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;
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 /* Longjmp back to here if have to switch in midstream to utf8 */
4302 if (! RExC_orig_utf8) {
4303 JMPENV_PUSH(jump_ret);
4306 if (jump_ret == 0) { /* First time through */
4307 exp = SvPV(pattern, plen);
4311 SV *dsv= sv_newmortal();
4312 RE_PV_QUOTED_DECL(s, RExC_utf8,
4313 dsv, exp, plen, 60);
4314 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4315 PL_colors[4],PL_colors[5],s);
4318 else { /* longjumped back */
4321 /* If the cause for the longjmp was other than changing to utf8, pop
4322 * our own setjmp, and longjmp to the correct handler */
4323 if (jump_ret != UTF8_LONGJMP) {
4325 JMPENV_JUMP(jump_ret);
4330 /* It's possible to write a regexp in ascii that represents Unicode
4331 codepoints outside of the byte range, such as via \x{100}. If we
4332 detect such a sequence we have to convert the entire pattern to utf8
4333 and then recompile, as our sizing calculation will have been based
4334 on 1 byte == 1 character, but we will need to use utf8 to encode
4335 at least some part of the pattern, and therefore must convert the whole
4338 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4339 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4340 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4342 RExC_orig_utf8 = RExC_utf8 = 1;
4346 #ifdef TRIE_STUDY_OPT
4351 RExC_flags = pm_flags;
4355 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4356 RExC_seen_evals = 0;
4359 /* First pass: determine size, legality. */
4367 RExC_emit = &PL_regdummy;
4368 RExC_whilem_seen = 0;
4369 RExC_open_parens = NULL;
4370 RExC_close_parens = NULL;
4372 RExC_paren_names = NULL;
4374 RExC_paren_name_list = NULL;
4376 RExC_recurse = NULL;
4377 RExC_recurse_count = 0;
4379 #if 0 /* REGC() is (currently) a NOP at the first pass.
4380 * Clever compilers notice this and complain. --jhi */
4381 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4383 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4384 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4385 RExC_precomp = NULL;
4389 /* Here, finished first pass. Get rid of our setjmp, which we added for
4390 * efficiency only if the passed-in string wasn't in utf8, as shown by
4391 * RExC_orig_utf8. But if the first pass was redone, that variable will be
4392 * 1 here even though the original string wasn't utf8, but in this case
4393 * there will have been a long jump */
4394 if (jump_ret == UTF8_LONGJMP || ! RExC_orig_utf8) {
4398 PerlIO_printf(Perl_debug_log,
4399 "Required size %"IVdf" nodes\n"
4400 "Starting second pass (creation)\n",
4403 RExC_lastparse=NULL;
4405 /* Small enough for pointer-storage convention?
4406 If extralen==0, this means that we will not need long jumps. */
4407 if (RExC_size >= 0x10000L && RExC_extralen)
4408 RExC_size += RExC_extralen;
4411 if (RExC_whilem_seen > 15)
4412 RExC_whilem_seen = 15;
4414 /* Allocate space and zero-initialize. Note, the two step process
4415 of zeroing when in debug mode, thus anything assigned has to
4416 happen after that */
4417 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4418 r = (struct regexp*)SvANY(rx);
4419 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4420 char, regexp_internal);
4421 if ( r == NULL || ri == NULL )
4422 FAIL("Regexp out of space");
4424 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4425 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4427 /* bulk initialize base fields with 0. */
4428 Zero(ri, sizeof(regexp_internal), char);
4431 /* non-zero initialization begins here */
4433 r->engine= RE_ENGINE_PTR;
4434 r->extflags = pm_flags;
4436 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4437 bool has_charset = (r->extflags & (RXf_PMf_LOCALE|RXf_PMf_UNICODE));
4439 /* The caret is output if there are any defaults: if not all the STD
4440 * flags are set, or if no character set specifier is needed */
4442 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4444 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4445 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4446 >> RXf_PMf_STD_PMMOD_SHIFT);
4447 const char *fptr = STD_PAT_MODS; /*"msix"*/
4449 /* Allocate for the worst case, which is all the std flags are turned
4450 * on. If more precision is desired, we could do a population count of
4451 * the flags set. This could be done with a small lookup table, or by
4452 * shifting, masking and adding, or even, when available, assembly
4453 * language for a machine-language population count.
4454 * We never output a minus, as all those are defaults, so are
4455 * covered by the caret */
4456 const STRLEN wraplen = plen + has_p + has_runon
4457 + has_default /* If needs a caret */
4458 + has_charset /* If needs a character set specifier */
4459 + (sizeof(STD_PAT_MODS) - 1)
4460 + (sizeof("(?:)") - 1);
4462 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4464 SvFLAGS(rx) |= SvUTF8(pattern);
4467 /* If a default, cover it using the caret */
4469 *p++= DEFAULT_PAT_MOD;
4472 if (r->extflags & RXf_PMf_LOCALE) {
4473 *p++ = LOCALE_PAT_MOD;
4475 *p++ = UNICODE_PAT_MOD;
4479 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4482 while((ch = *fptr++)) {
4490 Copy(RExC_precomp, p, plen, char);
4491 assert ((RX_WRAPPED(rx) - p) < 16);
4492 r->pre_prefix = p - RX_WRAPPED(rx);
4498 SvCUR_set(rx, p - SvPVX_const(rx));
4502 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4504 if (RExC_seen & REG_SEEN_RECURSE) {
4505 Newxz(RExC_open_parens, RExC_npar,regnode *);
4506 SAVEFREEPV(RExC_open_parens);
4507 Newxz(RExC_close_parens,RExC_npar,regnode *);
4508 SAVEFREEPV(RExC_close_parens);
4511 /* Useful during FAIL. */
4512 #ifdef RE_TRACK_PATTERN_OFFSETS
4513 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4514 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4515 "%s %"UVuf" bytes for offset annotations.\n",
4516 ri->u.offsets ? "Got" : "Couldn't get",
4517 (UV)((2*RExC_size+1) * sizeof(U32))));
4519 SetProgLen(ri,RExC_size);
4524 /* Second pass: emit code. */
4525 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4530 RExC_emit_start = ri->program;
4531 RExC_emit = ri->program;
4532 RExC_emit_bound = ri->program + RExC_size + 1;
4534 /* Store the count of eval-groups for security checks: */
4535 RExC_rx->seen_evals = RExC_seen_evals;
4536 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4537 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4541 /* XXXX To minimize changes to RE engine we always allocate
4542 3-units-long substrs field. */
4543 Newx(r->substrs, 1, struct reg_substr_data);
4544 if (RExC_recurse_count) {
4545 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4546 SAVEFREEPV(RExC_recurse);
4550 r->minlen = minlen = sawplus = sawopen = 0;
4551 Zero(r->substrs, 1, struct reg_substr_data);
4553 #ifdef TRIE_STUDY_OPT
4555 StructCopy(&zero_scan_data, &data, scan_data_t);
4556 copyRExC_state = RExC_state;
4559 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4561 RExC_state = copyRExC_state;
4562 if (seen & REG_TOP_LEVEL_BRANCHES)
4563 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4565 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4566 if (data.last_found) {
4567 SvREFCNT_dec(data.longest_fixed);
4568 SvREFCNT_dec(data.longest_float);
4569 SvREFCNT_dec(data.last_found);
4571 StructCopy(&zero_scan_data, &data, scan_data_t);
4574 StructCopy(&zero_scan_data, &data, scan_data_t);
4577 /* Dig out information for optimizations. */
4578 r->extflags = RExC_flags; /* was pm_op */
4579 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4582 SvUTF8_on(rx); /* Unicode in it? */
4583 ri->regstclass = NULL;
4584 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4585 r->intflags |= PREGf_NAUGHTY;
4586 scan = ri->program + 1; /* First BRANCH. */
4588 /* testing for BRANCH here tells us whether there is "must appear"
4589 data in the pattern. If there is then we can use it for optimisations */
4590 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4592 STRLEN longest_float_length, longest_fixed_length;
4593 struct regnode_charclass_class ch_class; /* pointed to by data */
4595 I32 last_close = 0; /* pointed to by data */
4596 regnode *first= scan;
4597 regnode *first_next= regnext(first);
4600 * Skip introductions and multiplicators >= 1
4601 * so that we can extract the 'meat' of the pattern that must
4602 * match in the large if() sequence following.
4603 * NOTE that EXACT is NOT covered here, as it is normally
4604 * picked up by the optimiser separately.
4606 * This is unfortunate as the optimiser isnt handling lookahead
4607 * properly currently.
4610 while ((OP(first) == OPEN && (sawopen = 1)) ||
4611 /* An OR of *one* alternative - should not happen now. */
4612 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4613 /* for now we can't handle lookbehind IFMATCH*/
4614 (OP(first) == IFMATCH && !first->flags) ||
4615 (OP(first) == PLUS) ||
4616 (OP(first) == MINMOD) ||
4617 /* An {n,m} with n>0 */
4618 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4619 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4622 * the only op that could be a regnode is PLUS, all the rest
4623 * will be regnode_1 or regnode_2.
4626 if (OP(first) == PLUS)
4629 first += regarglen[OP(first)];
4631 first = NEXTOPER(first);
4632 first_next= regnext(first);
4635 /* Starting-point info. */
4637 DEBUG_PEEP("first:",first,0);
4638 /* Ignore EXACT as we deal with it later. */
4639 if (PL_regkind[OP(first)] == EXACT) {
4640 if (OP(first) == EXACT)
4641 NOOP; /* Empty, get anchored substr later. */
4642 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4643 ri->regstclass = first;
4646 else if (PL_regkind[OP(first)] == TRIE &&
4647 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4650 /* this can happen only on restudy */
4651 if ( OP(first) == TRIE ) {
4652 struct regnode_1 *trieop = (struct regnode_1 *)
4653 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4654 StructCopy(first,trieop,struct regnode_1);
4655 trie_op=(regnode *)trieop;
4657 struct regnode_charclass *trieop = (struct regnode_charclass *)
4658 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4659 StructCopy(first,trieop,struct regnode_charclass);
4660 trie_op=(regnode *)trieop;
4663 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4664 ri->regstclass = trie_op;
4667 else if (REGNODE_SIMPLE(OP(first)))
4668 ri->regstclass = first;
4669 else if (PL_regkind[OP(first)] == BOUND ||
4670 PL_regkind[OP(first)] == NBOUND)
4671 ri->regstclass = first;
4672 else if (PL_regkind[OP(first)] == BOL) {
4673 r->extflags |= (OP(first) == MBOL
4675 : (OP(first) == SBOL
4678 first = NEXTOPER(first);
4681 else if (OP(first) == GPOS) {
4682 r->extflags |= RXf_ANCH_GPOS;
4683 first = NEXTOPER(first);
4686 else if ((!sawopen || !RExC_sawback) &&
4687 (OP(first) == STAR &&
4688 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4689 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4691 /* turn .* into ^.* with an implied $*=1 */
4693 (OP(NEXTOPER(first)) == REG_ANY)
4696 r->extflags |= type;
4697 r->intflags |= PREGf_IMPLICIT;
4698 first = NEXTOPER(first);
4701 if (sawplus && (!sawopen || !RExC_sawback)
4702 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4703 /* x+ must match at the 1st pos of run of x's */
4704 r->intflags |= PREGf_SKIP;
4706 /* Scan is after the zeroth branch, first is atomic matcher. */
4707 #ifdef TRIE_STUDY_OPT
4710 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4711 (IV)(first - scan + 1))
4715 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4716 (IV)(first - scan + 1))
4722 * If there's something expensive in the r.e., find the
4723 * longest literal string that must appear and make it the
4724 * regmust. Resolve ties in favor of later strings, since
4725 * the regstart check works with the beginning of the r.e.
4726 * and avoiding duplication strengthens checking. Not a
4727 * strong reason, but sufficient in the absence of others.
4728 * [Now we resolve ties in favor of the earlier string if
4729 * it happens that c_offset_min has been invalidated, since the
4730 * earlier string may buy us something the later one won't.]
4733 data.longest_fixed = newSVpvs("");
4734 data.longest_float = newSVpvs("");
4735 data.last_found = newSVpvs("");
4736 data.longest = &(data.longest_fixed);
4738 if (!ri->regstclass) {
4739 cl_init(pRExC_state, &ch_class);
4740 data.start_class = &ch_class;
4741 stclass_flag = SCF_DO_STCLASS_AND;
4742 } else /* XXXX Check for BOUND? */
4744 data.last_closep = &last_close;
4746 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4747 &data, -1, NULL, NULL,
4748 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4754 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4755 && data.last_start_min == 0 && data.last_end > 0
4756 && !RExC_seen_zerolen
4757 && !(RExC_seen & REG_SEEN_VERBARG)
4758 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4759 r->extflags |= RXf_CHECK_ALL;
4760 scan_commit(pRExC_state, &data,&minlen,0);
4761 SvREFCNT_dec(data.last_found);
4763 /* Note that code very similar to this but for anchored string
4764 follows immediately below, changes may need to be made to both.
4767 longest_float_length = CHR_SVLEN(data.longest_float);
4768 if (longest_float_length
4769 || (data.flags & SF_FL_BEFORE_EOL
4770 && (!(data.flags & SF_FL_BEFORE_MEOL)
4771 || (RExC_flags & RXf_PMf_MULTILINE))))
4775 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4776 && data.offset_fixed == data.offset_float_min
4777 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4778 goto remove_float; /* As in (a)+. */
4780 /* copy the information about the longest float from the reg_scan_data
4781 over to the program. */
4782 if (SvUTF8(data.longest_float)) {
4783 r->float_utf8 = data.longest_float;
4784 r->float_substr = NULL;
4786 r->float_substr = data.longest_float;
4787 r->float_utf8 = NULL;
4789 /* float_end_shift is how many chars that must be matched that
4790 follow this item. We calculate it ahead of time as once the
4791 lookbehind offset is added in we lose the ability to correctly
4793 ml = data.minlen_float ? *(data.minlen_float)
4794 : (I32)longest_float_length;
4795 r->float_end_shift = ml - data.offset_float_min
4796 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4797 + data.lookbehind_float;
4798 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4799 r->float_max_offset = data.offset_float_max;
4800 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4801 r->float_max_offset -= data.lookbehind_float;
4803 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4804 && (!(data.flags & SF_FL_BEFORE_MEOL)
4805 || (RExC_flags & RXf_PMf_MULTILINE)));
4806 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4810 r->float_substr = r->float_utf8 = NULL;
4811 SvREFCNT_dec(data.longest_float);
4812 longest_float_length = 0;
4815 /* Note that code very similar to this but for floating string
4816 is immediately above, changes may need to be made to both.
4819 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4820 if (longest_fixed_length
4821 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4822 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4823 || (RExC_flags & RXf_PMf_MULTILINE))))
4827 /* copy the information about the longest fixed
4828 from the reg_scan_data over to the program. */
4829 if (SvUTF8(data.longest_fixed)) {
4830 r->anchored_utf8 = data.longest_fixed;
4831 r->anchored_substr = NULL;
4833 r->anchored_substr = data.longest_fixed;
4834 r->anchored_utf8 = NULL;
4836 /* fixed_end_shift is how many chars that must be matched that
4837 follow this item. We calculate it ahead of time as once the
4838 lookbehind offset is added in we lose the ability to correctly
4840 ml = data.minlen_fixed ? *(data.minlen_fixed)
4841 : (I32)longest_fixed_length;
4842 r->anchored_end_shift = ml - data.offset_fixed
4843 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4844 + data.lookbehind_fixed;
4845 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4847 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4848 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4849 || (RExC_flags & RXf_PMf_MULTILINE)));
4850 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4853 r->anchored_substr = r->anchored_utf8 = NULL;
4854 SvREFCNT_dec(data.longest_fixed);
4855 longest_fixed_length = 0;
4858 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4859 ri->regstclass = NULL;
4860 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4862 && !(data.start_class->flags & ANYOF_EOS)
4863 && !cl_is_anything(data.start_class))
4865 const U32 n = add_data(pRExC_state, 1, "f");
4867 Newx(RExC_rxi->data->data[n], 1,
4868 struct regnode_charclass_class);
4869 StructCopy(data.start_class,
4870 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4871 struct regnode_charclass_class);
4872 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4873 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4874 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4875 regprop(r, sv, (regnode*)data.start_class);
4876 PerlIO_printf(Perl_debug_log,
4877 "synthetic stclass \"%s\".\n",
4878 SvPVX_const(sv));});
4881 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4882 if (longest_fixed_length > longest_float_length) {
4883 r->check_end_shift = r->anchored_end_shift;
4884 r->check_substr = r->anchored_substr;
4885 r->check_utf8 = r->anchored_utf8;
4886 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4887 if (r->extflags & RXf_ANCH_SINGLE)
4888 r->extflags |= RXf_NOSCAN;
4891 r->check_end_shift = r->float_end_shift;
4892 r->check_substr = r->float_substr;
4893 r->check_utf8 = r->float_utf8;
4894 r->check_offset_min = r->float_min_offset;
4895 r->check_offset_max = r->float_max_offset;
4897 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4898 This should be changed ASAP! */
4899 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4900 r->extflags |= RXf_USE_INTUIT;
4901 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4902 r->extflags |= RXf_INTUIT_TAIL;
4904 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4905 if ( (STRLEN)minlen < longest_float_length )
4906 minlen= longest_float_length;
4907 if ( (STRLEN)minlen < longest_fixed_length )
4908 minlen= longest_fixed_length;
4912 /* Several toplevels. Best we can is to set minlen. */
4914 struct regnode_charclass_class ch_class;
4917 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4919 scan = ri->program + 1;
4920 cl_init(pRExC_state, &ch_class);
4921 data.start_class = &ch_class;
4922 data.last_closep = &last_close;
4925 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4926 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4930 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4931 = r->float_substr = r->float_utf8 = NULL;
4932 if (!(data.start_class->flags & ANYOF_EOS)
4933 && !cl_is_anything(data.start_class))
4935 const U32 n = add_data(pRExC_state, 1, "f");
4937 Newx(RExC_rxi->data->data[n], 1,
4938 struct regnode_charclass_class);
4939 StructCopy(data.start_class,
4940 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4941 struct regnode_charclass_class);
4942 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4943 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4944 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4945 regprop(r, sv, (regnode*)data.start_class);
4946 PerlIO_printf(Perl_debug_log,
4947 "synthetic stclass \"%s\".\n",
4948 SvPVX_const(sv));});
4952 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4953 the "real" pattern. */
4955 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4956 (IV)minlen, (IV)r->minlen);
4958 r->minlenret = minlen;
4959 if (r->minlen < minlen)
4962 if (RExC_seen & REG_SEEN_GPOS)
4963 r->extflags |= RXf_GPOS_SEEN;
4964 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4965 r->extflags |= RXf_LOOKBEHIND_SEEN;
4966 if (RExC_seen & REG_SEEN_EVAL)
4967 r->extflags |= RXf_EVAL_SEEN;
4968 if (RExC_seen & REG_SEEN_CANY)
4969 r->extflags |= RXf_CANY_SEEN;
4970 if (RExC_seen & REG_SEEN_VERBARG)
4971 r->intflags |= PREGf_VERBARG_SEEN;
4972 if (RExC_seen & REG_SEEN_CUTGROUP)
4973 r->intflags |= PREGf_CUTGROUP_SEEN;
4974 if (RExC_paren_names)
4975 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
4977 RXp_PAREN_NAMES(r) = NULL;
4979 #ifdef STUPID_PATTERN_CHECKS
4980 if (RX_PRELEN(rx) == 0)
4981 r->extflags |= RXf_NULL;
4982 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4983 /* XXX: this should happen BEFORE we compile */
4984 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4985 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
4986 r->extflags |= RXf_WHITE;
4987 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
4988 r->extflags |= RXf_START_ONLY;
4990 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4991 /* XXX: this should happen BEFORE we compile */
4992 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4994 regnode *first = ri->program + 1;
4996 U8 nop = OP(NEXTOPER(first));
4998 if (PL_regkind[fop] == NOTHING && nop == END)
4999 r->extflags |= RXf_NULL;
5000 else if (PL_regkind[fop] == BOL && nop == END)
5001 r->extflags |= RXf_START_ONLY;
5002 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
5003 r->extflags |= RXf_WHITE;
5007 if (RExC_paren_names) {
5008 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5009 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5012 ri->name_list_idx = 0;
5014 if (RExC_recurse_count) {
5015 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5016 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5017 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5020 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5021 /* assume we don't need to swap parens around before we match */
5024 PerlIO_printf(Perl_debug_log,"Final program:\n");
5027 #ifdef RE_TRACK_PATTERN_OFFSETS
5028 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5029 const U32 len = ri->u.offsets[0];
5031 GET_RE_DEBUG_FLAGS_DECL;
5032 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5033 for (i = 1; i <= len; i++) {
5034 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5035 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5036 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5038 PerlIO_printf(Perl_debug_log, "\n");
5044 #undef RE_ENGINE_PTR
5048 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5051 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5053 PERL_UNUSED_ARG(value);
5055 if (flags & RXapif_FETCH) {
5056 return reg_named_buff_fetch(rx, key, flags);
5057 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5058 Perl_croak_no_modify(aTHX);
5060 } else if (flags & RXapif_EXISTS) {
5061 return reg_named_buff_exists(rx, key, flags)
5064 } else if (flags & RXapif_REGNAMES) {
5065 return reg_named_buff_all(rx, flags);
5066 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5067 return reg_named_buff_scalar(rx, flags);
5069 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5075 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5078 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5079 PERL_UNUSED_ARG(lastkey);
5081 if (flags & RXapif_FIRSTKEY)
5082 return reg_named_buff_firstkey(rx, flags);
5083 else if (flags & RXapif_NEXTKEY)
5084 return reg_named_buff_nextkey(rx, flags);
5086 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5092 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5095 AV *retarray = NULL;
5097 struct regexp *const rx = (struct regexp *)SvANY(r);
5099 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5101 if (flags & RXapif_ALL)
5104 if (rx && RXp_PAREN_NAMES(rx)) {
5105 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5108 SV* sv_dat=HeVAL(he_str);
5109 I32 *nums=(I32*)SvPVX(sv_dat);
5110 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5111 if ((I32)(rx->nparens) >= nums[i]
5112 && rx->offs[nums[i]].start != -1
5113 && rx->offs[nums[i]].end != -1)
5116 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5120 ret = newSVsv(&PL_sv_undef);
5123 av_push(retarray, ret);
5126 return newRV_noinc(MUTABLE_SV(retarray));
5133 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5136 struct regexp *const rx = (struct regexp *)SvANY(r);
5138 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5140 if (rx && RXp_PAREN_NAMES(rx)) {
5141 if (flags & RXapif_ALL) {
5142 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5144 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5158 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5160 struct regexp *const rx = (struct regexp *)SvANY(r);
5162 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5164 if ( rx && RXp_PAREN_NAMES(rx) ) {
5165 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5167 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5174 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5176 struct regexp *const rx = (struct regexp *)SvANY(r);
5177 GET_RE_DEBUG_FLAGS_DECL;
5179 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5181 if (rx && RXp_PAREN_NAMES(rx)) {
5182 HV *hv = RXp_PAREN_NAMES(rx);
5184 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5187 SV* sv_dat = HeVAL(temphe);
5188 I32 *nums = (I32*)SvPVX(sv_dat);
5189 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5190 if ((I32)(rx->lastparen) >= nums[i] &&
5191 rx->offs[nums[i]].start != -1 &&
5192 rx->offs[nums[i]].end != -1)
5198 if (parno || flags & RXapif_ALL) {
5199 return newSVhek(HeKEY_hek(temphe));
5207 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5212 struct regexp *const rx = (struct regexp *)SvANY(r);
5214 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5216 if (rx && RXp_PAREN_NAMES(rx)) {
5217 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5218 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5219 } else if (flags & RXapif_ONE) {
5220 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5221 av = MUTABLE_AV(SvRV(ret));
5222 length = av_len(av);
5224 return newSViv(length + 1);
5226 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5230 return &PL_sv_undef;
5234 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5236 struct regexp *const rx = (struct regexp *)SvANY(r);
5239 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5241 if (rx && RXp_PAREN_NAMES(rx)) {
5242 HV *hv= RXp_PAREN_NAMES(rx);
5244 (void)hv_iterinit(hv);
5245 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5248 SV* sv_dat = HeVAL(temphe);
5249 I32 *nums = (I32*)SvPVX(sv_dat);
5250 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5251 if ((I32)(rx->lastparen) >= nums[i] &&
5252 rx->offs[nums[i]].start != -1 &&
5253 rx->offs[nums[i]].end != -1)
5259 if (parno || flags & RXapif_ALL) {
5260 av_push(av, newSVhek(HeKEY_hek(temphe)));
5265 return newRV_noinc(MUTABLE_SV(av));
5269 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5272 struct regexp *const rx = (struct regexp *)SvANY(r);
5277 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5280 sv_setsv(sv,&PL_sv_undef);
5284 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5286 i = rx->offs[0].start;
5290 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5292 s = rx->subbeg + rx->offs[0].end;
5293 i = rx->sublen - rx->offs[0].end;
5296 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5297 (s1 = rx->offs[paren].start) != -1 &&
5298 (t1 = rx->offs[paren].end) != -1)
5302 s = rx->subbeg + s1;
5304 sv_setsv(sv,&PL_sv_undef);
5307 assert(rx->sublen >= (s - rx->subbeg) + i );
5309 const int oldtainted = PL_tainted;
5311 sv_setpvn(sv, s, i);
5312 PL_tainted = oldtainted;
5313 if ( (rx->extflags & RXf_CANY_SEEN)
5314 ? (RXp_MATCH_UTF8(rx)
5315 && (!i || is_utf8_string((U8*)s, i)))
5316 : (RXp_MATCH_UTF8(rx)) )
5323 if (RXp_MATCH_TAINTED(rx)) {
5324 if (SvTYPE(sv) >= SVt_PVMG) {
5325 MAGIC* const mg = SvMAGIC(sv);
5328 SvMAGIC_set(sv, mg->mg_moremagic);
5330 if ((mgt = SvMAGIC(sv))) {
5331 mg->mg_moremagic = mgt;
5332 SvMAGIC_set(sv, mg);
5342 sv_setsv(sv,&PL_sv_undef);
5348 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5349 SV const * const value)
5351 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5353 PERL_UNUSED_ARG(rx);
5354 PERL_UNUSED_ARG(paren);
5355 PERL_UNUSED_ARG(value);
5358 Perl_croak_no_modify(aTHX);
5362 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5365 struct regexp *const rx = (struct regexp *)SvANY(r);
5369 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5371 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5373 /* $` / ${^PREMATCH} */
5374 case RX_BUFF_IDX_PREMATCH:
5375 if (rx->offs[0].start != -1) {
5376 i = rx->offs[0].start;
5384 /* $' / ${^POSTMATCH} */
5385 case RX_BUFF_IDX_POSTMATCH:
5386 if (rx->offs[0].end != -1) {
5387 i = rx->sublen - rx->offs[0].end;
5389 s1 = rx->offs[0].end;
5395 /* $& / ${^MATCH}, $1, $2, ... */
5397 if (paren <= (I32)rx->nparens &&
5398 (s1 = rx->offs[paren].start) != -1 &&
5399 (t1 = rx->offs[paren].end) != -1)
5404 if (ckWARN(WARN_UNINITIALIZED))
5405 report_uninit((const SV *)sv);
5410 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5411 const char * const s = rx->subbeg + s1;
5416 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5423 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5425 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5426 PERL_UNUSED_ARG(rx);
5430 return newSVpvs("Regexp");
5433 /* Scans the name of a named buffer from the pattern.
5434 * If flags is REG_RSN_RETURN_NULL returns null.
5435 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5436 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5437 * to the parsed name as looked up in the RExC_paren_names hash.
5438 * If there is an error throws a vFAIL().. type exception.
5441 #define REG_RSN_RETURN_NULL 0
5442 #define REG_RSN_RETURN_NAME 1
5443 #define REG_RSN_RETURN_DATA 2
5446 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5448 char *name_start = RExC_parse;
5450 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5452 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5453 /* skip IDFIRST by using do...while */
5456 RExC_parse += UTF8SKIP(RExC_parse);
5457 } while (isALNUM_utf8((U8*)RExC_parse));
5461 } while (isALNUM(*RExC_parse));
5466 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5467 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5468 if ( flags == REG_RSN_RETURN_NAME)
5470 else if (flags==REG_RSN_RETURN_DATA) {
5473 if ( ! sv_name ) /* should not happen*/
5474 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5475 if (RExC_paren_names)
5476 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5478 sv_dat = HeVAL(he_str);
5480 vFAIL("Reference to nonexistent named group");
5484 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5491 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5492 int rem=(int)(RExC_end - RExC_parse); \
5501 if (RExC_lastparse!=RExC_parse) \
5502 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5505 iscut ? "..." : "<" \
5508 PerlIO_printf(Perl_debug_log,"%16s",""); \
5511 num = RExC_size + 1; \
5513 num=REG_NODE_NUM(RExC_emit); \
5514 if (RExC_lastnum!=num) \
5515 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5517 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5518 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5519 (int)((depth*2)), "", \
5523 RExC_lastparse=RExC_parse; \
5528 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5529 DEBUG_PARSE_MSG((funcname)); \
5530 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5532 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5533 DEBUG_PARSE_MSG((funcname)); \
5534 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5537 - reg - regular expression, i.e. main body or parenthesized thing
5539 * Caller must absorb opening parenthesis.
5541 * Combining parenthesis handling with the base level of regular expression
5542 * is a trifle forced, but the need to tie the tails of the branches to what
5543 * follows makes it hard to avoid.
5545 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5547 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5549 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5553 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5554 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5557 register regnode *ret; /* Will be the head of the group. */
5558 register regnode *br;
5559 register regnode *lastbr;
5560 register regnode *ender = NULL;
5561 register I32 parno = 0;
5563 U32 oregflags = RExC_flags;
5564 bool have_branch = 0;
5566 I32 freeze_paren = 0;
5567 I32 after_freeze = 0;
5569 /* for (?g), (?gc), and (?o) warnings; warning
5570 about (?c) will warn about (?g) -- japhy */
5572 #define WASTED_O 0x01
5573 #define WASTED_G 0x02
5574 #define WASTED_C 0x04
5575 #define WASTED_GC (0x02|0x04)
5576 I32 wastedflags = 0x00;
5578 char * parse_start = RExC_parse; /* MJD */
5579 char * const oregcomp_parse = RExC_parse;
5581 GET_RE_DEBUG_FLAGS_DECL;
5583 PERL_ARGS_ASSERT_REG;
5584 DEBUG_PARSE("reg ");
5586 *flagp = 0; /* Tentatively. */
5589 /* Make an OPEN node, if parenthesized. */
5591 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5592 char *start_verb = RExC_parse;
5593 STRLEN verb_len = 0;
5594 char *start_arg = NULL;
5595 unsigned char op = 0;
5597 int internal_argval = 0; /* internal_argval is only useful if !argok */
5598 while ( *RExC_parse && *RExC_parse != ')' ) {
5599 if ( *RExC_parse == ':' ) {
5600 start_arg = RExC_parse + 1;
5606 verb_len = RExC_parse - start_verb;
5609 while ( *RExC_parse && *RExC_parse != ')' )
5611 if ( *RExC_parse != ')' )
5612 vFAIL("Unterminated verb pattern argument");
5613 if ( RExC_parse == start_arg )
5616 if ( *RExC_parse != ')' )
5617 vFAIL("Unterminated verb pattern");
5620 switch ( *start_verb ) {
5621 case 'A': /* (*ACCEPT) */
5622 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5624 internal_argval = RExC_nestroot;
5627 case 'C': /* (*COMMIT) */
5628 if ( memEQs(start_verb,verb_len,"COMMIT") )
5631 case 'F': /* (*FAIL) */
5632 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5637 case ':': /* (*:NAME) */
5638 case 'M': /* (*MARK:NAME) */
5639 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5644 case 'P': /* (*PRUNE) */
5645 if ( memEQs(start_verb,verb_len,"PRUNE") )
5648 case 'S': /* (*SKIP) */
5649 if ( memEQs(start_verb,verb_len,"SKIP") )
5652 case 'T': /* (*THEN) */
5653 /* [19:06] <TimToady> :: is then */
5654 if ( memEQs(start_verb,verb_len,"THEN") ) {
5656 RExC_seen |= REG_SEEN_CUTGROUP;
5662 vFAIL3("Unknown verb pattern '%.*s'",
5663 verb_len, start_verb);
5666 if ( start_arg && internal_argval ) {
5667 vFAIL3("Verb pattern '%.*s' may not have an argument",
5668 verb_len, start_verb);
5669 } else if ( argok < 0 && !start_arg ) {
5670 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5671 verb_len, start_verb);
5673 ret = reganode(pRExC_state, op, internal_argval);
5674 if ( ! internal_argval && ! SIZE_ONLY ) {
5676 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5677 ARG(ret) = add_data( pRExC_state, 1, "S" );
5678 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5685 if (!internal_argval)
5686 RExC_seen |= REG_SEEN_VERBARG;
5687 } else if ( start_arg ) {
5688 vFAIL3("Verb pattern '%.*s' may not have an argument",
5689 verb_len, start_verb);
5691 ret = reg_node(pRExC_state, op);
5693 nextchar(pRExC_state);
5696 if (*RExC_parse == '?') { /* (?...) */
5697 bool is_logical = 0;
5698 const char * const seqstart = RExC_parse;
5699 bool has_use_defaults = FALSE;
5702 paren = *RExC_parse++;
5703 ret = NULL; /* For look-ahead/behind. */
5706 case 'P': /* (?P...) variants for those used to PCRE/Python */
5707 paren = *RExC_parse++;
5708 if ( paren == '<') /* (?P<...>) named capture */
5710 else if (paren == '>') { /* (?P>name) named recursion */
5711 goto named_recursion;
5713 else if (paren == '=') { /* (?P=...) named backref */
5714 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5715 you change this make sure you change that */
5716 char* name_start = RExC_parse;
5718 SV *sv_dat = reg_scan_name(pRExC_state,
5719 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5720 if (RExC_parse == name_start || *RExC_parse != ')')
5721 vFAIL2("Sequence %.3s... not terminated",parse_start);
5724 num = add_data( pRExC_state, 1, "S" );
5725 RExC_rxi->data->data[num]=(void*)sv_dat;
5726 SvREFCNT_inc_simple_void(sv_dat);
5729 ret = reganode(pRExC_state,
5730 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5734 Set_Node_Offset(ret, parse_start+1);
5735 Set_Node_Cur_Length(ret); /* MJD */
5737 nextchar(pRExC_state);
5741 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5743 case '<': /* (?<...) */
5744 if (*RExC_parse == '!')
5746 else if (*RExC_parse != '=')
5752 case '\'': /* (?'...') */
5753 name_start= RExC_parse;
5754 svname = reg_scan_name(pRExC_state,
5755 SIZE_ONLY ? /* reverse test from the others */
5756 REG_RSN_RETURN_NAME :
5757 REG_RSN_RETURN_NULL);
5758 if (RExC_parse == name_start) {
5760 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5763 if (*RExC_parse != paren)
5764 vFAIL2("Sequence (?%c... not terminated",
5765 paren=='>' ? '<' : paren);
5769 if (!svname) /* shouldnt happen */
5771 "panic: reg_scan_name returned NULL");
5772 if (!RExC_paren_names) {
5773 RExC_paren_names= newHV();
5774 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5776 RExC_paren_name_list= newAV();
5777 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5780 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5782 sv_dat = HeVAL(he_str);
5784 /* croak baby croak */
5786 "panic: paren_name hash element allocation failed");
5787 } else if ( SvPOK(sv_dat) ) {
5788 /* (?|...) can mean we have dupes so scan to check
5789 its already been stored. Maybe a flag indicating
5790 we are inside such a construct would be useful,
5791 but the arrays are likely to be quite small, so
5792 for now we punt -- dmq */
5793 IV count = SvIV(sv_dat);
5794 I32 *pv = (I32*)SvPVX(sv_dat);
5796 for ( i = 0 ; i < count ; i++ ) {
5797 if ( pv[i] == RExC_npar ) {
5803 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5804 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5805 pv[count] = RExC_npar;
5806 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5809 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5810 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5812 SvIV_set(sv_dat, 1);
5815 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5816 SvREFCNT_dec(svname);
5819 /*sv_dump(sv_dat);*/
5821 nextchar(pRExC_state);
5823 goto capturing_parens;
5825 RExC_seen |= REG_SEEN_LOOKBEHIND;
5827 case '=': /* (?=...) */
5828 RExC_seen_zerolen++;
5830 case '!': /* (?!...) */
5831 RExC_seen_zerolen++;
5832 if (*RExC_parse == ')') {
5833 ret=reg_node(pRExC_state, OPFAIL);
5834 nextchar(pRExC_state);
5838 case '|': /* (?|...) */
5839 /* branch reset, behave like a (?:...) except that
5840 buffers in alternations share the same numbers */
5842 after_freeze = freeze_paren = RExC_npar;
5844 case ':': /* (?:...) */
5845 case '>': /* (?>...) */
5847 case '$': /* (?$...) */
5848 case '@': /* (?@...) */
5849 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5851 case '#': /* (?#...) */
5852 while (*RExC_parse && *RExC_parse != ')')
5854 if (*RExC_parse != ')')
5855 FAIL("Sequence (?#... not terminated");
5856 nextchar(pRExC_state);
5859 case '0' : /* (?0) */
5860 case 'R' : /* (?R) */
5861 if (*RExC_parse != ')')
5862 FAIL("Sequence (?R) not terminated");
5863 ret = reg_node(pRExC_state, GOSTART);
5864 *flagp |= POSTPONED;
5865 nextchar(pRExC_state);
5868 { /* named and numeric backreferences */
5870 case '&': /* (?&NAME) */
5871 parse_start = RExC_parse - 1;
5874 SV *sv_dat = reg_scan_name(pRExC_state,
5875 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5876 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5878 goto gen_recurse_regop;
5881 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5883 vFAIL("Illegal pattern");
5885 goto parse_recursion;
5887 case '-': /* (?-1) */
5888 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5889 RExC_parse--; /* rewind to let it be handled later */
5893 case '1': case '2': case '3': case '4': /* (?1) */
5894 case '5': case '6': case '7': case '8': case '9':
5897 num = atoi(RExC_parse);
5898 parse_start = RExC_parse - 1; /* MJD */
5899 if (*RExC_parse == '-')
5901 while (isDIGIT(*RExC_parse))
5903 if (*RExC_parse!=')')
5904 vFAIL("Expecting close bracket");
5907 if ( paren == '-' ) {
5909 Diagram of capture buffer numbering.
5910 Top line is the normal capture buffer numbers
5911 Botton line is the negative indexing as from
5915 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5919 num = RExC_npar + num;
5922 vFAIL("Reference to nonexistent group");
5924 } else if ( paren == '+' ) {
5925 num = RExC_npar + num - 1;
5928 ret = reganode(pRExC_state, GOSUB, num);
5930 if (num > (I32)RExC_rx->nparens) {
5932 vFAIL("Reference to nonexistent group");
5934 ARG2L_SET( ret, RExC_recurse_count++);
5936 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5937 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5941 RExC_seen |= REG_SEEN_RECURSE;
5942 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5943 Set_Node_Offset(ret, parse_start); /* MJD */
5945 *flagp |= POSTPONED;
5946 nextchar(pRExC_state);
5948 } /* named and numeric backreferences */
5951 case '?': /* (??...) */
5953 if (*RExC_parse != '{') {
5955 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5958 *flagp |= POSTPONED;
5959 paren = *RExC_parse++;
5961 case '{': /* (?{...}) */
5966 char *s = RExC_parse;
5968 RExC_seen_zerolen++;
5969 RExC_seen |= REG_SEEN_EVAL;
5970 while (count && (c = *RExC_parse)) {
5981 if (*RExC_parse != ')') {
5983 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5987 OP_4tree *sop, *rop;
5988 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5991 Perl_save_re_context(aTHX);
5992 rop = sv_compile_2op(sv, &sop, "re", &pad);
5993 sop->op_private |= OPpREFCOUNTED;
5994 /* re_dup will OpREFCNT_inc */
5995 OpREFCNT_set(sop, 1);
5998 n = add_data(pRExC_state, 3, "nop");
5999 RExC_rxi->data->data[n] = (void*)rop;
6000 RExC_rxi->data->data[n+1] = (void*)sop;
6001 RExC_rxi->data->data[n+2] = (void*)pad;
6004 else { /* First pass */
6005 if (PL_reginterp_cnt < ++RExC_seen_evals
6007 /* No compiled RE interpolated, has runtime
6008 components ===> unsafe. */
6009 FAIL("Eval-group not allowed at runtime, use re 'eval'");
6010 if (PL_tainting && PL_tainted)
6011 FAIL("Eval-group in insecure regular expression");
6012 #if PERL_VERSION > 8
6013 if (IN_PERL_COMPILETIME)
6018 nextchar(pRExC_state);
6020 ret = reg_node(pRExC_state, LOGICAL);
6023 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
6024 /* deal with the length of this later - MJD */
6027 ret = reganode(pRExC_state, EVAL, n);
6028 Set_Node_Length(ret, RExC_parse - parse_start + 1);
6029 Set_Node_Offset(ret, parse_start);
6032 case '(': /* (?(?{...})...) and (?(?=...)...) */
6035 if (RExC_parse[0] == '?') { /* (?(?...)) */
6036 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
6037 || RExC_parse[1] == '<'
6038 || RExC_parse[1] == '{') { /* Lookahead or eval. */
6041 ret = reg_node(pRExC_state, LOGICAL);
6044 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
6048 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
6049 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
6051 char ch = RExC_parse[0] == '<' ? '>' : '\'';
6052 char *name_start= RExC_parse++;
6054 SV *sv_dat=reg_scan_name(pRExC_state,
6055 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6056 if (RExC_parse == name_start || *RExC_parse != ch)
6057 vFAIL2("Sequence (?(%c... not terminated",
6058 (ch == '>' ? '<' : ch));
6061 num = add_data( pRExC_state, 1, "S" );
6062 RExC_rxi->data->data[num]=(void*)sv_dat;
6063 SvREFCNT_inc_simple_void(sv_dat);
6065 ret = reganode(pRExC_state,NGROUPP,num);
6066 goto insert_if_check_paren;
6068 else if (RExC_parse[0] == 'D' &&
6069 RExC_parse[1] == 'E' &&
6070 RExC_parse[2] == 'F' &&
6071 RExC_parse[3] == 'I' &&
6072 RExC_parse[4] == 'N' &&
6073 RExC_parse[5] == 'E')
6075 ret = reganode(pRExC_state,DEFINEP,0);
6078 goto insert_if_check_paren;
6080 else if (RExC_parse[0] == 'R') {
6083 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6084 parno = atoi(RExC_parse++);
6085 while (isDIGIT(*RExC_parse))
6087 } else if (RExC_parse[0] == '&') {
6090 sv_dat = reg_scan_name(pRExC_state,
6091 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6092 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6094 ret = reganode(pRExC_state,INSUBP,parno);
6095 goto insert_if_check_paren;
6097 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6100 parno = atoi(RExC_parse++);
6102 while (isDIGIT(*RExC_parse))
6104 ret = reganode(pRExC_state, GROUPP, parno);
6106 insert_if_check_paren:
6107 if ((c = *nextchar(pRExC_state)) != ')')
6108 vFAIL("Switch condition not recognized");
6110 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
6111 br = regbranch(pRExC_state, &flags, 1,depth+1);
6113 br = reganode(pRExC_state, LONGJMP, 0);
6115 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
6116 c = *nextchar(pRExC_state);
6121 vFAIL("(?(DEFINE)....) does not allow branches");
6122 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6123 regbranch(pRExC_state, &flags, 1,depth+1);
6124 REGTAIL(pRExC_state, ret, lastbr);
6127 c = *nextchar(pRExC_state);
6132 vFAIL("Switch (?(condition)... contains too many branches");
6133 ender = reg_node(pRExC_state, TAIL);
6134 REGTAIL(pRExC_state, br, ender);
6136 REGTAIL(pRExC_state, lastbr, ender);
6137 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6140 REGTAIL(pRExC_state, ret, ender);
6141 RExC_size++; /* XXX WHY do we need this?!!
6142 For large programs it seems to be required
6143 but I can't figure out why. -- dmq*/
6147 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6151 RExC_parse--; /* for vFAIL to print correctly */
6152 vFAIL("Sequence (? incomplete");
6154 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
6156 has_use_defaults = TRUE;
6157 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
6158 RExC_flags &= ~(RXf_PMf_LOCALE|RXf_PMf_UNICODE);
6162 parse_flags: /* (?i) */
6164 U32 posflags = 0, negflags = 0;
6165 U32 *flagsp = &posflags;
6166 bool has_charset_modifier = 0;
6168 while (*RExC_parse) {
6169 /* && strchr("iogcmsx", *RExC_parse) */
6170 /* (?g), (?gc) and (?o) are useless here
6171 and must be globally applied -- japhy */
6172 switch (*RExC_parse) {
6173 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6174 case LOCALE_PAT_MOD:
6175 if (has_charset_modifier || flagsp == &negflags) {
6176 goto fail_modifiers;
6178 *flagsp &= ~RXf_PMf_UNICODE;
6179 *flagsp |= RXf_PMf_LOCALE;
6180 has_charset_modifier = 1;
6182 case UNICODE_PAT_MOD:
6183 if (has_charset_modifier || flagsp == &negflags) {
6184 goto fail_modifiers;
6186 *flagsp &= ~RXf_PMf_LOCALE;
6187 *flagsp |= RXf_PMf_UNICODE;
6188 has_charset_modifier = 1;
6191 if (has_use_defaults
6192 || has_charset_modifier
6193 || flagsp == &negflags)
6195 goto fail_modifiers;
6197 *flagsp &= ~(RXf_PMf_LOCALE|RXf_PMf_UNICODE);
6198 has_charset_modifier = 1;
6200 case ONCE_PAT_MOD: /* 'o' */
6201 case GLOBAL_PAT_MOD: /* 'g' */
6202 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6203 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6204 if (! (wastedflags & wflagbit) ) {
6205 wastedflags |= wflagbit;
6208 "Useless (%s%c) - %suse /%c modifier",
6209 flagsp == &negflags ? "?-" : "?",
6211 flagsp == &negflags ? "don't " : "",
6218 case CONTINUE_PAT_MOD: /* 'c' */
6219 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6220 if (! (wastedflags & WASTED_C) ) {
6221 wastedflags |= WASTED_GC;
6224 "Useless (%sc) - %suse /gc modifier",
6225 flagsp == &negflags ? "?-" : "?",
6226 flagsp == &negflags ? "don't " : ""
6231 case KEEPCOPY_PAT_MOD: /* 'p' */
6232 if (flagsp == &negflags) {
6234 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
6236 *flagsp |= RXf_PMf_KEEPCOPY;
6240 /* A flag is a default iff it is following a minus, so
6241 * if there is a minus, it means will be trying to
6242 * re-specify a default which is an error */
6243 if (has_use_defaults || flagsp == &negflags) {
6246 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6250 wastedflags = 0; /* reset so (?g-c) warns twice */
6256 RExC_flags |= posflags;
6257 RExC_flags &= ~negflags;
6259 oregflags |= posflags;
6260 oregflags &= ~negflags;
6262 nextchar(pRExC_state);
6273 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6278 }} /* one for the default block, one for the switch */
6285 ret = reganode(pRExC_state, OPEN, parno);
6288 RExC_nestroot = parno;
6289 if (RExC_seen & REG_SEEN_RECURSE
6290 && !RExC_open_parens[parno-1])
6292 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6293 "Setting open paren #%"IVdf" to %d\n",
6294 (IV)parno, REG_NODE_NUM(ret)));
6295 RExC_open_parens[parno-1]= ret;
6298 Set_Node_Length(ret, 1); /* MJD */
6299 Set_Node_Offset(ret, RExC_parse); /* MJD */
6307 /* Pick up the branches, linking them together. */
6308 parse_start = RExC_parse; /* MJD */
6309 br = regbranch(pRExC_state, &flags, 1,depth+1);
6312 if (RExC_npar > after_freeze)
6313 after_freeze = RExC_npar;
6314 RExC_npar = freeze_paren;
6317 /* branch_len = (paren != 0); */
6321 if (*RExC_parse == '|') {
6322 if (!SIZE_ONLY && RExC_extralen) {
6323 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6326 reginsert(pRExC_state, BRANCH, br, depth+1);
6327 Set_Node_Length(br, paren != 0);
6328 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6332 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6334 else if (paren == ':') {
6335 *flagp |= flags&SIMPLE;
6337 if (is_open) { /* Starts with OPEN. */
6338 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6340 else if (paren != '?') /* Not Conditional */
6342 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6344 while (*RExC_parse == '|') {
6345 if (!SIZE_ONLY && RExC_extralen) {
6346 ender = reganode(pRExC_state, LONGJMP,0);
6347 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6350 RExC_extralen += 2; /* Account for LONGJMP. */
6351 nextchar(pRExC_state);
6353 if (RExC_npar > after_freeze)
6354 after_freeze = RExC_npar;
6355 RExC_npar = freeze_paren;
6357 br = regbranch(pRExC_state, &flags, 0, depth+1);
6361 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6363 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6366 if (have_branch || paren != ':') {
6367 /* Make a closing node, and hook it on the end. */
6370 ender = reg_node(pRExC_state, TAIL);
6373 ender = reganode(pRExC_state, CLOSE, parno);
6374 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6375 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6376 "Setting close paren #%"IVdf" to %d\n",
6377 (IV)parno, REG_NODE_NUM(ender)));
6378 RExC_close_parens[parno-1]= ender;
6379 if (RExC_nestroot == parno)
6382 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6383 Set_Node_Length(ender,1); /* MJD */
6389 *flagp &= ~HASWIDTH;
6392 ender = reg_node(pRExC_state, SUCCEED);
6395 ender = reg_node(pRExC_state, END);
6397 assert(!RExC_opend); /* there can only be one! */
6402 REGTAIL(pRExC_state, lastbr, ender);
6404 if (have_branch && !SIZE_ONLY) {
6406 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6408 /* Hook the tails of the branches to the closing node. */
6409 for (br = ret; br; br = regnext(br)) {
6410 const U8 op = PL_regkind[OP(br)];
6412 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6414 else if (op == BRANCHJ) {
6415 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6423 static const char parens[] = "=!<,>";
6425 if (paren && (p = strchr(parens, paren))) {
6426 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6427 int flag = (p - parens) > 1;
6430 node = SUSPEND, flag = 0;
6431 reginsert(pRExC_state, node,ret, depth+1);
6432 Set_Node_Cur_Length(ret);
6433 Set_Node_Offset(ret, parse_start + 1);
6435 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6439 /* Check for proper termination. */
6441 RExC_flags = oregflags;
6442 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6443 RExC_parse = oregcomp_parse;
6444 vFAIL("Unmatched (");
6447 else if (!paren && RExC_parse < RExC_end) {
6448 if (*RExC_parse == ')') {
6450 vFAIL("Unmatched )");
6453 FAIL("Junk on end of regexp"); /* "Can't happen". */
6457 RExC_npar = after_freeze;
6462 - regbranch - one alternative of an | operator
6464 * Implements the concatenation operator.
6467 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6470 register regnode *ret;
6471 register regnode *chain = NULL;
6472 register regnode *latest;
6473 I32 flags = 0, c = 0;
6474 GET_RE_DEBUG_FLAGS_DECL;
6476 PERL_ARGS_ASSERT_REGBRANCH;
6478 DEBUG_PARSE("brnc");
6483 if (!SIZE_ONLY && RExC_extralen)
6484 ret = reganode(pRExC_state, BRANCHJ,0);
6486 ret = reg_node(pRExC_state, BRANCH);
6487 Set_Node_Length(ret, 1);
6491 if (!first && SIZE_ONLY)
6492 RExC_extralen += 1; /* BRANCHJ */
6494 *flagp = WORST; /* Tentatively. */
6497 nextchar(pRExC_state);
6498 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6500 latest = regpiece(pRExC_state, &flags,depth+1);
6501 if (latest == NULL) {
6502 if (flags & TRYAGAIN)
6506 else if (ret == NULL)
6508 *flagp |= flags&(HASWIDTH|POSTPONED);
6509 if (chain == NULL) /* First piece. */
6510 *flagp |= flags&SPSTART;
6513 REGTAIL(pRExC_state, chain, latest);
6518 if (chain == NULL) { /* Loop ran zero times. */
6519 chain = reg_node(pRExC_state, NOTHING);
6524 *flagp |= flags&SIMPLE;
6531 - regpiece - something followed by possible [*+?]
6533 * Note that the branching code sequences used for ? and the general cases
6534 * of * and + are somewhat optimized: they use the same NOTHING node as
6535 * both the endmarker for their branch list and the body of the last branch.
6536 * It might seem that this node could be dispensed with entirely, but the
6537 * endmarker role is not redundant.
6540 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6543 register regnode *ret;
6545 register char *next;
6547 const char * const origparse = RExC_parse;
6549 I32 max = REG_INFTY;
6551 const char *maxpos = NULL;
6552 GET_RE_DEBUG_FLAGS_DECL;
6554 PERL_ARGS_ASSERT_REGPIECE;
6556 DEBUG_PARSE("piec");
6558 ret = regatom(pRExC_state, &flags,depth+1);
6560 if (flags & TRYAGAIN)
6567 if (op == '{' && regcurly(RExC_parse)) {
6569 parse_start = RExC_parse; /* MJD */
6570 next = RExC_parse + 1;
6571 while (isDIGIT(*next) || *next == ',') {
6580 if (*next == '}') { /* got one */
6584 min = atoi(RExC_parse);
6588 maxpos = RExC_parse;
6590 if (!max && *maxpos != '0')
6591 max = REG_INFTY; /* meaning "infinity" */
6592 else if (max >= REG_INFTY)
6593 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6595 nextchar(pRExC_state);
6598 if ((flags&SIMPLE)) {
6599 RExC_naughty += 2 + RExC_naughty / 2;
6600 reginsert(pRExC_state, CURLY, ret, depth+1);
6601 Set_Node_Offset(ret, parse_start+1); /* MJD */
6602 Set_Node_Cur_Length(ret);
6605 regnode * const w = reg_node(pRExC_state, WHILEM);
6608 REGTAIL(pRExC_state, ret, w);
6609 if (!SIZE_ONLY && RExC_extralen) {
6610 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6611 reginsert(pRExC_state, NOTHING,ret, depth+1);
6612 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6614 reginsert(pRExC_state, CURLYX,ret, depth+1);
6616 Set_Node_Offset(ret, parse_start+1);
6617 Set_Node_Length(ret,
6618 op == '{' ? (RExC_parse - parse_start) : 1);
6620 if (!SIZE_ONLY && RExC_extralen)
6621 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6622 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6624 RExC_whilem_seen++, RExC_extralen += 3;
6625 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6634 vFAIL("Can't do {n,m} with n > m");
6636 ARG1_SET(ret, (U16)min);
6637 ARG2_SET(ret, (U16)max);
6649 #if 0 /* Now runtime fix should be reliable. */
6651 /* if this is reinstated, don't forget to put this back into perldiag:
6653 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6655 (F) The part of the regexp subject to either the * or + quantifier
6656 could match an empty string. The {#} shows in the regular
6657 expression about where the problem was discovered.
6661 if (!(flags&HASWIDTH) && op != '?')
6662 vFAIL("Regexp *+ operand could be empty");
6665 parse_start = RExC_parse;
6666 nextchar(pRExC_state);
6668 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6670 if (op == '*' && (flags&SIMPLE)) {
6671 reginsert(pRExC_state, STAR, ret, depth+1);
6675 else if (op == '*') {
6679 else if (op == '+' && (flags&SIMPLE)) {
6680 reginsert(pRExC_state, PLUS, ret, depth+1);
6684 else if (op == '+') {
6688 else if (op == '?') {
6693 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
6694 ckWARN3reg(RExC_parse,
6695 "%.*s matches null string many times",
6696 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6700 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6701 nextchar(pRExC_state);
6702 reginsert(pRExC_state, MINMOD, ret, depth+1);
6703 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6705 #ifndef REG_ALLOW_MINMOD_SUSPEND
6708 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6710 nextchar(pRExC_state);
6711 ender = reg_node(pRExC_state, SUCCEED);
6712 REGTAIL(pRExC_state, ret, ender);
6713 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6715 ender = reg_node(pRExC_state, TAIL);
6716 REGTAIL(pRExC_state, ret, ender);
6720 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6722 vFAIL("Nested quantifiers");
6729 /* reg_namedseq(pRExC_state,UVp)
6731 This is expected to be called by a parser routine that has
6732 recognized '\N' and needs to handle the rest. RExC_parse is
6733 expected to point at the first char following the N at the time
6736 The \N may be inside (indicated by valuep not being NULL) or outside a
6739 \N may begin either a named sequence, or if outside a character class, mean
6740 to match a non-newline. For non single-quoted regexes, the tokenizer has
6741 attempted to decide which, and in the case of a named sequence converted it
6742 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
6743 where c1... are the characters in the sequence. For single-quoted regexes,
6744 the tokenizer passes the \N sequence through unchanged; this code will not
6745 attempt to determine this nor expand those. The net effect is that if the
6746 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
6747 signals that this \N occurrence means to match a non-newline.
6749 Only the \N{U+...} form should occur in a character class, for the same
6750 reason that '.' inside a character class means to just match a period: it
6751 just doesn't make sense.
6753 If valuep is non-null then it is assumed that we are parsing inside
6754 of a charclass definition and the first codepoint in the resolved
6755 string is returned via *valuep and the routine will return NULL.
6756 In this mode if a multichar string is returned from the charnames
6757 handler, a warning will be issued, and only the first char in the
6758 sequence will be examined. If the string returned is zero length
6759 then the value of *valuep is undefined and NON-NULL will
6760 be returned to indicate failure. (This will NOT be a valid pointer
6763 If valuep is null then it is assumed that we are parsing normal text and a
6764 new EXACT node is inserted into the program containing the resolved string,
6765 and a pointer to the new node is returned. But if the string is zero length
6766 a NOTHING node is emitted instead.
6768 On success RExC_parse is set to the char following the endbrace.
6769 Parsing failures will generate a fatal error via vFAIL(...)
6772 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp)
6774 char * endbrace; /* '}' following the name */
6775 regnode *ret = NULL;
6777 char* parse_start = RExC_parse - 2; /* points to the '\N' */
6781 GET_RE_DEBUG_FLAGS_DECL;
6783 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6787 /* The [^\n] meaning of \N ignores spaces and comments under the /x
6788 * modifier. The other meaning does not */
6789 p = (RExC_flags & RXf_PMf_EXTENDED)
6790 ? regwhite( pRExC_state, RExC_parse )
6793 /* Disambiguate between \N meaning a named character versus \N meaning
6794 * [^\n]. The former is assumed when it can't be the latter. */
6795 if (*p != '{' || regcurly(p)) {
6798 /* no bare \N in a charclass */
6799 vFAIL("\\N in a character class must be a named character: \\N{...}");
6801 nextchar(pRExC_state);
6802 ret = reg_node(pRExC_state, REG_ANY);
6803 *flagp |= HASWIDTH|SIMPLE;
6806 Set_Node_Length(ret, 1); /* MJD */
6810 /* Here, we have decided it should be a named sequence */
6812 /* The test above made sure that the next real character is a '{', but
6813 * under the /x modifier, it could be separated by space (or a comment and
6814 * \n) and this is not allowed (for consistency with \x{...} and the
6815 * tokenizer handling of \N{NAME}). */
6816 if (*RExC_parse != '{') {
6817 vFAIL("Missing braces on \\N{}");
6820 RExC_parse++; /* Skip past the '{' */
6822 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
6823 || ! (endbrace == RExC_parse /* nothing between the {} */
6824 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
6825 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
6827 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
6828 vFAIL("\\N{NAME} must be resolved by the lexer");
6831 if (endbrace == RExC_parse) { /* empty: \N{} */
6833 RExC_parse = endbrace + 1;
6834 return reg_node(pRExC_state,NOTHING);
6838 ckWARNreg(RExC_parse,
6839 "Ignoring zero length \\N{} in character class"
6841 RExC_parse = endbrace + 1;
6844 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
6847 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
6848 RExC_parse += 2; /* Skip past the 'U+' */
6850 if (valuep) { /* In a bracketed char class */
6851 /* We only pay attention to the first char of
6852 multichar strings being returned. I kinda wonder
6853 if this makes sense as it does change the behaviour
6854 from earlier versions, OTOH that behaviour was broken
6855 as well. XXX Solution is to recharacterize as
6856 [rest-of-class]|multi1|multi2... */
6858 STRLEN length_of_hex;
6859 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6860 | PERL_SCAN_DISALLOW_PREFIX
6861 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6863 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
6864 if (endchar < endbrace) {
6865 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
6868 length_of_hex = (STRLEN)(endchar - RExC_parse);
6869 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
6871 /* The tokenizer should have guaranteed validity, but it's possible to
6872 * bypass it by using single quoting, so check */
6873 if (length_of_hex == 0
6874 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6876 RExC_parse += length_of_hex; /* Includes all the valid */
6877 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6878 ? UTF8SKIP(RExC_parse)
6880 /* Guard against malformed utf8 */
6881 if (RExC_parse >= endchar) RExC_parse = endchar;
6882 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6885 RExC_parse = endbrace + 1;
6886 if (endchar == endbrace) return NULL;
6888 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
6890 else { /* Not a char class */
6891 char *s; /* String to put in generated EXACT node */
6892 STRLEN len = 0; /* Its current byte length */
6893 char *endchar; /* Points to '.' or '}' ending cur char in the input
6896 ret = reg_node(pRExC_state,
6897 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6900 /* Exact nodes can hold only a U8 length's of text = 255. Loop through
6901 * the input which is of the form now 'c1.c2.c3...}' until find the
6902 * ending brace or exceed length 255. The characters that exceed this
6903 * limit are dropped. The limit could be relaxed should it become
6904 * desirable by reparsing this as (?:\N{NAME}), so could generate
6905 * multiple EXACT nodes, as is done for just regular input. But this
6906 * is primarily a named character, and not intended to be a huge long
6907 * string, so 255 bytes should be good enough */
6909 STRLEN length_of_hex;
6910 I32 grok_flags = PERL_SCAN_ALLOW_UNDERSCORES
6911 | PERL_SCAN_DISALLOW_PREFIX
6912 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6913 UV cp; /* Ord of current character */
6915 /* Code points are separated by dots. If none, there is only one
6916 * code point, and is terminated by the brace */
6917 endchar = RExC_parse + strcspn(RExC_parse, ".}");
6919 /* The values are Unicode even on EBCDIC machines */
6920 length_of_hex = (STRLEN)(endchar - RExC_parse);
6921 cp = grok_hex(RExC_parse, &length_of_hex, &grok_flags, NULL);
6922 if ( length_of_hex == 0
6923 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6925 RExC_parse += length_of_hex; /* Includes all the valid */
6926 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6927 ? UTF8SKIP(RExC_parse)
6929 /* Guard against malformed utf8 */
6930 if (RExC_parse >= endchar) RExC_parse = endchar;
6931 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6934 if (! FOLD) { /* Not folding, just append to the string */
6937 /* Quit before adding this character if would exceed limit */
6938 if (len + UNISKIP(cp) > U8_MAX) break;
6940 unilen = reguni(pRExC_state, cp, s);
6945 } else { /* Folding, output the folded equivalent */
6946 STRLEN foldlen,numlen;
6947 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6948 cp = toFOLD_uni(cp, tmpbuf, &foldlen);
6950 /* Quit before exceeding size limit */
6951 if (len + foldlen > U8_MAX) break;
6953 for (foldbuf = tmpbuf;
6957 cp = utf8_to_uvchr(foldbuf, &numlen);
6959 const STRLEN unilen = reguni(pRExC_state, cp, s);
6962 /* In EBCDIC the numlen and unilen can differ. */
6964 if (numlen >= foldlen)
6968 break; /* "Can't happen." */
6972 /* Point to the beginning of the next character in the sequence. */
6973 RExC_parse = endchar + 1;
6975 /* Quit if no more characters */
6976 if (RExC_parse >= endbrace) break;
6981 if (RExC_parse < endbrace) {
6982 ckWARNreg(RExC_parse - 1,
6983 "Using just the first characters returned by \\N{}");
6986 RExC_size += STR_SZ(len);
6989 RExC_emit += STR_SZ(len);
6992 RExC_parse = endbrace + 1;
6994 *flagp |= HASWIDTH; /* Not SIMPLE, as that causes the engine to fail
6995 with malformed in t/re/pat_advanced.t */
6997 Set_Node_Cur_Length(ret); /* MJD */
6998 nextchar(pRExC_state);
7008 * It returns the code point in utf8 for the value in *encp.
7009 * value: a code value in the source encoding
7010 * encp: a pointer to an Encode object
7012 * If the result from Encode is not a single character,
7013 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
7016 S_reg_recode(pTHX_ const char value, SV **encp)
7019 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
7020 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
7021 const STRLEN newlen = SvCUR(sv);
7022 UV uv = UNICODE_REPLACEMENT;
7024 PERL_ARGS_ASSERT_REG_RECODE;
7028 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
7031 if (!newlen || numlen != newlen) {
7032 uv = UNICODE_REPLACEMENT;
7040 - regatom - the lowest level
7042 Try to identify anything special at the start of the pattern. If there
7043 is, then handle it as required. This may involve generating a single regop,
7044 such as for an assertion; or it may involve recursing, such as to
7045 handle a () structure.
7047 If the string doesn't start with something special then we gobble up
7048 as much literal text as we can.
7050 Once we have been able to handle whatever type of thing started the
7051 sequence, we return.
7053 Note: we have to be careful with escapes, as they can be both literal
7054 and special, and in the case of \10 and friends can either, depending
7055 on context. Specifically there are two seperate switches for handling
7056 escape sequences, with the one for handling literal escapes requiring
7057 a dummy entry for all of the special escapes that are actually handled
7062 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7065 register regnode *ret = NULL;
7067 char *parse_start = RExC_parse;
7068 GET_RE_DEBUG_FLAGS_DECL;
7069 DEBUG_PARSE("atom");
7070 *flagp = WORST; /* Tentatively. */
7072 PERL_ARGS_ASSERT_REGATOM;
7075 switch ((U8)*RExC_parse) {
7077 RExC_seen_zerolen++;
7078 nextchar(pRExC_state);
7079 if (RExC_flags & RXf_PMf_MULTILINE)
7080 ret = reg_node(pRExC_state, MBOL);
7081 else if (RExC_flags & RXf_PMf_SINGLELINE)
7082 ret = reg_node(pRExC_state, SBOL);
7084 ret = reg_node(pRExC_state, BOL);
7085 Set_Node_Length(ret, 1); /* MJD */
7088 nextchar(pRExC_state);
7090 RExC_seen_zerolen++;
7091 if (RExC_flags & RXf_PMf_MULTILINE)
7092 ret = reg_node(pRExC_state, MEOL);
7093 else if (RExC_flags & RXf_PMf_SINGLELINE)
7094 ret = reg_node(pRExC_state, SEOL);
7096 ret = reg_node(pRExC_state, EOL);
7097 Set_Node_Length(ret, 1); /* MJD */
7100 nextchar(pRExC_state);
7101 if (RExC_flags & RXf_PMf_SINGLELINE)
7102 ret = reg_node(pRExC_state, SANY);
7104 ret = reg_node(pRExC_state, REG_ANY);
7105 *flagp |= HASWIDTH|SIMPLE;
7107 Set_Node_Length(ret, 1); /* MJD */
7111 char * const oregcomp_parse = ++RExC_parse;
7112 ret = regclass(pRExC_state,depth+1);
7113 if (*RExC_parse != ']') {
7114 RExC_parse = oregcomp_parse;
7115 vFAIL("Unmatched [");
7117 nextchar(pRExC_state);
7118 *flagp |= HASWIDTH|SIMPLE;
7119 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
7123 nextchar(pRExC_state);
7124 ret = reg(pRExC_state, 1, &flags,depth+1);
7126 if (flags & TRYAGAIN) {
7127 if (RExC_parse == RExC_end) {
7128 /* Make parent create an empty node if needed. */
7136 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7140 if (flags & TRYAGAIN) {
7144 vFAIL("Internal urp");
7145 /* Supposed to be caught earlier. */
7148 if (!regcurly(RExC_parse)) {
7157 vFAIL("Quantifier follows nothing");
7165 len=0; /* silence a spurious compiler warning */
7166 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
7167 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
7168 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
7169 ret = reganode(pRExC_state, FOLDCHAR, cp);
7170 Set_Node_Length(ret, 1); /* MJD */
7171 nextchar(pRExC_state); /* kill whitespace under /x */
7179 This switch handles escape sequences that resolve to some kind
7180 of special regop and not to literal text. Escape sequnces that
7181 resolve to literal text are handled below in the switch marked
7184 Every entry in this switch *must* have a corresponding entry
7185 in the literal escape switch. However, the opposite is not
7186 required, as the default for this switch is to jump to the
7187 literal text handling code.
7189 switch ((U8)*++RExC_parse) {
7194 /* Special Escapes */
7196 RExC_seen_zerolen++;
7197 ret = reg_node(pRExC_state, SBOL);
7199 goto finish_meta_pat;
7201 ret = reg_node(pRExC_state, GPOS);
7202 RExC_seen |= REG_SEEN_GPOS;
7204 goto finish_meta_pat;
7206 RExC_seen_zerolen++;
7207 ret = reg_node(pRExC_state, KEEPS);
7209 /* XXX:dmq : disabling in-place substitution seems to
7210 * be necessary here to avoid cases of memory corruption, as
7211 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7213 RExC_seen |= REG_SEEN_LOOKBEHIND;
7214 goto finish_meta_pat;
7216 ret = reg_node(pRExC_state, SEOL);
7218 RExC_seen_zerolen++; /* Do not optimize RE away */
7219 goto finish_meta_pat;
7221 ret = reg_node(pRExC_state, EOS);
7223 RExC_seen_zerolen++; /* Do not optimize RE away */
7224 goto finish_meta_pat;
7226 ret = reg_node(pRExC_state, CANY);
7227 RExC_seen |= REG_SEEN_CANY;
7228 *flagp |= HASWIDTH|SIMPLE;
7229 goto finish_meta_pat;
7231 ret = reg_node(pRExC_state, CLUMP);
7233 goto finish_meta_pat;
7235 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
7236 *flagp |= HASWIDTH|SIMPLE;
7237 goto finish_meta_pat;
7239 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
7240 *flagp |= HASWIDTH|SIMPLE;
7241 goto finish_meta_pat;
7243 RExC_seen_zerolen++;
7244 RExC_seen |= REG_SEEN_LOOKBEHIND;
7245 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
7247 goto finish_meta_pat;
7249 RExC_seen_zerolen++;
7250 RExC_seen |= REG_SEEN_LOOKBEHIND;
7251 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
7253 goto finish_meta_pat;
7255 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
7256 *flagp |= HASWIDTH|SIMPLE;
7257 goto finish_meta_pat;
7259 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
7260 *flagp |= HASWIDTH|SIMPLE;
7261 goto finish_meta_pat;
7263 ret = reg_node(pRExC_state, DIGIT);
7264 *flagp |= HASWIDTH|SIMPLE;
7265 goto finish_meta_pat;
7267 ret = reg_node(pRExC_state, NDIGIT);
7268 *flagp |= HASWIDTH|SIMPLE;
7269 goto finish_meta_pat;
7271 ret = reg_node(pRExC_state, LNBREAK);
7272 *flagp |= HASWIDTH|SIMPLE;
7273 goto finish_meta_pat;
7275 ret = reg_node(pRExC_state, HORIZWS);
7276 *flagp |= HASWIDTH|SIMPLE;
7277 goto finish_meta_pat;
7279 ret = reg_node(pRExC_state, NHORIZWS);
7280 *flagp |= HASWIDTH|SIMPLE;
7281 goto finish_meta_pat;
7283 ret = reg_node(pRExC_state, VERTWS);
7284 *flagp |= HASWIDTH|SIMPLE;
7285 goto finish_meta_pat;
7287 ret = reg_node(pRExC_state, NVERTWS);
7288 *flagp |= HASWIDTH|SIMPLE;
7290 nextchar(pRExC_state);
7291 Set_Node_Length(ret, 2); /* MJD */
7296 char* const oldregxend = RExC_end;
7298 char* parse_start = RExC_parse - 2;
7301 if (RExC_parse[1] == '{') {
7302 /* a lovely hack--pretend we saw [\pX] instead */
7303 RExC_end = strchr(RExC_parse, '}');
7305 const U8 c = (U8)*RExC_parse;
7307 RExC_end = oldregxend;
7308 vFAIL2("Missing right brace on \\%c{}", c);
7313 RExC_end = RExC_parse + 2;
7314 if (RExC_end > oldregxend)
7315 RExC_end = oldregxend;
7319 ret = regclass(pRExC_state,depth+1);
7321 RExC_end = oldregxend;
7324 Set_Node_Offset(ret, parse_start + 2);
7325 Set_Node_Cur_Length(ret);
7326 nextchar(pRExC_state);
7327 *flagp |= HASWIDTH|SIMPLE;
7331 /* Handle \N and \N{NAME} here and not below because it can be
7332 multicharacter. join_exact() will join them up later on.
7333 Also this makes sure that things like /\N{BLAH}+/ and
7334 \N{BLAH} being multi char Just Happen. dmq*/
7336 ret= reg_namedseq(pRExC_state, NULL, flagp);
7338 case 'k': /* Handle \k<NAME> and \k'NAME' */
7341 char ch= RExC_parse[1];
7342 if (ch != '<' && ch != '\'' && ch != '{') {
7344 vFAIL2("Sequence %.2s... not terminated",parse_start);
7346 /* this pretty much dupes the code for (?P=...) in reg(), if
7347 you change this make sure you change that */
7348 char* name_start = (RExC_parse += 2);
7350 SV *sv_dat = reg_scan_name(pRExC_state,
7351 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7352 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7353 if (RExC_parse == name_start || *RExC_parse != ch)
7354 vFAIL2("Sequence %.3s... not terminated",parse_start);
7357 num = add_data( pRExC_state, 1, "S" );
7358 RExC_rxi->data->data[num]=(void*)sv_dat;
7359 SvREFCNT_inc_simple_void(sv_dat);
7363 ret = reganode(pRExC_state,
7364 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7368 /* override incorrect value set in reganode MJD */
7369 Set_Node_Offset(ret, parse_start+1);
7370 Set_Node_Cur_Length(ret); /* MJD */
7371 nextchar(pRExC_state);
7377 case '1': case '2': case '3': case '4':
7378 case '5': case '6': case '7': case '8': case '9':
7381 bool isg = *RExC_parse == 'g';
7386 if (*RExC_parse == '{') {
7390 if (*RExC_parse == '-') {
7394 if (hasbrace && !isDIGIT(*RExC_parse)) {
7395 if (isrel) RExC_parse--;
7397 goto parse_named_seq;
7399 num = atoi(RExC_parse);
7400 if (isg && num == 0)
7401 vFAIL("Reference to invalid group 0");
7403 num = RExC_npar - num;
7405 vFAIL("Reference to nonexistent or unclosed group");
7407 if (!isg && num > 9 && num >= RExC_npar)
7410 char * const parse_start = RExC_parse - 1; /* MJD */
7411 while (isDIGIT(*RExC_parse))
7413 if (parse_start == RExC_parse - 1)
7414 vFAIL("Unterminated \\g... pattern");
7416 if (*RExC_parse != '}')
7417 vFAIL("Unterminated \\g{...} pattern");
7421 if (num > (I32)RExC_rx->nparens)
7422 vFAIL("Reference to nonexistent group");
7425 ret = reganode(pRExC_state,
7426 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7430 /* override incorrect value set in reganode MJD */
7431 Set_Node_Offset(ret, parse_start+1);
7432 Set_Node_Cur_Length(ret); /* MJD */
7434 nextchar(pRExC_state);
7439 if (RExC_parse >= RExC_end)
7440 FAIL("Trailing \\");
7443 /* Do not generate "unrecognized" warnings here, we fall
7444 back into the quick-grab loop below */
7451 if (RExC_flags & RXf_PMf_EXTENDED) {
7452 if ( reg_skipcomment( pRExC_state ) )
7459 register STRLEN len;
7464 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7466 parse_start = RExC_parse - 1;
7472 ret = reg_node(pRExC_state,
7473 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7475 for (len = 0, p = RExC_parse - 1;
7476 len < 127 && p < RExC_end;
7479 char * const oldp = p;
7481 if (RExC_flags & RXf_PMf_EXTENDED)
7482 p = regwhite( pRExC_state, p );
7487 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7488 goto normal_default;
7498 /* Literal Escapes Switch
7500 This switch is meant to handle escape sequences that
7501 resolve to a literal character.
7503 Every escape sequence that represents something
7504 else, like an assertion or a char class, is handled
7505 in the switch marked 'Special Escapes' above in this
7506 routine, but also has an entry here as anything that
7507 isn't explicitly mentioned here will be treated as
7508 an unescaped equivalent literal.
7512 /* These are all the special escapes. */
7516 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7517 goto normal_default;
7518 case 'A': /* Start assertion */
7519 case 'b': case 'B': /* Word-boundary assertion*/
7520 case 'C': /* Single char !DANGEROUS! */
7521 case 'd': case 'D': /* digit class */
7522 case 'g': case 'G': /* generic-backref, pos assertion */
7523 case 'h': case 'H': /* HORIZWS */
7524 case 'k': case 'K': /* named backref, keep marker */
7525 case 'N': /* named char sequence */
7526 case 'p': case 'P': /* Unicode property */
7527 case 'R': /* LNBREAK */
7528 case 's': case 'S': /* space class */
7529 case 'v': case 'V': /* VERTWS */
7530 case 'w': case 'W': /* word class */
7531 case 'X': /* eXtended Unicode "combining character sequence" */
7532 case 'z': case 'Z': /* End of line/string assertion */
7536 /* Anything after here is an escape that resolves to a
7537 literal. (Except digits, which may or may not)
7556 ender = ASCII_TO_NATIVE('\033');
7560 ender = ASCII_TO_NATIVE('\007');
7565 STRLEN brace_len = len;
7567 const char* error_msg;
7569 bool valid = grok_bslash_o(p,
7576 RExC_parse = p; /* going to die anyway; point
7577 to exact spot of failure */
7584 if (PL_encoding && ender < 0x100) {
7585 goto recode_encoding;
7594 char* const e = strchr(p, '}');
7598 vFAIL("Missing right brace on \\x{}");
7601 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7602 | PERL_SCAN_DISALLOW_PREFIX;
7603 STRLEN numlen = e - p - 1;
7604 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7611 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7613 ender = grok_hex(p, &numlen, &flags, NULL);
7616 if (PL_encoding && ender < 0x100)
7617 goto recode_encoding;
7621 ender = grok_bslash_c(*p++, SIZE_ONLY);
7623 case '0': case '1': case '2': case '3':case '4':
7624 case '5': case '6': case '7': case '8':case '9':
7626 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
7628 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
7630 ender = grok_oct(p, &numlen, &flags, NULL);
7640 if (PL_encoding && ender < 0x100)
7641 goto recode_encoding;
7645 SV* enc = PL_encoding;
7646 ender = reg_recode((const char)(U8)ender, &enc);
7647 if (!enc && SIZE_ONLY)
7648 ckWARNreg(p, "Invalid escape in the specified encoding");
7654 FAIL("Trailing \\");
7657 if (!SIZE_ONLY&& isALPHA(*p))
7658 ckWARN2reg(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7659 goto normal_default;
7664 if (UTF8_IS_START(*p) && UTF) {
7666 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7667 &numlen, UTF8_ALLOW_DEFAULT);
7674 if ( RExC_flags & RXf_PMf_EXTENDED)
7675 p = regwhite( pRExC_state, p );
7677 /* Prime the casefolded buffer. */
7678 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7680 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7685 /* Emit all the Unicode characters. */
7687 for (foldbuf = tmpbuf;
7689 foldlen -= numlen) {
7690 ender = utf8_to_uvchr(foldbuf, &numlen);
7692 const STRLEN unilen = reguni(pRExC_state, ender, s);
7695 /* In EBCDIC the numlen
7696 * and unilen can differ. */
7698 if (numlen >= foldlen)
7702 break; /* "Can't happen." */
7706 const STRLEN unilen = reguni(pRExC_state, ender, s);
7715 REGC((char)ender, s++);
7721 /* Emit all the Unicode characters. */
7723 for (foldbuf = tmpbuf;
7725 foldlen -= numlen) {
7726 ender = utf8_to_uvchr(foldbuf, &numlen);
7728 const STRLEN unilen = reguni(pRExC_state, ender, s);
7731 /* In EBCDIC the numlen
7732 * and unilen can differ. */
7734 if (numlen >= foldlen)
7742 const STRLEN unilen = reguni(pRExC_state, ender, s);
7751 REGC((char)ender, s++);
7755 Set_Node_Cur_Length(ret); /* MJD */
7756 nextchar(pRExC_state);
7758 /* len is STRLEN which is unsigned, need to copy to signed */
7761 vFAIL("Internal disaster");
7765 if (len == 1 && UNI_IS_INVARIANT(ender))
7769 RExC_size += STR_SZ(len);
7772 RExC_emit += STR_SZ(len);
7782 S_regwhite( RExC_state_t *pRExC_state, char *p )
7784 const char *e = RExC_end;
7786 PERL_ARGS_ASSERT_REGWHITE;
7791 else if (*p == '#') {
7800 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7808 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7809 Character classes ([:foo:]) can also be negated ([:^foo:]).
7810 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7811 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7812 but trigger failures because they are currently unimplemented. */
7814 #define POSIXCC_DONE(c) ((c) == ':')
7815 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7816 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7819 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7822 I32 namedclass = OOB_NAMEDCLASS;
7824 PERL_ARGS_ASSERT_REGPPOSIXCC;
7826 if (value == '[' && RExC_parse + 1 < RExC_end &&
7827 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7828 POSIXCC(UCHARAT(RExC_parse))) {
7829 const char c = UCHARAT(RExC_parse);
7830 char* const s = RExC_parse++;
7832 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7834 if (RExC_parse == RExC_end)
7835 /* Grandfather lone [:, [=, [. */
7838 const char* const t = RExC_parse++; /* skip over the c */
7841 if (UCHARAT(RExC_parse) == ']') {
7842 const char *posixcc = s + 1;
7843 RExC_parse++; /* skip over the ending ] */
7846 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7847 const I32 skip = t - posixcc;
7849 /* Initially switch on the length of the name. */
7852 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7853 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7856 /* Names all of length 5. */
7857 /* alnum alpha ascii blank cntrl digit graph lower
7858 print punct space upper */
7859 /* Offset 4 gives the best switch position. */
7860 switch (posixcc[4]) {
7862 if (memEQ(posixcc, "alph", 4)) /* alpha */
7863 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7866 if (memEQ(posixcc, "spac", 4)) /* space */
7867 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7870 if (memEQ(posixcc, "grap", 4)) /* graph */
7871 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7874 if (memEQ(posixcc, "asci", 4)) /* ascii */
7875 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7878 if (memEQ(posixcc, "blan", 4)) /* blank */
7879 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7882 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7883 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7886 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7887 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7890 if (memEQ(posixcc, "lowe", 4)) /* lower */
7891 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7892 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7893 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7896 if (memEQ(posixcc, "digi", 4)) /* digit */
7897 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7898 else if (memEQ(posixcc, "prin", 4)) /* print */
7899 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7900 else if (memEQ(posixcc, "punc", 4)) /* punct */
7901 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7906 if (memEQ(posixcc, "xdigit", 6))
7907 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7911 if (namedclass == OOB_NAMEDCLASS)
7912 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7914 assert (posixcc[skip] == ':');
7915 assert (posixcc[skip+1] == ']');
7916 } else if (!SIZE_ONLY) {
7917 /* [[=foo=]] and [[.foo.]] are still future. */
7919 /* adjust RExC_parse so the warning shows after
7921 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7923 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7926 /* Maternal grandfather:
7927 * "[:" ending in ":" but not in ":]" */
7937 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7941 PERL_ARGS_ASSERT_CHECKPOSIXCC;
7943 if (POSIXCC(UCHARAT(RExC_parse))) {
7944 const char *s = RExC_parse;
7945 const char c = *s++;
7949 if (*s && c == *s && s[1] == ']') {
7951 "POSIX syntax [%c %c] belongs inside character classes",
7954 /* [[=foo=]] and [[.foo.]] are still future. */
7955 if (POSIXCC_NOTYET(c)) {
7956 /* adjust RExC_parse so the error shows after
7958 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7960 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7967 #define _C_C_T_(NAME,TEST,WORD) \
7970 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7972 for (value = 0; value < 256; value++) \
7974 ANYOF_BITMAP_SET(ret, value); \
7979 case ANYOF_N##NAME: \
7981 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7983 for (value = 0; value < 256; value++) \
7985 ANYOF_BITMAP_SET(ret, value); \
7991 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7993 for (value = 0; value < 256; value++) \
7995 ANYOF_BITMAP_SET(ret, value); \
7999 case ANYOF_N##NAME: \
8000 for (value = 0; value < 256; value++) \
8002 ANYOF_BITMAP_SET(ret, value); \
8008 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
8009 so that it is possible to override the option here without having to
8010 rebuild the entire core. as we are required to do if we change regcomp.h
8011 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
8013 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
8014 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
8017 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8018 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
8020 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
8024 parse a class specification and produce either an ANYOF node that
8025 matches the pattern or if the pattern matches a single char only and
8026 that char is < 256 and we are case insensitive then we produce an
8031 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
8034 register UV nextvalue;
8035 register IV prevvalue = OOB_UNICODE;
8036 register IV range = 0;
8037 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
8038 register regnode *ret;
8041 char *rangebegin = NULL;
8042 bool need_class = 0;
8045 bool optimize_invert = TRUE;
8046 AV* unicode_alternate = NULL;
8048 UV literal_endpoint = 0;
8050 UV stored = 0; /* number of chars stored in the class */
8052 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
8053 case we need to change the emitted regop to an EXACT. */
8054 const char * orig_parse = RExC_parse;
8055 GET_RE_DEBUG_FLAGS_DECL;
8057 PERL_ARGS_ASSERT_REGCLASS;
8059 PERL_UNUSED_ARG(depth);
8062 DEBUG_PARSE("clas");
8064 /* Assume we are going to generate an ANYOF node. */
8065 ret = reganode(pRExC_state, ANYOF, 0);
8068 ANYOF_FLAGS(ret) = 0;
8070 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
8074 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
8078 RExC_size += ANYOF_SKIP;
8079 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
8082 RExC_emit += ANYOF_SKIP;
8084 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
8086 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
8087 ANYOF_BITMAP_ZERO(ret);
8088 listsv = newSVpvs("# comment\n");
8091 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8093 if (!SIZE_ONLY && POSIXCC(nextvalue))
8094 checkposixcc(pRExC_state);
8096 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
8097 if (UCHARAT(RExC_parse) == ']')
8101 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
8105 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
8108 rangebegin = RExC_parse;
8110 value = utf8n_to_uvchr((U8*)RExC_parse,
8111 RExC_end - RExC_parse,
8112 &numlen, UTF8_ALLOW_DEFAULT);
8113 RExC_parse += numlen;
8116 value = UCHARAT(RExC_parse++);
8118 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8119 if (value == '[' && POSIXCC(nextvalue))
8120 namedclass = regpposixcc(pRExC_state, value);
8121 else if (value == '\\') {
8123 value = utf8n_to_uvchr((U8*)RExC_parse,
8124 RExC_end - RExC_parse,
8125 &numlen, UTF8_ALLOW_DEFAULT);
8126 RExC_parse += numlen;
8129 value = UCHARAT(RExC_parse++);
8130 /* Some compilers cannot handle switching on 64-bit integer
8131 * values, therefore value cannot be an UV. Yes, this will
8132 * be a problem later if we want switch on Unicode.
8133 * A similar issue a little bit later when switching on
8134 * namedclass. --jhi */
8135 switch ((I32)value) {
8136 case 'w': namedclass = ANYOF_ALNUM; break;
8137 case 'W': namedclass = ANYOF_NALNUM; break;
8138 case 's': namedclass = ANYOF_SPACE; break;
8139 case 'S': namedclass = ANYOF_NSPACE; break;
8140 case 'd': namedclass = ANYOF_DIGIT; break;
8141 case 'D': namedclass = ANYOF_NDIGIT; break;
8142 case 'v': namedclass = ANYOF_VERTWS; break;
8143 case 'V': namedclass = ANYOF_NVERTWS; break;
8144 case 'h': namedclass = ANYOF_HORIZWS; break;
8145 case 'H': namedclass = ANYOF_NHORIZWS; break;
8146 case 'N': /* Handle \N{NAME} in class */
8148 /* We only pay attention to the first char of
8149 multichar strings being returned. I kinda wonder
8150 if this makes sense as it does change the behaviour
8151 from earlier versions, OTOH that behaviour was broken
8153 UV v; /* value is register so we cant & it /grrr */
8154 if (reg_namedseq(pRExC_state, &v, NULL)) {
8164 if (RExC_parse >= RExC_end)
8165 vFAIL2("Empty \\%c{}", (U8)value);
8166 if (*RExC_parse == '{') {
8167 const U8 c = (U8)value;
8168 e = strchr(RExC_parse++, '}');
8170 vFAIL2("Missing right brace on \\%c{}", c);
8171 while (isSPACE(UCHARAT(RExC_parse)))
8173 if (e == RExC_parse)
8174 vFAIL2("Empty \\%c{}", c);
8176 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
8184 if (UCHARAT(RExC_parse) == '^') {
8187 value = value == 'p' ? 'P' : 'p'; /* toggle */
8188 while (isSPACE(UCHARAT(RExC_parse))) {
8193 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
8194 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
8197 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8198 namedclass = ANYOF_MAX; /* no official name, but it's named */
8201 case 'n': value = '\n'; break;
8202 case 'r': value = '\r'; break;
8203 case 't': value = '\t'; break;
8204 case 'f': value = '\f'; break;
8205 case 'b': value = '\b'; break;
8206 case 'e': value = ASCII_TO_NATIVE('\033');break;
8207 case 'a': value = ASCII_TO_NATIVE('\007');break;
8209 RExC_parse--; /* function expects to be pointed at the 'o' */
8211 const char* error_msg;
8212 bool valid = grok_bslash_o(RExC_parse,
8217 RExC_parse += numlen;
8222 if (PL_encoding && value < 0x100) {
8223 goto recode_encoding;
8227 if (*RExC_parse == '{') {
8228 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8229 | PERL_SCAN_DISALLOW_PREFIX;
8230 char * const e = strchr(RExC_parse++, '}');
8232 vFAIL("Missing right brace on \\x{}");
8234 numlen = e - RExC_parse;
8235 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8239 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8241 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8242 RExC_parse += numlen;
8244 if (PL_encoding && value < 0x100)
8245 goto recode_encoding;
8248 value = grok_bslash_c(*RExC_parse++, SIZE_ONLY);
8250 case '0': case '1': case '2': case '3': case '4':
8251 case '5': case '6': case '7':
8253 /* Take 1-3 octal digits */
8254 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8256 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8257 RExC_parse += numlen;
8258 if (PL_encoding && value < 0x100)
8259 goto recode_encoding;
8264 SV* enc = PL_encoding;
8265 value = reg_recode((const char)(U8)value, &enc);
8266 if (!enc && SIZE_ONLY)
8267 ckWARNreg(RExC_parse,
8268 "Invalid escape in the specified encoding");
8272 /* Allow \_ to not give an error */
8273 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
8274 ckWARN2reg(RExC_parse,
8275 "Unrecognized escape \\%c in character class passed through",
8280 } /* end of \blah */
8286 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8288 if (!SIZE_ONLY && !need_class)
8289 ANYOF_CLASS_ZERO(ret);
8293 /* a bad range like a-\d, a-[:digit:] ? */
8297 RExC_parse >= rangebegin ?
8298 RExC_parse - rangebegin : 0;
8299 ckWARN4reg(RExC_parse,
8300 "False [] range \"%*.*s\"",
8303 if (prevvalue < 256) {
8304 ANYOF_BITMAP_SET(ret, prevvalue);
8305 ANYOF_BITMAP_SET(ret, '-');
8308 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8309 Perl_sv_catpvf(aTHX_ listsv,
8310 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8314 range = 0; /* this was not a true range */
8320 const char *what = NULL;
8323 if (namedclass > OOB_NAMEDCLASS)
8324 optimize_invert = FALSE;
8325 /* Possible truncation here but in some 64-bit environments
8326 * the compiler gets heartburn about switch on 64-bit values.
8327 * A similar issue a little earlier when switching on value.
8329 switch ((I32)namedclass) {
8331 case _C_C_T_(ALNUMC, isALNUMC(value), POSIX_CC_UNI_NAME("Alnum"));
8332 case _C_C_T_(ALPHA, isALPHA(value), POSIX_CC_UNI_NAME("Alpha"));
8333 case _C_C_T_(BLANK, isBLANK(value), POSIX_CC_UNI_NAME("Blank"));
8334 case _C_C_T_(CNTRL, isCNTRL(value), POSIX_CC_UNI_NAME("Cntrl"));
8335 case _C_C_T_(GRAPH, isGRAPH(value), POSIX_CC_UNI_NAME("Graph"));
8336 case _C_C_T_(LOWER, isLOWER(value), POSIX_CC_UNI_NAME("Lower"));
8337 case _C_C_T_(PRINT, isPRINT(value), POSIX_CC_UNI_NAME("Print"));
8338 case _C_C_T_(PSXSPC, isPSXSPC(value), POSIX_CC_UNI_NAME("Space"));
8339 case _C_C_T_(PUNCT, isPUNCT(value), POSIX_CC_UNI_NAME("Punct"));
8340 case _C_C_T_(UPPER, isUPPER(value), POSIX_CC_UNI_NAME("Upper"));
8341 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8342 case _C_C_T_(ALNUM, isALNUM(value), "Word");
8343 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
8345 case _C_C_T_(SPACE, isSPACE(value), "PerlSpace");
8346 case _C_C_T_(ALNUM, isALNUM(value), "PerlWord");
8348 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
8349 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8350 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8353 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8356 for (value = 0; value < 128; value++)
8357 ANYOF_BITMAP_SET(ret, value);
8359 for (value = 0; value < 256; value++) {
8361 ANYOF_BITMAP_SET(ret, value);
8370 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8373 for (value = 128; value < 256; value++)
8374 ANYOF_BITMAP_SET(ret, value);
8376 for (value = 0; value < 256; value++) {
8377 if (!isASCII(value))
8378 ANYOF_BITMAP_SET(ret, value);
8387 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8389 /* consecutive digits assumed */
8390 for (value = '0'; value <= '9'; value++)
8391 ANYOF_BITMAP_SET(ret, value);
8394 what = POSIX_CC_UNI_NAME("Digit");
8398 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8400 /* consecutive digits assumed */
8401 for (value = 0; value < '0'; value++)
8402 ANYOF_BITMAP_SET(ret, value);
8403 for (value = '9' + 1; value < 256; value++)
8404 ANYOF_BITMAP_SET(ret, value);
8407 what = POSIX_CC_UNI_NAME("Digit");
8410 /* this is to handle \p and \P */
8413 vFAIL("Invalid [::] class");
8417 /* Strings such as "+utf8::isWord\n" */
8418 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8421 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8424 } /* end of namedclass \blah */
8427 if (prevvalue > (IV)value) /* b-a */ {
8428 const int w = RExC_parse - rangebegin;
8429 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8430 range = 0; /* not a valid range */
8434 prevvalue = value; /* save the beginning of the range */
8435 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8436 RExC_parse[1] != ']') {
8439 /* a bad range like \w-, [:word:]- ? */
8440 if (namedclass > OOB_NAMEDCLASS) {
8441 if (ckWARN(WARN_REGEXP)) {
8443 RExC_parse >= rangebegin ?
8444 RExC_parse - rangebegin : 0;
8446 "False [] range \"%*.*s\"",
8450 ANYOF_BITMAP_SET(ret, '-');
8452 range = 1; /* yeah, it's a range! */
8453 continue; /* but do it the next time */
8457 /* now is the next time */
8458 /*stored += (value - prevvalue + 1);*/
8460 if (prevvalue < 256) {
8461 const IV ceilvalue = value < 256 ? value : 255;
8464 /* In EBCDIC [\x89-\x91] should include
8465 * the \x8e but [i-j] should not. */
8466 if (literal_endpoint == 2 &&
8467 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8468 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8470 if (isLOWER(prevvalue)) {
8471 for (i = prevvalue; i <= ceilvalue; i++)
8472 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8474 ANYOF_BITMAP_SET(ret, i);
8477 for (i = prevvalue; i <= ceilvalue; i++)
8478 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8480 ANYOF_BITMAP_SET(ret, i);
8486 for (i = prevvalue; i <= ceilvalue; i++) {
8487 if (!ANYOF_BITMAP_TEST(ret,i)) {
8489 ANYOF_BITMAP_SET(ret, i);
8493 if (value > 255 || UTF) {
8494 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8495 const UV natvalue = NATIVE_TO_UNI(value);
8496 stored+=2; /* can't optimize this class */
8497 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8498 if (prevnatvalue < natvalue) { /* what about > ? */
8499 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8500 prevnatvalue, natvalue);
8502 else if (prevnatvalue == natvalue) {
8503 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8505 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8507 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8509 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8510 if (RExC_precomp[0] == ':' &&
8511 RExC_precomp[1] == '[' &&
8512 (f == 0xDF || f == 0x92)) {
8513 f = NATIVE_TO_UNI(f);
8516 /* If folding and foldable and a single
8517 * character, insert also the folded version
8518 * to the charclass. */
8520 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8521 if ((RExC_precomp[0] == ':' &&
8522 RExC_precomp[1] == '[' &&
8524 (value == 0xFB05 || value == 0xFB06))) ?
8525 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8526 foldlen == (STRLEN)UNISKIP(f) )
8528 if (foldlen == (STRLEN)UNISKIP(f))
8530 Perl_sv_catpvf(aTHX_ listsv,
8533 /* Any multicharacter foldings
8534 * require the following transform:
8535 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8536 * where E folds into "pq" and F folds
8537 * into "rst", all other characters
8538 * fold to single characters. We save
8539 * away these multicharacter foldings,
8540 * to be later saved as part of the
8541 * additional "s" data. */
8544 if (!unicode_alternate)
8545 unicode_alternate = newAV();
8546 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8548 av_push(unicode_alternate, sv);
8552 /* If folding and the value is one of the Greek
8553 * sigmas insert a few more sigmas to make the
8554 * folding rules of the sigmas to work right.
8555 * Note that not all the possible combinations
8556 * are handled here: some of them are handled
8557 * by the standard folding rules, and some of
8558 * them (literal or EXACTF cases) are handled
8559 * during runtime in regexec.c:S_find_byclass(). */
8560 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8561 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8562 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8563 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8564 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8566 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8567 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8568 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8573 literal_endpoint = 0;
8577 range = 0; /* this range (if it was one) is done now */
8581 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8583 RExC_size += ANYOF_CLASS_ADD_SKIP;
8585 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8591 /****** !SIZE_ONLY AFTER HERE *********/
8593 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8594 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8596 /* optimize single char class to an EXACT node
8597 but *only* when its not a UTF/high char */
8598 const char * cur_parse= RExC_parse;
8599 RExC_emit = (regnode *)orig_emit;
8600 RExC_parse = (char *)orig_parse;
8601 ret = reg_node(pRExC_state,
8602 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8603 RExC_parse = (char *)cur_parse;
8604 *STRING(ret)= (char)value;
8606 RExC_emit += STR_SZ(1);
8607 SvREFCNT_dec(listsv);
8610 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8611 if ( /* If the only flag is folding (plus possibly inversion). */
8612 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8614 for (value = 0; value < 256; ++value) {
8615 if (ANYOF_BITMAP_TEST(ret, value)) {
8616 UV fold = PL_fold[value];
8619 ANYOF_BITMAP_SET(ret, fold);
8622 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8625 /* optimize inverted simple patterns (e.g. [^a-z]) */
8626 if (optimize_invert &&
8627 /* If the only flag is inversion. */
8628 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8629 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8630 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8631 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8634 AV * const av = newAV();
8636 /* The 0th element stores the character class description
8637 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8638 * to initialize the appropriate swash (which gets stored in
8639 * the 1st element), and also useful for dumping the regnode.
8640 * The 2nd element stores the multicharacter foldings,
8641 * used later (regexec.c:S_reginclass()). */
8642 av_store(av, 0, listsv);
8643 av_store(av, 1, NULL);
8644 av_store(av, 2, MUTABLE_SV(unicode_alternate));
8645 rv = newRV_noinc(MUTABLE_SV(av));
8646 n = add_data(pRExC_state, 1, "s");
8647 RExC_rxi->data->data[n] = (void*)rv;
8655 /* reg_skipcomment()
8657 Absorbs an /x style # comments from the input stream.
8658 Returns true if there is more text remaining in the stream.
8659 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8660 terminates the pattern without including a newline.
8662 Note its the callers responsibility to ensure that we are
8668 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8672 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8674 while (RExC_parse < RExC_end)
8675 if (*RExC_parse++ == '\n') {
8680 /* we ran off the end of the pattern without ending
8681 the comment, so we have to add an \n when wrapping */
8682 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8690 Advance that parse position, and optionall absorbs
8691 "whitespace" from the inputstream.
8693 Without /x "whitespace" means (?#...) style comments only,
8694 with /x this means (?#...) and # comments and whitespace proper.
8696 Returns the RExC_parse point from BEFORE the scan occurs.
8698 This is the /x friendly way of saying RExC_parse++.
8702 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8704 char* const retval = RExC_parse++;
8706 PERL_ARGS_ASSERT_NEXTCHAR;
8709 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8710 RExC_parse[2] == '#') {
8711 while (*RExC_parse != ')') {
8712 if (RExC_parse == RExC_end)
8713 FAIL("Sequence (?#... not terminated");
8719 if (RExC_flags & RXf_PMf_EXTENDED) {
8720 if (isSPACE(*RExC_parse)) {
8724 else if (*RExC_parse == '#') {
8725 if ( reg_skipcomment( pRExC_state ) )
8734 - reg_node - emit a node
8736 STATIC regnode * /* Location. */
8737 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8740 register regnode *ptr;
8741 regnode * const ret = RExC_emit;
8742 GET_RE_DEBUG_FLAGS_DECL;
8744 PERL_ARGS_ASSERT_REG_NODE;
8747 SIZE_ALIGN(RExC_size);
8751 if (RExC_emit >= RExC_emit_bound)
8752 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8754 NODE_ALIGN_FILL(ret);
8756 FILL_ADVANCE_NODE(ptr, op);
8757 #ifdef RE_TRACK_PATTERN_OFFSETS
8758 if (RExC_offsets) { /* MJD */
8759 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8760 "reg_node", __LINE__,
8762 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8763 ? "Overwriting end of array!\n" : "OK",
8764 (UV)(RExC_emit - RExC_emit_start),
8765 (UV)(RExC_parse - RExC_start),
8766 (UV)RExC_offsets[0]));
8767 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8775 - reganode - emit a node with an argument
8777 STATIC regnode * /* Location. */
8778 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8781 register regnode *ptr;
8782 regnode * const ret = RExC_emit;
8783 GET_RE_DEBUG_FLAGS_DECL;
8785 PERL_ARGS_ASSERT_REGANODE;
8788 SIZE_ALIGN(RExC_size);
8793 assert(2==regarglen[op]+1);
8795 Anything larger than this has to allocate the extra amount.
8796 If we changed this to be:
8798 RExC_size += (1 + regarglen[op]);
8800 then it wouldn't matter. Its not clear what side effect
8801 might come from that so its not done so far.
8806 if (RExC_emit >= RExC_emit_bound)
8807 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8809 NODE_ALIGN_FILL(ret);
8811 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8812 #ifdef RE_TRACK_PATTERN_OFFSETS
8813 if (RExC_offsets) { /* MJD */
8814 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8818 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8819 "Overwriting end of array!\n" : "OK",
8820 (UV)(RExC_emit - RExC_emit_start),
8821 (UV)(RExC_parse - RExC_start),
8822 (UV)RExC_offsets[0]));
8823 Set_Cur_Node_Offset;
8831 - reguni - emit (if appropriate) a Unicode character
8834 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8838 PERL_ARGS_ASSERT_REGUNI;
8840 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8844 - reginsert - insert an operator in front of already-emitted operand
8846 * Means relocating the operand.
8849 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8852 register regnode *src;
8853 register regnode *dst;
8854 register regnode *place;
8855 const int offset = regarglen[(U8)op];
8856 const int size = NODE_STEP_REGNODE + offset;
8857 GET_RE_DEBUG_FLAGS_DECL;
8859 PERL_ARGS_ASSERT_REGINSERT;
8860 PERL_UNUSED_ARG(depth);
8861 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8862 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8871 if (RExC_open_parens) {
8873 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8874 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8875 if ( RExC_open_parens[paren] >= opnd ) {
8876 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8877 RExC_open_parens[paren] += size;
8879 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8881 if ( RExC_close_parens[paren] >= opnd ) {
8882 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8883 RExC_close_parens[paren] += size;
8885 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8890 while (src > opnd) {
8891 StructCopy(--src, --dst, regnode);
8892 #ifdef RE_TRACK_PATTERN_OFFSETS
8893 if (RExC_offsets) { /* MJD 20010112 */
8894 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8898 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8899 ? "Overwriting end of array!\n" : "OK",
8900 (UV)(src - RExC_emit_start),
8901 (UV)(dst - RExC_emit_start),
8902 (UV)RExC_offsets[0]));
8903 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8904 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8910 place = opnd; /* Op node, where operand used to be. */
8911 #ifdef RE_TRACK_PATTERN_OFFSETS
8912 if (RExC_offsets) { /* MJD */
8913 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8917 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8918 ? "Overwriting end of array!\n" : "OK",
8919 (UV)(place - RExC_emit_start),
8920 (UV)(RExC_parse - RExC_start),
8921 (UV)RExC_offsets[0]));
8922 Set_Node_Offset(place, RExC_parse);
8923 Set_Node_Length(place, 1);
8926 src = NEXTOPER(place);
8927 FILL_ADVANCE_NODE(place, op);
8928 Zero(src, offset, regnode);
8932 - regtail - set the next-pointer at the end of a node chain of p to val.
8933 - SEE ALSO: regtail_study
8935 /* TODO: All three parms should be const */
8937 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8940 register regnode *scan;
8941 GET_RE_DEBUG_FLAGS_DECL;
8943 PERL_ARGS_ASSERT_REGTAIL;
8945 PERL_UNUSED_ARG(depth);
8951 /* Find last node. */
8954 regnode * const temp = regnext(scan);
8956 SV * const mysv=sv_newmortal();
8957 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8958 regprop(RExC_rx, mysv, scan);
8959 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8960 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8961 (temp == NULL ? "->" : ""),
8962 (temp == NULL ? PL_reg_name[OP(val)] : "")
8970 if (reg_off_by_arg[OP(scan)]) {
8971 ARG_SET(scan, val - scan);
8974 NEXT_OFF(scan) = val - scan;
8980 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8981 - Look for optimizable sequences at the same time.
8982 - currently only looks for EXACT chains.
8984 This is expermental code. The idea is to use this routine to perform
8985 in place optimizations on branches and groups as they are constructed,
8986 with the long term intention of removing optimization from study_chunk so
8987 that it is purely analytical.
8989 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8990 to control which is which.
8993 /* TODO: All four parms should be const */
8996 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8999 register regnode *scan;
9001 #ifdef EXPERIMENTAL_INPLACESCAN
9004 GET_RE_DEBUG_FLAGS_DECL;
9006 PERL_ARGS_ASSERT_REGTAIL_STUDY;
9012 /* Find last node. */
9016 regnode * const temp = regnext(scan);
9017 #ifdef EXPERIMENTAL_INPLACESCAN
9018 if (PL_regkind[OP(scan)] == EXACT)
9019 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
9027 if( exact == PSEUDO )
9029 else if ( exact != OP(scan) )
9038 SV * const mysv=sv_newmortal();
9039 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
9040 regprop(RExC_rx, mysv, scan);
9041 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
9042 SvPV_nolen_const(mysv),
9044 PL_reg_name[exact]);
9051 SV * const mysv_val=sv_newmortal();
9052 DEBUG_PARSE_MSG("");
9053 regprop(RExC_rx, mysv_val, val);
9054 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
9055 SvPV_nolen_const(mysv_val),
9056 (IV)REG_NODE_NUM(val),
9060 if (reg_off_by_arg[OP(scan)]) {
9061 ARG_SET(scan, val - scan);
9064 NEXT_OFF(scan) = val - scan;
9072 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
9076 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
9081 for (bit=0; bit<32; bit++) {
9082 if (flags & (1<<bit)) {
9084 PerlIO_printf(Perl_debug_log, "%s",lead);
9085 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
9090 PerlIO_printf(Perl_debug_log, "\n");
9092 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
9098 Perl_regdump(pTHX_ const regexp *r)
9102 SV * const sv = sv_newmortal();
9103 SV *dsv= sv_newmortal();
9105 GET_RE_DEBUG_FLAGS_DECL;
9107 PERL_ARGS_ASSERT_REGDUMP;
9109 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
9111 /* Header fields of interest. */
9112 if (r->anchored_substr) {
9113 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
9114 RE_SV_DUMPLEN(r->anchored_substr), 30);
9115 PerlIO_printf(Perl_debug_log,
9116 "anchored %s%s at %"IVdf" ",
9117 s, RE_SV_TAIL(r->anchored_substr),
9118 (IV)r->anchored_offset);
9119 } else if (r->anchored_utf8) {
9120 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
9121 RE_SV_DUMPLEN(r->anchored_utf8), 30);
9122 PerlIO_printf(Perl_debug_log,
9123 "anchored utf8 %s%s at %"IVdf" ",
9124 s, RE_SV_TAIL(r->anchored_utf8),
9125 (IV)r->anchored_offset);
9127 if (r->float_substr) {
9128 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
9129 RE_SV_DUMPLEN(r->float_substr), 30);
9130 PerlIO_printf(Perl_debug_log,
9131 "floating %s%s at %"IVdf"..%"UVuf" ",
9132 s, RE_SV_TAIL(r->float_substr),
9133 (IV)r->float_min_offset, (UV)r->float_max_offset);
9134 } else if (r->float_utf8) {
9135 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
9136 RE_SV_DUMPLEN(r->float_utf8), 30);
9137 PerlIO_printf(Perl_debug_log,
9138 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
9139 s, RE_SV_TAIL(r->float_utf8),
9140 (IV)r->float_min_offset, (UV)r->float_max_offset);
9142 if (r->check_substr || r->check_utf8)
9143 PerlIO_printf(Perl_debug_log,
9145 (r->check_substr == r->float_substr
9146 && r->check_utf8 == r->float_utf8
9147 ? "(checking floating" : "(checking anchored"));
9148 if (r->extflags & RXf_NOSCAN)
9149 PerlIO_printf(Perl_debug_log, " noscan");
9150 if (r->extflags & RXf_CHECK_ALL)
9151 PerlIO_printf(Perl_debug_log, " isall");
9152 if (r->check_substr || r->check_utf8)
9153 PerlIO_printf(Perl_debug_log, ") ");
9155 if (ri->regstclass) {
9156 regprop(r, sv, ri->regstclass);
9157 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
9159 if (r->extflags & RXf_ANCH) {
9160 PerlIO_printf(Perl_debug_log, "anchored");
9161 if (r->extflags & RXf_ANCH_BOL)
9162 PerlIO_printf(Perl_debug_log, "(BOL)");
9163 if (r->extflags & RXf_ANCH_MBOL)
9164 PerlIO_printf(Perl_debug_log, "(MBOL)");
9165 if (r->extflags & RXf_ANCH_SBOL)
9166 PerlIO_printf(Perl_debug_log, "(SBOL)");
9167 if (r->extflags & RXf_ANCH_GPOS)
9168 PerlIO_printf(Perl_debug_log, "(GPOS)");
9169 PerlIO_putc(Perl_debug_log, ' ');
9171 if (r->extflags & RXf_GPOS_SEEN)
9172 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
9173 if (r->intflags & PREGf_SKIP)
9174 PerlIO_printf(Perl_debug_log, "plus ");
9175 if (r->intflags & PREGf_IMPLICIT)
9176 PerlIO_printf(Perl_debug_log, "implicit ");
9177 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
9178 if (r->extflags & RXf_EVAL_SEEN)
9179 PerlIO_printf(Perl_debug_log, "with eval ");
9180 PerlIO_printf(Perl_debug_log, "\n");
9181 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
9183 PERL_ARGS_ASSERT_REGDUMP;
9184 PERL_UNUSED_CONTEXT;
9186 #endif /* DEBUGGING */
9190 - regprop - printable representation of opcode
9192 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9195 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9196 if (flags & ANYOF_INVERT) \
9197 /*make sure the invert info is in each */ \
9198 sv_catpvs(sv, "^"); \
9204 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9209 RXi_GET_DECL(prog,progi);
9210 GET_RE_DEBUG_FLAGS_DECL;
9212 PERL_ARGS_ASSERT_REGPROP;
9216 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9217 /* It would be nice to FAIL() here, but this may be called from
9218 regexec.c, and it would be hard to supply pRExC_state. */
9219 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9220 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9222 k = PL_regkind[OP(o)];
9226 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9227 * is a crude hack but it may be the best for now since
9228 * we have no flag "this EXACTish node was UTF-8"
9230 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9231 PERL_PV_ESCAPE_UNI_DETECT |
9232 PERL_PV_PRETTY_ELLIPSES |
9233 PERL_PV_PRETTY_LTGT |
9234 PERL_PV_PRETTY_NOCLEAR
9236 } else if (k == TRIE) {
9237 /* print the details of the trie in dumpuntil instead, as
9238 * progi->data isn't available here */
9239 const char op = OP(o);
9240 const U32 n = ARG(o);
9241 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9242 (reg_ac_data *)progi->data->data[n] :
9244 const reg_trie_data * const trie
9245 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9247 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9248 DEBUG_TRIE_COMPILE_r(
9249 Perl_sv_catpvf(aTHX_ sv,
9250 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9251 (UV)trie->startstate,
9252 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9253 (UV)trie->wordcount,
9256 (UV)TRIE_CHARCOUNT(trie),
9257 (UV)trie->uniquecharcount
9260 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9262 int rangestart = -1;
9263 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9265 for (i = 0; i <= 256; i++) {
9266 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9267 if (rangestart == -1)
9269 } else if (rangestart != -1) {
9270 if (i <= rangestart + 3)
9271 for (; rangestart < i; rangestart++)
9272 put_byte(sv, rangestart);
9274 put_byte(sv, rangestart);
9276 put_byte(sv, i - 1);
9284 } else if (k == CURLY) {
9285 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9286 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9287 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9289 else if (k == WHILEM && o->flags) /* Ordinal/of */
9290 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9291 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9292 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9293 if ( RXp_PAREN_NAMES(prog) ) {
9294 if ( k != REF || OP(o) < NREF) {
9295 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9296 SV **name= av_fetch(list, ARG(o), 0 );
9298 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9301 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9302 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9303 I32 *nums=(I32*)SvPVX(sv_dat);
9304 SV **name= av_fetch(list, nums[0], 0 );
9307 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9308 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9309 (n ? "," : ""), (IV)nums[n]);
9311 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9315 } else if (k == GOSUB)
9316 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9317 else if (k == VERB) {
9319 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9320 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9321 } else if (k == LOGICAL)
9322 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9323 else if (k == FOLDCHAR)
9324 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9325 else if (k == ANYOF) {
9326 int i, rangestart = -1;
9327 const U8 flags = ANYOF_FLAGS(o);
9330 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9331 static const char * const anyofs[] = {
9364 if (flags & ANYOF_LOCALE)
9365 sv_catpvs(sv, "{loc}");
9366 if (flags & ANYOF_FOLD)
9367 sv_catpvs(sv, "{i}");
9368 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9369 if (flags & ANYOF_INVERT)
9372 /* output what the standard cp 0-255 bitmap matches */
9373 for (i = 0; i <= 256; i++) {
9374 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9375 if (rangestart == -1)
9377 } else if (rangestart != -1) {
9378 if (i <= rangestart + 3)
9379 for (; rangestart < i; rangestart++)
9380 put_byte(sv, rangestart);
9382 put_byte(sv, rangestart);
9384 put_byte(sv, i - 1);
9391 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9392 /* output any special charclass tests (used mostly under use locale) */
9393 if (o->flags & ANYOF_CLASS)
9394 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9395 if (ANYOF_CLASS_TEST(o,i)) {
9396 sv_catpv(sv, anyofs[i]);
9400 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9402 /* output information about the unicode matching */
9403 if (flags & ANYOF_UNICODE)
9404 sv_catpvs(sv, "{unicode}");
9405 else if (flags & ANYOF_UNICODE_ALL)
9406 sv_catpvs(sv, "{unicode_all}");
9410 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9414 U8 s[UTF8_MAXBYTES_CASE+1];
9416 for (i = 0; i <= 256; i++) { /* just the first 256 */
9417 uvchr_to_utf8(s, i);
9419 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9420 if (rangestart == -1)
9422 } else if (rangestart != -1) {
9423 if (i <= rangestart + 3)
9424 for (; rangestart < i; rangestart++) {
9425 const U8 * const e = uvchr_to_utf8(s,rangestart);
9427 for(p = s; p < e; p++)
9431 const U8 *e = uvchr_to_utf8(s,rangestart);
9433 for (p = s; p < e; p++)
9436 e = uvchr_to_utf8(s, i-1);
9437 for (p = s; p < e; p++)
9444 sv_catpvs(sv, "..."); /* et cetera */
9448 char *s = savesvpv(lv);
9449 char * const origs = s;
9451 while (*s && *s != '\n')
9455 const char * const t = ++s;
9473 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9475 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9476 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9478 PERL_UNUSED_CONTEXT;
9479 PERL_UNUSED_ARG(sv);
9481 PERL_UNUSED_ARG(prog);
9482 #endif /* DEBUGGING */
9486 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9487 { /* Assume that RE_INTUIT is set */
9489 struct regexp *const prog = (struct regexp *)SvANY(r);
9490 GET_RE_DEBUG_FLAGS_DECL;
9492 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9493 PERL_UNUSED_CONTEXT;
9497 const char * const s = SvPV_nolen_const(prog->check_substr
9498 ? prog->check_substr : prog->check_utf8);
9500 if (!PL_colorset) reginitcolors();
9501 PerlIO_printf(Perl_debug_log,
9502 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9504 prog->check_substr ? "" : "utf8 ",
9505 PL_colors[5],PL_colors[0],
9508 (strlen(s) > 60 ? "..." : ""));
9511 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9517 handles refcounting and freeing the perl core regexp structure. When
9518 it is necessary to actually free the structure the first thing it
9519 does is call the 'free' method of the regexp_engine associated to to
9520 the regexp, allowing the handling of the void *pprivate; member
9521 first. (This routine is not overridable by extensions, which is why
9522 the extensions free is called first.)
9524 See regdupe and regdupe_internal if you change anything here.
9526 #ifndef PERL_IN_XSUB_RE
9528 Perl_pregfree(pTHX_ REGEXP *r)
9534 Perl_pregfree2(pTHX_ REGEXP *rx)
9537 struct regexp *const r = (struct regexp *)SvANY(rx);
9538 GET_RE_DEBUG_FLAGS_DECL;
9540 PERL_ARGS_ASSERT_PREGFREE2;
9543 ReREFCNT_dec(r->mother_re);
9545 CALLREGFREE_PVT(rx); /* free the private data */
9546 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9549 SvREFCNT_dec(r->anchored_substr);
9550 SvREFCNT_dec(r->anchored_utf8);
9551 SvREFCNT_dec(r->float_substr);
9552 SvREFCNT_dec(r->float_utf8);
9553 Safefree(r->substrs);
9555 RX_MATCH_COPY_FREE(rx);
9556 #ifdef PERL_OLD_COPY_ON_WRITE
9557 SvREFCNT_dec(r->saved_copy);
9564 This is a hacky workaround to the structural issue of match results
9565 being stored in the regexp structure which is in turn stored in
9566 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9567 could be PL_curpm in multiple contexts, and could require multiple
9568 result sets being associated with the pattern simultaneously, such
9569 as when doing a recursive match with (??{$qr})
9571 The solution is to make a lightweight copy of the regexp structure
9572 when a qr// is returned from the code executed by (??{$qr}) this
9573 lightweight copy doesnt actually own any of its data except for
9574 the starp/end and the actual regexp structure itself.
9580 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
9583 struct regexp *const r = (struct regexp *)SvANY(rx);
9584 register const I32 npar = r->nparens+1;
9586 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9589 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9590 ret = (struct regexp *)SvANY(ret_x);
9592 (void)ReREFCNT_inc(rx);
9593 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9594 by pointing directly at the buffer, but flagging that the allocated
9595 space in the copy is zero. As we've just done a struct copy, it's now
9596 a case of zero-ing that, rather than copying the current length. */
9597 SvPV_set(ret_x, RX_WRAPPED(rx));
9598 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9599 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
9600 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
9601 SvLEN_set(ret_x, 0);
9602 SvSTASH_set(ret_x, NULL);
9603 SvMAGIC_set(ret_x, NULL);
9604 Newx(ret->offs, npar, regexp_paren_pair);
9605 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9607 Newx(ret->substrs, 1, struct reg_substr_data);
9608 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9610 SvREFCNT_inc_void(ret->anchored_substr);
9611 SvREFCNT_inc_void(ret->anchored_utf8);
9612 SvREFCNT_inc_void(ret->float_substr);
9613 SvREFCNT_inc_void(ret->float_utf8);
9615 /* check_substr and check_utf8, if non-NULL, point to either their
9616 anchored or float namesakes, and don't hold a second reference. */
9618 RX_MATCH_COPIED_off(ret_x);
9619 #ifdef PERL_OLD_COPY_ON_WRITE
9620 ret->saved_copy = NULL;
9622 ret->mother_re = rx;
9628 /* regfree_internal()
9630 Free the private data in a regexp. This is overloadable by
9631 extensions. Perl takes care of the regexp structure in pregfree(),
9632 this covers the *pprivate pointer which technically perldoesnt
9633 know about, however of course we have to handle the
9634 regexp_internal structure when no extension is in use.
9636 Note this is called before freeing anything in the regexp
9641 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9644 struct regexp *const r = (struct regexp *)SvANY(rx);
9646 GET_RE_DEBUG_FLAGS_DECL;
9648 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9654 SV *dsv= sv_newmortal();
9655 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9656 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9657 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9658 PL_colors[4],PL_colors[5],s);
9661 #ifdef RE_TRACK_PATTERN_OFFSETS
9663 Safefree(ri->u.offsets); /* 20010421 MJD */
9666 int n = ri->data->count;
9667 PAD* new_comppad = NULL;
9672 /* If you add a ->what type here, update the comment in regcomp.h */
9673 switch (ri->data->what[n]) {
9678 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
9681 Safefree(ri->data->data[n]);
9684 new_comppad = MUTABLE_AV(ri->data->data[n]);
9687 if (new_comppad == NULL)
9688 Perl_croak(aTHX_ "panic: pregfree comppad");
9689 PAD_SAVE_LOCAL(old_comppad,
9690 /* Watch out for global destruction's random ordering. */
9691 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9694 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9697 op_free((OP_4tree*)ri->data->data[n]);
9699 PAD_RESTORE_LOCAL(old_comppad);
9700 SvREFCNT_dec(MUTABLE_SV(new_comppad));
9706 { /* Aho Corasick add-on structure for a trie node.
9707 Used in stclass optimization only */
9709 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9711 refcount = --aho->refcount;
9714 PerlMemShared_free(aho->states);
9715 PerlMemShared_free(aho->fail);
9716 /* do this last!!!! */
9717 PerlMemShared_free(ri->data->data[n]);
9718 PerlMemShared_free(ri->regstclass);
9724 /* trie structure. */
9726 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9728 refcount = --trie->refcount;
9731 PerlMemShared_free(trie->charmap);
9732 PerlMemShared_free(trie->states);
9733 PerlMemShared_free(trie->trans);
9735 PerlMemShared_free(trie->bitmap);
9737 PerlMemShared_free(trie->jump);
9738 PerlMemShared_free(trie->wordinfo);
9739 /* do this last!!!! */
9740 PerlMemShared_free(ri->data->data[n]);
9745 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9748 Safefree(ri->data->what);
9755 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
9756 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
9757 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9760 re_dup - duplicate a regexp.
9762 This routine is expected to clone a given regexp structure. It is only
9763 compiled under USE_ITHREADS.
9765 After all of the core data stored in struct regexp is duplicated
9766 the regexp_engine.dupe method is used to copy any private data
9767 stored in the *pprivate pointer. This allows extensions to handle
9768 any duplication it needs to do.
9770 See pregfree() and regfree_internal() if you change anything here.
9772 #if defined(USE_ITHREADS)
9773 #ifndef PERL_IN_XSUB_RE
9775 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9779 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9780 struct regexp *ret = (struct regexp *)SvANY(dstr);
9782 PERL_ARGS_ASSERT_RE_DUP_GUTS;
9784 npar = r->nparens+1;
9785 Newx(ret->offs, npar, regexp_paren_pair);
9786 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9788 /* no need to copy these */
9789 Newx(ret->swap, npar, regexp_paren_pair);
9793 /* Do it this way to avoid reading from *r after the StructCopy().
9794 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9795 cache, it doesn't matter. */
9796 const bool anchored = r->check_substr
9797 ? r->check_substr == r->anchored_substr
9798 : r->check_utf8 == r->anchored_utf8;
9799 Newx(ret->substrs, 1, struct reg_substr_data);
9800 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9802 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9803 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9804 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9805 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9807 /* check_substr and check_utf8, if non-NULL, point to either their
9808 anchored or float namesakes, and don't hold a second reference. */
9810 if (ret->check_substr) {
9812 assert(r->check_utf8 == r->anchored_utf8);
9813 ret->check_substr = ret->anchored_substr;
9814 ret->check_utf8 = ret->anchored_utf8;
9816 assert(r->check_substr == r->float_substr);
9817 assert(r->check_utf8 == r->float_utf8);
9818 ret->check_substr = ret->float_substr;
9819 ret->check_utf8 = ret->float_utf8;
9821 } else if (ret->check_utf8) {
9823 ret->check_utf8 = ret->anchored_utf8;
9825 ret->check_utf8 = ret->float_utf8;
9830 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9833 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9835 if (RX_MATCH_COPIED(dstr))
9836 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9839 #ifdef PERL_OLD_COPY_ON_WRITE
9840 ret->saved_copy = NULL;
9843 if (ret->mother_re) {
9844 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
9845 /* Our storage points directly to our mother regexp, but that's
9846 1: a buffer in a different thread
9847 2: something we no longer hold a reference on
9848 so we need to copy it locally. */
9849 /* Note we need to sue SvCUR() on our mother_re, because it, in
9850 turn, may well be pointing to its own mother_re. */
9851 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
9852 SvCUR(ret->mother_re)+1));
9853 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
9855 ret->mother_re = NULL;
9859 #endif /* PERL_IN_XSUB_RE */
9864 This is the internal complement to regdupe() which is used to copy
9865 the structure pointed to by the *pprivate pointer in the regexp.
9866 This is the core version of the extension overridable cloning hook.
9867 The regexp structure being duplicated will be copied by perl prior
9868 to this and will be provided as the regexp *r argument, however
9869 with the /old/ structures pprivate pointer value. Thus this routine
9870 may override any copying normally done by perl.
9872 It returns a pointer to the new regexp_internal structure.
9876 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
9879 struct regexp *const r = (struct regexp *)SvANY(rx);
9880 regexp_internal *reti;
9884 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
9886 npar = r->nparens+1;
9889 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
9890 Copy(ri->program, reti->program, len+1, regnode);
9893 reti->regstclass = NULL;
9897 const int count = ri->data->count;
9900 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9901 char, struct reg_data);
9902 Newx(d->what, count, U8);
9905 for (i = 0; i < count; i++) {
9906 d->what[i] = ri->data->what[i];
9907 switch (d->what[i]) {
9908 /* legal options are one of: sSfpontTua
9909 see also regcomp.h and pregfree() */
9910 case 'a': /* actually an AV, but the dup function is identical. */
9913 case 'p': /* actually an AV, but the dup function is identical. */
9914 case 'u': /* actually an HV, but the dup function is identical. */
9915 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
9918 /* This is cheating. */
9919 Newx(d->data[i], 1, struct regnode_charclass_class);
9920 StructCopy(ri->data->data[i], d->data[i],
9921 struct regnode_charclass_class);
9922 reti->regstclass = (regnode*)d->data[i];
9925 /* Compiled op trees are readonly and in shared memory,
9926 and can thus be shared without duplication. */
9928 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9932 /* Trie stclasses are readonly and can thus be shared
9933 * without duplication. We free the stclass in pregfree
9934 * when the corresponding reg_ac_data struct is freed.
9936 reti->regstclass= ri->regstclass;
9940 ((reg_trie_data*)ri->data->data[i])->refcount++;
9944 d->data[i] = ri->data->data[i];
9947 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9956 reti->name_list_idx = ri->name_list_idx;
9958 #ifdef RE_TRACK_PATTERN_OFFSETS
9959 if (ri->u.offsets) {
9960 Newx(reti->u.offsets, 2*len+1, U32);
9961 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9964 SetProgLen(reti,len);
9970 #endif /* USE_ITHREADS */
9972 #ifndef PERL_IN_XSUB_RE
9975 - regnext - dig the "next" pointer out of a node
9978 Perl_regnext(pTHX_ register regnode *p)
9981 register I32 offset;
9986 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
9987 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
9990 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9999 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
10002 STRLEN l1 = strlen(pat1);
10003 STRLEN l2 = strlen(pat2);
10006 const char *message;
10008 PERL_ARGS_ASSERT_RE_CROAK2;
10014 Copy(pat1, buf, l1 , char);
10015 Copy(pat2, buf + l1, l2 , char);
10016 buf[l1 + l2] = '\n';
10017 buf[l1 + l2 + 1] = '\0';
10019 /* ANSI variant takes additional second argument */
10020 va_start(args, pat2);
10024 msv = vmess(buf, &args);
10026 message = SvPV_const(msv,l1);
10029 Copy(message, buf, l1 , char);
10030 buf[l1-1] = '\0'; /* Overwrite \n */
10031 Perl_croak(aTHX_ "%s", buf);
10034 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
10036 #ifndef PERL_IN_XSUB_RE
10038 Perl_save_re_context(pTHX)
10042 struct re_save_state *state;
10044 SAVEVPTR(PL_curcop);
10045 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
10047 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
10048 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
10049 SSPUSHUV(SAVEt_RE_STATE);
10051 Copy(&PL_reg_state, state, 1, struct re_save_state);
10053 PL_reg_start_tmp = 0;
10054 PL_reg_start_tmpl = 0;
10055 PL_reg_oldsaved = NULL;
10056 PL_reg_oldsavedlen = 0;
10057 PL_reg_maxiter = 0;
10058 PL_reg_leftiter = 0;
10059 PL_reg_poscache = NULL;
10060 PL_reg_poscache_size = 0;
10061 #ifdef PERL_OLD_COPY_ON_WRITE
10065 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
10067 const REGEXP * const rx = PM_GETRE(PL_curpm);
10070 for (i = 1; i <= RX_NPARENS(rx); i++) {
10071 char digits[TYPE_CHARS(long)];
10072 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
10073 GV *const *const gvp
10074 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
10077 GV * const gv = *gvp;
10078 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
10088 clear_re(pTHX_ void *r)
10091 ReREFCNT_dec((REGEXP *)r);
10097 S_put_byte(pTHX_ SV *sv, int c)
10099 PERL_ARGS_ASSERT_PUT_BYTE;
10101 /* Our definition of isPRINT() ignores locales, so only bytes that are
10102 not part of UTF-8 are considered printable. I assume that the same
10103 holds for UTF-EBCDIC.
10104 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
10105 which Wikipedia says:
10107 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
10108 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
10109 identical, to the ASCII delete (DEL) or rubout control character.
10110 ) So the old condition can be simplified to !isPRINT(c) */
10112 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
10114 const char string = c;
10115 if (c == '-' || c == ']' || c == '\\' || c == '^')
10116 sv_catpvs(sv, "\\");
10117 sv_catpvn(sv, &string, 1);
10122 #define CLEAR_OPTSTART \
10123 if (optstart) STMT_START { \
10124 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
10128 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
10130 STATIC const regnode *
10131 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
10132 const regnode *last, const regnode *plast,
10133 SV* sv, I32 indent, U32 depth)
10136 register U8 op = PSEUDO; /* Arbitrary non-END op. */
10137 register const regnode *next;
10138 const regnode *optstart= NULL;
10140 RXi_GET_DECL(r,ri);
10141 GET_RE_DEBUG_FLAGS_DECL;
10143 PERL_ARGS_ASSERT_DUMPUNTIL;
10145 #ifdef DEBUG_DUMPUNTIL
10146 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
10147 last ? last-start : 0,plast ? plast-start : 0);
10150 if (plast && plast < last)
10153 while (PL_regkind[op] != END && (!last || node < last)) {
10154 /* While that wasn't END last time... */
10157 if (op == CLOSE || op == WHILEM)
10159 next = regnext((regnode *)node);
10162 if (OP(node) == OPTIMIZED) {
10163 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10170 regprop(r, sv, node);
10171 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10172 (int)(2*indent + 1), "", SvPVX_const(sv));
10174 if (OP(node) != OPTIMIZED) {
10175 if (next == NULL) /* Next ptr. */
10176 PerlIO_printf(Perl_debug_log, " (0)");
10177 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10178 PerlIO_printf(Perl_debug_log, " (FAIL)");
10180 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10181 (void)PerlIO_putc(Perl_debug_log, '\n');
10185 if (PL_regkind[(U8)op] == BRANCHJ) {
10188 register const regnode *nnode = (OP(next) == LONGJMP
10189 ? regnext((regnode *)next)
10191 if (last && nnode > last)
10193 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10196 else if (PL_regkind[(U8)op] == BRANCH) {
10198 DUMPUNTIL(NEXTOPER(node), next);
10200 else if ( PL_regkind[(U8)op] == TRIE ) {
10201 const regnode *this_trie = node;
10202 const char op = OP(node);
10203 const U32 n = ARG(node);
10204 const reg_ac_data * const ac = op>=AHOCORASICK ?
10205 (reg_ac_data *)ri->data->data[n] :
10207 const reg_trie_data * const trie =
10208 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10210 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10212 const regnode *nextbranch= NULL;
10215 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10216 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10218 PerlIO_printf(Perl_debug_log, "%*s%s ",
10219 (int)(2*(indent+3)), "",
10220 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10221 PL_colors[0], PL_colors[1],
10222 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10223 PERL_PV_PRETTY_ELLIPSES |
10224 PERL_PV_PRETTY_LTGT
10229 U16 dist= trie->jump[word_idx+1];
10230 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10231 (UV)((dist ? this_trie + dist : next) - start));
10234 nextbranch= this_trie + trie->jump[0];
10235 DUMPUNTIL(this_trie + dist, nextbranch);
10237 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10238 nextbranch= regnext((regnode *)nextbranch);
10240 PerlIO_printf(Perl_debug_log, "\n");
10243 if (last && next > last)
10248 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10249 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10250 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10252 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10254 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10256 else if ( op == PLUS || op == STAR) {
10257 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10259 else if (op == ANYOF) {
10260 /* arglen 1 + class block */
10261 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
10262 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10263 node = NEXTOPER(node);
10265 else if (PL_regkind[(U8)op] == EXACT) {
10266 /* Literal string, where present. */
10267 node += NODE_SZ_STR(node) - 1;
10268 node = NEXTOPER(node);
10271 node = NEXTOPER(node);
10272 node += regarglen[(U8)op];
10274 if (op == CURLYX || op == OPEN)
10278 #ifdef DEBUG_DUMPUNTIL
10279 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10284 #endif /* DEBUGGING */
10288 * c-indentation-style: bsd
10289 * c-basic-offset: 4
10290 * indent-tabs-mode: t
10293 * ex: set ts=8 sts=4 sw=4 noet: