5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
88 #include "dquote_static.c"
95 # if defined(BUGGY_MSC6)
96 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
97 # pragma optimize("a",off)
98 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
99 # pragma optimize("w",on )
100 # endif /* BUGGY_MSC6 */
104 #define STATIC static
107 typedef struct RExC_state_t {
108 U32 flags; /* are we folding, multilining? */
109 char *precomp; /* uncompiled string. */
110 REGEXP *rx_sv; /* The SV that is the regexp. */
111 regexp *rx; /* perl core regexp structure */
112 regexp_internal *rxi; /* internal data for regexp object pprivate field */
113 char *start; /* Start of input for compile */
114 char *end; /* End of input for compile */
115 char *parse; /* Input-scan pointer. */
116 I32 whilem_seen; /* number of WHILEM in this expr */
117 regnode *emit_start; /* Start of emitted-code area */
118 regnode *emit_bound; /* First regnode outside of the allocated space */
119 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
120 I32 naughty; /* How bad is this pattern? */
121 I32 sawback; /* Did we see \1, ...? */
123 I32 size; /* Code size. */
124 I32 npar; /* Capture buffer count, (OPEN). */
125 I32 cpar; /* Capture buffer count, (CLOSE). */
126 I32 nestroot; /* root parens we are in - used by accept */
130 regnode **open_parens; /* pointers to open parens */
131 regnode **close_parens; /* pointers to close parens */
132 regnode *opend; /* END node in program */
133 I32 utf8; /* whether the pattern is utf8 or not */
134 I32 orig_utf8; /* whether the pattern was originally in utf8 */
135 /* XXX use this for future optimisation of case
136 * where pattern must be upgraded to utf8. */
137 HV *paren_names; /* Paren names */
139 regnode **recurse; /* Recurse regops */
140 I32 recurse_count; /* Number of recurse regops */
142 char *starttry; /* -Dr: where regtry was called. */
143 #define RExC_starttry (pRExC_state->starttry)
146 const char *lastparse;
148 AV *paren_name_list; /* idx -> name */
149 #define RExC_lastparse (pRExC_state->lastparse)
150 #define RExC_lastnum (pRExC_state->lastnum)
151 #define RExC_paren_name_list (pRExC_state->paren_name_list)
155 #define RExC_flags (pRExC_state->flags)
156 #define RExC_precomp (pRExC_state->precomp)
157 #define RExC_rx_sv (pRExC_state->rx_sv)
158 #define RExC_rx (pRExC_state->rx)
159 #define RExC_rxi (pRExC_state->rxi)
160 #define RExC_start (pRExC_state->start)
161 #define RExC_end (pRExC_state->end)
162 #define RExC_parse (pRExC_state->parse)
163 #define RExC_whilem_seen (pRExC_state->whilem_seen)
164 #ifdef RE_TRACK_PATTERN_OFFSETS
165 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
167 #define RExC_emit (pRExC_state->emit)
168 #define RExC_emit_start (pRExC_state->emit_start)
169 #define RExC_emit_bound (pRExC_state->emit_bound)
170 #define RExC_naughty (pRExC_state->naughty)
171 #define RExC_sawback (pRExC_state->sawback)
172 #define RExC_seen (pRExC_state->seen)
173 #define RExC_size (pRExC_state->size)
174 #define RExC_npar (pRExC_state->npar)
175 #define RExC_nestroot (pRExC_state->nestroot)
176 #define RExC_extralen (pRExC_state->extralen)
177 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
178 #define RExC_seen_evals (pRExC_state->seen_evals)
179 #define RExC_utf8 (pRExC_state->utf8)
180 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
181 #define RExC_open_parens (pRExC_state->open_parens)
182 #define RExC_close_parens (pRExC_state->close_parens)
183 #define RExC_opend (pRExC_state->opend)
184 #define RExC_paren_names (pRExC_state->paren_names)
185 #define RExC_recurse (pRExC_state->recurse)
186 #define RExC_recurse_count (pRExC_state->recurse_count)
189 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
190 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
191 ((*s) == '{' && regcurly(s)))
194 #undef SPSTART /* dratted cpp namespace... */
197 * Flags to be passed up and down.
199 #define WORST 0 /* Worst case. */
200 #define HASWIDTH 0x01 /* Known to match non-null strings. */
202 /* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single
203 * character, and if utf8, must be invariant. */
205 #define SPSTART 0x04 /* Starts with * or +. */
206 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
207 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
209 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
211 /* whether trie related optimizations are enabled */
212 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
213 #define TRIE_STUDY_OPT
214 #define FULL_TRIE_STUDY
220 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
221 #define PBITVAL(paren) (1 << ((paren) & 7))
222 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
223 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
224 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
226 /* If not already in utf8, do a longjmp back to the beginning */
227 #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */
228 #define REQUIRE_UTF8 STMT_START { \
229 if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \
232 /* About scan_data_t.
234 During optimisation we recurse through the regexp program performing
235 various inplace (keyhole style) optimisations. In addition study_chunk
236 and scan_commit populate this data structure with information about
237 what strings MUST appear in the pattern. We look for the longest
238 string that must appear at a fixed location, and we look for the
239 longest string that may appear at a floating location. So for instance
244 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
245 strings (because they follow a .* construct). study_chunk will identify
246 both FOO and BAR as being the longest fixed and floating strings respectively.
248 The strings can be composites, for instance
252 will result in a composite fixed substring 'foo'.
254 For each string some basic information is maintained:
256 - offset or min_offset
257 This is the position the string must appear at, or not before.
258 It also implicitly (when combined with minlenp) tells us how many
259 characters must match before the string we are searching for.
260 Likewise when combined with minlenp and the length of the string it
261 tells us how many characters must appear after the string we have
265 Only used for floating strings. This is the rightmost point that
266 the string can appear at. If set to I32 max it indicates that the
267 string can occur infinitely far to the right.
270 A pointer to the minimum length of the pattern that the string
271 was found inside. This is important as in the case of positive
272 lookahead or positive lookbehind we can have multiple patterns
277 The minimum length of the pattern overall is 3, the minimum length
278 of the lookahead part is 3, but the minimum length of the part that
279 will actually match is 1. So 'FOO's minimum length is 3, but the
280 minimum length for the F is 1. This is important as the minimum length
281 is used to determine offsets in front of and behind the string being
282 looked for. Since strings can be composites this is the length of the
283 pattern at the time it was commited with a scan_commit. Note that
284 the length is calculated by study_chunk, so that the minimum lengths
285 are not known until the full pattern has been compiled, thus the
286 pointer to the value.
290 In the case of lookbehind the string being searched for can be
291 offset past the start point of the final matching string.
292 If this value was just blithely removed from the min_offset it would
293 invalidate some of the calculations for how many chars must match
294 before or after (as they are derived from min_offset and minlen and
295 the length of the string being searched for).
296 When the final pattern is compiled and the data is moved from the
297 scan_data_t structure into the regexp structure the information
298 about lookbehind is factored in, with the information that would
299 have been lost precalculated in the end_shift field for the
302 The fields pos_min and pos_delta are used to store the minimum offset
303 and the delta to the maximum offset at the current point in the pattern.
307 typedef struct scan_data_t {
308 /*I32 len_min; unused */
309 /*I32 len_delta; unused */
313 I32 last_end; /* min value, <0 unless valid. */
316 SV **longest; /* Either &l_fixed, or &l_float. */
317 SV *longest_fixed; /* longest fixed string found in pattern */
318 I32 offset_fixed; /* offset where it starts */
319 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
320 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
321 SV *longest_float; /* longest floating string found in pattern */
322 I32 offset_float_min; /* earliest point in string it can appear */
323 I32 offset_float_max; /* latest point in string it can appear */
324 I32 *minlen_float; /* pointer to the minlen relevent to the string */
325 I32 lookbehind_float; /* is the position of the string modified by LB */
329 struct regnode_charclass_class *start_class;
333 * Forward declarations for pregcomp()'s friends.
336 static const scan_data_t zero_scan_data =
337 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
339 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
340 #define SF_BEFORE_SEOL 0x0001
341 #define SF_BEFORE_MEOL 0x0002
342 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
343 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
346 # define SF_FIX_SHIFT_EOL (0+2)
347 # define SF_FL_SHIFT_EOL (0+4)
349 # define SF_FIX_SHIFT_EOL (+2)
350 # define SF_FL_SHIFT_EOL (+4)
353 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
354 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
356 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
357 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
358 #define SF_IS_INF 0x0040
359 #define SF_HAS_PAR 0x0080
360 #define SF_IN_PAR 0x0100
361 #define SF_HAS_EVAL 0x0200
362 #define SCF_DO_SUBSTR 0x0400
363 #define SCF_DO_STCLASS_AND 0x0800
364 #define SCF_DO_STCLASS_OR 0x1000
365 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
366 #define SCF_WHILEM_VISITED_POS 0x2000
368 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
369 #define SCF_SEEN_ACCEPT 0x8000
371 #define UTF cBOOL(RExC_utf8)
372 #define LOC cBOOL(RExC_flags & RXf_PMf_LOCALE)
373 #define UNI_SEMANTICS cBOOL(RExC_flags & RXf_PMf_UNICODE)
374 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
376 #define OOB_UNICODE 12345678
377 #define OOB_NAMEDCLASS -1
379 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
380 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
383 /* length of regex to show in messages that don't mark a position within */
384 #define RegexLengthToShowInErrorMessages 127
387 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
388 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
389 * op/pragma/warn/regcomp.
391 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
392 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
394 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
397 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
398 * arg. Show regex, up to a maximum length. If it's too long, chop and add
401 #define _FAIL(code) STMT_START { \
402 const char *ellipses = ""; \
403 IV len = RExC_end - RExC_precomp; \
406 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
407 if (len > RegexLengthToShowInErrorMessages) { \
408 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
409 len = RegexLengthToShowInErrorMessages - 10; \
415 #define FAIL(msg) _FAIL( \
416 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
417 msg, (int)len, RExC_precomp, ellipses))
419 #define FAIL2(msg,arg) _FAIL( \
420 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
421 arg, (int)len, RExC_precomp, ellipses))
424 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
426 #define Simple_vFAIL(m) STMT_START { \
427 const IV offset = RExC_parse - RExC_precomp; \
428 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
429 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
433 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
435 #define vFAIL(m) STMT_START { \
437 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
442 * Like Simple_vFAIL(), but accepts two arguments.
444 #define Simple_vFAIL2(m,a1) STMT_START { \
445 const IV offset = RExC_parse - RExC_precomp; \
446 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
447 (int)offset, RExC_precomp, RExC_precomp + offset); \
451 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
453 #define vFAIL2(m,a1) STMT_START { \
455 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
456 Simple_vFAIL2(m, a1); \
461 * Like Simple_vFAIL(), but accepts three arguments.
463 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
464 const IV offset = RExC_parse - RExC_precomp; \
465 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
466 (int)offset, RExC_precomp, RExC_precomp + offset); \
470 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
472 #define vFAIL3(m,a1,a2) STMT_START { \
474 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
475 Simple_vFAIL3(m, a1, a2); \
479 * Like Simple_vFAIL(), but accepts four arguments.
481 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
482 const IV offset = RExC_parse - RExC_precomp; \
483 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
484 (int)offset, RExC_precomp, RExC_precomp + offset); \
487 #define ckWARNreg(loc,m) STMT_START { \
488 const IV offset = loc - RExC_precomp; \
489 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 (int)offset, RExC_precomp, RExC_precomp + offset); \
493 #define ckWARNregdep(loc,m) STMT_START { \
494 const IV offset = loc - RExC_precomp; \
495 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
497 (int)offset, RExC_precomp, RExC_precomp + offset); \
500 #define ckWARN2reg(loc, m, a1) STMT_START { \
501 const IV offset = loc - RExC_precomp; \
502 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
503 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
506 #define vWARN3(loc, m, a1, a2) STMT_START { \
507 const IV offset = loc - RExC_precomp; \
508 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
509 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
512 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
513 const IV offset = loc - RExC_precomp; \
514 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
515 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
518 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
519 const IV offset = loc - RExC_precomp; \
520 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
521 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
524 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
525 const IV offset = loc - RExC_precomp; \
526 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
527 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
530 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
531 const IV offset = loc - RExC_precomp; \
532 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
533 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
537 /* Allow for side effects in s */
538 #define REGC(c,s) STMT_START { \
539 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
542 /* Macros for recording node offsets. 20001227 mjd@plover.com
543 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
544 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
545 * Element 0 holds the number n.
546 * Position is 1 indexed.
548 #ifndef RE_TRACK_PATTERN_OFFSETS
549 #define Set_Node_Offset_To_R(node,byte)
550 #define Set_Node_Offset(node,byte)
551 #define Set_Cur_Node_Offset
552 #define Set_Node_Length_To_R(node,len)
553 #define Set_Node_Length(node,len)
554 #define Set_Node_Cur_Length(node)
555 #define Node_Offset(n)
556 #define Node_Length(n)
557 #define Set_Node_Offset_Length(node,offset,len)
558 #define ProgLen(ri) ri->u.proglen
559 #define SetProgLen(ri,x) ri->u.proglen = x
561 #define ProgLen(ri) ri->u.offsets[0]
562 #define SetProgLen(ri,x) ri->u.offsets[0] = x
563 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
565 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
566 __LINE__, (int)(node), (int)(byte))); \
568 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
570 RExC_offsets[2*(node)-1] = (byte); \
575 #define Set_Node_Offset(node,byte) \
576 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
577 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
579 #define Set_Node_Length_To_R(node,len) STMT_START { \
581 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
582 __LINE__, (int)(node), (int)(len))); \
584 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
586 RExC_offsets[2*(node)] = (len); \
591 #define Set_Node_Length(node,len) \
592 Set_Node_Length_To_R((node)-RExC_emit_start, len)
593 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
594 #define Set_Node_Cur_Length(node) \
595 Set_Node_Length(node, RExC_parse - parse_start)
597 /* Get offsets and lengths */
598 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
599 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
601 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
602 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
603 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
607 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
608 #define EXPERIMENTAL_INPLACESCAN
609 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
611 #define DEBUG_STUDYDATA(str,data,depth) \
612 DEBUG_OPTIMISE_MORE_r(if(data){ \
613 PerlIO_printf(Perl_debug_log, \
614 "%*s" str "Pos:%"IVdf"/%"IVdf \
615 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
616 (int)(depth)*2, "", \
617 (IV)((data)->pos_min), \
618 (IV)((data)->pos_delta), \
619 (UV)((data)->flags), \
620 (IV)((data)->whilem_c), \
621 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
622 is_inf ? "INF " : "" \
624 if ((data)->last_found) \
625 PerlIO_printf(Perl_debug_log, \
626 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
627 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
628 SvPVX_const((data)->last_found), \
629 (IV)((data)->last_end), \
630 (IV)((data)->last_start_min), \
631 (IV)((data)->last_start_max), \
632 ((data)->longest && \
633 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
634 SvPVX_const((data)->longest_fixed), \
635 (IV)((data)->offset_fixed), \
636 ((data)->longest && \
637 (data)->longest==&((data)->longest_float)) ? "*" : "", \
638 SvPVX_const((data)->longest_float), \
639 (IV)((data)->offset_float_min), \
640 (IV)((data)->offset_float_max) \
642 PerlIO_printf(Perl_debug_log,"\n"); \
645 static void clear_re(pTHX_ void *r);
647 /* Mark that we cannot extend a found fixed substring at this point.
648 Update the longest found anchored substring and the longest found
649 floating substrings if needed. */
652 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
654 const STRLEN l = CHR_SVLEN(data->last_found);
655 const STRLEN old_l = CHR_SVLEN(*data->longest);
656 GET_RE_DEBUG_FLAGS_DECL;
658 PERL_ARGS_ASSERT_SCAN_COMMIT;
660 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
661 SvSetMagicSV(*data->longest, data->last_found);
662 if (*data->longest == data->longest_fixed) {
663 data->offset_fixed = l ? data->last_start_min : data->pos_min;
664 if (data->flags & SF_BEFORE_EOL)
666 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
668 data->flags &= ~SF_FIX_BEFORE_EOL;
669 data->minlen_fixed=minlenp;
670 data->lookbehind_fixed=0;
672 else { /* *data->longest == data->longest_float */
673 data->offset_float_min = l ? data->last_start_min : data->pos_min;
674 data->offset_float_max = (l
675 ? data->last_start_max
676 : data->pos_min + data->pos_delta);
677 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
678 data->offset_float_max = I32_MAX;
679 if (data->flags & SF_BEFORE_EOL)
681 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
683 data->flags &= ~SF_FL_BEFORE_EOL;
684 data->minlen_float=minlenp;
685 data->lookbehind_float=0;
688 SvCUR_set(data->last_found, 0);
690 SV * const sv = data->last_found;
691 if (SvUTF8(sv) && SvMAGICAL(sv)) {
692 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
698 data->flags &= ~SF_BEFORE_EOL;
699 DEBUG_STUDYDATA("commit: ",data,0);
702 /* Can match anything (initialization) */
704 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
706 PERL_ARGS_ASSERT_CL_ANYTHING;
708 ANYOF_CLASS_ZERO(cl);
709 ANYOF_BITMAP_SETALL(cl);
710 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
712 cl->flags |= ANYOF_LOCALE;
713 cl->flags |= ANYOF_FOLD;
716 /* Can match anything (initialization) */
718 S_cl_is_anything(const struct regnode_charclass_class *cl)
722 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
724 for (value = 0; value <= ANYOF_MAX; value += 2)
725 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
727 if (!(cl->flags & ANYOF_UNICODE_ALL))
729 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
734 /* Can match anything (initialization) */
736 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
738 PERL_ARGS_ASSERT_CL_INIT;
740 Zero(cl, 1, struct regnode_charclass_class);
742 cl_anything(pRExC_state, cl);
746 S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
748 PERL_ARGS_ASSERT_CL_INIT_ZERO;
750 Zero(cl, 1, struct regnode_charclass_class);
752 cl_anything(pRExC_state, cl);
754 cl->flags |= ANYOF_LOCALE;
757 /* 'And' a given class with another one. Can create false positives */
758 /* We assume that cl is not inverted */
760 S_cl_and(struct regnode_charclass_class *cl,
761 const struct regnode_charclass_class *and_with)
763 PERL_ARGS_ASSERT_CL_AND;
765 assert(and_with->type == ANYOF);
767 if (!(ANYOF_CLASS_TEST_ANY_SET(and_with))
768 && !(ANYOF_CLASS_TEST_ANY_SET(cl))
769 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
770 && !(and_with->flags & ANYOF_FOLD)
771 && !(cl->flags & ANYOF_FOLD)) {
774 if (and_with->flags & ANYOF_INVERT)
775 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
776 cl->bitmap[i] &= ~and_with->bitmap[i];
778 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
779 cl->bitmap[i] &= and_with->bitmap[i];
780 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
781 if (!(and_with->flags & ANYOF_EOS))
782 cl->flags &= ~ANYOF_EOS;
784 if (!(and_with->flags & ANYOF_FOLD))
785 cl->flags &= ~ANYOF_FOLD;
787 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_NONBITMAP &&
788 !(and_with->flags & ANYOF_INVERT)) {
789 cl->flags &= ~ANYOF_UNICODE_ALL;
790 cl->flags |= and_with->flags & ANYOF_NONBITMAP; /* field is 2 bits; use
793 ARG_SET(cl, ARG(and_with));
795 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
796 !(and_with->flags & ANYOF_INVERT))
797 cl->flags &= ~ANYOF_UNICODE_ALL;
798 if (!(and_with->flags & (ANYOF_NONBITMAP|ANYOF_UNICODE_ALL)) &&
799 !(and_with->flags & ANYOF_INVERT))
800 cl->flags &= ~ANYOF_NONBITMAP;
803 /* 'OR' a given class with another one. Can create false positives */
804 /* We assume that cl is not inverted */
806 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
808 PERL_ARGS_ASSERT_CL_OR;
810 if (or_with->flags & ANYOF_INVERT) {
812 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
813 * <= (B1 | !B2) | (CL1 | !CL2)
814 * which is wasteful if CL2 is small, but we ignore CL2:
815 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
816 * XXXX Can we handle case-fold? Unclear:
817 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
818 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
820 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
821 && !(or_with->flags & ANYOF_FOLD)
822 && !(cl->flags & ANYOF_FOLD) ) {
825 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
826 cl->bitmap[i] |= ~or_with->bitmap[i];
827 } /* XXXX: logic is complicated otherwise */
829 cl_anything(pRExC_state, cl);
832 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
833 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
834 && (!(or_with->flags & ANYOF_FOLD)
835 || (cl->flags & ANYOF_FOLD)) ) {
838 /* OR char bitmap and class bitmap separately */
839 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
840 cl->bitmap[i] |= or_with->bitmap[i];
841 if (ANYOF_CLASS_TEST_ANY_SET(or_with)) {
842 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
843 cl->classflags[i] |= or_with->classflags[i];
844 cl->flags |= ANYOF_CLASS;
847 else { /* XXXX: logic is complicated, leave it along for a moment. */
848 cl_anything(pRExC_state, cl);
851 if (or_with->flags & ANYOF_EOS)
852 cl->flags |= ANYOF_EOS;
854 if (or_with->flags & ANYOF_FOLD)
855 cl->flags |= ANYOF_FOLD;
857 /* If both nodes match something outside the bitmap, but what they match
858 * outside is not the same pointer, and hence not easily compared, give up
859 * and allow the start class to match everything outside the bitmap */
860 if (cl->flags & ANYOF_NONBITMAP && or_with->flags & ANYOF_NONBITMAP &&
861 ARG(cl) != ARG(or_with)) {
862 cl->flags |= ANYOF_UNICODE_ALL;
865 if (or_with->flags & ANYOF_UNICODE_ALL) {
866 cl->flags |= ANYOF_UNICODE_ALL;
870 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
871 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
872 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
873 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
878 dump_trie(trie,widecharmap,revcharmap)
879 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
880 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
882 These routines dump out a trie in a somewhat readable format.
883 The _interim_ variants are used for debugging the interim
884 tables that are used to generate the final compressed
885 representation which is what dump_trie expects.
887 Part of the reason for their existance is to provide a form
888 of documentation as to how the different representations function.
893 Dumps the final compressed table form of the trie to Perl_debug_log.
894 Used for debugging make_trie().
898 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
899 AV *revcharmap, U32 depth)
902 SV *sv=sv_newmortal();
903 int colwidth= widecharmap ? 6 : 4;
905 GET_RE_DEBUG_FLAGS_DECL;
907 PERL_ARGS_ASSERT_DUMP_TRIE;
909 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
910 (int)depth * 2 + 2,"",
911 "Match","Base","Ofs" );
913 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
914 SV ** const tmp = av_fetch( revcharmap, state, 0);
916 PerlIO_printf( Perl_debug_log, "%*s",
918 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
919 PL_colors[0], PL_colors[1],
920 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
921 PERL_PV_ESCAPE_FIRSTCHAR
926 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
927 (int)depth * 2 + 2,"");
929 for( state = 0 ; state < trie->uniquecharcount ; state++ )
930 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
931 PerlIO_printf( Perl_debug_log, "\n");
933 for( state = 1 ; state < trie->statecount ; state++ ) {
934 const U32 base = trie->states[ state ].trans.base;
936 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
938 if ( trie->states[ state ].wordnum ) {
939 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
941 PerlIO_printf( Perl_debug_log, "%6s", "" );
944 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
949 while( ( base + ofs < trie->uniquecharcount ) ||
950 ( base + ofs - trie->uniquecharcount < trie->lasttrans
951 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
954 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
956 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
957 if ( ( base + ofs >= trie->uniquecharcount ) &&
958 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
959 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
961 PerlIO_printf( Perl_debug_log, "%*"UVXf,
963 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
965 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
969 PerlIO_printf( Perl_debug_log, "]");
972 PerlIO_printf( Perl_debug_log, "\n" );
974 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
975 for (word=1; word <= trie->wordcount; word++) {
976 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
977 (int)word, (int)(trie->wordinfo[word].prev),
978 (int)(trie->wordinfo[word].len));
980 PerlIO_printf(Perl_debug_log, "\n" );
983 Dumps a fully constructed but uncompressed trie in list form.
984 List tries normally only are used for construction when the number of
985 possible chars (trie->uniquecharcount) is very high.
986 Used for debugging make_trie().
989 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
990 HV *widecharmap, AV *revcharmap, U32 next_alloc,
994 SV *sv=sv_newmortal();
995 int colwidth= widecharmap ? 6 : 4;
996 GET_RE_DEBUG_FLAGS_DECL;
998 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1000 /* print out the table precompression. */
1001 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1002 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1003 "------:-----+-----------------\n" );
1005 for( state=1 ; state < next_alloc ; state ++ ) {
1008 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1009 (int)depth * 2 + 2,"", (UV)state );
1010 if ( ! trie->states[ state ].wordnum ) {
1011 PerlIO_printf( Perl_debug_log, "%5s| ","");
1013 PerlIO_printf( Perl_debug_log, "W%4x| ",
1014 trie->states[ state ].wordnum
1017 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1018 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
1020 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1022 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1023 PL_colors[0], PL_colors[1],
1024 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1025 PERL_PV_ESCAPE_FIRSTCHAR
1027 TRIE_LIST_ITEM(state,charid).forid,
1028 (UV)TRIE_LIST_ITEM(state,charid).newstate
1031 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1032 (int)((depth * 2) + 14), "");
1035 PerlIO_printf( Perl_debug_log, "\n");
1040 Dumps a fully constructed but uncompressed trie in table form.
1041 This is the normal DFA style state transition table, with a few
1042 twists to facilitate compression later.
1043 Used for debugging make_trie().
1046 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1047 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1052 SV *sv=sv_newmortal();
1053 int colwidth= widecharmap ? 6 : 4;
1054 GET_RE_DEBUG_FLAGS_DECL;
1056 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1059 print out the table precompression so that we can do a visual check
1060 that they are identical.
1063 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1065 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1066 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1068 PerlIO_printf( Perl_debug_log, "%*s",
1070 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1071 PL_colors[0], PL_colors[1],
1072 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1073 PERL_PV_ESCAPE_FIRSTCHAR
1079 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1081 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1082 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1085 PerlIO_printf( Perl_debug_log, "\n" );
1087 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1089 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1090 (int)depth * 2 + 2,"",
1091 (UV)TRIE_NODENUM( state ) );
1093 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1094 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1096 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1098 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1100 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1101 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1103 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1104 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1112 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1113 startbranch: the first branch in the whole branch sequence
1114 first : start branch of sequence of branch-exact nodes.
1115 May be the same as startbranch
1116 last : Thing following the last branch.
1117 May be the same as tail.
1118 tail : item following the branch sequence
1119 count : words in the sequence
1120 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1121 depth : indent depth
1123 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1125 A trie is an N'ary tree where the branches are determined by digital
1126 decomposition of the key. IE, at the root node you look up the 1st character and
1127 follow that branch repeat until you find the end of the branches. Nodes can be
1128 marked as "accepting" meaning they represent a complete word. Eg:
1132 would convert into the following structure. Numbers represent states, letters
1133 following numbers represent valid transitions on the letter from that state, if
1134 the number is in square brackets it represents an accepting state, otherwise it
1135 will be in parenthesis.
1137 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1141 (1) +-i->(6)-+-s->[7]
1143 +-s->(3)-+-h->(4)-+-e->[5]
1145 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1147 This shows that when matching against the string 'hers' we will begin at state 1
1148 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1149 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1150 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1151 single traverse. We store a mapping from accepting to state to which word was
1152 matched, and then when we have multiple possibilities we try to complete the
1153 rest of the regex in the order in which they occured in the alternation.
1155 The only prior NFA like behaviour that would be changed by the TRIE support is
1156 the silent ignoring of duplicate alternations which are of the form:
1158 / (DUPE|DUPE) X? (?{ ... }) Y /x
1160 Thus EVAL blocks following a trie may be called a different number of times with
1161 and without the optimisation. With the optimisations dupes will be silently
1162 ignored. This inconsistant behaviour of EVAL type nodes is well established as
1163 the following demonstrates:
1165 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1167 which prints out 'word' three times, but
1169 'words'=~/(word|word|word)(?{ print $1 })S/
1171 which doesnt print it out at all. This is due to other optimisations kicking in.
1173 Example of what happens on a structural level:
1175 The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1177 1: CURLYM[1] {1,32767}(18)
1188 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1189 and should turn into:
1191 1: CURLYM[1] {1,32767}(18)
1193 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1201 Cases where tail != last would be like /(?foo|bar)baz/:
1211 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1212 and would end up looking like:
1215 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1222 d = uvuni_to_utf8_flags(d, uv, 0);
1224 is the recommended Unicode-aware way of saying
1229 #define TRIE_STORE_REVCHAR \
1232 SV *zlopp = newSV(2); \
1233 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1234 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1235 SvCUR_set(zlopp, kapow - flrbbbbb); \
1238 av_push(revcharmap, zlopp); \
1240 char ooooff = (char)uvc; \
1241 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1245 #define TRIE_READ_CHAR STMT_START { \
1249 if ( foldlen > 0 ) { \
1250 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1255 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1256 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1257 foldlen -= UNISKIP( uvc ); \
1258 scan = foldbuf + UNISKIP( uvc ); \
1261 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1271 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1272 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1273 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1274 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1276 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1277 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1278 TRIE_LIST_CUR( state )++; \
1281 #define TRIE_LIST_NEW(state) STMT_START { \
1282 Newxz( trie->states[ state ].trans.list, \
1283 4, reg_trie_trans_le ); \
1284 TRIE_LIST_CUR( state ) = 1; \
1285 TRIE_LIST_LEN( state ) = 4; \
1288 #define TRIE_HANDLE_WORD(state) STMT_START { \
1289 U16 dupe= trie->states[ state ].wordnum; \
1290 regnode * const noper_next = regnext( noper ); \
1293 /* store the word for dumping */ \
1295 if (OP(noper) != NOTHING) \
1296 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1298 tmp = newSVpvn_utf8( "", 0, UTF ); \
1299 av_push( trie_words, tmp ); \
1303 trie->wordinfo[curword].prev = 0; \
1304 trie->wordinfo[curword].len = wordlen; \
1305 trie->wordinfo[curword].accept = state; \
1307 if ( noper_next < tail ) { \
1309 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1310 trie->jump[curword] = (U16)(noper_next - convert); \
1312 jumper = noper_next; \
1314 nextbranch= regnext(cur); \
1318 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1319 /* chain, so that when the bits of chain are later */\
1320 /* linked together, the dups appear in the chain */\
1321 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1322 trie->wordinfo[dupe].prev = curword; \
1324 /* we haven't inserted this word yet. */ \
1325 trie->states[ state ].wordnum = curword; \
1330 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1331 ( ( base + charid >= ucharcount \
1332 && base + charid < ubound \
1333 && state == trie->trans[ base - ucharcount + charid ].check \
1334 && trie->trans[ base - ucharcount + charid ].next ) \
1335 ? trie->trans[ base - ucharcount + charid ].next \
1336 : ( state==1 ? special : 0 ) \
1340 #define MADE_JUMP_TRIE 2
1341 #define MADE_EXACT_TRIE 4
1344 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1347 /* first pass, loop through and scan words */
1348 reg_trie_data *trie;
1349 HV *widecharmap = NULL;
1350 AV *revcharmap = newAV();
1352 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1357 regnode *jumper = NULL;
1358 regnode *nextbranch = NULL;
1359 regnode *convert = NULL;
1360 U32 *prev_states; /* temp array mapping each state to previous one */
1361 /* we just use folder as a flag in utf8 */
1362 const U8 * folder = NULL;
1365 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1366 AV *trie_words = NULL;
1367 /* along with revcharmap, this only used during construction but both are
1368 * useful during debugging so we store them in the struct when debugging.
1371 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1372 STRLEN trie_charcount=0;
1374 SV *re_trie_maxbuff;
1375 GET_RE_DEBUG_FLAGS_DECL;
1377 PERL_ARGS_ASSERT_MAKE_TRIE;
1379 PERL_UNUSED_ARG(depth);
1383 case EXACTFU: folder = PL_fold_latin1; break;
1384 case EXACTF: folder = PL_fold; break;
1385 case EXACTFL: folder = PL_fold_locale; break;
1388 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1390 trie->startstate = 1;
1391 trie->wordcount = word_count;
1392 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1393 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1394 if (!(UTF && folder))
1395 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1396 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1397 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1400 trie_words = newAV();
1403 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1404 if (!SvIOK(re_trie_maxbuff)) {
1405 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1408 PerlIO_printf( Perl_debug_log,
1409 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1410 (int)depth * 2 + 2, "",
1411 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1412 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1416 /* Find the node we are going to overwrite */
1417 if ( first == startbranch && OP( last ) != BRANCH ) {
1418 /* whole branch chain */
1421 /* branch sub-chain */
1422 convert = NEXTOPER( first );
1425 /* -- First loop and Setup --
1427 We first traverse the branches and scan each word to determine if it
1428 contains widechars, and how many unique chars there are, this is
1429 important as we have to build a table with at least as many columns as we
1432 We use an array of integers to represent the character codes 0..255
1433 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1434 native representation of the character value as the key and IV's for the
1437 *TODO* If we keep track of how many times each character is used we can
1438 remap the columns so that the table compression later on is more
1439 efficient in terms of memory by ensuring the most common value is in the
1440 middle and the least common are on the outside. IMO this would be better
1441 than a most to least common mapping as theres a decent chance the most
1442 common letter will share a node with the least common, meaning the node
1443 will not be compressable. With a middle is most common approach the worst
1444 case is when we have the least common nodes twice.
1448 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1449 regnode * const noper = NEXTOPER( cur );
1450 const U8 *uc = (U8*)STRING( noper );
1451 const U8 * const e = uc + STR_LEN( noper );
1453 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1454 const U8 *scan = (U8*)NULL;
1455 U32 wordlen = 0; /* required init */
1457 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1459 if (OP(noper) == NOTHING) {
1463 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1464 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1465 regardless of encoding */
1467 for ( ; uc < e ; uc += len ) {
1468 TRIE_CHARCOUNT(trie)++;
1472 if ( !trie->charmap[ uvc ] ) {
1473 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1475 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1479 /* store the codepoint in the bitmap, and its folded
1481 TRIE_BITMAP_SET(trie,uvc);
1483 /* store the folded codepoint */
1484 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1487 /* store first byte of utf8 representation of
1488 variant codepoints */
1489 if (! UNI_IS_INVARIANT(uvc)) {
1490 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
1493 set_bit = 0; /* We've done our bit :-) */
1498 widecharmap = newHV();
1500 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1503 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1505 if ( !SvTRUE( *svpp ) ) {
1506 sv_setiv( *svpp, ++trie->uniquecharcount );
1511 if( cur == first ) {
1514 } else if (chars < trie->minlen) {
1516 } else if (chars > trie->maxlen) {
1520 } /* end first pass */
1521 DEBUG_TRIE_COMPILE_r(
1522 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1523 (int)depth * 2 + 2,"",
1524 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1525 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1526 (int)trie->minlen, (int)trie->maxlen )
1530 We now know what we are dealing with in terms of unique chars and
1531 string sizes so we can calculate how much memory a naive
1532 representation using a flat table will take. If it's over a reasonable
1533 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1534 conservative but potentially much slower representation using an array
1537 At the end we convert both representations into the same compressed
1538 form that will be used in regexec.c for matching with. The latter
1539 is a form that cannot be used to construct with but has memory
1540 properties similar to the list form and access properties similar
1541 to the table form making it both suitable for fast searches and
1542 small enough that its feasable to store for the duration of a program.
1544 See the comment in the code where the compressed table is produced
1545 inplace from the flat tabe representation for an explanation of how
1546 the compression works.
1551 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1554 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1556 Second Pass -- Array Of Lists Representation
1558 Each state will be represented by a list of charid:state records
1559 (reg_trie_trans_le) the first such element holds the CUR and LEN
1560 points of the allocated array. (See defines above).
1562 We build the initial structure using the lists, and then convert
1563 it into the compressed table form which allows faster lookups
1564 (but cant be modified once converted).
1567 STRLEN transcount = 1;
1569 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1570 "%*sCompiling trie using list compiler\n",
1571 (int)depth * 2 + 2, ""));
1573 trie->states = (reg_trie_state *)
1574 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1575 sizeof(reg_trie_state) );
1579 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1581 regnode * const noper = NEXTOPER( cur );
1582 U8 *uc = (U8*)STRING( noper );
1583 const U8 * const e = uc + STR_LEN( noper );
1584 U32 state = 1; /* required init */
1585 U16 charid = 0; /* sanity init */
1586 U8 *scan = (U8*)NULL; /* sanity init */
1587 STRLEN foldlen = 0; /* required init */
1588 U32 wordlen = 0; /* required init */
1589 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1591 if (OP(noper) != NOTHING) {
1592 for ( ; uc < e ; uc += len ) {
1597 charid = trie->charmap[ uvc ];
1599 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1603 charid=(U16)SvIV( *svpp );
1606 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1613 if ( !trie->states[ state ].trans.list ) {
1614 TRIE_LIST_NEW( state );
1616 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1617 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1618 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1623 newstate = next_alloc++;
1624 prev_states[newstate] = state;
1625 TRIE_LIST_PUSH( state, charid, newstate );
1630 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1634 TRIE_HANDLE_WORD(state);
1636 } /* end second pass */
1638 /* next alloc is the NEXT state to be allocated */
1639 trie->statecount = next_alloc;
1640 trie->states = (reg_trie_state *)
1641 PerlMemShared_realloc( trie->states,
1643 * sizeof(reg_trie_state) );
1645 /* and now dump it out before we compress it */
1646 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1647 revcharmap, next_alloc,
1651 trie->trans = (reg_trie_trans *)
1652 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1659 for( state=1 ; state < next_alloc ; state ++ ) {
1663 DEBUG_TRIE_COMPILE_MORE_r(
1664 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1668 if (trie->states[state].trans.list) {
1669 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1673 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1674 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1675 if ( forid < minid ) {
1677 } else if ( forid > maxid ) {
1681 if ( transcount < tp + maxid - minid + 1) {
1683 trie->trans = (reg_trie_trans *)
1684 PerlMemShared_realloc( trie->trans,
1686 * sizeof(reg_trie_trans) );
1687 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1689 base = trie->uniquecharcount + tp - minid;
1690 if ( maxid == minid ) {
1692 for ( ; zp < tp ; zp++ ) {
1693 if ( ! trie->trans[ zp ].next ) {
1694 base = trie->uniquecharcount + zp - minid;
1695 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1696 trie->trans[ zp ].check = state;
1702 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1703 trie->trans[ tp ].check = state;
1708 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1709 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1710 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1711 trie->trans[ tid ].check = state;
1713 tp += ( maxid - minid + 1 );
1715 Safefree(trie->states[ state ].trans.list);
1718 DEBUG_TRIE_COMPILE_MORE_r(
1719 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1722 trie->states[ state ].trans.base=base;
1724 trie->lasttrans = tp + 1;
1728 Second Pass -- Flat Table Representation.
1730 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1731 We know that we will need Charcount+1 trans at most to store the data
1732 (one row per char at worst case) So we preallocate both structures
1733 assuming worst case.
1735 We then construct the trie using only the .next slots of the entry
1738 We use the .check field of the first entry of the node temporarily to
1739 make compression both faster and easier by keeping track of how many non
1740 zero fields are in the node.
1742 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1745 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1746 number representing the first entry of the node, and state as a
1747 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1748 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1749 are 2 entrys per node. eg:
1757 The table is internally in the right hand, idx form. However as we also
1758 have to deal with the states array which is indexed by nodenum we have to
1759 use TRIE_NODENUM() to convert.
1762 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1763 "%*sCompiling trie using table compiler\n",
1764 (int)depth * 2 + 2, ""));
1766 trie->trans = (reg_trie_trans *)
1767 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1768 * trie->uniquecharcount + 1,
1769 sizeof(reg_trie_trans) );
1770 trie->states = (reg_trie_state *)
1771 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1772 sizeof(reg_trie_state) );
1773 next_alloc = trie->uniquecharcount + 1;
1776 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1778 regnode * const noper = NEXTOPER( cur );
1779 const U8 *uc = (U8*)STRING( noper );
1780 const U8 * const e = uc + STR_LEN( noper );
1782 U32 state = 1; /* required init */
1784 U16 charid = 0; /* sanity init */
1785 U32 accept_state = 0; /* sanity init */
1786 U8 *scan = (U8*)NULL; /* sanity init */
1788 STRLEN foldlen = 0; /* required init */
1789 U32 wordlen = 0; /* required init */
1790 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1792 if ( OP(noper) != NOTHING ) {
1793 for ( ; uc < e ; uc += len ) {
1798 charid = trie->charmap[ uvc ];
1800 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1801 charid = svpp ? (U16)SvIV(*svpp) : 0;
1805 if ( !trie->trans[ state + charid ].next ) {
1806 trie->trans[ state + charid ].next = next_alloc;
1807 trie->trans[ state ].check++;
1808 prev_states[TRIE_NODENUM(next_alloc)]
1809 = TRIE_NODENUM(state);
1810 next_alloc += trie->uniquecharcount;
1812 state = trie->trans[ state + charid ].next;
1814 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1816 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1819 accept_state = TRIE_NODENUM( state );
1820 TRIE_HANDLE_WORD(accept_state);
1822 } /* end second pass */
1824 /* and now dump it out before we compress it */
1825 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1827 next_alloc, depth+1));
1831 * Inplace compress the table.*
1833 For sparse data sets the table constructed by the trie algorithm will
1834 be mostly 0/FAIL transitions or to put it another way mostly empty.
1835 (Note that leaf nodes will not contain any transitions.)
1837 This algorithm compresses the tables by eliminating most such
1838 transitions, at the cost of a modest bit of extra work during lookup:
1840 - Each states[] entry contains a .base field which indicates the
1841 index in the state[] array wheres its transition data is stored.
1843 - If .base is 0 there are no valid transitions from that node.
1845 - If .base is nonzero then charid is added to it to find an entry in
1848 -If trans[states[state].base+charid].check!=state then the
1849 transition is taken to be a 0/Fail transition. Thus if there are fail
1850 transitions at the front of the node then the .base offset will point
1851 somewhere inside the previous nodes data (or maybe even into a node
1852 even earlier), but the .check field determines if the transition is
1856 The following process inplace converts the table to the compressed
1857 table: We first do not compress the root node 1,and mark all its
1858 .check pointers as 1 and set its .base pointer as 1 as well. This
1859 allows us to do a DFA construction from the compressed table later,
1860 and ensures that any .base pointers we calculate later are greater
1863 - We set 'pos' to indicate the first entry of the second node.
1865 - We then iterate over the columns of the node, finding the first and
1866 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1867 and set the .check pointers accordingly, and advance pos
1868 appropriately and repreat for the next node. Note that when we copy
1869 the next pointers we have to convert them from the original
1870 NODEIDX form to NODENUM form as the former is not valid post
1873 - If a node has no transitions used we mark its base as 0 and do not
1874 advance the pos pointer.
1876 - If a node only has one transition we use a second pointer into the
1877 structure to fill in allocated fail transitions from other states.
1878 This pointer is independent of the main pointer and scans forward
1879 looking for null transitions that are allocated to a state. When it
1880 finds one it writes the single transition into the "hole". If the
1881 pointer doesnt find one the single transition is appended as normal.
1883 - Once compressed we can Renew/realloc the structures to release the
1886 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1887 specifically Fig 3.47 and the associated pseudocode.
1891 const U32 laststate = TRIE_NODENUM( next_alloc );
1894 trie->statecount = laststate;
1896 for ( state = 1 ; state < laststate ; state++ ) {
1898 const U32 stateidx = TRIE_NODEIDX( state );
1899 const U32 o_used = trie->trans[ stateidx ].check;
1900 U32 used = trie->trans[ stateidx ].check;
1901 trie->trans[ stateidx ].check = 0;
1903 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1904 if ( flag || trie->trans[ stateidx + charid ].next ) {
1905 if ( trie->trans[ stateidx + charid ].next ) {
1907 for ( ; zp < pos ; zp++ ) {
1908 if ( ! trie->trans[ zp ].next ) {
1912 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1913 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1914 trie->trans[ zp ].check = state;
1915 if ( ++zp > pos ) pos = zp;
1922 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1924 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1925 trie->trans[ pos ].check = state;
1930 trie->lasttrans = pos + 1;
1931 trie->states = (reg_trie_state *)
1932 PerlMemShared_realloc( trie->states, laststate
1933 * sizeof(reg_trie_state) );
1934 DEBUG_TRIE_COMPILE_MORE_r(
1935 PerlIO_printf( Perl_debug_log,
1936 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1937 (int)depth * 2 + 2,"",
1938 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
1941 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1944 } /* end table compress */
1946 DEBUG_TRIE_COMPILE_MORE_r(
1947 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1948 (int)depth * 2 + 2, "",
1949 (UV)trie->statecount,
1950 (UV)trie->lasttrans)
1952 /* resize the trans array to remove unused space */
1953 trie->trans = (reg_trie_trans *)
1954 PerlMemShared_realloc( trie->trans, trie->lasttrans
1955 * sizeof(reg_trie_trans) );
1957 { /* Modify the program and insert the new TRIE node */
1958 U8 nodetype =(U8)(flags & 0xFF);
1962 regnode *optimize = NULL;
1963 #ifdef RE_TRACK_PATTERN_OFFSETS
1966 U32 mjd_nodelen = 0;
1967 #endif /* RE_TRACK_PATTERN_OFFSETS */
1968 #endif /* DEBUGGING */
1970 This means we convert either the first branch or the first Exact,
1971 depending on whether the thing following (in 'last') is a branch
1972 or not and whther first is the startbranch (ie is it a sub part of
1973 the alternation or is it the whole thing.)
1974 Assuming its a sub part we convert the EXACT otherwise we convert
1975 the whole branch sequence, including the first.
1977 /* Find the node we are going to overwrite */
1978 if ( first != startbranch || OP( last ) == BRANCH ) {
1979 /* branch sub-chain */
1980 NEXT_OFF( first ) = (U16)(last - first);
1981 #ifdef RE_TRACK_PATTERN_OFFSETS
1983 mjd_offset= Node_Offset((convert));
1984 mjd_nodelen= Node_Length((convert));
1987 /* whole branch chain */
1989 #ifdef RE_TRACK_PATTERN_OFFSETS
1992 const regnode *nop = NEXTOPER( convert );
1993 mjd_offset= Node_Offset((nop));
1994 mjd_nodelen= Node_Length((nop));
1998 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1999 (int)depth * 2 + 2, "",
2000 (UV)mjd_offset, (UV)mjd_nodelen)
2003 /* But first we check to see if there is a common prefix we can
2004 split out as an EXACT and put in front of the TRIE node. */
2005 trie->startstate= 1;
2006 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2008 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2012 const U32 base = trie->states[ state ].trans.base;
2014 if ( trie->states[state].wordnum )
2017 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2018 if ( ( base + ofs >= trie->uniquecharcount ) &&
2019 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2020 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2022 if ( ++count > 1 ) {
2023 SV **tmp = av_fetch( revcharmap, ofs, 0);
2024 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2025 if ( state == 1 ) break;
2027 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2029 PerlIO_printf(Perl_debug_log,
2030 "%*sNew Start State=%"UVuf" Class: [",
2031 (int)depth * 2 + 2, "",
2034 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2035 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2037 TRIE_BITMAP_SET(trie,*ch);
2039 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2041 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2045 TRIE_BITMAP_SET(trie,*ch);
2047 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2048 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2054 SV **tmp = av_fetch( revcharmap, idx, 0);
2056 char *ch = SvPV( *tmp, len );
2058 SV *sv=sv_newmortal();
2059 PerlIO_printf( Perl_debug_log,
2060 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2061 (int)depth * 2 + 2, "",
2063 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2064 PL_colors[0], PL_colors[1],
2065 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2066 PERL_PV_ESCAPE_FIRSTCHAR
2071 OP( convert ) = nodetype;
2072 str=STRING(convert);
2075 STR_LEN(convert) += len;
2081 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2086 trie->prefixlen = (state-1);
2088 regnode *n = convert+NODE_SZ_STR(convert);
2089 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2090 trie->startstate = state;
2091 trie->minlen -= (state - 1);
2092 trie->maxlen -= (state - 1);
2094 /* At least the UNICOS C compiler choked on this
2095 * being argument to DEBUG_r(), so let's just have
2098 #ifdef PERL_EXT_RE_BUILD
2104 regnode *fix = convert;
2105 U32 word = trie->wordcount;
2107 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2108 while( ++fix < n ) {
2109 Set_Node_Offset_Length(fix, 0, 0);
2112 SV ** const tmp = av_fetch( trie_words, word, 0 );
2114 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2115 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2117 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2125 NEXT_OFF(convert) = (U16)(tail - convert);
2126 DEBUG_r(optimize= n);
2132 if ( trie->maxlen ) {
2133 NEXT_OFF( convert ) = (U16)(tail - convert);
2134 ARG_SET( convert, data_slot );
2135 /* Store the offset to the first unabsorbed branch in
2136 jump[0], which is otherwise unused by the jump logic.
2137 We use this when dumping a trie and during optimisation. */
2139 trie->jump[0] = (U16)(nextbranch - convert);
2141 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2142 * and there is a bitmap
2143 * and the first "jump target" node we found leaves enough room
2144 * then convert the TRIE node into a TRIEC node, with the bitmap
2145 * embedded inline in the opcode - this is hypothetically faster.
2147 if ( !trie->states[trie->startstate].wordnum
2149 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2151 OP( convert ) = TRIEC;
2152 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2153 PerlMemShared_free(trie->bitmap);
2156 OP( convert ) = TRIE;
2158 /* store the type in the flags */
2159 convert->flags = nodetype;
2163 + regarglen[ OP( convert ) ];
2165 /* XXX We really should free up the resource in trie now,
2166 as we won't use them - (which resources?) dmq */
2168 /* needed for dumping*/
2169 DEBUG_r(if (optimize) {
2170 regnode *opt = convert;
2172 while ( ++opt < optimize) {
2173 Set_Node_Offset_Length(opt,0,0);
2176 Try to clean up some of the debris left after the
2179 while( optimize < jumper ) {
2180 mjd_nodelen += Node_Length((optimize));
2181 OP( optimize ) = OPTIMIZED;
2182 Set_Node_Offset_Length(optimize,0,0);
2185 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2187 } /* end node insert */
2188 REH_CALL_COMP_NODE_HOOK(pRExC_state->rx, convert);
2190 /* Finish populating the prev field of the wordinfo array. Walk back
2191 * from each accept state until we find another accept state, and if
2192 * so, point the first word's .prev field at the second word. If the
2193 * second already has a .prev field set, stop now. This will be the
2194 * case either if we've already processed that word's accept state,
2195 * or that state had multiple words, and the overspill words were
2196 * already linked up earlier.
2203 for (word=1; word <= trie->wordcount; word++) {
2205 if (trie->wordinfo[word].prev)
2207 state = trie->wordinfo[word].accept;
2209 state = prev_states[state];
2212 prev = trie->states[state].wordnum;
2216 trie->wordinfo[word].prev = prev;
2218 Safefree(prev_states);
2222 /* and now dump out the compressed format */
2223 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2225 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2227 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2228 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2230 SvREFCNT_dec(revcharmap);
2234 : trie->startstate>1
2240 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2242 /* The Trie is constructed and compressed now so we can build a fail array if it's needed
2244 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2245 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2248 We find the fail state for each state in the trie, this state is the longest proper
2249 suffix of the current state's 'word' that is also a proper prefix of another word in our
2250 trie. State 1 represents the word '' and is thus the default fail state. This allows
2251 the DFA not to have to restart after its tried and failed a word at a given point, it
2252 simply continues as though it had been matching the other word in the first place.
2254 'abcdgu'=~/abcdefg|cdgu/
2255 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2256 fail, which would bring us to the state representing 'd' in the second word where we would
2257 try 'g' and succeed, proceeding to match 'cdgu'.
2259 /* add a fail transition */
2260 const U32 trie_offset = ARG(source);
2261 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2263 const U32 ucharcount = trie->uniquecharcount;
2264 const U32 numstates = trie->statecount;
2265 const U32 ubound = trie->lasttrans + ucharcount;
2269 U32 base = trie->states[ 1 ].trans.base;
2272 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2273 GET_RE_DEBUG_FLAGS_DECL;
2275 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2277 PERL_UNUSED_ARG(depth);
2281 ARG_SET( stclass, data_slot );
2282 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2283 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2284 aho->trie=trie_offset;
2285 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2286 Copy( trie->states, aho->states, numstates, reg_trie_state );
2287 Newxz( q, numstates, U32);
2288 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2291 /* initialize fail[0..1] to be 1 so that we always have
2292 a valid final fail state */
2293 fail[ 0 ] = fail[ 1 ] = 1;
2295 for ( charid = 0; charid < ucharcount ; charid++ ) {
2296 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2298 q[ q_write ] = newstate;
2299 /* set to point at the root */
2300 fail[ q[ q_write++ ] ]=1;
2303 while ( q_read < q_write) {
2304 const U32 cur = q[ q_read++ % numstates ];
2305 base = trie->states[ cur ].trans.base;
2307 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2308 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2310 U32 fail_state = cur;
2313 fail_state = fail[ fail_state ];
2314 fail_base = aho->states[ fail_state ].trans.base;
2315 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2317 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2318 fail[ ch_state ] = fail_state;
2319 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2321 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2323 q[ q_write++ % numstates] = ch_state;
2327 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2328 when we fail in state 1, this allows us to use the
2329 charclass scan to find a valid start char. This is based on the principle
2330 that theres a good chance the string being searched contains lots of stuff
2331 that cant be a start char.
2333 fail[ 0 ] = fail[ 1 ] = 0;
2334 DEBUG_TRIE_COMPILE_r({
2335 PerlIO_printf(Perl_debug_log,
2336 "%*sStclass Failtable (%"UVuf" states): 0",
2337 (int)(depth * 2), "", (UV)numstates
2339 for( q_read=1; q_read<numstates; q_read++ ) {
2340 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2342 PerlIO_printf(Perl_debug_log, "\n");
2345 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2350 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2351 * These need to be revisited when a newer toolchain becomes available.
2353 #if defined(__sparc64__) && defined(__GNUC__)
2354 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2355 # undef SPARC64_GCC_WORKAROUND
2356 # define SPARC64_GCC_WORKAROUND 1
2360 #define DEBUG_PEEP(str,scan,depth) \
2361 DEBUG_OPTIMISE_r({if (scan){ \
2362 SV * const mysv=sv_newmortal(); \
2363 regnode *Next = regnext(scan); \
2364 regprop(RExC_rx, mysv, scan); \
2365 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2366 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2367 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2374 #define JOIN_EXACT(scan,min,flags) \
2375 if (PL_regkind[OP(scan)] == EXACT) \
2376 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2379 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2380 /* Merge several consecutive EXACTish nodes into one. */
2381 regnode *n = regnext(scan);
2383 regnode *next = scan + NODE_SZ_STR(scan);
2387 regnode *stop = scan;
2388 GET_RE_DEBUG_FLAGS_DECL;
2390 PERL_UNUSED_ARG(depth);
2393 PERL_ARGS_ASSERT_JOIN_EXACT;
2394 #ifndef EXPERIMENTAL_INPLACESCAN
2395 PERL_UNUSED_ARG(flags);
2396 PERL_UNUSED_ARG(val);
2398 DEBUG_PEEP("join",scan,depth);
2400 /* Skip NOTHING, merge EXACT*. */
2402 ( PL_regkind[OP(n)] == NOTHING ||
2403 (stringok && (OP(n) == OP(scan))))
2405 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2407 if (OP(n) == TAIL || n > next)
2409 if (PL_regkind[OP(n)] == NOTHING) {
2410 DEBUG_PEEP("skip:",n,depth);
2411 NEXT_OFF(scan) += NEXT_OFF(n);
2412 next = n + NODE_STEP_REGNODE;
2419 else if (stringok) {
2420 const unsigned int oldl = STR_LEN(scan);
2421 regnode * const nnext = regnext(n);
2423 DEBUG_PEEP("merg",n,depth);
2426 if (oldl + STR_LEN(n) > U8_MAX)
2428 NEXT_OFF(scan) += NEXT_OFF(n);
2429 STR_LEN(scan) += STR_LEN(n);
2430 next = n + NODE_SZ_STR(n);
2431 /* Now we can overwrite *n : */
2432 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2440 #ifdef EXPERIMENTAL_INPLACESCAN
2441 if (flags && !NEXT_OFF(n)) {
2442 DEBUG_PEEP("atch", val, depth);
2443 if (reg_off_by_arg[OP(n)]) {
2444 ARG_SET(n, val - n);
2447 NEXT_OFF(n) = val - n;
2453 #define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390
2454 #define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS
2455 #define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0
2456 #define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS
2459 && ( OP(scan) == EXACTF || OP(scan) == EXACTFU)
2460 && ( STR_LEN(scan) >= 6 ) )
2463 Two problematic code points in Unicode casefolding of EXACT nodes:
2465 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2466 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2472 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2473 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2475 This means that in case-insensitive matching (or "loose matching",
2476 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2477 length of the above casefolded versions) can match a target string
2478 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2479 This would rather mess up the minimum length computation.
2481 What we'll do is to look for the tail four bytes, and then peek
2482 at the preceding two bytes to see whether we need to decrease
2483 the minimum length by four (six minus two).
2485 Thanks to the design of UTF-8, there cannot be false matches:
2486 A sequence of valid UTF-8 bytes cannot be a subsequence of
2487 another valid sequence of UTF-8 bytes.
2490 char * const s0 = STRING(scan), *s, *t;
2491 char * const s1 = s0 + STR_LEN(scan) - 1;
2492 char * const s2 = s1 - 4;
2493 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2494 const char t0[] = "\xaf\x49\xaf\x42";
2496 const char t0[] = "\xcc\x88\xcc\x81";
2498 const char * const t1 = t0 + 3;
2501 s < s2 && (t = ninstr(s, s1, t0, t1));
2504 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2505 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2507 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2508 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2516 n = scan + NODE_SZ_STR(scan);
2518 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2525 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2529 /* REx optimizer. Converts nodes into quickier variants "in place".
2530 Finds fixed substrings. */
2532 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2533 to the position after last scanned or to NULL. */
2535 #define INIT_AND_WITHP \
2536 assert(!and_withp); \
2537 Newx(and_withp,1,struct regnode_charclass_class); \
2538 SAVEFREEPV(and_withp)
2540 /* this is a chain of data about sub patterns we are processing that
2541 need to be handled seperately/specially in study_chunk. Its so
2542 we can simulate recursion without losing state. */
2544 typedef struct scan_frame {
2545 regnode *last; /* last node to process in this frame */
2546 regnode *next; /* next node to process when last is reached */
2547 struct scan_frame *prev; /*previous frame*/
2548 I32 stop; /* what stopparen do we use */
2552 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2554 #define CASE_SYNST_FNC(nAmE) \
2556 if (flags & SCF_DO_STCLASS_AND) { \
2557 for (value = 0; value < 256; value++) \
2558 if (!is_ ## nAmE ## _cp(value)) \
2559 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2562 for (value = 0; value < 256; value++) \
2563 if (is_ ## nAmE ## _cp(value)) \
2564 ANYOF_BITMAP_SET(data->start_class, value); \
2568 if (flags & SCF_DO_STCLASS_AND) { \
2569 for (value = 0; value < 256; value++) \
2570 if (is_ ## nAmE ## _cp(value)) \
2571 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2574 for (value = 0; value < 256; value++) \
2575 if (!is_ ## nAmE ## _cp(value)) \
2576 ANYOF_BITMAP_SET(data->start_class, value); \
2583 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2584 I32 *minlenp, I32 *deltap,
2589 struct regnode_charclass_class *and_withp,
2590 U32 flags, U32 depth)
2591 /* scanp: Start here (read-write). */
2592 /* deltap: Write maxlen-minlen here. */
2593 /* last: Stop before this one. */
2594 /* data: string data about the pattern */
2595 /* stopparen: treat close N as END */
2596 /* recursed: which subroutines have we recursed into */
2597 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2600 I32 min = 0, pars = 0, code;
2601 regnode *scan = *scanp, *next;
2603 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2604 int is_inf_internal = 0; /* The studied chunk is infinite */
2605 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2606 scan_data_t data_fake;
2607 SV *re_trie_maxbuff = NULL;
2608 regnode *first_non_open = scan;
2609 I32 stopmin = I32_MAX;
2610 scan_frame *frame = NULL;
2611 GET_RE_DEBUG_FLAGS_DECL;
2613 PERL_ARGS_ASSERT_STUDY_CHUNK;
2616 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2620 while (first_non_open && OP(first_non_open) == OPEN)
2621 first_non_open=regnext(first_non_open);
2626 while ( scan && OP(scan) != END && scan < last ){
2627 /* Peephole optimizer: */
2628 DEBUG_STUDYDATA("Peep:", data,depth);
2629 DEBUG_PEEP("Peep",scan,depth);
2630 JOIN_EXACT(scan,&min,0);
2632 /* Follow the next-chain of the current node and optimize
2633 away all the NOTHINGs from it. */
2634 if (OP(scan) != CURLYX) {
2635 const int max = (reg_off_by_arg[OP(scan)]
2637 /* I32 may be smaller than U16 on CRAYs! */
2638 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2639 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2643 /* Skip NOTHING and LONGJMP. */
2644 while ((n = regnext(n))
2645 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2646 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2647 && off + noff < max)
2649 if (reg_off_by_arg[OP(scan)])
2652 NEXT_OFF(scan) = off;
2657 /* The principal pseudo-switch. Cannot be a switch, since we
2658 look into several different things. */
2659 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2660 || OP(scan) == IFTHEN) {
2661 next = regnext(scan);
2663 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2665 if (OP(next) == code || code == IFTHEN) {
2666 /* NOTE - There is similar code to this block below for handling
2667 TRIE nodes on a re-study. If you change stuff here check there
2669 I32 max1 = 0, min1 = I32_MAX, num = 0;
2670 struct regnode_charclass_class accum;
2671 regnode * const startbranch=scan;
2673 if (flags & SCF_DO_SUBSTR)
2674 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2675 if (flags & SCF_DO_STCLASS)
2676 cl_init_zero(pRExC_state, &accum);
2678 while (OP(scan) == code) {
2679 I32 deltanext, minnext, f = 0, fake;
2680 struct regnode_charclass_class this_class;
2683 data_fake.flags = 0;
2685 data_fake.whilem_c = data->whilem_c;
2686 data_fake.last_closep = data->last_closep;
2689 data_fake.last_closep = &fake;
2691 data_fake.pos_delta = delta;
2692 next = regnext(scan);
2693 scan = NEXTOPER(scan);
2695 scan = NEXTOPER(scan);
2696 if (flags & SCF_DO_STCLASS) {
2697 cl_init(pRExC_state, &this_class);
2698 data_fake.start_class = &this_class;
2699 f = SCF_DO_STCLASS_AND;
2701 if (flags & SCF_WHILEM_VISITED_POS)
2702 f |= SCF_WHILEM_VISITED_POS;
2704 /* we suppose the run is continuous, last=next...*/
2705 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2707 stopparen, recursed, NULL, f,depth+1);
2710 if (max1 < minnext + deltanext)
2711 max1 = minnext + deltanext;
2712 if (deltanext == I32_MAX)
2713 is_inf = is_inf_internal = 1;
2715 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2717 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2718 if ( stopmin > minnext)
2719 stopmin = min + min1;
2720 flags &= ~SCF_DO_SUBSTR;
2722 data->flags |= SCF_SEEN_ACCEPT;
2725 if (data_fake.flags & SF_HAS_EVAL)
2726 data->flags |= SF_HAS_EVAL;
2727 data->whilem_c = data_fake.whilem_c;
2729 if (flags & SCF_DO_STCLASS)
2730 cl_or(pRExC_state, &accum, &this_class);
2732 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2734 if (flags & SCF_DO_SUBSTR) {
2735 data->pos_min += min1;
2736 data->pos_delta += max1 - min1;
2737 if (max1 != min1 || is_inf)
2738 data->longest = &(data->longest_float);
2741 delta += max1 - min1;
2742 if (flags & SCF_DO_STCLASS_OR) {
2743 cl_or(pRExC_state, data->start_class, &accum);
2745 cl_and(data->start_class, and_withp);
2746 flags &= ~SCF_DO_STCLASS;
2749 else if (flags & SCF_DO_STCLASS_AND) {
2751 cl_and(data->start_class, &accum);
2752 flags &= ~SCF_DO_STCLASS;
2755 /* Switch to OR mode: cache the old value of
2756 * data->start_class */
2758 StructCopy(data->start_class, and_withp,
2759 struct regnode_charclass_class);
2760 flags &= ~SCF_DO_STCLASS_AND;
2761 StructCopy(&accum, data->start_class,
2762 struct regnode_charclass_class);
2763 flags |= SCF_DO_STCLASS_OR;
2764 data->start_class->flags |= ANYOF_EOS;
2768 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2771 Assuming this was/is a branch we are dealing with: 'scan' now
2772 points at the item that follows the branch sequence, whatever
2773 it is. We now start at the beginning of the sequence and look
2780 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2782 If we can find such a subseqence we need to turn the first
2783 element into a trie and then add the subsequent branch exact
2784 strings to the trie.
2788 1. patterns where the whole set of branches can be converted.
2790 2. patterns where only a subset can be converted.
2792 In case 1 we can replace the whole set with a single regop
2793 for the trie. In case 2 we need to keep the start and end
2796 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2797 becomes BRANCH TRIE; BRANCH X;
2799 There is an additional case, that being where there is a
2800 common prefix, which gets split out into an EXACT like node
2801 preceding the TRIE node.
2803 If x(1..n)==tail then we can do a simple trie, if not we make
2804 a "jump" trie, such that when we match the appropriate word
2805 we "jump" to the appopriate tail node. Essentailly we turn
2806 a nested if into a case structure of sorts.
2811 if (!re_trie_maxbuff) {
2812 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2813 if (!SvIOK(re_trie_maxbuff))
2814 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2816 if ( SvIV(re_trie_maxbuff)>=0 ) {
2818 regnode *first = (regnode *)NULL;
2819 regnode *last = (regnode *)NULL;
2820 regnode *tail = scan;
2825 SV * const mysv = sv_newmortal(); /* for dumping */
2827 /* var tail is used because there may be a TAIL
2828 regop in the way. Ie, the exacts will point to the
2829 thing following the TAIL, but the last branch will
2830 point at the TAIL. So we advance tail. If we
2831 have nested (?:) we may have to move through several
2835 while ( OP( tail ) == TAIL ) {
2836 /* this is the TAIL generated by (?:) */
2837 tail = regnext( tail );
2842 regprop(RExC_rx, mysv, tail );
2843 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2844 (int)depth * 2 + 2, "",
2845 "Looking for TRIE'able sequences. Tail node is: ",
2846 SvPV_nolen_const( mysv )
2852 step through the branches, cur represents each
2853 branch, noper is the first thing to be matched
2854 as part of that branch and noper_next is the
2855 regnext() of that node. if noper is an EXACT
2856 and noper_next is the same as scan (our current
2857 position in the regex) then the EXACT branch is
2858 a possible optimization target. Once we have
2859 two or more consequetive such branches we can
2860 create a trie of the EXACT's contents and stich
2861 it in place. If the sequence represents all of
2862 the branches we eliminate the whole thing and
2863 replace it with a single TRIE. If it is a
2864 subsequence then we need to stitch it in. This
2865 means the first branch has to remain, and needs
2866 to be repointed at the item on the branch chain
2867 following the last branch optimized. This could
2868 be either a BRANCH, in which case the
2869 subsequence is internal, or it could be the
2870 item following the branch sequence in which
2871 case the subsequence is at the end.
2875 /* dont use tail as the end marker for this traverse */
2876 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2877 regnode * const noper = NEXTOPER( cur );
2878 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
2879 regnode * const noper_next = regnext( noper );
2883 regprop(RExC_rx, mysv, cur);
2884 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2885 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2887 regprop(RExC_rx, mysv, noper);
2888 PerlIO_printf( Perl_debug_log, " -> %s",
2889 SvPV_nolen_const(mysv));
2892 regprop(RExC_rx, mysv, noper_next );
2893 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2894 SvPV_nolen_const(mysv));
2896 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2897 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2899 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2900 : PL_regkind[ OP( noper ) ] == EXACT )
2901 || OP(noper) == NOTHING )
2903 && noper_next == tail
2908 if ( !first || optype == NOTHING ) {
2909 if (!first) first = cur;
2910 optype = OP( noper );
2916 Currently we do not believe that the trie logic can
2917 handle case insensitive matching properly when the
2918 pattern is not unicode (thus forcing unicode semantics).
2920 If/when this is fixed the following define can be swapped
2921 in below to fully enable trie logic.
2923 #define TRIE_TYPE_IS_SAFE 1
2926 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2928 if ( last && TRIE_TYPE_IS_SAFE ) {
2929 make_trie( pRExC_state,
2930 startbranch, first, cur, tail, count,
2933 if ( PL_regkind[ OP( noper ) ] == EXACT
2935 && noper_next == tail
2940 optype = OP( noper );
2950 regprop(RExC_rx, mysv, cur);
2951 PerlIO_printf( Perl_debug_log,
2952 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2953 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2957 if ( last && TRIE_TYPE_IS_SAFE ) {
2958 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
2959 #ifdef TRIE_STUDY_OPT
2960 if ( ((made == MADE_EXACT_TRIE &&
2961 startbranch == first)
2962 || ( first_non_open == first )) &&
2964 flags |= SCF_TRIE_RESTUDY;
2965 if ( startbranch == first
2968 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2978 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2979 scan = NEXTOPER(NEXTOPER(scan));
2980 } else /* single branch is optimized. */
2981 scan = NEXTOPER(scan);
2983 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2984 scan_frame *newframe = NULL;
2989 if (OP(scan) != SUSPEND) {
2990 /* set the pointer */
2991 if (OP(scan) == GOSUB) {
2993 RExC_recurse[ARG2L(scan)] = scan;
2994 start = RExC_open_parens[paren-1];
2995 end = RExC_close_parens[paren-1];
2998 start = RExC_rxi->program + 1;
3002 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
3003 SAVEFREEPV(recursed);
3005 if (!PAREN_TEST(recursed,paren+1)) {
3006 PAREN_SET(recursed,paren+1);
3007 Newx(newframe,1,scan_frame);
3009 if (flags & SCF_DO_SUBSTR) {
3010 SCAN_COMMIT(pRExC_state,data,minlenp);
3011 data->longest = &(data->longest_float);
3013 is_inf = is_inf_internal = 1;
3014 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3015 cl_anything(pRExC_state, data->start_class);
3016 flags &= ~SCF_DO_STCLASS;
3019 Newx(newframe,1,scan_frame);
3022 end = regnext(scan);
3027 SAVEFREEPV(newframe);
3028 newframe->next = regnext(scan);
3029 newframe->last = last;
3030 newframe->stop = stopparen;
3031 newframe->prev = frame;
3041 else if (OP(scan) == EXACT) {
3042 I32 l = STR_LEN(scan);
3045 const U8 * const s = (U8*)STRING(scan);
3046 l = utf8_length(s, s + l);
3047 uc = utf8_to_uvchr(s, NULL);
3049 uc = *((U8*)STRING(scan));
3052 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3053 /* The code below prefers earlier match for fixed
3054 offset, later match for variable offset. */
3055 if (data->last_end == -1) { /* Update the start info. */
3056 data->last_start_min = data->pos_min;
3057 data->last_start_max = is_inf
3058 ? I32_MAX : data->pos_min + data->pos_delta;
3060 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3062 SvUTF8_on(data->last_found);
3064 SV * const sv = data->last_found;
3065 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3066 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3067 if (mg && mg->mg_len >= 0)
3068 mg->mg_len += utf8_length((U8*)STRING(scan),
3069 (U8*)STRING(scan)+STR_LEN(scan));
3071 data->last_end = data->pos_min + l;
3072 data->pos_min += l; /* As in the first entry. */
3073 data->flags &= ~SF_BEFORE_EOL;
3075 if (flags & SCF_DO_STCLASS_AND) {
3076 /* Check whether it is compatible with what we know already! */
3080 /* If compatibile, we or it in below. It is compatible if is
3081 * in the bitmp and either 1) its bit or its fold is set, or 2)
3082 * it's for a locale. Even if there isn't unicode semantics
3083 * here, at runtime there may be because of matching against a
3084 * utf8 string, so accept a possible false positive for
3085 * latin1-range folds */
3087 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3088 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3089 && (!(data->start_class->flags & ANYOF_FOLD)
3090 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3093 ANYOF_CLASS_ZERO(data->start_class);
3094 ANYOF_BITMAP_ZERO(data->start_class);
3096 ANYOF_BITMAP_SET(data->start_class, uc);
3097 data->start_class->flags &= ~ANYOF_EOS;
3099 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3101 else if (flags & SCF_DO_STCLASS_OR) {
3102 /* false positive possible if the class is case-folded */
3104 ANYOF_BITMAP_SET(data->start_class, uc);
3106 data->start_class->flags |= ANYOF_UNICODE_ALL;
3107 data->start_class->flags &= ~ANYOF_EOS;
3108 cl_and(data->start_class, and_withp);
3110 flags &= ~SCF_DO_STCLASS;
3112 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3113 I32 l = STR_LEN(scan);
3114 UV uc = *((U8*)STRING(scan));
3116 /* Search for fixed substrings supports EXACT only. */
3117 if (flags & SCF_DO_SUBSTR) {
3119 SCAN_COMMIT(pRExC_state, data, minlenp);
3122 const U8 * const s = (U8 *)STRING(scan);
3123 l = utf8_length(s, s + l);
3124 uc = utf8_to_uvchr(s, NULL);
3127 if (flags & SCF_DO_SUBSTR)
3129 if (flags & SCF_DO_STCLASS_AND) {
3130 /* Check whether it is compatible with what we know already! */
3133 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3134 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3135 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3139 ANYOF_CLASS_ZERO(data->start_class);
3140 ANYOF_BITMAP_ZERO(data->start_class);
3142 ANYOF_BITMAP_SET(data->start_class, uc);
3143 data->start_class->flags &= ~ANYOF_EOS;
3144 data->start_class->flags |= ANYOF_FOLD;
3145 if (OP(scan) == EXACTFL) {
3146 data->start_class->flags |= ANYOF_LOCALE;
3150 /* Also set the other member of the fold pair. In case
3151 * that unicode semantics is called for at runtime, use
3152 * the full latin1 fold. (Can't do this for locale,
3153 * because not known until runtime */
3154 ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]);
3158 else if (flags & SCF_DO_STCLASS_OR) {
3159 if (data->start_class->flags & ANYOF_FOLD) {
3160 /* false positive possible if the class is case-folded.
3161 Assume that the locale settings are the same... */
3163 ANYOF_BITMAP_SET(data->start_class, uc);
3164 if (OP(scan) != EXACTFL) {
3166 /* And set the other member of the fold pair, but
3167 * can't do that in locale because not known until
3169 ANYOF_BITMAP_SET(data->start_class,
3170 PL_fold_latin1[uc]);
3173 data->start_class->flags &= ~ANYOF_EOS;
3175 cl_and(data->start_class, and_withp);
3177 flags &= ~SCF_DO_STCLASS;
3179 else if (REGNODE_VARIES(OP(scan))) {
3180 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3181 I32 f = flags, pos_before = 0;
3182 regnode * const oscan = scan;
3183 struct regnode_charclass_class this_class;
3184 struct regnode_charclass_class *oclass = NULL;
3185 I32 next_is_eval = 0;
3187 switch (PL_regkind[OP(scan)]) {
3188 case WHILEM: /* End of (?:...)* . */
3189 scan = NEXTOPER(scan);
3192 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3193 next = NEXTOPER(scan);
3194 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3196 maxcount = REG_INFTY;
3197 next = regnext(scan);
3198 scan = NEXTOPER(scan);
3202 if (flags & SCF_DO_SUBSTR)
3207 if (flags & SCF_DO_STCLASS) {
3209 maxcount = REG_INFTY;
3210 next = regnext(scan);
3211 scan = NEXTOPER(scan);
3214 is_inf = is_inf_internal = 1;
3215 scan = regnext(scan);
3216 if (flags & SCF_DO_SUBSTR) {
3217 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3218 data->longest = &(data->longest_float);
3220 goto optimize_curly_tail;
3222 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3223 && (scan->flags == stopparen))
3228 mincount = ARG1(scan);
3229 maxcount = ARG2(scan);
3231 next = regnext(scan);
3232 if (OP(scan) == CURLYX) {
3233 I32 lp = (data ? *(data->last_closep) : 0);
3234 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3236 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3237 next_is_eval = (OP(scan) == EVAL);
3239 if (flags & SCF_DO_SUBSTR) {
3240 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3241 pos_before = data->pos_min;
3245 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3247 data->flags |= SF_IS_INF;
3249 if (flags & SCF_DO_STCLASS) {
3250 cl_init(pRExC_state, &this_class);
3251 oclass = data->start_class;
3252 data->start_class = &this_class;
3253 f |= SCF_DO_STCLASS_AND;
3254 f &= ~SCF_DO_STCLASS_OR;
3256 /* Exclude from super-linear cache processing any {n,m}
3257 regops for which the combination of input pos and regex
3258 pos is not enough information to determine if a match
3261 For example, in the regex /foo(bar\s*){4,8}baz/ with the
3262 regex pos at the \s*, the prospects for a match depend not
3263 only on the input position but also on how many (bar\s*)
3264 repeats into the {4,8} we are. */
3265 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
3266 f &= ~SCF_WHILEM_VISITED_POS;
3268 /* This will finish on WHILEM, setting scan, or on NULL: */
3269 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3270 last, data, stopparen, recursed, NULL,
3272 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3274 if (flags & SCF_DO_STCLASS)
3275 data->start_class = oclass;
3276 if (mincount == 0 || minnext == 0) {
3277 if (flags & SCF_DO_STCLASS_OR) {
3278 cl_or(pRExC_state, data->start_class, &this_class);
3280 else if (flags & SCF_DO_STCLASS_AND) {
3281 /* Switch to OR mode: cache the old value of
3282 * data->start_class */
3284 StructCopy(data->start_class, and_withp,
3285 struct regnode_charclass_class);
3286 flags &= ~SCF_DO_STCLASS_AND;
3287 StructCopy(&this_class, data->start_class,
3288 struct regnode_charclass_class);
3289 flags |= SCF_DO_STCLASS_OR;
3290 data->start_class->flags |= ANYOF_EOS;
3292 } else { /* Non-zero len */
3293 if (flags & SCF_DO_STCLASS_OR) {
3294 cl_or(pRExC_state, data->start_class, &this_class);
3295 cl_and(data->start_class, and_withp);
3297 else if (flags & SCF_DO_STCLASS_AND)
3298 cl_and(data->start_class, &this_class);
3299 flags &= ~SCF_DO_STCLASS;
3301 if (!scan) /* It was not CURLYX, but CURLY. */
3303 if ( /* ? quantifier ok, except for (?{ ... }) */
3304 (next_is_eval || !(mincount == 0 && maxcount == 1))
3305 && (minnext == 0) && (deltanext == 0)
3306 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3307 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3309 ckWARNreg(RExC_parse,
3310 "Quantifier unexpected on zero-length expression");
3313 min += minnext * mincount;
3314 is_inf_internal |= ((maxcount == REG_INFTY
3315 && (minnext + deltanext) > 0)
3316 || deltanext == I32_MAX);
3317 is_inf |= is_inf_internal;
3318 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3320 /* Try powerful optimization CURLYX => CURLYN. */
3321 if ( OP(oscan) == CURLYX && data
3322 && data->flags & SF_IN_PAR
3323 && !(data->flags & SF_HAS_EVAL)
3324 && !deltanext && minnext == 1 ) {
3325 /* Try to optimize to CURLYN. */
3326 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3327 regnode * const nxt1 = nxt;
3334 if (!REGNODE_SIMPLE(OP(nxt))
3335 && !(PL_regkind[OP(nxt)] == EXACT
3336 && STR_LEN(nxt) == 1))
3342 if (OP(nxt) != CLOSE)
3344 if (RExC_open_parens) {
3345 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3346 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3348 /* Now we know that nxt2 is the only contents: */
3349 oscan->flags = (U8)ARG(nxt);
3351 OP(nxt1) = NOTHING; /* was OPEN. */
3354 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3355 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3356 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
3357 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3358 OP(nxt + 1) = OPTIMIZED; /* was count. */
3359 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
3364 /* Try optimization CURLYX => CURLYM. */
3365 if ( OP(oscan) == CURLYX && data
3366 && !(data->flags & SF_HAS_PAR)
3367 && !(data->flags & SF_HAS_EVAL)
3368 && !deltanext /* atom is fixed width */
3369 && minnext != 0 /* CURLYM can't handle zero width */
3371 /* XXXX How to optimize if data == 0? */
3372 /* Optimize to a simpler form. */
3373 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3377 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3378 && (OP(nxt2) != WHILEM))
3380 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3381 /* Need to optimize away parenths. */
3382 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
3383 /* Set the parenth number. */
3384 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3386 oscan->flags = (U8)ARG(nxt);
3387 if (RExC_open_parens) {
3388 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3389 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3391 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3392 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3395 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3396 OP(nxt + 1) = OPTIMIZED; /* was count. */
3397 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3398 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
3401 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3402 regnode *nnxt = regnext(nxt1);
3404 if (reg_off_by_arg[OP(nxt1)])
3405 ARG_SET(nxt1, nxt2 - nxt1);
3406 else if (nxt2 - nxt1 < U16_MAX)
3407 NEXT_OFF(nxt1) = nxt2 - nxt1;
3409 OP(nxt) = NOTHING; /* Cannot beautify */
3414 /* Optimize again: */
3415 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3416 NULL, stopparen, recursed, NULL, 0,depth+1);
3421 else if ((OP(oscan) == CURLYX)
3422 && (flags & SCF_WHILEM_VISITED_POS)
3423 /* See the comment on a similar expression above.
3424 However, this time it's not a subexpression
3425 we care about, but the expression itself. */
3426 && (maxcount == REG_INFTY)
3427 && data && ++data->whilem_c < 16) {
3428 /* This stays as CURLYX, we can put the count/of pair. */
3429 /* Find WHILEM (as in regexec.c) */
3430 regnode *nxt = oscan + NEXT_OFF(oscan);
3432 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3434 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3435 | (RExC_whilem_seen << 4)); /* On WHILEM */
3437 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3439 if (flags & SCF_DO_SUBSTR) {
3440 SV *last_str = NULL;
3441 int counted = mincount != 0;
3443 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3444 #if defined(SPARC64_GCC_WORKAROUND)
3447 const char *s = NULL;
3450 if (pos_before >= data->last_start_min)
3453 b = data->last_start_min;
3456 s = SvPV_const(data->last_found, l);
3457 old = b - data->last_start_min;
3460 I32 b = pos_before >= data->last_start_min
3461 ? pos_before : data->last_start_min;
3463 const char * const s = SvPV_const(data->last_found, l);
3464 I32 old = b - data->last_start_min;
3468 old = utf8_hop((U8*)s, old) - (U8*)s;
3470 /* Get the added string: */
3471 last_str = newSVpvn_utf8(s + old, l, UTF);
3472 if (deltanext == 0 && pos_before == b) {
3473 /* What was added is a constant string */
3475 SvGROW(last_str, (mincount * l) + 1);
3476 repeatcpy(SvPVX(last_str) + l,
3477 SvPVX_const(last_str), l, mincount - 1);
3478 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3479 /* Add additional parts. */
3480 SvCUR_set(data->last_found,
3481 SvCUR(data->last_found) - l);
3482 sv_catsv(data->last_found, last_str);
3484 SV * sv = data->last_found;
3486 SvUTF8(sv) && SvMAGICAL(sv) ?
3487 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3488 if (mg && mg->mg_len >= 0)
3489 mg->mg_len += CHR_SVLEN(last_str) - l;
3491 data->last_end += l * (mincount - 1);
3494 /* start offset must point into the last copy */
3495 data->last_start_min += minnext * (mincount - 1);
3496 data->last_start_max += is_inf ? I32_MAX
3497 : (maxcount - 1) * (minnext + data->pos_delta);
3500 /* It is counted once already... */
3501 data->pos_min += minnext * (mincount - counted);
3502 data->pos_delta += - counted * deltanext +
3503 (minnext + deltanext) * maxcount - minnext * mincount;
3504 if (mincount != maxcount) {
3505 /* Cannot extend fixed substrings found inside
3507 SCAN_COMMIT(pRExC_state,data,minlenp);
3508 if (mincount && last_str) {
3509 SV * const sv = data->last_found;
3510 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3511 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3515 sv_setsv(sv, last_str);
3516 data->last_end = data->pos_min;
3517 data->last_start_min =
3518 data->pos_min - CHR_SVLEN(last_str);
3519 data->last_start_max = is_inf
3521 : data->pos_min + data->pos_delta
3522 - CHR_SVLEN(last_str);
3524 data->longest = &(data->longest_float);
3526 SvREFCNT_dec(last_str);
3528 if (data && (fl & SF_HAS_EVAL))
3529 data->flags |= SF_HAS_EVAL;
3530 optimize_curly_tail:
3531 if (OP(oscan) != CURLYX) {
3532 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3534 NEXT_OFF(oscan) += NEXT_OFF(next);
3537 default: /* REF and CLUMP only? */
3538 if (flags & SCF_DO_SUBSTR) {
3539 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3540 data->longest = &(data->longest_float);
3542 is_inf = is_inf_internal = 1;
3543 if (flags & SCF_DO_STCLASS_OR)
3544 cl_anything(pRExC_state, data->start_class);
3545 flags &= ~SCF_DO_STCLASS;
3549 else if (OP(scan) == LNBREAK) {
3550 if (flags & SCF_DO_STCLASS) {
3552 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3553 if (flags & SCF_DO_STCLASS_AND) {
3554 for (value = 0; value < 256; value++)
3555 if (!is_VERTWS_cp(value))
3556 ANYOF_BITMAP_CLEAR(data->start_class, value);
3559 for (value = 0; value < 256; value++)
3560 if (is_VERTWS_cp(value))
3561 ANYOF_BITMAP_SET(data->start_class, value);
3563 if (flags & SCF_DO_STCLASS_OR)
3564 cl_and(data->start_class, and_withp);
3565 flags &= ~SCF_DO_STCLASS;
3569 if (flags & SCF_DO_SUBSTR) {
3570 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3572 data->pos_delta += 1;
3573 data->longest = &(data->longest_float);
3576 else if (OP(scan) == FOLDCHAR) {
3577 int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2;
3578 flags &= ~SCF_DO_STCLASS;
3581 if (flags & SCF_DO_SUBSTR) {
3582 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3584 data->pos_delta += d;
3585 data->longest = &(data->longest_float);
3588 else if (REGNODE_SIMPLE(OP(scan))) {
3591 if (flags & SCF_DO_SUBSTR) {
3592 SCAN_COMMIT(pRExC_state,data,minlenp);
3596 if (flags & SCF_DO_STCLASS) {
3597 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3599 /* Some of the logic below assumes that switching
3600 locale on will only add false positives. */
3601 switch (PL_regkind[OP(scan)]) {
3605 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3606 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3607 cl_anything(pRExC_state, data->start_class);
3610 if (OP(scan) == SANY)
3612 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3613 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3614 || ANYOF_CLASS_TEST_ANY_SET(data->start_class));
3615 cl_anything(pRExC_state, data->start_class);
3617 if (flags & SCF_DO_STCLASS_AND || !value)
3618 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3621 if (flags & SCF_DO_STCLASS_AND)
3622 cl_and(data->start_class,
3623 (struct regnode_charclass_class*)scan);
3625 cl_or(pRExC_state, data->start_class,
3626 (struct regnode_charclass_class*)scan);
3629 if (flags & SCF_DO_STCLASS_AND) {
3630 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3631 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3632 if (FLAGS(scan) & USE_UNI) {
3633 for (value = 0; value < 256; value++) {
3634 if (!isWORDCHAR_L1(value)) {
3635 ANYOF_BITMAP_CLEAR(data->start_class, value);
3639 for (value = 0; value < 256; value++) {
3640 if (!isALNUM(value)) {
3641 ANYOF_BITMAP_CLEAR(data->start_class, value);
3648 if (data->start_class->flags & ANYOF_LOCALE)
3649 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3650 else if (FLAGS(scan) & USE_UNI) {
3651 for (value = 0; value < 256; value++) {
3652 if (isWORDCHAR_L1(value)) {
3653 ANYOF_BITMAP_SET(data->start_class, value);
3657 for (value = 0; value < 256; value++) {
3658 if (isALNUM(value)) {
3659 ANYOF_BITMAP_SET(data->start_class, value);
3666 if (flags & SCF_DO_STCLASS_AND) {
3667 if (data->start_class->flags & ANYOF_LOCALE)
3668 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3671 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3672 data->start_class->flags |= ANYOF_LOCALE;
3676 if (flags & SCF_DO_STCLASS_AND) {
3677 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3678 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3679 if (FLAGS(scan) & USE_UNI) {
3680 for (value = 0; value < 256; value++) {
3681 if (isWORDCHAR_L1(value)) {
3682 ANYOF_BITMAP_CLEAR(data->start_class, value);
3686 for (value = 0; value < 256; value++) {
3687 if (isALNUM(value)) {
3688 ANYOF_BITMAP_CLEAR(data->start_class, value);
3695 if (data->start_class->flags & ANYOF_LOCALE)
3696 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3698 for (value = 0; value < 256; value++)
3699 if (!isALNUM(value))
3700 ANYOF_BITMAP_SET(data->start_class, value);
3705 if (flags & SCF_DO_STCLASS_AND) {
3706 if (data->start_class->flags & ANYOF_LOCALE)
3707 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3710 data->start_class->flags |= ANYOF_LOCALE;
3711 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3715 if (flags & SCF_DO_STCLASS_AND) {
3716 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3717 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3718 if (FLAGS(scan) & USE_UNI) {
3719 for (value = 0; value < 256; value++) {
3720 if (!isSPACE_L1(value)) {
3721 ANYOF_BITMAP_CLEAR(data->start_class, value);
3725 for (value = 0; value < 256; value++) {
3726 if (!isSPACE(value)) {
3727 ANYOF_BITMAP_CLEAR(data->start_class, value);
3734 if (data->start_class->flags & ANYOF_LOCALE) {
3735 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3737 else if (FLAGS(scan) & USE_UNI) {
3738 for (value = 0; value < 256; value++) {
3739 if (isSPACE_L1(value)) {
3740 ANYOF_BITMAP_SET(data->start_class, value);
3744 for (value = 0; value < 256; value++) {
3745 if (isSPACE(value)) {
3746 ANYOF_BITMAP_SET(data->start_class, value);
3753 if (flags & SCF_DO_STCLASS_AND) {
3754 if (data->start_class->flags & ANYOF_LOCALE)
3755 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3758 data->start_class->flags |= ANYOF_LOCALE;
3759 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3763 if (flags & SCF_DO_STCLASS_AND) {
3764 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3765 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3766 if (FLAGS(scan) & USE_UNI) {
3767 for (value = 0; value < 256; value++) {
3768 if (isSPACE_L1(value)) {
3769 ANYOF_BITMAP_CLEAR(data->start_class, value);
3773 for (value = 0; value < 256; value++) {
3774 if (isSPACE(value)) {
3775 ANYOF_BITMAP_CLEAR(data->start_class, value);
3782 if (data->start_class->flags & ANYOF_LOCALE)
3783 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3784 else if (FLAGS(scan) & USE_UNI) {
3785 for (value = 0; value < 256; value++) {
3786 if (!isSPACE_L1(value)) {
3787 ANYOF_BITMAP_SET(data->start_class, value);
3792 for (value = 0; value < 256; value++) {
3793 if (!isSPACE(value)) {
3794 ANYOF_BITMAP_SET(data->start_class, value);
3801 if (flags & SCF_DO_STCLASS_AND) {
3802 if (data->start_class->flags & ANYOF_LOCALE) {
3803 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3804 for (value = 0; value < 256; value++)
3805 if (!isSPACE(value))
3806 ANYOF_BITMAP_CLEAR(data->start_class, value);
3810 data->start_class->flags |= ANYOF_LOCALE;
3811 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3815 if (flags & SCF_DO_STCLASS_AND) {
3816 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3817 for (value = 0; value < 256; value++)
3818 if (!isDIGIT(value))
3819 ANYOF_BITMAP_CLEAR(data->start_class, value);
3822 if (data->start_class->flags & ANYOF_LOCALE)
3823 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3825 for (value = 0; value < 256; value++)
3827 ANYOF_BITMAP_SET(data->start_class, value);
3832 if (flags & SCF_DO_STCLASS_AND) {
3833 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3834 for (value = 0; value < 256; value++)
3836 ANYOF_BITMAP_CLEAR(data->start_class, value);
3839 if (data->start_class->flags & ANYOF_LOCALE)
3840 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3842 for (value = 0; value < 256; value++)
3843 if (!isDIGIT(value))
3844 ANYOF_BITMAP_SET(data->start_class, value);
3848 CASE_SYNST_FNC(VERTWS);
3849 CASE_SYNST_FNC(HORIZWS);
3852 if (flags & SCF_DO_STCLASS_OR)
3853 cl_and(data->start_class, and_withp);
3854 flags &= ~SCF_DO_STCLASS;
3857 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3858 data->flags |= (OP(scan) == MEOL
3862 else if ( PL_regkind[OP(scan)] == BRANCHJ
3863 /* Lookbehind, or need to calculate parens/evals/stclass: */
3864 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3865 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3866 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3867 || OP(scan) == UNLESSM )
3869 /* Negative Lookahead/lookbehind
3870 In this case we can't do fixed string optimisation.
3873 I32 deltanext, minnext, fake = 0;
3875 struct regnode_charclass_class intrnl;
3878 data_fake.flags = 0;
3880 data_fake.whilem_c = data->whilem_c;
3881 data_fake.last_closep = data->last_closep;
3884 data_fake.last_closep = &fake;
3885 data_fake.pos_delta = delta;
3886 if ( flags & SCF_DO_STCLASS && !scan->flags
3887 && OP(scan) == IFMATCH ) { /* Lookahead */
3888 cl_init(pRExC_state, &intrnl);
3889 data_fake.start_class = &intrnl;
3890 f |= SCF_DO_STCLASS_AND;
3892 if (flags & SCF_WHILEM_VISITED_POS)
3893 f |= SCF_WHILEM_VISITED_POS;
3894 next = regnext(scan);
3895 nscan = NEXTOPER(NEXTOPER(scan));
3896 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3897 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3900 FAIL("Variable length lookbehind not implemented");
3902 else if (minnext > (I32)U8_MAX) {
3903 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3905 scan->flags = (U8)minnext;
3908 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3910 if (data_fake.flags & SF_HAS_EVAL)
3911 data->flags |= SF_HAS_EVAL;
3912 data->whilem_c = data_fake.whilem_c;
3914 if (f & SCF_DO_STCLASS_AND) {
3915 if (flags & SCF_DO_STCLASS_OR) {
3916 /* OR before, AND after: ideally we would recurse with
3917 * data_fake to get the AND applied by study of the
3918 * remainder of the pattern, and then derecurse;
3919 * *** HACK *** for now just treat as "no information".
3920 * See [perl #56690].
3922 cl_init(pRExC_state, data->start_class);
3924 /* AND before and after: combine and continue */
3925 const int was = (data->start_class->flags & ANYOF_EOS);
3927 cl_and(data->start_class, &intrnl);
3929 data->start_class->flags |= ANYOF_EOS;
3933 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3935 /* Positive Lookahead/lookbehind
3936 In this case we can do fixed string optimisation,
3937 but we must be careful about it. Note in the case of
3938 lookbehind the positions will be offset by the minimum
3939 length of the pattern, something we won't know about
3940 until after the recurse.
3942 I32 deltanext, fake = 0;
3944 struct regnode_charclass_class intrnl;
3946 /* We use SAVEFREEPV so that when the full compile
3947 is finished perl will clean up the allocated
3948 minlens when it's all done. This way we don't
3949 have to worry about freeing them when we know
3950 they wont be used, which would be a pain.
3953 Newx( minnextp, 1, I32 );
3954 SAVEFREEPV(minnextp);
3957 StructCopy(data, &data_fake, scan_data_t);
3958 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3961 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
3962 data_fake.last_found=newSVsv(data->last_found);
3966 data_fake.last_closep = &fake;
3967 data_fake.flags = 0;
3968 data_fake.pos_delta = delta;
3970 data_fake.flags |= SF_IS_INF;
3971 if ( flags & SCF_DO_STCLASS && !scan->flags
3972 && OP(scan) == IFMATCH ) { /* Lookahead */
3973 cl_init(pRExC_state, &intrnl);
3974 data_fake.start_class = &intrnl;
3975 f |= SCF_DO_STCLASS_AND;
3977 if (flags & SCF_WHILEM_VISITED_POS)
3978 f |= SCF_WHILEM_VISITED_POS;
3979 next = regnext(scan);
3980 nscan = NEXTOPER(NEXTOPER(scan));
3982 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3983 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3986 FAIL("Variable length lookbehind not implemented");
3988 else if (*minnextp > (I32)U8_MAX) {
3989 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
3991 scan->flags = (U8)*minnextp;
3996 if (f & SCF_DO_STCLASS_AND) {
3997 const int was = (data->start_class->flags & ANYOF_EOS);
3999 cl_and(data->start_class, &intrnl);
4001 data->start_class->flags |= ANYOF_EOS;
4004 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4006 if (data_fake.flags & SF_HAS_EVAL)
4007 data->flags |= SF_HAS_EVAL;
4008 data->whilem_c = data_fake.whilem_c;
4009 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
4010 if (RExC_rx->minlen<*minnextp)
4011 RExC_rx->minlen=*minnextp;
4012 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
4013 SvREFCNT_dec(data_fake.last_found);
4015 if ( data_fake.minlen_fixed != minlenp )
4017 data->offset_fixed= data_fake.offset_fixed;
4018 data->minlen_fixed= data_fake.minlen_fixed;
4019 data->lookbehind_fixed+= scan->flags;
4021 if ( data_fake.minlen_float != minlenp )
4023 data->minlen_float= data_fake.minlen_float;
4024 data->offset_float_min=data_fake.offset_float_min;
4025 data->offset_float_max=data_fake.offset_float_max;
4026 data->lookbehind_float+= scan->flags;
4035 else if (OP(scan) == OPEN) {
4036 if (stopparen != (I32)ARG(scan))
4039 else if (OP(scan) == CLOSE) {
4040 if (stopparen == (I32)ARG(scan)) {
4043 if ((I32)ARG(scan) == is_par) {
4044 next = regnext(scan);
4046 if ( next && (OP(next) != WHILEM) && next < last)
4047 is_par = 0; /* Disable optimization */
4050 *(data->last_closep) = ARG(scan);
4052 else if (OP(scan) == EVAL) {
4054 data->flags |= SF_HAS_EVAL;
4056 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
4057 if (flags & SCF_DO_SUBSTR) {
4058 SCAN_COMMIT(pRExC_state,data,minlenp);
4059 flags &= ~SCF_DO_SUBSTR;
4061 if (data && OP(scan)==ACCEPT) {
4062 data->flags |= SCF_SEEN_ACCEPT;
4067 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
4069 if (flags & SCF_DO_SUBSTR) {
4070 SCAN_COMMIT(pRExC_state,data,minlenp);
4071 data->longest = &(data->longest_float);
4073 is_inf = is_inf_internal = 1;
4074 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
4075 cl_anything(pRExC_state, data->start_class);
4076 flags &= ~SCF_DO_STCLASS;
4078 else if (OP(scan) == GPOS) {
4079 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
4080 !(delta || is_inf || (data && data->pos_delta)))
4082 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
4083 RExC_rx->extflags |= RXf_ANCH_GPOS;
4084 if (RExC_rx->gofs < (U32)min)
4085 RExC_rx->gofs = min;
4087 RExC_rx->extflags |= RXf_GPOS_FLOAT;
4091 #ifdef TRIE_STUDY_OPT
4092 #ifdef FULL_TRIE_STUDY
4093 else if (PL_regkind[OP(scan)] == TRIE) {
4094 /* NOTE - There is similar code to this block above for handling
4095 BRANCH nodes on the initial study. If you change stuff here
4097 regnode *trie_node= scan;
4098 regnode *tail= regnext(scan);
4099 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4100 I32 max1 = 0, min1 = I32_MAX;
4101 struct regnode_charclass_class accum;
4103 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
4104 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
4105 if (flags & SCF_DO_STCLASS)
4106 cl_init_zero(pRExC_state, &accum);
4112 const regnode *nextbranch= NULL;
4115 for ( word=1 ; word <= trie->wordcount ; word++)
4117 I32 deltanext=0, minnext=0, f = 0, fake;
4118 struct regnode_charclass_class this_class;
4120 data_fake.flags = 0;
4122 data_fake.whilem_c = data->whilem_c;
4123 data_fake.last_closep = data->last_closep;
4126 data_fake.last_closep = &fake;
4127 data_fake.pos_delta = delta;
4128 if (flags & SCF_DO_STCLASS) {
4129 cl_init(pRExC_state, &this_class);
4130 data_fake.start_class = &this_class;
4131 f = SCF_DO_STCLASS_AND;
4133 if (flags & SCF_WHILEM_VISITED_POS)
4134 f |= SCF_WHILEM_VISITED_POS;
4136 if (trie->jump[word]) {
4138 nextbranch = trie_node + trie->jump[0];
4139 scan= trie_node + trie->jump[word];
4140 /* We go from the jump point to the branch that follows
4141 it. Note this means we need the vestigal unused branches
4142 even though they arent otherwise used.
4144 minnext = study_chunk(pRExC_state, &scan, minlenp,
4145 &deltanext, (regnode *)nextbranch, &data_fake,
4146 stopparen, recursed, NULL, f,depth+1);
4148 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4149 nextbranch= regnext((regnode*)nextbranch);
4151 if (min1 > (I32)(minnext + trie->minlen))
4152 min1 = minnext + trie->minlen;
4153 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4154 max1 = minnext + deltanext + trie->maxlen;
4155 if (deltanext == I32_MAX)
4156 is_inf = is_inf_internal = 1;
4158 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4160 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4161 if ( stopmin > min + min1)
4162 stopmin = min + min1;
4163 flags &= ~SCF_DO_SUBSTR;
4165 data->flags |= SCF_SEEN_ACCEPT;
4168 if (data_fake.flags & SF_HAS_EVAL)
4169 data->flags |= SF_HAS_EVAL;
4170 data->whilem_c = data_fake.whilem_c;
4172 if (flags & SCF_DO_STCLASS)
4173 cl_or(pRExC_state, &accum, &this_class);
4176 if (flags & SCF_DO_SUBSTR) {
4177 data->pos_min += min1;
4178 data->pos_delta += max1 - min1;
4179 if (max1 != min1 || is_inf)
4180 data->longest = &(data->longest_float);
4183 delta += max1 - min1;
4184 if (flags & SCF_DO_STCLASS_OR) {
4185 cl_or(pRExC_state, data->start_class, &accum);
4187 cl_and(data->start_class, and_withp);
4188 flags &= ~SCF_DO_STCLASS;
4191 else if (flags & SCF_DO_STCLASS_AND) {
4193 cl_and(data->start_class, &accum);
4194 flags &= ~SCF_DO_STCLASS;
4197 /* Switch to OR mode: cache the old value of
4198 * data->start_class */
4200 StructCopy(data->start_class, and_withp,
4201 struct regnode_charclass_class);
4202 flags &= ~SCF_DO_STCLASS_AND;
4203 StructCopy(&accum, data->start_class,
4204 struct regnode_charclass_class);
4205 flags |= SCF_DO_STCLASS_OR;
4206 data->start_class->flags |= ANYOF_EOS;
4213 else if (PL_regkind[OP(scan)] == TRIE) {
4214 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4217 min += trie->minlen;
4218 delta += (trie->maxlen - trie->minlen);
4219 flags &= ~SCF_DO_STCLASS; /* xxx */
4220 if (flags & SCF_DO_SUBSTR) {
4221 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4222 data->pos_min += trie->minlen;
4223 data->pos_delta += (trie->maxlen - trie->minlen);
4224 if (trie->maxlen != trie->minlen)
4225 data->longest = &(data->longest_float);
4227 if (trie->jump) /* no more substrings -- for now /grr*/
4228 flags &= ~SCF_DO_SUBSTR;
4230 #endif /* old or new */
4231 #endif /* TRIE_STUDY_OPT */
4233 /* Else: zero-length, ignore. */
4234 scan = regnext(scan);
4239 stopparen = frame->stop;
4240 frame = frame->prev;
4241 goto fake_study_recurse;
4246 DEBUG_STUDYDATA("pre-fin:",data,depth);
4249 *deltap = is_inf_internal ? I32_MAX : delta;
4250 if (flags & SCF_DO_SUBSTR && is_inf)
4251 data->pos_delta = I32_MAX - data->pos_min;
4252 if (is_par > (I32)U8_MAX)
4254 if (is_par && pars==1 && data) {
4255 data->flags |= SF_IN_PAR;
4256 data->flags &= ~SF_HAS_PAR;
4258 else if (pars && data) {
4259 data->flags |= SF_HAS_PAR;
4260 data->flags &= ~SF_IN_PAR;
4262 if (flags & SCF_DO_STCLASS_OR)
4263 cl_and(data->start_class, and_withp);
4264 if (flags & SCF_TRIE_RESTUDY)
4265 data->flags |= SCF_TRIE_RESTUDY;
4267 DEBUG_STUDYDATA("post-fin:",data,depth);
4269 return min < stopmin ? min : stopmin;
4273 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4275 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4277 PERL_ARGS_ASSERT_ADD_DATA;
4279 Renewc(RExC_rxi->data,
4280 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4281 char, struct reg_data);
4283 Renew(RExC_rxi->data->what, count + n, U8);
4285 Newx(RExC_rxi->data->what, n, U8);
4286 RExC_rxi->data->count = count + n;
4287 Copy(s, RExC_rxi->data->what + count, n, U8);
4291 /*XXX: todo make this not included in a non debugging perl */
4292 #ifndef PERL_IN_XSUB_RE
4294 Perl_reginitcolors(pTHX)
4297 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4299 char *t = savepv(s);
4303 t = strchr(t, '\t');
4309 PL_colors[i] = t = (char *)"";
4314 PL_colors[i++] = (char *)"";
4321 #ifdef TRIE_STUDY_OPT
4322 #define CHECK_RESTUDY_GOTO \
4324 (data.flags & SCF_TRIE_RESTUDY) \
4328 #define CHECK_RESTUDY_GOTO
4332 - pregcomp - compile a regular expression into internal code
4334 * We can't allocate space until we know how big the compiled form will be,
4335 * but we can't compile it (and thus know how big it is) until we've got a
4336 * place to put the code. So we cheat: we compile it twice, once with code
4337 * generation turned off and size counting turned on, and once "for real".
4338 * This also means that we don't allocate space until we are sure that the
4339 * thing really will compile successfully, and we never have to move the
4340 * code and thus invalidate pointers into it. (Note that it has to be in
4341 * one piece because free() must be able to free it all.) [NB: not true in perl]
4343 * Beware that the optimization-preparation code in here knows about some
4344 * of the structure of the compiled regexp. [I'll say.]
4349 #ifndef PERL_IN_XSUB_RE
4350 #define RE_ENGINE_PTR &reh_regexp_engine
4352 extern const struct regexp_engine my_reg_engine;
4353 #define RE_ENGINE_PTR &my_reg_engine
4356 #ifndef PERL_IN_XSUB_RE
4358 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4361 HV * const table = GvHV(PL_hintgv);
4363 PERL_ARGS_ASSERT_PREGCOMP;
4365 /* Dispatch a request to compile a regexp to correct
4368 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4369 GET_RE_DEBUG_FLAGS_DECL;
4370 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4371 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4373 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4376 return CALLREGCOMP_ENG(eng, pattern, flags);
4379 return Perl_re_compile(aTHX_ pattern, flags);
4384 Perl_re_compile(pTHX_ SV * const pattern, U32 orig_pm_flags)
4389 register regexp_internal *ri;
4398 /* these are all flags - maybe they should be turned
4399 * into a single int with different bit masks */
4400 I32 sawlookahead = 0;
4403 bool used_setjump = FALSE;
4408 RExC_state_t RExC_state;
4409 RExC_state_t * const pRExC_state = &RExC_state;
4410 #ifdef TRIE_STUDY_OPT
4412 RExC_state_t copyRExC_state;
4414 GET_RE_DEBUG_FLAGS_DECL;
4416 PERL_ARGS_ASSERT_RE_COMPILE;
4418 DEBUG_r(if (!PL_colorset) reginitcolors());
4420 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4422 /****************** LONG JUMP TARGET HERE***********************/
4423 /* Longjmp back to here if have to switch in midstream to utf8 */
4424 if (! RExC_orig_utf8) {
4425 JMPENV_PUSH(jump_ret);
4426 used_setjump = TRUE;
4429 if (jump_ret == 0) { /* First time through */
4430 exp = SvPV(pattern, plen);
4432 /* ignore the utf8ness if the pattern is 0 length */
4434 RExC_utf8 = RExC_orig_utf8 = 0;
4438 SV *dsv= sv_newmortal();
4439 RE_PV_QUOTED_DECL(s, RExC_utf8,
4440 dsv, exp, plen, 60);
4441 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4442 PL_colors[4],PL_colors[5],s);
4445 else { /* longjumped back */
4448 /* If the cause for the longjmp was other than changing to utf8, pop
4449 * our own setjmp, and longjmp to the correct handler */
4450 if (jump_ret != UTF8_LONGJMP) {
4452 JMPENV_JUMP(jump_ret);
4457 /* It's possible to write a regexp in ascii that represents Unicode
4458 codepoints outside of the byte range, such as via \x{100}. If we
4459 detect such a sequence we have to convert the entire pattern to utf8
4460 and then recompile, as our sizing calculation will have been based
4461 on 1 byte == 1 character, but we will need to use utf8 to encode
4462 at least some part of the pattern, and therefore must convert the whole
4465 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4466 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4467 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4469 RExC_orig_utf8 = RExC_utf8 = 1;
4473 #ifdef TRIE_STUDY_OPT
4477 /* Set to use unicode semantics if the pattern is in utf8 and has the
4478 * 'dual' charset specified, as it means unicode when utf8 */
4479 pm_flags = orig_pm_flags;
4480 if (RExC_utf8 && ! (pm_flags & (RXf_PMf_LOCALE|RXf_PMf_UNICODE))) {
4481 pm_flags |= RXf_PMf_UNICODE;
4485 RExC_flags = pm_flags;
4489 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4490 RExC_seen_evals = 0;
4493 /* First pass: determine size, legality. */
4501 RExC_emit = &PL_regdummy;
4502 RExC_whilem_seen = 0;
4503 RExC_open_parens = NULL;
4504 RExC_close_parens = NULL;
4506 RExC_paren_names = NULL;
4508 RExC_paren_name_list = NULL;
4510 RExC_recurse = NULL;
4511 RExC_recurse_count = 0;
4513 #if 0 /* REGC() is (currently) a NOP at the first pass.
4514 * Clever compilers notice this and complain. --jhi */
4515 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4517 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4518 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4519 RExC_precomp = NULL;
4523 /* Here, finished first pass. Get rid of any added setjmp */
4528 PerlIO_printf(Perl_debug_log,
4529 "Required size %"IVdf" nodes\n"
4530 "Starting second pass (creation)\n",
4533 RExC_lastparse=NULL;
4535 /* Small enough for pointer-storage convention?
4536 If extralen==0, this means that we will not need long jumps. */
4537 if (RExC_size >= 0x10000L && RExC_extralen)
4538 RExC_size += RExC_extralen;
4541 if (RExC_whilem_seen > 15)
4542 RExC_whilem_seen = 15;
4544 /* Allocate space and zero-initialize. Note, the two step process
4545 of zeroing when in debug mode, thus anything assigned has to
4546 happen after that */
4547 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4548 r = (struct regexp*)SvANY(rx);
4549 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4550 char, regexp_internal);
4551 if ( r == NULL || ri == NULL )
4552 FAIL("Regexp out of space");
4554 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4555 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4557 /* bulk initialize base fields with 0. */
4558 Zero(ri, sizeof(regexp_internal), char);
4561 /* non-zero initialization begins here */
4563 r->engine= RE_ENGINE_PTR;
4564 r->extflags = pm_flags;
4566 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4567 bool has_charset = cBOOL(r->extflags & (RXf_PMf_LOCALE|RXf_PMf_UNICODE));
4569 /* The caret is output if there are any defaults: if not all the STD
4570 * flags are set, or if no character set specifier is needed */
4572 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4574 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4575 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4576 >> RXf_PMf_STD_PMMOD_SHIFT);
4577 const char *fptr = STD_PAT_MODS; /*"msix"*/
4579 /* Allocate for the worst case, which is all the std flags are turned
4580 * on. If more precision is desired, we could do a population count of
4581 * the flags set. This could be done with a small lookup table, or by
4582 * shifting, masking and adding, or even, when available, assembly
4583 * language for a machine-language population count.
4584 * We never output a minus, as all those are defaults, so are
4585 * covered by the caret */
4586 const STRLEN wraplen = plen + has_p + has_runon
4587 + has_default /* If needs a caret */
4588 + has_charset /* If needs a character set specifier */
4589 + (sizeof(STD_PAT_MODS) - 1)
4590 + (sizeof("(?:)") - 1);
4592 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4594 SvFLAGS(rx) |= SvUTF8(pattern);
4597 /* If a default, cover it using the caret */
4599 *p++= DEFAULT_PAT_MOD;
4602 if (r->extflags & RXf_PMf_LOCALE) {
4603 *p++ = LOCALE_PAT_MOD;
4605 *p++ = UNICODE_PAT_MOD;
4609 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4612 while((ch = *fptr++)) {
4620 Copy(RExC_precomp, p, plen, char);
4621 assert ((RX_WRAPPED(rx) - p) < 16);
4622 r->pre_prefix = p - RX_WRAPPED(rx);
4628 SvCUR_set(rx, p - SvPVX_const(rx));
4632 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4634 if (RExC_seen & REG_SEEN_RECURSE) {
4635 Newxz(RExC_open_parens, RExC_npar,regnode *);
4636 SAVEFREEPV(RExC_open_parens);
4637 Newxz(RExC_close_parens,RExC_npar,regnode *);
4638 SAVEFREEPV(RExC_close_parens);
4641 /* Useful during FAIL. */
4642 #ifdef RE_TRACK_PATTERN_OFFSETS
4643 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4644 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4645 "%s %"UVuf" bytes for offset annotations.\n",
4646 ri->u.offsets ? "Got" : "Couldn't get",
4647 (UV)((2*RExC_size+1) * sizeof(U32))));
4649 SetProgLen(ri,RExC_size);
4653 REH_CALL_COMP_BEGIN_HOOK(pRExC_state->rx);
4655 /* Second pass: emit code. */
4656 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4661 RExC_emit_start = ri->program;
4662 RExC_emit = ri->program;
4663 RExC_emit_bound = ri->program + RExC_size + 1;
4665 /* Store the count of eval-groups for security checks: */
4666 RExC_rx->seen_evals = RExC_seen_evals;
4667 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4668 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4672 /* XXXX To minimize changes to RE engine we always allocate
4673 3-units-long substrs field. */
4674 Newx(r->substrs, 1, struct reg_substr_data);
4675 if (RExC_recurse_count) {
4676 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4677 SAVEFREEPV(RExC_recurse);
4681 r->minlen = minlen = sawlookahead = sawplus = sawopen = 0;
4682 Zero(r->substrs, 1, struct reg_substr_data);
4684 #ifdef TRIE_STUDY_OPT
4686 StructCopy(&zero_scan_data, &data, scan_data_t);
4687 copyRExC_state = RExC_state;
4690 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4692 RExC_state = copyRExC_state;
4693 if (seen & REG_TOP_LEVEL_BRANCHES)
4694 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4696 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4697 if (data.last_found) {
4698 SvREFCNT_dec(data.longest_fixed);
4699 SvREFCNT_dec(data.longest_float);
4700 SvREFCNT_dec(data.last_found);
4702 StructCopy(&zero_scan_data, &data, scan_data_t);
4705 StructCopy(&zero_scan_data, &data, scan_data_t);
4708 /* Dig out information for optimizations. */
4709 r->extflags = RExC_flags; /* was pm_op */
4710 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4713 SvUTF8_on(rx); /* Unicode in it? */
4714 ri->regstclass = NULL;
4715 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4716 r->intflags |= PREGf_NAUGHTY;
4717 scan = ri->program + 1; /* First BRANCH. */
4719 /* testing for BRANCH here tells us whether there is "must appear"
4720 data in the pattern. If there is then we can use it for optimisations */
4721 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4723 STRLEN longest_float_length, longest_fixed_length;
4724 struct regnode_charclass_class ch_class; /* pointed to by data */
4726 I32 last_close = 0; /* pointed to by data */
4727 regnode *first= scan;
4728 regnode *first_next= regnext(first);
4730 * Skip introductions and multiplicators >= 1
4731 * so that we can extract the 'meat' of the pattern that must
4732 * match in the large if() sequence following.
4733 * NOTE that EXACT is NOT covered here, as it is normally
4734 * picked up by the optimiser separately.
4736 * This is unfortunate as the optimiser isnt handling lookahead
4737 * properly currently.
4740 while ((OP(first) == OPEN && (sawopen = 1)) ||
4741 /* An OR of *one* alternative - should not happen now. */
4742 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4743 /* for now we can't handle lookbehind IFMATCH*/
4744 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
4745 (OP(first) == PLUS) ||
4746 (OP(first) == MINMOD) ||
4747 /* An {n,m} with n>0 */
4748 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4749 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4752 * the only op that could be a regnode is PLUS, all the rest
4753 * will be regnode_1 or regnode_2.
4756 if (OP(first) == PLUS)
4759 first += regarglen[OP(first)];
4761 first = NEXTOPER(first);
4762 first_next= regnext(first);
4765 /* Starting-point info. */
4767 DEBUG_PEEP("first:",first,0);
4768 /* Ignore EXACT as we deal with it later. */
4769 if (PL_regkind[OP(first)] == EXACT) {
4770 if (OP(first) == EXACT)
4771 NOOP; /* Empty, get anchored substr later. */
4773 ri->regstclass = first;
4776 else if (PL_regkind[OP(first)] == TRIE &&
4777 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4780 /* this can happen only on restudy */
4781 if ( OP(first) == TRIE ) {
4782 struct regnode_1 *trieop = (struct regnode_1 *)
4783 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4784 StructCopy(first,trieop,struct regnode_1);
4785 trie_op=(regnode *)trieop;
4787 struct regnode_charclass *trieop = (struct regnode_charclass *)
4788 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4789 StructCopy(first,trieop,struct regnode_charclass);
4790 trie_op=(regnode *)trieop;
4793 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4794 ri->regstclass = trie_op;
4797 else if (REGNODE_SIMPLE(OP(first)))
4798 ri->regstclass = first;
4799 else if (PL_regkind[OP(first)] == BOUND ||
4800 PL_regkind[OP(first)] == NBOUND)
4801 ri->regstclass = first;
4802 else if (PL_regkind[OP(first)] == BOL) {
4803 r->extflags |= (OP(first) == MBOL
4805 : (OP(first) == SBOL
4808 first = NEXTOPER(first);
4811 else if (OP(first) == GPOS) {
4812 r->extflags |= RXf_ANCH_GPOS;
4813 first = NEXTOPER(first);
4816 else if ((!sawopen || !RExC_sawback) &&
4817 (OP(first) == STAR &&
4818 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4819 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4821 /* turn .* into ^.* with an implied $*=1 */
4823 (OP(NEXTOPER(first)) == REG_ANY)
4826 r->extflags |= type;
4827 r->intflags |= PREGf_IMPLICIT;
4828 first = NEXTOPER(first);
4831 if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback)
4832 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4833 /* x+ must match at the 1st pos of run of x's */
4834 r->intflags |= PREGf_SKIP;
4836 /* Scan is after the zeroth branch, first is atomic matcher. */
4837 #ifdef TRIE_STUDY_OPT
4840 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4841 (IV)(first - scan + 1))
4845 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4846 (IV)(first - scan + 1))
4852 * If there's something expensive in the r.e., find the
4853 * longest literal string that must appear and make it the
4854 * regmust. Resolve ties in favor of later strings, since
4855 * the regstart check works with the beginning of the r.e.
4856 * and avoiding duplication strengthens checking. Not a
4857 * strong reason, but sufficient in the absence of others.
4858 * [Now we resolve ties in favor of the earlier string if
4859 * it happens that c_offset_min has been invalidated, since the
4860 * earlier string may buy us something the later one won't.]
4863 data.longest_fixed = newSVpvs("");
4864 data.longest_float = newSVpvs("");
4865 data.last_found = newSVpvs("");
4866 data.longest = &(data.longest_fixed);
4868 if (!ri->regstclass) {
4869 cl_init(pRExC_state, &ch_class);
4870 data.start_class = &ch_class;
4871 stclass_flag = SCF_DO_STCLASS_AND;
4872 } else /* XXXX Check for BOUND? */
4874 data.last_closep = &last_close;
4876 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
4877 &data, -1, NULL, NULL,
4878 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
4884 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
4885 && data.last_start_min == 0 && data.last_end > 0
4886 && !RExC_seen_zerolen
4887 && !(RExC_seen & REG_SEEN_VERBARG)
4888 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
4889 r->extflags |= RXf_CHECK_ALL;
4890 scan_commit(pRExC_state, &data,&minlen,0);
4891 SvREFCNT_dec(data.last_found);
4893 /* Note that code very similar to this but for anchored string
4894 follows immediately below, changes may need to be made to both.
4897 longest_float_length = CHR_SVLEN(data.longest_float);
4898 if (longest_float_length
4899 || (data.flags & SF_FL_BEFORE_EOL
4900 && (!(data.flags & SF_FL_BEFORE_MEOL)
4901 || (RExC_flags & RXf_PMf_MULTILINE))))
4905 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
4906 && data.offset_fixed == data.offset_float_min
4907 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
4908 goto remove_float; /* As in (a)+. */
4910 /* copy the information about the longest float from the reg_scan_data
4911 over to the program. */
4912 if (SvUTF8(data.longest_float)) {
4913 r->float_utf8 = data.longest_float;
4914 r->float_substr = NULL;
4916 r->float_substr = data.longest_float;
4917 r->float_utf8 = NULL;
4919 /* float_end_shift is how many chars that must be matched that
4920 follow this item. We calculate it ahead of time as once the
4921 lookbehind offset is added in we lose the ability to correctly
4923 ml = data.minlen_float ? *(data.minlen_float)
4924 : (I32)longest_float_length;
4925 r->float_end_shift = ml - data.offset_float_min
4926 - longest_float_length + (SvTAIL(data.longest_float) != 0)
4927 + data.lookbehind_float;
4928 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
4929 r->float_max_offset = data.offset_float_max;
4930 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
4931 r->float_max_offset -= data.lookbehind_float;
4933 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
4934 && (!(data.flags & SF_FL_BEFORE_MEOL)
4935 || (RExC_flags & RXf_PMf_MULTILINE)));
4936 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
4940 r->float_substr = r->float_utf8 = NULL;
4941 SvREFCNT_dec(data.longest_float);
4942 longest_float_length = 0;
4945 /* Note that code very similar to this but for floating string
4946 is immediately above, changes may need to be made to both.
4949 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
4950 if (longest_fixed_length
4951 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
4952 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4953 || (RExC_flags & RXf_PMf_MULTILINE))))
4957 /* copy the information about the longest fixed
4958 from the reg_scan_data over to the program. */
4959 if (SvUTF8(data.longest_fixed)) {
4960 r->anchored_utf8 = data.longest_fixed;
4961 r->anchored_substr = NULL;
4963 r->anchored_substr = data.longest_fixed;
4964 r->anchored_utf8 = NULL;
4966 /* fixed_end_shift is how many chars that must be matched that
4967 follow this item. We calculate it ahead of time as once the
4968 lookbehind offset is added in we lose the ability to correctly
4970 ml = data.minlen_fixed ? *(data.minlen_fixed)
4971 : (I32)longest_fixed_length;
4972 r->anchored_end_shift = ml - data.offset_fixed
4973 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
4974 + data.lookbehind_fixed;
4975 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
4977 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
4978 && (!(data.flags & SF_FIX_BEFORE_MEOL)
4979 || (RExC_flags & RXf_PMf_MULTILINE)));
4980 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
4983 r->anchored_substr = r->anchored_utf8 = NULL;
4984 SvREFCNT_dec(data.longest_fixed);
4985 longest_fixed_length = 0;
4988 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
4989 ri->regstclass = NULL;
4990 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
4992 && !(data.start_class->flags & ANYOF_EOS)
4993 && !cl_is_anything(data.start_class))
4995 const U32 n = add_data(pRExC_state, 1, "f");
4997 Newx(RExC_rxi->data->data[n], 1,
4998 struct regnode_charclass_class);
4999 StructCopy(data.start_class,
5000 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5001 struct regnode_charclass_class);
5002 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5003 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5004 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
5005 regprop(r, sv, (regnode*)data.start_class);
5006 PerlIO_printf(Perl_debug_log,
5007 "synthetic stclass \"%s\".\n",
5008 SvPVX_const(sv));});
5011 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
5012 if (longest_fixed_length > longest_float_length) {
5013 r->check_end_shift = r->anchored_end_shift;
5014 r->check_substr = r->anchored_substr;
5015 r->check_utf8 = r->anchored_utf8;
5016 r->check_offset_min = r->check_offset_max = r->anchored_offset;
5017 if (r->extflags & RXf_ANCH_SINGLE)
5018 r->extflags |= RXf_NOSCAN;
5021 r->check_end_shift = r->float_end_shift;
5022 r->check_substr = r->float_substr;
5023 r->check_utf8 = r->float_utf8;
5024 r->check_offset_min = r->float_min_offset;
5025 r->check_offset_max = r->float_max_offset;
5027 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
5028 This should be changed ASAP! */
5029 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
5030 r->extflags |= RXf_USE_INTUIT;
5031 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
5032 r->extflags |= RXf_INTUIT_TAIL;
5034 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
5035 if ( (STRLEN)minlen < longest_float_length )
5036 minlen= longest_float_length;
5037 if ( (STRLEN)minlen < longest_fixed_length )
5038 minlen= longest_fixed_length;
5042 /* Several toplevels. Best we can is to set minlen. */
5044 struct regnode_charclass_class ch_class;
5047 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
5049 scan = ri->program + 1;
5050 cl_init(pRExC_state, &ch_class);
5051 data.start_class = &ch_class;
5052 data.last_closep = &last_close;
5055 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
5056 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
5060 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
5061 = r->float_substr = r->float_utf8 = NULL;
5062 if (!(data.start_class->flags & ANYOF_EOS)
5063 && !cl_is_anything(data.start_class))
5065 const U32 n = add_data(pRExC_state, 1, "f");
5067 Newx(RExC_rxi->data->data[n], 1,
5068 struct regnode_charclass_class);
5069 StructCopy(data.start_class,
5070 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5071 struct regnode_charclass_class);
5072 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5073 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5074 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
5075 regprop(r, sv, (regnode*)data.start_class);
5076 PerlIO_printf(Perl_debug_log,
5077 "synthetic stclass \"%s\".\n",
5078 SvPVX_const(sv));});
5082 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
5083 the "real" pattern. */
5085 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
5086 (IV)minlen, (IV)r->minlen);
5088 r->minlenret = minlen;
5089 if (r->minlen < minlen)
5092 if (RExC_seen & REG_SEEN_GPOS)
5093 r->extflags |= RXf_GPOS_SEEN;
5094 if (RExC_seen & REG_SEEN_LOOKBEHIND)
5095 r->extflags |= RXf_LOOKBEHIND_SEEN;
5096 if (RExC_seen & REG_SEEN_EVAL)
5097 r->extflags |= RXf_EVAL_SEEN;
5098 if (RExC_seen & REG_SEEN_CANY)
5099 r->extflags |= RXf_CANY_SEEN;
5100 if (RExC_seen & REG_SEEN_VERBARG)
5101 r->intflags |= PREGf_VERBARG_SEEN;
5102 if (RExC_seen & REG_SEEN_CUTGROUP)
5103 r->intflags |= PREGf_CUTGROUP_SEEN;
5104 if (RExC_paren_names)
5105 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
5107 RXp_PAREN_NAMES(r) = NULL;
5109 #ifdef STUPID_PATTERN_CHECKS
5110 if (RX_PRELEN(rx) == 0)
5111 r->extflags |= RXf_NULL;
5112 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5113 /* XXX: this should happen BEFORE we compile */
5114 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5115 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
5116 r->extflags |= RXf_WHITE;
5117 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
5118 r->extflags |= RXf_START_ONLY;
5120 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5121 /* XXX: this should happen BEFORE we compile */
5122 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5124 regnode *first = ri->program + 1;
5126 U8 nop = OP(NEXTOPER(first));
5128 if (PL_regkind[fop] == NOTHING && nop == END)
5129 r->extflags |= RXf_NULL;
5130 else if (PL_regkind[fop] == BOL && nop == END)
5131 r->extflags |= RXf_START_ONLY;
5132 else if (fop == PLUS && nop ==SPACE && OP(regnext(first))==END)
5133 r->extflags |= RXf_WHITE;
5137 if (RExC_paren_names) {
5138 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5139 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5142 ri->name_list_idx = 0;
5144 if (RExC_recurse_count) {
5145 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5146 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5147 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5150 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5151 /* assume we don't need to swap parens around before we match */
5154 PerlIO_printf(Perl_debug_log,"Final program:\n");
5157 #ifdef RE_TRACK_PATTERN_OFFSETS
5158 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5159 const U32 len = ri->u.offsets[0];
5161 GET_RE_DEBUG_FLAGS_DECL;
5162 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5163 for (i = 1; i <= len; i++) {
5164 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5165 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5166 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5168 PerlIO_printf(Perl_debug_log, "\n");
5174 #undef RE_ENGINE_PTR
5178 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5181 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5183 PERL_UNUSED_ARG(value);
5185 if (flags & RXapif_FETCH) {
5186 return reg_named_buff_fetch(rx, key, flags);
5187 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5188 Perl_croak_no_modify(aTHX);
5190 } else if (flags & RXapif_EXISTS) {
5191 return reg_named_buff_exists(rx, key, flags)
5194 } else if (flags & RXapif_REGNAMES) {
5195 return reg_named_buff_all(rx, flags);
5196 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5197 return reg_named_buff_scalar(rx, flags);
5199 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5205 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5208 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5209 PERL_UNUSED_ARG(lastkey);
5211 if (flags & RXapif_FIRSTKEY)
5212 return reg_named_buff_firstkey(rx, flags);
5213 else if (flags & RXapif_NEXTKEY)
5214 return reg_named_buff_nextkey(rx, flags);
5216 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5222 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5225 AV *retarray = NULL;
5227 struct regexp *const rx = (struct regexp *)SvANY(r);
5229 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5231 if (flags & RXapif_ALL)
5234 if (rx && RXp_PAREN_NAMES(rx)) {
5235 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5238 SV* sv_dat=HeVAL(he_str);
5239 I32 *nums=(I32*)SvPVX(sv_dat);
5240 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5241 if ((I32)(rx->nparens) >= nums[i]
5242 && rx->offs[nums[i]].start != -1
5243 && rx->offs[nums[i]].end != -1)
5246 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5250 ret = newSVsv(&PL_sv_undef);
5253 av_push(retarray, ret);
5256 return newRV_noinc(MUTABLE_SV(retarray));
5263 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5266 struct regexp *const rx = (struct regexp *)SvANY(r);
5268 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5270 if (rx && RXp_PAREN_NAMES(rx)) {
5271 if (flags & RXapif_ALL) {
5272 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5274 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5288 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5290 struct regexp *const rx = (struct regexp *)SvANY(r);
5292 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5294 if ( rx && RXp_PAREN_NAMES(rx) ) {
5295 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5297 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5304 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5306 struct regexp *const rx = (struct regexp *)SvANY(r);
5307 GET_RE_DEBUG_FLAGS_DECL;
5309 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5311 if (rx && RXp_PAREN_NAMES(rx)) {
5312 HV *hv = RXp_PAREN_NAMES(rx);
5314 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5317 SV* sv_dat = HeVAL(temphe);
5318 I32 *nums = (I32*)SvPVX(sv_dat);
5319 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5320 if ((I32)(rx->lastparen) >= nums[i] &&
5321 rx->offs[nums[i]].start != -1 &&
5322 rx->offs[nums[i]].end != -1)
5328 if (parno || flags & RXapif_ALL) {
5329 return newSVhek(HeKEY_hek(temphe));
5337 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5342 struct regexp *const rx = (struct regexp *)SvANY(r);
5344 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5346 if (rx && RXp_PAREN_NAMES(rx)) {
5347 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5348 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5349 } else if (flags & RXapif_ONE) {
5350 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5351 av = MUTABLE_AV(SvRV(ret));
5352 length = av_len(av);
5354 return newSViv(length + 1);
5356 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5360 return &PL_sv_undef;
5364 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5366 struct regexp *const rx = (struct regexp *)SvANY(r);
5369 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5371 if (rx && RXp_PAREN_NAMES(rx)) {
5372 HV *hv= RXp_PAREN_NAMES(rx);
5374 (void)hv_iterinit(hv);
5375 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5378 SV* sv_dat = HeVAL(temphe);
5379 I32 *nums = (I32*)SvPVX(sv_dat);
5380 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5381 if ((I32)(rx->lastparen) >= nums[i] &&
5382 rx->offs[nums[i]].start != -1 &&
5383 rx->offs[nums[i]].end != -1)
5389 if (parno || flags & RXapif_ALL) {
5390 av_push(av, newSVhek(HeKEY_hek(temphe)));
5395 return newRV_noinc(MUTABLE_SV(av));
5399 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5402 struct regexp *const rx = (struct regexp *)SvANY(r);
5407 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5410 sv_setsv(sv,&PL_sv_undef);
5414 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5416 i = rx->offs[0].start;
5420 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5422 s = rx->subbeg + rx->offs[0].end;
5423 i = rx->sublen - rx->offs[0].end;
5426 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5427 (s1 = rx->offs[paren].start) != -1 &&
5428 (t1 = rx->offs[paren].end) != -1)
5432 s = rx->subbeg + s1;
5434 sv_setsv(sv,&PL_sv_undef);
5437 assert(rx->sublen >= (s - rx->subbeg) + i );
5439 const int oldtainted = PL_tainted;
5441 sv_setpvn(sv, s, i);
5442 PL_tainted = oldtainted;
5443 if ( (rx->extflags & RXf_CANY_SEEN)
5444 ? (RXp_MATCH_UTF8(rx)
5445 && (!i || is_utf8_string((U8*)s, i)))
5446 : (RXp_MATCH_UTF8(rx)) )
5453 if (RXp_MATCH_TAINTED(rx)) {
5454 if (SvTYPE(sv) >= SVt_PVMG) {
5455 MAGIC* const mg = SvMAGIC(sv);
5458 SvMAGIC_set(sv, mg->mg_moremagic);
5460 if ((mgt = SvMAGIC(sv))) {
5461 mg->mg_moremagic = mgt;
5462 SvMAGIC_set(sv, mg);
5472 sv_setsv(sv,&PL_sv_undef);
5478 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5479 SV const * const value)
5481 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5483 PERL_UNUSED_ARG(rx);
5484 PERL_UNUSED_ARG(paren);
5485 PERL_UNUSED_ARG(value);
5488 Perl_croak_no_modify(aTHX);
5492 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5495 struct regexp *const rx = (struct regexp *)SvANY(r);
5499 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5501 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5503 /* $` / ${^PREMATCH} */
5504 case RX_BUFF_IDX_PREMATCH:
5505 if (rx->offs[0].start != -1) {
5506 i = rx->offs[0].start;
5514 /* $' / ${^POSTMATCH} */
5515 case RX_BUFF_IDX_POSTMATCH:
5516 if (rx->offs[0].end != -1) {
5517 i = rx->sublen - rx->offs[0].end;
5519 s1 = rx->offs[0].end;
5525 /* $& / ${^MATCH}, $1, $2, ... */
5527 if (paren <= (I32)rx->nparens &&
5528 (s1 = rx->offs[paren].start) != -1 &&
5529 (t1 = rx->offs[paren].end) != -1)
5534 if (ckWARN(WARN_UNINITIALIZED))
5535 report_uninit((const SV *)sv);
5540 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5541 const char * const s = rx->subbeg + s1;
5546 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5553 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5555 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5556 PERL_UNUSED_ARG(rx);
5560 return newSVpvs("Regexp");
5563 /* Scans the name of a named buffer from the pattern.
5564 * If flags is REG_RSN_RETURN_NULL returns null.
5565 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5566 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5567 * to the parsed name as looked up in the RExC_paren_names hash.
5568 * If there is an error throws a vFAIL().. type exception.
5571 #define REG_RSN_RETURN_NULL 0
5572 #define REG_RSN_RETURN_NAME 1
5573 #define REG_RSN_RETURN_DATA 2
5576 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5578 char *name_start = RExC_parse;
5580 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5582 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5583 /* skip IDFIRST by using do...while */
5586 RExC_parse += UTF8SKIP(RExC_parse);
5587 } while (isALNUM_utf8((U8*)RExC_parse));
5591 } while (isALNUM(*RExC_parse));
5596 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5597 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5598 if ( flags == REG_RSN_RETURN_NAME)
5600 else if (flags==REG_RSN_RETURN_DATA) {
5603 if ( ! sv_name ) /* should not happen*/
5604 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5605 if (RExC_paren_names)
5606 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5608 sv_dat = HeVAL(he_str);
5610 vFAIL("Reference to nonexistent named group");
5614 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5621 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5622 int rem=(int)(RExC_end - RExC_parse); \
5631 if (RExC_lastparse!=RExC_parse) \
5632 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5635 iscut ? "..." : "<" \
5638 PerlIO_printf(Perl_debug_log,"%16s",""); \
5641 num = RExC_size + 1; \
5643 num=REG_NODE_NUM(RExC_emit); \
5644 if (RExC_lastnum!=num) \
5645 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5647 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5648 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5649 (int)((depth*2)), "", \
5653 RExC_lastparse=RExC_parse; \
5658 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5659 DEBUG_PARSE_MSG((funcname)); \
5660 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5662 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5663 DEBUG_PARSE_MSG((funcname)); \
5664 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5667 - reg - regular expression, i.e. main body or parenthesized thing
5669 * Caller must absorb opening parenthesis.
5671 * Combining parenthesis handling with the base level of regular expression
5672 * is a trifle forced, but the need to tie the tails of the branches to what
5673 * follows makes it hard to avoid.
5675 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
5677 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
5679 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
5683 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
5684 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
5687 register regnode *ret; /* Will be the head of the group. */
5688 register regnode *br;
5689 register regnode *lastbr;
5690 register regnode *ender = NULL;
5691 register I32 parno = 0;
5693 U32 oregflags = RExC_flags;
5694 bool have_branch = 0;
5696 I32 freeze_paren = 0;
5697 I32 after_freeze = 0;
5699 /* for (?g), (?gc), and (?o) warnings; warning
5700 about (?c) will warn about (?g) -- japhy */
5702 #define WASTED_O 0x01
5703 #define WASTED_G 0x02
5704 #define WASTED_C 0x04
5705 #define WASTED_GC (0x02|0x04)
5706 I32 wastedflags = 0x00;
5708 char * parse_start = RExC_parse; /* MJD */
5709 char * const oregcomp_parse = RExC_parse;
5711 GET_RE_DEBUG_FLAGS_DECL;
5713 PERL_ARGS_ASSERT_REG;
5714 DEBUG_PARSE("reg ");
5716 *flagp = 0; /* Tentatively. */
5719 /* Make an OPEN node, if parenthesized. */
5721 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
5722 char *start_verb = RExC_parse;
5723 STRLEN verb_len = 0;
5724 char *start_arg = NULL;
5725 unsigned char op = 0;
5727 int internal_argval = 0; /* internal_argval is only useful if !argok */
5728 while ( *RExC_parse && *RExC_parse != ')' ) {
5729 if ( *RExC_parse == ':' ) {
5730 start_arg = RExC_parse + 1;
5736 verb_len = RExC_parse - start_verb;
5739 while ( *RExC_parse && *RExC_parse != ')' )
5741 if ( *RExC_parse != ')' )
5742 vFAIL("Unterminated verb pattern argument");
5743 if ( RExC_parse == start_arg )
5746 if ( *RExC_parse != ')' )
5747 vFAIL("Unterminated verb pattern");
5750 switch ( *start_verb ) {
5751 case 'A': /* (*ACCEPT) */
5752 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
5754 internal_argval = RExC_nestroot;
5757 case 'C': /* (*COMMIT) */
5758 if ( memEQs(start_verb,verb_len,"COMMIT") )
5761 case 'F': /* (*FAIL) */
5762 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
5767 case ':': /* (*:NAME) */
5768 case 'M': /* (*MARK:NAME) */
5769 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
5774 case 'P': /* (*PRUNE) */
5775 if ( memEQs(start_verb,verb_len,"PRUNE") )
5778 case 'S': /* (*SKIP) */
5779 if ( memEQs(start_verb,verb_len,"SKIP") )
5782 case 'T': /* (*THEN) */
5783 /* [19:06] <TimToady> :: is then */
5784 if ( memEQs(start_verb,verb_len,"THEN") ) {
5786 RExC_seen |= REG_SEEN_CUTGROUP;
5792 vFAIL3("Unknown verb pattern '%.*s'",
5793 verb_len, start_verb);
5796 if ( start_arg && internal_argval ) {
5797 vFAIL3("Verb pattern '%.*s' may not have an argument",
5798 verb_len, start_verb);
5799 } else if ( argok < 0 && !start_arg ) {
5800 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
5801 verb_len, start_verb);
5803 ret = reganode(pRExC_state, op, internal_argval);
5804 if ( ! internal_argval && ! SIZE_ONLY ) {
5806 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
5807 ARG(ret) = add_data( pRExC_state, 1, "S" );
5808 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
5815 if (!internal_argval)
5816 RExC_seen |= REG_SEEN_VERBARG;
5817 } else if ( start_arg ) {
5818 vFAIL3("Verb pattern '%.*s' may not have an argument",
5819 verb_len, start_verb);
5821 ret = reg_node(pRExC_state, op);
5823 nextchar(pRExC_state);
5826 if (*RExC_parse == '?') { /* (?...) */
5827 bool is_logical = 0;
5828 const char * const seqstart = RExC_parse;
5829 bool has_use_defaults = FALSE;
5832 paren = *RExC_parse++;
5833 ret = NULL; /* For look-ahead/behind. */
5836 case 'P': /* (?P...) variants for those used to PCRE/Python */
5837 paren = *RExC_parse++;
5838 if ( paren == '<') /* (?P<...>) named capture */
5840 else if (paren == '>') { /* (?P>name) named recursion */
5841 goto named_recursion;
5843 else if (paren == '=') { /* (?P=...) named backref */
5844 /* this pretty much dupes the code for \k<NAME> in regatom(), if
5845 you change this make sure you change that */
5846 char* name_start = RExC_parse;
5848 SV *sv_dat = reg_scan_name(pRExC_state,
5849 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
5850 if (RExC_parse == name_start || *RExC_parse != ')')
5851 vFAIL2("Sequence %.3s... not terminated",parse_start);
5854 num = add_data( pRExC_state, 1, "S" );
5855 RExC_rxi->data->data[num]=(void*)sv_dat;
5856 SvREFCNT_inc_simple_void(sv_dat);
5859 ret = reganode(pRExC_state,
5870 Set_Node_Offset(ret, parse_start+1);
5871 Set_Node_Cur_Length(ret); /* MJD */
5873 nextchar(pRExC_state);
5877 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5879 case '<': /* (?<...) */
5880 if (*RExC_parse == '!')
5882 else if (*RExC_parse != '=')
5888 case '\'': /* (?'...') */
5889 name_start= RExC_parse;
5890 svname = reg_scan_name(pRExC_state,
5891 SIZE_ONLY ? /* reverse test from the others */
5892 REG_RSN_RETURN_NAME :
5893 REG_RSN_RETURN_NULL);
5894 if (RExC_parse == name_start) {
5896 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
5899 if (*RExC_parse != paren)
5900 vFAIL2("Sequence (?%c... not terminated",
5901 paren=='>' ? '<' : paren);
5905 if (!svname) /* shouldnt happen */
5907 "panic: reg_scan_name returned NULL");
5908 if (!RExC_paren_names) {
5909 RExC_paren_names= newHV();
5910 sv_2mortal(MUTABLE_SV(RExC_paren_names));
5912 RExC_paren_name_list= newAV();
5913 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
5916 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
5918 sv_dat = HeVAL(he_str);
5920 /* croak baby croak */
5922 "panic: paren_name hash element allocation failed");
5923 } else if ( SvPOK(sv_dat) ) {
5924 /* (?|...) can mean we have dupes so scan to check
5925 its already been stored. Maybe a flag indicating
5926 we are inside such a construct would be useful,
5927 but the arrays are likely to be quite small, so
5928 for now we punt -- dmq */
5929 IV count = SvIV(sv_dat);
5930 I32 *pv = (I32*)SvPVX(sv_dat);
5932 for ( i = 0 ; i < count ; i++ ) {
5933 if ( pv[i] == RExC_npar ) {
5939 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
5940 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
5941 pv[count] = RExC_npar;
5942 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
5945 (void)SvUPGRADE(sv_dat,SVt_PVNV);
5946 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
5948 SvIV_set(sv_dat, 1);
5951 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
5952 SvREFCNT_dec(svname);
5955 /*sv_dump(sv_dat);*/
5957 nextchar(pRExC_state);
5959 goto capturing_parens;
5961 RExC_seen |= REG_SEEN_LOOKBEHIND;
5963 case '=': /* (?=...) */
5964 RExC_seen_zerolen++;
5966 case '!': /* (?!...) */
5967 RExC_seen_zerolen++;
5968 if (*RExC_parse == ')') {
5969 ret=reg_node(pRExC_state, OPFAIL);
5970 nextchar(pRExC_state);
5974 case '|': /* (?|...) */
5975 /* branch reset, behave like a (?:...) except that
5976 buffers in alternations share the same numbers */
5978 after_freeze = freeze_paren = RExC_npar;
5980 case ':': /* (?:...) */
5981 case '>': /* (?>...) */
5983 case '$': /* (?$...) */
5984 case '@': /* (?@...) */
5985 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
5987 case '#': /* (?#...) */
5988 while (*RExC_parse && *RExC_parse != ')')
5990 if (*RExC_parse != ')')
5991 FAIL("Sequence (?#... not terminated");
5992 nextchar(pRExC_state);
5995 case '0' : /* (?0) */
5996 case 'R' : /* (?R) */
5997 if (*RExC_parse != ')')
5998 FAIL("Sequence (?R) not terminated");
5999 ret = reg_node(pRExC_state, GOSTART);
6000 *flagp |= POSTPONED;
6001 nextchar(pRExC_state);
6004 { /* named and numeric backreferences */
6006 case '&': /* (?&NAME) */
6007 parse_start = RExC_parse - 1;
6010 SV *sv_dat = reg_scan_name(pRExC_state,
6011 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6012 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6014 goto gen_recurse_regop;
6017 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6019 vFAIL("Illegal pattern");
6021 goto parse_recursion;
6023 case '-': /* (?-1) */
6024 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6025 RExC_parse--; /* rewind to let it be handled later */
6029 case '1': case '2': case '3': case '4': /* (?1) */
6030 case '5': case '6': case '7': case '8': case '9':
6033 num = atoi(RExC_parse);
6034 parse_start = RExC_parse - 1; /* MJD */
6035 if (*RExC_parse == '-')
6037 while (isDIGIT(*RExC_parse))
6039 if (*RExC_parse!=')')
6040 vFAIL("Expecting close bracket");
6043 if ( paren == '-' ) {
6045 Diagram of capture buffer numbering.
6046 Top line is the normal capture buffer numbers
6047 Bottom line is the negative indexing as from
6051 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
6055 num = RExC_npar + num;
6058 vFAIL("Reference to nonexistent group");
6060 } else if ( paren == '+' ) {
6061 num = RExC_npar + num - 1;
6064 ret = reganode(pRExC_state, GOSUB, num);
6066 if (num > (I32)RExC_rx->nparens) {
6068 vFAIL("Reference to nonexistent group");
6070 ARG2L_SET( ret, RExC_recurse_count++);
6072 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6073 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
6077 RExC_seen |= REG_SEEN_RECURSE;
6078 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
6079 Set_Node_Offset(ret, parse_start); /* MJD */
6081 *flagp |= POSTPONED;
6082 nextchar(pRExC_state);
6084 } /* named and numeric backreferences */
6087 case '?': /* (??...) */
6089 if (*RExC_parse != '{') {
6091 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6094 *flagp |= POSTPONED;
6095 paren = *RExC_parse++;
6097 case '{': /* (?{...}) */
6102 char *s = RExC_parse;
6104 RExC_seen_zerolen++;
6105 RExC_seen |= REG_SEEN_EVAL;
6106 while (count && (c = *RExC_parse)) {
6117 if (*RExC_parse != ')') {
6119 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
6123 OP_4tree *sop, *rop;
6124 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
6127 Perl_save_re_context(aTHX);
6128 rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad);
6129 sop->op_private |= OPpREFCOUNTED;
6130 /* re_dup will OpREFCNT_inc */
6131 OpREFCNT_set(sop, 1);
6134 n = add_data(pRExC_state, 3, "nop");
6135 RExC_rxi->data->data[n] = (void*)rop;
6136 RExC_rxi->data->data[n+1] = (void*)sop;
6137 RExC_rxi->data->data[n+2] = (void*)pad;
6140 else { /* First pass */
6141 if (PL_reginterp_cnt < ++RExC_seen_evals
6143 /* No compiled RE interpolated, has runtime
6144 components ===> unsafe. */
6145 FAIL("Eval-group not allowed at runtime, use re 'eval'");
6146 if (PL_tainting && PL_tainted)
6147 FAIL("Eval-group in insecure regular expression");
6148 #if PERL_VERSION > 8
6149 if (IN_PERL_COMPILETIME)
6154 nextchar(pRExC_state);
6156 ret = reg_node(pRExC_state, LOGICAL);
6159 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
6160 /* deal with the length of this later - MJD */
6163 ret = reganode(pRExC_state, EVAL, n);
6164 Set_Node_Length(ret, RExC_parse - parse_start + 1);
6165 Set_Node_Offset(ret, parse_start);
6168 case '(': /* (?(?{...})...) and (?(?=...)...) */
6171 if (RExC_parse[0] == '?') { /* (?(?...)) */
6172 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
6173 || RExC_parse[1] == '<'
6174 || RExC_parse[1] == '{') { /* Lookahead or eval. */
6177 ret = reg_node(pRExC_state, LOGICAL);
6180 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
6184 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
6185 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
6187 char ch = RExC_parse[0] == '<' ? '>' : '\'';
6188 char *name_start= RExC_parse++;
6190 SV *sv_dat=reg_scan_name(pRExC_state,
6191 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6192 if (RExC_parse == name_start || *RExC_parse != ch)
6193 vFAIL2("Sequence (?(%c... not terminated",
6194 (ch == '>' ? '<' : ch));
6197 num = add_data( pRExC_state, 1, "S" );
6198 RExC_rxi->data->data[num]=(void*)sv_dat;
6199 SvREFCNT_inc_simple_void(sv_dat);
6201 ret = reganode(pRExC_state,NGROUPP,num);
6202 goto insert_if_check_paren;
6204 else if (RExC_parse[0] == 'D' &&
6205 RExC_parse[1] == 'E' &&
6206 RExC_parse[2] == 'F' &&
6207 RExC_parse[3] == 'I' &&
6208 RExC_parse[4] == 'N' &&
6209 RExC_parse[5] == 'E')
6211 ret = reganode(pRExC_state,DEFINEP,0);
6214 goto insert_if_check_paren;
6216 else if (RExC_parse[0] == 'R') {
6219 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6220 parno = atoi(RExC_parse++);
6221 while (isDIGIT(*RExC_parse))
6223 } else if (RExC_parse[0] == '&') {
6226 sv_dat = reg_scan_name(pRExC_state,
6227 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6228 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6230 ret = reganode(pRExC_state,INSUBP,parno);
6231 goto insert_if_check_paren;
6233 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6236 parno = atoi(RExC_parse++);
6238 while (isDIGIT(*RExC_parse))
6240 ret = reganode(pRExC_state, GROUPP, parno);
6242 insert_if_check_paren:
6243 if ((c = *nextchar(pRExC_state)) != ')')
6244 vFAIL("Switch condition not recognized");
6246 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
6247 br = regbranch(pRExC_state, &flags, 1,depth+1);
6249 br = reganode(pRExC_state, LONGJMP, 0);
6251 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
6252 c = *nextchar(pRExC_state);
6257 vFAIL("(?(DEFINE)....) does not allow branches");
6258 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6259 regbranch(pRExC_state, &flags, 1,depth+1);
6260 REGTAIL(pRExC_state, ret, lastbr);
6263 c = *nextchar(pRExC_state);
6268 vFAIL("Switch (?(condition)... contains too many branches");
6269 ender = reg_node(pRExC_state, TAIL);
6270 REGTAIL(pRExC_state, br, ender);
6272 REGTAIL(pRExC_state, lastbr, ender);
6273 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6276 REGTAIL(pRExC_state, ret, ender);
6277 RExC_size++; /* XXX WHY do we need this?!!
6278 For large programs it seems to be required
6279 but I can't figure out why. -- dmq*/
6283 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6287 RExC_parse--; /* for vFAIL to print correctly */
6288 vFAIL("Sequence (? incomplete");
6290 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
6292 has_use_defaults = TRUE;
6293 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
6294 if (RExC_utf8) { /* But the default for a utf8 pattern is
6295 unicode semantics */
6296 RExC_flags |= RXf_PMf_UNICODE;
6301 parse_flags: /* (?i) */
6303 U32 posflags = 0, negflags = 0;
6304 U32 *flagsp = &posflags;
6305 bool has_charset_modifier = 0;
6307 while (*RExC_parse) {
6308 /* && strchr("iogcmsx", *RExC_parse) */
6309 /* (?g), (?gc) and (?o) are useless here
6310 and must be globally applied -- japhy */
6311 switch (*RExC_parse) {
6312 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
6313 case LOCALE_PAT_MOD:
6314 if (has_charset_modifier || flagsp == &negflags) {
6315 goto fail_modifiers;
6317 posflags |= RXf_PMf_LOCALE;
6318 negflags |= RXf_PMf_UNICODE;
6319 has_charset_modifier = 1;
6321 case UNICODE_PAT_MOD:
6322 if (has_charset_modifier || flagsp == &negflags) {
6323 goto fail_modifiers;
6325 posflags |= RXf_PMf_UNICODE;
6326 negflags |= RXf_PMf_LOCALE;
6327 has_charset_modifier = 1;
6330 if (has_use_defaults
6331 || has_charset_modifier
6332 || flagsp == &negflags)
6334 goto fail_modifiers;
6337 /* The dual charset means unicode semantics if the
6338 * pattern (or target, not known until runtime) are
6341 posflags |= RXf_PMf_UNICODE;
6342 negflags |= RXf_PMf_LOCALE;
6345 negflags |= (RXf_PMf_LOCALE|RXf_PMf_UNICODE);
6347 has_charset_modifier = 1;
6349 case ONCE_PAT_MOD: /* 'o' */
6350 case GLOBAL_PAT_MOD: /* 'g' */
6351 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6352 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
6353 if (! (wastedflags & wflagbit) ) {
6354 wastedflags |= wflagbit;
6357 "Useless (%s%c) - %suse /%c modifier",
6358 flagsp == &negflags ? "?-" : "?",
6360 flagsp == &negflags ? "don't " : "",
6367 case CONTINUE_PAT_MOD: /* 'c' */
6368 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
6369 if (! (wastedflags & WASTED_C) ) {
6370 wastedflags |= WASTED_GC;
6373 "Useless (%sc) - %suse /gc modifier",
6374 flagsp == &negflags ? "?-" : "?",
6375 flagsp == &negflags ? "don't " : ""
6380 case KEEPCOPY_PAT_MOD: /* 'p' */
6381 if (flagsp == &negflags) {
6383 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
6385 *flagsp |= RXf_PMf_KEEPCOPY;
6389 /* A flag is a default iff it is following a minus, so
6390 * if there is a minus, it means will be trying to
6391 * re-specify a default which is an error */
6392 if (has_use_defaults || flagsp == &negflags) {
6395 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6399 wastedflags = 0; /* reset so (?g-c) warns twice */
6405 RExC_flags |= posflags;
6406 RExC_flags &= ~negflags;
6408 oregflags |= posflags;
6409 oregflags &= ~negflags;
6411 nextchar(pRExC_state);
6422 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6427 }} /* one for the default block, one for the switch */
6434 ret = reganode(pRExC_state, OPEN, parno);
6437 RExC_nestroot = parno;
6438 if (RExC_seen & REG_SEEN_RECURSE
6439 && !RExC_open_parens[parno-1])
6441 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6442 "Setting open paren #%"IVdf" to %d\n",
6443 (IV)parno, REG_NODE_NUM(ret)));
6444 RExC_open_parens[parno-1]= ret;
6447 Set_Node_Length(ret, 1); /* MJD */
6448 Set_Node_Offset(ret, RExC_parse); /* MJD */
6456 /* Pick up the branches, linking them together. */
6457 parse_start = RExC_parse; /* MJD */
6458 br = regbranch(pRExC_state, &flags, 1,depth+1);
6461 if (RExC_npar > after_freeze)
6462 after_freeze = RExC_npar;
6463 RExC_npar = freeze_paren;
6466 /* branch_len = (paren != 0); */
6470 if (*RExC_parse == '|') {
6471 if (!SIZE_ONLY && RExC_extralen) {
6472 reginsert(pRExC_state, BRANCHJ, br, depth+1);
6475 reginsert(pRExC_state, BRANCH, br, depth+1);
6476 Set_Node_Length(br, paren != 0);
6477 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
6481 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
6483 else if (paren == ':') {
6484 *flagp |= flags&SIMPLE;
6486 if (is_open) { /* Starts with OPEN. */
6487 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
6489 else if (paren != '?') /* Not Conditional */
6491 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6493 while (*RExC_parse == '|') {
6494 if (!SIZE_ONLY && RExC_extralen) {
6495 ender = reganode(pRExC_state, LONGJMP,0);
6496 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
6499 RExC_extralen += 2; /* Account for LONGJMP. */
6500 nextchar(pRExC_state);
6502 if (RExC_npar > after_freeze)
6503 after_freeze = RExC_npar;
6504 RExC_npar = freeze_paren;
6506 br = regbranch(pRExC_state, &flags, 0, depth+1);
6510 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
6512 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
6515 if (have_branch || paren != ':') {
6516 /* Make a closing node, and hook it on the end. */
6519 ender = reg_node(pRExC_state, TAIL);
6522 ender = reganode(pRExC_state, CLOSE, parno);
6523 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
6524 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6525 "Setting close paren #%"IVdf" to %d\n",
6526 (IV)parno, REG_NODE_NUM(ender)));
6527 RExC_close_parens[parno-1]= ender;
6528 if (RExC_nestroot == parno)
6531 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
6532 Set_Node_Length(ender,1); /* MJD */
6538 *flagp &= ~HASWIDTH;
6541 ender = reg_node(pRExC_state, SUCCEED);
6544 ender = reg_node(pRExC_state, END);
6546 assert(!RExC_opend); /* there can only be one! */
6551 REGTAIL(pRExC_state, lastbr, ender);
6553 if (have_branch && !SIZE_ONLY) {
6555 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
6557 /* Hook the tails of the branches to the closing node. */
6558 for (br = ret; br; br = regnext(br)) {
6559 const U8 op = PL_regkind[OP(br)];
6561 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
6563 else if (op == BRANCHJ) {
6564 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
6572 static const char parens[] = "=!<,>";
6574 if (paren && (p = strchr(parens, paren))) {
6575 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
6576 int flag = (p - parens) > 1;
6579 node = SUSPEND, flag = 0;
6580 reginsert(pRExC_state, node,ret, depth+1);
6581 Set_Node_Cur_Length(ret);
6582 Set_Node_Offset(ret, parse_start + 1);
6584 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
6588 /* Check for proper termination. */
6590 RExC_flags = oregflags;
6591 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
6592 RExC_parse = oregcomp_parse;
6593 vFAIL("Unmatched (");
6596 else if (!paren && RExC_parse < RExC_end) {
6597 if (*RExC_parse == ')') {
6599 vFAIL("Unmatched )");
6602 FAIL("Junk on end of regexp"); /* "Can't happen". */
6606 RExC_npar = after_freeze;
6611 - regbranch - one alternative of an | operator
6613 * Implements the concatenation operator.
6616 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
6619 register regnode *ret;
6620 register regnode *chain = NULL;
6621 register regnode *latest;
6622 I32 flags = 0, c = 0;
6623 GET_RE_DEBUG_FLAGS_DECL;
6625 PERL_ARGS_ASSERT_REGBRANCH;
6627 DEBUG_PARSE("brnc");
6632 if (!SIZE_ONLY && RExC_extralen)
6633 ret = reganode(pRExC_state, BRANCHJ,0);
6635 ret = reg_node(pRExC_state, BRANCH);
6636 Set_Node_Length(ret, 1);
6640 if (!first && SIZE_ONLY)
6641 RExC_extralen += 1; /* BRANCHJ */
6643 *flagp = WORST; /* Tentatively. */
6646 nextchar(pRExC_state);
6647 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
6649 latest = regpiece(pRExC_state, &flags,depth+1);
6650 if (latest == NULL) {
6651 if (flags & TRYAGAIN)
6655 else if (ret == NULL)
6657 *flagp |= flags&(HASWIDTH|POSTPONED);
6658 if (chain == NULL) /* First piece. */
6659 *flagp |= flags&SPSTART;
6662 REGTAIL(pRExC_state, chain, latest);
6667 if (chain == NULL) { /* Loop ran zero times. */
6668 chain = reg_node(pRExC_state, NOTHING);
6673 *flagp |= flags&SIMPLE;
6680 - regpiece - something followed by possible [*+?]
6682 * Note that the branching code sequences used for ? and the general cases
6683 * of * and + are somewhat optimized: they use the same NOTHING node as
6684 * both the endmarker for their branch list and the body of the last branch.
6685 * It might seem that this node could be dispensed with entirely, but the
6686 * endmarker role is not redundant.
6689 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
6692 register regnode *ret;
6694 register char *next;
6696 const char * const origparse = RExC_parse;
6698 I32 max = REG_INFTY;
6700 const char *maxpos = NULL;
6701 GET_RE_DEBUG_FLAGS_DECL;
6703 PERL_ARGS_ASSERT_REGPIECE;
6705 DEBUG_PARSE("piec");
6707 ret = regatom(pRExC_state, &flags,depth+1);
6709 if (flags & TRYAGAIN)
6716 if (op == '{' && regcurly(RExC_parse)) {
6718 parse_start = RExC_parse; /* MJD */
6719 next = RExC_parse + 1;
6720 while (isDIGIT(*next) || *next == ',') {
6729 if (*next == '}') { /* got one */
6733 min = atoi(RExC_parse);
6737 maxpos = RExC_parse;
6739 if (!max && *maxpos != '0')
6740 max = REG_INFTY; /* meaning "infinity" */
6741 else if (max >= REG_INFTY)
6742 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
6744 nextchar(pRExC_state);
6747 if ((flags&SIMPLE)) {
6748 RExC_naughty += 2 + RExC_naughty / 2;
6749 reginsert(pRExC_state, CURLY, ret, depth+1);
6750 Set_Node_Offset(ret, parse_start+1); /* MJD */
6751 Set_Node_Cur_Length(ret);
6754 regnode * const w = reg_node(pRExC_state, WHILEM);
6757 REGTAIL(pRExC_state, ret, w);
6758 if (!SIZE_ONLY && RExC_extralen) {
6759 reginsert(pRExC_state, LONGJMP,ret, depth+1);
6760 reginsert(pRExC_state, NOTHING,ret, depth+1);
6761 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
6763 reginsert(pRExC_state, CURLYX,ret, depth+1);
6765 Set_Node_Offset(ret, parse_start+1);
6766 Set_Node_Length(ret,
6767 op == '{' ? (RExC_parse - parse_start) : 1);
6769 if (!SIZE_ONLY && RExC_extralen)
6770 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
6771 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
6773 RExC_whilem_seen++, RExC_extralen += 3;
6774 RExC_naughty += 4 + RExC_naughty; /* compound interest */
6783 vFAIL("Can't do {n,m} with n > m");
6785 ARG1_SET(ret, (U16)min);
6786 ARG2_SET(ret, (U16)max);
6798 #if 0 /* Now runtime fix should be reliable. */
6800 /* if this is reinstated, don't forget to put this back into perldiag:
6802 =item Regexp *+ operand could be empty at {#} in regex m/%s/
6804 (F) The part of the regexp subject to either the * or + quantifier
6805 could match an empty string. The {#} shows in the regular
6806 expression about where the problem was discovered.
6810 if (!(flags&HASWIDTH) && op != '?')
6811 vFAIL("Regexp *+ operand could be empty");
6814 parse_start = RExC_parse;
6815 nextchar(pRExC_state);
6817 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
6819 if (op == '*' && (flags&SIMPLE)) {
6820 reginsert(pRExC_state, STAR, ret, depth+1);
6824 else if (op == '*') {
6828 else if (op == '+' && (flags&SIMPLE)) {
6829 reginsert(pRExC_state, PLUS, ret, depth+1);
6833 else if (op == '+') {
6837 else if (op == '?') {
6842 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
6843 ckWARN3reg(RExC_parse,
6844 "%.*s matches null string many times",
6845 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
6849 if (RExC_parse < RExC_end && *RExC_parse == '?') {
6850 nextchar(pRExC_state);
6851 reginsert(pRExC_state, MINMOD, ret, depth+1);
6852 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
6854 #ifndef REG_ALLOW_MINMOD_SUSPEND
6857 if (RExC_parse < RExC_end && *RExC_parse == '+') {
6859 nextchar(pRExC_state);
6860 ender = reg_node(pRExC_state, SUCCEED);
6861 REGTAIL(pRExC_state, ret, ender);
6862 reginsert(pRExC_state, SUSPEND, ret, depth+1);
6864 ender = reg_node(pRExC_state, TAIL);
6865 REGTAIL(pRExC_state, ret, ender);
6869 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
6871 vFAIL("Nested quantifiers");
6878 /* reg_namedseq(pRExC_state,UVp)
6880 This is expected to be called by a parser routine that has
6881 recognized '\N' and needs to handle the rest. RExC_parse is
6882 expected to point at the first char following the N at the time
6885 The \N may be inside (indicated by valuep not being NULL) or outside a
6888 \N may begin either a named sequence, or if outside a character class, mean
6889 to match a non-newline. For non single-quoted regexes, the tokenizer has
6890 attempted to decide which, and in the case of a named sequence converted it
6891 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
6892 where c1... are the characters in the sequence. For single-quoted regexes,
6893 the tokenizer passes the \N sequence through unchanged; this code will not
6894 attempt to determine this nor expand those. The net effect is that if the
6895 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
6896 signals that this \N occurrence means to match a non-newline.
6898 Only the \N{U+...} form should occur in a character class, for the same
6899 reason that '.' inside a character class means to just match a period: it
6900 just doesn't make sense.
6902 If valuep is non-null then it is assumed that we are parsing inside
6903 of a charclass definition and the first codepoint in the resolved
6904 string is returned via *valuep and the routine will return NULL.
6905 In this mode if a multichar string is returned from the charnames
6906 handler, a warning will be issued, and only the first char in the
6907 sequence will be examined. If the string returned is zero length
6908 then the value of *valuep is undefined and NON-NULL will
6909 be returned to indicate failure. (This will NOT be a valid pointer
6912 If valuep is null then it is assumed that we are parsing normal text and a
6913 new EXACT node is inserted into the program containing the resolved string,
6914 and a pointer to the new node is returned. But if the string is zero length
6915 a NOTHING node is emitted instead.
6917 On success RExC_parse is set to the char following the endbrace.
6918 Parsing failures will generate a fatal error via vFAIL(...)
6921 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp)
6923 char * endbrace; /* '}' following the name */
6924 regnode *ret = NULL;
6926 char* parse_start = RExC_parse - 2; /* points to the '\N' */
6930 GET_RE_DEBUG_FLAGS_DECL;
6932 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
6936 /* The [^\n] meaning of \N ignores spaces and comments under the /x
6937 * modifier. The other meaning does not */
6938 p = (RExC_flags & RXf_PMf_EXTENDED)
6939 ? regwhite( pRExC_state, RExC_parse )
6942 /* Disambiguate between \N meaning a named character versus \N meaning
6943 * [^\n]. The former is assumed when it can't be the latter. */
6944 if (*p != '{' || regcurly(p)) {
6947 /* no bare \N in a charclass */
6948 vFAIL("\\N in a character class must be a named character: \\N{...}");
6950 nextchar(pRExC_state);
6951 ret = reg_node(pRExC_state, REG_ANY);
6952 *flagp |= HASWIDTH|SIMPLE;
6955 Set_Node_Length(ret, 1); /* MJD */
6959 /* Here, we have decided it should be a named sequence */
6961 /* The test above made sure that the next real character is a '{', but
6962 * under the /x modifier, it could be separated by space (or a comment and
6963 * \n) and this is not allowed (for consistency with \x{...} and the
6964 * tokenizer handling of \N{NAME}). */
6965 if (*RExC_parse != '{') {
6966 vFAIL("Missing braces on \\N{}");
6969 RExC_parse++; /* Skip past the '{' */
6971 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
6972 || ! (endbrace == RExC_parse /* nothing between the {} */
6973 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
6974 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
6976 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
6977 vFAIL("\\N{NAME} must be resolved by the lexer");
6980 if (endbrace == RExC_parse) { /* empty: \N{} */
6982 RExC_parse = endbrace + 1;
6983 return reg_node(pRExC_state,NOTHING);
6987 ckWARNreg(RExC_parse,
6988 "Ignoring zero length \\N{} in character class"
6990 RExC_parse = endbrace + 1;
6993 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
6996 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
6997 RExC_parse += 2; /* Skip past the 'U+' */
6999 if (valuep) { /* In a bracketed char class */
7000 /* We only pay attention to the first char of
7001 multichar strings being returned. I kinda wonder
7002 if this makes sense as it does change the behaviour
7003 from earlier versions, OTOH that behaviour was broken
7004 as well. XXX Solution is to recharacterize as
7005 [rest-of-class]|multi1|multi2... */
7007 STRLEN length_of_hex;
7008 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7009 | PERL_SCAN_DISALLOW_PREFIX
7010 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
7012 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
7013 if (endchar < endbrace) {
7014 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
7017 length_of_hex = (STRLEN)(endchar - RExC_parse);
7018 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
7020 /* The tokenizer should have guaranteed validity, but it's possible to
7021 * bypass it by using single quoting, so check */
7022 if (length_of_hex == 0
7023 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
7025 RExC_parse += length_of_hex; /* Includes all the valid */
7026 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
7027 ? UTF8SKIP(RExC_parse)
7029 /* Guard against malformed utf8 */
7030 if (RExC_parse >= endchar) RExC_parse = endchar;
7031 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7034 RExC_parse = endbrace + 1;
7035 if (endchar == endbrace) return NULL;
7037 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
7039 else { /* Not a char class */
7040 char *s; /* String to put in generated EXACT node */
7041 STRLEN len = 0; /* Its current byte length */
7042 char *endchar; /* Points to '.' or '}' ending cur char in the input
7045 ret = reg_node(pRExC_state, (U8) ((! FOLD) ? EXACT
7053 /* Exact nodes can hold only a U8 length's of text = 255. Loop through
7054 * the input which is of the form now 'c1.c2.c3...}' until find the
7055 * ending brace or exceed length 255. The characters that exceed this
7056 * limit are dropped. The limit could be relaxed should it become
7057 * desirable by reparsing this as (?:\N{NAME}), so could generate
7058 * multiple EXACT nodes, as is done for just regular input. But this
7059 * is primarily a named character, and not intended to be a huge long
7060 * string, so 255 bytes should be good enough */
7062 STRLEN length_of_hex;
7063 I32 grok_flags = PERL_SCAN_ALLOW_UNDERSCORES
7064 | PERL_SCAN_DISALLOW_PREFIX
7065 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
7066 UV cp; /* Ord of current character */
7068 /* Code points are separated by dots. If none, there is only one
7069 * code point, and is terminated by the brace */
7070 endchar = RExC_parse + strcspn(RExC_parse, ".}");
7072 /* The values are Unicode even on EBCDIC machines */
7073 length_of_hex = (STRLEN)(endchar - RExC_parse);
7074 cp = grok_hex(RExC_parse, &length_of_hex, &grok_flags, NULL);
7075 if ( length_of_hex == 0
7076 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
7078 RExC_parse += length_of_hex; /* Includes all the valid */
7079 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
7080 ? UTF8SKIP(RExC_parse)
7082 /* Guard against malformed utf8 */
7083 if (RExC_parse >= endchar) RExC_parse = endchar;
7084 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7087 if (! FOLD) { /* Not folding, just append to the string */
7090 /* Quit before adding this character if would exceed limit */
7091 if (len + UNISKIP(cp) > U8_MAX) break;
7093 unilen = reguni(pRExC_state, cp, s);
7098 } else { /* Folding, output the folded equivalent */
7099 STRLEN foldlen,numlen;
7100 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7101 cp = toFOLD_uni(cp, tmpbuf, &foldlen);
7103 /* Quit before exceeding size limit */
7104 if (len + foldlen > U8_MAX) break;
7106 for (foldbuf = tmpbuf;
7110 cp = utf8_to_uvchr(foldbuf, &numlen);
7112 const STRLEN unilen = reguni(pRExC_state, cp, s);
7115 /* In EBCDIC the numlen and unilen can differ. */
7117 if (numlen >= foldlen)
7121 break; /* "Can't happen." */
7125 /* Point to the beginning of the next character in the sequence. */
7126 RExC_parse = endchar + 1;
7128 /* Quit if no more characters */
7129 if (RExC_parse >= endbrace) break;
7134 if (RExC_parse < endbrace) {
7135 ckWARNreg(RExC_parse - 1,
7136 "Using just the first characters returned by \\N{}");
7139 RExC_size += STR_SZ(len);
7142 RExC_emit += STR_SZ(len);
7145 RExC_parse = endbrace + 1;
7147 *flagp |= HASWIDTH; /* Not SIMPLE, as that causes the engine to fail
7148 with malformed in t/re/pat_advanced.t */
7150 Set_Node_Cur_Length(ret); /* MJD */
7151 nextchar(pRExC_state);
7161 * It returns the code point in utf8 for the value in *encp.
7162 * value: a code value in the source encoding
7163 * encp: a pointer to an Encode object
7165 * If the result from Encode is not a single character,
7166 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
7169 S_reg_recode(pTHX_ const char value, SV **encp)
7172 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
7173 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
7174 const STRLEN newlen = SvCUR(sv);
7175 UV uv = UNICODE_REPLACEMENT;
7177 PERL_ARGS_ASSERT_REG_RECODE;
7181 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
7184 if (!newlen || numlen != newlen) {
7185 uv = UNICODE_REPLACEMENT;
7193 - regatom - the lowest level
7195 Try to identify anything special at the start of the pattern. If there
7196 is, then handle it as required. This may involve generating a single regop,
7197 such as for an assertion; or it may involve recursing, such as to
7198 handle a () structure.
7200 If the string doesn't start with something special then we gobble up
7201 as much literal text as we can.
7203 Once we have been able to handle whatever type of thing started the
7204 sequence, we return.
7206 Note: we have to be careful with escapes, as they can be both literal
7207 and special, and in the case of \10 and friends can either, depending
7208 on context. Specifically there are two seperate switches for handling
7209 escape sequences, with the one for handling literal escapes requiring
7210 a dummy entry for all of the special escapes that are actually handled
7215 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7218 register regnode *ret = NULL;
7220 char *parse_start = RExC_parse;
7221 GET_RE_DEBUG_FLAGS_DECL;
7222 DEBUG_PARSE("atom");
7223 *flagp = WORST; /* Tentatively. */
7225 PERL_ARGS_ASSERT_REGATOM;
7228 switch ((U8)*RExC_parse) {
7230 RExC_seen_zerolen++;
7231 nextchar(pRExC_state);
7232 if (RExC_flags & RXf_PMf_MULTILINE)
7233 ret = reg_node(pRExC_state, MBOL);
7234 else if (RExC_flags & RXf_PMf_SINGLELINE)
7235 ret = reg_node(pRExC_state, SBOL);
7237 ret = reg_node(pRExC_state, BOL);
7238 Set_Node_Length(ret, 1); /* MJD */
7241 nextchar(pRExC_state);
7243 RExC_seen_zerolen++;
7244 if (RExC_flags & RXf_PMf_MULTILINE)
7245 ret = reg_node(pRExC_state, MEOL);
7246 else if (RExC_flags & RXf_PMf_SINGLELINE)
7247 ret = reg_node(pRExC_state, SEOL);
7249 ret = reg_node(pRExC_state, EOL);
7250 Set_Node_Length(ret, 1); /* MJD */
7253 nextchar(pRExC_state);
7254 if (RExC_flags & RXf_PMf_SINGLELINE)
7255 ret = reg_node(pRExC_state, SANY);
7257 ret = reg_node(pRExC_state, REG_ANY);
7258 *flagp |= HASWIDTH|SIMPLE;
7260 Set_Node_Length(ret, 1); /* MJD */
7264 char * const oregcomp_parse = ++RExC_parse;
7265 ret = regclass(pRExC_state,depth+1);
7266 if (*RExC_parse != ']') {
7267 RExC_parse = oregcomp_parse;
7268 vFAIL("Unmatched [");
7270 nextchar(pRExC_state);
7271 *flagp |= HASWIDTH|SIMPLE;
7272 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
7276 nextchar(pRExC_state);
7277 ret = reg(pRExC_state, 1, &flags,depth+1);
7279 if (flags & TRYAGAIN) {
7280 if (RExC_parse == RExC_end) {
7281 /* Make parent create an empty node if needed. */
7289 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7293 if (flags & TRYAGAIN) {
7297 vFAIL("Internal urp");
7298 /* Supposed to be caught earlier. */
7301 if (!regcurly(RExC_parse)) {
7310 vFAIL("Quantifier follows nothing");
7312 case LATIN_SMALL_LETTER_SHARP_S:
7313 case UTF8_TWO_BYTE_HI_nocast(LATIN_SMALL_LETTER_SHARP_S):
7314 case UTF8_TWO_BYTE_HI_nocast(IOTA_D_T):
7315 #if UTF8_TWO_BYTE_HI_nocast(UPSILON_D_T) != UTF8_TWO_BYTE_HI_nocast(IOTA_D_T)
7316 #error The beginning utf8 byte of IOTA_D_T and UPSILON_D_T unexpectedly differ. Other instances in this code should have the case statement below.
7317 case UTF8_TWO_BYTE_HI_nocast(UPSILON_D_T):
7322 len=0; /* silence a spurious compiler warning */
7323 if ((cp = what_len_TRICKYFOLD_safe(RExC_parse,RExC_end,UTF,len))) {
7324 *flagp |= HASWIDTH; /* could be SIMPLE too, but needs a handler in regexec.regrepeat */
7325 RExC_parse+=len-1; /* we get one from nextchar() as well. :-( */
7326 ret = reganode(pRExC_state, FOLDCHAR, cp);
7327 Set_Node_Length(ret, 1); /* MJD */
7328 nextchar(pRExC_state); /* kill whitespace under /x */
7336 This switch handles escape sequences that resolve to some kind
7337 of special regop and not to literal text. Escape sequnces that
7338 resolve to literal text are handled below in the switch marked
7341 Every entry in this switch *must* have a corresponding entry
7342 in the literal escape switch. However, the opposite is not
7343 required, as the default for this switch is to jump to the
7344 literal text handling code.
7346 switch ((U8)*++RExC_parse) {
7347 case LATIN_SMALL_LETTER_SHARP_S:
7348 case UTF8_TWO_BYTE_HI_nocast(LATIN_SMALL_LETTER_SHARP_S):
7349 case UTF8_TWO_BYTE_HI_nocast(IOTA_D_T):
7351 /* Special Escapes */
7353 RExC_seen_zerolen++;
7354 ret = reg_node(pRExC_state, SBOL);
7356 goto finish_meta_pat;
7358 ret = reg_node(pRExC_state, GPOS);
7359 RExC_seen |= REG_SEEN_GPOS;
7361 goto finish_meta_pat;
7363 RExC_seen_zerolen++;
7364 ret = reg_node(pRExC_state, KEEPS);
7366 /* XXX:dmq : disabling in-place substitution seems to
7367 * be necessary here to avoid cases of memory corruption, as
7368 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
7370 RExC_seen |= REG_SEEN_LOOKBEHIND;
7371 goto finish_meta_pat;
7373 ret = reg_node(pRExC_state, SEOL);
7375 RExC_seen_zerolen++; /* Do not optimize RE away */
7376 goto finish_meta_pat;
7378 ret = reg_node(pRExC_state, EOS);
7380 RExC_seen_zerolen++; /* Do not optimize RE away */
7381 goto finish_meta_pat;
7383 ret = reg_node(pRExC_state, CANY);
7384 RExC_seen |= REG_SEEN_CANY;
7385 *flagp |= HASWIDTH|SIMPLE;
7386 goto finish_meta_pat;
7388 ret = reg_node(pRExC_state, CLUMP);
7390 goto finish_meta_pat;
7393 ret = reg_node(pRExC_state, (U8)(ALNUML));
7395 ret = reg_node(pRExC_state, (U8)(ALNUM));
7396 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7398 *flagp |= HASWIDTH|SIMPLE;
7399 goto finish_meta_pat;
7402 ret = reg_node(pRExC_state, (U8)(NALNUML));
7404 ret = reg_node(pRExC_state, (U8)(NALNUM));
7405 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7407 *flagp |= HASWIDTH|SIMPLE;
7408 goto finish_meta_pat;
7410 RExC_seen_zerolen++;
7411 RExC_seen |= REG_SEEN_LOOKBEHIND;
7413 ret = reg_node(pRExC_state, (U8)(BOUNDL));
7415 ret = reg_node(pRExC_state, (U8)(BOUND));
7416 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7419 goto finish_meta_pat;
7421 RExC_seen_zerolen++;
7422 RExC_seen |= REG_SEEN_LOOKBEHIND;
7424 ret = reg_node(pRExC_state, (U8)(NBOUNDL));
7426 ret = reg_node(pRExC_state, (U8)(NBOUND));
7427 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7430 goto finish_meta_pat;
7433 ret = reg_node(pRExC_state, (U8)(SPACEL));
7435 ret = reg_node(pRExC_state, (U8)(SPACE));
7436 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7438 *flagp |= HASWIDTH|SIMPLE;
7439 goto finish_meta_pat;
7442 ret = reg_node(pRExC_state, (U8)(NSPACEL));
7444 ret = reg_node(pRExC_state, (U8)(NSPACE));
7445 FLAGS(ret) = (UNI_SEMANTICS) ? USE_UNI : 0;
7447 *flagp |= HASWIDTH|SIMPLE;
7448 goto finish_meta_pat;
7451 ret = reg_node(pRExC_state, (U8)(DIGITL));
7453 ret = reg_node(pRExC_state, (U8)(DIGIT));
7455 *flagp |= HASWIDTH|SIMPLE;
7456 goto finish_meta_pat;
7459 ret = reg_node(pRExC_state, (U8)(NDIGITL));
7461 ret = reg_node(pRExC_state, (U8)(NDIGIT));
7463 *flagp |= HASWIDTH|SIMPLE;
7464 goto finish_meta_pat;
7466 ret = reg_node(pRExC_state, LNBREAK);
7467 *flagp |= HASWIDTH|SIMPLE;
7468 goto finish_meta_pat;
7470 ret = reg_node(pRExC_state, HORIZWS);
7471 *flagp |= HASWIDTH|SIMPLE;
7472 goto finish_meta_pat;
7474 ret = reg_node(pRExC_state, NHORIZWS);
7475 *flagp |= HASWIDTH|SIMPLE;
7476 goto finish_meta_pat;
7478 ret = reg_node(pRExC_state, VERTWS);
7479 *flagp |= HASWIDTH|SIMPLE;
7480 goto finish_meta_pat;
7482 ret = reg_node(pRExC_state, NVERTWS);
7483 *flagp |= HASWIDTH|SIMPLE;
7485 nextchar(pRExC_state);
7486 Set_Node_Length(ret, 2); /* MJD */
7491 char* const oldregxend = RExC_end;
7493 char* parse_start = RExC_parse - 2;
7496 if (RExC_parse[1] == '{') {
7497 /* a lovely hack--pretend we saw [\pX] instead */
7498 RExC_end = strchr(RExC_parse, '}');
7500 const U8 c = (U8)*RExC_parse;
7502 RExC_end = oldregxend;
7503 vFAIL2("Missing right brace on \\%c{}", c);
7508 RExC_end = RExC_parse + 2;
7509 if (RExC_end > oldregxend)
7510 RExC_end = oldregxend;
7514 ret = regclass(pRExC_state,depth+1);
7516 RExC_end = oldregxend;
7519 Set_Node_Offset(ret, parse_start + 2);
7520 Set_Node_Cur_Length(ret);
7521 nextchar(pRExC_state);
7522 *flagp |= HASWIDTH|SIMPLE;
7526 /* Handle \N and \N{NAME} here and not below because it can be
7527 multicharacter. join_exact() will join them up later on.
7528 Also this makes sure that things like /\N{BLAH}+/ and
7529 \N{BLAH} being multi char Just Happen. dmq*/
7531 ret= reg_namedseq(pRExC_state, NULL, flagp);
7533 case 'k': /* Handle \k<NAME> and \k'NAME' */
7536 char ch= RExC_parse[1];
7537 if (ch != '<' && ch != '\'' && ch != '{') {
7539 vFAIL2("Sequence %.2s... not terminated",parse_start);
7541 /* this pretty much dupes the code for (?P=...) in reg(), if
7542 you change this make sure you change that */
7543 char* name_start = (RExC_parse += 2);
7545 SV *sv_dat = reg_scan_name(pRExC_state,
7546 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7547 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
7548 if (RExC_parse == name_start || *RExC_parse != ch)
7549 vFAIL2("Sequence %.3s... not terminated",parse_start);
7552 num = add_data( pRExC_state, 1, "S" );
7553 RExC_rxi->data->data[num]=(void*)sv_dat;
7554 SvREFCNT_inc_simple_void(sv_dat);
7558 ret = reganode(pRExC_state,
7569 /* override incorrect value set in reganode MJD */
7570 Set_Node_Offset(ret, parse_start+1);
7571 Set_Node_Cur_Length(ret); /* MJD */
7572 nextchar(pRExC_state);
7578 case '1': case '2': case '3': case '4':
7579 case '5': case '6': case '7': case '8': case '9':
7582 bool isg = *RExC_parse == 'g';
7587 if (*RExC_parse == '{') {
7591 if (*RExC_parse == '-') {
7595 if (hasbrace && !isDIGIT(*RExC_parse)) {
7596 if (isrel) RExC_parse--;
7598 goto parse_named_seq;
7600 num = atoi(RExC_parse);
7601 if (isg && num == 0)
7602 vFAIL("Reference to invalid group 0");
7604 num = RExC_npar - num;
7606 vFAIL("Reference to nonexistent or unclosed group");
7608 if (!isg && num > 9 && num >= RExC_npar)
7611 char * const parse_start = RExC_parse - 1; /* MJD */
7612 while (isDIGIT(*RExC_parse))
7614 if (parse_start == RExC_parse - 1)
7615 vFAIL("Unterminated \\g... pattern");
7617 if (*RExC_parse != '}')
7618 vFAIL("Unterminated \\g{...} pattern");
7622 if (num > (I32)RExC_rx->nparens)
7623 vFAIL("Reference to nonexistent group");
7626 ret = reganode(pRExC_state,
7637 /* override incorrect value set in reganode MJD */
7638 Set_Node_Offset(ret, parse_start+1);
7639 Set_Node_Cur_Length(ret); /* MJD */
7641 nextchar(pRExC_state);
7646 if (RExC_parse >= RExC_end)
7647 FAIL("Trailing \\");
7650 /* Do not generate "unrecognized" warnings here, we fall
7651 back into the quick-grab loop below */
7658 if (RExC_flags & RXf_PMf_EXTENDED) {
7659 if ( reg_skipcomment( pRExC_state ) )
7666 register STRLEN len;
7671 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
7673 parse_start = RExC_parse - 1;
7679 ret = reg_node(pRExC_state,
7680 (U8) ((! FOLD) ? EXACT
7688 for (len = 0, p = RExC_parse - 1;
7689 len < 127 && p < RExC_end;
7692 char * const oldp = p;
7694 if (RExC_flags & RXf_PMf_EXTENDED)
7695 p = regwhite( pRExC_state, p );
7697 case LATIN_SMALL_LETTER_SHARP_S:
7698 case UTF8_TWO_BYTE_HI_nocast(LATIN_SMALL_LETTER_SHARP_S):
7699 case UTF8_TWO_BYTE_HI_nocast(IOTA_D_T):
7700 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7701 goto normal_default;
7711 /* Literal Escapes Switch
7713 This switch is meant to handle escape sequences that
7714 resolve to a literal character.
7716 Every escape sequence that represents something
7717 else, like an assertion or a char class, is handled
7718 in the switch marked 'Special Escapes' above in this
7719 routine, but also has an entry here as anything that
7720 isn't explicitly mentioned here will be treated as
7721 an unescaped equivalent literal.
7725 /* These are all the special escapes. */
7726 case LATIN_SMALL_LETTER_SHARP_S:
7727 case UTF8_TWO_BYTE_HI_nocast(LATIN_SMALL_LETTER_SHARP_S):
7728 case UTF8_TWO_BYTE_HI_nocast(IOTA_D_T):
7729 if (LOC || !FOLD || !is_TRICKYFOLD_safe(p,RExC_end,UTF))
7730 goto normal_default;
7731 case 'A': /* Start assertion */
7732 case 'b': case 'B': /* Word-boundary assertion*/
7733 case 'C': /* Single char !DANGEROUS! */
7734 case 'd': case 'D': /* digit class */
7735 case 'g': case 'G': /* generic-backref, pos assertion */
7736 case 'h': case 'H': /* HORIZWS */
7737 case 'k': case 'K': /* named backref, keep marker */
7738 case 'N': /* named char sequence */
7739 case 'p': case 'P': /* Unicode property */
7740 case 'R': /* LNBREAK */
7741 case 's': case 'S': /* space class */
7742 case 'v': case 'V': /* VERTWS */
7743 case 'w': case 'W': /* word class */
7744 case 'X': /* eXtended Unicode "combining character sequence" */
7745 case 'z': case 'Z': /* End of line/string assertion */
7749 /* Anything after here is an escape that resolves to a
7750 literal. (Except digits, which may or may not)
7769 ender = ASCII_TO_NATIVE('\033');
7773 ender = ASCII_TO_NATIVE('\007');
7778 STRLEN brace_len = len;
7780 const char* error_msg;
7782 bool valid = grok_bslash_o(p,
7789 RExC_parse = p; /* going to die anyway; point
7790 to exact spot of failure */
7797 if (PL_encoding && ender < 0x100) {
7798 goto recode_encoding;
7807 char* const e = strchr(p, '}');
7811 vFAIL("Missing right brace on \\x{}");
7814 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7815 | PERL_SCAN_DISALLOW_PREFIX;
7816 STRLEN numlen = e - p - 1;
7817 ender = grok_hex(p + 1, &numlen, &flags, NULL);
7824 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
7826 ender = grok_hex(p, &numlen, &flags, NULL);
7829 if (PL_encoding && ender < 0x100)
7830 goto recode_encoding;
7834 ender = grok_bslash_c(*p++, SIZE_ONLY);
7836 case '0': case '1': case '2': case '3':case '4':
7837 case '5': case '6': case '7': case '8':case '9':
7839 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
7841 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
7843 ender = grok_oct(p, &numlen, &flags, NULL);
7853 if (PL_encoding && ender < 0x100)
7854 goto recode_encoding;
7858 SV* enc = PL_encoding;
7859 ender = reg_recode((const char)(U8)ender, &enc);
7860 if (!enc && SIZE_ONLY)
7861 ckWARNreg(p, "Invalid escape in the specified encoding");
7867 FAIL("Trailing \\");
7870 if (!SIZE_ONLY&& isALPHA(*p))
7871 ckWARN2reg(p + 1, "Unrecognized escape \\%c passed through", UCHARAT(p));
7872 goto normal_default;
7877 if (UTF8_IS_START(*p) && UTF) {
7879 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
7880 &numlen, UTF8_ALLOW_DEFAULT);
7887 if ( RExC_flags & RXf_PMf_EXTENDED)
7888 p = regwhite( pRExC_state, p );
7890 /* Prime the casefolded buffer. */
7891 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
7893 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
7898 /* Emit all the Unicode characters. */
7900 for (foldbuf = tmpbuf;
7902 foldlen -= numlen) {
7903 ender = utf8_to_uvchr(foldbuf, &numlen);
7905 const STRLEN unilen = reguni(pRExC_state, ender, s);
7908 /* In EBCDIC the numlen
7909 * and unilen can differ. */
7911 if (numlen >= foldlen)
7915 break; /* "Can't happen." */
7919 const STRLEN unilen = reguni(pRExC_state, ender, s);
7928 REGC((char)ender, s++);
7934 /* Emit all the Unicode characters. */
7936 for (foldbuf = tmpbuf;
7938 foldlen -= numlen) {
7939 ender = utf8_to_uvchr(foldbuf, &numlen);
7941 const STRLEN unilen = reguni(pRExC_state, ender, s);
7944 /* In EBCDIC the numlen
7945 * and unilen can differ. */
7947 if (numlen >= foldlen)
7955 const STRLEN unilen = reguni(pRExC_state, ender, s);
7964 REGC((char)ender, s++);
7968 Set_Node_Cur_Length(ret); /* MJD */
7969 nextchar(pRExC_state);
7971 /* len is STRLEN which is unsigned, need to copy to signed */
7974 vFAIL("Internal disaster");
7978 if (len == 1 && UNI_IS_INVARIANT(ender))
7982 RExC_size += STR_SZ(len);
7985 RExC_emit += STR_SZ(len);
7995 S_regwhite( RExC_state_t *pRExC_state, char *p )
7997 const char *e = RExC_end;
7999 PERL_ARGS_ASSERT_REGWHITE;
8004 else if (*p == '#') {
8013 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
8021 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
8022 Character classes ([:foo:]) can also be negated ([:^foo:]).
8023 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
8024 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
8025 but trigger failures because they are currently unimplemented. */
8027 #define POSIXCC_DONE(c) ((c) == ':')
8028 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
8029 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
8032 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
8035 I32 namedclass = OOB_NAMEDCLASS;
8037 PERL_ARGS_ASSERT_REGPPOSIXCC;
8039 if (value == '[' && RExC_parse + 1 < RExC_end &&
8040 /* I smell either [: or [= or [. -- POSIX has been here, right? */
8041 POSIXCC(UCHARAT(RExC_parse))) {
8042 const char c = UCHARAT(RExC_parse);
8043 char* const s = RExC_parse++;
8045 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
8047 if (RExC_parse == RExC_end)
8048 /* Grandfather lone [:, [=, [. */
8051 const char* const t = RExC_parse++; /* skip over the c */
8054 if (UCHARAT(RExC_parse) == ']') {
8055 const char *posixcc = s + 1;
8056 RExC_parse++; /* skip over the ending ] */
8059 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
8060 const I32 skip = t - posixcc;
8062 /* Initially switch on the length of the name. */
8065 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
8066 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
8069 /* Names all of length 5. */
8070 /* alnum alpha ascii blank cntrl digit graph lower
8071 print punct space upper */
8072 /* Offset 4 gives the best switch position. */
8073 switch (posixcc[4]) {
8075 if (memEQ(posixcc, "alph", 4)) /* alpha */
8076 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
8079 if (memEQ(posixcc, "spac", 4)) /* space */
8080 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
8083 if (memEQ(posixcc, "grap", 4)) /* graph */
8084 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
8087 if (memEQ(posixcc, "asci", 4)) /* ascii */
8088 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
8091 if (memEQ(posixcc, "blan", 4)) /* blank */
8092 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
8095 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
8096 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
8099 if (memEQ(posixcc, "alnu", 4)) /* alnum */
8100 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
8103 if (memEQ(posixcc, "lowe", 4)) /* lower */
8104 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
8105 else if (memEQ(posixcc, "uppe", 4)) /* upper */
8106 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
8109 if (memEQ(posixcc, "digi", 4)) /* digit */
8110 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
8111 else if (memEQ(posixcc, "prin", 4)) /* print */
8112 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
8113 else if (memEQ(posixcc, "punc", 4)) /* punct */
8114 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
8119 if (memEQ(posixcc, "xdigit", 6))
8120 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
8124 if (namedclass == OOB_NAMEDCLASS)
8125 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
8127 assert (posixcc[skip] == ':');
8128 assert (posixcc[skip+1] == ']');
8129 } else if (!SIZE_ONLY) {
8130 /* [[=foo=]] and [[.foo.]] are still future. */
8132 /* adjust RExC_parse so the warning shows after
8134 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
8136 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
8139 /* Maternal grandfather:
8140 * "[:" ending in ":" but not in ":]" */
8150 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
8154 PERL_ARGS_ASSERT_CHECKPOSIXCC;
8156 if (POSIXCC(UCHARAT(RExC_parse))) {
8157 const char *s = RExC_parse;
8158 const char c = *s++;
8162 if (*s && c == *s && s[1] == ']') {
8164 "POSIX syntax [%c %c] belongs inside character classes",
8167 /* [[=foo=]] and [[.foo.]] are still future. */
8168 if (POSIXCC_NOTYET(c)) {
8169 /* adjust RExC_parse so the error shows after
8171 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
8173 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
8179 /* No locale test */
8180 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
8182 for (value = 0; value < 256; value++) \
8184 stored += S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value); \
8188 case ANYOF_N##NAME: \
8189 for (value = 0; value < 256; value++) \
8191 stored += S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value); \
8196 /* Like the above, but there are differences if we are in uni-8-bit or not, so
8197 * there are two tests passed in, to use depending on that. There aren't any
8198 * cases where the label is different from the name, so no need for that
8200 #define _C_C_T_(NAME,TEST_8,TEST_7,WORD) \
8202 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
8203 else if (UNI_SEMANTICS) { \
8204 for (value = 0; value < 256; value++) { \
8205 if (TEST_8) stored += \
8206 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value); \
8210 for (value = 0; value < 128; value++) { \
8211 if (TEST_7) stored += \
8212 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) UNI_TO_NATIVE(value)); \
8218 case ANYOF_N##NAME: \
8219 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
8220 else if (UNI_SEMANTICS) { \
8221 for (value = 0; value < 256; value++) { \
8222 if (! TEST_8) stored += \
8223 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value); \
8227 for (value = 0; value < 128; value++) { \
8228 if (! TEST_7) stored += \
8229 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value); \
8231 for (value = 128; value < 256; value++) { \
8232 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value); \
8240 We dont use PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS as the direct test
8241 so that it is possible to override the option here without having to
8242 rebuild the entire core. as we are required to do if we change regcomp.h
8243 which is where PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS is defined.
8245 #if PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS
8246 #define BROKEN_UNICODE_CHARCLASS_MAPPINGS
8249 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8250 #define POSIX_CC_UNI_NAME(CCNAME) CCNAME
8252 #define POSIX_CC_UNI_NAME(CCNAME) "Posix" CCNAME
8256 S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value)
8259 /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes.
8260 * Locale folding is done at run-time, so this function should not be
8261 * called for nodes that are for locales.
8263 * This function simply sets the bit corresponding to the fold of the input
8264 * 'value', if not already set. The fold of 'f' is 'F', and the fold of
8267 * It also sets any necessary flags, and returns the number of bits that
8268 * actually changed from 0 to 1 */
8273 fold = (UNI_SEMANTICS) ? PL_fold_latin1[value]
8276 /* It assumes the bit for 'value' has already been set */
8277 if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) {
8278 ANYOF_BITMAP_SET(node, fold);
8282 /* The fold of the German sharp s is two ASCII characters, so isn't in the
8283 * bitmap and doesn't have to be in utf8, but we only process it if unicode
8284 * semantics are called for */
8285 if (UNI_SEMANTICS && value == LATIN_SMALL_LETTER_SHARP_S) {
8286 ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8;
8288 else if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value)
8291 && PL_fold_latin1[value] != value))
8292 { /* A character that has a fold outside of Latin1 matches outside the
8293 bitmap, but only when the target string is utf8. Similarly when we
8294 don't have unicode semantics for the above ASCII Latin-1 characters,
8295 and they have a fold, they should match if the target is utf8, and
8297 ANYOF_FLAGS(node) |= ANYOF_UTF8;
8304 PERL_STATIC_INLINE U8
8305 S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value)
8307 /* This inline function sets a bit in the bitmap if not already set, and if
8308 * appropriate, its fold, returning the number of bits that actually
8309 * changed from 0 to 1 */
8313 if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */
8317 ANYOF_BITMAP_SET(node, value);
8320 if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */
8321 stored += S_set_regclass_bit_fold(aTHX_ pRExC_state, node, value);
8328 parse a class specification and produce either an ANYOF node that
8329 matches the pattern or if the pattern matches a single char only and
8330 that char is < 256 and we are case insensitive then we produce an
8335 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
8338 register UV nextvalue;
8339 register IV prevvalue = OOB_UNICODE;
8340 register IV range = 0;
8341 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
8342 register regnode *ret;
8345 char *rangebegin = NULL;
8346 bool need_class = 0;
8349 AV* unicode_alternate = NULL;
8351 UV literal_endpoint = 0;
8353 UV stored = 0; /* how many chars stored in the bitmap */
8355 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
8356 case we need to change the emitted regop to an EXACT. */
8357 const char * orig_parse = RExC_parse;
8358 GET_RE_DEBUG_FLAGS_DECL;
8360 PERL_ARGS_ASSERT_REGCLASS;
8362 PERL_UNUSED_ARG(depth);
8365 DEBUG_PARSE("clas");
8367 /* Assume we are going to generate an ANYOF node. */
8368 ret = reganode(pRExC_state, ANYOF, 0);
8371 ANYOF_FLAGS(ret) = 0;
8373 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
8377 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
8381 RExC_size += ANYOF_SKIP;
8382 #ifdef ANYOF_ADD_LOC_SKIP
8384 RExC_size += ANYOF_ADD_LOC_SKIP;
8387 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
8390 RExC_emit += ANYOF_SKIP;
8392 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
8393 #ifdef ANYOF_ADD_LOC_SKIP
8394 RExC_emit += ANYOF_ADD_LOC_SKIP;
8397 ANYOF_BITMAP_ZERO(ret);
8398 listsv = newSVpvs("# comment\n");
8401 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8403 if (!SIZE_ONLY && POSIXCC(nextvalue))
8404 checkposixcc(pRExC_state);
8406 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
8407 if (UCHARAT(RExC_parse) == ']')
8411 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
8415 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
8418 rangebegin = RExC_parse;
8420 value = utf8n_to_uvchr((U8*)RExC_parse,
8421 RExC_end - RExC_parse,
8422 &numlen, UTF8_ALLOW_DEFAULT);
8423 RExC_parse += numlen;
8426 value = UCHARAT(RExC_parse++);
8428 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
8429 if (value == '[' && POSIXCC(nextvalue))
8430 namedclass = regpposixcc(pRExC_state, value);
8431 else if (value == '\\') {
8433 value = utf8n_to_uvchr((U8*)RExC_parse,
8434 RExC_end - RExC_parse,
8435 &numlen, UTF8_ALLOW_DEFAULT);
8436 RExC_parse += numlen;
8439 value = UCHARAT(RExC_parse++);
8440 /* Some compilers cannot handle switching on 64-bit integer
8441 * values, therefore value cannot be an UV. Yes, this will
8442 * be a problem later if we want switch on Unicode.
8443 * A similar issue a little bit later when switching on
8444 * namedclass. --jhi */
8445 switch ((I32)value) {
8446 case 'w': namedclass = ANYOF_ALNUM; break;
8447 case 'W': namedclass = ANYOF_NALNUM; break;
8448 case 's': namedclass = ANYOF_SPACE; break;
8449 case 'S': namedclass = ANYOF_NSPACE; break;
8450 case 'd': namedclass = ANYOF_DIGIT; break;
8451 case 'D': namedclass = ANYOF_NDIGIT; break;
8452 case 'v': namedclass = ANYOF_VERTWS; break;
8453 case 'V': namedclass = ANYOF_NVERTWS; break;
8454 case 'h': namedclass = ANYOF_HORIZWS; break;
8455 case 'H': namedclass = ANYOF_NHORIZWS; break;
8456 case 'N': /* Handle \N{NAME} in class */
8458 /* We only pay attention to the first char of
8459 multichar strings being returned. I kinda wonder
8460 if this makes sense as it does change the behaviour
8461 from earlier versions, OTOH that behaviour was broken
8463 UV v; /* value is register so we cant & it /grrr */
8464 if (reg_namedseq(pRExC_state, &v, NULL)) {
8474 if (RExC_parse >= RExC_end)
8475 vFAIL2("Empty \\%c{}", (U8)value);
8476 if (*RExC_parse == '{') {
8477 const U8 c = (U8)value;
8478 e = strchr(RExC_parse++, '}');
8480 vFAIL2("Missing right brace on \\%c{}", c);
8481 while (isSPACE(UCHARAT(RExC_parse)))
8483 if (e == RExC_parse)
8484 vFAIL2("Empty \\%c{}", c);
8486 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
8494 if (UCHARAT(RExC_parse) == '^') {
8497 value = value == 'p' ? 'P' : 'p'; /* toggle */
8498 while (isSPACE(UCHARAT(RExC_parse))) {
8503 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%.*s\n",
8504 (value=='p' ? '+' : '!'), (int)n, RExC_parse);
8508 /* The \p could match something in the Latin1 range, hence
8509 * something that isn't utf8 */
8510 ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP;
8511 namedclass = ANYOF_MAX; /* no official name, but it's named */
8514 case 'n': value = '\n'; break;
8515 case 'r': value = '\r'; break;
8516 case 't': value = '\t'; break;
8517 case 'f': value = '\f'; break;
8518 case 'b': value = '\b'; break;
8519 case 'e': value = ASCII_TO_NATIVE('\033');break;
8520 case 'a': value = ASCII_TO_NATIVE('\007');break;
8522 RExC_parse--; /* function expects to be pointed at the 'o' */
8524 const char* error_msg;
8525 bool valid = grok_bslash_o(RExC_parse,
8530 RExC_parse += numlen;
8535 if (PL_encoding && value < 0x100) {
8536 goto recode_encoding;
8540 if (*RExC_parse == '{') {
8541 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8542 | PERL_SCAN_DISALLOW_PREFIX;
8543 char * const e = strchr(RExC_parse++, '}');
8545 vFAIL("Missing right brace on \\x{}");
8547 numlen = e - RExC_parse;
8548 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8552 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8554 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
8555 RExC_parse += numlen;
8557 if (PL_encoding && value < 0x100)
8558 goto recode_encoding;
8561 value = grok_bslash_c(*RExC_parse++, SIZE_ONLY);
8563 case '0': case '1': case '2': case '3': case '4':
8564 case '5': case '6': case '7':
8566 /* Take 1-3 octal digits */
8567 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8569 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
8570 RExC_parse += numlen;
8571 if (PL_encoding && value < 0x100)
8572 goto recode_encoding;
8577 SV* enc = PL_encoding;
8578 value = reg_recode((const char)(U8)value, &enc);
8579 if (!enc && SIZE_ONLY)
8580 ckWARNreg(RExC_parse,
8581 "Invalid escape in the specified encoding");
8585 /* Allow \_ to not give an error */
8586 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
8587 ckWARN2reg(RExC_parse,
8588 "Unrecognized escape \\%c in character class passed through",
8593 } /* end of \blah */
8599 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
8601 /* What matches in a locale is not known until runtime, so need to
8602 * (one time per class) allocate extra space to pass to regexec.
8603 * The space will contain a bit for each named class that is to be
8604 * matched against. This isn't needed for \p{} and pseudo-classes,
8605 * as they are not affected by locale, and hence are dealt with
8607 if (LOC && namedclass < ANYOF_MAX && ! need_class) {
8610 #ifdef ANYOF_CLASS_ADD_SKIP
8611 RExC_size += ANYOF_CLASS_ADD_SKIP;
8615 #ifdef ANYOF_CLASS_ADD_SKIP
8616 RExC_emit += ANYOF_CLASS_ADD_SKIP;
8618 ANYOF_CLASS_ZERO(ret);
8620 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
8623 /* a bad range like a-\d, a-[:digit:] ? */
8627 RExC_parse >= rangebegin ?
8628 RExC_parse - rangebegin : 0;
8629 ckWARN4reg(RExC_parse,
8630 "False [] range \"%*.*s\"",
8633 if (prevvalue < 256) {
8635 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) prevvalue);
8637 S_set_regclass_bit(aTHX_ pRExC_state, ret, '-');
8640 ANYOF_FLAGS(ret) |= ANYOF_UTF8;
8641 Perl_sv_catpvf(aTHX_ listsv,
8642 "%04"UVxf"\n%04"UVxf"\n", (UV)prevvalue, (UV) '-');
8646 range = 0; /* this was not a true range */
8652 const char *what = NULL;
8655 /* Possible truncation here but in some 64-bit environments
8656 * the compiler gets heartburn about switch on 64-bit values.
8657 * A similar issue a little earlier when switching on value.
8659 switch ((I32)namedclass) {
8661 case _C_C_T_(ALNUMC, isALNUMC_L1(value), isALNUMC(value), "XPosixAlnum");
8662 case _C_C_T_(ALPHA, isALPHA_L1(value), isALPHA(value), "XPosixAlpha");
8663 case _C_C_T_(BLANK, isBLANK_L1(value), isBLANK(value), "XPosixBlank");
8664 case _C_C_T_(CNTRL, isCNTRL_L1(value), isCNTRL(value), "XPosixCntrl");
8665 case _C_C_T_(GRAPH, isGRAPH_L1(value), isGRAPH(value), "XPosixGraph");
8666 case _C_C_T_(LOWER, isLOWER_L1(value), isLOWER(value), "XPosixLower");
8667 case _C_C_T_(PRINT, isPRINT_L1(value), isPRINT(value), "XPosixPrint");
8668 case _C_C_T_(PSXSPC, isPSXSPC_L1(value), isPSXSPC(value), "XPosixSpace");
8669 case _C_C_T_(PUNCT, isPUNCT_L1(value), isPUNCT(value), "XPosixPunct");
8670 case _C_C_T_(UPPER, isUPPER_L1(value), isUPPER(value), "XPosixUpper");
8671 #ifdef BROKEN_UNICODE_CHARCLASS_MAPPINGS
8672 /* \s, \w match all unicode if utf8. */
8673 case _C_C_T_(SPACE, isSPACE_L1(value), isSPACE(value), "SpacePerl");
8674 case _C_C_T_(ALNUM, isWORDCHAR_L1(value), isALNUM(value), "Word");
8676 /* \s, \w match ascii and locale only */
8677 case _C_C_T_(SPACE, isSPACE_L1(value), isSPACE(value), "PerlSpace");
8678 case _C_C_T_(ALNUM, isWORDCHAR_L1(value), isALNUM(value), "PerlWord");
8680 case _C_C_T_(XDIGIT, isXDIGIT_L1(value), isXDIGIT(value), "XPosixXDigit");
8681 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
8682 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
8685 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
8687 for (value = 0; value < 128; value++)
8689 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) ASCII_TO_NATIVE(value));
8692 what = NULL; /* Doesn't match outside ascii, so
8693 don't want to add +utf8:: */
8697 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
8699 for (value = 128; value < 256; value++)
8701 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) ASCII_TO_NATIVE(value));
8708 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
8710 /* consecutive digits assumed */
8711 for (value = '0'; value <= '9'; value++)
8713 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value);
8716 what = POSIX_CC_UNI_NAME("Digit");
8720 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
8722 /* consecutive digits assumed */
8723 for (value = 0; value < '0'; value++)
8725 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value);
8726 for (value = '9' + 1; value < 256; value++)
8728 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) value);
8731 what = POSIX_CC_UNI_NAME("Digit");
8734 /* this is to handle \p and \P */
8737 vFAIL("Invalid [::] class");
8741 /* Strings such as "+utf8::isWord\n" */
8742 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
8743 ANYOF_FLAGS(ret) |= ANYOF_UTF8;
8748 } /* end of namedclass \blah */
8751 if (prevvalue > (IV)value) /* b-a */ {
8752 const int w = RExC_parse - rangebegin;
8753 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
8754 range = 0; /* not a valid range */
8758 prevvalue = value; /* save the beginning of the range */
8759 if (*RExC_parse == '-' && RExC_parse+1 < RExC_end &&
8760 RExC_parse[1] != ']') {
8763 /* a bad range like \w-, [:word:]- ? */
8764 if (namedclass > OOB_NAMEDCLASS) {
8765 if (ckWARN(WARN_REGEXP)) {
8767 RExC_parse >= rangebegin ?
8768 RExC_parse - rangebegin : 0;
8770 "False [] range \"%*.*s\"",
8775 S_set_regclass_bit(aTHX_ pRExC_state, ret, '-');
8777 range = 1; /* yeah, it's a range! */
8778 continue; /* but do it the next time */
8782 /* now is the next time */
8784 if (prevvalue < 256) {
8785 const IV ceilvalue = value < 256 ? value : 255;
8788 /* In EBCDIC [\x89-\x91] should include
8789 * the \x8e but [i-j] should not. */
8790 if (literal_endpoint == 2 &&
8791 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
8792 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
8794 if (isLOWER(prevvalue)) {
8795 for (i = prevvalue; i <= ceilvalue; i++)
8796 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8798 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) i);
8801 for (i = prevvalue; i <= ceilvalue; i++)
8802 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
8804 S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) i);
8810 for (i = prevvalue; i <= ceilvalue; i++) {
8811 stored += S_set_regclass_bit(aTHX_ pRExC_state, ret, (U8) i);
8814 if (value > 255 || UTF) {
8815 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
8816 const UV natvalue = NATIVE_TO_UNI(value);
8818 /* If the code point requires utf8 to represent, and we are not
8819 * folding, it can't match unless the target is in utf8. Only
8820 * a few code points above 255 fold to below it, so XXX an
8821 * optimization would be to know which ones and set the flag
8823 ANYOF_FLAGS(ret) |= (FOLD || value < 256)
8826 if (prevnatvalue < natvalue) { /* what about > ? */
8827 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
8828 prevnatvalue, natvalue);
8830 else if (prevnatvalue == natvalue) {
8831 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", natvalue);
8833 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
8835 const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
8837 #ifdef EBCDIC /* RD t/uni/fold ff and 6b */
8838 if (RExC_precomp[0] == ':' &&
8839 RExC_precomp[1] == '[' &&
8840 (f == 0xDF || f == 0x92)) {
8841 f = NATIVE_TO_UNI(f);
8844 /* If folding and foldable and a single
8845 * character, insert also the folded version
8846 * to the charclass. */
8848 #ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
8849 if ((RExC_precomp[0] == ':' &&
8850 RExC_precomp[1] == '[' &&
8852 (value == 0xFB05 || value == 0xFB06))) ?
8853 foldlen == ((STRLEN)UNISKIP(f) - 1) :
8854 foldlen == (STRLEN)UNISKIP(f) )
8856 if (foldlen == (STRLEN)UNISKIP(f))
8858 Perl_sv_catpvf(aTHX_ listsv,
8861 /* Any multicharacter foldings
8862 * require the following transform:
8863 * [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst)
8864 * where E folds into "pq" and F folds
8865 * into "rst", all other characters
8866 * fold to single characters. We save
8867 * away these multicharacter foldings,
8868 * to be later saved as part of the
8869 * additional "s" data. */
8872 if (!unicode_alternate)
8873 unicode_alternate = newAV();
8874 sv = newSVpvn_utf8((char*)foldbuf, foldlen,
8876 av_push(unicode_alternate, sv);
8880 /* If folding and the value is one of the Greek
8881 * sigmas insert a few more sigmas to make the
8882 * folding rules of the sigmas to work right.
8883 * Note that not all the possible combinations
8884 * are handled here: some of them are handled
8885 * by the standard folding rules, and some of
8886 * them (literal or EXACTF cases) are handled
8887 * during runtime in regexec.c:S_find_byclass(). */
8888 if (value == UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA) {
8889 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8890 (UV)UNICODE_GREEK_CAPITAL_LETTER_SIGMA);
8891 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8892 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8894 else if (value == UNICODE_GREEK_CAPITAL_LETTER_SIGMA)
8895 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n",
8896 (UV)UNICODE_GREEK_SMALL_LETTER_SIGMA);
8901 literal_endpoint = 0;
8905 range = 0; /* this range (if it was one) is done now */
8912 /****** !SIZE_ONLY AFTER HERE *********/
8914 /* Folding in the bitmap is taken care of above, but not for locale, for
8915 * which we have to wait to see what folding is in effect at runtime, and
8916 * for things not in the bitmap */
8917 if (FOLD && (LOC || ANYOF_FLAGS(ret) & ANYOF_NONBITMAP)) {
8918 ANYOF_FLAGS(ret) |= ANYOF_FOLD;
8921 /* Optimize inverted simple patterns (e.g. [^a-z]). Note that this doesn't
8922 * optimize locale. Doing so perhaps could be done as long as there is
8923 * nothing like \w in it; some thought also would have to be given to the
8924 * interaction with above 0x100 chars */
8925 if ((ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT) {
8926 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
8927 ANYOF_BITMAP(ret)[value] ^= 0xFF;
8928 stored = 256 - stored;
8930 /* The inversion means that everything above 255 is matched */
8931 ANYOF_FLAGS(ret) = ANYOF_UTF8|ANYOF_UNICODE_ALL;
8934 /* A single character class can be "optimized" into an EXACTish node.
8935 * Note that since we don't currently count how many characters there are
8936 * outside the bitmap, we are XXX missing optimization possibilities for
8937 * them. This optimization can't happen unless this is a truly single
8938 * character class, which means that it can't be an inversion into a
8939 * many-character class, and there must be no possibility of there being
8940 * things outside the bitmap. 'stored' (only) for locales doesn't include
8941 * \w, etc, so have to make a special test that they aren't present
8943 * Similarly A 2-character class of the very special form like [bB] can be
8944 * optimized into an EXACTFish node, but only for non-locales, and for
8945 * characters which only have the two folds; so things like 'fF' and 'Ii'
8946 * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE
8948 if (! (ANYOF_FLAGS(ret) & (ANYOF_NONBITMAP|ANYOF_INVERT|ANYOF_UNICODE_ALL))
8949 && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
8950 || (! ANYOF_CLASS_TEST_ANY_SET(ret)))))
8951 || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
8952 && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value))
8953 /* If the latest code point has a fold whose
8954 * bit is set, it must be the only other one */
8955 && ((prevvalue = PL_fold_latin1[value]) != (IV)value)
8956 && ANYOF_BITMAP_TEST(ret, prevvalue)))))
8958 /* Note that the information needed to decide to do this optimization
8959 * is not currently available until the 2nd pass, and that the actually
8960 * used EXACTish node takes less space than the calculated ANYOF node,
8961 * and hence the amount of space calculated in the first pass is larger
8962 * than actually used, so this optimization doesn't gain us any space.
8963 * But an EXACT node is faster than an ANYOF node, and can be combined
8964 * with any adjacent EXACT nodes later by the optimizer for further
8965 * gains. The speed of executing an EXACTF is similar to an ANYOF
8966 * node, so the optimization advantage comes from the ability to join
8967 * it to adjacent EXACT nodes */
8969 const char * cur_parse= RExC_parse;
8971 RExC_emit = (regnode *)orig_emit;
8972 RExC_parse = (char *)orig_parse;
8976 /* A locale node with one point can be folded; all the other cases
8977 * with folding will have two points, since we calculate them above
8979 if (ANYOF_FLAGS(ret) & ANYOF_FOLD) {
8985 } /* else 2 chars in the bit map: the folds of each other */
8986 else if (UNI_SEMANTICS || !isASCII(value)) {
8988 /* To join adjacent nodes, they must be the exact EXACTish type.
8989 * Try to use the most likely type, by using EXACTFU if the regex
8990 * calls for them, or is required because the character is
8994 else { /* Otherwise, more likely to be EXACTF type */
8998 ret = reg_node(pRExC_state, op);
8999 RExC_parse = (char *)cur_parse;
9000 if (UTF && ! NATIVE_IS_INVARIANT(value)) {
9001 *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value);
9002 *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value);
9004 RExC_emit += STR_SZ(2);
9007 *STRING(ret)= (char)value;
9009 RExC_emit += STR_SZ(1);
9011 SvREFCNT_dec(listsv);
9016 AV * const av = newAV();
9018 /* The 0th element stores the character class description
9019 * in its textual form: used later (regexec.c:Perl_regclass_swash())
9020 * to initialize the appropriate swash (which gets stored in
9021 * the 1st element), and also useful for dumping the regnode.
9022 * The 2nd element stores the multicharacter foldings,
9023 * used later (regexec.c:S_reginclass()). */
9024 av_store(av, 0, listsv);
9025 av_store(av, 1, NULL);
9026 av_store(av, 2, MUTABLE_SV(unicode_alternate));
9027 rv = newRV_noinc(MUTABLE_SV(av));
9028 n = add_data(pRExC_state, 1, "s");
9029 RExC_rxi->data->data[n] = (void*)rv;
9037 /* reg_skipcomment()
9039 Absorbs an /x style # comments from the input stream.
9040 Returns true if there is more text remaining in the stream.
9041 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
9042 terminates the pattern without including a newline.
9044 Note its the callers responsibility to ensure that we are
9050 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
9054 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
9056 while (RExC_parse < RExC_end)
9057 if (*RExC_parse++ == '\n') {
9062 /* we ran off the end of the pattern without ending
9063 the comment, so we have to add an \n when wrapping */
9064 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
9072 Advances the parse position, and optionally absorbs
9073 "whitespace" from the inputstream.
9075 Without /x "whitespace" means (?#...) style comments only,
9076 with /x this means (?#...) and # comments and whitespace proper.
9078 Returns the RExC_parse point from BEFORE the scan occurs.
9080 This is the /x friendly way of saying RExC_parse++.
9084 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
9086 char* const retval = RExC_parse++;
9088 PERL_ARGS_ASSERT_NEXTCHAR;
9091 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
9092 RExC_parse[2] == '#') {
9093 while (*RExC_parse != ')') {
9094 if (RExC_parse == RExC_end)
9095 FAIL("Sequence (?#... not terminated");
9101 if (RExC_flags & RXf_PMf_EXTENDED) {
9102 if (isSPACE(*RExC_parse)) {
9106 else if (*RExC_parse == '#') {
9107 if ( reg_skipcomment( pRExC_state ) )
9116 - reg_node - emit a node
9118 STATIC regnode * /* Location. */
9119 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
9122 register regnode *ptr;
9123 regnode * const ret = RExC_emit;
9124 GET_RE_DEBUG_FLAGS_DECL;
9126 PERL_ARGS_ASSERT_REG_NODE;
9129 SIZE_ALIGN(RExC_size);
9133 if (RExC_emit >= RExC_emit_bound)
9134 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
9136 NODE_ALIGN_FILL(ret);
9138 FILL_ADVANCE_NODE(ptr, op);
9139 REH_CALL_COMP_NODE_HOOK(pRExC_state->rx, (ptr) - 1);
9140 #ifdef RE_TRACK_PATTERN_OFFSETS
9141 if (RExC_offsets) { /* MJD */
9142 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
9143 "reg_node", __LINE__,
9145 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
9146 ? "Overwriting end of array!\n" : "OK",
9147 (UV)(RExC_emit - RExC_emit_start),
9148 (UV)(RExC_parse - RExC_start),
9149 (UV)RExC_offsets[0]));
9150 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
9158 - reganode - emit a node with an argument
9160 STATIC regnode * /* Location. */
9161 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
9164 register regnode *ptr;
9165 regnode * const ret = RExC_emit;
9166 GET_RE_DEBUG_FLAGS_DECL;
9168 PERL_ARGS_ASSERT_REGANODE;
9171 SIZE_ALIGN(RExC_size);
9176 assert(2==regarglen[op]+1);
9178 Anything larger than this has to allocate the extra amount.
9179 If we changed this to be:
9181 RExC_size += (1 + regarglen[op]);
9183 then it wouldn't matter. Its not clear what side effect
9184 might come from that so its not done so far.
9189 if (RExC_emit >= RExC_emit_bound)
9190 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
9192 NODE_ALIGN_FILL(ret);
9194 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
9195 REH_CALL_COMP_NODE_HOOK(pRExC_state->rx, (ptr) - 2);
9196 #ifdef RE_TRACK_PATTERN_OFFSETS
9197 if (RExC_offsets) { /* MJD */
9198 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
9202 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
9203 "Overwriting end of array!\n" : "OK",
9204 (UV)(RExC_emit - RExC_emit_start),
9205 (UV)(RExC_parse - RExC_start),
9206 (UV)RExC_offsets[0]));
9207 Set_Cur_Node_Offset;
9215 - reguni - emit (if appropriate) a Unicode character
9218 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
9222 PERL_ARGS_ASSERT_REGUNI;
9224 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
9228 - reginsert - insert an operator in front of already-emitted operand
9230 * Means relocating the operand.
9233 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
9236 register regnode *src;
9237 register regnode *dst;
9238 register regnode *place;
9239 const int offset = regarglen[(U8)op];
9240 const int size = NODE_STEP_REGNODE + offset;
9241 GET_RE_DEBUG_FLAGS_DECL;
9243 PERL_ARGS_ASSERT_REGINSERT;
9244 PERL_UNUSED_ARG(depth);
9245 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
9246 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
9255 if (RExC_open_parens) {
9257 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
9258 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
9259 if ( RExC_open_parens[paren] >= opnd ) {
9260 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
9261 RExC_open_parens[paren] += size;
9263 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
9265 if ( RExC_close_parens[paren] >= opnd ) {
9266 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
9267 RExC_close_parens[paren] += size;
9269 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
9274 while (src > opnd) {
9275 StructCopy(--src, --dst, regnode);
9276 #ifdef RE_TRACK_PATTERN_OFFSETS
9277 if (RExC_offsets) { /* MJD 20010112 */
9278 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
9282 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
9283 ? "Overwriting end of array!\n" : "OK",
9284 (UV)(src - RExC_emit_start),
9285 (UV)(dst - RExC_emit_start),
9286 (UV)RExC_offsets[0]));
9287 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
9288 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
9294 place = opnd; /* Op node, where operand used to be. */
9295 #ifdef RE_TRACK_PATTERN_OFFSETS
9296 if (RExC_offsets) { /* MJD */
9297 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
9301 (UV)(place - RExC_emit_start) > RExC_offsets[0]
9302 ? "Overwriting end of array!\n" : "OK",
9303 (UV)(place - RExC_emit_start),
9304 (UV)(RExC_parse - RExC_start),
9305 (UV)RExC_offsets[0]));
9306 Set_Node_Offset(place, RExC_parse);
9307 Set_Node_Length(place, 1);
9310 src = NEXTOPER(place);
9311 FILL_ADVANCE_NODE(place, op);
9312 REH_CALL_COMP_NODE_HOOK(pRExC_state->rx, (place) - 1);
9313 Zero(src, offset, regnode);
9317 - regtail - set the next-pointer at the end of a node chain of p to val.
9318 - SEE ALSO: regtail_study
9320 /* TODO: All three parms should be const */
9322 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
9325 register regnode *scan;
9326 GET_RE_DEBUG_FLAGS_DECL;
9328 PERL_ARGS_ASSERT_REGTAIL;
9330 PERL_UNUSED_ARG(depth);
9336 /* Find last node. */
9339 regnode * const temp = regnext(scan);
9341 SV * const mysv=sv_newmortal();
9342 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
9343 regprop(RExC_rx, mysv, scan);
9344 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
9345 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
9346 (temp == NULL ? "->" : ""),
9347 (temp == NULL ? PL_reg_name[OP(val)] : "")
9355 if (reg_off_by_arg[OP(scan)]) {
9356 ARG_SET(scan, val - scan);
9359 NEXT_OFF(scan) = val - scan;
9365 - regtail_study - set the next-pointer at the end of a node chain of p to val.
9366 - Look for optimizable sequences at the same time.
9367 - currently only looks for EXACT chains.
9369 This is expermental code. The idea is to use this routine to perform
9370 in place optimizations on branches and groups as they are constructed,
9371 with the long term intention of removing optimization from study_chunk so
9372 that it is purely analytical.
9374 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
9375 to control which is which.
9378 /* TODO: All four parms should be const */
9381 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
9384 register regnode *scan;
9386 #ifdef EXPERIMENTAL_INPLACESCAN
9389 GET_RE_DEBUG_FLAGS_DECL;
9391 PERL_ARGS_ASSERT_REGTAIL_STUDY;
9397 /* Find last node. */
9401 regnode * const temp = regnext(scan);
9402 #ifdef EXPERIMENTAL_INPLACESCAN
9403 if (PL_regkind[OP(scan)] == EXACT)
9404 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
9413 if( exact == PSEUDO )
9415 else if ( exact != OP(scan) )
9424 SV * const mysv=sv_newmortal();
9425 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
9426 regprop(RExC_rx, mysv, scan);
9427 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
9428 SvPV_nolen_const(mysv),
9430 PL_reg_name[exact]);
9437 SV * const mysv_val=sv_newmortal();
9438 DEBUG_PARSE_MSG("");
9439 regprop(RExC_rx, mysv_val, val);
9440 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
9441 SvPV_nolen_const(mysv_val),
9442 (IV)REG_NODE_NUM(val),
9446 if (reg_off_by_arg[OP(scan)]) {
9447 ARG_SET(scan, val - scan);
9450 NEXT_OFF(scan) = val - scan;
9458 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
9462 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
9467 for (bit=0; bit<32; bit++) {
9468 if (flags & (1<<bit)) {
9470 PerlIO_printf(Perl_debug_log, "%s",lead);
9471 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
9476 PerlIO_printf(Perl_debug_log, "\n");
9478 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
9484 Perl_regdump(pTHX_ const regexp *r)
9488 SV * const sv = sv_newmortal();
9489 SV *dsv= sv_newmortal();
9491 GET_RE_DEBUG_FLAGS_DECL;
9493 PERL_ARGS_ASSERT_REGDUMP;
9495 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
9497 /* Header fields of interest. */
9498 if (r->anchored_substr) {
9499 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
9500 RE_SV_DUMPLEN(r->anchored_substr), 30);
9501 PerlIO_printf(Perl_debug_log,
9502 "anchored %s%s at %"IVdf" ",
9503 s, RE_SV_TAIL(r->anchored_substr),
9504 (IV)r->anchored_offset);
9505 } else if (r->anchored_utf8) {
9506 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
9507 RE_SV_DUMPLEN(r->anchored_utf8), 30);
9508 PerlIO_printf(Perl_debug_log,
9509 "anchored utf8 %s%s at %"IVdf" ",
9510 s, RE_SV_TAIL(r->anchored_utf8),
9511 (IV)r->anchored_offset);
9513 if (r->float_substr) {
9514 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
9515 RE_SV_DUMPLEN(r->float_substr), 30);
9516 PerlIO_printf(Perl_debug_log,
9517 "floating %s%s at %"IVdf"..%"UVuf" ",
9518 s, RE_SV_TAIL(r->float_substr),
9519 (IV)r->float_min_offset, (UV)r->float_max_offset);
9520 } else if (r->float_utf8) {
9521 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
9522 RE_SV_DUMPLEN(r->float_utf8), 30);
9523 PerlIO_printf(Perl_debug_log,
9524 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
9525 s, RE_SV_TAIL(r->float_utf8),
9526 (IV)r->float_min_offset, (UV)r->float_max_offset);
9528 if (r->check_substr || r->check_utf8)
9529 PerlIO_printf(Perl_debug_log,
9531 (r->check_substr == r->float_substr
9532 && r->check_utf8 == r->float_utf8
9533 ? "(checking floating" : "(checking anchored"));
9534 if (r->extflags & RXf_NOSCAN)
9535 PerlIO_printf(Perl_debug_log, " noscan");
9536 if (r->extflags & RXf_CHECK_ALL)
9537 PerlIO_printf(Perl_debug_log, " isall");
9538 if (r->check_substr || r->check_utf8)
9539 PerlIO_printf(Perl_debug_log, ") ");
9541 if (ri->regstclass) {
9542 regprop(r, sv, ri->regstclass);
9543 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
9545 if (r->extflags & RXf_ANCH) {
9546 PerlIO_printf(Perl_debug_log, "anchored");
9547 if (r->extflags & RXf_ANCH_BOL)
9548 PerlIO_printf(Perl_debug_log, "(BOL)");
9549 if (r->extflags & RXf_ANCH_MBOL)
9550 PerlIO_printf(Perl_debug_log, "(MBOL)");
9551 if (r->extflags & RXf_ANCH_SBOL)
9552 PerlIO_printf(Perl_debug_log, "(SBOL)");
9553 if (r->extflags & RXf_ANCH_GPOS)
9554 PerlIO_printf(Perl_debug_log, "(GPOS)");
9555 PerlIO_putc(Perl_debug_log, ' ');
9557 if (r->extflags & RXf_GPOS_SEEN)
9558 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
9559 if (r->intflags & PREGf_SKIP)
9560 PerlIO_printf(Perl_debug_log, "plus ");
9561 if (r->intflags & PREGf_IMPLICIT)
9562 PerlIO_printf(Perl_debug_log, "implicit ");
9563 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
9564 if (r->extflags & RXf_EVAL_SEEN)
9565 PerlIO_printf(Perl_debug_log, "with eval ");
9566 PerlIO_printf(Perl_debug_log, "\n");
9567 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
9569 PERL_ARGS_ASSERT_REGDUMP;
9570 PERL_UNUSED_CONTEXT;
9572 #endif /* DEBUGGING */
9576 - regprop - printable representation of opcode
9578 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
9581 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
9582 if (flags & ANYOF_INVERT) \
9583 /*make sure the invert info is in each */ \
9584 sv_catpvs(sv, "^"); \
9590 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
9595 RXi_GET_DECL(prog,progi);
9596 GET_RE_DEBUG_FLAGS_DECL;
9598 PERL_ARGS_ASSERT_REGPROP;
9602 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
9603 /* It would be nice to FAIL() here, but this may be called from
9604 regexec.c, and it would be hard to supply pRExC_state. */
9605 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
9606 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
9608 k = PL_regkind[OP(o)];
9612 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
9613 * is a crude hack but it may be the best for now since
9614 * we have no flag "this EXACTish node was UTF-8"
9616 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
9617 PERL_PV_ESCAPE_UNI_DETECT |
9618 PERL_PV_PRETTY_ELLIPSES |
9619 PERL_PV_PRETTY_LTGT |
9620 PERL_PV_PRETTY_NOCLEAR
9622 } else if (k == TRIE) {
9623 /* print the details of the trie in dumpuntil instead, as
9624 * progi->data isn't available here */
9625 const char op = OP(o);
9626 const U32 n = ARG(o);
9627 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
9628 (reg_ac_data *)progi->data->data[n] :
9630 const reg_trie_data * const trie
9631 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
9633 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
9634 DEBUG_TRIE_COMPILE_r(
9635 Perl_sv_catpvf(aTHX_ sv,
9636 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
9637 (UV)trie->startstate,
9638 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
9639 (UV)trie->wordcount,
9642 (UV)TRIE_CHARCOUNT(trie),
9643 (UV)trie->uniquecharcount
9646 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
9648 int rangestart = -1;
9649 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
9651 for (i = 0; i <= 256; i++) {
9652 if (i < 256 && BITMAP_TEST(bitmap,i)) {
9653 if (rangestart == -1)
9655 } else if (rangestart != -1) {
9656 if (i <= rangestart + 3)
9657 for (; rangestart < i; rangestart++)
9658 put_byte(sv, rangestart);
9660 put_byte(sv, rangestart);
9662 put_byte(sv, i - 1);
9670 } else if (k == CURLY) {
9671 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
9672 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
9673 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
9675 else if (k == WHILEM && o->flags) /* Ordinal/of */
9676 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
9677 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
9678 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
9679 if ( RXp_PAREN_NAMES(prog) ) {
9680 if ( k != REF || (OP(o) < NREF)) {
9681 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
9682 SV **name= av_fetch(list, ARG(o), 0 );
9684 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9687 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
9688 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
9689 I32 *nums=(I32*)SvPVX(sv_dat);
9690 SV **name= av_fetch(list, nums[0], 0 );
9693 for ( n=0; n<SvIVX(sv_dat); n++ ) {
9694 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
9695 (n ? "," : ""), (IV)nums[n]);
9697 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
9701 } else if (k == GOSUB)
9702 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
9703 else if (k == VERB) {
9705 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
9706 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
9707 } else if (k == LOGICAL)
9708 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
9709 else if (k == FOLDCHAR)
9710 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
9711 else if (k == ANYOF) {
9712 int i, rangestart = -1;
9713 const U8 flags = ANYOF_FLAGS(o);
9716 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
9717 static const char * const anyofs[] = {
9750 if (flags & ANYOF_LOCALE)
9751 sv_catpvs(sv, "{loc}");
9752 if (flags & ANYOF_FOLD)
9753 sv_catpvs(sv, "{i}");
9754 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
9755 if (flags & ANYOF_INVERT)
9758 /* output what the standard cp 0-255 bitmap matches */
9759 for (i = 0; i <= 256; i++) {
9760 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
9761 if (rangestart == -1)
9763 } else if (rangestart != -1) {
9764 if (i <= rangestart + 3)
9765 for (; rangestart < i; rangestart++)
9766 put_byte(sv, rangestart);
9768 put_byte(sv, rangestart);
9770 put_byte(sv, i - 1);
9777 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9778 /* output any special charclass tests (used entirely under use locale) */
9779 if (ANYOF_CLASS_TEST_ANY_SET(o))
9780 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
9781 if (ANYOF_CLASS_TEST(o,i)) {
9782 sv_catpv(sv, anyofs[i]);
9786 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
9788 /* output information about the unicode matching */
9789 if (flags & ANYOF_UNICODE_ALL)
9790 sv_catpvs(sv, "{unicode_all}");
9791 else if (flags & ANYOF_UTF8)
9792 sv_catpvs(sv, "{unicode}");
9793 if (flags & ANYOF_NONBITMAP_NON_UTF8)
9794 sv_catpvs(sv, "{outside bitmap}");
9798 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
9802 U8 s[UTF8_MAXBYTES_CASE+1];
9804 for (i = 0; i <= 256; i++) { /* just the first 256 */
9805 uvchr_to_utf8(s, i);
9807 if (i < 256 && swash_fetch(sw, s, TRUE)) {
9808 if (rangestart == -1)
9810 } else if (rangestart != -1) {
9811 if (i <= rangestart + 3)
9812 for (; rangestart < i; rangestart++) {
9813 const U8 * const e = uvchr_to_utf8(s,rangestart);
9815 for(p = s; p < e; p++)
9819 const U8 *e = uvchr_to_utf8(s,rangestart);
9821 for (p = s; p < e; p++)
9824 e = uvchr_to_utf8(s, i-1);
9825 for (p = s; p < e; p++)
9832 sv_catpvs(sv, "..."); /* et cetera */
9836 char *s = savesvpv(lv);
9837 char * const origs = s;
9839 while (*s && *s != '\n')
9843 const char * const t = ++s;
9861 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
9863 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
9864 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
9866 PERL_UNUSED_CONTEXT;
9867 PERL_UNUSED_ARG(sv);
9869 PERL_UNUSED_ARG(prog);
9870 #endif /* DEBUGGING */
9874 Perl_re_intuit_string(pTHX_ REGEXP * const r)
9875 { /* Assume that RE_INTUIT is set */
9877 struct regexp *const prog = (struct regexp *)SvANY(r);
9878 GET_RE_DEBUG_FLAGS_DECL;
9880 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
9881 PERL_UNUSED_CONTEXT;
9885 const char * const s = SvPV_nolen_const(prog->check_substr
9886 ? prog->check_substr : prog->check_utf8);
9888 if (!PL_colorset) reginitcolors();
9889 PerlIO_printf(Perl_debug_log,
9890 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
9892 prog->check_substr ? "" : "utf8 ",
9893 PL_colors[5],PL_colors[0],
9896 (strlen(s) > 60 ? "..." : ""));
9899 return prog->check_substr ? prog->check_substr : prog->check_utf8;
9905 handles refcounting and freeing the perl core regexp structure. When
9906 it is necessary to actually free the structure the first thing it
9907 does is call the 'free' method of the regexp_engine associated to
9908 the regexp, allowing the handling of the void *pprivate; member
9909 first. (This routine is not overridable by extensions, which is why
9910 the extensions free is called first.)
9912 See regdupe and regdupe_internal if you change anything here.
9914 #ifndef PERL_IN_XSUB_RE
9916 Perl_pregfree(pTHX_ REGEXP *r)
9922 Perl_pregfree2(pTHX_ REGEXP *rx)
9925 struct regexp *const r = (struct regexp *)SvANY(rx);
9926 GET_RE_DEBUG_FLAGS_DECL;
9928 PERL_ARGS_ASSERT_PREGFREE2;
9931 ReREFCNT_dec(r->mother_re);
9933 CALLREGFREE_PVT(rx); /* free the private data */
9934 SvREFCNT_dec(RXp_PAREN_NAMES(r));
9937 SvREFCNT_dec(r->anchored_substr);
9938 SvREFCNT_dec(r->anchored_utf8);
9939 SvREFCNT_dec(r->float_substr);
9940 SvREFCNT_dec(r->float_utf8);
9941 Safefree(r->substrs);
9943 RX_MATCH_COPY_FREE(rx);
9944 #ifdef PERL_OLD_COPY_ON_WRITE
9945 SvREFCNT_dec(r->saved_copy);
9952 This is a hacky workaround to the structural issue of match results
9953 being stored in the regexp structure which is in turn stored in
9954 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
9955 could be PL_curpm in multiple contexts, and could require multiple
9956 result sets being associated with the pattern simultaneously, such
9957 as when doing a recursive match with (??{$qr})
9959 The solution is to make a lightweight copy of the regexp structure
9960 when a qr// is returned from the code executed by (??{$qr}) this
9961 lightweight copy doesnt actually own any of its data except for
9962 the starp/end and the actual regexp structure itself.
9968 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
9971 struct regexp *const r = (struct regexp *)SvANY(rx);
9972 register const I32 npar = r->nparens+1;
9974 PERL_ARGS_ASSERT_REG_TEMP_COPY;
9977 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
9978 ret = (struct regexp *)SvANY(ret_x);
9980 (void)ReREFCNT_inc(rx);
9981 /* We can take advantage of the existing "copied buffer" mechanism in SVs
9982 by pointing directly at the buffer, but flagging that the allocated
9983 space in the copy is zero. As we've just done a struct copy, it's now
9984 a case of zero-ing that, rather than copying the current length. */
9985 SvPV_set(ret_x, RX_WRAPPED(rx));
9986 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
9987 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
9988 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
9989 SvLEN_set(ret_x, 0);
9990 SvSTASH_set(ret_x, NULL);
9991 SvMAGIC_set(ret_x, NULL);
9992 Newx(ret->offs, npar, regexp_paren_pair);
9993 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
9995 Newx(ret->substrs, 1, struct reg_substr_data);
9996 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
9998 SvREFCNT_inc_void(ret->anchored_substr);
9999 SvREFCNT_inc_void(ret->anchored_utf8);
10000 SvREFCNT_inc_void(ret->float_substr);
10001 SvREFCNT_inc_void(ret->float_utf8);
10003 /* check_substr and check_utf8, if non-NULL, point to either their
10004 anchored or float namesakes, and don't hold a second reference. */
10006 RX_MATCH_COPIED_off(ret_x);
10007 #ifdef PERL_OLD_COPY_ON_WRITE
10008 ret->saved_copy = NULL;
10010 ret->mother_re = rx;
10016 /* regfree_internal()
10018 Free the private data in a regexp. This is overloadable by
10019 extensions. Perl takes care of the regexp structure in pregfree(),
10020 this covers the *pprivate pointer which technically perl doesn't
10021 know about, however of course we have to handle the
10022 regexp_internal structure when no extension is in use.
10024 Note this is called before freeing anything in the regexp
10029 Perl_regfree_internal(pTHX_ REGEXP * const rx)
10032 struct regexp *const r = (struct regexp *)SvANY(rx);
10033 RXi_GET_DECL(r,ri);
10034 GET_RE_DEBUG_FLAGS_DECL;
10036 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
10042 SV *dsv= sv_newmortal();
10043 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
10044 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
10045 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
10046 PL_colors[4],PL_colors[5],s);
10049 #ifdef RE_TRACK_PATTERN_OFFSETS
10051 Safefree(ri->u.offsets); /* 20010421 MJD */
10054 int n = ri->data->count;
10055 PAD* new_comppad = NULL;
10060 /* If you add a ->what type here, update the comment in regcomp.h */
10061 switch (ri->data->what[n]) {
10066 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
10069 Safefree(ri->data->data[n]);
10072 new_comppad = MUTABLE_AV(ri->data->data[n]);
10075 if (new_comppad == NULL)
10076 Perl_croak(aTHX_ "panic: pregfree comppad");
10077 PAD_SAVE_LOCAL(old_comppad,
10078 /* Watch out for global destruction's random ordering. */
10079 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
10082 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
10085 op_free((OP_4tree*)ri->data->data[n]);
10087 PAD_RESTORE_LOCAL(old_comppad);
10088 SvREFCNT_dec(MUTABLE_SV(new_comppad));
10089 new_comppad = NULL;
10094 { /* Aho Corasick add-on structure for a trie node.
10095 Used in stclass optimization only */
10097 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
10099 refcount = --aho->refcount;
10102 PerlMemShared_free(aho->states);
10103 PerlMemShared_free(aho->fail);
10104 /* do this last!!!! */
10105 PerlMemShared_free(ri->data->data[n]);
10106 PerlMemShared_free(ri->regstclass);
10112 /* trie structure. */
10114 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
10116 refcount = --trie->refcount;
10119 PerlMemShared_free(trie->charmap);
10120 PerlMemShared_free(trie->states);
10121 PerlMemShared_free(trie->trans);
10123 PerlMemShared_free(trie->bitmap);
10125 PerlMemShared_free(trie->jump);
10126 PerlMemShared_free(trie->wordinfo);
10127 /* do this last!!!! */
10128 PerlMemShared_free(ri->data->data[n]);
10133 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
10136 Safefree(ri->data->what);
10137 Safefree(ri->data);
10143 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
10144 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
10145 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
10148 re_dup - duplicate a regexp.
10150 This routine is expected to clone a given regexp structure. It is only
10151 compiled under USE_ITHREADS.
10153 After all of the core data stored in struct regexp is duplicated
10154 the regexp_engine.dupe method is used to copy any private data
10155 stored in the *pprivate pointer. This allows extensions to handle
10156 any duplication it needs to do.
10158 See pregfree() and regfree_internal() if you change anything here.
10160 #if defined(USE_ITHREADS)
10161 #ifndef PERL_IN_XSUB_RE
10163 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
10167 const struct regexp *r = (const struct regexp *)SvANY(sstr);
10168 struct regexp *ret = (struct regexp *)SvANY(dstr);
10170 PERL_ARGS_ASSERT_RE_DUP_GUTS;
10172 npar = r->nparens+1;
10173 Newx(ret->offs, npar, regexp_paren_pair);
10174 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
10176 /* no need to copy these */
10177 Newx(ret->swap, npar, regexp_paren_pair);
10180 if (ret->substrs) {
10181 /* Do it this way to avoid reading from *r after the StructCopy().
10182 That way, if any of the sv_dup_inc()s dislodge *r from the L1
10183 cache, it doesn't matter. */
10184 const bool anchored = r->check_substr
10185 ? r->check_substr == r->anchored_substr
10186 : r->check_utf8 == r->anchored_utf8;
10187 Newx(ret->substrs, 1, struct reg_substr_data);
10188 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
10190 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
10191 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
10192 ret->float_substr = sv_dup_inc(ret->float_substr, param);
10193 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
10195 /* check_substr and check_utf8, if non-NULL, point to either their
10196 anchored or float namesakes, and don't hold a second reference. */
10198 if (ret->check_substr) {
10200 assert(r->check_utf8 == r->anchored_utf8);
10201 ret->check_substr = ret->anchored_substr;
10202 ret->check_utf8 = ret->anchored_utf8;
10204 assert(r->check_substr == r->float_substr);
10205 assert(r->check_utf8 == r->float_utf8);
10206 ret->check_substr = ret->float_substr;
10207 ret->check_utf8 = ret->float_utf8;
10209 } else if (ret->check_utf8) {
10211 ret->check_utf8 = ret->anchored_utf8;
10213 ret->check_utf8 = ret->float_utf8;
10218 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
10221 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
10223 if (RX_MATCH_COPIED(dstr))
10224 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
10226 ret->subbeg = NULL;
10227 #ifdef PERL_OLD_COPY_ON_WRITE
10228 ret->saved_copy = NULL;
10231 if (ret->mother_re) {
10232 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
10233 /* Our storage points directly to our mother regexp, but that's
10234 1: a buffer in a different thread
10235 2: something we no longer hold a reference on
10236 so we need to copy it locally. */
10237 /* Note we need to sue SvCUR() on our mother_re, because it, in
10238 turn, may well be pointing to its own mother_re. */
10239 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
10240 SvCUR(ret->mother_re)+1));
10241 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
10243 ret->mother_re = NULL;
10247 #endif /* PERL_IN_XSUB_RE */
10252 This is the internal complement to regdupe() which is used to copy
10253 the structure pointed to by the *pprivate pointer in the regexp.
10254 This is the core version of the extension overridable cloning hook.
10255 The regexp structure being duplicated will be copied by perl prior
10256 to this and will be provided as the regexp *r argument, however
10257 with the /old/ structures pprivate pointer value. Thus this routine
10258 may override any copying normally done by perl.
10260 It returns a pointer to the new regexp_internal structure.
10264 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
10267 struct regexp *const r = (struct regexp *)SvANY(rx);
10268 regexp_internal *reti;
10270 RXi_GET_DECL(r,ri);
10272 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
10274 npar = r->nparens+1;
10277 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
10278 Copy(ri->program, reti->program, len+1, regnode);
10281 reti->regstclass = NULL;
10284 struct reg_data *d;
10285 const int count = ri->data->count;
10288 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
10289 char, struct reg_data);
10290 Newx(d->what, count, U8);
10293 for (i = 0; i < count; i++) {
10294 d->what[i] = ri->data->what[i];
10295 switch (d->what[i]) {
10296 /* legal options are one of: sSfpontTua
10297 see also regcomp.h and pregfree() */
10298 case 'a': /* actually an AV, but the dup function is identical. */
10301 case 'p': /* actually an AV, but the dup function is identical. */
10302 case 'u': /* actually an HV, but the dup function is identical. */
10303 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
10306 /* This is cheating. */
10307 Newx(d->data[i], 1, struct regnode_charclass_class);
10308 StructCopy(ri->data->data[i], d->data[i],
10309 struct regnode_charclass_class);
10310 reti->regstclass = (regnode*)d->data[i];
10313 /* Compiled op trees are readonly and in shared memory,
10314 and can thus be shared without duplication. */
10316 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
10320 /* Trie stclasses are readonly and can thus be shared
10321 * without duplication. We free the stclass in pregfree
10322 * when the corresponding reg_ac_data struct is freed.
10324 reti->regstclass= ri->regstclass;
10328 ((reg_trie_data*)ri->data->data[i])->refcount++;
10332 d->data[i] = ri->data->data[i];
10335 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
10344 reti->name_list_idx = ri->name_list_idx;
10346 #ifdef RE_TRACK_PATTERN_OFFSETS
10347 if (ri->u.offsets) {
10348 Newx(reti->u.offsets, 2*len+1, U32);
10349 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
10352 SetProgLen(reti,len);
10355 return (void*)reti;
10358 #endif /* USE_ITHREADS */
10360 #ifndef PERL_IN_XSUB_RE
10363 - regnext - dig the "next" pointer out of a node
10366 Perl_regnext(pTHX_ register regnode *p)
10369 register I32 offset;
10374 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
10375 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
10378 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
10387 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
10390 STRLEN l1 = strlen(pat1);
10391 STRLEN l2 = strlen(pat2);
10394 const char *message;
10396 PERL_ARGS_ASSERT_RE_CROAK2;
10402 Copy(pat1, buf, l1 , char);
10403 Copy(pat2, buf + l1, l2 , char);
10404 buf[l1 + l2] = '\n';
10405 buf[l1 + l2 + 1] = '\0';
10407 /* ANSI variant takes additional second argument */
10408 va_start(args, pat2);
10412 msv = vmess(buf, &args);
10414 message = SvPV_const(msv,l1);
10417 Copy(message, buf, l1 , char);
10418 buf[l1-1] = '\0'; /* Overwrite \n */
10419 Perl_croak(aTHX_ "%s", buf);
10422 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
10424 #ifndef PERL_IN_XSUB_RE
10426 Perl_save_re_context(pTHX)
10430 struct re_save_state *state;
10432 SAVEVPTR(PL_curcop);
10433 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
10435 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
10436 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
10437 SSPUSHUV(SAVEt_RE_STATE);
10439 Copy(&PL_reg_state, state, 1, struct re_save_state);
10441 PL_reg_start_tmp = 0;
10442 PL_reg_start_tmpl = 0;
10443 PL_reg_oldsaved = NULL;
10444 PL_reg_oldsavedlen = 0;
10445 PL_reg_maxiter = 0;
10446 PL_reg_leftiter = 0;
10447 PL_reg_poscache = NULL;
10448 PL_reg_poscache_size = 0;
10449 #ifdef PERL_OLD_COPY_ON_WRITE
10453 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
10455 const REGEXP * const rx = PM_GETRE(PL_curpm);
10458 for (i = 1; i <= RX_NPARENS(rx); i++) {
10459 char digits[TYPE_CHARS(long)];
10460 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
10461 GV *const *const gvp
10462 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
10465 GV * const gv = *gvp;
10466 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
10476 clear_re(pTHX_ void *r)
10479 ReREFCNT_dec((REGEXP *)r);
10485 S_put_byte(pTHX_ SV *sv, int c)
10487 PERL_ARGS_ASSERT_PUT_BYTE;
10489 /* Our definition of isPRINT() ignores locales, so only bytes that are
10490 not part of UTF-8 are considered printable. I assume that the same
10491 holds for UTF-EBCDIC.
10492 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
10493 which Wikipedia says:
10495 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
10496 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
10497 identical, to the ASCII delete (DEL) or rubout control character.
10498 ) So the old condition can be simplified to !isPRINT(c) */
10501 Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c);
10504 Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c);
10508 const char string = c;
10509 if (c == '-' || c == ']' || c == '\\' || c == '^')
10510 sv_catpvs(sv, "\\");
10511 sv_catpvn(sv, &string, 1);
10516 #define CLEAR_OPTSTART \
10517 if (optstart) STMT_START { \
10518 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
10522 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
10524 STATIC const regnode *
10525 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
10526 const regnode *last, const regnode *plast,
10527 SV* sv, I32 indent, U32 depth)
10530 register U8 op = PSEUDO; /* Arbitrary non-END op. */
10531 register const regnode *next;
10532 const regnode *optstart= NULL;
10534 RXi_GET_DECL(r,ri);
10535 GET_RE_DEBUG_FLAGS_DECL;
10537 PERL_ARGS_ASSERT_DUMPUNTIL;
10539 #ifdef DEBUG_DUMPUNTIL
10540 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
10541 last ? last-start : 0,plast ? plast-start : 0);
10544 if (plast && plast < last)
10547 while (PL_regkind[op] != END && (!last || node < last)) {
10548 /* While that wasn't END last time... */
10551 if (op == CLOSE || op == WHILEM)
10553 next = regnext((regnode *)node);
10556 if (OP(node) == OPTIMIZED) {
10557 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
10564 regprop(r, sv, node);
10565 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
10566 (int)(2*indent + 1), "", SvPVX_const(sv));
10568 if (OP(node) != OPTIMIZED) {
10569 if (next == NULL) /* Next ptr. */
10570 PerlIO_printf(Perl_debug_log, " (0)");
10571 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
10572 PerlIO_printf(Perl_debug_log, " (FAIL)");
10574 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
10575 (void)PerlIO_putc(Perl_debug_log, '\n');
10579 if (PL_regkind[(U8)op] == BRANCHJ) {
10582 register const regnode *nnode = (OP(next) == LONGJMP
10583 ? regnext((regnode *)next)
10585 if (last && nnode > last)
10587 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
10590 else if (PL_regkind[(U8)op] == BRANCH) {
10592 DUMPUNTIL(NEXTOPER(node), next);
10594 else if ( PL_regkind[(U8)op] == TRIE ) {
10595 const regnode *this_trie = node;
10596 const char op = OP(node);
10597 const U32 n = ARG(node);
10598 const reg_ac_data * const ac = op>=AHOCORASICK ?
10599 (reg_ac_data *)ri->data->data[n] :
10601 const reg_trie_data * const trie =
10602 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
10604 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
10606 const regnode *nextbranch= NULL;
10609 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
10610 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
10612 PerlIO_printf(Perl_debug_log, "%*s%s ",
10613 (int)(2*(indent+3)), "",
10614 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
10615 PL_colors[0], PL_colors[1],
10616 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
10617 PERL_PV_PRETTY_ELLIPSES |
10618 PERL_PV_PRETTY_LTGT
10623 U16 dist= trie->jump[word_idx+1];
10624 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
10625 (UV)((dist ? this_trie + dist : next) - start));
10628 nextbranch= this_trie + trie->jump[0];
10629 DUMPUNTIL(this_trie + dist, nextbranch);
10631 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
10632 nextbranch= regnext((regnode *)nextbranch);
10634 PerlIO_printf(Perl_debug_log, "\n");
10637 if (last && next > last)
10642 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
10643 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
10644 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
10646 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
10648 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
10650 else if ( op == PLUS || op == STAR) {
10651 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
10653 else if (op == ANYOF) {
10654 /* arglen 1 + class block */
10655 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS)
10656 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
10657 node = NEXTOPER(node);
10659 else if (PL_regkind[(U8)op] == EXACT) {
10660 /* Literal string, where present. */
10661 node += NODE_SZ_STR(node) - 1;
10662 node = NEXTOPER(node);
10665 node = NEXTOPER(node);
10666 node += regarglen[(U8)op];
10668 if (op == CURLYX || op == OPEN)
10672 #ifdef DEBUG_DUMPUNTIL
10673 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
10678 #endif /* DEBUGGING */
10682 * c-indentation-style: bsd
10683 * c-basic-offset: 4
10684 * indent-tabs-mode: t
10687 * ex: set ts=8 sts=4 sw=4 noet: