various_then_PO_or_PR;
4928 case LB_RI_then_RI + LB_NOBREAK:
4929 case LB_RI_then_RI + LB_BREAKABLE:
4933 /* LB30a Break between two regional indicator symbols if and
4934 * only if there are an even number of regional indicators
4935 * preceding the position of the break.
4937 * sot (RI RI)* RI × RI
4938 * [^RI] (RI RI)* RI × RI */
4940 while (backup_one_LB(strbeg,
4942 utf8_target) == LB_Regional_Indicator)
4947 return RI_count % 2 == 0;
4955 Perl_re_printf( aTHX_ "Unhandled LB pair: LB_table[%d, %d] = %d\n",
4956 before, after, LB_table[before][after]);
4963 S_advance_one_LB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
4969 PERL_ARGS_ASSERT_ADVANCE_ONE_LB;
4971 if (*curpos >= strend) {
4976 *curpos += UTF8SKIP(*curpos);
4977 if (*curpos >= strend) {
4980 lb = getLB_VAL_UTF8(*curpos, strend);
4984 if (*curpos >= strend) {
4987 lb = getLB_VAL_CP(**curpos);
4994 S_backup_one_LB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
4999 PERL_ARGS_ASSERT_BACKUP_ONE_LB;
5001 if (*curpos < strbeg) {
5006 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5007 U8 * prev_prev_char_pos;
5009 if (! prev_char_pos) {
5013 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1, strbeg))) {
5014 lb = getLB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5015 *curpos = prev_char_pos;
5016 prev_char_pos = prev_prev_char_pos;
5019 *curpos = (U8 *) strbeg;
5024 if (*curpos - 2 < strbeg) {
5025 *curpos = (U8 *) strbeg;
5029 lb = getLB_VAL_CP(*(*curpos - 1));
5036 S_isSB(pTHX_ SB_enum before,
5038 const U8 * const strbeg,
5039 const U8 * const curpos,
5040 const U8 * const strend,
5041 const bool utf8_target)
5043 /* returns a boolean indicating if there is a Sentence Boundary Break
5044 * between the inputs. See http://www.unicode.org/reports/tr29/ */
5046 U8 * lpos = (U8 *) curpos;
5047 bool has_para_sep = FALSE;
5048 bool has_sp = FALSE;
5050 PERL_ARGS_ASSERT_ISSB;
5052 /* Break at the start and end of text.
5055 But unstated in Unicode is don't break if the text is empty */
5056 if (before == SB_EDGE || after == SB_EDGE) {
5057 return before != after;
5060 /* SB 3: Do not break within CRLF. */
5061 if (before == SB_CR && after == SB_LF) {
5065 /* Break after paragraph separators. CR and LF are considered
5066 * so because Unicode views text as like word processing text where there
5067 * are no newlines except between paragraphs, and the word processor takes
5068 * care of wrapping without there being hard line-breaks in the text *./
5069 SB4. Sep | CR | LF ÷ */
5070 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
5074 /* Ignore Format and Extend characters, except after sot, Sep, CR, or LF.
5075 * (See Section 6.2, Replacing Ignore Rules.)
5076 SB5. X (Extend | Format)* → X */
5077 if (after == SB_Extend || after == SB_Format) {
5079 /* Implied is that the these characters attach to everything
5080 * immediately prior to them except for those separator-type
5081 * characters. And the rules earlier have already handled the case
5082 * when one of those immediately precedes the extend char */
5086 if (before == SB_Extend || before == SB_Format) {
5087 U8 * temp_pos = lpos;
5088 const SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
5089 if ( backup != SB_EDGE
5098 /* Here, both 'before' and 'backup' are these types; implied is that we
5099 * don't break between them */
5100 if (backup == SB_Extend || backup == SB_Format) {
5105 /* Do not break after ambiguous terminators like period, if they are
5106 * immediately followed by a number or lowercase letter, if they are
5107 * between uppercase letters, if the first following letter (optionally
5108 * after certain punctuation) is lowercase, or if they are followed by
5109 * "continuation" punctuation such as comma, colon, or semicolon. For
5110 * example, a period may be an abbreviation or numeric period, and thus may
5111 * not mark the end of a sentence.
5113 * SB6. ATerm × Numeric */
5114 if (before == SB_ATerm && after == SB_Numeric) {
5118 /* SB7. (Upper | Lower) ATerm × Upper */
5119 if (before == SB_ATerm && after == SB_Upper) {
5120 U8 * temp_pos = lpos;
5121 SB_enum backup = backup_one_SB(strbeg, &temp_pos, utf8_target);
5122 if (backup == SB_Upper || backup == SB_Lower) {
5127 /* The remaining rules that aren't the final one, all require an STerm or
5128 * an ATerm after having backed up over some Close* Sp*, and in one case an
5129 * optional Paragraph separator, although one rule doesn't have any Sp's in it.
5130 * So do that backup now, setting flags if either Sp or a paragraph
5131 * separator are found */
5133 if (before == SB_Sep || before == SB_CR || before == SB_LF) {
5134 has_para_sep = TRUE;
5135 before = backup_one_SB(strbeg, &lpos, utf8_target);
5138 if (before == SB_Sp) {
5141 before = backup_one_SB(strbeg, &lpos, utf8_target);
5143 while (before == SB_Sp);
5146 while (before == SB_Close) {
5147 before = backup_one_SB(strbeg, &lpos, utf8_target);
5150 /* The next few rules apply only when the backed-up-to is an ATerm, and in
5151 * most cases an STerm */
5152 if (before == SB_STerm || before == SB_ATerm) {
5154 /* So, here the lhs matches
5155 * (STerm | ATerm) Close* Sp* (Sep | CR | LF)?
5156 * and we have set flags if we found an Sp, or the optional Sep,CR,LF.
5157 * The rules that apply here are:
5159 * SB8 ATerm Close* Sp* × ( ¬(OLetter | Upper | Lower | Sep | CR
5160 | LF | STerm | ATerm) )* Lower
5161 SB8a (STerm | ATerm) Close* Sp* × (SContinue | STerm | ATerm)
5162 SB9 (STerm | ATerm) Close* × (Close | Sp | Sep | CR | LF)
5163 SB10 (STerm | ATerm) Close* Sp* × (Sp | Sep | CR | LF)
5164 SB11 (STerm | ATerm) Close* Sp* (Sep | CR | LF)? ÷
5167 /* And all but SB11 forbid having seen a paragraph separator */
5168 if (! has_para_sep) {
5169 if (before == SB_ATerm) { /* SB8 */
5170 U8 * rpos = (U8 *) curpos;
5171 SB_enum later = after;
5173 while ( later != SB_OLetter
5174 && later != SB_Upper
5175 && later != SB_Lower
5179 && later != SB_STerm
5180 && later != SB_ATerm
5181 && later != SB_EDGE)
5183 later = advance_one_SB(&rpos, strend, utf8_target);
5185 if (later == SB_Lower) {
5190 if ( after == SB_SContinue /* SB8a */
5191 || after == SB_STerm
5192 || after == SB_ATerm)
5197 if (! has_sp) { /* SB9 applies only if there was no Sp* */
5198 if ( after == SB_Close
5208 /* SB10. This and SB9 could probably be combined some way, but khw
5209 * has decided to follow the Unicode rule book precisely for
5210 * simplified maintenance */
5224 /* Otherwise, do not break.
5231 S_advance_one_SB(pTHX_ U8 ** curpos, const U8 * const strend, const bool utf8_target)
5236 PERL_ARGS_ASSERT_ADVANCE_ONE_SB;
5238 if (*curpos >= strend) {
5244 *curpos += UTF8SKIP(*curpos);
5245 if (*curpos >= strend) {
5248 sb = getSB_VAL_UTF8(*curpos, strend);
5249 } while (sb == SB_Extend || sb == SB_Format);
5254 if (*curpos >= strend) {
5257 sb = getSB_VAL_CP(**curpos);
5258 } while (sb == SB_Extend || sb == SB_Format);
5265 S_backup_one_SB(pTHX_ const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5270 PERL_ARGS_ASSERT_BACKUP_ONE_SB;
5272 if (*curpos < strbeg) {
5277 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5278 if (! prev_char_pos) {
5282 /* Back up over Extend and Format. curpos is always just to the right
5283 * of the characater whose value we are getting */
5285 U8 * prev_prev_char_pos;
5286 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos, -1,
5289 sb = getSB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5290 *curpos = prev_char_pos;
5291 prev_char_pos = prev_prev_char_pos;
5294 *curpos = (U8 *) strbeg;
5297 } while (sb == SB_Extend || sb == SB_Format);
5301 if (*curpos - 2 < strbeg) {
5302 *curpos = (U8 *) strbeg;
5306 sb = getSB_VAL_CP(*(*curpos - 1));
5307 } while (sb == SB_Extend || sb == SB_Format);
5314 S_isWB(pTHX_ WB_enum previous,
5317 const U8 * const strbeg,
5318 const U8 * const curpos,
5319 const U8 * const strend,
5320 const bool utf8_target)
5322 /* Return a boolean as to if the boundary between 'before' and 'after' is
5323 * a Unicode word break, using their published algorithm, but tailored for
5324 * Perl by treating spans of white space as one unit. Context may be
5325 * needed to make this determination. If the value for the character
5326 * before 'before' is known, it is passed as 'previous'; otherwise that
5327 * should be set to WB_UNKNOWN. The other input parameters give the
5328 * boundaries and current position in the matching of the string. That
5329 * is, 'curpos' marks the position where the character whose wb value is
5330 * 'after' begins. See http://www.unicode.org/reports/tr29/ */
5332 U8 * before_pos = (U8 *) curpos;
5333 U8 * after_pos = (U8 *) curpos;
5334 WB_enum prev = before;
5337 PERL_ARGS_ASSERT_ISWB;
5339 /* Rule numbers in the comments below are as of Unicode 9.0 */
5343 switch (WB_table[before][after]) {
5350 case WB_hs_then_hs: /* 2 horizontal spaces in a row */
5351 next = advance_one_WB(&after_pos, strend, utf8_target,
5352 FALSE /* Don't skip Extend nor Format */ );
5353 /* A space immediately preceeding an Extend or Format is attached
5354 * to by them, and hence gets separated from previous spaces.
5355 * Otherwise don't break between horizontal white space */
5356 return next == WB_Extend || next == WB_Format;
5358 /* WB4 Ignore Format and Extend characters, except when they appear at
5359 * the beginning of a region of text. This code currently isn't
5360 * general purpose, but it works as the rules are currently and likely
5361 * to be laid out. The reason it works is that when 'they appear at
5362 * the beginning of a region of text', the rule is to break before
5363 * them, just like any other character. Therefore, the default rule
5364 * applies and we don't have to look in more depth. Should this ever
5365 * change, we would have to have 2 'case' statements, like in the rules
5366 * below, and backup a single character (not spacing over the extend
5367 * ones) and then see if that is one of the region-end characters and
5369 case WB_Ex_or_FO_or_ZWJ_then_foo:
5370 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5373 case WB_DQ_then_HL + WB_BREAKABLE:
5374 case WB_DQ_then_HL + WB_NOBREAK:
5376 /* WB7c Hebrew_Letter Double_Quote × Hebrew_Letter */
5378 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5379 == WB_Hebrew_Letter)
5384 return WB_table[before][after] - WB_DQ_then_HL == WB_BREAKABLE;
5386 case WB_HL_then_DQ + WB_BREAKABLE:
5387 case WB_HL_then_DQ + WB_NOBREAK:
5389 /* WB7b Hebrew_Letter × Double_Quote Hebrew_Letter */
5391 if (advance_one_WB(&after_pos, strend, utf8_target,
5392 TRUE /* Do skip Extend and Format */ )
5393 == WB_Hebrew_Letter)
5398 return WB_table[before][after] - WB_HL_then_DQ == WB_BREAKABLE;
5400 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_NOBREAK:
5401 case WB_LE_or_HL_then_MB_or_ML_or_SQ + WB_BREAKABLE:
5403 /* WB6 (ALetter | Hebrew_Letter) × (MidLetter | MidNumLet
5404 * | Single_Quote) (ALetter | Hebrew_Letter) */
5406 next = advance_one_WB(&after_pos, strend, utf8_target,
5407 TRUE /* Do skip Extend and Format */ );
5409 if (next == WB_ALetter || next == WB_Hebrew_Letter)
5414 return WB_table[before][after]
5415 - WB_LE_or_HL_then_MB_or_ML_or_SQ == WB_BREAKABLE;
5417 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_NOBREAK:
5418 case WB_MB_or_ML_or_SQ_then_LE_or_HL + WB_BREAKABLE:
5420 /* WB7 (ALetter | Hebrew_Letter) (MidLetter | MidNumLet
5421 * | Single_Quote) × (ALetter | Hebrew_Letter) */
5423 prev = backup_one_WB(&previous, strbeg, &before_pos, utf8_target);
5424 if (prev == WB_ALetter || prev == WB_Hebrew_Letter)
5429 return WB_table[before][after]
5430 - WB_MB_or_ML_or_SQ_then_LE_or_HL == WB_BREAKABLE;
5432 case WB_MB_or_MN_or_SQ_then_NU + WB_NOBREAK:
5433 case WB_MB_or_MN_or_SQ_then_NU + WB_BREAKABLE:
5435 /* WB11 Numeric (MidNum | (MidNumLet | Single_Quote)) × Numeric
5438 if (backup_one_WB(&previous, strbeg, &before_pos, utf8_target)
5444 return WB_table[before][after]
5445 - WB_MB_or_MN_or_SQ_then_NU == WB_BREAKABLE;
5447 case WB_NU_then_MB_or_MN_or_SQ + WB_NOBREAK:
5448 case WB_NU_then_MB_or_MN_or_SQ + WB_BREAKABLE:
5450 /* WB12 Numeric × (MidNum | MidNumLet | Single_Quote) Numeric */
5452 if (advance_one_WB(&after_pos, strend, utf8_target,
5453 TRUE /* Do skip Extend and Format */ )
5459 return WB_table[before][after]
5460 - WB_NU_then_MB_or_MN_or_SQ == WB_BREAKABLE;
5462 case WB_RI_then_RI + WB_NOBREAK:
5463 case WB_RI_then_RI + WB_BREAKABLE:
5467 /* Do not break within emoji flag sequences. That is, do not
5468 * break between regional indicator (RI) symbols if there is an
5469 * odd number of RI characters before the potential break
5472 * WB15 sot (RI RI)* RI × RI
5473 * WB16 [^RI] (RI RI)* RI × RI */
5475 while (backup_one_WB(&previous,
5478 utf8_target) == WB_Regional_Indicator)
5483 return RI_count % 2 != 1;
5491 Perl_re_printf( aTHX_ "Unhandled WB pair: WB_table[%d, %d] = %d\n",
5492 before, after, WB_table[before][after]);
5499 S_advance_one_WB(pTHX_ U8 ** curpos,
5500 const U8 * const strend,
5501 const bool utf8_target,
5502 const bool skip_Extend_Format)
5507 PERL_ARGS_ASSERT_ADVANCE_ONE_WB;
5509 if (*curpos >= strend) {
5515 /* Advance over Extend and Format */
5517 *curpos += UTF8SKIP(*curpos);
5518 if (*curpos >= strend) {
5521 wb = getWB_VAL_UTF8(*curpos, strend);
5522 } while ( skip_Extend_Format
5523 && (wb == WB_Extend || wb == WB_Format));
5528 if (*curpos >= strend) {
5531 wb = getWB_VAL_CP(**curpos);
5532 } while ( skip_Extend_Format
5533 && (wb == WB_Extend || wb == WB_Format));
5540 S_backup_one_WB(pTHX_ WB_enum * previous, const U8 * const strbeg, U8 ** curpos, const bool utf8_target)
5545 PERL_ARGS_ASSERT_BACKUP_ONE_WB;
5547 /* If we know what the previous character's break value is, don't have
5549 if (*previous != WB_UNKNOWN) {
5552 /* But we need to move backwards by one */
5554 *curpos = reghopmaybe3(*curpos, -1, strbeg);
5556 *previous = WB_EDGE;
5557 *curpos = (U8 *) strbeg;
5560 *previous = WB_UNKNOWN;
5565 *previous = (*curpos <= strbeg) ? WB_EDGE : WB_UNKNOWN;
5568 /* And we always back up over these three types */
5569 if (wb != WB_Extend && wb != WB_Format && wb != WB_ZWJ) {
5574 if (*curpos < strbeg) {
5579 U8 * prev_char_pos = reghopmaybe3(*curpos, -1, strbeg);
5580 if (! prev_char_pos) {
5584 /* Back up over Extend and Format. curpos is always just to the right
5585 * of the characater whose value we are getting */
5587 U8 * prev_prev_char_pos;
5588 if ((prev_prev_char_pos = reghopmaybe3((U8 *) prev_char_pos,
5592 wb = getWB_VAL_UTF8(prev_prev_char_pos, prev_char_pos);
5593 *curpos = prev_char_pos;
5594 prev_char_pos = prev_prev_char_pos;
5597 *curpos = (U8 *) strbeg;
5600 } while (wb == WB_Extend || wb == WB_Format || wb == WB_ZWJ);
5604 if (*curpos - 2 < strbeg) {
5605 *curpos = (U8 *) strbeg;
5609 wb = getWB_VAL_CP(*(*curpos - 1));
5610 } while (wb == WB_Extend || wb == WB_Format);
5616 #define EVAL_CLOSE_PAREN_IS(st,expr) \
5619 ( ( st )->u.eval.close_paren ) && \
5620 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5623 #define EVAL_CLOSE_PAREN_IS_TRUE(st,expr) \
5626 ( ( st )->u.eval.close_paren ) && \
5628 ( ( ( st )->u.eval.close_paren ) == ( (expr) + 1 ) ) \
5632 #define EVAL_CLOSE_PAREN_SET(st,expr) \
5633 (st)->u.eval.close_paren = ( (expr) + 1 )
5635 #define EVAL_CLOSE_PAREN_CLEAR(st) \
5636 (st)->u.eval.close_paren = 0
5638 /* returns -1 on failure, $+[0] on success */
5640 S_regmatch(pTHX_ regmatch_info *reginfo, char *startpos, regnode *prog)
5643 const bool utf8_target = reginfo->is_utf8_target;
5644 const U32 uniflags = UTF8_ALLOW_DEFAULT;
5645 REGEXP *rex_sv = reginfo->prog;
5646 regexp *rex = ReANY(rex_sv);
5647 RXi_GET_DECL(rex,rexi);
5648 /* the current state. This is a cached copy of PL_regmatch_state */
5650 /* cache heavy used fields of st in registers */
5653 U32 n = 0; /* general value; init to avoid compiler warning */
5654 SSize_t ln = 0; /* len or last; init to avoid compiler warning */
5655 SSize_t endref = 0; /* offset of end of backref when ln is start */
5656 char *locinput = startpos;
5657 char *pushinput; /* where to continue after a PUSH */
5658 I32 nextchr; /* is always set to UCHARAT(locinput), or -1 at EOS */
5660 bool result = 0; /* return value of S_regmatch */
5661 U32 depth = 0; /* depth of backtrack stack */
5662 U32 nochange_depth = 0; /* depth of GOSUB recursion with nochange */
5663 const U32 max_nochange_depth =
5664 (3 * rex->nparens > MAX_RECURSE_EVAL_NOCHANGE_DEPTH) ?
5665 3 * rex->nparens : MAX_RECURSE_EVAL_NOCHANGE_DEPTH;
5666 regmatch_state *yes_state = NULL; /* state to pop to on success of
5668 /* mark_state piggy backs on the yes_state logic so that when we unwind
5669 the stack on success we can update the mark_state as we go */
5670 regmatch_state *mark_state = NULL; /* last mark state we have seen */
5671 regmatch_state *cur_eval = NULL; /* most recent EVAL_AB state */
5672 struct regmatch_state *cur_curlyx = NULL; /* most recent curlyx */
5674 bool no_final = 0; /* prevent failure from backtracking? */
5675 bool do_cutgroup = 0; /* no_final only until next branch/trie entry */
5676 char *startpoint = locinput;
5677 SV *popmark = NULL; /* are we looking for a mark? */
5678 SV *sv_commit = NULL; /* last mark name seen in failure */
5679 SV *sv_yes_mark = NULL; /* last mark name we have seen
5680 during a successful match */
5681 U32 lastopen = 0; /* last open we saw */
5682 bool has_cutgroup = RXp_HAS_CUTGROUP(rex) ? 1 : 0;
5683 SV* const oreplsv = GvSVn(PL_replgv);
5684 /* these three flags are set by various ops to signal information to
5685 * the very next op. They have a useful lifetime of exactly one loop
5686 * iteration, and are not preserved or restored by state pushes/pops
5688 bool sw = 0; /* the condition value in (?(cond)a|b) */
5689 bool minmod = 0; /* the next "{n,m}" is a "{n,m}?" */
5690 int logical = 0; /* the following EVAL is:
5694 or the following IFMATCH/UNLESSM is:
5695 false: plain (?=foo)
5696 true: used as a condition: (?(?=foo))
5698 PAD* last_pad = NULL;
5700 U8 gimme = G_SCALAR;
5701 CV *caller_cv = NULL; /* who called us */
5702 CV *last_pushed_cv = NULL; /* most recently called (?{}) CV */
5703 U32 maxopenparen = 0; /* max '(' index seen so far */
5704 int to_complement; /* Invert the result? */
5705 _char_class_number classnum;
5706 bool is_utf8_pat = reginfo->is_utf8_pat;
5708 I32 orig_savestack_ix = PL_savestack_ix;
5709 U8 * script_run_begin = NULL;
5711 /* Solaris Studio 12.3 messes up fetching PL_charclass['\n'] */
5712 #if (defined(__SUNPRO_C) && (__SUNPRO_C == 0x5120) && defined(__x86_64) && defined(USE_64_BIT_ALL))
5713 # define SOLARIS_BAD_OPTIMIZER
5714 const U32 *pl_charclass_dup = PL_charclass;
5715 # define PL_charclass pl_charclass_dup
5719 GET_RE_DEBUG_FLAGS_DECL;
5722 /* protect against undef(*^R) */
5723 SAVEFREESV(SvREFCNT_inc_simple_NN(oreplsv));
5725 /* shut up 'may be used uninitialized' compiler warnings for dMULTICALL */
5726 multicall_oldcatch = 0;
5727 PERL_UNUSED_VAR(multicall_cop);
5729 PERL_ARGS_ASSERT_REGMATCH;
5731 st = PL_regmatch_state;
5733 /* Note that nextchr is a byte even in UTF */
5737 DEBUG_OPTIMISE_r( DEBUG_EXECUTE_r({
5738 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5739 Perl_re_printf( aTHX_ "regmatch start\n" );
5742 while (scan != NULL) {
5743 next = scan + NEXT_OFF(scan);
5746 state_num = OP(scan);
5750 if (state_num <= REGNODE_MAX) {
5751 SV * const prop = sv_newmortal();
5752 regnode *rnext = regnext(scan);
5754 DUMP_EXEC_POS( locinput, scan, utf8_target, depth );
5755 regprop(rex, prop, scan, reginfo, NULL);
5756 Perl_re_printf( aTHX_
5757 "%*s%" IVdf ":%s(%" IVdf ")\n",
5758 INDENT_CHARS(depth), "",
5759 (IV)(scan - rexi->program),
5761 (PL_regkind[OP(scan)] == END || !rnext) ?
5762 0 : (IV)(rnext - rexi->program));
5769 assert(nextchr < 256 && (nextchr >= 0 || nextchr == NEXTCHR_EOS));
5771 switch (state_num) {
5772 case SBOL: /* /^../ and /\A../ */
5773 if (locinput == reginfo->strbeg)
5777 case MBOL: /* /^../m */
5778 if (locinput == reginfo->strbeg ||
5779 (!NEXTCHR_IS_EOS && locinput[-1] == '\n'))
5786 if (locinput == reginfo->ganch)
5790 case KEEPS: /* \K */
5791 /* update the startpoint */
5792 st->u.keeper.val = rex->offs[0].start;
5793 rex->offs[0].start = locinput - reginfo->strbeg;
5794 PUSH_STATE_GOTO(KEEPS_next, next, locinput);
5795 NOT_REACHED; /* NOTREACHED */
5797 case KEEPS_next_fail:
5798 /* rollback the start point change */
5799 rex->offs[0].start = st->u.keeper.val;
5801 NOT_REACHED; /* NOTREACHED */
5803 case MEOL: /* /..$/m */
5804 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5808 case SEOL: /* /..$/ */
5809 if (!NEXTCHR_IS_EOS && nextchr != '\n')
5811 if (reginfo->strend - locinput > 1)
5816 if (!NEXTCHR_IS_EOS)
5820 case SANY: /* /./s */
5823 goto increment_locinput;
5825 case REG_ANY: /* /./ */
5826 if ((NEXTCHR_IS_EOS) || nextchr == '\n')
5828 goto increment_locinput;
5832 #define ST st->u.trie
5833 case TRIEC: /* (ab|cd) with known charclass */
5834 /* In this case the charclass data is available inline so
5835 we can fail fast without a lot of extra overhead.
5837 if(!NEXTCHR_IS_EOS && !ANYOF_BITMAP_TEST(scan, nextchr)) {
5839 Perl_re_exec_indentf( aTHX_ "%sTRIE: failed to match trie start class...%s\n",
5840 depth, PL_colors[4], PL_colors[5])
5843 NOT_REACHED; /* NOTREACHED */
5846 case TRIE: /* (ab|cd) */
5847 /* the basic plan of execution of the trie is:
5848 * At the beginning, run though all the states, and
5849 * find the longest-matching word. Also remember the position
5850 * of the shortest matching word. For example, this pattern:
5853 * when matched against the string "abcde", will generate
5854 * accept states for all words except 3, with the longest
5855 * matching word being 4, and the shortest being 2 (with
5856 * the position being after char 1 of the string).
5858 * Then for each matching word, in word order (i.e. 1,2,4,5),
5859 * we run the remainder of the pattern; on each try setting
5860 * the current position to the character following the word,
5861 * returning to try the next word on failure.
5863 * We avoid having to build a list of words at runtime by
5864 * using a compile-time structure, wordinfo[].prev, which
5865 * gives, for each word, the previous accepting word (if any).
5866 * In the case above it would contain the mappings 1->2, 2->0,
5867 * 3->0, 4->5, 5->1. We can use this table to generate, from
5868 * the longest word (4 above), a list of all words, by
5869 * following the list of prev pointers; this gives us the
5870 * unordered list 4,5,1,2. Then given the current word we have
5871 * just tried, we can go through the list and find the
5872 * next-biggest word to try (so if we just failed on word 2,
5873 * the next in the list is 4).
5875 * Since at runtime we don't record the matching position in
5876 * the string for each word, we have to work that out for
5877 * each word we're about to process. The wordinfo table holds
5878 * the character length of each word; given that we recorded
5879 * at the start: the position of the shortest word and its
5880 * length in chars, we just need to move the pointer the
5881 * difference between the two char lengths. Depending on
5882 * Unicode status and folding, that's cheap or expensive.
5884 * This algorithm is optimised for the case where are only a
5885 * small number of accept states, i.e. 0,1, or maybe 2.
5886 * With lots of accepts states, and having to try all of them,
5887 * it becomes quadratic on number of accept states to find all
5892 /* what type of TRIE am I? (utf8 makes this contextual) */
5893 DECL_TRIE_TYPE(scan);
5895 /* what trie are we using right now */
5896 reg_trie_data * const trie
5897 = (reg_trie_data*)rexi->data->data[ ARG( scan ) ];
5898 HV * widecharmap = MUTABLE_HV(rexi->data->data[ ARG( scan ) + 1 ]);
5899 U32 state = trie->startstate;
5901 if (scan->flags == EXACTL || scan->flags == EXACTFLU8) {
5902 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
5905 && UTF8_IS_ABOVE_LATIN1(nextchr)
5906 && scan->flags == EXACTL)
5908 /* We only output for EXACTL, as we let the folder
5909 * output this message for EXACTFLU8 to avoid
5911 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
5916 && (NEXTCHR_IS_EOS || !TRIE_BITMAP_TEST(trie, nextchr)))
5918 if (trie->states[ state ].wordnum) {
5920 Perl_re_exec_indentf( aTHX_ "%sTRIE: matched empty string...%s\n",
5921 depth, PL_colors[4], PL_colors[5])
5927 Perl_re_exec_indentf( aTHX_ "%sTRIE: failed to match trie start class...%s\n",
5928 depth, PL_colors[4], PL_colors[5])
5935 U8 *uc = ( U8* )locinput;
5939 U8 *uscan = (U8*)NULL;
5940 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
5941 U32 charcount = 0; /* how many input chars we have matched */
5942 U32 accepted = 0; /* have we seen any accepting states? */
5944 ST.jump = trie->jump;
5947 ST.longfold = FALSE; /* char longer if folded => it's harder */
5950 /* fully traverse the TRIE; note the position of the
5951 shortest accept state and the wordnum of the longest
5954 while ( state && uc <= (U8*)(reginfo->strend) ) {
5955 U32 base = trie->states[ state ].trans.base;
5959 wordnum = trie->states[ state ].wordnum;
5961 if (wordnum) { /* it's an accept state */
5964 /* record first match position */
5966 ST.firstpos = (U8*)locinput;
5971 ST.firstchars = charcount;
5974 if (!ST.nextword || wordnum < ST.nextword)
5975 ST.nextword = wordnum;
5976 ST.topword = wordnum;
5979 DEBUG_TRIE_EXECUTE_r({
5980 DUMP_EXEC_POS( (char *)uc, scan, utf8_target, depth );
5982 PerlIO_printf( Perl_debug_log,
5983 "%*s%sTRIE: State: %4" UVxf " Accepted: %c ",
5984 INDENT_CHARS(depth), "", PL_colors[4],
5985 (UV)state, (accepted ? 'Y' : 'N'));
5988 /* read a char and goto next state */
5989 if ( base && (foldlen || uc < (U8*)(reginfo->strend))) {
5991 REXEC_TRIE_READ_CHAR(trie_type, trie, widecharmap, uc,
5992 (U8 *) reginfo->strend, uscan,
5993 len, uvc, charid, foldlen,
6000 base + charid - 1 - trie->uniquecharcount)) >= 0)
6002 && ((U32)offset < trie->lasttrans)
6003 && trie->trans[offset].check == state)
6005 state = trie->trans[offset].next;
6016 DEBUG_TRIE_EXECUTE_r(
6017 Perl_re_printf( aTHX_
6018 "TRIE: Charid:%3x CP:%4" UVxf " After State: %4" UVxf "%s\n",
6019 charid, uvc, (UV)state, PL_colors[5] );
6025 /* calculate total number of accept states */
6030 w = trie->wordinfo[w].prev;
6033 ST.accepted = accepted;
6037 Perl_re_exec_indentf( aTHX_ "%sTRIE: got %" IVdf " possible matches%s\n",
6039 PL_colors[4], (IV)ST.accepted, PL_colors[5] );
6041 goto trie_first_try; /* jump into the fail handler */
6043 NOT_REACHED; /* NOTREACHED */
6045 case TRIE_next_fail: /* we failed - try next alternative */
6049 /* undo any captures done in the tail part of a branch,
6051 * /(?:X(.)(.)|Y(.)).../
6052 * where the trie just matches X then calls out to do the
6053 * rest of the branch */
6054 REGCP_UNWIND(ST.cp);
6055 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
6057 if (!--ST.accepted) {
6059 Perl_re_exec_indentf( aTHX_ "%sTRIE failed...%s\n",
6067 /* Find next-highest word to process. Note that this code
6068 * is O(N^2) per trie run (O(N) per branch), so keep tight */
6071 U16 const nextword = ST.nextword;
6072 reg_trie_wordinfo * const wordinfo
6073 = ((reg_trie_data*)rexi->data->data[ARG(ST.me)])->wordinfo;
6074 for (word=ST.topword; word; word=wordinfo[word].prev) {
6075 if (word > nextword && (!min || word < min))
6088 ST.lastparen = rex->lastparen;
6089 ST.lastcloseparen = rex->lastcloseparen;
6093 /* find start char of end of current word */
6095 U32 chars; /* how many chars to skip */
6096 reg_trie_data * const trie
6097 = (reg_trie_data*)rexi->data->data[ARG(ST.me)];
6099 assert((trie->wordinfo[ST.nextword].len - trie->prefixlen)
6101 chars = (trie->wordinfo[ST.nextword].len - trie->prefixlen)
6106 /* the hard option - fold each char in turn and find
6107 * its folded length (which may be different */
6108 U8 foldbuf[UTF8_MAXBYTES_CASE + 1];
6116 uvc = utf8n_to_uvchr((U8*)uc, UTF8_MAXLEN, &len,
6124 uvc = to_uni_fold(uvc, foldbuf, &foldlen);
6129 uvc = utf8n_to_uvchr(uscan, foldlen, &len,
6145 scan = ST.me + ((ST.jump && ST.jump[ST.nextword])
6146 ? ST.jump[ST.nextword]
6150 Perl_re_exec_indentf( aTHX_ "%sTRIE matched word #%d, continuing%s\n",
6158 if ( ST.accepted > 1 || has_cutgroup || ST.jump ) {
6159 PUSH_STATE_GOTO(TRIE_next, scan, (char*)uc);
6160 NOT_REACHED; /* NOTREACHED */
6162 /* only one choice left - just continue */
6164 AV *const trie_words
6165 = MUTABLE_AV(rexi->data->data[ARG(ST.me)+TRIE_WORDS_OFFSET]);
6166 SV ** const tmp = trie_words
6167 ? av_fetch(trie_words, ST.nextword - 1, 0) : NULL;
6168 SV *sv= tmp ? sv_newmortal() : NULL;
6170 Perl_re_exec_indentf( aTHX_ "%sTRIE: only one match left, short-circuiting: #%d <%s>%s\n",
6171 depth, PL_colors[4],
6173 tmp ? pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 0,
6174 PL_colors[0], PL_colors[1],
6175 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)|PERL_PV_ESCAPE_NONASCII
6177 : "not compiled under -Dr",
6181 locinput = (char*)uc;
6182 continue; /* execute rest of RE */
6187 case EXACTL: /* /abc/l */
6188 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6190 /* Complete checking would involve going through every character
6191 * matched by the string to see if any is above latin1. But the
6192 * comparision otherwise might very well be a fast assembly
6193 * language routine, and I (khw) don't think slowing things down
6194 * just to check for this warning is worth it. So this just checks
6195 * the first character */
6196 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*locinput)) {
6197 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput, reginfo->strend);
6201 if (! utf8_target) {
6205 case EXACT: { /* /abc/ */
6210 if (utf8_target != is_utf8_pat) {
6211 /* The target and the pattern have differing utf8ness. */
6213 const char * const e = s + ln;
6216 /* The target is utf8, the pattern is not utf8.
6217 * Above-Latin1 code points can't match the pattern;
6218 * invariants match exactly, and the other Latin1 ones need
6219 * to be downgraded to a single byte in order to do the
6220 * comparison. (If we could be confident that the target
6221 * is not malformed, this could be refactored to have fewer
6222 * tests by just assuming that if the first bytes match, it
6223 * is an invariant, but there are tests in the test suite
6224 * dealing with (??{...}) which violate this) */
6226 if (l >= reginfo->strend
6227 || UTF8_IS_ABOVE_LATIN1(* (U8*) l))
6231 if (UTF8_IS_INVARIANT(*(U8*)l)) {
6238 if (EIGHT_BIT_UTF8_TO_NATIVE(*l, *(l+1)) != * (U8*) s)
6248 /* The target is not utf8, the pattern is utf8. */
6250 if (l >= reginfo->strend
6251 || UTF8_IS_ABOVE_LATIN1(* (U8*) s))
6255 if (UTF8_IS_INVARIANT(*(U8*)s)) {
6262 if (EIGHT_BIT_UTF8_TO_NATIVE(*s, *(s+1)) != * (U8*) l)
6274 /* The target and the pattern have the same utf8ness. */
6275 /* Inline the first character, for speed. */
6276 if (reginfo->strend - locinput < ln
6277 || UCHARAT(s) != nextchr
6278 || (ln > 1 && memNE(s, locinput, ln)))
6287 case EXACTFL: /* /abc/il */
6290 const U8 * fold_array;
6292 U32 fold_utf8_flags;
6294 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6295 folder = foldEQ_locale;
6296 fold_array = PL_fold_locale;
6297 fold_utf8_flags = FOLDEQ_LOCALE;
6300 case EXACTFLU8: /* /abc/il; but all 'abc' are above 255, so
6301 is effectively /u; hence to match, target
6303 if (! utf8_target) {
6306 fold_utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
6307 | FOLDEQ_S2_FOLDS_SANE;
6308 folder = foldEQ_latin1_s2_folded;
6309 fold_array = PL_fold_latin1;
6312 case EXACTFU_ONLY8: /* /abc/iu with something in /abc/ > 255 */
6313 if (! utf8_target) {
6316 assert(is_utf8_pat);
6317 fold_utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
6320 case EXACTFUP: /* /foo/iu, and something is problematic in
6321 'foo' so can't take shortcuts. */
6322 assert(! is_utf8_pat);
6323 folder = foldEQ_latin1;
6324 fold_array = PL_fold_latin1;
6325 fold_utf8_flags = 0;
6328 case EXACTFU: /* /abc/iu */
6329 folder = foldEQ_latin1_s2_folded;
6330 fold_array = PL_fold_latin1;
6331 fold_utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
6334 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8
6336 assert(! is_utf8_pat);
6338 case EXACTFAA: /* /abc/iaa */
6339 folder = foldEQ_latin1_s2_folded;
6340 fold_array = PL_fold_latin1;
6341 fold_utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6342 if (is_utf8_pat || ! utf8_target) {
6344 /* The possible presence of a MICRO SIGN in the pattern forbids
6345 * us to view a non-UTF-8 pattern as folded when there is a
6347 fold_utf8_flags |= FOLDEQ_S2_ALREADY_FOLDED
6348 |FOLDEQ_S2_FOLDS_SANE;
6353 case EXACTF: /* /abc/i This node only generated for
6354 non-utf8 patterns */
6355 assert(! is_utf8_pat);
6357 fold_array = PL_fold;
6358 fold_utf8_flags = 0;
6366 || state_num == EXACTFUP
6367 || (state_num == EXACTFL && IN_UTF8_CTYPE_LOCALE))
6369 /* Either target or the pattern are utf8, or has the issue where
6370 * the fold lengths may differ. */
6371 const char * const l = locinput;
6372 char *e = reginfo->strend;
6374 if (! foldEQ_utf8_flags(l, &e, 0, utf8_target,
6375 s, 0, ln, is_utf8_pat,fold_utf8_flags))
6383 /* Neither the target nor the pattern are utf8 */
6384 if (UCHARAT(s) != nextchr
6386 && UCHARAT(s) != fold_array[nextchr])
6390 if (reginfo->strend - locinput < ln)
6392 if (ln > 1 && ! folder(locinput, s, ln))
6398 case NBOUNDL: /* /\B/l */
6402 case BOUNDL: /* /\b/l */
6405 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6407 if (FLAGS(scan) != TRADITIONAL_BOUND) {
6408 if (! IN_UTF8_CTYPE_LOCALE) {
6409 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
6410 B_ON_NON_UTF8_LOCALE_IS_WRONG);
6416 if (locinput == reginfo->strbeg)
6417 b1 = isWORDCHAR_LC('\n');
6419 b1 = isWORDCHAR_LC_utf8_safe(reghop3((U8*)locinput, -1,
6420 (U8*)(reginfo->strbeg)),
6421 (U8*)(reginfo->strend));
6423 b2 = (NEXTCHR_IS_EOS)
6424 ? isWORDCHAR_LC('\n')
6425 : isWORDCHAR_LC_utf8_safe((U8*) locinput,
6426 (U8*) reginfo->strend);
6428 else { /* Here the string isn't utf8 */
6429 b1 = (locinput == reginfo->strbeg)
6430 ? isWORDCHAR_LC('\n')
6431 : isWORDCHAR_LC(UCHARAT(locinput - 1));
6432 b2 = (NEXTCHR_IS_EOS)
6433 ? isWORDCHAR_LC('\n')
6434 : isWORDCHAR_LC(nextchr);
6436 if (to_complement ^ (b1 == b2)) {
6442 case NBOUND: /* /\B/ */
6446 case BOUND: /* /\b/ */
6450 goto bound_ascii_match_only;
6452 case NBOUNDA: /* /\B/a */
6456 case BOUNDA: /* /\b/a */
6460 bound_ascii_match_only:
6461 /* Here the string isn't utf8, or is utf8 and only ascii characters
6462 * are to match \w. In the latter case looking at the byte just
6463 * prior to the current one may be just the final byte of a
6464 * multi-byte character. This is ok. There are two cases:
6465 * 1) it is a single byte character, and then the test is doing
6466 * just what it's supposed to.
6467 * 2) it is a multi-byte character, in which case the final byte is
6468 * never mistakable for ASCII, and so the test will say it is
6469 * not a word character, which is the correct answer. */
6470 b1 = (locinput == reginfo->strbeg)
6471 ? isWORDCHAR_A('\n')
6472 : isWORDCHAR_A(UCHARAT(locinput - 1));
6473 b2 = (NEXTCHR_IS_EOS)
6474 ? isWORDCHAR_A('\n')
6475 : isWORDCHAR_A(nextchr);
6476 if (to_complement ^ (b1 == b2)) {
6482 case NBOUNDU: /* /\B/u */
6486 case BOUNDU: /* /\b/u */
6489 if (UNLIKELY(reginfo->strbeg >= reginfo->strend)) {
6492 else if (utf8_target) {
6494 switch((bound_type) FLAGS(scan)) {
6495 case TRADITIONAL_BOUND:
6498 b1 = (locinput == reginfo->strbeg)
6499 ? 0 /* isWORDCHAR_L1('\n') */
6500 : isWORDCHAR_utf8_safe(
6501 reghop3((U8*)locinput,
6503 (U8*)(reginfo->strbeg)),
6504 (U8*) reginfo->strend);
6505 b2 = (NEXTCHR_IS_EOS)
6506 ? 0 /* isWORDCHAR_L1('\n') */
6507 : isWORDCHAR_utf8_safe((U8*)locinput,
6508 (U8*) reginfo->strend);
6509 match = cBOOL(b1 != b2);
6513 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6514 match = TRUE; /* GCB always matches at begin and
6518 /* Find the gcb values of previous and current
6519 * chars, then see if is a break point */
6520 match = isGCB(getGCB_VAL_UTF8(
6521 reghop3((U8*)locinput,
6523 (U8*)(reginfo->strbeg)),
6524 (U8*) reginfo->strend),
6525 getGCB_VAL_UTF8((U8*) locinput,
6526 (U8*) reginfo->strend),
6527 (U8*) reginfo->strbeg,
6534 if (locinput == reginfo->strbeg) {
6537 else if (NEXTCHR_IS_EOS) {
6541 match = isLB(getLB_VAL_UTF8(
6542 reghop3((U8*)locinput,
6544 (U8*)(reginfo->strbeg)),
6545 (U8*) reginfo->strend),
6546 getLB_VAL_UTF8((U8*) locinput,
6547 (U8*) reginfo->strend),
6548 (U8*) reginfo->strbeg,
6550 (U8*) reginfo->strend,
6555 case SB_BOUND: /* Always matches at begin and end */
6556 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6560 match = isSB(getSB_VAL_UTF8(
6561 reghop3((U8*)locinput,
6563 (U8*)(reginfo->strbeg)),
6564 (U8*) reginfo->strend),
6565 getSB_VAL_UTF8((U8*) locinput,
6566 (U8*) reginfo->strend),
6567 (U8*) reginfo->strbeg,
6569 (U8*) reginfo->strend,
6575 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6579 match = isWB(WB_UNKNOWN,
6581 reghop3((U8*)locinput,
6583 (U8*)(reginfo->strbeg)),
6584 (U8*) reginfo->strend),
6585 getWB_VAL_UTF8((U8*) locinput,
6586 (U8*) reginfo->strend),
6587 (U8*) reginfo->strbeg,
6589 (U8*) reginfo->strend,
6595 else { /* Not utf8 target */
6596 switch((bound_type) FLAGS(scan)) {
6597 case TRADITIONAL_BOUND:
6600 b1 = (locinput == reginfo->strbeg)
6601 ? 0 /* isWORDCHAR_L1('\n') */
6602 : isWORDCHAR_L1(UCHARAT(locinput - 1));
6603 b2 = (NEXTCHR_IS_EOS)
6604 ? 0 /* isWORDCHAR_L1('\n') */
6605 : isWORDCHAR_L1(nextchr);
6606 match = cBOOL(b1 != b2);
6611 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6612 match = TRUE; /* GCB always matches at begin and
6615 else { /* Only CR-LF combo isn't a GCB in 0-255
6617 match = UCHARAT(locinput - 1) != '\r'
6618 || UCHARAT(locinput) != '\n';
6623 if (locinput == reginfo->strbeg) {
6626 else if (NEXTCHR_IS_EOS) {
6630 match = isLB(getLB_VAL_CP(UCHARAT(locinput -1)),
6631 getLB_VAL_CP(UCHARAT(locinput)),
6632 (U8*) reginfo->strbeg,
6634 (U8*) reginfo->strend,
6639 case SB_BOUND: /* Always matches at begin and end */
6640 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6644 match = isSB(getSB_VAL_CP(UCHARAT(locinput -1)),
6645 getSB_VAL_CP(UCHARAT(locinput)),
6646 (U8*) reginfo->strbeg,
6648 (U8*) reginfo->strend,
6654 if (locinput == reginfo->strbeg || NEXTCHR_IS_EOS) {
6658 match = isWB(WB_UNKNOWN,
6659 getWB_VAL_CP(UCHARAT(locinput -1)),
6660 getWB_VAL_CP(UCHARAT(locinput)),
6661 (U8*) reginfo->strbeg,
6663 (U8*) reginfo->strend,
6670 if (to_complement ^ ! match) {
6676 case ANYOFL: /* /[abc]/l */
6677 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6679 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(scan)) && ! IN_UTF8_CTYPE_LOCALE)
6681 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
6684 case ANYOFD: /* /[abc]/d */
6685 case ANYOF: /* /[abc]/ */
6688 if ( (! utf8_target || UTF8_IS_INVARIANT(*locinput))
6689 && ! (ANYOF_FLAGS(scan) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP))
6691 if (! ANYOF_BITMAP_TEST(scan, * (U8 *) (locinput))) {
6697 if (!reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
6702 goto increment_locinput;
6707 if (NEXTCHR_IS_EOS || (UCHARAT(locinput) & FLAGS(scan)) != ARG(scan)) {
6710 locinput++; /* ANYOFM is always single byte */
6714 if (NEXTCHR_IS_EOS || (UCHARAT(locinput) & FLAGS(scan)) == ARG(scan)) {
6717 goto increment_locinput;
6723 || ! reginclass(rex, scan, (U8*)locinput, (U8*)reginfo->strend,
6728 goto increment_locinput;
6731 /* The argument (FLAGS) to all the POSIX node types is the class number
6734 case NPOSIXL: /* \W or [:^punct:] etc. under /l */
6738 case POSIXL: /* \w or [:punct:] etc. under /l */
6739 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6743 /* Use isFOO_lc() for characters within Latin1. (Note that
6744 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6745 * wouldn't be invariant) */
6746 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6747 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan), (U8) nextchr)))) {
6755 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6756 /* An above Latin-1 code point, or malformed */
6757 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(locinput,
6759 goto utf8_posix_above_latin1;
6762 /* Here is a UTF-8 variant code point below 256 and the target is
6764 if (! (to_complement ^ cBOOL(isFOO_lc(FLAGS(scan),
6765 EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6766 *(locinput + 1))))))
6771 goto increment_locinput;
6773 case NPOSIXD: /* \W or [:^punct:] etc. under /d */
6777 case POSIXD: /* \w or [:punct:] etc. under /d */
6783 case NPOSIXA: /* \W or [:^punct:] etc. under /a */
6785 if (NEXTCHR_IS_EOS) {
6789 /* All UTF-8 variants match */
6790 if (! UTF8_IS_INVARIANT(nextchr)) {
6791 goto increment_locinput;
6797 case POSIXA: /* \w or [:punct:] etc. under /a */
6800 /* We get here through POSIXD, NPOSIXD, and NPOSIXA when not in
6801 * UTF-8, and also from NPOSIXA even in UTF-8 when the current
6802 * character is a single byte */
6804 if (NEXTCHR_IS_EOS) {
6810 if (! (to_complement ^ cBOOL(_generic_isCC_A(nextchr,
6816 /* Here we are either not in utf8, or we matched a utf8-invariant,
6817 * so the next char is the next byte */
6821 case NPOSIXU: /* \W or [:^punct:] etc. under /u */
6825 case POSIXU: /* \w or [:punct:] etc. under /u */
6827 if (NEXTCHR_IS_EOS) {
6831 /* Use _generic_isCC() for characters within Latin1. (Note that
6832 * UTF8_IS_INVARIANT works even on non-UTF-8 strings, or else
6833 * wouldn't be invariant) */
6834 if (UTF8_IS_INVARIANT(nextchr) || ! utf8_target) {
6835 if (! (to_complement ^ cBOOL(_generic_isCC(nextchr,
6842 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(locinput, reginfo->strend)) {
6843 if (! (to_complement
6844 ^ cBOOL(_generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(nextchr,
6852 else { /* Handle above Latin-1 code points */
6853 utf8_posix_above_latin1:
6854 classnum = (_char_class_number) FLAGS(scan);
6857 if (! (to_complement
6858 ^ cBOOL(_invlist_contains_cp(
6859 PL_XPosix_ptrs[classnum],
6860 utf8_to_uvchr_buf((U8 *) locinput,
6861 (U8 *) reginfo->strend,
6867 case _CC_ENUM_SPACE:
6868 if (! (to_complement
6869 ^ cBOOL(is_XPERLSPACE_high(locinput))))
6874 case _CC_ENUM_BLANK:
6875 if (! (to_complement
6876 ^ cBOOL(is_HORIZWS_high(locinput))))
6881 case _CC_ENUM_XDIGIT:
6882 if (! (to_complement
6883 ^ cBOOL(is_XDIGIT_high(locinput))))
6888 case _CC_ENUM_VERTSPACE:
6889 if (! (to_complement
6890 ^ cBOOL(is_VERTWS_high(locinput))))
6895 case _CC_ENUM_CNTRL: /* These can't match above Latin1 */
6896 case _CC_ENUM_ASCII:
6897 if (! to_complement) {
6902 locinput += UTF8SKIP(locinput);
6906 case CLUMP: /* Match \X: logical Unicode character. This is defined as
6907 a Unicode extended Grapheme Cluster */
6910 if (! utf8_target) {
6912 /* Match either CR LF or '.', as all the other possibilities
6914 locinput++; /* Match the . or CR */
6915 if (nextchr == '\r' /* And if it was CR, and the next is LF,
6917 && locinput < reginfo->strend
6918 && UCHARAT(locinput) == '\n')
6925 /* Get the gcb type for the current character */
6926 GCB_enum prev_gcb = getGCB_VAL_UTF8((U8*) locinput,
6927 (U8*) reginfo->strend);
6929 /* Then scan through the input until we get to the first
6930 * character whose type is supposed to be a gcb with the
6931 * current character. (There is always a break at the
6933 locinput += UTF8SKIP(locinput);
6934 while (locinput < reginfo->strend) {
6935 GCB_enum cur_gcb = getGCB_VAL_UTF8((U8*) locinput,
6936 (U8*) reginfo->strend);
6937 if (isGCB(prev_gcb, cur_gcb,
6938 (U8*) reginfo->strbeg, (U8*) locinput,
6945 locinput += UTF8SKIP(locinput);
6952 case NREFFL: /* /\g{name}/il */
6953 { /* The capture buffer cases. The ones beginning with N for the
6954 named buffers just convert to the equivalent numbered and
6955 pretend they were called as the corresponding numbered buffer
6957 /* don't initialize these in the declaration, it makes C++
6962 const U8 *fold_array;
6965 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
6966 folder = foldEQ_locale;
6967 fold_array = PL_fold_locale;
6969 utf8_fold_flags = FOLDEQ_LOCALE;
6972 case NREFFA: /* /\g{name}/iaa */
6973 folder = foldEQ_latin1;
6974 fold_array = PL_fold_latin1;
6976 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
6979 case NREFFU: /* /\g{name}/iu */
6980 folder = foldEQ_latin1;
6981 fold_array = PL_fold_latin1;
6983 utf8_fold_flags = 0;
6986 case NREFF: /* /\g{name}/i */
6988 fold_array = PL_fold;
6990 utf8_fold_flags = 0;
6993 case NREF: /* /\g{name}/ */
6997 utf8_fold_flags = 0;
7000 /* For the named back references, find the corresponding buffer
7002 n = reg_check_named_buff_matched(rex,scan);
7007 goto do_nref_ref_common;
7009 case REFFL: /* /\1/il */
7010 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
7011 folder = foldEQ_locale;
7012 fold_array = PL_fold_locale;
7013 utf8_fold_flags = FOLDEQ_LOCALE;
7016 case REFFA: /* /\1/iaa */
7017 folder = foldEQ_latin1;
7018 fold_array = PL_fold_latin1;
7019 utf8_fold_flags = FOLDEQ_UTF8_NOMIX_ASCII;
7022 case REFFU: /* /\1/iu */
7023 folder = foldEQ_latin1;
7024 fold_array = PL_fold_latin1;
7025 utf8_fold_flags = 0;
7028 case REFF: /* /\1/i */
7030 fold_array = PL_fold;
7031 utf8_fold_flags = 0;
7034 case REF: /* /\1/ */
7037 utf8_fold_flags = 0;
7041 n = ARG(scan); /* which paren pair */
7044 ln = rex->offs[n].start;
7045 endref = rex->offs[n].end;
7046 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7047 if (rex->lastparen < n || ln == -1 || endref == -1)
7048 sayNO; /* Do not match unless seen CLOSEn. */
7052 s = reginfo->strbeg + ln;
7053 if (type != REF /* REF can do byte comparison */
7054 && (utf8_target || type == REFFU || type == REFFL))
7056 char * limit = reginfo->strend;
7058 /* This call case insensitively compares the entire buffer
7059 * at s, with the current input starting at locinput, but
7060 * not going off the end given by reginfo->strend, and
7061 * returns in <limit> upon success, how much of the
7062 * current input was matched */
7063 if (! foldEQ_utf8_flags(s, NULL, endref - ln, utf8_target,
7064 locinput, &limit, 0, utf8_target, utf8_fold_flags))
7072 /* Not utf8: Inline the first character, for speed. */
7073 if (!NEXTCHR_IS_EOS &&
7074 UCHARAT(s) != nextchr &&
7076 UCHARAT(s) != fold_array[nextchr]))
7079 if (locinput + ln > reginfo->strend)
7081 if (ln > 1 && (type == REF
7082 ? memNE(s, locinput, ln)
7083 : ! folder(locinput, s, ln)))
7089 case NOTHING: /* null op; e.g. the 'nothing' following
7090 * the '*' in m{(a+|b)*}' */
7092 case TAIL: /* placeholder while compiling (A|B|C) */
7096 #define ST st->u.eval
7097 #define CUR_EVAL cur_eval->u.eval
7103 regexp_internal *rei;
7104 regnode *startpoint;
7107 case GOSUB: /* /(...(?1))/ /(...(?&foo))/ */
7108 arg= (U32)ARG(scan);
7109 if (cur_eval && cur_eval->locinput == locinput) {
7110 if ( ++nochange_depth > max_nochange_depth )
7112 "Pattern subroutine nesting without pos change"
7113 " exceeded limit in regex");
7120 startpoint = scan + ARG2L(scan);
7121 EVAL_CLOSE_PAREN_SET( st, arg );
7122 /* Detect infinite recursion
7124 * A pattern like /(?R)foo/ or /(?<x>(?&y)foo)(?<y>(?&x)bar)/
7125 * or "a"=~/(.(?2))((?<=(?=(?1)).))/ could recurse forever.
7126 * So we track the position in the string we are at each time
7127 * we recurse and if we try to enter the same routine twice from
7128 * the same position we throw an error.
7130 if ( rex->recurse_locinput[arg] == locinput ) {
7131 /* FIXME: we should show the regop that is failing as part
7132 * of the error message. */
7133 Perl_croak(aTHX_ "Infinite recursion in regex");
7135 ST.prev_recurse_locinput= rex->recurse_locinput[arg];
7136 rex->recurse_locinput[arg]= locinput;
7139 GET_RE_DEBUG_FLAGS_DECL;
7141 Perl_re_exec_indentf( aTHX_
7142 "entering GOSUB, prev_recurse_locinput=%p recurse_locinput[%d]=%p\n",
7143 depth, ST.prev_recurse_locinput, arg, rex->recurse_locinput[arg]
7149 /* Save all the positions seen so far. */
7150 ST.cp = regcppush(rex, 0, maxopenparen);
7151 REGCP_SET(ST.lastcp);
7153 /* and then jump to the code we share with EVAL */
7154 goto eval_recurse_doit;
7157 case EVAL: /* /(?{...})B/ /(??{A})B/ and /(?(?{...})X|Y)B/ */
7158 if (cur_eval && cur_eval->locinput==locinput) {
7159 if ( ++nochange_depth > max_nochange_depth )
7160 Perl_croak(aTHX_ "EVAL without pos change exceeded limit in regex");
7165 /* execute the code in the {...} */
7169 OP * const oop = PL_op;
7170 COP * const ocurcop = PL_curcop;
7174 /* save *all* paren positions */
7175 regcppush(rex, 0, maxopenparen);
7176 REGCP_SET(ST.lastcp);
7179 caller_cv = find_runcv(NULL);
7183 if (rexi->data->what[n] == 'r') { /* code from an external qr */
7185 (REGEXP*)(rexi->data->data[n])
7187 nop = (OP*)rexi->data->data[n+1];
7189 else if (rexi->data->what[n] == 'l') { /* literal code */
7191 nop = (OP*)rexi->data->data[n];
7192 assert(CvDEPTH(newcv));
7195 /* literal with own CV */
7196 assert(rexi->data->what[n] == 'L');
7197 newcv = rex->qr_anoncv;
7198 nop = (OP*)rexi->data->data[n];
7201 /* Some notes about MULTICALL and the context and save stacks.
7204 * /...(?{ my $x)}...(?{ my $y)}...(?{ my $z)}.../
7205 * since codeblocks don't introduce a new scope (so that
7206 * local() etc accumulate), at the end of a successful
7207 * match there will be a SAVEt_CLEARSV on the savestack
7208 * for each of $x, $y, $z. If the three code blocks above
7209 * happen to have come from different CVs (e.g. via
7210 * embedded qr//s), then we must ensure that during any
7211 * savestack unwinding, PL_comppad always points to the
7212 * right pad at each moment. We achieve this by
7213 * interleaving SAVEt_COMPPAD's on the savestack whenever
7214 * there is a change of pad.
7215 * In theory whenever we call a code block, we should
7216 * push a CXt_SUB context, then pop it on return from
7217 * that code block. This causes a bit of an issue in that
7218 * normally popping a context also clears the savestack
7219 * back to cx->blk_oldsaveix, but here we specifically
7220 * don't want to clear the save stack on exit from the
7222 * Also for efficiency we don't want to keep pushing and
7223 * popping the single SUB context as we backtrack etc.
7224 * So instead, we push a single context the first time
7225 * we need, it, then hang onto it until the end of this
7226 * function. Whenever we encounter a new code block, we
7227 * update the CV etc if that's changed. During the times
7228 * in this function where we're not executing a code
7229 * block, having the SUB context still there is a bit
7230 * naughty - but we hope that no-one notices.
7231 * When the SUB context is initially pushed, we fake up
7232 * cx->blk_oldsaveix to be as if we'd pushed this context
7233 * on first entry to S_regmatch rather than at some random
7234 * point during the regexe execution. That way if we
7235 * croak, popping the context stack will ensure that
7236 * *everything* SAVEd by this function is undone and then
7237 * the context popped, rather than e.g., popping the
7238 * context (and restoring the original PL_comppad) then
7239 * popping more of the savestack and restoring a bad
7243 /* If this is the first EVAL, push a MULTICALL. On
7244 * subsequent calls, if we're executing a different CV, or
7245 * if PL_comppad has got messed up from backtracking
7246 * through SAVECOMPPADs, then refresh the context.
7248 if (newcv != last_pushed_cv || PL_comppad != last_pad)
7250 U8 flags = (CXp_SUB_RE |
7251 ((newcv == caller_cv) ? CXp_SUB_RE_FAKE : 0));
7253 if (last_pushed_cv) {
7254 CHANGE_MULTICALL_FLAGS(newcv, flags);
7257 PUSH_MULTICALL_FLAGS(newcv, flags);
7259 /* see notes above */
7260 CX_CUR()->blk_oldsaveix = orig_savestack_ix;
7262 last_pushed_cv = newcv;
7265 /* these assignments are just to silence compiler
7267 multicall_cop = NULL;
7269 last_pad = PL_comppad;
7271 /* the initial nextstate you would normally execute
7272 * at the start of an eval (which would cause error
7273 * messages to come from the eval), may be optimised
7274 * away from the execution path in the regex code blocks;
7275 * so manually set PL_curcop to it initially */
7277 OP *o = cUNOPx(nop)->op_first;
7278 assert(o->op_type == OP_NULL);
7279 if (o->op_targ == OP_SCOPE) {
7280 o = cUNOPo->op_first;
7283 assert(o->op_targ == OP_LEAVE);
7284 o = cUNOPo->op_first;
7285 assert(o->op_type == OP_ENTER);
7289 if (o->op_type != OP_STUB) {
7290 assert( o->op_type == OP_NEXTSTATE
7291 || o->op_type == OP_DBSTATE
7292 || (o->op_type == OP_NULL
7293 && ( o->op_targ == OP_NEXTSTATE
7294 || o->op_targ == OP_DBSTATE
7298 PL_curcop = (COP*)o;
7303 DEBUG_STATE_r( Perl_re_printf( aTHX_
7304 " re EVAL PL_op=0x%" UVxf "\n", PTR2UV(nop)) );
7306 rex->offs[0].end = locinput - reginfo->strbeg;
7307 if (reginfo->info_aux_eval->pos_magic)
7308 MgBYTEPOS_set(reginfo->info_aux_eval->pos_magic,
7309 reginfo->sv, reginfo->strbeg,
7310 locinput - reginfo->strbeg);
7313 SV *sv_mrk = get_sv("REGMARK", 1);
7314 sv_setsv(sv_mrk, sv_yes_mark);
7317 /* we don't use MULTICALL here as we want to call the
7318 * first op of the block of interest, rather than the
7319 * first op of the sub. Also, we don't want to free
7320 * the savestack frame */
7321 before = (IV)(SP-PL_stack_base);
7323 CALLRUNOPS(aTHX); /* Scalar context. */
7325 if ((IV)(SP-PL_stack_base) == before)
7326 ret = &PL_sv_undef; /* protect against empty (?{}) blocks. */
7332 /* before restoring everything, evaluate the returned
7333 * value, so that 'uninit' warnings don't use the wrong
7334 * PL_op or pad. Also need to process any magic vars
7335 * (e.g. $1) *before* parentheses are restored */
7340 if (logical == 0) { /* (?{})/ */
7341 SV *replsv = save_scalar(PL_replgv);
7342 sv_setsv(replsv, ret); /* $^R */
7345 else if (logical == 1) { /* /(?(?{...})X|Y)/ */
7346 sw = cBOOL(SvTRUE_NN(ret));
7349 else { /* /(??{}) */
7350 /* if its overloaded, let the regex compiler handle
7351 * it; otherwise extract regex, or stringify */
7352 if (SvGMAGICAL(ret))
7353 ret = sv_mortalcopy(ret);
7354 if (!SvAMAGIC(ret)) {
7358 if (SvTYPE(sv) == SVt_REGEXP)
7359 re_sv = (REGEXP*) sv;
7360 else if (SvSMAGICAL(ret)) {
7361 MAGIC *mg = mg_find(ret, PERL_MAGIC_qr);
7363 re_sv = (REGEXP *) mg->mg_obj;
7366 /* force any undef warnings here */
7367 if (!re_sv && !SvPOK(ret) && !SvNIOK(ret)) {
7368 ret = sv_mortalcopy(ret);
7369 (void) SvPV_force_nolen(ret);
7375 /* *** Note that at this point we don't restore
7376 * PL_comppad, (or pop the CxSUB) on the assumption it may
7377 * be used again soon. This is safe as long as nothing
7378 * in the regexp code uses the pad ! */
7380 PL_curcop = ocurcop;
7381 regcp_restore(rex, ST.lastcp, &maxopenparen);
7382 PL_curpm_under = PL_curpm;
7383 PL_curpm = PL_reg_curpm;
7386 PUSH_STATE_GOTO(EVAL_B, next, locinput);
7391 /* only /(??{})/ from now on */
7394 /* extract RE object from returned value; compiling if
7398 re_sv = reg_temp_copy(NULL, re_sv);
7403 if (SvUTF8(ret) && IN_BYTES) {
7404 /* In use 'bytes': make a copy of the octet
7405 * sequence, but without the flag on */
7407 const char *const p = SvPV(ret, len);
7408 ret = newSVpvn_flags(p, len, SVs_TEMP);
7410 if (rex->intflags & PREGf_USE_RE_EVAL)
7411 pm_flags |= PMf_USE_RE_EVAL;
7413 /* if we got here, it should be an engine which
7414 * supports compiling code blocks and stuff */
7415 assert(rex->engine && rex->engine->op_comp);
7416 assert(!(scan->flags & ~RXf_PMf_COMPILETIME));
7417 re_sv = rex->engine->op_comp(aTHX_ &ret, 1, NULL,
7418 rex->engine, NULL, NULL,
7419 /* copy /msixn etc to inner pattern */
7424 & (SVs_TEMP | SVs_GMG | SVf_ROK))
7425 && (!SvPADTMP(ret) || SvREADONLY(ret))) {
7426 /* This isn't a first class regexp. Instead, it's
7427 caching a regexp onto an existing, Perl visible
7429 sv_magic(ret, MUTABLE_SV(re_sv), PERL_MAGIC_qr, 0, 0);
7435 RXp_MATCH_COPIED_off(re);
7436 re->subbeg = rex->subbeg;
7437 re->sublen = rex->sublen;
7438 re->suboffset = rex->suboffset;
7439 re->subcoffset = rex->subcoffset;
7441 re->lastcloseparen = 0;
7444 debug_start_match(re_sv, utf8_target, locinput,
7445 reginfo->strend, "EVAL/GOSUB: Matching embedded");
7447 startpoint = rei->program + 1;
7448 EVAL_CLOSE_PAREN_CLEAR(st); /* ST.close_paren = 0;
7449 * close_paren only for GOSUB */
7450 ST.prev_recurse_locinput= NULL; /* only used for GOSUB */
7451 /* Save all the seen positions so far. */
7452 ST.cp = regcppush(rex, 0, maxopenparen);
7453 REGCP_SET(ST.lastcp);
7454 /* and set maxopenparen to 0, since we are starting a "fresh" match */
7456 /* run the pattern returned from (??{...}) */
7458 eval_recurse_doit: /* Share code with GOSUB below this line
7459 * At this point we expect the stack context to be
7460 * set up correctly */
7462 /* invalidate the S-L poscache. We're now executing a
7463 * different set of WHILEM ops (and their associated
7464 * indexes) against the same string, so the bits in the
7465 * cache are meaningless. Setting maxiter to zero forces
7466 * the cache to be invalidated and zeroed before reuse.
7467 * XXX This is too dramatic a measure. Ideally we should
7468 * save the old cache and restore when running the outer
7470 reginfo->poscache_maxiter = 0;
7472 /* the new regexp might have a different is_utf8_pat than we do */
7473 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(re_sv));
7475 ST.prev_rex = rex_sv;
7476 ST.prev_curlyx = cur_curlyx;
7478 SET_reg_curpm(rex_sv);
7483 ST.prev_eval = cur_eval;
7485 /* now continue from first node in postoned RE */
7486 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, startpoint, locinput);
7487 NOT_REACHED; /* NOTREACHED */
7490 case EVAL_postponed_AB: /* cleanup after a successful (??{A})B */
7491 /* note: this is called twice; first after popping B, then A */
7493 Perl_re_exec_indentf( aTHX_ "EVAL_AB cur_eval=%p prev_eval=%p\n",
7494 depth, cur_eval, ST.prev_eval);
7497 #define SET_RECURSE_LOCINPUT(STR,VAL)\
7498 if ( cur_eval && CUR_EVAL.close_paren ) {\
7500 Perl_re_exec_indentf( aTHX_ STR " GOSUB%d ce=%p recurse_locinput=%p\n",\
7502 CUR_EVAL.close_paren - 1,\
7506 rex->recurse_locinput[CUR_EVAL.close_paren - 1] = VAL;\
7509 SET_RECURSE_LOCINPUT("EVAL_AB[before]", CUR_EVAL.prev_recurse_locinput);
7511 rex_sv = ST.prev_rex;
7512 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7513 SET_reg_curpm(rex_sv);
7514 rex = ReANY(rex_sv);
7515 rexi = RXi_GET(rex);
7517 /* preserve $^R across LEAVE's. See Bug 121070. */
7518 SV *save_sv= GvSV(PL_replgv);
7520 SvREFCNT_inc(save_sv);
7521 regcpblow(ST.cp); /* LEAVE in disguise */
7522 /* don't move this initialization up */
7523 replsv = GvSV(PL_replgv);
7524 sv_setsv(replsv, save_sv);
7526 SvREFCNT_dec(save_sv);
7528 cur_eval = ST.prev_eval;
7529 cur_curlyx = ST.prev_curlyx;
7531 /* Invalidate cache. See "invalidate" comment above. */
7532 reginfo->poscache_maxiter = 0;
7533 if ( nochange_depth )
7536 SET_RECURSE_LOCINPUT("EVAL_AB[after]", cur_eval->locinput);
7540 case EVAL_B_fail: /* unsuccessful B in (?{...})B */
7541 REGCP_UNWIND(ST.lastcp);
7544 case EVAL_postponed_AB_fail: /* unsuccessfully ran A or B in (??{A})B */
7545 /* note: this is called twice; first after popping B, then A */
7547 Perl_re_exec_indentf( aTHX_ "EVAL_AB_fail cur_eval=%p prev_eval=%p\n",
7548 depth, cur_eval, ST.prev_eval);
7551 SET_RECURSE_LOCINPUT("EVAL_AB_fail[before]", CUR_EVAL.prev_recurse_locinput);
7553 rex_sv = ST.prev_rex;
7554 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
7555 SET_reg_curpm(rex_sv);
7556 rex = ReANY(rex_sv);
7557 rexi = RXi_GET(rex);
7559 REGCP_UNWIND(ST.lastcp);
7560 regcppop(rex, &maxopenparen);
7561 cur_eval = ST.prev_eval;
7562 cur_curlyx = ST.prev_curlyx;
7564 /* Invalidate cache. See "invalidate" comment above. */
7565 reginfo->poscache_maxiter = 0;
7566 if ( nochange_depth )
7569 SET_RECURSE_LOCINPUT("EVAL_AB_fail[after]", cur_eval->locinput);
7574 n = ARG(scan); /* which paren pair */
7575 rex->offs[n].start_tmp = locinput - reginfo->strbeg;
7576 if (n > maxopenparen)
7578 DEBUG_BUFFERS_r(Perl_re_exec_indentf( aTHX_
7579 "OPEN: rex=0x%" UVxf " offs=0x%" UVxf ": \\%" UVuf ": set %" IVdf " tmp; maxopenparen=%" UVuf "\n",
7584 (IV)rex->offs[n].start_tmp,
7590 case SROPEN: /* (*SCRIPT_RUN: */
7591 script_run_begin = (U8 *) locinput;
7596 n = ARG(scan); /* which paren pair */
7597 CLOSE_CAPTURE(n, rex->offs[n].start_tmp,
7598 locinput - reginfo->strbeg);
7599 if ( EVAL_CLOSE_PAREN_IS( cur_eval, n ) )
7604 case SRCLOSE: /* (*SCRIPT_RUN: ... ) */
7606 if (! isSCRIPT_RUN(script_run_begin, (U8 *) locinput, utf8_target))
7614 case ACCEPT: /* (*ACCEPT) */
7616 sv_yes_mark = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
7620 cursor && OP(cursor)!=END;
7621 cursor=regnext(cursor))
7623 if ( OP(cursor)==CLOSE ){
7625 if ( n <= lastopen ) {
7626 CLOSE_CAPTURE(n, rex->offs[n].start_tmp,
7627 locinput - reginfo->strbeg);
7628 if ( n == ARG(scan) || EVAL_CLOSE_PAREN_IS(cur_eval, n) )
7637 case GROUPP: /* (?(1)) */
7638 n = ARG(scan); /* which paren pair */
7639 sw = cBOOL(rex->lastparen >= n && rex->offs[n].end != -1);
7642 case NGROUPP: /* (?(<name>)) */
7643 /* reg_check_named_buff_matched returns 0 for no match */
7644 sw = cBOOL(0 < reg_check_named_buff_matched(rex,scan));
7647 case INSUBP: /* (?(R)) */
7649 /* this does not need to use EVAL_CLOSE_PAREN macros, as the arg
7650 * of SCAN is already set up as matches a eval.close_paren */
7651 sw = cur_eval && (n == 0 || CUR_EVAL.close_paren == n);
7654 case DEFINEP: /* (?(DEFINE)) */
7658 case IFTHEN: /* (?(cond)A|B) */
7659 reginfo->poscache_iter = reginfo->poscache_maxiter; /* Void cache */
7661 next = NEXTOPER(NEXTOPER(scan));
7663 next = scan + ARG(scan);
7664 if (OP(next) == IFTHEN) /* Fake one. */
7665 next = NEXTOPER(NEXTOPER(next));
7669 case LOGICAL: /* modifier for EVAL and IFMATCH */
7670 logical = scan->flags;
7673 /*******************************************************************
7675 The CURLYX/WHILEM pair of ops handle the most generic case of the /A*B/
7676 pattern, where A and B are subpatterns. (For simple A, CURLYM or
7677 STAR/PLUS/CURLY/CURLYN are used instead.)
7679 A*B is compiled as <CURLYX><A><WHILEM><B>
7681 On entry to the subpattern, CURLYX is called. This pushes a CURLYX
7682 state, which contains the current count, initialised to -1. It also sets
7683 cur_curlyx to point to this state, with any previous value saved in the
7686 CURLYX then jumps straight to the WHILEM op, rather than executing A,
7687 since the pattern may possibly match zero times (i.e. it's a while {} loop
7688 rather than a do {} while loop).
7690 Each entry to WHILEM represents a successful match of A. The count in the
7691 CURLYX block is incremented, another WHILEM state is pushed, and execution
7692 passes to A or B depending on greediness and the current count.
7694 For example, if matching against the string a1a2a3b (where the aN are
7695 substrings that match /A/), then the match progresses as follows: (the
7696 pushed states are interspersed with the bits of strings matched so far):
7699 <CURLYX cnt=0><WHILEM>
7700 <CURLYX cnt=1><WHILEM> a1 <WHILEM>
7701 <CURLYX cnt=2><WHILEM> a1 <WHILEM> a2 <WHILEM>
7702 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM>
7703 <CURLYX cnt=3><WHILEM> a1 <WHILEM> a2 <WHILEM> a3 <WHILEM> b
7705 (Contrast this with something like CURLYM, which maintains only a single
7709 a1 <CURLYM cnt=1> a2
7710 a1 a2 <CURLYM cnt=2> a3
7711 a1 a2 a3 <CURLYM cnt=3> b
7714 Each WHILEM state block marks a point to backtrack to upon partial failure
7715 of A or B, and also contains some minor state data related to that
7716 iteration. The CURLYX block, pointed to by cur_curlyx, contains the
7717 overall state, such as the count, and pointers to the A and B ops.
7719 This is complicated slightly by nested CURLYX/WHILEM's. Since cur_curlyx
7720 must always point to the *current* CURLYX block, the rules are:
7722 When executing CURLYX, save the old cur_curlyx in the CURLYX state block,
7723 and set cur_curlyx to point the new block.
7725 When popping the CURLYX block after a successful or unsuccessful match,
7726 restore the previous cur_curlyx.
7728 When WHILEM is about to execute B, save the current cur_curlyx, and set it
7729 to the outer one saved in the CURLYX block.
7731 When popping the WHILEM block after a successful or unsuccessful B match,
7732 restore the previous cur_curlyx.
7734 Here's an example for the pattern (AI* BI)*BO
7735 I and O refer to inner and outer, C and W refer to CURLYX and WHILEM:
7738 curlyx backtrack stack
7739 ------ ---------------
7741 CO <CO prev=NULL> <WO>
7742 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7743 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7744 NULL <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi <WO prev=CO> bo
7746 At this point the pattern succeeds, and we work back down the stack to
7747 clean up, restoring as we go:
7749 CO <CO prev=NULL> <WO> <CI prev=CO> <WI> ai <WI prev=CI> bi
7750 CI <CO prev=NULL> <WO> <CI prev=CO> <WI> ai
7751 CO <CO prev=NULL> <WO>
7754 *******************************************************************/
7756 #define ST st->u.curlyx
7758 case CURLYX: /* start of /A*B/ (for complex A) */
7760 /* No need to save/restore up to this paren */
7761 I32 parenfloor = scan->flags;
7763 assert(next); /* keep Coverity happy */
7764 if (OP(PREVOPER(next)) == NOTHING) /* LONGJMP */
7767 /* XXXX Probably it is better to teach regpush to support
7768 parenfloor > maxopenparen ... */
7769 if (parenfloor > (I32)rex->lastparen)
7770 parenfloor = rex->lastparen; /* Pessimization... */
7772 ST.prev_curlyx= cur_curlyx;
7774 ST.cp = PL_savestack_ix;
7776 /* these fields contain the state of the current curly.
7777 * they are accessed by subsequent WHILEMs */
7778 ST.parenfloor = parenfloor;
7783 ST.count = -1; /* this will be updated by WHILEM */
7784 ST.lastloc = NULL; /* this will be updated by WHILEM */
7786 PUSH_YES_STATE_GOTO(CURLYX_end, PREVOPER(next), locinput);
7787 NOT_REACHED; /* NOTREACHED */
7790 case CURLYX_end: /* just finished matching all of A*B */
7791 cur_curlyx = ST.prev_curlyx;
7793 NOT_REACHED; /* NOTREACHED */
7795 case CURLYX_end_fail: /* just failed to match all of A*B */
7797 cur_curlyx = ST.prev_curlyx;
7799 NOT_REACHED; /* NOTREACHED */
7803 #define ST st->u.whilem
7805 case WHILEM: /* just matched an A in /A*B/ (for complex A) */
7807 /* see the discussion above about CURLYX/WHILEM */
7812 assert(cur_curlyx); /* keep Coverity happy */
7814 min = ARG1(cur_curlyx->u.curlyx.me);
7815 max = ARG2(cur_curlyx->u.curlyx.me);
7816 A = NEXTOPER(cur_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS;
7817 n = ++cur_curlyx->u.curlyx.count; /* how many A's matched */
7818 ST.save_lastloc = cur_curlyx->u.curlyx.lastloc;
7819 ST.cache_offset = 0;
7823 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: matched %ld out of %d..%d\n",
7824 depth, (long)n, min, max)
7827 /* First just match a string of min A's. */
7830 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor, maxopenparen);
7831 cur_curlyx->u.curlyx.lastloc = locinput;
7832 REGCP_SET(ST.lastcp);
7834 PUSH_STATE_GOTO(WHILEM_A_pre, A, locinput);
7835 NOT_REACHED; /* NOTREACHED */
7838 /* If degenerate A matches "", assume A done. */
7840 if (locinput == cur_curlyx->u.curlyx.lastloc) {
7841 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: empty match detected, trying continuation...\n",
7844 goto do_whilem_B_max;
7847 /* super-linear cache processing.
7849 * The idea here is that for certain types of CURLYX/WHILEM -
7850 * principally those whose upper bound is infinity (and
7851 * excluding regexes that have things like \1 and other very
7852 * non-regular expresssiony things), then if a pattern like
7853 * /....A*.../ fails and we backtrack to the WHILEM, then we
7854 * make a note that this particular WHILEM op was at string
7855 * position 47 (say) when the rest of pattern failed. Then, if
7856 * we ever find ourselves back at that WHILEM, and at string
7857 * position 47 again, we can just fail immediately rather than
7858 * running the rest of the pattern again.
7860 * This is very handy when patterns start to go
7861 * 'super-linear', like in (a+)*(a+)*(a+)*, where you end up
7862 * with a combinatorial explosion of backtracking.
7864 * The cache is implemented as a bit array, with one bit per
7865 * string byte position per WHILEM op (up to 16) - so its
7866 * between 0.25 and 2x the string size.
7868 * To avoid allocating a poscache buffer every time, we do an
7869 * initially countdown; only after we have executed a WHILEM
7870 * op (string-length x #WHILEMs) times do we allocate the
7873 * The top 4 bits of scan->flags byte say how many different
7874 * relevant CURLLYX/WHILEM op pairs there are, while the
7875 * bottom 4-bits is the identifying index number of this
7881 if (!reginfo->poscache_maxiter) {
7882 /* start the countdown: Postpone detection until we
7883 * know the match is not *that* much linear. */
7884 reginfo->poscache_maxiter
7885 = (reginfo->strend - reginfo->strbeg + 1)
7887 /* possible overflow for long strings and many CURLYX's */
7888 if (reginfo->poscache_maxiter < 0)
7889 reginfo->poscache_maxiter = I32_MAX;
7890 reginfo->poscache_iter = reginfo->poscache_maxiter;
7893 if (reginfo->poscache_iter-- == 0) {
7894 /* initialise cache */
7895 const SSize_t size = (reginfo->poscache_maxiter + 7)/8;
7896 regmatch_info_aux *const aux = reginfo->info_aux;
7897 if (aux->poscache) {
7898 if ((SSize_t)reginfo->poscache_size < size) {
7899 Renew(aux->poscache, size, char);
7900 reginfo->poscache_size = size;
7902 Zero(aux->poscache, size, char);
7905 reginfo->poscache_size = size;
7906 Newxz(aux->poscache, size, char);
7908 DEBUG_EXECUTE_r( Perl_re_printf( aTHX_
7909 "%sWHILEM: Detected a super-linear match, switching on caching%s...\n",
7910 PL_colors[4], PL_colors[5])
7914 if (reginfo->poscache_iter < 0) {
7915 /* have we already failed at this position? */
7916 SSize_t offset, mask;
7918 reginfo->poscache_iter = -1; /* stop eventual underflow */
7919 offset = (scan->flags & 0xf) - 1
7920 + (locinput - reginfo->strbeg)
7922 mask = 1 << (offset % 8);
7924 if (reginfo->info_aux->poscache[offset] & mask) {
7925 DEBUG_EXECUTE_r( Perl_re_exec_indentf( aTHX_ "WHILEM: (cache) already tried at this position...\n",
7928 cur_curlyx->u.curlyx.count--;
7929 sayNO; /* cache records failure */
7931 ST.cache_offset = offset;
7932 ST.cache_mask = mask;
7936 /* Prefer B over A for minimal matching. */
7938 if (cur_curlyx->u.curlyx.minmod) {
7939 ST.save_curlyx = cur_curlyx;
7940 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
7941 PUSH_YES_STATE_GOTO(WHILEM_B_min, ST.save_curlyx->u.curlyx.B,
7943 NOT_REACHED; /* NOTREACHED */
7946 /* Prefer A over B for maximal matching. */
7948 if (n < max) { /* More greed allowed? */
7949 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
7951 cur_curlyx->u.curlyx.lastloc = locinput;
7952 REGCP_SET(ST.lastcp);
7953 PUSH_STATE_GOTO(WHILEM_A_max, A, locinput);
7954 NOT_REACHED; /* NOTREACHED */
7956 goto do_whilem_B_max;
7958 NOT_REACHED; /* NOTREACHED */
7960 case WHILEM_B_min: /* just matched B in a minimal match */
7961 case WHILEM_B_max: /* just matched B in a maximal match */
7962 cur_curlyx = ST.save_curlyx;
7964 NOT_REACHED; /* NOTREACHED */
7966 case WHILEM_B_max_fail: /* just failed to match B in a maximal match */
7967 cur_curlyx = ST.save_curlyx;
7968 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7969 cur_curlyx->u.curlyx.count--;
7971 NOT_REACHED; /* NOTREACHED */
7973 case WHILEM_A_min_fail: /* just failed to match A in a minimal match */
7975 case WHILEM_A_pre_fail: /* just failed to match even minimal A */
7976 REGCP_UNWIND(ST.lastcp);
7977 regcppop(rex, &maxopenparen);
7978 cur_curlyx->u.curlyx.lastloc = ST.save_lastloc;
7979 cur_curlyx->u.curlyx.count--;
7981 NOT_REACHED; /* NOTREACHED */
7983 case WHILEM_A_max_fail: /* just failed to match A in a maximal match */
7984 REGCP_UNWIND(ST.lastcp);
7985 regcppop(rex, &maxopenparen); /* Restore some previous $<digit>s? */
7986 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "WHILEM: failed, trying continuation...\n",
7990 if (cur_curlyx->u.curlyx.count >= REG_INFTY
7991 && ckWARN(WARN_REGEXP)
7992 && !reginfo->warned)
7994 reginfo->warned = TRUE;
7995 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
7996 "Complex regular subexpression recursion limit (%d) "
8002 ST.save_curlyx = cur_curlyx;
8003 cur_curlyx = cur_curlyx->u.curlyx.prev_curlyx;
8004 PUSH_YES_STATE_GOTO(WHILEM_B_max, ST.save_curlyx->u.curlyx.B,
8006 NOT_REACHED; /* NOTREACHED */
8008 case WHILEM_B_min_fail: /* just failed to match B in a minimal match */
8009 cur_curlyx = ST.save_curlyx;
8011 if (cur_curlyx->u.curlyx.count >= /*max*/ARG2(cur_curlyx->u.curlyx.me)) {
8012 /* Maximum greed exceeded */
8013 if (cur_curlyx->u.curlyx.count >= REG_INFTY
8014 && ckWARN(WARN_REGEXP)
8015 && !reginfo->warned)
8017 reginfo->warned = TRUE;
8018 Perl_warner(aTHX_ packWARN(WARN_REGEXP),
8019 "Complex regular subexpression recursion "
8020 "limit (%d) exceeded",
8023 cur_curlyx->u.curlyx.count--;
8027 DEBUG_EXECUTE_r(Perl_re_exec_indentf( aTHX_ "WHILEM: B min fail: trying longer...\n", depth)
8029 /* Try grabbing another A and see if it helps. */
8030 cur_curlyx->u.curlyx.lastloc = locinput;
8031 ST.cp = regcppush(rex, cur_curlyx->u.curlyx.parenfloor,
8033 REGCP_SET(ST.lastcp);
8034 PUSH_STATE_GOTO(WHILEM_A_min,
8035 /*A*/ NEXTOPER(ST.save_curlyx->u.curlyx.me) + EXTRA_STEP_2ARGS,
8037 NOT_REACHED; /* NOTREACHED */
8040 #define ST st->u.branch
8042 case BRANCHJ: /* /(...|A|...)/ with long next pointer */
8043 next = scan + ARG(scan);
8046 scan = NEXTOPER(scan);
8049 case BRANCH: /* /(...|A|...)/ */
8050 scan = NEXTOPER(scan); /* scan now points to inner node */
8051 ST.lastparen = rex->lastparen;
8052 ST.lastcloseparen = rex->lastcloseparen;
8053 ST.next_branch = next;
8056 /* Now go into the branch */
8058 PUSH_YES_STATE_GOTO(BRANCH_next, scan, locinput);
8060 PUSH_STATE_GOTO(BRANCH_next, scan, locinput);
8062 NOT_REACHED; /* NOTREACHED */
8064 case CUTGROUP: /* /(*THEN)/ */
8065 sv_yes_mark = st->u.mark.mark_name = scan->flags
8066 ? MUTABLE_SV(rexi->data->data[ ARG( scan ) ])
8068 PUSH_STATE_GOTO(CUTGROUP_next, next, locinput);
8069 NOT_REACHED; /* NOTREACHED */
8071 case CUTGROUP_next_fail:
8074 if (st->u.mark.mark_name)
8075 sv_commit = st->u.mark.mark_name;
8077 NOT_REACHED; /* NOTREACHED */
8081 NOT_REACHED; /* NOTREACHED */
8083 case BRANCH_next_fail: /* that branch failed; try the next, if any */
8088 REGCP_UNWIND(ST.cp);
8089 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8090 scan = ST.next_branch;
8091 /* no more branches? */
8092 if (!scan || (OP(scan) != BRANCH && OP(scan) != BRANCHJ)) {
8094 Perl_re_exec_indentf( aTHX_ "%sBRANCH failed...%s\n",
8101 continue; /* execute next BRANCH[J] op */
8104 case MINMOD: /* next op will be non-greedy, e.g. A*? */
8109 #define ST st->u.curlym
8111 case CURLYM: /* /A{m,n}B/ where A is fixed-length */
8113 /* This is an optimisation of CURLYX that enables us to push
8114 * only a single backtracking state, no matter how many matches
8115 * there are in {m,n}. It relies on the pattern being constant
8116 * length, with no parens to influence future backrefs
8120 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
8122 ST.lastparen = rex->lastparen;
8123 ST.lastcloseparen = rex->lastcloseparen;
8125 /* if paren positive, emulate an OPEN/CLOSE around A */
8127 U32 paren = ST.me->flags;
8128 if (paren > maxopenparen)
8129 maxopenparen = paren;
8130 scan += NEXT_OFF(scan); /* Skip former OPEN. */
8138 ST.c1 = CHRTEST_UNINIT;
8141 if (!(ST.minmod ? ARG1(ST.me) : ARG2(ST.me))) /* min/max */
8144 curlym_do_A: /* execute the A in /A{m,n}B/ */
8145 PUSH_YES_STATE_GOTO(CURLYM_A, ST.A, locinput); /* match A */
8146 NOT_REACHED; /* NOTREACHED */
8148 case CURLYM_A: /* we've just matched an A */
8150 /* after first match, determine A's length: u.curlym.alen */
8151 if (ST.count == 1) {
8152 if (reginfo->is_utf8_target) {
8153 char *s = st->locinput;
8154 while (s < locinput) {
8160 ST.alen = locinput - st->locinput;
8163 ST.count = ST.minmod ? ARG1(ST.me) : ARG2(ST.me);
8166 Perl_re_exec_indentf( aTHX_ "CURLYM now matched %" IVdf " times, len=%" IVdf "...\n",
8167 depth, (IV) ST.count, (IV)ST.alen)
8170 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8174 I32 max = (ST.minmod ? ARG1(ST.me) : ARG2(ST.me));
8175 if ( max == REG_INFTY || ST.count < max )
8176 goto curlym_do_A; /* try to match another A */
8178 goto curlym_do_B; /* try to match B */
8180 case CURLYM_A_fail: /* just failed to match an A */
8181 REGCP_UNWIND(ST.cp);
8184 if (ST.minmod || ST.count < ARG1(ST.me) /* min*/
8185 || EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8188 curlym_do_B: /* execute the B in /A{m,n}B/ */
8189 if (ST.c1 == CHRTEST_UNINIT) {
8190 /* calculate c1 and c2 for possible match of 1st char
8191 * following curly */
8192 ST.c1 = ST.c2 = CHRTEST_VOID;
8194 if (HAS_TEXT(ST.B) || JUMPABLE(ST.B)) {
8195 regnode *text_node = ST.B;
8196 if (! HAS_TEXT(text_node))
8197 FIND_NEXT_IMPT(text_node);
8198 if (PL_regkind[OP(text_node)] == EXACT) {
8199 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
8200 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
8210 Perl_re_exec_indentf( aTHX_ "CURLYM trying tail with matches=%" IVdf "...\n",
8211 depth, (IV)ST.count)
8213 if (! NEXTCHR_IS_EOS && ST.c1 != CHRTEST_VOID) {
8214 if (! UTF8_IS_INVARIANT(nextchr) && utf8_target) {
8215 if (memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
8216 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
8218 /* simulate B failing */
8220 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%" UVXf " c1=0x%" UVXf " c2=0x%" UVXf "\n",
8222 valid_utf8_to_uvchr((U8 *) locinput, NULL),
8223 valid_utf8_to_uvchr(ST.c1_utf8, NULL),
8224 valid_utf8_to_uvchr(ST.c2_utf8, NULL))
8226 state_num = CURLYM_B_fail;
8227 goto reenter_switch;
8230 else if (nextchr != ST.c1 && nextchr != ST.c2) {
8231 /* simulate B failing */
8233 Perl_re_exec_indentf( aTHX_ "CURLYM Fast bail next target=0x%X c1=0x%X c2=0x%X\n",
8235 (int) nextchr, ST.c1, ST.c2)
8237 state_num = CURLYM_B_fail;
8238 goto reenter_switch;
8243 /* emulate CLOSE: mark current A as captured */
8244 U32 paren = (U32)ST.me->flags;
8246 CLOSE_CAPTURE(paren,
8247 HOPc(locinput, -ST.alen) - reginfo->strbeg,
8248 locinput - reginfo->strbeg);
8251 rex->offs[paren].end = -1;
8253 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.me->flags))
8262 PUSH_STATE_GOTO(CURLYM_B, ST.B, locinput); /* match B */
8263 NOT_REACHED; /* NOTREACHED */
8265 case CURLYM_B_fail: /* just failed to match a B */
8266 REGCP_UNWIND(ST.cp);
8267 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8269 I32 max = ARG2(ST.me);
8270 if (max != REG_INFTY && ST.count == max)
8272 goto curlym_do_A; /* try to match a further A */
8274 /* backtrack one A */
8275 if (ST.count == ARG1(ST.me) /* min */)
8278 SET_locinput(HOPc(locinput, -ST.alen));
8279 goto curlym_do_B; /* try to match B */
8282 #define ST st->u.curly
8284 #define CURLY_SETPAREN(paren, success) \
8287 CLOSE_CAPTURE(paren, HOPc(locinput, -1) - reginfo->strbeg, \
8288 locinput - reginfo->strbeg); \
8291 rex->offs[paren].end = -1; \
8292 rex->lastparen = ST.lastparen; \
8293 rex->lastcloseparen = ST.lastcloseparen; \
8297 case STAR: /* /A*B/ where A is width 1 char */
8301 scan = NEXTOPER(scan);
8304 case PLUS: /* /A+B/ where A is width 1 char */
8308 scan = NEXTOPER(scan);
8311 case CURLYN: /* /(A){m,n}B/ where A is width 1 char */
8312 ST.paren = scan->flags; /* Which paren to set */
8313 ST.lastparen = rex->lastparen;
8314 ST.lastcloseparen = rex->lastcloseparen;
8315 if (ST.paren > maxopenparen)
8316 maxopenparen = ST.paren;
8317 ST.min = ARG1(scan); /* min to match */
8318 ST.max = ARG2(scan); /* max to match */
8319 scan = regnext(NEXTOPER(scan) + NODE_STEP_REGNODE);
8321 /* handle the single-char capture called as a GOSUB etc */
8322 if (EVAL_CLOSE_PAREN_IS_TRUE(cur_eval,(U32)ST.paren))
8324 char *li = locinput;
8325 if (!regrepeat(rex, &li, scan, reginfo, 1))
8333 case CURLY: /* /A{m,n}B/ where A is width 1 char */
8335 ST.min = ARG1(scan); /* min to match */
8336 ST.max = ARG2(scan); /* max to match */
8337 scan = NEXTOPER(scan) + NODE_STEP_REGNODE;
8340 * Lookahead to avoid useless match attempts
8341 * when we know what character comes next.
8343 * Used to only do .*x and .*?x, but now it allows
8344 * for )'s, ('s and (?{ ... })'s to be in the way
8345 * of the quantifier and the EXACT-like node. -- japhy
8348 assert(ST.min <= ST.max);
8349 if (! HAS_TEXT(next) && ! JUMPABLE(next)) {
8350 ST.c1 = ST.c2 = CHRTEST_VOID;
8353 regnode *text_node = next;
8355 if (! HAS_TEXT(text_node))
8356 FIND_NEXT_IMPT(text_node);
8358 if (! HAS_TEXT(text_node))
8359 ST.c1 = ST.c2 = CHRTEST_VOID;
8361 if ( PL_regkind[OP(text_node)] != EXACT ) {
8362 ST.c1 = ST.c2 = CHRTEST_VOID;
8365 if (! S_setup_EXACTISH_ST_c1_c2(aTHX_
8366 text_node, &ST.c1, ST.c1_utf8, &ST.c2, ST.c2_utf8,
8378 char *li = locinput;
8381 regrepeat(rex, &li, ST.A, reginfo, ST.min)
8387 if (ST.c1 == CHRTEST_VOID)
8388 goto curly_try_B_min;
8390 ST.oldloc = locinput;
8392 /* set ST.maxpos to the furthest point along the
8393 * string that could possibly match */
8394 if (ST.max == REG_INFTY) {
8395 ST.maxpos = reginfo->strend - 1;
8397 while (UTF8_IS_CONTINUATION(*(U8*)ST.maxpos))
8400 else if (utf8_target) {
8401 int m = ST.max - ST.min;
8402 for (ST.maxpos = locinput;
8403 m >0 && ST.maxpos < reginfo->strend; m--)
8404 ST.maxpos += UTF8SKIP(ST.maxpos);
8407 ST.maxpos = locinput + ST.max - ST.min;
8408 if (ST.maxpos >= reginfo->strend)
8409 ST.maxpos = reginfo->strend - 1;
8411 goto curly_try_B_min_known;
8415 /* avoid taking address of locinput, so it can remain
8417 char *li = locinput;
8418 ST.count = regrepeat(rex, &li, ST.A, reginfo, ST.max);
8419 if (ST.count < ST.min)
8422 if ((ST.count > ST.min)
8423 && (PL_regkind[OP(ST.B)] == EOL) && (OP(ST.B) != MEOL))
8425 /* A{m,n} must come at the end of the string, there's
8426 * no point in backing off ... */
8428 /* ...except that $ and \Z can match before *and* after
8429 newline at the end. Consider "\n\n" =~ /\n+\Z\n/.
8430 We may back off by one in this case. */
8431 if (UCHARAT(locinput - 1) == '\n' && OP(ST.B) != EOS)
8435 goto curly_try_B_max;
8437 NOT_REACHED; /* NOTREACHED */
8439 case CURLY_B_min_fail:
8440 /* failed to find B in a non-greedy match.
8441 * Handles both cases where c1,c2 valid or not */
8443 REGCP_UNWIND(ST.cp);
8445 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8448 if (ST.c1 == CHRTEST_VOID) {
8449 /* failed -- move forward one */
8450 char *li = locinput;
8451 if (!regrepeat(rex, &li, ST.A, reginfo, 1)) {
8456 if (!( ST.count <= ST.max
8457 /* count overflow ? */
8458 || (ST.max == REG_INFTY && ST.count > 0))
8464 /* Couldn't or didn't -- move forward. */
8465 ST.oldloc = locinput;
8467 locinput += UTF8SKIP(locinput);
8472 curly_try_B_min_known:
8473 /* find the next place where 'B' could work, then call B */
8475 n = (ST.oldloc == locinput) ? 0 : 1;
8476 if (ST.c1 == ST.c2) {
8477 /* set n to utf8_distance(oldloc, locinput) */
8478 while (locinput <= ST.maxpos
8479 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput)))
8481 locinput += UTF8SKIP(locinput);
8486 /* set n to utf8_distance(oldloc, locinput) */
8487 while (locinput <= ST.maxpos
8488 && memNE(locinput, ST.c1_utf8, UTF8SKIP(locinput))
8489 && memNE(locinput, ST.c2_utf8, UTF8SKIP(locinput)))
8491 locinput += UTF8SKIP(locinput);
8496 else { /* Not utf8_target */
8497 if (ST.c1 == ST.c2) {
8498 locinput = (char *) memchr(locinput,
8500 ST.maxpos + 1 - locinput);
8502 locinput = ST.maxpos + 1;
8506 U8 c1_c2_bits_differing = ST.c1 ^ ST.c2;
8508 if (! isPOWER_OF_2(c1_c2_bits_differing)) {
8509 while ( locinput <= ST.maxpos
8510 && UCHARAT(locinput) != ST.c1
8511 && UCHARAT(locinput) != ST.c2)
8517 /* If c1 and c2 only differ by a single bit, we can
8518 * avoid a conditional each time through the loop,
8519 * at the expense of a little preliminary setup and
8520 * an extra mask each iteration. By masking out
8521 * that bit, we match exactly two characters, c1
8522 * and c2, and so we don't have to test for both.
8523 * On both ASCII and EBCDIC platforms, most of the
8524 * ASCII-range and Latin1-range folded equivalents
8525 * differ only in a single bit, so this is actually
8526 * the most common case. (e.g. 'A' 0x41 vs 'a'
8528 U8 c1_masked = ST.c1 &~ c1_c2_bits_differing;
8529 U8 c1_c2_mask = ~ c1_c2_bits_differing;
8530 while ( locinput <= ST.maxpos
8531 && (UCHARAT(locinput) & c1_c2_mask)
8538 n = locinput - ST.oldloc;
8540 if (locinput > ST.maxpos)
8543 /* In /a{m,n}b/, ST.oldloc is at "a" x m, locinput is
8544 * at b; check that everything between oldloc and
8545 * locinput matches */
8546 char *li = ST.oldloc;
8548 if (regrepeat(rex, &li, ST.A, reginfo, n) < n)
8550 assert(n == REG_INFTY || locinput == li);
8555 CURLY_SETPAREN(ST.paren, ST.count);
8556 PUSH_STATE_GOTO(CURLY_B_min, ST.B, locinput);
8557 NOT_REACHED; /* NOTREACHED */
8561 /* a successful greedy match: now try to match B */
8563 bool could_match = locinput < reginfo->strend;
8565 /* If it could work, try it. */
8566 if (ST.c1 != CHRTEST_VOID && could_match) {
8567 if (! UTF8_IS_INVARIANT(UCHARAT(locinput)) && utf8_target)
8569 could_match = memEQ(locinput,
8574 UTF8SKIP(locinput));
8577 could_match = UCHARAT(locinput) == ST.c1
8578 || UCHARAT(locinput) == ST.c2;
8581 if (ST.c1 == CHRTEST_VOID || could_match) {
8582 CURLY_SETPAREN(ST.paren, ST.count);
8583 PUSH_STATE_GOTO(CURLY_B_max, ST.B, locinput);
8584 NOT_REACHED; /* NOTREACHED */
8589 case CURLY_B_max_fail:
8590 /* failed to find B in a greedy match */
8592 REGCP_UNWIND(ST.cp);
8594 UNWIND_PAREN(ST.lastparen, ST.lastcloseparen);
8597 if (--ST.count < ST.min)
8599 locinput = HOPc(locinput, -1);
8600 goto curly_try_B_max;
8604 case END: /* last op of main pattern */
8607 /* we've just finished A in /(??{A})B/; now continue with B */
8608 SET_RECURSE_LOCINPUT("FAKE-END[before]", CUR_EVAL.prev_recurse_locinput);
8609 st->u.eval.prev_rex = rex_sv; /* inner */
8611 /* Save *all* the positions. */
8612 st->u.eval.cp = regcppush(rex, 0, maxopenparen);
8613 rex_sv = CUR_EVAL.prev_rex;
8614 is_utf8_pat = reginfo->is_utf8_pat = cBOOL(RX_UTF8(rex_sv));
8615 SET_reg_curpm(rex_sv);
8616 rex = ReANY(rex_sv);
8617 rexi = RXi_GET(rex);
8619 st->u.eval.prev_curlyx = cur_curlyx;
8620 cur_curlyx = CUR_EVAL.prev_curlyx;
8622 REGCP_SET(st->u.eval.lastcp);
8624 /* Restore parens of the outer rex without popping the
8626 regcp_restore(rex, CUR_EVAL.lastcp, &maxopenparen);
8628 st->u.eval.prev_eval = cur_eval;
8629 cur_eval = CUR_EVAL.prev_eval;
8631 Perl_re_exec_indentf( aTHX_ "END: EVAL trying tail ... (cur_eval=%p)\n",
8633 if ( nochange_depth )
8636 SET_RECURSE_LOCINPUT("FAKE-END[after]", cur_eval->locinput);
8638 PUSH_YES_STATE_GOTO(EVAL_postponed_AB, st->u.eval.prev_eval->u.eval.B,
8639 locinput); /* match B */
8642 if (locinput < reginfo->till) {
8643 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_
8644 "%sEND: Match possible, but length=%ld is smaller than requested=%ld, failing!%s\n",
8646 (long)(locinput - startpos),
8647 (long)(reginfo->till - startpos),
8650 sayNO_SILENT; /* Cannot match: too short. */
8652 sayYES; /* Success! */
8654 case SUCCEED: /* successful SUSPEND/UNLESSM/IFMATCH/CURLYM */
8656 Perl_re_exec_indentf( aTHX_ "%sSUCCEED: subpattern success...%s\n",
8657 depth, PL_colors[4], PL_colors[5]));
8658 sayYES; /* Success! */
8661 #define ST st->u.ifmatch
8666 case SUSPEND: /* (?>A) */
8668 newstart = locinput;
8671 case UNLESSM: /* -ve lookaround: (?!A), or with flags, (?<!A) */
8673 goto ifmatch_trivial_fail_test;
8675 case IFMATCH: /* +ve lookaround: (?=A), or with flags, (?<=A) */
8677 ifmatch_trivial_fail_test:
8679 char * const s = HOPBACKc(locinput, scan->flags);
8684 sw = 1 - cBOOL(ST.wanted);
8688 next = scan + ARG(scan);
8696 newstart = locinput;
8700 ST.logical = logical;
8701 logical = 0; /* XXX: reset state of logical once it has been saved into ST */
8703 /* execute body of (?...A) */
8704 PUSH_YES_STATE_GOTO(IFMATCH_A, NEXTOPER(NEXTOPER(scan)), newstart);
8705 NOT_REACHED; /* NOTREACHED */
8708 case IFMATCH_A_fail: /* body of (?...A) failed */
8709 ST.wanted = !ST.wanted;
8712 case IFMATCH_A: /* body of (?...A) succeeded */
8714 sw = cBOOL(ST.wanted);
8716 else if (!ST.wanted)
8719 if (OP(ST.me) != SUSPEND) {
8720 /* restore old position except for (?>...) */
8721 locinput = st->locinput;
8723 scan = ST.me + ARG(ST.me);
8726 continue; /* execute B */
8730 case LONGJMP: /* alternative with many branches compiles to
8731 * (BRANCHJ; EXACT ...; LONGJMP ) x N */
8732 next = scan + ARG(scan);
8737 case COMMIT: /* (*COMMIT) */
8738 reginfo->cutpoint = reginfo->strend;
8741 case PRUNE: /* (*PRUNE) */
8743 sv_yes_mark = sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8744 PUSH_STATE_GOTO(COMMIT_next, next, locinput);
8745 NOT_REACHED; /* NOTREACHED */
8747 case COMMIT_next_fail:
8751 NOT_REACHED; /* NOTREACHED */
8753 case OPFAIL: /* (*FAIL) */
8755 sv_commit = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8757 /* deal with (?(?!)X|Y) properly,
8758 * make sure we trigger the no branch
8759 * of the trailing IFTHEN structure*/
8765 NOT_REACHED; /* NOTREACHED */
8767 #define ST st->u.mark
8768 case MARKPOINT: /* (*MARK:foo) */
8769 ST.prev_mark = mark_state;
8770 ST.mark_name = sv_commit = sv_yes_mark
8771 = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8773 ST.mark_loc = locinput;
8774 PUSH_YES_STATE_GOTO(MARKPOINT_next, next, locinput);
8775 NOT_REACHED; /* NOTREACHED */
8777 case MARKPOINT_next:
8778 mark_state = ST.prev_mark;
8780 NOT_REACHED; /* NOTREACHED */
8782 case MARKPOINT_next_fail:
8783 if (popmark && sv_eq(ST.mark_name,popmark))
8785 if (ST.mark_loc > startpoint)
8786 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8787 popmark = NULL; /* we found our mark */
8788 sv_commit = ST.mark_name;
8791 Perl_re_exec_indentf( aTHX_ "%sMARKPOINT: next fail: setting cutpoint to mark:%" SVf "...%s\n",
8793 PL_colors[4], SVfARG(sv_commit), PL_colors[5]);
8796 mark_state = ST.prev_mark;
8797 sv_yes_mark = mark_state ?
8798 mark_state->u.mark.mark_name : NULL;
8800 NOT_REACHED; /* NOTREACHED */
8802 case SKIP: /* (*SKIP) */
8804 /* (*SKIP) : if we fail we cut here*/
8805 ST.mark_name = NULL;
8806 ST.mark_loc = locinput;
8807 PUSH_STATE_GOTO(SKIP_next,next, locinput);
8809 /* (*SKIP:NAME) : if there is a (*MARK:NAME) fail where it was,
8810 otherwise do nothing. Meaning we need to scan
8812 regmatch_state *cur = mark_state;
8813 SV *find = MUTABLE_SV(rexi->data->data[ ARG( scan ) ]);
8816 if ( sv_eq( cur->u.mark.mark_name,
8819 ST.mark_name = find;
8820 PUSH_STATE_GOTO( SKIP_next, next, locinput);
8822 cur = cur->u.mark.prev_mark;
8825 /* Didn't find our (*MARK:NAME) so ignore this (*SKIP:NAME) */
8828 case SKIP_next_fail:
8830 /* (*CUT:NAME) - Set up to search for the name as we
8831 collapse the stack*/
8832 popmark = ST.mark_name;
8834 /* (*CUT) - No name, we cut here.*/
8835 if (ST.mark_loc > startpoint)
8836 reginfo->cutpoint = HOPBACKc(ST.mark_loc, 1);
8837 /* but we set sv_commit to latest mark_name if there
8838 is one so they can test to see how things lead to this
8841 sv_commit=mark_state->u.mark.mark_name;
8845 NOT_REACHED; /* NOTREACHED */
8848 case LNBREAK: /* \R */
8849 if ((n=is_LNBREAK_safe(locinput, reginfo->strend, utf8_target))) {
8856 PerlIO_printf(Perl_error_log, "%" UVxf " %d\n",
8857 PTR2UV(scan), OP(scan));
8858 Perl_croak(aTHX_ "regexp memory corruption");
8860 /* this is a point to jump to in order to increment
8861 * locinput by one character */
8863 assert(!NEXTCHR_IS_EOS);
8865 locinput += PL_utf8skip[nextchr];
8866 /* locinput is allowed to go 1 char off the end (signifying
8867 * EOS), but not 2+ */
8868 if (locinput > reginfo->strend)
8877 /* switch break jumps here */
8878 scan = next; /* prepare to execute the next op and ... */
8879 continue; /* ... jump back to the top, reusing st */
8883 /* push a state that backtracks on success */
8884 st->u.yes.prev_yes_state = yes_state;
8888 /* push a new regex state, then continue at scan */
8890 regmatch_state *newst;
8893 regmatch_state *cur = st;
8894 regmatch_state *curyes = yes_state;
8896 regmatch_slab *slab = PL_regmatch_slab;
8897 for (i = 0; i < 3 && i <= depth; cur--,i++) {
8898 if (cur < SLAB_FIRST(slab)) {
8900 cur = SLAB_LAST(slab);
8902 Perl_re_exec_indentf( aTHX_ "%4s #%-3d %-10s %s\n",
8905 depth - i, PL_reg_name[cur->resume_state],
8906 (curyes == cur) ? "yes" : ""
8909 curyes = cur->u.yes.prev_yes_state;
8912 DEBUG_STATE_pp("push")
8915 st->locinput = locinput;
8917 if (newst > SLAB_LAST(PL_regmatch_slab))
8918 newst = S_push_slab(aTHX);
8919 PL_regmatch_state = newst;
8921 locinput = pushinput;
8927 #ifdef SOLARIS_BAD_OPTIMIZER
8928 # undef PL_charclass
8932 * We get here only if there's trouble -- normally "case END" is
8933 * the terminating point.
8935 Perl_croak(aTHX_ "corrupted regexp pointers");
8936 NOT_REACHED; /* NOTREACHED */
8940 /* we have successfully completed a subexpression, but we must now
8941 * pop to the state marked by yes_state and continue from there */
8942 assert(st != yes_state);
8944 while (st != yes_state) {
8946 if (st < SLAB_FIRST(PL_regmatch_slab)) {
8947 PL_regmatch_slab = PL_regmatch_slab->prev;
8948 st = SLAB_LAST(PL_regmatch_slab);
8952 DEBUG_STATE_pp("pop (no final)");
8954 DEBUG_STATE_pp("pop (yes)");
8960 while (yes_state < SLAB_FIRST(PL_regmatch_slab)
8961 || yes_state > SLAB_LAST(PL_regmatch_slab))
8963 /* not in this slab, pop slab */
8964 depth -= (st - SLAB_FIRST(PL_regmatch_slab) + 1);
8965 PL_regmatch_slab = PL_regmatch_slab->prev;
8966 st = SLAB_LAST(PL_regmatch_slab);
8968 depth -= (st - yes_state);
8971 yes_state = st->u.yes.prev_yes_state;
8972 PL_regmatch_state = st;
8975 locinput= st->locinput;
8976 state_num = st->resume_state + no_final;
8977 goto reenter_switch;
8980 DEBUG_EXECUTE_r(Perl_re_printf( aTHX_ "%sMatch successful!%s\n",
8981 PL_colors[4], PL_colors[5]));
8983 if (reginfo->info_aux_eval) {
8984 /* each successfully executed (?{...}) block does the equivalent of
8985 * local $^R = do {...}
8986 * When popping the save stack, all these locals would be undone;
8987 * bypass this by setting the outermost saved $^R to the latest
8989 /* I dont know if this is needed or works properly now.
8990 * see code related to PL_replgv elsewhere in this file.
8993 if (oreplsv != GvSV(PL_replgv)) {
8994 sv_setsv(oreplsv, GvSV(PL_replgv));
8995 SvSETMAGIC(oreplsv);
9003 Perl_re_exec_indentf( aTHX_ "%sfailed...%s\n",
9005 PL_colors[4], PL_colors[5])
9017 /* there's a previous state to backtrack to */
9019 if (st < SLAB_FIRST(PL_regmatch_slab)) {
9020 PL_regmatch_slab = PL_regmatch_slab->prev;
9021 st = SLAB_LAST(PL_regmatch_slab);
9023 PL_regmatch_state = st;
9024 locinput= st->locinput;
9026 DEBUG_STATE_pp("pop");
9028 if (yes_state == st)
9029 yes_state = st->u.yes.prev_yes_state;
9031 state_num = st->resume_state + 1; /* failure = success + 1 */
9033 goto reenter_switch;
9038 if (rex->intflags & PREGf_VERBARG_SEEN) {
9039 SV *sv_err = get_sv("REGERROR", 1);
9040 SV *sv_mrk = get_sv("REGMARK", 1);
9042 sv_commit = &PL_sv_no;
9044 sv_yes_mark = &PL_sv_yes;
9047 sv_commit = &PL_sv_yes;
9048 sv_yes_mark = &PL_sv_no;
9052 sv_setsv(sv_err, sv_commit);
9053 sv_setsv(sv_mrk, sv_yes_mark);
9057 if (last_pushed_cv) {
9059 /* see "Some notes about MULTICALL" above */
9061 PERL_UNUSED_VAR(SP);
9064 LEAVE_SCOPE(orig_savestack_ix);
9066 assert(!result || locinput - reginfo->strbeg >= 0);
9067 return result ? locinput - reginfo->strbeg : -1;
9071 - regrepeat - repeatedly match something simple, report how many
9073 * What 'simple' means is a node which can be the operand of a quantifier like
9076 * startposp - pointer a pointer to the start position. This is updated
9077 * to point to the byte following the highest successful
9079 * p - the regnode to be repeatedly matched against.
9080 * reginfo - struct holding match state, such as strend
9081 * max - maximum number of things to match.
9082 * depth - (for debugging) backtracking depth.
9085 S_regrepeat(pTHX_ regexp *prog, char **startposp, const regnode *p,
9086 regmatch_info *const reginfo, I32 max _pDEPTH)
9089 char *scan; /* Pointer to current position in target string */
9091 char *loceol = reginfo->strend; /* local version */
9092 I32 hardcount = 0; /* How many matches so far */
9093 bool utf8_target = reginfo->is_utf8_target;
9094 unsigned int to_complement = 0; /* Invert the result? */
9096 _char_class_number classnum;
9098 PERL_ARGS_ASSERT_REGREPEAT;
9101 if (max == REG_INFTY)
9103 else if (! utf8_target && loceol - scan > max)
9104 loceol = scan + max;
9106 /* Here, for the case of a non-UTF-8 target we have adjusted <loceol> down
9107 * to the maximum of how far we should go in it (leaving it set to the real
9108 * end, if the maximum permissible would take us beyond that). This allows
9109 * us to make the loop exit condition that we haven't gone past <loceol> to
9110 * also mean that we haven't exceeded the max permissible count, saving a
9111 * test each time through the loop. But it assumes that the OP matches a
9112 * single byte, which is true for most of the OPs below when applied to a
9113 * non-UTF-8 target. Those relatively few OPs that don't have this
9114 * characteristic will have to compensate.
9116 * There is no adjustment for UTF-8 targets, as the number of bytes per
9117 * character varies. OPs will have to test both that the count is less
9118 * than the max permissible (using <hardcount> to keep track), and that we
9119 * are still within the bounds of the string (using <loceol>. A few OPs
9120 * match a single byte no matter what the encoding. They can omit the max
9121 * test if, for the UTF-8 case, they do the adjustment that was skipped
9124 * Thus, the code above sets things up for the common case; and exceptional
9125 * cases need extra work; the common case is to make sure <scan> doesn't
9126 * go past <loceol>, and for UTF-8 to also use <hardcount> to make sure the
9127 * count doesn't exceed the maximum permissible */
9132 while (scan < loceol && hardcount < max && *scan != '\n') {
9133 scan += UTF8SKIP(scan);
9137 scan = (char *) memchr(scan, '\n', loceol - scan);
9145 while (scan < loceol && hardcount < max) {
9146 scan += UTF8SKIP(scan);
9154 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9155 if (utf8_target && UTF8_IS_ABOVE_LATIN1(*scan)) {
9156 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(scan, loceol);
9161 if (! utf8_target) {
9167 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
9171 /* Can use a simple find if the pattern char to match on is invariant
9172 * under UTF-8, or both target and pattern aren't UTF-8. Note that we
9173 * can use UTF8_IS_INVARIANT() even if the pattern isn't UTF-8, as it's
9174 * true iff it doesn't matter if the argument is in UTF-8 or not */
9175 if (UTF8_IS_INVARIANT(c) || (! utf8_target && ! reginfo->is_utf8_pat)) {
9176 if (utf8_target && loceol - scan > max) {
9177 /* We didn't adjust <loceol> because is UTF-8, but ok to do so,
9178 * since here, to match at all, 1 char == 1 byte */
9179 loceol = scan + max;
9181 scan = (char *) find_span_end((U8 *) scan, (U8 *) loceol, (U8) c);
9183 else if (reginfo->is_utf8_pat) {
9185 STRLEN scan_char_len;
9187 /* When both target and pattern are UTF-8, we have to do
9189 while (hardcount < max
9191 && (scan_char_len = UTF8SKIP(scan)) <= STR_LEN(p)
9192 && memEQ(scan, STRING(p), scan_char_len))
9194 scan += scan_char_len;
9198 else if (! UTF8_IS_ABOVE_LATIN1(c)) {
9200 /* Target isn't utf8; convert the character in the UTF-8
9201 * pattern to non-UTF8, and do a simple find */
9202 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *(STRING(p) + 1));
9203 scan = (char *) find_span_end((U8 *) scan, (U8 *) loceol, (U8) c);
9204 } /* else pattern char is above Latin1, can't possibly match the
9209 /* Here, the string must be utf8; pattern isn't, and <c> is
9210 * different in utf8 than not, so can't compare them directly.
9211 * Outside the loop, find the two utf8 bytes that represent c, and
9212 * then look for those in sequence in the utf8 string */
9213 U8 high = UTF8_TWO_BYTE_HI(c);
9214 U8 low = UTF8_TWO_BYTE_LO(c);
9216 while (hardcount < max
9217 && scan + 1 < loceol
9218 && UCHARAT(scan) == high
9219 && UCHARAT(scan + 1) == low)
9227 case EXACTFAA_NO_TRIE: /* This node only generated for non-utf8 patterns */
9228 assert(! reginfo->is_utf8_pat);
9231 utf8_flags = FOLDEQ_UTF8_NOMIX_ASCII;
9232 if (reginfo->is_utf8_pat || ! utf8_target) {
9234 /* The possible presence of a MICRO SIGN in the pattern forbids us
9235 * to view a non-UTF-8 pattern as folded when there is a UTF-8
9237 utf8_flags |= FOLDEQ_S2_ALREADY_FOLDED|FOLDEQ_S2_FOLDS_SANE;
9242 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9243 utf8_flags = FOLDEQ_LOCALE;
9246 case EXACTF: /* This node only generated for non-utf8 patterns */
9247 assert(! reginfo->is_utf8_pat);
9251 if (! utf8_target) {
9254 utf8_flags = FOLDEQ_LOCALE | FOLDEQ_S2_ALREADY_FOLDED
9255 | FOLDEQ_S2_FOLDS_SANE;
9259 if (! utf8_target) {
9262 assert(reginfo->is_utf8_pat);
9263 utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
9267 utf8_flags = FOLDEQ_S2_ALREADY_FOLDED;
9274 U8 c1_utf8[UTF8_MAXBYTES+1], c2_utf8[UTF8_MAXBYTES+1];
9276 assert(STR_LEN(p) == reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1);
9278 if (S_setup_EXACTISH_ST_c1_c2(aTHX_ p, &c1, c1_utf8, &c2, c2_utf8,
9281 if (c1 == CHRTEST_VOID) {
9282 /* Use full Unicode fold matching */
9283 char *tmpeol = reginfo->strend;
9284 STRLEN pat_len = reginfo->is_utf8_pat ? UTF8SKIP(STRING(p)) : 1;
9285 while (hardcount < max
9286 && foldEQ_utf8_flags(scan, &tmpeol, 0, utf8_target,
9287 STRING(p), NULL, pat_len,
9288 reginfo->is_utf8_pat, utf8_flags))
9291 tmpeol = reginfo->strend;
9295 else if (utf8_target) {
9297 while (scan < loceol
9299 && memEQ(scan, c1_utf8, UTF8SKIP(scan)))
9301 scan += UTF8SKIP(scan);
9306 while (scan < loceol
9308 && (memEQ(scan, c1_utf8, UTF8SKIP(scan))
9309 || memEQ(scan, c2_utf8, UTF8SKIP(scan))))
9311 scan += UTF8SKIP(scan);
9316 else if (c1 == c2) {
9317 scan = (char *) find_span_end((U8 *) scan, (U8 *) loceol, (U8) c1);
9320 /* See comments in regmatch() CURLY_B_min_known_fail. We avoid
9321 * a conditional each time through the loop if the characters
9322 * differ only in a single bit, as is the usual situation */
9323 U8 c1_c2_bits_differing = c1 ^ c2;
9325 if (isPOWER_OF_2(c1_c2_bits_differing)) {
9326 U8 c1_c2_mask = ~ c1_c2_bits_differing;
9328 scan = (char *) find_span_end_mask((U8 *) scan,
9334 while ( scan < loceol
9335 && (UCHARAT(scan) == c1 || UCHARAT(scan) == c2))
9346 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9348 if (ANYOFL_UTF8_LOCALE_REQD(FLAGS(p)) && ! IN_UTF8_CTYPE_LOCALE) {
9349 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE), utf8_locale_required);
9355 while (hardcount < max
9357 && reginclass(prog, p, (U8*)scan, (U8*) loceol, utf8_target))
9359 scan += UTF8SKIP(scan);
9363 else if (ANYOF_FLAGS(p) & ~ ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
9364 while (scan < loceol
9365 && reginclass(prog, p, (U8*)scan, (U8*)scan+1, 0))
9369 while (scan < loceol && ANYOF_BITMAP_TEST(p, *((U8*)scan)))
9375 if (utf8_target && loceol - scan > max) {
9377 /* We didn't adjust <loceol> at the beginning of this routine
9378 * because is UTF-8, but it is actually ok to do so, since here, to
9379 * match, 1 char == 1 byte. */
9380 loceol = scan + max;
9383 scan = (char *) find_span_end_mask((U8 *) scan, (U8 *) loceol, (U8) ARG(p), FLAGS(p));
9388 while ( hardcount < max
9390 && (*scan & FLAGS(p)) != ARG(p))
9392 scan += UTF8SKIP(scan);
9397 scan = (char *) find_next_masked((U8 *) scan, (U8 *) loceol, (U8) ARG(p), FLAGS(p));
9402 if (utf8_target) while ( hardcount < max
9404 && reginclass(prog, p, (U8*)scan, (U8*) loceol,
9407 scan += UTF8SKIP(scan);
9412 /* The argument (FLAGS) to all the POSIX node types is the class number */
9419 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9420 if (! utf8_target) {
9421 while (scan < loceol && to_complement ^ cBOOL(isFOO_lc(FLAGS(p),
9427 while (hardcount < max && scan < loceol
9428 && to_complement ^ cBOOL(isFOO_utf8_lc(FLAGS(p),
9432 scan += UTF8SKIP(scan);
9445 if (utf8_target && loceol - scan > max) {
9447 /* We didn't adjust <loceol> at the beginning of this routine
9448 * because is UTF-8, but it is actually ok to do so, since here, to
9449 * match, 1 char == 1 byte. */
9450 loceol = scan + max;
9452 while (scan < loceol && _generic_isCC_A((U8) *scan, FLAGS(p))) {
9465 if (! utf8_target) {
9466 while (scan < loceol && ! _generic_isCC_A((U8) *scan, FLAGS(p))) {
9472 /* The complement of something that matches only ASCII matches all
9473 * non-ASCII, plus everything in ASCII that isn't in the class. */
9474 while (hardcount < max && scan < loceol
9475 && ( ! isASCII_utf8_safe(scan, reginfo->strend)
9476 || ! _generic_isCC_A((U8) *scan, FLAGS(p))))
9478 scan += UTF8SKIP(scan);
9489 if (! utf8_target) {
9490 while (scan < loceol && to_complement
9491 ^ cBOOL(_generic_isCC((U8) *scan, FLAGS(p))))
9498 classnum = (_char_class_number) FLAGS(p);
9501 while ( hardcount < max && scan < loceol
9502 && to_complement ^ cBOOL(_invlist_contains_cp(
9503 PL_XPosix_ptrs[classnum],
9504 utf8_to_uvchr_buf((U8 *) scan,
9508 scan += UTF8SKIP(scan);
9513 /* For the classes below, the knowledge of how to handle
9514 * every code point is compiled in to Perl via a macro.
9515 * This code is written for making the loops as tight as
9516 * possible. It could be refactored to save space instead.
9519 case _CC_ENUM_SPACE:
9520 while (hardcount < max
9523 ^ cBOOL(isSPACE_utf8_safe(scan, loceol))))
9525 scan += UTF8SKIP(scan);
9529 case _CC_ENUM_BLANK:
9530 while (hardcount < max
9533 ^ cBOOL(isBLANK_utf8_safe(scan, loceol))))
9535 scan += UTF8SKIP(scan);
9539 case _CC_ENUM_XDIGIT:
9540 while (hardcount < max
9543 ^ cBOOL(isXDIGIT_utf8_safe(scan, loceol))))
9545 scan += UTF8SKIP(scan);
9549 case _CC_ENUM_VERTSPACE:
9550 while (hardcount < max
9553 ^ cBOOL(isVERTWS_utf8_safe(scan, loceol))))
9555 scan += UTF8SKIP(scan);
9559 case _CC_ENUM_CNTRL:
9560 while (hardcount < max
9563 ^ cBOOL(isCNTRL_utf8_safe(scan, loceol))))
9565 scan += UTF8SKIP(scan);
9575 while (hardcount < max && scan < loceol &&
9576 (c=is_LNBREAK_utf8_safe(scan, loceol))) {
9581 /* LNBREAK can match one or two latin chars, which is ok, but we
9582 * have to use hardcount in this situation, and throw away the
9583 * adjustment to <loceol> done before the switch statement */
9584 loceol = reginfo->strend;
9585 while (scan < loceol && (c=is_LNBREAK_latin1_safe(scan, loceol))) {
9594 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
9608 /* These are all 0 width, so match right here or not at all. */
9612 Perl_croak(aTHX_ "panic: regrepeat() called with unrecognized node type %d='%s'", OP(p), PL_reg_name[OP(p)]);
9613 NOT_REACHED; /* NOTREACHED */
9620 c = scan - *startposp;
9624 GET_RE_DEBUG_FLAGS_DECL;
9626 SV * const prop = sv_newmortal();
9627 regprop(prog, prop, p, reginfo, NULL);
9628 Perl_re_exec_indentf( aTHX_ "%s can match %" IVdf " times out of %" IVdf "...\n",
9629 depth, SvPVX_const(prop),(IV)c,(IV)max);
9637 - reginclass - determine if a character falls into a character class
9639 n is the ANYOF-type regnode
9640 p is the target string
9641 p_end points to one byte beyond the end of the target string
9642 utf8_target tells whether p is in UTF-8.
9644 Returns true if matched; false otherwise.
9646 Note that this can be a synthetic start class, a combination of various
9647 nodes, so things you think might be mutually exclusive, such as locale,
9648 aren't. It can match both locale and non-locale
9653 S_reginclass(pTHX_ regexp * const prog, const regnode * const n, const U8* const p, const U8* const p_end, const bool utf8_target)
9656 const char flags = ANYOF_FLAGS(n);
9660 PERL_ARGS_ASSERT_REGINCLASS;
9662 /* If c is not already the code point, get it. Note that
9663 * UTF8_IS_INVARIANT() works even if not in UTF-8 */
9664 if (! UTF8_IS_INVARIANT(c) && utf8_target) {
9666 const U32 utf8n_flags = UTF8_ALLOW_DEFAULT;
9667 c = utf8n_to_uvchr(p, p_end - p, &c_len, utf8n_flags | UTF8_CHECK_ONLY);
9668 if (c_len == (STRLEN)-1) {
9669 _force_out_malformed_utf8_message(p, p_end,
9671 1 /* 1 means die */ );
9672 NOT_REACHED; /* NOTREACHED */
9675 && (OP(n) == ANYOFL || OP(n) == ANYOFPOSIXL)
9676 && ! ANYOFL_UTF8_LOCALE_REQD(flags))
9678 _CHECK_AND_OUTPUT_WIDE_LOCALE_CP_MSG(c);
9682 /* If this character is potentially in the bitmap, check it */
9683 if (c < NUM_ANYOF_CODE_POINTS && OP(n) != ANYOFH) {
9684 if (ANYOF_BITMAP_TEST(n, c))
9687 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9694 else if (flags & ANYOF_LOCALE_FLAGS) {
9695 if ( (flags & ANYOFL_FOLD)
9696 && c < sizeof(PL_fold_locale)
9697 && ANYOF_BITMAP_TEST(n, PL_fold_locale[c]))
9701 else if ( ANYOF_POSIXL_TEST_ANY_SET(n)
9702 && c <= U8_MAX /* param to isFOO_lc() */
9705 /* The data structure is arranged so bits 0, 2, 4, ... are set
9706 * if the class includes the Posix character class given by
9707 * bit/2; and 1, 3, 5, ... are set if the class includes the
9708 * complemented Posix class given by int(bit/2). So we loop
9709 * through the bits, each time changing whether we complement
9710 * the result or not. Suppose for the sake of illustration
9711 * that bits 0-3 mean respectively, \w, \W, \s, \S. If bit 0
9712 * is set, it means there is a match for this ANYOF node if the
9713 * character is in the class given by the expression (0 / 2 = 0
9714 * = \w). If it is in that class, isFOO_lc() will return 1,
9715 * and since 'to_complement' is 0, the result will stay TRUE,
9716 * and we exit the loop. Suppose instead that bit 0 is 0, but
9717 * bit 1 is 1. That means there is a match if the character
9718 * matches \W. We won't bother to call isFOO_lc() on bit 0,
9719 * but will on bit 1. On the second iteration 'to_complement'
9720 * will be 1, so the exclusive or will reverse things, so we
9721 * are testing for \W. On the third iteration, 'to_complement'
9722 * will be 0, and we would be testing for \s; the fourth
9723 * iteration would test for \S, etc.
9725 * Note that this code assumes that all the classes are closed
9726 * under folding. For example, if a character matches \w, then
9727 * its fold does too; and vice versa. This should be true for
9728 * any well-behaved locale for all the currently defined Posix
9729 * classes, except for :lower: and :upper:, which are handled
9730 * by the pseudo-class :cased: which matches if either of the
9731 * other two does. To get rid of this assumption, an outer
9732 * loop could be used below to iterate over both the source
9733 * character, and its fold (if different) */
9736 int to_complement = 0;
9738 while (count < ANYOF_MAX) {
9739 if (ANYOF_POSIXL_TEST(n, count)
9740 && to_complement ^ cBOOL(isFOO_lc(count/2, (U8) c)))
9753 /* If the bitmap didn't (or couldn't) match, and something outside the
9754 * bitmap could match, try that. */
9756 if (c >= NUM_ANYOF_CODE_POINTS
9757 && (flags & ANYOF_MATCHES_ALL_ABOVE_BITMAP))
9759 match = TRUE; /* Everything above the bitmap matches */
9761 /* Here doesn't match everything above the bitmap. If there is
9762 * some information available beyond the bitmap, we may find a
9763 * match in it. If so, this is most likely because the code point
9764 * is outside the bitmap range. But rarely, it could be because of
9765 * some other reason. If so, various flags are set to indicate
9766 * this possibility. On ANYOFD nodes, there may be matches that
9767 * happen only when the target string is UTF-8; or for other node
9768 * types, because runtime lookup is needed, regardless of the
9769 * UTF-8ness of the target string. Finally, under /il, there may
9770 * be some matches only possible if the locale is a UTF-8 one. */
9771 else if ( ARG(n) != ANYOF_ONLY_HAS_BITMAP
9772 && ( c >= NUM_ANYOF_CODE_POINTS
9773 || ( (flags & ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)
9774 && ( UNLIKELY(OP(n) != ANYOFD)
9775 || (utf8_target && ! isASCII_uni(c)
9776 # if NUM_ANYOF_CODE_POINTS > 256
9780 || ( ANYOFL_SOME_FOLDS_ONLY_IN_UTF8_LOCALE(flags)
9781 && IN_UTF8_CTYPE_LOCALE)))
9783 SV* only_utf8_locale = NULL;
9784 SV * const definition = _get_regclass_nonbitmap_data(prog, n, TRUE,
9785 0, &only_utf8_locale, NULL);
9791 } else { /* Convert to utf8 */
9792 utf8_p = utf8_buffer;
9793 append_utf8_from_native_byte(*p, &utf8_p);
9794 utf8_p = utf8_buffer;
9797 /* Turkish locales have these hard-coded rules overriding
9799 if ( UNLIKELY(PL_in_utf8_turkic_locale)
9800 && isALPHA_FOLD_EQ(*p, 'i'))
9803 if (_invlist_contains_cp(definition,
9804 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE))
9809 else if (*p == 'I') {
9810 if (_invlist_contains_cp(definition,
9811 LATIN_SMALL_LETTER_DOTLESS_I))
9817 else if (_invlist_contains_cp(definition, c)) {
9821 if (! match && only_utf8_locale && IN_UTF8_CTYPE_LOCALE) {
9822 match = _invlist_contains_cp(only_utf8_locale, c);
9826 /* In a Turkic locale under folding, hard-code the I i case pair
9828 if ( UNLIKELY(PL_in_utf8_turkic_locale)
9830 && (flags & ANYOFL_FOLD)
9833 if (c == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
9834 if (ANYOF_BITMAP_TEST(n, 'i')) {
9838 else if (c == LATIN_SMALL_LETTER_DOTLESS_I) {
9839 if (ANYOF_BITMAP_TEST(n, 'I')) {
9845 if (UNICODE_IS_SUPER(c)
9847 & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER)
9849 && ckWARN_d(WARN_NON_UNICODE))
9851 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
9852 "Matched non-Unicode code point 0x%04" UVXf " against Unicode property; may not be portable", c);
9856 #if ANYOF_INVERT != 1
9857 /* Depending on compiler optimization cBOOL takes time, so if don't have to
9859 # error ANYOF_INVERT needs to be set to 1, or guarded with cBOOL below,
9862 /* The xor complements the return if to invert: 1^1 = 0, 1^0 = 1 */
9863 return (flags & ANYOF_INVERT) ^ match;
9867 S_reghop3(U8 *s, SSize_t off, const U8* lim)
9869 /* return the position 'off' UTF-8 characters away from 's', forward if
9870 * 'off' >= 0, backwards if negative. But don't go outside of position
9871 * 'lim', which better be < s if off < 0 */
9873 PERL_ARGS_ASSERT_REGHOP3;
9876 while (off-- && s < lim) {
9877 /* XXX could check well-formedness here */
9878 U8 *new_s = s + UTF8SKIP(s);
9879 if (new_s > lim) /* lim may be in the middle of a long character */
9885 while (off++ && s > lim) {
9887 if (UTF8_IS_CONTINUED(*s)) {
9888 while (s > lim && UTF8_IS_CONTINUATION(*s))
9890 if (! UTF8_IS_START(*s)) {
9891 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9894 /* XXX could check well-formedness here */
9901 S_reghop4(U8 *s, SSize_t off, const U8* llim, const U8* rlim)
9903 PERL_ARGS_ASSERT_REGHOP4;
9906 while (off-- && s < rlim) {
9907 /* XXX could check well-formedness here */
9912 while (off++ && s > llim) {
9914 if (UTF8_IS_CONTINUED(*s)) {
9915 while (s > llim && UTF8_IS_CONTINUATION(*s))
9917 if (! UTF8_IS_START(*s)) {
9918 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9921 /* XXX could check well-formedness here */
9927 /* like reghop3, but returns NULL on overrun, rather than returning last
9931 S_reghopmaybe3(U8* s, SSize_t off, const U8* const lim)
9933 PERL_ARGS_ASSERT_REGHOPMAYBE3;
9936 while (off-- && s < lim) {
9937 /* XXX could check well-formedness here */
9944 while (off++ && s > lim) {
9946 if (UTF8_IS_CONTINUED(*s)) {
9947 while (s > lim && UTF8_IS_CONTINUATION(*s))
9949 if (! UTF8_IS_START(*s)) {
9950 Perl_croak_nocontext("Malformed UTF-8 character (fatal)");
9953 /* XXX could check well-formedness here */
9962 /* when executing a regex that may have (?{}), extra stuff needs setting
9963 up that will be visible to the called code, even before the current
9964 match has finished. In particular:
9966 * $_ is localised to the SV currently being matched;
9967 * pos($_) is created if necessary, ready to be updated on each call-out
9969 * a fake PMOP is created that can be set to PL_curpm (normally PL_curpm
9970 isn't set until the current pattern is successfully finished), so that
9971 $1 etc of the match-so-far can be seen;
9972 * save the old values of subbeg etc of the current regex, and set then
9973 to the current string (again, this is normally only done at the end
9978 S_setup_eval_state(pTHX_ regmatch_info *const reginfo)
9981 regexp *const rex = ReANY(reginfo->prog);
9982 regmatch_info_aux_eval *eval_state = reginfo->info_aux_eval;
9984 eval_state->rex = rex;
9987 /* Make $_ available to executed code. */
9988 if (reginfo->sv != DEFSV) {
9990 DEFSV_set(reginfo->sv);
9993 if (!(mg = mg_find_mglob(reginfo->sv))) {
9994 /* prepare for quick setting of pos */
9995 mg = sv_magicext_mglob(reginfo->sv);
9998 eval_state->pos_magic = mg;
9999 eval_state->pos = mg->mg_len;
10000 eval_state->pos_flags = mg->mg_flags;
10003 eval_state->pos_magic = NULL;
10005 if (!PL_reg_curpm) {
10006 /* PL_reg_curpm is a fake PMOP that we can attach the current
10007 * regex to and point PL_curpm at, so that $1 et al are visible
10008 * within a /(?{})/. It's just allocated once per interpreter the
10009 * first time its needed */
10010 Newxz(PL_reg_curpm, 1, PMOP);
10011 #ifdef USE_ITHREADS
10013 SV* const repointer = &PL_sv_undef;
10014 /* this regexp is also owned by the new PL_reg_curpm, which
10015 will try to free it. */
10016 av_push(PL_regex_padav, repointer);
10017 PL_reg_curpm->op_pmoffset = av_tindex(PL_regex_padav);
10018 PL_regex_pad = AvARRAY(PL_regex_padav);
10022 SET_reg_curpm(reginfo->prog);
10023 eval_state->curpm = PL_curpm;
10024 PL_curpm_under = PL_curpm;
10025 PL_curpm = PL_reg_curpm;
10026 if (RXp_MATCH_COPIED(rex)) {
10027 /* Here is a serious problem: we cannot rewrite subbeg,
10028 since it may be needed if this match fails. Thus
10029 $` inside (?{}) could fail... */
10030 eval_state->subbeg = rex->subbeg;
10031 eval_state->sublen = rex->sublen;
10032 eval_state->suboffset = rex->suboffset;
10033 eval_state->subcoffset = rex->subcoffset;
10034 #ifdef PERL_ANY_COW
10035 eval_state->saved_copy = rex->saved_copy;
10037 RXp_MATCH_COPIED_off(rex);
10040 eval_state->subbeg = NULL;
10041 rex->subbeg = (char *)reginfo->strbeg;
10042 rex->suboffset = 0;
10043 rex->subcoffset = 0;
10044 rex->sublen = reginfo->strend - reginfo->strbeg;
10048 /* destructor to clear up regmatch_info_aux and regmatch_info_aux_eval */
10051 S_cleanup_regmatch_info_aux(pTHX_ void *arg)
10053 regmatch_info_aux *aux = (regmatch_info_aux *) arg;
10054 regmatch_info_aux_eval *eval_state = aux->info_aux_eval;
10057 Safefree(aux->poscache);
10061 /* undo the effects of S_setup_eval_state() */
10063 if (eval_state->subbeg) {
10064 regexp * const rex = eval_state->rex;
10065 rex->subbeg = eval_state->subbeg;
10066 rex->sublen = eval_state->sublen;
10067 rex->suboffset = eval_state->suboffset;
10068 rex->subcoffset = eval_state->subcoffset;
10069 #ifdef PERL_ANY_COW
10070 rex->saved_copy = eval_state->saved_copy;
10072 RXp_MATCH_COPIED_on(rex);
10074 if (eval_state->pos_magic)
10076 eval_state->pos_magic->mg_len = eval_state->pos;
10077 eval_state->pos_magic->mg_flags =
10078 (eval_state->pos_magic->mg_flags & ~MGf_BYTES)
10079 | (eval_state->pos_flags & MGf_BYTES);
10082 PL_curpm = eval_state->curpm;
10085 PL_regmatch_state = aux->old_regmatch_state;
10086 PL_regmatch_slab = aux->old_regmatch_slab;
10088 /* free all slabs above current one - this must be the last action
10089 * of this function, as aux and eval_state are allocated within
10090 * slabs and may be freed here */
10092 s = PL_regmatch_slab->next;
10094 PL_regmatch_slab->next = NULL;
10096 regmatch_slab * const osl = s;
10105 S_to_utf8_substr(pTHX_ regexp *prog)
10107 /* Converts substr fields in prog from bytes to UTF-8, calling fbm_compile
10108 * on the converted value */
10112 PERL_ARGS_ASSERT_TO_UTF8_SUBSTR;
10115 if (prog->substrs->data[i].substr
10116 && !prog->substrs->data[i].utf8_substr) {
10117 SV* const sv = newSVsv(prog->substrs->data[i].substr);
10118 prog->substrs->data[i].utf8_substr = sv;
10119 sv_utf8_upgrade(sv);
10120 if (SvVALID(prog->substrs->data[i].substr)) {
10121 if (SvTAIL(prog->substrs->data[i].substr)) {
10122 /* Trim the trailing \n that fbm_compile added last
10124 SvCUR_set(sv, SvCUR(sv) - 1);
10125 /* Whilst this makes the SV technically "invalid" (as its
10126 buffer is no longer followed by "\0") when fbm_compile()
10127 adds the "\n" back, a "\0" is restored. */
10128 fbm_compile(sv, FBMcf_TAIL);
10130 fbm_compile(sv, 0);
10132 if (prog->substrs->data[i].substr == prog->check_substr)
10133 prog->check_utf8 = sv;
10139 S_to_byte_substr(pTHX_ regexp *prog)
10141 /* Converts substr fields in prog from UTF-8 to bytes, calling fbm_compile
10142 * on the converted value; returns FALSE if can't be converted. */
10146 PERL_ARGS_ASSERT_TO_BYTE_SUBSTR;
10149 if (prog->substrs->data[i].utf8_substr
10150 && !prog->substrs->data[i].substr) {
10151 SV* sv = newSVsv(prog->substrs->data[i].utf8_substr);
10152 if (! sv_utf8_downgrade(sv, TRUE)) {
10155 if (SvVALID(prog->substrs->data[i].utf8_substr)) {
10156 if (SvTAIL(prog->substrs->data[i].utf8_substr)) {
10157 /* Trim the trailing \n that fbm_compile added last
10159 SvCUR_set(sv, SvCUR(sv) - 1);
10160 fbm_compile(sv, FBMcf_TAIL);
10162 fbm_compile(sv, 0);
10164 prog->substrs->data[i].substr = sv;
10165 if (prog->substrs->data[i].utf8_substr == prog->check_utf8)
10166 prog->check_substr = sv;
10173 #ifndef PERL_IN_XSUB_RE
10176 Perl__is_grapheme(pTHX_ const U8 * strbeg, const U8 * s, const U8 * strend, const UV cp)
10178 /* Temporary helper function for toke.c. Verify that the code point 'cp'
10179 * is a stand-alone grapheme. The UTF-8 for 'cp' begins at position 's' in
10180 * the larger string bounded by 'strbeg' and 'strend'.
10182 * 'cp' needs to be assigned (if not a future version of the Unicode
10183 * Standard could make it something that combines with adjacent characters,
10184 * so code using it would then break), and there has to be a GCB break
10185 * before and after the character. */
10189 GCB_enum cp_gcb_val, prev_cp_gcb_val, next_cp_gcb_val;
10190 const U8 * prev_cp_start;
10192 PERL_ARGS_ASSERT__IS_GRAPHEME;
10194 if ( UNLIKELY(UNICODE_IS_SUPER(cp))
10195 || UNLIKELY(UNICODE_IS_NONCHAR(cp)))
10197 /* These are considered graphemes */
10201 /* Otherwise, unassigned code points are forbidden */
10202 if (UNLIKELY(! ELEMENT_RANGE_MATCHES_INVLIST(
10203 _invlist_search(PL_Assigned_invlist, cp))))
10208 cp_gcb_val = getGCB_VAL_CP(cp);
10210 /* Find the GCB value of the previous code point in the input */
10211 prev_cp_start = utf8_hop_back(s, -1, strbeg);
10212 if (UNLIKELY(prev_cp_start == s)) {
10213 prev_cp_gcb_val = GCB_EDGE;
10216 prev_cp_gcb_val = getGCB_VAL_UTF8(prev_cp_start, strend);
10219 /* And check that is a grapheme boundary */
10220 if (! isGCB(prev_cp_gcb_val, cp_gcb_val, strbeg, s,
10221 TRUE /* is UTF-8 encoded */ ))
10226 /* Similarly verify there is a break between the current character and the
10230 next_cp_gcb_val = GCB_EDGE;
10233 next_cp_gcb_val = getGCB_VAL_UTF8(s, strend);
10236 return isGCB(cp_gcb_val, next_cp_gcb_val, strbeg, s, TRUE);
10240 =head1 Unicode Support
10242 =for apidoc isSCRIPT_RUN
10244 Returns a bool as to whether or not the sequence of bytes from C<s> up to but
10245 not including C<send> form a "script run". C<utf8_target> is TRUE iff the
10246 sequence starting at C<s> is to be treated as UTF-8. To be precise, except for
10247 two degenerate cases given below, this function returns TRUE iff all code
10248 points in it come from any combination of three "scripts" given by the Unicode
10249 "Script Extensions" property: Common, Inherited, and possibly one other.
10250 Additionally all decimal digits must come from the same consecutive sequence of
10253 For example, if all the characters in the sequence are Greek, or Common, or
10254 Inherited, this function will return TRUE, provided any decimal digits in it
10255 are from the same block of digits in Common. (These are the ASCII digits
10256 "0".."9" and additionally a block for full width forms of these, and several
10257 others used in mathematical notation.) For scripts (unlike Greek) that have
10258 their own digits defined this will accept either digits from that set or from
10259 one of the Common digit sets, but not a combination of the two. Some scripts,
10260 such as Arabic, have more than one set of digits. All digits must come from
10261 the same set for this function to return TRUE.
10263 C<*ret_script>, if C<ret_script> is not NULL, will on return of TRUE
10264 contain the script found, using the C<SCX_enum> typedef. Its value will be
10265 C<SCX_INVALID> if the function returns FALSE.
10267 If the sequence is empty, TRUE is returned, but C<*ret_script> (if asked for)
10268 will be C<SCX_INVALID>.
10270 If the sequence contains a single code point which is unassigned to a character
10271 in the version of Unicode being used, the function will return TRUE, and the
10272 script will be C<SCX_Unknown>. Any other combination of unassigned code points
10273 in the input sequence will result in the function treating the input as not
10274 being a script run.
10276 The returned script will be C<SCX_Inherited> iff all the code points in it are
10277 from the Inherited script.
10279 Otherwise, the returned script will be C<SCX_Common> iff all the code points in
10280 it are from the Inherited or Common scripts.
10287 Perl_isSCRIPT_RUN(pTHX_ const U8 * s, const U8 * send, const bool utf8_target)
10289 /* Basically, it looks at each character in the sequence to see if the
10290 * above conditions are met; if not it fails. It uses an inversion map to
10291 * find the enum corresponding to the script of each character. But this
10292 * is complicated by the fact that a few code points can be in any of
10293 * several scripts. The data has been constructed so that there are
10294 * additional enum values (all negative) for these situations. The
10295 * absolute value of those is an index into another table which contains
10296 * pointers to auxiliary tables for each such situation. Each aux array
10297 * lists all the scripts for the given situation. There is another,
10298 * parallel, table that gives the number of entries in each aux table.
10299 * These are all defined in charclass_invlists.h */
10301 /* XXX Here are the additional things UTS 39 says could be done:
10303 * Forbid sequences of the same nonspacing mark
10305 * Check to see that all the characters are in the sets of exemplar
10306 * characters for at least one language in the Unicode Common Locale Data
10307 * Repository [CLDR]. */
10311 /* Things that match /\d/u */
10312 SV * decimals_invlist = PL_XPosix_ptrs[_CC_DIGIT];
10313 UV * decimals_array = invlist_array(decimals_invlist);
10315 /* What code point is the digit '0' of the script run? (0 meaning FALSE if
10316 * not currently known) */
10317 UV zero_of_run = 0;
10319 SCX_enum script_of_run = SCX_INVALID; /* Illegal value */
10320 SCX_enum script_of_char = SCX_INVALID;
10322 /* If the script remains not fully determined from iteration to iteration,
10323 * this is the current intersection of the possiblities. */
10324 SCX_enum * intersection = NULL;
10325 PERL_UINT_FAST8_T intersection_len = 0;
10327 bool retval = TRUE;
10328 SCX_enum * ret_script = NULL;
10332 PERL_ARGS_ASSERT_ISSCRIPT_RUN;
10334 /* All code points in 0..255 are either Common or Latin, so must be a
10335 * script run. We can return immediately unless we need to know which
10337 if (! utf8_target && LIKELY(send > s)) {
10338 if (ret_script == NULL) {
10342 /* If any character is Latin, the run is Latin */
10344 if (isALPHA_L1(*s) && LIKELY(*s != MICRO_SIGN_NATIVE)) {
10345 *ret_script = SCX_Latin;
10350 /* Here, all are Common */
10351 *ret_script = SCX_Common;
10355 /* Look at each character in the sequence */
10357 /* If the current character being examined is a digit, this is the code
10358 * point of the zero for its sequence of 10 */
10363 /* The code allows all scripts to use the ASCII digits. This is
10364 * because they are in the Common script. Hence any ASCII ones found
10365 * are ok, unless and until a digit from another set has already been
10366 * encountered. digit ranges in Common are not similarly blessed) */
10367 if (UNLIKELY(isDIGIT(*s))) {
10368 if (UNLIKELY(script_of_run == SCX_Unknown)) {
10373 if (zero_of_run != '0') {
10385 /* Here, isn't an ASCII digit. Find the code point of the character */
10386 if (! UTF8_IS_INVARIANT(*s)) {
10388 cp = valid_utf8_to_uvchr((U8 *) s, &len);
10395 /* If is within the range [+0 .. +9] of the script's zero, it also is a
10396 * digit in that script. We can skip the rest of this code for this
10398 if (UNLIKELY( zero_of_run
10399 && cp >= zero_of_run
10400 && cp - zero_of_run <= 9))
10405 /* Find the character's script. The correct values are hard-coded here
10406 * for small-enough code points. */
10407 if (cp < 0x2B9) { /* From inspection of Unicode db; extremely
10408 unlikely to change */
10410 || ( isALPHA_L1(cp)
10411 && LIKELY(cp != MICRO_SIGN_NATIVE)))
10413 script_of_char = SCX_Latin;
10416 script_of_char = SCX_Common;
10420 script_of_char = _Perl_SCX_invmap[
10421 _invlist_search(PL_SCX_invlist, cp)];
10424 /* We arbitrarily accept a single unassigned character, but not in
10425 * combination with anything else, and not a run of them. */
10426 if ( UNLIKELY(script_of_run == SCX_Unknown)
10427 || UNLIKELY( script_of_run != SCX_INVALID
10428 && script_of_char == SCX_Unknown))
10434 /* For the first character, or the run is inherited, the run's script
10435 * is set to the char's */
10436 if ( UNLIKELY(script_of_run == SCX_INVALID)
10437 || UNLIKELY(script_of_run == SCX_Inherited))
10439 script_of_run = script_of_char;
10442 /* For the character's script to be Unknown, it must be the first
10443 * character in the sequence (for otherwise a test above would have
10444 * prevented us from reaching here), and we have set the run's script
10445 * to it. Nothing further to be done for this character */
10446 if (UNLIKELY(script_of_char == SCX_Unknown)) {
10450 /* We accept 'inherited' script characters currently even at the
10451 * beginning. (We know that no characters in Inherited are digits, or
10452 * we'd have to check for that) */
10453 if (UNLIKELY(script_of_char == SCX_Inherited)) {
10457 /* If the run so far is Common, and the new character isn't, change the
10458 * run's script to that of this character */
10459 if (script_of_run == SCX_Common && script_of_char != SCX_Common) {
10460 script_of_run = script_of_char;
10463 /* Now we can see if the script of the new character is the same as
10464 * that of the run */
10465 if (LIKELY(script_of_char == script_of_run)) {
10466 /* By far the most common case */
10467 goto scripts_match;
10470 /* Here, the script of the run isn't Common. But characters in Common
10471 * match any script */
10472 if (script_of_char == SCX_Common) {
10473 goto scripts_match;
10476 #ifndef HAS_SCX_AUX_TABLES
10478 /* Too early a Unicode version to have a code point belonging to more
10479 * than one script, so, if the scripts don't exactly match, fail */
10480 PERL_UNUSED_VAR(intersection_len);
10486 /* Here there is no exact match between the character's script and the
10487 * run's. And we've handled the special cases of scripts Unknown,
10488 * Inherited, and Common.
10490 * Negative script numbers signify that the value may be any of several
10491 * scripts, and we need to look at auxiliary information to make our
10492 * deterimination. But if both are non-negative, we can fail now */
10493 if (LIKELY(script_of_char >= 0)) {
10494 const SCX_enum * search_in;
10495 PERL_UINT_FAST8_T search_in_len;
10496 PERL_UINT_FAST8_T i;
10498 if (LIKELY(script_of_run >= 0)) {
10503 /* Use the previously constructed set of possible scripts, if any.
10505 if (intersection) {
10506 search_in = intersection;
10507 search_in_len = intersection_len;
10510 search_in = SCX_AUX_TABLE_ptrs[-script_of_run];
10511 search_in_len = SCX_AUX_TABLE_lengths[-script_of_run];
10514 for (i = 0; i < search_in_len; i++) {
10515 if (search_in[i] == script_of_char) {
10516 script_of_run = script_of_char;
10517 goto scripts_match;
10524 else if (LIKELY(script_of_run >= 0)) {
10525 /* script of character could be one of several, but run is a single
10527 const SCX_enum * search_in = SCX_AUX_TABLE_ptrs[-script_of_char];
10528 const PERL_UINT_FAST8_T search_in_len
10529 = SCX_AUX_TABLE_lengths[-script_of_char];
10530 PERL_UINT_FAST8_T i;
10532 for (i = 0; i < search_in_len; i++) {
10533 if (search_in[i] == script_of_run) {
10534 script_of_char = script_of_run;
10535 goto scripts_match;
10543 /* Both run and char could be in one of several scripts. If the
10544 * intersection is empty, then this character isn't in this script
10545 * run. Otherwise, we need to calculate the intersection to use
10546 * for future iterations of the loop, unless we are already at the
10547 * final character */
10548 const SCX_enum * search_char = SCX_AUX_TABLE_ptrs[-script_of_char];
10549 const PERL_UINT_FAST8_T char_len
10550 = SCX_AUX_TABLE_lengths[-script_of_char];
10551 const SCX_enum * search_run;
10552 PERL_UINT_FAST8_T run_len;
10554 SCX_enum * new_overlap = NULL;
10555 PERL_UINT_FAST8_T i, j;
10557 if (intersection) {
10558 search_run = intersection;
10559 run_len = intersection_len;
10562 search_run = SCX_AUX_TABLE_ptrs[-script_of_run];
10563 run_len = SCX_AUX_TABLE_lengths[-script_of_run];
10566 intersection_len = 0;
10568 for (i = 0; i < run_len; i++) {
10569 for (j = 0; j < char_len; j++) {
10570 if (search_run[i] == search_char[j]) {
10572 /* Here, the script at i,j matches. That means this
10573 * character is in the run. But continue on to find
10574 * the complete intersection, for the next loop
10575 * iteration, and for the digit check after it.
10577 * On the first found common script, we malloc space
10578 * for the intersection list for the worst case of the
10579 * intersection, which is the minimum of the number of
10580 * scripts remaining in each set. */
10581 if (intersection_len == 0) {
10583 MIN(run_len - i, char_len - j),
10586 new_overlap[intersection_len++] = search_run[i];
10591 /* Here we've looked through everything. If they have no scripts
10592 * in common, not a run */
10593 if (intersection_len == 0) {
10598 /* If there is only a single script in common, set to that.
10599 * Otherwise, use the intersection going forward */
10600 Safefree(intersection);
10601 intersection = NULL;
10602 if (intersection_len == 1) {
10603 script_of_run = script_of_char = new_overlap[0];
10604 Safefree(new_overlap);
10605 new_overlap = NULL;
10608 intersection = new_overlap;
10616 /* Here, the script of the character is compatible with that of the
10617 * run. That means that in most cases, it continues the script run.
10618 * Either it and the run match exactly, or one or both can be in any of
10619 * several scripts, and the intersection is not empty. However, if the
10620 * character is a decimal digit, it could still mean failure if it is
10621 * from the wrong sequence of 10. So, we need to look at if it's a
10622 * digit. We've already handled the 10 decimal digits, and the next
10623 * lowest one is this one: */
10624 if (cp < FIRST_NON_ASCII_DECIMAL_DIGIT) {
10625 continue; /* Not a digit; this character is part of the run */
10628 /* If we have a definitive '0' for the script of this character, we
10629 * know that for this to be a digit, it must be in the range of +0..+9
10631 if ( script_of_char >= 0
10632 && (zero_of_char = script_zeros[script_of_char]))
10634 if ( cp < zero_of_char
10635 || cp > zero_of_char + 9)
10637 continue; /* Not a digit; this character is part of the run
10642 else { /* Need to look up if this character is a digit or not */
10643 SSize_t index_of_zero_of_char;
10644 index_of_zero_of_char = _invlist_search(decimals_invlist, cp);
10645 if ( UNLIKELY(index_of_zero_of_char < 0)
10646 || ! ELEMENT_RANGE_MATCHES_INVLIST(index_of_zero_of_char))
10648 continue; /* Not a digit; this character is part of the run.
10652 zero_of_char = decimals_array[index_of_zero_of_char];
10655 /* Here, the character is a decimal digit, and the zero of its sequence
10656 * of 10 is in 'zero_of_char'. If we already have a zero for this run,
10657 * they better be the same. */
10659 if (zero_of_run != zero_of_char) {
10664 else { /* Otherwise we now have a zero for this run */
10665 zero_of_run = zero_of_char;
10667 } /* end of looping through CLOSESR text */
10669 Safefree(intersection);
10671 if (ret_script != NULL) {
10673 *ret_script = script_of_run;
10676 *ret_script = SCX_INVALID;
10683 #endif /* ifndef PERL_IN_XSUB_RE */
10686 * ex: set ts=8 sts=4 sw=4 et: