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) && OP(nxt) == CLOSE) {
3320 /* Set the parenth number. */
3321 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3323 oscan->flags = (U8)ARG(nxt);
3324 if (RExC_open_parens) {
3325 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3326 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3328 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3329 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3332 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3333 OP(nxt + 1) = OPTIMIZED; /* was count. */
3334 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3335 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3338 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3339 regnode *nnxt = regnext(nxt1);
3342 if (reg_off_by_arg[OP(nxt1)])
3343 ARG_SET(nxt1, nxt2 - nxt1);
3344 else if (nxt2 - nxt1 < U16_MAX)
3345 NEXT_OFF(nxt1) = nxt2 - nxt1;
3347 OP(nxt) = NOTHING; /* Cannot beautify */
3352 /* Optimize again: */
3353 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3354 NULL, stopparen, recursed, NULL, 0,depth+1);
3359 else if ((OP(oscan) == CURLYX)
3360 && (flags & SCF_WHILEM_VISITED_POS)
3361 /* See the comment on a similar expression above.
3362 However, this time it not a subexpression
3363 we care about, but the expression itself. */
3364 && (maxcount == REG_INFTY)
3365 && data && ++data->whilem_c < 16) {
3366 /* This stays as CURLYX, we can put the count/of pair. */
3367 /* Find WHILEM (as in regexec.c) */
3368 regnode *nxt = oscan + NEXT_OFF(oscan);
3370 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3372 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3373 | (RExC_whilem_seen << 4)); /* On WHILEM */
3375 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3377 if (flags & SCF_DO_SUBSTR) {
3378 SV *last_str = NULL;
3379 int counted = mincount != 0;
3381 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3382 #if defined(SPARC64_GCC_WORKAROUND)
3385 const char *s = NULL;
3388 if (pos_before >= data->last_start_min)
3391 b = data->last_start_min;
3394 s = SvPV_const(data->last_found, l);
3395 old = b - data->last_start_min;
3398 I32 b = pos_before >= data->last_start_min
3399 ? pos_before : data->last_start_min;
3401 const char * const s = SvPV_const(data->last_found, l);
3402 I32 old = b - data->last_start_min;
3406 old = utf8_hop((U8*)s, old) - (U8*)s;
3409 /* Get the added string: */
3410 last_str = newSVpvn_utf8(s + old, l, UTF);
3411 if (deltanext == 0 && pos_before == b) {
3412 /* What was added is a constant string */
3414 SvGROW(last_str, (mincount * l) + 1);
3415 repeatcpy(SvPVX(last_str) + l,
3416 SvPVX_const(last_str), l, mincount - 1);
3417 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3418 /* Add additional parts. */
3419 SvCUR_set(data->last_found,
3420 SvCUR(data->last_found) - l);
3421 sv_catsv(data->last_found, last_str);
3423 SV * sv = data->last_found;
3425 SvUTF8(sv) && SvMAGICAL(sv) ?
3426 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3427 if (mg && mg->mg_len >= 0)
3428 mg->mg_len += CHR_SVLEN(last_str) - l;
3430 data->last_end += l * (mincount - 1);
3433 /* start offset must point into the last copy */
3434 data->last_start_min += minnext * (mincount - 1);
3435 data->last_start_max += is_inf ? I32_MAX
3436 : (maxcount - 1) * (minnext + data->pos_delta);
3439 /* It is counted once already... */
3440 data->pos_min += minnext * (mincount - counted);
3441 data->pos_delta += - counted * deltanext +
3442 (minnext + deltanext) * maxcount - minnext * mincount;
3443 if (mincount != maxcount) {
3444 /* Cannot extend fixed substrings found inside
3446 SCAN_COMMIT(pRExC_state,data,minlenp);
3447 if (mincount && last_str) {
3448 SV * const sv = data->last_found;
3449 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3450 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3454 sv_setsv(sv, last_str);
3455 data->last_end = data->pos_min;
3456 data->last_start_min =
3457 data->pos_min - CHR_SVLEN(last_str);
3458 data->last_start_max = is_inf
3460 : data->pos_min + data->pos_delta
3461 - CHR_SVLEN(last_str);
3463 data->longest = &(data->longest_float);
3465 SvREFCNT_dec(last_str);
3467 if (data && (fl & SF_HAS_EVAL))
3468 data->flags |= SF_HAS_EVAL;
3469 optimize_curly_tail:
3470 if (OP(oscan) != CURLYX) {
3471 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3473 NEXT_OFF(oscan) += NEXT_OFF(next);
3476 default: /* REF and CLUMP only? */
3477 if (flags & SCF_DO_SUBSTR) {
3478 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3479 data->longest = &(data->longest_float);
3481 is_inf = is_inf_internal = 1;
3482 if (flags & SCF_DO_STCLASS_OR)
3483 cl_anything(pRExC_state, data->start_class);
3484 flags &= ~SCF_DO_STCLASS;
3488 else if (OP(scan) == LNBREAK) {
3489 if (flags & SCF_DO_STCLASS) {
3491 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3492 if (flags & SCF_DO_STCLASS_AND) {
3493 for (value = 0; value < 256; value++)
3494 if (!is_VERTWS_cp(value))
3495 ANYOF_BITMAP_CLEAR(data->start_class, value);
3498 for (value = 0; value < 256; value++)
3499 if (is_VERTWS_cp(value))
3500 ANYOF_BITMAP_SET(data->start_class, value);
3502 if (flags & SCF_DO_STCLASS_OR)
3503 cl_and(data->start_class, and_withp);
3504 flags &= ~SCF_DO_STCLASS;
3508 if (flags & SCF_DO_SUBSTR) {
3509 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3511 data->pos_delta += 1;
3512 data->longest = &(data->longest_float);
3516 else if (OP(scan) == FOLDCHAR) {
3517 int d = ARG(scan)==0xDF ? 1 : 2;
3518 flags &= ~SCF_DO_STCLASS;
3521 if (flags & SCF_DO_SUBSTR) {
3522 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3524 data->pos_delta += d;
3525 data->longest = &(data->longest_float);
3528 else if (REGNODE_SIMPLE(OP(scan))) {
3531 if (flags & SCF_DO_SUBSTR) {
3532 SCAN_COMMIT(pRExC_state,data,minlenp);
3536 if (flags & SCF_DO_STCLASS) {
3537 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3539 /* Some of the logic below assumes that switching
3540 locale on will only add false positives. */
3541 switch (PL_regkind[OP(scan)]) {
3545 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3546 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3547 cl_anything(pRExC_state, data->start_class);
3550 if (OP(scan) == SANY)
3552 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3553 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3554 || (data->start_class->flags & ANYOF_CLASS));
3555 cl_anything(pRExC_state, data->start_class);
3557 if (flags & SCF_DO_STCLASS_AND || !value)
3558 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3561 if (flags & SCF_DO_STCLASS_AND)
3562 cl_and(data->start_class,
3563 (struct regnode_charclass_class*)scan);
3565 cl_or(pRExC_state, data->start_class,
3566 (struct regnode_charclass_class*)scan);
3569 if (flags & SCF_DO_STCLASS_AND) {
3570 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3571 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3572 for (value = 0; value < 256; value++)
3573 if (!isALNUM(value))
3574 ANYOF_BITMAP_CLEAR(data->start_class, value);
3578 if (data->start_class->flags & ANYOF_LOCALE)
3579 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3581 for (value = 0; value < 256; value++)
3583 ANYOF_BITMAP_SET(data->start_class, value);
3588 if (flags & SCF_DO_STCLASS_AND) {
3589 if (data->start_class->flags & ANYOF_LOCALE)
3590 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3593 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3594 data->start_class->flags |= ANYOF_LOCALE;
3598 if (flags & SCF_DO_STCLASS_AND) {
3599 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3600 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3601 for (value = 0; value < 256; value++)
3603 ANYOF_BITMAP_CLEAR(data->start_class, value);
3607 if (data->start_class->flags & ANYOF_LOCALE)
3608 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3610 for (value = 0; value < 256; value++)
3611 if (!isALNUM(value))
3612 ANYOF_BITMAP_SET(data->start_class, value);
3617 if (flags & SCF_DO_STCLASS_AND) {
3618 if (data->start_class->flags & ANYOF_LOCALE)
3619 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3622 data->start_class->flags |= ANYOF_LOCALE;
3623 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3627 if (flags & SCF_DO_STCLASS_AND) {
3628 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3629 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3630 for (value = 0; value < 256; value++)
3631 if (!isSPACE(value))
3632 ANYOF_BITMAP_CLEAR(data->start_class, value);
3636 if (data->start_class->flags & ANYOF_LOCALE)
3637 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3639 for (value = 0; value < 256; value++)
3641 ANYOF_BITMAP_SET(data->start_class, value);
3646 if (flags & SCF_DO_STCLASS_AND) {
3647 if (data->start_class->flags & ANYOF_LOCALE)
3648 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3651 data->start_class->flags |= ANYOF_LOCALE;
3652 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3656 if (flags & SCF_DO_STCLASS_AND) {
3657 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3658 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3659 for (value = 0; value < 256; value++)
3661 ANYOF_BITMAP_CLEAR(data->start_class, value);
3665 if (data->start_class->flags & ANYOF_LOCALE)
3666 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3668 for (value = 0; value < 256; value++)
3669 if (!isSPACE(value))
3670 ANYOF_BITMAP_SET(data->start_class, value);
3675 if (flags & SCF_DO_STCLASS_AND) {
3676 if (data->start_class->flags & ANYOF_LOCALE) {
3677 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3678 for (value = 0; value < 256; value++)
3679 if (!isSPACE(value))
3680 ANYOF_BITMAP_CLEAR(data->start_class, value);
3684 data->start_class->flags |= ANYOF_LOCALE;
3685 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3689 if (flags & SCF_DO_STCLASS_AND) {
3690 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3691 for (value = 0; value < 256; value++)
3692 if (!isDIGIT(value))
3693 ANYOF_BITMAP_CLEAR(data->start_class, value);
3696 if (data->start_class->flags & ANYOF_LOCALE)
3697 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3699 for (value = 0; value < 256; value++)
3701 ANYOF_BITMAP_SET(data->start_class, value);
3706 if (flags & SCF_DO_STCLASS_AND) {
3707 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3708 for (value = 0; value < 256; value++)
3710 ANYOF_BITMAP_CLEAR(data->start_class, value);
3713 if (data->start_class->flags & ANYOF_LOCALE)
3714 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3716 for (value = 0; value < 256; value++)
3717 if (!isDIGIT(value))
3718 ANYOF_BITMAP_SET(data->start_class, value);
3722 CASE_SYNST_FNC(VERTWS);
3723 CASE_SYNST_FNC(HORIZWS);
3726 if (flags & SCF_DO_STCLASS_OR)
3727 cl_and(data->start_class, and_withp);
3728 flags &= ~SCF_DO_STCLASS;
3731 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3732 data->flags |= (OP(scan) == MEOL
3736 else if ( PL_regkind[OP(scan)] == BRANCHJ
3737 /* Lookbehind, or need to calculate parens/evals/stclass: */
3738 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3739 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3740 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3741 || OP(scan) == UNLESSM )
3743 /* Negative Lookahead/lookbehind
3744 In this case we can't do fixed string optimisation.
3747 I32 deltanext, minnext, fake = 0;
3749 struct regnode_charclass_class intrnl;
3752 data_fake.flags = 0;
3754 data_fake.whilem_c = data->whilem_c;
3755 data_fake.last_closep = data->last_closep;
3758 data_fake.last_closep = &fake;
3759 data_fake.pos_delta = delta;
3760 if ( flags & SCF_DO_STCLASS && !scan->flags
3761 && OP(scan) == IFMATCH ) { /* Lookahead */
3762 cl_init(pRExC_state, &intrnl);
3763 data_fake.start_class = &intrnl;
3764 f |= SCF_DO_STCLASS_AND;
3766 if (flags & SCF_WHILEM_VISITED_POS)
3767 f |= SCF_WHILEM_VISITED_POS;
3768 next = regnext(scan);
3769 nscan = NEXTOPER(NEXTOPER(scan));
3770 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3771 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3774 FAIL("Variable length lookbehind not implemented");
3776 else if (minnext > (I32)U8_MAX) {
3777 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3779 scan->flags = (U8)minnext;
3782 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3784 if (data_fake.flags & SF_HAS_EVAL)
3785 data->flags |= SF_HAS_EVAL;
3786 data->whilem_c = data_fake.whilem_c;
3788 if (f & SCF_DO_STCLASS_AND) {
3789 if (flags & SCF_DO_STCLASS_OR) {
3790 /* OR before, AND after: ideally we would recurse with
3791 * data_fake to get the AND applied by study of the
3792 * remainder of the pattern, and then derecurse;
3793 * *** HACK *** for now just treat as "no information".
3794 * See [perl #56690].
3796 cl_init(pRExC_state, data->start_class);
3798 /* AND before and after: combine and continue */
3799 const int was = (data->start_class->flags & ANYOF_EOS);
3801 cl_and(data->start_class, &intrnl);
3803 data->start_class->flags |= ANYOF_EOS;
3807 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3809 /* Positive Lookahead/lookbehind
3810 In this case we can do fixed string optimisation,
3811 but we must be careful about it. Note in the case of
3812 lookbehind the positions will be offset by the minimum
3813 length of the pattern, something we won't know about
3814 until after the recurse.
3816 I32 deltanext, fake = 0;
3818 struct regnode_charclass_class intrnl;
3820 /* We use SAVEFREEPV so that when the full compile
3821 is finished perl will clean up the allocated
3822 minlens when its all done. This was we don't
3823 have to worry about freeing them when we know
3824 they wont be used, which would be a pain.
3827 Newx( minnextp, 1, I32 );
3828 SAVEFREEPV(minnextp);
3831 StructCopy(data, &data_fake, scan_data_t);
3832 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3835 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3836 data_fake.last_found=newSVsv(data->last_found);
3840 data_fake.last_closep = &fake;
3841 data_fake.flags = 0;
3842 data_fake.pos_delta = delta;
3844 data_fake.flags |= SF_IS_INF;
3845 if ( flags & SCF_DO_STCLASS && !scan->flags
3846 && OP(scan) == IFMATCH ) { /* Lookahead */
3847 cl_init(pRExC_state, &intrnl);
3848 data_fake.start_class = &intrnl;
3849 f |= SCF_DO_STCLASS_AND;
3851 if (flags & SCF_WHILEM_VISITED_POS)
3852 f |= SCF_WHILEM_VISITED_POS;
3853 next = regnext(scan);
3854 nscan = NEXTOPER(NEXTOPER(scan));
3856 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3857 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3860 FAIL("Variable length lookbehind not implemented");
3862 else if (*minnextp > (I32)U8_MAX) {
3863 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3865 scan->flags = (U8)*minnextp;
3870 if (f & SCF_DO_STCLASS_AND) {
3871 const int was = (data->start_class->flags & ANYOF_EOS);
3873 cl_and(data->start_class, &intrnl);
3875 data->start_class->flags |= ANYOF_EOS;
3878 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3880 if (data_fake.flags & SF_HAS_EVAL)
3881 data->flags |= SF_HAS_EVAL;
3882 data->whilem_c = data_fake.whilem_c;
3883 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3884 if (RExC_rx->minlen<*minnextp)
3885 RExC_rx->minlen=*minnextp;
3886 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
3887 SvREFCNT_dec(data_fake.last_found);
3889 if ( data_fake.minlen_fixed != minlenp )
3891 data->offset_fixed= data_fake.offset_fixed;
3892 data->minlen_fixed= data_fake.minlen_fixed;
3893 data->lookbehind_fixed+= scan->flags;
3895 if ( data_fake.minlen_float != minlenp )
3897 data->minlen_float= data_fake.minlen_float;
3898 data->offset_float_min=data_fake.offset_float_min;
3899 data->offset_float_max=data_fake.offset_float_max;
3900 data->lookbehind_float+= scan->flags;
3909 else if (OP(scan) == OPEN) {
3910 if (stopparen != (I32)ARG(scan))
3913 else if (OP(scan) == CLOSE) {
3914 if (stopparen == (I32)ARG(scan)) {
3917 if ((I32)ARG(scan) == is_par) {
3918 next = regnext(scan);
3920 if ( next && (OP(next) != WHILEM) && next < last)
3921 is_par = 0; /* Disable optimization */
3924 *(data->last_closep) = ARG(scan);
3926 else if (OP(scan) == EVAL) {
3928 data->flags |= SF_HAS_EVAL;
3930 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3931 if (flags & SCF_DO_SUBSTR) {
3932 SCAN_COMMIT(pRExC_state,data,minlenp);
3933 flags &= ~SCF_DO_SUBSTR;
3935 if (data && OP(scan)==ACCEPT) {
3936 data->flags |= SCF_SEEN_ACCEPT;
3941 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3943 if (flags & SCF_DO_SUBSTR) {
3944 SCAN_COMMIT(pRExC_state,data,minlenp);
3945 data->longest = &(data->longest_float);
3947 is_inf = is_inf_internal = 1;
3948 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3949 cl_anything(pRExC_state, data->start_class);
3950 flags &= ~SCF_DO_STCLASS;
3952 else if (OP(scan) == GPOS) {
3953 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
3954 !(delta || is_inf || (data && data->pos_delta)))
3956 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
3957 RExC_rx->extflags |= RXf_ANCH_GPOS;
3958 if (RExC_rx->gofs < (U32)min)
3959 RExC_rx->gofs = min;
3961 RExC_rx->extflags |= RXf_GPOS_FLOAT;
3965 #ifdef TRIE_STUDY_OPT
3966 #ifdef FULL_TRIE_STUDY
3967 else if (PL_regkind[OP(scan)] == TRIE) {
3968 /* NOTE - There is similar code to this block above for handling
3969 BRANCH nodes on the initial study. If you change stuff here
3971 regnode *trie_node= scan;
3972 regnode *tail= regnext(scan);
3973 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
3974 I32 max1 = 0, min1 = I32_MAX;
3975 struct regnode_charclass_class accum;
3977 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
3978 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
3979 if (flags & SCF_DO_STCLASS)
3980 cl_init_zero(pRExC_state, &accum);
3986 const regnode *nextbranch= NULL;
3989 for ( word=1 ; word <= trie->wordcount ; word++)
3991 I32 deltanext=0, minnext=0, f = 0, fake;
3992 struct regnode_charclass_class this_class;
3994 data_fake.flags = 0;
3996 data_fake.whilem_c = data->whilem_c;
3997 data_fake.last_closep = data->last_closep;
4000 data_fake.last_closep = &fake;
4001 data_fake.pos_delta = delta;
4002 if (flags & SCF_DO_STCLASS) {
4003 cl_init(pRExC_state, &this_class);
4004 data_fake.start_class = &this_class;
4005 f = SCF_DO_STCLASS_AND;
4007 if (flags & SCF_WHILEM_VISITED_POS)
4008 f |= SCF_WHILEM_VISITED_POS;
4010 if (trie->jump[word]) {
4012 nextbranch = trie_node + trie->jump[0];
4013 scan= trie_node + trie->jump[word];
4014 /* We go from the jump point to the branch that follows
4015 it. Note this means we need the vestigal unused branches
4016 even though they arent otherwise used.
4018 minnext = study_chunk(pRExC_state, &scan, minlenp,
4019 &deltanext, (regnode *)nextbranch, &data_fake,
4020 stopparen, recursed, NULL, f,depth+1);
4022 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4023 nextbranch= regnext((regnode*)nextbranch);
4025 if (min1 > (I32)(minnext + trie->minlen))
4026 min1 = minnext + trie->minlen;
4027 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4028 max1 = minnext + deltanext + trie->maxlen;
4029 if (deltanext == I32_MAX)
4030 is_inf = is_inf_internal = 1;
4032 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4034 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4035 if ( stopmin > min + min1)
4036 stopmin = min + min1;
4037 flags &= ~SCF_DO_SUBSTR;
4039 data->flags |= SCF_SEEN_ACCEPT;
4042 if (data_fake.flags & SF_HAS_EVAL)
4043 data->flags |= SF_HAS_EVAL;
4044 data->whilem_c = data_fake.whilem_c;
4046 if (flags & SCF_DO_STCLASS)
4047 cl_or(pRExC_state, &accum, &this_class);
4050 if (flags & SCF_DO_SUBSTR) {
4051 data->pos_min += min1;
4052 data->pos_delta += max1 - min1;
4053 if (max1 != min1 || is_inf)
4054 data->longest = &(data->longest_float);
4057 delta += max1 - min1;
4058 if (flags & SCF_DO_STCLASS_OR) {
4059 cl_or(pRExC_state, data->start_class, &accum);
4061 cl_and(data->start_class, and_withp);
4062 flags &= ~SCF_DO_STCLASS;
4065 else if (flags & SCF_DO_STCLASS_AND) {
4067 cl_and(data->start_class, &accum);
4068 flags &= ~SCF_DO_STCLASS;
4071 /* Switch to OR mode: cache the old value of
4072 * data->start_class */
4074 StructCopy(data->start_class, and_withp,
4075 struct regnode_charclass_class);
4076 flags &= ~SCF_DO_STCLASS_AND;
4077 StructCopy(&accum, data->start_class,
4078 struct regnode_charclass_class);
4079 flags |= SCF_DO_STCLASS_OR;
4080 data->start_class->flags |= ANYOF_EOS;
4087 else if (PL_regkind[OP(scan)] == TRIE) {
4088 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4091 min += trie->minlen;
4092 delta += (trie->maxlen - trie->minlen);
4093 flags &= ~SCF_DO_STCLASS; /* xxx */
4094 if (flags & SCF_DO_SUBSTR) {
4095 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4096 data->pos_min += trie->minlen;
4097 data->pos_delta += (trie->maxlen - trie->minlen);
4098 if (trie->maxlen != trie->minlen)
4099 data->longest = &(data->longest_float);
4101 if (trie->jump) /* no more substrings -- for now /grr*/
4102 flags &= ~SCF_DO_SUBSTR;
4104 #endif /* old or new */
4105 #endif /* TRIE_STUDY_OPT */
4107 /* Else: zero-length, ignore. */
4108 scan = regnext(scan);
4113 stopparen = frame->stop;
4114 frame = frame->prev;
4115 goto fake_study_recurse;
4120 DEBUG_STUDYDATA("pre-fin:",data,depth);
4123 *deltap = is_inf_internal ? I32_MAX : delta;
4124 if (flags & SCF_DO_SUBSTR && is_inf)
4125 data->pos_delta = I32_MAX - data->pos_min;
4126 if (is_par > (I32)U8_MAX)
4128 if (is_par && pars==1 && data) {
4129 data->flags |= SF_IN_PAR;
4130 data->flags &= ~SF_HAS_PAR;
4132 else if (pars && data) {
4133 data->flags |= SF_HAS_PAR;
4134 data->flags &= ~SF_IN_PAR;
4136 if (flags & SCF_DO_STCLASS_OR)
4137 cl_and(data->start_class, and_withp);
4138 if (flags & SCF_TRIE_RESTUDY)
4139 data->flags |= SCF_TRIE_RESTUDY;
4141 DEBUG_STUDYDATA("post-fin:",data,depth);
4143 return min < stopmin ? min : stopmin;
4147 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4149 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4151 PERL_ARGS_ASSERT_ADD_DATA;
4153 Renewc(RExC_rxi->data,
4154 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4155 char, struct reg_data);
4157 Renew(RExC_rxi->data->what, count + n, U8);
4159 Newx(RExC_rxi->data->what, n, U8);
4160 RExC_rxi->data->count = count + n;
4161 Copy(s, RExC_rxi->data->what + count, n, U8);
4165 /*XXX: todo make this not included in a non debugging perl */
4166 #ifndef PERL_IN_XSUB_RE
4168 Perl_reginitcolors(pTHX)
4171 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4173 char *t = savepv(s);
4177 t = strchr(t, '\t');
4183 PL_colors[i] = t = (char *)"";
4188 PL_colors[i++] = (char *)"";
4195 #ifdef TRIE_STUDY_OPT
4196 #define CHECK_RESTUDY_GOTO \
4198 (data.flags & SCF_TRIE_RESTUDY) \
4202 #define CHECK_RESTUDY_GOTO
4206 - pregcomp - compile a regular expression into internal code
4208 * We can't allocate space until we know how big the compiled form will be,
4209 * but we can't compile it (and thus know how big it is) until we've got a
4210 * place to put the code. So we cheat: we compile it twice, once with code
4211 * generation turned off and size counting turned on, and once "for real".
4212 * This also means that we don't allocate space until we are sure that the
4213 * thing really will compile successfully, and we never have to move the
4214 * code and thus invalidate pointers into it. (Note that it has to be in
4215 * one piece because free() must be able to free it all.) [NB: not true in perl]
4217 * Beware that the optimization-preparation code in here knows about some
4218 * of the structure of the compiled regexp. [I'll say.]
4223 #ifndef PERL_IN_XSUB_RE
4224 #define RE_ENGINE_PTR &PL_core_reg_engine
4226 extern const struct regexp_engine my_reg_engine;
4227 #define RE_ENGINE_PTR &my_reg_engine
4230 #ifndef PERL_IN_XSUB_RE
4232 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4235 HV * const table = GvHV(PL_hintgv);
4237 PERL_ARGS_ASSERT_PREGCOMP;
4239 /* Dispatch a request to compile a regexp to correct
4242 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4243 GET_RE_DEBUG_FLAGS_DECL;
4244 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4245 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4247 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4250 return CALLREGCOMP_ENG(eng, pattern, flags);
4253 return Perl_re_compile(aTHX_ pattern, flags);
4258 Perl_re_compile(pTHX_ SV * const pattern, U32 pm_flags)
4263 register regexp_internal *ri;
4265 char *exp = SvPV(pattern, plen);
4266 char* xend = exp + plen;
4273 RExC_state_t RExC_state;
4274 RExC_state_t * const pRExC_state = &RExC_state;
4275 #ifdef TRIE_STUDY_OPT
4277 RExC_state_t copyRExC_state;
4279 GET_RE_DEBUG_FLAGS_DECL;
4281 PERL_ARGS_ASSERT_RE_COMPILE;
4283 DEBUG_r(if (!PL_colorset) reginitcolors());
4285 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4288 SV *dsv= sv_newmortal();
4289 RE_PV_QUOTED_DECL(s, RExC_utf8,
4290 dsv, exp, plen, 60);
4291 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4292 PL_colors[4],PL_colors[5],s);
4297 RExC_flags = pm_flags;
4301 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4302 RExC_seen_evals = 0;
4305 /* First pass: determine size, legality. */
4313 RExC_emit = &PL_regdummy;
4314 RExC_whilem_seen = 0;
4315 RExC_open_parens = NULL;
4316 RExC_close_parens = NULL;
4318 RExC_paren_names = NULL;
4320 RExC_paren_name_list = NULL;
4322 RExC_recurse = NULL;
4323 RExC_recurse_count = 0;
4325 #if 0 /* REGC() is (currently) a NOP at the first pass.
4326 * Clever compilers notice this and complain. --jhi */
4327 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4329 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4330 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4331 RExC_precomp = NULL;
4334 if (RExC_utf8 && !RExC_orig_utf8) {
4335 /* It's possible to write a regexp in ascii that represents Unicode
4336 codepoints outside of the byte range, such as via \x{100}. If we
4337 detect such a sequence we have to convert the entire pattern to utf8
4338 and then recompile, as our sizing calculation will have been based
4339 on 1 byte == 1 character, but we will need to use utf8 to encode
4340 at least some part of the pattern, and therefore must convert the whole
4342 XXX: somehow figure out how to make this less expensive...
4345 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4346 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4347 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)exp, &len);
4349 RExC_orig_utf8 = RExC_utf8;
4351 goto redo_first_pass;
4354 PerlIO_printf(Perl_debug_log,
4355 "Required size %"IVdf" nodes\n"
4356 "Starting second pass (creation)\n",
4359 RExC_lastparse=NULL;
4361 /* Small enough for pointer-storage convention?
4362 If extralen==0, this means that we will not need long jumps. */
4363 if (RExC_size >= 0x10000L && RExC_extralen)
4364 RExC_size += RExC_extralen;
4367 if (RExC_whilem_seen > 15)
4368 RExC_whilem_seen = 15;
4370 /* Allocate space and zero-initialize. Note, the two step process
4371 of zeroing when in debug mode, thus anything assigned has to
4372 happen after that */
4373 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4374 r = (struct regexp*)SvANY(rx);
4375 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4376 char, regexp_internal);
4377 if ( r == NULL || ri == NULL )
4378 FAIL("Regexp out of space");
4380 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4381 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4383 /* bulk initialize base fields with 0. */
4384 Zero(ri, sizeof(regexp_internal), char);
4387 /* non-zero initialization begins here */
4389 r->engine= RE_ENGINE_PTR;
4390 r->extflags = pm_flags;
4392 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4393 bool has_minus = ((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD);
4394 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4395 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4396 >> RXf_PMf_STD_PMMOD_SHIFT);
4397 const char *fptr = STD_PAT_MODS; /*"msix"*/
4399 const STRLEN wraplen = plen + has_minus + has_p + has_runon
4400 + (sizeof(STD_PAT_MODS) - 1)
4401 + (sizeof("(?:)") - 1);
4403 p = sv_grow(MUTABLE_SV(rx), wraplen + 1);
4404 SvCUR_set(rx, wraplen);
4406 SvFLAGS(rx) |= SvUTF8(pattern);
4409 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4411 char *r = p + (sizeof(STD_PAT_MODS) - 1) + has_minus - 1;
4412 char *colon = r + 1;
4415 while((ch = *fptr++)) {
4429 Copy(RExC_precomp, p, plen, char);
4430 assert ((RX_WRAPPED(rx) - p) < 16);
4431 r->pre_prefix = p - RX_WRAPPED(rx);
4440 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4442 if (RExC_seen & REG_SEEN_RECURSE) {
4443 Newxz(RExC_open_parens, RExC_npar,regnode *);
4444 SAVEFREEPV(RExC_open_parens);
4445 Newxz(RExC_close_parens,RExC_npar,regnode *);
4446 SAVEFREEPV(RExC_close_parens);
4449 /* Useful during FAIL. */
4450 #ifdef RE_TRACK_PATTERN_OFFSETS
4451 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4452 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4453 "%s %"UVuf" bytes for offset annotations.\n",
4454 ri->u.offsets ? "Got" : "Couldn't get",
4455 (UV)((2*RExC_size+1) * sizeof(U32))));
4457 SetProgLen(ri,RExC_size);
4462 /* Second pass: emit code. */
4463 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4468 RExC_emit_start = ri->program;
4469 RExC_emit = ri->program;
4470 RExC_emit_bound = ri->program + RExC_size + 1;
4472 /* Store the count of eval-groups for security checks: */
4473 RExC_rx->seen_evals = RExC_seen_evals;
4474 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4475 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4479 /* XXXX To minimize changes to RE engine we always allocate
4480 3-units-long substrs field. */
4481 Newx(r->substrs, 1, struct reg_substr_data);
4482 if (RExC_recurse_count) {
4483 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4484 SAVEFREEPV(RExC_recurse);
4488 r->minlen = minlen = sawplus = sawopen = 0;
4489 Zero(r->substrs, 1, struct reg_substr_data);
4491 #ifdef TRIE_STUDY_OPT
4493 StructCopy(&zero_scan_data, &data, scan_data_t);
4494 copyRExC_state = RExC_state;
4497 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4499 RExC_state = copyRExC_state;
4500 if (seen & REG_TOP_LEVEL_BRANCHES)
4501 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4503 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4504 if (data.last_found) {
4505 SvREFCNT_dec(data.longest_fixed);
4506 SvREFCNT_dec(data.longest_float);
4507 SvREFCNT_dec(data.last_found);
4509 StructCopy(&zero_scan_data, &data, scan_data_t);
4512 StructCopy(&zero_scan_data, &data, scan_data_t);
4515 /* Dig out information for optimizations. */
4516 r->extflags = RExC_flags; /* was pm_op */
4517 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4520 SvUTF8_on(rx); /* Unicode in it? */
4521 ri->regstclass = NULL;
4522 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4523 r->intflags |= PREGf_NAUGHTY;
4524 scan = ri->program + 1; /* First BRANCH. */
4526 /* testing for BRANCH here tells us whether there is "must appear"
4527 data in the pattern. If there is then we can use it for optimisations */
4528 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4530 STRLEN longest_float_length, longest_fixed_length;
4531 struct regnode_charclass_class ch_class; /* pointed to by data */
4533 I32 last_close = 0; /* pointed to by data */
4534 regnode *first= scan;
4535 regnode *first_next= regnext(first);
4538 * Skip introductions and multiplicators >= 1
4539 * so that we can extract the 'meat' of the pattern that must
4540 * match in the large if() sequence following.
4541 * NOTE that EXACT is NOT covered here, as it is normally
4542 * picked up by the optimiser separately.
4544 * This is unfortunate as the optimiser isnt handling lookahead
4545 * properly currently.
4548 while ((OP(first) == OPEN && (sawopen = 1)) ||
4549 /* An OR of *one* alternative - should not happen now. */
4550 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4551 /* for now we can't handle lookbehind IFMATCH*/
4552 (OP(first) == IFMATCH && !first->flags) ||
4553 (OP(first) == PLUS) ||
4554 (OP(first) == MINMOD) ||
4555 /* An {n,m} with n>0 */
4556 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4557 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4560 * the only op that could be a regnode is PLUS, all the rest
4561 * will be regnode_1 or regnode_2.
4564 if (OP(first) == PLUS)
4567 first += regarglen[OP(first)];
4569 first = NEXTOPER(first);
4570 first_next= regnext(first);
4573 /* Starting-point info. */
4575 DEBUG_PEEP("first:",first,0);
4576 /* Ignore EXACT as we deal with it later. */
4577 if (PL_regkind[OP(first)] == EXACT) {
4578 if (OP(first) == EXACT)
4579 NOOP; /* Empty, get anchored substr later. */
4580 else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
4581 ri->regstclass = first;
4584 else if (PL_regkind[OP(first)] == TRIE &&
4585 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4588 /* this can happen only on restudy */
4589 if ( OP(first) == TRIE ) {
4590 struct regnode_1 *trieop = (struct regnode_1 *)
4591 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4592 StructCopy(first,trieop,struct regnode_1);
4593 trie_op=(regnode *)trieop;
4595 struct regnode_charclass *trieop = (struct regnode_charclass *)
4596 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4597 StructCopy(first,trieop,struct regnode_charclass);
4598 trie_op=(regnode *)trieop;
4601 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4602 ri->regstclass = trie_op;
4605 else if (REGNODE_SIMPLE(OP(first)))
4606 ri->regstclass = first;
4607 else if (PL_regkind[OP(first)] == BOUND ||
4608 PL_regkind[OP(first)] == NBOUND)
4609 ri->regstclass = first;
4610 else if (PL_regkind[OP(first)] == BOL) {
4611 r->extflags |= (OP(first) == MBOL
4613 : (OP(first) == SBOL
4616 first = NEXTOPER(first);
4619 else if (OP(first) == GPOS) {
4620 r->extflags |= RXf_ANCH_GPOS;
4621 first = NEXTOPER(first);
4624 else if ((!sawopen || !RExC_sawback) &&
4625 (OP(first) == STAR &&
4626 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4627 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4629 /* turn .* into ^.* with an implied $*=1 */
4631 (OP(NEXTOPER(first)) == REG_ANY)
4634 r->extflags |= type;
4635 r->intflags |= PREGf_IMPLICIT;
4636 first = NEXTOPER(first);
4639 if (sawplus && (!sawopen || !RExC_sawback)
4640 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4641 /* x+ must match at the 1st pos of run of x's */
4642 r->intflags |= PREGf_SKIP;
4644 /* Scan is after the zeroth branch, first is atomic matcher. */
4645 #ifdef TRIE_STUDY_OPT
4648 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4649 (IV)(first - scan + 1))
4653 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4654 (IV)(first - scan + 1))
4660 * If there's something expensive in the r.e., find the
4661 * longest literal string that must appear and make it the
4662 * regmust. Resolve ties in favor of later strings, since
4663 * the regstart check works with the beginning of the r.e.
4664 * and avoiding duplication strengthens checking. Not a
4665 * strong reason, but sufficient in the absence of others.
4666 * [Now we resolve ties in favor of the earlier string if
4667 * it happens that c_offset_min has been invalidated, since the
4668 * earlier string may buy us something the later one won't.]
4671 data.longest_fixed = newSVpvs("");
4672 data.longest_float = newSVpvs("");
4673 data.last_found = newSVpvs("");
4674 data.longest = &(data.longest_fixed);
4676 if (!ri->regstclass) {
4677 cl_init(pRExC_state, &ch_class);
4678 data.start_class = &ch_class;
4679 stclass_flag = SCF_DO_STCLASS_AND;
4680 } else /* XXXX Check for BOUND? */
4682 data.last_closep = &last_close;
4684 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4685 &data, -1, NULL, NULL,
4686 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4692 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4693 && data.last_start_min == 0 && data.last_end > 0
4694 && !RExC_seen_zerolen
4695 && !(RExC_seen & REG_SEEN_VERBARG)
4696 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4697 r->extflags |= RXf_CHECK_ALL;
4698 scan_commit(pRExC_state, &data,&minlen,0);
4699 SvREFCNT_dec(data.last_found);
4701 /* Note that code very similar to this but for anchored string
4702 follows immediately below, changes may need to be made to both.
4705 longest_float_length = CHR_SVLEN(data.longest_float);
4706 if (longest_float_length
4707 || (data.flags & SF_FL_BEFORE_EOL
4708 && (!(data.flags & SF_FL_BEFORE_MEOL)
4709 || (RExC_flags & RXf_PMf_MULTILINE))))
4713 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4714 && data.offset_fixed == data.offset_float_min
4715 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4716 goto remove_float; /* As in (a)+. */
4718 /* copy the information about the longest float from the reg_scan_data
4719 over to the program. */
4720 if (SvUTF8(data.longest_float)) {
4721 r->float_utf8 = data.longest_float;
4722 r->float_substr = NULL;
4724 r->float_substr = data.longest_float;
4725 r->float_utf8 = NULL;
4727 /* float_end_shift is how many chars that must be matched that
4728 follow this item. We calculate it ahead of time as once the
4729 lookbehind offset is added in we lose the ability to correctly
4731 ml = data.minlen_float ? *(data.minlen_float)
4732 : (I32)longest_float_length;
4733 r->float_end_shift = ml - data.offset_float_min
4734 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4735 + data.lookbehind_float;
4736 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4737 r->float_max_offset = data.offset_float_max;
4738 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4739 r->float_max_offset -= data.lookbehind_float;
4741 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4742 && (!(data.flags & SF_FL_BEFORE_MEOL)
4743 || (RExC_flags & RXf_PMf_MULTILINE)));
4744 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4748 r->float_substr = r->float_utf8 = NULL;
4749 SvREFCNT_dec(data.longest_float);
4750 longest_float_length = 0;
4753 /* Note that code very similar to this but for floating string
4754 is immediately above, changes may need to be made to both.
4757 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4758 if (longest_fixed_length
4759 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4760 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4761 || (RExC_flags & RXf_PMf_MULTILINE))))
4765 /* copy the information about the longest fixed
4766 from the reg_scan_data over to the program. */
4767 if (SvUTF8(data.longest_fixed)) {
4768 r->anchored_utf8 = data.longest_fixed;
4769 r->anchored_substr = NULL;
4771 r->anchored_substr = data.longest_fixed;
4772 r->anchored_utf8 = NULL;
4774 /* fixed_end_shift is how many chars that must be matched that
4775 follow this item. We calculate it ahead of time as once the
4776 lookbehind offset is added in we lose the ability to correctly
4778 ml = data.minlen_fixed ? *(data.minlen_fixed)
4779 : (I32)longest_fixed_length;
4780 r->anchored_end_shift = ml - data.offset_fixed
4781 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4782 + data.lookbehind_fixed;
4783 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4785 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4786 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4787 || (RExC_flags & RXf_PMf_MULTILINE)));
4788 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4791 r->anchored_substr = r->anchored_utf8 = NULL;
4792 SvREFCNT_dec(data.longest_fixed);
4793 longest_fixed_length = 0;
4796 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4797 ri->regstclass = NULL;
4798 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4800 && !(data.start_class->flags & ANYOF_EOS)
4801 && !cl_is_anything(data.start_class))
4803 const U32 n = add_data(pRExC_state, 1, "f");
4805 Newx(RExC_rxi->data->data[n], 1,
4806 struct regnode_charclass_class);
4807 StructCopy(data.start_class,
4808 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4809 struct regnode_charclass_class);
4810 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4811 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4812 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
4813 regprop(r, sv, (regnode*)data.start_class);
4814 PerlIO_printf(Perl_debug_log,
4815 "synthetic stclass \"%s\".\n",
4816 SvPVX_const(sv));});
4819 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
4820 if (longest_fixed_length > longest_float_length) {
4821 r->check_end_shift = r->anchored_end_shift;
4822 r->check_substr = r->anchored_substr;
4823 r->check_utf8 = r->anchored_utf8;
4824 r->check_offset_min = r->check_offset_max = r->anchored_offset;
4825 if (r->extflags & RXf_ANCH_SINGLE)
4826 r->extflags |= RXf_NOSCAN;
4829 r->check_end_shift = r->float_end_shift;
4830 r->check_substr = r->float_substr;
4831 r->check_utf8 = r->float_utf8;
4832 r->check_offset_min = r->float_min_offset;
4833 r->check_offset_max = r->float_max_offset;
4835 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
4836 This should be changed ASAP! */
4837 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
4838 r->extflags |= RXf_USE_INTUIT;
4839 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
4840 r->extflags |= RXf_INTUIT_TAIL;
4842 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
4843 if ( (STRLEN)minlen < longest_float_length )
4844 minlen= longest_float_length;
4845 if ( (STRLEN)minlen < longest_fixed_length )
4846 minlen= longest_fixed_length;
4850 /* Several toplevels. Best we can is to set minlen. */
4852 struct regnode_charclass_class ch_class;
4855 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
4857 scan = ri->program + 1;
4858 cl_init(pRExC_state, &ch_class);
4859 data.start_class = &ch_class;
4860 data.last_closep = &last_close;
4863 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
4864 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
4868 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
4869 = r->float_substr = r->float_utf8 = NULL;
4870 if (!(data.start_class->flags & ANYOF_EOS)
4871 && !cl_is_anything(data.start_class))
4873 const U32 n = add_data(pRExC_state, 1, "f");
4875 Newx(RExC_rxi->data->data[n], 1,
4876 struct regnode_charclass_class);
4877 StructCopy(data.start_class,
4878 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
4879 struct regnode_charclass_class);
4880 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
4881 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
4882 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
4883 regprop(r, sv, (regnode*)data.start_class);
4884 PerlIO_printf(Perl_debug_log,
4885 "synthetic stclass \"%s\".\n",
4886 SvPVX_const(sv));});
4890 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
4891 the "real" pattern. */
4893 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
4894 (IV)minlen, (IV)r->minlen);
4896 r->minlenret = minlen;
4897 if (r->minlen < minlen)
4900 if (RExC_seen & REG_SEEN_GPOS)
4901 r->extflags |= RXf_GPOS_SEEN;
4902 if (RExC_seen & REG_SEEN_LOOKBEHIND)
4903 r->extflags |= RXf_LOOKBEHIND_SEEN;
4904 if (RExC_seen & REG_SEEN_EVAL)
4905 r->extflags |= RXf_EVAL_SEEN;
4906 if (RExC_seen & REG_SEEN_CANY)
4907 r->extflags |= RXf_CANY_SEEN;
4908 if (RExC_seen & REG_SEEN_VERBARG)
4909 r->intflags |= PREGf_VERBARG_SEEN;
4910 if (RExC_seen & REG_SEEN_CUTGROUP)
4911 r->intflags |= PREGf_CUTGROUP_SEEN;
4912 if (RExC_paren_names)
4913 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
4915 RXp_PAREN_NAMES(r) = NULL;
4917 #ifdef STUPID_PATTERN_CHECKS
4918 if (RX_PRELEN(rx) == 0)
4919 r->extflags |= RXf_NULL;
4920 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4921 /* XXX: this should happen BEFORE we compile */
4922 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4923 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
4924 r->extflags |= RXf_WHITE;
4925 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
4926 r->extflags |= RXf_START_ONLY;
4928 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
4929 /* XXX: this should happen BEFORE we compile */
4930 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
4932 regnode *first = ri->program + 1;
4934 U8 nop = OP(NEXTOPER(first));
4936 if (PL_regkind[fop] == NOTHING && nop == END)
4937 r->extflags |= RXf_NULL;
4938 else if (PL_regkind[fop] == BOL && nop == END)
4939 r->extflags |= RXf_START_ONLY;
4940 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
4941 r->extflags |= RXf_WHITE;
4945 if (RExC_paren_names) {
4946 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
4947 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
4950 ri->name_list_idx = 0;
4952 if (RExC_recurse_count) {
4953 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
4954 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
4955 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
4958 Newxz(r->offs, RExC_npar, regexp_paren_pair);
4959 /* assume we don't need to swap parens around before we match */
4962 PerlIO_printf(Perl_debug_log,"Final program:\n");
4965 #ifdef RE_TRACK_PATTERN_OFFSETS
4966 DEBUG_OFFSETS_r(if (ri->u.offsets) {
4967 const U32 len = ri->u.offsets[0];
4969 GET_RE_DEBUG_FLAGS_DECL;
4970 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
4971 for (i = 1; i <= len; i++) {
4972 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
4973 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
4974 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
4976 PerlIO_printf(Perl_debug_log, "\n");
4982 #undef RE_ENGINE_PTR
4986 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
4989 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
4991 PERL_UNUSED_ARG(value);
4993 if (flags & RXapif_FETCH) {
4994 return reg_named_buff_fetch(rx, key, flags);
4995 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
4996 Perl_croak_no_modify(aTHX);
4998 } else if (flags & RXapif_EXISTS) {
4999 return reg_named_buff_exists(rx, key, flags)
5002 } else if (flags & RXapif_REGNAMES) {
5003 return reg_named_buff_all(rx, flags);
5004 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5005 return reg_named_buff_scalar(rx, flags);
5007 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5013 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5016 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5017 PERL_UNUSED_ARG(lastkey);
5019 if (flags & RXapif_FIRSTKEY)
5020 return reg_named_buff_firstkey(rx, flags);
5021 else if (flags & RXapif_NEXTKEY)
5022 return reg_named_buff_nextkey(rx, flags);
5024 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5030 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5033 AV *retarray = NULL;
5035 struct regexp *const rx = (struct regexp *)SvANY(r);
5037 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5039 if (flags & RXapif_ALL)
5042 if (rx && RXp_PAREN_NAMES(rx)) {
5043 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5046 SV* sv_dat=HeVAL(he_str);
5047 I32 *nums=(I32*)SvPVX(sv_dat);
5048 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5049 if ((I32)(rx->nparens) >= nums[i]
5050 && rx->offs[nums[i]].start != -1
5051 && rx->offs[nums[i]].end != -1)
5054 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5058 ret = newSVsv(&PL_sv_undef);
5061 av_push(retarray, ret);
5064 return newRV_noinc(MUTABLE_SV(retarray));
5071 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5074 struct regexp *const rx = (struct regexp *)SvANY(r);
5076 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5078 if (rx && RXp_PAREN_NAMES(rx)) {
5079 if (flags & RXapif_ALL) {
5080 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5082 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5096 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5098 struct regexp *const rx = (struct regexp *)SvANY(r);
5100 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5102 if ( rx && RXp_PAREN_NAMES(rx) ) {
5103 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5105 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5112 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5114 struct regexp *const rx = (struct regexp *)SvANY(r);
5115 GET_RE_DEBUG_FLAGS_DECL;
5117 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5119 if (rx && RXp_PAREN_NAMES(rx)) {
5120 HV *hv = RXp_PAREN_NAMES(rx);
5122 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5125 SV* sv_dat = HeVAL(temphe);
5126 I32 *nums = (I32*)SvPVX(sv_dat);
5127 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5128 if ((I32)(rx->lastparen) >= nums[i] &&
5129 rx->offs[nums[i]].start != -1 &&
5130 rx->offs[nums[i]].end != -1)
5136 if (parno || flags & RXapif_ALL) {
5137 return newSVhek(HeKEY_hek(temphe));
5145 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5150 struct regexp *const rx = (struct regexp *)SvANY(r);
5152 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5154 if (rx && RXp_PAREN_NAMES(rx)) {
5155 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5156 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5157 } else if (flags & RXapif_ONE) {
5158 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5159 av = MUTABLE_AV(SvRV(ret));
5160 length = av_len(av);
5162 return newSViv(length + 1);
5164 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5168 return &PL_sv_undef;
5172 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5174 struct regexp *const rx = (struct regexp *)SvANY(r);
5177 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5179 if (rx && RXp_PAREN_NAMES(rx)) {
5180 HV *hv= RXp_PAREN_NAMES(rx);
5182 (void)hv_iterinit(hv);
5183 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5186 SV* sv_dat = HeVAL(temphe);
5187 I32 *nums = (I32*)SvPVX(sv_dat);
5188 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5189 if ((I32)(rx->lastparen) >= nums[i] &&
5190 rx->offs[nums[i]].start != -1 &&
5191 rx->offs[nums[i]].end != -1)
5197 if (parno || flags & RXapif_ALL) {
5198 av_push(av, newSVhek(HeKEY_hek(temphe)));
5203 return newRV_noinc(MUTABLE_SV(av));
5207 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5210 struct regexp *const rx = (struct regexp *)SvANY(r);
5215 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5218 sv_setsv(sv,&PL_sv_undef);
5222 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5224 i = rx->offs[0].start;
5228 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5230 s = rx->subbeg + rx->offs[0].end;
5231 i = rx->sublen - rx->offs[0].end;
5234 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5235 (s1 = rx->offs[paren].start) != -1 &&
5236 (t1 = rx->offs[paren].end) != -1)
5240 s = rx->subbeg + s1;
5242 sv_setsv(sv,&PL_sv_undef);
5245 assert(rx->sublen >= (s - rx->subbeg) + i );
5247 const int oldtainted = PL_tainted;
5249 sv_setpvn(sv, s, i);
5250 PL_tainted = oldtainted;
5251 if ( (rx->extflags & RXf_CANY_SEEN)
5252 ? (RXp_MATCH_UTF8(rx)
5253 && (!i || is_utf8_string((U8*)s, i)))
5254 : (RXp_MATCH_UTF8(rx)) )
5261 if (RXp_MATCH_TAINTED(rx)) {
5262 if (SvTYPE(sv) >= SVt_PVMG) {
5263 MAGIC* const mg = SvMAGIC(sv);
5266 SvMAGIC_set(sv, mg->mg_moremagic);
5268 if ((mgt = SvMAGIC(sv))) {
5269 mg->mg_moremagic = mgt;
5270 SvMAGIC_set(sv, mg);
5280 sv_setsv(sv,&PL_sv_undef);
5286 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5287 SV const * const value)
5289 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5291 PERL_UNUSED_ARG(rx);
5292 PERL_UNUSED_ARG(paren);
5293 PERL_UNUSED_ARG(value);
5296 Perl_croak_no_modify(aTHX);
5300 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5303 struct regexp *const rx = (struct regexp *)SvANY(r);
5307 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5309 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5311 /* $` / ${^PREMATCH} */
5312 case RX_BUFF_IDX_PREMATCH:
5313 if (rx->offs[0].start != -1) {
5314 i = rx->offs[0].start;
5322 /* $' / ${^POSTMATCH} */
5323 case RX_BUFF_IDX_POSTMATCH:
5324 if (rx->offs[0].end != -1) {
5325 i = rx->sublen - rx->offs[0].end;
5327 s1 = rx->offs[0].end;
5333 /* $& / ${^MATCH}, $1, $2, ... */
5335 if (paren <= (I32)rx->nparens &&
5336 (s1 = rx->offs[paren].start) != -1 &&
5337 (t1 = rx->offs[paren].end) != -1)
5342 if (ckWARN(WARN_UNINITIALIZED))
5343 report_uninit((const SV *)sv);
5348 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5349 const char * const s = rx->subbeg + s1;
5354 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5361 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5363 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5364 PERL_UNUSED_ARG(rx);
5368 return newSVpvs("Regexp");
5371 /* Scans the name of a named buffer from the pattern.
5372 * If flags is REG_RSN_RETURN_NULL returns null.
5373 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5374 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5375 * to the parsed name as looked up in the RExC_paren_names hash.
5376 * If there is an error throws a vFAIL().. type exception.
5379 #define REG_RSN_RETURN_NULL 0
5380 #define REG_RSN_RETURN_NAME 1
5381 #define REG_RSN_RETURN_DATA 2
5384 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5386 char *name_start = RExC_parse;
5388 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5390 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5391 /* skip IDFIRST by using do...while */
5394 RExC_parse += UTF8SKIP(RExC_parse);
5395 } while (isALNUM_utf8((U8*)RExC_parse));
5399 } while (isALNUM(*RExC_parse));
5404 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5405 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5406 if ( flags == REG_RSN_RETURN_NAME)
5408 else if (flags==REG_RSN_RETURN_DATA) {
5411 if ( ! sv_name ) /* should not happen*/
5412 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5413 if (RExC_paren_names)
5414 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5416 sv_dat = HeVAL(he_str);
5418 vFAIL("Reference to nonexistent named group");
5422 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5429 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5430 int rem=(int)(RExC_end - RExC_parse); \
5439 if (RExC_lastparse!=RExC_parse) \
5440 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5443 iscut ? "..." : "<" \
5446 PerlIO_printf(Perl_debug_log,"%16s",""); \
5449 num = RExC_size + 1; \
5451 num=REG_NODE_NUM(RExC_emit); \
5452 if (RExC_lastnum!=num) \
5453 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5455 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5456 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5457 (int)((depth*2)), "", \
5461 RExC_lastparse=RExC_parse; \
5466 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5467 DEBUG_PARSE_MSG((funcname)); \
5468 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5470 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5471 DEBUG_PARSE_MSG((funcname)); \
5472 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5475 - reg - regular expression, i.e. main body or parenthesized thing
5477 * Caller must absorb opening parenthesis.
5479 * Combining parenthesis handling with the base level of regular expression
5480 * is a trifle forced, but the need to tie the tails of the branches to what
5481 * follows makes it hard to avoid.
5483 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5485 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5487 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5491 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5492 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5495 register regnode *ret; /* Will be the head of the group. */
5496 register regnode *br;
5497 register regnode *lastbr;
5498 register regnode *ender = NULL;
5499 register I32 parno = 0;
5501 U32 oregflags = RExC_flags;
5502 bool have_branch = 0;
5504 I32 freeze_paren = 0;
5505 I32 after_freeze = 0;
5507 /* for (?g), (?gc), and (?o) warnings; warning
5508 about (?c) will warn about (?g) -- japhy */
5510 #define WASTED_O 0x01
5511 #define WASTED_G 0x02
5512 #define WASTED_C 0x04
5513 #define WASTED_GC (0x02|0x04)
5514 I32 wastedflags = 0x00;
5516 char * parse_start = RExC_parse; /* MJD */
5517 char * const oregcomp_parse = RExC_parse;
5519 GET_RE_DEBUG_FLAGS_DECL;
5521 PERL_ARGS_ASSERT_REG;
5522 DEBUG_PARSE("reg ");
5524 *flagp = 0; /* Tentatively. */
5527 /* Make an OPEN node, if parenthesized. */
5529 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5530 char *start_verb = RExC_parse;
5531 STRLEN verb_len = 0;
5532 char *start_arg = NULL;
5533 unsigned char op = 0;
5535 int internal_argval = 0; /* internal_argval is only useful if !argok */
5536 while ( *RExC_parse && *RExC_parse != ')' ) {
5537 if ( *RExC_parse == ':' ) {
5538 start_arg = RExC_parse + 1;
5544 verb_len = RExC_parse - start_verb;
5547 while ( *RExC_parse && *RExC_parse != ')' )
5549 if ( *RExC_parse != ')' )
5550 vFAIL("Unterminated verb pattern argument");
5551 if ( RExC_parse == start_arg )
5554 if ( *RExC_parse != ')' )
5555 vFAIL("Unterminated verb pattern");
5558 switch ( *start_verb ) {
5559 case 'A': /* (*ACCEPT) */
5560 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5562 internal_argval = RExC_nestroot;
5565 case 'C': /* (*COMMIT) */
5566 if ( memEQs(start_verb,verb_len,"COMMIT") )
5569 case 'F': /* (*FAIL) */
5570 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5575 case ':': /* (*:NAME) */
5576 case 'M': /* (*MARK:NAME) */
5577 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5582 case 'P': /* (*PRUNE) */
5583 if ( memEQs(start_verb,verb_len,"PRUNE") )
5586 case 'S': /* (*SKIP) */
5587 if ( memEQs(start_verb,verb_len,"SKIP") )
5590 case 'T': /* (*THEN) */
5591 /* [19:06] <TimToady> :: is then */
5592 if ( memEQs(start_verb,verb_len,"THEN") ) {
5594 RExC_seen |= REG_SEEN_CUTGROUP;
5600 vFAIL3("Unknown verb pattern '%.*s'",
5601 verb_len, start_verb);
5604 if ( start_arg && internal_argval ) {
5605 vFAIL3("Verb pattern '%.*s' may not have an argument",
5606 verb_len, start_verb);
5607 } else if ( argok < 0 && !start_arg ) {
5608 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5609 verb_len, start_verb);
5611 ret = reganode(pRExC_state, op, internal_argval);
5612 if ( ! internal_argval && ! SIZE_ONLY ) {
5614 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5615 ARG(ret) = add_data( pRExC_state, 1, "S" );
5616 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5623 if (!internal_argval)
5624 RExC_seen |= REG_SEEN_VERBARG;
5625 } else if ( start_arg ) {
5626 vFAIL3("Verb pattern '%.*s' may not have an argument",
5627 verb_len, start_verb);
5629 ret = reg_node(pRExC_state, op);
5631 nextchar(pRExC_state);
5634 if (*RExC_parse == '?') { /* (?...) */
5635 bool is_logical = 0;
5636 const char * const seqstart = RExC_parse;
5639 paren = *RExC_parse++;
5640 ret = NULL; /* For look-ahead/behind. */
5643 case 'P': /* (?P...) variants for those used to PCRE/Python */
5644 paren = *RExC_parse++;
5645 if ( paren == '<') /* (?P<...>) named capture */
5647 else if (paren == '>') { /* (?P>name) named recursion */
5648 goto named_recursion;
5650 else if (paren == '=') { /* (?P=...) named backref */
5651 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5652 you change this make sure you change that */
5653 char* name_start = RExC_parse;
5655 SV *sv_dat = reg_scan_name(pRExC_state,
5656 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5657 if (RExC_parse == name_start || *RExC_parse != ')')
5658 vFAIL2("Sequence %.3s... not terminated",parse_start);
5661 num = add_data( pRExC_state, 1, "S" );
5662 RExC_rxi->data->data[num]=(void*)sv_dat;
5663 SvREFCNT_inc_simple_void(sv_dat);
5666 ret = reganode(pRExC_state,
5667 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
5671 Set_Node_Offset(ret, parse_start+1);
5672 Set_Node_Cur_Length(ret); /* MJD */
5674 nextchar(pRExC_state);
5678 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5680 case '<': /* (?<...) */
5681 if (*RExC_parse == '!')
5683 else if (*RExC_parse != '=')
5689 case '\'': /* (?'...') */
5690 name_start= RExC_parse;
5691 svname = reg_scan_name(pRExC_state,
5692 SIZE_ONLY ? /* reverse test from the others */
5693 REG_RSN_RETURN_NAME :
5694 REG_RSN_RETURN_NULL);
5695 if (RExC_parse == name_start) {
5697 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5700 if (*RExC_parse != paren)
5701 vFAIL2("Sequence (?%c... not terminated",
5702 paren=='>' ? '<' : paren);
5706 if (!svname) /* shouldnt happen */
5708 "panic: reg_scan_name returned NULL");
5709 if (!RExC_paren_names) {
5710 RExC_paren_names= newHV();
5711 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5713 RExC_paren_name_list= newAV();
5714 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5717 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5719 sv_dat = HeVAL(he_str);
5721 /* croak baby croak */
5723 "panic: paren_name hash element allocation failed");
5724 } else if ( SvPOK(sv_dat) ) {
5725 /* (?|...) can mean we have dupes so scan to check
5726 its already been stored. Maybe a flag indicating
5727 we are inside such a construct would be useful,
5728 but the arrays are likely to be quite small, so
5729 for now we punt -- dmq */
5730 IV count = SvIV(sv_dat);
5731 I32 *pv = (I32*)SvPVX(sv_dat);
5733 for ( i = 0 ; i < count ; i++ ) {
5734 if ( pv[i] == RExC_npar ) {
5740 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5741 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5742 pv[count] = RExC_npar;
5743 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5746 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5747 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5749 SvIV_set(sv_dat, 1);
5752 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5753 SvREFCNT_dec(svname);
5756 /*sv_dump(sv_dat);*/
5758 nextchar(pRExC_state);
5760 goto capturing_parens;
5762 RExC_seen |= REG_SEEN_LOOKBEHIND;
5764 case '=': /* (?=...) */
5765 RExC_seen_zerolen++;
5767 case '!': /* (?!...) */
5768 RExC_seen_zerolen++;
5769 if (*RExC_parse == ')') {
5770 ret=reg_node(pRExC_state, OPFAIL);
5771 nextchar(pRExC_state);
5775 case '|': /* (?|...) */
5776 /* branch reset, behave like a (?:...) except that
5777 buffers in alternations share the same numbers */
5779 after_freeze = freeze_paren = RExC_npar;
5781 case ':': /* (?:...) */
5782 case '>': /* (?>...) */
5784 case '$': /* (?$...) */
5785 case '@': /* (?@...) */
5786 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5788 case '#': /* (?#...) */
5789 while (*RExC_parse && *RExC_parse != ')')
5791 if (*RExC_parse != ')')
5792 FAIL("Sequence (?#... not terminated");
5793 nextchar(pRExC_state);
5796 case '0' : /* (?0) */
5797 case 'R' : /* (?R) */
5798 if (*RExC_parse != ')')
5799 FAIL("Sequence (?R) not terminated");
5800 ret = reg_node(pRExC_state, GOSTART);
5801 *flagp |= POSTPONED;
5802 nextchar(pRExC_state);
5805 { /* named and numeric backreferences */
5807 case '&': /* (?&NAME) */
5808 parse_start = RExC_parse - 1;
5811 SV *sv_dat = reg_scan_name(pRExC_state,
5812 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5813 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
5815 goto gen_recurse_regop;
5818 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5820 vFAIL("Illegal pattern");
5822 goto parse_recursion;
5824 case '-': /* (?-1) */
5825 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
5826 RExC_parse--; /* rewind to let it be handled later */
5830 case '1': case '2': case '3': case '4': /* (?1) */
5831 case '5': case '6': case '7': case '8': case '9':
5834 num = atoi(RExC_parse);
5835 parse_start = RExC_parse - 1; /* MJD */
5836 if (*RExC_parse == '-')
5838 while (isDIGIT(*RExC_parse))
5840 if (*RExC_parse!=')')
5841 vFAIL("Expecting close bracket");
5844 if ( paren == '-' ) {
5846 Diagram of capture buffer numbering.
5847 Top line is the normal capture buffer numbers
5848 Botton line is the negative indexing as from
5852 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
5856 num = RExC_npar + num;
5859 vFAIL("Reference to nonexistent group");
5861 } else if ( paren == '+' ) {
5862 num = RExC_npar + num - 1;
5865 ret = reganode(pRExC_state, GOSUB, num);
5867 if (num > (I32)RExC_rx->nparens) {
5869 vFAIL("Reference to nonexistent group");
5871 ARG2L_SET( ret, RExC_recurse_count++);
5873 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
5874 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
5878 RExC_seen |= REG_SEEN_RECURSE;
5879 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
5880 Set_Node_Offset(ret, parse_start); /* MJD */
5882 *flagp |= POSTPONED;
5883 nextchar(pRExC_state);
5885 } /* named and numeric backreferences */
5888 case '?': /* (??...) */
5890 if (*RExC_parse != '{') {
5892 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5895 *flagp |= POSTPONED;
5896 paren = *RExC_parse++;
5898 case '{': /* (?{...}) */
5903 char *s = RExC_parse;
5905 RExC_seen_zerolen++;
5906 RExC_seen |= REG_SEEN_EVAL;
5907 while (count && (c = *RExC_parse)) {
5918 if (*RExC_parse != ')') {
5920 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
5924 OP_4tree *sop, *rop;
5925 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
5928 Perl_save_re_context(aTHX);
5929 rop = sv_compile_2op(sv, &sop, "re", &pad);
5930 sop->op_private |= OPpREFCOUNTED;
5931 /* re_dup will OpREFCNT_inc */
5932 OpREFCNT_set(sop, 1);
5935 n = add_data(pRExC_state, 3, "nop");
5936 RExC_rxi->data->data[n] = (void*)rop;
5937 RExC_rxi->data->data[n+1] = (void*)sop;
5938 RExC_rxi->data->data[n+2] = (void*)pad;
5941 else { /* First pass */
5942 if (PL_reginterp_cnt < ++RExC_seen_evals
5944 /* No compiled RE interpolated, has runtime
5945 components ===> unsafe. */
5946 FAIL("Eval-group not allowed at runtime, use re 'eval'");
5947 if (PL_tainting && PL_tainted)
5948 FAIL("Eval-group in insecure regular expression");
5949 #if PERL_VERSION > 8
5950 if (IN_PERL_COMPILETIME)
5955 nextchar(pRExC_state);
5957 ret = reg_node(pRExC_state, LOGICAL);
5960 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
5961 /* deal with the length of this later - MJD */
5964 ret = reganode(pRExC_state, EVAL, n);
5965 Set_Node_Length(ret, RExC_parse - parse_start + 1);
5966 Set_Node_Offset(ret, parse_start);
5969 case '(': /* (?(?{...})...) and (?(?=...)...) */
5972 if (RExC_parse[0] == '?') { /* (?(?...)) */
5973 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
5974 || RExC_parse[1] == '<'
5975 || RExC_parse[1] == '{') { /* Lookahead or eval. */
5978 ret = reg_node(pRExC_state, LOGICAL);
5981 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
5985 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
5986 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
5988 char ch = RExC_parse[0] == '<' ? '>' : '\'';
5989 char *name_start= RExC_parse++;
5991 SV *sv_dat=reg_scan_name(pRExC_state,
5992 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5993 if (RExC_parse == name_start || *RExC_parse != ch)
5994 vFAIL2("Sequence (?(%c... not terminated",
5995 (ch == '>' ? '<' : ch));
5998 num = add_data( pRExC_state, 1, "S" );
5999 RExC_rxi->data->data[num]=(void*)sv_dat;
6000 SvREFCNT_inc_simple_void(sv_dat);
6002 ret = reganode(pRExC_state,NGROUPP,num);
6003 goto insert_if_check_paren;
6005 else if (RExC_parse[0] == 'D' &&
6006 RExC_parse[1] == 'E' &&
6007 RExC_parse[2] == 'F' &&
6008 RExC_parse[3] == 'I' &&
6009 RExC_parse[4] == 'N' &&
6010 RExC_parse[5] == 'E')
6012 ret = reganode(pRExC_state,DEFINEP,0);
6015 goto insert_if_check_paren;
6017 else if (RExC_parse[0] == 'R') {
6020 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6021 parno = atoi(RExC_parse++);
6022 while (isDIGIT(*RExC_parse))
6024 } else if (RExC_parse[0] == '&') {
6027 sv_dat = reg_scan_name(pRExC_state,
6028 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6029 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6031 ret = reganode(pRExC_state,INSUBP,parno);
6032 goto insert_if_check_paren;
6034 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6037 parno = atoi(RExC_parse++);
6039 while (isDIGIT(*RExC_parse))
6041 ret = reganode(pRExC_state, GROUPP, parno);
6043 insert_if_check_paren:
6044 if ((c = *nextchar(pRExC_state)) != ')')
6045 vFAIL("Switch condition not recognized");
6047 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
6048 br = regbranch(pRExC_state, &flags, 1,depth+1);
6050 br = reganode(pRExC_state, LONGJMP, 0);
6052 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
6053 c = *nextchar(pRExC_state);
6058 vFAIL("(?(DEFINE)....) does not allow branches");
6059 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6060 regbranch(pRExC_state, &flags, 1,depth+1);
6061 REGTAIL(pRExC_state, ret, lastbr);
6064 c = *nextchar(pRExC_state);
6069 vFAIL("Switch (?(condition)... contains too many branches");
6070 ender = reg_node(pRExC_state, TAIL);
6071 REGTAIL(pRExC_state, br, ender);
6073 REGTAIL(pRExC_state, lastbr, ender);
6074 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6077 REGTAIL(pRExC_state, ret, ender);
6078 RExC_size++; /* XXX WHY do we need this?!!
6079 For large programs it seems to be required
6080 but I can't figure out why. -- dmq*/
6084 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6088 RExC_parse--; /* for vFAIL to print correctly */
6089 vFAIL("Sequence (? incomplete");
6093 parse_flags: /* (?i) */
6095 U32 posflags = 0, negflags = 0;
6096 U32 *flagsp = &posflags;
6098 while (*RExC_parse) {
6099 /* && strchr("iogcmsx", *RExC_parse) */
6100 /* (?g), (?gc) and (?o) are useless here
6101 and must be globally applied -- japhy */
6102 switch (*RExC_parse) {
6103 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6104 case ONCE_PAT_MOD: /* 'o' */
6105 case GLOBAL_PAT_MOD: /* 'g' */
6106 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6107 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6108 if (! (wastedflags & wflagbit) ) {
6109 wastedflags |= wflagbit;
6112 "Useless (%s%c) - %suse /%c modifier",
6113 flagsp == &negflags ? "?-" : "?",
6115 flagsp == &negflags ? "don't " : "",
6122 case CONTINUE_PAT_MOD: /* 'c' */
6123 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6124 if (! (wastedflags & WASTED_C) ) {
6125 wastedflags |= WASTED_GC;
6128 "Useless (%sc) - %suse /gc modifier",
6129 flagsp == &negflags ? "?-" : "?",
6130 flagsp == &negflags ? "don't " : ""
6135 case KEEPCOPY_PAT_MOD: /* 'p' */
6136 if (flagsp == &negflags) {
6138 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
6140 *flagsp |= RXf_PMf_KEEPCOPY;
6144 if (flagsp == &negflags) {
6146 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6150 wastedflags = 0; /* reset so (?g-c) warns twice */
6156 RExC_flags |= posflags;
6157 RExC_flags &= ~negflags;
6159 oregflags |= posflags;
6160 oregflags &= ~negflags;
6162 nextchar(pRExC_state);
6173 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6178 }} /* one for the default block, one for the switch */
6185 ret = reganode(pRExC_state, OPEN, parno);
6188 RExC_nestroot = parno;
6189 if (RExC_seen & REG_SEEN_RECURSE
6190 && !RExC_open_parens[parno-1])
6192 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6193 "Setting open paren #%"IVdf" to %d\n",
6194 (IV)parno, REG_NODE_NUM(ret)));
6195 RExC_open_parens[parno-1]= ret;
6198 Set_Node_Length(ret, 1); /* MJD */
6199 Set_Node_Offset(ret, RExC_parse); /* MJD */
6207 /* Pick up the branches, linking them together. */
6208 parse_start = RExC_parse; /* MJD */
6209 br = regbranch(pRExC_state, &flags, 1,depth+1);
6212 if (RExC_npar > after_freeze)
6213 after_freeze = RExC_npar;
6214 RExC_npar = freeze_paren;
6217 /* branch_len = (paren != 0); */
6221 if (*RExC_parse == '|') {
6222 if (!SIZE_ONLY && RExC_extralen) {
6223 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6226 reginsert(pRExC_state, BRANCH, br, depth+1);
6227 Set_Node_Length(br, paren != 0);
6228 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6232 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6234 else if (paren == ':') {
6235 *flagp |= flags&SIMPLE;
6237 if (is_open) { /* Starts with OPEN. */
6238 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6240 else if (paren != '?') /* Not Conditional */
6242 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6244 while (*RExC_parse == '|') {
6245 if (!SIZE_ONLY && RExC_extralen) {
6246 ender = reganode(pRExC_state, LONGJMP,0);
6247 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6250 RExC_extralen += 2; /* Account for LONGJMP. */
6251 nextchar(pRExC_state);
6253 if (RExC_npar > after_freeze)
6254 after_freeze = RExC_npar;
6255 RExC_npar = freeze_paren;
6257 br = regbranch(pRExC_state, &flags, 0, depth+1);
6261 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6263 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6266 if (have_branch || paren != ':') {
6267 /* Make a closing node, and hook it on the end. */
6270 ender = reg_node(pRExC_state, TAIL);
6273 ender = reganode(pRExC_state, CLOSE, parno);
6274 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6275 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6276 "Setting close paren #%"IVdf" to %d\n",
6277 (IV)parno, REG_NODE_NUM(ender)));
6278 RExC_close_parens[parno-1]= ender;
6279 if (RExC_nestroot == parno)
6282 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6283 Set_Node_Length(ender,1); /* MJD */
6289 *flagp &= ~HASWIDTH;
6292 ender = reg_node(pRExC_state, SUCCEED);
6295 ender = reg_node(pRExC_state, END);
6297 assert(!RExC_opend); /* there can only be one! */
6302 REGTAIL(pRExC_state, lastbr, ender);
6304 if (have_branch && !SIZE_ONLY) {
6306 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6308 /* Hook the tails of the branches to the closing node. */
6309 for (br = ret; br; br = regnext(br)) {
6310 const U8 op = PL_regkind[OP(br)];
6312 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6314 else if (op == BRANCHJ) {
6315 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6323 static const char parens[] = "=!<,>";
6325 if (paren && (p = strchr(parens, paren))) {
6326 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6327 int flag = (p - parens) > 1;
6330 node = SUSPEND, flag = 0;
6331 reginsert(pRExC_state, node,ret, depth+1);
6332 Set_Node_Cur_Length(ret);
6333 Set_Node_Offset(ret, parse_start + 1);
6335 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6339 /* Check for proper termination. */
6341 RExC_flags = oregflags;
6342 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6343 RExC_parse = oregcomp_parse;
6344 vFAIL("Unmatched (");
6347 else if (!paren && RExC_parse < RExC_end) {
6348 if (*RExC_parse == ')') {
6350 vFAIL("Unmatched )");
6353 FAIL("Junk on end of regexp"); /* "Can't happen". */
6357 RExC_npar = after_freeze;
6362 - regbranch - one alternative of an | operator
6364 * Implements the concatenation operator.
6367 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6370 register regnode *ret;
6371 register regnode *chain = NULL;
6372 register regnode *latest;
6373 I32 flags = 0, c = 0;
6374 GET_RE_DEBUG_FLAGS_DECL;
6376 PERL_ARGS_ASSERT_REGBRANCH;
6378 DEBUG_PARSE("brnc");
6383 if (!SIZE_ONLY && RExC_extralen)
6384 ret = reganode(pRExC_state, BRANCHJ,0);
6386 ret = reg_node(pRExC_state, BRANCH);
6387 Set_Node_Length(ret, 1);
6391 if (!first && SIZE_ONLY)
6392 RExC_extralen += 1; /* BRANCHJ */
6394 *flagp = WORST; /* Tentatively. */
6397 nextchar(pRExC_state);
6398 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6400 latest = regpiece(pRExC_state, &flags,depth+1);
6401 if (latest == NULL) {
6402 if (flags & TRYAGAIN)
6406 else if (ret == NULL)
6408 *flagp |= flags&(HASWIDTH|POSTPONED);
6409 if (chain == NULL) /* First piece. */
6410 *flagp |= flags&SPSTART;
6413 REGTAIL(pRExC_state, chain, latest);
6418 if (chain == NULL) { /* Loop ran zero times. */
6419 chain = reg_node(pRExC_state, NOTHING);
6424 *flagp |= flags&SIMPLE;
6431 - regpiece - something followed by possible [*+?]
6433 * Note that the branching code sequences used for ? and the general cases
6434 * of * and + are somewhat optimized: they use the same NOTHING node as
6435 * both the endmarker for their branch list and the body of the last branch.
6436 * It might seem that this node could be dispensed with entirely, but the
6437 * endmarker role is not redundant.
6440 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6443 register regnode *ret;
6445 register char *next;
6447 const char * const origparse = RExC_parse;
6449 I32 max = REG_INFTY;
6451 const char *maxpos = NULL;
6452 GET_RE_DEBUG_FLAGS_DECL;
6454 PERL_ARGS_ASSERT_REGPIECE;
6456 DEBUG_PARSE("piec");
6458 ret = regatom(pRExC_state, &flags,depth+1);
6460 if (flags & TRYAGAIN)
6467 if (op == '{' && regcurly(RExC_parse)) {
6469 parse_start = RExC_parse; /* MJD */
6470 next = RExC_parse + 1;
6471 while (isDIGIT(*next) || *next == ',') {
6480 if (*next == '}') { /* got one */
6484 min = atoi(RExC_parse);
6488 maxpos = RExC_parse;
6490 if (!max && *maxpos != '0')
6491 max = REG_INFTY; /* meaning "infinity" */
6492 else if (max >= REG_INFTY)
6493 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6495 nextchar(pRExC_state);
6498 if ((flags&SIMPLE)) {
6499 RExC_naughty += 2 + RExC_naughty / 2;
6500 reginsert(pRExC_state, CURLY, ret, depth+1);
6501 Set_Node_Offset(ret, parse_start+1); /* MJD */
6502 Set_Node_Cur_Length(ret);
6505 regnode * const w = reg_node(pRExC_state, WHILEM);
6508 REGTAIL(pRExC_state, ret, w);
6509 if (!SIZE_ONLY && RExC_extralen) {
6510 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6511 reginsert(pRExC_state, NOTHING,ret, depth+1);
6512 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6514 reginsert(pRExC_state, CURLYX,ret, depth+1);
6516 Set_Node_Offset(ret, parse_start+1);
6517 Set_Node_Length(ret,
6518 op == '{' ? (RExC_parse - parse_start) : 1);
6520 if (!SIZE_ONLY && RExC_extralen)
6521 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6522 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6524 RExC_whilem_seen++, RExC_extralen += 3;
6525 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6534 vFAIL("Can't do {n,m} with n > m");
6536 ARG1_SET(ret, (U16)min);
6537 ARG2_SET(ret, (U16)max);
6549 #if 0 /* Now runtime fix should be reliable. */
6551 /* if this is reinstated, don't forget to put this back into perldiag:
6553 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6555 (F) The part of the regexp subject to either the * or + quantifier
6556 could match an empty string. The {#} shows in the regular
6557 expression about where the problem was discovered.
6561 if (!(flags&HASWIDTH) && op != '?')
6562 vFAIL("Regexp *+ operand could be empty");
6565 parse_start = RExC_parse;
6566 nextchar(pRExC_state);
6568 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6570 if (op == '*' && (flags&SIMPLE)) {
6571 reginsert(pRExC_state, STAR, ret, depth+1);
6575 else if (op == '*') {
6579 else if (op == '+' && (flags&SIMPLE)) {
6580 reginsert(pRExC_state, PLUS, ret, depth+1);
6584 else if (op == '+') {
6588 else if (op == '?') {
6593 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
6594 ckWARN3reg(RExC_parse,
6595 "%.*s matches null string many times",
6596 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6600 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6601 nextchar(pRExC_state);
6602 reginsert(pRExC_state, MINMOD, ret, depth+1);
6603 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6605 #ifndef REG_ALLOW_MINMOD_SUSPEND
6608 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6610 nextchar(pRExC_state);
6611 ender = reg_node(pRExC_state, SUCCEED);
6612 REGTAIL(pRExC_state, ret, ender);
6613 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6615 ender = reg_node(pRExC_state, TAIL);
6616 REGTAIL(pRExC_state, ret, ender);
6620 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6622 vFAIL("Nested quantifiers");
6629 /* reg_namedseq(pRExC_state,UVp)
6631 This is expected to be called by a parser routine that has
6632 recognized '\N' and needs to handle the rest. RExC_parse is
6633 expected to point at the first char following the N at the time
6636 The \N may be inside (indicated by valuep not being NULL) or outside a
6639 \N may begin either a named sequence, or if outside a character class, mean
6640 to match a non-newline. For non single-quoted regexes, the tokenizer has
6641 attempted to decide which, and in the case of a named sequence converted it
6642 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
6643 where c1... are the characters in the sequence. For single-quoted regexes,
6644 the tokenizer passes the \N sequence through unchanged; this code will not
6645 attempt to determine this nor expand those. The net effect is that if the
6646 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
6647 signals that this \N occurrence means to match a non-newline.
6649 Only the \N{U+...} form should occur in a character class, for the same
6650 reason that '.' inside a character class means to just match a period: it
6651 just doesn't make sense.
6653 If valuep is non-null then it is assumed that we are parsing inside
6654 of a charclass definition and the first codepoint in the resolved
6655 string is returned via *valuep and the routine will return NULL.
6656 In this mode if a multichar string is returned from the charnames
6657 handler, a warning will be issued, and only the first char in the
6658 sequence will be examined. If the string returned is zero length
6659 then the value of *valuep is undefined and NON-NULL will
6660 be returned to indicate failure. (This will NOT be a valid pointer
6663 If valuep is null then it is assumed that we are parsing normal text and a
6664 new EXACT node is inserted into the program containing the resolved string,
6665 and a pointer to the new node is returned. But if the string is zero length
6666 a NOTHING node is emitted instead.
6668 On success RExC_parse is set to the char following the endbrace.
6669 Parsing failures will generate a fatal error via vFAIL(...)
6672 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp)
6674 char * endbrace; /* '}' following the name */
6675 regnode *ret = NULL;
6677 char* parse_start = RExC_parse - 2; /* points to the '\N' */
6681 GET_RE_DEBUG_FLAGS_DECL;
6683 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6687 /* The [^\n] meaning of \N ignores spaces and comments under the /x
6688 * modifier. The other meaning does not */
6689 p = (RExC_flags & RXf_PMf_EXTENDED)
6690 ? regwhite( pRExC_state, RExC_parse )
6693 /* Disambiguate between \N meaning a named character versus \N meaning
6694 * [^\n]. The former is assumed when it can't be the latter. */
6695 if (*p != '{' || regcurly(p)) {
6698 /* no bare \N in a charclass */
6699 vFAIL("\\N in a character class must be a named character: \\N{...}");
6701 nextchar(pRExC_state);
6702 ret = reg_node(pRExC_state, REG_ANY);
6703 *flagp |= HASWIDTH|SIMPLE;
6706 Set_Node_Length(ret, 1); /* MJD */
6710 /* Here, we have decided it should be a named sequence */
6712 /* The test above made sure that the next real character is a '{', but
6713 * under the /x modifier, it could be separated by space (or a comment and
6714 * \n) and this is not allowed (for consistency with \x{...} and the
6715 * tokenizer handling of \N{NAME}). */
6716 if (*RExC_parse != '{') {
6717 vFAIL("Missing braces on \\N{}");
6720 RExC_parse++; /* Skip past the '{' */
6722 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
6723 || ! (endbrace == RExC_parse /* nothing between the {} */
6724 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
6725 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
6727 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
6728 vFAIL("\\N{NAME} must be resolved by the lexer");
6731 if (endbrace == RExC_parse) { /* empty: \N{} */
6733 RExC_parse = endbrace + 1;
6734 return reg_node(pRExC_state,NOTHING);
6738 ckWARNreg(RExC_parse,
6739 "Ignoring zero length \\N{} in character class"
6741 RExC_parse = endbrace + 1;
6744 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
6747 RExC_utf8 = 1; /* named sequences imply Unicode semantics */
6748 RExC_parse += 2; /* Skip past the 'U+' */
6750 if (valuep) { /* In a bracketed char class */
6751 /* We only pay attention to the first char of
6752 multichar strings being returned. I kinda wonder
6753 if this makes sense as it does change the behaviour
6754 from earlier versions, OTOH that behaviour was broken
6755 as well. XXX Solution is to recharacterize as
6756 [rest-of-class]|multi1|multi2... */
6758 STRLEN length_of_hex;
6759 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
6760 | PERL_SCAN_DISALLOW_PREFIX
6761 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6763 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
6764 if (endchar < endbrace) {
6765 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
6768 length_of_hex = (STRLEN)(endchar - RExC_parse);
6769 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
6771 /* The tokenizer should have guaranteed validity, but it's possible to
6772 * bypass it by using single quoting, so check */
6773 if (length_of_hex == 0
6774 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6776 RExC_parse += length_of_hex; /* Includes all the valid */
6777 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6778 ? UTF8SKIP(RExC_parse)
6780 /* Guard against malformed utf8 */
6781 if (RExC_parse >= endchar) RExC_parse = endchar;
6782 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6785 RExC_parse = endbrace + 1;
6786 if (endchar == endbrace) return NULL;
6788 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
6790 else { /* Not a char class */
6791 char *s; /* String to put in generated EXACT node */
6792 STRLEN len = 0; /* Its current length */
6793 char *endchar; /* Points to '.' or '}' ending cur char in the input
6796 ret = reg_node(pRExC_state,
6797 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
6800 /* Exact nodes can hold only a U8 length's of text = 255. Loop through
6801 * the input which is of the form now 'c1.c2.c3...}' until find the
6802 * ending brace or exeed length 255. The characters that exceed this
6803 * limit are dropped. The limit could be relaxed should it become
6804 * desirable by reparsing this as (?:\N{NAME}), so could generate
6805 * multiple EXACT nodes, as is done for just regular input. But this
6806 * is primarily a named character, and not intended to be a huge long
6807 * string, so 255 bytes should be good enough */
6809 STRLEN length_of_hex;
6810 I32 grok_flags = PERL_SCAN_ALLOW_UNDERSCORES
6811 | PERL_SCAN_DISALLOW_PREFIX
6812 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
6813 UV cp; /* Ord of current character */
6815 /* Code points are separated by dots. If none, there is only one
6816 * code point, and is terminated by the brace */
6817 endchar = RExC_parse + strcspn(RExC_parse, ".}");
6819 /* The values are Unicode even on EBCDIC machines */
6820 length_of_hex = (STRLEN)(endchar - RExC_parse);
6821 cp = grok_hex(RExC_parse, &length_of_hex, &grok_flags, NULL);
6822 if ( length_of_hex == 0
6823 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
6825 RExC_parse += length_of_hex; /* Includes all the valid */
6826 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
6827 ? UTF8SKIP(RExC_parse)
6829 /* Guard against malformed utf8 */
6830 if (RExC_parse >= endchar) RExC_parse = endchar;
6831 vFAIL("Invalid hexadecimal number in \\N{U+...}");
6834 if (! FOLD) { /* Not folding, just append to the string */
6837 /* Quit before adding this character if would exceed limit */
6838 if (len + UNISKIP(cp) > U8_MAX) break;
6840 unilen = reguni(pRExC_state, cp, s);
6845 } else { /* Folding, output the folded equivalent */
6846 STRLEN foldlen,numlen;
6847 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
6848 cp = toFOLD_uni(cp, tmpbuf, &foldlen);
6850 /* Quit before exceeding size limit */
6851 if (len + foldlen > U8_MAX) break;
6853 for (foldbuf = tmpbuf;
6857 cp = utf8_to_uvchr(foldbuf, &numlen);
6859 const STRLEN unilen = reguni(pRExC_state, cp, s);
6862 /* In EBCDIC the numlen and unilen can differ. */
6864 if (numlen >= foldlen)
6868 break; /* "Can't happen." */
6872 /* Point to the beginning of the next character in the sequence. */
6873 RExC_parse = endchar + 1;
6875 /* Quit if no more characters */
6876 if (RExC_parse >= endbrace) break;
6881 if (RExC_parse < endbrace) {
6882 ckWARNreg(RExC_parse - 1,
6883 "Using just the first characters returned by \\N{}");
6886 RExC_size += STR_SZ(len);
6889 RExC_emit += STR_SZ(len);
6892 RExC_parse = endbrace + 1;
6894 *flagp |= HASWIDTH; /* Not SIMPLE, as that causes the engine to fail
6895 with malformed in t/re/pat_advanced.t */
6897 Set_Node_Cur_Length(ret); /* MJD */
6898 nextchar(pRExC_state);
6908 * It returns the code point in utf8 for the value in *encp.
6909 * value: a code value in the source encoding
6910 * encp: a pointer to an Encode object
6912 * If the result from Encode is not a single character,
6913 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
6916 S_reg_recode(pTHX_ const char value, SV **encp)
6919 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
6920 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
6921 const STRLEN newlen = SvCUR(sv);
6922 UV uv = UNICODE_REPLACEMENT;
6924 PERL_ARGS_ASSERT_REG_RECODE;
6928 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
6931 if (!newlen || numlen != newlen) {
6932 uv = UNICODE_REPLACEMENT;
6940 - regatom - the lowest level
6942 Try to identify anything special at the start of the pattern. If there
6943 is, then handle it as required. This may involve generating a single regop,
6944 such as for an assertion; or it may involve recursing, such as to
6945 handle a () structure.
6947 If the string doesn't start with something special then we gobble up
6948 as much literal text as we can.
6950 Once we have been able to handle whatever type of thing started the
6951 sequence, we return.
6953 Note: we have to be careful with escapes, as they can be both literal
6954 and special, and in the case of \10 and friends can either, depending
6955 on context. Specifically there are two seperate switches for handling
6956 escape sequences, with the one for handling literal escapes requiring
6957 a dummy entry for all of the special escapes that are actually handled
6962 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6965 register regnode *ret = NULL;
6967 char *parse_start = RExC_parse;
6968 GET_RE_DEBUG_FLAGS_DECL;
6969 DEBUG_PARSE("atom");
6970 *flagp = WORST; /* Tentatively. */
6972 PERL_ARGS_ASSERT_REGATOM;
6975 switch ((U8)*RExC_parse) {
6977 RExC_seen_zerolen++;
6978 nextchar(pRExC_state);
6979 if (RExC_flags & RXf_PMf_MULTILINE)
6980 ret = reg_node(pRExC_state, MBOL);
6981 else if (RExC_flags & RXf_PMf_SINGLELINE)
6982 ret = reg_node(pRExC_state, SBOL);
6984 ret = reg_node(pRExC_state, BOL);
6985 Set_Node_Length(ret, 1); /* MJD */
6988 nextchar(pRExC_state);
6990 RExC_seen_zerolen++;
6991 if (RExC_flags & RXf_PMf_MULTILINE)
6992 ret = reg_node(pRExC_state, MEOL);
6993 else if (RExC_flags & RXf_PMf_SINGLELINE)
6994 ret = reg_node(pRExC_state, SEOL);
6996 ret = reg_node(pRExC_state, EOL);
6997 Set_Node_Length(ret, 1); /* MJD */
7000 nextchar(pRExC_state);
7001 if (RExC_flags & RXf_PMf_SINGLELINE)
7002 ret = reg_node(pRExC_state, SANY);
7004 ret = reg_node(pRExC_state, REG_ANY);
7005 *flagp |= HASWIDTH|SIMPLE;
7007 Set_Node_Length(ret, 1); /* MJD */
7011 char * const oregcomp_parse = ++RExC_parse;
7012 ret = regclass(pRExC_state,depth+1);
7013 if (*RExC_parse != ']') {
7014 RExC_parse = oregcomp_parse;
7015 vFAIL("Unmatched [");
7017 nextchar(pRExC_state);
7018 *flagp |= HASWIDTH|SIMPLE;
7019 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
7023 nextchar(pRExC_state);
7024 ret = reg(pRExC_state, 1, &flags,depth+1);
7026 if (flags & TRYAGAIN) {
7027 if (RExC_parse == RExC_end) {
7028 /* Make parent create an empty node if needed. */
7036 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7040 if (flags & TRYAGAIN) {
7044 vFAIL("Internal urp");
7045 /* Supposed to be caught earlier. */
7048 if (!regcurly(RExC_parse)) {
7057 vFAIL("Quantifier follows nothing");
7065 len=0; /* silence a spurious compiler warning */
7066 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
7067 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
7068 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
7069 ret = reganode(pRExC_state, FOLDCHAR, cp);
7070 Set_Node_Length(ret, 1); /* MJD */
7071 nextchar(pRExC_state); /* kill whitespace under /x */
7079 This switch handles escape sequences that resolve to some kind
7080 of special regop and not to literal text. Escape sequnces that
7081 resolve to literal text are handled below in the switch marked
7084 Every entry in this switch *must* have a corresponding entry
7085 in the literal escape switch. However, the opposite is not
7086 required, as the default for this switch is to jump to the
7087 literal text handling code.
7089 switch ((U8)*++RExC_parse) {
7094 /* Special Escapes */
7096 RExC_seen_zerolen++;
7097 ret = reg_node(pRExC_state, SBOL);
7099 goto finish_meta_pat;
7101 ret = reg_node(pRExC_state, GPOS);
7102 RExC_seen |= REG_SEEN_GPOS;
7104 goto finish_meta_pat;
7106 RExC_seen_zerolen++;
7107 ret = reg_node(pRExC_state, KEEPS);
7109 /* XXX:dmq : disabling in-place substitution seems to
7110 * be necessary here to avoid cases of memory corruption, as
7111 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7113 RExC_seen |= REG_SEEN_LOOKBEHIND;
7114 goto finish_meta_pat;
7116 ret = reg_node(pRExC_state, SEOL);
7118 RExC_seen_zerolen++; /* Do not optimize RE away */
7119 goto finish_meta_pat;
7121 ret = reg_node(pRExC_state, EOS);
7123 RExC_seen_zerolen++; /* Do not optimize RE away */
7124 goto finish_meta_pat;
7126 ret = reg_node(pRExC_state, CANY);
7127 RExC_seen |= REG_SEEN_CANY;
7128 *flagp |= HASWIDTH|SIMPLE;
7129 goto finish_meta_pat;
7131 ret = reg_node(pRExC_state, CLUMP);
7133 goto finish_meta_pat;
7135 ret = reg_node(pRExC_state, (U8)(LOC ? ALNUML : ALNUM));
7136 *flagp |= HASWIDTH|SIMPLE;
7137 goto finish_meta_pat;
7139 ret = reg_node(pRExC_state, (U8)(LOC ? NALNUML : NALNUM));
7140 *flagp |= HASWIDTH|SIMPLE;
7141 goto finish_meta_pat;
7143 RExC_seen_zerolen++;
7144 RExC_seen |= REG_SEEN_LOOKBEHIND;
7145 ret = reg_node(pRExC_state, (U8)(LOC ? BOUNDL : BOUND));
7147 goto finish_meta_pat;
7149 RExC_seen_zerolen++;
7150 RExC_seen |= REG_SEEN_LOOKBEHIND;
7151 ret = reg_node(pRExC_state, (U8)(LOC ? NBOUNDL : NBOUND));
7153 goto finish_meta_pat;
7155 ret = reg_node(pRExC_state, (U8)(LOC ? SPACEL : SPACE));
7156 *flagp |= HASWIDTH|SIMPLE;
7157 goto finish_meta_pat;
7159 ret = reg_node(pRExC_state, (U8)(LOC ? NSPACEL : NSPACE));
7160 *flagp |= HASWIDTH|SIMPLE;
7161 goto finish_meta_pat;
7163 ret = reg_node(pRExC_state, DIGIT);
7164 *flagp |= HASWIDTH|SIMPLE;
7165 goto finish_meta_pat;
7167 ret = reg_node(pRExC_state, NDIGIT);
7168 *flagp |= HASWIDTH|SIMPLE;
7169 goto finish_meta_pat;
7171 ret = reg_node(pRExC_state, LNBREAK);
7172 *flagp |= HASWIDTH|SIMPLE;
7173 goto finish_meta_pat;
7175 ret = reg_node(pRExC_state, HORIZWS);
7176 *flagp |= HASWIDTH|SIMPLE;
7177 goto finish_meta_pat;
7179 ret = reg_node(pRExC_state, NHORIZWS);
7180 *flagp |= HASWIDTH|SIMPLE;
7181 goto finish_meta_pat;
7183 ret = reg_node(pRExC_state, VERTWS);
7184 *flagp |= HASWIDTH|SIMPLE;
7185 goto finish_meta_pat;
7187 ret = reg_node(pRExC_state, NVERTWS);
7188 *flagp |= HASWIDTH|SIMPLE;
7190 nextchar(pRExC_state);
7191 Set_Node_Length(ret, 2); /* MJD */
7196 char* const oldregxend = RExC_end;
7198 char* parse_start = RExC_parse - 2;
7201 if (RExC_parse[1] == '{') {
7202 /* a lovely hack--pretend we saw [\pX] instead */
7203 RExC_end = strchr(RExC_parse, '}');
7205 const U8 c = (U8)*RExC_parse;
7207 RExC_end = oldregxend;
7208 vFAIL2("Missing right brace on \\%c{}", c);
7213 RExC_end = RExC_parse + 2;
7214 if (RExC_end > oldregxend)
7215 RExC_end = oldregxend;
7219 ret = regclass(pRExC_state,depth+1);
7221 RExC_end = oldregxend;
7224 Set_Node_Offset(ret, parse_start + 2);
7225 Set_Node_Cur_Length(ret);
7226 nextchar(pRExC_state);
7227 *flagp |= HASWIDTH|SIMPLE;
7231 /* Handle \N and \N{NAME} here and not below because it can be
7232 multicharacter. join_exact() will join them up later on.
7233 Also this makes sure that things like /\N{BLAH}+/ and
7234 \N{BLAH} being multi char Just Happen. dmq*/
7236 ret= reg_namedseq(pRExC_state, NULL, flagp);
7238 case 'k': /* Handle \k<NAME> and \k'NAME' */
7241 char ch= RExC_parse[1];
7242 if (ch != '<' && ch != '\'' && ch != '{') {
7244 vFAIL2("Sequence %.2s... not terminated",parse_start);
7246 /* this pretty much dupes the code for (?P=...) in reg(), if
7247 you change this make sure you change that */
7248 char* name_start = (RExC_parse += 2);
7250 SV *sv_dat = reg_scan_name(pRExC_state,
7251 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7252 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7253 if (RExC_parse == name_start || *RExC_parse != ch)
7254 vFAIL2("Sequence %.3s... not terminated",parse_start);
7257 num = add_data( pRExC_state, 1, "S" );
7258 RExC_rxi->data->data[num]=(void*)sv_dat;
7259 SvREFCNT_inc_simple_void(sv_dat);
7263 ret = reganode(pRExC_state,
7264 (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
7268 /* override incorrect value set in reganode MJD */
7269 Set_Node_Offset(ret, parse_start+1);
7270 Set_Node_Cur_Length(ret); /* MJD */
7271 nextchar(pRExC_state);
7277 case '1': case '2': case '3': case '4':
7278 case '5': case '6': case '7': case '8': case '9':
7281 bool isg = *RExC_parse == 'g';
7286 if (*RExC_parse == '{') {
7290 if (*RExC_parse == '-') {
7294 if (hasbrace && !isDIGIT(*RExC_parse)) {
7295 if (isrel) RExC_parse--;
7297 goto parse_named_seq;
7299 num = atoi(RExC_parse);
7300 if (isg && num == 0)
7301 vFAIL("Reference to invalid group 0");
7303 num = RExC_npar - num;
7305 vFAIL("Reference to nonexistent or unclosed group");
7307 if (!isg && num > 9 && num >= RExC_npar)
7310 char * const parse_start = RExC_parse - 1; /* MJD */
7311 while (isDIGIT(*RExC_parse))
7313 if (parse_start == RExC_parse - 1)
7314 vFAIL("Unterminated \\g... pattern");
7316 if (*RExC_parse != '}')
7317 vFAIL("Unterminated \\g{...} pattern");
7321 if (num > (I32)RExC_rx->nparens)
7322 vFAIL("Reference to nonexistent group");
7325 ret = reganode(pRExC_state,
7326 (U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
7330 /* override incorrect value set in reganode MJD */
7331 Set_Node_Offset(ret, parse_start+1);
7332 Set_Node_Cur_Length(ret); /* MJD */
7334 nextchar(pRExC_state);
7339 if (RExC_parse >= RExC_end)
7340 FAIL("Trailing \\");
7343 /* Do not generate "unrecognized" warnings here, we fall
7344 back into the quick-grab loop below */
7351 if (RExC_flags & RXf_PMf_EXTENDED) {
7352 if ( reg_skipcomment( pRExC_state ) )
7359 register STRLEN len;
7364 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7366 parse_start = RExC_parse - 1;
7372 ret = reg_node(pRExC_state,
7373 (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
7375 for (len = 0, p = RExC_parse - 1;
7376 len < 127 && p < RExC_end;
7379 char * const oldp = p;
7381 if (RExC_flags & RXf_PMf_EXTENDED)
7382 p = regwhite( pRExC_state, p );
7387 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7388 goto normal_default;
7398 /* Literal Escapes Switch
7400 This switch is meant to handle escape sequences that
7401 resolve to a literal character.
7403 Every escape sequence that represents something
7404 else, like an assertion or a char class, is handled
7405 in the switch marked 'Special Escapes' above in this
7406 routine, but also has an entry here as anything that
7407 isn't explicitly mentioned here will be treated as
7408 an unescaped equivalent literal.
7412 /* These are all the special escapes. */
7416 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7417 goto normal_default;
7418 case 'A': /* Start assertion */
7419 case 'b': case 'B': /* Word-boundary assertion*/
7420 case 'C': /* Single char !DANGEROUS! */
7421 case 'd': case 'D': /* digit class */
7422 case 'g': case 'G': /* generic-backref, pos assertion */
7423 case 'h': case 'H': /* HORIZWS */
7424 case 'k': case 'K': /* named backref, keep marker */
7425 case 'N': /* named char sequence */
7426 case 'p': case 'P': /* Unicode property */
7427 case 'R': /* LNBREAK */
7428 case 's': case 'S': /* space class */
7429 case 'v': case 'V': /* VERTWS */
7430 case 'w': case 'W': /* word class */
7431 case 'X': /* eXtended Unicode "combining character sequence" */
7432 case 'z': case 'Z': /* End of line/string assertion */
7436 /* Anything after here is an escape that resolves to a
7437 literal. (Except digits, which may or may not)
7456 ender = ASCII_TO_NATIVE('\033');
7460 ender = ASCII_TO_NATIVE('\007');
7465 STRLEN brace_len = len;
7467 const char* error_msg;
7469 bool valid = grok_bslash_o(p,
7476 RExC_parse = p; /* going to die anyway; point
7477 to exact spot of failure */
7484 if (PL_encoding && ender < 0x100) {
7485 goto recode_encoding;
7494 char* const e = strchr(p, '}');
7498 vFAIL("Missing right brace on \\x{}");
7501 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7502 | PERL_SCAN_DISALLOW_PREFIX;
7503 STRLEN numlen = e - p - 1;
7504 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7511 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7513 ender = grok_hex(p, &numlen, &flags, NULL);
7516 if (PL_encoding && ender < 0x100)
7517 goto recode_encoding;
7521 ender = grok_bslash_c(*p++, SIZE_ONLY);
7523 case '0': case '1': case '2': case '3':case '4':
7524 case '5': case '6': case '7': case '8':case '9':
7526 (isDIGIT(p[1]) && atoi(p) >= RExC_npar) ) {
7529 ender = grok_oct(p, &numlen, &flags, NULL);
7539 if (PL_encoding && ender < 0x100)
7540 goto recode_encoding;
7544 SV* enc = PL_encoding;
7545 ender = reg_recode((const char)(U8)ender, &enc);
7546 if (!enc && SIZE_ONLY)
7547 ckWARNreg(p, "Invalid escape in the specified encoding");
7553 FAIL("Trailing \\");
7556 if (!SIZE_ONLY&& isALPHA(*p))
7557 ckWARN2reg(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7558 goto normal_default;
7563 if (UTF8_IS_START(*p) && UTF) {
7565 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7566 &numlen, UTF8_ALLOW_DEFAULT);
7573 if ( RExC_flags & RXf_PMf_EXTENDED)
7574 p = regwhite( pRExC_state, p );
7576 /* Prime the casefolded buffer. */
7577 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7579 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7584 /* Emit all the Unicode characters. */
7586 for (foldbuf = tmpbuf;
7588 foldlen -= numlen) {
7589 ender = utf8_to_uvchr(foldbuf, &numlen);
7591 const STRLEN unilen = reguni(pRExC_state, ender, s);
7594 /* In EBCDIC the numlen
7595 * and unilen can differ. */
7597 if (numlen >= foldlen)
7601 break; /* "Can't happen." */
7605 const STRLEN unilen = reguni(pRExC_state, ender, s);
7614 REGC((char)ender, s++);
7620 /* Emit all the Unicode characters. */
7622 for (foldbuf = tmpbuf;
7624 foldlen -= numlen) {
7625 ender = utf8_to_uvchr(foldbuf, &numlen);
7627 const STRLEN unilen = reguni(pRExC_state, ender, s);
7630 /* In EBCDIC the numlen
7631 * and unilen can differ. */
7633 if (numlen >= foldlen)
7641 const STRLEN unilen = reguni(pRExC_state, ender, s);
7650 REGC((char)ender, s++);
7654 Set_Node_Cur_Length(ret); /* MJD */
7655 nextchar(pRExC_state);
7657 /* len is STRLEN which is unsigned, need to copy to signed */
7660 vFAIL("Internal disaster");
7664 if (len == 1 && UNI_IS_INVARIANT(ender))
7668 RExC_size += STR_SZ(len);
7671 RExC_emit += STR_SZ(len);
7681 S_regwhite( RExC_state_t *pRExC_state, char *p )
7683 const char *e = RExC_end;
7685 PERL_ARGS_ASSERT_REGWHITE;
7690 else if (*p == '#') {
7699 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
7707 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
7708 Character classes ([:foo:]) can also be negated ([:^foo:]).
7709 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
7710 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
7711 but trigger failures because they are currently unimplemented. */
7713 #define POSIXCC_DONE(c) ((c) == ':')
7714 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
7715 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
7718 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
7721 I32 namedclass = OOB_NAMEDCLASS;
7723 PERL_ARGS_ASSERT_REGPPOSIXCC;
7725 if (value == '[' && RExC_parse + 1 < RExC_end &&
7726 /* I smell either [: or [= or [. -- POSIX has been here, right? */
7727 POSIXCC(UCHARAT(RExC_parse))) {
7728 const char c = UCHARAT(RExC_parse);
7729 char* const s = RExC_parse++;
7731 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
7733 if (RExC_parse == RExC_end)
7734 /* Grandfather lone [:, [=, [. */
7737 const char* const t = RExC_parse++; /* skip over the c */
7740 if (UCHARAT(RExC_parse) == ']') {
7741 const char *posixcc = s + 1;
7742 RExC_parse++; /* skip over the ending ] */
7745 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
7746 const I32 skip = t - posixcc;
7748 /* Initially switch on the length of the name. */
7751 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
7752 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
7755 /* Names all of length 5. */
7756 /* alnum alpha ascii blank cntrl digit graph lower
7757 print punct space upper */
7758 /* Offset 4 gives the best switch position. */
7759 switch (posixcc[4]) {
7761 if (memEQ(posixcc, "alph", 4)) /* alpha */
7762 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
7765 if (memEQ(posixcc, "spac", 4)) /* space */
7766 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
7769 if (memEQ(posixcc, "grap", 4)) /* graph */
7770 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
7773 if (memEQ(posixcc, "asci", 4)) /* ascii */
7774 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
7777 if (memEQ(posixcc, "blan", 4)) /* blank */
7778 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
7781 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
7782 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
7785 if (memEQ(posixcc, "alnu", 4)) /* alnum */
7786 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
7789 if (memEQ(posixcc, "lowe", 4)) /* lower */
7790 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
7791 else if (memEQ(posixcc, "uppe", 4)) /* upper */
7792 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
7795 if (memEQ(posixcc, "digi", 4)) /* digit */
7796 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
7797 else if (memEQ(posixcc, "prin", 4)) /* print */
7798 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
7799 else if (memEQ(posixcc, "punc", 4)) /* punct */
7800 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
7805 if (memEQ(posixcc, "xdigit", 6))
7806 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
7810 if (namedclass == OOB_NAMEDCLASS)
7811 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
7813 assert (posixcc[skip] == ':');
7814 assert (posixcc[skip+1] == ']');
7815 } else if (!SIZE_ONLY) {
7816 /* [[=foo=]] and [[.foo.]] are still future. */
7818 /* adjust RExC_parse so the warning shows after
7820 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
7822 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7825 /* Maternal grandfather:
7826 * "[:" ending in ":" but not in ":]" */
7836 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
7840 PERL_ARGS_ASSERT_CHECKPOSIXCC;
7842 if (POSIXCC(UCHARAT(RExC_parse))) {
7843 const char *s = RExC_parse;
7844 const char c = *s++;
7848 if (*s && c == *s && s[1] == ']') {
7850 "POSIX syntax [%c %c] belongs inside character classes",
7853 /* [[=foo=]] and [[.foo.]] are still future. */
7854 if (POSIXCC_NOTYET(c)) {
7855 /* adjust RExC_parse so the error shows after
7857 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
7859 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
7866 #define _C_C_T_(NAME,TEST,WORD) \
7869 ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
7871 for (value = 0; value < 256; value++) \
7873 ANYOF_BITMAP_SET(ret, value); \
7878 case ANYOF_N##NAME: \
7880 ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
7882 for (value = 0; value < 256; value++) \
7884 ANYOF_BITMAP_SET(ret, value); \
7890 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
7892 for (value = 0; value < 256; value++) \
7894 ANYOF_BITMAP_SET(ret, value); \
7898 case ANYOF_N##NAME: \
7899 for (value = 0; value < 256; value++) \
7901 ANYOF_BITMAP_SET(ret, value); \
7907 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
7908 so that it is possible to override the option here without having to
7909 rebuild the entire core. as we are required to do if we change regcomp.h
7910 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
7912 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
7913 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
7916 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
7917 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
7919 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
7923 parse a class specification and produce either an ANYOF node that
7924 matches the pattern or if the pattern matches a single char only and
7925 that char is < 256 and we are case insensitive then we produce an
7930 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
7933 register UV nextvalue;
7934 register IV prevvalue = OOB_UNICODE;
7935 register IV range = 0;
7936 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
7937 register regnode *ret;
7940 char *rangebegin = NULL;
7941 bool need_class = 0;
7944 bool optimize_invert = TRUE;
7945 AV* unicode_alternate = NULL;
7947 UV literal_endpoint = 0;
7949 UV stored = 0; /* number of chars stored in the class */
7951 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
7952 case we need to change the emitted regop to an EXACT. */
7953 const char * orig_parse = RExC_parse;
7954 GET_RE_DEBUG_FLAGS_DECL;
7956 PERL_ARGS_ASSERT_REGCLASS;
7958 PERL_UNUSED_ARG(depth);
7961 DEBUG_PARSE("clas");
7963 /* Assume we are going to generate an ANYOF node. */
7964 ret = reganode(pRExC_state, ANYOF, 0);
7967 ANYOF_FLAGS(ret) = 0;
7969 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
7973 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
7977 RExC_size += ANYOF_SKIP;
7978 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
7981 RExC_emit += ANYOF_SKIP;
7983 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
7985 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
7986 ANYOF_BITMAP_ZERO(ret);
7987 listsv = newSVpvs("# comment\n");
7990 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
7992 if (!SIZE_ONLY && POSIXCC(nextvalue))
7993 checkposixcc(pRExC_state);
7995 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
7996 if (UCHARAT(RExC_parse) == ']')
8000 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
8004 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
8007 rangebegin = RExC_parse;
8009 value = utf8n_to_uvchr((U8*)RExC_parse,
8010 RExC_end - RExC_parse,
8011 &numlen, UTF8_ALLOW_DEFAULT);
8012 RExC_parse += numlen;
8015 value = UCHARAT(RExC_parse++);
8017 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8018 if (value == '[' && POSIXCC(nextvalue))
8019 namedclass = regpposixcc(pRExC_state, value);
8020 else if (value == '\\') {
8022 value = utf8n_to_uvchr((U8*)RExC_parse,
8023 RExC_end - RExC_parse,
8024 &numlen, UTF8_ALLOW_DEFAULT);
8025 RExC_parse += numlen;
8028 value = UCHARAT(RExC_parse++);
8029 /* Some compilers cannot handle switching on 64-bit integer
8030 * values, therefore value cannot be an UV. Yes, this will
8031 * be a problem later if we want switch on Unicode.
8032 * A similar issue a little bit later when switching on
8033 * namedclass. --jhi */
8034 switch ((I32)value) {
8035 case 'w': namedclass = ANYOF_ALNUM; break;
8036 case 'W': namedclass = ANYOF_NALNUM; break;
8037 case 's': namedclass = ANYOF_SPACE; break;
8038 case 'S': namedclass = ANYOF_NSPACE; break;
8039 case 'd': namedclass = ANYOF_DIGIT; break;
8040 case 'D': namedclass = ANYOF_NDIGIT; break;
8041 case 'v': namedclass = ANYOF_VERTWS; break;
8042 case 'V': namedclass = ANYOF_NVERTWS; break;
8043 case 'h': namedclass = ANYOF_HORIZWS; break;
8044 case 'H': namedclass = ANYOF_NHORIZWS; break;
8045 case 'N': /* Handle \N{NAME} in class */
8047 /* We only pay attention to the first char of
8048 multichar strings being returned. I kinda wonder
8049 if this makes sense as it does change the behaviour
8050 from earlier versions, OTOH that behaviour was broken
8052 UV v; /* value is register so we cant & it /grrr */
8053 if (reg_namedseq(pRExC_state, &v, NULL)) {
8063 if (RExC_parse >= RExC_end)
8064 vFAIL2("Empty \\%c{}", (U8)value);
8065 if (*RExC_parse == '{') {
8066 const U8 c = (U8)value;
8067 e = strchr(RExC_parse++, '}');
8069 vFAIL2("Missing right brace on \\%c{}", c);
8070 while (isSPACE(UCHARAT(RExC_parse)))
8072 if (e == RExC_parse)
8073 vFAIL2("Empty \\%c{}", c);
8075 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
8083 if (UCHARAT(RExC_parse) == '^') {
8086 value = value == 'p' ? 'P' : 'p'; /* toggle */
8087 while (isSPACE(UCHARAT(RExC_parse))) {
8092 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
8093 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
8096 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8097 namedclass = ANYOF_MAX; /* no official name, but it's named */
8100 case 'n': value = '\n'; break;
8101 case 'r': value = '\r'; break;
8102 case 't': value = '\t'; break;
8103 case 'f': value = '\f'; break;
8104 case 'b': value = '\b'; break;
8105 case 'e': value = ASCII_TO_NATIVE('\033');break;
8106 case 'a': value = ASCII_TO_NATIVE('\007');break;
8108 RExC_parse--; /* function expects to be pointed at the 'o' */
8110 const char* error_msg;
8111 bool valid = grok_bslash_o(RExC_parse,
8116 RExC_parse += numlen;
8121 if (PL_encoding && value < 0x100) {
8122 goto recode_encoding;
8126 if (*RExC_parse == '{') {
8127 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8128 | PERL_SCAN_DISALLOW_PREFIX;
8129 char * const e = strchr(RExC_parse++, '}');
8131 vFAIL("Missing right brace on \\x{}");
8133 numlen = e - RExC_parse;
8134 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8138 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8140 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8141 RExC_parse += numlen;
8143 if (PL_encoding && value < 0x100)
8144 goto recode_encoding;
8147 value = grok_bslash_c(*RExC_parse++, SIZE_ONLY);
8149 case '0': case '1': case '2': case '3': case '4':
8150 case '5': case '6': case '7': case '8': case '9':
8154 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8155 RExC_parse += numlen;
8156 if (PL_encoding && value < 0x100)
8157 goto recode_encoding;
8162 SV* enc = PL_encoding;
8163 value = reg_recode((const char)(U8)value, &enc);
8164 if (!enc && SIZE_ONLY)
8165 ckWARNreg(RExC_parse,
8166 "Invalid escape in the specified encoding");
8170 if (!SIZE_ONLY && isALPHA(value))
8171 ckWARN2reg(RExC_parse,
8172 "Unrecognized escape \\%c in character class passed through",
8176 } /* end of \blah */
8182 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8184 if (!SIZE_ONLY && !need_class)
8185 ANYOF_CLASS_ZERO(ret);
8189 /* a bad range like a-\d, a-[:digit:] ? */
8193 RExC_parse >= rangebegin ?
8194 RExC_parse - rangebegin : 0;
8195 ckWARN4reg(RExC_parse,
8196 "False [] range \"%*.*s\"",
8199 if (prevvalue < 256) {
8200 ANYOF_BITMAP_SET(ret, prevvalue);
8201 ANYOF_BITMAP_SET(ret, '-');
8204 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8205 Perl_sv_catpvf(aTHX_ listsv,
8206 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8210 range = 0; /* this was not a true range */
8216 const char *what = NULL;
8219 if (namedclass > OOB_NAMEDCLASS)
8220 optimize_invert = FALSE;
8221 /* Possible truncation here but in some 64-bit environments
8222 * the compiler gets heartburn about switch on 64-bit values.
8223 * A similar issue a little earlier when switching on value.
8225 switch ((I32)namedclass) {
8227 case _C_C_T_(ALNUMC, isALNUMC(value), POSIX_CC_UNI_NAME("Alnum"));
8228 case _C_C_T_(ALPHA, isALPHA(value), POSIX_CC_UNI_NAME("Alpha"));
8229 case _C_C_T_(BLANK, isBLANK(value), POSIX_CC_UNI_NAME("Blank"));
8230 case _C_C_T_(CNTRL, isCNTRL(value), POSIX_CC_UNI_NAME("Cntrl"));
8231 case _C_C_T_(GRAPH, isGRAPH(value), POSIX_CC_UNI_NAME("Graph"));
8232 case _C_C_T_(LOWER, isLOWER(value), POSIX_CC_UNI_NAME("Lower"));
8233 case _C_C_T_(PRINT, isPRINT(value), POSIX_CC_UNI_NAME("Print"));
8234 case _C_C_T_(PSXSPC, isPSXSPC(value), POSIX_CC_UNI_NAME("Space"));
8235 case _C_C_T_(PUNCT, isPUNCT(value), POSIX_CC_UNI_NAME("Punct"));
8236 case _C_C_T_(UPPER, isUPPER(value), POSIX_CC_UNI_NAME("Upper"));
8237 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8238 case _C_C_T_(ALNUM, isALNUM(value), "Word");
8239 case _C_C_T_(SPACE, isSPACE(value), "SpacePerl");
8241 case _C_C_T_(SPACE, isSPACE(value), "PerlSpace");
8242 case _C_C_T_(ALNUM, isALNUM(value), "PerlWord");
8244 case _C_C_T_(XDIGIT, isXDIGIT(value), "XDigit");
8245 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8246 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8249 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8252 for (value = 0; value < 128; value++)
8253 ANYOF_BITMAP_SET(ret, value);
8255 for (value = 0; value < 256; value++) {
8257 ANYOF_BITMAP_SET(ret, value);
8266 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8269 for (value = 128; value < 256; value++)
8270 ANYOF_BITMAP_SET(ret, value);
8272 for (value = 0; value < 256; value++) {
8273 if (!isASCII(value))
8274 ANYOF_BITMAP_SET(ret, value);
8283 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8285 /* consecutive digits assumed */
8286 for (value = '0'; value <= '9'; value++)
8287 ANYOF_BITMAP_SET(ret, value);
8290 what = POSIX_CC_UNI_NAME("Digit");
8294 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8296 /* consecutive digits assumed */
8297 for (value = 0; value < '0'; value++)
8298 ANYOF_BITMAP_SET(ret, value);
8299 for (value = '9' + 1; value < 256; value++)
8300 ANYOF_BITMAP_SET(ret, value);
8303 what = POSIX_CC_UNI_NAME("Digit");
8306 /* this is to handle \p and \P */
8309 vFAIL("Invalid [::] class");
8313 /* Strings such as "+utf8::isWord\n" */
8314 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8317 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8320 } /* end of namedclass \blah */
8323 if (prevvalue > (IV)value) /* b-a */ {
8324 const int w = RExC_parse - rangebegin;
8325 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8326 range = 0; /* not a valid range */
8330 prevvalue = value; /* save the beginning of the range */
8331 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8332 RExC_parse[1] != ']') {
8335 /* a bad range like \w-, [:word:]- ? */
8336 if (namedclass > OOB_NAMEDCLASS) {
8337 if (ckWARN(WARN_REGEXP)) {
8339 RExC_parse >= rangebegin ?
8340 RExC_parse - rangebegin : 0;
8342 "False [] range \"%*.*s\"",
8346 ANYOF_BITMAP_SET(ret, '-');
8348 range = 1; /* yeah, it's a range! */
8349 continue; /* but do it the next time */
8353 /* now is the next time */
8354 /*stored += (value - prevvalue + 1);*/
8356 if (prevvalue < 256) {
8357 const IV ceilvalue = value < 256 ? value : 255;
8360 /* In EBCDIC [\x89-\x91] should include
8361 * the \x8e but [i-j] should not. */
8362 if (literal_endpoint == 2 &&
8363 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8364 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8366 if (isLOWER(prevvalue)) {
8367 for (i = prevvalue; i <= ceilvalue; i++)
8368 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8370 ANYOF_BITMAP_SET(ret, i);
8373 for (i = prevvalue; i <= ceilvalue; i++)
8374 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8376 ANYOF_BITMAP_SET(ret, i);
8382 for (i = prevvalue; i <= ceilvalue; i++) {
8383 if (!ANYOF_BITMAP_TEST(ret,i)) {
8385 ANYOF_BITMAP_SET(ret, i);
8389 if (value > 255 || UTF) {
8390 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8391 const UV natvalue = NATIVE_TO_UNI(value);
8392 stored+=2; /* can't optimize this class */
8393 ANYOF_FLAGS(ret) |= ANYOF_UNICODE;
8394 if (prevnatvalue < natvalue) { /* what about > ? */
8395 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8396 prevnatvalue, natvalue);
8398 else if (prevnatvalue == natvalue) {
8399 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8401 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8403 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8405 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8406 if (RExC_precomp[0] == ':' &&
8407 RExC_precomp[1] == '[' &&
8408 (f == 0xDF || f == 0x92)) {
8409 f = NATIVE_TO_UNI(f);
8412 /* If folding and foldable and a single
8413 * character, insert also the folded version
8414 * to the charclass. */
8416 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8417 if ((RExC_precomp[0] == ':' &&
8418 RExC_precomp[1] == '[' &&
8420 (value == 0xFB05 || value == 0xFB06))) ?
8421 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8422 foldlen == (STRLEN)UNISKIP(f) )
8424 if (foldlen == (STRLEN)UNISKIP(f))
8426 Perl_sv_catpvf(aTHX_ listsv,
8429 /* Any multicharacter foldings
8430 * require the following transform:
8431 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8432 * where E folds into "pq" and F folds
8433 * into "rst", all other characters
8434 * fold to single characters. We save
8435 * away these multicharacter foldings,
8436 * to be later saved as part of the
8437 * additional "s" data. */
8440 if (!unicode_alternate)
8441 unicode_alternate = newAV();
8442 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8444 av_push(unicode_alternate, sv);
8448 /* If folding and the value is one of the Greek
8449 * sigmas insert a few more sigmas to make the
8450 * folding rules of the sigmas to work right.
8451 * Note that not all the possible combinations
8452 * are handled here: some of them are handled
8453 * by the standard folding rules, and some of
8454 * them (literal or EXACTF cases) are handled
8455 * during runtime in regexec.c:S_find_byclass(). */
8456 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8457 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8458 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8459 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8460 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8462 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8463 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8464 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8469 literal_endpoint = 0;
8473 range = 0; /* this range (if it was one) is done now */
8477 ANYOF_FLAGS(ret) |= ANYOF_LARGE;
8479 RExC_size += ANYOF_CLASS_ADD_SKIP;
8481 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8487 /****** !SIZE_ONLY AFTER HERE *********/
8489 if( stored == 1 && (value < 128 || (value < 256 && !UTF))
8490 && !( ANYOF_FLAGS(ret) & ( ANYOF_FLAGS_ALL ^ ANYOF_FOLD ) )
8492 /* optimize single char class to an EXACT node
8493 but *only* when its not a UTF/high char */
8494 const char * cur_parse= RExC_parse;
8495 RExC_emit = (regnode *)orig_emit;
8496 RExC_parse = (char *)orig_parse;
8497 ret = reg_node(pRExC_state,
8498 (U8)((ANYOF_FLAGS(ret) & ANYOF_FOLD) ? EXACTF : EXACT));
8499 RExC_parse = (char *)cur_parse;
8500 *STRING(ret)= (char)value;
8502 RExC_emit += STR_SZ(1);
8503 SvREFCNT_dec(listsv);
8506 /* optimize case-insensitive simple patterns (e.g. /[a-z]/i) */
8507 if ( /* If the only flag is folding (plus possibly inversion). */
8508 ((ANYOF_FLAGS(ret) & (ANYOF_FLAGS_ALL ^ ANYOF_INVERT)) == ANYOF_FOLD)
8510 for (value = 0; value < 256; ++value) {
8511 if (ANYOF_BITMAP_TEST(ret, value)) {
8512 UV fold = PL_fold[value];
8515 ANYOF_BITMAP_SET(ret, fold);
8518 ANYOF_FLAGS(ret) &= ~ANYOF_FOLD;
8521 /* optimize inverted simple patterns (e.g. [^a-z]) */
8522 if (optimize_invert &&
8523 /* If the only flag is inversion. */
8524 (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8525 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8526 ANYOF_BITMAP(ret)[value] ^= ANYOF_FLAGS_ALL;
8527 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
8530 AV * const av = newAV();
8532 /* The 0th element stores the character class description
8533 * in its textual form: used later (regexec.c:Perl_regclass_swash())
8534 * to initialize the appropriate swash (which gets stored in
8535 * the 1st element), and also useful for dumping the regnode.
8536 * The 2nd element stores the multicharacter foldings,
8537 * used later (regexec.c:S_reginclass()). */
8538 av_store(av, 0, listsv);
8539 av_store(av, 1, NULL);
8540 av_store(av, 2, MUTABLE_SV(unicode_alternate));
8541 rv = newRV_noinc(MUTABLE_SV(av));
8542 n = add_data(pRExC_state, 1, "s");
8543 RExC_rxi->data->data[n] = (void*)rv;
8551 /* reg_skipcomment()
8553 Absorbs an /x style # comments from the input stream.
8554 Returns true if there is more text remaining in the stream.
8555 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
8556 terminates the pattern without including a newline.
8558 Note its the callers responsibility to ensure that we are
8564 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
8568 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
8570 while (RExC_parse < RExC_end)
8571 if (*RExC_parse++ == '\n') {
8576 /* we ran off the end of the pattern without ending
8577 the comment, so we have to add an \n when wrapping */
8578 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8586 Advance that parse position, and optionall absorbs
8587 "whitespace" from the inputstream.
8589 Without /x "whitespace" means (?#...) style comments only,
8590 with /x this means (?#...) and # comments and whitespace proper.
8592 Returns the RExC_parse point from BEFORE the scan occurs.
8594 This is the /x friendly way of saying RExC_parse++.
8598 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
8600 char* const retval = RExC_parse++;
8602 PERL_ARGS_ASSERT_NEXTCHAR;
8605 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
8606 RExC_parse[2] == '#') {
8607 while (*RExC_parse != ')') {
8608 if (RExC_parse == RExC_end)
8609 FAIL("Sequence (?#... not terminated");
8615 if (RExC_flags & RXf_PMf_EXTENDED) {
8616 if (isSPACE(*RExC_parse)) {
8620 else if (*RExC_parse == '#') {
8621 if ( reg_skipcomment( pRExC_state ) )
8630 - reg_node - emit a node
8632 STATIC regnode * /* Location. */
8633 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
8636 register regnode *ptr;
8637 regnode * const ret = RExC_emit;
8638 GET_RE_DEBUG_FLAGS_DECL;
8640 PERL_ARGS_ASSERT_REG_NODE;
8643 SIZE_ALIGN(RExC_size);
8647 if (RExC_emit >= RExC_emit_bound)
8648 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8650 NODE_ALIGN_FILL(ret);
8652 FILL_ADVANCE_NODE(ptr, op);
8653 #ifdef RE_TRACK_PATTERN_OFFSETS
8654 if (RExC_offsets) { /* MJD */
8655 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
8656 "reg_node", __LINE__,
8658 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
8659 ? "Overwriting end of array!\n" : "OK",
8660 (UV)(RExC_emit - RExC_emit_start),
8661 (UV)(RExC_parse - RExC_start),
8662 (UV)RExC_offsets[0]));
8663 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
8671 - reganode - emit a node with an argument
8673 STATIC regnode * /* Location. */
8674 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
8677 register regnode *ptr;
8678 regnode * const ret = RExC_emit;
8679 GET_RE_DEBUG_FLAGS_DECL;
8681 PERL_ARGS_ASSERT_REGANODE;
8684 SIZE_ALIGN(RExC_size);
8689 assert(2==regarglen[op]+1);
8691 Anything larger than this has to allocate the extra amount.
8692 If we changed this to be:
8694 RExC_size += (1 + regarglen[op]);
8696 then it wouldn't matter. Its not clear what side effect
8697 might come from that so its not done so far.
8702 if (RExC_emit >= RExC_emit_bound)
8703 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
8705 NODE_ALIGN_FILL(ret);
8707 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
8708 #ifdef RE_TRACK_PATTERN_OFFSETS
8709 if (RExC_offsets) { /* MJD */
8710 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8714 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
8715 "Overwriting end of array!\n" : "OK",
8716 (UV)(RExC_emit - RExC_emit_start),
8717 (UV)(RExC_parse - RExC_start),
8718 (UV)RExC_offsets[0]));
8719 Set_Cur_Node_Offset;
8727 - reguni - emit (if appropriate) a Unicode character
8730 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
8734 PERL_ARGS_ASSERT_REGUNI;
8736 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
8740 - reginsert - insert an operator in front of already-emitted operand
8742 * Means relocating the operand.
8745 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
8748 register regnode *src;
8749 register regnode *dst;
8750 register regnode *place;
8751 const int offset = regarglen[(U8)op];
8752 const int size = NODE_STEP_REGNODE + offset;
8753 GET_RE_DEBUG_FLAGS_DECL;
8755 PERL_ARGS_ASSERT_REGINSERT;
8756 PERL_UNUSED_ARG(depth);
8757 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
8758 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
8767 if (RExC_open_parens) {
8769 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
8770 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
8771 if ( RExC_open_parens[paren] >= opnd ) {
8772 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
8773 RExC_open_parens[paren] += size;
8775 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
8777 if ( RExC_close_parens[paren] >= opnd ) {
8778 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
8779 RExC_close_parens[paren] += size;
8781 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
8786 while (src > opnd) {
8787 StructCopy(--src, --dst, regnode);
8788 #ifdef RE_TRACK_PATTERN_OFFSETS
8789 if (RExC_offsets) { /* MJD 20010112 */
8790 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
8794 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
8795 ? "Overwriting end of array!\n" : "OK",
8796 (UV)(src - RExC_emit_start),
8797 (UV)(dst - RExC_emit_start),
8798 (UV)RExC_offsets[0]));
8799 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
8800 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
8806 place = opnd; /* Op node, where operand used to be. */
8807 #ifdef RE_TRACK_PATTERN_OFFSETS
8808 if (RExC_offsets) { /* MJD */
8809 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
8813 (UV)(place - RExC_emit_start) > RExC_offsets[0]
8814 ? "Overwriting end of array!\n" : "OK",
8815 (UV)(place - RExC_emit_start),
8816 (UV)(RExC_parse - RExC_start),
8817 (UV)RExC_offsets[0]));
8818 Set_Node_Offset(place, RExC_parse);
8819 Set_Node_Length(place, 1);
8822 src = NEXTOPER(place);
8823 FILL_ADVANCE_NODE(place, op);
8824 Zero(src, offset, regnode);
8828 - regtail - set the next-pointer at the end of a node chain of p to val.
8829 - SEE ALSO: regtail_study
8831 /* TODO: All three parms should be const */
8833 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8836 register regnode *scan;
8837 GET_RE_DEBUG_FLAGS_DECL;
8839 PERL_ARGS_ASSERT_REGTAIL;
8841 PERL_UNUSED_ARG(depth);
8847 /* Find last node. */
8850 regnode * const temp = regnext(scan);
8852 SV * const mysv=sv_newmortal();
8853 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
8854 regprop(RExC_rx, mysv, scan);
8855 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
8856 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
8857 (temp == NULL ? "->" : ""),
8858 (temp == NULL ? PL_reg_name[OP(val)] : "")
8866 if (reg_off_by_arg[OP(scan)]) {
8867 ARG_SET(scan, val - scan);
8870 NEXT_OFF(scan) = val - scan;
8876 - regtail_study - set the next-pointer at the end of a node chain of p to val.
8877 - Look for optimizable sequences at the same time.
8878 - currently only looks for EXACT chains.
8880 This is expermental code. The idea is to use this routine to perform
8881 in place optimizations on branches and groups as they are constructed,
8882 with the long term intention of removing optimization from study_chunk so
8883 that it is purely analytical.
8885 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
8886 to control which is which.
8889 /* TODO: All four parms should be const */
8892 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
8895 register regnode *scan;
8897 #ifdef EXPERIMENTAL_INPLACESCAN
8900 GET_RE_DEBUG_FLAGS_DECL;
8902 PERL_ARGS_ASSERT_REGTAIL_STUDY;
8908 /* Find last node. */
8912 regnode * const temp = regnext(scan);
8913 #ifdef EXPERIMENTAL_INPLACESCAN
8914 if (PL_regkind[OP(scan)] == EXACT)
8915 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
8923 if( exact == PSEUDO )
8925 else if ( exact != OP(scan) )
8934 SV * const mysv=sv_newmortal();
8935 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
8936 regprop(RExC_rx, mysv, scan);
8937 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
8938 SvPV_nolen_const(mysv),
8940 PL_reg_name[exact]);
8947 SV * const mysv_val=sv_newmortal();
8948 DEBUG_PARSE_MSG("");
8949 regprop(RExC_rx, mysv_val, val);
8950 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
8951 SvPV_nolen_const(mysv_val),
8952 (IV)REG_NODE_NUM(val),
8956 if (reg_off_by_arg[OP(scan)]) {
8957 ARG_SET(scan, val - scan);
8960 NEXT_OFF(scan) = val - scan;
8968 - regcurly - a little FSA that accepts {\d+,?\d*}
8970 #ifndef PERL_IN_XSUB_RE
8972 Perl_regcurly(register const char *s)
8974 PERL_ARGS_ASSERT_REGCURLY;
8993 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
8997 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
9002 for (bit=0; bit<32; bit++) {
9003 if (flags & (1<<bit)) {
9005 PerlIO_printf(Perl_debug_log, "%s",lead);
9006 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
9011 PerlIO_printf(Perl_debug_log, "\n");
9013 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
9019 Perl_regdump(pTHX_ const regexp *r)
9023 SV * const sv = sv_newmortal();
9024 SV *dsv= sv_newmortal();
9026 GET_RE_DEBUG_FLAGS_DECL;
9028 PERL_ARGS_ASSERT_REGDUMP;
9030 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
9032 /* Header fields of interest. */
9033 if (r->anchored_substr) {
9034 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
9035 RE_SV_DUMPLEN(r->anchored_substr), 30);
9036 PerlIO_printf(Perl_debug_log,
9037 "anchored %s%s at %"IVdf" ",
9038 s, RE_SV_TAIL(r->anchored_substr),
9039 (IV)r->anchored_offset);
9040 } else if (r->anchored_utf8) {
9041 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
9042 RE_SV_DUMPLEN(r->anchored_utf8), 30);
9043 PerlIO_printf(Perl_debug_log,
9044 "anchored utf8 %s%s at %"IVdf" ",
9045 s, RE_SV_TAIL(r->anchored_utf8),
9046 (IV)r->anchored_offset);
9048 if (r->float_substr) {
9049 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
9050 RE_SV_DUMPLEN(r->float_substr), 30);
9051 PerlIO_printf(Perl_debug_log,
9052 "floating %s%s at %"IVdf"..%"UVuf" ",
9053 s, RE_SV_TAIL(r->float_substr),
9054 (IV)r->float_min_offset, (UV)r->float_max_offset);
9055 } else if (r->float_utf8) {
9056 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
9057 RE_SV_DUMPLEN(r->float_utf8), 30);
9058 PerlIO_printf(Perl_debug_log,
9059 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
9060 s, RE_SV_TAIL(r->float_utf8),
9061 (IV)r->float_min_offset, (UV)r->float_max_offset);
9063 if (r->check_substr || r->check_utf8)
9064 PerlIO_printf(Perl_debug_log,
9066 (r->check_substr == r->float_substr
9067 && r->check_utf8 == r->float_utf8
9068 ? "(checking floating" : "(checking anchored"));
9069 if (r->extflags & RXf_NOSCAN)
9070 PerlIO_printf(Perl_debug_log, " noscan");
9071 if (r->extflags & RXf_CHECK_ALL)
9072 PerlIO_printf(Perl_debug_log, " isall");
9073 if (r->check_substr || r->check_utf8)
9074 PerlIO_printf(Perl_debug_log, ") ");
9076 if (ri->regstclass) {
9077 regprop(r, sv, ri->regstclass);
9078 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
9080 if (r->extflags & RXf_ANCH) {
9081 PerlIO_printf(Perl_debug_log, "anchored");
9082 if (r->extflags & RXf_ANCH_BOL)
9083 PerlIO_printf(Perl_debug_log, "(BOL)");
9084 if (r->extflags & RXf_ANCH_MBOL)
9085 PerlIO_printf(Perl_debug_log, "(MBOL)");
9086 if (r->extflags & RXf_ANCH_SBOL)
9087 PerlIO_printf(Perl_debug_log, "(SBOL)");
9088 if (r->extflags & RXf_ANCH_GPOS)
9089 PerlIO_printf(Perl_debug_log, "(GPOS)");
9090 PerlIO_putc(Perl_debug_log, ' ');
9092 if (r->extflags & RXf_GPOS_SEEN)
9093 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
9094 if (r->intflags & PREGf_SKIP)
9095 PerlIO_printf(Perl_debug_log, "plus ");
9096 if (r->intflags & PREGf_IMPLICIT)
9097 PerlIO_printf(Perl_debug_log, "implicit ");
9098 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
9099 if (r->extflags & RXf_EVAL_SEEN)
9100 PerlIO_printf(Perl_debug_log, "with eval ");
9101 PerlIO_printf(Perl_debug_log, "\n");
9102 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
9104 PERL_ARGS_ASSERT_REGDUMP;
9105 PERL_UNUSED_CONTEXT;
9107 #endif /* DEBUGGING */
9111 - regprop - printable representation of opcode
9113 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9116 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9117 if (flags & ANYOF_INVERT) \
9118 /*make sure the invert info is in each */ \
9119 sv_catpvs(sv, "^"); \
9125 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9130 RXi_GET_DECL(prog,progi);
9131 GET_RE_DEBUG_FLAGS_DECL;
9133 PERL_ARGS_ASSERT_REGPROP;
9137 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9138 /* It would be nice to FAIL() here, but this may be called from
9139 regexec.c, and it would be hard to supply pRExC_state. */
9140 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9141 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9143 k = PL_regkind[OP(o)];
9147 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9148 * is a crude hack but it may be the best for now since
9149 * we have no flag "this EXACTish node was UTF-8"
9151 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9152 PERL_PV_ESCAPE_UNI_DETECT |
9153 PERL_PV_PRETTY_ELLIPSES |
9154 PERL_PV_PRETTY_LTGT |
9155 PERL_PV_PRETTY_NOCLEAR
9157 } else if (k == TRIE) {
9158 /* print the details of the trie in dumpuntil instead, as
9159 * progi->data isn't available here */
9160 const char op = OP(o);
9161 const U32 n = ARG(o);
9162 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9163 (reg_ac_data *)progi->data->data[n] :
9165 const reg_trie_data * const trie
9166 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9168 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9169 DEBUG_TRIE_COMPILE_r(
9170 Perl_sv_catpvf(aTHX_ sv,
9171 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9172 (UV)trie->startstate,
9173 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9174 (UV)trie->wordcount,
9177 (UV)TRIE_CHARCOUNT(trie),
9178 (UV)trie->uniquecharcount
9181 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9183 int rangestart = -1;
9184 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9186 for (i = 0; i <= 256; i++) {
9187 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9188 if (rangestart == -1)
9190 } else if (rangestart != -1) {
9191 if (i <= rangestart + 3)
9192 for (; rangestart < i; rangestart++)
9193 put_byte(sv, rangestart);
9195 put_byte(sv, rangestart);
9197 put_byte(sv, i - 1);
9205 } else if (k == CURLY) {
9206 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9207 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9208 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9210 else if (k == WHILEM && o->flags) /* Ordinal/of */
9211 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9212 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9213 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9214 if ( RXp_PAREN_NAMES(prog) ) {
9215 if ( k != REF || OP(o) < NREF) {
9216 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9217 SV **name= av_fetch(list, ARG(o), 0 );
9219 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9222 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9223 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9224 I32 *nums=(I32*)SvPVX(sv_dat);
9225 SV **name= av_fetch(list, nums[0], 0 );
9228 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9229 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9230 (n ? "," : ""), (IV)nums[n]);
9232 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9236 } else if (k == GOSUB)
9237 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9238 else if (k == VERB) {
9240 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9241 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9242 } else if (k == LOGICAL)
9243 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9244 else if (k == FOLDCHAR)
9245 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9246 else if (k == ANYOF) {
9247 int i, rangestart = -1;
9248 const U8 flags = ANYOF_FLAGS(o);
9251 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9252 static const char * const anyofs[] = {
9285 if (flags & ANYOF_LOCALE)
9286 sv_catpvs(sv, "{loc}");
9287 if (flags & ANYOF_FOLD)
9288 sv_catpvs(sv, "{i}");
9289 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9290 if (flags & ANYOF_INVERT)
9293 /* output what the standard cp 0-255 bitmap matches */
9294 for (i = 0; i <= 256; i++) {
9295 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9296 if (rangestart == -1)
9298 } else if (rangestart != -1) {
9299 if (i <= rangestart + 3)
9300 for (; rangestart < i; rangestart++)
9301 put_byte(sv, rangestart);
9303 put_byte(sv, rangestart);
9305 put_byte(sv, i - 1);
9312 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9313 /* output any special charclass tests (used mostly under use locale) */
9314 if (o->flags & ANYOF_CLASS)
9315 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9316 if (ANYOF_CLASS_TEST(o,i)) {
9317 sv_catpv(sv, anyofs[i]);
9321 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9323 /* output information about the unicode matching */
9324 if (flags & ANYOF_UNICODE)
9325 sv_catpvs(sv, "{unicode}");
9326 else if (flags & ANYOF_UNICODE_ALL)
9327 sv_catpvs(sv, "{unicode_all}");
9331 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9335 U8 s[UTF8_MAXBYTES_CASE+1];
9337 for (i = 0; i <= 256; i++) { /* just the first 256 */
9338 uvchr_to_utf8(s, i);
9340 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9341 if (rangestart == -1)
9343 } else if (rangestart != -1) {
9344 if (i <= rangestart + 3)
9345 for (; rangestart < i; rangestart++) {
9346 const U8 * const e = uvchr_to_utf8(s,rangestart);
9348 for(p = s; p < e; p++)
9352 const U8 *e = uvchr_to_utf8(s,rangestart);
9354 for (p = s; p < e; p++)
9357 e = uvchr_to_utf8(s, i-1);
9358 for (p = s; p < e; p++)
9365 sv_catpvs(sv, "..."); /* et cetera */
9369 char *s = savesvpv(lv);
9370 char * const origs = s;
9372 while (*s && *s != '\n')
9376 const char * const t = ++s;
9394 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9396 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9397 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9399 PERL_UNUSED_CONTEXT;
9400 PERL_UNUSED_ARG(sv);
9402 PERL_UNUSED_ARG(prog);
9403 #endif /* DEBUGGING */
9407 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9408 { /* Assume that RE_INTUIT is set */
9410 struct regexp *const prog = (struct regexp *)SvANY(r);
9411 GET_RE_DEBUG_FLAGS_DECL;
9413 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9414 PERL_UNUSED_CONTEXT;
9418 const char * const s = SvPV_nolen_const(prog->check_substr
9419 ? prog->check_substr : prog->check_utf8);
9421 if (!PL_colorset) reginitcolors();
9422 PerlIO_printf(Perl_debug_log,
9423 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9425 prog->check_substr ? "" : "utf8 ",
9426 PL_colors[5],PL_colors[0],
9429 (strlen(s) > 60 ? "..." : ""));
9432 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9438 handles refcounting and freeing the perl core regexp structure. When
9439 it is necessary to actually free the structure the first thing it
9440 does is call the 'free' method of the regexp_engine associated to to
9441 the regexp, allowing the handling of the void *pprivate; member
9442 first. (This routine is not overridable by extensions, which is why
9443 the extensions free is called first.)
9445 See regdupe and regdupe_internal if you change anything here.
9447 #ifndef PERL_IN_XSUB_RE
9449 Perl_pregfree(pTHX_ REGEXP *r)
9455 Perl_pregfree2(pTHX_ REGEXP *rx)
9458 struct regexp *const r = (struct regexp *)SvANY(rx);
9459 GET_RE_DEBUG_FLAGS_DECL;
9461 PERL_ARGS_ASSERT_PREGFREE2;
9464 ReREFCNT_dec(r->mother_re);
9466 CALLREGFREE_PVT(rx); /* free the private data */
9467 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9470 SvREFCNT_dec(r->anchored_substr);
9471 SvREFCNT_dec(r->anchored_utf8);
9472 SvREFCNT_dec(r->float_substr);
9473 SvREFCNT_dec(r->float_utf8);
9474 Safefree(r->substrs);
9476 RX_MATCH_COPY_FREE(rx);
9477 #ifdef PERL_OLD_COPY_ON_WRITE
9478 SvREFCNT_dec(r->saved_copy);
9485 This is a hacky workaround to the structural issue of match results
9486 being stored in the regexp structure which is in turn stored in
9487 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9488 could be PL_curpm in multiple contexts, and could require multiple
9489 result sets being associated with the pattern simultaneously, such
9490 as when doing a recursive match with (??{$qr})
9492 The solution is to make a lightweight copy of the regexp structure
9493 when a qr// is returned from the code executed by (??{$qr}) this
9494 lightweight copy doesnt actually own any of its data except for
9495 the starp/end and the actual regexp structure itself.
9501 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
9504 struct regexp *const r = (struct regexp *)SvANY(rx);
9505 register const I32 npar = r->nparens+1;
9507 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9510 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9511 ret = (struct regexp *)SvANY(ret_x);
9513 (void)ReREFCNT_inc(rx);
9514 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9515 by pointing directly at the buffer, but flagging that the allocated
9516 space in the copy is zero. As we've just done a struct copy, it's now
9517 a case of zero-ing that, rather than copying the current length. */
9518 SvPV_set(ret_x, RX_WRAPPED(rx));
9519 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9520 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
9521 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
9522 SvLEN_set(ret_x, 0);
9523 SvSTASH_set(ret_x, NULL);
9524 SvMAGIC_set(ret_x, NULL);
9525 Newx(ret->offs, npar, regexp_paren_pair);
9526 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9528 Newx(ret->substrs, 1, struct reg_substr_data);
9529 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9531 SvREFCNT_inc_void(ret->anchored_substr);
9532 SvREFCNT_inc_void(ret->anchored_utf8);
9533 SvREFCNT_inc_void(ret->float_substr);
9534 SvREFCNT_inc_void(ret->float_utf8);
9536 /* check_substr and check_utf8, if non-NULL, point to either their
9537 anchored or float namesakes, and don't hold a second reference. */
9539 RX_MATCH_COPIED_off(ret_x);
9540 #ifdef PERL_OLD_COPY_ON_WRITE
9541 ret->saved_copy = NULL;
9543 ret->mother_re = rx;
9549 /* regfree_internal()
9551 Free the private data in a regexp. This is overloadable by
9552 extensions. Perl takes care of the regexp structure in pregfree(),
9553 this covers the *pprivate pointer which technically perldoesnt
9554 know about, however of course we have to handle the
9555 regexp_internal structure when no extension is in use.
9557 Note this is called before freeing anything in the regexp
9562 Perl_regfree_internal(pTHX_ REGEXP * const rx)
9565 struct regexp *const r = (struct regexp *)SvANY(rx);
9567 GET_RE_DEBUG_FLAGS_DECL;
9569 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
9575 SV *dsv= sv_newmortal();
9576 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
9577 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
9578 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
9579 PL_colors[4],PL_colors[5],s);
9582 #ifdef RE_TRACK_PATTERN_OFFSETS
9584 Safefree(ri->u.offsets); /* 20010421 MJD */
9587 int n = ri->data->count;
9588 PAD* new_comppad = NULL;
9593 /* If you add a ->what type here, update the comment in regcomp.h */
9594 switch (ri->data->what[n]) {
9599 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
9602 Safefree(ri->data->data[n]);
9605 new_comppad = MUTABLE_AV(ri->data->data[n]);
9608 if (new_comppad == NULL)
9609 Perl_croak(aTHX_ "panic: pregfree comppad");
9610 PAD_SAVE_LOCAL(old_comppad,
9611 /* Watch out for global destruction's random ordering. */
9612 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
9615 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
9618 op_free((OP_4tree*)ri->data->data[n]);
9620 PAD_RESTORE_LOCAL(old_comppad);
9621 SvREFCNT_dec(MUTABLE_SV(new_comppad));
9627 { /* Aho Corasick add-on structure for a trie node.
9628 Used in stclass optimization only */
9630 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
9632 refcount = --aho->refcount;
9635 PerlMemShared_free(aho->states);
9636 PerlMemShared_free(aho->fail);
9637 /* do this last!!!! */
9638 PerlMemShared_free(ri->data->data[n]);
9639 PerlMemShared_free(ri->regstclass);
9645 /* trie structure. */
9647 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
9649 refcount = --trie->refcount;
9652 PerlMemShared_free(trie->charmap);
9653 PerlMemShared_free(trie->states);
9654 PerlMemShared_free(trie->trans);
9656 PerlMemShared_free(trie->bitmap);
9658 PerlMemShared_free(trie->jump);
9659 PerlMemShared_free(trie->wordinfo);
9660 /* do this last!!!! */
9661 PerlMemShared_free(ri->data->data[n]);
9666 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
9669 Safefree(ri->data->what);
9676 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
9677 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
9678 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
9681 re_dup - duplicate a regexp.
9683 This routine is expected to clone a given regexp structure. It is only
9684 compiled under USE_ITHREADS.
9686 After all of the core data stored in struct regexp is duplicated
9687 the regexp_engine.dupe method is used to copy any private data
9688 stored in the *pprivate pointer. This allows extensions to handle
9689 any duplication it needs to do.
9691 See pregfree() and regfree_internal() if you change anything here.
9693 #if defined(USE_ITHREADS)
9694 #ifndef PERL_IN_XSUB_RE
9696 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
9700 const struct regexp *r = (const struct regexp *)SvANY(sstr);
9701 struct regexp *ret = (struct regexp *)SvANY(dstr);
9703 PERL_ARGS_ASSERT_RE_DUP_GUTS;
9705 npar = r->nparens+1;
9706 Newx(ret->offs, npar, regexp_paren_pair);
9707 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9709 /* no need to copy these */
9710 Newx(ret->swap, npar, regexp_paren_pair);
9714 /* Do it this way to avoid reading from *r after the StructCopy().
9715 That way, if any of the sv_dup_inc()s dislodge *r from the L1
9716 cache, it doesn't matter. */
9717 const bool anchored = r->check_substr
9718 ? r->check_substr == r->anchored_substr
9719 : r->check_utf8 == r->anchored_utf8;
9720 Newx(ret->substrs, 1, struct reg_substr_data);
9721 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9723 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
9724 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
9725 ret->float_substr = sv_dup_inc(ret->float_substr, param);
9726 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
9728 /* check_substr and check_utf8, if non-NULL, point to either their
9729 anchored or float namesakes, and don't hold a second reference. */
9731 if (ret->check_substr) {
9733 assert(r->check_utf8 == r->anchored_utf8);
9734 ret->check_substr = ret->anchored_substr;
9735 ret->check_utf8 = ret->anchored_utf8;
9737 assert(r->check_substr == r->float_substr);
9738 assert(r->check_utf8 == r->float_utf8);
9739 ret->check_substr = ret->float_substr;
9740 ret->check_utf8 = ret->float_utf8;
9742 } else if (ret->check_utf8) {
9744 ret->check_utf8 = ret->anchored_utf8;
9746 ret->check_utf8 = ret->float_utf8;
9751 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
9754 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
9756 if (RX_MATCH_COPIED(dstr))
9757 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
9760 #ifdef PERL_OLD_COPY_ON_WRITE
9761 ret->saved_copy = NULL;
9764 if (ret->mother_re) {
9765 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
9766 /* Our storage points directly to our mother regexp, but that's
9767 1: a buffer in a different thread
9768 2: something we no longer hold a reference on
9769 so we need to copy it locally. */
9770 /* Note we need to sue SvCUR() on our mother_re, because it, in
9771 turn, may well be pointing to its own mother_re. */
9772 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
9773 SvCUR(ret->mother_re)+1));
9774 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
9776 ret->mother_re = NULL;
9780 #endif /* PERL_IN_XSUB_RE */
9785 This is the internal complement to regdupe() which is used to copy
9786 the structure pointed to by the *pprivate pointer in the regexp.
9787 This is the core version of the extension overridable cloning hook.
9788 The regexp structure being duplicated will be copied by perl prior
9789 to this and will be provided as the regexp *r argument, however
9790 with the /old/ structures pprivate pointer value. Thus this routine
9791 may override any copying normally done by perl.
9793 It returns a pointer to the new regexp_internal structure.
9797 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
9800 struct regexp *const r = (struct regexp *)SvANY(rx);
9801 regexp_internal *reti;
9805 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
9807 npar = r->nparens+1;
9810 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
9811 Copy(ri->program, reti->program, len+1, regnode);
9814 reti->regstclass = NULL;
9818 const int count = ri->data->count;
9821 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
9822 char, struct reg_data);
9823 Newx(d->what, count, U8);
9826 for (i = 0; i < count; i++) {
9827 d->what[i] = ri->data->what[i];
9828 switch (d->what[i]) {
9829 /* legal options are one of: sSfpontTua
9830 see also regcomp.h and pregfree() */
9831 case 'a': /* actually an AV, but the dup function is identical. */
9834 case 'p': /* actually an AV, but the dup function is identical. */
9835 case 'u': /* actually an HV, but the dup function is identical. */
9836 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
9839 /* This is cheating. */
9840 Newx(d->data[i], 1, struct regnode_charclass_class);
9841 StructCopy(ri->data->data[i], d->data[i],
9842 struct regnode_charclass_class);
9843 reti->regstclass = (regnode*)d->data[i];
9846 /* Compiled op trees are readonly and in shared memory,
9847 and can thus be shared without duplication. */
9849 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
9853 /* Trie stclasses are readonly and can thus be shared
9854 * without duplication. We free the stclass in pregfree
9855 * when the corresponding reg_ac_data struct is freed.
9857 reti->regstclass= ri->regstclass;
9861 ((reg_trie_data*)ri->data->data[i])->refcount++;
9865 d->data[i] = ri->data->data[i];
9868 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
9877 reti->name_list_idx = ri->name_list_idx;
9879 #ifdef RE_TRACK_PATTERN_OFFSETS
9880 if (ri->u.offsets) {
9881 Newx(reti->u.offsets, 2*len+1, U32);
9882 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
9885 SetProgLen(reti,len);
9891 #endif /* USE_ITHREADS */
9893 #ifndef PERL_IN_XSUB_RE
9896 - regnext - dig the "next" pointer out of a node
9899 Perl_regnext(pTHX_ register regnode *p)
9902 register I32 offset;
9907 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
9908 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
9911 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
9920 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
9923 STRLEN l1 = strlen(pat1);
9924 STRLEN l2 = strlen(pat2);
9927 const char *message;
9929 PERL_ARGS_ASSERT_RE_CROAK2;
9935 Copy(pat1, buf, l1 , char);
9936 Copy(pat2, buf + l1, l2 , char);
9937 buf[l1 + l2] = '\n';
9938 buf[l1 + l2 + 1] = '\0';
9940 /* ANSI variant takes additional second argument */
9941 va_start(args, pat2);
9945 msv = vmess(buf, &args);
9947 message = SvPV_const(msv,l1);
9950 Copy(message, buf, l1 , char);
9951 buf[l1-1] = '\0'; /* Overwrite \n */
9952 Perl_croak(aTHX_ "%s", buf);
9955 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
9957 #ifndef PERL_IN_XSUB_RE
9959 Perl_save_re_context(pTHX)
9963 struct re_save_state *state;
9965 SAVEVPTR(PL_curcop);
9966 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
9968 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
9969 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
9970 SSPUSHUV(SAVEt_RE_STATE);
9972 Copy(&PL_reg_state, state, 1, struct re_save_state);
9974 PL_reg_start_tmp = 0;
9975 PL_reg_start_tmpl = 0;
9976 PL_reg_oldsaved = NULL;
9977 PL_reg_oldsavedlen = 0;
9979 PL_reg_leftiter = 0;
9980 PL_reg_poscache = NULL;
9981 PL_reg_poscache_size = 0;
9982 #ifdef PERL_OLD_COPY_ON_WRITE
9986 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
9988 const REGEXP * const rx = PM_GETRE(PL_curpm);
9991 for (i = 1; i <= RX_NPARENS(rx); i++) {
9992 char digits[TYPE_CHARS(long)];
9993 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
9994 GV *const *const gvp
9995 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
9998 GV * const gv = *gvp;
9999 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
10009 clear_re(pTHX_ void *r)
10012 ReREFCNT_dec((REGEXP *)r);
10018 S_put_byte(pTHX_ SV *sv, int c)
10020 PERL_ARGS_ASSERT_PUT_BYTE;
10022 /* Our definition of isPRINT() ignores locales, so only bytes that are
10023 not part of UTF-8 are considered printable. I assume that the same
10024 holds for UTF-EBCDIC.
10025 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
10026 which Wikipedia says:
10028 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
10029 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
10030 identical, to the ASCII delete (DEL) or rubout control character.
10031 ) So the old condition can be simplified to !isPRINT(c) */
10033 Perl_sv_catpvf(aTHX_ sv, "\\%o", c);
10035 const char string = c;
10036 if (c == '-' || c == ']' || c == '\\' || c == '^')
10037 sv_catpvs(sv, "\\");
10038 sv_catpvn(sv, &string, 1);
10043 #define CLEAR_OPTSTART \
10044 if (optstart) STMT_START { \
10045 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
10049 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
10051 STATIC const regnode *
10052 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
10053 const regnode *last, const regnode *plast,
10054 SV* sv, I32 indent, U32 depth)
10057 register U8 op = PSEUDO; /* Arbitrary non-END op. */
10058 register const regnode *next;
10059 const regnode *optstart= NULL;
10061 RXi_GET_DECL(r,ri);
10062 GET_RE_DEBUG_FLAGS_DECL;
10064 PERL_ARGS_ASSERT_DUMPUNTIL;
10066 #ifdef DEBUG_DUMPUNTIL
10067 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
10068 last ? last-start : 0,plast ? plast-start : 0);
10071 if (plast && plast < last)
10074 while (PL_regkind[op] != END && (!last || node < last)) {
10075 /* While that wasn't END last time... */
10078 if (op == CLOSE || op == WHILEM)
10080 next = regnext((regnode *)node);
10083 if (OP(node) == OPTIMIZED) {
10084 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10091 regprop(r, sv, node);
10092 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10093 (int)(2*indent + 1), "", SvPVX_const(sv));
10095 if (OP(node) != OPTIMIZED) {
10096 if (next == NULL) /* Next ptr. */
10097 PerlIO_printf(Perl_debug_log, " (0)");
10098 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10099 PerlIO_printf(Perl_debug_log, " (FAIL)");
10101 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10102 (void)PerlIO_putc(Perl_debug_log, '\n');
10106 if (PL_regkind[(U8)op] == BRANCHJ) {
10109 register const regnode *nnode = (OP(next) == LONGJMP
10110 ? regnext((regnode *)next)
10112 if (last && nnode > last)
10114 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10117 else if (PL_regkind[(U8)op] == BRANCH) {
10119 DUMPUNTIL(NEXTOPER(node), next);
10121 else if ( PL_regkind[(U8)op] == TRIE ) {
10122 const regnode *this_trie = node;
10123 const char op = OP(node);
10124 const U32 n = ARG(node);
10125 const reg_ac_data * const ac = op>=AHOCORASICK ?
10126 (reg_ac_data *)ri->data->data[n] :
10128 const reg_trie_data * const trie =
10129 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10131 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10133 const regnode *nextbranch= NULL;
10136 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10137 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10139 PerlIO_printf(Perl_debug_log, "%*s%s ",
10140 (int)(2*(indent+3)), "",
10141 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10142 PL_colors[0], PL_colors[1],
10143 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10144 PERL_PV_PRETTY_ELLIPSES |
10145 PERL_PV_PRETTY_LTGT
10150 U16 dist= trie->jump[word_idx+1];
10151 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10152 (UV)((dist ? this_trie + dist : next) - start));
10155 nextbranch= this_trie + trie->jump[0];
10156 DUMPUNTIL(this_trie + dist, nextbranch);
10158 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10159 nextbranch= regnext((regnode *)nextbranch);
10161 PerlIO_printf(Perl_debug_log, "\n");
10164 if (last && next > last)
10169 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10170 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10171 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10173 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10175 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10177 else if ( op == PLUS || op == STAR) {
10178 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10180 else if (op == ANYOF) {
10181 /* arglen 1 + class block */
10182 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_LARGE)
10183 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10184 node = NEXTOPER(node);
10186 else if (PL_regkind[(U8)op] == EXACT) {
10187 /* Literal string, where present. */
10188 node += NODE_SZ_STR(node) - 1;
10189 node = NEXTOPER(node);
10192 node = NEXTOPER(node);
10193 node += regarglen[(U8)op];
10195 if (op == CURLYX || op == OPEN)
10199 #ifdef DEBUG_DUMPUNTIL
10200 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10205 #endif /* DEBUGGING */
10209 * c-indentation-style: bsd
10210 * c-basic-offset: 4
10211 * indent-tabs-mode: t
10214 * ex: set ts=8 sts=4 sw=4 noet: