5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
93 # if defined(BUGGY_MSC6)
94 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
95 # pragma optimize("a",off)
96 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
97 # pragma optimize("w",on )
98 # endif /* BUGGY_MSC6 */
102 #define STATIC static
105 typedef struct RExC_state_t {
106 U32 flags; /* are we folding, multilining? */
107 char *precomp; /* uncompiled string. */
108 regexp *rx; /* perl core regexp structure */
109 regexp_internal *rxi; /* internal data for regexp object pprivate field */
110 char *start; /* Start of input for compile */
111 char *end; /* End of input for compile */
112 char *parse; /* Input-scan pointer. */
113 I32 whilem_seen; /* number of WHILEM in this expr */
114 regnode *emit_start; /* Start of emitted-code area */
115 regnode *emit_bound; /* First regnode outside of the allocated space */
116 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
117 I32 naughty; /* How bad is this pattern? */
118 I32 sawback; /* Did we see \1, ...? */
120 I32 size; /* Code size. */
121 I32 npar; /* Capture buffer count, (OPEN). */
122 I32 cpar; /* Capture buffer count, (CLOSE). */
123 I32 nestroot; /* root parens we are in - used by accept */
127 regnode **open_parens; /* pointers to open parens */
128 regnode **close_parens; /* pointers to close parens */
129 regnode *opend; /* END node in program */
130 I32 utf8; /* whether the pattern is utf8 or not */
131 I32 orig_utf8; /* whether the pattern was originally in utf8 */
132 /* XXX use this for future optimisation of case
133 * where pattern must be upgraded to utf8. */
134 HV *charnames; /* cache of named sequences */
135 HV *paren_names; /* Paren names */
137 regnode **recurse; /* Recurse regops */
138 I32 recurse_count; /* Number of recurse regops */
140 char *starttry; /* -Dr: where regtry was called. */
141 #define RExC_starttry (pRExC_state->starttry)
144 const char *lastparse;
146 AV *paren_name_list; /* idx -> name */
147 #define RExC_lastparse (pRExC_state->lastparse)
148 #define RExC_lastnum (pRExC_state->lastnum)
149 #define RExC_paren_name_list (pRExC_state->paren_name_list)
153 #define RExC_flags (pRExC_state->flags)
154 #define RExC_precomp (pRExC_state->precomp)
155 #define RExC_rx (pRExC_state->rx)
156 #define RExC_rxi (pRExC_state->rxi)
157 #define RExC_start (pRExC_state->start)
158 #define RExC_end (pRExC_state->end)
159 #define RExC_parse (pRExC_state->parse)
160 #define RExC_whilem_seen (pRExC_state->whilem_seen)
161 #ifdef RE_TRACK_PATTERN_OFFSETS
162 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
164 #define RExC_emit (pRExC_state->emit)
165 #define RExC_emit_start (pRExC_state->emit_start)
166 #define RExC_emit_bound (pRExC_state->emit_bound)
167 #define RExC_naughty (pRExC_state->naughty)
168 #define RExC_sawback (pRExC_state->sawback)
169 #define RExC_seen (pRExC_state->seen)
170 #define RExC_size (pRExC_state->size)
171 #define RExC_npar (pRExC_state->npar)
172 #define RExC_nestroot (pRExC_state->nestroot)
173 #define RExC_extralen (pRExC_state->extralen)
174 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
175 #define RExC_seen_evals (pRExC_state->seen_evals)
176 #define RExC_utf8 (pRExC_state->utf8)
177 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
178 #define RExC_charnames (pRExC_state->charnames)
179 #define RExC_open_parens (pRExC_state->open_parens)
180 #define RExC_close_parens (pRExC_state->close_parens)
181 #define RExC_opend (pRExC_state->opend)
182 #define RExC_paren_names (pRExC_state->paren_names)
183 #define RExC_recurse (pRExC_state->recurse)
184 #define RExC_recurse_count (pRExC_state->recurse_count)
187 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
188 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
189 ((*s) == '{' && regcurly(s)))
192 #undef SPSTART /* dratted cpp namespace... */
195 * Flags to be passed up and down.
197 #define WORST 0 /* Worst case. */
198 #define HASWIDTH 0x01 /* Known to match non-null strings. */
199 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
200 #define SPSTART 0x04 /* Starts with * or +. */
201 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
202 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
204 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
206 /* whether trie related optimizations are enabled */
207 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
208 #define TRIE_STUDY_OPT
209 #define FULL_TRIE_STUDY
215 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
216 #define PBITVAL(paren) (1 << ((paren) & 7))
217 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
218 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
219 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
222 /* About scan_data_t.
224 During optimisation we recurse through the regexp program performing
225 various inplace (keyhole style) optimisations. In addition study_chunk
226 and scan_commit populate this data structure with information about
227 what strings MUST appear in the pattern. We look for the longest
228 string that must appear for at a fixed location, and we look for the
229 longest string that may appear at a floating location. So for instance
234 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
235 strings (because they follow a .* construct). study_chunk will identify
236 both FOO and BAR as being the longest fixed and floating strings respectively.
238 The strings can be composites, for instance
242 will result in a composite fixed substring 'foo'.
244 For each string some basic information is maintained:
246 - offset or min_offset
247 This is the position the string must appear at, or not before.
248 It also implicitly (when combined with minlenp) tells us how many
249 character must match before the string we are searching.
250 Likewise when combined with minlenp and the length of the string
251 tells us how many characters must appear after the string we have
255 Only used for floating strings. This is the rightmost point that
256 the string can appear at. Ifset to I32 max it indicates that the
257 string can occur infinitely far to the right.
260 A pointer to the minimum length of the pattern that the string
261 was found inside. This is important as in the case of positive
262 lookahead or positive lookbehind we can have multiple patterns
267 The minimum length of the pattern overall is 3, the minimum length
268 of the lookahead part is 3, but the minimum length of the part that
269 will actually match is 1. So 'FOO's minimum length is 3, but the
270 minimum length for the F is 1. This is important as the minimum length
271 is used to determine offsets in front of and behind the string being
272 looked for. Since strings can be composites this is the length of the
273 pattern at the time it was commited with a scan_commit. Note that
274 the length is calculated by study_chunk, so that the minimum lengths
275 are not known until the full pattern has been compiled, thus the
276 pointer to the value.
280 In the case of lookbehind the string being searched for can be
281 offset past the start point of the final matching string.
282 If this value was just blithely removed from the min_offset it would
283 invalidate some of the calculations for how many chars must match
284 before or after (as they are derived from min_offset and minlen and
285 the length of the string being searched for).
286 When the final pattern is compiled and the data is moved from the
287 scan_data_t structure into the regexp structure the information
288 about lookbehind is factored in, with the information that would
289 have been lost precalculated in the end_shift field for the
292 The fields pos_min and pos_delta are used to store the minimum offset
293 and the delta to the maximum offset at the current point in the pattern.
297 typedef struct scan_data_t {
298 /*I32 len_min; unused */
299 /*I32 len_delta; unused */
303 I32 last_end; /* min value, <0 unless valid. */
306 SV **longest; /* Either &l_fixed, or &l_float. */
307 SV *longest_fixed; /* longest fixed string found in pattern */
308 I32 offset_fixed; /* offset where it starts */
309 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
310 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
311 SV *longest_float; /* longest floating string found in pattern */
312 I32 offset_float_min; /* earliest point in string it can appear */
313 I32 offset_float_max; /* latest point in string it can appear */
314 I32 *minlen_float; /* pointer to the minlen relevent to the string */
315 I32 lookbehind_float; /* is the position of the string modified by LB */
319 struct regnode_charclass_class *start_class;
323 * Forward declarations for pregcomp()'s friends.
326 static const scan_data_t zero_scan_data =
327 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
329 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
330 #define SF_BEFORE_SEOL 0x0001
331 #define SF_BEFORE_MEOL 0x0002
332 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
333 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
336 # define SF_FIX_SHIFT_EOL (0+2)
337 # define SF_FL_SHIFT_EOL (0+4)
339 # define SF_FIX_SHIFT_EOL (+2)
340 # define SF_FL_SHIFT_EOL (+4)
343 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
344 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
346 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
347 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
348 #define SF_IS_INF 0x0040
349 #define SF_HAS_PAR 0x0080
350 #define SF_IN_PAR 0x0100
351 #define SF_HAS_EVAL 0x0200
352 #define SCF_DO_SUBSTR 0x0400
353 #define SCF_DO_STCLASS_AND 0x0800
354 #define SCF_DO_STCLASS_OR 0x1000
355 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
356 #define SCF_WHILEM_VISITED_POS 0x2000
358 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
359 #define SCF_SEEN_ACCEPT 0x8000
361 #define UTF (RExC_utf8 != 0)
362 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
363 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
365 #define OOB_UNICODE 12345678
366 #define OOB_NAMEDCLASS -1
368 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
369 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
372 /* length of regex to show in messages that don't mark a position within */
373 #define RegexLengthToShowInErrorMessages 127
376 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
377 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
378 * op/pragma/warn/regcomp.
380 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
381 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
383 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
386 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
387 * arg. Show regex, up to a maximum length. If it's too long, chop and add
390 #define _FAIL(code) STMT_START { \
391 const char *ellipses = ""; \
392 IV len = RExC_end - RExC_precomp; \
395 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
396 if (len > RegexLengthToShowInErrorMessages) { \
397 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
398 len = RegexLengthToShowInErrorMessages - 10; \
404 #define FAIL(msg) _FAIL( \
405 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
406 msg, (int)len, RExC_precomp, ellipses))
408 #define FAIL2(msg,arg) _FAIL( \
409 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
410 arg, (int)len, RExC_precomp, ellipses))
413 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
415 #define Simple_vFAIL(m) STMT_START { \
416 const IV offset = RExC_parse - RExC_precomp; \
417 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
418 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
422 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
424 #define vFAIL(m) STMT_START { \
426 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
431 * Like Simple_vFAIL(), but accepts two arguments.
433 #define Simple_vFAIL2(m,a1) STMT_START { \
434 const IV offset = RExC_parse - RExC_precomp; \
435 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
436 (int)offset, RExC_precomp, RExC_precomp + offset); \
440 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
442 #define vFAIL2(m,a1) STMT_START { \
444 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
445 Simple_vFAIL2(m, a1); \
450 * Like Simple_vFAIL(), but accepts three arguments.
452 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
453 const IV offset = RExC_parse - RExC_precomp; \
454 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
455 (int)offset, RExC_precomp, RExC_precomp + offset); \
459 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
461 #define vFAIL3(m,a1,a2) STMT_START { \
463 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
464 Simple_vFAIL3(m, a1, a2); \
468 * Like Simple_vFAIL(), but accepts four arguments.
470 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
471 const IV offset = RExC_parse - RExC_precomp; \
472 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
473 (int)offset, RExC_precomp, RExC_precomp + offset); \
476 #define vWARN(loc,m) STMT_START { \
477 const IV offset = loc - RExC_precomp; \
478 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
479 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
482 #define vWARNdep(loc,m) STMT_START { \
483 const IV offset = loc - RExC_precomp; \
484 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
485 "%s" REPORT_LOCATION, \
486 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
490 #define vWARN2(loc, m, a1) STMT_START { \
491 const IV offset = loc - RExC_precomp; \
492 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
493 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
496 #define vWARN3(loc, m, a1, a2) STMT_START { \
497 const IV offset = loc - RExC_precomp; \
498 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
499 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
502 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
503 const IV offset = loc - RExC_precomp; \
504 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
505 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
508 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
509 const IV offset = loc - RExC_precomp; \
510 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
511 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
515 /* Allow for side effects in s */
516 #define REGC(c,s) STMT_START { \
517 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
520 /* Macros for recording node offsets. 20001227 mjd@plover.com
521 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
522 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
523 * Element 0 holds the number n.
524 * Position is 1 indexed.
526 #ifndef RE_TRACK_PATTERN_OFFSETS
527 #define Set_Node_Offset_To_R(node,byte)
528 #define Set_Node_Offset(node,byte)
529 #define Set_Cur_Node_Offset
530 #define Set_Node_Length_To_R(node,len)
531 #define Set_Node_Length(node,len)
532 #define Set_Node_Cur_Length(node)
533 #define Node_Offset(n)
534 #define Node_Length(n)
535 #define Set_Node_Offset_Length(node,offset,len)
536 #define ProgLen(ri) ri->u.proglen
537 #define SetProgLen(ri,x) ri->u.proglen = x
539 #define ProgLen(ri) ri->u.offsets[0]
540 #define SetProgLen(ri,x) ri->u.offsets[0] = x
541 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
543 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
544 __LINE__, (int)(node), (int)(byte))); \
546 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
548 RExC_offsets[2*(node)-1] = (byte); \
553 #define Set_Node_Offset(node,byte) \
554 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
555 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
557 #define Set_Node_Length_To_R(node,len) STMT_START { \
559 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
560 __LINE__, (int)(node), (int)(len))); \
562 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
564 RExC_offsets[2*(node)] = (len); \
569 #define Set_Node_Length(node,len) \
570 Set_Node_Length_To_R((node)-RExC_emit_start, len)
571 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
572 #define Set_Node_Cur_Length(node) \
573 Set_Node_Length(node, RExC_parse - parse_start)
575 /* Get offsets and lengths */
576 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
577 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
579 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
580 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
581 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
585 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
586 #define EXPERIMENTAL_INPLACESCAN
587 #endif /*RE_TRACK_PATTERN_OFFSETS*/
589 #define DEBUG_STUDYDATA(str,data,depth) \
590 DEBUG_OPTIMISE_MORE_r(if(data){ \
591 PerlIO_printf(Perl_debug_log, \
592 "%*s" str "Pos:%"IVdf"/%"IVdf \
593 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
594 (int)(depth)*2, "", \
595 (IV)((data)->pos_min), \
596 (IV)((data)->pos_delta), \
597 (UV)((data)->flags), \
598 (IV)((data)->whilem_c), \
599 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
600 is_inf ? "INF " : "" \
602 if ((data)->last_found) \
603 PerlIO_printf(Perl_debug_log, \
604 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
605 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
606 SvPVX_const((data)->last_found), \
607 (IV)((data)->last_end), \
608 (IV)((data)->last_start_min), \
609 (IV)((data)->last_start_max), \
610 ((data)->longest && \
611 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
612 SvPVX_const((data)->longest_fixed), \
613 (IV)((data)->offset_fixed), \
614 ((data)->longest && \
615 (data)->longest==&((data)->longest_float)) ? "*" : "", \
616 SvPVX_const((data)->longest_float), \
617 (IV)((data)->offset_float_min), \
618 (IV)((data)->offset_float_max) \
620 PerlIO_printf(Perl_debug_log,"\n"); \
623 static void clear_re(pTHX_ void *r);
625 /* Mark that we cannot extend a found fixed substring at this point.
626 Update the longest found anchored substring and the longest found
627 floating substrings if needed. */
630 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
632 const STRLEN l = CHR_SVLEN(data->last_found);
633 const STRLEN old_l = CHR_SVLEN(*data->longest);
634 GET_RE_DEBUG_FLAGS_DECL;
636 PERL_ARGS_ASSERT_SCAN_COMMIT;
638 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
639 SvSetMagicSV(*data->longest, data->last_found);
640 if (*data->longest == data->longest_fixed) {
641 data->offset_fixed = l ? data->last_start_min : data->pos_min;
642 if (data->flags & SF_BEFORE_EOL)
644 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
646 data->flags &= ~SF_FIX_BEFORE_EOL;
647 data->minlen_fixed=minlenp;
648 data->lookbehind_fixed=0;
650 else { /* *data->longest == data->longest_float */
651 data->offset_float_min = l ? data->last_start_min : data->pos_min;
652 data->offset_float_max = (l
653 ? data->last_start_max
654 : data->pos_min + data->pos_delta);
655 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
656 data->offset_float_max = I32_MAX;
657 if (data->flags & SF_BEFORE_EOL)
659 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
661 data->flags &= ~SF_FL_BEFORE_EOL;
662 data->minlen_float=minlenp;
663 data->lookbehind_float=0;
666 SvCUR_set(data->last_found, 0);
668 SV * const sv = data->last_found;
669 if (SvUTF8(sv) && SvMAGICAL(sv)) {
670 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
676 data->flags &= ~SF_BEFORE_EOL;
677 DEBUG_STUDYDATA("commit: ",data,0);
680 /* Can match anything (initialization) */
682 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
684 PERL_ARGS_ASSERT_CL_ANYTHING;
686 ANYOF_CLASS_ZERO(cl);
687 ANYOF_BITMAP_SETALL(cl);
688 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
690 cl->flags |= ANYOF_LOCALE;
693 /* Can match anything (initialization) */
695 S_cl_is_anything(const struct regnode_charclass_class *cl)
699 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
701 for (value = 0; value <= ANYOF_MAX; value += 2)
702 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
704 if (!(cl->flags & ANYOF_UNICODE_ALL))
706 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
711 /* Can match anything (initialization) */
713 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
715 PERL_ARGS_ASSERT_CL_INIT;
717 Zero(cl, 1, struct regnode_charclass_class);
719 cl_anything(pRExC_state, cl);
723 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
725 PERL_ARGS_ASSERT_CL_INIT_ZERO;
727 Zero(cl, 1, struct regnode_charclass_class);
729 cl_anything(pRExC_state, cl);
731 cl->flags |= ANYOF_LOCALE;
734 /* 'And' a given class with another one. Can create false positives */
735 /* We assume that cl is not inverted */
737 S_cl_and(struct regnode_charclass_class *cl,
738 const struct regnode_charclass_class *and_with)
740 PERL_ARGS_ASSERT_CL_AND;
742 assert(and_with->type == ANYOF);
743 if (!(and_with->flags & ANYOF_CLASS)
744 && !(cl->flags & ANYOF_CLASS)
745 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
746 && !(and_with->flags & ANYOF_FOLD)
747 && !(cl->flags & ANYOF_FOLD)) {
750 if (and_with->flags & ANYOF_INVERT)
751 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
752 cl->bitmap[i] &= ~and_with->bitmap[i];
754 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
755 cl->bitmap[i] &= and_with->bitmap[i];
756 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
757 if (!(and_with->flags & ANYOF_EOS))
758 cl->flags &= ~ANYOF_EOS;
760 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
761 !(and_with->flags & ANYOF_INVERT)) {
762 cl->flags &= ~ANYOF_UNICODE_ALL;
763 cl->flags |= ANYOF_UNICODE;
764 ARG_SET(cl, ARG(and_with));
766 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
767 !(and_with->flags & ANYOF_INVERT))
768 cl->flags &= ~ANYOF_UNICODE_ALL;
769 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
770 !(and_with->flags & ANYOF_INVERT))
771 cl->flags &= ~ANYOF_UNICODE;
774 /* 'OR' a given class with another one. Can create false positives */
775 /* We assume that cl is not inverted */
777 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
779 PERL_ARGS_ASSERT_CL_OR;
781 if (or_with->flags & ANYOF_INVERT) {
783 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
784 * <= (B1 | !B2) | (CL1 | !CL2)
785 * which is wasteful if CL2 is small, but we ignore CL2:
786 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
787 * XXXX Can we handle case-fold? Unclear:
788 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
789 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
791 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
792 && !(or_with->flags & ANYOF_FOLD)
793 && !(cl->flags & ANYOF_FOLD) ) {
796 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
797 cl->bitmap[i] |= ~or_with->bitmap[i];
798 } /* XXXX: logic is complicated otherwise */
800 cl_anything(pRExC_state, cl);
803 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
804 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
805 && (!(or_with->flags & ANYOF_FOLD)
806 || (cl->flags & ANYOF_FOLD)) ) {
809 /* OR char bitmap and class bitmap separately */
810 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
811 cl->bitmap[i] |= or_with->bitmap[i];
812 if (or_with->flags & ANYOF_CLASS) {
813 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
814 cl->classflags[i] |= or_with->classflags[i];
815 cl->flags |= ANYOF_CLASS;
818 else { /* XXXX: logic is complicated, leave it along for a moment. */
819 cl_anything(pRExC_state, cl);
822 if (or_with->flags & ANYOF_EOS)
823 cl->flags |= ANYOF_EOS;
825 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
826 ARG(cl) != ARG(or_with)) {
827 cl->flags |= ANYOF_UNICODE_ALL;
828 cl->flags &= ~ANYOF_UNICODE;
830 if (or_with->flags & ANYOF_UNICODE_ALL) {
831 cl->flags |= ANYOF_UNICODE_ALL;
832 cl->flags &= ~ANYOF_UNICODE;
836 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
837 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
838 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
839 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
844 dump_trie(trie,widecharmap,revcharmap)
845 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
846 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
848 These routines dump out a trie in a somewhat readable format.
849 The _interim_ variants are used for debugging the interim
850 tables that are used to generate the final compressed
851 representation which is what dump_trie expects.
853 Part of the reason for their existance is to provide a form
854 of documentation as to how the different representations function.
859 Dumps the final compressed table form of the trie to Perl_debug_log.
860 Used for debugging make_trie().
864 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
865 AV *revcharmap, U32 depth)
868 SV *sv=sv_newmortal();
869 int colwidth= widecharmap ? 6 : 4;
870 GET_RE_DEBUG_FLAGS_DECL;
872 PERL_ARGS_ASSERT_DUMP_TRIE;
874 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
875 (int)depth * 2 + 2,"",
876 "Match","Base","Ofs" );
878 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
879 SV ** const tmp = av_fetch( revcharmap, state, 0);
881 PerlIO_printf( Perl_debug_log, "%*s",
883 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
884 PL_colors[0], PL_colors[1],
885 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
886 PERL_PV_ESCAPE_FIRSTCHAR
891 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
892 (int)depth * 2 + 2,"");
894 for( state = 0 ; state < trie->uniquecharcount ; state++ )
895 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
896 PerlIO_printf( Perl_debug_log, "\n");
898 for( state = 1 ; state < trie->statecount ; state++ ) {
899 const U32 base = trie->states[ state ].trans.base;
901 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
903 if ( trie->states[ state ].wordnum ) {
904 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
906 PerlIO_printf( Perl_debug_log, "%6s", "" );
909 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
914 while( ( base + ofs < trie->uniquecharcount ) ||
915 ( base + ofs - trie->uniquecharcount < trie->lasttrans
916 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
919 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
921 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
922 if ( ( base + ofs >= trie->uniquecharcount ) &&
923 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
924 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
926 PerlIO_printf( Perl_debug_log, "%*"UVXf,
928 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
930 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
934 PerlIO_printf( Perl_debug_log, "]");
937 PerlIO_printf( Perl_debug_log, "\n" );
941 Dumps a fully constructed but uncompressed trie in list form.
942 List tries normally only are used for construction when the number of
943 possible chars (trie->uniquecharcount) is very high.
944 Used for debugging make_trie().
947 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
948 HV *widecharmap, AV *revcharmap, U32 next_alloc,
952 SV *sv=sv_newmortal();
953 int colwidth= widecharmap ? 6 : 4;
954 GET_RE_DEBUG_FLAGS_DECL;
956 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
958 /* print out the table precompression. */
959 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
960 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
961 "------:-----+-----------------\n" );
963 for( state=1 ; state < next_alloc ; state ++ ) {
966 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
967 (int)depth * 2 + 2,"", (UV)state );
968 if ( ! trie->states[ state ].wordnum ) {
969 PerlIO_printf( Perl_debug_log, "%5s| ","");
971 PerlIO_printf( Perl_debug_log, "W%4x| ",
972 trie->states[ state ].wordnum
975 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
976 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
978 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
980 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
981 PL_colors[0], PL_colors[1],
982 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
983 PERL_PV_ESCAPE_FIRSTCHAR
985 TRIE_LIST_ITEM(state,charid).forid,
986 (UV)TRIE_LIST_ITEM(state,charid).newstate
989 PerlIO_printf(Perl_debug_log, "\n%*s| ",
990 (int)((depth * 2) + 14), "");
993 PerlIO_printf( Perl_debug_log, "\n");
998 Dumps a fully constructed but uncompressed trie in table form.
999 This is the normal DFA style state transition table, with a few
1000 twists to facilitate compression later.
1001 Used for debugging make_trie().
1004 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1005 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1010 SV *sv=sv_newmortal();
1011 int colwidth= widecharmap ? 6 : 4;
1012 GET_RE_DEBUG_FLAGS_DECL;
1014 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1017 print out the table precompression so that we can do a visual check
1018 that they are identical.
1021 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1023 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1024 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1026 PerlIO_printf( Perl_debug_log, "%*s",
1028 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1029 PL_colors[0], PL_colors[1],
1030 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1031 PERL_PV_ESCAPE_FIRSTCHAR
1037 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1039 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1040 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1043 PerlIO_printf( Perl_debug_log, "\n" );
1045 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1047 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1048 (int)depth * 2 + 2,"",
1049 (UV)TRIE_NODENUM( state ) );
1051 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1052 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1054 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1056 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1058 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1059 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1061 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1062 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1069 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1070 startbranch: the first branch in the whole branch sequence
1071 first : start branch of sequence of branch-exact nodes.
1072 May be the same as startbranch
1073 last : Thing following the last branch.
1074 May be the same as tail.
1075 tail : item following the branch sequence
1076 count : words in the sequence
1077 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1078 depth : indent depth
1080 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1082 A trie is an N'ary tree where the branches are determined by digital
1083 decomposition of the key. IE, at the root node you look up the 1st character and
1084 follow that branch repeat until you find the end of the branches. Nodes can be
1085 marked as "accepting" meaning they represent a complete word. Eg:
1089 would convert into the following structure. Numbers represent states, letters
1090 following numbers represent valid transitions on the letter from that state, if
1091 the number is in square brackets it represents an accepting state, otherwise it
1092 will be in parenthesis.
1094 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1098 (1) +-i->(6)-+-s->[7]
1100 +-s->(3)-+-h->(4)-+-e->[5]
1102 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1104 This shows that when matching against the string 'hers' we will begin at state 1
1105 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1106 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1107 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1108 single traverse. We store a mapping from accepting to state to which word was
1109 matched, and then when we have multiple possibilities we try to complete the
1110 rest of the regex in the order in which they occured in the alternation.
1112 The only prior NFA like behaviour that would be changed by the TRIE support is
1113 the silent ignoring of duplicate alternations which are of the form:
1115 / (DUPE|DUPE) X? (?{ ... }) Y /x
1117 Thus EVAL blocks follwing a trie may be called a different number of times with
1118 and without the optimisation. With the optimisations dupes will be silently
1119 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1120 the following demonstrates:
1122 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1124 which prints out 'word' three times, but
1126 'words'=~/(word|word|word)(?{ print $1 })S/
1128 which doesnt print it out at all. This is due to other optimisations kicking in.
1130 Example of what happens on a structural level:
1132 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1134 1: CURLYM[1] {1,32767}(18)
1145 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1146 and should turn into:
1148 1: CURLYM[1] {1,32767}(18)
1150 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1158 Cases where tail != last would be like /(?foo|bar)baz/:
1168 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1169 and would end up looking like:
1172 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1179 d = uvuni_to_utf8_flags(d, uv, 0);
1181 is the recommended Unicode-aware way of saying
1186 #define TRIE_STORE_REVCHAR \
1189 SV *zlopp = newSV(2); \
1190 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1191 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1192 SvCUR_set(zlopp, kapow - flrbbbbb); \
1195 av_push(revcharmap, zlopp); \
1197 char ooooff = (char)uvc; \
1198 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1202 #define TRIE_READ_CHAR STMT_START { \
1206 if ( foldlen > 0 ) { \
1207 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1212 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1213 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1214 foldlen -= UNISKIP( uvc ); \
1215 scan = foldbuf + UNISKIP( uvc ); \
1218 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1228 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1229 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1230 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1231 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1233 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1234 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1235 TRIE_LIST_CUR( state )++; \
1238 #define TRIE_LIST_NEW(state) STMT_START { \
1239 Newxz( trie->states[ state ].trans.list, \
1240 4, reg_trie_trans_le ); \
1241 TRIE_LIST_CUR( state ) = 1; \
1242 TRIE_LIST_LEN( state ) = 4; \
1245 #define TRIE_HANDLE_WORD(state) STMT_START { \
1246 U16 dupe= trie->states[ state ].wordnum; \
1247 regnode * const noper_next = regnext( noper ); \
1249 if (trie->wordlen) \
1250 trie->wordlen[ curword ] = wordlen; \
1252 /* store the word for dumping */ \
1254 if (OP(noper) != NOTHING) \
1255 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1257 tmp = newSVpvn_utf8( "", 0, UTF ); \
1258 av_push( trie_words, tmp ); \
1263 if ( noper_next < tail ) { \
1265 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1266 trie->jump[curword] = (U16)(noper_next - convert); \
1268 jumper = noper_next; \
1270 nextbranch= regnext(cur); \
1274 /* So it's a dupe. This means we need to maintain a */\
1275 /* linked-list from the first to the next. */\
1276 /* we only allocate the nextword buffer when there */\
1277 /* a dupe, so first time we have to do the allocation */\
1278 if (!trie->nextword) \
1279 trie->nextword = (U16 *) \
1280 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1281 while ( trie->nextword[dupe] ) \
1282 dupe= trie->nextword[dupe]; \
1283 trie->nextword[dupe]= curword; \
1285 /* we haven't inserted this word yet. */ \
1286 trie->states[ state ].wordnum = curword; \
1291 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1292 ( ( base + charid >= ucharcount \
1293 && base + charid < ubound \
1294 && state == trie->trans[ base - ucharcount + charid ].check \
1295 && trie->trans[ base - ucharcount + charid ].next ) \
1296 ? trie->trans[ base - ucharcount + charid ].next \
1297 : ( state==1 ? special : 0 ) \
1301 #define MADE_JUMP_TRIE 2
1302 #define MADE_EXACT_TRIE 4
1305 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1308 /* first pass, loop through and scan words */
1309 reg_trie_data *trie;
1310 HV *widecharmap = NULL;
1311 AV *revcharmap = newAV();
1313 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1318 regnode *jumper = NULL;
1319 regnode *nextbranch = NULL;
1320 regnode *convert = NULL;
1321 /* we just use folder as a flag in utf8 */
1322 const U8 * const folder = ( flags == EXACTF
1324 : ( flags == EXACTFL
1331 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1332 AV *trie_words = NULL;
1333 /* along with revcharmap, this only used during construction but both are
1334 * useful during debugging so we store them in the struct when debugging.
1337 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1338 STRLEN trie_charcount=0;
1340 SV *re_trie_maxbuff;
1341 GET_RE_DEBUG_FLAGS_DECL;
1343 PERL_ARGS_ASSERT_MAKE_TRIE;
1345 PERL_UNUSED_ARG(depth);
1348 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1350 trie->startstate = 1;
1351 trie->wordcount = word_count;
1352 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1353 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1354 if (!(UTF && folder))
1355 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1357 trie_words = newAV();
1360 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1361 if (!SvIOK(re_trie_maxbuff)) {
1362 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1365 PerlIO_printf( Perl_debug_log,
1366 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1367 (int)depth * 2 + 2, "",
1368 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1369 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1373 /* Find the node we are going to overwrite */
1374 if ( first == startbranch && OP( last ) != BRANCH ) {
1375 /* whole branch chain */
1378 /* branch sub-chain */
1379 convert = NEXTOPER( first );
1382 /* -- First loop and Setup --
1384 We first traverse the branches and scan each word to determine if it
1385 contains widechars, and how many unique chars there are, this is
1386 important as we have to build a table with at least as many columns as we
1389 We use an array of integers to represent the character codes 0..255
1390 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1391 native representation of the character value as the key and IV's for the
1394 *TODO* If we keep track of how many times each character is used we can
1395 remap the columns so that the table compression later on is more
1396 efficient in terms of memory by ensuring most common value is in the
1397 middle and the least common are on the outside. IMO this would be better
1398 than a most to least common mapping as theres a decent chance the most
1399 common letter will share a node with the least common, meaning the node
1400 will not be compressable. With a middle is most common approach the worst
1401 case is when we have the least common nodes twice.
1405 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1406 regnode * const noper = NEXTOPER( cur );
1407 const U8 *uc = (U8*)STRING( noper );
1408 const U8 * const e = uc + STR_LEN( noper );
1410 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1411 const U8 *scan = (U8*)NULL;
1412 U32 wordlen = 0; /* required init */
1414 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1416 if (OP(noper) == NOTHING) {
1420 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1421 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1422 regardless of encoding */
1424 for ( ; uc < e ; uc += len ) {
1425 TRIE_CHARCOUNT(trie)++;
1429 if ( !trie->charmap[ uvc ] ) {
1430 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1432 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1436 /* store the codepoint in the bitmap, and if its ascii
1437 also store its folded equivelent. */
1438 TRIE_BITMAP_SET(trie,uvc);
1440 /* store the folded codepoint */
1441 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1444 /* store first byte of utf8 representation of
1445 codepoints in the 127 < uvc < 256 range */
1446 if (127 < uvc && uvc < 192) {
1447 TRIE_BITMAP_SET(trie,194);
1448 } else if (191 < uvc ) {
1449 TRIE_BITMAP_SET(trie,195);
1450 /* && uvc < 256 -- we know uvc is < 256 already */
1453 set_bit = 0; /* We've done our bit :-) */
1458 widecharmap = newHV();
1460 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1463 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1465 if ( !SvTRUE( *svpp ) ) {
1466 sv_setiv( *svpp, ++trie->uniquecharcount );
1471 if( cur == first ) {
1474 } else if (chars < trie->minlen) {
1476 } else if (chars > trie->maxlen) {
1480 } /* end first pass */
1481 DEBUG_TRIE_COMPILE_r(
1482 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1483 (int)depth * 2 + 2,"",
1484 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1485 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1486 (int)trie->minlen, (int)trie->maxlen )
1488 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1491 We now know what we are dealing with in terms of unique chars and
1492 string sizes so we can calculate how much memory a naive
1493 representation using a flat table will take. If it's over a reasonable
1494 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1495 conservative but potentially much slower representation using an array
1498 At the end we convert both representations into the same compressed
1499 form that will be used in regexec.c for matching with. The latter
1500 is a form that cannot be used to construct with but has memory
1501 properties similar to the list form and access properties similar
1502 to the table form making it both suitable for fast searches and
1503 small enough that its feasable to store for the duration of a program.
1505 See the comment in the code where the compressed table is produced
1506 inplace from the flat tabe representation for an explanation of how
1507 the compression works.
1512 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1514 Second Pass -- Array Of Lists Representation
1516 Each state will be represented by a list of charid:state records
1517 (reg_trie_trans_le) the first such element holds the CUR and LEN
1518 points of the allocated array. (See defines above).
1520 We build the initial structure using the lists, and then convert
1521 it into the compressed table form which allows faster lookups
1522 (but cant be modified once converted).
1525 STRLEN transcount = 1;
1527 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1528 "%*sCompiling trie using list compiler\n",
1529 (int)depth * 2 + 2, ""));
1531 trie->states = (reg_trie_state *)
1532 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1533 sizeof(reg_trie_state) );
1537 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1539 regnode * const noper = NEXTOPER( cur );
1540 U8 *uc = (U8*)STRING( noper );
1541 const U8 * const e = uc + STR_LEN( noper );
1542 U32 state = 1; /* required init */
1543 U16 charid = 0; /* sanity init */
1544 U8 *scan = (U8*)NULL; /* sanity init */
1545 STRLEN foldlen = 0; /* required init */
1546 U32 wordlen = 0; /* required init */
1547 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1549 if (OP(noper) != NOTHING) {
1550 for ( ; uc < e ; uc += len ) {
1555 charid = trie->charmap[ uvc ];
1557 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1561 charid=(U16)SvIV( *svpp );
1564 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1571 if ( !trie->states[ state ].trans.list ) {
1572 TRIE_LIST_NEW( state );
1574 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1575 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1576 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1581 newstate = next_alloc++;
1582 TRIE_LIST_PUSH( state, charid, newstate );
1587 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1591 TRIE_HANDLE_WORD(state);
1593 } /* end second pass */
1595 /* next alloc is the NEXT state to be allocated */
1596 trie->statecount = next_alloc;
1597 trie->states = (reg_trie_state *)
1598 PerlMemShared_realloc( trie->states,
1600 * sizeof(reg_trie_state) );
1602 /* and now dump it out before we compress it */
1603 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1604 revcharmap, next_alloc,
1608 trie->trans = (reg_trie_trans *)
1609 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1616 for( state=1 ; state < next_alloc ; state ++ ) {
1620 DEBUG_TRIE_COMPILE_MORE_r(
1621 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1625 if (trie->states[state].trans.list) {
1626 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1630 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1631 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1632 if ( forid < minid ) {
1634 } else if ( forid > maxid ) {
1638 if ( transcount < tp + maxid - minid + 1) {
1640 trie->trans = (reg_trie_trans *)
1641 PerlMemShared_realloc( trie->trans,
1643 * sizeof(reg_trie_trans) );
1644 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1646 base = trie->uniquecharcount + tp - minid;
1647 if ( maxid == minid ) {
1649 for ( ; zp < tp ; zp++ ) {
1650 if ( ! trie->trans[ zp ].next ) {
1651 base = trie->uniquecharcount + zp - minid;
1652 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1653 trie->trans[ zp ].check = state;
1659 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1660 trie->trans[ tp ].check = state;
1665 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1666 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1667 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1668 trie->trans[ tid ].check = state;
1670 tp += ( maxid - minid + 1 );
1672 Safefree(trie->states[ state ].trans.list);
1675 DEBUG_TRIE_COMPILE_MORE_r(
1676 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1679 trie->states[ state ].trans.base=base;
1681 trie->lasttrans = tp + 1;
1685 Second Pass -- Flat Table Representation.
1687 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1688 We know that we will need Charcount+1 trans at most to store the data
1689 (one row per char at worst case) So we preallocate both structures
1690 assuming worst case.
1692 We then construct the trie using only the .next slots of the entry
1695 We use the .check field of the first entry of the node temporarily to
1696 make compression both faster and easier by keeping track of how many non
1697 zero fields are in the node.
1699 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1702 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1703 number representing the first entry of the node, and state as a
1704 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1705 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1706 are 2 entrys per node. eg:
1714 The table is internally in the right hand, idx form. However as we also
1715 have to deal with the states array which is indexed by nodenum we have to
1716 use TRIE_NODENUM() to convert.
1719 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1720 "%*sCompiling trie using table compiler\n",
1721 (int)depth * 2 + 2, ""));
1723 trie->trans = (reg_trie_trans *)
1724 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1725 * trie->uniquecharcount + 1,
1726 sizeof(reg_trie_trans) );
1727 trie->states = (reg_trie_state *)
1728 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1729 sizeof(reg_trie_state) );
1730 next_alloc = trie->uniquecharcount + 1;
1733 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1735 regnode * const noper = NEXTOPER( cur );
1736 const U8 *uc = (U8*)STRING( noper );
1737 const U8 * const e = uc + STR_LEN( noper );
1739 U32 state = 1; /* required init */
1741 U16 charid = 0; /* sanity init */
1742 U32 accept_state = 0; /* sanity init */
1743 U8 *scan = (U8*)NULL; /* sanity init */
1745 STRLEN foldlen = 0; /* required init */
1746 U32 wordlen = 0; /* required init */
1747 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1749 if ( OP(noper) != NOTHING ) {
1750 for ( ; uc < e ; uc += len ) {
1755 charid = trie->charmap[ uvc ];
1757 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1758 charid = svpp ? (U16)SvIV(*svpp) : 0;
1762 if ( !trie->trans[ state + charid ].next ) {
1763 trie->trans[ state + charid ].next = next_alloc;
1764 trie->trans[ state ].check++;
1765 next_alloc += trie->uniquecharcount;
1767 state = trie->trans[ state + charid ].next;
1769 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1771 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1774 accept_state = TRIE_NODENUM( state );
1775 TRIE_HANDLE_WORD(accept_state);
1777 } /* end second pass */
1779 /* and now dump it out before we compress it */
1780 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1782 next_alloc, depth+1));
1786 * Inplace compress the table.*
1788 For sparse data sets the table constructed by the trie algorithm will
1789 be mostly 0/FAIL transitions or to put it another way mostly empty.
1790 (Note that leaf nodes will not contain any transitions.)
1792 This algorithm compresses the tables by eliminating most such
1793 transitions, at the cost of a modest bit of extra work during lookup:
1795 - Each states[] entry contains a .base field which indicates the
1796 index in the state[] array wheres its transition data is stored.
1798 - If .base is 0 there are no valid transitions from that node.
1800 - If .base is nonzero then charid is added to it to find an entry in
1803 -If trans[states[state].base+charid].check!=state then the
1804 transition is taken to be a 0/Fail transition. Thus if there are fail
1805 transitions at the front of the node then the .base offset will point
1806 somewhere inside the previous nodes data (or maybe even into a node
1807 even earlier), but the .check field determines if the transition is
1811 The following process inplace converts the table to the compressed
1812 table: We first do not compress the root node 1,and mark its all its
1813 .check pointers as 1 and set its .base pointer as 1 as well. This
1814 allows to do a DFA construction from the compressed table later, and
1815 ensures that any .base pointers we calculate later are greater than
1818 - We set 'pos' to indicate the first entry of the second node.
1820 - We then iterate over the columns of the node, finding the first and
1821 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1822 and set the .check pointers accordingly, and advance pos
1823 appropriately and repreat for the next node. Note that when we copy
1824 the next pointers we have to convert them from the original
1825 NODEIDX form to NODENUM form as the former is not valid post
1828 - If a node has no transitions used we mark its base as 0 and do not
1829 advance the pos pointer.
1831 - If a node only has one transition we use a second pointer into the
1832 structure to fill in allocated fail transitions from other states.
1833 This pointer is independent of the main pointer and scans forward
1834 looking for null transitions that are allocated to a state. When it
1835 finds one it writes the single transition into the "hole". If the
1836 pointer doesnt find one the single transition is appended as normal.
1838 - Once compressed we can Renew/realloc the structures to release the
1841 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1842 specifically Fig 3.47 and the associated pseudocode.
1846 const U32 laststate = TRIE_NODENUM( next_alloc );
1849 trie->statecount = laststate;
1851 for ( state = 1 ; state < laststate ; state++ ) {
1853 const U32 stateidx = TRIE_NODEIDX( state );
1854 const U32 o_used = trie->trans[ stateidx ].check;
1855 U32 used = trie->trans[ stateidx ].check;
1856 trie->trans[ stateidx ].check = 0;
1858 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1859 if ( flag || trie->trans[ stateidx + charid ].next ) {
1860 if ( trie->trans[ stateidx + charid ].next ) {
1862 for ( ; zp < pos ; zp++ ) {
1863 if ( ! trie->trans[ zp ].next ) {
1867 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1868 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1869 trie->trans[ zp ].check = state;
1870 if ( ++zp > pos ) pos = zp;
1877 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1879 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1880 trie->trans[ pos ].check = state;
1885 trie->lasttrans = pos + 1;
1886 trie->states = (reg_trie_state *)
1887 PerlMemShared_realloc( trie->states, laststate
1888 * sizeof(reg_trie_state) );
1889 DEBUG_TRIE_COMPILE_MORE_r(
1890 PerlIO_printf( Perl_debug_log,
1891 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1892 (int)depth * 2 + 2,"",
1893 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1896 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1899 } /* end table compress */
1901 DEBUG_TRIE_COMPILE_MORE_r(
1902 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1903 (int)depth * 2 + 2, "",
1904 (UV)trie->statecount,
1905 (UV)trie->lasttrans)
1907 /* resize the trans array to remove unused space */
1908 trie->trans = (reg_trie_trans *)
1909 PerlMemShared_realloc( trie->trans, trie->lasttrans
1910 * sizeof(reg_trie_trans) );
1912 /* and now dump out the compressed format */
1913 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1915 { /* Modify the program and insert the new TRIE node*/
1916 U8 nodetype =(U8)(flags & 0xFF);
1920 regnode *optimize = NULL;
1921 #ifdef RE_TRACK_PATTERN_OFFSETS
1924 U32 mjd_nodelen = 0;
1925 #endif /* RE_TRACK_PATTERN_OFFSETS */
1926 #endif /* DEBUGGING */
1928 This means we convert either the first branch or the first Exact,
1929 depending on whether the thing following (in 'last') is a branch
1930 or not and whther first is the startbranch (ie is it a sub part of
1931 the alternation or is it the whole thing.)
1932 Assuming its a sub part we conver the EXACT otherwise we convert
1933 the whole branch sequence, including the first.
1935 /* Find the node we are going to overwrite */
1936 if ( first != startbranch || OP( last ) == BRANCH ) {
1937 /* branch sub-chain */
1938 NEXT_OFF( first ) = (U16)(last - first);
1939 #ifdef RE_TRACK_PATTERN_OFFSETS
1941 mjd_offset= Node_Offset((convert));
1942 mjd_nodelen= Node_Length((convert));
1945 /* whole branch chain */
1947 #ifdef RE_TRACK_PATTERN_OFFSETS
1950 const regnode *nop = NEXTOPER( convert );
1951 mjd_offset= Node_Offset((nop));
1952 mjd_nodelen= Node_Length((nop));
1956 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1957 (int)depth * 2 + 2, "",
1958 (UV)mjd_offset, (UV)mjd_nodelen)
1961 /* But first we check to see if there is a common prefix we can
1962 split out as an EXACT and put in front of the TRIE node. */
1963 trie->startstate= 1;
1964 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1966 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1970 const U32 base = trie->states[ state ].trans.base;
1972 if ( trie->states[state].wordnum )
1975 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1976 if ( ( base + ofs >= trie->uniquecharcount ) &&
1977 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1978 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1980 if ( ++count > 1 ) {
1981 SV **tmp = av_fetch( revcharmap, ofs, 0);
1982 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1983 if ( state == 1 ) break;
1985 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1987 PerlIO_printf(Perl_debug_log,
1988 "%*sNew Start State=%"UVuf" Class: [",
1989 (int)depth * 2 + 2, "",
1992 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1993 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1995 TRIE_BITMAP_SET(trie,*ch);
1997 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1999 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2003 TRIE_BITMAP_SET(trie,*ch);
2005 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2006 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2012 SV **tmp = av_fetch( revcharmap, idx, 0);
2014 char *ch = SvPV( *tmp, len );
2016 SV *sv=sv_newmortal();
2017 PerlIO_printf( Perl_debug_log,
2018 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2019 (int)depth * 2 + 2, "",
2021 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2022 PL_colors[0], PL_colors[1],
2023 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2024 PERL_PV_ESCAPE_FIRSTCHAR
2029 OP( convert ) = nodetype;
2030 str=STRING(convert);
2033 STR_LEN(convert) += len;
2039 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2045 regnode *n = convert+NODE_SZ_STR(convert);
2046 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2047 trie->startstate = state;
2048 trie->minlen -= (state - 1);
2049 trie->maxlen -= (state - 1);
2051 /* At least the UNICOS C compiler choked on this
2052 * being argument to DEBUG_r(), so let's just have
2055 #ifdef PERL_EXT_RE_BUILD
2061 regnode *fix = convert;
2062 U32 word = trie->wordcount;
2064 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2065 while( ++fix < n ) {
2066 Set_Node_Offset_Length(fix, 0, 0);
2069 SV ** const tmp = av_fetch( trie_words, word, 0 );
2071 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2072 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2074 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2082 NEXT_OFF(convert) = (U16)(tail - convert);
2083 DEBUG_r(optimize= n);
2089 if ( trie->maxlen ) {
2090 NEXT_OFF( convert ) = (U16)(tail - convert);
2091 ARG_SET( convert, data_slot );
2092 /* Store the offset to the first unabsorbed branch in
2093 jump[0], which is otherwise unused by the jump logic.
2094 We use this when dumping a trie and during optimisation. */
2096 trie->jump[0] = (U16)(nextbranch - convert);
2099 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2100 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2102 OP( convert ) = TRIEC;
2103 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2104 PerlMemShared_free(trie->bitmap);
2107 OP( convert ) = TRIE;
2109 /* store the type in the flags */
2110 convert->flags = nodetype;
2114 + regarglen[ OP( convert ) ];
2116 /* XXX We really should free up the resource in trie now,
2117 as we won't use them - (which resources?) dmq */
2119 /* needed for dumping*/
2120 DEBUG_r(if (optimize) {
2121 regnode *opt = convert;
2123 while ( ++opt < optimize) {
2124 Set_Node_Offset_Length(opt,0,0);
2127 Try to clean up some of the debris left after the
2130 while( optimize < jumper ) {
2131 mjd_nodelen += Node_Length((optimize));
2132 OP( optimize ) = OPTIMIZED;
2133 Set_Node_Offset_Length(optimize,0,0);
2136 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2138 } /* end node insert */
2139 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2141 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2142 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2144 SvREFCNT_dec(revcharmap);
2148 : trie->startstate>1
2154 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2156 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2158 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2159 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2162 We find the fail state for each state in the trie, this state is the longest proper
2163 suffix of the current states 'word' that is also a proper prefix of another word in our
2164 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2165 the DFA not to have to restart after its tried and failed a word at a given point, it
2166 simply continues as though it had been matching the other word in the first place.
2168 'abcdgu'=~/abcdefg|cdgu/
2169 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2170 fail, which would bring use to the state representing 'd' in the second word where we would
2171 try 'g' and succeed, prodceding to match 'cdgu'.
2173 /* add a fail transition */
2174 const U32 trie_offset = ARG(source);
2175 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2177 const U32 ucharcount = trie->uniquecharcount;
2178 const U32 numstates = trie->statecount;
2179 const U32 ubound = trie->lasttrans + ucharcount;
2183 U32 base = trie->states[ 1 ].trans.base;
2186 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2187 GET_RE_DEBUG_FLAGS_DECL;
2189 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2191 PERL_UNUSED_ARG(depth);
2195 ARG_SET( stclass, data_slot );
2196 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2197 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2198 aho->trie=trie_offset;
2199 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2200 Copy( trie->states, aho->states, numstates, reg_trie_state );
2201 Newxz( q, numstates, U32);
2202 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2205 /* initialize fail[0..1] to be 1 so that we always have
2206 a valid final fail state */
2207 fail[ 0 ] = fail[ 1 ] = 1;
2209 for ( charid = 0; charid < ucharcount ; charid++ ) {
2210 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2212 q[ q_write ] = newstate;
2213 /* set to point at the root */
2214 fail[ q[ q_write++ ] ]=1;
2217 while ( q_read < q_write) {
2218 const U32 cur = q[ q_read++ % numstates ];
2219 base = trie->states[ cur ].trans.base;
2221 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2222 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2224 U32 fail_state = cur;
2227 fail_state = fail[ fail_state ];
2228 fail_base = aho->states[ fail_state ].trans.base;
2229 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2231 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2232 fail[ ch_state ] = fail_state;
2233 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2235 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2237 q[ q_write++ % numstates] = ch_state;
2241 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2242 when we fail in state 1, this allows us to use the
2243 charclass scan to find a valid start char. This is based on the principle
2244 that theres a good chance the string being searched contains lots of stuff
2245 that cant be a start char.
2247 fail[ 0 ] = fail[ 1 ] = 0;
2248 DEBUG_TRIE_COMPILE_r({
2249 PerlIO_printf(Perl_debug_log,
2250 "%*sStclass Failtable (%"UVuf" states): 0",
2251 (int)(depth * 2), "", (UV)numstates
2253 for( q_read=1; q_read<numstates; q_read++ ) {
2254 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2256 PerlIO_printf(Perl_debug_log, "\n");
2259 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2264 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2265 * These need to be revisited when a newer toolchain becomes available.
2267 #if defined(__sparc64__) && defined(__GNUC__)
2268 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2269 # undef SPARC64_GCC_WORKAROUND
2270 # define SPARC64_GCC_WORKAROUND 1
2274 #define DEBUG_PEEP(str,scan,depth) \
2275 DEBUG_OPTIMISE_r({if (scan){ \
2276 SV * const mysv=sv_newmortal(); \
2277 regnode *Next = regnext(scan); \
2278 regprop(RExC_rx, mysv, scan); \
2279 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2280 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2281 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2288 #define JOIN_EXACT(scan,min,flags) \
2289 if (PL_regkind[OP(scan)] == EXACT) \
2290 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2293 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2294 /* Merge several consecutive EXACTish nodes into one. */
2295 regnode *n = regnext(scan);
2297 regnode *next = scan + NODE_SZ_STR(scan);
2301 regnode *stop = scan;
2302 GET_RE_DEBUG_FLAGS_DECL;
2304 PERL_UNUSED_ARG(depth);
2307 PERL_ARGS_ASSERT_JOIN_EXACT;
2308 #ifndef EXPERIMENTAL_INPLACESCAN
2309 PERL_UNUSED_ARG(flags);
2310 PERL_UNUSED_ARG(val);
2312 DEBUG_PEEP("join",scan,depth);
2314 /* Skip NOTHING, merge EXACT*. */
2316 ( PL_regkind[OP(n)] == NOTHING ||
2317 (stringok && (OP(n) == OP(scan))))
2319 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2321 if (OP(n) == TAIL || n > next)
2323 if (PL_regkind[OP(n)] == NOTHING) {
2324 DEBUG_PEEP("skip:",n,depth);
2325 NEXT_OFF(scan) += NEXT_OFF(n);
2326 next = n + NODE_STEP_REGNODE;
2333 else if (stringok) {
2334 const unsigned int oldl = STR_LEN(scan);
2335 regnode * const nnext = regnext(n);
2337 DEBUG_PEEP("merg",n,depth);
2340 if (oldl + STR_LEN(n) > U8_MAX)
2342 NEXT_OFF(scan) += NEXT_OFF(n);
2343 STR_LEN(scan) += STR_LEN(n);
2344 next = n + NODE_SZ_STR(n);
2345 /* Now we can overwrite *n : */
2346 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2354 #ifdef EXPERIMENTAL_INPLACESCAN
2355 if (flags && !NEXT_OFF(n)) {
2356 DEBUG_PEEP("atch", val, depth);
2357 if (reg_off_by_arg[OP(n)]) {
2358 ARG_SET(n, val - n);
2361 NEXT_OFF(n) = val - n;
2368 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2370 Two problematic code points in Unicode casefolding of EXACT nodes:
2372 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2373 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2379 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2380 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2382 This means that in case-insensitive matching (or "loose matching",
2383 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2384 length of the above casefolded versions) can match a target string
2385 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2386 This would rather mess up the minimum length computation.
2388 What we'll do is to look for the tail four bytes, and then peek
2389 at the preceding two bytes to see whether we need to decrease
2390 the minimum length by four (six minus two).
2392 Thanks to the design of UTF-8, there cannot be false matches:
2393 A sequence of valid UTF-8 bytes cannot be a subsequence of
2394 another valid sequence of UTF-8 bytes.
2397 char * const s0 = STRING(scan), *s, *t;
2398 char * const s1 = s0 + STR_LEN(scan) - 1;
2399 char * const s2 = s1 - 4;
2400 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2401 const char t0[] = "\xaf\x49\xaf\x42";
2403 const char t0[] = "\xcc\x88\xcc\x81";
2405 const char * const t1 = t0 + 3;
2408 s < s2 && (t = ninstr(s, s1, t0, t1));
2411 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2412 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2414 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2415 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2423 n = scan + NODE_SZ_STR(scan);
2425 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2432 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2436 /* REx optimizer. Converts nodes into quickier variants "in place".
2437 Finds fixed substrings. */
2439 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2440 to the position after last scanned or to NULL. */
2442 #define INIT_AND_WITHP \
2443 assert(!and_withp); \
2444 Newx(and_withp,1,struct regnode_charclass_class); \
2445 SAVEFREEPV(and_withp)
2447 /* this is a chain of data about sub patterns we are processing that
2448 need to be handled seperately/specially in study_chunk. Its so
2449 we can simulate recursion without losing state. */
2451 typedef struct scan_frame {
2452 regnode *last; /* last node to process in this frame */
2453 regnode *next; /* next node to process when last is reached */
2454 struct scan_frame *prev; /*previous frame*/
2455 I32 stop; /* what stopparen do we use */
2459 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2461 #define CASE_SYNST_FNC(nAmE) \
2463 if (flags & SCF_DO_STCLASS_AND) { \
2464 for (value = 0; value < 256; value++) \
2465 if (!is_ ## nAmE ## _cp(value)) \
2466 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2469 for (value = 0; value < 256; value++) \
2470 if (is_ ## nAmE ## _cp(value)) \
2471 ANYOF_BITMAP_SET(data->start_class, value); \
2475 if (flags & SCF_DO_STCLASS_AND) { \
2476 for (value = 0; value < 256; value++) \
2477 if (is_ ## nAmE ## _cp(value)) \
2478 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2481 for (value = 0; value < 256; value++) \
2482 if (!is_ ## nAmE ## _cp(value)) \
2483 ANYOF_BITMAP_SET(data->start_class, value); \
2490 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2491 I32 *minlenp, I32 *deltap,
2496 struct regnode_charclass_class *and_withp,
2497 U32 flags, U32 depth)
2498 /* scanp: Start here (read-write). */
2499 /* deltap: Write maxlen-minlen here. */
2500 /* last: Stop before this one. */
2501 /* data: string data about the pattern */
2502 /* stopparen: treat close N as END */
2503 /* recursed: which subroutines have we recursed into */
2504 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2507 I32 min = 0, pars = 0, code;
2508 regnode *scan = *scanp, *next;
2510 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2511 int is_inf_internal = 0; /* The studied chunk is infinite */
2512 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2513 scan_data_t data_fake;
2514 SV *re_trie_maxbuff = NULL;
2515 regnode *first_non_open = scan;
2516 I32 stopmin = I32_MAX;
2517 scan_frame *frame = NULL;
2518 GET_RE_DEBUG_FLAGS_DECL;
2520 PERL_ARGS_ASSERT_STUDY_CHUNK;
2523 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2527 while (first_non_open && OP(first_non_open) == OPEN)
2528 first_non_open=regnext(first_non_open);
2533 while ( scan && OP(scan) != END && scan < last ){
2534 /* Peephole optimizer: */
2535 DEBUG_STUDYDATA("Peep:", data,depth);
2536 DEBUG_PEEP("Peep",scan,depth);
2537 JOIN_EXACT(scan,&min,0);
2539 /* Follow the next-chain of the current node and optimize
2540 away all the NOTHINGs from it. */
2541 if (OP(scan) != CURLYX) {
2542 const int max = (reg_off_by_arg[OP(scan)]
2544 /* I32 may be smaller than U16 on CRAYs! */
2545 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2546 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2550 /* Skip NOTHING and LONGJMP. */
2551 while ((n = regnext(n))
2552 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2553 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2554 && off + noff < max)
2556 if (reg_off_by_arg[OP(scan)])
2559 NEXT_OFF(scan) = off;
2564 /* The principal pseudo-switch. Cannot be a switch, since we
2565 look into several different things. */
2566 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2567 || OP(scan) == IFTHEN) {
2568 next = regnext(scan);
2570 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2572 if (OP(next) == code || code == IFTHEN) {
2573 /* NOTE - There is similar code to this block below for handling
2574 TRIE nodes on a re-study. If you change stuff here check there
2576 I32 max1 = 0, min1 = I32_MAX, num = 0;
2577 struct regnode_charclass_class accum;
2578 regnode * const startbranch=scan;
2580 if (flags & SCF_DO_SUBSTR)
2581 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2582 if (flags & SCF_DO_STCLASS)
2583 cl_init_zero(pRExC_state, &accum);
2585 while (OP(scan) == code) {
2586 I32 deltanext, minnext, f = 0, fake;
2587 struct regnode_charclass_class this_class;
2590 data_fake.flags = 0;
2592 data_fake.whilem_c = data->whilem_c;
2593 data_fake.last_closep = data->last_closep;
2596 data_fake.last_closep = &fake;
2598 data_fake.pos_delta = delta;
2599 next = regnext(scan);
2600 scan = NEXTOPER(scan);
2602 scan = NEXTOPER(scan);
2603 if (flags & SCF_DO_STCLASS) {
2604 cl_init(pRExC_state, &this_class);
2605 data_fake.start_class = &this_class;
2606 f = SCF_DO_STCLASS_AND;
2608 if (flags & SCF_WHILEM_VISITED_POS)
2609 f |= SCF_WHILEM_VISITED_POS;
2611 /* we suppose the run is continuous, last=next...*/
2612 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2614 stopparen, recursed, NULL, f,depth+1);
2617 if (max1 < minnext + deltanext)
2618 max1 = minnext + deltanext;
2619 if (deltanext == I32_MAX)
2620 is_inf = is_inf_internal = 1;
2622 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2624 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2625 if ( stopmin > minnext)
2626 stopmin = min + min1;
2627 flags &= ~SCF_DO_SUBSTR;
2629 data->flags |= SCF_SEEN_ACCEPT;
2632 if (data_fake.flags & SF_HAS_EVAL)
2633 data->flags |= SF_HAS_EVAL;
2634 data->whilem_c = data_fake.whilem_c;
2636 if (flags & SCF_DO_STCLASS)
2637 cl_or(pRExC_state, &accum, &this_class);
2639 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2641 if (flags & SCF_DO_SUBSTR) {
2642 data->pos_min += min1;
2643 data->pos_delta += max1 - min1;
2644 if (max1 != min1 || is_inf)
2645 data->longest = &(data->longest_float);
2648 delta += max1 - min1;
2649 if (flags & SCF_DO_STCLASS_OR) {
2650 cl_or(pRExC_state, data->start_class, &accum);
2652 cl_and(data->start_class, and_withp);
2653 flags &= ~SCF_DO_STCLASS;
2656 else if (flags & SCF_DO_STCLASS_AND) {
2658 cl_and(data->start_class, &accum);
2659 flags &= ~SCF_DO_STCLASS;
2662 /* Switch to OR mode: cache the old value of
2663 * data->start_class */
2665 StructCopy(data->start_class, and_withp,
2666 struct regnode_charclass_class);
2667 flags &= ~SCF_DO_STCLASS_AND;
2668 StructCopy(&accum, data->start_class,
2669 struct regnode_charclass_class);
2670 flags |= SCF_DO_STCLASS_OR;
2671 data->start_class->flags |= ANYOF_EOS;
2675 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2678 Assuming this was/is a branch we are dealing with: 'scan' now
2679 points at the item that follows the branch sequence, whatever
2680 it is. We now start at the beginning of the sequence and look
2687 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2689 If we can find such a subseqence we need to turn the first
2690 element into a trie and then add the subsequent branch exact
2691 strings to the trie.
2695 1. patterns where the whole set of branch can be converted.
2697 2. patterns where only a subset can be converted.
2699 In case 1 we can replace the whole set with a single regop
2700 for the trie. In case 2 we need to keep the start and end
2703 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2704 becomes BRANCH TRIE; BRANCH X;
2706 There is an additional case, that being where there is a
2707 common prefix, which gets split out into an EXACT like node
2708 preceding the TRIE node.
2710 If x(1..n)==tail then we can do a simple trie, if not we make
2711 a "jump" trie, such that when we match the appropriate word
2712 we "jump" to the appopriate tail node. Essentailly we turn
2713 a nested if into a case structure of sorts.
2718 if (!re_trie_maxbuff) {
2719 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2720 if (!SvIOK(re_trie_maxbuff))
2721 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2723 if ( SvIV(re_trie_maxbuff)>=0 ) {
2725 regnode *first = (regnode *)NULL;
2726 regnode *last = (regnode *)NULL;
2727 regnode *tail = scan;
2732 SV * const mysv = sv_newmortal(); /* for dumping */
2734 /* var tail is used because there may be a TAIL
2735 regop in the way. Ie, the exacts will point to the
2736 thing following the TAIL, but the last branch will
2737 point at the TAIL. So we advance tail. If we
2738 have nested (?:) we may have to move through several
2742 while ( OP( tail ) == TAIL ) {
2743 /* this is the TAIL generated by (?:) */
2744 tail = regnext( tail );
2749 regprop(RExC_rx, mysv, tail );
2750 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2751 (int)depth * 2 + 2, "",
2752 "Looking for TRIE'able sequences. Tail node is: ",
2753 SvPV_nolen_const( mysv )
2759 step through the branches, cur represents each
2760 branch, noper is the first thing to be matched
2761 as part of that branch and noper_next is the
2762 regnext() of that node. if noper is an EXACT
2763 and noper_next is the same as scan (our current
2764 position in the regex) then the EXACT branch is
2765 a possible optimization target. Once we have
2766 two or more consequetive such branches we can
2767 create a trie of the EXACT's contents and stich
2768 it in place. If the sequence represents all of
2769 the branches we eliminate the whole thing and
2770 replace it with a single TRIE. If it is a
2771 subsequence then we need to stitch it in. This
2772 means the first branch has to remain, and needs
2773 to be repointed at the item on the branch chain
2774 following the last branch optimized. This could
2775 be either a BRANCH, in which case the
2776 subsequence is internal, or it could be the
2777 item following the branch sequence in which
2778 case the subsequence is at the end.
2782 /* dont use tail as the end marker for this traverse */
2783 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2784 regnode * const noper = NEXTOPER( cur );
2785 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2786 regnode * const noper_next = regnext( noper );
2790 regprop(RExC_rx, mysv, cur);
2791 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2792 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2794 regprop(RExC_rx, mysv, noper);
2795 PerlIO_printf( Perl_debug_log, " -> %s",
2796 SvPV_nolen_const(mysv));
2799 regprop(RExC_rx, mysv, noper_next );
2800 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2801 SvPV_nolen_const(mysv));
2803 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2804 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2806 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2807 : PL_regkind[ OP( noper ) ] == EXACT )
2808 || OP(noper) == NOTHING )
2810 && noper_next == tail
2815 if ( !first || optype == NOTHING ) {
2816 if (!first) first = cur;
2817 optype = OP( noper );
2823 Currently we assume that the trie can handle unicode and ascii
2824 matches fold cased matches. If this proves true then the following
2825 define will prevent tries in this situation.
2827 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2829 #define TRIE_TYPE_IS_SAFE 1
2830 if ( last && TRIE_TYPE_IS_SAFE ) {
2831 make_trie( pRExC_state,
2832 startbranch, first, cur, tail, count,
2835 if ( PL_regkind[ OP( noper ) ] == EXACT
2837 && noper_next == tail
2842 optype = OP( noper );
2852 regprop(RExC_rx, mysv, cur);
2853 PerlIO_printf( Perl_debug_log,
2854 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2855 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2859 if ( last && TRIE_TYPE_IS_SAFE ) {
2860 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2861 #ifdef TRIE_STUDY_OPT
2862 if ( ((made == MADE_EXACT_TRIE &&
2863 startbranch == first)
2864 || ( first_non_open == first )) &&
2866 flags |= SCF_TRIE_RESTUDY;
2867 if ( startbranch == first
2870 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2880 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2881 scan = NEXTOPER(NEXTOPER(scan));
2882 } else /* single branch is optimized. */
2883 scan = NEXTOPER(scan);
2885 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2886 scan_frame *newframe = NULL;
2891 if (OP(scan) != SUSPEND) {
2892 /* set the pointer */
2893 if (OP(scan) == GOSUB) {
2895 RExC_recurse[ARG2L(scan)] = scan;
2896 start = RExC_open_parens[paren-1];
2897 end = RExC_close_parens[paren-1];
2900 start = RExC_rxi->program + 1;
2904 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2905 SAVEFREEPV(recursed);
2907 if (!PAREN_TEST(recursed,paren+1)) {
2908 PAREN_SET(recursed,paren+1);
2909 Newx(newframe,1,scan_frame);
2911 if (flags & SCF_DO_SUBSTR) {
2912 SCAN_COMMIT(pRExC_state,data,minlenp);
2913 data->longest = &(data->longest_float);
2915 is_inf = is_inf_internal = 1;
2916 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2917 cl_anything(pRExC_state, data->start_class);
2918 flags &= ~SCF_DO_STCLASS;
2921 Newx(newframe,1,scan_frame);
2924 end = regnext(scan);
2929 SAVEFREEPV(newframe);
2930 newframe->next = regnext(scan);
2931 newframe->last = last;
2932 newframe->stop = stopparen;
2933 newframe->prev = frame;
2943 else if (OP(scan) == EXACT) {
2944 I32 l = STR_LEN(scan);
2947 const U8 * const s = (U8*)STRING(scan);
2948 l = utf8_length(s, s + l);
2949 uc = utf8_to_uvchr(s, NULL);
2951 uc = *((U8*)STRING(scan));
2954 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2955 /* The code below prefers earlier match for fixed
2956 offset, later match for variable offset. */
2957 if (data->last_end == -1) { /* Update the start info. */
2958 data->last_start_min = data->pos_min;
2959 data->last_start_max = is_inf
2960 ? I32_MAX : data->pos_min + data->pos_delta;
2962 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2964 SvUTF8_on(data->last_found);
2966 SV * const sv = data->last_found;
2967 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2968 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2969 if (mg && mg->mg_len >= 0)
2970 mg->mg_len += utf8_length((U8*)STRING(scan),
2971 (U8*)STRING(scan)+STR_LEN(scan));
2973 data->last_end = data->pos_min + l;
2974 data->pos_min += l; /* As in the first entry. */
2975 data->flags &= ~SF_BEFORE_EOL;
2977 if (flags & SCF_DO_STCLASS_AND) {
2978 /* Check whether it is compatible with what we know already! */
2982 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2983 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2984 && (!(data->start_class->flags & ANYOF_FOLD)
2985 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2988 ANYOF_CLASS_ZERO(data->start_class);
2989 ANYOF_BITMAP_ZERO(data->start_class);
2991 ANYOF_BITMAP_SET(data->start_class, uc);
2992 data->start_class->flags &= ~ANYOF_EOS;
2994 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2996 else if (flags & SCF_DO_STCLASS_OR) {
2997 /* false positive possible if the class is case-folded */
2999 ANYOF_BITMAP_SET(data->start_class, uc);
3001 data->start_class->flags |= ANYOF_UNICODE_ALL;
3002 data->start_class->flags &= ~ANYOF_EOS;
3003 cl_and(data->start_class, and_withp);
3005 flags &= ~SCF_DO_STCLASS;
3007 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3008 I32 l = STR_LEN(scan);
3009 UV uc = *((U8*)STRING(scan));
3011 /* Search for fixed substrings supports EXACT only. */
3012 if (flags & SCF_DO_SUBSTR) {
3014 SCAN_COMMIT(pRExC_state, data, minlenp);
3017 const U8 * const s = (U8 *)STRING(scan);
3018 l = utf8_length(s, s + l);
3019 uc = utf8_to_uvchr(s, NULL);
3022 if (flags & SCF_DO_SUBSTR)
3024 if (flags & SCF_DO_STCLASS_AND) {
3025 /* Check whether it is compatible with what we know already! */
3029 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3030 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3031 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3033 ANYOF_CLASS_ZERO(data->start_class);
3034 ANYOF_BITMAP_ZERO(data->start_class);
3036 ANYOF_BITMAP_SET(data->start_class, uc);
3037 data->start_class->flags &= ~ANYOF_EOS;
3038 data->start_class->flags |= ANYOF_FOLD;
3039 if (OP(scan) == EXACTFL)
3040 data->start_class->flags |= ANYOF_LOCALE;
3043 else if (flags & SCF_DO_STCLASS_OR) {
3044 if (data->start_class->flags & ANYOF_FOLD) {
3045 /* false positive possible if the class is case-folded.
3046 Assume that the locale settings are the same... */
3048 ANYOF_BITMAP_SET(data->start_class, uc);
3049 data->start_class->flags &= ~ANYOF_EOS;
3051 cl_and(data->start_class, and_withp);
3053 flags &= ~SCF_DO_STCLASS;
3055 else if (strchr((const char*)PL_varies,OP(scan))) {
3056 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3057 I32 f = flags, pos_before = 0;
3058 regnode * const oscan = scan;
3059 struct regnode_charclass_class this_class;
3060 struct regnode_charclass_class *oclass = NULL;
3061 I32 next_is_eval = 0;
3063 switch (PL_regkind[OP(scan)]) {
3064 case WHILEM: /* End of (?:...)* . */
3065 scan = NEXTOPER(scan);
3068 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3069 next = NEXTOPER(scan);
3070 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3072 maxcount = REG_INFTY;
3073 next = regnext(scan);
3074 scan = NEXTOPER(scan);
3078 if (flags & SCF_DO_SUBSTR)
3083 if (flags & SCF_DO_STCLASS) {
3085 maxcount = REG_INFTY;
3086 next = regnext(scan);
3087 scan = NEXTOPER(scan);
3090 is_inf = is_inf_internal = 1;
3091 scan = regnext(scan);
3092 if (flags & SCF_DO_SUBSTR) {
3093 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3094 data->longest = &(data->longest_float);
3096 goto optimize_curly_tail;
3098 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3099 && (scan->flags == stopparen))
3104 mincount = ARG1(scan);
3105 maxcount = ARG2(scan);
3107 next = regnext(scan);
3108 if (OP(scan) == CURLYX) {
3109 I32 lp = (data ? *(data->last_closep) : 0);
3110 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3112 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3113 next_is_eval = (OP(scan) == EVAL);
3115 if (flags & SCF_DO_SUBSTR) {
3116 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3117 pos_before = data->pos_min;
3121 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3123 data->flags |= SF_IS_INF;
3125 if (flags & SCF_DO_STCLASS) {
3126 cl_init(pRExC_state, &this_class);
3127 oclass = data->start_class;
3128 data->start_class = &this_class;
3129 f |= SCF_DO_STCLASS_AND;
3130 f &= ~SCF_DO_STCLASS_OR;
3132 /* These are the cases when once a subexpression
3133 fails at a particular position, it cannot succeed
3134 even after backtracking at the enclosing scope.
3136 XXXX what if minimal match and we are at the
3137 initial run of {n,m}? */
3138 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3139 f &= ~SCF_WHILEM_VISITED_POS;
3141 /* This will finish on WHILEM, setting scan, or on NULL: */
3142 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3143 last, data, stopparen, recursed, NULL,
3145 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3147 if (flags & SCF_DO_STCLASS)
3148 data->start_class = oclass;
3149 if (mincount == 0 || minnext == 0) {
3150 if (flags & SCF_DO_STCLASS_OR) {
3151 cl_or(pRExC_state, data->start_class, &this_class);
3153 else if (flags & SCF_DO_STCLASS_AND) {
3154 /* Switch to OR mode: cache the old value of
3155 * data->start_class */
3157 StructCopy(data->start_class, and_withp,
3158 struct regnode_charclass_class);
3159 flags &= ~SCF_DO_STCLASS_AND;
3160 StructCopy(&this_class, data->start_class,
3161 struct regnode_charclass_class);
3162 flags |= SCF_DO_STCLASS_OR;
3163 data->start_class->flags |= ANYOF_EOS;
3165 } else { /* Non-zero len */
3166 if (flags & SCF_DO_STCLASS_OR) {
3167 cl_or(pRExC_state, data->start_class, &this_class);
3168 cl_and(data->start_class, and_withp);
3170 else if (flags & SCF_DO_STCLASS_AND)
3171 cl_and(data->start_class, &this_class);
3172 flags &= ~SCF_DO_STCLASS;
3174 if (!scan) /* It was not CURLYX, but CURLY. */
3176 if ( /* ? quantifier ok, except for (?{ ... }) */
3177 (next_is_eval || !(mincount == 0 && maxcount == 1))
3178 && (minnext == 0) && (deltanext == 0)
3179 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3180 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3181 && ckWARN(WARN_REGEXP))
3184 "Quantifier unexpected on zero-length expression");
3187 min += minnext * mincount;
3188 is_inf_internal |= ((maxcount == REG_INFTY
3189 && (minnext + deltanext) > 0)
3190 || deltanext == I32_MAX);
3191 is_inf |= is_inf_internal;
3192 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3194 /* Try powerful optimization CURLYX => CURLYN. */
3195 if ( OP(oscan) == CURLYX && data
3196 && data->flags & SF_IN_PAR
3197 && !(data->flags & SF_HAS_EVAL)
3198 && !deltanext && minnext == 1 ) {
3199 /* Try to optimize to CURLYN. */
3200 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3201 regnode * const nxt1 = nxt;
3208 if (!strchr((const char*)PL_simple,OP(nxt))
3209 && !(PL_regkind[OP(nxt)] == EXACT
3210 && STR_LEN(nxt) == 1))
3216 if (OP(nxt) != CLOSE)
3218 if (RExC_open_parens) {
3219 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3220 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3222 /* Now we know that nxt2 is the only contents: */
3223 oscan->flags = (U8)ARG(nxt);
3225 OP(nxt1) = NOTHING; /* was OPEN. */
3228 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3229 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3230 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3231 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3232 OP(nxt + 1) = OPTIMIZED; /* was count. */
3233 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3238 /* Try optimization CURLYX => CURLYM. */
3239 if ( OP(oscan) == CURLYX && data
3240 && !(data->flags & SF_HAS_PAR)
3241 && !(data->flags & SF_HAS_EVAL)
3242 && !deltanext /* atom is fixed width */
3243 && minnext != 0 /* CURLYM can't handle zero width */
3245 /* XXXX How to optimize if data == 0? */
3246 /* Optimize to a simpler form. */
3247 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3251 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3252 && (OP(nxt2) != WHILEM))
3254 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3255 /* Need to optimize away parenths. */
3256 if (data->flags & SF_IN_PAR) {
3257 /* Set the parenth number. */
3258 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3260 if (OP(nxt) != CLOSE)
3261 FAIL("Panic opt close");
3262 oscan->flags = (U8)ARG(nxt);
3263 if (RExC_open_parens) {
3264 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3265 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3267 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3268 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3271 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3272 OP(nxt + 1) = OPTIMIZED; /* was count. */
3273 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3274 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3277 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3278 regnode *nnxt = regnext(nxt1);
3281 if (reg_off_by_arg[OP(nxt1)])
3282 ARG_SET(nxt1, nxt2 - nxt1);
3283 else if (nxt2 - nxt1 < U16_MAX)
3284 NEXT_OFF(nxt1) = nxt2 - nxt1;
3286 OP(nxt) = NOTHING; /* Cannot beautify */
3291 /* Optimize again: */
3292 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3293 NULL, stopparen, recursed, NULL, 0,depth+1);
3298 else if ((OP(oscan) == CURLYX)
3299 && (flags & SCF_WHILEM_VISITED_POS)
3300 /* See the comment on a similar expression above.
3301 However, this time it not a subexpression
3302 we care about, but the expression itself. */
3303 && (maxcount == REG_INFTY)
3304 && data && ++data->whilem_c < 16) {
3305 /* This stays as CURLYX, we can put the count/of pair. */
3306 /* Find WHILEM (as in regexec.c) */
3307 regnode *nxt = oscan + NEXT_OFF(oscan);
3309 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3311 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3312 | (RExC_whilem_seen << 4)); /* On WHILEM */
3314 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3316 if (flags & SCF_DO_SUBSTR) {
3317 SV *last_str = NULL;
3318 int counted = mincount != 0;
3320 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3321 #if defined(SPARC64_GCC_WORKAROUND)
3324 const char *s = NULL;
3327 if (pos_before >= data->last_start_min)
3330 b = data->last_start_min;
3333 s = SvPV_const(data->last_found, l);
3334 old = b - data->last_start_min;
3337 I32 b = pos_before >= data->last_start_min
3338 ? pos_before : data->last_start_min;
3340 const char * const s = SvPV_const(data->last_found, l);
3341 I32 old = b - data->last_start_min;
3345 old = utf8_hop((U8*)s, old) - (U8*)s;
3348 /* Get the added string: */
3349 last_str = newSVpvn_utf8(s + old, l, UTF);
3350 if (deltanext == 0 && pos_before == b) {
3351 /* What was added is a constant string */
3353 SvGROW(last_str, (mincount * l) + 1);
3354 repeatcpy(SvPVX(last_str) + l,
3355 SvPVX_const(last_str), l, mincount - 1);
3356 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3357 /* Add additional parts. */
3358 SvCUR_set(data->last_found,
3359 SvCUR(data->last_found) - l);
3360 sv_catsv(data->last_found, last_str);
3362 SV * sv = data->last_found;
3364 SvUTF8(sv) && SvMAGICAL(sv) ?
3365 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3366 if (mg && mg->mg_len >= 0)
3367 mg->mg_len += CHR_SVLEN(last_str) - l;
3369 data->last_end += l * (mincount - 1);
3372 /* start offset must point into the last copy */
3373 data->last_start_min += minnext * (mincount - 1);
3374 data->last_start_max += is_inf ? I32_MAX
3375 : (maxcount - 1) * (minnext + data->pos_delta);
3378 /* It is counted once already... */
3379 data->pos_min += minnext * (mincount - counted);
3380 data->pos_delta += - counted * deltanext +
3381 (minnext + deltanext) * maxcount - minnext * mincount;
3382 if (mincount != maxcount) {
3383 /* Cannot extend fixed substrings found inside
3385 SCAN_COMMIT(pRExC_state,data,minlenp);
3386 if (mincount && last_str) {
3387 SV * const sv = data->last_found;
3388 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3389 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3393 sv_setsv(sv, last_str);
3394 data->last_end = data->pos_min;
3395 data->last_start_min =
3396 data->pos_min - CHR_SVLEN(last_str);
3397 data->last_start_max = is_inf
3399 : data->pos_min + data->pos_delta
3400 - CHR_SVLEN(last_str);
3402 data->longest = &(data->longest_float);
3404 SvREFCNT_dec(last_str);
3406 if (data && (fl & SF_HAS_EVAL))
3407 data->flags |= SF_HAS_EVAL;
3408 optimize_curly_tail:
3409 if (OP(oscan) != CURLYX) {
3410 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3412 NEXT_OFF(oscan) += NEXT_OFF(next);
3415 default: /* REF and CLUMP only? */
3416 if (flags & SCF_DO_SUBSTR) {
3417 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3418 data->longest = &(data->longest_float);
3420 is_inf = is_inf_internal = 1;
3421 if (flags & SCF_DO_STCLASS_OR)
3422 cl_anything(pRExC_state, data->start_class);
3423 flags &= ~SCF_DO_STCLASS;
3427 else if (OP(scan) == LNBREAK) {
3428 if (flags & SCF_DO_STCLASS) {
3430 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3431 if (flags & SCF_DO_STCLASS_AND) {
3432 for (value = 0; value < 256; value++)
3433 if (!is_VERTWS_cp(value))
3434 ANYOF_BITMAP_CLEAR(data->start_class, value);
3437 for (value = 0; value < 256; value++)
3438 if (is_VERTWS_cp(value))
3439 ANYOF_BITMAP_SET(data->start_class, value);
3441 if (flags & SCF_DO_STCLASS_OR)
3442 cl_and(data->start_class, and_withp);
3443 flags &= ~SCF_DO_STCLASS;
3447 if (flags & SCF_DO_SUBSTR) {
3448 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3450 data->pos_delta += 1;
3451 data->longest = &(data->longest_float);
3455 else if (OP(scan) == FOLDCHAR) {
3456 int d = ARG(scan)==0xDF ? 1 : 2;
3457 flags &= ~SCF_DO_STCLASS;
3460 if (flags & SCF_DO_SUBSTR) {
3461 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3463 data->pos_delta += d;
3464 data->longest = &(data->longest_float);
3467 else if (strchr((const char*)PL_simple,OP(scan))) {
3470 if (flags & SCF_DO_SUBSTR) {
3471 SCAN_COMMIT(pRExC_state,data,minlenp);
3475 if (flags & SCF_DO_STCLASS) {
3476 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3478 /* Some of the logic below assumes that switching
3479 locale on will only add false positives. */
3480 switch (PL_regkind[OP(scan)]) {
3484 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3485 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3486 cl_anything(pRExC_state, data->start_class);
3489 if (OP(scan) == SANY)
3491 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3492 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3493 || (data->start_class->flags & ANYOF_CLASS));
3494 cl_anything(pRExC_state, data->start_class);
3496 if (flags & SCF_DO_STCLASS_AND || !value)
3497 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3500 if (flags & SCF_DO_STCLASS_AND)
3501 cl_and(data->start_class,
3502 (struct regnode_charclass_class*)scan);
3504 cl_or(pRExC_state, data->start_class,
3505 (struct regnode_charclass_class*)scan);
3508 if (flags & SCF_DO_STCLASS_AND) {
3509 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3510 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3511 for (value = 0; value < 256; value++)
3512 if (!isALNUM(value))
3513 ANYOF_BITMAP_CLEAR(data->start_class, value);
3517 if (data->start_class->flags & ANYOF_LOCALE)
3518 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3520 for (value = 0; value < 256; value++)
3522 ANYOF_BITMAP_SET(data->start_class, value);
3527 if (flags & SCF_DO_STCLASS_AND) {
3528 if (data->start_class->flags & ANYOF_LOCALE)
3529 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3532 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3533 data->start_class->flags |= ANYOF_LOCALE;
3537 if (flags & SCF_DO_STCLASS_AND) {
3538 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3539 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3540 for (value = 0; value < 256; value++)
3542 ANYOF_BITMAP_CLEAR(data->start_class, value);
3546 if (data->start_class->flags & ANYOF_LOCALE)
3547 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3549 for (value = 0; value < 256; value++)
3550 if (!isALNUM(value))
3551 ANYOF_BITMAP_SET(data->start_class, value);
3556 if (flags & SCF_DO_STCLASS_AND) {
3557 if (data->start_class->flags & ANYOF_LOCALE)
3558 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3561 data->start_class->flags |= ANYOF_LOCALE;
3562 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3566 if (flags & SCF_DO_STCLASS_AND) {
3567 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3568 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3569 for (value = 0; value < 256; value++)
3570 if (!isSPACE(value))
3571 ANYOF_BITMAP_CLEAR(data->start_class, value);
3575 if (data->start_class->flags & ANYOF_LOCALE)
3576 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3578 for (value = 0; value < 256; value++)
3580 ANYOF_BITMAP_SET(data->start_class, value);
3585 if (flags & SCF_DO_STCLASS_AND) {
3586 if (data->start_class->flags & ANYOF_LOCALE)
3587 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3590 data->start_class->flags |= ANYOF_LOCALE;
3591 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3595 if (flags & SCF_DO_STCLASS_AND) {
3596 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3597 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3598 for (value = 0; value < 256; value++)
3600 ANYOF_BITMAP_CLEAR(data->start_class, value);
3604 if (data->start_class->flags & ANYOF_LOCALE)
3605 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3607 for (value = 0; value < 256; value++)
3608 if (!isSPACE(value))
3609 ANYOF_BITMAP_SET(data->start_class, value);
3614 if (flags & SCF_DO_STCLASS_AND) {
3615 if (data->start_class->flags & ANYOF_LOCALE) {
3616 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3617 for (value = 0; value < 256; value++)
3618 if (!isSPACE(value))
3619 ANYOF_BITMAP_CLEAR(data->start_class, value);
3623 data->start_class->flags |= ANYOF_LOCALE;
3624 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3628 if (flags & SCF_DO_STCLASS_AND) {
3629 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3630 for (value = 0; value < 256; value++)
3631 if (!isDIGIT(value))
3632 ANYOF_BITMAP_CLEAR(data->start_class, value);
3635 if (data->start_class->flags & ANYOF_LOCALE)
3636 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3638 for (value = 0; value < 256; value++)
3640 ANYOF_BITMAP_SET(data->start_class, value);
3645 if (flags & SCF_DO_STCLASS_AND) {
3646 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3647 for (value = 0; value < 256; value++)
3649 ANYOF_BITMAP_CLEAR(data->start_class, value);
3652 if (data->start_class->flags & ANYOF_LOCALE)
3653 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3655 for (value = 0; value < 256; value++)
3656 if (!isDIGIT(value))
3657 ANYOF_BITMAP_SET(data->start_class, value);
3661 CASE_SYNST_FNC(VERTWS);
3662 CASE_SYNST_FNC(HORIZWS);
3665 if (flags & SCF_DO_STCLASS_OR)
3666 cl_and(data->start_class, and_withp);
3667 flags &= ~SCF_DO_STCLASS;
3670 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3671 data->flags |= (OP(scan) == MEOL
3675 else if ( PL_regkind[OP(scan)] == BRANCHJ
3676 /* Lookbehind, or need to calculate parens/evals/stclass: */
3677 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3678 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3679 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3680 || OP(scan) == UNLESSM )
3682 /* Negative Lookahead/lookbehind
3683 In this case we can't do fixed string optimisation.
3686 I32 deltanext, minnext, fake = 0;
3688 struct regnode_charclass_class intrnl;
3691 data_fake.flags = 0;
3693 data_fake.whilem_c = data->whilem_c;
3694 data_fake.last_closep = data->last_closep;
3697 data_fake.last_closep = &fake;
3698 data_fake.pos_delta = delta;
3699 if ( flags & SCF_DO_STCLASS && !scan->flags
3700 && OP(scan) == IFMATCH ) { /* Lookahead */
3701 cl_init(pRExC_state, &intrnl);
3702 data_fake.start_class = &intrnl;
3703 f |= SCF_DO_STCLASS_AND;
3705 if (flags & SCF_WHILEM_VISITED_POS)
3706 f |= SCF_WHILEM_VISITED_POS;
3707 next = regnext(scan);
3708 nscan = NEXTOPER(NEXTOPER(scan));
3709 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3710 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3713 FAIL("Variable length lookbehind not implemented");
3715 else if (minnext > (I32)U8_MAX) {
3716 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3718 scan->flags = (U8)minnext;
3721 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3723 if (data_fake.flags & SF_HAS_EVAL)
3724 data->flags |= SF_HAS_EVAL;
3725 data->whilem_c = data_fake.whilem_c;
3727 if (f & SCF_DO_STCLASS_AND) {
3728 if (flags & SCF_DO_STCLASS_OR) {
3729 /* OR before, AND after: ideally we would recurse with
3730 * data_fake to get the AND applied by study of the
3731 * remainder of the pattern, and then derecurse;
3732 * *** HACK *** for now just treat as "no information".
3733 * See [perl #56690].
3735 cl_init(pRExC_state, data->start_class);
3737 /* AND before and after: combine and continue */
3738 const int was = (data->start_class->flags & ANYOF_EOS);
3740 cl_and(data->start_class, &intrnl);
3742 data->start_class->flags |= ANYOF_EOS;
3746 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3748 /* Positive Lookahead/lookbehind
3749 In this case we can do fixed string optimisation,
3750 but we must be careful about it. Note in the case of
3751 lookbehind the positions will be offset by the minimum
3752 length of the pattern, something we won't know about
3753 until after the recurse.
3755 I32 deltanext, fake = 0;
3757 struct regnode_charclass_class intrnl;
3759 /* We use SAVEFREEPV so that when the full compile
3760 is finished perl will clean up the allocated
3761 minlens when its all done. This was we don't
3762 have to worry about freeing them when we know
3763 they wont be used, which would be a pain.
3766 Newx( minnextp, 1, I32 );
3767 SAVEFREEPV(minnextp);
3770 StructCopy(data, &data_fake, scan_data_t);
3771 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3774 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3775 data_fake.last_found=newSVsv(data->last_found);
3779 data_fake.last_closep = &fake;
3780 data_fake.flags = 0;
3781 data_fake.pos_delta = delta;
3783 data_fake.flags |= SF_IS_INF;
3784 if ( flags & SCF_DO_STCLASS && !scan->flags
3785 && OP(scan) == IFMATCH ) { /* Lookahead */
3786 cl_init(pRExC_state, &intrnl);
3787 data_fake.start_class = &intrnl;
3788 f |= SCF_DO_STCLASS_AND;
3790 if (flags & SCF_WHILEM_VISITED_POS)
3791 f |= SCF_WHILEM_VISITED_POS;
3792 next = regnext(scan);
3793 nscan = NEXTOPER(NEXTOPER(scan));
3795 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3796 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3799 FAIL("Variable length lookbehind not implemented");
3801 else if (*minnextp > (I32)U8_MAX) {
3802 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3804 scan->flags = (U8)*minnextp;
3809 if (f & SCF_DO_STCLASS_AND) {
3810 const int was = (data->start_class->flags & ANYOF_EOS);
3812 cl_and(data->start_class, &intrnl);
3814 data->start_class->flags |= ANYOF_EOS;
3817 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3819 if (data_fake.flags & SF_HAS_EVAL)
3820 data->flags |= SF_HAS_EVAL;
3821 data->whilem_c = data_fake.whilem_c;
3822 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3823 if (RExC_rx->minlen<*minnextp)
3824 RExC_rx->minlen=*minnextp;
3825 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3826 SvREFCNT_dec(data_fake.last_found);
3828 if ( data_fake.minlen_fixed != minlenp )
3830 data->offset_fixed= data_fake.offset_fixed;
3831 data->minlen_fixed= data_fake.minlen_fixed;
3832 data->lookbehind_fixed+= scan->flags;
3834 if ( data_fake.minlen_float != minlenp )
3836 data->minlen_float= data_fake.minlen_float;
3837 data->offset_float_min=data_fake.offset_float_min;
3838 data->offset_float_max=data_fake.offset_float_max;
3839 data->lookbehind_float+= scan->flags;
3848 else if (OP(scan) == OPEN) {
3849 if (stopparen != (I32)ARG(scan))
3852 else if (OP(scan) == CLOSE) {
3853 if (stopparen == (I32)ARG(scan)) {
3856 if ((I32)ARG(scan) == is_par) {
3857 next = regnext(scan);
3859 if ( next && (OP(next) != WHILEM) && next < last)
3860 is_par = 0; /* Disable optimization */
3863 *(data->last_closep) = ARG(scan);
3865 else if (OP(scan) == EVAL) {
3867 data->flags |= SF_HAS_EVAL;
3869 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3870 if (flags & SCF_DO_SUBSTR) {
3871 SCAN_COMMIT(pRExC_state,data,minlenp);
3872 flags &= ~SCF_DO_SUBSTR;
3874 if (data && OP(scan)==ACCEPT) {
3875 data->flags |= SCF_SEEN_ACCEPT;
3880 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3882 if (flags & SCF_DO_SUBSTR) {
3883 SCAN_COMMIT(pRExC_state,data,minlenp);
3884 data->longest = &(data->longest_float);
3886 is_inf = is_inf_internal = 1;
3887 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3888 cl_anything(pRExC_state, data->start_class);
3889 flags &= ~SCF_DO_STCLASS;
3891 else if (OP(scan) == GPOS) {
3892 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3893 !(delta || is_inf || (data && data->pos_delta)))
3895 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3896 RExC_rx->extflags |= RXf_ANCH_GPOS;
3897 if (RExC_rx->gofs < (U32)min)
3898 RExC_rx->gofs = min;
3900 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3904 #ifdef TRIE_STUDY_OPT
3905 #ifdef FULL_TRIE_STUDY
3906 else if (PL_regkind[OP(scan)] == TRIE) {
3907 /* NOTE - There is similar code to this block above for handling
3908 BRANCH nodes on the initial study. If you change stuff here
3910 regnode *trie_node= scan;
3911 regnode *tail= regnext(scan);
3912 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3913 I32 max1 = 0, min1 = I32_MAX;
3914 struct regnode_charclass_class accum;
3916 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3917 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3918 if (flags & SCF_DO_STCLASS)
3919 cl_init_zero(pRExC_state, &accum);
3925 const regnode *nextbranch= NULL;
3928 for ( word=1 ; word <= trie->wordcount ; word++)
3930 I32 deltanext=0, minnext=0, f = 0, fake;
3931 struct regnode_charclass_class this_class;
3933 data_fake.flags = 0;
3935 data_fake.whilem_c = data->whilem_c;
3936 data_fake.last_closep = data->last_closep;
3939 data_fake.last_closep = &fake;
3940 data_fake.pos_delta = delta;
3941 if (flags & SCF_DO_STCLASS) {
3942 cl_init(pRExC_state, &this_class);
3943 data_fake.start_class = &this_class;
3944 f = SCF_DO_STCLASS_AND;
3946 if (flags & SCF_WHILEM_VISITED_POS)
3947 f |= SCF_WHILEM_VISITED_POS;
3949 if (trie->jump[word]) {
3951 nextbranch = trie_node + trie->jump[0];
3952 scan= trie_node + trie->jump[word];
3953 /* We go from the jump point to the branch that follows
3954 it. Note this means we need the vestigal unused branches
3955 even though they arent otherwise used.
3957 minnext = study_chunk(pRExC_state, &scan, minlenp,
3958 &deltanext, (regnode *)nextbranch, &data_fake,
3959 stopparen, recursed, NULL, f,depth+1);
3961 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3962 nextbranch= regnext((regnode*)nextbranch);
3964 if (min1 > (I32)(minnext + trie->minlen))
3965 min1 = minnext + trie->minlen;
3966 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3967 max1 = minnext + deltanext + trie->maxlen;
3968 if (deltanext == I32_MAX)
3969 is_inf = is_inf_internal = 1;
3971 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3973 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3974 if ( stopmin > min + min1)
3975 stopmin = min + min1;
3976 flags &= ~SCF_DO_SUBSTR;
3978 data->flags |= SCF_SEEN_ACCEPT;
3981 if (data_fake.flags & SF_HAS_EVAL)
3982 data->flags |= SF_HAS_EVAL;
3983 data->whilem_c = data_fake.whilem_c;
3985 if (flags & SCF_DO_STCLASS)
3986 cl_or(pRExC_state, &accum, &this_class);
3989 if (flags & SCF_DO_SUBSTR) {
3990 data->pos_min += min1;
3991 data->pos_delta += max1 - min1;
3992 if (max1 != min1 || is_inf)
3993 data->longest = &(data->longest_float);
3996 delta += max1 - min1;
3997 if (flags & SCF_DO_STCLASS_OR) {
3998 cl_or(pRExC_state, data->start_class, &accum);
4000 cl_and(data->start_class, and_withp);
4001 flags &= ~SCF_DO_STCLASS;
4004 else if (flags & SCF_DO_STCLASS_AND) {
4006 cl_and(data->start_class, &accum);
4007 flags &= ~SCF_DO_STCLASS;
4010 /* Switch to OR mode: cache the old value of
4011 * data->start_class */
4013 StructCopy(data->start_class, and_withp,
4014 struct regnode_charclass_class);
4015 flags &= ~SCF_DO_STCLASS_AND;
4016 StructCopy(&accum, data->start_class,
4017 struct regnode_charclass_class);
4018 flags |= SCF_DO_STCLASS_OR;
4019 data->start_class->flags |= ANYOF_EOS;
4026 else if (PL_regkind[OP(scan)] == TRIE) {
4027 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4030 min += trie->minlen;
4031 delta += (trie->maxlen - trie->minlen);
4032 flags &= ~SCF_DO_STCLASS; /* xxx */
4033 if (flags & SCF_DO_SUBSTR) {
4034 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4035 data->pos_min += trie->minlen;
4036 data->pos_delta += (trie->maxlen - trie->minlen);
4037 if (trie->maxlen != trie->minlen)
4038 data->longest = &(data->longest_float);
4040 if (trie->jump) /* no more substrings -- for now /grr*/
4041 flags &= ~SCF_DO_SUBSTR;
4043 #endif /* old or new */
4044 #endif /* TRIE_STUDY_OPT */
4046 /* Else: zero-length, ignore. */
4047 scan = regnext(scan);
4052 stopparen = frame->stop;
4053 frame = frame->prev;
4054 goto fake_study_recurse;
4059 DEBUG_STUDYDATA("pre-fin:",data,depth);
4062 *deltap = is_inf_internal ? I32_MAX : delta;
4063 if (flags & SCF_DO_SUBSTR && is_inf)
4064 data->pos_delta = I32_MAX - data->pos_min;
4065 if (is_par > (I32)U8_MAX)
4067 if (is_par && pars==1 && data) {
4068 data->flags |= SF_IN_PAR;
4069 data->flags &= ~SF_HAS_PAR;
4071 else if (pars && data) {
4072 data->flags |= SF_HAS_PAR;
4073 data->flags &= ~SF_IN_PAR;
4075 if (flags & SCF_DO_STCLASS_OR)
4076 cl_and(data->start_class, and_withp);
4077 if (flags & SCF_TRIE_RESTUDY)
4078 data->flags |= SCF_TRIE_RESTUDY;
4080 DEBUG_STUDYDATA("post-fin:",data,depth);
4082 return min < stopmin ? min : stopmin;
4086 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4088 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4090 PERL_ARGS_ASSERT_ADD_DATA;
4092 Renewc(RExC_rxi->data,
4093 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4094 char, struct reg_data);
4096 Renew(RExC_rxi->data->what, count + n, U8);
4098 Newx(RExC_rxi->data->what, n, U8);
4099 RExC_rxi->data->count = count + n;
4100 Copy(s, RExC_rxi->data->what + count, n, U8);
4104 /*XXX: todo make this not included in a non debugging perl */
4105 #ifndef PERL_IN_XSUB_RE
4107 Perl_reginitcolors(pTHX)
4110 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4112 char *t = savepv(s);
4116 t = strchr(t, '\t');
4122 PL_colors[i] = t = (char *)"";
4127 PL_colors[i++] = (char *)"";
4134 #ifdef TRIE_STUDY_OPT
4135 #define CHECK_RESTUDY_GOTO \
4137 (data.flags & SCF_TRIE_RESTUDY) \
4141 #define CHECK_RESTUDY_GOTO
4145 - pregcomp - compile a regular expression into internal code
4147 * We can't allocate space until we know how big the compiled form will be,
4148 * but we can't compile it (and thus know how big it is) until we've got a
4149 * place to put the code. So we cheat: we compile it twice, once with code
4150 * generation turned off and size counting turned on, and once "for real".
4151 * This also means that we don't allocate space until we are sure that the
4152 * thing really will compile successfully, and we never have to move the
4153 * code and thus invalidate pointers into it. (Note that it has to be in
4154 * one piece because free() must be able to free it all.) [NB: not true in perl]
4156 * Beware that the optimization-preparation code in here knows about some
4157 * of the structure of the compiled regexp. [I'll say.]
4162 #ifndef PERL_IN_XSUB_RE
4163 #define RE_ENGINE_PTR &PL_core_reg_engine
4165 extern const struct regexp_engine my_reg_engine;
4166 #define RE_ENGINE_PTR &my_reg_engine
4169 #ifndef PERL_IN_XSUB_RE
4171 Perl_pregcomp(pTHX_ const SV * const pattern, const U32 flags)
4174 HV * const table = GvHV(PL_hintgv);
4176 PERL_ARGS_ASSERT_PREGCOMP;
4178 /* Dispatch a request to compile a regexp to correct
4181 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4182 GET_RE_DEBUG_FLAGS_DECL;
4183 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4184 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4186 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4189 return CALLREGCOMP_ENG(eng, pattern, flags);
4192 return Perl_re_compile(aTHX_ pattern, flags);
4197 Perl_re_compile(pTHX_ const SV * const pattern, const U32 pm_flags)
4201 register regexp_internal *ri;
4203 char* exp = SvPV((SV*)pattern, plen);
4204 char* xend = exp + plen;
4211 RExC_state_t RExC_state;
4212 RExC_state_t * const pRExC_state = &RExC_state;
4213 #ifdef TRIE_STUDY_OPT
4215 RExC_state_t copyRExC_state;
4217 GET_RE_DEBUG_FLAGS_DECL;
4219 PERL_ARGS_ASSERT_RE_COMPILE;
4221 DEBUG_r(if (!PL_colorset) reginitcolors());
4223 RExC_utf8 = RExC_orig_utf8 = pm_flags & RXf_UTF8;
4226 SV *dsv= sv_newmortal();
4227 RE_PV_QUOTED_DECL(s, RExC_utf8,
4228 dsv, exp, plen, 60);
4229 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4230 PL_colors[4],PL_colors[5],s);
4235 RExC_flags = pm_flags;
4239 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4240 RExC_seen_evals = 0;
4243 /* First pass: determine size, legality. */
4251 RExC_emit = &PL_regdummy;
4252 RExC_whilem_seen = 0;
4253 RExC_charnames = NULL;
4254 RExC_open_parens = NULL;
4255 RExC_close_parens = NULL;
4257 RExC_paren_names = NULL;
4259 RExC_paren_name_list = NULL;
4261 RExC_recurse = NULL;
4262 RExC_recurse_count = 0;
4264 #if 0 /* REGC() is (currently) a NOP at the first pass.
4265 * Clever compilers notice this and complain. --jhi */
4266 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4268 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4269 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4270 RExC_precomp = NULL;
4273 if (RExC_utf8 && !RExC_orig_utf8) {
4274 /* It's possible to write a regexp in ascii that represents Unicode
4275 codepoints outside of the byte range, such as via \x{100}. If we
4276 detect such a sequence we have to convert the entire pattern to utf8
4277 and then recompile, as our sizing calculation will have been based
4278 on 1 byte == 1 character, but we will need to use utf8 to encode
4279 at least some part of the pattern, and therefore must convert the whole
4281 XXX: somehow figure out how to make this less expensive...
4284 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4285 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4286 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4288 RExC_orig_utf8 = RExC_utf8;
4290 goto redo_first_pass;
4293 PerlIO_printf(Perl_debug_log,
4294 "Required size %"IVdf" nodes\n"
4295 "Starting second pass (creation)\n",
4298 RExC_lastparse=NULL;
4300 /* Small enough for pointer-storage convention?
4301 If extralen==0, this means that we will not need long jumps. */
4302 if (RExC_size >= 0x10000L && RExC_extralen)
4303 RExC_size += RExC_extralen;
4306 if (RExC_whilem_seen > 15)
4307 RExC_whilem_seen = 15;
4309 /* Allocate space and zero-initialize. Note, the two step process
4310 of zeroing when in debug mode, thus anything assigned has to
4311 happen after that */
4312 Newxz(r, 1, regexp);
4313 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4314 char, regexp_internal);
4315 if ( r == NULL || ri == NULL )
4316 FAIL("Regexp out of space");
4318 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4319 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4321 /* bulk initialize base fields with 0. */
4322 Zero(ri, sizeof(regexp_internal), char);
4325 /* non-zero initialization begins here */
4327 r->engine= RE_ENGINE_PTR;
4329 RX_PRELEN(r) = plen;
4330 r->extflags = pm_flags;
4332 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4333 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4334 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4335 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4336 >> RXf_PMf_STD_PMMOD_SHIFT);
4337 const char *fptr = STD_PAT_MODS; /*"msix"*/
4339 RX_WRAPLEN(r) = plen + has_minus + has_p + has_runon
4340 + (sizeof(STD_PAT_MODS) - 1)
4341 + (sizeof("(?:)") - 1);
4343 Newx(RX_WRAPPED(r), RX_WRAPLEN(r) + 1, char );
4347 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4349 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4350 char *colon = r + 1;
4353 while((ch = *fptr++)) {
4367 Copy(RExC_precomp, p, plen, char);
4377 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4379 if (RExC_seen & REG_SEEN_RECURSE) {
4380 Newxz(RExC_open_parens, RExC_npar,regnode *);
4381 SAVEFREEPV(RExC_open_parens);
4382 Newxz(RExC_close_parens,RExC_npar,regnode *);
4383 SAVEFREEPV(RExC_close_parens);
4386 /* Useful during FAIL. */
4387 #ifdef RE_TRACK_PATTERN_OFFSETS
4388 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4389 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4390 "%s %"UVuf" bytes for offset annotations.\n",
4391 ri->u.offsets ? "Got" : "Couldn't get",
4392 (UV)((2*RExC_size+1) * sizeof(U32))));
4394 SetProgLen(ri,RExC_size);
4398 /* Second pass: emit code. */
4399 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4404 RExC_emit_start = ri->program;
4405 RExC_emit = ri->program;
4406 RExC_emit_bound = ri->program + RExC_size + 1;
4408 /* Store the count of eval-groups for security checks: */
4409 RExC_rx->seen_evals = RExC_seen_evals;
4410 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4411 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4415 /* XXXX To minimize changes to RE engine we always allocate
4416 3-units-long substrs field. */
4417 Newx(r->substrs, 1, struct reg_substr_data);
4418 if (RExC_recurse_count) {
4419 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4420 SAVEFREEPV(RExC_recurse);
4424 r->minlen = minlen = sawplus = sawopen = 0;
4425 Zero(r->substrs, 1, struct reg_substr_data);
4427 #ifdef TRIE_STUDY_OPT
4429 StructCopy(&zero_scan_data, &data, scan_data_t);
4430 copyRExC_state = RExC_state;
4433 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4435 RExC_state = copyRExC_state;
4436 if (seen & REG_TOP_LEVEL_BRANCHES)
4437 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4439 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4440 if (data.last_found) {
4441 SvREFCNT_dec(data.longest_fixed);
4442 SvREFCNT_dec(data.longest_float);
4443 SvREFCNT_dec(data.last_found);
4445 StructCopy(&zero_scan_data, &data, scan_data_t);
4448 StructCopy(&zero_scan_data, &data, scan_data_t);
4451 /* Dig out information for optimizations. */
4452 r->extflags = RExC_flags; /* was pm_op */
4453 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4456 r->extflags |= RXf_UTF8; /* Unicode in it? */
4457 ri->regstclass = NULL;
4458 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4459 r->intflags |= PREGf_NAUGHTY;
4460 scan = ri->program + 1; /* First BRANCH. */
4462 /* testing for BRANCH here tells us whether there is "must appear"
4463 data in the pattern. If there is then we can use it for optimisations */
4464 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4466 STRLEN longest_float_length, longest_fixed_length;
4467 struct regnode_charclass_class ch_class; /* pointed to by data */
4469 I32 last_close = 0; /* pointed to by data */
4470 regnode *first= scan;
4471 regnode *first_next= regnext(first);
4474 * Skip introductions and multiplicators >= 1
4475 * so that we can extract the 'meat' of the pattern that must
4476 * match in the large if() sequence following.
4477 * NOTE that EXACT is NOT covered here, as it is normally
4478 * picked up by the optimiser separately.
4480 * This is unfortunate as the optimiser isnt handling lookahead
4481 * properly currently.
4484 while ((OP(first) == OPEN && (sawopen = 1)) ||
4485 /* An OR of *one* alternative - should not happen now. */
4486 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4487 /* for now we can't handle lookbehind IFMATCH*/
4488 (OP(first) == IFMATCH && !first->flags) ||
4489 (OP(first) == PLUS) ||
4490 (OP(first) == MINMOD) ||
4491 /* An {n,m} with n>0 */
4492 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4493 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4496 * the only op that could be a regnode is PLUS, all the rest
4497 * will be regnode_1 or regnode_2.
4500 if (OP(first) == PLUS)
4503 first += regarglen[OP(first)];
4505 first = NEXTOPER(first);
4506 first_next= regnext(first);
4509 /* Starting-point info. */
4511 DEBUG_PEEP("first:",first,0);
4512 /* Ignore EXACT as we deal with it later. */
4513 if (PL_regkind[OP(first)] == EXACT) {
4514 if (OP(first) == EXACT)
4515 NOOP; /* Empty, get anchored substr later. */
4516 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4517 ri->regstclass = first;
4520 else if (PL_regkind[OP(first)] == TRIE &&
4521 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4524 /* this can happen only on restudy */
4525 if ( OP(first) == TRIE ) {
4526 struct regnode_1 *trieop = (struct regnode_1 *)
4527 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4528 StructCopy(first,trieop,struct regnode_1);
4529 trie_op=(regnode *)trieop;
4531 struct regnode_charclass *trieop = (struct regnode_charclass *)
4532 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4533 StructCopy(first,trieop,struct regnode_charclass);
4534 trie_op=(regnode *)trieop;
4537 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4538 ri->regstclass = trie_op;
4541 else if (strchr((const char*)PL_simple,OP(first)))
4542 ri->regstclass = first;
4543 else if (PL_regkind[OP(first)] == BOUND ||
4544 PL_regkind[OP(first)] == NBOUND)
4545 ri->regstclass = first;
4546 else if (PL_regkind[OP(first)] == BOL) {
4547 r->extflags |= (OP(first) == MBOL
4549 : (OP(first) == SBOL
4552 first = NEXTOPER(first);
4555 else if (OP(first) == GPOS) {
4556 r->extflags |= RXf_ANCH_GPOS;
4557 first = NEXTOPER(first);
4560 else if ((!sawopen || !RExC_sawback) &&
4561 (OP(first) == STAR &&
4562 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4563 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4565 /* turn .* into ^.* with an implied $*=1 */
4567 (OP(NEXTOPER(first)) == REG_ANY)
4570 r->extflags |= type;
4571 r->intflags |= PREGf_IMPLICIT;
4572 first = NEXTOPER(first);
4575 if (sawplus && (!sawopen || !RExC_sawback)
4576 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4577 /* x+ must match at the 1st pos of run of x's */
4578 r->intflags |= PREGf_SKIP;
4580 /* Scan is after the zeroth branch, first is atomic matcher. */
4581 #ifdef TRIE_STUDY_OPT
4584 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4585 (IV)(first - scan + 1))
4589 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4590 (IV)(first - scan + 1))
4596 * If there's something expensive in the r.e., find the
4597 * longest literal string that must appear and make it the
4598 * regmust. Resolve ties in favor of later strings, since
4599 * the regstart check works with the beginning of the r.e.
4600 * and avoiding duplication strengthens checking. Not a
4601 * strong reason, but sufficient in the absence of others.
4602 * [Now we resolve ties in favor of the earlier string if
4603 * it happens that c_offset_min has been invalidated, since the
4604 * earlier string may buy us something the later one won't.]
4607 data.longest_fixed = newSVpvs("");
4608 data.longest_float = newSVpvs("");
4609 data.last_found = newSVpvs("");
4610 data.longest = &(data.longest_fixed);
4612 if (!ri->regstclass) {
4613 cl_init(pRExC_state, &ch_class);
4614 data.start_class = &ch_class;
4615 stclass_flag = SCF_DO_STCLASS_AND;
4616 } else /* XXXX Check for BOUND? */
4618 data.last_closep = &last_close;
4620 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4621 &data, -1, NULL, NULL,
4622 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4628 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4629 && data.last_start_min == 0 && data.last_end > 0
4630 && !RExC_seen_zerolen
4631 && !(RExC_seen & REG_SEEN_VERBARG)
4632 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4633 r->extflags |= RXf_CHECK_ALL;
4634 scan_commit(pRExC_state, &data,&minlen,0);
4635 SvREFCNT_dec(data.last_found);
4637 /* Note that code very similar to this but for anchored string
4638 follows immediately below, changes may need to be made to both.
4641 longest_float_length = CHR_SVLEN(data.longest_float);
4642 if (longest_float_length
4643 || (data.flags & SF_FL_BEFORE_EOL
4644 && (!(data.flags & SF_FL_BEFORE_MEOL)
4645 || (RExC_flags & RXf_PMf_MULTILINE))))
4649 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4650 && data.offset_fixed == data.offset_float_min
4651 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4652 goto remove_float; /* As in (a)+. */
4654 /* copy the information about the longest float from the reg_scan_data
4655 over to the program. */
4656 if (SvUTF8(data.longest_float)) {
4657 r->float_utf8 = data.longest_float;
4658 r->float_substr = NULL;
4660 r->float_substr = data.longest_float;
4661 r->float_utf8 = NULL;
4663 /* float_end_shift is how many chars that must be matched that
4664 follow this item. We calculate it ahead of time as once the
4665 lookbehind offset is added in we lose the ability to correctly
4667 ml = data.minlen_float ? *(data.minlen_float)
4668 : (I32)longest_float_length;
4669 r->float_end_shift = ml - data.offset_float_min
4670 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4671 + data.lookbehind_float;
4672 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4673 r->float_max_offset = data.offset_float_max;
4674 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4675 r->float_max_offset -= data.lookbehind_float;
4677 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4678 && (!(data.flags & SF_FL_BEFORE_MEOL)
4679 || (RExC_flags & RXf_PMf_MULTILINE)));
4680 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4684 r->float_substr = r->float_utf8 = NULL;
4685 SvREFCNT_dec(data.longest_float);
4686 longest_float_length = 0;
4689 /* Note that code very similar to this but for floating string
4690 is immediately above, changes may need to be made to both.
4693 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4694 if (longest_fixed_length
4695 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4696 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4697 || (RExC_flags & RXf_PMf_MULTILINE))))
4701 /* copy the information about the longest fixed
4702 from the reg_scan_data over to the program. */
4703 if (SvUTF8(data.longest_fixed)) {
4704 r->anchored_utf8 = data.longest_fixed;
4705 r->anchored_substr = NULL;
4707 r->anchored_substr = data.longest_fixed;
4708 r->anchored_utf8 = NULL;
4710 /* fixed_end_shift is how many chars that must be matched that
4711 follow this item. We calculate it ahead of time as once the
4712 lookbehind offset is added in we lose the ability to correctly
4714 ml = data.minlen_fixed ? *(data.minlen_fixed)
4715 : (I32)longest_fixed_length;
4716 r->anchored_end_shift = ml - data.offset_fixed
4717 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4718 + data.lookbehind_fixed;
4719 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4721 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4722 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4723 || (RExC_flags & RXf_PMf_MULTILINE)));
4724 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4727 r->anchored_substr = r->anchored_utf8 = NULL;
4728 SvREFCNT_dec(data.longest_fixed);
4729 longest_fixed_length = 0;
4732 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4733 ri->regstclass = NULL;
4734 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4736 && !(data.start_class->flags & ANYOF_EOS)
4737 && !cl_is_anything(data.start_class))
4739 const U32 n = add_data(pRExC_state, 1, "f");
4741 Newx(RExC_rxi->data->data[n], 1,
4742 struct regnode_charclass_class);
4743 StructCopy(data.start_class,
4744 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4745 struct regnode_charclass_class);
4746 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4747 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4748 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4749 regprop(r, sv, (regnode*)data.start_class);
4750 PerlIO_printf(Perl_debug_log,
4751 "synthetic stclass \"%s\".\n",
4752 SvPVX_const(sv));});
4755 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4756 if (longest_fixed_length > longest_float_length) {
4757 r->check_end_shift = r->anchored_end_shift;
4758 r->check_substr = r->anchored_substr;
4759 r->check_utf8 = r->anchored_utf8;
4760 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4761 if (r->extflags & RXf_ANCH_SINGLE)
4762 r->extflags |= RXf_NOSCAN;
4765 r->check_end_shift = r->float_end_shift;
4766 r->check_substr = r->float_substr;
4767 r->check_utf8 = r->float_utf8;
4768 r->check_offset_min = r->float_min_offset;
4769 r->check_offset_max = r->float_max_offset;
4771 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4772 This should be changed ASAP! */
4773 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4774 r->extflags |= RXf_USE_INTUIT;
4775 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4776 r->extflags |= RXf_INTUIT_TAIL;
4778 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4779 if ( (STRLEN)minlen < longest_float_length )
4780 minlen= longest_float_length;
4781 if ( (STRLEN)minlen < longest_fixed_length )
4782 minlen= longest_fixed_length;
4786 /* Several toplevels. Best we can is to set minlen. */
4788 struct regnode_charclass_class ch_class;
4791 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4793 scan = ri->program + 1;
4794 cl_init(pRExC_state, &ch_class);
4795 data.start_class = &ch_class;
4796 data.last_closep = &last_close;
4799 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4800 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4804 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4805 = r->float_substr = r->float_utf8 = NULL;
4806 if (!(data.start_class->flags & ANYOF_EOS)
4807 && !cl_is_anything(data.start_class))
4809 const U32 n = add_data(pRExC_state, 1, "f");
4811 Newx(RExC_rxi->data->data[n], 1,
4812 struct regnode_charclass_class);
4813 StructCopy(data.start_class,
4814 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4815 struct regnode_charclass_class);
4816 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4817 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4818 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4819 regprop(r, sv, (regnode*)data.start_class);
4820 PerlIO_printf(Perl_debug_log,
4821 "synthetic stclass \"%s\".\n",
4822 SvPVX_const(sv));});
4826 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4827 the "real" pattern. */
4829 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4830 (IV)minlen, (IV)r->minlen);
4832 r->minlenret = minlen;
4833 if (r->minlen < minlen)
4836 if (RExC_seen & REG_SEEN_GPOS)
4837 r->extflags |= RXf_GPOS_SEEN;
4838 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4839 r->extflags |= RXf_LOOKBEHIND_SEEN;
4840 if (RExC_seen & REG_SEEN_EVAL)
4841 r->extflags |= RXf_EVAL_SEEN;
4842 if (RExC_seen & REG_SEEN_CANY)
4843 r->extflags |= RXf_CANY_SEEN;
4844 if (RExC_seen & REG_SEEN_VERBARG)
4845 r->intflags |= PREGf_VERBARG_SEEN;
4846 if (RExC_seen & REG_SEEN_CUTGROUP)
4847 r->intflags |= PREGf_CUTGROUP_SEEN;
4848 if (RExC_paren_names)
4849 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
4851 RXp_PAREN_NAMES(r) = NULL;
4853 #ifdef STUPID_PATTERN_CHECKS
4854 if (RX_PRELEN(r) == 0)
4855 r->extflags |= RXf_NULL;
4856 if (r->extflags & RXf_SPLIT && RX_PRELEN(r) == 1 && RX_PRECOMP(r)[0] == ' ')
4857 /* XXX: this should happen BEFORE we compile */
4858 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4859 else if (RX_PRELEN(r) == 3 && memEQ("\\s+", RX_PRECOMP(r), 3))
4860 r->extflags |= RXf_WHITE;
4861 else if (RX_PRELEN(r) == 1 && RXp_PRECOMP(r)[0] == '^')
4862 r->extflags |= RXf_START_ONLY;
4864 if (r->extflags & RXf_SPLIT && RX_PRELEN(r) == 1 && RX_PRECOMP(r)[0] == ' ')
4865 /* XXX: this should happen BEFORE we compile */
4866 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4868 regnode *first = ri->program + 1;
4870 U8 nop = OP(NEXTOPER(first));
4872 if (PL_regkind[fop] == NOTHING && nop == END)
4873 r->extflags |= RXf_NULL;
4874 else if (PL_regkind[fop] == BOL && nop == END)
4875 r->extflags |= RXf_START_ONLY;
4876 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4877 r->extflags |= RXf_WHITE;
4881 if (RExC_paren_names) {
4882 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4883 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4886 ri->name_list_idx = 0;
4888 if (RExC_recurse_count) {
4889 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4890 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4891 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4894 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4895 /* assume we don't need to swap parens around before we match */
4898 PerlIO_printf(Perl_debug_log,"Final program:\n");
4901 #ifdef RE_TRACK_PATTERN_OFFSETS
4902 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4903 const U32 len = ri->u.offsets[0];
4905 GET_RE_DEBUG_FLAGS_DECL;
4906 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4907 for (i = 1; i <= len; i++) {
4908 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4909 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4910 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4912 PerlIO_printf(Perl_debug_log, "\n");
4918 #undef RE_ENGINE_PTR
4922 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4925 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
4927 PERL_UNUSED_ARG(value);
4929 if (flags & RXapif_FETCH) {
4930 return reg_named_buff_fetch(rx, key, flags);
4931 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
4932 Perl_croak(aTHX_ "%s", PL_no_modify);
4934 } else if (flags & RXapif_EXISTS) {
4935 return reg_named_buff_exists(rx, key, flags)
4938 } else if (flags & RXapif_REGNAMES) {
4939 return reg_named_buff_all(rx, flags);
4940 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
4941 return reg_named_buff_scalar(rx, flags);
4943 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
4949 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
4952 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
4953 PERL_UNUSED_ARG(lastkey);
4955 if (flags & RXapif_FIRSTKEY)
4956 return reg_named_buff_firstkey(rx, flags);
4957 else if (flags & RXapif_NEXTKEY)
4958 return reg_named_buff_nextkey(rx, flags);
4960 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
4966 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const rx, SV * const namesv, const U32 flags)
4968 AV *retarray = NULL;
4971 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
4973 if (flags & RXapif_ALL)
4976 if (rx && RXp_PAREN_NAMES(rx)) {
4977 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
4980 SV* sv_dat=HeVAL(he_str);
4981 I32 *nums=(I32*)SvPVX(sv_dat);
4982 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4983 if ((I32)(rx->nparens) >= nums[i]
4984 && rx->offs[nums[i]].start != -1
4985 && rx->offs[nums[i]].end != -1)
4988 CALLREG_NUMBUF_FETCH(rx,nums[i],ret);
4992 ret = newSVsv(&PL_sv_undef);
4995 av_push(retarray, ret);
4998 return newRV_noinc(MUTABLE_SV(retarray));
5005 Perl_reg_named_buff_exists(pTHX_ REGEXP * const rx, SV * const key,
5009 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5011 if (rx && RXp_PAREN_NAMES(rx)) {
5012 if (flags & RXapif_ALL) {
5013 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5015 SV *sv = CALLREG_NAMED_BUFF_FETCH(rx, key, flags);
5029 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const rx, const U32 flags)
5032 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5034 if ( rx && RXp_PAREN_NAMES(rx) ) {
5035 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5037 return CALLREG_NAMED_BUFF_NEXTKEY(rx, NULL, flags & ~RXapif_FIRSTKEY);
5044 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const rx, const U32 flags)
5046 GET_RE_DEBUG_FLAGS_DECL;
5048 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5050 if (rx && RXp_PAREN_NAMES(rx)) {
5051 HV *hv = RXp_PAREN_NAMES(rx);
5053 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5056 SV* sv_dat = HeVAL(temphe);
5057 I32 *nums = (I32*)SvPVX(sv_dat);
5058 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5059 if ((I32)(rx->lastparen) >= nums[i] &&
5060 rx->offs[nums[i]].start != -1 &&
5061 rx->offs[nums[i]].end != -1)
5067 if (parno || flags & RXapif_ALL) {
5068 return newSVhek(HeKEY_hek(temphe));
5076 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const rx, const U32 flags)
5082 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5084 if (rx && RXp_PAREN_NAMES(rx)) {
5085 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5086 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5087 } else if (flags & RXapif_ONE) {
5088 ret = CALLREG_NAMED_BUFF_ALL(rx, (flags | RXapif_REGNAMES));
5089 av = MUTABLE_AV(SvRV(ret));
5090 length = av_len(av);
5092 return newSViv(length + 1);
5094 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5098 return &PL_sv_undef;
5102 Perl_reg_named_buff_all(pTHX_ REGEXP * const rx, const U32 flags)
5106 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5108 if (rx && RXp_PAREN_NAMES(rx)) {
5109 HV *hv= RXp_PAREN_NAMES(rx);
5111 (void)hv_iterinit(hv);
5112 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5115 SV* sv_dat = HeVAL(temphe);
5116 I32 *nums = (I32*)SvPVX(sv_dat);
5117 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5118 if ((I32)(rx->lastparen) >= nums[i] &&
5119 rx->offs[nums[i]].start != -1 &&
5120 rx->offs[nums[i]].end != -1)
5126 if (parno || flags & RXapif_ALL) {
5127 av_push(av, newSVhek(HeKEY_hek(temphe)));
5132 return newRV_noinc(MUTABLE_SV(av));
5136 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const rx, const I32 paren, SV * const sv)
5142 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5145 sv_setsv(sv,&PL_sv_undef);
5149 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5151 i = rx->offs[0].start;
5155 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5157 s = rx->subbeg + rx->offs[0].end;
5158 i = rx->sublen - rx->offs[0].end;
5161 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5162 (s1 = rx->offs[paren].start) != -1 &&
5163 (t1 = rx->offs[paren].end) != -1)
5167 s = rx->subbeg + s1;
5169 sv_setsv(sv,&PL_sv_undef);
5172 assert(rx->sublen >= (s - rx->subbeg) + i );
5174 const int oldtainted = PL_tainted;
5176 sv_setpvn(sv, s, i);
5177 PL_tainted = oldtainted;
5178 if ( (rx->extflags & RXf_CANY_SEEN)
5179 ? (RXp_MATCH_UTF8(rx)
5180 && (!i || is_utf8_string((U8*)s, i)))
5181 : (RXp_MATCH_UTF8(rx)) )
5188 if (RXp_MATCH_TAINTED(rx)) {
5189 if (SvTYPE(sv) >= SVt_PVMG) {
5190 MAGIC* const mg = SvMAGIC(sv);
5193 SvMAGIC_set(sv, mg->mg_moremagic);
5195 if ((mgt = SvMAGIC(sv))) {
5196 mg->mg_moremagic = mgt;
5197 SvMAGIC_set(sv, mg);
5207 sv_setsv(sv,&PL_sv_undef);
5213 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5214 SV const * const value)
5216 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5218 PERL_UNUSED_ARG(rx);
5219 PERL_UNUSED_ARG(paren);
5220 PERL_UNUSED_ARG(value);
5223 Perl_croak(aTHX_ "%s", PL_no_modify);
5227 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const rx, const SV * const sv,
5233 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5235 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5237 /* $` / ${^PREMATCH} */
5238 case RX_BUFF_IDX_PREMATCH:
5239 if (rx->offs[0].start != -1) {
5240 i = rx->offs[0].start;
5248 /* $' / ${^POSTMATCH} */
5249 case RX_BUFF_IDX_POSTMATCH:
5250 if (rx->offs[0].end != -1) {
5251 i = rx->sublen - rx->offs[0].end;
5253 s1 = rx->offs[0].end;
5259 /* $& / ${^MATCH}, $1, $2, ... */
5261 if (paren <= (I32)rx->nparens &&
5262 (s1 = rx->offs[paren].start) != -1 &&
5263 (t1 = rx->offs[paren].end) != -1)
5268 if (ckWARN(WARN_UNINITIALIZED))
5269 report_uninit((SV *)sv);
5274 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5275 const char * const s = rx->subbeg + s1;
5280 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5287 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5289 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5290 PERL_UNUSED_ARG(rx);
5291 return newSVpvs("Regexp");
5294 /* Scans the name of a named buffer from the pattern.
5295 * If flags is REG_RSN_RETURN_NULL returns null.
5296 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5297 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5298 * to the parsed name as looked up in the RExC_paren_names hash.
5299 * If there is an error throws a vFAIL().. type exception.
5302 #define REG_RSN_RETURN_NULL 0
5303 #define REG_RSN_RETURN_NAME 1
5304 #define REG_RSN_RETURN_DATA 2
5307 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5309 char *name_start = RExC_parse;
5311 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5313 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5314 /* skip IDFIRST by using do...while */
5317 RExC_parse += UTF8SKIP(RExC_parse);
5318 } while (isALNUM_utf8((U8*)RExC_parse));
5322 } while (isALNUM(*RExC_parse));
5327 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5328 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5329 if ( flags == REG_RSN_RETURN_NAME)
5331 else if (flags==REG_RSN_RETURN_DATA) {
5334 if ( ! sv_name ) /* should not happen*/
5335 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5336 if (RExC_paren_names)
5337 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5339 sv_dat = HeVAL(he_str);
5341 vFAIL("Reference to nonexistent named group");
5345 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5352 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5353 int rem=(int)(RExC_end - RExC_parse); \
5362 if (RExC_lastparse!=RExC_parse) \
5363 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5366 iscut ? "..." : "<" \
5369 PerlIO_printf(Perl_debug_log,"%16s",""); \
5372 num = RExC_size + 1; \
5374 num=REG_NODE_NUM(RExC_emit); \
5375 if (RExC_lastnum!=num) \
5376 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5378 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5379 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5380 (int)((depth*2)), "", \
5384 RExC_lastparse=RExC_parse; \
5389 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5390 DEBUG_PARSE_MSG((funcname)); \
5391 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5393 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5394 DEBUG_PARSE_MSG((funcname)); \
5395 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5398 - reg - regular expression, i.e. main body or parenthesized thing
5400 * Caller must absorb opening parenthesis.
5402 * Combining parenthesis handling with the base level of regular expression
5403 * is a trifle forced, but the need to tie the tails of the branches to what
5404 * follows makes it hard to avoid.
5406 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5408 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5410 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5414 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5415 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5418 register regnode *ret; /* Will be the head of the group. */
5419 register regnode *br;
5420 register regnode *lastbr;
5421 register regnode *ender = NULL;
5422 register I32 parno = 0;
5424 U32 oregflags = RExC_flags;
5425 bool have_branch = 0;
5427 I32 freeze_paren = 0;
5428 I32 after_freeze = 0;
5430 /* for (?g), (?gc), and (?o) warnings; warning
5431 about (?c) will warn about (?g) -- japhy */
5433 #define WASTED_O 0x01
5434 #define WASTED_G 0x02
5435 #define WASTED_C 0x04
5436 #define WASTED_GC (0x02|0x04)
5437 I32 wastedflags = 0x00;
5439 char * parse_start = RExC_parse; /* MJD */
5440 char * const oregcomp_parse = RExC_parse;
5442 GET_RE_DEBUG_FLAGS_DECL;
5444 PERL_ARGS_ASSERT_REG;
5445 DEBUG_PARSE("reg ");
5447 *flagp = 0; /* Tentatively. */
5450 /* Make an OPEN node, if parenthesized. */
5452 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5453 char *start_verb = RExC_parse;
5454 STRLEN verb_len = 0;
5455 char *start_arg = NULL;
5456 unsigned char op = 0;
5458 int internal_argval = 0; /* internal_argval is only useful if !argok */
5459 while ( *RExC_parse && *RExC_parse != ')' ) {
5460 if ( *RExC_parse == ':' ) {
5461 start_arg = RExC_parse + 1;
5467 verb_len = RExC_parse - start_verb;
5470 while ( *RExC_parse && *RExC_parse != ')' )
5472 if ( *RExC_parse != ')' )
5473 vFAIL("Unterminated verb pattern argument");
5474 if ( RExC_parse == start_arg )
5477 if ( *RExC_parse != ')' )
5478 vFAIL("Unterminated verb pattern");
5481 switch ( *start_verb ) {
5482 case 'A': /* (*ACCEPT) */
5483 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5485 internal_argval = RExC_nestroot;
5488 case 'C': /* (*COMMIT) */
5489 if ( memEQs(start_verb,verb_len,"COMMIT") )
5492 case 'F': /* (*FAIL) */
5493 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5498 case ':': /* (*:NAME) */
5499 case 'M': /* (*MARK:NAME) */
5500 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5505 case 'P': /* (*PRUNE) */
5506 if ( memEQs(start_verb,verb_len,"PRUNE") )
5509 case 'S': /* (*SKIP) */
5510 if ( memEQs(start_verb,verb_len,"SKIP") )
5513 case 'T': /* (*THEN) */
5514 /* [19:06] <TimToady> :: is then */
5515 if ( memEQs(start_verb,verb_len,"THEN") ) {
5517 RExC_seen |= REG_SEEN_CUTGROUP;
5523 vFAIL3("Unknown verb pattern '%.*s'",
5524 verb_len, start_verb);
5527 if ( start_arg && internal_argval ) {
5528 vFAIL3("Verb pattern '%.*s' may not have an argument",
5529 verb_len, start_verb);
5530 } else if ( argok < 0 && !start_arg ) {
5531 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5532 verb_len, start_verb);
5534 ret = reganode(pRExC_state, op, internal_argval);
5535 if ( ! internal_argval && ! SIZE_ONLY ) {
5537 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5538 ARG(ret) = add_data( pRExC_state, 1, "S" );
5539 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5546 if (!internal_argval)
5547 RExC_seen |= REG_SEEN_VERBARG;
5548 } else if ( start_arg ) {
5549 vFAIL3("Verb pattern '%.*s' may not have an argument",
5550 verb_len, start_verb);
5552 ret = reg_node(pRExC_state, op);
5554 nextchar(pRExC_state);
5557 if (*RExC_parse == '?') { /* (?...) */
5558 bool is_logical = 0;
5559 const char * const seqstart = RExC_parse;
5562 paren = *RExC_parse++;
5563 ret = NULL; /* For look-ahead/behind. */
5566 case 'P': /* (?P...) variants for those used to PCRE/Python */
5567 paren = *RExC_parse++;
5568 if ( paren == '<') /* (?P<...>) named capture */
5570 else if (paren == '>') { /* (?P>name) named recursion */
5571 goto named_recursion;
5573 else if (paren == '=') { /* (?P=...) named backref */
5574 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5575 you change this make sure you change that */
5576 char* name_start = RExC_parse;
5578 SV *sv_dat = reg_scan_name(pRExC_state,
5579 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5580 if (RExC_parse == name_start || *RExC_parse != ')')
5581 vFAIL2("Sequence %.3s... not terminated",parse_start);
5584 num = add_data( pRExC_state, 1, "S" );
5585 RExC_rxi->data->data[num]=(void*)sv_dat;
5586 SvREFCNT_inc_simple_void(sv_dat);
5589 ret = reganode(pRExC_state,
5590 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5594 Set_Node_Offset(ret, parse_start+1);
5595 Set_Node_Cur_Length(ret); /* MJD */
5597 nextchar(pRExC_state);
5601 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5603 case '<': /* (?<...) */
5604 if (*RExC_parse == '!')
5606 else if (*RExC_parse != '=')
5612 case '\'': /* (?'...') */
5613 name_start= RExC_parse;
5614 svname = reg_scan_name(pRExC_state,
5615 SIZE_ONLY ? /* reverse test from the others */
5616 REG_RSN_RETURN_NAME :
5617 REG_RSN_RETURN_NULL);
5618 if (RExC_parse == name_start) {
5620 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5623 if (*RExC_parse != paren)
5624 vFAIL2("Sequence (?%c... not terminated",
5625 paren=='>' ? '<' : paren);
5629 if (!svname) /* shouldnt happen */
5631 "panic: reg_scan_name returned NULL");
5632 if (!RExC_paren_names) {
5633 RExC_paren_names= newHV();
5634 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5636 RExC_paren_name_list= newAV();
5637 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5640 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5642 sv_dat = HeVAL(he_str);
5644 /* croak baby croak */
5646 "panic: paren_name hash element allocation failed");
5647 } else if ( SvPOK(sv_dat) ) {
5648 /* (?|...) can mean we have dupes so scan to check
5649 its already been stored. Maybe a flag indicating
5650 we are inside such a construct would be useful,
5651 but the arrays are likely to be quite small, so
5652 for now we punt -- dmq */
5653 IV count = SvIV(sv_dat);
5654 I32 *pv = (I32*)SvPVX(sv_dat);
5656 for ( i = 0 ; i < count ; i++ ) {
5657 if ( pv[i] == RExC_npar ) {
5663 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5664 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5665 pv[count] = RExC_npar;
5666 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5669 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5670 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5672 SvIV_set(sv_dat, 1);
5675 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5676 SvREFCNT_dec(svname);
5679 /*sv_dump(sv_dat);*/
5681 nextchar(pRExC_state);
5683 goto capturing_parens;
5685 RExC_seen |= REG_SEEN_LOOKBEHIND;
5687 case '=': /* (?=...) */
5688 RExC_seen_zerolen++;
5690 case '!': /* (?!...) */
5691 RExC_seen_zerolen++;
5692 if (*RExC_parse == ')') {
5693 ret=reg_node(pRExC_state, OPFAIL);
5694 nextchar(pRExC_state);
5698 case '|': /* (?|...) */
5699 /* branch reset, behave like a (?:...) except that
5700 buffers in alternations share the same numbers */
5702 after_freeze = freeze_paren = RExC_npar;
5704 case ':': /* (?:...) */
5705 case '>': /* (?>...) */
5707 case '$': /* (?$...) */
5708 case '@': /* (?@...) */
5709 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5711 case '#': /* (?#...) */
5712 while (*RExC_parse && *RExC_parse != ')')
5714 if (*RExC_parse != ')')
5715 FAIL("Sequence (?#... not terminated");
5716 nextchar(pRExC_state);
5719 case '0' : /* (?0) */
5720 case 'R' : /* (?R) */
5721 if (*RExC_parse != ')')
5722 FAIL("Sequence (?R) not terminated");
5723 ret = reg_node(pRExC_state, GOSTART);
5724 *flagp |= POSTPONED;
5725 nextchar(pRExC_state);
5728 { /* named and numeric backreferences */
5730 case '&': /* (?&NAME) */
5731 parse_start = RExC_parse - 1;
5734 SV *sv_dat = reg_scan_name(pRExC_state,
5735 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5736 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5738 goto gen_recurse_regop;
5741 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5743 vFAIL("Illegal pattern");
5745 goto parse_recursion;
5747 case '-': /* (?-1) */
5748 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5749 RExC_parse--; /* rewind to let it be handled later */
5753 case '1': case '2': case '3': case '4': /* (?1) */
5754 case '5': case '6': case '7': case '8': case '9':
5757 num = atoi(RExC_parse);
5758 parse_start = RExC_parse - 1; /* MJD */
5759 if (*RExC_parse == '-')
5761 while (isDIGIT(*RExC_parse))
5763 if (*RExC_parse!=')')
5764 vFAIL("Expecting close bracket");
5767 if ( paren == '-' ) {
5769 Diagram of capture buffer numbering.
5770 Top line is the normal capture buffer numbers
5771 Botton line is the negative indexing as from
5775 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5779 num = RExC_npar + num;
5782 vFAIL("Reference to nonexistent group");
5784 } else if ( paren == '+' ) {
5785 num = RExC_npar + num - 1;
5788 ret = reganode(pRExC_state, GOSUB, num);
5790 if (num > (I32)RExC_rx->nparens) {
5792 vFAIL("Reference to nonexistent group");
5794 ARG2L_SET( ret, RExC_recurse_count++);
5796 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5797 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5801 RExC_seen |= REG_SEEN_RECURSE;
5802 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5803 Set_Node_Offset(ret, parse_start); /* MJD */
5805 *flagp |= POSTPONED;
5806 nextchar(pRExC_state);
5808 } /* named and numeric backreferences */
5811 case '?': /* (??...) */
5813 if (*RExC_parse != '{') {
5815 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5818 *flagp |= POSTPONED;
5819 paren = *RExC_parse++;
5821 case '{': /* (?{...}) */
5826 char *s = RExC_parse;
5828 RExC_seen_zerolen++;
5829 RExC_seen |= REG_SEEN_EVAL;
5830 while (count && (c = *RExC_parse)) {
5841 if (*RExC_parse != ')') {
5843 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5847 OP_4tree *sop, *rop;
5848 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5851 Perl_save_re_context(aTHX);
5852 rop = sv_compile_2op(sv, &sop, "re", &pad);
5853 sop->op_private |= OPpREFCOUNTED;
5854 /* re_dup will OpREFCNT_inc */
5855 OpREFCNT_set(sop, 1);
5858 n = add_data(pRExC_state, 3, "nop");
5859 RExC_rxi->data->data[n] = (void*)rop;
5860 RExC_rxi->data->data[n+1] = (void*)sop;
5861 RExC_rxi->data->data[n+2] = (void*)pad;
5864 else { /* First pass */
5865 if (PL_reginterp_cnt < ++RExC_seen_evals
5867 /* No compiled RE interpolated, has runtime
5868 components ===> unsafe. */
5869 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5870 if (PL_tainting && PL_tainted)
5871 FAIL("Eval-group in insecure regular expression");
5872 #if PERL_VERSION > 8
5873 if (IN_PERL_COMPILETIME)
5878 nextchar(pRExC_state);
5880 ret = reg_node(pRExC_state, LOGICAL);
5883 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5884 /* deal with the length of this later - MJD */
5887 ret = reganode(pRExC_state, EVAL, n);
5888 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5889 Set_Node_Offset(ret, parse_start);
5892 case '(': /* (?(?{...})...) and (?(?=...)...) */
5895 if (RExC_parse[0] == '?') { /* (?(?...)) */
5896 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5897 || RExC_parse[1] == '<'
5898 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5901 ret = reg_node(pRExC_state, LOGICAL);
5904 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5908 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5909 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5911 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5912 char *name_start= RExC_parse++;
5914 SV *sv_dat=reg_scan_name(pRExC_state,
5915 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5916 if (RExC_parse == name_start || *RExC_parse != ch)
5917 vFAIL2("Sequence (?(%c... not terminated",
5918 (ch == '>' ? '<' : ch));
5921 num = add_data( pRExC_state, 1, "S" );
5922 RExC_rxi->data->data[num]=(void*)sv_dat;
5923 SvREFCNT_inc_simple_void(sv_dat);
5925 ret = reganode(pRExC_state,NGROUPP,num);
5926 goto insert_if_check_paren;
5928 else if (RExC_parse[0] == 'D' &&
5929 RExC_parse[1] == 'E' &&
5930 RExC_parse[2] == 'F' &&
5931 RExC_parse[3] == 'I' &&
5932 RExC_parse[4] == 'N' &&
5933 RExC_parse[5] == 'E')
5935 ret = reganode(pRExC_state,DEFINEP,0);
5938 goto insert_if_check_paren;
5940 else if (RExC_parse[0] == 'R') {
5943 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5944 parno = atoi(RExC_parse++);
5945 while (isDIGIT(*RExC_parse))
5947 } else if (RExC_parse[0] == '&') {
5950 sv_dat = reg_scan_name(pRExC_state,
5951 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5952 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5954 ret = reganode(pRExC_state,INSUBP,parno);
5955 goto insert_if_check_paren;
5957 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5960 parno = atoi(RExC_parse++);
5962 while (isDIGIT(*RExC_parse))
5964 ret = reganode(pRExC_state, GROUPP, parno);
5966 insert_if_check_paren:
5967 if ((c = *nextchar(pRExC_state)) != ')')
5968 vFAIL("Switch condition not recognized");
5970 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5971 br = regbranch(pRExC_state, &flags, 1,depth+1);
5973 br = reganode(pRExC_state, LONGJMP, 0);
5975 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5976 c = *nextchar(pRExC_state);
5981 vFAIL("(?(DEFINE)....) does not allow branches");
5982 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5983 regbranch(pRExC_state, &flags, 1,depth+1);
5984 REGTAIL(pRExC_state, ret, lastbr);
5987 c = *nextchar(pRExC_state);
5992 vFAIL("Switch (?(condition)... contains too many branches");
5993 ender = reg_node(pRExC_state, TAIL);
5994 REGTAIL(pRExC_state, br, ender);
5996 REGTAIL(pRExC_state, lastbr, ender);
5997 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6000 REGTAIL(pRExC_state, ret, ender);
6001 RExC_size++; /* XXX WHY do we need this?!!
6002 For large programs it seems to be required
6003 but I can't figure out why. -- dmq*/
6007 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6011 RExC_parse--; /* for vFAIL to print correctly */
6012 vFAIL("Sequence (? incomplete");
6016 parse_flags: /* (?i) */
6018 U32 posflags = 0, negflags = 0;
6019 U32 *flagsp = &posflags;
6021 while (*RExC_parse) {
6022 /* && strchr("iogcmsx", *RExC_parse) */
6023 /* (?g), (?gc) and (?o) are useless here
6024 and must be globally applied -- japhy */
6025 switch (*RExC_parse) {
6026 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6027 case ONCE_PAT_MOD: /* 'o' */
6028 case GLOBAL_PAT_MOD: /* 'g' */
6029 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6030 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6031 if (! (wastedflags & wflagbit) ) {
6032 wastedflags |= wflagbit;
6035 "Useless (%s%c) - %suse /%c modifier",
6036 flagsp == &negflags ? "?-" : "?",
6038 flagsp == &negflags ? "don't " : "",
6045 case CONTINUE_PAT_MOD: /* 'c' */
6046 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6047 if (! (wastedflags & WASTED_C) ) {
6048 wastedflags |= WASTED_GC;
6051 "Useless (%sc) - %suse /gc modifier",
6052 flagsp == &negflags ? "?-" : "?",
6053 flagsp == &negflags ? "don't " : ""
6058 case KEEPCOPY_PAT_MOD: /* 'p' */
6059 if (flagsp == &negflags) {
6060 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
6061 vWARN(RExC_parse + 1,"Useless use of (?-p)");
6063 *flagsp |= RXf_PMf_KEEPCOPY;
6067 if (flagsp == &negflags) {
6069 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6073 wastedflags = 0; /* reset so (?g-c) warns twice */
6079 RExC_flags |= posflags;
6080 RExC_flags &= ~negflags;
6082 oregflags |= posflags;
6083 oregflags &= ~negflags;
6085 nextchar(pRExC_state);
6096 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6101 }} /* one for the default block, one for the switch */
6108 ret = reganode(pRExC_state, OPEN, parno);
6111 RExC_nestroot = parno;
6112 if (RExC_seen & REG_SEEN_RECURSE
6113 && !RExC_open_parens[parno-1])
6115 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6116 "Setting open paren #%"IVdf" to %d\n",
6117 (IV)parno, REG_NODE_NUM(ret)));
6118 RExC_open_parens[parno-1]= ret;
6121 Set_Node_Length(ret, 1); /* MJD */
6122 Set_Node_Offset(ret, RExC_parse); /* MJD */
6130 /* Pick up the branches, linking them together. */
6131 parse_start = RExC_parse; /* MJD */
6132 br = regbranch(pRExC_state, &flags, 1,depth+1);
6135 if (RExC_npar > after_freeze)
6136 after_freeze = RExC_npar;
6137 RExC_npar = freeze_paren;
6140 /* branch_len = (paren != 0); */
6144 if (*RExC_parse == '|') {
6145 if (!SIZE_ONLY && RExC_extralen) {
6146 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6149 reginsert(pRExC_state, BRANCH, br, depth+1);
6150 Set_Node_Length(br, paren != 0);
6151 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6155 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6157 else if (paren == ':') {
6158 *flagp |= flags&SIMPLE;
6160 if (is_open) { /* Starts with OPEN. */
6161 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6163 else if (paren != '?') /* Not Conditional */
6165 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6167 while (*RExC_parse == '|') {
6168 if (!SIZE_ONLY && RExC_extralen) {
6169 ender = reganode(pRExC_state, LONGJMP,0);
6170 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6173 RExC_extralen += 2; /* Account for LONGJMP. */
6174 nextchar(pRExC_state);
6176 if (RExC_npar > after_freeze)
6177 after_freeze = RExC_npar;
6178 RExC_npar = freeze_paren;
6180 br = regbranch(pRExC_state, &flags, 0, depth+1);
6184 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6186 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6189 if (have_branch || paren != ':') {
6190 /* Make a closing node, and hook it on the end. */
6193 ender = reg_node(pRExC_state, TAIL);
6196 ender = reganode(pRExC_state, CLOSE, parno);
6197 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6198 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6199 "Setting close paren #%"IVdf" to %d\n",
6200 (IV)parno, REG_NODE_NUM(ender)));
6201 RExC_close_parens[parno-1]= ender;
6202 if (RExC_nestroot == parno)
6205 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6206 Set_Node_Length(ender,1); /* MJD */
6212 *flagp &= ~HASWIDTH;
6215 ender = reg_node(pRExC_state, SUCCEED);
6218 ender = reg_node(pRExC_state, END);
6220 assert(!RExC_opend); /* there can only be one! */
6225 REGTAIL(pRExC_state, lastbr, ender);
6227 if (have_branch && !SIZE_ONLY) {
6229 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6231 /* Hook the tails of the branches to the closing node. */
6232 for (br = ret; br; br = regnext(br)) {
6233 const U8 op = PL_regkind[OP(br)];
6235 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6237 else if (op == BRANCHJ) {
6238 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6246 static const char parens[] = "=!<,>";
6248 if (paren && (p = strchr(parens, paren))) {
6249 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6250 int flag = (p - parens) > 1;
6253 node = SUSPEND, flag = 0;
6254 reginsert(pRExC_state, node,ret, depth+1);
6255 Set_Node_Cur_Length(ret);
6256 Set_Node_Offset(ret, parse_start + 1);
6258 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6262 /* Check for proper termination. */
6264 RExC_flags = oregflags;
6265 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6266 RExC_parse = oregcomp_parse;
6267 vFAIL("Unmatched (");
6270 else if (!paren && RExC_parse < RExC_end) {
6271 if (*RExC_parse == ')') {
6273 vFAIL("Unmatched )");
6276 FAIL("Junk on end of regexp"); /* "Can't happen". */
6280 RExC_npar = after_freeze;
6285 - regbranch - one alternative of an | operator
6287 * Implements the concatenation operator.
6290 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6293 register regnode *ret;
6294 register regnode *chain = NULL;
6295 register regnode *latest;
6296 I32 flags = 0, c = 0;
6297 GET_RE_DEBUG_FLAGS_DECL;
6299 PERL_ARGS_ASSERT_REGBRANCH;
6301 DEBUG_PARSE("brnc");
6306 if (!SIZE_ONLY && RExC_extralen)
6307 ret = reganode(pRExC_state, BRANCHJ,0);
6309 ret = reg_node(pRExC_state, BRANCH);
6310 Set_Node_Length(ret, 1);
6314 if (!first && SIZE_ONLY)
6315 RExC_extralen += 1; /* BRANCHJ */
6317 *flagp = WORST; /* Tentatively. */
6320 nextchar(pRExC_state);
6321 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6323 latest = regpiece(pRExC_state, &flags,depth+1);
6324 if (latest == NULL) {
6325 if (flags & TRYAGAIN)
6329 else if (ret == NULL)
6331 *flagp |= flags&(HASWIDTH|POSTPONED);
6332 if (chain == NULL) /* First piece. */
6333 *flagp |= flags&SPSTART;
6336 REGTAIL(pRExC_state, chain, latest);
6341 if (chain == NULL) { /* Loop ran zero times. */
6342 chain = reg_node(pRExC_state, NOTHING);
6347 *flagp |= flags&SIMPLE;
6354 - regpiece - something followed by possible [*+?]
6356 * Note that the branching code sequences used for ? and the general cases
6357 * of * and + are somewhat optimized: they use the same NOTHING node as
6358 * both the endmarker for their branch list and the body of the last branch.
6359 * It might seem that this node could be dispensed with entirely, but the
6360 * endmarker role is not redundant.
6363 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6366 register regnode *ret;
6368 register char *next;
6370 const char * const origparse = RExC_parse;
6372 I32 max = REG_INFTY;
6374 const char *maxpos = NULL;
6375 GET_RE_DEBUG_FLAGS_DECL;
6377 PERL_ARGS_ASSERT_REGPIECE;
6379 DEBUG_PARSE("piec");
6381 ret = regatom(pRExC_state, &flags,depth+1);
6383 if (flags & TRYAGAIN)
6390 if (op == '{' && regcurly(RExC_parse)) {
6392 parse_start = RExC_parse; /* MJD */
6393 next = RExC_parse + 1;
6394 while (isDIGIT(*next) || *next == ',') {
6403 if (*next == '}') { /* got one */
6407 min = atoi(RExC_parse);
6411 maxpos = RExC_parse;
6413 if (!max && *maxpos != '0')
6414 max = REG_INFTY; /* meaning "infinity" */
6415 else if (max >= REG_INFTY)
6416 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6418 nextchar(pRExC_state);
6421 if ((flags&SIMPLE)) {
6422 RExC_naughty += 2 + RExC_naughty / 2;
6423 reginsert(pRExC_state, CURLY, ret, depth+1);
6424 Set_Node_Offset(ret, parse_start+1); /* MJD */
6425 Set_Node_Cur_Length(ret);
6428 regnode * const w = reg_node(pRExC_state, WHILEM);
6431 REGTAIL(pRExC_state, ret, w);
6432 if (!SIZE_ONLY && RExC_extralen) {
6433 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6434 reginsert(pRExC_state, NOTHING,ret, depth+1);
6435 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6437 reginsert(pRExC_state, CURLYX,ret, depth+1);
6439 Set_Node_Offset(ret, parse_start+1);
6440 Set_Node_Length(ret,
6441 op == '{' ? (RExC_parse - parse_start) : 1);
6443 if (!SIZE_ONLY && RExC_extralen)
6444 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6445 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6447 RExC_whilem_seen++, RExC_extralen += 3;
6448 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6457 vFAIL("Can't do {n,m} with n > m");
6459 ARG1_SET(ret, (U16)min);
6460 ARG2_SET(ret, (U16)max);
6472 #if 0 /* Now runtime fix should be reliable. */
6474 /* if this is reinstated, don't forget to put this back into perldiag:
6476 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6478 (F) The part of the regexp subject to either the * or + quantifier
6479 could match an empty string. The {#} shows in the regular
6480 expression about where the problem was discovered.
6484 if (!(flags&HASWIDTH) && op != '?')
6485 vFAIL("Regexp *+ operand could be empty");
6488 parse_start = RExC_parse;
6489 nextchar(pRExC_state);
6491 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6493 if (op == '*' && (flags&SIMPLE)) {
6494 reginsert(pRExC_state, STAR, ret, depth+1);
6498 else if (op == '*') {
6502 else if (op == '+' && (flags&SIMPLE)) {
6503 reginsert(pRExC_state, PLUS, ret, depth+1);
6507 else if (op == '+') {
6511 else if (op == '?') {
6516 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6518 "%.*s matches null string many times",
6519 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6523 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6524 nextchar(pRExC_state);
6525 reginsert(pRExC_state, MINMOD, ret, depth+1);
6526 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6528 #ifndef REG_ALLOW_MINMOD_SUSPEND
6531 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6533 nextchar(pRExC_state);
6534 ender = reg_node(pRExC_state, SUCCEED);
6535 REGTAIL(pRExC_state, ret, ender);
6536 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6538 ender = reg_node(pRExC_state, TAIL);
6539 REGTAIL(pRExC_state, ret, ender);
6543 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6545 vFAIL("Nested quantifiers");
6552 /* reg_namedseq(pRExC_state,UVp)
6554 This is expected to be called by a parser routine that has
6555 recognized'\N' and needs to handle the rest. RExC_parse is
6556 expected to point at the first char following the N at the time
6559 If valuep is non-null then it is assumed that we are parsing inside
6560 of a charclass definition and the first codepoint in the resolved
6561 string is returned via *valuep and the routine will return NULL.
6562 In this mode if a multichar string is returned from the charnames
6563 handler a warning will be issued, and only the first char in the
6564 sequence will be examined. If the string returned is zero length
6565 then the value of *valuep is undefined and NON-NULL will
6566 be returned to indicate failure. (This will NOT be a valid pointer
6569 If value is null then it is assumed that we are parsing normal text
6570 and inserts a new EXACT node into the program containing the resolved
6571 string and returns a pointer to the new node. If the string is
6572 zerolength a NOTHING node is emitted.
6574 On success RExC_parse is set to the char following the endbrace.
6575 Parsing failures will generate a fatal errorvia vFAIL(...)
6577 NOTE: We cache all results from the charnames handler locally in
6578 the RExC_charnames hash (created on first use) to prevent a charnames
6579 handler from playing silly-buggers and returning a short string and
6580 then a long string for a given pattern. Since the regexp program
6581 size is calculated during an initial parse this would result
6582 in a buffer overrun so we cache to prevent the charname result from
6583 changing during the course of the parse.
6587 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6589 char * name; /* start of the content of the name */
6590 char * endbrace; /* endbrace following the name */
6593 STRLEN len; /* this has various purposes throughout the code */
6594 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6595 regnode *ret = NULL;
6597 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6599 if (*RExC_parse != '{') {
6600 vFAIL("Missing braces on \\N{}");
6602 name = RExC_parse+1;
6603 endbrace = strchr(RExC_parse, '}');
6606 vFAIL("Missing right brace on \\N{}");
6608 RExC_parse = endbrace + 1;
6611 /* RExC_parse points at the beginning brace,
6612 endbrace points at the last */
6613 if ( name[0]=='U' && name[1]=='+' ) {
6614 /* its a "Unicode hex" notation {U+89AB} */
6615 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6616 | PERL_SCAN_DISALLOW_PREFIX
6617 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6619 len = (STRLEN)(endbrace - name - 2);
6620 cp = grok_hex(name + 2, &len, &fl, NULL);
6621 if ( len != (STRLEN)(endbrace - name - 2) ) {
6625 if (cp > 0xff) RExC_utf8 = 1;
6630 /* Need to convert to utf8 if either: won't fit into a byte, or the re
6631 * is going to be in utf8 and the representation changes under utf8. */
6632 if (cp > 0xff || (RExC_utf8 && ! UNI_IS_INVARIANT(cp))) {
6633 U8 string[UTF8_MAXBYTES+1];
6636 tmps = uvuni_to_utf8(string, cp);
6637 sv_str = newSVpvn_utf8((char*)string, tmps - string, TRUE);
6638 } else { /* Otherwise, no need for utf8, can skip that step */
6641 sv_str= newSVpvn(&string, 1);
6644 /* fetch the charnames handler for this scope */
6645 HV * const table = GvHV(PL_hintgv);
6647 hv_fetchs(table, "charnames", FALSE) :
6649 SV *cv= cvp ? *cvp : NULL;
6652 /* create an SV with the name as argument */
6653 sv_name = newSVpvn(name, endbrace - name);
6655 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6656 vFAIL2("Constant(\\N{%s}) unknown: "
6657 "(possibly a missing \"use charnames ...\")",
6660 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6661 vFAIL2("Constant(\\N{%s}): "
6662 "$^H{charnames} is not defined",SvPVX(sv_name));
6667 if (!RExC_charnames) {
6668 /* make sure our cache is allocated */
6669 RExC_charnames = newHV();
6670 sv_2mortal(MUTABLE_SV(RExC_charnames));
6672 /* see if we have looked this one up before */
6673 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6675 sv_str = HeVAL(he_str);
6688 count= call_sv(cv, G_SCALAR);
6690 if (count == 1) { /* XXXX is this right? dmq */
6692 SvREFCNT_inc_simple_void(sv_str);
6700 if ( !sv_str || !SvOK(sv_str) ) {
6701 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6702 "did not return a defined value",SvPVX(sv_name));
6704 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6709 char *p = SvPV(sv_str, len);
6712 if ( SvUTF8(sv_str) ) {
6713 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6717 We have to turn on utf8 for high bit chars otherwise
6718 we get failures with
6720 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6721 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6723 This is different from what \x{} would do with the same
6724 codepoint, where the condition is > 0xFF.
6731 /* warn if we havent used the whole string? */
6733 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6735 "Ignoring excess chars from \\N{%s} in character class",
6739 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6741 "Ignoring zero length \\N{%s} in character class",
6746 SvREFCNT_dec(sv_name);
6748 SvREFCNT_dec(sv_str);
6749 return len ? NULL : (regnode *)&len;
6750 } else if(SvCUR(sv_str)) {
6756 char * parse_start = name-3; /* needed for the offsets */
6758 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6760 ret = reg_node(pRExC_state,
6761 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6764 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6765 sv_utf8_upgrade(sv_str);
6766 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6770 p = SvPV(sv_str, len);
6772 /* len is the length written, charlen is the size the char read */
6773 for ( len = 0; p < pend; p += charlen ) {
6775 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6777 STRLEN foldlen,numlen;
6778 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6779 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6780 /* Emit all the Unicode characters. */
6782 for (foldbuf = tmpbuf;
6786 uvc = utf8_to_uvchr(foldbuf, &numlen);
6788 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6791 /* In EBCDIC the numlen
6792 * and unilen can differ. */
6794 if (numlen >= foldlen)
6798 break; /* "Can't happen." */
6801 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6813 RExC_size += STR_SZ(len);
6816 RExC_emit += STR_SZ(len);
6818 Set_Node_Cur_Length(ret); /* MJD */
6820 nextchar(pRExC_state);
6821 } else { /* zero length */
6822 ret = reg_node(pRExC_state,NOTHING);
6825 SvREFCNT_dec(sv_str);
6828 SvREFCNT_dec(sv_name);
6838 * It returns the code point in utf8 for the value in *encp.
6839 * value: a code value in the source encoding
6840 * encp: a pointer to an Encode object
6842 * If the result from Encode is not a single character,
6843 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6846 S_reg_recode(pTHX_ const char value, SV **encp)
6849 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
6850 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6851 const STRLEN newlen = SvCUR(sv);
6852 UV uv = UNICODE_REPLACEMENT;
6854 PERL_ARGS_ASSERT_REG_RECODE;
6858 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6861 if (!newlen || numlen != newlen) {
6862 uv = UNICODE_REPLACEMENT;
6870 - regatom - the lowest level
6872 Try to identify anything special at the start of the pattern. If there
6873 is, then handle it as required. This may involve generating a single regop,
6874 such as for an assertion; or it may involve recursing, such as to
6875 handle a () structure.
6877 If the string doesn't start with something special then we gobble up
6878 as much literal text as we can.
6880 Once we have been able to handle whatever type of thing started the
6881 sequence, we return.
6883 Note: we have to be careful with escapes, as they can be both literal
6884 and special, and in the case of \10 and friends can either, depending
6885 on context. Specifically there are two seperate switches for handling
6886 escape sequences, with the one for handling literal escapes requiring
6887 a dummy entry for all of the special escapes that are actually handled
6892 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6895 register regnode *ret = NULL;
6897 char *parse_start = RExC_parse;
6898 GET_RE_DEBUG_FLAGS_DECL;
6899 DEBUG_PARSE("atom");
6900 *flagp = WORST; /* Tentatively. */
6902 PERL_ARGS_ASSERT_REGATOM;
6905 switch ((U8)*RExC_parse) {
6907 RExC_seen_zerolen++;
6908 nextchar(pRExC_state);
6909 if (RExC_flags & RXf_PMf_MULTILINE)
6910 ret = reg_node(pRExC_state, MBOL);
6911 else if (RExC_flags & RXf_PMf_SINGLELINE)
6912 ret = reg_node(pRExC_state, SBOL);
6914 ret = reg_node(pRExC_state, BOL);
6915 Set_Node_Length(ret, 1); /* MJD */
6918 nextchar(pRExC_state);
6920 RExC_seen_zerolen++;
6921 if (RExC_flags & RXf_PMf_MULTILINE)
6922 ret = reg_node(pRExC_state, MEOL);
6923 else if (RExC_flags & RXf_PMf_SINGLELINE)
6924 ret = reg_node(pRExC_state, SEOL);
6926 ret = reg_node(pRExC_state, EOL);
6927 Set_Node_Length(ret, 1); /* MJD */
6930 nextchar(pRExC_state);
6931 if (RExC_flags & RXf_PMf_SINGLELINE)
6932 ret = reg_node(pRExC_state, SANY);
6934 ret = reg_node(pRExC_state, REG_ANY);
6935 *flagp |= HASWIDTH|SIMPLE;
6937 Set_Node_Length(ret, 1); /* MJD */
6941 char * const oregcomp_parse = ++RExC_parse;
6942 ret = regclass(pRExC_state,depth+1);
6943 if (*RExC_parse != ']') {
6944 RExC_parse = oregcomp_parse;
6945 vFAIL("Unmatched [");
6947 nextchar(pRExC_state);
6948 *flagp |= HASWIDTH|SIMPLE;
6949 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6953 nextchar(pRExC_state);
6954 ret = reg(pRExC_state, 1, &flags,depth+1);
6956 if (flags & TRYAGAIN) {
6957 if (RExC_parse == RExC_end) {
6958 /* Make parent create an empty node if needed. */
6966 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6970 if (flags & TRYAGAIN) {
6974 vFAIL("Internal urp");
6975 /* Supposed to be caught earlier. */
6978 if (!regcurly(RExC_parse)) {
6987 vFAIL("Quantifier follows nothing");
6995 len=0; /* silence a spurious compiler warning */
6996 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
6997 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
6998 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
6999 ret = reganode(pRExC_state, FOLDCHAR, cp);
7000 Set_Node_Length(ret, 1); /* MJD */
7001 nextchar(pRExC_state); /* kill whitespace under /x */
7009 This switch handles escape sequences that resolve to some kind
7010 of special regop and not to literal text. Escape sequnces that
7011 resolve to literal text are handled below in the switch marked
7014 Every entry in this switch *must* have a corresponding entry
7015 in the literal escape switch. However, the opposite is not
7016 required, as the default for this switch is to jump to the
7017 literal text handling code.
7019 switch ((U8)*++RExC_parse) {
7024 /* Special Escapes */
7026 RExC_seen_zerolen++;
7027 ret = reg_node(pRExC_state, SBOL);
7029 goto finish_meta_pat;
7031 ret = reg_node(pRExC_state, GPOS);
7032 RExC_seen |= REG_SEEN_GPOS;
7034 goto finish_meta_pat;
7036 RExC_seen_zerolen++;
7037 ret = reg_node(pRExC_state, KEEPS);
7039 /* XXX:dmq : disabling in-place substitution seems to
7040 * be necessary here to avoid cases of memory corruption, as
7041 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7043 RExC_seen |= REG_SEEN_LOOKBEHIND;
7044 goto finish_meta_pat;
7046 ret = reg_node(pRExC_state, SEOL);
7048 RExC_seen_zerolen++; /* Do not optimize RE away */
7049 goto finish_meta_pat;
7051 ret = reg_node(pRExC_state, EOS);
7053 RExC_seen_zerolen++; /* Do not optimize RE away */
7054 goto finish_meta_pat;
7056 ret = reg_node(pRExC_state, CANY);
7057 RExC_seen |= REG_SEEN_CANY;
7058 *flagp |= HASWIDTH|SIMPLE;
7059 goto finish_meta_pat;
7061 ret = reg_node(pRExC_state, CLUMP);
7063 goto finish_meta_pat;
7065 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
7066 *flagp |= HASWIDTH|SIMPLE;
7067 goto finish_meta_pat;
7069 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
7070 *flagp |= HASWIDTH|SIMPLE;
7071 goto finish_meta_pat;
7073 RExC_seen_zerolen++;
7074 RExC_seen |= REG_SEEN_LOOKBEHIND;
7075 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
7077 goto finish_meta_pat;
7079 RExC_seen_zerolen++;
7080 RExC_seen |= REG_SEEN_LOOKBEHIND;
7081 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
7083 goto finish_meta_pat;
7085 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
7086 *flagp |= HASWIDTH|SIMPLE;
7087 goto finish_meta_pat;
7089 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
7090 *flagp |= HASWIDTH|SIMPLE;
7091 goto finish_meta_pat;
7093 ret = reg_node(pRExC_state, DIGIT);
7094 *flagp |= HASWIDTH|SIMPLE;
7095 goto finish_meta_pat;
7097 ret = reg_node(pRExC_state, NDIGIT);
7098 *flagp |= HASWIDTH|SIMPLE;
7099 goto finish_meta_pat;
7101 ret = reg_node(pRExC_state, LNBREAK);
7102 *flagp |= HASWIDTH|SIMPLE;
7103 goto finish_meta_pat;
7105 ret = reg_node(pRExC_state, HORIZWS);
7106 *flagp |= HASWIDTH|SIMPLE;
7107 goto finish_meta_pat;
7109 ret = reg_node(pRExC_state, NHORIZWS);
7110 *flagp |= HASWIDTH|SIMPLE;
7111 goto finish_meta_pat;
7113 ret = reg_node(pRExC_state, VERTWS);
7114 *flagp |= HASWIDTH|SIMPLE;
7115 goto finish_meta_pat;
7117 ret = reg_node(pRExC_state, NVERTWS);
7118 *flagp |= HASWIDTH|SIMPLE;
7120 nextchar(pRExC_state);
7121 Set_Node_Length(ret, 2); /* MJD */
7126 char* const oldregxend = RExC_end;
7128 char* parse_start = RExC_parse - 2;
7131 if (RExC_parse[1] == '{') {
7132 /* a lovely hack--pretend we saw [\pX] instead */
7133 RExC_end = strchr(RExC_parse, '}');
7135 const U8 c = (U8)*RExC_parse;
7137 RExC_end = oldregxend;
7138 vFAIL2("Missing right brace on \\%c{}", c);
7143 RExC_end = RExC_parse + 2;
7144 if (RExC_end > oldregxend)
7145 RExC_end = oldregxend;
7149 ret = regclass(pRExC_state,depth+1);
7151 RExC_end = oldregxend;
7154 Set_Node_Offset(ret, parse_start + 2);
7155 Set_Node_Cur_Length(ret);
7156 nextchar(pRExC_state);
7157 *flagp |= HASWIDTH|SIMPLE;
7161 /* Handle \N{NAME} here and not below because it can be
7162 multicharacter. join_exact() will join them up later on.
7163 Also this makes sure that things like /\N{BLAH}+/ and
7164 \N{BLAH} being multi char Just Happen. dmq*/
7166 ret= reg_namedseq(pRExC_state, NULL);
7168 case 'k': /* Handle \k<NAME> and \k'NAME' */
7171 char ch= RExC_parse[1];
7172 if (ch != '<' && ch != '\'' && ch != '{') {
7174 vFAIL2("Sequence %.2s... not terminated",parse_start);
7176 /* this pretty much dupes the code for (?P=...) in reg(), if
7177 you change this make sure you change that */
7178 char* name_start = (RExC_parse += 2);
7180 SV *sv_dat = reg_scan_name(pRExC_state,
7181 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7182 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7183 if (RExC_parse == name_start || *RExC_parse != ch)
7184 vFAIL2("Sequence %.3s... not terminated",parse_start);
7187 num = add_data( pRExC_state, 1, "S" );
7188 RExC_rxi->data->data[num]=(void*)sv_dat;
7189 SvREFCNT_inc_simple_void(sv_dat);
7193 ret = reganode(pRExC_state,
7194 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7198 /* override incorrect value set in reganode MJD */
7199 Set_Node_Offset(ret, parse_start+1);
7200 Set_Node_Cur_Length(ret); /* MJD */
7201 nextchar(pRExC_state);
7207 case '1': case '2': case '3': case '4':
7208 case '5': case '6': case '7': case '8': case '9':
7211 bool isg = *RExC_parse == 'g';
7216 if (*RExC_parse == '{') {
7220 if (*RExC_parse == '-') {
7224 if (hasbrace && !isDIGIT(*RExC_parse)) {
7225 if (isrel) RExC_parse--;
7227 goto parse_named_seq;
7229 num = atoi(RExC_parse);
7230 if (isg && num == 0)
7231 vFAIL("Reference to invalid group 0");
7233 num = RExC_npar - num;
7235 vFAIL("Reference to nonexistent or unclosed group");
7237 if (!isg && num > 9 && num >= RExC_npar)
7240 char * const parse_start = RExC_parse - 1; /* MJD */
7241 while (isDIGIT(*RExC_parse))
7243 if (parse_start == RExC_parse - 1)
7244 vFAIL("Unterminated \\g... pattern");
7246 if (*RExC_parse != '}')
7247 vFAIL("Unterminated \\g{...} pattern");
7251 if (num > (I32)RExC_rx->nparens)
7252 vFAIL("Reference to nonexistent group");
7255 ret = reganode(pRExC_state,
7256 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7260 /* override incorrect value set in reganode MJD */
7261 Set_Node_Offset(ret, parse_start+1);
7262 Set_Node_Cur_Length(ret); /* MJD */
7264 nextchar(pRExC_state);
7269 if (RExC_parse >= RExC_end)
7270 FAIL("Trailing \\");
7273 /* Do not generate "unrecognized" warnings here, we fall
7274 back into the quick-grab loop below */
7281 if (RExC_flags & RXf_PMf_EXTENDED) {
7282 if ( reg_skipcomment( pRExC_state ) )
7289 register STRLEN len;
7294 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7296 parse_start = RExC_parse - 1;
7302 ret = reg_node(pRExC_state,
7303 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7305 for (len = 0, p = RExC_parse - 1;
7306 len < 127 && p < RExC_end;
7309 char * const oldp = p;
7311 if (RExC_flags & RXf_PMf_EXTENDED)
7312 p = regwhite( pRExC_state, p );
7317 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7318 goto normal_default;
7328 /* Literal Escapes Switch
7330 This switch is meant to handle escape sequences that
7331 resolve to a literal character.
7333 Every escape sequence that represents something
7334 else, like an assertion or a char class, is handled
7335 in the switch marked 'Special Escapes' above in this
7336 routine, but also has an entry here as anything that
7337 isn't explicitly mentioned here will be treated as
7338 an unescaped equivalent literal.
7342 /* These are all the special escapes. */
7346 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7347 goto normal_default;
7348 case 'A': /* Start assertion */
7349 case 'b': case 'B': /* Word-boundary assertion*/
7350 case 'C': /* Single char !DANGEROUS! */
7351 case 'd': case 'D': /* digit class */
7352 case 'g': case 'G': /* generic-backref, pos assertion */
7353 case 'h': case 'H': /* HORIZWS */
7354 case 'k': case 'K': /* named backref, keep marker */
7355 case 'N': /* named char sequence */
7356 case 'p': case 'P': /* Unicode property */
7357 case 'R': /* LNBREAK */
7358 case 's': case 'S': /* space class */
7359 case 'v': case 'V': /* VERTWS */
7360 case 'w': case 'W': /* word class */
7361 case 'X': /* eXtended Unicode "combining character sequence" */
7362 case 'z': case 'Z': /* End of line/string assertion */
7366 /* Anything after here is an escape that resolves to a
7367 literal. (Except digits, which may or may not)
7386 ender = ASCII_TO_NATIVE('\033');
7390 ender = ASCII_TO_NATIVE('\007');
7395 char* const e = strchr(p, '}');
7399 vFAIL("Missing right brace on \\x{}");
7402 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7403 | PERL_SCAN_DISALLOW_PREFIX;
7404 STRLEN numlen = e - p - 1;
7405 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7412 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7414 ender = grok_hex(p, &numlen, &flags, NULL);
7417 if (PL_encoding && ender < 0x100)
7418 goto recode_encoding;
7422 ender = UCHARAT(p++);
7423 ender = toCTRL(ender);
7425 case '0': case '1': case '2': case '3':case '4':
7426 case '5': case '6': case '7': case '8':case '9':
7428 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7431 ender = grok_oct(p, &numlen, &flags, NULL);
7438 if (PL_encoding && ender < 0x100)
7439 goto recode_encoding;
7443 SV* enc = PL_encoding;
7444 ender = reg_recode((const char)(U8)ender, &enc);
7445 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7446 vWARN(p, "Invalid escape in the specified encoding");
7452 FAIL("Trailing \\");
7455 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
7456 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7457 goto normal_default;
7462 if (UTF8_IS_START(*p) && UTF) {
7464 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7465 &numlen, UTF8_ALLOW_DEFAULT);
7472 if ( RExC_flags & RXf_PMf_EXTENDED)
7473 p = regwhite( pRExC_state, p );
7475 /* Prime the casefolded buffer. */
7476 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7478 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7483 /* Emit all the Unicode characters. */
7485 for (foldbuf = tmpbuf;
7487 foldlen -= numlen) {
7488 ender = utf8_to_uvchr(foldbuf, &numlen);
7490 const STRLEN unilen = reguni(pRExC_state, ender, s);
7493 /* In EBCDIC the numlen
7494 * and unilen can differ. */
7496 if (numlen >= foldlen)
7500 break; /* "Can't happen." */
7504 const STRLEN unilen = reguni(pRExC_state, ender, s);
7513 REGC((char)ender, s++);
7519 /* Emit all the Unicode characters. */
7521 for (foldbuf = tmpbuf;
7523 foldlen -= numlen) {
7524 ender = utf8_to_uvchr(foldbuf, &numlen);
7526 const STRLEN unilen = reguni(pRExC_state, ender, s);
7529 /* In EBCDIC the numlen
7530 * and unilen can differ. */
7532 if (numlen >= foldlen)
7540 const STRLEN unilen = reguni(pRExC_state, ender, s);
7549 REGC((char)ender, s++);
7553 Set_Node_Cur_Length(ret); /* MJD */
7554 nextchar(pRExC_state);
7556 /* len is STRLEN which is unsigned, need to copy to signed */
7559 vFAIL("Internal disaster");
7563 if (len == 1 && UNI_IS_INVARIANT(ender))
7567 RExC_size += STR_SZ(len);
7570 RExC_emit += STR_SZ(len);
7580 S_regwhite( RExC_state_t *pRExC_state, char *p )
7582 const char *e = RExC_end;
7584 PERL_ARGS_ASSERT_REGWHITE;
7589 else if (*p == '#') {
7598 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7606 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7607 Character classes ([:foo:]) can also be negated ([:^foo:]).
7608 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7609 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7610 but trigger failures because they are currently unimplemented. */
7612 #define POSIXCC_DONE(c) ((c) == ':')
7613 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7614 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7617 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7620 I32 namedclass = OOB_NAMEDCLASS;
7622 PERL_ARGS_ASSERT_REGPPOSIXCC;
7624 if (value == '[' && RExC_parse + 1 < RExC_end &&
7625 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7626 POSIXCC(UCHARAT(RExC_parse))) {
7627 const char c = UCHARAT(RExC_parse);
7628 char* const s = RExC_parse++;
7630 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7632 if (RExC_parse == RExC_end)
7633 /* Grandfather lone [:, [=, [. */
7636 const char* const t = RExC_parse++; /* skip over the c */
7639 if (UCHARAT(RExC_parse) == ']') {
7640 const char *posixcc = s + 1;
7641 RExC_parse++; /* skip over the ending ] */
7644 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7645 const I32 skip = t - posixcc;
7647 /* Initially switch on the length of the name. */
7650 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7651 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7654 /* Names all of length 5. */
7655 /* alnum alpha ascii blank cntrl digit graph lower
7656 print punct space upper */
7657 /* Offset 4 gives the best switch position. */
7658 switch (posixcc[4]) {
7660 if (memEQ(posixcc, "alph", 4)) /* alpha */
7661 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7664 if (memEQ(posixcc, "spac", 4)) /* space */
7665 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7668 if (memEQ(posixcc, "grap", 4)) /* graph */
7669 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7672 if (memEQ(posixcc, "asci", 4)) /* ascii */
7673 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7676 if (memEQ(posixcc, "blan", 4)) /* blank */
7677 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7680 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7681 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7684 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7685 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7688 if (memEQ(posixcc, "lowe", 4)) /* lower */
7689 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7690 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7691 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7694 if (memEQ(posixcc, "digi", 4)) /* digit */
7695 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7696 else if (memEQ(posixcc, "prin", 4)) /* print */
7697 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7698 else if (memEQ(posixcc, "punc", 4)) /* punct */
7699 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7704 if (memEQ(posixcc, "xdigit", 6))
7705 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7709 if (namedclass == OOB_NAMEDCLASS)
7710 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7712 assert (posixcc[skip] == ':');
7713 assert (posixcc[skip+1] == ']');
7714 } else if (!SIZE_ONLY) {
7715 /* [[=foo=]] and [[.foo.]] are still future. */
7717 /* adjust RExC_parse so the warning shows after
7719 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7721 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7724 /* Maternal grandfather:
7725 * "[:" ending in ":" but not in ":]" */
7735 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7739 PERL_ARGS_ASSERT_CHECKPOSIXCC;
7741 if (POSIXCC(UCHARAT(RExC_parse))) {
7742 const char *s = RExC_parse;
7743 const char c = *s++;
7747 if (*s && c == *s && s[1] == ']') {
7748 if (ckWARN(WARN_REGEXP))
7750 "POSIX syntax [%c %c] belongs inside character classes",
7753 /* [[=foo=]] and [[.foo.]] are still future. */
7754 if (POSIXCC_NOTYET(c)) {
7755 /* adjust RExC_parse so the error shows after
7757 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7759 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7766 #define _C_C_T_(NAME,TEST,WORD) \
7769 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7771 for (value = 0; value < 256; value++) \
7773 ANYOF_BITMAP_SET(ret, value); \
7778 case ANYOF_N##NAME: \
7780 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7782 for (value = 0; value < 256; value++) \
7784 ANYOF_BITMAP_SET(ret, value); \
7790 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7792 for (value = 0; value < 256; value++) \
7794 ANYOF_BITMAP_SET(ret, value); \
7798 case ANYOF_N##NAME: \
7799 for (value = 0; value < 256; value++) \
7801 ANYOF_BITMAP_SET(ret, value); \
7807 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
7808 so that it is possible to override the option here without having to
7809 rebuild the entire core. as we are required to do if we change regcomp.h
7810 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
7812 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
7813 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
7816 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
7817 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
7819 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
7823 parse a class specification and produce either an ANYOF node that
7824 matches the pattern or if the pattern matches a single char only and
7825 that char is < 256 and we are case insensitive then we produce an
7830 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7833 register UV nextvalue;
7834 register IV prevvalue = OOB_UNICODE;
7835 register IV range = 0;
7836 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7837 register regnode *ret;
7840 char *rangebegin = NULL;
7841 bool need_class = 0;
7844 bool optimize_invert = TRUE;
7845 AV* unicode_alternate = NULL;
7847 UV literal_endpoint = 0;
7849 UV stored = 0; /* number of chars stored in the class */
7851 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7852 case we need to change the emitted regop to an EXACT. */
7853 const char * orig_parse = RExC_parse;
7854 GET_RE_DEBUG_FLAGS_DECL;
7856 PERL_ARGS_ASSERT_REGCLASS;
7858 PERL_UNUSED_ARG(depth);
7861 DEBUG_PARSE("clas");
7863 /* Assume we are going to generate an ANYOF node. */
7864 ret = reganode(pRExC_state, ANYOF, 0);
7867 ANYOF_FLAGS(ret) = 0;
7869 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7873 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7877 RExC_size += ANYOF_SKIP;
7878 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7881 RExC_emit += ANYOF_SKIP;
7883 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7885 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7886 ANYOF_BITMAP_ZERO(ret);
7887 listsv = newSVpvs("# comment\n");
7890 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7892 if (!SIZE_ONLY && POSIXCC(nextvalue))
7893 checkposixcc(pRExC_state);
7895 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7896 if (UCHARAT(RExC_parse) == ']')
7900 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7904 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7907 rangebegin = RExC_parse;
7909 value = utf8n_to_uvchr((U8*)RExC_parse,
7910 RExC_end - RExC_parse,
7911 &numlen, UTF8_ALLOW_DEFAULT);
7912 RExC_parse += numlen;
7915 value = UCHARAT(RExC_parse++);
7917 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7918 if (value == '[' && POSIXCC(nextvalue))
7919 namedclass = regpposixcc(pRExC_state, value);
7920 else if (value == '\\') {
7922 value = utf8n_to_uvchr((U8*)RExC_parse,
7923 RExC_end - RExC_parse,
7924 &numlen, UTF8_ALLOW_DEFAULT);
7925 RExC_parse += numlen;
7928 value = UCHARAT(RExC_parse++);
7929 /* Some compilers cannot handle switching on 64-bit integer
7930 * values, therefore value cannot be an UV. Yes, this will
7931 * be a problem later if we want switch on Unicode.
7932 * A similar issue a little bit later when switching on
7933 * namedclass. --jhi */
7934 switch ((I32)value) {
7935 case 'w': namedclass = ANYOF_ALNUM; break;
7936 case 'W': namedclass = ANYOF_NALNUM; break;
7937 case 's': namedclass = ANYOF_SPACE; break;
7938 case 'S': namedclass = ANYOF_NSPACE; break;
7939 case 'd': namedclass = ANYOF_DIGIT; break;
7940 case 'D': namedclass = ANYOF_NDIGIT; break;
7941 case 'v': namedclass = ANYOF_VERTWS; break;
7942 case 'V': namedclass = ANYOF_NVERTWS; break;
7943 case 'h': namedclass = ANYOF_HORIZWS; break;
7944 case 'H': namedclass = ANYOF_NHORIZWS; break;
7945 case 'N': /* Handle \N{NAME} in class */
7947 /* We only pay attention to the first char of
7948 multichar strings being returned. I kinda wonder
7949 if this makes sense as it does change the behaviour
7950 from earlier versions, OTOH that behaviour was broken
7952 UV v; /* value is register so we cant & it /grrr */
7953 if (reg_namedseq(pRExC_state, &v)) {
7963 if (RExC_parse >= RExC_end)
7964 vFAIL2("Empty \\%c{}", (U8)value);
7965 if (*RExC_parse == '{') {
7966 const U8 c = (U8)value;
7967 e = strchr(RExC_parse++, '}');
7969 vFAIL2("Missing right brace on \\%c{}", c);
7970 while (isSPACE(UCHARAT(RExC_parse)))
7972 if (e == RExC_parse)
7973 vFAIL2("Empty \\%c{}", c);
7975 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7983 if (UCHARAT(RExC_parse) == '^') {
7986 value = value == 'p' ? 'P' : 'p'; /* toggle */
7987 while (isSPACE(UCHARAT(RExC_parse))) {
7992 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7993 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7996 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7997 namedclass = ANYOF_MAX; /* no official name, but it's named */
8000 case 'n': value = '\n'; break;
8001 case 'r': value = '\r'; break;
8002 case 't': value = '\t'; break;
8003 case 'f': value = '\f'; break;
8004 case 'b': value = '\b'; break;
8005 case 'e': value = ASCII_TO_NATIVE('\033');break;
8006 case 'a': value = ASCII_TO_NATIVE('\007');break;
8008 if (*RExC_parse == '{') {
8009 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8010 | PERL_SCAN_DISALLOW_PREFIX;
8011 char * const e = strchr(RExC_parse++, '}');
8013 vFAIL("Missing right brace on \\x{}");
8015 numlen = e - RExC_parse;
8016 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8020 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8022 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8023 RExC_parse += numlen;
8025 if (PL_encoding && value < 0x100)
8026 goto recode_encoding;
8029 value = UCHARAT(RExC_parse++);
8030 value = toCTRL(value);
8032 case '0': case '1': case '2': case '3': case '4':
8033 case '5': case '6': case '7': case '8': case '9':
8037 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8038 RExC_parse += numlen;
8039 if (PL_encoding && value < 0x100)
8040 goto recode_encoding;
8045 SV* enc = PL_encoding;
8046 value = reg_recode((const char)(U8)value, &enc);
8047 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
8049 "Invalid escape in the specified encoding");
8053 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
8055 "Unrecognized escape \\%c in character class passed through",
8059 } /* end of \blah */
8065 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8067 if (!SIZE_ONLY && !need_class)
8068 ANYOF_CLASS_ZERO(ret);
8072 /* a bad range like a-\d, a-[:digit:] ? */
8075 if (ckWARN(WARN_REGEXP)) {
8077 RExC_parse >= rangebegin ?
8078 RExC_parse - rangebegin : 0;
8080 "False [] range \"%*.*s\"",
8083 if (prevvalue < 256) {
8084 ANYOF_BITMAP_SET(ret, prevvalue);
8085 ANYOF_BITMAP_SET(ret, '-');
8088 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8089 Perl_sv_catpvf(aTHX_ listsv,
8090 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8094 range = 0; /* this was not a true range */
8100 const char *what = NULL;
8103 if (namedclass > OOB_NAMEDCLASS)
8104 optimize_invert = FALSE;
8105 /* Possible truncation here but in some 64-bit environments
8106 * the compiler gets heartburn about switch on 64-bit values.
8107 * A similar issue a little earlier when switching on value.
8109 switch ((I32)namedclass) {
8111 case _C_C_T_(ALNUMC, isALNUMC(value), POSIX_CC_UNI_NAME("Alnum"));
8112 case _C_C_T_(ALPHA, isALPHA(value), POSIX_CC_UNI_NAME("Alpha"));
8113 case _C_C_T_(BLANK, isBLANK(value), POSIX_CC_UNI_NAME("Blank"));
8114 case _C_C_T_(CNTRL, isCNTRL(value), POSIX_CC_UNI_NAME("Cntrl"));
8115 case _C_C_T_(GRAPH, isGRAPH(value), POSIX_CC_UNI_NAME("Graph"));
8116 case _C_C_T_(LOWER, isLOWER(value), POSIX_CC_UNI_NAME("Lower"));
8117 case _C_C_T_(PRINT, isPRINT(value), POSIX_CC_UNI_NAME("Print"));
8118 case _C_C_T_(PSXSPC, isPSXSPC(value), POSIX_CC_UNI_NAME("Space"));
8119 case _C_C_T_(PUNCT, isPUNCT(value), POSIX_CC_UNI_NAME("Punct"));
8120 case _C_C_T_(UPPER, isUPPER(value), POSIX_CC_UNI_NAME("Upper"));
8121 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8122 case _C_C_T_(ALNUM, isALNUM(value), "Word");
8123 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
8125 case _C_C_T_(SPACE, isSPACE(value), "PerlSpace");
8126 case _C_C_T_(ALNUM, isALNUM(value), "PerlWord");
8128 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
8129 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8130 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8133 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8136 for (value = 0; value < 128; value++)
8137 ANYOF_BITMAP_SET(ret, value);
8139 for (value = 0; value < 256; value++) {
8141 ANYOF_BITMAP_SET(ret, value);
8150 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8153 for (value = 128; value < 256; value++)
8154 ANYOF_BITMAP_SET(ret, value);
8156 for (value = 0; value < 256; value++) {
8157 if (!isASCII(value))
8158 ANYOF_BITMAP_SET(ret, value);
8167 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8169 /* consecutive digits assumed */
8170 for (value = '0'; value <= '9'; value++)
8171 ANYOF_BITMAP_SET(ret, value);
8174 what = POSIX_CC_UNI_NAME("Digit");
8178 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8180 /* consecutive digits assumed */
8181 for (value = 0; value < '0'; value++)
8182 ANYOF_BITMAP_SET(ret, value);
8183 for (value = '9' + 1; value < 256; value++)
8184 ANYOF_BITMAP_SET(ret, value);
8187 what = POSIX_CC_UNI_NAME("Digit");
8190 /* this is to handle \p and \P */
8193 vFAIL("Invalid [::] class");
8197 /* Strings such as "+utf8::isWord\n" */
8198 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8201 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8204 } /* end of namedclass \blah */
8207 if (prevvalue > (IV)value) /* b-a */ {
8208 const int w = RExC_parse - rangebegin;
8209 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8210 range = 0; /* not a valid range */
8214 prevvalue = value; /* save the beginning of the range */
8215 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8216 RExC_parse[1] != ']') {
8219 /* a bad range like \w-, [:word:]- ? */
8220 if (namedclass > OOB_NAMEDCLASS) {
8221 if (ckWARN(WARN_REGEXP)) {
8223 RExC_parse >= rangebegin ?
8224 RExC_parse - rangebegin : 0;
8226 "False [] range \"%*.*s\"",
8230 ANYOF_BITMAP_SET(ret, '-');
8232 range = 1; /* yeah, it's a range! */
8233 continue; /* but do it the next time */
8237 /* now is the next time */
8238 /*stored += (value - prevvalue + 1);*/
8240 if (prevvalue < 256) {
8241 const IV ceilvalue = value < 256 ? value : 255;
8244 /* In EBCDIC [\x89-\x91] should include
8245 * the \x8e but [i-j] should not. */
8246 if (literal_endpoint == 2 &&
8247 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8248 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8250 if (isLOWER(prevvalue)) {
8251 for (i = prevvalue; i <= ceilvalue; i++)
8252 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8254 ANYOF_BITMAP_SET(ret, i);
8257 for (i = prevvalue; i <= ceilvalue; i++)
8258 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8260 ANYOF_BITMAP_SET(ret, i);
8266 for (i = prevvalue; i <= ceilvalue; i++) {
8267 if (!ANYOF_BITMAP_TEST(ret,i)) {
8269 ANYOF_BITMAP_SET(ret, i);
8273 if (value > 255 || UTF) {
8274 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8275 const UV natvalue = NATIVE_TO_UNI(value);
8276 stored+=2; /* can't optimize this class */
8277 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8278 if (prevnatvalue < natvalue) { /* what about > ? */
8279 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8280 prevnatvalue, natvalue);
8282 else if (prevnatvalue == natvalue) {
8283 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8285 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8287 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8289 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8290 if (RExC_precomp[0] == ':' &&
8291 RExC_precomp[1] == '[' &&
8292 (f == 0xDF || f == 0x92)) {
8293 f = NATIVE_TO_UNI(f);
8296 /* If folding and foldable and a single
8297 * character, insert also the folded version
8298 * to the charclass. */
8300 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8301 if ((RExC_precomp[0] == ':' &&
8302 RExC_precomp[1] == '[' &&
8304 (value == 0xFB05 || value == 0xFB06))) ?
8305 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8306 foldlen == (STRLEN)UNISKIP(f) )
8308 if (foldlen == (STRLEN)UNISKIP(f))
8310 Perl_sv_catpvf(aTHX_ listsv,
8313 /* Any multicharacter foldings
8314 * require the following transform:
8315 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8316 * where E folds into "pq" and F folds
8317 * into "rst", all other characters
8318 * fold to single characters. We save
8319 * away these multicharacter foldings,
8320 * to be later saved as part of the
8321 * additional "s" data. */
8324 if (!unicode_alternate)
8325 unicode_alternate = newAV();
8326 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8328 av_push(unicode_alternate, sv);
8332 /* If folding and the value is one of the Greek
8333 * sigmas insert a few more sigmas to make the
8334 * folding rules of the sigmas to work right.
8335 * Note that not all the possible combinations
8336 * are handled here: some of them are handled
8337 * by the standard folding rules, and some of
8338 * them (literal or EXACTF cases) are handled
8339 * during runtime in regexec.c:S_find_byclass(). */
8340 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8341 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8342 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8343 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8344 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8346 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8347 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8348 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8353 literal_endpoint = 0;
8357 range = 0; /* this range (if it was one) is done now */
8361 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8363 RExC_size += ANYOF_CLASS_ADD_SKIP;
8365 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8371 /****** !SIZE_ONLY AFTER HERE *********/
8373 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8374 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8376 /* optimize single char class to an EXACT node
8377 but *only* when its not a UTF/high char */
8378 const char * cur_parse= RExC_parse;
8379 RExC_emit = (regnode *)orig_emit;
8380 RExC_parse = (char *)orig_parse;
8381 ret = reg_node(pRExC_state,
8382 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8383 RExC_parse = (char *)cur_parse;
8384 *STRING(ret)= (char)value;
8386 RExC_emit += STR_SZ(1);
8388 SvREFCNT_dec(listsv);
8392 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8393 if ( /* If the only flag is folding (plus possibly inversion). */
8394 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8396 for (value = 0; value < 256; ++value) {
8397 if (ANYOF_BITMAP_TEST(ret, value)) {
8398 UV fold = PL_fold[value];
8401 ANYOF_BITMAP_SET(ret, fold);
8404 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8407 /* optimize inverted simple patterns (e.g. [^a-z]) */
8408 if (optimize_invert &&
8409 /* If the only flag is inversion. */
8410 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8411 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8412 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8413 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8416 AV * const av = newAV();
8418 /* The 0th element stores the character class description
8419 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8420 * to initialize the appropriate swash (which gets stored in
8421 * the 1st element), and also useful for dumping the regnode.
8422 * The 2nd element stores the multicharacter foldings,
8423 * used later (regexec.c:S_reginclass()). */
8424 av_store(av, 0, listsv);
8425 av_store(av, 1, NULL);
8426 av_store(av, 2, MUTABLE_SV(unicode_alternate));
8427 rv = newRV_noinc(MUTABLE_SV(av));
8428 n = add_data(pRExC_state, 1, "s");
8429 RExC_rxi->data->data[n] = (void*)rv;
8437 /* reg_skipcomment()
8439 Absorbs an /x style # comments from the input stream.
8440 Returns true if there is more text remaining in the stream.
8441 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8442 terminates the pattern without including a newline.
8444 Note its the callers responsibility to ensure that we are
8450 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8454 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8456 while (RExC_parse < RExC_end)
8457 if (*RExC_parse++ == '\n') {
8462 /* we ran off the end of the pattern without ending
8463 the comment, so we have to add an \n when wrapping */
8464 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8472 Advance that parse position, and optionall absorbs
8473 "whitespace" from the inputstream.
8475 Without /x "whitespace" means (?#...) style comments only,
8476 with /x this means (?#...) and # comments and whitespace proper.
8478 Returns the RExC_parse point from BEFORE the scan occurs.
8480 This is the /x friendly way of saying RExC_parse++.
8484 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8486 char* const retval = RExC_parse++;
8488 PERL_ARGS_ASSERT_NEXTCHAR;
8491 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8492 RExC_parse[2] == '#') {
8493 while (*RExC_parse != ')') {
8494 if (RExC_parse == RExC_end)
8495 FAIL("Sequence (?#... not terminated");
8501 if (RExC_flags & RXf_PMf_EXTENDED) {
8502 if (isSPACE(*RExC_parse)) {
8506 else if (*RExC_parse == '#') {
8507 if ( reg_skipcomment( pRExC_state ) )
8516 - reg_node - emit a node
8518 STATIC regnode * /* Location. */
8519 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8522 register regnode *ptr;
8523 regnode * const ret = RExC_emit;
8524 GET_RE_DEBUG_FLAGS_DECL;
8526 PERL_ARGS_ASSERT_REG_NODE;
8529 SIZE_ALIGN(RExC_size);
8533 if (RExC_emit >= RExC_emit_bound)
8534 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8536 NODE_ALIGN_FILL(ret);
8538 FILL_ADVANCE_NODE(ptr, op);
8539 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 1);
8540 #ifdef RE_TRACK_PATTERN_OFFSETS
8541 if (RExC_offsets) { /* MJD */
8542 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8543 "reg_node", __LINE__,
8545 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8546 ? "Overwriting end of array!\n" : "OK",
8547 (UV)(RExC_emit - RExC_emit_start),
8548 (UV)(RExC_parse - RExC_start),
8549 (UV)RExC_offsets[0]));
8550 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8558 - reganode - emit a node with an argument
8560 STATIC regnode * /* Location. */
8561 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8564 register regnode *ptr;
8565 regnode * const ret = RExC_emit;
8566 GET_RE_DEBUG_FLAGS_DECL;
8568 PERL_ARGS_ASSERT_REGANODE;
8571 SIZE_ALIGN(RExC_size);
8576 assert(2==regarglen[op]+1);
8578 Anything larger than this has to allocate the extra amount.
8579 If we changed this to be:
8581 RExC_size += (1 + regarglen[op]);
8583 then it wouldn't matter. Its not clear what side effect
8584 might come from that so its not done so far.
8589 if (RExC_emit >= RExC_emit_bound)
8590 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8592 NODE_ALIGN_FILL(ret);
8594 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8595 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 2);
8596 #ifdef RE_TRACK_PATTERN_OFFSETS
8597 if (RExC_offsets) { /* MJD */
8598 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8602 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8603 "Overwriting end of array!\n" : "OK",
8604 (UV)(RExC_emit - RExC_emit_start),
8605 (UV)(RExC_parse - RExC_start),
8606 (UV)RExC_offsets[0]));
8607 Set_Cur_Node_Offset;
8615 - reguni - emit (if appropriate) a Unicode character
8618 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8622 PERL_ARGS_ASSERT_REGUNI;
8624 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8628 - reginsert - insert an operator in front of already-emitted operand
8630 * Means relocating the operand.
8633 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8636 register regnode *src;
8637 register regnode *dst;
8638 register regnode *place;
8639 const int offset = regarglen[(U8)op];
8640 const int size = NODE_STEP_REGNODE + offset;
8641 GET_RE_DEBUG_FLAGS_DECL;
8643 PERL_ARGS_ASSERT_REGINSERT;
8644 PERL_UNUSED_ARG(depth);
8645 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8646 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8655 if (RExC_open_parens) {
8657 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8658 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8659 if ( RExC_open_parens[paren] >= opnd ) {
8660 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8661 RExC_open_parens[paren] += size;
8663 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8665 if ( RExC_close_parens[paren] >= opnd ) {
8666 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8667 RExC_close_parens[paren] += size;
8669 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8674 while (src > opnd) {
8675 StructCopy(--src, --dst, regnode);
8676 #ifdef RE_TRACK_PATTERN_OFFSETS
8677 if (RExC_offsets) { /* MJD 20010112 */
8678 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8682 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8683 ? "Overwriting end of array!\n" : "OK",
8684 (UV)(src - RExC_emit_start),
8685 (UV)(dst - RExC_emit_start),
8686 (UV)RExC_offsets[0]));
8687 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8688 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8694 place = opnd; /* Op node, where operand used to be. */
8695 #ifdef RE_TRACK_PATTERN_OFFSETS
8696 if (RExC_offsets) { /* MJD */
8697 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8701 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8702 ? "Overwriting end of array!\n" : "OK",
8703 (UV)(place - RExC_emit_start),
8704 (UV)(RExC_parse - RExC_start),
8705 (UV)RExC_offsets[0]));
8706 Set_Node_Offset(place, RExC_parse);
8707 Set_Node_Length(place, 1);
8710 src = NEXTOPER(place);
8711 FILL_ADVANCE_NODE(place, op);
8712 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (place) - 1);
8713 Zero(src, offset, regnode);
8717 - regtail - set the next-pointer at the end of a node chain of p to val.
8718 - SEE ALSO: regtail_study
8720 /* TODO: All three parms should be const */
8722 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8725 register regnode *scan;
8726 GET_RE_DEBUG_FLAGS_DECL;
8728 PERL_ARGS_ASSERT_REGTAIL;
8730 PERL_UNUSED_ARG(depth);
8736 /* Find last node. */
8739 regnode * const temp = regnext(scan);
8741 SV * const mysv=sv_newmortal();
8742 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8743 regprop(RExC_rx, mysv, scan);
8744 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8745 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8746 (temp == NULL ? "->" : ""),
8747 (temp == NULL ? PL_reg_name[OP(val)] : "")
8755 if (reg_off_by_arg[OP(scan)]) {
8756 ARG_SET(scan, val - scan);
8759 NEXT_OFF(scan) = val - scan;
8765 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8766 - Look for optimizable sequences at the same time.
8767 - currently only looks for EXACT chains.
8769 This is expermental code. The idea is to use this routine to perform
8770 in place optimizations on branches and groups as they are constructed,
8771 with the long term intention of removing optimization from study_chunk so
8772 that it is purely analytical.
8774 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8775 to control which is which.
8778 /* TODO: All four parms should be const */
8781 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8784 register regnode *scan;
8786 #ifdef EXPERIMENTAL_INPLACESCAN
8789 GET_RE_DEBUG_FLAGS_DECL;
8791 PERL_ARGS_ASSERT_REGTAIL_STUDY;
8797 /* Find last node. */
8801 regnode * const temp = regnext(scan);
8802 #ifdef EXPERIMENTAL_INPLACESCAN
8803 if (PL_regkind[OP(scan)] == EXACT)
8804 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8812 if( exact == PSEUDO )
8814 else if ( exact != OP(scan) )
8823 SV * const mysv=sv_newmortal();
8824 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8825 regprop(RExC_rx, mysv, scan);
8826 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8827 SvPV_nolen_const(mysv),
8829 PL_reg_name[exact]);
8836 SV * const mysv_val=sv_newmortal();
8837 DEBUG_PARSE_MSG("");
8838 regprop(RExC_rx, mysv_val, val);
8839 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8840 SvPV_nolen_const(mysv_val),
8841 (IV)REG_NODE_NUM(val),
8845 if (reg_off_by_arg[OP(scan)]) {
8846 ARG_SET(scan, val - scan);
8849 NEXT_OFF(scan) = val - scan;
8857 - regcurly - a little FSA that accepts {\d+,?\d*}
8860 S_regcurly(register const char *s)
8862 PERL_ARGS_ASSERT_REGCURLY;
8881 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8885 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
8890 for (bit=0; bit<32; bit++) {
8891 if (flags & (1<<bit)) {
8893 PerlIO_printf(Perl_debug_log, "%s",lead);
8894 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
8899 PerlIO_printf(Perl_debug_log, "\n");
8901 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
8907 Perl_regdump(pTHX_ const regexp *r)
8911 SV * const sv = sv_newmortal();
8912 SV *dsv= sv_newmortal();
8914 GET_RE_DEBUG_FLAGS_DECL;
8916 PERL_ARGS_ASSERT_REGDUMP;
8918 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8920 /* Header fields of interest. */
8921 if (r->anchored_substr) {
8922 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8923 RE_SV_DUMPLEN(r->anchored_substr), 30);
8924 PerlIO_printf(Perl_debug_log,
8925 "anchored %s%s at %"IVdf" ",
8926 s, RE_SV_TAIL(r->anchored_substr),
8927 (IV)r->anchored_offset);
8928 } else if (r->anchored_utf8) {
8929 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8930 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8931 PerlIO_printf(Perl_debug_log,
8932 "anchored utf8 %s%s at %"IVdf" ",
8933 s, RE_SV_TAIL(r->anchored_utf8),
8934 (IV)r->anchored_offset);
8936 if (r->float_substr) {
8937 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8938 RE_SV_DUMPLEN(r->float_substr), 30);
8939 PerlIO_printf(Perl_debug_log,
8940 "floating %s%s at %"IVdf"..%"UVuf" ",
8941 s, RE_SV_TAIL(r->float_substr),
8942 (IV)r->float_min_offset, (UV)r->float_max_offset);
8943 } else if (r->float_utf8) {
8944 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8945 RE_SV_DUMPLEN(r->float_utf8), 30);
8946 PerlIO_printf(Perl_debug_log,
8947 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8948 s, RE_SV_TAIL(r->float_utf8),
8949 (IV)r->float_min_offset, (UV)r->float_max_offset);
8951 if (r->check_substr || r->check_utf8)
8952 PerlIO_printf(Perl_debug_log,
8954 (r->check_substr == r->float_substr
8955 && r->check_utf8 == r->float_utf8
8956 ? "(checking floating" : "(checking anchored"));
8957 if (r->extflags & RXf_NOSCAN)
8958 PerlIO_printf(Perl_debug_log, " noscan");
8959 if (r->extflags & RXf_CHECK_ALL)
8960 PerlIO_printf(Perl_debug_log, " isall");
8961 if (r->check_substr || r->check_utf8)
8962 PerlIO_printf(Perl_debug_log, ") ");
8964 if (ri->regstclass) {
8965 regprop(r, sv, ri->regstclass);
8966 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8968 if (r->extflags & RXf_ANCH) {
8969 PerlIO_printf(Perl_debug_log, "anchored");
8970 if (r->extflags & RXf_ANCH_BOL)
8971 PerlIO_printf(Perl_debug_log, "(BOL)");
8972 if (r->extflags & RXf_ANCH_MBOL)
8973 PerlIO_printf(Perl_debug_log, "(MBOL)");
8974 if (r->extflags & RXf_ANCH_SBOL)
8975 PerlIO_printf(Perl_debug_log, "(SBOL)");
8976 if (r->extflags & RXf_ANCH_GPOS)
8977 PerlIO_printf(Perl_debug_log, "(GPOS)");
8978 PerlIO_putc(Perl_debug_log, ' ');
8980 if (r->extflags & RXf_GPOS_SEEN)
8981 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8982 if (r->intflags & PREGf_SKIP)
8983 PerlIO_printf(Perl_debug_log, "plus ");
8984 if (r->intflags & PREGf_IMPLICIT)
8985 PerlIO_printf(Perl_debug_log, "implicit ");
8986 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8987 if (r->extflags & RXf_EVAL_SEEN)
8988 PerlIO_printf(Perl_debug_log, "with eval ");
8989 PerlIO_printf(Perl_debug_log, "\n");
8990 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
8992 PERL_ARGS_ASSERT_REGDUMP;
8993 PERL_UNUSED_CONTEXT;
8995 #endif /* DEBUGGING */
8999 - regprop - printable representation of opcode
9001 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9004 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9005 if (flags & ANYOF_INVERT) \
9006 /*make sure the invert info is in each */ \
9007 sv_catpvs(sv, "^"); \
9013 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9018 RXi_GET_DECL(prog,progi);
9019 GET_RE_DEBUG_FLAGS_DECL;
9021 PERL_ARGS_ASSERT_REGPROP;
9025 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9026 /* It would be nice to FAIL() here, but this may be called from
9027 regexec.c, and it would be hard to supply pRExC_state. */
9028 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9029 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9031 k = PL_regkind[OP(o)];
9035 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9036 * is a crude hack but it may be the best for now since
9037 * we have no flag "this EXACTish node was UTF-8"
9039 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9040 PERL_PV_ESCAPE_UNI_DETECT |
9041 PERL_PV_PRETTY_ELLIPSES |
9042 PERL_PV_PRETTY_LTGT |
9043 PERL_PV_PRETTY_NOCLEAR
9045 } else if (k == TRIE) {
9046 /* print the details of the trie in dumpuntil instead, as
9047 * progi->data isn't available here */
9048 const char op = OP(o);
9049 const U32 n = ARG(o);
9050 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9051 (reg_ac_data *)progi->data->data[n] :
9053 const reg_trie_data * const trie
9054 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9056 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9057 DEBUG_TRIE_COMPILE_r(
9058 Perl_sv_catpvf(aTHX_ sv,
9059 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9060 (UV)trie->startstate,
9061 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9062 (UV)trie->wordcount,
9065 (UV)TRIE_CHARCOUNT(trie),
9066 (UV)trie->uniquecharcount
9069 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9071 int rangestart = -1;
9072 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9074 for (i = 0; i <= 256; i++) {
9075 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9076 if (rangestart == -1)
9078 } else if (rangestart != -1) {
9079 if (i <= rangestart + 3)
9080 for (; rangestart < i; rangestart++)
9081 put_byte(sv, rangestart);
9083 put_byte(sv, rangestart);
9085 put_byte(sv, i - 1);
9093 } else if (k == CURLY) {
9094 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9095 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9096 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9098 else if (k == WHILEM && o->flags) /* Ordinal/of */
9099 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9100 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9101 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9102 if ( RXp_PAREN_NAMES(prog) ) {
9103 if ( k != REF || OP(o) < NREF) {
9104 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9105 SV **name= av_fetch(list, ARG(o), 0 );
9107 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9110 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9111 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9112 I32 *nums=(I32*)SvPVX(sv_dat);
9113 SV **name= av_fetch(list, nums[0], 0 );
9116 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9117 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9118 (n ? "," : ""), (IV)nums[n]);
9120 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9124 } else if (k == GOSUB)
9125 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9126 else if (k == VERB) {
9128 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9129 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9130 } else if (k == LOGICAL)
9131 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9132 else if (k == FOLDCHAR)
9133 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9134 else if (k == ANYOF) {
9135 int i, rangestart = -1;
9136 const U8 flags = ANYOF_FLAGS(o);
9139 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9140 static const char * const anyofs[] = {
9173 if (flags & ANYOF_LOCALE)
9174 sv_catpvs(sv, "{loc}");
9175 if (flags & ANYOF_FOLD)
9176 sv_catpvs(sv, "{i}");
9177 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9178 if (flags & ANYOF_INVERT)
9181 /* output what the standard cp 0-255 bitmap matches */
9182 for (i = 0; i <= 256; i++) {
9183 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9184 if (rangestart == -1)
9186 } else if (rangestart != -1) {
9187 if (i <= rangestart + 3)
9188 for (; rangestart < i; rangestart++)
9189 put_byte(sv, rangestart);
9191 put_byte(sv, rangestart);
9193 put_byte(sv, i - 1);
9200 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9201 /* output any special charclass tests (used mostly under use locale) */
9202 if (o->flags & ANYOF_CLASS)
9203 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9204 if (ANYOF_CLASS_TEST(o,i)) {
9205 sv_catpv(sv, anyofs[i]);
9209 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9211 /* output information about the unicode matching */
9212 if (flags & ANYOF_UNICODE)
9213 sv_catpvs(sv, "{unicode}");
9214 else if (flags & ANYOF_UNICODE_ALL)
9215 sv_catpvs(sv, "{unicode_all}");
9219 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9223 U8 s[UTF8_MAXBYTES_CASE+1];
9225 for (i = 0; i <= 256; i++) { /* just the first 256 */
9226 uvchr_to_utf8(s, i);
9228 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9229 if (rangestart == -1)
9231 } else if (rangestart != -1) {
9232 if (i <= rangestart + 3)
9233 for (; rangestart < i; rangestart++) {
9234 const U8 * const e = uvchr_to_utf8(s,rangestart);
9236 for(p = s; p < e; p++)
9240 const U8 *e = uvchr_to_utf8(s,rangestart);
9242 for (p = s; p < e; p++)
9245 e = uvchr_to_utf8(s, i-1);
9246 for (p = s; p < e; p++)
9253 sv_catpvs(sv, "..."); /* et cetera */
9257 char *s = savesvpv(lv);
9258 char * const origs = s;
9260 while (*s && *s != '\n')
9264 const char * const t = ++s;
9282 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9284 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9285 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9287 PERL_UNUSED_CONTEXT;
9288 PERL_UNUSED_ARG(sv);
9290 PERL_UNUSED_ARG(prog);
9291 #endif /* DEBUGGING */
9295 Perl_re_intuit_string(pTHX_ REGEXP * const prog)
9296 { /* Assume that RE_INTUIT is set */
9298 GET_RE_DEBUG_FLAGS_DECL;
9300 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9301 PERL_UNUSED_CONTEXT;
9305 const char * const s = SvPV_nolen_const(prog->check_substr
9306 ? prog->check_substr : prog->check_utf8);
9308 if (!PL_colorset) reginitcolors();
9309 PerlIO_printf(Perl_debug_log,
9310 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9312 prog->check_substr ? "" : "utf8 ",
9313 PL_colors[5],PL_colors[0],
9316 (strlen(s) > 60 ? "..." : ""));
9319 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9325 handles refcounting and freeing the perl core regexp structure. When
9326 it is necessary to actually free the structure the first thing it
9327 does is call the 'free' method of the regexp_engine associated to to
9328 the regexp, allowing the handling of the void *pprivate; member
9329 first. (This routine is not overridable by extensions, which is why
9330 the extensions free is called first.)
9332 See regdupe and regdupe_internal if you change anything here.
9334 #ifndef PERL_IN_XSUB_RE
9336 Perl_pregfree(pTHX_ REGEXP *r)
9339 GET_RE_DEBUG_FLAGS_DECL;
9341 if (!r || (--r->refcnt > 0))
9344 ReREFCNT_dec(r->mother_re);
9346 CALLREGFREE_PVT(r); /* free the private data */
9347 if (RXp_PAREN_NAMES(r))
9348 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9349 Safefree(RX_WRAPPED(r));
9352 if (r->anchored_substr)
9353 SvREFCNT_dec(r->anchored_substr);
9354 if (r->anchored_utf8)
9355 SvREFCNT_dec(r->anchored_utf8);
9356 if (r->float_substr)
9357 SvREFCNT_dec(r->float_substr);
9359 SvREFCNT_dec(r->float_utf8);
9360 Safefree(r->substrs);
9362 RX_MATCH_COPY_FREE(r);
9363 #ifdef PERL_OLD_COPY_ON_WRITE
9365 SvREFCNT_dec(r->saved_copy);
9374 This is a hacky workaround to the structural issue of match results
9375 being stored in the regexp structure which is in turn stored in
9376 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9377 could be PL_curpm in multiple contexts, and could require multiple
9378 result sets being associated with the pattern simultaneously, such
9379 as when doing a recursive match with (??{$qr})
9381 The solution is to make a lightweight copy of the regexp structure
9382 when a qr// is returned from the code executed by (??{$qr}) this
9383 lightweight copy doesnt actually own any of its data except for
9384 the starp/end and the actual regexp structure itself.
9390 Perl_reg_temp_copy (pTHX_ REGEXP *r) {
9392 register const I32 npar = r->nparens+1;
9394 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9396 (void)ReREFCNT_inc(r);
9397 Newx(ret, 1, regexp);
9398 StructCopy(r, ret, regexp);
9399 Newx(ret->offs, npar, regexp_paren_pair);
9400 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9403 Newx(ret->substrs, 1, struct reg_substr_data);
9404 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9406 SvREFCNT_inc_void(ret->anchored_substr);
9407 SvREFCNT_inc_void(ret->anchored_utf8);
9408 SvREFCNT_inc_void(ret->float_substr);
9409 SvREFCNT_inc_void(ret->float_utf8);
9411 /* check_substr and check_utf8, if non-NULL, point to either their
9412 anchored or float namesakes, and don't hold a second reference. */
9414 RX_MATCH_COPIED_off(ret);
9415 #ifdef PERL_OLD_COPY_ON_WRITE
9416 ret->saved_copy = NULL;
9425 /* regfree_internal()
9427 Free the private data in a regexp. This is overloadable by
9428 extensions. Perl takes care of the regexp structure in pregfree(),
9429 this covers the *pprivate pointer which technically perldoesnt
9430 know about, however of course we have to handle the
9431 regexp_internal structure when no extension is in use.
9433 Note this is called before freeing anything in the regexp
9438 Perl_regfree_internal(pTHX_ REGEXP * const r)
9442 GET_RE_DEBUG_FLAGS_DECL;
9444 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9450 SV *dsv= sv_newmortal();
9451 RE_PV_QUOTED_DECL(s, RX_UTF8(r),
9452 dsv, RX_PRECOMP(r), RX_PRELEN(r), 60);
9453 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9454 PL_colors[4],PL_colors[5],s);
9457 #ifdef RE_TRACK_PATTERN_OFFSETS
9459 Safefree(ri->u.offsets); /* 20010421 MJD */
9462 int n = ri->data->count;
9463 PAD* new_comppad = NULL;
9468 /* If you add a ->what type here, update the comment in regcomp.h */
9469 switch (ri->data->what[n]) {
9473 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
9476 Safefree(ri->data->data[n]);
9479 new_comppad = MUTABLE_AV(ri->data->data[n]);
9482 if (new_comppad == NULL)
9483 Perl_croak(aTHX_ "panic: pregfree comppad");
9484 PAD_SAVE_LOCAL(old_comppad,
9485 /* Watch out for global destruction's random ordering. */
9486 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9489 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9492 op_free((OP_4tree*)ri->data->data[n]);
9494 PAD_RESTORE_LOCAL(old_comppad);
9495 SvREFCNT_dec(MUTABLE_SV(new_comppad));
9501 { /* Aho Corasick add-on structure for a trie node.
9502 Used in stclass optimization only */
9504 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9506 refcount = --aho->refcount;
9509 PerlMemShared_free(aho->states);
9510 PerlMemShared_free(aho->fail);
9511 /* do this last!!!! */
9512 PerlMemShared_free(ri->data->data[n]);
9513 PerlMemShared_free(ri->regstclass);
9519 /* trie structure. */
9521 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9523 refcount = --trie->refcount;
9526 PerlMemShared_free(trie->charmap);
9527 PerlMemShared_free(trie->states);
9528 PerlMemShared_free(trie->trans);
9530 PerlMemShared_free(trie->bitmap);
9532 PerlMemShared_free(trie->wordlen);
9534 PerlMemShared_free(trie->jump);
9536 PerlMemShared_free(trie->nextword);
9537 /* do this last!!!! */
9538 PerlMemShared_free(ri->data->data[n]);
9543 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9546 Safefree(ri->data->what);
9553 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
9554 #define av_dup_inc(s,t) MUTABLE_AV(SvREFCNT_inc(sv_dup((const SV *)s,t)))
9555 #define hv_dup_inc(s,t) MUTABLE_HV(SvREFCNT_inc(sv_dup((const SV *)s,t)))
9556 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9559 re_dup - duplicate a regexp.
9561 This routine is expected to clone a given regexp structure. It is only
9562 compiled under USE_ITHREADS.
9564 After all of the core data stored in struct regexp is duplicated
9565 the regexp_engine.dupe method is used to copy any private data
9566 stored in the *pprivate pointer. This allows extensions to handle
9567 any duplication it needs to do.
9569 See pregfree() and regfree_internal() if you change anything here.
9571 #if defined(USE_ITHREADS)
9572 #ifndef PERL_IN_XSUB_RE
9574 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
9581 PERL_ARGS_ASSERT_RE_DUP;
9584 return (REGEXP *)NULL;
9586 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
9589 npar = r->nparens+1;
9590 Newx(ret, 1, regexp);
9591 StructCopy(r, ret, regexp);
9592 Newx(ret->offs, npar, regexp_paren_pair);
9593 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9595 /* no need to copy these */
9596 Newx(ret->swap, npar, regexp_paren_pair);
9600 /* Do it this way to avoid reading from *r after the StructCopy().
9601 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9602 cache, it doesn't matter. */
9603 const bool anchored = r->check_substr
9604 ? r->check_substr == r->anchored_substr
9605 : r->check_utf8 == r->anchored_utf8;
9606 Newx(ret->substrs, 1, struct reg_substr_data);
9607 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9609 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9610 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9611 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9612 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9614 /* check_substr and check_utf8, if non-NULL, point to either their
9615 anchored or float namesakes, and don't hold a second reference. */
9617 if (ret->check_substr) {
9619 assert(r->check_utf8 == r->anchored_utf8);
9620 ret->check_substr = ret->anchored_substr;
9621 ret->check_utf8 = ret->anchored_utf8;
9623 assert(r->check_substr == r->float_substr);
9624 assert(r->check_utf8 == r->float_utf8);
9625 ret->check_substr = ret->float_substr;
9626 ret->check_utf8 = ret->float_utf8;
9628 } else if (ret->check_utf8) {
9630 ret->check_utf8 = ret->anchored_utf8;
9632 ret->check_utf8 = ret->float_utf8;
9637 precomp_offset = RX_PRECOMP(ret) - ret->wrapped;
9639 RX_WRAPPED(ret) = SAVEPVN(RX_WRAPPED(ret), RX_WRAPLEN(ret)+1);
9640 RX_PRECOMP(ret) = ret->wrapped + precomp_offset;
9641 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9644 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
9646 if (RX_MATCH_COPIED(ret))
9647 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9650 #ifdef PERL_OLD_COPY_ON_WRITE
9651 ret->saved_copy = NULL;
9654 ret->mother_re = NULL;
9657 ptr_table_store(PL_ptr_table, r, ret);
9660 #endif /* PERL_IN_XSUB_RE */
9665 This is the internal complement to regdupe() which is used to copy
9666 the structure pointed to by the *pprivate pointer in the regexp.
9667 This is the core version of the extension overridable cloning hook.
9668 The regexp structure being duplicated will be copied by perl prior
9669 to this and will be provided as the regexp *r argument, however
9670 with the /old/ structures pprivate pointer value. Thus this routine
9671 may override any copying normally done by perl.
9673 It returns a pointer to the new regexp_internal structure.
9677 Perl_regdupe_internal(pTHX_ REGEXP * const r, CLONE_PARAMS *param)
9680 regexp_internal *reti;
9684 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
9686 npar = r->nparens+1;
9689 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
9690 Copy(ri->program, reti->program, len+1, regnode);
9693 reti->regstclass = NULL;
9697 const int count = ri->data->count;
9700 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9701 char, struct reg_data);
9702 Newx(d->what, count, U8);
9705 for (i = 0; i < count; i++) {
9706 d->what[i] = ri->data->what[i];
9707 switch (d->what[i]) {
9708 /* legal options are one of: sSfpontTu
9709 see also regcomp.h and pregfree() */
9712 case 'p': /* actually an AV, but the dup function is identical. */
9713 case 'u': /* actually an HV, but the dup function is identical. */
9714 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
9717 /* This is cheating. */
9718 Newx(d->data[i], 1, struct regnode_charclass_class);
9719 StructCopy(ri->data->data[i], d->data[i],
9720 struct regnode_charclass_class);
9721 reti->regstclass = (regnode*)d->data[i];
9724 /* Compiled op trees are readonly and in shared memory,
9725 and can thus be shared without duplication. */
9727 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9731 /* Trie stclasses are readonly and can thus be shared
9732 * without duplication. We free the stclass in pregfree
9733 * when the corresponding reg_ac_data struct is freed.
9735 reti->regstclass= ri->regstclass;
9739 ((reg_trie_data*)ri->data->data[i])->refcount++;
9743 d->data[i] = ri->data->data[i];
9746 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9755 reti->name_list_idx = ri->name_list_idx;
9757 #ifdef RE_TRACK_PATTERN_OFFSETS
9758 if (ri->u.offsets) {
9759 Newx(reti->u.offsets, 2*len+1, U32);
9760 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9763 SetProgLen(reti,len);
9769 #endif /* USE_ITHREADS */
9774 converts a regexp embedded in a MAGIC struct to its stringified form,
9775 caching the converted form in the struct and returns the cached
9778 If lp is nonnull then it is used to return the length of the
9781 If flags is nonnull and the returned string contains UTF8 then
9782 (*flags & 1) will be true.
9784 If haseval is nonnull then it is used to return whether the pattern
9787 Normally called via macro:
9789 CALLREG_STRINGIFY(mg,&len,&utf8);
9793 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9795 See sv_2pv_flags() in sv.c for an example of internal usage.
9798 #ifndef PERL_IN_XSUB_RE
9801 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9803 const REGEXP * const re = (REGEXP *)mg->mg_obj;
9805 *haseval = RX_SEEN_EVALS(re);
9807 *flags = RX_UTF8(re) ? 1 : 0;
9809 *lp = RX_WRAPLEN(re);
9810 return RX_WRAPPED(re);
9814 - regnext - dig the "next" pointer out of a node
9817 Perl_regnext(pTHX_ register regnode *p)
9820 register I32 offset;
9825 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9834 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9837 STRLEN l1 = strlen(pat1);
9838 STRLEN l2 = strlen(pat2);
9841 const char *message;
9843 PERL_ARGS_ASSERT_RE_CROAK2;
9849 Copy(pat1, buf, l1 , char);
9850 Copy(pat2, buf + l1, l2 , char);
9851 buf[l1 + l2] = '\n';
9852 buf[l1 + l2 + 1] = '\0';
9854 /* ANSI variant takes additional second argument */
9855 va_start(args, pat2);
9859 msv = vmess(buf, &args);
9861 message = SvPV_const(msv,l1);
9864 Copy(message, buf, l1 , char);
9865 buf[l1-1] = '\0'; /* Overwrite \n */
9866 Perl_croak(aTHX_ "%s", buf);
9869 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9871 #ifndef PERL_IN_XSUB_RE
9873 Perl_save_re_context(pTHX)
9877 struct re_save_state *state;
9879 SAVEVPTR(PL_curcop);
9880 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9882 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9883 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9884 SSPUSHINT(SAVEt_RE_STATE);
9886 Copy(&PL_reg_state, state, 1, struct re_save_state);
9888 PL_reg_start_tmp = 0;
9889 PL_reg_start_tmpl = 0;
9890 PL_reg_oldsaved = NULL;
9891 PL_reg_oldsavedlen = 0;
9893 PL_reg_leftiter = 0;
9894 PL_reg_poscache = NULL;
9895 PL_reg_poscache_size = 0;
9896 #ifdef PERL_OLD_COPY_ON_WRITE
9900 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9902 const REGEXP * const rx = PM_GETRE(PL_curpm);
9905 for (i = 1; i <= RX_NPARENS(rx); i++) {
9906 char digits[TYPE_CHARS(long)];
9907 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9908 GV *const *const gvp
9909 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9912 GV * const gv = *gvp;
9913 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9923 clear_re(pTHX_ void *r)
9926 ReREFCNT_dec((REGEXP *)r);
9932 S_put_byte(pTHX_ SV *sv, int c)
9934 PERL_ARGS_ASSERT_PUT_BYTE;
9936 /* Our definition of isPRINT() ignores locales, so only bytes that are
9937 not part of UTF-8 are considered printable. I assume that the same
9938 holds for UTF-EBCDIC.
9939 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
9940 which Wikipedia says:
9942 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
9943 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
9944 identical, to the ASCII delete (DEL) or rubout control character.
9945 ) So the old condition can be simplified to !isPRINT(c) */
9947 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9949 const char string = c;
9950 if (c == '-' || c == ']' || c == '\\' || c == '^')
9951 sv_catpvs(sv, "\\");
9952 sv_catpvn(sv, &string, 1);
9957 #define CLEAR_OPTSTART \
9958 if (optstart) STMT_START { \
9959 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9963 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9965 STATIC const regnode *
9966 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9967 const regnode *last, const regnode *plast,
9968 SV* sv, I32 indent, U32 depth)
9971 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9972 register const regnode *next;
9973 const regnode *optstart= NULL;
9976 GET_RE_DEBUG_FLAGS_DECL;
9978 PERL_ARGS_ASSERT_DUMPUNTIL;
9980 #ifdef DEBUG_DUMPUNTIL
9981 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9982 last ? last-start : 0,plast ? plast-start : 0);
9985 if (plast && plast < last)
9988 while (PL_regkind[op] != END && (!last || node < last)) {
9989 /* While that wasn't END last time... */
9992 if (op == CLOSE || op == WHILEM)
9994 next = regnext((regnode *)node);
9997 if (OP(node) == OPTIMIZED) {
9998 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10005 regprop(r, sv, node);
10006 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10007 (int)(2*indent + 1), "", SvPVX_const(sv));
10009 if (OP(node) != OPTIMIZED) {
10010 if (next == NULL) /* Next ptr. */
10011 PerlIO_printf(Perl_debug_log, " (0)");
10012 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10013 PerlIO_printf(Perl_debug_log, " (FAIL)");
10015 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10016 (void)PerlIO_putc(Perl_debug_log, '\n');
10020 if (PL_regkind[(U8)op] == BRANCHJ) {
10023 register const regnode *nnode = (OP(next) == LONGJMP
10024 ? regnext((regnode *)next)
10026 if (last && nnode > last)
10028 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10031 else if (PL_regkind[(U8)op] == BRANCH) {
10033 DUMPUNTIL(NEXTOPER(node), next);
10035 else if ( PL_regkind[(U8)op] == TRIE ) {
10036 const regnode *this_trie = node;
10037 const char op = OP(node);
10038 const U32 n = ARG(node);
10039 const reg_ac_data * const ac = op>=AHOCORASICK ?
10040 (reg_ac_data *)ri->data->data[n] :
10042 const reg_trie_data * const trie =
10043 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10045 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10047 const regnode *nextbranch= NULL;
10050 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10051 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10053 PerlIO_printf(Perl_debug_log, "%*s%s ",
10054 (int)(2*(indent+3)), "",
10055 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10056 PL_colors[0], PL_colors[1],
10057 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10058 PERL_PV_PRETTY_ELLIPSES |
10059 PERL_PV_PRETTY_LTGT
10064 U16 dist= trie->jump[word_idx+1];
10065 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10066 (UV)((dist ? this_trie + dist : next) - start));
10069 nextbranch= this_trie + trie->jump[0];
10070 DUMPUNTIL(this_trie + dist, nextbranch);
10072 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10073 nextbranch= regnext((regnode *)nextbranch);
10075 PerlIO_printf(Perl_debug_log, "\n");
10078 if (last && next > last)
10083 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10084 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10085 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10087 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10089 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10091 else if ( op == PLUS || op == STAR) {
10092 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10094 else if (op == ANYOF) {
10095 /* arglen 1 + class block */
10096 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
10097 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10098 node = NEXTOPER(node);
10100 else if (PL_regkind[(U8)op] == EXACT) {
10101 /* Literal string, where present. */
10102 node += NODE_SZ_STR(node) - 1;
10103 node = NEXTOPER(node);
10106 node = NEXTOPER(node);
10107 node += regarglen[(U8)op];
10109 if (op == CURLYX || op == OPEN)
10113 #ifdef DEBUG_DUMPUNTIL
10114 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10119 #endif /* DEBUGGING */
10123 * c-indentation-style: bsd
10124 * c-basic-offset: 4
10125 * indent-tabs-mode: t
10128 * ex: set ts=8 sts=4 sw=4 noet:
projects / perl / modules / re-engine-Hooks.git / blob