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)
2403 #define CASE_SYNST_FNC(nAmE) \
2405 if (flags & SCF_DO_STCLASS_AND) { \
2406 for (value = 0; value < 256; value++) \
2407 if (!is_ ## nAmE ## _cp(value)) \
2408 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2411 for (value = 0; value < 256; value++) \
2412 if (is_ ## nAmE ## _cp(value)) \
2413 ANYOF_BITMAP_SET(data->start_class, value); \
2417 if (flags & SCF_DO_STCLASS_AND) { \
2418 for (value = 0; value < 256; value++) \
2419 if (is_ ## nAmE ## _cp(value)) \
2420 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2423 for (value = 0; value < 256; value++) \
2424 if (!is_ ## nAmE ## _cp(value)) \
2425 ANYOF_BITMAP_SET(data->start_class, value); \
2432 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2433 I32 *minlenp, I32 *deltap,
2438 struct regnode_charclass_class *and_withp,
2439 U32 flags, U32 depth)
2440 /* scanp: Start here (read-write). */
2441 /* deltap: Write maxlen-minlen here. */
2442 /* last: Stop before this one. */
2443 /* data: string data about the pattern */
2444 /* stopparen: treat close N as END */
2445 /* recursed: which subroutines have we recursed into */
2446 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2449 I32 min = 0, pars = 0, code;
2450 regnode *scan = *scanp, *next;
2452 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2453 int is_inf_internal = 0; /* The studied chunk is infinite */
2454 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2455 scan_data_t data_fake;
2456 SV *re_trie_maxbuff = NULL;
2457 regnode *first_non_open = scan;
2458 I32 stopmin = I32_MAX;
2459 scan_frame *frame = NULL;
2461 GET_RE_DEBUG_FLAGS_DECL;
2464 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2468 while (first_non_open && OP(first_non_open) == OPEN)
2469 first_non_open=regnext(first_non_open);
2474 while ( scan && OP(scan) != END && scan < last ){
2475 /* Peephole optimizer: */
2476 DEBUG_STUDYDATA("Peep:", data,depth);
2477 DEBUG_PEEP("Peep",scan,depth);
2478 JOIN_EXACT(scan,&min,0);
2480 /* Follow the next-chain of the current node and optimize
2481 away all the NOTHINGs from it. */
2482 if (OP(scan) != CURLYX) {
2483 const int max = (reg_off_by_arg[OP(scan)]
2485 /* I32 may be smaller than U16 on CRAYs! */
2486 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2487 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2491 /* Skip NOTHING and LONGJMP. */
2492 while ((n = regnext(n))
2493 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2494 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2495 && off + noff < max)
2497 if (reg_off_by_arg[OP(scan)])
2500 NEXT_OFF(scan) = off;
2505 /* The principal pseudo-switch. Cannot be a switch, since we
2506 look into several different things. */
2507 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2508 || OP(scan) == IFTHEN) {
2509 next = regnext(scan);
2511 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2513 if (OP(next) == code || code == IFTHEN) {
2514 /* NOTE - There is similar code to this block below for handling
2515 TRIE nodes on a re-study. If you change stuff here check there
2517 I32 max1 = 0, min1 = I32_MAX, num = 0;
2518 struct regnode_charclass_class accum;
2519 regnode * const startbranch=scan;
2521 if (flags & SCF_DO_SUBSTR)
2522 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2523 if (flags & SCF_DO_STCLASS)
2524 cl_init_zero(pRExC_state, &accum);
2526 while (OP(scan) == code) {
2527 I32 deltanext, minnext, f = 0, fake;
2528 struct regnode_charclass_class this_class;
2531 data_fake.flags = 0;
2533 data_fake.whilem_c = data->whilem_c;
2534 data_fake.last_closep = data->last_closep;
2537 data_fake.last_closep = &fake;
2539 data_fake.pos_delta = delta;
2540 next = regnext(scan);
2541 scan = NEXTOPER(scan);
2543 scan = NEXTOPER(scan);
2544 if (flags & SCF_DO_STCLASS) {
2545 cl_init(pRExC_state, &this_class);
2546 data_fake.start_class = &this_class;
2547 f = SCF_DO_STCLASS_AND;
2549 if (flags & SCF_WHILEM_VISITED_POS)
2550 f |= SCF_WHILEM_VISITED_POS;
2552 /* we suppose the run is continuous, last=next...*/
2553 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2555 stopparen, recursed, NULL, f,depth+1);
2558 if (max1 < minnext + deltanext)
2559 max1 = minnext + deltanext;
2560 if (deltanext == I32_MAX)
2561 is_inf = is_inf_internal = 1;
2563 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2565 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2566 if ( stopmin > minnext)
2567 stopmin = min + min1;
2568 flags &= ~SCF_DO_SUBSTR;
2570 data->flags |= SCF_SEEN_ACCEPT;
2573 if (data_fake.flags & SF_HAS_EVAL)
2574 data->flags |= SF_HAS_EVAL;
2575 data->whilem_c = data_fake.whilem_c;
2577 if (flags & SCF_DO_STCLASS)
2578 cl_or(pRExC_state, &accum, &this_class);
2580 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2582 if (flags & SCF_DO_SUBSTR) {
2583 data->pos_min += min1;
2584 data->pos_delta += max1 - min1;
2585 if (max1 != min1 || is_inf)
2586 data->longest = &(data->longest_float);
2589 delta += max1 - min1;
2590 if (flags & SCF_DO_STCLASS_OR) {
2591 cl_or(pRExC_state, data->start_class, &accum);
2593 cl_and(data->start_class, and_withp);
2594 flags &= ~SCF_DO_STCLASS;
2597 else if (flags & SCF_DO_STCLASS_AND) {
2599 cl_and(data->start_class, &accum);
2600 flags &= ~SCF_DO_STCLASS;
2603 /* Switch to OR mode: cache the old value of
2604 * data->start_class */
2606 StructCopy(data->start_class, and_withp,
2607 struct regnode_charclass_class);
2608 flags &= ~SCF_DO_STCLASS_AND;
2609 StructCopy(&accum, data->start_class,
2610 struct regnode_charclass_class);
2611 flags |= SCF_DO_STCLASS_OR;
2612 data->start_class->flags |= ANYOF_EOS;
2616 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2619 Assuming this was/is a branch we are dealing with: 'scan' now
2620 points at the item that follows the branch sequence, whatever
2621 it is. We now start at the beginning of the sequence and look
2628 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2630 If we can find such a subseqence we need to turn the first
2631 element into a trie and then add the subsequent branch exact
2632 strings to the trie.
2636 1. patterns where the whole set of branch can be converted.
2638 2. patterns where only a subset can be converted.
2640 In case 1 we can replace the whole set with a single regop
2641 for the trie. In case 2 we need to keep the start and end
2644 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2645 becomes BRANCH TRIE; BRANCH X;
2647 There is an additional case, that being where there is a
2648 common prefix, which gets split out into an EXACT like node
2649 preceding the TRIE node.
2651 If x(1..n)==tail then we can do a simple trie, if not we make
2652 a "jump" trie, such that when we match the appropriate word
2653 we "jump" to the appopriate tail node. Essentailly we turn
2654 a nested if into a case structure of sorts.
2659 if (!re_trie_maxbuff) {
2660 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2661 if (!SvIOK(re_trie_maxbuff))
2662 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2664 if ( SvIV(re_trie_maxbuff)>=0 ) {
2666 regnode *first = (regnode *)NULL;
2667 regnode *last = (regnode *)NULL;
2668 regnode *tail = scan;
2673 SV * const mysv = sv_newmortal(); /* for dumping */
2675 /* var tail is used because there may be a TAIL
2676 regop in the way. Ie, the exacts will point to the
2677 thing following the TAIL, but the last branch will
2678 point at the TAIL. So we advance tail. If we
2679 have nested (?:) we may have to move through several
2683 while ( OP( tail ) == TAIL ) {
2684 /* this is the TAIL generated by (?:) */
2685 tail = regnext( tail );
2690 regprop(RExC_rx, mysv, tail );
2691 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2692 (int)depth * 2 + 2, "",
2693 "Looking for TRIE'able sequences. Tail node is: ",
2694 SvPV_nolen_const( mysv )
2700 step through the branches, cur represents each
2701 branch, noper is the first thing to be matched
2702 as part of that branch and noper_next is the
2703 regnext() of that node. if noper is an EXACT
2704 and noper_next is the same as scan (our current
2705 position in the regex) then the EXACT branch is
2706 a possible optimization target. Once we have
2707 two or more consequetive such branches we can
2708 create a trie of the EXACT's contents and stich
2709 it in place. If the sequence represents all of
2710 the branches we eliminate the whole thing and
2711 replace it with a single TRIE. If it is a
2712 subsequence then we need to stitch it in. This
2713 means the first branch has to remain, and needs
2714 to be repointed at the item on the branch chain
2715 following the last branch optimized. This could
2716 be either a BRANCH, in which case the
2717 subsequence is internal, or it could be the
2718 item following the branch sequence in which
2719 case the subsequence is at the end.
2723 /* dont use tail as the end marker for this traverse */
2724 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2725 regnode * const noper = NEXTOPER( cur );
2726 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2727 regnode * const noper_next = regnext( noper );
2731 regprop(RExC_rx, mysv, cur);
2732 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2733 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2735 regprop(RExC_rx, mysv, noper);
2736 PerlIO_printf( Perl_debug_log, " -> %s",
2737 SvPV_nolen_const(mysv));
2740 regprop(RExC_rx, mysv, noper_next );
2741 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2742 SvPV_nolen_const(mysv));
2744 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2745 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2747 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2748 : PL_regkind[ OP( noper ) ] == EXACT )
2749 || OP(noper) == NOTHING )
2751 && noper_next == tail
2756 if ( !first || optype == NOTHING ) {
2757 if (!first) first = cur;
2758 optype = OP( noper );
2764 make_trie( pRExC_state,
2765 startbranch, first, cur, tail, count,
2768 if ( PL_regkind[ OP( noper ) ] == EXACT
2770 && noper_next == tail
2775 optype = OP( noper );
2785 regprop(RExC_rx, mysv, cur);
2786 PerlIO_printf( Perl_debug_log,
2787 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2788 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2792 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2793 #ifdef TRIE_STUDY_OPT
2794 if ( ((made == MADE_EXACT_TRIE &&
2795 startbranch == first)
2796 || ( first_non_open == first )) &&
2798 flags |= SCF_TRIE_RESTUDY;
2799 if ( startbranch == first
2802 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2812 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2813 scan = NEXTOPER(NEXTOPER(scan));
2814 } else /* single branch is optimized. */
2815 scan = NEXTOPER(scan);
2817 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2818 scan_frame *newframe = NULL;
2823 if (OP(scan) != SUSPEND) {
2824 /* set the pointer */
2825 if (OP(scan) == GOSUB) {
2827 RExC_recurse[ARG2L(scan)] = scan;
2828 start = RExC_open_parens[paren-1];
2829 end = RExC_close_parens[paren-1];
2832 start = RExC_rxi->program + 1;
2836 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2837 SAVEFREEPV(recursed);
2839 if (!PAREN_TEST(recursed,paren+1)) {
2840 PAREN_SET(recursed,paren+1);
2841 Newx(newframe,1,scan_frame);
2843 if (flags & SCF_DO_SUBSTR) {
2844 SCAN_COMMIT(pRExC_state,data,minlenp);
2845 data->longest = &(data->longest_float);
2847 is_inf = is_inf_internal = 1;
2848 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2849 cl_anything(pRExC_state, data->start_class);
2850 flags &= ~SCF_DO_STCLASS;
2853 Newx(newframe,1,scan_frame);
2856 end = regnext(scan);
2861 SAVEFREEPV(newframe);
2862 newframe->next = regnext(scan);
2863 newframe->last = last;
2864 newframe->stop = stopparen;
2865 newframe->prev = frame;
2875 else if (OP(scan) == EXACT) {
2876 I32 l = STR_LEN(scan);
2879 const U8 * const s = (U8*)STRING(scan);
2880 l = utf8_length(s, s + l);
2881 uc = utf8_to_uvchr(s, NULL);
2883 uc = *((U8*)STRING(scan));
2886 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2887 /* The code below prefers earlier match for fixed
2888 offset, later match for variable offset. */
2889 if (data->last_end == -1) { /* Update the start info. */
2890 data->last_start_min = data->pos_min;
2891 data->last_start_max = is_inf
2892 ? I32_MAX : data->pos_min + data->pos_delta;
2894 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2896 SvUTF8_on(data->last_found);
2898 SV * const sv = data->last_found;
2899 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2900 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2901 if (mg && mg->mg_len >= 0)
2902 mg->mg_len += utf8_length((U8*)STRING(scan),
2903 (U8*)STRING(scan)+STR_LEN(scan));
2905 data->last_end = data->pos_min + l;
2906 data->pos_min += l; /* As in the first entry. */
2907 data->flags &= ~SF_BEFORE_EOL;
2909 if (flags & SCF_DO_STCLASS_AND) {
2910 /* Check whether it is compatible with what we know already! */
2914 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2915 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2916 && (!(data->start_class->flags & ANYOF_FOLD)
2917 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2920 ANYOF_CLASS_ZERO(data->start_class);
2921 ANYOF_BITMAP_ZERO(data->start_class);
2923 ANYOF_BITMAP_SET(data->start_class, uc);
2924 data->start_class->flags &= ~ANYOF_EOS;
2926 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2928 else if (flags & SCF_DO_STCLASS_OR) {
2929 /* false positive possible if the class is case-folded */
2931 ANYOF_BITMAP_SET(data->start_class, uc);
2933 data->start_class->flags |= ANYOF_UNICODE_ALL;
2934 data->start_class->flags &= ~ANYOF_EOS;
2935 cl_and(data->start_class, and_withp);
2937 flags &= ~SCF_DO_STCLASS;
2939 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2940 I32 l = STR_LEN(scan);
2941 UV uc = *((U8*)STRING(scan));
2943 /* Search for fixed substrings supports EXACT only. */
2944 if (flags & SCF_DO_SUBSTR) {
2946 SCAN_COMMIT(pRExC_state, data, minlenp);
2949 const U8 * const s = (U8 *)STRING(scan);
2950 l = utf8_length(s, s + l);
2951 uc = utf8_to_uvchr(s, NULL);
2954 if (flags & SCF_DO_SUBSTR)
2956 if (flags & SCF_DO_STCLASS_AND) {
2957 /* Check whether it is compatible with what we know already! */
2961 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2962 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2963 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2965 ANYOF_CLASS_ZERO(data->start_class);
2966 ANYOF_BITMAP_ZERO(data->start_class);
2968 ANYOF_BITMAP_SET(data->start_class, uc);
2969 data->start_class->flags &= ~ANYOF_EOS;
2970 data->start_class->flags |= ANYOF_FOLD;
2971 if (OP(scan) == EXACTFL)
2972 data->start_class->flags |= ANYOF_LOCALE;
2975 else if (flags & SCF_DO_STCLASS_OR) {
2976 if (data->start_class->flags & ANYOF_FOLD) {
2977 /* false positive possible if the class is case-folded.
2978 Assume that the locale settings are the same... */
2980 ANYOF_BITMAP_SET(data->start_class, uc);
2981 data->start_class->flags &= ~ANYOF_EOS;
2983 cl_and(data->start_class, and_withp);
2985 flags &= ~SCF_DO_STCLASS;
2987 else if (strchr((const char*)PL_varies,OP(scan))) {
2988 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2989 I32 f = flags, pos_before = 0;
2990 regnode * const oscan = scan;
2991 struct regnode_charclass_class this_class;
2992 struct regnode_charclass_class *oclass = NULL;
2993 I32 next_is_eval = 0;
2995 switch (PL_regkind[OP(scan)]) {
2996 case WHILEM: /* End of (?:...)* . */
2997 scan = NEXTOPER(scan);
3000 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3001 next = NEXTOPER(scan);
3002 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3004 maxcount = REG_INFTY;
3005 next = regnext(scan);
3006 scan = NEXTOPER(scan);
3010 if (flags & SCF_DO_SUBSTR)
3015 if (flags & SCF_DO_STCLASS) {
3017 maxcount = REG_INFTY;
3018 next = regnext(scan);
3019 scan = NEXTOPER(scan);
3022 is_inf = is_inf_internal = 1;
3023 scan = regnext(scan);
3024 if (flags & SCF_DO_SUBSTR) {
3025 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3026 data->longest = &(data->longest_float);
3028 goto optimize_curly_tail;
3030 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3031 && (scan->flags == stopparen))
3036 mincount = ARG1(scan);
3037 maxcount = ARG2(scan);
3039 next = regnext(scan);
3040 if (OP(scan) == CURLYX) {
3041 I32 lp = (data ? *(data->last_closep) : 0);
3042 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3044 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3045 next_is_eval = (OP(scan) == EVAL);
3047 if (flags & SCF_DO_SUBSTR) {
3048 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3049 pos_before = data->pos_min;
3053 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3055 data->flags |= SF_IS_INF;
3057 if (flags & SCF_DO_STCLASS) {
3058 cl_init(pRExC_state, &this_class);
3059 oclass = data->start_class;
3060 data->start_class = &this_class;
3061 f |= SCF_DO_STCLASS_AND;
3062 f &= ~SCF_DO_STCLASS_OR;
3064 /* These are the cases when once a subexpression
3065 fails at a particular position, it cannot succeed
3066 even after backtracking at the enclosing scope.
3068 XXXX what if minimal match and we are at the
3069 initial run of {n,m}? */
3070 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3071 f &= ~SCF_WHILEM_VISITED_POS;
3073 /* This will finish on WHILEM, setting scan, or on NULL: */
3074 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3075 last, data, stopparen, recursed, NULL,
3077 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3079 if (flags & SCF_DO_STCLASS)
3080 data->start_class = oclass;
3081 if (mincount == 0 || minnext == 0) {
3082 if (flags & SCF_DO_STCLASS_OR) {
3083 cl_or(pRExC_state, data->start_class, &this_class);
3085 else if (flags & SCF_DO_STCLASS_AND) {
3086 /* Switch to OR mode: cache the old value of
3087 * data->start_class */
3089 StructCopy(data->start_class, and_withp,
3090 struct regnode_charclass_class);
3091 flags &= ~SCF_DO_STCLASS_AND;
3092 StructCopy(&this_class, data->start_class,
3093 struct regnode_charclass_class);
3094 flags |= SCF_DO_STCLASS_OR;
3095 data->start_class->flags |= ANYOF_EOS;
3097 } else { /* Non-zero len */
3098 if (flags & SCF_DO_STCLASS_OR) {
3099 cl_or(pRExC_state, data->start_class, &this_class);
3100 cl_and(data->start_class, and_withp);
3102 else if (flags & SCF_DO_STCLASS_AND)
3103 cl_and(data->start_class, &this_class);
3104 flags &= ~SCF_DO_STCLASS;
3106 if (!scan) /* It was not CURLYX, but CURLY. */
3108 if ( /* ? quantifier ok, except for (?{ ... }) */
3109 (next_is_eval || !(mincount == 0 && maxcount == 1))
3110 && (minnext == 0) && (deltanext == 0)
3111 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3112 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3113 && ckWARN(WARN_REGEXP))
3116 "Quantifier unexpected on zero-length expression");
3119 min += minnext * mincount;
3120 is_inf_internal |= ((maxcount == REG_INFTY
3121 && (minnext + deltanext) > 0)
3122 || deltanext == I32_MAX);
3123 is_inf |= is_inf_internal;
3124 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3126 /* Try powerful optimization CURLYX => CURLYN. */
3127 if ( OP(oscan) == CURLYX && data
3128 && data->flags & SF_IN_PAR
3129 && !(data->flags & SF_HAS_EVAL)
3130 && !deltanext && minnext == 1 ) {
3131 /* Try to optimize to CURLYN. */
3132 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3133 regnode * const nxt1 = nxt;
3140 if (!strchr((const char*)PL_simple,OP(nxt))
3141 && !(PL_regkind[OP(nxt)] == EXACT
3142 && STR_LEN(nxt) == 1))
3148 if (OP(nxt) != CLOSE)
3150 if (RExC_open_parens) {
3151 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3152 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3154 /* Now we know that nxt2 is the only contents: */
3155 oscan->flags = (U8)ARG(nxt);
3157 OP(nxt1) = NOTHING; /* was OPEN. */
3160 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3161 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3162 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3163 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3164 OP(nxt + 1) = OPTIMIZED; /* was count. */
3165 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3170 /* Try optimization CURLYX => CURLYM. */
3171 if ( OP(oscan) == CURLYX && data
3172 && !(data->flags & SF_HAS_PAR)
3173 && !(data->flags & SF_HAS_EVAL)
3174 && !deltanext /* atom is fixed width */
3175 && minnext != 0 /* CURLYM can't handle zero width */
3177 /* XXXX How to optimize if data == 0? */
3178 /* Optimize to a simpler form. */
3179 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3183 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3184 && (OP(nxt2) != WHILEM))
3186 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3187 /* Need to optimize away parenths. */
3188 if (data->flags & SF_IN_PAR) {
3189 /* Set the parenth number. */
3190 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3192 if (OP(nxt) != CLOSE)
3193 FAIL("Panic opt close");
3194 oscan->flags = (U8)ARG(nxt);
3195 if (RExC_open_parens) {
3196 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3197 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3199 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3200 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3203 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3204 OP(nxt + 1) = OPTIMIZED; /* was count. */
3205 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3206 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3209 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3210 regnode *nnxt = regnext(nxt1);
3213 if (reg_off_by_arg[OP(nxt1)])
3214 ARG_SET(nxt1, nxt2 - nxt1);
3215 else if (nxt2 - nxt1 < U16_MAX)
3216 NEXT_OFF(nxt1) = nxt2 - nxt1;
3218 OP(nxt) = NOTHING; /* Cannot beautify */
3223 /* Optimize again: */
3224 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3225 NULL, stopparen, recursed, NULL, 0,depth+1);
3230 else if ((OP(oscan) == CURLYX)
3231 && (flags & SCF_WHILEM_VISITED_POS)
3232 /* See the comment on a similar expression above.
3233 However, this time it not a subexpression
3234 we care about, but the expression itself. */
3235 && (maxcount == REG_INFTY)
3236 && data && ++data->whilem_c < 16) {
3237 /* This stays as CURLYX, we can put the count/of pair. */
3238 /* Find WHILEM (as in regexec.c) */
3239 regnode *nxt = oscan + NEXT_OFF(oscan);
3241 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3243 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3244 | (RExC_whilem_seen << 4)); /* On WHILEM */
3246 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3248 if (flags & SCF_DO_SUBSTR) {
3249 SV *last_str = NULL;
3250 int counted = mincount != 0;
3252 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3253 #if defined(SPARC64_GCC_WORKAROUND)
3256 const char *s = NULL;
3259 if (pos_before >= data->last_start_min)
3262 b = data->last_start_min;
3265 s = SvPV_const(data->last_found, l);
3266 old = b - data->last_start_min;
3269 I32 b = pos_before >= data->last_start_min
3270 ? pos_before : data->last_start_min;
3272 const char * const s = SvPV_const(data->last_found, l);
3273 I32 old = b - data->last_start_min;
3277 old = utf8_hop((U8*)s, old) - (U8*)s;
3280 /* Get the added string: */
3281 last_str = newSVpvn(s + old, l);
3283 SvUTF8_on(last_str);
3284 if (deltanext == 0 && pos_before == b) {
3285 /* What was added is a constant string */
3287 SvGROW(last_str, (mincount * l) + 1);
3288 repeatcpy(SvPVX(last_str) + l,
3289 SvPVX_const(last_str), l, mincount - 1);
3290 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3291 /* Add additional parts. */
3292 SvCUR_set(data->last_found,
3293 SvCUR(data->last_found) - l);
3294 sv_catsv(data->last_found, last_str);
3296 SV * sv = data->last_found;
3298 SvUTF8(sv) && SvMAGICAL(sv) ?
3299 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3300 if (mg && mg->mg_len >= 0)
3301 mg->mg_len += CHR_SVLEN(last_str);
3303 data->last_end += l * (mincount - 1);
3306 /* start offset must point into the last copy */
3307 data->last_start_min += minnext * (mincount - 1);
3308 data->last_start_max += is_inf ? I32_MAX
3309 : (maxcount - 1) * (minnext + data->pos_delta);
3312 /* It is counted once already... */
3313 data->pos_min += minnext * (mincount - counted);
3314 data->pos_delta += - counted * deltanext +
3315 (minnext + deltanext) * maxcount - minnext * mincount;
3316 if (mincount != maxcount) {
3317 /* Cannot extend fixed substrings found inside
3319 SCAN_COMMIT(pRExC_state,data,minlenp);
3320 if (mincount && last_str) {
3321 SV * const sv = data->last_found;
3322 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3323 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3327 sv_setsv(sv, last_str);
3328 data->last_end = data->pos_min;
3329 data->last_start_min =
3330 data->pos_min - CHR_SVLEN(last_str);
3331 data->last_start_max = is_inf
3333 : data->pos_min + data->pos_delta
3334 - CHR_SVLEN(last_str);
3336 data->longest = &(data->longest_float);
3338 SvREFCNT_dec(last_str);
3340 if (data && (fl & SF_HAS_EVAL))
3341 data->flags |= SF_HAS_EVAL;
3342 optimize_curly_tail:
3343 if (OP(oscan) != CURLYX) {
3344 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3346 NEXT_OFF(oscan) += NEXT_OFF(next);
3349 default: /* REF and CLUMP only? */
3350 if (flags & SCF_DO_SUBSTR) {
3351 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3352 data->longest = &(data->longest_float);
3354 is_inf = is_inf_internal = 1;
3355 if (flags & SCF_DO_STCLASS_OR)
3356 cl_anything(pRExC_state, data->start_class);
3357 flags &= ~SCF_DO_STCLASS;
3361 else if (OP(scan) == LNBREAK) {
3362 if (flags & SCF_DO_STCLASS) {
3364 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3365 if (flags & SCF_DO_STCLASS_AND) {
3366 for (value = 0; value < 256; value++)
3367 if (!is_VERTWS_cp(value))
3368 ANYOF_BITMAP_CLEAR(data->start_class, value);
3371 for (value = 0; value < 256; value++)
3372 if (is_VERTWS_cp(value))
3373 ANYOF_BITMAP_SET(data->start_class, value);
3375 if (flags & SCF_DO_STCLASS_OR)
3376 cl_and(data->start_class, and_withp);
3377 flags &= ~SCF_DO_STCLASS;
3381 if (flags & SCF_DO_SUBSTR) {
3382 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3384 data->pos_delta += 1;
3385 data->longest = &(data->longest_float);
3389 else if (OP(scan) == FOLDCHAR) {
3390 int d = ARG(scan)==0xDF ? 1 : 2;
3391 flags &= ~SCF_DO_STCLASS;
3394 if (flags & SCF_DO_SUBSTR) {
3395 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3397 data->pos_delta += d;
3398 data->longest = &(data->longest_float);
3401 else if (strchr((const char*)PL_simple,OP(scan))) {
3404 if (flags & SCF_DO_SUBSTR) {
3405 SCAN_COMMIT(pRExC_state,data,minlenp);
3409 if (flags & SCF_DO_STCLASS) {
3410 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3412 /* Some of the logic below assumes that switching
3413 locale on will only add false positives. */
3414 switch (PL_regkind[OP(scan)]) {
3418 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3419 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3420 cl_anything(pRExC_state, data->start_class);
3423 if (OP(scan) == SANY)
3425 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3426 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3427 || (data->start_class->flags & ANYOF_CLASS));
3428 cl_anything(pRExC_state, data->start_class);
3430 if (flags & SCF_DO_STCLASS_AND || !value)
3431 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3434 if (flags & SCF_DO_STCLASS_AND)
3435 cl_and(data->start_class,
3436 (struct regnode_charclass_class*)scan);
3438 cl_or(pRExC_state, data->start_class,
3439 (struct regnode_charclass_class*)scan);
3442 if (flags & SCF_DO_STCLASS_AND) {
3443 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3444 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3445 for (value = 0; value < 256; value++)
3446 if (!isALNUM(value))
3447 ANYOF_BITMAP_CLEAR(data->start_class, value);
3451 if (data->start_class->flags & ANYOF_LOCALE)
3452 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3454 for (value = 0; value < 256; value++)
3456 ANYOF_BITMAP_SET(data->start_class, value);
3461 if (flags & SCF_DO_STCLASS_AND) {
3462 if (data->start_class->flags & ANYOF_LOCALE)
3463 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3466 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3467 data->start_class->flags |= ANYOF_LOCALE;
3471 if (flags & SCF_DO_STCLASS_AND) {
3472 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3473 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3474 for (value = 0; value < 256; value++)
3476 ANYOF_BITMAP_CLEAR(data->start_class, value);
3480 if (data->start_class->flags & ANYOF_LOCALE)
3481 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3483 for (value = 0; value < 256; value++)
3484 if (!isALNUM(value))
3485 ANYOF_BITMAP_SET(data->start_class, value);
3490 if (flags & SCF_DO_STCLASS_AND) {
3491 if (data->start_class->flags & ANYOF_LOCALE)
3492 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3495 data->start_class->flags |= ANYOF_LOCALE;
3496 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3500 if (flags & SCF_DO_STCLASS_AND) {
3501 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3502 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3503 for (value = 0; value < 256; value++)
3504 if (!isSPACE(value))
3505 ANYOF_BITMAP_CLEAR(data->start_class, value);
3509 if (data->start_class->flags & ANYOF_LOCALE)
3510 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3512 for (value = 0; value < 256; value++)
3514 ANYOF_BITMAP_SET(data->start_class, value);
3519 if (flags & SCF_DO_STCLASS_AND) {
3520 if (data->start_class->flags & ANYOF_LOCALE)
3521 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3524 data->start_class->flags |= ANYOF_LOCALE;
3525 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3529 if (flags & SCF_DO_STCLASS_AND) {
3530 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3531 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3532 for (value = 0; value < 256; value++)
3534 ANYOF_BITMAP_CLEAR(data->start_class, value);
3538 if (data->start_class->flags & ANYOF_LOCALE)
3539 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3541 for (value = 0; value < 256; value++)
3542 if (!isSPACE(value))
3543 ANYOF_BITMAP_SET(data->start_class, value);
3548 if (flags & SCF_DO_STCLASS_AND) {
3549 if (data->start_class->flags & ANYOF_LOCALE) {
3550 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3551 for (value = 0; value < 256; value++)
3552 if (!isSPACE(value))
3553 ANYOF_BITMAP_CLEAR(data->start_class, value);
3557 data->start_class->flags |= ANYOF_LOCALE;
3558 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3562 if (flags & SCF_DO_STCLASS_AND) {
3563 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3564 for (value = 0; value < 256; value++)
3565 if (!isDIGIT(value))
3566 ANYOF_BITMAP_CLEAR(data->start_class, value);
3569 if (data->start_class->flags & ANYOF_LOCALE)
3570 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3572 for (value = 0; value < 256; value++)
3574 ANYOF_BITMAP_SET(data->start_class, value);
3579 if (flags & SCF_DO_STCLASS_AND) {
3580 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3581 for (value = 0; value < 256; value++)
3583 ANYOF_BITMAP_CLEAR(data->start_class, value);
3586 if (data->start_class->flags & ANYOF_LOCALE)
3587 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3589 for (value = 0; value < 256; value++)
3590 if (!isDIGIT(value))
3591 ANYOF_BITMAP_SET(data->start_class, value);
3595 CASE_SYNST_FNC(VERTWS);
3596 CASE_SYNST_FNC(HORIZWS);
3599 if (flags & SCF_DO_STCLASS_OR)
3600 cl_and(data->start_class, and_withp);
3601 flags &= ~SCF_DO_STCLASS;
3604 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3605 data->flags |= (OP(scan) == MEOL
3609 else if ( PL_regkind[OP(scan)] == BRANCHJ
3610 /* Lookbehind, or need to calculate parens/evals/stclass: */
3611 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3612 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3613 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3614 || OP(scan) == UNLESSM )
3616 /* Negative Lookahead/lookbehind
3617 In this case we can't do fixed string optimisation.
3620 I32 deltanext, minnext, fake = 0;
3622 struct regnode_charclass_class intrnl;
3625 data_fake.flags = 0;
3627 data_fake.whilem_c = data->whilem_c;
3628 data_fake.last_closep = data->last_closep;
3631 data_fake.last_closep = &fake;
3632 data_fake.pos_delta = delta;
3633 if ( flags & SCF_DO_STCLASS && !scan->flags
3634 && OP(scan) == IFMATCH ) { /* Lookahead */
3635 cl_init(pRExC_state, &intrnl);
3636 data_fake.start_class = &intrnl;
3637 f |= SCF_DO_STCLASS_AND;
3639 if (flags & SCF_WHILEM_VISITED_POS)
3640 f |= SCF_WHILEM_VISITED_POS;
3641 next = regnext(scan);
3642 nscan = NEXTOPER(NEXTOPER(scan));
3643 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3644 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3647 FAIL("Variable length lookbehind not implemented");
3649 else if (minnext > (I32)U8_MAX) {
3650 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3652 scan->flags = (U8)minnext;
3655 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3657 if (data_fake.flags & SF_HAS_EVAL)
3658 data->flags |= SF_HAS_EVAL;
3659 data->whilem_c = data_fake.whilem_c;
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 PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3671 /* Positive Lookahead/lookbehind
3672 In this case we can do fixed string optimisation,
3673 but we must be careful about it. Note in the case of
3674 lookbehind the positions will be offset by the minimum
3675 length of the pattern, something we won't know about
3676 until after the recurse.
3678 I32 deltanext, fake = 0;
3680 struct regnode_charclass_class intrnl;
3682 /* We use SAVEFREEPV so that when the full compile
3683 is finished perl will clean up the allocated
3684 minlens when its all done. This was we don't
3685 have to worry about freeing them when we know
3686 they wont be used, which would be a pain.
3689 Newx( minnextp, 1, I32 );
3690 SAVEFREEPV(minnextp);
3693 StructCopy(data, &data_fake, scan_data_t);
3694 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3697 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3698 data_fake.last_found=newSVsv(data->last_found);
3702 data_fake.last_closep = &fake;
3703 data_fake.flags = 0;
3704 data_fake.pos_delta = delta;
3706 data_fake.flags |= SF_IS_INF;
3707 if ( flags & SCF_DO_STCLASS && !scan->flags
3708 && OP(scan) == IFMATCH ) { /* Lookahead */
3709 cl_init(pRExC_state, &intrnl);
3710 data_fake.start_class = &intrnl;
3711 f |= SCF_DO_STCLASS_AND;
3713 if (flags & SCF_WHILEM_VISITED_POS)
3714 f |= SCF_WHILEM_VISITED_POS;
3715 next = regnext(scan);
3716 nscan = NEXTOPER(NEXTOPER(scan));
3718 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3719 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3722 FAIL("Variable length lookbehind not implemented");
3724 else if (*minnextp > (I32)U8_MAX) {
3725 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3727 scan->flags = (U8)*minnextp;
3732 if (f & SCF_DO_STCLASS_AND) {
3733 const int was = (data->start_class->flags & ANYOF_EOS);
3735 cl_and(data->start_class, &intrnl);
3737 data->start_class->flags |= ANYOF_EOS;
3740 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3742 if (data_fake.flags & SF_HAS_EVAL)
3743 data->flags |= SF_HAS_EVAL;
3744 data->whilem_c = data_fake.whilem_c;
3745 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3746 if (RExC_rx->minlen<*minnextp)
3747 RExC_rx->minlen=*minnextp;
3748 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3749 SvREFCNT_dec(data_fake.last_found);
3751 if ( data_fake.minlen_fixed != minlenp )
3753 data->offset_fixed= data_fake.offset_fixed;
3754 data->minlen_fixed= data_fake.minlen_fixed;
3755 data->lookbehind_fixed+= scan->flags;
3757 if ( data_fake.minlen_float != minlenp )
3759 data->minlen_float= data_fake.minlen_float;
3760 data->offset_float_min=data_fake.offset_float_min;
3761 data->offset_float_max=data_fake.offset_float_max;
3762 data->lookbehind_float+= scan->flags;
3771 else if (OP(scan) == OPEN) {
3772 if (stopparen != (I32)ARG(scan))
3775 else if (OP(scan) == CLOSE) {
3776 if (stopparen == (I32)ARG(scan)) {
3779 if ((I32)ARG(scan) == is_par) {
3780 next = regnext(scan);
3782 if ( next && (OP(next) != WHILEM) && next < last)
3783 is_par = 0; /* Disable optimization */
3786 *(data->last_closep) = ARG(scan);
3788 else if (OP(scan) == EVAL) {
3790 data->flags |= SF_HAS_EVAL;
3792 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3793 if (flags & SCF_DO_SUBSTR) {
3794 SCAN_COMMIT(pRExC_state,data,minlenp);
3795 flags &= ~SCF_DO_SUBSTR;
3797 if (data && OP(scan)==ACCEPT) {
3798 data->flags |= SCF_SEEN_ACCEPT;
3803 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3805 if (flags & SCF_DO_SUBSTR) {
3806 SCAN_COMMIT(pRExC_state,data,minlenp);
3807 data->longest = &(data->longest_float);
3809 is_inf = is_inf_internal = 1;
3810 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3811 cl_anything(pRExC_state, data->start_class);
3812 flags &= ~SCF_DO_STCLASS;
3814 else if (OP(scan) == GPOS) {
3815 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3816 !(delta || is_inf || (data && data->pos_delta)))
3818 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3819 RExC_rx->extflags |= RXf_ANCH_GPOS;
3820 if (RExC_rx->gofs < (U32)min)
3821 RExC_rx->gofs = min;
3823 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3827 #ifdef TRIE_STUDY_OPT
3828 #ifdef FULL_TRIE_STUDY
3829 else if (PL_regkind[OP(scan)] == TRIE) {
3830 /* NOTE - There is similar code to this block above for handling
3831 BRANCH nodes on the initial study. If you change stuff here
3833 regnode *trie_node= scan;
3834 regnode *tail= regnext(scan);
3835 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3836 I32 max1 = 0, min1 = I32_MAX;
3837 struct regnode_charclass_class accum;
3839 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3840 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3841 if (flags & SCF_DO_STCLASS)
3842 cl_init_zero(pRExC_state, &accum);
3848 const regnode *nextbranch= NULL;
3851 for ( word=1 ; word <= trie->wordcount ; word++)
3853 I32 deltanext=0, minnext=0, f = 0, fake;
3854 struct regnode_charclass_class this_class;
3856 data_fake.flags = 0;
3858 data_fake.whilem_c = data->whilem_c;
3859 data_fake.last_closep = data->last_closep;
3862 data_fake.last_closep = &fake;
3863 data_fake.pos_delta = delta;
3864 if (flags & SCF_DO_STCLASS) {
3865 cl_init(pRExC_state, &this_class);
3866 data_fake.start_class = &this_class;
3867 f = SCF_DO_STCLASS_AND;
3869 if (flags & SCF_WHILEM_VISITED_POS)
3870 f |= SCF_WHILEM_VISITED_POS;
3872 if (trie->jump[word]) {
3874 nextbranch = trie_node + trie->jump[0];
3875 scan= trie_node + trie->jump[word];
3876 /* We go from the jump point to the branch that follows
3877 it. Note this means we need the vestigal unused branches
3878 even though they arent otherwise used.
3880 minnext = study_chunk(pRExC_state, &scan, minlenp,
3881 &deltanext, (regnode *)nextbranch, &data_fake,
3882 stopparen, recursed, NULL, f,depth+1);
3884 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3885 nextbranch= regnext((regnode*)nextbranch);
3887 if (min1 > (I32)(minnext + trie->minlen))
3888 min1 = minnext + trie->minlen;
3889 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3890 max1 = minnext + deltanext + trie->maxlen;
3891 if (deltanext == I32_MAX)
3892 is_inf = is_inf_internal = 1;
3894 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3896 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3897 if ( stopmin > min + min1)
3898 stopmin = min + min1;
3899 flags &= ~SCF_DO_SUBSTR;
3901 data->flags |= SCF_SEEN_ACCEPT;
3904 if (data_fake.flags & SF_HAS_EVAL)
3905 data->flags |= SF_HAS_EVAL;
3906 data->whilem_c = data_fake.whilem_c;
3908 if (flags & SCF_DO_STCLASS)
3909 cl_or(pRExC_state, &accum, &this_class);
3912 if (flags & SCF_DO_SUBSTR) {
3913 data->pos_min += min1;
3914 data->pos_delta += max1 - min1;
3915 if (max1 != min1 || is_inf)
3916 data->longest = &(data->longest_float);
3919 delta += max1 - min1;
3920 if (flags & SCF_DO_STCLASS_OR) {
3921 cl_or(pRExC_state, data->start_class, &accum);
3923 cl_and(data->start_class, and_withp);
3924 flags &= ~SCF_DO_STCLASS;
3927 else if (flags & SCF_DO_STCLASS_AND) {
3929 cl_and(data->start_class, &accum);
3930 flags &= ~SCF_DO_STCLASS;
3933 /* Switch to OR mode: cache the old value of
3934 * data->start_class */
3936 StructCopy(data->start_class, and_withp,
3937 struct regnode_charclass_class);
3938 flags &= ~SCF_DO_STCLASS_AND;
3939 StructCopy(&accum, data->start_class,
3940 struct regnode_charclass_class);
3941 flags |= SCF_DO_STCLASS_OR;
3942 data->start_class->flags |= ANYOF_EOS;
3949 else if (PL_regkind[OP(scan)] == TRIE) {
3950 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3953 min += trie->minlen;
3954 delta += (trie->maxlen - trie->minlen);
3955 flags &= ~SCF_DO_STCLASS; /* xxx */
3956 if (flags & SCF_DO_SUBSTR) {
3957 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3958 data->pos_min += trie->minlen;
3959 data->pos_delta += (trie->maxlen - trie->minlen);
3960 if (trie->maxlen != trie->minlen)
3961 data->longest = &(data->longest_float);
3963 if (trie->jump) /* no more substrings -- for now /grr*/
3964 flags &= ~SCF_DO_SUBSTR;
3966 #endif /* old or new */
3967 #endif /* TRIE_STUDY_OPT */
3969 /* Else: zero-length, ignore. */
3970 scan = regnext(scan);
3975 stopparen = frame->stop;
3976 frame = frame->prev;
3977 goto fake_study_recurse;
3982 DEBUG_STUDYDATA("pre-fin:",data,depth);
3985 *deltap = is_inf_internal ? I32_MAX : delta;
3986 if (flags & SCF_DO_SUBSTR && is_inf)
3987 data->pos_delta = I32_MAX - data->pos_min;
3988 if (is_par > (I32)U8_MAX)
3990 if (is_par && pars==1 && data) {
3991 data->flags |= SF_IN_PAR;
3992 data->flags &= ~SF_HAS_PAR;
3994 else if (pars && data) {
3995 data->flags |= SF_HAS_PAR;
3996 data->flags &= ~SF_IN_PAR;
3998 if (flags & SCF_DO_STCLASS_OR)
3999 cl_and(data->start_class, and_withp);
4000 if (flags & SCF_TRIE_RESTUDY)
4001 data->flags |= SCF_TRIE_RESTUDY;
4003 DEBUG_STUDYDATA("post-fin:",data,depth);
4005 return min < stopmin ? min : stopmin;
4009 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4011 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4013 Renewc(RExC_rxi->data,
4014 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4015 char, struct reg_data);
4017 Renew(RExC_rxi->data->what, count + n, U8);
4019 Newx(RExC_rxi->data->what, n, U8);
4020 RExC_rxi->data->count = count + n;
4021 Copy(s, RExC_rxi->data->what + count, n, U8);
4025 /*XXX: todo make this not included in a non debugging perl */
4026 #ifndef PERL_IN_XSUB_RE
4028 Perl_reginitcolors(pTHX)
4031 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4033 char *t = savepv(s);
4037 t = strchr(t, '\t');
4043 PL_colors[i] = t = (char *)"";
4048 PL_colors[i++] = (char *)"";
4055 #ifdef TRIE_STUDY_OPT
4056 #define CHECK_RESTUDY_GOTO \
4058 (data.flags & SCF_TRIE_RESTUDY) \
4062 #define CHECK_RESTUDY_GOTO
4066 - pregcomp - compile a regular expression into internal code
4068 * We can't allocate space until we know how big the compiled form will be,
4069 * but we can't compile it (and thus know how big it is) until we've got a
4070 * place to put the code. So we cheat: we compile it twice, once with code
4071 * generation turned off and size counting turned on, and once "for real".
4072 * This also means that we don't allocate space until we are sure that the
4073 * thing really will compile successfully, and we never have to move the
4074 * code and thus invalidate pointers into it. (Note that it has to be in
4075 * one piece because free() must be able to free it all.) [NB: not true in perl]
4077 * Beware that the optimization-preparation code in here knows about some
4078 * of the structure of the compiled regexp. [I'll say.]
4083 #ifndef PERL_IN_XSUB_RE
4084 #define RE_ENGINE_PTR &PL_core_reg_engine
4086 extern const struct regexp_engine my_reg_engine;
4087 #define RE_ENGINE_PTR &my_reg_engine
4090 #ifndef PERL_IN_XSUB_RE
4092 Perl_pregcomp(pTHX_ const SV * const pattern, const U32 flags)
4095 HV * const table = GvHV(PL_hintgv);
4096 /* Dispatch a request to compile a regexp to correct
4099 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4100 GET_RE_DEBUG_FLAGS_DECL;
4101 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4102 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4104 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4107 return CALLREGCOMP_ENG(eng, pattern, flags);
4110 return Perl_re_compile(aTHX_ pattern, flags);
4115 Perl_re_compile(pTHX_ const SV * const pattern, const U32 pm_flags)
4119 register regexp_internal *ri;
4121 char* exp = SvPV((SV*)pattern, plen);
4122 char* xend = exp + plen;
4130 RExC_state_t RExC_state;
4131 RExC_state_t * const pRExC_state = &RExC_state;
4132 #ifdef TRIE_STUDY_OPT
4134 RExC_state_t copyRExC_state;
4136 GET_RE_DEBUG_FLAGS_DECL;
4137 DEBUG_r(if (!PL_colorset) reginitcolors());
4139 RExC_utf8 = RExC_orig_utf8 = pm_flags & RXf_UTF8;
4142 SV *dsv= sv_newmortal();
4143 RE_PV_QUOTED_DECL(s, RExC_utf8,
4144 dsv, exp, plen, 60);
4145 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4146 PL_colors[4],PL_colors[5],s);
4151 RExC_flags = pm_flags;
4155 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4156 RExC_seen_evals = 0;
4159 /* First pass: determine size, legality. */
4167 RExC_emit = &PL_regdummy;
4168 RExC_whilem_seen = 0;
4169 RExC_charnames = NULL;
4170 RExC_open_parens = NULL;
4171 RExC_close_parens = NULL;
4173 RExC_paren_names = NULL;
4175 RExC_paren_name_list = NULL;
4177 RExC_recurse = NULL;
4178 RExC_recurse_count = 0;
4180 #if 0 /* REGC() is (currently) a NOP at the first pass.
4181 * Clever compilers notice this and complain. --jhi */
4182 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4184 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4185 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4186 RExC_precomp = NULL;
4189 if (RExC_utf8 && !RExC_orig_utf8) {
4190 /* It's possible to write a regexp in ascii that represents unicode
4191 codepoints outside of the byte range, such as via \x{100}. If we
4192 detect such a sequence we have to convert the entire pattern to utf8
4193 and then recompile, as our sizing calculation will have been based
4194 on 1 byte == 1 character, but we will need to use utf8 to encode
4195 at least some part of the pattern, and therefore must convert the whole
4197 XXX: somehow figure out how to make this less expensive...
4200 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4201 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4202 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4204 RExC_orig_utf8 = RExC_utf8;
4206 goto redo_first_pass;
4209 PerlIO_printf(Perl_debug_log,
4210 "Required size %"IVdf" nodes\n"
4211 "Starting second pass (creation)\n",
4214 RExC_lastparse=NULL;
4216 /* Small enough for pointer-storage convention?
4217 If extralen==0, this means that we will not need long jumps. */
4218 if (RExC_size >= 0x10000L && RExC_extralen)
4219 RExC_size += RExC_extralen;
4222 if (RExC_whilem_seen > 15)
4223 RExC_whilem_seen = 15;
4225 /* Allocate space and zero-initialize. Note, the two step process
4226 of zeroing when in debug mode, thus anything assigned has to
4227 happen after that */
4228 Newxz(r, 1, regexp);
4229 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4230 char, regexp_internal);
4231 if ( r == NULL || ri == NULL )
4232 FAIL("Regexp out of space");
4234 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4235 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4237 /* bulk initialize base fields with 0. */
4238 Zero(ri, sizeof(regexp_internal), char);
4241 /* non-zero initialization begins here */
4243 r->engine= RE_ENGINE_PTR;
4246 r->extflags = pm_flags;
4248 bool has_k = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4249 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4250 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4251 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) >> 12);
4252 const char *fptr = STD_PAT_MODS; /*"msix"*/
4254 r->wraplen = r->prelen + has_minus + has_k + has_runon
4255 + (sizeof(STD_PAT_MODS) - 1)
4256 + (sizeof("(?:)") - 1);
4258 Newx(r->wrapped, r->wraplen + 1, char );
4262 *p++ = KEEPCOPY_PAT_MOD; /*'k'*/
4264 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4265 char *colon = r + 1;
4268 while((ch = *fptr++)) {
4282 Copy(RExC_precomp, p, r->prelen, char);
4292 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4294 if (RExC_seen & REG_SEEN_RECURSE) {
4295 Newxz(RExC_open_parens, RExC_npar,regnode *);
4296 SAVEFREEPV(RExC_open_parens);
4297 Newxz(RExC_close_parens,RExC_npar,regnode *);
4298 SAVEFREEPV(RExC_close_parens);
4301 /* Useful during FAIL. */
4302 #ifdef RE_TRACK_PATTERN_OFFSETS
4303 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4304 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4305 "%s %"UVuf" bytes for offset annotations.\n",
4306 ri->u.offsets ? "Got" : "Couldn't get",
4307 (UV)((2*RExC_size+1) * sizeof(U32))));
4309 SetProgLen(ri,RExC_size);
4313 /* Second pass: emit code. */
4314 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4319 RExC_emit_start = ri->program;
4320 RExC_emit = ri->program;
4321 RExC_emit_bound = ri->program + RExC_size + 1;
4323 /* Store the count of eval-groups for security checks: */
4324 RExC_rx->seen_evals = RExC_seen_evals;
4325 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4326 if (reg(pRExC_state, 0, &flags,1) == NULL)
4329 /* XXXX To minimize changes to RE engine we always allocate
4330 3-units-long substrs field. */
4331 Newx(r->substrs, 1, struct reg_substr_data);
4332 if (RExC_recurse_count) {
4333 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4334 SAVEFREEPV(RExC_recurse);
4338 r->minlen = minlen = sawplus = sawopen = 0;
4339 Zero(r->substrs, 1, struct reg_substr_data);
4341 #ifdef TRIE_STUDY_OPT
4344 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4346 RExC_state = copyRExC_state;
4347 if (seen & REG_TOP_LEVEL_BRANCHES)
4348 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4350 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4351 if (data.last_found) {
4352 SvREFCNT_dec(data.longest_fixed);
4353 SvREFCNT_dec(data.longest_float);
4354 SvREFCNT_dec(data.last_found);
4356 StructCopy(&zero_scan_data, &data, scan_data_t);
4358 StructCopy(&zero_scan_data, &data, scan_data_t);
4359 copyRExC_state = RExC_state;
4362 StructCopy(&zero_scan_data, &data, scan_data_t);
4365 /* Dig out information for optimizations. */
4366 r->extflags = pm_flags; /* Again? */
4367 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4370 r->extflags |= RXf_UTF8; /* Unicode in it? */
4371 ri->regstclass = NULL;
4372 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4373 r->intflags |= PREGf_NAUGHTY;
4374 scan = ri->program + 1; /* First BRANCH. */
4376 /* testing for BRANCH here tells us whether there is "must appear"
4377 data in the pattern. If there is then we can use it for optimisations */
4378 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4380 STRLEN longest_float_length, longest_fixed_length;
4381 struct regnode_charclass_class ch_class; /* pointed to by data */
4383 I32 last_close = 0; /* pointed to by data */
4386 /* Skip introductions and multiplicators >= 1. */
4387 while ((OP(first) == OPEN && (sawopen = 1)) ||
4388 /* An OR of *one* alternative - should not happen now. */
4389 (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) ||
4390 /* for now we can't handle lookbehind IFMATCH*/
4391 (OP(first) == IFMATCH && !first->flags) ||
4392 (OP(first) == PLUS) ||
4393 (OP(first) == MINMOD) ||
4394 /* An {n,m} with n>0 */
4395 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
4398 if (OP(first) == PLUS)
4401 first += regarglen[OP(first)];
4402 if (OP(first) == IFMATCH) {
4403 first = NEXTOPER(first);
4404 first += EXTRA_STEP_2ARGS;
4405 } else /* XXX possible optimisation for /(?=)/ */
4406 first = NEXTOPER(first);
4409 /* Starting-point info. */
4411 DEBUG_PEEP("first:",first,0);
4412 /* Ignore EXACT as we deal with it later. */
4413 if (PL_regkind[OP(first)] == EXACT) {
4414 if (OP(first) == EXACT)
4415 NOOP; /* Empty, get anchored substr later. */
4416 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4417 ri->regstclass = first;
4420 else if (PL_regkind[OP(first)] == TRIE &&
4421 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4424 /* this can happen only on restudy */
4425 if ( OP(first) == TRIE ) {
4426 struct regnode_1 *trieop = (struct regnode_1 *)
4427 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4428 StructCopy(first,trieop,struct regnode_1);
4429 trie_op=(regnode *)trieop;
4431 struct regnode_charclass *trieop = (struct regnode_charclass *)
4432 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4433 StructCopy(first,trieop,struct regnode_charclass);
4434 trie_op=(regnode *)trieop;
4437 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4438 ri->regstclass = trie_op;
4441 else if (strchr((const char*)PL_simple,OP(first)))
4442 ri->regstclass = first;
4443 else if (PL_regkind[OP(first)] == BOUND ||
4444 PL_regkind[OP(first)] == NBOUND)
4445 ri->regstclass = first;
4446 else if (PL_regkind[OP(first)] == BOL) {
4447 r->extflags |= (OP(first) == MBOL
4449 : (OP(first) == SBOL
4452 first = NEXTOPER(first);
4455 else if (OP(first) == GPOS) {
4456 r->extflags |= RXf_ANCH_GPOS;
4457 first = NEXTOPER(first);
4460 else if ((!sawopen || !RExC_sawback) &&
4461 (OP(first) == STAR &&
4462 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4463 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4465 /* turn .* into ^.* with an implied $*=1 */
4467 (OP(NEXTOPER(first)) == REG_ANY)
4470 r->extflags |= type;
4471 r->intflags |= PREGf_IMPLICIT;
4472 first = NEXTOPER(first);
4475 if (sawplus && (!sawopen || !RExC_sawback)
4476 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4477 /* x+ must match at the 1st pos of run of x's */
4478 r->intflags |= PREGf_SKIP;
4480 /* Scan is after the zeroth branch, first is atomic matcher. */
4481 #ifdef TRIE_STUDY_OPT
4484 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4485 (IV)(first - scan + 1))
4489 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4490 (IV)(first - scan + 1))
4496 * If there's something expensive in the r.e., find the
4497 * longest literal string that must appear and make it the
4498 * regmust. Resolve ties in favor of later strings, since
4499 * the regstart check works with the beginning of the r.e.
4500 * and avoiding duplication strengthens checking. Not a
4501 * strong reason, but sufficient in the absence of others.
4502 * [Now we resolve ties in favor of the earlier string if
4503 * it happens that c_offset_min has been invalidated, since the
4504 * earlier string may buy us something the later one won't.]
4507 data.longest_fixed = newSVpvs("");
4508 data.longest_float = newSVpvs("");
4509 data.last_found = newSVpvs("");
4510 data.longest = &(data.longest_fixed);
4512 if (!ri->regstclass) {
4513 cl_init(pRExC_state, &ch_class);
4514 data.start_class = &ch_class;
4515 stclass_flag = SCF_DO_STCLASS_AND;
4516 } else /* XXXX Check for BOUND? */
4518 data.last_closep = &last_close;
4520 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4521 &data, -1, NULL, NULL,
4522 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4528 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4529 && data.last_start_min == 0 && data.last_end > 0
4530 && !RExC_seen_zerolen
4531 && !(RExC_seen & REG_SEEN_VERBARG)
4532 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4533 r->extflags |= RXf_CHECK_ALL;
4534 scan_commit(pRExC_state, &data,&minlen,0);
4535 SvREFCNT_dec(data.last_found);
4537 /* Note that code very similar to this but for anchored string
4538 follows immediately below, changes may need to be made to both.
4541 longest_float_length = CHR_SVLEN(data.longest_float);
4542 if (longest_float_length
4543 || (data.flags & SF_FL_BEFORE_EOL
4544 && (!(data.flags & SF_FL_BEFORE_MEOL)
4545 || (RExC_flags & RXf_PMf_MULTILINE))))
4549 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4550 && data.offset_fixed == data.offset_float_min
4551 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4552 goto remove_float; /* As in (a)+. */
4554 /* copy the information about the longest float from the reg_scan_data
4555 over to the program. */
4556 if (SvUTF8(data.longest_float)) {
4557 r->float_utf8 = data.longest_float;
4558 r->float_substr = NULL;
4560 r->float_substr = data.longest_float;
4561 r->float_utf8 = NULL;
4563 /* float_end_shift is how many chars that must be matched that
4564 follow this item. We calculate it ahead of time as once the
4565 lookbehind offset is added in we lose the ability to correctly
4567 ml = data.minlen_float ? *(data.minlen_float)
4568 : (I32)longest_float_length;
4569 r->float_end_shift = ml - data.offset_float_min
4570 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4571 + data.lookbehind_float;
4572 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4573 r->float_max_offset = data.offset_float_max;
4574 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4575 r->float_max_offset -= data.lookbehind_float;
4577 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4578 && (!(data.flags & SF_FL_BEFORE_MEOL)
4579 || (RExC_flags & RXf_PMf_MULTILINE)));
4580 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4584 r->float_substr = r->float_utf8 = NULL;
4585 SvREFCNT_dec(data.longest_float);
4586 longest_float_length = 0;
4589 /* Note that code very similar to this but for floating string
4590 is immediately above, changes may need to be made to both.
4593 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4594 if (longest_fixed_length
4595 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4596 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4597 || (RExC_flags & RXf_PMf_MULTILINE))))
4601 /* copy the information about the longest fixed
4602 from the reg_scan_data over to the program. */
4603 if (SvUTF8(data.longest_fixed)) {
4604 r->anchored_utf8 = data.longest_fixed;
4605 r->anchored_substr = NULL;
4607 r->anchored_substr = data.longest_fixed;
4608 r->anchored_utf8 = NULL;
4610 /* fixed_end_shift is how many chars that must be matched that
4611 follow this item. We calculate it ahead of time as once the
4612 lookbehind offset is added in we lose the ability to correctly
4614 ml = data.minlen_fixed ? *(data.minlen_fixed)
4615 : (I32)longest_fixed_length;
4616 r->anchored_end_shift = ml - data.offset_fixed
4617 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4618 + data.lookbehind_fixed;
4619 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4621 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4622 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4623 || (RExC_flags & RXf_PMf_MULTILINE)));
4624 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4627 r->anchored_substr = r->anchored_utf8 = NULL;
4628 SvREFCNT_dec(data.longest_fixed);
4629 longest_fixed_length = 0;
4632 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4633 ri->regstclass = NULL;
4634 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4636 && !(data.start_class->flags & ANYOF_EOS)
4637 && !cl_is_anything(data.start_class))
4639 const U32 n = add_data(pRExC_state, 1, "f");
4641 Newx(RExC_rxi->data->data[n], 1,
4642 struct regnode_charclass_class);
4643 StructCopy(data.start_class,
4644 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4645 struct regnode_charclass_class);
4646 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4647 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4648 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4649 regprop(r, sv, (regnode*)data.start_class);
4650 PerlIO_printf(Perl_debug_log,
4651 "synthetic stclass \"%s\".\n",
4652 SvPVX_const(sv));});
4655 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4656 if (longest_fixed_length > longest_float_length) {
4657 r->check_end_shift = r->anchored_end_shift;
4658 r->check_substr = r->anchored_substr;
4659 r->check_utf8 = r->anchored_utf8;
4660 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4661 if (r->extflags & RXf_ANCH_SINGLE)
4662 r->extflags |= RXf_NOSCAN;
4665 r->check_end_shift = r->float_end_shift;
4666 r->check_substr = r->float_substr;
4667 r->check_utf8 = r->float_utf8;
4668 r->check_offset_min = r->float_min_offset;
4669 r->check_offset_max = r->float_max_offset;
4671 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4672 This should be changed ASAP! */
4673 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4674 r->extflags |= RXf_USE_INTUIT;
4675 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4676 r->extflags |= RXf_INTUIT_TAIL;
4678 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4679 if ( (STRLEN)minlen < longest_float_length )
4680 minlen= longest_float_length;
4681 if ( (STRLEN)minlen < longest_fixed_length )
4682 minlen= longest_fixed_length;
4686 /* Several toplevels. Best we can is to set minlen. */
4688 struct regnode_charclass_class ch_class;
4691 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4693 scan = ri->program + 1;
4694 cl_init(pRExC_state, &ch_class);
4695 data.start_class = &ch_class;
4696 data.last_closep = &last_close;
4699 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4700 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4704 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4705 = r->float_substr = r->float_utf8 = NULL;
4706 if (!(data.start_class->flags & ANYOF_EOS)
4707 && !cl_is_anything(data.start_class))
4709 const U32 n = add_data(pRExC_state, 1, "f");
4711 Newx(RExC_rxi->data->data[n], 1,
4712 struct regnode_charclass_class);
4713 StructCopy(data.start_class,
4714 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4715 struct regnode_charclass_class);
4716 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4717 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4718 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4719 regprop(r, sv, (regnode*)data.start_class);
4720 PerlIO_printf(Perl_debug_log,
4721 "synthetic stclass \"%s\".\n",
4722 SvPVX_const(sv));});
4726 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4727 the "real" pattern. */
4729 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4730 (IV)minlen, (IV)r->minlen);
4732 r->minlenret = minlen;
4733 if (r->minlen < minlen)
4736 if (RExC_seen & REG_SEEN_GPOS)
4737 r->extflags |= RXf_GPOS_SEEN;
4738 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4739 r->extflags |= RXf_LOOKBEHIND_SEEN;
4740 if (RExC_seen & REG_SEEN_EVAL)
4741 r->extflags |= RXf_EVAL_SEEN;
4742 if (RExC_seen & REG_SEEN_CANY)
4743 r->extflags |= RXf_CANY_SEEN;
4744 if (RExC_seen & REG_SEEN_VERBARG)
4745 r->intflags |= PREGf_VERBARG_SEEN;
4746 if (RExC_seen & REG_SEEN_CUTGROUP)
4747 r->intflags |= PREGf_CUTGROUP_SEEN;
4748 if (RExC_paren_names)
4749 r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
4751 r->paren_names = NULL;
4752 if (r->prelen == 3 && strnEQ("\\s+", r->precomp, 3)) /* precomp = "\\s+)" */
4753 r->extflags |= RXf_WHITE;
4754 else if (r->prelen == 1 && r->precomp[0] == '^')
4755 r->extflags |= RXf_START_ONLY;
4758 if (RExC_paren_names) {
4759 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4760 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4763 ri->name_list_idx = 0;
4765 if (RExC_recurse_count) {
4766 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4767 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4768 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4771 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4772 /* assume we don't need to swap parens around before we match */
4775 PerlIO_printf(Perl_debug_log,"Final program:\n");
4778 #ifdef RE_TRACK_PATTERN_OFFSETS
4779 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4780 const U32 len = ri->u.offsets[0];
4782 GET_RE_DEBUG_FLAGS_DECL;
4783 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4784 for (i = 1; i <= len; i++) {
4785 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4786 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4787 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4789 PerlIO_printf(Perl_debug_log, "\n");
4795 #undef RE_ENGINE_PTR
4799 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const rx, SV * const namesv, const U32 flags)
4801 AV *retarray = NULL;
4806 if (rx && rx->paren_names) {
4807 HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
4810 SV* sv_dat=HeVAL(he_str);
4811 I32 *nums=(I32*)SvPVX(sv_dat);
4812 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4813 if ((I32)(rx->nparens) >= nums[i]
4814 && rx->offs[nums[i]].start != -1
4815 && rx->offs[nums[i]].end != -1)
4818 CALLREG_NUMBUF_FETCH(rx,nums[i],ret);
4822 ret = newSVsv(&PL_sv_undef);
4825 SvREFCNT_inc_simple_void(ret);
4826 av_push(retarray, ret);
4830 return (SV*)retarray;
4837 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const rx, const I32 paren, SV * const sv)
4844 sv_setsv(sv,&PL_sv_undef);
4848 if (paren == -2 && rx->offs[0].start != -1) {
4850 i = rx->offs[0].start;
4854 if (paren == -1 && rx->offs[0].end != -1) {
4856 s = rx->subbeg + rx->offs[0].end;
4857 i = rx->sublen - rx->offs[0].end;
4860 if ( 0 <= paren && paren <= (I32)rx->nparens &&
4861 (s1 = rx->offs[paren].start) != -1 &&
4862 (t1 = rx->offs[paren].end) != -1)
4866 s = rx->subbeg + s1;
4868 sv_setsv(sv,&PL_sv_undef);
4871 assert(rx->sublen >= (s - rx->subbeg) + i );
4873 const int oldtainted = PL_tainted;
4875 sv_setpvn(sv, s, i);
4876 PL_tainted = oldtainted;
4877 if ( (rx->extflags & RXf_CANY_SEEN)
4878 ? (RX_MATCH_UTF8(rx)
4879 && (!i || is_utf8_string((U8*)s, i)))
4880 : (RX_MATCH_UTF8(rx)) )
4887 if (RX_MATCH_TAINTED(rx)) {
4888 if (SvTYPE(sv) >= SVt_PVMG) {
4889 MAGIC* const mg = SvMAGIC(sv);
4892 SvMAGIC_set(sv, mg->mg_moremagic);
4894 if ((mgt = SvMAGIC(sv))) {
4895 mg->mg_moremagic = mgt;
4896 SvMAGIC_set(sv, mg);
4906 sv_setsv(sv,&PL_sv_undef);
4912 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
4913 SV const * const value)
4915 PERL_UNUSED_ARG(rx);
4916 PERL_UNUSED_ARG(paren);
4917 PERL_UNUSED_ARG(value);
4920 Perl_croak(aTHX_ PL_no_modify);
4924 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const rx, const SV * const sv,
4930 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
4933 if (rx->offs[0].start != -1) {
4934 i = rx->offs[0].start;
4943 if (rx->offs[0].end != -1) {
4944 i = rx->sublen - rx->offs[0].end;
4946 s1 = rx->offs[0].end;
4952 default: /* $&, $1, $2, ... */
4953 if (paren <= (I32)rx->nparens &&
4954 (s1 = rx->offs[paren].start) != -1 &&
4955 (t1 = rx->offs[paren].end) != -1)
4960 if (ckWARN(WARN_UNINITIALIZED))
4961 report_uninit((SV*)sv);
4966 if (i > 0 && RX_MATCH_UTF8(rx)) {
4967 const char * const s = rx->subbeg + s1;
4972 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
4979 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
4981 PERL_UNUSED_ARG(rx);
4982 return newSVpvs("Regexp");
4985 /* Scans the name of a named buffer from the pattern.
4986 * If flags is REG_RSN_RETURN_NULL returns null.
4987 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
4988 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
4989 * to the parsed name as looked up in the RExC_paren_names hash.
4990 * If there is an error throws a vFAIL().. type exception.
4993 #define REG_RSN_RETURN_NULL 0
4994 #define REG_RSN_RETURN_NAME 1
4995 #define REG_RSN_RETURN_DATA 2
4998 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
4999 char *name_start = RExC_parse;
5001 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5002 /* skip IDFIRST by using do...while */
5005 RExC_parse += UTF8SKIP(RExC_parse);
5006 } while (isALNUM_utf8((U8*)RExC_parse));
5010 } while (isALNUM(*RExC_parse));
5014 SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
5015 (int)(RExC_parse - name_start)));
5018 if ( flags == REG_RSN_RETURN_NAME)
5020 else if (flags==REG_RSN_RETURN_DATA) {
5023 if ( ! sv_name ) /* should not happen*/
5024 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5025 if (RExC_paren_names)
5026 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5028 sv_dat = HeVAL(he_str);
5030 vFAIL("Reference to nonexistent named group");
5034 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5041 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5042 int rem=(int)(RExC_end - RExC_parse); \
5051 if (RExC_lastparse!=RExC_parse) \
5052 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5055 iscut ? "..." : "<" \
5058 PerlIO_printf(Perl_debug_log,"%16s",""); \
5061 num = RExC_size + 1; \
5063 num=REG_NODE_NUM(RExC_emit); \
5064 if (RExC_lastnum!=num) \
5065 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5067 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5068 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5069 (int)((depth*2)), "", \
5073 RExC_lastparse=RExC_parse; \
5078 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5079 DEBUG_PARSE_MSG((funcname)); \
5080 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5082 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5083 DEBUG_PARSE_MSG((funcname)); \
5084 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5087 - reg - regular expression, i.e. main body or parenthesized thing
5089 * Caller must absorb opening parenthesis.
5091 * Combining parenthesis handling with the base level of regular expression
5092 * is a trifle forced, but the need to tie the tails of the branches to what
5093 * follows makes it hard to avoid.
5095 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5097 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5099 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5103 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5104 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5107 register regnode *ret; /* Will be the head of the group. */
5108 register regnode *br;
5109 register regnode *lastbr;
5110 register regnode *ender = NULL;
5111 register I32 parno = 0;
5113 const I32 oregflags = RExC_flags;
5114 bool have_branch = 0;
5116 I32 freeze_paren = 0;
5117 I32 after_freeze = 0;
5119 /* for (?g), (?gc), and (?o) warnings; warning
5120 about (?c) will warn about (?g) -- japhy */
5122 #define WASTED_O 0x01
5123 #define WASTED_G 0x02
5124 #define WASTED_C 0x04
5125 #define WASTED_GC (0x02|0x04)
5126 I32 wastedflags = 0x00;
5128 char * parse_start = RExC_parse; /* MJD */
5129 char * const oregcomp_parse = RExC_parse;
5131 GET_RE_DEBUG_FLAGS_DECL;
5132 DEBUG_PARSE("reg ");
5134 *flagp = 0; /* Tentatively. */
5137 /* Make an OPEN node, if parenthesized. */
5139 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5140 char *start_verb = RExC_parse;
5141 STRLEN verb_len = 0;
5142 char *start_arg = NULL;
5143 unsigned char op = 0;
5145 int internal_argval = 0; /* internal_argval is only useful if !argok */
5146 while ( *RExC_parse && *RExC_parse != ')' ) {
5147 if ( *RExC_parse == ':' ) {
5148 start_arg = RExC_parse + 1;
5154 verb_len = RExC_parse - start_verb;
5157 while ( *RExC_parse && *RExC_parse != ')' )
5159 if ( *RExC_parse != ')' )
5160 vFAIL("Unterminated verb pattern argument");
5161 if ( RExC_parse == start_arg )
5164 if ( *RExC_parse != ')' )
5165 vFAIL("Unterminated verb pattern");
5168 switch ( *start_verb ) {
5169 case 'A': /* (*ACCEPT) */
5170 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5172 internal_argval = RExC_nestroot;
5175 case 'C': /* (*COMMIT) */
5176 if ( memEQs(start_verb,verb_len,"COMMIT") )
5179 case 'F': /* (*FAIL) */
5180 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5185 case ':': /* (*:NAME) */
5186 case 'M': /* (*MARK:NAME) */
5187 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5192 case 'P': /* (*PRUNE) */
5193 if ( memEQs(start_verb,verb_len,"PRUNE") )
5196 case 'S': /* (*SKIP) */
5197 if ( memEQs(start_verb,verb_len,"SKIP") )
5200 case 'T': /* (*THEN) */
5201 /* [19:06] <TimToady> :: is then */
5202 if ( memEQs(start_verb,verb_len,"THEN") ) {
5204 RExC_seen |= REG_SEEN_CUTGROUP;
5210 vFAIL3("Unknown verb pattern '%.*s'",
5211 verb_len, start_verb);
5214 if ( start_arg && internal_argval ) {
5215 vFAIL3("Verb pattern '%.*s' may not have an argument",
5216 verb_len, start_verb);
5217 } else if ( argok < 0 && !start_arg ) {
5218 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5219 verb_len, start_verb);
5221 ret = reganode(pRExC_state, op, internal_argval);
5222 if ( ! internal_argval && ! SIZE_ONLY ) {
5224 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5225 ARG(ret) = add_data( pRExC_state, 1, "S" );
5226 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5233 if (!internal_argval)
5234 RExC_seen |= REG_SEEN_VERBARG;
5235 } else if ( start_arg ) {
5236 vFAIL3("Verb pattern '%.*s' may not have an argument",
5237 verb_len, start_verb);
5239 ret = reg_node(pRExC_state, op);
5241 nextchar(pRExC_state);
5244 if (*RExC_parse == '?') { /* (?...) */
5245 bool is_logical = 0;
5246 const char * const seqstart = RExC_parse;
5249 paren = *RExC_parse++;
5250 ret = NULL; /* For look-ahead/behind. */
5253 case 'P': /* (?P...) variants for those used to PCRE/Python */
5254 paren = *RExC_parse++;
5255 if ( paren == '<') /* (?P<...>) named capture */
5257 else if (paren == '>') { /* (?P>name) named recursion */
5258 goto named_recursion;
5260 else if (paren == '=') { /* (?P=...) named backref */
5261 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5262 you change this make sure you change that */
5263 char* name_start = RExC_parse;
5265 SV *sv_dat = reg_scan_name(pRExC_state,
5266 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5267 if (RExC_parse == name_start || *RExC_parse != ')')
5268 vFAIL2("Sequence %.3s... not terminated",parse_start);
5271 num = add_data( pRExC_state, 1, "S" );
5272 RExC_rxi->data->data[num]=(void*)sv_dat;
5273 SvREFCNT_inc_simple_void(sv_dat);
5276 ret = reganode(pRExC_state,
5277 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5281 Set_Node_Offset(ret, parse_start+1);
5282 Set_Node_Cur_Length(ret); /* MJD */
5284 nextchar(pRExC_state);
5288 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5290 case '<': /* (?<...) */
5291 if (*RExC_parse == '!')
5293 else if (*RExC_parse != '=')
5299 case '\'': /* (?'...') */
5300 name_start= RExC_parse;
5301 svname = reg_scan_name(pRExC_state,
5302 SIZE_ONLY ? /* reverse test from the others */
5303 REG_RSN_RETURN_NAME :
5304 REG_RSN_RETURN_NULL);
5305 if (RExC_parse == name_start) {
5307 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5310 if (*RExC_parse != paren)
5311 vFAIL2("Sequence (?%c... not terminated",
5312 paren=='>' ? '<' : paren);
5316 if (!svname) /* shouldnt happen */
5318 "panic: reg_scan_name returned NULL");
5319 if (!RExC_paren_names) {
5320 RExC_paren_names= newHV();
5321 sv_2mortal((SV*)RExC_paren_names);
5323 RExC_paren_name_list= newAV();
5324 sv_2mortal((SV*)RExC_paren_name_list);
5327 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5329 sv_dat = HeVAL(he_str);
5331 /* croak baby croak */
5333 "panic: paren_name hash element allocation failed");
5334 } else if ( SvPOK(sv_dat) ) {
5335 /* (?|...) can mean we have dupes so scan to check
5336 its already been stored. Maybe a flag indicating
5337 we are inside such a construct would be useful,
5338 but the arrays are likely to be quite small, so
5339 for now we punt -- dmq */
5340 IV count = SvIV(sv_dat);
5341 I32 *pv = (I32*)SvPVX(sv_dat);
5343 for ( i = 0 ; i < count ; i++ ) {
5344 if ( pv[i] == RExC_npar ) {
5350 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5351 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5352 pv[count] = RExC_npar;
5356 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5357 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5362 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5363 SvREFCNT_dec(svname);
5366 /*sv_dump(sv_dat);*/
5368 nextchar(pRExC_state);
5370 goto capturing_parens;
5372 RExC_seen |= REG_SEEN_LOOKBEHIND;
5374 case '=': /* (?=...) */
5375 case '!': /* (?!...) */
5376 RExC_seen_zerolen++;
5377 if (*RExC_parse == ')') {
5378 ret=reg_node(pRExC_state, OPFAIL);
5379 nextchar(pRExC_state);
5383 case '|': /* (?|...) */
5384 /* branch reset, behave like a (?:...) except that
5385 buffers in alternations share the same numbers */
5387 after_freeze = freeze_paren = RExC_npar;
5389 case ':': /* (?:...) */
5390 case '>': /* (?>...) */
5392 case '$': /* (?$...) */
5393 case '@': /* (?@...) */
5394 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5396 case '#': /* (?#...) */
5397 while (*RExC_parse && *RExC_parse != ')')
5399 if (*RExC_parse != ')')
5400 FAIL("Sequence (?#... not terminated");
5401 nextchar(pRExC_state);
5404 case '0' : /* (?0) */
5405 case 'R' : /* (?R) */
5406 if (*RExC_parse != ')')
5407 FAIL("Sequence (?R) not terminated");
5408 ret = reg_node(pRExC_state, GOSTART);
5409 *flagp |= POSTPONED;
5410 nextchar(pRExC_state);
5413 { /* named and numeric backreferences */
5415 case '&': /* (?&NAME) */
5416 parse_start = RExC_parse - 1;
5419 SV *sv_dat = reg_scan_name(pRExC_state,
5420 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5421 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5423 goto gen_recurse_regop;
5426 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5428 vFAIL("Illegal pattern");
5430 goto parse_recursion;
5432 case '-': /* (?-1) */
5433 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5434 RExC_parse--; /* rewind to let it be handled later */
5438 case '1': case '2': case '3': case '4': /* (?1) */
5439 case '5': case '6': case '7': case '8': case '9':
5442 num = atoi(RExC_parse);
5443 parse_start = RExC_parse - 1; /* MJD */
5444 if (*RExC_parse == '-')
5446 while (isDIGIT(*RExC_parse))
5448 if (*RExC_parse!=')')
5449 vFAIL("Expecting close bracket");
5452 if ( paren == '-' ) {
5454 Diagram of capture buffer numbering.
5455 Top line is the normal capture buffer numbers
5456 Botton line is the negative indexing as from
5460 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5464 num = RExC_npar + num;
5467 vFAIL("Reference to nonexistent group");
5469 } else if ( paren == '+' ) {
5470 num = RExC_npar + num - 1;
5473 ret = reganode(pRExC_state, GOSUB, num);
5475 if (num > (I32)RExC_rx->nparens) {
5477 vFAIL("Reference to nonexistent group");
5479 ARG2L_SET( ret, RExC_recurse_count++);
5481 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5482 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5486 RExC_seen |= REG_SEEN_RECURSE;
5487 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5488 Set_Node_Offset(ret, parse_start); /* MJD */
5490 *flagp |= POSTPONED;
5491 nextchar(pRExC_state);
5493 } /* named and numeric backreferences */
5496 case '?': /* (??...) */
5498 if (*RExC_parse != '{') {
5500 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5503 *flagp |= POSTPONED;
5504 paren = *RExC_parse++;
5506 case '{': /* (?{...}) */
5511 char *s = RExC_parse;
5513 RExC_seen_zerolen++;
5514 RExC_seen |= REG_SEEN_EVAL;
5515 while (count && (c = *RExC_parse)) {
5526 if (*RExC_parse != ')') {
5528 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5532 OP_4tree *sop, *rop;
5533 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5536 Perl_save_re_context(aTHX);
5537 rop = sv_compile_2op(sv, &sop, "re", &pad);
5538 sop->op_private |= OPpREFCOUNTED;
5539 /* re_dup will OpREFCNT_inc */
5540 OpREFCNT_set(sop, 1);
5543 n = add_data(pRExC_state, 3, "nop");
5544 RExC_rxi->data->data[n] = (void*)rop;
5545 RExC_rxi->data->data[n+1] = (void*)sop;
5546 RExC_rxi->data->data[n+2] = (void*)pad;
5549 else { /* First pass */
5550 if (PL_reginterp_cnt < ++RExC_seen_evals
5552 /* No compiled RE interpolated, has runtime
5553 components ===> unsafe. */
5554 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5555 if (PL_tainting && PL_tainted)
5556 FAIL("Eval-group in insecure regular expression");
5557 #if PERL_VERSION > 8
5558 if (IN_PERL_COMPILETIME)
5563 nextchar(pRExC_state);
5565 ret = reg_node(pRExC_state, LOGICAL);
5568 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5569 /* deal with the length of this later - MJD */
5572 ret = reganode(pRExC_state, EVAL, n);
5573 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5574 Set_Node_Offset(ret, parse_start);
5577 case '(': /* (?(?{...})...) and (?(?=...)...) */
5580 if (RExC_parse[0] == '?') { /* (?(?...)) */
5581 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5582 || RExC_parse[1] == '<'
5583 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5586 ret = reg_node(pRExC_state, LOGICAL);
5589 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5593 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5594 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5596 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5597 char *name_start= RExC_parse++;
5599 SV *sv_dat=reg_scan_name(pRExC_state,
5600 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5601 if (RExC_parse == name_start || *RExC_parse != ch)
5602 vFAIL2("Sequence (?(%c... not terminated",
5603 (ch == '>' ? '<' : ch));
5606 num = add_data( pRExC_state, 1, "S" );
5607 RExC_rxi->data->data[num]=(void*)sv_dat;
5608 SvREFCNT_inc_simple_void(sv_dat);
5610 ret = reganode(pRExC_state,NGROUPP,num);
5611 goto insert_if_check_paren;
5613 else if (RExC_parse[0] == 'D' &&
5614 RExC_parse[1] == 'E' &&
5615 RExC_parse[2] == 'F' &&
5616 RExC_parse[3] == 'I' &&
5617 RExC_parse[4] == 'N' &&
5618 RExC_parse[5] == 'E')
5620 ret = reganode(pRExC_state,DEFINEP,0);
5623 goto insert_if_check_paren;
5625 else if (RExC_parse[0] == 'R') {
5628 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5629 parno = atoi(RExC_parse++);
5630 while (isDIGIT(*RExC_parse))
5632 } else if (RExC_parse[0] == '&') {
5635 sv_dat = reg_scan_name(pRExC_state,
5636 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5637 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5639 ret = reganode(pRExC_state,INSUBP,parno);
5640 goto insert_if_check_paren;
5642 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5645 parno = atoi(RExC_parse++);
5647 while (isDIGIT(*RExC_parse))
5649 ret = reganode(pRExC_state, GROUPP, parno);
5651 insert_if_check_paren:
5652 if ((c = *nextchar(pRExC_state)) != ')')
5653 vFAIL("Switch condition not recognized");
5655 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5656 br = regbranch(pRExC_state, &flags, 1,depth+1);
5658 br = reganode(pRExC_state, LONGJMP, 0);
5660 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5661 c = *nextchar(pRExC_state);
5666 vFAIL("(?(DEFINE)....) does not allow branches");
5667 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
5668 regbranch(pRExC_state, &flags, 1,depth+1);
5669 REGTAIL(pRExC_state, ret, lastbr);
5672 c = *nextchar(pRExC_state);
5677 vFAIL("Switch (?(condition)... contains too many branches");
5678 ender = reg_node(pRExC_state, TAIL);
5679 REGTAIL(pRExC_state, br, ender);
5681 REGTAIL(pRExC_state, lastbr, ender);
5682 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
5685 REGTAIL(pRExC_state, ret, ender);
5686 RExC_size++; /* XXX WHY do we need this?!!
5687 For large programs it seems to be required
5688 but I can't figure out why. -- dmq*/
5692 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
5696 RExC_parse--; /* for vFAIL to print correctly */
5697 vFAIL("Sequence (? incomplete");
5701 parse_flags: /* (?i) */
5703 U32 posflags = 0, negflags = 0;
5704 U32 *flagsp = &posflags;
5706 while (*RExC_parse) {
5707 /* && strchr("iogcmsx", *RExC_parse) */
5708 /* (?g), (?gc) and (?o) are useless here
5709 and must be globally applied -- japhy */
5710 switch (*RExC_parse) {
5711 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
5714 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5715 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
5716 if (! (wastedflags & wflagbit) ) {
5717 wastedflags |= wflagbit;
5720 "Useless (%s%c) - %suse /%c modifier",
5721 flagsp == &negflags ? "?-" : "?",
5723 flagsp == &negflags ? "don't " : "",
5731 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
5732 if (! (wastedflags & WASTED_C) ) {
5733 wastedflags |= WASTED_GC;
5736 "Useless (%sc) - %suse /gc modifier",
5737 flagsp == &negflags ? "?-" : "?",
5738 flagsp == &negflags ? "don't " : ""
5744 if (flagsp == &negflags) {
5745 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
5746 vWARN(RExC_parse + 1,"Useless use of (?-k)");
5748 *flagsp |= RXf_PMf_KEEPCOPY;
5752 if (flagsp == &negflags) {
5754 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5758 wastedflags = 0; /* reset so (?g-c) warns twice */
5764 RExC_flags |= posflags;
5765 RExC_flags &= ~negflags;
5766 nextchar(pRExC_state);
5777 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5782 }} /* one for the default block, one for the switch */
5789 ret = reganode(pRExC_state, OPEN, parno);
5792 RExC_nestroot = parno;
5793 if (RExC_seen & REG_SEEN_RECURSE
5794 && !RExC_open_parens[parno-1])
5796 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5797 "Setting open paren #%"IVdf" to %d\n",
5798 (IV)parno, REG_NODE_NUM(ret)));
5799 RExC_open_parens[parno-1]= ret;
5802 Set_Node_Length(ret, 1); /* MJD */
5803 Set_Node_Offset(ret, RExC_parse); /* MJD */
5811 /* Pick up the branches, linking them together. */
5812 parse_start = RExC_parse; /* MJD */
5813 br = regbranch(pRExC_state, &flags, 1,depth+1);
5814 /* branch_len = (paren != 0); */
5818 if (*RExC_parse == '|') {
5819 if (!SIZE_ONLY && RExC_extralen) {
5820 reginsert(pRExC_state, BRANCHJ, br, depth+1);
5823 reginsert(pRExC_state, BRANCH, br, depth+1);
5824 Set_Node_Length(br, paren != 0);
5825 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
5829 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
5831 else if (paren == ':') {
5832 *flagp |= flags&SIMPLE;
5834 if (is_open) { /* Starts with OPEN. */
5835 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
5837 else if (paren != '?') /* Not Conditional */
5839 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5841 while (*RExC_parse == '|') {
5842 if (!SIZE_ONLY && RExC_extralen) {
5843 ender = reganode(pRExC_state, LONGJMP,0);
5844 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
5847 RExC_extralen += 2; /* Account for LONGJMP. */
5848 nextchar(pRExC_state);
5850 if (RExC_npar > after_freeze)
5851 after_freeze = RExC_npar;
5852 RExC_npar = freeze_paren;
5854 br = regbranch(pRExC_state, &flags, 0, depth+1);
5858 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
5860 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
5863 if (have_branch || paren != ':') {
5864 /* Make a closing node, and hook it on the end. */
5867 ender = reg_node(pRExC_state, TAIL);
5870 ender = reganode(pRExC_state, CLOSE, parno);
5871 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
5872 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5873 "Setting close paren #%"IVdf" to %d\n",
5874 (IV)parno, REG_NODE_NUM(ender)));
5875 RExC_close_parens[parno-1]= ender;
5876 if (RExC_nestroot == parno)
5879 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
5880 Set_Node_Length(ender,1); /* MJD */
5886 *flagp &= ~HASWIDTH;
5889 ender = reg_node(pRExC_state, SUCCEED);
5892 ender = reg_node(pRExC_state, END);
5894 assert(!RExC_opend); /* there can only be one! */
5899 REGTAIL(pRExC_state, lastbr, ender);
5901 if (have_branch && !SIZE_ONLY) {
5903 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
5905 /* Hook the tails of the branches to the closing node. */
5906 for (br = ret; br; br = regnext(br)) {
5907 const U8 op = PL_regkind[OP(br)];
5909 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
5911 else if (op == BRANCHJ) {
5912 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
5920 static const char parens[] = "=!<,>";
5922 if (paren && (p = strchr(parens, paren))) {
5923 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
5924 int flag = (p - parens) > 1;
5927 node = SUSPEND, flag = 0;
5928 reginsert(pRExC_state, node,ret, depth+1);
5929 Set_Node_Cur_Length(ret);
5930 Set_Node_Offset(ret, parse_start + 1);
5932 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
5936 /* Check for proper termination. */
5938 RExC_flags = oregflags;
5939 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
5940 RExC_parse = oregcomp_parse;
5941 vFAIL("Unmatched (");
5944 else if (!paren && RExC_parse < RExC_end) {
5945 if (*RExC_parse == ')') {
5947 vFAIL("Unmatched )");
5950 FAIL("Junk on end of regexp"); /* "Can't happen". */
5954 RExC_npar = after_freeze;
5959 - regbranch - one alternative of an | operator
5961 * Implements the concatenation operator.
5964 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
5967 register regnode *ret;
5968 register regnode *chain = NULL;
5969 register regnode *latest;
5970 I32 flags = 0, c = 0;
5971 GET_RE_DEBUG_FLAGS_DECL;
5972 DEBUG_PARSE("brnc");
5977 if (!SIZE_ONLY && RExC_extralen)
5978 ret = reganode(pRExC_state, BRANCHJ,0);
5980 ret = reg_node(pRExC_state, BRANCH);
5981 Set_Node_Length(ret, 1);
5985 if (!first && SIZE_ONLY)
5986 RExC_extralen += 1; /* BRANCHJ */
5988 *flagp = WORST; /* Tentatively. */
5991 nextchar(pRExC_state);
5992 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
5994 latest = regpiece(pRExC_state, &flags,depth+1);
5995 if (latest == NULL) {
5996 if (flags & TRYAGAIN)
6000 else if (ret == NULL)
6002 *flagp |= flags&(HASWIDTH|POSTPONED);
6003 if (chain == NULL) /* First piece. */
6004 *flagp |= flags&SPSTART;
6007 REGTAIL(pRExC_state, chain, latest);
6012 if (chain == NULL) { /* Loop ran zero times. */
6013 chain = reg_node(pRExC_state, NOTHING);
6018 *flagp |= flags&SIMPLE;
6025 - regpiece - something followed by possible [*+?]
6027 * Note that the branching code sequences used for ? and the general cases
6028 * of * and + are somewhat optimized: they use the same NOTHING node as
6029 * both the endmarker for their branch list and the body of the last branch.
6030 * It might seem that this node could be dispensed with entirely, but the
6031 * endmarker role is not redundant.
6034 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6037 register regnode *ret;
6039 register char *next;
6041 const char * const origparse = RExC_parse;
6043 I32 max = REG_INFTY;
6045 const char *maxpos = NULL;
6046 GET_RE_DEBUG_FLAGS_DECL;
6047 DEBUG_PARSE("piec");
6049 ret = regatom(pRExC_state, &flags,depth+1);
6051 if (flags & TRYAGAIN)
6058 if (op == '{' && regcurly(RExC_parse)) {
6060 parse_start = RExC_parse; /* MJD */
6061 next = RExC_parse + 1;
6062 while (isDIGIT(*next) || *next == ',') {
6071 if (*next == '}') { /* got one */
6075 min = atoi(RExC_parse);
6079 maxpos = RExC_parse;
6081 if (!max && *maxpos != '0')
6082 max = REG_INFTY; /* meaning "infinity" */
6083 else if (max >= REG_INFTY)
6084 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6086 nextchar(pRExC_state);
6089 if ((flags&SIMPLE)) {
6090 RExC_naughty += 2 + RExC_naughty / 2;
6091 reginsert(pRExC_state, CURLY, ret, depth+1);
6092 Set_Node_Offset(ret, parse_start+1); /* MJD */
6093 Set_Node_Cur_Length(ret);
6096 regnode * const w = reg_node(pRExC_state, WHILEM);
6099 REGTAIL(pRExC_state, ret, w);
6100 if (!SIZE_ONLY && RExC_extralen) {
6101 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6102 reginsert(pRExC_state, NOTHING,ret, depth+1);
6103 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6105 reginsert(pRExC_state, CURLYX,ret, depth+1);
6107 Set_Node_Offset(ret, parse_start+1);
6108 Set_Node_Length(ret,
6109 op == '{' ? (RExC_parse - parse_start) : 1);
6111 if (!SIZE_ONLY && RExC_extralen)
6112 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6113 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6115 RExC_whilem_seen++, RExC_extralen += 3;
6116 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6124 if (max && max < min)
6125 vFAIL("Can't do {n,m} with n > m");
6127 ARG1_SET(ret, (U16)min);
6128 ARG2_SET(ret, (U16)max);
6140 #if 0 /* Now runtime fix should be reliable. */
6142 /* if this is reinstated, don't forget to put this back into perldiag:
6144 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6146 (F) The part of the regexp subject to either the * or + quantifier
6147 could match an empty string. The {#} shows in the regular
6148 expression about where the problem was discovered.
6152 if (!(flags&HASWIDTH) && op != '?')
6153 vFAIL("Regexp *+ operand could be empty");
6156 parse_start = RExC_parse;
6157 nextchar(pRExC_state);
6159 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6161 if (op == '*' && (flags&SIMPLE)) {
6162 reginsert(pRExC_state, STAR, ret, depth+1);
6166 else if (op == '*') {
6170 else if (op == '+' && (flags&SIMPLE)) {
6171 reginsert(pRExC_state, PLUS, ret, depth+1);
6175 else if (op == '+') {
6179 else if (op == '?') {
6184 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6186 "%.*s matches null string many times",
6187 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6191 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6192 nextchar(pRExC_state);
6193 reginsert(pRExC_state, MINMOD, ret, depth+1);
6194 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6196 #ifndef REG_ALLOW_MINMOD_SUSPEND
6199 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6201 nextchar(pRExC_state);
6202 ender = reg_node(pRExC_state, SUCCEED);
6203 REGTAIL(pRExC_state, ret, ender);
6204 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6206 ender = reg_node(pRExC_state, TAIL);
6207 REGTAIL(pRExC_state, ret, ender);
6211 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6213 vFAIL("Nested quantifiers");
6220 /* reg_namedseq(pRExC_state,UVp)
6222 This is expected to be called by a parser routine that has
6223 recognized'\N' and needs to handle the rest. RExC_parse is
6224 expected to point at the first char following the N at the time
6227 If valuep is non-null then it is assumed that we are parsing inside
6228 of a charclass definition and the first codepoint in the resolved
6229 string is returned via *valuep and the routine will return NULL.
6230 In this mode if a multichar string is returned from the charnames
6231 handler a warning will be issued, and only the first char in the
6232 sequence will be examined. If the string returned is zero length
6233 then the value of *valuep is undefined and NON-NULL will
6234 be returned to indicate failure. (This will NOT be a valid pointer
6237 If value is null then it is assumed that we are parsing normal text
6238 and inserts a new EXACT node into the program containing the resolved
6239 string and returns a pointer to the new node. If the string is
6240 zerolength a NOTHING node is emitted.
6242 On success RExC_parse is set to the char following the endbrace.
6243 Parsing failures will generate a fatal errorvia vFAIL(...)
6245 NOTE: We cache all results from the charnames handler locally in
6246 the RExC_charnames hash (created on first use) to prevent a charnames
6247 handler from playing silly-buggers and returning a short string and
6248 then a long string for a given pattern. Since the regexp program
6249 size is calculated during an initial parse this would result
6250 in a buffer overrun so we cache to prevent the charname result from
6251 changing during the course of the parse.
6255 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
6257 char * name; /* start of the content of the name */
6258 char * endbrace; /* endbrace following the name */
6261 STRLEN len; /* this has various purposes throughout the code */
6262 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6263 regnode *ret = NULL;
6265 if (*RExC_parse != '{') {
6266 vFAIL("Missing braces on \\N{}");
6268 name = RExC_parse+1;
6269 endbrace = strchr(RExC_parse, '}');
6272 vFAIL("Missing right brace on \\N{}");
6274 RExC_parse = endbrace + 1;
6277 /* RExC_parse points at the beginning brace,
6278 endbrace points at the last */
6279 if ( name[0]=='U' && name[1]=='+' ) {
6280 /* its a "unicode hex" notation {U+89AB} */
6281 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6282 | PERL_SCAN_DISALLOW_PREFIX
6283 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6285 len = (STRLEN)(endbrace - name - 2);
6286 cp = grok_hex(name + 2, &len, &fl, NULL);
6287 if ( len != (STRLEN)(endbrace - name - 2) ) {
6296 sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
6298 /* fetch the charnames handler for this scope */
6299 HV * const table = GvHV(PL_hintgv);
6301 hv_fetchs(table, "charnames", FALSE) :
6303 SV *cv= cvp ? *cvp : NULL;
6306 /* create an SV with the name as argument */
6307 sv_name = newSVpvn(name, endbrace - name);
6309 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6310 vFAIL2("Constant(\\N{%s}) unknown: "
6311 "(possibly a missing \"use charnames ...\")",
6314 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6315 vFAIL2("Constant(\\N{%s}): "
6316 "$^H{charnames} is not defined",SvPVX(sv_name));
6321 if (!RExC_charnames) {
6322 /* make sure our cache is allocated */
6323 RExC_charnames = newHV();
6324 sv_2mortal((SV*)RExC_charnames);
6326 /* see if we have looked this one up before */
6327 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6329 sv_str = HeVAL(he_str);
6342 count= call_sv(cv, G_SCALAR);
6344 if (count == 1) { /* XXXX is this right? dmq */
6346 SvREFCNT_inc_simple_void(sv_str);
6354 if ( !sv_str || !SvOK(sv_str) ) {
6355 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6356 "did not return a defined value",SvPVX(sv_name));
6358 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6363 char *p = SvPV(sv_str, len);
6366 if ( SvUTF8(sv_str) ) {
6367 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6371 We have to turn on utf8 for high bit chars otherwise
6372 we get failures with
6374 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6375 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6377 This is different from what \x{} would do with the same
6378 codepoint, where the condition is > 0xFF.
6385 /* warn if we havent used the whole string? */
6387 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6389 "Ignoring excess chars from \\N{%s} in character class",
6393 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6395 "Ignoring zero length \\N{%s} in character class",
6400 SvREFCNT_dec(sv_name);
6402 SvREFCNT_dec(sv_str);
6403 return len ? NULL : (regnode *)&len;
6404 } else if(SvCUR(sv_str)) {
6410 char * parse_start = name-3; /* needed for the offsets */
6412 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6414 ret = reg_node(pRExC_state,
6415 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6418 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6419 sv_utf8_upgrade(sv_str);
6420 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6424 p = SvPV(sv_str, len);
6426 /* len is the length written, charlen is the size the char read */
6427 for ( len = 0; p < pend; p += charlen ) {
6429 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6431 STRLEN foldlen,numlen;
6432 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6433 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6434 /* Emit all the Unicode characters. */
6436 for (foldbuf = tmpbuf;
6440 uvc = utf8_to_uvchr(foldbuf, &numlen);
6442 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6445 /* In EBCDIC the numlen
6446 * and unilen can differ. */
6448 if (numlen >= foldlen)
6452 break; /* "Can't happen." */
6455 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6467 RExC_size += STR_SZ(len);
6470 RExC_emit += STR_SZ(len);
6472 Set_Node_Cur_Length(ret); /* MJD */
6474 nextchar(pRExC_state);
6476 ret = reg_node(pRExC_state,NOTHING);
6479 SvREFCNT_dec(sv_str);
6482 SvREFCNT_dec(sv_name);
6492 * It returns the code point in utf8 for the value in *encp.
6493 * value: a code value in the source encoding
6494 * encp: a pointer to an Encode object
6496 * If the result from Encode is not a single character,
6497 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6500 S_reg_recode(pTHX_ const char value, SV **encp)
6503 SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
6504 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6505 const STRLEN newlen = SvCUR(sv);
6506 UV uv = UNICODE_REPLACEMENT;
6510 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6513 if (!newlen || numlen != newlen) {
6514 uv = UNICODE_REPLACEMENT;
6522 - regatom - the lowest level
6524 Try to identify anything special at the start of the pattern. If there
6525 is, then handle it as required. This may involve generating a single regop,
6526 such as for an assertion; or it may involve recursing, such as to
6527 handle a () structure.
6529 If the string doesn't start with something special then we gobble up
6530 as much literal text as we can.
6532 Once we have been able to handle whatever type of thing started the
6533 sequence, we return.
6535 Note: we have to be careful with escapes, as they can be both literal
6536 and special, and in the case of \10 and friends can either, depending
6537 on context. Specifically there are two seperate switches for handling
6538 escape sequences, with the one for handling literal escapes requiring
6539 a dummy entry for all of the special escapes that are actually handled
6544 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6547 register regnode *ret = NULL;
6549 char *parse_start = RExC_parse;
6550 GET_RE_DEBUG_FLAGS_DECL;
6551 DEBUG_PARSE("atom");
6552 *flagp = WORST; /* Tentatively. */
6556 switch ((U8)*RExC_parse) {
6558 RExC_seen_zerolen++;
6559 nextchar(pRExC_state);
6560 if (RExC_flags & RXf_PMf_MULTILINE)
6561 ret = reg_node(pRExC_state, MBOL);
6562 else if (RExC_flags & RXf_PMf_SINGLELINE)
6563 ret = reg_node(pRExC_state, SBOL);
6565 ret = reg_node(pRExC_state, BOL);
6566 Set_Node_Length(ret, 1); /* MJD */
6569 nextchar(pRExC_state);
6571 RExC_seen_zerolen++;
6572 if (RExC_flags & RXf_PMf_MULTILINE)
6573 ret = reg_node(pRExC_state, MEOL);
6574 else if (RExC_flags & RXf_PMf_SINGLELINE)
6575 ret = reg_node(pRExC_state, SEOL);
6577 ret = reg_node(pRExC_state, EOL);
6578 Set_Node_Length(ret, 1); /* MJD */
6581 nextchar(pRExC_state);
6582 if (RExC_flags & RXf_PMf_SINGLELINE)
6583 ret = reg_node(pRExC_state, SANY);
6585 ret = reg_node(pRExC_state, REG_ANY);
6586 *flagp |= HASWIDTH|SIMPLE;
6588 Set_Node_Length(ret, 1); /* MJD */
6592 char * const oregcomp_parse = ++RExC_parse;
6593 ret = regclass(pRExC_state,depth+1);
6594 if (*RExC_parse != ']') {
6595 RExC_parse = oregcomp_parse;
6596 vFAIL("Unmatched [");
6598 nextchar(pRExC_state);
6599 *flagp |= HASWIDTH|SIMPLE;
6600 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6604 nextchar(pRExC_state);
6605 ret = reg(pRExC_state, 1, &flags,depth+1);
6607 if (flags & TRYAGAIN) {
6608 if (RExC_parse == RExC_end) {
6609 /* Make parent create an empty node if needed. */
6617 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
6621 if (flags & TRYAGAIN) {
6625 vFAIL("Internal urp");
6626 /* Supposed to be caught earlier. */
6629 if (!regcurly(RExC_parse)) {
6638 vFAIL("Quantifier follows nothing");
6645 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
6646 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
6647 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
6648 ret = reganode(pRExC_state, FOLDCHAR, cp);
6649 Set_Node_Length(ret, 1); /* MJD */
6650 nextchar(pRExC_state); /* kill whitespace under /x */
6658 This switch handles escape sequences that resolve to some kind
6659 of special regop and not to literal text. Escape sequnces that
6660 resolve to literal text are handled below in the switch marked
6663 Every entry in this switch *must* have a corresponding entry
6664 in the literal escape switch. However, the opposite is not
6665 required, as the default for this switch is to jump to the
6666 literal text handling code.
6668 switch (*++RExC_parse) {
6669 /* Special Escapes */
6671 RExC_seen_zerolen++;
6672 ret = reg_node(pRExC_state, SBOL);
6674 goto finish_meta_pat;
6676 ret = reg_node(pRExC_state, GPOS);
6677 RExC_seen |= REG_SEEN_GPOS;
6679 goto finish_meta_pat;
6681 RExC_seen_zerolen++;
6682 ret = reg_node(pRExC_state, KEEPS);
6684 goto finish_meta_pat;
6686 ret = reg_node(pRExC_state, SEOL);
6688 RExC_seen_zerolen++; /* Do not optimize RE away */
6689 goto finish_meta_pat;
6691 ret = reg_node(pRExC_state, EOS);
6693 RExC_seen_zerolen++; /* Do not optimize RE away */
6694 goto finish_meta_pat;
6696 ret = reg_node(pRExC_state, CANY);
6697 RExC_seen |= REG_SEEN_CANY;
6698 *flagp |= HASWIDTH|SIMPLE;
6699 goto finish_meta_pat;
6701 ret = reg_node(pRExC_state, CLUMP);
6703 goto finish_meta_pat;
6705 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
6706 *flagp |= HASWIDTH|SIMPLE;
6707 goto finish_meta_pat;
6709 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
6710 *flagp |= HASWIDTH|SIMPLE;
6711 goto finish_meta_pat;
6713 RExC_seen_zerolen++;
6714 RExC_seen |= REG_SEEN_LOOKBEHIND;
6715 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
6717 goto finish_meta_pat;
6719 RExC_seen_zerolen++;
6720 RExC_seen |= REG_SEEN_LOOKBEHIND;
6721 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
6723 goto finish_meta_pat;
6725 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
6726 *flagp |= HASWIDTH|SIMPLE;
6727 goto finish_meta_pat;
6729 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
6730 *flagp |= HASWIDTH|SIMPLE;
6731 goto finish_meta_pat;
6733 ret = reg_node(pRExC_state, DIGIT);
6734 *flagp |= HASWIDTH|SIMPLE;
6735 goto finish_meta_pat;
6737 ret = reg_node(pRExC_state, NDIGIT);
6738 *flagp |= HASWIDTH|SIMPLE;
6739 goto finish_meta_pat;
6741 ret = reg_node(pRExC_state, LNBREAK);
6742 *flagp |= HASWIDTH|SIMPLE;
6743 goto finish_meta_pat;
6745 ret = reg_node(pRExC_state, HORIZWS);
6746 *flagp |= HASWIDTH|SIMPLE;
6747 goto finish_meta_pat;
6749 ret = reg_node(pRExC_state, NHORIZWS);
6750 *flagp |= HASWIDTH|SIMPLE;
6751 goto finish_meta_pat;
6753 ret = reg_node(pRExC_state, VERTWS);
6754 *flagp |= HASWIDTH|SIMPLE;
6755 goto finish_meta_pat;
6757 ret = reg_node(pRExC_state, NVERTWS);
6758 *flagp |= HASWIDTH|SIMPLE;
6760 nextchar(pRExC_state);
6761 Set_Node_Length(ret, 2); /* MJD */
6766 char* const oldregxend = RExC_end;
6768 char* parse_start = RExC_parse - 2;
6771 if (RExC_parse[1] == '{') {
6772 /* a lovely hack--pretend we saw [\pX] instead */
6773 RExC_end = strchr(RExC_parse, '}');
6775 const U8 c = (U8)*RExC_parse;
6777 RExC_end = oldregxend;
6778 vFAIL2("Missing right brace on \\%c{}", c);
6783 RExC_end = RExC_parse + 2;
6784 if (RExC_end > oldregxend)
6785 RExC_end = oldregxend;
6789 ret = regclass(pRExC_state,depth+1);
6791 RExC_end = oldregxend;
6794 Set_Node_Offset(ret, parse_start + 2);
6795 Set_Node_Cur_Length(ret);
6796 nextchar(pRExC_state);
6797 *flagp |= HASWIDTH|SIMPLE;
6801 /* Handle \N{NAME} here and not below because it can be
6802 multicharacter. join_exact() will join them up later on.
6803 Also this makes sure that things like /\N{BLAH}+/ and
6804 \N{BLAH} being multi char Just Happen. dmq*/
6806 ret= reg_namedseq(pRExC_state, NULL);
6808 case 'k': /* Handle \k<NAME> and \k'NAME' */
6811 char ch= RExC_parse[1];
6812 if (ch != '<' && ch != '\'' && ch != '{') {
6814 vFAIL2("Sequence %.2s... not terminated",parse_start);
6816 /* this pretty much dupes the code for (?P=...) in reg(), if
6817 you change this make sure you change that */
6818 char* name_start = (RExC_parse += 2);
6820 SV *sv_dat = reg_scan_name(pRExC_state,
6821 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6822 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
6823 if (RExC_parse == name_start || *RExC_parse != ch)
6824 vFAIL2("Sequence %.3s... not terminated",parse_start);
6827 num = add_data( pRExC_state, 1, "S" );
6828 RExC_rxi->data->data[num]=(void*)sv_dat;
6829 SvREFCNT_inc_simple_void(sv_dat);
6833 ret = reganode(pRExC_state,
6834 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
6838 /* override incorrect value set in reganode MJD */
6839 Set_Node_Offset(ret, parse_start+1);
6840 Set_Node_Cur_Length(ret); /* MJD */
6841 nextchar(pRExC_state);
6847 case '1': case '2': case '3': case '4':
6848 case '5': case '6': case '7': case '8': case '9':
6851 bool isg = *RExC_parse == 'g';
6856 if (*RExC_parse == '{') {
6860 if (*RExC_parse == '-') {
6864 if (hasbrace && !isDIGIT(*RExC_parse)) {
6865 if (isrel) RExC_parse--;
6867 goto parse_named_seq;
6869 num = atoi(RExC_parse);
6871 num = RExC_npar - num;
6873 vFAIL("Reference to nonexistent or unclosed group");
6875 if (!isg && num > 9 && num >= RExC_npar)
6878 char * const parse_start = RExC_parse - 1; /* MJD */
6879 while (isDIGIT(*RExC_parse))
6881 if (parse_start == RExC_parse - 1)
6882 vFAIL("Unterminated \\g... pattern");
6884 if (*RExC_parse != '}')
6885 vFAIL("Unterminated \\g{...} pattern");
6889 if (num > (I32)RExC_rx->nparens)
6890 vFAIL("Reference to nonexistent group");
6893 ret = reganode(pRExC_state,
6894 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
6898 /* override incorrect value set in reganode MJD */
6899 Set_Node_Offset(ret, parse_start+1);
6900 Set_Node_Cur_Length(ret); /* MJD */
6902 nextchar(pRExC_state);
6907 if (RExC_parse >= RExC_end)
6908 FAIL("Trailing \\");
6911 /* Do not generate "unrecognized" warnings here, we fall
6912 back into the quick-grab loop below */
6919 if (RExC_flags & RXf_PMf_EXTENDED) {
6920 if ( reg_skipcomment( pRExC_state ) )
6927 register STRLEN len;
6932 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6934 parse_start = RExC_parse - 1;
6940 ret = reg_node(pRExC_state,
6941 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6943 for (len = 0, p = RExC_parse - 1;
6944 len < 127 && p < RExC_end;
6947 char * const oldp = p;
6949 if (RExC_flags & RXf_PMf_EXTENDED)
6950 p = regwhite( pRExC_state, p );
6955 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
6956 goto normal_default;
6966 /* Literal Escapes Switch
6968 This switch is meant to handle escape sequences that
6969 resolve to a literal character.
6971 Every escape sequence that represents something
6972 else, like an assertion or a char class, is handled
6973 in the switch marked 'Special Escapes' above in this
6974 routine, but also has an entry here as anything that
6975 isn't explicitly mentioned here will be treated as
6976 an unescaped equivalent literal.
6980 /* These are all the special escapes. */
6981 case 'A': /* Start assertion */
6982 case 'b': case 'B': /* Word-boundary assertion*/
6983 case 'C': /* Single char !DANGEROUS! */
6984 case 'd': case 'D': /* digit class */
6985 case 'g': case 'G': /* generic-backref, pos assertion */
6986 case 'h': case 'H': /* HORIZWS */
6987 case 'k': case 'K': /* named backref, keep marker */
6988 case 'N': /* named char sequence */
6989 case 'p': case 'P': /* unicode property */
6990 case 'R': /* LNBREAK */
6991 case 's': case 'S': /* space class */
6992 case 'v': case 'V': /* VERTWS */
6993 case 'w': case 'W': /* word class */
6994 case 'X': /* eXtended Unicode "combining character sequence" */
6995 case 'z': case 'Z': /* End of line/string assertion */
6999 /* Anything after here is an escape that resolves to a
7000 literal. (Except digits, which may or may not)
7019 ender = ASCII_TO_NATIVE('\033');
7023 ender = ASCII_TO_NATIVE('\007');
7028 char* const e = strchr(p, '}');
7032 vFAIL("Missing right brace on \\x{}");
7035 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7036 | PERL_SCAN_DISALLOW_PREFIX;
7037 STRLEN numlen = e - p - 1;
7038 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7045 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7047 ender = grok_hex(p, &numlen, &flags, NULL);
7050 if (PL_encoding && ender < 0x100)
7051 goto recode_encoding;
7055 ender = UCHARAT(p++);
7056 ender = toCTRL(ender);
7058 case '0': case '1': case '2': case '3':case '4':
7059 case '5': case '6': case '7': case '8':case '9':
7061 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7064 ender = grok_oct(p, &numlen, &flags, NULL);
7071 if (PL_encoding && ender < 0x100)
7072 goto recode_encoding;
7076 SV* enc = PL_encoding;
7077 ender = reg_recode((const char)(U8)ender, &enc);
7078 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7079 vWARN(p, "Invalid escape in the specified encoding");
7085 FAIL("Trailing \\");
7088 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
7089 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7090 goto normal_default;
7095 if (UTF8_IS_START(*p) && UTF) {
7097 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7098 &numlen, UTF8_ALLOW_DEFAULT);
7105 if ( RExC_flags & RXf_PMf_EXTENDED)
7106 p = regwhite( pRExC_state, p );
7108 /* Prime the casefolded buffer. */
7109 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7111 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7116 /* Emit all the Unicode characters. */
7118 for (foldbuf = tmpbuf;
7120 foldlen -= numlen) {
7121 ender = utf8_to_uvchr(foldbuf, &numlen);
7123 const STRLEN unilen = reguni(pRExC_state, ender, s);
7126 /* In EBCDIC the numlen
7127 * and unilen can differ. */
7129 if (numlen >= foldlen)
7133 break; /* "Can't happen." */
7137 const STRLEN unilen = reguni(pRExC_state, ender, s);
7146 REGC((char)ender, s++);
7152 /* Emit all the Unicode characters. */
7154 for (foldbuf = tmpbuf;
7156 foldlen -= numlen) {
7157 ender = utf8_to_uvchr(foldbuf, &numlen);
7159 const STRLEN unilen = reguni(pRExC_state, ender, s);
7162 /* In EBCDIC the numlen
7163 * and unilen can differ. */
7165 if (numlen >= foldlen)
7173 const STRLEN unilen = reguni(pRExC_state, ender, s);
7182 REGC((char)ender, s++);
7186 Set_Node_Cur_Length(ret); /* MJD */
7187 nextchar(pRExC_state);
7189 /* len is STRLEN which is unsigned, need to copy to signed */
7192 vFAIL("Internal disaster");
7196 if (len == 1 && UNI_IS_INVARIANT(ender))
7200 RExC_size += STR_SZ(len);
7203 RExC_emit += STR_SZ(len);
7213 S_regwhite( RExC_state_t *pRExC_state, char *p )
7215 const char *e = RExC_end;
7219 else if (*p == '#') {
7228 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7236 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7237 Character classes ([:foo:]) can also be negated ([:^foo:]).
7238 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7239 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7240 but trigger failures because they are currently unimplemented. */
7242 #define POSIXCC_DONE(c) ((c) == ':')
7243 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7244 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7247 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7250 I32 namedclass = OOB_NAMEDCLASS;
7252 if (value == '[' && RExC_parse + 1 < RExC_end &&
7253 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7254 POSIXCC(UCHARAT(RExC_parse))) {
7255 const char c = UCHARAT(RExC_parse);
7256 char* const s = RExC_parse++;
7258 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7260 if (RExC_parse == RExC_end)
7261 /* Grandfather lone [:, [=, [. */
7264 const char* const t = RExC_parse++; /* skip over the c */
7267 if (UCHARAT(RExC_parse) == ']') {
7268 const char *posixcc = s + 1;
7269 RExC_parse++; /* skip over the ending ] */
7272 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7273 const I32 skip = t - posixcc;
7275 /* Initially switch on the length of the name. */
7278 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7279 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7282 /* Names all of length 5. */
7283 /* alnum alpha ascii blank cntrl digit graph lower
7284 print punct space upper */
7285 /* Offset 4 gives the best switch position. */
7286 switch (posixcc[4]) {
7288 if (memEQ(posixcc, "alph", 4)) /* alpha */
7289 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7292 if (memEQ(posixcc, "spac", 4)) /* space */
7293 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7296 if (memEQ(posixcc, "grap", 4)) /* graph */
7297 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7300 if (memEQ(posixcc, "asci", 4)) /* ascii */
7301 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7304 if (memEQ(posixcc, "blan", 4)) /* blank */
7305 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7308 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7309 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7312 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7313 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7316 if (memEQ(posixcc, "lowe", 4)) /* lower */
7317 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7318 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7319 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7322 if (memEQ(posixcc, "digi", 4)) /* digit */
7323 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7324 else if (memEQ(posixcc, "prin", 4)) /* print */
7325 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7326 else if (memEQ(posixcc, "punc", 4)) /* punct */
7327 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7332 if (memEQ(posixcc, "xdigit", 6))
7333 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7337 if (namedclass == OOB_NAMEDCLASS)
7338 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7340 assert (posixcc[skip] == ':');
7341 assert (posixcc[skip+1] == ']');
7342 } else if (!SIZE_ONLY) {
7343 /* [[=foo=]] and [[.foo.]] are still future. */
7345 /* adjust RExC_parse so the warning shows after
7347 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7349 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7352 /* Maternal grandfather:
7353 * "[:" ending in ":" but not in ":]" */
7363 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7366 if (POSIXCC(UCHARAT(RExC_parse))) {
7367 const char *s = RExC_parse;
7368 const char c = *s++;
7372 if (*s && c == *s && s[1] == ']') {
7373 if (ckWARN(WARN_REGEXP))
7375 "POSIX syntax [%c %c] belongs inside character classes",
7378 /* [[=foo=]] and [[.foo.]] are still future. */
7379 if (POSIXCC_NOTYET(c)) {
7380 /* adjust RExC_parse so the error shows after
7382 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7384 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7391 #define _C_C_T_(NAME,TEST,WORD) \
7394 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7396 for (value = 0; value < 256; value++) \
7398 ANYOF_BITMAP_SET(ret, value); \
7403 case ANYOF_N##NAME: \
7405 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7407 for (value = 0; value < 256; value++) \
7409 ANYOF_BITMAP_SET(ret, value); \
7415 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7417 for (value = 0; value < 256; value++) \
7419 ANYOF_BITMAP_SET(ret, value); \
7423 case ANYOF_N##NAME: \
7424 for (value = 0; value < 256; value++) \
7426 ANYOF_BITMAP_SET(ret, value); \
7432 parse a class specification and produce either an ANYOF node that
7433 matches the pattern or if the pattern matches a single char only and
7434 that char is < 256 and we are case insensitive then we produce an
7439 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7442 register UV nextvalue;
7443 register IV prevvalue = OOB_UNICODE;
7444 register IV range = 0;
7445 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7446 register regnode *ret;
7449 char *rangebegin = NULL;
7450 bool need_class = 0;
7453 bool optimize_invert = TRUE;
7454 AV* unicode_alternate = NULL;
7456 UV literal_endpoint = 0;
7458 UV stored = 0; /* number of chars stored in the class */
7460 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7461 case we need to change the emitted regop to an EXACT. */
7462 const char * orig_parse = RExC_parse;
7463 GET_RE_DEBUG_FLAGS_DECL;
7465 PERL_UNUSED_ARG(depth);
7468 DEBUG_PARSE("clas");
7470 /* Assume we are going to generate an ANYOF node. */
7471 ret = reganode(pRExC_state, ANYOF, 0);
7474 ANYOF_FLAGS(ret) = 0;
7476 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7480 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7484 RExC_size += ANYOF_SKIP;
7485 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7488 RExC_emit += ANYOF_SKIP;
7490 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7492 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7493 ANYOF_BITMAP_ZERO(ret);
7494 listsv = newSVpvs("# comment\n");
7497 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7499 if (!SIZE_ONLY && POSIXCC(nextvalue))
7500 checkposixcc(pRExC_state);
7502 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7503 if (UCHARAT(RExC_parse) == ']')
7507 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7511 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7514 rangebegin = RExC_parse;
7516 value = utf8n_to_uvchr((U8*)RExC_parse,
7517 RExC_end - RExC_parse,
7518 &numlen, UTF8_ALLOW_DEFAULT);
7519 RExC_parse += numlen;
7522 value = UCHARAT(RExC_parse++);
7524 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7525 if (value == '[' && POSIXCC(nextvalue))
7526 namedclass = regpposixcc(pRExC_state, value);
7527 else if (value == '\\') {
7529 value = utf8n_to_uvchr((U8*)RExC_parse,
7530 RExC_end - RExC_parse,
7531 &numlen, UTF8_ALLOW_DEFAULT);
7532 RExC_parse += numlen;
7535 value = UCHARAT(RExC_parse++);
7536 /* Some compilers cannot handle switching on 64-bit integer
7537 * values, therefore value cannot be an UV. Yes, this will
7538 * be a problem later if we want switch on Unicode.
7539 * A similar issue a little bit later when switching on
7540 * namedclass. --jhi */
7541 switch ((I32)value) {
7542 case 'w': namedclass = ANYOF_ALNUM; break;
7543 case 'W': namedclass = ANYOF_NALNUM; break;
7544 case 's': namedclass = ANYOF_SPACE; break;
7545 case 'S': namedclass = ANYOF_NSPACE; break;
7546 case 'd': namedclass = ANYOF_DIGIT; break;
7547 case 'D': namedclass = ANYOF_NDIGIT; break;
7548 case 'v': namedclass = ANYOF_VERTWS; break;
7549 case 'V': namedclass = ANYOF_NVERTWS; break;
7550 case 'h': namedclass = ANYOF_HORIZWS; break;
7551 case 'H': namedclass = ANYOF_NHORIZWS; break;
7552 case 'N': /* Handle \N{NAME} in class */
7554 /* We only pay attention to the first char of
7555 multichar strings being returned. I kinda wonder
7556 if this makes sense as it does change the behaviour
7557 from earlier versions, OTOH that behaviour was broken
7559 UV v; /* value is register so we cant & it /grrr */
7560 if (reg_namedseq(pRExC_state, &v)) {
7570 if (RExC_parse >= RExC_end)
7571 vFAIL2("Empty \\%c{}", (U8)value);
7572 if (*RExC_parse == '{') {
7573 const U8 c = (U8)value;
7574 e = strchr(RExC_parse++, '}');
7576 vFAIL2("Missing right brace on \\%c{}", c);
7577 while (isSPACE(UCHARAT(RExC_parse)))
7579 if (e == RExC_parse)
7580 vFAIL2("Empty \\%c{}", c);
7582 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
7590 if (UCHARAT(RExC_parse) == '^') {
7593 value = value == 'p' ? 'P' : 'p'; /* toggle */
7594 while (isSPACE(UCHARAT(RExC_parse))) {
7599 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
7600 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
7603 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7604 namedclass = ANYOF_MAX; /* no official name, but it's named */
7607 case 'n': value = '\n'; break;
7608 case 'r': value = '\r'; break;
7609 case 't': value = '\t'; break;
7610 case 'f': value = '\f'; break;
7611 case 'b': value = '\b'; break;
7612 case 'e': value = ASCII_TO_NATIVE('\033');break;
7613 case 'a': value = ASCII_TO_NATIVE('\007');break;
7615 if (*RExC_parse == '{') {
7616 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7617 | PERL_SCAN_DISALLOW_PREFIX;
7618 char * const e = strchr(RExC_parse++, '}');
7620 vFAIL("Missing right brace on \\x{}");
7622 numlen = e - RExC_parse;
7623 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7627 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7629 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
7630 RExC_parse += numlen;
7632 if (PL_encoding && value < 0x100)
7633 goto recode_encoding;
7636 value = UCHARAT(RExC_parse++);
7637 value = toCTRL(value);
7639 case '0': case '1': case '2': case '3': case '4':
7640 case '5': case '6': case '7': case '8': case '9':
7644 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
7645 RExC_parse += numlen;
7646 if (PL_encoding && value < 0x100)
7647 goto recode_encoding;
7652 SV* enc = PL_encoding;
7653 value = reg_recode((const char)(U8)value, &enc);
7654 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7656 "Invalid escape in the specified encoding");
7660 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
7662 "Unrecognized escape \\%c in character class passed through",
7666 } /* end of \blah */
7672 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
7674 if (!SIZE_ONLY && !need_class)
7675 ANYOF_CLASS_ZERO(ret);
7679 /* a bad range like a-\d, a-[:digit:] ? */
7682 if (ckWARN(WARN_REGEXP)) {
7684 RExC_parse >= rangebegin ?
7685 RExC_parse - rangebegin : 0;
7687 "False [] range \"%*.*s\"",
7690 if (prevvalue < 256) {
7691 ANYOF_BITMAP_SET(ret, prevvalue);
7692 ANYOF_BITMAP_SET(ret, '-');
7695 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7696 Perl_sv_catpvf(aTHX_ listsv,
7697 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
7701 range = 0; /* this was not a true range */
7707 const char *what = NULL;
7710 if (namedclass > OOB_NAMEDCLASS)
7711 optimize_invert = FALSE;
7712 /* Possible truncation here but in some 64-bit environments
7713 * the compiler gets heartburn about switch on 64-bit values.
7714 * A similar issue a little earlier when switching on value.
7716 switch ((I32)namedclass) {
7717 case _C_C_T_(ALNUM, isALNUM(value), "Word");
7718 case _C_C_T_(ALNUMC, isALNUMC(value), "Alnum");
7719 case _C_C_T_(ALPHA, isALPHA(value), "Alpha");
7720 case _C_C_T_(BLANK, isBLANK(value), "Blank");
7721 case _C_C_T_(CNTRL, isCNTRL(value), "Cntrl");
7722 case _C_C_T_(GRAPH, isGRAPH(value), "Graph");
7723 case _C_C_T_(LOWER, isLOWER(value), "Lower");
7724 case _C_C_T_(PRINT, isPRINT(value), "Print");
7725 case _C_C_T_(PSXSPC, isPSXSPC(value), "Space");
7726 case _C_C_T_(PUNCT, isPUNCT(value), "Punct");
7727 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
7728 case _C_C_T_(UPPER, isUPPER(value), "Upper");
7729 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
7730 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
7731 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
7734 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
7737 for (value = 0; value < 128; value++)
7738 ANYOF_BITMAP_SET(ret, value);
7740 for (value = 0; value < 256; value++) {
7742 ANYOF_BITMAP_SET(ret, value);
7751 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
7754 for (value = 128; value < 256; value++)
7755 ANYOF_BITMAP_SET(ret, value);
7757 for (value = 0; value < 256; value++) {
7758 if (!isASCII(value))
7759 ANYOF_BITMAP_SET(ret, value);
7768 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
7770 /* consecutive digits assumed */
7771 for (value = '0'; value <= '9'; value++)
7772 ANYOF_BITMAP_SET(ret, value);
7779 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
7781 /* consecutive digits assumed */
7782 for (value = 0; value < '0'; value++)
7783 ANYOF_BITMAP_SET(ret, value);
7784 for (value = '9' + 1; value < 256; value++)
7785 ANYOF_BITMAP_SET(ret, value);
7791 /* this is to handle \p and \P */
7794 vFAIL("Invalid [::] class");
7798 /* Strings such as "+utf8::isWord\n" */
7799 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
7802 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
7805 } /* end of namedclass \blah */
7808 if (prevvalue > (IV)value) /* b-a */ {
7809 const int w = RExC_parse - rangebegin;
7810 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
7811 range = 0; /* not a valid range */
7815 prevvalue = value; /* save the beginning of the range */
7816 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
7817 RExC_parse[1] != ']') {
7820 /* a bad range like \w-, [:word:]- ? */
7821 if (namedclass > OOB_NAMEDCLASS) {
7822 if (ckWARN(WARN_REGEXP)) {
7824 RExC_parse >= rangebegin ?
7825 RExC_parse - rangebegin : 0;
7827 "False [] range \"%*.*s\"",
7831 ANYOF_BITMAP_SET(ret, '-');
7833 range = 1; /* yeah, it's a range! */
7834 continue; /* but do it the next time */
7838 /* now is the next time */
7839 /*stored += (value - prevvalue + 1);*/
7841 if (prevvalue < 256) {
7842 const IV ceilvalue = value < 256 ? value : 255;
7845 /* In EBCDIC [\x89-\x91] should include
7846 * the \x8e but [i-j] should not. */
7847 if (literal_endpoint == 2 &&
7848 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
7849 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
7851 if (isLOWER(prevvalue)) {
7852 for (i = prevvalue; i <= ceilvalue; i++)
7854 ANYOF_BITMAP_SET(ret, i);
7856 for (i = prevvalue; i <= ceilvalue; i++)
7858 ANYOF_BITMAP_SET(ret, i);
7863 for (i = prevvalue; i <= ceilvalue; i++) {
7864 if (!ANYOF_BITMAP_TEST(ret,i)) {
7866 ANYOF_BITMAP_SET(ret, i);
7870 if (value > 255 || UTF) {
7871 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
7872 const UV natvalue = NATIVE_TO_UNI(value);
7873 stored+=2; /* can't optimize this class */
7874 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
7875 if (prevnatvalue < natvalue) { /* what about > ? */
7876 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
7877 prevnatvalue, natvalue);
7879 else if (prevnatvalue == natvalue) {
7880 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
7882 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
7884 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
7886 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
7887 if (RExC_precomp[0] == ':' &&
7888 RExC_precomp[1] == '[' &&
7889 (f == 0xDF || f == 0x92)) {
7890 f = NATIVE_TO_UNI(f);
7893 /* If folding and foldable and a single
7894 * character, insert also the folded version
7895 * to the charclass. */
7897 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
7898 if ((RExC_precomp[0] == ':' &&
7899 RExC_precomp[1] == '[' &&
7901 (value == 0xFB05 || value == 0xFB06))) ?
7902 foldlen == ((STRLEN)UNISKIP(f) - 1) :
7903 foldlen == (STRLEN)UNISKIP(f) )
7905 if (foldlen == (STRLEN)UNISKIP(f))
7907 Perl_sv_catpvf(aTHX_ listsv,
7910 /* Any multicharacter foldings
7911 * require the following transform:
7912 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
7913 * where E folds into "pq" and F folds
7914 * into "rst", all other characters
7915 * fold to single characters. We save
7916 * away these multicharacter foldings,
7917 * to be later saved as part of the
7918 * additional "s" data. */
7921 if (!unicode_alternate)
7922 unicode_alternate = newAV();
7923 sv = newSVpvn((char*)foldbuf, foldlen);
7925 av_push(unicode_alternate, sv);
7929 /* If folding and the value is one of the Greek
7930 * sigmas insert a few more sigmas to make the
7931 * folding rules of the sigmas to work right.
7932 * Note that not all the possible combinations
7933 * are handled here: some of them are handled
7934 * by the standard folding rules, and some of
7935 * them (literal or EXACTF cases) are handled
7936 * during runtime in regexec.c:S_find_byclass(). */
7937 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
7938 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7939 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
7940 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7941 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7943 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
7944 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
7945 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
7950 literal_endpoint = 0;
7954 range = 0; /* this range (if it was one) is done now */
7958 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
7960 RExC_size += ANYOF_CLASS_ADD_SKIP;
7962 RExC_emit += ANYOF_CLASS_ADD_SKIP;
7968 /****** !SIZE_ONLY AFTER HERE *********/
7970 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
7971 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
7973 /* optimize single char class to an EXACT node
7974 but *only* when its not a UTF/high char */
7975 const char * cur_parse= RExC_parse;
7976 RExC_emit = (regnode *)orig_emit;
7977 RExC_parse = (char *)orig_parse;
7978 ret = reg_node(pRExC_state,
7979 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
7980 RExC_parse = (char *)cur_parse;
7981 *STRING(ret)= (char)value;
7983 RExC_emit += STR_SZ(1);
7986 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
7987 if ( /* If the only flag is folding (plus possibly inversion). */
7988 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
7990 for (value = 0; value < 256; ++value) {
7991 if (ANYOF_BITMAP_TEST(ret, value)) {
7992 UV fold = PL_fold[value];
7995 ANYOF_BITMAP_SET(ret, fold);
7998 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8001 /* optimize inverted simple patterns (e.g. [^a-z]) */
8002 if (optimize_invert &&
8003 /* If the only flag is inversion. */
8004 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8005 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8006 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8007 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8010 AV * const av = newAV();
8012 /* The 0th element stores the character class description
8013 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8014 * to initialize the appropriate swash (which gets stored in
8015 * the 1st element), and also useful for dumping the regnode.
8016 * The 2nd element stores the multicharacter foldings,
8017 * used later (regexec.c:S_reginclass()). */
8018 av_store(av, 0, listsv);
8019 av_store(av, 1, NULL);
8020 av_store(av, 2, (SV*)unicode_alternate);
8021 rv = newRV_noinc((SV*)av);
8022 n = add_data(pRExC_state, 1, "s");
8023 RExC_rxi->data->data[n] = (void*)rv;
8031 /* reg_skipcomment()
8033 Absorbs an /x style # comments from the input stream.
8034 Returns true if there is more text remaining in the stream.
8035 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8036 terminates the pattern without including a newline.
8038 Note its the callers responsibility to ensure that we are
8044 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8047 while (RExC_parse < RExC_end)
8048 if (*RExC_parse++ == '\n') {
8053 /* we ran off the end of the pattern without ending
8054 the comment, so we have to add an \n when wrapping */
8055 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8063 Advance that parse position, and optionall absorbs
8064 "whitespace" from the inputstream.
8066 Without /x "whitespace" means (?#...) style comments only,
8067 with /x this means (?#...) and # comments and whitespace proper.
8069 Returns the RExC_parse point from BEFORE the scan occurs.
8071 This is the /x friendly way of saying RExC_parse++.
8075 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8077 char* const retval = RExC_parse++;
8080 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8081 RExC_parse[2] == '#') {
8082 while (*RExC_parse != ')') {
8083 if (RExC_parse == RExC_end)
8084 FAIL("Sequence (?#... not terminated");
8090 if (RExC_flags & RXf_PMf_EXTENDED) {
8091 if (isSPACE(*RExC_parse)) {
8095 else if (*RExC_parse == '#') {
8096 if ( reg_skipcomment( pRExC_state ) )
8105 - reg_node - emit a node
8107 STATIC regnode * /* Location. */
8108 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8111 register regnode *ptr;
8112 regnode * const ret = RExC_emit;
8113 GET_RE_DEBUG_FLAGS_DECL;
8116 SIZE_ALIGN(RExC_size);
8120 if (RExC_emit >= RExC_emit_bound)
8121 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8123 NODE_ALIGN_FILL(ret);
8125 FILL_ADVANCE_NODE(ptr, op);
8126 #ifdef RE_TRACK_PATTERN_OFFSETS
8127 if (RExC_offsets) { /* MJD */
8128 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8129 "reg_node", __LINE__,
8131 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8132 ? "Overwriting end of array!\n" : "OK",
8133 (UV)(RExC_emit - RExC_emit_start),
8134 (UV)(RExC_parse - RExC_start),
8135 (UV)RExC_offsets[0]));
8136 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8144 - reganode - emit a node with an argument
8146 STATIC regnode * /* Location. */
8147 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8150 register regnode *ptr;
8151 regnode * const ret = RExC_emit;
8152 GET_RE_DEBUG_FLAGS_DECL;
8155 SIZE_ALIGN(RExC_size);
8160 assert(2==regarglen[op]+1);
8162 Anything larger than this has to allocate the extra amount.
8163 If we changed this to be:
8165 RExC_size += (1 + regarglen[op]);
8167 then it wouldn't matter. Its not clear what side effect
8168 might come from that so its not done so far.
8173 if (RExC_emit >= RExC_emit_bound)
8174 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8176 NODE_ALIGN_FILL(ret);
8178 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8179 #ifdef RE_TRACK_PATTERN_OFFSETS
8180 if (RExC_offsets) { /* MJD */
8181 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8185 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8186 "Overwriting end of array!\n" : "OK",
8187 (UV)(RExC_emit - RExC_emit_start),
8188 (UV)(RExC_parse - RExC_start),
8189 (UV)RExC_offsets[0]));
8190 Set_Cur_Node_Offset;
8198 - reguni - emit (if appropriate) a Unicode character
8201 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8204 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8208 - reginsert - insert an operator in front of already-emitted operand
8210 * Means relocating the operand.
8213 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8216 register regnode *src;
8217 register regnode *dst;
8218 register regnode *place;
8219 const int offset = regarglen[(U8)op];
8220 const int size = NODE_STEP_REGNODE + offset;
8221 GET_RE_DEBUG_FLAGS_DECL;
8222 PERL_UNUSED_ARG(depth);
8223 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8224 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8233 if (RExC_open_parens) {
8235 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8236 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8237 if ( RExC_open_parens[paren] >= opnd ) {
8238 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8239 RExC_open_parens[paren] += size;
8241 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8243 if ( RExC_close_parens[paren] >= opnd ) {
8244 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8245 RExC_close_parens[paren] += size;
8247 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8252 while (src > opnd) {
8253 StructCopy(--src, --dst, regnode);
8254 #ifdef RE_TRACK_PATTERN_OFFSETS
8255 if (RExC_offsets) { /* MJD 20010112 */
8256 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8260 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8261 ? "Overwriting end of array!\n" : "OK",
8262 (UV)(src - RExC_emit_start),
8263 (UV)(dst - RExC_emit_start),
8264 (UV)RExC_offsets[0]));
8265 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8266 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8272 place = opnd; /* Op node, where operand used to be. */
8273 #ifdef RE_TRACK_PATTERN_OFFSETS
8274 if (RExC_offsets) { /* MJD */
8275 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8279 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8280 ? "Overwriting end of array!\n" : "OK",
8281 (UV)(place - RExC_emit_start),
8282 (UV)(RExC_parse - RExC_start),
8283 (UV)RExC_offsets[0]));
8284 Set_Node_Offset(place, RExC_parse);
8285 Set_Node_Length(place, 1);
8288 src = NEXTOPER(place);
8289 FILL_ADVANCE_NODE(place, op);
8290 Zero(src, offset, regnode);
8294 - regtail - set the next-pointer at the end of a node chain of p to val.
8295 - SEE ALSO: regtail_study
8297 /* TODO: All three parms should be const */
8299 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8302 register regnode *scan;
8303 GET_RE_DEBUG_FLAGS_DECL;
8305 PERL_UNUSED_ARG(depth);
8311 /* Find last node. */
8314 regnode * const temp = regnext(scan);
8316 SV * const mysv=sv_newmortal();
8317 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8318 regprop(RExC_rx, mysv, scan);
8319 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8320 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8321 (temp == NULL ? "->" : ""),
8322 (temp == NULL ? PL_reg_name[OP(val)] : "")
8330 if (reg_off_by_arg[OP(scan)]) {
8331 ARG_SET(scan, val - scan);
8334 NEXT_OFF(scan) = val - scan;
8340 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8341 - Look for optimizable sequences at the same time.
8342 - currently only looks for EXACT chains.
8344 This is expermental code. The idea is to use this routine to perform
8345 in place optimizations on branches and groups as they are constructed,
8346 with the long term intention of removing optimization from study_chunk so
8347 that it is purely analytical.
8349 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8350 to control which is which.
8353 /* TODO: All four parms should be const */
8356 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8359 register regnode *scan;
8361 #ifdef EXPERIMENTAL_INPLACESCAN
8365 GET_RE_DEBUG_FLAGS_DECL;
8371 /* Find last node. */
8375 regnode * const temp = regnext(scan);
8376 #ifdef EXPERIMENTAL_INPLACESCAN
8377 if (PL_regkind[OP(scan)] == EXACT)
8378 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8386 if( exact == PSEUDO )
8388 else if ( exact != OP(scan) )
8397 SV * const mysv=sv_newmortal();
8398 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8399 regprop(RExC_rx, mysv, scan);
8400 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8401 SvPV_nolen_const(mysv),
8403 PL_reg_name[exact]);
8410 SV * const mysv_val=sv_newmortal();
8411 DEBUG_PARSE_MSG("");
8412 regprop(RExC_rx, mysv_val, val);
8413 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8414 SvPV_nolen_const(mysv_val),
8415 (IV)REG_NODE_NUM(val),
8419 if (reg_off_by_arg[OP(scan)]) {
8420 ARG_SET(scan, val - scan);
8423 NEXT_OFF(scan) = val - scan;
8431 - regcurly - a little FSA that accepts {\d+,?\d*}
8434 S_regcurly(register const char *s)
8453 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8456 Perl_regdump(pTHX_ const regexp *r)
8460 SV * const sv = sv_newmortal();
8461 SV *dsv= sv_newmortal();
8464 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8466 /* Header fields of interest. */
8467 if (r->anchored_substr) {
8468 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8469 RE_SV_DUMPLEN(r->anchored_substr), 30);
8470 PerlIO_printf(Perl_debug_log,
8471 "anchored %s%s at %"IVdf" ",
8472 s, RE_SV_TAIL(r->anchored_substr),
8473 (IV)r->anchored_offset);
8474 } else if (r->anchored_utf8) {
8475 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8476 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8477 PerlIO_printf(Perl_debug_log,
8478 "anchored utf8 %s%s at %"IVdf" ",
8479 s, RE_SV_TAIL(r->anchored_utf8),
8480 (IV)r->anchored_offset);
8482 if (r->float_substr) {
8483 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8484 RE_SV_DUMPLEN(r->float_substr), 30);
8485 PerlIO_printf(Perl_debug_log,
8486 "floating %s%s at %"IVdf"..%"UVuf" ",
8487 s, RE_SV_TAIL(r->float_substr),
8488 (IV)r->float_min_offset, (UV)r->float_max_offset);
8489 } else if (r->float_utf8) {
8490 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8491 RE_SV_DUMPLEN(r->float_utf8), 30);
8492 PerlIO_printf(Perl_debug_log,
8493 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8494 s, RE_SV_TAIL(r->float_utf8),
8495 (IV)r->float_min_offset, (UV)r->float_max_offset);
8497 if (r->check_substr || r->check_utf8)
8498 PerlIO_printf(Perl_debug_log,
8500 (r->check_substr == r->float_substr
8501 && r->check_utf8 == r->float_utf8
8502 ? "(checking floating" : "(checking anchored"));
8503 if (r->extflags & RXf_NOSCAN)
8504 PerlIO_printf(Perl_debug_log, " noscan");
8505 if (r->extflags & RXf_CHECK_ALL)
8506 PerlIO_printf(Perl_debug_log, " isall");
8507 if (r->check_substr || r->check_utf8)
8508 PerlIO_printf(Perl_debug_log, ") ");
8510 if (ri->regstclass) {
8511 regprop(r, sv, ri->regstclass);
8512 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
8514 if (r->extflags & RXf_ANCH) {
8515 PerlIO_printf(Perl_debug_log, "anchored");
8516 if (r->extflags & RXf_ANCH_BOL)
8517 PerlIO_printf(Perl_debug_log, "(BOL)");
8518 if (r->extflags & RXf_ANCH_MBOL)
8519 PerlIO_printf(Perl_debug_log, "(MBOL)");
8520 if (r->extflags & RXf_ANCH_SBOL)
8521 PerlIO_printf(Perl_debug_log, "(SBOL)");
8522 if (r->extflags & RXf_ANCH_GPOS)
8523 PerlIO_printf(Perl_debug_log, "(GPOS)");
8524 PerlIO_putc(Perl_debug_log, ' ');
8526 if (r->extflags & RXf_GPOS_SEEN)
8527 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
8528 if (r->intflags & PREGf_SKIP)
8529 PerlIO_printf(Perl_debug_log, "plus ");
8530 if (r->intflags & PREGf_IMPLICIT)
8531 PerlIO_printf(Perl_debug_log, "implicit ");
8532 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
8533 if (r->extflags & RXf_EVAL_SEEN)
8534 PerlIO_printf(Perl_debug_log, "with eval ");
8535 PerlIO_printf(Perl_debug_log, "\n");
8537 PERL_UNUSED_CONTEXT;
8539 #endif /* DEBUGGING */
8543 - regprop - printable representation of opcode
8546 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
8551 RXi_GET_DECL(prog,progi);
8552 GET_RE_DEBUG_FLAGS_DECL;
8555 sv_setpvn(sv, "", 0);
8557 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
8558 /* It would be nice to FAIL() here, but this may be called from
8559 regexec.c, and it would be hard to supply pRExC_state. */
8560 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
8561 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
8563 k = PL_regkind[OP(o)];
8566 SV * const dsv = sv_2mortal(newSVpvs(""));
8567 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
8568 * is a crude hack but it may be the best for now since
8569 * we have no flag "this EXACTish node was UTF-8"
8571 const char * const s =
8572 pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
8573 PL_colors[0], PL_colors[1],
8574 PERL_PV_ESCAPE_UNI_DETECT |
8575 PERL_PV_PRETTY_ELIPSES |
8578 Perl_sv_catpvf(aTHX_ sv, " %s", s );
8579 } else if (k == TRIE) {
8580 /* print the details of the trie in dumpuntil instead, as
8581 * progi->data isn't available here */
8582 const char op = OP(o);
8583 const U32 n = ARG(o);
8584 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
8585 (reg_ac_data *)progi->data->data[n] :
8587 const reg_trie_data * const trie
8588 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
8590 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
8591 DEBUG_TRIE_COMPILE_r(
8592 Perl_sv_catpvf(aTHX_ sv,
8593 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
8594 (UV)trie->startstate,
8595 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
8596 (UV)trie->wordcount,
8599 (UV)TRIE_CHARCOUNT(trie),
8600 (UV)trie->uniquecharcount
8603 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
8605 int rangestart = -1;
8606 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
8607 Perl_sv_catpvf(aTHX_ sv, "[");
8608 for (i = 0; i <= 256; i++) {
8609 if (i < 256 && BITMAP_TEST(bitmap,i)) {
8610 if (rangestart == -1)
8612 } else if (rangestart != -1) {
8613 if (i <= rangestart + 3)
8614 for (; rangestart < i; rangestart++)
8615 put_byte(sv, rangestart);
8617 put_byte(sv, rangestart);
8619 put_byte(sv, i - 1);
8624 Perl_sv_catpvf(aTHX_ sv, "]");
8627 } else if (k == CURLY) {
8628 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
8629 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
8630 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
8632 else if (k == WHILEM && o->flags) /* Ordinal/of */
8633 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
8634 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
8635 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
8636 if ( prog->paren_names ) {
8637 if ( k != REF || OP(o) < NREF) {
8638 AV *list= (AV *)progi->data->data[progi->name_list_idx];
8639 SV **name= av_fetch(list, ARG(o), 0 );
8641 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8644 AV *list= (AV *)progi->data->data[ progi->name_list_idx ];
8645 SV *sv_dat=(SV*)progi->data->data[ ARG( o ) ];
8646 I32 *nums=(I32*)SvPVX(sv_dat);
8647 SV **name= av_fetch(list, nums[0], 0 );
8650 for ( n=0; n<SvIVX(sv_dat); n++ ) {
8651 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
8652 (n ? "," : ""), (IV)nums[n]);
8654 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
8658 } else if (k == GOSUB)
8659 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
8660 else if (k == VERB) {
8662 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
8663 SVfARG((SV*)progi->data->data[ ARG( o ) ]));
8664 } else if (k == LOGICAL)
8665 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
8666 else if (k == FOLDCHAR)
8667 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]",ARG(o) );
8668 else if (k == ANYOF) {
8669 int i, rangestart = -1;
8670 const U8 flags = ANYOF_FLAGS(o);
8672 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
8673 static const char * const anyofs[] = {
8706 if (flags & ANYOF_LOCALE)
8707 sv_catpvs(sv, "{loc}");
8708 if (flags & ANYOF_FOLD)
8709 sv_catpvs(sv, "{i}");
8710 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
8711 if (flags & ANYOF_INVERT)
8713 for (i = 0; i <= 256; i++) {
8714 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
8715 if (rangestart == -1)
8717 } else if (rangestart != -1) {
8718 if (i <= rangestart + 3)
8719 for (; rangestart < i; rangestart++)
8720 put_byte(sv, rangestart);
8722 put_byte(sv, rangestart);
8724 put_byte(sv, i - 1);
8730 if (o->flags & ANYOF_CLASS)
8731 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
8732 if (ANYOF_CLASS_TEST(o,i))
8733 sv_catpv(sv, anyofs[i]);
8735 if (flags & ANYOF_UNICODE)
8736 sv_catpvs(sv, "{unicode}");
8737 else if (flags & ANYOF_UNICODE_ALL)
8738 sv_catpvs(sv, "{unicode_all}");
8742 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
8746 U8 s[UTF8_MAXBYTES_CASE+1];
8748 for (i = 0; i <= 256; i++) { /* just the first 256 */
8749 uvchr_to_utf8(s, i);
8751 if (i < 256 && swash_fetch(sw, s, TRUE)) {
8752 if (rangestart == -1)
8754 } else if (rangestart != -1) {
8755 if (i <= rangestart + 3)
8756 for (; rangestart < i; rangestart++) {
8757 const U8 * const e = uvchr_to_utf8(s,rangestart);
8759 for(p = s; p < e; p++)
8763 const U8 *e = uvchr_to_utf8(s,rangestart);
8765 for (p = s; p < e; p++)
8768 e = uvchr_to_utf8(s, i-1);
8769 for (p = s; p < e; p++)
8776 sv_catpvs(sv, "..."); /* et cetera */
8780 char *s = savesvpv(lv);
8781 char * const origs = s;
8783 while (*s && *s != '\n')
8787 const char * const t = ++s;
8805 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
8807 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
8808 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
8810 PERL_UNUSED_CONTEXT;
8811 PERL_UNUSED_ARG(sv);
8813 PERL_UNUSED_ARG(prog);
8814 #endif /* DEBUGGING */
8818 Perl_re_intuit_string(pTHX_ REGEXP * const prog)
8819 { /* Assume that RE_INTUIT is set */
8821 GET_RE_DEBUG_FLAGS_DECL;
8822 PERL_UNUSED_CONTEXT;
8826 const char * const s = SvPV_nolen_const(prog->check_substr
8827 ? prog->check_substr : prog->check_utf8);
8829 if (!PL_colorset) reginitcolors();
8830 PerlIO_printf(Perl_debug_log,
8831 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
8833 prog->check_substr ? "" : "utf8 ",
8834 PL_colors[5],PL_colors[0],
8837 (strlen(s) > 60 ? "..." : ""));
8840 return prog->check_substr ? prog->check_substr : prog->check_utf8;
8846 handles refcounting and freeing the perl core regexp structure. When
8847 it is necessary to actually free the structure the first thing it
8848 does is call the 'free' method of the regexp_engine associated to to
8849 the regexp, allowing the handling of the void *pprivate; member
8850 first. (This routine is not overridable by extensions, which is why
8851 the extensions free is called first.)
8853 See regdupe and regdupe_internal if you change anything here.
8855 #ifndef PERL_IN_XSUB_RE
8857 Perl_pregfree(pTHX_ struct regexp *r)
8860 GET_RE_DEBUG_FLAGS_DECL;
8862 if (!r || (--r->refcnt > 0))
8865 ReREFCNT_dec(r->mother_re);
8867 CALLREGFREE_PVT(r); /* free the private data */
8869 SvREFCNT_dec(r->paren_names);
8870 Safefree(r->wrapped);
8873 if (r->anchored_substr)
8874 SvREFCNT_dec(r->anchored_substr);
8875 if (r->anchored_utf8)
8876 SvREFCNT_dec(r->anchored_utf8);
8877 if (r->float_substr)
8878 SvREFCNT_dec(r->float_substr);
8880 SvREFCNT_dec(r->float_utf8);
8881 Safefree(r->substrs);
8883 RX_MATCH_COPY_FREE(r);
8884 #ifdef PERL_OLD_COPY_ON_WRITE
8886 SvREFCNT_dec(r->saved_copy);
8895 This is a hacky workaround to the structural issue of match results
8896 being stored in the regexp structure which is in turn stored in
8897 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
8898 could be PL_curpm in multiple contexts, and could require multiple
8899 result sets being associated with the pattern simultaneously, such
8900 as when doing a recursive match with (??{$qr})
8902 The solution is to make a lightweight copy of the regexp structure
8903 when a qr// is returned from the code executed by (??{$qr}) this
8904 lightweight copy doesnt actually own any of its data except for
8905 the starp/end and the actual regexp structure itself.
8911 Perl_reg_temp_copy (pTHX_ struct regexp *r) {
8913 register const I32 npar = r->nparens+1;
8914 (void)ReREFCNT_inc(r);
8915 Newx(ret, 1, regexp);
8916 StructCopy(r, ret, regexp);
8917 Newx(ret->offs, npar, regexp_paren_pair);
8918 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
8921 Newx(ret->substrs, 1, struct reg_substr_data);
8922 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
8924 SvREFCNT_inc_void(ret->anchored_substr);
8925 SvREFCNT_inc_void(ret->anchored_utf8);
8926 SvREFCNT_inc_void(ret->float_substr);
8927 SvREFCNT_inc_void(ret->float_utf8);
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 RX_MATCH_COPIED_off(ret);
8933 #ifdef PERL_OLD_COPY_ON_WRITE
8934 ret->saved_copy = NULL;
8943 /* regfree_internal()
8945 Free the private data in a regexp. This is overloadable by
8946 extensions. Perl takes care of the regexp structure in pregfree(),
8947 this covers the *pprivate pointer which technically perldoesnt
8948 know about, however of course we have to handle the
8949 regexp_internal structure when no extension is in use.
8951 Note this is called before freeing anything in the regexp
8956 Perl_regfree_internal(pTHX_ REGEXP * const r)
8960 GET_RE_DEBUG_FLAGS_DECL;
8966 SV *dsv= sv_newmortal();
8967 RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
8968 dsv, r->precomp, r->prelen, 60);
8969 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
8970 PL_colors[4],PL_colors[5],s);
8973 #ifdef RE_TRACK_PATTERN_OFFSETS
8975 Safefree(ri->u.offsets); /* 20010421 MJD */
8978 int n = ri->data->count;
8979 PAD* new_comppad = NULL;
8984 /* If you add a ->what type here, update the comment in regcomp.h */
8985 switch (ri->data->what[n]) {
8989 SvREFCNT_dec((SV*)ri->data->data[n]);
8992 Safefree(ri->data->data[n]);
8995 new_comppad = (AV*)ri->data->data[n];
8998 if (new_comppad == NULL)
8999 Perl_croak(aTHX_ "panic: pregfree comppad");
9000 PAD_SAVE_LOCAL(old_comppad,
9001 /* Watch out for global destruction's random ordering. */
9002 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9005 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9008 op_free((OP_4tree*)ri->data->data[n]);
9010 PAD_RESTORE_LOCAL(old_comppad);
9011 SvREFCNT_dec((SV*)new_comppad);
9017 { /* Aho Corasick add-on structure for a trie node.
9018 Used in stclass optimization only */
9020 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9022 refcount = --aho->refcount;
9025 PerlMemShared_free(aho->states);
9026 PerlMemShared_free(aho->fail);
9027 /* do this last!!!! */
9028 PerlMemShared_free(ri->data->data[n]);
9029 PerlMemShared_free(ri->regstclass);
9035 /* trie structure. */
9037 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9039 refcount = --trie->refcount;
9042 PerlMemShared_free(trie->charmap);
9043 PerlMemShared_free(trie->states);
9044 PerlMemShared_free(trie->trans);
9046 PerlMemShared_free(trie->bitmap);
9048 PerlMemShared_free(trie->wordlen);
9050 PerlMemShared_free(trie->jump);
9052 PerlMemShared_free(trie->nextword);
9053 /* do this last!!!! */
9054 PerlMemShared_free(ri->data->data[n]);
9059 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9062 Safefree(ri->data->what);
9069 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
9070 #define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9071 #define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
9072 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9075 re_dup - duplicate a regexp.
9077 This routine is expected to clone a given regexp structure. It is not
9078 compiler under USE_ITHREADS.
9080 After all of the core data stored in struct regexp is duplicated
9081 the regexp_engine.dupe method is used to copy any private data
9082 stored in the *pprivate pointer. This allows extensions to handle
9083 any duplication it needs to do.
9085 See pregfree() and regfree_internal() if you change anything here.
9087 #if defined(USE_ITHREADS)
9088 #ifndef PERL_IN_XSUB_RE
9090 Perl_re_dup(pTHX_ const regexp *r, CLONE_PARAMS *param)
9097 return (REGEXP *)NULL;
9099 if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
9103 npar = r->nparens+1;
9104 Newx(ret, 1, regexp);
9105 StructCopy(r, ret, regexp);
9106 Newx(ret->offs, npar, regexp_paren_pair);
9107 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9109 /* no need to copy these */
9110 Newx(ret->swap, npar, regexp_paren_pair);
9114 /* Do it this way to avoid reading from *r after the StructCopy().
9115 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9116 cache, it doesn't matter. */
9117 const bool anchored = r->check_substr == r->anchored_substr;
9118 Newx(ret->substrs, 1, struct reg_substr_data);
9119 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9121 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9122 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9123 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9124 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9126 /* check_substr and check_utf8, if non-NULL, point to either their
9127 anchored or float namesakes, and don't hold a second reference. */
9129 if (ret->check_substr) {
9131 assert(r->check_utf8 == r->anchored_utf8);
9132 ret->check_substr = ret->anchored_substr;
9133 ret->check_utf8 = ret->anchored_utf8;
9135 assert(r->check_substr == r->float_substr);
9136 assert(r->check_utf8 == r->float_utf8);
9137 ret->check_substr = ret->float_substr;
9138 ret->check_utf8 = ret->float_utf8;
9143 ret->wrapped = SAVEPVN(ret->wrapped, ret->wraplen+1);
9144 ret->precomp = ret->wrapped + (ret->precomp - ret->wrapped);
9145 ret->paren_names = hv_dup_inc(ret->paren_names, param);
9148 RXi_SET(ret,CALLREGDUPE_PVT(ret,param));
9150 if (RX_MATCH_COPIED(ret))
9151 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9154 #ifdef PERL_OLD_COPY_ON_WRITE
9155 ret->saved_copy = NULL;
9158 ret->mother_re = NULL;
9160 ret->seen_evals = 0;
9162 ptr_table_store(PL_ptr_table, r, ret);
9165 #endif /* PERL_IN_XSUB_RE */
9170 This is the internal complement to regdupe() which is used to copy
9171 the structure pointed to by the *pprivate pointer in the regexp.
9172 This is the core version of the extension overridable cloning hook.
9173 The regexp structure being duplicated will be copied by perl prior
9174 to this and will be provided as the regexp *r argument, however
9175 with the /old/ structures pprivate pointer value. Thus this routine
9176 may override any copying normally done by perl.
9178 It returns a pointer to the new regexp_internal structure.
9182 Perl_regdupe_internal(pTHX_ REGEXP * const r, CLONE_PARAMS *param)
9185 regexp_internal *reti;
9189 npar = r->nparens+1;
9192 Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
9193 Copy(ri->program, reti->program, len+1, regnode);
9196 reti->regstclass = NULL;
9200 const int count = ri->data->count;
9203 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9204 char, struct reg_data);
9205 Newx(d->what, count, U8);
9208 for (i = 0; i < count; i++) {
9209 d->what[i] = ri->data->what[i];
9210 switch (d->what[i]) {
9211 /* legal options are one of: sSfpontTu
9212 see also regcomp.h and pregfree() */
9215 case 'p': /* actually an AV, but the dup function is identical. */
9216 case 'u': /* actually an HV, but the dup function is identical. */
9217 d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
9220 /* This is cheating. */
9221 Newx(d->data[i], 1, struct regnode_charclass_class);
9222 StructCopy(ri->data->data[i], d->data[i],
9223 struct regnode_charclass_class);
9224 reti->regstclass = (regnode*)d->data[i];
9227 /* Compiled op trees are readonly and in shared memory,
9228 and can thus be shared without duplication. */
9230 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9234 /* Trie stclasses are readonly and can thus be shared
9235 * without duplication. We free the stclass in pregfree
9236 * when the corresponding reg_ac_data struct is freed.
9238 reti->regstclass= ri->regstclass;
9242 ((reg_trie_data*)ri->data->data[i])->refcount++;
9246 d->data[i] = ri->data->data[i];
9249 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9258 reti->name_list_idx = ri->name_list_idx;
9260 #ifdef RE_TRACK_PATTERN_OFFSETS
9261 if (ri->u.offsets) {
9262 Newx(reti->u.offsets, 2*len+1, U32);
9263 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9266 SetProgLen(reti,len);
9272 #endif /* USE_ITHREADS */
9277 converts a regexp embedded in a MAGIC struct to its stringified form,
9278 caching the converted form in the struct and returns the cached
9281 If lp is nonnull then it is used to return the length of the
9284 If flags is nonnull and the returned string contains UTF8 then
9285 (*flags & 1) will be true.
9287 If haseval is nonnull then it is used to return whether the pattern
9290 Normally called via macro:
9292 CALLREG_STRINGIFY(mg,&len,&utf8);
9296 CALLREG_AS_STR(mg,&lp,&flags,&haseval)
9298 See sv_2pv_flags() in sv.c for an example of internal usage.
9301 #ifndef PERL_IN_XSUB_RE
9304 Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
9306 const regexp * const re = (regexp *)mg->mg_obj;
9308 *haseval = re->seen_evals;
9310 *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
9317 - regnext - dig the "next" pointer out of a node
9320 Perl_regnext(pTHX_ register regnode *p)
9323 register I32 offset;
9328 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9337 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9340 STRLEN l1 = strlen(pat1);
9341 STRLEN l2 = strlen(pat2);
9344 const char *message;
9350 Copy(pat1, buf, l1 , char);
9351 Copy(pat2, buf + l1, l2 , char);
9352 buf[l1 + l2] = '\n';
9353 buf[l1 + l2 + 1] = '\0';
9355 /* ANSI variant takes additional second argument */
9356 va_start(args, pat2);
9360 msv = vmess(buf, &args);
9362 message = SvPV_const(msv,l1);
9365 Copy(message, buf, l1 , char);
9366 buf[l1-1] = '\0'; /* Overwrite \n */
9367 Perl_croak(aTHX_ "%s", buf);
9370 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9372 #ifndef PERL_IN_XSUB_RE
9374 Perl_save_re_context(pTHX)
9378 struct re_save_state *state;
9380 SAVEVPTR(PL_curcop);
9381 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9383 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9384 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9385 SSPUSHINT(SAVEt_RE_STATE);
9387 Copy(&PL_reg_state, state, 1, struct re_save_state);
9389 PL_reg_start_tmp = 0;
9390 PL_reg_start_tmpl = 0;
9391 PL_reg_oldsaved = NULL;
9392 PL_reg_oldsavedlen = 0;
9394 PL_reg_leftiter = 0;
9395 PL_reg_poscache = NULL;
9396 PL_reg_poscache_size = 0;
9397 #ifdef PERL_OLD_COPY_ON_WRITE
9401 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9403 const REGEXP * const rx = PM_GETRE(PL_curpm);
9406 for (i = 1; i <= rx->nparens; i++) {
9407 char digits[TYPE_CHARS(long)];
9408 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9409 GV *const *const gvp
9410 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9413 GV * const gv = *gvp;
9414 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9424 clear_re(pTHX_ void *r)
9427 ReREFCNT_dec((regexp *)r);
9433 S_put_byte(pTHX_ SV *sv, int c)
9435 if (isCNTRL(c) || c == 255 || !isPRINT(c))
9436 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9437 else if (c == '-' || c == ']' || c == '\\' || c == '^')
9438 Perl_sv_catpvf(aTHX_ sv, "\\%c", c);
9440 Perl_sv_catpvf(aTHX_ sv, "%c", c);
9444 #define CLEAR_OPTSTART \
9445 if (optstart) STMT_START { \
9446 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9450 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9452 STATIC const regnode *
9453 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9454 const regnode *last, const regnode *plast,
9455 SV* sv, I32 indent, U32 depth)
9458 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9459 register const regnode *next;
9460 const regnode *optstart= NULL;
9463 GET_RE_DEBUG_FLAGS_DECL;
9465 #ifdef DEBUG_DUMPUNTIL
9466 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9467 last ? last-start : 0,plast ? plast-start : 0);
9470 if (plast && plast < last)
9473 while (PL_regkind[op] != END && (!last || node < last)) {
9474 /* While that wasn't END last time... */
9477 if (op == CLOSE || op == WHILEM)
9479 next = regnext((regnode *)node);
9482 if (OP(node) == OPTIMIZED) {
9483 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
9490 regprop(r, sv, node);
9491 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
9492 (int)(2*indent + 1), "", SvPVX_const(sv));
9494 if (OP(node) != OPTIMIZED) {
9495 if (next == NULL) /* Next ptr. */
9496 PerlIO_printf(Perl_debug_log, " (0)");
9497 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
9498 PerlIO_printf(Perl_debug_log, " (FAIL)");
9500 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
9501 (void)PerlIO_putc(Perl_debug_log, '\n');
9505 if (PL_regkind[(U8)op] == BRANCHJ) {
9508 register const regnode *nnode = (OP(next) == LONGJMP
9509 ? regnext((regnode *)next)
9511 if (last && nnode > last)
9513 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
9516 else if (PL_regkind[(U8)op] == BRANCH) {
9518 DUMPUNTIL(NEXTOPER(node), next);
9520 else if ( PL_regkind[(U8)op] == TRIE ) {
9521 const regnode *this_trie = node;
9522 const char op = OP(node);
9523 const U32 n = ARG(node);
9524 const reg_ac_data * const ac = op>=AHOCORASICK ?
9525 (reg_ac_data *)ri->data->data[n] :
9527 const reg_trie_data * const trie =
9528 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
9530 AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
9532 const regnode *nextbranch= NULL;
9534 sv_setpvn(sv, "", 0);
9535 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
9536 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
9538 PerlIO_printf(Perl_debug_log, "%*s%s ",
9539 (int)(2*(indent+3)), "",
9540 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
9541 PL_colors[0], PL_colors[1],
9542 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
9543 PERL_PV_PRETTY_ELIPSES |
9549 U16 dist= trie->jump[word_idx+1];
9550 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
9551 (UV)((dist ? this_trie + dist : next) - start));
9554 nextbranch= this_trie + trie->jump[0];
9555 DUMPUNTIL(this_trie + dist, nextbranch);
9557 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
9558 nextbranch= regnext((regnode *)nextbranch);
9560 PerlIO_printf(Perl_debug_log, "\n");
9563 if (last && next > last)
9568 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
9569 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
9570 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
9572 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
9574 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
9576 else if ( op == PLUS || op == STAR) {
9577 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
9579 else if (op == ANYOF) {
9580 /* arglen 1 + class block */
9581 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
9582 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
9583 node = NEXTOPER(node);
9585 else if (PL_regkind[(U8)op] == EXACT) {
9586 /* Literal string, where present. */
9587 node += NODE_SZ_STR(node) - 1;
9588 node = NEXTOPER(node);
9591 node = NEXTOPER(node);
9592 node += regarglen[(U8)op];
9594 if (op == CURLYX || op == OPEN)
9598 #ifdef DEBUG_DUMPUNTIL
9599 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
9604 #endif /* DEBUGGING */
9608 * c-indentation-style: bsd
9610 * indent-tabs-mode: t
9613 * ex: set ts=8 sts=4 sw=4 noet: