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_sv; /* The SV that is the regexp. */
106 regexp *rx; /* perl core regexp structure */
107 regexp_internal *rxi; /* internal data for regexp object pprivate field */
108 char *start; /* Start of input for compile */
109 char *end; /* End of input for compile */
110 char *parse; /* Input-scan pointer. */
111 I32 whilem_seen; /* number of WHILEM in this expr */
112 regnode *emit_start; /* Start of emitted-code area */
113 regnode *emit_bound; /* First regnode outside of the allocated space */
114 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
115 I32 naughty; /* How bad is this pattern? */
116 I32 sawback; /* Did we see \1, ...? */
118 I32 size; /* Code size. */
119 I32 npar; /* Capture buffer count, (OPEN). */
120 I32 cpar; /* Capture buffer count, (CLOSE). */
121 I32 nestroot; /* root parens we are in - used by accept */
125 regnode **open_parens; /* pointers to open parens */
126 regnode **close_parens; /* pointers to close parens */
127 regnode *opend; /* END node in program */
128 I32 utf8; /* whether the pattern is utf8 or not */
129 I32 orig_utf8; /* whether the pattern was originally in utf8 */
130 /* XXX use this for future optimisation of case
131 * where pattern must be upgraded to utf8. */
132 HV *charnames; /* cache of named sequences */
133 HV *paren_names; /* Paren names */
135 regnode **recurse; /* Recurse regops */
136 I32 recurse_count; /* Number of recurse regops */
138 char *starttry; /* -Dr: where regtry was called. */
139 #define RExC_starttry (pRExC_state->starttry)
142 const char *lastparse;
144 AV *paren_name_list; /* idx -> name */
145 #define RExC_lastparse (pRExC_state->lastparse)
146 #define RExC_lastnum (pRExC_state->lastnum)
147 #define RExC_paren_name_list (pRExC_state->paren_name_list)
151 #define RExC_flags (pRExC_state->flags)
152 #define RExC_precomp (pRExC_state->precomp)
153 #define RExC_rx_sv (pRExC_state->rx_sv)
154 #define RExC_rx (pRExC_state->rx)
155 #define RExC_rxi (pRExC_state->rxi)
156 #define RExC_start (pRExC_state->start)
157 #define RExC_end (pRExC_state->end)
158 #define RExC_parse (pRExC_state->parse)
159 #define RExC_whilem_seen (pRExC_state->whilem_seen)
160 #ifdef RE_TRACK_PATTERN_OFFSETS
161 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
163 #define RExC_emit (pRExC_state->emit)
164 #define RExC_emit_start (pRExC_state->emit_start)
165 #define RExC_emit_bound (pRExC_state->emit_bound)
166 #define RExC_naughty (pRExC_state->naughty)
167 #define RExC_sawback (pRExC_state->sawback)
168 #define RExC_seen (pRExC_state->seen)
169 #define RExC_size (pRExC_state->size)
170 #define RExC_npar (pRExC_state->npar)
171 #define RExC_nestroot (pRExC_state->nestroot)
172 #define RExC_extralen (pRExC_state->extralen)
173 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
174 #define RExC_seen_evals (pRExC_state->seen_evals)
175 #define RExC_utf8 (pRExC_state->utf8)
176 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
177 #define RExC_charnames (pRExC_state->charnames)
178 #define RExC_open_parens (pRExC_state->open_parens)
179 #define RExC_close_parens (pRExC_state->close_parens)
180 #define RExC_opend (pRExC_state->opend)
181 #define RExC_paren_names (pRExC_state->paren_names)
182 #define RExC_recurse (pRExC_state->recurse)
183 #define RExC_recurse_count (pRExC_state->recurse_count)
186 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
187 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
188 ((*s) == '{' && regcurly(s)))
191 #undef SPSTART /* dratted cpp namespace... */
194 * Flags to be passed up and down.
196 #define WORST 0 /* Worst case. */
197 #define HASWIDTH 0x01 /* Known to match non-null strings. */
198 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
199 #define SPSTART 0x04 /* Starts with * or +. */
200 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
201 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
203 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
205 /* whether trie related optimizations are enabled */
206 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
207 #define TRIE_STUDY_OPT
208 #define FULL_TRIE_STUDY
214 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
215 #define PBITVAL(paren) (1 << ((paren) & 7))
216 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
217 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
218 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
221 /* About scan_data_t.
223 During optimisation we recurse through the regexp program performing
224 various inplace (keyhole style) optimisations. In addition study_chunk
225 and scan_commit populate this data structure with information about
226 what strings MUST appear in the pattern. We look for the longest
227 string that must appear for at a fixed location, and we look for the
228 longest string that may appear at a floating location. So for instance
233 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
234 strings (because they follow a .* construct). study_chunk will identify
235 both FOO and BAR as being the longest fixed and floating strings respectively.
237 The strings can be composites, for instance
241 will result in a composite fixed substring 'foo'.
243 For each string some basic information is maintained:
245 - offset or min_offset
246 This is the position the string must appear at, or not before.
247 It also implicitly (when combined with minlenp) tells us how many
248 character must match before the string we are searching.
249 Likewise when combined with minlenp and the length of the string
250 tells us how many characters must appear after the string we have
254 Only used for floating strings. This is the rightmost point that
255 the string can appear at. Ifset to I32 max it indicates that the
256 string can occur infinitely far to the right.
259 A pointer to the minimum length of the pattern that the string
260 was found inside. This is important as in the case of positive
261 lookahead or positive lookbehind we can have multiple patterns
266 The minimum length of the pattern overall is 3, the minimum length
267 of the lookahead part is 3, but the minimum length of the part that
268 will actually match is 1. So 'FOO's minimum length is 3, but the
269 minimum length for the F is 1. This is important as the minimum length
270 is used to determine offsets in front of and behind the string being
271 looked for. Since strings can be composites this is the length of the
272 pattern at the time it was commited with a scan_commit. Note that
273 the length is calculated by study_chunk, so that the minimum lengths
274 are not known until the full pattern has been compiled, thus the
275 pointer to the value.
279 In the case of lookbehind the string being searched for can be
280 offset past the start point of the final matching string.
281 If this value was just blithely removed from the min_offset it would
282 invalidate some of the calculations for how many chars must match
283 before or after (as they are derived from min_offset and minlen and
284 the length of the string being searched for).
285 When the final pattern is compiled and the data is moved from the
286 scan_data_t structure into the regexp structure the information
287 about lookbehind is factored in, with the information that would
288 have been lost precalculated in the end_shift field for the
291 The fields pos_min and pos_delta are used to store the minimum offset
292 and the delta to the maximum offset at the current point in the pattern.
296 typedef struct scan_data_t {
297 /*I32 len_min; unused */
298 /*I32 len_delta; unused */
302 I32 last_end; /* min value, <0 unless valid. */
305 SV **longest; /* Either &l_fixed, or &l_float. */
306 SV *longest_fixed; /* longest fixed string found in pattern */
307 I32 offset_fixed; /* offset where it starts */
308 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
309 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
310 SV *longest_float; /* longest floating string found in pattern */
311 I32 offset_float_min; /* earliest point in string it can appear */
312 I32 offset_float_max; /* latest point in string it can appear */
313 I32 *minlen_float; /* pointer to the minlen relevent to the string */
314 I32 lookbehind_float; /* is the position of the string modified by LB */
318 struct regnode_charclass_class *start_class;
322 * Forward declarations for pregcomp()'s friends.
325 static const scan_data_t zero_scan_data =
326 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
328 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
329 #define SF_BEFORE_SEOL 0x0001
330 #define SF_BEFORE_MEOL 0x0002
331 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
332 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
335 # define SF_FIX_SHIFT_EOL (0+2)
336 # define SF_FL_SHIFT_EOL (0+4)
338 # define SF_FIX_SHIFT_EOL (+2)
339 # define SF_FL_SHIFT_EOL (+4)
342 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
343 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
345 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
346 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
347 #define SF_IS_INF 0x0040
348 #define SF_HAS_PAR 0x0080
349 #define SF_IN_PAR 0x0100
350 #define SF_HAS_EVAL 0x0200
351 #define SCF_DO_SUBSTR 0x0400
352 #define SCF_DO_STCLASS_AND 0x0800
353 #define SCF_DO_STCLASS_OR 0x1000
354 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
355 #define SCF_WHILEM_VISITED_POS 0x2000
357 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
358 #define SCF_SEEN_ACCEPT 0x8000
360 #define UTF (RExC_utf8 != 0)
361 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
362 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
364 #define OOB_UNICODE 12345678
365 #define OOB_NAMEDCLASS -1
367 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
368 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
371 /* length of regex to show in messages that don't mark a position within */
372 #define RegexLengthToShowInErrorMessages 127
375 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
376 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
377 * op/pragma/warn/regcomp.
379 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
380 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
382 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
385 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
386 * arg. Show regex, up to a maximum length. If it's too long, chop and add
389 #define _FAIL(code) STMT_START { \
390 const char *ellipses = ""; \
391 IV len = RExC_end - RExC_precomp; \
394 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
395 if (len > RegexLengthToShowInErrorMessages) { \
396 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
397 len = RegexLengthToShowInErrorMessages - 10; \
403 #define FAIL(msg) _FAIL( \
404 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
405 msg, (int)len, RExC_precomp, ellipses))
407 #define FAIL2(msg,arg) _FAIL( \
408 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
409 arg, (int)len, RExC_precomp, ellipses))
412 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
414 #define Simple_vFAIL(m) STMT_START { \
415 const IV offset = RExC_parse - RExC_precomp; \
416 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
417 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
421 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
423 #define vFAIL(m) STMT_START { \
425 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
430 * Like Simple_vFAIL(), but accepts two arguments.
432 #define Simple_vFAIL2(m,a1) STMT_START { \
433 const IV offset = RExC_parse - RExC_precomp; \
434 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
435 (int)offset, RExC_precomp, RExC_precomp + offset); \
439 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
441 #define vFAIL2(m,a1) STMT_START { \
443 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
444 Simple_vFAIL2(m, a1); \
449 * Like Simple_vFAIL(), but accepts three arguments.
451 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
452 const IV offset = RExC_parse - RExC_precomp; \
453 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
454 (int)offset, RExC_precomp, RExC_precomp + offset); \
458 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
460 #define vFAIL3(m,a1,a2) STMT_START { \
462 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
463 Simple_vFAIL3(m, a1, a2); \
467 * Like Simple_vFAIL(), but accepts four arguments.
469 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
470 const IV offset = RExC_parse - RExC_precomp; \
471 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
472 (int)offset, RExC_precomp, RExC_precomp + offset); \
475 #define vWARN(loc,m) STMT_START { \
476 const IV offset = loc - RExC_precomp; \
477 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
478 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
481 #define vWARNdep(loc,m) STMT_START { \
482 const IV offset = loc - RExC_precomp; \
483 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
484 "%s" REPORT_LOCATION, \
485 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
489 #define vWARN2(loc, m, a1) STMT_START { \
490 const IV offset = loc - RExC_precomp; \
491 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
492 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
495 #define vWARN3(loc, m, a1, a2) STMT_START { \
496 const IV offset = loc - RExC_precomp; \
497 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
498 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
501 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
502 const IV offset = loc - RExC_precomp; \
503 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
504 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
507 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
508 const IV offset = loc - RExC_precomp; \
509 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
510 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
514 /* Allow for side effects in s */
515 #define REGC(c,s) STMT_START { \
516 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
519 /* Macros for recording node offsets. 20001227 mjd@plover.com
520 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
521 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
522 * Element 0 holds the number n.
523 * Position is 1 indexed.
525 #ifndef RE_TRACK_PATTERN_OFFSETS
526 #define Set_Node_Offset_To_R(node,byte)
527 #define Set_Node_Offset(node,byte)
528 #define Set_Cur_Node_Offset
529 #define Set_Node_Length_To_R(node,len)
530 #define Set_Node_Length(node,len)
531 #define Set_Node_Cur_Length(node)
532 #define Node_Offset(n)
533 #define Node_Length(n)
534 #define Set_Node_Offset_Length(node,offset,len)
535 #define ProgLen(ri) ri->u.proglen
536 #define SetProgLen(ri,x) ri->u.proglen = x
538 #define ProgLen(ri) ri->u.offsets[0]
539 #define SetProgLen(ri,x) ri->u.offsets[0] = x
540 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
542 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
543 __LINE__, (int)(node), (int)(byte))); \
545 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
547 RExC_offsets[2*(node)-1] = (byte); \
552 #define Set_Node_Offset(node,byte) \
553 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
554 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
556 #define Set_Node_Length_To_R(node,len) STMT_START { \
558 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
559 __LINE__, (int)(node), (int)(len))); \
561 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
563 RExC_offsets[2*(node)] = (len); \
568 #define Set_Node_Length(node,len) \
569 Set_Node_Length_To_R((node)-RExC_emit_start, len)
570 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
571 #define Set_Node_Cur_Length(node) \
572 Set_Node_Length(node, RExC_parse - parse_start)
574 /* Get offsets and lengths */
575 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
576 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
578 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
579 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
580 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
584 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
585 #define EXPERIMENTAL_INPLACESCAN
586 #endif /*RE_TRACK_PATTERN_OFFSETS*/
588 #define DEBUG_STUDYDATA(str,data,depth) \
589 DEBUG_OPTIMISE_MORE_r(if(data){ \
590 PerlIO_printf(Perl_debug_log, \
591 "%*s" str "Pos:%"IVdf"/%"IVdf \
592 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
593 (int)(depth)*2, "", \
594 (IV)((data)->pos_min), \
595 (IV)((data)->pos_delta), \
596 (UV)((data)->flags), \
597 (IV)((data)->whilem_c), \
598 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
599 is_inf ? "INF " : "" \
601 if ((data)->last_found) \
602 PerlIO_printf(Perl_debug_log, \
603 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
604 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
605 SvPVX_const((data)->last_found), \
606 (IV)((data)->last_end), \
607 (IV)((data)->last_start_min), \
608 (IV)((data)->last_start_max), \
609 ((data)->longest && \
610 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
611 SvPVX_const((data)->longest_fixed), \
612 (IV)((data)->offset_fixed), \
613 ((data)->longest && \
614 (data)->longest==&((data)->longest_float)) ? "*" : "", \
615 SvPVX_const((data)->longest_float), \
616 (IV)((data)->offset_float_min), \
617 (IV)((data)->offset_float_max) \
619 PerlIO_printf(Perl_debug_log,"\n"); \
622 static void clear_re(pTHX_ void *r);
624 /* Mark that we cannot extend a found fixed substring at this point.
625 Update the longest found anchored substring and the longest found
626 floating substrings if needed. */
629 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
631 const STRLEN l = CHR_SVLEN(data->last_found);
632 const STRLEN old_l = CHR_SVLEN(*data->longest);
633 GET_RE_DEBUG_FLAGS_DECL;
635 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
636 SvSetMagicSV(*data->longest, data->last_found);
637 if (*data->longest == data->longest_fixed) {
638 data->offset_fixed = l ? data->last_start_min : data->pos_min;
639 if (data->flags & SF_BEFORE_EOL)
641 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
643 data->flags &= ~SF_FIX_BEFORE_EOL;
644 data->minlen_fixed=minlenp;
645 data->lookbehind_fixed=0;
647 else { /* *data->longest == data->longest_float */
648 data->offset_float_min = l ? data->last_start_min : data->pos_min;
649 data->offset_float_max = (l
650 ? data->last_start_max
651 : data->pos_min + data->pos_delta);
652 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
653 data->offset_float_max = I32_MAX;
654 if (data->flags & SF_BEFORE_EOL)
656 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
658 data->flags &= ~SF_FL_BEFORE_EOL;
659 data->minlen_float=minlenp;
660 data->lookbehind_float=0;
663 SvCUR_set(data->last_found, 0);
665 SV * const sv = data->last_found;
666 if (SvUTF8(sv) && SvMAGICAL(sv)) {
667 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
673 data->flags &= ~SF_BEFORE_EOL;
674 DEBUG_STUDYDATA("commit: ",data,0);
677 /* Can match anything (initialization) */
679 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
681 ANYOF_CLASS_ZERO(cl);
682 ANYOF_BITMAP_SETALL(cl);
683 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
685 cl->flags |= ANYOF_LOCALE;
688 /* Can match anything (initialization) */
690 S_cl_is_anything(const struct regnode_charclass_class *cl)
694 for (value = 0; value <= ANYOF_MAX; value += 2)
695 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
697 if (!(cl->flags & ANYOF_UNICODE_ALL))
699 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
704 /* Can match anything (initialization) */
706 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
708 Zero(cl, 1, struct regnode_charclass_class);
710 cl_anything(pRExC_state, cl);
714 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
716 Zero(cl, 1, struct regnode_charclass_class);
718 cl_anything(pRExC_state, cl);
720 cl->flags |= ANYOF_LOCALE;
723 /* 'And' a given class with another one. Can create false positives */
724 /* We assume that cl is not inverted */
726 S_cl_and(struct regnode_charclass_class *cl,
727 const struct regnode_charclass_class *and_with)
730 assert(and_with->type == ANYOF);
731 if (!(and_with->flags & ANYOF_CLASS)
732 && !(cl->flags & ANYOF_CLASS)
733 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
734 && !(and_with->flags & ANYOF_FOLD)
735 && !(cl->flags & ANYOF_FOLD)) {
738 if (and_with->flags & ANYOF_INVERT)
739 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
740 cl->bitmap[i] &= ~and_with->bitmap[i];
742 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
743 cl->bitmap[i] &= and_with->bitmap[i];
744 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
745 if (!(and_with->flags & ANYOF_EOS))
746 cl->flags &= ~ANYOF_EOS;
748 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
749 !(and_with->flags & ANYOF_INVERT)) {
750 cl->flags &= ~ANYOF_UNICODE_ALL;
751 cl->flags |= ANYOF_UNICODE;
752 ARG_SET(cl, ARG(and_with));
754 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
755 !(and_with->flags & ANYOF_INVERT))
756 cl->flags &= ~ANYOF_UNICODE_ALL;
757 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
758 !(and_with->flags & ANYOF_INVERT))
759 cl->flags &= ~ANYOF_UNICODE;
762 /* 'OR' a given class with another one. Can create false positives */
763 /* We assume that cl is not inverted */
765 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
767 if (or_with->flags & ANYOF_INVERT) {
769 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
770 * <= (B1 | !B2) | (CL1 | !CL2)
771 * which is wasteful if CL2 is small, but we ignore CL2:
772 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
773 * XXXX Can we handle case-fold? Unclear:
774 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
775 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
777 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
778 && !(or_with->flags & ANYOF_FOLD)
779 && !(cl->flags & ANYOF_FOLD) ) {
782 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
783 cl->bitmap[i] |= ~or_with->bitmap[i];
784 } /* XXXX: logic is complicated otherwise */
786 cl_anything(pRExC_state, cl);
789 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
790 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
791 && (!(or_with->flags & ANYOF_FOLD)
792 || (cl->flags & ANYOF_FOLD)) ) {
795 /* OR char bitmap and class bitmap separately */
796 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
797 cl->bitmap[i] |= or_with->bitmap[i];
798 if (or_with->flags & ANYOF_CLASS) {
799 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
800 cl->classflags[i] |= or_with->classflags[i];
801 cl->flags |= ANYOF_CLASS;
804 else { /* XXXX: logic is complicated, leave it along for a moment. */
805 cl_anything(pRExC_state, cl);
808 if (or_with->flags & ANYOF_EOS)
809 cl->flags |= ANYOF_EOS;
811 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
812 ARG(cl) != ARG(or_with)) {
813 cl->flags |= ANYOF_UNICODE_ALL;
814 cl->flags &= ~ANYOF_UNICODE;
816 if (or_with->flags & ANYOF_UNICODE_ALL) {
817 cl->flags |= ANYOF_UNICODE_ALL;
818 cl->flags &= ~ANYOF_UNICODE;
822 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
823 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
824 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
825 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
830 dump_trie(trie,widecharmap,revcharmap)
831 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
832 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
834 These routines dump out a trie in a somewhat readable format.
835 The _interim_ variants are used for debugging the interim
836 tables that are used to generate the final compressed
837 representation which is what dump_trie expects.
839 Part of the reason for their existance is to provide a form
840 of documentation as to how the different representations function.
845 Dumps the final compressed table form of the trie to Perl_debug_log.
846 Used for debugging make_trie().
850 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
851 AV *revcharmap, U32 depth)
854 SV *sv=sv_newmortal();
855 int colwidth= widecharmap ? 6 : 4;
856 GET_RE_DEBUG_FLAGS_DECL;
859 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
860 (int)depth * 2 + 2,"",
861 "Match","Base","Ofs" );
863 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
864 SV ** const tmp = av_fetch( revcharmap, state, 0);
866 PerlIO_printf( Perl_debug_log, "%*s",
868 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
869 PL_colors[0], PL_colors[1],
870 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
871 PERL_PV_ESCAPE_FIRSTCHAR
876 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
877 (int)depth * 2 + 2,"");
879 for( state = 0 ; state < trie->uniquecharcount ; state++ )
880 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
881 PerlIO_printf( Perl_debug_log, "\n");
883 for( state = 1 ; state < trie->statecount ; state++ ) {
884 const U32 base = trie->states[ state ].trans.base;
886 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
888 if ( trie->states[ state ].wordnum ) {
889 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
891 PerlIO_printf( Perl_debug_log, "%6s", "" );
894 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
899 while( ( base + ofs < trie->uniquecharcount ) ||
900 ( base + ofs - trie->uniquecharcount < trie->lasttrans
901 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
904 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
906 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
907 if ( ( base + ofs >= trie->uniquecharcount ) &&
908 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
909 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
911 PerlIO_printf( Perl_debug_log, "%*"UVXf,
913 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
915 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
919 PerlIO_printf( Perl_debug_log, "]");
922 PerlIO_printf( Perl_debug_log, "\n" );
926 Dumps a fully constructed but uncompressed trie in list form.
927 List tries normally only are used for construction when the number of
928 possible chars (trie->uniquecharcount) is very high.
929 Used for debugging make_trie().
932 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
933 HV *widecharmap, AV *revcharmap, U32 next_alloc,
937 SV *sv=sv_newmortal();
938 int colwidth= widecharmap ? 6 : 4;
939 GET_RE_DEBUG_FLAGS_DECL;
940 /* print out the table precompression. */
941 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
942 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
943 "------:-----+-----------------\n" );
945 for( state=1 ; state < next_alloc ; state ++ ) {
948 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
949 (int)depth * 2 + 2,"", (UV)state );
950 if ( ! trie->states[ state ].wordnum ) {
951 PerlIO_printf( Perl_debug_log, "%5s| ","");
953 PerlIO_printf( Perl_debug_log, "W%4x| ",
954 trie->states[ state ].wordnum
957 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
958 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
960 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
962 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
963 PL_colors[0], PL_colors[1],
964 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
965 PERL_PV_ESCAPE_FIRSTCHAR
967 TRIE_LIST_ITEM(state,charid).forid,
968 (UV)TRIE_LIST_ITEM(state,charid).newstate
971 PerlIO_printf(Perl_debug_log, "\n%*s| ",
972 (int)((depth * 2) + 14), "");
975 PerlIO_printf( Perl_debug_log, "\n");
980 Dumps a fully constructed but uncompressed trie in table form.
981 This is the normal DFA style state transition table, with a few
982 twists to facilitate compression later.
983 Used for debugging make_trie().
986 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
987 HV *widecharmap, AV *revcharmap, U32 next_alloc,
992 SV *sv=sv_newmortal();
993 int colwidth= widecharmap ? 6 : 4;
994 GET_RE_DEBUG_FLAGS_DECL;
997 print out the table precompression so that we can do a visual check
998 that they are identical.
1001 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1003 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1004 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1006 PerlIO_printf( Perl_debug_log, "%*s",
1008 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1009 PL_colors[0], PL_colors[1],
1010 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1011 PERL_PV_ESCAPE_FIRSTCHAR
1017 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1019 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1020 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1023 PerlIO_printf( Perl_debug_log, "\n" );
1025 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1027 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1028 (int)depth * 2 + 2,"",
1029 (UV)TRIE_NODENUM( state ) );
1031 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1032 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1034 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1036 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1038 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1039 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1041 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1042 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1049 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1050 startbranch: the first branch in the whole branch sequence
1051 first : start branch of sequence of branch-exact nodes.
1052 May be the same as startbranch
1053 last : Thing following the last branch.
1054 May be the same as tail.
1055 tail : item following the branch sequence
1056 count : words in the sequence
1057 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1058 depth : indent depth
1060 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1062 A trie is an N'ary tree where the branches are determined by digital
1063 decomposition of the key. IE, at the root node you look up the 1st character and
1064 follow that branch repeat until you find the end of the branches. Nodes can be
1065 marked as "accepting" meaning they represent a complete word. Eg:
1069 would convert into the following structure. Numbers represent states, letters
1070 following numbers represent valid transitions on the letter from that state, if
1071 the number is in square brackets it represents an accepting state, otherwise it
1072 will be in parenthesis.
1074 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1078 (1) +-i->(6)-+-s->[7]
1080 +-s->(3)-+-h->(4)-+-e->[5]
1082 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1084 This shows that when matching against the string 'hers' we will begin at state 1
1085 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1086 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1087 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1088 single traverse. We store a mapping from accepting to state to which word was
1089 matched, and then when we have multiple possibilities we try to complete the
1090 rest of the regex in the order in which they occured in the alternation.
1092 The only prior NFA like behaviour that would be changed by the TRIE support is
1093 the silent ignoring of duplicate alternations which are of the form:
1095 / (DUPE|DUPE) X? (?{ ... }) Y /x
1097 Thus EVAL blocks follwing a trie may be called a different number of times with
1098 and without the optimisation. With the optimisations dupes will be silently
1099 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1100 the following demonstrates:
1102 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1104 which prints out 'word' three times, but
1106 'words'=~/(word|word|word)(?{ print $1 })S/
1108 which doesnt print it out at all. This is due to other optimisations kicking in.
1110 Example of what happens on a structural level:
1112 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1114 1: CURLYM[1] {1,32767}(18)
1125 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1126 and should turn into:
1128 1: CURLYM[1] {1,32767}(18)
1130 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1138 Cases where tail != last would be like /(?foo|bar)baz/:
1148 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1149 and would end up looking like:
1152 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1159 d = uvuni_to_utf8_flags(d, uv, 0);
1161 is the recommended Unicode-aware way of saying
1166 #define TRIE_STORE_REVCHAR \
1169 SV *zlopp = newSV(2); \
1170 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1171 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1172 SvCUR_set(zlopp, kapow - flrbbbbb); \
1175 av_push(revcharmap, zlopp); \
1177 char ooooff = (char)uvc; \
1178 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1182 #define TRIE_READ_CHAR STMT_START { \
1186 if ( foldlen > 0 ) { \
1187 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1192 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1193 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1194 foldlen -= UNISKIP( uvc ); \
1195 scan = foldbuf + UNISKIP( uvc ); \
1198 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1208 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1209 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1210 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1211 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1213 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1214 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1215 TRIE_LIST_CUR( state )++; \
1218 #define TRIE_LIST_NEW(state) STMT_START { \
1219 Newxz( trie->states[ state ].trans.list, \
1220 4, reg_trie_trans_le ); \
1221 TRIE_LIST_CUR( state ) = 1; \
1222 TRIE_LIST_LEN( state ) = 4; \
1225 #define TRIE_HANDLE_WORD(state) STMT_START { \
1226 U16 dupe= trie->states[ state ].wordnum; \
1227 regnode * const noper_next = regnext( noper ); \
1229 if (trie->wordlen) \
1230 trie->wordlen[ curword ] = wordlen; \
1232 /* store the word for dumping */ \
1234 if (OP(noper) != NOTHING) \
1235 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1237 tmp = newSVpvn_utf8( "", 0, UTF ); \
1238 av_push( trie_words, tmp ); \
1243 if ( noper_next < tail ) { \
1245 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1246 trie->jump[curword] = (U16)(noper_next - convert); \
1248 jumper = noper_next; \
1250 nextbranch= regnext(cur); \
1254 /* So it's a dupe. This means we need to maintain a */\
1255 /* linked-list from the first to the next. */\
1256 /* we only allocate the nextword buffer when there */\
1257 /* a dupe, so first time we have to do the allocation */\
1258 if (!trie->nextword) \
1259 trie->nextword = (U16 *) \
1260 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1261 while ( trie->nextword[dupe] ) \
1262 dupe= trie->nextword[dupe]; \
1263 trie->nextword[dupe]= curword; \
1265 /* we haven't inserted this word yet. */ \
1266 trie->states[ state ].wordnum = curword; \
1271 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1272 ( ( base + charid >= ucharcount \
1273 && base + charid < ubound \
1274 && state == trie->trans[ base - ucharcount + charid ].check \
1275 && trie->trans[ base - ucharcount + charid ].next ) \
1276 ? trie->trans[ base - ucharcount + charid ].next \
1277 : ( state==1 ? special : 0 ) \
1281 #define MADE_JUMP_TRIE 2
1282 #define MADE_EXACT_TRIE 4
1285 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1288 /* first pass, loop through and scan words */
1289 reg_trie_data *trie;
1290 HV *widecharmap = NULL;
1291 AV *revcharmap = newAV();
1293 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1298 regnode *jumper = NULL;
1299 regnode *nextbranch = NULL;
1300 regnode *convert = NULL;
1301 /* we just use folder as a flag in utf8 */
1302 const U8 * const folder = ( flags == EXACTF
1304 : ( flags == EXACTFL
1311 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1312 AV *trie_words = NULL;
1313 /* along with revcharmap, this only used during construction but both are
1314 * useful during debugging so we store them in the struct when debugging.
1317 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1318 STRLEN trie_charcount=0;
1320 SV *re_trie_maxbuff;
1321 GET_RE_DEBUG_FLAGS_DECL;
1323 PERL_UNUSED_ARG(depth);
1326 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1328 trie->startstate = 1;
1329 trie->wordcount = word_count;
1330 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1331 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1332 if (!(UTF && folder))
1333 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1335 trie_words = newAV();
1338 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1339 if (!SvIOK(re_trie_maxbuff)) {
1340 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1343 PerlIO_printf( Perl_debug_log,
1344 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1345 (int)depth * 2 + 2, "",
1346 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1347 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1351 /* Find the node we are going to overwrite */
1352 if ( first == startbranch && OP( last ) != BRANCH ) {
1353 /* whole branch chain */
1356 /* branch sub-chain */
1357 convert = NEXTOPER( first );
1360 /* -- First loop and Setup --
1362 We first traverse the branches and scan each word to determine if it
1363 contains widechars, and how many unique chars there are, this is
1364 important as we have to build a table with at least as many columns as we
1367 We use an array of integers to represent the character codes 0..255
1368 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1369 native representation of the character value as the key and IV's for the
1372 *TODO* If we keep track of how many times each character is used we can
1373 remap the columns so that the table compression later on is more
1374 efficient in terms of memory by ensuring most common value is in the
1375 middle and the least common are on the outside. IMO this would be better
1376 than a most to least common mapping as theres a decent chance the most
1377 common letter will share a node with the least common, meaning the node
1378 will not be compressable. With a middle is most common approach the worst
1379 case is when we have the least common nodes twice.
1383 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1384 regnode * const noper = NEXTOPER( cur );
1385 const U8 *uc = (U8*)STRING( noper );
1386 const U8 * const e = uc + STR_LEN( noper );
1388 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1389 const U8 *scan = (U8*)NULL;
1390 U32 wordlen = 0; /* required init */
1392 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1394 if (OP(noper) == NOTHING) {
1398 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1399 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1400 regardless of encoding */
1402 for ( ; uc < e ; uc += len ) {
1403 TRIE_CHARCOUNT(trie)++;
1407 if ( !trie->charmap[ uvc ] ) {
1408 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1410 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1414 /* store the codepoint in the bitmap, and if its ascii
1415 also store its folded equivelent. */
1416 TRIE_BITMAP_SET(trie,uvc);
1418 /* store the folded codepoint */
1419 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1422 /* store first byte of utf8 representation of
1423 codepoints in the 127 < uvc < 256 range */
1424 if (127 < uvc && uvc < 192) {
1425 TRIE_BITMAP_SET(trie,194);
1426 } else if (191 < uvc ) {
1427 TRIE_BITMAP_SET(trie,195);
1428 /* && uvc < 256 -- we know uvc is < 256 already */
1431 set_bit = 0; /* We've done our bit :-) */
1436 widecharmap = newHV();
1438 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1441 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1443 if ( !SvTRUE( *svpp ) ) {
1444 sv_setiv( *svpp, ++trie->uniquecharcount );
1449 if( cur == first ) {
1452 } else if (chars < trie->minlen) {
1454 } else if (chars > trie->maxlen) {
1458 } /* end first pass */
1459 DEBUG_TRIE_COMPILE_r(
1460 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1461 (int)depth * 2 + 2,"",
1462 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1463 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1464 (int)trie->minlen, (int)trie->maxlen )
1466 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1469 We now know what we are dealing with in terms of unique chars and
1470 string sizes so we can calculate how much memory a naive
1471 representation using a flat table will take. If it's over a reasonable
1472 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1473 conservative but potentially much slower representation using an array
1476 At the end we convert both representations into the same compressed
1477 form that will be used in regexec.c for matching with. The latter
1478 is a form that cannot be used to construct with but has memory
1479 properties similar to the list form and access properties similar
1480 to the table form making it both suitable for fast searches and
1481 small enough that its feasable to store for the duration of a program.
1483 See the comment in the code where the compressed table is produced
1484 inplace from the flat tabe representation for an explanation of how
1485 the compression works.
1490 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1492 Second Pass -- Array Of Lists Representation
1494 Each state will be represented by a list of charid:state records
1495 (reg_trie_trans_le) the first such element holds the CUR and LEN
1496 points of the allocated array. (See defines above).
1498 We build the initial structure using the lists, and then convert
1499 it into the compressed table form which allows faster lookups
1500 (but cant be modified once converted).
1503 STRLEN transcount = 1;
1505 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1506 "%*sCompiling trie using list compiler\n",
1507 (int)depth * 2 + 2, ""));
1509 trie->states = (reg_trie_state *)
1510 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1511 sizeof(reg_trie_state) );
1515 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1517 regnode * const noper = NEXTOPER( cur );
1518 U8 *uc = (U8*)STRING( noper );
1519 const U8 * const e = uc + STR_LEN( noper );
1520 U32 state = 1; /* required init */
1521 U16 charid = 0; /* sanity init */
1522 U8 *scan = (U8*)NULL; /* sanity init */
1523 STRLEN foldlen = 0; /* required init */
1524 U32 wordlen = 0; /* required init */
1525 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1527 if (OP(noper) != NOTHING) {
1528 for ( ; uc < e ; uc += len ) {
1533 charid = trie->charmap[ uvc ];
1535 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1539 charid=(U16)SvIV( *svpp );
1542 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1549 if ( !trie->states[ state ].trans.list ) {
1550 TRIE_LIST_NEW( state );
1552 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1553 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1554 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1559 newstate = next_alloc++;
1560 TRIE_LIST_PUSH( state, charid, newstate );
1565 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1569 TRIE_HANDLE_WORD(state);
1571 } /* end second pass */
1573 /* next alloc is the NEXT state to be allocated */
1574 trie->statecount = next_alloc;
1575 trie->states = (reg_trie_state *)
1576 PerlMemShared_realloc( trie->states,
1578 * sizeof(reg_trie_state) );
1580 /* and now dump it out before we compress it */
1581 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1582 revcharmap, next_alloc,
1586 trie->trans = (reg_trie_trans *)
1587 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1594 for( state=1 ; state < next_alloc ; state ++ ) {
1598 DEBUG_TRIE_COMPILE_MORE_r(
1599 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1603 if (trie->states[state].trans.list) {
1604 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1608 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1609 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1610 if ( forid < minid ) {
1612 } else if ( forid > maxid ) {
1616 if ( transcount < tp + maxid - minid + 1) {
1618 trie->trans = (reg_trie_trans *)
1619 PerlMemShared_realloc( trie->trans,
1621 * sizeof(reg_trie_trans) );
1622 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1624 base = trie->uniquecharcount + tp - minid;
1625 if ( maxid == minid ) {
1627 for ( ; zp < tp ; zp++ ) {
1628 if ( ! trie->trans[ zp ].next ) {
1629 base = trie->uniquecharcount + zp - minid;
1630 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1631 trie->trans[ zp ].check = state;
1637 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1638 trie->trans[ tp ].check = state;
1643 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1644 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1645 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1646 trie->trans[ tid ].check = state;
1648 tp += ( maxid - minid + 1 );
1650 Safefree(trie->states[ state ].trans.list);
1653 DEBUG_TRIE_COMPILE_MORE_r(
1654 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1657 trie->states[ state ].trans.base=base;
1659 trie->lasttrans = tp + 1;
1663 Second Pass -- Flat Table Representation.
1665 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1666 We know that we will need Charcount+1 trans at most to store the data
1667 (one row per char at worst case) So we preallocate both structures
1668 assuming worst case.
1670 We then construct the trie using only the .next slots of the entry
1673 We use the .check field of the first entry of the node temporarily to
1674 make compression both faster and easier by keeping track of how many non
1675 zero fields are in the node.
1677 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1680 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1681 number representing the first entry of the node, and state as a
1682 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1683 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1684 are 2 entrys per node. eg:
1692 The table is internally in the right hand, idx form. However as we also
1693 have to deal with the states array which is indexed by nodenum we have to
1694 use TRIE_NODENUM() to convert.
1697 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1698 "%*sCompiling trie using table compiler\n",
1699 (int)depth * 2 + 2, ""));
1701 trie->trans = (reg_trie_trans *)
1702 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1703 * trie->uniquecharcount + 1,
1704 sizeof(reg_trie_trans) );
1705 trie->states = (reg_trie_state *)
1706 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1707 sizeof(reg_trie_state) );
1708 next_alloc = trie->uniquecharcount + 1;
1711 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1713 regnode * const noper = NEXTOPER( cur );
1714 const U8 *uc = (U8*)STRING( noper );
1715 const U8 * const e = uc + STR_LEN( noper );
1717 U32 state = 1; /* required init */
1719 U16 charid = 0; /* sanity init */
1720 U32 accept_state = 0; /* sanity init */
1721 U8 *scan = (U8*)NULL; /* sanity init */
1723 STRLEN foldlen = 0; /* required init */
1724 U32 wordlen = 0; /* required init */
1725 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1727 if ( OP(noper) != NOTHING ) {
1728 for ( ; uc < e ; uc += len ) {
1733 charid = trie->charmap[ uvc ];
1735 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1736 charid = svpp ? (U16)SvIV(*svpp) : 0;
1740 if ( !trie->trans[ state + charid ].next ) {
1741 trie->trans[ state + charid ].next = next_alloc;
1742 trie->trans[ state ].check++;
1743 next_alloc += trie->uniquecharcount;
1745 state = trie->trans[ state + charid ].next;
1747 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1749 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1752 accept_state = TRIE_NODENUM( state );
1753 TRIE_HANDLE_WORD(accept_state);
1755 } /* end second pass */
1757 /* and now dump it out before we compress it */
1758 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1760 next_alloc, depth+1));
1764 * Inplace compress the table.*
1766 For sparse data sets the table constructed by the trie algorithm will
1767 be mostly 0/FAIL transitions or to put it another way mostly empty.
1768 (Note that leaf nodes will not contain any transitions.)
1770 This algorithm compresses the tables by eliminating most such
1771 transitions, at the cost of a modest bit of extra work during lookup:
1773 - Each states[] entry contains a .base field which indicates the
1774 index in the state[] array wheres its transition data is stored.
1776 - If .base is 0 there are no valid transitions from that node.
1778 - If .base is nonzero then charid is added to it to find an entry in
1781 -If trans[states[state].base+charid].check!=state then the
1782 transition is taken to be a 0/Fail transition. Thus if there are fail
1783 transitions at the front of the node then the .base offset will point
1784 somewhere inside the previous nodes data (or maybe even into a node
1785 even earlier), but the .check field determines if the transition is
1789 The following process inplace converts the table to the compressed
1790 table: We first do not compress the root node 1,and mark its all its
1791 .check pointers as 1 and set its .base pointer as 1 as well. This
1792 allows to do a DFA construction from the compressed table later, and
1793 ensures that any .base pointers we calculate later are greater than
1796 - We set 'pos' to indicate the first entry of the second node.
1798 - We then iterate over the columns of the node, finding the first and
1799 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1800 and set the .check pointers accordingly, and advance pos
1801 appropriately and repreat for the next node. Note that when we copy
1802 the next pointers we have to convert them from the original
1803 NODEIDX form to NODENUM form as the former is not valid post
1806 - If a node has no transitions used we mark its base as 0 and do not
1807 advance the pos pointer.
1809 - If a node only has one transition we use a second pointer into the
1810 structure to fill in allocated fail transitions from other states.
1811 This pointer is independent of the main pointer and scans forward
1812 looking for null transitions that are allocated to a state. When it
1813 finds one it writes the single transition into the "hole". If the
1814 pointer doesnt find one the single transition is appended as normal.
1816 - Once compressed we can Renew/realloc the structures to release the
1819 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1820 specifically Fig 3.47 and the associated pseudocode.
1824 const U32 laststate = TRIE_NODENUM( next_alloc );
1827 trie->statecount = laststate;
1829 for ( state = 1 ; state < laststate ; state++ ) {
1831 const U32 stateidx = TRIE_NODEIDX( state );
1832 const U32 o_used = trie->trans[ stateidx ].check;
1833 U32 used = trie->trans[ stateidx ].check;
1834 trie->trans[ stateidx ].check = 0;
1836 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1837 if ( flag || trie->trans[ stateidx + charid ].next ) {
1838 if ( trie->trans[ stateidx + charid ].next ) {
1840 for ( ; zp < pos ; zp++ ) {
1841 if ( ! trie->trans[ zp ].next ) {
1845 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1846 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1847 trie->trans[ zp ].check = state;
1848 if ( ++zp > pos ) pos = zp;
1855 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1857 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1858 trie->trans[ pos ].check = state;
1863 trie->lasttrans = pos + 1;
1864 trie->states = (reg_trie_state *)
1865 PerlMemShared_realloc( trie->states, laststate
1866 * sizeof(reg_trie_state) );
1867 DEBUG_TRIE_COMPILE_MORE_r(
1868 PerlIO_printf( Perl_debug_log,
1869 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1870 (int)depth * 2 + 2,"",
1871 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1874 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1877 } /* end table compress */
1879 DEBUG_TRIE_COMPILE_MORE_r(
1880 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1881 (int)depth * 2 + 2, "",
1882 (UV)trie->statecount,
1883 (UV)trie->lasttrans)
1885 /* resize the trans array to remove unused space */
1886 trie->trans = (reg_trie_trans *)
1887 PerlMemShared_realloc( trie->trans, trie->lasttrans
1888 * sizeof(reg_trie_trans) );
1890 /* and now dump out the compressed format */
1891 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1893 { /* Modify the program and insert the new TRIE node*/
1894 U8 nodetype =(U8)(flags & 0xFF);
1898 regnode *optimize = NULL;
1899 #ifdef RE_TRACK_PATTERN_OFFSETS
1902 U32 mjd_nodelen = 0;
1903 #endif /* RE_TRACK_PATTERN_OFFSETS */
1904 #endif /* DEBUGGING */
1906 This means we convert either the first branch or the first Exact,
1907 depending on whether the thing following (in 'last') is a branch
1908 or not and whther first is the startbranch (ie is it a sub part of
1909 the alternation or is it the whole thing.)
1910 Assuming its a sub part we conver the EXACT otherwise we convert
1911 the whole branch sequence, including the first.
1913 /* Find the node we are going to overwrite */
1914 if ( first != startbranch || OP( last ) == BRANCH ) {
1915 /* branch sub-chain */
1916 NEXT_OFF( first ) = (U16)(last - first);
1917 #ifdef RE_TRACK_PATTERN_OFFSETS
1919 mjd_offset= Node_Offset((convert));
1920 mjd_nodelen= Node_Length((convert));
1923 /* whole branch chain */
1925 #ifdef RE_TRACK_PATTERN_OFFSETS
1928 const regnode *nop = NEXTOPER( convert );
1929 mjd_offset= Node_Offset((nop));
1930 mjd_nodelen= Node_Length((nop));
1934 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1935 (int)depth * 2 + 2, "",
1936 (UV)mjd_offset, (UV)mjd_nodelen)
1939 /* But first we check to see if there is a common prefix we can
1940 split out as an EXACT and put in front of the TRIE node. */
1941 trie->startstate= 1;
1942 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1944 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1948 const U32 base = trie->states[ state ].trans.base;
1950 if ( trie->states[state].wordnum )
1953 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1954 if ( ( base + ofs >= trie->uniquecharcount ) &&
1955 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1956 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1958 if ( ++count > 1 ) {
1959 SV **tmp = av_fetch( revcharmap, ofs, 0);
1960 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1961 if ( state == 1 ) break;
1963 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1965 PerlIO_printf(Perl_debug_log,
1966 "%*sNew Start State=%"UVuf" Class: [",
1967 (int)depth * 2 + 2, "",
1970 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1971 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1973 TRIE_BITMAP_SET(trie,*ch);
1975 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1977 PerlIO_printf(Perl_debug_log, (char*)ch)
1981 TRIE_BITMAP_SET(trie,*ch);
1983 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1984 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1990 SV **tmp = av_fetch( revcharmap, idx, 0);
1992 char *ch = SvPV( *tmp, len );
1994 SV *sv=sv_newmortal();
1995 PerlIO_printf( Perl_debug_log,
1996 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1997 (int)depth * 2 + 2, "",
1999 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2000 PL_colors[0], PL_colors[1],
2001 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2002 PERL_PV_ESCAPE_FIRSTCHAR
2007 OP( convert ) = nodetype;
2008 str=STRING(convert);
2011 STR_LEN(convert) += len;
2017 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2023 regnode *n = convert+NODE_SZ_STR(convert);
2024 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2025 trie->startstate = state;
2026 trie->minlen -= (state - 1);
2027 trie->maxlen -= (state - 1);
2029 /* At least the UNICOS C compiler choked on this
2030 * being argument to DEBUG_r(), so let's just have
2033 #ifdef PERL_EXT_RE_BUILD
2039 regnode *fix = convert;
2040 U32 word = trie->wordcount;
2042 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2043 while( ++fix < n ) {
2044 Set_Node_Offset_Length(fix, 0, 0);
2047 SV ** const tmp = av_fetch( trie_words, word, 0 );
2049 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2050 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2052 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2060 NEXT_OFF(convert) = (U16)(tail - convert);
2061 DEBUG_r(optimize= n);
2067 if ( trie->maxlen ) {
2068 NEXT_OFF( convert ) = (U16)(tail - convert);
2069 ARG_SET( convert, data_slot );
2070 /* Store the offset to the first unabsorbed branch in
2071 jump[0], which is otherwise unused by the jump logic.
2072 We use this when dumping a trie and during optimisation. */
2074 trie->jump[0] = (U16)(nextbranch - convert);
2077 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2078 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2080 OP( convert ) = TRIEC;
2081 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2082 PerlMemShared_free(trie->bitmap);
2085 OP( convert ) = TRIE;
2087 /* store the type in the flags */
2088 convert->flags = nodetype;
2092 + regarglen[ OP( convert ) ];
2094 /* XXX We really should free up the resource in trie now,
2095 as we won't use them - (which resources?) dmq */
2097 /* needed for dumping*/
2098 DEBUG_r(if (optimize) {
2099 regnode *opt = convert;
2101 while ( ++opt < optimize) {
2102 Set_Node_Offset_Length(opt,0,0);
2105 Try to clean up some of the debris left after the
2108 while( optimize < jumper ) {
2109 mjd_nodelen += Node_Length((optimize));
2110 OP( optimize ) = OPTIMIZED;
2111 Set_Node_Offset_Length(optimize,0,0);
2114 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2116 } /* end node insert */
2117 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2119 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2120 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2122 SvREFCNT_dec(revcharmap);
2126 : trie->startstate>1
2132 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2134 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2136 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2137 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2140 We find the fail state for each state in the trie, this state is the longest proper
2141 suffix of the current states 'word' that is also a proper prefix of another word in our
2142 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2143 the DFA not to have to restart after its tried and failed a word at a given point, it
2144 simply continues as though it had been matching the other word in the first place.
2146 'abcdgu'=~/abcdefg|cdgu/
2147 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2148 fail, which would bring use to the state representing 'd' in the second word where we would
2149 try 'g' and succeed, prodceding to match 'cdgu'.
2151 /* add a fail transition */
2152 const U32 trie_offset = ARG(source);
2153 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2155 const U32 ucharcount = trie->uniquecharcount;
2156 const U32 numstates = trie->statecount;
2157 const U32 ubound = trie->lasttrans + ucharcount;
2161 U32 base = trie->states[ 1 ].trans.base;
2164 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2165 GET_RE_DEBUG_FLAGS_DECL;
2167 PERL_UNUSED_ARG(depth);
2171 ARG_SET( stclass, data_slot );
2172 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2173 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2174 aho->trie=trie_offset;
2175 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2176 Copy( trie->states, aho->states, numstates, reg_trie_state );
2177 Newxz( q, numstates, U32);
2178 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2181 /* initialize fail[0..1] to be 1 so that we always have
2182 a valid final fail state */
2183 fail[ 0 ] = fail[ 1 ] = 1;
2185 for ( charid = 0; charid < ucharcount ; charid++ ) {
2186 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2188 q[ q_write ] = newstate;
2189 /* set to point at the root */
2190 fail[ q[ q_write++ ] ]=1;
2193 while ( q_read < q_write) {
2194 const U32 cur = q[ q_read++ % numstates ];
2195 base = trie->states[ cur ].trans.base;
2197 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2198 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2200 U32 fail_state = cur;
2203 fail_state = fail[ fail_state ];
2204 fail_base = aho->states[ fail_state ].trans.base;
2205 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2207 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2208 fail[ ch_state ] = fail_state;
2209 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2211 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2213 q[ q_write++ % numstates] = ch_state;
2217 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2218 when we fail in state 1, this allows us to use the
2219 charclass scan to find a valid start char. This is based on the principle
2220 that theres a good chance the string being searched contains lots of stuff
2221 that cant be a start char.
2223 fail[ 0 ] = fail[ 1 ] = 0;
2224 DEBUG_TRIE_COMPILE_r({
2225 PerlIO_printf(Perl_debug_log,
2226 "%*sStclass Failtable (%"UVuf" states): 0",
2227 (int)(depth * 2), "", (UV)numstates
2229 for( q_read=1; q_read<numstates; q_read++ ) {
2230 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2232 PerlIO_printf(Perl_debug_log, "\n");
2235 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2240 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2241 * These need to be revisited when a newer toolchain becomes available.
2243 #if defined(__sparc64__) && defined(__GNUC__)
2244 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2245 # undef SPARC64_GCC_WORKAROUND
2246 # define SPARC64_GCC_WORKAROUND 1
2250 #define DEBUG_PEEP(str,scan,depth) \
2251 DEBUG_OPTIMISE_r({if (scan){ \
2252 SV * const mysv=sv_newmortal(); \
2253 regnode *Next = regnext(scan); \
2254 regprop(RExC_rx, mysv, scan); \
2255 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2256 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2257 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2264 #define JOIN_EXACT(scan,min,flags) \
2265 if (PL_regkind[OP(scan)] == EXACT) \
2266 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2269 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2270 /* Merge several consecutive EXACTish nodes into one. */
2271 regnode *n = regnext(scan);
2273 regnode *next = scan + NODE_SZ_STR(scan);
2277 regnode *stop = scan;
2278 GET_RE_DEBUG_FLAGS_DECL;
2280 PERL_UNUSED_ARG(depth);
2282 #ifndef EXPERIMENTAL_INPLACESCAN
2283 PERL_UNUSED_ARG(flags);
2284 PERL_UNUSED_ARG(val);
2286 DEBUG_PEEP("join",scan,depth);
2288 /* Skip NOTHING, merge EXACT*. */
2290 ( PL_regkind[OP(n)] == NOTHING ||
2291 (stringok && (OP(n) == OP(scan))))
2293 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2295 if (OP(n) == TAIL || n > next)
2297 if (PL_regkind[OP(n)] == NOTHING) {
2298 DEBUG_PEEP("skip:",n,depth);
2299 NEXT_OFF(scan) += NEXT_OFF(n);
2300 next = n + NODE_STEP_REGNODE;
2307 else if (stringok) {
2308 const unsigned int oldl = STR_LEN(scan);
2309 regnode * const nnext = regnext(n);
2311 DEBUG_PEEP("merg",n,depth);
2314 if (oldl + STR_LEN(n) > U8_MAX)
2316 NEXT_OFF(scan) += NEXT_OFF(n);
2317 STR_LEN(scan) += STR_LEN(n);
2318 next = n + NODE_SZ_STR(n);
2319 /* Now we can overwrite *n : */
2320 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2328 #ifdef EXPERIMENTAL_INPLACESCAN
2329 if (flags && !NEXT_OFF(n)) {
2330 DEBUG_PEEP("atch", val, depth);
2331 if (reg_off_by_arg[OP(n)]) {
2332 ARG_SET(n, val - n);
2335 NEXT_OFF(n) = val - n;
2342 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2344 Two problematic code points in Unicode casefolding of EXACT nodes:
2346 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2347 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2353 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2354 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2356 This means that in case-insensitive matching (or "loose matching",
2357 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2358 length of the above casefolded versions) can match a target string
2359 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2360 This would rather mess up the minimum length computation.
2362 What we'll do is to look for the tail four bytes, and then peek
2363 at the preceding two bytes to see whether we need to decrease
2364 the minimum length by four (six minus two).
2366 Thanks to the design of UTF-8, there cannot be false matches:
2367 A sequence of valid UTF-8 bytes cannot be a subsequence of
2368 another valid sequence of UTF-8 bytes.
2371 char * const s0 = STRING(scan), *s, *t;
2372 char * const s1 = s0 + STR_LEN(scan) - 1;
2373 char * const s2 = s1 - 4;
2374 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2375 const char t0[] = "\xaf\x49\xaf\x42";
2377 const char t0[] = "\xcc\x88\xcc\x81";
2379 const char * const t1 = t0 + 3;
2382 s < s2 && (t = ninstr(s, s1, t0, t1));
2385 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2386 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2388 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2389 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2397 n = scan + NODE_SZ_STR(scan);
2399 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2406 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2410 /* REx optimizer. Converts nodes into quickier variants "in place".
2411 Finds fixed substrings. */
2413 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2414 to the position after last scanned or to NULL. */
2416 #define INIT_AND_WITHP \
2417 assert(!and_withp); \
2418 Newx(and_withp,1,struct regnode_charclass_class); \
2419 SAVEFREEPV(and_withp)
2421 /* this is a chain of data about sub patterns we are processing that
2422 need to be handled seperately/specially in study_chunk. Its so
2423 we can simulate recursion without losing state. */
2425 typedef struct scan_frame {
2426 regnode *last; /* last node to process in this frame */
2427 regnode *next; /* next node to process when last is reached */
2428 struct scan_frame *prev; /*previous frame*/
2429 I32 stop; /* what stopparen do we use */
2433 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2435 #define CASE_SYNST_FNC(nAmE) \
2437 if (flags & SCF_DO_STCLASS_AND) { \
2438 for (value = 0; value < 256; value++) \
2439 if (!is_ ## nAmE ## _cp(value)) \
2440 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2443 for (value = 0; value < 256; value++) \
2444 if (is_ ## nAmE ## _cp(value)) \
2445 ANYOF_BITMAP_SET(data->start_class, value); \
2449 if (flags & SCF_DO_STCLASS_AND) { \
2450 for (value = 0; value < 256; value++) \
2451 if (is_ ## nAmE ## _cp(value)) \
2452 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2455 for (value = 0; value < 256; value++) \
2456 if (!is_ ## nAmE ## _cp(value)) \
2457 ANYOF_BITMAP_SET(data->start_class, value); \
2464 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2465 I32 *minlenp, I32 *deltap,
2470 struct regnode_charclass_class *and_withp,
2471 U32 flags, U32 depth)
2472 /* scanp: Start here (read-write). */
2473 /* deltap: Write maxlen-minlen here. */
2474 /* last: Stop before this one. */
2475 /* data: string data about the pattern */
2476 /* stopparen: treat close N as END */
2477 /* recursed: which subroutines have we recursed into */
2478 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2481 I32 min = 0, pars = 0, code;
2482 regnode *scan = *scanp, *next;
2484 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2485 int is_inf_internal = 0; /* The studied chunk is infinite */
2486 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2487 scan_data_t data_fake;
2488 SV *re_trie_maxbuff = NULL;
2489 regnode *first_non_open = scan;
2490 I32 stopmin = I32_MAX;
2491 scan_frame *frame = NULL;
2493 GET_RE_DEBUG_FLAGS_DECL;
2496 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2500 while (first_non_open && OP(first_non_open) == OPEN)
2501 first_non_open=regnext(first_non_open);
2506 while ( scan && OP(scan) != END && scan < last ){
2507 /* Peephole optimizer: */
2508 DEBUG_STUDYDATA("Peep:", data,depth);
2509 DEBUG_PEEP("Peep",scan,depth);
2510 JOIN_EXACT(scan,&min,0);
2512 /* Follow the next-chain of the current node and optimize
2513 away all the NOTHINGs from it. */
2514 if (OP(scan) != CURLYX) {
2515 const int max = (reg_off_by_arg[OP(scan)]
2517 /* I32 may be smaller than U16 on CRAYs! */
2518 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2519 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2523 /* Skip NOTHING and LONGJMP. */
2524 while ((n = regnext(n))
2525 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2526 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2527 && off + noff < max)
2529 if (reg_off_by_arg[OP(scan)])
2532 NEXT_OFF(scan) = off;
2537 /* The principal pseudo-switch. Cannot be a switch, since we
2538 look into several different things. */
2539 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2540 || OP(scan) == IFTHEN) {
2541 next = regnext(scan);
2543 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2545 if (OP(next) == code || code == IFTHEN) {
2546 /* NOTE - There is similar code to this block below for handling
2547 TRIE nodes on a re-study. If you change stuff here check there
2549 I32 max1 = 0, min1 = I32_MAX, num = 0;
2550 struct regnode_charclass_class accum;
2551 regnode * const startbranch=scan;
2553 if (flags & SCF_DO_SUBSTR)
2554 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2555 if (flags & SCF_DO_STCLASS)
2556 cl_init_zero(pRExC_state, &accum);
2558 while (OP(scan) == code) {
2559 I32 deltanext, minnext, f = 0, fake;
2560 struct regnode_charclass_class this_class;
2563 data_fake.flags = 0;
2565 data_fake.whilem_c = data->whilem_c;
2566 data_fake.last_closep = data->last_closep;
2569 data_fake.last_closep = &fake;
2571 data_fake.pos_delta = delta;
2572 next = regnext(scan);
2573 scan = NEXTOPER(scan);
2575 scan = NEXTOPER(scan);
2576 if (flags & SCF_DO_STCLASS) {
2577 cl_init(pRExC_state, &this_class);
2578 data_fake.start_class = &this_class;
2579 f = SCF_DO_STCLASS_AND;
2581 if (flags & SCF_WHILEM_VISITED_POS)
2582 f |= SCF_WHILEM_VISITED_POS;
2584 /* we suppose the run is continuous, last=next...*/
2585 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2587 stopparen, recursed, NULL, f,depth+1);
2590 if (max1 < minnext + deltanext)
2591 max1 = minnext + deltanext;
2592 if (deltanext == I32_MAX)
2593 is_inf = is_inf_internal = 1;
2595 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2597 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2598 if ( stopmin > minnext)
2599 stopmin = min + min1;
2600 flags &= ~SCF_DO_SUBSTR;
2602 data->flags |= SCF_SEEN_ACCEPT;
2605 if (data_fake.flags & SF_HAS_EVAL)
2606 data->flags |= SF_HAS_EVAL;
2607 data->whilem_c = data_fake.whilem_c;
2609 if (flags & SCF_DO_STCLASS)
2610 cl_or(pRExC_state, &accum, &this_class);
2612 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2614 if (flags & SCF_DO_SUBSTR) {
2615 data->pos_min += min1;
2616 data->pos_delta += max1 - min1;
2617 if (max1 != min1 || is_inf)
2618 data->longest = &(data->longest_float);
2621 delta += max1 - min1;
2622 if (flags & SCF_DO_STCLASS_OR) {
2623 cl_or(pRExC_state, data->start_class, &accum);
2625 cl_and(data->start_class, and_withp);
2626 flags &= ~SCF_DO_STCLASS;
2629 else if (flags & SCF_DO_STCLASS_AND) {
2631 cl_and(data->start_class, &accum);
2632 flags &= ~SCF_DO_STCLASS;
2635 /* Switch to OR mode: cache the old value of
2636 * data->start_class */
2638 StructCopy(data->start_class, and_withp,
2639 struct regnode_charclass_class);
2640 flags &= ~SCF_DO_STCLASS_AND;
2641 StructCopy(&accum, data->start_class,
2642 struct regnode_charclass_class);
2643 flags |= SCF_DO_STCLASS_OR;
2644 data->start_class->flags |= ANYOF_EOS;
2648 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2651 Assuming this was/is a branch we are dealing with: 'scan' now
2652 points at the item that follows the branch sequence, whatever
2653 it is. We now start at the beginning of the sequence and look
2660 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2662 If we can find such a subseqence we need to turn the first
2663 element into a trie and then add the subsequent branch exact
2664 strings to the trie.
2668 1. patterns where the whole set of branch can be converted.
2670 2. patterns where only a subset can be converted.
2672 In case 1 we can replace the whole set with a single regop
2673 for the trie. In case 2 we need to keep the start and end
2676 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2677 becomes BRANCH TRIE; BRANCH X;
2679 There is an additional case, that being where there is a
2680 common prefix, which gets split out into an EXACT like node
2681 preceding the TRIE node.
2683 If x(1..n)==tail then we can do a simple trie, if not we make
2684 a "jump" trie, such that when we match the appropriate word
2685 we "jump" to the appopriate tail node. Essentailly we turn
2686 a nested if into a case structure of sorts.
2691 if (!re_trie_maxbuff) {
2692 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2693 if (!SvIOK(re_trie_maxbuff))
2694 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2696 if ( SvIV(re_trie_maxbuff)>=0 ) {
2698 regnode *first = (regnode *)NULL;
2699 regnode *last = (regnode *)NULL;
2700 regnode *tail = scan;
2705 SV * const mysv = sv_newmortal(); /* for dumping */
2707 /* var tail is used because there may be a TAIL
2708 regop in the way. Ie, the exacts will point to the
2709 thing following the TAIL, but the last branch will
2710 point at the TAIL. So we advance tail. If we
2711 have nested (?:) we may have to move through several
2715 while ( OP( tail ) == TAIL ) {
2716 /* this is the TAIL generated by (?:) */
2717 tail = regnext( tail );
2722 regprop(RExC_rx, mysv, tail );
2723 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2724 (int)depth * 2 + 2, "",
2725 "Looking for TRIE'able sequences. Tail node is: ",
2726 SvPV_nolen_const( mysv )
2732 step through the branches, cur represents each
2733 branch, noper is the first thing to be matched
2734 as part of that branch and noper_next is the
2735 regnext() of that node. if noper is an EXACT
2736 and noper_next is the same as scan (our current
2737 position in the regex) then the EXACT branch is
2738 a possible optimization target. Once we have
2739 two or more consequetive such branches we can
2740 create a trie of the EXACT's contents and stich
2741 it in place. If the sequence represents all of
2742 the branches we eliminate the whole thing and
2743 replace it with a single TRIE. If it is a
2744 subsequence then we need to stitch it in. This
2745 means the first branch has to remain, and needs
2746 to be repointed at the item on the branch chain
2747 following the last branch optimized. This could
2748 be either a BRANCH, in which case the
2749 subsequence is internal, or it could be the
2750 item following the branch sequence in which
2751 case the subsequence is at the end.
2755 /* dont use tail as the end marker for this traverse */
2756 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2757 regnode * const noper = NEXTOPER( cur );
2758 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2759 regnode * const noper_next = regnext( noper );
2763 regprop(RExC_rx, mysv, cur);
2764 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2765 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2767 regprop(RExC_rx, mysv, noper);
2768 PerlIO_printf( Perl_debug_log, " -> %s",
2769 SvPV_nolen_const(mysv));
2772 regprop(RExC_rx, mysv, noper_next );
2773 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2774 SvPV_nolen_const(mysv));
2776 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2777 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2779 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2780 : PL_regkind[ OP( noper ) ] == EXACT )
2781 || OP(noper) == NOTHING )
2783 && noper_next == tail
2788 if ( !first || optype == NOTHING ) {
2789 if (!first) first = cur;
2790 optype = OP( noper );
2796 Currently we assume that the trie can handle unicode and ascii
2797 matches fold cased matches. If this proves true then the following
2798 define will prevent tries in this situation.
2800 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2802 #define TRIE_TYPE_IS_SAFE 1
2803 if ( last && TRIE_TYPE_IS_SAFE ) {
2804 make_trie( pRExC_state,
2805 startbranch, first, cur, tail, count,
2808 if ( PL_regkind[ OP( noper ) ] == EXACT
2810 && noper_next == tail
2815 optype = OP( noper );
2825 regprop(RExC_rx, mysv, cur);
2826 PerlIO_printf( Perl_debug_log,
2827 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2828 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2832 if ( last && TRIE_TYPE_IS_SAFE ) {
2833 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2834 #ifdef TRIE_STUDY_OPT
2835 if ( ((made == MADE_EXACT_TRIE &&
2836 startbranch == first)
2837 || ( first_non_open == first )) &&
2839 flags |= SCF_TRIE_RESTUDY;
2840 if ( startbranch == first
2843 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2853 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2854 scan = NEXTOPER(NEXTOPER(scan));
2855 } else /* single branch is optimized. */
2856 scan = NEXTOPER(scan);
2858 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2859 scan_frame *newframe = NULL;
2864 if (OP(scan) != SUSPEND) {
2865 /* set the pointer */
2866 if (OP(scan) == GOSUB) {
2868 RExC_recurse[ARG2L(scan)] = scan;
2869 start = RExC_open_parens[paren-1];
2870 end = RExC_close_parens[paren-1];
2873 start = RExC_rxi->program + 1;
2877 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2878 SAVEFREEPV(recursed);
2880 if (!PAREN_TEST(recursed,paren+1)) {
2881 PAREN_SET(recursed,paren+1);
2882 Newx(newframe,1,scan_frame);
2884 if (flags & SCF_DO_SUBSTR) {
2885 SCAN_COMMIT(pRExC_state,data,minlenp);
2886 data->longest = &(data->longest_float);
2888 is_inf = is_inf_internal = 1;
2889 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2890 cl_anything(pRExC_state, data->start_class);
2891 flags &= ~SCF_DO_STCLASS;
2894 Newx(newframe,1,scan_frame);
2897 end = regnext(scan);
2902 SAVEFREEPV(newframe);
2903 newframe->next = regnext(scan);
2904 newframe->last = last;
2905 newframe->stop = stopparen;
2906 newframe->prev = frame;
2916 else if (OP(scan) == EXACT) {
2917 I32 l = STR_LEN(scan);
2920 const U8 * const s = (U8*)STRING(scan);
2921 l = utf8_length(s, s + l);
2922 uc = utf8_to_uvchr(s, NULL);
2924 uc = *((U8*)STRING(scan));
2927 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2928 /* The code below prefers earlier match for fixed
2929 offset, later match for variable offset. */
2930 if (data->last_end == -1) { /* Update the start info. */
2931 data->last_start_min = data->pos_min;
2932 data->last_start_max = is_inf
2933 ? I32_MAX : data->pos_min + data->pos_delta;
2935 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2937 SvUTF8_on(data->last_found);
2939 SV * const sv = data->last_found;
2940 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2941 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2942 if (mg && mg->mg_len >= 0)
2943 mg->mg_len += utf8_length((U8*)STRING(scan),
2944 (U8*)STRING(scan)+STR_LEN(scan));
2946 data->last_end = data->pos_min + l;
2947 data->pos_min += l; /* As in the first entry. */
2948 data->flags &= ~SF_BEFORE_EOL;
2950 if (flags & SCF_DO_STCLASS_AND) {
2951 /* Check whether it is compatible with what we know already! */
2955 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2956 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2957 && (!(data->start_class->flags & ANYOF_FOLD)
2958 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2961 ANYOF_CLASS_ZERO(data->start_class);
2962 ANYOF_BITMAP_ZERO(data->start_class);
2964 ANYOF_BITMAP_SET(data->start_class, uc);
2965 data->start_class->flags &= ~ANYOF_EOS;
2967 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2969 else if (flags & SCF_DO_STCLASS_OR) {
2970 /* false positive possible if the class is case-folded */
2972 ANYOF_BITMAP_SET(data->start_class, uc);
2974 data->start_class->flags |= ANYOF_UNICODE_ALL;
2975 data->start_class->flags &= ~ANYOF_EOS;
2976 cl_and(data->start_class, and_withp);
2978 flags &= ~SCF_DO_STCLASS;
2980 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2981 I32 l = STR_LEN(scan);
2982 UV uc = *((U8*)STRING(scan));
2984 /* Search for fixed substrings supports EXACT only. */
2985 if (flags & SCF_DO_SUBSTR) {
2987 SCAN_COMMIT(pRExC_state, data, minlenp);
2990 const U8 * const s = (U8 *)STRING(scan);
2991 l = utf8_length(s, s + l);
2992 uc = utf8_to_uvchr(s, NULL);
2995 if (flags & SCF_DO_SUBSTR)
2997 if (flags & SCF_DO_STCLASS_AND) {
2998 /* Check whether it is compatible with what we know already! */
3002 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3003 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3004 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3006 ANYOF_CLASS_ZERO(data->start_class);
3007 ANYOF_BITMAP_ZERO(data->start_class);
3009 ANYOF_BITMAP_SET(data->start_class, uc);
3010 data->start_class->flags &= ~ANYOF_EOS;
3011 data->start_class->flags |= ANYOF_FOLD;
3012 if (OP(scan) == EXACTFL)
3013 data->start_class->flags |= ANYOF_LOCALE;
3016 else if (flags & SCF_DO_STCLASS_OR) {
3017 if (data->start_class->flags & ANYOF_FOLD) {
3018 /* false positive possible if the class is case-folded.
3019 Assume that the locale settings are the same... */
3021 ANYOF_BITMAP_SET(data->start_class, uc);
3022 data->start_class->flags &= ~ANYOF_EOS;
3024 cl_and(data->start_class, and_withp);
3026 flags &= ~SCF_DO_STCLASS;
3028 else if (strchr((const char*)PL_varies,OP(scan))) {
3029 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3030 I32 f = flags, pos_before = 0;
3031 regnode * const oscan = scan;
3032 struct regnode_charclass_class this_class;
3033 struct regnode_charclass_class *oclass = NULL;
3034 I32 next_is_eval = 0;
3036 switch (PL_regkind[OP(scan)]) {
3037 case WHILEM: /* End of (?:...)* . */
3038 scan = NEXTOPER(scan);
3041 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3042 next = NEXTOPER(scan);
3043 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3045 maxcount = REG_INFTY;
3046 next = regnext(scan);
3047 scan = NEXTOPER(scan);
3051 if (flags & SCF_DO_SUBSTR)
3056 if (flags & SCF_DO_STCLASS) {
3058 maxcount = REG_INFTY;
3059 next = regnext(scan);
3060 scan = NEXTOPER(scan);
3063 is_inf = is_inf_internal = 1;
3064 scan = regnext(scan);
3065 if (flags & SCF_DO_SUBSTR) {
3066 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3067 data->longest = &(data->longest_float);
3069 goto optimize_curly_tail;
3071 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3072 && (scan->flags == stopparen))
3077 mincount = ARG1(scan);
3078 maxcount = ARG2(scan);
3080 next = regnext(scan);
3081 if (OP(scan) == CURLYX) {
3082 I32 lp = (data ? *(data->last_closep) : 0);
3083 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3085 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3086 next_is_eval = (OP(scan) == EVAL);
3088 if (flags & SCF_DO_SUBSTR) {
3089 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3090 pos_before = data->pos_min;
3094 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3096 data->flags |= SF_IS_INF;
3098 if (flags & SCF_DO_STCLASS) {
3099 cl_init(pRExC_state, &this_class);
3100 oclass = data->start_class;
3101 data->start_class = &this_class;
3102 f |= SCF_DO_STCLASS_AND;
3103 f &= ~SCF_DO_STCLASS_OR;
3105 /* These are the cases when once a subexpression
3106 fails at a particular position, it cannot succeed
3107 even after backtracking at the enclosing scope.
3109 XXXX what if minimal match and we are at the
3110 initial run of {n,m}? */
3111 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3112 f &= ~SCF_WHILEM_VISITED_POS;
3114 /* This will finish on WHILEM, setting scan, or on NULL: */
3115 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3116 last, data, stopparen, recursed, NULL,
3118 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3120 if (flags & SCF_DO_STCLASS)
3121 data->start_class = oclass;
3122 if (mincount == 0 || minnext == 0) {
3123 if (flags & SCF_DO_STCLASS_OR) {
3124 cl_or(pRExC_state, data->start_class, &this_class);
3126 else if (flags & SCF_DO_STCLASS_AND) {
3127 /* Switch to OR mode: cache the old value of
3128 * data->start_class */
3130 StructCopy(data->start_class, and_withp,
3131 struct regnode_charclass_class);
3132 flags &= ~SCF_DO_STCLASS_AND;
3133 StructCopy(&this_class, data->start_class,
3134 struct regnode_charclass_class);
3135 flags |= SCF_DO_STCLASS_OR;
3136 data->start_class->flags |= ANYOF_EOS;
3138 } else { /* Non-zero len */
3139 if (flags & SCF_DO_STCLASS_OR) {
3140 cl_or(pRExC_state, data->start_class, &this_class);
3141 cl_and(data->start_class, and_withp);
3143 else if (flags & SCF_DO_STCLASS_AND)
3144 cl_and(data->start_class, &this_class);
3145 flags &= ~SCF_DO_STCLASS;
3147 if (!scan) /* It was not CURLYX, but CURLY. */
3149 if ( /* ? quantifier ok, except for (?{ ... }) */
3150 (next_is_eval || !(mincount == 0 && maxcount == 1))
3151 && (minnext == 0) && (deltanext == 0)
3152 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3153 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3154 && ckWARN(WARN_REGEXP))
3157 "Quantifier unexpected on zero-length expression");
3160 min += minnext * mincount;
3161 is_inf_internal |= ((maxcount == REG_INFTY
3162 && (minnext + deltanext) > 0)
3163 || deltanext == I32_MAX);
3164 is_inf |= is_inf_internal;
3165 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3167 /* Try powerful optimization CURLYX => CURLYN. */
3168 if ( OP(oscan) == CURLYX && data
3169 && data->flags & SF_IN_PAR
3170 && !(data->flags & SF_HAS_EVAL)
3171 && !deltanext && minnext == 1 ) {
3172 /* Try to optimize to CURLYN. */
3173 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3174 regnode * const nxt1 = nxt;
3181 if (!strchr((const char*)PL_simple,OP(nxt))
3182 && !(PL_regkind[OP(nxt)] == EXACT
3183 && STR_LEN(nxt) == 1))
3189 if (OP(nxt) != CLOSE)
3191 if (RExC_open_parens) {
3192 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3193 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3195 /* Now we know that nxt2 is the only contents: */
3196 oscan->flags = (U8)ARG(nxt);
3198 OP(nxt1) = NOTHING; /* was OPEN. */
3201 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3202 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3203 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3204 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3205 OP(nxt + 1) = OPTIMIZED; /* was count. */
3206 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3211 /* Try optimization CURLYX => CURLYM. */
3212 if ( OP(oscan) == CURLYX && data
3213 && !(data->flags & SF_HAS_PAR)
3214 && !(data->flags & SF_HAS_EVAL)
3215 && !deltanext /* atom is fixed width */
3216 && minnext != 0 /* CURLYM can't handle zero width */
3218 /* XXXX How to optimize if data == 0? */
3219 /* Optimize to a simpler form. */
3220 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3224 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3225 && (OP(nxt2) != WHILEM))
3227 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3228 /* Need to optimize away parenths. */
3229 if (data->flags & SF_IN_PAR) {
3230 /* Set the parenth number. */
3231 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3233 if (OP(nxt) != CLOSE)
3234 FAIL("Panic opt close");
3235 oscan->flags = (U8)ARG(nxt);
3236 if (RExC_open_parens) {
3237 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3238 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3240 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3241 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3244 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3245 OP(nxt + 1) = OPTIMIZED; /* was count. */
3246 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3247 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3250 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3251 regnode *nnxt = regnext(nxt1);
3254 if (reg_off_by_arg[OP(nxt1)])
3255 ARG_SET(nxt1, nxt2 - nxt1);
3256 else if (nxt2 - nxt1 < U16_MAX)
3257 NEXT_OFF(nxt1) = nxt2 - nxt1;
3259 OP(nxt) = NOTHING; /* Cannot beautify */
3264 /* Optimize again: */
3265 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3266 NULL, stopparen, recursed, NULL, 0,depth+1);
3271 else if ((OP(oscan) == CURLYX)
3272 && (flags & SCF_WHILEM_VISITED_POS)
3273 /* See the comment on a similar expression above.
3274 However, this time it not a subexpression
3275 we care about, but the expression itself. */
3276 && (maxcount == REG_INFTY)
3277 && data && ++data->whilem_c < 16) {
3278 /* This stays as CURLYX, we can put the count/of pair. */
3279 /* Find WHILEM (as in regexec.c) */
3280 regnode *nxt = oscan + NEXT_OFF(oscan);
3282 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3284 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3285 | (RExC_whilem_seen << 4)); /* On WHILEM */
3287 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3289 if (flags & SCF_DO_SUBSTR) {
3290 SV *last_str = NULL;
3291 int counted = mincount != 0;
3293 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3294 #if defined(SPARC64_GCC_WORKAROUND)
3297 const char *s = NULL;
3300 if (pos_before >= data->last_start_min)
3303 b = data->last_start_min;
3306 s = SvPV_const(data->last_found, l);
3307 old = b - data->last_start_min;
3310 I32 b = pos_before >= data->last_start_min
3311 ? pos_before : data->last_start_min;
3313 const char * const s = SvPV_const(data->last_found, l);
3314 I32 old = b - data->last_start_min;
3318 old = utf8_hop((U8*)s, old) - (U8*)s;
3321 /* Get the added string: */
3322 last_str = newSVpvn_utf8(s + old, l, UTF);
3323 if (deltanext == 0 && pos_before == b) {
3324 /* What was added is a constant string */
3326 SvGROW(last_str, (mincount * l) + 1);
3327 repeatcpy(SvPVX(last_str) + l,
3328 SvPVX_const(last_str), l, mincount - 1);
3329 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3330 /* Add additional parts. */
3331 SvCUR_set(data->last_found,
3332 SvCUR(data->last_found) - l);
3333 sv_catsv(data->last_found, last_str);
3335 SV * sv = data->last_found;
3337 SvUTF8(sv) && SvMAGICAL(sv) ?
3338 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3339 if (mg && mg->mg_len >= 0)
3340 mg->mg_len += CHR_SVLEN(last_str) - l;
3342 data->last_end += l * (mincount - 1);
3345 /* start offset must point into the last copy */
3346 data->last_start_min += minnext * (mincount - 1);
3347 data->last_start_max += is_inf ? I32_MAX
3348 : (maxcount - 1) * (minnext + data->pos_delta);
3351 /* It is counted once already... */
3352 data->pos_min += minnext * (mincount - counted);
3353 data->pos_delta += - counted * deltanext +
3354 (minnext + deltanext) * maxcount - minnext * mincount;
3355 if (mincount != maxcount) {
3356 /* Cannot extend fixed substrings found inside
3358 SCAN_COMMIT(pRExC_state,data,minlenp);
3359 if (mincount && last_str) {
3360 SV * const sv = data->last_found;
3361 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3362 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3366 sv_setsv(sv, last_str);
3367 data->last_end = data->pos_min;
3368 data->last_start_min =
3369 data->pos_min - CHR_SVLEN(last_str);
3370 data->last_start_max = is_inf
3372 : data->pos_min + data->pos_delta
3373 - CHR_SVLEN(last_str);
3375 data->longest = &(data->longest_float);
3377 SvREFCNT_dec(last_str);
3379 if (data && (fl & SF_HAS_EVAL))
3380 data->flags |= SF_HAS_EVAL;
3381 optimize_curly_tail:
3382 if (OP(oscan) != CURLYX) {
3383 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3385 NEXT_OFF(oscan) += NEXT_OFF(next);
3388 default: /* REF and CLUMP only? */
3389 if (flags & SCF_DO_SUBSTR) {
3390 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3391 data->longest = &(data->longest_float);
3393 is_inf = is_inf_internal = 1;
3394 if (flags & SCF_DO_STCLASS_OR)
3395 cl_anything(pRExC_state, data->start_class);
3396 flags &= ~SCF_DO_STCLASS;
3400 else if (OP(scan) == LNBREAK) {
3401 if (flags & SCF_DO_STCLASS) {
3403 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3404 if (flags & SCF_DO_STCLASS_AND) {
3405 for (value = 0; value < 256; value++)
3406 if (!is_VERTWS_cp(value))
3407 ANYOF_BITMAP_CLEAR(data->start_class, value);
3410 for (value = 0; value < 256; value++)
3411 if (is_VERTWS_cp(value))
3412 ANYOF_BITMAP_SET(data->start_class, value);
3414 if (flags & SCF_DO_STCLASS_OR)
3415 cl_and(data->start_class, and_withp);
3416 flags &= ~SCF_DO_STCLASS;
3420 if (flags & SCF_DO_SUBSTR) {
3421 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3423 data->pos_delta += 1;
3424 data->longest = &(data->longest_float);
3428 else if (OP(scan) == FOLDCHAR) {
3429 int d = ARG(scan)==0xDF ? 1 : 2;
3430 flags &= ~SCF_DO_STCLASS;
3433 if (flags & SCF_DO_SUBSTR) {
3434 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3436 data->pos_delta += d;
3437 data->longest = &(data->longest_float);
3440 else if (strchr((const char*)PL_simple,OP(scan))) {
3443 if (flags & SCF_DO_SUBSTR) {
3444 SCAN_COMMIT(pRExC_state,data,minlenp);
3448 if (flags & SCF_DO_STCLASS) {
3449 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3451 /* Some of the logic below assumes that switching
3452 locale on will only add false positives. */
3453 switch (PL_regkind[OP(scan)]) {
3457 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3458 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3459 cl_anything(pRExC_state, data->start_class);
3462 if (OP(scan) == SANY)
3464 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3465 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3466 || (data->start_class->flags & ANYOF_CLASS));
3467 cl_anything(pRExC_state, data->start_class);
3469 if (flags & SCF_DO_STCLASS_AND || !value)
3470 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3473 if (flags & SCF_DO_STCLASS_AND)
3474 cl_and(data->start_class,
3475 (struct regnode_charclass_class*)scan);
3477 cl_or(pRExC_state, data->start_class,
3478 (struct regnode_charclass_class*)scan);
3481 if (flags & SCF_DO_STCLASS_AND) {
3482 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3483 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3484 for (value = 0; value < 256; value++)
3485 if (!isALNUM(value))
3486 ANYOF_BITMAP_CLEAR(data->start_class, value);
3490 if (data->start_class->flags & ANYOF_LOCALE)
3491 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3493 for (value = 0; value < 256; value++)
3495 ANYOF_BITMAP_SET(data->start_class, value);
3500 if (flags & SCF_DO_STCLASS_AND) {
3501 if (data->start_class->flags & ANYOF_LOCALE)
3502 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3505 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3506 data->start_class->flags |= ANYOF_LOCALE;
3510 if (flags & SCF_DO_STCLASS_AND) {
3511 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3512 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3513 for (value = 0; value < 256; value++)
3515 ANYOF_BITMAP_CLEAR(data->start_class, value);
3519 if (data->start_class->flags & ANYOF_LOCALE)
3520 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3522 for (value = 0; value < 256; value++)
3523 if (!isALNUM(value))
3524 ANYOF_BITMAP_SET(data->start_class, value);
3529 if (flags & SCF_DO_STCLASS_AND) {
3530 if (data->start_class->flags & ANYOF_LOCALE)
3531 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3534 data->start_class->flags |= ANYOF_LOCALE;
3535 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3539 if (flags & SCF_DO_STCLASS_AND) {
3540 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3541 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3542 for (value = 0; value < 256; value++)
3543 if (!isSPACE(value))
3544 ANYOF_BITMAP_CLEAR(data->start_class, value);
3548 if (data->start_class->flags & ANYOF_LOCALE)
3549 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3551 for (value = 0; value < 256; value++)
3553 ANYOF_BITMAP_SET(data->start_class, value);
3558 if (flags & SCF_DO_STCLASS_AND) {
3559 if (data->start_class->flags & ANYOF_LOCALE)
3560 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3563 data->start_class->flags |= ANYOF_LOCALE;
3564 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3568 if (flags & SCF_DO_STCLASS_AND) {
3569 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3570 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3571 for (value = 0; value < 256; value++)
3573 ANYOF_BITMAP_CLEAR(data->start_class, value);
3577 if (data->start_class->flags & ANYOF_LOCALE)
3578 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3580 for (value = 0; value < 256; value++)
3581 if (!isSPACE(value))
3582 ANYOF_BITMAP_SET(data->start_class, value);
3587 if (flags & SCF_DO_STCLASS_AND) {
3588 if (data->start_class->flags & ANYOF_LOCALE) {
3589 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3590 for (value = 0; value < 256; value++)
3591 if (!isSPACE(value))
3592 ANYOF_BITMAP_CLEAR(data->start_class, value);
3596 data->start_class->flags |= ANYOF_LOCALE;
3597 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3601 if (flags & SCF_DO_STCLASS_AND) {
3602 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3603 for (value = 0; value < 256; value++)
3604 if (!isDIGIT(value))
3605 ANYOF_BITMAP_CLEAR(data->start_class, value);
3608 if (data->start_class->flags & ANYOF_LOCALE)
3609 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3611 for (value = 0; value < 256; value++)
3613 ANYOF_BITMAP_SET(data->start_class, value);
3618 if (flags & SCF_DO_STCLASS_AND) {
3619 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3620 for (value = 0; value < 256; value++)
3622 ANYOF_BITMAP_CLEAR(data->start_class, value);
3625 if (data->start_class->flags & ANYOF_LOCALE)
3626 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3628 for (value = 0; value < 256; value++)
3629 if (!isDIGIT(value))
3630 ANYOF_BITMAP_SET(data->start_class, value);
3634 CASE_SYNST_FNC(VERTWS);
3635 CASE_SYNST_FNC(HORIZWS);
3638 if (flags & SCF_DO_STCLASS_OR)
3639 cl_and(data->start_class, and_withp);
3640 flags &= ~SCF_DO_STCLASS;
3643 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3644 data->flags |= (OP(scan) == MEOL
3648 else if ( PL_regkind[OP(scan)] == BRANCHJ
3649 /* Lookbehind, or need to calculate parens/evals/stclass: */
3650 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3651 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3652 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3653 || OP(scan) == UNLESSM )
3655 /* Negative Lookahead/lookbehind
3656 In this case we can't do fixed string optimisation.
3659 I32 deltanext, minnext, fake = 0;
3661 struct regnode_charclass_class intrnl;
3664 data_fake.flags = 0;
3666 data_fake.whilem_c = data->whilem_c;
3667 data_fake.last_closep = data->last_closep;
3670 data_fake.last_closep = &fake;
3671 data_fake.pos_delta = delta;
3672 if ( flags & SCF_DO_STCLASS && !scan->flags
3673 && OP(scan) == IFMATCH ) { /* Lookahead */
3674 cl_init(pRExC_state, &intrnl);
3675 data_fake.start_class = &intrnl;
3676 f |= SCF_DO_STCLASS_AND;
3678 if (flags & SCF_WHILEM_VISITED_POS)
3679 f |= SCF_WHILEM_VISITED_POS;
3680 next = regnext(scan);
3681 nscan = NEXTOPER(NEXTOPER(scan));
3682 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3683 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3686 FAIL("Variable length lookbehind not implemented");
3688 else if (minnext > (I32)U8_MAX) {
3689 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3691 scan->flags = (U8)minnext;
3694 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3696 if (data_fake.flags & SF_HAS_EVAL)
3697 data->flags |= SF_HAS_EVAL;
3698 data->whilem_c = data_fake.whilem_c;
3700 if (f & SCF_DO_STCLASS_AND) {
3701 const int was = (data->start_class->flags & ANYOF_EOS);
3703 cl_and(data->start_class, &intrnl);
3705 data->start_class->flags |= ANYOF_EOS;
3708 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3710 /* Positive Lookahead/lookbehind
3711 In this case we can do fixed string optimisation,
3712 but we must be careful about it. Note in the case of
3713 lookbehind the positions will be offset by the minimum
3714 length of the pattern, something we won't know about
3715 until after the recurse.
3717 I32 deltanext, fake = 0;
3719 struct regnode_charclass_class intrnl;
3721 /* We use SAVEFREEPV so that when the full compile
3722 is finished perl will clean up the allocated
3723 minlens when its all done. This was we don't
3724 have to worry about freeing them when we know
3725 they wont be used, which would be a pain.
3728 Newx( minnextp, 1, I32 );
3729 SAVEFREEPV(minnextp);
3732 StructCopy(data, &data_fake, scan_data_t);
3733 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3736 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3737 data_fake.last_found=newSVsv(data->last_found);
3741 data_fake.last_closep = &fake;
3742 data_fake.flags = 0;
3743 data_fake.pos_delta = delta;
3745 data_fake.flags |= SF_IS_INF;
3746 if ( flags & SCF_DO_STCLASS && !scan->flags
3747 && OP(scan) == IFMATCH ) { /* Lookahead */
3748 cl_init(pRExC_state, &intrnl);
3749 data_fake.start_class = &intrnl;
3750 f |= SCF_DO_STCLASS_AND;
3752 if (flags & SCF_WHILEM_VISITED_POS)
3753 f |= SCF_WHILEM_VISITED_POS;
3754 next = regnext(scan);
3755 nscan = NEXTOPER(NEXTOPER(scan));
3757 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3758 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3761 FAIL("Variable length lookbehind not implemented");
3763 else if (*minnextp > (I32)U8_MAX) {
3764 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3766 scan->flags = (U8)*minnextp;
3771 if (f & SCF_DO_STCLASS_AND) {
3772 const int was = (data->start_class->flags & ANYOF_EOS);
3774 cl_and(data->start_class, &intrnl);
3776 data->start_class->flags |= ANYOF_EOS;
3779 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3781 if (data_fake.flags & SF_HAS_EVAL)
3782 data->flags |= SF_HAS_EVAL;
3783 data->whilem_c = data_fake.whilem_c;
3784 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3785 if (RExC_rx->minlen<*minnextp)
3786 RExC_rx->minlen=*minnextp;
3787 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3788 SvREFCNT_dec(data_fake.last_found);
3790 if ( data_fake.minlen_fixed != minlenp )
3792 data->offset_fixed= data_fake.offset_fixed;
3793 data->minlen_fixed= data_fake.minlen_fixed;
3794 data->lookbehind_fixed+= scan->flags;
3796 if ( data_fake.minlen_float != minlenp )
3798 data->minlen_float= data_fake.minlen_float;
3799 data->offset_float_min=data_fake.offset_float_min;
3800 data->offset_float_max=data_fake.offset_float_max;
3801 data->lookbehind_float+= scan->flags;
3810 else if (OP(scan) == OPEN) {
3811 if (stopparen != (I32)ARG(scan))
3814 else if (OP(scan) == CLOSE) {
3815 if (stopparen == (I32)ARG(scan)) {
3818 if ((I32)ARG(scan) == is_par) {
3819 next = regnext(scan);
3821 if ( next && (OP(next) != WHILEM) && next < last)
3822 is_par = 0; /* Disable optimization */
3825 *(data->last_closep) = ARG(scan);
3827 else if (OP(scan) == EVAL) {
3829 data->flags |= SF_HAS_EVAL;
3831 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3832 if (flags & SCF_DO_SUBSTR) {
3833 SCAN_COMMIT(pRExC_state,data,minlenp);
3834 flags &= ~SCF_DO_SUBSTR;
3836 if (data && OP(scan)==ACCEPT) {
3837 data->flags |= SCF_SEEN_ACCEPT;
3842 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3844 if (flags & SCF_DO_SUBSTR) {
3845 SCAN_COMMIT(pRExC_state,data,minlenp);
3846 data->longest = &(data->longest_float);
3848 is_inf = is_inf_internal = 1;
3849 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3850 cl_anything(pRExC_state, data->start_class);
3851 flags &= ~SCF_DO_STCLASS;
3853 else if (OP(scan) == GPOS) {
3854 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3855 !(delta || is_inf || (data && data->pos_delta)))
3857 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3858 RExC_rx->extflags |= RXf_ANCH_GPOS;
3859 if (RExC_rx->gofs < (U32)min)
3860 RExC_rx->gofs = min;
3862 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3866 #ifdef TRIE_STUDY_OPT
3867 #ifdef FULL_TRIE_STUDY
3868 else if (PL_regkind[OP(scan)] == TRIE) {
3869 /* NOTE - There is similar code to this block above for handling
3870 BRANCH nodes on the initial study. If you change stuff here
3872 regnode *trie_node= scan;
3873 regnode *tail= regnext(scan);
3874 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3875 I32 max1 = 0, min1 = I32_MAX;
3876 struct regnode_charclass_class accum;
3878 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3879 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3880 if (flags & SCF_DO_STCLASS)
3881 cl_init_zero(pRExC_state, &accum);
3887 const regnode *nextbranch= NULL;
3890 for ( word=1 ; word <= trie->wordcount ; word++)
3892 I32 deltanext=0, minnext=0, f = 0, fake;
3893 struct regnode_charclass_class this_class;
3895 data_fake.flags = 0;
3897 data_fake.whilem_c = data->whilem_c;
3898 data_fake.last_closep = data->last_closep;
3901 data_fake.last_closep = &fake;
3902 data_fake.pos_delta = delta;
3903 if (flags & SCF_DO_STCLASS) {
3904 cl_init(pRExC_state, &this_class);
3905 data_fake.start_class = &this_class;
3906 f = SCF_DO_STCLASS_AND;
3908 if (flags & SCF_WHILEM_VISITED_POS)
3909 f |= SCF_WHILEM_VISITED_POS;
3911 if (trie->jump[word]) {
3913 nextbranch = trie_node + trie->jump[0];
3914 scan= trie_node + trie->jump[word];
3915 /* We go from the jump point to the branch that follows
3916 it. Note this means we need the vestigal unused branches
3917 even though they arent otherwise used.
3919 minnext = study_chunk(pRExC_state, &scan, minlenp,
3920 &deltanext, (regnode *)nextbranch, &data_fake,
3921 stopparen, recursed, NULL, f,depth+1);
3923 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3924 nextbranch= regnext((regnode*)nextbranch);
3926 if (min1 > (I32)(minnext + trie->minlen))
3927 min1 = minnext + trie->minlen;
3928 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3929 max1 = minnext + deltanext + trie->maxlen;
3930 if (deltanext == I32_MAX)
3931 is_inf = is_inf_internal = 1;
3933 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3935 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3936 if ( stopmin > min + min1)
3937 stopmin = min + min1;
3938 flags &= ~SCF_DO_SUBSTR;
3940 data->flags |= SCF_SEEN_ACCEPT;
3943 if (data_fake.flags & SF_HAS_EVAL)
3944 data->flags |= SF_HAS_EVAL;
3945 data->whilem_c = data_fake.whilem_c;
3947 if (flags & SCF_DO_STCLASS)
3948 cl_or(pRExC_state, &accum, &this_class);
3951 if (flags & SCF_DO_SUBSTR) {
3952 data->pos_min += min1;
3953 data->pos_delta += max1 - min1;
3954 if (max1 != min1 || is_inf)
3955 data->longest = &(data->longest_float);
3958 delta += max1 - min1;
3959 if (flags & SCF_DO_STCLASS_OR) {
3960 cl_or(pRExC_state, data->start_class, &accum);
3962 cl_and(data->start_class, and_withp);
3963 flags &= ~SCF_DO_STCLASS;
3966 else if (flags & SCF_DO_STCLASS_AND) {
3968 cl_and(data->start_class, &accum);
3969 flags &= ~SCF_DO_STCLASS;
3972 /* Switch to OR mode: cache the old value of
3973 * data->start_class */
3975 StructCopy(data->start_class, and_withp,
3976 struct regnode_charclass_class);
3977 flags &= ~SCF_DO_STCLASS_AND;
3978 StructCopy(&accum, data->start_class,
3979 struct regnode_charclass_class);
3980 flags |= SCF_DO_STCLASS_OR;
3981 data->start_class->flags |= ANYOF_EOS;
3988 else if (PL_regkind[OP(scan)] == TRIE) {
3989 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3992 min += trie->minlen;
3993 delta += (trie->maxlen - trie->minlen);
3994 flags &= ~SCF_DO_STCLASS; /* xxx */
3995 if (flags & SCF_DO_SUBSTR) {
3996 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3997 data->pos_min += trie->minlen;
3998 data->pos_delta += (trie->maxlen - trie->minlen);
3999 if (trie->maxlen != trie->minlen)
4000 data->longest = &(data->longest_float);
4002 if (trie->jump) /* no more substrings -- for now /grr*/
4003 flags &= ~SCF_DO_SUBSTR;
4005 #endif /* old or new */
4006 #endif /* TRIE_STUDY_OPT */
4008 /* Else: zero-length, ignore. */
4009 scan = regnext(scan);
4014 stopparen = frame->stop;
4015 frame = frame->prev;
4016 goto fake_study_recurse;
4021 DEBUG_STUDYDATA("pre-fin:",data,depth);
4024 *deltap = is_inf_internal ? I32_MAX : delta;
4025 if (flags & SCF_DO_SUBSTR && is_inf)
4026 data->pos_delta = I32_MAX - data->pos_min;
4027 if (is_par > (I32)U8_MAX)
4029 if (is_par && pars==1 && data) {
4030 data->flags |= SF_IN_PAR;
4031 data->flags &= ~SF_HAS_PAR;
4033 else if (pars && data) {
4034 data->flags |= SF_HAS_PAR;
4035 data->flags &= ~SF_IN_PAR;
4037 if (flags & SCF_DO_STCLASS_OR)
4038 cl_and(data->start_class, and_withp);
4039 if (flags & SCF_TRIE_RESTUDY)
4040 data->flags |= SCF_TRIE_RESTUDY;
4042 DEBUG_STUDYDATA("post-fin:",data,depth);
4044 return min < stopmin ? min : stopmin;
4048 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4050 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4052 Renewc(RExC_rxi->data,
4053 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4054 char, struct reg_data);
4056 Renew(RExC_rxi->data->what, count + n, U8);
4058 Newx(RExC_rxi->data->what, n, U8);
4059 RExC_rxi->data->count = count + n;
4060 Copy(s, RExC_rxi->data->what + count, n, U8);
4064 /*XXX: todo make this not included in a non debugging perl */
4065 #ifndef PERL_IN_XSUB_RE
4067 Perl_reginitcolors(pTHX)
4070 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4072 char *t = savepv(s);
4076 t = strchr(t, '\t');
4082 PL_colors[i] = t = (char *)"";
4087 PL_colors[i++] = (char *)"";
4094 #ifdef TRIE_STUDY_OPT
4095 #define CHECK_RESTUDY_GOTO \
4097 (data.flags & SCF_TRIE_RESTUDY) \
4101 #define CHECK_RESTUDY_GOTO
4105 - pregcomp - compile a regular expression into internal code
4107 * We can't allocate space until we know how big the compiled form will be,
4108 * but we can't compile it (and thus know how big it is) until we've got a
4109 * place to put the code. So we cheat: we compile it twice, once with code
4110 * generation turned off and size counting turned on, and once "for real".
4111 * This also means that we don't allocate space until we are sure that the
4112 * thing really will compile successfully, and we never have to move the
4113 * code and thus invalidate pointers into it. (Note that it has to be in
4114 * one piece because free() must be able to free it all.) [NB: not true in perl]
4116 * Beware that the optimization-preparation code in here knows about some
4117 * of the structure of the compiled regexp. [I'll say.]
4122 #ifndef PERL_IN_XSUB_RE
4123 #define RE_ENGINE_PTR &PL_core_reg_engine
4125 extern const struct regexp_engine my_reg_engine;
4126 #define RE_ENGINE_PTR &my_reg_engine
4129 #ifndef PERL_IN_XSUB_RE
4131 Perl_pregcomp(pTHX_ const SV * const pattern, const U32 flags)
4134 HV * const table = GvHV(PL_hintgv);
4135 /* Dispatch a request to compile a regexp to correct
4138 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4139 GET_RE_DEBUG_FLAGS_DECL;
4140 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4141 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4143 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4146 return CALLREGCOMP_ENG(eng, pattern, flags);
4149 return Perl_re_compile(aTHX_ pattern, flags);
4154 Perl_re_compile(pTHX_ const SV * const pattern, const U32 pm_flags)
4159 register regexp_internal *ri;
4161 char* exp = SvPV((SV*)pattern, plen);
4162 char* xend = exp + plen;
4169 RExC_state_t RExC_state;
4170 RExC_state_t * const pRExC_state = &RExC_state;
4171 #ifdef TRIE_STUDY_OPT
4173 RExC_state_t copyRExC_state;
4175 GET_RE_DEBUG_FLAGS_DECL;
4176 DEBUG_r(if (!PL_colorset) reginitcolors());
4178 RExC_utf8 = RExC_orig_utf8 = pm_flags & RXf_UTF8;
4181 SV *dsv= sv_newmortal();
4182 RE_PV_QUOTED_DECL(s, RExC_utf8,
4183 dsv, exp, plen, 60);
4184 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4185 PL_colors[4],PL_colors[5],s);
4190 RExC_flags = pm_flags;
4194 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4195 RExC_seen_evals = 0;
4198 /* First pass: determine size, legality. */
4206 RExC_emit = &PL_regdummy;
4207 RExC_whilem_seen = 0;
4208 RExC_charnames = NULL;
4209 RExC_open_parens = NULL;
4210 RExC_close_parens = NULL;
4212 RExC_paren_names = NULL;
4214 RExC_paren_name_list = NULL;
4216 RExC_recurse = NULL;
4217 RExC_recurse_count = 0;
4219 #if 0 /* REGC() is (currently) a NOP at the first pass.
4220 * Clever compilers notice this and complain. --jhi */
4221 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4223 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4224 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4225 RExC_precomp = NULL;
4228 if (RExC_utf8 && !RExC_orig_utf8) {
4229 /* It's possible to write a regexp in ascii that represents Unicode
4230 codepoints outside of the byte range, such as via \x{100}. If we
4231 detect such a sequence we have to convert the entire pattern to utf8
4232 and then recompile, as our sizing calculation will have been based
4233 on 1 byte == 1 character, but we will need to use utf8 to encode
4234 at least some part of the pattern, and therefore must convert the whole
4236 XXX: somehow figure out how to make this less expensive...
4239 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4240 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4241 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4243 RExC_orig_utf8 = RExC_utf8;
4245 goto redo_first_pass;
4248 PerlIO_printf(Perl_debug_log,
4249 "Required size %"IVdf" nodes\n"
4250 "Starting second pass (creation)\n",
4253 RExC_lastparse=NULL;
4255 /* Small enough for pointer-storage convention?
4256 If extralen==0, this means that we will not need long jumps. */
4257 if (RExC_size >= 0x10000L && RExC_extralen)
4258 RExC_size += RExC_extralen;
4261 if (RExC_whilem_seen > 15)
4262 RExC_whilem_seen = 15;
4264 /* Allocate space and zero-initialize. Note, the two step process
4265 of zeroing when in debug mode, thus anything assigned has to
4266 happen after that */
4267 rx = newSV_type(SVt_REGEXP);
4268 r = (struct regexp*)SvANY(rx);
4269 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4270 char, regexp_internal);
4271 if ( r == NULL || ri == NULL )
4272 FAIL("Regexp out of space");
4274 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4275 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4277 /* bulk initialize base fields with 0. */
4278 Zero(ri, sizeof(regexp_internal), char);
4281 /* non-zero initialization begins here */
4283 r->engine= RE_ENGINE_PTR;
4284 r->extflags = pm_flags;
4286 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4287 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4288 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4289 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4290 >> RXf_PMf_STD_PMMOD_SHIFT);
4291 const char *fptr = STD_PAT_MODS; /*"msix"*/
4293 const STRLEN wraplen = plen + has_minus + has_p + has_runon
4294 + (sizeof(STD_PAT_MODS) - 1)
4295 + (sizeof("(?:)") - 1);
4297 p = sv_grow(rx, wraplen + 1);
4298 SvCUR_set(rx, wraplen);
4302 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4304 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4305 char *colon = r + 1;
4308 while((ch = *fptr++)) {
4322 Copy(RExC_precomp, p, plen, char);
4323 assert ((RX_WRAPPED(rx) - p) < 16);
4324 r->pre_prefix = p - RX_WRAPPED(rx);
4333 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4335 if (RExC_seen & REG_SEEN_RECURSE) {
4336 Newxz(RExC_open_parens, RExC_npar,regnode *);
4337 SAVEFREEPV(RExC_open_parens);
4338 Newxz(RExC_close_parens,RExC_npar,regnode *);
4339 SAVEFREEPV(RExC_close_parens);
4342 /* Useful during FAIL. */
4343 #ifdef RE_TRACK_PATTERN_OFFSETS
4344 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4345 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4346 "%s %"UVuf" bytes for offset annotations.\n",
4347 ri->u.offsets ? "Got" : "Couldn't get",
4348 (UV)((2*RExC_size+1) * sizeof(U32))));
4350 SetProgLen(ri,RExC_size);
4355 /* Second pass: emit code. */
4356 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4361 RExC_emit_start = ri->program;
4362 RExC_emit = ri->program;
4363 RExC_emit_bound = ri->program + RExC_size + 1;
4365 /* Store the count of eval-groups for security checks: */
4366 RExC_rx->seen_evals = RExC_seen_evals;
4367 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4368 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4372 /* XXXX To minimize changes to RE engine we always allocate
4373 3-units-long substrs field. */
4374 Newx(r->substrs, 1, struct reg_substr_data);
4375 if (RExC_recurse_count) {
4376 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4377 SAVEFREEPV(RExC_recurse);
4381 r->minlen = minlen = sawplus = sawopen = 0;
4382 Zero(r->substrs, 1, struct reg_substr_data);
4384 #ifdef TRIE_STUDY_OPT
4387 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4389 RExC_state = copyRExC_state;
4390 if (seen & REG_TOP_LEVEL_BRANCHES)
4391 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4393 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4394 if (data.last_found) {
4395 SvREFCNT_dec(data.longest_fixed);
4396 SvREFCNT_dec(data.longest_float);
4397 SvREFCNT_dec(data.last_found);
4399 StructCopy(&zero_scan_data, &data, scan_data_t);
4401 StructCopy(&zero_scan_data, &data, scan_data_t);
4402 copyRExC_state = RExC_state;
4405 StructCopy(&zero_scan_data, &data, scan_data_t);
4408 /* Dig out information for optimizations. */
4409 r->extflags = RExC_flags; /* was pm_op */
4410 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4413 r->extflags |= RXf_UTF8; /* Unicode in it? */
4414 ri->regstclass = NULL;
4415 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4416 r->intflags |= PREGf_NAUGHTY;
4417 scan = ri->program + 1; /* First BRANCH. */
4419 /* testing for BRANCH here tells us whether there is "must appear"
4420 data in the pattern. If there is then we can use it for optimisations */
4421 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4423 STRLEN longest_float_length, longest_fixed_length;
4424 struct regnode_charclass_class ch_class; /* pointed to by data */
4426 I32 last_close = 0; /* pointed to by data */
4427 regnode *first= scan;
4428 regnode *first_next= regnext(first);
4430 /* Skip introductions and multiplicators >= 1. */
4431 while ((OP(first) == OPEN && (sawopen = 1)) ||
4432 /* An OR of *one* alternative - should not happen now. */
4433 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4434 /* for now we can't handle lookbehind IFMATCH*/
4435 (OP(first) == IFMATCH && !first->flags) ||
4436 (OP(first) == PLUS) ||
4437 (OP(first) == MINMOD) ||
4438 /* An {n,m} with n>0 */
4439 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4440 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4443 if (OP(first) == PLUS)
4446 first += regarglen[OP(first)];
4447 if (OP(first) == IFMATCH) {
4448 first = NEXTOPER(first);
4449 first += EXTRA_STEP_2ARGS;
4450 } else /* XXX possible optimisation for /(?=)/ */
4451 first = NEXTOPER(first);
4452 first_next= regnext(first);
4455 /* Starting-point info. */
4457 DEBUG_PEEP("first:",first,0);
4458 /* Ignore EXACT as we deal with it later. */
4459 if (PL_regkind[OP(first)] == EXACT) {
4460 if (OP(first) == EXACT)
4461 NOOP; /* Empty, get anchored substr later. */
4462 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4463 ri->regstclass = first;
4466 else if (PL_regkind[OP(first)] == TRIE &&
4467 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4470 /* this can happen only on restudy */
4471 if ( OP(first) == TRIE ) {
4472 struct regnode_1 *trieop = (struct regnode_1 *)
4473 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4474 StructCopy(first,trieop,struct regnode_1);
4475 trie_op=(regnode *)trieop;
4477 struct regnode_charclass *trieop = (struct regnode_charclass *)
4478 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4479 StructCopy(first,trieop,struct regnode_charclass);
4480 trie_op=(regnode *)trieop;
4483 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4484 ri->regstclass = trie_op;
4487 else if (strchr((const char*)PL_simple,OP(first)))
4488 ri->regstclass = first;
4489 else if (PL_regkind[OP(first)] == BOUND ||
4490 PL_regkind[OP(first)] == NBOUND)
4491 ri->regstclass = first;
4492 else if (PL_regkind[OP(first)] == BOL) {
4493 r->extflags |= (OP(first) == MBOL
4495 : (OP(first) == SBOL
4498 first = NEXTOPER(first);
4501 else if (OP(first) == GPOS) {
4502 r->extflags |= RXf_ANCH_GPOS;
4503 first = NEXTOPER(first);
4506 else if ((!sawopen || !RExC_sawback) &&
4507 (OP(first) == STAR &&
4508 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4509 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4511 /* turn .* into ^.* with an implied $*=1 */
4513 (OP(NEXTOPER(first)) == REG_ANY)
4516 r->extflags |= type;
4517 r->intflags |= PREGf_IMPLICIT;
4518 first = NEXTOPER(first);
4521 if (sawplus && (!sawopen || !RExC_sawback)
4522 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4523 /* x+ must match at the 1st pos of run of x's */
4524 r->intflags |= PREGf_SKIP;
4526 /* Scan is after the zeroth branch, first is atomic matcher. */
4527 #ifdef TRIE_STUDY_OPT
4530 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4531 (IV)(first - scan + 1))
4535 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4536 (IV)(first - scan + 1))
4542 * If there's something expensive in the r.e., find the
4543 * longest literal string that must appear and make it the
4544 * regmust. Resolve ties in favor of later strings, since
4545 * the regstart check works with the beginning of the r.e.
4546 * and avoiding duplication strengthens checking. Not a
4547 * strong reason, but sufficient in the absence of others.
4548 * [Now we resolve ties in favor of the earlier string if
4549 * it happens that c_offset_min has been invalidated, since the
4550 * earlier string may buy us something the later one won't.]
4553 data.longest_fixed = newSVpvs("");
4554 data.longest_float = newSVpvs("");
4555 data.last_found = newSVpvs("");
4556 data.longest = &(data.longest_fixed);
4558 if (!ri->regstclass) {
4559 cl_init(pRExC_state, &ch_class);
4560 data.start_class = &ch_class;
4561 stclass_flag = SCF_DO_STCLASS_AND;
4562 } else /* XXXX Check for BOUND? */
4564 data.last_closep = &last_close;
4566 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4567 &data, -1, NULL, NULL,
4568 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4574 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4575 && data.last_start_min == 0 && data.last_end > 0
4576 && !RExC_seen_zerolen
4577 && !(RExC_seen & REG_SEEN_VERBARG)
4578 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4579 r->extflags |= RXf_CHECK_ALL;
4580 scan_commit(pRExC_state, &data,&minlen,0);
4581 SvREFCNT_dec(data.last_found);
4583 /* Note that code very similar to this but for anchored string
4584 follows immediately below, changes may need to be made to both.
4587 longest_float_length = CHR_SVLEN(data.longest_float);
4588 if (longest_float_length
4589 || (data.flags & SF_FL_BEFORE_EOL
4590 && (!(data.flags & SF_FL_BEFORE_MEOL)
4591 || (RExC_flags & RXf_PMf_MULTILINE))))
4595 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4596 && data.offset_fixed == data.offset_float_min
4597 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4598 goto remove_float; /* As in (a)+. */
4600 /* copy the information about the longest float from the reg_scan_data
4601 over to the program. */
4602 if (SvUTF8(data.longest_float)) {
4603 r->float_utf8 = data.longest_float;
4604 r->float_substr = NULL;
4606 r->float_substr = data.longest_float;
4607 r->float_utf8 = NULL;
4609 /* float_end_shift is how many chars that must be matched that
4610 follow this item. We calculate it ahead of time as once the
4611 lookbehind offset is added in we lose the ability to correctly
4613 ml = data.minlen_float ? *(data.minlen_float)
4614 : (I32)longest_float_length;
4615 r->float_end_shift = ml - data.offset_float_min
4616 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4617 + data.lookbehind_float;
4618 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4619 r->float_max_offset = data.offset_float_max;
4620 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4621 r->float_max_offset -= data.lookbehind_float;
4623 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4624 && (!(data.flags & SF_FL_BEFORE_MEOL)
4625 || (RExC_flags & RXf_PMf_MULTILINE)));
4626 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4630 r->float_substr = r->float_utf8 = NULL;
4631 SvREFCNT_dec(data.longest_float);
4632 longest_float_length = 0;
4635 /* Note that code very similar to this but for floating string
4636 is immediately above, changes may need to be made to both.
4639 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4640 if (longest_fixed_length
4641 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4642 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4643 || (RExC_flags & RXf_PMf_MULTILINE))))
4647 /* copy the information about the longest fixed
4648 from the reg_scan_data over to the program. */
4649 if (SvUTF8(data.longest_fixed)) {
4650 r->anchored_utf8 = data.longest_fixed;
4651 r->anchored_substr = NULL;
4653 r->anchored_substr = data.longest_fixed;
4654 r->anchored_utf8 = NULL;
4656 /* fixed_end_shift is how many chars that must be matched that
4657 follow this item. We calculate it ahead of time as once the
4658 lookbehind offset is added in we lose the ability to correctly
4660 ml = data.minlen_fixed ? *(data.minlen_fixed)
4661 : (I32)longest_fixed_length;
4662 r->anchored_end_shift = ml - data.offset_fixed
4663 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4664 + data.lookbehind_fixed;
4665 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4667 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4668 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4669 || (RExC_flags & RXf_PMf_MULTILINE)));
4670 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4673 r->anchored_substr = r->anchored_utf8 = NULL;
4674 SvREFCNT_dec(data.longest_fixed);
4675 longest_fixed_length = 0;
4678 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4679 ri->regstclass = NULL;
4680 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4682 && !(data.start_class->flags & ANYOF_EOS)
4683 && !cl_is_anything(data.start_class))
4685 const U32 n = add_data(pRExC_state, 1, "f");
4687 Newx(RExC_rxi->data->data[n], 1,
4688 struct regnode_charclass_class);
4689 StructCopy(data.start_class,
4690 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4691 struct regnode_charclass_class);
4692 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4693 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4694 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4695 regprop(r, sv, (regnode*)data.start_class);
4696 PerlIO_printf(Perl_debug_log,
4697 "synthetic stclass \"%s\".\n",
4698 SvPVX_const(sv));});
4701 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4702 if (longest_fixed_length > longest_float_length) {
4703 r->check_end_shift = r->anchored_end_shift;
4704 r->check_substr = r->anchored_substr;
4705 r->check_utf8 = r->anchored_utf8;
4706 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4707 if (r->extflags & RXf_ANCH_SINGLE)
4708 r->extflags |= RXf_NOSCAN;
4711 r->check_end_shift = r->float_end_shift;
4712 r->check_substr = r->float_substr;
4713 r->check_utf8 = r->float_utf8;
4714 r->check_offset_min = r->float_min_offset;
4715 r->check_offset_max = r->float_max_offset;
4717 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4718 This should be changed ASAP! */
4719 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4720 r->extflags |= RXf_USE_INTUIT;
4721 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4722 r->extflags |= RXf_INTUIT_TAIL;
4724 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4725 if ( (STRLEN)minlen < longest_float_length )
4726 minlen= longest_float_length;
4727 if ( (STRLEN)minlen < longest_fixed_length )
4728 minlen= longest_fixed_length;
4732 /* Several toplevels. Best we can is to set minlen. */
4734 struct regnode_charclass_class ch_class;
4737 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4739 scan = ri->program + 1;
4740 cl_init(pRExC_state, &ch_class);
4741 data.start_class = &ch_class;
4742 data.last_closep = &last_close;
4745 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4746 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4750 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4751 = r->float_substr = r->float_utf8 = NULL;
4752 if (!(data.start_class->flags & ANYOF_EOS)
4753 && !cl_is_anything(data.start_class))
4755 const U32 n = add_data(pRExC_state, 1, "f");
4757 Newx(RExC_rxi->data->data[n], 1,
4758 struct regnode_charclass_class);
4759 StructCopy(data.start_class,
4760 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4761 struct regnode_charclass_class);
4762 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4763 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4764 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4765 regprop(r, sv, (regnode*)data.start_class);
4766 PerlIO_printf(Perl_debug_log,
4767 "synthetic stclass \"%s\".\n",
4768 SvPVX_const(sv));});
4772 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4773 the "real" pattern. */
4775 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4776 (IV)minlen, (IV)r->minlen);
4778 r->minlenret = minlen;
4779 if (r->minlen < minlen)
4782 if (RExC_seen & REG_SEEN_GPOS)
4783 r->extflags |= RXf_GPOS_SEEN;
4784 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4785 r->extflags |= RXf_LOOKBEHIND_SEEN;
4786 if (RExC_seen & REG_SEEN_EVAL)
4787 r->extflags |= RXf_EVAL_SEEN;
4788 if (RExC_seen & REG_SEEN_CANY)
4789 r->extflags |= RXf_CANY_SEEN;
4790 if (RExC_seen & REG_SEEN_VERBARG)
4791 r->intflags |= PREGf_VERBARG_SEEN;
4792 if (RExC_seen & REG_SEEN_CUTGROUP)
4793 r->intflags |= PREGf_CUTGROUP_SEEN;
4794 if (RExC_paren_names)
4795 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4797 r->paren_names = NULL;
4799 #ifdef STUPID_PATTERN_CHECKS
4800 if (RX_PRELEN(rx) == 0)
4801 r->extflags |= RXf_NULL;
4802 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4803 /* XXX: this should happen BEFORE we compile */
4804 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4805 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
4806 r->extflags |= RXf_WHITE;
4807 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
4808 r->extflags |= RXf_START_ONLY;
4810 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4811 /* XXX: this should happen BEFORE we compile */
4812 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4814 regnode *first = ri->program + 1;
4816 U8 nop = OP(NEXTOPER(first));
4818 if (PL_regkind[fop] == NOTHING && nop == END)
4819 r->extflags |= RXf_NULL;
4820 else if (PL_regkind[fop] == BOL && nop == END)
4821 r->extflags |= RXf_START_ONLY;
4822 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4823 r->extflags |= RXf_WHITE;
4827 if (RExC_paren_names) {
4828 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4829 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4832 ri->name_list_idx = 0;
4834 if (RExC_recurse_count) {
4835 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4836 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4837 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4840 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4841 /* assume we don't need to swap parens around before we match */
4844 PerlIO_printf(Perl_debug_log,"Final program:\n");
4847 #ifdef RE_TRACK_PATTERN_OFFSETS
4848 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4849 const U32 len = ri->u.offsets[0];
4851 GET_RE_DEBUG_FLAGS_DECL;
4852 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4853 for (i = 1; i <= len; i++) {
4854 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4855 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4856 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4858 PerlIO_printf(Perl_debug_log, "\n");
4864 #undef RE_ENGINE_PTR
4868 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4871 PERL_UNUSED_ARG(value);
4873 if (flags & RXapif_FETCH) {
4874 return reg_named_buff_fetch(rx, key, flags);
4875 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
4876 Perl_croak(aTHX_ PL_no_modify);
4878 } else if (flags & RXapif_EXISTS) {
4879 return reg_named_buff_exists(rx, key, flags)
4882 } else if (flags & RXapif_REGNAMES) {
4883 return reg_named_buff_all(rx, flags);
4884 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
4885 return reg_named_buff_scalar(rx, flags);
4887 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
4893 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
4896 PERL_UNUSED_ARG(lastkey);
4898 if (flags & RXapif_FIRSTKEY)
4899 return reg_named_buff_firstkey(rx, flags);
4900 else if (flags & RXapif_NEXTKEY)
4901 return reg_named_buff_nextkey(rx, flags);
4903 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
4909 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
4912 AV *retarray = NULL;
4914 struct regexp *const rx = (struct regexp *)SvANY(r);
4915 if (flags & RXapif_ALL)
4918 if (rx && rx->paren_names) {
4919 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4922 SV* sv_dat=HeVAL(he_str);
4923 I32 *nums=(I32*)SvPVX(sv_dat);
4924 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4925 if ((I32)(rx->nparens) >= nums[i]
4926 && rx->offs[nums[i]].start != -1
4927 && rx->offs[nums[i]].end != -1)
4930 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
4934 ret = newSVsv(&PL_sv_undef);
4937 SvREFCNT_inc_simple_void(ret);
4938 av_push(retarray, ret);
4942 return newRV((SV*)retarray);
4949 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
4952 struct regexp *const rx = (struct regexp *)SvANY(r);
4953 if (rx && rx->paren_names) {
4954 if (flags & RXapif_ALL) {
4955 return hv_exists_ent(rx->paren_names, key, 0);
4957 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
4971 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
4973 struct regexp *const rx = (struct regexp *)SvANY(r);
4974 if ( rx && rx->paren_names ) {
4975 (void)hv_iterinit(rx->paren_names);
4977 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
4984 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
4986 struct regexp *const rx = (struct regexp *)SvANY(r);
4987 if (rx && rx->paren_names) {
4988 HV *hv = rx->paren_names;
4990 while ( (temphe = hv_iternext_flags(hv,0)) ) {
4993 SV* sv_dat = HeVAL(temphe);
4994 I32 *nums = (I32*)SvPVX(sv_dat);
4995 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
4996 if ((I32)(rx->lastcloseparen) >= nums[i] &&
4997 rx->offs[nums[i]].start != -1 &&
4998 rx->offs[nums[i]].end != -1)
5004 if (parno || flags & RXapif_ALL) {
5005 return newSVhek(HeKEY_hek(temphe));
5013 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5018 struct regexp *const rx = (struct regexp *)SvANY(r);
5020 if (rx && rx->paren_names) {
5021 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5022 return newSViv(HvTOTALKEYS(rx->paren_names));
5023 } else if (flags & RXapif_ONE) {
5024 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5025 av = (AV*)SvRV(ret);
5026 length = av_len(av);
5027 return newSViv(length + 1);
5029 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5033 return &PL_sv_undef;
5037 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5039 struct regexp *const rx = (struct regexp *)SvANY(r);
5042 if (rx && rx->paren_names) {
5043 HV *hv= rx->paren_names;
5045 (void)hv_iterinit(hv);
5046 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5049 SV* sv_dat = HeVAL(temphe);
5050 I32 *nums = (I32*)SvPVX(sv_dat);
5051 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5052 if ((I32)(rx->lastcloseparen) >= nums[i] &&
5053 rx->offs[nums[i]].start != -1 &&
5054 rx->offs[nums[i]].end != -1)
5060 if (parno || flags & RXapif_ALL) {
5061 av_push(av, newSVhek(HeKEY_hek(temphe)));
5066 return newRV((SV*)av);
5070 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5073 struct regexp *const rx = (struct regexp *)SvANY(r);
5079 sv_setsv(sv,&PL_sv_undef);
5083 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5085 i = rx->offs[0].start;
5089 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5091 s = rx->subbeg + rx->offs[0].end;
5092 i = rx->sublen - rx->offs[0].end;
5095 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5096 (s1 = rx->offs[paren].start) != -1 &&
5097 (t1 = rx->offs[paren].end) != -1)
5101 s = rx->subbeg + s1;
5103 sv_setsv(sv,&PL_sv_undef);
5106 assert(rx->sublen >= (s - rx->subbeg) + i );
5108 const int oldtainted = PL_tainted;
5110 sv_setpvn(sv, s, i);
5111 PL_tainted = oldtainted;
5112 if ( (rx->extflags & RXf_CANY_SEEN)
5113 ? (RXp_MATCH_UTF8(rx)
5114 && (!i || is_utf8_string((U8*)s, i)))
5115 : (RXp_MATCH_UTF8(rx)) )
5122 if (RXp_MATCH_TAINTED(rx)) {
5123 if (SvTYPE(sv) >= SVt_PVMG) {
5124 MAGIC* const mg = SvMAGIC(sv);
5127 SvMAGIC_set(sv, mg->mg_moremagic);
5129 if ((mgt = SvMAGIC(sv))) {
5130 mg->mg_moremagic = mgt;
5131 SvMAGIC_set(sv, mg);
5141 sv_setsv(sv,&PL_sv_undef);
5147 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5148 SV const * const value)
5150 PERL_UNUSED_ARG(rx);
5151 PERL_UNUSED_ARG(paren);
5152 PERL_UNUSED_ARG(value);
5155 Perl_croak(aTHX_ PL_no_modify);
5159 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5162 struct regexp *const rx = (struct regexp *)SvANY(r);
5166 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5168 /* $` / ${^PREMATCH} */
5169 case RX_BUFF_IDX_PREMATCH:
5170 if (rx->offs[0].start != -1) {
5171 i = rx->offs[0].start;
5179 /* $' / ${^POSTMATCH} */
5180 case RX_BUFF_IDX_POSTMATCH:
5181 if (rx->offs[0].end != -1) {
5182 i = rx->sublen - rx->offs[0].end;
5184 s1 = rx->offs[0].end;
5190 /* $& / ${^MATCH}, $1, $2, ... */
5192 if (paren <= (I32)rx->nparens &&
5193 (s1 = rx->offs[paren].start) != -1 &&
5194 (t1 = rx->offs[paren].end) != -1)
5199 if (ckWARN(WARN_UNINITIALIZED))
5200 report_uninit((SV*)sv);
5205 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5206 const char * const s = rx->subbeg + s1;
5211 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5218 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5220 PERL_UNUSED_ARG(rx);
5224 /* Scans the name of a named buffer from the pattern.
5225 * If flags is REG_RSN_RETURN_NULL returns null.
5226 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5227 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5228 * to the parsed name as looked up in the RExC_paren_names hash.
5229 * If there is an error throws a vFAIL().. type exception.
5232 #define REG_RSN_RETURN_NULL 0
5233 #define REG_RSN_RETURN_NAME 1
5234 #define REG_RSN_RETURN_DATA 2
5237 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
5238 char *name_start = RExC_parse;
5240 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5241 /* skip IDFIRST by using do...while */
5244 RExC_parse += UTF8SKIP(RExC_parse);
5245 } while (isALNUM_utf8((U8*)RExC_parse));
5249 } while (isALNUM(*RExC_parse));
5254 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5255 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5256 if ( flags == REG_RSN_RETURN_NAME)
5258 else if (flags==REG_RSN_RETURN_DATA) {
5261 if ( ! sv_name ) /* should not happen*/
5262 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5263 if (RExC_paren_names)
5264 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5266 sv_dat = HeVAL(he_str);
5268 vFAIL("Reference to nonexistent named group");
5272 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5279 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5280 int rem=(int)(RExC_end - RExC_parse); \
5289 if (RExC_lastparse!=RExC_parse) \
5290 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5293 iscut ? "..." : "<" \
5296 PerlIO_printf(Perl_debug_log,"%16s",""); \
5299 num = RExC_size + 1; \
5301 num=REG_NODE_NUM(RExC_emit); \
5302 if (RExC_lastnum!=num) \
5303 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5305 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5306 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5307 (int)((depth*2)), "", \
5311 RExC_lastparse=RExC_parse; \
5316 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5317 DEBUG_PARSE_MSG((funcname)); \
5318 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5320 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5321 DEBUG_PARSE_MSG((funcname)); \
5322 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5325 - reg - regular expression, i.e. main body or parenthesized thing
5327 * Caller must absorb opening parenthesis.
5329 * Combining parenthesis handling with the base level of regular expression
5330 * is a trifle forced, but the need to tie the tails of the branches to what
5331 * follows makes it hard to avoid.
5333 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5335 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5337 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5341 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5342 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5345 register regnode *ret; /* Will be the head of the group. */
5346 register regnode *br;
5347 register regnode *lastbr;
5348 register regnode *ender = NULL;
5349 register I32 parno = 0;
5351 U32 oregflags = RExC_flags;
5352 bool have_branch = 0;
5354 I32 freeze_paren = 0;
5355 I32 after_freeze = 0;
5357 /* for (?g), (?gc), and (?o) warnings; warning
5358 about (?c) will warn about (?g) -- japhy */
5360 #define WASTED_O 0x01
5361 #define WASTED_G 0x02
5362 #define WASTED_C 0x04
5363 #define WASTED_GC (0x02|0x04)
5364 I32 wastedflags = 0x00;
5366 char * parse_start = RExC_parse; /* MJD */
5367 char * const oregcomp_parse = RExC_parse;
5369 GET_RE_DEBUG_FLAGS_DECL;
5370 DEBUG_PARSE("reg ");
5372 *flagp = 0; /* Tentatively. */
5375 /* Make an OPEN node, if parenthesized. */
5377 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5378 char *start_verb = RExC_parse;
5379 STRLEN verb_len = 0;
5380 char *start_arg = NULL;
5381 unsigned char op = 0;
5383 int internal_argval = 0; /* internal_argval is only useful if !argok */
5384 while ( *RExC_parse && *RExC_parse != ')' ) {
5385 if ( *RExC_parse == ':' ) {
5386 start_arg = RExC_parse + 1;
5392 verb_len = RExC_parse - start_verb;
5395 while ( *RExC_parse && *RExC_parse != ')' )
5397 if ( *RExC_parse != ')' )
5398 vFAIL("Unterminated verb pattern argument");
5399 if ( RExC_parse == start_arg )
5402 if ( *RExC_parse != ')' )
5403 vFAIL("Unterminated verb pattern");
5406 switch ( *start_verb ) {
5407 case 'A': /* (*ACCEPT) */
5408 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5410 internal_argval = RExC_nestroot;
5413 case 'C': /* (*COMMIT) */
5414 if ( memEQs(start_verb,verb_len,"COMMIT") )
5417 case 'F': /* (*FAIL) */
5418 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5423 case ':': /* (*:NAME) */
5424 case 'M': /* (*MARK:NAME) */
5425 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5430 case 'P': /* (*PRUNE) */
5431 if ( memEQs(start_verb,verb_len,"PRUNE") )
5434 case 'S': /* (*SKIP) */
5435 if ( memEQs(start_verb,verb_len,"SKIP") )
5438 case 'T': /* (*THEN) */
5439 /* [19:06] <TimToady> :: is then */
5440 if ( memEQs(start_verb,verb_len,"THEN") ) {
5442 RExC_seen |= REG_SEEN_CUTGROUP;
5448 vFAIL3("Unknown verb pattern '%.*s'",
5449 verb_len, start_verb);
5452 if ( start_arg && internal_argval ) {
5453 vFAIL3("Verb pattern '%.*s' may not have an argument",
5454 verb_len, start_verb);
5455 } else if ( argok < 0 && !start_arg ) {
5456 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5457 verb_len, start_verb);
5459 ret = reganode(pRExC_state, op, internal_argval);
5460 if ( ! internal_argval && ! SIZE_ONLY ) {
5462 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5463 ARG(ret) = add_data( pRExC_state, 1, "S" );
5464 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5471 if (!internal_argval)
5472 RExC_seen |= REG_SEEN_VERBARG;
5473 } else if ( start_arg ) {
5474 vFAIL3("Verb pattern '%.*s' may not have an argument",
5475 verb_len, start_verb);
5477 ret = reg_node(pRExC_state, op);
5479 nextchar(pRExC_state);
5482 if (*RExC_parse == '?') { /* (?...) */
5483 bool is_logical = 0;
5484 const char * const seqstart = RExC_parse;
5487 paren = *RExC_parse++;
5488 ret = NULL; /* For look-ahead/behind. */
5491 case 'P': /* (?P...) variants for those used to PCRE/Python */
5492 paren = *RExC_parse++;
5493 if ( paren == '<') /* (?P<...>) named capture */
5495 else if (paren == '>') { /* (?P>name) named recursion */
5496 goto named_recursion;
5498 else if (paren == '=') { /* (?P=...) named backref */
5499 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5500 you change this make sure you change that */
5501 char* name_start = RExC_parse;
5503 SV *sv_dat = reg_scan_name(pRExC_state,
5504 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5505 if (RExC_parse == name_start || *RExC_parse != ')')
5506 vFAIL2("Sequence %.3s... not terminated",parse_start);
5509 num = add_data( pRExC_state, 1, "S" );
5510 RExC_rxi->data->data[num]=(void*)sv_dat;
5511 SvREFCNT_inc_simple_void(sv_dat);
5514 ret = reganode(pRExC_state,
5515 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5519 Set_Node_Offset(ret, parse_start+1);
5520 Set_Node_Cur_Length(ret); /* MJD */
5522 nextchar(pRExC_state);
5526 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5528 case '<': /* (?<...) */
5529 if (*RExC_parse == '!')
5531 else if (*RExC_parse != '=')
5537 case '\'': /* (?'...') */
5538 name_start= RExC_parse;
5539 svname = reg_scan_name(pRExC_state,
5540 SIZE_ONLY ? /* reverse test from the others */
5541 REG_RSN_RETURN_NAME :
5542 REG_RSN_RETURN_NULL);
5543 if (RExC_parse == name_start) {
5545 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5548 if (*RExC_parse != paren)
5549 vFAIL2("Sequence (?%c... not terminated",
5550 paren=='>' ? '<' : paren);
5554 if (!svname) /* shouldnt happen */
5556 "panic: reg_scan_name returned NULL");
5557 if (!RExC_paren_names) {
5558 RExC_paren_names= newHV();
5559 sv_2mortal((SV*)RExC_paren_names);
5561 RExC_paren_name_list= newAV();
5562 sv_2mortal((SV*)RExC_paren_name_list);
5565 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5567 sv_dat = HeVAL(he_str);
5569 /* croak baby croak */
5571 "panic: paren_name hash element allocation failed");
5572 } else if ( SvPOK(sv_dat) ) {
5573 /* (?|...) can mean we have dupes so scan to check
5574 its already been stored. Maybe a flag indicating
5575 we are inside such a construct would be useful,
5576 but the arrays are likely to be quite small, so
5577 for now we punt -- dmq */
5578 IV count = SvIV(sv_dat);
5579 I32 *pv = (I32*)SvPVX(sv_dat);
5581 for ( i = 0 ; i < count ; i++ ) {
5582 if ( pv[i] == RExC_npar ) {
5588 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5589 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5590 pv[count] = RExC_npar;
5594 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5595 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5600 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5601 SvREFCNT_dec(svname);
5604 /*sv_dump(sv_dat);*/
5606 nextchar(pRExC_state);
5608 goto capturing_parens;
5610 RExC_seen |= REG_SEEN_LOOKBEHIND;
5612 case '=': /* (?=...) */
5613 case '!': /* (?!...) */
5614 RExC_seen_zerolen++;
5615 if (*RExC_parse == ')') {
5616 ret=reg_node(pRExC_state, OPFAIL);
5617 nextchar(pRExC_state);
5621 case '|': /* (?|...) */
5622 /* branch reset, behave like a (?:...) except that
5623 buffers in alternations share the same numbers */
5625 after_freeze = freeze_paren = RExC_npar;
5627 case ':': /* (?:...) */
5628 case '>': /* (?>...) */
5630 case '$': /* (?$...) */
5631 case '@': /* (?@...) */
5632 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5634 case '#': /* (?#...) */
5635 while (*RExC_parse && *RExC_parse != ')')
5637 if (*RExC_parse != ')')
5638 FAIL("Sequence (?#... not terminated");
5639 nextchar(pRExC_state);
5642 case '0' : /* (?0) */
5643 case 'R' : /* (?R) */
5644 if (*RExC_parse != ')')
5645 FAIL("Sequence (?R) not terminated");
5646 ret = reg_node(pRExC_state, GOSTART);
5647 *flagp |= POSTPONED;
5648 nextchar(pRExC_state);
5651 { /* named and numeric backreferences */
5653 case '&': /* (?&NAME) */
5654 parse_start = RExC_parse - 1;
5657 SV *sv_dat = reg_scan_name(pRExC_state,
5658 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5659 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5661 goto gen_recurse_regop;
5664 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5666 vFAIL("Illegal pattern");
5668 goto parse_recursion;
5670 case '-': /* (?-1) */
5671 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5672 RExC_parse--; /* rewind to let it be handled later */
5676 case '1': case '2': case '3': case '4': /* (?1) */
5677 case '5': case '6': case '7': case '8': case '9':
5680 num = atoi(RExC_parse);
5681 parse_start = RExC_parse - 1; /* MJD */
5682 if (*RExC_parse == '-')
5684 while (isDIGIT(*RExC_parse))
5686 if (*RExC_parse!=')')
5687 vFAIL("Expecting close bracket");
5690 if ( paren == '-' ) {
5692 Diagram of capture buffer numbering.
5693 Top line is the normal capture buffer numbers
5694 Botton line is the negative indexing as from
5698 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5702 num = RExC_npar + num;
5705 vFAIL("Reference to nonexistent group");
5707 } else if ( paren == '+' ) {
5708 num = RExC_npar + num - 1;
5711 ret = reganode(pRExC_state, GOSUB, num);
5713 if (num > (I32)RExC_rx->nparens) {
5715 vFAIL("Reference to nonexistent group");
5717 ARG2L_SET( ret, RExC_recurse_count++);
5719 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5720 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5724 RExC_seen |= REG_SEEN_RECURSE;
5725 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5726 Set_Node_Offset(ret, parse_start); /* MJD */
5728 *flagp |= POSTPONED;
5729 nextchar(pRExC_state);
5731 } /* named and numeric backreferences */
5734 case '?': /* (??...) */
5736 if (*RExC_parse != '{') {
5738 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5741 *flagp |= POSTPONED;
5742 paren = *RExC_parse++;
5744 case '{': /* (?{...}) */
5749 char *s = RExC_parse;
5751 RExC_seen_zerolen++;
5752 RExC_seen |= REG_SEEN_EVAL;
5753 while (count && (c = *RExC_parse)) {
5764 if (*RExC_parse != ')') {
5766 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5770 OP_4tree *sop, *rop;
5771 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5774 Perl_save_re_context(aTHX);
5775 rop = sv_compile_2op(sv, &sop, "re", &pad);
5776 sop->op_private |= OPpREFCOUNTED;
5777 /* re_dup will OpREFCNT_inc */
5778 OpREFCNT_set(sop, 1);
5781 n = add_data(pRExC_state, 3, "nop");
5782 RExC_rxi->data->data[n] = (void*)rop;
5783 RExC_rxi->data->data[n+1] = (void*)sop;
5784 RExC_rxi->data->data[n+2] = (void*)pad;
5787 else { /* First pass */
5788 if (PL_reginterp_cnt < ++RExC_seen_evals
5790 /* No compiled RE interpolated, has runtime
5791 components ===> unsafe. */
5792 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5793 if (PL_tainting && PL_tainted)
5794 FAIL("Eval-group in insecure regular expression");
5795 #if PERL_VERSION > 8
5796 if (IN_PERL_COMPILETIME)
5801 nextchar(pRExC_state);
5803 ret = reg_node(pRExC_state, LOGICAL);
5806 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5807 /* deal with the length of this later - MJD */
5810 ret = reganode(pRExC_state, EVAL, n);
5811 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5812 Set_Node_Offset(ret, parse_start);
5815 case '(': /* (?(?{...})...) and (?(?=...)...) */
5818 if (RExC_parse[0] == '?') { /* (?(?...)) */
5819 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5820 || RExC_parse[1] == '<'
5821 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5824 ret = reg_node(pRExC_state, LOGICAL);
5827 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5831 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5832 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5834 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5835 char *name_start= RExC_parse++;
5837 SV *sv_dat=reg_scan_name(pRExC_state,
5838 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5839 if (RExC_parse == name_start || *RExC_parse != ch)
5840 vFAIL2("Sequence (?(%c... not terminated",
5841 (ch == '>' ? '<' : ch));
5844 num = add_data( pRExC_state, 1, "S" );
5845 RExC_rxi->data->data[num]=(void*)sv_dat;
5846 SvREFCNT_inc_simple_void(sv_dat);
5848 ret = reganode(pRExC_state,NGROUPP,num);
5849 goto insert_if_check_paren;
5851 else if (RExC_parse[0] == 'D' &&
5852 RExC_parse[1] == 'E' &&
5853 RExC_parse[2] == 'F' &&
5854 RExC_parse[3] == 'I' &&
5855 RExC_parse[4] == 'N' &&
5856 RExC_parse[5] == 'E')
5858 ret = reganode(pRExC_state,DEFINEP,0);
5861 goto insert_if_check_paren;
5863 else if (RExC_parse[0] == 'R') {
5866 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5867 parno = atoi(RExC_parse++);
5868 while (isDIGIT(*RExC_parse))
5870 } else if (RExC_parse[0] == '&') {
5873 sv_dat = reg_scan_name(pRExC_state,
5874 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5875 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5877 ret = reganode(pRExC_state,INSUBP,parno);
5878 goto insert_if_check_paren;
5880 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5883 parno = atoi(RExC_parse++);
5885 while (isDIGIT(*RExC_parse))
5887 ret = reganode(pRExC_state, GROUPP, parno);
5889 insert_if_check_paren:
5890 if ((c = *nextchar(pRExC_state)) != ')')
5891 vFAIL("Switch condition not recognized");
5893 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5894 br = regbranch(pRExC_state, &flags, 1,depth+1);
5896 br = reganode(pRExC_state, LONGJMP, 0);
5898 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5899 c = *nextchar(pRExC_state);
5904 vFAIL("(?(DEFINE)....) does not allow branches");
5905 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5906 regbranch(pRExC_state, &flags, 1,depth+1);
5907 REGTAIL(pRExC_state, ret, lastbr);
5910 c = *nextchar(pRExC_state);
5915 vFAIL("Switch (?(condition)... contains too many branches");
5916 ender = reg_node(pRExC_state, TAIL);
5917 REGTAIL(pRExC_state, br, ender);
5919 REGTAIL(pRExC_state, lastbr, ender);
5920 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5923 REGTAIL(pRExC_state, ret, ender);
5924 RExC_size++; /* XXX WHY do we need this?!!
5925 For large programs it seems to be required
5926 but I can't figure out why. -- dmq*/
5930 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5934 RExC_parse--; /* for vFAIL to print correctly */
5935 vFAIL("Sequence (? incomplete");
5939 parse_flags: /* (?i) */
5941 U32 posflags = 0, negflags = 0;
5942 U32 *flagsp = &posflags;
5944 while (*RExC_parse) {
5945 /* && strchr("iogcmsx", *RExC_parse) */
5946 /* (?g), (?gc) and (?o) are useless here
5947 and must be globally applied -- japhy */
5948 switch (*RExC_parse) {
5949 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5950 case ONCE_PAT_MOD: /* 'o' */
5951 case GLOBAL_PAT_MOD: /* 'g' */
5952 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5953 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5954 if (! (wastedflags & wflagbit) ) {
5955 wastedflags |= wflagbit;
5958 "Useless (%s%c) - %suse /%c modifier",
5959 flagsp == &negflags ? "?-" : "?",
5961 flagsp == &negflags ? "don't " : "",
5968 case CONTINUE_PAT_MOD: /* 'c' */
5969 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5970 if (! (wastedflags & WASTED_C) ) {
5971 wastedflags |= WASTED_GC;
5974 "Useless (%sc) - %suse /gc modifier",
5975 flagsp == &negflags ? "?-" : "?",
5976 flagsp == &negflags ? "don't " : ""
5981 case KEEPCOPY_PAT_MOD: /* 'p' */
5982 if (flagsp == &negflags) {
5983 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5984 vWARN(RExC_parse + 1,"Useless use of (?-p)");
5986 *flagsp |= RXf_PMf_KEEPCOPY;
5990 if (flagsp == &negflags) {
5992 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5996 wastedflags = 0; /* reset so (?g-c) warns twice */
6002 RExC_flags |= posflags;
6003 RExC_flags &= ~negflags;
6005 oregflags |= posflags;
6006 oregflags &= ~negflags;
6008 nextchar(pRExC_state);
6019 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6024 }} /* one for the default block, one for the switch */
6031 ret = reganode(pRExC_state, OPEN, parno);
6034 RExC_nestroot = parno;
6035 if (RExC_seen & REG_SEEN_RECURSE
6036 && !RExC_open_parens[parno-1])
6038 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6039 "Setting open paren #%"IVdf" to %d\n",
6040 (IV)parno, REG_NODE_NUM(ret)));
6041 RExC_open_parens[parno-1]= ret;
6044 Set_Node_Length(ret, 1); /* MJD */
6045 Set_Node_Offset(ret, RExC_parse); /* MJD */
6053 /* Pick up the branches, linking them together. */
6054 parse_start = RExC_parse; /* MJD */
6055 br = regbranch(pRExC_state, &flags, 1,depth+1);
6056 /* branch_len = (paren != 0); */
6060 if (*RExC_parse == '|') {
6061 if (!SIZE_ONLY && RExC_extralen) {
6062 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6065 reginsert(pRExC_state, BRANCH, br, depth+1);
6066 Set_Node_Length(br, paren != 0);
6067 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6071 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6073 else if (paren == ':') {
6074 *flagp |= flags&SIMPLE;
6076 if (is_open) { /* Starts with OPEN. */
6077 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6079 else if (paren != '?') /* Not Conditional */
6081 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6083 while (*RExC_parse == '|') {
6084 if (!SIZE_ONLY && RExC_extralen) {
6085 ender = reganode(pRExC_state, LONGJMP,0);
6086 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6089 RExC_extralen += 2; /* Account for LONGJMP. */
6090 nextchar(pRExC_state);
6092 if (RExC_npar > after_freeze)
6093 after_freeze = RExC_npar;
6094 RExC_npar = freeze_paren;
6096 br = regbranch(pRExC_state, &flags, 0, depth+1);
6100 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6102 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6105 if (have_branch || paren != ':') {
6106 /* Make a closing node, and hook it on the end. */
6109 ender = reg_node(pRExC_state, TAIL);
6112 ender = reganode(pRExC_state, CLOSE, parno);
6113 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6114 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6115 "Setting close paren #%"IVdf" to %d\n",
6116 (IV)parno, REG_NODE_NUM(ender)));
6117 RExC_close_parens[parno-1]= ender;
6118 if (RExC_nestroot == parno)
6121 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6122 Set_Node_Length(ender,1); /* MJD */
6128 *flagp &= ~HASWIDTH;
6131 ender = reg_node(pRExC_state, SUCCEED);
6134 ender = reg_node(pRExC_state, END);
6136 assert(!RExC_opend); /* there can only be one! */
6141 REGTAIL(pRExC_state, lastbr, ender);
6143 if (have_branch && !SIZE_ONLY) {
6145 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6147 /* Hook the tails of the branches to the closing node. */
6148 for (br = ret; br; br = regnext(br)) {
6149 const U8 op = PL_regkind[OP(br)];
6151 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6153 else if (op == BRANCHJ) {
6154 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6162 static const char parens[] = "=!<,>";
6164 if (paren && (p = strchr(parens, paren))) {
6165 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6166 int flag = (p - parens) > 1;
6169 node = SUSPEND, flag = 0;
6170 reginsert(pRExC_state, node,ret, depth+1);
6171 Set_Node_Cur_Length(ret);
6172 Set_Node_Offset(ret, parse_start + 1);
6174 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6178 /* Check for proper termination. */
6180 RExC_flags = oregflags;
6181 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6182 RExC_parse = oregcomp_parse;
6183 vFAIL("Unmatched (");
6186 else if (!paren && RExC_parse < RExC_end) {
6187 if (*RExC_parse == ')') {
6189 vFAIL("Unmatched )");
6192 FAIL("Junk on end of regexp"); /* "Can't happen". */
6196 RExC_npar = after_freeze;
6201 - regbranch - one alternative of an | operator
6203 * Implements the concatenation operator.
6206 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6209 register regnode *ret;
6210 register regnode *chain = NULL;
6211 register regnode *latest;
6212 I32 flags = 0, c = 0;
6213 GET_RE_DEBUG_FLAGS_DECL;
6214 DEBUG_PARSE("brnc");
6219 if (!SIZE_ONLY && RExC_extralen)
6220 ret = reganode(pRExC_state, BRANCHJ,0);
6222 ret = reg_node(pRExC_state, BRANCH);
6223 Set_Node_Length(ret, 1);
6227 if (!first && SIZE_ONLY)
6228 RExC_extralen += 1; /* BRANCHJ */
6230 *flagp = WORST; /* Tentatively. */
6233 nextchar(pRExC_state);
6234 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6236 latest = regpiece(pRExC_state, &flags,depth+1);
6237 if (latest == NULL) {
6238 if (flags & TRYAGAIN)
6242 else if (ret == NULL)
6244 *flagp |= flags&(HASWIDTH|POSTPONED);
6245 if (chain == NULL) /* First piece. */
6246 *flagp |= flags&SPSTART;
6249 REGTAIL(pRExC_state, chain, latest);
6254 if (chain == NULL) { /* Loop ran zero times. */
6255 chain = reg_node(pRExC_state, NOTHING);
6260 *flagp |= flags&SIMPLE;
6267 - regpiece - something followed by possible [*+?]
6269 * Note that the branching code sequences used for ? and the general cases
6270 * of * and + are somewhat optimized: they use the same NOTHING node as
6271 * both the endmarker for their branch list and the body of the last branch.
6272 * It might seem that this node could be dispensed with entirely, but the
6273 * endmarker role is not redundant.
6276 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6279 register regnode *ret;
6281 register char *next;
6283 const char * const origparse = RExC_parse;
6285 I32 max = REG_INFTY;
6287 const char *maxpos = NULL;
6288 GET_RE_DEBUG_FLAGS_DECL;
6289 DEBUG_PARSE("piec");
6291 ret = regatom(pRExC_state, &flags,depth+1);
6293 if (flags & TRYAGAIN)
6300 if (op == '{' && regcurly(RExC_parse)) {
6302 parse_start = RExC_parse; /* MJD */
6303 next = RExC_parse + 1;
6304 while (isDIGIT(*next) || *next == ',') {
6313 if (*next == '}') { /* got one */
6317 min = atoi(RExC_parse);
6321 maxpos = RExC_parse;
6323 if (!max && *maxpos != '0')
6324 max = REG_INFTY; /* meaning "infinity" */
6325 else if (max >= REG_INFTY)
6326 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6328 nextchar(pRExC_state);
6331 if ((flags&SIMPLE)) {
6332 RExC_naughty += 2 + RExC_naughty / 2;
6333 reginsert(pRExC_state, CURLY, ret, depth+1);
6334 Set_Node_Offset(ret, parse_start+1); /* MJD */
6335 Set_Node_Cur_Length(ret);
6338 regnode * const w = reg_node(pRExC_state, WHILEM);
6341 REGTAIL(pRExC_state, ret, w);
6342 if (!SIZE_ONLY && RExC_extralen) {
6343 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6344 reginsert(pRExC_state, NOTHING,ret, depth+1);
6345 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6347 reginsert(pRExC_state, CURLYX,ret, depth+1);
6349 Set_Node_Offset(ret, parse_start+1);
6350 Set_Node_Length(ret,
6351 op == '{' ? (RExC_parse - parse_start) : 1);
6353 if (!SIZE_ONLY && RExC_extralen)
6354 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6355 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6357 RExC_whilem_seen++, RExC_extralen += 3;
6358 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6366 if (max && max < min)
6367 vFAIL("Can't do {n,m} with n > m");
6369 ARG1_SET(ret, (U16)min);
6370 ARG2_SET(ret, (U16)max);
6382 #if 0 /* Now runtime fix should be reliable. */
6384 /* if this is reinstated, don't forget to put this back into perldiag:
6386 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6388 (F) The part of the regexp subject to either the * or + quantifier
6389 could match an empty string. The {#} shows in the regular
6390 expression about where the problem was discovered.
6394 if (!(flags&HASWIDTH) && op != '?')
6395 vFAIL("Regexp *+ operand could be empty");
6398 parse_start = RExC_parse;
6399 nextchar(pRExC_state);
6401 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6403 if (op == '*' && (flags&SIMPLE)) {
6404 reginsert(pRExC_state, STAR, ret, depth+1);
6408 else if (op == '*') {
6412 else if (op == '+' && (flags&SIMPLE)) {
6413 reginsert(pRExC_state, PLUS, ret, depth+1);
6417 else if (op == '+') {
6421 else if (op == '?') {
6426 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6428 "%.*s matches null string many times",
6429 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6433 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6434 nextchar(pRExC_state);
6435 reginsert(pRExC_state, MINMOD, ret, depth+1);
6436 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6438 #ifndef REG_ALLOW_MINMOD_SUSPEND
6441 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6443 nextchar(pRExC_state);
6444 ender = reg_node(pRExC_state, SUCCEED);
6445 REGTAIL(pRExC_state, ret, ender);
6446 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6448 ender = reg_node(pRExC_state, TAIL);
6449 REGTAIL(pRExC_state, ret, ender);
6453 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6455 vFAIL("Nested quantifiers");
6462 /* reg_namedseq(pRExC_state,UVp)
6464 This is expected to be called by a parser routine that has
6465 recognized'\N' and needs to handle the rest. RExC_parse is
6466 expected to point at the first char following the N at the time
6469 If valuep is non-null then it is assumed that we are parsing inside
6470 of a charclass definition and the first codepoint in the resolved
6471 string is returned via *valuep and the routine will return NULL.
6472 In this mode if a multichar string is returned from the charnames
6473 handler a warning will be issued, and only the first char in the
6474 sequence will be examined. If the string returned is zero length
6475 then the value of *valuep is undefined and NON-NULL will
6476 be returned to indicate failure. (This will NOT be a valid pointer
6479 If value is null then it is assumed that we are parsing normal text
6480 and inserts a new EXACT node into the program containing the resolved
6481 string and returns a pointer to the new node. If the string is
6482 zerolength a NOTHING node is emitted.
6484 On success RExC_parse is set to the char following the endbrace.
6485 Parsing failures will generate a fatal errorvia vFAIL(...)
6487 NOTE: We cache all results from the charnames handler locally in
6488 the RExC_charnames hash (created on first use) to prevent a charnames
6489 handler from playing silly-buggers and returning a short string and
6490 then a long string for a given pattern. Since the regexp program
6491 size is calculated during an initial parse this would result
6492 in a buffer overrun so we cache to prevent the charname result from
6493 changing during the course of the parse.
6497 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6499 char * name; /* start of the content of the name */
6500 char * endbrace; /* endbrace following the name */
6503 STRLEN len; /* this has various purposes throughout the code */
6504 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6505 regnode *ret = NULL;
6507 if (*RExC_parse != '{') {
6508 vFAIL("Missing braces on \\N{}");
6510 name = RExC_parse+1;
6511 endbrace = strchr(RExC_parse, '}');
6514 vFAIL("Missing right brace on \\N{}");
6516 RExC_parse = endbrace + 1;
6519 /* RExC_parse points at the beginning brace,
6520 endbrace points at the last */
6521 if ( name[0]=='U' && name[1]=='+' ) {
6522 /* its a "Unicode hex" notation {U+89AB} */
6523 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6524 | PERL_SCAN_DISALLOW_PREFIX
6525 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6528 len = (STRLEN)(endbrace - name - 2);
6529 cp = grok_hex(name + 2, &len, &fl, NULL);
6530 if ( len != (STRLEN)(endbrace - name - 2) ) {
6540 sv_str= newSVpvn(&string, 1);
6542 /* fetch the charnames handler for this scope */
6543 HV * const table = GvHV(PL_hintgv);
6545 hv_fetchs(table, "charnames", FALSE) :
6547 SV *cv= cvp ? *cvp : NULL;
6550 /* create an SV with the name as argument */
6551 sv_name = newSVpvn(name, endbrace - name);
6553 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6554 vFAIL2("Constant(\\N{%s}) unknown: "
6555 "(possibly a missing \"use charnames ...\")",
6558 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6559 vFAIL2("Constant(\\N{%s}): "
6560 "$^H{charnames} is not defined",SvPVX(sv_name));
6565 if (!RExC_charnames) {
6566 /* make sure our cache is allocated */
6567 RExC_charnames = newHV();
6568 sv_2mortal((SV*)RExC_charnames);
6570 /* see if we have looked this one up before */
6571 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6573 sv_str = HeVAL(he_str);
6586 count= call_sv(cv, G_SCALAR);
6588 if (count == 1) { /* XXXX is this right? dmq */
6590 SvREFCNT_inc_simple_void(sv_str);
6598 if ( !sv_str || !SvOK(sv_str) ) {
6599 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6600 "did not return a defined value",SvPVX(sv_name));
6602 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6607 char *p = SvPV(sv_str, len);
6610 if ( SvUTF8(sv_str) ) {
6611 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6615 We have to turn on utf8 for high bit chars otherwise
6616 we get failures with
6618 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6619 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6621 This is different from what \x{} would do with the same
6622 codepoint, where the condition is > 0xFF.
6629 /* warn if we havent used the whole string? */
6631 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6633 "Ignoring excess chars from \\N{%s} in character class",
6637 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6639 "Ignoring zero length \\N{%s} in character class",
6644 SvREFCNT_dec(sv_name);
6646 SvREFCNT_dec(sv_str);
6647 return len ? NULL : (regnode *)&len;
6648 } else if(SvCUR(sv_str)) {
6654 char * parse_start = name-3; /* needed for the offsets */
6656 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6658 ret = reg_node(pRExC_state,
6659 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6662 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6663 sv_utf8_upgrade(sv_str);
6664 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6668 p = SvPV(sv_str, len);
6670 /* len is the length written, charlen is the size the char read */
6671 for ( len = 0; p < pend; p += charlen ) {
6673 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6675 STRLEN foldlen,numlen;
6676 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6677 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6678 /* Emit all the Unicode characters. */
6680 for (foldbuf = tmpbuf;
6684 uvc = utf8_to_uvchr(foldbuf, &numlen);
6686 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6689 /* In EBCDIC the numlen
6690 * and unilen can differ. */
6692 if (numlen >= foldlen)
6696 break; /* "Can't happen." */
6699 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6711 RExC_size += STR_SZ(len);
6714 RExC_emit += STR_SZ(len);
6716 Set_Node_Cur_Length(ret); /* MJD */
6718 nextchar(pRExC_state);
6720 ret = reg_node(pRExC_state,NOTHING);
6723 SvREFCNT_dec(sv_str);
6726 SvREFCNT_dec(sv_name);
6736 * It returns the code point in utf8 for the value in *encp.
6737 * value: a code value in the source encoding
6738 * encp: a pointer to an Encode object
6740 * If the result from Encode is not a single character,
6741 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6744 S_reg_recode(pTHX_ const char value, SV **encp)
6747 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
6748 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6749 const STRLEN newlen = SvCUR(sv);
6750 UV uv = UNICODE_REPLACEMENT;
6754 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6757 if (!newlen || numlen != newlen) {
6758 uv = UNICODE_REPLACEMENT;
6766 - regatom - the lowest level
6768 Try to identify anything special at the start of the pattern. If there
6769 is, then handle it as required. This may involve generating a single regop,
6770 such as for an assertion; or it may involve recursing, such as to
6771 handle a () structure.
6773 If the string doesn't start with something special then we gobble up
6774 as much literal text as we can.
6776 Once we have been able to handle whatever type of thing started the
6777 sequence, we return.
6779 Note: we have to be careful with escapes, as they can be both literal
6780 and special, and in the case of \10 and friends can either, depending
6781 on context. Specifically there are two seperate switches for handling
6782 escape sequences, with the one for handling literal escapes requiring
6783 a dummy entry for all of the special escapes that are actually handled
6788 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6791 register regnode *ret = NULL;
6793 char *parse_start = RExC_parse;
6794 GET_RE_DEBUG_FLAGS_DECL;
6795 DEBUG_PARSE("atom");
6796 *flagp = WORST; /* Tentatively. */
6800 switch ((U8)*RExC_parse) {
6802 RExC_seen_zerolen++;
6803 nextchar(pRExC_state);
6804 if (RExC_flags & RXf_PMf_MULTILINE)
6805 ret = reg_node(pRExC_state, MBOL);
6806 else if (RExC_flags & RXf_PMf_SINGLELINE)
6807 ret = reg_node(pRExC_state, SBOL);
6809 ret = reg_node(pRExC_state, BOL);
6810 Set_Node_Length(ret, 1); /* MJD */
6813 nextchar(pRExC_state);
6815 RExC_seen_zerolen++;
6816 if (RExC_flags & RXf_PMf_MULTILINE)
6817 ret = reg_node(pRExC_state, MEOL);
6818 else if (RExC_flags & RXf_PMf_SINGLELINE)
6819 ret = reg_node(pRExC_state, SEOL);
6821 ret = reg_node(pRExC_state, EOL);
6822 Set_Node_Length(ret, 1); /* MJD */
6825 nextchar(pRExC_state);
6826 if (RExC_flags & RXf_PMf_SINGLELINE)
6827 ret = reg_node(pRExC_state, SANY);
6829 ret = reg_node(pRExC_state, REG_ANY);
6830 *flagp |= HASWIDTH|SIMPLE;
6832 Set_Node_Length(ret, 1); /* MJD */
6836 char * const oregcomp_parse = ++RExC_parse;
6837 ret = regclass(pRExC_state,depth+1);
6838 if (*RExC_parse != ']') {
6839 RExC_parse = oregcomp_parse;
6840 vFAIL("Unmatched [");
6842 nextchar(pRExC_state);
6843 *flagp |= HASWIDTH|SIMPLE;
6844 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6848 nextchar(pRExC_state);
6849 ret = reg(pRExC_state, 1, &flags,depth+1);
6851 if (flags & TRYAGAIN) {
6852 if (RExC_parse == RExC_end) {
6853 /* Make parent create an empty node if needed. */
6861 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6865 if (flags & TRYAGAIN) {
6869 vFAIL("Internal urp");
6870 /* Supposed to be caught earlier. */
6873 if (!regcurly(RExC_parse)) {
6882 vFAIL("Quantifier follows nothing");
6890 len=0; /* silence a spurious compiler warning */
6891 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
6892 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
6893 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
6894 ret = reganode(pRExC_state, FOLDCHAR, cp);
6895 Set_Node_Length(ret, 1); /* MJD */
6896 nextchar(pRExC_state); /* kill whitespace under /x */
6904 This switch handles escape sequences that resolve to some kind
6905 of special regop and not to literal text. Escape sequnces that
6906 resolve to literal text are handled below in the switch marked
6909 Every entry in this switch *must* have a corresponding entry
6910 in the literal escape switch. However, the opposite is not
6911 required, as the default for this switch is to jump to the
6912 literal text handling code.
6914 switch ((U8)*++RExC_parse) {
6919 /* Special Escapes */
6921 RExC_seen_zerolen++;
6922 ret = reg_node(pRExC_state, SBOL);
6924 goto finish_meta_pat;
6926 ret = reg_node(pRExC_state, GPOS);
6927 RExC_seen |= REG_SEEN_GPOS;
6929 goto finish_meta_pat;
6931 RExC_seen_zerolen++;
6932 ret = reg_node(pRExC_state, KEEPS);
6934 /* XXX:dmq : disabling in-place substitution seems to
6935 * be necessary here to avoid cases of memory corruption, as
6936 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
6938 RExC_seen |= REG_SEEN_LOOKBEHIND;
6939 goto finish_meta_pat;
6941 ret = reg_node(pRExC_state, SEOL);
6943 RExC_seen_zerolen++; /* Do not optimize RE away */
6944 goto finish_meta_pat;
6946 ret = reg_node(pRExC_state, EOS);
6948 RExC_seen_zerolen++; /* Do not optimize RE away */
6949 goto finish_meta_pat;
6951 ret = reg_node(pRExC_state, CANY);
6952 RExC_seen |= REG_SEEN_CANY;
6953 *flagp |= HASWIDTH|SIMPLE;
6954 goto finish_meta_pat;
6956 ret = reg_node(pRExC_state, CLUMP);
6958 goto finish_meta_pat;
6960 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6961 *flagp |= HASWIDTH|SIMPLE;
6962 goto finish_meta_pat;
6964 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6965 *flagp |= HASWIDTH|SIMPLE;
6966 goto finish_meta_pat;
6968 RExC_seen_zerolen++;
6969 RExC_seen |= REG_SEEN_LOOKBEHIND;
6970 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6972 goto finish_meta_pat;
6974 RExC_seen_zerolen++;
6975 RExC_seen |= REG_SEEN_LOOKBEHIND;
6976 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6978 goto finish_meta_pat;
6980 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6981 *flagp |= HASWIDTH|SIMPLE;
6982 goto finish_meta_pat;
6984 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6985 *flagp |= HASWIDTH|SIMPLE;
6986 goto finish_meta_pat;
6988 ret = reg_node(pRExC_state, DIGIT);
6989 *flagp |= HASWIDTH|SIMPLE;
6990 goto finish_meta_pat;
6992 ret = reg_node(pRExC_state, NDIGIT);
6993 *flagp |= HASWIDTH|SIMPLE;
6994 goto finish_meta_pat;
6996 ret = reg_node(pRExC_state, LNBREAK);
6997 *flagp |= HASWIDTH|SIMPLE;
6998 goto finish_meta_pat;
7000 ret = reg_node(pRExC_state, HORIZWS);
7001 *flagp |= HASWIDTH|SIMPLE;
7002 goto finish_meta_pat;
7004 ret = reg_node(pRExC_state, NHORIZWS);
7005 *flagp |= HASWIDTH|SIMPLE;
7006 goto finish_meta_pat;
7008 ret = reg_node(pRExC_state, VERTWS);
7009 *flagp |= HASWIDTH|SIMPLE;
7010 goto finish_meta_pat;
7012 ret = reg_node(pRExC_state, NVERTWS);
7013 *flagp |= HASWIDTH|SIMPLE;
7015 nextchar(pRExC_state);
7016 Set_Node_Length(ret, 2); /* MJD */
7021 char* const oldregxend = RExC_end;
7023 char* parse_start = RExC_parse - 2;
7026 if (RExC_parse[1] == '{') {
7027 /* a lovely hack--pretend we saw [\pX] instead */
7028 RExC_end = strchr(RExC_parse, '}');
7030 const U8 c = (U8)*RExC_parse;
7032 RExC_end = oldregxend;
7033 vFAIL2("Missing right brace on \\%c{}", c);
7038 RExC_end = RExC_parse + 2;
7039 if (RExC_end > oldregxend)
7040 RExC_end = oldregxend;
7044 ret = regclass(pRExC_state,depth+1);
7046 RExC_end = oldregxend;
7049 Set_Node_Offset(ret, parse_start + 2);
7050 Set_Node_Cur_Length(ret);
7051 nextchar(pRExC_state);
7052 *flagp |= HASWIDTH|SIMPLE;
7056 /* Handle \N{NAME} here and not below because it can be
7057 multicharacter. join_exact() will join them up later on.
7058 Also this makes sure that things like /\N{BLAH}+/ and
7059 \N{BLAH} being multi char Just Happen. dmq*/
7061 ret= reg_namedseq(pRExC_state, NULL);
7063 case 'k': /* Handle \k<NAME> and \k'NAME' */
7066 char ch= RExC_parse[1];
7067 if (ch != '<' && ch != '\'' && ch != '{') {
7069 vFAIL2("Sequence %.2s... not terminated",parse_start);
7071 /* this pretty much dupes the code for (?P=...) in reg(), if
7072 you change this make sure you change that */
7073 char* name_start = (RExC_parse += 2);
7075 SV *sv_dat = reg_scan_name(pRExC_state,
7076 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7077 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7078 if (RExC_parse == name_start || *RExC_parse != ch)
7079 vFAIL2("Sequence %.3s... not terminated",parse_start);
7082 num = add_data( pRExC_state, 1, "S" );
7083 RExC_rxi->data->data[num]=(void*)sv_dat;
7084 SvREFCNT_inc_simple_void(sv_dat);
7088 ret = reganode(pRExC_state,
7089 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7093 /* override incorrect value set in reganode MJD */
7094 Set_Node_Offset(ret, parse_start+1);
7095 Set_Node_Cur_Length(ret); /* MJD */
7096 nextchar(pRExC_state);
7102 case '1': case '2': case '3': case '4':
7103 case '5': case '6': case '7': case '8': case '9':
7106 bool isg = *RExC_parse == 'g';
7111 if (*RExC_parse == '{') {
7115 if (*RExC_parse == '-') {
7119 if (hasbrace && !isDIGIT(*RExC_parse)) {
7120 if (isrel) RExC_parse--;
7122 goto parse_named_seq;
7124 num = atoi(RExC_parse);
7125 if (isg && num == 0)
7126 vFAIL("Reference to invalid group 0");
7128 num = RExC_npar - num;
7130 vFAIL("Reference to nonexistent or unclosed group");
7132 if (!isg && num > 9 && num >= RExC_npar)
7135 char * const parse_start = RExC_parse - 1; /* MJD */
7136 while (isDIGIT(*RExC_parse))
7138 if (parse_start == RExC_parse - 1)
7139 vFAIL("Unterminated \\g... pattern");
7141 if (*RExC_parse != '}')
7142 vFAIL("Unterminated \\g{...} pattern");
7146 if (num > (I32)RExC_rx->nparens)
7147 vFAIL("Reference to nonexistent group");
7150 ret = reganode(pRExC_state,
7151 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7155 /* override incorrect value set in reganode MJD */
7156 Set_Node_Offset(ret, parse_start+1);
7157 Set_Node_Cur_Length(ret); /* MJD */
7159 nextchar(pRExC_state);
7164 if (RExC_parse >= RExC_end)
7165 FAIL("Trailing \\");
7168 /* Do not generate "unrecognized" warnings here, we fall
7169 back into the quick-grab loop below */
7176 if (RExC_flags & RXf_PMf_EXTENDED) {
7177 if ( reg_skipcomment( pRExC_state ) )
7184 register STRLEN len;
7189 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7191 parse_start = RExC_parse - 1;
7197 ret = reg_node(pRExC_state,
7198 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7200 for (len = 0, p = RExC_parse - 1;
7201 len < 127 && p < RExC_end;
7204 char * const oldp = p;
7206 if (RExC_flags & RXf_PMf_EXTENDED)
7207 p = regwhite( pRExC_state, p );
7212 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7213 goto normal_default;
7223 /* Literal Escapes Switch
7225 This switch is meant to handle escape sequences that
7226 resolve to a literal character.
7228 Every escape sequence that represents something
7229 else, like an assertion or a char class, is handled
7230 in the switch marked 'Special Escapes' above in this
7231 routine, but also has an entry here as anything that
7232 isn't explicitly mentioned here will be treated as
7233 an unescaped equivalent literal.
7237 /* These are all the special escapes. */
7241 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7242 goto normal_default;
7243 case 'A': /* Start assertion */
7244 case 'b': case 'B': /* Word-boundary assertion*/
7245 case 'C': /* Single char !DANGEROUS! */
7246 case 'd': case 'D': /* digit class */
7247 case 'g': case 'G': /* generic-backref, pos assertion */
7248 case 'h': case 'H': /* HORIZWS */
7249 case 'k': case 'K': /* named backref, keep marker */
7250 case 'N': /* named char sequence */
7251 case 'p': case 'P': /* Unicode property */
7252 case 'R': /* LNBREAK */
7253 case 's': case 'S': /* space class */
7254 case 'v': case 'V': /* VERTWS */
7255 case 'w': case 'W': /* word class */
7256 case 'X': /* eXtended Unicode "combining character sequence" */
7257 case 'z': case 'Z': /* End of line/string assertion */
7261 /* Anything after here is an escape that resolves to a
7262 literal. (Except digits, which may or may not)
7281 ender = ASCII_TO_NATIVE('\033');
7285 ender = ASCII_TO_NATIVE('\007');
7290 char* const e = strchr(p, '}');
7294 vFAIL("Missing right brace on \\x{}");
7297 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7298 | PERL_SCAN_DISALLOW_PREFIX;
7299 STRLEN numlen = e - p - 1;
7300 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7307 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7309 ender = grok_hex(p, &numlen, &flags, NULL);
7312 if (PL_encoding && ender < 0x100)
7313 goto recode_encoding;
7317 ender = UCHARAT(p++);
7318 ender = toCTRL(ender);
7320 case '0': case '1': case '2': case '3':case '4':
7321 case '5': case '6': case '7': case '8':case '9':
7323 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7326 ender = grok_oct(p, &numlen, &flags, NULL);
7333 if (PL_encoding && ender < 0x100)
7334 goto recode_encoding;
7338 SV* enc = PL_encoding;
7339 ender = reg_recode((const char)(U8)ender, &enc);
7340 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7341 vWARN(p, "Invalid escape in the specified encoding");
7347 FAIL("Trailing \\");
7350 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
7351 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7352 goto normal_default;
7357 if (UTF8_IS_START(*p) && UTF) {
7359 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7360 &numlen, UTF8_ALLOW_DEFAULT);
7367 if ( RExC_flags & RXf_PMf_EXTENDED)
7368 p = regwhite( pRExC_state, p );
7370 /* Prime the casefolded buffer. */
7371 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7373 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7378 /* Emit all the Unicode characters. */
7380 for (foldbuf = tmpbuf;
7382 foldlen -= numlen) {
7383 ender = utf8_to_uvchr(foldbuf, &numlen);
7385 const STRLEN unilen = reguni(pRExC_state, ender, s);
7388 /* In EBCDIC the numlen
7389 * and unilen can differ. */
7391 if (numlen >= foldlen)
7395 break; /* "Can't happen." */
7399 const STRLEN unilen = reguni(pRExC_state, ender, s);
7408 REGC((char)ender, s++);
7414 /* Emit all the Unicode characters. */
7416 for (foldbuf = tmpbuf;
7418 foldlen -= numlen) {
7419 ender = utf8_to_uvchr(foldbuf, &numlen);
7421 const STRLEN unilen = reguni(pRExC_state, ender, s);
7424 /* In EBCDIC the numlen
7425 * and unilen can differ. */
7427 if (numlen >= foldlen)
7435 const STRLEN unilen = reguni(pRExC_state, ender, s);
7444 REGC((char)ender, s++);
7448 Set_Node_Cur_Length(ret); /* MJD */
7449 nextchar(pRExC_state);
7451 /* len is STRLEN which is unsigned, need to copy to signed */
7454 vFAIL("Internal disaster");
7458 if (len == 1 && UNI_IS_INVARIANT(ender))
7462 RExC_size += STR_SZ(len);
7465 RExC_emit += STR_SZ(len);
7475 S_regwhite( RExC_state_t *pRExC_state, char *p )
7477 const char *e = RExC_end;
7481 else if (*p == '#') {
7490 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7498 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7499 Character classes ([:foo:]) can also be negated ([:^foo:]).
7500 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7501 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7502 but trigger failures because they are currently unimplemented. */
7504 #define POSIXCC_DONE(c) ((c) == ':')
7505 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7506 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7509 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7512 I32 namedclass = OOB_NAMEDCLASS;
7514 if (value == '[' && RExC_parse + 1 < RExC_end &&
7515 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7516 POSIXCC(UCHARAT(RExC_parse))) {
7517 const char c = UCHARAT(RExC_parse);
7518 char* const s = RExC_parse++;
7520 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7522 if (RExC_parse == RExC_end)
7523 /* Grandfather lone [:, [=, [. */
7526 const char* const t = RExC_parse++; /* skip over the c */
7529 if (UCHARAT(RExC_parse) == ']') {
7530 const char *posixcc = s + 1;
7531 RExC_parse++; /* skip over the ending ] */
7534 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7535 const I32 skip = t - posixcc;
7537 /* Initially switch on the length of the name. */
7540 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7541 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7544 /* Names all of length 5. */
7545 /* alnum alpha ascii blank cntrl digit graph lower
7546 print punct space upper */
7547 /* Offset 4 gives the best switch position. */
7548 switch (posixcc[4]) {
7550 if (memEQ(posixcc, "alph", 4)) /* alpha */
7551 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7554 if (memEQ(posixcc, "spac", 4)) /* space */
7555 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7558 if (memEQ(posixcc, "grap", 4)) /* graph */
7559 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7562 if (memEQ(posixcc, "asci", 4)) /* ascii */
7563 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7566 if (memEQ(posixcc, "blan", 4)) /* blank */
7567 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7570 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7571 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7574 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7575 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7578 if (memEQ(posixcc, "lowe", 4)) /* lower */
7579 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7580 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7581 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7584 if (memEQ(posixcc, "digi", 4)) /* digit */
7585 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7586 else if (memEQ(posixcc, "prin", 4)) /* print */
7587 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7588 else if (memEQ(posixcc, "punc", 4)) /* punct */
7589 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7594 if (memEQ(posixcc, "xdigit", 6))
7595 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7599 if (namedclass == OOB_NAMEDCLASS)
7600 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7602 assert (posixcc[skip] == ':');
7603 assert (posixcc[skip+1] == ']');
7604 } else if (!SIZE_ONLY) {
7605 /* [[=foo=]] and [[.foo.]] are still future. */
7607 /* adjust RExC_parse so the warning shows after
7609 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7611 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7614 /* Maternal grandfather:
7615 * "[:" ending in ":" but not in ":]" */
7625 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7628 if (POSIXCC(UCHARAT(RExC_parse))) {
7629 const char *s = RExC_parse;
7630 const char c = *s++;
7634 if (*s && c == *s && s[1] == ']') {
7635 if (ckWARN(WARN_REGEXP))
7637 "POSIX syntax [%c %c] belongs inside character classes",
7640 /* [[=foo=]] and [[.foo.]] are still future. */
7641 if (POSIXCC_NOTYET(c)) {
7642 /* adjust RExC_parse so the error shows after
7644 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7646 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7653 #define _C_C_T_(NAME,TEST,WORD) \
7656 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7658 for (value = 0; value < 256; value++) \
7660 ANYOF_BITMAP_SET(ret, value); \
7665 case ANYOF_N##NAME: \
7667 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7669 for (value = 0; value < 256; value++) \
7671 ANYOF_BITMAP_SET(ret, value); \
7677 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7679 for (value = 0; value < 256; value++) \
7681 ANYOF_BITMAP_SET(ret, value); \
7685 case ANYOF_N##NAME: \
7686 for (value = 0; value < 256; value++) \
7688 ANYOF_BITMAP_SET(ret, value); \
7694 parse a class specification and produce either an ANYOF node that
7695 matches the pattern or if the pattern matches a single char only and
7696 that char is < 256 and we are case insensitive then we produce an
7701 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7704 register UV nextvalue;
7705 register IV prevvalue = OOB_UNICODE;
7706 register IV range = 0;
7707 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7708 register regnode *ret;
7711 char *rangebegin = NULL;
7712 bool need_class = 0;
7715 bool optimize_invert = TRUE;
7716 AV* unicode_alternate = NULL;
7718 UV literal_endpoint = 0;
7720 UV stored = 0; /* number of chars stored in the class */
7722 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7723 case we need to change the emitted regop to an EXACT. */
7724 const char * orig_parse = RExC_parse;
7725 GET_RE_DEBUG_FLAGS_DECL;
7727 PERL_UNUSED_ARG(depth);
7730 DEBUG_PARSE("clas");
7732 /* Assume we are going to generate an ANYOF node. */
7733 ret = reganode(pRExC_state, ANYOF, 0);
7736 ANYOF_FLAGS(ret) = 0;
7738 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7742 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7746 RExC_size += ANYOF_SKIP;
7747 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7750 RExC_emit += ANYOF_SKIP;
7752 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7754 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7755 ANYOF_BITMAP_ZERO(ret);
7756 listsv = newSVpvs("# comment\n");
7759 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7761 if (!SIZE_ONLY && POSIXCC(nextvalue))
7762 checkposixcc(pRExC_state);
7764 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7765 if (UCHARAT(RExC_parse) == ']')
7769 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7773 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7776 rangebegin = RExC_parse;
7778 value = utf8n_to_uvchr((U8*)RExC_parse,
7779 RExC_end - RExC_parse,
7780 &numlen, UTF8_ALLOW_DEFAULT);
7781 RExC_parse += numlen;
7784 value = UCHARAT(RExC_parse++);
7786 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7787 if (value == '[' && POSIXCC(nextvalue))
7788 namedclass = regpposixcc(pRExC_state, value);
7789 else if (value == '\\') {
7791 value = utf8n_to_uvchr((U8*)RExC_parse,
7792 RExC_end - RExC_parse,
7793 &numlen, UTF8_ALLOW_DEFAULT);
7794 RExC_parse += numlen;
7797 value = UCHARAT(RExC_parse++);
7798 /* Some compilers cannot handle switching on 64-bit integer
7799 * values, therefore value cannot be an UV. Yes, this will
7800 * be a problem later if we want switch on Unicode.
7801 * A similar issue a little bit later when switching on
7802 * namedclass. --jhi */
7803 switch ((I32)value) {
7804 case 'w': namedclass = ANYOF_ALNUM; break;
7805 case 'W': namedclass = ANYOF_NALNUM; break;
7806 case 's': namedclass = ANYOF_SPACE; break;
7807 case 'S': namedclass = ANYOF_NSPACE; break;
7808 case 'd': namedclass = ANYOF_DIGIT; break;
7809 case 'D': namedclass = ANYOF_NDIGIT; break;
7810 case 'v': namedclass = ANYOF_VERTWS; break;
7811 case 'V': namedclass = ANYOF_NVERTWS; break;
7812 case 'h': namedclass = ANYOF_HORIZWS; break;
7813 case 'H': namedclass = ANYOF_NHORIZWS; break;
7814 case 'N': /* Handle \N{NAME} in class */
7816 /* We only pay attention to the first char of
7817 multichar strings being returned. I kinda wonder
7818 if this makes sense as it does change the behaviour
7819 from earlier versions, OTOH that behaviour was broken
7821 UV v; /* value is register so we cant & it /grrr */
7822 if (reg_namedseq(pRExC_state, &v)) {
7832 if (RExC_parse >= RExC_end)
7833 vFAIL2("Empty \\%c{}", (U8)value);
7834 if (*RExC_parse == '{') {
7835 const U8 c = (U8)value;
7836 e = strchr(RExC_parse++, '}');
7838 vFAIL2("Missing right brace on \\%c{}", c);
7839 while (isSPACE(UCHARAT(RExC_parse)))
7841 if (e == RExC_parse)
7842 vFAIL2("Empty \\%c{}", c);
7844 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7852 if (UCHARAT(RExC_parse) == '^') {
7855 value = value == 'p' ? 'P' : 'p'; /* toggle */
7856 while (isSPACE(UCHARAT(RExC_parse))) {
7861 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7862 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7865 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7866 namedclass = ANYOF_MAX; /* no official name, but it's named */
7869 case 'n': value = '\n'; break;
7870 case 'r': value = '\r'; break;
7871 case 't': value = '\t'; break;
7872 case 'f': value = '\f'; break;
7873 case 'b': value = '\b'; break;
7874 case 'e': value = ASCII_TO_NATIVE('\033');break;
7875 case 'a': value = ASCII_TO_NATIVE('\007');break;
7877 if (*RExC_parse == '{') {
7878 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7879 | PERL_SCAN_DISALLOW_PREFIX;
7880 char * const e = strchr(RExC_parse++, '}');
7882 vFAIL("Missing right brace on \\x{}");
7884 numlen = e - RExC_parse;
7885 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7889 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7891 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7892 RExC_parse += numlen;
7894 if (PL_encoding && value < 0x100)
7895 goto recode_encoding;
7898 value = UCHARAT(RExC_parse++);
7899 value = toCTRL(value);
7901 case '0': case '1': case '2': case '3': case '4':
7902 case '5': case '6': case '7': case '8': case '9':
7906 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7907 RExC_parse += numlen;
7908 if (PL_encoding && value < 0x100)
7909 goto recode_encoding;
7914 SV* enc = PL_encoding;
7915 value = reg_recode((const char)(U8)value, &enc);
7916 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7918 "Invalid escape in the specified encoding");
7922 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7924 "Unrecognized escape \\%c in character class passed through",
7928 } /* end of \blah */
7934 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7936 if (!SIZE_ONLY && !need_class)
7937 ANYOF_CLASS_ZERO(ret);
7941 /* a bad range like a-\d, a-[:digit:] ? */
7944 if (ckWARN(WARN_REGEXP)) {
7946 RExC_parse >= rangebegin ?
7947 RExC_parse - rangebegin : 0;
7949 "False [] range \"%*.*s\"",
7952 if (prevvalue < 256) {
7953 ANYOF_BITMAP_SET(ret, prevvalue);
7954 ANYOF_BITMAP_SET(ret, '-');
7957 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7958 Perl_sv_catpvf(aTHX_ listsv,
7959 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7963 range = 0; /* this was not a true range */
7969 const char *what = NULL;
7972 if (namedclass > OOB_NAMEDCLASS)
7973 optimize_invert = FALSE;
7974 /* Possible truncation here but in some 64-bit environments
7975 * the compiler gets heartburn about switch on 64-bit values.
7976 * A similar issue a little earlier when switching on value.
7978 switch ((I32)namedclass) {
7979 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7980 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7981 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7982 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7983 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7984 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7985 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7986 case _C_C_T_(PRINT, isPRINT(value), "Print");
7987 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7988 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7989 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7990 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7991 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7992 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
7993 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
7996 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7999 for (value = 0; value < 128; value++)
8000 ANYOF_BITMAP_SET(ret, value);
8002 for (value = 0; value < 256; value++) {
8004 ANYOF_BITMAP_SET(ret, value);
8013 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8016 for (value = 128; value < 256; value++)
8017 ANYOF_BITMAP_SET(ret, value);
8019 for (value = 0; value < 256; value++) {
8020 if (!isASCII(value))
8021 ANYOF_BITMAP_SET(ret, value);
8030 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8032 /* consecutive digits assumed */
8033 for (value = '0'; value <= '9'; value++)
8034 ANYOF_BITMAP_SET(ret, value);
8041 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8043 /* consecutive digits assumed */
8044 for (value = 0; value < '0'; value++)
8045 ANYOF_BITMAP_SET(ret, value);
8046 for (value = '9' + 1; value < 256; value++)
8047 ANYOF_BITMAP_SET(ret, value);
8053 /* this is to handle \p and \P */
8056 vFAIL("Invalid [::] class");
8060 /* Strings such as "+utf8::isWord\n" */
8061 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8064 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8067 } /* end of namedclass \blah */
8070 if (prevvalue > (IV)value) /* b-a */ {
8071 const int w = RExC_parse - rangebegin;
8072 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8073 range = 0; /* not a valid range */
8077 prevvalue = value; /* save the beginning of the range */
8078 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8079 RExC_parse[1] != ']') {
8082 /* a bad range like \w-, [:word:]- ? */
8083 if (namedclass > OOB_NAMEDCLASS) {
8084 if (ckWARN(WARN_REGEXP)) {
8086 RExC_parse >= rangebegin ?
8087 RExC_parse - rangebegin : 0;
8089 "False [] range \"%*.*s\"",
8093 ANYOF_BITMAP_SET(ret, '-');
8095 range = 1; /* yeah, it's a range! */
8096 continue; /* but do it the next time */
8100 /* now is the next time */
8101 /*stored += (value - prevvalue + 1);*/
8103 if (prevvalue < 256) {
8104 const IV ceilvalue = value < 256 ? value : 255;
8107 /* In EBCDIC [\x89-\x91] should include
8108 * the \x8e but [i-j] should not. */
8109 if (literal_endpoint == 2 &&
8110 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8111 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8113 if (isLOWER(prevvalue)) {
8114 for (i = prevvalue; i <= ceilvalue; i++)
8115 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8117 ANYOF_BITMAP_SET(ret, i);
8120 for (i = prevvalue; i <= ceilvalue; i++)
8121 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8123 ANYOF_BITMAP_SET(ret, i);
8129 for (i = prevvalue; i <= ceilvalue; i++) {
8130 if (!ANYOF_BITMAP_TEST(ret,i)) {
8132 ANYOF_BITMAP_SET(ret, i);
8136 if (value > 255 || UTF) {
8137 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8138 const UV natvalue = NATIVE_TO_UNI(value);
8139 stored+=2; /* can't optimize this class */
8140 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8141 if (prevnatvalue < natvalue) { /* what about > ? */
8142 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8143 prevnatvalue, natvalue);
8145 else if (prevnatvalue == natvalue) {
8146 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8148 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8150 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8152 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8153 if (RExC_precomp[0] == ':' &&
8154 RExC_precomp[1] == '[' &&
8155 (f == 0xDF || f == 0x92)) {
8156 f = NATIVE_TO_UNI(f);
8159 /* If folding and foldable and a single
8160 * character, insert also the folded version
8161 * to the charclass. */
8163 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8164 if ((RExC_precomp[0] == ':' &&
8165 RExC_precomp[1] == '[' &&
8167 (value == 0xFB05 || value == 0xFB06))) ?
8168 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8169 foldlen == (STRLEN)UNISKIP(f) )
8171 if (foldlen == (STRLEN)UNISKIP(f))
8173 Perl_sv_catpvf(aTHX_ listsv,
8176 /* Any multicharacter foldings
8177 * require the following transform:
8178 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8179 * where E folds into "pq" and F folds
8180 * into "rst", all other characters
8181 * fold to single characters. We save
8182 * away these multicharacter foldings,
8183 * to be later saved as part of the
8184 * additional "s" data. */
8187 if (!unicode_alternate)
8188 unicode_alternate = newAV();
8189 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8191 av_push(unicode_alternate, sv);
8195 /* If folding and the value is one of the Greek
8196 * sigmas insert a few more sigmas to make the
8197 * folding rules of the sigmas to work right.
8198 * Note that not all the possible combinations
8199 * are handled here: some of them are handled
8200 * by the standard folding rules, and some of
8201 * them (literal or EXACTF cases) are handled
8202 * during runtime in regexec.c:S_find_byclass(). */
8203 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8204 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8205 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8206 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8207 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8209 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8210 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8211 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8216 literal_endpoint = 0;
8220 range = 0; /* this range (if it was one) is done now */
8224 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8226 RExC_size += ANYOF_CLASS_ADD_SKIP;
8228 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8234 /****** !SIZE_ONLY AFTER HERE *********/
8236 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8237 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8239 /* optimize single char class to an EXACT node
8240 but *only* when its not a UTF/high char */
8241 const char * cur_parse= RExC_parse;
8242 RExC_emit = (regnode *)orig_emit;
8243 RExC_parse = (char *)orig_parse;
8244 ret = reg_node(pRExC_state,
8245 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8246 RExC_parse = (char *)cur_parse;
8247 *STRING(ret)= (char)value;
8249 RExC_emit += STR_SZ(1);
8252 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8253 if ( /* If the only flag is folding (plus possibly inversion). */
8254 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8256 for (value = 0; value < 256; ++value) {
8257 if (ANYOF_BITMAP_TEST(ret, value)) {
8258 UV fold = PL_fold[value];
8261 ANYOF_BITMAP_SET(ret, fold);
8264 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8267 /* optimize inverted simple patterns (e.g. [^a-z]) */
8268 if (optimize_invert &&
8269 /* If the only flag is inversion. */
8270 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8271 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8272 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8273 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8276 AV * const av = newAV();
8278 /* The 0th element stores the character class description
8279 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8280 * to initialize the appropriate swash (which gets stored in
8281 * the 1st element), and also useful for dumping the regnode.
8282 * The 2nd element stores the multicharacter foldings,
8283 * used later (regexec.c:S_reginclass()). */
8284 av_store(av, 0, listsv);
8285 av_store(av, 1, NULL);
8286 av_store(av, 2, (SV*)unicode_alternate);
8287 rv = newRV_noinc((SV*)av);
8288 n = add_data(pRExC_state, 1, "s");
8289 RExC_rxi->data->data[n] = (void*)rv;
8297 /* reg_skipcomment()
8299 Absorbs an /x style # comments from the input stream.
8300 Returns true if there is more text remaining in the stream.
8301 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8302 terminates the pattern without including a newline.
8304 Note its the callers responsibility to ensure that we are
8310 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8313 while (RExC_parse < RExC_end)
8314 if (*RExC_parse++ == '\n') {
8319 /* we ran off the end of the pattern without ending
8320 the comment, so we have to add an \n when wrapping */
8321 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8329 Advance that parse position, and optionall absorbs
8330 "whitespace" from the inputstream.
8332 Without /x "whitespace" means (?#...) style comments only,
8333 with /x this means (?#...) and # comments and whitespace proper.
8335 Returns the RExC_parse point from BEFORE the scan occurs.
8337 This is the /x friendly way of saying RExC_parse++.
8341 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8343 char* const retval = RExC_parse++;
8346 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8347 RExC_parse[2] == '#') {
8348 while (*RExC_parse != ')') {
8349 if (RExC_parse == RExC_end)
8350 FAIL("Sequence (?#... not terminated");
8356 if (RExC_flags & RXf_PMf_EXTENDED) {
8357 if (isSPACE(*RExC_parse)) {
8361 else if (*RExC_parse == '#') {
8362 if ( reg_skipcomment( pRExC_state ) )
8371 - reg_node - emit a node
8373 STATIC regnode * /* Location. */
8374 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8377 register regnode *ptr;
8378 regnode * const ret = RExC_emit;
8379 GET_RE_DEBUG_FLAGS_DECL;
8382 SIZE_ALIGN(RExC_size);
8386 if (RExC_emit >= RExC_emit_bound)
8387 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8389 NODE_ALIGN_FILL(ret);
8391 FILL_ADVANCE_NODE(ptr, op);
8392 #ifdef RE_TRACK_PATTERN_OFFSETS
8393 if (RExC_offsets) { /* MJD */
8394 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8395 "reg_node", __LINE__,
8397 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8398 ? "Overwriting end of array!\n" : "OK",
8399 (UV)(RExC_emit - RExC_emit_start),
8400 (UV)(RExC_parse - RExC_start),
8401 (UV)RExC_offsets[0]));
8402 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8410 - reganode - emit a node with an argument
8412 STATIC regnode * /* Location. */
8413 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8416 register regnode *ptr;
8417 regnode * const ret = RExC_emit;
8418 GET_RE_DEBUG_FLAGS_DECL;
8421 SIZE_ALIGN(RExC_size);
8426 assert(2==regarglen[op]+1);
8428 Anything larger than this has to allocate the extra amount.
8429 If we changed this to be:
8431 RExC_size += (1 + regarglen[op]);
8433 then it wouldn't matter. Its not clear what side effect
8434 might come from that so its not done so far.
8439 if (RExC_emit >= RExC_emit_bound)
8440 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8442 NODE_ALIGN_FILL(ret);
8444 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8445 #ifdef RE_TRACK_PATTERN_OFFSETS
8446 if (RExC_offsets) { /* MJD */
8447 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8451 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8452 "Overwriting end of array!\n" : "OK",
8453 (UV)(RExC_emit - RExC_emit_start),
8454 (UV)(RExC_parse - RExC_start),
8455 (UV)RExC_offsets[0]));
8456 Set_Cur_Node_Offset;
8464 - reguni - emit (if appropriate) a Unicode character
8467 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8470 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8474 - reginsert - insert an operator in front of already-emitted operand
8476 * Means relocating the operand.
8479 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8482 register regnode *src;
8483 register regnode *dst;
8484 register regnode *place;
8485 const int offset = regarglen[(U8)op];
8486 const int size = NODE_STEP_REGNODE + offset;
8487 GET_RE_DEBUG_FLAGS_DECL;
8488 PERL_UNUSED_ARG(depth);
8489 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8490 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8499 if (RExC_open_parens) {
8501 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8502 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8503 if ( RExC_open_parens[paren] >= opnd ) {
8504 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8505 RExC_open_parens[paren] += size;
8507 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8509 if ( RExC_close_parens[paren] >= opnd ) {
8510 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8511 RExC_close_parens[paren] += size;
8513 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8518 while (src > opnd) {
8519 StructCopy(--src, --dst, regnode);
8520 #ifdef RE_TRACK_PATTERN_OFFSETS
8521 if (RExC_offsets) { /* MJD 20010112 */
8522 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8526 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8527 ? "Overwriting end of array!\n" : "OK",
8528 (UV)(src - RExC_emit_start),
8529 (UV)(dst - RExC_emit_start),
8530 (UV)RExC_offsets[0]));
8531 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8532 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8538 place = opnd; /* Op node, where operand used to be. */
8539 #ifdef RE_TRACK_PATTERN_OFFSETS
8540 if (RExC_offsets) { /* MJD */
8541 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8545 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8546 ? "Overwriting end of array!\n" : "OK",
8547 (UV)(place - RExC_emit_start),
8548 (UV)(RExC_parse - RExC_start),
8549 (UV)RExC_offsets[0]));
8550 Set_Node_Offset(place, RExC_parse);
8551 Set_Node_Length(place, 1);
8554 src = NEXTOPER(place);
8555 FILL_ADVANCE_NODE(place, op);
8556 Zero(src, offset, regnode);
8560 - regtail - set the next-pointer at the end of a node chain of p to val.
8561 - SEE ALSO: regtail_study
8563 /* TODO: All three parms should be const */
8565 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8568 register regnode *scan;
8569 GET_RE_DEBUG_FLAGS_DECL;
8571 PERL_UNUSED_ARG(depth);
8577 /* Find last node. */
8580 regnode * const temp = regnext(scan);
8582 SV * const mysv=sv_newmortal();
8583 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8584 regprop(RExC_rx, mysv, scan);
8585 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8586 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8587 (temp == NULL ? "->" : ""),
8588 (temp == NULL ? PL_reg_name[OP(val)] : "")
8596 if (reg_off_by_arg[OP(scan)]) {
8597 ARG_SET(scan, val - scan);
8600 NEXT_OFF(scan) = val - scan;
8606 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8607 - Look for optimizable sequences at the same time.
8608 - currently only looks for EXACT chains.
8610 This is expermental code. The idea is to use this routine to perform
8611 in place optimizations on branches and groups as they are constructed,
8612 with the long term intention of removing optimization from study_chunk so
8613 that it is purely analytical.
8615 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8616 to control which is which.
8619 /* TODO: All four parms should be const */
8622 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8625 register regnode *scan;
8627 #ifdef EXPERIMENTAL_INPLACESCAN
8631 GET_RE_DEBUG_FLAGS_DECL;
8637 /* Find last node. */
8641 regnode * const temp = regnext(scan);
8642 #ifdef EXPERIMENTAL_INPLACESCAN
8643 if (PL_regkind[OP(scan)] == EXACT)
8644 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8652 if( exact == PSEUDO )
8654 else if ( exact != OP(scan) )
8663 SV * const mysv=sv_newmortal();
8664 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8665 regprop(RExC_rx, mysv, scan);
8666 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8667 SvPV_nolen_const(mysv),
8669 PL_reg_name[exact]);
8676 SV * const mysv_val=sv_newmortal();
8677 DEBUG_PARSE_MSG("");
8678 regprop(RExC_rx, mysv_val, val);
8679 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8680 SvPV_nolen_const(mysv_val),
8681 (IV)REG_NODE_NUM(val),
8685 if (reg_off_by_arg[OP(scan)]) {
8686 ARG_SET(scan, val - scan);
8689 NEXT_OFF(scan) = val - scan;
8697 - regcurly - a little FSA that accepts {\d+,?\d*}
8700 S_regcurly(register const char *s)
8719 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8723 S_regdump_extflags(pTHX_ const char *lead, const U32 flags) {
8726 for (bit=0; bit<32; bit++) {
8727 if (flags & (1<<bit)) {
8729 PerlIO_printf(Perl_debug_log, "%s",lead);
8730 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
8735 PerlIO_printf(Perl_debug_log, "\n");
8737 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
8743 Perl_regdump(pTHX_ const regexp *r)
8747 SV * const sv = sv_newmortal();
8748 SV *dsv= sv_newmortal();
8750 GET_RE_DEBUG_FLAGS_DECL;
8752 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8754 /* Header fields of interest. */
8755 if (r->anchored_substr) {
8756 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8757 RE_SV_DUMPLEN(r->anchored_substr), 30);
8758 PerlIO_printf(Perl_debug_log,
8759 "anchored %s%s at %"IVdf" ",
8760 s, RE_SV_TAIL(r->anchored_substr),
8761 (IV)r->anchored_offset);
8762 } else if (r->anchored_utf8) {
8763 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8764 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8765 PerlIO_printf(Perl_debug_log,
8766 "anchored utf8 %s%s at %"IVdf" ",
8767 s, RE_SV_TAIL(r->anchored_utf8),
8768 (IV)r->anchored_offset);
8770 if (r->float_substr) {
8771 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8772 RE_SV_DUMPLEN(r->float_substr), 30);
8773 PerlIO_printf(Perl_debug_log,
8774 "floating %s%s at %"IVdf"..%"UVuf" ",
8775 s, RE_SV_TAIL(r->float_substr),
8776 (IV)r->float_min_offset, (UV)r->float_max_offset);
8777 } else if (r->float_utf8) {
8778 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8779 RE_SV_DUMPLEN(r->float_utf8), 30);
8780 PerlIO_printf(Perl_debug_log,
8781 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8782 s, RE_SV_TAIL(r->float_utf8),
8783 (IV)r->float_min_offset, (UV)r->float_max_offset);
8785 if (r->check_substr || r->check_utf8)
8786 PerlIO_printf(Perl_debug_log,
8788 (r->check_substr == r->float_substr
8789 && r->check_utf8 == r->float_utf8
8790 ? "(checking floating" : "(checking anchored"));
8791 if (r->extflags & RXf_NOSCAN)
8792 PerlIO_printf(Perl_debug_log, " noscan");
8793 if (r->extflags & RXf_CHECK_ALL)
8794 PerlIO_printf(Perl_debug_log, " isall");
8795 if (r->check_substr || r->check_utf8)
8796 PerlIO_printf(Perl_debug_log, ") ");
8798 if (ri->regstclass) {
8799 regprop(r, sv, ri->regstclass);
8800 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8802 if (r->extflags & RXf_ANCH) {
8803 PerlIO_printf(Perl_debug_log, "anchored");
8804 if (r->extflags & RXf_ANCH_BOL)
8805 PerlIO_printf(Perl_debug_log, "(BOL)");
8806 if (r->extflags & RXf_ANCH_MBOL)
8807 PerlIO_printf(Perl_debug_log, "(MBOL)");
8808 if (r->extflags & RXf_ANCH_SBOL)
8809 PerlIO_printf(Perl_debug_log, "(SBOL)");
8810 if (r->extflags & RXf_ANCH_GPOS)
8811 PerlIO_printf(Perl_debug_log, "(GPOS)");
8812 PerlIO_putc(Perl_debug_log, ' ');
8814 if (r->extflags & RXf_GPOS_SEEN)
8815 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8816 if (r->intflags & PREGf_SKIP)
8817 PerlIO_printf(Perl_debug_log, "plus ");
8818 if (r->intflags & PREGf_IMPLICIT)
8819 PerlIO_printf(Perl_debug_log, "implicit ");
8820 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8821 if (r->extflags & RXf_EVAL_SEEN)
8822 PerlIO_printf(Perl_debug_log, "with eval ");
8823 PerlIO_printf(Perl_debug_log, "\n");
8824 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
8826 PERL_UNUSED_CONTEXT;
8828 #endif /* DEBUGGING */
8832 - regprop - printable representation of opcode
8835 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8840 RXi_GET_DECL(prog,progi);
8841 GET_RE_DEBUG_FLAGS_DECL;
8844 sv_setpvn(sv, "", 0);
8846 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8847 /* It would be nice to FAIL() here, but this may be called from
8848 regexec.c, and it would be hard to supply pRExC_state. */
8849 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8850 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
8852 k = PL_regkind[OP(o)];
8856 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8857 * is a crude hack but it may be the best for now since
8858 * we have no flag "this EXACTish node was UTF-8"
8860 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
8861 PERL_PV_ESCAPE_UNI_DETECT |
8862 PERL_PV_PRETTY_ELLIPSES |
8863 PERL_PV_PRETTY_LTGT |
8864 PERL_PV_PRETTY_NOCLEAR
8866 } else if (k == TRIE) {
8867 /* print the details of the trie in dumpuntil instead, as
8868 * progi->data isn't available here */
8869 const char op = OP(o);
8870 const U32 n = ARG(o);
8871 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8872 (reg_ac_data *)progi->data->data[n] :
8874 const reg_trie_data * const trie
8875 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8877 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
8878 DEBUG_TRIE_COMPILE_r(
8879 Perl_sv_catpvf(aTHX_ sv,
8880 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8881 (UV)trie->startstate,
8882 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8883 (UV)trie->wordcount,
8886 (UV)TRIE_CHARCOUNT(trie),
8887 (UV)trie->uniquecharcount
8890 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8892 int rangestart = -1;
8893 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8895 for (i = 0; i <= 256; i++) {
8896 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8897 if (rangestart == -1)
8899 } else if (rangestart != -1) {
8900 if (i <= rangestart + 3)
8901 for (; rangestart < i; rangestart++)
8902 put_byte(sv, rangestart);
8904 put_byte(sv, rangestart);
8906 put_byte(sv, i - 1);
8914 } else if (k == CURLY) {
8915 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8916 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8917 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8919 else if (k == WHILEM && o->flags) /* Ordinal/of */
8920 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8921 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8922 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8923 if ( prog->paren_names ) {
8924 if ( k != REF || OP(o) < NREF) {
8925 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8926 SV **name= av_fetch(list, ARG(o), 0 );
8928 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8931 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8932 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8933 I32 *nums=(I32*)SvPVX(sv_dat);
8934 SV **name= av_fetch(list, nums[0], 0 );
8937 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8938 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8939 (n ? "," : ""), (IV)nums[n]);
8941 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8945 } else if (k == GOSUB)
8946 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8947 else if (k == VERB) {
8949 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8950 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8951 } else if (k == LOGICAL)
8952 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8953 else if (k == FOLDCHAR)
8954 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
8955 else if (k == ANYOF) {
8956 int i, rangestart = -1;
8957 const U8 flags = ANYOF_FLAGS(o);
8959 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8960 static const char * const anyofs[] = {
8993 if (flags & ANYOF_LOCALE)
8994 sv_catpvs(sv, "{loc}");
8995 if (flags & ANYOF_FOLD)
8996 sv_catpvs(sv, "{i}");
8997 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8998 if (flags & ANYOF_INVERT)
9000 for (i = 0; i <= 256; i++) {
9001 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9002 if (rangestart == -1)
9004 } else if (rangestart != -1) {
9005 if (i <= rangestart + 3)
9006 for (; rangestart < i; rangestart++)
9007 put_byte(sv, rangestart);
9009 put_byte(sv, rangestart);
9011 put_byte(sv, i - 1);
9017 if (o->flags & ANYOF_CLASS)
9018 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9019 if (ANYOF_CLASS_TEST(o,i))
9020 sv_catpv(sv, anyofs[i]);
9022 if (flags & ANYOF_UNICODE)
9023 sv_catpvs(sv, "{unicode}");
9024 else if (flags & ANYOF_UNICODE_ALL)
9025 sv_catpvs(sv, "{unicode_all}");
9029 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9033 U8 s[UTF8_MAXBYTES_CASE+1];
9035 for (i = 0; i <= 256; i++) { /* just the first 256 */
9036 uvchr_to_utf8(s, i);
9038 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9039 if (rangestart == -1)
9041 } else if (rangestart != -1) {
9042 if (i <= rangestart + 3)
9043 for (; rangestart < i; rangestart++) {
9044 const U8 * const e = uvchr_to_utf8(s,rangestart);
9046 for(p = s; p < e; p++)
9050 const U8 *e = uvchr_to_utf8(s,rangestart);
9052 for (p = s; p < e; p++)
9055 e = uvchr_to_utf8(s, i-1);
9056 for (p = s; p < e; p++)
9063 sv_catpvs(sv, "..."); /* et cetera */
9067 char *s = savesvpv(lv);
9068 char * const origs = s;
9070 while (*s && *s != '\n')
9074 const char * const t = ++s;
9092 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9094 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9095 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9097 PERL_UNUSED_CONTEXT;
9098 PERL_UNUSED_ARG(sv);
9100 PERL_UNUSED_ARG(prog);
9101 #endif /* DEBUGGING */
9105 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9106 { /* Assume that RE_INTUIT is set */
9108 struct regexp *const prog = (struct regexp *)SvANY(r);
9109 GET_RE_DEBUG_FLAGS_DECL;
9110 PERL_UNUSED_CONTEXT;
9114 const char * const s = SvPV_nolen_const(prog->check_substr
9115 ? prog->check_substr : prog->check_utf8);
9117 if (!PL_colorset) reginitcolors();
9118 PerlIO_printf(Perl_debug_log,
9119 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9121 prog->check_substr ? "" : "utf8 ",
9122 PL_colors[5],PL_colors[0],
9125 (strlen(s) > 60 ? "..." : ""));
9128 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9134 handles refcounting and freeing the perl core regexp structure. When
9135 it is necessary to actually free the structure the first thing it
9136 does is call the 'free' method of the regexp_engine associated to to
9137 the regexp, allowing the handling of the void *pprivate; member
9138 first. (This routine is not overridable by extensions, which is why
9139 the extensions free is called first.)
9141 See regdupe and regdupe_internal if you change anything here.
9143 #ifndef PERL_IN_XSUB_RE
9145 Perl_pregfree(pTHX_ REGEXP *r)
9151 Perl_pregfree2(pTHX_ REGEXP *rx)
9154 struct regexp *const r = (struct regexp *)SvANY(rx);
9155 GET_RE_DEBUG_FLAGS_DECL;
9158 ReREFCNT_dec(r->mother_re);
9160 CALLREGFREE_PVT(rx); /* free the private data */
9162 SvREFCNT_dec(r->paren_names);
9165 if (r->anchored_substr)
9166 SvREFCNT_dec(r->anchored_substr);
9167 if (r->anchored_utf8)
9168 SvREFCNT_dec(r->anchored_utf8);
9169 if (r->float_substr)
9170 SvREFCNT_dec(r->float_substr);
9172 SvREFCNT_dec(r->float_utf8);
9173 Safefree(r->substrs);
9175 RX_MATCH_COPY_FREE(rx);
9176 #ifdef PERL_OLD_COPY_ON_WRITE
9178 SvREFCNT_dec(r->saved_copy);
9186 This is a hacky workaround to the structural issue of match results
9187 being stored in the regexp structure which is in turn stored in
9188 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9189 could be PL_curpm in multiple contexts, and could require multiple
9190 result sets being associated with the pattern simultaneously, such
9191 as when doing a recursive match with (??{$qr})
9193 The solution is to make a lightweight copy of the regexp structure
9194 when a qr// is returned from the code executed by (??{$qr}) this
9195 lightweight copy doesnt actually own any of its data except for
9196 the starp/end and the actual regexp structure itself.
9202 Perl_reg_temp_copy (pTHX_ REGEXP *rx) {
9203 REGEXP *ret_x = newSV_type(SVt_REGEXP);
9204 struct regexp *ret = (struct regexp *)SvANY(ret_x);
9205 struct regexp *const r = (struct regexp *)SvANY(rx);
9206 register const I32 npar = r->nparens+1;
9207 (void)ReREFCNT_inc(rx);
9208 /* FIXME ORANGE (once we start actually using the regular SV fields.) */
9209 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9210 by pointing directly at the buffer, but flagging that the allocated
9211 space in the copy is zero. As we've just done a struct copy, it's now
9212 a case of zero-ing that, rather than copying the current length. */
9213 SvPV_set(ret_x, RX_WRAPPED(rx));
9214 StructCopy(r, ret, regexp);
9215 SvLEN_set(ret_x, 0);
9216 Newx(ret->offs, npar, regexp_paren_pair);
9217 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9219 Newx(ret->substrs, 1, struct reg_substr_data);
9220 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9222 SvREFCNT_inc_void(ret->anchored_substr);
9223 SvREFCNT_inc_void(ret->anchored_utf8);
9224 SvREFCNT_inc_void(ret->float_substr);
9225 SvREFCNT_inc_void(ret->float_utf8);
9227 /* check_substr and check_utf8, if non-NULL, point to either their
9228 anchored or float namesakes, and don't hold a second reference. */
9230 RX_MATCH_COPIED_off(ret_x);
9231 #ifdef PERL_OLD_COPY_ON_WRITE
9232 ret->saved_copy = NULL;
9234 ret->mother_re = rx;
9241 /* regfree_internal()
9243 Free the private data in a regexp. This is overloadable by
9244 extensions. Perl takes care of the regexp structure in pregfree(),
9245 this covers the *pprivate pointer which technically perldoesnt
9246 know about, however of course we have to handle the
9247 regexp_internal structure when no extension is in use.
9249 Note this is called before freeing anything in the regexp
9254 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9257 struct regexp *const r = (struct regexp *)SvANY(rx);
9259 GET_RE_DEBUG_FLAGS_DECL;
9265 SV *dsv= sv_newmortal();
9266 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9267 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9268 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9269 PL_colors[4],PL_colors[5],s);
9272 #ifdef RE_TRACK_PATTERN_OFFSETS
9274 Safefree(ri->u.offsets); /* 20010421 MJD */
9277 int n = ri->data->count;
9278 PAD* new_comppad = NULL;
9283 /* If you add a ->what type here, update the comment in regcomp.h */
9284 switch (ri->data->what[n]) {
9288 SvREFCNT_dec((SV*)ri->data->data[n]);
9291 Safefree(ri->data->data[n]);
9294 new_comppad = (AV*)ri->data->data[n];
9297 if (new_comppad == NULL)
9298 Perl_croak(aTHX_ "panic: pregfree comppad");
9299 PAD_SAVE_LOCAL(old_comppad,
9300 /* Watch out for global destruction's random ordering. */
9301 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9304 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9307 op_free((OP_4tree*)ri->data->data[n]);
9309 PAD_RESTORE_LOCAL(old_comppad);
9310 SvREFCNT_dec((SV*)new_comppad);
9316 { /* Aho Corasick add-on structure for a trie node.
9317 Used in stclass optimization only */
9319 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9321 refcount = --aho->refcount;
9324 PerlMemShared_free(aho->states);
9325 PerlMemShared_free(aho->fail);
9326 /* do this last!!!! */
9327 PerlMemShared_free(ri->data->data[n]);
9328 PerlMemShared_free(ri->regstclass);
9334 /* trie structure. */
9336 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9338 refcount = --trie->refcount;
9341 PerlMemShared_free(trie->charmap);
9342 PerlMemShared_free(trie->states);
9343 PerlMemShared_free(trie->trans);
9345 PerlMemShared_free(trie->bitmap);
9347 PerlMemShared_free(trie->wordlen);
9349 PerlMemShared_free(trie->jump);
9351 PerlMemShared_free(trie->nextword);
9352 /* do this last!!!! */
9353 PerlMemShared_free(ri->data->data[n]);
9358 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9361 Safefree(ri->data->what);
9368 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
9369 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9370 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9371 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9374 re_dup - duplicate a regexp.
9376 This routine is expected to clone a given regexp structure. It is not
9377 compiler under USE_ITHREADS.
9379 After all of the core data stored in struct regexp is duplicated
9380 the regexp_engine.dupe method is used to copy any private data
9381 stored in the *pprivate pointer. This allows extensions to handle
9382 any duplication it needs to do.
9384 See pregfree() and regfree_internal() if you change anything here.
9386 #if defined(USE_ITHREADS)
9387 #ifndef PERL_IN_XSUB_RE
9389 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9393 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9394 struct regexp *ret = (struct regexp *)SvANY(dstr);
9396 npar = r->nparens+1;
9397 Newx(ret->offs, npar, regexp_paren_pair);
9398 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9400 /* no need to copy these */
9401 Newx(ret->swap, npar, regexp_paren_pair);
9405 /* Do it this way to avoid reading from *r after the StructCopy().
9406 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9407 cache, it doesn't matter. */
9408 const bool anchored = r->check_substr == r->anchored_substr;
9409 Newx(ret->substrs, 1, struct reg_substr_data);
9410 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9412 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9413 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9414 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9415 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9417 /* check_substr and check_utf8, if non-NULL, point to either their
9418 anchored or float namesakes, and don't hold a second reference. */
9420 if (ret->check_substr) {
9422 assert(r->check_utf8 == r->anchored_utf8);
9423 ret->check_substr = ret->anchored_substr;
9424 ret->check_utf8 = ret->anchored_utf8;
9426 assert(r->check_substr == r->float_substr);
9427 assert(r->check_utf8 == r->float_utf8);
9428 ret->check_substr = ret->float_substr;
9429 ret->check_utf8 = ret->float_utf8;
9434 ret->paren_names = hv_dup_inc(ret->paren_names, param);
9437 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9439 if (RX_MATCH_COPIED(dstr))
9440 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9443 #ifdef PERL_OLD_COPY_ON_WRITE
9444 ret->saved_copy = NULL;
9447 ret->mother_re = NULL;
9449 ret->seen_evals = 0;
9451 #endif /* PERL_IN_XSUB_RE */
9456 This is the internal complement to regdupe() which is used to copy
9457 the structure pointed to by the *pprivate pointer in the regexp.
9458 This is the core version of the extension overridable cloning hook.
9459 The regexp structure being duplicated will be copied by perl prior
9460 to this and will be provided as the regexp *r argument, however
9461 with the /old/ structures pprivate pointer value. Thus this routine
9462 may override any copying normally done by perl.
9464 It returns a pointer to the new regexp_internal structure.
9468 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
9471 struct regexp *const r = (struct regexp *)SvANY(rx);
9472 regexp_internal *reti;
9476 npar = r->nparens+1;
9479 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
9480 Copy(ri->program, reti->program, len+1, regnode);
9483 reti->regstclass = NULL;
9487 const int count = ri->data->count;
9490 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9491 char, struct reg_data);
9492 Newx(d->what, count, U8);
9495 for (i = 0; i < count; i++) {
9496 d->what[i] = ri->data->what[i];
9497 switch (d->what[i]) {
9498 /* legal options are one of: sSfpontTu
9499 see also regcomp.h and pregfree() */
9502 case 'p': /* actually an AV, but the dup function is identical. */
9503 case 'u': /* actually an HV, but the dup function is identical. */
9504 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
9507 /* This is cheating. */
9508 Newx(d->data[i], 1, struct regnode_charclass_class);
9509 StructCopy(ri->data->data[i], d->data[i],
9510 struct regnode_charclass_class);
9511 reti->regstclass = (regnode*)d->data[i];
9514 /* Compiled op trees are readonly and in shared memory,
9515 and can thus be shared without duplication. */
9517 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9521 /* Trie stclasses are readonly and can thus be shared
9522 * without duplication. We free the stclass in pregfree
9523 * when the corresponding reg_ac_data struct is freed.
9525 reti->regstclass= ri->regstclass;
9529 ((reg_trie_data*)ri->data->data[i])->refcount++;
9533 d->data[i] = ri->data->data[i];
9536 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9545 reti->name_list_idx = ri->name_list_idx;
9547 #ifdef RE_TRACK_PATTERN_OFFSETS
9548 if (ri->u.offsets) {
9549 Newx(reti->u.offsets, 2*len+1, U32);
9550 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9553 SetProgLen(reti,len);
9559 #endif /* USE_ITHREADS */
9564 converts a regexp embedded in a MAGIC struct to its stringified form,
9565 caching the converted form in the struct and returns the cached
9568 If lp is nonnull then it is used to return the length of the
9571 If flags is nonnull and the returned string contains UTF8 then
9572 (*flags & 1) will be true.
9574 If haseval is nonnull then it is used to return whether the pattern
9577 Normally called via macro:
9579 CALLREG_STRINGIFY(mg,&len,&utf8);
9583 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9585 See sv_2pv_flags() in sv.c for an example of internal usage.
9588 #ifndef PERL_IN_XSUB_RE
9591 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9593 const REGEXP * const re = (REGEXP *)mg->mg_obj;
9595 *haseval = RX_SEEN_EVALS(re);
9597 *flags = RX_UTF8(re) ? 1 : 0;
9599 *lp = RX_WRAPLEN(re);
9600 return RX_WRAPPED(re);
9604 - regnext - dig the "next" pointer out of a node
9607 Perl_regnext(pTHX_ register regnode *p)
9610 register I32 offset;
9615 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9624 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9627 STRLEN l1 = strlen(pat1);
9628 STRLEN l2 = strlen(pat2);
9631 const char *message;
9637 Copy(pat1, buf, l1 , char);
9638 Copy(pat2, buf + l1, l2 , char);
9639 buf[l1 + l2] = '\n';
9640 buf[l1 + l2 + 1] = '\0';
9642 /* ANSI variant takes additional second argument */
9643 va_start(args, pat2);
9647 msv = vmess(buf, &args);
9649 message = SvPV_const(msv,l1);
9652 Copy(message, buf, l1 , char);
9653 buf[l1-1] = '\0'; /* Overwrite \n */
9654 Perl_croak(aTHX_ "%s", buf);
9657 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9659 #ifndef PERL_IN_XSUB_RE
9661 Perl_save_re_context(pTHX)
9665 struct re_save_state *state;
9667 SAVEVPTR(PL_curcop);
9668 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9670 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9671 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9672 SSPUSHINT(SAVEt_RE_STATE);
9674 Copy(&PL_reg_state, state, 1, struct re_save_state);
9676 PL_reg_start_tmp = 0;
9677 PL_reg_start_tmpl = 0;
9678 PL_reg_oldsaved = NULL;
9679 PL_reg_oldsavedlen = 0;
9681 PL_reg_leftiter = 0;
9682 PL_reg_poscache = NULL;
9683 PL_reg_poscache_size = 0;
9684 #ifdef PERL_OLD_COPY_ON_WRITE
9688 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9690 const REGEXP * const rx = PM_GETRE(PL_curpm);
9693 for (i = 1; i <= RX_NPARENS(rx); i++) {
9694 char digits[TYPE_CHARS(long)];
9695 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9696 GV *const *const gvp
9697 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9700 GV * const gv = *gvp;
9701 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9711 clear_re(pTHX_ void *r)
9714 ReREFCNT_dec((REGEXP *)r);
9720 S_put_byte(pTHX_ SV *sv, int c)
9722 /* Our definition of isPRINT() ignores locales, so only bytes that are
9723 not part of UTF-8 are considered printable. I assume that the same
9724 holds for UTF-EBCDIC.
9725 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
9726 which Wikipedia says:
9728 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
9729 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
9730 identical, to the ASCII delete (DEL) or rubout control character.
9731 ) So the old condition can be simplified to !isPRINT(c) */
9733 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9735 const char string = c;
9736 if (c == '-' || c == ']' || c == '\\' || c == '^')
9737 sv_catpvs(sv, "\\");
9738 sv_catpvn(sv, &string, 1);
9743 #define CLEAR_OPTSTART \
9744 if (optstart) STMT_START { \
9745 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9749 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9751 STATIC const regnode *
9752 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9753 const regnode *last, const regnode *plast,
9754 SV* sv, I32 indent, U32 depth)
9757 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9758 register const regnode *next;
9759 const regnode *optstart= NULL;
9762 GET_RE_DEBUG_FLAGS_DECL;
9764 #ifdef DEBUG_DUMPUNTIL
9765 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9766 last ? last-start : 0,plast ? plast-start : 0);
9769 if (plast && plast < last)
9772 while (PL_regkind[op] != END && (!last || node < last)) {
9773 /* While that wasn't END last time... */
9776 if (op == CLOSE || op == WHILEM)
9778 next = regnext((regnode *)node);
9781 if (OP(node) == OPTIMIZED) {
9782 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9789 regprop(r, sv, node);
9790 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9791 (int)(2*indent + 1), "", SvPVX_const(sv));
9793 if (OP(node) != OPTIMIZED) {
9794 if (next == NULL) /* Next ptr. */
9795 PerlIO_printf(Perl_debug_log, " (0)");
9796 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9797 PerlIO_printf(Perl_debug_log, " (FAIL)");
9799 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9800 (void)PerlIO_putc(Perl_debug_log, '\n');
9804 if (PL_regkind[(U8)op] == BRANCHJ) {
9807 register const regnode *nnode = (OP(next) == LONGJMP
9808 ? regnext((regnode *)next)
9810 if (last && nnode > last)
9812 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9815 else if (PL_regkind[(U8)op] == BRANCH) {
9817 DUMPUNTIL(NEXTOPER(node), next);
9819 else if ( PL_regkind[(U8)op] == TRIE ) {
9820 const regnode *this_trie = node;
9821 const char op = OP(node);
9822 const U32 n = ARG(node);
9823 const reg_ac_data * const ac = op>=AHOCORASICK ?
9824 (reg_ac_data *)ri->data->data[n] :
9826 const reg_trie_data * const trie =
9827 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9829 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9831 const regnode *nextbranch= NULL;
9833 sv_setpvn(sv, "", 0);
9834 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9835 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9837 PerlIO_printf(Perl_debug_log, "%*s%s ",
9838 (int)(2*(indent+3)), "",
9839 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9840 PL_colors[0], PL_colors[1],
9841 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9842 PERL_PV_PRETTY_ELLIPSES |
9848 U16 dist= trie->jump[word_idx+1];
9849 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9850 (UV)((dist ? this_trie + dist : next) - start));
9853 nextbranch= this_trie + trie->jump[0];
9854 DUMPUNTIL(this_trie + dist, nextbranch);
9856 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9857 nextbranch= regnext((regnode *)nextbranch);
9859 PerlIO_printf(Perl_debug_log, "\n");
9862 if (last && next > last)
9867 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9868 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9869 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9871 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9873 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9875 else if ( op == PLUS || op == STAR) {
9876 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9878 else if (op == ANYOF) {
9879 /* arglen 1 + class block */
9880 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9881 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9882 node = NEXTOPER(node);
9884 else if (PL_regkind[(U8)op] == EXACT) {
9885 /* Literal string, where present. */
9886 node += NODE_SZ_STR(node) - 1;
9887 node = NEXTOPER(node);
9890 node = NEXTOPER(node);
9891 node += regarglen[(U8)op];
9893 if (op == CURLYX || op == OPEN)
9897 #ifdef DEBUG_DUMPUNTIL
9898 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9903 #endif /* DEBUGGING */
9907 * c-indentation-style: bsd
9909 * indent-tabs-mode: t
9912 * ex: set ts=8 sts=4 sw=4 noet:
projects / p5sagit/p5-mst-13.2.git / blob