5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
8 /* This file contains functions for compiling a regular expression. See
9 * also regexec.c which funnily enough, contains functions for executing
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_comp.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
57 **** Alterations to Henry's code are...
59 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
60 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 by Larry Wall and others
62 **** You may distribute under the terms of either the GNU General Public
63 **** License or the Artistic License, as specified in the README file.
66 * Beware that some of this code is subtly aware of the way operator
67 * precedence is structured in regular expressions. Serious changes in
68 * regular-expression syntax might require a total rethink.
71 #define PERL_IN_REGCOMP_C
74 #ifndef PERL_IN_XSUB_RE
79 #ifdef PERL_IN_XSUB_RE
90 # if defined(BUGGY_MSC6)
91 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
92 # pragma optimize("a",off)
93 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
94 # pragma optimize("w",on )
95 # endif /* BUGGY_MSC6 */
102 typedef struct RExC_state_t {
103 U32 flags; /* are we folding, multilining? */
104 char *precomp; /* uncompiled string. */
105 regexp *rx; /* perl core regexp structure */
106 regexp_internal *rxi; /* internal data for regexp object pprivate field */
107 char *start; /* Start of input for compile */
108 char *end; /* End of input for compile */
109 char *parse; /* Input-scan pointer. */
110 I32 whilem_seen; /* number of WHILEM in this expr */
111 regnode *emit_start; /* Start of emitted-code area */
112 regnode *emit_bound; /* First regnode outside of the allocated space */
113 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
114 I32 naughty; /* How bad is this pattern? */
115 I32 sawback; /* Did we see \1, ...? */
117 I32 size; /* Code size. */
118 I32 npar; /* Capture buffer count, (OPEN). */
119 I32 cpar; /* Capture buffer count, (CLOSE). */
120 I32 nestroot; /* root parens we are in - used by accept */
124 regnode **open_parens; /* pointers to open parens */
125 regnode **close_parens; /* pointers to close parens */
126 regnode *opend; /* END node in program */
127 I32 utf8; /* whether the pattern is utf8 or not */
128 I32 orig_utf8; /* whether the pattern was originally in utf8 */
129 /* XXX use this for future optimisation of case
130 * where pattern must be upgraded to utf8. */
131 HV *charnames; /* cache of named sequences */
132 HV *paren_names; /* Paren names */
134 regnode **recurse; /* Recurse regops */
135 I32 recurse_count; /* Number of recurse regops */
137 char *starttry; /* -Dr: where regtry was called. */
138 #define RExC_starttry (pRExC_state->starttry)
141 const char *lastparse;
143 AV *paren_name_list; /* idx -> name */
144 #define RExC_lastparse (pRExC_state->lastparse)
145 #define RExC_lastnum (pRExC_state->lastnum)
146 #define RExC_paren_name_list (pRExC_state->paren_name_list)
150 #define RExC_flags (pRExC_state->flags)
151 #define RExC_precomp (pRExC_state->precomp)
152 #define RExC_rx (pRExC_state->rx)
153 #define RExC_rxi (pRExC_state->rxi)
154 #define RExC_start (pRExC_state->start)
155 #define RExC_end (pRExC_state->end)
156 #define RExC_parse (pRExC_state->parse)
157 #define RExC_whilem_seen (pRExC_state->whilem_seen)
158 #ifdef RE_TRACK_PATTERN_OFFSETS
159 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
161 #define RExC_emit (pRExC_state->emit)
162 #define RExC_emit_start (pRExC_state->emit_start)
163 #define RExC_emit_bound (pRExC_state->emit_bound)
164 #define RExC_naughty (pRExC_state->naughty)
165 #define RExC_sawback (pRExC_state->sawback)
166 #define RExC_seen (pRExC_state->seen)
167 #define RExC_size (pRExC_state->size)
168 #define RExC_npar (pRExC_state->npar)
169 #define RExC_nestroot (pRExC_state->nestroot)
170 #define RExC_extralen (pRExC_state->extralen)
171 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
172 #define RExC_seen_evals (pRExC_state->seen_evals)
173 #define RExC_utf8 (pRExC_state->utf8)
174 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
175 #define RExC_charnames (pRExC_state->charnames)
176 #define RExC_open_parens (pRExC_state->open_parens)
177 #define RExC_close_parens (pRExC_state->close_parens)
178 #define RExC_opend (pRExC_state->opend)
179 #define RExC_paren_names (pRExC_state->paren_names)
180 #define RExC_recurse (pRExC_state->recurse)
181 #define RExC_recurse_count (pRExC_state->recurse_count)
184 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
185 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
186 ((*s) == '{' && regcurly(s)))
189 #undef SPSTART /* dratted cpp namespace... */
192 * Flags to be passed up and down.
194 #define WORST 0 /* Worst case. */
195 #define HASWIDTH 0x01 /* Known to match non-null strings. */
196 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
197 #define SPSTART 0x04 /* Starts with * or +. */
198 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
199 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
201 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
203 /* whether trie related optimizations are enabled */
204 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
205 #define TRIE_STUDY_OPT
206 #define FULL_TRIE_STUDY
212 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
213 #define PBITVAL(paren) (1 << ((paren) & 7))
214 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
215 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
216 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
219 /* About scan_data_t.
221 During optimisation we recurse through the regexp program performing
222 various inplace (keyhole style) optimisations. In addition study_chunk
223 and scan_commit populate this data structure with information about
224 what strings MUST appear in the pattern. We look for the longest
225 string that must appear for at a fixed location, and we look for the
226 longest string that may appear at a floating location. So for instance
231 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
232 strings (because they follow a .* construct). study_chunk will identify
233 both FOO and BAR as being the longest fixed and floating strings respectively.
235 The strings can be composites, for instance
239 will result in a composite fixed substring 'foo'.
241 For each string some basic information is maintained:
243 - offset or min_offset
244 This is the position the string must appear at, or not before.
245 It also implicitly (when combined with minlenp) tells us how many
246 character must match before the string we are searching.
247 Likewise when combined with minlenp and the length of the string
248 tells us how many characters must appear after the string we have
252 Only used for floating strings. This is the rightmost point that
253 the string can appear at. Ifset to I32 max it indicates that the
254 string can occur infinitely far to the right.
257 A pointer to the minimum length of the pattern that the string
258 was found inside. This is important as in the case of positive
259 lookahead or positive lookbehind we can have multiple patterns
264 The minimum length of the pattern overall is 3, the minimum length
265 of the lookahead part is 3, but the minimum length of the part that
266 will actually match is 1. So 'FOO's minimum length is 3, but the
267 minimum length for the F is 1. This is important as the minimum length
268 is used to determine offsets in front of and behind the string being
269 looked for. Since strings can be composites this is the length of the
270 pattern at the time it was commited with a scan_commit. Note that
271 the length is calculated by study_chunk, so that the minimum lengths
272 are not known until the full pattern has been compiled, thus the
273 pointer to the value.
277 In the case of lookbehind the string being searched for can be
278 offset past the start point of the final matching string.
279 If this value was just blithely removed from the min_offset it would
280 invalidate some of the calculations for how many chars must match
281 before or after (as they are derived from min_offset and minlen and
282 the length of the string being searched for).
283 When the final pattern is compiled and the data is moved from the
284 scan_data_t structure into the regexp structure the information
285 about lookbehind is factored in, with the information that would
286 have been lost precalculated in the end_shift field for the
289 The fields pos_min and pos_delta are used to store the minimum offset
290 and the delta to the maximum offset at the current point in the pattern.
294 typedef struct scan_data_t {
295 /*I32 len_min; unused */
296 /*I32 len_delta; unused */
300 I32 last_end; /* min value, <0 unless valid. */
303 SV **longest; /* Either &l_fixed, or &l_float. */
304 SV *longest_fixed; /* longest fixed string found in pattern */
305 I32 offset_fixed; /* offset where it starts */
306 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
307 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
308 SV *longest_float; /* longest floating string found in pattern */
309 I32 offset_float_min; /* earliest point in string it can appear */
310 I32 offset_float_max; /* latest point in string it can appear */
311 I32 *minlen_float; /* pointer to the minlen relevent to the string */
312 I32 lookbehind_float; /* is the position of the string modified by LB */
316 struct regnode_charclass_class *start_class;
320 * Forward declarations for pregcomp()'s friends.
323 static const scan_data_t zero_scan_data =
324 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
326 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
327 #define SF_BEFORE_SEOL 0x0001
328 #define SF_BEFORE_MEOL 0x0002
329 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
330 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
333 # define SF_FIX_SHIFT_EOL (0+2)
334 # define SF_FL_SHIFT_EOL (0+4)
336 # define SF_FIX_SHIFT_EOL (+2)
337 # define SF_FL_SHIFT_EOL (+4)
340 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
341 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
343 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
344 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
345 #define SF_IS_INF 0x0040
346 #define SF_HAS_PAR 0x0080
347 #define SF_IN_PAR 0x0100
348 #define SF_HAS_EVAL 0x0200
349 #define SCF_DO_SUBSTR 0x0400
350 #define SCF_DO_STCLASS_AND 0x0800
351 #define SCF_DO_STCLASS_OR 0x1000
352 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
353 #define SCF_WHILEM_VISITED_POS 0x2000
355 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
356 #define SCF_SEEN_ACCEPT 0x8000
358 #define UTF (RExC_utf8 != 0)
359 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
360 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
362 #define OOB_UNICODE 12345678
363 #define OOB_NAMEDCLASS -1
365 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
366 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
369 /* length of regex to show in messages that don't mark a position within */
370 #define RegexLengthToShowInErrorMessages 127
373 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
374 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
375 * op/pragma/warn/regcomp.
377 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
378 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
380 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
383 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
384 * arg. Show regex, up to a maximum length. If it's too long, chop and add
387 #define _FAIL(code) STMT_START { \
388 const char *ellipses = ""; \
389 IV len = RExC_end - RExC_precomp; \
392 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
393 if (len > RegexLengthToShowInErrorMessages) { \
394 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
395 len = RegexLengthToShowInErrorMessages - 10; \
401 #define FAIL(msg) _FAIL( \
402 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
403 msg, (int)len, RExC_precomp, ellipses))
405 #define FAIL2(msg,arg) _FAIL( \
406 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
407 arg, (int)len, RExC_precomp, ellipses))
410 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
412 #define Simple_vFAIL(m) STMT_START { \
413 const IV offset = RExC_parse - RExC_precomp; \
414 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
415 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
419 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
421 #define vFAIL(m) STMT_START { \
423 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
428 * Like Simple_vFAIL(), but accepts two arguments.
430 #define Simple_vFAIL2(m,a1) STMT_START { \
431 const IV offset = RExC_parse - RExC_precomp; \
432 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
433 (int)offset, RExC_precomp, RExC_precomp + offset); \
437 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
439 #define vFAIL2(m,a1) STMT_START { \
441 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
442 Simple_vFAIL2(m, a1); \
447 * Like Simple_vFAIL(), but accepts three arguments.
449 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
450 const IV offset = RExC_parse - RExC_precomp; \
451 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
452 (int)offset, RExC_precomp, RExC_precomp + offset); \
456 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
458 #define vFAIL3(m,a1,a2) STMT_START { \
460 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
461 Simple_vFAIL3(m, a1, a2); \
465 * Like Simple_vFAIL(), but accepts four arguments.
467 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
468 const IV offset = RExC_parse - RExC_precomp; \
469 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
470 (int)offset, RExC_precomp, RExC_precomp + offset); \
473 #define vWARN(loc,m) STMT_START { \
474 const IV offset = loc - RExC_precomp; \
475 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
476 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
479 #define vWARNdep(loc,m) STMT_START { \
480 const IV offset = loc - RExC_precomp; \
481 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
482 "%s" REPORT_LOCATION, \
483 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
487 #define vWARN2(loc, m, a1) STMT_START { \
488 const IV offset = loc - RExC_precomp; \
489 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
493 #define vWARN3(loc, m, a1, a2) STMT_START { \
494 const IV offset = loc - RExC_precomp; \
495 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
496 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
499 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
500 const IV offset = loc - RExC_precomp; \
501 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
502 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
505 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
506 const IV offset = loc - RExC_precomp; \
507 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
508 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
512 /* Allow for side effects in s */
513 #define REGC(c,s) STMT_START { \
514 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
517 /* Macros for recording node offsets. 20001227 mjd@plover.com
518 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
519 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
520 * Element 0 holds the number n.
521 * Position is 1 indexed.
523 #ifndef RE_TRACK_PATTERN_OFFSETS
524 #define Set_Node_Offset_To_R(node,byte)
525 #define Set_Node_Offset(node,byte)
526 #define Set_Cur_Node_Offset
527 #define Set_Node_Length_To_R(node,len)
528 #define Set_Node_Length(node,len)
529 #define Set_Node_Cur_Length(node)
530 #define Node_Offset(n)
531 #define Node_Length(n)
532 #define Set_Node_Offset_Length(node,offset,len)
533 #define ProgLen(ri) ri->u.proglen
534 #define SetProgLen(ri,x) ri->u.proglen = x
536 #define ProgLen(ri) ri->u.offsets[0]
537 #define SetProgLen(ri,x) ri->u.offsets[0] = x
538 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
540 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
541 __LINE__, (int)(node), (int)(byte))); \
543 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
545 RExC_offsets[2*(node)-1] = (byte); \
550 #define Set_Node_Offset(node,byte) \
551 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
552 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
554 #define Set_Node_Length_To_R(node,len) STMT_START { \
556 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
557 __LINE__, (int)(node), (int)(len))); \
559 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
561 RExC_offsets[2*(node)] = (len); \
566 #define Set_Node_Length(node,len) \
567 Set_Node_Length_To_R((node)-RExC_emit_start, len)
568 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
569 #define Set_Node_Cur_Length(node) \
570 Set_Node_Length(node, RExC_parse - parse_start)
572 /* Get offsets and lengths */
573 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
574 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
576 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
577 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
578 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
582 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
583 #define EXPERIMENTAL_INPLACESCAN
584 #endif /*RE_TRACK_PATTERN_OFFSETS*/
586 #define DEBUG_STUDYDATA(str,data,depth) \
587 DEBUG_OPTIMISE_MORE_r(if(data){ \
588 PerlIO_printf(Perl_debug_log, \
589 "%*s" str "Pos:%"IVdf"/%"IVdf \
590 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
591 (int)(depth)*2, "", \
592 (IV)((data)->pos_min), \
593 (IV)((data)->pos_delta), \
594 (UV)((data)->flags), \
595 (IV)((data)->whilem_c), \
596 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
597 is_inf ? "INF " : "" \
599 if ((data)->last_found) \
600 PerlIO_printf(Perl_debug_log, \
601 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
602 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
603 SvPVX_const((data)->last_found), \
604 (IV)((data)->last_end), \
605 (IV)((data)->last_start_min), \
606 (IV)((data)->last_start_max), \
607 ((data)->longest && \
608 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
609 SvPVX_const((data)->longest_fixed), \
610 (IV)((data)->offset_fixed), \
611 ((data)->longest && \
612 (data)->longest==&((data)->longest_float)) ? "*" : "", \
613 SvPVX_const((data)->longest_float), \
614 (IV)((data)->offset_float_min), \
615 (IV)((data)->offset_float_max) \
617 PerlIO_printf(Perl_debug_log,"\n"); \
620 static void clear_re(pTHX_ void *r);
622 /* Mark that we cannot extend a found fixed substring at this point.
623 Update the longest found anchored substring and the longest found
624 floating substrings if needed. */
627 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
629 const STRLEN l = CHR_SVLEN(data->last_found);
630 const STRLEN old_l = CHR_SVLEN(*data->longest);
631 GET_RE_DEBUG_FLAGS_DECL;
633 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
634 SvSetMagicSV(*data->longest, data->last_found);
635 if (*data->longest == data->longest_fixed) {
636 data->offset_fixed = l ? data->last_start_min : data->pos_min;
637 if (data->flags & SF_BEFORE_EOL)
639 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
641 data->flags &= ~SF_FIX_BEFORE_EOL;
642 data->minlen_fixed=minlenp;
643 data->lookbehind_fixed=0;
645 else { /* *data->longest == data->longest_float */
646 data->offset_float_min = l ? data->last_start_min : data->pos_min;
647 data->offset_float_max = (l
648 ? data->last_start_max
649 : data->pos_min + data->pos_delta);
650 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
651 data->offset_float_max = I32_MAX;
652 if (data->flags & SF_BEFORE_EOL)
654 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
656 data->flags &= ~SF_FL_BEFORE_EOL;
657 data->minlen_float=minlenp;
658 data->lookbehind_float=0;
661 SvCUR_set(data->last_found, 0);
663 SV * const sv = data->last_found;
664 if (SvUTF8(sv) && SvMAGICAL(sv)) {
665 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
671 data->flags &= ~SF_BEFORE_EOL;
672 DEBUG_STUDYDATA("commit: ",data,0);
675 /* Can match anything (initialization) */
677 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
679 ANYOF_CLASS_ZERO(cl);
680 ANYOF_BITMAP_SETALL(cl);
681 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
683 cl->flags |= ANYOF_LOCALE;
686 /* Can match anything (initialization) */
688 S_cl_is_anything(const struct regnode_charclass_class *cl)
692 for (value = 0; value <= ANYOF_MAX; value += 2)
693 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
695 if (!(cl->flags & ANYOF_UNICODE_ALL))
697 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
702 /* Can match anything (initialization) */
704 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
706 Zero(cl, 1, struct regnode_charclass_class);
708 cl_anything(pRExC_state, cl);
712 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
714 Zero(cl, 1, struct regnode_charclass_class);
716 cl_anything(pRExC_state, cl);
718 cl->flags |= ANYOF_LOCALE;
721 /* 'And' a given class with another one. Can create false positives */
722 /* We assume that cl is not inverted */
724 S_cl_and(struct regnode_charclass_class *cl,
725 const struct regnode_charclass_class *and_with)
728 assert(and_with->type == ANYOF);
729 if (!(and_with->flags & ANYOF_CLASS)
730 && !(cl->flags & ANYOF_CLASS)
731 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
732 && !(and_with->flags & ANYOF_FOLD)
733 && !(cl->flags & ANYOF_FOLD)) {
736 if (and_with->flags & ANYOF_INVERT)
737 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
738 cl->bitmap[i] &= ~and_with->bitmap[i];
740 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
741 cl->bitmap[i] &= and_with->bitmap[i];
742 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
743 if (!(and_with->flags & ANYOF_EOS))
744 cl->flags &= ~ANYOF_EOS;
746 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
747 !(and_with->flags & ANYOF_INVERT)) {
748 cl->flags &= ~ANYOF_UNICODE_ALL;
749 cl->flags |= ANYOF_UNICODE;
750 ARG_SET(cl, ARG(and_with));
752 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
753 !(and_with->flags & ANYOF_INVERT))
754 cl->flags &= ~ANYOF_UNICODE_ALL;
755 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
756 !(and_with->flags & ANYOF_INVERT))
757 cl->flags &= ~ANYOF_UNICODE;
760 /* 'OR' a given class with another one. Can create false positives */
761 /* We assume that cl is not inverted */
763 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
765 if (or_with->flags & ANYOF_INVERT) {
767 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
768 * <= (B1 | !B2) | (CL1 | !CL2)
769 * which is wasteful if CL2 is small, but we ignore CL2:
770 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
771 * XXXX Can we handle case-fold? Unclear:
772 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
773 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
775 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
776 && !(or_with->flags & ANYOF_FOLD)
777 && !(cl->flags & ANYOF_FOLD) ) {
780 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
781 cl->bitmap[i] |= ~or_with->bitmap[i];
782 } /* XXXX: logic is complicated otherwise */
784 cl_anything(pRExC_state, cl);
787 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
788 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
789 && (!(or_with->flags & ANYOF_FOLD)
790 || (cl->flags & ANYOF_FOLD)) ) {
793 /* OR char bitmap and class bitmap separately */
794 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
795 cl->bitmap[i] |= or_with->bitmap[i];
796 if (or_with->flags & ANYOF_CLASS) {
797 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
798 cl->classflags[i] |= or_with->classflags[i];
799 cl->flags |= ANYOF_CLASS;
802 else { /* XXXX: logic is complicated, leave it along for a moment. */
803 cl_anything(pRExC_state, cl);
806 if (or_with->flags & ANYOF_EOS)
807 cl->flags |= ANYOF_EOS;
809 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
810 ARG(cl) != ARG(or_with)) {
811 cl->flags |= ANYOF_UNICODE_ALL;
812 cl->flags &= ~ANYOF_UNICODE;
814 if (or_with->flags & ANYOF_UNICODE_ALL) {
815 cl->flags |= ANYOF_UNICODE_ALL;
816 cl->flags &= ~ANYOF_UNICODE;
820 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
821 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
822 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
823 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
828 dump_trie(trie,widecharmap,revcharmap)
829 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
830 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
832 These routines dump out a trie in a somewhat readable format.
833 The _interim_ variants are used for debugging the interim
834 tables that are used to generate the final compressed
835 representation which is what dump_trie expects.
837 Part of the reason for their existance is to provide a form
838 of documentation as to how the different representations function.
843 Dumps the final compressed table form of the trie to Perl_debug_log.
844 Used for debugging make_trie().
848 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
849 AV *revcharmap, U32 depth)
852 SV *sv=sv_newmortal();
853 int colwidth= widecharmap ? 6 : 4;
854 GET_RE_DEBUG_FLAGS_DECL;
857 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
858 (int)depth * 2 + 2,"",
859 "Match","Base","Ofs" );
861 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
862 SV ** const tmp = av_fetch( revcharmap, state, 0);
864 PerlIO_printf( Perl_debug_log, "%*s",
866 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
867 PL_colors[0], PL_colors[1],
868 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
869 PERL_PV_ESCAPE_FIRSTCHAR
874 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
875 (int)depth * 2 + 2,"");
877 for( state = 0 ; state < trie->uniquecharcount ; state++ )
878 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
879 PerlIO_printf( Perl_debug_log, "\n");
881 for( state = 1 ; state < trie->statecount ; state++ ) {
882 const U32 base = trie->states[ state ].trans.base;
884 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
886 if ( trie->states[ state ].wordnum ) {
887 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
889 PerlIO_printf( Perl_debug_log, "%6s", "" );
892 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
897 while( ( base + ofs < trie->uniquecharcount ) ||
898 ( base + ofs - trie->uniquecharcount < trie->lasttrans
899 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
902 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
904 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
905 if ( ( base + ofs >= trie->uniquecharcount ) &&
906 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
907 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
909 PerlIO_printf( Perl_debug_log, "%*"UVXf,
911 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
913 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
917 PerlIO_printf( Perl_debug_log, "]");
920 PerlIO_printf( Perl_debug_log, "\n" );
924 Dumps a fully constructed but uncompressed trie in list form.
925 List tries normally only are used for construction when the number of
926 possible chars (trie->uniquecharcount) is very high.
927 Used for debugging make_trie().
930 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
931 HV *widecharmap, AV *revcharmap, U32 next_alloc,
935 SV *sv=sv_newmortal();
936 int colwidth= widecharmap ? 6 : 4;
937 GET_RE_DEBUG_FLAGS_DECL;
938 /* print out the table precompression. */
939 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
940 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
941 "------:-----+-----------------\n" );
943 for( state=1 ; state < next_alloc ; state ++ ) {
946 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
947 (int)depth * 2 + 2,"", (UV)state );
948 if ( ! trie->states[ state ].wordnum ) {
949 PerlIO_printf( Perl_debug_log, "%5s| ","");
951 PerlIO_printf( Perl_debug_log, "W%4x| ",
952 trie->states[ state ].wordnum
955 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
956 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
958 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
960 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
961 PL_colors[0], PL_colors[1],
962 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
963 PERL_PV_ESCAPE_FIRSTCHAR
965 TRIE_LIST_ITEM(state,charid).forid,
966 (UV)TRIE_LIST_ITEM(state,charid).newstate
969 PerlIO_printf(Perl_debug_log, "\n%*s| ",
970 (int)((depth * 2) + 14), "");
973 PerlIO_printf( Perl_debug_log, "\n");
978 Dumps a fully constructed but uncompressed trie in table form.
979 This is the normal DFA style state transition table, with a few
980 twists to facilitate compression later.
981 Used for debugging make_trie().
984 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
985 HV *widecharmap, AV *revcharmap, U32 next_alloc,
990 SV *sv=sv_newmortal();
991 int colwidth= widecharmap ? 6 : 4;
992 GET_RE_DEBUG_FLAGS_DECL;
995 print out the table precompression so that we can do a visual check
996 that they are identical.
999 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1001 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1002 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1004 PerlIO_printf( Perl_debug_log, "%*s",
1006 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1007 PL_colors[0], PL_colors[1],
1008 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1009 PERL_PV_ESCAPE_FIRSTCHAR
1015 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1017 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1018 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1021 PerlIO_printf( Perl_debug_log, "\n" );
1023 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1025 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1026 (int)depth * 2 + 2,"",
1027 (UV)TRIE_NODENUM( state ) );
1029 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1030 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1032 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1034 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1036 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1037 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1039 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1040 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1047 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1048 startbranch: the first branch in the whole branch sequence
1049 first : start branch of sequence of branch-exact nodes.
1050 May be the same as startbranch
1051 last : Thing following the last branch.
1052 May be the same as tail.
1053 tail : item following the branch sequence
1054 count : words in the sequence
1055 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1056 depth : indent depth
1058 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1060 A trie is an N'ary tree where the branches are determined by digital
1061 decomposition of the key. IE, at the root node you look up the 1st character and
1062 follow that branch repeat until you find the end of the branches. Nodes can be
1063 marked as "accepting" meaning they represent a complete word. Eg:
1067 would convert into the following structure. Numbers represent states, letters
1068 following numbers represent valid transitions on the letter from that state, if
1069 the number is in square brackets it represents an accepting state, otherwise it
1070 will be in parenthesis.
1072 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1076 (1) +-i->(6)-+-s->[7]
1078 +-s->(3)-+-h->(4)-+-e->[5]
1080 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1082 This shows that when matching against the string 'hers' we will begin at state 1
1083 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1084 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1085 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1086 single traverse. We store a mapping from accepting to state to which word was
1087 matched, and then when we have multiple possibilities we try to complete the
1088 rest of the regex in the order in which they occured in the alternation.
1090 The only prior NFA like behaviour that would be changed by the TRIE support is
1091 the silent ignoring of duplicate alternations which are of the form:
1093 / (DUPE|DUPE) X? (?{ ... }) Y /x
1095 Thus EVAL blocks follwing a trie may be called a different number of times with
1096 and without the optimisation. With the optimisations dupes will be silently
1097 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1098 the following demonstrates:
1100 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1102 which prints out 'word' three times, but
1104 'words'=~/(word|word|word)(?{ print $1 })S/
1106 which doesnt print it out at all. This is due to other optimisations kicking in.
1108 Example of what happens on a structural level:
1110 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1112 1: CURLYM[1] {1,32767}(18)
1123 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1124 and should turn into:
1126 1: CURLYM[1] {1,32767}(18)
1128 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1136 Cases where tail != last would be like /(?foo|bar)baz/:
1146 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1147 and would end up looking like:
1150 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1157 d = uvuni_to_utf8_flags(d, uv, 0);
1159 is the recommended Unicode-aware way of saying
1164 #define TRIE_STORE_REVCHAR \
1166 SV *tmp = newSVpvs(""); \
1167 if (UTF) SvUTF8_on(tmp); \
1168 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1169 av_push( revcharmap, tmp ); \
1172 #define TRIE_READ_CHAR STMT_START { \
1176 if ( foldlen > 0 ) { \
1177 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1182 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1183 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1184 foldlen -= UNISKIP( uvc ); \
1185 scan = foldbuf + UNISKIP( uvc ); \
1188 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1198 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1199 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1200 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1201 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1203 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1204 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1205 TRIE_LIST_CUR( state )++; \
1208 #define TRIE_LIST_NEW(state) STMT_START { \
1209 Newxz( trie->states[ state ].trans.list, \
1210 4, reg_trie_trans_le ); \
1211 TRIE_LIST_CUR( state ) = 1; \
1212 TRIE_LIST_LEN( state ) = 4; \
1215 #define TRIE_HANDLE_WORD(state) STMT_START { \
1216 U16 dupe= trie->states[ state ].wordnum; \
1217 regnode * const noper_next = regnext( noper ); \
1219 if (trie->wordlen) \
1220 trie->wordlen[ curword ] = wordlen; \
1222 /* store the word for dumping */ \
1224 if (OP(noper) != NOTHING) \
1225 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1227 tmp = newSVpvn( "", 0 ); \
1228 if ( UTF ) SvUTF8_on( tmp ); \
1229 av_push( trie_words, tmp ); \
1234 if ( noper_next < tail ) { \
1236 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1237 trie->jump[curword] = (U16)(noper_next - convert); \
1239 jumper = noper_next; \
1241 nextbranch= regnext(cur); \
1245 /* So it's a dupe. This means we need to maintain a */\
1246 /* linked-list from the first to the next. */\
1247 /* we only allocate the nextword buffer when there */\
1248 /* a dupe, so first time we have to do the allocation */\
1249 if (!trie->nextword) \
1250 trie->nextword = (U16 *) \
1251 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1252 while ( trie->nextword[dupe] ) \
1253 dupe= trie->nextword[dupe]; \
1254 trie->nextword[dupe]= curword; \
1256 /* we haven't inserted this word yet. */ \
1257 trie->states[ state ].wordnum = curword; \
1262 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1263 ( ( base + charid >= ucharcount \
1264 && base + charid < ubound \
1265 && state == trie->trans[ base - ucharcount + charid ].check \
1266 && trie->trans[ base - ucharcount + charid ].next ) \
1267 ? trie->trans[ base - ucharcount + charid ].next \
1268 : ( state==1 ? special : 0 ) \
1272 #define MADE_JUMP_TRIE 2
1273 #define MADE_EXACT_TRIE 4
1276 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1279 /* first pass, loop through and scan words */
1280 reg_trie_data *trie;
1281 HV *widecharmap = NULL;
1282 AV *revcharmap = newAV();
1284 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1289 regnode *jumper = NULL;
1290 regnode *nextbranch = NULL;
1291 regnode *convert = NULL;
1292 /* we just use folder as a flag in utf8 */
1293 const U8 * const folder = ( flags == EXACTF
1295 : ( flags == EXACTFL
1302 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1303 AV *trie_words = NULL;
1304 /* along with revcharmap, this only used during construction but both are
1305 * useful during debugging so we store them in the struct when debugging.
1308 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1309 STRLEN trie_charcount=0;
1311 SV *re_trie_maxbuff;
1312 GET_RE_DEBUG_FLAGS_DECL;
1314 PERL_UNUSED_ARG(depth);
1317 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1319 trie->startstate = 1;
1320 trie->wordcount = word_count;
1321 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1322 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1323 if (!(UTF && folder))
1324 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1326 trie_words = newAV();
1329 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1330 if (!SvIOK(re_trie_maxbuff)) {
1331 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1334 PerlIO_printf( Perl_debug_log,
1335 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1336 (int)depth * 2 + 2, "",
1337 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1338 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1342 /* Find the node we are going to overwrite */
1343 if ( first == startbranch && OP( last ) != BRANCH ) {
1344 /* whole branch chain */
1347 /* branch sub-chain */
1348 convert = NEXTOPER( first );
1351 /* -- First loop and Setup --
1353 We first traverse the branches and scan each word to determine if it
1354 contains widechars, and how many unique chars there are, this is
1355 important as we have to build a table with at least as many columns as we
1358 We use an array of integers to represent the character codes 0..255
1359 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1360 native representation of the character value as the key and IV's for the
1363 *TODO* If we keep track of how many times each character is used we can
1364 remap the columns so that the table compression later on is more
1365 efficient in terms of memory by ensuring most common value is in the
1366 middle and the least common are on the outside. IMO this would be better
1367 than a most to least common mapping as theres a decent chance the most
1368 common letter will share a node with the least common, meaning the node
1369 will not be compressable. With a middle is most common approach the worst
1370 case is when we have the least common nodes twice.
1374 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1375 regnode * const noper = NEXTOPER( cur );
1376 const U8 *uc = (U8*)STRING( noper );
1377 const U8 * const e = uc + STR_LEN( noper );
1379 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1380 const U8 *scan = (U8*)NULL;
1381 U32 wordlen = 0; /* required init */
1383 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1385 if (OP(noper) == NOTHING) {
1389 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1390 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1391 regardless of encoding */
1393 for ( ; uc < e ; uc += len ) {
1394 TRIE_CHARCOUNT(trie)++;
1398 if ( !trie->charmap[ uvc ] ) {
1399 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1401 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1405 /* store the codepoint in the bitmap, and if its ascii
1406 also store its folded equivelent. */
1407 TRIE_BITMAP_SET(trie,uvc);
1409 /* store the folded codepoint */
1410 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1413 /* store first byte of utf8 representation of
1414 codepoints in the 127 < uvc < 256 range */
1415 if (127 < uvc && uvc < 192) {
1416 TRIE_BITMAP_SET(trie,194);
1417 } else if (191 < uvc ) {
1418 TRIE_BITMAP_SET(trie,195);
1419 /* && uvc < 256 -- we know uvc is < 256 already */
1422 set_bit = 0; /* We've done our bit :-) */
1427 widecharmap = newHV();
1429 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1432 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1434 if ( !SvTRUE( *svpp ) ) {
1435 sv_setiv( *svpp, ++trie->uniquecharcount );
1440 if( cur == first ) {
1443 } else if (chars < trie->minlen) {
1445 } else if (chars > trie->maxlen) {
1449 } /* end first pass */
1450 DEBUG_TRIE_COMPILE_r(
1451 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1452 (int)depth * 2 + 2,"",
1453 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1454 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1455 (int)trie->minlen, (int)trie->maxlen )
1457 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1460 We now know what we are dealing with in terms of unique chars and
1461 string sizes so we can calculate how much memory a naive
1462 representation using a flat table will take. If it's over a reasonable
1463 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1464 conservative but potentially much slower representation using an array
1467 At the end we convert both representations into the same compressed
1468 form that will be used in regexec.c for matching with. The latter
1469 is a form that cannot be used to construct with but has memory
1470 properties similar to the list form and access properties similar
1471 to the table form making it both suitable for fast searches and
1472 small enough that its feasable to store for the duration of a program.
1474 See the comment in the code where the compressed table is produced
1475 inplace from the flat tabe representation for an explanation of how
1476 the compression works.
1481 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1483 Second Pass -- Array Of Lists Representation
1485 Each state will be represented by a list of charid:state records
1486 (reg_trie_trans_le) the first such element holds the CUR and LEN
1487 points of the allocated array. (See defines above).
1489 We build the initial structure using the lists, and then convert
1490 it into the compressed table form which allows faster lookups
1491 (but cant be modified once converted).
1494 STRLEN transcount = 1;
1496 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1497 "%*sCompiling trie using list compiler\n",
1498 (int)depth * 2 + 2, ""));
1500 trie->states = (reg_trie_state *)
1501 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1502 sizeof(reg_trie_state) );
1506 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1508 regnode * const noper = NEXTOPER( cur );
1509 U8 *uc = (U8*)STRING( noper );
1510 const U8 * const e = uc + STR_LEN( noper );
1511 U32 state = 1; /* required init */
1512 U16 charid = 0; /* sanity init */
1513 U8 *scan = (U8*)NULL; /* sanity init */
1514 STRLEN foldlen = 0; /* required init */
1515 U32 wordlen = 0; /* required init */
1516 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1518 if (OP(noper) != NOTHING) {
1519 for ( ; uc < e ; uc += len ) {
1524 charid = trie->charmap[ uvc ];
1526 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1530 charid=(U16)SvIV( *svpp );
1533 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1540 if ( !trie->states[ state ].trans.list ) {
1541 TRIE_LIST_NEW( state );
1543 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1544 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1545 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1550 newstate = next_alloc++;
1551 TRIE_LIST_PUSH( state, charid, newstate );
1556 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1560 TRIE_HANDLE_WORD(state);
1562 } /* end second pass */
1564 /* next alloc is the NEXT state to be allocated */
1565 trie->statecount = next_alloc;
1566 trie->states = (reg_trie_state *)
1567 PerlMemShared_realloc( trie->states,
1569 * sizeof(reg_trie_state) );
1571 /* and now dump it out before we compress it */
1572 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1573 revcharmap, next_alloc,
1577 trie->trans = (reg_trie_trans *)
1578 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1585 for( state=1 ; state < next_alloc ; state ++ ) {
1589 DEBUG_TRIE_COMPILE_MORE_r(
1590 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1594 if (trie->states[state].trans.list) {
1595 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1599 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1600 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1601 if ( forid < minid ) {
1603 } else if ( forid > maxid ) {
1607 if ( transcount < tp + maxid - minid + 1) {
1609 trie->trans = (reg_trie_trans *)
1610 PerlMemShared_realloc( trie->trans,
1612 * sizeof(reg_trie_trans) );
1613 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1615 base = trie->uniquecharcount + tp - minid;
1616 if ( maxid == minid ) {
1618 for ( ; zp < tp ; zp++ ) {
1619 if ( ! trie->trans[ zp ].next ) {
1620 base = trie->uniquecharcount + zp - minid;
1621 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1622 trie->trans[ zp ].check = state;
1628 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1629 trie->trans[ tp ].check = state;
1634 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1635 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1636 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1637 trie->trans[ tid ].check = state;
1639 tp += ( maxid - minid + 1 );
1641 Safefree(trie->states[ state ].trans.list);
1644 DEBUG_TRIE_COMPILE_MORE_r(
1645 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1648 trie->states[ state ].trans.base=base;
1650 trie->lasttrans = tp + 1;
1654 Second Pass -- Flat Table Representation.
1656 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1657 We know that we will need Charcount+1 trans at most to store the data
1658 (one row per char at worst case) So we preallocate both structures
1659 assuming worst case.
1661 We then construct the trie using only the .next slots of the entry
1664 We use the .check field of the first entry of the node temporarily to
1665 make compression both faster and easier by keeping track of how many non
1666 zero fields are in the node.
1668 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1671 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1672 number representing the first entry of the node, and state as a
1673 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1674 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1675 are 2 entrys per node. eg:
1683 The table is internally in the right hand, idx form. However as we also
1684 have to deal with the states array which is indexed by nodenum we have to
1685 use TRIE_NODENUM() to convert.
1688 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1689 "%*sCompiling trie using table compiler\n",
1690 (int)depth * 2 + 2, ""));
1692 trie->trans = (reg_trie_trans *)
1693 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1694 * trie->uniquecharcount + 1,
1695 sizeof(reg_trie_trans) );
1696 trie->states = (reg_trie_state *)
1697 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1698 sizeof(reg_trie_state) );
1699 next_alloc = trie->uniquecharcount + 1;
1702 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1704 regnode * const noper = NEXTOPER( cur );
1705 const U8 *uc = (U8*)STRING( noper );
1706 const U8 * const e = uc + STR_LEN( noper );
1708 U32 state = 1; /* required init */
1710 U16 charid = 0; /* sanity init */
1711 U32 accept_state = 0; /* sanity init */
1712 U8 *scan = (U8*)NULL; /* sanity init */
1714 STRLEN foldlen = 0; /* required init */
1715 U32 wordlen = 0; /* required init */
1716 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1718 if ( OP(noper) != NOTHING ) {
1719 for ( ; uc < e ; uc += len ) {
1724 charid = trie->charmap[ uvc ];
1726 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1727 charid = svpp ? (U16)SvIV(*svpp) : 0;
1731 if ( !trie->trans[ state + charid ].next ) {
1732 trie->trans[ state + charid ].next = next_alloc;
1733 trie->trans[ state ].check++;
1734 next_alloc += trie->uniquecharcount;
1736 state = trie->trans[ state + charid ].next;
1738 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1740 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1743 accept_state = TRIE_NODENUM( state );
1744 TRIE_HANDLE_WORD(accept_state);
1746 } /* end second pass */
1748 /* and now dump it out before we compress it */
1749 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1751 next_alloc, depth+1));
1755 * Inplace compress the table.*
1757 For sparse data sets the table constructed by the trie algorithm will
1758 be mostly 0/FAIL transitions or to put it another way mostly empty.
1759 (Note that leaf nodes will not contain any transitions.)
1761 This algorithm compresses the tables by eliminating most such
1762 transitions, at the cost of a modest bit of extra work during lookup:
1764 - Each states[] entry contains a .base field which indicates the
1765 index in the state[] array wheres its transition data is stored.
1767 - If .base is 0 there are no valid transitions from that node.
1769 - If .base is nonzero then charid is added to it to find an entry in
1772 -If trans[states[state].base+charid].check!=state then the
1773 transition is taken to be a 0/Fail transition. Thus if there are fail
1774 transitions at the front of the node then the .base offset will point
1775 somewhere inside the previous nodes data (or maybe even into a node
1776 even earlier), but the .check field determines if the transition is
1780 The following process inplace converts the table to the compressed
1781 table: We first do not compress the root node 1,and mark its all its
1782 .check pointers as 1 and set its .base pointer as 1 as well. This
1783 allows to do a DFA construction from the compressed table later, and
1784 ensures that any .base pointers we calculate later are greater than
1787 - We set 'pos' to indicate the first entry of the second node.
1789 - We then iterate over the columns of the node, finding the first and
1790 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1791 and set the .check pointers accordingly, and advance pos
1792 appropriately and repreat for the next node. Note that when we copy
1793 the next pointers we have to convert them from the original
1794 NODEIDX form to NODENUM form as the former is not valid post
1797 - If a node has no transitions used we mark its base as 0 and do not
1798 advance the pos pointer.
1800 - If a node only has one transition we use a second pointer into the
1801 structure to fill in allocated fail transitions from other states.
1802 This pointer is independent of the main pointer and scans forward
1803 looking for null transitions that are allocated to a state. When it
1804 finds one it writes the single transition into the "hole". If the
1805 pointer doesnt find one the single transition is appended as normal.
1807 - Once compressed we can Renew/realloc the structures to release the
1810 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1811 specifically Fig 3.47 and the associated pseudocode.
1815 const U32 laststate = TRIE_NODENUM( next_alloc );
1818 trie->statecount = laststate;
1820 for ( state = 1 ; state < laststate ; state++ ) {
1822 const U32 stateidx = TRIE_NODEIDX( state );
1823 const U32 o_used = trie->trans[ stateidx ].check;
1824 U32 used = trie->trans[ stateidx ].check;
1825 trie->trans[ stateidx ].check = 0;
1827 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1828 if ( flag || trie->trans[ stateidx + charid ].next ) {
1829 if ( trie->trans[ stateidx + charid ].next ) {
1831 for ( ; zp < pos ; zp++ ) {
1832 if ( ! trie->trans[ zp ].next ) {
1836 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1837 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1838 trie->trans[ zp ].check = state;
1839 if ( ++zp > pos ) pos = zp;
1846 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1848 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1849 trie->trans[ pos ].check = state;
1854 trie->lasttrans = pos + 1;
1855 trie->states = (reg_trie_state *)
1856 PerlMemShared_realloc( trie->states, laststate
1857 * sizeof(reg_trie_state) );
1858 DEBUG_TRIE_COMPILE_MORE_r(
1859 PerlIO_printf( Perl_debug_log,
1860 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1861 (int)depth * 2 + 2,"",
1862 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1865 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1868 } /* end table compress */
1870 DEBUG_TRIE_COMPILE_MORE_r(
1871 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1872 (int)depth * 2 + 2, "",
1873 (UV)trie->statecount,
1874 (UV)trie->lasttrans)
1876 /* resize the trans array to remove unused space */
1877 trie->trans = (reg_trie_trans *)
1878 PerlMemShared_realloc( trie->trans, trie->lasttrans
1879 * sizeof(reg_trie_trans) );
1881 /* and now dump out the compressed format */
1882 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1884 { /* Modify the program and insert the new TRIE node*/
1885 U8 nodetype =(U8)(flags & 0xFF);
1889 regnode *optimize = NULL;
1890 #ifdef RE_TRACK_PATTERN_OFFSETS
1893 U32 mjd_nodelen = 0;
1894 #endif /* RE_TRACK_PATTERN_OFFSETS */
1895 #endif /* DEBUGGING */
1897 This means we convert either the first branch or the first Exact,
1898 depending on whether the thing following (in 'last') is a branch
1899 or not and whther first is the startbranch (ie is it a sub part of
1900 the alternation or is it the whole thing.)
1901 Assuming its a sub part we conver the EXACT otherwise we convert
1902 the whole branch sequence, including the first.
1904 /* Find the node we are going to overwrite */
1905 if ( first != startbranch || OP( last ) == BRANCH ) {
1906 /* branch sub-chain */
1907 NEXT_OFF( first ) = (U16)(last - first);
1908 #ifdef RE_TRACK_PATTERN_OFFSETS
1910 mjd_offset= Node_Offset((convert));
1911 mjd_nodelen= Node_Length((convert));
1914 /* whole branch chain */
1916 #ifdef RE_TRACK_PATTERN_OFFSETS
1919 const regnode *nop = NEXTOPER( convert );
1920 mjd_offset= Node_Offset((nop));
1921 mjd_nodelen= Node_Length((nop));
1925 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1926 (int)depth * 2 + 2, "",
1927 (UV)mjd_offset, (UV)mjd_nodelen)
1930 /* But first we check to see if there is a common prefix we can
1931 split out as an EXACT and put in front of the TRIE node. */
1932 trie->startstate= 1;
1933 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1935 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1939 const U32 base = trie->states[ state ].trans.base;
1941 if ( trie->states[state].wordnum )
1944 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1945 if ( ( base + ofs >= trie->uniquecharcount ) &&
1946 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1947 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1949 if ( ++count > 1 ) {
1950 SV **tmp = av_fetch( revcharmap, ofs, 0);
1951 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1952 if ( state == 1 ) break;
1954 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1956 PerlIO_printf(Perl_debug_log,
1957 "%*sNew Start State=%"UVuf" Class: [",
1958 (int)depth * 2 + 2, "",
1961 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1962 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1964 TRIE_BITMAP_SET(trie,*ch);
1966 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1968 PerlIO_printf(Perl_debug_log, (char*)ch)
1972 TRIE_BITMAP_SET(trie,*ch);
1974 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1975 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1981 SV **tmp = av_fetch( revcharmap, idx, 0);
1983 char *ch = SvPV( *tmp, len );
1985 SV *sv=sv_newmortal();
1986 PerlIO_printf( Perl_debug_log,
1987 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1988 (int)depth * 2 + 2, "",
1990 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1991 PL_colors[0], PL_colors[1],
1992 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1993 PERL_PV_ESCAPE_FIRSTCHAR
1998 OP( convert ) = nodetype;
1999 str=STRING(convert);
2002 STR_LEN(convert) += len;
2008 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2014 regnode *n = convert+NODE_SZ_STR(convert);
2015 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2016 trie->startstate = state;
2017 trie->minlen -= (state - 1);
2018 trie->maxlen -= (state - 1);
2020 regnode *fix = convert;
2021 U32 word = trie->wordcount;
2023 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2024 while( ++fix < n ) {
2025 Set_Node_Offset_Length(fix, 0, 0);
2028 SV ** const tmp = av_fetch( trie_words, word, 0 );
2030 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2031 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2033 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2040 NEXT_OFF(convert) = (U16)(tail - convert);
2041 DEBUG_r(optimize= n);
2047 if ( trie->maxlen ) {
2048 NEXT_OFF( convert ) = (U16)(tail - convert);
2049 ARG_SET( convert, data_slot );
2050 /* Store the offset to the first unabsorbed branch in
2051 jump[0], which is otherwise unused by the jump logic.
2052 We use this when dumping a trie and during optimisation. */
2054 trie->jump[0] = (U16)(nextbranch - convert);
2057 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2058 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2060 OP( convert ) = TRIEC;
2061 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2062 PerlMemShared_free(trie->bitmap);
2065 OP( convert ) = TRIE;
2067 /* store the type in the flags */
2068 convert->flags = nodetype;
2072 + regarglen[ OP( convert ) ];
2074 /* XXX We really should free up the resource in trie now,
2075 as we won't use them - (which resources?) dmq */
2077 /* needed for dumping*/
2078 DEBUG_r(if (optimize) {
2079 regnode *opt = convert;
2081 while ( ++opt < optimize) {
2082 Set_Node_Offset_Length(opt,0,0);
2085 Try to clean up some of the debris left after the
2088 while( optimize < jumper ) {
2089 mjd_nodelen += Node_Length((optimize));
2090 OP( optimize ) = OPTIMIZED;
2091 Set_Node_Offset_Length(optimize,0,0);
2094 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2096 } /* end node insert */
2097 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2099 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2100 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2102 SvREFCNT_dec(revcharmap);
2106 : trie->startstate>1
2112 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2114 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2116 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2117 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2120 We find the fail state for each state in the trie, this state is the longest proper
2121 suffix of the current states 'word' that is also a proper prefix of another word in our
2122 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2123 the DFA not to have to restart after its tried and failed a word at a given point, it
2124 simply continues as though it had been matching the other word in the first place.
2126 'abcdgu'=~/abcdefg|cdgu/
2127 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2128 fail, which would bring use to the state representing 'd' in the second word where we would
2129 try 'g' and succeed, prodceding to match 'cdgu'.
2131 /* add a fail transition */
2132 const U32 trie_offset = ARG(source);
2133 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2135 const U32 ucharcount = trie->uniquecharcount;
2136 const U32 numstates = trie->statecount;
2137 const U32 ubound = trie->lasttrans + ucharcount;
2141 U32 base = trie->states[ 1 ].trans.base;
2144 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2145 GET_RE_DEBUG_FLAGS_DECL;
2147 PERL_UNUSED_ARG(depth);
2151 ARG_SET( stclass, data_slot );
2152 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2153 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2154 aho->trie=trie_offset;
2155 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2156 Copy( trie->states, aho->states, numstates, reg_trie_state );
2157 Newxz( q, numstates, U32);
2158 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2161 /* initialize fail[0..1] to be 1 so that we always have
2162 a valid final fail state */
2163 fail[ 0 ] = fail[ 1 ] = 1;
2165 for ( charid = 0; charid < ucharcount ; charid++ ) {
2166 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2168 q[ q_write ] = newstate;
2169 /* set to point at the root */
2170 fail[ q[ q_write++ ] ]=1;
2173 while ( q_read < q_write) {
2174 const U32 cur = q[ q_read++ % numstates ];
2175 base = trie->states[ cur ].trans.base;
2177 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2178 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2180 U32 fail_state = cur;
2183 fail_state = fail[ fail_state ];
2184 fail_base = aho->states[ fail_state ].trans.base;
2185 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2187 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2188 fail[ ch_state ] = fail_state;
2189 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2191 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2193 q[ q_write++ % numstates] = ch_state;
2197 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2198 when we fail in state 1, this allows us to use the
2199 charclass scan to find a valid start char. This is based on the principle
2200 that theres a good chance the string being searched contains lots of stuff
2201 that cant be a start char.
2203 fail[ 0 ] = fail[ 1 ] = 0;
2204 DEBUG_TRIE_COMPILE_r({
2205 PerlIO_printf(Perl_debug_log,
2206 "%*sStclass Failtable (%"UVuf" states): 0",
2207 (int)(depth * 2), "", (UV)numstates
2209 for( q_read=1; q_read<numstates; q_read++ ) {
2210 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2212 PerlIO_printf(Perl_debug_log, "\n");
2215 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2220 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2221 * These need to be revisited when a newer toolchain becomes available.
2223 #if defined(__sparc64__) && defined(__GNUC__)
2224 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2225 # undef SPARC64_GCC_WORKAROUND
2226 # define SPARC64_GCC_WORKAROUND 1
2230 #define DEBUG_PEEP(str,scan,depth) \
2231 DEBUG_OPTIMISE_r({if (scan){ \
2232 SV * const mysv=sv_newmortal(); \
2233 regnode *Next = regnext(scan); \
2234 regprop(RExC_rx, mysv, scan); \
2235 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2236 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2237 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2244 #define JOIN_EXACT(scan,min,flags) \
2245 if (PL_regkind[OP(scan)] == EXACT) \
2246 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2249 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2250 /* Merge several consecutive EXACTish nodes into one. */
2251 regnode *n = regnext(scan);
2253 regnode *next = scan + NODE_SZ_STR(scan);
2257 regnode *stop = scan;
2258 GET_RE_DEBUG_FLAGS_DECL;
2260 PERL_UNUSED_ARG(depth);
2262 #ifndef EXPERIMENTAL_INPLACESCAN
2263 PERL_UNUSED_ARG(flags);
2264 PERL_UNUSED_ARG(val);
2266 DEBUG_PEEP("join",scan,depth);
2268 /* Skip NOTHING, merge EXACT*. */
2270 ( PL_regkind[OP(n)] == NOTHING ||
2271 (stringok && (OP(n) == OP(scan))))
2273 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2275 if (OP(n) == TAIL || n > next)
2277 if (PL_regkind[OP(n)] == NOTHING) {
2278 DEBUG_PEEP("skip:",n,depth);
2279 NEXT_OFF(scan) += NEXT_OFF(n);
2280 next = n + NODE_STEP_REGNODE;
2287 else if (stringok) {
2288 const unsigned int oldl = STR_LEN(scan);
2289 regnode * const nnext = regnext(n);
2291 DEBUG_PEEP("merg",n,depth);
2294 if (oldl + STR_LEN(n) > U8_MAX)
2296 NEXT_OFF(scan) += NEXT_OFF(n);
2297 STR_LEN(scan) += STR_LEN(n);
2298 next = n + NODE_SZ_STR(n);
2299 /* Now we can overwrite *n : */
2300 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2308 #ifdef EXPERIMENTAL_INPLACESCAN
2309 if (flags && !NEXT_OFF(n)) {
2310 DEBUG_PEEP("atch", val, depth);
2311 if (reg_off_by_arg[OP(n)]) {
2312 ARG_SET(n, val - n);
2315 NEXT_OFF(n) = val - n;
2322 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2324 Two problematic code points in Unicode casefolding of EXACT nodes:
2326 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2327 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2333 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2334 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2336 This means that in case-insensitive matching (or "loose matching",
2337 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2338 length of the above casefolded versions) can match a target string
2339 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2340 This would rather mess up the minimum length computation.
2342 What we'll do is to look for the tail four bytes, and then peek
2343 at the preceding two bytes to see whether we need to decrease
2344 the minimum length by four (six minus two).
2346 Thanks to the design of UTF-8, there cannot be false matches:
2347 A sequence of valid UTF-8 bytes cannot be a subsequence of
2348 another valid sequence of UTF-8 bytes.
2351 char * const s0 = STRING(scan), *s, *t;
2352 char * const s1 = s0 + STR_LEN(scan) - 1;
2353 char * const s2 = s1 - 4;
2354 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2355 const char t0[] = "\xaf\x49\xaf\x42";
2357 const char t0[] = "\xcc\x88\xcc\x81";
2359 const char * const t1 = t0 + 3;
2362 s < s2 && (t = ninstr(s, s1, t0, t1));
2365 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2366 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2368 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2369 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2377 n = scan + NODE_SZ_STR(scan);
2379 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2386 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2390 /* REx optimizer. Converts nodes into quickier variants "in place".
2391 Finds fixed substrings. */
2393 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2394 to the position after last scanned or to NULL. */
2396 #define INIT_AND_WITHP \
2397 assert(!and_withp); \
2398 Newx(and_withp,1,struct regnode_charclass_class); \
2399 SAVEFREEPV(and_withp)
2401 /* this is a chain of data about sub patterns we are processing that
2402 need to be handled seperately/specially in study_chunk. Its so
2403 we can simulate recursion without losing state. */
2405 typedef struct scan_frame {
2406 regnode *last; /* last node to process in this frame */
2407 regnode *next; /* next node to process when last is reached */
2408 struct scan_frame *prev; /*previous frame*/
2409 I32 stop; /* what stopparen do we use */
2413 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2415 #define CASE_SYNST_FNC(nAmE) \
2417 if (flags & SCF_DO_STCLASS_AND) { \
2418 for (value = 0; value < 256; value++) \
2419 if (!is_ ## nAmE ## _cp(value)) \
2420 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2423 for (value = 0; value < 256; value++) \
2424 if (is_ ## nAmE ## _cp(value)) \
2425 ANYOF_BITMAP_SET(data->start_class, value); \
2429 if (flags & SCF_DO_STCLASS_AND) { \
2430 for (value = 0; value < 256; value++) \
2431 if (is_ ## nAmE ## _cp(value)) \
2432 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2435 for (value = 0; value < 256; value++) \
2436 if (!is_ ## nAmE ## _cp(value)) \
2437 ANYOF_BITMAP_SET(data->start_class, value); \
2444 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2445 I32 *minlenp, I32 *deltap,
2450 struct regnode_charclass_class *and_withp,
2451 U32 flags, U32 depth)
2452 /* scanp: Start here (read-write). */
2453 /* deltap: Write maxlen-minlen here. */
2454 /* last: Stop before this one. */
2455 /* data: string data about the pattern */
2456 /* stopparen: treat close N as END */
2457 /* recursed: which subroutines have we recursed into */
2458 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2461 I32 min = 0, pars = 0, code;
2462 regnode *scan = *scanp, *next;
2464 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2465 int is_inf_internal = 0; /* The studied chunk is infinite */
2466 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2467 scan_data_t data_fake;
2468 SV *re_trie_maxbuff = NULL;
2469 regnode *first_non_open = scan;
2470 I32 stopmin = I32_MAX;
2471 scan_frame *frame = NULL;
2473 GET_RE_DEBUG_FLAGS_DECL;
2476 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2480 while (first_non_open && OP(first_non_open) == OPEN)
2481 first_non_open=regnext(first_non_open);
2486 while ( scan && OP(scan) != END && scan < last ){
2487 /* Peephole optimizer: */
2488 DEBUG_STUDYDATA("Peep:", data,depth);
2489 DEBUG_PEEP("Peep",scan,depth);
2490 JOIN_EXACT(scan,&min,0);
2492 /* Follow the next-chain of the current node and optimize
2493 away all the NOTHINGs from it. */
2494 if (OP(scan) != CURLYX) {
2495 const int max = (reg_off_by_arg[OP(scan)]
2497 /* I32 may be smaller than U16 on CRAYs! */
2498 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2499 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2503 /* Skip NOTHING and LONGJMP. */
2504 while ((n = regnext(n))
2505 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2506 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2507 && off + noff < max)
2509 if (reg_off_by_arg[OP(scan)])
2512 NEXT_OFF(scan) = off;
2517 /* The principal pseudo-switch. Cannot be a switch, since we
2518 look into several different things. */
2519 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2520 || OP(scan) == IFTHEN) {
2521 next = regnext(scan);
2523 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2525 if (OP(next) == code || code == IFTHEN) {
2526 /* NOTE - There is similar code to this block below for handling
2527 TRIE nodes on a re-study. If you change stuff here check there
2529 I32 max1 = 0, min1 = I32_MAX, num = 0;
2530 struct regnode_charclass_class accum;
2531 regnode * const startbranch=scan;
2533 if (flags & SCF_DO_SUBSTR)
2534 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2535 if (flags & SCF_DO_STCLASS)
2536 cl_init_zero(pRExC_state, &accum);
2538 while (OP(scan) == code) {
2539 I32 deltanext, minnext, f = 0, fake;
2540 struct regnode_charclass_class this_class;
2543 data_fake.flags = 0;
2545 data_fake.whilem_c = data->whilem_c;
2546 data_fake.last_closep = data->last_closep;
2549 data_fake.last_closep = &fake;
2551 data_fake.pos_delta = delta;
2552 next = regnext(scan);
2553 scan = NEXTOPER(scan);
2555 scan = NEXTOPER(scan);
2556 if (flags & SCF_DO_STCLASS) {
2557 cl_init(pRExC_state, &this_class);
2558 data_fake.start_class = &this_class;
2559 f = SCF_DO_STCLASS_AND;
2561 if (flags & SCF_WHILEM_VISITED_POS)
2562 f |= SCF_WHILEM_VISITED_POS;
2564 /* we suppose the run is continuous, last=next...*/
2565 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2567 stopparen, recursed, NULL, f,depth+1);
2570 if (max1 < minnext + deltanext)
2571 max1 = minnext + deltanext;
2572 if (deltanext == I32_MAX)
2573 is_inf = is_inf_internal = 1;
2575 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2577 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2578 if ( stopmin > minnext)
2579 stopmin = min + min1;
2580 flags &= ~SCF_DO_SUBSTR;
2582 data->flags |= SCF_SEEN_ACCEPT;
2585 if (data_fake.flags & SF_HAS_EVAL)
2586 data->flags |= SF_HAS_EVAL;
2587 data->whilem_c = data_fake.whilem_c;
2589 if (flags & SCF_DO_STCLASS)
2590 cl_or(pRExC_state, &accum, &this_class);
2592 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2594 if (flags & SCF_DO_SUBSTR) {
2595 data->pos_min += min1;
2596 data->pos_delta += max1 - min1;
2597 if (max1 != min1 || is_inf)
2598 data->longest = &(data->longest_float);
2601 delta += max1 - min1;
2602 if (flags & SCF_DO_STCLASS_OR) {
2603 cl_or(pRExC_state, data->start_class, &accum);
2605 cl_and(data->start_class, and_withp);
2606 flags &= ~SCF_DO_STCLASS;
2609 else if (flags & SCF_DO_STCLASS_AND) {
2611 cl_and(data->start_class, &accum);
2612 flags &= ~SCF_DO_STCLASS;
2615 /* Switch to OR mode: cache the old value of
2616 * data->start_class */
2618 StructCopy(data->start_class, and_withp,
2619 struct regnode_charclass_class);
2620 flags &= ~SCF_DO_STCLASS_AND;
2621 StructCopy(&accum, data->start_class,
2622 struct regnode_charclass_class);
2623 flags |= SCF_DO_STCLASS_OR;
2624 data->start_class->flags |= ANYOF_EOS;
2628 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2631 Assuming this was/is a branch we are dealing with: 'scan' now
2632 points at the item that follows the branch sequence, whatever
2633 it is. We now start at the beginning of the sequence and look
2640 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2642 If we can find such a subseqence we need to turn the first
2643 element into a trie and then add the subsequent branch exact
2644 strings to the trie.
2648 1. patterns where the whole set of branch can be converted.
2650 2. patterns where only a subset can be converted.
2652 In case 1 we can replace the whole set with a single regop
2653 for the trie. In case 2 we need to keep the start and end
2656 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2657 becomes BRANCH TRIE; BRANCH X;
2659 There is an additional case, that being where there is a
2660 common prefix, which gets split out into an EXACT like node
2661 preceding the TRIE node.
2663 If x(1..n)==tail then we can do a simple trie, if not we make
2664 a "jump" trie, such that when we match the appropriate word
2665 we "jump" to the appopriate tail node. Essentailly we turn
2666 a nested if into a case structure of sorts.
2671 if (!re_trie_maxbuff) {
2672 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2673 if (!SvIOK(re_trie_maxbuff))
2674 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2676 if ( SvIV(re_trie_maxbuff)>=0 ) {
2678 regnode *first = (regnode *)NULL;
2679 regnode *last = (regnode *)NULL;
2680 regnode *tail = scan;
2685 SV * const mysv = sv_newmortal(); /* for dumping */
2687 /* var tail is used because there may be a TAIL
2688 regop in the way. Ie, the exacts will point to the
2689 thing following the TAIL, but the last branch will
2690 point at the TAIL. So we advance tail. If we
2691 have nested (?:) we may have to move through several
2695 while ( OP( tail ) == TAIL ) {
2696 /* this is the TAIL generated by (?:) */
2697 tail = regnext( tail );
2702 regprop(RExC_rx, mysv, tail );
2703 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2704 (int)depth * 2 + 2, "",
2705 "Looking for TRIE'able sequences. Tail node is: ",
2706 SvPV_nolen_const( mysv )
2712 step through the branches, cur represents each
2713 branch, noper is the first thing to be matched
2714 as part of that branch and noper_next is the
2715 regnext() of that node. if noper is an EXACT
2716 and noper_next is the same as scan (our current
2717 position in the regex) then the EXACT branch is
2718 a possible optimization target. Once we have
2719 two or more consequetive such branches we can
2720 create a trie of the EXACT's contents and stich
2721 it in place. If the sequence represents all of
2722 the branches we eliminate the whole thing and
2723 replace it with a single TRIE. If it is a
2724 subsequence then we need to stitch it in. This
2725 means the first branch has to remain, and needs
2726 to be repointed at the item on the branch chain
2727 following the last branch optimized. This could
2728 be either a BRANCH, in which case the
2729 subsequence is internal, or it could be the
2730 item following the branch sequence in which
2731 case the subsequence is at the end.
2735 /* dont use tail as the end marker for this traverse */
2736 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2737 regnode * const noper = NEXTOPER( cur );
2738 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2739 regnode * const noper_next = regnext( noper );
2743 regprop(RExC_rx, mysv, cur);
2744 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2745 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2747 regprop(RExC_rx, mysv, noper);
2748 PerlIO_printf( Perl_debug_log, " -> %s",
2749 SvPV_nolen_const(mysv));
2752 regprop(RExC_rx, mysv, noper_next );
2753 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2754 SvPV_nolen_const(mysv));
2756 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2757 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2759 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2760 : PL_regkind[ OP( noper ) ] == EXACT )
2761 || OP(noper) == NOTHING )
2763 && noper_next == tail
2768 if ( !first || optype == NOTHING ) {
2769 if (!first) first = cur;
2770 optype = OP( noper );
2776 make_trie( pRExC_state,
2777 startbranch, first, cur, tail, count,
2780 if ( PL_regkind[ OP( noper ) ] == EXACT
2782 && noper_next == tail
2787 optype = OP( noper );
2797 regprop(RExC_rx, mysv, cur);
2798 PerlIO_printf( Perl_debug_log,
2799 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2800 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2804 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2805 #ifdef TRIE_STUDY_OPT
2806 if ( ((made == MADE_EXACT_TRIE &&
2807 startbranch == first)
2808 || ( first_non_open == first )) &&
2810 flags |= SCF_TRIE_RESTUDY;
2811 if ( startbranch == first
2814 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2824 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2825 scan = NEXTOPER(NEXTOPER(scan));
2826 } else /* single branch is optimized. */
2827 scan = NEXTOPER(scan);
2829 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2830 scan_frame *newframe = NULL;
2835 if (OP(scan) != SUSPEND) {
2836 /* set the pointer */
2837 if (OP(scan) == GOSUB) {
2839 RExC_recurse[ARG2L(scan)] = scan;
2840 start = RExC_open_parens[paren-1];
2841 end = RExC_close_parens[paren-1];
2844 start = RExC_rxi->program + 1;
2848 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2849 SAVEFREEPV(recursed);
2851 if (!PAREN_TEST(recursed,paren+1)) {
2852 PAREN_SET(recursed,paren+1);
2853 Newx(newframe,1,scan_frame);
2855 if (flags & SCF_DO_SUBSTR) {
2856 SCAN_COMMIT(pRExC_state,data,minlenp);
2857 data->longest = &(data->longest_float);
2859 is_inf = is_inf_internal = 1;
2860 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2861 cl_anything(pRExC_state, data->start_class);
2862 flags &= ~SCF_DO_STCLASS;
2865 Newx(newframe,1,scan_frame);
2868 end = regnext(scan);
2873 SAVEFREEPV(newframe);
2874 newframe->next = regnext(scan);
2875 newframe->last = last;
2876 newframe->stop = stopparen;
2877 newframe->prev = frame;
2887 else if (OP(scan) == EXACT) {
2888 I32 l = STR_LEN(scan);
2891 const U8 * const s = (U8*)STRING(scan);
2892 l = utf8_length(s, s + l);
2893 uc = utf8_to_uvchr(s, NULL);
2895 uc = *((U8*)STRING(scan));
2898 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2899 /* The code below prefers earlier match for fixed
2900 offset, later match for variable offset. */
2901 if (data->last_end == -1) { /* Update the start info. */
2902 data->last_start_min = data->pos_min;
2903 data->last_start_max = is_inf
2904 ? I32_MAX : data->pos_min + data->pos_delta;
2906 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2908 SvUTF8_on(data->last_found);
2910 SV * const sv = data->last_found;
2911 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2912 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2913 if (mg && mg->mg_len >= 0)
2914 mg->mg_len += utf8_length((U8*)STRING(scan),
2915 (U8*)STRING(scan)+STR_LEN(scan));
2917 data->last_end = data->pos_min + l;
2918 data->pos_min += l; /* As in the first entry. */
2919 data->flags &= ~SF_BEFORE_EOL;
2921 if (flags & SCF_DO_STCLASS_AND) {
2922 /* Check whether it is compatible with what we know already! */
2926 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2927 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2928 && (!(data->start_class->flags & ANYOF_FOLD)
2929 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2932 ANYOF_CLASS_ZERO(data->start_class);
2933 ANYOF_BITMAP_ZERO(data->start_class);
2935 ANYOF_BITMAP_SET(data->start_class, uc);
2936 data->start_class->flags &= ~ANYOF_EOS;
2938 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2940 else if (flags & SCF_DO_STCLASS_OR) {
2941 /* false positive possible if the class is case-folded */
2943 ANYOF_BITMAP_SET(data->start_class, uc);
2945 data->start_class->flags |= ANYOF_UNICODE_ALL;
2946 data->start_class->flags &= ~ANYOF_EOS;
2947 cl_and(data->start_class, and_withp);
2949 flags &= ~SCF_DO_STCLASS;
2951 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2952 I32 l = STR_LEN(scan);
2953 UV uc = *((U8*)STRING(scan));
2955 /* Search for fixed substrings supports EXACT only. */
2956 if (flags & SCF_DO_SUBSTR) {
2958 SCAN_COMMIT(pRExC_state, data, minlenp);
2961 const U8 * const s = (U8 *)STRING(scan);
2962 l = utf8_length(s, s + l);
2963 uc = utf8_to_uvchr(s, NULL);
2966 if (flags & SCF_DO_SUBSTR)
2968 if (flags & SCF_DO_STCLASS_AND) {
2969 /* Check whether it is compatible with what we know already! */
2973 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2974 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2975 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2977 ANYOF_CLASS_ZERO(data->start_class);
2978 ANYOF_BITMAP_ZERO(data->start_class);
2980 ANYOF_BITMAP_SET(data->start_class, uc);
2981 data->start_class->flags &= ~ANYOF_EOS;
2982 data->start_class->flags |= ANYOF_FOLD;
2983 if (OP(scan) == EXACTFL)
2984 data->start_class->flags |= ANYOF_LOCALE;
2987 else if (flags & SCF_DO_STCLASS_OR) {
2988 if (data->start_class->flags & ANYOF_FOLD) {
2989 /* false positive possible if the class is case-folded.
2990 Assume that the locale settings are the same... */
2992 ANYOF_BITMAP_SET(data->start_class, uc);
2993 data->start_class->flags &= ~ANYOF_EOS;
2995 cl_and(data->start_class, and_withp);
2997 flags &= ~SCF_DO_STCLASS;
2999 else if (strchr((const char*)PL_varies,OP(scan))) {
3000 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3001 I32 f = flags, pos_before = 0;
3002 regnode * const oscan = scan;
3003 struct regnode_charclass_class this_class;
3004 struct regnode_charclass_class *oclass = NULL;
3005 I32 next_is_eval = 0;
3007 switch (PL_regkind[OP(scan)]) {
3008 case WHILEM: /* End of (?:...)* . */
3009 scan = NEXTOPER(scan);
3012 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3013 next = NEXTOPER(scan);
3014 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3016 maxcount = REG_INFTY;
3017 next = regnext(scan);
3018 scan = NEXTOPER(scan);
3022 if (flags & SCF_DO_SUBSTR)
3027 if (flags & SCF_DO_STCLASS) {
3029 maxcount = REG_INFTY;
3030 next = regnext(scan);
3031 scan = NEXTOPER(scan);
3034 is_inf = is_inf_internal = 1;
3035 scan = regnext(scan);
3036 if (flags & SCF_DO_SUBSTR) {
3037 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3038 data->longest = &(data->longest_float);
3040 goto optimize_curly_tail;
3042 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3043 && (scan->flags == stopparen))
3048 mincount = ARG1(scan);
3049 maxcount = ARG2(scan);
3051 next = regnext(scan);
3052 if (OP(scan) == CURLYX) {
3053 I32 lp = (data ? *(data->last_closep) : 0);
3054 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3056 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3057 next_is_eval = (OP(scan) == EVAL);
3059 if (flags & SCF_DO_SUBSTR) {
3060 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3061 pos_before = data->pos_min;
3065 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3067 data->flags |= SF_IS_INF;
3069 if (flags & SCF_DO_STCLASS) {
3070 cl_init(pRExC_state, &this_class);
3071 oclass = data->start_class;
3072 data->start_class = &this_class;
3073 f |= SCF_DO_STCLASS_AND;
3074 f &= ~SCF_DO_STCLASS_OR;
3076 /* These are the cases when once a subexpression
3077 fails at a particular position, it cannot succeed
3078 even after backtracking at the enclosing scope.
3080 XXXX what if minimal match and we are at the
3081 initial run of {n,m}? */
3082 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3083 f &= ~SCF_WHILEM_VISITED_POS;
3085 /* This will finish on WHILEM, setting scan, or on NULL: */
3086 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3087 last, data, stopparen, recursed, NULL,
3089 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3091 if (flags & SCF_DO_STCLASS)
3092 data->start_class = oclass;
3093 if (mincount == 0 || minnext == 0) {
3094 if (flags & SCF_DO_STCLASS_OR) {
3095 cl_or(pRExC_state, data->start_class, &this_class);
3097 else if (flags & SCF_DO_STCLASS_AND) {
3098 /* Switch to OR mode: cache the old value of
3099 * data->start_class */
3101 StructCopy(data->start_class, and_withp,
3102 struct regnode_charclass_class);
3103 flags &= ~SCF_DO_STCLASS_AND;
3104 StructCopy(&this_class, data->start_class,
3105 struct regnode_charclass_class);
3106 flags |= SCF_DO_STCLASS_OR;
3107 data->start_class->flags |= ANYOF_EOS;
3109 } else { /* Non-zero len */
3110 if (flags & SCF_DO_STCLASS_OR) {
3111 cl_or(pRExC_state, data->start_class, &this_class);
3112 cl_and(data->start_class, and_withp);
3114 else if (flags & SCF_DO_STCLASS_AND)
3115 cl_and(data->start_class, &this_class);
3116 flags &= ~SCF_DO_STCLASS;
3118 if (!scan) /* It was not CURLYX, but CURLY. */
3120 if ( /* ? quantifier ok, except for (?{ ... }) */
3121 (next_is_eval || !(mincount == 0 && maxcount == 1))
3122 && (minnext == 0) && (deltanext == 0)
3123 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3124 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3125 && ckWARN(WARN_REGEXP))
3128 "Quantifier unexpected on zero-length expression");
3131 min += minnext * mincount;
3132 is_inf_internal |= ((maxcount == REG_INFTY
3133 && (minnext + deltanext) > 0)
3134 || deltanext == I32_MAX);
3135 is_inf |= is_inf_internal;
3136 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3138 /* Try powerful optimization CURLYX => CURLYN. */
3139 if ( OP(oscan) == CURLYX && data
3140 && data->flags & SF_IN_PAR
3141 && !(data->flags & SF_HAS_EVAL)
3142 && !deltanext && minnext == 1 ) {
3143 /* Try to optimize to CURLYN. */
3144 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3145 regnode * const nxt1 = nxt;
3152 if (!strchr((const char*)PL_simple,OP(nxt))
3153 && !(PL_regkind[OP(nxt)] == EXACT
3154 && STR_LEN(nxt) == 1))
3160 if (OP(nxt) != CLOSE)
3162 if (RExC_open_parens) {
3163 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3164 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3166 /* Now we know that nxt2 is the only contents: */
3167 oscan->flags = (U8)ARG(nxt);
3169 OP(nxt1) = NOTHING; /* was OPEN. */
3172 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3173 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3174 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3175 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3176 OP(nxt + 1) = OPTIMIZED; /* was count. */
3177 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3182 /* Try optimization CURLYX => CURLYM. */
3183 if ( OP(oscan) == CURLYX && data
3184 && !(data->flags & SF_HAS_PAR)
3185 && !(data->flags & SF_HAS_EVAL)
3186 && !deltanext /* atom is fixed width */
3187 && minnext != 0 /* CURLYM can't handle zero width */
3189 /* XXXX How to optimize if data == 0? */
3190 /* Optimize to a simpler form. */
3191 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3195 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3196 && (OP(nxt2) != WHILEM))
3198 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3199 /* Need to optimize away parenths. */
3200 if (data->flags & SF_IN_PAR) {
3201 /* Set the parenth number. */
3202 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3204 if (OP(nxt) != CLOSE)
3205 FAIL("Panic opt close");
3206 oscan->flags = (U8)ARG(nxt);
3207 if (RExC_open_parens) {
3208 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3209 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3211 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3212 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3215 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3216 OP(nxt + 1) = OPTIMIZED; /* was count. */
3217 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3218 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3221 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3222 regnode *nnxt = regnext(nxt1);
3225 if (reg_off_by_arg[OP(nxt1)])
3226 ARG_SET(nxt1, nxt2 - nxt1);
3227 else if (nxt2 - nxt1 < U16_MAX)
3228 NEXT_OFF(nxt1) = nxt2 - nxt1;
3230 OP(nxt) = NOTHING; /* Cannot beautify */
3235 /* Optimize again: */
3236 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3237 NULL, stopparen, recursed, NULL, 0,depth+1);
3242 else if ((OP(oscan) == CURLYX)
3243 && (flags & SCF_WHILEM_VISITED_POS)
3244 /* See the comment on a similar expression above.
3245 However, this time it not a subexpression
3246 we care about, but the expression itself. */
3247 && (maxcount == REG_INFTY)
3248 && data && ++data->whilem_c < 16) {
3249 /* This stays as CURLYX, we can put the count/of pair. */
3250 /* Find WHILEM (as in regexec.c) */
3251 regnode *nxt = oscan + NEXT_OFF(oscan);
3253 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3255 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3256 | (RExC_whilem_seen << 4)); /* On WHILEM */
3258 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3260 if (flags & SCF_DO_SUBSTR) {
3261 SV *last_str = NULL;
3262 int counted = mincount != 0;
3264 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3265 #if defined(SPARC64_GCC_WORKAROUND)
3268 const char *s = NULL;
3271 if (pos_before >= data->last_start_min)
3274 b = data->last_start_min;
3277 s = SvPV_const(data->last_found, l);
3278 old = b - data->last_start_min;
3281 I32 b = pos_before >= data->last_start_min
3282 ? pos_before : data->last_start_min;
3284 const char * const s = SvPV_const(data->last_found, l);
3285 I32 old = b - data->last_start_min;
3289 old = utf8_hop((U8*)s, old) - (U8*)s;
3292 /* Get the added string: */
3293 last_str = newSVpvn(s + old, l);
3295 SvUTF8_on(last_str);
3296 if (deltanext == 0 && pos_before == b) {
3297 /* What was added is a constant string */
3299 SvGROW(last_str, (mincount * l) + 1);
3300 repeatcpy(SvPVX(last_str) + l,
3301 SvPVX_const(last_str), l, mincount - 1);
3302 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3303 /* Add additional parts. */
3304 SvCUR_set(data->last_found,
3305 SvCUR(data->last_found) - l);
3306 sv_catsv(data->last_found, last_str);
3308 SV * sv = data->last_found;
3310 SvUTF8(sv) && SvMAGICAL(sv) ?
3311 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3312 if (mg && mg->mg_len >= 0)
3313 mg->mg_len += CHR_SVLEN(last_str) - l;
3315 data->last_end += l * (mincount - 1);
3318 /* start offset must point into the last copy */
3319 data->last_start_min += minnext * (mincount - 1);
3320 data->last_start_max += is_inf ? I32_MAX
3321 : (maxcount - 1) * (minnext + data->pos_delta);
3324 /* It is counted once already... */
3325 data->pos_min += minnext * (mincount - counted);
3326 data->pos_delta += - counted * deltanext +
3327 (minnext + deltanext) * maxcount - minnext * mincount;
3328 if (mincount != maxcount) {
3329 /* Cannot extend fixed substrings found inside
3331 SCAN_COMMIT(pRExC_state,data,minlenp);
3332 if (mincount && last_str) {
3333 SV * const sv = data->last_found;
3334 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3335 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3339 sv_setsv(sv, last_str);
3340 data->last_end = data->pos_min;
3341 data->last_start_min =
3342 data->pos_min - CHR_SVLEN(last_str);
3343 data->last_start_max = is_inf
3345 : data->pos_min + data->pos_delta
3346 - CHR_SVLEN(last_str);
3348 data->longest = &(data->longest_float);
3350 SvREFCNT_dec(last_str);
3352 if (data && (fl & SF_HAS_EVAL))
3353 data->flags |= SF_HAS_EVAL;
3354 optimize_curly_tail:
3355 if (OP(oscan) != CURLYX) {
3356 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3358 NEXT_OFF(oscan) += NEXT_OFF(next);
3361 default: /* REF and CLUMP only? */
3362 if (flags & SCF_DO_SUBSTR) {
3363 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3364 data->longest = &(data->longest_float);
3366 is_inf = is_inf_internal = 1;
3367 if (flags & SCF_DO_STCLASS_OR)
3368 cl_anything(pRExC_state, data->start_class);
3369 flags &= ~SCF_DO_STCLASS;
3373 else if (OP(scan) == LNBREAK) {
3374 if (flags & SCF_DO_STCLASS) {
3376 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3377 if (flags & SCF_DO_STCLASS_AND) {
3378 for (value = 0; value < 256; value++)
3379 if (!is_VERTWS_cp(value))
3380 ANYOF_BITMAP_CLEAR(data->start_class, value);
3383 for (value = 0; value < 256; value++)
3384 if (is_VERTWS_cp(value))
3385 ANYOF_BITMAP_SET(data->start_class, value);
3387 if (flags & SCF_DO_STCLASS_OR)
3388 cl_and(data->start_class, and_withp);
3389 flags &= ~SCF_DO_STCLASS;
3393 if (flags & SCF_DO_SUBSTR) {
3394 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3396 data->pos_delta += 1;
3397 data->longest = &(data->longest_float);
3401 else if (OP(scan) == FOLDCHAR) {
3402 int d = ARG(scan)==0xDF ? 1 : 2;
3403 flags &= ~SCF_DO_STCLASS;
3406 if (flags & SCF_DO_SUBSTR) {
3407 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3409 data->pos_delta += d;
3410 data->longest = &(data->longest_float);
3413 else if (strchr((const char*)PL_simple,OP(scan))) {
3416 if (flags & SCF_DO_SUBSTR) {
3417 SCAN_COMMIT(pRExC_state,data,minlenp);
3421 if (flags & SCF_DO_STCLASS) {
3422 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3424 /* Some of the logic below assumes that switching
3425 locale on will only add false positives. */
3426 switch (PL_regkind[OP(scan)]) {
3430 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3431 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3432 cl_anything(pRExC_state, data->start_class);
3435 if (OP(scan) == SANY)
3437 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3438 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3439 || (data->start_class->flags & ANYOF_CLASS));
3440 cl_anything(pRExC_state, data->start_class);
3442 if (flags & SCF_DO_STCLASS_AND || !value)
3443 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3446 if (flags & SCF_DO_STCLASS_AND)
3447 cl_and(data->start_class,
3448 (struct regnode_charclass_class*)scan);
3450 cl_or(pRExC_state, data->start_class,
3451 (struct regnode_charclass_class*)scan);
3454 if (flags & SCF_DO_STCLASS_AND) {
3455 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3456 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3457 for (value = 0; value < 256; value++)
3458 if (!isALNUM(value))
3459 ANYOF_BITMAP_CLEAR(data->start_class, value);
3463 if (data->start_class->flags & ANYOF_LOCALE)
3464 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3466 for (value = 0; value < 256; value++)
3468 ANYOF_BITMAP_SET(data->start_class, value);
3473 if (flags & SCF_DO_STCLASS_AND) {
3474 if (data->start_class->flags & ANYOF_LOCALE)
3475 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3478 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3479 data->start_class->flags |= ANYOF_LOCALE;
3483 if (flags & SCF_DO_STCLASS_AND) {
3484 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3485 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3486 for (value = 0; value < 256; value++)
3488 ANYOF_BITMAP_CLEAR(data->start_class, value);
3492 if (data->start_class->flags & ANYOF_LOCALE)
3493 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3495 for (value = 0; value < 256; value++)
3496 if (!isALNUM(value))
3497 ANYOF_BITMAP_SET(data->start_class, value);
3502 if (flags & SCF_DO_STCLASS_AND) {
3503 if (data->start_class->flags & ANYOF_LOCALE)
3504 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3507 data->start_class->flags |= ANYOF_LOCALE;
3508 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3512 if (flags & SCF_DO_STCLASS_AND) {
3513 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3514 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3515 for (value = 0; value < 256; value++)
3516 if (!isSPACE(value))
3517 ANYOF_BITMAP_CLEAR(data->start_class, value);
3521 if (data->start_class->flags & ANYOF_LOCALE)
3522 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3524 for (value = 0; value < 256; value++)
3526 ANYOF_BITMAP_SET(data->start_class, value);
3531 if (flags & SCF_DO_STCLASS_AND) {
3532 if (data->start_class->flags & ANYOF_LOCALE)
3533 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3536 data->start_class->flags |= ANYOF_LOCALE;
3537 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3541 if (flags & SCF_DO_STCLASS_AND) {
3542 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3543 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3544 for (value = 0; value < 256; value++)
3546 ANYOF_BITMAP_CLEAR(data->start_class, value);
3550 if (data->start_class->flags & ANYOF_LOCALE)
3551 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3553 for (value = 0; value < 256; value++)
3554 if (!isSPACE(value))
3555 ANYOF_BITMAP_SET(data->start_class, value);
3560 if (flags & SCF_DO_STCLASS_AND) {
3561 if (data->start_class->flags & ANYOF_LOCALE) {
3562 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3563 for (value = 0; value < 256; value++)
3564 if (!isSPACE(value))
3565 ANYOF_BITMAP_CLEAR(data->start_class, value);
3569 data->start_class->flags |= ANYOF_LOCALE;
3570 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3574 if (flags & SCF_DO_STCLASS_AND) {
3575 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3576 for (value = 0; value < 256; value++)
3577 if (!isDIGIT(value))
3578 ANYOF_BITMAP_CLEAR(data->start_class, value);
3581 if (data->start_class->flags & ANYOF_LOCALE)
3582 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3584 for (value = 0; value < 256; value++)
3586 ANYOF_BITMAP_SET(data->start_class, value);
3591 if (flags & SCF_DO_STCLASS_AND) {
3592 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3593 for (value = 0; value < 256; value++)
3595 ANYOF_BITMAP_CLEAR(data->start_class, value);
3598 if (data->start_class->flags & ANYOF_LOCALE)
3599 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3601 for (value = 0; value < 256; value++)
3602 if (!isDIGIT(value))
3603 ANYOF_BITMAP_SET(data->start_class, value);
3607 CASE_SYNST_FNC(VERTWS);
3608 CASE_SYNST_FNC(HORIZWS);
3611 if (flags & SCF_DO_STCLASS_OR)
3612 cl_and(data->start_class, and_withp);
3613 flags &= ~SCF_DO_STCLASS;
3616 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3617 data->flags |= (OP(scan) == MEOL
3621 else if ( PL_regkind[OP(scan)] == BRANCHJ
3622 /* Lookbehind, or need to calculate parens/evals/stclass: */
3623 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3624 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3625 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3626 || OP(scan) == UNLESSM )
3628 /* Negative Lookahead/lookbehind
3629 In this case we can't do fixed string optimisation.
3632 I32 deltanext, minnext, fake = 0;
3634 struct regnode_charclass_class intrnl;
3637 data_fake.flags = 0;
3639 data_fake.whilem_c = data->whilem_c;
3640 data_fake.last_closep = data->last_closep;
3643 data_fake.last_closep = &fake;
3644 data_fake.pos_delta = delta;
3645 if ( flags & SCF_DO_STCLASS && !scan->flags
3646 && OP(scan) == IFMATCH ) { /* Lookahead */
3647 cl_init(pRExC_state, &intrnl);
3648 data_fake.start_class = &intrnl;
3649 f |= SCF_DO_STCLASS_AND;
3651 if (flags & SCF_WHILEM_VISITED_POS)
3652 f |= SCF_WHILEM_VISITED_POS;
3653 next = regnext(scan);
3654 nscan = NEXTOPER(NEXTOPER(scan));
3655 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3656 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3659 FAIL("Variable length lookbehind not implemented");
3661 else if (minnext > (I32)U8_MAX) {
3662 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3664 scan->flags = (U8)minnext;
3667 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3669 if (data_fake.flags & SF_HAS_EVAL)
3670 data->flags |= SF_HAS_EVAL;
3671 data->whilem_c = data_fake.whilem_c;
3673 if (f & SCF_DO_STCLASS_AND) {
3674 const int was = (data->start_class->flags & ANYOF_EOS);
3676 cl_and(data->start_class, &intrnl);
3678 data->start_class->flags |= ANYOF_EOS;
3681 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3683 /* Positive Lookahead/lookbehind
3684 In this case we can do fixed string optimisation,
3685 but we must be careful about it. Note in the case of
3686 lookbehind the positions will be offset by the minimum
3687 length of the pattern, something we won't know about
3688 until after the recurse.
3690 I32 deltanext, fake = 0;
3692 struct regnode_charclass_class intrnl;
3694 /* We use SAVEFREEPV so that when the full compile
3695 is finished perl will clean up the allocated
3696 minlens when its all done. This was we don't
3697 have to worry about freeing them when we know
3698 they wont be used, which would be a pain.
3701 Newx( minnextp, 1, I32 );
3702 SAVEFREEPV(minnextp);
3705 StructCopy(data, &data_fake, scan_data_t);
3706 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3709 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3710 data_fake.last_found=newSVsv(data->last_found);
3714 data_fake.last_closep = &fake;
3715 data_fake.flags = 0;
3716 data_fake.pos_delta = delta;
3718 data_fake.flags |= SF_IS_INF;
3719 if ( flags & SCF_DO_STCLASS && !scan->flags
3720 && OP(scan) == IFMATCH ) { /* Lookahead */
3721 cl_init(pRExC_state, &intrnl);
3722 data_fake.start_class = &intrnl;
3723 f |= SCF_DO_STCLASS_AND;
3725 if (flags & SCF_WHILEM_VISITED_POS)
3726 f |= SCF_WHILEM_VISITED_POS;
3727 next = regnext(scan);
3728 nscan = NEXTOPER(NEXTOPER(scan));
3730 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3731 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3734 FAIL("Variable length lookbehind not implemented");
3736 else if (*minnextp > (I32)U8_MAX) {
3737 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3739 scan->flags = (U8)*minnextp;
3744 if (f & SCF_DO_STCLASS_AND) {
3745 const int was = (data->start_class->flags & ANYOF_EOS);
3747 cl_and(data->start_class, &intrnl);
3749 data->start_class->flags |= ANYOF_EOS;
3752 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3754 if (data_fake.flags & SF_HAS_EVAL)
3755 data->flags |= SF_HAS_EVAL;
3756 data->whilem_c = data_fake.whilem_c;
3757 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3758 if (RExC_rx->minlen<*minnextp)
3759 RExC_rx->minlen=*minnextp;
3760 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3761 SvREFCNT_dec(data_fake.last_found);
3763 if ( data_fake.minlen_fixed != minlenp )
3765 data->offset_fixed= data_fake.offset_fixed;
3766 data->minlen_fixed= data_fake.minlen_fixed;
3767 data->lookbehind_fixed+= scan->flags;
3769 if ( data_fake.minlen_float != minlenp )
3771 data->minlen_float= data_fake.minlen_float;
3772 data->offset_float_min=data_fake.offset_float_min;
3773 data->offset_float_max=data_fake.offset_float_max;
3774 data->lookbehind_float+= scan->flags;
3783 else if (OP(scan) == OPEN) {
3784 if (stopparen != (I32)ARG(scan))
3787 else if (OP(scan) == CLOSE) {
3788 if (stopparen == (I32)ARG(scan)) {
3791 if ((I32)ARG(scan) == is_par) {
3792 next = regnext(scan);
3794 if ( next && (OP(next) != WHILEM) && next < last)
3795 is_par = 0; /* Disable optimization */
3798 *(data->last_closep) = ARG(scan);
3800 else if (OP(scan) == EVAL) {
3802 data->flags |= SF_HAS_EVAL;
3804 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3805 if (flags & SCF_DO_SUBSTR) {
3806 SCAN_COMMIT(pRExC_state,data,minlenp);
3807 flags &= ~SCF_DO_SUBSTR;
3809 if (data && OP(scan)==ACCEPT) {
3810 data->flags |= SCF_SEEN_ACCEPT;
3815 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3817 if (flags & SCF_DO_SUBSTR) {
3818 SCAN_COMMIT(pRExC_state,data,minlenp);
3819 data->longest = &(data->longest_float);
3821 is_inf = is_inf_internal = 1;
3822 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3823 cl_anything(pRExC_state, data->start_class);
3824 flags &= ~SCF_DO_STCLASS;
3826 else if (OP(scan) == GPOS) {
3827 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3828 !(delta || is_inf || (data && data->pos_delta)))
3830 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3831 RExC_rx->extflags |= RXf_ANCH_GPOS;
3832 if (RExC_rx->gofs < (U32)min)
3833 RExC_rx->gofs = min;
3835 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3839 #ifdef TRIE_STUDY_OPT
3840 #ifdef FULL_TRIE_STUDY
3841 else if (PL_regkind[OP(scan)] == TRIE) {
3842 /* NOTE - There is similar code to this block above for handling
3843 BRANCH nodes on the initial study. If you change stuff here
3845 regnode *trie_node= scan;
3846 regnode *tail= regnext(scan);
3847 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3848 I32 max1 = 0, min1 = I32_MAX;
3849 struct regnode_charclass_class accum;
3851 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3852 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3853 if (flags & SCF_DO_STCLASS)
3854 cl_init_zero(pRExC_state, &accum);
3860 const regnode *nextbranch= NULL;
3863 for ( word=1 ; word <= trie->wordcount ; word++)
3865 I32 deltanext=0, minnext=0, f = 0, fake;
3866 struct regnode_charclass_class this_class;
3868 data_fake.flags = 0;
3870 data_fake.whilem_c = data->whilem_c;
3871 data_fake.last_closep = data->last_closep;
3874 data_fake.last_closep = &fake;
3875 data_fake.pos_delta = delta;
3876 if (flags & SCF_DO_STCLASS) {
3877 cl_init(pRExC_state, &this_class);
3878 data_fake.start_class = &this_class;
3879 f = SCF_DO_STCLASS_AND;
3881 if (flags & SCF_WHILEM_VISITED_POS)
3882 f |= SCF_WHILEM_VISITED_POS;
3884 if (trie->jump[word]) {
3886 nextbranch = trie_node + trie->jump[0];
3887 scan= trie_node + trie->jump[word];
3888 /* We go from the jump point to the branch that follows
3889 it. Note this means we need the vestigal unused branches
3890 even though they arent otherwise used.
3892 minnext = study_chunk(pRExC_state, &scan, minlenp,
3893 &deltanext, (regnode *)nextbranch, &data_fake,
3894 stopparen, recursed, NULL, f,depth+1);
3896 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3897 nextbranch= regnext((regnode*)nextbranch);
3899 if (min1 > (I32)(minnext + trie->minlen))
3900 min1 = minnext + trie->minlen;
3901 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3902 max1 = minnext + deltanext + trie->maxlen;
3903 if (deltanext == I32_MAX)
3904 is_inf = is_inf_internal = 1;
3906 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3908 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3909 if ( stopmin > min + min1)
3910 stopmin = min + min1;
3911 flags &= ~SCF_DO_SUBSTR;
3913 data->flags |= SCF_SEEN_ACCEPT;
3916 if (data_fake.flags & SF_HAS_EVAL)
3917 data->flags |= SF_HAS_EVAL;
3918 data->whilem_c = data_fake.whilem_c;
3920 if (flags & SCF_DO_STCLASS)
3921 cl_or(pRExC_state, &accum, &this_class);
3924 if (flags & SCF_DO_SUBSTR) {
3925 data->pos_min += min1;
3926 data->pos_delta += max1 - min1;
3927 if (max1 != min1 || is_inf)
3928 data->longest = &(data->longest_float);
3931 delta += max1 - min1;
3932 if (flags & SCF_DO_STCLASS_OR) {
3933 cl_or(pRExC_state, data->start_class, &accum);
3935 cl_and(data->start_class, and_withp);
3936 flags &= ~SCF_DO_STCLASS;
3939 else if (flags & SCF_DO_STCLASS_AND) {
3941 cl_and(data->start_class, &accum);
3942 flags &= ~SCF_DO_STCLASS;
3945 /* Switch to OR mode: cache the old value of
3946 * data->start_class */
3948 StructCopy(data->start_class, and_withp,
3949 struct regnode_charclass_class);
3950 flags &= ~SCF_DO_STCLASS_AND;
3951 StructCopy(&accum, data->start_class,
3952 struct regnode_charclass_class);
3953 flags |= SCF_DO_STCLASS_OR;
3954 data->start_class->flags |= ANYOF_EOS;
3961 else if (PL_regkind[OP(scan)] == TRIE) {
3962 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3965 min += trie->minlen;
3966 delta += (trie->maxlen - trie->minlen);
3967 flags &= ~SCF_DO_STCLASS; /* xxx */
3968 if (flags & SCF_DO_SUBSTR) {
3969 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3970 data->pos_min += trie->minlen;
3971 data->pos_delta += (trie->maxlen - trie->minlen);
3972 if (trie->maxlen != trie->minlen)
3973 data->longest = &(data->longest_float);
3975 if (trie->jump) /* no more substrings -- for now /grr*/
3976 flags &= ~SCF_DO_SUBSTR;
3978 #endif /* old or new */
3979 #endif /* TRIE_STUDY_OPT */
3981 /* Else: zero-length, ignore. */
3982 scan = regnext(scan);
3987 stopparen = frame->stop;
3988 frame = frame->prev;
3989 goto fake_study_recurse;
3994 DEBUG_STUDYDATA("pre-fin:",data,depth);
3997 *deltap = is_inf_internal ? I32_MAX : delta;
3998 if (flags & SCF_DO_SUBSTR && is_inf)
3999 data->pos_delta = I32_MAX - data->pos_min;
4000 if (is_par > (I32)U8_MAX)
4002 if (is_par && pars==1 && data) {
4003 data->flags |= SF_IN_PAR;
4004 data->flags &= ~SF_HAS_PAR;
4006 else if (pars && data) {
4007 data->flags |= SF_HAS_PAR;
4008 data->flags &= ~SF_IN_PAR;
4010 if (flags & SCF_DO_STCLASS_OR)
4011 cl_and(data->start_class, and_withp);
4012 if (flags & SCF_TRIE_RESTUDY)
4013 data->flags |= SCF_TRIE_RESTUDY;
4015 DEBUG_STUDYDATA("post-fin:",data,depth);
4017 return min < stopmin ? min : stopmin;
4021 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4023 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4025 Renewc(RExC_rxi->data,
4026 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4027 char, struct reg_data);
4029 Renew(RExC_rxi->data->what, count + n, U8);
4031 Newx(RExC_rxi->data->what, n, U8);
4032 RExC_rxi->data->count = count + n;
4033 Copy(s, RExC_rxi->data->what + count, n, U8);
4037 /*XXX: todo make this not included in a non debugging perl */
4038 #ifndef PERL_IN_XSUB_RE
4040 Perl_reginitcolors(pTHX)
4043 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4045 char *t = savepv(s);
4049 t = strchr(t, '\t');
4055 PL_colors[i] = t = (char *)"";
4060 PL_colors[i++] = (char *)"";
4067 #ifdef TRIE_STUDY_OPT
4068 #define CHECK_RESTUDY_GOTO \
4070 (data.flags & SCF_TRIE_RESTUDY) \
4074 #define CHECK_RESTUDY_GOTO
4078 - pregcomp - compile a regular expression into internal code
4080 * We can't allocate space until we know how big the compiled form will be,
4081 * but we can't compile it (and thus know how big it is) until we've got a
4082 * place to put the code. So we cheat: we compile it twice, once with code
4083 * generation turned off and size counting turned on, and once "for real".
4084 * This also means that we don't allocate space until we are sure that the
4085 * thing really will compile successfully, and we never have to move the
4086 * code and thus invalidate pointers into it. (Note that it has to be in
4087 * one piece because free() must be able to free it all.) [NB: not true in perl]
4089 * Beware that the optimization-preparation code in here knows about some
4090 * of the structure of the compiled regexp. [I'll say.]
4095 #ifndef PERL_IN_XSUB_RE
4096 #define RE_ENGINE_PTR &PL_core_reg_engine
4098 extern const struct regexp_engine my_reg_engine;
4099 #define RE_ENGINE_PTR &my_reg_engine
4102 #ifndef PERL_IN_XSUB_RE
4104 Perl_pregcomp(pTHX_ const SV * const pattern, const U32 flags)
4107 HV * const table = GvHV(PL_hintgv);
4108 /* Dispatch a request to compile a regexp to correct
4111 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4112 GET_RE_DEBUG_FLAGS_DECL;
4113 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4114 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4116 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4119 return CALLREGCOMP_ENG(eng, pattern, flags);
4122 return Perl_re_compile(aTHX_ pattern, flags);
4127 Perl_re_compile(pTHX_ const SV * const pattern, const U32 pm_flags)
4131 register regexp_internal *ri;
4133 char* exp = SvPV((SV*)pattern, plen);
4134 char* xend = exp + plen;
4141 RExC_state_t RExC_state;
4142 RExC_state_t * const pRExC_state = &RExC_state;
4143 #ifdef TRIE_STUDY_OPT
4145 RExC_state_t copyRExC_state;
4147 GET_RE_DEBUG_FLAGS_DECL;
4148 DEBUG_r(if (!PL_colorset) reginitcolors());
4150 RExC_utf8 = RExC_orig_utf8 = pm_flags & RXf_UTF8;
4153 SV *dsv= sv_newmortal();
4154 RE_PV_QUOTED_DECL(s, RExC_utf8,
4155 dsv, exp, plen, 60);
4156 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4157 PL_colors[4],PL_colors[5],s);
4162 RExC_flags = pm_flags;
4166 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4167 RExC_seen_evals = 0;
4170 /* First pass: determine size, legality. */
4178 RExC_emit = &PL_regdummy;
4179 RExC_whilem_seen = 0;
4180 RExC_charnames = NULL;
4181 RExC_open_parens = NULL;
4182 RExC_close_parens = NULL;
4184 RExC_paren_names = NULL;
4186 RExC_paren_name_list = NULL;
4188 RExC_recurse = NULL;
4189 RExC_recurse_count = 0;
4191 #if 0 /* REGC() is (currently) a NOP at the first pass.
4192 * Clever compilers notice this and complain. --jhi */
4193 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4195 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4196 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4197 RExC_precomp = NULL;
4200 if (RExC_utf8 && !RExC_orig_utf8) {
4201 /* It's possible to write a regexp in ascii that represents Unicode
4202 codepoints outside of the byte range, such as via \x{100}. If we
4203 detect such a sequence we have to convert the entire pattern to utf8
4204 and then recompile, as our sizing calculation will have been based
4205 on 1 byte == 1 character, but we will need to use utf8 to encode
4206 at least some part of the pattern, and therefore must convert the whole
4208 XXX: somehow figure out how to make this less expensive...
4211 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4212 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4213 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4215 RExC_orig_utf8 = RExC_utf8;
4217 goto redo_first_pass;
4220 PerlIO_printf(Perl_debug_log,
4221 "Required size %"IVdf" nodes\n"
4222 "Starting second pass (creation)\n",
4225 RExC_lastparse=NULL;
4227 /* Small enough for pointer-storage convention?
4228 If extralen==0, this means that we will not need long jumps. */
4229 if (RExC_size >= 0x10000L && RExC_extralen)
4230 RExC_size += RExC_extralen;
4233 if (RExC_whilem_seen > 15)
4234 RExC_whilem_seen = 15;
4236 /* Allocate space and zero-initialize. Note, the two step process
4237 of zeroing when in debug mode, thus anything assigned has to
4238 happen after that */
4239 Newxz(r, 1, regexp);
4240 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4241 char, regexp_internal);
4242 if ( r == NULL || ri == NULL )
4243 FAIL("Regexp out of space");
4245 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4246 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4248 /* bulk initialize base fields with 0. */
4249 Zero(ri, sizeof(regexp_internal), char);
4252 /* non-zero initialization begins here */
4254 r->engine= RE_ENGINE_PTR;
4257 r->extflags = pm_flags;
4259 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4260 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4261 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4262 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) >> 12);
4263 const char *fptr = STD_PAT_MODS; /*"msix"*/
4265 r->wraplen = r->prelen + has_minus + has_p + has_runon
4266 + (sizeof(STD_PAT_MODS) - 1)
4267 + (sizeof("(?:)") - 1);
4269 Newx(r->wrapped, r->wraplen + 1, char );
4273 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4275 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4276 char *colon = r + 1;
4279 while((ch = *fptr++)) {
4293 Copy(RExC_precomp, p, r->prelen, char);
4303 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4305 if (RExC_seen & REG_SEEN_RECURSE) {
4306 Newxz(RExC_open_parens, RExC_npar,regnode *);
4307 SAVEFREEPV(RExC_open_parens);
4308 Newxz(RExC_close_parens,RExC_npar,regnode *);
4309 SAVEFREEPV(RExC_close_parens);
4312 /* Useful during FAIL. */
4313 #ifdef RE_TRACK_PATTERN_OFFSETS
4314 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4315 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4316 "%s %"UVuf" bytes for offset annotations.\n",
4317 ri->u.offsets ? "Got" : "Couldn't get",
4318 (UV)((2*RExC_size+1) * sizeof(U32))));
4320 SetProgLen(ri,RExC_size);
4324 /* Second pass: emit code. */
4325 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4330 RExC_emit_start = ri->program;
4331 RExC_emit = ri->program;
4332 RExC_emit_bound = ri->program + RExC_size + 1;
4334 /* Store the count of eval-groups for security checks: */
4335 RExC_rx->seen_evals = RExC_seen_evals;
4336 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4337 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4341 /* XXXX To minimize changes to RE engine we always allocate
4342 3-units-long substrs field. */
4343 Newx(r->substrs, 1, struct reg_substr_data);
4344 if (RExC_recurse_count) {
4345 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4346 SAVEFREEPV(RExC_recurse);
4350 r->minlen = minlen = sawplus = sawopen = 0;
4351 Zero(r->substrs, 1, struct reg_substr_data);
4353 #ifdef TRIE_STUDY_OPT
4356 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4358 RExC_state = copyRExC_state;
4359 if (seen & REG_TOP_LEVEL_BRANCHES)
4360 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4362 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4363 if (data.last_found) {
4364 SvREFCNT_dec(data.longest_fixed);
4365 SvREFCNT_dec(data.longest_float);
4366 SvREFCNT_dec(data.last_found);
4368 StructCopy(&zero_scan_data, &data, scan_data_t);
4370 StructCopy(&zero_scan_data, &data, scan_data_t);
4371 copyRExC_state = RExC_state;
4374 StructCopy(&zero_scan_data, &data, scan_data_t);
4377 /* Dig out information for optimizations. */
4378 r->extflags = RExC_flags; /* was pm_op */
4379 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4382 r->extflags |= RXf_UTF8; /* Unicode in it? */
4383 ri->regstclass = NULL;
4384 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4385 r->intflags |= PREGf_NAUGHTY;
4386 scan = ri->program + 1; /* First BRANCH. */
4388 /* testing for BRANCH here tells us whether there is "must appear"
4389 data in the pattern. If there is then we can use it for optimisations */
4390 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4392 STRLEN longest_float_length, longest_fixed_length;
4393 struct regnode_charclass_class ch_class; /* pointed to by data */
4395 I32 last_close = 0; /* pointed to by data */
4396 regnode *first= scan;
4397 regnode *first_next= regnext(first);
4399 /* Skip introductions and multiplicators >= 1. */
4400 while ((OP(first) == OPEN && (sawopen = 1)) ||
4401 /* An OR of *one* alternative - should not happen now. */
4402 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4403 /* for now we can't handle lookbehind IFMATCH*/
4404 (OP(first) == IFMATCH && !first->flags) ||
4405 (OP(first) == PLUS) ||
4406 (OP(first) == MINMOD) ||
4407 /* An {n,m} with n>0 */
4408 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4409 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4412 if (OP(first) == PLUS)
4415 first += regarglen[OP(first)];
4416 if (OP(first) == IFMATCH) {
4417 first = NEXTOPER(first);
4418 first += EXTRA_STEP_2ARGS;
4419 } else /* XXX possible optimisation for /(?=)/ */
4420 first = NEXTOPER(first);
4421 first_next= regnext(first);
4424 /* Starting-point info. */
4426 DEBUG_PEEP("first:",first,0);
4427 /* Ignore EXACT as we deal with it later. */
4428 if (PL_regkind[OP(first)] == EXACT) {
4429 if (OP(first) == EXACT)
4430 NOOP; /* Empty, get anchored substr later. */
4431 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4432 ri->regstclass = first;
4435 else if (PL_regkind[OP(first)] == TRIE &&
4436 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4439 /* this can happen only on restudy */
4440 if ( OP(first) == TRIE ) {
4441 struct regnode_1 *trieop = (struct regnode_1 *)
4442 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4443 StructCopy(first,trieop,struct regnode_1);
4444 trie_op=(regnode *)trieop;
4446 struct regnode_charclass *trieop = (struct regnode_charclass *)
4447 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4448 StructCopy(first,trieop,struct regnode_charclass);
4449 trie_op=(regnode *)trieop;
4452 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4453 ri->regstclass = trie_op;
4456 else if (strchr((const char*)PL_simple,OP(first)))
4457 ri->regstclass = first;
4458 else if (PL_regkind[OP(first)] == BOUND ||
4459 PL_regkind[OP(first)] == NBOUND)
4460 ri->regstclass = first;
4461 else if (PL_regkind[OP(first)] == BOL) {
4462 r->extflags |= (OP(first) == MBOL
4464 : (OP(first) == SBOL
4467 first = NEXTOPER(first);
4470 else if (OP(first) == GPOS) {
4471 r->extflags |= RXf_ANCH_GPOS;
4472 first = NEXTOPER(first);
4475 else if ((!sawopen || !RExC_sawback) &&
4476 (OP(first) == STAR &&
4477 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4478 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4480 /* turn .* into ^.* with an implied $*=1 */
4482 (OP(NEXTOPER(first)) == REG_ANY)
4485 r->extflags |= type;
4486 r->intflags |= PREGf_IMPLICIT;
4487 first = NEXTOPER(first);
4490 if (sawplus && (!sawopen || !RExC_sawback)
4491 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4492 /* x+ must match at the 1st pos of run of x's */
4493 r->intflags |= PREGf_SKIP;
4495 /* Scan is after the zeroth branch, first is atomic matcher. */
4496 #ifdef TRIE_STUDY_OPT
4499 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4500 (IV)(first - scan + 1))
4504 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4505 (IV)(first - scan + 1))
4511 * If there's something expensive in the r.e., find the
4512 * longest literal string that must appear and make it the
4513 * regmust. Resolve ties in favor of later strings, since
4514 * the regstart check works with the beginning of the r.e.
4515 * and avoiding duplication strengthens checking. Not a
4516 * strong reason, but sufficient in the absence of others.
4517 * [Now we resolve ties in favor of the earlier string if
4518 * it happens that c_offset_min has been invalidated, since the
4519 * earlier string may buy us something the later one won't.]
4522 data.longest_fixed = newSVpvs("");
4523 data.longest_float = newSVpvs("");
4524 data.last_found = newSVpvs("");
4525 data.longest = &(data.longest_fixed);
4527 if (!ri->regstclass) {
4528 cl_init(pRExC_state, &ch_class);
4529 data.start_class = &ch_class;
4530 stclass_flag = SCF_DO_STCLASS_AND;
4531 } else /* XXXX Check for BOUND? */
4533 data.last_closep = &last_close;
4535 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4536 &data, -1, NULL, NULL,
4537 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4543 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4544 && data.last_start_min == 0 && data.last_end > 0
4545 && !RExC_seen_zerolen
4546 && !(RExC_seen & REG_SEEN_VERBARG)
4547 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4548 r->extflags |= RXf_CHECK_ALL;
4549 scan_commit(pRExC_state, &data,&minlen,0);
4550 SvREFCNT_dec(data.last_found);
4552 /* Note that code very similar to this but for anchored string
4553 follows immediately below, changes may need to be made to both.
4556 longest_float_length = CHR_SVLEN(data.longest_float);
4557 if (longest_float_length
4558 || (data.flags & SF_FL_BEFORE_EOL
4559 && (!(data.flags & SF_FL_BEFORE_MEOL)
4560 || (RExC_flags & RXf_PMf_MULTILINE))))
4564 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4565 && data.offset_fixed == data.offset_float_min
4566 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4567 goto remove_float; /* As in (a)+. */
4569 /* copy the information about the longest float from the reg_scan_data
4570 over to the program. */
4571 if (SvUTF8(data.longest_float)) {
4572 r->float_utf8 = data.longest_float;
4573 r->float_substr = NULL;
4575 r->float_substr = data.longest_float;
4576 r->float_utf8 = NULL;
4578 /* float_end_shift is how many chars that must be matched that
4579 follow this item. We calculate it ahead of time as once the
4580 lookbehind offset is added in we lose the ability to correctly
4582 ml = data.minlen_float ? *(data.minlen_float)
4583 : (I32)longest_float_length;
4584 r->float_end_shift = ml - data.offset_float_min
4585 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4586 + data.lookbehind_float;
4587 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4588 r->float_max_offset = data.offset_float_max;
4589 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4590 r->float_max_offset -= data.lookbehind_float;
4592 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4593 && (!(data.flags & SF_FL_BEFORE_MEOL)
4594 || (RExC_flags & RXf_PMf_MULTILINE)));
4595 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4599 r->float_substr = r->float_utf8 = NULL;
4600 SvREFCNT_dec(data.longest_float);
4601 longest_float_length = 0;
4604 /* Note that code very similar to this but for floating string
4605 is immediately above, changes may need to be made to both.
4608 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4609 if (longest_fixed_length
4610 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4611 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4612 || (RExC_flags & RXf_PMf_MULTILINE))))
4616 /* copy the information about the longest fixed
4617 from the reg_scan_data over to the program. */
4618 if (SvUTF8(data.longest_fixed)) {
4619 r->anchored_utf8 = data.longest_fixed;
4620 r->anchored_substr = NULL;
4622 r->anchored_substr = data.longest_fixed;
4623 r->anchored_utf8 = NULL;
4625 /* fixed_end_shift is how many chars that must be matched that
4626 follow this item. We calculate it ahead of time as once the
4627 lookbehind offset is added in we lose the ability to correctly
4629 ml = data.minlen_fixed ? *(data.minlen_fixed)
4630 : (I32)longest_fixed_length;
4631 r->anchored_end_shift = ml - data.offset_fixed
4632 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4633 + data.lookbehind_fixed;
4634 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4636 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4637 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4638 || (RExC_flags & RXf_PMf_MULTILINE)));
4639 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4642 r->anchored_substr = r->anchored_utf8 = NULL;
4643 SvREFCNT_dec(data.longest_fixed);
4644 longest_fixed_length = 0;
4647 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4648 ri->regstclass = NULL;
4649 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4651 && !(data.start_class->flags & ANYOF_EOS)
4652 && !cl_is_anything(data.start_class))
4654 const U32 n = add_data(pRExC_state, 1, "f");
4656 Newx(RExC_rxi->data->data[n], 1,
4657 struct regnode_charclass_class);
4658 StructCopy(data.start_class,
4659 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4660 struct regnode_charclass_class);
4661 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4662 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4663 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4664 regprop(r, sv, (regnode*)data.start_class);
4665 PerlIO_printf(Perl_debug_log,
4666 "synthetic stclass \"%s\".\n",
4667 SvPVX_const(sv));});
4670 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4671 if (longest_fixed_length > longest_float_length) {
4672 r->check_end_shift = r->anchored_end_shift;
4673 r->check_substr = r->anchored_substr;
4674 r->check_utf8 = r->anchored_utf8;
4675 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4676 if (r->extflags & RXf_ANCH_SINGLE)
4677 r->extflags |= RXf_NOSCAN;
4680 r->check_end_shift = r->float_end_shift;
4681 r->check_substr = r->float_substr;
4682 r->check_utf8 = r->float_utf8;
4683 r->check_offset_min = r->float_min_offset;
4684 r->check_offset_max = r->float_max_offset;
4686 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4687 This should be changed ASAP! */
4688 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4689 r->extflags |= RXf_USE_INTUIT;
4690 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4691 r->extflags |= RXf_INTUIT_TAIL;
4693 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4694 if ( (STRLEN)minlen < longest_float_length )
4695 minlen= longest_float_length;
4696 if ( (STRLEN)minlen < longest_fixed_length )
4697 minlen= longest_fixed_length;
4701 /* Several toplevels. Best we can is to set minlen. */
4703 struct regnode_charclass_class ch_class;
4706 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4708 scan = ri->program + 1;
4709 cl_init(pRExC_state, &ch_class);
4710 data.start_class = &ch_class;
4711 data.last_closep = &last_close;
4714 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4715 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4719 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4720 = r->float_substr = r->float_utf8 = NULL;
4721 if (!(data.start_class->flags & ANYOF_EOS)
4722 && !cl_is_anything(data.start_class))
4724 const U32 n = add_data(pRExC_state, 1, "f");
4726 Newx(RExC_rxi->data->data[n], 1,
4727 struct regnode_charclass_class);
4728 StructCopy(data.start_class,
4729 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4730 struct regnode_charclass_class);
4731 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4732 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4733 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4734 regprop(r, sv, (regnode*)data.start_class);
4735 PerlIO_printf(Perl_debug_log,
4736 "synthetic stclass \"%s\".\n",
4737 SvPVX_const(sv));});
4741 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4742 the "real" pattern. */
4744 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4745 (IV)minlen, (IV)r->minlen);
4747 r->minlenret = minlen;
4748 if (r->minlen < minlen)
4751 if (RExC_seen & REG_SEEN_GPOS)
4752 r->extflags |= RXf_GPOS_SEEN;
4753 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4754 r->extflags |= RXf_LOOKBEHIND_SEEN;
4755 if (RExC_seen & REG_SEEN_EVAL)
4756 r->extflags |= RXf_EVAL_SEEN;
4757 if (RExC_seen & REG_SEEN_CANY)
4758 r->extflags |= RXf_CANY_SEEN;
4759 if (RExC_seen & REG_SEEN_VERBARG)
4760 r->intflags |= PREGf_VERBARG_SEEN;
4761 if (RExC_seen & REG_SEEN_CUTGROUP)
4762 r->intflags |= PREGf_CUTGROUP_SEEN;
4763 if (RExC_paren_names)
4764 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4766 r->paren_names = NULL;
4768 #ifdef STUPID_PATTERN_CHECKS
4770 r->extflags |= RXf_NULL;
4771 if (r->extflags & RXf_SPLIT && r->prelen == 1 && r->precomp[0] == ' ')
4772 /* XXX: this should happen BEFORE we compile */
4773 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4774 else if (r->prelen == 3 && memEQ("\\s+", r->precomp, 3))
4775 r->extflags |= RXf_WHITE;
4776 else if (r->prelen == 1 && r->precomp[0] == '^')
4777 r->extflags |= RXf_START_ONLY;
4779 if (r->extflags & RXf_SPLIT && r->prelen == 1 && r->precomp[0] == ' ')
4780 /* XXX: this should happen BEFORE we compile */
4781 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4783 regnode *first = ri->program + 1;
4785 U8 nop = OP(NEXTOPER(first));
4787 if (PL_regkind[fop] == NOTHING && nop == END)
4788 r->extflags |= RXf_NULL;
4789 else if (PL_regkind[fop] == BOL && nop == END)
4790 r->extflags |= RXf_START_ONLY;
4791 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4792 r->extflags |= RXf_WHITE;
4796 if (RExC_paren_names) {
4797 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4798 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4801 ri->name_list_idx = 0;
4803 if (RExC_recurse_count) {
4804 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4805 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4806 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4809 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4810 /* assume we don't need to swap parens around before we match */
4813 PerlIO_printf(Perl_debug_log,"Final program:\n");
4816 #ifdef RE_TRACK_PATTERN_OFFSETS
4817 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4818 const U32 len = ri->u.offsets[0];
4820 GET_RE_DEBUG_FLAGS_DECL;
4821 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4822 for (i = 1; i <= len; i++) {
4823 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4824 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4825 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4827 PerlIO_printf(Perl_debug_log, "\n");
4833 #undef RE_ENGINE_PTR
4837 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4840 PERL_UNUSED_ARG(value);
4842 if (flags & RXapif_FETCH) {
4843 return reg_named_buff_fetch(rx, key, flags);
4844 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
4845 Perl_croak(aTHX_ PL_no_modify);
4847 } else if (flags & RXapif_EXISTS) {
4848 return reg_named_buff_exists(rx, key, flags)
4851 } else if (flags & RXapif_REGNAMES) {
4852 return reg_named_buff_all(rx, flags);
4853 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
4854 return reg_named_buff_scalar(rx, flags);
4856 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
4862 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
4865 PERL_UNUSED_ARG(lastkey);
4867 if (flags & RXapif_FIRSTKEY)
4868 return reg_named_buff_firstkey(rx, flags);
4869 else if (flags & RXapif_NEXTKEY)
4870 return reg_named_buff_nextkey(rx, flags);
4872 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
4878 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const rx, SV * const namesv, const U32 flags)
4880 AV *retarray = NULL;
4882 if (flags & RXapif_ALL)
4885 if (rx && rx->paren_names) {
4886 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4889 SV* sv_dat=HeVAL(he_str);
4890 I32 *nums=(I32*)SvPVX(sv_dat);
4891 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4892 if ((I32)(rx->nparens) >= nums[i]
4893 && rx->offs[nums[i]].start != -1
4894 && rx->offs[nums[i]].end != -1)
4897 CALLREG_NUMBUF_FETCH(rx,nums[i],ret);
4901 ret = newSVsv(&PL_sv_undef);
4904 SvREFCNT_inc_simple_void(ret);
4905 av_push(retarray, ret);
4909 return newRV((SV*)retarray);
4916 Perl_reg_named_buff_exists(pTHX_ REGEXP * const rx, SV * const key,
4919 if (rx && rx->paren_names) {
4920 if (flags & RXapif_ALL) {
4921 return hv_exists_ent(rx->paren_names, key, 0);
4923 SV *sv = CALLREG_NAMED_BUFF_FETCH(rx, key, flags);
4937 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const rx, const U32 flags)
4939 (void)hv_iterinit(rx->paren_names);
4941 return CALLREG_NAMED_BUFF_NEXTKEY(rx, NULL, flags & ~RXapif_FIRSTKEY);
4945 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const rx, const U32 flags)
4947 if (rx && rx->paren_names) {
4948 HV *hv = rx->paren_names;
4950 while ( (temphe = hv_iternext_flags(hv,0)) ) {
4953 SV* sv_dat = HeVAL(temphe);
4954 I32 *nums = (I32*)SvPVX(sv_dat);
4955 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
4956 if ((I32)(rx->lastcloseparen) >= nums[i] &&
4957 rx->offs[nums[i]].start != -1 &&
4958 rx->offs[nums[i]].end != -1)
4964 if (parno || flags & RXapif_ALL) {
4966 char *pv = HePV(temphe, len);
4967 return newSVpvn(pv,len);
4975 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const rx, const U32 flags)
4981 if (rx && rx->paren_names) {
4982 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
4983 return newSViv(HvTOTALKEYS(rx->paren_names));
4984 } else if (flags & RXapif_ONE) {
4985 ret = CALLREG_NAMED_BUFF_ALL(rx, (flags | RXapif_REGNAMES));
4986 av = (AV*)SvRV(ret);
4987 length = av_len(av);
4988 return newSViv(length + 1);
4990 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
4994 return &PL_sv_undef;
4998 Perl_reg_named_buff_all(pTHX_ REGEXP * const rx, const U32 flags)
5002 if (rx && rx->paren_names) {
5003 HV *hv= rx->paren_names;
5005 (void)hv_iterinit(hv);
5006 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5009 SV* sv_dat = HeVAL(temphe);
5010 I32 *nums = (I32*)SvPVX(sv_dat);
5011 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5012 if ((I32)(rx->lastcloseparen) >= nums[i] &&
5013 rx->offs[nums[i]].start != -1 &&
5014 rx->offs[nums[i]].end != -1)
5020 if (parno || flags & RXapif_ALL) {
5022 char *pv = HePV(temphe, len);
5023 av_push(av, newSVpvn(pv,len));
5028 return newRV((SV*)av);
5032 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const rx, const I32 paren, SV * const sv)
5039 sv_setsv(sv,&PL_sv_undef);
5043 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5045 i = rx->offs[0].start;
5049 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5051 s = rx->subbeg + rx->offs[0].end;
5052 i = rx->sublen - rx->offs[0].end;
5055 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5056 (s1 = rx->offs[paren].start) != -1 &&
5057 (t1 = rx->offs[paren].end) != -1)
5061 s = rx->subbeg + s1;
5063 sv_setsv(sv,&PL_sv_undef);
5066 assert(rx->sublen >= (s - rx->subbeg) + i );
5068 const int oldtainted = PL_tainted;
5070 sv_setpvn(sv, s, i);
5071 PL_tainted = oldtainted;
5072 if ( (rx->extflags & RXf_CANY_SEEN)
5073 ? (RX_MATCH_UTF8(rx)
5074 && (!i || is_utf8_string((U8*)s, i)))
5075 : (RX_MATCH_UTF8(rx)) )
5082 if (RX_MATCH_TAINTED(rx)) {
5083 if (SvTYPE(sv) >= SVt_PVMG) {
5084 MAGIC* const mg = SvMAGIC(sv);
5087 SvMAGIC_set(sv, mg->mg_moremagic);
5089 if ((mgt = SvMAGIC(sv))) {
5090 mg->mg_moremagic = mgt;
5091 SvMAGIC_set(sv, mg);
5101 sv_setsv(sv,&PL_sv_undef);
5107 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5108 SV const * const value)
5110 PERL_UNUSED_ARG(rx);
5111 PERL_UNUSED_ARG(paren);
5112 PERL_UNUSED_ARG(value);
5115 Perl_croak(aTHX_ PL_no_modify);
5119 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const rx, const SV * const sv,
5125 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5127 /* $` / ${^PREMATCH} */
5128 case RX_BUFF_IDX_PREMATCH:
5129 if (rx->offs[0].start != -1) {
5130 i = rx->offs[0].start;
5138 /* $' / ${^POSTMATCH} */
5139 case RX_BUFF_IDX_POSTMATCH:
5140 if (rx->offs[0].end != -1) {
5141 i = rx->sublen - rx->offs[0].end;
5143 s1 = rx->offs[0].end;
5149 /* $& / ${^MATCH}, $1, $2, ... */
5151 if (paren <= (I32)rx->nparens &&
5152 (s1 = rx->offs[paren].start) != -1 &&
5153 (t1 = rx->offs[paren].end) != -1)
5158 if (ckWARN(WARN_UNINITIALIZED))
5159 report_uninit((SV*)sv);
5164 if (i > 0 && RX_MATCH_UTF8(rx)) {
5165 const char * const s = rx->subbeg + s1;
5170 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5177 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5179 PERL_UNUSED_ARG(rx);
5180 return newSVpvs("Regexp");
5183 /* Scans the name of a named buffer from the pattern.
5184 * If flags is REG_RSN_RETURN_NULL returns null.
5185 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5186 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5187 * to the parsed name as looked up in the RExC_paren_names hash.
5188 * If there is an error throws a vFAIL().. type exception.
5191 #define REG_RSN_RETURN_NULL 0
5192 #define REG_RSN_RETURN_NAME 1
5193 #define REG_RSN_RETURN_DATA 2
5196 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
5197 char *name_start = RExC_parse;
5199 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5200 /* skip IDFIRST by using do...while */
5203 RExC_parse += UTF8SKIP(RExC_parse);
5204 } while (isALNUM_utf8((U8*)RExC_parse));
5208 } while (isALNUM(*RExC_parse));
5212 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
5213 (int)(RExC_parse - name_start)));
5216 if ( flags == REG_RSN_RETURN_NAME)
5218 else if (flags==REG_RSN_RETURN_DATA) {
5221 if ( ! sv_name ) /* should not happen*/
5222 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5223 if (RExC_paren_names)
5224 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5226 sv_dat = HeVAL(he_str);
5228 vFAIL("Reference to nonexistent named group");
5232 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5239 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5240 int rem=(int)(RExC_end - RExC_parse); \
5249 if (RExC_lastparse!=RExC_parse) \
5250 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5253 iscut ? "..." : "<" \
5256 PerlIO_printf(Perl_debug_log,"%16s",""); \
5259 num = RExC_size + 1; \
5261 num=REG_NODE_NUM(RExC_emit); \
5262 if (RExC_lastnum!=num) \
5263 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5265 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5266 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5267 (int)((depth*2)), "", \
5271 RExC_lastparse=RExC_parse; \
5276 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5277 DEBUG_PARSE_MSG((funcname)); \
5278 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5280 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5281 DEBUG_PARSE_MSG((funcname)); \
5282 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5285 - reg - regular expression, i.e. main body or parenthesized thing
5287 * Caller must absorb opening parenthesis.
5289 * Combining parenthesis handling with the base level of regular expression
5290 * is a trifle forced, but the need to tie the tails of the branches to what
5291 * follows makes it hard to avoid.
5293 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5295 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5297 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5301 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5302 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5305 register regnode *ret; /* Will be the head of the group. */
5306 register regnode *br;
5307 register regnode *lastbr;
5308 register regnode *ender = NULL;
5309 register I32 parno = 0;
5311 U32 oregflags = RExC_flags;
5312 bool have_branch = 0;
5314 I32 freeze_paren = 0;
5315 I32 after_freeze = 0;
5317 /* for (?g), (?gc), and (?o) warnings; warning
5318 about (?c) will warn about (?g) -- japhy */
5320 #define WASTED_O 0x01
5321 #define WASTED_G 0x02
5322 #define WASTED_C 0x04
5323 #define WASTED_GC (0x02|0x04)
5324 I32 wastedflags = 0x00;
5326 char * parse_start = RExC_parse; /* MJD */
5327 char * const oregcomp_parse = RExC_parse;
5329 GET_RE_DEBUG_FLAGS_DECL;
5330 DEBUG_PARSE("reg ");
5332 *flagp = 0; /* Tentatively. */
5335 /* Make an OPEN node, if parenthesized. */
5337 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5338 char *start_verb = RExC_parse;
5339 STRLEN verb_len = 0;
5340 char *start_arg = NULL;
5341 unsigned char op = 0;
5343 int internal_argval = 0; /* internal_argval is only useful if !argok */
5344 while ( *RExC_parse && *RExC_parse != ')' ) {
5345 if ( *RExC_parse == ':' ) {
5346 start_arg = RExC_parse + 1;
5352 verb_len = RExC_parse - start_verb;
5355 while ( *RExC_parse && *RExC_parse != ')' )
5357 if ( *RExC_parse != ')' )
5358 vFAIL("Unterminated verb pattern argument");
5359 if ( RExC_parse == start_arg )
5362 if ( *RExC_parse != ')' )
5363 vFAIL("Unterminated verb pattern");
5366 switch ( *start_verb ) {
5367 case 'A': /* (*ACCEPT) */
5368 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5370 internal_argval = RExC_nestroot;
5373 case 'C': /* (*COMMIT) */
5374 if ( memEQs(start_verb,verb_len,"COMMIT") )
5377 case 'F': /* (*FAIL) */
5378 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5383 case ':': /* (*:NAME) */
5384 case 'M': /* (*MARK:NAME) */
5385 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5390 case 'P': /* (*PRUNE) */
5391 if ( memEQs(start_verb,verb_len,"PRUNE") )
5394 case 'S': /* (*SKIP) */
5395 if ( memEQs(start_verb,verb_len,"SKIP") )
5398 case 'T': /* (*THEN) */
5399 /* [19:06] <TimToady> :: is then */
5400 if ( memEQs(start_verb,verb_len,"THEN") ) {
5402 RExC_seen |= REG_SEEN_CUTGROUP;
5408 vFAIL3("Unknown verb pattern '%.*s'",
5409 verb_len, start_verb);
5412 if ( start_arg && internal_argval ) {
5413 vFAIL3("Verb pattern '%.*s' may not have an argument",
5414 verb_len, start_verb);
5415 } else if ( argok < 0 && !start_arg ) {
5416 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5417 verb_len, start_verb);
5419 ret = reganode(pRExC_state, op, internal_argval);
5420 if ( ! internal_argval && ! SIZE_ONLY ) {
5422 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5423 ARG(ret) = add_data( pRExC_state, 1, "S" );
5424 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5431 if (!internal_argval)
5432 RExC_seen |= REG_SEEN_VERBARG;
5433 } else if ( start_arg ) {
5434 vFAIL3("Verb pattern '%.*s' may not have an argument",
5435 verb_len, start_verb);
5437 ret = reg_node(pRExC_state, op);
5439 nextchar(pRExC_state);
5442 if (*RExC_parse == '?') { /* (?...) */
5443 bool is_logical = 0;
5444 const char * const seqstart = RExC_parse;
5447 paren = *RExC_parse++;
5448 ret = NULL; /* For look-ahead/behind. */
5451 case 'P': /* (?P...) variants for those used to PCRE/Python */
5452 paren = *RExC_parse++;
5453 if ( paren == '<') /* (?P<...>) named capture */
5455 else if (paren == '>') { /* (?P>name) named recursion */
5456 goto named_recursion;
5458 else if (paren == '=') { /* (?P=...) named backref */
5459 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5460 you change this make sure you change that */
5461 char* name_start = RExC_parse;
5463 SV *sv_dat = reg_scan_name(pRExC_state,
5464 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5465 if (RExC_parse == name_start || *RExC_parse != ')')
5466 vFAIL2("Sequence %.3s... not terminated",parse_start);
5469 num = add_data( pRExC_state, 1, "S" );
5470 RExC_rxi->data->data[num]=(void*)sv_dat;
5471 SvREFCNT_inc_simple_void(sv_dat);
5474 ret = reganode(pRExC_state,
5475 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5479 Set_Node_Offset(ret, parse_start+1);
5480 Set_Node_Cur_Length(ret); /* MJD */
5482 nextchar(pRExC_state);
5486 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5488 case '<': /* (?<...) */
5489 if (*RExC_parse == '!')
5491 else if (*RExC_parse != '=')
5497 case '\'': /* (?'...') */
5498 name_start= RExC_parse;
5499 svname = reg_scan_name(pRExC_state,
5500 SIZE_ONLY ? /* reverse test from the others */
5501 REG_RSN_RETURN_NAME :
5502 REG_RSN_RETURN_NULL);
5503 if (RExC_parse == name_start) {
5505 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5508 if (*RExC_parse != paren)
5509 vFAIL2("Sequence (?%c... not terminated",
5510 paren=='>' ? '<' : paren);
5514 if (!svname) /* shouldnt happen */
5516 "panic: reg_scan_name returned NULL");
5517 if (!RExC_paren_names) {
5518 RExC_paren_names= newHV();
5519 sv_2mortal((SV*)RExC_paren_names);
5521 RExC_paren_name_list= newAV();
5522 sv_2mortal((SV*)RExC_paren_name_list);
5525 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5527 sv_dat = HeVAL(he_str);
5529 /* croak baby croak */
5531 "panic: paren_name hash element allocation failed");
5532 } else if ( SvPOK(sv_dat) ) {
5533 /* (?|...) can mean we have dupes so scan to check
5534 its already been stored. Maybe a flag indicating
5535 we are inside such a construct would be useful,
5536 but the arrays are likely to be quite small, so
5537 for now we punt -- dmq */
5538 IV count = SvIV(sv_dat);
5539 I32 *pv = (I32*)SvPVX(sv_dat);
5541 for ( i = 0 ; i < count ; i++ ) {
5542 if ( pv[i] == RExC_npar ) {
5548 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5549 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5550 pv[count] = RExC_npar;
5554 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5555 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5560 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5561 SvREFCNT_dec(svname);
5564 /*sv_dump(sv_dat);*/
5566 nextchar(pRExC_state);
5568 goto capturing_parens;
5570 RExC_seen |= REG_SEEN_LOOKBEHIND;
5572 case '=': /* (?=...) */
5573 case '!': /* (?!...) */
5574 RExC_seen_zerolen++;
5575 if (*RExC_parse == ')') {
5576 ret=reg_node(pRExC_state, OPFAIL);
5577 nextchar(pRExC_state);
5581 case '|': /* (?|...) */
5582 /* branch reset, behave like a (?:...) except that
5583 buffers in alternations share the same numbers */
5585 after_freeze = freeze_paren = RExC_npar;
5587 case ':': /* (?:...) */
5588 case '>': /* (?>...) */
5590 case '$': /* (?$...) */
5591 case '@': /* (?@...) */
5592 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5594 case '#': /* (?#...) */
5595 while (*RExC_parse && *RExC_parse != ')')
5597 if (*RExC_parse != ')')
5598 FAIL("Sequence (?#... not terminated");
5599 nextchar(pRExC_state);
5602 case '0' : /* (?0) */
5603 case 'R' : /* (?R) */
5604 if (*RExC_parse != ')')
5605 FAIL("Sequence (?R) not terminated");
5606 ret = reg_node(pRExC_state, GOSTART);
5607 *flagp |= POSTPONED;
5608 nextchar(pRExC_state);
5611 { /* named and numeric backreferences */
5613 case '&': /* (?&NAME) */
5614 parse_start = RExC_parse - 1;
5617 SV *sv_dat = reg_scan_name(pRExC_state,
5618 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5619 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5621 goto gen_recurse_regop;
5624 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5626 vFAIL("Illegal pattern");
5628 goto parse_recursion;
5630 case '-': /* (?-1) */
5631 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5632 RExC_parse--; /* rewind to let it be handled later */
5636 case '1': case '2': case '3': case '4': /* (?1) */
5637 case '5': case '6': case '7': case '8': case '9':
5640 num = atoi(RExC_parse);
5641 parse_start = RExC_parse - 1; /* MJD */
5642 if (*RExC_parse == '-')
5644 while (isDIGIT(*RExC_parse))
5646 if (*RExC_parse!=')')
5647 vFAIL("Expecting close bracket");
5650 if ( paren == '-' ) {
5652 Diagram of capture buffer numbering.
5653 Top line is the normal capture buffer numbers
5654 Botton line is the negative indexing as from
5658 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5662 num = RExC_npar + num;
5665 vFAIL("Reference to nonexistent group");
5667 } else if ( paren == '+' ) {
5668 num = RExC_npar + num - 1;
5671 ret = reganode(pRExC_state, GOSUB, num);
5673 if (num > (I32)RExC_rx->nparens) {
5675 vFAIL("Reference to nonexistent group");
5677 ARG2L_SET( ret, RExC_recurse_count++);
5679 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5680 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5684 RExC_seen |= REG_SEEN_RECURSE;
5685 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5686 Set_Node_Offset(ret, parse_start); /* MJD */
5688 *flagp |= POSTPONED;
5689 nextchar(pRExC_state);
5691 } /* named and numeric backreferences */
5694 case '?': /* (??...) */
5696 if (*RExC_parse != '{') {
5698 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5701 *flagp |= POSTPONED;
5702 paren = *RExC_parse++;
5704 case '{': /* (?{...}) */
5709 char *s = RExC_parse;
5711 RExC_seen_zerolen++;
5712 RExC_seen |= REG_SEEN_EVAL;
5713 while (count && (c = *RExC_parse)) {
5724 if (*RExC_parse != ')') {
5726 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5730 OP_4tree *sop, *rop;
5731 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5734 Perl_save_re_context(aTHX);
5735 rop = sv_compile_2op(sv, &sop, "re", &pad);
5736 sop->op_private |= OPpREFCOUNTED;
5737 /* re_dup will OpREFCNT_inc */
5738 OpREFCNT_set(sop, 1);
5741 n = add_data(pRExC_state, 3, "nop");
5742 RExC_rxi->data->data[n] = (void*)rop;
5743 RExC_rxi->data->data[n+1] = (void*)sop;
5744 RExC_rxi->data->data[n+2] = (void*)pad;
5747 else { /* First pass */
5748 if (PL_reginterp_cnt < ++RExC_seen_evals
5750 /* No compiled RE interpolated, has runtime
5751 components ===> unsafe. */
5752 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5753 if (PL_tainting && PL_tainted)
5754 FAIL("Eval-group in insecure regular expression");
5755 #if PERL_VERSION > 8
5756 if (IN_PERL_COMPILETIME)
5761 nextchar(pRExC_state);
5763 ret = reg_node(pRExC_state, LOGICAL);
5766 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5767 /* deal with the length of this later - MJD */
5770 ret = reganode(pRExC_state, EVAL, n);
5771 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5772 Set_Node_Offset(ret, parse_start);
5775 case '(': /* (?(?{...})...) and (?(?=...)...) */
5778 if (RExC_parse[0] == '?') { /* (?(?...)) */
5779 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5780 || RExC_parse[1] == '<'
5781 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5784 ret = reg_node(pRExC_state, LOGICAL);
5787 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5791 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5792 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5794 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5795 char *name_start= RExC_parse++;
5797 SV *sv_dat=reg_scan_name(pRExC_state,
5798 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5799 if (RExC_parse == name_start || *RExC_parse != ch)
5800 vFAIL2("Sequence (?(%c... not terminated",
5801 (ch == '>' ? '<' : ch));
5804 num = add_data( pRExC_state, 1, "S" );
5805 RExC_rxi->data->data[num]=(void*)sv_dat;
5806 SvREFCNT_inc_simple_void(sv_dat);
5808 ret = reganode(pRExC_state,NGROUPP,num);
5809 goto insert_if_check_paren;
5811 else if (RExC_parse[0] == 'D' &&
5812 RExC_parse[1] == 'E' &&
5813 RExC_parse[2] == 'F' &&
5814 RExC_parse[3] == 'I' &&
5815 RExC_parse[4] == 'N' &&
5816 RExC_parse[5] == 'E')
5818 ret = reganode(pRExC_state,DEFINEP,0);
5821 goto insert_if_check_paren;
5823 else if (RExC_parse[0] == 'R') {
5826 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5827 parno = atoi(RExC_parse++);
5828 while (isDIGIT(*RExC_parse))
5830 } else if (RExC_parse[0] == '&') {
5833 sv_dat = reg_scan_name(pRExC_state,
5834 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5835 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5837 ret = reganode(pRExC_state,INSUBP,parno);
5838 goto insert_if_check_paren;
5840 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5843 parno = atoi(RExC_parse++);
5845 while (isDIGIT(*RExC_parse))
5847 ret = reganode(pRExC_state, GROUPP, parno);
5849 insert_if_check_paren:
5850 if ((c = *nextchar(pRExC_state)) != ')')
5851 vFAIL("Switch condition not recognized");
5853 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5854 br = regbranch(pRExC_state, &flags, 1,depth+1);
5856 br = reganode(pRExC_state, LONGJMP, 0);
5858 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5859 c = *nextchar(pRExC_state);
5864 vFAIL("(?(DEFINE)....) does not allow branches");
5865 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5866 regbranch(pRExC_state, &flags, 1,depth+1);
5867 REGTAIL(pRExC_state, ret, lastbr);
5870 c = *nextchar(pRExC_state);
5875 vFAIL("Switch (?(condition)... contains too many branches");
5876 ender = reg_node(pRExC_state, TAIL);
5877 REGTAIL(pRExC_state, br, ender);
5879 REGTAIL(pRExC_state, lastbr, ender);
5880 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5883 REGTAIL(pRExC_state, ret, ender);
5884 RExC_size++; /* XXX WHY do we need this?!!
5885 For large programs it seems to be required
5886 but I can't figure out why. -- dmq*/
5890 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5894 RExC_parse--; /* for vFAIL to print correctly */
5895 vFAIL("Sequence (? incomplete");
5899 parse_flags: /* (?i) */
5901 U32 posflags = 0, negflags = 0;
5902 U32 *flagsp = &posflags;
5904 while (*RExC_parse) {
5905 /* && strchr("iogcmsx", *RExC_parse) */
5906 /* (?g), (?gc) and (?o) are useless here
5907 and must be globally applied -- japhy */
5908 switch (*RExC_parse) {
5909 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5910 case ONCE_PAT_MOD: /* 'o' */
5911 case GLOBAL_PAT_MOD: /* 'g' */
5912 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5913 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5914 if (! (wastedflags & wflagbit) ) {
5915 wastedflags |= wflagbit;
5918 "Useless (%s%c) - %suse /%c modifier",
5919 flagsp == &negflags ? "?-" : "?",
5921 flagsp == &negflags ? "don't " : "",
5928 case CONTINUE_PAT_MOD: /* 'c' */
5929 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5930 if (! (wastedflags & WASTED_C) ) {
5931 wastedflags |= WASTED_GC;
5934 "Useless (%sc) - %suse /gc modifier",
5935 flagsp == &negflags ? "?-" : "?",
5936 flagsp == &negflags ? "don't " : ""
5941 case KEEPCOPY_PAT_MOD: /* 'p' */
5942 if (flagsp == &negflags) {
5943 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5944 vWARN(RExC_parse + 1,"Useless use of (?-p)");
5946 *flagsp |= RXf_PMf_KEEPCOPY;
5950 if (flagsp == &negflags) {
5952 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5956 wastedflags = 0; /* reset so (?g-c) warns twice */
5962 RExC_flags |= posflags;
5963 RExC_flags &= ~negflags;
5965 oregflags |= posflags;
5966 oregflags &= ~negflags;
5968 nextchar(pRExC_state);
5979 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5984 }} /* one for the default block, one for the switch */
5991 ret = reganode(pRExC_state, OPEN, parno);
5994 RExC_nestroot = parno;
5995 if (RExC_seen & REG_SEEN_RECURSE
5996 && !RExC_open_parens[parno-1])
5998 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5999 "Setting open paren #%"IVdf" to %d\n",
6000 (IV)parno, REG_NODE_NUM(ret)));
6001 RExC_open_parens[parno-1]= ret;
6004 Set_Node_Length(ret, 1); /* MJD */
6005 Set_Node_Offset(ret, RExC_parse); /* MJD */
6013 /* Pick up the branches, linking them together. */
6014 parse_start = RExC_parse; /* MJD */
6015 br = regbranch(pRExC_state, &flags, 1,depth+1);
6016 /* branch_len = (paren != 0); */
6020 if (*RExC_parse == '|') {
6021 if (!SIZE_ONLY && RExC_extralen) {
6022 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6025 reginsert(pRExC_state, BRANCH, br, depth+1);
6026 Set_Node_Length(br, paren != 0);
6027 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6031 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6033 else if (paren == ':') {
6034 *flagp |= flags&SIMPLE;
6036 if (is_open) { /* Starts with OPEN. */
6037 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6039 else if (paren != '?') /* Not Conditional */
6041 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6043 while (*RExC_parse == '|') {
6044 if (!SIZE_ONLY && RExC_extralen) {
6045 ender = reganode(pRExC_state, LONGJMP,0);
6046 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6049 RExC_extralen += 2; /* Account for LONGJMP. */
6050 nextchar(pRExC_state);
6052 if (RExC_npar > after_freeze)
6053 after_freeze = RExC_npar;
6054 RExC_npar = freeze_paren;
6056 br = regbranch(pRExC_state, &flags, 0, depth+1);
6060 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6062 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6065 if (have_branch || paren != ':') {
6066 /* Make a closing node, and hook it on the end. */
6069 ender = reg_node(pRExC_state, TAIL);
6072 ender = reganode(pRExC_state, CLOSE, parno);
6073 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6074 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6075 "Setting close paren #%"IVdf" to %d\n",
6076 (IV)parno, REG_NODE_NUM(ender)));
6077 RExC_close_parens[parno-1]= ender;
6078 if (RExC_nestroot == parno)
6081 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6082 Set_Node_Length(ender,1); /* MJD */
6088 *flagp &= ~HASWIDTH;
6091 ender = reg_node(pRExC_state, SUCCEED);
6094 ender = reg_node(pRExC_state, END);
6096 assert(!RExC_opend); /* there can only be one! */
6101 REGTAIL(pRExC_state, lastbr, ender);
6103 if (have_branch && !SIZE_ONLY) {
6105 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6107 /* Hook the tails of the branches to the closing node. */
6108 for (br = ret; br; br = regnext(br)) {
6109 const U8 op = PL_regkind[OP(br)];
6111 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6113 else if (op == BRANCHJ) {
6114 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6122 static const char parens[] = "=!<,>";
6124 if (paren && (p = strchr(parens, paren))) {
6125 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6126 int flag = (p - parens) > 1;
6129 node = SUSPEND, flag = 0;
6130 reginsert(pRExC_state, node,ret, depth+1);
6131 Set_Node_Cur_Length(ret);
6132 Set_Node_Offset(ret, parse_start + 1);
6134 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6138 /* Check for proper termination. */
6140 RExC_flags = oregflags;
6141 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6142 RExC_parse = oregcomp_parse;
6143 vFAIL("Unmatched (");
6146 else if (!paren && RExC_parse < RExC_end) {
6147 if (*RExC_parse == ')') {
6149 vFAIL("Unmatched )");
6152 FAIL("Junk on end of regexp"); /* "Can't happen". */
6156 RExC_npar = after_freeze;
6161 - regbranch - one alternative of an | operator
6163 * Implements the concatenation operator.
6166 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6169 register regnode *ret;
6170 register regnode *chain = NULL;
6171 register regnode *latest;
6172 I32 flags = 0, c = 0;
6173 GET_RE_DEBUG_FLAGS_DECL;
6174 DEBUG_PARSE("brnc");
6179 if (!SIZE_ONLY && RExC_extralen)
6180 ret = reganode(pRExC_state, BRANCHJ,0);
6182 ret = reg_node(pRExC_state, BRANCH);
6183 Set_Node_Length(ret, 1);
6187 if (!first && SIZE_ONLY)
6188 RExC_extralen += 1; /* BRANCHJ */
6190 *flagp = WORST; /* Tentatively. */
6193 nextchar(pRExC_state);
6194 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6196 latest = regpiece(pRExC_state, &flags,depth+1);
6197 if (latest == NULL) {
6198 if (flags & TRYAGAIN)
6202 else if (ret == NULL)
6204 *flagp |= flags&(HASWIDTH|POSTPONED);
6205 if (chain == NULL) /* First piece. */
6206 *flagp |= flags&SPSTART;
6209 REGTAIL(pRExC_state, chain, latest);
6214 if (chain == NULL) { /* Loop ran zero times. */
6215 chain = reg_node(pRExC_state, NOTHING);
6220 *flagp |= flags&SIMPLE;
6227 - regpiece - something followed by possible [*+?]
6229 * Note that the branching code sequences used for ? and the general cases
6230 * of * and + are somewhat optimized: they use the same NOTHING node as
6231 * both the endmarker for their branch list and the body of the last branch.
6232 * It might seem that this node could be dispensed with entirely, but the
6233 * endmarker role is not redundant.
6236 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6239 register regnode *ret;
6241 register char *next;
6243 const char * const origparse = RExC_parse;
6245 I32 max = REG_INFTY;
6247 const char *maxpos = NULL;
6248 GET_RE_DEBUG_FLAGS_DECL;
6249 DEBUG_PARSE("piec");
6251 ret = regatom(pRExC_state, &flags,depth+1);
6253 if (flags & TRYAGAIN)
6260 if (op == '{' && regcurly(RExC_parse)) {
6262 parse_start = RExC_parse; /* MJD */
6263 next = RExC_parse + 1;
6264 while (isDIGIT(*next) || *next == ',') {
6273 if (*next == '}') { /* got one */
6277 min = atoi(RExC_parse);
6281 maxpos = RExC_parse;
6283 if (!max && *maxpos != '0')
6284 max = REG_INFTY; /* meaning "infinity" */
6285 else if (max >= REG_INFTY)
6286 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6288 nextchar(pRExC_state);
6291 if ((flags&SIMPLE)) {
6292 RExC_naughty += 2 + RExC_naughty / 2;
6293 reginsert(pRExC_state, CURLY, ret, depth+1);
6294 Set_Node_Offset(ret, parse_start+1); /* MJD */
6295 Set_Node_Cur_Length(ret);
6298 regnode * const w = reg_node(pRExC_state, WHILEM);
6301 REGTAIL(pRExC_state, ret, w);
6302 if (!SIZE_ONLY && RExC_extralen) {
6303 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6304 reginsert(pRExC_state, NOTHING,ret, depth+1);
6305 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6307 reginsert(pRExC_state, CURLYX,ret, depth+1);
6309 Set_Node_Offset(ret, parse_start+1);
6310 Set_Node_Length(ret,
6311 op == '{' ? (RExC_parse - parse_start) : 1);
6313 if (!SIZE_ONLY && RExC_extralen)
6314 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6315 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6317 RExC_whilem_seen++, RExC_extralen += 3;
6318 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6326 if (max && max < min)
6327 vFAIL("Can't do {n,m} with n > m");
6329 ARG1_SET(ret, (U16)min);
6330 ARG2_SET(ret, (U16)max);
6342 #if 0 /* Now runtime fix should be reliable. */
6344 /* if this is reinstated, don't forget to put this back into perldiag:
6346 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6348 (F) The part of the regexp subject to either the * or + quantifier
6349 could match an empty string. The {#} shows in the regular
6350 expression about where the problem was discovered.
6354 if (!(flags&HASWIDTH) && op != '?')
6355 vFAIL("Regexp *+ operand could be empty");
6358 parse_start = RExC_parse;
6359 nextchar(pRExC_state);
6361 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6363 if (op == '*' && (flags&SIMPLE)) {
6364 reginsert(pRExC_state, STAR, ret, depth+1);
6368 else if (op == '*') {
6372 else if (op == '+' && (flags&SIMPLE)) {
6373 reginsert(pRExC_state, PLUS, ret, depth+1);
6377 else if (op == '+') {
6381 else if (op == '?') {
6386 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6388 "%.*s matches null string many times",
6389 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6393 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6394 nextchar(pRExC_state);
6395 reginsert(pRExC_state, MINMOD, ret, depth+1);
6396 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6398 #ifndef REG_ALLOW_MINMOD_SUSPEND
6401 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6403 nextchar(pRExC_state);
6404 ender = reg_node(pRExC_state, SUCCEED);
6405 REGTAIL(pRExC_state, ret, ender);
6406 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6408 ender = reg_node(pRExC_state, TAIL);
6409 REGTAIL(pRExC_state, ret, ender);
6413 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6415 vFAIL("Nested quantifiers");
6422 /* reg_namedseq(pRExC_state,UVp)
6424 This is expected to be called by a parser routine that has
6425 recognized'\N' and needs to handle the rest. RExC_parse is
6426 expected to point at the first char following the N at the time
6429 If valuep is non-null then it is assumed that we are parsing inside
6430 of a charclass definition and the first codepoint in the resolved
6431 string is returned via *valuep and the routine will return NULL.
6432 In this mode if a multichar string is returned from the charnames
6433 handler a warning will be issued, and only the first char in the
6434 sequence will be examined. If the string returned is zero length
6435 then the value of *valuep is undefined and NON-NULL will
6436 be returned to indicate failure. (This will NOT be a valid pointer
6439 If value is null then it is assumed that we are parsing normal text
6440 and inserts a new EXACT node into the program containing the resolved
6441 string and returns a pointer to the new node. If the string is
6442 zerolength a NOTHING node is emitted.
6444 On success RExC_parse is set to the char following the endbrace.
6445 Parsing failures will generate a fatal errorvia vFAIL(...)
6447 NOTE: We cache all results from the charnames handler locally in
6448 the RExC_charnames hash (created on first use) to prevent a charnames
6449 handler from playing silly-buggers and returning a short string and
6450 then a long string for a given pattern. Since the regexp program
6451 size is calculated during an initial parse this would result
6452 in a buffer overrun so we cache to prevent the charname result from
6453 changing during the course of the parse.
6457 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6459 char * name; /* start of the content of the name */
6460 char * endbrace; /* endbrace following the name */
6463 STRLEN len; /* this has various purposes throughout the code */
6464 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6465 regnode *ret = NULL;
6467 if (*RExC_parse != '{') {
6468 vFAIL("Missing braces on \\N{}");
6470 name = RExC_parse+1;
6471 endbrace = strchr(RExC_parse, '}');
6474 vFAIL("Missing right brace on \\N{}");
6476 RExC_parse = endbrace + 1;
6479 /* RExC_parse points at the beginning brace,
6480 endbrace points at the last */
6481 if ( name[0]=='U' && name[1]=='+' ) {
6482 /* its a "Unicode hex" notation {U+89AB} */
6483 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6484 | PERL_SCAN_DISALLOW_PREFIX
6485 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6487 len = (STRLEN)(endbrace - name - 2);
6488 cp = grok_hex(name + 2, &len, &fl, NULL);
6489 if ( len != (STRLEN)(endbrace - name - 2) ) {
6498 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6500 /* fetch the charnames handler for this scope */
6501 HV * const table = GvHV(PL_hintgv);
6503 hv_fetchs(table, "charnames", FALSE) :
6505 SV *cv= cvp ? *cvp : NULL;
6508 /* create an SV with the name as argument */
6509 sv_name = newSVpvn(name, endbrace - name);
6511 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6512 vFAIL2("Constant(\\N{%s}) unknown: "
6513 "(possibly a missing \"use charnames ...\")",
6516 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6517 vFAIL2("Constant(\\N{%s}): "
6518 "$^H{charnames} is not defined",SvPVX(sv_name));
6523 if (!RExC_charnames) {
6524 /* make sure our cache is allocated */
6525 RExC_charnames = newHV();
6526 sv_2mortal((SV*)RExC_charnames);
6528 /* see if we have looked this one up before */
6529 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6531 sv_str = HeVAL(he_str);
6544 count= call_sv(cv, G_SCALAR);
6546 if (count == 1) { /* XXXX is this right? dmq */
6548 SvREFCNT_inc_simple_void(sv_str);
6556 if ( !sv_str || !SvOK(sv_str) ) {
6557 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6558 "did not return a defined value",SvPVX(sv_name));
6560 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6565 char *p = SvPV(sv_str, len);
6568 if ( SvUTF8(sv_str) ) {
6569 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6573 We have to turn on utf8 for high bit chars otherwise
6574 we get failures with
6576 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6577 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6579 This is different from what \x{} would do with the same
6580 codepoint, where the condition is > 0xFF.
6587 /* warn if we havent used the whole string? */
6589 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6591 "Ignoring excess chars from \\N{%s} in character class",
6595 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6597 "Ignoring zero length \\N{%s} in character class",
6602 SvREFCNT_dec(sv_name);
6604 SvREFCNT_dec(sv_str);
6605 return len ? NULL : (regnode *)&len;
6606 } else if(SvCUR(sv_str)) {
6612 char * parse_start = name-3; /* needed for the offsets */
6614 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6616 ret = reg_node(pRExC_state,
6617 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6620 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6621 sv_utf8_upgrade(sv_str);
6622 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6626 p = SvPV(sv_str, len);
6628 /* len is the length written, charlen is the size the char read */
6629 for ( len = 0; p < pend; p += charlen ) {
6631 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6633 STRLEN foldlen,numlen;
6634 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6635 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6636 /* Emit all the Unicode characters. */
6638 for (foldbuf = tmpbuf;
6642 uvc = utf8_to_uvchr(foldbuf, &numlen);
6644 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6647 /* In EBCDIC the numlen
6648 * and unilen can differ. */
6650 if (numlen >= foldlen)
6654 break; /* "Can't happen." */
6657 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6669 RExC_size += STR_SZ(len);
6672 RExC_emit += STR_SZ(len);
6674 Set_Node_Cur_Length(ret); /* MJD */
6676 nextchar(pRExC_state);
6678 ret = reg_node(pRExC_state,NOTHING);
6681 SvREFCNT_dec(sv_str);
6684 SvREFCNT_dec(sv_name);
6694 * It returns the code point in utf8 for the value in *encp.
6695 * value: a code value in the source encoding
6696 * encp: a pointer to an Encode object
6698 * If the result from Encode is not a single character,
6699 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6702 S_reg_recode(pTHX_ const char value, SV **encp)
6705 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6706 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6707 const STRLEN newlen = SvCUR(sv);
6708 UV uv = UNICODE_REPLACEMENT;
6712 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6715 if (!newlen || numlen != newlen) {
6716 uv = UNICODE_REPLACEMENT;
6724 - regatom - the lowest level
6726 Try to identify anything special at the start of the pattern. If there
6727 is, then handle it as required. This may involve generating a single regop,
6728 such as for an assertion; or it may involve recursing, such as to
6729 handle a () structure.
6731 If the string doesn't start with something special then we gobble up
6732 as much literal text as we can.
6734 Once we have been able to handle whatever type of thing started the
6735 sequence, we return.
6737 Note: we have to be careful with escapes, as they can be both literal
6738 and special, and in the case of \10 and friends can either, depending
6739 on context. Specifically there are two seperate switches for handling
6740 escape sequences, with the one for handling literal escapes requiring
6741 a dummy entry for all of the special escapes that are actually handled
6746 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6749 register regnode *ret = NULL;
6751 char *parse_start = RExC_parse;
6752 GET_RE_DEBUG_FLAGS_DECL;
6753 DEBUG_PARSE("atom");
6754 *flagp = WORST; /* Tentatively. */
6758 switch ((U8)*RExC_parse) {
6760 RExC_seen_zerolen++;
6761 nextchar(pRExC_state);
6762 if (RExC_flags & RXf_PMf_MULTILINE)
6763 ret = reg_node(pRExC_state, MBOL);
6764 else if (RExC_flags & RXf_PMf_SINGLELINE)
6765 ret = reg_node(pRExC_state, SBOL);
6767 ret = reg_node(pRExC_state, BOL);
6768 Set_Node_Length(ret, 1); /* MJD */
6771 nextchar(pRExC_state);
6773 RExC_seen_zerolen++;
6774 if (RExC_flags & RXf_PMf_MULTILINE)
6775 ret = reg_node(pRExC_state, MEOL);
6776 else if (RExC_flags & RXf_PMf_SINGLELINE)
6777 ret = reg_node(pRExC_state, SEOL);
6779 ret = reg_node(pRExC_state, EOL);
6780 Set_Node_Length(ret, 1); /* MJD */
6783 nextchar(pRExC_state);
6784 if (RExC_flags & RXf_PMf_SINGLELINE)
6785 ret = reg_node(pRExC_state, SANY);
6787 ret = reg_node(pRExC_state, REG_ANY);
6788 *flagp |= HASWIDTH|SIMPLE;
6790 Set_Node_Length(ret, 1); /* MJD */
6794 char * const oregcomp_parse = ++RExC_parse;
6795 ret = regclass(pRExC_state,depth+1);
6796 if (*RExC_parse != ']') {
6797 RExC_parse = oregcomp_parse;
6798 vFAIL("Unmatched [");
6800 nextchar(pRExC_state);
6801 *flagp |= HASWIDTH|SIMPLE;
6802 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6806 nextchar(pRExC_state);
6807 ret = reg(pRExC_state, 1, &flags,depth+1);
6809 if (flags & TRYAGAIN) {
6810 if (RExC_parse == RExC_end) {
6811 /* Make parent create an empty node if needed. */
6819 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6823 if (flags & TRYAGAIN) {
6827 vFAIL("Internal urp");
6828 /* Supposed to be caught earlier. */
6831 if (!regcurly(RExC_parse)) {
6840 vFAIL("Quantifier follows nothing");
6847 len=0; /* silence a spurious compiler warning */
6848 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
6849 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
6850 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
6851 ret = reganode(pRExC_state, FOLDCHAR, cp);
6852 Set_Node_Length(ret, 1); /* MJD */
6853 nextchar(pRExC_state); /* kill whitespace under /x */
6861 This switch handles escape sequences that resolve to some kind
6862 of special regop and not to literal text. Escape sequnces that
6863 resolve to literal text are handled below in the switch marked
6866 Every entry in this switch *must* have a corresponding entry
6867 in the literal escape switch. However, the opposite is not
6868 required, as the default for this switch is to jump to the
6869 literal text handling code.
6871 switch (*++RExC_parse) {
6872 /* Special Escapes */
6874 RExC_seen_zerolen++;
6875 ret = reg_node(pRExC_state, SBOL);
6877 goto finish_meta_pat;
6879 ret = reg_node(pRExC_state, GPOS);
6880 RExC_seen |= REG_SEEN_GPOS;
6882 goto finish_meta_pat;
6884 RExC_seen_zerolen++;
6885 ret = reg_node(pRExC_state, KEEPS);
6887 goto finish_meta_pat;
6889 ret = reg_node(pRExC_state, SEOL);
6891 RExC_seen_zerolen++; /* Do not optimize RE away */
6892 goto finish_meta_pat;
6894 ret = reg_node(pRExC_state, EOS);
6896 RExC_seen_zerolen++; /* Do not optimize RE away */
6897 goto finish_meta_pat;
6899 ret = reg_node(pRExC_state, CANY);
6900 RExC_seen |= REG_SEEN_CANY;
6901 *flagp |= HASWIDTH|SIMPLE;
6902 goto finish_meta_pat;
6904 ret = reg_node(pRExC_state, CLUMP);
6906 goto finish_meta_pat;
6908 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6909 *flagp |= HASWIDTH|SIMPLE;
6910 goto finish_meta_pat;
6912 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6913 *flagp |= HASWIDTH|SIMPLE;
6914 goto finish_meta_pat;
6916 RExC_seen_zerolen++;
6917 RExC_seen |= REG_SEEN_LOOKBEHIND;
6918 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6920 goto finish_meta_pat;
6922 RExC_seen_zerolen++;
6923 RExC_seen |= REG_SEEN_LOOKBEHIND;
6924 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6926 goto finish_meta_pat;
6928 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6929 *flagp |= HASWIDTH|SIMPLE;
6930 goto finish_meta_pat;
6932 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6933 *flagp |= HASWIDTH|SIMPLE;
6934 goto finish_meta_pat;
6936 ret = reg_node(pRExC_state, DIGIT);
6937 *flagp |= HASWIDTH|SIMPLE;
6938 goto finish_meta_pat;
6940 ret = reg_node(pRExC_state, NDIGIT);
6941 *flagp |= HASWIDTH|SIMPLE;
6942 goto finish_meta_pat;
6944 ret = reg_node(pRExC_state, LNBREAK);
6945 *flagp |= HASWIDTH|SIMPLE;
6946 goto finish_meta_pat;
6948 ret = reg_node(pRExC_state, HORIZWS);
6949 *flagp |= HASWIDTH|SIMPLE;
6950 goto finish_meta_pat;
6952 ret = reg_node(pRExC_state, NHORIZWS);
6953 *flagp |= HASWIDTH|SIMPLE;
6954 goto finish_meta_pat;
6956 ret = reg_node(pRExC_state, VERTWS);
6957 *flagp |= HASWIDTH|SIMPLE;
6958 goto finish_meta_pat;
6960 ret = reg_node(pRExC_state, NVERTWS);
6961 *flagp |= HASWIDTH|SIMPLE;
6963 nextchar(pRExC_state);
6964 Set_Node_Length(ret, 2); /* MJD */
6969 char* const oldregxend = RExC_end;
6971 char* parse_start = RExC_parse - 2;
6974 if (RExC_parse[1] == '{') {
6975 /* a lovely hack--pretend we saw [\pX] instead */
6976 RExC_end = strchr(RExC_parse, '}');
6978 const U8 c = (U8)*RExC_parse;
6980 RExC_end = oldregxend;
6981 vFAIL2("Missing right brace on \\%c{}", c);
6986 RExC_end = RExC_parse + 2;
6987 if (RExC_end > oldregxend)
6988 RExC_end = oldregxend;
6992 ret = regclass(pRExC_state,depth+1);
6994 RExC_end = oldregxend;
6997 Set_Node_Offset(ret, parse_start + 2);
6998 Set_Node_Cur_Length(ret);
6999 nextchar(pRExC_state);
7000 *flagp |= HASWIDTH|SIMPLE;
7004 /* Handle \N{NAME} here and not below because it can be
7005 multicharacter. join_exact() will join them up later on.
7006 Also this makes sure that things like /\N{BLAH}+/ and
7007 \N{BLAH} being multi char Just Happen. dmq*/
7009 ret= reg_namedseq(pRExC_state, NULL);
7011 case 'k': /* Handle \k<NAME> and \k'NAME' */
7014 char ch= RExC_parse[1];
7015 if (ch != '<' && ch != '\'' && ch != '{') {
7017 vFAIL2("Sequence %.2s... not terminated",parse_start);
7019 /* this pretty much dupes the code for (?P=...) in reg(), if
7020 you change this make sure you change that */
7021 char* name_start = (RExC_parse += 2);
7023 SV *sv_dat = reg_scan_name(pRExC_state,
7024 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7025 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7026 if (RExC_parse == name_start || *RExC_parse != ch)
7027 vFAIL2("Sequence %.3s... not terminated",parse_start);
7030 num = add_data( pRExC_state, 1, "S" );
7031 RExC_rxi->data->data[num]=(void*)sv_dat;
7032 SvREFCNT_inc_simple_void(sv_dat);
7036 ret = reganode(pRExC_state,
7037 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7041 /* override incorrect value set in reganode MJD */
7042 Set_Node_Offset(ret, parse_start+1);
7043 Set_Node_Cur_Length(ret); /* MJD */
7044 nextchar(pRExC_state);
7050 case '1': case '2': case '3': case '4':
7051 case '5': case '6': case '7': case '8': case '9':
7054 bool isg = *RExC_parse == 'g';
7059 if (*RExC_parse == '{') {
7063 if (*RExC_parse == '-') {
7067 if (hasbrace && !isDIGIT(*RExC_parse)) {
7068 if (isrel) RExC_parse--;
7070 goto parse_named_seq;
7072 num = atoi(RExC_parse);
7073 if (isg && num == 0)
7074 vFAIL("Reference to invalid group 0");
7076 num = RExC_npar - num;
7078 vFAIL("Reference to nonexistent or unclosed group");
7080 if (!isg && num > 9 && num >= RExC_npar)
7083 char * const parse_start = RExC_parse - 1; /* MJD */
7084 while (isDIGIT(*RExC_parse))
7086 if (parse_start == RExC_parse - 1)
7087 vFAIL("Unterminated \\g... pattern");
7089 if (*RExC_parse != '}')
7090 vFAIL("Unterminated \\g{...} pattern");
7094 if (num > (I32)RExC_rx->nparens)
7095 vFAIL("Reference to nonexistent group");
7098 ret = reganode(pRExC_state,
7099 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7103 /* override incorrect value set in reganode MJD */
7104 Set_Node_Offset(ret, parse_start+1);
7105 Set_Node_Cur_Length(ret); /* MJD */
7107 nextchar(pRExC_state);
7112 if (RExC_parse >= RExC_end)
7113 FAIL("Trailing \\");
7116 /* Do not generate "unrecognized" warnings here, we fall
7117 back into the quick-grab loop below */
7124 if (RExC_flags & RXf_PMf_EXTENDED) {
7125 if ( reg_skipcomment( pRExC_state ) )
7132 register STRLEN len;
7137 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7139 parse_start = RExC_parse - 1;
7145 ret = reg_node(pRExC_state,
7146 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7148 for (len = 0, p = RExC_parse - 1;
7149 len < 127 && p < RExC_end;
7152 char * const oldp = p;
7154 if (RExC_flags & RXf_PMf_EXTENDED)
7155 p = regwhite( pRExC_state, p );
7160 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7161 goto normal_default;
7171 /* Literal Escapes Switch
7173 This switch is meant to handle escape sequences that
7174 resolve to a literal character.
7176 Every escape sequence that represents something
7177 else, like an assertion or a char class, is handled
7178 in the switch marked 'Special Escapes' above in this
7179 routine, but also has an entry here as anything that
7180 isn't explicitly mentioned here will be treated as
7181 an unescaped equivalent literal.
7185 /* These are all the special escapes. */
7186 case 'A': /* Start assertion */
7187 case 'b': case 'B': /* Word-boundary assertion*/
7188 case 'C': /* Single char !DANGEROUS! */
7189 case 'd': case 'D': /* digit class */
7190 case 'g': case 'G': /* generic-backref, pos assertion */
7191 case 'h': case 'H': /* HORIZWS */
7192 case 'k': case 'K': /* named backref, keep marker */
7193 case 'N': /* named char sequence */
7194 case 'p': case 'P': /* Unicode property */
7195 case 'R': /* LNBREAK */
7196 case 's': case 'S': /* space class */
7197 case 'v': case 'V': /* VERTWS */
7198 case 'w': case 'W': /* word class */
7199 case 'X': /* eXtended Unicode "combining character sequence" */
7200 case 'z': case 'Z': /* End of line/string assertion */
7204 /* Anything after here is an escape that resolves to a
7205 literal. (Except digits, which may or may not)
7224 ender = ASCII_TO_NATIVE('\033');
7228 ender = ASCII_TO_NATIVE('\007');
7233 char* const e = strchr(p, '}');
7237 vFAIL("Missing right brace on \\x{}");
7240 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7241 | PERL_SCAN_DISALLOW_PREFIX;
7242 STRLEN numlen = e - p - 1;
7243 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7250 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7252 ender = grok_hex(p, &numlen, &flags, NULL);
7255 if (PL_encoding && ender < 0x100)
7256 goto recode_encoding;
7260 ender = UCHARAT(p++);
7261 ender = toCTRL(ender);
7263 case '0': case '1': case '2': case '3':case '4':
7264 case '5': case '6': case '7': case '8':case '9':
7266 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7269 ender = grok_oct(p, &numlen, &flags, NULL);
7276 if (PL_encoding && ender < 0x100)
7277 goto recode_encoding;
7281 SV* enc = PL_encoding;
7282 ender = reg_recode((const char)(U8)ender, &enc);
7283 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7284 vWARN(p, "Invalid escape in the specified encoding");
7290 FAIL("Trailing \\");
7293 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
7294 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7295 goto normal_default;
7300 if (UTF8_IS_START(*p) && UTF) {
7302 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7303 &numlen, UTF8_ALLOW_DEFAULT);
7310 if ( RExC_flags & RXf_PMf_EXTENDED)
7311 p = regwhite( pRExC_state, p );
7313 /* Prime the casefolded buffer. */
7314 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7316 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7321 /* Emit all the Unicode characters. */
7323 for (foldbuf = tmpbuf;
7325 foldlen -= numlen) {
7326 ender = utf8_to_uvchr(foldbuf, &numlen);
7328 const STRLEN unilen = reguni(pRExC_state, ender, s);
7331 /* In EBCDIC the numlen
7332 * and unilen can differ. */
7334 if (numlen >= foldlen)
7338 break; /* "Can't happen." */
7342 const STRLEN unilen = reguni(pRExC_state, ender, s);
7351 REGC((char)ender, s++);
7357 /* Emit all the Unicode characters. */
7359 for (foldbuf = tmpbuf;
7361 foldlen -= numlen) {
7362 ender = utf8_to_uvchr(foldbuf, &numlen);
7364 const STRLEN unilen = reguni(pRExC_state, ender, s);
7367 /* In EBCDIC the numlen
7368 * and unilen can differ. */
7370 if (numlen >= foldlen)
7378 const STRLEN unilen = reguni(pRExC_state, ender, s);
7387 REGC((char)ender, s++);
7391 Set_Node_Cur_Length(ret); /* MJD */
7392 nextchar(pRExC_state);
7394 /* len is STRLEN which is unsigned, need to copy to signed */
7397 vFAIL("Internal disaster");
7401 if (len == 1 && UNI_IS_INVARIANT(ender))
7405 RExC_size += STR_SZ(len);
7408 RExC_emit += STR_SZ(len);
7418 S_regwhite( RExC_state_t *pRExC_state, char *p )
7420 const char *e = RExC_end;
7424 else if (*p == '#') {
7433 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7441 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7442 Character classes ([:foo:]) can also be negated ([:^foo:]).
7443 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7444 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7445 but trigger failures because they are currently unimplemented. */
7447 #define POSIXCC_DONE(c) ((c) == ':')
7448 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7449 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7452 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7455 I32 namedclass = OOB_NAMEDCLASS;
7457 if (value == '[' && RExC_parse + 1 < RExC_end &&
7458 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7459 POSIXCC(UCHARAT(RExC_parse))) {
7460 const char c = UCHARAT(RExC_parse);
7461 char* const s = RExC_parse++;
7463 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7465 if (RExC_parse == RExC_end)
7466 /* Grandfather lone [:, [=, [. */
7469 const char* const t = RExC_parse++; /* skip over the c */
7472 if (UCHARAT(RExC_parse) == ']') {
7473 const char *posixcc = s + 1;
7474 RExC_parse++; /* skip over the ending ] */
7477 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7478 const I32 skip = t - posixcc;
7480 /* Initially switch on the length of the name. */
7483 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7484 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7487 /* Names all of length 5. */
7488 /* alnum alpha ascii blank cntrl digit graph lower
7489 print punct space upper */
7490 /* Offset 4 gives the best switch position. */
7491 switch (posixcc[4]) {
7493 if (memEQ(posixcc, "alph", 4)) /* alpha */
7494 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7497 if (memEQ(posixcc, "spac", 4)) /* space */
7498 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7501 if (memEQ(posixcc, "grap", 4)) /* graph */
7502 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7505 if (memEQ(posixcc, "asci", 4)) /* ascii */
7506 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7509 if (memEQ(posixcc, "blan", 4)) /* blank */
7510 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7513 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7514 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7517 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7518 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7521 if (memEQ(posixcc, "lowe", 4)) /* lower */
7522 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7523 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7524 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7527 if (memEQ(posixcc, "digi", 4)) /* digit */
7528 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7529 else if (memEQ(posixcc, "prin", 4)) /* print */
7530 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7531 else if (memEQ(posixcc, "punc", 4)) /* punct */
7532 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7537 if (memEQ(posixcc, "xdigit", 6))
7538 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7542 if (namedclass == OOB_NAMEDCLASS)
7543 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7545 assert (posixcc[skip] == ':');
7546 assert (posixcc[skip+1] == ']');
7547 } else if (!SIZE_ONLY) {
7548 /* [[=foo=]] and [[.foo.]] are still future. */
7550 /* adjust RExC_parse so the warning shows after
7552 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7554 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7557 /* Maternal grandfather:
7558 * "[:" ending in ":" but not in ":]" */
7568 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7571 if (POSIXCC(UCHARAT(RExC_parse))) {
7572 const char *s = RExC_parse;
7573 const char c = *s++;
7577 if (*s && c == *s && s[1] == ']') {
7578 if (ckWARN(WARN_REGEXP))
7580 "POSIX syntax [%c %c] belongs inside character classes",
7583 /* [[=foo=]] and [[.foo.]] are still future. */
7584 if (POSIXCC_NOTYET(c)) {
7585 /* adjust RExC_parse so the error shows after
7587 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7589 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7596 #define _C_C_T_(NAME,TEST,WORD) \
7599 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7601 for (value = 0; value < 256; value++) \
7603 ANYOF_BITMAP_SET(ret, value); \
7608 case ANYOF_N##NAME: \
7610 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7612 for (value = 0; value < 256; value++) \
7614 ANYOF_BITMAP_SET(ret, value); \
7620 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7622 for (value = 0; value < 256; value++) \
7624 ANYOF_BITMAP_SET(ret, value); \
7628 case ANYOF_N##NAME: \
7629 for (value = 0; value < 256; value++) \
7631 ANYOF_BITMAP_SET(ret, value); \
7637 parse a class specification and produce either an ANYOF node that
7638 matches the pattern or if the pattern matches a single char only and
7639 that char is < 256 and we are case insensitive then we produce an
7644 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7647 register UV nextvalue;
7648 register IV prevvalue = OOB_UNICODE;
7649 register IV range = 0;
7650 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7651 register regnode *ret;
7654 char *rangebegin = NULL;
7655 bool need_class = 0;
7658 bool optimize_invert = TRUE;
7659 AV* unicode_alternate = NULL;
7661 UV literal_endpoint = 0;
7663 UV stored = 0; /* number of chars stored in the class */
7665 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7666 case we need to change the emitted regop to an EXACT. */
7667 const char * orig_parse = RExC_parse;
7668 GET_RE_DEBUG_FLAGS_DECL;
7670 PERL_UNUSED_ARG(depth);
7673 DEBUG_PARSE("clas");
7675 /* Assume we are going to generate an ANYOF node. */
7676 ret = reganode(pRExC_state, ANYOF, 0);
7679 ANYOF_FLAGS(ret) = 0;
7681 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7685 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7689 RExC_size += ANYOF_SKIP;
7690 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7693 RExC_emit += ANYOF_SKIP;
7695 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7697 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7698 ANYOF_BITMAP_ZERO(ret);
7699 listsv = newSVpvs("# comment\n");
7702 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7704 if (!SIZE_ONLY && POSIXCC(nextvalue))
7705 checkposixcc(pRExC_state);
7707 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7708 if (UCHARAT(RExC_parse) == ']')
7712 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7716 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7719 rangebegin = RExC_parse;
7721 value = utf8n_to_uvchr((U8*)RExC_parse,
7722 RExC_end - RExC_parse,
7723 &numlen, UTF8_ALLOW_DEFAULT);
7724 RExC_parse += numlen;
7727 value = UCHARAT(RExC_parse++);
7729 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7730 if (value == '[' && POSIXCC(nextvalue))
7731 namedclass = regpposixcc(pRExC_state, value);
7732 else if (value == '\\') {
7734 value = utf8n_to_uvchr((U8*)RExC_parse,
7735 RExC_end - RExC_parse,
7736 &numlen, UTF8_ALLOW_DEFAULT);
7737 RExC_parse += numlen;
7740 value = UCHARAT(RExC_parse++);
7741 /* Some compilers cannot handle switching on 64-bit integer
7742 * values, therefore value cannot be an UV. Yes, this will
7743 * be a problem later if we want switch on Unicode.
7744 * A similar issue a little bit later when switching on
7745 * namedclass. --jhi */
7746 switch ((I32)value) {
7747 case 'w': namedclass = ANYOF_ALNUM; break;
7748 case 'W': namedclass = ANYOF_NALNUM; break;
7749 case 's': namedclass = ANYOF_SPACE; break;
7750 case 'S': namedclass = ANYOF_NSPACE; break;
7751 case 'd': namedclass = ANYOF_DIGIT; break;
7752 case 'D': namedclass = ANYOF_NDIGIT; break;
7753 case 'v': namedclass = ANYOF_VERTWS; break;
7754 case 'V': namedclass = ANYOF_NVERTWS; break;
7755 case 'h': namedclass = ANYOF_HORIZWS; break;
7756 case 'H': namedclass = ANYOF_NHORIZWS; break;
7757 case 'N': /* Handle \N{NAME} in class */
7759 /* We only pay attention to the first char of
7760 multichar strings being returned. I kinda wonder
7761 if this makes sense as it does change the behaviour
7762 from earlier versions, OTOH that behaviour was broken
7764 UV v; /* value is register so we cant & it /grrr */
7765 if (reg_namedseq(pRExC_state, &v)) {
7775 if (RExC_parse >= RExC_end)
7776 vFAIL2("Empty \\%c{}", (U8)value);
7777 if (*RExC_parse == '{') {
7778 const U8 c = (U8)value;
7779 e = strchr(RExC_parse++, '}');
7781 vFAIL2("Missing right brace on \\%c{}", c);
7782 while (isSPACE(UCHARAT(RExC_parse)))
7784 if (e == RExC_parse)
7785 vFAIL2("Empty \\%c{}", c);
7787 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7795 if (UCHARAT(RExC_parse) == '^') {
7798 value = value == 'p' ? 'P' : 'p'; /* toggle */
7799 while (isSPACE(UCHARAT(RExC_parse))) {
7804 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7805 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7808 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7809 namedclass = ANYOF_MAX; /* no official name, but it's named */
7812 case 'n': value = '\n'; break;
7813 case 'r': value = '\r'; break;
7814 case 't': value = '\t'; break;
7815 case 'f': value = '\f'; break;
7816 case 'b': value = '\b'; break;
7817 case 'e': value = ASCII_TO_NATIVE('\033');break;
7818 case 'a': value = ASCII_TO_NATIVE('\007');break;
7820 if (*RExC_parse == '{') {
7821 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7822 | PERL_SCAN_DISALLOW_PREFIX;
7823 char * const e = strchr(RExC_parse++, '}');
7825 vFAIL("Missing right brace on \\x{}");
7827 numlen = e - RExC_parse;
7828 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7832 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7834 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7835 RExC_parse += numlen;
7837 if (PL_encoding && value < 0x100)
7838 goto recode_encoding;
7841 value = UCHARAT(RExC_parse++);
7842 value = toCTRL(value);
7844 case '0': case '1': case '2': case '3': case '4':
7845 case '5': case '6': case '7': case '8': case '9':
7849 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7850 RExC_parse += numlen;
7851 if (PL_encoding && value < 0x100)
7852 goto recode_encoding;
7857 SV* enc = PL_encoding;
7858 value = reg_recode((const char)(U8)value, &enc);
7859 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7861 "Invalid escape in the specified encoding");
7865 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7867 "Unrecognized escape \\%c in character class passed through",
7871 } /* end of \blah */
7877 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7879 if (!SIZE_ONLY && !need_class)
7880 ANYOF_CLASS_ZERO(ret);
7884 /* a bad range like a-\d, a-[:digit:] ? */
7887 if (ckWARN(WARN_REGEXP)) {
7889 RExC_parse >= rangebegin ?
7890 RExC_parse - rangebegin : 0;
7892 "False [] range \"%*.*s\"",
7895 if (prevvalue < 256) {
7896 ANYOF_BITMAP_SET(ret, prevvalue);
7897 ANYOF_BITMAP_SET(ret, '-');
7900 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7901 Perl_sv_catpvf(aTHX_ listsv,
7902 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7906 range = 0; /* this was not a true range */
7912 const char *what = NULL;
7915 if (namedclass > OOB_NAMEDCLASS)
7916 optimize_invert = FALSE;
7917 /* Possible truncation here but in some 64-bit environments
7918 * the compiler gets heartburn about switch on 64-bit values.
7919 * A similar issue a little earlier when switching on value.
7921 switch ((I32)namedclass) {
7922 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7923 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7924 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7925 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7926 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7927 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7928 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7929 case _C_C_T_(PRINT, isPRINT(value), "Print");
7930 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7931 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7932 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7933 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7934 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7935 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
7936 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
7939 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7942 for (value = 0; value < 128; value++)
7943 ANYOF_BITMAP_SET(ret, value);
7945 for (value = 0; value < 256; value++) {
7947 ANYOF_BITMAP_SET(ret, value);
7956 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7959 for (value = 128; value < 256; value++)
7960 ANYOF_BITMAP_SET(ret, value);
7962 for (value = 0; value < 256; value++) {
7963 if (!isASCII(value))
7964 ANYOF_BITMAP_SET(ret, value);
7973 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7975 /* consecutive digits assumed */
7976 for (value = '0'; value <= '9'; value++)
7977 ANYOF_BITMAP_SET(ret, value);
7984 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7986 /* consecutive digits assumed */
7987 for (value = 0; value < '0'; value++)
7988 ANYOF_BITMAP_SET(ret, value);
7989 for (value = '9' + 1; value < 256; value++)
7990 ANYOF_BITMAP_SET(ret, value);
7996 /* this is to handle \p and \P */
7999 vFAIL("Invalid [::] class");
8003 /* Strings such as "+utf8::isWord\n" */
8004 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8007 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8010 } /* end of namedclass \blah */
8013 if (prevvalue > (IV)value) /* b-a */ {
8014 const int w = RExC_parse - rangebegin;
8015 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8016 range = 0; /* not a valid range */
8020 prevvalue = value; /* save the beginning of the range */
8021 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8022 RExC_parse[1] != ']') {
8025 /* a bad range like \w-, [:word:]- ? */
8026 if (namedclass > OOB_NAMEDCLASS) {
8027 if (ckWARN(WARN_REGEXP)) {
8029 RExC_parse >= rangebegin ?
8030 RExC_parse - rangebegin : 0;
8032 "False [] range \"%*.*s\"",
8036 ANYOF_BITMAP_SET(ret, '-');
8038 range = 1; /* yeah, it's a range! */
8039 continue; /* but do it the next time */
8043 /* now is the next time */
8044 /*stored += (value - prevvalue + 1);*/
8046 if (prevvalue < 256) {
8047 const IV ceilvalue = value < 256 ? value : 255;
8050 /* In EBCDIC [\x89-\x91] should include
8051 * the \x8e but [i-j] should not. */
8052 if (literal_endpoint == 2 &&
8053 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8054 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8056 if (isLOWER(prevvalue)) {
8057 for (i = prevvalue; i <= ceilvalue; i++)
8058 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8060 ANYOF_BITMAP_SET(ret, i);
8063 for (i = prevvalue; i <= ceilvalue; i++)
8064 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8066 ANYOF_BITMAP_SET(ret, i);
8072 for (i = prevvalue; i <= ceilvalue; i++) {
8073 if (!ANYOF_BITMAP_TEST(ret,i)) {
8075 ANYOF_BITMAP_SET(ret, i);
8079 if (value > 255 || UTF) {
8080 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8081 const UV natvalue = NATIVE_TO_UNI(value);
8082 stored+=2; /* can't optimize this class */
8083 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8084 if (prevnatvalue < natvalue) { /* what about > ? */
8085 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8086 prevnatvalue, natvalue);
8088 else if (prevnatvalue == natvalue) {
8089 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8091 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8093 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8095 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8096 if (RExC_precomp[0] == ':' &&
8097 RExC_precomp[1] == '[' &&
8098 (f == 0xDF || f == 0x92)) {
8099 f = NATIVE_TO_UNI(f);
8102 /* If folding and foldable and a single
8103 * character, insert also the folded version
8104 * to the charclass. */
8106 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8107 if ((RExC_precomp[0] == ':' &&
8108 RExC_precomp[1] == '[' &&
8110 (value == 0xFB05 || value == 0xFB06))) ?
8111 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8112 foldlen == (STRLEN)UNISKIP(f) )
8114 if (foldlen == (STRLEN)UNISKIP(f))
8116 Perl_sv_catpvf(aTHX_ listsv,
8119 /* Any multicharacter foldings
8120 * require the following transform:
8121 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8122 * where E folds into "pq" and F folds
8123 * into "rst", all other characters
8124 * fold to single characters. We save
8125 * away these multicharacter foldings,
8126 * to be later saved as part of the
8127 * additional "s" data. */
8130 if (!unicode_alternate)
8131 unicode_alternate = newAV();
8132 sv = newSVpvn((char*)foldbuf, foldlen);
8134 av_push(unicode_alternate, sv);
8138 /* If folding and the value is one of the Greek
8139 * sigmas insert a few more sigmas to make the
8140 * folding rules of the sigmas to work right.
8141 * Note that not all the possible combinations
8142 * are handled here: some of them are handled
8143 * by the standard folding rules, and some of
8144 * them (literal or EXACTF cases) are handled
8145 * during runtime in regexec.c:S_find_byclass(). */
8146 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8147 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8148 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8149 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8150 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8152 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8153 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8154 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8159 literal_endpoint = 0;
8163 range = 0; /* this range (if it was one) is done now */
8167 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8169 RExC_size += ANYOF_CLASS_ADD_SKIP;
8171 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8177 /****** !SIZE_ONLY AFTER HERE *********/
8179 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8180 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8182 /* optimize single char class to an EXACT node
8183 but *only* when its not a UTF/high char */
8184 const char * cur_parse= RExC_parse;
8185 RExC_emit = (regnode *)orig_emit;
8186 RExC_parse = (char *)orig_parse;
8187 ret = reg_node(pRExC_state,
8188 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8189 RExC_parse = (char *)cur_parse;
8190 *STRING(ret)= (char)value;
8192 RExC_emit += STR_SZ(1);
8195 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8196 if ( /* If the only flag is folding (plus possibly inversion). */
8197 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8199 for (value = 0; value < 256; ++value) {
8200 if (ANYOF_BITMAP_TEST(ret, value)) {
8201 UV fold = PL_fold[value];
8204 ANYOF_BITMAP_SET(ret, fold);
8207 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8210 /* optimize inverted simple patterns (e.g. [^a-z]) */
8211 if (optimize_invert &&
8212 /* If the only flag is inversion. */
8213 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8214 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8215 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8216 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8219 AV * const av = newAV();
8221 /* The 0th element stores the character class description
8222 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8223 * to initialize the appropriate swash (which gets stored in
8224 * the 1st element), and also useful for dumping the regnode.
8225 * The 2nd element stores the multicharacter foldings,
8226 * used later (regexec.c:S_reginclass()). */
8227 av_store(av, 0, listsv);
8228 av_store(av, 1, NULL);
8229 av_store(av, 2, (SV*)unicode_alternate);
8230 rv = newRV_noinc((SV*)av);
8231 n = add_data(pRExC_state, 1, "s");
8232 RExC_rxi->data->data[n] = (void*)rv;
8240 /* reg_skipcomment()
8242 Absorbs an /x style # comments from the input stream.
8243 Returns true if there is more text remaining in the stream.
8244 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8245 terminates the pattern without including a newline.
8247 Note its the callers responsibility to ensure that we are
8253 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8256 while (RExC_parse < RExC_end)
8257 if (*RExC_parse++ == '\n') {
8262 /* we ran off the end of the pattern without ending
8263 the comment, so we have to add an \n when wrapping */
8264 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8272 Advance that parse position, and optionall absorbs
8273 "whitespace" from the inputstream.
8275 Without /x "whitespace" means (?#...) style comments only,
8276 with /x this means (?#...) and # comments and whitespace proper.
8278 Returns the RExC_parse point from BEFORE the scan occurs.
8280 This is the /x friendly way of saying RExC_parse++.
8284 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8286 char* const retval = RExC_parse++;
8289 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8290 RExC_parse[2] == '#') {
8291 while (*RExC_parse != ')') {
8292 if (RExC_parse == RExC_end)
8293 FAIL("Sequence (?#... not terminated");
8299 if (RExC_flags & RXf_PMf_EXTENDED) {
8300 if (isSPACE(*RExC_parse)) {
8304 else if (*RExC_parse == '#') {
8305 if ( reg_skipcomment( pRExC_state ) )
8314 - reg_node - emit a node
8316 STATIC regnode * /* Location. */
8317 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8320 register regnode *ptr;
8321 regnode * const ret = RExC_emit;
8322 GET_RE_DEBUG_FLAGS_DECL;
8325 SIZE_ALIGN(RExC_size);
8329 if (RExC_emit >= RExC_emit_bound)
8330 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8332 NODE_ALIGN_FILL(ret);
8334 FILL_ADVANCE_NODE(ptr, op);
8335 #ifdef RE_TRACK_PATTERN_OFFSETS
8336 if (RExC_offsets) { /* MJD */
8337 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8338 "reg_node", __LINE__,
8340 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8341 ? "Overwriting end of array!\n" : "OK",
8342 (UV)(RExC_emit - RExC_emit_start),
8343 (UV)(RExC_parse - RExC_start),
8344 (UV)RExC_offsets[0]));
8345 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8353 - reganode - emit a node with an argument
8355 STATIC regnode * /* Location. */
8356 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8359 register regnode *ptr;
8360 regnode * const ret = RExC_emit;
8361 GET_RE_DEBUG_FLAGS_DECL;
8364 SIZE_ALIGN(RExC_size);
8369 assert(2==regarglen[op]+1);
8371 Anything larger than this has to allocate the extra amount.
8372 If we changed this to be:
8374 RExC_size += (1 + regarglen[op]);
8376 then it wouldn't matter. Its not clear what side effect
8377 might come from that so its not done so far.
8382 if (RExC_emit >= RExC_emit_bound)
8383 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8385 NODE_ALIGN_FILL(ret);
8387 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8388 #ifdef RE_TRACK_PATTERN_OFFSETS
8389 if (RExC_offsets) { /* MJD */
8390 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8394 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8395 "Overwriting end of array!\n" : "OK",
8396 (UV)(RExC_emit - RExC_emit_start),
8397 (UV)(RExC_parse - RExC_start),
8398 (UV)RExC_offsets[0]));
8399 Set_Cur_Node_Offset;
8407 - reguni - emit (if appropriate) a Unicode character
8410 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8413 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8417 - reginsert - insert an operator in front of already-emitted operand
8419 * Means relocating the operand.
8422 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8425 register regnode *src;
8426 register regnode *dst;
8427 register regnode *place;
8428 const int offset = regarglen[(U8)op];
8429 const int size = NODE_STEP_REGNODE + offset;
8430 GET_RE_DEBUG_FLAGS_DECL;
8431 PERL_UNUSED_ARG(depth);
8432 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8433 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8442 if (RExC_open_parens) {
8444 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8445 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8446 if ( RExC_open_parens[paren] >= opnd ) {
8447 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8448 RExC_open_parens[paren] += size;
8450 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8452 if ( RExC_close_parens[paren] >= opnd ) {
8453 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8454 RExC_close_parens[paren] += size;
8456 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8461 while (src > opnd) {
8462 StructCopy(--src, --dst, regnode);
8463 #ifdef RE_TRACK_PATTERN_OFFSETS
8464 if (RExC_offsets) { /* MJD 20010112 */
8465 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8469 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8470 ? "Overwriting end of array!\n" : "OK",
8471 (UV)(src - RExC_emit_start),
8472 (UV)(dst - RExC_emit_start),
8473 (UV)RExC_offsets[0]));
8474 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8475 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8481 place = opnd; /* Op node, where operand used to be. */
8482 #ifdef RE_TRACK_PATTERN_OFFSETS
8483 if (RExC_offsets) { /* MJD */
8484 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8488 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8489 ? "Overwriting end of array!\n" : "OK",
8490 (UV)(place - RExC_emit_start),
8491 (UV)(RExC_parse - RExC_start),
8492 (UV)RExC_offsets[0]));
8493 Set_Node_Offset(place, RExC_parse);
8494 Set_Node_Length(place, 1);
8497 src = NEXTOPER(place);
8498 FILL_ADVANCE_NODE(place, op);
8499 Zero(src, offset, regnode);
8503 - regtail - set the next-pointer at the end of a node chain of p to val.
8504 - SEE ALSO: regtail_study
8506 /* TODO: All three parms should be const */
8508 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8511 register regnode *scan;
8512 GET_RE_DEBUG_FLAGS_DECL;
8514 PERL_UNUSED_ARG(depth);
8520 /* Find last node. */
8523 regnode * const temp = regnext(scan);
8525 SV * const mysv=sv_newmortal();
8526 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8527 regprop(RExC_rx, mysv, scan);
8528 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8529 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8530 (temp == NULL ? "->" : ""),
8531 (temp == NULL ? PL_reg_name[OP(val)] : "")
8539 if (reg_off_by_arg[OP(scan)]) {
8540 ARG_SET(scan, val - scan);
8543 NEXT_OFF(scan) = val - scan;
8549 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8550 - Look for optimizable sequences at the same time.
8551 - currently only looks for EXACT chains.
8553 This is expermental code. The idea is to use this routine to perform
8554 in place optimizations on branches and groups as they are constructed,
8555 with the long term intention of removing optimization from study_chunk so
8556 that it is purely analytical.
8558 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8559 to control which is which.
8562 /* TODO: All four parms should be const */
8565 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8568 register regnode *scan;
8570 #ifdef EXPERIMENTAL_INPLACESCAN
8574 GET_RE_DEBUG_FLAGS_DECL;
8580 /* Find last node. */
8584 regnode * const temp = regnext(scan);
8585 #ifdef EXPERIMENTAL_INPLACESCAN
8586 if (PL_regkind[OP(scan)] == EXACT)
8587 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8595 if( exact == PSEUDO )
8597 else if ( exact != OP(scan) )
8606 SV * const mysv=sv_newmortal();
8607 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8608 regprop(RExC_rx, mysv, scan);
8609 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8610 SvPV_nolen_const(mysv),
8612 PL_reg_name[exact]);
8619 SV * const mysv_val=sv_newmortal();
8620 DEBUG_PARSE_MSG("");
8621 regprop(RExC_rx, mysv_val, val);
8622 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8623 SvPV_nolen_const(mysv_val),
8624 (IV)REG_NODE_NUM(val),
8628 if (reg_off_by_arg[OP(scan)]) {
8629 ARG_SET(scan, val - scan);
8632 NEXT_OFF(scan) = val - scan;
8640 - regcurly - a little FSA that accepts {\d+,?\d*}
8643 S_regcurly(register const char *s)
8662 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8666 S_regdump_extflags(pTHX_ const char *lead, const U32 flags) {
8669 for (bit=0; bit<32; bit++) {
8670 if (flags & (1<<bit)) {
8672 PerlIO_printf(Perl_debug_log, "%s",lead);
8673 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
8678 PerlIO_printf(Perl_debug_log, "\n");
8680 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
8686 Perl_regdump(pTHX_ const regexp *r)
8690 SV * const sv = sv_newmortal();
8691 SV *dsv= sv_newmortal();
8693 GET_RE_DEBUG_FLAGS_DECL;
8695 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8697 /* Header fields of interest. */
8698 if (r->anchored_substr) {
8699 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8700 RE_SV_DUMPLEN(r->anchored_substr), 30);
8701 PerlIO_printf(Perl_debug_log,
8702 "anchored %s%s at %"IVdf" ",
8703 s, RE_SV_TAIL(r->anchored_substr),
8704 (IV)r->anchored_offset);
8705 } else if (r->anchored_utf8) {
8706 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8707 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8708 PerlIO_printf(Perl_debug_log,
8709 "anchored utf8 %s%s at %"IVdf" ",
8710 s, RE_SV_TAIL(r->anchored_utf8),
8711 (IV)r->anchored_offset);
8713 if (r->float_substr) {
8714 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8715 RE_SV_DUMPLEN(r->float_substr), 30);
8716 PerlIO_printf(Perl_debug_log,
8717 "floating %s%s at %"IVdf"..%"UVuf" ",
8718 s, RE_SV_TAIL(r->float_substr),
8719 (IV)r->float_min_offset, (UV)r->float_max_offset);
8720 } else if (r->float_utf8) {
8721 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8722 RE_SV_DUMPLEN(r->float_utf8), 30);
8723 PerlIO_printf(Perl_debug_log,
8724 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8725 s, RE_SV_TAIL(r->float_utf8),
8726 (IV)r->float_min_offset, (UV)r->float_max_offset);
8728 if (r->check_substr || r->check_utf8)
8729 PerlIO_printf(Perl_debug_log,
8731 (r->check_substr == r->float_substr
8732 && r->check_utf8 == r->float_utf8
8733 ? "(checking floating" : "(checking anchored"));
8734 if (r->extflags & RXf_NOSCAN)
8735 PerlIO_printf(Perl_debug_log, " noscan");
8736 if (r->extflags & RXf_CHECK_ALL)
8737 PerlIO_printf(Perl_debug_log, " isall");
8738 if (r->check_substr || r->check_utf8)
8739 PerlIO_printf(Perl_debug_log, ") ");
8741 if (ri->regstclass) {
8742 regprop(r, sv, ri->regstclass);
8743 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8745 if (r->extflags & RXf_ANCH) {
8746 PerlIO_printf(Perl_debug_log, "anchored");
8747 if (r->extflags & RXf_ANCH_BOL)
8748 PerlIO_printf(Perl_debug_log, "(BOL)");
8749 if (r->extflags & RXf_ANCH_MBOL)
8750 PerlIO_printf(Perl_debug_log, "(MBOL)");
8751 if (r->extflags & RXf_ANCH_SBOL)
8752 PerlIO_printf(Perl_debug_log, "(SBOL)");
8753 if (r->extflags & RXf_ANCH_GPOS)
8754 PerlIO_printf(Perl_debug_log, "(GPOS)");
8755 PerlIO_putc(Perl_debug_log, ' ');
8757 if (r->extflags & RXf_GPOS_SEEN)
8758 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8759 if (r->intflags & PREGf_SKIP)
8760 PerlIO_printf(Perl_debug_log, "plus ");
8761 if (r->intflags & PREGf_IMPLICIT)
8762 PerlIO_printf(Perl_debug_log, "implicit ");
8763 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8764 if (r->extflags & RXf_EVAL_SEEN)
8765 PerlIO_printf(Perl_debug_log, "with eval ");
8766 PerlIO_printf(Perl_debug_log, "\n");
8767 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
8769 PERL_UNUSED_CONTEXT;
8771 #endif /* DEBUGGING */
8775 - regprop - printable representation of opcode
8778 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8783 RXi_GET_DECL(prog,progi);
8784 GET_RE_DEBUG_FLAGS_DECL;
8787 sv_setpvn(sv, "", 0);
8789 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8790 /* It would be nice to FAIL() here, but this may be called from
8791 regexec.c, and it would be hard to supply pRExC_state. */
8792 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8793 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
8795 k = PL_regkind[OP(o)];
8798 SV * const dsv = sv_2mortal(newSVpvs(""));
8799 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8800 * is a crude hack but it may be the best for now since
8801 * we have no flag "this EXACTish node was UTF-8"
8803 const char * const s =
8804 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8805 PL_colors[0], PL_colors[1],
8806 PERL_PV_ESCAPE_UNI_DETECT |
8807 PERL_PV_PRETTY_ELLIPSES |
8810 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8811 } else if (k == TRIE) {
8812 /* print the details of the trie in dumpuntil instead, as
8813 * progi->data isn't available here */
8814 const char op = OP(o);
8815 const U32 n = ARG(o);
8816 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8817 (reg_ac_data *)progi->data->data[n] :
8819 const reg_trie_data * const trie
8820 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8822 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
8823 DEBUG_TRIE_COMPILE_r(
8824 Perl_sv_catpvf(aTHX_ sv,
8825 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8826 (UV)trie->startstate,
8827 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8828 (UV)trie->wordcount,
8831 (UV)TRIE_CHARCOUNT(trie),
8832 (UV)trie->uniquecharcount
8835 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8837 int rangestart = -1;
8838 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8839 Perl_sv_catpvf(aTHX_ sv, "[");
8840 for (i = 0; i <= 256; i++) {
8841 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8842 if (rangestart == -1)
8844 } else if (rangestart != -1) {
8845 if (i <= rangestart + 3)
8846 for (; rangestart < i; rangestart++)
8847 put_byte(sv, rangestart);
8849 put_byte(sv, rangestart);
8851 put_byte(sv, i - 1);
8856 Perl_sv_catpvf(aTHX_ sv, "]");
8859 } else if (k == CURLY) {
8860 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8861 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8862 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8864 else if (k == WHILEM && o->flags) /* Ordinal/of */
8865 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8866 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8867 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8868 if ( prog->paren_names ) {
8869 if ( k != REF || OP(o) < NREF) {
8870 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8871 SV **name= av_fetch(list, ARG(o), 0 );
8873 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8876 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8877 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8878 I32 *nums=(I32*)SvPVX(sv_dat);
8879 SV **name= av_fetch(list, nums[0], 0 );
8882 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8883 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8884 (n ? "," : ""), (IV)nums[n]);
8886 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8890 } else if (k == GOSUB)
8891 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8892 else if (k == VERB) {
8894 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8895 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8896 } else if (k == LOGICAL)
8897 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8898 else if (k == FOLDCHAR)
8899 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
8900 else if (k == ANYOF) {
8901 int i, rangestart = -1;
8902 const U8 flags = ANYOF_FLAGS(o);
8904 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8905 static const char * const anyofs[] = {
8938 if (flags & ANYOF_LOCALE)
8939 sv_catpvs(sv, "{loc}");
8940 if (flags & ANYOF_FOLD)
8941 sv_catpvs(sv, "{i}");
8942 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8943 if (flags & ANYOF_INVERT)
8945 for (i = 0; i <= 256; i++) {
8946 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8947 if (rangestart == -1)
8949 } else if (rangestart != -1) {
8950 if (i <= rangestart + 3)
8951 for (; rangestart < i; rangestart++)
8952 put_byte(sv, rangestart);
8954 put_byte(sv, rangestart);
8956 put_byte(sv, i - 1);
8962 if (o->flags & ANYOF_CLASS)
8963 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8964 if (ANYOF_CLASS_TEST(o,i))
8965 sv_catpv(sv, anyofs[i]);
8967 if (flags & ANYOF_UNICODE)
8968 sv_catpvs(sv, "{unicode}");
8969 else if (flags & ANYOF_UNICODE_ALL)
8970 sv_catpvs(sv, "{unicode_all}");
8974 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8978 U8 s[UTF8_MAXBYTES_CASE+1];
8980 for (i = 0; i <= 256; i++) { /* just the first 256 */
8981 uvchr_to_utf8(s, i);
8983 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8984 if (rangestart == -1)
8986 } else if (rangestart != -1) {
8987 if (i <= rangestart + 3)
8988 for (; rangestart < i; rangestart++) {
8989 const U8 * const e = uvchr_to_utf8(s,rangestart);
8991 for(p = s; p < e; p++)
8995 const U8 *e = uvchr_to_utf8(s,rangestart);
8997 for (p = s; p < e; p++)
9000 e = uvchr_to_utf8(s, i-1);
9001 for (p = s; p < e; p++)
9008 sv_catpvs(sv, "..."); /* et cetera */
9012 char *s = savesvpv(lv);
9013 char * const origs = s;
9015 while (*s && *s != '\n')
9019 const char * const t = ++s;
9037 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9039 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9040 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9042 PERL_UNUSED_CONTEXT;
9043 PERL_UNUSED_ARG(sv);
9045 PERL_UNUSED_ARG(prog);
9046 #endif /* DEBUGGING */
9050 Perl_re_intuit_string(pTHX_ REGEXP * const prog)
9051 { /* Assume that RE_INTUIT is set */
9053 GET_RE_DEBUG_FLAGS_DECL;
9054 PERL_UNUSED_CONTEXT;
9058 const char * const s = SvPV_nolen_const(prog->check_substr
9059 ? prog->check_substr : prog->check_utf8);
9061 if (!PL_colorset) reginitcolors();
9062 PerlIO_printf(Perl_debug_log,
9063 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9065 prog->check_substr ? "" : "utf8 ",
9066 PL_colors[5],PL_colors[0],
9069 (strlen(s) > 60 ? "..." : ""));
9072 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9078 handles refcounting and freeing the perl core regexp structure. When
9079 it is necessary to actually free the structure the first thing it
9080 does is call the 'free' method of the regexp_engine associated to to
9081 the regexp, allowing the handling of the void *pprivate; member
9082 first. (This routine is not overridable by extensions, which is why
9083 the extensions free is called first.)
9085 See regdupe and regdupe_internal if you change anything here.
9087 #ifndef PERL_IN_XSUB_RE
9089 Perl_pregfree(pTHX_ struct regexp *r)
9092 GET_RE_DEBUG_FLAGS_DECL;
9094 if (!r || (--r->refcnt > 0))
9097 ReREFCNT_dec(r->mother_re);
9099 CALLREGFREE_PVT(r); /* free the private data */
9101 SvREFCNT_dec(r->paren_names);
9102 Safefree(r->wrapped);
9105 if (r->anchored_substr)
9106 SvREFCNT_dec(r->anchored_substr);
9107 if (r->anchored_utf8)
9108 SvREFCNT_dec(r->anchored_utf8);
9109 if (r->float_substr)
9110 SvREFCNT_dec(r->float_substr);
9112 SvREFCNT_dec(r->float_utf8);
9113 Safefree(r->substrs);
9115 RX_MATCH_COPY_FREE(r);
9116 #ifdef PERL_OLD_COPY_ON_WRITE
9118 SvREFCNT_dec(r->saved_copy);
9127 This is a hacky workaround to the structural issue of match results
9128 being stored in the regexp structure which is in turn stored in
9129 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9130 could be PL_curpm in multiple contexts, and could require multiple
9131 result sets being associated with the pattern simultaneously, such
9132 as when doing a recursive match with (??{$qr})
9134 The solution is to make a lightweight copy of the regexp structure
9135 when a qr// is returned from the code executed by (??{$qr}) this
9136 lightweight copy doesnt actually own any of its data except for
9137 the starp/end and the actual regexp structure itself.
9143 Perl_reg_temp_copy (pTHX_ struct regexp *r) {
9145 register const I32 npar = r->nparens+1;
9146 (void)ReREFCNT_inc(r);
9147 Newx(ret, 1, regexp);
9148 StructCopy(r, ret, regexp);
9149 Newx(ret->offs, npar, regexp_paren_pair);
9150 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9153 Newx(ret->substrs, 1, struct reg_substr_data);
9154 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9156 SvREFCNT_inc_void(ret->anchored_substr);
9157 SvREFCNT_inc_void(ret->anchored_utf8);
9158 SvREFCNT_inc_void(ret->float_substr);
9159 SvREFCNT_inc_void(ret->float_utf8);
9161 /* check_substr and check_utf8, if non-NULL, point to either their
9162 anchored or float namesakes, and don't hold a second reference. */
9164 RX_MATCH_COPIED_off(ret);
9165 #ifdef PERL_OLD_COPY_ON_WRITE
9166 ret->saved_copy = NULL;
9175 /* regfree_internal()
9177 Free the private data in a regexp. This is overloadable by
9178 extensions. Perl takes care of the regexp structure in pregfree(),
9179 this covers the *pprivate pointer which technically perldoesnt
9180 know about, however of course we have to handle the
9181 regexp_internal structure when no extension is in use.
9183 Note this is called before freeing anything in the regexp
9188 Perl_regfree_internal(pTHX_ REGEXP * const r)
9192 GET_RE_DEBUG_FLAGS_DECL;
9198 SV *dsv= sv_newmortal();
9199 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
9200 dsv, r->precomp, r->prelen, 60);
9201 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9202 PL_colors[4],PL_colors[5],s);
9205 #ifdef RE_TRACK_PATTERN_OFFSETS
9207 Safefree(ri->u.offsets); /* 20010421 MJD */
9210 int n = ri->data->count;
9211 PAD* new_comppad = NULL;
9216 /* If you add a ->what type here, update the comment in regcomp.h */
9217 switch (ri->data->what[n]) {
9221 SvREFCNT_dec((SV*)ri->data->data[n]);
9224 Safefree(ri->data->data[n]);
9227 new_comppad = (AV*)ri->data->data[n];
9230 if (new_comppad == NULL)
9231 Perl_croak(aTHX_ "panic: pregfree comppad");
9232 PAD_SAVE_LOCAL(old_comppad,
9233 /* Watch out for global destruction's random ordering. */
9234 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9237 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9240 op_free((OP_4tree*)ri->data->data[n]);
9242 PAD_RESTORE_LOCAL(old_comppad);
9243 SvREFCNT_dec((SV*)new_comppad);
9249 { /* Aho Corasick add-on structure for a trie node.
9250 Used in stclass optimization only */
9252 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9254 refcount = --aho->refcount;
9257 PerlMemShared_free(aho->states);
9258 PerlMemShared_free(aho->fail);
9259 /* do this last!!!! */
9260 PerlMemShared_free(ri->data->data[n]);
9261 PerlMemShared_free(ri->regstclass);
9267 /* trie structure. */
9269 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9271 refcount = --trie->refcount;
9274 PerlMemShared_free(trie->charmap);
9275 PerlMemShared_free(trie->states);
9276 PerlMemShared_free(trie->trans);
9278 PerlMemShared_free(trie->bitmap);
9280 PerlMemShared_free(trie->wordlen);
9282 PerlMemShared_free(trie->jump);
9284 PerlMemShared_free(trie->nextword);
9285 /* do this last!!!! */
9286 PerlMemShared_free(ri->data->data[n]);
9291 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9294 Safefree(ri->data->what);
9301 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
9302 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9303 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9304 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9307 re_dup - duplicate a regexp.
9309 This routine is expected to clone a given regexp structure. It is not
9310 compiler under USE_ITHREADS.
9312 After all of the core data stored in struct regexp is duplicated
9313 the regexp_engine.dupe method is used to copy any private data
9314 stored in the *pprivate pointer. This allows extensions to handle
9315 any duplication it needs to do.
9317 See pregfree() and regfree_internal() if you change anything here.
9319 #if defined(USE_ITHREADS)
9320 #ifndef PERL_IN_XSUB_RE
9322 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
9329 return (REGEXP *)NULL;
9331 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
9335 npar = r->nparens+1;
9336 Newx(ret, 1, regexp);
9337 StructCopy(r, ret, regexp);
9338 Newx(ret->offs, npar, regexp_paren_pair);
9339 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9341 /* no need to copy these */
9342 Newx(ret->swap, npar, regexp_paren_pair);
9346 /* Do it this way to avoid reading from *r after the StructCopy().
9347 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9348 cache, it doesn't matter. */
9349 const bool anchored = r->check_substr == r->anchored_substr;
9350 Newx(ret->substrs, 1, struct reg_substr_data);
9351 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9353 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9354 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9355 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9356 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9358 /* check_substr and check_utf8, if non-NULL, point to either their
9359 anchored or float namesakes, and don't hold a second reference. */
9361 if (ret->check_substr) {
9363 assert(r->check_utf8 == r->anchored_utf8);
9364 ret->check_substr = ret->anchored_substr;
9365 ret->check_utf8 = ret->anchored_utf8;
9367 assert(r->check_substr == r->float_substr);
9368 assert(r->check_utf8 == r->float_utf8);
9369 ret->check_substr = ret->float_substr;
9370 ret->check_utf8 = ret->float_utf8;
9375 ret->wrapped = SAVEPVN(ret->wrapped, ret->wraplen+1);
9376 ret->precomp = ret->wrapped + (ret->precomp - ret->wrapped);
9377 ret->paren_names = hv_dup_inc(ret->paren_names, param);
9380 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
9382 if (RX_MATCH_COPIED(ret))
9383 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9386 #ifdef PERL_OLD_COPY_ON_WRITE
9387 ret->saved_copy = NULL;
9390 ret->mother_re = NULL;
9392 ret->seen_evals = 0;
9394 ptr_table_store(PL_ptr_table, r, ret);
9397 #endif /* PERL_IN_XSUB_RE */
9402 This is the internal complement to regdupe() which is used to copy
9403 the structure pointed to by the *pprivate pointer in the regexp.
9404 This is the core version of the extension overridable cloning hook.
9405 The regexp structure being duplicated will be copied by perl prior
9406 to this and will be provided as the regexp *r argument, however
9407 with the /old/ structures pprivate pointer value. Thus this routine
9408 may override any copying normally done by perl.
9410 It returns a pointer to the new regexp_internal structure.
9414 Perl_regdupe_internal(pTHX_ REGEXP * const r, CLONE_PARAMS *param)
9417 regexp_internal *reti;
9421 npar = r->nparens+1;
9424 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
9425 Copy(ri->program, reti->program, len+1, regnode);
9428 reti->regstclass = NULL;
9432 const int count = ri->data->count;
9435 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9436 char, struct reg_data);
9437 Newx(d->what, count, U8);
9440 for (i = 0; i < count; i++) {
9441 d->what[i] = ri->data->what[i];
9442 switch (d->what[i]) {
9443 /* legal options are one of: sSfpontTu
9444 see also regcomp.h and pregfree() */
9447 case 'p': /* actually an AV, but the dup function is identical. */
9448 case 'u': /* actually an HV, but the dup function is identical. */
9449 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
9452 /* This is cheating. */
9453 Newx(d->data[i], 1, struct regnode_charclass_class);
9454 StructCopy(ri->data->data[i], d->data[i],
9455 struct regnode_charclass_class);
9456 reti->regstclass = (regnode*)d->data[i];
9459 /* Compiled op trees are readonly and in shared memory,
9460 and can thus be shared without duplication. */
9462 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9466 /* Trie stclasses are readonly and can thus be shared
9467 * without duplication. We free the stclass in pregfree
9468 * when the corresponding reg_ac_data struct is freed.
9470 reti->regstclass= ri->regstclass;
9474 ((reg_trie_data*)ri->data->data[i])->refcount++;
9478 d->data[i] = ri->data->data[i];
9481 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9490 reti->name_list_idx = ri->name_list_idx;
9492 #ifdef RE_TRACK_PATTERN_OFFSETS
9493 if (ri->u.offsets) {
9494 Newx(reti->u.offsets, 2*len+1, U32);
9495 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9498 SetProgLen(reti,len);
9504 #endif /* USE_ITHREADS */
9509 converts a regexp embedded in a MAGIC struct to its stringified form,
9510 caching the converted form in the struct and returns the cached
9513 If lp is nonnull then it is used to return the length of the
9516 If flags is nonnull and the returned string contains UTF8 then
9517 (*flags & 1) will be true.
9519 If haseval is nonnull then it is used to return whether the pattern
9522 Normally called via macro:
9524 CALLREG_STRINGIFY(mg,&len,&utf8);
9528 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9530 See sv_2pv_flags() in sv.c for an example of internal usage.
9533 #ifndef PERL_IN_XSUB_RE
9536 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9538 const regexp * const re = (regexp *)mg->mg_obj;
9540 *haseval = re->seen_evals;
9542 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
9549 - regnext - dig the "next" pointer out of a node
9552 Perl_regnext(pTHX_ register regnode *p)
9555 register I32 offset;
9560 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9569 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9572 STRLEN l1 = strlen(pat1);
9573 STRLEN l2 = strlen(pat2);
9576 const char *message;
9582 Copy(pat1, buf, l1 , char);
9583 Copy(pat2, buf + l1, l2 , char);
9584 buf[l1 + l2] = '\n';
9585 buf[l1 + l2 + 1] = '\0';
9587 /* ANSI variant takes additional second argument */
9588 va_start(args, pat2);
9592 msv = vmess(buf, &args);
9594 message = SvPV_const(msv,l1);
9597 Copy(message, buf, l1 , char);
9598 buf[l1-1] = '\0'; /* Overwrite \n */
9599 Perl_croak(aTHX_ "%s", buf);
9602 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9604 #ifndef PERL_IN_XSUB_RE
9606 Perl_save_re_context(pTHX)
9610 struct re_save_state *state;
9612 SAVEVPTR(PL_curcop);
9613 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9615 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9616 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9617 SSPUSHINT(SAVEt_RE_STATE);
9619 Copy(&PL_reg_state, state, 1, struct re_save_state);
9621 PL_reg_start_tmp = 0;
9622 PL_reg_start_tmpl = 0;
9623 PL_reg_oldsaved = NULL;
9624 PL_reg_oldsavedlen = 0;
9626 PL_reg_leftiter = 0;
9627 PL_reg_poscache = NULL;
9628 PL_reg_poscache_size = 0;
9629 #ifdef PERL_OLD_COPY_ON_WRITE
9633 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9635 const REGEXP * const rx = PM_GETRE(PL_curpm);
9638 for (i = 1; i <= rx->nparens; i++) {
9639 char digits[TYPE_CHARS(long)];
9640 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9641 GV *const *const gvp
9642 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9645 GV * const gv = *gvp;
9646 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9656 clear_re(pTHX_ void *r)
9659 ReREFCNT_dec((regexp *)r);
9665 S_put_byte(pTHX_ SV *sv, int c)
9667 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9668 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9669 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9670 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9672 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9676 #define CLEAR_OPTSTART \
9677 if (optstart) STMT_START { \
9678 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9682 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9684 STATIC const regnode *
9685 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9686 const regnode *last, const regnode *plast,
9687 SV* sv, I32 indent, U32 depth)
9690 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9691 register const regnode *next;
9692 const regnode *optstart= NULL;
9695 GET_RE_DEBUG_FLAGS_DECL;
9697 #ifdef DEBUG_DUMPUNTIL
9698 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9699 last ? last-start : 0,plast ? plast-start : 0);
9702 if (plast && plast < last)
9705 while (PL_regkind[op] != END && (!last || node < last)) {
9706 /* While that wasn't END last time... */
9709 if (op == CLOSE || op == WHILEM)
9711 next = regnext((regnode *)node);
9714 if (OP(node) == OPTIMIZED) {
9715 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9722 regprop(r, sv, node);
9723 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9724 (int)(2*indent + 1), "", SvPVX_const(sv));
9726 if (OP(node) != OPTIMIZED) {
9727 if (next == NULL) /* Next ptr. */
9728 PerlIO_printf(Perl_debug_log, " (0)");
9729 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9730 PerlIO_printf(Perl_debug_log, " (FAIL)");
9732 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9733 (void)PerlIO_putc(Perl_debug_log, '\n');
9737 if (PL_regkind[(U8)op] == BRANCHJ) {
9740 register const regnode *nnode = (OP(next) == LONGJMP
9741 ? regnext((regnode *)next)
9743 if (last && nnode > last)
9745 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9748 else if (PL_regkind[(U8)op] == BRANCH) {
9750 DUMPUNTIL(NEXTOPER(node), next);
9752 else if ( PL_regkind[(U8)op] == TRIE ) {
9753 const regnode *this_trie = node;
9754 const char op = OP(node);
9755 const U32 n = ARG(node);
9756 const reg_ac_data * const ac = op>=AHOCORASICK ?
9757 (reg_ac_data *)ri->data->data[n] :
9759 const reg_trie_data * const trie =
9760 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9762 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9764 const regnode *nextbranch= NULL;
9766 sv_setpvn(sv, "", 0);
9767 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9768 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9770 PerlIO_printf(Perl_debug_log, "%*s%s ",
9771 (int)(2*(indent+3)), "",
9772 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9773 PL_colors[0], PL_colors[1],
9774 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9775 PERL_PV_PRETTY_ELLIPSES |
9781 U16 dist= trie->jump[word_idx+1];
9782 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9783 (UV)((dist ? this_trie + dist : next) - start));
9786 nextbranch= this_trie + trie->jump[0];
9787 DUMPUNTIL(this_trie + dist, nextbranch);
9789 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9790 nextbranch= regnext((regnode *)nextbranch);
9792 PerlIO_printf(Perl_debug_log, "\n");
9795 if (last && next > last)
9800 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9801 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9802 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9804 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9806 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9808 else if ( op == PLUS || op == STAR) {
9809 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9811 else if (op == ANYOF) {
9812 /* arglen 1 + class block */
9813 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9814 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9815 node = NEXTOPER(node);
9817 else if (PL_regkind[(U8)op] == EXACT) {
9818 /* Literal string, where present. */
9819 node += NODE_SZ_STR(node) - 1;
9820 node = NEXTOPER(node);
9823 node = NEXTOPER(node);
9824 node += regarglen[(U8)op];
9826 if (op == CURLYX || op == OPEN)
9830 #ifdef DEBUG_DUMPUNTIL
9831 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9836 #endif /* DEBUGGING */
9840 * c-indentation-style: bsd
9842 * indent-tabs-mode: t
9845 * ex: set ts=8 sts=4 sw=4 noet: