5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
93 # if defined(BUGGY_MSC6)
94 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
95 # pragma optimize("a",off)
96 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
97 # pragma optimize("w",on )
98 # endif /* BUGGY_MSC6 */
102 #define STATIC static
105 typedef struct RExC_state_t {
106 U32 flags; /* are we folding, multilining? */
107 char *precomp; /* uncompiled string. */
108 REGEXP *rx_sv; /* The SV that is the regexp. */
109 regexp *rx; /* perl core regexp structure */
110 regexp_internal *rxi; /* internal data for regexp object pprivate field */
111 char *start; /* Start of input for compile */
112 char *end; /* End of input for compile */
113 char *parse; /* Input-scan pointer. */
114 I32 whilem_seen; /* number of WHILEM in this expr */
115 regnode *emit_start; /* Start of emitted-code area */
116 regnode *emit_bound; /* First regnode outside of the allocated space */
117 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
118 I32 naughty; /* How bad is this pattern? */
119 I32 sawback; /* Did we see \1, ...? */
121 I32 size; /* Code size. */
122 I32 npar; /* Capture buffer count, (OPEN). */
123 I32 cpar; /* Capture buffer count, (CLOSE). */
124 I32 nestroot; /* root parens we are in - used by accept */
128 regnode **open_parens; /* pointers to open parens */
129 regnode **close_parens; /* pointers to close parens */
130 regnode *opend; /* END node in program */
131 I32 utf8; /* whether the pattern is utf8 or not */
132 I32 orig_utf8; /* whether the pattern was originally in utf8 */
133 /* XXX use this for future optimisation of case
134 * where pattern must be upgraded to utf8. */
135 HV *charnames; /* cache of named sequences */
136 HV *paren_names; /* Paren names */
138 regnode **recurse; /* Recurse regops */
139 I32 recurse_count; /* Number of recurse regops */
141 char *starttry; /* -Dr: where regtry was called. */
142 #define RExC_starttry (pRExC_state->starttry)
145 const char *lastparse;
147 AV *paren_name_list; /* idx -> name */
148 #define RExC_lastparse (pRExC_state->lastparse)
149 #define RExC_lastnum (pRExC_state->lastnum)
150 #define RExC_paren_name_list (pRExC_state->paren_name_list)
154 #define RExC_flags (pRExC_state->flags)
155 #define RExC_precomp (pRExC_state->precomp)
156 #define RExC_rx_sv (pRExC_state->rx_sv)
157 #define RExC_rx (pRExC_state->rx)
158 #define RExC_rxi (pRExC_state->rxi)
159 #define RExC_start (pRExC_state->start)
160 #define RExC_end (pRExC_state->end)
161 #define RExC_parse (pRExC_state->parse)
162 #define RExC_whilem_seen (pRExC_state->whilem_seen)
163 #ifdef RE_TRACK_PATTERN_OFFSETS
164 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
166 #define RExC_emit (pRExC_state->emit)
167 #define RExC_emit_start (pRExC_state->emit_start)
168 #define RExC_emit_bound (pRExC_state->emit_bound)
169 #define RExC_naughty (pRExC_state->naughty)
170 #define RExC_sawback (pRExC_state->sawback)
171 #define RExC_seen (pRExC_state->seen)
172 #define RExC_size (pRExC_state->size)
173 #define RExC_npar (pRExC_state->npar)
174 #define RExC_nestroot (pRExC_state->nestroot)
175 #define RExC_extralen (pRExC_state->extralen)
176 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
177 #define RExC_seen_evals (pRExC_state->seen_evals)
178 #define RExC_utf8 (pRExC_state->utf8)
179 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
180 #define RExC_charnames (pRExC_state->charnames)
181 #define RExC_open_parens (pRExC_state->open_parens)
182 #define RExC_close_parens (pRExC_state->close_parens)
183 #define RExC_opend (pRExC_state->opend)
184 #define RExC_paren_names (pRExC_state->paren_names)
185 #define RExC_recurse (pRExC_state->recurse)
186 #define RExC_recurse_count (pRExC_state->recurse_count)
189 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
190 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
191 ((*s) == '{' && regcurly(s)))
194 #undef SPSTART /* dratted cpp namespace... */
197 * Flags to be passed up and down.
199 #define WORST 0 /* Worst case. */
200 #define HASWIDTH 0x01 /* Known to match non-null strings. */
201 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
202 #define SPSTART 0x04 /* Starts with * or +. */
203 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
204 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
206 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
208 /* whether trie related optimizations are enabled */
209 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
210 #define TRIE_STUDY_OPT
211 #define FULL_TRIE_STUDY
217 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
218 #define PBITVAL(paren) (1 << ((paren) & 7))
219 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
220 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
221 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
224 /* About scan_data_t.
226 During optimisation we recurse through the regexp program performing
227 various inplace (keyhole style) optimisations. In addition study_chunk
228 and scan_commit populate this data structure with information about
229 what strings MUST appear in the pattern. We look for the longest
230 string that must appear for at a fixed location, and we look for the
231 longest string that may appear at a floating location. So for instance
236 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
237 strings (because they follow a .* construct). study_chunk will identify
238 both FOO and BAR as being the longest fixed and floating strings respectively.
240 The strings can be composites, for instance
244 will result in a composite fixed substring 'foo'.
246 For each string some basic information is maintained:
248 - offset or min_offset
249 This is the position the string must appear at, or not before.
250 It also implicitly (when combined with minlenp) tells us how many
251 character must match before the string we are searching.
252 Likewise when combined with minlenp and the length of the string
253 tells us how many characters must appear after the string we have
257 Only used for floating strings. This is the rightmost point that
258 the string can appear at. Ifset to I32 max it indicates that the
259 string can occur infinitely far to the right.
262 A pointer to the minimum length of the pattern that the string
263 was found inside. This is important as in the case of positive
264 lookahead or positive lookbehind we can have multiple patterns
269 The minimum length of the pattern overall is 3, the minimum length
270 of the lookahead part is 3, but the minimum length of the part that
271 will actually match is 1. So 'FOO's minimum length is 3, but the
272 minimum length for the F is 1. This is important as the minimum length
273 is used to determine offsets in front of and behind the string being
274 looked for. Since strings can be composites this is the length of the
275 pattern at the time it was commited with a scan_commit. Note that
276 the length is calculated by study_chunk, so that the minimum lengths
277 are not known until the full pattern has been compiled, thus the
278 pointer to the value.
282 In the case of lookbehind the string being searched for can be
283 offset past the start point of the final matching string.
284 If this value was just blithely removed from the min_offset it would
285 invalidate some of the calculations for how many chars must match
286 before or after (as they are derived from min_offset and minlen and
287 the length of the string being searched for).
288 When the final pattern is compiled and the data is moved from the
289 scan_data_t structure into the regexp structure the information
290 about lookbehind is factored in, with the information that would
291 have been lost precalculated in the end_shift field for the
294 The fields pos_min and pos_delta are used to store the minimum offset
295 and the delta to the maximum offset at the current point in the pattern.
299 typedef struct scan_data_t {
300 /*I32 len_min; unused */
301 /*I32 len_delta; unused */
305 I32 last_end; /* min value, <0 unless valid. */
308 SV **longest; /* Either &l_fixed, or &l_float. */
309 SV *longest_fixed; /* longest fixed string found in pattern */
310 I32 offset_fixed; /* offset where it starts */
311 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
312 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
313 SV *longest_float; /* longest floating string found in pattern */
314 I32 offset_float_min; /* earliest point in string it can appear */
315 I32 offset_float_max; /* latest point in string it can appear */
316 I32 *minlen_float; /* pointer to the minlen relevent to the string */
317 I32 lookbehind_float; /* is the position of the string modified by LB */
321 struct regnode_charclass_class *start_class;
325 * Forward declarations for pregcomp()'s friends.
328 static const scan_data_t zero_scan_data =
329 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
331 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
332 #define SF_BEFORE_SEOL 0x0001
333 #define SF_BEFORE_MEOL 0x0002
334 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
335 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
338 # define SF_FIX_SHIFT_EOL (0+2)
339 # define SF_FL_SHIFT_EOL (0+4)
341 # define SF_FIX_SHIFT_EOL (+2)
342 # define SF_FL_SHIFT_EOL (+4)
345 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
346 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
348 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
349 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
350 #define SF_IS_INF 0x0040
351 #define SF_HAS_PAR 0x0080
352 #define SF_IN_PAR 0x0100
353 #define SF_HAS_EVAL 0x0200
354 #define SCF_DO_SUBSTR 0x0400
355 #define SCF_DO_STCLASS_AND 0x0800
356 #define SCF_DO_STCLASS_OR 0x1000
357 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
358 #define SCF_WHILEM_VISITED_POS 0x2000
360 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
361 #define SCF_SEEN_ACCEPT 0x8000
363 #define UTF (RExC_utf8 != 0)
364 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
365 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
367 #define OOB_UNICODE 12345678
368 #define OOB_NAMEDCLASS -1
370 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
371 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
374 /* length of regex to show in messages that don't mark a position within */
375 #define RegexLengthToShowInErrorMessages 127
378 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
379 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
380 * op/pragma/warn/regcomp.
382 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
383 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
385 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
388 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
389 * arg. Show regex, up to a maximum length. If it's too long, chop and add
392 #define _FAIL(code) STMT_START { \
393 const char *ellipses = ""; \
394 IV len = RExC_end - RExC_precomp; \
397 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
398 if (len > RegexLengthToShowInErrorMessages) { \
399 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
400 len = RegexLengthToShowInErrorMessages - 10; \
406 #define FAIL(msg) _FAIL( \
407 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
408 msg, (int)len, RExC_precomp, ellipses))
410 #define FAIL2(msg,arg) _FAIL( \
411 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
412 arg, (int)len, RExC_precomp, ellipses))
415 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
417 #define Simple_vFAIL(m) STMT_START { \
418 const IV offset = RExC_parse - RExC_precomp; \
419 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
420 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
424 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
426 #define vFAIL(m) STMT_START { \
428 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
433 * Like Simple_vFAIL(), but accepts two arguments.
435 #define Simple_vFAIL2(m,a1) STMT_START { \
436 const IV offset = RExC_parse - RExC_precomp; \
437 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
438 (int)offset, RExC_precomp, RExC_precomp + offset); \
442 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
444 #define vFAIL2(m,a1) STMT_START { \
446 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
447 Simple_vFAIL2(m, a1); \
452 * Like Simple_vFAIL(), but accepts three arguments.
454 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
455 const IV offset = RExC_parse - RExC_precomp; \
456 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
457 (int)offset, RExC_precomp, RExC_precomp + offset); \
461 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
463 #define vFAIL3(m,a1,a2) STMT_START { \
465 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
466 Simple_vFAIL3(m, a1, a2); \
470 * Like Simple_vFAIL(), but accepts four arguments.
472 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
473 const IV offset = RExC_parse - RExC_precomp; \
474 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
475 (int)offset, RExC_precomp, RExC_precomp + offset); \
478 #define vWARN(loc,m) STMT_START { \
479 const IV offset = loc - RExC_precomp; \
480 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
481 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
484 #define vWARNdep(loc,m) STMT_START { \
485 const IV offset = loc - RExC_precomp; \
486 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
487 "%s" REPORT_LOCATION, \
488 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
492 #define vWARN2(loc, m, a1) STMT_START { \
493 const IV offset = loc - RExC_precomp; \
494 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
495 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
498 #define vWARN3(loc, m, a1, a2) STMT_START { \
499 const IV offset = loc - RExC_precomp; \
500 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
501 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
504 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
505 const IV offset = loc - RExC_precomp; \
506 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
507 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
510 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
511 const IV offset = loc - RExC_precomp; \
512 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
513 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
517 /* Allow for side effects in s */
518 #define REGC(c,s) STMT_START { \
519 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
522 /* Macros for recording node offsets. 20001227 mjd@plover.com
523 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
524 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
525 * Element 0 holds the number n.
526 * Position is 1 indexed.
528 #ifndef RE_TRACK_PATTERN_OFFSETS
529 #define Set_Node_Offset_To_R(node,byte)
530 #define Set_Node_Offset(node,byte)
531 #define Set_Cur_Node_Offset
532 #define Set_Node_Length_To_R(node,len)
533 #define Set_Node_Length(node,len)
534 #define Set_Node_Cur_Length(node)
535 #define Node_Offset(n)
536 #define Node_Length(n)
537 #define Set_Node_Offset_Length(node,offset,len)
538 #define ProgLen(ri) ri->u.proglen
539 #define SetProgLen(ri,x) ri->u.proglen = x
541 #define ProgLen(ri) ri->u.offsets[0]
542 #define SetProgLen(ri,x) ri->u.offsets[0] = x
543 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
545 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
546 __LINE__, (int)(node), (int)(byte))); \
548 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
550 RExC_offsets[2*(node)-1] = (byte); \
555 #define Set_Node_Offset(node,byte) \
556 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
557 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
559 #define Set_Node_Length_To_R(node,len) STMT_START { \
561 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
562 __LINE__, (int)(node), (int)(len))); \
564 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
566 RExC_offsets[2*(node)] = (len); \
571 #define Set_Node_Length(node,len) \
572 Set_Node_Length_To_R((node)-RExC_emit_start, len)
573 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
574 #define Set_Node_Cur_Length(node) \
575 Set_Node_Length(node, RExC_parse - parse_start)
577 /* Get offsets and lengths */
578 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
579 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
581 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
582 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
583 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
587 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
588 #define EXPERIMENTAL_INPLACESCAN
589 #endif /*RE_TRACK_PATTERN_OFFSETS*/
591 #define DEBUG_STUDYDATA(str,data,depth) \
592 DEBUG_OPTIMISE_MORE_r(if(data){ \
593 PerlIO_printf(Perl_debug_log, \
594 "%*s" str "Pos:%"IVdf"/%"IVdf \
595 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
596 (int)(depth)*2, "", \
597 (IV)((data)->pos_min), \
598 (IV)((data)->pos_delta), \
599 (UV)((data)->flags), \
600 (IV)((data)->whilem_c), \
601 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
602 is_inf ? "INF " : "" \
604 if ((data)->last_found) \
605 PerlIO_printf(Perl_debug_log, \
606 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
607 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
608 SvPVX_const((data)->last_found), \
609 (IV)((data)->last_end), \
610 (IV)((data)->last_start_min), \
611 (IV)((data)->last_start_max), \
612 ((data)->longest && \
613 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
614 SvPVX_const((data)->longest_fixed), \
615 (IV)((data)->offset_fixed), \
616 ((data)->longest && \
617 (data)->longest==&((data)->longest_float)) ? "*" : "", \
618 SvPVX_const((data)->longest_float), \
619 (IV)((data)->offset_float_min), \
620 (IV)((data)->offset_float_max) \
622 PerlIO_printf(Perl_debug_log,"\n"); \
625 static void clear_re(pTHX_ void *r);
627 /* Mark that we cannot extend a found fixed substring at this point.
628 Update the longest found anchored substring and the longest found
629 floating substrings if needed. */
632 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
634 const STRLEN l = CHR_SVLEN(data->last_found);
635 const STRLEN old_l = CHR_SVLEN(*data->longest);
636 GET_RE_DEBUG_FLAGS_DECL;
638 PERL_ARGS_ASSERT_SCAN_COMMIT;
640 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
641 SvSetMagicSV(*data->longest, data->last_found);
642 if (*data->longest == data->longest_fixed) {
643 data->offset_fixed = l ? data->last_start_min : data->pos_min;
644 if (data->flags & SF_BEFORE_EOL)
646 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
648 data->flags &= ~SF_FIX_BEFORE_EOL;
649 data->minlen_fixed=minlenp;
650 data->lookbehind_fixed=0;
652 else { /* *data->longest == data->longest_float */
653 data->offset_float_min = l ? data->last_start_min : data->pos_min;
654 data->offset_float_max = (l
655 ? data->last_start_max
656 : data->pos_min + data->pos_delta);
657 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
658 data->offset_float_max = I32_MAX;
659 if (data->flags & SF_BEFORE_EOL)
661 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
663 data->flags &= ~SF_FL_BEFORE_EOL;
664 data->minlen_float=minlenp;
665 data->lookbehind_float=0;
668 SvCUR_set(data->last_found, 0);
670 SV * const sv = data->last_found;
671 if (SvUTF8(sv) && SvMAGICAL(sv)) {
672 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
678 data->flags &= ~SF_BEFORE_EOL;
679 DEBUG_STUDYDATA("commit: ",data,0);
682 /* Can match anything (initialization) */
684 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
686 PERL_ARGS_ASSERT_CL_ANYTHING;
688 ANYOF_CLASS_ZERO(cl);
689 ANYOF_BITMAP_SETALL(cl);
690 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
692 cl->flags |= ANYOF_LOCALE;
695 /* Can match anything (initialization) */
697 S_cl_is_anything(const struct regnode_charclass_class *cl)
701 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
703 for (value = 0; value <= ANYOF_MAX; value += 2)
704 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
706 if (!(cl->flags & ANYOF_UNICODE_ALL))
708 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
713 /* Can match anything (initialization) */
715 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
717 PERL_ARGS_ASSERT_CL_INIT;
719 Zero(cl, 1, struct regnode_charclass_class);
721 cl_anything(pRExC_state, cl);
725 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
727 PERL_ARGS_ASSERT_CL_INIT_ZERO;
729 Zero(cl, 1, struct regnode_charclass_class);
731 cl_anything(pRExC_state, cl);
733 cl->flags |= ANYOF_LOCALE;
736 /* 'And' a given class with another one. Can create false positives */
737 /* We assume that cl is not inverted */
739 S_cl_and(struct regnode_charclass_class *cl,
740 const struct regnode_charclass_class *and_with)
742 PERL_ARGS_ASSERT_CL_AND;
744 assert(and_with->type == ANYOF);
745 if (!(and_with->flags & ANYOF_CLASS)
746 && !(cl->flags & ANYOF_CLASS)
747 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
748 && !(and_with->flags & ANYOF_FOLD)
749 && !(cl->flags & ANYOF_FOLD)) {
752 if (and_with->flags & ANYOF_INVERT)
753 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
754 cl->bitmap[i] &= ~and_with->bitmap[i];
756 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
757 cl->bitmap[i] &= and_with->bitmap[i];
758 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
759 if (!(and_with->flags & ANYOF_EOS))
760 cl->flags &= ~ANYOF_EOS;
762 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
763 !(and_with->flags & ANYOF_INVERT)) {
764 cl->flags &= ~ANYOF_UNICODE_ALL;
765 cl->flags |= ANYOF_UNICODE;
766 ARG_SET(cl, ARG(and_with));
768 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
769 !(and_with->flags & ANYOF_INVERT))
770 cl->flags &= ~ANYOF_UNICODE_ALL;
771 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
772 !(and_with->flags & ANYOF_INVERT))
773 cl->flags &= ~ANYOF_UNICODE;
776 /* 'OR' a given class with another one. Can create false positives */
777 /* We assume that cl is not inverted */
779 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
781 PERL_ARGS_ASSERT_CL_OR;
783 if (or_with->flags & ANYOF_INVERT) {
785 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
786 * <= (B1 | !B2) | (CL1 | !CL2)
787 * which is wasteful if CL2 is small, but we ignore CL2:
788 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
789 * XXXX Can we handle case-fold? Unclear:
790 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
791 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
793 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
794 && !(or_with->flags & ANYOF_FOLD)
795 && !(cl->flags & ANYOF_FOLD) ) {
798 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
799 cl->bitmap[i] |= ~or_with->bitmap[i];
800 } /* XXXX: logic is complicated otherwise */
802 cl_anything(pRExC_state, cl);
805 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
806 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
807 && (!(or_with->flags & ANYOF_FOLD)
808 || (cl->flags & ANYOF_FOLD)) ) {
811 /* OR char bitmap and class bitmap separately */
812 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
813 cl->bitmap[i] |= or_with->bitmap[i];
814 if (or_with->flags & ANYOF_CLASS) {
815 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
816 cl->classflags[i] |= or_with->classflags[i];
817 cl->flags |= ANYOF_CLASS;
820 else { /* XXXX: logic is complicated, leave it along for a moment. */
821 cl_anything(pRExC_state, cl);
824 if (or_with->flags & ANYOF_EOS)
825 cl->flags |= ANYOF_EOS;
827 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
828 ARG(cl) != ARG(or_with)) {
829 cl->flags |= ANYOF_UNICODE_ALL;
830 cl->flags &= ~ANYOF_UNICODE;
832 if (or_with->flags & ANYOF_UNICODE_ALL) {
833 cl->flags |= ANYOF_UNICODE_ALL;
834 cl->flags &= ~ANYOF_UNICODE;
838 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
839 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
840 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
841 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
846 dump_trie(trie,widecharmap,revcharmap)
847 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
848 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
850 These routines dump out a trie in a somewhat readable format.
851 The _interim_ variants are used for debugging the interim
852 tables that are used to generate the final compressed
853 representation which is what dump_trie expects.
855 Part of the reason for their existance is to provide a form
856 of documentation as to how the different representations function.
861 Dumps the final compressed table form of the trie to Perl_debug_log.
862 Used for debugging make_trie().
866 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
867 AV *revcharmap, U32 depth)
870 SV *sv=sv_newmortal();
871 int colwidth= widecharmap ? 6 : 4;
872 GET_RE_DEBUG_FLAGS_DECL;
874 PERL_ARGS_ASSERT_DUMP_TRIE;
876 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
877 (int)depth * 2 + 2,"",
878 "Match","Base","Ofs" );
880 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
881 SV ** const tmp = av_fetch( revcharmap, state, 0);
883 PerlIO_printf( Perl_debug_log, "%*s",
885 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
886 PL_colors[0], PL_colors[1],
887 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
888 PERL_PV_ESCAPE_FIRSTCHAR
893 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
894 (int)depth * 2 + 2,"");
896 for( state = 0 ; state < trie->uniquecharcount ; state++ )
897 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
898 PerlIO_printf( Perl_debug_log, "\n");
900 for( state = 1 ; state < trie->statecount ; state++ ) {
901 const U32 base = trie->states[ state ].trans.base;
903 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
905 if ( trie->states[ state ].wordnum ) {
906 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
908 PerlIO_printf( Perl_debug_log, "%6s", "" );
911 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
916 while( ( base + ofs < trie->uniquecharcount ) ||
917 ( base + ofs - trie->uniquecharcount < trie->lasttrans
918 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
921 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
923 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
924 if ( ( base + ofs >= trie->uniquecharcount ) &&
925 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
926 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
928 PerlIO_printf( Perl_debug_log, "%*"UVXf,
930 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
932 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
936 PerlIO_printf( Perl_debug_log, "]");
939 PerlIO_printf( Perl_debug_log, "\n" );
943 Dumps a fully constructed but uncompressed trie in list form.
944 List tries normally only are used for construction when the number of
945 possible chars (trie->uniquecharcount) is very high.
946 Used for debugging make_trie().
949 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
950 HV *widecharmap, AV *revcharmap, U32 next_alloc,
954 SV *sv=sv_newmortal();
955 int colwidth= widecharmap ? 6 : 4;
956 GET_RE_DEBUG_FLAGS_DECL;
958 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
960 /* print out the table precompression. */
961 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
962 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
963 "------:-----+-----------------\n" );
965 for( state=1 ; state < next_alloc ; state ++ ) {
968 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
969 (int)depth * 2 + 2,"", (UV)state );
970 if ( ! trie->states[ state ].wordnum ) {
971 PerlIO_printf( Perl_debug_log, "%5s| ","");
973 PerlIO_printf( Perl_debug_log, "W%4x| ",
974 trie->states[ state ].wordnum
977 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
978 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
980 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
982 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
983 PL_colors[0], PL_colors[1],
984 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
985 PERL_PV_ESCAPE_FIRSTCHAR
987 TRIE_LIST_ITEM(state,charid).forid,
988 (UV)TRIE_LIST_ITEM(state,charid).newstate
991 PerlIO_printf(Perl_debug_log, "\n%*s| ",
992 (int)((depth * 2) + 14), "");
995 PerlIO_printf( Perl_debug_log, "\n");
1000 Dumps a fully constructed but uncompressed trie in table form.
1001 This is the normal DFA style state transition table, with a few
1002 twists to facilitate compression later.
1003 Used for debugging make_trie().
1006 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1007 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1012 SV *sv=sv_newmortal();
1013 int colwidth= widecharmap ? 6 : 4;
1014 GET_RE_DEBUG_FLAGS_DECL;
1016 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1019 print out the table precompression so that we can do a visual check
1020 that they are identical.
1023 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1025 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1026 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1028 PerlIO_printf( Perl_debug_log, "%*s",
1030 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1031 PL_colors[0], PL_colors[1],
1032 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1033 PERL_PV_ESCAPE_FIRSTCHAR
1039 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1041 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1042 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1045 PerlIO_printf( Perl_debug_log, "\n" );
1047 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1049 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1050 (int)depth * 2 + 2,"",
1051 (UV)TRIE_NODENUM( state ) );
1053 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1054 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1056 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1058 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1060 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1061 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1063 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1064 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1071 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1072 startbranch: the first branch in the whole branch sequence
1073 first : start branch of sequence of branch-exact nodes.
1074 May be the same as startbranch
1075 last : Thing following the last branch.
1076 May be the same as tail.
1077 tail : item following the branch sequence
1078 count : words in the sequence
1079 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1080 depth : indent depth
1082 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1084 A trie is an N'ary tree where the branches are determined by digital
1085 decomposition of the key. IE, at the root node you look up the 1st character and
1086 follow that branch repeat until you find the end of the branches. Nodes can be
1087 marked as "accepting" meaning they represent a complete word. Eg:
1091 would convert into the following structure. Numbers represent states, letters
1092 following numbers represent valid transitions on the letter from that state, if
1093 the number is in square brackets it represents an accepting state, otherwise it
1094 will be in parenthesis.
1096 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1100 (1) +-i->(6)-+-s->[7]
1102 +-s->(3)-+-h->(4)-+-e->[5]
1104 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1106 This shows that when matching against the string 'hers' we will begin at state 1
1107 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1108 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1109 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1110 single traverse. We store a mapping from accepting to state to which word was
1111 matched, and then when we have multiple possibilities we try to complete the
1112 rest of the regex in the order in which they occured in the alternation.
1114 The only prior NFA like behaviour that would be changed by the TRIE support is
1115 the silent ignoring of duplicate alternations which are of the form:
1117 / (DUPE|DUPE) X? (?{ ... }) Y /x
1119 Thus EVAL blocks follwing a trie may be called a different number of times with
1120 and without the optimisation. With the optimisations dupes will be silently
1121 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1122 the following demonstrates:
1124 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1126 which prints out 'word' three times, but
1128 'words'=~/(word|word|word)(?{ print $1 })S/
1130 which doesnt print it out at all. This is due to other optimisations kicking in.
1132 Example of what happens on a structural level:
1134 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1136 1: CURLYM[1] {1,32767}(18)
1147 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1148 and should turn into:
1150 1: CURLYM[1] {1,32767}(18)
1152 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1160 Cases where tail != last would be like /(?foo|bar)baz/:
1170 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1171 and would end up looking like:
1174 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1181 d = uvuni_to_utf8_flags(d, uv, 0);
1183 is the recommended Unicode-aware way of saying
1188 #define TRIE_STORE_REVCHAR \
1191 SV *zlopp = newSV(2); \
1192 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1193 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1194 SvCUR_set(zlopp, kapow - flrbbbbb); \
1197 av_push(revcharmap, zlopp); \
1199 char ooooff = (char)uvc; \
1200 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1204 #define TRIE_READ_CHAR STMT_START { \
1208 if ( foldlen > 0 ) { \
1209 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1214 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1215 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1216 foldlen -= UNISKIP( uvc ); \
1217 scan = foldbuf + UNISKIP( uvc ); \
1220 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1230 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1231 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1232 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1233 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1235 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1236 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1237 TRIE_LIST_CUR( state )++; \
1240 #define TRIE_LIST_NEW(state) STMT_START { \
1241 Newxz( trie->states[ state ].trans.list, \
1242 4, reg_trie_trans_le ); \
1243 TRIE_LIST_CUR( state ) = 1; \
1244 TRIE_LIST_LEN( state ) = 4; \
1247 #define TRIE_HANDLE_WORD(state) STMT_START { \
1248 U16 dupe= trie->states[ state ].wordnum; \
1249 regnode * const noper_next = regnext( noper ); \
1251 if (trie->wordlen) \
1252 trie->wordlen[ curword ] = wordlen; \
1254 /* store the word for dumping */ \
1256 if (OP(noper) != NOTHING) \
1257 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1259 tmp = newSVpvn_utf8( "", 0, UTF ); \
1260 av_push( trie_words, tmp ); \
1265 if ( noper_next < tail ) { \
1267 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1268 trie->jump[curword] = (U16)(noper_next - convert); \
1270 jumper = noper_next; \
1272 nextbranch= regnext(cur); \
1276 /* So it's a dupe. This means we need to maintain a */\
1277 /* linked-list from the first to the next. */\
1278 /* we only allocate the nextword buffer when there */\
1279 /* a dupe, so first time we have to do the allocation */\
1280 if (!trie->nextword) \
1281 trie->nextword = (U16 *) \
1282 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
1283 while ( trie->nextword[dupe] ) \
1284 dupe= trie->nextword[dupe]; \
1285 trie->nextword[dupe]= curword; \
1287 /* we haven't inserted this word yet. */ \
1288 trie->states[ state ].wordnum = curword; \
1293 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1294 ( ( base + charid >= ucharcount \
1295 && base + charid < ubound \
1296 && state == trie->trans[ base - ucharcount + charid ].check \
1297 && trie->trans[ base - ucharcount + charid ].next ) \
1298 ? trie->trans[ base - ucharcount + charid ].next \
1299 : ( state==1 ? special : 0 ) \
1303 #define MADE_JUMP_TRIE 2
1304 #define MADE_EXACT_TRIE 4
1307 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1310 /* first pass, loop through and scan words */
1311 reg_trie_data *trie;
1312 HV *widecharmap = NULL;
1313 AV *revcharmap = newAV();
1315 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1320 regnode *jumper = NULL;
1321 regnode *nextbranch = NULL;
1322 regnode *convert = NULL;
1323 /* we just use folder as a flag in utf8 */
1324 const U8 * const folder = ( flags == EXACTF
1326 : ( flags == EXACTFL
1333 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1334 AV *trie_words = NULL;
1335 /* along with revcharmap, this only used during construction but both are
1336 * useful during debugging so we store them in the struct when debugging.
1339 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1340 STRLEN trie_charcount=0;
1342 SV *re_trie_maxbuff;
1343 GET_RE_DEBUG_FLAGS_DECL;
1345 PERL_ARGS_ASSERT_MAKE_TRIE;
1347 PERL_UNUSED_ARG(depth);
1350 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1352 trie->startstate = 1;
1353 trie->wordcount = word_count;
1354 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1355 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1356 if (!(UTF && folder))
1357 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1359 trie_words = newAV();
1362 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1363 if (!SvIOK(re_trie_maxbuff)) {
1364 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1367 PerlIO_printf( Perl_debug_log,
1368 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1369 (int)depth * 2 + 2, "",
1370 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1371 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1375 /* Find the node we are going to overwrite */
1376 if ( first == startbranch && OP( last ) != BRANCH ) {
1377 /* whole branch chain */
1380 /* branch sub-chain */
1381 convert = NEXTOPER( first );
1384 /* -- First loop and Setup --
1386 We first traverse the branches and scan each word to determine if it
1387 contains widechars, and how many unique chars there are, this is
1388 important as we have to build a table with at least as many columns as we
1391 We use an array of integers to represent the character codes 0..255
1392 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1393 native representation of the character value as the key and IV's for the
1396 *TODO* If we keep track of how many times each character is used we can
1397 remap the columns so that the table compression later on is more
1398 efficient in terms of memory by ensuring most common value is in the
1399 middle and the least common are on the outside. IMO this would be better
1400 than a most to least common mapping as theres a decent chance the most
1401 common letter will share a node with the least common, meaning the node
1402 will not be compressable. With a middle is most common approach the worst
1403 case is when we have the least common nodes twice.
1407 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1408 regnode * const noper = NEXTOPER( cur );
1409 const U8 *uc = (U8*)STRING( noper );
1410 const U8 * const e = uc + STR_LEN( noper );
1412 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1413 const U8 *scan = (U8*)NULL;
1414 U32 wordlen = 0; /* required init */
1416 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1418 if (OP(noper) == NOTHING) {
1422 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1423 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1424 regardless of encoding */
1426 for ( ; uc < e ; uc += len ) {
1427 TRIE_CHARCOUNT(trie)++;
1431 if ( !trie->charmap[ uvc ] ) {
1432 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1434 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1438 /* store the codepoint in the bitmap, and if its ascii
1439 also store its folded equivelent. */
1440 TRIE_BITMAP_SET(trie,uvc);
1442 /* store the folded codepoint */
1443 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1446 /* store first byte of utf8 representation of
1447 codepoints in the 127 < uvc < 256 range */
1448 if (127 < uvc && uvc < 192) {
1449 TRIE_BITMAP_SET(trie,194);
1450 } else if (191 < uvc ) {
1451 TRIE_BITMAP_SET(trie,195);
1452 /* && uvc < 256 -- we know uvc is < 256 already */
1455 set_bit = 0; /* We've done our bit :-) */
1460 widecharmap = newHV();
1462 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1465 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1467 if ( !SvTRUE( *svpp ) ) {
1468 sv_setiv( *svpp, ++trie->uniquecharcount );
1473 if( cur == first ) {
1476 } else if (chars < trie->minlen) {
1478 } else if (chars > trie->maxlen) {
1482 } /* end first pass */
1483 DEBUG_TRIE_COMPILE_r(
1484 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1485 (int)depth * 2 + 2,"",
1486 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1487 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1488 (int)trie->minlen, (int)trie->maxlen )
1490 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
1493 We now know what we are dealing with in terms of unique chars and
1494 string sizes so we can calculate how much memory a naive
1495 representation using a flat table will take. If it's over a reasonable
1496 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1497 conservative but potentially much slower representation using an array
1500 At the end we convert both representations into the same compressed
1501 form that will be used in regexec.c for matching with. The latter
1502 is a form that cannot be used to construct with but has memory
1503 properties similar to the list form and access properties similar
1504 to the table form making it both suitable for fast searches and
1505 small enough that its feasable to store for the duration of a program.
1507 See the comment in the code where the compressed table is produced
1508 inplace from the flat tabe representation for an explanation of how
1509 the compression works.
1514 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1516 Second Pass -- Array Of Lists Representation
1518 Each state will be represented by a list of charid:state records
1519 (reg_trie_trans_le) the first such element holds the CUR and LEN
1520 points of the allocated array. (See defines above).
1522 We build the initial structure using the lists, and then convert
1523 it into the compressed table form which allows faster lookups
1524 (but cant be modified once converted).
1527 STRLEN transcount = 1;
1529 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1530 "%*sCompiling trie using list compiler\n",
1531 (int)depth * 2 + 2, ""));
1533 trie->states = (reg_trie_state *)
1534 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1535 sizeof(reg_trie_state) );
1539 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1541 regnode * const noper = NEXTOPER( cur );
1542 U8 *uc = (U8*)STRING( noper );
1543 const U8 * const e = uc + STR_LEN( noper );
1544 U32 state = 1; /* required init */
1545 U16 charid = 0; /* sanity init */
1546 U8 *scan = (U8*)NULL; /* sanity init */
1547 STRLEN foldlen = 0; /* required init */
1548 U32 wordlen = 0; /* required init */
1549 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1551 if (OP(noper) != NOTHING) {
1552 for ( ; uc < e ; uc += len ) {
1557 charid = trie->charmap[ uvc ];
1559 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1563 charid=(U16)SvIV( *svpp );
1566 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1573 if ( !trie->states[ state ].trans.list ) {
1574 TRIE_LIST_NEW( state );
1576 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1577 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1578 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1583 newstate = next_alloc++;
1584 TRIE_LIST_PUSH( state, charid, newstate );
1589 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1593 TRIE_HANDLE_WORD(state);
1595 } /* end second pass */
1597 /* next alloc is the NEXT state to be allocated */
1598 trie->statecount = next_alloc;
1599 trie->states = (reg_trie_state *)
1600 PerlMemShared_realloc( trie->states,
1602 * sizeof(reg_trie_state) );
1604 /* and now dump it out before we compress it */
1605 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1606 revcharmap, next_alloc,
1610 trie->trans = (reg_trie_trans *)
1611 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1618 for( state=1 ; state < next_alloc ; state ++ ) {
1622 DEBUG_TRIE_COMPILE_MORE_r(
1623 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1627 if (trie->states[state].trans.list) {
1628 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1632 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1633 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1634 if ( forid < minid ) {
1636 } else if ( forid > maxid ) {
1640 if ( transcount < tp + maxid - minid + 1) {
1642 trie->trans = (reg_trie_trans *)
1643 PerlMemShared_realloc( trie->trans,
1645 * sizeof(reg_trie_trans) );
1646 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1648 base = trie->uniquecharcount + tp - minid;
1649 if ( maxid == minid ) {
1651 for ( ; zp < tp ; zp++ ) {
1652 if ( ! trie->trans[ zp ].next ) {
1653 base = trie->uniquecharcount + zp - minid;
1654 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1655 trie->trans[ zp ].check = state;
1661 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1662 trie->trans[ tp ].check = state;
1667 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1668 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1669 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1670 trie->trans[ tid ].check = state;
1672 tp += ( maxid - minid + 1 );
1674 Safefree(trie->states[ state ].trans.list);
1677 DEBUG_TRIE_COMPILE_MORE_r(
1678 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1681 trie->states[ state ].trans.base=base;
1683 trie->lasttrans = tp + 1;
1687 Second Pass -- Flat Table Representation.
1689 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1690 We know that we will need Charcount+1 trans at most to store the data
1691 (one row per char at worst case) So we preallocate both structures
1692 assuming worst case.
1694 We then construct the trie using only the .next slots of the entry
1697 We use the .check field of the first entry of the node temporarily to
1698 make compression both faster and easier by keeping track of how many non
1699 zero fields are in the node.
1701 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1704 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1705 number representing the first entry of the node, and state as a
1706 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1707 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1708 are 2 entrys per node. eg:
1716 The table is internally in the right hand, idx form. However as we also
1717 have to deal with the states array which is indexed by nodenum we have to
1718 use TRIE_NODENUM() to convert.
1721 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1722 "%*sCompiling trie using table compiler\n",
1723 (int)depth * 2 + 2, ""));
1725 trie->trans = (reg_trie_trans *)
1726 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1727 * trie->uniquecharcount + 1,
1728 sizeof(reg_trie_trans) );
1729 trie->states = (reg_trie_state *)
1730 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1731 sizeof(reg_trie_state) );
1732 next_alloc = trie->uniquecharcount + 1;
1735 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1737 regnode * const noper = NEXTOPER( cur );
1738 const U8 *uc = (U8*)STRING( noper );
1739 const U8 * const e = uc + STR_LEN( noper );
1741 U32 state = 1; /* required init */
1743 U16 charid = 0; /* sanity init */
1744 U32 accept_state = 0; /* sanity init */
1745 U8 *scan = (U8*)NULL; /* sanity init */
1747 STRLEN foldlen = 0; /* required init */
1748 U32 wordlen = 0; /* required init */
1749 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1751 if ( OP(noper) != NOTHING ) {
1752 for ( ; uc < e ; uc += len ) {
1757 charid = trie->charmap[ uvc ];
1759 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1760 charid = svpp ? (U16)SvIV(*svpp) : 0;
1764 if ( !trie->trans[ state + charid ].next ) {
1765 trie->trans[ state + charid ].next = next_alloc;
1766 trie->trans[ state ].check++;
1767 next_alloc += trie->uniquecharcount;
1769 state = trie->trans[ state + charid ].next;
1771 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1773 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1776 accept_state = TRIE_NODENUM( state );
1777 TRIE_HANDLE_WORD(accept_state);
1779 } /* end second pass */
1781 /* and now dump it out before we compress it */
1782 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1784 next_alloc, depth+1));
1788 * Inplace compress the table.*
1790 For sparse data sets the table constructed by the trie algorithm will
1791 be mostly 0/FAIL transitions or to put it another way mostly empty.
1792 (Note that leaf nodes will not contain any transitions.)
1794 This algorithm compresses the tables by eliminating most such
1795 transitions, at the cost of a modest bit of extra work during lookup:
1797 - Each states[] entry contains a .base field which indicates the
1798 index in the state[] array wheres its transition data is stored.
1800 - If .base is 0 there are no valid transitions from that node.
1802 - If .base is nonzero then charid is added to it to find an entry in
1805 -If trans[states[state].base+charid].check!=state then the
1806 transition is taken to be a 0/Fail transition. Thus if there are fail
1807 transitions at the front of the node then the .base offset will point
1808 somewhere inside the previous nodes data (or maybe even into a node
1809 even earlier), but the .check field determines if the transition is
1813 The following process inplace converts the table to the compressed
1814 table: We first do not compress the root node 1,and mark its all its
1815 .check pointers as 1 and set its .base pointer as 1 as well. This
1816 allows to do a DFA construction from the compressed table later, and
1817 ensures that any .base pointers we calculate later are greater than
1820 - We set 'pos' to indicate the first entry of the second node.
1822 - We then iterate over the columns of the node, finding the first and
1823 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1824 and set the .check pointers accordingly, and advance pos
1825 appropriately and repreat for the next node. Note that when we copy
1826 the next pointers we have to convert them from the original
1827 NODEIDX form to NODENUM form as the former is not valid post
1830 - If a node has no transitions used we mark its base as 0 and do not
1831 advance the pos pointer.
1833 - If a node only has one transition we use a second pointer into the
1834 structure to fill in allocated fail transitions from other states.
1835 This pointer is independent of the main pointer and scans forward
1836 looking for null transitions that are allocated to a state. When it
1837 finds one it writes the single transition into the "hole". If the
1838 pointer doesnt find one the single transition is appended as normal.
1840 - Once compressed we can Renew/realloc the structures to release the
1843 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1844 specifically Fig 3.47 and the associated pseudocode.
1848 const U32 laststate = TRIE_NODENUM( next_alloc );
1851 trie->statecount = laststate;
1853 for ( state = 1 ; state < laststate ; state++ ) {
1855 const U32 stateidx = TRIE_NODEIDX( state );
1856 const U32 o_used = trie->trans[ stateidx ].check;
1857 U32 used = trie->trans[ stateidx ].check;
1858 trie->trans[ stateidx ].check = 0;
1860 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1861 if ( flag || trie->trans[ stateidx + charid ].next ) {
1862 if ( trie->trans[ stateidx + charid ].next ) {
1864 for ( ; zp < pos ; zp++ ) {
1865 if ( ! trie->trans[ zp ].next ) {
1869 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1870 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1871 trie->trans[ zp ].check = state;
1872 if ( ++zp > pos ) pos = zp;
1879 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1881 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1882 trie->trans[ pos ].check = state;
1887 trie->lasttrans = pos + 1;
1888 trie->states = (reg_trie_state *)
1889 PerlMemShared_realloc( trie->states, laststate
1890 * sizeof(reg_trie_state) );
1891 DEBUG_TRIE_COMPILE_MORE_r(
1892 PerlIO_printf( Perl_debug_log,
1893 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1894 (int)depth * 2 + 2,"",
1895 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1898 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1901 } /* end table compress */
1903 DEBUG_TRIE_COMPILE_MORE_r(
1904 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1905 (int)depth * 2 + 2, "",
1906 (UV)trie->statecount,
1907 (UV)trie->lasttrans)
1909 /* resize the trans array to remove unused space */
1910 trie->trans = (reg_trie_trans *)
1911 PerlMemShared_realloc( trie->trans, trie->lasttrans
1912 * sizeof(reg_trie_trans) );
1914 /* and now dump out the compressed format */
1915 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
1917 { /* Modify the program and insert the new TRIE node*/
1918 U8 nodetype =(U8)(flags & 0xFF);
1922 regnode *optimize = NULL;
1923 #ifdef RE_TRACK_PATTERN_OFFSETS
1926 U32 mjd_nodelen = 0;
1927 #endif /* RE_TRACK_PATTERN_OFFSETS */
1928 #endif /* DEBUGGING */
1930 This means we convert either the first branch or the first Exact,
1931 depending on whether the thing following (in 'last') is a branch
1932 or not and whther first is the startbranch (ie is it a sub part of
1933 the alternation or is it the whole thing.)
1934 Assuming its a sub part we conver the EXACT otherwise we convert
1935 the whole branch sequence, including the first.
1937 /* Find the node we are going to overwrite */
1938 if ( first != startbranch || OP( last ) == BRANCH ) {
1939 /* branch sub-chain */
1940 NEXT_OFF( first ) = (U16)(last - first);
1941 #ifdef RE_TRACK_PATTERN_OFFSETS
1943 mjd_offset= Node_Offset((convert));
1944 mjd_nodelen= Node_Length((convert));
1947 /* whole branch chain */
1949 #ifdef RE_TRACK_PATTERN_OFFSETS
1952 const regnode *nop = NEXTOPER( convert );
1953 mjd_offset= Node_Offset((nop));
1954 mjd_nodelen= Node_Length((nop));
1958 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1959 (int)depth * 2 + 2, "",
1960 (UV)mjd_offset, (UV)mjd_nodelen)
1963 /* But first we check to see if there is a common prefix we can
1964 split out as an EXACT and put in front of the TRIE node. */
1965 trie->startstate= 1;
1966 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1968 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1972 const U32 base = trie->states[ state ].trans.base;
1974 if ( trie->states[state].wordnum )
1977 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1978 if ( ( base + ofs >= trie->uniquecharcount ) &&
1979 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1980 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1982 if ( ++count > 1 ) {
1983 SV **tmp = av_fetch( revcharmap, ofs, 0);
1984 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
1985 if ( state == 1 ) break;
1987 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1989 PerlIO_printf(Perl_debug_log,
1990 "%*sNew Start State=%"UVuf" Class: [",
1991 (int)depth * 2 + 2, "",
1994 SV ** const tmp = av_fetch( revcharmap, idx, 0);
1995 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
1997 TRIE_BITMAP_SET(trie,*ch);
1999 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2001 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2005 TRIE_BITMAP_SET(trie,*ch);
2007 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2008 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2014 SV **tmp = av_fetch( revcharmap, idx, 0);
2016 char *ch = SvPV( *tmp, len );
2018 SV *sv=sv_newmortal();
2019 PerlIO_printf( Perl_debug_log,
2020 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2021 (int)depth * 2 + 2, "",
2023 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2024 PL_colors[0], PL_colors[1],
2025 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2026 PERL_PV_ESCAPE_FIRSTCHAR
2031 OP( convert ) = nodetype;
2032 str=STRING(convert);
2035 STR_LEN(convert) += len;
2041 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2047 regnode *n = convert+NODE_SZ_STR(convert);
2048 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2049 trie->startstate = state;
2050 trie->minlen -= (state - 1);
2051 trie->maxlen -= (state - 1);
2053 /* At least the UNICOS C compiler choked on this
2054 * being argument to DEBUG_r(), so let's just have
2057 #ifdef PERL_EXT_RE_BUILD
2063 regnode *fix = convert;
2064 U32 word = trie->wordcount;
2066 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2067 while( ++fix < n ) {
2068 Set_Node_Offset_Length(fix, 0, 0);
2071 SV ** const tmp = av_fetch( trie_words, word, 0 );
2073 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2074 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2076 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2084 NEXT_OFF(convert) = (U16)(tail - convert);
2085 DEBUG_r(optimize= n);
2091 if ( trie->maxlen ) {
2092 NEXT_OFF( convert ) = (U16)(tail - convert);
2093 ARG_SET( convert, data_slot );
2094 /* Store the offset to the first unabsorbed branch in
2095 jump[0], which is otherwise unused by the jump logic.
2096 We use this when dumping a trie and during optimisation. */
2098 trie->jump[0] = (U16)(nextbranch - convert);
2101 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2102 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2104 OP( convert ) = TRIEC;
2105 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2106 PerlMemShared_free(trie->bitmap);
2109 OP( convert ) = TRIE;
2111 /* store the type in the flags */
2112 convert->flags = nodetype;
2116 + regarglen[ OP( convert ) ];
2118 /* XXX We really should free up the resource in trie now,
2119 as we won't use them - (which resources?) dmq */
2121 /* needed for dumping*/
2122 DEBUG_r(if (optimize) {
2123 regnode *opt = convert;
2125 while ( ++opt < optimize) {
2126 Set_Node_Offset_Length(opt,0,0);
2129 Try to clean up some of the debris left after the
2132 while( optimize < jumper ) {
2133 mjd_nodelen += Node_Length((optimize));
2134 OP( optimize ) = OPTIMIZED;
2135 Set_Node_Offset_Length(optimize,0,0);
2138 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2140 } /* end node insert */
2141 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2143 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2144 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2146 SvREFCNT_dec(revcharmap);
2150 : trie->startstate>1
2156 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2158 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2160 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2161 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2164 We find the fail state for each state in the trie, this state is the longest proper
2165 suffix of the current states 'word' that is also a proper prefix of another word in our
2166 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2167 the DFA not to have to restart after its tried and failed a word at a given point, it
2168 simply continues as though it had been matching the other word in the first place.
2170 'abcdgu'=~/abcdefg|cdgu/
2171 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2172 fail, which would bring use to the state representing 'd' in the second word where we would
2173 try 'g' and succeed, prodceding to match 'cdgu'.
2175 /* add a fail transition */
2176 const U32 trie_offset = ARG(source);
2177 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2179 const U32 ucharcount = trie->uniquecharcount;
2180 const U32 numstates = trie->statecount;
2181 const U32 ubound = trie->lasttrans + ucharcount;
2185 U32 base = trie->states[ 1 ].trans.base;
2188 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2189 GET_RE_DEBUG_FLAGS_DECL;
2191 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2193 PERL_UNUSED_ARG(depth);
2197 ARG_SET( stclass, data_slot );
2198 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2199 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2200 aho->trie=trie_offset;
2201 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2202 Copy( trie->states, aho->states, numstates, reg_trie_state );
2203 Newxz( q, numstates, U32);
2204 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2207 /* initialize fail[0..1] to be 1 so that we always have
2208 a valid final fail state */
2209 fail[ 0 ] = fail[ 1 ] = 1;
2211 for ( charid = 0; charid < ucharcount ; charid++ ) {
2212 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2214 q[ q_write ] = newstate;
2215 /* set to point at the root */
2216 fail[ q[ q_write++ ] ]=1;
2219 while ( q_read < q_write) {
2220 const U32 cur = q[ q_read++ % numstates ];
2221 base = trie->states[ cur ].trans.base;
2223 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2224 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2226 U32 fail_state = cur;
2229 fail_state = fail[ fail_state ];
2230 fail_base = aho->states[ fail_state ].trans.base;
2231 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2233 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2234 fail[ ch_state ] = fail_state;
2235 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2237 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2239 q[ q_write++ % numstates] = ch_state;
2243 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2244 when we fail in state 1, this allows us to use the
2245 charclass scan to find a valid start char. This is based on the principle
2246 that theres a good chance the string being searched contains lots of stuff
2247 that cant be a start char.
2249 fail[ 0 ] = fail[ 1 ] = 0;
2250 DEBUG_TRIE_COMPILE_r({
2251 PerlIO_printf(Perl_debug_log,
2252 "%*sStclass Failtable (%"UVuf" states): 0",
2253 (int)(depth * 2), "", (UV)numstates
2255 for( q_read=1; q_read<numstates; q_read++ ) {
2256 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2258 PerlIO_printf(Perl_debug_log, "\n");
2261 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2266 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2267 * These need to be revisited when a newer toolchain becomes available.
2269 #if defined(__sparc64__) && defined(__GNUC__)
2270 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2271 # undef SPARC64_GCC_WORKAROUND
2272 # define SPARC64_GCC_WORKAROUND 1
2276 #define DEBUG_PEEP(str,scan,depth) \
2277 DEBUG_OPTIMISE_r({if (scan){ \
2278 SV * const mysv=sv_newmortal(); \
2279 regnode *Next = regnext(scan); \
2280 regprop(RExC_rx, mysv, scan); \
2281 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2282 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2283 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2290 #define JOIN_EXACT(scan,min,flags) \
2291 if (PL_regkind[OP(scan)] == EXACT) \
2292 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2295 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2296 /* Merge several consecutive EXACTish nodes into one. */
2297 regnode *n = regnext(scan);
2299 regnode *next = scan + NODE_SZ_STR(scan);
2303 regnode *stop = scan;
2304 GET_RE_DEBUG_FLAGS_DECL;
2306 PERL_UNUSED_ARG(depth);
2309 PERL_ARGS_ASSERT_JOIN_EXACT;
2310 #ifndef EXPERIMENTAL_INPLACESCAN
2311 PERL_UNUSED_ARG(flags);
2312 PERL_UNUSED_ARG(val);
2314 DEBUG_PEEP("join",scan,depth);
2316 /* Skip NOTHING, merge EXACT*. */
2318 ( PL_regkind[OP(n)] == NOTHING ||
2319 (stringok && (OP(n) == OP(scan))))
2321 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2323 if (OP(n) == TAIL || n > next)
2325 if (PL_regkind[OP(n)] == NOTHING) {
2326 DEBUG_PEEP("skip:",n,depth);
2327 NEXT_OFF(scan) += NEXT_OFF(n);
2328 next = n + NODE_STEP_REGNODE;
2335 else if (stringok) {
2336 const unsigned int oldl = STR_LEN(scan);
2337 regnode * const nnext = regnext(n);
2339 DEBUG_PEEP("merg",n,depth);
2342 if (oldl + STR_LEN(n) > U8_MAX)
2344 NEXT_OFF(scan) += NEXT_OFF(n);
2345 STR_LEN(scan) += STR_LEN(n);
2346 next = n + NODE_SZ_STR(n);
2347 /* Now we can overwrite *n : */
2348 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2356 #ifdef EXPERIMENTAL_INPLACESCAN
2357 if (flags && !NEXT_OFF(n)) {
2358 DEBUG_PEEP("atch", val, depth);
2359 if (reg_off_by_arg[OP(n)]) {
2360 ARG_SET(n, val - n);
2363 NEXT_OFF(n) = val - n;
2370 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2372 Two problematic code points in Unicode casefolding of EXACT nodes:
2374 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2375 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2381 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2382 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2384 This means that in case-insensitive matching (or "loose matching",
2385 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2386 length of the above casefolded versions) can match a target string
2387 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2388 This would rather mess up the minimum length computation.
2390 What we'll do is to look for the tail four bytes, and then peek
2391 at the preceding two bytes to see whether we need to decrease
2392 the minimum length by four (six minus two).
2394 Thanks to the design of UTF-8, there cannot be false matches:
2395 A sequence of valid UTF-8 bytes cannot be a subsequence of
2396 another valid sequence of UTF-8 bytes.
2399 char * const s0 = STRING(scan), *s, *t;
2400 char * const s1 = s0 + STR_LEN(scan) - 1;
2401 char * const s2 = s1 - 4;
2402 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2403 const char t0[] = "\xaf\x49\xaf\x42";
2405 const char t0[] = "\xcc\x88\xcc\x81";
2407 const char * const t1 = t0 + 3;
2410 s < s2 && (t = ninstr(s, s1, t0, t1));
2413 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2414 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2416 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2417 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2425 n = scan + NODE_SZ_STR(scan);
2427 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2434 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2438 /* REx optimizer. Converts nodes into quickier variants "in place".
2439 Finds fixed substrings. */
2441 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2442 to the position after last scanned or to NULL. */
2444 #define INIT_AND_WITHP \
2445 assert(!and_withp); \
2446 Newx(and_withp,1,struct regnode_charclass_class); \
2447 SAVEFREEPV(and_withp)
2449 /* this is a chain of data about sub patterns we are processing that
2450 need to be handled seperately/specially in study_chunk. Its so
2451 we can simulate recursion without losing state. */
2453 typedef struct scan_frame {
2454 regnode *last; /* last node to process in this frame */
2455 regnode *next; /* next node to process when last is reached */
2456 struct scan_frame *prev; /*previous frame*/
2457 I32 stop; /* what stopparen do we use */
2461 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2463 #define CASE_SYNST_FNC(nAmE) \
2465 if (flags & SCF_DO_STCLASS_AND) { \
2466 for (value = 0; value < 256; value++) \
2467 if (!is_ ## nAmE ## _cp(value)) \
2468 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2471 for (value = 0; value < 256; value++) \
2472 if (is_ ## nAmE ## _cp(value)) \
2473 ANYOF_BITMAP_SET(data->start_class, value); \
2477 if (flags & SCF_DO_STCLASS_AND) { \
2478 for (value = 0; value < 256; value++) \
2479 if (is_ ## nAmE ## _cp(value)) \
2480 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2483 for (value = 0; value < 256; value++) \
2484 if (!is_ ## nAmE ## _cp(value)) \
2485 ANYOF_BITMAP_SET(data->start_class, value); \
2492 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2493 I32 *minlenp, I32 *deltap,
2498 struct regnode_charclass_class *and_withp,
2499 U32 flags, U32 depth)
2500 /* scanp: Start here (read-write). */
2501 /* deltap: Write maxlen-minlen here. */
2502 /* last: Stop before this one. */
2503 /* data: string data about the pattern */
2504 /* stopparen: treat close N as END */
2505 /* recursed: which subroutines have we recursed into */
2506 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2509 I32 min = 0, pars = 0, code;
2510 regnode *scan = *scanp, *next;
2512 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2513 int is_inf_internal = 0; /* The studied chunk is infinite */
2514 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2515 scan_data_t data_fake;
2516 SV *re_trie_maxbuff = NULL;
2517 regnode *first_non_open = scan;
2518 I32 stopmin = I32_MAX;
2519 scan_frame *frame = NULL;
2520 GET_RE_DEBUG_FLAGS_DECL;
2522 PERL_ARGS_ASSERT_STUDY_CHUNK;
2525 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2529 while (first_non_open && OP(first_non_open) == OPEN)
2530 first_non_open=regnext(first_non_open);
2535 while ( scan && OP(scan) != END && scan < last ){
2536 /* Peephole optimizer: */
2537 DEBUG_STUDYDATA("Peep:", data,depth);
2538 DEBUG_PEEP("Peep",scan,depth);
2539 JOIN_EXACT(scan,&min,0);
2541 /* Follow the next-chain of the current node and optimize
2542 away all the NOTHINGs from it. */
2543 if (OP(scan) != CURLYX) {
2544 const int max = (reg_off_by_arg[OP(scan)]
2546 /* I32 may be smaller than U16 on CRAYs! */
2547 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2548 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2552 /* Skip NOTHING and LONGJMP. */
2553 while ((n = regnext(n))
2554 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2555 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2556 && off + noff < max)
2558 if (reg_off_by_arg[OP(scan)])
2561 NEXT_OFF(scan) = off;
2566 /* The principal pseudo-switch. Cannot be a switch, since we
2567 look into several different things. */
2568 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2569 || OP(scan) == IFTHEN) {
2570 next = regnext(scan);
2572 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2574 if (OP(next) == code || code == IFTHEN) {
2575 /* NOTE - There is similar code to this block below for handling
2576 TRIE nodes on a re-study. If you change stuff here check there
2578 I32 max1 = 0, min1 = I32_MAX, num = 0;
2579 struct regnode_charclass_class accum;
2580 regnode * const startbranch=scan;
2582 if (flags & SCF_DO_SUBSTR)
2583 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2584 if (flags & SCF_DO_STCLASS)
2585 cl_init_zero(pRExC_state, &accum);
2587 while (OP(scan) == code) {
2588 I32 deltanext, minnext, f = 0, fake;
2589 struct regnode_charclass_class this_class;
2592 data_fake.flags = 0;
2594 data_fake.whilem_c = data->whilem_c;
2595 data_fake.last_closep = data->last_closep;
2598 data_fake.last_closep = &fake;
2600 data_fake.pos_delta = delta;
2601 next = regnext(scan);
2602 scan = NEXTOPER(scan);
2604 scan = NEXTOPER(scan);
2605 if (flags & SCF_DO_STCLASS) {
2606 cl_init(pRExC_state, &this_class);
2607 data_fake.start_class = &this_class;
2608 f = SCF_DO_STCLASS_AND;
2610 if (flags & SCF_WHILEM_VISITED_POS)
2611 f |= SCF_WHILEM_VISITED_POS;
2613 /* we suppose the run is continuous, last=next...*/
2614 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2616 stopparen, recursed, NULL, f,depth+1);
2619 if (max1 < minnext + deltanext)
2620 max1 = minnext + deltanext;
2621 if (deltanext == I32_MAX)
2622 is_inf = is_inf_internal = 1;
2624 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2626 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2627 if ( stopmin > minnext)
2628 stopmin = min + min1;
2629 flags &= ~SCF_DO_SUBSTR;
2631 data->flags |= SCF_SEEN_ACCEPT;
2634 if (data_fake.flags & SF_HAS_EVAL)
2635 data->flags |= SF_HAS_EVAL;
2636 data->whilem_c = data_fake.whilem_c;
2638 if (flags & SCF_DO_STCLASS)
2639 cl_or(pRExC_state, &accum, &this_class);
2641 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2643 if (flags & SCF_DO_SUBSTR) {
2644 data->pos_min += min1;
2645 data->pos_delta += max1 - min1;
2646 if (max1 != min1 || is_inf)
2647 data->longest = &(data->longest_float);
2650 delta += max1 - min1;
2651 if (flags & SCF_DO_STCLASS_OR) {
2652 cl_or(pRExC_state, data->start_class, &accum);
2654 cl_and(data->start_class, and_withp);
2655 flags &= ~SCF_DO_STCLASS;
2658 else if (flags & SCF_DO_STCLASS_AND) {
2660 cl_and(data->start_class, &accum);
2661 flags &= ~SCF_DO_STCLASS;
2664 /* Switch to OR mode: cache the old value of
2665 * data->start_class */
2667 StructCopy(data->start_class, and_withp,
2668 struct regnode_charclass_class);
2669 flags &= ~SCF_DO_STCLASS_AND;
2670 StructCopy(&accum, data->start_class,
2671 struct regnode_charclass_class);
2672 flags |= SCF_DO_STCLASS_OR;
2673 data->start_class->flags |= ANYOF_EOS;
2677 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2680 Assuming this was/is a branch we are dealing with: 'scan' now
2681 points at the item that follows the branch sequence, whatever
2682 it is. We now start at the beginning of the sequence and look
2689 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2691 If we can find such a subseqence we need to turn the first
2692 element into a trie and then add the subsequent branch exact
2693 strings to the trie.
2697 1. patterns where the whole set of branch can be converted.
2699 2. patterns where only a subset can be converted.
2701 In case 1 we can replace the whole set with a single regop
2702 for the trie. In case 2 we need to keep the start and end
2705 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2706 becomes BRANCH TRIE; BRANCH X;
2708 There is an additional case, that being where there is a
2709 common prefix, which gets split out into an EXACT like node
2710 preceding the TRIE node.
2712 If x(1..n)==tail then we can do a simple trie, if not we make
2713 a "jump" trie, such that when we match the appropriate word
2714 we "jump" to the appopriate tail node. Essentailly we turn
2715 a nested if into a case structure of sorts.
2720 if (!re_trie_maxbuff) {
2721 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2722 if (!SvIOK(re_trie_maxbuff))
2723 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2725 if ( SvIV(re_trie_maxbuff)>=0 ) {
2727 regnode *first = (regnode *)NULL;
2728 regnode *last = (regnode *)NULL;
2729 regnode *tail = scan;
2734 SV * const mysv = sv_newmortal(); /* for dumping */
2736 /* var tail is used because there may be a TAIL
2737 regop in the way. Ie, the exacts will point to the
2738 thing following the TAIL, but the last branch will
2739 point at the TAIL. So we advance tail. If we
2740 have nested (?:) we may have to move through several
2744 while ( OP( tail ) == TAIL ) {
2745 /* this is the TAIL generated by (?:) */
2746 tail = regnext( tail );
2751 regprop(RExC_rx, mysv, tail );
2752 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2753 (int)depth * 2 + 2, "",
2754 "Looking for TRIE'able sequences. Tail node is: ",
2755 SvPV_nolen_const( mysv )
2761 step through the branches, cur represents each
2762 branch, noper is the first thing to be matched
2763 as part of that branch and noper_next is the
2764 regnext() of that node. if noper is an EXACT
2765 and noper_next is the same as scan (our current
2766 position in the regex) then the EXACT branch is
2767 a possible optimization target. Once we have
2768 two or more consequetive such branches we can
2769 create a trie of the EXACT's contents and stich
2770 it in place. If the sequence represents all of
2771 the branches we eliminate the whole thing and
2772 replace it with a single TRIE. If it is a
2773 subsequence then we need to stitch it in. This
2774 means the first branch has to remain, and needs
2775 to be repointed at the item on the branch chain
2776 following the last branch optimized. This could
2777 be either a BRANCH, in which case the
2778 subsequence is internal, or it could be the
2779 item following the branch sequence in which
2780 case the subsequence is at the end.
2784 /* dont use tail as the end marker for this traverse */
2785 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2786 regnode * const noper = NEXTOPER( cur );
2787 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2788 regnode * const noper_next = regnext( noper );
2792 regprop(RExC_rx, mysv, cur);
2793 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2794 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2796 regprop(RExC_rx, mysv, noper);
2797 PerlIO_printf( Perl_debug_log, " -> %s",
2798 SvPV_nolen_const(mysv));
2801 regprop(RExC_rx, mysv, noper_next );
2802 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2803 SvPV_nolen_const(mysv));
2805 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2806 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2808 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2809 : PL_regkind[ OP( noper ) ] == EXACT )
2810 || OP(noper) == NOTHING )
2812 && noper_next == tail
2817 if ( !first || optype == NOTHING ) {
2818 if (!first) first = cur;
2819 optype = OP( noper );
2825 Currently we assume that the trie can handle unicode and ascii
2826 matches fold cased matches. If this proves true then the following
2827 define will prevent tries in this situation.
2829 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2831 #define TRIE_TYPE_IS_SAFE 1
2832 if ( last && TRIE_TYPE_IS_SAFE ) {
2833 make_trie( pRExC_state,
2834 startbranch, first, cur, tail, count,
2837 if ( PL_regkind[ OP( noper ) ] == EXACT
2839 && noper_next == tail
2844 optype = OP( noper );
2854 regprop(RExC_rx, mysv, cur);
2855 PerlIO_printf( Perl_debug_log,
2856 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2857 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2861 if ( last && TRIE_TYPE_IS_SAFE ) {
2862 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2863 #ifdef TRIE_STUDY_OPT
2864 if ( ((made == MADE_EXACT_TRIE &&
2865 startbranch == first)
2866 || ( first_non_open == first )) &&
2868 flags |= SCF_TRIE_RESTUDY;
2869 if ( startbranch == first
2872 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2882 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2883 scan = NEXTOPER(NEXTOPER(scan));
2884 } else /* single branch is optimized. */
2885 scan = NEXTOPER(scan);
2887 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2888 scan_frame *newframe = NULL;
2893 if (OP(scan) != SUSPEND) {
2894 /* set the pointer */
2895 if (OP(scan) == GOSUB) {
2897 RExC_recurse[ARG2L(scan)] = scan;
2898 start = RExC_open_parens[paren-1];
2899 end = RExC_close_parens[paren-1];
2902 start = RExC_rxi->program + 1;
2906 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2907 SAVEFREEPV(recursed);
2909 if (!PAREN_TEST(recursed,paren+1)) {
2910 PAREN_SET(recursed,paren+1);
2911 Newx(newframe,1,scan_frame);
2913 if (flags & SCF_DO_SUBSTR) {
2914 SCAN_COMMIT(pRExC_state,data,minlenp);
2915 data->longest = &(data->longest_float);
2917 is_inf = is_inf_internal = 1;
2918 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2919 cl_anything(pRExC_state, data->start_class);
2920 flags &= ~SCF_DO_STCLASS;
2923 Newx(newframe,1,scan_frame);
2926 end = regnext(scan);
2931 SAVEFREEPV(newframe);
2932 newframe->next = regnext(scan);
2933 newframe->last = last;
2934 newframe->stop = stopparen;
2935 newframe->prev = frame;
2945 else if (OP(scan) == EXACT) {
2946 I32 l = STR_LEN(scan);
2949 const U8 * const s = (U8*)STRING(scan);
2950 l = utf8_length(s, s + l);
2951 uc = utf8_to_uvchr(s, NULL);
2953 uc = *((U8*)STRING(scan));
2956 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2957 /* The code below prefers earlier match for fixed
2958 offset, later match for variable offset. */
2959 if (data->last_end == -1) { /* Update the start info. */
2960 data->last_start_min = data->pos_min;
2961 data->last_start_max = is_inf
2962 ? I32_MAX : data->pos_min + data->pos_delta;
2964 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2966 SvUTF8_on(data->last_found);
2968 SV * const sv = data->last_found;
2969 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2970 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2971 if (mg && mg->mg_len >= 0)
2972 mg->mg_len += utf8_length((U8*)STRING(scan),
2973 (U8*)STRING(scan)+STR_LEN(scan));
2975 data->last_end = data->pos_min + l;
2976 data->pos_min += l; /* As in the first entry. */
2977 data->flags &= ~SF_BEFORE_EOL;
2979 if (flags & SCF_DO_STCLASS_AND) {
2980 /* Check whether it is compatible with what we know already! */
2984 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2985 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2986 && (!(data->start_class->flags & ANYOF_FOLD)
2987 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2990 ANYOF_CLASS_ZERO(data->start_class);
2991 ANYOF_BITMAP_ZERO(data->start_class);
2993 ANYOF_BITMAP_SET(data->start_class, uc);
2994 data->start_class->flags &= ~ANYOF_EOS;
2996 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2998 else if (flags & SCF_DO_STCLASS_OR) {
2999 /* false positive possible if the class is case-folded */
3001 ANYOF_BITMAP_SET(data->start_class, uc);
3003 data->start_class->flags |= ANYOF_UNICODE_ALL;
3004 data->start_class->flags &= ~ANYOF_EOS;
3005 cl_and(data->start_class, and_withp);
3007 flags &= ~SCF_DO_STCLASS;
3009 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3010 I32 l = STR_LEN(scan);
3011 UV uc = *((U8*)STRING(scan));
3013 /* Search for fixed substrings supports EXACT only. */
3014 if (flags & SCF_DO_SUBSTR) {
3016 SCAN_COMMIT(pRExC_state, data, minlenp);
3019 const U8 * const s = (U8 *)STRING(scan);
3020 l = utf8_length(s, s + l);
3021 uc = utf8_to_uvchr(s, NULL);
3024 if (flags & SCF_DO_SUBSTR)
3026 if (flags & SCF_DO_STCLASS_AND) {
3027 /* Check whether it is compatible with what we know already! */
3031 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3032 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3033 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3035 ANYOF_CLASS_ZERO(data->start_class);
3036 ANYOF_BITMAP_ZERO(data->start_class);
3038 ANYOF_BITMAP_SET(data->start_class, uc);
3039 data->start_class->flags &= ~ANYOF_EOS;
3040 data->start_class->flags |= ANYOF_FOLD;
3041 if (OP(scan) == EXACTFL)
3042 data->start_class->flags |= ANYOF_LOCALE;
3045 else if (flags & SCF_DO_STCLASS_OR) {
3046 if (data->start_class->flags & ANYOF_FOLD) {
3047 /* false positive possible if the class is case-folded.
3048 Assume that the locale settings are the same... */
3050 ANYOF_BITMAP_SET(data->start_class, uc);
3051 data->start_class->flags &= ~ANYOF_EOS;
3053 cl_and(data->start_class, and_withp);
3055 flags &= ~SCF_DO_STCLASS;
3057 else if (strchr((const char*)PL_varies,OP(scan))) {
3058 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3059 I32 f = flags, pos_before = 0;
3060 regnode * const oscan = scan;
3061 struct regnode_charclass_class this_class;
3062 struct regnode_charclass_class *oclass = NULL;
3063 I32 next_is_eval = 0;
3065 switch (PL_regkind[OP(scan)]) {
3066 case WHILEM: /* End of (?:...)* . */
3067 scan = NEXTOPER(scan);
3070 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3071 next = NEXTOPER(scan);
3072 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3074 maxcount = REG_INFTY;
3075 next = regnext(scan);
3076 scan = NEXTOPER(scan);
3080 if (flags & SCF_DO_SUBSTR)
3085 if (flags & SCF_DO_STCLASS) {
3087 maxcount = REG_INFTY;
3088 next = regnext(scan);
3089 scan = NEXTOPER(scan);
3092 is_inf = is_inf_internal = 1;
3093 scan = regnext(scan);
3094 if (flags & SCF_DO_SUBSTR) {
3095 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3096 data->longest = &(data->longest_float);
3098 goto optimize_curly_tail;
3100 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3101 && (scan->flags == stopparen))
3106 mincount = ARG1(scan);
3107 maxcount = ARG2(scan);
3109 next = regnext(scan);
3110 if (OP(scan) == CURLYX) {
3111 I32 lp = (data ? *(data->last_closep) : 0);
3112 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3114 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3115 next_is_eval = (OP(scan) == EVAL);
3117 if (flags & SCF_DO_SUBSTR) {
3118 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3119 pos_before = data->pos_min;
3123 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3125 data->flags |= SF_IS_INF;
3127 if (flags & SCF_DO_STCLASS) {
3128 cl_init(pRExC_state, &this_class);
3129 oclass = data->start_class;
3130 data->start_class = &this_class;
3131 f |= SCF_DO_STCLASS_AND;
3132 f &= ~SCF_DO_STCLASS_OR;
3134 /* These are the cases when once a subexpression
3135 fails at a particular position, it cannot succeed
3136 even after backtracking at the enclosing scope.
3138 XXXX what if minimal match and we are at the
3139 initial run of {n,m}? */
3140 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3141 f &= ~SCF_WHILEM_VISITED_POS;
3143 /* This will finish on WHILEM, setting scan, or on NULL: */
3144 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3145 last, data, stopparen, recursed, NULL,
3147 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3149 if (flags & SCF_DO_STCLASS)
3150 data->start_class = oclass;
3151 if (mincount == 0 || minnext == 0) {
3152 if (flags & SCF_DO_STCLASS_OR) {
3153 cl_or(pRExC_state, data->start_class, &this_class);
3155 else if (flags & SCF_DO_STCLASS_AND) {
3156 /* Switch to OR mode: cache the old value of
3157 * data->start_class */
3159 StructCopy(data->start_class, and_withp,
3160 struct regnode_charclass_class);
3161 flags &= ~SCF_DO_STCLASS_AND;
3162 StructCopy(&this_class, data->start_class,
3163 struct regnode_charclass_class);
3164 flags |= SCF_DO_STCLASS_OR;
3165 data->start_class->flags |= ANYOF_EOS;
3167 } else { /* Non-zero len */
3168 if (flags & SCF_DO_STCLASS_OR) {
3169 cl_or(pRExC_state, data->start_class, &this_class);
3170 cl_and(data->start_class, and_withp);
3172 else if (flags & SCF_DO_STCLASS_AND)
3173 cl_and(data->start_class, &this_class);
3174 flags &= ~SCF_DO_STCLASS;
3176 if (!scan) /* It was not CURLYX, but CURLY. */
3178 if ( /* ? quantifier ok, except for (?{ ... }) */
3179 (next_is_eval || !(mincount == 0 && maxcount == 1))
3180 && (minnext == 0) && (deltanext == 0)
3181 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3182 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3183 && ckWARN(WARN_REGEXP))
3186 "Quantifier unexpected on zero-length expression");
3189 min += minnext * mincount;
3190 is_inf_internal |= ((maxcount == REG_INFTY
3191 && (minnext + deltanext) > 0)
3192 || deltanext == I32_MAX);
3193 is_inf |= is_inf_internal;
3194 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3196 /* Try powerful optimization CURLYX => CURLYN. */
3197 if ( OP(oscan) == CURLYX && data
3198 && data->flags & SF_IN_PAR
3199 && !(data->flags & SF_HAS_EVAL)
3200 && !deltanext && minnext == 1 ) {
3201 /* Try to optimize to CURLYN. */
3202 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3203 regnode * const nxt1 = nxt;
3210 if (!strchr((const char*)PL_simple,OP(nxt))
3211 && !(PL_regkind[OP(nxt)] == EXACT
3212 && STR_LEN(nxt) == 1))
3218 if (OP(nxt) != CLOSE)
3220 if (RExC_open_parens) {
3221 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3222 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3224 /* Now we know that nxt2 is the only contents: */
3225 oscan->flags = (U8)ARG(nxt);
3227 OP(nxt1) = NOTHING; /* was OPEN. */
3230 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3231 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3232 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3233 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3234 OP(nxt + 1) = OPTIMIZED; /* was count. */
3235 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3240 /* Try optimization CURLYX => CURLYM. */
3241 if ( OP(oscan) == CURLYX && data
3242 && !(data->flags & SF_HAS_PAR)
3243 && !(data->flags & SF_HAS_EVAL)
3244 && !deltanext /* atom is fixed width */
3245 && minnext != 0 /* CURLYM can't handle zero width */
3247 /* XXXX How to optimize if data == 0? */
3248 /* Optimize to a simpler form. */
3249 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3253 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3254 && (OP(nxt2) != WHILEM))
3256 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3257 /* Need to optimize away parenths. */
3258 if (data->flags & SF_IN_PAR) {
3259 /* Set the parenth number. */
3260 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3262 if (OP(nxt) != CLOSE)
3263 FAIL("Panic opt close");
3264 oscan->flags = (U8)ARG(nxt);
3265 if (RExC_open_parens) {
3266 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3267 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3269 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3270 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3273 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3274 OP(nxt + 1) = OPTIMIZED; /* was count. */
3275 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3276 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3279 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3280 regnode *nnxt = regnext(nxt1);
3283 if (reg_off_by_arg[OP(nxt1)])
3284 ARG_SET(nxt1, nxt2 - nxt1);
3285 else if (nxt2 - nxt1 < U16_MAX)
3286 NEXT_OFF(nxt1) = nxt2 - nxt1;
3288 OP(nxt) = NOTHING; /* Cannot beautify */
3293 /* Optimize again: */
3294 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3295 NULL, stopparen, recursed, NULL, 0,depth+1);
3300 else if ((OP(oscan) == CURLYX)
3301 && (flags & SCF_WHILEM_VISITED_POS)
3302 /* See the comment on a similar expression above.
3303 However, this time it not a subexpression
3304 we care about, but the expression itself. */
3305 && (maxcount == REG_INFTY)
3306 && data && ++data->whilem_c < 16) {
3307 /* This stays as CURLYX, we can put the count/of pair. */
3308 /* Find WHILEM (as in regexec.c) */
3309 regnode *nxt = oscan + NEXT_OFF(oscan);
3311 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3313 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3314 | (RExC_whilem_seen << 4)); /* On WHILEM */
3316 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3318 if (flags & SCF_DO_SUBSTR) {
3319 SV *last_str = NULL;
3320 int counted = mincount != 0;
3322 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3323 #if defined(SPARC64_GCC_WORKAROUND)
3326 const char *s = NULL;
3329 if (pos_before >= data->last_start_min)
3332 b = data->last_start_min;
3335 s = SvPV_const(data->last_found, l);
3336 old = b - data->last_start_min;
3339 I32 b = pos_before >= data->last_start_min
3340 ? pos_before : data->last_start_min;
3342 const char * const s = SvPV_const(data->last_found, l);
3343 I32 old = b - data->last_start_min;
3347 old = utf8_hop((U8*)s, old) - (U8*)s;
3350 /* Get the added string: */
3351 last_str = newSVpvn_utf8(s + old, l, UTF);
3352 if (deltanext == 0 && pos_before == b) {
3353 /* What was added is a constant string */
3355 SvGROW(last_str, (mincount * l) + 1);
3356 repeatcpy(SvPVX(last_str) + l,
3357 SvPVX_const(last_str), l, mincount - 1);
3358 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3359 /* Add additional parts. */
3360 SvCUR_set(data->last_found,
3361 SvCUR(data->last_found) - l);
3362 sv_catsv(data->last_found, last_str);
3364 SV * sv = data->last_found;
3366 SvUTF8(sv) && SvMAGICAL(sv) ?
3367 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3368 if (mg && mg->mg_len >= 0)
3369 mg->mg_len += CHR_SVLEN(last_str) - l;
3371 data->last_end += l * (mincount - 1);
3374 /* start offset must point into the last copy */
3375 data->last_start_min += minnext * (mincount - 1);
3376 data->last_start_max += is_inf ? I32_MAX
3377 : (maxcount - 1) * (minnext + data->pos_delta);
3380 /* It is counted once already... */
3381 data->pos_min += minnext * (mincount - counted);
3382 data->pos_delta += - counted * deltanext +
3383 (minnext + deltanext) * maxcount - minnext * mincount;
3384 if (mincount != maxcount) {
3385 /* Cannot extend fixed substrings found inside
3387 SCAN_COMMIT(pRExC_state,data,minlenp);
3388 if (mincount && last_str) {
3389 SV * const sv = data->last_found;
3390 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3391 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3395 sv_setsv(sv, last_str);
3396 data->last_end = data->pos_min;
3397 data->last_start_min =
3398 data->pos_min - CHR_SVLEN(last_str);
3399 data->last_start_max = is_inf
3401 : data->pos_min + data->pos_delta
3402 - CHR_SVLEN(last_str);
3404 data->longest = &(data->longest_float);
3406 SvREFCNT_dec(last_str);
3408 if (data && (fl & SF_HAS_EVAL))
3409 data->flags |= SF_HAS_EVAL;
3410 optimize_curly_tail:
3411 if (OP(oscan) != CURLYX) {
3412 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3414 NEXT_OFF(oscan) += NEXT_OFF(next);
3417 default: /* REF and CLUMP only? */
3418 if (flags & SCF_DO_SUBSTR) {
3419 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3420 data->longest = &(data->longest_float);
3422 is_inf = is_inf_internal = 1;
3423 if (flags & SCF_DO_STCLASS_OR)
3424 cl_anything(pRExC_state, data->start_class);
3425 flags &= ~SCF_DO_STCLASS;
3429 else if (OP(scan) == LNBREAK) {
3430 if (flags & SCF_DO_STCLASS) {
3432 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3433 if (flags & SCF_DO_STCLASS_AND) {
3434 for (value = 0; value < 256; value++)
3435 if (!is_VERTWS_cp(value))
3436 ANYOF_BITMAP_CLEAR(data->start_class, value);
3439 for (value = 0; value < 256; value++)
3440 if (is_VERTWS_cp(value))
3441 ANYOF_BITMAP_SET(data->start_class, value);
3443 if (flags & SCF_DO_STCLASS_OR)
3444 cl_and(data->start_class, and_withp);
3445 flags &= ~SCF_DO_STCLASS;
3449 if (flags & SCF_DO_SUBSTR) {
3450 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3452 data->pos_delta += 1;
3453 data->longest = &(data->longest_float);
3457 else if (OP(scan) == FOLDCHAR) {
3458 int d = ARG(scan)==0xDF ? 1 : 2;
3459 flags &= ~SCF_DO_STCLASS;
3462 if (flags & SCF_DO_SUBSTR) {
3463 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3465 data->pos_delta += d;
3466 data->longest = &(data->longest_float);
3469 else if (strchr((const char*)PL_simple,OP(scan))) {
3472 if (flags & SCF_DO_SUBSTR) {
3473 SCAN_COMMIT(pRExC_state,data,minlenp);
3477 if (flags & SCF_DO_STCLASS) {
3478 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3480 /* Some of the logic below assumes that switching
3481 locale on will only add false positives. */
3482 switch (PL_regkind[OP(scan)]) {
3486 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3487 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3488 cl_anything(pRExC_state, data->start_class);
3491 if (OP(scan) == SANY)
3493 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3494 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3495 || (data->start_class->flags & ANYOF_CLASS));
3496 cl_anything(pRExC_state, data->start_class);
3498 if (flags & SCF_DO_STCLASS_AND || !value)
3499 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3502 if (flags & SCF_DO_STCLASS_AND)
3503 cl_and(data->start_class,
3504 (struct regnode_charclass_class*)scan);
3506 cl_or(pRExC_state, data->start_class,
3507 (struct regnode_charclass_class*)scan);
3510 if (flags & SCF_DO_STCLASS_AND) {
3511 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3512 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3513 for (value = 0; value < 256; value++)
3514 if (!isALNUM(value))
3515 ANYOF_BITMAP_CLEAR(data->start_class, value);
3519 if (data->start_class->flags & ANYOF_LOCALE)
3520 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3522 for (value = 0; value < 256; value++)
3524 ANYOF_BITMAP_SET(data->start_class, value);
3529 if (flags & SCF_DO_STCLASS_AND) {
3530 if (data->start_class->flags & ANYOF_LOCALE)
3531 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3534 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3535 data->start_class->flags |= ANYOF_LOCALE;
3539 if (flags & SCF_DO_STCLASS_AND) {
3540 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3541 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3542 for (value = 0; value < 256; value++)
3544 ANYOF_BITMAP_CLEAR(data->start_class, value);
3548 if (data->start_class->flags & ANYOF_LOCALE)
3549 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3551 for (value = 0; value < 256; value++)
3552 if (!isALNUM(value))
3553 ANYOF_BITMAP_SET(data->start_class, value);
3558 if (flags & SCF_DO_STCLASS_AND) {
3559 if (data->start_class->flags & ANYOF_LOCALE)
3560 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3563 data->start_class->flags |= ANYOF_LOCALE;
3564 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3568 if (flags & SCF_DO_STCLASS_AND) {
3569 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3570 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3571 for (value = 0; value < 256; value++)
3572 if (!isSPACE(value))
3573 ANYOF_BITMAP_CLEAR(data->start_class, value);
3577 if (data->start_class->flags & ANYOF_LOCALE)
3578 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3580 for (value = 0; value < 256; value++)
3582 ANYOF_BITMAP_SET(data->start_class, value);
3587 if (flags & SCF_DO_STCLASS_AND) {
3588 if (data->start_class->flags & ANYOF_LOCALE)
3589 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3592 data->start_class->flags |= ANYOF_LOCALE;
3593 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3597 if (flags & SCF_DO_STCLASS_AND) {
3598 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3599 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3600 for (value = 0; value < 256; value++)
3602 ANYOF_BITMAP_CLEAR(data->start_class, value);
3606 if (data->start_class->flags & ANYOF_LOCALE)
3607 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3609 for (value = 0; value < 256; value++)
3610 if (!isSPACE(value))
3611 ANYOF_BITMAP_SET(data->start_class, value);
3616 if (flags & SCF_DO_STCLASS_AND) {
3617 if (data->start_class->flags & ANYOF_LOCALE) {
3618 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3619 for (value = 0; value < 256; value++)
3620 if (!isSPACE(value))
3621 ANYOF_BITMAP_CLEAR(data->start_class, value);
3625 data->start_class->flags |= ANYOF_LOCALE;
3626 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3630 if (flags & SCF_DO_STCLASS_AND) {
3631 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3632 for (value = 0; value < 256; value++)
3633 if (!isDIGIT(value))
3634 ANYOF_BITMAP_CLEAR(data->start_class, value);
3637 if (data->start_class->flags & ANYOF_LOCALE)
3638 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3640 for (value = 0; value < 256; value++)
3642 ANYOF_BITMAP_SET(data->start_class, value);
3647 if (flags & SCF_DO_STCLASS_AND) {
3648 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3649 for (value = 0; value < 256; value++)
3651 ANYOF_BITMAP_CLEAR(data->start_class, value);
3654 if (data->start_class->flags & ANYOF_LOCALE)
3655 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3657 for (value = 0; value < 256; value++)
3658 if (!isDIGIT(value))
3659 ANYOF_BITMAP_SET(data->start_class, value);
3663 CASE_SYNST_FNC(VERTWS);
3664 CASE_SYNST_FNC(HORIZWS);
3667 if (flags & SCF_DO_STCLASS_OR)
3668 cl_and(data->start_class, and_withp);
3669 flags &= ~SCF_DO_STCLASS;
3672 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3673 data->flags |= (OP(scan) == MEOL
3677 else if ( PL_regkind[OP(scan)] == BRANCHJ
3678 /* Lookbehind, or need to calculate parens/evals/stclass: */
3679 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3680 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3681 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3682 || OP(scan) == UNLESSM )
3684 /* Negative Lookahead/lookbehind
3685 In this case we can't do fixed string optimisation.
3688 I32 deltanext, minnext, fake = 0;
3690 struct regnode_charclass_class intrnl;
3693 data_fake.flags = 0;
3695 data_fake.whilem_c = data->whilem_c;
3696 data_fake.last_closep = data->last_closep;
3699 data_fake.last_closep = &fake;
3700 data_fake.pos_delta = delta;
3701 if ( flags & SCF_DO_STCLASS && !scan->flags
3702 && OP(scan) == IFMATCH ) { /* Lookahead */
3703 cl_init(pRExC_state, &intrnl);
3704 data_fake.start_class = &intrnl;
3705 f |= SCF_DO_STCLASS_AND;
3707 if (flags & SCF_WHILEM_VISITED_POS)
3708 f |= SCF_WHILEM_VISITED_POS;
3709 next = regnext(scan);
3710 nscan = NEXTOPER(NEXTOPER(scan));
3711 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3712 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3715 FAIL("Variable length lookbehind not implemented");
3717 else if (minnext > (I32)U8_MAX) {
3718 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3720 scan->flags = (U8)minnext;
3723 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3725 if (data_fake.flags & SF_HAS_EVAL)
3726 data->flags |= SF_HAS_EVAL;
3727 data->whilem_c = data_fake.whilem_c;
3729 if (f & SCF_DO_STCLASS_AND) {
3730 if (flags & SCF_DO_STCLASS_OR) {
3731 /* OR before, AND after: ideally we would recurse with
3732 * data_fake to get the AND applied by study of the
3733 * remainder of the pattern, and then derecurse;
3734 * *** HACK *** for now just treat as "no information".
3735 * See [perl #56690].
3737 cl_init(pRExC_state, data->start_class);
3739 /* AND before and after: combine and continue */
3740 const int was = (data->start_class->flags & ANYOF_EOS);
3742 cl_and(data->start_class, &intrnl);
3744 data->start_class->flags |= ANYOF_EOS;
3748 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3750 /* Positive Lookahead/lookbehind
3751 In this case we can do fixed string optimisation,
3752 but we must be careful about it. Note in the case of
3753 lookbehind the positions will be offset by the minimum
3754 length of the pattern, something we won't know about
3755 until after the recurse.
3757 I32 deltanext, fake = 0;
3759 struct regnode_charclass_class intrnl;
3761 /* We use SAVEFREEPV so that when the full compile
3762 is finished perl will clean up the allocated
3763 minlens when its all done. This was we don't
3764 have to worry about freeing them when we know
3765 they wont be used, which would be a pain.
3768 Newx( minnextp, 1, I32 );
3769 SAVEFREEPV(minnextp);
3772 StructCopy(data, &data_fake, scan_data_t);
3773 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3776 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3777 data_fake.last_found=newSVsv(data->last_found);
3781 data_fake.last_closep = &fake;
3782 data_fake.flags = 0;
3783 data_fake.pos_delta = delta;
3785 data_fake.flags |= SF_IS_INF;
3786 if ( flags & SCF_DO_STCLASS && !scan->flags
3787 && OP(scan) == IFMATCH ) { /* Lookahead */
3788 cl_init(pRExC_state, &intrnl);
3789 data_fake.start_class = &intrnl;
3790 f |= SCF_DO_STCLASS_AND;
3792 if (flags & SCF_WHILEM_VISITED_POS)
3793 f |= SCF_WHILEM_VISITED_POS;
3794 next = regnext(scan);
3795 nscan = NEXTOPER(NEXTOPER(scan));
3797 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3798 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3801 FAIL("Variable length lookbehind not implemented");
3803 else if (*minnextp > (I32)U8_MAX) {
3804 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3806 scan->flags = (U8)*minnextp;
3811 if (f & SCF_DO_STCLASS_AND) {
3812 const int was = (data->start_class->flags & ANYOF_EOS);
3814 cl_and(data->start_class, &intrnl);
3816 data->start_class->flags |= ANYOF_EOS;
3819 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3821 if (data_fake.flags & SF_HAS_EVAL)
3822 data->flags |= SF_HAS_EVAL;
3823 data->whilem_c = data_fake.whilem_c;
3824 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3825 if (RExC_rx->minlen<*minnextp)
3826 RExC_rx->minlen=*minnextp;
3827 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3828 SvREFCNT_dec(data_fake.last_found);
3830 if ( data_fake.minlen_fixed != minlenp )
3832 data->offset_fixed= data_fake.offset_fixed;
3833 data->minlen_fixed= data_fake.minlen_fixed;
3834 data->lookbehind_fixed+= scan->flags;
3836 if ( data_fake.minlen_float != minlenp )
3838 data->minlen_float= data_fake.minlen_float;
3839 data->offset_float_min=data_fake.offset_float_min;
3840 data->offset_float_max=data_fake.offset_float_max;
3841 data->lookbehind_float+= scan->flags;
3850 else if (OP(scan) == OPEN) {
3851 if (stopparen != (I32)ARG(scan))
3854 else if (OP(scan) == CLOSE) {
3855 if (stopparen == (I32)ARG(scan)) {
3858 if ((I32)ARG(scan) == is_par) {
3859 next = regnext(scan);
3861 if ( next && (OP(next) != WHILEM) && next < last)
3862 is_par = 0; /* Disable optimization */
3865 *(data->last_closep) = ARG(scan);
3867 else if (OP(scan) == EVAL) {
3869 data->flags |= SF_HAS_EVAL;
3871 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3872 if (flags & SCF_DO_SUBSTR) {
3873 SCAN_COMMIT(pRExC_state,data,minlenp);
3874 flags &= ~SCF_DO_SUBSTR;
3876 if (data && OP(scan)==ACCEPT) {
3877 data->flags |= SCF_SEEN_ACCEPT;
3882 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3884 if (flags & SCF_DO_SUBSTR) {
3885 SCAN_COMMIT(pRExC_state,data,minlenp);
3886 data->longest = &(data->longest_float);
3888 is_inf = is_inf_internal = 1;
3889 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3890 cl_anything(pRExC_state, data->start_class);
3891 flags &= ~SCF_DO_STCLASS;
3893 else if (OP(scan) == GPOS) {
3894 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3895 !(delta || is_inf || (data && data->pos_delta)))
3897 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3898 RExC_rx->extflags |= RXf_ANCH_GPOS;
3899 if (RExC_rx->gofs < (U32)min)
3900 RExC_rx->gofs = min;
3902 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3906 #ifdef TRIE_STUDY_OPT
3907 #ifdef FULL_TRIE_STUDY
3908 else if (PL_regkind[OP(scan)] == TRIE) {
3909 /* NOTE - There is similar code to this block above for handling
3910 BRANCH nodes on the initial study. If you change stuff here
3912 regnode *trie_node= scan;
3913 regnode *tail= regnext(scan);
3914 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3915 I32 max1 = 0, min1 = I32_MAX;
3916 struct regnode_charclass_class accum;
3918 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3919 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3920 if (flags & SCF_DO_STCLASS)
3921 cl_init_zero(pRExC_state, &accum);
3927 const regnode *nextbranch= NULL;
3930 for ( word=1 ; word <= trie->wordcount ; word++)
3932 I32 deltanext=0, minnext=0, f = 0, fake;
3933 struct regnode_charclass_class this_class;
3935 data_fake.flags = 0;
3937 data_fake.whilem_c = data->whilem_c;
3938 data_fake.last_closep = data->last_closep;
3941 data_fake.last_closep = &fake;
3942 data_fake.pos_delta = delta;
3943 if (flags & SCF_DO_STCLASS) {
3944 cl_init(pRExC_state, &this_class);
3945 data_fake.start_class = &this_class;
3946 f = SCF_DO_STCLASS_AND;
3948 if (flags & SCF_WHILEM_VISITED_POS)
3949 f |= SCF_WHILEM_VISITED_POS;
3951 if (trie->jump[word]) {
3953 nextbranch = trie_node + trie->jump[0];
3954 scan= trie_node + trie->jump[word];
3955 /* We go from the jump point to the branch that follows
3956 it. Note this means we need the vestigal unused branches
3957 even though they arent otherwise used.
3959 minnext = study_chunk(pRExC_state, &scan, minlenp,
3960 &deltanext, (regnode *)nextbranch, &data_fake,
3961 stopparen, recursed, NULL, f,depth+1);
3963 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
3964 nextbranch= regnext((regnode*)nextbranch);
3966 if (min1 > (I32)(minnext + trie->minlen))
3967 min1 = minnext + trie->minlen;
3968 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
3969 max1 = minnext + deltanext + trie->maxlen;
3970 if (deltanext == I32_MAX)
3971 is_inf = is_inf_internal = 1;
3973 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3975 if (data_fake.flags & SCF_SEEN_ACCEPT) {
3976 if ( stopmin > min + min1)
3977 stopmin = min + min1;
3978 flags &= ~SCF_DO_SUBSTR;
3980 data->flags |= SCF_SEEN_ACCEPT;
3983 if (data_fake.flags & SF_HAS_EVAL)
3984 data->flags |= SF_HAS_EVAL;
3985 data->whilem_c = data_fake.whilem_c;
3987 if (flags & SCF_DO_STCLASS)
3988 cl_or(pRExC_state, &accum, &this_class);
3991 if (flags & SCF_DO_SUBSTR) {
3992 data->pos_min += min1;
3993 data->pos_delta += max1 - min1;
3994 if (max1 != min1 || is_inf)
3995 data->longest = &(data->longest_float);
3998 delta += max1 - min1;
3999 if (flags & SCF_DO_STCLASS_OR) {
4000 cl_or(pRExC_state, data->start_class, &accum);
4002 cl_and(data->start_class, and_withp);
4003 flags &= ~SCF_DO_STCLASS;
4006 else if (flags & SCF_DO_STCLASS_AND) {
4008 cl_and(data->start_class, &accum);
4009 flags &= ~SCF_DO_STCLASS;
4012 /* Switch to OR mode: cache the old value of
4013 * data->start_class */
4015 StructCopy(data->start_class, and_withp,
4016 struct regnode_charclass_class);
4017 flags &= ~SCF_DO_STCLASS_AND;
4018 StructCopy(&accum, data->start_class,
4019 struct regnode_charclass_class);
4020 flags |= SCF_DO_STCLASS_OR;
4021 data->start_class->flags |= ANYOF_EOS;
4028 else if (PL_regkind[OP(scan)] == TRIE) {
4029 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4032 min += trie->minlen;
4033 delta += (trie->maxlen - trie->minlen);
4034 flags &= ~SCF_DO_STCLASS; /* xxx */
4035 if (flags & SCF_DO_SUBSTR) {
4036 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4037 data->pos_min += trie->minlen;
4038 data->pos_delta += (trie->maxlen - trie->minlen);
4039 if (trie->maxlen != trie->minlen)
4040 data->longest = &(data->longest_float);
4042 if (trie->jump) /* no more substrings -- for now /grr*/
4043 flags &= ~SCF_DO_SUBSTR;
4045 #endif /* old or new */
4046 #endif /* TRIE_STUDY_OPT */
4048 /* Else: zero-length, ignore. */
4049 scan = regnext(scan);
4054 stopparen = frame->stop;
4055 frame = frame->prev;
4056 goto fake_study_recurse;
4061 DEBUG_STUDYDATA("pre-fin:",data,depth);
4064 *deltap = is_inf_internal ? I32_MAX : delta;
4065 if (flags & SCF_DO_SUBSTR && is_inf)
4066 data->pos_delta = I32_MAX - data->pos_min;
4067 if (is_par > (I32)U8_MAX)
4069 if (is_par && pars==1 && data) {
4070 data->flags |= SF_IN_PAR;
4071 data->flags &= ~SF_HAS_PAR;
4073 else if (pars && data) {
4074 data->flags |= SF_HAS_PAR;
4075 data->flags &= ~SF_IN_PAR;
4077 if (flags & SCF_DO_STCLASS_OR)
4078 cl_and(data->start_class, and_withp);
4079 if (flags & SCF_TRIE_RESTUDY)
4080 data->flags |= SCF_TRIE_RESTUDY;
4082 DEBUG_STUDYDATA("post-fin:",data,depth);
4084 return min < stopmin ? min : stopmin;
4088 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4090 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4092 PERL_ARGS_ASSERT_ADD_DATA;
4094 Renewc(RExC_rxi->data,
4095 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4096 char, struct reg_data);
4098 Renew(RExC_rxi->data->what, count + n, U8);
4100 Newx(RExC_rxi->data->what, n, U8);
4101 RExC_rxi->data->count = count + n;
4102 Copy(s, RExC_rxi->data->what + count, n, U8);
4106 /*XXX: todo make this not included in a non debugging perl */
4107 #ifndef PERL_IN_XSUB_RE
4109 Perl_reginitcolors(pTHX)
4112 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4114 char *t = savepv(s);
4118 t = strchr(t, '\t');
4124 PL_colors[i] = t = (char *)"";
4129 PL_colors[i++] = (char *)"";
4136 #ifdef TRIE_STUDY_OPT
4137 #define CHECK_RESTUDY_GOTO \
4139 (data.flags & SCF_TRIE_RESTUDY) \
4143 #define CHECK_RESTUDY_GOTO
4147 - pregcomp - compile a regular expression into internal code
4149 * We can't allocate space until we know how big the compiled form will be,
4150 * but we can't compile it (and thus know how big it is) until we've got a
4151 * place to put the code. So we cheat: we compile it twice, once with code
4152 * generation turned off and size counting turned on, and once "for real".
4153 * This also means that we don't allocate space until we are sure that the
4154 * thing really will compile successfully, and we never have to move the
4155 * code and thus invalidate pointers into it. (Note that it has to be in
4156 * one piece because free() must be able to free it all.) [NB: not true in perl]
4158 * Beware that the optimization-preparation code in here knows about some
4159 * of the structure of the compiled regexp. [I'll say.]
4164 #ifndef PERL_IN_XSUB_RE
4165 #define RE_ENGINE_PTR &PL_core_reg_engine
4167 extern const struct regexp_engine my_reg_engine;
4168 #define RE_ENGINE_PTR &my_reg_engine
4171 #ifndef PERL_IN_XSUB_RE
4173 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4176 HV * const table = GvHV(PL_hintgv);
4178 PERL_ARGS_ASSERT_PREGCOMP;
4180 /* Dispatch a request to compile a regexp to correct
4183 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4184 GET_RE_DEBUG_FLAGS_DECL;
4185 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4186 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4188 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4191 return CALLREGCOMP_ENG(eng, pattern, flags);
4194 return Perl_re_compile(aTHX_ pattern, flags);
4199 Perl_re_compile(pTHX_ SV * const pattern, U32 pm_flags)
4204 register regexp_internal *ri;
4206 char *exp = SvPV(pattern, plen);
4207 char* xend = exp + plen;
4214 RExC_state_t RExC_state;
4215 RExC_state_t * const pRExC_state = &RExC_state;
4216 #ifdef TRIE_STUDY_OPT
4218 RExC_state_t copyRExC_state;
4220 GET_RE_DEBUG_FLAGS_DECL;
4222 PERL_ARGS_ASSERT_RE_COMPILE;
4224 DEBUG_r(if (!PL_colorset) reginitcolors());
4226 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4229 SV *dsv= sv_newmortal();
4230 RE_PV_QUOTED_DECL(s, RExC_utf8,
4231 dsv, exp, plen, 60);
4232 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4233 PL_colors[4],PL_colors[5],s);
4238 RExC_flags = pm_flags;
4242 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4243 RExC_seen_evals = 0;
4246 /* First pass: determine size, legality. */
4254 RExC_emit = &PL_regdummy;
4255 RExC_whilem_seen = 0;
4256 RExC_charnames = NULL;
4257 RExC_open_parens = NULL;
4258 RExC_close_parens = NULL;
4260 RExC_paren_names = NULL;
4262 RExC_paren_name_list = NULL;
4264 RExC_recurse = NULL;
4265 RExC_recurse_count = 0;
4267 #if 0 /* REGC() is (currently) a NOP at the first pass.
4268 * Clever compilers notice this and complain. --jhi */
4269 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4271 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4272 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4273 RExC_precomp = NULL;
4276 if (RExC_utf8 && !RExC_orig_utf8) {
4277 /* It's possible to write a regexp in ascii that represents Unicode
4278 codepoints outside of the byte range, such as via \x{100}. If we
4279 detect such a sequence we have to convert the entire pattern to utf8
4280 and then recompile, as our sizing calculation will have been based
4281 on 1 byte == 1 character, but we will need to use utf8 to encode
4282 at least some part of the pattern, and therefore must convert the whole
4284 XXX: somehow figure out how to make this less expensive...
4287 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4288 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4289 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4291 RExC_orig_utf8 = RExC_utf8;
4293 goto redo_first_pass;
4296 PerlIO_printf(Perl_debug_log,
4297 "Required size %"IVdf" nodes\n"
4298 "Starting second pass (creation)\n",
4301 RExC_lastparse=NULL;
4303 /* Small enough for pointer-storage convention?
4304 If extralen==0, this means that we will not need long jumps. */
4305 if (RExC_size >= 0x10000L && RExC_extralen)
4306 RExC_size += RExC_extralen;
4309 if (RExC_whilem_seen > 15)
4310 RExC_whilem_seen = 15;
4312 /* Allocate space and zero-initialize. Note, the two step process
4313 of zeroing when in debug mode, thus anything assigned has to
4314 happen after that */
4315 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4316 r = (struct regexp*)SvANY(rx);
4317 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4318 char, regexp_internal);
4319 if ( r == NULL || ri == NULL )
4320 FAIL("Regexp out of space");
4322 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4323 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4325 /* bulk initialize base fields with 0. */
4326 Zero(ri, sizeof(regexp_internal), char);
4329 /* non-zero initialization begins here */
4331 r->engine= RE_ENGINE_PTR;
4332 r->extflags = pm_flags;
4334 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4335 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4336 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4337 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4338 >> RXf_PMf_STD_PMMOD_SHIFT);
4339 const char *fptr = STD_PAT_MODS; /*"msix"*/
4341 const STRLEN wraplen = plen + has_minus + has_p + has_runon
4342 + (sizeof(STD_PAT_MODS) - 1)
4343 + (sizeof("(?:)") - 1);
4345 p = sv_grow(MUTABLE_SV(rx), wraplen + 1);
4346 SvCUR_set(rx, wraplen);
4348 SvFLAGS(rx) |= SvUTF8(pattern);
4351 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4353 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4354 char *colon = r + 1;
4357 while((ch = *fptr++)) {
4371 Copy(RExC_precomp, p, plen, char);
4372 assert ((RX_WRAPPED(rx) - p) < 16);
4373 r->pre_prefix = p - RX_WRAPPED(rx);
4382 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4384 if (RExC_seen & REG_SEEN_RECURSE) {
4385 Newxz(RExC_open_parens, RExC_npar,regnode *);
4386 SAVEFREEPV(RExC_open_parens);
4387 Newxz(RExC_close_parens,RExC_npar,regnode *);
4388 SAVEFREEPV(RExC_close_parens);
4391 /* Useful during FAIL. */
4392 #ifdef RE_TRACK_PATTERN_OFFSETS
4393 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4394 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4395 "%s %"UVuf" bytes for offset annotations.\n",
4396 ri->u.offsets ? "Got" : "Couldn't get",
4397 (UV)((2*RExC_size+1) * sizeof(U32))));
4399 SetProgLen(ri,RExC_size);
4404 /* Second pass: emit code. */
4405 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4410 RExC_emit_start = ri->program;
4411 RExC_emit = ri->program;
4412 RExC_emit_bound = ri->program + RExC_size + 1;
4414 /* Store the count of eval-groups for security checks: */
4415 RExC_rx->seen_evals = RExC_seen_evals;
4416 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4417 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4421 /* XXXX To minimize changes to RE engine we always allocate
4422 3-units-long substrs field. */
4423 Newx(r->substrs, 1, struct reg_substr_data);
4424 if (RExC_recurse_count) {
4425 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4426 SAVEFREEPV(RExC_recurse);
4430 r->minlen = minlen = sawplus = sawopen = 0;
4431 Zero(r->substrs, 1, struct reg_substr_data);
4433 #ifdef TRIE_STUDY_OPT
4435 StructCopy(&zero_scan_data, &data, scan_data_t);
4436 copyRExC_state = RExC_state;
4439 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4441 RExC_state = copyRExC_state;
4442 if (seen & REG_TOP_LEVEL_BRANCHES)
4443 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4445 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4446 if (data.last_found) {
4447 SvREFCNT_dec(data.longest_fixed);
4448 SvREFCNT_dec(data.longest_float);
4449 SvREFCNT_dec(data.last_found);
4451 StructCopy(&zero_scan_data, &data, scan_data_t);
4454 StructCopy(&zero_scan_data, &data, scan_data_t);
4457 /* Dig out information for optimizations. */
4458 r->extflags = RExC_flags; /* was pm_op */
4459 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4462 SvUTF8_on(rx); /* Unicode in it? */
4463 ri->regstclass = NULL;
4464 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4465 r->intflags |= PREGf_NAUGHTY;
4466 scan = ri->program + 1; /* First BRANCH. */
4468 /* testing for BRANCH here tells us whether there is "must appear"
4469 data in the pattern. If there is then we can use it for optimisations */
4470 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4472 STRLEN longest_float_length, longest_fixed_length;
4473 struct regnode_charclass_class ch_class; /* pointed to by data */
4475 I32 last_close = 0; /* pointed to by data */
4476 regnode *first= scan;
4477 regnode *first_next= regnext(first);
4480 * Skip introductions and multiplicators >= 1
4481 * so that we can extract the 'meat' of the pattern that must
4482 * match in the large if() sequence following.
4483 * NOTE that EXACT is NOT covered here, as it is normally
4484 * picked up by the optimiser separately.
4486 * This is unfortunate as the optimiser isnt handling lookahead
4487 * properly currently.
4490 while ((OP(first) == OPEN && (sawopen = 1)) ||
4491 /* An OR of *one* alternative - should not happen now. */
4492 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4493 /* for now we can't handle lookbehind IFMATCH*/
4494 (OP(first) == IFMATCH && !first->flags) ||
4495 (OP(first) == PLUS) ||
4496 (OP(first) == MINMOD) ||
4497 /* An {n,m} with n>0 */
4498 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4499 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4502 * the only op that could be a regnode is PLUS, all the rest
4503 * will be regnode_1 or regnode_2.
4506 if (OP(first) == PLUS)
4509 first += regarglen[OP(first)];
4511 first = NEXTOPER(first);
4512 first_next= regnext(first);
4515 /* Starting-point info. */
4517 DEBUG_PEEP("first:",first,0);
4518 /* Ignore EXACT as we deal with it later. */
4519 if (PL_regkind[OP(first)] == EXACT) {
4520 if (OP(first) == EXACT)
4521 NOOP; /* Empty, get anchored substr later. */
4522 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4523 ri->regstclass = first;
4526 else if (PL_regkind[OP(first)] == TRIE &&
4527 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4530 /* this can happen only on restudy */
4531 if ( OP(first) == TRIE ) {
4532 struct regnode_1 *trieop = (struct regnode_1 *)
4533 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4534 StructCopy(first,trieop,struct regnode_1);
4535 trie_op=(regnode *)trieop;
4537 struct regnode_charclass *trieop = (struct regnode_charclass *)
4538 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4539 StructCopy(first,trieop,struct regnode_charclass);
4540 trie_op=(regnode *)trieop;
4543 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4544 ri->regstclass = trie_op;
4547 else if (strchr((const char*)PL_simple,OP(first)))
4548 ri->regstclass = first;
4549 else if (PL_regkind[OP(first)] == BOUND ||
4550 PL_regkind[OP(first)] == NBOUND)
4551 ri->regstclass = first;
4552 else if (PL_regkind[OP(first)] == BOL) {
4553 r->extflags |= (OP(first) == MBOL
4555 : (OP(first) == SBOL
4558 first = NEXTOPER(first);
4561 else if (OP(first) == GPOS) {
4562 r->extflags |= RXf_ANCH_GPOS;
4563 first = NEXTOPER(first);
4566 else if ((!sawopen || !RExC_sawback) &&
4567 (OP(first) == STAR &&
4568 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4569 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4571 /* turn .* into ^.* with an implied $*=1 */
4573 (OP(NEXTOPER(first)) == REG_ANY)
4576 r->extflags |= type;
4577 r->intflags |= PREGf_IMPLICIT;
4578 first = NEXTOPER(first);
4581 if (sawplus && (!sawopen || !RExC_sawback)
4582 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4583 /* x+ must match at the 1st pos of run of x's */
4584 r->intflags |= PREGf_SKIP;
4586 /* Scan is after the zeroth branch, first is atomic matcher. */
4587 #ifdef TRIE_STUDY_OPT
4590 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4591 (IV)(first - scan + 1))
4595 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4596 (IV)(first - scan + 1))
4602 * If there's something expensive in the r.e., find the
4603 * longest literal string that must appear and make it the
4604 * regmust. Resolve ties in favor of later strings, since
4605 * the regstart check works with the beginning of the r.e.
4606 * and avoiding duplication strengthens checking. Not a
4607 * strong reason, but sufficient in the absence of others.
4608 * [Now we resolve ties in favor of the earlier string if
4609 * it happens that c_offset_min has been invalidated, since the
4610 * earlier string may buy us something the later one won't.]
4613 data.longest_fixed = newSVpvs("");
4614 data.longest_float = newSVpvs("");
4615 data.last_found = newSVpvs("");
4616 data.longest = &(data.longest_fixed);
4618 if (!ri->regstclass) {
4619 cl_init(pRExC_state, &ch_class);
4620 data.start_class = &ch_class;
4621 stclass_flag = SCF_DO_STCLASS_AND;
4622 } else /* XXXX Check for BOUND? */
4624 data.last_closep = &last_close;
4626 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4627 &data, -1, NULL, NULL,
4628 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4634 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4635 && data.last_start_min == 0 && data.last_end > 0
4636 && !RExC_seen_zerolen
4637 && !(RExC_seen & REG_SEEN_VERBARG)
4638 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4639 r->extflags |= RXf_CHECK_ALL;
4640 scan_commit(pRExC_state, &data,&minlen,0);
4641 SvREFCNT_dec(data.last_found);
4643 /* Note that code very similar to this but for anchored string
4644 follows immediately below, changes may need to be made to both.
4647 longest_float_length = CHR_SVLEN(data.longest_float);
4648 if (longest_float_length
4649 || (data.flags & SF_FL_BEFORE_EOL
4650 && (!(data.flags & SF_FL_BEFORE_MEOL)
4651 || (RExC_flags & RXf_PMf_MULTILINE))))
4655 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4656 && data.offset_fixed == data.offset_float_min
4657 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4658 goto remove_float; /* As in (a)+. */
4660 /* copy the information about the longest float from the reg_scan_data
4661 over to the program. */
4662 if (SvUTF8(data.longest_float)) {
4663 r->float_utf8 = data.longest_float;
4664 r->float_substr = NULL;
4666 r->float_substr = data.longest_float;
4667 r->float_utf8 = NULL;
4669 /* float_end_shift is how many chars that must be matched that
4670 follow this item. We calculate it ahead of time as once the
4671 lookbehind offset is added in we lose the ability to correctly
4673 ml = data.minlen_float ? *(data.minlen_float)
4674 : (I32)longest_float_length;
4675 r->float_end_shift = ml - data.offset_float_min
4676 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4677 + data.lookbehind_float;
4678 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4679 r->float_max_offset = data.offset_float_max;
4680 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4681 r->float_max_offset -= data.lookbehind_float;
4683 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4684 && (!(data.flags & SF_FL_BEFORE_MEOL)
4685 || (RExC_flags & RXf_PMf_MULTILINE)));
4686 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4690 r->float_substr = r->float_utf8 = NULL;
4691 SvREFCNT_dec(data.longest_float);
4692 longest_float_length = 0;
4695 /* Note that code very similar to this but for floating string
4696 is immediately above, changes may need to be made to both.
4699 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4700 if (longest_fixed_length
4701 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4702 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4703 || (RExC_flags & RXf_PMf_MULTILINE))))
4707 /* copy the information about the longest fixed
4708 from the reg_scan_data over to the program. */
4709 if (SvUTF8(data.longest_fixed)) {
4710 r->anchored_utf8 = data.longest_fixed;
4711 r->anchored_substr = NULL;
4713 r->anchored_substr = data.longest_fixed;
4714 r->anchored_utf8 = NULL;
4716 /* fixed_end_shift is how many chars that must be matched that
4717 follow this item. We calculate it ahead of time as once the
4718 lookbehind offset is added in we lose the ability to correctly
4720 ml = data.minlen_fixed ? *(data.minlen_fixed)
4721 : (I32)longest_fixed_length;
4722 r->anchored_end_shift = ml - data.offset_fixed
4723 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4724 + data.lookbehind_fixed;
4725 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4727 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4728 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4729 || (RExC_flags & RXf_PMf_MULTILINE)));
4730 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4733 r->anchored_substr = r->anchored_utf8 = NULL;
4734 SvREFCNT_dec(data.longest_fixed);
4735 longest_fixed_length = 0;
4738 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4739 ri->regstclass = NULL;
4740 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4742 && !(data.start_class->flags & ANYOF_EOS)
4743 && !cl_is_anything(data.start_class))
4745 const U32 n = add_data(pRExC_state, 1, "f");
4747 Newx(RExC_rxi->data->data[n], 1,
4748 struct regnode_charclass_class);
4749 StructCopy(data.start_class,
4750 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4751 struct regnode_charclass_class);
4752 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4753 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4754 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4755 regprop(r, sv, (regnode*)data.start_class);
4756 PerlIO_printf(Perl_debug_log,
4757 "synthetic stclass \"%s\".\n",
4758 SvPVX_const(sv));});
4761 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4762 if (longest_fixed_length > longest_float_length) {
4763 r->check_end_shift = r->anchored_end_shift;
4764 r->check_substr = r->anchored_substr;
4765 r->check_utf8 = r->anchored_utf8;
4766 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4767 if (r->extflags & RXf_ANCH_SINGLE)
4768 r->extflags |= RXf_NOSCAN;
4771 r->check_end_shift = r->float_end_shift;
4772 r->check_substr = r->float_substr;
4773 r->check_utf8 = r->float_utf8;
4774 r->check_offset_min = r->float_min_offset;
4775 r->check_offset_max = r->float_max_offset;
4777 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4778 This should be changed ASAP! */
4779 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4780 r->extflags |= RXf_USE_INTUIT;
4781 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4782 r->extflags |= RXf_INTUIT_TAIL;
4784 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4785 if ( (STRLEN)minlen < longest_float_length )
4786 minlen= longest_float_length;
4787 if ( (STRLEN)minlen < longest_fixed_length )
4788 minlen= longest_fixed_length;
4792 /* Several toplevels. Best we can is to set minlen. */
4794 struct regnode_charclass_class ch_class;
4797 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4799 scan = ri->program + 1;
4800 cl_init(pRExC_state, &ch_class);
4801 data.start_class = &ch_class;
4802 data.last_closep = &last_close;
4805 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4806 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4810 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4811 = r->float_substr = r->float_utf8 = NULL;
4812 if (!(data.start_class->flags & ANYOF_EOS)
4813 && !cl_is_anything(data.start_class))
4815 const U32 n = add_data(pRExC_state, 1, "f");
4817 Newx(RExC_rxi->data->data[n], 1,
4818 struct regnode_charclass_class);
4819 StructCopy(data.start_class,
4820 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4821 struct regnode_charclass_class);
4822 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4823 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4824 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4825 regprop(r, sv, (regnode*)data.start_class);
4826 PerlIO_printf(Perl_debug_log,
4827 "synthetic stclass \"%s\".\n",
4828 SvPVX_const(sv));});
4832 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4833 the "real" pattern. */
4835 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4836 (IV)minlen, (IV)r->minlen);
4838 r->minlenret = minlen;
4839 if (r->minlen < minlen)
4842 if (RExC_seen & REG_SEEN_GPOS)
4843 r->extflags |= RXf_GPOS_SEEN;
4844 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4845 r->extflags |= RXf_LOOKBEHIND_SEEN;
4846 if (RExC_seen & REG_SEEN_EVAL)
4847 r->extflags |= RXf_EVAL_SEEN;
4848 if (RExC_seen & REG_SEEN_CANY)
4849 r->extflags |= RXf_CANY_SEEN;
4850 if (RExC_seen & REG_SEEN_VERBARG)
4851 r->intflags |= PREGf_VERBARG_SEEN;
4852 if (RExC_seen & REG_SEEN_CUTGROUP)
4853 r->intflags |= PREGf_CUTGROUP_SEEN;
4854 if (RExC_paren_names)
4855 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
4857 RXp_PAREN_NAMES(r) = NULL;
4859 #ifdef STUPID_PATTERN_CHECKS
4860 if (RX_PRELEN(rx) == 0)
4861 r->extflags |= RXf_NULL;
4862 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4863 /* XXX: this should happen BEFORE we compile */
4864 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4865 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
4866 r->extflags |= RXf_WHITE;
4867 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
4868 r->extflags |= RXf_START_ONLY;
4870 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4871 /* XXX: this should happen BEFORE we compile */
4872 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4874 regnode *first = ri->program + 1;
4876 U8 nop = OP(NEXTOPER(first));
4878 if (PL_regkind[fop] == NOTHING && nop == END)
4879 r->extflags |= RXf_NULL;
4880 else if (PL_regkind[fop] == BOL && nop == END)
4881 r->extflags |= RXf_START_ONLY;
4882 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4883 r->extflags |= RXf_WHITE;
4887 if (RExC_paren_names) {
4888 ri->name_list_idx = add_data( pRExC_state, 1, "p" );
4889 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4892 ri->name_list_idx = 0;
4894 if (RExC_recurse_count) {
4895 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4896 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4897 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4900 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4901 /* assume we don't need to swap parens around before we match */
4904 PerlIO_printf(Perl_debug_log,"Final program:\n");
4907 #ifdef RE_TRACK_PATTERN_OFFSETS
4908 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4909 const U32 len = ri->u.offsets[0];
4911 GET_RE_DEBUG_FLAGS_DECL;
4912 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4913 for (i = 1; i <= len; i++) {
4914 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4915 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4916 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4918 PerlIO_printf(Perl_debug_log, "\n");
4924 #undef RE_ENGINE_PTR
4928 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4931 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
4933 PERL_UNUSED_ARG(value);
4935 if (flags & RXapif_FETCH) {
4936 return reg_named_buff_fetch(rx, key, flags);
4937 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
4938 Perl_croak(aTHX_ "%s", PL_no_modify);
4940 } else if (flags & RXapif_EXISTS) {
4941 return reg_named_buff_exists(rx, key, flags)
4944 } else if (flags & RXapif_REGNAMES) {
4945 return reg_named_buff_all(rx, flags);
4946 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
4947 return reg_named_buff_scalar(rx, flags);
4949 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
4955 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
4958 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
4959 PERL_UNUSED_ARG(lastkey);
4961 if (flags & RXapif_FIRSTKEY)
4962 return reg_named_buff_firstkey(rx, flags);
4963 else if (flags & RXapif_NEXTKEY)
4964 return reg_named_buff_nextkey(rx, flags);
4966 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
4972 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
4975 AV *retarray = NULL;
4977 struct regexp *const rx = (struct regexp *)SvANY(r);
4979 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
4981 if (flags & RXapif_ALL)
4984 if (rx && RXp_PAREN_NAMES(rx)) {
4985 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
4988 SV* sv_dat=HeVAL(he_str);
4989 I32 *nums=(I32*)SvPVX(sv_dat);
4990 for ( i=0; i<SvIVX(sv_dat); i++ ) {
4991 if ((I32)(rx->nparens) >= nums[i]
4992 && rx->offs[nums[i]].start != -1
4993 && rx->offs[nums[i]].end != -1)
4996 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5000 ret = newSVsv(&PL_sv_undef);
5003 av_push(retarray, ret);
5006 return newRV_noinc(MUTABLE_SV(retarray));
5013 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5016 struct regexp *const rx = (struct regexp *)SvANY(r);
5018 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5020 if (rx && RXp_PAREN_NAMES(rx)) {
5021 if (flags & RXapif_ALL) {
5022 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5024 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5038 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5040 struct regexp *const rx = (struct regexp *)SvANY(r);
5042 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5044 if ( rx && RXp_PAREN_NAMES(rx) ) {
5045 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5047 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5054 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5056 struct regexp *const rx = (struct regexp *)SvANY(r);
5057 GET_RE_DEBUG_FLAGS_DECL;
5059 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5061 if (rx && RXp_PAREN_NAMES(rx)) {
5062 HV *hv = RXp_PAREN_NAMES(rx);
5064 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5067 SV* sv_dat = HeVAL(temphe);
5068 I32 *nums = (I32*)SvPVX(sv_dat);
5069 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5070 if ((I32)(rx->lastparen) >= nums[i] &&
5071 rx->offs[nums[i]].start != -1 &&
5072 rx->offs[nums[i]].end != -1)
5078 if (parno || flags & RXapif_ALL) {
5079 return newSVhek(HeKEY_hek(temphe));
5087 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5092 struct regexp *const rx = (struct regexp *)SvANY(r);
5094 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5096 if (rx && RXp_PAREN_NAMES(rx)) {
5097 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5098 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5099 } else if (flags & RXapif_ONE) {
5100 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5101 av = MUTABLE_AV(SvRV(ret));
5102 length = av_len(av);
5104 return newSViv(length + 1);
5106 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5110 return &PL_sv_undef;
5114 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5116 struct regexp *const rx = (struct regexp *)SvANY(r);
5119 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5121 if (rx && RXp_PAREN_NAMES(rx)) {
5122 HV *hv= RXp_PAREN_NAMES(rx);
5124 (void)hv_iterinit(hv);
5125 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5128 SV* sv_dat = HeVAL(temphe);
5129 I32 *nums = (I32*)SvPVX(sv_dat);
5130 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5131 if ((I32)(rx->lastparen) >= nums[i] &&
5132 rx->offs[nums[i]].start != -1 &&
5133 rx->offs[nums[i]].end != -1)
5139 if (parno || flags & RXapif_ALL) {
5140 av_push(av, newSVhek(HeKEY_hek(temphe)));
5145 return newRV_noinc(MUTABLE_SV(av));
5149 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5152 struct regexp *const rx = (struct regexp *)SvANY(r);
5157 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5160 sv_setsv(sv,&PL_sv_undef);
5164 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5166 i = rx->offs[0].start;
5170 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5172 s = rx->subbeg + rx->offs[0].end;
5173 i = rx->sublen - rx->offs[0].end;
5176 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5177 (s1 = rx->offs[paren].start) != -1 &&
5178 (t1 = rx->offs[paren].end) != -1)
5182 s = rx->subbeg + s1;
5184 sv_setsv(sv,&PL_sv_undef);
5187 assert(rx->sublen >= (s - rx->subbeg) + i );
5189 const int oldtainted = PL_tainted;
5191 sv_setpvn(sv, s, i);
5192 PL_tainted = oldtainted;
5193 if ( (rx->extflags & RXf_CANY_SEEN)
5194 ? (RXp_MATCH_UTF8(rx)
5195 && (!i || is_utf8_string((U8*)s, i)))
5196 : (RXp_MATCH_UTF8(rx)) )
5203 if (RXp_MATCH_TAINTED(rx)) {
5204 if (SvTYPE(sv) >= SVt_PVMG) {
5205 MAGIC* const mg = SvMAGIC(sv);
5208 SvMAGIC_set(sv, mg->mg_moremagic);
5210 if ((mgt = SvMAGIC(sv))) {
5211 mg->mg_moremagic = mgt;
5212 SvMAGIC_set(sv, mg);
5222 sv_setsv(sv,&PL_sv_undef);
5228 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5229 SV const * const value)
5231 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5233 PERL_UNUSED_ARG(rx);
5234 PERL_UNUSED_ARG(paren);
5235 PERL_UNUSED_ARG(value);
5238 Perl_croak(aTHX_ "%s", PL_no_modify);
5242 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5245 struct regexp *const rx = (struct regexp *)SvANY(r);
5249 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5251 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5253 /* $` / ${^PREMATCH} */
5254 case RX_BUFF_IDX_PREMATCH:
5255 if (rx->offs[0].start != -1) {
5256 i = rx->offs[0].start;
5264 /* $' / ${^POSTMATCH} */
5265 case RX_BUFF_IDX_POSTMATCH:
5266 if (rx->offs[0].end != -1) {
5267 i = rx->sublen - rx->offs[0].end;
5269 s1 = rx->offs[0].end;
5275 /* $& / ${^MATCH}, $1, $2, ... */
5277 if (paren <= (I32)rx->nparens &&
5278 (s1 = rx->offs[paren].start) != -1 &&
5279 (t1 = rx->offs[paren].end) != -1)
5284 if (ckWARN(WARN_UNINITIALIZED))
5285 report_uninit((const SV *)sv);
5290 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5291 const char * const s = rx->subbeg + s1;
5296 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5303 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5305 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5306 PERL_UNUSED_ARG(rx);
5310 return newSVpvs("Regexp");
5313 /* Scans the name of a named buffer from the pattern.
5314 * If flags is REG_RSN_RETURN_NULL returns null.
5315 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5316 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5317 * to the parsed name as looked up in the RExC_paren_names hash.
5318 * If there is an error throws a vFAIL().. type exception.
5321 #define REG_RSN_RETURN_NULL 0
5322 #define REG_RSN_RETURN_NAME 1
5323 #define REG_RSN_RETURN_DATA 2
5326 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5328 char *name_start = RExC_parse;
5330 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5332 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5333 /* skip IDFIRST by using do...while */
5336 RExC_parse += UTF8SKIP(RExC_parse);
5337 } while (isALNUM_utf8((U8*)RExC_parse));
5341 } while (isALNUM(*RExC_parse));
5346 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5347 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5348 if ( flags == REG_RSN_RETURN_NAME)
5350 else if (flags==REG_RSN_RETURN_DATA) {
5353 if ( ! sv_name ) /* should not happen*/
5354 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5355 if (RExC_paren_names)
5356 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5358 sv_dat = HeVAL(he_str);
5360 vFAIL("Reference to nonexistent named group");
5364 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5371 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5372 int rem=(int)(RExC_end - RExC_parse); \
5381 if (RExC_lastparse!=RExC_parse) \
5382 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5385 iscut ? "..." : "<" \
5388 PerlIO_printf(Perl_debug_log,"%16s",""); \
5391 num = RExC_size + 1; \
5393 num=REG_NODE_NUM(RExC_emit); \
5394 if (RExC_lastnum!=num) \
5395 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5397 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5398 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5399 (int)((depth*2)), "", \
5403 RExC_lastparse=RExC_parse; \
5408 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5409 DEBUG_PARSE_MSG((funcname)); \
5410 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5412 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5413 DEBUG_PARSE_MSG((funcname)); \
5414 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5417 - reg - regular expression, i.e. main body or parenthesized thing
5419 * Caller must absorb opening parenthesis.
5421 * Combining parenthesis handling with the base level of regular expression
5422 * is a trifle forced, but the need to tie the tails of the branches to what
5423 * follows makes it hard to avoid.
5425 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5427 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5429 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5433 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5434 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5437 register regnode *ret; /* Will be the head of the group. */
5438 register regnode *br;
5439 register regnode *lastbr;
5440 register regnode *ender = NULL;
5441 register I32 parno = 0;
5443 U32 oregflags = RExC_flags;
5444 bool have_branch = 0;
5446 I32 freeze_paren = 0;
5447 I32 after_freeze = 0;
5449 /* for (?g), (?gc), and (?o) warnings; warning
5450 about (?c) will warn about (?g) -- japhy */
5452 #define WASTED_O 0x01
5453 #define WASTED_G 0x02
5454 #define WASTED_C 0x04
5455 #define WASTED_GC (0x02|0x04)
5456 I32 wastedflags = 0x00;
5458 char * parse_start = RExC_parse; /* MJD */
5459 char * const oregcomp_parse = RExC_parse;
5461 GET_RE_DEBUG_FLAGS_DECL;
5463 PERL_ARGS_ASSERT_REG;
5464 DEBUG_PARSE("reg ");
5466 *flagp = 0; /* Tentatively. */
5469 /* Make an OPEN node, if parenthesized. */
5471 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5472 char *start_verb = RExC_parse;
5473 STRLEN verb_len = 0;
5474 char *start_arg = NULL;
5475 unsigned char op = 0;
5477 int internal_argval = 0; /* internal_argval is only useful if !argok */
5478 while ( *RExC_parse && *RExC_parse != ')' ) {
5479 if ( *RExC_parse == ':' ) {
5480 start_arg = RExC_parse + 1;
5486 verb_len = RExC_parse - start_verb;
5489 while ( *RExC_parse && *RExC_parse != ')' )
5491 if ( *RExC_parse != ')' )
5492 vFAIL("Unterminated verb pattern argument");
5493 if ( RExC_parse == start_arg )
5496 if ( *RExC_parse != ')' )
5497 vFAIL("Unterminated verb pattern");
5500 switch ( *start_verb ) {
5501 case 'A': /* (*ACCEPT) */
5502 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5504 internal_argval = RExC_nestroot;
5507 case 'C': /* (*COMMIT) */
5508 if ( memEQs(start_verb,verb_len,"COMMIT") )
5511 case 'F': /* (*FAIL) */
5512 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5517 case ':': /* (*:NAME) */
5518 case 'M': /* (*MARK:NAME) */
5519 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5524 case 'P': /* (*PRUNE) */
5525 if ( memEQs(start_verb,verb_len,"PRUNE") )
5528 case 'S': /* (*SKIP) */
5529 if ( memEQs(start_verb,verb_len,"SKIP") )
5532 case 'T': /* (*THEN) */
5533 /* [19:06] <TimToady> :: is then */
5534 if ( memEQs(start_verb,verb_len,"THEN") ) {
5536 RExC_seen |= REG_SEEN_CUTGROUP;
5542 vFAIL3("Unknown verb pattern '%.*s'",
5543 verb_len, start_verb);
5546 if ( start_arg && internal_argval ) {
5547 vFAIL3("Verb pattern '%.*s' may not have an argument",
5548 verb_len, start_verb);
5549 } else if ( argok < 0 && !start_arg ) {
5550 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5551 verb_len, start_verb);
5553 ret = reganode(pRExC_state, op, internal_argval);
5554 if ( ! internal_argval && ! SIZE_ONLY ) {
5556 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5557 ARG(ret) = add_data( pRExC_state, 1, "S" );
5558 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5565 if (!internal_argval)
5566 RExC_seen |= REG_SEEN_VERBARG;
5567 } else if ( start_arg ) {
5568 vFAIL3("Verb pattern '%.*s' may not have an argument",
5569 verb_len, start_verb);
5571 ret = reg_node(pRExC_state, op);
5573 nextchar(pRExC_state);
5576 if (*RExC_parse == '?') { /* (?...) */
5577 bool is_logical = 0;
5578 const char * const seqstart = RExC_parse;
5581 paren = *RExC_parse++;
5582 ret = NULL; /* For look-ahead/behind. */
5585 case 'P': /* (?P...) variants for those used to PCRE/Python */
5586 paren = *RExC_parse++;
5587 if ( paren == '<') /* (?P<...>) named capture */
5589 else if (paren == '>') { /* (?P>name) named recursion */
5590 goto named_recursion;
5592 else if (paren == '=') { /* (?P=...) named backref */
5593 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5594 you change this make sure you change that */
5595 char* name_start = RExC_parse;
5597 SV *sv_dat = reg_scan_name(pRExC_state,
5598 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5599 if (RExC_parse == name_start || *RExC_parse != ')')
5600 vFAIL2("Sequence %.3s... not terminated",parse_start);
5603 num = add_data( pRExC_state, 1, "S" );
5604 RExC_rxi->data->data[num]=(void*)sv_dat;
5605 SvREFCNT_inc_simple_void(sv_dat);
5608 ret = reganode(pRExC_state,
5609 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5613 Set_Node_Offset(ret, parse_start+1);
5614 Set_Node_Cur_Length(ret); /* MJD */
5616 nextchar(pRExC_state);
5620 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5622 case '<': /* (?<...) */
5623 if (*RExC_parse == '!')
5625 else if (*RExC_parse != '=')
5631 case '\'': /* (?'...') */
5632 name_start= RExC_parse;
5633 svname = reg_scan_name(pRExC_state,
5634 SIZE_ONLY ? /* reverse test from the others */
5635 REG_RSN_RETURN_NAME :
5636 REG_RSN_RETURN_NULL);
5637 if (RExC_parse == name_start) {
5639 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5642 if (*RExC_parse != paren)
5643 vFAIL2("Sequence (?%c... not terminated",
5644 paren=='>' ? '<' : paren);
5648 if (!svname) /* shouldnt happen */
5650 "panic: reg_scan_name returned NULL");
5651 if (!RExC_paren_names) {
5652 RExC_paren_names= newHV();
5653 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5655 RExC_paren_name_list= newAV();
5656 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5659 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5661 sv_dat = HeVAL(he_str);
5663 /* croak baby croak */
5665 "panic: paren_name hash element allocation failed");
5666 } else if ( SvPOK(sv_dat) ) {
5667 /* (?|...) can mean we have dupes so scan to check
5668 its already been stored. Maybe a flag indicating
5669 we are inside such a construct would be useful,
5670 but the arrays are likely to be quite small, so
5671 for now we punt -- dmq */
5672 IV count = SvIV(sv_dat);
5673 I32 *pv = (I32*)SvPVX(sv_dat);
5675 for ( i = 0 ; i < count ; i++ ) {
5676 if ( pv[i] == RExC_npar ) {
5682 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5683 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5684 pv[count] = RExC_npar;
5685 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5688 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5689 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5691 SvIV_set(sv_dat, 1);
5694 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5695 SvREFCNT_dec(svname);
5698 /*sv_dump(sv_dat);*/
5700 nextchar(pRExC_state);
5702 goto capturing_parens;
5704 RExC_seen |= REG_SEEN_LOOKBEHIND;
5706 case '=': /* (?=...) */
5707 RExC_seen_zerolen++;
5709 case '!': /* (?!...) */
5710 RExC_seen_zerolen++;
5711 if (*RExC_parse == ')') {
5712 ret=reg_node(pRExC_state, OPFAIL);
5713 nextchar(pRExC_state);
5717 case '|': /* (?|...) */
5718 /* branch reset, behave like a (?:...) except that
5719 buffers in alternations share the same numbers */
5721 after_freeze = freeze_paren = RExC_npar;
5723 case ':': /* (?:...) */
5724 case '>': /* (?>...) */
5726 case '$': /* (?$...) */
5727 case '@': /* (?@...) */
5728 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5730 case '#': /* (?#...) */
5731 while (*RExC_parse && *RExC_parse != ')')
5733 if (*RExC_parse != ')')
5734 FAIL("Sequence (?#... not terminated");
5735 nextchar(pRExC_state);
5738 case '0' : /* (?0) */
5739 case 'R' : /* (?R) */
5740 if (*RExC_parse != ')')
5741 FAIL("Sequence (?R) not terminated");
5742 ret = reg_node(pRExC_state, GOSTART);
5743 *flagp |= POSTPONED;
5744 nextchar(pRExC_state);
5747 { /* named and numeric backreferences */
5749 case '&': /* (?&NAME) */
5750 parse_start = RExC_parse - 1;
5753 SV *sv_dat = reg_scan_name(pRExC_state,
5754 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5755 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5757 goto gen_recurse_regop;
5760 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5762 vFAIL("Illegal pattern");
5764 goto parse_recursion;
5766 case '-': /* (?-1) */
5767 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5768 RExC_parse--; /* rewind to let it be handled later */
5772 case '1': case '2': case '3': case '4': /* (?1) */
5773 case '5': case '6': case '7': case '8': case '9':
5776 num = atoi(RExC_parse);
5777 parse_start = RExC_parse - 1; /* MJD */
5778 if (*RExC_parse == '-')
5780 while (isDIGIT(*RExC_parse))
5782 if (*RExC_parse!=')')
5783 vFAIL("Expecting close bracket");
5786 if ( paren == '-' ) {
5788 Diagram of capture buffer numbering.
5789 Top line is the normal capture buffer numbers
5790 Botton line is the negative indexing as from
5794 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5798 num = RExC_npar + num;
5801 vFAIL("Reference to nonexistent group");
5803 } else if ( paren == '+' ) {
5804 num = RExC_npar + num - 1;
5807 ret = reganode(pRExC_state, GOSUB, num);
5809 if (num > (I32)RExC_rx->nparens) {
5811 vFAIL("Reference to nonexistent group");
5813 ARG2L_SET( ret, RExC_recurse_count++);
5815 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5816 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5820 RExC_seen |= REG_SEEN_RECURSE;
5821 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5822 Set_Node_Offset(ret, parse_start); /* MJD */
5824 *flagp |= POSTPONED;
5825 nextchar(pRExC_state);
5827 } /* named and numeric backreferences */
5830 case '?': /* (??...) */
5832 if (*RExC_parse != '{') {
5834 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5837 *flagp |= POSTPONED;
5838 paren = *RExC_parse++;
5840 case '{': /* (?{...}) */
5845 char *s = RExC_parse;
5847 RExC_seen_zerolen++;
5848 RExC_seen |= REG_SEEN_EVAL;
5849 while (count && (c = *RExC_parse)) {
5860 if (*RExC_parse != ')') {
5862 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5866 OP_4tree *sop, *rop;
5867 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5870 Perl_save_re_context(aTHX);
5871 rop = sv_compile_2op(sv, &sop, "re", &pad);
5872 sop->op_private |= OPpREFCOUNTED;
5873 /* re_dup will OpREFCNT_inc */
5874 OpREFCNT_set(sop, 1);
5877 n = add_data(pRExC_state, 3, "nop");
5878 RExC_rxi->data->data[n] = (void*)rop;
5879 RExC_rxi->data->data[n+1] = (void*)sop;
5880 RExC_rxi->data->data[n+2] = (void*)pad;
5883 else { /* First pass */
5884 if (PL_reginterp_cnt < ++RExC_seen_evals
5886 /* No compiled RE interpolated, has runtime
5887 components ===> unsafe. */
5888 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5889 if (PL_tainting && PL_tainted)
5890 FAIL("Eval-group in insecure regular expression");
5891 #if PERL_VERSION > 8
5892 if (IN_PERL_COMPILETIME)
5897 nextchar(pRExC_state);
5899 ret = reg_node(pRExC_state, LOGICAL);
5902 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5903 /* deal with the length of this later - MJD */
5906 ret = reganode(pRExC_state, EVAL, n);
5907 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5908 Set_Node_Offset(ret, parse_start);
5911 case '(': /* (?(?{...})...) and (?(?=...)...) */
5914 if (RExC_parse[0] == '?') { /* (?(?...)) */
5915 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5916 || RExC_parse[1] == '<'
5917 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5920 ret = reg_node(pRExC_state, LOGICAL);
5923 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5927 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5928 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5930 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5931 char *name_start= RExC_parse++;
5933 SV *sv_dat=reg_scan_name(pRExC_state,
5934 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5935 if (RExC_parse == name_start || *RExC_parse != ch)
5936 vFAIL2("Sequence (?(%c... not terminated",
5937 (ch == '>' ? '<' : ch));
5940 num = add_data( pRExC_state, 1, "S" );
5941 RExC_rxi->data->data[num]=(void*)sv_dat;
5942 SvREFCNT_inc_simple_void(sv_dat);
5944 ret = reganode(pRExC_state,NGROUPP,num);
5945 goto insert_if_check_paren;
5947 else if (RExC_parse[0] == 'D' &&
5948 RExC_parse[1] == 'E' &&
5949 RExC_parse[2] == 'F' &&
5950 RExC_parse[3] == 'I' &&
5951 RExC_parse[4] == 'N' &&
5952 RExC_parse[5] == 'E')
5954 ret = reganode(pRExC_state,DEFINEP,0);
5957 goto insert_if_check_paren;
5959 else if (RExC_parse[0] == 'R') {
5962 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5963 parno = atoi(RExC_parse++);
5964 while (isDIGIT(*RExC_parse))
5966 } else if (RExC_parse[0] == '&') {
5969 sv_dat = reg_scan_name(pRExC_state,
5970 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5971 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5973 ret = reganode(pRExC_state,INSUBP,parno);
5974 goto insert_if_check_paren;
5976 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
5979 parno = atoi(RExC_parse++);
5981 while (isDIGIT(*RExC_parse))
5983 ret = reganode(pRExC_state, GROUPP, parno);
5985 insert_if_check_paren:
5986 if ((c = *nextchar(pRExC_state)) != ')')
5987 vFAIL("Switch condition not recognized");
5989 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
5990 br = regbranch(pRExC_state, &flags, 1,depth+1);
5992 br = reganode(pRExC_state, LONGJMP, 0);
5994 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
5995 c = *nextchar(pRExC_state);
6000 vFAIL("(?(DEFINE)....) does not allow branches");
6001 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6002 regbranch(pRExC_state, &flags, 1,depth+1);
6003 REGTAIL(pRExC_state, ret, lastbr);
6006 c = *nextchar(pRExC_state);
6011 vFAIL("Switch (?(condition)... contains too many branches");
6012 ender = reg_node(pRExC_state, TAIL);
6013 REGTAIL(pRExC_state, br, ender);
6015 REGTAIL(pRExC_state, lastbr, ender);
6016 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6019 REGTAIL(pRExC_state, ret, ender);
6020 RExC_size++; /* XXX WHY do we need this?!!
6021 For large programs it seems to be required
6022 but I can't figure out why. -- dmq*/
6026 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6030 RExC_parse--; /* for vFAIL to print correctly */
6031 vFAIL("Sequence (? incomplete");
6035 parse_flags: /* (?i) */
6037 U32 posflags = 0, negflags = 0;
6038 U32 *flagsp = &posflags;
6040 while (*RExC_parse) {
6041 /* && strchr("iogcmsx", *RExC_parse) */
6042 /* (?g), (?gc) and (?o) are useless here
6043 and must be globally applied -- japhy */
6044 switch (*RExC_parse) {
6045 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6046 case ONCE_PAT_MOD: /* 'o' */
6047 case GLOBAL_PAT_MOD: /* 'g' */
6048 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6049 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6050 if (! (wastedflags & wflagbit) ) {
6051 wastedflags |= wflagbit;
6054 "Useless (%s%c) - %suse /%c modifier",
6055 flagsp == &negflags ? "?-" : "?",
6057 flagsp == &negflags ? "don't " : "",
6064 case CONTINUE_PAT_MOD: /* 'c' */
6065 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6066 if (! (wastedflags & WASTED_C) ) {
6067 wastedflags |= WASTED_GC;
6070 "Useless (%sc) - %suse /gc modifier",
6071 flagsp == &negflags ? "?-" : "?",
6072 flagsp == &negflags ? "don't " : ""
6077 case KEEPCOPY_PAT_MOD: /* 'p' */
6078 if (flagsp == &negflags) {
6079 if (SIZE_ONLY && ckWARN(WARN_REGEXP))
6080 vWARN(RExC_parse + 1,"Useless use of (?-p)");
6082 *flagsp |= RXf_PMf_KEEPCOPY;
6086 if (flagsp == &negflags) {
6088 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6092 wastedflags = 0; /* reset so (?g-c) warns twice */
6098 RExC_flags |= posflags;
6099 RExC_flags &= ~negflags;
6101 oregflags |= posflags;
6102 oregflags &= ~negflags;
6104 nextchar(pRExC_state);
6115 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6120 }} /* one for the default block, one for the switch */
6127 ret = reganode(pRExC_state, OPEN, parno);
6130 RExC_nestroot = parno;
6131 if (RExC_seen & REG_SEEN_RECURSE
6132 && !RExC_open_parens[parno-1])
6134 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6135 "Setting open paren #%"IVdf" to %d\n",
6136 (IV)parno, REG_NODE_NUM(ret)));
6137 RExC_open_parens[parno-1]= ret;
6140 Set_Node_Length(ret, 1); /* MJD */
6141 Set_Node_Offset(ret, RExC_parse); /* MJD */
6149 /* Pick up the branches, linking them together. */
6150 parse_start = RExC_parse; /* MJD */
6151 br = regbranch(pRExC_state, &flags, 1,depth+1);
6154 if (RExC_npar > after_freeze)
6155 after_freeze = RExC_npar;
6156 RExC_npar = freeze_paren;
6159 /* branch_len = (paren != 0); */
6163 if (*RExC_parse == '|') {
6164 if (!SIZE_ONLY && RExC_extralen) {
6165 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6168 reginsert(pRExC_state, BRANCH, br, depth+1);
6169 Set_Node_Length(br, paren != 0);
6170 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6174 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6176 else if (paren == ':') {
6177 *flagp |= flags&SIMPLE;
6179 if (is_open) { /* Starts with OPEN. */
6180 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6182 else if (paren != '?') /* Not Conditional */
6184 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6186 while (*RExC_parse == '|') {
6187 if (!SIZE_ONLY && RExC_extralen) {
6188 ender = reganode(pRExC_state, LONGJMP,0);
6189 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6192 RExC_extralen += 2; /* Account for LONGJMP. */
6193 nextchar(pRExC_state);
6195 if (RExC_npar > after_freeze)
6196 after_freeze = RExC_npar;
6197 RExC_npar = freeze_paren;
6199 br = regbranch(pRExC_state, &flags, 0, depth+1);
6203 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6205 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6208 if (have_branch || paren != ':') {
6209 /* Make a closing node, and hook it on the end. */
6212 ender = reg_node(pRExC_state, TAIL);
6215 ender = reganode(pRExC_state, CLOSE, parno);
6216 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6217 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6218 "Setting close paren #%"IVdf" to %d\n",
6219 (IV)parno, REG_NODE_NUM(ender)));
6220 RExC_close_parens[parno-1]= ender;
6221 if (RExC_nestroot == parno)
6224 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6225 Set_Node_Length(ender,1); /* MJD */
6231 *flagp &= ~HASWIDTH;
6234 ender = reg_node(pRExC_state, SUCCEED);
6237 ender = reg_node(pRExC_state, END);
6239 assert(!RExC_opend); /* there can only be one! */
6244 REGTAIL(pRExC_state, lastbr, ender);
6246 if (have_branch && !SIZE_ONLY) {
6248 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6250 /* Hook the tails of the branches to the closing node. */
6251 for (br = ret; br; br = regnext(br)) {
6252 const U8 op = PL_regkind[OP(br)];
6254 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6256 else if (op == BRANCHJ) {
6257 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6265 static const char parens[] = "=!<,>";
6267 if (paren && (p = strchr(parens, paren))) {
6268 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6269 int flag = (p - parens) > 1;
6272 node = SUSPEND, flag = 0;
6273 reginsert(pRExC_state, node,ret, depth+1);
6274 Set_Node_Cur_Length(ret);
6275 Set_Node_Offset(ret, parse_start + 1);
6277 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6281 /* Check for proper termination. */
6283 RExC_flags = oregflags;
6284 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6285 RExC_parse = oregcomp_parse;
6286 vFAIL("Unmatched (");
6289 else if (!paren && RExC_parse < RExC_end) {
6290 if (*RExC_parse == ')') {
6292 vFAIL("Unmatched )");
6295 FAIL("Junk on end of regexp"); /* "Can't happen". */
6299 RExC_npar = after_freeze;
6304 - regbranch - one alternative of an | operator
6306 * Implements the concatenation operator.
6309 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6312 register regnode *ret;
6313 register regnode *chain = NULL;
6314 register regnode *latest;
6315 I32 flags = 0, c = 0;
6316 GET_RE_DEBUG_FLAGS_DECL;
6318 PERL_ARGS_ASSERT_REGBRANCH;
6320 DEBUG_PARSE("brnc");
6325 if (!SIZE_ONLY && RExC_extralen)
6326 ret = reganode(pRExC_state, BRANCHJ,0);
6328 ret = reg_node(pRExC_state, BRANCH);
6329 Set_Node_Length(ret, 1);
6333 if (!first && SIZE_ONLY)
6334 RExC_extralen += 1; /* BRANCHJ */
6336 *flagp = WORST; /* Tentatively. */
6339 nextchar(pRExC_state);
6340 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6342 latest = regpiece(pRExC_state, &flags,depth+1);
6343 if (latest == NULL) {
6344 if (flags & TRYAGAIN)
6348 else if (ret == NULL)
6350 *flagp |= flags&(HASWIDTH|POSTPONED);
6351 if (chain == NULL) /* First piece. */
6352 *flagp |= flags&SPSTART;
6355 REGTAIL(pRExC_state, chain, latest);
6360 if (chain == NULL) { /* Loop ran zero times. */
6361 chain = reg_node(pRExC_state, NOTHING);
6366 *flagp |= flags&SIMPLE;
6373 - regpiece - something followed by possible [*+?]
6375 * Note that the branching code sequences used for ? and the general cases
6376 * of * and + are somewhat optimized: they use the same NOTHING node as
6377 * both the endmarker for their branch list and the body of the last branch.
6378 * It might seem that this node could be dispensed with entirely, but the
6379 * endmarker role is not redundant.
6382 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6385 register regnode *ret;
6387 register char *next;
6389 const char * const origparse = RExC_parse;
6391 I32 max = REG_INFTY;
6393 const char *maxpos = NULL;
6394 GET_RE_DEBUG_FLAGS_DECL;
6396 PERL_ARGS_ASSERT_REGPIECE;
6398 DEBUG_PARSE("piec");
6400 ret = regatom(pRExC_state, &flags,depth+1);
6402 if (flags & TRYAGAIN)
6409 if (op == '{' && regcurly(RExC_parse)) {
6411 parse_start = RExC_parse; /* MJD */
6412 next = RExC_parse + 1;
6413 while (isDIGIT(*next) || *next == ',') {
6422 if (*next == '}') { /* got one */
6426 min = atoi(RExC_parse);
6430 maxpos = RExC_parse;
6432 if (!max && *maxpos != '0')
6433 max = REG_INFTY; /* meaning "infinity" */
6434 else if (max >= REG_INFTY)
6435 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6437 nextchar(pRExC_state);
6440 if ((flags&SIMPLE)) {
6441 RExC_naughty += 2 + RExC_naughty / 2;
6442 reginsert(pRExC_state, CURLY, ret, depth+1);
6443 Set_Node_Offset(ret, parse_start+1); /* MJD */
6444 Set_Node_Cur_Length(ret);
6447 regnode * const w = reg_node(pRExC_state, WHILEM);
6450 REGTAIL(pRExC_state, ret, w);
6451 if (!SIZE_ONLY && RExC_extralen) {
6452 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6453 reginsert(pRExC_state, NOTHING,ret, depth+1);
6454 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6456 reginsert(pRExC_state, CURLYX,ret, depth+1);
6458 Set_Node_Offset(ret, parse_start+1);
6459 Set_Node_Length(ret,
6460 op == '{' ? (RExC_parse - parse_start) : 1);
6462 if (!SIZE_ONLY && RExC_extralen)
6463 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6464 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6466 RExC_whilem_seen++, RExC_extralen += 3;
6467 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6476 vFAIL("Can't do {n,m} with n > m");
6478 ARG1_SET(ret, (U16)min);
6479 ARG2_SET(ret, (U16)max);
6491 #if 0 /* Now runtime fix should be reliable. */
6493 /* if this is reinstated, don't forget to put this back into perldiag:
6495 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6497 (F) The part of the regexp subject to either the * or + quantifier
6498 could match an empty string. The {#} shows in the regular
6499 expression about where the problem was discovered.
6503 if (!(flags&HASWIDTH) && op != '?')
6504 vFAIL("Regexp *+ operand could be empty");
6507 parse_start = RExC_parse;
6508 nextchar(pRExC_state);
6510 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6512 if (op == '*' && (flags&SIMPLE)) {
6513 reginsert(pRExC_state, STAR, ret, depth+1);
6517 else if (op == '*') {
6521 else if (op == '+' && (flags&SIMPLE)) {
6522 reginsert(pRExC_state, PLUS, ret, depth+1);
6526 else if (op == '+') {
6530 else if (op == '?') {
6535 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3 && ckWARN(WARN_REGEXP)) {
6537 "%.*s matches null string many times",
6538 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6542 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6543 nextchar(pRExC_state);
6544 reginsert(pRExC_state, MINMOD, ret, depth+1);
6545 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6547 #ifndef REG_ALLOW_MINMOD_SUSPEND
6550 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6552 nextchar(pRExC_state);
6553 ender = reg_node(pRExC_state, SUCCEED);
6554 REGTAIL(pRExC_state, ret, ender);
6555 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6557 ender = reg_node(pRExC_state, TAIL);
6558 REGTAIL(pRExC_state, ret, ender);
6562 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6564 vFAIL("Nested quantifiers");
6571 /* reg_namedseq(pRExC_state,UVp)
6573 This is expected to be called by a parser routine that has
6574 recognized '\N' and needs to handle the rest. RExC_parse is
6575 expected to point at the first char following the N at the time
6578 If valuep is non-null then it is assumed that we are parsing inside
6579 of a charclass definition and the first codepoint in the resolved
6580 string is returned via *valuep and the routine will return NULL.
6581 In this mode if a multichar string is returned from the charnames
6582 handler a warning will be issued, and only the first char in the
6583 sequence will be examined. If the string returned is zero length
6584 then the value of *valuep is undefined and NON-NULL will
6585 be returned to indicate failure. (This will NOT be a valid pointer
6588 If valuep is null then it is assumed that we are parsing normal text
6589 and inserts a new EXACT node into the program containing the resolved
6590 string and returns a pointer to the new node. If the string is
6591 zerolength a NOTHING node is emitted.
6593 On success RExC_parse is set to the char following the endbrace.
6594 Parsing failures will generate a fatal errorvia vFAIL(...)
6596 NOTE: We cache all results from the charnames handler locally in
6597 the RExC_charnames hash (created on first use) to prevent a charnames
6598 handler from playing silly-buggers and returning a short string and
6599 then a long string for a given pattern. Since the regexp program
6600 size is calculated during an initial parse this would result
6601 in a buffer overrun so we cache to prevent the charname result from
6602 changing during the course of the parse.
6606 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp)
6608 char * name; /* start of the content of the name */
6609 char * endbrace; /* endbrace following the name */
6612 STRLEN len; /* this has various purposes throughout the code */
6613 bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
6614 regnode *ret = NULL;
6616 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6618 if (*RExC_parse != '{' ||
6619 (*RExC_parse == '{' && RExC_parse[1]
6620 && strchr("0123456789", RExC_parse[1])))
6622 GET_RE_DEBUG_FLAGS_DECL;
6624 /* no bare \N in a charclass */
6625 vFAIL("Missing braces on \\N{}");
6627 nextchar(pRExC_state);
6628 ret = reg_node(pRExC_state, REG_ANY);
6629 *flagp |= HASWIDTH|SIMPLE;
6632 Set_Node_Length(ret, 1); /* MJD */
6635 name = RExC_parse+1;
6636 endbrace = strchr(RExC_parse, '}');
6639 vFAIL("Missing right brace on \\N{}");
6641 RExC_parse = endbrace + 1;
6644 /* RExC_parse points at the beginning brace,
6645 endbrace points at the last */
6646 if ( name[0]=='U' && name[1]=='+' ) {
6647 /* its a "Unicode hex" notation {U+89AB} */
6648 I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
6649 | PERL_SCAN_DISALLOW_PREFIX
6650 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6652 len = (STRLEN)(endbrace - name - 2);
6653 cp = grok_hex(name + 2, &len, &fl, NULL);
6654 if ( len != (STRLEN)(endbrace - name - 2) ) {
6658 if (cp > 0xff) RExC_utf8 = 1;
6663 /* Need to convert to utf8 if either: won't fit into a byte, or the re
6664 * is going to be in utf8 and the representation changes under utf8. */
6665 if (cp > 0xff || (RExC_utf8 && ! UNI_IS_INVARIANT(cp))) {
6666 U8 string[UTF8_MAXBYTES+1];
6669 tmps = uvuni_to_utf8(string, cp);
6670 sv_str = newSVpvn_utf8((char*)string, tmps - string, TRUE);
6671 } else { /* Otherwise, no need for utf8, can skip that step */
6674 sv_str= newSVpvn(&string, 1);
6677 /* fetch the charnames handler for this scope */
6678 HV * const table = GvHV(PL_hintgv);
6680 hv_fetchs(table, "charnames", FALSE) :
6682 SV *cv= cvp ? *cvp : NULL;
6685 /* create an SV with the name as argument */
6686 sv_name = newSVpvn(name, endbrace - name);
6688 if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
6689 vFAIL2("Constant(\\N{%s}) unknown: "
6690 "(possibly a missing \"use charnames ...\")",
6693 if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
6694 vFAIL2("Constant(\\N{%s}): "
6695 "$^H{charnames} is not defined",SvPVX(sv_name));
6700 if (!RExC_charnames) {
6701 /* make sure our cache is allocated */
6702 RExC_charnames = newHV();
6703 sv_2mortal(MUTABLE_SV(RExC_charnames));
6705 /* see if we have looked this one up before */
6706 he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
6708 sv_str = HeVAL(he_str);
6721 count= call_sv(cv, G_SCALAR);
6723 if (count == 1) { /* XXXX is this right? dmq */
6725 SvREFCNT_inc_simple_void(sv_str);
6733 if ( !sv_str || !SvOK(sv_str) ) {
6734 vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
6735 "did not return a defined value",SvPVX(sv_name));
6737 if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
6742 char *p = SvPV(sv_str, len);
6745 if ( SvUTF8(sv_str) ) {
6746 *valuep = utf8_to_uvchr((U8*)p, &numlen);
6750 We have to turn on utf8 for high bit chars otherwise
6751 we get failures with
6753 "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6754 "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
6756 This is different from what \x{} would do with the same
6757 codepoint, where the condition is > 0xFF.
6764 /* warn if we havent used the whole string? */
6766 if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6768 "Ignoring excess chars from \\N{%s} in character class",
6772 } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6774 "Ignoring zero length \\N{%s} in character class",
6779 SvREFCNT_dec(sv_name);
6781 SvREFCNT_dec(sv_str);
6782 return len ? NULL : (regnode *)&len;
6783 } else if(SvCUR(sv_str)) {
6789 char * parse_start = name-3; /* needed for the offsets */
6791 GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
6793 ret = reg_node(pRExC_state,
6794 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6797 if ( RExC_utf8 && !SvUTF8(sv_str) ) {
6798 sv_utf8_upgrade(sv_str);
6799 } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
6803 p = SvPV(sv_str, len);
6805 /* len is the length written, charlen is the size the char read */
6806 for ( len = 0; p < pend; p += charlen ) {
6808 UV uvc = utf8_to_uvchr((U8*)p, &charlen);
6810 STRLEN foldlen,numlen;
6811 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6812 uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
6813 /* Emit all the Unicode characters. */
6815 for (foldbuf = tmpbuf;
6819 uvc = utf8_to_uvchr(foldbuf, &numlen);
6821 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6824 /* In EBCDIC the numlen
6825 * and unilen can differ. */
6827 if (numlen >= foldlen)
6831 break; /* "Can't happen." */
6834 const STRLEN unilen = reguni(pRExC_state, uvc, s);
6846 RExC_size += STR_SZ(len);
6849 RExC_emit += STR_SZ(len);
6851 Set_Node_Cur_Length(ret); /* MJD */
6853 nextchar(pRExC_state);
6854 } else { /* zero length */
6855 ret = reg_node(pRExC_state,NOTHING);
6858 SvREFCNT_dec(sv_str);
6861 SvREFCNT_dec(sv_name);
6871 * It returns the code point in utf8 for the value in *encp.
6872 * value: a code value in the source encoding
6873 * encp: a pointer to an Encode object
6875 * If the result from Encode is not a single character,
6876 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6879 S_reg_recode(pTHX_ const char value, SV **encp)
6882 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
6883 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6884 const STRLEN newlen = SvCUR(sv);
6885 UV uv = UNICODE_REPLACEMENT;
6887 PERL_ARGS_ASSERT_REG_RECODE;
6891 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6894 if (!newlen || numlen != newlen) {
6895 uv = UNICODE_REPLACEMENT;
6903 - regatom - the lowest level
6905 Try to identify anything special at the start of the pattern. If there
6906 is, then handle it as required. This may involve generating a single regop,
6907 such as for an assertion; or it may involve recursing, such as to
6908 handle a () structure.
6910 If the string doesn't start with something special then we gobble up
6911 as much literal text as we can.
6913 Once we have been able to handle whatever type of thing started the
6914 sequence, we return.
6916 Note: we have to be careful with escapes, as they can be both literal
6917 and special, and in the case of \10 and friends can either, depending
6918 on context. Specifically there are two seperate switches for handling
6919 escape sequences, with the one for handling literal escapes requiring
6920 a dummy entry for all of the special escapes that are actually handled
6925 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6928 register regnode *ret = NULL;
6930 char *parse_start = RExC_parse;
6931 GET_RE_DEBUG_FLAGS_DECL;
6932 DEBUG_PARSE("atom");
6933 *flagp = WORST; /* Tentatively. */
6935 PERL_ARGS_ASSERT_REGATOM;
6938 switch ((U8)*RExC_parse) {
6940 RExC_seen_zerolen++;
6941 nextchar(pRExC_state);
6942 if (RExC_flags & RXf_PMf_MULTILINE)
6943 ret = reg_node(pRExC_state, MBOL);
6944 else if (RExC_flags & RXf_PMf_SINGLELINE)
6945 ret = reg_node(pRExC_state, SBOL);
6947 ret = reg_node(pRExC_state, BOL);
6948 Set_Node_Length(ret, 1); /* MJD */
6951 nextchar(pRExC_state);
6953 RExC_seen_zerolen++;
6954 if (RExC_flags & RXf_PMf_MULTILINE)
6955 ret = reg_node(pRExC_state, MEOL);
6956 else if (RExC_flags & RXf_PMf_SINGLELINE)
6957 ret = reg_node(pRExC_state, SEOL);
6959 ret = reg_node(pRExC_state, EOL);
6960 Set_Node_Length(ret, 1); /* MJD */
6963 nextchar(pRExC_state);
6964 if (RExC_flags & RXf_PMf_SINGLELINE)
6965 ret = reg_node(pRExC_state, SANY);
6967 ret = reg_node(pRExC_state, REG_ANY);
6968 *flagp |= HASWIDTH|SIMPLE;
6970 Set_Node_Length(ret, 1); /* MJD */
6974 char * const oregcomp_parse = ++RExC_parse;
6975 ret = regclass(pRExC_state,depth+1);
6976 if (*RExC_parse != ']') {
6977 RExC_parse = oregcomp_parse;
6978 vFAIL("Unmatched [");
6980 nextchar(pRExC_state);
6981 *flagp |= HASWIDTH|SIMPLE;
6982 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
6986 nextchar(pRExC_state);
6987 ret = reg(pRExC_state, 1, &flags,depth+1);
6989 if (flags & TRYAGAIN) {
6990 if (RExC_parse == RExC_end) {
6991 /* Make parent create an empty node if needed. */
6999 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7003 if (flags & TRYAGAIN) {
7007 vFAIL("Internal urp");
7008 /* Supposed to be caught earlier. */
7011 if (!regcurly(RExC_parse)) {
7020 vFAIL("Quantifier follows nothing");
7028 len=0; /* silence a spurious compiler warning */
7029 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
7030 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
7031 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
7032 ret = reganode(pRExC_state, FOLDCHAR, cp);
7033 Set_Node_Length(ret, 1); /* MJD */
7034 nextchar(pRExC_state); /* kill whitespace under /x */
7042 This switch handles escape sequences that resolve to some kind
7043 of special regop and not to literal text. Escape sequnces that
7044 resolve to literal text are handled below in the switch marked
7047 Every entry in this switch *must* have a corresponding entry
7048 in the literal escape switch. However, the opposite is not
7049 required, as the default for this switch is to jump to the
7050 literal text handling code.
7052 switch ((U8)*++RExC_parse) {
7057 /* Special Escapes */
7059 RExC_seen_zerolen++;
7060 ret = reg_node(pRExC_state, SBOL);
7062 goto finish_meta_pat;
7064 ret = reg_node(pRExC_state, GPOS);
7065 RExC_seen |= REG_SEEN_GPOS;
7067 goto finish_meta_pat;
7069 RExC_seen_zerolen++;
7070 ret = reg_node(pRExC_state, KEEPS);
7072 /* XXX:dmq : disabling in-place substitution seems to
7073 * be necessary here to avoid cases of memory corruption, as
7074 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7076 RExC_seen |= REG_SEEN_LOOKBEHIND;
7077 goto finish_meta_pat;
7079 ret = reg_node(pRExC_state, SEOL);
7081 RExC_seen_zerolen++; /* Do not optimize RE away */
7082 goto finish_meta_pat;
7084 ret = reg_node(pRExC_state, EOS);
7086 RExC_seen_zerolen++; /* Do not optimize RE away */
7087 goto finish_meta_pat;
7089 ret = reg_node(pRExC_state, CANY);
7090 RExC_seen |= REG_SEEN_CANY;
7091 *flagp |= HASWIDTH|SIMPLE;
7092 goto finish_meta_pat;
7094 ret = reg_node(pRExC_state, CLUMP);
7096 goto finish_meta_pat;
7098 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
7099 *flagp |= HASWIDTH|SIMPLE;
7100 goto finish_meta_pat;
7102 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
7103 *flagp |= HASWIDTH|SIMPLE;
7104 goto finish_meta_pat;
7106 RExC_seen_zerolen++;
7107 RExC_seen |= REG_SEEN_LOOKBEHIND;
7108 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
7110 goto finish_meta_pat;
7112 RExC_seen_zerolen++;
7113 RExC_seen |= REG_SEEN_LOOKBEHIND;
7114 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
7116 goto finish_meta_pat;
7118 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
7119 *flagp |= HASWIDTH|SIMPLE;
7120 goto finish_meta_pat;
7122 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
7123 *flagp |= HASWIDTH|SIMPLE;
7124 goto finish_meta_pat;
7126 ret = reg_node(pRExC_state, DIGIT);
7127 *flagp |= HASWIDTH|SIMPLE;
7128 goto finish_meta_pat;
7130 ret = reg_node(pRExC_state, NDIGIT);
7131 *flagp |= HASWIDTH|SIMPLE;
7132 goto finish_meta_pat;
7134 ret = reg_node(pRExC_state, LNBREAK);
7135 *flagp |= HASWIDTH|SIMPLE;
7136 goto finish_meta_pat;
7138 ret = reg_node(pRExC_state, HORIZWS);
7139 *flagp |= HASWIDTH|SIMPLE;
7140 goto finish_meta_pat;
7142 ret = reg_node(pRExC_state, NHORIZWS);
7143 *flagp |= HASWIDTH|SIMPLE;
7144 goto finish_meta_pat;
7146 ret = reg_node(pRExC_state, VERTWS);
7147 *flagp |= HASWIDTH|SIMPLE;
7148 goto finish_meta_pat;
7150 ret = reg_node(pRExC_state, NVERTWS);
7151 *flagp |= HASWIDTH|SIMPLE;
7153 nextchar(pRExC_state);
7154 Set_Node_Length(ret, 2); /* MJD */
7159 char* const oldregxend = RExC_end;
7161 char* parse_start = RExC_parse - 2;
7164 if (RExC_parse[1] == '{') {
7165 /* a lovely hack--pretend we saw [\pX] instead */
7166 RExC_end = strchr(RExC_parse, '}');
7168 const U8 c = (U8)*RExC_parse;
7170 RExC_end = oldregxend;
7171 vFAIL2("Missing right brace on \\%c{}", c);
7176 RExC_end = RExC_parse + 2;
7177 if (RExC_end > oldregxend)
7178 RExC_end = oldregxend;
7182 ret = regclass(pRExC_state,depth+1);
7184 RExC_end = oldregxend;
7187 Set_Node_Offset(ret, parse_start + 2);
7188 Set_Node_Cur_Length(ret);
7189 nextchar(pRExC_state);
7190 *flagp |= HASWIDTH|SIMPLE;
7194 /* Handle \N and \N{NAME} here and not below because it can be
7195 multicharacter. join_exact() will join them up later on.
7196 Also this makes sure that things like /\N{BLAH}+/ and
7197 \N{BLAH} being multi char Just Happen. dmq*/
7199 ret= reg_namedseq(pRExC_state, NULL, flagp);
7201 case 'k': /* Handle \k<NAME> and \k'NAME' */
7204 char ch= RExC_parse[1];
7205 if (ch != '<' && ch != '\'' && ch != '{') {
7207 vFAIL2("Sequence %.2s... not terminated",parse_start);
7209 /* this pretty much dupes the code for (?P=...) in reg(), if
7210 you change this make sure you change that */
7211 char* name_start = (RExC_parse += 2);
7213 SV *sv_dat = reg_scan_name(pRExC_state,
7214 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7215 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7216 if (RExC_parse == name_start || *RExC_parse != ch)
7217 vFAIL2("Sequence %.3s... not terminated",parse_start);
7220 num = add_data( pRExC_state, 1, "S" );
7221 RExC_rxi->data->data[num]=(void*)sv_dat;
7222 SvREFCNT_inc_simple_void(sv_dat);
7226 ret = reganode(pRExC_state,
7227 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7231 /* override incorrect value set in reganode MJD */
7232 Set_Node_Offset(ret, parse_start+1);
7233 Set_Node_Cur_Length(ret); /* MJD */
7234 nextchar(pRExC_state);
7240 case '1': case '2': case '3': case '4':
7241 case '5': case '6': case '7': case '8': case '9':
7244 bool isg = *RExC_parse == 'g';
7249 if (*RExC_parse == '{') {
7253 if (*RExC_parse == '-') {
7257 if (hasbrace && !isDIGIT(*RExC_parse)) {
7258 if (isrel) RExC_parse--;
7260 goto parse_named_seq;
7262 num = atoi(RExC_parse);
7263 if (isg && num == 0)
7264 vFAIL("Reference to invalid group 0");
7266 num = RExC_npar - num;
7268 vFAIL("Reference to nonexistent or unclosed group");
7270 if (!isg && num > 9 && num >= RExC_npar)
7273 char * const parse_start = RExC_parse - 1; /* MJD */
7274 while (isDIGIT(*RExC_parse))
7276 if (parse_start == RExC_parse - 1)
7277 vFAIL("Unterminated \\g... pattern");
7279 if (*RExC_parse != '}')
7280 vFAIL("Unterminated \\g{...} pattern");
7284 if (num > (I32)RExC_rx->nparens)
7285 vFAIL("Reference to nonexistent group");
7288 ret = reganode(pRExC_state,
7289 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7293 /* override incorrect value set in reganode MJD */
7294 Set_Node_Offset(ret, parse_start+1);
7295 Set_Node_Cur_Length(ret); /* MJD */
7297 nextchar(pRExC_state);
7302 if (RExC_parse >= RExC_end)
7303 FAIL("Trailing \\");
7306 /* Do not generate "unrecognized" warnings here, we fall
7307 back into the quick-grab loop below */
7314 if (RExC_flags & RXf_PMf_EXTENDED) {
7315 if ( reg_skipcomment( pRExC_state ) )
7322 register STRLEN len;
7327 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7329 parse_start = RExC_parse - 1;
7335 ret = reg_node(pRExC_state,
7336 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7338 for (len = 0, p = RExC_parse - 1;
7339 len < 127 && p < RExC_end;
7342 char * const oldp = p;
7344 if (RExC_flags & RXf_PMf_EXTENDED)
7345 p = regwhite( pRExC_state, p );
7350 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7351 goto normal_default;
7361 /* Literal Escapes Switch
7363 This switch is meant to handle escape sequences that
7364 resolve to a literal character.
7366 Every escape sequence that represents something
7367 else, like an assertion or a char class, is handled
7368 in the switch marked 'Special Escapes' above in this
7369 routine, but also has an entry here as anything that
7370 isn't explicitly mentioned here will be treated as
7371 an unescaped equivalent literal.
7375 /* These are all the special escapes. */
7379 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7380 goto normal_default;
7381 case 'A': /* Start assertion */
7382 case 'b': case 'B': /* Word-boundary assertion*/
7383 case 'C': /* Single char !DANGEROUS! */
7384 case 'd': case 'D': /* digit class */
7385 case 'g': case 'G': /* generic-backref, pos assertion */
7386 case 'h': case 'H': /* HORIZWS */
7387 case 'k': case 'K': /* named backref, keep marker */
7388 case 'N': /* named char sequence */
7389 case 'p': case 'P': /* Unicode property */
7390 case 'R': /* LNBREAK */
7391 case 's': case 'S': /* space class */
7392 case 'v': case 'V': /* VERTWS */
7393 case 'w': case 'W': /* word class */
7394 case 'X': /* eXtended Unicode "combining character sequence" */
7395 case 'z': case 'Z': /* End of line/string assertion */
7399 /* Anything after here is an escape that resolves to a
7400 literal. (Except digits, which may or may not)
7419 ender = ASCII_TO_NATIVE('\033');
7423 ender = ASCII_TO_NATIVE('\007');
7428 char* const e = strchr(p, '}');
7432 vFAIL("Missing right brace on \\x{}");
7435 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7436 | PERL_SCAN_DISALLOW_PREFIX;
7437 STRLEN numlen = e - p - 1;
7438 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7445 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7447 ender = grok_hex(p, &numlen, &flags, NULL);
7450 if (PL_encoding && ender < 0x100)
7451 goto recode_encoding;
7455 ender = UCHARAT(p++);
7456 ender = toCTRL(ender);
7458 case '0': case '1': case '2': case '3':case '4':
7459 case '5': case '6': case '7': case '8':case '9':
7461 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7464 ender = grok_oct(p, &numlen, &flags, NULL);
7466 /* An octal above 0xff is interpreted differently
7467 * depending on if the re is in utf8 or not. If it
7468 * is in utf8, the value will be itself, otherwise
7469 * it is interpreted as modulo 0x100. It has been
7470 * decided to discourage the use of octal above the
7471 * single-byte range. For now, warn only when
7472 * it ends up modulo */
7473 if (SIZE_ONLY && ender >= 0x100
7474 && ! UTF && ! PL_encoding
7475 && ckWARN2(WARN_DEPRECATED, WARN_REGEXP)) {
7476 vWARNdep(p, "Use of octal value above 377 is deprecated");
7484 if (PL_encoding && ender < 0x100)
7485 goto recode_encoding;
7489 SV* enc = PL_encoding;
7490 ender = reg_recode((const char)(U8)ender, &enc);
7491 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
7492 vWARN(p, "Invalid escape in the specified encoding");
7498 FAIL("Trailing \\");
7501 if (!SIZE_ONLY&& isALPHA(*p) && ckWARN(WARN_REGEXP))
7502 vWARN2(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7503 goto normal_default;
7508 if (UTF8_IS_START(*p) && UTF) {
7510 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7511 &numlen, UTF8_ALLOW_DEFAULT);
7518 if ( RExC_flags & RXf_PMf_EXTENDED)
7519 p = regwhite( pRExC_state, p );
7521 /* Prime the casefolded buffer. */
7522 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7524 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7529 /* Emit all the Unicode characters. */
7531 for (foldbuf = tmpbuf;
7533 foldlen -= numlen) {
7534 ender = utf8_to_uvchr(foldbuf, &numlen);
7536 const STRLEN unilen = reguni(pRExC_state, ender, s);
7539 /* In EBCDIC the numlen
7540 * and unilen can differ. */
7542 if (numlen >= foldlen)
7546 break; /* "Can't happen." */
7550 const STRLEN unilen = reguni(pRExC_state, ender, s);
7559 REGC((char)ender, s++);
7565 /* Emit all the Unicode characters. */
7567 for (foldbuf = tmpbuf;
7569 foldlen -= numlen) {
7570 ender = utf8_to_uvchr(foldbuf, &numlen);
7572 const STRLEN unilen = reguni(pRExC_state, ender, s);
7575 /* In EBCDIC the numlen
7576 * and unilen can differ. */
7578 if (numlen >= foldlen)
7586 const STRLEN unilen = reguni(pRExC_state, ender, s);
7595 REGC((char)ender, s++);
7599 Set_Node_Cur_Length(ret); /* MJD */
7600 nextchar(pRExC_state);
7602 /* len is STRLEN which is unsigned, need to copy to signed */
7605 vFAIL("Internal disaster");
7609 if (len == 1 && UNI_IS_INVARIANT(ender))
7613 RExC_size += STR_SZ(len);
7616 RExC_emit += STR_SZ(len);
7626 S_regwhite( RExC_state_t *pRExC_state, char *p )
7628 const char *e = RExC_end;
7630 PERL_ARGS_ASSERT_REGWHITE;
7635 else if (*p == '#') {
7644 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7652 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7653 Character classes ([:foo:]) can also be negated ([:^foo:]).
7654 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7655 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7656 but trigger failures because they are currently unimplemented. */
7658 #define POSIXCC_DONE(c) ((c) == ':')
7659 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7660 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7663 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7666 I32 namedclass = OOB_NAMEDCLASS;
7668 PERL_ARGS_ASSERT_REGPPOSIXCC;
7670 if (value == '[' && RExC_parse + 1 < RExC_end &&
7671 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7672 POSIXCC(UCHARAT(RExC_parse))) {
7673 const char c = UCHARAT(RExC_parse);
7674 char* const s = RExC_parse++;
7676 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7678 if (RExC_parse == RExC_end)
7679 /* Grandfather lone [:, [=, [. */
7682 const char* const t = RExC_parse++; /* skip over the c */
7685 if (UCHARAT(RExC_parse) == ']') {
7686 const char *posixcc = s + 1;
7687 RExC_parse++; /* skip over the ending ] */
7690 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7691 const I32 skip = t - posixcc;
7693 /* Initially switch on the length of the name. */
7696 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7697 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7700 /* Names all of length 5. */
7701 /* alnum alpha ascii blank cntrl digit graph lower
7702 print punct space upper */
7703 /* Offset 4 gives the best switch position. */
7704 switch (posixcc[4]) {
7706 if (memEQ(posixcc, "alph", 4)) /* alpha */
7707 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7710 if (memEQ(posixcc, "spac", 4)) /* space */
7711 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7714 if (memEQ(posixcc, "grap", 4)) /* graph */
7715 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7718 if (memEQ(posixcc, "asci", 4)) /* ascii */
7719 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7722 if (memEQ(posixcc, "blan", 4)) /* blank */
7723 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7726 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7727 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7730 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7731 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7734 if (memEQ(posixcc, "lowe", 4)) /* lower */
7735 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7736 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7737 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7740 if (memEQ(posixcc, "digi", 4)) /* digit */
7741 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7742 else if (memEQ(posixcc, "prin", 4)) /* print */
7743 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7744 else if (memEQ(posixcc, "punc", 4)) /* punct */
7745 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7750 if (memEQ(posixcc, "xdigit", 6))
7751 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7755 if (namedclass == OOB_NAMEDCLASS)
7756 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7758 assert (posixcc[skip] == ':');
7759 assert (posixcc[skip+1] == ']');
7760 } else if (!SIZE_ONLY) {
7761 /* [[=foo=]] and [[.foo.]] are still future. */
7763 /* adjust RExC_parse so the warning shows after
7765 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7767 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7770 /* Maternal grandfather:
7771 * "[:" ending in ":" but not in ":]" */
7781 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7785 PERL_ARGS_ASSERT_CHECKPOSIXCC;
7787 if (POSIXCC(UCHARAT(RExC_parse))) {
7788 const char *s = RExC_parse;
7789 const char c = *s++;
7793 if (*s && c == *s && s[1] == ']') {
7794 if (ckWARN(WARN_REGEXP))
7796 "POSIX syntax [%c %c] belongs inside character classes",
7799 /* [[=foo=]] and [[.foo.]] are still future. */
7800 if (POSIXCC_NOTYET(c)) {
7801 /* adjust RExC_parse so the error shows after
7803 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7805 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7812 #define _C_C_T_(NAME,TEST,WORD) \
7815 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7817 for (value = 0; value < 256; value++) \
7819 ANYOF_BITMAP_SET(ret, value); \
7824 case ANYOF_N##NAME: \
7826 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7828 for (value = 0; value < 256; value++) \
7830 ANYOF_BITMAP_SET(ret, value); \
7836 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7838 for (value = 0; value < 256; value++) \
7840 ANYOF_BITMAP_SET(ret, value); \
7844 case ANYOF_N##NAME: \
7845 for (value = 0; value < 256; value++) \
7847 ANYOF_BITMAP_SET(ret, value); \
7853 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
7854 so that it is possible to override the option here without having to
7855 rebuild the entire core. as we are required to do if we change regcomp.h
7856 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
7858 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
7859 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
7862 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
7863 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
7865 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
7869 parse a class specification and produce either an ANYOF node that
7870 matches the pattern or if the pattern matches a single char only and
7871 that char is < 256 and we are case insensitive then we produce an
7876 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7879 register UV nextvalue;
7880 register IV prevvalue = OOB_UNICODE;
7881 register IV range = 0;
7882 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7883 register regnode *ret;
7886 char *rangebegin = NULL;
7887 bool need_class = 0;
7890 bool optimize_invert = TRUE;
7891 AV* unicode_alternate = NULL;
7893 UV literal_endpoint = 0;
7895 UV stored = 0; /* number of chars stored in the class */
7897 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7898 case we need to change the emitted regop to an EXACT. */
7899 const char * orig_parse = RExC_parse;
7900 GET_RE_DEBUG_FLAGS_DECL;
7902 PERL_ARGS_ASSERT_REGCLASS;
7904 PERL_UNUSED_ARG(depth);
7907 DEBUG_PARSE("clas");
7909 /* Assume we are going to generate an ANYOF node. */
7910 ret = reganode(pRExC_state, ANYOF, 0);
7913 ANYOF_FLAGS(ret) = 0;
7915 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7919 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7923 RExC_size += ANYOF_SKIP;
7924 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7927 RExC_emit += ANYOF_SKIP;
7929 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7931 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7932 ANYOF_BITMAP_ZERO(ret);
7933 listsv = newSVpvs("# comment\n");
7936 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7938 if (!SIZE_ONLY && POSIXCC(nextvalue))
7939 checkposixcc(pRExC_state);
7941 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7942 if (UCHARAT(RExC_parse) == ']')
7946 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7950 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7953 rangebegin = RExC_parse;
7955 value = utf8n_to_uvchr((U8*)RExC_parse,
7956 RExC_end - RExC_parse,
7957 &numlen, UTF8_ALLOW_DEFAULT);
7958 RExC_parse += numlen;
7961 value = UCHARAT(RExC_parse++);
7963 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7964 if (value == '[' && POSIXCC(nextvalue))
7965 namedclass = regpposixcc(pRExC_state, value);
7966 else if (value == '\\') {
7968 value = utf8n_to_uvchr((U8*)RExC_parse,
7969 RExC_end - RExC_parse,
7970 &numlen, UTF8_ALLOW_DEFAULT);
7971 RExC_parse += numlen;
7974 value = UCHARAT(RExC_parse++);
7975 /* Some compilers cannot handle switching on 64-bit integer
7976 * values, therefore value cannot be an UV. Yes, this will
7977 * be a problem later if we want switch on Unicode.
7978 * A similar issue a little bit later when switching on
7979 * namedclass. --jhi */
7980 switch ((I32)value) {
7981 case 'w': namedclass = ANYOF_ALNUM; break;
7982 case 'W': namedclass = ANYOF_NALNUM; break;
7983 case 's': namedclass = ANYOF_SPACE; break;
7984 case 'S': namedclass = ANYOF_NSPACE; break;
7985 case 'd': namedclass = ANYOF_DIGIT; break;
7986 case 'D': namedclass = ANYOF_NDIGIT; break;
7987 case 'v': namedclass = ANYOF_VERTWS; break;
7988 case 'V': namedclass = ANYOF_NVERTWS; break;
7989 case 'h': namedclass = ANYOF_HORIZWS; break;
7990 case 'H': namedclass = ANYOF_NHORIZWS; break;
7991 case 'N': /* Handle \N{NAME} in class */
7993 /* We only pay attention to the first char of
7994 multichar strings being returned. I kinda wonder
7995 if this makes sense as it does change the behaviour
7996 from earlier versions, OTOH that behaviour was broken
7998 UV v; /* value is register so we cant & it /grrr */
7999 if (reg_namedseq(pRExC_state, &v, NULL)) {
8009 if (RExC_parse >= RExC_end)
8010 vFAIL2("Empty \\%c{}", (U8)value);
8011 if (*RExC_parse == '{') {
8012 const U8 c = (U8)value;
8013 e = strchr(RExC_parse++, '}');
8015 vFAIL2("Missing right brace on \\%c{}", c);
8016 while (isSPACE(UCHARAT(RExC_parse)))
8018 if (e == RExC_parse)
8019 vFAIL2("Empty \\%c{}", c);
8021 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
8029 if (UCHARAT(RExC_parse) == '^') {
8032 value = value == 'p' ? 'P' : 'p'; /* toggle */
8033 while (isSPACE(UCHARAT(RExC_parse))) {
8038 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
8039 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
8042 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8043 namedclass = ANYOF_MAX; /* no official name, but it's named */
8046 case 'n': value = '\n'; break;
8047 case 'r': value = '\r'; break;
8048 case 't': value = '\t'; break;
8049 case 'f': value = '\f'; break;
8050 case 'b': value = '\b'; break;
8051 case 'e': value = ASCII_TO_NATIVE('\033');break;
8052 case 'a': value = ASCII_TO_NATIVE('\007');break;
8054 if (*RExC_parse == '{') {
8055 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8056 | PERL_SCAN_DISALLOW_PREFIX;
8057 char * const e = strchr(RExC_parse++, '}');
8059 vFAIL("Missing right brace on \\x{}");
8061 numlen = e - RExC_parse;
8062 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8066 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8068 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8069 RExC_parse += numlen;
8071 if (PL_encoding && value < 0x100)
8072 goto recode_encoding;
8075 value = UCHARAT(RExC_parse++);
8076 value = toCTRL(value);
8078 case '0': case '1': case '2': case '3': case '4':
8079 case '5': case '6': case '7': case '8': case '9':
8083 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8084 RExC_parse += numlen;
8085 if (PL_encoding && value < 0x100)
8086 goto recode_encoding;
8091 SV* enc = PL_encoding;
8092 value = reg_recode((const char)(U8)value, &enc);
8093 if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
8095 "Invalid escape in the specified encoding");
8099 if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
8101 "Unrecognized escape \\%c in character class passed through",
8105 } /* end of \blah */
8111 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8113 if (!SIZE_ONLY && !need_class)
8114 ANYOF_CLASS_ZERO(ret);
8118 /* a bad range like a-\d, a-[:digit:] ? */
8121 if (ckWARN(WARN_REGEXP)) {
8123 RExC_parse >= rangebegin ?
8124 RExC_parse - rangebegin : 0;
8126 "False [] range \"%*.*s\"",
8129 if (prevvalue < 256) {
8130 ANYOF_BITMAP_SET(ret, prevvalue);
8131 ANYOF_BITMAP_SET(ret, '-');
8134 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8135 Perl_sv_catpvf(aTHX_ listsv,
8136 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8140 range = 0; /* this was not a true range */
8146 const char *what = NULL;
8149 if (namedclass > OOB_NAMEDCLASS)
8150 optimize_invert = FALSE;
8151 /* Possible truncation here but in some 64-bit environments
8152 * the compiler gets heartburn about switch on 64-bit values.
8153 * A similar issue a little earlier when switching on value.
8155 switch ((I32)namedclass) {
8157 case _C_C_T_(ALNUMC, isALNUMC(value), POSIX_CC_UNI_NAME("Alnum"));
8158 case _C_C_T_(ALPHA, isALPHA(value), POSIX_CC_UNI_NAME("Alpha"));
8159 case _C_C_T_(BLANK, isBLANK(value), POSIX_CC_UNI_NAME("Blank"));
8160 case _C_C_T_(CNTRL, isCNTRL(value), POSIX_CC_UNI_NAME("Cntrl"));
8161 case _C_C_T_(GRAPH, isGRAPH(value), POSIX_CC_UNI_NAME("Graph"));
8162 case _C_C_T_(LOWER, isLOWER(value), POSIX_CC_UNI_NAME("Lower"));
8163 case _C_C_T_(PRINT, isPRINT(value), POSIX_CC_UNI_NAME("Print"));
8164 case _C_C_T_(PSXSPC, isPSXSPC(value), POSIX_CC_UNI_NAME("Space"));
8165 case _C_C_T_(PUNCT, isPUNCT(value), POSIX_CC_UNI_NAME("Punct"));
8166 case _C_C_T_(UPPER, isUPPER(value), POSIX_CC_UNI_NAME("Upper"));
8167 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8168 case _C_C_T_(ALNUM, isALNUM(value), "Word");
8169 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
8171 case _C_C_T_(SPACE, isSPACE(value), "PerlSpace");
8172 case _C_C_T_(ALNUM, isALNUM(value), "PerlWord");
8174 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
8175 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8176 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8179 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8182 for (value = 0; value < 128; value++)
8183 ANYOF_BITMAP_SET(ret, value);
8185 for (value = 0; value < 256; value++) {
8187 ANYOF_BITMAP_SET(ret, value);
8196 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8199 for (value = 128; value < 256; value++)
8200 ANYOF_BITMAP_SET(ret, value);
8202 for (value = 0; value < 256; value++) {
8203 if (!isASCII(value))
8204 ANYOF_BITMAP_SET(ret, value);
8213 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8215 /* consecutive digits assumed */
8216 for (value = '0'; value <= '9'; value++)
8217 ANYOF_BITMAP_SET(ret, value);
8220 what = POSIX_CC_UNI_NAME("Digit");
8224 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8226 /* consecutive digits assumed */
8227 for (value = 0; value < '0'; value++)
8228 ANYOF_BITMAP_SET(ret, value);
8229 for (value = '9' + 1; value < 256; value++)
8230 ANYOF_BITMAP_SET(ret, value);
8233 what = POSIX_CC_UNI_NAME("Digit");
8236 /* this is to handle \p and \P */
8239 vFAIL("Invalid [::] class");
8243 /* Strings such as "+utf8::isWord\n" */
8244 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8247 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8250 } /* end of namedclass \blah */
8253 if (prevvalue > (IV)value) /* b-a */ {
8254 const int w = RExC_parse - rangebegin;
8255 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8256 range = 0; /* not a valid range */
8260 prevvalue = value; /* save the beginning of the range */
8261 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8262 RExC_parse[1] != ']') {
8265 /* a bad range like \w-, [:word:]- ? */
8266 if (namedclass > OOB_NAMEDCLASS) {
8267 if (ckWARN(WARN_REGEXP)) {
8269 RExC_parse >= rangebegin ?
8270 RExC_parse - rangebegin : 0;
8272 "False [] range \"%*.*s\"",
8276 ANYOF_BITMAP_SET(ret, '-');
8278 range = 1; /* yeah, it's a range! */
8279 continue; /* but do it the next time */
8283 /* now is the next time */
8284 /*stored += (value - prevvalue + 1);*/
8286 if (prevvalue < 256) {
8287 const IV ceilvalue = value < 256 ? value : 255;
8290 /* In EBCDIC [\x89-\x91] should include
8291 * the \x8e but [i-j] should not. */
8292 if (literal_endpoint == 2 &&
8293 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8294 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8296 if (isLOWER(prevvalue)) {
8297 for (i = prevvalue; i <= ceilvalue; i++)
8298 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8300 ANYOF_BITMAP_SET(ret, i);
8303 for (i = prevvalue; i <= ceilvalue; i++)
8304 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8306 ANYOF_BITMAP_SET(ret, i);
8312 for (i = prevvalue; i <= ceilvalue; i++) {
8313 if (!ANYOF_BITMAP_TEST(ret,i)) {
8315 ANYOF_BITMAP_SET(ret, i);
8319 if (value > 255 || UTF) {
8320 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8321 const UV natvalue = NATIVE_TO_UNI(value);
8322 stored+=2; /* can't optimize this class */
8323 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8324 if (prevnatvalue < natvalue) { /* what about > ? */
8325 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8326 prevnatvalue, natvalue);
8328 else if (prevnatvalue == natvalue) {
8329 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8331 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8333 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8335 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8336 if (RExC_precomp[0] == ':' &&
8337 RExC_precomp[1] == '[' &&
8338 (f == 0xDF || f == 0x92)) {
8339 f = NATIVE_TO_UNI(f);
8342 /* If folding and foldable and a single
8343 * character, insert also the folded version
8344 * to the charclass. */
8346 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8347 if ((RExC_precomp[0] == ':' &&
8348 RExC_precomp[1] == '[' &&
8350 (value == 0xFB05 || value == 0xFB06))) ?
8351 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8352 foldlen == (STRLEN)UNISKIP(f) )
8354 if (foldlen == (STRLEN)UNISKIP(f))
8356 Perl_sv_catpvf(aTHX_ listsv,
8359 /* Any multicharacter foldings
8360 * require the following transform:
8361 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8362 * where E folds into "pq" and F folds
8363 * into "rst", all other characters
8364 * fold to single characters. We save
8365 * away these multicharacter foldings,
8366 * to be later saved as part of the
8367 * additional "s" data. */
8370 if (!unicode_alternate)
8371 unicode_alternate = newAV();
8372 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8374 av_push(unicode_alternate, sv);
8378 /* If folding and the value is one of the Greek
8379 * sigmas insert a few more sigmas to make the
8380 * folding rules of the sigmas to work right.
8381 * Note that not all the possible combinations
8382 * are handled here: some of them are handled
8383 * by the standard folding rules, and some of
8384 * them (literal or EXACTF cases) are handled
8385 * during runtime in regexec.c:S_find_byclass(). */
8386 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8387 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8388 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8389 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8390 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8392 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8393 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8394 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8399 literal_endpoint = 0;
8403 range = 0; /* this range (if it was one) is done now */
8407 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8409 RExC_size += ANYOF_CLASS_ADD_SKIP;
8411 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8417 /****** !SIZE_ONLY AFTER HERE *********/
8419 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8420 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8422 /* optimize single char class to an EXACT node
8423 but *only* when its not a UTF/high char */
8424 const char * cur_parse= RExC_parse;
8425 RExC_emit = (regnode *)orig_emit;
8426 RExC_parse = (char *)orig_parse;
8427 ret = reg_node(pRExC_state,
8428 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8429 RExC_parse = (char *)cur_parse;
8430 *STRING(ret)= (char)value;
8432 RExC_emit += STR_SZ(1);
8434 SvREFCNT_dec(listsv);
8438 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8439 if ( /* If the only flag is folding (plus possibly inversion). */
8440 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8442 for (value = 0; value < 256; ++value) {
8443 if (ANYOF_BITMAP_TEST(ret, value)) {
8444 UV fold = PL_fold[value];
8447 ANYOF_BITMAP_SET(ret, fold);
8450 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8453 /* optimize inverted simple patterns (e.g. [^a-z]) */
8454 if (optimize_invert &&
8455 /* If the only flag is inversion. */
8456 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8457 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8458 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8459 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8462 AV * const av = newAV();
8464 /* The 0th element stores the character class description
8465 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8466 * to initialize the appropriate swash (which gets stored in
8467 * the 1st element), and also useful for dumping the regnode.
8468 * The 2nd element stores the multicharacter foldings,
8469 * used later (regexec.c:S_reginclass()). */
8470 av_store(av, 0, listsv);
8471 av_store(av, 1, NULL);
8472 av_store(av, 2, MUTABLE_SV(unicode_alternate));
8473 rv = newRV_noinc(MUTABLE_SV(av));
8474 n = add_data(pRExC_state, 1, "s");
8475 RExC_rxi->data->data[n] = (void*)rv;
8483 /* reg_skipcomment()
8485 Absorbs an /x style # comments from the input stream.
8486 Returns true if there is more text remaining in the stream.
8487 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8488 terminates the pattern without including a newline.
8490 Note its the callers responsibility to ensure that we are
8496 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8500 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8502 while (RExC_parse < RExC_end)
8503 if (*RExC_parse++ == '\n') {
8508 /* we ran off the end of the pattern without ending
8509 the comment, so we have to add an \n when wrapping */
8510 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8518 Advance that parse position, and optionall absorbs
8519 "whitespace" from the inputstream.
8521 Without /x "whitespace" means (?#...) style comments only,
8522 with /x this means (?#...) and # comments and whitespace proper.
8524 Returns the RExC_parse point from BEFORE the scan occurs.
8526 This is the /x friendly way of saying RExC_parse++.
8530 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8532 char* const retval = RExC_parse++;
8534 PERL_ARGS_ASSERT_NEXTCHAR;
8537 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8538 RExC_parse[2] == '#') {
8539 while (*RExC_parse != ')') {
8540 if (RExC_parse == RExC_end)
8541 FAIL("Sequence (?#... not terminated");
8547 if (RExC_flags & RXf_PMf_EXTENDED) {
8548 if (isSPACE(*RExC_parse)) {
8552 else if (*RExC_parse == '#') {
8553 if ( reg_skipcomment( pRExC_state ) )
8562 - reg_node - emit a node
8564 STATIC regnode * /* Location. */
8565 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8568 register regnode *ptr;
8569 regnode * const ret = RExC_emit;
8570 GET_RE_DEBUG_FLAGS_DECL;
8572 PERL_ARGS_ASSERT_REG_NODE;
8575 SIZE_ALIGN(RExC_size);
8579 if (RExC_emit >= RExC_emit_bound)
8580 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8582 NODE_ALIGN_FILL(ret);
8584 FILL_ADVANCE_NODE(ptr, op);
8585 #ifdef RE_TRACK_PATTERN_OFFSETS
8586 if (RExC_offsets) { /* MJD */
8587 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8588 "reg_node", __LINE__,
8590 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8591 ? "Overwriting end of array!\n" : "OK",
8592 (UV)(RExC_emit - RExC_emit_start),
8593 (UV)(RExC_parse - RExC_start),
8594 (UV)RExC_offsets[0]));
8595 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8603 - reganode - emit a node with an argument
8605 STATIC regnode * /* Location. */
8606 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8609 register regnode *ptr;
8610 regnode * const ret = RExC_emit;
8611 GET_RE_DEBUG_FLAGS_DECL;
8613 PERL_ARGS_ASSERT_REGANODE;
8616 SIZE_ALIGN(RExC_size);
8621 assert(2==regarglen[op]+1);
8623 Anything larger than this has to allocate the extra amount.
8624 If we changed this to be:
8626 RExC_size += (1 + regarglen[op]);
8628 then it wouldn't matter. Its not clear what side effect
8629 might come from that so its not done so far.
8634 if (RExC_emit >= RExC_emit_bound)
8635 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8637 NODE_ALIGN_FILL(ret);
8639 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8640 #ifdef RE_TRACK_PATTERN_OFFSETS
8641 if (RExC_offsets) { /* MJD */
8642 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8646 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8647 "Overwriting end of array!\n" : "OK",
8648 (UV)(RExC_emit - RExC_emit_start),
8649 (UV)(RExC_parse - RExC_start),
8650 (UV)RExC_offsets[0]));
8651 Set_Cur_Node_Offset;
8659 - reguni - emit (if appropriate) a Unicode character
8662 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8666 PERL_ARGS_ASSERT_REGUNI;
8668 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8672 - reginsert - insert an operator in front of already-emitted operand
8674 * Means relocating the operand.
8677 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8680 register regnode *src;
8681 register regnode *dst;
8682 register regnode *place;
8683 const int offset = regarglen[(U8)op];
8684 const int size = NODE_STEP_REGNODE + offset;
8685 GET_RE_DEBUG_FLAGS_DECL;
8687 PERL_ARGS_ASSERT_REGINSERT;
8688 PERL_UNUSED_ARG(depth);
8689 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8690 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8699 if (RExC_open_parens) {
8701 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8702 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8703 if ( RExC_open_parens[paren] >= opnd ) {
8704 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8705 RExC_open_parens[paren] += size;
8707 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8709 if ( RExC_close_parens[paren] >= opnd ) {
8710 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8711 RExC_close_parens[paren] += size;
8713 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8718 while (src > opnd) {
8719 StructCopy(--src, --dst, regnode);
8720 #ifdef RE_TRACK_PATTERN_OFFSETS
8721 if (RExC_offsets) { /* MJD 20010112 */
8722 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8726 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8727 ? "Overwriting end of array!\n" : "OK",
8728 (UV)(src - RExC_emit_start),
8729 (UV)(dst - RExC_emit_start),
8730 (UV)RExC_offsets[0]));
8731 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8732 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8738 place = opnd; /* Op node, where operand used to be. */
8739 #ifdef RE_TRACK_PATTERN_OFFSETS
8740 if (RExC_offsets) { /* MJD */
8741 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8745 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8746 ? "Overwriting end of array!\n" : "OK",
8747 (UV)(place - RExC_emit_start),
8748 (UV)(RExC_parse - RExC_start),
8749 (UV)RExC_offsets[0]));
8750 Set_Node_Offset(place, RExC_parse);
8751 Set_Node_Length(place, 1);
8754 src = NEXTOPER(place);
8755 FILL_ADVANCE_NODE(place, op);
8756 Zero(src, offset, regnode);
8760 - regtail - set the next-pointer at the end of a node chain of p to val.
8761 - SEE ALSO: regtail_study
8763 /* TODO: All three parms should be const */
8765 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8768 register regnode *scan;
8769 GET_RE_DEBUG_FLAGS_DECL;
8771 PERL_ARGS_ASSERT_REGTAIL;
8773 PERL_UNUSED_ARG(depth);
8779 /* Find last node. */
8782 regnode * const temp = regnext(scan);
8784 SV * const mysv=sv_newmortal();
8785 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8786 regprop(RExC_rx, mysv, scan);
8787 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8788 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8789 (temp == NULL ? "->" : ""),
8790 (temp == NULL ? PL_reg_name[OP(val)] : "")
8798 if (reg_off_by_arg[OP(scan)]) {
8799 ARG_SET(scan, val - scan);
8802 NEXT_OFF(scan) = val - scan;
8808 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8809 - Look for optimizable sequences at the same time.
8810 - currently only looks for EXACT chains.
8812 This is expermental code. The idea is to use this routine to perform
8813 in place optimizations on branches and groups as they are constructed,
8814 with the long term intention of removing optimization from study_chunk so
8815 that it is purely analytical.
8817 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8818 to control which is which.
8821 /* TODO: All four parms should be const */
8824 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8827 register regnode *scan;
8829 #ifdef EXPERIMENTAL_INPLACESCAN
8832 GET_RE_DEBUG_FLAGS_DECL;
8834 PERL_ARGS_ASSERT_REGTAIL_STUDY;
8840 /* Find last node. */
8844 regnode * const temp = regnext(scan);
8845 #ifdef EXPERIMENTAL_INPLACESCAN
8846 if (PL_regkind[OP(scan)] == EXACT)
8847 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8855 if( exact == PSEUDO )
8857 else if ( exact != OP(scan) )
8866 SV * const mysv=sv_newmortal();
8867 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8868 regprop(RExC_rx, mysv, scan);
8869 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8870 SvPV_nolen_const(mysv),
8872 PL_reg_name[exact]);
8879 SV * const mysv_val=sv_newmortal();
8880 DEBUG_PARSE_MSG("");
8881 regprop(RExC_rx, mysv_val, val);
8882 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8883 SvPV_nolen_const(mysv_val),
8884 (IV)REG_NODE_NUM(val),
8888 if (reg_off_by_arg[OP(scan)]) {
8889 ARG_SET(scan, val - scan);
8892 NEXT_OFF(scan) = val - scan;
8900 - regcurly - a little FSA that accepts {\d+,?\d*}
8903 S_regcurly(register const char *s)
8905 PERL_ARGS_ASSERT_REGCURLY;
8924 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8928 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
8933 for (bit=0; bit<32; bit++) {
8934 if (flags & (1<<bit)) {
8936 PerlIO_printf(Perl_debug_log, "%s",lead);
8937 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
8942 PerlIO_printf(Perl_debug_log, "\n");
8944 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
8950 Perl_regdump(pTHX_ const regexp *r)
8954 SV * const sv = sv_newmortal();
8955 SV *dsv= sv_newmortal();
8957 GET_RE_DEBUG_FLAGS_DECL;
8959 PERL_ARGS_ASSERT_REGDUMP;
8961 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8963 /* Header fields of interest. */
8964 if (r->anchored_substr) {
8965 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8966 RE_SV_DUMPLEN(r->anchored_substr), 30);
8967 PerlIO_printf(Perl_debug_log,
8968 "anchored %s%s at %"IVdf" ",
8969 s, RE_SV_TAIL(r->anchored_substr),
8970 (IV)r->anchored_offset);
8971 } else if (r->anchored_utf8) {
8972 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
8973 RE_SV_DUMPLEN(r->anchored_utf8), 30);
8974 PerlIO_printf(Perl_debug_log,
8975 "anchored utf8 %s%s at %"IVdf" ",
8976 s, RE_SV_TAIL(r->anchored_utf8),
8977 (IV)r->anchored_offset);
8979 if (r->float_substr) {
8980 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
8981 RE_SV_DUMPLEN(r->float_substr), 30);
8982 PerlIO_printf(Perl_debug_log,
8983 "floating %s%s at %"IVdf"..%"UVuf" ",
8984 s, RE_SV_TAIL(r->float_substr),
8985 (IV)r->float_min_offset, (UV)r->float_max_offset);
8986 } else if (r->float_utf8) {
8987 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
8988 RE_SV_DUMPLEN(r->float_utf8), 30);
8989 PerlIO_printf(Perl_debug_log,
8990 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
8991 s, RE_SV_TAIL(r->float_utf8),
8992 (IV)r->float_min_offset, (UV)r->float_max_offset);
8994 if (r->check_substr || r->check_utf8)
8995 PerlIO_printf(Perl_debug_log,
8997 (r->check_substr == r->float_substr
8998 && r->check_utf8 == r->float_utf8
8999 ? "(checking floating" : "(checking anchored"));
9000 if (r->extflags & RXf_NOSCAN)
9001 PerlIO_printf(Perl_debug_log, " noscan");
9002 if (r->extflags & RXf_CHECK_ALL)
9003 PerlIO_printf(Perl_debug_log, " isall");
9004 if (r->check_substr || r->check_utf8)
9005 PerlIO_printf(Perl_debug_log, ") ");
9007 if (ri->regstclass) {
9008 regprop(r, sv, ri->regstclass);
9009 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
9011 if (r->extflags & RXf_ANCH) {
9012 PerlIO_printf(Perl_debug_log, "anchored");
9013 if (r->extflags & RXf_ANCH_BOL)
9014 PerlIO_printf(Perl_debug_log, "(BOL)");
9015 if (r->extflags & RXf_ANCH_MBOL)
9016 PerlIO_printf(Perl_debug_log, "(MBOL)");
9017 if (r->extflags & RXf_ANCH_SBOL)
9018 PerlIO_printf(Perl_debug_log, "(SBOL)");
9019 if (r->extflags & RXf_ANCH_GPOS)
9020 PerlIO_printf(Perl_debug_log, "(GPOS)");
9021 PerlIO_putc(Perl_debug_log, ' ');
9023 if (r->extflags & RXf_GPOS_SEEN)
9024 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
9025 if (r->intflags & PREGf_SKIP)
9026 PerlIO_printf(Perl_debug_log, "plus ");
9027 if (r->intflags & PREGf_IMPLICIT)
9028 PerlIO_printf(Perl_debug_log, "implicit ");
9029 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
9030 if (r->extflags & RXf_EVAL_SEEN)
9031 PerlIO_printf(Perl_debug_log, "with eval ");
9032 PerlIO_printf(Perl_debug_log, "\n");
9033 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
9035 PERL_ARGS_ASSERT_REGDUMP;
9036 PERL_UNUSED_CONTEXT;
9038 #endif /* DEBUGGING */
9042 - regprop - printable representation of opcode
9044 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9047 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9048 if (flags & ANYOF_INVERT) \
9049 /*make sure the invert info is in each */ \
9050 sv_catpvs(sv, "^"); \
9056 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9061 RXi_GET_DECL(prog,progi);
9062 GET_RE_DEBUG_FLAGS_DECL;
9064 PERL_ARGS_ASSERT_REGPROP;
9068 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9069 /* It would be nice to FAIL() here, but this may be called from
9070 regexec.c, and it would be hard to supply pRExC_state. */
9071 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9072 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9074 k = PL_regkind[OP(o)];
9078 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9079 * is a crude hack but it may be the best for now since
9080 * we have no flag "this EXACTish node was UTF-8"
9082 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9083 PERL_PV_ESCAPE_UNI_DETECT |
9084 PERL_PV_PRETTY_ELLIPSES |
9085 PERL_PV_PRETTY_LTGT |
9086 PERL_PV_PRETTY_NOCLEAR
9088 } else if (k == TRIE) {
9089 /* print the details of the trie in dumpuntil instead, as
9090 * progi->data isn't available here */
9091 const char op = OP(o);
9092 const U32 n = ARG(o);
9093 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9094 (reg_ac_data *)progi->data->data[n] :
9096 const reg_trie_data * const trie
9097 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9099 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9100 DEBUG_TRIE_COMPILE_r(
9101 Perl_sv_catpvf(aTHX_ sv,
9102 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9103 (UV)trie->startstate,
9104 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9105 (UV)trie->wordcount,
9108 (UV)TRIE_CHARCOUNT(trie),
9109 (UV)trie->uniquecharcount
9112 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9114 int rangestart = -1;
9115 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9117 for (i = 0; i <= 256; i++) {
9118 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9119 if (rangestart == -1)
9121 } else if (rangestart != -1) {
9122 if (i <= rangestart + 3)
9123 for (; rangestart < i; rangestart++)
9124 put_byte(sv, rangestart);
9126 put_byte(sv, rangestart);
9128 put_byte(sv, i - 1);
9136 } else if (k == CURLY) {
9137 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9138 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9139 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9141 else if (k == WHILEM && o->flags) /* Ordinal/of */
9142 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9143 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9144 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9145 if ( RXp_PAREN_NAMES(prog) ) {
9146 if ( k != REF || OP(o) < NREF) {
9147 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9148 SV **name= av_fetch(list, ARG(o), 0 );
9150 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9153 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9154 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9155 I32 *nums=(I32*)SvPVX(sv_dat);
9156 SV **name= av_fetch(list, nums[0], 0 );
9159 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9160 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9161 (n ? "," : ""), (IV)nums[n]);
9163 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9167 } else if (k == GOSUB)
9168 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9169 else if (k == VERB) {
9171 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9172 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9173 } else if (k == LOGICAL)
9174 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9175 else if (k == FOLDCHAR)
9176 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9177 else if (k == ANYOF) {
9178 int i, rangestart = -1;
9179 const U8 flags = ANYOF_FLAGS(o);
9182 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9183 static const char * const anyofs[] = {
9216 if (flags & ANYOF_LOCALE)
9217 sv_catpvs(sv, "{loc}");
9218 if (flags & ANYOF_FOLD)
9219 sv_catpvs(sv, "{i}");
9220 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9221 if (flags & ANYOF_INVERT)
9224 /* output what the standard cp 0-255 bitmap matches */
9225 for (i = 0; i <= 256; i++) {
9226 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9227 if (rangestart == -1)
9229 } else if (rangestart != -1) {
9230 if (i <= rangestart + 3)
9231 for (; rangestart < i; rangestart++)
9232 put_byte(sv, rangestart);
9234 put_byte(sv, rangestart);
9236 put_byte(sv, i - 1);
9243 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9244 /* output any special charclass tests (used mostly under use locale) */
9245 if (o->flags & ANYOF_CLASS)
9246 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9247 if (ANYOF_CLASS_TEST(o,i)) {
9248 sv_catpv(sv, anyofs[i]);
9252 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9254 /* output information about the unicode matching */
9255 if (flags & ANYOF_UNICODE)
9256 sv_catpvs(sv, "{unicode}");
9257 else if (flags & ANYOF_UNICODE_ALL)
9258 sv_catpvs(sv, "{unicode_all}");
9262 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9266 U8 s[UTF8_MAXBYTES_CASE+1];
9268 for (i = 0; i <= 256; i++) { /* just the first 256 */
9269 uvchr_to_utf8(s, i);
9271 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9272 if (rangestart == -1)
9274 } else if (rangestart != -1) {
9275 if (i <= rangestart + 3)
9276 for (; rangestart < i; rangestart++) {
9277 const U8 * const e = uvchr_to_utf8(s,rangestart);
9279 for(p = s; p < e; p++)
9283 const U8 *e = uvchr_to_utf8(s,rangestart);
9285 for (p = s; p < e; p++)
9288 e = uvchr_to_utf8(s, i-1);
9289 for (p = s; p < e; p++)
9296 sv_catpvs(sv, "..."); /* et cetera */
9300 char *s = savesvpv(lv);
9301 char * const origs = s;
9303 while (*s && *s != '\n')
9307 const char * const t = ++s;
9325 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9327 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9328 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9330 PERL_UNUSED_CONTEXT;
9331 PERL_UNUSED_ARG(sv);
9333 PERL_UNUSED_ARG(prog);
9334 #endif /* DEBUGGING */
9338 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9339 { /* Assume that RE_INTUIT is set */
9341 struct regexp *const prog = (struct regexp *)SvANY(r);
9342 GET_RE_DEBUG_FLAGS_DECL;
9344 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9345 PERL_UNUSED_CONTEXT;
9349 const char * const s = SvPV_nolen_const(prog->check_substr
9350 ? prog->check_substr : prog->check_utf8);
9352 if (!PL_colorset) reginitcolors();
9353 PerlIO_printf(Perl_debug_log,
9354 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9356 prog->check_substr ? "" : "utf8 ",
9357 PL_colors[5],PL_colors[0],
9360 (strlen(s) > 60 ? "..." : ""));
9363 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9369 handles refcounting and freeing the perl core regexp structure. When
9370 it is necessary to actually free the structure the first thing it
9371 does is call the 'free' method of the regexp_engine associated to to
9372 the regexp, allowing the handling of the void *pprivate; member
9373 first. (This routine is not overridable by extensions, which is why
9374 the extensions free is called first.)
9376 See regdupe and regdupe_internal if you change anything here.
9378 #ifndef PERL_IN_XSUB_RE
9380 Perl_pregfree(pTHX_ REGEXP *r)
9386 Perl_pregfree2(pTHX_ REGEXP *rx)
9389 struct regexp *const r = (struct regexp *)SvANY(rx);
9390 GET_RE_DEBUG_FLAGS_DECL;
9392 PERL_ARGS_ASSERT_PREGFREE2;
9395 ReREFCNT_dec(r->mother_re);
9397 CALLREGFREE_PVT(rx); /* free the private data */
9398 if (RXp_PAREN_NAMES(r))
9399 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9402 if (r->anchored_substr)
9403 SvREFCNT_dec(r->anchored_substr);
9404 if (r->anchored_utf8)
9405 SvREFCNT_dec(r->anchored_utf8);
9406 if (r->float_substr)
9407 SvREFCNT_dec(r->float_substr);
9409 SvREFCNT_dec(r->float_utf8);
9410 Safefree(r->substrs);
9412 RX_MATCH_COPY_FREE(rx);
9413 #ifdef PERL_OLD_COPY_ON_WRITE
9415 SvREFCNT_dec(r->saved_copy);
9422 This is a hacky workaround to the structural issue of match results
9423 being stored in the regexp structure which is in turn stored in
9424 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9425 could be PL_curpm in multiple contexts, and could require multiple
9426 result sets being associated with the pattern simultaneously, such
9427 as when doing a recursive match with (??{$qr})
9429 The solution is to make a lightweight copy of the regexp structure
9430 when a qr// is returned from the code executed by (??{$qr}) this
9431 lightweight copy doesnt actually own any of its data except for
9432 the starp/end and the actual regexp structure itself.
9438 Perl_reg_temp_copy (pTHX_ REGEXP *rx)
9440 REGEXP *ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9441 struct regexp *ret = (struct regexp *)SvANY(ret_x);
9442 struct regexp *const r = (struct regexp *)SvANY(rx);
9443 register const I32 npar = r->nparens+1;
9445 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9447 (void)ReREFCNT_inc(rx);
9448 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9449 by pointing directly at the buffer, but flagging that the allocated
9450 space in the copy is zero. As we've just done a struct copy, it's now
9451 a case of zero-ing that, rather than copying the current length. */
9452 SvPV_set(ret_x, RX_WRAPPED(rx));
9453 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9454 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
9455 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
9456 SvLEN_set(ret_x, 0);
9457 Newx(ret->offs, npar, regexp_paren_pair);
9458 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9460 Newx(ret->substrs, 1, struct reg_substr_data);
9461 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9463 SvREFCNT_inc_void(ret->anchored_substr);
9464 SvREFCNT_inc_void(ret->anchored_utf8);
9465 SvREFCNT_inc_void(ret->float_substr);
9466 SvREFCNT_inc_void(ret->float_utf8);
9468 /* check_substr and check_utf8, if non-NULL, point to either their
9469 anchored or float namesakes, and don't hold a second reference. */
9471 RX_MATCH_COPIED_off(ret_x);
9472 #ifdef PERL_OLD_COPY_ON_WRITE
9473 ret->saved_copy = NULL;
9475 ret->mother_re = rx;
9481 /* regfree_internal()
9483 Free the private data in a regexp. This is overloadable by
9484 extensions. Perl takes care of the regexp structure in pregfree(),
9485 this covers the *pprivate pointer which technically perldoesnt
9486 know about, however of course we have to handle the
9487 regexp_internal structure when no extension is in use.
9489 Note this is called before freeing anything in the regexp
9494 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9497 struct regexp *const r = (struct regexp *)SvANY(rx);
9499 GET_RE_DEBUG_FLAGS_DECL;
9501 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9507 SV *dsv= sv_newmortal();
9508 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9509 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9510 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9511 PL_colors[4],PL_colors[5],s);
9514 #ifdef RE_TRACK_PATTERN_OFFSETS
9516 Safefree(ri->u.offsets); /* 20010421 MJD */
9519 int n = ri->data->count;
9520 PAD* new_comppad = NULL;
9525 /* If you add a ->what type here, update the comment in regcomp.h */
9526 switch (ri->data->what[n]) {
9530 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
9533 Safefree(ri->data->data[n]);
9536 new_comppad = MUTABLE_AV(ri->data->data[n]);
9539 if (new_comppad == NULL)
9540 Perl_croak(aTHX_ "panic: pregfree comppad");
9541 PAD_SAVE_LOCAL(old_comppad,
9542 /* Watch out for global destruction's random ordering. */
9543 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9546 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9549 op_free((OP_4tree*)ri->data->data[n]);
9551 PAD_RESTORE_LOCAL(old_comppad);
9552 SvREFCNT_dec(MUTABLE_SV(new_comppad));
9558 { /* Aho Corasick add-on structure for a trie node.
9559 Used in stclass optimization only */
9561 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9563 refcount = --aho->refcount;
9566 PerlMemShared_free(aho->states);
9567 PerlMemShared_free(aho->fail);
9568 /* do this last!!!! */
9569 PerlMemShared_free(ri->data->data[n]);
9570 PerlMemShared_free(ri->regstclass);
9576 /* trie structure. */
9578 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9580 refcount = --trie->refcount;
9583 PerlMemShared_free(trie->charmap);
9584 PerlMemShared_free(trie->states);
9585 PerlMemShared_free(trie->trans);
9587 PerlMemShared_free(trie->bitmap);
9589 PerlMemShared_free(trie->wordlen);
9591 PerlMemShared_free(trie->jump);
9593 PerlMemShared_free(trie->nextword);
9594 /* do this last!!!! */
9595 PerlMemShared_free(ri->data->data[n]);
9600 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9603 Safefree(ri->data->what);
9610 #define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
9611 #define av_dup_inc(s,t) MUTABLE_AV(SvREFCNT_inc(sv_dup((const SV *)s,t)))
9612 #define hv_dup_inc(s,t) MUTABLE_HV(SvREFCNT_inc(sv_dup((const SV *)s,t)))
9613 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9616 re_dup - duplicate a regexp.
9618 This routine is expected to clone a given regexp structure. It is only
9619 compiled under USE_ITHREADS.
9621 After all of the core data stored in struct regexp is duplicated
9622 the regexp_engine.dupe method is used to copy any private data
9623 stored in the *pprivate pointer. This allows extensions to handle
9624 any duplication it needs to do.
9626 See pregfree() and regfree_internal() if you change anything here.
9628 #if defined(USE_ITHREADS)
9629 #ifndef PERL_IN_XSUB_RE
9631 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9635 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9636 struct regexp *ret = (struct regexp *)SvANY(dstr);
9638 PERL_ARGS_ASSERT_RE_DUP_GUTS;
9640 npar = r->nparens+1;
9641 Newx(ret->offs, npar, regexp_paren_pair);
9642 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9644 /* no need to copy these */
9645 Newx(ret->swap, npar, regexp_paren_pair);
9649 /* Do it this way to avoid reading from *r after the StructCopy().
9650 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9651 cache, it doesn't matter. */
9652 const bool anchored = r->check_substr
9653 ? r->check_substr == r->anchored_substr
9654 : r->check_utf8 == r->anchored_utf8;
9655 Newx(ret->substrs, 1, struct reg_substr_data);
9656 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9658 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9659 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9660 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9661 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9663 /* check_substr and check_utf8, if non-NULL, point to either their
9664 anchored or float namesakes, and don't hold a second reference. */
9666 if (ret->check_substr) {
9668 assert(r->check_utf8 == r->anchored_utf8);
9669 ret->check_substr = ret->anchored_substr;
9670 ret->check_utf8 = ret->anchored_utf8;
9672 assert(r->check_substr == r->float_substr);
9673 assert(r->check_utf8 == r->float_utf8);
9674 ret->check_substr = ret->float_substr;
9675 ret->check_utf8 = ret->float_utf8;
9677 } else if (ret->check_utf8) {
9679 ret->check_utf8 = ret->anchored_utf8;
9681 ret->check_utf8 = ret->float_utf8;
9686 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9689 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9691 if (RX_MATCH_COPIED(dstr))
9692 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9695 #ifdef PERL_OLD_COPY_ON_WRITE
9696 ret->saved_copy = NULL;
9699 ret->mother_re = NULL;
9702 #endif /* PERL_IN_XSUB_RE */
9707 This is the internal complement to regdupe() which is used to copy
9708 the structure pointed to by the *pprivate pointer in the regexp.
9709 This is the core version of the extension overridable cloning hook.
9710 The regexp structure being duplicated will be copied by perl prior
9711 to this and will be provided as the regexp *r argument, however
9712 with the /old/ structures pprivate pointer value. Thus this routine
9713 may override any copying normally done by perl.
9715 It returns a pointer to the new regexp_internal structure.
9719 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
9722 struct regexp *const r = (struct regexp *)SvANY(rx);
9723 regexp_internal *reti;
9727 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
9729 npar = r->nparens+1;
9732 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
9733 Copy(ri->program, reti->program, len+1, regnode);
9736 reti->regstclass = NULL;
9740 const int count = ri->data->count;
9743 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9744 char, struct reg_data);
9745 Newx(d->what, count, U8);
9748 for (i = 0; i < count; i++) {
9749 d->what[i] = ri->data->what[i];
9750 switch (d->what[i]) {
9751 /* legal options are one of: sSfpontTu
9752 see also regcomp.h and pregfree() */
9755 case 'p': /* actually an AV, but the dup function is identical. */
9756 case 'u': /* actually an HV, but the dup function is identical. */
9757 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
9760 /* This is cheating. */
9761 Newx(d->data[i], 1, struct regnode_charclass_class);
9762 StructCopy(ri->data->data[i], d->data[i],
9763 struct regnode_charclass_class);
9764 reti->regstclass = (regnode*)d->data[i];
9767 /* Compiled op trees are readonly and in shared memory,
9768 and can thus be shared without duplication. */
9770 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9774 /* Trie stclasses are readonly and can thus be shared
9775 * without duplication. We free the stclass in pregfree
9776 * when the corresponding reg_ac_data struct is freed.
9778 reti->regstclass= ri->regstclass;
9782 ((reg_trie_data*)ri->data->data[i])->refcount++;
9786 d->data[i] = ri->data->data[i];
9789 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9798 reti->name_list_idx = ri->name_list_idx;
9800 #ifdef RE_TRACK_PATTERN_OFFSETS
9801 if (ri->u.offsets) {
9802 Newx(reti->u.offsets, 2*len+1, U32);
9803 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9806 SetProgLen(reti,len);
9812 #endif /* USE_ITHREADS */
9814 #ifndef PERL_IN_XSUB_RE
9817 - regnext - dig the "next" pointer out of a node
9820 Perl_regnext(pTHX_ register regnode *p)
9823 register I32 offset;
9828 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9837 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9840 STRLEN l1 = strlen(pat1);
9841 STRLEN l2 = strlen(pat2);
9844 const char *message;
9846 PERL_ARGS_ASSERT_RE_CROAK2;
9852 Copy(pat1, buf, l1 , char);
9853 Copy(pat2, buf + l1, l2 , char);
9854 buf[l1 + l2] = '\n';
9855 buf[l1 + l2 + 1] = '\0';
9857 /* ANSI variant takes additional second argument */
9858 va_start(args, pat2);
9862 msv = vmess(buf, &args);
9864 message = SvPV_const(msv,l1);
9867 Copy(message, buf, l1 , char);
9868 buf[l1-1] = '\0'; /* Overwrite \n */
9869 Perl_croak(aTHX_ "%s", buf);
9872 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9874 #ifndef PERL_IN_XSUB_RE
9876 Perl_save_re_context(pTHX)
9880 struct re_save_state *state;
9882 SAVEVPTR(PL_curcop);
9883 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9885 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9886 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9887 SSPUSHINT(SAVEt_RE_STATE);
9889 Copy(&PL_reg_state, state, 1, struct re_save_state);
9891 PL_reg_start_tmp = 0;
9892 PL_reg_start_tmpl = 0;
9893 PL_reg_oldsaved = NULL;
9894 PL_reg_oldsavedlen = 0;
9896 PL_reg_leftiter = 0;
9897 PL_reg_poscache = NULL;
9898 PL_reg_poscache_size = 0;
9899 #ifdef PERL_OLD_COPY_ON_WRITE
9903 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9905 const REGEXP * const rx = PM_GETRE(PL_curpm);
9908 for (i = 1; i <= RX_NPARENS(rx); i++) {
9909 char digits[TYPE_CHARS(long)];
9910 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9911 GV *const *const gvp
9912 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9915 GV * const gv = *gvp;
9916 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9926 clear_re(pTHX_ void *r)
9929 ReREFCNT_dec((REGEXP *)r);
9935 S_put_byte(pTHX_ SV *sv, int c)
9937 PERL_ARGS_ASSERT_PUT_BYTE;
9939 /* Our definition of isPRINT() ignores locales, so only bytes that are
9940 not part of UTF-8 are considered printable. I assume that the same
9941 holds for UTF-EBCDIC.
9942 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
9943 which Wikipedia says:
9945 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
9946 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
9947 identical, to the ASCII delete (DEL) or rubout control character.
9948 ) So the old condition can be simplified to !isPRINT(c) */
9950 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9952 const char string = c;
9953 if (c == '-' || c == ']' || c == '\\' || c == '^')
9954 sv_catpvs(sv, "\\");
9955 sv_catpvn(sv, &string, 1);
9960 #define CLEAR_OPTSTART \
9961 if (optstart) STMT_START { \
9962 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
9966 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
9968 STATIC const regnode *
9969 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
9970 const regnode *last, const regnode *plast,
9971 SV* sv, I32 indent, U32 depth)
9974 register U8 op = PSEUDO; /* Arbitrary non-END op. */
9975 register const regnode *next;
9976 const regnode *optstart= NULL;
9979 GET_RE_DEBUG_FLAGS_DECL;
9981 PERL_ARGS_ASSERT_DUMPUNTIL;
9983 #ifdef DEBUG_DUMPUNTIL
9984 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
9985 last ? last-start : 0,plast ? plast-start : 0);
9988 if (plast && plast < last)
9991 while (PL_regkind[op] != END && (!last || node < last)) {
9992 /* While that wasn't END last time... */
9995 if (op == CLOSE || op == WHILEM)
9997 next = regnext((regnode *)node);
10000 if (OP(node) == OPTIMIZED) {
10001 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10008 regprop(r, sv, node);
10009 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10010 (int)(2*indent + 1), "", SvPVX_const(sv));
10012 if (OP(node) != OPTIMIZED) {
10013 if (next == NULL) /* Next ptr. */
10014 PerlIO_printf(Perl_debug_log, " (0)");
10015 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10016 PerlIO_printf(Perl_debug_log, " (FAIL)");
10018 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10019 (void)PerlIO_putc(Perl_debug_log, '\n');
10023 if (PL_regkind[(U8)op] == BRANCHJ) {
10026 register const regnode *nnode = (OP(next) == LONGJMP
10027 ? regnext((regnode *)next)
10029 if (last && nnode > last)
10031 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10034 else if (PL_regkind[(U8)op] == BRANCH) {
10036 DUMPUNTIL(NEXTOPER(node), next);
10038 else if ( PL_regkind[(U8)op] == TRIE ) {
10039 const regnode *this_trie = node;
10040 const char op = OP(node);
10041 const U32 n = ARG(node);
10042 const reg_ac_data * const ac = op>=AHOCORASICK ?
10043 (reg_ac_data *)ri->data->data[n] :
10045 const reg_trie_data * const trie =
10046 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10048 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10050 const regnode *nextbranch= NULL;
10053 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10054 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10056 PerlIO_printf(Perl_debug_log, "%*s%s ",
10057 (int)(2*(indent+3)), "",
10058 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10059 PL_colors[0], PL_colors[1],
10060 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10061 PERL_PV_PRETTY_ELLIPSES |
10062 PERL_PV_PRETTY_LTGT
10067 U16 dist= trie->jump[word_idx+1];
10068 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10069 (UV)((dist ? this_trie + dist : next) - start));
10072 nextbranch= this_trie + trie->jump[0];
10073 DUMPUNTIL(this_trie + dist, nextbranch);
10075 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10076 nextbranch= regnext((regnode *)nextbranch);
10078 PerlIO_printf(Perl_debug_log, "\n");
10081 if (last && next > last)
10086 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10087 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10088 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10090 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10092 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10094 else if ( op == PLUS || op == STAR) {
10095 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10097 else if (op == ANYOF) {
10098 /* arglen 1 + class block */
10099 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
10100 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10101 node = NEXTOPER(node);
10103 else if (PL_regkind[(U8)op] == EXACT) {
10104 /* Literal string, where present. */
10105 node += NODE_SZ_STR(node) - 1;
10106 node = NEXTOPER(node);
10109 node = NEXTOPER(node);
10110 node += regarglen[(U8)op];
10112 if (op == CURLYX || op == OPEN)
10116 #ifdef DEBUG_DUMPUNTIL
10117 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10122 #endif /* DEBUGGING */
10126 * c-indentation-style: bsd
10127 * c-basic-offset: 4
10128 * indent-tabs-mode: t
10131 * ex: set ts=8 sts=4 sw=4 noet:
projects / p5sagit/p5-mst-13.2.git / blob