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
88 #include "dquote_static.c"
95 # if defined(BUGGY_MSC6)
96 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
97 # pragma optimize("a",off)
98 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
99 # pragma optimize("w",on )
100 # endif /* BUGGY_MSC6 */
104 #define STATIC static
107 typedef struct RExC_state_t {
108 U32 flags; /* are we folding, multilining? */
109 char *precomp; /* uncompiled string. */
110 REGEXP *rx_sv; /* The SV that is the regexp. */
111 regexp *rx; /* perl core regexp structure */
112 regexp_internal *rxi; /* internal data for regexp object pprivate field */
113 char *start; /* Start of input for compile */
114 char *end; /* End of input for compile */
115 char *parse; /* Input-scan pointer. */
116 I32 whilem_seen; /* number of WHILEM in this expr */
117 regnode *emit_start; /* Start of emitted-code area */
118 regnode *emit_bound; /* First regnode outside of the allocated space */
119 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
120 I32 naughty; /* How bad is this pattern? */
121 I32 sawback; /* Did we see \1, ...? */
123 I32 size; /* Code size. */
124 I32 npar; /* Capture buffer count, (OPEN). */
125 I32 cpar; /* Capture buffer count, (CLOSE). */
126 I32 nestroot; /* root parens we are in - used by accept */
130 regnode **open_parens; /* pointers to open parens */
131 regnode **close_parens; /* pointers to close parens */
132 regnode *opend; /* END node in program */
133 I32 utf8; /* whether the pattern is utf8 or not */
134 I32 orig_utf8; /* whether the pattern was originally in utf8 */
135 /* XXX use this for future optimisation of case
136 * where pattern must be upgraded to utf8. */
137 HV *paren_names; /* Paren names */
139 regnode **recurse; /* Recurse regops */
140 I32 recurse_count; /* Number of recurse regops */
142 char *starttry; /* -Dr: where regtry was called. */
143 #define RExC_starttry (pRExC_state->starttry)
146 const char *lastparse;
148 AV *paren_name_list; /* idx -> name */
149 #define RExC_lastparse (pRExC_state->lastparse)
150 #define RExC_lastnum (pRExC_state->lastnum)
151 #define RExC_paren_name_list (pRExC_state->paren_name_list)
155 #define RExC_flags (pRExC_state->flags)
156 #define RExC_precomp (pRExC_state->precomp)
157 #define RExC_rx_sv (pRExC_state->rx_sv)
158 #define RExC_rx (pRExC_state->rx)
159 #define RExC_rxi (pRExC_state->rxi)
160 #define RExC_start (pRExC_state->start)
161 #define RExC_end (pRExC_state->end)
162 #define RExC_parse (pRExC_state->parse)
163 #define RExC_whilem_seen (pRExC_state->whilem_seen)
164 #ifdef RE_TRACK_PATTERN_OFFSETS
165 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
167 #define RExC_emit (pRExC_state->emit)
168 #define RExC_emit_start (pRExC_state->emit_start)
169 #define RExC_emit_bound (pRExC_state->emit_bound)
170 #define RExC_naughty (pRExC_state->naughty)
171 #define RExC_sawback (pRExC_state->sawback)
172 #define RExC_seen (pRExC_state->seen)
173 #define RExC_size (pRExC_state->size)
174 #define RExC_npar (pRExC_state->npar)
175 #define RExC_nestroot (pRExC_state->nestroot)
176 #define RExC_extralen (pRExC_state->extralen)
177 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
178 #define RExC_seen_evals (pRExC_state->seen_evals)
179 #define RExC_utf8 (pRExC_state->utf8)
180 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
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. */
202 /* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single
203 * character, and if utf8, must be invariant. */
205 #define SPSTART 0x04 /* Starts with * or +. */
206 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
207 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
209 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
211 /* whether trie related optimizations are enabled */
212 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
213 #define TRIE_STUDY_OPT
214 #define FULL_TRIE_STUDY
220 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
221 #define PBITVAL(paren) (1 << ((paren) & 7))
222 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
223 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
224 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
226 /* If not already in utf8, do a longjmp back to the beginning */
227 #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */
228 #define REQUIRE_UTF8 STMT_START { \
229 if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \
232 /* About scan_data_t.
234 During optimisation we recurse through the regexp program performing
235 various inplace (keyhole style) optimisations. In addition study_chunk
236 and scan_commit populate this data structure with information about
237 what strings MUST appear in the pattern. We look for the longest
238 string that must appear for at a fixed location, and we look for the
239 longest string that may appear at a floating location. So for instance
244 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
245 strings (because they follow a .* construct). study_chunk will identify
246 both FOO and BAR as being the longest fixed and floating strings respectively.
248 The strings can be composites, for instance
252 will result in a composite fixed substring 'foo'.
254 For each string some basic information is maintained:
256 - offset or min_offset
257 This is the position the string must appear at, or not before.
258 It also implicitly (when combined with minlenp) tells us how many
259 character must match before the string we are searching.
260 Likewise when combined with minlenp and the length of the string
261 tells us how many characters must appear after the string we have
265 Only used for floating strings. This is the rightmost point that
266 the string can appear at. Ifset to I32 max it indicates that the
267 string can occur infinitely far to the right.
270 A pointer to the minimum length of the pattern that the string
271 was found inside. This is important as in the case of positive
272 lookahead or positive lookbehind we can have multiple patterns
277 The minimum length of the pattern overall is 3, the minimum length
278 of the lookahead part is 3, but the minimum length of the part that
279 will actually match is 1. So 'FOO's minimum length is 3, but the
280 minimum length for the F is 1. This is important as the minimum length
281 is used to determine offsets in front of and behind the string being
282 looked for. Since strings can be composites this is the length of the
283 pattern at the time it was commited with a scan_commit. Note that
284 the length is calculated by study_chunk, so that the minimum lengths
285 are not known until the full pattern has been compiled, thus the
286 pointer to the value.
290 In the case of lookbehind the string being searched for can be
291 offset past the start point of the final matching string.
292 If this value was just blithely removed from the min_offset it would
293 invalidate some of the calculations for how many chars must match
294 before or after (as they are derived from min_offset and minlen and
295 the length of the string being searched for).
296 When the final pattern is compiled and the data is moved from the
297 scan_data_t structure into the regexp structure the information
298 about lookbehind is factored in, with the information that would
299 have been lost precalculated in the end_shift field for the
302 The fields pos_min and pos_delta are used to store the minimum offset
303 and the delta to the maximum offset at the current point in the pattern.
307 typedef struct scan_data_t {
308 /*I32 len_min; unused */
309 /*I32 len_delta; unused */
313 I32 last_end; /* min value, <0 unless valid. */
316 SV **longest; /* Either &l_fixed, or &l_float. */
317 SV *longest_fixed; /* longest fixed string found in pattern */
318 I32 offset_fixed; /* offset where it starts */
319 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
320 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
321 SV *longest_float; /* longest floating string found in pattern */
322 I32 offset_float_min; /* earliest point in string it can appear */
323 I32 offset_float_max; /* latest point in string it can appear */
324 I32 *minlen_float; /* pointer to the minlen relevent to the string */
325 I32 lookbehind_float; /* is the position of the string modified by LB */
329 struct regnode_charclass_class *start_class;
333 * Forward declarations for pregcomp()'s friends.
336 static const scan_data_t zero_scan_data =
337 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
339 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
340 #define SF_BEFORE_SEOL 0x0001
341 #define SF_BEFORE_MEOL 0x0002
342 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
343 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
346 # define SF_FIX_SHIFT_EOL (0+2)
347 # define SF_FL_SHIFT_EOL (0+4)
349 # define SF_FIX_SHIFT_EOL (+2)
350 # define SF_FL_SHIFT_EOL (+4)
353 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
354 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
356 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
357 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
358 #define SF_IS_INF 0x0040
359 #define SF_HAS_PAR 0x0080
360 #define SF_IN_PAR 0x0100
361 #define SF_HAS_EVAL 0x0200
362 #define SCF_DO_SUBSTR 0x0400
363 #define SCF_DO_STCLASS_AND 0x0800
364 #define SCF_DO_STCLASS_OR 0x1000
365 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
366 #define SCF_WHILEM_VISITED_POS 0x2000
368 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
369 #define SCF_SEEN_ACCEPT 0x8000
371 #define UTF cBOOL(RExC_utf8)
372 #define LOC cBOOL(RExC_flags & RXf_PMf_LOCALE)
373 #define UNI_SEMANTICS cBOOL(RExC_flags & RXf_PMf_UNICODE)
374 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
376 #define OOB_UNICODE 12345678
377 #define OOB_NAMEDCLASS -1
379 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
380 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
383 /* length of regex to show in messages that don't mark a position within */
384 #define RegexLengthToShowInErrorMessages 127
387 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
388 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
389 * op/pragma/warn/regcomp.
391 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
392 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
394 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
397 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
398 * arg. Show regex, up to a maximum length. If it's too long, chop and add
401 #define _FAIL(code) STMT_START { \
402 const char *ellipses = ""; \
403 IV len = RExC_end - RExC_precomp; \
406 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
407 if (len > RegexLengthToShowInErrorMessages) { \
408 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
409 len = RegexLengthToShowInErrorMessages - 10; \
415 #define FAIL(msg) _FAIL( \
416 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
417 msg, (int)len, RExC_precomp, ellipses))
419 #define FAIL2(msg,arg) _FAIL( \
420 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
421 arg, (int)len, RExC_precomp, ellipses))
424 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
426 #define Simple_vFAIL(m) STMT_START { \
427 const IV offset = RExC_parse - RExC_precomp; \
428 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
429 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
433 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
435 #define vFAIL(m) STMT_START { \
437 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
442 * Like Simple_vFAIL(), but accepts two arguments.
444 #define Simple_vFAIL2(m,a1) STMT_START { \
445 const IV offset = RExC_parse - RExC_precomp; \
446 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
447 (int)offset, RExC_precomp, RExC_precomp + offset); \
451 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
453 #define vFAIL2(m,a1) STMT_START { \
455 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
456 Simple_vFAIL2(m, a1); \
461 * Like Simple_vFAIL(), but accepts three arguments.
463 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
464 const IV offset = RExC_parse - RExC_precomp; \
465 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
466 (int)offset, RExC_precomp, RExC_precomp + offset); \
470 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
472 #define vFAIL3(m,a1,a2) STMT_START { \
474 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
475 Simple_vFAIL3(m, a1, a2); \
479 * Like Simple_vFAIL(), but accepts four arguments.
481 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
482 const IV offset = RExC_parse - RExC_precomp; \
483 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
484 (int)offset, RExC_precomp, RExC_precomp + offset); \
487 #define ckWARNreg(loc,m) STMT_START { \
488 const IV offset = loc - RExC_precomp; \
489 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 (int)offset, RExC_precomp, RExC_precomp + offset); \
493 #define ckWARNregdep(loc,m) STMT_START { \
494 const IV offset = loc - RExC_precomp; \
495 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
497 (int)offset, RExC_precomp, RExC_precomp + offset); \
500 #define ckWARN2reg(loc, m, a1) STMT_START { \
501 const IV offset = loc - RExC_precomp; \
502 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
503 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
506 #define vWARN3(loc, m, a1, a2) STMT_START { \
507 const IV offset = loc - RExC_precomp; \
508 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
509 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
512 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
513 const IV offset = loc - RExC_precomp; \
514 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
515 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
518 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
519 const IV offset = loc - RExC_precomp; \
520 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
521 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
524 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
525 const IV offset = loc - RExC_precomp; \
526 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
527 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
530 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
531 const IV offset = loc - RExC_precomp; \
532 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
533 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
537 /* Allow for side effects in s */
538 #define REGC(c,s) STMT_START { \
539 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
542 /* Macros for recording node offsets. 20001227 mjd@plover.com
543 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
544 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
545 * Element 0 holds the number n.
546 * Position is 1 indexed.
548 #ifndef RE_TRACK_PATTERN_OFFSETS
549 #define Set_Node_Offset_To_R(node,byte)
550 #define Set_Node_Offset(node,byte)
551 #define Set_Cur_Node_Offset
552 #define Set_Node_Length_To_R(node,len)
553 #define Set_Node_Length(node,len)
554 #define Set_Node_Cur_Length(node)
555 #define Node_Offset(n)
556 #define Node_Length(n)
557 #define Set_Node_Offset_Length(node,offset,len)
558 #define ProgLen(ri) ri->u.proglen
559 #define SetProgLen(ri,x) ri->u.proglen = x
561 #define ProgLen(ri) ri->u.offsets[0]
562 #define SetProgLen(ri,x) ri->u.offsets[0] = x
563 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
565 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
566 __LINE__, (int)(node), (int)(byte))); \
568 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
570 RExC_offsets[2*(node)-1] = (byte); \
575 #define Set_Node_Offset(node,byte) \
576 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
577 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
579 #define Set_Node_Length_To_R(node,len) STMT_START { \
581 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
582 __LINE__, (int)(node), (int)(len))); \
584 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
586 RExC_offsets[2*(node)] = (len); \
591 #define Set_Node_Length(node,len) \
592 Set_Node_Length_To_R((node)-RExC_emit_start, len)
593 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
594 #define Set_Node_Cur_Length(node) \
595 Set_Node_Length(node, RExC_parse - parse_start)
597 /* Get offsets and lengths */
598 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
599 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
601 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
602 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
603 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
607 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
608 #define EXPERIMENTAL_INPLACESCAN
609 #endif /*RE_TRACK_PATTERN_OFFSETS*/
611 #define DEBUG_STUDYDATA(str,data,depth) \
612 DEBUG_OPTIMISE_MORE_r(if(data){ \
613 PerlIO_printf(Perl_debug_log, \
614 "%*s" str "Pos:%"IVdf"/%"IVdf \
615 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
616 (int)(depth)*2, "", \
617 (IV)((data)->pos_min), \
618 (IV)((data)->pos_delta), \
619 (UV)((data)->flags), \
620 (IV)((data)->whilem_c), \
621 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
622 is_inf ? "INF " : "" \
624 if ((data)->last_found) \
625 PerlIO_printf(Perl_debug_log, \
626 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
627 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
628 SvPVX_const((data)->last_found), \
629 (IV)((data)->last_end), \
630 (IV)((data)->last_start_min), \
631 (IV)((data)->last_start_max), \
632 ((data)->longest && \
633 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
634 SvPVX_const((data)->longest_fixed), \
635 (IV)((data)->offset_fixed), \
636 ((data)->longest && \
637 (data)->longest==&((data)->longest_float)) ? "*" : "", \
638 SvPVX_const((data)->longest_float), \
639 (IV)((data)->offset_float_min), \
640 (IV)((data)->offset_float_max) \
642 PerlIO_printf(Perl_debug_log,"\n"); \
645 static void clear_re(pTHX_ void *r);
647 /* Mark that we cannot extend a found fixed substring at this point.
648 Update the longest found anchored substring and the longest found
649 floating substrings if needed. */
652 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
654 const STRLEN l = CHR_SVLEN(data->last_found);
655 const STRLEN old_l = CHR_SVLEN(*data->longest);
656 GET_RE_DEBUG_FLAGS_DECL;
658 PERL_ARGS_ASSERT_SCAN_COMMIT;
660 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
661 SvSetMagicSV(*data->longest, data->last_found);
662 if (*data->longest == data->longest_fixed) {
663 data->offset_fixed = l ? data->last_start_min : data->pos_min;
664 if (data->flags & SF_BEFORE_EOL)
666 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
668 data->flags &= ~SF_FIX_BEFORE_EOL;
669 data->minlen_fixed=minlenp;
670 data->lookbehind_fixed=0;
672 else { /* *data->longest == data->longest_float */
673 data->offset_float_min = l ? data->last_start_min : data->pos_min;
674 data->offset_float_max = (l
675 ? data->last_start_max
676 : data->pos_min + data->pos_delta);
677 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
678 data->offset_float_max = I32_MAX;
679 if (data->flags & SF_BEFORE_EOL)
681 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
683 data->flags &= ~SF_FL_BEFORE_EOL;
684 data->minlen_float=minlenp;
685 data->lookbehind_float=0;
688 SvCUR_set(data->last_found, 0);
690 SV * const sv = data->last_found;
691 if (SvUTF8(sv) && SvMAGICAL(sv)) {
692 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
698 data->flags &= ~SF_BEFORE_EOL;
699 DEBUG_STUDYDATA("commit: ",data,0);
702 /* Can match anything (initialization) */
704 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
706 PERL_ARGS_ASSERT_CL_ANYTHING;
708 ANYOF_CLASS_ZERO(cl);
709 ANYOF_BITMAP_SETALL(cl);
710 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
712 cl->flags |= ANYOF_LOCALE;
715 /* Can match anything (initialization) */
717 S_cl_is_anything(const struct regnode_charclass_class *cl)
721 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
723 for (value = 0; value <= ANYOF_MAX; value += 2)
724 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
726 if (!(cl->flags & ANYOF_UNICODE_ALL))
728 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
733 /* Can match anything (initialization) */
735 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
737 PERL_ARGS_ASSERT_CL_INIT;
739 Zero(cl, 1, struct regnode_charclass_class);
741 cl_anything(pRExC_state, cl);
745 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
747 PERL_ARGS_ASSERT_CL_INIT_ZERO;
749 Zero(cl, 1, struct regnode_charclass_class);
751 cl_anything(pRExC_state, cl);
753 cl->flags |= ANYOF_LOCALE;
756 /* 'And' a given class with another one. Can create false positives */
757 /* We assume that cl is not inverted */
759 S_cl_and(struct regnode_charclass_class *cl,
760 const struct regnode_charclass_class *and_with)
762 PERL_ARGS_ASSERT_CL_AND;
764 assert(and_with->type == ANYOF);
765 if (!(and_with->flags & ANYOF_CLASS)
766 && !(cl->flags & ANYOF_CLASS)
767 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
768 && !(and_with->flags & ANYOF_FOLD)
769 && !(cl->flags & ANYOF_FOLD)) {
772 if (and_with->flags & ANYOF_INVERT)
773 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
774 cl->bitmap[i] &= ~and_with->bitmap[i];
776 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
777 cl->bitmap[i] &= and_with->bitmap[i];
778 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
779 if (!(and_with->flags & ANYOF_EOS))
780 cl->flags &= ~ANYOF_EOS;
782 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
783 !(and_with->flags & ANYOF_INVERT)) {
784 cl->flags &= ~ANYOF_UNICODE_ALL;
785 cl->flags |= ANYOF_UNICODE;
786 ARG_SET(cl, ARG(and_with));
788 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
789 !(and_with->flags & ANYOF_INVERT))
790 cl->flags &= ~ANYOF_UNICODE_ALL;
791 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
792 !(and_with->flags & ANYOF_INVERT))
793 cl->flags &= ~ANYOF_UNICODE;
796 /* 'OR' a given class with another one. Can create false positives */
797 /* We assume that cl is not inverted */
799 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
801 PERL_ARGS_ASSERT_CL_OR;
803 if (or_with->flags & ANYOF_INVERT) {
805 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
806 * <= (B1 | !B2) | (CL1 | !CL2)
807 * which is wasteful if CL2 is small, but we ignore CL2:
808 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
809 * XXXX Can we handle case-fold? Unclear:
810 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
811 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
813 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
814 && !(or_with->flags & ANYOF_FOLD)
815 && !(cl->flags & ANYOF_FOLD) ) {
818 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
819 cl->bitmap[i] |= ~or_with->bitmap[i];
820 } /* XXXX: logic is complicated otherwise */
822 cl_anything(pRExC_state, cl);
825 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
826 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
827 && (!(or_with->flags & ANYOF_FOLD)
828 || (cl->flags & ANYOF_FOLD)) ) {
831 /* OR char bitmap and class bitmap separately */
832 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
833 cl->bitmap[i] |= or_with->bitmap[i];
834 if (or_with->flags & ANYOF_CLASS) {
835 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
836 cl->classflags[i] |= or_with->classflags[i];
837 cl->flags |= ANYOF_CLASS;
840 else { /* XXXX: logic is complicated, leave it along for a moment. */
841 cl_anything(pRExC_state, cl);
844 if (or_with->flags & ANYOF_EOS)
845 cl->flags |= ANYOF_EOS;
847 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
848 ARG(cl) != ARG(or_with)) {
849 cl->flags |= ANYOF_UNICODE_ALL;
850 cl->flags &= ~ANYOF_UNICODE;
852 if (or_with->flags & ANYOF_UNICODE_ALL) {
853 cl->flags |= ANYOF_UNICODE_ALL;
854 cl->flags &= ~ANYOF_UNICODE;
858 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
859 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
860 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
861 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
866 dump_trie(trie,widecharmap,revcharmap)
867 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
868 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
870 These routines dump out a trie in a somewhat readable format.
871 The _interim_ variants are used for debugging the interim
872 tables that are used to generate the final compressed
873 representation which is what dump_trie expects.
875 Part of the reason for their existance is to provide a form
876 of documentation as to how the different representations function.
881 Dumps the final compressed table form of the trie to Perl_debug_log.
882 Used for debugging make_trie().
886 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
887 AV *revcharmap, U32 depth)
890 SV *sv=sv_newmortal();
891 int colwidth= widecharmap ? 6 : 4;
893 GET_RE_DEBUG_FLAGS_DECL;
895 PERL_ARGS_ASSERT_DUMP_TRIE;
897 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
898 (int)depth * 2 + 2,"",
899 "Match","Base","Ofs" );
901 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
902 SV ** const tmp = av_fetch( revcharmap, state, 0);
904 PerlIO_printf( Perl_debug_log, "%*s",
906 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
907 PL_colors[0], PL_colors[1],
908 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
909 PERL_PV_ESCAPE_FIRSTCHAR
914 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
915 (int)depth * 2 + 2,"");
917 for( state = 0 ; state < trie->uniquecharcount ; state++ )
918 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
919 PerlIO_printf( Perl_debug_log, "\n");
921 for( state = 1 ; state < trie->statecount ; state++ ) {
922 const U32 base = trie->states[ state ].trans.base;
924 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
926 if ( trie->states[ state ].wordnum ) {
927 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
929 PerlIO_printf( Perl_debug_log, "%6s", "" );
932 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
937 while( ( base + ofs < trie->uniquecharcount ) ||
938 ( base + ofs - trie->uniquecharcount < trie->lasttrans
939 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
942 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
944 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
945 if ( ( base + ofs >= trie->uniquecharcount ) &&
946 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
947 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
949 PerlIO_printf( Perl_debug_log, "%*"UVXf,
951 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
953 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
957 PerlIO_printf( Perl_debug_log, "]");
960 PerlIO_printf( Perl_debug_log, "\n" );
962 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
963 for (word=1; word <= trie->wordcount; word++) {
964 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
965 (int)word, (int)(trie->wordinfo[word].prev),
966 (int)(trie->wordinfo[word].len));
968 PerlIO_printf(Perl_debug_log, "\n" );
971 Dumps a fully constructed but uncompressed trie in list form.
972 List tries normally only are used for construction when the number of
973 possible chars (trie->uniquecharcount) is very high.
974 Used for debugging make_trie().
977 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
978 HV *widecharmap, AV *revcharmap, U32 next_alloc,
982 SV *sv=sv_newmortal();
983 int colwidth= widecharmap ? 6 : 4;
984 GET_RE_DEBUG_FLAGS_DECL;
986 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
988 /* print out the table precompression. */
989 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
990 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
991 "------:-----+-----------------\n" );
993 for( state=1 ; state < next_alloc ; state ++ ) {
996 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
997 (int)depth * 2 + 2,"", (UV)state );
998 if ( ! trie->states[ state ].wordnum ) {
999 PerlIO_printf( Perl_debug_log, "%5s| ","");
1001 PerlIO_printf( Perl_debug_log, "W%4x| ",
1002 trie->states[ state ].wordnum
1005 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1006 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
1008 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1010 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1011 PL_colors[0], PL_colors[1],
1012 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1013 PERL_PV_ESCAPE_FIRSTCHAR
1015 TRIE_LIST_ITEM(state,charid).forid,
1016 (UV)TRIE_LIST_ITEM(state,charid).newstate
1019 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1020 (int)((depth * 2) + 14), "");
1023 PerlIO_printf( Perl_debug_log, "\n");
1028 Dumps a fully constructed but uncompressed trie in table form.
1029 This is the normal DFA style state transition table, with a few
1030 twists to facilitate compression later.
1031 Used for debugging make_trie().
1034 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1035 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1040 SV *sv=sv_newmortal();
1041 int colwidth= widecharmap ? 6 : 4;
1042 GET_RE_DEBUG_FLAGS_DECL;
1044 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1047 print out the table precompression so that we can do a visual check
1048 that they are identical.
1051 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1053 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1054 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1056 PerlIO_printf( Perl_debug_log, "%*s",
1058 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1059 PL_colors[0], PL_colors[1],
1060 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1061 PERL_PV_ESCAPE_FIRSTCHAR
1067 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1069 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1070 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1073 PerlIO_printf( Perl_debug_log, "\n" );
1075 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1077 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1078 (int)depth * 2 + 2,"",
1079 (UV)TRIE_NODENUM( state ) );
1081 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1082 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1084 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1086 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1088 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1089 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1091 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1092 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1100 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1101 startbranch: the first branch in the whole branch sequence
1102 first : start branch of sequence of branch-exact nodes.
1103 May be the same as startbranch
1104 last : Thing following the last branch.
1105 May be the same as tail.
1106 tail : item following the branch sequence
1107 count : words in the sequence
1108 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1109 depth : indent depth
1111 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1113 A trie is an N'ary tree where the branches are determined by digital
1114 decomposition of the key. IE, at the root node you look up the 1st character and
1115 follow that branch repeat until you find the end of the branches. Nodes can be
1116 marked as "accepting" meaning they represent a complete word. Eg:
1120 would convert into the following structure. Numbers represent states, letters
1121 following numbers represent valid transitions on the letter from that state, if
1122 the number is in square brackets it represents an accepting state, otherwise it
1123 will be in parenthesis.
1125 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1129 (1) +-i->(6)-+-s->[7]
1131 +-s->(3)-+-h->(4)-+-e->[5]
1133 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1135 This shows that when matching against the string 'hers' we will begin at state 1
1136 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1137 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1138 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1139 single traverse. We store a mapping from accepting to state to which word was
1140 matched, and then when we have multiple possibilities we try to complete the
1141 rest of the regex in the order in which they occured in the alternation.
1143 The only prior NFA like behaviour that would be changed by the TRIE support is
1144 the silent ignoring of duplicate alternations which are of the form:
1146 / (DUPE|DUPE) X? (?{ ... }) Y /x
1148 Thus EVAL blocks follwing a trie may be called a different number of times with
1149 and without the optimisation. With the optimisations dupes will be silently
1150 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1151 the following demonstrates:
1153 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1155 which prints out 'word' three times, but
1157 'words'=~/(word|word|word)(?{ print $1 })S/
1159 which doesnt print it out at all. This is due to other optimisations kicking in.
1161 Example of what happens on a structural level:
1163 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1165 1: CURLYM[1] {1,32767}(18)
1176 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1177 and should turn into:
1179 1: CURLYM[1] {1,32767}(18)
1181 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1189 Cases where tail != last would be like /(?foo|bar)baz/:
1199 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1200 and would end up looking like:
1203 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1210 d = uvuni_to_utf8_flags(d, uv, 0);
1212 is the recommended Unicode-aware way of saying
1217 #define TRIE_STORE_REVCHAR \
1220 SV *zlopp = newSV(2); \
1221 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1222 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1223 SvCUR_set(zlopp, kapow - flrbbbbb); \
1226 av_push(revcharmap, zlopp); \
1228 char ooooff = (char)uvc; \
1229 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1233 #define TRIE_READ_CHAR STMT_START { \
1237 if ( foldlen > 0 ) { \
1238 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1243 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1244 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1245 foldlen -= UNISKIP( uvc ); \
1246 scan = foldbuf + UNISKIP( uvc ); \
1249 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1259 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1260 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1261 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1262 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1264 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1265 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1266 TRIE_LIST_CUR( state )++; \
1269 #define TRIE_LIST_NEW(state) STMT_START { \
1270 Newxz( trie->states[ state ].trans.list, \
1271 4, reg_trie_trans_le ); \
1272 TRIE_LIST_CUR( state ) = 1; \
1273 TRIE_LIST_LEN( state ) = 4; \
1276 #define TRIE_HANDLE_WORD(state) STMT_START { \
1277 U16 dupe= trie->states[ state ].wordnum; \
1278 regnode * const noper_next = regnext( noper ); \
1281 /* store the word for dumping */ \
1283 if (OP(noper) != NOTHING) \
1284 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1286 tmp = newSVpvn_utf8( "", 0, UTF ); \
1287 av_push( trie_words, tmp ); \
1291 trie->wordinfo[curword].prev = 0; \
1292 trie->wordinfo[curword].len = wordlen; \
1293 trie->wordinfo[curword].accept = state; \
1295 if ( noper_next < tail ) { \
1297 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1298 trie->jump[curword] = (U16)(noper_next - convert); \
1300 jumper = noper_next; \
1302 nextbranch= regnext(cur); \
1306 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1307 /* chain, so that when the bits of chain are later */\
1308 /* linked together, the dups appear in the chain */\
1309 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1310 trie->wordinfo[dupe].prev = curword; \
1312 /* we haven't inserted this word yet. */ \
1313 trie->states[ state ].wordnum = curword; \
1318 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1319 ( ( base + charid >= ucharcount \
1320 && base + charid < ubound \
1321 && state == trie->trans[ base - ucharcount + charid ].check \
1322 && trie->trans[ base - ucharcount + charid ].next ) \
1323 ? trie->trans[ base - ucharcount + charid ].next \
1324 : ( state==1 ? special : 0 ) \
1328 #define MADE_JUMP_TRIE 2
1329 #define MADE_EXACT_TRIE 4
1332 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1335 /* first pass, loop through and scan words */
1336 reg_trie_data *trie;
1337 HV *widecharmap = NULL;
1338 AV *revcharmap = newAV();
1340 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1345 regnode *jumper = NULL;
1346 regnode *nextbranch = NULL;
1347 regnode *convert = NULL;
1348 U32 *prev_states; /* temp array mapping each state to previous one */
1349 /* we just use folder as a flag in utf8 */
1350 const U8 * const folder = ( flags == EXACTF
1352 : ( flags == EXACTFL
1359 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1360 AV *trie_words = NULL;
1361 /* along with revcharmap, this only used during construction but both are
1362 * useful during debugging so we store them in the struct when debugging.
1365 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1366 STRLEN trie_charcount=0;
1368 SV *re_trie_maxbuff;
1369 GET_RE_DEBUG_FLAGS_DECL;
1371 PERL_ARGS_ASSERT_MAKE_TRIE;
1373 PERL_UNUSED_ARG(depth);
1376 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1378 trie->startstate = 1;
1379 trie->wordcount = word_count;
1380 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1381 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1382 if (!(UTF && folder))
1383 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1384 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1385 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1388 trie_words = newAV();
1391 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1392 if (!SvIOK(re_trie_maxbuff)) {
1393 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1396 PerlIO_printf( Perl_debug_log,
1397 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1398 (int)depth * 2 + 2, "",
1399 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1400 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1404 /* Find the node we are going to overwrite */
1405 if ( first == startbranch && OP( last ) != BRANCH ) {
1406 /* whole branch chain */
1409 /* branch sub-chain */
1410 convert = NEXTOPER( first );
1413 /* -- First loop and Setup --
1415 We first traverse the branches and scan each word to determine if it
1416 contains widechars, and how many unique chars there are, this is
1417 important as we have to build a table with at least as many columns as we
1420 We use an array of integers to represent the character codes 0..255
1421 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1422 native representation of the character value as the key and IV's for the
1425 *TODO* If we keep track of how many times each character is used we can
1426 remap the columns so that the table compression later on is more
1427 efficient in terms of memory by ensuring most common value is in the
1428 middle and the least common are on the outside. IMO this would be better
1429 than a most to least common mapping as theres a decent chance the most
1430 common letter will share a node with the least common, meaning the node
1431 will not be compressable. With a middle is most common approach the worst
1432 case is when we have the least common nodes twice.
1436 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1437 regnode * const noper = NEXTOPER( cur );
1438 const U8 *uc = (U8*)STRING( noper );
1439 const U8 * const e = uc + STR_LEN( noper );
1441 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1442 const U8 *scan = (U8*)NULL;
1443 U32 wordlen = 0; /* required init */
1445 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1447 if (OP(noper) == NOTHING) {
1451 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1452 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1453 regardless of encoding */
1455 for ( ; uc < e ; uc += len ) {
1456 TRIE_CHARCOUNT(trie)++;
1460 if ( !trie->charmap[ uvc ] ) {
1461 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1463 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1467 /* store the codepoint in the bitmap, and if its ascii
1468 also store its folded equivelent. */
1469 TRIE_BITMAP_SET(trie,uvc);
1471 /* store the folded codepoint */
1472 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1475 /* store first byte of utf8 representation of
1476 codepoints in the 127 < uvc < 256 range */
1477 if (127 < uvc && uvc < 192) {
1478 TRIE_BITMAP_SET(trie,194);
1479 } else if (191 < uvc ) {
1480 TRIE_BITMAP_SET(trie,195);
1481 /* && uvc < 256 -- we know uvc is < 256 already */
1484 set_bit = 0; /* We've done our bit :-) */
1489 widecharmap = newHV();
1491 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1494 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1496 if ( !SvTRUE( *svpp ) ) {
1497 sv_setiv( *svpp, ++trie->uniquecharcount );
1502 if( cur == first ) {
1505 } else if (chars < trie->minlen) {
1507 } else if (chars > trie->maxlen) {
1511 } /* end first pass */
1512 DEBUG_TRIE_COMPILE_r(
1513 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1514 (int)depth * 2 + 2,"",
1515 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1516 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1517 (int)trie->minlen, (int)trie->maxlen )
1521 We now know what we are dealing with in terms of unique chars and
1522 string sizes so we can calculate how much memory a naive
1523 representation using a flat table will take. If it's over a reasonable
1524 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1525 conservative but potentially much slower representation using an array
1528 At the end we convert both representations into the same compressed
1529 form that will be used in regexec.c for matching with. The latter
1530 is a form that cannot be used to construct with but has memory
1531 properties similar to the list form and access properties similar
1532 to the table form making it both suitable for fast searches and
1533 small enough that its feasable to store for the duration of a program.
1535 See the comment in the code where the compressed table is produced
1536 inplace from the flat tabe representation for an explanation of how
1537 the compression works.
1542 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1545 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1547 Second Pass -- Array Of Lists Representation
1549 Each state will be represented by a list of charid:state records
1550 (reg_trie_trans_le) the first such element holds the CUR and LEN
1551 points of the allocated array. (See defines above).
1553 We build the initial structure using the lists, and then convert
1554 it into the compressed table form which allows faster lookups
1555 (but cant be modified once converted).
1558 STRLEN transcount = 1;
1560 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1561 "%*sCompiling trie using list compiler\n",
1562 (int)depth * 2 + 2, ""));
1564 trie->states = (reg_trie_state *)
1565 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1566 sizeof(reg_trie_state) );
1570 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1572 regnode * const noper = NEXTOPER( cur );
1573 U8 *uc = (U8*)STRING( noper );
1574 const U8 * const e = uc + STR_LEN( noper );
1575 U32 state = 1; /* required init */
1576 U16 charid = 0; /* sanity init */
1577 U8 *scan = (U8*)NULL; /* sanity init */
1578 STRLEN foldlen = 0; /* required init */
1579 U32 wordlen = 0; /* required init */
1580 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1582 if (OP(noper) != NOTHING) {
1583 for ( ; uc < e ; uc += len ) {
1588 charid = trie->charmap[ uvc ];
1590 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1594 charid=(U16)SvIV( *svpp );
1597 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1604 if ( !trie->states[ state ].trans.list ) {
1605 TRIE_LIST_NEW( state );
1607 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1608 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1609 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1614 newstate = next_alloc++;
1615 prev_states[newstate] = state;
1616 TRIE_LIST_PUSH( state, charid, newstate );
1621 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1625 TRIE_HANDLE_WORD(state);
1627 } /* end second pass */
1629 /* next alloc is the NEXT state to be allocated */
1630 trie->statecount = next_alloc;
1631 trie->states = (reg_trie_state *)
1632 PerlMemShared_realloc( trie->states,
1634 * sizeof(reg_trie_state) );
1636 /* and now dump it out before we compress it */
1637 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1638 revcharmap, next_alloc,
1642 trie->trans = (reg_trie_trans *)
1643 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1650 for( state=1 ; state < next_alloc ; state ++ ) {
1654 DEBUG_TRIE_COMPILE_MORE_r(
1655 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1659 if (trie->states[state].trans.list) {
1660 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1664 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1665 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1666 if ( forid < minid ) {
1668 } else if ( forid > maxid ) {
1672 if ( transcount < tp + maxid - minid + 1) {
1674 trie->trans = (reg_trie_trans *)
1675 PerlMemShared_realloc( trie->trans,
1677 * sizeof(reg_trie_trans) );
1678 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1680 base = trie->uniquecharcount + tp - minid;
1681 if ( maxid == minid ) {
1683 for ( ; zp < tp ; zp++ ) {
1684 if ( ! trie->trans[ zp ].next ) {
1685 base = trie->uniquecharcount + zp - minid;
1686 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1687 trie->trans[ zp ].check = state;
1693 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1694 trie->trans[ tp ].check = state;
1699 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1700 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1701 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1702 trie->trans[ tid ].check = state;
1704 tp += ( maxid - minid + 1 );
1706 Safefree(trie->states[ state ].trans.list);
1709 DEBUG_TRIE_COMPILE_MORE_r(
1710 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1713 trie->states[ state ].trans.base=base;
1715 trie->lasttrans = tp + 1;
1719 Second Pass -- Flat Table Representation.
1721 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1722 We know that we will need Charcount+1 trans at most to store the data
1723 (one row per char at worst case) So we preallocate both structures
1724 assuming worst case.
1726 We then construct the trie using only the .next slots of the entry
1729 We use the .check field of the first entry of the node temporarily to
1730 make compression both faster and easier by keeping track of how many non
1731 zero fields are in the node.
1733 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1736 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1737 number representing the first entry of the node, and state as a
1738 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1739 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1740 are 2 entrys per node. eg:
1748 The table is internally in the right hand, idx form. However as we also
1749 have to deal with the states array which is indexed by nodenum we have to
1750 use TRIE_NODENUM() to convert.
1753 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1754 "%*sCompiling trie using table compiler\n",
1755 (int)depth * 2 + 2, ""));
1757 trie->trans = (reg_trie_trans *)
1758 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1759 * trie->uniquecharcount + 1,
1760 sizeof(reg_trie_trans) );
1761 trie->states = (reg_trie_state *)
1762 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1763 sizeof(reg_trie_state) );
1764 next_alloc = trie->uniquecharcount + 1;
1767 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1769 regnode * const noper = NEXTOPER( cur );
1770 const U8 *uc = (U8*)STRING( noper );
1771 const U8 * const e = uc + STR_LEN( noper );
1773 U32 state = 1; /* required init */
1775 U16 charid = 0; /* sanity init */
1776 U32 accept_state = 0; /* sanity init */
1777 U8 *scan = (U8*)NULL; /* sanity init */
1779 STRLEN foldlen = 0; /* required init */
1780 U32 wordlen = 0; /* required init */
1781 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1783 if ( OP(noper) != NOTHING ) {
1784 for ( ; uc < e ; uc += len ) {
1789 charid = trie->charmap[ uvc ];
1791 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1792 charid = svpp ? (U16)SvIV(*svpp) : 0;
1796 if ( !trie->trans[ state + charid ].next ) {
1797 trie->trans[ state + charid ].next = next_alloc;
1798 trie->trans[ state ].check++;
1799 prev_states[TRIE_NODENUM(next_alloc)]
1800 = TRIE_NODENUM(state);
1801 next_alloc += trie->uniquecharcount;
1803 state = trie->trans[ state + charid ].next;
1805 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1807 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1810 accept_state = TRIE_NODENUM( state );
1811 TRIE_HANDLE_WORD(accept_state);
1813 } /* end second pass */
1815 /* and now dump it out before we compress it */
1816 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1818 next_alloc, depth+1));
1822 * Inplace compress the table.*
1824 For sparse data sets the table constructed by the trie algorithm will
1825 be mostly 0/FAIL transitions or to put it another way mostly empty.
1826 (Note that leaf nodes will not contain any transitions.)
1828 This algorithm compresses the tables by eliminating most such
1829 transitions, at the cost of a modest bit of extra work during lookup:
1831 - Each states[] entry contains a .base field which indicates the
1832 index in the state[] array wheres its transition data is stored.
1834 - If .base is 0 there are no valid transitions from that node.
1836 - If .base is nonzero then charid is added to it to find an entry in
1839 -If trans[states[state].base+charid].check!=state then the
1840 transition is taken to be a 0/Fail transition. Thus if there are fail
1841 transitions at the front of the node then the .base offset will point
1842 somewhere inside the previous nodes data (or maybe even into a node
1843 even earlier), but the .check field determines if the transition is
1847 The following process inplace converts the table to the compressed
1848 table: We first do not compress the root node 1,and mark its all its
1849 .check pointers as 1 and set its .base pointer as 1 as well. This
1850 allows to do a DFA construction from the compressed table later, and
1851 ensures that any .base pointers we calculate later are greater than
1854 - We set 'pos' to indicate the first entry of the second node.
1856 - We then iterate over the columns of the node, finding the first and
1857 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1858 and set the .check pointers accordingly, and advance pos
1859 appropriately and repreat for the next node. Note that when we copy
1860 the next pointers we have to convert them from the original
1861 NODEIDX form to NODENUM form as the former is not valid post
1864 - If a node has no transitions used we mark its base as 0 and do not
1865 advance the pos pointer.
1867 - If a node only has one transition we use a second pointer into the
1868 structure to fill in allocated fail transitions from other states.
1869 This pointer is independent of the main pointer and scans forward
1870 looking for null transitions that are allocated to a state. When it
1871 finds one it writes the single transition into the "hole". If the
1872 pointer doesnt find one the single transition is appended as normal.
1874 - Once compressed we can Renew/realloc the structures to release the
1877 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1878 specifically Fig 3.47 and the associated pseudocode.
1882 const U32 laststate = TRIE_NODENUM( next_alloc );
1885 trie->statecount = laststate;
1887 for ( state = 1 ; state < laststate ; state++ ) {
1889 const U32 stateidx = TRIE_NODEIDX( state );
1890 const U32 o_used = trie->trans[ stateidx ].check;
1891 U32 used = trie->trans[ stateidx ].check;
1892 trie->trans[ stateidx ].check = 0;
1894 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1895 if ( flag || trie->trans[ stateidx + charid ].next ) {
1896 if ( trie->trans[ stateidx + charid ].next ) {
1898 for ( ; zp < pos ; zp++ ) {
1899 if ( ! trie->trans[ zp ].next ) {
1903 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1904 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1905 trie->trans[ zp ].check = state;
1906 if ( ++zp > pos ) pos = zp;
1913 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1915 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1916 trie->trans[ pos ].check = state;
1921 trie->lasttrans = pos + 1;
1922 trie->states = (reg_trie_state *)
1923 PerlMemShared_realloc( trie->states, laststate
1924 * sizeof(reg_trie_state) );
1925 DEBUG_TRIE_COMPILE_MORE_r(
1926 PerlIO_printf( Perl_debug_log,
1927 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1928 (int)depth * 2 + 2,"",
1929 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1932 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1935 } /* end table compress */
1937 DEBUG_TRIE_COMPILE_MORE_r(
1938 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1939 (int)depth * 2 + 2, "",
1940 (UV)trie->statecount,
1941 (UV)trie->lasttrans)
1943 /* resize the trans array to remove unused space */
1944 trie->trans = (reg_trie_trans *)
1945 PerlMemShared_realloc( trie->trans, trie->lasttrans
1946 * sizeof(reg_trie_trans) );
1948 { /* Modify the program and insert the new TRIE node*/
1949 U8 nodetype =(U8)(flags & 0xFF);
1953 regnode *optimize = NULL;
1954 #ifdef RE_TRACK_PATTERN_OFFSETS
1957 U32 mjd_nodelen = 0;
1958 #endif /* RE_TRACK_PATTERN_OFFSETS */
1959 #endif /* DEBUGGING */
1961 This means we convert either the first branch or the first Exact,
1962 depending on whether the thing following (in 'last') is a branch
1963 or not and whther first is the startbranch (ie is it a sub part of
1964 the alternation or is it the whole thing.)
1965 Assuming its a sub part we conver the EXACT otherwise we convert
1966 the whole branch sequence, including the first.
1968 /* Find the node we are going to overwrite */
1969 if ( first != startbranch || OP( last ) == BRANCH ) {
1970 /* branch sub-chain */
1971 NEXT_OFF( first ) = (U16)(last - first);
1972 #ifdef RE_TRACK_PATTERN_OFFSETS
1974 mjd_offset= Node_Offset((convert));
1975 mjd_nodelen= Node_Length((convert));
1978 /* whole branch chain */
1980 #ifdef RE_TRACK_PATTERN_OFFSETS
1983 const regnode *nop = NEXTOPER( convert );
1984 mjd_offset= Node_Offset((nop));
1985 mjd_nodelen= Node_Length((nop));
1989 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1990 (int)depth * 2 + 2, "",
1991 (UV)mjd_offset, (UV)mjd_nodelen)
1994 /* But first we check to see if there is a common prefix we can
1995 split out as an EXACT and put in front of the TRIE node. */
1996 trie->startstate= 1;
1997 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1999 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2003 const U32 base = trie->states[ state ].trans.base;
2005 if ( trie->states[state].wordnum )
2008 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2009 if ( ( base + ofs >= trie->uniquecharcount ) &&
2010 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2011 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2013 if ( ++count > 1 ) {
2014 SV **tmp = av_fetch( revcharmap, ofs, 0);
2015 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2016 if ( state == 1 ) break;
2018 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2020 PerlIO_printf(Perl_debug_log,
2021 "%*sNew Start State=%"UVuf" Class: [",
2022 (int)depth * 2 + 2, "",
2025 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2026 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2028 TRIE_BITMAP_SET(trie,*ch);
2030 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2032 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2036 TRIE_BITMAP_SET(trie,*ch);
2038 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2039 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2045 SV **tmp = av_fetch( revcharmap, idx, 0);
2047 char *ch = SvPV( *tmp, len );
2049 SV *sv=sv_newmortal();
2050 PerlIO_printf( Perl_debug_log,
2051 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2052 (int)depth * 2 + 2, "",
2054 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2055 PL_colors[0], PL_colors[1],
2056 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2057 PERL_PV_ESCAPE_FIRSTCHAR
2062 OP( convert ) = nodetype;
2063 str=STRING(convert);
2066 STR_LEN(convert) += len;
2072 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2077 trie->prefixlen = (state-1);
2079 regnode *n = convert+NODE_SZ_STR(convert);
2080 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2081 trie->startstate = state;
2082 trie->minlen -= (state - 1);
2083 trie->maxlen -= (state - 1);
2085 /* At least the UNICOS C compiler choked on this
2086 * being argument to DEBUG_r(), so let's just have
2089 #ifdef PERL_EXT_RE_BUILD
2095 regnode *fix = convert;
2096 U32 word = trie->wordcount;
2098 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2099 while( ++fix < n ) {
2100 Set_Node_Offset_Length(fix, 0, 0);
2103 SV ** const tmp = av_fetch( trie_words, word, 0 );
2105 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2106 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2108 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2116 NEXT_OFF(convert) = (U16)(tail - convert);
2117 DEBUG_r(optimize= n);
2123 if ( trie->maxlen ) {
2124 NEXT_OFF( convert ) = (U16)(tail - convert);
2125 ARG_SET( convert, data_slot );
2126 /* Store the offset to the first unabsorbed branch in
2127 jump[0], which is otherwise unused by the jump logic.
2128 We use this when dumping a trie and during optimisation. */
2130 trie->jump[0] = (U16)(nextbranch - convert);
2133 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2134 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2136 OP( convert ) = TRIEC;
2137 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2138 PerlMemShared_free(trie->bitmap);
2141 OP( convert ) = TRIE;
2143 /* store the type in the flags */
2144 convert->flags = nodetype;
2148 + regarglen[ OP( convert ) ];
2150 /* XXX We really should free up the resource in trie now,
2151 as we won't use them - (which resources?) dmq */
2153 /* needed for dumping*/
2154 DEBUG_r(if (optimize) {
2155 regnode *opt = convert;
2157 while ( ++opt < optimize) {
2158 Set_Node_Offset_Length(opt,0,0);
2161 Try to clean up some of the debris left after the
2164 while( optimize < jumper ) {
2165 mjd_nodelen += Node_Length((optimize));
2166 OP( optimize ) = OPTIMIZED;
2167 Set_Node_Offset_Length(optimize,0,0);
2170 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2172 } /* end node insert */
2174 /* Finish populating the prev field of the wordinfo array. Walk back
2175 * from each accept state until we find another accept state, and if
2176 * so, point the first word's .prev field at the second word. If the
2177 * second already has a .prev field set, stop now. This will be the
2178 * case either if we've already processed that word's accept state,
2179 * or that that state had multiple words, and the overspill words
2180 * were already linked up earlier.
2187 for (word=1; word <= trie->wordcount; word++) {
2189 if (trie->wordinfo[word].prev)
2191 state = trie->wordinfo[word].accept;
2193 state = prev_states[state];
2196 prev = trie->states[state].wordnum;
2200 trie->wordinfo[word].prev = prev;
2202 Safefree(prev_states);
2206 /* and now dump out the compressed format */
2207 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2209 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2211 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2212 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2214 SvREFCNT_dec(revcharmap);
2218 : trie->startstate>1
2224 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2226 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2228 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2229 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2232 We find the fail state for each state in the trie, this state is the longest proper
2233 suffix of the current states 'word' that is also a proper prefix of another word in our
2234 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2235 the DFA not to have to restart after its tried and failed a word at a given point, it
2236 simply continues as though it had been matching the other word in the first place.
2238 'abcdgu'=~/abcdefg|cdgu/
2239 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2240 fail, which would bring use to the state representing 'd' in the second word where we would
2241 try 'g' and succeed, prodceding to match 'cdgu'.
2243 /* add a fail transition */
2244 const U32 trie_offset = ARG(source);
2245 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2247 const U32 ucharcount = trie->uniquecharcount;
2248 const U32 numstates = trie->statecount;
2249 const U32 ubound = trie->lasttrans + ucharcount;
2253 U32 base = trie->states[ 1 ].trans.base;
2256 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2257 GET_RE_DEBUG_FLAGS_DECL;
2259 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2261 PERL_UNUSED_ARG(depth);
2265 ARG_SET( stclass, data_slot );
2266 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2267 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2268 aho->trie=trie_offset;
2269 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2270 Copy( trie->states, aho->states, numstates, reg_trie_state );
2271 Newxz( q, numstates, U32);
2272 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2275 /* initialize fail[0..1] to be 1 so that we always have
2276 a valid final fail state */
2277 fail[ 0 ] = fail[ 1 ] = 1;
2279 for ( charid = 0; charid < ucharcount ; charid++ ) {
2280 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2282 q[ q_write ] = newstate;
2283 /* set to point at the root */
2284 fail[ q[ q_write++ ] ]=1;
2287 while ( q_read < q_write) {
2288 const U32 cur = q[ q_read++ % numstates ];
2289 base = trie->states[ cur ].trans.base;
2291 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2292 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2294 U32 fail_state = cur;
2297 fail_state = fail[ fail_state ];
2298 fail_base = aho->states[ fail_state ].trans.base;
2299 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2301 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2302 fail[ ch_state ] = fail_state;
2303 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2305 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2307 q[ q_write++ % numstates] = ch_state;
2311 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2312 when we fail in state 1, this allows us to use the
2313 charclass scan to find a valid start char. This is based on the principle
2314 that theres a good chance the string being searched contains lots of stuff
2315 that cant be a start char.
2317 fail[ 0 ] = fail[ 1 ] = 0;
2318 DEBUG_TRIE_COMPILE_r({
2319 PerlIO_printf(Perl_debug_log,
2320 "%*sStclass Failtable (%"UVuf" states): 0",
2321 (int)(depth * 2), "", (UV)numstates
2323 for( q_read=1; q_read<numstates; q_read++ ) {
2324 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2326 PerlIO_printf(Perl_debug_log, "\n");
2329 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2334 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2335 * These need to be revisited when a newer toolchain becomes available.
2337 #if defined(__sparc64__) && defined(__GNUC__)
2338 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2339 # undef SPARC64_GCC_WORKAROUND
2340 # define SPARC64_GCC_WORKAROUND 1
2344 #define DEBUG_PEEP(str,scan,depth) \
2345 DEBUG_OPTIMISE_r({if (scan){ \
2346 SV * const mysv=sv_newmortal(); \
2347 regnode *Next = regnext(scan); \
2348 regprop(RExC_rx, mysv, scan); \
2349 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2350 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2351 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2358 #define JOIN_EXACT(scan,min,flags) \
2359 if (PL_regkind[OP(scan)] == EXACT) \
2360 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2363 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2364 /* Merge several consecutive EXACTish nodes into one. */
2365 regnode *n = regnext(scan);
2367 regnode *next = scan + NODE_SZ_STR(scan);
2371 regnode *stop = scan;
2372 GET_RE_DEBUG_FLAGS_DECL;
2374 PERL_UNUSED_ARG(depth);
2377 PERL_ARGS_ASSERT_JOIN_EXACT;
2378 #ifndef EXPERIMENTAL_INPLACESCAN
2379 PERL_UNUSED_ARG(flags);
2380 PERL_UNUSED_ARG(val);
2382 DEBUG_PEEP("join",scan,depth);
2384 /* Skip NOTHING, merge EXACT*. */
2386 ( PL_regkind[OP(n)] == NOTHING ||
2387 (stringok && (OP(n) == OP(scan))))
2389 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2391 if (OP(n) == TAIL || n > next)
2393 if (PL_regkind[OP(n)] == NOTHING) {
2394 DEBUG_PEEP("skip:",n,depth);
2395 NEXT_OFF(scan) += NEXT_OFF(n);
2396 next = n + NODE_STEP_REGNODE;
2403 else if (stringok) {
2404 const unsigned int oldl = STR_LEN(scan);
2405 regnode * const nnext = regnext(n);
2407 DEBUG_PEEP("merg",n,depth);
2410 if (oldl + STR_LEN(n) > U8_MAX)
2412 NEXT_OFF(scan) += NEXT_OFF(n);
2413 STR_LEN(scan) += STR_LEN(n);
2414 next = n + NODE_SZ_STR(n);
2415 /* Now we can overwrite *n : */
2416 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2424 #ifdef EXPERIMENTAL_INPLACESCAN
2425 if (flags && !NEXT_OFF(n)) {
2426 DEBUG_PEEP("atch", val, depth);
2427 if (reg_off_by_arg[OP(n)]) {
2428 ARG_SET(n, val - n);
2431 NEXT_OFF(n) = val - n;
2438 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2440 Two problematic code points in Unicode casefolding of EXACT nodes:
2442 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2443 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2449 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2450 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2452 This means that in case-insensitive matching (or "loose matching",
2453 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2454 length of the above casefolded versions) can match a target string
2455 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2456 This would rather mess up the minimum length computation.
2458 What we'll do is to look for the tail four bytes, and then peek
2459 at the preceding two bytes to see whether we need to decrease
2460 the minimum length by four (six minus two).
2462 Thanks to the design of UTF-8, there cannot be false matches:
2463 A sequence of valid UTF-8 bytes cannot be a subsequence of
2464 another valid sequence of UTF-8 bytes.
2467 char * const s0 = STRING(scan), *s, *t;
2468 char * const s1 = s0 + STR_LEN(scan) - 1;
2469 char * const s2 = s1 - 4;
2470 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2471 const char t0[] = "\xaf\x49\xaf\x42";
2473 const char t0[] = "\xcc\x88\xcc\x81";
2475 const char * const t1 = t0 + 3;
2478 s < s2 && (t = ninstr(s, s1, t0, t1));
2481 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2482 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2484 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2485 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2493 n = scan + NODE_SZ_STR(scan);
2495 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2502 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2506 /* REx optimizer. Converts nodes into quickier variants "in place".
2507 Finds fixed substrings. */
2509 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2510 to the position after last scanned or to NULL. */
2512 #define INIT_AND_WITHP \
2513 assert(!and_withp); \
2514 Newx(and_withp,1,struct regnode_charclass_class); \
2515 SAVEFREEPV(and_withp)
2517 /* this is a chain of data about sub patterns we are processing that
2518 need to be handled seperately/specially in study_chunk. Its so
2519 we can simulate recursion without losing state. */
2521 typedef struct scan_frame {
2522 regnode *last; /* last node to process in this frame */
2523 regnode *next; /* next node to process when last is reached */
2524 struct scan_frame *prev; /*previous frame*/
2525 I32 stop; /* what stopparen do we use */
2529 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2531 #define CASE_SYNST_FNC(nAmE) \
2533 if (flags & SCF_DO_STCLASS_AND) { \
2534 for (value = 0; value < 256; value++) \
2535 if (!is_ ## nAmE ## _cp(value)) \
2536 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2539 for (value = 0; value < 256; value++) \
2540 if (is_ ## nAmE ## _cp(value)) \
2541 ANYOF_BITMAP_SET(data->start_class, value); \
2545 if (flags & SCF_DO_STCLASS_AND) { \
2546 for (value = 0; value < 256; value++) \
2547 if (is_ ## nAmE ## _cp(value)) \
2548 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2551 for (value = 0; value < 256; value++) \
2552 if (!is_ ## nAmE ## _cp(value)) \
2553 ANYOF_BITMAP_SET(data->start_class, value); \
2560 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2561 I32 *minlenp, I32 *deltap,
2566 struct regnode_charclass_class *and_withp,
2567 U32 flags, U32 depth)
2568 /* scanp: Start here (read-write). */
2569 /* deltap: Write maxlen-minlen here. */
2570 /* last: Stop before this one. */
2571 /* data: string data about the pattern */
2572 /* stopparen: treat close N as END */
2573 /* recursed: which subroutines have we recursed into */
2574 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2577 I32 min = 0, pars = 0, code;
2578 regnode *scan = *scanp, *next;
2580 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2581 int is_inf_internal = 0; /* The studied chunk is infinite */
2582 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2583 scan_data_t data_fake;
2584 SV *re_trie_maxbuff = NULL;
2585 regnode *first_non_open = scan;
2586 I32 stopmin = I32_MAX;
2587 scan_frame *frame = NULL;
2588 GET_RE_DEBUG_FLAGS_DECL;
2590 PERL_ARGS_ASSERT_STUDY_CHUNK;
2593 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2597 while (first_non_open && OP(first_non_open) == OPEN)
2598 first_non_open=regnext(first_non_open);
2603 while ( scan && OP(scan) != END && scan < last ){
2604 /* Peephole optimizer: */
2605 DEBUG_STUDYDATA("Peep:", data,depth);
2606 DEBUG_PEEP("Peep",scan,depth);
2607 JOIN_EXACT(scan,&min,0);
2609 /* Follow the next-chain of the current node and optimize
2610 away all the NOTHINGs from it. */
2611 if (OP(scan) != CURLYX) {
2612 const int max = (reg_off_by_arg[OP(scan)]
2614 /* I32 may be smaller than U16 on CRAYs! */
2615 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2616 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2620 /* Skip NOTHING and LONGJMP. */
2621 while ((n = regnext(n))
2622 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2623 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2624 && off + noff < max)
2626 if (reg_off_by_arg[OP(scan)])
2629 NEXT_OFF(scan) = off;
2634 /* The principal pseudo-switch. Cannot be a switch, since we
2635 look into several different things. */
2636 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2637 || OP(scan) == IFTHEN) {
2638 next = regnext(scan);
2640 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2642 if (OP(next) == code || code == IFTHEN) {
2643 /* NOTE - There is similar code to this block below for handling
2644 TRIE nodes on a re-study. If you change stuff here check there
2646 I32 max1 = 0, min1 = I32_MAX, num = 0;
2647 struct regnode_charclass_class accum;
2648 regnode * const startbranch=scan;
2650 if (flags & SCF_DO_SUBSTR)
2651 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2652 if (flags & SCF_DO_STCLASS)
2653 cl_init_zero(pRExC_state, &accum);
2655 while (OP(scan) == code) {
2656 I32 deltanext, minnext, f = 0, fake;
2657 struct regnode_charclass_class this_class;
2660 data_fake.flags = 0;
2662 data_fake.whilem_c = data->whilem_c;
2663 data_fake.last_closep = data->last_closep;
2666 data_fake.last_closep = &fake;
2668 data_fake.pos_delta = delta;
2669 next = regnext(scan);
2670 scan = NEXTOPER(scan);
2672 scan = NEXTOPER(scan);
2673 if (flags & SCF_DO_STCLASS) {
2674 cl_init(pRExC_state, &this_class);
2675 data_fake.start_class = &this_class;
2676 f = SCF_DO_STCLASS_AND;
2678 if (flags & SCF_WHILEM_VISITED_POS)
2679 f |= SCF_WHILEM_VISITED_POS;
2681 /* we suppose the run is continuous, last=next...*/
2682 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2684 stopparen, recursed, NULL, f,depth+1);
2687 if (max1 < minnext + deltanext)
2688 max1 = minnext + deltanext;
2689 if (deltanext == I32_MAX)
2690 is_inf = is_inf_internal = 1;
2692 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2694 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2695 if ( stopmin > minnext)
2696 stopmin = min + min1;
2697 flags &= ~SCF_DO_SUBSTR;
2699 data->flags |= SCF_SEEN_ACCEPT;
2702 if (data_fake.flags & SF_HAS_EVAL)
2703 data->flags |= SF_HAS_EVAL;
2704 data->whilem_c = data_fake.whilem_c;
2706 if (flags & SCF_DO_STCLASS)
2707 cl_or(pRExC_state, &accum, &this_class);
2709 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2711 if (flags & SCF_DO_SUBSTR) {
2712 data->pos_min += min1;
2713 data->pos_delta += max1 - min1;
2714 if (max1 != min1 || is_inf)
2715 data->longest = &(data->longest_float);
2718 delta += max1 - min1;
2719 if (flags & SCF_DO_STCLASS_OR) {
2720 cl_or(pRExC_state, data->start_class, &accum);
2722 cl_and(data->start_class, and_withp);
2723 flags &= ~SCF_DO_STCLASS;
2726 else if (flags & SCF_DO_STCLASS_AND) {
2728 cl_and(data->start_class, &accum);
2729 flags &= ~SCF_DO_STCLASS;
2732 /* Switch to OR mode: cache the old value of
2733 * data->start_class */
2735 StructCopy(data->start_class, and_withp,
2736 struct regnode_charclass_class);
2737 flags &= ~SCF_DO_STCLASS_AND;
2738 StructCopy(&accum, data->start_class,
2739 struct regnode_charclass_class);
2740 flags |= SCF_DO_STCLASS_OR;
2741 data->start_class->flags |= ANYOF_EOS;
2745 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2748 Assuming this was/is a branch we are dealing with: 'scan' now
2749 points at the item that follows the branch sequence, whatever
2750 it is. We now start at the beginning of the sequence and look
2757 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2759 If we can find such a subseqence we need to turn the first
2760 element into a trie and then add the subsequent branch exact
2761 strings to the trie.
2765 1. patterns where the whole set of branch can be converted.
2767 2. patterns where only a subset can be converted.
2769 In case 1 we can replace the whole set with a single regop
2770 for the trie. In case 2 we need to keep the start and end
2773 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2774 becomes BRANCH TRIE; BRANCH X;
2776 There is an additional case, that being where there is a
2777 common prefix, which gets split out into an EXACT like node
2778 preceding the TRIE node.
2780 If x(1..n)==tail then we can do a simple trie, if not we make
2781 a "jump" trie, such that when we match the appropriate word
2782 we "jump" to the appopriate tail node. Essentailly we turn
2783 a nested if into a case structure of sorts.
2788 if (!re_trie_maxbuff) {
2789 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2790 if (!SvIOK(re_trie_maxbuff))
2791 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2793 if ( SvIV(re_trie_maxbuff)>=0 ) {
2795 regnode *first = (regnode *)NULL;
2796 regnode *last = (regnode *)NULL;
2797 regnode *tail = scan;
2802 SV * const mysv = sv_newmortal(); /* for dumping */
2804 /* var tail is used because there may be a TAIL
2805 regop in the way. Ie, the exacts will point to the
2806 thing following the TAIL, but the last branch will
2807 point at the TAIL. So we advance tail. If we
2808 have nested (?:) we may have to move through several
2812 while ( OP( tail ) == TAIL ) {
2813 /* this is the TAIL generated by (?:) */
2814 tail = regnext( tail );
2819 regprop(RExC_rx, mysv, tail );
2820 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2821 (int)depth * 2 + 2, "",
2822 "Looking for TRIE'able sequences. Tail node is: ",
2823 SvPV_nolen_const( mysv )
2829 step through the branches, cur represents each
2830 branch, noper is the first thing to be matched
2831 as part of that branch and noper_next is the
2832 regnext() of that node. if noper is an EXACT
2833 and noper_next is the same as scan (our current
2834 position in the regex) then the EXACT branch is
2835 a possible optimization target. Once we have
2836 two or more consequetive such branches we can
2837 create a trie of the EXACT's contents and stich
2838 it in place. If the sequence represents all of
2839 the branches we eliminate the whole thing and
2840 replace it with a single TRIE. If it is a
2841 subsequence then we need to stitch it in. This
2842 means the first branch has to remain, and needs
2843 to be repointed at the item on the branch chain
2844 following the last branch optimized. This could
2845 be either a BRANCH, in which case the
2846 subsequence is internal, or it could be the
2847 item following the branch sequence in which
2848 case the subsequence is at the end.
2852 /* dont use tail as the end marker for this traverse */
2853 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2854 regnode * const noper = NEXTOPER( cur );
2855 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2856 regnode * const noper_next = regnext( noper );
2860 regprop(RExC_rx, mysv, cur);
2861 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2862 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2864 regprop(RExC_rx, mysv, noper);
2865 PerlIO_printf( Perl_debug_log, " -> %s",
2866 SvPV_nolen_const(mysv));
2869 regprop(RExC_rx, mysv, noper_next );
2870 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2871 SvPV_nolen_const(mysv));
2873 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2874 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2876 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2877 : PL_regkind[ OP( noper ) ] == EXACT )
2878 || OP(noper) == NOTHING )
2880 && noper_next == tail
2885 if ( !first || optype == NOTHING ) {
2886 if (!first) first = cur;
2887 optype = OP( noper );
2893 Currently we do not believe that the trie logic can
2894 handle case insensitive matching properly when the
2895 pattern is not unicode (thus forcing unicode semantics).
2897 If/when this is fixed the following define can be swapped
2898 in below to fully enable trie logic.
2900 #define TRIE_TYPE_IS_SAFE 1
2903 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2905 if ( last && TRIE_TYPE_IS_SAFE ) {
2906 make_trie( pRExC_state,
2907 startbranch, first, cur, tail, count,
2910 if ( PL_regkind[ OP( noper ) ] == EXACT
2912 && noper_next == tail
2917 optype = OP( noper );
2927 regprop(RExC_rx, mysv, cur);
2928 PerlIO_printf( Perl_debug_log,
2929 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2930 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2934 if ( last && TRIE_TYPE_IS_SAFE ) {
2935 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2936 #ifdef TRIE_STUDY_OPT
2937 if ( ((made == MADE_EXACT_TRIE &&
2938 startbranch == first)
2939 || ( first_non_open == first )) &&
2941 flags |= SCF_TRIE_RESTUDY;
2942 if ( startbranch == first
2945 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2955 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2956 scan = NEXTOPER(NEXTOPER(scan));
2957 } else /* single branch is optimized. */
2958 scan = NEXTOPER(scan);
2960 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2961 scan_frame *newframe = NULL;
2966 if (OP(scan) != SUSPEND) {
2967 /* set the pointer */
2968 if (OP(scan) == GOSUB) {
2970 RExC_recurse[ARG2L(scan)] = scan;
2971 start = RExC_open_parens[paren-1];
2972 end = RExC_close_parens[paren-1];
2975 start = RExC_rxi->program + 1;
2979 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2980 SAVEFREEPV(recursed);
2982 if (!PAREN_TEST(recursed,paren+1)) {
2983 PAREN_SET(recursed,paren+1);
2984 Newx(newframe,1,scan_frame);
2986 if (flags & SCF_DO_SUBSTR) {
2987 SCAN_COMMIT(pRExC_state,data,minlenp);
2988 data->longest = &(data->longest_float);
2990 is_inf = is_inf_internal = 1;
2991 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2992 cl_anything(pRExC_state, data->start_class);
2993 flags &= ~SCF_DO_STCLASS;
2996 Newx(newframe,1,scan_frame);
2999 end = regnext(scan);
3004 SAVEFREEPV(newframe);
3005 newframe->next = regnext(scan);
3006 newframe->last = last;
3007 newframe->stop = stopparen;
3008 newframe->prev = frame;
3018 else if (OP(scan) == EXACT) {
3019 I32 l = STR_LEN(scan);
3022 const U8 * const s = (U8*)STRING(scan);
3023 l = utf8_length(s, s + l);
3024 uc = utf8_to_uvchr(s, NULL);
3026 uc = *((U8*)STRING(scan));
3029 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3030 /* The code below prefers earlier match for fixed
3031 offset, later match for variable offset. */
3032 if (data->last_end == -1) { /* Update the start info. */
3033 data->last_start_min = data->pos_min;
3034 data->last_start_max = is_inf
3035 ? I32_MAX : data->pos_min + data->pos_delta;
3037 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3039 SvUTF8_on(data->last_found);
3041 SV * const sv = data->last_found;
3042 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3043 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3044 if (mg && mg->mg_len >= 0)
3045 mg->mg_len += utf8_length((U8*)STRING(scan),
3046 (U8*)STRING(scan)+STR_LEN(scan));
3048 data->last_end = data->pos_min + l;
3049 data->pos_min += l; /* As in the first entry. */
3050 data->flags &= ~SF_BEFORE_EOL;
3052 if (flags & SCF_DO_STCLASS_AND) {
3053 /* Check whether it is compatible with what we know already! */
3057 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3058 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3059 && (!(data->start_class->flags & ANYOF_FOLD)
3060 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3063 ANYOF_CLASS_ZERO(data->start_class);
3064 ANYOF_BITMAP_ZERO(data->start_class);
3066 ANYOF_BITMAP_SET(data->start_class, uc);
3067 data->start_class->flags &= ~ANYOF_EOS;
3069 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3071 else if (flags & SCF_DO_STCLASS_OR) {
3072 /* false positive possible if the class is case-folded */
3074 ANYOF_BITMAP_SET(data->start_class, uc);
3076 data->start_class->flags |= ANYOF_UNICODE_ALL;
3077 data->start_class->flags &= ~ANYOF_EOS;
3078 cl_and(data->start_class, and_withp);
3080 flags &= ~SCF_DO_STCLASS;
3082 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3083 I32 l = STR_LEN(scan);
3084 UV uc = *((U8*)STRING(scan));
3086 /* Search for fixed substrings supports EXACT only. */
3087 if (flags & SCF_DO_SUBSTR) {
3089 SCAN_COMMIT(pRExC_state, data, minlenp);
3092 const U8 * const s = (U8 *)STRING(scan);
3093 l = utf8_length(s, s + l);
3094 uc = utf8_to_uvchr(s, NULL);
3097 if (flags & SCF_DO_SUBSTR)
3099 if (flags & SCF_DO_STCLASS_AND) {
3100 /* Check whether it is compatible with what we know already! */
3104 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3105 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3106 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3108 ANYOF_CLASS_ZERO(data->start_class);
3109 ANYOF_BITMAP_ZERO(data->start_class);
3111 ANYOF_BITMAP_SET(data->start_class, uc);
3112 data->start_class->flags &= ~ANYOF_EOS;
3113 data->start_class->flags |= ANYOF_FOLD;
3114 if (OP(scan) == EXACTFL)
3115 data->start_class->flags |= ANYOF_LOCALE;
3118 else if (flags & SCF_DO_STCLASS_OR) {
3119 if (data->start_class->flags & ANYOF_FOLD) {
3120 /* false positive possible if the class is case-folded.
3121 Assume that the locale settings are the same... */
3123 ANYOF_BITMAP_SET(data->start_class, uc);
3124 data->start_class->flags &= ~ANYOF_EOS;
3126 cl_and(data->start_class, and_withp);
3128 flags &= ~SCF_DO_STCLASS;
3130 else if (REGNODE_VARIES(OP(scan))) {
3131 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3132 I32 f = flags, pos_before = 0;
3133 regnode * const oscan = scan;
3134 struct regnode_charclass_class this_class;
3135 struct regnode_charclass_class *oclass = NULL;
3136 I32 next_is_eval = 0;
3138 switch (PL_regkind[OP(scan)]) {
3139 case WHILEM: /* End of (?:...)* . */
3140 scan = NEXTOPER(scan);
3143 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3144 next = NEXTOPER(scan);
3145 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3147 maxcount = REG_INFTY;
3148 next = regnext(scan);
3149 scan = NEXTOPER(scan);
3153 if (flags & SCF_DO_SUBSTR)
3158 if (flags & SCF_DO_STCLASS) {
3160 maxcount = REG_INFTY;
3161 next = regnext(scan);
3162 scan = NEXTOPER(scan);
3165 is_inf = is_inf_internal = 1;
3166 scan = regnext(scan);
3167 if (flags & SCF_DO_SUBSTR) {
3168 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3169 data->longest = &(data->longest_float);
3171 goto optimize_curly_tail;
3173 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3174 && (scan->flags == stopparen))
3179 mincount = ARG1(scan);
3180 maxcount = ARG2(scan);
3182 next = regnext(scan);
3183 if (OP(scan) == CURLYX) {
3184 I32 lp = (data ? *(data->last_closep) : 0);
3185 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3187 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3188 next_is_eval = (OP(scan) == EVAL);
3190 if (flags & SCF_DO_SUBSTR) {
3191 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3192 pos_before = data->pos_min;
3196 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3198 data->flags |= SF_IS_INF;
3200 if (flags & SCF_DO_STCLASS) {
3201 cl_init(pRExC_state, &this_class);
3202 oclass = data->start_class;
3203 data->start_class = &this_class;
3204 f |= SCF_DO_STCLASS_AND;
3205 f &= ~SCF_DO_STCLASS_OR;
3207 /* These are the cases when once a subexpression
3208 fails at a particular position, it cannot succeed
3209 even after backtracking at the enclosing scope.
3211 XXXX what if minimal match and we are at the
3212 initial run of {n,m}? */
3213 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3214 f &= ~SCF_WHILEM_VISITED_POS;
3216 /* This will finish on WHILEM, setting scan, or on NULL: */
3217 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3218 last, data, stopparen, recursed, NULL,
3220 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3222 if (flags & SCF_DO_STCLASS)
3223 data->start_class = oclass;
3224 if (mincount == 0 || minnext == 0) {
3225 if (flags & SCF_DO_STCLASS_OR) {
3226 cl_or(pRExC_state, data->start_class, &this_class);
3228 else if (flags & SCF_DO_STCLASS_AND) {
3229 /* Switch to OR mode: cache the old value of
3230 * data->start_class */
3232 StructCopy(data->start_class, and_withp,
3233 struct regnode_charclass_class);
3234 flags &= ~SCF_DO_STCLASS_AND;
3235 StructCopy(&this_class, data->start_class,
3236 struct regnode_charclass_class);
3237 flags |= SCF_DO_STCLASS_OR;
3238 data->start_class->flags |= ANYOF_EOS;
3240 } else { /* Non-zero len */
3241 if (flags & SCF_DO_STCLASS_OR) {
3242 cl_or(pRExC_state, data->start_class, &this_class);
3243 cl_and(data->start_class, and_withp);
3245 else if (flags & SCF_DO_STCLASS_AND)
3246 cl_and(data->start_class, &this_class);
3247 flags &= ~SCF_DO_STCLASS;
3249 if (!scan) /* It was not CURLYX, but CURLY. */
3251 if ( /* ? quantifier ok, except for (?{ ... }) */
3252 (next_is_eval || !(mincount == 0 && maxcount == 1))
3253 && (minnext == 0) && (deltanext == 0)
3254 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3255 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3257 ckWARNreg(RExC_parse,
3258 "Quantifier unexpected on zero-length expression");
3261 min += minnext * mincount;
3262 is_inf_internal |= ((maxcount == REG_INFTY
3263 && (minnext + deltanext) > 0)
3264 || deltanext == I32_MAX);
3265 is_inf |= is_inf_internal;
3266 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3268 /* Try powerful optimization CURLYX => CURLYN. */
3269 if ( OP(oscan) == CURLYX && data
3270 && data->flags & SF_IN_PAR
3271 && !(data->flags & SF_HAS_EVAL)
3272 && !deltanext && minnext == 1 ) {
3273 /* Try to optimize to CURLYN. */
3274 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3275 regnode * const nxt1 = nxt;
3282 if (!REGNODE_SIMPLE(OP(nxt))
3283 && !(PL_regkind[OP(nxt)] == EXACT
3284 && STR_LEN(nxt) == 1))
3290 if (OP(nxt) != CLOSE)
3292 if (RExC_open_parens) {
3293 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3294 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3296 /* Now we know that nxt2 is the only contents: */
3297 oscan->flags = (U8)ARG(nxt);
3299 OP(nxt1) = NOTHING; /* was OPEN. */
3302 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3303 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3304 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
3305 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3306 OP(nxt + 1) = OPTIMIZED; /* was count. */
3307 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
3312 /* Try optimization CURLYX => CURLYM. */
3313 if ( OP(oscan) == CURLYX && data
3314 && !(data->flags & SF_HAS_PAR)
3315 && !(data->flags & SF_HAS_EVAL)
3316 && !deltanext /* atom is fixed width */
3317 && minnext != 0 /* CURLYM can't handle zero width */
3319 /* XXXX How to optimize if data == 0? */
3320 /* Optimize to a simpler form. */
3321 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3325 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3326 && (OP(nxt2) != WHILEM))
3328 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3329 /* Need to optimize away parenths. */
3330 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
3331 /* Set the parenth number. */
3332 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3334 oscan->flags = (U8)ARG(nxt);
3335 if (RExC_open_parens) {
3336 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3337 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3339 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3340 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3343 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3344 OP(nxt + 1) = OPTIMIZED; /* was count. */
3345 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3346 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3349 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3350 regnode *nnxt = regnext(nxt1);
3353 if (reg_off_by_arg[OP(nxt1)])
3354 ARG_SET(nxt1, nxt2 - nxt1);
3355 else if (nxt2 - nxt1 < U16_MAX)
3356 NEXT_OFF(nxt1) = nxt2 - nxt1;
3358 OP(nxt) = NOTHING; /* Cannot beautify */
3363 /* Optimize again: */
3364 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3365 NULL, stopparen, recursed, NULL, 0,depth+1);
3370 else if ((OP(oscan) == CURLYX)
3371 && (flags & SCF_WHILEM_VISITED_POS)
3372 /* See the comment on a similar expression above.
3373 However, this time it not a subexpression
3374 we care about, but the expression itself. */
3375 && (maxcount == REG_INFTY)
3376 && data && ++data->whilem_c < 16) {
3377 /* This stays as CURLYX, we can put the count/of pair. */
3378 /* Find WHILEM (as in regexec.c) */
3379 regnode *nxt = oscan + NEXT_OFF(oscan);
3381 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3383 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3384 | (RExC_whilem_seen << 4)); /* On WHILEM */
3386 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3388 if (flags & SCF_DO_SUBSTR) {
3389 SV *last_str = NULL;
3390 int counted = mincount != 0;
3392 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3393 #if defined(SPARC64_GCC_WORKAROUND)
3396 const char *s = NULL;
3399 if (pos_before >= data->last_start_min)
3402 b = data->last_start_min;
3405 s = SvPV_const(data->last_found, l);
3406 old = b - data->last_start_min;
3409 I32 b = pos_before >= data->last_start_min
3410 ? pos_before : data->last_start_min;
3412 const char * const s = SvPV_const(data->last_found, l);
3413 I32 old = b - data->last_start_min;
3417 old = utf8_hop((U8*)s, old) - (U8*)s;
3420 /* Get the added string: */
3421 last_str = newSVpvn_utf8(s + old, l, UTF);
3422 if (deltanext == 0 && pos_before == b) {
3423 /* What was added is a constant string */
3425 SvGROW(last_str, (mincount * l) + 1);
3426 repeatcpy(SvPVX(last_str) + l,
3427 SvPVX_const(last_str), l, mincount - 1);
3428 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3429 /* Add additional parts. */
3430 SvCUR_set(data->last_found,
3431 SvCUR(data->last_found) - l);
3432 sv_catsv(data->last_found, last_str);
3434 SV * sv = data->last_found;
3436 SvUTF8(sv) && SvMAGICAL(sv) ?
3437 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3438 if (mg && mg->mg_len >= 0)
3439 mg->mg_len += CHR_SVLEN(last_str) - l;
3441 data->last_end += l * (mincount - 1);
3444 /* start offset must point into the last copy */
3445 data->last_start_min += minnext * (mincount - 1);
3446 data->last_start_max += is_inf ? I32_MAX
3447 : (maxcount - 1) * (minnext + data->pos_delta);
3450 /* It is counted once already... */
3451 data->pos_min += minnext * (mincount - counted);
3452 data->pos_delta += - counted * deltanext +
3453 (minnext + deltanext) * maxcount - minnext * mincount;
3454 if (mincount != maxcount) {
3455 /* Cannot extend fixed substrings found inside
3457 SCAN_COMMIT(pRExC_state,data,minlenp);
3458 if (mincount && last_str) {
3459 SV * const sv = data->last_found;
3460 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3461 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3465 sv_setsv(sv, last_str);
3466 data->last_end = data->pos_min;
3467 data->last_start_min =
3468 data->pos_min - CHR_SVLEN(last_str);
3469 data->last_start_max = is_inf
3471 : data->pos_min + data->pos_delta
3472 - CHR_SVLEN(last_str);
3474 data->longest = &(data->longest_float);
3476 SvREFCNT_dec(last_str);
3478 if (data && (fl & SF_HAS_EVAL))
3479 data->flags |= SF_HAS_EVAL;
3480 optimize_curly_tail:
3481 if (OP(oscan) != CURLYX) {
3482 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3484 NEXT_OFF(oscan) += NEXT_OFF(next);
3487 default: /* REF and CLUMP only? */
3488 if (flags & SCF_DO_SUBSTR) {
3489 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3490 data->longest = &(data->longest_float);
3492 is_inf = is_inf_internal = 1;
3493 if (flags & SCF_DO_STCLASS_OR)
3494 cl_anything(pRExC_state, data->start_class);
3495 flags &= ~SCF_DO_STCLASS;
3499 else if (OP(scan) == LNBREAK) {
3500 if (flags & SCF_DO_STCLASS) {
3502 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3503 if (flags & SCF_DO_STCLASS_AND) {
3504 for (value = 0; value < 256; value++)
3505 if (!is_VERTWS_cp(value))
3506 ANYOF_BITMAP_CLEAR(data->start_class, value);
3509 for (value = 0; value < 256; value++)
3510 if (is_VERTWS_cp(value))
3511 ANYOF_BITMAP_SET(data->start_class, value);
3513 if (flags & SCF_DO_STCLASS_OR)
3514 cl_and(data->start_class, and_withp);
3515 flags &= ~SCF_DO_STCLASS;
3519 if (flags & SCF_DO_SUBSTR) {
3520 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3522 data->pos_delta += 1;
3523 data->longest = &(data->longest_float);
3527 else if (OP(scan) == FOLDCHAR) {
3528 int d = ARG(scan)==0xDF ? 1 : 2;
3529 flags &= ~SCF_DO_STCLASS;
3532 if (flags & SCF_DO_SUBSTR) {
3533 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3535 data->pos_delta += d;
3536 data->longest = &(data->longest_float);
3539 else if (REGNODE_SIMPLE(OP(scan))) {
3542 if (flags & SCF_DO_SUBSTR) {
3543 SCAN_COMMIT(pRExC_state,data,minlenp);
3547 if (flags & SCF_DO_STCLASS) {
3548 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3550 /* Some of the logic below assumes that switching
3551 locale on will only add false positives. */
3552 switch (PL_regkind[OP(scan)]) {
3556 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3557 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3558 cl_anything(pRExC_state, data->start_class);
3561 if (OP(scan) == SANY)
3563 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3564 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3565 || (data->start_class->flags & ANYOF_CLASS));
3566 cl_anything(pRExC_state, data->start_class);
3568 if (flags & SCF_DO_STCLASS_AND || !value)
3569 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3572 if (flags & SCF_DO_STCLASS_AND)
3573 cl_and(data->start_class,
3574 (struct regnode_charclass_class*)scan);
3576 cl_or(pRExC_state, data->start_class,
3577 (struct regnode_charclass_class*)scan);
3580 if (flags & SCF_DO_STCLASS_AND) {
3581 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3582 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3583 for (value = 0; value < 256; value++)
3584 if (!isALNUM(value))
3585 ANYOF_BITMAP_CLEAR(data->start_class, value);
3589 if (data->start_class->flags & ANYOF_LOCALE)
3590 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3592 for (value = 0; value < 256; value++)
3594 ANYOF_BITMAP_SET(data->start_class, value);
3599 if (flags & SCF_DO_STCLASS_AND) {
3600 if (data->start_class->flags & ANYOF_LOCALE)
3601 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3604 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3605 data->start_class->flags |= ANYOF_LOCALE;
3609 if (flags & SCF_DO_STCLASS_AND) {
3610 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3611 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3612 for (value = 0; value < 256; value++)
3614 ANYOF_BITMAP_CLEAR(data->start_class, value);
3618 if (data->start_class->flags & ANYOF_LOCALE)
3619 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3621 for (value = 0; value < 256; value++)
3622 if (!isALNUM(value))
3623 ANYOF_BITMAP_SET(data->start_class, value);
3628 if (flags & SCF_DO_STCLASS_AND) {
3629 if (data->start_class->flags & ANYOF_LOCALE)
3630 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3633 data->start_class->flags |= ANYOF_LOCALE;
3634 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3638 if (flags & SCF_DO_STCLASS_AND) {
3639 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3640 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3641 for (value = 0; value < 256; value++)
3642 if (!isSPACE(value))
3643 ANYOF_BITMAP_CLEAR(data->start_class, value);
3647 if (data->start_class->flags & ANYOF_LOCALE)
3648 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3650 for (value = 0; value < 256; value++)
3652 ANYOF_BITMAP_SET(data->start_class, value);
3657 if (flags & SCF_DO_STCLASS_AND) {
3658 if (data->start_class->flags & ANYOF_LOCALE)
3659 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3662 data->start_class->flags |= ANYOF_LOCALE;
3663 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3667 if (flags & SCF_DO_STCLASS_AND) {
3668 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3669 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3670 for (value = 0; value < 256; value++)
3672 ANYOF_BITMAP_CLEAR(data->start_class, value);
3676 if (data->start_class->flags & ANYOF_LOCALE)
3677 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3679 for (value = 0; value < 256; value++)
3680 if (!isSPACE(value))
3681 ANYOF_BITMAP_SET(data->start_class, value);
3686 if (flags & SCF_DO_STCLASS_AND) {
3687 if (data->start_class->flags & ANYOF_LOCALE) {
3688 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3689 for (value = 0; value < 256; value++)
3690 if (!isSPACE(value))
3691 ANYOF_BITMAP_CLEAR(data->start_class, value);
3695 data->start_class->flags |= ANYOF_LOCALE;
3696 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3700 if (flags & SCF_DO_STCLASS_AND) {
3701 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3702 for (value = 0; value < 256; value++)
3703 if (!isDIGIT(value))
3704 ANYOF_BITMAP_CLEAR(data->start_class, value);
3707 if (data->start_class->flags & ANYOF_LOCALE)
3708 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3710 for (value = 0; value < 256; value++)
3712 ANYOF_BITMAP_SET(data->start_class, value);
3717 if (flags & SCF_DO_STCLASS_AND) {
3718 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3719 for (value = 0; value < 256; value++)
3721 ANYOF_BITMAP_CLEAR(data->start_class, value);
3724 if (data->start_class->flags & ANYOF_LOCALE)
3725 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3727 for (value = 0; value < 256; value++)
3728 if (!isDIGIT(value))
3729 ANYOF_BITMAP_SET(data->start_class, value);
3733 CASE_SYNST_FNC(VERTWS);
3734 CASE_SYNST_FNC(HORIZWS);
3737 if (flags & SCF_DO_STCLASS_OR)
3738 cl_and(data->start_class, and_withp);
3739 flags &= ~SCF_DO_STCLASS;
3742 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3743 data->flags |= (OP(scan) == MEOL
3747 else if ( PL_regkind[OP(scan)] == BRANCHJ
3748 /* Lookbehind, or need to calculate parens/evals/stclass: */
3749 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3750 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3751 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3752 || OP(scan) == UNLESSM )
3754 /* Negative Lookahead/lookbehind
3755 In this case we can't do fixed string optimisation.
3758 I32 deltanext, minnext, fake = 0;
3760 struct regnode_charclass_class intrnl;
3763 data_fake.flags = 0;
3765 data_fake.whilem_c = data->whilem_c;
3766 data_fake.last_closep = data->last_closep;
3769 data_fake.last_closep = &fake;
3770 data_fake.pos_delta = delta;
3771 if ( flags & SCF_DO_STCLASS && !scan->flags
3772 && OP(scan) == IFMATCH ) { /* Lookahead */
3773 cl_init(pRExC_state, &intrnl);
3774 data_fake.start_class = &intrnl;
3775 f |= SCF_DO_STCLASS_AND;
3777 if (flags & SCF_WHILEM_VISITED_POS)
3778 f |= SCF_WHILEM_VISITED_POS;
3779 next = regnext(scan);
3780 nscan = NEXTOPER(NEXTOPER(scan));
3781 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3782 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3785 FAIL("Variable length lookbehind not implemented");
3787 else if (minnext > (I32)U8_MAX) {
3788 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3790 scan->flags = (U8)minnext;
3793 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3795 if (data_fake.flags & SF_HAS_EVAL)
3796 data->flags |= SF_HAS_EVAL;
3797 data->whilem_c = data_fake.whilem_c;
3799 if (f & SCF_DO_STCLASS_AND) {
3800 if (flags & SCF_DO_STCLASS_OR) {
3801 /* OR before, AND after: ideally we would recurse with
3802 * data_fake to get the AND applied by study of the
3803 * remainder of the pattern, and then derecurse;
3804 * *** HACK *** for now just treat as "no information".
3805 * See [perl #56690].
3807 cl_init(pRExC_state, data->start_class);
3809 /* AND before and after: combine and continue */
3810 const int was = (data->start_class->flags & ANYOF_EOS);
3812 cl_and(data->start_class, &intrnl);
3814 data->start_class->flags |= ANYOF_EOS;
3818 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3820 /* Positive Lookahead/lookbehind
3821 In this case we can do fixed string optimisation,
3822 but we must be careful about it. Note in the case of
3823 lookbehind the positions will be offset by the minimum
3824 length of the pattern, something we won't know about
3825 until after the recurse.
3827 I32 deltanext, fake = 0;
3829 struct regnode_charclass_class intrnl;
3831 /* We use SAVEFREEPV so that when the full compile
3832 is finished perl will clean up the allocated
3833 minlens when its all done. This was we don't
3834 have to worry about freeing them when we know
3835 they wont be used, which would be a pain.
3838 Newx( minnextp, 1, I32 );
3839 SAVEFREEPV(minnextp);
3842 StructCopy(data, &data_fake, scan_data_t);
3843 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3846 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3847 data_fake.last_found=newSVsv(data->last_found);
3851 data_fake.last_closep = &fake;
3852 data_fake.flags = 0;
3853 data_fake.pos_delta = delta;
3855 data_fake.flags |= SF_IS_INF;
3856 if ( flags & SCF_DO_STCLASS && !scan->flags
3857 && OP(scan) == IFMATCH ) { /* Lookahead */
3858 cl_init(pRExC_state, &intrnl);
3859 data_fake.start_class = &intrnl;
3860 f |= SCF_DO_STCLASS_AND;
3862 if (flags & SCF_WHILEM_VISITED_POS)
3863 f |= SCF_WHILEM_VISITED_POS;
3864 next = regnext(scan);
3865 nscan = NEXTOPER(NEXTOPER(scan));
3867 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3868 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3871 FAIL("Variable length lookbehind not implemented");
3873 else if (*minnextp > (I32)U8_MAX) {
3874 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3876 scan->flags = (U8)*minnextp;
3881 if (f & SCF_DO_STCLASS_AND) {
3882 const int was = (data->start_class->flags & ANYOF_EOS);
3884 cl_and(data->start_class, &intrnl);
3886 data->start_class->flags |= ANYOF_EOS;
3889 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3891 if (data_fake.flags & SF_HAS_EVAL)
3892 data->flags |= SF_HAS_EVAL;
3893 data->whilem_c = data_fake.whilem_c;
3894 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3895 if (RExC_rx->minlen<*minnextp)
3896 RExC_rx->minlen=*minnextp;
3897 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3898 SvREFCNT_dec(data_fake.last_found);
3900 if ( data_fake.minlen_fixed != minlenp )
3902 data->offset_fixed= data_fake.offset_fixed;
3903 data->minlen_fixed= data_fake.minlen_fixed;
3904 data->lookbehind_fixed+= scan->flags;
3906 if ( data_fake.minlen_float != minlenp )
3908 data->minlen_float= data_fake.minlen_float;
3909 data->offset_float_min=data_fake.offset_float_min;
3910 data->offset_float_max=data_fake.offset_float_max;
3911 data->lookbehind_float+= scan->flags;
3920 else if (OP(scan) == OPEN) {
3921 if (stopparen != (I32)ARG(scan))
3924 else if (OP(scan) == CLOSE) {
3925 if (stopparen == (I32)ARG(scan)) {
3928 if ((I32)ARG(scan) == is_par) {
3929 next = regnext(scan);
3931 if ( next && (OP(next) != WHILEM) && next < last)
3932 is_par = 0; /* Disable optimization */
3935 *(data->last_closep) = ARG(scan);
3937 else if (OP(scan) == EVAL) {
3939 data->flags |= SF_HAS_EVAL;
3941 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3942 if (flags & SCF_DO_SUBSTR) {
3943 SCAN_COMMIT(pRExC_state,data,minlenp);
3944 flags &= ~SCF_DO_SUBSTR;
3946 if (data && OP(scan)==ACCEPT) {
3947 data->flags |= SCF_SEEN_ACCEPT;
3952 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3954 if (flags & SCF_DO_SUBSTR) {
3955 SCAN_COMMIT(pRExC_state,data,minlenp);
3956 data->longest = &(data->longest_float);
3958 is_inf = is_inf_internal = 1;
3959 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3960 cl_anything(pRExC_state, data->start_class);
3961 flags &= ~SCF_DO_STCLASS;
3963 else if (OP(scan) == GPOS) {
3964 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3965 !(delta || is_inf || (data && data->pos_delta)))
3967 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3968 RExC_rx->extflags |= RXf_ANCH_GPOS;
3969 if (RExC_rx->gofs < (U32)min)
3970 RExC_rx->gofs = min;
3972 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3976 #ifdef TRIE_STUDY_OPT
3977 #ifdef FULL_TRIE_STUDY
3978 else if (PL_regkind[OP(scan)] == TRIE) {
3979 /* NOTE - There is similar code to this block above for handling
3980 BRANCH nodes on the initial study. If you change stuff here
3982 regnode *trie_node= scan;
3983 regnode *tail= regnext(scan);
3984 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3985 I32 max1 = 0, min1 = I32_MAX;
3986 struct regnode_charclass_class accum;
3988 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3989 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3990 if (flags & SCF_DO_STCLASS)
3991 cl_init_zero(pRExC_state, &accum);
3997 const regnode *nextbranch= NULL;
4000 for ( word=1 ; word <= trie->wordcount ; word++)
4002 I32 deltanext=0, minnext=0, f = 0, fake;
4003 struct regnode_charclass_class this_class;
4005 data_fake.flags = 0;
4007 data_fake.whilem_c = data->whilem_c;
4008 data_fake.last_closep = data->last_closep;
4011 data_fake.last_closep = &fake;
4012 data_fake.pos_delta = delta;
4013 if (flags & SCF_DO_STCLASS) {
4014 cl_init(pRExC_state, &this_class);
4015 data_fake.start_class = &this_class;
4016 f = SCF_DO_STCLASS_AND;
4018 if (flags & SCF_WHILEM_VISITED_POS)
4019 f |= SCF_WHILEM_VISITED_POS;
4021 if (trie->jump[word]) {
4023 nextbranch = trie_node + trie->jump[0];
4024 scan= trie_node + trie->jump[word];
4025 /* We go from the jump point to the branch that follows
4026 it. Note this means we need the vestigal unused branches
4027 even though they arent otherwise used.
4029 minnext = study_chunk(pRExC_state, &scan, minlenp,
4030 &deltanext, (regnode *)nextbranch, &data_fake,
4031 stopparen, recursed, NULL, f,depth+1);
4033 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4034 nextbranch= regnext((regnode*)nextbranch);
4036 if (min1 > (I32)(minnext + trie->minlen))
4037 min1 = minnext + trie->minlen;
4038 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4039 max1 = minnext + deltanext + trie->maxlen;
4040 if (deltanext == I32_MAX)
4041 is_inf = is_inf_internal = 1;
4043 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4045 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4046 if ( stopmin > min + min1)
4047 stopmin = min + min1;
4048 flags &= ~SCF_DO_SUBSTR;
4050 data->flags |= SCF_SEEN_ACCEPT;
4053 if (data_fake.flags & SF_HAS_EVAL)
4054 data->flags |= SF_HAS_EVAL;
4055 data->whilem_c = data_fake.whilem_c;
4057 if (flags & SCF_DO_STCLASS)
4058 cl_or(pRExC_state, &accum, &this_class);
4061 if (flags & SCF_DO_SUBSTR) {
4062 data->pos_min += min1;
4063 data->pos_delta += max1 - min1;
4064 if (max1 != min1 || is_inf)
4065 data->longest = &(data->longest_float);
4068 delta += max1 - min1;
4069 if (flags & SCF_DO_STCLASS_OR) {
4070 cl_or(pRExC_state, data->start_class, &accum);
4072 cl_and(data->start_class, and_withp);
4073 flags &= ~SCF_DO_STCLASS;
4076 else if (flags & SCF_DO_STCLASS_AND) {
4078 cl_and(data->start_class, &accum);
4079 flags &= ~SCF_DO_STCLASS;
4082 /* Switch to OR mode: cache the old value of
4083 * data->start_class */
4085 StructCopy(data->start_class, and_withp,
4086 struct regnode_charclass_class);
4087 flags &= ~SCF_DO_STCLASS_AND;
4088 StructCopy(&accum, data->start_class,
4089 struct regnode_charclass_class);
4090 flags |= SCF_DO_STCLASS_OR;
4091 data->start_class->flags |= ANYOF_EOS;
4098 else if (PL_regkind[OP(scan)] == TRIE) {
4099 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4102 min += trie->minlen;
4103 delta += (trie->maxlen - trie->minlen);
4104 flags &= ~SCF_DO_STCLASS; /* xxx */
4105 if (flags & SCF_DO_SUBSTR) {
4106 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4107 data->pos_min += trie->minlen;
4108 data->pos_delta += (trie->maxlen - trie->minlen);
4109 if (trie->maxlen != trie->minlen)
4110 data->longest = &(data->longest_float);
4112 if (trie->jump) /* no more substrings -- for now /grr*/
4113 flags &= ~SCF_DO_SUBSTR;
4115 #endif /* old or new */
4116 #endif /* TRIE_STUDY_OPT */
4118 /* Else: zero-length, ignore. */
4119 scan = regnext(scan);
4124 stopparen = frame->stop;
4125 frame = frame->prev;
4126 goto fake_study_recurse;
4131 DEBUG_STUDYDATA("pre-fin:",data,depth);
4134 *deltap = is_inf_internal ? I32_MAX : delta;
4135 if (flags & SCF_DO_SUBSTR && is_inf)
4136 data->pos_delta = I32_MAX - data->pos_min;
4137 if (is_par > (I32)U8_MAX)
4139 if (is_par && pars==1 && data) {
4140 data->flags |= SF_IN_PAR;
4141 data->flags &= ~SF_HAS_PAR;
4143 else if (pars && data) {
4144 data->flags |= SF_HAS_PAR;
4145 data->flags &= ~SF_IN_PAR;
4147 if (flags & SCF_DO_STCLASS_OR)
4148 cl_and(data->start_class, and_withp);
4149 if (flags & SCF_TRIE_RESTUDY)
4150 data->flags |= SCF_TRIE_RESTUDY;
4152 DEBUG_STUDYDATA("post-fin:",data,depth);
4154 return min < stopmin ? min : stopmin;
4158 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4160 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4162 PERL_ARGS_ASSERT_ADD_DATA;
4164 Renewc(RExC_rxi->data,
4165 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4166 char, struct reg_data);
4168 Renew(RExC_rxi->data->what, count + n, U8);
4170 Newx(RExC_rxi->data->what, n, U8);
4171 RExC_rxi->data->count = count + n;
4172 Copy(s, RExC_rxi->data->what + count, n, U8);
4176 /*XXX: todo make this not included in a non debugging perl */
4177 #ifndef PERL_IN_XSUB_RE
4179 Perl_reginitcolors(pTHX)
4182 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4184 char *t = savepv(s);
4188 t = strchr(t, '\t');
4194 PL_colors[i] = t = (char *)"";
4199 PL_colors[i++] = (char *)"";
4206 #ifdef TRIE_STUDY_OPT
4207 #define CHECK_RESTUDY_GOTO \
4209 (data.flags & SCF_TRIE_RESTUDY) \
4213 #define CHECK_RESTUDY_GOTO
4217 - pregcomp - compile a regular expression into internal code
4219 * We can't allocate space until we know how big the compiled form will be,
4220 * but we can't compile it (and thus know how big it is) until we've got a
4221 * place to put the code. So we cheat: we compile it twice, once with code
4222 * generation turned off and size counting turned on, and once "for real".
4223 * This also means that we don't allocate space until we are sure that the
4224 * thing really will compile successfully, and we never have to move the
4225 * code and thus invalidate pointers into it. (Note that it has to be in
4226 * one piece because free() must be able to free it all.) [NB: not true in perl]
4228 * Beware that the optimization-preparation code in here knows about some
4229 * of the structure of the compiled regexp. [I'll say.]
4234 #ifndef PERL_IN_XSUB_RE
4235 #define RE_ENGINE_PTR &PL_core_reg_engine
4237 extern const struct regexp_engine my_reg_engine;
4238 #define RE_ENGINE_PTR &my_reg_engine
4241 #ifndef PERL_IN_XSUB_RE
4243 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4246 HV * const table = GvHV(PL_hintgv);
4248 PERL_ARGS_ASSERT_PREGCOMP;
4250 /* Dispatch a request to compile a regexp to correct
4253 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4254 GET_RE_DEBUG_FLAGS_DECL;
4255 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4256 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4258 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4261 return CALLREGCOMP_ENG(eng, pattern, flags);
4264 return Perl_re_compile(aTHX_ pattern, flags);
4269 Perl_re_compile(pTHX_ SV * const pattern, U32 pm_flags)
4274 register regexp_internal *ri;
4276 char *exp = SvPV(pattern, plen);
4277 char* xend = exp + plen;
4286 RExC_state_t RExC_state;
4287 RExC_state_t * const pRExC_state = &RExC_state;
4288 #ifdef TRIE_STUDY_OPT
4290 RExC_state_t copyRExC_state;
4292 GET_RE_DEBUG_FLAGS_DECL;
4294 PERL_ARGS_ASSERT_RE_COMPILE;
4296 DEBUG_r(if (!PL_colorset) reginitcolors());
4298 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4301 SV *dsv= sv_newmortal();
4302 RE_PV_QUOTED_DECL(s, RExC_utf8,
4303 dsv, exp, plen, 60);
4304 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4305 PL_colors[4],PL_colors[5],s);
4308 /* Longjmp back to here if have to switch in midstream to utf8 */
4309 if (! RExC_orig_utf8) {
4310 JMPENV_PUSH(jump_ret);
4313 if (jump_ret != 0) {
4316 /* Here, we longjmped back. If the cause was other than changing to
4317 * utf8, pop our own setjmp, and longjmp to the correct handler */
4318 if (jump_ret != UTF8_LONGJMP) {
4320 JMPENV_JUMP(jump_ret);
4325 /* It's possible to write a regexp in ascii that represents Unicode
4326 codepoints outside of the byte range, such as via \x{100}. If we
4327 detect such a sequence we have to convert the entire pattern to utf8
4328 and then recompile, as our sizing calculation will have been based
4329 on 1 byte == 1 character, but we will need to use utf8 to encode
4330 at least some part of the pattern, and therefore must convert the whole
4333 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4334 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4335 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4337 RExC_orig_utf8 = RExC_utf8 = 1;
4342 RExC_flags = pm_flags;
4346 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4347 RExC_seen_evals = 0;
4350 /* First pass: determine size, legality. */
4358 RExC_emit = &PL_regdummy;
4359 RExC_whilem_seen = 0;
4360 RExC_open_parens = NULL;
4361 RExC_close_parens = NULL;
4363 RExC_paren_names = NULL;
4365 RExC_paren_name_list = NULL;
4367 RExC_recurse = NULL;
4368 RExC_recurse_count = 0;
4370 #if 0 /* REGC() is (currently) a NOP at the first pass.
4371 * Clever compilers notice this and complain. --jhi */
4372 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4374 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4375 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4376 RExC_precomp = NULL;
4380 /* Here, finished first pass. Get rid of our setjmp, which we added for
4381 * efficiency only if the passed-in string wasn't in utf8, as shown by
4382 * RExC_orig_utf8. But if the first pass was redone, that variable will be
4383 * 1 here even though the original string wasn't utf8, but in this case
4384 * there will have been a long jump */
4385 if (jump_ret == UTF8_LONGJMP || ! RExC_orig_utf8) {
4389 PerlIO_printf(Perl_debug_log,
4390 "Required size %"IVdf" nodes\n"
4391 "Starting second pass (creation)\n",
4394 RExC_lastparse=NULL;
4396 /* Small enough for pointer-storage convention?
4397 If extralen==0, this means that we will not need long jumps. */
4398 if (RExC_size >= 0x10000L && RExC_extralen)
4399 RExC_size += RExC_extralen;
4402 if (RExC_whilem_seen > 15)
4403 RExC_whilem_seen = 15;
4405 /* Allocate space and zero-initialize. Note, the two step process
4406 of zeroing when in debug mode, thus anything assigned has to
4407 happen after that */
4408 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4409 r = (struct regexp*)SvANY(rx);
4410 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4411 char, regexp_internal);
4412 if ( r == NULL || ri == NULL )
4413 FAIL("Regexp out of space");
4415 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4416 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4418 /* bulk initialize base fields with 0. */
4419 Zero(ri, sizeof(regexp_internal), char);
4422 /* non-zero initialization begins here */
4424 r->engine= RE_ENGINE_PTR;
4425 r->extflags = pm_flags;
4427 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4428 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4429 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4430 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4431 >> RXf_PMf_STD_PMMOD_SHIFT);
4432 const char *fptr = STD_PAT_MODS; /*"msix"*/
4434 /* Allocate for the worst case, which is all the std flags are turned
4435 * on, but this means no caret. We never output a minus, as all those
4436 * are defaults, so are covered by the caret */
4437 const STRLEN wraplen = plen + has_p + has_runon
4438 + (sizeof(STD_PAT_MODS) - 1)
4439 + (sizeof("(?:)") - 1);
4441 p = sv_grow(MUTABLE_SV(rx), wraplen + 1);
4443 SvFLAGS(rx) |= SvUTF8(pattern);
4446 /* If a default, cover it using the caret */
4447 if (has_minus || (r->extflags & ~(RXf_PMf_LOCALE|RXf_PMf_UNICODE))) {
4448 *p++= DEFAULT_PAT_MOD;
4450 if (r->extflags & RXf_PMf_LOCALE) {
4451 *p++ = LOCALE_PAT_MOD;
4453 else if (r->extflags & RXf_PMf_UNICODE) {
4454 *p++ = UNICODE_PAT_MOD;
4457 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4460 while((ch = *fptr++)) {
4468 Copy(RExC_precomp, p, plen, char);
4469 assert ((RX_WRAPPED(rx) - p) < 16);
4470 r->pre_prefix = p - RX_WRAPPED(rx);
4476 SvCUR_set(rx, p - SvPVX_const(rx));
4480 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4482 if (RExC_seen & REG_SEEN_RECURSE) {
4483 Newxz(RExC_open_parens, RExC_npar,regnode *);
4484 SAVEFREEPV(RExC_open_parens);
4485 Newxz(RExC_close_parens,RExC_npar,regnode *);
4486 SAVEFREEPV(RExC_close_parens);
4489 /* Useful during FAIL. */
4490 #ifdef RE_TRACK_PATTERN_OFFSETS
4491 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4492 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4493 "%s %"UVuf" bytes for offset annotations.\n",
4494 ri->u.offsets ? "Got" : "Couldn't get",
4495 (UV)((2*RExC_size+1) * sizeof(U32))));
4497 SetProgLen(ri,RExC_size);
4502 /* Second pass: emit code. */
4503 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4508 RExC_emit_start = ri->program;
4509 RExC_emit = ri->program;
4510 RExC_emit_bound = ri->program + RExC_size + 1;
4512 /* Store the count of eval-groups for security checks: */
4513 RExC_rx->seen_evals = RExC_seen_evals;
4514 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4515 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4519 /* XXXX To minimize changes to RE engine we always allocate
4520 3-units-long substrs field. */
4521 Newx(r->substrs, 1, struct reg_substr_data);
4522 if (RExC_recurse_count) {
4523 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4524 SAVEFREEPV(RExC_recurse);
4528 r->minlen = minlen = sawplus = sawopen = 0;
4529 Zero(r->substrs, 1, struct reg_substr_data);
4531 #ifdef TRIE_STUDY_OPT
4533 StructCopy(&zero_scan_data, &data, scan_data_t);
4534 copyRExC_state = RExC_state;
4537 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4539 RExC_state = copyRExC_state;
4540 if (seen & REG_TOP_LEVEL_BRANCHES)
4541 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4543 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4544 if (data.last_found) {
4545 SvREFCNT_dec(data.longest_fixed);
4546 SvREFCNT_dec(data.longest_float);
4547 SvREFCNT_dec(data.last_found);
4549 StructCopy(&zero_scan_data, &data, scan_data_t);
4552 StructCopy(&zero_scan_data, &data, scan_data_t);
4555 /* Dig out information for optimizations. */
4556 r->extflags = RExC_flags; /* was pm_op */
4557 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4560 SvUTF8_on(rx); /* Unicode in it? */
4561 ri->regstclass = NULL;
4562 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4563 r->intflags |= PREGf_NAUGHTY;
4564 scan = ri->program + 1; /* First BRANCH. */
4566 /* testing for BRANCH here tells us whether there is "must appear"
4567 data in the pattern. If there is then we can use it for optimisations */
4568 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4570 STRLEN longest_float_length, longest_fixed_length;
4571 struct regnode_charclass_class ch_class; /* pointed to by data */
4573 I32 last_close = 0; /* pointed to by data */
4574 regnode *first= scan;
4575 regnode *first_next= regnext(first);
4578 * Skip introductions and multiplicators >= 1
4579 * so that we can extract the 'meat' of the pattern that must
4580 * match in the large if() sequence following.
4581 * NOTE that EXACT is NOT covered here, as it is normally
4582 * picked up by the optimiser separately.
4584 * This is unfortunate as the optimiser isnt handling lookahead
4585 * properly currently.
4588 while ((OP(first) == OPEN && (sawopen = 1)) ||
4589 /* An OR of *one* alternative - should not happen now. */
4590 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4591 /* for now we can't handle lookbehind IFMATCH*/
4592 (OP(first) == IFMATCH && !first->flags) ||
4593 (OP(first) == PLUS) ||
4594 (OP(first) == MINMOD) ||
4595 /* An {n,m} with n>0 */
4596 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4597 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4600 * the only op that could be a regnode is PLUS, all the rest
4601 * will be regnode_1 or regnode_2.
4604 if (OP(first) == PLUS)
4607 first += regarglen[OP(first)];
4609 first = NEXTOPER(first);
4610 first_next= regnext(first);
4613 /* Starting-point info. */
4615 DEBUG_PEEP("first:",first,0);
4616 /* Ignore EXACT as we deal with it later. */
4617 if (PL_regkind[OP(first)] == EXACT) {
4618 if (OP(first) == EXACT)
4619 NOOP; /* Empty, get anchored substr later. */
4620 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4621 ri->regstclass = first;
4624 else if (PL_regkind[OP(first)] == TRIE &&
4625 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4628 /* this can happen only on restudy */
4629 if ( OP(first) == TRIE ) {
4630 struct regnode_1 *trieop = (struct regnode_1 *)
4631 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4632 StructCopy(first,trieop,struct regnode_1);
4633 trie_op=(regnode *)trieop;
4635 struct regnode_charclass *trieop = (struct regnode_charclass *)
4636 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4637 StructCopy(first,trieop,struct regnode_charclass);
4638 trie_op=(regnode *)trieop;
4641 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4642 ri->regstclass = trie_op;
4645 else if (REGNODE_SIMPLE(OP(first)))
4646 ri->regstclass = first;
4647 else if (PL_regkind[OP(first)] == BOUND ||
4648 PL_regkind[OP(first)] == NBOUND)
4649 ri->regstclass = first;
4650 else if (PL_regkind[OP(first)] == BOL) {
4651 r->extflags |= (OP(first) == MBOL
4653 : (OP(first) == SBOL
4656 first = NEXTOPER(first);
4659 else if (OP(first) == GPOS) {
4660 r->extflags |= RXf_ANCH_GPOS;
4661 first = NEXTOPER(first);
4664 else if ((!sawopen || !RExC_sawback) &&
4665 (OP(first) == STAR &&
4666 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4667 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4669 /* turn .* into ^.* with an implied $*=1 */
4671 (OP(NEXTOPER(first)) == REG_ANY)
4674 r->extflags |= type;
4675 r->intflags |= PREGf_IMPLICIT;
4676 first = NEXTOPER(first);
4679 if (sawplus && (!sawopen || !RExC_sawback)
4680 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4681 /* x+ must match at the 1st pos of run of x's */
4682 r->intflags |= PREGf_SKIP;
4684 /* Scan is after the zeroth branch, first is atomic matcher. */
4685 #ifdef TRIE_STUDY_OPT
4688 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4689 (IV)(first - scan + 1))
4693 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4694 (IV)(first - scan + 1))
4700 * If there's something expensive in the r.e., find the
4701 * longest literal string that must appear and make it the
4702 * regmust. Resolve ties in favor of later strings, since
4703 * the regstart check works with the beginning of the r.e.
4704 * and avoiding duplication strengthens checking. Not a
4705 * strong reason, but sufficient in the absence of others.
4706 * [Now we resolve ties in favor of the earlier string if
4707 * it happens that c_offset_min has been invalidated, since the
4708 * earlier string may buy us something the later one won't.]
4711 data.longest_fixed = newSVpvs("");
4712 data.longest_float = newSVpvs("");
4713 data.last_found = newSVpvs("");
4714 data.longest = &(data.longest_fixed);
4716 if (!ri->regstclass) {
4717 cl_init(pRExC_state, &ch_class);
4718 data.start_class = &ch_class;
4719 stclass_flag = SCF_DO_STCLASS_AND;
4720 } else /* XXXX Check for BOUND? */
4722 data.last_closep = &last_close;
4724 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4725 &data, -1, NULL, NULL,
4726 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4732 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4733 && data.last_start_min == 0 && data.last_end > 0
4734 && !RExC_seen_zerolen
4735 && !(RExC_seen & REG_SEEN_VERBARG)
4736 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4737 r->extflags |= RXf_CHECK_ALL;
4738 scan_commit(pRExC_state, &data,&minlen,0);
4739 SvREFCNT_dec(data.last_found);
4741 /* Note that code very similar to this but for anchored string
4742 follows immediately below, changes may need to be made to both.
4745 longest_float_length = CHR_SVLEN(data.longest_float);
4746 if (longest_float_length
4747 || (data.flags & SF_FL_BEFORE_EOL
4748 && (!(data.flags & SF_FL_BEFORE_MEOL)
4749 || (RExC_flags & RXf_PMf_MULTILINE))))
4753 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4754 && data.offset_fixed == data.offset_float_min
4755 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4756 goto remove_float; /* As in (a)+. */
4758 /* copy the information about the longest float from the reg_scan_data
4759 over to the program. */
4760 if (SvUTF8(data.longest_float)) {
4761 r->float_utf8 = data.longest_float;
4762 r->float_substr = NULL;
4764 r->float_substr = data.longest_float;
4765 r->float_utf8 = NULL;
4767 /* float_end_shift is how many chars that must be matched that
4768 follow this item. We calculate it ahead of time as once the
4769 lookbehind offset is added in we lose the ability to correctly
4771 ml = data.minlen_float ? *(data.minlen_float)
4772 : (I32)longest_float_length;
4773 r->float_end_shift = ml - data.offset_float_min
4774 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4775 + data.lookbehind_float;
4776 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4777 r->float_max_offset = data.offset_float_max;
4778 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4779 r->float_max_offset -= data.lookbehind_float;
4781 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4782 && (!(data.flags & SF_FL_BEFORE_MEOL)
4783 || (RExC_flags & RXf_PMf_MULTILINE)));
4784 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4788 r->float_substr = r->float_utf8 = NULL;
4789 SvREFCNT_dec(data.longest_float);
4790 longest_float_length = 0;
4793 /* Note that code very similar to this but for floating string
4794 is immediately above, changes may need to be made to both.
4797 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4798 if (longest_fixed_length
4799 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4800 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4801 || (RExC_flags & RXf_PMf_MULTILINE))))
4805 /* copy the information about the longest fixed
4806 from the reg_scan_data over to the program. */
4807 if (SvUTF8(data.longest_fixed)) {
4808 r->anchored_utf8 = data.longest_fixed;
4809 r->anchored_substr = NULL;
4811 r->anchored_substr = data.longest_fixed;
4812 r->anchored_utf8 = NULL;
4814 /* fixed_end_shift is how many chars that must be matched that
4815 follow this item. We calculate it ahead of time as once the
4816 lookbehind offset is added in we lose the ability to correctly
4818 ml = data.minlen_fixed ? *(data.minlen_fixed)
4819 : (I32)longest_fixed_length;
4820 r->anchored_end_shift = ml - data.offset_fixed
4821 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4822 + data.lookbehind_fixed;
4823 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4825 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4826 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4827 || (RExC_flags & RXf_PMf_MULTILINE)));
4828 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4831 r->anchored_substr = r->anchored_utf8 = NULL;
4832 SvREFCNT_dec(data.longest_fixed);
4833 longest_fixed_length = 0;
4836 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4837 ri->regstclass = NULL;
4838 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4840 && !(data.start_class->flags & ANYOF_EOS)
4841 && !cl_is_anything(data.start_class))
4843 const U32 n = add_data(pRExC_state, 1, "f");
4845 Newx(RExC_rxi->data->data[n], 1,
4846 struct regnode_charclass_class);
4847 StructCopy(data.start_class,
4848 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4849 struct regnode_charclass_class);
4850 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4851 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4852 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4853 regprop(r, sv, (regnode*)data.start_class);
4854 PerlIO_printf(Perl_debug_log,
4855 "synthetic stclass \"%s\".\n",
4856 SvPVX_const(sv));});
4859 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4860 if (longest_fixed_length > longest_float_length) {
4861 r->check_end_shift = r->anchored_end_shift;
4862 r->check_substr = r->anchored_substr;
4863 r->check_utf8 = r->anchored_utf8;
4864 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4865 if (r->extflags & RXf_ANCH_SINGLE)
4866 r->extflags |= RXf_NOSCAN;
4869 r->check_end_shift = r->float_end_shift;
4870 r->check_substr = r->float_substr;
4871 r->check_utf8 = r->float_utf8;
4872 r->check_offset_min = r->float_min_offset;
4873 r->check_offset_max = r->float_max_offset;
4875 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4876 This should be changed ASAP! */
4877 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4878 r->extflags |= RXf_USE_INTUIT;
4879 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4880 r->extflags |= RXf_INTUIT_TAIL;
4882 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4883 if ( (STRLEN)minlen < longest_float_length )
4884 minlen= longest_float_length;
4885 if ( (STRLEN)minlen < longest_fixed_length )
4886 minlen= longest_fixed_length;
4890 /* Several toplevels. Best we can is to set minlen. */
4892 struct regnode_charclass_class ch_class;
4895 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4897 scan = ri->program + 1;
4898 cl_init(pRExC_state, &ch_class);
4899 data.start_class = &ch_class;
4900 data.last_closep = &last_close;
4903 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4904 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4908 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4909 = r->float_substr = r->float_utf8 = NULL;
4910 if (!(data.start_class->flags & ANYOF_EOS)
4911 && !cl_is_anything(data.start_class))
4913 const U32 n = add_data(pRExC_state, 1, "f");
4915 Newx(RExC_rxi->data->data[n], 1,
4916 struct regnode_charclass_class);
4917 StructCopy(data.start_class,
4918 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4919 struct regnode_charclass_class);
4920 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4921 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4922 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4923 regprop(r, sv, (regnode*)data.start_class);
4924 PerlIO_printf(Perl_debug_log,
4925 "synthetic stclass \"%s\".\n",
4926 SvPVX_const(sv));});
4930 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4931 the "real" pattern. */
4933 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4934 (IV)minlen, (IV)r->minlen);
4936 r->minlenret = minlen;
4937 if (r->minlen < minlen)
4940 if (RExC_seen & REG_SEEN_GPOS)
4941 r->extflags |= RXf_GPOS_SEEN;
4942 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4943 r->extflags |= RXf_LOOKBEHIND_SEEN;
4944 if (RExC_seen & REG_SEEN_EVAL)
4945 r->extflags |= RXf_EVAL_SEEN;
4946 if (RExC_seen & REG_SEEN_CANY)
4947 r->extflags |= RXf_CANY_SEEN;
4948 if (RExC_seen & REG_SEEN_VERBARG)
4949 r->intflags |= PREGf_VERBARG_SEEN;
4950 if (RExC_seen & REG_SEEN_CUTGROUP)
4951 r->intflags |= PREGf_CUTGROUP_SEEN;
4952 if (RExC_paren_names)
4953 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
4955 RXp_PAREN_NAMES(r) = NULL;
4957 #ifdef STUPID_PATTERN_CHECKS
4958 if (RX_PRELEN(rx) == 0)
4959 r->extflags |= RXf_NULL;
4960 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4961 /* XXX: this should happen BEFORE we compile */
4962 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4963 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
4964 r->extflags |= RXf_WHITE;
4965 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
4966 r->extflags |= RXf_START_ONLY;
4968 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4969 /* XXX: this should happen BEFORE we compile */
4970 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4972 regnode *first = ri->program + 1;
4974 U8 nop = OP(NEXTOPER(first));
4976 if (PL_regkind[fop] == NOTHING && nop == END)
4977 r->extflags |= RXf_NULL;
4978 else if (PL_regkind[fop] == BOL && nop == END)
4979 r->extflags |= RXf_START_ONLY;
4980 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4981 r->extflags |= RXf_WHITE;
4985 if (RExC_paren_names) {
4986 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
4987 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4990 ri->name_list_idx = 0;
4992 if (RExC_recurse_count) {
4993 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4994 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4995 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4998 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4999 /* assume we don't need to swap parens around before we match */
5002 PerlIO_printf(Perl_debug_log,"Final program:\n");
5005 #ifdef RE_TRACK_PATTERN_OFFSETS
5006 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5007 const U32 len = ri->u.offsets[0];
5009 GET_RE_DEBUG_FLAGS_DECL;
5010 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5011 for (i = 1; i <= len; i++) {
5012 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5013 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5014 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5016 PerlIO_printf(Perl_debug_log, "\n");
5022 #undef RE_ENGINE_PTR
5026 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5029 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5031 PERL_UNUSED_ARG(value);
5033 if (flags & RXapif_FETCH) {
5034 return reg_named_buff_fetch(rx, key, flags);
5035 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5036 Perl_croak_no_modify(aTHX);
5038 } else if (flags & RXapif_EXISTS) {
5039 return reg_named_buff_exists(rx, key, flags)
5042 } else if (flags & RXapif_REGNAMES) {
5043 return reg_named_buff_all(rx, flags);
5044 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5045 return reg_named_buff_scalar(rx, flags);
5047 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5053 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5056 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5057 PERL_UNUSED_ARG(lastkey);
5059 if (flags & RXapif_FIRSTKEY)
5060 return reg_named_buff_firstkey(rx, flags);
5061 else if (flags & RXapif_NEXTKEY)
5062 return reg_named_buff_nextkey(rx, flags);
5064 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5070 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5073 AV *retarray = NULL;
5075 struct regexp *const rx = (struct regexp *)SvANY(r);
5077 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5079 if (flags & RXapif_ALL)
5082 if (rx && RXp_PAREN_NAMES(rx)) {
5083 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5086 SV* sv_dat=HeVAL(he_str);
5087 I32 *nums=(I32*)SvPVX(sv_dat);
5088 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5089 if ((I32)(rx->nparens) >= nums[i]
5090 && rx->offs[nums[i]].start != -1
5091 && rx->offs[nums[i]].end != -1)
5094 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5098 ret = newSVsv(&PL_sv_undef);
5101 av_push(retarray, ret);
5104 return newRV_noinc(MUTABLE_SV(retarray));
5111 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5114 struct regexp *const rx = (struct regexp *)SvANY(r);
5116 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5118 if (rx && RXp_PAREN_NAMES(rx)) {
5119 if (flags & RXapif_ALL) {
5120 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5122 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5136 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5138 struct regexp *const rx = (struct regexp *)SvANY(r);
5140 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5142 if ( rx && RXp_PAREN_NAMES(rx) ) {
5143 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5145 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5152 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5154 struct regexp *const rx = (struct regexp *)SvANY(r);
5155 GET_RE_DEBUG_FLAGS_DECL;
5157 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5159 if (rx && RXp_PAREN_NAMES(rx)) {
5160 HV *hv = RXp_PAREN_NAMES(rx);
5162 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5165 SV* sv_dat = HeVAL(temphe);
5166 I32 *nums = (I32*)SvPVX(sv_dat);
5167 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5168 if ((I32)(rx->lastparen) >= nums[i] &&
5169 rx->offs[nums[i]].start != -1 &&
5170 rx->offs[nums[i]].end != -1)
5176 if (parno || flags & RXapif_ALL) {
5177 return newSVhek(HeKEY_hek(temphe));
5185 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5190 struct regexp *const rx = (struct regexp *)SvANY(r);
5192 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5194 if (rx && RXp_PAREN_NAMES(rx)) {
5195 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5196 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5197 } else if (flags & RXapif_ONE) {
5198 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5199 av = MUTABLE_AV(SvRV(ret));
5200 length = av_len(av);
5202 return newSViv(length + 1);
5204 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5208 return &PL_sv_undef;
5212 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5214 struct regexp *const rx = (struct regexp *)SvANY(r);
5217 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5219 if (rx && RXp_PAREN_NAMES(rx)) {
5220 HV *hv= RXp_PAREN_NAMES(rx);
5222 (void)hv_iterinit(hv);
5223 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5226 SV* sv_dat = HeVAL(temphe);
5227 I32 *nums = (I32*)SvPVX(sv_dat);
5228 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5229 if ((I32)(rx->lastparen) >= nums[i] &&
5230 rx->offs[nums[i]].start != -1 &&
5231 rx->offs[nums[i]].end != -1)
5237 if (parno || flags & RXapif_ALL) {
5238 av_push(av, newSVhek(HeKEY_hek(temphe)));
5243 return newRV_noinc(MUTABLE_SV(av));
5247 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5250 struct regexp *const rx = (struct regexp *)SvANY(r);
5255 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5258 sv_setsv(sv,&PL_sv_undef);
5262 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5264 i = rx->offs[0].start;
5268 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5270 s = rx->subbeg + rx->offs[0].end;
5271 i = rx->sublen - rx->offs[0].end;
5274 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5275 (s1 = rx->offs[paren].start) != -1 &&
5276 (t1 = rx->offs[paren].end) != -1)
5280 s = rx->subbeg + s1;
5282 sv_setsv(sv,&PL_sv_undef);
5285 assert(rx->sublen >= (s - rx->subbeg) + i );
5287 const int oldtainted = PL_tainted;
5289 sv_setpvn(sv, s, i);
5290 PL_tainted = oldtainted;
5291 if ( (rx->extflags & RXf_CANY_SEEN)
5292 ? (RXp_MATCH_UTF8(rx)
5293 && (!i || is_utf8_string((U8*)s, i)))
5294 : (RXp_MATCH_UTF8(rx)) )
5301 if (RXp_MATCH_TAINTED(rx)) {
5302 if (SvTYPE(sv) >= SVt_PVMG) {
5303 MAGIC* const mg = SvMAGIC(sv);
5306 SvMAGIC_set(sv, mg->mg_moremagic);
5308 if ((mgt = SvMAGIC(sv))) {
5309 mg->mg_moremagic = mgt;
5310 SvMAGIC_set(sv, mg);
5320 sv_setsv(sv,&PL_sv_undef);
5326 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5327 SV const * const value)
5329 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5331 PERL_UNUSED_ARG(rx);
5332 PERL_UNUSED_ARG(paren);
5333 PERL_UNUSED_ARG(value);
5336 Perl_croak_no_modify(aTHX);
5340 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5343 struct regexp *const rx = (struct regexp *)SvANY(r);
5347 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5349 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5351 /* $` / ${^PREMATCH} */
5352 case RX_BUFF_IDX_PREMATCH:
5353 if (rx->offs[0].start != -1) {
5354 i = rx->offs[0].start;
5362 /* $' / ${^POSTMATCH} */
5363 case RX_BUFF_IDX_POSTMATCH:
5364 if (rx->offs[0].end != -1) {
5365 i = rx->sublen - rx->offs[0].end;
5367 s1 = rx->offs[0].end;
5373 /* $& / ${^MATCH}, $1, $2, ... */
5375 if (paren <= (I32)rx->nparens &&
5376 (s1 = rx->offs[paren].start) != -1 &&
5377 (t1 = rx->offs[paren].end) != -1)
5382 if (ckWARN(WARN_UNINITIALIZED))
5383 report_uninit((const SV *)sv);
5388 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5389 const char * const s = rx->subbeg + s1;
5394 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5401 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5403 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5404 PERL_UNUSED_ARG(rx);
5408 return newSVpvs("Regexp");
5411 /* Scans the name of a named buffer from the pattern.
5412 * If flags is REG_RSN_RETURN_NULL returns null.
5413 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5414 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5415 * to the parsed name as looked up in the RExC_paren_names hash.
5416 * If there is an error throws a vFAIL().. type exception.
5419 #define REG_RSN_RETURN_NULL 0
5420 #define REG_RSN_RETURN_NAME 1
5421 #define REG_RSN_RETURN_DATA 2
5424 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5426 char *name_start = RExC_parse;
5428 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5430 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5431 /* skip IDFIRST by using do...while */
5434 RExC_parse += UTF8SKIP(RExC_parse);
5435 } while (isALNUM_utf8((U8*)RExC_parse));
5439 } while (isALNUM(*RExC_parse));
5444 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5445 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5446 if ( flags == REG_RSN_RETURN_NAME)
5448 else if (flags==REG_RSN_RETURN_DATA) {
5451 if ( ! sv_name ) /* should not happen*/
5452 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5453 if (RExC_paren_names)
5454 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5456 sv_dat = HeVAL(he_str);
5458 vFAIL("Reference to nonexistent named group");
5462 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5469 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5470 int rem=(int)(RExC_end - RExC_parse); \
5479 if (RExC_lastparse!=RExC_parse) \
5480 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5483 iscut ? "..." : "<" \
5486 PerlIO_printf(Perl_debug_log,"%16s",""); \
5489 num = RExC_size + 1; \
5491 num=REG_NODE_NUM(RExC_emit); \
5492 if (RExC_lastnum!=num) \
5493 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5495 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5496 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5497 (int)((depth*2)), "", \
5501 RExC_lastparse=RExC_parse; \
5506 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5507 DEBUG_PARSE_MSG((funcname)); \
5508 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5510 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5511 DEBUG_PARSE_MSG((funcname)); \
5512 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5515 - reg - regular expression, i.e. main body or parenthesized thing
5517 * Caller must absorb opening parenthesis.
5519 * Combining parenthesis handling with the base level of regular expression
5520 * is a trifle forced, but the need to tie the tails of the branches to what
5521 * follows makes it hard to avoid.
5523 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5525 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5527 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5531 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5532 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5535 register regnode *ret; /* Will be the head of the group. */
5536 register regnode *br;
5537 register regnode *lastbr;
5538 register regnode *ender = NULL;
5539 register I32 parno = 0;
5541 U32 oregflags = RExC_flags;
5542 bool have_branch = 0;
5544 I32 freeze_paren = 0;
5545 I32 after_freeze = 0;
5547 /* for (?g), (?gc), and (?o) warnings; warning
5548 about (?c) will warn about (?g) -- japhy */
5550 #define WASTED_O 0x01
5551 #define WASTED_G 0x02
5552 #define WASTED_C 0x04
5553 #define WASTED_GC (0x02|0x04)
5554 I32 wastedflags = 0x00;
5556 char * parse_start = RExC_parse; /* MJD */
5557 char * const oregcomp_parse = RExC_parse;
5559 GET_RE_DEBUG_FLAGS_DECL;
5561 PERL_ARGS_ASSERT_REG;
5562 DEBUG_PARSE("reg ");
5564 *flagp = 0; /* Tentatively. */
5567 /* Make an OPEN node, if parenthesized. */
5569 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5570 char *start_verb = RExC_parse;
5571 STRLEN verb_len = 0;
5572 char *start_arg = NULL;
5573 unsigned char op = 0;
5575 int internal_argval = 0; /* internal_argval is only useful if !argok */
5576 while ( *RExC_parse && *RExC_parse != ')' ) {
5577 if ( *RExC_parse == ':' ) {
5578 start_arg = RExC_parse + 1;
5584 verb_len = RExC_parse - start_verb;
5587 while ( *RExC_parse && *RExC_parse != ')' )
5589 if ( *RExC_parse != ')' )
5590 vFAIL("Unterminated verb pattern argument");
5591 if ( RExC_parse == start_arg )
5594 if ( *RExC_parse != ')' )
5595 vFAIL("Unterminated verb pattern");
5598 switch ( *start_verb ) {
5599 case 'A': /* (*ACCEPT) */
5600 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5602 internal_argval = RExC_nestroot;
5605 case 'C': /* (*COMMIT) */
5606 if ( memEQs(start_verb,verb_len,"COMMIT") )
5609 case 'F': /* (*FAIL) */
5610 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5615 case ':': /* (*:NAME) */
5616 case 'M': /* (*MARK:NAME) */
5617 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5622 case 'P': /* (*PRUNE) */
5623 if ( memEQs(start_verb,verb_len,"PRUNE") )
5626 case 'S': /* (*SKIP) */
5627 if ( memEQs(start_verb,verb_len,"SKIP") )
5630 case 'T': /* (*THEN) */
5631 /* [19:06] <TimToady> :: is then */
5632 if ( memEQs(start_verb,verb_len,"THEN") ) {
5634 RExC_seen |= REG_SEEN_CUTGROUP;
5640 vFAIL3("Unknown verb pattern '%.*s'",
5641 verb_len, start_verb);
5644 if ( start_arg && internal_argval ) {
5645 vFAIL3("Verb pattern '%.*s' may not have an argument",
5646 verb_len, start_verb);
5647 } else if ( argok < 0 && !start_arg ) {
5648 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5649 verb_len, start_verb);
5651 ret = reganode(pRExC_state, op, internal_argval);
5652 if ( ! internal_argval && ! SIZE_ONLY ) {
5654 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5655 ARG(ret) = add_data( pRExC_state, 1, "S" );
5656 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5663 if (!internal_argval)
5664 RExC_seen |= REG_SEEN_VERBARG;
5665 } else if ( start_arg ) {
5666 vFAIL3("Verb pattern '%.*s' may not have an argument",
5667 verb_len, start_verb);
5669 ret = reg_node(pRExC_state, op);
5671 nextchar(pRExC_state);
5674 if (*RExC_parse == '?') { /* (?...) */
5675 bool is_logical = 0;
5676 const char * const seqstart = RExC_parse;
5677 bool has_use_defaults = FALSE;
5680 paren = *RExC_parse++;
5681 ret = NULL; /* For look-ahead/behind. */
5684 case 'P': /* (?P...) variants for those used to PCRE/Python */
5685 paren = *RExC_parse++;
5686 if ( paren == '<') /* (?P<...>) named capture */
5688 else if (paren == '>') { /* (?P>name) named recursion */
5689 goto named_recursion;
5691 else if (paren == '=') { /* (?P=...) named backref */
5692 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5693 you change this make sure you change that */
5694 char* name_start = RExC_parse;
5696 SV *sv_dat = reg_scan_name(pRExC_state,
5697 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5698 if (RExC_parse == name_start || *RExC_parse != ')')
5699 vFAIL2("Sequence %.3s... not terminated",parse_start);
5702 num = add_data( pRExC_state, 1, "S" );
5703 RExC_rxi->data->data[num]=(void*)sv_dat;
5704 SvREFCNT_inc_simple_void(sv_dat);
5707 ret = reganode(pRExC_state,
5708 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5712 Set_Node_Offset(ret, parse_start+1);
5713 Set_Node_Cur_Length(ret); /* MJD */
5715 nextchar(pRExC_state);
5719 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5721 case '<': /* (?<...) */
5722 if (*RExC_parse == '!')
5724 else if (*RExC_parse != '=')
5730 case '\'': /* (?'...') */
5731 name_start= RExC_parse;
5732 svname = reg_scan_name(pRExC_state,
5733 SIZE_ONLY ? /* reverse test from the others */
5734 REG_RSN_RETURN_NAME :
5735 REG_RSN_RETURN_NULL);
5736 if (RExC_parse == name_start) {
5738 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5741 if (*RExC_parse != paren)
5742 vFAIL2("Sequence (?%c... not terminated",
5743 paren=='>' ? '<' : paren);
5747 if (!svname) /* shouldnt happen */
5749 "panic: reg_scan_name returned NULL");
5750 if (!RExC_paren_names) {
5751 RExC_paren_names= newHV();
5752 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5754 RExC_paren_name_list= newAV();
5755 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5758 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5760 sv_dat = HeVAL(he_str);
5762 /* croak baby croak */
5764 "panic: paren_name hash element allocation failed");
5765 } else if ( SvPOK(sv_dat) ) {
5766 /* (?|...) can mean we have dupes so scan to check
5767 its already been stored. Maybe a flag indicating
5768 we are inside such a construct would be useful,
5769 but the arrays are likely to be quite small, so
5770 for now we punt -- dmq */
5771 IV count = SvIV(sv_dat);
5772 I32 *pv = (I32*)SvPVX(sv_dat);
5774 for ( i = 0 ; i < count ; i++ ) {
5775 if ( pv[i] == RExC_npar ) {
5781 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5782 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5783 pv[count] = RExC_npar;
5784 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5787 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5788 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5790 SvIV_set(sv_dat, 1);
5793 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5794 SvREFCNT_dec(svname);
5797 /*sv_dump(sv_dat);*/
5799 nextchar(pRExC_state);
5801 goto capturing_parens;
5803 RExC_seen |= REG_SEEN_LOOKBEHIND;
5805 case '=': /* (?=...) */
5806 RExC_seen_zerolen++;
5808 case '!': /* (?!...) */
5809 RExC_seen_zerolen++;
5810 if (*RExC_parse == ')') {
5811 ret=reg_node(pRExC_state, OPFAIL);
5812 nextchar(pRExC_state);
5816 case '|': /* (?|...) */
5817 /* branch reset, behave like a (?:...) except that
5818 buffers in alternations share the same numbers */
5820 after_freeze = freeze_paren = RExC_npar;
5822 case ':': /* (?:...) */
5823 case '>': /* (?>...) */
5825 case '$': /* (?$...) */
5826 case '@': /* (?@...) */
5827 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5829 case '#': /* (?#...) */
5830 while (*RExC_parse && *RExC_parse != ')')
5832 if (*RExC_parse != ')')
5833 FAIL("Sequence (?#... not terminated");
5834 nextchar(pRExC_state);
5837 case '0' : /* (?0) */
5838 case 'R' : /* (?R) */
5839 if (*RExC_parse != ')')
5840 FAIL("Sequence (?R) not terminated");
5841 ret = reg_node(pRExC_state, GOSTART);
5842 *flagp |= POSTPONED;
5843 nextchar(pRExC_state);
5846 { /* named and numeric backreferences */
5848 case '&': /* (?&NAME) */
5849 parse_start = RExC_parse - 1;
5852 SV *sv_dat = reg_scan_name(pRExC_state,
5853 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5854 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5856 goto gen_recurse_regop;
5859 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5861 vFAIL("Illegal pattern");
5863 goto parse_recursion;
5865 case '-': /* (?-1) */
5866 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5867 RExC_parse--; /* rewind to let it be handled later */
5871 case '1': case '2': case '3': case '4': /* (?1) */
5872 case '5': case '6': case '7': case '8': case '9':
5875 num = atoi(RExC_parse);
5876 parse_start = RExC_parse - 1; /* MJD */
5877 if (*RExC_parse == '-')
5879 while (isDIGIT(*RExC_parse))
5881 if (*RExC_parse!=')')
5882 vFAIL("Expecting close bracket");
5885 if ( paren == '-' ) {
5887 Diagram of capture buffer numbering.
5888 Top line is the normal capture buffer numbers
5889 Botton line is the negative indexing as from
5893 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5897 num = RExC_npar + num;
5900 vFAIL("Reference to nonexistent group");
5902 } else if ( paren == '+' ) {
5903 num = RExC_npar + num - 1;
5906 ret = reganode(pRExC_state, GOSUB, num);
5908 if (num > (I32)RExC_rx->nparens) {
5910 vFAIL("Reference to nonexistent group");
5912 ARG2L_SET( ret, RExC_recurse_count++);
5914 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5915 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5919 RExC_seen |= REG_SEEN_RECURSE;
5920 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5921 Set_Node_Offset(ret, parse_start); /* MJD */
5923 *flagp |= POSTPONED;
5924 nextchar(pRExC_state);
5926 } /* named and numeric backreferences */
5929 case '?': /* (??...) */
5931 if (*RExC_parse != '{') {
5933 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5936 *flagp |= POSTPONED;
5937 paren = *RExC_parse++;
5939 case '{': /* (?{...}) */
5944 char *s = RExC_parse;
5946 RExC_seen_zerolen++;
5947 RExC_seen |= REG_SEEN_EVAL;
5948 while (count && (c = *RExC_parse)) {
5959 if (*RExC_parse != ')') {
5961 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5965 OP_4tree *sop, *rop;
5966 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5969 Perl_save_re_context(aTHX);
5970 rop = sv_compile_2op(sv, &sop, "re", &pad);
5971 sop->op_private |= OPpREFCOUNTED;
5972 /* re_dup will OpREFCNT_inc */
5973 OpREFCNT_set(sop, 1);
5976 n = add_data(pRExC_state, 3, "nop");
5977 RExC_rxi->data->data[n] = (void*)rop;
5978 RExC_rxi->data->data[n+1] = (void*)sop;
5979 RExC_rxi->data->data[n+2] = (void*)pad;
5982 else { /* First pass */
5983 if (PL_reginterp_cnt < ++RExC_seen_evals
5985 /* No compiled RE interpolated, has runtime
5986 components ===> unsafe. */
5987 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5988 if (PL_tainting && PL_tainted)
5989 FAIL("Eval-group in insecure regular expression");
5990 #if PERL_VERSION > 8
5991 if (IN_PERL_COMPILETIME)
5996 nextchar(pRExC_state);
5998 ret = reg_node(pRExC_state, LOGICAL);
6001 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
6002 /* deal with the length of this later - MJD */
6005 ret = reganode(pRExC_state, EVAL, n);
6006 Set_Node_Length(ret, RExC_parse - parse_start + 1);
6007 Set_Node_Offset(ret, parse_start);
6010 case '(': /* (?(?{...})...) and (?(?=...)...) */
6013 if (RExC_parse[0] == '?') { /* (?(?...)) */
6014 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
6015 || RExC_parse[1] == '<'
6016 || RExC_parse[1] == '{') { /* Lookahead or eval. */
6019 ret = reg_node(pRExC_state, LOGICAL);
6022 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
6026 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
6027 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
6029 char ch = RExC_parse[0] == '<' ? '>' : '\'';
6030 char *name_start= RExC_parse++;
6032 SV *sv_dat=reg_scan_name(pRExC_state,
6033 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6034 if (RExC_parse == name_start || *RExC_parse != ch)
6035 vFAIL2("Sequence (?(%c... not terminated",
6036 (ch == '>' ? '<' : ch));
6039 num = add_data( pRExC_state, 1, "S" );
6040 RExC_rxi->data->data[num]=(void*)sv_dat;
6041 SvREFCNT_inc_simple_void(sv_dat);
6043 ret = reganode(pRExC_state,NGROUPP,num);
6044 goto insert_if_check_paren;
6046 else if (RExC_parse[0] == 'D' &&
6047 RExC_parse[1] == 'E' &&
6048 RExC_parse[2] == 'F' &&
6049 RExC_parse[3] == 'I' &&
6050 RExC_parse[4] == 'N' &&
6051 RExC_parse[5] == 'E')
6053 ret = reganode(pRExC_state,DEFINEP,0);
6056 goto insert_if_check_paren;
6058 else if (RExC_parse[0] == 'R') {
6061 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6062 parno = atoi(RExC_parse++);
6063 while (isDIGIT(*RExC_parse))
6065 } else if (RExC_parse[0] == '&') {
6068 sv_dat = reg_scan_name(pRExC_state,
6069 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6070 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6072 ret = reganode(pRExC_state,INSUBP,parno);
6073 goto insert_if_check_paren;
6075 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6078 parno = atoi(RExC_parse++);
6080 while (isDIGIT(*RExC_parse))
6082 ret = reganode(pRExC_state, GROUPP, parno);
6084 insert_if_check_paren:
6085 if ((c = *nextchar(pRExC_state)) != ')')
6086 vFAIL("Switch condition not recognized");
6088 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
6089 br = regbranch(pRExC_state, &flags, 1,depth+1);
6091 br = reganode(pRExC_state, LONGJMP, 0);
6093 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
6094 c = *nextchar(pRExC_state);
6099 vFAIL("(?(DEFINE)....) does not allow branches");
6100 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6101 regbranch(pRExC_state, &flags, 1,depth+1);
6102 REGTAIL(pRExC_state, ret, lastbr);
6105 c = *nextchar(pRExC_state);
6110 vFAIL("Switch (?(condition)... contains too many branches");
6111 ender = reg_node(pRExC_state, TAIL);
6112 REGTAIL(pRExC_state, br, ender);
6114 REGTAIL(pRExC_state, lastbr, ender);
6115 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6118 REGTAIL(pRExC_state, ret, ender);
6119 RExC_size++; /* XXX WHY do we need this?!!
6120 For large programs it seems to be required
6121 but I can't figure out why. -- dmq*/
6125 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6129 RExC_parse--; /* for vFAIL to print correctly */
6130 vFAIL("Sequence (? incomplete");
6132 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
6134 has_use_defaults = TRUE;
6135 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
6136 RExC_flags &= ~(RXf_PMf_LOCALE|RXf_PMf_UNICODE);
6140 parse_flags: /* (?i) */
6142 U32 posflags = 0, negflags = 0;
6143 U32 *flagsp = &posflags;
6144 bool has_charset_modifier = 0;
6146 while (*RExC_parse) {
6147 /* && strchr("iogcmsx", *RExC_parse) */
6148 /* (?g), (?gc) and (?o) are useless here
6149 and must be globally applied -- japhy */
6150 switch (*RExC_parse) {
6151 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6152 case LOCALE_PAT_MOD:
6153 if (has_charset_modifier || flagsp == &negflags) {
6154 goto fail_modifiers;
6156 *flagsp &= ~RXf_PMf_UNICODE;
6157 *flagsp |= RXf_PMf_LOCALE;
6158 has_charset_modifier = 1;
6160 case UNICODE_PAT_MOD:
6161 if (has_charset_modifier || flagsp == &negflags) {
6162 goto fail_modifiers;
6164 *flagsp &= ~RXf_PMf_LOCALE;
6165 *flagsp |= RXf_PMf_UNICODE;
6166 has_charset_modifier = 1;
6169 if (has_use_defaults
6170 || has_charset_modifier
6171 || flagsp == &negflags)
6173 goto fail_modifiers;
6175 *flagsp &= ~(RXf_PMf_LOCALE|RXf_PMf_UNICODE);
6176 has_charset_modifier = 1;
6178 case ONCE_PAT_MOD: /* 'o' */
6179 case GLOBAL_PAT_MOD: /* 'g' */
6180 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6181 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6182 if (! (wastedflags & wflagbit) ) {
6183 wastedflags |= wflagbit;
6186 "Useless (%s%c) - %suse /%c modifier",
6187 flagsp == &negflags ? "?-" : "?",
6189 flagsp == &negflags ? "don't " : "",
6196 case CONTINUE_PAT_MOD: /* 'c' */
6197 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6198 if (! (wastedflags & WASTED_C) ) {
6199 wastedflags |= WASTED_GC;
6202 "Useless (%sc) - %suse /gc modifier",
6203 flagsp == &negflags ? "?-" : "?",
6204 flagsp == &negflags ? "don't " : ""
6209 case KEEPCOPY_PAT_MOD: /* 'p' */
6210 if (flagsp == &negflags) {
6212 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
6214 *flagsp |= RXf_PMf_KEEPCOPY;
6218 /* A flag is a default iff it is following a minus, so
6219 * if there is a minus, it means will be trying to
6220 * re-specify a default which is an error */
6221 if (has_use_defaults || flagsp == &negflags) {
6224 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6228 wastedflags = 0; /* reset so (?g-c) warns twice */
6234 RExC_flags |= posflags;
6235 RExC_flags &= ~negflags;
6237 oregflags |= posflags;
6238 oregflags &= ~negflags;
6240 nextchar(pRExC_state);
6251 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6256 }} /* one for the default block, one for the switch */
6263 ret = reganode(pRExC_state, OPEN, parno);
6266 RExC_nestroot = parno;
6267 if (RExC_seen & REG_SEEN_RECURSE
6268 && !RExC_open_parens[parno-1])
6270 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6271 "Setting open paren #%"IVdf" to %d\n",
6272 (IV)parno, REG_NODE_NUM(ret)));
6273 RExC_open_parens[parno-1]= ret;
6276 Set_Node_Length(ret, 1); /* MJD */
6277 Set_Node_Offset(ret, RExC_parse); /* MJD */
6285 /* Pick up the branches, linking them together. */
6286 parse_start = RExC_parse; /* MJD */
6287 br = regbranch(pRExC_state, &flags, 1,depth+1);
6290 if (RExC_npar > after_freeze)
6291 after_freeze = RExC_npar;
6292 RExC_npar = freeze_paren;
6295 /* branch_len = (paren != 0); */
6299 if (*RExC_parse == '|') {
6300 if (!SIZE_ONLY && RExC_extralen) {
6301 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6304 reginsert(pRExC_state, BRANCH, br, depth+1);
6305 Set_Node_Length(br, paren != 0);
6306 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6310 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6312 else if (paren == ':') {
6313 *flagp |= flags&SIMPLE;
6315 if (is_open) { /* Starts with OPEN. */
6316 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6318 else if (paren != '?') /* Not Conditional */
6320 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6322 while (*RExC_parse == '|') {
6323 if (!SIZE_ONLY && RExC_extralen) {
6324 ender = reganode(pRExC_state, LONGJMP,0);
6325 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6328 RExC_extralen += 2; /* Account for LONGJMP. */
6329 nextchar(pRExC_state);
6331 if (RExC_npar > after_freeze)
6332 after_freeze = RExC_npar;
6333 RExC_npar = freeze_paren;
6335 br = regbranch(pRExC_state, &flags, 0, depth+1);
6339 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6341 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6344 if (have_branch || paren != ':') {
6345 /* Make a closing node, and hook it on the end. */
6348 ender = reg_node(pRExC_state, TAIL);
6351 ender = reganode(pRExC_state, CLOSE, parno);
6352 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6353 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6354 "Setting close paren #%"IVdf" to %d\n",
6355 (IV)parno, REG_NODE_NUM(ender)));
6356 RExC_close_parens[parno-1]= ender;
6357 if (RExC_nestroot == parno)
6360 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6361 Set_Node_Length(ender,1); /* MJD */
6367 *flagp &= ~HASWIDTH;
6370 ender = reg_node(pRExC_state, SUCCEED);
6373 ender = reg_node(pRExC_state, END);
6375 assert(!RExC_opend); /* there can only be one! */
6380 REGTAIL(pRExC_state, lastbr, ender);
6382 if (have_branch && !SIZE_ONLY) {
6384 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6386 /* Hook the tails of the branches to the closing node. */
6387 for (br = ret; br; br = regnext(br)) {
6388 const U8 op = PL_regkind[OP(br)];
6390 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6392 else if (op == BRANCHJ) {
6393 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6401 static const char parens[] = "=!<,>";
6403 if (paren && (p = strchr(parens, paren))) {
6404 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6405 int flag = (p - parens) > 1;
6408 node = SUSPEND, flag = 0;
6409 reginsert(pRExC_state, node,ret, depth+1);
6410 Set_Node_Cur_Length(ret);
6411 Set_Node_Offset(ret, parse_start + 1);
6413 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6417 /* Check for proper termination. */
6419 RExC_flags = oregflags;
6420 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6421 RExC_parse = oregcomp_parse;
6422 vFAIL("Unmatched (");
6425 else if (!paren && RExC_parse < RExC_end) {
6426 if (*RExC_parse == ')') {
6428 vFAIL("Unmatched )");
6431 FAIL("Junk on end of regexp"); /* "Can't happen". */
6435 RExC_npar = after_freeze;
6440 - regbranch - one alternative of an | operator
6442 * Implements the concatenation operator.
6445 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6448 register regnode *ret;
6449 register regnode *chain = NULL;
6450 register regnode *latest;
6451 I32 flags = 0, c = 0;
6452 GET_RE_DEBUG_FLAGS_DECL;
6454 PERL_ARGS_ASSERT_REGBRANCH;
6456 DEBUG_PARSE("brnc");
6461 if (!SIZE_ONLY && RExC_extralen)
6462 ret = reganode(pRExC_state, BRANCHJ,0);
6464 ret = reg_node(pRExC_state, BRANCH);
6465 Set_Node_Length(ret, 1);
6469 if (!first && SIZE_ONLY)
6470 RExC_extralen += 1; /* BRANCHJ */
6472 *flagp = WORST; /* Tentatively. */
6475 nextchar(pRExC_state);
6476 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6478 latest = regpiece(pRExC_state, &flags,depth+1);
6479 if (latest == NULL) {
6480 if (flags & TRYAGAIN)
6484 else if (ret == NULL)
6486 *flagp |= flags&(HASWIDTH|POSTPONED);
6487 if (chain == NULL) /* First piece. */
6488 *flagp |= flags&SPSTART;
6491 REGTAIL(pRExC_state, chain, latest);
6496 if (chain == NULL) { /* Loop ran zero times. */
6497 chain = reg_node(pRExC_state, NOTHING);
6502 *flagp |= flags&SIMPLE;
6509 - regpiece - something followed by possible [*+?]
6511 * Note that the branching code sequences used for ? and the general cases
6512 * of * and + are somewhat optimized: they use the same NOTHING node as
6513 * both the endmarker for their branch list and the body of the last branch.
6514 * It might seem that this node could be dispensed with entirely, but the
6515 * endmarker role is not redundant.
6518 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6521 register regnode *ret;
6523 register char *next;
6525 const char * const origparse = RExC_parse;
6527 I32 max = REG_INFTY;
6529 const char *maxpos = NULL;
6530 GET_RE_DEBUG_FLAGS_DECL;
6532 PERL_ARGS_ASSERT_REGPIECE;
6534 DEBUG_PARSE("piec");
6536 ret = regatom(pRExC_state, &flags,depth+1);
6538 if (flags & TRYAGAIN)
6545 if (op == '{' && regcurly(RExC_parse)) {
6547 parse_start = RExC_parse; /* MJD */
6548 next = RExC_parse + 1;
6549 while (isDIGIT(*next) || *next == ',') {
6558 if (*next == '}') { /* got one */
6562 min = atoi(RExC_parse);
6566 maxpos = RExC_parse;
6568 if (!max && *maxpos != '0')
6569 max = REG_INFTY; /* meaning "infinity" */
6570 else if (max >= REG_INFTY)
6571 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6573 nextchar(pRExC_state);
6576 if ((flags&SIMPLE)) {
6577 RExC_naughty += 2 + RExC_naughty / 2;
6578 reginsert(pRExC_state, CURLY, ret, depth+1);
6579 Set_Node_Offset(ret, parse_start+1); /* MJD */
6580 Set_Node_Cur_Length(ret);
6583 regnode * const w = reg_node(pRExC_state, WHILEM);
6586 REGTAIL(pRExC_state, ret, w);
6587 if (!SIZE_ONLY && RExC_extralen) {
6588 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6589 reginsert(pRExC_state, NOTHING,ret, depth+1);
6590 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6592 reginsert(pRExC_state, CURLYX,ret, depth+1);
6594 Set_Node_Offset(ret, parse_start+1);
6595 Set_Node_Length(ret,
6596 op == '{' ? (RExC_parse - parse_start) : 1);
6598 if (!SIZE_ONLY && RExC_extralen)
6599 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6600 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6602 RExC_whilem_seen++, RExC_extralen += 3;
6603 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6612 vFAIL("Can't do {n,m} with n > m");
6614 ARG1_SET(ret, (U16)min);
6615 ARG2_SET(ret, (U16)max);
6627 #if 0 /* Now runtime fix should be reliable. */
6629 /* if this is reinstated, don't forget to put this back into perldiag:
6631 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6633 (F) The part of the regexp subject to either the * or + quantifier
6634 could match an empty string. The {#} shows in the regular
6635 expression about where the problem was discovered.
6639 if (!(flags&HASWIDTH) && op != '?')
6640 vFAIL("Regexp *+ operand could be empty");
6643 parse_start = RExC_parse;
6644 nextchar(pRExC_state);
6646 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6648 if (op == '*' && (flags&SIMPLE)) {
6649 reginsert(pRExC_state, STAR, ret, depth+1);
6653 else if (op == '*') {
6657 else if (op == '+' && (flags&SIMPLE)) {
6658 reginsert(pRExC_state, PLUS, ret, depth+1);
6662 else if (op == '+') {
6666 else if (op == '?') {
6671 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
6672 ckWARN3reg(RExC_parse,
6673 "%.*s matches null string many times",
6674 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6678 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6679 nextchar(pRExC_state);
6680 reginsert(pRExC_state, MINMOD, ret, depth+1);
6681 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6683 #ifndef REG_ALLOW_MINMOD_SUSPEND
6686 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6688 nextchar(pRExC_state);
6689 ender = reg_node(pRExC_state, SUCCEED);
6690 REGTAIL(pRExC_state, ret, ender);
6691 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6693 ender = reg_node(pRExC_state, TAIL);
6694 REGTAIL(pRExC_state, ret, ender);
6698 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6700 vFAIL("Nested quantifiers");
6707 /* reg_namedseq(pRExC_state,UVp)
6709 This is expected to be called by a parser routine that has
6710 recognized '\N' and needs to handle the rest. RExC_parse is
6711 expected to point at the first char following the N at the time
6714 The \N may be inside (indicated by valuep not being NULL) or outside a
6717 \N may begin either a named sequence, or if outside a character class, mean
6718 to match a non-newline. For non single-quoted regexes, the tokenizer has
6719 attempted to decide which, and in the case of a named sequence converted it
6720 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
6721 where c1... are the characters in the sequence. For single-quoted regexes,
6722 the tokenizer passes the \N sequence through unchanged; this code will not
6723 attempt to determine this nor expand those. The net effect is that if the
6724 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
6725 signals that this \N occurrence means to match a non-newline.
6727 Only the \N{U+...} form should occur in a character class, for the same
6728 reason that '.' inside a character class means to just match a period: it
6729 just doesn't make sense.
6731 If valuep is non-null then it is assumed that we are parsing inside
6732 of a charclass definition and the first codepoint in the resolved
6733 string is returned via *valuep and the routine will return NULL.
6734 In this mode if a multichar string is returned from the charnames
6735 handler, a warning will be issued, and only the first char in the
6736 sequence will be examined. If the string returned is zero length
6737 then the value of *valuep is undefined and NON-NULL will
6738 be returned to indicate failure. (This will NOT be a valid pointer
6741 If valuep is null then it is assumed that we are parsing normal text and a
6742 new EXACT node is inserted into the program containing the resolved string,
6743 and a pointer to the new node is returned. But if the string is zero length
6744 a NOTHING node is emitted instead.
6746 On success RExC_parse is set to the char following the endbrace.
6747 Parsing failures will generate a fatal error via vFAIL(...)
6750 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp)
6752 char * endbrace; /* '}' following the name */
6753 regnode *ret = NULL;
6755 char* parse_start = RExC_parse - 2; /* points to the '\N' */
6759 GET_RE_DEBUG_FLAGS_DECL;
6761 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6765 /* The [^\n] meaning of \N ignores spaces and comments under the /x
6766 * modifier. The other meaning does not */
6767 p = (RExC_flags & RXf_PMf_EXTENDED)
6768 ? regwhite( pRExC_state, RExC_parse )
6771 /* Disambiguate between \N meaning a named character versus \N meaning
6772 * [^\n]. The former is assumed when it can't be the latter. */
6773 if (*p != '{' || regcurly(p)) {
6776 /* no bare \N in a charclass */
6777 vFAIL("\\N in a character class must be a named character: \\N{...}");
6779 nextchar(pRExC_state);
6780 ret = reg_node(pRExC_state, REG_ANY);
6781 *flagp |= HASWIDTH|SIMPLE;
6784 Set_Node_Length(ret, 1); /* MJD */
6788 /* Here, we have decided it should be a named sequence */
6790 /* The test above made sure that the next real character is a '{', but
6791 * under the /x modifier, it could be separated by space (or a comment and
6792 * \n) and this is not allowed (for consistency with \x{...} and the
6793 * tokenizer handling of \N{NAME}). */
6794 if (*RExC_parse != '{') {
6795 vFAIL("Missing braces on \\N{}");
6798 RExC_parse++; /* Skip past the '{' */
6800 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
6801 || ! (endbrace == RExC_parse /* nothing between the {} */
6802 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
6803 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
6805 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
6806 vFAIL("\\N{NAME} must be resolved by the lexer");
6809 if (endbrace == RExC_parse) { /* empty: \N{} */
6811 RExC_parse = endbrace + 1;
6812 return reg_node(pRExC_state,NOTHING);
6816 ckWARNreg(RExC_parse,
6817 "Ignoring zero length \\N{} in character class"
6819 RExC_parse = endbrace + 1;
6822 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
6825 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
6826 RExC_parse += 2; /* Skip past the 'U+' */
6828 if (valuep) { /* In a bracketed char class */
6829 /* We only pay attention to the first char of
6830 multichar strings being returned. I kinda wonder
6831 if this makes sense as it does change the behaviour
6832 from earlier versions, OTOH that behaviour was broken
6833 as well. XXX Solution is to recharacterize as
6834 [rest-of-class]|multi1|multi2... */
6836 STRLEN length_of_hex;
6837 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6838 | PERL_SCAN_DISALLOW_PREFIX
6839 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6841 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
6842 if (endchar < endbrace) {
6843 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
6846 length_of_hex = (STRLEN)(endchar - RExC_parse);
6847 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
6849 /* The tokenizer should have guaranteed validity, but it's possible to
6850 * bypass it by using single quoting, so check */
6851 if (length_of_hex == 0
6852 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6854 RExC_parse += length_of_hex; /* Includes all the valid */
6855 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6856 ? UTF8SKIP(RExC_parse)
6858 /* Guard against malformed utf8 */
6859 if (RExC_parse >= endchar) RExC_parse = endchar;
6860 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6863 RExC_parse = endbrace + 1;
6864 if (endchar == endbrace) return NULL;
6866 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
6868 else { /* Not a char class */
6869 char *s; /* String to put in generated EXACT node */
6870 STRLEN len = 0; /* Its current byte length */
6871 char *endchar; /* Points to '.' or '}' ending cur char in the input
6874 ret = reg_node(pRExC_state,
6875 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6878 /* Exact nodes can hold only a U8 length's of text = 255. Loop through
6879 * the input which is of the form now 'c1.c2.c3...}' until find the
6880 * ending brace or exceed length 255. The characters that exceed this
6881 * limit are dropped. The limit could be relaxed should it become
6882 * desirable by reparsing this as (?:\N{NAME}), so could generate
6883 * multiple EXACT nodes, as is done for just regular input. But this
6884 * is primarily a named character, and not intended to be a huge long
6885 * string, so 255 bytes should be good enough */
6887 STRLEN length_of_hex;
6888 I32 grok_flags = PERL_SCAN_ALLOW_UNDERSCORES
6889 | PERL_SCAN_DISALLOW_PREFIX
6890 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6891 UV cp; /* Ord of current character */
6893 /* Code points are separated by dots. If none, there is only one
6894 * code point, and is terminated by the brace */
6895 endchar = RExC_parse + strcspn(RExC_parse, ".}");
6897 /* The values are Unicode even on EBCDIC machines */
6898 length_of_hex = (STRLEN)(endchar - RExC_parse);
6899 cp = grok_hex(RExC_parse, &length_of_hex, &grok_flags, NULL);
6900 if ( length_of_hex == 0
6901 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6903 RExC_parse += length_of_hex; /* Includes all the valid */
6904 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6905 ? UTF8SKIP(RExC_parse)
6907 /* Guard against malformed utf8 */
6908 if (RExC_parse >= endchar) RExC_parse = endchar;
6909 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6912 if (! FOLD) { /* Not folding, just append to the string */
6915 /* Quit before adding this character if would exceed limit */
6916 if (len + UNISKIP(cp) > U8_MAX) break;
6918 unilen = reguni(pRExC_state, cp, s);
6923 } else { /* Folding, output the folded equivalent */
6924 STRLEN foldlen,numlen;
6925 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6926 cp = toFOLD_uni(cp, tmpbuf, &foldlen);
6928 /* Quit before exceeding size limit */
6929 if (len + foldlen > U8_MAX) break;
6931 for (foldbuf = tmpbuf;
6935 cp = utf8_to_uvchr(foldbuf, &numlen);
6937 const STRLEN unilen = reguni(pRExC_state, cp, s);
6940 /* In EBCDIC the numlen and unilen can differ. */
6942 if (numlen >= foldlen)
6946 break; /* "Can't happen." */
6950 /* Point to the beginning of the next character in the sequence. */
6951 RExC_parse = endchar + 1;
6953 /* Quit if no more characters */
6954 if (RExC_parse >= endbrace) break;
6959 if (RExC_parse < endbrace) {
6960 ckWARNreg(RExC_parse - 1,
6961 "Using just the first characters returned by \\N{}");
6964 RExC_size += STR_SZ(len);
6967 RExC_emit += STR_SZ(len);
6970 RExC_parse = endbrace + 1;
6972 *flagp |= HASWIDTH; /* Not SIMPLE, as that causes the engine to fail
6973 with malformed in t/re/pat_advanced.t */
6975 Set_Node_Cur_Length(ret); /* MJD */
6976 nextchar(pRExC_state);
6986 * It returns the code point in utf8 for the value in *encp.
6987 * value: a code value in the source encoding
6988 * encp: a pointer to an Encode object
6990 * If the result from Encode is not a single character,
6991 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6994 S_reg_recode(pTHX_ const char value, SV **encp)
6997 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
6998 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6999 const STRLEN newlen = SvCUR(sv);
7000 UV uv = UNICODE_REPLACEMENT;
7002 PERL_ARGS_ASSERT_REG_RECODE;
7006 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
7009 if (!newlen || numlen != newlen) {
7010 uv = UNICODE_REPLACEMENT;
7018 - regatom - the lowest level
7020 Try to identify anything special at the start of the pattern. If there
7021 is, then handle it as required. This may involve generating a single regop,
7022 such as for an assertion; or it may involve recursing, such as to
7023 handle a () structure.
7025 If the string doesn't start with something special then we gobble up
7026 as much literal text as we can.
7028 Once we have been able to handle whatever type of thing started the
7029 sequence, we return.
7031 Note: we have to be careful with escapes, as they can be both literal
7032 and special, and in the case of \10 and friends can either, depending
7033 on context. Specifically there are two seperate switches for handling
7034 escape sequences, with the one for handling literal escapes requiring
7035 a dummy entry for all of the special escapes that are actually handled
7040 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7043 register regnode *ret = NULL;
7045 char *parse_start = RExC_parse;
7046 GET_RE_DEBUG_FLAGS_DECL;
7047 DEBUG_PARSE("atom");
7048 *flagp = WORST; /* Tentatively. */
7050 PERL_ARGS_ASSERT_REGATOM;
7053 switch ((U8)*RExC_parse) {
7055 RExC_seen_zerolen++;
7056 nextchar(pRExC_state);
7057 if (RExC_flags & RXf_PMf_MULTILINE)
7058 ret = reg_node(pRExC_state, MBOL);
7059 else if (RExC_flags & RXf_PMf_SINGLELINE)
7060 ret = reg_node(pRExC_state, SBOL);
7062 ret = reg_node(pRExC_state, BOL);
7063 Set_Node_Length(ret, 1); /* MJD */
7066 nextchar(pRExC_state);
7068 RExC_seen_zerolen++;
7069 if (RExC_flags & RXf_PMf_MULTILINE)
7070 ret = reg_node(pRExC_state, MEOL);
7071 else if (RExC_flags & RXf_PMf_SINGLELINE)
7072 ret = reg_node(pRExC_state, SEOL);
7074 ret = reg_node(pRExC_state, EOL);
7075 Set_Node_Length(ret, 1); /* MJD */
7078 nextchar(pRExC_state);
7079 if (RExC_flags & RXf_PMf_SINGLELINE)
7080 ret = reg_node(pRExC_state, SANY);
7082 ret = reg_node(pRExC_state, REG_ANY);
7083 *flagp |= HASWIDTH|SIMPLE;
7085 Set_Node_Length(ret, 1); /* MJD */
7089 char * const oregcomp_parse = ++RExC_parse;
7090 ret = regclass(pRExC_state,depth+1);
7091 if (*RExC_parse != ']') {
7092 RExC_parse = oregcomp_parse;
7093 vFAIL("Unmatched [");
7095 nextchar(pRExC_state);
7096 *flagp |= HASWIDTH|SIMPLE;
7097 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
7101 nextchar(pRExC_state);
7102 ret = reg(pRExC_state, 1, &flags,depth+1);
7104 if (flags & TRYAGAIN) {
7105 if (RExC_parse == RExC_end) {
7106 /* Make parent create an empty node if needed. */
7114 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7118 if (flags & TRYAGAIN) {
7122 vFAIL("Internal urp");
7123 /* Supposed to be caught earlier. */
7126 if (!regcurly(RExC_parse)) {
7135 vFAIL("Quantifier follows nothing");
7143 len=0; /* silence a spurious compiler warning */
7144 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
7145 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
7146 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
7147 ret = reganode(pRExC_state, FOLDCHAR, cp);
7148 Set_Node_Length(ret, 1); /* MJD */
7149 nextchar(pRExC_state); /* kill whitespace under /x */
7157 This switch handles escape sequences that resolve to some kind
7158 of special regop and not to literal text. Escape sequnces that
7159 resolve to literal text are handled below in the switch marked
7162 Every entry in this switch *must* have a corresponding entry
7163 in the literal escape switch. However, the opposite is not
7164 required, as the default for this switch is to jump to the
7165 literal text handling code.
7167 switch ((U8)*++RExC_parse) {
7172 /* Special Escapes */
7174 RExC_seen_zerolen++;
7175 ret = reg_node(pRExC_state, SBOL);
7177 goto finish_meta_pat;
7179 ret = reg_node(pRExC_state, GPOS);
7180 RExC_seen |= REG_SEEN_GPOS;
7182 goto finish_meta_pat;
7184 RExC_seen_zerolen++;
7185 ret = reg_node(pRExC_state, KEEPS);
7187 /* XXX:dmq : disabling in-place substitution seems to
7188 * be necessary here to avoid cases of memory corruption, as
7189 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7191 RExC_seen |= REG_SEEN_LOOKBEHIND;
7192 goto finish_meta_pat;
7194 ret = reg_node(pRExC_state, SEOL);
7196 RExC_seen_zerolen++; /* Do not optimize RE away */
7197 goto finish_meta_pat;
7199 ret = reg_node(pRExC_state, EOS);
7201 RExC_seen_zerolen++; /* Do not optimize RE away */
7202 goto finish_meta_pat;
7204 ret = reg_node(pRExC_state, CANY);
7205 RExC_seen |= REG_SEEN_CANY;
7206 *flagp |= HASWIDTH|SIMPLE;
7207 goto finish_meta_pat;
7209 ret = reg_node(pRExC_state, CLUMP);
7211 goto finish_meta_pat;
7213 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
7214 *flagp |= HASWIDTH|SIMPLE;
7215 goto finish_meta_pat;
7217 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
7218 *flagp |= HASWIDTH|SIMPLE;
7219 goto finish_meta_pat;
7221 RExC_seen_zerolen++;
7222 RExC_seen |= REG_SEEN_LOOKBEHIND;
7223 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
7225 goto finish_meta_pat;
7227 RExC_seen_zerolen++;
7228 RExC_seen |= REG_SEEN_LOOKBEHIND;
7229 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
7231 goto finish_meta_pat;
7233 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
7234 *flagp |= HASWIDTH|SIMPLE;
7235 goto finish_meta_pat;
7237 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
7238 *flagp |= HASWIDTH|SIMPLE;
7239 goto finish_meta_pat;
7241 ret = reg_node(pRExC_state, DIGIT);
7242 *flagp |= HASWIDTH|SIMPLE;
7243 goto finish_meta_pat;
7245 ret = reg_node(pRExC_state, NDIGIT);
7246 *flagp |= HASWIDTH|SIMPLE;
7247 goto finish_meta_pat;
7249 ret = reg_node(pRExC_state, LNBREAK);
7250 *flagp |= HASWIDTH|SIMPLE;
7251 goto finish_meta_pat;
7253 ret = reg_node(pRExC_state, HORIZWS);
7254 *flagp |= HASWIDTH|SIMPLE;
7255 goto finish_meta_pat;
7257 ret = reg_node(pRExC_state, NHORIZWS);
7258 *flagp |= HASWIDTH|SIMPLE;
7259 goto finish_meta_pat;
7261 ret = reg_node(pRExC_state, VERTWS);
7262 *flagp |= HASWIDTH|SIMPLE;
7263 goto finish_meta_pat;
7265 ret = reg_node(pRExC_state, NVERTWS);
7266 *flagp |= HASWIDTH|SIMPLE;
7268 nextchar(pRExC_state);
7269 Set_Node_Length(ret, 2); /* MJD */
7274 char* const oldregxend = RExC_end;
7276 char* parse_start = RExC_parse - 2;
7279 if (RExC_parse[1] == '{') {
7280 /* a lovely hack--pretend we saw [\pX] instead */
7281 RExC_end = strchr(RExC_parse, '}');
7283 const U8 c = (U8)*RExC_parse;
7285 RExC_end = oldregxend;
7286 vFAIL2("Missing right brace on \\%c{}", c);
7291 RExC_end = RExC_parse + 2;
7292 if (RExC_end > oldregxend)
7293 RExC_end = oldregxend;
7297 ret = regclass(pRExC_state,depth+1);
7299 RExC_end = oldregxend;
7302 Set_Node_Offset(ret, parse_start + 2);
7303 Set_Node_Cur_Length(ret);
7304 nextchar(pRExC_state);
7305 *flagp |= HASWIDTH|SIMPLE;
7309 /* Handle \N and \N{NAME} here and not below because it can be
7310 multicharacter. join_exact() will join them up later on.
7311 Also this makes sure that things like /\N{BLAH}+/ and
7312 \N{BLAH} being multi char Just Happen. dmq*/
7314 ret= reg_namedseq(pRExC_state, NULL, flagp);
7316 case 'k': /* Handle \k<NAME> and \k'NAME' */
7319 char ch= RExC_parse[1];
7320 if (ch != '<' && ch != '\'' && ch != '{') {
7322 vFAIL2("Sequence %.2s... not terminated",parse_start);
7324 /* this pretty much dupes the code for (?P=...) in reg(), if
7325 you change this make sure you change that */
7326 char* name_start = (RExC_parse += 2);
7328 SV *sv_dat = reg_scan_name(pRExC_state,
7329 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7330 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7331 if (RExC_parse == name_start || *RExC_parse != ch)
7332 vFAIL2("Sequence %.3s... not terminated",parse_start);
7335 num = add_data( pRExC_state, 1, "S" );
7336 RExC_rxi->data->data[num]=(void*)sv_dat;
7337 SvREFCNT_inc_simple_void(sv_dat);
7341 ret = reganode(pRExC_state,
7342 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7346 /* override incorrect value set in reganode MJD */
7347 Set_Node_Offset(ret, parse_start+1);
7348 Set_Node_Cur_Length(ret); /* MJD */
7349 nextchar(pRExC_state);
7355 case '1': case '2': case '3': case '4':
7356 case '5': case '6': case '7': case '8': case '9':
7359 bool isg = *RExC_parse == 'g';
7364 if (*RExC_parse == '{') {
7368 if (*RExC_parse == '-') {
7372 if (hasbrace && !isDIGIT(*RExC_parse)) {
7373 if (isrel) RExC_parse--;
7375 goto parse_named_seq;
7377 num = atoi(RExC_parse);
7378 if (isg && num == 0)
7379 vFAIL("Reference to invalid group 0");
7381 num = RExC_npar - num;
7383 vFAIL("Reference to nonexistent or unclosed group");
7385 if (!isg && num > 9 && num >= RExC_npar)
7388 char * const parse_start = RExC_parse - 1; /* MJD */
7389 while (isDIGIT(*RExC_parse))
7391 if (parse_start == RExC_parse - 1)
7392 vFAIL("Unterminated \\g... pattern");
7394 if (*RExC_parse != '}')
7395 vFAIL("Unterminated \\g{...} pattern");
7399 if (num > (I32)RExC_rx->nparens)
7400 vFAIL("Reference to nonexistent group");
7403 ret = reganode(pRExC_state,
7404 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7408 /* override incorrect value set in reganode MJD */
7409 Set_Node_Offset(ret, parse_start+1);
7410 Set_Node_Cur_Length(ret); /* MJD */
7412 nextchar(pRExC_state);
7417 if (RExC_parse >= RExC_end)
7418 FAIL("Trailing \\");
7421 /* Do not generate "unrecognized" warnings here, we fall
7422 back into the quick-grab loop below */
7429 if (RExC_flags & RXf_PMf_EXTENDED) {
7430 if ( reg_skipcomment( pRExC_state ) )
7437 register STRLEN len;
7442 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7444 parse_start = RExC_parse - 1;
7450 ret = reg_node(pRExC_state,
7451 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7453 for (len = 0, p = RExC_parse - 1;
7454 len < 127 && p < RExC_end;
7457 char * const oldp = p;
7459 if (RExC_flags & RXf_PMf_EXTENDED)
7460 p = regwhite( pRExC_state, p );
7465 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7466 goto normal_default;
7476 /* Literal Escapes Switch
7478 This switch is meant to handle escape sequences that
7479 resolve to a literal character.
7481 Every escape sequence that represents something
7482 else, like an assertion or a char class, is handled
7483 in the switch marked 'Special Escapes' above in this
7484 routine, but also has an entry here as anything that
7485 isn't explicitly mentioned here will be treated as
7486 an unescaped equivalent literal.
7490 /* These are all the special escapes. */
7494 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7495 goto normal_default;
7496 case 'A': /* Start assertion */
7497 case 'b': case 'B': /* Word-boundary assertion*/
7498 case 'C': /* Single char !DANGEROUS! */
7499 case 'd': case 'D': /* digit class */
7500 case 'g': case 'G': /* generic-backref, pos assertion */
7501 case 'h': case 'H': /* HORIZWS */
7502 case 'k': case 'K': /* named backref, keep marker */
7503 case 'N': /* named char sequence */
7504 case 'p': case 'P': /* Unicode property */
7505 case 'R': /* LNBREAK */
7506 case 's': case 'S': /* space class */
7507 case 'v': case 'V': /* VERTWS */
7508 case 'w': case 'W': /* word class */
7509 case 'X': /* eXtended Unicode "combining character sequence" */
7510 case 'z': case 'Z': /* End of line/string assertion */
7514 /* Anything after here is an escape that resolves to a
7515 literal. (Except digits, which may or may not)
7534 ender = ASCII_TO_NATIVE('\033');
7538 ender = ASCII_TO_NATIVE('\007');
7543 STRLEN brace_len = len;
7545 const char* error_msg;
7547 bool valid = grok_bslash_o(p,
7554 RExC_parse = p; /* going to die anyway; point
7555 to exact spot of failure */
7562 if (PL_encoding && ender < 0x100) {
7563 goto recode_encoding;
7572 char* const e = strchr(p, '}');
7576 vFAIL("Missing right brace on \\x{}");
7579 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7580 | PERL_SCAN_DISALLOW_PREFIX;
7581 STRLEN numlen = e - p - 1;
7582 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7589 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7591 ender = grok_hex(p, &numlen, &flags, NULL);
7594 if (PL_encoding && ender < 0x100)
7595 goto recode_encoding;
7599 ender = grok_bslash_c(*p++, SIZE_ONLY);
7601 case '0': case '1': case '2': case '3':case '4':
7602 case '5': case '6': case '7': case '8':case '9':
7604 (isOCTAL(p[1]) && atoi(p) >= RExC_npar))
7606 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
7608 ender = grok_oct(p, &numlen, &flags, NULL);
7618 if (PL_encoding && ender < 0x100)
7619 goto recode_encoding;
7623 SV* enc = PL_encoding;
7624 ender = reg_recode((const char)(U8)ender, &enc);
7625 if (!enc && SIZE_ONLY)
7626 ckWARNreg(p, "Invalid escape in the specified encoding");
7632 FAIL("Trailing \\");
7635 if (!SIZE_ONLY&& isALPHA(*p))
7636 ckWARN2reg(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7637 goto normal_default;
7642 if (UTF8_IS_START(*p) && UTF) {
7644 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7645 &numlen, UTF8_ALLOW_DEFAULT);
7652 if ( RExC_flags & RXf_PMf_EXTENDED)
7653 p = regwhite( pRExC_state, p );
7655 /* Prime the casefolded buffer. */
7656 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7658 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7663 /* Emit all the Unicode characters. */
7665 for (foldbuf = tmpbuf;
7667 foldlen -= numlen) {
7668 ender = utf8_to_uvchr(foldbuf, &numlen);
7670 const STRLEN unilen = reguni(pRExC_state, ender, s);
7673 /* In EBCDIC the numlen
7674 * and unilen can differ. */
7676 if (numlen >= foldlen)
7680 break; /* "Can't happen." */
7684 const STRLEN unilen = reguni(pRExC_state, ender, s);
7693 REGC((char)ender, s++);
7699 /* Emit all the Unicode characters. */
7701 for (foldbuf = tmpbuf;
7703 foldlen -= numlen) {
7704 ender = utf8_to_uvchr(foldbuf, &numlen);
7706 const STRLEN unilen = reguni(pRExC_state, ender, s);
7709 /* In EBCDIC the numlen
7710 * and unilen can differ. */
7712 if (numlen >= foldlen)
7720 const STRLEN unilen = reguni(pRExC_state, ender, s);
7729 REGC((char)ender, s++);
7733 Set_Node_Cur_Length(ret); /* MJD */
7734 nextchar(pRExC_state);
7736 /* len is STRLEN which is unsigned, need to copy to signed */
7739 vFAIL("Internal disaster");
7743 if (len == 1 && UNI_IS_INVARIANT(ender))
7747 RExC_size += STR_SZ(len);
7750 RExC_emit += STR_SZ(len);
7760 S_regwhite( RExC_state_t *pRExC_state, char *p )
7762 const char *e = RExC_end;
7764 PERL_ARGS_ASSERT_REGWHITE;
7769 else if (*p == '#') {
7778 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7786 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7787 Character classes ([:foo:]) can also be negated ([:^foo:]).
7788 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7789 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7790 but trigger failures because they are currently unimplemented. */
7792 #define POSIXCC_DONE(c) ((c) == ':')
7793 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7794 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7797 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7800 I32 namedclass = OOB_NAMEDCLASS;
7802 PERL_ARGS_ASSERT_REGPPOSIXCC;
7804 if (value == '[' && RExC_parse + 1 < RExC_end &&
7805 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7806 POSIXCC(UCHARAT(RExC_parse))) {
7807 const char c = UCHARAT(RExC_parse);
7808 char* const s = RExC_parse++;
7810 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7812 if (RExC_parse == RExC_end)
7813 /* Grandfather lone [:, [=, [. */
7816 const char* const t = RExC_parse++; /* skip over the c */
7819 if (UCHARAT(RExC_parse) == ']') {
7820 const char *posixcc = s + 1;
7821 RExC_parse++; /* skip over the ending ] */
7824 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7825 const I32 skip = t - posixcc;
7827 /* Initially switch on the length of the name. */
7830 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7831 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7834 /* Names all of length 5. */
7835 /* alnum alpha ascii blank cntrl digit graph lower
7836 print punct space upper */
7837 /* Offset 4 gives the best switch position. */
7838 switch (posixcc[4]) {
7840 if (memEQ(posixcc, "alph", 4)) /* alpha */
7841 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7844 if (memEQ(posixcc, "spac", 4)) /* space */
7845 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7848 if (memEQ(posixcc, "grap", 4)) /* graph */
7849 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7852 if (memEQ(posixcc, "asci", 4)) /* ascii */
7853 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7856 if (memEQ(posixcc, "blan", 4)) /* blank */
7857 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7860 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7861 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7864 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7865 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7868 if (memEQ(posixcc, "lowe", 4)) /* lower */
7869 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7870 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7871 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7874 if (memEQ(posixcc, "digi", 4)) /* digit */
7875 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7876 else if (memEQ(posixcc, "prin", 4)) /* print */
7877 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7878 else if (memEQ(posixcc, "punc", 4)) /* punct */
7879 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7884 if (memEQ(posixcc, "xdigit", 6))
7885 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7889 if (namedclass == OOB_NAMEDCLASS)
7890 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7892 assert (posixcc[skip] == ':');
7893 assert (posixcc[skip+1] == ']');
7894 } else if (!SIZE_ONLY) {
7895 /* [[=foo=]] and [[.foo.]] are still future. */
7897 /* adjust RExC_parse so the warning shows after
7899 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7901 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7904 /* Maternal grandfather:
7905 * "[:" ending in ":" but not in ":]" */
7915 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7919 PERL_ARGS_ASSERT_CHECKPOSIXCC;
7921 if (POSIXCC(UCHARAT(RExC_parse))) {
7922 const char *s = RExC_parse;
7923 const char c = *s++;
7927 if (*s && c == *s && s[1] == ']') {
7929 "POSIX syntax [%c %c] belongs inside character classes",
7932 /* [[=foo=]] and [[.foo.]] are still future. */
7933 if (POSIXCC_NOTYET(c)) {
7934 /* adjust RExC_parse so the error shows after
7936 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7938 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7945 #define _C_C_T_(NAME,TEST,WORD) \
7948 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7950 for (value = 0; value < 256; value++) \
7952 ANYOF_BITMAP_SET(ret, value); \
7957 case ANYOF_N##NAME: \
7959 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7961 for (value = 0; value < 256; value++) \
7963 ANYOF_BITMAP_SET(ret, value); \
7969 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7971 for (value = 0; value < 256; value++) \
7973 ANYOF_BITMAP_SET(ret, value); \
7977 case ANYOF_N##NAME: \
7978 for (value = 0; value < 256; value++) \
7980 ANYOF_BITMAP_SET(ret, value); \
7986 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
7987 so that it is possible to override the option here without having to
7988 rebuild the entire core. as we are required to do if we change regcomp.h
7989 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
7991 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
7992 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
7995 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
7996 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
7998 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
8002 parse a class specification and produce either an ANYOF node that
8003 matches the pattern or if the pattern matches a single char only and
8004 that char is < 256 and we are case insensitive then we produce an
8009 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
8012 register UV nextvalue;
8013 register IV prevvalue = OOB_UNICODE;
8014 register IV range = 0;
8015 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
8016 register regnode *ret;
8019 char *rangebegin = NULL;
8020 bool need_class = 0;
8023 bool optimize_invert = TRUE;
8024 AV* unicode_alternate = NULL;
8026 UV literal_endpoint = 0;
8028 UV stored = 0; /* number of chars stored in the class */
8030 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
8031 case we need to change the emitted regop to an EXACT. */
8032 const char * orig_parse = RExC_parse;
8033 GET_RE_DEBUG_FLAGS_DECL;
8035 PERL_ARGS_ASSERT_REGCLASS;
8037 PERL_UNUSED_ARG(depth);
8040 DEBUG_PARSE("clas");
8042 /* Assume we are going to generate an ANYOF node. */
8043 ret = reganode(pRExC_state, ANYOF, 0);
8046 ANYOF_FLAGS(ret) = 0;
8048 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
8052 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
8056 RExC_size += ANYOF_SKIP;
8057 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
8060 RExC_emit += ANYOF_SKIP;
8062 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
8064 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
8065 ANYOF_BITMAP_ZERO(ret);
8066 listsv = newSVpvs("# comment\n");
8069 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8071 if (!SIZE_ONLY && POSIXCC(nextvalue))
8072 checkposixcc(pRExC_state);
8074 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
8075 if (UCHARAT(RExC_parse) == ']')
8079 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
8083 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
8086 rangebegin = RExC_parse;
8088 value = utf8n_to_uvchr((U8*)RExC_parse,
8089 RExC_end - RExC_parse,
8090 &numlen, UTF8_ALLOW_DEFAULT);
8091 RExC_parse += numlen;
8094 value = UCHARAT(RExC_parse++);
8096 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8097 if (value == '[' && POSIXCC(nextvalue))
8098 namedclass = regpposixcc(pRExC_state, value);
8099 else if (value == '\\') {
8101 value = utf8n_to_uvchr((U8*)RExC_parse,
8102 RExC_end - RExC_parse,
8103 &numlen, UTF8_ALLOW_DEFAULT);
8104 RExC_parse += numlen;
8107 value = UCHARAT(RExC_parse++);
8108 /* Some compilers cannot handle switching on 64-bit integer
8109 * values, therefore value cannot be an UV. Yes, this will
8110 * be a problem later if we want switch on Unicode.
8111 * A similar issue a little bit later when switching on
8112 * namedclass. --jhi */
8113 switch ((I32)value) {
8114 case 'w': namedclass = ANYOF_ALNUM; break;
8115 case 'W': namedclass = ANYOF_NALNUM; break;
8116 case 's': namedclass = ANYOF_SPACE; break;
8117 case 'S': namedclass = ANYOF_NSPACE; break;
8118 case 'd': namedclass = ANYOF_DIGIT; break;
8119 case 'D': namedclass = ANYOF_NDIGIT; break;
8120 case 'v': namedclass = ANYOF_VERTWS; break;
8121 case 'V': namedclass = ANYOF_NVERTWS; break;
8122 case 'h': namedclass = ANYOF_HORIZWS; break;
8123 case 'H': namedclass = ANYOF_NHORIZWS; break;
8124 case 'N': /* Handle \N{NAME} in class */
8126 /* We only pay attention to the first char of
8127 multichar strings being returned. I kinda wonder
8128 if this makes sense as it does change the behaviour
8129 from earlier versions, OTOH that behaviour was broken
8131 UV v; /* value is register so we cant & it /grrr */
8132 if (reg_namedseq(pRExC_state, &v, NULL)) {
8142 if (RExC_parse >= RExC_end)
8143 vFAIL2("Empty \\%c{}", (U8)value);
8144 if (*RExC_parse == '{') {
8145 const U8 c = (U8)value;
8146 e = strchr(RExC_parse++, '}');
8148 vFAIL2("Missing right brace on \\%c{}", c);
8149 while (isSPACE(UCHARAT(RExC_parse)))
8151 if (e == RExC_parse)
8152 vFAIL2("Empty \\%c{}", c);
8154 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
8162 if (UCHARAT(RExC_parse) == '^') {
8165 value = value == 'p' ? 'P' : 'p'; /* toggle */
8166 while (isSPACE(UCHARAT(RExC_parse))) {
8171 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
8172 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
8175 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8176 namedclass = ANYOF_MAX; /* no official name, but it's named */
8179 case 'n': value = '\n'; break;
8180 case 'r': value = '\r'; break;
8181 case 't': value = '\t'; break;
8182 case 'f': value = '\f'; break;
8183 case 'b': value = '\b'; break;
8184 case 'e': value = ASCII_TO_NATIVE('\033');break;
8185 case 'a': value = ASCII_TO_NATIVE('\007');break;
8187 RExC_parse--; /* function expects to be pointed at the 'o' */
8189 const char* error_msg;
8190 bool valid = grok_bslash_o(RExC_parse,
8195 RExC_parse += numlen;
8200 if (PL_encoding && value < 0x100) {
8201 goto recode_encoding;
8205 if (*RExC_parse == '{') {
8206 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8207 | PERL_SCAN_DISALLOW_PREFIX;
8208 char * const e = strchr(RExC_parse++, '}');
8210 vFAIL("Missing right brace on \\x{}");
8212 numlen = e - RExC_parse;
8213 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8217 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8219 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8220 RExC_parse += numlen;
8222 if (PL_encoding && value < 0x100)
8223 goto recode_encoding;
8226 value = grok_bslash_c(*RExC_parse++, SIZE_ONLY);
8228 case '0': case '1': case '2': case '3': case '4':
8229 case '5': case '6': case '7':
8231 /* Take 1-3 octal digits */
8232 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8234 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8235 RExC_parse += numlen;
8236 if (PL_encoding && value < 0x100)
8237 goto recode_encoding;
8242 SV* enc = PL_encoding;
8243 value = reg_recode((const char)(U8)value, &enc);
8244 if (!enc && SIZE_ONLY)
8245 ckWARNreg(RExC_parse,
8246 "Invalid escape in the specified encoding");
8250 /* Allow \_ to not give an error */
8251 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
8252 ckWARN2reg(RExC_parse,
8253 "Unrecognized escape \\%c in character class passed through",
8258 } /* end of \blah */
8264 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8266 if (!SIZE_ONLY && !need_class)
8267 ANYOF_CLASS_ZERO(ret);
8271 /* a bad range like a-\d, a-[:digit:] ? */
8275 RExC_parse >= rangebegin ?
8276 RExC_parse - rangebegin : 0;
8277 ckWARN4reg(RExC_parse,
8278 "False [] range \"%*.*s\"",
8281 if (prevvalue < 256) {
8282 ANYOF_BITMAP_SET(ret, prevvalue);
8283 ANYOF_BITMAP_SET(ret, '-');
8286 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8287 Perl_sv_catpvf(aTHX_ listsv,
8288 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8292 range = 0; /* this was not a true range */
8298 const char *what = NULL;
8301 if (namedclass > OOB_NAMEDCLASS)
8302 optimize_invert = FALSE;
8303 /* Possible truncation here but in some 64-bit environments
8304 * the compiler gets heartburn about switch on 64-bit values.
8305 * A similar issue a little earlier when switching on value.
8307 switch ((I32)namedclass) {
8309 case _C_C_T_(ALNUMC, isALNUMC(value), POSIX_CC_UNI_NAME("Alnum"));
8310 case _C_C_T_(ALPHA, isALPHA(value), POSIX_CC_UNI_NAME("Alpha"));
8311 case _C_C_T_(BLANK, isBLANK(value), POSIX_CC_UNI_NAME("Blank"));
8312 case _C_C_T_(CNTRL, isCNTRL(value), POSIX_CC_UNI_NAME("Cntrl"));
8313 case _C_C_T_(GRAPH, isGRAPH(value), POSIX_CC_UNI_NAME("Graph"));
8314 case _C_C_T_(LOWER, isLOWER(value), POSIX_CC_UNI_NAME("Lower"));
8315 case _C_C_T_(PRINT, isPRINT(value), POSIX_CC_UNI_NAME("Print"));
8316 case _C_C_T_(PSXSPC, isPSXSPC(value), POSIX_CC_UNI_NAME("Space"));
8317 case _C_C_T_(PUNCT, isPUNCT(value), POSIX_CC_UNI_NAME("Punct"));
8318 case _C_C_T_(UPPER, isUPPER(value), POSIX_CC_UNI_NAME("Upper"));
8319 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8320 case _C_C_T_(ALNUM, isALNUM(value), "Word");
8321 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
8323 case _C_C_T_(SPACE, isSPACE(value), "PerlSpace");
8324 case _C_C_T_(ALNUM, isALNUM(value), "PerlWord");
8326 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
8327 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8328 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8331 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8334 for (value = 0; value < 128; value++)
8335 ANYOF_BITMAP_SET(ret, value);
8337 for (value = 0; value < 256; value++) {
8339 ANYOF_BITMAP_SET(ret, value);
8348 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8351 for (value = 128; value < 256; value++)
8352 ANYOF_BITMAP_SET(ret, value);
8354 for (value = 0; value < 256; value++) {
8355 if (!isASCII(value))
8356 ANYOF_BITMAP_SET(ret, value);
8365 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8367 /* consecutive digits assumed */
8368 for (value = '0'; value <= '9'; value++)
8369 ANYOF_BITMAP_SET(ret, value);
8372 what = POSIX_CC_UNI_NAME("Digit");
8376 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8378 /* consecutive digits assumed */
8379 for (value = 0; value < '0'; value++)
8380 ANYOF_BITMAP_SET(ret, value);
8381 for (value = '9' + 1; value < 256; value++)
8382 ANYOF_BITMAP_SET(ret, value);
8385 what = POSIX_CC_UNI_NAME("Digit");
8388 /* this is to handle \p and \P */
8391 vFAIL("Invalid [::] class");
8395 /* Strings such as "+utf8::isWord\n" */
8396 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8399 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8402 } /* end of namedclass \blah */
8405 if (prevvalue > (IV)value) /* b-a */ {
8406 const int w = RExC_parse - rangebegin;
8407 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8408 range = 0; /* not a valid range */
8412 prevvalue = value; /* save the beginning of the range */
8413 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8414 RExC_parse[1] != ']') {
8417 /* a bad range like \w-, [:word:]- ? */
8418 if (namedclass > OOB_NAMEDCLASS) {
8419 if (ckWARN(WARN_REGEXP)) {
8421 RExC_parse >= rangebegin ?
8422 RExC_parse - rangebegin : 0;
8424 "False [] range \"%*.*s\"",
8428 ANYOF_BITMAP_SET(ret, '-');
8430 range = 1; /* yeah, it's a range! */
8431 continue; /* but do it the next time */
8435 /* now is the next time */
8436 /*stored += (value - prevvalue + 1);*/
8438 if (prevvalue < 256) {
8439 const IV ceilvalue = value < 256 ? value : 255;
8442 /* In EBCDIC [\x89-\x91] should include
8443 * the \x8e but [i-j] should not. */
8444 if (literal_endpoint == 2 &&
8445 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8446 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8448 if (isLOWER(prevvalue)) {
8449 for (i = prevvalue; i <= ceilvalue; i++)
8450 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8452 ANYOF_BITMAP_SET(ret, i);
8455 for (i = prevvalue; i <= ceilvalue; i++)
8456 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8458 ANYOF_BITMAP_SET(ret, i);
8464 for (i = prevvalue; i <= ceilvalue; i++) {
8465 if (!ANYOF_BITMAP_TEST(ret,i)) {
8467 ANYOF_BITMAP_SET(ret, i);
8471 if (value > 255 || UTF) {
8472 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8473 const UV natvalue = NATIVE_TO_UNI(value);
8474 stored+=2; /* can't optimize this class */
8475 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8476 if (prevnatvalue < natvalue) { /* what about > ? */
8477 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8478 prevnatvalue, natvalue);
8480 else if (prevnatvalue == natvalue) {
8481 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8483 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8485 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8487 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8488 if (RExC_precomp[0] == ':' &&
8489 RExC_precomp[1] == '[' &&
8490 (f == 0xDF || f == 0x92)) {
8491 f = NATIVE_TO_UNI(f);
8494 /* If folding and foldable and a single
8495 * character, insert also the folded version
8496 * to the charclass. */
8498 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8499 if ((RExC_precomp[0] == ':' &&
8500 RExC_precomp[1] == '[' &&
8502 (value == 0xFB05 || value == 0xFB06))) ?
8503 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8504 foldlen == (STRLEN)UNISKIP(f) )
8506 if (foldlen == (STRLEN)UNISKIP(f))
8508 Perl_sv_catpvf(aTHX_ listsv,
8511 /* Any multicharacter foldings
8512 * require the following transform:
8513 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8514 * where E folds into "pq" and F folds
8515 * into "rst", all other characters
8516 * fold to single characters. We save
8517 * away these multicharacter foldings,
8518 * to be later saved as part of the
8519 * additional "s" data. */
8522 if (!unicode_alternate)
8523 unicode_alternate = newAV();
8524 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8526 av_push(unicode_alternate, sv);
8530 /* If folding and the value is one of the Greek
8531 * sigmas insert a few more sigmas to make the
8532 * folding rules of the sigmas to work right.
8533 * Note that not all the possible combinations
8534 * are handled here: some of them are handled
8535 * by the standard folding rules, and some of
8536 * them (literal or EXACTF cases) are handled
8537 * during runtime in regexec.c:S_find_byclass(). */
8538 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8539 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8540 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8541 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8542 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8544 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8545 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8546 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8551 literal_endpoint = 0;
8555 range = 0; /* this range (if it was one) is done now */
8559 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8561 RExC_size += ANYOF_CLASS_ADD_SKIP;
8563 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8569 /****** !SIZE_ONLY AFTER HERE *********/
8571 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8572 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8574 /* optimize single char class to an EXACT node
8575 but *only* when its not a UTF/high char */
8576 const char * cur_parse= RExC_parse;
8577 RExC_emit = (regnode *)orig_emit;
8578 RExC_parse = (char *)orig_parse;
8579 ret = reg_node(pRExC_state,
8580 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8581 RExC_parse = (char *)cur_parse;
8582 *STRING(ret)= (char)value;
8584 RExC_emit += STR_SZ(1);
8585 SvREFCNT_dec(listsv);
8588 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8589 if ( /* If the only flag is folding (plus possibly inversion). */
8590 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8592 for (value = 0; value < 256; ++value) {
8593 if (ANYOF_BITMAP_TEST(ret, value)) {
8594 UV fold = PL_fold[value];
8597 ANYOF_BITMAP_SET(ret, fold);
8600 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8603 /* optimize inverted simple patterns (e.g. [^a-z]) */
8604 if (optimize_invert &&
8605 /* If the only flag is inversion. */
8606 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8607 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8608 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8609 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8612 AV * const av = newAV();
8614 /* The 0th element stores the character class description
8615 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8616 * to initialize the appropriate swash (which gets stored in
8617 * the 1st element), and also useful for dumping the regnode.
8618 * The 2nd element stores the multicharacter foldings,
8619 * used later (regexec.c:S_reginclass()). */
8620 av_store(av, 0, listsv);
8621 av_store(av, 1, NULL);
8622 av_store(av, 2, MUTABLE_SV(unicode_alternate));
8623 rv = newRV_noinc(MUTABLE_SV(av));
8624 n = add_data(pRExC_state, 1, "s");
8625 RExC_rxi->data->data[n] = (void*)rv;
8633 /* reg_skipcomment()
8635 Absorbs an /x style # comments from the input stream.
8636 Returns true if there is more text remaining in the stream.
8637 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8638 terminates the pattern without including a newline.
8640 Note its the callers responsibility to ensure that we are
8646 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8650 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8652 while (RExC_parse < RExC_end)
8653 if (*RExC_parse++ == '\n') {
8658 /* we ran off the end of the pattern without ending
8659 the comment, so we have to add an \n when wrapping */
8660 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8668 Advance that parse position, and optionall absorbs
8669 "whitespace" from the inputstream.
8671 Without /x "whitespace" means (?#...) style comments only,
8672 with /x this means (?#...) and # comments and whitespace proper.
8674 Returns the RExC_parse point from BEFORE the scan occurs.
8676 This is the /x friendly way of saying RExC_parse++.
8680 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8682 char* const retval = RExC_parse++;
8684 PERL_ARGS_ASSERT_NEXTCHAR;
8687 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8688 RExC_parse[2] == '#') {
8689 while (*RExC_parse != ')') {
8690 if (RExC_parse == RExC_end)
8691 FAIL("Sequence (?#... not terminated");
8697 if (RExC_flags & RXf_PMf_EXTENDED) {
8698 if (isSPACE(*RExC_parse)) {
8702 else if (*RExC_parse == '#') {
8703 if ( reg_skipcomment( pRExC_state ) )
8712 - reg_node - emit a node
8714 STATIC regnode * /* Location. */
8715 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8718 register regnode *ptr;
8719 regnode * const ret = RExC_emit;
8720 GET_RE_DEBUG_FLAGS_DECL;
8722 PERL_ARGS_ASSERT_REG_NODE;
8725 SIZE_ALIGN(RExC_size);
8729 if (RExC_emit >= RExC_emit_bound)
8730 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8732 NODE_ALIGN_FILL(ret);
8734 FILL_ADVANCE_NODE(ptr, op);
8735 #ifdef RE_TRACK_PATTERN_OFFSETS
8736 if (RExC_offsets) { /* MJD */
8737 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8738 "reg_node", __LINE__,
8740 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8741 ? "Overwriting end of array!\n" : "OK",
8742 (UV)(RExC_emit - RExC_emit_start),
8743 (UV)(RExC_parse - RExC_start),
8744 (UV)RExC_offsets[0]));
8745 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8753 - reganode - emit a node with an argument
8755 STATIC regnode * /* Location. */
8756 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8759 register regnode *ptr;
8760 regnode * const ret = RExC_emit;
8761 GET_RE_DEBUG_FLAGS_DECL;
8763 PERL_ARGS_ASSERT_REGANODE;
8766 SIZE_ALIGN(RExC_size);
8771 assert(2==regarglen[op]+1);
8773 Anything larger than this has to allocate the extra amount.
8774 If we changed this to be:
8776 RExC_size += (1 + regarglen[op]);
8778 then it wouldn't matter. Its not clear what side effect
8779 might come from that so its not done so far.
8784 if (RExC_emit >= RExC_emit_bound)
8785 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8787 NODE_ALIGN_FILL(ret);
8789 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8790 #ifdef RE_TRACK_PATTERN_OFFSETS
8791 if (RExC_offsets) { /* MJD */
8792 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8796 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8797 "Overwriting end of array!\n" : "OK",
8798 (UV)(RExC_emit - RExC_emit_start),
8799 (UV)(RExC_parse - RExC_start),
8800 (UV)RExC_offsets[0]));
8801 Set_Cur_Node_Offset;
8809 - reguni - emit (if appropriate) a Unicode character
8812 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8816 PERL_ARGS_ASSERT_REGUNI;
8818 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8822 - reginsert - insert an operator in front of already-emitted operand
8824 * Means relocating the operand.
8827 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8830 register regnode *src;
8831 register regnode *dst;
8832 register regnode *place;
8833 const int offset = regarglen[(U8)op];
8834 const int size = NODE_STEP_REGNODE + offset;
8835 GET_RE_DEBUG_FLAGS_DECL;
8837 PERL_ARGS_ASSERT_REGINSERT;
8838 PERL_UNUSED_ARG(depth);
8839 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8840 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8849 if (RExC_open_parens) {
8851 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8852 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8853 if ( RExC_open_parens[paren] >= opnd ) {
8854 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8855 RExC_open_parens[paren] += size;
8857 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8859 if ( RExC_close_parens[paren] >= opnd ) {
8860 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8861 RExC_close_parens[paren] += size;
8863 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8868 while (src > opnd) {
8869 StructCopy(--src, --dst, regnode);
8870 #ifdef RE_TRACK_PATTERN_OFFSETS
8871 if (RExC_offsets) { /* MJD 20010112 */
8872 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8876 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8877 ? "Overwriting end of array!\n" : "OK",
8878 (UV)(src - RExC_emit_start),
8879 (UV)(dst - RExC_emit_start),
8880 (UV)RExC_offsets[0]));
8881 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8882 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8888 place = opnd; /* Op node, where operand used to be. */
8889 #ifdef RE_TRACK_PATTERN_OFFSETS
8890 if (RExC_offsets) { /* MJD */
8891 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8895 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8896 ? "Overwriting end of array!\n" : "OK",
8897 (UV)(place - RExC_emit_start),
8898 (UV)(RExC_parse - RExC_start),
8899 (UV)RExC_offsets[0]));
8900 Set_Node_Offset(place, RExC_parse);
8901 Set_Node_Length(place, 1);
8904 src = NEXTOPER(place);
8905 FILL_ADVANCE_NODE(place, op);
8906 Zero(src, offset, regnode);
8910 - regtail - set the next-pointer at the end of a node chain of p to val.
8911 - SEE ALSO: regtail_study
8913 /* TODO: All three parms should be const */
8915 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8918 register regnode *scan;
8919 GET_RE_DEBUG_FLAGS_DECL;
8921 PERL_ARGS_ASSERT_REGTAIL;
8923 PERL_UNUSED_ARG(depth);
8929 /* Find last node. */
8932 regnode * const temp = regnext(scan);
8934 SV * const mysv=sv_newmortal();
8935 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8936 regprop(RExC_rx, mysv, scan);
8937 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8938 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8939 (temp == NULL ? "->" : ""),
8940 (temp == NULL ? PL_reg_name[OP(val)] : "")
8948 if (reg_off_by_arg[OP(scan)]) {
8949 ARG_SET(scan, val - scan);
8952 NEXT_OFF(scan) = val - scan;
8958 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8959 - Look for optimizable sequences at the same time.
8960 - currently only looks for EXACT chains.
8962 This is expermental code. The idea is to use this routine to perform
8963 in place optimizations on branches and groups as they are constructed,
8964 with the long term intention of removing optimization from study_chunk so
8965 that it is purely analytical.
8967 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8968 to control which is which.
8971 /* TODO: All four parms should be const */
8974 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8977 register regnode *scan;
8979 #ifdef EXPERIMENTAL_INPLACESCAN
8982 GET_RE_DEBUG_FLAGS_DECL;
8984 PERL_ARGS_ASSERT_REGTAIL_STUDY;
8990 /* Find last node. */
8994 regnode * const temp = regnext(scan);
8995 #ifdef EXPERIMENTAL_INPLACESCAN
8996 if (PL_regkind[OP(scan)] == EXACT)
8997 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
9005 if( exact == PSEUDO )
9007 else if ( exact != OP(scan) )
9016 SV * const mysv=sv_newmortal();
9017 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
9018 regprop(RExC_rx, mysv, scan);
9019 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
9020 SvPV_nolen_const(mysv),
9022 PL_reg_name[exact]);
9029 SV * const mysv_val=sv_newmortal();
9030 DEBUG_PARSE_MSG("");
9031 regprop(RExC_rx, mysv_val, val);
9032 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
9033 SvPV_nolen_const(mysv_val),
9034 (IV)REG_NODE_NUM(val),
9038 if (reg_off_by_arg[OP(scan)]) {
9039 ARG_SET(scan, val - scan);
9042 NEXT_OFF(scan) = val - scan;
9050 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
9054 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
9059 for (bit=0; bit<32; bit++) {
9060 if (flags & (1<<bit)) {
9062 PerlIO_printf(Perl_debug_log, "%s",lead);
9063 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
9068 PerlIO_printf(Perl_debug_log, "\n");
9070 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
9076 Perl_regdump(pTHX_ const regexp *r)
9080 SV * const sv = sv_newmortal();
9081 SV *dsv= sv_newmortal();
9083 GET_RE_DEBUG_FLAGS_DECL;
9085 PERL_ARGS_ASSERT_REGDUMP;
9087 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
9089 /* Header fields of interest. */
9090 if (r->anchored_substr) {
9091 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
9092 RE_SV_DUMPLEN(r->anchored_substr), 30);
9093 PerlIO_printf(Perl_debug_log,
9094 "anchored %s%s at %"IVdf" ",
9095 s, RE_SV_TAIL(r->anchored_substr),
9096 (IV)r->anchored_offset);
9097 } else if (r->anchored_utf8) {
9098 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
9099 RE_SV_DUMPLEN(r->anchored_utf8), 30);
9100 PerlIO_printf(Perl_debug_log,
9101 "anchored utf8 %s%s at %"IVdf" ",
9102 s, RE_SV_TAIL(r->anchored_utf8),
9103 (IV)r->anchored_offset);
9105 if (r->float_substr) {
9106 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
9107 RE_SV_DUMPLEN(r->float_substr), 30);
9108 PerlIO_printf(Perl_debug_log,
9109 "floating %s%s at %"IVdf"..%"UVuf" ",
9110 s, RE_SV_TAIL(r->float_substr),
9111 (IV)r->float_min_offset, (UV)r->float_max_offset);
9112 } else if (r->float_utf8) {
9113 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
9114 RE_SV_DUMPLEN(r->float_utf8), 30);
9115 PerlIO_printf(Perl_debug_log,
9116 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
9117 s, RE_SV_TAIL(r->float_utf8),
9118 (IV)r->float_min_offset, (UV)r->float_max_offset);
9120 if (r->check_substr || r->check_utf8)
9121 PerlIO_printf(Perl_debug_log,
9123 (r->check_substr == r->float_substr
9124 && r->check_utf8 == r->float_utf8
9125 ? "(checking floating" : "(checking anchored"));
9126 if (r->extflags & RXf_NOSCAN)
9127 PerlIO_printf(Perl_debug_log, " noscan");
9128 if (r->extflags & RXf_CHECK_ALL)
9129 PerlIO_printf(Perl_debug_log, " isall");
9130 if (r->check_substr || r->check_utf8)
9131 PerlIO_printf(Perl_debug_log, ") ");
9133 if (ri->regstclass) {
9134 regprop(r, sv, ri->regstclass);
9135 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
9137 if (r->extflags & RXf_ANCH) {
9138 PerlIO_printf(Perl_debug_log, "anchored");
9139 if (r->extflags & RXf_ANCH_BOL)
9140 PerlIO_printf(Perl_debug_log, "(BOL)");
9141 if (r->extflags & RXf_ANCH_MBOL)
9142 PerlIO_printf(Perl_debug_log, "(MBOL)");
9143 if (r->extflags & RXf_ANCH_SBOL)
9144 PerlIO_printf(Perl_debug_log, "(SBOL)");
9145 if (r->extflags & RXf_ANCH_GPOS)
9146 PerlIO_printf(Perl_debug_log, "(GPOS)");
9147 PerlIO_putc(Perl_debug_log, ' ');
9149 if (r->extflags & RXf_GPOS_SEEN)
9150 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
9151 if (r->intflags & PREGf_SKIP)
9152 PerlIO_printf(Perl_debug_log, "plus ");
9153 if (r->intflags & PREGf_IMPLICIT)
9154 PerlIO_printf(Perl_debug_log, "implicit ");
9155 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
9156 if (r->extflags & RXf_EVAL_SEEN)
9157 PerlIO_printf(Perl_debug_log, "with eval ");
9158 PerlIO_printf(Perl_debug_log, "\n");
9159 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
9161 PERL_ARGS_ASSERT_REGDUMP;
9162 PERL_UNUSED_CONTEXT;
9164 #endif /* DEBUGGING */
9168 - regprop - printable representation of opcode
9170 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9173 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9174 if (flags & ANYOF_INVERT) \
9175 /*make sure the invert info is in each */ \
9176 sv_catpvs(sv, "^"); \
9182 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9187 RXi_GET_DECL(prog,progi);
9188 GET_RE_DEBUG_FLAGS_DECL;
9190 PERL_ARGS_ASSERT_REGPROP;
9194 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9195 /* It would be nice to FAIL() here, but this may be called from
9196 regexec.c, and it would be hard to supply pRExC_state. */
9197 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9198 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9200 k = PL_regkind[OP(o)];
9204 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9205 * is a crude hack but it may be the best for now since
9206 * we have no flag "this EXACTish node was UTF-8"
9208 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9209 PERL_PV_ESCAPE_UNI_DETECT |
9210 PERL_PV_PRETTY_ELLIPSES |
9211 PERL_PV_PRETTY_LTGT |
9212 PERL_PV_PRETTY_NOCLEAR
9214 } else if (k == TRIE) {
9215 /* print the details of the trie in dumpuntil instead, as
9216 * progi->data isn't available here */
9217 const char op = OP(o);
9218 const U32 n = ARG(o);
9219 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9220 (reg_ac_data *)progi->data->data[n] :
9222 const reg_trie_data * const trie
9223 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9225 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9226 DEBUG_TRIE_COMPILE_r(
9227 Perl_sv_catpvf(aTHX_ sv,
9228 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9229 (UV)trie->startstate,
9230 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9231 (UV)trie->wordcount,
9234 (UV)TRIE_CHARCOUNT(trie),
9235 (UV)trie->uniquecharcount
9238 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9240 int rangestart = -1;
9241 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9243 for (i = 0; i <= 256; i++) {
9244 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9245 if (rangestart == -1)
9247 } else if (rangestart != -1) {
9248 if (i <= rangestart + 3)
9249 for (; rangestart < i; rangestart++)
9250 put_byte(sv, rangestart);
9252 put_byte(sv, rangestart);
9254 put_byte(sv, i - 1);
9262 } else if (k == CURLY) {
9263 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9264 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9265 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9267 else if (k == WHILEM && o->flags) /* Ordinal/of */
9268 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9269 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9270 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9271 if ( RXp_PAREN_NAMES(prog) ) {
9272 if ( k != REF || OP(o) < NREF) {
9273 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9274 SV **name= av_fetch(list, ARG(o), 0 );
9276 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9279 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9280 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9281 I32 *nums=(I32*)SvPVX(sv_dat);
9282 SV **name= av_fetch(list, nums[0], 0 );
9285 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9286 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9287 (n ? "," : ""), (IV)nums[n]);
9289 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9293 } else if (k == GOSUB)
9294 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9295 else if (k == VERB) {
9297 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9298 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9299 } else if (k == LOGICAL)
9300 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9301 else if (k == FOLDCHAR)
9302 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9303 else if (k == ANYOF) {
9304 int i, rangestart = -1;
9305 const U8 flags = ANYOF_FLAGS(o);
9308 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9309 static const char * const anyofs[] = {
9342 if (flags & ANYOF_LOCALE)
9343 sv_catpvs(sv, "{loc}");
9344 if (flags & ANYOF_FOLD)
9345 sv_catpvs(sv, "{i}");
9346 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9347 if (flags & ANYOF_INVERT)
9350 /* output what the standard cp 0-255 bitmap matches */
9351 for (i = 0; i <= 256; i++) {
9352 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9353 if (rangestart == -1)
9355 } else if (rangestart != -1) {
9356 if (i <= rangestart + 3)
9357 for (; rangestart < i; rangestart++)
9358 put_byte(sv, rangestart);
9360 put_byte(sv, rangestart);
9362 put_byte(sv, i - 1);
9369 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9370 /* output any special charclass tests (used mostly under use locale) */
9371 if (o->flags & ANYOF_CLASS)
9372 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9373 if (ANYOF_CLASS_TEST(o,i)) {
9374 sv_catpv(sv, anyofs[i]);
9378 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9380 /* output information about the unicode matching */
9381 if (flags & ANYOF_UNICODE)
9382 sv_catpvs(sv, "{unicode}");
9383 else if (flags & ANYOF_UNICODE_ALL)
9384 sv_catpvs(sv, "{unicode_all}");
9388 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9392 U8 s[UTF8_MAXBYTES_CASE+1];
9394 for (i = 0; i <= 256; i++) { /* just the first 256 */
9395 uvchr_to_utf8(s, i);
9397 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9398 if (rangestart == -1)
9400 } else if (rangestart != -1) {
9401 if (i <= rangestart + 3)
9402 for (; rangestart < i; rangestart++) {
9403 const U8 * const e = uvchr_to_utf8(s,rangestart);
9405 for(p = s; p < e; p++)
9409 const U8 *e = uvchr_to_utf8(s,rangestart);
9411 for (p = s; p < e; p++)
9414 e = uvchr_to_utf8(s, i-1);
9415 for (p = s; p < e; p++)
9422 sv_catpvs(sv, "..."); /* et cetera */
9426 char *s = savesvpv(lv);
9427 char * const origs = s;
9429 while (*s && *s != '\n')
9433 const char * const t = ++s;
9451 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9453 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9454 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9456 PERL_UNUSED_CONTEXT;
9457 PERL_UNUSED_ARG(sv);
9459 PERL_UNUSED_ARG(prog);
9460 #endif /* DEBUGGING */
9464 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9465 { /* Assume that RE_INTUIT is set */
9467 struct regexp *const prog = (struct regexp *)SvANY(r);
9468 GET_RE_DEBUG_FLAGS_DECL;
9470 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9471 PERL_UNUSED_CONTEXT;
9475 const char * const s = SvPV_nolen_const(prog->check_substr
9476 ? prog->check_substr : prog->check_utf8);
9478 if (!PL_colorset) reginitcolors();
9479 PerlIO_printf(Perl_debug_log,
9480 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9482 prog->check_substr ? "" : "utf8 ",
9483 PL_colors[5],PL_colors[0],
9486 (strlen(s) > 60 ? "..." : ""));
9489 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9495 handles refcounting and freeing the perl core regexp structure. When
9496 it is necessary to actually free the structure the first thing it
9497 does is call the 'free' method of the regexp_engine associated to to
9498 the regexp, allowing the handling of the void *pprivate; member
9499 first. (This routine is not overridable by extensions, which is why
9500 the extensions free is called first.)
9502 See regdupe and regdupe_internal if you change anything here.
9504 #ifndef PERL_IN_XSUB_RE
9506 Perl_pregfree(pTHX_ REGEXP *r)
9512 Perl_pregfree2(pTHX_ REGEXP *rx)
9515 struct regexp *const r = (struct regexp *)SvANY(rx);
9516 GET_RE_DEBUG_FLAGS_DECL;
9518 PERL_ARGS_ASSERT_PREGFREE2;
9521 ReREFCNT_dec(r->mother_re);
9523 CALLREGFREE_PVT(rx); /* free the private data */
9524 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9527 SvREFCNT_dec(r->anchored_substr);
9528 SvREFCNT_dec(r->anchored_utf8);
9529 SvREFCNT_dec(r->float_substr);
9530 SvREFCNT_dec(r->float_utf8);
9531 Safefree(r->substrs);
9533 RX_MATCH_COPY_FREE(rx);
9534 #ifdef PERL_OLD_COPY_ON_WRITE
9535 SvREFCNT_dec(r->saved_copy);
9542 This is a hacky workaround to the structural issue of match results
9543 being stored in the regexp structure which is in turn stored in
9544 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9545 could be PL_curpm in multiple contexts, and could require multiple
9546 result sets being associated with the pattern simultaneously, such
9547 as when doing a recursive match with (??{$qr})
9549 The solution is to make a lightweight copy of the regexp structure
9550 when a qr// is returned from the code executed by (??{$qr}) this
9551 lightweight copy doesnt actually own any of its data except for
9552 the starp/end and the actual regexp structure itself.
9558 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
9561 struct regexp *const r = (struct regexp *)SvANY(rx);
9562 register const I32 npar = r->nparens+1;
9564 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9567 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9568 ret = (struct regexp *)SvANY(ret_x);
9570 (void)ReREFCNT_inc(rx);
9571 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9572 by pointing directly at the buffer, but flagging that the allocated
9573 space in the copy is zero. As we've just done a struct copy, it's now
9574 a case of zero-ing that, rather than copying the current length. */
9575 SvPV_set(ret_x, RX_WRAPPED(rx));
9576 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9577 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
9578 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
9579 SvLEN_set(ret_x, 0);
9580 SvSTASH_set(ret_x, NULL);
9581 SvMAGIC_set(ret_x, NULL);
9582 Newx(ret->offs, npar, regexp_paren_pair);
9583 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9585 Newx(ret->substrs, 1, struct reg_substr_data);
9586 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9588 SvREFCNT_inc_void(ret->anchored_substr);
9589 SvREFCNT_inc_void(ret->anchored_utf8);
9590 SvREFCNT_inc_void(ret->float_substr);
9591 SvREFCNT_inc_void(ret->float_utf8);
9593 /* check_substr and check_utf8, if non-NULL, point to either their
9594 anchored or float namesakes, and don't hold a second reference. */
9596 RX_MATCH_COPIED_off(ret_x);
9597 #ifdef PERL_OLD_COPY_ON_WRITE
9598 ret->saved_copy = NULL;
9600 ret->mother_re = rx;
9606 /* regfree_internal()
9608 Free the private data in a regexp. This is overloadable by
9609 extensions. Perl takes care of the regexp structure in pregfree(),
9610 this covers the *pprivate pointer which technically perldoesnt
9611 know about, however of course we have to handle the
9612 regexp_internal structure when no extension is in use.
9614 Note this is called before freeing anything in the regexp
9619 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9622 struct regexp *const r = (struct regexp *)SvANY(rx);
9624 GET_RE_DEBUG_FLAGS_DECL;
9626 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9632 SV *dsv= sv_newmortal();
9633 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9634 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9635 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9636 PL_colors[4],PL_colors[5],s);
9639 #ifdef RE_TRACK_PATTERN_OFFSETS
9641 Safefree(ri->u.offsets); /* 20010421 MJD */
9644 int n = ri->data->count;
9645 PAD* new_comppad = NULL;
9650 /* If you add a ->what type here, update the comment in regcomp.h */
9651 switch (ri->data->what[n]) {
9656 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
9659 Safefree(ri->data->data[n]);
9662 new_comppad = MUTABLE_AV(ri->data->data[n]);
9665 if (new_comppad == NULL)
9666 Perl_croak(aTHX_ "panic: pregfree comppad");
9667 PAD_SAVE_LOCAL(old_comppad,
9668 /* Watch out for global destruction's random ordering. */
9669 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9672 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9675 op_free((OP_4tree*)ri->data->data[n]);
9677 PAD_RESTORE_LOCAL(old_comppad);
9678 SvREFCNT_dec(MUTABLE_SV(new_comppad));
9684 { /* Aho Corasick add-on structure for a trie node.
9685 Used in stclass optimization only */
9687 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9689 refcount = --aho->refcount;
9692 PerlMemShared_free(aho->states);
9693 PerlMemShared_free(aho->fail);
9694 /* do this last!!!! */
9695 PerlMemShared_free(ri->data->data[n]);
9696 PerlMemShared_free(ri->regstclass);
9702 /* trie structure. */
9704 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9706 refcount = --trie->refcount;
9709 PerlMemShared_free(trie->charmap);
9710 PerlMemShared_free(trie->states);
9711 PerlMemShared_free(trie->trans);
9713 PerlMemShared_free(trie->bitmap);
9715 PerlMemShared_free(trie->jump);
9716 PerlMemShared_free(trie->wordinfo);
9717 /* do this last!!!! */
9718 PerlMemShared_free(ri->data->data[n]);
9723 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9726 Safefree(ri->data->what);
9733 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
9734 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
9735 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9738 re_dup - duplicate a regexp.
9740 This routine is expected to clone a given regexp structure. It is only
9741 compiled under USE_ITHREADS.
9743 After all of the core data stored in struct regexp is duplicated
9744 the regexp_engine.dupe method is used to copy any private data
9745 stored in the *pprivate pointer. This allows extensions to handle
9746 any duplication it needs to do.
9748 See pregfree() and regfree_internal() if you change anything here.
9750 #if defined(USE_ITHREADS)
9751 #ifndef PERL_IN_XSUB_RE
9753 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9757 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9758 struct regexp *ret = (struct regexp *)SvANY(dstr);
9760 PERL_ARGS_ASSERT_RE_DUP_GUTS;
9762 npar = r->nparens+1;
9763 Newx(ret->offs, npar, regexp_paren_pair);
9764 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9766 /* no need to copy these */
9767 Newx(ret->swap, npar, regexp_paren_pair);
9771 /* Do it this way to avoid reading from *r after the StructCopy().
9772 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9773 cache, it doesn't matter. */
9774 const bool anchored = r->check_substr
9775 ? r->check_substr == r->anchored_substr
9776 : r->check_utf8 == r->anchored_utf8;
9777 Newx(ret->substrs, 1, struct reg_substr_data);
9778 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9780 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9781 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9782 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9783 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9785 /* check_substr and check_utf8, if non-NULL, point to either their
9786 anchored or float namesakes, and don't hold a second reference. */
9788 if (ret->check_substr) {
9790 assert(r->check_utf8 == r->anchored_utf8);
9791 ret->check_substr = ret->anchored_substr;
9792 ret->check_utf8 = ret->anchored_utf8;
9794 assert(r->check_substr == r->float_substr);
9795 assert(r->check_utf8 == r->float_utf8);
9796 ret->check_substr = ret->float_substr;
9797 ret->check_utf8 = ret->float_utf8;
9799 } else if (ret->check_utf8) {
9801 ret->check_utf8 = ret->anchored_utf8;
9803 ret->check_utf8 = ret->float_utf8;
9808 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9811 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9813 if (RX_MATCH_COPIED(dstr))
9814 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9817 #ifdef PERL_OLD_COPY_ON_WRITE
9818 ret->saved_copy = NULL;
9821 if (ret->mother_re) {
9822 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
9823 /* Our storage points directly to our mother regexp, but that's
9824 1: a buffer in a different thread
9825 2: something we no longer hold a reference on
9826 so we need to copy it locally. */
9827 /* Note we need to sue SvCUR() on our mother_re, because it, in
9828 turn, may well be pointing to its own mother_re. */
9829 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
9830 SvCUR(ret->mother_re)+1));
9831 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
9833 ret->mother_re = NULL;
9837 #endif /* PERL_IN_XSUB_RE */
9842 This is the internal complement to regdupe() which is used to copy
9843 the structure pointed to by the *pprivate pointer in the regexp.
9844 This is the core version of the extension overridable cloning hook.
9845 The regexp structure being duplicated will be copied by perl prior
9846 to this and will be provided as the regexp *r argument, however
9847 with the /old/ structures pprivate pointer value. Thus this routine
9848 may override any copying normally done by perl.
9850 It returns a pointer to the new regexp_internal structure.
9854 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
9857 struct regexp *const r = (struct regexp *)SvANY(rx);
9858 regexp_internal *reti;
9862 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
9864 npar = r->nparens+1;
9867 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
9868 Copy(ri->program, reti->program, len+1, regnode);
9871 reti->regstclass = NULL;
9875 const int count = ri->data->count;
9878 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9879 char, struct reg_data);
9880 Newx(d->what, count, U8);
9883 for (i = 0; i < count; i++) {
9884 d->what[i] = ri->data->what[i];
9885 switch (d->what[i]) {
9886 /* legal options are one of: sSfpontTua
9887 see also regcomp.h and pregfree() */
9888 case 'a': /* actually an AV, but the dup function is identical. */
9891 case 'p': /* actually an AV, but the dup function is identical. */
9892 case 'u': /* actually an HV, but the dup function is identical. */
9893 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
9896 /* This is cheating. */
9897 Newx(d->data[i], 1, struct regnode_charclass_class);
9898 StructCopy(ri->data->data[i], d->data[i],
9899 struct regnode_charclass_class);
9900 reti->regstclass = (regnode*)d->data[i];
9903 /* Compiled op trees are readonly and in shared memory,
9904 and can thus be shared without duplication. */
9906 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9910 /* Trie stclasses are readonly and can thus be shared
9911 * without duplication. We free the stclass in pregfree
9912 * when the corresponding reg_ac_data struct is freed.
9914 reti->regstclass= ri->regstclass;
9918 ((reg_trie_data*)ri->data->data[i])->refcount++;
9922 d->data[i] = ri->data->data[i];
9925 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9934 reti->name_list_idx = ri->name_list_idx;
9936 #ifdef RE_TRACK_PATTERN_OFFSETS
9937 if (ri->u.offsets) {
9938 Newx(reti->u.offsets, 2*len+1, U32);
9939 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9942 SetProgLen(reti,len);
9948 #endif /* USE_ITHREADS */
9950 #ifndef PERL_IN_XSUB_RE
9953 - regnext - dig the "next" pointer out of a node
9956 Perl_regnext(pTHX_ register regnode *p)
9959 register I32 offset;
9964 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
9965 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
9968 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9977 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9980 STRLEN l1 = strlen(pat1);
9981 STRLEN l2 = strlen(pat2);
9984 const char *message;
9986 PERL_ARGS_ASSERT_RE_CROAK2;
9992 Copy(pat1, buf, l1 , char);
9993 Copy(pat2, buf + l1, l2 , char);
9994 buf[l1 + l2] = '\n';
9995 buf[l1 + l2 + 1] = '\0';
9997 /* ANSI variant takes additional second argument */
9998 va_start(args, pat2);
10002 msv = vmess(buf, &args);
10004 message = SvPV_const(msv,l1);
10007 Copy(message, buf, l1 , char);
10008 buf[l1-1] = '\0'; /* Overwrite \n */
10009 Perl_croak(aTHX_ "%s", buf);
10012 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
10014 #ifndef PERL_IN_XSUB_RE
10016 Perl_save_re_context(pTHX)
10020 struct re_save_state *state;
10022 SAVEVPTR(PL_curcop);
10023 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
10025 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
10026 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
10027 SSPUSHUV(SAVEt_RE_STATE);
10029 Copy(&PL_reg_state, state, 1, struct re_save_state);
10031 PL_reg_start_tmp = 0;
10032 PL_reg_start_tmpl = 0;
10033 PL_reg_oldsaved = NULL;
10034 PL_reg_oldsavedlen = 0;
10035 PL_reg_maxiter = 0;
10036 PL_reg_leftiter = 0;
10037 PL_reg_poscache = NULL;
10038 PL_reg_poscache_size = 0;
10039 #ifdef PERL_OLD_COPY_ON_WRITE
10043 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
10045 const REGEXP * const rx = PM_GETRE(PL_curpm);
10048 for (i = 1; i <= RX_NPARENS(rx); i++) {
10049 char digits[TYPE_CHARS(long)];
10050 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
10051 GV *const *const gvp
10052 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
10055 GV * const gv = *gvp;
10056 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
10066 clear_re(pTHX_ void *r)
10069 ReREFCNT_dec((REGEXP *)r);
10075 S_put_byte(pTHX_ SV *sv, int c)
10077 PERL_ARGS_ASSERT_PUT_BYTE;
10079 /* Our definition of isPRINT() ignores locales, so only bytes that are
10080 not part of UTF-8 are considered printable. I assume that the same
10081 holds for UTF-EBCDIC.
10082 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
10083 which Wikipedia says:
10085 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
10086 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
10087 identical, to the ASCII delete (DEL) or rubout control character.
10088 ) So the old condition can be simplified to !isPRINT(c) */
10090 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
10092 const char string = c;
10093 if (c == '-' || c == ']' || c == '\\' || c == '^')
10094 sv_catpvs(sv, "\\");
10095 sv_catpvn(sv, &string, 1);
10100 #define CLEAR_OPTSTART \
10101 if (optstart) STMT_START { \
10102 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
10106 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
10108 STATIC const regnode *
10109 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
10110 const regnode *last, const regnode *plast,
10111 SV* sv, I32 indent, U32 depth)
10114 register U8 op = PSEUDO; /* Arbitrary non-END op. */
10115 register const regnode *next;
10116 const regnode *optstart= NULL;
10118 RXi_GET_DECL(r,ri);
10119 GET_RE_DEBUG_FLAGS_DECL;
10121 PERL_ARGS_ASSERT_DUMPUNTIL;
10123 #ifdef DEBUG_DUMPUNTIL
10124 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
10125 last ? last-start : 0,plast ? plast-start : 0);
10128 if (plast && plast < last)
10131 while (PL_regkind[op] != END && (!last || node < last)) {
10132 /* While that wasn't END last time... */
10135 if (op == CLOSE || op == WHILEM)
10137 next = regnext((regnode *)node);
10140 if (OP(node) == OPTIMIZED) {
10141 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10148 regprop(r, sv, node);
10149 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10150 (int)(2*indent + 1), "", SvPVX_const(sv));
10152 if (OP(node) != OPTIMIZED) {
10153 if (next == NULL) /* Next ptr. */
10154 PerlIO_printf(Perl_debug_log, " (0)");
10155 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10156 PerlIO_printf(Perl_debug_log, " (FAIL)");
10158 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10159 (void)PerlIO_putc(Perl_debug_log, '\n');
10163 if (PL_regkind[(U8)op] == BRANCHJ) {
10166 register const regnode *nnode = (OP(next) == LONGJMP
10167 ? regnext((regnode *)next)
10169 if (last && nnode > last)
10171 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10174 else if (PL_regkind[(U8)op] == BRANCH) {
10176 DUMPUNTIL(NEXTOPER(node), next);
10178 else if ( PL_regkind[(U8)op] == TRIE ) {
10179 const regnode *this_trie = node;
10180 const char op = OP(node);
10181 const U32 n = ARG(node);
10182 const reg_ac_data * const ac = op>=AHOCORASICK ?
10183 (reg_ac_data *)ri->data->data[n] :
10185 const reg_trie_data * const trie =
10186 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10188 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10190 const regnode *nextbranch= NULL;
10193 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10194 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10196 PerlIO_printf(Perl_debug_log, "%*s%s ",
10197 (int)(2*(indent+3)), "",
10198 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10199 PL_colors[0], PL_colors[1],
10200 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10201 PERL_PV_PRETTY_ELLIPSES |
10202 PERL_PV_PRETTY_LTGT
10207 U16 dist= trie->jump[word_idx+1];
10208 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10209 (UV)((dist ? this_trie + dist : next) - start));
10212 nextbranch= this_trie + trie->jump[0];
10213 DUMPUNTIL(this_trie + dist, nextbranch);
10215 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10216 nextbranch= regnext((regnode *)nextbranch);
10218 PerlIO_printf(Perl_debug_log, "\n");
10221 if (last && next > last)
10226 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10227 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10228 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10230 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10232 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10234 else if ( op == PLUS || op == STAR) {
10235 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10237 else if (op == ANYOF) {
10238 /* arglen 1 + class block */
10239 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
10240 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10241 node = NEXTOPER(node);
10243 else if (PL_regkind[(U8)op] == EXACT) {
10244 /* Literal string, where present. */
10245 node += NODE_SZ_STR(node) - 1;
10246 node = NEXTOPER(node);
10249 node = NEXTOPER(node);
10250 node += regarglen[(U8)op];
10252 if (op == CURLYX || op == OPEN)
10256 #ifdef DEBUG_DUMPUNTIL
10257 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10262 #endif /* DEBUGGING */
10266 * c-indentation-style: bsd
10267 * c-basic-offset: 4
10268 * indent-tabs-mode: t
10271 * ex: set ts=8 sts=4 sw=4 noet: