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 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 characters must match before the string we are searching for.
260 Likewise when combined with minlenp and the length of the string it
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. If set 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 /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
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;
713 cl->flags |= ANYOF_FOLD;
716 /* Can match anything (initialization) */
718 S_cl_is_anything(const struct regnode_charclass_class *cl)
722 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
724 for (value = 0; value <= ANYOF_MAX; value += 2)
725 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
727 if (!(cl->flags & ANYOF_UNICODE_ALL))
729 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
734 /* Can match anything (initialization) */
736 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
738 PERL_ARGS_ASSERT_CL_INIT;
740 Zero(cl, 1, struct regnode_charclass_class);
742 cl_anything(pRExC_state, cl);
746 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
748 PERL_ARGS_ASSERT_CL_INIT_ZERO;
750 Zero(cl, 1, struct regnode_charclass_class);
752 cl_anything(pRExC_state, cl);
754 cl->flags |= ANYOF_LOCALE;
757 /* 'And' a given class with another one. Can create false positives */
758 /* We assume that cl is not inverted */
760 S_cl_and(struct regnode_charclass_class *cl,
761 const struct regnode_charclass_class *and_with)
763 PERL_ARGS_ASSERT_CL_AND;
765 assert(and_with->type == ANYOF);
766 if (!(and_with->flags & ANYOF_CLASS)
767 && !(cl->flags & ANYOF_CLASS)
768 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
769 && !(and_with->flags & ANYOF_FOLD)
770 && !(cl->flags & ANYOF_FOLD)) {
773 if (and_with->flags & ANYOF_INVERT)
774 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
775 cl->bitmap[i] &= ~and_with->bitmap[i];
777 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
778 cl->bitmap[i] &= and_with->bitmap[i];
779 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
780 if (!(and_with->flags & ANYOF_EOS))
781 cl->flags &= ~ANYOF_EOS;
783 if (!(and_with->flags & ANYOF_FOLD))
784 cl->flags &= ~ANYOF_FOLD;
786 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
787 !(and_with->flags & ANYOF_INVERT)) {
788 cl->flags &= ~ANYOF_UNICODE_ALL;
789 cl->flags |= ANYOF_UNICODE;
790 ARG_SET(cl, ARG(and_with));
792 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
793 !(and_with->flags & ANYOF_INVERT))
794 cl->flags &= ~ANYOF_UNICODE_ALL;
795 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
796 !(and_with->flags & ANYOF_INVERT))
797 cl->flags &= ~ANYOF_UNICODE;
800 /* 'OR' a given class with another one. Can create false positives */
801 /* We assume that cl is not inverted */
803 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
805 PERL_ARGS_ASSERT_CL_OR;
807 if (or_with->flags & ANYOF_INVERT) {
809 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
810 * <= (B1 | !B2) | (CL1 | !CL2)
811 * which is wasteful if CL2 is small, but we ignore CL2:
812 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
813 * XXXX Can we handle case-fold? Unclear:
814 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
815 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
817 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
818 && !(or_with->flags & ANYOF_FOLD)
819 && !(cl->flags & ANYOF_FOLD) ) {
822 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
823 cl->bitmap[i] |= ~or_with->bitmap[i];
824 } /* XXXX: logic is complicated otherwise */
826 cl_anything(pRExC_state, cl);
829 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
830 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
831 && (!(or_with->flags & ANYOF_FOLD)
832 || (cl->flags & ANYOF_FOLD)) ) {
835 /* OR char bitmap and class bitmap separately */
836 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
837 cl->bitmap[i] |= or_with->bitmap[i];
838 if (or_with->flags & ANYOF_CLASS) {
839 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
840 cl->classflags[i] |= or_with->classflags[i];
841 cl->flags |= ANYOF_CLASS;
844 else { /* XXXX: logic is complicated, leave it along for a moment. */
845 cl_anything(pRExC_state, cl);
848 if (or_with->flags & ANYOF_EOS)
849 cl->flags |= ANYOF_EOS;
851 if (or_with->flags & ANYOF_FOLD)
852 cl->flags |= ANYOF_FOLD;
854 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
855 ARG(cl) != ARG(or_with)) {
856 cl->flags |= ANYOF_UNICODE_ALL;
857 cl->flags &= ~ANYOF_UNICODE;
859 if (or_with->flags & ANYOF_UNICODE_ALL) {
860 cl->flags |= ANYOF_UNICODE_ALL;
861 cl->flags &= ~ANYOF_UNICODE;
865 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
866 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
867 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
868 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
873 dump_trie(trie,widecharmap,revcharmap)
874 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
875 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
877 These routines dump out a trie in a somewhat readable format.
878 The _interim_ variants are used for debugging the interim
879 tables that are used to generate the final compressed
880 representation which is what dump_trie expects.
882 Part of the reason for their existance is to provide a form
883 of documentation as to how the different representations function.
888 Dumps the final compressed table form of the trie to Perl_debug_log.
889 Used for debugging make_trie().
893 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
894 AV *revcharmap, U32 depth)
897 SV *sv=sv_newmortal();
898 int colwidth= widecharmap ? 6 : 4;
900 GET_RE_DEBUG_FLAGS_DECL;
902 PERL_ARGS_ASSERT_DUMP_TRIE;
904 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
905 (int)depth * 2 + 2,"",
906 "Match","Base","Ofs" );
908 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
909 SV ** const tmp = av_fetch( revcharmap, state, 0);
911 PerlIO_printf( Perl_debug_log, "%*s",
913 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
914 PL_colors[0], PL_colors[1],
915 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
916 PERL_PV_ESCAPE_FIRSTCHAR
921 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
922 (int)depth * 2 + 2,"");
924 for( state = 0 ; state < trie->uniquecharcount ; state++ )
925 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
926 PerlIO_printf( Perl_debug_log, "\n");
928 for( state = 1 ; state < trie->statecount ; state++ ) {
929 const U32 base = trie->states[ state ].trans.base;
931 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
933 if ( trie->states[ state ].wordnum ) {
934 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
936 PerlIO_printf( Perl_debug_log, "%6s", "" );
939 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
944 while( ( base + ofs < trie->uniquecharcount ) ||
945 ( base + ofs - trie->uniquecharcount < trie->lasttrans
946 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
949 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
951 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
952 if ( ( base + ofs >= trie->uniquecharcount ) &&
953 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
954 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
956 PerlIO_printf( Perl_debug_log, "%*"UVXf,
958 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
960 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
964 PerlIO_printf( Perl_debug_log, "]");
967 PerlIO_printf( Perl_debug_log, "\n" );
969 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
970 for (word=1; word <= trie->wordcount; word++) {
971 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
972 (int)word, (int)(trie->wordinfo[word].prev),
973 (int)(trie->wordinfo[word].len));
975 PerlIO_printf(Perl_debug_log, "\n" );
978 Dumps a fully constructed but uncompressed trie in list form.
979 List tries normally only are used for construction when the number of
980 possible chars (trie->uniquecharcount) is very high.
981 Used for debugging make_trie().
984 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
985 HV *widecharmap, AV *revcharmap, U32 next_alloc,
989 SV *sv=sv_newmortal();
990 int colwidth= widecharmap ? 6 : 4;
991 GET_RE_DEBUG_FLAGS_DECL;
993 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
995 /* print out the table precompression. */
996 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
997 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
998 "------:-----+-----------------\n" );
1000 for( state=1 ; state < next_alloc ; state ++ ) {
1003 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1004 (int)depth * 2 + 2,"", (UV)state );
1005 if ( ! trie->states[ state ].wordnum ) {
1006 PerlIO_printf( Perl_debug_log, "%5s| ","");
1008 PerlIO_printf( Perl_debug_log, "W%4x| ",
1009 trie->states[ state ].wordnum
1012 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1013 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
1015 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1017 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1018 PL_colors[0], PL_colors[1],
1019 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1020 PERL_PV_ESCAPE_FIRSTCHAR
1022 TRIE_LIST_ITEM(state,charid).forid,
1023 (UV)TRIE_LIST_ITEM(state,charid).newstate
1026 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1027 (int)((depth * 2) + 14), "");
1030 PerlIO_printf( Perl_debug_log, "\n");
1035 Dumps a fully constructed but uncompressed trie in table form.
1036 This is the normal DFA style state transition table, with a few
1037 twists to facilitate compression later.
1038 Used for debugging make_trie().
1041 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1042 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1047 SV *sv=sv_newmortal();
1048 int colwidth= widecharmap ? 6 : 4;
1049 GET_RE_DEBUG_FLAGS_DECL;
1051 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1054 print out the table precompression so that we can do a visual check
1055 that they are identical.
1058 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1060 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1061 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1063 PerlIO_printf( Perl_debug_log, "%*s",
1065 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1066 PL_colors[0], PL_colors[1],
1067 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1068 PERL_PV_ESCAPE_FIRSTCHAR
1074 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1076 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1077 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1080 PerlIO_printf( Perl_debug_log, "\n" );
1082 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1084 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1085 (int)depth * 2 + 2,"",
1086 (UV)TRIE_NODENUM( state ) );
1088 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1089 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1091 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1093 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1095 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1096 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1098 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1099 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1107 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1108 startbranch: the first branch in the whole branch sequence
1109 first : start branch of sequence of branch-exact nodes.
1110 May be the same as startbranch
1111 last : Thing following the last branch.
1112 May be the same as tail.
1113 tail : item following the branch sequence
1114 count : words in the sequence
1115 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1116 depth : indent depth
1118 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1120 A trie is an N'ary tree where the branches are determined by digital
1121 decomposition of the key. IE, at the root node you look up the 1st character and
1122 follow that branch repeat until you find the end of the branches. Nodes can be
1123 marked as "accepting" meaning they represent a complete word. Eg:
1127 would convert into the following structure. Numbers represent states, letters
1128 following numbers represent valid transitions on the letter from that state, if
1129 the number is in square brackets it represents an accepting state, otherwise it
1130 will be in parenthesis.
1132 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1136 (1) +-i->(6)-+-s->[7]
1138 +-s->(3)-+-h->(4)-+-e->[5]
1140 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1142 This shows that when matching against the string 'hers' we will begin at state 1
1143 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1144 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1145 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1146 single traverse. We store a mapping from accepting to state to which word was
1147 matched, and then when we have multiple possibilities we try to complete the
1148 rest of the regex in the order in which they occured in the alternation.
1150 The only prior NFA like behaviour that would be changed by the TRIE support is
1151 the silent ignoring of duplicate alternations which are of the form:
1153 / (DUPE|DUPE) X? (?{ ... }) Y /x
1155 Thus EVAL blocks follwing a trie may be called a different number of times with
1156 and without the optimisation. With the optimisations dupes will be silently
1157 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1158 the following demonstrates:
1160 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1162 which prints out 'word' three times, but
1164 'words'=~/(word|word|word)(?{ print $1 })S/
1166 which doesnt print it out at all. This is due to other optimisations kicking in.
1168 Example of what happens on a structural level:
1170 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1172 1: CURLYM[1] {1,32767}(18)
1183 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1184 and should turn into:
1186 1: CURLYM[1] {1,32767}(18)
1188 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1196 Cases where tail != last would be like /(?foo|bar)baz/:
1206 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1207 and would end up looking like:
1210 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1217 d = uvuni_to_utf8_flags(d, uv, 0);
1219 is the recommended Unicode-aware way of saying
1224 #define TRIE_STORE_REVCHAR \
1227 SV *zlopp = newSV(2); \
1228 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1229 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1230 SvCUR_set(zlopp, kapow - flrbbbbb); \
1233 av_push(revcharmap, zlopp); \
1235 char ooooff = (char)uvc; \
1236 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1240 #define TRIE_READ_CHAR STMT_START { \
1244 if ( foldlen > 0 ) { \
1245 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1250 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1251 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1252 foldlen -= UNISKIP( uvc ); \
1253 scan = foldbuf + UNISKIP( uvc ); \
1256 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1266 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1267 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1268 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1269 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1271 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1272 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1273 TRIE_LIST_CUR( state )++; \
1276 #define TRIE_LIST_NEW(state) STMT_START { \
1277 Newxz( trie->states[ state ].trans.list, \
1278 4, reg_trie_trans_le ); \
1279 TRIE_LIST_CUR( state ) = 1; \
1280 TRIE_LIST_LEN( state ) = 4; \
1283 #define TRIE_HANDLE_WORD(state) STMT_START { \
1284 U16 dupe= trie->states[ state ].wordnum; \
1285 regnode * const noper_next = regnext( noper ); \
1288 /* store the word for dumping */ \
1290 if (OP(noper) != NOTHING) \
1291 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1293 tmp = newSVpvn_utf8( "", 0, UTF ); \
1294 av_push( trie_words, tmp ); \
1298 trie->wordinfo[curword].prev = 0; \
1299 trie->wordinfo[curword].len = wordlen; \
1300 trie->wordinfo[curword].accept = state; \
1302 if ( noper_next < tail ) { \
1304 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1305 trie->jump[curword] = (U16)(noper_next - convert); \
1307 jumper = noper_next; \
1309 nextbranch= regnext(cur); \
1313 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1314 /* chain, so that when the bits of chain are later */\
1315 /* linked together, the dups appear in the chain */\
1316 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1317 trie->wordinfo[dupe].prev = curword; \
1319 /* we haven't inserted this word yet. */ \
1320 trie->states[ state ].wordnum = curword; \
1325 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1326 ( ( base + charid >= ucharcount \
1327 && base + charid < ubound \
1328 && state == trie->trans[ base - ucharcount + charid ].check \
1329 && trie->trans[ base - ucharcount + charid ].next ) \
1330 ? trie->trans[ base - ucharcount + charid ].next \
1331 : ( state==1 ? special : 0 ) \
1335 #define MADE_JUMP_TRIE 2
1336 #define MADE_EXACT_TRIE 4
1339 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1342 /* first pass, loop through and scan words */
1343 reg_trie_data *trie;
1344 HV *widecharmap = NULL;
1345 AV *revcharmap = newAV();
1347 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1352 regnode *jumper = NULL;
1353 regnode *nextbranch = NULL;
1354 regnode *convert = NULL;
1355 U32 *prev_states; /* temp array mapping each state to previous one */
1356 /* we just use folder as a flag in utf8 */
1357 const U8 * const folder = ( flags == EXACTF
1359 : ( flags == EXACTFL
1366 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1367 AV *trie_words = NULL;
1368 /* along with revcharmap, this only used during construction but both are
1369 * useful during debugging so we store them in the struct when debugging.
1372 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1373 STRLEN trie_charcount=0;
1375 SV *re_trie_maxbuff;
1376 GET_RE_DEBUG_FLAGS_DECL;
1378 PERL_ARGS_ASSERT_MAKE_TRIE;
1380 PERL_UNUSED_ARG(depth);
1383 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1385 trie->startstate = 1;
1386 trie->wordcount = word_count;
1387 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1388 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1389 if (!(UTF && folder))
1390 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1391 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1392 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1395 trie_words = newAV();
1398 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1399 if (!SvIOK(re_trie_maxbuff)) {
1400 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1403 PerlIO_printf( Perl_debug_log,
1404 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1405 (int)depth * 2 + 2, "",
1406 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1407 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1411 /* Find the node we are going to overwrite */
1412 if ( first == startbranch && OP( last ) != BRANCH ) {
1413 /* whole branch chain */
1416 /* branch sub-chain */
1417 convert = NEXTOPER( first );
1420 /* -- First loop and Setup --
1422 We first traverse the branches and scan each word to determine if it
1423 contains widechars, and how many unique chars there are, this is
1424 important as we have to build a table with at least as many columns as we
1427 We use an array of integers to represent the character codes 0..255
1428 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1429 native representation of the character value as the key and IV's for the
1432 *TODO* If we keep track of how many times each character is used we can
1433 remap the columns so that the table compression later on is more
1434 efficient in terms of memory by ensuring the most common value is in the
1435 middle and the least common are on the outside. IMO this would be better
1436 than a most to least common mapping as theres a decent chance the most
1437 common letter will share a node with the least common, meaning the node
1438 will not be compressable. With a middle is most common approach the worst
1439 case is when we have the least common nodes twice.
1443 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1444 regnode * const noper = NEXTOPER( cur );
1445 const U8 *uc = (U8*)STRING( noper );
1446 const U8 * const e = uc + STR_LEN( noper );
1448 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1449 const U8 *scan = (U8*)NULL;
1450 U32 wordlen = 0; /* required init */
1452 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1454 if (OP(noper) == NOTHING) {
1458 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1459 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1460 regardless of encoding */
1462 for ( ; uc < e ; uc += len ) {
1463 TRIE_CHARCOUNT(trie)++;
1467 if ( !trie->charmap[ uvc ] ) {
1468 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1470 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1474 /* store the codepoint in the bitmap, and if its ascii
1475 also store its folded equivelent. */
1476 TRIE_BITMAP_SET(trie,uvc);
1478 /* store the folded codepoint */
1479 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1482 /* store first byte of utf8 representation of
1483 codepoints in the 127 < uvc < 256 range */
1484 if (127 < uvc && uvc < 192) {
1485 TRIE_BITMAP_SET(trie,194);
1486 } else if (191 < uvc ) {
1487 TRIE_BITMAP_SET(trie,195);
1488 /* && uvc < 256 -- we know uvc is < 256 already */
1491 set_bit = 0; /* We've done our bit :-) */
1496 widecharmap = newHV();
1498 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1501 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1503 if ( !SvTRUE( *svpp ) ) {
1504 sv_setiv( *svpp, ++trie->uniquecharcount );
1509 if( cur == first ) {
1512 } else if (chars < trie->minlen) {
1514 } else if (chars > trie->maxlen) {
1518 } /* end first pass */
1519 DEBUG_TRIE_COMPILE_r(
1520 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1521 (int)depth * 2 + 2,"",
1522 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1523 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1524 (int)trie->minlen, (int)trie->maxlen )
1528 We now know what we are dealing with in terms of unique chars and
1529 string sizes so we can calculate how much memory a naive
1530 representation using a flat table will take. If it's over a reasonable
1531 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1532 conservative but potentially much slower representation using an array
1535 At the end we convert both representations into the same compressed
1536 form that will be used in regexec.c for matching with. The latter
1537 is a form that cannot be used to construct with but has memory
1538 properties similar to the list form and access properties similar
1539 to the table form making it both suitable for fast searches and
1540 small enough that its feasable to store for the duration of a program.
1542 See the comment in the code where the compressed table is produced
1543 inplace from the flat tabe representation for an explanation of how
1544 the compression works.
1549 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1552 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1554 Second Pass -- Array Of Lists Representation
1556 Each state will be represented by a list of charid:state records
1557 (reg_trie_trans_le) the first such element holds the CUR and LEN
1558 points of the allocated array. (See defines above).
1560 We build the initial structure using the lists, and then convert
1561 it into the compressed table form which allows faster lookups
1562 (but cant be modified once converted).
1565 STRLEN transcount = 1;
1567 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1568 "%*sCompiling trie using list compiler\n",
1569 (int)depth * 2 + 2, ""));
1571 trie->states = (reg_trie_state *)
1572 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1573 sizeof(reg_trie_state) );
1577 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1579 regnode * const noper = NEXTOPER( cur );
1580 U8 *uc = (U8*)STRING( noper );
1581 const U8 * const e = uc + STR_LEN( noper );
1582 U32 state = 1; /* required init */
1583 U16 charid = 0; /* sanity init */
1584 U8 *scan = (U8*)NULL; /* sanity init */
1585 STRLEN foldlen = 0; /* required init */
1586 U32 wordlen = 0; /* required init */
1587 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1589 if (OP(noper) != NOTHING) {
1590 for ( ; uc < e ; uc += len ) {
1595 charid = trie->charmap[ uvc ];
1597 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1601 charid=(U16)SvIV( *svpp );
1604 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1611 if ( !trie->states[ state ].trans.list ) {
1612 TRIE_LIST_NEW( state );
1614 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1615 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1616 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1621 newstate = next_alloc++;
1622 prev_states[newstate] = state;
1623 TRIE_LIST_PUSH( state, charid, newstate );
1628 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1632 TRIE_HANDLE_WORD(state);
1634 } /* end second pass */
1636 /* next alloc is the NEXT state to be allocated */
1637 trie->statecount = next_alloc;
1638 trie->states = (reg_trie_state *)
1639 PerlMemShared_realloc( trie->states,
1641 * sizeof(reg_trie_state) );
1643 /* and now dump it out before we compress it */
1644 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1645 revcharmap, next_alloc,
1649 trie->trans = (reg_trie_trans *)
1650 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1657 for( state=1 ; state < next_alloc ; state ++ ) {
1661 DEBUG_TRIE_COMPILE_MORE_r(
1662 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1666 if (trie->states[state].trans.list) {
1667 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1671 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1672 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1673 if ( forid < minid ) {
1675 } else if ( forid > maxid ) {
1679 if ( transcount < tp + maxid - minid + 1) {
1681 trie->trans = (reg_trie_trans *)
1682 PerlMemShared_realloc( trie->trans,
1684 * sizeof(reg_trie_trans) );
1685 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1687 base = trie->uniquecharcount + tp - minid;
1688 if ( maxid == minid ) {
1690 for ( ; zp < tp ; zp++ ) {
1691 if ( ! trie->trans[ zp ].next ) {
1692 base = trie->uniquecharcount + zp - minid;
1693 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1694 trie->trans[ zp ].check = state;
1700 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1701 trie->trans[ tp ].check = state;
1706 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1707 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1708 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1709 trie->trans[ tid ].check = state;
1711 tp += ( maxid - minid + 1 );
1713 Safefree(trie->states[ state ].trans.list);
1716 DEBUG_TRIE_COMPILE_MORE_r(
1717 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1720 trie->states[ state ].trans.base=base;
1722 trie->lasttrans = tp + 1;
1726 Second Pass -- Flat Table Representation.
1728 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1729 We know that we will need Charcount+1 trans at most to store the data
1730 (one row per char at worst case) So we preallocate both structures
1731 assuming worst case.
1733 We then construct the trie using only the .next slots of the entry
1736 We use the .check field of the first entry of the node temporarily to
1737 make compression both faster and easier by keeping track of how many non
1738 zero fields are in the node.
1740 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1743 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1744 number representing the first entry of the node, and state as a
1745 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1746 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1747 are 2 entrys per node. eg:
1755 The table is internally in the right hand, idx form. However as we also
1756 have to deal with the states array which is indexed by nodenum we have to
1757 use TRIE_NODENUM() to convert.
1760 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1761 "%*sCompiling trie using table compiler\n",
1762 (int)depth * 2 + 2, ""));
1764 trie->trans = (reg_trie_trans *)
1765 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1766 * trie->uniquecharcount + 1,
1767 sizeof(reg_trie_trans) );
1768 trie->states = (reg_trie_state *)
1769 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1770 sizeof(reg_trie_state) );
1771 next_alloc = trie->uniquecharcount + 1;
1774 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1776 regnode * const noper = NEXTOPER( cur );
1777 const U8 *uc = (U8*)STRING( noper );
1778 const U8 * const e = uc + STR_LEN( noper );
1780 U32 state = 1; /* required init */
1782 U16 charid = 0; /* sanity init */
1783 U32 accept_state = 0; /* sanity init */
1784 U8 *scan = (U8*)NULL; /* sanity init */
1786 STRLEN foldlen = 0; /* required init */
1787 U32 wordlen = 0; /* required init */
1788 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1790 if ( OP(noper) != NOTHING ) {
1791 for ( ; uc < e ; uc += len ) {
1796 charid = trie->charmap[ uvc ];
1798 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1799 charid = svpp ? (U16)SvIV(*svpp) : 0;
1803 if ( !trie->trans[ state + charid ].next ) {
1804 trie->trans[ state + charid ].next = next_alloc;
1805 trie->trans[ state ].check++;
1806 prev_states[TRIE_NODENUM(next_alloc)]
1807 = TRIE_NODENUM(state);
1808 next_alloc += trie->uniquecharcount;
1810 state = trie->trans[ state + charid ].next;
1812 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1814 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1817 accept_state = TRIE_NODENUM( state );
1818 TRIE_HANDLE_WORD(accept_state);
1820 } /* end second pass */
1822 /* and now dump it out before we compress it */
1823 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1825 next_alloc, depth+1));
1829 * Inplace compress the table.*
1831 For sparse data sets the table constructed by the trie algorithm will
1832 be mostly 0/FAIL transitions or to put it another way mostly empty.
1833 (Note that leaf nodes will not contain any transitions.)
1835 This algorithm compresses the tables by eliminating most such
1836 transitions, at the cost of a modest bit of extra work during lookup:
1838 - Each states[] entry contains a .base field which indicates the
1839 index in the state[] array wheres its transition data is stored.
1841 - If .base is 0 there are no valid transitions from that node.
1843 - If .base is nonzero then charid is added to it to find an entry in
1846 -If trans[states[state].base+charid].check!=state then the
1847 transition is taken to be a 0/Fail transition. Thus if there are fail
1848 transitions at the front of the node then the .base offset will point
1849 somewhere inside the previous nodes data (or maybe even into a node
1850 even earlier), but the .check field determines if the transition is
1854 The following process inplace converts the table to the compressed
1855 table: We first do not compress the root node 1,and mark all its
1856 .check pointers as 1 and set its .base pointer as 1 as well. This
1857 allows us to do a DFA construction from the compressed table later,
1858 and ensures that any .base pointers we calculate later are greater
1861 - We set 'pos' to indicate the first entry of the second node.
1863 - We then iterate over the columns of the node, finding the first and
1864 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1865 and set the .check pointers accordingly, and advance pos
1866 appropriately and repreat for the next node. Note that when we copy
1867 the next pointers we have to convert them from the original
1868 NODEIDX form to NODENUM form as the former is not valid post
1871 - If a node has no transitions used we mark its base as 0 and do not
1872 advance the pos pointer.
1874 - If a node only has one transition we use a second pointer into the
1875 structure to fill in allocated fail transitions from other states.
1876 This pointer is independent of the main pointer and scans forward
1877 looking for null transitions that are allocated to a state. When it
1878 finds one it writes the single transition into the "hole". If the
1879 pointer doesnt find one the single transition is appended as normal.
1881 - Once compressed we can Renew/realloc the structures to release the
1884 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1885 specifically Fig 3.47 and the associated pseudocode.
1889 const U32 laststate = TRIE_NODENUM( next_alloc );
1892 trie->statecount = laststate;
1894 for ( state = 1 ; state < laststate ; state++ ) {
1896 const U32 stateidx = TRIE_NODEIDX( state );
1897 const U32 o_used = trie->trans[ stateidx ].check;
1898 U32 used = trie->trans[ stateidx ].check;
1899 trie->trans[ stateidx ].check = 0;
1901 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1902 if ( flag || trie->trans[ stateidx + charid ].next ) {
1903 if ( trie->trans[ stateidx + charid ].next ) {
1905 for ( ; zp < pos ; zp++ ) {
1906 if ( ! trie->trans[ zp ].next ) {
1910 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1911 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1912 trie->trans[ zp ].check = state;
1913 if ( ++zp > pos ) pos = zp;
1920 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1922 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1923 trie->trans[ pos ].check = state;
1928 trie->lasttrans = pos + 1;
1929 trie->states = (reg_trie_state *)
1930 PerlMemShared_realloc( trie->states, laststate
1931 * sizeof(reg_trie_state) );
1932 DEBUG_TRIE_COMPILE_MORE_r(
1933 PerlIO_printf( Perl_debug_log,
1934 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1935 (int)depth * 2 + 2,"",
1936 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1939 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1942 } /* end table compress */
1944 DEBUG_TRIE_COMPILE_MORE_r(
1945 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1946 (int)depth * 2 + 2, "",
1947 (UV)trie->statecount,
1948 (UV)trie->lasttrans)
1950 /* resize the trans array to remove unused space */
1951 trie->trans = (reg_trie_trans *)
1952 PerlMemShared_realloc( trie->trans, trie->lasttrans
1953 * sizeof(reg_trie_trans) );
1955 { /* Modify the program and insert the new TRIE node */
1956 U8 nodetype =(U8)(flags & 0xFF);
1960 regnode *optimize = NULL;
1961 #ifdef RE_TRACK_PATTERN_OFFSETS
1964 U32 mjd_nodelen = 0;
1965 #endif /* RE_TRACK_PATTERN_OFFSETS */
1966 #endif /* DEBUGGING */
1968 This means we convert either the first branch or the first Exact,
1969 depending on whether the thing following (in 'last') is a branch
1970 or not and whther first is the startbranch (ie is it a sub part of
1971 the alternation or is it the whole thing.)
1972 Assuming its a sub part we convert the EXACT otherwise we convert
1973 the whole branch sequence, including the first.
1975 /* Find the node we are going to overwrite */
1976 if ( first != startbranch || OP( last ) == BRANCH ) {
1977 /* branch sub-chain */
1978 NEXT_OFF( first ) = (U16)(last - first);
1979 #ifdef RE_TRACK_PATTERN_OFFSETS
1981 mjd_offset= Node_Offset((convert));
1982 mjd_nodelen= Node_Length((convert));
1985 /* whole branch chain */
1987 #ifdef RE_TRACK_PATTERN_OFFSETS
1990 const regnode *nop = NEXTOPER( convert );
1991 mjd_offset= Node_Offset((nop));
1992 mjd_nodelen= Node_Length((nop));
1996 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1997 (int)depth * 2 + 2, "",
1998 (UV)mjd_offset, (UV)mjd_nodelen)
2001 /* But first we check to see if there is a common prefix we can
2002 split out as an EXACT and put in front of the TRIE node. */
2003 trie->startstate= 1;
2004 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2006 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2010 const U32 base = trie->states[ state ].trans.base;
2012 if ( trie->states[state].wordnum )
2015 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2016 if ( ( base + ofs >= trie->uniquecharcount ) &&
2017 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2018 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2020 if ( ++count > 1 ) {
2021 SV **tmp = av_fetch( revcharmap, ofs, 0);
2022 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2023 if ( state == 1 ) break;
2025 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2027 PerlIO_printf(Perl_debug_log,
2028 "%*sNew Start State=%"UVuf" Class: [",
2029 (int)depth * 2 + 2, "",
2032 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2033 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2035 TRIE_BITMAP_SET(trie,*ch);
2037 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2039 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2043 TRIE_BITMAP_SET(trie,*ch);
2045 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2046 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2052 SV **tmp = av_fetch( revcharmap, idx, 0);
2054 char *ch = SvPV( *tmp, len );
2056 SV *sv=sv_newmortal();
2057 PerlIO_printf( Perl_debug_log,
2058 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2059 (int)depth * 2 + 2, "",
2061 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2062 PL_colors[0], PL_colors[1],
2063 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2064 PERL_PV_ESCAPE_FIRSTCHAR
2069 OP( convert ) = nodetype;
2070 str=STRING(convert);
2073 STR_LEN(convert) += len;
2079 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2084 trie->prefixlen = (state-1);
2086 regnode *n = convert+NODE_SZ_STR(convert);
2087 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2088 trie->startstate = state;
2089 trie->minlen -= (state - 1);
2090 trie->maxlen -= (state - 1);
2092 /* At least the UNICOS C compiler choked on this
2093 * being argument to DEBUG_r(), so let's just have
2096 #ifdef PERL_EXT_RE_BUILD
2102 regnode *fix = convert;
2103 U32 word = trie->wordcount;
2105 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2106 while( ++fix < n ) {
2107 Set_Node_Offset_Length(fix, 0, 0);
2110 SV ** const tmp = av_fetch( trie_words, word, 0 );
2112 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2113 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2115 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2123 NEXT_OFF(convert) = (U16)(tail - convert);
2124 DEBUG_r(optimize= n);
2130 if ( trie->maxlen ) {
2131 NEXT_OFF( convert ) = (U16)(tail - convert);
2132 ARG_SET( convert, data_slot );
2133 /* Store the offset to the first unabsorbed branch in
2134 jump[0], which is otherwise unused by the jump logic.
2135 We use this when dumping a trie and during optimisation. */
2137 trie->jump[0] = (U16)(nextbranch - convert);
2139 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2140 * and there is a bitmap
2141 * and the first "jump target" node we found leaves enough room
2142 * then convert the TRIE node into a TRIEC node, with the bitmap
2143 * embedded inline in the opcode - this is hypothetically faster.
2145 if ( !trie->states[trie->startstate].wordnum
2147 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2149 OP( convert ) = TRIEC;
2150 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2151 PerlMemShared_free(trie->bitmap);
2154 OP( convert ) = TRIE;
2156 /* store the type in the flags */
2157 convert->flags = nodetype;
2161 + regarglen[ OP( convert ) ];
2163 /* XXX We really should free up the resource in trie now,
2164 as we won't use them - (which resources?) dmq */
2166 /* needed for dumping*/
2167 DEBUG_r(if (optimize) {
2168 regnode *opt = convert;
2170 while ( ++opt < optimize) {
2171 Set_Node_Offset_Length(opt,0,0);
2174 Try to clean up some of the debris left after the
2177 while( optimize < jumper ) {
2178 mjd_nodelen += Node_Length((optimize));
2179 OP( optimize ) = OPTIMIZED;
2180 Set_Node_Offset_Length(optimize,0,0);
2183 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2185 } /* end node insert */
2187 /* Finish populating the prev field of the wordinfo array. Walk back
2188 * from each accept state until we find another accept state, and if
2189 * so, point the first word's .prev field at the second word. If the
2190 * second already has a .prev field set, stop now. This will be the
2191 * case either if we've already processed that word's accept state,
2192 * or that state had multiple words, and the overspill words were
2193 * already linked up earlier.
2200 for (word=1; word <= trie->wordcount; word++) {
2202 if (trie->wordinfo[word].prev)
2204 state = trie->wordinfo[word].accept;
2206 state = prev_states[state];
2209 prev = trie->states[state].wordnum;
2213 trie->wordinfo[word].prev = prev;
2215 Safefree(prev_states);
2219 /* and now dump out the compressed format */
2220 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2222 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2224 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2225 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2227 SvREFCNT_dec(revcharmap);
2231 : trie->startstate>1
2237 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2239 /* The Trie is constructed and compressed now so we can build a fail array if it's needed
2241 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2242 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2245 We find the fail state for each state in the trie, this state is the longest proper
2246 suffix of the current state's 'word' that is also a proper prefix of another word in our
2247 trie. State 1 represents the word '' and is thus the default fail state. This allows
2248 the DFA not to have to restart after its tried and failed a word at a given point, it
2249 simply continues as though it had been matching the other word in the first place.
2251 'abcdgu'=~/abcdefg|cdgu/
2252 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2253 fail, which would bring us to the state representing 'd' in the second word where we would
2254 try 'g' and succeed, proceeding to match 'cdgu'.
2256 /* add a fail transition */
2257 const U32 trie_offset = ARG(source);
2258 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2260 const U32 ucharcount = trie->uniquecharcount;
2261 const U32 numstates = trie->statecount;
2262 const U32 ubound = trie->lasttrans + ucharcount;
2266 U32 base = trie->states[ 1 ].trans.base;
2269 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2270 GET_RE_DEBUG_FLAGS_DECL;
2272 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2274 PERL_UNUSED_ARG(depth);
2278 ARG_SET( stclass, data_slot );
2279 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2280 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2281 aho->trie=trie_offset;
2282 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2283 Copy( trie->states, aho->states, numstates, reg_trie_state );
2284 Newxz( q, numstates, U32);
2285 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2288 /* initialize fail[0..1] to be 1 so that we always have
2289 a valid final fail state */
2290 fail[ 0 ] = fail[ 1 ] = 1;
2292 for ( charid = 0; charid < ucharcount ; charid++ ) {
2293 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2295 q[ q_write ] = newstate;
2296 /* set to point at the root */
2297 fail[ q[ q_write++ ] ]=1;
2300 while ( q_read < q_write) {
2301 const U32 cur = q[ q_read++ % numstates ];
2302 base = trie->states[ cur ].trans.base;
2304 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2305 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2307 U32 fail_state = cur;
2310 fail_state = fail[ fail_state ];
2311 fail_base = aho->states[ fail_state ].trans.base;
2312 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2314 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2315 fail[ ch_state ] = fail_state;
2316 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2318 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2320 q[ q_write++ % numstates] = ch_state;
2324 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2325 when we fail in state 1, this allows us to use the
2326 charclass scan to find a valid start char. This is based on the principle
2327 that theres a good chance the string being searched contains lots of stuff
2328 that cant be a start char.
2330 fail[ 0 ] = fail[ 1 ] = 0;
2331 DEBUG_TRIE_COMPILE_r({
2332 PerlIO_printf(Perl_debug_log,
2333 "%*sStclass Failtable (%"UVuf" states): 0",
2334 (int)(depth * 2), "", (UV)numstates
2336 for( q_read=1; q_read<numstates; q_read++ ) {
2337 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2339 PerlIO_printf(Perl_debug_log, "\n");
2342 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2347 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2348 * These need to be revisited when a newer toolchain becomes available.
2350 #if defined(__sparc64__) && defined(__GNUC__)
2351 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2352 # undef SPARC64_GCC_WORKAROUND
2353 # define SPARC64_GCC_WORKAROUND 1
2357 #define DEBUG_PEEP(str,scan,depth) \
2358 DEBUG_OPTIMISE_r({if (scan){ \
2359 SV * const mysv=sv_newmortal(); \
2360 regnode *Next = regnext(scan); \
2361 regprop(RExC_rx, mysv, scan); \
2362 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2363 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2364 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2371 #define JOIN_EXACT(scan,min,flags) \
2372 if (PL_regkind[OP(scan)] == EXACT) \
2373 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2376 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2377 /* Merge several consecutive EXACTish nodes into one. */
2378 regnode *n = regnext(scan);
2380 regnode *next = scan + NODE_SZ_STR(scan);
2384 regnode *stop = scan;
2385 GET_RE_DEBUG_FLAGS_DECL;
2387 PERL_UNUSED_ARG(depth);
2390 PERL_ARGS_ASSERT_JOIN_EXACT;
2391 #ifndef EXPERIMENTAL_INPLACESCAN
2392 PERL_UNUSED_ARG(flags);
2393 PERL_UNUSED_ARG(val);
2395 DEBUG_PEEP("join",scan,depth);
2397 /* Skip NOTHING, merge EXACT*. */
2399 ( PL_regkind[OP(n)] == NOTHING ||
2400 (stringok && (OP(n) == OP(scan))))
2402 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2404 if (OP(n) == TAIL || n > next)
2406 if (PL_regkind[OP(n)] == NOTHING) {
2407 DEBUG_PEEP("skip:",n,depth);
2408 NEXT_OFF(scan) += NEXT_OFF(n);
2409 next = n + NODE_STEP_REGNODE;
2416 else if (stringok) {
2417 const unsigned int oldl = STR_LEN(scan);
2418 regnode * const nnext = regnext(n);
2420 DEBUG_PEEP("merg",n,depth);
2423 if (oldl + STR_LEN(n) > U8_MAX)
2425 NEXT_OFF(scan) += NEXT_OFF(n);
2426 STR_LEN(scan) += STR_LEN(n);
2427 next = n + NODE_SZ_STR(n);
2428 /* Now we can overwrite *n : */
2429 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2437 #ifdef EXPERIMENTAL_INPLACESCAN
2438 if (flags && !NEXT_OFF(n)) {
2439 DEBUG_PEEP("atch", val, depth);
2440 if (reg_off_by_arg[OP(n)]) {
2441 ARG_SET(n, val - n);
2444 NEXT_OFF(n) = val - n;
2451 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2453 Two problematic code points in Unicode casefolding of EXACT nodes:
2455 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2456 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2462 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2463 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2465 This means that in case-insensitive matching (or "loose matching",
2466 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2467 length of the above casefolded versions) can match a target string
2468 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2469 This would rather mess up the minimum length computation.
2471 What we'll do is to look for the tail four bytes, and then peek
2472 at the preceding two bytes to see whether we need to decrease
2473 the minimum length by four (six minus two).
2475 Thanks to the design of UTF-8, there cannot be false matches:
2476 A sequence of valid UTF-8 bytes cannot be a subsequence of
2477 another valid sequence of UTF-8 bytes.
2480 char * const s0 = STRING(scan), *s, *t;
2481 char * const s1 = s0 + STR_LEN(scan) - 1;
2482 char * const s2 = s1 - 4;
2483 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2484 const char t0[] = "\xaf\x49\xaf\x42";
2486 const char t0[] = "\xcc\x88\xcc\x81";
2488 const char * const t1 = t0 + 3;
2491 s < s2 && (t = ninstr(s, s1, t0, t1));
2494 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2495 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2497 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2498 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2506 n = scan + NODE_SZ_STR(scan);
2508 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2515 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2519 /* REx optimizer. Converts nodes into quickier variants "in place".
2520 Finds fixed substrings. */
2522 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2523 to the position after last scanned or to NULL. */
2525 #define INIT_AND_WITHP \
2526 assert(!and_withp); \
2527 Newx(and_withp,1,struct regnode_charclass_class); \
2528 SAVEFREEPV(and_withp)
2530 /* this is a chain of data about sub patterns we are processing that
2531 need to be handled seperately/specially in study_chunk. Its so
2532 we can simulate recursion without losing state. */
2534 typedef struct scan_frame {
2535 regnode *last; /* last node to process in this frame */
2536 regnode *next; /* next node to process when last is reached */
2537 struct scan_frame *prev; /*previous frame*/
2538 I32 stop; /* what stopparen do we use */
2542 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2544 #define CASE_SYNST_FNC(nAmE) \
2546 if (flags & SCF_DO_STCLASS_AND) { \
2547 for (value = 0; value < 256; value++) \
2548 if (!is_ ## nAmE ## _cp(value)) \
2549 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2552 for (value = 0; value < 256; value++) \
2553 if (is_ ## nAmE ## _cp(value)) \
2554 ANYOF_BITMAP_SET(data->start_class, value); \
2558 if (flags & SCF_DO_STCLASS_AND) { \
2559 for (value = 0; value < 256; value++) \
2560 if (is_ ## nAmE ## _cp(value)) \
2561 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2564 for (value = 0; value < 256; value++) \
2565 if (!is_ ## nAmE ## _cp(value)) \
2566 ANYOF_BITMAP_SET(data->start_class, value); \
2573 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2574 I32 *minlenp, I32 *deltap,
2579 struct regnode_charclass_class *and_withp,
2580 U32 flags, U32 depth)
2581 /* scanp: Start here (read-write). */
2582 /* deltap: Write maxlen-minlen here. */
2583 /* last: Stop before this one. */
2584 /* data: string data about the pattern */
2585 /* stopparen: treat close N as END */
2586 /* recursed: which subroutines have we recursed into */
2587 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2590 I32 min = 0, pars = 0, code;
2591 regnode *scan = *scanp, *next;
2593 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2594 int is_inf_internal = 0; /* The studied chunk is infinite */
2595 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2596 scan_data_t data_fake;
2597 SV *re_trie_maxbuff = NULL;
2598 regnode *first_non_open = scan;
2599 I32 stopmin = I32_MAX;
2600 scan_frame *frame = NULL;
2601 GET_RE_DEBUG_FLAGS_DECL;
2603 PERL_ARGS_ASSERT_STUDY_CHUNK;
2606 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2610 while (first_non_open && OP(first_non_open) == OPEN)
2611 first_non_open=regnext(first_non_open);
2616 while ( scan && OP(scan) != END && scan < last ){
2617 /* Peephole optimizer: */
2618 DEBUG_STUDYDATA("Peep:", data,depth);
2619 DEBUG_PEEP("Peep",scan,depth);
2620 JOIN_EXACT(scan,&min,0);
2622 /* Follow the next-chain of the current node and optimize
2623 away all the NOTHINGs from it. */
2624 if (OP(scan) != CURLYX) {
2625 const int max = (reg_off_by_arg[OP(scan)]
2627 /* I32 may be smaller than U16 on CRAYs! */
2628 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2629 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2633 /* Skip NOTHING and LONGJMP. */
2634 while ((n = regnext(n))
2635 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2636 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2637 && off + noff < max)
2639 if (reg_off_by_arg[OP(scan)])
2642 NEXT_OFF(scan) = off;
2647 /* The principal pseudo-switch. Cannot be a switch, since we
2648 look into several different things. */
2649 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2650 || OP(scan) == IFTHEN) {
2651 next = regnext(scan);
2653 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2655 if (OP(next) == code || code == IFTHEN) {
2656 /* NOTE - There is similar code to this block below for handling
2657 TRIE nodes on a re-study. If you change stuff here check there
2659 I32 max1 = 0, min1 = I32_MAX, num = 0;
2660 struct regnode_charclass_class accum;
2661 regnode * const startbranch=scan;
2663 if (flags & SCF_DO_SUBSTR)
2664 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2665 if (flags & SCF_DO_STCLASS)
2666 cl_init_zero(pRExC_state, &accum);
2668 while (OP(scan) == code) {
2669 I32 deltanext, minnext, f = 0, fake;
2670 struct regnode_charclass_class this_class;
2673 data_fake.flags = 0;
2675 data_fake.whilem_c = data->whilem_c;
2676 data_fake.last_closep = data->last_closep;
2679 data_fake.last_closep = &fake;
2681 data_fake.pos_delta = delta;
2682 next = regnext(scan);
2683 scan = NEXTOPER(scan);
2685 scan = NEXTOPER(scan);
2686 if (flags & SCF_DO_STCLASS) {
2687 cl_init(pRExC_state, &this_class);
2688 data_fake.start_class = &this_class;
2689 f = SCF_DO_STCLASS_AND;
2691 if (flags & SCF_WHILEM_VISITED_POS)
2692 f |= SCF_WHILEM_VISITED_POS;
2694 /* we suppose the run is continuous, last=next...*/
2695 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2697 stopparen, recursed, NULL, f,depth+1);
2700 if (max1 < minnext + deltanext)
2701 max1 = minnext + deltanext;
2702 if (deltanext == I32_MAX)
2703 is_inf = is_inf_internal = 1;
2705 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2707 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2708 if ( stopmin > minnext)
2709 stopmin = min + min1;
2710 flags &= ~SCF_DO_SUBSTR;
2712 data->flags |= SCF_SEEN_ACCEPT;
2715 if (data_fake.flags & SF_HAS_EVAL)
2716 data->flags |= SF_HAS_EVAL;
2717 data->whilem_c = data_fake.whilem_c;
2719 if (flags & SCF_DO_STCLASS)
2720 cl_or(pRExC_state, &accum, &this_class);
2722 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2724 if (flags & SCF_DO_SUBSTR) {
2725 data->pos_min += min1;
2726 data->pos_delta += max1 - min1;
2727 if (max1 != min1 || is_inf)
2728 data->longest = &(data->longest_float);
2731 delta += max1 - min1;
2732 if (flags & SCF_DO_STCLASS_OR) {
2733 cl_or(pRExC_state, data->start_class, &accum);
2735 cl_and(data->start_class, and_withp);
2736 flags &= ~SCF_DO_STCLASS;
2739 else if (flags & SCF_DO_STCLASS_AND) {
2741 cl_and(data->start_class, &accum);
2742 flags &= ~SCF_DO_STCLASS;
2745 /* Switch to OR mode: cache the old value of
2746 * data->start_class */
2748 StructCopy(data->start_class, and_withp,
2749 struct regnode_charclass_class);
2750 flags &= ~SCF_DO_STCLASS_AND;
2751 StructCopy(&accum, data->start_class,
2752 struct regnode_charclass_class);
2753 flags |= SCF_DO_STCLASS_OR;
2754 data->start_class->flags |= ANYOF_EOS;
2758 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2761 Assuming this was/is a branch we are dealing with: 'scan' now
2762 points at the item that follows the branch sequence, whatever
2763 it is. We now start at the beginning of the sequence and look
2770 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2772 If we can find such a subseqence we need to turn the first
2773 element into a trie and then add the subsequent branch exact
2774 strings to the trie.
2778 1. patterns where the whole set of branches can be converted.
2780 2. patterns where only a subset can be converted.
2782 In case 1 we can replace the whole set with a single regop
2783 for the trie. In case 2 we need to keep the start and end
2786 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2787 becomes BRANCH TRIE; BRANCH X;
2789 There is an additional case, that being where there is a
2790 common prefix, which gets split out into an EXACT like node
2791 preceding the TRIE node.
2793 If x(1..n)==tail then we can do a simple trie, if not we make
2794 a "jump" trie, such that when we match the appropriate word
2795 we "jump" to the appopriate tail node. Essentailly we turn
2796 a nested if into a case structure of sorts.
2801 if (!re_trie_maxbuff) {
2802 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2803 if (!SvIOK(re_trie_maxbuff))
2804 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2806 if ( SvIV(re_trie_maxbuff)>=0 ) {
2808 regnode *first = (regnode *)NULL;
2809 regnode *last = (regnode *)NULL;
2810 regnode *tail = scan;
2815 SV * const mysv = sv_newmortal(); /* for dumping */
2817 /* var tail is used because there may be a TAIL
2818 regop in the way. Ie, the exacts will point to the
2819 thing following the TAIL, but the last branch will
2820 point at the TAIL. So we advance tail. If we
2821 have nested (?:) we may have to move through several
2825 while ( OP( tail ) == TAIL ) {
2826 /* this is the TAIL generated by (?:) */
2827 tail = regnext( tail );
2832 regprop(RExC_rx, mysv, tail );
2833 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2834 (int)depth * 2 + 2, "",
2835 "Looking for TRIE'able sequences. Tail node is: ",
2836 SvPV_nolen_const( mysv )
2842 step through the branches, cur represents each
2843 branch, noper is the first thing to be matched
2844 as part of that branch and noper_next is the
2845 regnext() of that node. if noper is an EXACT
2846 and noper_next is the same as scan (our current
2847 position in the regex) then the EXACT branch is
2848 a possible optimization target. Once we have
2849 two or more consequetive such branches we can
2850 create a trie of the EXACT's contents and stich
2851 it in place. If the sequence represents all of
2852 the branches we eliminate the whole thing and
2853 replace it with a single TRIE. If it is a
2854 subsequence then we need to stitch it in. This
2855 means the first branch has to remain, and needs
2856 to be repointed at the item on the branch chain
2857 following the last branch optimized. This could
2858 be either a BRANCH, in which case the
2859 subsequence is internal, or it could be the
2860 item following the branch sequence in which
2861 case the subsequence is at the end.
2865 /* dont use tail as the end marker for this traverse */
2866 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2867 regnode * const noper = NEXTOPER( cur );
2868 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2869 regnode * const noper_next = regnext( noper );
2873 regprop(RExC_rx, mysv, cur);
2874 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2875 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2877 regprop(RExC_rx, mysv, noper);
2878 PerlIO_printf( Perl_debug_log, " -> %s",
2879 SvPV_nolen_const(mysv));
2882 regprop(RExC_rx, mysv, noper_next );
2883 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2884 SvPV_nolen_const(mysv));
2886 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2887 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2889 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2890 : PL_regkind[ OP( noper ) ] == EXACT )
2891 || OP(noper) == NOTHING )
2893 && noper_next == tail
2898 if ( !first || optype == NOTHING ) {
2899 if (!first) first = cur;
2900 optype = OP( noper );
2906 Currently we do not believe that the trie logic can
2907 handle case insensitive matching properly when the
2908 pattern is not unicode (thus forcing unicode semantics).
2910 If/when this is fixed the following define can be swapped
2911 in below to fully enable trie logic.
2913 #define TRIE_TYPE_IS_SAFE 1
2916 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2918 if ( last && TRIE_TYPE_IS_SAFE ) {
2919 make_trie( pRExC_state,
2920 startbranch, first, cur, tail, count,
2923 if ( PL_regkind[ OP( noper ) ] == EXACT
2925 && noper_next == tail
2930 optype = OP( noper );
2940 regprop(RExC_rx, mysv, cur);
2941 PerlIO_printf( Perl_debug_log,
2942 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2943 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2947 if ( last && TRIE_TYPE_IS_SAFE ) {
2948 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2949 #ifdef TRIE_STUDY_OPT
2950 if ( ((made == MADE_EXACT_TRIE &&
2951 startbranch == first)
2952 || ( first_non_open == first )) &&
2954 flags |= SCF_TRIE_RESTUDY;
2955 if ( startbranch == first
2958 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2968 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2969 scan = NEXTOPER(NEXTOPER(scan));
2970 } else /* single branch is optimized. */
2971 scan = NEXTOPER(scan);
2973 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2974 scan_frame *newframe = NULL;
2979 if (OP(scan) != SUSPEND) {
2980 /* set the pointer */
2981 if (OP(scan) == GOSUB) {
2983 RExC_recurse[ARG2L(scan)] = scan;
2984 start = RExC_open_parens[paren-1];
2985 end = RExC_close_parens[paren-1];
2988 start = RExC_rxi->program + 1;
2992 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2993 SAVEFREEPV(recursed);
2995 if (!PAREN_TEST(recursed,paren+1)) {
2996 PAREN_SET(recursed,paren+1);
2997 Newx(newframe,1,scan_frame);
2999 if (flags & SCF_DO_SUBSTR) {
3000 SCAN_COMMIT(pRExC_state,data,minlenp);
3001 data->longest = &(data->longest_float);
3003 is_inf = is_inf_internal = 1;
3004 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3005 cl_anything(pRExC_state, data->start_class);
3006 flags &= ~SCF_DO_STCLASS;
3009 Newx(newframe,1,scan_frame);
3012 end = regnext(scan);
3017 SAVEFREEPV(newframe);
3018 newframe->next = regnext(scan);
3019 newframe->last = last;
3020 newframe->stop = stopparen;
3021 newframe->prev = frame;
3031 else if (OP(scan) == EXACT) {
3032 I32 l = STR_LEN(scan);
3035 const U8 * const s = (U8*)STRING(scan);
3036 l = utf8_length(s, s + l);
3037 uc = utf8_to_uvchr(s, NULL);
3039 uc = *((U8*)STRING(scan));
3042 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3043 /* The code below prefers earlier match for fixed
3044 offset, later match for variable offset. */
3045 if (data->last_end == -1) { /* Update the start info. */
3046 data->last_start_min = data->pos_min;
3047 data->last_start_max = is_inf
3048 ? I32_MAX : data->pos_min + data->pos_delta;
3050 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3052 SvUTF8_on(data->last_found);
3054 SV * const sv = data->last_found;
3055 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3056 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3057 if (mg && mg->mg_len >= 0)
3058 mg->mg_len += utf8_length((U8*)STRING(scan),
3059 (U8*)STRING(scan)+STR_LEN(scan));
3061 data->last_end = data->pos_min + l;
3062 data->pos_min += l; /* As in the first entry. */
3063 data->flags &= ~SF_BEFORE_EOL;
3065 if (flags & SCF_DO_STCLASS_AND) {
3066 /* Check whether it is compatible with what we know already! */
3070 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3071 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3072 && (!(data->start_class->flags & ANYOF_FOLD)
3073 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3076 ANYOF_CLASS_ZERO(data->start_class);
3077 ANYOF_BITMAP_ZERO(data->start_class);
3079 ANYOF_BITMAP_SET(data->start_class, uc);
3080 data->start_class->flags &= ~ANYOF_EOS;
3082 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3084 else if (flags & SCF_DO_STCLASS_OR) {
3085 /* false positive possible if the class is case-folded */
3087 ANYOF_BITMAP_SET(data->start_class, uc);
3089 data->start_class->flags |= ANYOF_UNICODE_ALL;
3090 data->start_class->flags &= ~ANYOF_EOS;
3091 cl_and(data->start_class, and_withp);
3093 flags &= ~SCF_DO_STCLASS;
3095 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3096 I32 l = STR_LEN(scan);
3097 UV uc = *((U8*)STRING(scan));
3099 /* Search for fixed substrings supports EXACT only. */
3100 if (flags & SCF_DO_SUBSTR) {
3102 SCAN_COMMIT(pRExC_state, data, minlenp);
3105 const U8 * const s = (U8 *)STRING(scan);
3106 l = utf8_length(s, s + l);
3107 uc = utf8_to_uvchr(s, NULL);
3110 if (flags & SCF_DO_SUBSTR)
3112 if (flags & SCF_DO_STCLASS_AND) {
3113 /* Check whether it is compatible with what we know already! */
3117 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3118 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3119 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3121 ANYOF_CLASS_ZERO(data->start_class);
3122 ANYOF_BITMAP_ZERO(data->start_class);
3124 ANYOF_BITMAP_SET(data->start_class, uc);
3125 data->start_class->flags &= ~ANYOF_EOS;
3126 data->start_class->flags |= ANYOF_FOLD;
3127 if (OP(scan) == EXACTFL)
3128 data->start_class->flags |= ANYOF_LOCALE;
3131 else if (flags & SCF_DO_STCLASS_OR) {
3132 if (data->start_class->flags & ANYOF_FOLD) {
3133 /* false positive possible if the class is case-folded.
3134 Assume that the locale settings are the same... */
3136 ANYOF_BITMAP_SET(data->start_class, uc);
3137 data->start_class->flags &= ~ANYOF_EOS;
3139 cl_and(data->start_class, and_withp);
3141 flags &= ~SCF_DO_STCLASS;
3143 else if (REGNODE_VARIES(OP(scan))) {
3144 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3145 I32 f = flags, pos_before = 0;
3146 regnode * const oscan = scan;
3147 struct regnode_charclass_class this_class;
3148 struct regnode_charclass_class *oclass = NULL;
3149 I32 next_is_eval = 0;
3151 switch (PL_regkind[OP(scan)]) {
3152 case WHILEM: /* End of (?:...)* . */
3153 scan = NEXTOPER(scan);
3156 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3157 next = NEXTOPER(scan);
3158 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3160 maxcount = REG_INFTY;
3161 next = regnext(scan);
3162 scan = NEXTOPER(scan);
3166 if (flags & SCF_DO_SUBSTR)
3171 if (flags & SCF_DO_STCLASS) {
3173 maxcount = REG_INFTY;
3174 next = regnext(scan);
3175 scan = NEXTOPER(scan);
3178 is_inf = is_inf_internal = 1;
3179 scan = regnext(scan);
3180 if (flags & SCF_DO_SUBSTR) {
3181 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3182 data->longest = &(data->longest_float);
3184 goto optimize_curly_tail;
3186 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3187 && (scan->flags == stopparen))
3192 mincount = ARG1(scan);
3193 maxcount = ARG2(scan);
3195 next = regnext(scan);
3196 if (OP(scan) == CURLYX) {
3197 I32 lp = (data ? *(data->last_closep) : 0);
3198 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3200 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3201 next_is_eval = (OP(scan) == EVAL);
3203 if (flags & SCF_DO_SUBSTR) {
3204 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3205 pos_before = data->pos_min;
3209 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3211 data->flags |= SF_IS_INF;
3213 if (flags & SCF_DO_STCLASS) {
3214 cl_init(pRExC_state, &this_class);
3215 oclass = data->start_class;
3216 data->start_class = &this_class;
3217 f |= SCF_DO_STCLASS_AND;
3218 f &= ~SCF_DO_STCLASS_OR;
3220 /* These are the cases when once a subexpression
3221 fails at a particular position, it cannot succeed
3222 even after backtracking at the enclosing scope.
3224 XXXX what if minimal match and we are at the
3225 initial run of {n,m}? */
3226 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3227 f &= ~SCF_WHILEM_VISITED_POS;
3229 /* This will finish on WHILEM, setting scan, or on NULL: */
3230 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3231 last, data, stopparen, recursed, NULL,
3233 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3235 if (flags & SCF_DO_STCLASS)
3236 data->start_class = oclass;
3237 if (mincount == 0 || minnext == 0) {
3238 if (flags & SCF_DO_STCLASS_OR) {
3239 cl_or(pRExC_state, data->start_class, &this_class);
3241 else if (flags & SCF_DO_STCLASS_AND) {
3242 /* Switch to OR mode: cache the old value of
3243 * data->start_class */
3245 StructCopy(data->start_class, and_withp,
3246 struct regnode_charclass_class);
3247 flags &= ~SCF_DO_STCLASS_AND;
3248 StructCopy(&this_class, data->start_class,
3249 struct regnode_charclass_class);
3250 flags |= SCF_DO_STCLASS_OR;
3251 data->start_class->flags |= ANYOF_EOS;
3253 } else { /* Non-zero len */
3254 if (flags & SCF_DO_STCLASS_OR) {
3255 cl_or(pRExC_state, data->start_class, &this_class);
3256 cl_and(data->start_class, and_withp);
3258 else if (flags & SCF_DO_STCLASS_AND)
3259 cl_and(data->start_class, &this_class);
3260 flags &= ~SCF_DO_STCLASS;
3262 if (!scan) /* It was not CURLYX, but CURLY. */
3264 if ( /* ? quantifier ok, except for (?{ ... }) */
3265 (next_is_eval || !(mincount == 0 && maxcount == 1))
3266 && (minnext == 0) && (deltanext == 0)
3267 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3268 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3270 ckWARNreg(RExC_parse,
3271 "Quantifier unexpected on zero-length expression");
3274 min += minnext * mincount;
3275 is_inf_internal |= ((maxcount == REG_INFTY
3276 && (minnext + deltanext) > 0)
3277 || deltanext == I32_MAX);
3278 is_inf |= is_inf_internal;
3279 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3281 /* Try powerful optimization CURLYX => CURLYN. */
3282 if ( OP(oscan) == CURLYX && data
3283 && data->flags & SF_IN_PAR
3284 && !(data->flags & SF_HAS_EVAL)
3285 && !deltanext && minnext == 1 ) {
3286 /* Try to optimize to CURLYN. */
3287 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3288 regnode * const nxt1 = nxt;
3295 if (!REGNODE_SIMPLE(OP(nxt))
3296 && !(PL_regkind[OP(nxt)] == EXACT
3297 && STR_LEN(nxt) == 1))
3303 if (OP(nxt) != CLOSE)
3305 if (RExC_open_parens) {
3306 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3307 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3309 /* Now we know that nxt2 is the only contents: */
3310 oscan->flags = (U8)ARG(nxt);
3312 OP(nxt1) = NOTHING; /* was OPEN. */
3315 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3316 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3317 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
3318 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3319 OP(nxt + 1) = OPTIMIZED; /* was count. */
3320 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
3325 /* Try optimization CURLYX => CURLYM. */
3326 if ( OP(oscan) == CURLYX && data
3327 && !(data->flags & SF_HAS_PAR)
3328 && !(data->flags & SF_HAS_EVAL)
3329 && !deltanext /* atom is fixed width */
3330 && minnext != 0 /* CURLYM can't handle zero width */
3332 /* XXXX How to optimize if data == 0? */
3333 /* Optimize to a simpler form. */
3334 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3338 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3339 && (OP(nxt2) != WHILEM))
3341 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3342 /* Need to optimize away parenths. */
3343 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
3344 /* Set the parenth number. */
3345 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3347 oscan->flags = (U8)ARG(nxt);
3348 if (RExC_open_parens) {
3349 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3350 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3352 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3353 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3356 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3357 OP(nxt + 1) = OPTIMIZED; /* was count. */
3358 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3359 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3362 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3363 regnode *nnxt = regnext(nxt1);
3365 if (reg_off_by_arg[OP(nxt1)])
3366 ARG_SET(nxt1, nxt2 - nxt1);
3367 else if (nxt2 - nxt1 < U16_MAX)
3368 NEXT_OFF(nxt1) = nxt2 - nxt1;
3370 OP(nxt) = NOTHING; /* Cannot beautify */
3375 /* Optimize again: */
3376 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3377 NULL, stopparen, recursed, NULL, 0,depth+1);
3382 else if ((OP(oscan) == CURLYX)
3383 && (flags & SCF_WHILEM_VISITED_POS)
3384 /* See the comment on a similar expression above.
3385 However, this time it's not a subexpression
3386 we care about, but the expression itself. */
3387 && (maxcount == REG_INFTY)
3388 && data && ++data->whilem_c < 16) {
3389 /* This stays as CURLYX, we can put the count/of pair. */
3390 /* Find WHILEM (as in regexec.c) */
3391 regnode *nxt = oscan + NEXT_OFF(oscan);
3393 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3395 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3396 | (RExC_whilem_seen << 4)); /* On WHILEM */
3398 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3400 if (flags & SCF_DO_SUBSTR) {
3401 SV *last_str = NULL;
3402 int counted = mincount != 0;
3404 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3405 #if defined(SPARC64_GCC_WORKAROUND)
3408 const char *s = NULL;
3411 if (pos_before >= data->last_start_min)
3414 b = data->last_start_min;
3417 s = SvPV_const(data->last_found, l);
3418 old = b - data->last_start_min;
3421 I32 b = pos_before >= data->last_start_min
3422 ? pos_before : data->last_start_min;
3424 const char * const s = SvPV_const(data->last_found, l);
3425 I32 old = b - data->last_start_min;
3429 old = utf8_hop((U8*)s, old) - (U8*)s;
3431 /* Get the added string: */
3432 last_str = newSVpvn_utf8(s + old, l, UTF);
3433 if (deltanext == 0 && pos_before == b) {
3434 /* What was added is a constant string */
3436 SvGROW(last_str, (mincount * l) + 1);
3437 repeatcpy(SvPVX(last_str) + l,
3438 SvPVX_const(last_str), l, mincount - 1);
3439 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3440 /* Add additional parts. */
3441 SvCUR_set(data->last_found,
3442 SvCUR(data->last_found) - l);
3443 sv_catsv(data->last_found, last_str);
3445 SV * sv = data->last_found;
3447 SvUTF8(sv) && SvMAGICAL(sv) ?
3448 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3449 if (mg && mg->mg_len >= 0)
3450 mg->mg_len += CHR_SVLEN(last_str) - l;
3452 data->last_end += l * (mincount - 1);
3455 /* start offset must point into the last copy */
3456 data->last_start_min += minnext * (mincount - 1);
3457 data->last_start_max += is_inf ? I32_MAX
3458 : (maxcount - 1) * (minnext + data->pos_delta);
3461 /* It is counted once already... */
3462 data->pos_min += minnext * (mincount - counted);
3463 data->pos_delta += - counted * deltanext +
3464 (minnext + deltanext) * maxcount - minnext * mincount;
3465 if (mincount != maxcount) {
3466 /* Cannot extend fixed substrings found inside
3468 SCAN_COMMIT(pRExC_state,data,minlenp);
3469 if (mincount && last_str) {
3470 SV * const sv = data->last_found;
3471 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3472 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3476 sv_setsv(sv, last_str);
3477 data->last_end = data->pos_min;
3478 data->last_start_min =
3479 data->pos_min - CHR_SVLEN(last_str);
3480 data->last_start_max = is_inf
3482 : data->pos_min + data->pos_delta
3483 - CHR_SVLEN(last_str);
3485 data->longest = &(data->longest_float);
3487 SvREFCNT_dec(last_str);
3489 if (data && (fl & SF_HAS_EVAL))
3490 data->flags |= SF_HAS_EVAL;
3491 optimize_curly_tail:
3492 if (OP(oscan) != CURLYX) {
3493 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3495 NEXT_OFF(oscan) += NEXT_OFF(next);
3498 default: /* REF and CLUMP only? */
3499 if (flags & SCF_DO_SUBSTR) {
3500 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3501 data->longest = &(data->longest_float);
3503 is_inf = is_inf_internal = 1;
3504 if (flags & SCF_DO_STCLASS_OR)
3505 cl_anything(pRExC_state, data->start_class);
3506 flags &= ~SCF_DO_STCLASS;
3510 else if (OP(scan) == LNBREAK) {
3511 if (flags & SCF_DO_STCLASS) {
3513 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3514 if (flags & SCF_DO_STCLASS_AND) {
3515 for (value = 0; value < 256; value++)
3516 if (!is_VERTWS_cp(value))
3517 ANYOF_BITMAP_CLEAR(data->start_class, value);
3520 for (value = 0; value < 256; value++)
3521 if (is_VERTWS_cp(value))
3522 ANYOF_BITMAP_SET(data->start_class, value);
3524 if (flags & SCF_DO_STCLASS_OR)
3525 cl_and(data->start_class, and_withp);
3526 flags &= ~SCF_DO_STCLASS;
3530 if (flags & SCF_DO_SUBSTR) {
3531 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3533 data->pos_delta += 1;
3534 data->longest = &(data->longest_float);
3537 else if (OP(scan) == FOLDCHAR) {
3538 int d = ARG(scan)==0xDF ? 1 : 2;
3539 flags &= ~SCF_DO_STCLASS;
3542 if (flags & SCF_DO_SUBSTR) {
3543 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3545 data->pos_delta += d;
3546 data->longest = &(data->longest_float);
3549 else if (REGNODE_SIMPLE(OP(scan))) {
3552 if (flags & SCF_DO_SUBSTR) {
3553 SCAN_COMMIT(pRExC_state,data,minlenp);
3557 if (flags & SCF_DO_STCLASS) {
3558 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3560 /* Some of the logic below assumes that switching
3561 locale on will only add false positives. */
3562 switch (PL_regkind[OP(scan)]) {
3566 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3567 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3568 cl_anything(pRExC_state, data->start_class);
3571 if (OP(scan) == SANY)
3573 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3574 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3575 || (data->start_class->flags & ANYOF_CLASS));
3576 cl_anything(pRExC_state, data->start_class);
3578 if (flags & SCF_DO_STCLASS_AND || !value)
3579 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3582 if (flags & SCF_DO_STCLASS_AND)
3583 cl_and(data->start_class,
3584 (struct regnode_charclass_class*)scan);
3586 cl_or(pRExC_state, data->start_class,
3587 (struct regnode_charclass_class*)scan);
3590 if (flags & SCF_DO_STCLASS_AND) {
3591 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3592 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3593 if (FLAGS(scan) & USE_UNI) {
3594 for (value = 0; value < 256; value++) {
3595 if (!isWORDCHAR_L1(value)) {
3596 ANYOF_BITMAP_CLEAR(data->start_class, value);
3600 for (value = 0; value < 256; value++) {
3601 if (!isALNUM(value)) {
3602 ANYOF_BITMAP_CLEAR(data->start_class, value);
3609 if (data->start_class->flags & ANYOF_LOCALE)
3610 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3611 else if (FLAGS(scan) & USE_UNI) {
3612 for (value = 0; value < 256; value++) {
3613 if (isWORDCHAR_L1(value)) {
3614 ANYOF_BITMAP_SET(data->start_class, value);
3618 for (value = 0; value < 256; value++) {
3619 if (isALNUM(value)) {
3620 ANYOF_BITMAP_SET(data->start_class, value);
3627 if (flags & SCF_DO_STCLASS_AND) {
3628 if (data->start_class->flags & ANYOF_LOCALE)
3629 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3632 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3633 data->start_class->flags |= ANYOF_LOCALE;
3637 if (flags & SCF_DO_STCLASS_AND) {
3638 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3639 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3640 if (FLAGS(scan) & USE_UNI) {
3641 for (value = 0; value < 256; value++) {
3642 if (isWORDCHAR_L1(value)) {
3643 ANYOF_BITMAP_CLEAR(data->start_class, value);
3647 for (value = 0; value < 256; value++) {
3648 if (isALNUM(value)) {
3649 ANYOF_BITMAP_CLEAR(data->start_class, value);
3656 if (data->start_class->flags & ANYOF_LOCALE)
3657 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3659 for (value = 0; value < 256; value++)
3660 if (!isALNUM(value))
3661 ANYOF_BITMAP_SET(data->start_class, value);
3666 if (flags & SCF_DO_STCLASS_AND) {
3667 if (data->start_class->flags & ANYOF_LOCALE)
3668 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3671 data->start_class->flags |= ANYOF_LOCALE;
3672 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3676 if (flags & SCF_DO_STCLASS_AND) {
3677 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3678 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3679 if (FLAGS(scan) & USE_UNI) {
3680 for (value = 0; value < 256; value++) {
3681 if (!isSPACE_L1(value)) {
3682 ANYOF_BITMAP_CLEAR(data->start_class, value);
3686 for (value = 0; value < 256; value++) {
3687 if (!isSPACE(value)) {
3688 ANYOF_BITMAP_CLEAR(data->start_class, value);
3695 if (data->start_class->flags & ANYOF_LOCALE) {
3696 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3698 else if (FLAGS(scan) & USE_UNI) {
3699 for (value = 0; value < 256; value++) {
3700 if (isSPACE_L1(value)) {
3701 ANYOF_BITMAP_SET(data->start_class, value);
3705 for (value = 0; value < 256; value++) {
3706 if (isSPACE(value)) {
3707 ANYOF_BITMAP_SET(data->start_class, value);
3714 if (flags & SCF_DO_STCLASS_AND) {
3715 if (data->start_class->flags & ANYOF_LOCALE)
3716 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3719 data->start_class->flags |= ANYOF_LOCALE;
3720 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3724 if (flags & SCF_DO_STCLASS_AND) {
3725 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3726 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3727 if (FLAGS(scan) & USE_UNI) {
3728 for (value = 0; value < 256; value++) {
3729 if (isSPACE_L1(value)) {
3730 ANYOF_BITMAP_CLEAR(data->start_class, value);
3734 for (value = 0; value < 256; value++) {
3735 if (isSPACE(value)) {
3736 ANYOF_BITMAP_CLEAR(data->start_class, value);
3743 if (data->start_class->flags & ANYOF_LOCALE)
3744 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3745 else if (FLAGS(scan) & USE_UNI) {
3746 for (value = 0; value < 256; value++) {
3747 if (!isSPACE_L1(value)) {
3748 ANYOF_BITMAP_SET(data->start_class, value);
3753 for (value = 0; value < 256; value++) {
3754 if (!isSPACE(value)) {
3755 ANYOF_BITMAP_SET(data->start_class, value);
3762 if (flags & SCF_DO_STCLASS_AND) {
3763 if (data->start_class->flags & ANYOF_LOCALE) {
3764 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3765 for (value = 0; value < 256; value++)
3766 if (!isSPACE(value))
3767 ANYOF_BITMAP_CLEAR(data->start_class, value);
3771 data->start_class->flags |= ANYOF_LOCALE;
3772 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3776 if (flags & SCF_DO_STCLASS_AND) {
3777 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3778 for (value = 0; value < 256; value++)
3779 if (!isDIGIT(value))
3780 ANYOF_BITMAP_CLEAR(data->start_class, value);
3783 if (data->start_class->flags & ANYOF_LOCALE)
3784 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3786 for (value = 0; value < 256; value++)
3788 ANYOF_BITMAP_SET(data->start_class, value);
3793 if (flags & SCF_DO_STCLASS_AND) {
3794 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3795 for (value = 0; value < 256; value++)
3797 ANYOF_BITMAP_CLEAR(data->start_class, value);
3800 if (data->start_class->flags & ANYOF_LOCALE)
3801 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3803 for (value = 0; value < 256; value++)
3804 if (!isDIGIT(value))
3805 ANYOF_BITMAP_SET(data->start_class, value);
3809 CASE_SYNST_FNC(VERTWS);
3810 CASE_SYNST_FNC(HORIZWS);
3813 if (flags & SCF_DO_STCLASS_OR)
3814 cl_and(data->start_class, and_withp);
3815 flags &= ~SCF_DO_STCLASS;
3818 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3819 data->flags |= (OP(scan) == MEOL
3823 else if ( PL_regkind[OP(scan)] == BRANCHJ
3824 /* Lookbehind, or need to calculate parens/evals/stclass: */
3825 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3826 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3827 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3828 || OP(scan) == UNLESSM )
3830 /* Negative Lookahead/lookbehind
3831 In this case we can't do fixed string optimisation.
3834 I32 deltanext, minnext, fake = 0;
3836 struct regnode_charclass_class intrnl;
3839 data_fake.flags = 0;
3841 data_fake.whilem_c = data->whilem_c;
3842 data_fake.last_closep = data->last_closep;
3845 data_fake.last_closep = &fake;
3846 data_fake.pos_delta = delta;
3847 if ( flags & SCF_DO_STCLASS && !scan->flags
3848 && OP(scan) == IFMATCH ) { /* Lookahead */
3849 cl_init(pRExC_state, &intrnl);
3850 data_fake.start_class = &intrnl;
3851 f |= SCF_DO_STCLASS_AND;
3853 if (flags & SCF_WHILEM_VISITED_POS)
3854 f |= SCF_WHILEM_VISITED_POS;
3855 next = regnext(scan);
3856 nscan = NEXTOPER(NEXTOPER(scan));
3857 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3858 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3861 FAIL("Variable length lookbehind not implemented");
3863 else if (minnext > (I32)U8_MAX) {
3864 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3866 scan->flags = (U8)minnext;
3869 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3871 if (data_fake.flags & SF_HAS_EVAL)
3872 data->flags |= SF_HAS_EVAL;
3873 data->whilem_c = data_fake.whilem_c;
3875 if (f & SCF_DO_STCLASS_AND) {
3876 if (flags & SCF_DO_STCLASS_OR) {
3877 /* OR before, AND after: ideally we would recurse with
3878 * data_fake to get the AND applied by study of the
3879 * remainder of the pattern, and then derecurse;
3880 * *** HACK *** for now just treat as "no information".
3881 * See [perl #56690].
3883 cl_init(pRExC_state, data->start_class);
3885 /* AND before and after: combine and continue */
3886 const int was = (data->start_class->flags & ANYOF_EOS);
3888 cl_and(data->start_class, &intrnl);
3890 data->start_class->flags |= ANYOF_EOS;
3894 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3896 /* Positive Lookahead/lookbehind
3897 In this case we can do fixed string optimisation,
3898 but we must be careful about it. Note in the case of
3899 lookbehind the positions will be offset by the minimum
3900 length of the pattern, something we won't know about
3901 until after the recurse.
3903 I32 deltanext, fake = 0;
3905 struct regnode_charclass_class intrnl;
3907 /* We use SAVEFREEPV so that when the full compile
3908 is finished perl will clean up the allocated
3909 minlens when it's all done. This way we don't
3910 have to worry about freeing them when we know
3911 they wont be used, which would be a pain.
3914 Newx( minnextp, 1, I32 );
3915 SAVEFREEPV(minnextp);
3918 StructCopy(data, &data_fake, scan_data_t);
3919 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3922 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3923 data_fake.last_found=newSVsv(data->last_found);
3927 data_fake.last_closep = &fake;
3928 data_fake.flags = 0;
3929 data_fake.pos_delta = delta;
3931 data_fake.flags |= SF_IS_INF;
3932 if ( flags & SCF_DO_STCLASS && !scan->flags
3933 && OP(scan) == IFMATCH ) { /* Lookahead */
3934 cl_init(pRExC_state, &intrnl);
3935 data_fake.start_class = &intrnl;
3936 f |= SCF_DO_STCLASS_AND;
3938 if (flags & SCF_WHILEM_VISITED_POS)
3939 f |= SCF_WHILEM_VISITED_POS;
3940 next = regnext(scan);
3941 nscan = NEXTOPER(NEXTOPER(scan));
3943 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3944 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3947 FAIL("Variable length lookbehind not implemented");
3949 else if (*minnextp > (I32)U8_MAX) {
3950 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3952 scan->flags = (U8)*minnextp;
3957 if (f & SCF_DO_STCLASS_AND) {
3958 const int was = (data->start_class->flags & ANYOF_EOS);
3960 cl_and(data->start_class, &intrnl);
3962 data->start_class->flags |= ANYOF_EOS;
3965 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3967 if (data_fake.flags & SF_HAS_EVAL)
3968 data->flags |= SF_HAS_EVAL;
3969 data->whilem_c = data_fake.whilem_c;
3970 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3971 if (RExC_rx->minlen<*minnextp)
3972 RExC_rx->minlen=*minnextp;
3973 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3974 SvREFCNT_dec(data_fake.last_found);
3976 if ( data_fake.minlen_fixed != minlenp )
3978 data->offset_fixed= data_fake.offset_fixed;
3979 data->minlen_fixed= data_fake.minlen_fixed;
3980 data->lookbehind_fixed+= scan->flags;
3982 if ( data_fake.minlen_float != minlenp )
3984 data->minlen_float= data_fake.minlen_float;
3985 data->offset_float_min=data_fake.offset_float_min;
3986 data->offset_float_max=data_fake.offset_float_max;
3987 data->lookbehind_float+= scan->flags;
3996 else if (OP(scan) == OPEN) {
3997 if (stopparen != (I32)ARG(scan))
4000 else if (OP(scan) == CLOSE) {
4001 if (stopparen == (I32)ARG(scan)) {
4004 if ((I32)ARG(scan) == is_par) {
4005 next = regnext(scan);
4007 if ( next && (OP(next) != WHILEM) && next < last)
4008 is_par = 0; /* Disable optimization */
4011 *(data->last_closep) = ARG(scan);
4013 else if (OP(scan) == EVAL) {
4015 data->flags |= SF_HAS_EVAL;
4017 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
4018 if (flags & SCF_DO_SUBSTR) {
4019 SCAN_COMMIT(pRExC_state,data,minlenp);
4020 flags &= ~SCF_DO_SUBSTR;
4022 if (data && OP(scan)==ACCEPT) {
4023 data->flags |= SCF_SEEN_ACCEPT;
4028 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
4030 if (flags & SCF_DO_SUBSTR) {
4031 SCAN_COMMIT(pRExC_state,data,minlenp);
4032 data->longest = &(data->longest_float);
4034 is_inf = is_inf_internal = 1;
4035 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
4036 cl_anything(pRExC_state, data->start_class);
4037 flags &= ~SCF_DO_STCLASS;
4039 else if (OP(scan) == GPOS) {
4040 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
4041 !(delta || is_inf || (data && data->pos_delta)))
4043 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
4044 RExC_rx->extflags |= RXf_ANCH_GPOS;
4045 if (RExC_rx->gofs < (U32)min)
4046 RExC_rx->gofs = min;
4048 RExC_rx->extflags |= RXf_GPOS_FLOAT;
4052 #ifdef TRIE_STUDY_OPT
4053 #ifdef FULL_TRIE_STUDY
4054 else if (PL_regkind[OP(scan)] == TRIE) {
4055 /* NOTE - There is similar code to this block above for handling
4056 BRANCH nodes on the initial study. If you change stuff here
4058 regnode *trie_node= scan;
4059 regnode *tail= regnext(scan);
4060 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4061 I32 max1 = 0, min1 = I32_MAX;
4062 struct regnode_charclass_class accum;
4064 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
4065 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
4066 if (flags & SCF_DO_STCLASS)
4067 cl_init_zero(pRExC_state, &accum);
4073 const regnode *nextbranch= NULL;
4076 for ( word=1 ; word <= trie->wordcount ; word++)
4078 I32 deltanext=0, minnext=0, f = 0, fake;
4079 struct regnode_charclass_class this_class;
4081 data_fake.flags = 0;
4083 data_fake.whilem_c = data->whilem_c;
4084 data_fake.last_closep = data->last_closep;
4087 data_fake.last_closep = &fake;
4088 data_fake.pos_delta = delta;
4089 if (flags & SCF_DO_STCLASS) {
4090 cl_init(pRExC_state, &this_class);
4091 data_fake.start_class = &this_class;
4092 f = SCF_DO_STCLASS_AND;
4094 if (flags & SCF_WHILEM_VISITED_POS)
4095 f |= SCF_WHILEM_VISITED_POS;
4097 if (trie->jump[word]) {
4099 nextbranch = trie_node + trie->jump[0];
4100 scan= trie_node + trie->jump[word];
4101 /* We go from the jump point to the branch that follows
4102 it. Note this means we need the vestigal unused branches
4103 even though they arent otherwise used.
4105 minnext = study_chunk(pRExC_state, &scan, minlenp,
4106 &deltanext, (regnode *)nextbranch, &data_fake,
4107 stopparen, recursed, NULL, f,depth+1);
4109 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4110 nextbranch= regnext((regnode*)nextbranch);
4112 if (min1 > (I32)(minnext + trie->minlen))
4113 min1 = minnext + trie->minlen;
4114 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4115 max1 = minnext + deltanext + trie->maxlen;
4116 if (deltanext == I32_MAX)
4117 is_inf = is_inf_internal = 1;
4119 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4121 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4122 if ( stopmin > min + min1)
4123 stopmin = min + min1;
4124 flags &= ~SCF_DO_SUBSTR;
4126 data->flags |= SCF_SEEN_ACCEPT;
4129 if (data_fake.flags & SF_HAS_EVAL)
4130 data->flags |= SF_HAS_EVAL;
4131 data->whilem_c = data_fake.whilem_c;
4133 if (flags & SCF_DO_STCLASS)
4134 cl_or(pRExC_state, &accum, &this_class);
4137 if (flags & SCF_DO_SUBSTR) {
4138 data->pos_min += min1;
4139 data->pos_delta += max1 - min1;
4140 if (max1 != min1 || is_inf)
4141 data->longest = &(data->longest_float);
4144 delta += max1 - min1;
4145 if (flags & SCF_DO_STCLASS_OR) {
4146 cl_or(pRExC_state, data->start_class, &accum);
4148 cl_and(data->start_class, and_withp);
4149 flags &= ~SCF_DO_STCLASS;
4152 else if (flags & SCF_DO_STCLASS_AND) {
4154 cl_and(data->start_class, &accum);
4155 flags &= ~SCF_DO_STCLASS;
4158 /* Switch to OR mode: cache the old value of
4159 * data->start_class */
4161 StructCopy(data->start_class, and_withp,
4162 struct regnode_charclass_class);
4163 flags &= ~SCF_DO_STCLASS_AND;
4164 StructCopy(&accum, data->start_class,
4165 struct regnode_charclass_class);
4166 flags |= SCF_DO_STCLASS_OR;
4167 data->start_class->flags |= ANYOF_EOS;
4174 else if (PL_regkind[OP(scan)] == TRIE) {
4175 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4178 min += trie->minlen;
4179 delta += (trie->maxlen - trie->minlen);
4180 flags &= ~SCF_DO_STCLASS; /* xxx */
4181 if (flags & SCF_DO_SUBSTR) {
4182 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4183 data->pos_min += trie->minlen;
4184 data->pos_delta += (trie->maxlen - trie->minlen);
4185 if (trie->maxlen != trie->minlen)
4186 data->longest = &(data->longest_float);
4188 if (trie->jump) /* no more substrings -- for now /grr*/
4189 flags &= ~SCF_DO_SUBSTR;
4191 #endif /* old or new */
4192 #endif /* TRIE_STUDY_OPT */
4194 /* Else: zero-length, ignore. */
4195 scan = regnext(scan);
4200 stopparen = frame->stop;
4201 frame = frame->prev;
4202 goto fake_study_recurse;
4207 DEBUG_STUDYDATA("pre-fin:",data,depth);
4210 *deltap = is_inf_internal ? I32_MAX : delta;
4211 if (flags & SCF_DO_SUBSTR && is_inf)
4212 data->pos_delta = I32_MAX - data->pos_min;
4213 if (is_par > (I32)U8_MAX)
4215 if (is_par && pars==1 && data) {
4216 data->flags |= SF_IN_PAR;
4217 data->flags &= ~SF_HAS_PAR;
4219 else if (pars && data) {
4220 data->flags |= SF_HAS_PAR;
4221 data->flags &= ~SF_IN_PAR;
4223 if (flags & SCF_DO_STCLASS_OR)
4224 cl_and(data->start_class, and_withp);
4225 if (flags & SCF_TRIE_RESTUDY)
4226 data->flags |= SCF_TRIE_RESTUDY;
4228 DEBUG_STUDYDATA("post-fin:",data,depth);
4230 return min < stopmin ? min : stopmin;
4234 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4236 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4238 PERL_ARGS_ASSERT_ADD_DATA;
4240 Renewc(RExC_rxi->data,
4241 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4242 char, struct reg_data);
4244 Renew(RExC_rxi->data->what, count + n, U8);
4246 Newx(RExC_rxi->data->what, n, U8);
4247 RExC_rxi->data->count = count + n;
4248 Copy(s, RExC_rxi->data->what + count, n, U8);
4252 /*XXX: todo make this not included in a non debugging perl */
4253 #ifndef PERL_IN_XSUB_RE
4255 Perl_reginitcolors(pTHX)
4258 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4260 char *t = savepv(s);
4264 t = strchr(t, '\t');
4270 PL_colors[i] = t = (char *)"";
4275 PL_colors[i++] = (char *)"";
4282 #ifdef TRIE_STUDY_OPT
4283 #define CHECK_RESTUDY_GOTO \
4285 (data.flags & SCF_TRIE_RESTUDY) \
4289 #define CHECK_RESTUDY_GOTO
4293 - pregcomp - compile a regular expression into internal code
4295 * We can't allocate space until we know how big the compiled form will be,
4296 * but we can't compile it (and thus know how big it is) until we've got a
4297 * place to put the code. So we cheat: we compile it twice, once with code
4298 * generation turned off and size counting turned on, and once "for real".
4299 * This also means that we don't allocate space until we are sure that the
4300 * thing really will compile successfully, and we never have to move the
4301 * code and thus invalidate pointers into it. (Note that it has to be in
4302 * one piece because free() must be able to free it all.) [NB: not true in perl]
4304 * Beware that the optimization-preparation code in here knows about some
4305 * of the structure of the compiled regexp. [I'll say.]
4310 #ifndef PERL_IN_XSUB_RE
4311 #define RE_ENGINE_PTR &PL_core_reg_engine
4313 extern const struct regexp_engine my_reg_engine;
4314 #define RE_ENGINE_PTR &my_reg_engine
4317 #ifndef PERL_IN_XSUB_RE
4319 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4322 HV * const table = GvHV(PL_hintgv);
4324 PERL_ARGS_ASSERT_PREGCOMP;
4326 /* Dispatch a request to compile a regexp to correct
4329 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4330 GET_RE_DEBUG_FLAGS_DECL;
4331 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4332 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4334 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4337 return CALLREGCOMP_ENG(eng, pattern, flags);
4340 return Perl_re_compile(aTHX_ pattern, flags);
4345 Perl_re_compile(pTHX_ SV * const pattern, U32 pm_flags)
4350 register regexp_internal *ri;
4358 /* these are all flags - maybe they should be turned
4359 * into a single int with different bit masks */
4360 I32 sawlookahead = 0;
4367 RExC_state_t RExC_state;
4368 RExC_state_t * const pRExC_state = &RExC_state;
4369 #ifdef TRIE_STUDY_OPT
4371 RExC_state_t copyRExC_state;
4373 GET_RE_DEBUG_FLAGS_DECL;
4375 PERL_ARGS_ASSERT_RE_COMPILE;
4377 DEBUG_r(if (!PL_colorset) reginitcolors());
4379 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4382 /* Longjmp back to here if have to switch in midstream to utf8 */
4383 if (! RExC_orig_utf8) {
4384 JMPENV_PUSH(jump_ret);
4387 if (jump_ret == 0) { /* First time through */
4388 exp = SvPV(pattern, plen);
4392 SV *dsv= sv_newmortal();
4393 RE_PV_QUOTED_DECL(s, RExC_utf8,
4394 dsv, exp, plen, 60);
4395 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4396 PL_colors[4],PL_colors[5],s);
4399 else { /* longjumped back */
4402 /* If the cause for the longjmp was other than changing to utf8, pop
4403 * our own setjmp, and longjmp to the correct handler */
4404 if (jump_ret != UTF8_LONGJMP) {
4406 JMPENV_JUMP(jump_ret);
4411 /* It's possible to write a regexp in ascii that represents Unicode
4412 codepoints outside of the byte range, such as via \x{100}. If we
4413 detect such a sequence we have to convert the entire pattern to utf8
4414 and then recompile, as our sizing calculation will have been based
4415 on 1 byte == 1 character, but we will need to use utf8 to encode
4416 at least some part of the pattern, and therefore must convert the whole
4419 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4420 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4421 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4423 RExC_orig_utf8 = RExC_utf8 = 1;
4427 #ifdef TRIE_STUDY_OPT
4432 RExC_flags = pm_flags;
4436 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4437 RExC_seen_evals = 0;
4440 /* First pass: determine size, legality. */
4448 RExC_emit = &PL_regdummy;
4449 RExC_whilem_seen = 0;
4450 RExC_open_parens = NULL;
4451 RExC_close_parens = NULL;
4453 RExC_paren_names = NULL;
4455 RExC_paren_name_list = NULL;
4457 RExC_recurse = NULL;
4458 RExC_recurse_count = 0;
4460 #if 0 /* REGC() is (currently) a NOP at the first pass.
4461 * Clever compilers notice this and complain. --jhi */
4462 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4464 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4465 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4466 RExC_precomp = NULL;
4470 /* Here, finished first pass. Get rid of our setjmp, which we added for
4471 * efficiency only if the passed-in string wasn't in utf8, as shown by
4472 * RExC_orig_utf8. But if the first pass was redone, that variable will be
4473 * 1 here even though the original string wasn't utf8, but in this case
4474 * there will have been a long jump */
4475 if (jump_ret == UTF8_LONGJMP || ! RExC_orig_utf8) {
4479 PerlIO_printf(Perl_debug_log,
4480 "Required size %"IVdf" nodes\n"
4481 "Starting second pass (creation)\n",
4484 RExC_lastparse=NULL;
4486 /* Small enough for pointer-storage convention?
4487 If extralen==0, this means that we will not need long jumps. */
4488 if (RExC_size >= 0x10000L && RExC_extralen)
4489 RExC_size += RExC_extralen;
4492 if (RExC_whilem_seen > 15)
4493 RExC_whilem_seen = 15;
4495 /* Allocate space and zero-initialize. Note, the two step process
4496 of zeroing when in debug mode, thus anything assigned has to
4497 happen after that */
4498 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4499 r = (struct regexp*)SvANY(rx);
4500 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4501 char, regexp_internal);
4502 if ( r == NULL || ri == NULL )
4503 FAIL("Regexp out of space");
4505 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4506 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4508 /* bulk initialize base fields with 0. */
4509 Zero(ri, sizeof(regexp_internal), char);
4512 /* non-zero initialization begins here */
4514 r->engine= RE_ENGINE_PTR;
4515 r->extflags = pm_flags;
4517 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4518 bool has_charset = cBOOL(r->extflags & (RXf_PMf_LOCALE|RXf_PMf_UNICODE));
4520 /* The caret is output if there are any defaults: if not all the STD
4521 * flags are set, or if no character set specifier is needed */
4523 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4525 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4526 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4527 >> RXf_PMf_STD_PMMOD_SHIFT);
4528 const char *fptr = STD_PAT_MODS; /*"msix"*/
4530 /* Allocate for the worst case, which is all the std flags are turned
4531 * on. If more precision is desired, we could do a population count of
4532 * the flags set. This could be done with a small lookup table, or by
4533 * shifting, masking and adding, or even, when available, assembly
4534 * language for a machine-language population count.
4535 * We never output a minus, as all those are defaults, so are
4536 * covered by the caret */
4537 const STRLEN wraplen = plen + has_p + has_runon
4538 + has_default /* If needs a caret */
4539 + has_charset /* If needs a character set specifier */
4540 + (sizeof(STD_PAT_MODS) - 1)
4541 + (sizeof("(?:)") - 1);
4543 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4545 SvFLAGS(rx) |= SvUTF8(pattern);
4548 /* If a default, cover it using the caret */
4550 *p++= DEFAULT_PAT_MOD;
4553 if (r->extflags & RXf_PMf_LOCALE) {
4554 *p++ = LOCALE_PAT_MOD;
4556 *p++ = UNICODE_PAT_MOD;
4560 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4563 while((ch = *fptr++)) {
4571 Copy(RExC_precomp, p, plen, char);
4572 assert ((RX_WRAPPED(rx) - p) < 16);
4573 r->pre_prefix = p - RX_WRAPPED(rx);
4579 SvCUR_set(rx, p - SvPVX_const(rx));
4583 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4585 if (RExC_seen & REG_SEEN_RECURSE) {
4586 Newxz(RExC_open_parens, RExC_npar,regnode *);
4587 SAVEFREEPV(RExC_open_parens);
4588 Newxz(RExC_close_parens,RExC_npar,regnode *);
4589 SAVEFREEPV(RExC_close_parens);
4592 /* Useful during FAIL. */
4593 #ifdef RE_TRACK_PATTERN_OFFSETS
4594 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4595 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4596 "%s %"UVuf" bytes for offset annotations.\n",
4597 ri->u.offsets ? "Got" : "Couldn't get",
4598 (UV)((2*RExC_size+1) * sizeof(U32))));
4600 SetProgLen(ri,RExC_size);
4604 REH_CALL_COMP_BEGIN_HOOK(pRExC_state->rx);
4606 /* Second pass: emit code. */
4607 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4612 RExC_emit_start = ri->program;
4613 RExC_emit = ri->program;
4614 RExC_emit_bound = ri->program + RExC_size + 1;
4616 /* Store the count of eval-groups for security checks: */
4617 RExC_rx->seen_evals = RExC_seen_evals;
4618 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4619 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4623 /* XXXX To minimize changes to RE engine we always allocate
4624 3-units-long substrs field. */
4625 Newx(r->substrs, 1, struct reg_substr_data);
4626 if (RExC_recurse_count) {
4627 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4628 SAVEFREEPV(RExC_recurse);
4632 r->minlen = minlen = sawlookahead = sawplus = sawopen = 0;
4633 Zero(r->substrs, 1, struct reg_substr_data);
4635 #ifdef TRIE_STUDY_OPT
4637 StructCopy(&zero_scan_data, &data, scan_data_t);
4638 copyRExC_state = RExC_state;
4641 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4643 RExC_state = copyRExC_state;
4644 if (seen & REG_TOP_LEVEL_BRANCHES)
4645 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4647 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4648 if (data.last_found) {
4649 SvREFCNT_dec(data.longest_fixed);
4650 SvREFCNT_dec(data.longest_float);
4651 SvREFCNT_dec(data.last_found);
4653 StructCopy(&zero_scan_data, &data, scan_data_t);
4656 StructCopy(&zero_scan_data, &data, scan_data_t);
4659 /* Dig out information for optimizations. */
4660 r->extflags = RExC_flags; /* was pm_op */
4661 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4664 SvUTF8_on(rx); /* Unicode in it? */
4665 ri->regstclass = NULL;
4666 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4667 r->intflags |= PREGf_NAUGHTY;
4668 scan = ri->program + 1; /* First BRANCH. */
4670 /* testing for BRANCH here tells us whether there is "must appear"
4671 data in the pattern. If there is then we can use it for optimisations */
4672 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4674 STRLEN longest_float_length, longest_fixed_length;
4675 struct regnode_charclass_class ch_class; /* pointed to by data */
4677 I32 last_close = 0; /* pointed to by data */
4678 regnode *first= scan;
4679 regnode *first_next= regnext(first);
4681 * Skip introductions and multiplicators >= 1
4682 * so that we can extract the 'meat' of the pattern that must
4683 * match in the large if() sequence following.
4684 * NOTE that EXACT is NOT covered here, as it is normally
4685 * picked up by the optimiser separately.
4687 * This is unfortunate as the optimiser isnt handling lookahead
4688 * properly currently.
4691 while ((OP(first) == OPEN && (sawopen = 1)) ||
4692 /* An OR of *one* alternative - should not happen now. */
4693 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4694 /* for now we can't handle lookbehind IFMATCH*/
4695 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
4696 (OP(first) == PLUS) ||
4697 (OP(first) == MINMOD) ||
4698 /* An {n,m} with n>0 */
4699 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4700 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4703 * the only op that could be a regnode is PLUS, all the rest
4704 * will be regnode_1 or regnode_2.
4707 if (OP(first) == PLUS)
4710 first += regarglen[OP(first)];
4712 first = NEXTOPER(first);
4713 first_next= regnext(first);
4716 /* Starting-point info. */
4718 DEBUG_PEEP("first:",first,0);
4719 /* Ignore EXACT as we deal with it later. */
4720 if (PL_regkind[OP(first)] == EXACT) {
4721 if (OP(first) == EXACT)
4722 NOOP; /* Empty, get anchored substr later. */
4723 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4724 ri->regstclass = first;
4727 else if (PL_regkind[OP(first)] == TRIE &&
4728 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4731 /* this can happen only on restudy */
4732 if ( OP(first) == TRIE ) {
4733 struct regnode_1 *trieop = (struct regnode_1 *)
4734 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4735 StructCopy(first,trieop,struct regnode_1);
4736 trie_op=(regnode *)trieop;
4738 struct regnode_charclass *trieop = (struct regnode_charclass *)
4739 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4740 StructCopy(first,trieop,struct regnode_charclass);
4741 trie_op=(regnode *)trieop;
4744 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4745 ri->regstclass = trie_op;
4748 else if (REGNODE_SIMPLE(OP(first)))
4749 ri->regstclass = first;
4750 else if (PL_regkind[OP(first)] == BOUND ||
4751 PL_regkind[OP(first)] == NBOUND)
4752 ri->regstclass = first;
4753 else if (PL_regkind[OP(first)] == BOL) {
4754 r->extflags |= (OP(first) == MBOL
4756 : (OP(first) == SBOL
4759 first = NEXTOPER(first);
4762 else if (OP(first) == GPOS) {
4763 r->extflags |= RXf_ANCH_GPOS;
4764 first = NEXTOPER(first);
4767 else if ((!sawopen || !RExC_sawback) &&
4768 (OP(first) == STAR &&
4769 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4770 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4772 /* turn .* into ^.* with an implied $*=1 */
4774 (OP(NEXTOPER(first)) == REG_ANY)
4777 r->extflags |= type;
4778 r->intflags |= PREGf_IMPLICIT;
4779 first = NEXTOPER(first);
4782 if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback)
4783 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4784 /* x+ must match at the 1st pos of run of x's */
4785 r->intflags |= PREGf_SKIP;
4787 /* Scan is after the zeroth branch, first is atomic matcher. */
4788 #ifdef TRIE_STUDY_OPT
4791 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4792 (IV)(first - scan + 1))
4796 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4797 (IV)(first - scan + 1))
4803 * If there's something expensive in the r.e., find the
4804 * longest literal string that must appear and make it the
4805 * regmust. Resolve ties in favor of later strings, since
4806 * the regstart check works with the beginning of the r.e.
4807 * and avoiding duplication strengthens checking. Not a
4808 * strong reason, but sufficient in the absence of others.
4809 * [Now we resolve ties in favor of the earlier string if
4810 * it happens that c_offset_min has been invalidated, since the
4811 * earlier string may buy us something the later one won't.]
4814 data.longest_fixed = newSVpvs("");
4815 data.longest_float = newSVpvs("");
4816 data.last_found = newSVpvs("");
4817 data.longest = &(data.longest_fixed);
4819 if (!ri->regstclass) {
4820 cl_init(pRExC_state, &ch_class);
4821 data.start_class = &ch_class;
4822 stclass_flag = SCF_DO_STCLASS_AND;
4823 } else /* XXXX Check for BOUND? */
4825 data.last_closep = &last_close;
4827 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4828 &data, -1, NULL, NULL,
4829 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4835 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4836 && data.last_start_min == 0 && data.last_end > 0
4837 && !RExC_seen_zerolen
4838 && !(RExC_seen & REG_SEEN_VERBARG)
4839 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4840 r->extflags |= RXf_CHECK_ALL;
4841 scan_commit(pRExC_state, &data,&minlen,0);
4842 SvREFCNT_dec(data.last_found);
4844 /* Note that code very similar to this but for anchored string
4845 follows immediately below, changes may need to be made to both.
4848 longest_float_length = CHR_SVLEN(data.longest_float);
4849 if (longest_float_length
4850 || (data.flags & SF_FL_BEFORE_EOL
4851 && (!(data.flags & SF_FL_BEFORE_MEOL)
4852 || (RExC_flags & RXf_PMf_MULTILINE))))
4856 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4857 && data.offset_fixed == data.offset_float_min
4858 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4859 goto remove_float; /* As in (a)+. */
4861 /* copy the information about the longest float from the reg_scan_data
4862 over to the program. */
4863 if (SvUTF8(data.longest_float)) {
4864 r->float_utf8 = data.longest_float;
4865 r->float_substr = NULL;
4867 r->float_substr = data.longest_float;
4868 r->float_utf8 = NULL;
4870 /* float_end_shift is how many chars that must be matched that
4871 follow this item. We calculate it ahead of time as once the
4872 lookbehind offset is added in we lose the ability to correctly
4874 ml = data.minlen_float ? *(data.minlen_float)
4875 : (I32)longest_float_length;
4876 r->float_end_shift = ml - data.offset_float_min
4877 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4878 + data.lookbehind_float;
4879 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4880 r->float_max_offset = data.offset_float_max;
4881 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4882 r->float_max_offset -= data.lookbehind_float;
4884 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4885 && (!(data.flags & SF_FL_BEFORE_MEOL)
4886 || (RExC_flags & RXf_PMf_MULTILINE)));
4887 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4891 r->float_substr = r->float_utf8 = NULL;
4892 SvREFCNT_dec(data.longest_float);
4893 longest_float_length = 0;
4896 /* Note that code very similar to this but for floating string
4897 is immediately above, changes may need to be made to both.
4900 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4901 if (longest_fixed_length
4902 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4903 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4904 || (RExC_flags & RXf_PMf_MULTILINE))))
4908 /* copy the information about the longest fixed
4909 from the reg_scan_data over to the program. */
4910 if (SvUTF8(data.longest_fixed)) {
4911 r->anchored_utf8 = data.longest_fixed;
4912 r->anchored_substr = NULL;
4914 r->anchored_substr = data.longest_fixed;
4915 r->anchored_utf8 = NULL;
4917 /* fixed_end_shift is how many chars that must be matched that
4918 follow this item. We calculate it ahead of time as once the
4919 lookbehind offset is added in we lose the ability to correctly
4921 ml = data.minlen_fixed ? *(data.minlen_fixed)
4922 : (I32)longest_fixed_length;
4923 r->anchored_end_shift = ml - data.offset_fixed
4924 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4925 + data.lookbehind_fixed;
4926 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4928 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4929 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4930 || (RExC_flags & RXf_PMf_MULTILINE)));
4931 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4934 r->anchored_substr = r->anchored_utf8 = NULL;
4935 SvREFCNT_dec(data.longest_fixed);
4936 longest_fixed_length = 0;
4939 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4940 ri->regstclass = NULL;
4941 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4943 && !(data.start_class->flags & ANYOF_EOS)
4944 && !cl_is_anything(data.start_class))
4946 const U32 n = add_data(pRExC_state, 1, "f");
4948 Newx(RExC_rxi->data->data[n], 1,
4949 struct regnode_charclass_class);
4950 StructCopy(data.start_class,
4951 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4952 struct regnode_charclass_class);
4953 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4954 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4955 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4956 regprop(r, sv, (regnode*)data.start_class);
4957 PerlIO_printf(Perl_debug_log,
4958 "synthetic stclass \"%s\".\n",
4959 SvPVX_const(sv));});
4962 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4963 if (longest_fixed_length > longest_float_length) {
4964 r->check_end_shift = r->anchored_end_shift;
4965 r->check_substr = r->anchored_substr;
4966 r->check_utf8 = r->anchored_utf8;
4967 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4968 if (r->extflags & RXf_ANCH_SINGLE)
4969 r->extflags |= RXf_NOSCAN;
4972 r->check_end_shift = r->float_end_shift;
4973 r->check_substr = r->float_substr;
4974 r->check_utf8 = r->float_utf8;
4975 r->check_offset_min = r->float_min_offset;
4976 r->check_offset_max = r->float_max_offset;
4978 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4979 This should be changed ASAP! */
4980 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4981 r->extflags |= RXf_USE_INTUIT;
4982 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4983 r->extflags |= RXf_INTUIT_TAIL;
4985 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4986 if ( (STRLEN)minlen < longest_float_length )
4987 minlen= longest_float_length;
4988 if ( (STRLEN)minlen < longest_fixed_length )
4989 minlen= longest_fixed_length;
4993 /* Several toplevels. Best we can is to set minlen. */
4995 struct regnode_charclass_class ch_class;
4998 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
5000 scan = ri->program + 1;
5001 cl_init(pRExC_state, &ch_class);
5002 data.start_class = &ch_class;
5003 data.last_closep = &last_close;
5006 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
5007 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
5011 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
5012 = r->float_substr = r->float_utf8 = NULL;
5013 if (!(data.start_class->flags & ANYOF_EOS)
5014 && !cl_is_anything(data.start_class))
5016 const U32 n = add_data(pRExC_state, 1, "f");
5018 Newx(RExC_rxi->data->data[n], 1,
5019 struct regnode_charclass_class);
5020 StructCopy(data.start_class,
5021 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5022 struct regnode_charclass_class);
5023 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5024 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5025 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
5026 regprop(r, sv, (regnode*)data.start_class);
5027 PerlIO_printf(Perl_debug_log,
5028 "synthetic stclass \"%s\".\n",
5029 SvPVX_const(sv));});
5033 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
5034 the "real" pattern. */
5036 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
5037 (IV)minlen, (IV)r->minlen);
5039 r->minlenret = minlen;
5040 if (r->minlen < minlen)
5043 if (RExC_seen & REG_SEEN_GPOS)
5044 r->extflags |= RXf_GPOS_SEEN;
5045 if (RExC_seen & REG_SEEN_LOOKBEHIND)
5046 r->extflags |= RXf_LOOKBEHIND_SEEN;
5047 if (RExC_seen & REG_SEEN_EVAL)
5048 r->extflags |= RXf_EVAL_SEEN;
5049 if (RExC_seen & REG_SEEN_CANY)
5050 r->extflags |= RXf_CANY_SEEN;
5051 if (RExC_seen & REG_SEEN_VERBARG)
5052 r->intflags |= PREGf_VERBARG_SEEN;
5053 if (RExC_seen & REG_SEEN_CUTGROUP)
5054 r->intflags |= PREGf_CUTGROUP_SEEN;
5055 if (RExC_paren_names)
5056 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
5058 RXp_PAREN_NAMES(r) = NULL;
5060 #ifdef STUPID_PATTERN_CHECKS
5061 if (RX_PRELEN(rx) == 0)
5062 r->extflags |= RXf_NULL;
5063 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5064 /* XXX: this should happen BEFORE we compile */
5065 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5066 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
5067 r->extflags |= RXf_WHITE;
5068 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
5069 r->extflags |= RXf_START_ONLY;
5071 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5072 /* XXX: this should happen BEFORE we compile */
5073 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5075 regnode *first = ri->program + 1;
5077 U8 nop = OP(NEXTOPER(first));
5079 if (PL_regkind[fop] == NOTHING && nop == END)
5080 r->extflags |= RXf_NULL;
5081 else if (PL_regkind[fop] == BOL && nop == END)
5082 r->extflags |= RXf_START_ONLY;
5083 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
5084 r->extflags |= RXf_WHITE;
5088 if (RExC_paren_names) {
5089 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5090 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5093 ri->name_list_idx = 0;
5095 if (RExC_recurse_count) {
5096 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5097 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5098 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5101 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5102 /* assume we don't need to swap parens around before we match */
5105 PerlIO_printf(Perl_debug_log,"Final program:\n");
5108 #ifdef RE_TRACK_PATTERN_OFFSETS
5109 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5110 const U32 len = ri->u.offsets[0];
5112 GET_RE_DEBUG_FLAGS_DECL;
5113 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5114 for (i = 1; i <= len; i++) {
5115 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5116 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5117 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5119 PerlIO_printf(Perl_debug_log, "\n");
5125 #undef RE_ENGINE_PTR
5129 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5132 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5134 PERL_UNUSED_ARG(value);
5136 if (flags & RXapif_FETCH) {
5137 return reg_named_buff_fetch(rx, key, flags);
5138 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5139 Perl_croak_no_modify(aTHX);
5141 } else if (flags & RXapif_EXISTS) {
5142 return reg_named_buff_exists(rx, key, flags)
5145 } else if (flags & RXapif_REGNAMES) {
5146 return reg_named_buff_all(rx, flags);
5147 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5148 return reg_named_buff_scalar(rx, flags);
5150 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5156 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5159 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5160 PERL_UNUSED_ARG(lastkey);
5162 if (flags & RXapif_FIRSTKEY)
5163 return reg_named_buff_firstkey(rx, flags);
5164 else if (flags & RXapif_NEXTKEY)
5165 return reg_named_buff_nextkey(rx, flags);
5167 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5173 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5176 AV *retarray = NULL;
5178 struct regexp *const rx = (struct regexp *)SvANY(r);
5180 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5182 if (flags & RXapif_ALL)
5185 if (rx && RXp_PAREN_NAMES(rx)) {
5186 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5189 SV* sv_dat=HeVAL(he_str);
5190 I32 *nums=(I32*)SvPVX(sv_dat);
5191 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5192 if ((I32)(rx->nparens) >= nums[i]
5193 && rx->offs[nums[i]].start != -1
5194 && rx->offs[nums[i]].end != -1)
5197 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5201 ret = newSVsv(&PL_sv_undef);
5204 av_push(retarray, ret);
5207 return newRV_noinc(MUTABLE_SV(retarray));
5214 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5217 struct regexp *const rx = (struct regexp *)SvANY(r);
5219 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5221 if (rx && RXp_PAREN_NAMES(rx)) {
5222 if (flags & RXapif_ALL) {
5223 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5225 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5239 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5241 struct regexp *const rx = (struct regexp *)SvANY(r);
5243 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5245 if ( rx && RXp_PAREN_NAMES(rx) ) {
5246 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5248 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5255 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5257 struct regexp *const rx = (struct regexp *)SvANY(r);
5258 GET_RE_DEBUG_FLAGS_DECL;
5260 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5262 if (rx && RXp_PAREN_NAMES(rx)) {
5263 HV *hv = RXp_PAREN_NAMES(rx);
5265 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5268 SV* sv_dat = HeVAL(temphe);
5269 I32 *nums = (I32*)SvPVX(sv_dat);
5270 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5271 if ((I32)(rx->lastparen) >= nums[i] &&
5272 rx->offs[nums[i]].start != -1 &&
5273 rx->offs[nums[i]].end != -1)
5279 if (parno || flags & RXapif_ALL) {
5280 return newSVhek(HeKEY_hek(temphe));
5288 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5293 struct regexp *const rx = (struct regexp *)SvANY(r);
5295 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5297 if (rx && RXp_PAREN_NAMES(rx)) {
5298 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5299 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5300 } else if (flags & RXapif_ONE) {
5301 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5302 av = MUTABLE_AV(SvRV(ret));
5303 length = av_len(av);
5305 return newSViv(length + 1);
5307 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5311 return &PL_sv_undef;
5315 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5317 struct regexp *const rx = (struct regexp *)SvANY(r);
5320 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5322 if (rx && RXp_PAREN_NAMES(rx)) {
5323 HV *hv= RXp_PAREN_NAMES(rx);
5325 (void)hv_iterinit(hv);
5326 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5329 SV* sv_dat = HeVAL(temphe);
5330 I32 *nums = (I32*)SvPVX(sv_dat);
5331 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5332 if ((I32)(rx->lastparen) >= nums[i] &&
5333 rx->offs[nums[i]].start != -1 &&
5334 rx->offs[nums[i]].end != -1)
5340 if (parno || flags & RXapif_ALL) {
5341 av_push(av, newSVhek(HeKEY_hek(temphe)));
5346 return newRV_noinc(MUTABLE_SV(av));
5350 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5353 struct regexp *const rx = (struct regexp *)SvANY(r);
5358 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5361 sv_setsv(sv,&PL_sv_undef);
5365 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5367 i = rx->offs[0].start;
5371 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5373 s = rx->subbeg + rx->offs[0].end;
5374 i = rx->sublen - rx->offs[0].end;
5377 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5378 (s1 = rx->offs[paren].start) != -1 &&
5379 (t1 = rx->offs[paren].end) != -1)
5383 s = rx->subbeg + s1;
5385 sv_setsv(sv,&PL_sv_undef);
5388 assert(rx->sublen >= (s - rx->subbeg) + i );
5390 const int oldtainted = PL_tainted;
5392 sv_setpvn(sv, s, i);
5393 PL_tainted = oldtainted;
5394 if ( (rx->extflags & RXf_CANY_SEEN)
5395 ? (RXp_MATCH_UTF8(rx)
5396 && (!i || is_utf8_string((U8*)s, i)))
5397 : (RXp_MATCH_UTF8(rx)) )
5404 if (RXp_MATCH_TAINTED(rx)) {
5405 if (SvTYPE(sv) >= SVt_PVMG) {
5406 MAGIC* const mg = SvMAGIC(sv);
5409 SvMAGIC_set(sv, mg->mg_moremagic);
5411 if ((mgt = SvMAGIC(sv))) {
5412 mg->mg_moremagic = mgt;
5413 SvMAGIC_set(sv, mg);
5423 sv_setsv(sv,&PL_sv_undef);
5429 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5430 SV const * const value)
5432 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5434 PERL_UNUSED_ARG(rx);
5435 PERL_UNUSED_ARG(paren);
5436 PERL_UNUSED_ARG(value);
5439 Perl_croak_no_modify(aTHX);
5443 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5446 struct regexp *const rx = (struct regexp *)SvANY(r);
5450 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5452 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5454 /* $` / ${^PREMATCH} */
5455 case RX_BUFF_IDX_PREMATCH:
5456 if (rx->offs[0].start != -1) {
5457 i = rx->offs[0].start;
5465 /* $' / ${^POSTMATCH} */
5466 case RX_BUFF_IDX_POSTMATCH:
5467 if (rx->offs[0].end != -1) {
5468 i = rx->sublen - rx->offs[0].end;
5470 s1 = rx->offs[0].end;
5476 /* $& / ${^MATCH}, $1, $2, ... */
5478 if (paren <= (I32)rx->nparens &&
5479 (s1 = rx->offs[paren].start) != -1 &&
5480 (t1 = rx->offs[paren].end) != -1)
5485 if (ckWARN(WARN_UNINITIALIZED))
5486 report_uninit((const SV *)sv);
5491 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5492 const char * const s = rx->subbeg + s1;
5497 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5504 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5506 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5507 PERL_UNUSED_ARG(rx);
5511 return newSVpvs("Regexp");
5514 /* Scans the name of a named buffer from the pattern.
5515 * If flags is REG_RSN_RETURN_NULL returns null.
5516 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5517 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5518 * to the parsed name as looked up in the RExC_paren_names hash.
5519 * If there is an error throws a vFAIL().. type exception.
5522 #define REG_RSN_RETURN_NULL 0
5523 #define REG_RSN_RETURN_NAME 1
5524 #define REG_RSN_RETURN_DATA 2
5527 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5529 char *name_start = RExC_parse;
5531 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5533 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5534 /* skip IDFIRST by using do...while */
5537 RExC_parse += UTF8SKIP(RExC_parse);
5538 } while (isALNUM_utf8((U8*)RExC_parse));
5542 } while (isALNUM(*RExC_parse));
5547 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5548 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5549 if ( flags == REG_RSN_RETURN_NAME)
5551 else if (flags==REG_RSN_RETURN_DATA) {
5554 if ( ! sv_name ) /* should not happen*/
5555 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5556 if (RExC_paren_names)
5557 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5559 sv_dat = HeVAL(he_str);
5561 vFAIL("Reference to nonexistent named group");
5565 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5572 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5573 int rem=(int)(RExC_end - RExC_parse); \
5582 if (RExC_lastparse!=RExC_parse) \
5583 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5586 iscut ? "..." : "<" \
5589 PerlIO_printf(Perl_debug_log,"%16s",""); \
5592 num = RExC_size + 1; \
5594 num=REG_NODE_NUM(RExC_emit); \
5595 if (RExC_lastnum!=num) \
5596 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5598 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5599 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5600 (int)((depth*2)), "", \
5604 RExC_lastparse=RExC_parse; \
5609 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5610 DEBUG_PARSE_MSG((funcname)); \
5611 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5613 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5614 DEBUG_PARSE_MSG((funcname)); \
5615 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5618 - reg - regular expression, i.e. main body or parenthesized thing
5620 * Caller must absorb opening parenthesis.
5622 * Combining parenthesis handling with the base level of regular expression
5623 * is a trifle forced, but the need to tie the tails of the branches to what
5624 * follows makes it hard to avoid.
5626 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5628 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5630 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5634 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5635 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5638 register regnode *ret; /* Will be the head of the group. */
5639 register regnode *br;
5640 register regnode *lastbr;
5641 register regnode *ender = NULL;
5642 register I32 parno = 0;
5644 U32 oregflags = RExC_flags;
5645 bool have_branch = 0;
5647 I32 freeze_paren = 0;
5648 I32 after_freeze = 0;
5650 /* for (?g), (?gc), and (?o) warnings; warning
5651 about (?c) will warn about (?g) -- japhy */
5653 #define WASTED_O 0x01
5654 #define WASTED_G 0x02
5655 #define WASTED_C 0x04
5656 #define WASTED_GC (0x02|0x04)
5657 I32 wastedflags = 0x00;
5659 char * parse_start = RExC_parse; /* MJD */
5660 char * const oregcomp_parse = RExC_parse;
5662 GET_RE_DEBUG_FLAGS_DECL;
5664 PERL_ARGS_ASSERT_REG;
5665 DEBUG_PARSE("reg ");
5667 *flagp = 0; /* Tentatively. */
5670 /* Make an OPEN node, if parenthesized. */
5672 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5673 char *start_verb = RExC_parse;
5674 STRLEN verb_len = 0;
5675 char *start_arg = NULL;
5676 unsigned char op = 0;
5678 int internal_argval = 0; /* internal_argval is only useful if !argok */
5679 while ( *RExC_parse && *RExC_parse != ')' ) {
5680 if ( *RExC_parse == ':' ) {
5681 start_arg = RExC_parse + 1;
5687 verb_len = RExC_parse - start_verb;
5690 while ( *RExC_parse && *RExC_parse != ')' )
5692 if ( *RExC_parse != ')' )
5693 vFAIL("Unterminated verb pattern argument");
5694 if ( RExC_parse == start_arg )
5697 if ( *RExC_parse != ')' )
5698 vFAIL("Unterminated verb pattern");
5701 switch ( *start_verb ) {
5702 case 'A': /* (*ACCEPT) */
5703 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5705 internal_argval = RExC_nestroot;
5708 case 'C': /* (*COMMIT) */
5709 if ( memEQs(start_verb,verb_len,"COMMIT") )
5712 case 'F': /* (*FAIL) */
5713 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5718 case ':': /* (*:NAME) */
5719 case 'M': /* (*MARK:NAME) */
5720 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5725 case 'P': /* (*PRUNE) */
5726 if ( memEQs(start_verb,verb_len,"PRUNE") )
5729 case 'S': /* (*SKIP) */
5730 if ( memEQs(start_verb,verb_len,"SKIP") )
5733 case 'T': /* (*THEN) */
5734 /* [19:06] <TimToady> :: is then */
5735 if ( memEQs(start_verb,verb_len,"THEN") ) {
5737 RExC_seen |= REG_SEEN_CUTGROUP;
5743 vFAIL3("Unknown verb pattern '%.*s'",
5744 verb_len, start_verb);
5747 if ( start_arg && internal_argval ) {
5748 vFAIL3("Verb pattern '%.*s' may not have an argument",
5749 verb_len, start_verb);
5750 } else if ( argok < 0 && !start_arg ) {
5751 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5752 verb_len, start_verb);
5754 ret = reganode(pRExC_state, op, internal_argval);
5755 if ( ! internal_argval && ! SIZE_ONLY ) {
5757 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5758 ARG(ret) = add_data( pRExC_state, 1, "S" );
5759 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5766 if (!internal_argval)
5767 RExC_seen |= REG_SEEN_VERBARG;
5768 } else if ( start_arg ) {
5769 vFAIL3("Verb pattern '%.*s' may not have an argument",
5770 verb_len, start_verb);
5772 ret = reg_node(pRExC_state, op);
5774 nextchar(pRExC_state);
5777 if (*RExC_parse == '?') { /* (?...) */
5778 bool is_logical = 0;
5779 const char * const seqstart = RExC_parse;
5780 bool has_use_defaults = FALSE;
5783 paren = *RExC_parse++;
5784 ret = NULL; /* For look-ahead/behind. */
5787 case 'P': /* (?P...) variants for those used to PCRE/Python */
5788 paren = *RExC_parse++;
5789 if ( paren == '<') /* (?P<...>) named capture */
5791 else if (paren == '>') { /* (?P>name) named recursion */
5792 goto named_recursion;
5794 else if (paren == '=') { /* (?P=...) named backref */
5795 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5796 you change this make sure you change that */
5797 char* name_start = RExC_parse;
5799 SV *sv_dat = reg_scan_name(pRExC_state,
5800 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5801 if (RExC_parse == name_start || *RExC_parse != ')')
5802 vFAIL2("Sequence %.3s... not terminated",parse_start);
5805 num = add_data( pRExC_state, 1, "S" );
5806 RExC_rxi->data->data[num]=(void*)sv_dat;
5807 SvREFCNT_inc_simple_void(sv_dat);
5810 ret = reganode(pRExC_state,
5811 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5815 Set_Node_Offset(ret, parse_start+1);
5816 Set_Node_Cur_Length(ret); /* MJD */
5818 nextchar(pRExC_state);
5822 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5824 case '<': /* (?<...) */
5825 if (*RExC_parse == '!')
5827 else if (*RExC_parse != '=')
5833 case '\'': /* (?'...') */
5834 name_start= RExC_parse;
5835 svname = reg_scan_name(pRExC_state,
5836 SIZE_ONLY ? /* reverse test from the others */
5837 REG_RSN_RETURN_NAME :
5838 REG_RSN_RETURN_NULL);
5839 if (RExC_parse == name_start) {
5841 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5844 if (*RExC_parse != paren)
5845 vFAIL2("Sequence (?%c... not terminated",
5846 paren=='>' ? '<' : paren);
5850 if (!svname) /* shouldnt happen */
5852 "panic: reg_scan_name returned NULL");
5853 if (!RExC_paren_names) {
5854 RExC_paren_names= newHV();
5855 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5857 RExC_paren_name_list= newAV();
5858 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5861 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5863 sv_dat = HeVAL(he_str);
5865 /* croak baby croak */
5867 "panic: paren_name hash element allocation failed");
5868 } else if ( SvPOK(sv_dat) ) {
5869 /* (?|...) can mean we have dupes so scan to check
5870 its already been stored. Maybe a flag indicating
5871 we are inside such a construct would be useful,
5872 but the arrays are likely to be quite small, so
5873 for now we punt -- dmq */
5874 IV count = SvIV(sv_dat);
5875 I32 *pv = (I32*)SvPVX(sv_dat);
5877 for ( i = 0 ; i < count ; i++ ) {
5878 if ( pv[i] == RExC_npar ) {
5884 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5885 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5886 pv[count] = RExC_npar;
5887 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5890 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5891 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5893 SvIV_set(sv_dat, 1);
5896 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5897 SvREFCNT_dec(svname);
5900 /*sv_dump(sv_dat);*/
5902 nextchar(pRExC_state);
5904 goto capturing_parens;
5906 RExC_seen |= REG_SEEN_LOOKBEHIND;
5908 case '=': /* (?=...) */
5909 RExC_seen_zerolen++;
5911 case '!': /* (?!...) */
5912 RExC_seen_zerolen++;
5913 if (*RExC_parse == ')') {
5914 ret=reg_node(pRExC_state, OPFAIL);
5915 nextchar(pRExC_state);
5919 case '|': /* (?|...) */
5920 /* branch reset, behave like a (?:...) except that
5921 buffers in alternations share the same numbers */
5923 after_freeze = freeze_paren = RExC_npar;
5925 case ':': /* (?:...) */
5926 case '>': /* (?>...) */
5928 case '$': /* (?$...) */
5929 case '@': /* (?@...) */
5930 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5932 case '#': /* (?#...) */
5933 while (*RExC_parse && *RExC_parse != ')')
5935 if (*RExC_parse != ')')
5936 FAIL("Sequence (?#... not terminated");
5937 nextchar(pRExC_state);
5940 case '0' : /* (?0) */
5941 case 'R' : /* (?R) */
5942 if (*RExC_parse != ')')
5943 FAIL("Sequence (?R) not terminated");
5944 ret = reg_node(pRExC_state, GOSTART);
5945 *flagp |= POSTPONED;
5946 nextchar(pRExC_state);
5949 { /* named and numeric backreferences */
5951 case '&': /* (?&NAME) */
5952 parse_start = RExC_parse - 1;
5955 SV *sv_dat = reg_scan_name(pRExC_state,
5956 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5957 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5959 goto gen_recurse_regop;
5962 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5964 vFAIL("Illegal pattern");
5966 goto parse_recursion;
5968 case '-': /* (?-1) */
5969 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5970 RExC_parse--; /* rewind to let it be handled later */
5974 case '1': case '2': case '3': case '4': /* (?1) */
5975 case '5': case '6': case '7': case '8': case '9':
5978 num = atoi(RExC_parse);
5979 parse_start = RExC_parse - 1; /* MJD */
5980 if (*RExC_parse == '-')
5982 while (isDIGIT(*RExC_parse))
5984 if (*RExC_parse!=')')
5985 vFAIL("Expecting close bracket");
5988 if ( paren == '-' ) {
5990 Diagram of capture buffer numbering.
5991 Top line is the normal capture buffer numbers
5992 Bottom line is the negative indexing as from
5996 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
6000 num = RExC_npar + num;
6003 vFAIL("Reference to nonexistent group");
6005 } else if ( paren == '+' ) {
6006 num = RExC_npar + num - 1;
6009 ret = reganode(pRExC_state, GOSUB, num);
6011 if (num > (I32)RExC_rx->nparens) {
6013 vFAIL("Reference to nonexistent group");
6015 ARG2L_SET( ret, RExC_recurse_count++);
6017 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6018 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
6022 RExC_seen |= REG_SEEN_RECURSE;
6023 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
6024 Set_Node_Offset(ret, parse_start); /* MJD */
6026 *flagp |= POSTPONED;
6027 nextchar(pRExC_state);
6029 } /* named and numeric backreferences */
6032 case '?': /* (??...) */
6034 if (*RExC_parse != '{') {
6036 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6039 *flagp |= POSTPONED;
6040 paren = *RExC_parse++;
6042 case '{': /* (?{...}) */
6047 char *s = RExC_parse;
6049 RExC_seen_zerolen++;
6050 RExC_seen |= REG_SEEN_EVAL;
6051 while (count && (c = *RExC_parse)) {
6062 if (*RExC_parse != ')') {
6064 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
6068 OP_4tree *sop, *rop;
6069 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
6072 Perl_save_re_context(aTHX);
6073 rop = sv_compile_2op(sv, &sop, "re", &pad);
6074 sop->op_private |= OPpREFCOUNTED;
6075 /* re_dup will OpREFCNT_inc */
6076 OpREFCNT_set(sop, 1);
6079 n = add_data(pRExC_state, 3, "nop");
6080 RExC_rxi->data->data[n] = (void*)rop;
6081 RExC_rxi->data->data[n+1] = (void*)sop;
6082 RExC_rxi->data->data[n+2] = (void*)pad;
6085 else { /* First pass */
6086 if (PL_reginterp_cnt < ++RExC_seen_evals
6088 /* No compiled RE interpolated, has runtime
6089 components ===> unsafe. */
6090 FAIL("Eval-group not allowed at runtime, use re 'eval'");
6091 if (PL_tainting && PL_tainted)
6092 FAIL("Eval-group in insecure regular expression");
6093 #if PERL_VERSION > 8
6094 if (IN_PERL_COMPILETIME)
6099 nextchar(pRExC_state);
6101 ret = reg_node(pRExC_state, LOGICAL);
6104 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
6105 /* deal with the length of this later - MJD */
6108 ret = reganode(pRExC_state, EVAL, n);
6109 Set_Node_Length(ret, RExC_parse - parse_start + 1);
6110 Set_Node_Offset(ret, parse_start);
6113 case '(': /* (?(?{...})...) and (?(?=...)...) */
6116 if (RExC_parse[0] == '?') { /* (?(?...)) */
6117 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
6118 || RExC_parse[1] == '<'
6119 || RExC_parse[1] == '{') { /* Lookahead or eval. */
6122 ret = reg_node(pRExC_state, LOGICAL);
6125 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
6129 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
6130 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
6132 char ch = RExC_parse[0] == '<' ? '>' : '\'';
6133 char *name_start= RExC_parse++;
6135 SV *sv_dat=reg_scan_name(pRExC_state,
6136 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6137 if (RExC_parse == name_start || *RExC_parse != ch)
6138 vFAIL2("Sequence (?(%c... not terminated",
6139 (ch == '>' ? '<' : ch));
6142 num = add_data( pRExC_state, 1, "S" );
6143 RExC_rxi->data->data[num]=(void*)sv_dat;
6144 SvREFCNT_inc_simple_void(sv_dat);
6146 ret = reganode(pRExC_state,NGROUPP,num);
6147 goto insert_if_check_paren;
6149 else if (RExC_parse[0] == 'D' &&
6150 RExC_parse[1] == 'E' &&
6151 RExC_parse[2] == 'F' &&
6152 RExC_parse[3] == 'I' &&
6153 RExC_parse[4] == 'N' &&
6154 RExC_parse[5] == 'E')
6156 ret = reganode(pRExC_state,DEFINEP,0);
6159 goto insert_if_check_paren;
6161 else if (RExC_parse[0] == 'R') {
6164 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6165 parno = atoi(RExC_parse++);
6166 while (isDIGIT(*RExC_parse))
6168 } else if (RExC_parse[0] == '&') {
6171 sv_dat = reg_scan_name(pRExC_state,
6172 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6173 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6175 ret = reganode(pRExC_state,INSUBP,parno);
6176 goto insert_if_check_paren;
6178 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6181 parno = atoi(RExC_parse++);
6183 while (isDIGIT(*RExC_parse))
6185 ret = reganode(pRExC_state, GROUPP, parno);
6187 insert_if_check_paren:
6188 if ((c = *nextchar(pRExC_state)) != ')')
6189 vFAIL("Switch condition not recognized");
6191 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
6192 br = regbranch(pRExC_state, &flags, 1,depth+1);
6194 br = reganode(pRExC_state, LONGJMP, 0);
6196 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
6197 c = *nextchar(pRExC_state);
6202 vFAIL("(?(DEFINE)....) does not allow branches");
6203 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6204 regbranch(pRExC_state, &flags, 1,depth+1);
6205 REGTAIL(pRExC_state, ret, lastbr);
6208 c = *nextchar(pRExC_state);
6213 vFAIL("Switch (?(condition)... contains too many branches");
6214 ender = reg_node(pRExC_state, TAIL);
6215 REGTAIL(pRExC_state, br, ender);
6217 REGTAIL(pRExC_state, lastbr, ender);
6218 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6221 REGTAIL(pRExC_state, ret, ender);
6222 RExC_size++; /* XXX WHY do we need this?!!
6223 For large programs it seems to be required
6224 but I can't figure out why. -- dmq*/
6228 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6232 RExC_parse--; /* for vFAIL to print correctly */
6233 vFAIL("Sequence (? incomplete");
6235 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
6237 has_use_defaults = TRUE;
6238 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
6239 RExC_flags &= ~(RXf_PMf_LOCALE|RXf_PMf_UNICODE);
6243 parse_flags: /* (?i) */
6245 U32 posflags = 0, negflags = 0;
6246 U32 *flagsp = &posflags;
6247 bool has_charset_modifier = 0;
6249 while (*RExC_parse) {
6250 /* && strchr("iogcmsx", *RExC_parse) */
6251 /* (?g), (?gc) and (?o) are useless here
6252 and must be globally applied -- japhy */
6253 switch (*RExC_parse) {
6254 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6255 case LOCALE_PAT_MOD:
6256 if (has_charset_modifier || flagsp == &negflags) {
6257 goto fail_modifiers;
6259 posflags |= RXf_PMf_LOCALE;
6260 negflags |= RXf_PMf_UNICODE;
6261 has_charset_modifier = 1;
6263 case UNICODE_PAT_MOD:
6264 if (has_charset_modifier || flagsp == &negflags) {
6265 goto fail_modifiers;
6267 posflags |= RXf_PMf_UNICODE;
6268 negflags |= RXf_PMf_LOCALE;
6269 has_charset_modifier = 1;
6272 if (has_use_defaults
6273 || has_charset_modifier
6274 || flagsp == &negflags)
6276 goto fail_modifiers;
6278 negflags |= (RXf_PMf_LOCALE|RXf_PMf_UNICODE);
6279 has_charset_modifier = 1;
6281 case ONCE_PAT_MOD: /* 'o' */
6282 case GLOBAL_PAT_MOD: /* 'g' */
6283 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6284 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6285 if (! (wastedflags & wflagbit) ) {
6286 wastedflags |= wflagbit;
6289 "Useless (%s%c) - %suse /%c modifier",
6290 flagsp == &negflags ? "?-" : "?",
6292 flagsp == &negflags ? "don't " : "",
6299 case CONTINUE_PAT_MOD: /* 'c' */
6300 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6301 if (! (wastedflags & WASTED_C) ) {
6302 wastedflags |= WASTED_GC;
6305 "Useless (%sc) - %suse /gc modifier",
6306 flagsp == &negflags ? "?-" : "?",
6307 flagsp == &negflags ? "don't " : ""
6312 case KEEPCOPY_PAT_MOD: /* 'p' */
6313 if (flagsp == &negflags) {
6315 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
6317 *flagsp |= RXf_PMf_KEEPCOPY;
6321 /* A flag is a default iff it is following a minus, so
6322 * if there is a minus, it means will be trying to
6323 * re-specify a default which is an error */
6324 if (has_use_defaults || flagsp == &negflags) {
6327 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6331 wastedflags = 0; /* reset so (?g-c) warns twice */
6337 RExC_flags |= posflags;
6338 RExC_flags &= ~negflags;
6340 oregflags |= posflags;
6341 oregflags &= ~negflags;
6343 nextchar(pRExC_state);
6354 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6359 }} /* one for the default block, one for the switch */
6366 ret = reganode(pRExC_state, OPEN, parno);
6369 RExC_nestroot = parno;
6370 if (RExC_seen & REG_SEEN_RECURSE
6371 && !RExC_open_parens[parno-1])
6373 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6374 "Setting open paren #%"IVdf" to %d\n",
6375 (IV)parno, REG_NODE_NUM(ret)));
6376 RExC_open_parens[parno-1]= ret;
6379 Set_Node_Length(ret, 1); /* MJD */
6380 Set_Node_Offset(ret, RExC_parse); /* MJD */
6388 /* Pick up the branches, linking them together. */
6389 parse_start = RExC_parse; /* MJD */
6390 br = regbranch(pRExC_state, &flags, 1,depth+1);
6393 if (RExC_npar > after_freeze)
6394 after_freeze = RExC_npar;
6395 RExC_npar = freeze_paren;
6398 /* branch_len = (paren != 0); */
6402 if (*RExC_parse == '|') {
6403 if (!SIZE_ONLY && RExC_extralen) {
6404 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6407 reginsert(pRExC_state, BRANCH, br, depth+1);
6408 Set_Node_Length(br, paren != 0);
6409 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6413 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6415 else if (paren == ':') {
6416 *flagp |= flags&SIMPLE;
6418 if (is_open) { /* Starts with OPEN. */
6419 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6421 else if (paren != '?') /* Not Conditional */
6423 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6425 while (*RExC_parse == '|') {
6426 if (!SIZE_ONLY && RExC_extralen) {
6427 ender = reganode(pRExC_state, LONGJMP,0);
6428 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6431 RExC_extralen += 2; /* Account for LONGJMP. */
6432 nextchar(pRExC_state);
6434 if (RExC_npar > after_freeze)
6435 after_freeze = RExC_npar;
6436 RExC_npar = freeze_paren;
6438 br = regbranch(pRExC_state, &flags, 0, depth+1);
6442 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6444 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6447 if (have_branch || paren != ':') {
6448 /* Make a closing node, and hook it on the end. */
6451 ender = reg_node(pRExC_state, TAIL);
6454 ender = reganode(pRExC_state, CLOSE, parno);
6455 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6456 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6457 "Setting close paren #%"IVdf" to %d\n",
6458 (IV)parno, REG_NODE_NUM(ender)));
6459 RExC_close_parens[parno-1]= ender;
6460 if (RExC_nestroot == parno)
6463 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6464 Set_Node_Length(ender,1); /* MJD */
6470 *flagp &= ~HASWIDTH;
6473 ender = reg_node(pRExC_state, SUCCEED);
6476 ender = reg_node(pRExC_state, END);
6478 assert(!RExC_opend); /* there can only be one! */
6483 REGTAIL(pRExC_state, lastbr, ender);
6485 if (have_branch && !SIZE_ONLY) {
6487 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6489 /* Hook the tails of the branches to the closing node. */
6490 for (br = ret; br; br = regnext(br)) {
6491 const U8 op = PL_regkind[OP(br)];
6493 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6495 else if (op == BRANCHJ) {
6496 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6504 static const char parens[] = "=!<,>";
6506 if (paren && (p = strchr(parens, paren))) {
6507 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6508 int flag = (p - parens) > 1;
6511 node = SUSPEND, flag = 0;
6512 reginsert(pRExC_state, node,ret, depth+1);
6513 Set_Node_Cur_Length(ret);
6514 Set_Node_Offset(ret, parse_start + 1);
6516 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6520 /* Check for proper termination. */
6522 RExC_flags = oregflags;
6523 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6524 RExC_parse = oregcomp_parse;
6525 vFAIL("Unmatched (");
6528 else if (!paren && RExC_parse < RExC_end) {
6529 if (*RExC_parse == ')') {
6531 vFAIL("Unmatched )");
6534 FAIL("Junk on end of regexp"); /* "Can't happen". */
6538 RExC_npar = after_freeze;
6543 - regbranch - one alternative of an | operator
6545 * Implements the concatenation operator.
6548 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6551 register regnode *ret;
6552 register regnode *chain = NULL;
6553 register regnode *latest;
6554 I32 flags = 0, c = 0;
6555 GET_RE_DEBUG_FLAGS_DECL;
6557 PERL_ARGS_ASSERT_REGBRANCH;
6559 DEBUG_PARSE("brnc");
6564 if (!SIZE_ONLY && RExC_extralen)
6565 ret = reganode(pRExC_state, BRANCHJ,0);
6567 ret = reg_node(pRExC_state, BRANCH);
6568 Set_Node_Length(ret, 1);
6572 if (!first && SIZE_ONLY)
6573 RExC_extralen += 1; /* BRANCHJ */
6575 *flagp = WORST; /* Tentatively. */
6578 nextchar(pRExC_state);
6579 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6581 latest = regpiece(pRExC_state, &flags,depth+1);
6582 if (latest == NULL) {
6583 if (flags & TRYAGAIN)
6587 else if (ret == NULL)
6589 *flagp |= flags&(HASWIDTH|POSTPONED);
6590 if (chain == NULL) /* First piece. */
6591 *flagp |= flags&SPSTART;
6594 REGTAIL(pRExC_state, chain, latest);
6599 if (chain == NULL) { /* Loop ran zero times. */
6600 chain = reg_node(pRExC_state, NOTHING);
6605 *flagp |= flags&SIMPLE;
6612 - regpiece - something followed by possible [*+?]
6614 * Note that the branching code sequences used for ? and the general cases
6615 * of * and + are somewhat optimized: they use the same NOTHING node as
6616 * both the endmarker for their branch list and the body of the last branch.
6617 * It might seem that this node could be dispensed with entirely, but the
6618 * endmarker role is not redundant.
6621 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6624 register regnode *ret;
6626 register char *next;
6628 const char * const origparse = RExC_parse;
6630 I32 max = REG_INFTY;
6632 const char *maxpos = NULL;
6633 GET_RE_DEBUG_FLAGS_DECL;
6635 PERL_ARGS_ASSERT_REGPIECE;
6637 DEBUG_PARSE("piec");
6639 ret = regatom(pRExC_state, &flags,depth+1);
6641 if (flags & TRYAGAIN)
6648 if (op == '{' && regcurly(RExC_parse)) {
6650 parse_start = RExC_parse; /* MJD */
6651 next = RExC_parse + 1;
6652 while (isDIGIT(*next) || *next == ',') {
6661 if (*next == '}') { /* got one */
6665 min = atoi(RExC_parse);
6669 maxpos = RExC_parse;
6671 if (!max && *maxpos != '0')
6672 max = REG_INFTY; /* meaning "infinity" */
6673 else if (max >= REG_INFTY)
6674 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6676 nextchar(pRExC_state);
6679 if ((flags&SIMPLE)) {
6680 RExC_naughty += 2 + RExC_naughty / 2;
6681 reginsert(pRExC_state, CURLY, ret, depth+1);
6682 Set_Node_Offset(ret, parse_start+1); /* MJD */
6683 Set_Node_Cur_Length(ret);
6686 regnode * const w = reg_node(pRExC_state, WHILEM);
6689 REGTAIL(pRExC_state, ret, w);
6690 if (!SIZE_ONLY && RExC_extralen) {
6691 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6692 reginsert(pRExC_state, NOTHING,ret, depth+1);
6693 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6695 reginsert(pRExC_state, CURLYX,ret, depth+1);
6697 Set_Node_Offset(ret, parse_start+1);
6698 Set_Node_Length(ret,
6699 op == '{' ? (RExC_parse - parse_start) : 1);
6701 if (!SIZE_ONLY && RExC_extralen)
6702 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6703 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6705 RExC_whilem_seen++, RExC_extralen += 3;
6706 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6715 vFAIL("Can't do {n,m} with n > m");
6717 ARG1_SET(ret, (U16)min);
6718 ARG2_SET(ret, (U16)max);
6730 #if 0 /* Now runtime fix should be reliable. */
6732 /* if this is reinstated, don't forget to put this back into perldiag:
6734 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6736 (F) The part of the regexp subject to either the * or + quantifier
6737 could match an empty string. The {#} shows in the regular
6738 expression about where the problem was discovered.
6742 if (!(flags&HASWIDTH) && op != '?')
6743 vFAIL("Regexp *+ operand could be empty");
6746 parse_start = RExC_parse;
6747 nextchar(pRExC_state);
6749 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6751 if (op == '*' && (flags&SIMPLE)) {
6752 reginsert(pRExC_state, STAR, ret, depth+1);
6756 else if (op == '*') {
6760 else if (op == '+' && (flags&SIMPLE)) {
6761 reginsert(pRExC_state, PLUS, ret, depth+1);
6765 else if (op == '+') {
6769 else if (op == '?') {
6774 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
6775 ckWARN3reg(RExC_parse,
6776 "%.*s matches null string many times",
6777 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6781 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6782 nextchar(pRExC_state);
6783 reginsert(pRExC_state, MINMOD, ret, depth+1);
6784 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6786 #ifndef REG_ALLOW_MINMOD_SUSPEND
6789 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6791 nextchar(pRExC_state);
6792 ender = reg_node(pRExC_state, SUCCEED);
6793 REGTAIL(pRExC_state, ret, ender);
6794 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6796 ender = reg_node(pRExC_state, TAIL);
6797 REGTAIL(pRExC_state, ret, ender);
6801 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6803 vFAIL("Nested quantifiers");
6810 /* reg_namedseq(pRExC_state,UVp)
6812 This is expected to be called by a parser routine that has
6813 recognized '\N' and needs to handle the rest. RExC_parse is
6814 expected to point at the first char following the N at the time
6817 The \N may be inside (indicated by valuep not being NULL) or outside a
6820 \N may begin either a named sequence, or if outside a character class, mean
6821 to match a non-newline. For non single-quoted regexes, the tokenizer has
6822 attempted to decide which, and in the case of a named sequence converted it
6823 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
6824 where c1... are the characters in the sequence. For single-quoted regexes,
6825 the tokenizer passes the \N sequence through unchanged; this code will not
6826 attempt to determine this nor expand those. The net effect is that if the
6827 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
6828 signals that this \N occurrence means to match a non-newline.
6830 Only the \N{U+...} form should occur in a character class, for the same
6831 reason that '.' inside a character class means to just match a period: it
6832 just doesn't make sense.
6834 If valuep is non-null then it is assumed that we are parsing inside
6835 of a charclass definition and the first codepoint in the resolved
6836 string is returned via *valuep and the routine will return NULL.
6837 In this mode if a multichar string is returned from the charnames
6838 handler, a warning will be issued, and only the first char in the
6839 sequence will be examined. If the string returned is zero length
6840 then the value of *valuep is undefined and NON-NULL will
6841 be returned to indicate failure. (This will NOT be a valid pointer
6844 If valuep is null then it is assumed that we are parsing normal text and a
6845 new EXACT node is inserted into the program containing the resolved string,
6846 and a pointer to the new node is returned. But if the string is zero length
6847 a NOTHING node is emitted instead.
6849 On success RExC_parse is set to the char following the endbrace.
6850 Parsing failures will generate a fatal error via vFAIL(...)
6853 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp)
6855 char * endbrace; /* '}' following the name */
6856 regnode *ret = NULL;
6858 char* parse_start = RExC_parse - 2; /* points to the '\N' */
6862 GET_RE_DEBUG_FLAGS_DECL;
6864 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6868 /* The [^\n] meaning of \N ignores spaces and comments under the /x
6869 * modifier. The other meaning does not */
6870 p = (RExC_flags & RXf_PMf_EXTENDED)
6871 ? regwhite( pRExC_state, RExC_parse )
6874 /* Disambiguate between \N meaning a named character versus \N meaning
6875 * [^\n]. The former is assumed when it can't be the latter. */
6876 if (*p != '{' || regcurly(p)) {
6879 /* no bare \N in a charclass */
6880 vFAIL("\\N in a character class must be a named character: \\N{...}");
6882 nextchar(pRExC_state);
6883 ret = reg_node(pRExC_state, REG_ANY);
6884 *flagp |= HASWIDTH|SIMPLE;
6887 Set_Node_Length(ret, 1); /* MJD */
6891 /* Here, we have decided it should be a named sequence */
6893 /* The test above made sure that the next real character is a '{', but
6894 * under the /x modifier, it could be separated by space (or a comment and
6895 * \n) and this is not allowed (for consistency with \x{...} and the
6896 * tokenizer handling of \N{NAME}). */
6897 if (*RExC_parse != '{') {
6898 vFAIL("Missing braces on \\N{}");
6901 RExC_parse++; /* Skip past the '{' */
6903 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
6904 || ! (endbrace == RExC_parse /* nothing between the {} */
6905 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
6906 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
6908 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
6909 vFAIL("\\N{NAME} must be resolved by the lexer");
6912 if (endbrace == RExC_parse) { /* empty: \N{} */
6914 RExC_parse = endbrace + 1;
6915 return reg_node(pRExC_state,NOTHING);
6919 ckWARNreg(RExC_parse,
6920 "Ignoring zero length \\N{} in character class"
6922 RExC_parse = endbrace + 1;
6925 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
6928 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
6929 RExC_parse += 2; /* Skip past the 'U+' */
6931 if (valuep) { /* In a bracketed char class */
6932 /* We only pay attention to the first char of
6933 multichar strings being returned. I kinda wonder
6934 if this makes sense as it does change the behaviour
6935 from earlier versions, OTOH that behaviour was broken
6936 as well. XXX Solution is to recharacterize as
6937 [rest-of-class]|multi1|multi2... */
6939 STRLEN length_of_hex;
6940 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6941 | PERL_SCAN_DISALLOW_PREFIX
6942 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6944 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
6945 if (endchar < endbrace) {
6946 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
6949 length_of_hex = (STRLEN)(endchar - RExC_parse);
6950 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
6952 /* The tokenizer should have guaranteed validity, but it's possible to
6953 * bypass it by using single quoting, so check */
6954 if (length_of_hex == 0
6955 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6957 RExC_parse += length_of_hex; /* Includes all the valid */
6958 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6959 ? UTF8SKIP(RExC_parse)
6961 /* Guard against malformed utf8 */
6962 if (RExC_parse >= endchar) RExC_parse = endchar;
6963 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6966 RExC_parse = endbrace + 1;
6967 if (endchar == endbrace) return NULL;
6969 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
6971 else { /* Not a char class */
6972 char *s; /* String to put in generated EXACT node */
6973 STRLEN len = 0; /* Its current byte length */
6974 char *endchar; /* Points to '.' or '}' ending cur char in the input
6977 ret = reg_node(pRExC_state,
6978 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6981 /* Exact nodes can hold only a U8 length's of text = 255. Loop through
6982 * the input which is of the form now 'c1.c2.c3...}' until find the
6983 * ending brace or exceed length 255. The characters that exceed this
6984 * limit are dropped. The limit could be relaxed should it become
6985 * desirable by reparsing this as (?:\N{NAME}), so could generate
6986 * multiple EXACT nodes, as is done for just regular input. But this
6987 * is primarily a named character, and not intended to be a huge long
6988 * string, so 255 bytes should be good enough */
6990 STRLEN length_of_hex;
6991 I32 grok_flags = PERL_SCAN_ALLOW_UNDERSCORES
6992 | PERL_SCAN_DISALLOW_PREFIX
6993 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6994 UV cp; /* Ord of current character */
6996 /* Code points are separated by dots. If none, there is only one
6997 * code point, and is terminated by the brace */
6998 endchar = RExC_parse + strcspn(RExC_parse, ".}");
7000 /* The values are Unicode even on EBCDIC machines */
7001 length_of_hex = (STRLEN)(endchar - RExC_parse);
7002 cp = grok_hex(RExC_parse, &length_of_hex, &grok_flags, NULL);
7003 if ( length_of_hex == 0
7004 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
7006 RExC_parse += length_of_hex; /* Includes all the valid */
7007 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
7008 ? UTF8SKIP(RExC_parse)
7010 /* Guard against malformed utf8 */
7011 if (RExC_parse >= endchar) RExC_parse = endchar;
7012 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7015 if (! FOLD) { /* Not folding, just append to the string */
7018 /* Quit before adding this character if would exceed limit */
7019 if (len + UNISKIP(cp) > U8_MAX) break;
7021 unilen = reguni(pRExC_state, cp, s);
7026 } else { /* Folding, output the folded equivalent */
7027 STRLEN foldlen,numlen;
7028 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7029 cp = toFOLD_uni(cp, tmpbuf, &foldlen);
7031 /* Quit before exceeding size limit */
7032 if (len + foldlen > U8_MAX) break;
7034 for (foldbuf = tmpbuf;
7038 cp = utf8_to_uvchr(foldbuf, &numlen);
7040 const STRLEN unilen = reguni(pRExC_state, cp, s);
7043 /* In EBCDIC the numlen and unilen can differ. */
7045 if (numlen >= foldlen)
7049 break; /* "Can't happen." */
7053 /* Point to the beginning of the next character in the sequence. */
7054 RExC_parse = endchar + 1;
7056 /* Quit if no more characters */
7057 if (RExC_parse >= endbrace) break;
7062 if (RExC_parse < endbrace) {
7063 ckWARNreg(RExC_parse - 1,
7064 "Using just the first characters returned by \\N{}");
7067 RExC_size += STR_SZ(len);
7070 RExC_emit += STR_SZ(len);
7073 RExC_parse = endbrace + 1;
7075 *flagp |= HASWIDTH; /* Not SIMPLE, as that causes the engine to fail
7076 with malformed in t/re/pat_advanced.t */
7078 Set_Node_Cur_Length(ret); /* MJD */
7079 nextchar(pRExC_state);
7089 * It returns the code point in utf8 for the value in *encp.
7090 * value: a code value in the source encoding
7091 * encp: a pointer to an Encode object
7093 * If the result from Encode is not a single character,
7094 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
7097 S_reg_recode(pTHX_ const char value, SV **encp)
7100 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
7101 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
7102 const STRLEN newlen = SvCUR(sv);
7103 UV uv = UNICODE_REPLACEMENT;
7105 PERL_ARGS_ASSERT_REG_RECODE;
7109 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
7112 if (!newlen || numlen != newlen) {
7113 uv = UNICODE_REPLACEMENT;
7121 - regatom - the lowest level
7123 Try to identify anything special at the start of the pattern. If there
7124 is, then handle it as required. This may involve generating a single regop,
7125 such as for an assertion; or it may involve recursing, such as to
7126 handle a () structure.
7128 If the string doesn't start with something special then we gobble up
7129 as much literal text as we can.
7131 Once we have been able to handle whatever type of thing started the
7132 sequence, we return.
7134 Note: we have to be careful with escapes, as they can be both literal
7135 and special, and in the case of \10 and friends can either, depending
7136 on context. Specifically there are two seperate switches for handling
7137 escape sequences, with the one for handling literal escapes requiring
7138 a dummy entry for all of the special escapes that are actually handled
7143 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7146 register regnode *ret = NULL;
7148 char *parse_start = RExC_parse;
7149 GET_RE_DEBUG_FLAGS_DECL;
7150 DEBUG_PARSE("atom");
7151 *flagp = WORST; /* Tentatively. */
7153 PERL_ARGS_ASSERT_REGATOM;
7156 switch ((U8)*RExC_parse) {
7158 RExC_seen_zerolen++;
7159 nextchar(pRExC_state);
7160 if (RExC_flags & RXf_PMf_MULTILINE)
7161 ret = reg_node(pRExC_state, MBOL);
7162 else if (RExC_flags & RXf_PMf_SINGLELINE)
7163 ret = reg_node(pRExC_state, SBOL);
7165 ret = reg_node(pRExC_state, BOL);
7166 Set_Node_Length(ret, 1); /* MJD */
7169 nextchar(pRExC_state);
7171 RExC_seen_zerolen++;
7172 if (RExC_flags & RXf_PMf_MULTILINE)
7173 ret = reg_node(pRExC_state, MEOL);
7174 else if (RExC_flags & RXf_PMf_SINGLELINE)
7175 ret = reg_node(pRExC_state, SEOL);
7177 ret = reg_node(pRExC_state, EOL);
7178 Set_Node_Length(ret, 1); /* MJD */
7181 nextchar(pRExC_state);
7182 if (RExC_flags & RXf_PMf_SINGLELINE)
7183 ret = reg_node(pRExC_state, SANY);
7185 ret = reg_node(pRExC_state, REG_ANY);
7186 *flagp |= HASWIDTH|SIMPLE;
7188 Set_Node_Length(ret, 1); /* MJD */
7192 char * const oregcomp_parse = ++RExC_parse;
7193 ret = regclass(pRExC_state,depth+1);
7194 if (*RExC_parse != ']') {
7195 RExC_parse = oregcomp_parse;
7196 vFAIL("Unmatched [");
7198 nextchar(pRExC_state);
7199 *flagp |= HASWIDTH|SIMPLE;
7200 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
7204 nextchar(pRExC_state);
7205 ret = reg(pRExC_state, 1, &flags,depth+1);
7207 if (flags & TRYAGAIN) {
7208 if (RExC_parse == RExC_end) {
7209 /* Make parent create an empty node if needed. */
7217 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7221 if (flags & TRYAGAIN) {
7225 vFAIL("Internal urp");
7226 /* Supposed to be caught earlier. */
7229 if (!regcurly(RExC_parse)) {
7238 vFAIL("Quantifier follows nothing");
7246 len=0; /* silence a spurious compiler warning */
7247 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
7248 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
7249 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
7250 ret = reganode(pRExC_state, FOLDCHAR, cp);
7251 Set_Node_Length(ret, 1); /* MJD */
7252 nextchar(pRExC_state); /* kill whitespace under /x */
7260 This switch handles escape sequences that resolve to some kind
7261 of special regop and not to literal text. Escape sequnces that
7262 resolve to literal text are handled below in the switch marked
7265 Every entry in this switch *must* have a corresponding entry
7266 in the literal escape switch. However, the opposite is not
7267 required, as the default for this switch is to jump to the
7268 literal text handling code.
7270 switch ((U8)*++RExC_parse) {
7275 /* Special Escapes */
7277 RExC_seen_zerolen++;
7278 ret = reg_node(pRExC_state, SBOL);
7280 goto finish_meta_pat;
7282 ret = reg_node(pRExC_state, GPOS);
7283 RExC_seen |= REG_SEEN_GPOS;
7285 goto finish_meta_pat;
7287 RExC_seen_zerolen++;
7288 ret = reg_node(pRExC_state, KEEPS);
7290 /* XXX:dmq : disabling in-place substitution seems to
7291 * be necessary here to avoid cases of memory corruption, as
7292 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7294 RExC_seen |= REG_SEEN_LOOKBEHIND;
7295 goto finish_meta_pat;
7297 ret = reg_node(pRExC_state, SEOL);
7299 RExC_seen_zerolen++; /* Do not optimize RE away */
7300 goto finish_meta_pat;
7302 ret = reg_node(pRExC_state, EOS);
7304 RExC_seen_zerolen++; /* Do not optimize RE away */
7305 goto finish_meta_pat;
7307 ret = reg_node(pRExC_state, CANY);
7308 RExC_seen |= REG_SEEN_CANY;
7309 *flagp |= HASWIDTH|SIMPLE;
7310 goto finish_meta_pat;
7312 ret = reg_node(pRExC_state, CLUMP);
7314 goto finish_meta_pat;
7317 ret = reg_node(pRExC_state, (U8)(ALNUML));
7319 ret = reg_node(pRExC_state, (U8)(ALNUM));
7320 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7322 *flagp |= HASWIDTH|SIMPLE;
7323 goto finish_meta_pat;
7326 ret = reg_node(pRExC_state, (U8)(NALNUML));
7328 ret = reg_node(pRExC_state, (U8)(NALNUM));
7329 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7331 *flagp |= HASWIDTH|SIMPLE;
7332 goto finish_meta_pat;
7334 RExC_seen_zerolen++;
7335 RExC_seen |= REG_SEEN_LOOKBEHIND;
7337 ret = reg_node(pRExC_state, (U8)(BOUNDL));
7339 ret = reg_node(pRExC_state, (U8)(BOUND));
7340 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7343 goto finish_meta_pat;
7345 RExC_seen_zerolen++;
7346 RExC_seen |= REG_SEEN_LOOKBEHIND;
7348 ret = reg_node(pRExC_state, (U8)(NBOUNDL));
7350 ret = reg_node(pRExC_state, (U8)(NBOUND));
7351 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7354 goto finish_meta_pat;
7357 ret = reg_node(pRExC_state, (U8)(SPACEL));
7359 ret = reg_node(pRExC_state, (U8)(SPACE));
7360 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7362 *flagp |= HASWIDTH|SIMPLE;
7363 goto finish_meta_pat;
7366 ret = reg_node(pRExC_state, (U8)(NSPACEL));
7368 ret = reg_node(pRExC_state, (U8)(NSPACE));
7369 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7371 *flagp |= HASWIDTH|SIMPLE;
7372 goto finish_meta_pat;
7374 ret = reg_node(pRExC_state, DIGIT);
7375 *flagp |= HASWIDTH|SIMPLE;
7376 goto finish_meta_pat;
7378 ret = reg_node(pRExC_state, NDIGIT);
7379 *flagp |= HASWIDTH|SIMPLE;
7380 goto finish_meta_pat;
7382 ret = reg_node(pRExC_state, LNBREAK);
7383 *flagp |= HASWIDTH|SIMPLE;
7384 goto finish_meta_pat;
7386 ret = reg_node(pRExC_state, HORIZWS);
7387 *flagp |= HASWIDTH|SIMPLE;
7388 goto finish_meta_pat;
7390 ret = reg_node(pRExC_state, NHORIZWS);
7391 *flagp |= HASWIDTH|SIMPLE;
7392 goto finish_meta_pat;
7394 ret = reg_node(pRExC_state, VERTWS);
7395 *flagp |= HASWIDTH|SIMPLE;
7396 goto finish_meta_pat;
7398 ret = reg_node(pRExC_state, NVERTWS);
7399 *flagp |= HASWIDTH|SIMPLE;
7401 nextchar(pRExC_state);
7402 Set_Node_Length(ret, 2); /* MJD */
7407 char* const oldregxend = RExC_end;
7409 char* parse_start = RExC_parse - 2;
7412 if (RExC_parse[1] == '{') {
7413 /* a lovely hack--pretend we saw [\pX] instead */
7414 RExC_end = strchr(RExC_parse, '}');
7416 const U8 c = (U8)*RExC_parse;
7418 RExC_end = oldregxend;
7419 vFAIL2("Missing right brace on \\%c{}", c);
7424 RExC_end = RExC_parse + 2;
7425 if (RExC_end > oldregxend)
7426 RExC_end = oldregxend;
7430 ret = regclass(pRExC_state,depth+1);
7432 RExC_end = oldregxend;
7435 Set_Node_Offset(ret, parse_start + 2);
7436 Set_Node_Cur_Length(ret);
7437 nextchar(pRExC_state);
7438 *flagp |= HASWIDTH|SIMPLE;
7442 /* Handle \N and \N{NAME} here and not below because it can be
7443 multicharacter. join_exact() will join them up later on.
7444 Also this makes sure that things like /\N{BLAH}+/ and
7445 \N{BLAH} being multi char Just Happen. dmq*/
7447 ret= reg_namedseq(pRExC_state, NULL, flagp);
7449 case 'k': /* Handle \k<NAME> and \k'NAME' */
7452 char ch= RExC_parse[1];
7453 if (ch != '<' && ch != '\'' && ch != '{') {
7455 vFAIL2("Sequence %.2s... not terminated",parse_start);
7457 /* this pretty much dupes the code for (?P=...) in reg(), if
7458 you change this make sure you change that */
7459 char* name_start = (RExC_parse += 2);
7461 SV *sv_dat = reg_scan_name(pRExC_state,
7462 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7463 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7464 if (RExC_parse == name_start || *RExC_parse != ch)
7465 vFAIL2("Sequence %.3s... not terminated",parse_start);
7468 num = add_data( pRExC_state, 1, "S" );
7469 RExC_rxi->data->data[num]=(void*)sv_dat;
7470 SvREFCNT_inc_simple_void(sv_dat);
7474 ret = reganode(pRExC_state,
7475 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7479 /* override incorrect value set in reganode MJD */
7480 Set_Node_Offset(ret, parse_start+1);
7481 Set_Node_Cur_Length(ret); /* MJD */
7482 nextchar(pRExC_state);
7488 case '1': case '2': case '3': case '4':
7489 case '5': case '6': case '7': case '8': case '9':
7492 bool isg = *RExC_parse == 'g';
7497 if (*RExC_parse == '{') {
7501 if (*RExC_parse == '-') {
7505 if (hasbrace && !isDIGIT(*RExC_parse)) {
7506 if (isrel) RExC_parse--;
7508 goto parse_named_seq;
7510 num = atoi(RExC_parse);
7511 if (isg && num == 0)
7512 vFAIL("Reference to invalid group 0");
7514 num = RExC_npar - num;
7516 vFAIL("Reference to nonexistent or unclosed group");
7518 if (!isg && num > 9 && num >= RExC_npar)
7521 char * const parse_start = RExC_parse - 1; /* MJD */
7522 while (isDIGIT(*RExC_parse))
7524 if (parse_start == RExC_parse - 1)
7525 vFAIL("Unterminated \\g... pattern");
7527 if (*RExC_parse != '}')
7528 vFAIL("Unterminated \\g{...} pattern");
7532 if (num > (I32)RExC_rx->nparens)
7533 vFAIL("Reference to nonexistent group");
7536 ret = reganode(pRExC_state,
7537 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7541 /* override incorrect value set in reganode MJD */
7542 Set_Node_Offset(ret, parse_start+1);
7543 Set_Node_Cur_Length(ret); /* MJD */
7545 nextchar(pRExC_state);
7550 if (RExC_parse >= RExC_end)
7551 FAIL("Trailing \\");
7554 /* Do not generate "unrecognized" warnings here, we fall
7555 back into the quick-grab loop below */
7562 if (RExC_flags & RXf_PMf_EXTENDED) {
7563 if ( reg_skipcomment( pRExC_state ) )
7570 register STRLEN len;
7575 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7577 parse_start = RExC_parse - 1;
7583 ret = reg_node(pRExC_state,
7584 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7586 for (len = 0, p = RExC_parse - 1;
7587 len < 127 && p < RExC_end;
7590 char * const oldp = p;
7592 if (RExC_flags & RXf_PMf_EXTENDED)
7593 p = regwhite( pRExC_state, p );
7598 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7599 goto normal_default;
7609 /* Literal Escapes Switch
7611 This switch is meant to handle escape sequences that
7612 resolve to a literal character.
7614 Every escape sequence that represents something
7615 else, like an assertion or a char class, is handled
7616 in the switch marked 'Special Escapes' above in this
7617 routine, but also has an entry here as anything that
7618 isn't explicitly mentioned here will be treated as
7619 an unescaped equivalent literal.
7623 /* These are all the special escapes. */
7627 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7628 goto normal_default;
7629 case 'A': /* Start assertion */
7630 case 'b': case 'B': /* Word-boundary assertion*/
7631 case 'C': /* Single char !DANGEROUS! */
7632 case 'd': case 'D': /* digit class */
7633 case 'g': case 'G': /* generic-backref, pos assertion */
7634 case 'h': case 'H': /* HORIZWS */
7635 case 'k': case 'K': /* named backref, keep marker */
7636 case 'N': /* named char sequence */
7637 case 'p': case 'P': /* Unicode property */
7638 case 'R': /* LNBREAK */
7639 case 's': case 'S': /* space class */
7640 case 'v': case 'V': /* VERTWS */
7641 case 'w': case 'W': /* word class */
7642 case 'X': /* eXtended Unicode "combining character sequence" */
7643 case 'z': case 'Z': /* End of line/string assertion */
7647 /* Anything after here is an escape that resolves to a
7648 literal. (Except digits, which may or may not)
7667 ender = ASCII_TO_NATIVE('\033');
7671 ender = ASCII_TO_NATIVE('\007');
7676 STRLEN brace_len = len;
7678 const char* error_msg;
7680 bool valid = grok_bslash_o(p,
7687 RExC_parse = p; /* going to die anyway; point
7688 to exact spot of failure */
7695 if (PL_encoding && ender < 0x100) {
7696 goto recode_encoding;
7705 char* const e = strchr(p, '}');
7709 vFAIL("Missing right brace on \\x{}");
7712 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7713 | PERL_SCAN_DISALLOW_PREFIX;
7714 STRLEN numlen = e - p - 1;
7715 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7722 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7724 ender = grok_hex(p, &numlen, &flags, NULL);
7727 if (PL_encoding && ender < 0x100)
7728 goto recode_encoding;
7732 ender = grok_bslash_c(*p++, SIZE_ONLY);
7734 case '0': case '1': case '2': case '3':case '4':
7735 case '5': case '6': case '7': case '8':case '9':
7737 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
7739 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
7741 ender = grok_oct(p, &numlen, &flags, NULL);
7751 if (PL_encoding && ender < 0x100)
7752 goto recode_encoding;
7756 SV* enc = PL_encoding;
7757 ender = reg_recode((const char)(U8)ender, &enc);
7758 if (!enc && SIZE_ONLY)
7759 ckWARNreg(p, "Invalid escape in the specified encoding");
7765 FAIL("Trailing \\");
7768 if (!SIZE_ONLY&& isALPHA(*p))
7769 ckWARN2reg(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7770 goto normal_default;
7775 if (UTF8_IS_START(*p) && UTF) {
7777 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7778 &numlen, UTF8_ALLOW_DEFAULT);
7785 if ( RExC_flags & RXf_PMf_EXTENDED)
7786 p = regwhite( pRExC_state, p );
7788 /* Prime the casefolded buffer. */
7789 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7791 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7796 /* Emit all the Unicode characters. */
7798 for (foldbuf = tmpbuf;
7800 foldlen -= numlen) {
7801 ender = utf8_to_uvchr(foldbuf, &numlen);
7803 const STRLEN unilen = reguni(pRExC_state, ender, s);
7806 /* In EBCDIC the numlen
7807 * and unilen can differ. */
7809 if (numlen >= foldlen)
7813 break; /* "Can't happen." */
7817 const STRLEN unilen = reguni(pRExC_state, ender, s);
7826 REGC((char)ender, s++);
7832 /* Emit all the Unicode characters. */
7834 for (foldbuf = tmpbuf;
7836 foldlen -= numlen) {
7837 ender = utf8_to_uvchr(foldbuf, &numlen);
7839 const STRLEN unilen = reguni(pRExC_state, ender, s);
7842 /* In EBCDIC the numlen
7843 * and unilen can differ. */
7845 if (numlen >= foldlen)
7853 const STRLEN unilen = reguni(pRExC_state, ender, s);
7862 REGC((char)ender, s++);
7866 Set_Node_Cur_Length(ret); /* MJD */
7867 nextchar(pRExC_state);
7869 /* len is STRLEN which is unsigned, need to copy to signed */
7872 vFAIL("Internal disaster");
7876 if (len == 1 && UNI_IS_INVARIANT(ender))
7880 RExC_size += STR_SZ(len);
7883 RExC_emit += STR_SZ(len);
7893 S_regwhite( RExC_state_t *pRExC_state, char *p )
7895 const char *e = RExC_end;
7897 PERL_ARGS_ASSERT_REGWHITE;
7902 else if (*p == '#') {
7911 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7919 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7920 Character classes ([:foo:]) can also be negated ([:^foo:]).
7921 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7922 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7923 but trigger failures because they are currently unimplemented. */
7925 #define POSIXCC_DONE(c) ((c) == ':')
7926 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7927 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7930 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7933 I32 namedclass = OOB_NAMEDCLASS;
7935 PERL_ARGS_ASSERT_REGPPOSIXCC;
7937 if (value == '[' && RExC_parse + 1 < RExC_end &&
7938 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7939 POSIXCC(UCHARAT(RExC_parse))) {
7940 const char c = UCHARAT(RExC_parse);
7941 char* const s = RExC_parse++;
7943 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7945 if (RExC_parse == RExC_end)
7946 /* Grandfather lone [:, [=, [. */
7949 const char* const t = RExC_parse++; /* skip over the c */
7952 if (UCHARAT(RExC_parse) == ']') {
7953 const char *posixcc = s + 1;
7954 RExC_parse++; /* skip over the ending ] */
7957 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7958 const I32 skip = t - posixcc;
7960 /* Initially switch on the length of the name. */
7963 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7964 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7967 /* Names all of length 5. */
7968 /* alnum alpha ascii blank cntrl digit graph lower
7969 print punct space upper */
7970 /* Offset 4 gives the best switch position. */
7971 switch (posixcc[4]) {
7973 if (memEQ(posixcc, "alph", 4)) /* alpha */
7974 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7977 if (memEQ(posixcc, "spac", 4)) /* space */
7978 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7981 if (memEQ(posixcc, "grap", 4)) /* graph */
7982 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7985 if (memEQ(posixcc, "asci", 4)) /* ascii */
7986 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7989 if (memEQ(posixcc, "blan", 4)) /* blank */
7990 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7993 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7994 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7997 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7998 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
8001 if (memEQ(posixcc, "lowe", 4)) /* lower */
8002 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
8003 else if (memEQ(posixcc, "uppe", 4)) /* upper */
8004 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
8007 if (memEQ(posixcc, "digi", 4)) /* digit */
8008 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
8009 else if (memEQ(posixcc, "prin", 4)) /* print */
8010 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
8011 else if (memEQ(posixcc, "punc", 4)) /* punct */
8012 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
8017 if (memEQ(posixcc, "xdigit", 6))
8018 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
8022 if (namedclass == OOB_NAMEDCLASS)
8023 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
8025 assert (posixcc[skip] == ':');
8026 assert (posixcc[skip+1] == ']');
8027 } else if (!SIZE_ONLY) {
8028 /* [[=foo=]] and [[.foo.]] are still future. */
8030 /* adjust RExC_parse so the warning shows after
8032 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
8034 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
8037 /* Maternal grandfather:
8038 * "[:" ending in ":" but not in ":]" */
8048 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
8052 PERL_ARGS_ASSERT_CHECKPOSIXCC;
8054 if (POSIXCC(UCHARAT(RExC_parse))) {
8055 const char *s = RExC_parse;
8056 const char c = *s++;
8060 if (*s && c == *s && s[1] == ']') {
8062 "POSIX syntax [%c %c] belongs inside character classes",
8065 /* [[=foo=]] and [[.foo.]] are still future. */
8066 if (POSIXCC_NOTYET(c)) {
8067 /* adjust RExC_parse so the error shows after
8069 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
8071 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
8077 /* No locale test */
8078 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
8080 for (value = 0; value < 256; value++) \
8082 ANYOF_BITMAP_SET(ret, value); \
8086 case ANYOF_N##NAME: \
8087 for (value = 0; value < 256; value++) \
8089 ANYOF_BITMAP_SET(ret, value); \
8094 /* Like the above, but there are differences if we are in uni-8-bit or not, so
8095 * there are two tests passed in, to use depending on that. There aren't any
8096 * cases where the label is different from the name, so no need for that
8098 #define _C_C_T_(NAME,TEST_8,TEST_7,WORD) \
8100 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
8101 else if (UNI_SEMANTICS) { \
8102 for (value = 0; value < 256; value++) { \
8103 if (TEST_8) ANYOF_BITMAP_SET(ret, value); \
8107 for (value = 0; value < 256; value++) { \
8108 if (TEST_7) ANYOF_BITMAP_SET(ret, value); \
8114 case ANYOF_N##NAME: \
8115 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
8116 else if (UNI_SEMANTICS) { \
8117 for (value = 0; value < 256; value++) { \
8118 if (! TEST_8) ANYOF_BITMAP_SET(ret, value); \
8122 for (value = 0; value < 256; value++) { \
8123 if (! TEST_7) ANYOF_BITMAP_SET(ret, value); \
8131 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
8132 so that it is possible to override the option here without having to
8133 rebuild the entire core. as we are required to do if we change regcomp.h
8134 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
8136 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
8137 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
8140 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8141 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
8143 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
8147 parse a class specification and produce either an ANYOF node that
8148 matches the pattern or if the pattern matches a single char only and
8149 that char is < 256 and we are case insensitive then we produce an
8154 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
8157 register UV nextvalue;
8158 register IV prevvalue = OOB_UNICODE;
8159 register IV range = 0;
8160 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
8161 register regnode *ret;
8164 char *rangebegin = NULL;
8165 bool need_class = 0;
8168 bool optimize_invert = TRUE;
8169 AV* unicode_alternate = NULL;
8171 UV literal_endpoint = 0;
8173 UV stored = 0; /* 0, 1, or more than 1 chars stored in the class */
8175 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
8176 case we need to change the emitted regop to an EXACT. */
8177 const char * orig_parse = RExC_parse;
8178 GET_RE_DEBUG_FLAGS_DECL;
8180 PERL_ARGS_ASSERT_REGCLASS;
8182 PERL_UNUSED_ARG(depth);
8185 DEBUG_PARSE("clas");
8187 /* Assume we are going to generate an ANYOF node. */
8188 ret = reganode(pRExC_state, ANYOF, 0);
8191 ANYOF_FLAGS(ret) = 0;
8193 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
8197 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
8201 RExC_size += ANYOF_SKIP;
8202 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
8205 RExC_emit += ANYOF_SKIP;
8207 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
8209 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
8210 ANYOF_BITMAP_ZERO(ret);
8211 listsv = newSVpvs("# comment\n");
8214 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8216 if (!SIZE_ONLY && POSIXCC(nextvalue))
8217 checkposixcc(pRExC_state);
8219 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
8220 if (UCHARAT(RExC_parse) == ']')
8224 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
8228 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
8231 rangebegin = RExC_parse;
8233 value = utf8n_to_uvchr((U8*)RExC_parse,
8234 RExC_end - RExC_parse,
8235 &numlen, UTF8_ALLOW_DEFAULT);
8236 RExC_parse += numlen;
8239 value = UCHARAT(RExC_parse++);
8241 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8242 if (value == '[' && POSIXCC(nextvalue))
8243 namedclass = regpposixcc(pRExC_state, value);
8244 else if (value == '\\') {
8246 value = utf8n_to_uvchr((U8*)RExC_parse,
8247 RExC_end - RExC_parse,
8248 &numlen, UTF8_ALLOW_DEFAULT);
8249 RExC_parse += numlen;
8252 value = UCHARAT(RExC_parse++);
8253 /* Some compilers cannot handle switching on 64-bit integer
8254 * values, therefore value cannot be an UV. Yes, this will
8255 * be a problem later if we want switch on Unicode.
8256 * A similar issue a little bit later when switching on
8257 * namedclass. --jhi */
8258 switch ((I32)value) {
8259 case 'w': namedclass = ANYOF_ALNUM; break;
8260 case 'W': namedclass = ANYOF_NALNUM; break;
8261 case 's': namedclass = ANYOF_SPACE; break;
8262 case 'S': namedclass = ANYOF_NSPACE; break;
8263 case 'd': namedclass = ANYOF_DIGIT; break;
8264 case 'D': namedclass = ANYOF_NDIGIT; break;
8265 case 'v': namedclass = ANYOF_VERTWS; break;
8266 case 'V': namedclass = ANYOF_NVERTWS; break;
8267 case 'h': namedclass = ANYOF_HORIZWS; break;
8268 case 'H': namedclass = ANYOF_NHORIZWS; break;
8269 case 'N': /* Handle \N{NAME} in class */
8271 /* We only pay attention to the first char of
8272 multichar strings being returned. I kinda wonder
8273 if this makes sense as it does change the behaviour
8274 from earlier versions, OTOH that behaviour was broken
8276 UV v; /* value is register so we cant & it /grrr */
8277 if (reg_namedseq(pRExC_state, &v, NULL)) {
8287 if (RExC_parse >= RExC_end)
8288 vFAIL2("Empty \\%c{}", (U8)value);
8289 if (*RExC_parse == '{') {
8290 const U8 c = (U8)value;
8291 e = strchr(RExC_parse++, '}');
8293 vFAIL2("Missing right brace on \\%c{}", c);
8294 while (isSPACE(UCHARAT(RExC_parse)))
8296 if (e == RExC_parse)
8297 vFAIL2("Empty \\%c{}", c);
8299 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
8307 if (UCHARAT(RExC_parse) == '^') {
8310 value = value == 'p' ? 'P' : 'p'; /* toggle */
8311 while (isSPACE(UCHARAT(RExC_parse))) {
8316 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
8317 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
8320 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8321 namedclass = ANYOF_MAX; /* no official name, but it's named */
8324 case 'n': value = '\n'; break;
8325 case 'r': value = '\r'; break;
8326 case 't': value = '\t'; break;
8327 case 'f': value = '\f'; break;
8328 case 'b': value = '\b'; break;
8329 case 'e': value = ASCII_TO_NATIVE('\033');break;
8330 case 'a': value = ASCII_TO_NATIVE('\007');break;
8332 RExC_parse--; /* function expects to be pointed at the 'o' */
8334 const char* error_msg;
8335 bool valid = grok_bslash_o(RExC_parse,
8340 RExC_parse += numlen;
8345 if (PL_encoding && value < 0x100) {
8346 goto recode_encoding;
8350 if (*RExC_parse == '{') {
8351 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8352 | PERL_SCAN_DISALLOW_PREFIX;
8353 char * const e = strchr(RExC_parse++, '}');
8355 vFAIL("Missing right brace on \\x{}");
8357 numlen = e - RExC_parse;
8358 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8362 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8364 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8365 RExC_parse += numlen;
8367 if (PL_encoding && value < 0x100)
8368 goto recode_encoding;
8371 value = grok_bslash_c(*RExC_parse++, SIZE_ONLY);
8373 case '0': case '1': case '2': case '3': case '4':
8374 case '5': case '6': case '7':
8376 /* Take 1-3 octal digits */
8377 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8379 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8380 RExC_parse += numlen;
8381 if (PL_encoding && value < 0x100)
8382 goto recode_encoding;
8387 SV* enc = PL_encoding;
8388 value = reg_recode((const char)(U8)value, &enc);
8389 if (!enc && SIZE_ONLY)
8390 ckWARNreg(RExC_parse,
8391 "Invalid escape in the specified encoding");
8395 /* Allow \_ to not give an error */
8396 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
8397 ckWARN2reg(RExC_parse,
8398 "Unrecognized escape \\%c in character class passed through",
8403 } /* end of \blah */
8409 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8411 /* What matches in a locale is not known until runtime, so need to
8412 * (one time per class) allocate extra space to pass to regexec.
8413 * The space will contain a bit for each named class that is to be
8414 * matched against. This isn't needed for \p{} and pseudo-classes,
8415 * as they are not affected by locale, and hence are dealt with
8417 if (LOC && namedclass < ANYOF_MAX && ! need_class) {
8420 RExC_size += ANYOF_CLASS_ADD_SKIP;
8423 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8424 ANYOF_CLASS_ZERO(ret);
8426 ANYOF_FLAGS(ret) |= ANYOF_CLASS|ANYOF_LARGE;
8429 /* a bad range like a-\d, a-[:digit:] ? */
8433 RExC_parse >= rangebegin ?
8434 RExC_parse - rangebegin : 0;
8435 ckWARN4reg(RExC_parse,
8436 "False [] range \"%*.*s\"",
8439 if (prevvalue < 256) {
8440 ANYOF_BITMAP_SET(ret, prevvalue);
8441 ANYOF_BITMAP_SET(ret, '-');
8444 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8445 Perl_sv_catpvf(aTHX_ listsv,
8446 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8450 range = 0; /* this was not a true range */
8456 const char *what = NULL;
8459 if (namedclass > OOB_NAMEDCLASS)
8460 optimize_invert = FALSE;
8461 /* Possible truncation here but in some 64-bit environments
8462 * the compiler gets heartburn about switch on 64-bit values.
8463 * A similar issue a little earlier when switching on value.
8465 switch ((I32)namedclass) {
8467 case _C_C_T_(ALNUMC, isALNUMC_L1(value), isALNUMC(value), "XPosixAlnum");
8468 case _C_C_T_(ALPHA, isALPHA_L1(value), isALPHA(value), "XPosixAlpha");
8469 case _C_C_T_(BLANK, isBLANK_L1(value), isBLANK(value), "XPosixBlank");
8470 case _C_C_T_(CNTRL, isCNTRL_L1(value), isCNTRL(value), "XPosixCntrl");
8471 case _C_C_T_(GRAPH, isGRAPH_L1(value), isGRAPH(value), "XPosixGraph");
8472 case _C_C_T_(LOWER, isLOWER_L1(value), isLOWER(value), "XPosixLower");
8473 case _C_C_T_(PRINT, isPRINT_L1(value), isPRINT(value), "XPosixPrint");
8474 case _C_C_T_(PSXSPC, isPSXSPC_L1(value), isPSXSPC(value), "XPosixSpace");
8475 case _C_C_T_(PUNCT, isPUNCT_L1(value), isPUNCT(value), "XPosixPunct");
8476 case _C_C_T_(UPPER, isUPPER_L1(value), isUPPER(value), "XPosixUpper");
8477 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8478 /* \s, \w match all unicode if utf8. */
8479 case _C_C_T_(SPACE, isSPACE_L1(value), isSPACE(value), "SpacePerl");
8480 case _C_C_T_(ALNUM, isWORDCHAR_L1(value), isALNUM(value), "Word");
8482 /* \s, \w match ascii and locale only */
8483 case _C_C_T_(SPACE, isSPACE_L1(value), isSPACE(value), "PerlSpace");
8484 case _C_C_T_(ALNUM, isWORDCHAR_L1(value), isALNUM(value), "PerlWord");
8486 case _C_C_T_(XDIGIT, isXDIGIT_L1(value), isXDIGIT(value), "XPosixXDigit");
8487 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8488 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8491 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8494 for (value = 0; value < 128; value++)
8495 ANYOF_BITMAP_SET(ret, value);
8497 for (value = 0; value < 256; value++) {
8499 ANYOF_BITMAP_SET(ret, value);
8508 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8511 for (value = 128; value < 256; value++)
8512 ANYOF_BITMAP_SET(ret, value);
8514 for (value = 0; value < 256; value++) {
8515 if (!isASCII(value))
8516 ANYOF_BITMAP_SET(ret, value);
8525 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8527 /* consecutive digits assumed */
8528 for (value = '0'; value <= '9'; value++)
8529 ANYOF_BITMAP_SET(ret, value);
8532 what = POSIX_CC_UNI_NAME("Digit");
8536 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8538 /* consecutive digits assumed */
8539 for (value = 0; value < '0'; value++)
8540 ANYOF_BITMAP_SET(ret, value);
8541 for (value = '9' + 1; value < 256; value++)
8542 ANYOF_BITMAP_SET(ret, value);
8545 what = POSIX_CC_UNI_NAME("Digit");
8548 /* this is to handle \p and \P */
8551 vFAIL("Invalid [::] class");
8555 /* Strings such as "+utf8::isWord\n" */
8556 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8558 stored+=2; /* can't optimize this class */
8561 } /* end of namedclass \blah */
8564 if (prevvalue > (IV)value) /* b-a */ {
8565 const int w = RExC_parse - rangebegin;
8566 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8567 range = 0; /* not a valid range */
8571 prevvalue = value; /* save the beginning of the range */
8572 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8573 RExC_parse[1] != ']') {
8576 /* a bad range like \w-, [:word:]- ? */
8577 if (namedclass > OOB_NAMEDCLASS) {
8578 if (ckWARN(WARN_REGEXP)) {
8580 RExC_parse >= rangebegin ?
8581 RExC_parse - rangebegin : 0;
8583 "False [] range \"%*.*s\"",
8587 ANYOF_BITMAP_SET(ret, '-');
8589 range = 1; /* yeah, it's a range! */
8590 continue; /* but do it the next time */
8594 /* now is the next time */
8595 /*stored += (value - prevvalue + 1);*/
8597 if (prevvalue < 256) {
8598 const IV ceilvalue = value < 256 ? value : 255;
8601 /* In EBCDIC [\x89-\x91] should include
8602 * the \x8e but [i-j] should not. */
8603 if (literal_endpoint == 2 &&
8604 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8605 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8607 if (isLOWER(prevvalue)) {
8608 for (i = prevvalue; i <= ceilvalue; i++)
8609 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8611 ANYOF_BITMAP_SET(ret, i);
8614 for (i = prevvalue; i <= ceilvalue; i++)
8615 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8617 ANYOF_BITMAP_SET(ret, i);
8623 for (i = prevvalue; i <= ceilvalue; i++) {
8624 if (!ANYOF_BITMAP_TEST(ret,i)) {
8626 ANYOF_BITMAP_SET(ret, i);
8630 if (value > 255 || UTF) {
8631 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8632 const UV natvalue = NATIVE_TO_UNI(value);
8633 stored+=2; /* can't optimize this class */
8634 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8635 if (prevnatvalue < natvalue) { /* what about > ? */
8636 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8637 prevnatvalue, natvalue);
8639 else if (prevnatvalue == natvalue) {
8640 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8642 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8644 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8646 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8647 if (RExC_precomp[0] == ':' &&
8648 RExC_precomp[1] == '[' &&
8649 (f == 0xDF || f == 0x92)) {
8650 f = NATIVE_TO_UNI(f);
8653 /* If folding and foldable and a single
8654 * character, insert also the folded version
8655 * to the charclass. */
8657 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8658 if ((RExC_precomp[0] == ':' &&
8659 RExC_precomp[1] == '[' &&
8661 (value == 0xFB05 || value == 0xFB06))) ?
8662 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8663 foldlen == (STRLEN)UNISKIP(f) )
8665 if (foldlen == (STRLEN)UNISKIP(f))
8667 Perl_sv_catpvf(aTHX_ listsv,
8670 /* Any multicharacter foldings
8671 * require the following transform:
8672 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8673 * where E folds into "pq" and F folds
8674 * into "rst", all other characters
8675 * fold to single characters. We save
8676 * away these multicharacter foldings,
8677 * to be later saved as part of the
8678 * additional "s" data. */
8681 if (!unicode_alternate)
8682 unicode_alternate = newAV();
8683 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8685 av_push(unicode_alternate, sv);
8689 /* If folding and the value is one of the Greek
8690 * sigmas insert a few more sigmas to make the
8691 * folding rules of the sigmas to work right.
8692 * Note that not all the possible combinations
8693 * are handled here: some of them are handled
8694 * by the standard folding rules, and some of
8695 * them (literal or EXACTF cases) are handled
8696 * during runtime in regexec.c:S_find_byclass(). */
8697 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8698 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8699 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8700 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8701 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8703 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8704 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8705 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8710 literal_endpoint = 0;
8714 range = 0; /* this range (if it was one) is done now */
8721 /****** !SIZE_ONLY AFTER HERE *********/
8723 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8724 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8726 /* optimize single char class to an EXACT node
8727 but *only* when its not a UTF/high char */
8728 const char * cur_parse= RExC_parse;
8729 RExC_emit = (regnode *)orig_emit;
8730 RExC_parse = (char *)orig_parse;
8731 ret = reg_node(pRExC_state,
8732 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8733 RExC_parse = (char *)cur_parse;
8734 *STRING(ret)= (char)value;
8736 RExC_emit += STR_SZ(1);
8737 SvREFCNT_dec(listsv);
8740 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8741 if ( /* If the only flag is folding (plus possibly inversion). */
8742 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8744 for (value = 0; value < 256; ++value) {
8745 if (ANYOF_BITMAP_TEST(ret, value)) {
8746 UV fold = PL_fold[value];
8749 ANYOF_BITMAP_SET(ret, fold);
8752 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8755 /* optimize inverted simple patterns (e.g. [^a-z]) */
8756 if (optimize_invert &&
8757 /* If the only flag is inversion. */
8758 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8759 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8760 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8761 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8764 AV * const av = newAV();
8766 /* The 0th element stores the character class description
8767 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8768 * to initialize the appropriate swash (which gets stored in
8769 * the 1st element), and also useful for dumping the regnode.
8770 * The 2nd element stores the multicharacter foldings,
8771 * used later (regexec.c:S_reginclass()). */
8772 av_store(av, 0, listsv);
8773 av_store(av, 1, NULL);
8774 av_store(av, 2, MUTABLE_SV(unicode_alternate));
8775 rv = newRV_noinc(MUTABLE_SV(av));
8776 n = add_data(pRExC_state, 1, "s");
8777 RExC_rxi->data->data[n] = (void*)rv;
8785 /* reg_skipcomment()
8787 Absorbs an /x style # comments from the input stream.
8788 Returns true if there is more text remaining in the stream.
8789 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8790 terminates the pattern without including a newline.
8792 Note its the callers responsibility to ensure that we are
8798 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8802 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8804 while (RExC_parse < RExC_end)
8805 if (*RExC_parse++ == '\n') {
8810 /* we ran off the end of the pattern without ending
8811 the comment, so we have to add an \n when wrapping */
8812 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8820 Advances the parse position, and optionally absorbs
8821 "whitespace" from the inputstream.
8823 Without /x "whitespace" means (?#...) style comments only,
8824 with /x this means (?#...) and # comments and whitespace proper.
8826 Returns the RExC_parse point from BEFORE the scan occurs.
8828 This is the /x friendly way of saying RExC_parse++.
8832 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8834 char* const retval = RExC_parse++;
8836 PERL_ARGS_ASSERT_NEXTCHAR;
8839 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8840 RExC_parse[2] == '#') {
8841 while (*RExC_parse != ')') {
8842 if (RExC_parse == RExC_end)
8843 FAIL("Sequence (?#... not terminated");
8849 if (RExC_flags & RXf_PMf_EXTENDED) {
8850 if (isSPACE(*RExC_parse)) {
8854 else if (*RExC_parse == '#') {
8855 if ( reg_skipcomment( pRExC_state ) )
8864 - reg_node - emit a node
8866 STATIC regnode * /* Location. */
8867 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8870 register regnode *ptr;
8871 regnode * const ret = RExC_emit;
8872 GET_RE_DEBUG_FLAGS_DECL;
8874 PERL_ARGS_ASSERT_REG_NODE;
8877 SIZE_ALIGN(RExC_size);
8881 if (RExC_emit >= RExC_emit_bound)
8882 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8884 NODE_ALIGN_FILL(ret);
8886 FILL_ADVANCE_NODE(ptr, op);
8887 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 1);
8888 #ifdef RE_TRACK_PATTERN_OFFSETS
8889 if (RExC_offsets) { /* MJD */
8890 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8891 "reg_node", __LINE__,
8893 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8894 ? "Overwriting end of array!\n" : "OK",
8895 (UV)(RExC_emit - RExC_emit_start),
8896 (UV)(RExC_parse - RExC_start),
8897 (UV)RExC_offsets[0]));
8898 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8906 - reganode - emit a node with an argument
8908 STATIC regnode * /* Location. */
8909 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8912 register regnode *ptr;
8913 regnode * const ret = RExC_emit;
8914 GET_RE_DEBUG_FLAGS_DECL;
8916 PERL_ARGS_ASSERT_REGANODE;
8919 SIZE_ALIGN(RExC_size);
8924 assert(2==regarglen[op]+1);
8926 Anything larger than this has to allocate the extra amount.
8927 If we changed this to be:
8929 RExC_size += (1 + regarglen[op]);
8931 then it wouldn't matter. Its not clear what side effect
8932 might come from that so its not done so far.
8937 if (RExC_emit >= RExC_emit_bound)
8938 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8940 NODE_ALIGN_FILL(ret);
8942 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8943 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 2);
8944 #ifdef RE_TRACK_PATTERN_OFFSETS
8945 if (RExC_offsets) { /* MJD */
8946 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8950 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8951 "Overwriting end of array!\n" : "OK",
8952 (UV)(RExC_emit - RExC_emit_start),
8953 (UV)(RExC_parse - RExC_start),
8954 (UV)RExC_offsets[0]));
8955 Set_Cur_Node_Offset;
8963 - reguni - emit (if appropriate) a Unicode character
8966 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8970 PERL_ARGS_ASSERT_REGUNI;
8972 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8976 - reginsert - insert an operator in front of already-emitted operand
8978 * Means relocating the operand.
8981 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8984 register regnode *src;
8985 register regnode *dst;
8986 register regnode *place;
8987 const int offset = regarglen[(U8)op];
8988 const int size = NODE_STEP_REGNODE + offset;
8989 GET_RE_DEBUG_FLAGS_DECL;
8991 PERL_ARGS_ASSERT_REGINSERT;
8992 PERL_UNUSED_ARG(depth);
8993 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8994 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
9003 if (RExC_open_parens) {
9005 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
9006 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
9007 if ( RExC_open_parens[paren] >= opnd ) {
9008 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
9009 RExC_open_parens[paren] += size;
9011 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
9013 if ( RExC_close_parens[paren] >= opnd ) {
9014 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
9015 RExC_close_parens[paren] += size;
9017 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
9022 while (src > opnd) {
9023 StructCopy(--src, --dst, regnode);
9024 #ifdef RE_TRACK_PATTERN_OFFSETS
9025 if (RExC_offsets) { /* MJD 20010112 */
9026 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
9030 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
9031 ? "Overwriting end of array!\n" : "OK",
9032 (UV)(src - RExC_emit_start),
9033 (UV)(dst - RExC_emit_start),
9034 (UV)RExC_offsets[0]));
9035 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
9036 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
9042 place = opnd; /* Op node, where operand used to be. */
9043 #ifdef RE_TRACK_PATTERN_OFFSETS
9044 if (RExC_offsets) { /* MJD */
9045 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
9049 (UV)(place - RExC_emit_start) > RExC_offsets[0]
9050 ? "Overwriting end of array!\n" : "OK",
9051 (UV)(place - RExC_emit_start),
9052 (UV)(RExC_parse - RExC_start),
9053 (UV)RExC_offsets[0]));
9054 Set_Node_Offset(place, RExC_parse);
9055 Set_Node_Length(place, 1);
9058 src = NEXTOPER(place);
9059 FILL_ADVANCE_NODE(place, op);
9060 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (place) - 1);
9061 Zero(src, offset, regnode);
9065 - regtail - set the next-pointer at the end of a node chain of p to val.
9066 - SEE ALSO: regtail_study
9068 /* TODO: All three parms should be const */
9070 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
9073 register regnode *scan;
9074 GET_RE_DEBUG_FLAGS_DECL;
9076 PERL_ARGS_ASSERT_REGTAIL;
9078 PERL_UNUSED_ARG(depth);
9084 /* Find last node. */
9087 regnode * const temp = regnext(scan);
9089 SV * const mysv=sv_newmortal();
9090 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
9091 regprop(RExC_rx, mysv, scan);
9092 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
9093 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
9094 (temp == NULL ? "->" : ""),
9095 (temp == NULL ? PL_reg_name[OP(val)] : "")
9103 if (reg_off_by_arg[OP(scan)]) {
9104 ARG_SET(scan, val - scan);
9107 NEXT_OFF(scan) = val - scan;
9113 - regtail_study - set the next-pointer at the end of a node chain of p to val.
9114 - Look for optimizable sequences at the same time.
9115 - currently only looks for EXACT chains.
9117 This is expermental code. The idea is to use this routine to perform
9118 in place optimizations on branches and groups as they are constructed,
9119 with the long term intention of removing optimization from study_chunk so
9120 that it is purely analytical.
9122 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
9123 to control which is which.
9126 /* TODO: All four parms should be const */
9129 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
9132 register regnode *scan;
9134 #ifdef EXPERIMENTAL_INPLACESCAN
9137 GET_RE_DEBUG_FLAGS_DECL;
9139 PERL_ARGS_ASSERT_REGTAIL_STUDY;
9145 /* Find last node. */
9149 regnode * const temp = regnext(scan);
9150 #ifdef EXPERIMENTAL_INPLACESCAN
9151 if (PL_regkind[OP(scan)] == EXACT)
9152 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
9160 if( exact == PSEUDO )
9162 else if ( exact != OP(scan) )
9171 SV * const mysv=sv_newmortal();
9172 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
9173 regprop(RExC_rx, mysv, scan);
9174 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
9175 SvPV_nolen_const(mysv),
9177 PL_reg_name[exact]);
9184 SV * const mysv_val=sv_newmortal();
9185 DEBUG_PARSE_MSG("");
9186 regprop(RExC_rx, mysv_val, val);
9187 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
9188 SvPV_nolen_const(mysv_val),
9189 (IV)REG_NODE_NUM(val),
9193 if (reg_off_by_arg[OP(scan)]) {
9194 ARG_SET(scan, val - scan);
9197 NEXT_OFF(scan) = val - scan;
9205 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
9209 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
9214 for (bit=0; bit<32; bit++) {
9215 if (flags & (1<<bit)) {
9217 PerlIO_printf(Perl_debug_log, "%s",lead);
9218 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
9223 PerlIO_printf(Perl_debug_log, "\n");
9225 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
9231 Perl_regdump(pTHX_ const regexp *r)
9235 SV * const sv = sv_newmortal();
9236 SV *dsv= sv_newmortal();
9238 GET_RE_DEBUG_FLAGS_DECL;
9240 PERL_ARGS_ASSERT_REGDUMP;
9242 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
9244 /* Header fields of interest. */
9245 if (r->anchored_substr) {
9246 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
9247 RE_SV_DUMPLEN(r->anchored_substr), 30);
9248 PerlIO_printf(Perl_debug_log,
9249 "anchored %s%s at %"IVdf" ",
9250 s, RE_SV_TAIL(r->anchored_substr),
9251 (IV)r->anchored_offset);
9252 } else if (r->anchored_utf8) {
9253 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
9254 RE_SV_DUMPLEN(r->anchored_utf8), 30);
9255 PerlIO_printf(Perl_debug_log,
9256 "anchored utf8 %s%s at %"IVdf" ",
9257 s, RE_SV_TAIL(r->anchored_utf8),
9258 (IV)r->anchored_offset);
9260 if (r->float_substr) {
9261 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
9262 RE_SV_DUMPLEN(r->float_substr), 30);
9263 PerlIO_printf(Perl_debug_log,
9264 "floating %s%s at %"IVdf"..%"UVuf" ",
9265 s, RE_SV_TAIL(r->float_substr),
9266 (IV)r->float_min_offset, (UV)r->float_max_offset);
9267 } else if (r->float_utf8) {
9268 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
9269 RE_SV_DUMPLEN(r->float_utf8), 30);
9270 PerlIO_printf(Perl_debug_log,
9271 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
9272 s, RE_SV_TAIL(r->float_utf8),
9273 (IV)r->float_min_offset, (UV)r->float_max_offset);
9275 if (r->check_substr || r->check_utf8)
9276 PerlIO_printf(Perl_debug_log,
9278 (r->check_substr == r->float_substr
9279 && r->check_utf8 == r->float_utf8
9280 ? "(checking floating" : "(checking anchored"));
9281 if (r->extflags & RXf_NOSCAN)
9282 PerlIO_printf(Perl_debug_log, " noscan");
9283 if (r->extflags & RXf_CHECK_ALL)
9284 PerlIO_printf(Perl_debug_log, " isall");
9285 if (r->check_substr || r->check_utf8)
9286 PerlIO_printf(Perl_debug_log, ") ");
9288 if (ri->regstclass) {
9289 regprop(r, sv, ri->regstclass);
9290 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
9292 if (r->extflags & RXf_ANCH) {
9293 PerlIO_printf(Perl_debug_log, "anchored");
9294 if (r->extflags & RXf_ANCH_BOL)
9295 PerlIO_printf(Perl_debug_log, "(BOL)");
9296 if (r->extflags & RXf_ANCH_MBOL)
9297 PerlIO_printf(Perl_debug_log, "(MBOL)");
9298 if (r->extflags & RXf_ANCH_SBOL)
9299 PerlIO_printf(Perl_debug_log, "(SBOL)");
9300 if (r->extflags & RXf_ANCH_GPOS)
9301 PerlIO_printf(Perl_debug_log, "(GPOS)");
9302 PerlIO_putc(Perl_debug_log, ' ');
9304 if (r->extflags & RXf_GPOS_SEEN)
9305 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
9306 if (r->intflags & PREGf_SKIP)
9307 PerlIO_printf(Perl_debug_log, "plus ");
9308 if (r->intflags & PREGf_IMPLICIT)
9309 PerlIO_printf(Perl_debug_log, "implicit ");
9310 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
9311 if (r->extflags & RXf_EVAL_SEEN)
9312 PerlIO_printf(Perl_debug_log, "with eval ");
9313 PerlIO_printf(Perl_debug_log, "\n");
9314 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
9316 PERL_ARGS_ASSERT_REGDUMP;
9317 PERL_UNUSED_CONTEXT;
9319 #endif /* DEBUGGING */
9323 - regprop - printable representation of opcode
9325 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9328 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9329 if (flags & ANYOF_INVERT) \
9330 /*make sure the invert info is in each */ \
9331 sv_catpvs(sv, "^"); \
9337 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9342 RXi_GET_DECL(prog,progi);
9343 GET_RE_DEBUG_FLAGS_DECL;
9345 PERL_ARGS_ASSERT_REGPROP;
9349 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9350 /* It would be nice to FAIL() here, but this may be called from
9351 regexec.c, and it would be hard to supply pRExC_state. */
9352 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9353 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9355 k = PL_regkind[OP(o)];
9359 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9360 * is a crude hack but it may be the best for now since
9361 * we have no flag "this EXACTish node was UTF-8"
9363 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9364 PERL_PV_ESCAPE_UNI_DETECT |
9365 PERL_PV_PRETTY_ELLIPSES |
9366 PERL_PV_PRETTY_LTGT |
9367 PERL_PV_PRETTY_NOCLEAR
9369 } else if (k == TRIE) {
9370 /* print the details of the trie in dumpuntil instead, as
9371 * progi->data isn't available here */
9372 const char op = OP(o);
9373 const U32 n = ARG(o);
9374 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9375 (reg_ac_data *)progi->data->data[n] :
9377 const reg_trie_data * const trie
9378 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9380 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9381 DEBUG_TRIE_COMPILE_r(
9382 Perl_sv_catpvf(aTHX_ sv,
9383 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9384 (UV)trie->startstate,
9385 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9386 (UV)trie->wordcount,
9389 (UV)TRIE_CHARCOUNT(trie),
9390 (UV)trie->uniquecharcount
9393 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9395 int rangestart = -1;
9396 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9398 for (i = 0; i <= 256; i++) {
9399 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9400 if (rangestart == -1)
9402 } else if (rangestart != -1) {
9403 if (i <= rangestart + 3)
9404 for (; rangestart < i; rangestart++)
9405 put_byte(sv, rangestart);
9407 put_byte(sv, rangestart);
9409 put_byte(sv, i - 1);
9417 } else if (k == CURLY) {
9418 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9419 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9420 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9422 else if (k == WHILEM && o->flags) /* Ordinal/of */
9423 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9424 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9425 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9426 if ( RXp_PAREN_NAMES(prog) ) {
9427 if ( k != REF || OP(o) < NREF) {
9428 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9429 SV **name= av_fetch(list, ARG(o), 0 );
9431 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9434 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9435 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9436 I32 *nums=(I32*)SvPVX(sv_dat);
9437 SV **name= av_fetch(list, nums[0], 0 );
9440 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9441 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9442 (n ? "," : ""), (IV)nums[n]);
9444 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9448 } else if (k == GOSUB)
9449 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9450 else if (k == VERB) {
9452 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9453 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9454 } else if (k == LOGICAL)
9455 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9456 else if (k == FOLDCHAR)
9457 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9458 else if (k == ANYOF) {
9459 int i, rangestart = -1;
9460 const U8 flags = ANYOF_FLAGS(o);
9463 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9464 static const char * const anyofs[] = {
9497 if (flags & ANYOF_LOCALE)
9498 sv_catpvs(sv, "{loc}");
9499 if (flags & ANYOF_FOLD)
9500 sv_catpvs(sv, "{i}");
9501 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9502 if (flags & ANYOF_INVERT)
9505 /* output what the standard cp 0-255 bitmap matches */
9506 for (i = 0; i <= 256; i++) {
9507 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9508 if (rangestart == -1)
9510 } else if (rangestart != -1) {
9511 if (i <= rangestart + 3)
9512 for (; rangestart < i; rangestart++)
9513 put_byte(sv, rangestart);
9515 put_byte(sv, rangestart);
9517 put_byte(sv, i - 1);
9524 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9525 /* output any special charclass tests (used mostly under use locale) */
9526 if (o->flags & ANYOF_CLASS)
9527 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9528 if (ANYOF_CLASS_TEST(o,i)) {
9529 sv_catpv(sv, anyofs[i]);
9533 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9535 /* output information about the unicode matching */
9536 if (flags & ANYOF_UNICODE)
9537 sv_catpvs(sv, "{unicode}");
9538 else if (flags & ANYOF_UNICODE_ALL)
9539 sv_catpvs(sv, "{unicode_all}");
9543 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9547 U8 s[UTF8_MAXBYTES_CASE+1];
9549 for (i = 0; i <= 256; i++) { /* just the first 256 */
9550 uvchr_to_utf8(s, i);
9552 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9553 if (rangestart == -1)
9555 } else if (rangestart != -1) {
9556 if (i <= rangestart + 3)
9557 for (; rangestart < i; rangestart++) {
9558 const U8 * const e = uvchr_to_utf8(s,rangestart);
9560 for(p = s; p < e; p++)
9564 const U8 *e = uvchr_to_utf8(s,rangestart);
9566 for (p = s; p < e; p++)
9569 e = uvchr_to_utf8(s, i-1);
9570 for (p = s; p < e; p++)
9577 sv_catpvs(sv, "..."); /* et cetera */
9581 char *s = savesvpv(lv);
9582 char * const origs = s;
9584 while (*s && *s != '\n')
9588 const char * const t = ++s;
9606 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9608 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9609 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9611 PERL_UNUSED_CONTEXT;
9612 PERL_UNUSED_ARG(sv);
9614 PERL_UNUSED_ARG(prog);
9615 #endif /* DEBUGGING */
9619 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9620 { /* Assume that RE_INTUIT is set */
9622 struct regexp *const prog = (struct regexp *)SvANY(r);
9623 GET_RE_DEBUG_FLAGS_DECL;
9625 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9626 PERL_UNUSED_CONTEXT;
9630 const char * const s = SvPV_nolen_const(prog->check_substr
9631 ? prog->check_substr : prog->check_utf8);
9633 if (!PL_colorset) reginitcolors();
9634 PerlIO_printf(Perl_debug_log,
9635 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9637 prog->check_substr ? "" : "utf8 ",
9638 PL_colors[5],PL_colors[0],
9641 (strlen(s) > 60 ? "..." : ""));
9644 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9650 handles refcounting and freeing the perl core regexp structure. When
9651 it is necessary to actually free the structure the first thing it
9652 does is call the 'free' method of the regexp_engine associated to
9653 the regexp, allowing the handling of the void *pprivate; member
9654 first. (This routine is not overridable by extensions, which is why
9655 the extensions free is called first.)
9657 See regdupe and regdupe_internal if you change anything here.
9659 #ifndef PERL_IN_XSUB_RE
9661 Perl_pregfree(pTHX_ REGEXP *r)
9667 Perl_pregfree2(pTHX_ REGEXP *rx)
9670 struct regexp *const r = (struct regexp *)SvANY(rx);
9671 GET_RE_DEBUG_FLAGS_DECL;
9673 PERL_ARGS_ASSERT_PREGFREE2;
9676 ReREFCNT_dec(r->mother_re);
9678 CALLREGFREE_PVT(rx); /* free the private data */
9679 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9682 SvREFCNT_dec(r->anchored_substr);
9683 SvREFCNT_dec(r->anchored_utf8);
9684 SvREFCNT_dec(r->float_substr);
9685 SvREFCNT_dec(r->float_utf8);
9686 Safefree(r->substrs);
9688 RX_MATCH_COPY_FREE(rx);
9689 #ifdef PERL_OLD_COPY_ON_WRITE
9690 SvREFCNT_dec(r->saved_copy);
9697 This is a hacky workaround to the structural issue of match results
9698 being stored in the regexp structure which is in turn stored in
9699 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9700 could be PL_curpm in multiple contexts, and could require multiple
9701 result sets being associated with the pattern simultaneously, such
9702 as when doing a recursive match with (??{$qr})
9704 The solution is to make a lightweight copy of the regexp structure
9705 when a qr// is returned from the code executed by (??{$qr}) this
9706 lightweight copy doesnt actually own any of its data except for
9707 the starp/end and the actual regexp structure itself.
9713 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
9716 struct regexp *const r = (struct regexp *)SvANY(rx);
9717 register const I32 npar = r->nparens+1;
9719 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9722 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9723 ret = (struct regexp *)SvANY(ret_x);
9725 (void)ReREFCNT_inc(rx);
9726 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9727 by pointing directly at the buffer, but flagging that the allocated
9728 space in the copy is zero. As we've just done a struct copy, it's now
9729 a case of zero-ing that, rather than copying the current length. */
9730 SvPV_set(ret_x, RX_WRAPPED(rx));
9731 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9732 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
9733 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
9734 SvLEN_set(ret_x, 0);
9735 SvSTASH_set(ret_x, NULL);
9736 SvMAGIC_set(ret_x, NULL);
9737 Newx(ret->offs, npar, regexp_paren_pair);
9738 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9740 Newx(ret->substrs, 1, struct reg_substr_data);
9741 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9743 SvREFCNT_inc_void(ret->anchored_substr);
9744 SvREFCNT_inc_void(ret->anchored_utf8);
9745 SvREFCNT_inc_void(ret->float_substr);
9746 SvREFCNT_inc_void(ret->float_utf8);
9748 /* check_substr and check_utf8, if non-NULL, point to either their
9749 anchored or float namesakes, and don't hold a second reference. */
9751 RX_MATCH_COPIED_off(ret_x);
9752 #ifdef PERL_OLD_COPY_ON_WRITE
9753 ret->saved_copy = NULL;
9755 ret->mother_re = rx;
9761 /* regfree_internal()
9763 Free the private data in a regexp. This is overloadable by
9764 extensions. Perl takes care of the regexp structure in pregfree(),
9765 this covers the *pprivate pointer which technically perl doesn't
9766 know about, however of course we have to handle the
9767 regexp_internal structure when no extension is in use.
9769 Note this is called before freeing anything in the regexp
9774 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9777 struct regexp *const r = (struct regexp *)SvANY(rx);
9779 GET_RE_DEBUG_FLAGS_DECL;
9781 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9787 SV *dsv= sv_newmortal();
9788 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9789 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9790 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9791 PL_colors[4],PL_colors[5],s);
9794 #ifdef RE_TRACK_PATTERN_OFFSETS
9796 Safefree(ri->u.offsets); /* 20010421 MJD */
9799 int n = ri->data->count;
9800 PAD* new_comppad = NULL;
9805 /* If you add a ->what type here, update the comment in regcomp.h */
9806 switch (ri->data->what[n]) {
9811 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
9814 Safefree(ri->data->data[n]);
9817 new_comppad = MUTABLE_AV(ri->data->data[n]);
9820 if (new_comppad == NULL)
9821 Perl_croak(aTHX_ "panic: pregfree comppad");
9822 PAD_SAVE_LOCAL(old_comppad,
9823 /* Watch out for global destruction's random ordering. */
9824 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9827 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9830 op_free((OP_4tree*)ri->data->data[n]);
9832 PAD_RESTORE_LOCAL(old_comppad);
9833 SvREFCNT_dec(MUTABLE_SV(new_comppad));
9839 { /* Aho Corasick add-on structure for a trie node.
9840 Used in stclass optimization only */
9842 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9844 refcount = --aho->refcount;
9847 PerlMemShared_free(aho->states);
9848 PerlMemShared_free(aho->fail);
9849 /* do this last!!!! */
9850 PerlMemShared_free(ri->data->data[n]);
9851 PerlMemShared_free(ri->regstclass);
9857 /* trie structure. */
9859 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9861 refcount = --trie->refcount;
9864 PerlMemShared_free(trie->charmap);
9865 PerlMemShared_free(trie->states);
9866 PerlMemShared_free(trie->trans);
9868 PerlMemShared_free(trie->bitmap);
9870 PerlMemShared_free(trie->jump);
9871 PerlMemShared_free(trie->wordinfo);
9872 /* do this last!!!! */
9873 PerlMemShared_free(ri->data->data[n]);
9878 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9881 Safefree(ri->data->what);
9888 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
9889 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
9890 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9893 re_dup - duplicate a regexp.
9895 This routine is expected to clone a given regexp structure. It is only
9896 compiled under USE_ITHREADS.
9898 After all of the core data stored in struct regexp is duplicated
9899 the regexp_engine.dupe method is used to copy any private data
9900 stored in the *pprivate pointer. This allows extensions to handle
9901 any duplication it needs to do.
9903 See pregfree() and regfree_internal() if you change anything here.
9905 #if defined(USE_ITHREADS)
9906 #ifndef PERL_IN_XSUB_RE
9908 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9912 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9913 struct regexp *ret = (struct regexp *)SvANY(dstr);
9915 PERL_ARGS_ASSERT_RE_DUP_GUTS;
9917 npar = r->nparens+1;
9918 Newx(ret->offs, npar, regexp_paren_pair);
9919 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9921 /* no need to copy these */
9922 Newx(ret->swap, npar, regexp_paren_pair);
9926 /* Do it this way to avoid reading from *r after the StructCopy().
9927 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9928 cache, it doesn't matter. */
9929 const bool anchored = r->check_substr
9930 ? r->check_substr == r->anchored_substr
9931 : r->check_utf8 == r->anchored_utf8;
9932 Newx(ret->substrs, 1, struct reg_substr_data);
9933 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9935 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9936 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9937 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9938 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9940 /* check_substr and check_utf8, if non-NULL, point to either their
9941 anchored or float namesakes, and don't hold a second reference. */
9943 if (ret->check_substr) {
9945 assert(r->check_utf8 == r->anchored_utf8);
9946 ret->check_substr = ret->anchored_substr;
9947 ret->check_utf8 = ret->anchored_utf8;
9949 assert(r->check_substr == r->float_substr);
9950 assert(r->check_utf8 == r->float_utf8);
9951 ret->check_substr = ret->float_substr;
9952 ret->check_utf8 = ret->float_utf8;
9954 } else if (ret->check_utf8) {
9956 ret->check_utf8 = ret->anchored_utf8;
9958 ret->check_utf8 = ret->float_utf8;
9963 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9966 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9968 if (RX_MATCH_COPIED(dstr))
9969 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9972 #ifdef PERL_OLD_COPY_ON_WRITE
9973 ret->saved_copy = NULL;
9976 if (ret->mother_re) {
9977 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
9978 /* Our storage points directly to our mother regexp, but that's
9979 1: a buffer in a different thread
9980 2: something we no longer hold a reference on
9981 so we need to copy it locally. */
9982 /* Note we need to sue SvCUR() on our mother_re, because it, in
9983 turn, may well be pointing to its own mother_re. */
9984 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
9985 SvCUR(ret->mother_re)+1));
9986 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
9988 ret->mother_re = NULL;
9992 #endif /* PERL_IN_XSUB_RE */
9997 This is the internal complement to regdupe() which is used to copy
9998 the structure pointed to by the *pprivate pointer in the regexp.
9999 This is the core version of the extension overridable cloning hook.
10000 The regexp structure being duplicated will be copied by perl prior
10001 to this and will be provided as the regexp *r argument, however
10002 with the /old/ structures pprivate pointer value. Thus this routine
10003 may override any copying normally done by perl.
10005 It returns a pointer to the new regexp_internal structure.
10009 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
10012 struct regexp *const r = (struct regexp *)SvANY(rx);
10013 regexp_internal *reti;
10015 RXi_GET_DECL(r,ri);
10017 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
10019 npar = r->nparens+1;
10022 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
10023 Copy(ri->program, reti->program, len+1, regnode);
10026 reti->regstclass = NULL;
10029 struct reg_data *d;
10030 const int count = ri->data->count;
10033 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
10034 char, struct reg_data);
10035 Newx(d->what, count, U8);
10038 for (i = 0; i < count; i++) {
10039 d->what[i] = ri->data->what[i];
10040 switch (d->what[i]) {
10041 /* legal options are one of: sSfpontTua
10042 see also regcomp.h and pregfree() */
10043 case 'a': /* actually an AV, but the dup function is identical. */
10046 case 'p': /* actually an AV, but the dup function is identical. */
10047 case 'u': /* actually an HV, but the dup function is identical. */
10048 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
10051 /* This is cheating. */
10052 Newx(d->data[i], 1, struct regnode_charclass_class);
10053 StructCopy(ri->data->data[i], d->data[i],
10054 struct regnode_charclass_class);
10055 reti->regstclass = (regnode*)d->data[i];
10058 /* Compiled op trees are readonly and in shared memory,
10059 and can thus be shared without duplication. */
10061 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
10065 /* Trie stclasses are readonly and can thus be shared
10066 * without duplication. We free the stclass in pregfree
10067 * when the corresponding reg_ac_data struct is freed.
10069 reti->regstclass= ri->regstclass;
10073 ((reg_trie_data*)ri->data->data[i])->refcount++;
10077 d->data[i] = ri->data->data[i];
10080 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
10089 reti->name_list_idx = ri->name_list_idx;
10091 #ifdef RE_TRACK_PATTERN_OFFSETS
10092 if (ri->u.offsets) {
10093 Newx(reti->u.offsets, 2*len+1, U32);
10094 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
10097 SetProgLen(reti,len);
10100 return (void*)reti;
10103 #endif /* USE_ITHREADS */
10105 #ifndef PERL_IN_XSUB_RE
10108 - regnext - dig the "next" pointer out of a node
10111 Perl_regnext(pTHX_ register regnode *p)
10114 register I32 offset;
10119 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
10120 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
10123 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
10132 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
10135 STRLEN l1 = strlen(pat1);
10136 STRLEN l2 = strlen(pat2);
10139 const char *message;
10141 PERL_ARGS_ASSERT_RE_CROAK2;
10147 Copy(pat1, buf, l1 , char);
10148 Copy(pat2, buf + l1, l2 , char);
10149 buf[l1 + l2] = '\n';
10150 buf[l1 + l2 + 1] = '\0';
10152 /* ANSI variant takes additional second argument */
10153 va_start(args, pat2);
10157 msv = vmess(buf, &args);
10159 message = SvPV_const(msv,l1);
10162 Copy(message, buf, l1 , char);
10163 buf[l1-1] = '\0'; /* Overwrite \n */
10164 Perl_croak(aTHX_ "%s", buf);
10167 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
10169 #ifndef PERL_IN_XSUB_RE
10171 Perl_save_re_context(pTHX)
10175 struct re_save_state *state;
10177 SAVEVPTR(PL_curcop);
10178 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
10180 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
10181 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
10182 SSPUSHUV(SAVEt_RE_STATE);
10184 Copy(&PL_reg_state, state, 1, struct re_save_state);
10186 PL_reg_start_tmp = 0;
10187 PL_reg_start_tmpl = 0;
10188 PL_reg_oldsaved = NULL;
10189 PL_reg_oldsavedlen = 0;
10190 PL_reg_maxiter = 0;
10191 PL_reg_leftiter = 0;
10192 PL_reg_poscache = NULL;
10193 PL_reg_poscache_size = 0;
10194 #ifdef PERL_OLD_COPY_ON_WRITE
10198 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
10200 const REGEXP * const rx = PM_GETRE(PL_curpm);
10203 for (i = 1; i <= RX_NPARENS(rx); i++) {
10204 char digits[TYPE_CHARS(long)];
10205 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
10206 GV *const *const gvp
10207 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
10210 GV * const gv = *gvp;
10211 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
10221 clear_re(pTHX_ void *r)
10224 ReREFCNT_dec((REGEXP *)r);
10230 S_put_byte(pTHX_ SV *sv, int c)
10232 PERL_ARGS_ASSERT_PUT_BYTE;
10234 /* Our definition of isPRINT() ignores locales, so only bytes that are
10235 not part of UTF-8 are considered printable. I assume that the same
10236 holds for UTF-EBCDIC.
10237 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
10238 which Wikipedia says:
10240 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
10241 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
10242 identical, to the ASCII delete (DEL) or rubout control character.
10243 ) So the old condition can be simplified to !isPRINT(c) */
10245 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
10247 const char string = c;
10248 if (c == '-' || c == ']' || c == '\\' || c == '^')
10249 sv_catpvs(sv, "\\");
10250 sv_catpvn(sv, &string, 1);
10255 #define CLEAR_OPTSTART \
10256 if (optstart) STMT_START { \
10257 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
10261 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
10263 STATIC const regnode *
10264 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
10265 const regnode *last, const regnode *plast,
10266 SV* sv, I32 indent, U32 depth)
10269 register U8 op = PSEUDO; /* Arbitrary non-END op. */
10270 register const regnode *next;
10271 const regnode *optstart= NULL;
10273 RXi_GET_DECL(r,ri);
10274 GET_RE_DEBUG_FLAGS_DECL;
10276 PERL_ARGS_ASSERT_DUMPUNTIL;
10278 #ifdef DEBUG_DUMPUNTIL
10279 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
10280 last ? last-start : 0,plast ? plast-start : 0);
10283 if (plast && plast < last)
10286 while (PL_regkind[op] != END && (!last || node < last)) {
10287 /* While that wasn't END last time... */
10290 if (op == CLOSE || op == WHILEM)
10292 next = regnext((regnode *)node);
10295 if (OP(node) == OPTIMIZED) {
10296 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10303 regprop(r, sv, node);
10304 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10305 (int)(2*indent + 1), "", SvPVX_const(sv));
10307 if (OP(node) != OPTIMIZED) {
10308 if (next == NULL) /* Next ptr. */
10309 PerlIO_printf(Perl_debug_log, " (0)");
10310 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10311 PerlIO_printf(Perl_debug_log, " (FAIL)");
10313 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10314 (void)PerlIO_putc(Perl_debug_log, '\n');
10318 if (PL_regkind[(U8)op] == BRANCHJ) {
10321 register const regnode *nnode = (OP(next) == LONGJMP
10322 ? regnext((regnode *)next)
10324 if (last && nnode > last)
10326 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10329 else if (PL_regkind[(U8)op] == BRANCH) {
10331 DUMPUNTIL(NEXTOPER(node), next);
10333 else if ( PL_regkind[(U8)op] == TRIE ) {
10334 const regnode *this_trie = node;
10335 const char op = OP(node);
10336 const U32 n = ARG(node);
10337 const reg_ac_data * const ac = op>=AHOCORASICK ?
10338 (reg_ac_data *)ri->data->data[n] :
10340 const reg_trie_data * const trie =
10341 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10343 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10345 const regnode *nextbranch= NULL;
10348 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10349 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10351 PerlIO_printf(Perl_debug_log, "%*s%s ",
10352 (int)(2*(indent+3)), "",
10353 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10354 PL_colors[0], PL_colors[1],
10355 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10356 PERL_PV_PRETTY_ELLIPSES |
10357 PERL_PV_PRETTY_LTGT
10362 U16 dist= trie->jump[word_idx+1];
10363 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10364 (UV)((dist ? this_trie + dist : next) - start));
10367 nextbranch= this_trie + trie->jump[0];
10368 DUMPUNTIL(this_trie + dist, nextbranch);
10370 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10371 nextbranch= regnext((regnode *)nextbranch);
10373 PerlIO_printf(Perl_debug_log, "\n");
10376 if (last && next > last)
10381 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10382 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10383 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10385 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10387 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10389 else if ( op == PLUS || op == STAR) {
10390 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10392 else if (op == ANYOF) {
10393 /* arglen 1 + class block */
10394 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
10395 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10396 node = NEXTOPER(node);
10398 else if (PL_regkind[(U8)op] == EXACT) {
10399 /* Literal string, where present. */
10400 node += NODE_SZ_STR(node) - 1;
10401 node = NEXTOPER(node);
10404 node = NEXTOPER(node);
10405 node += regarglen[(U8)op];
10407 if (op == CURLYX || op == OPEN)
10411 #ifdef DEBUG_DUMPUNTIL
10412 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10417 #endif /* DEBUGGING */
10421 * c-indentation-style: bsd
10422 * c-basic-offset: 4
10423 * indent-tabs-mode: t
10426 * ex: set ts=8 sts=4 sw=4 noet: