5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
8 /* This file contains functions for compiling a regular expression. See
9 * also regexec.c which funnily enough, contains functions for executing
10 * a regular expression.
12 * This file is also copied at build time to ext/re/re_comp.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
18 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
22 /* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
27 /* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
32 #ifdef PERL_EXT_RE_BUILD
37 * pregcomp and pregexec -- regsub and regerror are not used in perl
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
57 **** Alterations to Henry's code are...
59 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
60 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 by Larry Wall and others
62 **** You may distribute under the terms of either the GNU General Public
63 **** License or the Artistic License, as specified in the README file.
66 * Beware that some of this code is subtly aware of the way operator
67 * precedence is structured in regular expressions. Serious changes in
68 * regular-expression syntax might require a total rethink.
71 #define PERL_IN_REGCOMP_C
74 #ifndef PERL_IN_XSUB_RE
79 #ifdef PERL_IN_XSUB_RE
90 # if defined(BUGGY_MSC6)
91 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
92 # pragma optimize("a",off)
93 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
94 # pragma optimize("w",on )
95 # endif /* BUGGY_MSC6 */
102 typedef struct RExC_state_t {
103 U32 flags; /* are we folding, multilining? */
104 char *precomp; /* uncompiled string. */
105 regexp *rx; /* perl core regexp structure */
106 regexp_internal *rxi; /* internal data for regexp object pprivate field */
107 char *start; /* Start of input for compile */
108 char *end; /* End of input for compile */
109 char *parse; /* Input-scan pointer. */
110 I32 whilem_seen; /* number of WHILEM in this expr */
111 regnode *emit_start; /* Start of emitted-code area */
112 regnode *emit_bound; /* First regnode outside of the allocated space */
113 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
114 I32 naughty; /* How bad is this pattern? */
115 I32 sawback; /* Did we see \1, ...? */
117 I32 size; /* Code size. */
118 I32 npar; /* Capture buffer count, (OPEN). */
119 I32 cpar; /* Capture buffer count, (CLOSE). */
120 I32 nestroot; /* root parens we are in - used by accept */
124 regnode **open_parens; /* pointers to open parens */
125 regnode **close_parens; /* pointers to close parens */
126 regnode *opend; /* END node in program */
127 I32 utf8; /* whether the pattern is utf8 or not */
128 I32 orig_utf8; /* whether the pattern was originally in utf8 */
129 /* XXX use this for future optimisation of case
130 * where pattern must be upgraded to utf8. */
131 HV *charnames; /* cache of named sequences */
132 HV *paren_names; /* Paren names */
134 regnode **recurse; /* Recurse regops */
135 I32 recurse_count; /* Number of recurse regops */
137 char *starttry; /* -Dr: where regtry was called. */
138 #define RExC_starttry (pRExC_state->starttry)
141 const char *lastparse;
143 AV *paren_name_list; /* idx -> name */
144 #define RExC_lastparse (pRExC_state->lastparse)
145 #define RExC_lastnum (pRExC_state->lastnum)
146 #define RExC_paren_name_list (pRExC_state->paren_name_list)
150 #define RExC_flags (pRExC_state->flags)
151 #define RExC_precomp (pRExC_state->precomp)
152 #define RExC_rx (pRExC_state->rx)
153 #define RExC_rxi (pRExC_state->rxi)
154 #define RExC_start (pRExC_state->start)
155 #define RExC_end (pRExC_state->end)
156 #define RExC_parse (pRExC_state->parse)
157 #define RExC_whilem_seen (pRExC_state->whilem_seen)
158 #ifdef RE_TRACK_PATTERN_OFFSETS
159 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
161 #define RExC_emit (pRExC_state->emit)
162 #define RExC_emit_start (pRExC_state->emit_start)
163 #define RExC_emit_bound (pRExC_state->emit_bound)
164 #define RExC_naughty (pRExC_state->naughty)
165 #define RExC_sawback (pRExC_state->sawback)
166 #define RExC_seen (pRExC_state->seen)
167 #define RExC_size (pRExC_state->size)
168 #define RExC_npar (pRExC_state->npar)
169 #define RExC_nestroot (pRExC_state->nestroot)
170 #define RExC_extralen (pRExC_state->extralen)
171 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
172 #define RExC_seen_evals (pRExC_state->seen_evals)
173 #define RExC_utf8 (pRExC_state->utf8)
174 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
175 #define RExC_charnames (pRExC_state->charnames)
176 #define RExC_open_parens (pRExC_state->open_parens)
177 #define RExC_close_parens (pRExC_state->close_parens)
178 #define RExC_opend (pRExC_state->opend)
179 #define RExC_paren_names (pRExC_state->paren_names)
180 #define RExC_recurse (pRExC_state->recurse)
181 #define RExC_recurse_count (pRExC_state->recurse_count)
184 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
185 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
186 ((*s) == '{' && regcurly(s)))
189 #undef SPSTART /* dratted cpp namespace... */
192 * Flags to be passed up and down.
194 #define WORST 0 /* Worst case. */
195 #define HASWIDTH 0x01 /* Known to match non-null strings. */
196 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
197 #define SPSTART 0x04 /* Starts with * or +. */
198 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
199 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
201 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
203 /* whether trie related optimizations are enabled */
204 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
205 #define TRIE_STUDY_OPT
206 #define FULL_TRIE_STUDY
212 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
213 #define PBITVAL(paren) (1 << ((paren) & 7))
214 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
215 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
216 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
219 /* About scan_data_t.
221 During optimisation we recurse through the regexp program performing
222 various inplace (keyhole style) optimisations. In addition study_chunk
223 and scan_commit populate this data structure with information about
224 what strings MUST appear in the pattern. We look for the longest
225 string that must appear for at a fixed location, and we look for the
226 longest string that may appear at a floating location. So for instance
231 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
232 strings (because they follow a .* construct). study_chunk will identify
233 both FOO and BAR as being the longest fixed and floating strings respectively.
235 The strings can be composites, for instance
239 will result in a composite fixed substring 'foo'.
241 For each string some basic information is maintained:
243 - offset or min_offset
244 This is the position the string must appear at, or not before.
245 It also implicitly (when combined with minlenp) tells us how many
246 character must match before the string we are searching.
247 Likewise when combined with minlenp and the length of the string
248 tells us how many characters must appear after the string we have
252 Only used for floating strings. This is the rightmost point that
253 the string can appear at. Ifset to I32 max it indicates that the
254 string can occur infinitely far to the right.
257 A pointer to the minimum length of the pattern that the string
258 was found inside. This is important as in the case of positive
259 lookahead or positive lookbehind we can have multiple patterns
264 The minimum length of the pattern overall is 3, the minimum length
265 of the lookahead part is 3, but the minimum length of the part that
266 will actually match is 1. So 'FOO's minimum length is 3, but the
267 minimum length for the F is 1. This is important as the minimum length
268 is used to determine offsets in front of and behind the string being
269 looked for. Since strings can be composites this is the length of the
270 pattern at the time it was commited with a scan_commit. Note that
271 the length is calculated by study_chunk, so that the minimum lengths
272 are not known until the full pattern has been compiled, thus the
273 pointer to the value.
277 In the case of lookbehind the string being searched for can be
278 offset past the start point of the final matching string.
279 If this value was just blithely removed from the min_offset it would
280 invalidate some of the calculations for how many chars must match
281 before or after (as they are derived from min_offset and minlen and
282 the length of the string being searched for).
283 When the final pattern is compiled and the data is moved from the
284 scan_data_t structure into the regexp structure the information
285 about lookbehind is factored in, with the information that would
286 have been lost precalculated in the end_shift field for the
289 The fields pos_min and pos_delta are used to store the minimum offset
290 and the delta to the maximum offset at the current point in the pattern.
294 typedef struct scan_data_t {
295 /*I32 len_min; unused */
296 /*I32 len_delta; unused */
300 I32 last_end; /* min value, <0 unless valid. */
303 SV **longest; /* Either &l_fixed, or &l_float. */
304 SV *longest_fixed; /* longest fixed string found in pattern */
305 I32 offset_fixed; /* offset where it starts */
306 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
307 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
308 SV *longest_float; /* longest floating string found in pattern */
309 I32 offset_float_min; /* earliest point in string it can appear */
310 I32 offset_float_max; /* latest point in string it can appear */
311 I32 *minlen_float; /* pointer to the minlen relevent to the string */
312 I32 lookbehind_float; /* is the position of the string modified by LB */
316 struct regnode_charclass_class *start_class;
320 * Forward declarations for pregcomp()'s friends.
323 static const scan_data_t zero_scan_data =
324 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
326 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
327 #define SF_BEFORE_SEOL 0x0001
328 #define SF_BEFORE_MEOL 0x0002
329 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
330 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
333 # define SF_FIX_SHIFT_EOL (0+2)
334 # define SF_FL_SHIFT_EOL (0+4)
336 # define SF_FIX_SHIFT_EOL (+2)
337 # define SF_FL_SHIFT_EOL (+4)
340 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
341 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
343 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
344 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
345 #define SF_IS_INF 0x0040
346 #define SF_HAS_PAR 0x0080
347 #define SF_IN_PAR 0x0100
348 #define SF_HAS_EVAL 0x0200
349 #define SCF_DO_SUBSTR 0x0400
350 #define SCF_DO_STCLASS_AND 0x0800
351 #define SCF_DO_STCLASS_OR 0x1000
352 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
353 #define SCF_WHILEM_VISITED_POS 0x2000
355 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
356 #define SCF_SEEN_ACCEPT 0x8000
358 #define UTF (RExC_utf8 != 0)
359 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
360 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
362 #define OOB_UNICODE 12345678
363 #define OOB_NAMEDCLASS -1
365 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
366 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
369 /* length of regex to show in messages that don't mark a position within */
370 #define RegexLengthToShowInErrorMessages 127
373 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
374 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
375 * op/pragma/warn/regcomp.
377 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
378 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
380 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
383 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
384 * arg. Show regex, up to a maximum length. If it's too long, chop and add
387 #define _FAIL(code) STMT_START { \
388 const char *ellipses = ""; \
389 IV len = RExC_end - RExC_precomp; \
392 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
393 if (len > RegexLengthToShowInErrorMessages) { \
394 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
395 len = RegexLengthToShowInErrorMessages - 10; \
401 #define FAIL(msg) _FAIL( \
402 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
403 msg, (int)len, RExC_precomp, ellipses))
405 #define FAIL2(msg,arg) _FAIL( \
406 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
407 arg, (int)len, RExC_precomp, ellipses))
410 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
412 #define Simple_vFAIL(m) STMT_START { \
413 const IV offset = RExC_parse - RExC_precomp; \
414 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
415 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
419 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
421 #define vFAIL(m) STMT_START { \
423 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
428 * Like Simple_vFAIL(), but accepts two arguments.
430 #define Simple_vFAIL2(m,a1) STMT_START { \
431 const IV offset = RExC_parse - RExC_precomp; \
432 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
433 (int)offset, RExC_precomp, RExC_precomp + offset); \
437 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
439 #define vFAIL2(m,a1) STMT_START { \
441 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
442 Simple_vFAIL2(m, a1); \
447 * Like Simple_vFAIL(), but accepts three arguments.
449 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
450 const IV offset = RExC_parse - RExC_precomp; \
451 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
452 (int)offset, RExC_precomp, RExC_precomp + offset); \
456 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
458 #define vFAIL3(m,a1,a2) STMT_START { \
460 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
461 Simple_vFAIL3(m, a1, a2); \
465 * Like Simple_vFAIL(), but accepts four arguments.
467 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
468 const IV offset = RExC_parse - RExC_precomp; \
469 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
470 (int)offset, RExC_precomp, RExC_precomp + offset); \
473 #define vWARN(loc,m) STMT_START { \
474 const IV offset = loc - RExC_precomp; \
475 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
476 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
479 #define vWARNdep(loc,m) STMT_START { \
480 const IV offset = loc - RExC_precomp; \
481 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
482 "%s" REPORT_LOCATION, \
483 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
487 #define vWARN2(loc, m, a1) STMT_START { \
488 const IV offset = loc - RExC_precomp; \
489 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
493 #define vWARN3(loc, m, a1, a2) STMT_START { \
494 const IV offset = loc - RExC_precomp; \
495 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
496 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
499 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
500 const IV offset = loc - RExC_precomp; \
501 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
502 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
505 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
506 const IV offset = loc - RExC_precomp; \
507 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
508 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
512 /* Allow for side effects in s */
513 #define REGC(c,s) STMT_START { \
514 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
517 /* Macros for recording node offsets. 20001227 mjd@plover.com
518 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
519 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
520 * Element 0 holds the number n.
521 * Position is 1 indexed.
523 #ifndef RE_TRACK_PATTERN_OFFSETS
524 #define Set_Node_Offset_To_R(node,byte)
525 #define Set_Node_Offset(node,byte)
526 #define Set_Cur_Node_Offset
527 #define Set_Node_Length_To_R(node,len)
528 #define Set_Node_Length(node,len)
529 #define Set_Node_Cur_Length(node)
530 #define Node_Offset(n)
531 #define Node_Length(n)
532 #define Set_Node_Offset_Length(node,offset,len)
533 #define ProgLen(ri) ri->u.proglen
534 #define SetProgLen(ri,x) ri->u.proglen = x
536 #define ProgLen(ri) ri->u.offsets[0]
537 #define SetProgLen(ri,x) ri->u.offsets[0] = x
538 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
540 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
541 __LINE__, (int)(node), (int)(byte))); \
543 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
545 RExC_offsets[2*(node)-1] = (byte); \
550 #define Set_Node_Offset(node,byte) \
551 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
552 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
554 #define Set_Node_Length_To_R(node,len) STMT_START { \
556 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
557 __LINE__, (int)(node), (int)(len))); \
559 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
561 RExC_offsets[2*(node)] = (len); \
566 #define Set_Node_Length(node,len) \
567 Set_Node_Length_To_R((node)-RExC_emit_start, len)
568 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
569 #define Set_Node_Cur_Length(node) \
570 Set_Node_Length(node, RExC_parse - parse_start)
572 /* Get offsets and lengths */
573 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
574 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
576 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
577 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
578 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
582 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
583 #define EXPERIMENTAL_INPLACESCAN
584 #endif /*RE_TRACK_PATTERN_OFFSETS*/
586 #define DEBUG_STUDYDATA(str,data,depth) \
587 DEBUG_OPTIMISE_MORE_r(if(data){ \
588 PerlIO_printf(Perl_debug_log, \
589 "%*s" str "Pos:%"IVdf"/%"IVdf \
590 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
591 (int)(depth)*2, "", \
592 (IV)((data)->pos_min), \
593 (IV)((data)->pos_delta), \
594 (UV)((data)->flags), \
595 (IV)((data)->whilem_c), \
596 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
597 is_inf ? "INF " : "" \
599 if ((data)->last_found) \
600 PerlIO_printf(Perl_debug_log, \
601 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
602 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
603 SvPVX_const((data)->last_found), \
604 (IV)((data)->last_end), \
605 (IV)((data)->last_start_min), \
606 (IV)((data)->last_start_max), \
607 ((data)->longest && \
608 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
609 SvPVX_const((data)->longest_fixed), \
610 (IV)((data)->offset_fixed), \
611 ((data)->longest && \
612 (data)->longest==&((data)->longest_float)) ? "*" : "", \
613 SvPVX_const((data)->longest_float), \
614 (IV)((data)->offset_float_min), \
615 (IV)((data)->offset_float_max) \
617 PerlIO_printf(Perl_debug_log,"\n"); \
620 static void clear_re(pTHX_ void *r);
622 /* Mark that we cannot extend a found fixed substring at this point.
623 Update the longest found anchored substring and the longest found
624 floating substrings if needed. */
627 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
629 const STRLEN l = CHR_SVLEN(data->last_found);
630 const STRLEN old_l = CHR_SVLEN(*data->longest);
631 GET_RE_DEBUG_FLAGS_DECL;
633 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
634 SvSetMagicSV(*data->longest, data->last_found);
635 if (*data->longest == data->longest_fixed) {
636 data->offset_fixed = l ? data->last_start_min : data->pos_min;
637 if (data->flags & SF_BEFORE_EOL)
639 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
641 data->flags &= ~SF_FIX_BEFORE_EOL;
642 data->minlen_fixed=minlenp;
643 data->lookbehind_fixed=0;
645 else { /* *data->longest == data->longest_float */
646 data->offset_float_min = l ? data->last_start_min : data->pos_min;
647 data->offset_float_max = (l
648 ? data->last_start_max
649 : data->pos_min + data->pos_delta);
650 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
651 data->offset_float_max = I32_MAX;
652 if (data->flags & SF_BEFORE_EOL)
654 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
656 data->flags &= ~SF_FL_BEFORE_EOL;
657 data->minlen_float=minlenp;
658 data->lookbehind_float=0;
661 SvCUR_set(data->last_found, 0);
663 SV * const sv = data->last_found;
664 if (SvUTF8(sv) && SvMAGICAL(sv)) {
665 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
671 data->flags &= ~SF_BEFORE_EOL;
672 DEBUG_STUDYDATA("commit: ",data,0);
675 /* Can match anything (initialization) */
677 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
679 ANYOF_CLASS_ZERO(cl);
680 ANYOF_BITMAP_SETALL(cl);
681 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
683 cl->flags |= ANYOF_LOCALE;
686 /* Can match anything (initialization) */
688 S_cl_is_anything(const struct regnode_charclass_class *cl)
692 for (value = 0; value <= ANYOF_MAX; value += 2)
693 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
695 if (!(cl->flags & ANYOF_UNICODE_ALL))
697 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
702 /* Can match anything (initialization) */
704 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
706 Zero(cl, 1, struct regnode_charclass_class);
708 cl_anything(pRExC_state, cl);
712 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
714 Zero(cl, 1, struct regnode_charclass_class);
716 cl_anything(pRExC_state, cl);
718 cl->flags |= ANYOF_LOCALE;
721 /* 'And' a given class with another one. Can create false positives */
722 /* We assume that cl is not inverted */
724 S_cl_and(struct regnode_charclass_class *cl,
725 const struct regnode_charclass_class *and_with)
728 assert(and_with->type == ANYOF);
729 if (!(and_with->flags & ANYOF_CLASS)
730 && !(cl->flags & ANYOF_CLASS)
731 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
732 && !(and_with->flags & ANYOF_FOLD)
733 && !(cl->flags & ANYOF_FOLD)) {
736 if (and_with->flags & ANYOF_INVERT)
737 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
738 cl->bitmap[i] &= ~and_with->bitmap[i];
740 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
741 cl->bitmap[i] &= and_with->bitmap[i];
742 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
743 if (!(and_with->flags & ANYOF_EOS))
744 cl->flags &= ~ANYOF_EOS;
746 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
747 !(and_with->flags & ANYOF_INVERT)) {
748 cl->flags &= ~ANYOF_UNICODE_ALL;
749 cl->flags |= ANYOF_UNICODE;
750 ARG_SET(cl, ARG(and_with));
752 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
753 !(and_with->flags & ANYOF_INVERT))
754 cl->flags &= ~ANYOF_UNICODE_ALL;
755 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
756 !(and_with->flags & ANYOF_INVERT))
757 cl->flags &= ~ANYOF_UNICODE;
760 /* 'OR' a given class with another one. Can create false positives */
761 /* We assume that cl is not inverted */
763 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
765 if (or_with->flags & ANYOF_INVERT) {
767 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
768 * <= (B1 | !B2) | (CL1 | !CL2)
769 * which is wasteful if CL2 is small, but we ignore CL2:
770 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
771 * XXXX Can we handle case-fold? Unclear:
772 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
773 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
775 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
776 && !(or_with->flags & ANYOF_FOLD)
777 && !(cl->flags & ANYOF_FOLD) ) {
780 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
781 cl->bitmap[i] |= ~or_with->bitmap[i];
782 } /* XXXX: logic is complicated otherwise */
784 cl_anything(pRExC_state, cl);
787 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
788 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
789 && (!(or_with->flags & ANYOF_FOLD)
790 || (cl->flags & ANYOF_FOLD)) ) {
793 /* OR char bitmap and class bitmap separately */
794 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
795 cl->bitmap[i] |= or_with->bitmap[i];
796 if (or_with->flags & ANYOF_CLASS) {
797 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
798 cl->classflags[i] |= or_with->classflags[i];
799 cl->flags |= ANYOF_CLASS;
802 else { /* XXXX: logic is complicated, leave it along for a moment. */
803 cl_anything(pRExC_state, cl);
806 if (or_with->flags & ANYOF_EOS)
807 cl->flags |= ANYOF_EOS;
809 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
810 ARG(cl) != ARG(or_with)) {
811 cl->flags |= ANYOF_UNICODE_ALL;
812 cl->flags &= ~ANYOF_UNICODE;
814 if (or_with->flags & ANYOF_UNICODE_ALL) {
815 cl->flags |= ANYOF_UNICODE_ALL;
816 cl->flags &= ~ANYOF_UNICODE;
820 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
821 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
822 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
823 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
828 dump_trie(trie,widecharmap,revcharmap)
829 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
830 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
832 These routines dump out a trie in a somewhat readable format.
833 The _interim_ variants are used for debugging the interim
834 tables that are used to generate the final compressed
835 representation which is what dump_trie expects.
837 Part of the reason for their existance is to provide a form
838 of documentation as to how the different representations function.
843 Dumps the final compressed table form of the trie to Perl_debug_log.
844 Used for debugging make_trie().
848 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
849 AV *revcharmap, U32 depth)
852 SV *sv=sv_newmortal();
853 int colwidth= widecharmap ? 6 : 4;
854 GET_RE_DEBUG_FLAGS_DECL;
857 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
858 (int)depth * 2 + 2,"",
859 "Match","Base","Ofs" );
861 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
862 SV ** const tmp = av_fetch( revcharmap, state, 0);
864 PerlIO_printf( Perl_debug_log, "%*s",
866 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
867 PL_colors[0], PL_colors[1],
868 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
869 PERL_PV_ESCAPE_FIRSTCHAR
874 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
875 (int)depth * 2 + 2,"");
877 for( state = 0 ; state < trie->uniquecharcount ; state++ )
878 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
879 PerlIO_printf( Perl_debug_log, "\n");
881 for( state = 1 ; state < trie->statecount ; state++ ) {
882 const U32 base = trie->states[ state ].trans.base;
884 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
886 if ( trie->states[ state ].wordnum ) {
887 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
889 PerlIO_printf( Perl_debug_log, "%6s", "" );
892 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
897 while( ( base + ofs < trie->uniquecharcount ) ||
898 ( base + ofs - trie->uniquecharcount < trie->lasttrans
899 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
902 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
904 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
905 if ( ( base + ofs >= trie->uniquecharcount ) &&
906 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
907 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
909 PerlIO_printf( Perl_debug_log, "%*"UVXf,
911 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
913 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
917 PerlIO_printf( Perl_debug_log, "]");
920 PerlIO_printf( Perl_debug_log, "\n" );
924 Dumps a fully constructed but uncompressed trie in list form.
925 List tries normally only are used for construction when the number of
926 possible chars (trie->uniquecharcount) is very high.
927 Used for debugging make_trie().
930 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
931 HV *widecharmap, AV *revcharmap, U32 next_alloc,
935 SV *sv=sv_newmortal();
936 int colwidth= widecharmap ? 6 : 4;
937 GET_RE_DEBUG_FLAGS_DECL;
938 /* print out the table precompression. */
939 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
940 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
941 "------:-----+-----------------\n" );
943 for( state=1 ; state < next_alloc ; state ++ ) {
946 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
947 (int)depth * 2 + 2,"", (UV)state );
948 if ( ! trie->states[ state ].wordnum ) {
949 PerlIO_printf( Perl_debug_log, "%5s| ","");
951 PerlIO_printf( Perl_debug_log, "W%4x| ",
952 trie->states[ state ].wordnum
955 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
956 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
958 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
960 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
961 PL_colors[0], PL_colors[1],
962 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
963 PERL_PV_ESCAPE_FIRSTCHAR
965 TRIE_LIST_ITEM(state,charid).forid,
966 (UV)TRIE_LIST_ITEM(state,charid).newstate
969 PerlIO_printf(Perl_debug_log, "\n%*s| ",
970 (int)((depth * 2) + 14), "");
973 PerlIO_printf( Perl_debug_log, "\n");
978 Dumps a fully constructed but uncompressed trie in table form.
979 This is the normal DFA style state transition table, with a few
980 twists to facilitate compression later.
981 Used for debugging make_trie().
984 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
985 HV *widecharmap, AV *revcharmap, U32 next_alloc,
990 SV *sv=sv_newmortal();
991 int colwidth= widecharmap ? 6 : 4;
992 GET_RE_DEBUG_FLAGS_DECL;
995 print out the table precompression so that we can do a visual check
996 that they are identical.
999 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1001 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1002 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1004 PerlIO_printf( Perl_debug_log, "%*s",
1006 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1007 PL_colors[0], PL_colors[1],
1008 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1009 PERL_PV_ESCAPE_FIRSTCHAR
1015 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1017 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1018 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1021 PerlIO_printf( Perl_debug_log, "\n" );
1023 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1025 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1026 (int)depth * 2 + 2,"",
1027 (UV)TRIE_NODENUM( state ) );
1029 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1030 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1032 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1034 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1036 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1037 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1039 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1040 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1047 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1048 startbranch: the first branch in the whole branch sequence
1049 first : start branch of sequence of branch-exact nodes.
1050 May be the same as startbranch
1051 last : Thing following the last branch.
1052 May be the same as tail.
1053 tail : item following the branch sequence
1054 count : words in the sequence
1055 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1056 depth : indent depth
1058 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1060 A trie is an N'ary tree where the branches are determined by digital
1061 decomposition of the key. IE, at the root node you look up the 1st character and
1062 follow that branch repeat until you find the end of the branches. Nodes can be
1063 marked as "accepting" meaning they represent a complete word. Eg:
1067 would convert into the following structure. Numbers represent states, letters
1068 following numbers represent valid transitions on the letter from that state, if
1069 the number is in square brackets it represents an accepting state, otherwise it
1070 will be in parenthesis.
1072 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1076 (1) +-i->(6)-+-s->[7]
1078 +-s->(3)-+-h->(4)-+-e->[5]
1080 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1082 This shows that when matching against the string 'hers' we will begin at state 1
1083 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1084 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1085 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1086 single traverse. We store a mapping from accepting to state to which word was
1087 matched, and then when we have multiple possibilities we try to complete the
1088 rest of the regex in the order in which they occured in the alternation.
1090 The only prior NFA like behaviour that would be changed by the TRIE support is
1091 the silent ignoring of duplicate alternations which are of the form:
1093 / (DUPE|DUPE) X? (?{ ... }) Y /x
1095 Thus EVAL blocks follwing a trie may be called a different number of times with
1096 and without the optimisation. With the optimisations dupes will be silently
1097 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1098 the following demonstrates:
1100 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1102 which prints out 'word' three times, but
1104 'words'=~/(word|word|word)(?{ print $1 })S/
1106 which doesnt print it out at all. This is due to other optimisations kicking in.
1108 Example of what happens on a structural level:
1110 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1112 1: CURLYM[1] {1,32767}(18)
1123 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1124 and should turn into:
1126 1: CURLYM[1] {1,32767}(18)
1128 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1136 Cases where tail != last would be like /(?foo|bar)baz/:
1146 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1147 and would end up looking like:
1150 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1157 d = uvuni_to_utf8_flags(d, uv, 0);
1159 is the recommended Unicode-aware way of saying
1164 #define TRIE_STORE_REVCHAR \
1166 SV *tmp = newSVpvs(""); \
1167 if (UTF) SvUTF8_on(tmp); \
1168 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
1169 av_push( revcharmap, tmp ); \
1172 #define TRIE_READ_CHAR STMT_START { \
1176 if ( foldlen > 0 ) { \
1177 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1182 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1183 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1184 foldlen -= UNISKIP( uvc ); \
1185 scan = foldbuf + UNISKIP( uvc ); \
1188 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1198 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1199 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1200 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1201 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1203 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1204 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1205 TRIE_LIST_CUR( state )++; \
1208 #define TRIE_LIST_NEW(state) STMT_START { \
1209 Newxz( trie->states[ state ].trans.list, \
1210 4, reg_trie_trans_le ); \
1211 TRIE_LIST_CUR( state ) = 1; \
1212 TRIE_LIST_LEN( state ) = 4; \
1215 #define TRIE_HANDLE_WORD(state) STMT_START { \
1216 U16 dupe= trie->states[ state ].wordnum; \
1217 regnode * const noper_next = regnext( noper ); \
1219 if (trie->wordlen) \
1220 trie->wordlen[ curword ] = wordlen; \
1222 /* store the word for dumping */ \
1224 if (OP(noper) != NOTHING) \
1225 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1227 tmp = newSVpvn( "", 0 ); \
1228 if ( UTF ) SvUTF8_on( tmp ); \
1229 av_push( trie_words, tmp ); \
1234 if ( noper_next < tail ) { \
1236 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1237 trie->jump[curword] = (U16)(noper_next - convert); \
1239 jumper = noper_next; \
1241 nextbranch= regnext(cur); \
1245 /* So it's a dupe. This means we need to maintain a */\
1246 /* linked-list from the first to the next. */\
1247 /* we only allocate the nextword buffer when there */\
1248 /* a dupe, so first time we have to do the allocation */\
1249 if (!trie->nextword) \
1250 trie->nextword = (U16 *) \
1251 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1252 while ( trie->nextword[dupe] ) \
1253 dupe= trie->nextword[dupe]; \
1254 trie->nextword[dupe]= curword; \
1256 /* we haven't inserted this word yet. */ \
1257 trie->states[ state ].wordnum = curword; \
1262 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1263 ( ( base + charid >= ucharcount \
1264 && base + charid < ubound \
1265 && state == trie->trans[ base - ucharcount + charid ].check \
1266 && trie->trans[ base - ucharcount + charid ].next ) \
1267 ? trie->trans[ base - ucharcount + charid ].next \
1268 : ( state==1 ? special : 0 ) \
1272 #define MADE_JUMP_TRIE 2
1273 #define MADE_EXACT_TRIE 4
1276 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1279 /* first pass, loop through and scan words */
1280 reg_trie_data *trie;
1281 HV *widecharmap = NULL;
1282 AV *revcharmap = newAV();
1284 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1289 regnode *jumper = NULL;
1290 regnode *nextbranch = NULL;
1291 regnode *convert = NULL;
1292 /* we just use folder as a flag in utf8 */
1293 const U8 * const folder = ( flags == EXACTF
1295 : ( flags == EXACTFL
1302 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1303 AV *trie_words = NULL;
1304 /* along with revcharmap, this only used during construction but both are
1305 * useful during debugging so we store them in the struct when debugging.
1308 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1309 STRLEN trie_charcount=0;
1311 SV *re_trie_maxbuff;
1312 GET_RE_DEBUG_FLAGS_DECL;
1314 PERL_UNUSED_ARG(depth);
1317 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1319 trie->startstate = 1;
1320 trie->wordcount = word_count;
1321 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1322 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1323 if (!(UTF && folder))
1324 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1326 trie_words = newAV();
1329 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1330 if (!SvIOK(re_trie_maxbuff)) {
1331 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1334 PerlIO_printf( Perl_debug_log,
1335 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1336 (int)depth * 2 + 2, "",
1337 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1338 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1342 /* Find the node we are going to overwrite */
1343 if ( first == startbranch && OP( last ) != BRANCH ) {
1344 /* whole branch chain */
1347 /* branch sub-chain */
1348 convert = NEXTOPER( first );
1351 /* -- First loop and Setup --
1353 We first traverse the branches and scan each word to determine if it
1354 contains widechars, and how many unique chars there are, this is
1355 important as we have to build a table with at least as many columns as we
1358 We use an array of integers to represent the character codes 0..255
1359 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1360 native representation of the character value as the key and IV's for the
1363 *TODO* If we keep track of how many times each character is used we can
1364 remap the columns so that the table compression later on is more
1365 efficient in terms of memory by ensuring most common value is in the
1366 middle and the least common are on the outside. IMO this would be better
1367 than a most to least common mapping as theres a decent chance the most
1368 common letter will share a node with the least common, meaning the node
1369 will not be compressable. With a middle is most common approach the worst
1370 case is when we have the least common nodes twice.
1374 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1375 regnode * const noper = NEXTOPER( cur );
1376 const U8 *uc = (U8*)STRING( noper );
1377 const U8 * const e = uc + STR_LEN( noper );
1379 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1380 const U8 *scan = (U8*)NULL;
1381 U32 wordlen = 0; /* required init */
1383 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1385 if (OP(noper) == NOTHING) {
1389 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1390 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1391 regardless of encoding */
1393 for ( ; uc < e ; uc += len ) {
1394 TRIE_CHARCOUNT(trie)++;
1398 if ( !trie->charmap[ uvc ] ) {
1399 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1401 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1405 /* store the codepoint in the bitmap, and if its ascii
1406 also store its folded equivelent. */
1407 TRIE_BITMAP_SET(trie,uvc);
1408 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1409 set_bit = 0; /* We've done our bit :-) */
1414 widecharmap = newHV();
1416 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1419 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1421 if ( !SvTRUE( *svpp ) ) {
1422 sv_setiv( *svpp, ++trie->uniquecharcount );
1427 if( cur == first ) {
1430 } else if (chars < trie->minlen) {
1432 } else if (chars > trie->maxlen) {
1436 } /* end first pass */
1437 DEBUG_TRIE_COMPILE_r(
1438 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1439 (int)depth * 2 + 2,"",
1440 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1441 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1442 (int)trie->minlen, (int)trie->maxlen )
1444 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1447 We now know what we are dealing with in terms of unique chars and
1448 string sizes so we can calculate how much memory a naive
1449 representation using a flat table will take. If it's over a reasonable
1450 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1451 conservative but potentially much slower representation using an array
1454 At the end we convert both representations into the same compressed
1455 form that will be used in regexec.c for matching with. The latter
1456 is a form that cannot be used to construct with but has memory
1457 properties similar to the list form and access properties similar
1458 to the table form making it both suitable for fast searches and
1459 small enough that its feasable to store for the duration of a program.
1461 See the comment in the code where the compressed table is produced
1462 inplace from the flat tabe representation for an explanation of how
1463 the compression works.
1468 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1470 Second Pass -- Array Of Lists Representation
1472 Each state will be represented by a list of charid:state records
1473 (reg_trie_trans_le) the first such element holds the CUR and LEN
1474 points of the allocated array. (See defines above).
1476 We build the initial structure using the lists, and then convert
1477 it into the compressed table form which allows faster lookups
1478 (but cant be modified once converted).
1481 STRLEN transcount = 1;
1483 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1484 "%*sCompiling trie using list compiler\n",
1485 (int)depth * 2 + 2, ""));
1487 trie->states = (reg_trie_state *)
1488 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1489 sizeof(reg_trie_state) );
1493 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1495 regnode * const noper = NEXTOPER( cur );
1496 U8 *uc = (U8*)STRING( noper );
1497 const U8 * const e = uc + STR_LEN( noper );
1498 U32 state = 1; /* required init */
1499 U16 charid = 0; /* sanity init */
1500 U8 *scan = (U8*)NULL; /* sanity init */
1501 STRLEN foldlen = 0; /* required init */
1502 U32 wordlen = 0; /* required init */
1503 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1505 if (OP(noper) != NOTHING) {
1506 for ( ; uc < e ; uc += len ) {
1511 charid = trie->charmap[ uvc ];
1513 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1517 charid=(U16)SvIV( *svpp );
1520 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1527 if ( !trie->states[ state ].trans.list ) {
1528 TRIE_LIST_NEW( state );
1530 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1531 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1532 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1537 newstate = next_alloc++;
1538 TRIE_LIST_PUSH( state, charid, newstate );
1543 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1547 TRIE_HANDLE_WORD(state);
1549 } /* end second pass */
1551 /* next alloc is the NEXT state to be allocated */
1552 trie->statecount = next_alloc;
1553 trie->states = (reg_trie_state *)
1554 PerlMemShared_realloc( trie->states,
1556 * sizeof(reg_trie_state) );
1558 /* and now dump it out before we compress it */
1559 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1560 revcharmap, next_alloc,
1564 trie->trans = (reg_trie_trans *)
1565 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1572 for( state=1 ; state < next_alloc ; state ++ ) {
1576 DEBUG_TRIE_COMPILE_MORE_r(
1577 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1581 if (trie->states[state].trans.list) {
1582 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1586 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1587 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1588 if ( forid < minid ) {
1590 } else if ( forid > maxid ) {
1594 if ( transcount < tp + maxid - minid + 1) {
1596 trie->trans = (reg_trie_trans *)
1597 PerlMemShared_realloc( trie->trans,
1599 * sizeof(reg_trie_trans) );
1600 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1602 base = trie->uniquecharcount + tp - minid;
1603 if ( maxid == minid ) {
1605 for ( ; zp < tp ; zp++ ) {
1606 if ( ! trie->trans[ zp ].next ) {
1607 base = trie->uniquecharcount + zp - minid;
1608 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1609 trie->trans[ zp ].check = state;
1615 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1616 trie->trans[ tp ].check = state;
1621 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1622 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1623 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1624 trie->trans[ tid ].check = state;
1626 tp += ( maxid - minid + 1 );
1628 Safefree(trie->states[ state ].trans.list);
1631 DEBUG_TRIE_COMPILE_MORE_r(
1632 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1635 trie->states[ state ].trans.base=base;
1637 trie->lasttrans = tp + 1;
1641 Second Pass -- Flat Table Representation.
1643 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1644 We know that we will need Charcount+1 trans at most to store the data
1645 (one row per char at worst case) So we preallocate both structures
1646 assuming worst case.
1648 We then construct the trie using only the .next slots of the entry
1651 We use the .check field of the first entry of the node temporarily to
1652 make compression both faster and easier by keeping track of how many non
1653 zero fields are in the node.
1655 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1658 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1659 number representing the first entry of the node, and state as a
1660 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1661 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1662 are 2 entrys per node. eg:
1670 The table is internally in the right hand, idx form. However as we also
1671 have to deal with the states array which is indexed by nodenum we have to
1672 use TRIE_NODENUM() to convert.
1675 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1676 "%*sCompiling trie using table compiler\n",
1677 (int)depth * 2 + 2, ""));
1679 trie->trans = (reg_trie_trans *)
1680 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1681 * trie->uniquecharcount + 1,
1682 sizeof(reg_trie_trans) );
1683 trie->states = (reg_trie_state *)
1684 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1685 sizeof(reg_trie_state) );
1686 next_alloc = trie->uniquecharcount + 1;
1689 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1691 regnode * const noper = NEXTOPER( cur );
1692 const U8 *uc = (U8*)STRING( noper );
1693 const U8 * const e = uc + STR_LEN( noper );
1695 U32 state = 1; /* required init */
1697 U16 charid = 0; /* sanity init */
1698 U32 accept_state = 0; /* sanity init */
1699 U8 *scan = (U8*)NULL; /* sanity init */
1701 STRLEN foldlen = 0; /* required init */
1702 U32 wordlen = 0; /* required init */
1703 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1705 if ( OP(noper) != NOTHING ) {
1706 for ( ; uc < e ; uc += len ) {
1711 charid = trie->charmap[ uvc ];
1713 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1714 charid = svpp ? (U16)SvIV(*svpp) : 0;
1718 if ( !trie->trans[ state + charid ].next ) {
1719 trie->trans[ state + charid ].next = next_alloc;
1720 trie->trans[ state ].check++;
1721 next_alloc += trie->uniquecharcount;
1723 state = trie->trans[ state + charid ].next;
1725 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1727 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1730 accept_state = TRIE_NODENUM( state );
1731 TRIE_HANDLE_WORD(accept_state);
1733 } /* end second pass */
1735 /* and now dump it out before we compress it */
1736 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1738 next_alloc, depth+1));
1742 * Inplace compress the table.*
1744 For sparse data sets the table constructed by the trie algorithm will
1745 be mostly 0/FAIL transitions or to put it another way mostly empty.
1746 (Note that leaf nodes will not contain any transitions.)
1748 This algorithm compresses the tables by eliminating most such
1749 transitions, at the cost of a modest bit of extra work during lookup:
1751 - Each states[] entry contains a .base field which indicates the
1752 index in the state[] array wheres its transition data is stored.
1754 - If .base is 0 there are no valid transitions from that node.
1756 - If .base is nonzero then charid is added to it to find an entry in
1759 -If trans[states[state].base+charid].check!=state then the
1760 transition is taken to be a 0/Fail transition. Thus if there are fail
1761 transitions at the front of the node then the .base offset will point
1762 somewhere inside the previous nodes data (or maybe even into a node
1763 even earlier), but the .check field determines if the transition is
1767 The following process inplace converts the table to the compressed
1768 table: We first do not compress the root node 1,and mark its all its
1769 .check pointers as 1 and set its .base pointer as 1 as well. This
1770 allows to do a DFA construction from the compressed table later, and
1771 ensures that any .base pointers we calculate later are greater than
1774 - We set 'pos' to indicate the first entry of the second node.
1776 - We then iterate over the columns of the node, finding the first and
1777 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1778 and set the .check pointers accordingly, and advance pos
1779 appropriately and repreat for the next node. Note that when we copy
1780 the next pointers we have to convert them from the original
1781 NODEIDX form to NODENUM form as the former is not valid post
1784 - If a node has no transitions used we mark its base as 0 and do not
1785 advance the pos pointer.
1787 - If a node only has one transition we use a second pointer into the
1788 structure to fill in allocated fail transitions from other states.
1789 This pointer is independent of the main pointer and scans forward
1790 looking for null transitions that are allocated to a state. When it
1791 finds one it writes the single transition into the "hole". If the
1792 pointer doesnt find one the single transition is appended as normal.
1794 - Once compressed we can Renew/realloc the structures to release the
1797 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1798 specifically Fig 3.47 and the associated pseudocode.
1802 const U32 laststate = TRIE_NODENUM( next_alloc );
1805 trie->statecount = laststate;
1807 for ( state = 1 ; state < laststate ; state++ ) {
1809 const U32 stateidx = TRIE_NODEIDX( state );
1810 const U32 o_used = trie->trans[ stateidx ].check;
1811 U32 used = trie->trans[ stateidx ].check;
1812 trie->trans[ stateidx ].check = 0;
1814 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1815 if ( flag || trie->trans[ stateidx + charid ].next ) {
1816 if ( trie->trans[ stateidx + charid ].next ) {
1818 for ( ; zp < pos ; zp++ ) {
1819 if ( ! trie->trans[ zp ].next ) {
1823 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1824 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1825 trie->trans[ zp ].check = state;
1826 if ( ++zp > pos ) pos = zp;
1833 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1835 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1836 trie->trans[ pos ].check = state;
1841 trie->lasttrans = pos + 1;
1842 trie->states = (reg_trie_state *)
1843 PerlMemShared_realloc( trie->states, laststate
1844 * sizeof(reg_trie_state) );
1845 DEBUG_TRIE_COMPILE_MORE_r(
1846 PerlIO_printf( Perl_debug_log,
1847 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1848 (int)depth * 2 + 2,"",
1849 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1852 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1855 } /* end table compress */
1857 DEBUG_TRIE_COMPILE_MORE_r(
1858 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1859 (int)depth * 2 + 2, "",
1860 (UV)trie->statecount,
1861 (UV)trie->lasttrans)
1863 /* resize the trans array to remove unused space */
1864 trie->trans = (reg_trie_trans *)
1865 PerlMemShared_realloc( trie->trans, trie->lasttrans
1866 * sizeof(reg_trie_trans) );
1868 /* and now dump out the compressed format */
1869 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1871 { /* Modify the program and insert the new TRIE node*/
1872 U8 nodetype =(U8)(flags & 0xFF);
1876 regnode *optimize = NULL;
1877 #ifdef RE_TRACK_PATTERN_OFFSETS
1880 U32 mjd_nodelen = 0;
1881 #endif /* RE_TRACK_PATTERN_OFFSETS */
1882 #endif /* DEBUGGING */
1884 This means we convert either the first branch or the first Exact,
1885 depending on whether the thing following (in 'last') is a branch
1886 or not and whther first is the startbranch (ie is it a sub part of
1887 the alternation or is it the whole thing.)
1888 Assuming its a sub part we conver the EXACT otherwise we convert
1889 the whole branch sequence, including the first.
1891 /* Find the node we are going to overwrite */
1892 if ( first != startbranch || OP( last ) == BRANCH ) {
1893 /* branch sub-chain */
1894 NEXT_OFF( first ) = (U16)(last - first);
1895 #ifdef RE_TRACK_PATTERN_OFFSETS
1897 mjd_offset= Node_Offset((convert));
1898 mjd_nodelen= Node_Length((convert));
1901 /* whole branch chain */
1903 #ifdef RE_TRACK_PATTERN_OFFSETS
1906 const regnode *nop = NEXTOPER( convert );
1907 mjd_offset= Node_Offset((nop));
1908 mjd_nodelen= Node_Length((nop));
1912 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1913 (int)depth * 2 + 2, "",
1914 (UV)mjd_offset, (UV)mjd_nodelen)
1917 /* But first we check to see if there is a common prefix we can
1918 split out as an EXACT and put in front of the TRIE node. */
1919 trie->startstate= 1;
1920 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1922 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1926 const U32 base = trie->states[ state ].trans.base;
1928 if ( trie->states[state].wordnum )
1931 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1932 if ( ( base + ofs >= trie->uniquecharcount ) &&
1933 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1934 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1936 if ( ++count > 1 ) {
1937 SV **tmp = av_fetch( revcharmap, ofs, 0);
1938 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1939 if ( state == 1 ) break;
1941 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1943 PerlIO_printf(Perl_debug_log,
1944 "%*sNew Start State=%"UVuf" Class: [",
1945 (int)depth * 2 + 2, "",
1948 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1949 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1951 TRIE_BITMAP_SET(trie,*ch);
1953 TRIE_BITMAP_SET(trie, folder[ *ch ]);
1955 PerlIO_printf(Perl_debug_log, (char*)ch)
1959 TRIE_BITMAP_SET(trie,*ch);
1961 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1962 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1968 SV **tmp = av_fetch( revcharmap, idx, 0);
1969 char *ch = SvPV_nolen( *tmp );
1971 SV *sv=sv_newmortal();
1972 PerlIO_printf( Perl_debug_log,
1973 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1974 (int)depth * 2 + 2, "",
1976 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1977 PL_colors[0], PL_colors[1],
1978 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1979 PERL_PV_ESCAPE_FIRSTCHAR
1984 OP( convert ) = nodetype;
1985 str=STRING(convert);
1996 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2002 regnode *n = convert+NODE_SZ_STR(convert);
2003 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2004 trie->startstate = state;
2005 trie->minlen -= (state - 1);
2006 trie->maxlen -= (state - 1);
2008 regnode *fix = convert;
2009 U32 word = trie->wordcount;
2011 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2012 while( ++fix < n ) {
2013 Set_Node_Offset_Length(fix, 0, 0);
2016 SV ** const tmp = av_fetch( trie_words, word, 0 );
2018 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2019 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2021 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2028 NEXT_OFF(convert) = (U16)(tail - convert);
2029 DEBUG_r(optimize= n);
2035 if ( trie->maxlen ) {
2036 NEXT_OFF( convert ) = (U16)(tail - convert);
2037 ARG_SET( convert, data_slot );
2038 /* Store the offset to the first unabsorbed branch in
2039 jump[0], which is otherwise unused by the jump logic.
2040 We use this when dumping a trie and during optimisation. */
2042 trie->jump[0] = (U16)(nextbranch - convert);
2045 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2046 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2048 OP( convert ) = TRIEC;
2049 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2050 PerlMemShared_free(trie->bitmap);
2053 OP( convert ) = TRIE;
2055 /* store the type in the flags */
2056 convert->flags = nodetype;
2060 + regarglen[ OP( convert ) ];
2062 /* XXX We really should free up the resource in trie now,
2063 as we won't use them - (which resources?) dmq */
2065 /* needed for dumping*/
2066 DEBUG_r(if (optimize) {
2067 regnode *opt = convert;
2069 while ( ++opt < optimize) {
2070 Set_Node_Offset_Length(opt,0,0);
2073 Try to clean up some of the debris left after the
2076 while( optimize < jumper ) {
2077 mjd_nodelen += Node_Length((optimize));
2078 OP( optimize ) = OPTIMIZED;
2079 Set_Node_Offset_Length(optimize,0,0);
2082 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2084 } /* end node insert */
2085 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2087 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2088 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2090 SvREFCNT_dec(revcharmap);
2094 : trie->startstate>1
2100 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2102 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2104 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2105 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2108 We find the fail state for each state in the trie, this state is the longest proper
2109 suffix of the current states 'word' that is also a proper prefix of another word in our
2110 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2111 the DFA not to have to restart after its tried and failed a word at a given point, it
2112 simply continues as though it had been matching the other word in the first place.
2114 'abcdgu'=~/abcdefg|cdgu/
2115 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2116 fail, which would bring use to the state representing 'd' in the second word where we would
2117 try 'g' and succeed, prodceding to match 'cdgu'.
2119 /* add a fail transition */
2120 const U32 trie_offset = ARG(source);
2121 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2123 const U32 ucharcount = trie->uniquecharcount;
2124 const U32 numstates = trie->statecount;
2125 const U32 ubound = trie->lasttrans + ucharcount;
2129 U32 base = trie->states[ 1 ].trans.base;
2132 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2133 GET_RE_DEBUG_FLAGS_DECL;
2135 PERL_UNUSED_ARG(depth);
2139 ARG_SET( stclass, data_slot );
2140 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2141 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2142 aho->trie=trie_offset;
2143 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2144 Copy( trie->states, aho->states, numstates, reg_trie_state );
2145 Newxz( q, numstates, U32);
2146 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2149 /* initialize fail[0..1] to be 1 so that we always have
2150 a valid final fail state */
2151 fail[ 0 ] = fail[ 1 ] = 1;
2153 for ( charid = 0; charid < ucharcount ; charid++ ) {
2154 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2156 q[ q_write ] = newstate;
2157 /* set to point at the root */
2158 fail[ q[ q_write++ ] ]=1;
2161 while ( q_read < q_write) {
2162 const U32 cur = q[ q_read++ % numstates ];
2163 base = trie->states[ cur ].trans.base;
2165 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2166 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2168 U32 fail_state = cur;
2171 fail_state = fail[ fail_state ];
2172 fail_base = aho->states[ fail_state ].trans.base;
2173 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2175 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2176 fail[ ch_state ] = fail_state;
2177 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2179 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2181 q[ q_write++ % numstates] = ch_state;
2185 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2186 when we fail in state 1, this allows us to use the
2187 charclass scan to find a valid start char. This is based on the principle
2188 that theres a good chance the string being searched contains lots of stuff
2189 that cant be a start char.
2191 fail[ 0 ] = fail[ 1 ] = 0;
2192 DEBUG_TRIE_COMPILE_r({
2193 PerlIO_printf(Perl_debug_log,
2194 "%*sStclass Failtable (%"UVuf" states): 0",
2195 (int)(depth * 2), "", (UV)numstates
2197 for( q_read=1; q_read<numstates; q_read++ ) {
2198 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2200 PerlIO_printf(Perl_debug_log, "\n");
2203 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2208 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2209 * These need to be revisited when a newer toolchain becomes available.
2211 #if defined(__sparc64__) && defined(__GNUC__)
2212 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2213 # undef SPARC64_GCC_WORKAROUND
2214 # define SPARC64_GCC_WORKAROUND 1
2218 #define DEBUG_PEEP(str,scan,depth) \
2219 DEBUG_OPTIMISE_r({if (scan){ \
2220 SV * const mysv=sv_newmortal(); \
2221 regnode *Next = regnext(scan); \
2222 regprop(RExC_rx, mysv, scan); \
2223 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2224 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2225 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2232 #define JOIN_EXACT(scan,min,flags) \
2233 if (PL_regkind[OP(scan)] == EXACT) \
2234 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2237 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2238 /* Merge several consecutive EXACTish nodes into one. */
2239 regnode *n = regnext(scan);
2241 regnode *next = scan + NODE_SZ_STR(scan);
2245 regnode *stop = scan;
2246 GET_RE_DEBUG_FLAGS_DECL;
2248 PERL_UNUSED_ARG(depth);
2250 #ifndef EXPERIMENTAL_INPLACESCAN
2251 PERL_UNUSED_ARG(flags);
2252 PERL_UNUSED_ARG(val);
2254 DEBUG_PEEP("join",scan,depth);
2256 /* Skip NOTHING, merge EXACT*. */
2258 ( PL_regkind[OP(n)] == NOTHING ||
2259 (stringok && (OP(n) == OP(scan))))
2261 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2263 if (OP(n) == TAIL || n > next)
2265 if (PL_regkind[OP(n)] == NOTHING) {
2266 DEBUG_PEEP("skip:",n,depth);
2267 NEXT_OFF(scan) += NEXT_OFF(n);
2268 next = n + NODE_STEP_REGNODE;
2275 else if (stringok) {
2276 const unsigned int oldl = STR_LEN(scan);
2277 regnode * const nnext = regnext(n);
2279 DEBUG_PEEP("merg",n,depth);
2282 if (oldl + STR_LEN(n) > U8_MAX)
2284 NEXT_OFF(scan) += NEXT_OFF(n);
2285 STR_LEN(scan) += STR_LEN(n);
2286 next = n + NODE_SZ_STR(n);
2287 /* Now we can overwrite *n : */
2288 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2296 #ifdef EXPERIMENTAL_INPLACESCAN
2297 if (flags && !NEXT_OFF(n)) {
2298 DEBUG_PEEP("atch", val, depth);
2299 if (reg_off_by_arg[OP(n)]) {
2300 ARG_SET(n, val - n);
2303 NEXT_OFF(n) = val - n;
2310 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2312 Two problematic code points in Unicode casefolding of EXACT nodes:
2314 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2315 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2321 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2322 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2324 This means that in case-insensitive matching (or "loose matching",
2325 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2326 length of the above casefolded versions) can match a target string
2327 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2328 This would rather mess up the minimum length computation.
2330 What we'll do is to look for the tail four bytes, and then peek
2331 at the preceding two bytes to see whether we need to decrease
2332 the minimum length by four (six minus two).
2334 Thanks to the design of UTF-8, there cannot be false matches:
2335 A sequence of valid UTF-8 bytes cannot be a subsequence of
2336 another valid sequence of UTF-8 bytes.
2339 char * const s0 = STRING(scan), *s, *t;
2340 char * const s1 = s0 + STR_LEN(scan) - 1;
2341 char * const s2 = s1 - 4;
2342 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2343 const char t0[] = "\xaf\x49\xaf\x42";
2345 const char t0[] = "\xcc\x88\xcc\x81";
2347 const char * const t1 = t0 + 3;
2350 s < s2 && (t = ninstr(s, s1, t0, t1));
2353 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2354 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2356 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2357 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2365 n = scan + NODE_SZ_STR(scan);
2367 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2374 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2378 /* REx optimizer. Converts nodes into quickier variants "in place".
2379 Finds fixed substrings. */
2381 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2382 to the position after last scanned or to NULL. */
2384 #define INIT_AND_WITHP \
2385 assert(!and_withp); \
2386 Newx(and_withp,1,struct regnode_charclass_class); \
2387 SAVEFREEPV(and_withp)
2389 /* this is a chain of data about sub patterns we are processing that
2390 need to be handled seperately/specially in study_chunk. Its so
2391 we can simulate recursion without losing state. */
2393 typedef struct scan_frame {
2394 regnode *last; /* last node to process in this frame */
2395 regnode *next; /* next node to process when last is reached */
2396 struct scan_frame *prev; /*previous frame*/
2397 I32 stop; /* what stopparen do we use */
2401 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2404 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2405 I32 *minlenp, I32 *deltap,
2410 struct regnode_charclass_class *and_withp,
2411 U32 flags, U32 depth)
2412 /* scanp: Start here (read-write). */
2413 /* deltap: Write maxlen-minlen here. */
2414 /* last: Stop before this one. */
2415 /* data: string data about the pattern */
2416 /* stopparen: treat close N as END */
2417 /* recursed: which subroutines have we recursed into */
2418 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2421 I32 min = 0, pars = 0, code;
2422 regnode *scan = *scanp, *next;
2424 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2425 int is_inf_internal = 0; /* The studied chunk is infinite */
2426 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2427 scan_data_t data_fake;
2428 SV *re_trie_maxbuff = NULL;
2429 regnode *first_non_open = scan;
2430 I32 stopmin = I32_MAX;
2431 scan_frame *frame = NULL;
2433 GET_RE_DEBUG_FLAGS_DECL;
2436 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2440 while (first_non_open && OP(first_non_open) == OPEN)
2441 first_non_open=regnext(first_non_open);
2446 while ( scan && OP(scan) != END && scan < last ){
2447 /* Peephole optimizer: */
2448 DEBUG_STUDYDATA("Peep:", data,depth);
2449 DEBUG_PEEP("Peep",scan,depth);
2450 JOIN_EXACT(scan,&min,0);
2452 /* Follow the next-chain of the current node and optimize
2453 away all the NOTHINGs from it. */
2454 if (OP(scan) != CURLYX) {
2455 const int max = (reg_off_by_arg[OP(scan)]
2457 /* I32 may be smaller than U16 on CRAYs! */
2458 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2459 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2463 /* Skip NOTHING and LONGJMP. */
2464 while ((n = regnext(n))
2465 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2466 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2467 && off + noff < max)
2469 if (reg_off_by_arg[OP(scan)])
2472 NEXT_OFF(scan) = off;
2477 /* The principal pseudo-switch. Cannot be a switch, since we
2478 look into several different things. */
2479 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2480 || OP(scan) == IFTHEN) {
2481 next = regnext(scan);
2483 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2485 if (OP(next) == code || code == IFTHEN) {
2486 /* NOTE - There is similar code to this block below for handling
2487 TRIE nodes on a re-study. If you change stuff here check there
2489 I32 max1 = 0, min1 = I32_MAX, num = 0;
2490 struct regnode_charclass_class accum;
2491 regnode * const startbranch=scan;
2493 if (flags & SCF_DO_SUBSTR)
2494 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2495 if (flags & SCF_DO_STCLASS)
2496 cl_init_zero(pRExC_state, &accum);
2498 while (OP(scan) == code) {
2499 I32 deltanext, minnext, f = 0, fake;
2500 struct regnode_charclass_class this_class;
2503 data_fake.flags = 0;
2505 data_fake.whilem_c = data->whilem_c;
2506 data_fake.last_closep = data->last_closep;
2509 data_fake.last_closep = &fake;
2511 data_fake.pos_delta = delta;
2512 next = regnext(scan);
2513 scan = NEXTOPER(scan);
2515 scan = NEXTOPER(scan);
2516 if (flags & SCF_DO_STCLASS) {
2517 cl_init(pRExC_state, &this_class);
2518 data_fake.start_class = &this_class;
2519 f = SCF_DO_STCLASS_AND;
2521 if (flags & SCF_WHILEM_VISITED_POS)
2522 f |= SCF_WHILEM_VISITED_POS;
2524 /* we suppose the run is continuous, last=next...*/
2525 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2527 stopparen, recursed, NULL, f,depth+1);
2530 if (max1 < minnext + deltanext)
2531 max1 = minnext + deltanext;
2532 if (deltanext == I32_MAX)
2533 is_inf = is_inf_internal = 1;
2535 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2537 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2538 if ( stopmin > minnext)
2539 stopmin = min + min1;
2540 flags &= ~SCF_DO_SUBSTR;
2542 data->flags |= SCF_SEEN_ACCEPT;
2545 if (data_fake.flags & SF_HAS_EVAL)
2546 data->flags |= SF_HAS_EVAL;
2547 data->whilem_c = data_fake.whilem_c;
2549 if (flags & SCF_DO_STCLASS)
2550 cl_or(pRExC_state, &accum, &this_class);
2552 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2554 if (flags & SCF_DO_SUBSTR) {
2555 data->pos_min += min1;
2556 data->pos_delta += max1 - min1;
2557 if (max1 != min1 || is_inf)
2558 data->longest = &(data->longest_float);
2561 delta += max1 - min1;
2562 if (flags & SCF_DO_STCLASS_OR) {
2563 cl_or(pRExC_state, data->start_class, &accum);
2565 cl_and(data->start_class, and_withp);
2566 flags &= ~SCF_DO_STCLASS;
2569 else if (flags & SCF_DO_STCLASS_AND) {
2571 cl_and(data->start_class, &accum);
2572 flags &= ~SCF_DO_STCLASS;
2575 /* Switch to OR mode: cache the old value of
2576 * data->start_class */
2578 StructCopy(data->start_class, and_withp,
2579 struct regnode_charclass_class);
2580 flags &= ~SCF_DO_STCLASS_AND;
2581 StructCopy(&accum, data->start_class,
2582 struct regnode_charclass_class);
2583 flags |= SCF_DO_STCLASS_OR;
2584 data->start_class->flags |= ANYOF_EOS;
2588 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2591 Assuming this was/is a branch we are dealing with: 'scan' now
2592 points at the item that follows the branch sequence, whatever
2593 it is. We now start at the beginning of the sequence and look
2600 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2602 If we can find such a subseqence we need to turn the first
2603 element into a trie and then add the subsequent branch exact
2604 strings to the trie.
2608 1. patterns where the whole set of branch can be converted.
2610 2. patterns where only a subset can be converted.
2612 In case 1 we can replace the whole set with a single regop
2613 for the trie. In case 2 we need to keep the start and end
2616 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2617 becomes BRANCH TRIE; BRANCH X;
2619 There is an additional case, that being where there is a
2620 common prefix, which gets split out into an EXACT like node
2621 preceding the TRIE node.
2623 If x(1..n)==tail then we can do a simple trie, if not we make
2624 a "jump" trie, such that when we match the appropriate word
2625 we "jump" to the appopriate tail node. Essentailly we turn
2626 a nested if into a case structure of sorts.
2631 if (!re_trie_maxbuff) {
2632 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2633 if (!SvIOK(re_trie_maxbuff))
2634 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2636 if ( SvIV(re_trie_maxbuff)>=0 ) {
2638 regnode *first = (regnode *)NULL;
2639 regnode *last = (regnode *)NULL;
2640 regnode *tail = scan;
2645 SV * const mysv = sv_newmortal(); /* for dumping */
2647 /* var tail is used because there may be a TAIL
2648 regop in the way. Ie, the exacts will point to the
2649 thing following the TAIL, but the last branch will
2650 point at the TAIL. So we advance tail. If we
2651 have nested (?:) we may have to move through several
2655 while ( OP( tail ) == TAIL ) {
2656 /* this is the TAIL generated by (?:) */
2657 tail = regnext( tail );
2662 regprop(RExC_rx, mysv, tail );
2663 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2664 (int)depth * 2 + 2, "",
2665 "Looking for TRIE'able sequences. Tail node is: ",
2666 SvPV_nolen_const( mysv )
2672 step through the branches, cur represents each
2673 branch, noper is the first thing to be matched
2674 as part of that branch and noper_next is the
2675 regnext() of that node. if noper is an EXACT
2676 and noper_next is the same as scan (our current
2677 position in the regex) then the EXACT branch is
2678 a possible optimization target. Once we have
2679 two or more consequetive such branches we can
2680 create a trie of the EXACT's contents and stich
2681 it in place. If the sequence represents all of
2682 the branches we eliminate the whole thing and
2683 replace it with a single TRIE. If it is a
2684 subsequence then we need to stitch it in. This
2685 means the first branch has to remain, and needs
2686 to be repointed at the item on the branch chain
2687 following the last branch optimized. This could
2688 be either a BRANCH, in which case the
2689 subsequence is internal, or it could be the
2690 item following the branch sequence in which
2691 case the subsequence is at the end.
2695 /* dont use tail as the end marker for this traverse */
2696 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2697 regnode * const noper = NEXTOPER( cur );
2698 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2699 regnode * const noper_next = regnext( noper );
2703 regprop(RExC_rx, mysv, cur);
2704 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2705 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2707 regprop(RExC_rx, mysv, noper);
2708 PerlIO_printf( Perl_debug_log, " -> %s",
2709 SvPV_nolen_const(mysv));
2712 regprop(RExC_rx, mysv, noper_next );
2713 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2714 SvPV_nolen_const(mysv));
2716 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2717 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2719 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2720 : PL_regkind[ OP( noper ) ] == EXACT )
2721 || OP(noper) == NOTHING )
2723 && noper_next == tail
2728 if ( !first || optype == NOTHING ) {
2729 if (!first) first = cur;
2730 optype = OP( noper );
2736 make_trie( pRExC_state,
2737 startbranch, first, cur, tail, count,
2740 if ( PL_regkind[ OP( noper ) ] == EXACT
2742 && noper_next == tail
2747 optype = OP( noper );
2757 regprop(RExC_rx, mysv, cur);
2758 PerlIO_printf( Perl_debug_log,
2759 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2760 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2764 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2765 #ifdef TRIE_STUDY_OPT
2766 if ( ((made == MADE_EXACT_TRIE &&
2767 startbranch == first)
2768 || ( first_non_open == first )) &&
2770 flags |= SCF_TRIE_RESTUDY;
2771 if ( startbranch == first
2774 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2784 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2785 scan = NEXTOPER(NEXTOPER(scan));
2786 } else /* single branch is optimized. */
2787 scan = NEXTOPER(scan);
2789 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2790 scan_frame *newframe = NULL;
2795 if (OP(scan) != SUSPEND) {
2796 /* set the pointer */
2797 if (OP(scan) == GOSUB) {
2799 RExC_recurse[ARG2L(scan)] = scan;
2800 start = RExC_open_parens[paren-1];
2801 end = RExC_close_parens[paren-1];
2804 start = RExC_rxi->program + 1;
2808 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2809 SAVEFREEPV(recursed);
2811 if (!PAREN_TEST(recursed,paren+1)) {
2812 PAREN_SET(recursed,paren+1);
2813 Newx(newframe,1,scan_frame);
2815 if (flags & SCF_DO_SUBSTR) {
2816 SCAN_COMMIT(pRExC_state,data,minlenp);
2817 data->longest = &(data->longest_float);
2819 is_inf = is_inf_internal = 1;
2820 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2821 cl_anything(pRExC_state, data->start_class);
2822 flags &= ~SCF_DO_STCLASS;
2825 Newx(newframe,1,scan_frame);
2828 end = regnext(scan);
2833 SAVEFREEPV(newframe);
2834 newframe->next = regnext(scan);
2835 newframe->last = last;
2836 newframe->stop = stopparen;
2837 newframe->prev = frame;
2847 else if (OP(scan) == EXACT) {
2848 I32 l = STR_LEN(scan);
2851 const U8 * const s = (U8*)STRING(scan);
2852 l = utf8_length(s, s + l);
2853 uc = utf8_to_uvchr(s, NULL);
2855 uc = *((U8*)STRING(scan));
2858 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2859 /* The code below prefers earlier match for fixed
2860 offset, later match for variable offset. */
2861 if (data->last_end == -1) { /* Update the start info. */
2862 data->last_start_min = data->pos_min;
2863 data->last_start_max = is_inf
2864 ? I32_MAX : data->pos_min + data->pos_delta;
2866 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2868 SvUTF8_on(data->last_found);
2870 SV * const sv = data->last_found;
2871 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2872 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2873 if (mg && mg->mg_len >= 0)
2874 mg->mg_len += utf8_length((U8*)STRING(scan),
2875 (U8*)STRING(scan)+STR_LEN(scan));
2877 data->last_end = data->pos_min + l;
2878 data->pos_min += l; /* As in the first entry. */
2879 data->flags &= ~SF_BEFORE_EOL;
2881 if (flags & SCF_DO_STCLASS_AND) {
2882 /* Check whether it is compatible with what we know already! */
2886 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2887 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2888 && (!(data->start_class->flags & ANYOF_FOLD)
2889 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2892 ANYOF_CLASS_ZERO(data->start_class);
2893 ANYOF_BITMAP_ZERO(data->start_class);
2895 ANYOF_BITMAP_SET(data->start_class, uc);
2896 data->start_class->flags &= ~ANYOF_EOS;
2898 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2900 else if (flags & SCF_DO_STCLASS_OR) {
2901 /* false positive possible if the class is case-folded */
2903 ANYOF_BITMAP_SET(data->start_class, uc);
2905 data->start_class->flags |= ANYOF_UNICODE_ALL;
2906 data->start_class->flags &= ~ANYOF_EOS;
2907 cl_and(data->start_class, and_withp);
2909 flags &= ~SCF_DO_STCLASS;
2911 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2912 I32 l = STR_LEN(scan);
2913 UV uc = *((U8*)STRING(scan));
2915 /* Search for fixed substrings supports EXACT only. */
2916 if (flags & SCF_DO_SUBSTR) {
2918 SCAN_COMMIT(pRExC_state, data, minlenp);
2921 const U8 * const s = (U8 *)STRING(scan);
2922 l = utf8_length(s, s + l);
2923 uc = utf8_to_uvchr(s, NULL);
2926 if (flags & SCF_DO_SUBSTR)
2928 if (flags & SCF_DO_STCLASS_AND) {
2929 /* Check whether it is compatible with what we know already! */
2933 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2934 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2935 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2937 ANYOF_CLASS_ZERO(data->start_class);
2938 ANYOF_BITMAP_ZERO(data->start_class);
2940 ANYOF_BITMAP_SET(data->start_class, uc);
2941 data->start_class->flags &= ~ANYOF_EOS;
2942 data->start_class->flags |= ANYOF_FOLD;
2943 if (OP(scan) == EXACTFL)
2944 data->start_class->flags |= ANYOF_LOCALE;
2947 else if (flags & SCF_DO_STCLASS_OR) {
2948 if (data->start_class->flags & ANYOF_FOLD) {
2949 /* false positive possible if the class is case-folded.
2950 Assume that the locale settings are the same... */
2952 ANYOF_BITMAP_SET(data->start_class, uc);
2953 data->start_class->flags &= ~ANYOF_EOS;
2955 cl_and(data->start_class, and_withp);
2957 flags &= ~SCF_DO_STCLASS;
2959 else if (strchr((const char*)PL_varies,OP(scan))) {
2960 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2961 I32 f = flags, pos_before = 0;
2962 regnode * const oscan = scan;
2963 struct regnode_charclass_class this_class;
2964 struct regnode_charclass_class *oclass = NULL;
2965 I32 next_is_eval = 0;
2967 switch (PL_regkind[OP(scan)]) {
2968 case WHILEM: /* End of (?:...)* . */
2969 scan = NEXTOPER(scan);
2972 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
2973 next = NEXTOPER(scan);
2974 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
2976 maxcount = REG_INFTY;
2977 next = regnext(scan);
2978 scan = NEXTOPER(scan);
2982 if (flags & SCF_DO_SUBSTR)
2987 if (flags & SCF_DO_STCLASS) {
2989 maxcount = REG_INFTY;
2990 next = regnext(scan);
2991 scan = NEXTOPER(scan);
2994 is_inf = is_inf_internal = 1;
2995 scan = regnext(scan);
2996 if (flags & SCF_DO_SUBSTR) {
2997 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
2998 data->longest = &(data->longest_float);
3000 goto optimize_curly_tail;
3002 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3003 && (scan->flags == stopparen))
3008 mincount = ARG1(scan);
3009 maxcount = ARG2(scan);
3011 next = regnext(scan);
3012 if (OP(scan) == CURLYX) {
3013 I32 lp = (data ? *(data->last_closep) : 0);
3014 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3016 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3017 next_is_eval = (OP(scan) == EVAL);
3019 if (flags & SCF_DO_SUBSTR) {
3020 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3021 pos_before = data->pos_min;
3025 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3027 data->flags |= SF_IS_INF;
3029 if (flags & SCF_DO_STCLASS) {
3030 cl_init(pRExC_state, &this_class);
3031 oclass = data->start_class;
3032 data->start_class = &this_class;
3033 f |= SCF_DO_STCLASS_AND;
3034 f &= ~SCF_DO_STCLASS_OR;
3036 /* These are the cases when once a subexpression
3037 fails at a particular position, it cannot succeed
3038 even after backtracking at the enclosing scope.
3040 XXXX what if minimal match and we are at the
3041 initial run of {n,m}? */
3042 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3043 f &= ~SCF_WHILEM_VISITED_POS;
3045 /* This will finish on WHILEM, setting scan, or on NULL: */
3046 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3047 last, data, stopparen, recursed, NULL,
3049 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3051 if (flags & SCF_DO_STCLASS)
3052 data->start_class = oclass;
3053 if (mincount == 0 || minnext == 0) {
3054 if (flags & SCF_DO_STCLASS_OR) {
3055 cl_or(pRExC_state, data->start_class, &this_class);
3057 else if (flags & SCF_DO_STCLASS_AND) {
3058 /* Switch to OR mode: cache the old value of
3059 * data->start_class */
3061 StructCopy(data->start_class, and_withp,
3062 struct regnode_charclass_class);
3063 flags &= ~SCF_DO_STCLASS_AND;
3064 StructCopy(&this_class, data->start_class,
3065 struct regnode_charclass_class);
3066 flags |= SCF_DO_STCLASS_OR;
3067 data->start_class->flags |= ANYOF_EOS;
3069 } else { /* Non-zero len */
3070 if (flags & SCF_DO_STCLASS_OR) {
3071 cl_or(pRExC_state, data->start_class, &this_class);
3072 cl_and(data->start_class, and_withp);
3074 else if (flags & SCF_DO_STCLASS_AND)
3075 cl_and(data->start_class, &this_class);
3076 flags &= ~SCF_DO_STCLASS;
3078 if (!scan) /* It was not CURLYX, but CURLY. */
3080 if ( /* ? quantifier ok, except for (?{ ... }) */
3081 (next_is_eval || !(mincount == 0 && maxcount == 1))
3082 && (minnext == 0) && (deltanext == 0)
3083 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3084 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3085 && ckWARN(WARN_REGEXP))
3088 "Quantifier unexpected on zero-length expression");
3091 min += minnext * mincount;
3092 is_inf_internal |= ((maxcount == REG_INFTY
3093 && (minnext + deltanext) > 0)
3094 || deltanext == I32_MAX);
3095 is_inf |= is_inf_internal;
3096 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3098 /* Try powerful optimization CURLYX => CURLYN. */
3099 if ( OP(oscan) == CURLYX && data
3100 && data->flags & SF_IN_PAR
3101 && !(data->flags & SF_HAS_EVAL)
3102 && !deltanext && minnext == 1 ) {
3103 /* Try to optimize to CURLYN. */
3104 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3105 regnode * const nxt1 = nxt;
3112 if (!strchr((const char*)PL_simple,OP(nxt))
3113 && !(PL_regkind[OP(nxt)] == EXACT
3114 && STR_LEN(nxt) == 1))
3120 if (OP(nxt) != CLOSE)
3122 if (RExC_open_parens) {
3123 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3124 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3126 /* Now we know that nxt2 is the only contents: */
3127 oscan->flags = (U8)ARG(nxt);
3129 OP(nxt1) = NOTHING; /* was OPEN. */
3132 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3133 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3134 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3135 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3136 OP(nxt + 1) = OPTIMIZED; /* was count. */
3137 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3142 /* Try optimization CURLYX => CURLYM. */
3143 if ( OP(oscan) == CURLYX && data
3144 && !(data->flags & SF_HAS_PAR)
3145 && !(data->flags & SF_HAS_EVAL)
3146 && !deltanext /* atom is fixed width */
3147 && minnext != 0 /* CURLYM can't handle zero width */
3149 /* XXXX How to optimize if data == 0? */
3150 /* Optimize to a simpler form. */
3151 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3155 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3156 && (OP(nxt2) != WHILEM))
3158 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3159 /* Need to optimize away parenths. */
3160 if (data->flags & SF_IN_PAR) {
3161 /* Set the parenth number. */
3162 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3164 if (OP(nxt) != CLOSE)
3165 FAIL("Panic opt close");
3166 oscan->flags = (U8)ARG(nxt);
3167 if (RExC_open_parens) {
3168 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3169 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3171 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3172 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3175 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3176 OP(nxt + 1) = OPTIMIZED; /* was count. */
3177 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3178 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3181 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3182 regnode *nnxt = regnext(nxt1);
3185 if (reg_off_by_arg[OP(nxt1)])
3186 ARG_SET(nxt1, nxt2 - nxt1);
3187 else if (nxt2 - nxt1 < U16_MAX)
3188 NEXT_OFF(nxt1) = nxt2 - nxt1;
3190 OP(nxt) = NOTHING; /* Cannot beautify */
3195 /* Optimize again: */
3196 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3197 NULL, stopparen, recursed, NULL, 0,depth+1);
3202 else if ((OP(oscan) == CURLYX)
3203 && (flags & SCF_WHILEM_VISITED_POS)
3204 /* See the comment on a similar expression above.
3205 However, this time it not a subexpression
3206 we care about, but the expression itself. */
3207 && (maxcount == REG_INFTY)
3208 && data && ++data->whilem_c < 16) {
3209 /* This stays as CURLYX, we can put the count/of pair. */
3210 /* Find WHILEM (as in regexec.c) */
3211 regnode *nxt = oscan + NEXT_OFF(oscan);
3213 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3215 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3216 | (RExC_whilem_seen << 4)); /* On WHILEM */
3218 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3220 if (flags & SCF_DO_SUBSTR) {
3221 SV *last_str = NULL;
3222 int counted = mincount != 0;
3224 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3225 #if defined(SPARC64_GCC_WORKAROUND)
3228 const char *s = NULL;
3231 if (pos_before >= data->last_start_min)
3234 b = data->last_start_min;
3237 s = SvPV_const(data->last_found, l);
3238 old = b - data->last_start_min;
3241 I32 b = pos_before >= data->last_start_min
3242 ? pos_before : data->last_start_min;
3244 const char * const s = SvPV_const(data->last_found, l);
3245 I32 old = b - data->last_start_min;
3249 old = utf8_hop((U8*)s, old) - (U8*)s;
3252 /* Get the added string: */
3253 last_str = newSVpvn(s + old, l);
3255 SvUTF8_on(last_str);
3256 if (deltanext == 0 && pos_before == b) {
3257 /* What was added is a constant string */
3259 SvGROW(last_str, (mincount * l) + 1);
3260 repeatcpy(SvPVX(last_str) + l,
3261 SvPVX_const(last_str), l, mincount - 1);
3262 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3263 /* Add additional parts. */
3264 SvCUR_set(data->last_found,
3265 SvCUR(data->last_found) - l);
3266 sv_catsv(data->last_found, last_str);
3268 SV * sv = data->last_found;
3270 SvUTF8(sv) && SvMAGICAL(sv) ?
3271 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3272 if (mg && mg->mg_len >= 0)
3273 mg->mg_len += CHR_SVLEN(last_str);
3275 data->last_end += l * (mincount - 1);
3278 /* start offset must point into the last copy */
3279 data->last_start_min += minnext * (mincount - 1);
3280 data->last_start_max += is_inf ? I32_MAX
3281 : (maxcount - 1) * (minnext + data->pos_delta);
3284 /* It is counted once already... */
3285 data->pos_min += minnext * (mincount - counted);
3286 data->pos_delta += - counted * deltanext +
3287 (minnext + deltanext) * maxcount - minnext * mincount;
3288 if (mincount != maxcount) {
3289 /* Cannot extend fixed substrings found inside
3291 SCAN_COMMIT(pRExC_state,data,minlenp);
3292 if (mincount && last_str) {
3293 SV * const sv = data->last_found;
3294 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3295 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3299 sv_setsv(sv, last_str);
3300 data->last_end = data->pos_min;
3301 data->last_start_min =
3302 data->pos_min - CHR_SVLEN(last_str);
3303 data->last_start_max = is_inf
3305 : data->pos_min + data->pos_delta
3306 - CHR_SVLEN(last_str);
3308 data->longest = &(data->longest_float);
3310 SvREFCNT_dec(last_str);
3312 if (data && (fl & SF_HAS_EVAL))
3313 data->flags |= SF_HAS_EVAL;
3314 optimize_curly_tail:
3315 if (OP(oscan) != CURLYX) {
3316 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3318 NEXT_OFF(oscan) += NEXT_OFF(next);
3321 default: /* REF and CLUMP only? */
3322 if (flags & SCF_DO_SUBSTR) {
3323 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3324 data->longest = &(data->longest_float);
3326 is_inf = is_inf_internal = 1;
3327 if (flags & SCF_DO_STCLASS_OR)
3328 cl_anything(pRExC_state, data->start_class);
3329 flags &= ~SCF_DO_STCLASS;
3333 else if (strchr((const char*)PL_simple,OP(scan))) {
3336 if (flags & SCF_DO_SUBSTR) {
3337 SCAN_COMMIT(pRExC_state,data,minlenp);
3341 if (flags & SCF_DO_STCLASS) {
3342 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3344 /* Some of the logic below assumes that switching
3345 locale on will only add false positives. */
3346 switch (PL_regkind[OP(scan)]) {
3350 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3351 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3352 cl_anything(pRExC_state, data->start_class);
3355 if (OP(scan) == SANY)
3357 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3358 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3359 || (data->start_class->flags & ANYOF_CLASS));
3360 cl_anything(pRExC_state, data->start_class);
3362 if (flags & SCF_DO_STCLASS_AND || !value)
3363 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3366 if (flags & SCF_DO_STCLASS_AND)
3367 cl_and(data->start_class,
3368 (struct regnode_charclass_class*)scan);
3370 cl_or(pRExC_state, data->start_class,
3371 (struct regnode_charclass_class*)scan);
3374 if (flags & SCF_DO_STCLASS_AND) {
3375 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3376 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3377 for (value = 0; value < 256; value++)
3378 if (!isALNUM(value))
3379 ANYOF_BITMAP_CLEAR(data->start_class, value);
3383 if (data->start_class->flags & ANYOF_LOCALE)
3384 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3386 for (value = 0; value < 256; value++)
3388 ANYOF_BITMAP_SET(data->start_class, value);
3393 if (flags & SCF_DO_STCLASS_AND) {
3394 if (data->start_class->flags & ANYOF_LOCALE)
3395 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3398 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3399 data->start_class->flags |= ANYOF_LOCALE;
3403 if (flags & SCF_DO_STCLASS_AND) {
3404 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3405 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3406 for (value = 0; value < 256; value++)
3408 ANYOF_BITMAP_CLEAR(data->start_class, value);
3412 if (data->start_class->flags & ANYOF_LOCALE)
3413 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3415 for (value = 0; value < 256; value++)
3416 if (!isALNUM(value))
3417 ANYOF_BITMAP_SET(data->start_class, value);
3422 if (flags & SCF_DO_STCLASS_AND) {
3423 if (data->start_class->flags & ANYOF_LOCALE)
3424 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3427 data->start_class->flags |= ANYOF_LOCALE;
3428 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3432 if (flags & SCF_DO_STCLASS_AND) {
3433 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3434 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3435 for (value = 0; value < 256; value++)
3436 if (!isSPACE(value))
3437 ANYOF_BITMAP_CLEAR(data->start_class, value);
3441 if (data->start_class->flags & ANYOF_LOCALE)
3442 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3444 for (value = 0; value < 256; value++)
3446 ANYOF_BITMAP_SET(data->start_class, value);
3451 if (flags & SCF_DO_STCLASS_AND) {
3452 if (data->start_class->flags & ANYOF_LOCALE)
3453 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3456 data->start_class->flags |= ANYOF_LOCALE;
3457 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3461 if (flags & SCF_DO_STCLASS_AND) {
3462 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3463 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3464 for (value = 0; value < 256; value++)
3466 ANYOF_BITMAP_CLEAR(data->start_class, value);
3470 if (data->start_class->flags & ANYOF_LOCALE)
3471 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3473 for (value = 0; value < 256; value++)
3474 if (!isSPACE(value))
3475 ANYOF_BITMAP_SET(data->start_class, value);
3480 if (flags & SCF_DO_STCLASS_AND) {
3481 if (data->start_class->flags & ANYOF_LOCALE) {
3482 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3483 for (value = 0; value < 256; value++)
3484 if (!isSPACE(value))
3485 ANYOF_BITMAP_CLEAR(data->start_class, value);
3489 data->start_class->flags |= ANYOF_LOCALE;
3490 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3494 if (flags & SCF_DO_STCLASS_AND) {
3495 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3496 for (value = 0; value < 256; value++)
3497 if (!isDIGIT(value))
3498 ANYOF_BITMAP_CLEAR(data->start_class, value);
3501 if (data->start_class->flags & ANYOF_LOCALE)
3502 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3504 for (value = 0; value < 256; value++)
3506 ANYOF_BITMAP_SET(data->start_class, value);
3511 if (flags & SCF_DO_STCLASS_AND) {
3512 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3513 for (value = 0; value < 256; value++)
3515 ANYOF_BITMAP_CLEAR(data->start_class, value);
3518 if (data->start_class->flags & ANYOF_LOCALE)
3519 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3521 for (value = 0; value < 256; value++)
3522 if (!isDIGIT(value))
3523 ANYOF_BITMAP_SET(data->start_class, value);
3528 if (flags & SCF_DO_STCLASS_OR)
3529 cl_and(data->start_class, and_withp);
3530 flags &= ~SCF_DO_STCLASS;
3533 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3534 data->flags |= (OP(scan) == MEOL
3538 else if ( PL_regkind[OP(scan)] == BRANCHJ
3539 /* Lookbehind, or need to calculate parens/evals/stclass: */
3540 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3541 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3542 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3543 || OP(scan) == UNLESSM )
3545 /* Negative Lookahead/lookbehind
3546 In this case we can't do fixed string optimisation.
3549 I32 deltanext, minnext, fake = 0;
3551 struct regnode_charclass_class intrnl;
3554 data_fake.flags = 0;
3556 data_fake.whilem_c = data->whilem_c;
3557 data_fake.last_closep = data->last_closep;
3560 data_fake.last_closep = &fake;
3561 data_fake.pos_delta = delta;
3562 if ( flags & SCF_DO_STCLASS && !scan->flags
3563 && OP(scan) == IFMATCH ) { /* Lookahead */
3564 cl_init(pRExC_state, &intrnl);
3565 data_fake.start_class = &intrnl;
3566 f |= SCF_DO_STCLASS_AND;
3568 if (flags & SCF_WHILEM_VISITED_POS)
3569 f |= SCF_WHILEM_VISITED_POS;
3570 next = regnext(scan);
3571 nscan = NEXTOPER(NEXTOPER(scan));
3572 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3573 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3576 FAIL("Variable length lookbehind not implemented");
3578 else if (minnext > (I32)U8_MAX) {
3579 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3581 scan->flags = (U8)minnext;
3584 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3586 if (data_fake.flags & SF_HAS_EVAL)
3587 data->flags |= SF_HAS_EVAL;
3588 data->whilem_c = data_fake.whilem_c;
3590 if (f & SCF_DO_STCLASS_AND) {
3591 const int was = (data->start_class->flags & ANYOF_EOS);
3593 cl_and(data->start_class, &intrnl);
3595 data->start_class->flags |= ANYOF_EOS;
3598 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3600 /* Positive Lookahead/lookbehind
3601 In this case we can do fixed string optimisation,
3602 but we must be careful about it. Note in the case of
3603 lookbehind the positions will be offset by the minimum
3604 length of the pattern, something we won't know about
3605 until after the recurse.
3607 I32 deltanext, fake = 0;
3609 struct regnode_charclass_class intrnl;
3611 /* We use SAVEFREEPV so that when the full compile
3612 is finished perl will clean up the allocated
3613 minlens when its all done. This was we don't
3614 have to worry about freeing them when we know
3615 they wont be used, which would be a pain.
3618 Newx( minnextp, 1, I32 );
3619 SAVEFREEPV(minnextp);
3622 StructCopy(data, &data_fake, scan_data_t);
3623 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3626 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3627 data_fake.last_found=newSVsv(data->last_found);
3631 data_fake.last_closep = &fake;
3632 data_fake.flags = 0;
3633 data_fake.pos_delta = delta;
3635 data_fake.flags |= SF_IS_INF;
3636 if ( flags & SCF_DO_STCLASS && !scan->flags
3637 && OP(scan) == IFMATCH ) { /* Lookahead */
3638 cl_init(pRExC_state, &intrnl);
3639 data_fake.start_class = &intrnl;
3640 f |= SCF_DO_STCLASS_AND;
3642 if (flags & SCF_WHILEM_VISITED_POS)
3643 f |= SCF_WHILEM_VISITED_POS;
3644 next = regnext(scan);
3645 nscan = NEXTOPER(NEXTOPER(scan));
3647 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3648 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3651 FAIL("Variable length lookbehind not implemented");
3653 else if (*minnextp > (I32)U8_MAX) {
3654 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3656 scan->flags = (U8)*minnextp;
3661 if (f & SCF_DO_STCLASS_AND) {
3662 const int was = (data->start_class->flags & ANYOF_EOS);
3664 cl_and(data->start_class, &intrnl);
3666 data->start_class->flags |= ANYOF_EOS;
3669 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3671 if (data_fake.flags & SF_HAS_EVAL)
3672 data->flags |= SF_HAS_EVAL;
3673 data->whilem_c = data_fake.whilem_c;
3674 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3675 if (RExC_rx->minlen<*minnextp)
3676 RExC_rx->minlen=*minnextp;
3677 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3678 SvREFCNT_dec(data_fake.last_found);
3680 if ( data_fake.minlen_fixed != minlenp )
3682 data->offset_fixed= data_fake.offset_fixed;
3683 data->minlen_fixed= data_fake.minlen_fixed;
3684 data->lookbehind_fixed+= scan->flags;
3686 if ( data_fake.minlen_float != minlenp )
3688 data->minlen_float= data_fake.minlen_float;
3689 data->offset_float_min=data_fake.offset_float_min;
3690 data->offset_float_max=data_fake.offset_float_max;
3691 data->lookbehind_float+= scan->flags;
3700 else if (OP(scan) == OPEN) {
3701 if (stopparen != (I32)ARG(scan))
3704 else if (OP(scan) == CLOSE) {
3705 if (stopparen == (I32)ARG(scan)) {
3708 if ((I32)ARG(scan) == is_par) {
3709 next = regnext(scan);
3711 if ( next && (OP(next) != WHILEM) && next < last)
3712 is_par = 0; /* Disable optimization */
3715 *(data->last_closep) = ARG(scan);
3717 else if (OP(scan) == EVAL) {
3719 data->flags |= SF_HAS_EVAL;
3721 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3722 if (flags & SCF_DO_SUBSTR) {
3723 SCAN_COMMIT(pRExC_state,data,minlenp);
3724 flags &= ~SCF_DO_SUBSTR;
3726 if (data && OP(scan)==ACCEPT) {
3727 data->flags |= SCF_SEEN_ACCEPT;
3732 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3734 if (flags & SCF_DO_SUBSTR) {
3735 SCAN_COMMIT(pRExC_state,data,minlenp);
3736 data->longest = &(data->longest_float);
3738 is_inf = is_inf_internal = 1;
3739 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3740 cl_anything(pRExC_state, data->start_class);
3741 flags &= ~SCF_DO_STCLASS;
3743 else if (OP(scan) == GPOS) {
3744 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3745 !(delta || is_inf || (data && data->pos_delta)))
3747 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3748 RExC_rx->extflags |= RXf_ANCH_GPOS;
3749 if (RExC_rx->gofs < (U32)min)
3750 RExC_rx->gofs = min;
3752 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3756 #ifdef TRIE_STUDY_OPT
3757 #ifdef FULL_TRIE_STUDY
3758 else if (PL_regkind[OP(scan)] == TRIE) {
3759 /* NOTE - There is similar code to this block above for handling
3760 BRANCH nodes on the initial study. If you change stuff here
3762 regnode *trie_node= scan;
3763 regnode *tail= regnext(scan);
3764 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3765 I32 max1 = 0, min1 = I32_MAX;
3766 struct regnode_charclass_class accum;
3768 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3769 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3770 if (flags & SCF_DO_STCLASS)
3771 cl_init_zero(pRExC_state, &accum);
3777 const regnode *nextbranch= NULL;
3780 for ( word=1 ; word <= trie->wordcount ; word++)
3782 I32 deltanext=0, minnext=0, f = 0, fake;
3783 struct regnode_charclass_class this_class;
3785 data_fake.flags = 0;
3787 data_fake.whilem_c = data->whilem_c;
3788 data_fake.last_closep = data->last_closep;
3791 data_fake.last_closep = &fake;
3792 data_fake.pos_delta = delta;
3793 if (flags & SCF_DO_STCLASS) {
3794 cl_init(pRExC_state, &this_class);
3795 data_fake.start_class = &this_class;
3796 f = SCF_DO_STCLASS_AND;
3798 if (flags & SCF_WHILEM_VISITED_POS)
3799 f |= SCF_WHILEM_VISITED_POS;
3801 if (trie->jump[word]) {
3803 nextbranch = trie_node + trie->jump[0];
3804 scan= trie_node + trie->jump[word];
3805 /* We go from the jump point to the branch that follows
3806 it. Note this means we need the vestigal unused branches
3807 even though they arent otherwise used.
3809 minnext = study_chunk(pRExC_state, &scan, minlenp,
3810 &deltanext, (regnode *)nextbranch, &data_fake,
3811 stopparen, recursed, NULL, f,depth+1);
3813 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3814 nextbranch= regnext((regnode*)nextbranch);
3816 if (min1 > (I32)(minnext + trie->minlen))
3817 min1 = minnext + trie->minlen;
3818 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3819 max1 = minnext + deltanext + trie->maxlen;
3820 if (deltanext == I32_MAX)
3821 is_inf = is_inf_internal = 1;
3823 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3825 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3826 if ( stopmin > min + min1)
3827 stopmin = min + min1;
3828 flags &= ~SCF_DO_SUBSTR;
3830 data->flags |= SCF_SEEN_ACCEPT;
3833 if (data_fake.flags & SF_HAS_EVAL)
3834 data->flags |= SF_HAS_EVAL;
3835 data->whilem_c = data_fake.whilem_c;
3837 if (flags & SCF_DO_STCLASS)
3838 cl_or(pRExC_state, &accum, &this_class);
3841 if (flags & SCF_DO_SUBSTR) {
3842 data->pos_min += min1;
3843 data->pos_delta += max1 - min1;
3844 if (max1 != min1 || is_inf)
3845 data->longest = &(data->longest_float);
3848 delta += max1 - min1;
3849 if (flags & SCF_DO_STCLASS_OR) {
3850 cl_or(pRExC_state, data->start_class, &accum);
3852 cl_and(data->start_class, and_withp);
3853 flags &= ~SCF_DO_STCLASS;
3856 else if (flags & SCF_DO_STCLASS_AND) {
3858 cl_and(data->start_class, &accum);
3859 flags &= ~SCF_DO_STCLASS;
3862 /* Switch to OR mode: cache the old value of
3863 * data->start_class */
3865 StructCopy(data->start_class, and_withp,
3866 struct regnode_charclass_class);
3867 flags &= ~SCF_DO_STCLASS_AND;
3868 StructCopy(&accum, data->start_class,
3869 struct regnode_charclass_class);
3870 flags |= SCF_DO_STCLASS_OR;
3871 data->start_class->flags |= ANYOF_EOS;
3878 else if (PL_regkind[OP(scan)] == TRIE) {
3879 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3882 min += trie->minlen;
3883 delta += (trie->maxlen - trie->minlen);
3884 flags &= ~SCF_DO_STCLASS; /* xxx */
3885 if (flags & SCF_DO_SUBSTR) {
3886 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3887 data->pos_min += trie->minlen;
3888 data->pos_delta += (trie->maxlen - trie->minlen);
3889 if (trie->maxlen != trie->minlen)
3890 data->longest = &(data->longest_float);
3892 if (trie->jump) /* no more substrings -- for now /grr*/
3893 flags &= ~SCF_DO_SUBSTR;
3895 #endif /* old or new */
3896 #endif /* TRIE_STUDY_OPT */
3897 /* Else: zero-length, ignore. */
3898 scan = regnext(scan);
3903 stopparen = frame->stop;
3904 frame = frame->prev;
3905 goto fake_study_recurse;
3910 DEBUG_STUDYDATA("pre-fin:",data,depth);
3913 *deltap = is_inf_internal ? I32_MAX : delta;
3914 if (flags & SCF_DO_SUBSTR && is_inf)
3915 data->pos_delta = I32_MAX - data->pos_min;
3916 if (is_par > (I32)U8_MAX)
3918 if (is_par && pars==1 && data) {
3919 data->flags |= SF_IN_PAR;
3920 data->flags &= ~SF_HAS_PAR;
3922 else if (pars && data) {
3923 data->flags |= SF_HAS_PAR;
3924 data->flags &= ~SF_IN_PAR;
3926 if (flags & SCF_DO_STCLASS_OR)
3927 cl_and(data->start_class, and_withp);
3928 if (flags & SCF_TRIE_RESTUDY)
3929 data->flags |= SCF_TRIE_RESTUDY;
3931 DEBUG_STUDYDATA("post-fin:",data,depth);
3933 return min < stopmin ? min : stopmin;
3937 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
3939 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
3941 Renewc(RExC_rxi->data,
3942 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
3943 char, struct reg_data);
3945 Renew(RExC_rxi->data->what, count + n, U8);
3947 Newx(RExC_rxi->data->what, n, U8);
3948 RExC_rxi->data->count = count + n;
3949 Copy(s, RExC_rxi->data->what + count, n, U8);
3953 /*XXX: todo make this not included in a non debugging perl */
3954 #ifndef PERL_IN_XSUB_RE
3956 Perl_reginitcolors(pTHX)
3959 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
3961 char *t = savepv(s);
3965 t = strchr(t, '\t');
3971 PL_colors[i] = t = (char *)"";
3976 PL_colors[i++] = (char *)"";
3983 #ifdef TRIE_STUDY_OPT
3984 #define CHECK_RESTUDY_GOTO \
3986 (data.flags & SCF_TRIE_RESTUDY) \
3990 #define CHECK_RESTUDY_GOTO
3994 - pregcomp - compile a regular expression into internal code
3996 * We can't allocate space until we know how big the compiled form will be,
3997 * but we can't compile it (and thus know how big it is) until we've got a
3998 * place to put the code. So we cheat: we compile it twice, once with code
3999 * generation turned off and size counting turned on, and once "for real".
4000 * This also means that we don't allocate space until we are sure that the
4001 * thing really will compile successfully, and we never have to move the
4002 * code and thus invalidate pointers into it. (Note that it has to be in
4003 * one piece because free() must be able to free it all.) [NB: not true in perl]
4005 * Beware that the optimization-preparation code in here knows about some
4006 * of the structure of the compiled regexp. [I'll say.]
4011 #ifndef PERL_IN_XSUB_RE
4012 #define RE_ENGINE_PTR &PL_core_reg_engine
4014 extern const struct regexp_engine my_reg_engine;
4015 #define RE_ENGINE_PTR &my_reg_engine
4018 #ifndef PERL_IN_XSUB_RE
4020 Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
4023 HV * const table = GvHV(PL_hintgv);
4024 /* Dispatch a request to compile a regexp to correct
4027 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4028 GET_RE_DEBUG_FLAGS_DECL;
4029 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4030 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4032 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4035 return CALLREGCOMP_ENG(eng, exp, xend, pm);
4038 return Perl_re_compile(aTHX_ exp, xend, pm);
4043 Perl_re_compile(pTHX_ char *exp, char *xend, PMOP *pm)
4047 register regexp_internal *ri;
4055 RExC_state_t RExC_state;
4056 RExC_state_t * const pRExC_state = &RExC_state;
4057 #ifdef TRIE_STUDY_OPT
4059 RExC_state_t copyRExC_state;
4061 GET_RE_DEBUG_FLAGS_DECL;
4062 DEBUG_r(if (!PL_colorset) reginitcolors());
4065 FAIL("NULL regexp argument");
4067 RExC_utf8 = RExC_orig_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8;
4070 SV *dsv= sv_newmortal();
4071 RE_PV_QUOTED_DECL(s, RExC_utf8,
4072 dsv, exp, (xend - exp), 60);
4073 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4074 PL_colors[4],PL_colors[5],s);
4079 RExC_flags = pm->op_pmflags;
4083 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4084 RExC_seen_evals = 0;
4087 /* First pass: determine size, legality. */
4095 RExC_emit = &PL_regdummy;
4096 RExC_whilem_seen = 0;
4097 RExC_charnames = NULL;
4098 RExC_open_parens = NULL;
4099 RExC_close_parens = NULL;
4101 RExC_paren_names = NULL;
4103 RExC_paren_name_list = NULL;
4105 RExC_recurse = NULL;
4106 RExC_recurse_count = 0;
4108 #if 0 /* REGC() is (currently) a NOP at the first pass.
4109 * Clever compilers notice this and complain. --jhi */
4110 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4112 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4113 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4114 RExC_precomp = NULL;
4117 if (RExC_utf8 && !RExC_orig_utf8) {
4118 /* It's possible to write a regexp in ascii that represents unicode
4119 codepoints outside of the byte range, such as via \x{100}. If we
4120 detect such a sequence we have to convert the entire pattern to utf8
4121 and then recompile, as our sizing calculation will have been based
4122 on 1 byte == 1 character, but we will need to use utf8 to encode
4123 at least some part of the pattern, and therefore must convert the whole
4125 XXX: somehow figure out how to make this less expensive...
4127 STRLEN len = xend-exp;
4128 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4129 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4130 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4132 RExC_orig_utf8 = RExC_utf8;
4134 goto redo_first_pass;
4137 PerlIO_printf(Perl_debug_log,
4138 "Required size %"IVdf" nodes\n"
4139 "Starting second pass (creation)\n",
4142 RExC_lastparse=NULL;
4144 /* Small enough for pointer-storage convention?
4145 If extralen==0, this means that we will not need long jumps. */
4146 if (RExC_size >= 0x10000L && RExC_extralen)
4147 RExC_size += RExC_extralen;
4150 if (RExC_whilem_seen > 15)
4151 RExC_whilem_seen = 15;
4153 /* Allocate space and zero-initialize. Note, the two step process
4154 of zeroing when in debug mode, thus anything assigned has to
4155 happen after that */
4156 Newxz(r, 1, regexp);
4157 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4158 char, regexp_internal);
4159 if ( r == NULL || ri == NULL )
4160 FAIL("Regexp out of space");
4162 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4163 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4165 /* bulk initialize base fields with 0. */
4166 Zero(ri, sizeof(regexp_internal), char);
4169 /* non-zero initialization begins here */
4171 r->engine= RE_ENGINE_PTR;
4173 r->prelen = xend - exp;
4174 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
4176 bool has_k = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4177 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4178 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4179 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) >> 12);
4180 const char *fptr = STD_PAT_MODS; /*"msix"*/
4182 r->wraplen = r->prelen + has_minus + has_k + has_runon
4183 + (sizeof(STD_PAT_MODS) - 1)
4184 + (sizeof("(?:)") - 1);
4186 Newx(r->wrapped, r->wraplen + 1, char );
4190 *p++ = KEEPCOPY_PAT_MOD; /*'k'*/
4192 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4193 char *colon = r + 1;
4196 while((ch = *fptr++)) {
4210 Copy(RExC_precomp, p, r->prelen, char);
4220 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4222 if (RExC_seen & REG_SEEN_RECURSE) {
4223 Newxz(RExC_open_parens, RExC_npar,regnode *);
4224 SAVEFREEPV(RExC_open_parens);
4225 Newxz(RExC_close_parens,RExC_npar,regnode *);
4226 SAVEFREEPV(RExC_close_parens);
4229 /* Useful during FAIL. */
4230 #ifdef RE_TRACK_PATTERN_OFFSETS
4231 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4232 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4233 "%s %"UVuf" bytes for offset annotations.\n",
4234 ri->u.offsets ? "Got" : "Couldn't get",
4235 (UV)((2*RExC_size+1) * sizeof(U32))));
4237 SetProgLen(ri,RExC_size);
4241 /* Second pass: emit code. */
4242 RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
4247 RExC_emit_start = ri->program;
4248 RExC_emit = ri->program;
4249 RExC_emit_bound = ri->program + RExC_size + 1;
4251 /* Store the count of eval-groups for security checks: */
4252 RExC_rx->seen_evals = RExC_seen_evals;
4253 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4254 if (reg(pRExC_state, 0, &flags,1) == NULL)
4257 /* XXXX To minimize changes to RE engine we always allocate
4258 3-units-long substrs field. */
4259 Newx(r->substrs, 1, struct reg_substr_data);
4260 if (RExC_recurse_count) {
4261 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4262 SAVEFREEPV(RExC_recurse);
4266 r->minlen = minlen = sawplus = sawopen = 0;
4267 Zero(r->substrs, 1, struct reg_substr_data);
4269 #ifdef TRIE_STUDY_OPT
4272 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4274 RExC_state = copyRExC_state;
4275 if (seen & REG_TOP_LEVEL_BRANCHES)
4276 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4278 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4279 if (data.last_found) {
4280 SvREFCNT_dec(data.longest_fixed);
4281 SvREFCNT_dec(data.longest_float);
4282 SvREFCNT_dec(data.last_found);
4284 StructCopy(&zero_scan_data, &data, scan_data_t);
4286 StructCopy(&zero_scan_data, &data, scan_data_t);
4287 copyRExC_state = RExC_state;
4290 StructCopy(&zero_scan_data, &data, scan_data_t);
4293 /* Dig out information for optimizations. */
4294 r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
4295 pm->op_pmflags = RExC_flags;
4297 r->extflags |= RXf_UTF8; /* Unicode in it? */
4298 ri->regstclass = NULL;
4299 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4300 r->intflags |= PREGf_NAUGHTY;
4301 scan = ri->program + 1; /* First BRANCH. */
4303 /* testing for BRANCH here tells us whether there is "must appear"
4304 data in the pattern. If there is then we can use it for optimisations */
4305 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4307 STRLEN longest_float_length, longest_fixed_length;
4308 struct regnode_charclass_class ch_class; /* pointed to by data */
4310 I32 last_close = 0; /* pointed to by data */
4313 /* Skip introductions and multiplicators >= 1. */
4314 while ((OP(first) == OPEN && (sawopen = 1)) ||
4315 /* An OR of *one* alternative - should not happen now. */
4316 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4317 /* for now we can't handle lookbehind IFMATCH*/
4318 (OP(first) == IFMATCH && !first->flags) ||
4319 (OP(first) == PLUS) ||
4320 (OP(first) == MINMOD) ||
4321 /* An {n,m} with n>0 */
4322 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4325 if (OP(first) == PLUS)
4328 first += regarglen[OP(first)];
4329 if (OP(first) == IFMATCH) {
4330 first = NEXTOPER(first);
4331 first += EXTRA_STEP_2ARGS;
4332 } else /* XXX possible optimisation for /(?=)/ */
4333 first = NEXTOPER(first);
4336 /* Starting-point info. */
4338 DEBUG_PEEP("first:",first,0);
4339 /* Ignore EXACT as we deal with it later. */
4340 if (PL_regkind[OP(first)] == EXACT) {
4341 if (OP(first) == EXACT)
4342 NOOP; /* Empty, get anchored substr later. */
4343 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4344 ri->regstclass = first;
4347 else if (PL_regkind[OP(first)] == TRIE &&
4348 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4351 /* this can happen only on restudy */
4352 if ( OP(first) == TRIE ) {
4353 struct regnode_1 *trieop = (struct regnode_1 *)
4354 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4355 StructCopy(first,trieop,struct regnode_1);
4356 trie_op=(regnode *)trieop;
4358 struct regnode_charclass *trieop = (struct regnode_charclass *)
4359 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4360 StructCopy(first,trieop,struct regnode_charclass);
4361 trie_op=(regnode *)trieop;
4364 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4365 ri->regstclass = trie_op;
4368 else if (strchr((const char*)PL_simple,OP(first)))
4369 ri->regstclass = first;
4370 else if (PL_regkind[OP(first)] == BOUND ||
4371 PL_regkind[OP(first)] == NBOUND)
4372 ri->regstclass = first;
4373 else if (PL_regkind[OP(first)] == BOL) {
4374 r->extflags |= (OP(first) == MBOL
4376 : (OP(first) == SBOL
4379 first = NEXTOPER(first);
4382 else if (OP(first) == GPOS) {
4383 r->extflags |= RXf_ANCH_GPOS;
4384 first = NEXTOPER(first);
4387 else if ((!sawopen || !RExC_sawback) &&
4388 (OP(first) == STAR &&
4389 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4390 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4392 /* turn .* into ^.* with an implied $*=1 */
4394 (OP(NEXTOPER(first)) == REG_ANY)
4397 r->extflags |= type;
4398 r->intflags |= PREGf_IMPLICIT;
4399 first = NEXTOPER(first);
4402 if (sawplus && (!sawopen || !RExC_sawback)
4403 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4404 /* x+ must match at the 1st pos of run of x's */
4405 r->intflags |= PREGf_SKIP;
4407 /* Scan is after the zeroth branch, first is atomic matcher. */
4408 #ifdef TRIE_STUDY_OPT
4411 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4412 (IV)(first - scan + 1))
4416 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4417 (IV)(first - scan + 1))
4423 * If there's something expensive in the r.e., find the
4424 * longest literal string that must appear and make it the
4425 * regmust. Resolve ties in favor of later strings, since
4426 * the regstart check works with the beginning of the r.e.
4427 * and avoiding duplication strengthens checking. Not a
4428 * strong reason, but sufficient in the absence of others.
4429 * [Now we resolve ties in favor of the earlier string if
4430 * it happens that c_offset_min has been invalidated, since the
4431 * earlier string may buy us something the later one won't.]
4434 data.longest_fixed = newSVpvs("");
4435 data.longest_float = newSVpvs("");
4436 data.last_found = newSVpvs("");
4437 data.longest = &(data.longest_fixed);
4439 if (!ri->regstclass) {
4440 cl_init(pRExC_state, &ch_class);
4441 data.start_class = &ch_class;
4442 stclass_flag = SCF_DO_STCLASS_AND;
4443 } else /* XXXX Check for BOUND? */
4445 data.last_closep = &last_close;
4447 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4448 &data, -1, NULL, NULL,
4449 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4455 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4456 && data.last_start_min == 0 && data.last_end > 0
4457 && !RExC_seen_zerolen
4458 && !(RExC_seen & REG_SEEN_VERBARG)
4459 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4460 r->extflags |= RXf_CHECK_ALL;
4461 scan_commit(pRExC_state, &data,&minlen,0);
4462 SvREFCNT_dec(data.last_found);
4464 /* Note that code very similar to this but for anchored string
4465 follows immediately below, changes may need to be made to both.
4468 longest_float_length = CHR_SVLEN(data.longest_float);
4469 if (longest_float_length
4470 || (data.flags & SF_FL_BEFORE_EOL
4471 && (!(data.flags & SF_FL_BEFORE_MEOL)
4472 || (RExC_flags & RXf_PMf_MULTILINE))))
4476 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4477 && data.offset_fixed == data.offset_float_min
4478 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4479 goto remove_float; /* As in (a)+. */
4481 /* copy the information about the longest float from the reg_scan_data
4482 over to the program. */
4483 if (SvUTF8(data.longest_float)) {
4484 r->float_utf8 = data.longest_float;
4485 r->float_substr = NULL;
4487 r->float_substr = data.longest_float;
4488 r->float_utf8 = NULL;
4490 /* float_end_shift is how many chars that must be matched that
4491 follow this item. We calculate it ahead of time as once the
4492 lookbehind offset is added in we lose the ability to correctly
4494 ml = data.minlen_float ? *(data.minlen_float)
4495 : (I32)longest_float_length;
4496 r->float_end_shift = ml - data.offset_float_min
4497 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4498 + data.lookbehind_float;
4499 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4500 r->float_max_offset = data.offset_float_max;
4501 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4502 r->float_max_offset -= data.lookbehind_float;
4504 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4505 && (!(data.flags & SF_FL_BEFORE_MEOL)
4506 || (RExC_flags & RXf_PMf_MULTILINE)));
4507 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4511 r->float_substr = r->float_utf8 = NULL;
4512 SvREFCNT_dec(data.longest_float);
4513 longest_float_length = 0;
4516 /* Note that code very similar to this but for floating string
4517 is immediately above, changes may need to be made to both.
4520 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4521 if (longest_fixed_length
4522 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4523 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4524 || (RExC_flags & RXf_PMf_MULTILINE))))
4528 /* copy the information about the longest fixed
4529 from the reg_scan_data over to the program. */
4530 if (SvUTF8(data.longest_fixed)) {
4531 r->anchored_utf8 = data.longest_fixed;
4532 r->anchored_substr = NULL;
4534 r->anchored_substr = data.longest_fixed;
4535 r->anchored_utf8 = NULL;
4537 /* fixed_end_shift is how many chars that must be matched that
4538 follow this item. We calculate it ahead of time as once the
4539 lookbehind offset is added in we lose the ability to correctly
4541 ml = data.minlen_fixed ? *(data.minlen_fixed)
4542 : (I32)longest_fixed_length;
4543 r->anchored_end_shift = ml - data.offset_fixed
4544 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4545 + data.lookbehind_fixed;
4546 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4548 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4549 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4550 || (RExC_flags & RXf_PMf_MULTILINE)));
4551 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4554 r->anchored_substr = r->anchored_utf8 = NULL;
4555 SvREFCNT_dec(data.longest_fixed);
4556 longest_fixed_length = 0;
4559 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4560 ri->regstclass = NULL;
4561 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4563 && !(data.start_class->flags & ANYOF_EOS)
4564 && !cl_is_anything(data.start_class))
4566 const U32 n = add_data(pRExC_state, 1, "f");
4568 Newx(RExC_rxi->data->data[n], 1,
4569 struct regnode_charclass_class);
4570 StructCopy(data.start_class,
4571 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4572 struct regnode_charclass_class);
4573 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4574 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4575 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4576 regprop(r, sv, (regnode*)data.start_class);
4577 PerlIO_printf(Perl_debug_log,
4578 "synthetic stclass \"%s\".\n",
4579 SvPVX_const(sv));});
4582 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4583 if (longest_fixed_length > longest_float_length) {
4584 r->check_end_shift = r->anchored_end_shift;
4585 r->check_substr = r->anchored_substr;
4586 r->check_utf8 = r->anchored_utf8;
4587 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4588 if (r->extflags & RXf_ANCH_SINGLE)
4589 r->extflags |= RXf_NOSCAN;
4592 r->check_end_shift = r->float_end_shift;
4593 r->check_substr = r->float_substr;
4594 r->check_utf8 = r->float_utf8;
4595 r->check_offset_min = r->float_min_offset;
4596 r->check_offset_max = r->float_max_offset;
4598 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4599 This should be changed ASAP! */
4600 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4601 r->extflags |= RXf_USE_INTUIT;
4602 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4603 r->extflags |= RXf_INTUIT_TAIL;
4605 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4606 if ( (STRLEN)minlen < longest_float_length )
4607 minlen= longest_float_length;
4608 if ( (STRLEN)minlen < longest_fixed_length )
4609 minlen= longest_fixed_length;
4613 /* Several toplevels. Best we can is to set minlen. */
4615 struct regnode_charclass_class ch_class;
4618 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4620 scan = ri->program + 1;
4621 cl_init(pRExC_state, &ch_class);
4622 data.start_class = &ch_class;
4623 data.last_closep = &last_close;
4626 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4627 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4631 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4632 = r->float_substr = r->float_utf8 = NULL;
4633 if (!(data.start_class->flags & ANYOF_EOS)
4634 && !cl_is_anything(data.start_class))
4636 const U32 n = add_data(pRExC_state, 1, "f");
4638 Newx(RExC_rxi->data->data[n], 1,
4639 struct regnode_charclass_class);
4640 StructCopy(data.start_class,
4641 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4642 struct regnode_charclass_class);
4643 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4644 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4645 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4646 regprop(r, sv, (regnode*)data.start_class);
4647 PerlIO_printf(Perl_debug_log,
4648 "synthetic stclass \"%s\".\n",
4649 SvPVX_const(sv));});
4653 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4654 the "real" pattern. */
4656 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4657 (IV)minlen, (IV)r->minlen);
4659 r->minlenret = minlen;
4660 if (r->minlen < minlen)
4663 if (RExC_seen & REG_SEEN_GPOS)
4664 r->extflags |= RXf_GPOS_SEEN;
4665 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4666 r->extflags |= RXf_LOOKBEHIND_SEEN;
4667 if (RExC_seen & REG_SEEN_EVAL)
4668 r->extflags |= RXf_EVAL_SEEN;
4669 if (RExC_seen & REG_SEEN_CANY)
4670 r->extflags |= RXf_CANY_SEEN;
4671 if (RExC_seen & REG_SEEN_VERBARG)
4672 r->intflags |= PREGf_VERBARG_SEEN;
4673 if (RExC_seen & REG_SEEN_CUTGROUP)
4674 r->intflags |= PREGf_CUTGROUP_SEEN;
4675 if (RExC_paren_names)
4676 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4678 r->paren_names = NULL;
4679 if (r->prelen == 3 && strEQ("\\s+", r->precomp))
4680 r->extflags |= RXf_WHITE;
4681 else if (r->prelen == 1 && r->precomp[0] == '^')
4682 r->extflags |= RXf_START_ONLY;
4685 if (RExC_paren_names) {
4686 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4687 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4690 ri->name_list_idx = 0;
4692 if (RExC_recurse_count) {
4693 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4694 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4695 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4698 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4699 /* assume we don't need to swap parens around before we match */
4702 PerlIO_printf(Perl_debug_log,"Final program:\n");
4705 #ifdef RE_TRACK_PATTERN_OFFSETS
4706 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4707 const U32 len = ri->u.offsets[0];
4709 GET_RE_DEBUG_FLAGS_DECL;
4710 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4711 for (i = 1; i <= len; i++) {
4712 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4713 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4714 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4716 PerlIO_printf(Perl_debug_log, "\n");
4722 #undef RE_ENGINE_PTR
4726 Perl_reg_named_buff_get(pTHX_ const REGEXP * const rx, SV* namesv, U32 flags)
4728 AV *retarray = NULL;
4733 if (rx && rx->paren_names) {
4734 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4737 SV* sv_dat=HeVAL(he_str);
4738 I32 *nums=(I32*)SvPVX(sv_dat);
4739 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4740 if ((I32)(rx->nparens) >= nums[i]
4741 && rx->offs[nums[i]].start != -1
4742 && rx->offs[nums[i]].end != -1)
4744 ret = CALLREG_NUMBUF(rx,nums[i],NULL);
4748 ret = newSVsv(&PL_sv_undef);
4752 av_push(retarray, ret);
4756 return (SV*)retarray;
4763 Perl_reg_numbered_buff_get(pTHX_ const REGEXP * const rx, I32 paren, SV* usesv)
4768 SV *sv = usesv ? usesv : newSVpvs("");
4771 sv_setsv(sv,&PL_sv_undef);
4775 if (paren == -2 && rx->offs[0].start != -1) {
4777 i = rx->offs[0].start;
4781 if (paren == -1 && rx->offs[0].end != -1) {
4783 s = rx->subbeg + rx->offs[0].end;
4784 i = rx->sublen - rx->offs[0].end;
4787 if ( 0 <= paren && paren <= (I32)rx->nparens &&
4788 (s1 = rx->offs[paren].start) != -1 &&
4789 (t1 = rx->offs[paren].end) != -1)
4793 s = rx->subbeg + s1;
4795 sv_setsv(sv,&PL_sv_undef);
4798 assert(rx->sublen >= (s - rx->subbeg) + i );
4800 const int oldtainted = PL_tainted;
4802 sv_setpvn(sv, s, i);
4803 PL_tainted = oldtainted;
4804 if ( (rx->extflags & RXf_CANY_SEEN)
4805 ? (RX_MATCH_UTF8(rx)
4806 && (!i || is_utf8_string((U8*)s, i)))
4807 : (RX_MATCH_UTF8(rx)) )
4814 if (RX_MATCH_TAINTED(rx)) {
4815 if (SvTYPE(sv) >= SVt_PVMG) {
4816 MAGIC* const mg = SvMAGIC(sv);
4819 SvMAGIC_set(sv, mg->mg_moremagic);
4821 if ((mgt = SvMAGIC(sv))) {
4822 mg->mg_moremagic = mgt;
4823 SvMAGIC_set(sv, mg);
4833 sv_setsv(sv,&PL_sv_undef);
4839 /* Scans the name of a named buffer from the pattern.
4840 * If flags is REG_RSN_RETURN_NULL returns null.
4841 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4842 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4843 * to the parsed name as looked up in the RExC_paren_names hash.
4844 * If there is an error throws a vFAIL().. type exception.
4847 #define REG_RSN_RETURN_NULL 0
4848 #define REG_RSN_RETURN_NAME 1
4849 #define REG_RSN_RETURN_DATA 2
4852 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4853 char *name_start = RExC_parse;
4855 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
4856 /* skip IDFIRST by using do...while */
4859 RExC_parse += UTF8SKIP(RExC_parse);
4860 } while (isALNUM_utf8((U8*)RExC_parse));
4864 } while (isALNUM(*RExC_parse));
4868 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
4869 (int)(RExC_parse - name_start)));
4872 if ( flags == REG_RSN_RETURN_NAME)
4874 else if (flags==REG_RSN_RETURN_DATA) {
4877 if ( ! sv_name ) /* should not happen*/
4878 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
4879 if (RExC_paren_names)
4880 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
4882 sv_dat = HeVAL(he_str);
4884 vFAIL("Reference to nonexistent named group");
4888 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
4895 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
4896 int rem=(int)(RExC_end - RExC_parse); \
4905 if (RExC_lastparse!=RExC_parse) \
4906 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
4909 iscut ? "..." : "<" \
4912 PerlIO_printf(Perl_debug_log,"%16s",""); \
4915 num = RExC_size + 1; \
4917 num=REG_NODE_NUM(RExC_emit); \
4918 if (RExC_lastnum!=num) \
4919 PerlIO_printf(Perl_debug_log,"|%4d",num); \
4921 PerlIO_printf(Perl_debug_log,"|%4s",""); \
4922 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
4923 (int)((depth*2)), "", \
4927 RExC_lastparse=RExC_parse; \
4932 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
4933 DEBUG_PARSE_MSG((funcname)); \
4934 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
4936 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
4937 DEBUG_PARSE_MSG((funcname)); \
4938 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
4941 - reg - regular expression, i.e. main body or parenthesized thing
4943 * Caller must absorb opening parenthesis.
4945 * Combining parenthesis handling with the base level of regular expression
4946 * is a trifle forced, but the need to tie the tails of the branches to what
4947 * follows makes it hard to avoid.
4949 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
4951 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
4953 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
4957 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
4958 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
4961 register regnode *ret; /* Will be the head of the group. */
4962 register regnode *br;
4963 register regnode *lastbr;
4964 register regnode *ender = NULL;
4965 register I32 parno = 0;
4967 const I32 oregflags = RExC_flags;
4968 bool have_branch = 0;
4970 I32 freeze_paren = 0;
4971 I32 after_freeze = 0;
4973 /* for (?g), (?gc), and (?o) warnings; warning
4974 about (?c) will warn about (?g) -- japhy */
4976 #define WASTED_O 0x01
4977 #define WASTED_G 0x02
4978 #define WASTED_C 0x04
4979 #define WASTED_GC (0x02|0x04)
4980 I32 wastedflags = 0x00;
4982 char * parse_start = RExC_parse; /* MJD */
4983 char * const oregcomp_parse = RExC_parse;
4985 GET_RE_DEBUG_FLAGS_DECL;
4986 DEBUG_PARSE("reg ");
4988 *flagp = 0; /* Tentatively. */
4991 /* Make an OPEN node, if parenthesized. */
4993 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
4994 char *start_verb = RExC_parse;
4995 STRLEN verb_len = 0;
4996 char *start_arg = NULL;
4997 unsigned char op = 0;
4999 int internal_argval = 0; /* internal_argval is only useful if !argok */
5000 while ( *RExC_parse && *RExC_parse != ')' ) {
5001 if ( *RExC_parse == ':' ) {
5002 start_arg = RExC_parse + 1;
5008 verb_len = RExC_parse - start_verb;
5011 while ( *RExC_parse && *RExC_parse != ')' )
5013 if ( *RExC_parse != ')' )
5014 vFAIL("Unterminated verb pattern argument");
5015 if ( RExC_parse == start_arg )
5018 if ( *RExC_parse != ')' )
5019 vFAIL("Unterminated verb pattern");
5022 switch ( *start_verb ) {
5023 case 'A': /* (*ACCEPT) */
5024 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5026 internal_argval = RExC_nestroot;
5029 case 'C': /* (*COMMIT) */
5030 if ( memEQs(start_verb,verb_len,"COMMIT") )
5033 case 'F': /* (*FAIL) */
5034 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5039 case ':': /* (*:NAME) */
5040 case 'M': /* (*MARK:NAME) */
5041 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5046 case 'P': /* (*PRUNE) */
5047 if ( memEQs(start_verb,verb_len,"PRUNE") )
5050 case 'S': /* (*SKIP) */
5051 if ( memEQs(start_verb,verb_len,"SKIP") )
5054 case 'T': /* (*THEN) */
5055 /* [19:06] <TimToady> :: is then */
5056 if ( memEQs(start_verb,verb_len,"THEN") ) {
5058 RExC_seen |= REG_SEEN_CUTGROUP;
5064 vFAIL3("Unknown verb pattern '%.*s'",
5065 verb_len, start_verb);
5068 if ( start_arg && internal_argval ) {
5069 vFAIL3("Verb pattern '%.*s' may not have an argument",
5070 verb_len, start_verb);
5071 } else if ( argok < 0 && !start_arg ) {
5072 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5073 verb_len, start_verb);
5075 ret = reganode(pRExC_state, op, internal_argval);
5076 if ( ! internal_argval && ! SIZE_ONLY ) {
5078 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5079 ARG(ret) = add_data( pRExC_state, 1, "S" );
5080 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5087 if (!internal_argval)
5088 RExC_seen |= REG_SEEN_VERBARG;
5089 } else if ( start_arg ) {
5090 vFAIL3("Verb pattern '%.*s' may not have an argument",
5091 verb_len, start_verb);
5093 ret = reg_node(pRExC_state, op);
5095 nextchar(pRExC_state);
5098 if (*RExC_parse == '?') { /* (?...) */
5099 bool is_logical = 0;
5100 const char * const seqstart = RExC_parse;
5103 paren = *RExC_parse++;
5104 ret = NULL; /* For look-ahead/behind. */
5107 case 'P': /* (?P...) variants for those used to PCRE/Python */
5108 paren = *RExC_parse++;
5109 if ( paren == '<') /* (?P<...>) named capture */
5111 else if (paren == '>') { /* (?P>name) named recursion */
5112 goto named_recursion;
5114 else if (paren == '=') { /* (?P=...) named backref */
5115 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5116 you change this make sure you change that */
5117 char* name_start = RExC_parse;
5119 SV *sv_dat = reg_scan_name(pRExC_state,
5120 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5121 if (RExC_parse == name_start || *RExC_parse != ')')
5122 vFAIL2("Sequence %.3s... not terminated",parse_start);
5125 num = add_data( pRExC_state, 1, "S" );
5126 RExC_rxi->data->data[num]=(void*)sv_dat;
5127 SvREFCNT_inc(sv_dat);
5130 ret = reganode(pRExC_state,
5131 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5135 Set_Node_Offset(ret, parse_start+1);
5136 Set_Node_Cur_Length(ret); /* MJD */
5138 nextchar(pRExC_state);
5142 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5144 case '<': /* (?<...) */
5145 if (*RExC_parse == '!')
5147 else if (*RExC_parse != '=')
5153 case '\'': /* (?'...') */
5154 name_start= RExC_parse;
5155 svname = reg_scan_name(pRExC_state,
5156 SIZE_ONLY ? /* reverse test from the others */
5157 REG_RSN_RETURN_NAME :
5158 REG_RSN_RETURN_NULL);
5159 if (RExC_parse == name_start) {
5161 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5164 if (*RExC_parse != paren)
5165 vFAIL2("Sequence (?%c... not terminated",
5166 paren=='>' ? '<' : paren);
5170 if (!svname) /* shouldnt happen */
5172 "panic: reg_scan_name returned NULL");
5173 if (!RExC_paren_names) {
5174 RExC_paren_names= newHV();
5175 sv_2mortal((SV*)RExC_paren_names);
5177 RExC_paren_name_list= newAV();
5178 sv_2mortal((SV*)RExC_paren_name_list);
5181 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5183 sv_dat = HeVAL(he_str);
5185 /* croak baby croak */
5187 "panic: paren_name hash element allocation failed");
5188 } else if ( SvPOK(sv_dat) ) {
5189 /* (?|...) can mean we have dupes so scan to check
5190 its already been stored. Maybe a flag indicating
5191 we are inside such a construct would be useful,
5192 but the arrays are likely to be quite small, so
5193 for now we punt -- dmq */
5194 IV count = SvIV(sv_dat);
5195 I32 *pv = (I32*)SvPVX(sv_dat);
5197 for ( i = 0 ; i < count ; i++ ) {
5198 if ( pv[i] == RExC_npar ) {
5204 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5205 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5206 pv[count] = RExC_npar;
5210 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5211 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5216 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5217 SvREFCNT_dec(svname);
5220 /*sv_dump(sv_dat);*/
5222 nextchar(pRExC_state);
5224 goto capturing_parens;
5226 RExC_seen |= REG_SEEN_LOOKBEHIND;
5228 case '=': /* (?=...) */
5229 case '!': /* (?!...) */
5230 RExC_seen_zerolen++;
5231 if (*RExC_parse == ')') {
5232 ret=reg_node(pRExC_state, OPFAIL);
5233 nextchar(pRExC_state);
5237 case '|': /* (?|...) */
5238 /* branch reset, behave like a (?:...) except that
5239 buffers in alternations share the same numbers */
5241 after_freeze = freeze_paren = RExC_npar;
5243 case ':': /* (?:...) */
5244 case '>': /* (?>...) */
5246 case '$': /* (?$...) */
5247 case '@': /* (?@...) */
5248 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5250 case '#': /* (?#...) */
5251 while (*RExC_parse && *RExC_parse != ')')
5253 if (*RExC_parse != ')')
5254 FAIL("Sequence (?#... not terminated");
5255 nextchar(pRExC_state);
5258 case '0' : /* (?0) */
5259 case 'R' : /* (?R) */
5260 if (*RExC_parse != ')')
5261 FAIL("Sequence (?R) not terminated");
5262 ret = reg_node(pRExC_state, GOSTART);
5263 *flagp |= POSTPONED;
5264 nextchar(pRExC_state);
5267 { /* named and numeric backreferences */
5269 case '&': /* (?&NAME) */
5270 parse_start = RExC_parse - 1;
5273 SV *sv_dat = reg_scan_name(pRExC_state,
5274 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5275 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5277 goto gen_recurse_regop;
5280 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5282 vFAIL("Illegal pattern");
5284 goto parse_recursion;
5286 case '-': /* (?-1) */
5287 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5288 RExC_parse--; /* rewind to let it be handled later */
5292 case '1': case '2': case '3': case '4': /* (?1) */
5293 case '5': case '6': case '7': case '8': case '9':
5296 num = atoi(RExC_parse);
5297 parse_start = RExC_parse - 1; /* MJD */
5298 if (*RExC_parse == '-')
5300 while (isDIGIT(*RExC_parse))
5302 if (*RExC_parse!=')')
5303 vFAIL("Expecting close bracket");
5306 if ( paren == '-' ) {
5308 Diagram of capture buffer numbering.
5309 Top line is the normal capture buffer numbers
5310 Botton line is the negative indexing as from
5314 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5318 num = RExC_npar + num;
5321 vFAIL("Reference to nonexistent group");
5323 } else if ( paren == '+' ) {
5324 num = RExC_npar + num - 1;
5327 ret = reganode(pRExC_state, GOSUB, num);
5329 if (num > (I32)RExC_rx->nparens) {
5331 vFAIL("Reference to nonexistent group");
5333 ARG2L_SET( ret, RExC_recurse_count++);
5335 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5336 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5340 RExC_seen |= REG_SEEN_RECURSE;
5341 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5342 Set_Node_Offset(ret, parse_start); /* MJD */
5344 *flagp |= POSTPONED;
5345 nextchar(pRExC_state);
5347 } /* named and numeric backreferences */
5350 case '?': /* (??...) */
5352 if (*RExC_parse != '{') {
5354 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5357 *flagp |= POSTPONED;
5358 paren = *RExC_parse++;
5360 case '{': /* (?{...}) */
5365 char *s = RExC_parse;
5367 RExC_seen_zerolen++;
5368 RExC_seen |= REG_SEEN_EVAL;
5369 while (count && (c = *RExC_parse)) {
5380 if (*RExC_parse != ')') {
5382 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5386 OP_4tree *sop, *rop;
5387 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5390 Perl_save_re_context(aTHX);
5391 rop = sv_compile_2op(sv, &sop, "re", &pad);
5392 sop->op_private |= OPpREFCOUNTED;
5393 /* re_dup will OpREFCNT_inc */
5394 OpREFCNT_set(sop, 1);
5397 n = add_data(pRExC_state, 3, "nop");
5398 RExC_rxi->data->data[n] = (void*)rop;
5399 RExC_rxi->data->data[n+1] = (void*)sop;
5400 RExC_rxi->data->data[n+2] = (void*)pad;
5403 else { /* First pass */
5404 if (PL_reginterp_cnt < ++RExC_seen_evals
5406 /* No compiled RE interpolated, has runtime
5407 components ===> unsafe. */
5408 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5409 if (PL_tainting && PL_tainted)
5410 FAIL("Eval-group in insecure regular expression");
5411 #if PERL_VERSION > 8
5412 if (IN_PERL_COMPILETIME)
5417 nextchar(pRExC_state);
5419 ret = reg_node(pRExC_state, LOGICAL);
5422 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5423 /* deal with the length of this later - MJD */
5426 ret = reganode(pRExC_state, EVAL, n);
5427 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5428 Set_Node_Offset(ret, parse_start);
5431 case '(': /* (?(?{...})...) and (?(?=...)...) */
5434 if (RExC_parse[0] == '?') { /* (?(?...)) */
5435 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5436 || RExC_parse[1] == '<'
5437 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5440 ret = reg_node(pRExC_state, LOGICAL);
5443 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5447 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5448 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5450 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5451 char *name_start= RExC_parse++;
5453 SV *sv_dat=reg_scan_name(pRExC_state,
5454 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5455 if (RExC_parse == name_start || *RExC_parse != ch)
5456 vFAIL2("Sequence (?(%c... not terminated",
5457 (ch == '>' ? '<' : ch));
5460 num = add_data( pRExC_state, 1, "S" );
5461 RExC_rxi->data->data[num]=(void*)sv_dat;
5462 SvREFCNT_inc(sv_dat);
5464 ret = reganode(pRExC_state,NGROUPP,num);
5465 goto insert_if_check_paren;
5467 else if (RExC_parse[0] == 'D' &&
5468 RExC_parse[1] == 'E' &&
5469 RExC_parse[2] == 'F' &&
5470 RExC_parse[3] == 'I' &&
5471 RExC_parse[4] == 'N' &&
5472 RExC_parse[5] == 'E')
5474 ret = reganode(pRExC_state,DEFINEP,0);
5477 goto insert_if_check_paren;
5479 else if (RExC_parse[0] == 'R') {
5482 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5483 parno = atoi(RExC_parse++);
5484 while (isDIGIT(*RExC_parse))
5486 } else if (RExC_parse[0] == '&') {
5489 sv_dat = reg_scan_name(pRExC_state,
5490 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5491 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5493 ret = reganode(pRExC_state,INSUBP,parno);
5494 goto insert_if_check_paren;
5496 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5499 parno = atoi(RExC_parse++);
5501 while (isDIGIT(*RExC_parse))
5503 ret = reganode(pRExC_state, GROUPP, parno);
5505 insert_if_check_paren:
5506 if ((c = *nextchar(pRExC_state)) != ')')
5507 vFAIL("Switch condition not recognized");
5509 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5510 br = regbranch(pRExC_state, &flags, 1,depth+1);
5512 br = reganode(pRExC_state, LONGJMP, 0);
5514 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5515 c = *nextchar(pRExC_state);
5520 vFAIL("(?(DEFINE)....) does not allow branches");
5521 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5522 regbranch(pRExC_state, &flags, 1,depth+1);
5523 REGTAIL(pRExC_state, ret, lastbr);
5526 c = *nextchar(pRExC_state);
5531 vFAIL("Switch (?(condition)... contains too many branches");
5532 ender = reg_node(pRExC_state, TAIL);
5533 REGTAIL(pRExC_state, br, ender);
5535 REGTAIL(pRExC_state, lastbr, ender);
5536 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5539 REGTAIL(pRExC_state, ret, ender);
5540 RExC_size++; /* XXX WHY do we need this?!!
5541 For large programs it seems to be required
5542 but I can't figure out why. -- dmq*/
5546 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5550 RExC_parse--; /* for vFAIL to print correctly */
5551 vFAIL("Sequence (? incomplete");
5555 parse_flags: /* (?i) */
5557 U32 posflags = 0, negflags = 0;
5558 U32 *flagsp = &posflags;
5560 while (*RExC_parse) {
5561 /* && strchr("iogcmsx", *RExC_parse) */
5562 /* (?g), (?gc) and (?o) are useless here
5563 and must be globally applied -- japhy */
5564 switch (*RExC_parse) {
5565 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5568 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5569 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5570 if (! (wastedflags & wflagbit) ) {
5571 wastedflags |= wflagbit;
5574 "Useless (%s%c) - %suse /%c modifier",
5575 flagsp == &negflags ? "?-" : "?",
5577 flagsp == &negflags ? "don't " : "",
5585 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5586 if (! (wastedflags & WASTED_C) ) {
5587 wastedflags |= WASTED_GC;
5590 "Useless (%sc) - %suse /gc modifier",
5591 flagsp == &negflags ? "?-" : "?",
5592 flagsp == &negflags ? "don't " : ""
5598 if (flagsp == &negflags) {
5599 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5600 vWARN(RExC_parse + 1,"Useless use of (?-k)");
5602 *flagsp |= RXf_PMf_KEEPCOPY;
5606 if (flagsp == &negflags) {
5608 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5612 wastedflags = 0; /* reset so (?g-c) warns twice */
5618 RExC_flags |= posflags;
5619 RExC_flags &= ~negflags;
5620 nextchar(pRExC_state);
5631 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5636 }} /* one for the default block, one for the switch */
5643 ret = reganode(pRExC_state, OPEN, parno);
5646 RExC_nestroot = parno;
5647 if (RExC_seen & REG_SEEN_RECURSE
5648 && !RExC_open_parens[parno-1])
5650 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5651 "Setting open paren #%"IVdf" to %d\n",
5652 (IV)parno, REG_NODE_NUM(ret)));
5653 RExC_open_parens[parno-1]= ret;
5656 Set_Node_Length(ret, 1); /* MJD */
5657 Set_Node_Offset(ret, RExC_parse); /* MJD */
5665 /* Pick up the branches, linking them together. */
5666 parse_start = RExC_parse; /* MJD */
5667 br = regbranch(pRExC_state, &flags, 1,depth+1);
5668 /* branch_len = (paren != 0); */
5672 if (*RExC_parse == '|') {
5673 if (!SIZE_ONLY && RExC_extralen) {
5674 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5677 reginsert(pRExC_state, BRANCH, br, depth+1);
5678 Set_Node_Length(br, paren != 0);
5679 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5683 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5685 else if (paren == ':') {
5686 *flagp |= flags&SIMPLE;
5688 if (is_open) { /* Starts with OPEN. */
5689 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5691 else if (paren != '?') /* Not Conditional */
5693 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5695 while (*RExC_parse == '|') {
5696 if (!SIZE_ONLY && RExC_extralen) {
5697 ender = reganode(pRExC_state, LONGJMP,0);
5698 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5701 RExC_extralen += 2; /* Account for LONGJMP. */
5702 nextchar(pRExC_state);
5704 if (RExC_npar > after_freeze)
5705 after_freeze = RExC_npar;
5706 RExC_npar = freeze_paren;
5708 br = regbranch(pRExC_state, &flags, 0, depth+1);
5712 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5714 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5717 if (have_branch || paren != ':') {
5718 /* Make a closing node, and hook it on the end. */
5721 ender = reg_node(pRExC_state, TAIL);
5724 ender = reganode(pRExC_state, CLOSE, parno);
5725 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5726 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5727 "Setting close paren #%"IVdf" to %d\n",
5728 (IV)parno, REG_NODE_NUM(ender)));
5729 RExC_close_parens[parno-1]= ender;
5730 if (RExC_nestroot == parno)
5733 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5734 Set_Node_Length(ender,1); /* MJD */
5740 *flagp &= ~HASWIDTH;
5743 ender = reg_node(pRExC_state, SUCCEED);
5746 ender = reg_node(pRExC_state, END);
5748 assert(!RExC_opend); /* there can only be one! */
5753 REGTAIL(pRExC_state, lastbr, ender);
5755 if (have_branch && !SIZE_ONLY) {
5757 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5759 /* Hook the tails of the branches to the closing node. */
5760 for (br = ret; br; br = regnext(br)) {
5761 const U8 op = PL_regkind[OP(br)];
5763 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5765 else if (op == BRANCHJ) {
5766 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5774 static const char parens[] = "=!<,>";
5776 if (paren && (p = strchr(parens, paren))) {
5777 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5778 int flag = (p - parens) > 1;
5781 node = SUSPEND, flag = 0;
5782 reginsert(pRExC_state, node,ret, depth+1);
5783 Set_Node_Cur_Length(ret);
5784 Set_Node_Offset(ret, parse_start + 1);
5786 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5790 /* Check for proper termination. */
5792 RExC_flags = oregflags;
5793 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5794 RExC_parse = oregcomp_parse;
5795 vFAIL("Unmatched (");
5798 else if (!paren && RExC_parse < RExC_end) {
5799 if (*RExC_parse == ')') {
5801 vFAIL("Unmatched )");
5804 FAIL("Junk on end of regexp"); /* "Can't happen". */
5808 RExC_npar = after_freeze;
5813 - regbranch - one alternative of an | operator
5815 * Implements the concatenation operator.
5818 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5821 register regnode *ret;
5822 register regnode *chain = NULL;
5823 register regnode *latest;
5824 I32 flags = 0, c = 0;
5825 GET_RE_DEBUG_FLAGS_DECL;
5826 DEBUG_PARSE("brnc");
5831 if (!SIZE_ONLY && RExC_extralen)
5832 ret = reganode(pRExC_state, BRANCHJ,0);
5834 ret = reg_node(pRExC_state, BRANCH);
5835 Set_Node_Length(ret, 1);
5839 if (!first && SIZE_ONLY)
5840 RExC_extralen += 1; /* BRANCHJ */
5842 *flagp = WORST; /* Tentatively. */
5845 nextchar(pRExC_state);
5846 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5848 latest = regpiece(pRExC_state, &flags,depth+1);
5849 if (latest == NULL) {
5850 if (flags & TRYAGAIN)
5854 else if (ret == NULL)
5856 *flagp |= flags&(HASWIDTH|POSTPONED);
5857 if (chain == NULL) /* First piece. */
5858 *flagp |= flags&SPSTART;
5861 REGTAIL(pRExC_state, chain, latest);
5866 if (chain == NULL) { /* Loop ran zero times. */
5867 chain = reg_node(pRExC_state, NOTHING);
5872 *flagp |= flags&SIMPLE;
5879 - regpiece - something followed by possible [*+?]
5881 * Note that the branching code sequences used for ? and the general cases
5882 * of * and + are somewhat optimized: they use the same NOTHING node as
5883 * both the endmarker for their branch list and the body of the last branch.
5884 * It might seem that this node could be dispensed with entirely, but the
5885 * endmarker role is not redundant.
5888 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
5891 register regnode *ret;
5893 register char *next;
5895 const char * const origparse = RExC_parse;
5897 I32 max = REG_INFTY;
5899 const char *maxpos = NULL;
5900 GET_RE_DEBUG_FLAGS_DECL;
5901 DEBUG_PARSE("piec");
5903 ret = regatom(pRExC_state, &flags,depth+1);
5905 if (flags & TRYAGAIN)
5912 if (op == '{' && regcurly(RExC_parse)) {
5914 parse_start = RExC_parse; /* MJD */
5915 next = RExC_parse + 1;
5916 while (isDIGIT(*next) || *next == ',') {
5925 if (*next == '}') { /* got one */
5929 min = atoi(RExC_parse);
5933 maxpos = RExC_parse;
5935 if (!max && *maxpos != '0')
5936 max = REG_INFTY; /* meaning "infinity" */
5937 else if (max >= REG_INFTY)
5938 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
5940 nextchar(pRExC_state);
5943 if ((flags&SIMPLE)) {
5944 RExC_naughty += 2 + RExC_naughty / 2;
5945 reginsert(pRExC_state, CURLY, ret, depth+1);
5946 Set_Node_Offset(ret, parse_start+1); /* MJD */
5947 Set_Node_Cur_Length(ret);
5950 regnode * const w = reg_node(pRExC_state, WHILEM);
5953 REGTAIL(pRExC_state, ret, w);
5954 if (!SIZE_ONLY && RExC_extralen) {
5955 reginsert(pRExC_state, LONGJMP,ret, depth+1);
5956 reginsert(pRExC_state, NOTHING,ret, depth+1);
5957 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
5959 reginsert(pRExC_state, CURLYX,ret, depth+1);
5961 Set_Node_Offset(ret, parse_start+1);
5962 Set_Node_Length(ret,
5963 op == '{' ? (RExC_parse - parse_start) : 1);
5965 if (!SIZE_ONLY && RExC_extralen)
5966 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
5967 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
5969 RExC_whilem_seen++, RExC_extralen += 3;
5970 RExC_naughty += 4 + RExC_naughty; /* compound interest */
5978 if (max && max < min)
5979 vFAIL("Can't do {n,m} with n > m");
5981 ARG1_SET(ret, (U16)min);
5982 ARG2_SET(ret, (U16)max);
5994 #if 0 /* Now runtime fix should be reliable. */
5996 /* if this is reinstated, don't forget to put this back into perldiag:
5998 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6000 (F) The part of the regexp subject to either the * or + quantifier
6001 could match an empty string. The {#} shows in the regular
6002 expression about where the problem was discovered.
6006 if (!(flags&HASWIDTH) && op != '?')
6007 vFAIL("Regexp *+ operand could be empty");
6010 parse_start = RExC_parse;
6011 nextchar(pRExC_state);
6013 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6015 if (op == '*' && (flags&SIMPLE)) {
6016 reginsert(pRExC_state, STAR, ret, depth+1);
6020 else if (op == '*') {
6024 else if (op == '+' && (flags&SIMPLE)) {
6025 reginsert(pRExC_state, PLUS, ret, depth+1);
6029 else if (op == '+') {
6033 else if (op == '?') {
6038 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6040 "%.*s matches null string many times",
6041 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6045 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6046 nextchar(pRExC_state);
6047 reginsert(pRExC_state, MINMOD, ret, depth+1);
6048 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6050 #ifndef REG_ALLOW_MINMOD_SUSPEND
6053 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6055 nextchar(pRExC_state);
6056 ender = reg_node(pRExC_state, SUCCEED);
6057 REGTAIL(pRExC_state, ret, ender);
6058 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6060 ender = reg_node(pRExC_state, TAIL);
6061 REGTAIL(pRExC_state, ret, ender);
6065 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6067 vFAIL("Nested quantifiers");
6074 /* reg_namedseq(pRExC_state,UVp)
6076 This is expected to be called by a parser routine that has
6077 recognized'\N' and needs to handle the rest. RExC_parse is
6078 expected to point at the first char following the N at the time
6081 If valuep is non-null then it is assumed that we are parsing inside
6082 of a charclass definition and the first codepoint in the resolved
6083 string is returned via *valuep and the routine will return NULL.
6084 In this mode if a multichar string is returned from the charnames
6085 handler a warning will be issued, and only the first char in the
6086 sequence will be examined. If the string returned is zero length
6087 then the value of *valuep is undefined and NON-NULL will
6088 be returned to indicate failure. (This will NOT be a valid pointer
6091 If value is null then it is assumed that we are parsing normal text
6092 and inserts a new EXACT node into the program containing the resolved
6093 string and returns a pointer to the new node. If the string is
6094 zerolength a NOTHING node is emitted.
6096 On success RExC_parse is set to the char following the endbrace.
6097 Parsing failures will generate a fatal errorvia vFAIL(...)
6099 NOTE: We cache all results from the charnames handler locally in
6100 the RExC_charnames hash (created on first use) to prevent a charnames
6101 handler from playing silly-buggers and returning a short string and
6102 then a long string for a given pattern. Since the regexp program
6103 size is calculated during an initial parse this would result
6104 in a buffer overrun so we cache to prevent the charname result from
6105 changing during the course of the parse.
6109 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6111 char * name; /* start of the content of the name */
6112 char * endbrace; /* endbrace following the name */
6115 STRLEN len; /* this has various purposes throughout the code */
6116 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6117 regnode *ret = NULL;
6119 if (*RExC_parse != '{') {
6120 vFAIL("Missing braces on \\N{}");
6122 name = RExC_parse+1;
6123 endbrace = strchr(RExC_parse, '}');
6126 vFAIL("Missing right brace on \\N{}");
6128 RExC_parse = endbrace + 1;
6131 /* RExC_parse points at the beginning brace,
6132 endbrace points at the last */
6133 if ( name[0]=='U' && name[1]=='+' ) {
6134 /* its a "unicode hex" notation {U+89AB} */
6135 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6136 | PERL_SCAN_DISALLOW_PREFIX
6137 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6139 len = (STRLEN)(endbrace - name - 2);
6140 cp = grok_hex(name + 2, &len, &fl, NULL);
6141 if ( len != (STRLEN)(endbrace - name - 2) ) {
6150 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6152 /* fetch the charnames handler for this scope */
6153 HV * const table = GvHV(PL_hintgv);
6155 hv_fetchs(table, "charnames", FALSE) :
6157 SV *cv= cvp ? *cvp : NULL;
6160 /* create an SV with the name as argument */
6161 sv_name = newSVpvn(name, endbrace - name);
6163 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6164 vFAIL2("Constant(\\N{%s}) unknown: "
6165 "(possibly a missing \"use charnames ...\")",
6168 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6169 vFAIL2("Constant(\\N{%s}): "
6170 "$^H{charnames} is not defined",SvPVX(sv_name));
6175 if (!RExC_charnames) {
6176 /* make sure our cache is allocated */
6177 RExC_charnames = newHV();
6178 sv_2mortal((SV*)RExC_charnames);
6180 /* see if we have looked this one up before */
6181 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6183 sv_str = HeVAL(he_str);
6196 count= call_sv(cv, G_SCALAR);
6198 if (count == 1) { /* XXXX is this right? dmq */
6200 SvREFCNT_inc_simple_void(sv_str);
6208 if ( !sv_str || !SvOK(sv_str) ) {
6209 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6210 "did not return a defined value",SvPVX(sv_name));
6212 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6217 char *p = SvPV(sv_str, len);
6220 if ( SvUTF8(sv_str) ) {
6221 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6225 We have to turn on utf8 for high bit chars otherwise
6226 we get failures with
6228 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6229 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6231 This is different from what \x{} would do with the same
6232 codepoint, where the condition is > 0xFF.
6239 /* warn if we havent used the whole string? */
6241 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6243 "Ignoring excess chars from \\N{%s} in character class",
6247 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6249 "Ignoring zero length \\N{%s} in character class",
6254 SvREFCNT_dec(sv_name);
6256 SvREFCNT_dec(sv_str);
6257 return len ? NULL : (regnode *)&len;
6258 } else if(SvCUR(sv_str)) {
6264 char * parse_start = name-3; /* needed for the offsets */
6266 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6268 ret = reg_node(pRExC_state,
6269 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6272 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6273 sv_utf8_upgrade(sv_str);
6274 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6278 p = SvPV(sv_str, len);
6280 /* len is the length written, charlen is the size the char read */
6281 for ( len = 0; p < pend; p += charlen ) {
6283 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6285 STRLEN foldlen,numlen;
6286 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6287 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6288 /* Emit all the Unicode characters. */
6290 for (foldbuf = tmpbuf;
6294 uvc = utf8_to_uvchr(foldbuf, &numlen);
6296 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6299 /* In EBCDIC the numlen
6300 * and unilen can differ. */
6302 if (numlen >= foldlen)
6306 break; /* "Can't happen." */
6309 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6321 RExC_size += STR_SZ(len);
6324 RExC_emit += STR_SZ(len);
6326 Set_Node_Cur_Length(ret); /* MJD */
6328 nextchar(pRExC_state);
6330 ret = reg_node(pRExC_state,NOTHING);
6333 SvREFCNT_dec(sv_str);
6336 SvREFCNT_dec(sv_name);
6346 * It returns the code point in utf8 for the value in *encp.
6347 * value: a code value in the source encoding
6348 * encp: a pointer to an Encode object
6350 * If the result from Encode is not a single character,
6351 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6354 S_reg_recode(pTHX_ const char value, SV **encp)
6357 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6358 const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
6360 const STRLEN newlen = SvCUR(sv);
6361 UV uv = UNICODE_REPLACEMENT;
6365 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6368 if (!newlen || numlen != newlen) {
6369 uv = UNICODE_REPLACEMENT;
6378 - regatom - the lowest level
6380 Try to identify anything special at the start of the pattern. If there
6381 is, then handle it as required. This may involve generating a single regop,
6382 such as for an assertion; or it may involve recursing, such as to
6383 handle a () structure.
6385 If the string doesn't start with something special then we gobble up
6386 as much literal text as we can.
6388 Once we have been able to handle whatever type of thing started the
6389 sequence, we return.
6391 Note: we have to be careful with escapes, as they can be both literal
6392 and special, and in the case of \10 and friends can either, depending
6393 on context. Specifically there are two seperate switches for handling
6394 escape sequences, with the one for handling literal escapes requiring
6395 a dummy entry for all of the special escapes that are actually handled
6400 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6403 register regnode *ret = NULL;
6405 char *parse_start = RExC_parse;
6406 GET_RE_DEBUG_FLAGS_DECL;
6407 DEBUG_PARSE("atom");
6408 *flagp = WORST; /* Tentatively. */
6412 switch (*RExC_parse) {
6414 RExC_seen_zerolen++;
6415 nextchar(pRExC_state);
6416 if (RExC_flags & RXf_PMf_MULTILINE)
6417 ret = reg_node(pRExC_state, MBOL);
6418 else if (RExC_flags & RXf_PMf_SINGLELINE)
6419 ret = reg_node(pRExC_state, SBOL);
6421 ret = reg_node(pRExC_state, BOL);
6422 Set_Node_Length(ret, 1); /* MJD */
6425 nextchar(pRExC_state);
6427 RExC_seen_zerolen++;
6428 if (RExC_flags & RXf_PMf_MULTILINE)
6429 ret = reg_node(pRExC_state, MEOL);
6430 else if (RExC_flags & RXf_PMf_SINGLELINE)
6431 ret = reg_node(pRExC_state, SEOL);
6433 ret = reg_node(pRExC_state, EOL);
6434 Set_Node_Length(ret, 1); /* MJD */
6437 nextchar(pRExC_state);
6438 if (RExC_flags & RXf_PMf_SINGLELINE)
6439 ret = reg_node(pRExC_state, SANY);
6441 ret = reg_node(pRExC_state, REG_ANY);
6442 *flagp |= HASWIDTH|SIMPLE;
6444 Set_Node_Length(ret, 1); /* MJD */
6448 char * const oregcomp_parse = ++RExC_parse;
6449 ret = regclass(pRExC_state,depth+1);
6450 if (*RExC_parse != ']') {
6451 RExC_parse = oregcomp_parse;
6452 vFAIL("Unmatched [");
6454 nextchar(pRExC_state);
6455 *flagp |= HASWIDTH|SIMPLE;
6456 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6460 nextchar(pRExC_state);
6461 ret = reg(pRExC_state, 1, &flags,depth+1);
6463 if (flags & TRYAGAIN) {
6464 if (RExC_parse == RExC_end) {
6465 /* Make parent create an empty node if needed. */
6473 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6477 if (flags & TRYAGAIN) {
6481 vFAIL("Internal urp");
6482 /* Supposed to be caught earlier. */
6485 if (!regcurly(RExC_parse)) {
6494 vFAIL("Quantifier follows nothing");
6499 This switch handles escape sequences that resolve to some kind
6500 of special regop and not to literal text. Escape sequnces that
6501 resolve to literal text are handled below in the switch marked
6504 Every entry in this switch *must* have a corresponding entry
6505 in the literal escape switch. However, the opposite is not
6506 required, as the default for this switch is to jump to the
6507 literal text handling code.
6509 switch (*++RExC_parse) {
6510 /* Special Escapes */
6512 RExC_seen_zerolen++;
6513 ret = reg_node(pRExC_state, SBOL);
6515 goto finish_meta_pat;
6517 ret = reg_node(pRExC_state, GPOS);
6518 RExC_seen |= REG_SEEN_GPOS;
6520 goto finish_meta_pat;
6522 RExC_seen_zerolen++;
6523 ret = reg_node(pRExC_state, KEEPS);
6525 goto finish_meta_pat;
6527 ret = reg_node(pRExC_state, SEOL);
6529 RExC_seen_zerolen++; /* Do not optimize RE away */
6530 goto finish_meta_pat;
6532 ret = reg_node(pRExC_state, EOS);
6534 RExC_seen_zerolen++; /* Do not optimize RE away */
6535 goto finish_meta_pat;
6537 ret = reg_node(pRExC_state, CANY);
6538 RExC_seen |= REG_SEEN_CANY;
6539 *flagp |= HASWIDTH|SIMPLE;
6540 goto finish_meta_pat;
6542 ret = reg_node(pRExC_state, CLUMP);
6544 goto finish_meta_pat;
6546 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6547 *flagp |= HASWIDTH|SIMPLE;
6548 goto finish_meta_pat;
6550 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6551 *flagp |= HASWIDTH|SIMPLE;
6552 goto finish_meta_pat;
6554 RExC_seen_zerolen++;
6555 RExC_seen |= REG_SEEN_LOOKBEHIND;
6556 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6558 goto finish_meta_pat;
6560 RExC_seen_zerolen++;
6561 RExC_seen |= REG_SEEN_LOOKBEHIND;
6562 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6564 goto finish_meta_pat;
6566 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6567 *flagp |= HASWIDTH|SIMPLE;
6568 goto finish_meta_pat;
6570 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6571 *flagp |= HASWIDTH|SIMPLE;
6572 goto finish_meta_pat;
6574 ret = reg_node(pRExC_state, DIGIT);
6575 *flagp |= HASWIDTH|SIMPLE;
6576 goto finish_meta_pat;
6578 ret = reg_node(pRExC_state, NDIGIT);
6579 *flagp |= HASWIDTH|SIMPLE;
6580 goto finish_meta_pat;
6582 ret = reganode(pRExC_state, PRUNE, 0);
6585 goto finish_meta_pat;
6587 ret = reganode(pRExC_state, SKIP, 0);
6591 nextchar(pRExC_state);
6592 Set_Node_Length(ret, 2); /* MJD */
6597 char* const oldregxend = RExC_end;
6599 char* parse_start = RExC_parse - 2;
6602 if (RExC_parse[1] == '{') {
6603 /* a lovely hack--pretend we saw [\pX] instead */
6604 RExC_end = strchr(RExC_parse, '}');
6606 const U8 c = (U8)*RExC_parse;
6608 RExC_end = oldregxend;
6609 vFAIL2("Missing right brace on \\%c{}", c);
6614 RExC_end = RExC_parse + 2;
6615 if (RExC_end > oldregxend)
6616 RExC_end = oldregxend;
6620 ret = regclass(pRExC_state,depth+1);
6622 RExC_end = oldregxend;
6625 Set_Node_Offset(ret, parse_start + 2);
6626 Set_Node_Cur_Length(ret);
6627 nextchar(pRExC_state);
6628 *flagp |= HASWIDTH|SIMPLE;
6632 /* Handle \N{NAME} here and not below because it can be
6633 multicharacter. join_exact() will join them up later on.
6634 Also this makes sure that things like /\N{BLAH}+/ and
6635 \N{BLAH} being multi char Just Happen. dmq*/
6637 ret= reg_namedseq(pRExC_state, NULL);
6639 case 'k': /* Handle \k<NAME> and \k'NAME' */
6642 char ch= RExC_parse[1];
6643 if (ch != '<' && ch != '\'' && ch != '{') {
6645 vFAIL2("Sequence %.2s... not terminated",parse_start);
6647 /* this pretty much dupes the code for (?P=...) in reg(), if
6648 you change this make sure you change that */
6649 char* name_start = (RExC_parse += 2);
6651 SV *sv_dat = reg_scan_name(pRExC_state,
6652 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6653 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
6654 if (RExC_parse == name_start || *RExC_parse != ch)
6655 vFAIL2("Sequence %.3s... not terminated",parse_start);
6658 num = add_data( pRExC_state, 1, "S" );
6659 RExC_rxi->data->data[num]=(void*)sv_dat;
6660 SvREFCNT_inc(sv_dat);
6664 ret = reganode(pRExC_state,
6665 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6669 /* override incorrect value set in reganode MJD */
6670 Set_Node_Offset(ret, parse_start+1);
6671 Set_Node_Cur_Length(ret); /* MJD */
6672 nextchar(pRExC_state);
6678 case '1': case '2': case '3': case '4':
6679 case '5': case '6': case '7': case '8': case '9':
6682 bool isg = *RExC_parse == 'g';
6687 if (*RExC_parse == '{') {
6691 if (*RExC_parse == '-') {
6695 if (hasbrace && !isDIGIT(*RExC_parse)) {
6696 if (isrel) RExC_parse--;
6698 goto parse_named_seq;
6700 num = atoi(RExC_parse);
6702 num = RExC_npar - num;
6704 vFAIL("Reference to nonexistent or unclosed group");
6706 if (!isg && num > 9 && num >= RExC_npar)
6709 char * const parse_start = RExC_parse - 1; /* MJD */
6710 while (isDIGIT(*RExC_parse))
6712 if (parse_start == RExC_parse - 1)
6713 vFAIL("Unterminated \\g... pattern");
6715 if (*RExC_parse != '}')
6716 vFAIL("Unterminated \\g{...} pattern");
6720 if (num > (I32)RExC_rx->nparens)
6721 vFAIL("Reference to nonexistent group");
6724 ret = reganode(pRExC_state,
6725 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6729 /* override incorrect value set in reganode MJD */
6730 Set_Node_Offset(ret, parse_start+1);
6731 Set_Node_Cur_Length(ret); /* MJD */
6733 nextchar(pRExC_state);
6738 if (RExC_parse >= RExC_end)
6739 FAIL("Trailing \\");
6742 /* Do not generate "unrecognized" warnings here, we fall
6743 back into the quick-grab loop below */
6750 if (RExC_flags & RXf_PMf_EXTENDED) {
6751 if ( reg_skipcomment( pRExC_state ) )
6757 register STRLEN len;
6762 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6764 parse_start = RExC_parse - 1;
6770 ret = reg_node(pRExC_state,
6771 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6773 for (len = 0, p = RExC_parse - 1;
6774 len < 127 && p < RExC_end;
6777 char * const oldp = p;
6779 if (RExC_flags & RXf_PMf_EXTENDED)
6780 p = regwhite( pRExC_state, p );
6791 /* Literal Escapes Switch
6793 This switch is meant to handle escape sequences that
6794 resolve to a literal character.
6796 Every escape sequence that represents something
6797 else, like an assertion or a char class, is handled
6798 in the switch marked 'Special Escapes' above in this
6799 routine, but also has an entry here as anything that
6800 isn't explicitly mentioned here will be treated as
6801 an unescaped equivalent literal.
6805 /* These are all the special escapes. */
6806 case 'A': /* Start assertion */
6807 case 'b': case 'B': /* Word-boundary assertion*/
6808 case 'C': /* Single char !DANGEROUS! */
6809 case 'd': case 'D': /* digit class */
6810 case 'g': case 'G': /* generic-backref, pos assertion */
6811 case 'k': case 'K': /* named backref, keep marker */
6812 case 'N': /* named char sequence */
6813 case 'p': case 'P': /* unicode property */
6814 case 's': case 'S': /* space class */
6815 case 'v': case 'V': /* (*PRUNE) and (*SKIP) */
6816 case 'w': case 'W': /* word class */
6817 case 'X': /* eXtended Unicode "combining character sequence" */
6818 case 'z': case 'Z': /* End of line/string assertion */
6822 /* Anything after here is an escape that resolves to a
6823 literal. (Except digits, which may or may not)
6842 ender = ASCII_TO_NATIVE('\033');
6846 ender = ASCII_TO_NATIVE('\007');
6851 char* const e = strchr(p, '}');
6855 vFAIL("Missing right brace on \\x{}");
6858 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6859 | PERL_SCAN_DISALLOW_PREFIX;
6860 STRLEN numlen = e - p - 1;
6861 ender = grok_hex(p + 1, &numlen, &flags, NULL);
6868 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
6870 ender = grok_hex(p, &numlen, &flags, NULL);
6873 if (PL_encoding && ender < 0x100)
6874 goto recode_encoding;
6878 ender = UCHARAT(p++);
6879 ender = toCTRL(ender);
6881 case '0': case '1': case '2': case '3':case '4':
6882 case '5': case '6': case '7': case '8':case '9':
6884 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
6887 ender = grok_oct(p, &numlen, &flags, NULL);
6894 if (PL_encoding && ender < 0x100)
6895 goto recode_encoding;
6899 SV* enc = PL_encoding;
6900 ender = reg_recode((const char)(U8)ender, &enc);
6901 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
6902 vWARN(p, "Invalid escape in the specified encoding");
6908 FAIL("Trailing \\");
6911 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
6912 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
6913 goto normal_default;
6918 if (UTF8_IS_START(*p) && UTF) {
6920 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
6921 &numlen, UTF8_ALLOW_DEFAULT);
6928 if ( RExC_flags & RXf_PMf_EXTENDED)
6929 p = regwhite( pRExC_state, p );
6931 /* Prime the casefolded buffer. */
6932 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
6934 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
6939 /* Emit all the Unicode characters. */
6941 for (foldbuf = tmpbuf;
6943 foldlen -= numlen) {
6944 ender = utf8_to_uvchr(foldbuf, &numlen);
6946 const STRLEN unilen = reguni(pRExC_state, ender, s);
6949 /* In EBCDIC the numlen
6950 * and unilen can differ. */
6952 if (numlen >= foldlen)
6956 break; /* "Can't happen." */
6960 const STRLEN unilen = reguni(pRExC_state, ender, s);
6969 REGC((char)ender, s++);
6975 /* Emit all the Unicode characters. */
6977 for (foldbuf = tmpbuf;
6979 foldlen -= numlen) {
6980 ender = utf8_to_uvchr(foldbuf, &numlen);
6982 const STRLEN unilen = reguni(pRExC_state, ender, s);
6985 /* In EBCDIC the numlen
6986 * and unilen can differ. */
6988 if (numlen >= foldlen)
6996 const STRLEN unilen = reguni(pRExC_state, ender, s);
7005 REGC((char)ender, s++);
7009 Set_Node_Cur_Length(ret); /* MJD */
7010 nextchar(pRExC_state);
7012 /* len is STRLEN which is unsigned, need to copy to signed */
7015 vFAIL("Internal disaster");
7019 if (len == 1 && UNI_IS_INVARIANT(ender))
7023 RExC_size += STR_SZ(len);
7026 RExC_emit += STR_SZ(len);
7036 S_regwhite( RExC_state_t *pRExC_state, char *p )
7038 const char *e = RExC_end;
7042 else if (*p == '#') {
7051 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7059 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7060 Character classes ([:foo:]) can also be negated ([:^foo:]).
7061 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7062 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7063 but trigger failures because they are currently unimplemented. */
7065 #define POSIXCC_DONE(c) ((c) == ':')
7066 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7067 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7070 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7073 I32 namedclass = OOB_NAMEDCLASS;
7075 if (value == '[' && RExC_parse + 1 < RExC_end &&
7076 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7077 POSIXCC(UCHARAT(RExC_parse))) {
7078 const char c = UCHARAT(RExC_parse);
7079 char* const s = RExC_parse++;
7081 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7083 if (RExC_parse == RExC_end)
7084 /* Grandfather lone [:, [=, [. */
7087 const char* const t = RExC_parse++; /* skip over the c */
7090 if (UCHARAT(RExC_parse) == ']') {
7091 const char *posixcc = s + 1;
7092 RExC_parse++; /* skip over the ending ] */
7095 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7096 const I32 skip = t - posixcc;
7098 /* Initially switch on the length of the name. */
7101 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7102 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7105 /* Names all of length 5. */
7106 /* alnum alpha ascii blank cntrl digit graph lower
7107 print punct space upper */
7108 /* Offset 4 gives the best switch position. */
7109 switch (posixcc[4]) {
7111 if (memEQ(posixcc, "alph", 4)) /* alpha */
7112 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7115 if (memEQ(posixcc, "spac", 4)) /* space */
7116 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7119 if (memEQ(posixcc, "grap", 4)) /* graph */
7120 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7123 if (memEQ(posixcc, "asci", 4)) /* ascii */
7124 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7127 if (memEQ(posixcc, "blan", 4)) /* blank */
7128 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7131 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7132 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7135 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7136 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7139 if (memEQ(posixcc, "lowe", 4)) /* lower */
7140 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7141 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7142 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7145 if (memEQ(posixcc, "digi", 4)) /* digit */
7146 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7147 else if (memEQ(posixcc, "prin", 4)) /* print */
7148 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7149 else if (memEQ(posixcc, "punc", 4)) /* punct */
7150 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7155 if (memEQ(posixcc, "xdigit", 6))
7156 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7160 if (namedclass == OOB_NAMEDCLASS)
7161 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7163 assert (posixcc[skip] == ':');
7164 assert (posixcc[skip+1] == ']');
7165 } else if (!SIZE_ONLY) {
7166 /* [[=foo=]] and [[.foo.]] are still future. */
7168 /* adjust RExC_parse so the warning shows after
7170 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7172 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7175 /* Maternal grandfather:
7176 * "[:" ending in ":" but not in ":]" */
7186 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7189 if (POSIXCC(UCHARAT(RExC_parse))) {
7190 const char *s = RExC_parse;
7191 const char c = *s++;
7195 if (*s && c == *s && s[1] == ']') {
7196 if (ckWARN(WARN_REGEXP))
7198 "POSIX syntax [%c %c] belongs inside character classes",
7201 /* [[=foo=]] and [[.foo.]] are still future. */
7202 if (POSIXCC_NOTYET(c)) {
7203 /* adjust RExC_parse so the error shows after
7205 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7207 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7214 #define _C_C_T_(NAME,TEST,WORD) \
7217 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7219 for (value = 0; value < 256; value++) \
7221 ANYOF_BITMAP_SET(ret, value); \
7226 case ANYOF_N##NAME: \
7228 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7230 for (value = 0; value < 256; value++) \
7232 ANYOF_BITMAP_SET(ret, value); \
7240 parse a class specification and produce either an ANYOF node that
7241 matches the pattern or if the pattern matches a single char only and
7242 that char is < 256 and we are case insensitive then we produce an
7247 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7250 register UV value = 0;
7251 register UV nextvalue;
7252 register IV prevvalue = OOB_UNICODE;
7253 register IV range = 0;
7254 register regnode *ret;
7257 char *rangebegin = NULL;
7258 bool need_class = 0;
7261 bool optimize_invert = TRUE;
7262 AV* unicode_alternate = NULL;
7264 UV literal_endpoint = 0;
7266 UV stored = 0; /* number of chars stored in the class */
7268 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7269 case we need to change the emitted regop to an EXACT. */
7270 const char * orig_parse = RExC_parse;
7271 GET_RE_DEBUG_FLAGS_DECL;
7273 PERL_UNUSED_ARG(depth);
7276 DEBUG_PARSE("clas");
7278 /* Assume we are going to generate an ANYOF node. */
7279 ret = reganode(pRExC_state, ANYOF, 0);
7282 ANYOF_FLAGS(ret) = 0;
7284 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7288 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7292 RExC_size += ANYOF_SKIP;
7293 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7296 RExC_emit += ANYOF_SKIP;
7298 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7300 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7301 ANYOF_BITMAP_ZERO(ret);
7302 listsv = newSVpvs("# comment\n");
7305 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7307 if (!SIZE_ONLY && POSIXCC(nextvalue))
7308 checkposixcc(pRExC_state);
7310 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7311 if (UCHARAT(RExC_parse) == ']')
7315 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7319 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7322 rangebegin = RExC_parse;
7324 value = utf8n_to_uvchr((U8*)RExC_parse,
7325 RExC_end - RExC_parse,
7326 &numlen, UTF8_ALLOW_DEFAULT);
7327 RExC_parse += numlen;
7330 value = UCHARAT(RExC_parse++);
7332 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7333 if (value == '[' && POSIXCC(nextvalue))
7334 namedclass = regpposixcc(pRExC_state, value);
7335 else if (value == '\\') {
7337 value = utf8n_to_uvchr((U8*)RExC_parse,
7338 RExC_end - RExC_parse,
7339 &numlen, UTF8_ALLOW_DEFAULT);
7340 RExC_parse += numlen;
7343 value = UCHARAT(RExC_parse++);
7344 /* Some compilers cannot handle switching on 64-bit integer
7345 * values, therefore value cannot be an UV. Yes, this will
7346 * be a problem later if we want switch on Unicode.
7347 * A similar issue a little bit later when switching on
7348 * namedclass. --jhi */
7349 switch ((I32)value) {
7350 case 'w': namedclass = ANYOF_ALNUM; break;
7351 case 'W': namedclass = ANYOF_NALNUM; break;
7352 case 's': namedclass = ANYOF_SPACE; break;
7353 case 'S': namedclass = ANYOF_NSPACE; break;
7354 case 'd': namedclass = ANYOF_DIGIT; break;
7355 case 'D': namedclass = ANYOF_NDIGIT; break;
7356 case 'N': /* Handle \N{NAME} in class */
7358 /* We only pay attention to the first char of
7359 multichar strings being returned. I kinda wonder
7360 if this makes sense as it does change the behaviour
7361 from earlier versions, OTOH that behaviour was broken
7363 UV v; /* value is register so we cant & it /grrr */
7364 if (reg_namedseq(pRExC_state, &v)) {
7374 if (RExC_parse >= RExC_end)
7375 vFAIL2("Empty \\%c{}", (U8)value);
7376 if (*RExC_parse == '{') {
7377 const U8 c = (U8)value;
7378 e = strchr(RExC_parse++, '}');
7380 vFAIL2("Missing right brace on \\%c{}", c);
7381 while (isSPACE(UCHARAT(RExC_parse)))
7383 if (e == RExC_parse)
7384 vFAIL2("Empty \\%c{}", c);
7386 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7394 if (UCHARAT(RExC_parse) == '^') {
7397 value = value == 'p' ? 'P' : 'p'; /* toggle */
7398 while (isSPACE(UCHARAT(RExC_parse))) {
7403 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7404 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7407 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7408 namedclass = ANYOF_MAX; /* no official name, but it's named */
7411 case 'n': value = '\n'; break;
7412 case 'r': value = '\r'; break;
7413 case 't': value = '\t'; break;
7414 case 'f': value = '\f'; break;
7415 case 'b': value = '\b'; break;
7416 case 'e': value = ASCII_TO_NATIVE('\033');break;
7417 case 'a': value = ASCII_TO_NATIVE('\007');break;
7419 if (*RExC_parse == '{') {
7420 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7421 | PERL_SCAN_DISALLOW_PREFIX;
7422 char * const e = strchr(RExC_parse++, '}');
7424 vFAIL("Missing right brace on \\x{}");
7426 numlen = e - RExC_parse;
7427 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7431 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7433 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7434 RExC_parse += numlen;
7436 if (PL_encoding && value < 0x100)
7437 goto recode_encoding;
7440 value = UCHARAT(RExC_parse++);
7441 value = toCTRL(value);
7443 case '0': case '1': case '2': case '3': case '4':
7444 case '5': case '6': case '7': case '8': case '9':
7448 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7449 RExC_parse += numlen;
7450 if (PL_encoding && value < 0x100)
7451 goto recode_encoding;
7456 SV* enc = PL_encoding;
7457 value = reg_recode((const char)(U8)value, &enc);
7458 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7460 "Invalid escape in the specified encoding");
7464 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7466 "Unrecognized escape \\%c in character class passed through",
7470 } /* end of \blah */
7476 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7478 if (!SIZE_ONLY && !need_class)
7479 ANYOF_CLASS_ZERO(ret);
7483 /* a bad range like a-\d, a-[:digit:] ? */
7486 if (ckWARN(WARN_REGEXP)) {
7488 RExC_parse >= rangebegin ?
7489 RExC_parse - rangebegin : 0;
7491 "False [] range \"%*.*s\"",
7494 if (prevvalue < 256) {
7495 ANYOF_BITMAP_SET(ret, prevvalue);
7496 ANYOF_BITMAP_SET(ret, '-');
7499 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7500 Perl_sv_catpvf(aTHX_ listsv,
7501 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7505 range = 0; /* this was not a true range */
7511 const char *what = NULL;
7514 if (namedclass > OOB_NAMEDCLASS)
7515 optimize_invert = FALSE;
7516 /* Possible truncation here but in some 64-bit environments
7517 * the compiler gets heartburn about switch on 64-bit values.
7518 * A similar issue a little earlier when switching on value.
7520 switch ((I32)namedclass) {
7521 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7522 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7523 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7524 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7525 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7526 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7527 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7528 case _C_C_T_(PRINT, isPRINT(value), "Print");
7529 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7530 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7531 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7532 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7533 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7536 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7539 for (value = 0; value < 128; value++)
7540 ANYOF_BITMAP_SET(ret, value);
7542 for (value = 0; value < 256; value++) {
7544 ANYOF_BITMAP_SET(ret, value);
7553 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7556 for (value = 128; value < 256; value++)
7557 ANYOF_BITMAP_SET(ret, value);
7559 for (value = 0; value < 256; value++) {
7560 if (!isASCII(value))
7561 ANYOF_BITMAP_SET(ret, value);
7570 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7572 /* consecutive digits assumed */
7573 for (value = '0'; value <= '9'; value++)
7574 ANYOF_BITMAP_SET(ret, value);
7581 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7583 /* consecutive digits assumed */
7584 for (value = 0; value < '0'; value++)
7585 ANYOF_BITMAP_SET(ret, value);
7586 for (value = '9' + 1; value < 256; value++)
7587 ANYOF_BITMAP_SET(ret, value);
7593 /* this is to handle \p and \P */
7596 vFAIL("Invalid [::] class");
7600 /* Strings such as "+utf8::isWord\n" */
7601 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7604 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7607 } /* end of namedclass \blah */
7610 if (prevvalue > (IV)value) /* b-a */ {
7611 const int w = RExC_parse - rangebegin;
7612 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7613 range = 0; /* not a valid range */
7617 prevvalue = value; /* save the beginning of the range */
7618 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7619 RExC_parse[1] != ']') {
7622 /* a bad range like \w-, [:word:]- ? */
7623 if (namedclass > OOB_NAMEDCLASS) {
7624 if (ckWARN(WARN_REGEXP)) {
7626 RExC_parse >= rangebegin ?
7627 RExC_parse - rangebegin : 0;
7629 "False [] range \"%*.*s\"",
7633 ANYOF_BITMAP_SET(ret, '-');
7635 range = 1; /* yeah, it's a range! */
7636 continue; /* but do it the next time */
7640 /* now is the next time */
7641 /*stored += (value - prevvalue + 1);*/
7643 if (prevvalue < 256) {
7644 const IV ceilvalue = value < 256 ? value : 255;
7647 /* In EBCDIC [\x89-\x91] should include
7648 * the \x8e but [i-j] should not. */
7649 if (literal_endpoint == 2 &&
7650 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7651 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7653 if (isLOWER(prevvalue)) {
7654 for (i = prevvalue; i <= ceilvalue; i++)
7656 ANYOF_BITMAP_SET(ret, i);
7658 for (i = prevvalue; i <= ceilvalue; i++)
7660 ANYOF_BITMAP_SET(ret, i);
7665 for (i = prevvalue; i <= ceilvalue; i++) {
7666 if (!ANYOF_BITMAP_TEST(ret,i)) {
7668 ANYOF_BITMAP_SET(ret, i);
7672 if (value > 255 || UTF) {
7673 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7674 const UV natvalue = NATIVE_TO_UNI(value);
7675 stored+=2; /* can't optimize this class */
7676 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7677 if (prevnatvalue < natvalue) { /* what about > ? */
7678 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7679 prevnatvalue, natvalue);
7681 else if (prevnatvalue == natvalue) {
7682 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7684 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7686 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7688 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7689 if (RExC_precomp[0] == ':' &&
7690 RExC_precomp[1] == '[' &&
7691 (f == 0xDF || f == 0x92)) {
7692 f = NATIVE_TO_UNI(f);
7695 /* If folding and foldable and a single
7696 * character, insert also the folded version
7697 * to the charclass. */
7699 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7700 if ((RExC_precomp[0] == ':' &&
7701 RExC_precomp[1] == '[' &&
7703 (value == 0xFB05 || value == 0xFB06))) ?
7704 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7705 foldlen == (STRLEN)UNISKIP(f) )
7707 if (foldlen == (STRLEN)UNISKIP(f))
7709 Perl_sv_catpvf(aTHX_ listsv,
7712 /* Any multicharacter foldings
7713 * require the following transform:
7714 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7715 * where E folds into "pq" and F folds
7716 * into "rst", all other characters
7717 * fold to single characters. We save
7718 * away these multicharacter foldings,
7719 * to be later saved as part of the
7720 * additional "s" data. */
7723 if (!unicode_alternate)
7724 unicode_alternate = newAV();
7725 sv = newSVpvn((char*)foldbuf, foldlen);
7727 av_push(unicode_alternate, sv);
7731 /* If folding and the value is one of the Greek
7732 * sigmas insert a few more sigmas to make the
7733 * folding rules of the sigmas to work right.
7734 * Note that not all the possible combinations
7735 * are handled here: some of them are handled
7736 * by the standard folding rules, and some of
7737 * them (literal or EXACTF cases) are handled
7738 * during runtime in regexec.c:S_find_byclass(). */
7739 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7740 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7741 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7742 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7743 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7745 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7746 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7747 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7752 literal_endpoint = 0;
7756 range = 0; /* this range (if it was one) is done now */
7760 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7762 RExC_size += ANYOF_CLASS_ADD_SKIP;
7764 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7770 /****** !SIZE_ONLY AFTER HERE *********/
7772 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
7773 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7775 /* optimize single char class to an EXACT node
7776 but *only* when its not a UTF/high char */
7777 const char * cur_parse= RExC_parse;
7778 RExC_emit = (regnode *)orig_emit;
7779 RExC_parse = (char *)orig_parse;
7780 ret = reg_node(pRExC_state,
7781 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7782 RExC_parse = (char *)cur_parse;
7783 *STRING(ret)= (char)value;
7785 RExC_emit += STR_SZ(1);
7788 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7789 if ( /* If the only flag is folding (plus possibly inversion). */
7790 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7792 for (value = 0; value < 256; ++value) {
7793 if (ANYOF_BITMAP_TEST(ret, value)) {
7794 UV fold = PL_fold[value];
7797 ANYOF_BITMAP_SET(ret, fold);
7800 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
7803 /* optimize inverted simple patterns (e.g. [^a-z]) */
7804 if (optimize_invert &&
7805 /* If the only flag is inversion. */
7806 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
7807 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
7808 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
7809 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
7812 AV * const av = newAV();
7814 /* The 0th element stores the character class description
7815 * in its textual form: used later (regexec.c:Perl_regclass_swash())
7816 * to initialize the appropriate swash (which gets stored in
7817 * the 1st element), and also useful for dumping the regnode.
7818 * The 2nd element stores the multicharacter foldings,
7819 * used later (regexec.c:S_reginclass()). */
7820 av_store(av, 0, listsv);
7821 av_store(av, 1, NULL);
7822 av_store(av, 2, (SV*)unicode_alternate);
7823 rv = newRV_noinc((SV*)av);
7824 n = add_data(pRExC_state, 1, "s");
7825 RExC_rxi->data->data[n] = (void*)rv;
7833 /* reg_skipcomment()
7835 Absorbs an /x style # comments from the input stream.
7836 Returns true if there is more text remaining in the stream.
7837 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
7838 terminates the pattern without including a newline.
7840 Note its the callers responsibility to ensure that we are
7846 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
7849 while (RExC_parse < RExC_end)
7850 if (*RExC_parse++ == '\n') {
7855 /* we ran off the end of the pattern without ending
7856 the comment, so we have to add an \n when wrapping */
7857 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7865 Advance that parse position, and optionall absorbs
7866 "whitespace" from the inputstream.
7868 Without /x "whitespace" means (?#...) style comments only,
7869 with /x this means (?#...) and # comments and whitespace proper.
7871 Returns the RExC_parse point from BEFORE the scan occurs.
7873 This is the /x friendly way of saying RExC_parse++.
7877 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
7879 char* const retval = RExC_parse++;
7882 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
7883 RExC_parse[2] == '#') {
7884 while (*RExC_parse != ')') {
7885 if (RExC_parse == RExC_end)
7886 FAIL("Sequence (?#... not terminated");
7892 if (RExC_flags & RXf_PMf_EXTENDED) {
7893 if (isSPACE(*RExC_parse)) {
7897 else if (*RExC_parse == '#') {
7898 if ( reg_skipcomment( pRExC_state ) )
7907 - reg_node - emit a node
7909 STATIC regnode * /* Location. */
7910 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
7913 register regnode *ptr;
7914 regnode * const ret = RExC_emit;
7915 GET_RE_DEBUG_FLAGS_DECL;
7918 SIZE_ALIGN(RExC_size);
7922 if (RExC_emit >= RExC_emit_bound)
7923 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
7925 NODE_ALIGN_FILL(ret);
7927 FILL_ADVANCE_NODE(ptr, op);
7928 #ifdef RE_TRACK_PATTERN_OFFSETS
7929 if (RExC_offsets) { /* MJD */
7930 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
7931 "reg_node", __LINE__,
7933 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
7934 ? "Overwriting end of array!\n" : "OK",
7935 (UV)(RExC_emit - RExC_emit_start),
7936 (UV)(RExC_parse - RExC_start),
7937 (UV)RExC_offsets[0]));
7938 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
7946 - reganode - emit a node with an argument
7948 STATIC regnode * /* Location. */
7949 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
7952 register regnode *ptr;
7953 regnode * const ret = RExC_emit;
7954 GET_RE_DEBUG_FLAGS_DECL;
7957 SIZE_ALIGN(RExC_size);
7962 assert(2==regarglen[op]+1);
7964 Anything larger than this has to allocate the extra amount.
7965 If we changed this to be:
7967 RExC_size += (1 + regarglen[op]);
7969 then it wouldn't matter. Its not clear what side effect
7970 might come from that so its not done so far.
7975 if (RExC_emit >= RExC_emit_bound)
7976 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
7978 NODE_ALIGN_FILL(ret);
7980 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
7981 #ifdef RE_TRACK_PATTERN_OFFSETS
7982 if (RExC_offsets) { /* MJD */
7983 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
7987 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
7988 "Overwriting end of array!\n" : "OK",
7989 (UV)(RExC_emit - RExC_emit_start),
7990 (UV)(RExC_parse - RExC_start),
7991 (UV)RExC_offsets[0]));
7992 Set_Cur_Node_Offset;
8000 - reguni - emit (if appropriate) a Unicode character
8003 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8006 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8010 - reginsert - insert an operator in front of already-emitted operand
8012 * Means relocating the operand.
8015 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8018 register regnode *src;
8019 register regnode *dst;
8020 register regnode *place;
8021 const int offset = regarglen[(U8)op];
8022 const int size = NODE_STEP_REGNODE + offset;
8023 GET_RE_DEBUG_FLAGS_DECL;
8024 PERL_UNUSED_ARG(depth);
8025 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8026 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8035 if (RExC_open_parens) {
8037 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8038 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8039 if ( RExC_open_parens[paren] >= opnd ) {
8040 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8041 RExC_open_parens[paren] += size;
8043 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8045 if ( RExC_close_parens[paren] >= opnd ) {
8046 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8047 RExC_close_parens[paren] += size;
8049 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8054 while (src > opnd) {
8055 StructCopy(--src, --dst, regnode);
8056 #ifdef RE_TRACK_PATTERN_OFFSETS
8057 if (RExC_offsets) { /* MJD 20010112 */
8058 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8062 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8063 ? "Overwriting end of array!\n" : "OK",
8064 (UV)(src - RExC_emit_start),
8065 (UV)(dst - RExC_emit_start),
8066 (UV)RExC_offsets[0]));
8067 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8068 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8074 place = opnd; /* Op node, where operand used to be. */
8075 #ifdef RE_TRACK_PATTERN_OFFSETS
8076 if (RExC_offsets) { /* MJD */
8077 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8081 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8082 ? "Overwriting end of array!\n" : "OK",
8083 (UV)(place - RExC_emit_start),
8084 (UV)(RExC_parse - RExC_start),
8085 (UV)RExC_offsets[0]));
8086 Set_Node_Offset(place, RExC_parse);
8087 Set_Node_Length(place, 1);
8090 src = NEXTOPER(place);
8091 FILL_ADVANCE_NODE(place, op);
8092 Zero(src, offset, regnode);
8096 - regtail - set the next-pointer at the end of a node chain of p to val.
8097 - SEE ALSO: regtail_study
8099 /* TODO: All three parms should be const */
8101 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8104 register regnode *scan;
8105 GET_RE_DEBUG_FLAGS_DECL;
8107 PERL_UNUSED_ARG(depth);
8113 /* Find last node. */
8116 regnode * const temp = regnext(scan);
8118 SV * const mysv=sv_newmortal();
8119 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8120 regprop(RExC_rx, mysv, scan);
8121 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8122 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8123 (temp == NULL ? "->" : ""),
8124 (temp == NULL ? PL_reg_name[OP(val)] : "")
8132 if (reg_off_by_arg[OP(scan)]) {
8133 ARG_SET(scan, val - scan);
8136 NEXT_OFF(scan) = val - scan;
8142 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8143 - Look for optimizable sequences at the same time.
8144 - currently only looks for EXACT chains.
8146 This is expermental code. The idea is to use this routine to perform
8147 in place optimizations on branches and groups as they are constructed,
8148 with the long term intention of removing optimization from study_chunk so
8149 that it is purely analytical.
8151 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8152 to control which is which.
8155 /* TODO: All four parms should be const */
8158 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8161 register regnode *scan;
8163 #ifdef EXPERIMENTAL_INPLACESCAN
8167 GET_RE_DEBUG_FLAGS_DECL;
8173 /* Find last node. */
8177 regnode * const temp = regnext(scan);
8178 #ifdef EXPERIMENTAL_INPLACESCAN
8179 if (PL_regkind[OP(scan)] == EXACT)
8180 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8188 if( exact == PSEUDO )
8190 else if ( exact != OP(scan) )
8199 SV * const mysv=sv_newmortal();
8200 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8201 regprop(RExC_rx, mysv, scan);
8202 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8203 SvPV_nolen_const(mysv),
8205 PL_reg_name[exact]);
8212 SV * const mysv_val=sv_newmortal();
8213 DEBUG_PARSE_MSG("");
8214 regprop(RExC_rx, mysv_val, val);
8215 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8216 SvPV_nolen_const(mysv_val),
8217 (IV)REG_NODE_NUM(val),
8221 if (reg_off_by_arg[OP(scan)]) {
8222 ARG_SET(scan, val - scan);
8225 NEXT_OFF(scan) = val - scan;
8233 - regcurly - a little FSA that accepts {\d+,?\d*}
8236 S_regcurly(register const char *s)
8255 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8258 Perl_regdump(pTHX_ const regexp *r)
8262 SV * const sv = sv_newmortal();
8263 SV *dsv= sv_newmortal();
8266 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8268 /* Header fields of interest. */
8269 if (r->anchored_substr) {
8270 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8271 RE_SV_DUMPLEN(r->anchored_substr), 30);
8272 PerlIO_printf(Perl_debug_log,
8273 "anchored %s%s at %"IVdf" ",
8274 s, RE_SV_TAIL(r->anchored_substr),
8275 (IV)r->anchored_offset);
8276 } else if (r->anchored_utf8) {
8277 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8278 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8279 PerlIO_printf(Perl_debug_log,
8280 "anchored utf8 %s%s at %"IVdf" ",
8281 s, RE_SV_TAIL(r->anchored_utf8),
8282 (IV)r->anchored_offset);
8284 if (r->float_substr) {
8285 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8286 RE_SV_DUMPLEN(r->float_substr), 30);
8287 PerlIO_printf(Perl_debug_log,
8288 "floating %s%s at %"IVdf"..%"UVuf" ",
8289 s, RE_SV_TAIL(r->float_substr),
8290 (IV)r->float_min_offset, (UV)r->float_max_offset);
8291 } else if (r->float_utf8) {
8292 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8293 RE_SV_DUMPLEN(r->float_utf8), 30);
8294 PerlIO_printf(Perl_debug_log,
8295 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8296 s, RE_SV_TAIL(r->float_utf8),
8297 (IV)r->float_min_offset, (UV)r->float_max_offset);
8299 if (r->check_substr || r->check_utf8)
8300 PerlIO_printf(Perl_debug_log,
8302 (r->check_substr == r->float_substr
8303 && r->check_utf8 == r->float_utf8
8304 ? "(checking floating" : "(checking anchored"));
8305 if (r->extflags & RXf_NOSCAN)
8306 PerlIO_printf(Perl_debug_log, " noscan");
8307 if (r->extflags & RXf_CHECK_ALL)
8308 PerlIO_printf(Perl_debug_log, " isall");
8309 if (r->check_substr || r->check_utf8)
8310 PerlIO_printf(Perl_debug_log, ") ");
8312 if (ri->regstclass) {
8313 regprop(r, sv, ri->regstclass);
8314 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8316 if (r->extflags & RXf_ANCH) {
8317 PerlIO_printf(Perl_debug_log, "anchored");
8318 if (r->extflags & RXf_ANCH_BOL)
8319 PerlIO_printf(Perl_debug_log, "(BOL)");
8320 if (r->extflags & RXf_ANCH_MBOL)
8321 PerlIO_printf(Perl_debug_log, "(MBOL)");
8322 if (r->extflags & RXf_ANCH_SBOL)
8323 PerlIO_printf(Perl_debug_log, "(SBOL)");
8324 if (r->extflags & RXf_ANCH_GPOS)
8325 PerlIO_printf(Perl_debug_log, "(GPOS)");
8326 PerlIO_putc(Perl_debug_log, ' ');
8328 if (r->extflags & RXf_GPOS_SEEN)
8329 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8330 if (r->intflags & PREGf_SKIP)
8331 PerlIO_printf(Perl_debug_log, "plus ");
8332 if (r->intflags & PREGf_IMPLICIT)
8333 PerlIO_printf(Perl_debug_log, "implicit ");
8334 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8335 if (r->extflags & RXf_EVAL_SEEN)
8336 PerlIO_printf(Perl_debug_log, "with eval ");
8337 PerlIO_printf(Perl_debug_log, "\n");
8339 PERL_UNUSED_CONTEXT;
8341 #endif /* DEBUGGING */
8345 - regprop - printable representation of opcode
8348 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8353 RXi_GET_DECL(prog,progi);
8354 GET_RE_DEBUG_FLAGS_DECL;
8357 sv_setpvn(sv, "", 0);
8359 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8360 /* It would be nice to FAIL() here, but this may be called from
8361 regexec.c, and it would be hard to supply pRExC_state. */
8362 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8363 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
8365 k = PL_regkind[OP(o)];
8368 SV * const dsv = sv_2mortal(newSVpvs(""));
8369 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8370 * is a crude hack but it may be the best for now since
8371 * we have no flag "this EXACTish node was UTF-8"
8373 const char * const s =
8374 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8375 PL_colors[0], PL_colors[1],
8376 PERL_PV_ESCAPE_UNI_DETECT |
8377 PERL_PV_PRETTY_ELIPSES |
8380 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8381 } else if (k == TRIE) {
8382 /* print the details of the trie in dumpuntil instead, as
8383 * progi->data isn't available here */
8384 const char op = OP(o);
8385 const U32 n = ARG(o);
8386 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8387 (reg_ac_data *)progi->data->data[n] :
8389 const reg_trie_data * const trie
8390 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8392 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
8393 DEBUG_TRIE_COMPILE_r(
8394 Perl_sv_catpvf(aTHX_ sv,
8395 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8396 (UV)trie->startstate,
8397 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8398 (UV)trie->wordcount,
8401 (UV)TRIE_CHARCOUNT(trie),
8402 (UV)trie->uniquecharcount
8405 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8407 int rangestart = -1;
8408 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8409 Perl_sv_catpvf(aTHX_ sv, "[");
8410 for (i = 0; i <= 256; i++) {
8411 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8412 if (rangestart == -1)
8414 } else if (rangestart != -1) {
8415 if (i <= rangestart + 3)
8416 for (; rangestart < i; rangestart++)
8417 put_byte(sv, rangestart);
8419 put_byte(sv, rangestart);
8421 put_byte(sv, i - 1);
8426 Perl_sv_catpvf(aTHX_ sv, "]");
8429 } else if (k == CURLY) {
8430 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8431 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8432 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8434 else if (k == WHILEM && o->flags) /* Ordinal/of */
8435 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8436 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8437 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8438 if ( prog->paren_names ) {
8439 if ( k != REF || OP(o) < NREF) {
8440 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8441 SV **name= av_fetch(list, ARG(o), 0 );
8443 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8446 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8447 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8448 I32 *nums=(I32*)SvPVX(sv_dat);
8449 SV **name= av_fetch(list, nums[0], 0 );
8452 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8453 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8454 (n ? "," : ""), (IV)nums[n]);
8456 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8460 } else if (k == GOSUB)
8461 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8462 else if (k == VERB) {
8464 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8465 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8466 } else if (k == LOGICAL)
8467 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8468 else if (k == ANYOF) {
8469 int i, rangestart = -1;
8470 const U8 flags = ANYOF_FLAGS(o);
8472 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8473 static const char * const anyofs[] = {
8506 if (flags & ANYOF_LOCALE)
8507 sv_catpvs(sv, "{loc}");
8508 if (flags & ANYOF_FOLD)
8509 sv_catpvs(sv, "{i}");
8510 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8511 if (flags & ANYOF_INVERT)
8513 for (i = 0; i <= 256; i++) {
8514 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8515 if (rangestart == -1)
8517 } else if (rangestart != -1) {
8518 if (i <= rangestart + 3)
8519 for (; rangestart < i; rangestart++)
8520 put_byte(sv, rangestart);
8522 put_byte(sv, rangestart);
8524 put_byte(sv, i - 1);
8530 if (o->flags & ANYOF_CLASS)
8531 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8532 if (ANYOF_CLASS_TEST(o,i))
8533 sv_catpv(sv, anyofs[i]);
8535 if (flags & ANYOF_UNICODE)
8536 sv_catpvs(sv, "{unicode}");
8537 else if (flags & ANYOF_UNICODE_ALL)
8538 sv_catpvs(sv, "{unicode_all}");
8542 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8546 U8 s[UTF8_MAXBYTES_CASE+1];
8548 for (i = 0; i <= 256; i++) { /* just the first 256 */
8549 uvchr_to_utf8(s, i);
8551 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8552 if (rangestart == -1)
8554 } else if (rangestart != -1) {
8555 if (i <= rangestart + 3)
8556 for (; rangestart < i; rangestart++) {
8557 const U8 * const e = uvchr_to_utf8(s,rangestart);
8559 for(p = s; p < e; p++)
8563 const U8 *e = uvchr_to_utf8(s,rangestart);
8565 for (p = s; p < e; p++)
8568 e = uvchr_to_utf8(s, i-1);
8569 for (p = s; p < e; p++)
8576 sv_catpvs(sv, "..."); /* et cetera */
8580 char *s = savesvpv(lv);
8581 char * const origs = s;
8583 while (*s && *s != '\n')
8587 const char * const t = ++s;
8605 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8607 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8608 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8610 PERL_UNUSED_CONTEXT;
8611 PERL_UNUSED_ARG(sv);
8613 PERL_UNUSED_ARG(prog);
8614 #endif /* DEBUGGING */
8618 Perl_re_intuit_string(pTHX_ regexp *prog)
8619 { /* Assume that RE_INTUIT is set */
8621 GET_RE_DEBUG_FLAGS_DECL;
8622 PERL_UNUSED_CONTEXT;
8626 const char * const s = SvPV_nolen_const(prog->check_substr
8627 ? prog->check_substr : prog->check_utf8);
8629 if (!PL_colorset) reginitcolors();
8630 PerlIO_printf(Perl_debug_log,
8631 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8633 prog->check_substr ? "" : "utf8 ",
8634 PL_colors[5],PL_colors[0],
8637 (strlen(s) > 60 ? "..." : ""));
8640 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8646 handles refcounting and freeing the perl core regexp structure. When
8647 it is necessary to actually free the structure the first thing it
8648 does is call the 'free' method of the regexp_engine associated to to
8649 the regexp, allowing the handling of the void *pprivate; member
8650 first. (This routine is not overridable by extensions, which is why
8651 the extensions free is called first.)
8653 See regdupe and regdupe_internal if you change anything here.
8655 #ifndef PERL_IN_XSUB_RE
8657 Perl_pregfree(pTHX_ struct regexp *r)
8660 GET_RE_DEBUG_FLAGS_DECL;
8662 if (!r || (--r->refcnt > 0))
8665 ReREFCNT_dec(r->mother_re);
8667 CALLREGFREE_PVT(r); /* free the private data */
8669 SvREFCNT_dec(r->paren_names);
8670 Safefree(r->wrapped);
8673 if (r->anchored_substr)
8674 SvREFCNT_dec(r->anchored_substr);
8675 if (r->anchored_utf8)
8676 SvREFCNT_dec(r->anchored_utf8);
8677 if (r->float_substr)
8678 SvREFCNT_dec(r->float_substr);
8680 SvREFCNT_dec(r->float_utf8);
8681 Safefree(r->substrs);
8683 RX_MATCH_COPY_FREE(r);
8684 #ifdef PERL_OLD_COPY_ON_WRITE
8686 SvREFCNT_dec(r->saved_copy);
8695 This is a hacky workaround to the structural issue of match results
8696 being stored in the regexp structure which is in turn stored in
8697 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
8698 could be PL_curpm in multiple contexts, and could require multiple
8699 result sets being associated with the pattern simultaneously, such
8700 as when doing a recursive match with (??{$qr})
8702 The solution is to make a lightweight copy of the regexp structure
8703 when a qr// is returned from the code executed by (??{$qr}) this
8704 lightweight copy doesnt actually own any of its data except for
8705 the starp/end and the actual regexp structure itself.
8711 Perl_reg_temp_copy (pTHX_ struct regexp *r) {
8713 register const I32 npar = r->nparens+1;
8714 (void)ReREFCNT_inc(r);
8715 Newx(ret, 1, regexp);
8716 StructCopy(r, ret, regexp);
8717 Newx(ret->offs, npar, regexp_paren_pair);
8718 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
8721 Newx(ret->substrs, 1, struct reg_substr_data);
8722 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
8724 SvREFCNT_inc_void(ret->anchored_substr);
8725 SvREFCNT_inc_void(ret->anchored_utf8);
8726 SvREFCNT_inc_void(ret->float_substr);
8727 SvREFCNT_inc_void(ret->float_utf8);
8729 /* check_substr and check_utf8, if non-NULL, point to either their
8730 anchored or float namesakes, and don't hold a second reference. */
8732 RX_MATCH_COPIED_off(ret);
8733 #ifdef PERL_OLD_COPY_ON_WRITE
8734 /* this is broken. */
8736 if (ret->saved_copy)
8737 ret->saved_copy=NULL;
8746 /* regfree_internal()
8748 Free the private data in a regexp. This is overloadable by
8749 extensions. Perl takes care of the regexp structure in pregfree(),
8750 this covers the *pprivate pointer which technically perldoesnt
8751 know about, however of course we have to handle the
8752 regexp_internal structure when no extension is in use.
8754 Note this is called before freeing anything in the regexp
8759 Perl_regfree_internal(pTHX_ struct regexp *r)
8763 GET_RE_DEBUG_FLAGS_DECL;
8769 SV *dsv= sv_newmortal();
8770 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8771 dsv, r->precomp, r->prelen, 60);
8772 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8773 PL_colors[4],PL_colors[5],s);
8776 #ifdef RE_TRACK_PATTERN_OFFSETS
8778 Safefree(ri->u.offsets); /* 20010421 MJD */
8781 int n = ri->data->count;
8782 PAD* new_comppad = NULL;
8787 /* If you add a ->what type here, update the comment in regcomp.h */
8788 switch (ri->data->what[n]) {
8792 SvREFCNT_dec((SV*)ri->data->data[n]);
8795 Safefree(ri->data->data[n]);
8798 new_comppad = (AV*)ri->data->data[n];
8801 if (new_comppad == NULL)
8802 Perl_croak(aTHX_ "panic: pregfree comppad");
8803 PAD_SAVE_LOCAL(old_comppad,
8804 /* Watch out for global destruction's random ordering. */
8805 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
8808 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
8811 op_free((OP_4tree*)ri->data->data[n]);
8813 PAD_RESTORE_LOCAL(old_comppad);
8814 SvREFCNT_dec((SV*)new_comppad);
8820 { /* Aho Corasick add-on structure for a trie node.
8821 Used in stclass optimization only */
8823 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
8825 refcount = --aho->refcount;
8828 PerlMemShared_free(aho->states);
8829 PerlMemShared_free(aho->fail);
8830 /* do this last!!!! */
8831 PerlMemShared_free(ri->data->data[n]);
8832 PerlMemShared_free(ri->regstclass);
8838 /* trie structure. */
8840 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
8842 refcount = --trie->refcount;
8845 PerlMemShared_free(trie->charmap);
8846 PerlMemShared_free(trie->states);
8847 PerlMemShared_free(trie->trans);
8849 PerlMemShared_free(trie->bitmap);
8851 PerlMemShared_free(trie->wordlen);
8853 PerlMemShared_free(trie->jump);
8855 PerlMemShared_free(trie->nextword);
8856 /* do this last!!!! */
8857 PerlMemShared_free(ri->data->data[n]);
8862 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
8865 Safefree(ri->data->what);
8872 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
8873 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8874 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
8875 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
8878 re_dup - duplicate a regexp.
8880 This routine is expected to clone a given regexp structure. It is not
8881 compiler under USE_ITHREADS.
8883 After all of the core data stored in struct regexp is duplicated
8884 the regexp_engine.dupe method is used to copy any private data
8885 stored in the *pprivate pointer. This allows extensions to handle
8886 any duplication it needs to do.
8888 See pregfree() and regfree_internal() if you change anything here.
8890 #if defined(USE_ITHREADS)
8891 #ifndef PERL_IN_XSUB_RE
8893 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
8900 return (REGEXP *)NULL;
8902 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
8906 npar = r->nparens+1;
8907 Newx(ret, 1, regexp);
8908 StructCopy(r, ret, regexp);
8909 Newx(ret->offs, npar, regexp_paren_pair);
8910 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
8912 /* no need to copy these */
8913 Newx(ret->swap, npar, regexp_paren_pair);
8917 /* Do it this way to avoid reading from *r after the StructCopy().
8918 That way, if any of the sv_dup_inc()s dislodge *r from the L1
8919 cache, it doesn't matter. */
8920 const bool anchored = r->check_substr == r->anchored_substr;
8921 Newx(ret->substrs, 1, struct reg_substr_data);
8922 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
8924 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
8925 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
8926 ret->float_substr = sv_dup_inc(ret->float_substr, param);
8927 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
8929 /* check_substr and check_utf8, if non-NULL, point to either their
8930 anchored or float namesakes, and don't hold a second reference. */
8932 if (ret->check_substr) {
8934 assert(r->check_utf8 == r->anchored_utf8);
8935 ret->check_substr = ret->anchored_substr;
8936 ret->check_utf8 = ret->anchored_utf8;
8938 assert(r->check_substr == r->float_substr);
8939 assert(r->check_utf8 == r->float_utf8);
8940 ret->check_substr = ret->float_substr;
8941 ret->check_utf8 = ret->float_utf8;
8946 ret->wrapped = SAVEPVN(ret->wrapped, ret->wraplen+1);
8947 ret->precomp = ret->wrapped + (ret->precomp - ret->wrapped);
8948 ret->paren_names = hv_dup_inc(ret->paren_names, param);
8951 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
8953 if (RX_MATCH_COPIED(ret))
8954 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
8957 #ifdef PERL_OLD_COPY_ON_WRITE
8958 ret->saved_copy = NULL;
8961 ret->mother_re = NULL;
8963 ret->seen_evals = 0;
8965 ptr_table_store(PL_ptr_table, r, ret);
8968 #endif /* PERL_IN_XSUB_RE */
8973 This is the internal complement to regdupe() which is used to copy
8974 the structure pointed to by the *pprivate pointer in the regexp.
8975 This is the core version of the extension overridable cloning hook.
8976 The regexp structure being duplicated will be copied by perl prior
8977 to this and will be provided as the regexp *r argument, however
8978 with the /old/ structures pprivate pointer value. Thus this routine
8979 may override any copying normally done by perl.
8981 It returns a pointer to the new regexp_internal structure.
8985 Perl_regdupe_internal(pTHX_ const regexp *r, CLONE_PARAMS *param)
8988 regexp_internal *reti;
8992 npar = r->nparens+1;
8995 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
8996 Copy(ri->program, reti->program, len+1, regnode);
8999 reti->regstclass = NULL;
9003 const int count = ri->data->count;
9006 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9007 char, struct reg_data);
9008 Newx(d->what, count, U8);
9011 for (i = 0; i < count; i++) {
9012 d->what[i] = ri->data->what[i];
9013 switch (d->what[i]) {
9014 /* legal options are one of: sSfpontTu
9015 see also regcomp.h and pregfree() */
9018 case 'p': /* actually an AV, but the dup function is identical. */
9019 case 'u': /* actually an HV, but the dup function is identical. */
9020 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
9023 /* This is cheating. */
9024 Newx(d->data[i], 1, struct regnode_charclass_class);
9025 StructCopy(ri->data->data[i], d->data[i],
9026 struct regnode_charclass_class);
9027 reti->regstclass = (regnode*)d->data[i];
9030 /* Compiled op trees are readonly and in shared memory,
9031 and can thus be shared without duplication. */
9033 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9037 /* Trie stclasses are readonly and can thus be shared
9038 * without duplication. We free the stclass in pregfree
9039 * when the corresponding reg_ac_data struct is freed.
9041 reti->regstclass= ri->regstclass;
9045 ((reg_trie_data*)ri->data->data[i])->refcount++;
9049 d->data[i] = ri->data->data[i];
9052 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9061 reti->name_list_idx = ri->name_list_idx;
9063 #ifdef RE_TRACK_PATTERN_OFFSETS
9064 if (ri->u.offsets) {
9065 Newx(reti->u.offsets, 2*len+1, U32);
9066 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9069 SetProgLen(reti,len);
9075 #endif /* USE_ITHREADS */
9080 converts a regexp embedded in a MAGIC struct to its stringified form,
9081 caching the converted form in the struct and returns the cached
9084 If lp is nonnull then it is used to return the length of the
9087 If flags is nonnull and the returned string contains UTF8 then
9088 (*flags & 1) will be true.
9090 If haseval is nonnull then it is used to return whether the pattern
9093 Normally called via macro:
9095 CALLREG_STRINGIFY(mg,&len,&utf8);
9099 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9101 See sv_2pv_flags() in sv.c for an example of internal usage.
9104 #ifndef PERL_IN_XSUB_RE
9107 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9109 const regexp * const re = (regexp *)mg->mg_obj;
9111 *haseval = re->seen_evals;
9113 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
9120 - regnext - dig the "next" pointer out of a node
9123 Perl_regnext(pTHX_ register regnode *p)
9126 register I32 offset;
9131 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9140 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9143 STRLEN l1 = strlen(pat1);
9144 STRLEN l2 = strlen(pat2);
9147 const char *message;
9153 Copy(pat1, buf, l1 , char);
9154 Copy(pat2, buf + l1, l2 , char);
9155 buf[l1 + l2] = '\n';
9156 buf[l1 + l2 + 1] = '\0';
9158 /* ANSI variant takes additional second argument */
9159 va_start(args, pat2);
9163 msv = vmess(buf, &args);
9165 message = SvPV_const(msv,l1);
9168 Copy(message, buf, l1 , char);
9169 buf[l1-1] = '\0'; /* Overwrite \n */
9170 Perl_croak(aTHX_ "%s", buf);
9173 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9175 #ifndef PERL_IN_XSUB_RE
9177 Perl_save_re_context(pTHX)
9181 struct re_save_state *state;
9183 SAVEVPTR(PL_curcop);
9184 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9186 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9187 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9188 SSPUSHINT(SAVEt_RE_STATE);
9190 Copy(&PL_reg_state, state, 1, struct re_save_state);
9192 PL_reg_start_tmp = 0;
9193 PL_reg_start_tmpl = 0;
9194 PL_reg_oldsaved = NULL;
9195 PL_reg_oldsavedlen = 0;
9197 PL_reg_leftiter = 0;
9198 PL_reg_poscache = NULL;
9199 PL_reg_poscache_size = 0;
9200 #ifdef PERL_OLD_COPY_ON_WRITE
9204 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9206 const REGEXP * const rx = PM_GETRE(PL_curpm);
9209 for (i = 1; i <= rx->nparens; i++) {
9210 char digits[TYPE_CHARS(long)];
9211 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9212 GV *const *const gvp
9213 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9216 GV * const gv = *gvp;
9217 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9227 clear_re(pTHX_ void *r)
9230 ReREFCNT_dec((regexp *)r);
9236 S_put_byte(pTHX_ SV *sv, int c)
9238 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9239 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9240 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9241 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9243 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9247 #define CLEAR_OPTSTART \
9248 if (optstart) STMT_START { \
9249 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9253 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9255 STATIC const regnode *
9256 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9257 const regnode *last, const regnode *plast,
9258 SV* sv, I32 indent, U32 depth)
9261 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9262 register const regnode *next;
9263 const regnode *optstart= NULL;
9266 GET_RE_DEBUG_FLAGS_DECL;
9268 #ifdef DEBUG_DUMPUNTIL
9269 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9270 last ? last-start : 0,plast ? plast-start : 0);
9273 if (plast && plast < last)
9276 while (PL_regkind[op] != END && (!last || node < last)) {
9277 /* While that wasn't END last time... */
9280 if (op == CLOSE || op == WHILEM)
9282 next = regnext((regnode *)node);
9285 if (OP(node) == OPTIMIZED) {
9286 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9293 regprop(r, sv, node);
9294 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9295 (int)(2*indent + 1), "", SvPVX_const(sv));
9297 if (OP(node) != OPTIMIZED) {
9298 if (next == NULL) /* Next ptr. */
9299 PerlIO_printf(Perl_debug_log, " (0)");
9300 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9301 PerlIO_printf(Perl_debug_log, " (FAIL)");
9303 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9304 (void)PerlIO_putc(Perl_debug_log, '\n');
9308 if (PL_regkind[(U8)op] == BRANCHJ) {
9311 register const regnode *nnode = (OP(next) == LONGJMP
9312 ? regnext((regnode *)next)
9314 if (last && nnode > last)
9316 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9319 else if (PL_regkind[(U8)op] == BRANCH) {
9321 DUMPUNTIL(NEXTOPER(node), next);
9323 else if ( PL_regkind[(U8)op] == TRIE ) {
9324 const regnode *this_trie = node;
9325 const char op = OP(node);
9326 const U32 n = ARG(node);
9327 const reg_ac_data * const ac = op>=AHOCORASICK ?
9328 (reg_ac_data *)ri->data->data[n] :
9330 const reg_trie_data * const trie =
9331 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9333 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9335 const regnode *nextbranch= NULL;
9337 sv_setpvn(sv, "", 0);
9338 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9339 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9341 PerlIO_printf(Perl_debug_log, "%*s%s ",
9342 (int)(2*(indent+3)), "",
9343 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9344 PL_colors[0], PL_colors[1],
9345 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9346 PERL_PV_PRETTY_ELIPSES |
9352 U16 dist= trie->jump[word_idx+1];
9353 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9354 (UV)((dist ? this_trie + dist : next) - start));
9357 nextbranch= this_trie + trie->jump[0];
9358 DUMPUNTIL(this_trie + dist, nextbranch);
9360 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9361 nextbranch= regnext((regnode *)nextbranch);
9363 PerlIO_printf(Perl_debug_log, "\n");
9366 if (last && next > last)
9371 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9372 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9373 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9375 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9377 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9379 else if ( op == PLUS || op == STAR) {
9380 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9382 else if (op == ANYOF) {
9383 /* arglen 1 + class block */
9384 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9385 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9386 node = NEXTOPER(node);
9388 else if (PL_regkind[(U8)op] == EXACT) {
9389 /* Literal string, where present. */
9390 node += NODE_SZ_STR(node) - 1;
9391 node = NEXTOPER(node);
9394 node = NEXTOPER(node);
9395 node += regarglen[(U8)op];
9397 if (op == CURLYX || op == OPEN)
9401 #ifdef DEBUG_DUMPUNTIL
9402 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9407 #endif /* DEBUGGING */
9411 * c-indentation-style: bsd
9413 * indent-tabs-mode: t
9416 * ex: set ts=8 sts=4 sw=4 noet: