5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
93 # if defined(BUGGY_MSC6)
94 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
95 # pragma optimize("a",off)
96 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
97 # pragma optimize("w",on )
98 # endif /* BUGGY_MSC6 */
102 #define STATIC static
105 typedef struct RExC_state_t {
106 U32 flags; /* are we folding, multilining? */
107 char *precomp; /* uncompiled string. */
108 REGEXP *rx_sv; /* The SV that is the regexp. */
109 regexp *rx; /* perl core regexp structure */
110 regexp_internal *rxi; /* internal data for regexp object pprivate field */
111 char *start; /* Start of input for compile */
112 char *end; /* End of input for compile */
113 char *parse; /* Input-scan pointer. */
114 I32 whilem_seen; /* number of WHILEM in this expr */
115 regnode *emit_start; /* Start of emitted-code area */
116 regnode *emit_bound; /* First regnode outside of the allocated space */
117 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
118 I32 naughty; /* How bad is this pattern? */
119 I32 sawback; /* Did we see \1, ...? */
121 I32 size; /* Code size. */
122 I32 npar; /* Capture buffer count, (OPEN). */
123 I32 cpar; /* Capture buffer count, (CLOSE). */
124 I32 nestroot; /* root parens we are in - used by accept */
128 regnode **open_parens; /* pointers to open parens */
129 regnode **close_parens; /* pointers to close parens */
130 regnode *opend; /* END node in program */
131 I32 utf8; /* whether the pattern is utf8 or not */
132 I32 orig_utf8; /* whether the pattern was originally in utf8 */
133 /* XXX use this for future optimisation of case
134 * where pattern must be upgraded to utf8. */
135 HV *paren_names; /* Paren names */
137 regnode **recurse; /* Recurse regops */
138 I32 recurse_count; /* Number of recurse regops */
140 char *starttry; /* -Dr: where regtry was called. */
141 #define RExC_starttry (pRExC_state->starttry)
144 const char *lastparse;
146 AV *paren_name_list; /* idx -> name */
147 #define RExC_lastparse (pRExC_state->lastparse)
148 #define RExC_lastnum (pRExC_state->lastnum)
149 #define RExC_paren_name_list (pRExC_state->paren_name_list)
153 #define RExC_flags (pRExC_state->flags)
154 #define RExC_precomp (pRExC_state->precomp)
155 #define RExC_rx_sv (pRExC_state->rx_sv)
156 #define RExC_rx (pRExC_state->rx)
157 #define RExC_rxi (pRExC_state->rxi)
158 #define RExC_start (pRExC_state->start)
159 #define RExC_end (pRExC_state->end)
160 #define RExC_parse (pRExC_state->parse)
161 #define RExC_whilem_seen (pRExC_state->whilem_seen)
162 #ifdef RE_TRACK_PATTERN_OFFSETS
163 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
165 #define RExC_emit (pRExC_state->emit)
166 #define RExC_emit_start (pRExC_state->emit_start)
167 #define RExC_emit_bound (pRExC_state->emit_bound)
168 #define RExC_naughty (pRExC_state->naughty)
169 #define RExC_sawback (pRExC_state->sawback)
170 #define RExC_seen (pRExC_state->seen)
171 #define RExC_size (pRExC_state->size)
172 #define RExC_npar (pRExC_state->npar)
173 #define RExC_nestroot (pRExC_state->nestroot)
174 #define RExC_extralen (pRExC_state->extralen)
175 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
176 #define RExC_seen_evals (pRExC_state->seen_evals)
177 #define RExC_utf8 (pRExC_state->utf8)
178 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
179 #define RExC_open_parens (pRExC_state->open_parens)
180 #define RExC_close_parens (pRExC_state->close_parens)
181 #define RExC_opend (pRExC_state->opend)
182 #define RExC_paren_names (pRExC_state->paren_names)
183 #define RExC_recurse (pRExC_state->recurse)
184 #define RExC_recurse_count (pRExC_state->recurse_count)
187 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
188 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
189 ((*s) == '{' && regcurly(s)))
192 #undef SPSTART /* dratted cpp namespace... */
195 * Flags to be passed up and down.
197 #define WORST 0 /* Worst case. */
198 #define HASWIDTH 0x01 /* Known to match non-null strings. */
199 #define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
200 #define SPSTART 0x04 /* Starts with * or +. */
201 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
202 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
204 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
206 /* whether trie related optimizations are enabled */
207 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
208 #define TRIE_STUDY_OPT
209 #define FULL_TRIE_STUDY
215 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
216 #define PBITVAL(paren) (1 << ((paren) & 7))
217 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
218 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
219 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
222 /* About scan_data_t.
224 During optimisation we recurse through the regexp program performing
225 various inplace (keyhole style) optimisations. In addition study_chunk
226 and scan_commit populate this data structure with information about
227 what strings MUST appear in the pattern. We look for the longest
228 string that must appear for at a fixed location, and we look for the
229 longest string that may appear at a floating location. So for instance
234 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
235 strings (because they follow a .* construct). study_chunk will identify
236 both FOO and BAR as being the longest fixed and floating strings respectively.
238 The strings can be composites, for instance
242 will result in a composite fixed substring 'foo'.
244 For each string some basic information is maintained:
246 - offset or min_offset
247 This is the position the string must appear at, or not before.
248 It also implicitly (when combined with minlenp) tells us how many
249 character must match before the string we are searching.
250 Likewise when combined with minlenp and the length of the string
251 tells us how many characters must appear after the string we have
255 Only used for floating strings. This is the rightmost point that
256 the string can appear at. Ifset to I32 max it indicates that the
257 string can occur infinitely far to the right.
260 A pointer to the minimum length of the pattern that the string
261 was found inside. This is important as in the case of positive
262 lookahead or positive lookbehind we can have multiple patterns
267 The minimum length of the pattern overall is 3, the minimum length
268 of the lookahead part is 3, but the minimum length of the part that
269 will actually match is 1. So 'FOO's minimum length is 3, but the
270 minimum length for the F is 1. This is important as the minimum length
271 is used to determine offsets in front of and behind the string being
272 looked for. Since strings can be composites this is the length of the
273 pattern at the time it was commited with a scan_commit. Note that
274 the length is calculated by study_chunk, so that the minimum lengths
275 are not known until the full pattern has been compiled, thus the
276 pointer to the value.
280 In the case of lookbehind the string being searched for can be
281 offset past the start point of the final matching string.
282 If this value was just blithely removed from the min_offset it would
283 invalidate some of the calculations for how many chars must match
284 before or after (as they are derived from min_offset and minlen and
285 the length of the string being searched for).
286 When the final pattern is compiled and the data is moved from the
287 scan_data_t structure into the regexp structure the information
288 about lookbehind is factored in, with the information that would
289 have been lost precalculated in the end_shift field for the
292 The fields pos_min and pos_delta are used to store the minimum offset
293 and the delta to the maximum offset at the current point in the pattern.
297 typedef struct scan_data_t {
298 /*I32 len_min; unused */
299 /*I32 len_delta; unused */
303 I32 last_end; /* min value, <0 unless valid. */
306 SV **longest; /* Either &l_fixed, or &l_float. */
307 SV *longest_fixed; /* longest fixed string found in pattern */
308 I32 offset_fixed; /* offset where it starts */
309 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
310 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
311 SV *longest_float; /* longest floating string found in pattern */
312 I32 offset_float_min; /* earliest point in string it can appear */
313 I32 offset_float_max; /* latest point in string it can appear */
314 I32 *minlen_float; /* pointer to the minlen relevent to the string */
315 I32 lookbehind_float; /* is the position of the string modified by LB */
319 struct regnode_charclass_class *start_class;
323 * Forward declarations for pregcomp()'s friends.
326 static const scan_data_t zero_scan_data =
327 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
329 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
330 #define SF_BEFORE_SEOL 0x0001
331 #define SF_BEFORE_MEOL 0x0002
332 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
333 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
336 # define SF_FIX_SHIFT_EOL (0+2)
337 # define SF_FL_SHIFT_EOL (0+4)
339 # define SF_FIX_SHIFT_EOL (+2)
340 # define SF_FL_SHIFT_EOL (+4)
343 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
344 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
346 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
347 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
348 #define SF_IS_INF 0x0040
349 #define SF_HAS_PAR 0x0080
350 #define SF_IN_PAR 0x0100
351 #define SF_HAS_EVAL 0x0200
352 #define SCF_DO_SUBSTR 0x0400
353 #define SCF_DO_STCLASS_AND 0x0800
354 #define SCF_DO_STCLASS_OR 0x1000
355 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
356 #define SCF_WHILEM_VISITED_POS 0x2000
358 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
359 #define SCF_SEEN_ACCEPT 0x8000
361 #define UTF (RExC_utf8 != 0)
362 #define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
363 #define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
365 #define OOB_UNICODE 12345678
366 #define OOB_NAMEDCLASS -1
368 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
369 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
372 /* length of regex to show in messages that don't mark a position within */
373 #define RegexLengthToShowInErrorMessages 127
376 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
377 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
378 * op/pragma/warn/regcomp.
380 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
381 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
383 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
386 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
387 * arg. Show regex, up to a maximum length. If it's too long, chop and add
390 #define _FAIL(code) STMT_START { \
391 const char *ellipses = ""; \
392 IV len = RExC_end - RExC_precomp; \
395 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
396 if (len > RegexLengthToShowInErrorMessages) { \
397 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
398 len = RegexLengthToShowInErrorMessages - 10; \
404 #define FAIL(msg) _FAIL( \
405 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
406 msg, (int)len, RExC_precomp, ellipses))
408 #define FAIL2(msg,arg) _FAIL( \
409 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
410 arg, (int)len, RExC_precomp, ellipses))
413 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
415 #define Simple_vFAIL(m) STMT_START { \
416 const IV offset = RExC_parse - RExC_precomp; \
417 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
418 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
422 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
424 #define vFAIL(m) STMT_START { \
426 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
431 * Like Simple_vFAIL(), but accepts two arguments.
433 #define Simple_vFAIL2(m,a1) STMT_START { \
434 const IV offset = RExC_parse - RExC_precomp; \
435 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
436 (int)offset, RExC_precomp, RExC_precomp + offset); \
440 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
442 #define vFAIL2(m,a1) STMT_START { \
444 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
445 Simple_vFAIL2(m, a1); \
450 * Like Simple_vFAIL(), but accepts three arguments.
452 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
453 const IV offset = RExC_parse - RExC_precomp; \
454 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
455 (int)offset, RExC_precomp, RExC_precomp + offset); \
459 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
461 #define vFAIL3(m,a1,a2) STMT_START { \
463 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
464 Simple_vFAIL3(m, a1, a2); \
468 * Like Simple_vFAIL(), but accepts four arguments.
470 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
471 const IV offset = RExC_parse - RExC_precomp; \
472 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
473 (int)offset, RExC_precomp, RExC_precomp + offset); \
476 #define ckWARNreg(loc,m) STMT_START { \
477 const IV offset = loc - RExC_precomp; \
478 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
479 (int)offset, RExC_precomp, RExC_precomp + offset); \
482 #define ckWARNregdep(loc,m) STMT_START { \
483 const IV offset = loc - RExC_precomp; \
484 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
486 (int)offset, RExC_precomp, RExC_precomp + offset); \
489 #define ckWARN2reg(loc, m, a1) STMT_START { \
490 const IV offset = loc - RExC_precomp; \
491 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
492 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
495 #define vWARN3(loc, m, a1, a2) STMT_START { \
496 const IV offset = loc - RExC_precomp; \
497 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
498 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
501 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
502 const IV offset = loc - RExC_precomp; \
503 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
504 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
507 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
508 const IV offset = loc - RExC_precomp; \
509 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
510 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
513 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
514 const IV offset = loc - RExC_precomp; \
515 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
516 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
519 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
520 const IV offset = loc - RExC_precomp; \
521 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
522 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
526 /* Allow for side effects in s */
527 #define REGC(c,s) STMT_START { \
528 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
531 /* Macros for recording node offsets. 20001227 mjd@plover.com
532 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
533 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
534 * Element 0 holds the number n.
535 * Position is 1 indexed.
537 #ifndef RE_TRACK_PATTERN_OFFSETS
538 #define Set_Node_Offset_To_R(node,byte)
539 #define Set_Node_Offset(node,byte)
540 #define Set_Cur_Node_Offset
541 #define Set_Node_Length_To_R(node,len)
542 #define Set_Node_Length(node,len)
543 #define Set_Node_Cur_Length(node)
544 #define Node_Offset(n)
545 #define Node_Length(n)
546 #define Set_Node_Offset_Length(node,offset,len)
547 #define ProgLen(ri) ri->u.proglen
548 #define SetProgLen(ri,x) ri->u.proglen = x
550 #define ProgLen(ri) ri->u.offsets[0]
551 #define SetProgLen(ri,x) ri->u.offsets[0] = x
552 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
554 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
555 __LINE__, (int)(node), (int)(byte))); \
557 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
559 RExC_offsets[2*(node)-1] = (byte); \
564 #define Set_Node_Offset(node,byte) \
565 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
566 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
568 #define Set_Node_Length_To_R(node,len) STMT_START { \
570 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
571 __LINE__, (int)(node), (int)(len))); \
573 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
575 RExC_offsets[2*(node)] = (len); \
580 #define Set_Node_Length(node,len) \
581 Set_Node_Length_To_R((node)-RExC_emit_start, len)
582 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
583 #define Set_Node_Cur_Length(node) \
584 Set_Node_Length(node, RExC_parse - parse_start)
586 /* Get offsets and lengths */
587 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
588 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
590 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
591 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
592 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
596 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
597 #define EXPERIMENTAL_INPLACESCAN
598 #endif /*RE_TRACK_PATTERN_OFFSETS*/
600 #define DEBUG_STUDYDATA(str,data,depth) \
601 DEBUG_OPTIMISE_MORE_r(if(data){ \
602 PerlIO_printf(Perl_debug_log, \
603 "%*s" str "Pos:%"IVdf"/%"IVdf \
604 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
605 (int)(depth)*2, "", \
606 (IV)((data)->pos_min), \
607 (IV)((data)->pos_delta), \
608 (UV)((data)->flags), \
609 (IV)((data)->whilem_c), \
610 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
611 is_inf ? "INF " : "" \
613 if ((data)->last_found) \
614 PerlIO_printf(Perl_debug_log, \
615 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
616 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
617 SvPVX_const((data)->last_found), \
618 (IV)((data)->last_end), \
619 (IV)((data)->last_start_min), \
620 (IV)((data)->last_start_max), \
621 ((data)->longest && \
622 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
623 SvPVX_const((data)->longest_fixed), \
624 (IV)((data)->offset_fixed), \
625 ((data)->longest && \
626 (data)->longest==&((data)->longest_float)) ? "*" : "", \
627 SvPVX_const((data)->longest_float), \
628 (IV)((data)->offset_float_min), \
629 (IV)((data)->offset_float_max) \
631 PerlIO_printf(Perl_debug_log,"\n"); \
634 static void clear_re(pTHX_ void *r);
636 /* Mark that we cannot extend a found fixed substring at this point.
637 Update the longest found anchored substring and the longest found
638 floating substrings if needed. */
641 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
643 const STRLEN l = CHR_SVLEN(data->last_found);
644 const STRLEN old_l = CHR_SVLEN(*data->longest);
645 GET_RE_DEBUG_FLAGS_DECL;
647 PERL_ARGS_ASSERT_SCAN_COMMIT;
649 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
650 SvSetMagicSV(*data->longest, data->last_found);
651 if (*data->longest == data->longest_fixed) {
652 data->offset_fixed = l ? data->last_start_min : data->pos_min;
653 if (data->flags & SF_BEFORE_EOL)
655 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
657 data->flags &= ~SF_FIX_BEFORE_EOL;
658 data->minlen_fixed=minlenp;
659 data->lookbehind_fixed=0;
661 else { /* *data->longest == data->longest_float */
662 data->offset_float_min = l ? data->last_start_min : data->pos_min;
663 data->offset_float_max = (l
664 ? data->last_start_max
665 : data->pos_min + data->pos_delta);
666 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
667 data->offset_float_max = I32_MAX;
668 if (data->flags & SF_BEFORE_EOL)
670 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
672 data->flags &= ~SF_FL_BEFORE_EOL;
673 data->minlen_float=minlenp;
674 data->lookbehind_float=0;
677 SvCUR_set(data->last_found, 0);
679 SV * const sv = data->last_found;
680 if (SvUTF8(sv) && SvMAGICAL(sv)) {
681 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
687 data->flags &= ~SF_BEFORE_EOL;
688 DEBUG_STUDYDATA("commit: ",data,0);
691 /* Can match anything (initialization) */
693 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
695 PERL_ARGS_ASSERT_CL_ANYTHING;
697 ANYOF_CLASS_ZERO(cl);
698 ANYOF_BITMAP_SETALL(cl);
699 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
701 cl->flags |= ANYOF_LOCALE;
704 /* Can match anything (initialization) */
706 S_cl_is_anything(const struct regnode_charclass_class *cl)
710 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
712 for (value = 0; value <= ANYOF_MAX; value += 2)
713 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
715 if (!(cl->flags & ANYOF_UNICODE_ALL))
717 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
722 /* Can match anything (initialization) */
724 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
726 PERL_ARGS_ASSERT_CL_INIT;
728 Zero(cl, 1, struct regnode_charclass_class);
730 cl_anything(pRExC_state, cl);
734 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
736 PERL_ARGS_ASSERT_CL_INIT_ZERO;
738 Zero(cl, 1, struct regnode_charclass_class);
740 cl_anything(pRExC_state, cl);
742 cl->flags |= ANYOF_LOCALE;
745 /* 'And' a given class with another one. Can create false positives */
746 /* We assume that cl is not inverted */
748 S_cl_and(struct regnode_charclass_class *cl,
749 const struct regnode_charclass_class *and_with)
751 PERL_ARGS_ASSERT_CL_AND;
753 assert(and_with->type == ANYOF);
754 if (!(and_with->flags & ANYOF_CLASS)
755 && !(cl->flags & ANYOF_CLASS)
756 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
757 && !(and_with->flags & ANYOF_FOLD)
758 && !(cl->flags & ANYOF_FOLD)) {
761 if (and_with->flags & ANYOF_INVERT)
762 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
763 cl->bitmap[i] &= ~and_with->bitmap[i];
765 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
766 cl->bitmap[i] &= and_with->bitmap[i];
767 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
768 if (!(and_with->flags & ANYOF_EOS))
769 cl->flags &= ~ANYOF_EOS;
771 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
772 !(and_with->flags & ANYOF_INVERT)) {
773 cl->flags &= ~ANYOF_UNICODE_ALL;
774 cl->flags |= ANYOF_UNICODE;
775 ARG_SET(cl, ARG(and_with));
777 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
778 !(and_with->flags & ANYOF_INVERT))
779 cl->flags &= ~ANYOF_UNICODE_ALL;
780 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
781 !(and_with->flags & ANYOF_INVERT))
782 cl->flags &= ~ANYOF_UNICODE;
785 /* 'OR' a given class with another one. Can create false positives */
786 /* We assume that cl is not inverted */
788 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
790 PERL_ARGS_ASSERT_CL_OR;
792 if (or_with->flags & ANYOF_INVERT) {
794 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
795 * <= (B1 | !B2) | (CL1 | !CL2)
796 * which is wasteful if CL2 is small, but we ignore CL2:
797 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
798 * XXXX Can we handle case-fold? Unclear:
799 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
800 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
802 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
803 && !(or_with->flags & ANYOF_FOLD)
804 && !(cl->flags & ANYOF_FOLD) ) {
807 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
808 cl->bitmap[i] |= ~or_with->bitmap[i];
809 } /* XXXX: logic is complicated otherwise */
811 cl_anything(pRExC_state, cl);
814 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
815 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
816 && (!(or_with->flags & ANYOF_FOLD)
817 || (cl->flags & ANYOF_FOLD)) ) {
820 /* OR char bitmap and class bitmap separately */
821 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
822 cl->bitmap[i] |= or_with->bitmap[i];
823 if (or_with->flags & ANYOF_CLASS) {
824 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
825 cl->classflags[i] |= or_with->classflags[i];
826 cl->flags |= ANYOF_CLASS;
829 else { /* XXXX: logic is complicated, leave it along for a moment. */
830 cl_anything(pRExC_state, cl);
833 if (or_with->flags & ANYOF_EOS)
834 cl->flags |= ANYOF_EOS;
836 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
837 ARG(cl) != ARG(or_with)) {
838 cl->flags |= ANYOF_UNICODE_ALL;
839 cl->flags &= ~ANYOF_UNICODE;
841 if (or_with->flags & ANYOF_UNICODE_ALL) {
842 cl->flags |= ANYOF_UNICODE_ALL;
843 cl->flags &= ~ANYOF_UNICODE;
847 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
848 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
849 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
850 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
855 dump_trie(trie,widecharmap,revcharmap)
856 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
857 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
859 These routines dump out a trie in a somewhat readable format.
860 The _interim_ variants are used for debugging the interim
861 tables that are used to generate the final compressed
862 representation which is what dump_trie expects.
864 Part of the reason for their existance is to provide a form
865 of documentation as to how the different representations function.
870 Dumps the final compressed table form of the trie to Perl_debug_log.
871 Used for debugging make_trie().
875 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
876 AV *revcharmap, U32 depth)
879 SV *sv=sv_newmortal();
880 int colwidth= widecharmap ? 6 : 4;
882 GET_RE_DEBUG_FLAGS_DECL;
884 PERL_ARGS_ASSERT_DUMP_TRIE;
886 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
887 (int)depth * 2 + 2,"",
888 "Match","Base","Ofs" );
890 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
891 SV ** const tmp = av_fetch( revcharmap, state, 0);
893 PerlIO_printf( Perl_debug_log, "%*s",
895 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
896 PL_colors[0], PL_colors[1],
897 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
898 PERL_PV_ESCAPE_FIRSTCHAR
903 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
904 (int)depth * 2 + 2,"");
906 for( state = 0 ; state < trie->uniquecharcount ; state++ )
907 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
908 PerlIO_printf( Perl_debug_log, "\n");
910 for( state = 1 ; state < trie->statecount ; state++ ) {
911 const U32 base = trie->states[ state ].trans.base;
913 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
915 if ( trie->states[ state ].wordnum ) {
916 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
918 PerlIO_printf( Perl_debug_log, "%6s", "" );
921 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
926 while( ( base + ofs < trie->uniquecharcount ) ||
927 ( base + ofs - trie->uniquecharcount < trie->lasttrans
928 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
931 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
933 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
934 if ( ( base + ofs >= trie->uniquecharcount ) &&
935 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
936 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
938 PerlIO_printf( Perl_debug_log, "%*"UVXf,
940 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
942 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
946 PerlIO_printf( Perl_debug_log, "]");
949 PerlIO_printf( Perl_debug_log, "\n" );
951 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
952 for (word=1; word <= trie->wordcount; word++) {
953 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
954 (int)word, (int)(trie->wordinfo[word].prev),
955 (int)(trie->wordinfo[word].len));
957 PerlIO_printf(Perl_debug_log, "\n" );
960 Dumps a fully constructed but uncompressed trie in list form.
961 List tries normally only are used for construction when the number of
962 possible chars (trie->uniquecharcount) is very high.
963 Used for debugging make_trie().
966 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
967 HV *widecharmap, AV *revcharmap, U32 next_alloc,
971 SV *sv=sv_newmortal();
972 int colwidth= widecharmap ? 6 : 4;
973 GET_RE_DEBUG_FLAGS_DECL;
975 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
977 /* print out the table precompression. */
978 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
979 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
980 "------:-----+-----------------\n" );
982 for( state=1 ; state < next_alloc ; state ++ ) {
985 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
986 (int)depth * 2 + 2,"", (UV)state );
987 if ( ! trie->states[ state ].wordnum ) {
988 PerlIO_printf( Perl_debug_log, "%5s| ","");
990 PerlIO_printf( Perl_debug_log, "W%4x| ",
991 trie->states[ state ].wordnum
994 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
995 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
997 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
999 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1000 PL_colors[0], PL_colors[1],
1001 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1002 PERL_PV_ESCAPE_FIRSTCHAR
1004 TRIE_LIST_ITEM(state,charid).forid,
1005 (UV)TRIE_LIST_ITEM(state,charid).newstate
1008 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1009 (int)((depth * 2) + 14), "");
1012 PerlIO_printf( Perl_debug_log, "\n");
1017 Dumps a fully constructed but uncompressed trie in table form.
1018 This is the normal DFA style state transition table, with a few
1019 twists to facilitate compression later.
1020 Used for debugging make_trie().
1023 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1024 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1029 SV *sv=sv_newmortal();
1030 int colwidth= widecharmap ? 6 : 4;
1031 GET_RE_DEBUG_FLAGS_DECL;
1033 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1036 print out the table precompression so that we can do a visual check
1037 that they are identical.
1040 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1042 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1043 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1045 PerlIO_printf( Perl_debug_log, "%*s",
1047 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1048 PL_colors[0], PL_colors[1],
1049 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1050 PERL_PV_ESCAPE_FIRSTCHAR
1056 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1058 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1059 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1062 PerlIO_printf( Perl_debug_log, "\n" );
1064 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1066 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1067 (int)depth * 2 + 2,"",
1068 (UV)TRIE_NODENUM( state ) );
1070 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1071 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1073 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1075 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1077 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1078 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1080 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1081 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1089 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1090 startbranch: the first branch in the whole branch sequence
1091 first : start branch of sequence of branch-exact nodes.
1092 May be the same as startbranch
1093 last : Thing following the last branch.
1094 May be the same as tail.
1095 tail : item following the branch sequence
1096 count : words in the sequence
1097 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1098 depth : indent depth
1100 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1102 A trie is an N'ary tree where the branches are determined by digital
1103 decomposition of the key. IE, at the root node you look up the 1st character and
1104 follow that branch repeat until you find the end of the branches. Nodes can be
1105 marked as "accepting" meaning they represent a complete word. Eg:
1109 would convert into the following structure. Numbers represent states, letters
1110 following numbers represent valid transitions on the letter from that state, if
1111 the number is in square brackets it represents an accepting state, otherwise it
1112 will be in parenthesis.
1114 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1118 (1) +-i->(6)-+-s->[7]
1120 +-s->(3)-+-h->(4)-+-e->[5]
1122 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1124 This shows that when matching against the string 'hers' we will begin at state 1
1125 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1126 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1127 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1128 single traverse. We store a mapping from accepting to state to which word was
1129 matched, and then when we have multiple possibilities we try to complete the
1130 rest of the regex in the order in which they occured in the alternation.
1132 The only prior NFA like behaviour that would be changed by the TRIE support is
1133 the silent ignoring of duplicate alternations which are of the form:
1135 / (DUPE|DUPE) X? (?{ ... }) Y /x
1137 Thus EVAL blocks follwing a trie may be called a different number of times with
1138 and without the optimisation. With the optimisations dupes will be silently
1139 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1140 the following demonstrates:
1142 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1144 which prints out 'word' three times, but
1146 'words'=~/(word|word|word)(?{ print $1 })S/
1148 which doesnt print it out at all. This is due to other optimisations kicking in.
1150 Example of what happens on a structural level:
1152 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1154 1: CURLYM[1] {1,32767}(18)
1165 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1166 and should turn into:
1168 1: CURLYM[1] {1,32767}(18)
1170 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1178 Cases where tail != last would be like /(?foo|bar)baz/:
1188 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1189 and would end up looking like:
1192 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1199 d = uvuni_to_utf8_flags(d, uv, 0);
1201 is the recommended Unicode-aware way of saying
1206 #define TRIE_STORE_REVCHAR \
1209 SV *zlopp = newSV(2); \
1210 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1211 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1212 SvCUR_set(zlopp, kapow - flrbbbbb); \
1215 av_push(revcharmap, zlopp); \
1217 char ooooff = (char)uvc; \
1218 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1222 #define TRIE_READ_CHAR STMT_START { \
1226 if ( foldlen > 0 ) { \
1227 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1232 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1233 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1234 foldlen -= UNISKIP( uvc ); \
1235 scan = foldbuf + UNISKIP( uvc ); \
1238 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1248 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1249 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1250 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1251 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1253 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1254 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1255 TRIE_LIST_CUR( state )++; \
1258 #define TRIE_LIST_NEW(state) STMT_START { \
1259 Newxz( trie->states[ state ].trans.list, \
1260 4, reg_trie_trans_le ); \
1261 TRIE_LIST_CUR( state ) = 1; \
1262 TRIE_LIST_LEN( state ) = 4; \
1265 #define TRIE_HANDLE_WORD(state) STMT_START { \
1266 U16 dupe= trie->states[ state ].wordnum; \
1267 regnode * const noper_next = regnext( noper ); \
1270 /* store the word for dumping */ \
1272 if (OP(noper) != NOTHING) \
1273 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1275 tmp = newSVpvn_utf8( "", 0, UTF ); \
1276 av_push( trie_words, tmp ); \
1280 trie->wordinfo[curword].prev = 0; \
1281 trie->wordinfo[curword].len = wordlen; \
1282 trie->wordinfo[curword].accept = state; \
1284 if ( noper_next < tail ) { \
1286 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1287 trie->jump[curword] = (U16)(noper_next - convert); \
1289 jumper = noper_next; \
1291 nextbranch= regnext(cur); \
1295 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1296 /* chain, so that when the bits of chain are later */\
1297 /* linked together, the dups appear in the chain */\
1298 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1299 trie->wordinfo[dupe].prev = curword; \
1301 /* we haven't inserted this word yet. */ \
1302 trie->states[ state ].wordnum = curword; \
1307 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1308 ( ( base + charid >= ucharcount \
1309 && base + charid < ubound \
1310 && state == trie->trans[ base - ucharcount + charid ].check \
1311 && trie->trans[ base - ucharcount + charid ].next ) \
1312 ? trie->trans[ base - ucharcount + charid ].next \
1313 : ( state==1 ? special : 0 ) \
1317 #define MADE_JUMP_TRIE 2
1318 #define MADE_EXACT_TRIE 4
1321 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1324 /* first pass, loop through and scan words */
1325 reg_trie_data *trie;
1326 HV *widecharmap = NULL;
1327 AV *revcharmap = newAV();
1329 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1334 regnode *jumper = NULL;
1335 regnode *nextbranch = NULL;
1336 regnode *convert = NULL;
1337 U32 *prev_states; /* temp array mapping each state to previous one */
1338 /* we just use folder as a flag in utf8 */
1339 const U8 * const folder = ( flags == EXACTF
1341 : ( flags == EXACTFL
1348 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1349 AV *trie_words = NULL;
1350 /* along with revcharmap, this only used during construction but both are
1351 * useful during debugging so we store them in the struct when debugging.
1354 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1355 STRLEN trie_charcount=0;
1357 SV *re_trie_maxbuff;
1358 GET_RE_DEBUG_FLAGS_DECL;
1360 PERL_ARGS_ASSERT_MAKE_TRIE;
1362 PERL_UNUSED_ARG(depth);
1365 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1367 trie->startstate = 1;
1368 trie->wordcount = word_count;
1369 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1370 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1371 if (!(UTF && folder))
1372 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1373 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1374 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1377 trie_words = newAV();
1380 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1381 if (!SvIOK(re_trie_maxbuff)) {
1382 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1385 PerlIO_printf( Perl_debug_log,
1386 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1387 (int)depth * 2 + 2, "",
1388 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1389 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1393 /* Find the node we are going to overwrite */
1394 if ( first == startbranch && OP( last ) != BRANCH ) {
1395 /* whole branch chain */
1398 /* branch sub-chain */
1399 convert = NEXTOPER( first );
1402 /* -- First loop and Setup --
1404 We first traverse the branches and scan each word to determine if it
1405 contains widechars, and how many unique chars there are, this is
1406 important as we have to build a table with at least as many columns as we
1409 We use an array of integers to represent the character codes 0..255
1410 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1411 native representation of the character value as the key and IV's for the
1414 *TODO* If we keep track of how many times each character is used we can
1415 remap the columns so that the table compression later on is more
1416 efficient in terms of memory by ensuring most common value is in the
1417 middle and the least common are on the outside. IMO this would be better
1418 than a most to least common mapping as theres a decent chance the most
1419 common letter will share a node with the least common, meaning the node
1420 will not be compressable. With a middle is most common approach the worst
1421 case is when we have the least common nodes twice.
1425 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1426 regnode * const noper = NEXTOPER( cur );
1427 const U8 *uc = (U8*)STRING( noper );
1428 const U8 * const e = uc + STR_LEN( noper );
1430 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1431 const U8 *scan = (U8*)NULL;
1432 U32 wordlen = 0; /* required init */
1434 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1436 if (OP(noper) == NOTHING) {
1440 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1441 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1442 regardless of encoding */
1444 for ( ; uc < e ; uc += len ) {
1445 TRIE_CHARCOUNT(trie)++;
1449 if ( !trie->charmap[ uvc ] ) {
1450 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1452 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1456 /* store the codepoint in the bitmap, and if its ascii
1457 also store its folded equivelent. */
1458 TRIE_BITMAP_SET(trie,uvc);
1460 /* store the folded codepoint */
1461 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1464 /* store first byte of utf8 representation of
1465 codepoints in the 127 < uvc < 256 range */
1466 if (127 < uvc && uvc < 192) {
1467 TRIE_BITMAP_SET(trie,194);
1468 } else if (191 < uvc ) {
1469 TRIE_BITMAP_SET(trie,195);
1470 /* && uvc < 256 -- we know uvc is < 256 already */
1473 set_bit = 0; /* We've done our bit :-) */
1478 widecharmap = newHV();
1480 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1483 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1485 if ( !SvTRUE( *svpp ) ) {
1486 sv_setiv( *svpp, ++trie->uniquecharcount );
1491 if( cur == first ) {
1494 } else if (chars < trie->minlen) {
1496 } else if (chars > trie->maxlen) {
1500 } /* end first pass */
1501 DEBUG_TRIE_COMPILE_r(
1502 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1503 (int)depth * 2 + 2,"",
1504 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1505 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1506 (int)trie->minlen, (int)trie->maxlen )
1510 We now know what we are dealing with in terms of unique chars and
1511 string sizes so we can calculate how much memory a naive
1512 representation using a flat table will take. If it's over a reasonable
1513 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1514 conservative but potentially much slower representation using an array
1517 At the end we convert both representations into the same compressed
1518 form that will be used in regexec.c for matching with. The latter
1519 is a form that cannot be used to construct with but has memory
1520 properties similar to the list form and access properties similar
1521 to the table form making it both suitable for fast searches and
1522 small enough that its feasable to store for the duration of a program.
1524 See the comment in the code where the compressed table is produced
1525 inplace from the flat tabe representation for an explanation of how
1526 the compression works.
1531 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1534 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1536 Second Pass -- Array Of Lists Representation
1538 Each state will be represented by a list of charid:state records
1539 (reg_trie_trans_le) the first such element holds the CUR and LEN
1540 points of the allocated array. (See defines above).
1542 We build the initial structure using the lists, and then convert
1543 it into the compressed table form which allows faster lookups
1544 (but cant be modified once converted).
1547 STRLEN transcount = 1;
1549 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1550 "%*sCompiling trie using list compiler\n",
1551 (int)depth * 2 + 2, ""));
1553 trie->states = (reg_trie_state *)
1554 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1555 sizeof(reg_trie_state) );
1559 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1561 regnode * const noper = NEXTOPER( cur );
1562 U8 *uc = (U8*)STRING( noper );
1563 const U8 * const e = uc + STR_LEN( noper );
1564 U32 state = 1; /* required init */
1565 U16 charid = 0; /* sanity init */
1566 U8 *scan = (U8*)NULL; /* sanity init */
1567 STRLEN foldlen = 0; /* required init */
1568 U32 wordlen = 0; /* required init */
1569 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1571 if (OP(noper) != NOTHING) {
1572 for ( ; uc < e ; uc += len ) {
1577 charid = trie->charmap[ uvc ];
1579 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1583 charid=(U16)SvIV( *svpp );
1586 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1593 if ( !trie->states[ state ].trans.list ) {
1594 TRIE_LIST_NEW( state );
1596 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1597 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1598 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1603 newstate = next_alloc++;
1604 prev_states[newstate] = state;
1605 TRIE_LIST_PUSH( state, charid, newstate );
1610 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1614 TRIE_HANDLE_WORD(state);
1616 } /* end second pass */
1618 /* next alloc is the NEXT state to be allocated */
1619 trie->statecount = next_alloc;
1620 trie->states = (reg_trie_state *)
1621 PerlMemShared_realloc( trie->states,
1623 * sizeof(reg_trie_state) );
1625 /* and now dump it out before we compress it */
1626 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1627 revcharmap, next_alloc,
1631 trie->trans = (reg_trie_trans *)
1632 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1639 for( state=1 ; state < next_alloc ; state ++ ) {
1643 DEBUG_TRIE_COMPILE_MORE_r(
1644 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1648 if (trie->states[state].trans.list) {
1649 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1653 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1654 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1655 if ( forid < minid ) {
1657 } else if ( forid > maxid ) {
1661 if ( transcount < tp + maxid - minid + 1) {
1663 trie->trans = (reg_trie_trans *)
1664 PerlMemShared_realloc( trie->trans,
1666 * sizeof(reg_trie_trans) );
1667 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1669 base = trie->uniquecharcount + tp - minid;
1670 if ( maxid == minid ) {
1672 for ( ; zp < tp ; zp++ ) {
1673 if ( ! trie->trans[ zp ].next ) {
1674 base = trie->uniquecharcount + zp - minid;
1675 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1676 trie->trans[ zp ].check = state;
1682 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1683 trie->trans[ tp ].check = state;
1688 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1689 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1690 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1691 trie->trans[ tid ].check = state;
1693 tp += ( maxid - minid + 1 );
1695 Safefree(trie->states[ state ].trans.list);
1698 DEBUG_TRIE_COMPILE_MORE_r(
1699 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1702 trie->states[ state ].trans.base=base;
1704 trie->lasttrans = tp + 1;
1708 Second Pass -- Flat Table Representation.
1710 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1711 We know that we will need Charcount+1 trans at most to store the data
1712 (one row per char at worst case) So we preallocate both structures
1713 assuming worst case.
1715 We then construct the trie using only the .next slots of the entry
1718 We use the .check field of the first entry of the node temporarily to
1719 make compression both faster and easier by keeping track of how many non
1720 zero fields are in the node.
1722 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1725 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1726 number representing the first entry of the node, and state as a
1727 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1728 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1729 are 2 entrys per node. eg:
1737 The table is internally in the right hand, idx form. However as we also
1738 have to deal with the states array which is indexed by nodenum we have to
1739 use TRIE_NODENUM() to convert.
1742 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1743 "%*sCompiling trie using table compiler\n",
1744 (int)depth * 2 + 2, ""));
1746 trie->trans = (reg_trie_trans *)
1747 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1748 * trie->uniquecharcount + 1,
1749 sizeof(reg_trie_trans) );
1750 trie->states = (reg_trie_state *)
1751 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1752 sizeof(reg_trie_state) );
1753 next_alloc = trie->uniquecharcount + 1;
1756 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1758 regnode * const noper = NEXTOPER( cur );
1759 const U8 *uc = (U8*)STRING( noper );
1760 const U8 * const e = uc + STR_LEN( noper );
1762 U32 state = 1; /* required init */
1764 U16 charid = 0; /* sanity init */
1765 U32 accept_state = 0; /* sanity init */
1766 U8 *scan = (U8*)NULL; /* sanity init */
1768 STRLEN foldlen = 0; /* required init */
1769 U32 wordlen = 0; /* required init */
1770 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1772 if ( OP(noper) != NOTHING ) {
1773 for ( ; uc < e ; uc += len ) {
1778 charid = trie->charmap[ uvc ];
1780 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1781 charid = svpp ? (U16)SvIV(*svpp) : 0;
1785 if ( !trie->trans[ state + charid ].next ) {
1786 trie->trans[ state + charid ].next = next_alloc;
1787 trie->trans[ state ].check++;
1788 prev_states[TRIE_NODENUM(next_alloc)]
1789 = TRIE_NODENUM(state);
1790 next_alloc += trie->uniquecharcount;
1792 state = trie->trans[ state + charid ].next;
1794 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1796 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1799 accept_state = TRIE_NODENUM( state );
1800 TRIE_HANDLE_WORD(accept_state);
1802 } /* end second pass */
1804 /* and now dump it out before we compress it */
1805 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1807 next_alloc, depth+1));
1811 * Inplace compress the table.*
1813 For sparse data sets the table constructed by the trie algorithm will
1814 be mostly 0/FAIL transitions or to put it another way mostly empty.
1815 (Note that leaf nodes will not contain any transitions.)
1817 This algorithm compresses the tables by eliminating most such
1818 transitions, at the cost of a modest bit of extra work during lookup:
1820 - Each states[] entry contains a .base field which indicates the
1821 index in the state[] array wheres its transition data is stored.
1823 - If .base is 0 there are no valid transitions from that node.
1825 - If .base is nonzero then charid is added to it to find an entry in
1828 -If trans[states[state].base+charid].check!=state then the
1829 transition is taken to be a 0/Fail transition. Thus if there are fail
1830 transitions at the front of the node then the .base offset will point
1831 somewhere inside the previous nodes data (or maybe even into a node
1832 even earlier), but the .check field determines if the transition is
1836 The following process inplace converts the table to the compressed
1837 table: We first do not compress the root node 1,and mark its all its
1838 .check pointers as 1 and set its .base pointer as 1 as well. This
1839 allows to do a DFA construction from the compressed table later, and
1840 ensures that any .base pointers we calculate later are greater than
1843 - We set 'pos' to indicate the first entry of the second node.
1845 - We then iterate over the columns of the node, finding the first and
1846 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1847 and set the .check pointers accordingly, and advance pos
1848 appropriately and repreat for the next node. Note that when we copy
1849 the next pointers we have to convert them from the original
1850 NODEIDX form to NODENUM form as the former is not valid post
1853 - If a node has no transitions used we mark its base as 0 and do not
1854 advance the pos pointer.
1856 - If a node only has one transition we use a second pointer into the
1857 structure to fill in allocated fail transitions from other states.
1858 This pointer is independent of the main pointer and scans forward
1859 looking for null transitions that are allocated to a state. When it
1860 finds one it writes the single transition into the "hole". If the
1861 pointer doesnt find one the single transition is appended as normal.
1863 - Once compressed we can Renew/realloc the structures to release the
1866 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1867 specifically Fig 3.47 and the associated pseudocode.
1871 const U32 laststate = TRIE_NODENUM( next_alloc );
1874 trie->statecount = laststate;
1876 for ( state = 1 ; state < laststate ; state++ ) {
1878 const U32 stateidx = TRIE_NODEIDX( state );
1879 const U32 o_used = trie->trans[ stateidx ].check;
1880 U32 used = trie->trans[ stateidx ].check;
1881 trie->trans[ stateidx ].check = 0;
1883 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1884 if ( flag || trie->trans[ stateidx + charid ].next ) {
1885 if ( trie->trans[ stateidx + charid ].next ) {
1887 for ( ; zp < pos ; zp++ ) {
1888 if ( ! trie->trans[ zp ].next ) {
1892 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1893 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1894 trie->trans[ zp ].check = state;
1895 if ( ++zp > pos ) pos = zp;
1902 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1904 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1905 trie->trans[ pos ].check = state;
1910 trie->lasttrans = pos + 1;
1911 trie->states = (reg_trie_state *)
1912 PerlMemShared_realloc( trie->states, laststate
1913 * sizeof(reg_trie_state) );
1914 DEBUG_TRIE_COMPILE_MORE_r(
1915 PerlIO_printf( Perl_debug_log,
1916 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1917 (int)depth * 2 + 2,"",
1918 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1921 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1924 } /* end table compress */
1926 DEBUG_TRIE_COMPILE_MORE_r(
1927 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1928 (int)depth * 2 + 2, "",
1929 (UV)trie->statecount,
1930 (UV)trie->lasttrans)
1932 /* resize the trans array to remove unused space */
1933 trie->trans = (reg_trie_trans *)
1934 PerlMemShared_realloc( trie->trans, trie->lasttrans
1935 * sizeof(reg_trie_trans) );
1937 { /* Modify the program and insert the new TRIE node*/
1938 U8 nodetype =(U8)(flags & 0xFF);
1942 regnode *optimize = NULL;
1943 #ifdef RE_TRACK_PATTERN_OFFSETS
1946 U32 mjd_nodelen = 0;
1947 #endif /* RE_TRACK_PATTERN_OFFSETS */
1948 #endif /* DEBUGGING */
1950 This means we convert either the first branch or the first Exact,
1951 depending on whether the thing following (in 'last') is a branch
1952 or not and whther first is the startbranch (ie is it a sub part of
1953 the alternation or is it the whole thing.)
1954 Assuming its a sub part we conver the EXACT otherwise we convert
1955 the whole branch sequence, including the first.
1957 /* Find the node we are going to overwrite */
1958 if ( first != startbranch || OP( last ) == BRANCH ) {
1959 /* branch sub-chain */
1960 NEXT_OFF( first ) = (U16)(last - first);
1961 #ifdef RE_TRACK_PATTERN_OFFSETS
1963 mjd_offset= Node_Offset((convert));
1964 mjd_nodelen= Node_Length((convert));
1967 /* whole branch chain */
1969 #ifdef RE_TRACK_PATTERN_OFFSETS
1972 const regnode *nop = NEXTOPER( convert );
1973 mjd_offset= Node_Offset((nop));
1974 mjd_nodelen= Node_Length((nop));
1978 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1979 (int)depth * 2 + 2, "",
1980 (UV)mjd_offset, (UV)mjd_nodelen)
1983 /* But first we check to see if there is a common prefix we can
1984 split out as an EXACT and put in front of the TRIE node. */
1985 trie->startstate= 1;
1986 if ( trie->bitmap && !widecharmap && !trie->jump ) {
1988 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
1992 const U32 base = trie->states[ state ].trans.base;
1994 if ( trie->states[state].wordnum )
1997 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1998 if ( ( base + ofs >= trie->uniquecharcount ) &&
1999 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2000 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2002 if ( ++count > 1 ) {
2003 SV **tmp = av_fetch( revcharmap, ofs, 0);
2004 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2005 if ( state == 1 ) break;
2007 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2009 PerlIO_printf(Perl_debug_log,
2010 "%*sNew Start State=%"UVuf" Class: [",
2011 (int)depth * 2 + 2, "",
2014 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2015 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2017 TRIE_BITMAP_SET(trie,*ch);
2019 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2021 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2025 TRIE_BITMAP_SET(trie,*ch);
2027 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2028 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2034 SV **tmp = av_fetch( revcharmap, idx, 0);
2036 char *ch = SvPV( *tmp, len );
2038 SV *sv=sv_newmortal();
2039 PerlIO_printf( Perl_debug_log,
2040 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2041 (int)depth * 2 + 2, "",
2043 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2044 PL_colors[0], PL_colors[1],
2045 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2046 PERL_PV_ESCAPE_FIRSTCHAR
2051 OP( convert ) = nodetype;
2052 str=STRING(convert);
2055 STR_LEN(convert) += len;
2061 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2066 trie->prefixlen = (state-1);
2068 regnode *n = convert+NODE_SZ_STR(convert);
2069 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2070 trie->startstate = state;
2071 trie->minlen -= (state - 1);
2072 trie->maxlen -= (state - 1);
2074 /* At least the UNICOS C compiler choked on this
2075 * being argument to DEBUG_r(), so let's just have
2078 #ifdef PERL_EXT_RE_BUILD
2084 regnode *fix = convert;
2085 U32 word = trie->wordcount;
2087 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2088 while( ++fix < n ) {
2089 Set_Node_Offset_Length(fix, 0, 0);
2092 SV ** const tmp = av_fetch( trie_words, word, 0 );
2094 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2095 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2097 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2105 NEXT_OFF(convert) = (U16)(tail - convert);
2106 DEBUG_r(optimize= n);
2112 if ( trie->maxlen ) {
2113 NEXT_OFF( convert ) = (U16)(tail - convert);
2114 ARG_SET( convert, data_slot );
2115 /* Store the offset to the first unabsorbed branch in
2116 jump[0], which is otherwise unused by the jump logic.
2117 We use this when dumping a trie and during optimisation. */
2119 trie->jump[0] = (U16)(nextbranch - convert);
2122 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
2123 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2125 OP( convert ) = TRIEC;
2126 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2127 PerlMemShared_free(trie->bitmap);
2130 OP( convert ) = TRIE;
2132 /* store the type in the flags */
2133 convert->flags = nodetype;
2137 + regarglen[ OP( convert ) ];
2139 /* XXX We really should free up the resource in trie now,
2140 as we won't use them - (which resources?) dmq */
2142 /* needed for dumping*/
2143 DEBUG_r(if (optimize) {
2144 regnode *opt = convert;
2146 while ( ++opt < optimize) {
2147 Set_Node_Offset_Length(opt,0,0);
2150 Try to clean up some of the debris left after the
2153 while( optimize < jumper ) {
2154 mjd_nodelen += Node_Length((optimize));
2155 OP( optimize ) = OPTIMIZED;
2156 Set_Node_Offset_Length(optimize,0,0);
2159 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2161 } /* end node insert */
2163 /* Finish populating the prev field of the wordinfo array. Walk back
2164 * from each accept state until we find another accept state, and if
2165 * so, point the first word's .prev field at the second word. If the
2166 * second already has a .prev field set, stop now. This will be the
2167 * case either if we've already processed that word's accept state,
2168 * or that that state had multiple words, and the overspill words
2169 * were already linked up earlier.
2176 for (word=1; word <= trie->wordcount; word++) {
2178 if (trie->wordinfo[word].prev)
2180 state = trie->wordinfo[word].accept;
2182 state = prev_states[state];
2185 prev = trie->states[state].wordnum;
2189 trie->wordinfo[word].prev = prev;
2191 Safefree(prev_states);
2195 /* and now dump out the compressed format */
2196 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2198 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2200 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2201 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2203 SvREFCNT_dec(revcharmap);
2207 : trie->startstate>1
2213 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2215 /* The Trie is constructed and compressed now so we can build a fail array now if its needed
2217 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2218 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2221 We find the fail state for each state in the trie, this state is the longest proper
2222 suffix of the current states 'word' that is also a proper prefix of another word in our
2223 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2224 the DFA not to have to restart after its tried and failed a word at a given point, it
2225 simply continues as though it had been matching the other word in the first place.
2227 'abcdgu'=~/abcdefg|cdgu/
2228 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2229 fail, which would bring use to the state representing 'd' in the second word where we would
2230 try 'g' and succeed, prodceding to match 'cdgu'.
2232 /* add a fail transition */
2233 const U32 trie_offset = ARG(source);
2234 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2236 const U32 ucharcount = trie->uniquecharcount;
2237 const U32 numstates = trie->statecount;
2238 const U32 ubound = trie->lasttrans + ucharcount;
2242 U32 base = trie->states[ 1 ].trans.base;
2245 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2246 GET_RE_DEBUG_FLAGS_DECL;
2248 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2250 PERL_UNUSED_ARG(depth);
2254 ARG_SET( stclass, data_slot );
2255 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2256 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2257 aho->trie=trie_offset;
2258 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2259 Copy( trie->states, aho->states, numstates, reg_trie_state );
2260 Newxz( q, numstates, U32);
2261 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2264 /* initialize fail[0..1] to be 1 so that we always have
2265 a valid final fail state */
2266 fail[ 0 ] = fail[ 1 ] = 1;
2268 for ( charid = 0; charid < ucharcount ; charid++ ) {
2269 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2271 q[ q_write ] = newstate;
2272 /* set to point at the root */
2273 fail[ q[ q_write++ ] ]=1;
2276 while ( q_read < q_write) {
2277 const U32 cur = q[ q_read++ % numstates ];
2278 base = trie->states[ cur ].trans.base;
2280 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2281 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2283 U32 fail_state = cur;
2286 fail_state = fail[ fail_state ];
2287 fail_base = aho->states[ fail_state ].trans.base;
2288 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2290 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2291 fail[ ch_state ] = fail_state;
2292 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2294 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2296 q[ q_write++ % numstates] = ch_state;
2300 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2301 when we fail in state 1, this allows us to use the
2302 charclass scan to find a valid start char. This is based on the principle
2303 that theres a good chance the string being searched contains lots of stuff
2304 that cant be a start char.
2306 fail[ 0 ] = fail[ 1 ] = 0;
2307 DEBUG_TRIE_COMPILE_r({
2308 PerlIO_printf(Perl_debug_log,
2309 "%*sStclass Failtable (%"UVuf" states): 0",
2310 (int)(depth * 2), "", (UV)numstates
2312 for( q_read=1; q_read<numstates; q_read++ ) {
2313 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2315 PerlIO_printf(Perl_debug_log, "\n");
2318 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2323 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2324 * These need to be revisited when a newer toolchain becomes available.
2326 #if defined(__sparc64__) && defined(__GNUC__)
2327 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2328 # undef SPARC64_GCC_WORKAROUND
2329 # define SPARC64_GCC_WORKAROUND 1
2333 #define DEBUG_PEEP(str,scan,depth) \
2334 DEBUG_OPTIMISE_r({if (scan){ \
2335 SV * const mysv=sv_newmortal(); \
2336 regnode *Next = regnext(scan); \
2337 regprop(RExC_rx, mysv, scan); \
2338 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2339 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2340 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2347 #define JOIN_EXACT(scan,min,flags) \
2348 if (PL_regkind[OP(scan)] == EXACT) \
2349 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2352 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2353 /* Merge several consecutive EXACTish nodes into one. */
2354 regnode *n = regnext(scan);
2356 regnode *next = scan + NODE_SZ_STR(scan);
2360 regnode *stop = scan;
2361 GET_RE_DEBUG_FLAGS_DECL;
2363 PERL_UNUSED_ARG(depth);
2366 PERL_ARGS_ASSERT_JOIN_EXACT;
2367 #ifndef EXPERIMENTAL_INPLACESCAN
2368 PERL_UNUSED_ARG(flags);
2369 PERL_UNUSED_ARG(val);
2371 DEBUG_PEEP("join",scan,depth);
2373 /* Skip NOTHING, merge EXACT*. */
2375 ( PL_regkind[OP(n)] == NOTHING ||
2376 (stringok && (OP(n) == OP(scan))))
2378 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2380 if (OP(n) == TAIL || n > next)
2382 if (PL_regkind[OP(n)] == NOTHING) {
2383 DEBUG_PEEP("skip:",n,depth);
2384 NEXT_OFF(scan) += NEXT_OFF(n);
2385 next = n + NODE_STEP_REGNODE;
2392 else if (stringok) {
2393 const unsigned int oldl = STR_LEN(scan);
2394 regnode * const nnext = regnext(n);
2396 DEBUG_PEEP("merg",n,depth);
2399 if (oldl + STR_LEN(n) > U8_MAX)
2401 NEXT_OFF(scan) += NEXT_OFF(n);
2402 STR_LEN(scan) += STR_LEN(n);
2403 next = n + NODE_SZ_STR(n);
2404 /* Now we can overwrite *n : */
2405 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2413 #ifdef EXPERIMENTAL_INPLACESCAN
2414 if (flags && !NEXT_OFF(n)) {
2415 DEBUG_PEEP("atch", val, depth);
2416 if (reg_off_by_arg[OP(n)]) {
2417 ARG_SET(n, val - n);
2420 NEXT_OFF(n) = val - n;
2427 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2429 Two problematic code points in Unicode casefolding of EXACT nodes:
2431 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2432 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2438 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2439 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2441 This means that in case-insensitive matching (or "loose matching",
2442 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2443 length of the above casefolded versions) can match a target string
2444 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2445 This would rather mess up the minimum length computation.
2447 What we'll do is to look for the tail four bytes, and then peek
2448 at the preceding two bytes to see whether we need to decrease
2449 the minimum length by four (six minus two).
2451 Thanks to the design of UTF-8, there cannot be false matches:
2452 A sequence of valid UTF-8 bytes cannot be a subsequence of
2453 another valid sequence of UTF-8 bytes.
2456 char * const s0 = STRING(scan), *s, *t;
2457 char * const s1 = s0 + STR_LEN(scan) - 1;
2458 char * const s2 = s1 - 4;
2459 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2460 const char t0[] = "\xaf\x49\xaf\x42";
2462 const char t0[] = "\xcc\x88\xcc\x81";
2464 const char * const t1 = t0 + 3;
2467 s < s2 && (t = ninstr(s, s1, t0, t1));
2470 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2471 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2473 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2474 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2482 n = scan + NODE_SZ_STR(scan);
2484 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2491 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2495 /* REx optimizer. Converts nodes into quickier variants "in place".
2496 Finds fixed substrings. */
2498 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2499 to the position after last scanned or to NULL. */
2501 #define INIT_AND_WITHP \
2502 assert(!and_withp); \
2503 Newx(and_withp,1,struct regnode_charclass_class); \
2504 SAVEFREEPV(and_withp)
2506 /* this is a chain of data about sub patterns we are processing that
2507 need to be handled seperately/specially in study_chunk. Its so
2508 we can simulate recursion without losing state. */
2510 typedef struct scan_frame {
2511 regnode *last; /* last node to process in this frame */
2512 regnode *next; /* next node to process when last is reached */
2513 struct scan_frame *prev; /*previous frame*/
2514 I32 stop; /* what stopparen do we use */
2518 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2520 #define CASE_SYNST_FNC(nAmE) \
2522 if (flags & SCF_DO_STCLASS_AND) { \
2523 for (value = 0; value < 256; value++) \
2524 if (!is_ ## nAmE ## _cp(value)) \
2525 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2528 for (value = 0; value < 256; value++) \
2529 if (is_ ## nAmE ## _cp(value)) \
2530 ANYOF_BITMAP_SET(data->start_class, value); \
2534 if (flags & SCF_DO_STCLASS_AND) { \
2535 for (value = 0; value < 256; value++) \
2536 if (is_ ## nAmE ## _cp(value)) \
2537 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2540 for (value = 0; value < 256; value++) \
2541 if (!is_ ## nAmE ## _cp(value)) \
2542 ANYOF_BITMAP_SET(data->start_class, value); \
2549 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2550 I32 *minlenp, I32 *deltap,
2555 struct regnode_charclass_class *and_withp,
2556 U32 flags, U32 depth)
2557 /* scanp: Start here (read-write). */
2558 /* deltap: Write maxlen-minlen here. */
2559 /* last: Stop before this one. */
2560 /* data: string data about the pattern */
2561 /* stopparen: treat close N as END */
2562 /* recursed: which subroutines have we recursed into */
2563 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2566 I32 min = 0, pars = 0, code;
2567 regnode *scan = *scanp, *next;
2569 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2570 int is_inf_internal = 0; /* The studied chunk is infinite */
2571 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2572 scan_data_t data_fake;
2573 SV *re_trie_maxbuff = NULL;
2574 regnode *first_non_open = scan;
2575 I32 stopmin = I32_MAX;
2576 scan_frame *frame = NULL;
2577 GET_RE_DEBUG_FLAGS_DECL;
2579 PERL_ARGS_ASSERT_STUDY_CHUNK;
2582 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2586 while (first_non_open && OP(first_non_open) == OPEN)
2587 first_non_open=regnext(first_non_open);
2592 while ( scan && OP(scan) != END && scan < last ){
2593 /* Peephole optimizer: */
2594 DEBUG_STUDYDATA("Peep:", data,depth);
2595 DEBUG_PEEP("Peep",scan,depth);
2596 JOIN_EXACT(scan,&min,0);
2598 /* Follow the next-chain of the current node and optimize
2599 away all the NOTHINGs from it. */
2600 if (OP(scan) != CURLYX) {
2601 const int max = (reg_off_by_arg[OP(scan)]
2603 /* I32 may be smaller than U16 on CRAYs! */
2604 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2605 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2609 /* Skip NOTHING and LONGJMP. */
2610 while ((n = regnext(n))
2611 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2612 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2613 && off + noff < max)
2615 if (reg_off_by_arg[OP(scan)])
2618 NEXT_OFF(scan) = off;
2623 /* The principal pseudo-switch. Cannot be a switch, since we
2624 look into several different things. */
2625 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2626 || OP(scan) == IFTHEN) {
2627 next = regnext(scan);
2629 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2631 if (OP(next) == code || code == IFTHEN) {
2632 /* NOTE - There is similar code to this block below for handling
2633 TRIE nodes on a re-study. If you change stuff here check there
2635 I32 max1 = 0, min1 = I32_MAX, num = 0;
2636 struct regnode_charclass_class accum;
2637 regnode * const startbranch=scan;
2639 if (flags & SCF_DO_SUBSTR)
2640 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2641 if (flags & SCF_DO_STCLASS)
2642 cl_init_zero(pRExC_state, &accum);
2644 while (OP(scan) == code) {
2645 I32 deltanext, minnext, f = 0, fake;
2646 struct regnode_charclass_class this_class;
2649 data_fake.flags = 0;
2651 data_fake.whilem_c = data->whilem_c;
2652 data_fake.last_closep = data->last_closep;
2655 data_fake.last_closep = &fake;
2657 data_fake.pos_delta = delta;
2658 next = regnext(scan);
2659 scan = NEXTOPER(scan);
2661 scan = NEXTOPER(scan);
2662 if (flags & SCF_DO_STCLASS) {
2663 cl_init(pRExC_state, &this_class);
2664 data_fake.start_class = &this_class;
2665 f = SCF_DO_STCLASS_AND;
2667 if (flags & SCF_WHILEM_VISITED_POS)
2668 f |= SCF_WHILEM_VISITED_POS;
2670 /* we suppose the run is continuous, last=next...*/
2671 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2673 stopparen, recursed, NULL, f,depth+1);
2676 if (max1 < minnext + deltanext)
2677 max1 = minnext + deltanext;
2678 if (deltanext == I32_MAX)
2679 is_inf = is_inf_internal = 1;
2681 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2683 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2684 if ( stopmin > minnext)
2685 stopmin = min + min1;
2686 flags &= ~SCF_DO_SUBSTR;
2688 data->flags |= SCF_SEEN_ACCEPT;
2691 if (data_fake.flags & SF_HAS_EVAL)
2692 data->flags |= SF_HAS_EVAL;
2693 data->whilem_c = data_fake.whilem_c;
2695 if (flags & SCF_DO_STCLASS)
2696 cl_or(pRExC_state, &accum, &this_class);
2698 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2700 if (flags & SCF_DO_SUBSTR) {
2701 data->pos_min += min1;
2702 data->pos_delta += max1 - min1;
2703 if (max1 != min1 || is_inf)
2704 data->longest = &(data->longest_float);
2707 delta += max1 - min1;
2708 if (flags & SCF_DO_STCLASS_OR) {
2709 cl_or(pRExC_state, data->start_class, &accum);
2711 cl_and(data->start_class, and_withp);
2712 flags &= ~SCF_DO_STCLASS;
2715 else if (flags & SCF_DO_STCLASS_AND) {
2717 cl_and(data->start_class, &accum);
2718 flags &= ~SCF_DO_STCLASS;
2721 /* Switch to OR mode: cache the old value of
2722 * data->start_class */
2724 StructCopy(data->start_class, and_withp,
2725 struct regnode_charclass_class);
2726 flags &= ~SCF_DO_STCLASS_AND;
2727 StructCopy(&accum, data->start_class,
2728 struct regnode_charclass_class);
2729 flags |= SCF_DO_STCLASS_OR;
2730 data->start_class->flags |= ANYOF_EOS;
2734 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2737 Assuming this was/is a branch we are dealing with: 'scan' now
2738 points at the item that follows the branch sequence, whatever
2739 it is. We now start at the beginning of the sequence and look
2746 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2748 If we can find such a subseqence we need to turn the first
2749 element into a trie and then add the subsequent branch exact
2750 strings to the trie.
2754 1. patterns where the whole set of branch can be converted.
2756 2. patterns where only a subset can be converted.
2758 In case 1 we can replace the whole set with a single regop
2759 for the trie. In case 2 we need to keep the start and end
2762 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2763 becomes BRANCH TRIE; BRANCH X;
2765 There is an additional case, that being where there is a
2766 common prefix, which gets split out into an EXACT like node
2767 preceding the TRIE node.
2769 If x(1..n)==tail then we can do a simple trie, if not we make
2770 a "jump" trie, such that when we match the appropriate word
2771 we "jump" to the appopriate tail node. Essentailly we turn
2772 a nested if into a case structure of sorts.
2777 if (!re_trie_maxbuff) {
2778 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2779 if (!SvIOK(re_trie_maxbuff))
2780 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2782 if ( SvIV(re_trie_maxbuff)>=0 ) {
2784 regnode *first = (regnode *)NULL;
2785 regnode *last = (regnode *)NULL;
2786 regnode *tail = scan;
2791 SV * const mysv = sv_newmortal(); /* for dumping */
2793 /* var tail is used because there may be a TAIL
2794 regop in the way. Ie, the exacts will point to the
2795 thing following the TAIL, but the last branch will
2796 point at the TAIL. So we advance tail. If we
2797 have nested (?:) we may have to move through several
2801 while ( OP( tail ) == TAIL ) {
2802 /* this is the TAIL generated by (?:) */
2803 tail = regnext( tail );
2808 regprop(RExC_rx, mysv, tail );
2809 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2810 (int)depth * 2 + 2, "",
2811 "Looking for TRIE'able sequences. Tail node is: ",
2812 SvPV_nolen_const( mysv )
2818 step through the branches, cur represents each
2819 branch, noper is the first thing to be matched
2820 as part of that branch and noper_next is the
2821 regnext() of that node. if noper is an EXACT
2822 and noper_next is the same as scan (our current
2823 position in the regex) then the EXACT branch is
2824 a possible optimization target. Once we have
2825 two or more consequetive such branches we can
2826 create a trie of the EXACT's contents and stich
2827 it in place. If the sequence represents all of
2828 the branches we eliminate the whole thing and
2829 replace it with a single TRIE. If it is a
2830 subsequence then we need to stitch it in. This
2831 means the first branch has to remain, and needs
2832 to be repointed at the item on the branch chain
2833 following the last branch optimized. This could
2834 be either a BRANCH, in which case the
2835 subsequence is internal, or it could be the
2836 item following the branch sequence in which
2837 case the subsequence is at the end.
2841 /* dont use tail as the end marker for this traverse */
2842 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2843 regnode * const noper = NEXTOPER( cur );
2844 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2845 regnode * const noper_next = regnext( noper );
2849 regprop(RExC_rx, mysv, cur);
2850 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2851 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2853 regprop(RExC_rx, mysv, noper);
2854 PerlIO_printf( Perl_debug_log, " -> %s",
2855 SvPV_nolen_const(mysv));
2858 regprop(RExC_rx, mysv, noper_next );
2859 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2860 SvPV_nolen_const(mysv));
2862 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2863 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2865 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2866 : PL_regkind[ OP( noper ) ] == EXACT )
2867 || OP(noper) == NOTHING )
2869 && noper_next == tail
2874 if ( !first || optype == NOTHING ) {
2875 if (!first) first = cur;
2876 optype = OP( noper );
2882 Currently we do not believe that the trie logic can
2883 handle case insensitive matching properly when the
2884 pattern is not unicode (thus forcing unicode semantics).
2886 If/when this is fixed the following define can be swapped
2887 in below to fully enable trie logic.
2889 #define TRIE_TYPE_IS_SAFE 1
2892 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2894 if ( last && TRIE_TYPE_IS_SAFE ) {
2895 make_trie( pRExC_state,
2896 startbranch, first, cur, tail, count,
2899 if ( PL_regkind[ OP( noper ) ] == EXACT
2901 && noper_next == tail
2906 optype = OP( noper );
2916 regprop(RExC_rx, mysv, cur);
2917 PerlIO_printf( Perl_debug_log,
2918 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2919 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2923 if ( last && TRIE_TYPE_IS_SAFE ) {
2924 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2925 #ifdef TRIE_STUDY_OPT
2926 if ( ((made == MADE_EXACT_TRIE &&
2927 startbranch == first)
2928 || ( first_non_open == first )) &&
2930 flags |= SCF_TRIE_RESTUDY;
2931 if ( startbranch == first
2934 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2944 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2945 scan = NEXTOPER(NEXTOPER(scan));
2946 } else /* single branch is optimized. */
2947 scan = NEXTOPER(scan);
2949 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2950 scan_frame *newframe = NULL;
2955 if (OP(scan) != SUSPEND) {
2956 /* set the pointer */
2957 if (OP(scan) == GOSUB) {
2959 RExC_recurse[ARG2L(scan)] = scan;
2960 start = RExC_open_parens[paren-1];
2961 end = RExC_close_parens[paren-1];
2964 start = RExC_rxi->program + 1;
2968 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2969 SAVEFREEPV(recursed);
2971 if (!PAREN_TEST(recursed,paren+1)) {
2972 PAREN_SET(recursed,paren+1);
2973 Newx(newframe,1,scan_frame);
2975 if (flags & SCF_DO_SUBSTR) {
2976 SCAN_COMMIT(pRExC_state,data,minlenp);
2977 data->longest = &(data->longest_float);
2979 is_inf = is_inf_internal = 1;
2980 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2981 cl_anything(pRExC_state, data->start_class);
2982 flags &= ~SCF_DO_STCLASS;
2985 Newx(newframe,1,scan_frame);
2988 end = regnext(scan);
2993 SAVEFREEPV(newframe);
2994 newframe->next = regnext(scan);
2995 newframe->last = last;
2996 newframe->stop = stopparen;
2997 newframe->prev = frame;
3007 else if (OP(scan) == EXACT) {
3008 I32 l = STR_LEN(scan);
3011 const U8 * const s = (U8*)STRING(scan);
3012 l = utf8_length(s, s + l);
3013 uc = utf8_to_uvchr(s, NULL);
3015 uc = *((U8*)STRING(scan));
3018 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3019 /* The code below prefers earlier match for fixed
3020 offset, later match for variable offset. */
3021 if (data->last_end == -1) { /* Update the start info. */
3022 data->last_start_min = data->pos_min;
3023 data->last_start_max = is_inf
3024 ? I32_MAX : data->pos_min + data->pos_delta;
3026 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3028 SvUTF8_on(data->last_found);
3030 SV * const sv = data->last_found;
3031 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3032 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3033 if (mg && mg->mg_len >= 0)
3034 mg->mg_len += utf8_length((U8*)STRING(scan),
3035 (U8*)STRING(scan)+STR_LEN(scan));
3037 data->last_end = data->pos_min + l;
3038 data->pos_min += l; /* As in the first entry. */
3039 data->flags &= ~SF_BEFORE_EOL;
3041 if (flags & SCF_DO_STCLASS_AND) {
3042 /* Check whether it is compatible with what we know already! */
3046 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3047 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3048 && (!(data->start_class->flags & ANYOF_FOLD)
3049 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3052 ANYOF_CLASS_ZERO(data->start_class);
3053 ANYOF_BITMAP_ZERO(data->start_class);
3055 ANYOF_BITMAP_SET(data->start_class, uc);
3056 data->start_class->flags &= ~ANYOF_EOS;
3058 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3060 else if (flags & SCF_DO_STCLASS_OR) {
3061 /* false positive possible if the class is case-folded */
3063 ANYOF_BITMAP_SET(data->start_class, uc);
3065 data->start_class->flags |= ANYOF_UNICODE_ALL;
3066 data->start_class->flags &= ~ANYOF_EOS;
3067 cl_and(data->start_class, and_withp);
3069 flags &= ~SCF_DO_STCLASS;
3071 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3072 I32 l = STR_LEN(scan);
3073 UV uc = *((U8*)STRING(scan));
3075 /* Search for fixed substrings supports EXACT only. */
3076 if (flags & SCF_DO_SUBSTR) {
3078 SCAN_COMMIT(pRExC_state, data, minlenp);
3081 const U8 * const s = (U8 *)STRING(scan);
3082 l = utf8_length(s, s + l);
3083 uc = utf8_to_uvchr(s, NULL);
3086 if (flags & SCF_DO_SUBSTR)
3088 if (flags & SCF_DO_STCLASS_AND) {
3089 /* Check whether it is compatible with what we know already! */
3093 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3094 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3095 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3097 ANYOF_CLASS_ZERO(data->start_class);
3098 ANYOF_BITMAP_ZERO(data->start_class);
3100 ANYOF_BITMAP_SET(data->start_class, uc);
3101 data->start_class->flags &= ~ANYOF_EOS;
3102 data->start_class->flags |= ANYOF_FOLD;
3103 if (OP(scan) == EXACTFL)
3104 data->start_class->flags |= ANYOF_LOCALE;
3107 else if (flags & SCF_DO_STCLASS_OR) {
3108 if (data->start_class->flags & ANYOF_FOLD) {
3109 /* false positive possible if the class is case-folded.
3110 Assume that the locale settings are the same... */
3112 ANYOF_BITMAP_SET(data->start_class, uc);
3113 data->start_class->flags &= ~ANYOF_EOS;
3115 cl_and(data->start_class, and_withp);
3117 flags &= ~SCF_DO_STCLASS;
3119 else if (REGNODE_VARIES(OP(scan))) {
3120 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3121 I32 f = flags, pos_before = 0;
3122 regnode * const oscan = scan;
3123 struct regnode_charclass_class this_class;
3124 struct regnode_charclass_class *oclass = NULL;
3125 I32 next_is_eval = 0;
3127 switch (PL_regkind[OP(scan)]) {
3128 case WHILEM: /* End of (?:...)* . */
3129 scan = NEXTOPER(scan);
3132 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3133 next = NEXTOPER(scan);
3134 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3136 maxcount = REG_INFTY;
3137 next = regnext(scan);
3138 scan = NEXTOPER(scan);
3142 if (flags & SCF_DO_SUBSTR)
3147 if (flags & SCF_DO_STCLASS) {
3149 maxcount = REG_INFTY;
3150 next = regnext(scan);
3151 scan = NEXTOPER(scan);
3154 is_inf = is_inf_internal = 1;
3155 scan = regnext(scan);
3156 if (flags & SCF_DO_SUBSTR) {
3157 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3158 data->longest = &(data->longest_float);
3160 goto optimize_curly_tail;
3162 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3163 && (scan->flags == stopparen))
3168 mincount = ARG1(scan);
3169 maxcount = ARG2(scan);
3171 next = regnext(scan);
3172 if (OP(scan) == CURLYX) {
3173 I32 lp = (data ? *(data->last_closep) : 0);
3174 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3176 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3177 next_is_eval = (OP(scan) == EVAL);
3179 if (flags & SCF_DO_SUBSTR) {
3180 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3181 pos_before = data->pos_min;
3185 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3187 data->flags |= SF_IS_INF;
3189 if (flags & SCF_DO_STCLASS) {
3190 cl_init(pRExC_state, &this_class);
3191 oclass = data->start_class;
3192 data->start_class = &this_class;
3193 f |= SCF_DO_STCLASS_AND;
3194 f &= ~SCF_DO_STCLASS_OR;
3196 /* These are the cases when once a subexpression
3197 fails at a particular position, it cannot succeed
3198 even after backtracking at the enclosing scope.
3200 XXXX what if minimal match and we are at the
3201 initial run of {n,m}? */
3202 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3203 f &= ~SCF_WHILEM_VISITED_POS;
3205 /* This will finish on WHILEM, setting scan, or on NULL: */
3206 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3207 last, data, stopparen, recursed, NULL,
3209 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3211 if (flags & SCF_DO_STCLASS)
3212 data->start_class = oclass;
3213 if (mincount == 0 || minnext == 0) {
3214 if (flags & SCF_DO_STCLASS_OR) {
3215 cl_or(pRExC_state, data->start_class, &this_class);
3217 else if (flags & SCF_DO_STCLASS_AND) {
3218 /* Switch to OR mode: cache the old value of
3219 * data->start_class */
3221 StructCopy(data->start_class, and_withp,
3222 struct regnode_charclass_class);
3223 flags &= ~SCF_DO_STCLASS_AND;
3224 StructCopy(&this_class, data->start_class,
3225 struct regnode_charclass_class);
3226 flags |= SCF_DO_STCLASS_OR;
3227 data->start_class->flags |= ANYOF_EOS;
3229 } else { /* Non-zero len */
3230 if (flags & SCF_DO_STCLASS_OR) {
3231 cl_or(pRExC_state, data->start_class, &this_class);
3232 cl_and(data->start_class, and_withp);
3234 else if (flags & SCF_DO_STCLASS_AND)
3235 cl_and(data->start_class, &this_class);
3236 flags &= ~SCF_DO_STCLASS;
3238 if (!scan) /* It was not CURLYX, but CURLY. */
3240 if ( /* ? quantifier ok, except for (?{ ... }) */
3241 (next_is_eval || !(mincount == 0 && maxcount == 1))
3242 && (minnext == 0) && (deltanext == 0)
3243 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3244 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3246 ckWARNreg(RExC_parse,
3247 "Quantifier unexpected on zero-length expression");
3250 min += minnext * mincount;
3251 is_inf_internal |= ((maxcount == REG_INFTY
3252 && (minnext + deltanext) > 0)
3253 || deltanext == I32_MAX);
3254 is_inf |= is_inf_internal;
3255 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3257 /* Try powerful optimization CURLYX => CURLYN. */
3258 if ( OP(oscan) == CURLYX && data
3259 && data->flags & SF_IN_PAR
3260 && !(data->flags & SF_HAS_EVAL)
3261 && !deltanext && minnext == 1 ) {
3262 /* Try to optimize to CURLYN. */
3263 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3264 regnode * const nxt1 = nxt;
3271 if (!REGNODE_SIMPLE(OP(nxt))
3272 && !(PL_regkind[OP(nxt)] == EXACT
3273 && STR_LEN(nxt) == 1))
3279 if (OP(nxt) != CLOSE)
3281 if (RExC_open_parens) {
3282 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3283 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3285 /* Now we know that nxt2 is the only contents: */
3286 oscan->flags = (U8)ARG(nxt);
3288 OP(nxt1) = NOTHING; /* was OPEN. */
3291 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3292 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3293 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3294 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3295 OP(nxt + 1) = OPTIMIZED; /* was count. */
3296 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
3301 /* Try optimization CURLYX => CURLYM. */
3302 if ( OP(oscan) == CURLYX && data
3303 && !(data->flags & SF_HAS_PAR)
3304 && !(data->flags & SF_HAS_EVAL)
3305 && !deltanext /* atom is fixed width */
3306 && minnext != 0 /* CURLYM can't handle zero width */
3308 /* XXXX How to optimize if data == 0? */
3309 /* Optimize to a simpler form. */
3310 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3314 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3315 && (OP(nxt2) != WHILEM))
3317 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3318 /* Need to optimize away parenths. */
3319 if (data->flags & SF_IN_PAR) {
3320 /* Set the parenth number. */
3321 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3323 if (OP(nxt) != CLOSE)
3324 FAIL("Panic opt close");
3325 oscan->flags = (U8)ARG(nxt);
3326 if (RExC_open_parens) {
3327 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3328 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3330 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3331 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3334 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3335 OP(nxt + 1) = OPTIMIZED; /* was count. */
3336 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3337 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3340 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3341 regnode *nnxt = regnext(nxt1);
3344 if (reg_off_by_arg[OP(nxt1)])
3345 ARG_SET(nxt1, nxt2 - nxt1);
3346 else if (nxt2 - nxt1 < U16_MAX)
3347 NEXT_OFF(nxt1) = nxt2 - nxt1;
3349 OP(nxt) = NOTHING; /* Cannot beautify */
3354 /* Optimize again: */
3355 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3356 NULL, stopparen, recursed, NULL, 0,depth+1);
3361 else if ((OP(oscan) == CURLYX)
3362 && (flags & SCF_WHILEM_VISITED_POS)
3363 /* See the comment on a similar expression above.
3364 However, this time it not a subexpression
3365 we care about, but the expression itself. */
3366 && (maxcount == REG_INFTY)
3367 && data && ++data->whilem_c < 16) {
3368 /* This stays as CURLYX, we can put the count/of pair. */
3369 /* Find WHILEM (as in regexec.c) */
3370 regnode *nxt = oscan + NEXT_OFF(oscan);
3372 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3374 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3375 | (RExC_whilem_seen << 4)); /* On WHILEM */
3377 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3379 if (flags & SCF_DO_SUBSTR) {
3380 SV *last_str = NULL;
3381 int counted = mincount != 0;
3383 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3384 #if defined(SPARC64_GCC_WORKAROUND)
3387 const char *s = NULL;
3390 if (pos_before >= data->last_start_min)
3393 b = data->last_start_min;
3396 s = SvPV_const(data->last_found, l);
3397 old = b - data->last_start_min;
3400 I32 b = pos_before >= data->last_start_min
3401 ? pos_before : data->last_start_min;
3403 const char * const s = SvPV_const(data->last_found, l);
3404 I32 old = b - data->last_start_min;
3408 old = utf8_hop((U8*)s, old) - (U8*)s;
3411 /* Get the added string: */
3412 last_str = newSVpvn_utf8(s + old, l, UTF);
3413 if (deltanext == 0 && pos_before == b) {
3414 /* What was added is a constant string */
3416 SvGROW(last_str, (mincount * l) + 1);
3417 repeatcpy(SvPVX(last_str) + l,
3418 SvPVX_const(last_str), l, mincount - 1);
3419 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3420 /* Add additional parts. */
3421 SvCUR_set(data->last_found,
3422 SvCUR(data->last_found) - l);
3423 sv_catsv(data->last_found, last_str);
3425 SV * sv = data->last_found;
3427 SvUTF8(sv) && SvMAGICAL(sv) ?
3428 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3429 if (mg && mg->mg_len >= 0)
3430 mg->mg_len += CHR_SVLEN(last_str) - l;
3432 data->last_end += l * (mincount - 1);
3435 /* start offset must point into the last copy */
3436 data->last_start_min += minnext * (mincount - 1);
3437 data->last_start_max += is_inf ? I32_MAX
3438 : (maxcount - 1) * (minnext + data->pos_delta);
3441 /* It is counted once already... */
3442 data->pos_min += minnext * (mincount - counted);
3443 data->pos_delta += - counted * deltanext +
3444 (minnext + deltanext) * maxcount - minnext * mincount;
3445 if (mincount != maxcount) {
3446 /* Cannot extend fixed substrings found inside
3448 SCAN_COMMIT(pRExC_state,data,minlenp);
3449 if (mincount && last_str) {
3450 SV * const sv = data->last_found;
3451 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3452 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3456 sv_setsv(sv, last_str);
3457 data->last_end = data->pos_min;
3458 data->last_start_min =
3459 data->pos_min - CHR_SVLEN(last_str);
3460 data->last_start_max = is_inf
3462 : data->pos_min + data->pos_delta
3463 - CHR_SVLEN(last_str);
3465 data->longest = &(data->longest_float);
3467 SvREFCNT_dec(last_str);
3469 if (data && (fl & SF_HAS_EVAL))
3470 data->flags |= SF_HAS_EVAL;
3471 optimize_curly_tail:
3472 if (OP(oscan) != CURLYX) {
3473 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3475 NEXT_OFF(oscan) += NEXT_OFF(next);
3478 default: /* REF and CLUMP only? */
3479 if (flags & SCF_DO_SUBSTR) {
3480 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3481 data->longest = &(data->longest_float);
3483 is_inf = is_inf_internal = 1;
3484 if (flags & SCF_DO_STCLASS_OR)
3485 cl_anything(pRExC_state, data->start_class);
3486 flags &= ~SCF_DO_STCLASS;
3490 else if (OP(scan) == LNBREAK) {
3491 if (flags & SCF_DO_STCLASS) {
3493 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3494 if (flags & SCF_DO_STCLASS_AND) {
3495 for (value = 0; value < 256; value++)
3496 if (!is_VERTWS_cp(value))
3497 ANYOF_BITMAP_CLEAR(data->start_class, value);
3500 for (value = 0; value < 256; value++)
3501 if (is_VERTWS_cp(value))
3502 ANYOF_BITMAP_SET(data->start_class, value);
3504 if (flags & SCF_DO_STCLASS_OR)
3505 cl_and(data->start_class, and_withp);
3506 flags &= ~SCF_DO_STCLASS;
3510 if (flags & SCF_DO_SUBSTR) {
3511 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3513 data->pos_delta += 1;
3514 data->longest = &(data->longest_float);
3518 else if (OP(scan) == FOLDCHAR) {
3519 int d = ARG(scan)==0xDF ? 1 : 2;
3520 flags &= ~SCF_DO_STCLASS;
3523 if (flags & SCF_DO_SUBSTR) {
3524 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3526 data->pos_delta += d;
3527 data->longest = &(data->longest_float);
3530 else if (REGNODE_SIMPLE(OP(scan))) {
3533 if (flags & SCF_DO_SUBSTR) {
3534 SCAN_COMMIT(pRExC_state,data,minlenp);
3538 if (flags & SCF_DO_STCLASS) {
3539 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3541 /* Some of the logic below assumes that switching
3542 locale on will only add false positives. */
3543 switch (PL_regkind[OP(scan)]) {
3547 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3548 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3549 cl_anything(pRExC_state, data->start_class);
3552 if (OP(scan) == SANY)
3554 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3555 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3556 || (data->start_class->flags & ANYOF_CLASS));
3557 cl_anything(pRExC_state, data->start_class);
3559 if (flags & SCF_DO_STCLASS_AND || !value)
3560 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3563 if (flags & SCF_DO_STCLASS_AND)
3564 cl_and(data->start_class,
3565 (struct regnode_charclass_class*)scan);
3567 cl_or(pRExC_state, data->start_class,
3568 (struct regnode_charclass_class*)scan);
3571 if (flags & SCF_DO_STCLASS_AND) {
3572 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3573 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3574 for (value = 0; value < 256; value++)
3575 if (!isALNUM(value))
3576 ANYOF_BITMAP_CLEAR(data->start_class, value);
3580 if (data->start_class->flags & ANYOF_LOCALE)
3581 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3583 for (value = 0; value < 256; value++)
3585 ANYOF_BITMAP_SET(data->start_class, value);
3590 if (flags & SCF_DO_STCLASS_AND) {
3591 if (data->start_class->flags & ANYOF_LOCALE)
3592 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3595 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3596 data->start_class->flags |= ANYOF_LOCALE;
3600 if (flags & SCF_DO_STCLASS_AND) {
3601 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3602 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3603 for (value = 0; value < 256; value++)
3605 ANYOF_BITMAP_CLEAR(data->start_class, value);
3609 if (data->start_class->flags & ANYOF_LOCALE)
3610 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3612 for (value = 0; value < 256; value++)
3613 if (!isALNUM(value))
3614 ANYOF_BITMAP_SET(data->start_class, value);
3619 if (flags & SCF_DO_STCLASS_AND) {
3620 if (data->start_class->flags & ANYOF_LOCALE)
3621 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3624 data->start_class->flags |= ANYOF_LOCALE;
3625 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3629 if (flags & SCF_DO_STCLASS_AND) {
3630 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3631 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3632 for (value = 0; value < 256; value++)
3633 if (!isSPACE(value))
3634 ANYOF_BITMAP_CLEAR(data->start_class, value);
3638 if (data->start_class->flags & ANYOF_LOCALE)
3639 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3641 for (value = 0; value < 256; value++)
3643 ANYOF_BITMAP_SET(data->start_class, value);
3648 if (flags & SCF_DO_STCLASS_AND) {
3649 if (data->start_class->flags & ANYOF_LOCALE)
3650 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3653 data->start_class->flags |= ANYOF_LOCALE;
3654 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3658 if (flags & SCF_DO_STCLASS_AND) {
3659 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3660 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3661 for (value = 0; value < 256; value++)
3663 ANYOF_BITMAP_CLEAR(data->start_class, value);
3667 if (data->start_class->flags & ANYOF_LOCALE)
3668 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3670 for (value = 0; value < 256; value++)
3671 if (!isSPACE(value))
3672 ANYOF_BITMAP_SET(data->start_class, value);
3677 if (flags & SCF_DO_STCLASS_AND) {
3678 if (data->start_class->flags & ANYOF_LOCALE) {
3679 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3680 for (value = 0; value < 256; value++)
3681 if (!isSPACE(value))
3682 ANYOF_BITMAP_CLEAR(data->start_class, value);
3686 data->start_class->flags |= ANYOF_LOCALE;
3687 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3691 if (flags & SCF_DO_STCLASS_AND) {
3692 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3693 for (value = 0; value < 256; value++)
3694 if (!isDIGIT(value))
3695 ANYOF_BITMAP_CLEAR(data->start_class, value);
3698 if (data->start_class->flags & ANYOF_LOCALE)
3699 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3701 for (value = 0; value < 256; value++)
3703 ANYOF_BITMAP_SET(data->start_class, value);
3708 if (flags & SCF_DO_STCLASS_AND) {
3709 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3710 for (value = 0; value < 256; value++)
3712 ANYOF_BITMAP_CLEAR(data->start_class, value);
3715 if (data->start_class->flags & ANYOF_LOCALE)
3716 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3718 for (value = 0; value < 256; value++)
3719 if (!isDIGIT(value))
3720 ANYOF_BITMAP_SET(data->start_class, value);
3724 CASE_SYNST_FNC(VERTWS);
3725 CASE_SYNST_FNC(HORIZWS);
3728 if (flags & SCF_DO_STCLASS_OR)
3729 cl_and(data->start_class, and_withp);
3730 flags &= ~SCF_DO_STCLASS;
3733 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3734 data->flags |= (OP(scan) == MEOL
3738 else if ( PL_regkind[OP(scan)] == BRANCHJ
3739 /* Lookbehind, or need to calculate parens/evals/stclass: */
3740 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3741 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3742 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3743 || OP(scan) == UNLESSM )
3745 /* Negative Lookahead/lookbehind
3746 In this case we can't do fixed string optimisation.
3749 I32 deltanext, minnext, fake = 0;
3751 struct regnode_charclass_class intrnl;
3754 data_fake.flags = 0;
3756 data_fake.whilem_c = data->whilem_c;
3757 data_fake.last_closep = data->last_closep;
3760 data_fake.last_closep = &fake;
3761 data_fake.pos_delta = delta;
3762 if ( flags & SCF_DO_STCLASS && !scan->flags
3763 && OP(scan) == IFMATCH ) { /* Lookahead */
3764 cl_init(pRExC_state, &intrnl);
3765 data_fake.start_class = &intrnl;
3766 f |= SCF_DO_STCLASS_AND;
3768 if (flags & SCF_WHILEM_VISITED_POS)
3769 f |= SCF_WHILEM_VISITED_POS;
3770 next = regnext(scan);
3771 nscan = NEXTOPER(NEXTOPER(scan));
3772 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3773 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3776 FAIL("Variable length lookbehind not implemented");
3778 else if (minnext > (I32)U8_MAX) {
3779 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3781 scan->flags = (U8)minnext;
3784 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3786 if (data_fake.flags & SF_HAS_EVAL)
3787 data->flags |= SF_HAS_EVAL;
3788 data->whilem_c = data_fake.whilem_c;
3790 if (f & SCF_DO_STCLASS_AND) {
3791 if (flags & SCF_DO_STCLASS_OR) {
3792 /* OR before, AND after: ideally we would recurse with
3793 * data_fake to get the AND applied by study of the
3794 * remainder of the pattern, and then derecurse;
3795 * *** HACK *** for now just treat as "no information".
3796 * See [perl #56690].
3798 cl_init(pRExC_state, data->start_class);
3800 /* AND before and after: combine and continue */
3801 const int was = (data->start_class->flags & ANYOF_EOS);
3803 cl_and(data->start_class, &intrnl);
3805 data->start_class->flags |= ANYOF_EOS;
3809 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3811 /* Positive Lookahead/lookbehind
3812 In this case we can do fixed string optimisation,
3813 but we must be careful about it. Note in the case of
3814 lookbehind the positions will be offset by the minimum
3815 length of the pattern, something we won't know about
3816 until after the recurse.
3818 I32 deltanext, fake = 0;
3820 struct regnode_charclass_class intrnl;
3822 /* We use SAVEFREEPV so that when the full compile
3823 is finished perl will clean up the allocated
3824 minlens when its all done. This was we don't
3825 have to worry about freeing them when we know
3826 they wont be used, which would be a pain.
3829 Newx( minnextp, 1, I32 );
3830 SAVEFREEPV(minnextp);
3833 StructCopy(data, &data_fake, scan_data_t);
3834 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3837 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3838 data_fake.last_found=newSVsv(data->last_found);
3842 data_fake.last_closep = &fake;
3843 data_fake.flags = 0;
3844 data_fake.pos_delta = delta;
3846 data_fake.flags |= SF_IS_INF;
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));
3858 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3859 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3862 FAIL("Variable length lookbehind not implemented");
3864 else if (*minnextp > (I32)U8_MAX) {
3865 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3867 scan->flags = (U8)*minnextp;
3872 if (f & SCF_DO_STCLASS_AND) {
3873 const int was = (data->start_class->flags & ANYOF_EOS);
3875 cl_and(data->start_class, &intrnl);
3877 data->start_class->flags |= ANYOF_EOS;
3880 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3882 if (data_fake.flags & SF_HAS_EVAL)
3883 data->flags |= SF_HAS_EVAL;
3884 data->whilem_c = data_fake.whilem_c;
3885 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3886 if (RExC_rx->minlen<*minnextp)
3887 RExC_rx->minlen=*minnextp;
3888 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3889 SvREFCNT_dec(data_fake.last_found);
3891 if ( data_fake.minlen_fixed != minlenp )
3893 data->offset_fixed= data_fake.offset_fixed;
3894 data->minlen_fixed= data_fake.minlen_fixed;
3895 data->lookbehind_fixed+= scan->flags;
3897 if ( data_fake.minlen_float != minlenp )
3899 data->minlen_float= data_fake.minlen_float;
3900 data->offset_float_min=data_fake.offset_float_min;
3901 data->offset_float_max=data_fake.offset_float_max;
3902 data->lookbehind_float+= scan->flags;
3911 else if (OP(scan) == OPEN) {
3912 if (stopparen != (I32)ARG(scan))
3915 else if (OP(scan) == CLOSE) {
3916 if (stopparen == (I32)ARG(scan)) {
3919 if ((I32)ARG(scan) == is_par) {
3920 next = regnext(scan);
3922 if ( next && (OP(next) != WHILEM) && next < last)
3923 is_par = 0; /* Disable optimization */
3926 *(data->last_closep) = ARG(scan);
3928 else if (OP(scan) == EVAL) {
3930 data->flags |= SF_HAS_EVAL;
3932 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3933 if (flags & SCF_DO_SUBSTR) {
3934 SCAN_COMMIT(pRExC_state,data,minlenp);
3935 flags &= ~SCF_DO_SUBSTR;
3937 if (data && OP(scan)==ACCEPT) {
3938 data->flags |= SCF_SEEN_ACCEPT;
3943 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3945 if (flags & SCF_DO_SUBSTR) {
3946 SCAN_COMMIT(pRExC_state,data,minlenp);
3947 data->longest = &(data->longest_float);
3949 is_inf = is_inf_internal = 1;
3950 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3951 cl_anything(pRExC_state, data->start_class);
3952 flags &= ~SCF_DO_STCLASS;
3954 else if (OP(scan) == GPOS) {
3955 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3956 !(delta || is_inf || (data && data->pos_delta)))
3958 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3959 RExC_rx->extflags |= RXf_ANCH_GPOS;
3960 if (RExC_rx->gofs < (U32)min)
3961 RExC_rx->gofs = min;
3963 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3967 #ifdef TRIE_STUDY_OPT
3968 #ifdef FULL_TRIE_STUDY
3969 else if (PL_regkind[OP(scan)] == TRIE) {
3970 /* NOTE - There is similar code to this block above for handling
3971 BRANCH nodes on the initial study. If you change stuff here
3973 regnode *trie_node= scan;
3974 regnode *tail= regnext(scan);
3975 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3976 I32 max1 = 0, min1 = I32_MAX;
3977 struct regnode_charclass_class accum;
3979 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3980 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3981 if (flags & SCF_DO_STCLASS)
3982 cl_init_zero(pRExC_state, &accum);
3988 const regnode *nextbranch= NULL;
3991 for ( word=1 ; word <= trie->wordcount ; word++)
3993 I32 deltanext=0, minnext=0, f = 0, fake;
3994 struct regnode_charclass_class this_class;
3996 data_fake.flags = 0;
3998 data_fake.whilem_c = data->whilem_c;
3999 data_fake.last_closep = data->last_closep;
4002 data_fake.last_closep = &fake;
4003 data_fake.pos_delta = delta;
4004 if (flags & SCF_DO_STCLASS) {
4005 cl_init(pRExC_state, &this_class);
4006 data_fake.start_class = &this_class;
4007 f = SCF_DO_STCLASS_AND;
4009 if (flags & SCF_WHILEM_VISITED_POS)
4010 f |= SCF_WHILEM_VISITED_POS;
4012 if (trie->jump[word]) {
4014 nextbranch = trie_node + trie->jump[0];
4015 scan= trie_node + trie->jump[word];
4016 /* We go from the jump point to the branch that follows
4017 it. Note this means we need the vestigal unused branches
4018 even though they arent otherwise used.
4020 minnext = study_chunk(pRExC_state, &scan, minlenp,
4021 &deltanext, (regnode *)nextbranch, &data_fake,
4022 stopparen, recursed, NULL, f,depth+1);
4024 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4025 nextbranch= regnext((regnode*)nextbranch);
4027 if (min1 > (I32)(minnext + trie->minlen))
4028 min1 = minnext + trie->minlen;
4029 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4030 max1 = minnext + deltanext + trie->maxlen;
4031 if (deltanext == I32_MAX)
4032 is_inf = is_inf_internal = 1;
4034 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4036 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4037 if ( stopmin > min + min1)
4038 stopmin = min + min1;
4039 flags &= ~SCF_DO_SUBSTR;
4041 data->flags |= SCF_SEEN_ACCEPT;
4044 if (data_fake.flags & SF_HAS_EVAL)
4045 data->flags |= SF_HAS_EVAL;
4046 data->whilem_c = data_fake.whilem_c;
4048 if (flags & SCF_DO_STCLASS)
4049 cl_or(pRExC_state, &accum, &this_class);
4052 if (flags & SCF_DO_SUBSTR) {
4053 data->pos_min += min1;
4054 data->pos_delta += max1 - min1;
4055 if (max1 != min1 || is_inf)
4056 data->longest = &(data->longest_float);
4059 delta += max1 - min1;
4060 if (flags & SCF_DO_STCLASS_OR) {
4061 cl_or(pRExC_state, data->start_class, &accum);
4063 cl_and(data->start_class, and_withp);
4064 flags &= ~SCF_DO_STCLASS;
4067 else if (flags & SCF_DO_STCLASS_AND) {
4069 cl_and(data->start_class, &accum);
4070 flags &= ~SCF_DO_STCLASS;
4073 /* Switch to OR mode: cache the old value of
4074 * data->start_class */
4076 StructCopy(data->start_class, and_withp,
4077 struct regnode_charclass_class);
4078 flags &= ~SCF_DO_STCLASS_AND;
4079 StructCopy(&accum, data->start_class,
4080 struct regnode_charclass_class);
4081 flags |= SCF_DO_STCLASS_OR;
4082 data->start_class->flags |= ANYOF_EOS;
4089 else if (PL_regkind[OP(scan)] == TRIE) {
4090 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4093 min += trie->minlen;
4094 delta += (trie->maxlen - trie->minlen);
4095 flags &= ~SCF_DO_STCLASS; /* xxx */
4096 if (flags & SCF_DO_SUBSTR) {
4097 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4098 data->pos_min += trie->minlen;
4099 data->pos_delta += (trie->maxlen - trie->minlen);
4100 if (trie->maxlen != trie->minlen)
4101 data->longest = &(data->longest_float);
4103 if (trie->jump) /* no more substrings -- for now /grr*/
4104 flags &= ~SCF_DO_SUBSTR;
4106 #endif /* old or new */
4107 #endif /* TRIE_STUDY_OPT */
4109 /* Else: zero-length, ignore. */
4110 scan = regnext(scan);
4115 stopparen = frame->stop;
4116 frame = frame->prev;
4117 goto fake_study_recurse;
4122 DEBUG_STUDYDATA("pre-fin:",data,depth);
4125 *deltap = is_inf_internal ? I32_MAX : delta;
4126 if (flags & SCF_DO_SUBSTR && is_inf)
4127 data->pos_delta = I32_MAX - data->pos_min;
4128 if (is_par > (I32)U8_MAX)
4130 if (is_par && pars==1 && data) {
4131 data->flags |= SF_IN_PAR;
4132 data->flags &= ~SF_HAS_PAR;
4134 else if (pars && data) {
4135 data->flags |= SF_HAS_PAR;
4136 data->flags &= ~SF_IN_PAR;
4138 if (flags & SCF_DO_STCLASS_OR)
4139 cl_and(data->start_class, and_withp);
4140 if (flags & SCF_TRIE_RESTUDY)
4141 data->flags |= SCF_TRIE_RESTUDY;
4143 DEBUG_STUDYDATA("post-fin:",data,depth);
4145 return min < stopmin ? min : stopmin;
4149 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4151 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4153 PERL_ARGS_ASSERT_ADD_DATA;
4155 Renewc(RExC_rxi->data,
4156 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4157 char, struct reg_data);
4159 Renew(RExC_rxi->data->what, count + n, U8);
4161 Newx(RExC_rxi->data->what, n, U8);
4162 RExC_rxi->data->count = count + n;
4163 Copy(s, RExC_rxi->data->what + count, n, U8);
4167 /*XXX: todo make this not included in a non debugging perl */
4168 #ifndef PERL_IN_XSUB_RE
4170 Perl_reginitcolors(pTHX)
4173 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4175 char *t = savepv(s);
4179 t = strchr(t, '\t');
4185 PL_colors[i] = t = (char *)"";
4190 PL_colors[i++] = (char *)"";
4197 #ifdef TRIE_STUDY_OPT
4198 #define CHECK_RESTUDY_GOTO \
4200 (data.flags & SCF_TRIE_RESTUDY) \
4204 #define CHECK_RESTUDY_GOTO
4208 - pregcomp - compile a regular expression into internal code
4210 * We can't allocate space until we know how big the compiled form will be,
4211 * but we can't compile it (and thus know how big it is) until we've got a
4212 * place to put the code. So we cheat: we compile it twice, once with code
4213 * generation turned off and size counting turned on, and once "for real".
4214 * This also means that we don't allocate space until we are sure that the
4215 * thing really will compile successfully, and we never have to move the
4216 * code and thus invalidate pointers into it. (Note that it has to be in
4217 * one piece because free() must be able to free it all.) [NB: not true in perl]
4219 * Beware that the optimization-preparation code in here knows about some
4220 * of the structure of the compiled regexp. [I'll say.]
4225 #ifndef PERL_IN_XSUB_RE
4226 #define RE_ENGINE_PTR &PL_core_reg_engine
4228 extern const struct regexp_engine my_reg_engine;
4229 #define RE_ENGINE_PTR &my_reg_engine
4232 #ifndef PERL_IN_XSUB_RE
4234 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4237 HV * const table = GvHV(PL_hintgv);
4239 PERL_ARGS_ASSERT_PREGCOMP;
4241 /* Dispatch a request to compile a regexp to correct
4244 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4245 GET_RE_DEBUG_FLAGS_DECL;
4246 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4247 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4249 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4252 return CALLREGCOMP_ENG(eng, pattern, flags);
4255 return Perl_re_compile(aTHX_ pattern, flags);
4260 Perl_re_compile(pTHX_ SV * const pattern, U32 pm_flags)
4265 register regexp_internal *ri;
4267 char *exp = SvPV(pattern, plen);
4268 char* xend = exp + plen;
4275 RExC_state_t RExC_state;
4276 RExC_state_t * const pRExC_state = &RExC_state;
4277 #ifdef TRIE_STUDY_OPT
4279 RExC_state_t copyRExC_state;
4281 GET_RE_DEBUG_FLAGS_DECL;
4283 PERL_ARGS_ASSERT_RE_COMPILE;
4285 DEBUG_r(if (!PL_colorset) reginitcolors());
4287 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4290 SV *dsv= sv_newmortal();
4291 RE_PV_QUOTED_DECL(s, RExC_utf8,
4292 dsv, exp, plen, 60);
4293 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4294 PL_colors[4],PL_colors[5],s);
4299 RExC_flags = pm_flags;
4303 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4304 RExC_seen_evals = 0;
4307 /* First pass: determine size, legality. */
4315 RExC_emit = &PL_regdummy;
4316 RExC_whilem_seen = 0;
4317 RExC_open_parens = NULL;
4318 RExC_close_parens = NULL;
4320 RExC_paren_names = NULL;
4322 RExC_paren_name_list = NULL;
4324 RExC_recurse = NULL;
4325 RExC_recurse_count = 0;
4327 #if 0 /* REGC() is (currently) a NOP at the first pass.
4328 * Clever compilers notice this and complain. --jhi */
4329 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4331 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4332 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4333 RExC_precomp = NULL;
4336 if (RExC_utf8 && !RExC_orig_utf8) {
4337 /* It's possible to write a regexp in ascii that represents Unicode
4338 codepoints outside of the byte range, such as via \x{100}. If we
4339 detect such a sequence we have to convert the entire pattern to utf8
4340 and then recompile, as our sizing calculation will have been based
4341 on 1 byte == 1 character, but we will need to use utf8 to encode
4342 at least some part of the pattern, and therefore must convert the whole
4344 XXX: somehow figure out how to make this less expensive...
4347 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4348 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4349 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4351 RExC_orig_utf8 = RExC_utf8;
4353 goto redo_first_pass;
4356 PerlIO_printf(Perl_debug_log,
4357 "Required size %"IVdf" nodes\n"
4358 "Starting second pass (creation)\n",
4361 RExC_lastparse=NULL;
4363 /* Small enough for pointer-storage convention?
4364 If extralen==0, this means that we will not need long jumps. */
4365 if (RExC_size >= 0x10000L && RExC_extralen)
4366 RExC_size += RExC_extralen;
4369 if (RExC_whilem_seen > 15)
4370 RExC_whilem_seen = 15;
4372 /* Allocate space and zero-initialize. Note, the two step process
4373 of zeroing when in debug mode, thus anything assigned has to
4374 happen after that */
4375 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4376 r = (struct regexp*)SvANY(rx);
4377 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4378 char, regexp_internal);
4379 if ( r == NULL || ri == NULL )
4380 FAIL("Regexp out of space");
4382 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4383 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4385 /* bulk initialize base fields with 0. */
4386 Zero(ri, sizeof(regexp_internal), char);
4389 /* non-zero initialization begins here */
4391 r->engine= RE_ENGINE_PTR;
4392 r->extflags = pm_flags;
4394 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4395 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4396 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4397 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4398 >> RXf_PMf_STD_PMMOD_SHIFT);
4399 const char *fptr = STD_PAT_MODS; /*"msix"*/
4401 const STRLEN wraplen = plen + has_minus + has_p + has_runon
4402 + (sizeof(STD_PAT_MODS) - 1)
4403 + (sizeof("(?:)") - 1);
4405 p = sv_grow(MUTABLE_SV(rx), wraplen + 1);
4406 SvCUR_set(rx, wraplen);
4408 SvFLAGS(rx) |= SvUTF8(pattern);
4411 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4413 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4414 char *colon = r + 1;
4417 while((ch = *fptr++)) {
4431 Copy(RExC_precomp, p, plen, char);
4432 assert ((RX_WRAPPED(rx) - p) < 16);
4433 r->pre_prefix = p - RX_WRAPPED(rx);
4442 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4444 if (RExC_seen & REG_SEEN_RECURSE) {
4445 Newxz(RExC_open_parens, RExC_npar,regnode *);
4446 SAVEFREEPV(RExC_open_parens);
4447 Newxz(RExC_close_parens,RExC_npar,regnode *);
4448 SAVEFREEPV(RExC_close_parens);
4451 /* Useful during FAIL. */
4452 #ifdef RE_TRACK_PATTERN_OFFSETS
4453 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4454 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4455 "%s %"UVuf" bytes for offset annotations.\n",
4456 ri->u.offsets ? "Got" : "Couldn't get",
4457 (UV)((2*RExC_size+1) * sizeof(U32))));
4459 SetProgLen(ri,RExC_size);
4464 /* Second pass: emit code. */
4465 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4470 RExC_emit_start = ri->program;
4471 RExC_emit = ri->program;
4472 RExC_emit_bound = ri->program + RExC_size + 1;
4474 /* Store the count of eval-groups for security checks: */
4475 RExC_rx->seen_evals = RExC_seen_evals;
4476 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4477 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4481 /* XXXX To minimize changes to RE engine we always allocate
4482 3-units-long substrs field. */
4483 Newx(r->substrs, 1, struct reg_substr_data);
4484 if (RExC_recurse_count) {
4485 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4486 SAVEFREEPV(RExC_recurse);
4490 r->minlen = minlen = sawplus = sawopen = 0;
4491 Zero(r->substrs, 1, struct reg_substr_data);
4493 #ifdef TRIE_STUDY_OPT
4495 StructCopy(&zero_scan_data, &data, scan_data_t);
4496 copyRExC_state = RExC_state;
4499 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4501 RExC_state = copyRExC_state;
4502 if (seen & REG_TOP_LEVEL_BRANCHES)
4503 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4505 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4506 if (data.last_found) {
4507 SvREFCNT_dec(data.longest_fixed);
4508 SvREFCNT_dec(data.longest_float);
4509 SvREFCNT_dec(data.last_found);
4511 StructCopy(&zero_scan_data, &data, scan_data_t);
4514 StructCopy(&zero_scan_data, &data, scan_data_t);
4517 /* Dig out information for optimizations. */
4518 r->extflags = RExC_flags; /* was pm_op */
4519 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4522 SvUTF8_on(rx); /* Unicode in it? */
4523 ri->regstclass = NULL;
4524 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4525 r->intflags |= PREGf_NAUGHTY;
4526 scan = ri->program + 1; /* First BRANCH. */
4528 /* testing for BRANCH here tells us whether there is "must appear"
4529 data in the pattern. If there is then we can use it for optimisations */
4530 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4532 STRLEN longest_float_length, longest_fixed_length;
4533 struct regnode_charclass_class ch_class; /* pointed to by data */
4535 I32 last_close = 0; /* pointed to by data */
4536 regnode *first= scan;
4537 regnode *first_next= regnext(first);
4540 * Skip introductions and multiplicators >= 1
4541 * so that we can extract the 'meat' of the pattern that must
4542 * match in the large if() sequence following.
4543 * NOTE that EXACT is NOT covered here, as it is normally
4544 * picked up by the optimiser separately.
4546 * This is unfortunate as the optimiser isnt handling lookahead
4547 * properly currently.
4550 while ((OP(first) == OPEN && (sawopen = 1)) ||
4551 /* An OR of *one* alternative - should not happen now. */
4552 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4553 /* for now we can't handle lookbehind IFMATCH*/
4554 (OP(first) == IFMATCH && !first->flags) ||
4555 (OP(first) == PLUS) ||
4556 (OP(first) == MINMOD) ||
4557 /* An {n,m} with n>0 */
4558 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4559 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4562 * the only op that could be a regnode is PLUS, all the rest
4563 * will be regnode_1 or regnode_2.
4566 if (OP(first) == PLUS)
4569 first += regarglen[OP(first)];
4571 first = NEXTOPER(first);
4572 first_next= regnext(first);
4575 /* Starting-point info. */
4577 DEBUG_PEEP("first:",first,0);
4578 /* Ignore EXACT as we deal with it later. */
4579 if (PL_regkind[OP(first)] == EXACT) {
4580 if (OP(first) == EXACT)
4581 NOOP; /* Empty, get anchored substr later. */
4582 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4583 ri->regstclass = first;
4586 else if (PL_regkind[OP(first)] == TRIE &&
4587 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4590 /* this can happen only on restudy */
4591 if ( OP(first) == TRIE ) {
4592 struct regnode_1 *trieop = (struct regnode_1 *)
4593 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4594 StructCopy(first,trieop,struct regnode_1);
4595 trie_op=(regnode *)trieop;
4597 struct regnode_charclass *trieop = (struct regnode_charclass *)
4598 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4599 StructCopy(first,trieop,struct regnode_charclass);
4600 trie_op=(regnode *)trieop;
4603 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4604 ri->regstclass = trie_op;
4607 else if (REGNODE_SIMPLE(OP(first)))
4608 ri->regstclass = first;
4609 else if (PL_regkind[OP(first)] == BOUND ||
4610 PL_regkind[OP(first)] == NBOUND)
4611 ri->regstclass = first;
4612 else if (PL_regkind[OP(first)] == BOL) {
4613 r->extflags |= (OP(first) == MBOL
4615 : (OP(first) == SBOL
4618 first = NEXTOPER(first);
4621 else if (OP(first) == GPOS) {
4622 r->extflags |= RXf_ANCH_GPOS;
4623 first = NEXTOPER(first);
4626 else if ((!sawopen || !RExC_sawback) &&
4627 (OP(first) == STAR &&
4628 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4629 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4631 /* turn .* into ^.* with an implied $*=1 */
4633 (OP(NEXTOPER(first)) == REG_ANY)
4636 r->extflags |= type;
4637 r->intflags |= PREGf_IMPLICIT;
4638 first = NEXTOPER(first);
4641 if (sawplus && (!sawopen || !RExC_sawback)
4642 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4643 /* x+ must match at the 1st pos of run of x's */
4644 r->intflags |= PREGf_SKIP;
4646 /* Scan is after the zeroth branch, first is atomic matcher. */
4647 #ifdef TRIE_STUDY_OPT
4650 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4651 (IV)(first - scan + 1))
4655 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4656 (IV)(first - scan + 1))
4662 * If there's something expensive in the r.e., find the
4663 * longest literal string that must appear and make it the
4664 * regmust. Resolve ties in favor of later strings, since
4665 * the regstart check works with the beginning of the r.e.
4666 * and avoiding duplication strengthens checking. Not a
4667 * strong reason, but sufficient in the absence of others.
4668 * [Now we resolve ties in favor of the earlier string if
4669 * it happens that c_offset_min has been invalidated, since the
4670 * earlier string may buy us something the later one won't.]
4673 data.longest_fixed = newSVpvs("");
4674 data.longest_float = newSVpvs("");
4675 data.last_found = newSVpvs("");
4676 data.longest = &(data.longest_fixed);
4678 if (!ri->regstclass) {
4679 cl_init(pRExC_state, &ch_class);
4680 data.start_class = &ch_class;
4681 stclass_flag = SCF_DO_STCLASS_AND;
4682 } else /* XXXX Check for BOUND? */
4684 data.last_closep = &last_close;
4686 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4687 &data, -1, NULL, NULL,
4688 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4694 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4695 && data.last_start_min == 0 && data.last_end > 0
4696 && !RExC_seen_zerolen
4697 && !(RExC_seen & REG_SEEN_VERBARG)
4698 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4699 r->extflags |= RXf_CHECK_ALL;
4700 scan_commit(pRExC_state, &data,&minlen,0);
4701 SvREFCNT_dec(data.last_found);
4703 /* Note that code very similar to this but for anchored string
4704 follows immediately below, changes may need to be made to both.
4707 longest_float_length = CHR_SVLEN(data.longest_float);
4708 if (longest_float_length
4709 || (data.flags & SF_FL_BEFORE_EOL
4710 && (!(data.flags & SF_FL_BEFORE_MEOL)
4711 || (RExC_flags & RXf_PMf_MULTILINE))))
4715 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4716 && data.offset_fixed == data.offset_float_min
4717 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4718 goto remove_float; /* As in (a)+. */
4720 /* copy the information about the longest float from the reg_scan_data
4721 over to the program. */
4722 if (SvUTF8(data.longest_float)) {
4723 r->float_utf8 = data.longest_float;
4724 r->float_substr = NULL;
4726 r->float_substr = data.longest_float;
4727 r->float_utf8 = NULL;
4729 /* float_end_shift is how many chars that must be matched that
4730 follow this item. We calculate it ahead of time as once the
4731 lookbehind offset is added in we lose the ability to correctly
4733 ml = data.minlen_float ? *(data.minlen_float)
4734 : (I32)longest_float_length;
4735 r->float_end_shift = ml - data.offset_float_min
4736 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4737 + data.lookbehind_float;
4738 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4739 r->float_max_offset = data.offset_float_max;
4740 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4741 r->float_max_offset -= data.lookbehind_float;
4743 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4744 && (!(data.flags & SF_FL_BEFORE_MEOL)
4745 || (RExC_flags & RXf_PMf_MULTILINE)));
4746 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4750 r->float_substr = r->float_utf8 = NULL;
4751 SvREFCNT_dec(data.longest_float);
4752 longest_float_length = 0;
4755 /* Note that code very similar to this but for floating string
4756 is immediately above, changes may need to be made to both.
4759 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4760 if (longest_fixed_length
4761 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4762 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4763 || (RExC_flags & RXf_PMf_MULTILINE))))
4767 /* copy the information about the longest fixed
4768 from the reg_scan_data over to the program. */
4769 if (SvUTF8(data.longest_fixed)) {
4770 r->anchored_utf8 = data.longest_fixed;
4771 r->anchored_substr = NULL;
4773 r->anchored_substr = data.longest_fixed;
4774 r->anchored_utf8 = NULL;
4776 /* fixed_end_shift is how many chars that must be matched that
4777 follow this item. We calculate it ahead of time as once the
4778 lookbehind offset is added in we lose the ability to correctly
4780 ml = data.minlen_fixed ? *(data.minlen_fixed)
4781 : (I32)longest_fixed_length;
4782 r->anchored_end_shift = ml - data.offset_fixed
4783 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4784 + data.lookbehind_fixed;
4785 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4787 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4788 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4789 || (RExC_flags & RXf_PMf_MULTILINE)));
4790 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4793 r->anchored_substr = r->anchored_utf8 = NULL;
4794 SvREFCNT_dec(data.longest_fixed);
4795 longest_fixed_length = 0;
4798 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4799 ri->regstclass = NULL;
4800 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4802 && !(data.start_class->flags & ANYOF_EOS)
4803 && !cl_is_anything(data.start_class))
4805 const U32 n = add_data(pRExC_state, 1, "f");
4807 Newx(RExC_rxi->data->data[n], 1,
4808 struct regnode_charclass_class);
4809 StructCopy(data.start_class,
4810 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4811 struct regnode_charclass_class);
4812 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4813 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4814 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4815 regprop(r, sv, (regnode*)data.start_class);
4816 PerlIO_printf(Perl_debug_log,
4817 "synthetic stclass \"%s\".\n",
4818 SvPVX_const(sv));});
4821 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4822 if (longest_fixed_length > longest_float_length) {
4823 r->check_end_shift = r->anchored_end_shift;
4824 r->check_substr = r->anchored_substr;
4825 r->check_utf8 = r->anchored_utf8;
4826 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4827 if (r->extflags & RXf_ANCH_SINGLE)
4828 r->extflags |= RXf_NOSCAN;
4831 r->check_end_shift = r->float_end_shift;
4832 r->check_substr = r->float_substr;
4833 r->check_utf8 = r->float_utf8;
4834 r->check_offset_min = r->float_min_offset;
4835 r->check_offset_max = r->float_max_offset;
4837 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4838 This should be changed ASAP! */
4839 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4840 r->extflags |= RXf_USE_INTUIT;
4841 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4842 r->extflags |= RXf_INTUIT_TAIL;
4844 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4845 if ( (STRLEN)minlen < longest_float_length )
4846 minlen= longest_float_length;
4847 if ( (STRLEN)minlen < longest_fixed_length )
4848 minlen= longest_fixed_length;
4852 /* Several toplevels. Best we can is to set minlen. */
4854 struct regnode_charclass_class ch_class;
4857 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4859 scan = ri->program + 1;
4860 cl_init(pRExC_state, &ch_class);
4861 data.start_class = &ch_class;
4862 data.last_closep = &last_close;
4865 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4866 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4870 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4871 = r->float_substr = r->float_utf8 = NULL;
4872 if (!(data.start_class->flags & ANYOF_EOS)
4873 && !cl_is_anything(data.start_class))
4875 const U32 n = add_data(pRExC_state, 1, "f");
4877 Newx(RExC_rxi->data->data[n], 1,
4878 struct regnode_charclass_class);
4879 StructCopy(data.start_class,
4880 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4881 struct regnode_charclass_class);
4882 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4883 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4884 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4885 regprop(r, sv, (regnode*)data.start_class);
4886 PerlIO_printf(Perl_debug_log,
4887 "synthetic stclass \"%s\".\n",
4888 SvPVX_const(sv));});
4892 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4893 the "real" pattern. */
4895 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4896 (IV)minlen, (IV)r->minlen);
4898 r->minlenret = minlen;
4899 if (r->minlen < minlen)
4902 if (RExC_seen & REG_SEEN_GPOS)
4903 r->extflags |= RXf_GPOS_SEEN;
4904 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4905 r->extflags |= RXf_LOOKBEHIND_SEEN;
4906 if (RExC_seen & REG_SEEN_EVAL)
4907 r->extflags |= RXf_EVAL_SEEN;
4908 if (RExC_seen & REG_SEEN_CANY)
4909 r->extflags |= RXf_CANY_SEEN;
4910 if (RExC_seen & REG_SEEN_VERBARG)
4911 r->intflags |= PREGf_VERBARG_SEEN;
4912 if (RExC_seen & REG_SEEN_CUTGROUP)
4913 r->intflags |= PREGf_CUTGROUP_SEEN;
4914 if (RExC_paren_names)
4915 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
4917 RXp_PAREN_NAMES(r) = NULL;
4919 #ifdef STUPID_PATTERN_CHECKS
4920 if (RX_PRELEN(rx) == 0)
4921 r->extflags |= RXf_NULL;
4922 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4923 /* XXX: this should happen BEFORE we compile */
4924 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4925 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
4926 r->extflags |= RXf_WHITE;
4927 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
4928 r->extflags |= RXf_START_ONLY;
4930 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4931 /* XXX: this should happen BEFORE we compile */
4932 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4934 regnode *first = ri->program + 1;
4936 U8 nop = OP(NEXTOPER(first));
4938 if (PL_regkind[fop] == NOTHING && nop == END)
4939 r->extflags |= RXf_NULL;
4940 else if (PL_regkind[fop] == BOL && nop == END)
4941 r->extflags |= RXf_START_ONLY;
4942 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4943 r->extflags |= RXf_WHITE;
4947 if (RExC_paren_names) {
4948 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
4949 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4952 ri->name_list_idx = 0;
4954 if (RExC_recurse_count) {
4955 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4956 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4957 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4960 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4961 /* assume we don't need to swap parens around before we match */
4964 PerlIO_printf(Perl_debug_log,"Final program:\n");
4967 #ifdef RE_TRACK_PATTERN_OFFSETS
4968 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4969 const U32 len = ri->u.offsets[0];
4971 GET_RE_DEBUG_FLAGS_DECL;
4972 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4973 for (i = 1; i <= len; i++) {
4974 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4975 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4976 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4978 PerlIO_printf(Perl_debug_log, "\n");
4984 #undef RE_ENGINE_PTR
4988 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4991 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
4993 PERL_UNUSED_ARG(value);
4995 if (flags & RXapif_FETCH) {
4996 return reg_named_buff_fetch(rx, key, flags);
4997 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
4998 Perl_croak(aTHX_ "%s", PL_no_modify);
5000 } else if (flags & RXapif_EXISTS) {
5001 return reg_named_buff_exists(rx, key, flags)
5004 } else if (flags & RXapif_REGNAMES) {
5005 return reg_named_buff_all(rx, flags);
5006 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5007 return reg_named_buff_scalar(rx, flags);
5009 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5015 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5018 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5019 PERL_UNUSED_ARG(lastkey);
5021 if (flags & RXapif_FIRSTKEY)
5022 return reg_named_buff_firstkey(rx, flags);
5023 else if (flags & RXapif_NEXTKEY)
5024 return reg_named_buff_nextkey(rx, flags);
5026 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5032 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5035 AV *retarray = NULL;
5037 struct regexp *const rx = (struct regexp *)SvANY(r);
5039 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5041 if (flags & RXapif_ALL)
5044 if (rx && RXp_PAREN_NAMES(rx)) {
5045 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5048 SV* sv_dat=HeVAL(he_str);
5049 I32 *nums=(I32*)SvPVX(sv_dat);
5050 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5051 if ((I32)(rx->nparens) >= nums[i]
5052 && rx->offs[nums[i]].start != -1
5053 && rx->offs[nums[i]].end != -1)
5056 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5060 ret = newSVsv(&PL_sv_undef);
5063 av_push(retarray, ret);
5066 return newRV_noinc(MUTABLE_SV(retarray));
5073 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5076 struct regexp *const rx = (struct regexp *)SvANY(r);
5078 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5080 if (rx && RXp_PAREN_NAMES(rx)) {
5081 if (flags & RXapif_ALL) {
5082 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5084 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5098 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5100 struct regexp *const rx = (struct regexp *)SvANY(r);
5102 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5104 if ( rx && RXp_PAREN_NAMES(rx) ) {
5105 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5107 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5114 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5116 struct regexp *const rx = (struct regexp *)SvANY(r);
5117 GET_RE_DEBUG_FLAGS_DECL;
5119 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5121 if (rx && RXp_PAREN_NAMES(rx)) {
5122 HV *hv = RXp_PAREN_NAMES(rx);
5124 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5127 SV* sv_dat = HeVAL(temphe);
5128 I32 *nums = (I32*)SvPVX(sv_dat);
5129 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5130 if ((I32)(rx->lastparen) >= nums[i] &&
5131 rx->offs[nums[i]].start != -1 &&
5132 rx->offs[nums[i]].end != -1)
5138 if (parno || flags & RXapif_ALL) {
5139 return newSVhek(HeKEY_hek(temphe));
5147 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5152 struct regexp *const rx = (struct regexp *)SvANY(r);
5154 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5156 if (rx && RXp_PAREN_NAMES(rx)) {
5157 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5158 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5159 } else if (flags & RXapif_ONE) {
5160 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5161 av = MUTABLE_AV(SvRV(ret));
5162 length = av_len(av);
5164 return newSViv(length + 1);
5166 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5170 return &PL_sv_undef;
5174 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5176 struct regexp *const rx = (struct regexp *)SvANY(r);
5179 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5181 if (rx && RXp_PAREN_NAMES(rx)) {
5182 HV *hv= RXp_PAREN_NAMES(rx);
5184 (void)hv_iterinit(hv);
5185 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5188 SV* sv_dat = HeVAL(temphe);
5189 I32 *nums = (I32*)SvPVX(sv_dat);
5190 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5191 if ((I32)(rx->lastparen) >= nums[i] &&
5192 rx->offs[nums[i]].start != -1 &&
5193 rx->offs[nums[i]].end != -1)
5199 if (parno || flags & RXapif_ALL) {
5200 av_push(av, newSVhek(HeKEY_hek(temphe)));
5205 return newRV_noinc(MUTABLE_SV(av));
5209 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5212 struct regexp *const rx = (struct regexp *)SvANY(r);
5217 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5220 sv_setsv(sv,&PL_sv_undef);
5224 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5226 i = rx->offs[0].start;
5230 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5232 s = rx->subbeg + rx->offs[0].end;
5233 i = rx->sublen - rx->offs[0].end;
5236 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5237 (s1 = rx->offs[paren].start) != -1 &&
5238 (t1 = rx->offs[paren].end) != -1)
5242 s = rx->subbeg + s1;
5244 sv_setsv(sv,&PL_sv_undef);
5247 assert(rx->sublen >= (s - rx->subbeg) + i );
5249 const int oldtainted = PL_tainted;
5251 sv_setpvn(sv, s, i);
5252 PL_tainted = oldtainted;
5253 if ( (rx->extflags & RXf_CANY_SEEN)
5254 ? (RXp_MATCH_UTF8(rx)
5255 && (!i || is_utf8_string((U8*)s, i)))
5256 : (RXp_MATCH_UTF8(rx)) )
5263 if (RXp_MATCH_TAINTED(rx)) {
5264 if (SvTYPE(sv) >= SVt_PVMG) {
5265 MAGIC* const mg = SvMAGIC(sv);
5268 SvMAGIC_set(sv, mg->mg_moremagic);
5270 if ((mgt = SvMAGIC(sv))) {
5271 mg->mg_moremagic = mgt;
5272 SvMAGIC_set(sv, mg);
5282 sv_setsv(sv,&PL_sv_undef);
5288 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5289 SV const * const value)
5291 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5293 PERL_UNUSED_ARG(rx);
5294 PERL_UNUSED_ARG(paren);
5295 PERL_UNUSED_ARG(value);
5298 Perl_croak(aTHX_ "%s", PL_no_modify);
5302 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5305 struct regexp *const rx = (struct regexp *)SvANY(r);
5309 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5311 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5313 /* $` / ${^PREMATCH} */
5314 case RX_BUFF_IDX_PREMATCH:
5315 if (rx->offs[0].start != -1) {
5316 i = rx->offs[0].start;
5324 /* $' / ${^POSTMATCH} */
5325 case RX_BUFF_IDX_POSTMATCH:
5326 if (rx->offs[0].end != -1) {
5327 i = rx->sublen - rx->offs[0].end;
5329 s1 = rx->offs[0].end;
5335 /* $& / ${^MATCH}, $1, $2, ... */
5337 if (paren <= (I32)rx->nparens &&
5338 (s1 = rx->offs[paren].start) != -1 &&
5339 (t1 = rx->offs[paren].end) != -1)
5344 if (ckWARN(WARN_UNINITIALIZED))
5345 report_uninit((const SV *)sv);
5350 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5351 const char * const s = rx->subbeg + s1;
5356 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5363 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5365 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5366 PERL_UNUSED_ARG(rx);
5370 return newSVpvs("Regexp");
5373 /* Scans the name of a named buffer from the pattern.
5374 * If flags is REG_RSN_RETURN_NULL returns null.
5375 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5376 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5377 * to the parsed name as looked up in the RExC_paren_names hash.
5378 * If there is an error throws a vFAIL().. type exception.
5381 #define REG_RSN_RETURN_NULL 0
5382 #define REG_RSN_RETURN_NAME 1
5383 #define REG_RSN_RETURN_DATA 2
5386 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5388 char *name_start = RExC_parse;
5390 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5392 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5393 /* skip IDFIRST by using do...while */
5396 RExC_parse += UTF8SKIP(RExC_parse);
5397 } while (isALNUM_utf8((U8*)RExC_parse));
5401 } while (isALNUM(*RExC_parse));
5406 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5407 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5408 if ( flags == REG_RSN_RETURN_NAME)
5410 else if (flags==REG_RSN_RETURN_DATA) {
5413 if ( ! sv_name ) /* should not happen*/
5414 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5415 if (RExC_paren_names)
5416 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5418 sv_dat = HeVAL(he_str);
5420 vFAIL("Reference to nonexistent named group");
5424 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5431 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5432 int rem=(int)(RExC_end - RExC_parse); \
5441 if (RExC_lastparse!=RExC_parse) \
5442 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5445 iscut ? "..." : "<" \
5448 PerlIO_printf(Perl_debug_log,"%16s",""); \
5451 num = RExC_size + 1; \
5453 num=REG_NODE_NUM(RExC_emit); \
5454 if (RExC_lastnum!=num) \
5455 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5457 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5458 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5459 (int)((depth*2)), "", \
5463 RExC_lastparse=RExC_parse; \
5468 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5469 DEBUG_PARSE_MSG((funcname)); \
5470 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5472 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5473 DEBUG_PARSE_MSG((funcname)); \
5474 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5477 - reg - regular expression, i.e. main body or parenthesized thing
5479 * Caller must absorb opening parenthesis.
5481 * Combining parenthesis handling with the base level of regular expression
5482 * is a trifle forced, but the need to tie the tails of the branches to what
5483 * follows makes it hard to avoid.
5485 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5487 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5489 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5493 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5494 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5497 register regnode *ret; /* Will be the head of the group. */
5498 register regnode *br;
5499 register regnode *lastbr;
5500 register regnode *ender = NULL;
5501 register I32 parno = 0;
5503 U32 oregflags = RExC_flags;
5504 bool have_branch = 0;
5506 I32 freeze_paren = 0;
5507 I32 after_freeze = 0;
5509 /* for (?g), (?gc), and (?o) warnings; warning
5510 about (?c) will warn about (?g) -- japhy */
5512 #define WASTED_O 0x01
5513 #define WASTED_G 0x02
5514 #define WASTED_C 0x04
5515 #define WASTED_GC (0x02|0x04)
5516 I32 wastedflags = 0x00;
5518 char * parse_start = RExC_parse; /* MJD */
5519 char * const oregcomp_parse = RExC_parse;
5521 GET_RE_DEBUG_FLAGS_DECL;
5523 PERL_ARGS_ASSERT_REG;
5524 DEBUG_PARSE("reg ");
5526 *flagp = 0; /* Tentatively. */
5529 /* Make an OPEN node, if parenthesized. */
5531 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5532 char *start_verb = RExC_parse;
5533 STRLEN verb_len = 0;
5534 char *start_arg = NULL;
5535 unsigned char op = 0;
5537 int internal_argval = 0; /* internal_argval is only useful if !argok */
5538 while ( *RExC_parse && *RExC_parse != ')' ) {
5539 if ( *RExC_parse == ':' ) {
5540 start_arg = RExC_parse + 1;
5546 verb_len = RExC_parse - start_verb;
5549 while ( *RExC_parse && *RExC_parse != ')' )
5551 if ( *RExC_parse != ')' )
5552 vFAIL("Unterminated verb pattern argument");
5553 if ( RExC_parse == start_arg )
5556 if ( *RExC_parse != ')' )
5557 vFAIL("Unterminated verb pattern");
5560 switch ( *start_verb ) {
5561 case 'A': /* (*ACCEPT) */
5562 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5564 internal_argval = RExC_nestroot;
5567 case 'C': /* (*COMMIT) */
5568 if ( memEQs(start_verb,verb_len,"COMMIT") )
5571 case 'F': /* (*FAIL) */
5572 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5577 case ':': /* (*:NAME) */
5578 case 'M': /* (*MARK:NAME) */
5579 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5584 case 'P': /* (*PRUNE) */
5585 if ( memEQs(start_verb,verb_len,"PRUNE") )
5588 case 'S': /* (*SKIP) */
5589 if ( memEQs(start_verb,verb_len,"SKIP") )
5592 case 'T': /* (*THEN) */
5593 /* [19:06] <TimToady> :: is then */
5594 if ( memEQs(start_verb,verb_len,"THEN") ) {
5596 RExC_seen |= REG_SEEN_CUTGROUP;
5602 vFAIL3("Unknown verb pattern '%.*s'",
5603 verb_len, start_verb);
5606 if ( start_arg && internal_argval ) {
5607 vFAIL3("Verb pattern '%.*s' may not have an argument",
5608 verb_len, start_verb);
5609 } else if ( argok < 0 && !start_arg ) {
5610 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5611 verb_len, start_verb);
5613 ret = reganode(pRExC_state, op, internal_argval);
5614 if ( ! internal_argval && ! SIZE_ONLY ) {
5616 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5617 ARG(ret) = add_data( pRExC_state, 1, "S" );
5618 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5625 if (!internal_argval)
5626 RExC_seen |= REG_SEEN_VERBARG;
5627 } else if ( start_arg ) {
5628 vFAIL3("Verb pattern '%.*s' may not have an argument",
5629 verb_len, start_verb);
5631 ret = reg_node(pRExC_state, op);
5633 nextchar(pRExC_state);
5636 if (*RExC_parse == '?') { /* (?...) */
5637 bool is_logical = 0;
5638 const char * const seqstart = RExC_parse;
5641 paren = *RExC_parse++;
5642 ret = NULL; /* For look-ahead/behind. */
5645 case 'P': /* (?P...) variants for those used to PCRE/Python */
5646 paren = *RExC_parse++;
5647 if ( paren == '<') /* (?P<...>) named capture */
5649 else if (paren == '>') { /* (?P>name) named recursion */
5650 goto named_recursion;
5652 else if (paren == '=') { /* (?P=...) named backref */
5653 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5654 you change this make sure you change that */
5655 char* name_start = RExC_parse;
5657 SV *sv_dat = reg_scan_name(pRExC_state,
5658 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5659 if (RExC_parse == name_start || *RExC_parse != ')')
5660 vFAIL2("Sequence %.3s... not terminated",parse_start);
5663 num = add_data( pRExC_state, 1, "S" );
5664 RExC_rxi->data->data[num]=(void*)sv_dat;
5665 SvREFCNT_inc_simple_void(sv_dat);
5668 ret = reganode(pRExC_state,
5669 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5673 Set_Node_Offset(ret, parse_start+1);
5674 Set_Node_Cur_Length(ret); /* MJD */
5676 nextchar(pRExC_state);
5680 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5682 case '<': /* (?<...) */
5683 if (*RExC_parse == '!')
5685 else if (*RExC_parse != '=')
5691 case '\'': /* (?'...') */
5692 name_start= RExC_parse;
5693 svname = reg_scan_name(pRExC_state,
5694 SIZE_ONLY ? /* reverse test from the others */
5695 REG_RSN_RETURN_NAME :
5696 REG_RSN_RETURN_NULL);
5697 if (RExC_parse == name_start) {
5699 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5702 if (*RExC_parse != paren)
5703 vFAIL2("Sequence (?%c... not terminated",
5704 paren=='>' ? '<' : paren);
5708 if (!svname) /* shouldnt happen */
5710 "panic: reg_scan_name returned NULL");
5711 if (!RExC_paren_names) {
5712 RExC_paren_names= newHV();
5713 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5715 RExC_paren_name_list= newAV();
5716 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5719 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5721 sv_dat = HeVAL(he_str);
5723 /* croak baby croak */
5725 "panic: paren_name hash element allocation failed");
5726 } else if ( SvPOK(sv_dat) ) {
5727 /* (?|...) can mean we have dupes so scan to check
5728 its already been stored. Maybe a flag indicating
5729 we are inside such a construct would be useful,
5730 but the arrays are likely to be quite small, so
5731 for now we punt -- dmq */
5732 IV count = SvIV(sv_dat);
5733 I32 *pv = (I32*)SvPVX(sv_dat);
5735 for ( i = 0 ; i < count ; i++ ) {
5736 if ( pv[i] == RExC_npar ) {
5742 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5743 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5744 pv[count] = RExC_npar;
5745 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5748 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5749 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5751 SvIV_set(sv_dat, 1);
5754 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5755 SvREFCNT_dec(svname);
5758 /*sv_dump(sv_dat);*/
5760 nextchar(pRExC_state);
5762 goto capturing_parens;
5764 RExC_seen |= REG_SEEN_LOOKBEHIND;
5766 case '=': /* (?=...) */
5767 RExC_seen_zerolen++;
5769 case '!': /* (?!...) */
5770 RExC_seen_zerolen++;
5771 if (*RExC_parse == ')') {
5772 ret=reg_node(pRExC_state, OPFAIL);
5773 nextchar(pRExC_state);
5777 case '|': /* (?|...) */
5778 /* branch reset, behave like a (?:...) except that
5779 buffers in alternations share the same numbers */
5781 after_freeze = freeze_paren = RExC_npar;
5783 case ':': /* (?:...) */
5784 case '>': /* (?>...) */
5786 case '$': /* (?$...) */
5787 case '@': /* (?@...) */
5788 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5790 case '#': /* (?#...) */
5791 while (*RExC_parse && *RExC_parse != ')')
5793 if (*RExC_parse != ')')
5794 FAIL("Sequence (?#... not terminated");
5795 nextchar(pRExC_state);
5798 case '0' : /* (?0) */
5799 case 'R' : /* (?R) */
5800 if (*RExC_parse != ')')
5801 FAIL("Sequence (?R) not terminated");
5802 ret = reg_node(pRExC_state, GOSTART);
5803 *flagp |= POSTPONED;
5804 nextchar(pRExC_state);
5807 { /* named and numeric backreferences */
5809 case '&': /* (?&NAME) */
5810 parse_start = RExC_parse - 1;
5813 SV *sv_dat = reg_scan_name(pRExC_state,
5814 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5815 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5817 goto gen_recurse_regop;
5820 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5822 vFAIL("Illegal pattern");
5824 goto parse_recursion;
5826 case '-': /* (?-1) */
5827 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5828 RExC_parse--; /* rewind to let it be handled later */
5832 case '1': case '2': case '3': case '4': /* (?1) */
5833 case '5': case '6': case '7': case '8': case '9':
5836 num = atoi(RExC_parse);
5837 parse_start = RExC_parse - 1; /* MJD */
5838 if (*RExC_parse == '-')
5840 while (isDIGIT(*RExC_parse))
5842 if (*RExC_parse!=')')
5843 vFAIL("Expecting close bracket");
5846 if ( paren == '-' ) {
5848 Diagram of capture buffer numbering.
5849 Top line is the normal capture buffer numbers
5850 Botton line is the negative indexing as from
5854 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5858 num = RExC_npar + num;
5861 vFAIL("Reference to nonexistent group");
5863 } else if ( paren == '+' ) {
5864 num = RExC_npar + num - 1;
5867 ret = reganode(pRExC_state, GOSUB, num);
5869 if (num > (I32)RExC_rx->nparens) {
5871 vFAIL("Reference to nonexistent group");
5873 ARG2L_SET( ret, RExC_recurse_count++);
5875 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5876 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5880 RExC_seen |= REG_SEEN_RECURSE;
5881 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5882 Set_Node_Offset(ret, parse_start); /* MJD */
5884 *flagp |= POSTPONED;
5885 nextchar(pRExC_state);
5887 } /* named and numeric backreferences */
5890 case '?': /* (??...) */
5892 if (*RExC_parse != '{') {
5894 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5897 *flagp |= POSTPONED;
5898 paren = *RExC_parse++;
5900 case '{': /* (?{...}) */
5905 char *s = RExC_parse;
5907 RExC_seen_zerolen++;
5908 RExC_seen |= REG_SEEN_EVAL;
5909 while (count && (c = *RExC_parse)) {
5920 if (*RExC_parse != ')') {
5922 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5926 OP_4tree *sop, *rop;
5927 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5930 Perl_save_re_context(aTHX);
5931 rop = sv_compile_2op(sv, &sop, "re", &pad);
5932 sop->op_private |= OPpREFCOUNTED;
5933 /* re_dup will OpREFCNT_inc */
5934 OpREFCNT_set(sop, 1);
5937 n = add_data(pRExC_state, 3, "nop");
5938 RExC_rxi->data->data[n] = (void*)rop;
5939 RExC_rxi->data->data[n+1] = (void*)sop;
5940 RExC_rxi->data->data[n+2] = (void*)pad;
5943 else { /* First pass */
5944 if (PL_reginterp_cnt < ++RExC_seen_evals
5946 /* No compiled RE interpolated, has runtime
5947 components ===> unsafe. */
5948 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5949 if (PL_tainting && PL_tainted)
5950 FAIL("Eval-group in insecure regular expression");
5951 #if PERL_VERSION > 8
5952 if (IN_PERL_COMPILETIME)
5957 nextchar(pRExC_state);
5959 ret = reg_node(pRExC_state, LOGICAL);
5962 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5963 /* deal with the length of this later - MJD */
5966 ret = reganode(pRExC_state, EVAL, n);
5967 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5968 Set_Node_Offset(ret, parse_start);
5971 case '(': /* (?(?{...})...) and (?(?=...)...) */
5974 if (RExC_parse[0] == '?') { /* (?(?...)) */
5975 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5976 || RExC_parse[1] == '<'
5977 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5980 ret = reg_node(pRExC_state, LOGICAL);
5983 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5987 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5988 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5990 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5991 char *name_start= RExC_parse++;
5993 SV *sv_dat=reg_scan_name(pRExC_state,
5994 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5995 if (RExC_parse == name_start || *RExC_parse != ch)
5996 vFAIL2("Sequence (?(%c... not terminated",
5997 (ch == '>' ? '<' : ch));
6000 num = add_data( pRExC_state, 1, "S" );
6001 RExC_rxi->data->data[num]=(void*)sv_dat;
6002 SvREFCNT_inc_simple_void(sv_dat);
6004 ret = reganode(pRExC_state,NGROUPP,num);
6005 goto insert_if_check_paren;
6007 else if (RExC_parse[0] == 'D' &&
6008 RExC_parse[1] == 'E' &&
6009 RExC_parse[2] == 'F' &&
6010 RExC_parse[3] == 'I' &&
6011 RExC_parse[4] == 'N' &&
6012 RExC_parse[5] == 'E')
6014 ret = reganode(pRExC_state,DEFINEP,0);
6017 goto insert_if_check_paren;
6019 else if (RExC_parse[0] == 'R') {
6022 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6023 parno = atoi(RExC_parse++);
6024 while (isDIGIT(*RExC_parse))
6026 } else if (RExC_parse[0] == '&') {
6029 sv_dat = reg_scan_name(pRExC_state,
6030 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6031 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6033 ret = reganode(pRExC_state,INSUBP,parno);
6034 goto insert_if_check_paren;
6036 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6039 parno = atoi(RExC_parse++);
6041 while (isDIGIT(*RExC_parse))
6043 ret = reganode(pRExC_state, GROUPP, parno);
6045 insert_if_check_paren:
6046 if ((c = *nextchar(pRExC_state)) != ')')
6047 vFAIL("Switch condition not recognized");
6049 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
6050 br = regbranch(pRExC_state, &flags, 1,depth+1);
6052 br = reganode(pRExC_state, LONGJMP, 0);
6054 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
6055 c = *nextchar(pRExC_state);
6060 vFAIL("(?(DEFINE)....) does not allow branches");
6061 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6062 regbranch(pRExC_state, &flags, 1,depth+1);
6063 REGTAIL(pRExC_state, ret, lastbr);
6066 c = *nextchar(pRExC_state);
6071 vFAIL("Switch (?(condition)... contains too many branches");
6072 ender = reg_node(pRExC_state, TAIL);
6073 REGTAIL(pRExC_state, br, ender);
6075 REGTAIL(pRExC_state, lastbr, ender);
6076 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6079 REGTAIL(pRExC_state, ret, ender);
6080 RExC_size++; /* XXX WHY do we need this?!!
6081 For large programs it seems to be required
6082 but I can't figure out why. -- dmq*/
6086 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6090 RExC_parse--; /* for vFAIL to print correctly */
6091 vFAIL("Sequence (? incomplete");
6095 parse_flags: /* (?i) */
6097 U32 posflags = 0, negflags = 0;
6098 U32 *flagsp = &posflags;
6100 while (*RExC_parse) {
6101 /* && strchr("iogcmsx", *RExC_parse) */
6102 /* (?g), (?gc) and (?o) are useless here
6103 and must be globally applied -- japhy */
6104 switch (*RExC_parse) {
6105 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6106 case ONCE_PAT_MOD: /* 'o' */
6107 case GLOBAL_PAT_MOD: /* 'g' */
6108 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6109 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6110 if (! (wastedflags & wflagbit) ) {
6111 wastedflags |= wflagbit;
6114 "Useless (%s%c) - %suse /%c modifier",
6115 flagsp == &negflags ? "?-" : "?",
6117 flagsp == &negflags ? "don't " : "",
6124 case CONTINUE_PAT_MOD: /* 'c' */
6125 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6126 if (! (wastedflags & WASTED_C) ) {
6127 wastedflags |= WASTED_GC;
6130 "Useless (%sc) - %suse /gc modifier",
6131 flagsp == &negflags ? "?-" : "?",
6132 flagsp == &negflags ? "don't " : ""
6137 case KEEPCOPY_PAT_MOD: /* 'p' */
6138 if (flagsp == &negflags) {
6140 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
6142 *flagsp |= RXf_PMf_KEEPCOPY;
6146 if (flagsp == &negflags) {
6148 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6152 wastedflags = 0; /* reset so (?g-c) warns twice */
6158 RExC_flags |= posflags;
6159 RExC_flags &= ~negflags;
6161 oregflags |= posflags;
6162 oregflags &= ~negflags;
6164 nextchar(pRExC_state);
6175 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6180 }} /* one for the default block, one for the switch */
6187 ret = reganode(pRExC_state, OPEN, parno);
6190 RExC_nestroot = parno;
6191 if (RExC_seen & REG_SEEN_RECURSE
6192 && !RExC_open_parens[parno-1])
6194 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6195 "Setting open paren #%"IVdf" to %d\n",
6196 (IV)parno, REG_NODE_NUM(ret)));
6197 RExC_open_parens[parno-1]= ret;
6200 Set_Node_Length(ret, 1); /* MJD */
6201 Set_Node_Offset(ret, RExC_parse); /* MJD */
6209 /* Pick up the branches, linking them together. */
6210 parse_start = RExC_parse; /* MJD */
6211 br = regbranch(pRExC_state, &flags, 1,depth+1);
6214 if (RExC_npar > after_freeze)
6215 after_freeze = RExC_npar;
6216 RExC_npar = freeze_paren;
6219 /* branch_len = (paren != 0); */
6223 if (*RExC_parse == '|') {
6224 if (!SIZE_ONLY && RExC_extralen) {
6225 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6228 reginsert(pRExC_state, BRANCH, br, depth+1);
6229 Set_Node_Length(br, paren != 0);
6230 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6234 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6236 else if (paren == ':') {
6237 *flagp |= flags&SIMPLE;
6239 if (is_open) { /* Starts with OPEN. */
6240 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6242 else if (paren != '?') /* Not Conditional */
6244 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6246 while (*RExC_parse == '|') {
6247 if (!SIZE_ONLY && RExC_extralen) {
6248 ender = reganode(pRExC_state, LONGJMP,0);
6249 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6252 RExC_extralen += 2; /* Account for LONGJMP. */
6253 nextchar(pRExC_state);
6255 if (RExC_npar > after_freeze)
6256 after_freeze = RExC_npar;
6257 RExC_npar = freeze_paren;
6259 br = regbranch(pRExC_state, &flags, 0, depth+1);
6263 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6265 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6268 if (have_branch || paren != ':') {
6269 /* Make a closing node, and hook it on the end. */
6272 ender = reg_node(pRExC_state, TAIL);
6275 ender = reganode(pRExC_state, CLOSE, parno);
6276 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6277 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6278 "Setting close paren #%"IVdf" to %d\n",
6279 (IV)parno, REG_NODE_NUM(ender)));
6280 RExC_close_parens[parno-1]= ender;
6281 if (RExC_nestroot == parno)
6284 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6285 Set_Node_Length(ender,1); /* MJD */
6291 *flagp &= ~HASWIDTH;
6294 ender = reg_node(pRExC_state, SUCCEED);
6297 ender = reg_node(pRExC_state, END);
6299 assert(!RExC_opend); /* there can only be one! */
6304 REGTAIL(pRExC_state, lastbr, ender);
6306 if (have_branch && !SIZE_ONLY) {
6308 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6310 /* Hook the tails of the branches to the closing node. */
6311 for (br = ret; br; br = regnext(br)) {
6312 const U8 op = PL_regkind[OP(br)];
6314 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6316 else if (op == BRANCHJ) {
6317 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6325 static const char parens[] = "=!<,>";
6327 if (paren && (p = strchr(parens, paren))) {
6328 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6329 int flag = (p - parens) > 1;
6332 node = SUSPEND, flag = 0;
6333 reginsert(pRExC_state, node,ret, depth+1);
6334 Set_Node_Cur_Length(ret);
6335 Set_Node_Offset(ret, parse_start + 1);
6337 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6341 /* Check for proper termination. */
6343 RExC_flags = oregflags;
6344 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6345 RExC_parse = oregcomp_parse;
6346 vFAIL("Unmatched (");
6349 else if (!paren && RExC_parse < RExC_end) {
6350 if (*RExC_parse == ')') {
6352 vFAIL("Unmatched )");
6355 FAIL("Junk on end of regexp"); /* "Can't happen". */
6359 RExC_npar = after_freeze;
6364 - regbranch - one alternative of an | operator
6366 * Implements the concatenation operator.
6369 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6372 register regnode *ret;
6373 register regnode *chain = NULL;
6374 register regnode *latest;
6375 I32 flags = 0, c = 0;
6376 GET_RE_DEBUG_FLAGS_DECL;
6378 PERL_ARGS_ASSERT_REGBRANCH;
6380 DEBUG_PARSE("brnc");
6385 if (!SIZE_ONLY && RExC_extralen)
6386 ret = reganode(pRExC_state, BRANCHJ,0);
6388 ret = reg_node(pRExC_state, BRANCH);
6389 Set_Node_Length(ret, 1);
6393 if (!first && SIZE_ONLY)
6394 RExC_extralen += 1; /* BRANCHJ */
6396 *flagp = WORST; /* Tentatively. */
6399 nextchar(pRExC_state);
6400 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6402 latest = regpiece(pRExC_state, &flags,depth+1);
6403 if (latest == NULL) {
6404 if (flags & TRYAGAIN)
6408 else if (ret == NULL)
6410 *flagp |= flags&(HASWIDTH|POSTPONED);
6411 if (chain == NULL) /* First piece. */
6412 *flagp |= flags&SPSTART;
6415 REGTAIL(pRExC_state, chain, latest);
6420 if (chain == NULL) { /* Loop ran zero times. */
6421 chain = reg_node(pRExC_state, NOTHING);
6426 *flagp |= flags&SIMPLE;
6433 - regpiece - something followed by possible [*+?]
6435 * Note that the branching code sequences used for ? and the general cases
6436 * of * and + are somewhat optimized: they use the same NOTHING node as
6437 * both the endmarker for their branch list and the body of the last branch.
6438 * It might seem that this node could be dispensed with entirely, but the
6439 * endmarker role is not redundant.
6442 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6445 register regnode *ret;
6447 register char *next;
6449 const char * const origparse = RExC_parse;
6451 I32 max = REG_INFTY;
6453 const char *maxpos = NULL;
6454 GET_RE_DEBUG_FLAGS_DECL;
6456 PERL_ARGS_ASSERT_REGPIECE;
6458 DEBUG_PARSE("piec");
6460 ret = regatom(pRExC_state, &flags,depth+1);
6462 if (flags & TRYAGAIN)
6469 if (op == '{' && regcurly(RExC_parse)) {
6471 parse_start = RExC_parse; /* MJD */
6472 next = RExC_parse + 1;
6473 while (isDIGIT(*next) || *next == ',') {
6482 if (*next == '}') { /* got one */
6486 min = atoi(RExC_parse);
6490 maxpos = RExC_parse;
6492 if (!max && *maxpos != '0')
6493 max = REG_INFTY; /* meaning "infinity" */
6494 else if (max >= REG_INFTY)
6495 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6497 nextchar(pRExC_state);
6500 if ((flags&SIMPLE)) {
6501 RExC_naughty += 2 + RExC_naughty / 2;
6502 reginsert(pRExC_state, CURLY, ret, depth+1);
6503 Set_Node_Offset(ret, parse_start+1); /* MJD */
6504 Set_Node_Cur_Length(ret);
6507 regnode * const w = reg_node(pRExC_state, WHILEM);
6510 REGTAIL(pRExC_state, ret, w);
6511 if (!SIZE_ONLY && RExC_extralen) {
6512 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6513 reginsert(pRExC_state, NOTHING,ret, depth+1);
6514 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6516 reginsert(pRExC_state, CURLYX,ret, depth+1);
6518 Set_Node_Offset(ret, parse_start+1);
6519 Set_Node_Length(ret,
6520 op == '{' ? (RExC_parse - parse_start) : 1);
6522 if (!SIZE_ONLY && RExC_extralen)
6523 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6524 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6526 RExC_whilem_seen++, RExC_extralen += 3;
6527 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6536 vFAIL("Can't do {n,m} with n > m");
6538 ARG1_SET(ret, (U16)min);
6539 ARG2_SET(ret, (U16)max);
6551 #if 0 /* Now runtime fix should be reliable. */
6553 /* if this is reinstated, don't forget to put this back into perldiag:
6555 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6557 (F) The part of the regexp subject to either the * or + quantifier
6558 could match an empty string. The {#} shows in the regular
6559 expression about where the problem was discovered.
6563 if (!(flags&HASWIDTH) && op != '?')
6564 vFAIL("Regexp *+ operand could be empty");
6567 parse_start = RExC_parse;
6568 nextchar(pRExC_state);
6570 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6572 if (op == '*' && (flags&SIMPLE)) {
6573 reginsert(pRExC_state, STAR, ret, depth+1);
6577 else if (op == '*') {
6581 else if (op == '+' && (flags&SIMPLE)) {
6582 reginsert(pRExC_state, PLUS, ret, depth+1);
6586 else if (op == '+') {
6590 else if (op == '?') {
6595 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
6596 ckWARN3reg(RExC_parse,
6597 "%.*s matches null string many times",
6598 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6602 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6603 nextchar(pRExC_state);
6604 reginsert(pRExC_state, MINMOD, ret, depth+1);
6605 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6607 #ifndef REG_ALLOW_MINMOD_SUSPEND
6610 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6612 nextchar(pRExC_state);
6613 ender = reg_node(pRExC_state, SUCCEED);
6614 REGTAIL(pRExC_state, ret, ender);
6615 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6617 ender = reg_node(pRExC_state, TAIL);
6618 REGTAIL(pRExC_state, ret, ender);
6622 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6624 vFAIL("Nested quantifiers");
6631 /* reg_namedseq(pRExC_state,UVp)
6633 This is expected to be called by a parser routine that has
6634 recognized '\N' and needs to handle the rest. RExC_parse is
6635 expected to point at the first char following the N at the time
6638 The \N may be inside (indicated by valuep not being NULL) or outside a
6641 \N may begin either a named sequence, or if outside a character class, mean
6642 to match a non-newline. For non single-quoted regexes, the tokenizer has
6643 attempted to decide which, and in the case of a named sequence converted it
6644 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
6645 where c1... are the characters in the sequence. For single-quoted regexes,
6646 the tokenizer passes the \N sequence through unchanged; this code will not
6647 attempt to determine this nor expand those. The net effect is that if the
6648 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
6649 signals that this \N occurrence means to match a non-newline.
6651 Only the \N{U+...} form should occur in a character class, for the same
6652 reason that '.' inside a character class means to just match a period: it
6653 just doesn't make sense.
6655 If valuep is non-null then it is assumed that we are parsing inside
6656 of a charclass definition and the first codepoint in the resolved
6657 string is returned via *valuep and the routine will return NULL.
6658 In this mode if a multichar string is returned from the charnames
6659 handler, a warning will be issued, and only the first char in the
6660 sequence will be examined. If the string returned is zero length
6661 then the value of *valuep is undefined and NON-NULL will
6662 be returned to indicate failure. (This will NOT be a valid pointer
6665 If valuep is null then it is assumed that we are parsing normal text and a
6666 new EXACT node is inserted into the program containing the resolved string,
6667 and a pointer to the new node is returned. But if the string is zero length
6668 a NOTHING node is emitted instead.
6670 On success RExC_parse is set to the char following the endbrace.
6671 Parsing failures will generate a fatal error via vFAIL(...)
6674 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp)
6676 char * endbrace; /* '}' following the name */
6677 regnode *ret = NULL;
6679 char* parse_start = RExC_parse - 2; /* points to the '\N' */
6683 GET_RE_DEBUG_FLAGS_DECL;
6685 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6689 /* The [^\n] meaning of \N ignores spaces and comments under the /x
6690 * modifier. The other meaning does not */
6691 p = (RExC_flags & RXf_PMf_EXTENDED)
6692 ? regwhite( pRExC_state, RExC_parse )
6695 /* Disambiguate between \N meaning a named character versus \N meaning
6696 * [^\n]. The former is assumed when it can't be the latter. */
6697 if (*p != '{' || regcurly(p)) {
6700 /* no bare \N in a charclass */
6701 vFAIL("\\N in a character class must be a named character: \\N{...}");
6703 nextchar(pRExC_state);
6704 ret = reg_node(pRExC_state, REG_ANY);
6705 *flagp |= HASWIDTH|SIMPLE;
6708 Set_Node_Length(ret, 1); /* MJD */
6712 /* Here, we have decided it should be a named sequence */
6714 /* The test above made sure that the next real character is a '{', but
6715 * under the /x modifier, it could be separated by space (or a comment and
6716 * \n) and this is not allowed (for consistency with \x{...} and the
6717 * tokenizer handling of \N{NAME}). */
6718 if (*RExC_parse != '{') {
6719 vFAIL("Missing braces on \\N{}");
6722 RExC_parse++; /* Skip past the '{' */
6724 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
6725 || ! (endbrace == RExC_parse /* nothing between the {} */
6726 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
6727 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
6729 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
6730 vFAIL("\\N{NAME} must be resolved by the lexer");
6733 if (endbrace == RExC_parse) { /* empty: \N{} */
6735 RExC_parse = endbrace + 1;
6736 return reg_node(pRExC_state,NOTHING);
6740 ckWARNreg(RExC_parse,
6741 "Ignoring zero length \\N{} in character class"
6743 RExC_parse = endbrace + 1;
6746 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
6749 RExC_utf8 = 1; /* named sequences imply Unicode semantics */
6750 RExC_parse += 2; /* Skip past the 'U+' */
6752 if (valuep) { /* In a bracketed char class */
6753 /* We only pay attention to the first char of
6754 multichar strings being returned. I kinda wonder
6755 if this makes sense as it does change the behaviour
6756 from earlier versions, OTOH that behaviour was broken
6757 as well. XXX Solution is to recharacterize as
6758 [rest-of-class]|multi1|multi2... */
6760 STRLEN length_of_hex;
6761 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6762 | PERL_SCAN_DISALLOW_PREFIX
6763 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6765 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
6766 if (endchar < endbrace) {
6767 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
6770 length_of_hex = (STRLEN)(endchar - RExC_parse);
6771 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
6773 /* The tokenizer should have guaranteed validity, but it's possible to
6774 * bypass it by using single quoting, so check */
6775 if (length_of_hex == 0
6776 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6778 RExC_parse += length_of_hex; /* Includes all the valid */
6779 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6780 ? UTF8SKIP(RExC_parse)
6782 /* Guard against malformed utf8 */
6783 if (RExC_parse >= endchar) RExC_parse = endchar;
6784 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6787 RExC_parse = endbrace + 1;
6788 if (endchar == endbrace) return NULL;
6790 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
6792 else { /* Not a char class */
6793 char *s; /* String to put in generated EXACT node */
6794 STRLEN len = 0; /* Its current length */
6795 char *endchar; /* Points to '.' or '}' ending cur char in the input
6798 ret = reg_node(pRExC_state,
6799 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6802 /* Exact nodes can hold only a U8 length's of text = 255. Loop through
6803 * the input which is of the form now 'c1.c2.c3...}' until find the
6804 * ending brace or exeed length 255. The characters that exceed this
6805 * limit are dropped. The limit could be relaxed should it become
6806 * desirable by reparsing this as (?:\N{NAME}), so could generate
6807 * multiple EXACT nodes, as is done for just regular input. But this
6808 * is primarily a named character, and not intended to be a huge long
6809 * string, so 255 bytes should be good enough */
6811 STRLEN length_of_hex;
6812 I32 grok_flags = PERL_SCAN_ALLOW_UNDERSCORES
6813 | PERL_SCAN_DISALLOW_PREFIX
6814 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6815 UV cp; /* Ord of current character */
6817 /* Code points are separated by dots. If none, there is only one
6818 * code point, and is terminated by the brace */
6819 endchar = RExC_parse + strcspn(RExC_parse, ".}");
6821 /* The values are Unicode even on EBCDIC machines */
6822 length_of_hex = (STRLEN)(endchar - RExC_parse);
6823 cp = grok_hex(RExC_parse, &length_of_hex, &grok_flags, NULL);
6824 if ( length_of_hex == 0
6825 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6827 RExC_parse += length_of_hex; /* Includes all the valid */
6828 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6829 ? UTF8SKIP(RExC_parse)
6831 /* Guard against malformed utf8 */
6832 if (RExC_parse >= endchar) RExC_parse = endchar;
6833 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6836 if (! FOLD) { /* Not folding, just append to the string */
6839 /* Quit before adding this character if would exceed limit */
6840 if (len + UNISKIP(cp) > U8_MAX) break;
6842 unilen = reguni(pRExC_state, cp, s);
6847 } else { /* Folding, output the folded equivalent */
6848 STRLEN foldlen,numlen;
6849 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6850 cp = toFOLD_uni(cp, tmpbuf, &foldlen);
6852 /* Quit before exceeding size limit */
6853 if (len + foldlen > U8_MAX) break;
6855 for (foldbuf = tmpbuf;
6859 cp = utf8_to_uvchr(foldbuf, &numlen);
6861 const STRLEN unilen = reguni(pRExC_state, cp, s);
6864 /* In EBCDIC the numlen and unilen can differ. */
6866 if (numlen >= foldlen)
6870 break; /* "Can't happen." */
6874 /* Point to the beginning of the next character in the sequence. */
6875 RExC_parse = endchar + 1;
6877 /* Quit if no more characters */
6878 if (RExC_parse >= endbrace) break;
6883 if (RExC_parse < endbrace) {
6884 ckWARNreg(RExC_parse - 1,
6885 "Using just the first characters returned by \\N{}");
6888 RExC_size += STR_SZ(len);
6891 RExC_emit += STR_SZ(len);
6894 RExC_parse = endbrace + 1;
6896 *flagp |= HASWIDTH; /* Not SIMPLE, as that causes the engine to fail
6897 with malformed in t/re/pat_advanced.t */
6899 Set_Node_Cur_Length(ret); /* MJD */
6900 nextchar(pRExC_state);
6910 * It returns the code point in utf8 for the value in *encp.
6911 * value: a code value in the source encoding
6912 * encp: a pointer to an Encode object
6914 * If the result from Encode is not a single character,
6915 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6918 S_reg_recode(pTHX_ const char value, SV **encp)
6921 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
6922 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6923 const STRLEN newlen = SvCUR(sv);
6924 UV uv = UNICODE_REPLACEMENT;
6926 PERL_ARGS_ASSERT_REG_RECODE;
6930 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6933 if (!newlen || numlen != newlen) {
6934 uv = UNICODE_REPLACEMENT;
6942 - regatom - the lowest level
6944 Try to identify anything special at the start of the pattern. If there
6945 is, then handle it as required. This may involve generating a single regop,
6946 such as for an assertion; or it may involve recursing, such as to
6947 handle a () structure.
6949 If the string doesn't start with something special then we gobble up
6950 as much literal text as we can.
6952 Once we have been able to handle whatever type of thing started the
6953 sequence, we return.
6955 Note: we have to be careful with escapes, as they can be both literal
6956 and special, and in the case of \10 and friends can either, depending
6957 on context. Specifically there are two seperate switches for handling
6958 escape sequences, with the one for handling literal escapes requiring
6959 a dummy entry for all of the special escapes that are actually handled
6964 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6967 register regnode *ret = NULL;
6969 char *parse_start = RExC_parse;
6970 GET_RE_DEBUG_FLAGS_DECL;
6971 DEBUG_PARSE("atom");
6972 *flagp = WORST; /* Tentatively. */
6974 PERL_ARGS_ASSERT_REGATOM;
6977 switch ((U8)*RExC_parse) {
6979 RExC_seen_zerolen++;
6980 nextchar(pRExC_state);
6981 if (RExC_flags & RXf_PMf_MULTILINE)
6982 ret = reg_node(pRExC_state, MBOL);
6983 else if (RExC_flags & RXf_PMf_SINGLELINE)
6984 ret = reg_node(pRExC_state, SBOL);
6986 ret = reg_node(pRExC_state, BOL);
6987 Set_Node_Length(ret, 1); /* MJD */
6990 nextchar(pRExC_state);
6992 RExC_seen_zerolen++;
6993 if (RExC_flags & RXf_PMf_MULTILINE)
6994 ret = reg_node(pRExC_state, MEOL);
6995 else if (RExC_flags & RXf_PMf_SINGLELINE)
6996 ret = reg_node(pRExC_state, SEOL);
6998 ret = reg_node(pRExC_state, EOL);
6999 Set_Node_Length(ret, 1); /* MJD */
7002 nextchar(pRExC_state);
7003 if (RExC_flags & RXf_PMf_SINGLELINE)
7004 ret = reg_node(pRExC_state, SANY);
7006 ret = reg_node(pRExC_state, REG_ANY);
7007 *flagp |= HASWIDTH|SIMPLE;
7009 Set_Node_Length(ret, 1); /* MJD */
7013 char * const oregcomp_parse = ++RExC_parse;
7014 ret = regclass(pRExC_state,depth+1);
7015 if (*RExC_parse != ']') {
7016 RExC_parse = oregcomp_parse;
7017 vFAIL("Unmatched [");
7019 nextchar(pRExC_state);
7020 *flagp |= HASWIDTH|SIMPLE;
7021 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
7025 nextchar(pRExC_state);
7026 ret = reg(pRExC_state, 1, &flags,depth+1);
7028 if (flags & TRYAGAIN) {
7029 if (RExC_parse == RExC_end) {
7030 /* Make parent create an empty node if needed. */
7038 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7042 if (flags & TRYAGAIN) {
7046 vFAIL("Internal urp");
7047 /* Supposed to be caught earlier. */
7050 if (!regcurly(RExC_parse)) {
7059 vFAIL("Quantifier follows nothing");
7067 len=0; /* silence a spurious compiler warning */
7068 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
7069 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
7070 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
7071 ret = reganode(pRExC_state, FOLDCHAR, cp);
7072 Set_Node_Length(ret, 1); /* MJD */
7073 nextchar(pRExC_state); /* kill whitespace under /x */
7081 This switch handles escape sequences that resolve to some kind
7082 of special regop and not to literal text. Escape sequnces that
7083 resolve to literal text are handled below in the switch marked
7086 Every entry in this switch *must* have a corresponding entry
7087 in the literal escape switch. However, the opposite is not
7088 required, as the default for this switch is to jump to the
7089 literal text handling code.
7091 switch ((U8)*++RExC_parse) {
7096 /* Special Escapes */
7098 RExC_seen_zerolen++;
7099 ret = reg_node(pRExC_state, SBOL);
7101 goto finish_meta_pat;
7103 ret = reg_node(pRExC_state, GPOS);
7104 RExC_seen |= REG_SEEN_GPOS;
7106 goto finish_meta_pat;
7108 RExC_seen_zerolen++;
7109 ret = reg_node(pRExC_state, KEEPS);
7111 /* XXX:dmq : disabling in-place substitution seems to
7112 * be necessary here to avoid cases of memory corruption, as
7113 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7115 RExC_seen |= REG_SEEN_LOOKBEHIND;
7116 goto finish_meta_pat;
7118 ret = reg_node(pRExC_state, SEOL);
7120 RExC_seen_zerolen++; /* Do not optimize RE away */
7121 goto finish_meta_pat;
7123 ret = reg_node(pRExC_state, EOS);
7125 RExC_seen_zerolen++; /* Do not optimize RE away */
7126 goto finish_meta_pat;
7128 ret = reg_node(pRExC_state, CANY);
7129 RExC_seen |= REG_SEEN_CANY;
7130 *flagp |= HASWIDTH|SIMPLE;
7131 goto finish_meta_pat;
7133 ret = reg_node(pRExC_state, CLUMP);
7135 goto finish_meta_pat;
7137 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
7138 *flagp |= HASWIDTH|SIMPLE;
7139 goto finish_meta_pat;
7141 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
7142 *flagp |= HASWIDTH|SIMPLE;
7143 goto finish_meta_pat;
7145 RExC_seen_zerolen++;
7146 RExC_seen |= REG_SEEN_LOOKBEHIND;
7147 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
7149 goto finish_meta_pat;
7151 RExC_seen_zerolen++;
7152 RExC_seen |= REG_SEEN_LOOKBEHIND;
7153 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
7155 goto finish_meta_pat;
7157 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
7158 *flagp |= HASWIDTH|SIMPLE;
7159 goto finish_meta_pat;
7161 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
7162 *flagp |= HASWIDTH|SIMPLE;
7163 goto finish_meta_pat;
7165 ret = reg_node(pRExC_state, DIGIT);
7166 *flagp |= HASWIDTH|SIMPLE;
7167 goto finish_meta_pat;
7169 ret = reg_node(pRExC_state, NDIGIT);
7170 *flagp |= HASWIDTH|SIMPLE;
7171 goto finish_meta_pat;
7173 ret = reg_node(pRExC_state, LNBREAK);
7174 *flagp |= HASWIDTH|SIMPLE;
7175 goto finish_meta_pat;
7177 ret = reg_node(pRExC_state, HORIZWS);
7178 *flagp |= HASWIDTH|SIMPLE;
7179 goto finish_meta_pat;
7181 ret = reg_node(pRExC_state, NHORIZWS);
7182 *flagp |= HASWIDTH|SIMPLE;
7183 goto finish_meta_pat;
7185 ret = reg_node(pRExC_state, VERTWS);
7186 *flagp |= HASWIDTH|SIMPLE;
7187 goto finish_meta_pat;
7189 ret = reg_node(pRExC_state, NVERTWS);
7190 *flagp |= HASWIDTH|SIMPLE;
7192 nextchar(pRExC_state);
7193 Set_Node_Length(ret, 2); /* MJD */
7198 char* const oldregxend = RExC_end;
7200 char* parse_start = RExC_parse - 2;
7203 if (RExC_parse[1] == '{') {
7204 /* a lovely hack--pretend we saw [\pX] instead */
7205 RExC_end = strchr(RExC_parse, '}');
7207 const U8 c = (U8)*RExC_parse;
7209 RExC_end = oldregxend;
7210 vFAIL2("Missing right brace on \\%c{}", c);
7215 RExC_end = RExC_parse + 2;
7216 if (RExC_end > oldregxend)
7217 RExC_end = oldregxend;
7221 ret = regclass(pRExC_state,depth+1);
7223 RExC_end = oldregxend;
7226 Set_Node_Offset(ret, parse_start + 2);
7227 Set_Node_Cur_Length(ret);
7228 nextchar(pRExC_state);
7229 *flagp |= HASWIDTH|SIMPLE;
7233 /* Handle \N and \N{NAME} here and not below because it can be
7234 multicharacter. join_exact() will join them up later on.
7235 Also this makes sure that things like /\N{BLAH}+/ and
7236 \N{BLAH} being multi char Just Happen. dmq*/
7238 ret= reg_namedseq(pRExC_state, NULL, flagp);
7240 case 'k': /* Handle \k<NAME> and \k'NAME' */
7243 char ch= RExC_parse[1];
7244 if (ch != '<' && ch != '\'' && ch != '{') {
7246 vFAIL2("Sequence %.2s... not terminated",parse_start);
7248 /* this pretty much dupes the code for (?P=...) in reg(), if
7249 you change this make sure you change that */
7250 char* name_start = (RExC_parse += 2);
7252 SV *sv_dat = reg_scan_name(pRExC_state,
7253 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7254 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7255 if (RExC_parse == name_start || *RExC_parse != ch)
7256 vFAIL2("Sequence %.3s... not terminated",parse_start);
7259 num = add_data( pRExC_state, 1, "S" );
7260 RExC_rxi->data->data[num]=(void*)sv_dat;
7261 SvREFCNT_inc_simple_void(sv_dat);
7265 ret = reganode(pRExC_state,
7266 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7270 /* override incorrect value set in reganode MJD */
7271 Set_Node_Offset(ret, parse_start+1);
7272 Set_Node_Cur_Length(ret); /* MJD */
7273 nextchar(pRExC_state);
7279 case '1': case '2': case '3': case '4':
7280 case '5': case '6': case '7': case '8': case '9':
7283 bool isg = *RExC_parse == 'g';
7288 if (*RExC_parse == '{') {
7292 if (*RExC_parse == '-') {
7296 if (hasbrace && !isDIGIT(*RExC_parse)) {
7297 if (isrel) RExC_parse--;
7299 goto parse_named_seq;
7301 num = atoi(RExC_parse);
7302 if (isg && num == 0)
7303 vFAIL("Reference to invalid group 0");
7305 num = RExC_npar - num;
7307 vFAIL("Reference to nonexistent or unclosed group");
7309 if (!isg && num > 9 && num >= RExC_npar)
7312 char * const parse_start = RExC_parse - 1; /* MJD */
7313 while (isDIGIT(*RExC_parse))
7315 if (parse_start == RExC_parse - 1)
7316 vFAIL("Unterminated \\g... pattern");
7318 if (*RExC_parse != '}')
7319 vFAIL("Unterminated \\g{...} pattern");
7323 if (num > (I32)RExC_rx->nparens)
7324 vFAIL("Reference to nonexistent group");
7327 ret = reganode(pRExC_state,
7328 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7332 /* override incorrect value set in reganode MJD */
7333 Set_Node_Offset(ret, parse_start+1);
7334 Set_Node_Cur_Length(ret); /* MJD */
7336 nextchar(pRExC_state);
7341 if (RExC_parse >= RExC_end)
7342 FAIL("Trailing \\");
7345 /* Do not generate "unrecognized" warnings here, we fall
7346 back into the quick-grab loop below */
7353 if (RExC_flags & RXf_PMf_EXTENDED) {
7354 if ( reg_skipcomment( pRExC_state ) )
7361 register STRLEN len;
7366 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7368 parse_start = RExC_parse - 1;
7374 ret = reg_node(pRExC_state,
7375 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7377 for (len = 0, p = RExC_parse - 1;
7378 len < 127 && p < RExC_end;
7381 char * const oldp = p;
7383 if (RExC_flags & RXf_PMf_EXTENDED)
7384 p = regwhite( pRExC_state, p );
7389 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7390 goto normal_default;
7400 /* Literal Escapes Switch
7402 This switch is meant to handle escape sequences that
7403 resolve to a literal character.
7405 Every escape sequence that represents something
7406 else, like an assertion or a char class, is handled
7407 in the switch marked 'Special Escapes' above in this
7408 routine, but also has an entry here as anything that
7409 isn't explicitly mentioned here will be treated as
7410 an unescaped equivalent literal.
7414 /* These are all the special escapes. */
7418 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7419 goto normal_default;
7420 case 'A': /* Start assertion */
7421 case 'b': case 'B': /* Word-boundary assertion*/
7422 case 'C': /* Single char !DANGEROUS! */
7423 case 'd': case 'D': /* digit class */
7424 case 'g': case 'G': /* generic-backref, pos assertion */
7425 case 'h': case 'H': /* HORIZWS */
7426 case 'k': case 'K': /* named backref, keep marker */
7427 case 'N': /* named char sequence */
7428 case 'p': case 'P': /* Unicode property */
7429 case 'R': /* LNBREAK */
7430 case 's': case 'S': /* space class */
7431 case 'v': case 'V': /* VERTWS */
7432 case 'w': case 'W': /* word class */
7433 case 'X': /* eXtended Unicode "combining character sequence" */
7434 case 'z': case 'Z': /* End of line/string assertion */
7438 /* Anything after here is an escape that resolves to a
7439 literal. (Except digits, which may or may not)
7458 ender = ASCII_TO_NATIVE('\033');
7462 ender = ASCII_TO_NATIVE('\007');
7467 char* const e = strchr(p, '}');
7471 vFAIL("Missing right brace on \\x{}");
7474 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7475 | PERL_SCAN_DISALLOW_PREFIX;
7476 STRLEN numlen = e - p - 1;
7477 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7484 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7486 ender = grok_hex(p, &numlen, &flags, NULL);
7489 if (PL_encoding && ender < 0x100)
7490 goto recode_encoding;
7494 ender = grok_bslash_c(*p++, SIZE_ONLY);
7496 case '0': case '1': case '2': case '3':case '4':
7497 case '5': case '6': case '7': case '8':case '9':
7499 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7502 ender = grok_oct(p, &numlen, &flags, NULL);
7504 /* An octal above 0xff is interpreted differently
7505 * depending on if the re is in utf8 or not. If it
7506 * is in utf8, the value will be itself, otherwise
7507 * it is interpreted as modulo 0x100. It has been
7508 * decided to discourage the use of octal above the
7509 * single-byte range. For now, warn only when
7510 * it ends up modulo */
7511 if (SIZE_ONLY && ender >= 0x100
7512 && ! UTF && ! PL_encoding) {
7513 ckWARNregdep(p, "Use of octal value above 377 is deprecated");
7521 if (PL_encoding && ender < 0x100)
7522 goto recode_encoding;
7526 SV* enc = PL_encoding;
7527 ender = reg_recode((const char)(U8)ender, &enc);
7528 if (!enc && SIZE_ONLY)
7529 ckWARNreg(p, "Invalid escape in the specified encoding");
7535 FAIL("Trailing \\");
7538 if (!SIZE_ONLY&& isALPHA(*p))
7539 ckWARN2reg(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7540 goto normal_default;
7545 if (UTF8_IS_START(*p) && UTF) {
7547 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7548 &numlen, UTF8_ALLOW_DEFAULT);
7555 if ( RExC_flags & RXf_PMf_EXTENDED)
7556 p = regwhite( pRExC_state, p );
7558 /* Prime the casefolded buffer. */
7559 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7561 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7566 /* Emit all the Unicode characters. */
7568 for (foldbuf = tmpbuf;
7570 foldlen -= numlen) {
7571 ender = utf8_to_uvchr(foldbuf, &numlen);
7573 const STRLEN unilen = reguni(pRExC_state, ender, s);
7576 /* In EBCDIC the numlen
7577 * and unilen can differ. */
7579 if (numlen >= foldlen)
7583 break; /* "Can't happen." */
7587 const STRLEN unilen = reguni(pRExC_state, ender, s);
7596 REGC((char)ender, s++);
7602 /* Emit all the Unicode characters. */
7604 for (foldbuf = tmpbuf;
7606 foldlen -= numlen) {
7607 ender = utf8_to_uvchr(foldbuf, &numlen);
7609 const STRLEN unilen = reguni(pRExC_state, ender, s);
7612 /* In EBCDIC the numlen
7613 * and unilen can differ. */
7615 if (numlen >= foldlen)
7623 const STRLEN unilen = reguni(pRExC_state, ender, s);
7632 REGC((char)ender, s++);
7636 Set_Node_Cur_Length(ret); /* MJD */
7637 nextchar(pRExC_state);
7639 /* len is STRLEN which is unsigned, need to copy to signed */
7642 vFAIL("Internal disaster");
7646 if (len == 1 && UNI_IS_INVARIANT(ender))
7650 RExC_size += STR_SZ(len);
7653 RExC_emit += STR_SZ(len);
7663 S_regwhite( RExC_state_t *pRExC_state, char *p )
7665 const char *e = RExC_end;
7667 PERL_ARGS_ASSERT_REGWHITE;
7672 else if (*p == '#') {
7681 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7689 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7690 Character classes ([:foo:]) can also be negated ([:^foo:]).
7691 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7692 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7693 but trigger failures because they are currently unimplemented. */
7695 #define POSIXCC_DONE(c) ((c) == ':')
7696 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7697 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7700 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7703 I32 namedclass = OOB_NAMEDCLASS;
7705 PERL_ARGS_ASSERT_REGPPOSIXCC;
7707 if (value == '[' && RExC_parse + 1 < RExC_end &&
7708 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7709 POSIXCC(UCHARAT(RExC_parse))) {
7710 const char c = UCHARAT(RExC_parse);
7711 char* const s = RExC_parse++;
7713 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7715 if (RExC_parse == RExC_end)
7716 /* Grandfather lone [:, [=, [. */
7719 const char* const t = RExC_parse++; /* skip over the c */
7722 if (UCHARAT(RExC_parse) == ']') {
7723 const char *posixcc = s + 1;
7724 RExC_parse++; /* skip over the ending ] */
7727 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7728 const I32 skip = t - posixcc;
7730 /* Initially switch on the length of the name. */
7733 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7734 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7737 /* Names all of length 5. */
7738 /* alnum alpha ascii blank cntrl digit graph lower
7739 print punct space upper */
7740 /* Offset 4 gives the best switch position. */
7741 switch (posixcc[4]) {
7743 if (memEQ(posixcc, "alph", 4)) /* alpha */
7744 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7747 if (memEQ(posixcc, "spac", 4)) /* space */
7748 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7751 if (memEQ(posixcc, "grap", 4)) /* graph */
7752 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7755 if (memEQ(posixcc, "asci", 4)) /* ascii */
7756 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7759 if (memEQ(posixcc, "blan", 4)) /* blank */
7760 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7763 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7764 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7767 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7768 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7771 if (memEQ(posixcc, "lowe", 4)) /* lower */
7772 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7773 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7774 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7777 if (memEQ(posixcc, "digi", 4)) /* digit */
7778 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7779 else if (memEQ(posixcc, "prin", 4)) /* print */
7780 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7781 else if (memEQ(posixcc, "punc", 4)) /* punct */
7782 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7787 if (memEQ(posixcc, "xdigit", 6))
7788 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7792 if (namedclass == OOB_NAMEDCLASS)
7793 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7795 assert (posixcc[skip] == ':');
7796 assert (posixcc[skip+1] == ']');
7797 } else if (!SIZE_ONLY) {
7798 /* [[=foo=]] and [[.foo.]] are still future. */
7800 /* adjust RExC_parse so the warning shows after
7802 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7804 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7807 /* Maternal grandfather:
7808 * "[:" ending in ":" but not in ":]" */
7818 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7822 PERL_ARGS_ASSERT_CHECKPOSIXCC;
7824 if (POSIXCC(UCHARAT(RExC_parse))) {
7825 const char *s = RExC_parse;
7826 const char c = *s++;
7830 if (*s && c == *s && s[1] == ']') {
7832 "POSIX syntax [%c %c] belongs inside character classes",
7835 /* [[=foo=]] and [[.foo.]] are still future. */
7836 if (POSIXCC_NOTYET(c)) {
7837 /* adjust RExC_parse so the error shows after
7839 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7841 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7848 #define _C_C_T_(NAME,TEST,WORD) \
7851 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7853 for (value = 0; value < 256; value++) \
7855 ANYOF_BITMAP_SET(ret, value); \
7860 case ANYOF_N##NAME: \
7862 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7864 for (value = 0; value < 256; value++) \
7866 ANYOF_BITMAP_SET(ret, value); \
7872 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7874 for (value = 0; value < 256; value++) \
7876 ANYOF_BITMAP_SET(ret, value); \
7880 case ANYOF_N##NAME: \
7881 for (value = 0; value < 256; value++) \
7883 ANYOF_BITMAP_SET(ret, value); \
7889 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
7890 so that it is possible to override the option here without having to
7891 rebuild the entire core. as we are required to do if we change regcomp.h
7892 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
7894 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
7895 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
7898 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
7899 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
7901 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
7905 parse a class specification and produce either an ANYOF node that
7906 matches the pattern or if the pattern matches a single char only and
7907 that char is < 256 and we are case insensitive then we produce an
7912 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7915 register UV nextvalue;
7916 register IV prevvalue = OOB_UNICODE;
7917 register IV range = 0;
7918 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7919 register regnode *ret;
7922 char *rangebegin = NULL;
7923 bool need_class = 0;
7926 bool optimize_invert = TRUE;
7927 AV* unicode_alternate = NULL;
7929 UV literal_endpoint = 0;
7931 UV stored = 0; /* number of chars stored in the class */
7933 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7934 case we need to change the emitted regop to an EXACT. */
7935 const char * orig_parse = RExC_parse;
7936 GET_RE_DEBUG_FLAGS_DECL;
7938 PERL_ARGS_ASSERT_REGCLASS;
7940 PERL_UNUSED_ARG(depth);
7943 DEBUG_PARSE("clas");
7945 /* Assume we are going to generate an ANYOF node. */
7946 ret = reganode(pRExC_state, ANYOF, 0);
7949 ANYOF_FLAGS(ret) = 0;
7951 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7955 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7959 RExC_size += ANYOF_SKIP;
7960 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7963 RExC_emit += ANYOF_SKIP;
7965 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7967 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7968 ANYOF_BITMAP_ZERO(ret);
7969 listsv = newSVpvs("# comment\n");
7972 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7974 if (!SIZE_ONLY && POSIXCC(nextvalue))
7975 checkposixcc(pRExC_state);
7977 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7978 if (UCHARAT(RExC_parse) == ']')
7982 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
7986 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
7989 rangebegin = RExC_parse;
7991 value = utf8n_to_uvchr((U8*)RExC_parse,
7992 RExC_end - RExC_parse,
7993 &numlen, UTF8_ALLOW_DEFAULT);
7994 RExC_parse += numlen;
7997 value = UCHARAT(RExC_parse++);
7999 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8000 if (value == '[' && POSIXCC(nextvalue))
8001 namedclass = regpposixcc(pRExC_state, value);
8002 else if (value == '\\') {
8004 value = utf8n_to_uvchr((U8*)RExC_parse,
8005 RExC_end - RExC_parse,
8006 &numlen, UTF8_ALLOW_DEFAULT);
8007 RExC_parse += numlen;
8010 value = UCHARAT(RExC_parse++);
8011 /* Some compilers cannot handle switching on 64-bit integer
8012 * values, therefore value cannot be an UV. Yes, this will
8013 * be a problem later if we want switch on Unicode.
8014 * A similar issue a little bit later when switching on
8015 * namedclass. --jhi */
8016 switch ((I32)value) {
8017 case 'w': namedclass = ANYOF_ALNUM; break;
8018 case 'W': namedclass = ANYOF_NALNUM; break;
8019 case 's': namedclass = ANYOF_SPACE; break;
8020 case 'S': namedclass = ANYOF_NSPACE; break;
8021 case 'd': namedclass = ANYOF_DIGIT; break;
8022 case 'D': namedclass = ANYOF_NDIGIT; break;
8023 case 'v': namedclass = ANYOF_VERTWS; break;
8024 case 'V': namedclass = ANYOF_NVERTWS; break;
8025 case 'h': namedclass = ANYOF_HORIZWS; break;
8026 case 'H': namedclass = ANYOF_NHORIZWS; break;
8027 case 'N': /* Handle \N{NAME} in class */
8029 /* We only pay attention to the first char of
8030 multichar strings being returned. I kinda wonder
8031 if this makes sense as it does change the behaviour
8032 from earlier versions, OTOH that behaviour was broken
8034 UV v; /* value is register so we cant & it /grrr */
8035 if (reg_namedseq(pRExC_state, &v, NULL)) {
8045 if (RExC_parse >= RExC_end)
8046 vFAIL2("Empty \\%c{}", (U8)value);
8047 if (*RExC_parse == '{') {
8048 const U8 c = (U8)value;
8049 e = strchr(RExC_parse++, '}');
8051 vFAIL2("Missing right brace on \\%c{}", c);
8052 while (isSPACE(UCHARAT(RExC_parse)))
8054 if (e == RExC_parse)
8055 vFAIL2("Empty \\%c{}", c);
8057 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
8065 if (UCHARAT(RExC_parse) == '^') {
8068 value = value == 'p' ? 'P' : 'p'; /* toggle */
8069 while (isSPACE(UCHARAT(RExC_parse))) {
8074 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
8075 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
8078 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8079 namedclass = ANYOF_MAX; /* no official name, but it's named */
8082 case 'n': value = '\n'; break;
8083 case 'r': value = '\r'; break;
8084 case 't': value = '\t'; break;
8085 case 'f': value = '\f'; break;
8086 case 'b': value = '\b'; break;
8087 case 'e': value = ASCII_TO_NATIVE('\033');break;
8088 case 'a': value = ASCII_TO_NATIVE('\007');break;
8090 if (*RExC_parse == '{') {
8091 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8092 | PERL_SCAN_DISALLOW_PREFIX;
8093 char * const e = strchr(RExC_parse++, '}');
8095 vFAIL("Missing right brace on \\x{}");
8097 numlen = e - RExC_parse;
8098 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8102 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8104 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8105 RExC_parse += numlen;
8107 if (PL_encoding && value < 0x100)
8108 goto recode_encoding;
8111 value = grok_bslash_c(*RExC_parse++, SIZE_ONLY);
8113 case '0': case '1': case '2': case '3': case '4':
8114 case '5': case '6': case '7': case '8': case '9':
8118 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8119 RExC_parse += numlen;
8120 if (PL_encoding && value < 0x100)
8121 goto recode_encoding;
8126 SV* enc = PL_encoding;
8127 value = reg_recode((const char)(U8)value, &enc);
8128 if (!enc && SIZE_ONLY)
8129 ckWARNreg(RExC_parse,
8130 "Invalid escape in the specified encoding");
8134 if (!SIZE_ONLY && isALPHA(value))
8135 ckWARN2reg(RExC_parse,
8136 "Unrecognized escape \\%c in character class passed through",
8140 } /* end of \blah */
8146 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8148 if (!SIZE_ONLY && !need_class)
8149 ANYOF_CLASS_ZERO(ret);
8153 /* a bad range like a-\d, a-[:digit:] ? */
8157 RExC_parse >= rangebegin ?
8158 RExC_parse - rangebegin : 0;
8159 ckWARN4reg(RExC_parse,
8160 "False [] range \"%*.*s\"",
8163 if (prevvalue < 256) {
8164 ANYOF_BITMAP_SET(ret, prevvalue);
8165 ANYOF_BITMAP_SET(ret, '-');
8168 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8169 Perl_sv_catpvf(aTHX_ listsv,
8170 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8174 range = 0; /* this was not a true range */
8180 const char *what = NULL;
8183 if (namedclass > OOB_NAMEDCLASS)
8184 optimize_invert = FALSE;
8185 /* Possible truncation here but in some 64-bit environments
8186 * the compiler gets heartburn about switch on 64-bit values.
8187 * A similar issue a little earlier when switching on value.
8189 switch ((I32)namedclass) {
8191 case _C_C_T_(ALNUMC, isALNUMC(value), POSIX_CC_UNI_NAME("Alnum"));
8192 case _C_C_T_(ALPHA, isALPHA(value), POSIX_CC_UNI_NAME("Alpha"));
8193 case _C_C_T_(BLANK, isBLANK(value), POSIX_CC_UNI_NAME("Blank"));
8194 case _C_C_T_(CNTRL, isCNTRL(value), POSIX_CC_UNI_NAME("Cntrl"));
8195 case _C_C_T_(GRAPH, isGRAPH(value), POSIX_CC_UNI_NAME("Graph"));
8196 case _C_C_T_(LOWER, isLOWER(value), POSIX_CC_UNI_NAME("Lower"));
8197 case _C_C_T_(PRINT, isPRINT(value), POSIX_CC_UNI_NAME("Print"));
8198 case _C_C_T_(PSXSPC, isPSXSPC(value), POSIX_CC_UNI_NAME("Space"));
8199 case _C_C_T_(PUNCT, isPUNCT(value), POSIX_CC_UNI_NAME("Punct"));
8200 case _C_C_T_(UPPER, isUPPER(value), POSIX_CC_UNI_NAME("Upper"));
8201 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8202 case _C_C_T_(ALNUM, isALNUM(value), "Word");
8203 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
8205 case _C_C_T_(SPACE, isSPACE(value), "PerlSpace");
8206 case _C_C_T_(ALNUM, isALNUM(value), "PerlWord");
8208 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
8209 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8210 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8213 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8216 for (value = 0; value < 128; value++)
8217 ANYOF_BITMAP_SET(ret, value);
8219 for (value = 0; value < 256; value++) {
8221 ANYOF_BITMAP_SET(ret, value);
8230 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8233 for (value = 128; value < 256; value++)
8234 ANYOF_BITMAP_SET(ret, value);
8236 for (value = 0; value < 256; value++) {
8237 if (!isASCII(value))
8238 ANYOF_BITMAP_SET(ret, value);
8247 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8249 /* consecutive digits assumed */
8250 for (value = '0'; value <= '9'; value++)
8251 ANYOF_BITMAP_SET(ret, value);
8254 what = POSIX_CC_UNI_NAME("Digit");
8258 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8260 /* consecutive digits assumed */
8261 for (value = 0; value < '0'; value++)
8262 ANYOF_BITMAP_SET(ret, value);
8263 for (value = '9' + 1; value < 256; value++)
8264 ANYOF_BITMAP_SET(ret, value);
8267 what = POSIX_CC_UNI_NAME("Digit");
8270 /* this is to handle \p and \P */
8273 vFAIL("Invalid [::] class");
8277 /* Strings such as "+utf8::isWord\n" */
8278 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8281 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8284 } /* end of namedclass \blah */
8287 if (prevvalue > (IV)value) /* b-a */ {
8288 const int w = RExC_parse - rangebegin;
8289 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8290 range = 0; /* not a valid range */
8294 prevvalue = value; /* save the beginning of the range */
8295 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8296 RExC_parse[1] != ']') {
8299 /* a bad range like \w-, [:word:]- ? */
8300 if (namedclass > OOB_NAMEDCLASS) {
8301 if (ckWARN(WARN_REGEXP)) {
8303 RExC_parse >= rangebegin ?
8304 RExC_parse - rangebegin : 0;
8306 "False [] range \"%*.*s\"",
8310 ANYOF_BITMAP_SET(ret, '-');
8312 range = 1; /* yeah, it's a range! */
8313 continue; /* but do it the next time */
8317 /* now is the next time */
8318 /*stored += (value - prevvalue + 1);*/
8320 if (prevvalue < 256) {
8321 const IV ceilvalue = value < 256 ? value : 255;
8324 /* In EBCDIC [\x89-\x91] should include
8325 * the \x8e but [i-j] should not. */
8326 if (literal_endpoint == 2 &&
8327 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8328 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8330 if (isLOWER(prevvalue)) {
8331 for (i = prevvalue; i <= ceilvalue; i++)
8332 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8334 ANYOF_BITMAP_SET(ret, i);
8337 for (i = prevvalue; i <= ceilvalue; i++)
8338 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8340 ANYOF_BITMAP_SET(ret, i);
8346 for (i = prevvalue; i <= ceilvalue; i++) {
8347 if (!ANYOF_BITMAP_TEST(ret,i)) {
8349 ANYOF_BITMAP_SET(ret, i);
8353 if (value > 255 || UTF) {
8354 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8355 const UV natvalue = NATIVE_TO_UNI(value);
8356 stored+=2; /* can't optimize this class */
8357 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8358 if (prevnatvalue < natvalue) { /* what about > ? */
8359 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8360 prevnatvalue, natvalue);
8362 else if (prevnatvalue == natvalue) {
8363 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8365 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8367 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8369 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8370 if (RExC_precomp[0] == ':' &&
8371 RExC_precomp[1] == '[' &&
8372 (f == 0xDF || f == 0x92)) {
8373 f = NATIVE_TO_UNI(f);
8376 /* If folding and foldable and a single
8377 * character, insert also the folded version
8378 * to the charclass. */
8380 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8381 if ((RExC_precomp[0] == ':' &&
8382 RExC_precomp[1] == '[' &&
8384 (value == 0xFB05 || value == 0xFB06))) ?
8385 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8386 foldlen == (STRLEN)UNISKIP(f) )
8388 if (foldlen == (STRLEN)UNISKIP(f))
8390 Perl_sv_catpvf(aTHX_ listsv,
8393 /* Any multicharacter foldings
8394 * require the following transform:
8395 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8396 * where E folds into "pq" and F folds
8397 * into "rst", all other characters
8398 * fold to single characters. We save
8399 * away these multicharacter foldings,
8400 * to be later saved as part of the
8401 * additional "s" data. */
8404 if (!unicode_alternate)
8405 unicode_alternate = newAV();
8406 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8408 av_push(unicode_alternate, sv);
8412 /* If folding and the value is one of the Greek
8413 * sigmas insert a few more sigmas to make the
8414 * folding rules of the sigmas to work right.
8415 * Note that not all the possible combinations
8416 * are handled here: some of them are handled
8417 * by the standard folding rules, and some of
8418 * them (literal or EXACTF cases) are handled
8419 * during runtime in regexec.c:S_find_byclass(). */
8420 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8421 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8422 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8423 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8424 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8426 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8427 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8428 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8433 literal_endpoint = 0;
8437 range = 0; /* this range (if it was one) is done now */
8441 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8443 RExC_size += ANYOF_CLASS_ADD_SKIP;
8445 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8451 /****** !SIZE_ONLY AFTER HERE *********/
8453 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8454 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8456 /* optimize single char class to an EXACT node
8457 but *only* when its not a UTF/high char */
8458 const char * cur_parse= RExC_parse;
8459 RExC_emit = (regnode *)orig_emit;
8460 RExC_parse = (char *)orig_parse;
8461 ret = reg_node(pRExC_state,
8462 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8463 RExC_parse = (char *)cur_parse;
8464 *STRING(ret)= (char)value;
8466 RExC_emit += STR_SZ(1);
8467 SvREFCNT_dec(listsv);
8470 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8471 if ( /* If the only flag is folding (plus possibly inversion). */
8472 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8474 for (value = 0; value < 256; ++value) {
8475 if (ANYOF_BITMAP_TEST(ret, value)) {
8476 UV fold = PL_fold[value];
8479 ANYOF_BITMAP_SET(ret, fold);
8482 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8485 /* optimize inverted simple patterns (e.g. [^a-z]) */
8486 if (optimize_invert &&
8487 /* If the only flag is inversion. */
8488 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8489 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8490 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8491 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8494 AV * const av = newAV();
8496 /* The 0th element stores the character class description
8497 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8498 * to initialize the appropriate swash (which gets stored in
8499 * the 1st element), and also useful for dumping the regnode.
8500 * The 2nd element stores the multicharacter foldings,
8501 * used later (regexec.c:S_reginclass()). */
8502 av_store(av, 0, listsv);
8503 av_store(av, 1, NULL);
8504 av_store(av, 2, MUTABLE_SV(unicode_alternate));
8505 rv = newRV_noinc(MUTABLE_SV(av));
8506 n = add_data(pRExC_state, 1, "s");
8507 RExC_rxi->data->data[n] = (void*)rv;
8515 /* reg_skipcomment()
8517 Absorbs an /x style # comments from the input stream.
8518 Returns true if there is more text remaining in the stream.
8519 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8520 terminates the pattern without including a newline.
8522 Note its the callers responsibility to ensure that we are
8528 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8532 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8534 while (RExC_parse < RExC_end)
8535 if (*RExC_parse++ == '\n') {
8540 /* we ran off the end of the pattern without ending
8541 the comment, so we have to add an \n when wrapping */
8542 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8550 Advance that parse position, and optionall absorbs
8551 "whitespace" from the inputstream.
8553 Without /x "whitespace" means (?#...) style comments only,
8554 with /x this means (?#...) and # comments and whitespace proper.
8556 Returns the RExC_parse point from BEFORE the scan occurs.
8558 This is the /x friendly way of saying RExC_parse++.
8562 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8564 char* const retval = RExC_parse++;
8566 PERL_ARGS_ASSERT_NEXTCHAR;
8569 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8570 RExC_parse[2] == '#') {
8571 while (*RExC_parse != ')') {
8572 if (RExC_parse == RExC_end)
8573 FAIL("Sequence (?#... not terminated");
8579 if (RExC_flags & RXf_PMf_EXTENDED) {
8580 if (isSPACE(*RExC_parse)) {
8584 else if (*RExC_parse == '#') {
8585 if ( reg_skipcomment( pRExC_state ) )
8594 - reg_node - emit a node
8596 STATIC regnode * /* Location. */
8597 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8600 register regnode *ptr;
8601 regnode * const ret = RExC_emit;
8602 GET_RE_DEBUG_FLAGS_DECL;
8604 PERL_ARGS_ASSERT_REG_NODE;
8607 SIZE_ALIGN(RExC_size);
8611 if (RExC_emit >= RExC_emit_bound)
8612 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8614 NODE_ALIGN_FILL(ret);
8616 FILL_ADVANCE_NODE(ptr, op);
8617 #ifdef RE_TRACK_PATTERN_OFFSETS
8618 if (RExC_offsets) { /* MJD */
8619 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8620 "reg_node", __LINE__,
8622 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8623 ? "Overwriting end of array!\n" : "OK",
8624 (UV)(RExC_emit - RExC_emit_start),
8625 (UV)(RExC_parse - RExC_start),
8626 (UV)RExC_offsets[0]));
8627 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8635 - reganode - emit a node with an argument
8637 STATIC regnode * /* Location. */
8638 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8641 register regnode *ptr;
8642 regnode * const ret = RExC_emit;
8643 GET_RE_DEBUG_FLAGS_DECL;
8645 PERL_ARGS_ASSERT_REGANODE;
8648 SIZE_ALIGN(RExC_size);
8653 assert(2==regarglen[op]+1);
8655 Anything larger than this has to allocate the extra amount.
8656 If we changed this to be:
8658 RExC_size += (1 + regarglen[op]);
8660 then it wouldn't matter. Its not clear what side effect
8661 might come from that so its not done so far.
8666 if (RExC_emit >= RExC_emit_bound)
8667 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8669 NODE_ALIGN_FILL(ret);
8671 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8672 #ifdef RE_TRACK_PATTERN_OFFSETS
8673 if (RExC_offsets) { /* MJD */
8674 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8678 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8679 "Overwriting end of array!\n" : "OK",
8680 (UV)(RExC_emit - RExC_emit_start),
8681 (UV)(RExC_parse - RExC_start),
8682 (UV)RExC_offsets[0]));
8683 Set_Cur_Node_Offset;
8691 - reguni - emit (if appropriate) a Unicode character
8694 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8698 PERL_ARGS_ASSERT_REGUNI;
8700 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8704 - reginsert - insert an operator in front of already-emitted operand
8706 * Means relocating the operand.
8709 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8712 register regnode *src;
8713 register regnode *dst;
8714 register regnode *place;
8715 const int offset = regarglen[(U8)op];
8716 const int size = NODE_STEP_REGNODE + offset;
8717 GET_RE_DEBUG_FLAGS_DECL;
8719 PERL_ARGS_ASSERT_REGINSERT;
8720 PERL_UNUSED_ARG(depth);
8721 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8722 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8731 if (RExC_open_parens) {
8733 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8734 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8735 if ( RExC_open_parens[paren] >= opnd ) {
8736 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8737 RExC_open_parens[paren] += size;
8739 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8741 if ( RExC_close_parens[paren] >= opnd ) {
8742 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8743 RExC_close_parens[paren] += size;
8745 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8750 while (src > opnd) {
8751 StructCopy(--src, --dst, regnode);
8752 #ifdef RE_TRACK_PATTERN_OFFSETS
8753 if (RExC_offsets) { /* MJD 20010112 */
8754 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8758 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8759 ? "Overwriting end of array!\n" : "OK",
8760 (UV)(src - RExC_emit_start),
8761 (UV)(dst - RExC_emit_start),
8762 (UV)RExC_offsets[0]));
8763 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8764 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8770 place = opnd; /* Op node, where operand used to be. */
8771 #ifdef RE_TRACK_PATTERN_OFFSETS
8772 if (RExC_offsets) { /* MJD */
8773 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8777 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8778 ? "Overwriting end of array!\n" : "OK",
8779 (UV)(place - RExC_emit_start),
8780 (UV)(RExC_parse - RExC_start),
8781 (UV)RExC_offsets[0]));
8782 Set_Node_Offset(place, RExC_parse);
8783 Set_Node_Length(place, 1);
8786 src = NEXTOPER(place);
8787 FILL_ADVANCE_NODE(place, op);
8788 Zero(src, offset, regnode);
8792 - regtail - set the next-pointer at the end of a node chain of p to val.
8793 - SEE ALSO: regtail_study
8795 /* TODO: All three parms should be const */
8797 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8800 register regnode *scan;
8801 GET_RE_DEBUG_FLAGS_DECL;
8803 PERL_ARGS_ASSERT_REGTAIL;
8805 PERL_UNUSED_ARG(depth);
8811 /* Find last node. */
8814 regnode * const temp = regnext(scan);
8816 SV * const mysv=sv_newmortal();
8817 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8818 regprop(RExC_rx, mysv, scan);
8819 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8820 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8821 (temp == NULL ? "->" : ""),
8822 (temp == NULL ? PL_reg_name[OP(val)] : "")
8830 if (reg_off_by_arg[OP(scan)]) {
8831 ARG_SET(scan, val - scan);
8834 NEXT_OFF(scan) = val - scan;
8840 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8841 - Look for optimizable sequences at the same time.
8842 - currently only looks for EXACT chains.
8844 This is expermental code. The idea is to use this routine to perform
8845 in place optimizations on branches and groups as they are constructed,
8846 with the long term intention of removing optimization from study_chunk so
8847 that it is purely analytical.
8849 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8850 to control which is which.
8853 /* TODO: All four parms should be const */
8856 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8859 register regnode *scan;
8861 #ifdef EXPERIMENTAL_INPLACESCAN
8864 GET_RE_DEBUG_FLAGS_DECL;
8866 PERL_ARGS_ASSERT_REGTAIL_STUDY;
8872 /* Find last node. */
8876 regnode * const temp = regnext(scan);
8877 #ifdef EXPERIMENTAL_INPLACESCAN
8878 if (PL_regkind[OP(scan)] == EXACT)
8879 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8887 if( exact == PSEUDO )
8889 else if ( exact != OP(scan) )
8898 SV * const mysv=sv_newmortal();
8899 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8900 regprop(RExC_rx, mysv, scan);
8901 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8902 SvPV_nolen_const(mysv),
8904 PL_reg_name[exact]);
8911 SV * const mysv_val=sv_newmortal();
8912 DEBUG_PARSE_MSG("");
8913 regprop(RExC_rx, mysv_val, val);
8914 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8915 SvPV_nolen_const(mysv_val),
8916 (IV)REG_NODE_NUM(val),
8920 if (reg_off_by_arg[OP(scan)]) {
8921 ARG_SET(scan, val - scan);
8924 NEXT_OFF(scan) = val - scan;
8932 - regcurly - a little FSA that accepts {\d+,?\d*}
8934 #ifndef PERL_IN_XSUB_RE
8936 Perl_regcurly(register const char *s)
8938 PERL_ARGS_ASSERT_REGCURLY;
8957 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8961 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
8966 for (bit=0; bit<32; bit++) {
8967 if (flags & (1<<bit)) {
8969 PerlIO_printf(Perl_debug_log, "%s",lead);
8970 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
8975 PerlIO_printf(Perl_debug_log, "\n");
8977 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
8983 Perl_regdump(pTHX_ const regexp *r)
8987 SV * const sv = sv_newmortal();
8988 SV *dsv= sv_newmortal();
8990 GET_RE_DEBUG_FLAGS_DECL;
8992 PERL_ARGS_ASSERT_REGDUMP;
8994 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
8996 /* Header fields of interest. */
8997 if (r->anchored_substr) {
8998 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
8999 RE_SV_DUMPLEN(r->anchored_substr), 30);
9000 PerlIO_printf(Perl_debug_log,
9001 "anchored %s%s at %"IVdf" ",
9002 s, RE_SV_TAIL(r->anchored_substr),
9003 (IV)r->anchored_offset);
9004 } else if (r->anchored_utf8) {
9005 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
9006 RE_SV_DUMPLEN(r->anchored_utf8), 30);
9007 PerlIO_printf(Perl_debug_log,
9008 "anchored utf8 %s%s at %"IVdf" ",
9009 s, RE_SV_TAIL(r->anchored_utf8),
9010 (IV)r->anchored_offset);
9012 if (r->float_substr) {
9013 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
9014 RE_SV_DUMPLEN(r->float_substr), 30);
9015 PerlIO_printf(Perl_debug_log,
9016 "floating %s%s at %"IVdf"..%"UVuf" ",
9017 s, RE_SV_TAIL(r->float_substr),
9018 (IV)r->float_min_offset, (UV)r->float_max_offset);
9019 } else if (r->float_utf8) {
9020 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
9021 RE_SV_DUMPLEN(r->float_utf8), 30);
9022 PerlIO_printf(Perl_debug_log,
9023 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
9024 s, RE_SV_TAIL(r->float_utf8),
9025 (IV)r->float_min_offset, (UV)r->float_max_offset);
9027 if (r->check_substr || r->check_utf8)
9028 PerlIO_printf(Perl_debug_log,
9030 (r->check_substr == r->float_substr
9031 && r->check_utf8 == r->float_utf8
9032 ? "(checking floating" : "(checking anchored"));
9033 if (r->extflags & RXf_NOSCAN)
9034 PerlIO_printf(Perl_debug_log, " noscan");
9035 if (r->extflags & RXf_CHECK_ALL)
9036 PerlIO_printf(Perl_debug_log, " isall");
9037 if (r->check_substr || r->check_utf8)
9038 PerlIO_printf(Perl_debug_log, ") ");
9040 if (ri->regstclass) {
9041 regprop(r, sv, ri->regstclass);
9042 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
9044 if (r->extflags & RXf_ANCH) {
9045 PerlIO_printf(Perl_debug_log, "anchored");
9046 if (r->extflags & RXf_ANCH_BOL)
9047 PerlIO_printf(Perl_debug_log, "(BOL)");
9048 if (r->extflags & RXf_ANCH_MBOL)
9049 PerlIO_printf(Perl_debug_log, "(MBOL)");
9050 if (r->extflags & RXf_ANCH_SBOL)
9051 PerlIO_printf(Perl_debug_log, "(SBOL)");
9052 if (r->extflags & RXf_ANCH_GPOS)
9053 PerlIO_printf(Perl_debug_log, "(GPOS)");
9054 PerlIO_putc(Perl_debug_log, ' ');
9056 if (r->extflags & RXf_GPOS_SEEN)
9057 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
9058 if (r->intflags & PREGf_SKIP)
9059 PerlIO_printf(Perl_debug_log, "plus ");
9060 if (r->intflags & PREGf_IMPLICIT)
9061 PerlIO_printf(Perl_debug_log, "implicit ");
9062 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
9063 if (r->extflags & RXf_EVAL_SEEN)
9064 PerlIO_printf(Perl_debug_log, "with eval ");
9065 PerlIO_printf(Perl_debug_log, "\n");
9066 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
9068 PERL_ARGS_ASSERT_REGDUMP;
9069 PERL_UNUSED_CONTEXT;
9071 #endif /* DEBUGGING */
9075 - regprop - printable representation of opcode
9077 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9080 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9081 if (flags & ANYOF_INVERT) \
9082 /*make sure the invert info is in each */ \
9083 sv_catpvs(sv, "^"); \
9089 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9094 RXi_GET_DECL(prog,progi);
9095 GET_RE_DEBUG_FLAGS_DECL;
9097 PERL_ARGS_ASSERT_REGPROP;
9101 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9102 /* It would be nice to FAIL() here, but this may be called from
9103 regexec.c, and it would be hard to supply pRExC_state. */
9104 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9105 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9107 k = PL_regkind[OP(o)];
9111 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9112 * is a crude hack but it may be the best for now since
9113 * we have no flag "this EXACTish node was UTF-8"
9115 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9116 PERL_PV_ESCAPE_UNI_DETECT |
9117 PERL_PV_PRETTY_ELLIPSES |
9118 PERL_PV_PRETTY_LTGT |
9119 PERL_PV_PRETTY_NOCLEAR
9121 } else if (k == TRIE) {
9122 /* print the details of the trie in dumpuntil instead, as
9123 * progi->data isn't available here */
9124 const char op = OP(o);
9125 const U32 n = ARG(o);
9126 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9127 (reg_ac_data *)progi->data->data[n] :
9129 const reg_trie_data * const trie
9130 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9132 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9133 DEBUG_TRIE_COMPILE_r(
9134 Perl_sv_catpvf(aTHX_ sv,
9135 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9136 (UV)trie->startstate,
9137 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9138 (UV)trie->wordcount,
9141 (UV)TRIE_CHARCOUNT(trie),
9142 (UV)trie->uniquecharcount
9145 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9147 int rangestart = -1;
9148 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9150 for (i = 0; i <= 256; i++) {
9151 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9152 if (rangestart == -1)
9154 } else if (rangestart != -1) {
9155 if (i <= rangestart + 3)
9156 for (; rangestart < i; rangestart++)
9157 put_byte(sv, rangestart);
9159 put_byte(sv, rangestart);
9161 put_byte(sv, i - 1);
9169 } else if (k == CURLY) {
9170 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9171 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9172 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9174 else if (k == WHILEM && o->flags) /* Ordinal/of */
9175 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9176 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9177 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9178 if ( RXp_PAREN_NAMES(prog) ) {
9179 if ( k != REF || OP(o) < NREF) {
9180 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9181 SV **name= av_fetch(list, ARG(o), 0 );
9183 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9186 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9187 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9188 I32 *nums=(I32*)SvPVX(sv_dat);
9189 SV **name= av_fetch(list, nums[0], 0 );
9192 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9193 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9194 (n ? "," : ""), (IV)nums[n]);
9196 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9200 } else if (k == GOSUB)
9201 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9202 else if (k == VERB) {
9204 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9205 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9206 } else if (k == LOGICAL)
9207 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9208 else if (k == FOLDCHAR)
9209 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9210 else if (k == ANYOF) {
9211 int i, rangestart = -1;
9212 const U8 flags = ANYOF_FLAGS(o);
9215 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9216 static const char * const anyofs[] = {
9249 if (flags & ANYOF_LOCALE)
9250 sv_catpvs(sv, "{loc}");
9251 if (flags & ANYOF_FOLD)
9252 sv_catpvs(sv, "{i}");
9253 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9254 if (flags & ANYOF_INVERT)
9257 /* output what the standard cp 0-255 bitmap matches */
9258 for (i = 0; i <= 256; i++) {
9259 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9260 if (rangestart == -1)
9262 } else if (rangestart != -1) {
9263 if (i <= rangestart + 3)
9264 for (; rangestart < i; rangestart++)
9265 put_byte(sv, rangestart);
9267 put_byte(sv, rangestart);
9269 put_byte(sv, i - 1);
9276 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9277 /* output any special charclass tests (used mostly under use locale) */
9278 if (o->flags & ANYOF_CLASS)
9279 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9280 if (ANYOF_CLASS_TEST(o,i)) {
9281 sv_catpv(sv, anyofs[i]);
9285 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9287 /* output information about the unicode matching */
9288 if (flags & ANYOF_UNICODE)
9289 sv_catpvs(sv, "{unicode}");
9290 else if (flags & ANYOF_UNICODE_ALL)
9291 sv_catpvs(sv, "{unicode_all}");
9295 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9299 U8 s[UTF8_MAXBYTES_CASE+1];
9301 for (i = 0; i <= 256; i++) { /* just the first 256 */
9302 uvchr_to_utf8(s, i);
9304 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9305 if (rangestart == -1)
9307 } else if (rangestart != -1) {
9308 if (i <= rangestart + 3)
9309 for (; rangestart < i; rangestart++) {
9310 const U8 * const e = uvchr_to_utf8(s,rangestart);
9312 for(p = s; p < e; p++)
9316 const U8 *e = uvchr_to_utf8(s,rangestart);
9318 for (p = s; p < e; p++)
9321 e = uvchr_to_utf8(s, i-1);
9322 for (p = s; p < e; p++)
9329 sv_catpvs(sv, "..."); /* et cetera */
9333 char *s = savesvpv(lv);
9334 char * const origs = s;
9336 while (*s && *s != '\n')
9340 const char * const t = ++s;
9358 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9360 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9361 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9363 PERL_UNUSED_CONTEXT;
9364 PERL_UNUSED_ARG(sv);
9366 PERL_UNUSED_ARG(prog);
9367 #endif /* DEBUGGING */
9371 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9372 { /* Assume that RE_INTUIT is set */
9374 struct regexp *const prog = (struct regexp *)SvANY(r);
9375 GET_RE_DEBUG_FLAGS_DECL;
9377 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9378 PERL_UNUSED_CONTEXT;
9382 const char * const s = SvPV_nolen_const(prog->check_substr
9383 ? prog->check_substr : prog->check_utf8);
9385 if (!PL_colorset) reginitcolors();
9386 PerlIO_printf(Perl_debug_log,
9387 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9389 prog->check_substr ? "" : "utf8 ",
9390 PL_colors[5],PL_colors[0],
9393 (strlen(s) > 60 ? "..." : ""));
9396 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9402 handles refcounting and freeing the perl core regexp structure. When
9403 it is necessary to actually free the structure the first thing it
9404 does is call the 'free' method of the regexp_engine associated to to
9405 the regexp, allowing the handling of the void *pprivate; member
9406 first. (This routine is not overridable by extensions, which is why
9407 the extensions free is called first.)
9409 See regdupe and regdupe_internal if you change anything here.
9411 #ifndef PERL_IN_XSUB_RE
9413 Perl_pregfree(pTHX_ REGEXP *r)
9419 Perl_pregfree2(pTHX_ REGEXP *rx)
9422 struct regexp *const r = (struct regexp *)SvANY(rx);
9423 GET_RE_DEBUG_FLAGS_DECL;
9425 PERL_ARGS_ASSERT_PREGFREE2;
9428 ReREFCNT_dec(r->mother_re);
9430 CALLREGFREE_PVT(rx); /* free the private data */
9431 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9434 SvREFCNT_dec(r->anchored_substr);
9435 SvREFCNT_dec(r->anchored_utf8);
9436 SvREFCNT_dec(r->float_substr);
9437 SvREFCNT_dec(r->float_utf8);
9438 Safefree(r->substrs);
9440 RX_MATCH_COPY_FREE(rx);
9441 #ifdef PERL_OLD_COPY_ON_WRITE
9442 SvREFCNT_dec(r->saved_copy);
9449 This is a hacky workaround to the structural issue of match results
9450 being stored in the regexp structure which is in turn stored in
9451 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9452 could be PL_curpm in multiple contexts, and could require multiple
9453 result sets being associated with the pattern simultaneously, such
9454 as when doing a recursive match with (??{$qr})
9456 The solution is to make a lightweight copy of the regexp structure
9457 when a qr// is returned from the code executed by (??{$qr}) this
9458 lightweight copy doesnt actually own any of its data except for
9459 the starp/end and the actual regexp structure itself.
9465 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
9468 struct regexp *const r = (struct regexp *)SvANY(rx);
9469 register const I32 npar = r->nparens+1;
9471 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9474 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9475 ret = (struct regexp *)SvANY(ret_x);
9477 (void)ReREFCNT_inc(rx);
9478 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9479 by pointing directly at the buffer, but flagging that the allocated
9480 space in the copy is zero. As we've just done a struct copy, it's now
9481 a case of zero-ing that, rather than copying the current length. */
9482 SvPV_set(ret_x, RX_WRAPPED(rx));
9483 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9484 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
9485 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
9486 SvLEN_set(ret_x, 0);
9487 SvSTASH_set(ret_x, NULL);
9488 SvMAGIC_set(ret_x, NULL);
9489 Newx(ret->offs, npar, regexp_paren_pair);
9490 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9492 Newx(ret->substrs, 1, struct reg_substr_data);
9493 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9495 SvREFCNT_inc_void(ret->anchored_substr);
9496 SvREFCNT_inc_void(ret->anchored_utf8);
9497 SvREFCNT_inc_void(ret->float_substr);
9498 SvREFCNT_inc_void(ret->float_utf8);
9500 /* check_substr and check_utf8, if non-NULL, point to either their
9501 anchored or float namesakes, and don't hold a second reference. */
9503 RX_MATCH_COPIED_off(ret_x);
9504 #ifdef PERL_OLD_COPY_ON_WRITE
9505 ret->saved_copy = NULL;
9507 ret->mother_re = rx;
9513 /* regfree_internal()
9515 Free the private data in a regexp. This is overloadable by
9516 extensions. Perl takes care of the regexp structure in pregfree(),
9517 this covers the *pprivate pointer which technically perldoesnt
9518 know about, however of course we have to handle the
9519 regexp_internal structure when no extension is in use.
9521 Note this is called before freeing anything in the regexp
9526 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9529 struct regexp *const r = (struct regexp *)SvANY(rx);
9531 GET_RE_DEBUG_FLAGS_DECL;
9533 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9539 SV *dsv= sv_newmortal();
9540 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9541 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9542 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9543 PL_colors[4],PL_colors[5],s);
9546 #ifdef RE_TRACK_PATTERN_OFFSETS
9548 Safefree(ri->u.offsets); /* 20010421 MJD */
9551 int n = ri->data->count;
9552 PAD* new_comppad = NULL;
9557 /* If you add a ->what type here, update the comment in regcomp.h */
9558 switch (ri->data->what[n]) {
9563 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
9566 Safefree(ri->data->data[n]);
9569 new_comppad = MUTABLE_AV(ri->data->data[n]);
9572 if (new_comppad == NULL)
9573 Perl_croak(aTHX_ "panic: pregfree comppad");
9574 PAD_SAVE_LOCAL(old_comppad,
9575 /* Watch out for global destruction's random ordering. */
9576 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9579 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9582 op_free((OP_4tree*)ri->data->data[n]);
9584 PAD_RESTORE_LOCAL(old_comppad);
9585 SvREFCNT_dec(MUTABLE_SV(new_comppad));
9591 { /* Aho Corasick add-on structure for a trie node.
9592 Used in stclass optimization only */
9594 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9596 refcount = --aho->refcount;
9599 PerlMemShared_free(aho->states);
9600 PerlMemShared_free(aho->fail);
9601 /* do this last!!!! */
9602 PerlMemShared_free(ri->data->data[n]);
9603 PerlMemShared_free(ri->regstclass);
9609 /* trie structure. */
9611 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9613 refcount = --trie->refcount;
9616 PerlMemShared_free(trie->charmap);
9617 PerlMemShared_free(trie->states);
9618 PerlMemShared_free(trie->trans);
9620 PerlMemShared_free(trie->bitmap);
9622 PerlMemShared_free(trie->jump);
9623 PerlMemShared_free(trie->wordinfo);
9624 /* do this last!!!! */
9625 PerlMemShared_free(ri->data->data[n]);
9630 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9633 Safefree(ri->data->what);
9640 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
9641 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
9642 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9645 re_dup - duplicate a regexp.
9647 This routine is expected to clone a given regexp structure. It is only
9648 compiled under USE_ITHREADS.
9650 After all of the core data stored in struct regexp is duplicated
9651 the regexp_engine.dupe method is used to copy any private data
9652 stored in the *pprivate pointer. This allows extensions to handle
9653 any duplication it needs to do.
9655 See pregfree() and regfree_internal() if you change anything here.
9657 #if defined(USE_ITHREADS)
9658 #ifndef PERL_IN_XSUB_RE
9660 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9664 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9665 struct regexp *ret = (struct regexp *)SvANY(dstr);
9667 PERL_ARGS_ASSERT_RE_DUP_GUTS;
9669 npar = r->nparens+1;
9670 Newx(ret->offs, npar, regexp_paren_pair);
9671 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9673 /* no need to copy these */
9674 Newx(ret->swap, npar, regexp_paren_pair);
9678 /* Do it this way to avoid reading from *r after the StructCopy().
9679 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9680 cache, it doesn't matter. */
9681 const bool anchored = r->check_substr
9682 ? r->check_substr == r->anchored_substr
9683 : r->check_utf8 == r->anchored_utf8;
9684 Newx(ret->substrs, 1, struct reg_substr_data);
9685 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9687 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9688 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9689 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9690 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9692 /* check_substr and check_utf8, if non-NULL, point to either their
9693 anchored or float namesakes, and don't hold a second reference. */
9695 if (ret->check_substr) {
9697 assert(r->check_utf8 == r->anchored_utf8);
9698 ret->check_substr = ret->anchored_substr;
9699 ret->check_utf8 = ret->anchored_utf8;
9701 assert(r->check_substr == r->float_substr);
9702 assert(r->check_utf8 == r->float_utf8);
9703 ret->check_substr = ret->float_substr;
9704 ret->check_utf8 = ret->float_utf8;
9706 } else if (ret->check_utf8) {
9708 ret->check_utf8 = ret->anchored_utf8;
9710 ret->check_utf8 = ret->float_utf8;
9715 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9718 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9720 if (RX_MATCH_COPIED(dstr))
9721 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9724 #ifdef PERL_OLD_COPY_ON_WRITE
9725 ret->saved_copy = NULL;
9728 if (ret->mother_re) {
9729 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
9730 /* Our storage points directly to our mother regexp, but that's
9731 1: a buffer in a different thread
9732 2: something we no longer hold a reference on
9733 so we need to copy it locally. */
9734 /* Note we need to sue SvCUR() on our mother_re, because it, in
9735 turn, may well be pointing to its own mother_re. */
9736 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
9737 SvCUR(ret->mother_re)+1));
9738 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
9740 ret->mother_re = NULL;
9744 #endif /* PERL_IN_XSUB_RE */
9749 This is the internal complement to regdupe() which is used to copy
9750 the structure pointed to by the *pprivate pointer in the regexp.
9751 This is the core version of the extension overridable cloning hook.
9752 The regexp structure being duplicated will be copied by perl prior
9753 to this and will be provided as the regexp *r argument, however
9754 with the /old/ structures pprivate pointer value. Thus this routine
9755 may override any copying normally done by perl.
9757 It returns a pointer to the new regexp_internal structure.
9761 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
9764 struct regexp *const r = (struct regexp *)SvANY(rx);
9765 regexp_internal *reti;
9769 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
9771 npar = r->nparens+1;
9774 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
9775 Copy(ri->program, reti->program, len+1, regnode);
9778 reti->regstclass = NULL;
9782 const int count = ri->data->count;
9785 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9786 char, struct reg_data);
9787 Newx(d->what, count, U8);
9790 for (i = 0; i < count; i++) {
9791 d->what[i] = ri->data->what[i];
9792 switch (d->what[i]) {
9793 /* legal options are one of: sSfpontTua
9794 see also regcomp.h and pregfree() */
9795 case 'a': /* actually an AV, but the dup function is identical. */
9798 case 'p': /* actually an AV, but the dup function is identical. */
9799 case 'u': /* actually an HV, but the dup function is identical. */
9800 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
9803 /* This is cheating. */
9804 Newx(d->data[i], 1, struct regnode_charclass_class);
9805 StructCopy(ri->data->data[i], d->data[i],
9806 struct regnode_charclass_class);
9807 reti->regstclass = (regnode*)d->data[i];
9810 /* Compiled op trees are readonly and in shared memory,
9811 and can thus be shared without duplication. */
9813 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9817 /* Trie stclasses are readonly and can thus be shared
9818 * without duplication. We free the stclass in pregfree
9819 * when the corresponding reg_ac_data struct is freed.
9821 reti->regstclass= ri->regstclass;
9825 ((reg_trie_data*)ri->data->data[i])->refcount++;
9829 d->data[i] = ri->data->data[i];
9832 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9841 reti->name_list_idx = ri->name_list_idx;
9843 #ifdef RE_TRACK_PATTERN_OFFSETS
9844 if (ri->u.offsets) {
9845 Newx(reti->u.offsets, 2*len+1, U32);
9846 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9849 SetProgLen(reti,len);
9855 #endif /* USE_ITHREADS */
9857 #ifndef PERL_IN_XSUB_RE
9860 - regnext - dig the "next" pointer out of a node
9863 Perl_regnext(pTHX_ register regnode *p)
9866 register I32 offset;
9871 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
9872 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
9875 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9884 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9887 STRLEN l1 = strlen(pat1);
9888 STRLEN l2 = strlen(pat2);
9891 const char *message;
9893 PERL_ARGS_ASSERT_RE_CROAK2;
9899 Copy(pat1, buf, l1 , char);
9900 Copy(pat2, buf + l1, l2 , char);
9901 buf[l1 + l2] = '\n';
9902 buf[l1 + l2 + 1] = '\0';
9904 /* ANSI variant takes additional second argument */
9905 va_start(args, pat2);
9909 msv = vmess(buf, &args);
9911 message = SvPV_const(msv,l1);
9914 Copy(message, buf, l1 , char);
9915 buf[l1-1] = '\0'; /* Overwrite \n */
9916 Perl_croak(aTHX_ "%s", buf);
9919 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9921 #ifndef PERL_IN_XSUB_RE
9923 Perl_save_re_context(pTHX)
9927 struct re_save_state *state;
9929 SAVEVPTR(PL_curcop);
9930 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9932 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9933 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9934 SSPUSHUV(SAVEt_RE_STATE);
9936 Copy(&PL_reg_state, state, 1, struct re_save_state);
9938 PL_reg_start_tmp = 0;
9939 PL_reg_start_tmpl = 0;
9940 PL_reg_oldsaved = NULL;
9941 PL_reg_oldsavedlen = 0;
9943 PL_reg_leftiter = 0;
9944 PL_reg_poscache = NULL;
9945 PL_reg_poscache_size = 0;
9946 #ifdef PERL_OLD_COPY_ON_WRITE
9950 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9952 const REGEXP * const rx = PM_GETRE(PL_curpm);
9955 for (i = 1; i <= RX_NPARENS(rx); i++) {
9956 char digits[TYPE_CHARS(long)];
9957 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9958 GV *const *const gvp
9959 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9962 GV * const gv = *gvp;
9963 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
9973 clear_re(pTHX_ void *r)
9976 ReREFCNT_dec((REGEXP *)r);
9982 S_put_byte(pTHX_ SV *sv, int c)
9984 PERL_ARGS_ASSERT_PUT_BYTE;
9986 /* Our definition of isPRINT() ignores locales, so only bytes that are
9987 not part of UTF-8 are considered printable. I assume that the same
9988 holds for UTF-EBCDIC.
9989 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
9990 which Wikipedia says:
9992 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
9993 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
9994 identical, to the ASCII delete (DEL) or rubout control character.
9995 ) So the old condition can be simplified to !isPRINT(c) */
9997 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
9999 const char string = c;
10000 if (c == '-' || c == ']' || c == '\\' || c == '^')
10001 sv_catpvs(sv, "\\");
10002 sv_catpvn(sv, &string, 1);
10007 #define CLEAR_OPTSTART \
10008 if (optstart) STMT_START { \
10009 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
10013 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
10015 STATIC const regnode *
10016 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
10017 const regnode *last, const regnode *plast,
10018 SV* sv, I32 indent, U32 depth)
10021 register U8 op = PSEUDO; /* Arbitrary non-END op. */
10022 register const regnode *next;
10023 const regnode *optstart= NULL;
10025 RXi_GET_DECL(r,ri);
10026 GET_RE_DEBUG_FLAGS_DECL;
10028 PERL_ARGS_ASSERT_DUMPUNTIL;
10030 #ifdef DEBUG_DUMPUNTIL
10031 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
10032 last ? last-start : 0,plast ? plast-start : 0);
10035 if (plast && plast < last)
10038 while (PL_regkind[op] != END && (!last || node < last)) {
10039 /* While that wasn't END last time... */
10042 if (op == CLOSE || op == WHILEM)
10044 next = regnext((regnode *)node);
10047 if (OP(node) == OPTIMIZED) {
10048 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10055 regprop(r, sv, node);
10056 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10057 (int)(2*indent + 1), "", SvPVX_const(sv));
10059 if (OP(node) != OPTIMIZED) {
10060 if (next == NULL) /* Next ptr. */
10061 PerlIO_printf(Perl_debug_log, " (0)");
10062 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10063 PerlIO_printf(Perl_debug_log, " (FAIL)");
10065 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10066 (void)PerlIO_putc(Perl_debug_log, '\n');
10070 if (PL_regkind[(U8)op] == BRANCHJ) {
10073 register const regnode *nnode = (OP(next) == LONGJMP
10074 ? regnext((regnode *)next)
10076 if (last && nnode > last)
10078 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10081 else if (PL_regkind[(U8)op] == BRANCH) {
10083 DUMPUNTIL(NEXTOPER(node), next);
10085 else if ( PL_regkind[(U8)op] == TRIE ) {
10086 const regnode *this_trie = node;
10087 const char op = OP(node);
10088 const U32 n = ARG(node);
10089 const reg_ac_data * const ac = op>=AHOCORASICK ?
10090 (reg_ac_data *)ri->data->data[n] :
10092 const reg_trie_data * const trie =
10093 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10095 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10097 const regnode *nextbranch= NULL;
10100 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10101 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10103 PerlIO_printf(Perl_debug_log, "%*s%s ",
10104 (int)(2*(indent+3)), "",
10105 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10106 PL_colors[0], PL_colors[1],
10107 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10108 PERL_PV_PRETTY_ELLIPSES |
10109 PERL_PV_PRETTY_LTGT
10114 U16 dist= trie->jump[word_idx+1];
10115 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10116 (UV)((dist ? this_trie + dist : next) - start));
10119 nextbranch= this_trie + trie->jump[0];
10120 DUMPUNTIL(this_trie + dist, nextbranch);
10122 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10123 nextbranch= regnext((regnode *)nextbranch);
10125 PerlIO_printf(Perl_debug_log, "\n");
10128 if (last && next > last)
10133 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10134 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10135 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10137 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10139 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10141 else if ( op == PLUS || op == STAR) {
10142 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10144 else if (op == ANYOF) {
10145 /* arglen 1 + class block */
10146 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
10147 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10148 node = NEXTOPER(node);
10150 else if (PL_regkind[(U8)op] == EXACT) {
10151 /* Literal string, where present. */
10152 node += NODE_SZ_STR(node) - 1;
10153 node = NEXTOPER(node);
10156 node = NEXTOPER(node);
10157 node += regarglen[(U8)op];
10159 if (op == CURLYX || op == OPEN)
10163 #ifdef DEBUG_DUMPUNTIL
10164 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10169 #endif /* DEBUGGING */
10173 * c-indentation-style: bsd
10174 * c-basic-offset: 4
10175 * indent-tabs-mode: t
10178 * ex: set ts=8 sts=4 sw=4 noet: