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 I32 uni_semantics; /* If a d charset modifier should use unicode
138 rules, even if the pattern is not in
140 HV *paren_names; /* Paren names */
142 regnode **recurse; /* Recurse regops */
143 I32 recurse_count; /* Number of recurse regops */
146 I32 override_recoding;
148 char *starttry; /* -Dr: where regtry was called. */
149 #define RExC_starttry (pRExC_state->starttry)
152 const char *lastparse;
154 AV *paren_name_list; /* idx -> name */
155 #define RExC_lastparse (pRExC_state->lastparse)
156 #define RExC_lastnum (pRExC_state->lastnum)
157 #define RExC_paren_name_list (pRExC_state->paren_name_list)
161 #define RExC_flags (pRExC_state->flags)
162 #define RExC_precomp (pRExC_state->precomp)
163 #define RExC_rx_sv (pRExC_state->rx_sv)
164 #define RExC_rx (pRExC_state->rx)
165 #define RExC_rxi (pRExC_state->rxi)
166 #define RExC_start (pRExC_state->start)
167 #define RExC_end (pRExC_state->end)
168 #define RExC_parse (pRExC_state->parse)
169 #define RExC_whilem_seen (pRExC_state->whilem_seen)
170 #ifdef RE_TRACK_PATTERN_OFFSETS
171 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
173 #define RExC_emit (pRExC_state->emit)
174 #define RExC_emit_start (pRExC_state->emit_start)
175 #define RExC_emit_bound (pRExC_state->emit_bound)
176 #define RExC_naughty (pRExC_state->naughty)
177 #define RExC_sawback (pRExC_state->sawback)
178 #define RExC_seen (pRExC_state->seen)
179 #define RExC_size (pRExC_state->size)
180 #define RExC_npar (pRExC_state->npar)
181 #define RExC_nestroot (pRExC_state->nestroot)
182 #define RExC_extralen (pRExC_state->extralen)
183 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
184 #define RExC_seen_evals (pRExC_state->seen_evals)
185 #define RExC_utf8 (pRExC_state->utf8)
186 #define RExC_uni_semantics (pRExC_state->uni_semantics)
187 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
188 #define RExC_open_parens (pRExC_state->open_parens)
189 #define RExC_close_parens (pRExC_state->close_parens)
190 #define RExC_opend (pRExC_state->opend)
191 #define RExC_paren_names (pRExC_state->paren_names)
192 #define RExC_recurse (pRExC_state->recurse)
193 #define RExC_recurse_count (pRExC_state->recurse_count)
194 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
195 #define RExC_contains_locale (pRExC_state->contains_locale)
196 #define RExC_override_recoding (pRExC_state->override_recoding)
199 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
200 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
201 ((*s) == '{' && regcurly(s)))
204 #undef SPSTART /* dratted cpp namespace... */
207 * Flags to be passed up and down.
209 #define WORST 0 /* Worst case. */
210 #define HASWIDTH 0x01 /* Known to match non-null strings. */
212 /* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single
213 * character, and if utf8, must be invariant. Note that this is not the same thing as REGNODE_SIMPLE */
215 #define SPSTART 0x04 /* Starts with * or +. */
216 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
217 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
219 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
221 /* whether trie related optimizations are enabled */
222 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
223 #define TRIE_STUDY_OPT
224 #define FULL_TRIE_STUDY
230 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
231 #define PBITVAL(paren) (1 << ((paren) & 7))
232 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
233 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
234 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
236 /* If not already in utf8, do a longjmp back to the beginning */
237 #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */
238 #define REQUIRE_UTF8 STMT_START { \
239 if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \
242 /* About scan_data_t.
244 During optimisation we recurse through the regexp program performing
245 various inplace (keyhole style) optimisations. In addition study_chunk
246 and scan_commit populate this data structure with information about
247 what strings MUST appear in the pattern. We look for the longest
248 string that must appear at a fixed location, and we look for the
249 longest string that may appear at a floating location. So for instance
254 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
255 strings (because they follow a .* construct). study_chunk will identify
256 both FOO and BAR as being the longest fixed and floating strings respectively.
258 The strings can be composites, for instance
262 will result in a composite fixed substring 'foo'.
264 For each string some basic information is maintained:
266 - offset or min_offset
267 This is the position the string must appear at, or not before.
268 It also implicitly (when combined with minlenp) tells us how many
269 characters must match before the string we are searching for.
270 Likewise when combined with minlenp and the length of the string it
271 tells us how many characters must appear after the string we have
275 Only used for floating strings. This is the rightmost point that
276 the string can appear at. If set to I32 max it indicates that the
277 string can occur infinitely far to the right.
280 A pointer to the minimum length of the pattern that the string
281 was found inside. This is important as in the case of positive
282 lookahead or positive lookbehind we can have multiple patterns
287 The minimum length of the pattern overall is 3, the minimum length
288 of the lookahead part is 3, but the minimum length of the part that
289 will actually match is 1. So 'FOO's minimum length is 3, but the
290 minimum length for the F is 1. This is important as the minimum length
291 is used to determine offsets in front of and behind the string being
292 looked for. Since strings can be composites this is the length of the
293 pattern at the time it was committed with a scan_commit. Note that
294 the length is calculated by study_chunk, so that the minimum lengths
295 are not known until the full pattern has been compiled, thus the
296 pointer to the value.
300 In the case of lookbehind the string being searched for can be
301 offset past the start point of the final matching string.
302 If this value was just blithely removed from the min_offset it would
303 invalidate some of the calculations for how many chars must match
304 before or after (as they are derived from min_offset and minlen and
305 the length of the string being searched for).
306 When the final pattern is compiled and the data is moved from the
307 scan_data_t structure into the regexp structure the information
308 about lookbehind is factored in, with the information that would
309 have been lost precalculated in the end_shift field for the
312 The fields pos_min and pos_delta are used to store the minimum offset
313 and the delta to the maximum offset at the current point in the pattern.
317 typedef struct scan_data_t {
318 /*I32 len_min; unused */
319 /*I32 len_delta; unused */
323 I32 last_end; /* min value, <0 unless valid. */
326 SV **longest; /* Either &l_fixed, or &l_float. */
327 SV *longest_fixed; /* longest fixed string found in pattern */
328 I32 offset_fixed; /* offset where it starts */
329 I32 *minlen_fixed; /* pointer to the minlen relevant to the string */
330 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
331 SV *longest_float; /* longest floating string found in pattern */
332 I32 offset_float_min; /* earliest point in string it can appear */
333 I32 offset_float_max; /* latest point in string it can appear */
334 I32 *minlen_float; /* pointer to the minlen relevant to the string */
335 I32 lookbehind_float; /* is the position of the string modified by LB */
339 struct regnode_charclass_class *start_class;
343 * Forward declarations for pregcomp()'s friends.
346 static const scan_data_t zero_scan_data =
347 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
349 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
350 #define SF_BEFORE_SEOL 0x0001
351 #define SF_BEFORE_MEOL 0x0002
352 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
353 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
356 # define SF_FIX_SHIFT_EOL (0+2)
357 # define SF_FL_SHIFT_EOL (0+4)
359 # define SF_FIX_SHIFT_EOL (+2)
360 # define SF_FL_SHIFT_EOL (+4)
363 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
364 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
366 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
367 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
368 #define SF_IS_INF 0x0040
369 #define SF_HAS_PAR 0x0080
370 #define SF_IN_PAR 0x0100
371 #define SF_HAS_EVAL 0x0200
372 #define SCF_DO_SUBSTR 0x0400
373 #define SCF_DO_STCLASS_AND 0x0800
374 #define SCF_DO_STCLASS_OR 0x1000
375 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
376 #define SCF_WHILEM_VISITED_POS 0x2000
378 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
379 #define SCF_SEEN_ACCEPT 0x8000
381 #define UTF cBOOL(RExC_utf8)
382 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
383 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
384 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_DEPENDS_CHARSET)
385 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) >= REGEX_UNICODE_CHARSET)
386 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_RESTRICTED_CHARSET)
387 #define MORE_ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
388 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) >= REGEX_ASCII_RESTRICTED_CHARSET)
390 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
392 #define OOB_UNICODE 12345678
393 #define OOB_NAMEDCLASS -1
395 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
396 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
399 /* length of regex to show in messages that don't mark a position within */
400 #define RegexLengthToShowInErrorMessages 127
403 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
404 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
405 * op/pragma/warn/regcomp.
407 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
408 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
410 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
413 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
414 * arg. Show regex, up to a maximum length. If it's too long, chop and add
417 #define _FAIL(code) STMT_START { \
418 const char *ellipses = ""; \
419 IV len = RExC_end - RExC_precomp; \
422 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
423 if (len > RegexLengthToShowInErrorMessages) { \
424 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
425 len = RegexLengthToShowInErrorMessages - 10; \
431 #define FAIL(msg) _FAIL( \
432 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
433 msg, (int)len, RExC_precomp, ellipses))
435 #define FAIL2(msg,arg) _FAIL( \
436 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
437 arg, (int)len, RExC_precomp, ellipses))
440 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
442 #define Simple_vFAIL(m) STMT_START { \
443 const IV offset = RExC_parse - RExC_precomp; \
444 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
445 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
449 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
451 #define vFAIL(m) STMT_START { \
453 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
458 * Like Simple_vFAIL(), but accepts two arguments.
460 #define Simple_vFAIL2(m,a1) STMT_START { \
461 const IV offset = RExC_parse - RExC_precomp; \
462 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
463 (int)offset, RExC_precomp, RExC_precomp + offset); \
467 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
469 #define vFAIL2(m,a1) STMT_START { \
471 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
472 Simple_vFAIL2(m, a1); \
477 * Like Simple_vFAIL(), but accepts three arguments.
479 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
480 const IV offset = RExC_parse - RExC_precomp; \
481 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
482 (int)offset, RExC_precomp, RExC_precomp + offset); \
486 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
488 #define vFAIL3(m,a1,a2) STMT_START { \
490 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
491 Simple_vFAIL3(m, a1, a2); \
495 * Like Simple_vFAIL(), but accepts four arguments.
497 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
498 const IV offset = RExC_parse - RExC_precomp; \
499 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
500 (int)offset, RExC_precomp, RExC_precomp + offset); \
503 #define ckWARNreg(loc,m) STMT_START { \
504 const IV offset = loc - RExC_precomp; \
505 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
506 (int)offset, RExC_precomp, RExC_precomp + offset); \
509 #define ckWARNregdep(loc,m) STMT_START { \
510 const IV offset = loc - RExC_precomp; \
511 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
513 (int)offset, RExC_precomp, RExC_precomp + offset); \
516 #define ckWARN2regdep(loc,m, a1) STMT_START { \
517 const IV offset = loc - RExC_precomp; \
518 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
520 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
523 #define ckWARN2reg(loc, m, a1) STMT_START { \
524 const IV offset = loc - RExC_precomp; \
525 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
526 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
529 #define vWARN3(loc, m, a1, a2) STMT_START { \
530 const IV offset = loc - RExC_precomp; \
531 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
532 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
535 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
536 const IV offset = loc - RExC_precomp; \
537 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
538 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
541 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
542 const IV offset = loc - RExC_precomp; \
543 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
544 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
547 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
548 const IV offset = loc - RExC_precomp; \
549 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
550 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
553 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
554 const IV offset = loc - RExC_precomp; \
555 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
556 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
560 /* Allow for side effects in s */
561 #define REGC(c,s) STMT_START { \
562 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
565 /* Macros for recording node offsets. 20001227 mjd@plover.com
566 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
567 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
568 * Element 0 holds the number n.
569 * Position is 1 indexed.
571 #ifndef RE_TRACK_PATTERN_OFFSETS
572 #define Set_Node_Offset_To_R(node,byte)
573 #define Set_Node_Offset(node,byte)
574 #define Set_Cur_Node_Offset
575 #define Set_Node_Length_To_R(node,len)
576 #define Set_Node_Length(node,len)
577 #define Set_Node_Cur_Length(node)
578 #define Node_Offset(n)
579 #define Node_Length(n)
580 #define Set_Node_Offset_Length(node,offset,len)
581 #define ProgLen(ri) ri->u.proglen
582 #define SetProgLen(ri,x) ri->u.proglen = x
584 #define ProgLen(ri) ri->u.offsets[0]
585 #define SetProgLen(ri,x) ri->u.offsets[0] = x
586 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
588 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
589 __LINE__, (int)(node), (int)(byte))); \
591 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
593 RExC_offsets[2*(node)-1] = (byte); \
598 #define Set_Node_Offset(node,byte) \
599 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
600 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
602 #define Set_Node_Length_To_R(node,len) STMT_START { \
604 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
605 __LINE__, (int)(node), (int)(len))); \
607 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
609 RExC_offsets[2*(node)] = (len); \
614 #define Set_Node_Length(node,len) \
615 Set_Node_Length_To_R((node)-RExC_emit_start, len)
616 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
617 #define Set_Node_Cur_Length(node) \
618 Set_Node_Length(node, RExC_parse - parse_start)
620 /* Get offsets and lengths */
621 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
622 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
624 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
625 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
626 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
630 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
631 #define EXPERIMENTAL_INPLACESCAN
632 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
634 #define DEBUG_STUDYDATA(str,data,depth) \
635 DEBUG_OPTIMISE_MORE_r(if(data){ \
636 PerlIO_printf(Perl_debug_log, \
637 "%*s" str "Pos:%"IVdf"/%"IVdf \
638 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
639 (int)(depth)*2, "", \
640 (IV)((data)->pos_min), \
641 (IV)((data)->pos_delta), \
642 (UV)((data)->flags), \
643 (IV)((data)->whilem_c), \
644 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
645 is_inf ? "INF " : "" \
647 if ((data)->last_found) \
648 PerlIO_printf(Perl_debug_log, \
649 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
650 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
651 SvPVX_const((data)->last_found), \
652 (IV)((data)->last_end), \
653 (IV)((data)->last_start_min), \
654 (IV)((data)->last_start_max), \
655 ((data)->longest && \
656 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
657 SvPVX_const((data)->longest_fixed), \
658 (IV)((data)->offset_fixed), \
659 ((data)->longest && \
660 (data)->longest==&((data)->longest_float)) ? "*" : "", \
661 SvPVX_const((data)->longest_float), \
662 (IV)((data)->offset_float_min), \
663 (IV)((data)->offset_float_max) \
665 PerlIO_printf(Perl_debug_log,"\n"); \
668 static void clear_re(pTHX_ void *r);
670 /* Mark that we cannot extend a found fixed substring at this point.
671 Update the longest found anchored substring and the longest found
672 floating substrings if needed. */
675 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
677 const STRLEN l = CHR_SVLEN(data->last_found);
678 const STRLEN old_l = CHR_SVLEN(*data->longest);
679 GET_RE_DEBUG_FLAGS_DECL;
681 PERL_ARGS_ASSERT_SCAN_COMMIT;
683 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
684 SvSetMagicSV(*data->longest, data->last_found);
685 if (*data->longest == data->longest_fixed) {
686 data->offset_fixed = l ? data->last_start_min : data->pos_min;
687 if (data->flags & SF_BEFORE_EOL)
689 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
691 data->flags &= ~SF_FIX_BEFORE_EOL;
692 data->minlen_fixed=minlenp;
693 data->lookbehind_fixed=0;
695 else { /* *data->longest == data->longest_float */
696 data->offset_float_min = l ? data->last_start_min : data->pos_min;
697 data->offset_float_max = (l
698 ? data->last_start_max
699 : data->pos_min + data->pos_delta);
700 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
701 data->offset_float_max = I32_MAX;
702 if (data->flags & SF_BEFORE_EOL)
704 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
706 data->flags &= ~SF_FL_BEFORE_EOL;
707 data->minlen_float=minlenp;
708 data->lookbehind_float=0;
711 SvCUR_set(data->last_found, 0);
713 SV * const sv = data->last_found;
714 if (SvUTF8(sv) && SvMAGICAL(sv)) {
715 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
721 data->flags &= ~SF_BEFORE_EOL;
722 DEBUG_STUDYDATA("commit: ",data,0);
725 /* Can match anything (initialization) */
727 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
729 PERL_ARGS_ASSERT_CL_ANYTHING;
731 ANYOF_BITMAP_SETALL(cl);
732 cl->flags = ANYOF_CLASS|ANYOF_EOS|ANYOF_UNICODE_ALL
733 |ANYOF_LOC_NONBITMAP_FOLD|ANYOF_NON_UTF8_LATIN1_ALL;
735 /* If any portion of the regex is to operate under locale rules,
736 * initialization includes it. The reason this isn't done for all regexes
737 * is that the optimizer was written under the assumption that locale was
738 * all-or-nothing. Given the complexity and lack of documentation in the
739 * optimizer, and that there are inadequate test cases for locale, so many
740 * parts of it may not work properly, it is safest to avoid locale unless
742 if (RExC_contains_locale) {
743 ANYOF_CLASS_SETALL(cl); /* /l uses class */
744 cl->flags |= ANYOF_LOCALE;
747 ANYOF_CLASS_ZERO(cl); /* Only /l uses class now */
751 /* Can match anything (initialization) */
753 S_cl_is_anything(const struct regnode_charclass_class *cl)
757 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
759 for (value = 0; value <= ANYOF_MAX; value += 2)
760 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
762 if (!(cl->flags & ANYOF_UNICODE_ALL))
764 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
769 /* Can match anything (initialization) */
771 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
773 PERL_ARGS_ASSERT_CL_INIT;
775 Zero(cl, 1, struct regnode_charclass_class);
777 cl_anything(pRExC_state, cl);
778 ARG_SET(cl, ANYOF_NONBITMAP_EMPTY);
781 /* These two functions currently do the exact same thing */
782 #define cl_init_zero S_cl_init
784 /* 'AND' a given class with another one. Can create false positives. 'cl'
785 * should not be inverted. 'and_with->flags & ANYOF_CLASS' should be 0 if
786 * 'and_with' is a regnode_charclass instead of a regnode_charclass_class. */
788 S_cl_and(struct regnode_charclass_class *cl,
789 const struct regnode_charclass_class *and_with)
791 PERL_ARGS_ASSERT_CL_AND;
793 assert(and_with->type == ANYOF);
795 /* I (khw) am not sure all these restrictions are necessary XXX */
796 if (!(ANYOF_CLASS_TEST_ANY_SET(and_with))
797 && !(ANYOF_CLASS_TEST_ANY_SET(cl))
798 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
799 && !(and_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
800 && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) {
803 if (and_with->flags & ANYOF_INVERT)
804 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
805 cl->bitmap[i] &= ~and_with->bitmap[i];
807 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
808 cl->bitmap[i] &= and_with->bitmap[i];
809 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
811 if (and_with->flags & ANYOF_INVERT) {
813 /* Here, the and'ed node is inverted. Get the AND of the flags that
814 * aren't affected by the inversion. Those that are affected are
815 * handled individually below */
816 U8 affected_flags = cl->flags & ~INVERSION_UNAFFECTED_FLAGS;
817 cl->flags &= (and_with->flags & INVERSION_UNAFFECTED_FLAGS);
818 cl->flags |= affected_flags;
820 /* We currently don't know how to deal with things that aren't in the
821 * bitmap, but we know that the intersection is no greater than what
822 * is already in cl, so let there be false positives that get sorted
823 * out after the synthetic start class succeeds, and the node is
824 * matched for real. */
826 /* The inversion of these two flags indicate that the resulting
827 * intersection doesn't have them */
828 if (and_with->flags & ANYOF_UNICODE_ALL) {
829 cl->flags &= ~ANYOF_UNICODE_ALL;
831 if (and_with->flags & ANYOF_NON_UTF8_LATIN1_ALL) {
832 cl->flags &= ~ANYOF_NON_UTF8_LATIN1_ALL;
835 else { /* and'd node is not inverted */
836 U8 outside_bitmap_but_not_utf8; /* Temp variable */
838 if (! ANYOF_NONBITMAP(and_with)) {
840 /* Here 'and_with' doesn't match anything outside the bitmap
841 * (except possibly ANYOF_UNICODE_ALL), which means the
842 * intersection can't either, except for ANYOF_UNICODE_ALL, in
843 * which case we don't know what the intersection is, but it's no
844 * greater than what cl already has, so can just leave it alone,
845 * with possible false positives */
846 if (! (and_with->flags & ANYOF_UNICODE_ALL)) {
847 ARG_SET(cl, ANYOF_NONBITMAP_EMPTY);
848 cl->flags &= ~ANYOF_NONBITMAP_NON_UTF8;
851 else if (! ANYOF_NONBITMAP(cl)) {
853 /* Here, 'and_with' does match something outside the bitmap, and cl
854 * doesn't have a list of things to match outside the bitmap. If
855 * cl can match all code points above 255, the intersection will
856 * be those above-255 code points that 'and_with' matches. If cl
857 * can't match all Unicode code points, it means that it can't
858 * match anything outside the bitmap (since the 'if' that got us
859 * into this block tested for that), so we leave the bitmap empty.
861 if (cl->flags & ANYOF_UNICODE_ALL) {
862 ARG_SET(cl, ARG(and_with));
864 /* and_with's ARG may match things that don't require UTF8.
865 * And now cl's will too, in spite of this being an 'and'. See
866 * the comments below about the kludge */
867 cl->flags |= and_with->flags & ANYOF_NONBITMAP_NON_UTF8;
871 /* Here, both 'and_with' and cl match something outside the
872 * bitmap. Currently we do not do the intersection, so just match
873 * whatever cl had at the beginning. */
877 /* Take the intersection of the two sets of flags. However, the
878 * ANYOF_NONBITMAP_NON_UTF8 flag is treated as an 'or'. This is a
879 * kludge around the fact that this flag is not treated like the others
880 * which are initialized in cl_anything(). The way the optimizer works
881 * is that the synthetic start class (SSC) is initialized to match
882 * anything, and then the first time a real node is encountered, its
883 * values are AND'd with the SSC's with the result being the values of
884 * the real node. However, there are paths through the optimizer where
885 * the AND never gets called, so those initialized bits are set
886 * inappropriately, which is not usually a big deal, as they just cause
887 * false positives in the SSC, which will just mean a probably
888 * imperceptible slow down in execution. However this bit has a
889 * higher false positive consequence in that it can cause utf8.pm,
890 * utf8_heavy.pl ... to be loaded when not necessary, which is a much
891 * bigger slowdown and also causes significant extra memory to be used.
892 * In order to prevent this, the code now takes a different tack. The
893 * bit isn't set unless some part of the regular expression needs it,
894 * but once set it won't get cleared. This means that these extra
895 * modules won't get loaded unless there was some path through the
896 * pattern that would have required them anyway, and so any false
897 * positives that occur by not ANDing them out when they could be
898 * aren't as severe as they would be if we treated this bit like all
900 outside_bitmap_but_not_utf8 = (cl->flags | and_with->flags)
901 & ANYOF_NONBITMAP_NON_UTF8;
902 cl->flags &= and_with->flags;
903 cl->flags |= outside_bitmap_but_not_utf8;
907 /* 'OR' a given class with another one. Can create false positives. 'cl'
908 * should not be inverted. 'or_with->flags & ANYOF_CLASS' should be 0 if
909 * 'or_with' is a regnode_charclass instead of a regnode_charclass_class. */
911 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
913 PERL_ARGS_ASSERT_CL_OR;
915 if (or_with->flags & ANYOF_INVERT) {
917 /* Here, the or'd node is to be inverted. This means we take the
918 * complement of everything not in the bitmap, but currently we don't
919 * know what that is, so give up and match anything */
920 if (ANYOF_NONBITMAP(or_with)) {
921 cl_anything(pRExC_state, cl);
924 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
925 * <= (B1 | !B2) | (CL1 | !CL2)
926 * which is wasteful if CL2 is small, but we ignore CL2:
927 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
928 * XXXX Can we handle case-fold? Unclear:
929 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
930 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
932 else if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
933 && !(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
934 && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD) ) {
937 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
938 cl->bitmap[i] |= ~or_with->bitmap[i];
939 } /* XXXX: logic is complicated otherwise */
941 cl_anything(pRExC_state, cl);
944 /* And, we can just take the union of the flags that aren't affected
945 * by the inversion */
946 cl->flags |= or_with->flags & INVERSION_UNAFFECTED_FLAGS;
948 /* For the remaining flags:
949 ANYOF_UNICODE_ALL and inverted means to not match anything above
950 255, which means that the union with cl should just be
951 what cl has in it, so can ignore this flag
952 ANYOF_NON_UTF8_LATIN1_ALL and inverted means if not utf8 and ord
953 is 127-255 to match them, but then invert that, so the
954 union with cl should just be what cl has in it, so can
957 } else { /* 'or_with' is not inverted */
958 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
959 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
960 && (!(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
961 || (cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) ) {
964 /* OR char bitmap and class bitmap separately */
965 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
966 cl->bitmap[i] |= or_with->bitmap[i];
967 if (ANYOF_CLASS_TEST_ANY_SET(or_with)) {
968 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
969 cl->classflags[i] |= or_with->classflags[i];
970 cl->flags |= ANYOF_CLASS;
973 else { /* XXXX: logic is complicated, leave it along for a moment. */
974 cl_anything(pRExC_state, cl);
977 if (ANYOF_NONBITMAP(or_with)) {
979 /* Use the added node's outside-the-bit-map match if there isn't a
980 * conflict. If there is a conflict (both nodes match something
981 * outside the bitmap, but what they match outside is not the same
982 * pointer, and hence not easily compared until XXX we extend
983 * inversion lists this far), give up and allow the start class to
984 * match everything outside the bitmap. If that stuff is all above
985 * 255, can just set UNICODE_ALL, otherwise caould be anything. */
986 if (! ANYOF_NONBITMAP(cl)) {
987 ARG_SET(cl, ARG(or_with));
989 else if (ARG(cl) != ARG(or_with)) {
991 if ((or_with->flags & ANYOF_NONBITMAP_NON_UTF8)) {
992 cl_anything(pRExC_state, cl);
995 cl->flags |= ANYOF_UNICODE_ALL;
1000 /* Take the union */
1001 cl->flags |= or_with->flags;
1005 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
1006 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
1007 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
1008 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
1013 dump_trie(trie,widecharmap,revcharmap)
1014 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
1015 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
1017 These routines dump out a trie in a somewhat readable format.
1018 The _interim_ variants are used for debugging the interim
1019 tables that are used to generate the final compressed
1020 representation which is what dump_trie expects.
1022 Part of the reason for their existence is to provide a form
1023 of documentation as to how the different representations function.
1028 Dumps the final compressed table form of the trie to Perl_debug_log.
1029 Used for debugging make_trie().
1033 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
1034 AV *revcharmap, U32 depth)
1037 SV *sv=sv_newmortal();
1038 int colwidth= widecharmap ? 6 : 4;
1040 GET_RE_DEBUG_FLAGS_DECL;
1042 PERL_ARGS_ASSERT_DUMP_TRIE;
1044 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
1045 (int)depth * 2 + 2,"",
1046 "Match","Base","Ofs" );
1048 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
1049 SV ** const tmp = av_fetch( revcharmap, state, 0);
1051 PerlIO_printf( Perl_debug_log, "%*s",
1053 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1054 PL_colors[0], PL_colors[1],
1055 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1056 PERL_PV_ESCAPE_FIRSTCHAR
1061 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
1062 (int)depth * 2 + 2,"");
1064 for( state = 0 ; state < trie->uniquecharcount ; state++ )
1065 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
1066 PerlIO_printf( Perl_debug_log, "\n");
1068 for( state = 1 ; state < trie->statecount ; state++ ) {
1069 const U32 base = trie->states[ state ].trans.base;
1071 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
1073 if ( trie->states[ state ].wordnum ) {
1074 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
1076 PerlIO_printf( Perl_debug_log, "%6s", "" );
1079 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
1084 while( ( base + ofs < trie->uniquecharcount ) ||
1085 ( base + ofs - trie->uniquecharcount < trie->lasttrans
1086 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
1089 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
1091 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1092 if ( ( base + ofs >= trie->uniquecharcount ) &&
1093 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1094 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1096 PerlIO_printf( Perl_debug_log, "%*"UVXf,
1098 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
1100 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
1104 PerlIO_printf( Perl_debug_log, "]");
1107 PerlIO_printf( Perl_debug_log, "\n" );
1109 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
1110 for (word=1; word <= trie->wordcount; word++) {
1111 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
1112 (int)word, (int)(trie->wordinfo[word].prev),
1113 (int)(trie->wordinfo[word].len));
1115 PerlIO_printf(Perl_debug_log, "\n" );
1118 Dumps a fully constructed but uncompressed trie in list form.
1119 List tries normally only are used for construction when the number of
1120 possible chars (trie->uniquecharcount) is very high.
1121 Used for debugging make_trie().
1124 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
1125 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1129 SV *sv=sv_newmortal();
1130 int colwidth= widecharmap ? 6 : 4;
1131 GET_RE_DEBUG_FLAGS_DECL;
1133 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1135 /* print out the table precompression. */
1136 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1137 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1138 "------:-----+-----------------\n" );
1140 for( state=1 ; state < next_alloc ; state ++ ) {
1143 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1144 (int)depth * 2 + 2,"", (UV)state );
1145 if ( ! trie->states[ state ].wordnum ) {
1146 PerlIO_printf( Perl_debug_log, "%5s| ","");
1148 PerlIO_printf( Perl_debug_log, "W%4x| ",
1149 trie->states[ state ].wordnum
1152 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1153 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
1155 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1157 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1158 PL_colors[0], PL_colors[1],
1159 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1160 PERL_PV_ESCAPE_FIRSTCHAR
1162 TRIE_LIST_ITEM(state,charid).forid,
1163 (UV)TRIE_LIST_ITEM(state,charid).newstate
1166 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1167 (int)((depth * 2) + 14), "");
1170 PerlIO_printf( Perl_debug_log, "\n");
1175 Dumps a fully constructed but uncompressed trie in table form.
1176 This is the normal DFA style state transition table, with a few
1177 twists to facilitate compression later.
1178 Used for debugging make_trie().
1181 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1182 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1187 SV *sv=sv_newmortal();
1188 int colwidth= widecharmap ? 6 : 4;
1189 GET_RE_DEBUG_FLAGS_DECL;
1191 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1194 print out the table precompression so that we can do a visual check
1195 that they are identical.
1198 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1200 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1201 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1203 PerlIO_printf( Perl_debug_log, "%*s",
1205 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1206 PL_colors[0], PL_colors[1],
1207 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1208 PERL_PV_ESCAPE_FIRSTCHAR
1214 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1216 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1217 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1220 PerlIO_printf( Perl_debug_log, "\n" );
1222 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1224 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1225 (int)depth * 2 + 2,"",
1226 (UV)TRIE_NODENUM( state ) );
1228 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1229 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1231 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1233 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1235 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1236 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1238 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1239 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1247 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1248 startbranch: the first branch in the whole branch sequence
1249 first : start branch of sequence of branch-exact nodes.
1250 May be the same as startbranch
1251 last : Thing following the last branch.
1252 May be the same as tail.
1253 tail : item following the branch sequence
1254 count : words in the sequence
1255 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1256 depth : indent depth
1258 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1260 A trie is an N'ary tree where the branches are determined by digital
1261 decomposition of the key. IE, at the root node you look up the 1st character and
1262 follow that branch repeat until you find the end of the branches. Nodes can be
1263 marked as "accepting" meaning they represent a complete word. Eg:
1267 would convert into the following structure. Numbers represent states, letters
1268 following numbers represent valid transitions on the letter from that state, if
1269 the number is in square brackets it represents an accepting state, otherwise it
1270 will be in parenthesis.
1272 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1276 (1) +-i->(6)-+-s->[7]
1278 +-s->(3)-+-h->(4)-+-e->[5]
1280 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1282 This shows that when matching against the string 'hers' we will begin at state 1
1283 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1284 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1285 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1286 single traverse. We store a mapping from accepting to state to which word was
1287 matched, and then when we have multiple possibilities we try to complete the
1288 rest of the regex in the order in which they occured in the alternation.
1290 The only prior NFA like behaviour that would be changed by the TRIE support is
1291 the silent ignoring of duplicate alternations which are of the form:
1293 / (DUPE|DUPE) X? (?{ ... }) Y /x
1295 Thus EVAL blocks following a trie may be called a different number of times with
1296 and without the optimisation. With the optimisations dupes will be silently
1297 ignored. This inconsistent behaviour of EVAL type nodes is well established as
1298 the following demonstrates:
1300 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1302 which prints out 'word' three times, but
1304 'words'=~/(word|word|word)(?{ print $1 })S/
1306 which doesnt print it out at all. This is due to other optimisations kicking in.
1308 Example of what happens on a structural level:
1310 The regexp /(ac|ad|ab)+/ will produce the following debug output:
1312 1: CURLYM[1] {1,32767}(18)
1323 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1324 and should turn into:
1326 1: CURLYM[1] {1,32767}(18)
1328 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1336 Cases where tail != last would be like /(?foo|bar)baz/:
1346 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1347 and would end up looking like:
1350 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1357 d = uvuni_to_utf8_flags(d, uv, 0);
1359 is the recommended Unicode-aware way of saying
1364 #define TRIE_STORE_REVCHAR \
1367 SV *zlopp = newSV(2); \
1368 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1369 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1370 SvCUR_set(zlopp, kapow - flrbbbbb); \
1373 av_push(revcharmap, zlopp); \
1375 char ooooff = (char)uvc; \
1376 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1380 #define TRIE_READ_CHAR STMT_START { \
1384 if ( foldlen > 0 ) { \
1385 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1390 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1391 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1392 foldlen -= UNISKIP( uvc ); \
1393 scan = foldbuf + UNISKIP( uvc ); \
1396 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1406 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1407 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1408 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1409 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1411 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1412 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1413 TRIE_LIST_CUR( state )++; \
1416 #define TRIE_LIST_NEW(state) STMT_START { \
1417 Newxz( trie->states[ state ].trans.list, \
1418 4, reg_trie_trans_le ); \
1419 TRIE_LIST_CUR( state ) = 1; \
1420 TRIE_LIST_LEN( state ) = 4; \
1423 #define TRIE_HANDLE_WORD(state) STMT_START { \
1424 U16 dupe= trie->states[ state ].wordnum; \
1425 regnode * const noper_next = regnext( noper ); \
1428 /* store the word for dumping */ \
1430 if (OP(noper) != NOTHING) \
1431 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1433 tmp = newSVpvn_utf8( "", 0, UTF ); \
1434 av_push( trie_words, tmp ); \
1438 trie->wordinfo[curword].prev = 0; \
1439 trie->wordinfo[curword].len = wordlen; \
1440 trie->wordinfo[curword].accept = state; \
1442 if ( noper_next < tail ) { \
1444 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1445 trie->jump[curword] = (U16)(noper_next - convert); \
1447 jumper = noper_next; \
1449 nextbranch= regnext(cur); \
1453 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1454 /* chain, so that when the bits of chain are later */\
1455 /* linked together, the dups appear in the chain */\
1456 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1457 trie->wordinfo[dupe].prev = curword; \
1459 /* we haven't inserted this word yet. */ \
1460 trie->states[ state ].wordnum = curword; \
1465 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1466 ( ( base + charid >= ucharcount \
1467 && base + charid < ubound \
1468 && state == trie->trans[ base - ucharcount + charid ].check \
1469 && trie->trans[ base - ucharcount + charid ].next ) \
1470 ? trie->trans[ base - ucharcount + charid ].next \
1471 : ( state==1 ? special : 0 ) \
1475 #define MADE_JUMP_TRIE 2
1476 #define MADE_EXACT_TRIE 4
1479 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1482 /* first pass, loop through and scan words */
1483 reg_trie_data *trie;
1484 HV *widecharmap = NULL;
1485 AV *revcharmap = newAV();
1487 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1492 regnode *jumper = NULL;
1493 regnode *nextbranch = NULL;
1494 regnode *convert = NULL;
1495 U32 *prev_states; /* temp array mapping each state to previous one */
1496 /* we just use folder as a flag in utf8 */
1497 const U8 * folder = NULL;
1500 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1501 AV *trie_words = NULL;
1502 /* along with revcharmap, this only used during construction but both are
1503 * useful during debugging so we store them in the struct when debugging.
1506 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1507 STRLEN trie_charcount=0;
1509 SV *re_trie_maxbuff;
1510 GET_RE_DEBUG_FLAGS_DECL;
1512 PERL_ARGS_ASSERT_MAKE_TRIE;
1514 PERL_UNUSED_ARG(depth);
1519 case EXACTFU: folder = PL_fold_latin1; break;
1520 case EXACTF: folder = PL_fold; break;
1521 case EXACTFL: folder = PL_fold_locale; break;
1524 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1526 trie->startstate = 1;
1527 trie->wordcount = word_count;
1528 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1529 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1530 if (!(UTF && folder))
1531 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1532 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1533 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1536 trie_words = newAV();
1539 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1540 if (!SvIOK(re_trie_maxbuff)) {
1541 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1544 PerlIO_printf( Perl_debug_log,
1545 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1546 (int)depth * 2 + 2, "",
1547 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1548 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1552 /* Find the node we are going to overwrite */
1553 if ( first == startbranch && OP( last ) != BRANCH ) {
1554 /* whole branch chain */
1557 /* branch sub-chain */
1558 convert = NEXTOPER( first );
1561 /* -- First loop and Setup --
1563 We first traverse the branches and scan each word to determine if it
1564 contains widechars, and how many unique chars there are, this is
1565 important as we have to build a table with at least as many columns as we
1568 We use an array of integers to represent the character codes 0..255
1569 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1570 native representation of the character value as the key and IV's for the
1573 *TODO* If we keep track of how many times each character is used we can
1574 remap the columns so that the table compression later on is more
1575 efficient in terms of memory by ensuring the most common value is in the
1576 middle and the least common are on the outside. IMO this would be better
1577 than a most to least common mapping as theres a decent chance the most
1578 common letter will share a node with the least common, meaning the node
1579 will not be compressible. With a middle is most common approach the worst
1580 case is when we have the least common nodes twice.
1584 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1585 regnode * const noper = NEXTOPER( cur );
1586 const U8 *uc = (U8*)STRING( noper );
1587 const U8 * const e = uc + STR_LEN( noper );
1589 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1590 const U8 *scan = (U8*)NULL;
1591 U32 wordlen = 0; /* required init */
1593 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1595 if (OP(noper) == NOTHING) {
1599 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1600 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1601 regardless of encoding */
1603 for ( ; uc < e ; uc += len ) {
1604 TRIE_CHARCOUNT(trie)++;
1608 if ( !trie->charmap[ uvc ] ) {
1609 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1611 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1615 /* store the codepoint in the bitmap, and its folded
1617 TRIE_BITMAP_SET(trie,uvc);
1619 /* store the folded codepoint */
1620 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1623 /* store first byte of utf8 representation of
1624 variant codepoints */
1625 if (! UNI_IS_INVARIANT(uvc)) {
1626 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
1629 set_bit = 0; /* We've done our bit :-) */
1634 widecharmap = newHV();
1636 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1639 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1641 if ( !SvTRUE( *svpp ) ) {
1642 sv_setiv( *svpp, ++trie->uniquecharcount );
1647 if( cur == first ) {
1650 } else if (chars < trie->minlen) {
1652 } else if (chars > trie->maxlen) {
1656 } /* end first pass */
1657 DEBUG_TRIE_COMPILE_r(
1658 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1659 (int)depth * 2 + 2,"",
1660 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1661 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1662 (int)trie->minlen, (int)trie->maxlen )
1666 We now know what we are dealing with in terms of unique chars and
1667 string sizes so we can calculate how much memory a naive
1668 representation using a flat table will take. If it's over a reasonable
1669 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1670 conservative but potentially much slower representation using an array
1673 At the end we convert both representations into the same compressed
1674 form that will be used in regexec.c for matching with. The latter
1675 is a form that cannot be used to construct with but has memory
1676 properties similar to the list form and access properties similar
1677 to the table form making it both suitable for fast searches and
1678 small enough that its feasable to store for the duration of a program.
1680 See the comment in the code where the compressed table is produced
1681 inplace from the flat tabe representation for an explanation of how
1682 the compression works.
1687 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1690 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1692 Second Pass -- Array Of Lists Representation
1694 Each state will be represented by a list of charid:state records
1695 (reg_trie_trans_le) the first such element holds the CUR and LEN
1696 points of the allocated array. (See defines above).
1698 We build the initial structure using the lists, and then convert
1699 it into the compressed table form which allows faster lookups
1700 (but cant be modified once converted).
1703 STRLEN transcount = 1;
1705 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1706 "%*sCompiling trie using list compiler\n",
1707 (int)depth * 2 + 2, ""));
1709 trie->states = (reg_trie_state *)
1710 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1711 sizeof(reg_trie_state) );
1715 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1717 regnode * const noper = NEXTOPER( cur );
1718 U8 *uc = (U8*)STRING( noper );
1719 const U8 * const e = uc + STR_LEN( noper );
1720 U32 state = 1; /* required init */
1721 U16 charid = 0; /* sanity init */
1722 U8 *scan = (U8*)NULL; /* sanity init */
1723 STRLEN foldlen = 0; /* required init */
1724 U32 wordlen = 0; /* required init */
1725 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1727 if (OP(noper) != NOTHING) {
1728 for ( ; uc < e ; uc += len ) {
1733 charid = trie->charmap[ uvc ];
1735 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1739 charid=(U16)SvIV( *svpp );
1742 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1749 if ( !trie->states[ state ].trans.list ) {
1750 TRIE_LIST_NEW( state );
1752 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1753 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1754 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1759 newstate = next_alloc++;
1760 prev_states[newstate] = state;
1761 TRIE_LIST_PUSH( state, charid, newstate );
1766 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1770 TRIE_HANDLE_WORD(state);
1772 } /* end second pass */
1774 /* next alloc is the NEXT state to be allocated */
1775 trie->statecount = next_alloc;
1776 trie->states = (reg_trie_state *)
1777 PerlMemShared_realloc( trie->states,
1779 * sizeof(reg_trie_state) );
1781 /* and now dump it out before we compress it */
1782 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1783 revcharmap, next_alloc,
1787 trie->trans = (reg_trie_trans *)
1788 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1795 for( state=1 ; state < next_alloc ; state ++ ) {
1799 DEBUG_TRIE_COMPILE_MORE_r(
1800 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1804 if (trie->states[state].trans.list) {
1805 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1809 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1810 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1811 if ( forid < minid ) {
1813 } else if ( forid > maxid ) {
1817 if ( transcount < tp + maxid - minid + 1) {
1819 trie->trans = (reg_trie_trans *)
1820 PerlMemShared_realloc( trie->trans,
1822 * sizeof(reg_trie_trans) );
1823 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1825 base = trie->uniquecharcount + tp - minid;
1826 if ( maxid == minid ) {
1828 for ( ; zp < tp ; zp++ ) {
1829 if ( ! trie->trans[ zp ].next ) {
1830 base = trie->uniquecharcount + zp - minid;
1831 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1832 trie->trans[ zp ].check = state;
1838 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1839 trie->trans[ tp ].check = state;
1844 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1845 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1846 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1847 trie->trans[ tid ].check = state;
1849 tp += ( maxid - minid + 1 );
1851 Safefree(trie->states[ state ].trans.list);
1854 DEBUG_TRIE_COMPILE_MORE_r(
1855 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1858 trie->states[ state ].trans.base=base;
1860 trie->lasttrans = tp + 1;
1864 Second Pass -- Flat Table Representation.
1866 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1867 We know that we will need Charcount+1 trans at most to store the data
1868 (one row per char at worst case) So we preallocate both structures
1869 assuming worst case.
1871 We then construct the trie using only the .next slots of the entry
1874 We use the .check field of the first entry of the node temporarily to
1875 make compression both faster and easier by keeping track of how many non
1876 zero fields are in the node.
1878 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1881 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1882 number representing the first entry of the node, and state as a
1883 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1884 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1885 are 2 entrys per node. eg:
1893 The table is internally in the right hand, idx form. However as we also
1894 have to deal with the states array which is indexed by nodenum we have to
1895 use TRIE_NODENUM() to convert.
1898 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1899 "%*sCompiling trie using table compiler\n",
1900 (int)depth * 2 + 2, ""));
1902 trie->trans = (reg_trie_trans *)
1903 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1904 * trie->uniquecharcount + 1,
1905 sizeof(reg_trie_trans) );
1906 trie->states = (reg_trie_state *)
1907 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1908 sizeof(reg_trie_state) );
1909 next_alloc = trie->uniquecharcount + 1;
1912 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1914 regnode * const noper = NEXTOPER( cur );
1915 const U8 *uc = (U8*)STRING( noper );
1916 const U8 * const e = uc + STR_LEN( noper );
1918 U32 state = 1; /* required init */
1920 U16 charid = 0; /* sanity init */
1921 U32 accept_state = 0; /* sanity init */
1922 U8 *scan = (U8*)NULL; /* sanity init */
1924 STRLEN foldlen = 0; /* required init */
1925 U32 wordlen = 0; /* required init */
1926 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1928 if ( OP(noper) != NOTHING ) {
1929 for ( ; uc < e ; uc += len ) {
1934 charid = trie->charmap[ uvc ];
1936 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1937 charid = svpp ? (U16)SvIV(*svpp) : 0;
1941 if ( !trie->trans[ state + charid ].next ) {
1942 trie->trans[ state + charid ].next = next_alloc;
1943 trie->trans[ state ].check++;
1944 prev_states[TRIE_NODENUM(next_alloc)]
1945 = TRIE_NODENUM(state);
1946 next_alloc += trie->uniquecharcount;
1948 state = trie->trans[ state + charid ].next;
1950 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1952 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1955 accept_state = TRIE_NODENUM( state );
1956 TRIE_HANDLE_WORD(accept_state);
1958 } /* end second pass */
1960 /* and now dump it out before we compress it */
1961 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1963 next_alloc, depth+1));
1967 * Inplace compress the table.*
1969 For sparse data sets the table constructed by the trie algorithm will
1970 be mostly 0/FAIL transitions or to put it another way mostly empty.
1971 (Note that leaf nodes will not contain any transitions.)
1973 This algorithm compresses the tables by eliminating most such
1974 transitions, at the cost of a modest bit of extra work during lookup:
1976 - Each states[] entry contains a .base field which indicates the
1977 index in the state[] array wheres its transition data is stored.
1979 - If .base is 0 there are no valid transitions from that node.
1981 - If .base is nonzero then charid is added to it to find an entry in
1984 -If trans[states[state].base+charid].check!=state then the
1985 transition is taken to be a 0/Fail transition. Thus if there are fail
1986 transitions at the front of the node then the .base offset will point
1987 somewhere inside the previous nodes data (or maybe even into a node
1988 even earlier), but the .check field determines if the transition is
1992 The following process inplace converts the table to the compressed
1993 table: We first do not compress the root node 1,and mark all its
1994 .check pointers as 1 and set its .base pointer as 1 as well. This
1995 allows us to do a DFA construction from the compressed table later,
1996 and ensures that any .base pointers we calculate later are greater
1999 - We set 'pos' to indicate the first entry of the second node.
2001 - We then iterate over the columns of the node, finding the first and
2002 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
2003 and set the .check pointers accordingly, and advance pos
2004 appropriately and repreat for the next node. Note that when we copy
2005 the next pointers we have to convert them from the original
2006 NODEIDX form to NODENUM form as the former is not valid post
2009 - If a node has no transitions used we mark its base as 0 and do not
2010 advance the pos pointer.
2012 - If a node only has one transition we use a second pointer into the
2013 structure to fill in allocated fail transitions from other states.
2014 This pointer is independent of the main pointer and scans forward
2015 looking for null transitions that are allocated to a state. When it
2016 finds one it writes the single transition into the "hole". If the
2017 pointer doesnt find one the single transition is appended as normal.
2019 - Once compressed we can Renew/realloc the structures to release the
2022 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
2023 specifically Fig 3.47 and the associated pseudocode.
2027 const U32 laststate = TRIE_NODENUM( next_alloc );
2030 trie->statecount = laststate;
2032 for ( state = 1 ; state < laststate ; state++ ) {
2034 const U32 stateidx = TRIE_NODEIDX( state );
2035 const U32 o_used = trie->trans[ stateidx ].check;
2036 U32 used = trie->trans[ stateidx ].check;
2037 trie->trans[ stateidx ].check = 0;
2039 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
2040 if ( flag || trie->trans[ stateidx + charid ].next ) {
2041 if ( trie->trans[ stateidx + charid ].next ) {
2043 for ( ; zp < pos ; zp++ ) {
2044 if ( ! trie->trans[ zp ].next ) {
2048 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
2049 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
2050 trie->trans[ zp ].check = state;
2051 if ( ++zp > pos ) pos = zp;
2058 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
2060 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
2061 trie->trans[ pos ].check = state;
2066 trie->lasttrans = pos + 1;
2067 trie->states = (reg_trie_state *)
2068 PerlMemShared_realloc( trie->states, laststate
2069 * sizeof(reg_trie_state) );
2070 DEBUG_TRIE_COMPILE_MORE_r(
2071 PerlIO_printf( Perl_debug_log,
2072 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
2073 (int)depth * 2 + 2,"",
2074 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
2077 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
2080 } /* end table compress */
2082 DEBUG_TRIE_COMPILE_MORE_r(
2083 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
2084 (int)depth * 2 + 2, "",
2085 (UV)trie->statecount,
2086 (UV)trie->lasttrans)
2088 /* resize the trans array to remove unused space */
2089 trie->trans = (reg_trie_trans *)
2090 PerlMemShared_realloc( trie->trans, trie->lasttrans
2091 * sizeof(reg_trie_trans) );
2093 { /* Modify the program and insert the new TRIE node */
2094 U8 nodetype =(U8)(flags & 0xFF);
2098 regnode *optimize = NULL;
2099 #ifdef RE_TRACK_PATTERN_OFFSETS
2102 U32 mjd_nodelen = 0;
2103 #endif /* RE_TRACK_PATTERN_OFFSETS */
2104 #endif /* DEBUGGING */
2106 This means we convert either the first branch or the first Exact,
2107 depending on whether the thing following (in 'last') is a branch
2108 or not and whther first is the startbranch (ie is it a sub part of
2109 the alternation or is it the whole thing.)
2110 Assuming its a sub part we convert the EXACT otherwise we convert
2111 the whole branch sequence, including the first.
2113 /* Find the node we are going to overwrite */
2114 if ( first != startbranch || OP( last ) == BRANCH ) {
2115 /* branch sub-chain */
2116 NEXT_OFF( first ) = (U16)(last - first);
2117 #ifdef RE_TRACK_PATTERN_OFFSETS
2119 mjd_offset= Node_Offset((convert));
2120 mjd_nodelen= Node_Length((convert));
2123 /* whole branch chain */
2125 #ifdef RE_TRACK_PATTERN_OFFSETS
2128 const regnode *nop = NEXTOPER( convert );
2129 mjd_offset= Node_Offset((nop));
2130 mjd_nodelen= Node_Length((nop));
2134 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2135 (int)depth * 2 + 2, "",
2136 (UV)mjd_offset, (UV)mjd_nodelen)
2139 /* But first we check to see if there is a common prefix we can
2140 split out as an EXACT and put in front of the TRIE node. */
2141 trie->startstate= 1;
2142 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2144 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2148 const U32 base = trie->states[ state ].trans.base;
2150 if ( trie->states[state].wordnum )
2153 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2154 if ( ( base + ofs >= trie->uniquecharcount ) &&
2155 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2156 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2158 if ( ++count > 1 ) {
2159 SV **tmp = av_fetch( revcharmap, ofs, 0);
2160 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2161 if ( state == 1 ) break;
2163 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2165 PerlIO_printf(Perl_debug_log,
2166 "%*sNew Start State=%"UVuf" Class: [",
2167 (int)depth * 2 + 2, "",
2170 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2171 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2173 TRIE_BITMAP_SET(trie,*ch);
2175 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2177 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2181 TRIE_BITMAP_SET(trie,*ch);
2183 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2184 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2190 SV **tmp = av_fetch( revcharmap, idx, 0);
2192 char *ch = SvPV( *tmp, len );
2194 SV *sv=sv_newmortal();
2195 PerlIO_printf( Perl_debug_log,
2196 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2197 (int)depth * 2 + 2, "",
2199 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2200 PL_colors[0], PL_colors[1],
2201 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2202 PERL_PV_ESCAPE_FIRSTCHAR
2207 OP( convert ) = nodetype;
2208 str=STRING(convert);
2211 STR_LEN(convert) += len;
2217 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2222 trie->prefixlen = (state-1);
2224 regnode *n = convert+NODE_SZ_STR(convert);
2225 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2226 trie->startstate = state;
2227 trie->minlen -= (state - 1);
2228 trie->maxlen -= (state - 1);
2230 /* At least the UNICOS C compiler choked on this
2231 * being argument to DEBUG_r(), so let's just have
2234 #ifdef PERL_EXT_RE_BUILD
2240 regnode *fix = convert;
2241 U32 word = trie->wordcount;
2243 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2244 while( ++fix < n ) {
2245 Set_Node_Offset_Length(fix, 0, 0);
2248 SV ** const tmp = av_fetch( trie_words, word, 0 );
2250 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2251 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2253 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2261 NEXT_OFF(convert) = (U16)(tail - convert);
2262 DEBUG_r(optimize= n);
2268 if ( trie->maxlen ) {
2269 NEXT_OFF( convert ) = (U16)(tail - convert);
2270 ARG_SET( convert, data_slot );
2271 /* Store the offset to the first unabsorbed branch in
2272 jump[0], which is otherwise unused by the jump logic.
2273 We use this when dumping a trie and during optimisation. */
2275 trie->jump[0] = (U16)(nextbranch - convert);
2277 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2278 * and there is a bitmap
2279 * and the first "jump target" node we found leaves enough room
2280 * then convert the TRIE node into a TRIEC node, with the bitmap
2281 * embedded inline in the opcode - this is hypothetically faster.
2283 if ( !trie->states[trie->startstate].wordnum
2285 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2287 OP( convert ) = TRIEC;
2288 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2289 PerlMemShared_free(trie->bitmap);
2292 OP( convert ) = TRIE;
2294 /* store the type in the flags */
2295 convert->flags = nodetype;
2299 + regarglen[ OP( convert ) ];
2301 /* XXX We really should free up the resource in trie now,
2302 as we won't use them - (which resources?) dmq */
2304 /* needed for dumping*/
2305 DEBUG_r(if (optimize) {
2306 regnode *opt = convert;
2308 while ( ++opt < optimize) {
2309 Set_Node_Offset_Length(opt,0,0);
2312 Try to clean up some of the debris left after the
2315 while( optimize < jumper ) {
2316 mjd_nodelen += Node_Length((optimize));
2317 OP( optimize ) = OPTIMIZED;
2318 Set_Node_Offset_Length(optimize,0,0);
2321 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2323 } /* end node insert */
2325 /* Finish populating the prev field of the wordinfo array. Walk back
2326 * from each accept state until we find another accept state, and if
2327 * so, point the first word's .prev field at the second word. If the
2328 * second already has a .prev field set, stop now. This will be the
2329 * case either if we've already processed that word's accept state,
2330 * or that state had multiple words, and the overspill words were
2331 * already linked up earlier.
2338 for (word=1; word <= trie->wordcount; word++) {
2340 if (trie->wordinfo[word].prev)
2342 state = trie->wordinfo[word].accept;
2344 state = prev_states[state];
2347 prev = trie->states[state].wordnum;
2351 trie->wordinfo[word].prev = prev;
2353 Safefree(prev_states);
2357 /* and now dump out the compressed format */
2358 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2360 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2362 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2363 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2365 SvREFCNT_dec(revcharmap);
2369 : trie->startstate>1
2375 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2377 /* The Trie is constructed and compressed now so we can build a fail array if it's needed
2379 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2380 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2383 We find the fail state for each state in the trie, this state is the longest proper
2384 suffix of the current state's 'word' that is also a proper prefix of another word in our
2385 trie. State 1 represents the word '' and is thus the default fail state. This allows
2386 the DFA not to have to restart after its tried and failed a word at a given point, it
2387 simply continues as though it had been matching the other word in the first place.
2389 'abcdgu'=~/abcdefg|cdgu/
2390 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2391 fail, which would bring us to the state representing 'd' in the second word where we would
2392 try 'g' and succeed, proceeding to match 'cdgu'.
2394 /* add a fail transition */
2395 const U32 trie_offset = ARG(source);
2396 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2398 const U32 ucharcount = trie->uniquecharcount;
2399 const U32 numstates = trie->statecount;
2400 const U32 ubound = trie->lasttrans + ucharcount;
2404 U32 base = trie->states[ 1 ].trans.base;
2407 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2408 GET_RE_DEBUG_FLAGS_DECL;
2410 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2412 PERL_UNUSED_ARG(depth);
2416 ARG_SET( stclass, data_slot );
2417 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2418 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2419 aho->trie=trie_offset;
2420 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2421 Copy( trie->states, aho->states, numstates, reg_trie_state );
2422 Newxz( q, numstates, U32);
2423 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2426 /* initialize fail[0..1] to be 1 so that we always have
2427 a valid final fail state */
2428 fail[ 0 ] = fail[ 1 ] = 1;
2430 for ( charid = 0; charid < ucharcount ; charid++ ) {
2431 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2433 q[ q_write ] = newstate;
2434 /* set to point at the root */
2435 fail[ q[ q_write++ ] ]=1;
2438 while ( q_read < q_write) {
2439 const U32 cur = q[ q_read++ % numstates ];
2440 base = trie->states[ cur ].trans.base;
2442 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2443 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2445 U32 fail_state = cur;
2448 fail_state = fail[ fail_state ];
2449 fail_base = aho->states[ fail_state ].trans.base;
2450 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2452 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2453 fail[ ch_state ] = fail_state;
2454 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2456 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2458 q[ q_write++ % numstates] = ch_state;
2462 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2463 when we fail in state 1, this allows us to use the
2464 charclass scan to find a valid start char. This is based on the principle
2465 that theres a good chance the string being searched contains lots of stuff
2466 that cant be a start char.
2468 fail[ 0 ] = fail[ 1 ] = 0;
2469 DEBUG_TRIE_COMPILE_r({
2470 PerlIO_printf(Perl_debug_log,
2471 "%*sStclass Failtable (%"UVuf" states): 0",
2472 (int)(depth * 2), "", (UV)numstates
2474 for( q_read=1; q_read<numstates; q_read++ ) {
2475 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2477 PerlIO_printf(Perl_debug_log, "\n");
2480 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2485 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2486 * These need to be revisited when a newer toolchain becomes available.
2488 #if defined(__sparc64__) && defined(__GNUC__)
2489 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2490 # undef SPARC64_GCC_WORKAROUND
2491 # define SPARC64_GCC_WORKAROUND 1
2495 #define DEBUG_PEEP(str,scan,depth) \
2496 DEBUG_OPTIMISE_r({if (scan){ \
2497 SV * const mysv=sv_newmortal(); \
2498 regnode *Next = regnext(scan); \
2499 regprop(RExC_rx, mysv, scan); \
2500 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2501 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2502 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2509 #define JOIN_EXACT(scan,min,flags) \
2510 if (PL_regkind[OP(scan)] == EXACT) \
2511 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2514 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2515 /* Merge several consecutive EXACTish nodes into one. */
2516 regnode *n = regnext(scan);
2518 regnode *next = scan + NODE_SZ_STR(scan);
2522 regnode *stop = scan;
2523 GET_RE_DEBUG_FLAGS_DECL;
2525 PERL_UNUSED_ARG(depth);
2528 PERL_ARGS_ASSERT_JOIN_EXACT;
2529 #ifndef EXPERIMENTAL_INPLACESCAN
2530 PERL_UNUSED_ARG(flags);
2531 PERL_UNUSED_ARG(val);
2533 DEBUG_PEEP("join",scan,depth);
2535 /* Skip NOTHING, merge EXACT*. */
2537 ( PL_regkind[OP(n)] == NOTHING ||
2538 (stringok && (OP(n) == OP(scan))))
2540 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2542 if (OP(n) == TAIL || n > next)
2544 if (PL_regkind[OP(n)] == NOTHING) {
2545 DEBUG_PEEP("skip:",n,depth);
2546 NEXT_OFF(scan) += NEXT_OFF(n);
2547 next = n + NODE_STEP_REGNODE;
2554 else if (stringok) {
2555 const unsigned int oldl = STR_LEN(scan);
2556 regnode * const nnext = regnext(n);
2558 DEBUG_PEEP("merg",n,depth);
2561 if (oldl + STR_LEN(n) > U8_MAX)
2563 NEXT_OFF(scan) += NEXT_OFF(n);
2564 STR_LEN(scan) += STR_LEN(n);
2565 next = n + NODE_SZ_STR(n);
2566 /* Now we can overwrite *n : */
2567 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2575 #ifdef EXPERIMENTAL_INPLACESCAN
2576 if (flags && !NEXT_OFF(n)) {
2577 DEBUG_PEEP("atch", val, depth);
2578 if (reg_off_by_arg[OP(n)]) {
2579 ARG_SET(n, val - n);
2582 NEXT_OFF(n) = val - n;
2588 #define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390
2589 #define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS
2590 #define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0
2591 #define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS
2594 && ( OP(scan) == EXACTF || OP(scan) == EXACTFU || OP(scan) == EXACTFA)
2595 && ( STR_LEN(scan) >= 6 ) )
2598 Two problematic code points in Unicode casefolding of EXACT nodes:
2600 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2601 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2607 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2608 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2610 This means that in case-insensitive matching (or "loose matching",
2611 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2612 length of the above casefolded versions) can match a target string
2613 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2614 This would rather mess up the minimum length computation.
2616 What we'll do is to look for the tail four bytes, and then peek
2617 at the preceding two bytes to see whether we need to decrease
2618 the minimum length by four (six minus two).
2620 Thanks to the design of UTF-8, there cannot be false matches:
2621 A sequence of valid UTF-8 bytes cannot be a subsequence of
2622 another valid sequence of UTF-8 bytes.
2625 char * const s0 = STRING(scan), *s, *t;
2626 char * const s1 = s0 + STR_LEN(scan) - 1;
2627 char * const s2 = s1 - 4;
2628 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2629 const char t0[] = "\xaf\x49\xaf\x42";
2631 const char t0[] = "\xcc\x88\xcc\x81";
2633 const char * const t1 = t0 + 3;
2636 s < s2 && (t = ninstr(s, s1, t0, t1));
2639 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2640 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2642 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2643 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2650 /* Allow dumping but overwriting the collection of skipped
2651 * ops and/or strings with fake optimized ops */
2652 n = scan + NODE_SZ_STR(scan);
2660 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2664 /* REx optimizer. Converts nodes into quicker variants "in place".
2665 Finds fixed substrings. */
2667 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2668 to the position after last scanned or to NULL. */
2670 #define INIT_AND_WITHP \
2671 assert(!and_withp); \
2672 Newx(and_withp,1,struct regnode_charclass_class); \
2673 SAVEFREEPV(and_withp)
2675 /* this is a chain of data about sub patterns we are processing that
2676 need to be handled separately/specially in study_chunk. Its so
2677 we can simulate recursion without losing state. */
2679 typedef struct scan_frame {
2680 regnode *last; /* last node to process in this frame */
2681 regnode *next; /* next node to process when last is reached */
2682 struct scan_frame *prev; /*previous frame*/
2683 I32 stop; /* what stopparen do we use */
2687 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2689 #define CASE_SYNST_FNC(nAmE) \
2691 if (flags & SCF_DO_STCLASS_AND) { \
2692 for (value = 0; value < 256; value++) \
2693 if (!is_ ## nAmE ## _cp(value)) \
2694 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2697 for (value = 0; value < 256; value++) \
2698 if (is_ ## nAmE ## _cp(value)) \
2699 ANYOF_BITMAP_SET(data->start_class, value); \
2703 if (flags & SCF_DO_STCLASS_AND) { \
2704 for (value = 0; value < 256; value++) \
2705 if (is_ ## nAmE ## _cp(value)) \
2706 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2709 for (value = 0; value < 256; value++) \
2710 if (!is_ ## nAmE ## _cp(value)) \
2711 ANYOF_BITMAP_SET(data->start_class, value); \
2718 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2719 I32 *minlenp, I32 *deltap,
2724 struct regnode_charclass_class *and_withp,
2725 U32 flags, U32 depth)
2726 /* scanp: Start here (read-write). */
2727 /* deltap: Write maxlen-minlen here. */
2728 /* last: Stop before this one. */
2729 /* data: string data about the pattern */
2730 /* stopparen: treat close N as END */
2731 /* recursed: which subroutines have we recursed into */
2732 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2735 I32 min = 0, pars = 0, code;
2736 regnode *scan = *scanp, *next;
2738 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2739 int is_inf_internal = 0; /* The studied chunk is infinite */
2740 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2741 scan_data_t data_fake;
2742 SV *re_trie_maxbuff = NULL;
2743 regnode *first_non_open = scan;
2744 I32 stopmin = I32_MAX;
2745 scan_frame *frame = NULL;
2746 GET_RE_DEBUG_FLAGS_DECL;
2748 PERL_ARGS_ASSERT_STUDY_CHUNK;
2751 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2755 while (first_non_open && OP(first_non_open) == OPEN)
2756 first_non_open=regnext(first_non_open);
2761 while ( scan && OP(scan) != END && scan < last ){
2762 /* Peephole optimizer: */
2763 DEBUG_STUDYDATA("Peep:", data,depth);
2764 DEBUG_PEEP("Peep",scan,depth);
2765 JOIN_EXACT(scan,&min,0);
2767 /* Follow the next-chain of the current node and optimize
2768 away all the NOTHINGs from it. */
2769 if (OP(scan) != CURLYX) {
2770 const int max = (reg_off_by_arg[OP(scan)]
2772 /* I32 may be smaller than U16 on CRAYs! */
2773 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2774 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2778 /* Skip NOTHING and LONGJMP. */
2779 while ((n = regnext(n))
2780 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2781 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2782 && off + noff < max)
2784 if (reg_off_by_arg[OP(scan)])
2787 NEXT_OFF(scan) = off;
2792 /* The principal pseudo-switch. Cannot be a switch, since we
2793 look into several different things. */
2794 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2795 || OP(scan) == IFTHEN) {
2796 next = regnext(scan);
2798 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2800 if (OP(next) == code || code == IFTHEN) {
2801 /* NOTE - There is similar code to this block below for handling
2802 TRIE nodes on a re-study. If you change stuff here check there
2804 I32 max1 = 0, min1 = I32_MAX, num = 0;
2805 struct regnode_charclass_class accum;
2806 regnode * const startbranch=scan;
2808 if (flags & SCF_DO_SUBSTR)
2809 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2810 if (flags & SCF_DO_STCLASS)
2811 cl_init_zero(pRExC_state, &accum);
2813 while (OP(scan) == code) {
2814 I32 deltanext, minnext, f = 0, fake;
2815 struct regnode_charclass_class this_class;
2818 data_fake.flags = 0;
2820 data_fake.whilem_c = data->whilem_c;
2821 data_fake.last_closep = data->last_closep;
2824 data_fake.last_closep = &fake;
2826 data_fake.pos_delta = delta;
2827 next = regnext(scan);
2828 scan = NEXTOPER(scan);
2830 scan = NEXTOPER(scan);
2831 if (flags & SCF_DO_STCLASS) {
2832 cl_init(pRExC_state, &this_class);
2833 data_fake.start_class = &this_class;
2834 f = SCF_DO_STCLASS_AND;
2836 if (flags & SCF_WHILEM_VISITED_POS)
2837 f |= SCF_WHILEM_VISITED_POS;
2839 /* we suppose the run is continuous, last=next...*/
2840 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2842 stopparen, recursed, NULL, f,depth+1);
2845 if (max1 < minnext + deltanext)
2846 max1 = minnext + deltanext;
2847 if (deltanext == I32_MAX)
2848 is_inf = is_inf_internal = 1;
2850 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2852 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2853 if ( stopmin > minnext)
2854 stopmin = min + min1;
2855 flags &= ~SCF_DO_SUBSTR;
2857 data->flags |= SCF_SEEN_ACCEPT;
2860 if (data_fake.flags & SF_HAS_EVAL)
2861 data->flags |= SF_HAS_EVAL;
2862 data->whilem_c = data_fake.whilem_c;
2864 if (flags & SCF_DO_STCLASS)
2865 cl_or(pRExC_state, &accum, &this_class);
2867 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2869 if (flags & SCF_DO_SUBSTR) {
2870 data->pos_min += min1;
2871 data->pos_delta += max1 - min1;
2872 if (max1 != min1 || is_inf)
2873 data->longest = &(data->longest_float);
2876 delta += max1 - min1;
2877 if (flags & SCF_DO_STCLASS_OR) {
2878 cl_or(pRExC_state, data->start_class, &accum);
2880 cl_and(data->start_class, and_withp);
2881 flags &= ~SCF_DO_STCLASS;
2884 else if (flags & SCF_DO_STCLASS_AND) {
2886 cl_and(data->start_class, &accum);
2887 flags &= ~SCF_DO_STCLASS;
2890 /* Switch to OR mode: cache the old value of
2891 * data->start_class */
2893 StructCopy(data->start_class, and_withp,
2894 struct regnode_charclass_class);
2895 flags &= ~SCF_DO_STCLASS_AND;
2896 StructCopy(&accum, data->start_class,
2897 struct regnode_charclass_class);
2898 flags |= SCF_DO_STCLASS_OR;
2899 data->start_class->flags |= ANYOF_EOS;
2903 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2906 Assuming this was/is a branch we are dealing with: 'scan' now
2907 points at the item that follows the branch sequence, whatever
2908 it is. We now start at the beginning of the sequence and look
2915 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2917 If we can find such a subsequence we need to turn the first
2918 element into a trie and then add the subsequent branch exact
2919 strings to the trie.
2923 1. patterns where the whole set of branches can be converted.
2925 2. patterns where only a subset can be converted.
2927 In case 1 we can replace the whole set with a single regop
2928 for the trie. In case 2 we need to keep the start and end
2931 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2932 becomes BRANCH TRIE; BRANCH X;
2934 There is an additional case, that being where there is a
2935 common prefix, which gets split out into an EXACT like node
2936 preceding the TRIE node.
2938 If x(1..n)==tail then we can do a simple trie, if not we make
2939 a "jump" trie, such that when we match the appropriate word
2940 we "jump" to the appropriate tail node. Essentially we turn
2941 a nested if into a case structure of sorts.
2946 if (!re_trie_maxbuff) {
2947 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2948 if (!SvIOK(re_trie_maxbuff))
2949 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2951 if ( SvIV(re_trie_maxbuff)>=0 ) {
2953 regnode *first = (regnode *)NULL;
2954 regnode *last = (regnode *)NULL;
2955 regnode *tail = scan;
2960 SV * const mysv = sv_newmortal(); /* for dumping */
2962 /* var tail is used because there may be a TAIL
2963 regop in the way. Ie, the exacts will point to the
2964 thing following the TAIL, but the last branch will
2965 point at the TAIL. So we advance tail. If we
2966 have nested (?:) we may have to move through several
2970 while ( OP( tail ) == TAIL ) {
2971 /* this is the TAIL generated by (?:) */
2972 tail = regnext( tail );
2977 regprop(RExC_rx, mysv, tail );
2978 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2979 (int)depth * 2 + 2, "",
2980 "Looking for TRIE'able sequences. Tail node is: ",
2981 SvPV_nolen_const( mysv )
2987 step through the branches, cur represents each
2988 branch, noper is the first thing to be matched
2989 as part of that branch and noper_next is the
2990 regnext() of that node. if noper is an EXACT
2991 and noper_next is the same as scan (our current
2992 position in the regex) then the EXACT branch is
2993 a possible optimization target. Once we have
2994 two or more consecutive such branches we can
2995 create a trie of the EXACT's contents and stich
2996 it in place. If the sequence represents all of
2997 the branches we eliminate the whole thing and
2998 replace it with a single TRIE. If it is a
2999 subsequence then we need to stitch it in. This
3000 means the first branch has to remain, and needs
3001 to be repointed at the item on the branch chain
3002 following the last branch optimized. This could
3003 be either a BRANCH, in which case the
3004 subsequence is internal, or it could be the
3005 item following the branch sequence in which
3006 case the subsequence is at the end.
3010 /* dont use tail as the end marker for this traverse */
3011 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
3012 regnode * const noper = NEXTOPER( cur );
3013 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
3014 regnode * const noper_next = regnext( noper );
3018 regprop(RExC_rx, mysv, cur);
3019 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
3020 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
3022 regprop(RExC_rx, mysv, noper);
3023 PerlIO_printf( Perl_debug_log, " -> %s",
3024 SvPV_nolen_const(mysv));
3027 regprop(RExC_rx, mysv, noper_next );
3028 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
3029 SvPV_nolen_const(mysv));
3031 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
3032 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
3034 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
3035 : PL_regkind[ OP( noper ) ] == EXACT )
3036 || OP(noper) == NOTHING )
3038 && noper_next == tail
3043 if ( !first || optype == NOTHING ) {
3044 if (!first) first = cur;
3045 optype = OP( noper );
3051 Currently the trie logic handles case insensitive matching properly only
3052 when the pattern is UTF-8 and the node is EXACTFU (thus forcing unicode
3055 If/when this is fixed the following define can be swapped
3056 in below to fully enable trie logic.
3058 #define TRIE_TYPE_IS_SAFE 1
3061 #define TRIE_TYPE_IS_SAFE ((UTF && optype == EXACTFU) || optype==EXACT)
3063 if ( last && TRIE_TYPE_IS_SAFE ) {
3064 make_trie( pRExC_state,
3065 startbranch, first, cur, tail, count,
3068 if ( PL_regkind[ OP( noper ) ] == EXACT
3070 && noper_next == tail
3075 optype = OP( noper );
3085 regprop(RExC_rx, mysv, cur);
3086 PerlIO_printf( Perl_debug_log,
3087 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
3088 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
3092 if ( last && TRIE_TYPE_IS_SAFE ) {
3093 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3094 #ifdef TRIE_STUDY_OPT
3095 if ( ((made == MADE_EXACT_TRIE &&
3096 startbranch == first)
3097 || ( first_non_open == first )) &&
3099 flags |= SCF_TRIE_RESTUDY;
3100 if ( startbranch == first
3103 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
3113 else if ( code == BRANCHJ ) { /* single branch is optimized. */
3114 scan = NEXTOPER(NEXTOPER(scan));
3115 } else /* single branch is optimized. */
3116 scan = NEXTOPER(scan);
3118 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
3119 scan_frame *newframe = NULL;
3124 if (OP(scan) != SUSPEND) {
3125 /* set the pointer */
3126 if (OP(scan) == GOSUB) {
3128 RExC_recurse[ARG2L(scan)] = scan;
3129 start = RExC_open_parens[paren-1];
3130 end = RExC_close_parens[paren-1];
3133 start = RExC_rxi->program + 1;
3137 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
3138 SAVEFREEPV(recursed);
3140 if (!PAREN_TEST(recursed,paren+1)) {
3141 PAREN_SET(recursed,paren+1);
3142 Newx(newframe,1,scan_frame);
3144 if (flags & SCF_DO_SUBSTR) {
3145 SCAN_COMMIT(pRExC_state,data,minlenp);
3146 data->longest = &(data->longest_float);
3148 is_inf = is_inf_internal = 1;
3149 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3150 cl_anything(pRExC_state, data->start_class);
3151 flags &= ~SCF_DO_STCLASS;
3154 Newx(newframe,1,scan_frame);
3157 end = regnext(scan);
3162 SAVEFREEPV(newframe);
3163 newframe->next = regnext(scan);
3164 newframe->last = last;
3165 newframe->stop = stopparen;
3166 newframe->prev = frame;
3176 else if (OP(scan) == EXACT) {
3177 I32 l = STR_LEN(scan);
3180 const U8 * const s = (U8*)STRING(scan);
3181 l = utf8_length(s, s + l);
3182 uc = utf8_to_uvchr(s, NULL);
3184 uc = *((U8*)STRING(scan));
3187 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3188 /* The code below prefers earlier match for fixed
3189 offset, later match for variable offset. */
3190 if (data->last_end == -1) { /* Update the start info. */
3191 data->last_start_min = data->pos_min;
3192 data->last_start_max = is_inf
3193 ? I32_MAX : data->pos_min + data->pos_delta;
3195 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3197 SvUTF8_on(data->last_found);
3199 SV * const sv = data->last_found;
3200 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3201 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3202 if (mg && mg->mg_len >= 0)
3203 mg->mg_len += utf8_length((U8*)STRING(scan),
3204 (U8*)STRING(scan)+STR_LEN(scan));
3206 data->last_end = data->pos_min + l;
3207 data->pos_min += l; /* As in the first entry. */
3208 data->flags &= ~SF_BEFORE_EOL;
3210 if (flags & SCF_DO_STCLASS_AND) {
3211 /* Check whether it is compatible with what we know already! */
3215 /* If compatible, we or it in below. It is compatible if is
3216 * in the bitmp and either 1) its bit or its fold is set, or 2)
3217 * it's for a locale. Even if there isn't unicode semantics
3218 * here, at runtime there may be because of matching against a
3219 * utf8 string, so accept a possible false positive for
3220 * latin1-range folds */
3222 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3223 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3224 && (!(data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD)
3225 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3230 ANYOF_CLASS_ZERO(data->start_class);
3231 ANYOF_BITMAP_ZERO(data->start_class);
3233 ANYOF_BITMAP_SET(data->start_class, uc);
3234 else if (uc >= 0x100) {
3237 /* Some Unicode code points fold to the Latin1 range; as
3238 * XXX temporary code, instead of figuring out if this is
3239 * one, just assume it is and set all the start class bits
3240 * that could be some such above 255 code point's fold
3241 * which will generate fals positives. As the code
3242 * elsewhere that does compute the fold settles down, it
3243 * can be extracted out and re-used here */
3244 for (i = 0; i < 256; i++){
3245 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3246 ANYOF_BITMAP_SET(data->start_class, i);
3250 data->start_class->flags &= ~ANYOF_EOS;
3252 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3254 else if (flags & SCF_DO_STCLASS_OR) {
3255 /* false positive possible if the class is case-folded */
3257 ANYOF_BITMAP_SET(data->start_class, uc);
3259 data->start_class->flags |= ANYOF_UNICODE_ALL;
3260 data->start_class->flags &= ~ANYOF_EOS;
3261 cl_and(data->start_class, and_withp);
3263 flags &= ~SCF_DO_STCLASS;
3265 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3266 I32 l = STR_LEN(scan);
3267 UV uc = *((U8*)STRING(scan));
3269 /* Search for fixed substrings supports EXACT only. */
3270 if (flags & SCF_DO_SUBSTR) {
3272 SCAN_COMMIT(pRExC_state, data, minlenp);
3275 const U8 * const s = (U8 *)STRING(scan);
3276 l = utf8_length(s, s + l);
3277 uc = utf8_to_uvchr(s, NULL);
3280 if (flags & SCF_DO_SUBSTR)
3282 if (flags & SCF_DO_STCLASS_AND) {
3283 /* Check whether it is compatible with what we know already! */
3286 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3287 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3288 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3292 ANYOF_CLASS_ZERO(data->start_class);
3293 ANYOF_BITMAP_ZERO(data->start_class);
3295 ANYOF_BITMAP_SET(data->start_class, uc);
3296 data->start_class->flags &= ~ANYOF_EOS;
3297 data->start_class->flags |= ANYOF_LOC_NONBITMAP_FOLD;
3298 if (OP(scan) == EXACTFL) {
3299 /* XXX This set is probably no longer necessary, and
3300 * probably wrong as LOCALE now is on in the initial
3302 data->start_class->flags |= ANYOF_LOCALE;
3306 /* Also set the other member of the fold pair. In case
3307 * that unicode semantics is called for at runtime, use
3308 * the full latin1 fold. (Can't do this for locale,
3309 * because not known until runtime */
3310 ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]);
3313 else if (uc >= 0x100) {
3315 for (i = 0; i < 256; i++){
3316 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3317 ANYOF_BITMAP_SET(data->start_class, i);
3322 else if (flags & SCF_DO_STCLASS_OR) {
3323 if (data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) {
3324 /* false positive possible if the class is case-folded.
3325 Assume that the locale settings are the same... */
3327 ANYOF_BITMAP_SET(data->start_class, uc);
3328 if (OP(scan) != EXACTFL) {
3330 /* And set the other member of the fold pair, but
3331 * can't do that in locale because not known until
3333 ANYOF_BITMAP_SET(data->start_class,
3334 PL_fold_latin1[uc]);
3337 data->start_class->flags &= ~ANYOF_EOS;
3339 cl_and(data->start_class, and_withp);
3341 flags &= ~SCF_DO_STCLASS;
3343 else if (REGNODE_VARIES(OP(scan))) {
3344 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3345 I32 f = flags, pos_before = 0;
3346 regnode * const oscan = scan;
3347 struct regnode_charclass_class this_class;
3348 struct regnode_charclass_class *oclass = NULL;
3349 I32 next_is_eval = 0;
3351 switch (PL_regkind[OP(scan)]) {
3352 case WHILEM: /* End of (?:...)* . */
3353 scan = NEXTOPER(scan);
3356 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3357 next = NEXTOPER(scan);
3358 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3360 maxcount = REG_INFTY;
3361 next = regnext(scan);
3362 scan = NEXTOPER(scan);
3366 if (flags & SCF_DO_SUBSTR)
3371 if (flags & SCF_DO_STCLASS) {
3373 maxcount = REG_INFTY;
3374 next = regnext(scan);
3375 scan = NEXTOPER(scan);
3378 is_inf = is_inf_internal = 1;
3379 scan = regnext(scan);
3380 if (flags & SCF_DO_SUBSTR) {
3381 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3382 data->longest = &(data->longest_float);
3384 goto optimize_curly_tail;
3386 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3387 && (scan->flags == stopparen))
3392 mincount = ARG1(scan);
3393 maxcount = ARG2(scan);
3395 next = regnext(scan);
3396 if (OP(scan) == CURLYX) {
3397 I32 lp = (data ? *(data->last_closep) : 0);
3398 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3400 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3401 next_is_eval = (OP(scan) == EVAL);
3403 if (flags & SCF_DO_SUBSTR) {
3404 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3405 pos_before = data->pos_min;
3409 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3411 data->flags |= SF_IS_INF;
3413 if (flags & SCF_DO_STCLASS) {
3414 cl_init(pRExC_state, &this_class);
3415 oclass = data->start_class;
3416 data->start_class = &this_class;
3417 f |= SCF_DO_STCLASS_AND;
3418 f &= ~SCF_DO_STCLASS_OR;
3420 /* Exclude from super-linear cache processing any {n,m}
3421 regops for which the combination of input pos and regex
3422 pos is not enough information to determine if a match
3425 For example, in the regex /foo(bar\s*){4,8}baz/ with the
3426 regex pos at the \s*, the prospects for a match depend not
3427 only on the input position but also on how many (bar\s*)
3428 repeats into the {4,8} we are. */
3429 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
3430 f &= ~SCF_WHILEM_VISITED_POS;
3432 /* This will finish on WHILEM, setting scan, or on NULL: */
3433 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3434 last, data, stopparen, recursed, NULL,
3436 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3438 if (flags & SCF_DO_STCLASS)
3439 data->start_class = oclass;
3440 if (mincount == 0 || minnext == 0) {
3441 if (flags & SCF_DO_STCLASS_OR) {
3442 cl_or(pRExC_state, data->start_class, &this_class);
3444 else if (flags & SCF_DO_STCLASS_AND) {
3445 /* Switch to OR mode: cache the old value of
3446 * data->start_class */
3448 StructCopy(data->start_class, and_withp,
3449 struct regnode_charclass_class);
3450 flags &= ~SCF_DO_STCLASS_AND;
3451 StructCopy(&this_class, data->start_class,
3452 struct regnode_charclass_class);
3453 flags |= SCF_DO_STCLASS_OR;
3454 data->start_class->flags |= ANYOF_EOS;
3456 } else { /* Non-zero len */
3457 if (flags & SCF_DO_STCLASS_OR) {
3458 cl_or(pRExC_state, data->start_class, &this_class);
3459 cl_and(data->start_class, and_withp);
3461 else if (flags & SCF_DO_STCLASS_AND)
3462 cl_and(data->start_class, &this_class);
3463 flags &= ~SCF_DO_STCLASS;
3465 if (!scan) /* It was not CURLYX, but CURLY. */
3467 if ( /* ? quantifier ok, except for (?{ ... }) */
3468 (next_is_eval || !(mincount == 0 && maxcount == 1))
3469 && (minnext == 0) && (deltanext == 0)
3470 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3471 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3473 ckWARNreg(RExC_parse,
3474 "Quantifier unexpected on zero-length expression");
3477 min += minnext * mincount;
3478 is_inf_internal |= ((maxcount == REG_INFTY
3479 && (minnext + deltanext) > 0)
3480 || deltanext == I32_MAX);
3481 is_inf |= is_inf_internal;
3482 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3484 /* Try powerful optimization CURLYX => CURLYN. */
3485 if ( OP(oscan) == CURLYX && data
3486 && data->flags & SF_IN_PAR
3487 && !(data->flags & SF_HAS_EVAL)
3488 && !deltanext && minnext == 1 ) {
3489 /* Try to optimize to CURLYN. */
3490 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3491 regnode * const nxt1 = nxt;
3498 if (!REGNODE_SIMPLE(OP(nxt))
3499 && !(PL_regkind[OP(nxt)] == EXACT
3500 && STR_LEN(nxt) == 1))
3506 if (OP(nxt) != CLOSE)
3508 if (RExC_open_parens) {
3509 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3510 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3512 /* Now we know that nxt2 is the only contents: */
3513 oscan->flags = (U8)ARG(nxt);
3515 OP(nxt1) = NOTHING; /* was OPEN. */
3518 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3519 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3520 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
3521 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3522 OP(nxt + 1) = OPTIMIZED; /* was count. */
3523 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
3528 /* Try optimization CURLYX => CURLYM. */
3529 if ( OP(oscan) == CURLYX && data
3530 && !(data->flags & SF_HAS_PAR)
3531 && !(data->flags & SF_HAS_EVAL)
3532 && !deltanext /* atom is fixed width */
3533 && minnext != 0 /* CURLYM can't handle zero width */
3535 /* XXXX How to optimize if data == 0? */
3536 /* Optimize to a simpler form. */
3537 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3541 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3542 && (OP(nxt2) != WHILEM))
3544 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3545 /* Need to optimize away parenths. */
3546 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
3547 /* Set the parenth number. */
3548 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3550 oscan->flags = (U8)ARG(nxt);
3551 if (RExC_open_parens) {
3552 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3553 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3555 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3556 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3559 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3560 OP(nxt + 1) = OPTIMIZED; /* was count. */
3561 NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */
3562 NEXT_OFF(nxt + 1) = 0; /* just for consistency. */
3565 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3566 regnode *nnxt = regnext(nxt1);
3568 if (reg_off_by_arg[OP(nxt1)])
3569 ARG_SET(nxt1, nxt2 - nxt1);
3570 else if (nxt2 - nxt1 < U16_MAX)
3571 NEXT_OFF(nxt1) = nxt2 - nxt1;
3573 OP(nxt) = NOTHING; /* Cannot beautify */
3578 /* Optimize again: */
3579 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3580 NULL, stopparen, recursed, NULL, 0,depth+1);
3585 else if ((OP(oscan) == CURLYX)
3586 && (flags & SCF_WHILEM_VISITED_POS)
3587 /* See the comment on a similar expression above.
3588 However, this time it's not a subexpression
3589 we care about, but the expression itself. */
3590 && (maxcount == REG_INFTY)
3591 && data && ++data->whilem_c < 16) {
3592 /* This stays as CURLYX, we can put the count/of pair. */
3593 /* Find WHILEM (as in regexec.c) */
3594 regnode *nxt = oscan + NEXT_OFF(oscan);
3596 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3598 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3599 | (RExC_whilem_seen << 4)); /* On WHILEM */
3601 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3603 if (flags & SCF_DO_SUBSTR) {
3604 SV *last_str = NULL;
3605 int counted = mincount != 0;
3607 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3608 #if defined(SPARC64_GCC_WORKAROUND)
3611 const char *s = NULL;
3614 if (pos_before >= data->last_start_min)
3617 b = data->last_start_min;
3620 s = SvPV_const(data->last_found, l);
3621 old = b - data->last_start_min;
3624 I32 b = pos_before >= data->last_start_min
3625 ? pos_before : data->last_start_min;
3627 const char * const s = SvPV_const(data->last_found, l);
3628 I32 old = b - data->last_start_min;
3632 old = utf8_hop((U8*)s, old) - (U8*)s;
3634 /* Get the added string: */
3635 last_str = newSVpvn_utf8(s + old, l, UTF);
3636 if (deltanext == 0 && pos_before == b) {
3637 /* What was added is a constant string */
3639 SvGROW(last_str, (mincount * l) + 1);
3640 repeatcpy(SvPVX(last_str) + l,
3641 SvPVX_const(last_str), l, mincount - 1);
3642 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3643 /* Add additional parts. */
3644 SvCUR_set(data->last_found,
3645 SvCUR(data->last_found) - l);
3646 sv_catsv(data->last_found, last_str);
3648 SV * sv = data->last_found;
3650 SvUTF8(sv) && SvMAGICAL(sv) ?
3651 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3652 if (mg && mg->mg_len >= 0)
3653 mg->mg_len += CHR_SVLEN(last_str) - l;
3655 data->last_end += l * (mincount - 1);
3658 /* start offset must point into the last copy */
3659 data->last_start_min += minnext * (mincount - 1);
3660 data->last_start_max += is_inf ? I32_MAX
3661 : (maxcount - 1) * (minnext + data->pos_delta);
3664 /* It is counted once already... */
3665 data->pos_min += minnext * (mincount - counted);
3666 data->pos_delta += - counted * deltanext +
3667 (minnext + deltanext) * maxcount - minnext * mincount;
3668 if (mincount != maxcount) {
3669 /* Cannot extend fixed substrings found inside
3671 SCAN_COMMIT(pRExC_state,data,minlenp);
3672 if (mincount && last_str) {
3673 SV * const sv = data->last_found;
3674 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3675 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3679 sv_setsv(sv, last_str);
3680 data->last_end = data->pos_min;
3681 data->last_start_min =
3682 data->pos_min - CHR_SVLEN(last_str);
3683 data->last_start_max = is_inf
3685 : data->pos_min + data->pos_delta
3686 - CHR_SVLEN(last_str);
3688 data->longest = &(data->longest_float);
3690 SvREFCNT_dec(last_str);
3692 if (data && (fl & SF_HAS_EVAL))
3693 data->flags |= SF_HAS_EVAL;
3694 optimize_curly_tail:
3695 if (OP(oscan) != CURLYX) {
3696 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3698 NEXT_OFF(oscan) += NEXT_OFF(next);
3701 default: /* REF, ANYOFV, and CLUMP only? */
3702 if (flags & SCF_DO_SUBSTR) {
3703 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3704 data->longest = &(data->longest_float);
3706 is_inf = is_inf_internal = 1;
3707 if (flags & SCF_DO_STCLASS_OR)
3708 cl_anything(pRExC_state, data->start_class);
3709 flags &= ~SCF_DO_STCLASS;
3713 else if (OP(scan) == LNBREAK) {
3714 if (flags & SCF_DO_STCLASS) {
3716 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3717 if (flags & SCF_DO_STCLASS_AND) {
3718 for (value = 0; value < 256; value++)
3719 if (!is_VERTWS_cp(value))
3720 ANYOF_BITMAP_CLEAR(data->start_class, value);
3723 for (value = 0; value < 256; value++)
3724 if (is_VERTWS_cp(value))
3725 ANYOF_BITMAP_SET(data->start_class, value);
3727 if (flags & SCF_DO_STCLASS_OR)
3728 cl_and(data->start_class, and_withp);
3729 flags &= ~SCF_DO_STCLASS;
3733 if (flags & SCF_DO_SUBSTR) {
3734 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3736 data->pos_delta += 1;
3737 data->longest = &(data->longest_float);
3740 else if (OP(scan) == FOLDCHAR) {
3741 int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2;
3742 flags &= ~SCF_DO_STCLASS;
3745 if (flags & SCF_DO_SUBSTR) {
3746 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3748 data->pos_delta += d;
3749 data->longest = &(data->longest_float);
3752 else if (REGNODE_SIMPLE(OP(scan))) {
3755 if (flags & SCF_DO_SUBSTR) {
3756 SCAN_COMMIT(pRExC_state,data,minlenp);
3760 if (flags & SCF_DO_STCLASS) {
3761 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3763 /* Some of the logic below assumes that switching
3764 locale on will only add false positives. */
3765 switch (PL_regkind[OP(scan)]) {
3769 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3770 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3771 cl_anything(pRExC_state, data->start_class);
3774 if (OP(scan) == SANY)
3776 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3777 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3778 || ANYOF_CLASS_TEST_ANY_SET(data->start_class));
3779 cl_anything(pRExC_state, data->start_class);
3781 if (flags & SCF_DO_STCLASS_AND || !value)
3782 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3785 if (flags & SCF_DO_STCLASS_AND)
3786 cl_and(data->start_class,
3787 (struct regnode_charclass_class*)scan);
3789 cl_or(pRExC_state, data->start_class,
3790 (struct regnode_charclass_class*)scan);
3793 if (flags & SCF_DO_STCLASS_AND) {
3794 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3795 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3796 if (OP(scan) == ALNUMU) {
3797 for (value = 0; value < 256; value++) {
3798 if (!isWORDCHAR_L1(value)) {
3799 ANYOF_BITMAP_CLEAR(data->start_class, value);
3803 for (value = 0; value < 256; value++) {
3804 if (!isALNUM(value)) {
3805 ANYOF_BITMAP_CLEAR(data->start_class, value);
3812 if (data->start_class->flags & ANYOF_LOCALE)
3813 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3815 /* Even if under locale, set the bits for non-locale
3816 * in case it isn't a true locale-node. This will
3817 * create false positives if it truly is locale */
3818 if (OP(scan) == ALNUMU) {
3819 for (value = 0; value < 256; value++) {
3820 if (isWORDCHAR_L1(value)) {
3821 ANYOF_BITMAP_SET(data->start_class, value);
3825 for (value = 0; value < 256; value++) {
3826 if (isALNUM(value)) {
3827 ANYOF_BITMAP_SET(data->start_class, value);
3834 if (flags & SCF_DO_STCLASS_AND) {
3835 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3836 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3837 if (OP(scan) == NALNUMU) {
3838 for (value = 0; value < 256; value++) {
3839 if (isWORDCHAR_L1(value)) {
3840 ANYOF_BITMAP_CLEAR(data->start_class, value);
3844 for (value = 0; value < 256; value++) {
3845 if (isALNUM(value)) {
3846 ANYOF_BITMAP_CLEAR(data->start_class, value);
3853 if (data->start_class->flags & ANYOF_LOCALE)
3854 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3856 /* Even if under locale, set the bits for non-locale in
3857 * case it isn't a true locale-node. This will create
3858 * false positives if it truly is locale */
3859 if (OP(scan) == NALNUMU) {
3860 for (value = 0; value < 256; value++) {
3861 if (! isWORDCHAR_L1(value)) {
3862 ANYOF_BITMAP_SET(data->start_class, value);
3866 for (value = 0; value < 256; value++) {
3867 if (! isALNUM(value)) {
3868 ANYOF_BITMAP_SET(data->start_class, value);
3875 if (flags & SCF_DO_STCLASS_AND) {
3876 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3877 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3878 if (OP(scan) == SPACEU) {
3879 for (value = 0; value < 256; value++) {
3880 if (!isSPACE_L1(value)) {
3881 ANYOF_BITMAP_CLEAR(data->start_class, value);
3885 for (value = 0; value < 256; value++) {
3886 if (!isSPACE(value)) {
3887 ANYOF_BITMAP_CLEAR(data->start_class, value);
3894 if (data->start_class->flags & ANYOF_LOCALE) {
3895 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3897 if (OP(scan) == SPACEU) {
3898 for (value = 0; value < 256; value++) {
3899 if (isSPACE_L1(value)) {
3900 ANYOF_BITMAP_SET(data->start_class, value);
3904 for (value = 0; value < 256; value++) {
3905 if (isSPACE(value)) {
3906 ANYOF_BITMAP_SET(data->start_class, value);
3913 if (flags & SCF_DO_STCLASS_AND) {
3914 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3915 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3916 if (OP(scan) == NSPACEU) {
3917 for (value = 0; value < 256; value++) {
3918 if (isSPACE_L1(value)) {
3919 ANYOF_BITMAP_CLEAR(data->start_class, value);
3923 for (value = 0; value < 256; value++) {
3924 if (isSPACE(value)) {
3925 ANYOF_BITMAP_CLEAR(data->start_class, value);
3932 if (data->start_class->flags & ANYOF_LOCALE)
3933 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3934 if (OP(scan) == NSPACEU) {
3935 for (value = 0; value < 256; value++) {
3936 if (!isSPACE_L1(value)) {
3937 ANYOF_BITMAP_SET(data->start_class, value);
3942 for (value = 0; value < 256; value++) {
3943 if (!isSPACE(value)) {
3944 ANYOF_BITMAP_SET(data->start_class, value);
3951 if (flags & SCF_DO_STCLASS_AND) {
3952 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3953 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3954 for (value = 0; value < 256; value++)
3955 if (!isDIGIT(value))
3956 ANYOF_BITMAP_CLEAR(data->start_class, value);
3960 if (data->start_class->flags & ANYOF_LOCALE)
3961 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3962 for (value = 0; value < 256; value++)
3964 ANYOF_BITMAP_SET(data->start_class, value);
3968 if (flags & SCF_DO_STCLASS_AND) {
3969 if (!(data->start_class->flags & ANYOF_LOCALE))
3970 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3971 for (value = 0; value < 256; value++)
3973 ANYOF_BITMAP_CLEAR(data->start_class, value);
3976 if (data->start_class->flags & ANYOF_LOCALE)
3977 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3978 for (value = 0; value < 256; value++)
3979 if (!isDIGIT(value))
3980 ANYOF_BITMAP_SET(data->start_class, value);
3983 CASE_SYNST_FNC(VERTWS);
3984 CASE_SYNST_FNC(HORIZWS);
3987 if (flags & SCF_DO_STCLASS_OR)
3988 cl_and(data->start_class, and_withp);
3989 flags &= ~SCF_DO_STCLASS;
3992 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3993 data->flags |= (OP(scan) == MEOL
3997 else if ( PL_regkind[OP(scan)] == BRANCHJ
3998 /* Lookbehind, or need to calculate parens/evals/stclass: */
3999 && (scan->flags || data || (flags & SCF_DO_STCLASS))
4000 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
4001 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4002 || OP(scan) == UNLESSM )
4004 /* Negative Lookahead/lookbehind
4005 In this case we can't do fixed string optimisation.
4008 I32 deltanext, minnext, fake = 0;
4010 struct regnode_charclass_class intrnl;
4013 data_fake.flags = 0;
4015 data_fake.whilem_c = data->whilem_c;
4016 data_fake.last_closep = data->last_closep;
4019 data_fake.last_closep = &fake;
4020 data_fake.pos_delta = delta;
4021 if ( flags & SCF_DO_STCLASS && !scan->flags
4022 && OP(scan) == IFMATCH ) { /* Lookahead */
4023 cl_init(pRExC_state, &intrnl);
4024 data_fake.start_class = &intrnl;
4025 f |= SCF_DO_STCLASS_AND;
4027 if (flags & SCF_WHILEM_VISITED_POS)
4028 f |= SCF_WHILEM_VISITED_POS;
4029 next = regnext(scan);
4030 nscan = NEXTOPER(NEXTOPER(scan));
4031 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
4032 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
4035 FAIL("Variable length lookbehind not implemented");
4037 else if (minnext > (I32)U8_MAX) {
4038 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4040 scan->flags = (U8)minnext;
4043 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4045 if (data_fake.flags & SF_HAS_EVAL)
4046 data->flags |= SF_HAS_EVAL;
4047 data->whilem_c = data_fake.whilem_c;
4049 if (f & SCF_DO_STCLASS_AND) {
4050 if (flags & SCF_DO_STCLASS_OR) {
4051 /* OR before, AND after: ideally we would recurse with
4052 * data_fake to get the AND applied by study of the
4053 * remainder of the pattern, and then derecurse;
4054 * *** HACK *** for now just treat as "no information".
4055 * See [perl #56690].
4057 cl_init(pRExC_state, data->start_class);
4059 /* AND before and after: combine and continue */
4060 const int was = (data->start_class->flags & ANYOF_EOS);
4062 cl_and(data->start_class, &intrnl);
4064 data->start_class->flags |= ANYOF_EOS;
4068 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4070 /* Positive Lookahead/lookbehind
4071 In this case we can do fixed string optimisation,
4072 but we must be careful about it. Note in the case of
4073 lookbehind the positions will be offset by the minimum
4074 length of the pattern, something we won't know about
4075 until after the recurse.
4077 I32 deltanext, fake = 0;
4079 struct regnode_charclass_class intrnl;
4081 /* We use SAVEFREEPV so that when the full compile
4082 is finished perl will clean up the allocated
4083 minlens when it's all done. This way we don't
4084 have to worry about freeing them when we know
4085 they wont be used, which would be a pain.
4088 Newx( minnextp, 1, I32 );
4089 SAVEFREEPV(minnextp);
4092 StructCopy(data, &data_fake, scan_data_t);
4093 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
4096 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
4097 data_fake.last_found=newSVsv(data->last_found);
4101 data_fake.last_closep = &fake;
4102 data_fake.flags = 0;
4103 data_fake.pos_delta = delta;
4105 data_fake.flags |= SF_IS_INF;
4106 if ( flags & SCF_DO_STCLASS && !scan->flags
4107 && OP(scan) == IFMATCH ) { /* Lookahead */
4108 cl_init(pRExC_state, &intrnl);
4109 data_fake.start_class = &intrnl;
4110 f |= SCF_DO_STCLASS_AND;
4112 if (flags & SCF_WHILEM_VISITED_POS)
4113 f |= SCF_WHILEM_VISITED_POS;
4114 next = regnext(scan);
4115 nscan = NEXTOPER(NEXTOPER(scan));
4117 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
4118 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
4121 FAIL("Variable length lookbehind not implemented");
4123 else if (*minnextp > (I32)U8_MAX) {
4124 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4126 scan->flags = (U8)*minnextp;
4131 if (f & SCF_DO_STCLASS_AND) {
4132 const int was = (data->start_class->flags & ANYOF_EOS);
4134 cl_and(data->start_class, &intrnl);
4136 data->start_class->flags |= ANYOF_EOS;
4139 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4141 if (data_fake.flags & SF_HAS_EVAL)
4142 data->flags |= SF_HAS_EVAL;
4143 data->whilem_c = data_fake.whilem_c;
4144 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
4145 if (RExC_rx->minlen<*minnextp)
4146 RExC_rx->minlen=*minnextp;
4147 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
4148 SvREFCNT_dec(data_fake.last_found);
4150 if ( data_fake.minlen_fixed != minlenp )
4152 data->offset_fixed= data_fake.offset_fixed;
4153 data->minlen_fixed= data_fake.minlen_fixed;
4154 data->lookbehind_fixed+= scan->flags;
4156 if ( data_fake.minlen_float != minlenp )
4158 data->minlen_float= data_fake.minlen_float;
4159 data->offset_float_min=data_fake.offset_float_min;
4160 data->offset_float_max=data_fake.offset_float_max;
4161 data->lookbehind_float+= scan->flags;
4170 else if (OP(scan) == OPEN) {
4171 if (stopparen != (I32)ARG(scan))
4174 else if (OP(scan) == CLOSE) {
4175 if (stopparen == (I32)ARG(scan)) {
4178 if ((I32)ARG(scan) == is_par) {
4179 next = regnext(scan);
4181 if ( next && (OP(next) != WHILEM) && next < last)
4182 is_par = 0; /* Disable optimization */
4185 *(data->last_closep) = ARG(scan);
4187 else if (OP(scan) == EVAL) {
4189 data->flags |= SF_HAS_EVAL;
4191 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
4192 if (flags & SCF_DO_SUBSTR) {
4193 SCAN_COMMIT(pRExC_state,data,minlenp);
4194 flags &= ~SCF_DO_SUBSTR;
4196 if (data && OP(scan)==ACCEPT) {
4197 data->flags |= SCF_SEEN_ACCEPT;
4202 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
4204 if (flags & SCF_DO_SUBSTR) {
4205 SCAN_COMMIT(pRExC_state,data,minlenp);
4206 data->longest = &(data->longest_float);
4208 is_inf = is_inf_internal = 1;
4209 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
4210 cl_anything(pRExC_state, data->start_class);
4211 flags &= ~SCF_DO_STCLASS;
4213 else if (OP(scan) == GPOS) {
4214 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
4215 !(delta || is_inf || (data && data->pos_delta)))
4217 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
4218 RExC_rx->extflags |= RXf_ANCH_GPOS;
4219 if (RExC_rx->gofs < (U32)min)
4220 RExC_rx->gofs = min;
4222 RExC_rx->extflags |= RXf_GPOS_FLOAT;
4226 #ifdef TRIE_STUDY_OPT
4227 #ifdef FULL_TRIE_STUDY
4228 else if (PL_regkind[OP(scan)] == TRIE) {
4229 /* NOTE - There is similar code to this block above for handling
4230 BRANCH nodes on the initial study. If you change stuff here
4232 regnode *trie_node= scan;
4233 regnode *tail= regnext(scan);
4234 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4235 I32 max1 = 0, min1 = I32_MAX;
4236 struct regnode_charclass_class accum;
4238 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
4239 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
4240 if (flags & SCF_DO_STCLASS)
4241 cl_init_zero(pRExC_state, &accum);
4247 const regnode *nextbranch= NULL;
4250 for ( word=1 ; word <= trie->wordcount ; word++)
4252 I32 deltanext=0, minnext=0, f = 0, fake;
4253 struct regnode_charclass_class this_class;
4255 data_fake.flags = 0;
4257 data_fake.whilem_c = data->whilem_c;
4258 data_fake.last_closep = data->last_closep;
4261 data_fake.last_closep = &fake;
4262 data_fake.pos_delta = delta;
4263 if (flags & SCF_DO_STCLASS) {
4264 cl_init(pRExC_state, &this_class);
4265 data_fake.start_class = &this_class;
4266 f = SCF_DO_STCLASS_AND;
4268 if (flags & SCF_WHILEM_VISITED_POS)
4269 f |= SCF_WHILEM_VISITED_POS;
4271 if (trie->jump[word]) {
4273 nextbranch = trie_node + trie->jump[0];
4274 scan= trie_node + trie->jump[word];
4275 /* We go from the jump point to the branch that follows
4276 it. Note this means we need the vestigal unused branches
4277 even though they arent otherwise used.
4279 minnext = study_chunk(pRExC_state, &scan, minlenp,
4280 &deltanext, (regnode *)nextbranch, &data_fake,
4281 stopparen, recursed, NULL, f,depth+1);
4283 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4284 nextbranch= regnext((regnode*)nextbranch);
4286 if (min1 > (I32)(minnext + trie->minlen))
4287 min1 = minnext + trie->minlen;
4288 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4289 max1 = minnext + deltanext + trie->maxlen;
4290 if (deltanext == I32_MAX)
4291 is_inf = is_inf_internal = 1;
4293 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4295 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4296 if ( stopmin > min + min1)
4297 stopmin = min + min1;
4298 flags &= ~SCF_DO_SUBSTR;
4300 data->flags |= SCF_SEEN_ACCEPT;
4303 if (data_fake.flags & SF_HAS_EVAL)
4304 data->flags |= SF_HAS_EVAL;
4305 data->whilem_c = data_fake.whilem_c;
4307 if (flags & SCF_DO_STCLASS)
4308 cl_or(pRExC_state, &accum, &this_class);
4311 if (flags & SCF_DO_SUBSTR) {
4312 data->pos_min += min1;
4313 data->pos_delta += max1 - min1;
4314 if (max1 != min1 || is_inf)
4315 data->longest = &(data->longest_float);
4318 delta += max1 - min1;
4319 if (flags & SCF_DO_STCLASS_OR) {
4320 cl_or(pRExC_state, data->start_class, &accum);
4322 cl_and(data->start_class, and_withp);
4323 flags &= ~SCF_DO_STCLASS;
4326 else if (flags & SCF_DO_STCLASS_AND) {
4328 cl_and(data->start_class, &accum);
4329 flags &= ~SCF_DO_STCLASS;
4332 /* Switch to OR mode: cache the old value of
4333 * data->start_class */
4335 StructCopy(data->start_class, and_withp,
4336 struct regnode_charclass_class);
4337 flags &= ~SCF_DO_STCLASS_AND;
4338 StructCopy(&accum, data->start_class,
4339 struct regnode_charclass_class);
4340 flags |= SCF_DO_STCLASS_OR;
4341 data->start_class->flags |= ANYOF_EOS;
4348 else if (PL_regkind[OP(scan)] == TRIE) {
4349 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4352 min += trie->minlen;
4353 delta += (trie->maxlen - trie->minlen);
4354 flags &= ~SCF_DO_STCLASS; /* xxx */
4355 if (flags & SCF_DO_SUBSTR) {
4356 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4357 data->pos_min += trie->minlen;
4358 data->pos_delta += (trie->maxlen - trie->minlen);
4359 if (trie->maxlen != trie->minlen)
4360 data->longest = &(data->longest_float);
4362 if (trie->jump) /* no more substrings -- for now /grr*/
4363 flags &= ~SCF_DO_SUBSTR;
4365 #endif /* old or new */
4366 #endif /* TRIE_STUDY_OPT */
4368 /* Else: zero-length, ignore. */
4369 scan = regnext(scan);
4374 stopparen = frame->stop;
4375 frame = frame->prev;
4376 goto fake_study_recurse;
4381 DEBUG_STUDYDATA("pre-fin:",data,depth);
4384 *deltap = is_inf_internal ? I32_MAX : delta;
4385 if (flags & SCF_DO_SUBSTR && is_inf)
4386 data->pos_delta = I32_MAX - data->pos_min;
4387 if (is_par > (I32)U8_MAX)
4389 if (is_par && pars==1 && data) {
4390 data->flags |= SF_IN_PAR;
4391 data->flags &= ~SF_HAS_PAR;
4393 else if (pars && data) {
4394 data->flags |= SF_HAS_PAR;
4395 data->flags &= ~SF_IN_PAR;
4397 if (flags & SCF_DO_STCLASS_OR)
4398 cl_and(data->start_class, and_withp);
4399 if (flags & SCF_TRIE_RESTUDY)
4400 data->flags |= SCF_TRIE_RESTUDY;
4402 DEBUG_STUDYDATA("post-fin:",data,depth);
4404 return min < stopmin ? min : stopmin;
4408 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4410 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4412 PERL_ARGS_ASSERT_ADD_DATA;
4414 Renewc(RExC_rxi->data,
4415 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4416 char, struct reg_data);
4418 Renew(RExC_rxi->data->what, count + n, U8);
4420 Newx(RExC_rxi->data->what, n, U8);
4421 RExC_rxi->data->count = count + n;
4422 Copy(s, RExC_rxi->data->what + count, n, U8);
4426 /*XXX: todo make this not included in a non debugging perl */
4427 #ifndef PERL_IN_XSUB_RE
4429 Perl_reginitcolors(pTHX)
4432 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4434 char *t = savepv(s);
4438 t = strchr(t, '\t');
4444 PL_colors[i] = t = (char *)"";
4449 PL_colors[i++] = (char *)"";
4456 #ifdef TRIE_STUDY_OPT
4457 #define CHECK_RESTUDY_GOTO \
4459 (data.flags & SCF_TRIE_RESTUDY) \
4463 #define CHECK_RESTUDY_GOTO
4467 - pregcomp - compile a regular expression into internal code
4469 * We can't allocate space until we know how big the compiled form will be,
4470 * but we can't compile it (and thus know how big it is) until we've got a
4471 * place to put the code. So we cheat: we compile it twice, once with code
4472 * generation turned off and size counting turned on, and once "for real".
4473 * This also means that we don't allocate space until we are sure that the
4474 * thing really will compile successfully, and we never have to move the
4475 * code and thus invalidate pointers into it. (Note that it has to be in
4476 * one piece because free() must be able to free it all.) [NB: not true in perl]
4478 * Beware that the optimization-preparation code in here knows about some
4479 * of the structure of the compiled regexp. [I'll say.]
4484 #ifndef PERL_IN_XSUB_RE
4485 #define RE_ENGINE_PTR &PL_core_reg_engine
4487 extern const struct regexp_engine my_reg_engine;
4488 #define RE_ENGINE_PTR &my_reg_engine
4491 #ifndef PERL_IN_XSUB_RE
4493 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4496 HV * const table = GvHV(PL_hintgv);
4498 PERL_ARGS_ASSERT_PREGCOMP;
4500 /* Dispatch a request to compile a regexp to correct
4503 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4504 GET_RE_DEBUG_FLAGS_DECL;
4505 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4506 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4508 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4511 return CALLREGCOMP_ENG(eng, pattern, flags);
4514 return Perl_re_compile(aTHX_ pattern, flags);
4519 Perl_re_compile(pTHX_ SV * const pattern, U32 orig_pm_flags)
4524 register regexp_internal *ri;
4533 /* these are all flags - maybe they should be turned
4534 * into a single int with different bit masks */
4535 I32 sawlookahead = 0;
4538 bool used_setjump = FALSE;
4539 regex_charset initial_charset = get_regex_charset(orig_pm_flags);
4544 RExC_state_t RExC_state;
4545 RExC_state_t * const pRExC_state = &RExC_state;
4546 #ifdef TRIE_STUDY_OPT
4548 RExC_state_t copyRExC_state;
4550 GET_RE_DEBUG_FLAGS_DECL;
4552 PERL_ARGS_ASSERT_RE_COMPILE;
4554 DEBUG_r(if (!PL_colorset) reginitcolors());
4556 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4557 RExC_uni_semantics = 0;
4558 RExC_contains_locale = 0;
4560 /****************** LONG JUMP TARGET HERE***********************/
4561 /* Longjmp back to here if have to switch in midstream to utf8 */
4562 if (! RExC_orig_utf8) {
4563 JMPENV_PUSH(jump_ret);
4564 used_setjump = TRUE;
4567 if (jump_ret == 0) { /* First time through */
4568 exp = SvPV(pattern, plen);
4570 /* ignore the utf8ness if the pattern is 0 length */
4572 RExC_utf8 = RExC_orig_utf8 = 0;
4576 SV *dsv= sv_newmortal();
4577 RE_PV_QUOTED_DECL(s, RExC_utf8,
4578 dsv, exp, plen, 60);
4579 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4580 PL_colors[4],PL_colors[5],s);
4583 else { /* longjumped back */
4586 /* If the cause for the longjmp was other than changing to utf8, pop
4587 * our own setjmp, and longjmp to the correct handler */
4588 if (jump_ret != UTF8_LONGJMP) {
4590 JMPENV_JUMP(jump_ret);
4595 /* It's possible to write a regexp in ascii that represents Unicode
4596 codepoints outside of the byte range, such as via \x{100}. If we
4597 detect such a sequence we have to convert the entire pattern to utf8
4598 and then recompile, as our sizing calculation will have been based
4599 on 1 byte == 1 character, but we will need to use utf8 to encode
4600 at least some part of the pattern, and therefore must convert the whole
4603 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4604 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4605 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4607 RExC_orig_utf8 = RExC_utf8 = 1;
4611 #ifdef TRIE_STUDY_OPT
4615 pm_flags = orig_pm_flags;
4617 if (initial_charset == REGEX_LOCALE_CHARSET) {
4618 RExC_contains_locale = 1;
4620 else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) {
4622 /* Set to use unicode semantics if the pattern is in utf8 and has the
4623 * 'depends' charset specified, as it means unicode when utf8 */
4624 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4628 RExC_flags = pm_flags;
4632 RExC_in_lookbehind = 0;
4633 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4634 RExC_seen_evals = 0;
4636 RExC_override_recoding = 0;
4638 /* First pass: determine size, legality. */
4646 RExC_emit = &PL_regdummy;
4647 RExC_whilem_seen = 0;
4648 RExC_open_parens = NULL;
4649 RExC_close_parens = NULL;
4651 RExC_paren_names = NULL;
4653 RExC_paren_name_list = NULL;
4655 RExC_recurse = NULL;
4656 RExC_recurse_count = 0;
4658 #if 0 /* REGC() is (currently) a NOP at the first pass.
4659 * Clever compilers notice this and complain. --jhi */
4660 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4662 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4663 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4664 RExC_precomp = NULL;
4668 /* Here, finished first pass. Get rid of any added setjmp */
4674 PerlIO_printf(Perl_debug_log,
4675 "Required size %"IVdf" nodes\n"
4676 "Starting second pass (creation)\n",
4679 RExC_lastparse=NULL;
4682 /* The first pass could have found things that force Unicode semantics */
4683 if ((RExC_utf8 || RExC_uni_semantics)
4684 && get_regex_charset(pm_flags) == REGEX_DEPENDS_CHARSET)
4686 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4689 /* Small enough for pointer-storage convention?
4690 If extralen==0, this means that we will not need long jumps. */
4691 if (RExC_size >= 0x10000L && RExC_extralen)
4692 RExC_size += RExC_extralen;
4695 if (RExC_whilem_seen > 15)
4696 RExC_whilem_seen = 15;
4698 /* Allocate space and zero-initialize. Note, the two step process
4699 of zeroing when in debug mode, thus anything assigned has to
4700 happen after that */
4701 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4702 r = (struct regexp*)SvANY(rx);
4703 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4704 char, regexp_internal);
4705 if ( r == NULL || ri == NULL )
4706 FAIL("Regexp out of space");
4708 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4709 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4711 /* bulk initialize base fields with 0. */
4712 Zero(ri, sizeof(regexp_internal), char);
4715 /* non-zero initialization begins here */
4717 r->engine= RE_ENGINE_PTR;
4718 r->extflags = pm_flags;
4720 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4721 bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET);
4723 /* The caret is output if there are any defaults: if not all the STD
4724 * flags are set, or if no character set specifier is needed */
4726 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4728 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4729 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4730 >> RXf_PMf_STD_PMMOD_SHIFT);
4731 const char *fptr = STD_PAT_MODS; /*"msix"*/
4733 /* Allocate for the worst case, which is all the std flags are turned
4734 * on. If more precision is desired, we could do a population count of
4735 * the flags set. This could be done with a small lookup table, or by
4736 * shifting, masking and adding, or even, when available, assembly
4737 * language for a machine-language population count.
4738 * We never output a minus, as all those are defaults, so are
4739 * covered by the caret */
4740 const STRLEN wraplen = plen + has_p + has_runon
4741 + has_default /* If needs a caret */
4743 /* If needs a character set specifier */
4744 + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0)
4745 + (sizeof(STD_PAT_MODS) - 1)
4746 + (sizeof("(?:)") - 1);
4748 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4750 SvFLAGS(rx) |= SvUTF8(pattern);
4753 /* If a default, cover it using the caret */
4755 *p++= DEFAULT_PAT_MOD;
4759 const char* const name = get_regex_charset_name(r->extflags, &len);
4760 Copy(name, p, len, char);
4764 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4767 while((ch = *fptr++)) {
4775 Copy(RExC_precomp, p, plen, char);
4776 assert ((RX_WRAPPED(rx) - p) < 16);
4777 r->pre_prefix = p - RX_WRAPPED(rx);
4783 SvCUR_set(rx, p - SvPVX_const(rx));
4787 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4789 if (RExC_seen & REG_SEEN_RECURSE) {
4790 Newxz(RExC_open_parens, RExC_npar,regnode *);
4791 SAVEFREEPV(RExC_open_parens);
4792 Newxz(RExC_close_parens,RExC_npar,regnode *);
4793 SAVEFREEPV(RExC_close_parens);
4796 /* Useful during FAIL. */
4797 #ifdef RE_TRACK_PATTERN_OFFSETS
4798 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4799 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4800 "%s %"UVuf" bytes for offset annotations.\n",
4801 ri->u.offsets ? "Got" : "Couldn't get",
4802 (UV)((2*RExC_size+1) * sizeof(U32))));
4804 SetProgLen(ri,RExC_size);
4808 REH_CALL_COMP_BEGIN_HOOK(pRExC_state->rx);
4810 /* Second pass: emit code. */
4811 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4816 RExC_emit_start = ri->program;
4817 RExC_emit = ri->program;
4818 RExC_emit_bound = ri->program + RExC_size + 1;
4820 /* Store the count of eval-groups for security checks: */
4821 RExC_rx->seen_evals = RExC_seen_evals;
4822 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4823 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4827 /* XXXX To minimize changes to RE engine we always allocate
4828 3-units-long substrs field. */
4829 Newx(r->substrs, 1, struct reg_substr_data);
4830 if (RExC_recurse_count) {
4831 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4832 SAVEFREEPV(RExC_recurse);
4836 r->minlen = minlen = sawlookahead = sawplus = sawopen = 0;
4837 Zero(r->substrs, 1, struct reg_substr_data);
4839 #ifdef TRIE_STUDY_OPT
4841 StructCopy(&zero_scan_data, &data, scan_data_t);
4842 copyRExC_state = RExC_state;
4845 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4847 RExC_state = copyRExC_state;
4848 if (seen & REG_TOP_LEVEL_BRANCHES)
4849 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4851 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4852 if (data.last_found) {
4853 SvREFCNT_dec(data.longest_fixed);
4854 SvREFCNT_dec(data.longest_float);
4855 SvREFCNT_dec(data.last_found);
4857 StructCopy(&zero_scan_data, &data, scan_data_t);
4860 StructCopy(&zero_scan_data, &data, scan_data_t);
4863 /* Dig out information for optimizations. */
4864 r->extflags = RExC_flags; /* was pm_op */
4865 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4868 SvUTF8_on(rx); /* Unicode in it? */
4869 ri->regstclass = NULL;
4870 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4871 r->intflags |= PREGf_NAUGHTY;
4872 scan = ri->program + 1; /* First BRANCH. */
4874 /* testing for BRANCH here tells us whether there is "must appear"
4875 data in the pattern. If there is then we can use it for optimisations */
4876 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4878 STRLEN longest_float_length, longest_fixed_length;
4879 struct regnode_charclass_class ch_class; /* pointed to by data */
4881 I32 last_close = 0; /* pointed to by data */
4882 regnode *first= scan;
4883 regnode *first_next= regnext(first);
4885 * Skip introductions and multiplicators >= 1
4886 * so that we can extract the 'meat' of the pattern that must
4887 * match in the large if() sequence following.
4888 * NOTE that EXACT is NOT covered here, as it is normally
4889 * picked up by the optimiser separately.
4891 * This is unfortunate as the optimiser isnt handling lookahead
4892 * properly currently.
4895 while ((OP(first) == OPEN && (sawopen = 1)) ||
4896 /* An OR of *one* alternative - should not happen now. */
4897 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4898 /* for now we can't handle lookbehind IFMATCH*/
4899 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
4900 (OP(first) == PLUS) ||
4901 (OP(first) == MINMOD) ||
4902 /* An {n,m} with n>0 */
4903 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4904 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4907 * the only op that could be a regnode is PLUS, all the rest
4908 * will be regnode_1 or regnode_2.
4911 if (OP(first) == PLUS)
4914 first += regarglen[OP(first)];
4916 first = NEXTOPER(first);
4917 first_next= regnext(first);
4920 /* Starting-point info. */
4922 DEBUG_PEEP("first:",first,0);
4923 /* Ignore EXACT as we deal with it later. */
4924 if (PL_regkind[OP(first)] == EXACT) {
4925 if (OP(first) == EXACT)
4926 NOOP; /* Empty, get anchored substr later. */
4928 ri->regstclass = first;
4931 else if (PL_regkind[OP(first)] == TRIE &&
4932 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4935 /* this can happen only on restudy */
4936 if ( OP(first) == TRIE ) {
4937 struct regnode_1 *trieop = (struct regnode_1 *)
4938 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4939 StructCopy(first,trieop,struct regnode_1);
4940 trie_op=(regnode *)trieop;
4942 struct regnode_charclass *trieop = (struct regnode_charclass *)
4943 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4944 StructCopy(first,trieop,struct regnode_charclass);
4945 trie_op=(regnode *)trieop;
4948 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4949 ri->regstclass = trie_op;
4952 else if (REGNODE_SIMPLE(OP(first)))
4953 ri->regstclass = first;
4954 else if (PL_regkind[OP(first)] == BOUND ||
4955 PL_regkind[OP(first)] == NBOUND)
4956 ri->regstclass = first;
4957 else if (PL_regkind[OP(first)] == BOL) {
4958 r->extflags |= (OP(first) == MBOL
4960 : (OP(first) == SBOL
4963 first = NEXTOPER(first);
4966 else if (OP(first) == GPOS) {
4967 r->extflags |= RXf_ANCH_GPOS;
4968 first = NEXTOPER(first);
4971 else if ((!sawopen || !RExC_sawback) &&
4972 (OP(first) == STAR &&
4973 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4974 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4976 /* turn .* into ^.* with an implied $*=1 */
4978 (OP(NEXTOPER(first)) == REG_ANY)
4981 r->extflags |= type;
4982 r->intflags |= PREGf_IMPLICIT;
4983 first = NEXTOPER(first);
4986 if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback)
4987 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4988 /* x+ must match at the 1st pos of run of x's */
4989 r->intflags |= PREGf_SKIP;
4991 /* Scan is after the zeroth branch, first is atomic matcher. */
4992 #ifdef TRIE_STUDY_OPT
4995 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4996 (IV)(first - scan + 1))
5000 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
5001 (IV)(first - scan + 1))
5007 * If there's something expensive in the r.e., find the
5008 * longest literal string that must appear and make it the
5009 * regmust. Resolve ties in favor of later strings, since
5010 * the regstart check works with the beginning of the r.e.
5011 * and avoiding duplication strengthens checking. Not a
5012 * strong reason, but sufficient in the absence of others.
5013 * [Now we resolve ties in favor of the earlier string if
5014 * it happens that c_offset_min has been invalidated, since the
5015 * earlier string may buy us something the later one won't.]
5018 data.longest_fixed = newSVpvs("");
5019 data.longest_float = newSVpvs("");
5020 data.last_found = newSVpvs("");
5021 data.longest = &(data.longest_fixed);
5023 if (!ri->regstclass) {
5024 cl_init(pRExC_state, &ch_class);
5025 data.start_class = &ch_class;
5026 stclass_flag = SCF_DO_STCLASS_AND;
5027 } else /* XXXX Check for BOUND? */
5029 data.last_closep = &last_close;
5031 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
5032 &data, -1, NULL, NULL,
5033 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
5039 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
5040 && data.last_start_min == 0 && data.last_end > 0
5041 && !RExC_seen_zerolen
5042 && !(RExC_seen & REG_SEEN_VERBARG)
5043 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
5044 r->extflags |= RXf_CHECK_ALL;
5045 scan_commit(pRExC_state, &data,&minlen,0);
5046 SvREFCNT_dec(data.last_found);
5048 /* Note that code very similar to this but for anchored string
5049 follows immediately below, changes may need to be made to both.
5052 longest_float_length = CHR_SVLEN(data.longest_float);
5053 if (longest_float_length
5054 || (data.flags & SF_FL_BEFORE_EOL
5055 && (!(data.flags & SF_FL_BEFORE_MEOL)
5056 || (RExC_flags & RXf_PMf_MULTILINE))))
5060 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
5061 && data.offset_fixed == data.offset_float_min
5062 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
5063 goto remove_float; /* As in (a)+. */
5065 /* copy the information about the longest float from the reg_scan_data
5066 over to the program. */
5067 if (SvUTF8(data.longest_float)) {
5068 r->float_utf8 = data.longest_float;
5069 r->float_substr = NULL;
5071 r->float_substr = data.longest_float;
5072 r->float_utf8 = NULL;
5074 /* float_end_shift is how many chars that must be matched that
5075 follow this item. We calculate it ahead of time as once the
5076 lookbehind offset is added in we lose the ability to correctly
5078 ml = data.minlen_float ? *(data.minlen_float)
5079 : (I32)longest_float_length;
5080 r->float_end_shift = ml - data.offset_float_min
5081 - longest_float_length + (SvTAIL(data.longest_float) != 0)
5082 + data.lookbehind_float;
5083 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
5084 r->float_max_offset = data.offset_float_max;
5085 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
5086 r->float_max_offset -= data.lookbehind_float;
5088 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
5089 && (!(data.flags & SF_FL_BEFORE_MEOL)
5090 || (RExC_flags & RXf_PMf_MULTILINE)));
5091 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
5095 r->float_substr = r->float_utf8 = NULL;
5096 SvREFCNT_dec(data.longest_float);
5097 longest_float_length = 0;
5100 /* Note that code very similar to this but for floating string
5101 is immediately above, changes may need to be made to both.
5104 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
5105 if (longest_fixed_length
5106 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
5107 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5108 || (RExC_flags & RXf_PMf_MULTILINE))))
5112 /* copy the information about the longest fixed
5113 from the reg_scan_data over to the program. */
5114 if (SvUTF8(data.longest_fixed)) {
5115 r->anchored_utf8 = data.longest_fixed;
5116 r->anchored_substr = NULL;
5118 r->anchored_substr = data.longest_fixed;
5119 r->anchored_utf8 = NULL;
5121 /* fixed_end_shift is how many chars that must be matched that
5122 follow this item. We calculate it ahead of time as once the
5123 lookbehind offset is added in we lose the ability to correctly
5125 ml = data.minlen_fixed ? *(data.minlen_fixed)
5126 : (I32)longest_fixed_length;
5127 r->anchored_end_shift = ml - data.offset_fixed
5128 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
5129 + data.lookbehind_fixed;
5130 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
5132 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
5133 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5134 || (RExC_flags & RXf_PMf_MULTILINE)));
5135 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
5138 r->anchored_substr = r->anchored_utf8 = NULL;
5139 SvREFCNT_dec(data.longest_fixed);
5140 longest_fixed_length = 0;
5143 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
5144 ri->regstclass = NULL;
5146 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
5148 && !(data.start_class->flags & ANYOF_EOS)
5149 && !cl_is_anything(data.start_class))
5151 const U32 n = add_data(pRExC_state, 1, "f");
5152 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5154 Newx(RExC_rxi->data->data[n], 1,
5155 struct regnode_charclass_class);
5156 StructCopy(data.start_class,
5157 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5158 struct regnode_charclass_class);
5159 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5160 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5161 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
5162 regprop(r, sv, (regnode*)data.start_class);
5163 PerlIO_printf(Perl_debug_log,
5164 "synthetic stclass \"%s\".\n",
5165 SvPVX_const(sv));});
5168 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
5169 if (longest_fixed_length > longest_float_length) {
5170 r->check_end_shift = r->anchored_end_shift;
5171 r->check_substr = r->anchored_substr;
5172 r->check_utf8 = r->anchored_utf8;
5173 r->check_offset_min = r->check_offset_max = r->anchored_offset;
5174 if (r->extflags & RXf_ANCH_SINGLE)
5175 r->extflags |= RXf_NOSCAN;
5178 r->check_end_shift = r->float_end_shift;
5179 r->check_substr = r->float_substr;
5180 r->check_utf8 = r->float_utf8;
5181 r->check_offset_min = r->float_min_offset;
5182 r->check_offset_max = r->float_max_offset;
5184 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
5185 This should be changed ASAP! */
5186 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
5187 r->extflags |= RXf_USE_INTUIT;
5188 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
5189 r->extflags |= RXf_INTUIT_TAIL;
5191 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
5192 if ( (STRLEN)minlen < longest_float_length )
5193 minlen= longest_float_length;
5194 if ( (STRLEN)minlen < longest_fixed_length )
5195 minlen= longest_fixed_length;
5199 /* Several toplevels. Best we can is to set minlen. */
5201 struct regnode_charclass_class ch_class;
5204 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
5206 scan = ri->program + 1;
5207 cl_init(pRExC_state, &ch_class);
5208 data.start_class = &ch_class;
5209 data.last_closep = &last_close;
5212 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
5213 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
5217 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
5218 = r->float_substr = r->float_utf8 = NULL;
5220 if (!(data.start_class->flags & ANYOF_EOS)
5221 && !cl_is_anything(data.start_class))
5223 const U32 n = add_data(pRExC_state, 1, "f");
5224 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5226 Newx(RExC_rxi->data->data[n], 1,
5227 struct regnode_charclass_class);
5228 StructCopy(data.start_class,
5229 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5230 struct regnode_charclass_class);
5231 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5232 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5233 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
5234 regprop(r, sv, (regnode*)data.start_class);
5235 PerlIO_printf(Perl_debug_log,
5236 "synthetic stclass \"%s\".\n",
5237 SvPVX_const(sv));});
5241 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
5242 the "real" pattern. */
5244 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
5245 (IV)minlen, (IV)r->minlen);
5247 r->minlenret = minlen;
5248 if (r->minlen < minlen)
5251 if (RExC_seen & REG_SEEN_GPOS)
5252 r->extflags |= RXf_GPOS_SEEN;
5253 if (RExC_seen & REG_SEEN_LOOKBEHIND)
5254 r->extflags |= RXf_LOOKBEHIND_SEEN;
5255 if (RExC_seen & REG_SEEN_EVAL)
5256 r->extflags |= RXf_EVAL_SEEN;
5257 if (RExC_seen & REG_SEEN_CANY)
5258 r->extflags |= RXf_CANY_SEEN;
5259 if (RExC_seen & REG_SEEN_VERBARG)
5260 r->intflags |= PREGf_VERBARG_SEEN;
5261 if (RExC_seen & REG_SEEN_CUTGROUP)
5262 r->intflags |= PREGf_CUTGROUP_SEEN;
5263 if (RExC_paren_names)
5264 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
5266 RXp_PAREN_NAMES(r) = NULL;
5268 #ifdef STUPID_PATTERN_CHECKS
5269 if (RX_PRELEN(rx) == 0)
5270 r->extflags |= RXf_NULL;
5271 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5272 /* XXX: this should happen BEFORE we compile */
5273 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5274 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
5275 r->extflags |= RXf_WHITE;
5276 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
5277 r->extflags |= RXf_START_ONLY;
5279 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5280 /* XXX: this should happen BEFORE we compile */
5281 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5283 regnode *first = ri->program + 1;
5286 if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END)
5287 r->extflags |= RXf_NULL;
5288 else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END)
5289 r->extflags |= RXf_START_ONLY;
5290 else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE
5291 && OP(regnext(first)) == END)
5292 r->extflags |= RXf_WHITE;
5296 if (RExC_paren_names) {
5297 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5298 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5301 ri->name_list_idx = 0;
5303 if (RExC_recurse_count) {
5304 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5305 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5306 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5309 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5310 /* assume we don't need to swap parens around before we match */
5313 PerlIO_printf(Perl_debug_log,"Final program:\n");
5316 #ifdef RE_TRACK_PATTERN_OFFSETS
5317 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5318 const U32 len = ri->u.offsets[0];
5320 GET_RE_DEBUG_FLAGS_DECL;
5321 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5322 for (i = 1; i <= len; i++) {
5323 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5324 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5325 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5327 PerlIO_printf(Perl_debug_log, "\n");
5333 #undef RE_ENGINE_PTR
5337 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5340 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5342 PERL_UNUSED_ARG(value);
5344 if (flags & RXapif_FETCH) {
5345 return reg_named_buff_fetch(rx, key, flags);
5346 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5347 Perl_croak_no_modify(aTHX);
5349 } else if (flags & RXapif_EXISTS) {
5350 return reg_named_buff_exists(rx, key, flags)
5353 } else if (flags & RXapif_REGNAMES) {
5354 return reg_named_buff_all(rx, flags);
5355 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5356 return reg_named_buff_scalar(rx, flags);
5358 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5364 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5367 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5368 PERL_UNUSED_ARG(lastkey);
5370 if (flags & RXapif_FIRSTKEY)
5371 return reg_named_buff_firstkey(rx, flags);
5372 else if (flags & RXapif_NEXTKEY)
5373 return reg_named_buff_nextkey(rx, flags);
5375 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5381 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5384 AV *retarray = NULL;
5386 struct regexp *const rx = (struct regexp *)SvANY(r);
5388 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5390 if (flags & RXapif_ALL)
5393 if (rx && RXp_PAREN_NAMES(rx)) {
5394 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5397 SV* sv_dat=HeVAL(he_str);
5398 I32 *nums=(I32*)SvPVX(sv_dat);
5399 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5400 if ((I32)(rx->nparens) >= nums[i]
5401 && rx->offs[nums[i]].start != -1
5402 && rx->offs[nums[i]].end != -1)
5405 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5409 ret = newSVsv(&PL_sv_undef);
5412 av_push(retarray, ret);
5415 return newRV_noinc(MUTABLE_SV(retarray));
5422 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5425 struct regexp *const rx = (struct regexp *)SvANY(r);
5427 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5429 if (rx && RXp_PAREN_NAMES(rx)) {
5430 if (flags & RXapif_ALL) {
5431 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5433 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5447 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5449 struct regexp *const rx = (struct regexp *)SvANY(r);
5451 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5453 if ( rx && RXp_PAREN_NAMES(rx) ) {
5454 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5456 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5463 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5465 struct regexp *const rx = (struct regexp *)SvANY(r);
5466 GET_RE_DEBUG_FLAGS_DECL;
5468 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5470 if (rx && RXp_PAREN_NAMES(rx)) {
5471 HV *hv = RXp_PAREN_NAMES(rx);
5473 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5476 SV* sv_dat = HeVAL(temphe);
5477 I32 *nums = (I32*)SvPVX(sv_dat);
5478 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5479 if ((I32)(rx->lastparen) >= nums[i] &&
5480 rx->offs[nums[i]].start != -1 &&
5481 rx->offs[nums[i]].end != -1)
5487 if (parno || flags & RXapif_ALL) {
5488 return newSVhek(HeKEY_hek(temphe));
5496 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5501 struct regexp *const rx = (struct regexp *)SvANY(r);
5503 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5505 if (rx && RXp_PAREN_NAMES(rx)) {
5506 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5507 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5508 } else if (flags & RXapif_ONE) {
5509 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5510 av = MUTABLE_AV(SvRV(ret));
5511 length = av_len(av);
5513 return newSViv(length + 1);
5515 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5519 return &PL_sv_undef;
5523 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5525 struct regexp *const rx = (struct regexp *)SvANY(r);
5528 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5530 if (rx && RXp_PAREN_NAMES(rx)) {
5531 HV *hv= RXp_PAREN_NAMES(rx);
5533 (void)hv_iterinit(hv);
5534 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5537 SV* sv_dat = HeVAL(temphe);
5538 I32 *nums = (I32*)SvPVX(sv_dat);
5539 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5540 if ((I32)(rx->lastparen) >= nums[i] &&
5541 rx->offs[nums[i]].start != -1 &&
5542 rx->offs[nums[i]].end != -1)
5548 if (parno || flags & RXapif_ALL) {
5549 av_push(av, newSVhek(HeKEY_hek(temphe)));
5554 return newRV_noinc(MUTABLE_SV(av));
5558 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5561 struct regexp *const rx = (struct regexp *)SvANY(r);
5566 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5569 sv_setsv(sv,&PL_sv_undef);
5573 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5575 i = rx->offs[0].start;
5579 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5581 s = rx->subbeg + rx->offs[0].end;
5582 i = rx->sublen - rx->offs[0].end;
5585 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5586 (s1 = rx->offs[paren].start) != -1 &&
5587 (t1 = rx->offs[paren].end) != -1)
5591 s = rx->subbeg + s1;
5593 sv_setsv(sv,&PL_sv_undef);
5596 assert(rx->sublen >= (s - rx->subbeg) + i );
5598 const int oldtainted = PL_tainted;
5600 sv_setpvn(sv, s, i);
5601 PL_tainted = oldtainted;
5602 if ( (rx->extflags & RXf_CANY_SEEN)
5603 ? (RXp_MATCH_UTF8(rx)
5604 && (!i || is_utf8_string((U8*)s, i)))
5605 : (RXp_MATCH_UTF8(rx)) )
5612 if (RXp_MATCH_TAINTED(rx)) {
5613 if (SvTYPE(sv) >= SVt_PVMG) {
5614 MAGIC* const mg = SvMAGIC(sv);
5617 SvMAGIC_set(sv, mg->mg_moremagic);
5619 if ((mgt = SvMAGIC(sv))) {
5620 mg->mg_moremagic = mgt;
5621 SvMAGIC_set(sv, mg);
5631 sv_setsv(sv,&PL_sv_undef);
5637 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5638 SV const * const value)
5640 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5642 PERL_UNUSED_ARG(rx);
5643 PERL_UNUSED_ARG(paren);
5644 PERL_UNUSED_ARG(value);
5647 Perl_croak_no_modify(aTHX);
5651 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5654 struct regexp *const rx = (struct regexp *)SvANY(r);
5658 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5660 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5662 /* $` / ${^PREMATCH} */
5663 case RX_BUFF_IDX_PREMATCH:
5664 if (rx->offs[0].start != -1) {
5665 i = rx->offs[0].start;
5673 /* $' / ${^POSTMATCH} */
5674 case RX_BUFF_IDX_POSTMATCH:
5675 if (rx->offs[0].end != -1) {
5676 i = rx->sublen - rx->offs[0].end;
5678 s1 = rx->offs[0].end;
5684 /* $& / ${^MATCH}, $1, $2, ... */
5686 if (paren <= (I32)rx->nparens &&
5687 (s1 = rx->offs[paren].start) != -1 &&
5688 (t1 = rx->offs[paren].end) != -1)
5693 if (ckWARN(WARN_UNINITIALIZED))
5694 report_uninit((const SV *)sv);
5699 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5700 const char * const s = rx->subbeg + s1;
5705 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5712 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5714 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5715 PERL_UNUSED_ARG(rx);
5719 return newSVpvs("Regexp");
5722 /* Scans the name of a named buffer from the pattern.
5723 * If flags is REG_RSN_RETURN_NULL returns null.
5724 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5725 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5726 * to the parsed name as looked up in the RExC_paren_names hash.
5727 * If there is an error throws a vFAIL().. type exception.
5730 #define REG_RSN_RETURN_NULL 0
5731 #define REG_RSN_RETURN_NAME 1
5732 #define REG_RSN_RETURN_DATA 2
5735 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5737 char *name_start = RExC_parse;
5739 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5741 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5742 /* skip IDFIRST by using do...while */
5745 RExC_parse += UTF8SKIP(RExC_parse);
5746 } while (isALNUM_utf8((U8*)RExC_parse));
5750 } while (isALNUM(*RExC_parse));
5755 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5756 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5757 if ( flags == REG_RSN_RETURN_NAME)
5759 else if (flags==REG_RSN_RETURN_DATA) {
5762 if ( ! sv_name ) /* should not happen*/
5763 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5764 if (RExC_paren_names)
5765 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5767 sv_dat = HeVAL(he_str);
5769 vFAIL("Reference to nonexistent named group");
5773 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5780 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5781 int rem=(int)(RExC_end - RExC_parse); \
5790 if (RExC_lastparse!=RExC_parse) \
5791 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5794 iscut ? "..." : "<" \
5797 PerlIO_printf(Perl_debug_log,"%16s",""); \
5800 num = RExC_size + 1; \
5802 num=REG_NODE_NUM(RExC_emit); \
5803 if (RExC_lastnum!=num) \
5804 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5806 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5807 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5808 (int)((depth*2)), "", \
5812 RExC_lastparse=RExC_parse; \
5817 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5818 DEBUG_PARSE_MSG((funcname)); \
5819 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5821 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5822 DEBUG_PARSE_MSG((funcname)); \
5823 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5826 /* This section of code defines the inversion list object and its methods. The
5827 * interfaces are highly subject to change, so as much as possible is static to
5828 * this file. An inversion list is here implemented as a malloc'd C UV array
5829 * with some added info that is placed as UVs at the beginning in a header
5830 * portion. An inversion list for Unicode is an array of code points, sorted
5831 * by ordinal number. The zeroth element is the first code point in the list.
5832 * The 1th element is the first element beyond that not in the list. In other
5833 * words, the first range is
5834 * invlist[0]..(invlist[1]-1)
5835 * The other ranges follow. Thus every element that is divisible by two marks
5836 * the beginning of a range that is in the list, and every element not
5837 * divisible by two marks the beginning of a range not in the list. A single
5838 * element inversion list that contains the single code point N generally
5839 * consists of two elements
5842 * (The exception is when N is the highest representable value on the
5843 * machine, in which case the list containing just it would be a single
5844 * element, itself. By extension, if the last range in the list extends to
5845 * infinity, then the first element of that range will be in the inversion list
5846 * at a position that is divisible by two, and is the final element in the
5848 * Taking the complement (inverting) an inversion list is quite simple, if the
5849 * first element is 0, remove it; otherwise add a 0 element at the beginning.
5850 * This implementation reserves an element at the beginning of each inversion list
5851 * to contain 0 when the list contains 0, and contains 1 otherwise. The actual
5852 * beginning of the list is either that element if 0, or the next one if 1.
5854 * More about inversion lists can be found in "Unicode Demystified"
5855 * Chapter 13 by Richard Gillam, published by Addison-Wesley.
5856 * More will be coming when functionality is added later.
5858 * The inversion list data structure is currently implemented as an SV pointing
5859 * to an array of UVs that the SV thinks are bytes. This allows us to have an
5860 * array of UV whose memory management is automatically handled by the existing
5861 * facilities for SV's.
5863 * Some of the methods should always be private to the implementation, and some
5864 * should eventually be made public */
5866 #define INVLIST_LEN_OFFSET 0 /* Number of elements in the inversion list */
5867 #define INVLIST_ITER_OFFSET 1 /* Current iteration position */
5869 #define INVLIST_ZERO_OFFSET 2 /* 0 or 1; must be last element in header */
5870 /* The UV at position ZERO contains either 0 or 1. If 0, the inversion list
5871 * contains the code point U+00000, and begins here. If 1, the inversion list
5872 * doesn't contain U+0000, and it begins at the next UV in the array.
5873 * Inverting an inversion list consists of adding or removing the 0 at the
5874 * beginning of it. By reserving a space for that 0, inversion can be made
5877 #define HEADER_LENGTH (INVLIST_ZERO_OFFSET + 1)
5879 /* Internally things are UVs */
5880 #define TO_INTERNAL_SIZE(x) ((x + HEADER_LENGTH) * sizeof(UV))
5881 #define FROM_INTERNAL_SIZE(x) ((x / sizeof(UV)) - HEADER_LENGTH)
5883 #define INVLIST_INITIAL_LEN 10
5885 PERL_STATIC_INLINE UV*
5886 S__invlist_array_init(pTHX_ SV* const invlist, const bool will_have_0)
5888 /* Returns a pointer to the first element in the inversion list's array.
5889 * This is called upon initialization of an inversion list. Where the
5890 * array begins depends on whether the list has the code point U+0000
5891 * in it or not. The other parameter tells it whether the code that
5892 * follows this call is about to put a 0 in the inversion list or not.
5893 * The first element is either the element with 0, if 0, or the next one,
5896 UV* zero = get_invlist_zero_addr(invlist);
5898 PERL_ARGS_ASSERT__INVLIST_ARRAY_INIT;
5901 assert(! *get_invlist_len_addr(invlist));
5903 /* 1^1 = 0; 1^0 = 1 */
5904 *zero = 1 ^ will_have_0;
5905 return zero + *zero;
5908 PERL_STATIC_INLINE UV*
5909 S_invlist_array(pTHX_ SV* const invlist)
5911 /* Returns the pointer to the inversion list's array. Every time the
5912 * length changes, this needs to be called in case malloc or realloc moved
5915 PERL_ARGS_ASSERT_INVLIST_ARRAY;
5917 /* Must not be empty */
5918 assert(*get_invlist_len_addr(invlist));
5919 assert(*get_invlist_zero_addr(invlist) == 0
5920 || *get_invlist_zero_addr(invlist) == 1);
5922 /* The array begins either at the element reserved for zero if the
5923 * list contains 0 (that element will be set to 0), or otherwise the next
5924 * element (in which case the reserved element will be set to 1). */
5925 return (UV *) (get_invlist_zero_addr(invlist)
5926 + *get_invlist_zero_addr(invlist));
5929 PERL_STATIC_INLINE UV*
5930 S_get_invlist_len_addr(pTHX_ SV* invlist)
5932 /* Return the address of the UV that contains the current number
5933 * of used elements in the inversion list */
5935 PERL_ARGS_ASSERT_GET_INVLIST_LEN_ADDR;
5937 return (UV *) (SvPVX(invlist) + (INVLIST_LEN_OFFSET * sizeof (UV)));
5940 PERL_STATIC_INLINE UV
5941 S_invlist_len(pTHX_ SV* const invlist)
5943 /* Returns the current number of elements in the inversion list's array */
5945 PERL_ARGS_ASSERT_INVLIST_LEN;
5947 return *get_invlist_len_addr(invlist);
5950 PERL_STATIC_INLINE void
5951 S_invlist_set_len(pTHX_ SV* const invlist, const UV len)
5953 /* Sets the current number of elements stored in the inversion list */
5955 PERL_ARGS_ASSERT_INVLIST_SET_LEN;
5957 *get_invlist_len_addr(invlist) = len;
5959 assert(len <= SvLEN(invlist));
5961 SvCUR_set(invlist, TO_INTERNAL_SIZE(len));
5962 /* If the list contains U+0000, that element is part of the header,
5963 * and should not be counted as part of the array. It will contain
5964 * 0 in that case, and 1 otherwise. So we could flop 0=>1, 1=>0 and
5966 * SvCUR_set(invlist,
5967 * TO_INTERNAL_SIZE(len
5968 * - (*get_invlist_zero_addr(inv_list) ^ 1)));
5969 * But, this is only valid if len is not 0. The consequences of not doing
5970 * this is that the memory allocation code may think that 1 more UV is
5971 * being used than actually is, and so might do an unnecessary grow. That
5972 * seems worth not bothering to make this the precise amount.
5974 * Note that when inverting, SvCUR shouldn't change */
5977 PERL_STATIC_INLINE UV
5978 S_invlist_max(pTHX_ SV* const invlist)
5980 /* Returns the maximum number of elements storable in the inversion list's
5981 * array, without having to realloc() */
5983 PERL_ARGS_ASSERT_INVLIST_MAX;
5985 return FROM_INTERNAL_SIZE(SvLEN(invlist));
5988 PERL_STATIC_INLINE UV*
5989 S_get_invlist_zero_addr(pTHX_ SV* invlist)
5991 /* Return the address of the UV that is reserved to hold 0 if the inversion
5992 * list contains 0. This has to be the last element of the heading, as the
5993 * list proper starts with either it if 0, or the next element if not.
5994 * (But we force it to contain either 0 or 1) */
5996 PERL_ARGS_ASSERT_GET_INVLIST_ZERO_ADDR;
5998 return (UV *) (SvPVX(invlist) + (INVLIST_ZERO_OFFSET * sizeof (UV)));
6001 #ifndef PERL_IN_XSUB_RE
6003 Perl__new_invlist(pTHX_ IV initial_size)
6006 /* Return a pointer to a newly constructed inversion list, with enough
6007 * space to store 'initial_size' elements. If that number is negative, a
6008 * system default is used instead */
6012 if (initial_size < 0) {
6013 initial_size = INVLIST_INITIAL_LEN;
6016 /* Allocate the initial space */
6017 new_list = newSV(TO_INTERNAL_SIZE(initial_size));
6018 invlist_set_len(new_list, 0);
6020 /* Force iterinit() to be used to get iteration to work */
6021 *get_invlist_iter_addr(new_list) = UV_MAX;
6023 /* This should force a segfault if a method doesn't initialize this
6025 *get_invlist_zero_addr(new_list) = UV_MAX;
6032 S_invlist_extend(pTHX_ SV* const invlist, const UV new_max)
6034 /* Grow the maximum size of an inversion list */
6036 PERL_ARGS_ASSERT_INVLIST_EXTEND;
6038 SvGROW((SV *)invlist, TO_INTERNAL_SIZE(new_max));
6041 PERL_STATIC_INLINE void
6042 S_invlist_trim(pTHX_ SV* const invlist)
6044 PERL_ARGS_ASSERT_INVLIST_TRIM;
6046 /* Change the length of the inversion list to how many entries it currently
6049 SvPV_shrink_to_cur((SV *) invlist);
6052 /* An element is in an inversion list iff its index is even numbered: 0, 2, 4,
6055 #define ELEMENT_IN_INVLIST_SET(i) (! ((i) & 1))
6056 #define PREV_ELEMENT_IN_INVLIST_SET(i) (! ELEMENT_IN_INVLIST_SET(i))
6058 #ifndef PERL_IN_XSUB_RE
6060 Perl__append_range_to_invlist(pTHX_ SV* const invlist, const UV start, const UV end)
6062 /* Subject to change or removal. Append the range from 'start' to 'end' at
6063 * the end of the inversion list. The range must be above any existing
6067 UV max = invlist_max(invlist);
6068 UV len = invlist_len(invlist);
6070 PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST;
6072 if (len == 0) { /* Empty lists must be initialized */
6073 array = _invlist_array_init(invlist, start == 0);
6076 /* Here, the existing list is non-empty. The current max entry in the
6077 * list is generally the first value not in the set, except when the
6078 * set extends to the end of permissible values, in which case it is
6079 * the first entry in that final set, and so this call is an attempt to
6080 * append out-of-order */
6082 UV final_element = len - 1;
6083 array = invlist_array(invlist);
6084 if (array[final_element] > start
6085 || ELEMENT_IN_INVLIST_SET(final_element))
6087 Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list");
6090 /* Here, it is a legal append. If the new range begins with the first
6091 * value not in the set, it is extending the set, so the new first
6092 * value not in the set is one greater than the newly extended range.
6094 if (array[final_element] == start) {
6095 if (end != UV_MAX) {
6096 array[final_element] = end + 1;
6099 /* But if the end is the maximum representable on the machine,
6100 * just let the range that this would extend have no end */
6101 invlist_set_len(invlist, len - 1);
6107 /* Here the new range doesn't extend any existing set. Add it */
6109 len += 2; /* Includes an element each for the start and end of range */
6111 /* If overflows the existing space, extend, which may cause the array to be
6114 invlist_extend(invlist, len);
6115 invlist_set_len(invlist, len); /* Have to set len here to avoid assert
6116 failure in invlist_array() */
6117 array = invlist_array(invlist);
6120 invlist_set_len(invlist, len);
6123 /* The next item on the list starts the range, the one after that is
6124 * one past the new range. */
6125 array[len - 2] = start;
6126 if (end != UV_MAX) {
6127 array[len - 1] = end + 1;
6130 /* But if the end is the maximum representable on the machine, just let
6131 * the range have no end */
6132 invlist_set_len(invlist, len - 1);
6137 Perl__invlist_union(pTHX_ SV* const a, SV* const b, SV** output)
6139 /* Take the union of two inversion lists and point 'result' to it. If
6140 * 'result' on input points to one of the two lists, the reference count to
6141 * that list will be decremented.
6142 * The basis for this comes from "Unicode Demystified" Chapter 13 by
6143 * Richard Gillam, published by Addison-Wesley, and explained at some
6144 * length there. The preface says to incorporate its examples into your
6145 * code at your own risk.
6147 * The algorithm is like a merge sort.
6149 * XXX A potential performance improvement is to keep track as we go along
6150 * if only one of the inputs contributes to the result, meaning the other
6151 * is a subset of that one. In that case, we can skip the final copy and
6152 * return the larger of the input lists, but then outside code might need
6153 * to keep track of whether to free the input list or not */
6155 UV* array_a; /* a's array */
6157 UV len_a; /* length of a's array */
6160 SV* u; /* the resulting union */
6164 UV i_a = 0; /* current index into a's array */
6168 /* running count, as explained in the algorithm source book; items are
6169 * stopped accumulating and are output when the count changes to/from 0.
6170 * The count is incremented when we start a range that's in the set, and
6171 * decremented when we start a range that's not in the set. So its range
6172 * is 0 to 2. Only when the count is zero is something not in the set.
6176 PERL_ARGS_ASSERT__INVLIST_UNION;
6178 /* If either one is empty, the union is the other one */
6179 len_a = invlist_len(a);
6184 else if (output != &b) {
6185 *output = invlist_clone(b);
6187 /* else *output already = b; */
6190 else if ((len_b = invlist_len(b)) == 0) {
6194 else if (output != &a) {
6195 *output = invlist_clone(a);
6197 /* else *output already = a; */
6201 /* Here both lists exist and are non-empty */
6202 array_a = invlist_array(a);
6203 array_b = invlist_array(b);
6205 /* Size the union for the worst case: that the sets are completely
6207 u = _new_invlist(len_a + len_b);
6209 /* Will contain U+0000 if either component does */
6210 array_u = _invlist_array_init(u, (len_a > 0 && array_a[0] == 0)
6211 || (len_b > 0 && array_b[0] == 0));
6213 /* Go through each list item by item, stopping when exhausted one of
6215 while (i_a < len_a && i_b < len_b) {
6216 UV cp; /* The element to potentially add to the union's array */
6217 bool cp_in_set; /* is it in the the input list's set or not */
6219 /* We need to take one or the other of the two inputs for the union.
6220 * Since we are merging two sorted lists, we take the smaller of the
6221 * next items. In case of a tie, we take the one that is in its set
6222 * first. If we took one not in the set first, it would decrement the
6223 * count, possibly to 0 which would cause it to be output as ending the
6224 * range, and the next time through we would take the same number, and
6225 * output it again as beginning the next range. By doing it the
6226 * opposite way, there is no possibility that the count will be
6227 * momentarily decremented to 0, and thus the two adjoining ranges will
6228 * be seamlessly merged. (In a tie and both are in the set or both not
6229 * in the set, it doesn't matter which we take first.) */
6230 if (array_a[i_a] < array_b[i_b]
6231 || (array_a[i_a] == array_b[i_b] && ELEMENT_IN_INVLIST_SET(i_a)))
6233 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6237 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6241 /* Here, have chosen which of the two inputs to look at. Only output
6242 * if the running count changes to/from 0, which marks the
6243 * beginning/end of a range in that's in the set */
6246 array_u[i_u++] = cp;
6253 array_u[i_u++] = cp;
6258 /* Here, we are finished going through at least one of the lists, which
6259 * means there is something remaining in at most one. We check if the list
6260 * that hasn't been exhausted is positioned such that we are in the middle
6261 * of a range in its set or not. (i_a and i_b point to the element beyond
6262 * the one we care about.) If in the set, we decrement 'count'; if 0, there
6263 * is potentially more to output.
6264 * There are four cases:
6265 * 1) Both weren't in their sets, count is 0, and remains 0. What's left
6266 * in the union is entirely from the non-exhausted set.
6267 * 2) Both were in their sets, count is 2. Nothing further should
6268 * be output, as everything that remains will be in the exhausted
6269 * list's set, hence in the union; decrementing to 1 but not 0 insures
6271 * 3) the exhausted was in its set, non-exhausted isn't, count is 1.
6272 * Nothing further should be output because the union includes
6273 * everything from the exhausted set. Not decrementing ensures that.
6274 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1;
6275 * decrementing to 0 insures that we look at the remainder of the
6276 * non-exhausted set */
6277 if ((i_a != len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6278 || (i_b != len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6283 /* The final length is what we've output so far, plus what else is about to
6284 * be output. (If 'count' is non-zero, then the input list we exhausted
6285 * has everything remaining up to the machine's limit in its set, and hence
6286 * in the union, so there will be no further output. */
6289 /* At most one of the subexpressions will be non-zero */
6290 len_u += (len_a - i_a) + (len_b - i_b);
6293 /* Set result to final length, which can change the pointer to array_u, so
6295 if (len_u != invlist_len(u)) {
6296 invlist_set_len(u, len_u);
6298 array_u = invlist_array(u);
6301 /* When 'count' is 0, the list that was exhausted (if one was shorter than
6302 * the other) ended with everything above it not in its set. That means
6303 * that the remaining part of the union is precisely the same as the
6304 * non-exhausted list, so can just copy it unchanged. (If both list were
6305 * exhausted at the same time, then the operations below will be both 0.)
6308 IV copy_count; /* At most one will have a non-zero copy count */
6309 if ((copy_count = len_a - i_a) > 0) {
6310 Copy(array_a + i_a, array_u + i_u, copy_count, UV);
6312 else if ((copy_count = len_b - i_b) > 0) {
6313 Copy(array_b + i_b, array_u + i_u, copy_count, UV);
6317 /* We may be removing a reference to one of the inputs */
6318 if (&a == output || &b == output) {
6319 SvREFCNT_dec(*output);
6327 Perl__invlist_intersection(pTHX_ SV* const a, SV* const b, SV** i)
6329 /* Take the intersection of two inversion lists and point 'i' to it. If
6330 * 'i' on input points to one of the two lists, the reference count to that
6331 * list will be decremented.
6332 * The basis for this comes from "Unicode Demystified" Chapter 13 by
6333 * Richard Gillam, published by Addison-Wesley, and explained at some
6334 * length there. The preface says to incorporate its examples into your
6335 * code at your own risk. In fact, it had bugs
6337 * The algorithm is like a merge sort, and is essentially the same as the
6341 UV* array_a; /* a's array */
6343 UV len_a; /* length of a's array */
6346 SV* r; /* the resulting intersection */
6350 UV i_a = 0; /* current index into a's array */
6354 /* running count, as explained in the algorithm source book; items are
6355 * stopped accumulating and are output when the count changes to/from 2.
6356 * The count is incremented when we start a range that's in the set, and
6357 * decremented when we start a range that's not in the set. So its range
6358 * is 0 to 2. Only when the count is 2 is something in the intersection.
6362 PERL_ARGS_ASSERT__INVLIST_INTERSECTION;
6364 /* If either one is empty, the intersection is null */
6365 len_a = invlist_len(a);
6366 if ((len_a == 0) || ((len_b = invlist_len(b)) == 0)) {
6367 *i = _new_invlist(0);
6369 /* If the result is the same as one of the inputs, the input is being
6380 /* Here both lists exist and are non-empty */
6381 array_a = invlist_array(a);
6382 array_b = invlist_array(b);
6384 /* Size the intersection for the worst case: that the intersection ends up
6385 * fragmenting everything to be completely disjoint */
6386 r= _new_invlist(len_a + len_b);
6388 /* Will contain U+0000 iff both components do */
6389 array_r = _invlist_array_init(r, len_a > 0 && array_a[0] == 0
6390 && len_b > 0 && array_b[0] == 0);
6392 /* Go through each list item by item, stopping when exhausted one of
6394 while (i_a < len_a && i_b < len_b) {
6395 UV cp; /* The element to potentially add to the intersection's
6397 bool cp_in_set; /* Is it in the input list's set or not */
6399 /* We need to take one or the other of the two inputs for the
6400 * intersection. Since we are merging two sorted lists, we take the
6401 * smaller of the next items. In case of a tie, we take the one that
6402 * is not in its set first (a difference from the union algorithm). If
6403 * we took one in the set first, it would increment the count, possibly
6404 * to 2 which would cause it to be output as starting a range in the
6405 * intersection, and the next time through we would take that same
6406 * number, and output it again as ending the set. By doing it the
6407 * opposite of this, there is no possibility that the count will be
6408 * momentarily incremented to 2. (In a tie and both are in the set or
6409 * both not in the set, it doesn't matter which we take first.) */
6410 if (array_a[i_a] < array_b[i_b]
6411 || (array_a[i_a] == array_b[i_b] && ! ELEMENT_IN_INVLIST_SET(i_a)))
6413 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6417 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6421 /* Here, have chosen which of the two inputs to look at. Only output
6422 * if the running count changes to/from 2, which marks the
6423 * beginning/end of a range that's in the intersection */
6427 array_r[i_r++] = cp;
6432 array_r[i_r++] = cp;
6438 /* Here, we are finished going through at least one of the lists, which
6439 * means there is something remaining in at most one. We check if the list
6440 * that has been exhausted is positioned such that we are in the middle
6441 * of a range in its set or not. (i_a and i_b point to elements 1 beyond
6442 * the ones we care about.) There are four cases:
6443 * 1) Both weren't in their sets, count is 0, and remains 0. There's
6444 * nothing left in the intersection.
6445 * 2) Both were in their sets, count is 2 and perhaps is incremented to
6446 * above 2. What should be output is exactly that which is in the
6447 * non-exhausted set, as everything it has is also in the intersection
6448 * set, and everything it doesn't have can't be in the intersection
6449 * 3) The exhausted was in its set, non-exhausted isn't, count is 1, and
6450 * gets incremented to 2. Like the previous case, the intersection is
6451 * everything that remains in the non-exhausted set.
6452 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1, and
6453 * remains 1. And the intersection has nothing more. */
6454 if ((i_a == len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6455 || (i_b == len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6460 /* The final length is what we've output so far plus what else is in the
6461 * intersection. At most one of the subexpressions below will be non-zero */
6464 len_r += (len_a - i_a) + (len_b - i_b);
6467 /* Set result to final length, which can change the pointer to array_r, so
6469 if (len_r != invlist_len(r)) {
6470 invlist_set_len(r, len_r);
6472 array_r = invlist_array(r);
6475 /* Finish outputting any remaining */
6476 if (count >= 2) { /* At most one will have a non-zero copy count */
6478 if ((copy_count = len_a - i_a) > 0) {
6479 Copy(array_a + i_a, array_r + i_r, copy_count, UV);
6481 else if ((copy_count = len_b - i_b) > 0) {
6482 Copy(array_b + i_b, array_r + i_r, copy_count, UV);
6486 /* We may be removing a reference to one of the inputs */
6487 if (&a == i || &b == i) {
6498 S_add_range_to_invlist(pTHX_ SV* invlist, const UV start, const UV end)
6500 /* Add the range from 'start' to 'end' inclusive to the inversion list's
6501 * set. A pointer to the inversion list is returned. This may actually be
6502 * a new list, in which case the passed in one has been destroyed. The
6503 * passed in inversion list can be NULL, in which case a new one is created
6504 * with just the one range in it */
6509 if (invlist == NULL) {
6510 invlist = _new_invlist(2);
6514 len = invlist_len(invlist);
6517 /* If comes after the final entry, can just append it to the end */
6519 || start >= invlist_array(invlist)
6520 [invlist_len(invlist) - 1])
6522 _append_range_to_invlist(invlist, start, end);
6526 /* Here, can't just append things, create and return a new inversion list
6527 * which is the union of this range and the existing inversion list */
6528 range_invlist = _new_invlist(2);
6529 _append_range_to_invlist(range_invlist, start, end);
6531 _invlist_union(invlist, range_invlist, &invlist);
6533 /* The temporary can be freed */
6534 SvREFCNT_dec(range_invlist);
6539 PERL_STATIC_INLINE SV*
6540 S_add_cp_to_invlist(pTHX_ SV* invlist, const UV cp) {
6541 return add_range_to_invlist(invlist, cp, cp);
6544 #ifndef PERL_IN_XSUB_RE
6546 Perl__invlist_invert(pTHX_ SV* const invlist)
6548 /* Complement the input inversion list. This adds a 0 if the list didn't
6549 * have a zero; removes it otherwise. As described above, the data
6550 * structure is set up so that this is very efficient */
6552 UV* len_pos = get_invlist_len_addr(invlist);
6554 PERL_ARGS_ASSERT__INVLIST_INVERT;
6556 /* The inverse of matching nothing is matching everything */
6557 if (*len_pos == 0) {
6558 _append_range_to_invlist(invlist, 0, UV_MAX);
6562 /* The exclusive or complents 0 to 1; and 1 to 0. If the result is 1, the
6563 * zero element was a 0, so it is being removed, so the length decrements
6564 * by 1; and vice-versa. SvCUR is unaffected */
6565 if (*get_invlist_zero_addr(invlist) ^= 1) {
6574 Perl__invlist_invert_prop(pTHX_ SV* const invlist)
6576 /* Complement the input inversion list (which must be a Unicode property,
6577 * all of which don't match above the Unicode maximum code point.) And
6578 * Perl has chosen to not have the inversion match above that either. This
6579 * adds a 0x110000 if the list didn't end with it, and removes it if it did
6585 PERL_ARGS_ASSERT__INVLIST_INVERT_PROP;
6587 _invlist_invert(invlist);
6589 len = invlist_len(invlist);
6591 if (len != 0) { /* If empty do nothing */
6592 array = invlist_array(invlist);
6593 if (array[len - 1] != PERL_UNICODE_MAX + 1) {
6594 /* Add 0x110000. First, grow if necessary */
6596 if (invlist_max(invlist) < len) {
6597 invlist_extend(invlist, len);
6598 array = invlist_array(invlist);
6600 invlist_set_len(invlist, len);
6601 array[len - 1] = PERL_UNICODE_MAX + 1;
6603 else { /* Remove the 0x110000 */
6604 invlist_set_len(invlist, len - 1);
6612 PERL_STATIC_INLINE SV*
6613 S_invlist_clone(pTHX_ SV* const invlist)
6616 /* Return a new inversion list that is a copy of the input one, which is
6619 SV* new_invlist = _new_invlist(SvCUR(invlist));
6621 PERL_ARGS_ASSERT_INVLIST_CLONE;
6623 Copy(SvPVX(invlist), SvPVX(new_invlist), SvCUR(invlist), char);
6627 #ifndef PERL_IN_XSUB_RE
6629 Perl__invlist_subtract(pTHX_ SV* const a, SV* const b, SV** result)
6631 /* Point result to an inversion list which consists of all elements in 'a'
6632 * that aren't also in 'b' */
6634 PERL_ARGS_ASSERT__INVLIST_SUBTRACT;
6636 /* Subtracting nothing retains the original */
6637 if (invlist_len(b) == 0) {
6639 /* If the result is not to be the same variable as the original, create
6642 *result = invlist_clone(a);
6645 SV *b_copy = invlist_clone(b);
6646 _invlist_invert(b_copy); /* Everything not in 'b' */
6647 _invlist_intersection(a, b_copy, result); /* Everything in 'a' not in
6649 SvREFCNT_dec(b_copy);
6660 PERL_STATIC_INLINE UV*
6661 S_get_invlist_iter_addr(pTHX_ SV* invlist)
6663 /* Return the address of the UV that contains the current iteration
6666 PERL_ARGS_ASSERT_GET_INVLIST_ITER_ADDR;
6668 return (UV *) (SvPVX(invlist) + (INVLIST_ITER_OFFSET * sizeof (UV)));
6671 PERL_STATIC_INLINE void
6672 S_invlist_iterinit(pTHX_ SV* invlist) /* Initialize iterator for invlist */
6674 PERL_ARGS_ASSERT_INVLIST_ITERINIT;
6676 *get_invlist_iter_addr(invlist) = 0;
6680 S_invlist_iternext(pTHX_ SV* invlist, UV* start, UV* end)
6682 UV* pos = get_invlist_iter_addr(invlist);
6683 UV len = invlist_len(invlist);
6686 PERL_ARGS_ASSERT_INVLIST_ITERNEXT;
6689 *pos = UV_MAX; /* Force iternit() to be required next time */
6693 array = invlist_array(invlist);
6695 *start = array[(*pos)++];
6701 *end = array[(*pos)++] - 1;
6709 S_invlist_dump(pTHX_ SV* const invlist, const char * const header)
6711 /* Dumps out the ranges in an inversion list. The string 'header'
6712 * if present is output on a line before the first range */
6716 if (header && strlen(header)) {
6717 PerlIO_printf(Perl_debug_log, "%s\n", header);
6719 invlist_iterinit(invlist);
6720 while (invlist_iternext(invlist, &start, &end)) {
6721 if (end == UV_MAX) {
6722 PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. INFINITY\n", start);
6725 PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. 0x%04"UVXf"\n", start, end);
6731 #undef HEADER_LENGTH
6732 #undef INVLIST_INITIAL_LENGTH
6733 #undef TO_INTERNAL_SIZE
6734 #undef FROM_INTERNAL_SIZE
6735 #undef INVLIST_LEN_OFFSET
6736 #undef INVLIST_ZERO_OFFSET
6737 #undef INVLIST_ITER_OFFSET
6739 /* End of inversion list object */
6742 - reg - regular expression, i.e. main body or parenthesized thing
6744 * Caller must absorb opening parenthesis.
6746 * Combining parenthesis handling with the base level of regular expression
6747 * is a trifle forced, but the need to tie the tails of the branches to what
6748 * follows makes it hard to avoid.
6750 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
6752 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
6754 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
6758 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
6759 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
6762 register regnode *ret; /* Will be the head of the group. */
6763 register regnode *br;
6764 register regnode *lastbr;
6765 register regnode *ender = NULL;
6766 register I32 parno = 0;
6768 U32 oregflags = RExC_flags;
6769 bool have_branch = 0;
6771 I32 freeze_paren = 0;
6772 I32 after_freeze = 0;
6774 /* for (?g), (?gc), and (?o) warnings; warning
6775 about (?c) will warn about (?g) -- japhy */
6777 #define WASTED_O 0x01
6778 #define WASTED_G 0x02
6779 #define WASTED_C 0x04
6780 #define WASTED_GC (0x02|0x04)
6781 I32 wastedflags = 0x00;
6783 char * parse_start = RExC_parse; /* MJD */
6784 char * const oregcomp_parse = RExC_parse;
6786 GET_RE_DEBUG_FLAGS_DECL;
6788 PERL_ARGS_ASSERT_REG;
6789 DEBUG_PARSE("reg ");
6791 *flagp = 0; /* Tentatively. */
6794 /* Make an OPEN node, if parenthesized. */
6796 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
6797 char *start_verb = RExC_parse;
6798 STRLEN verb_len = 0;
6799 char *start_arg = NULL;
6800 unsigned char op = 0;
6802 int internal_argval = 0; /* internal_argval is only useful if !argok */
6803 while ( *RExC_parse && *RExC_parse != ')' ) {
6804 if ( *RExC_parse == ':' ) {
6805 start_arg = RExC_parse + 1;
6811 verb_len = RExC_parse - start_verb;
6814 while ( *RExC_parse && *RExC_parse != ')' )
6816 if ( *RExC_parse != ')' )
6817 vFAIL("Unterminated verb pattern argument");
6818 if ( RExC_parse == start_arg )
6821 if ( *RExC_parse != ')' )
6822 vFAIL("Unterminated verb pattern");
6825 switch ( *start_verb ) {
6826 case 'A': /* (*ACCEPT) */
6827 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
6829 internal_argval = RExC_nestroot;
6832 case 'C': /* (*COMMIT) */
6833 if ( memEQs(start_verb,verb_len,"COMMIT") )
6836 case 'F': /* (*FAIL) */
6837 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
6842 case ':': /* (*:NAME) */
6843 case 'M': /* (*MARK:NAME) */
6844 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
6849 case 'P': /* (*PRUNE) */
6850 if ( memEQs(start_verb,verb_len,"PRUNE") )
6853 case 'S': /* (*SKIP) */
6854 if ( memEQs(start_verb,verb_len,"SKIP") )
6857 case 'T': /* (*THEN) */
6858 /* [19:06] <TimToady> :: is then */
6859 if ( memEQs(start_verb,verb_len,"THEN") ) {
6861 RExC_seen |= REG_SEEN_CUTGROUP;
6867 vFAIL3("Unknown verb pattern '%.*s'",
6868 verb_len, start_verb);
6871 if ( start_arg && internal_argval ) {
6872 vFAIL3("Verb pattern '%.*s' may not have an argument",
6873 verb_len, start_verb);
6874 } else if ( argok < 0 && !start_arg ) {
6875 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
6876 verb_len, start_verb);
6878 ret = reganode(pRExC_state, op, internal_argval);
6879 if ( ! internal_argval && ! SIZE_ONLY ) {
6881 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
6882 ARG(ret) = add_data( pRExC_state, 1, "S" );
6883 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
6890 if (!internal_argval)
6891 RExC_seen |= REG_SEEN_VERBARG;
6892 } else if ( start_arg ) {
6893 vFAIL3("Verb pattern '%.*s' may not have an argument",
6894 verb_len, start_verb);
6896 ret = reg_node(pRExC_state, op);
6898 nextchar(pRExC_state);
6901 if (*RExC_parse == '?') { /* (?...) */
6902 bool is_logical = 0;
6903 const char * const seqstart = RExC_parse;
6904 bool has_use_defaults = FALSE;
6907 paren = *RExC_parse++;
6908 ret = NULL; /* For look-ahead/behind. */
6911 case 'P': /* (?P...) variants for those used to PCRE/Python */
6912 paren = *RExC_parse++;
6913 if ( paren == '<') /* (?P<...>) named capture */
6915 else if (paren == '>') { /* (?P>name) named recursion */
6916 goto named_recursion;
6918 else if (paren == '=') { /* (?P=...) named backref */
6919 /* this pretty much dupes the code for \k<NAME> in regatom(), if
6920 you change this make sure you change that */
6921 char* name_start = RExC_parse;
6923 SV *sv_dat = reg_scan_name(pRExC_state,
6924 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6925 if (RExC_parse == name_start || *RExC_parse != ')')
6926 vFAIL2("Sequence %.3s... not terminated",parse_start);
6929 num = add_data( pRExC_state, 1, "S" );
6930 RExC_rxi->data->data[num]=(void*)sv_dat;
6931 SvREFCNT_inc_simple_void(sv_dat);
6934 ret = reganode(pRExC_state,
6937 : (MORE_ASCII_RESTRICTED)
6939 : (AT_LEAST_UNI_SEMANTICS)
6947 Set_Node_Offset(ret, parse_start+1);
6948 Set_Node_Cur_Length(ret); /* MJD */
6950 nextchar(pRExC_state);
6954 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6956 case '<': /* (?<...) */
6957 if (*RExC_parse == '!')
6959 else if (*RExC_parse != '=')
6965 case '\'': /* (?'...') */
6966 name_start= RExC_parse;
6967 svname = reg_scan_name(pRExC_state,
6968 SIZE_ONLY ? /* reverse test from the others */
6969 REG_RSN_RETURN_NAME :
6970 REG_RSN_RETURN_NULL);
6971 if (RExC_parse == name_start) {
6973 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6976 if (*RExC_parse != paren)
6977 vFAIL2("Sequence (?%c... not terminated",
6978 paren=='>' ? '<' : paren);
6982 if (!svname) /* shouldn't happen */
6984 "panic: reg_scan_name returned NULL");
6985 if (!RExC_paren_names) {
6986 RExC_paren_names= newHV();
6987 sv_2mortal(MUTABLE_SV(RExC_paren_names));
6989 RExC_paren_name_list= newAV();
6990 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
6993 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
6995 sv_dat = HeVAL(he_str);
6997 /* croak baby croak */
6999 "panic: paren_name hash element allocation failed");
7000 } else if ( SvPOK(sv_dat) ) {
7001 /* (?|...) can mean we have dupes so scan to check
7002 its already been stored. Maybe a flag indicating
7003 we are inside such a construct would be useful,
7004 but the arrays are likely to be quite small, so
7005 for now we punt -- dmq */
7006 IV count = SvIV(sv_dat);
7007 I32 *pv = (I32*)SvPVX(sv_dat);
7009 for ( i = 0 ; i < count ; i++ ) {
7010 if ( pv[i] == RExC_npar ) {
7016 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
7017 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
7018 pv[count] = RExC_npar;
7019 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
7022 (void)SvUPGRADE(sv_dat,SVt_PVNV);
7023 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
7025 SvIV_set(sv_dat, 1);
7028 /* Yes this does cause a memory leak in debugging Perls */
7029 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
7030 SvREFCNT_dec(svname);
7033 /*sv_dump(sv_dat);*/
7035 nextchar(pRExC_state);
7037 goto capturing_parens;
7039 RExC_seen |= REG_SEEN_LOOKBEHIND;
7040 RExC_in_lookbehind++;
7042 case '=': /* (?=...) */
7043 RExC_seen_zerolen++;
7045 case '!': /* (?!...) */
7046 RExC_seen_zerolen++;
7047 if (*RExC_parse == ')') {
7048 ret=reg_node(pRExC_state, OPFAIL);
7049 nextchar(pRExC_state);
7053 case '|': /* (?|...) */
7054 /* branch reset, behave like a (?:...) except that
7055 buffers in alternations share the same numbers */
7057 after_freeze = freeze_paren = RExC_npar;
7059 case ':': /* (?:...) */
7060 case '>': /* (?>...) */
7062 case '$': /* (?$...) */
7063 case '@': /* (?@...) */
7064 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
7066 case '#': /* (?#...) */
7067 while (*RExC_parse && *RExC_parse != ')')
7069 if (*RExC_parse != ')')
7070 FAIL("Sequence (?#... not terminated");
7071 nextchar(pRExC_state);
7074 case '0' : /* (?0) */
7075 case 'R' : /* (?R) */
7076 if (*RExC_parse != ')')
7077 FAIL("Sequence (?R) not terminated");
7078 ret = reg_node(pRExC_state, GOSTART);
7079 *flagp |= POSTPONED;
7080 nextchar(pRExC_state);
7083 { /* named and numeric backreferences */
7085 case '&': /* (?&NAME) */
7086 parse_start = RExC_parse - 1;
7089 SV *sv_dat = reg_scan_name(pRExC_state,
7090 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7091 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
7093 goto gen_recurse_regop;
7096 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
7098 vFAIL("Illegal pattern");
7100 goto parse_recursion;
7102 case '-': /* (?-1) */
7103 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
7104 RExC_parse--; /* rewind to let it be handled later */
7108 case '1': case '2': case '3': case '4': /* (?1) */
7109 case '5': case '6': case '7': case '8': case '9':
7112 num = atoi(RExC_parse);
7113 parse_start = RExC_parse - 1; /* MJD */
7114 if (*RExC_parse == '-')
7116 while (isDIGIT(*RExC_parse))
7118 if (*RExC_parse!=')')
7119 vFAIL("Expecting close bracket");
7122 if ( paren == '-' ) {
7124 Diagram of capture buffer numbering.
7125 Top line is the normal capture buffer numbers
7126 Bottom line is the negative indexing as from
7130 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
7134 num = RExC_npar + num;
7137 vFAIL("Reference to nonexistent group");
7139 } else if ( paren == '+' ) {
7140 num = RExC_npar + num - 1;
7143 ret = reganode(pRExC_state, GOSUB, num);
7145 if (num > (I32)RExC_rx->nparens) {
7147 vFAIL("Reference to nonexistent group");
7149 ARG2L_SET( ret, RExC_recurse_count++);
7151 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7152 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
7156 RExC_seen |= REG_SEEN_RECURSE;
7157 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
7158 Set_Node_Offset(ret, parse_start); /* MJD */
7160 *flagp |= POSTPONED;
7161 nextchar(pRExC_state);
7163 } /* named and numeric backreferences */
7166 case '?': /* (??...) */
7168 if (*RExC_parse != '{') {
7170 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7173 *flagp |= POSTPONED;
7174 paren = *RExC_parse++;
7176 case '{': /* (?{...}) */
7181 char *s = RExC_parse;
7183 RExC_seen_zerolen++;
7184 RExC_seen |= REG_SEEN_EVAL;
7185 while (count && (c = *RExC_parse)) {
7196 if (*RExC_parse != ')') {
7198 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
7202 OP_4tree *sop, *rop;
7203 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
7206 Perl_save_re_context(aTHX);
7207 rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad);
7208 sop->op_private |= OPpREFCOUNTED;
7209 /* re_dup will OpREFCNT_inc */
7210 OpREFCNT_set(sop, 1);
7213 n = add_data(pRExC_state, 3, "nop");
7214 RExC_rxi->data->data[n] = (void*)rop;
7215 RExC_rxi->data->data[n+1] = (void*)sop;
7216 RExC_rxi->data->data[n+2] = (void*)pad;
7219 else { /* First pass */
7220 if (PL_reginterp_cnt < ++RExC_seen_evals
7222 /* No compiled RE interpolated, has runtime
7223 components ===> unsafe. */
7224 FAIL("Eval-group not allowed at runtime, use re 'eval'");
7225 if (PL_tainting && PL_tainted)
7226 FAIL("Eval-group in insecure regular expression");
7227 #if PERL_VERSION > 8
7228 if (IN_PERL_COMPILETIME)
7233 nextchar(pRExC_state);
7235 ret = reg_node(pRExC_state, LOGICAL);
7238 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
7239 /* deal with the length of this later - MJD */
7242 ret = reganode(pRExC_state, EVAL, n);
7243 Set_Node_Length(ret, RExC_parse - parse_start + 1);
7244 Set_Node_Offset(ret, parse_start);
7247 case '(': /* (?(?{...})...) and (?(?=...)...) */
7250 if (RExC_parse[0] == '?') { /* (?(?...)) */
7251 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
7252 || RExC_parse[1] == '<'
7253 || RExC_parse[1] == '{') { /* Lookahead or eval. */
7256 ret = reg_node(pRExC_state, LOGICAL);
7259 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
7263 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
7264 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
7266 char ch = RExC_parse[0] == '<' ? '>' : '\'';
7267 char *name_start= RExC_parse++;
7269 SV *sv_dat=reg_scan_name(pRExC_state,
7270 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7271 if (RExC_parse == name_start || *RExC_parse != ch)
7272 vFAIL2("Sequence (?(%c... not terminated",
7273 (ch == '>' ? '<' : ch));
7276 num = add_data( pRExC_state, 1, "S" );
7277 RExC_rxi->data->data[num]=(void*)sv_dat;
7278 SvREFCNT_inc_simple_void(sv_dat);
7280 ret = reganode(pRExC_state,NGROUPP,num);
7281 goto insert_if_check_paren;
7283 else if (RExC_parse[0] == 'D' &&
7284 RExC_parse[1] == 'E' &&
7285 RExC_parse[2] == 'F' &&
7286 RExC_parse[3] == 'I' &&
7287 RExC_parse[4] == 'N' &&
7288 RExC_parse[5] == 'E')
7290 ret = reganode(pRExC_state,DEFINEP,0);
7293 goto insert_if_check_paren;
7295 else if (RExC_parse[0] == 'R') {
7298 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
7299 parno = atoi(RExC_parse++);
7300 while (isDIGIT(*RExC_parse))
7302 } else if (RExC_parse[0] == '&') {
7305 sv_dat = reg_scan_name(pRExC_state,
7306 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7307 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
7309 ret = reganode(pRExC_state,INSUBP,parno);
7310 goto insert_if_check_paren;
7312 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
7315 parno = atoi(RExC_parse++);
7317 while (isDIGIT(*RExC_parse))
7319 ret = reganode(pRExC_state, GROUPP, parno);
7321 insert_if_check_paren:
7322 if ((c = *nextchar(pRExC_state)) != ')')
7323 vFAIL("Switch condition not recognized");
7325 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
7326 br = regbranch(pRExC_state, &flags, 1,depth+1);
7328 br = reganode(pRExC_state, LONGJMP, 0);
7330 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
7331 c = *nextchar(pRExC_state);
7336 vFAIL("(?(DEFINE)....) does not allow branches");
7337 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
7338 regbranch(pRExC_state, &flags, 1,depth+1);
7339 REGTAIL(pRExC_state, ret, lastbr);
7342 c = *nextchar(pRExC_state);
7347 vFAIL("Switch (?(condition)... contains too many branches");
7348 ender = reg_node(pRExC_state, TAIL);
7349 REGTAIL(pRExC_state, br, ender);
7351 REGTAIL(pRExC_state, lastbr, ender);
7352 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
7355 REGTAIL(pRExC_state, ret, ender);
7356 RExC_size++; /* XXX WHY do we need this?!!
7357 For large programs it seems to be required
7358 but I can't figure out why. -- dmq*/
7362 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
7366 RExC_parse--; /* for vFAIL to print correctly */
7367 vFAIL("Sequence (? incomplete");
7369 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
7371 has_use_defaults = TRUE;
7372 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
7373 set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics)
7374 ? REGEX_UNICODE_CHARSET
7375 : REGEX_DEPENDS_CHARSET);
7379 parse_flags: /* (?i) */
7381 U32 posflags = 0, negflags = 0;
7382 U32 *flagsp = &posflags;
7383 char has_charset_modifier = '\0';
7384 regex_charset cs = (RExC_utf8 || RExC_uni_semantics)
7385 ? REGEX_UNICODE_CHARSET
7386 : REGEX_DEPENDS_CHARSET;
7388 while (*RExC_parse) {
7389 /* && strchr("iogcmsx", *RExC_parse) */
7390 /* (?g), (?gc) and (?o) are useless here
7391 and must be globally applied -- japhy */
7392 switch (*RExC_parse) {
7393 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
7394 case LOCALE_PAT_MOD:
7395 if (has_charset_modifier) {
7396 goto excess_modifier;
7398 else if (flagsp == &negflags) {
7401 cs = REGEX_LOCALE_CHARSET;
7402 has_charset_modifier = LOCALE_PAT_MOD;
7403 RExC_contains_locale = 1;
7405 case UNICODE_PAT_MOD:
7406 if (has_charset_modifier) {
7407 goto excess_modifier;
7409 else if (flagsp == &negflags) {
7412 cs = REGEX_UNICODE_CHARSET;
7413 has_charset_modifier = UNICODE_PAT_MOD;
7415 case ASCII_RESTRICT_PAT_MOD:
7416 if (flagsp == &negflags) {
7419 if (has_charset_modifier) {
7420 if (cs != REGEX_ASCII_RESTRICTED_CHARSET) {
7421 goto excess_modifier;
7423 /* Doubled modifier implies more restricted */
7424 cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET;
7427 cs = REGEX_ASCII_RESTRICTED_CHARSET;
7429 has_charset_modifier = ASCII_RESTRICT_PAT_MOD;
7431 case DEPENDS_PAT_MOD:
7432 if (has_use_defaults) {
7433 goto fail_modifiers;
7435 else if (flagsp == &negflags) {
7438 else if (has_charset_modifier) {
7439 goto excess_modifier;
7442 /* The dual charset means unicode semantics if the
7443 * pattern (or target, not known until runtime) are
7444 * utf8, or something in the pattern indicates unicode
7446 cs = (RExC_utf8 || RExC_uni_semantics)
7447 ? REGEX_UNICODE_CHARSET
7448 : REGEX_DEPENDS_CHARSET;
7449 has_charset_modifier = DEPENDS_PAT_MOD;
7453 if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) {
7454 vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD);
7456 else if (has_charset_modifier == *(RExC_parse - 1)) {
7457 vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1));
7460 vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1));
7465 vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1));
7467 case ONCE_PAT_MOD: /* 'o' */
7468 case GLOBAL_PAT_MOD: /* 'g' */
7469 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7470 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
7471 if (! (wastedflags & wflagbit) ) {
7472 wastedflags |= wflagbit;
7475 "Useless (%s%c) - %suse /%c modifier",
7476 flagsp == &negflags ? "?-" : "?",
7478 flagsp == &negflags ? "don't " : "",
7485 case CONTINUE_PAT_MOD: /* 'c' */
7486 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7487 if (! (wastedflags & WASTED_C) ) {
7488 wastedflags |= WASTED_GC;
7491 "Useless (%sc) - %suse /gc modifier",
7492 flagsp == &negflags ? "?-" : "?",
7493 flagsp == &negflags ? "don't " : ""
7498 case KEEPCOPY_PAT_MOD: /* 'p' */
7499 if (flagsp == &negflags) {
7501 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
7503 *flagsp |= RXf_PMf_KEEPCOPY;
7507 /* A flag is a default iff it is following a minus, so
7508 * if there is a minus, it means will be trying to
7509 * re-specify a default which is an error */
7510 if (has_use_defaults || flagsp == &negflags) {
7513 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7517 wastedflags = 0; /* reset so (?g-c) warns twice */
7523 RExC_flags |= posflags;
7524 RExC_flags &= ~negflags;
7525 set_regex_charset(&RExC_flags, cs);
7527 oregflags |= posflags;
7528 oregflags &= ~negflags;
7529 set_regex_charset(&oregflags, cs);
7531 nextchar(pRExC_state);
7542 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7547 }} /* one for the default block, one for the switch */
7554 ret = reganode(pRExC_state, OPEN, parno);
7557 RExC_nestroot = parno;
7558 if (RExC_seen & REG_SEEN_RECURSE
7559 && !RExC_open_parens[parno-1])
7561 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7562 "Setting open paren #%"IVdf" to %d\n",
7563 (IV)parno, REG_NODE_NUM(ret)));
7564 RExC_open_parens[parno-1]= ret;
7567 Set_Node_Length(ret, 1); /* MJD */
7568 Set_Node_Offset(ret, RExC_parse); /* MJD */
7576 /* Pick up the branches, linking them together. */
7577 parse_start = RExC_parse; /* MJD */
7578 br = regbranch(pRExC_state, &flags, 1,depth+1);
7580 /* branch_len = (paren != 0); */
7584 if (*RExC_parse == '|') {
7585 if (!SIZE_ONLY && RExC_extralen) {
7586 reginsert(pRExC_state, BRANCHJ, br, depth+1);
7589 reginsert(pRExC_state, BRANCH, br, depth+1);
7590 Set_Node_Length(br, paren != 0);
7591 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
7595 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
7597 else if (paren == ':') {
7598 *flagp |= flags&SIMPLE;
7600 if (is_open) { /* Starts with OPEN. */
7601 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
7603 else if (paren != '?') /* Not Conditional */
7605 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7607 while (*RExC_parse == '|') {
7608 if (!SIZE_ONLY && RExC_extralen) {
7609 ender = reganode(pRExC_state, LONGJMP,0);
7610 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
7613 RExC_extralen += 2; /* Account for LONGJMP. */
7614 nextchar(pRExC_state);
7616 if (RExC_npar > after_freeze)
7617 after_freeze = RExC_npar;
7618 RExC_npar = freeze_paren;
7620 br = regbranch(pRExC_state, &flags, 0, depth+1);
7624 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
7626 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7629 if (have_branch || paren != ':') {
7630 /* Make a closing node, and hook it on the end. */
7633 ender = reg_node(pRExC_state, TAIL);
7636 ender = reganode(pRExC_state, CLOSE, parno);
7637 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
7638 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7639 "Setting close paren #%"IVdf" to %d\n",
7640 (IV)parno, REG_NODE_NUM(ender)));
7641 RExC_close_parens[parno-1]= ender;
7642 if (RExC_nestroot == parno)
7645 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
7646 Set_Node_Length(ender,1); /* MJD */
7652 *flagp &= ~HASWIDTH;
7655 ender = reg_node(pRExC_state, SUCCEED);
7658 ender = reg_node(pRExC_state, END);
7660 assert(!RExC_opend); /* there can only be one! */
7665 REGTAIL(pRExC_state, lastbr, ender);
7667 if (have_branch && !SIZE_ONLY) {
7669 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
7671 /* Hook the tails of the branches to the closing node. */
7672 for (br = ret; br; br = regnext(br)) {
7673 const U8 op = PL_regkind[OP(br)];
7675 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
7677 else if (op == BRANCHJ) {
7678 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
7686 static const char parens[] = "=!<,>";
7688 if (paren && (p = strchr(parens, paren))) {
7689 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
7690 int flag = (p - parens) > 1;
7693 node = SUSPEND, flag = 0;
7694 reginsert(pRExC_state, node,ret, depth+1);
7695 Set_Node_Cur_Length(ret);
7696 Set_Node_Offset(ret, parse_start + 1);
7698 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
7702 /* Check for proper termination. */
7704 RExC_flags = oregflags;
7705 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
7706 RExC_parse = oregcomp_parse;
7707 vFAIL("Unmatched (");
7710 else if (!paren && RExC_parse < RExC_end) {
7711 if (*RExC_parse == ')') {
7713 vFAIL("Unmatched )");
7716 FAIL("Junk on end of regexp"); /* "Can't happen". */
7720 if (RExC_in_lookbehind) {
7721 RExC_in_lookbehind--;
7723 if (after_freeze > RExC_npar)
7724 RExC_npar = after_freeze;
7729 - regbranch - one alternative of an | operator
7731 * Implements the concatenation operator.
7734 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
7737 register regnode *ret;
7738 register regnode *chain = NULL;
7739 register regnode *latest;
7740 I32 flags = 0, c = 0;
7741 GET_RE_DEBUG_FLAGS_DECL;
7743 PERL_ARGS_ASSERT_REGBRANCH;
7745 DEBUG_PARSE("brnc");
7750 if (!SIZE_ONLY && RExC_extralen)
7751 ret = reganode(pRExC_state, BRANCHJ,0);
7753 ret = reg_node(pRExC_state, BRANCH);
7754 Set_Node_Length(ret, 1);
7758 if (!first && SIZE_ONLY)
7759 RExC_extralen += 1; /* BRANCHJ */
7761 *flagp = WORST; /* Tentatively. */
7764 nextchar(pRExC_state);
7765 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
7767 latest = regpiece(pRExC_state, &flags,depth+1);
7768 if (latest == NULL) {
7769 if (flags & TRYAGAIN)
7773 else if (ret == NULL)
7775 *flagp |= flags&(HASWIDTH|POSTPONED);
7776 if (chain == NULL) /* First piece. */
7777 *flagp |= flags&SPSTART;
7780 REGTAIL(pRExC_state, chain, latest);
7785 if (chain == NULL) { /* Loop ran zero times. */
7786 chain = reg_node(pRExC_state, NOTHING);
7791 *flagp |= flags&SIMPLE;
7798 - regpiece - something followed by possible [*+?]
7800 * Note that the branching code sequences used for ? and the general cases
7801 * of * and + are somewhat optimized: they use the same NOTHING node as
7802 * both the endmarker for their branch list and the body of the last branch.
7803 * It might seem that this node could be dispensed with entirely, but the
7804 * endmarker role is not redundant.
7807 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7810 register regnode *ret;
7812 register char *next;
7814 const char * const origparse = RExC_parse;
7816 I32 max = REG_INFTY;
7817 #ifdef RE_TRACK_PATTERN_OFFSETS
7820 const char *maxpos = NULL;
7821 GET_RE_DEBUG_FLAGS_DECL;
7823 PERL_ARGS_ASSERT_REGPIECE;
7825 DEBUG_PARSE("piec");
7827 ret = regatom(pRExC_state, &flags,depth+1);
7829 if (flags & TRYAGAIN)
7836 if (op == '{' && regcurly(RExC_parse)) {
7838 #ifdef RE_TRACK_PATTERN_OFFSETS
7839 parse_start = RExC_parse; /* MJD */
7841 next = RExC_parse + 1;
7842 while (isDIGIT(*next) || *next == ',') {
7851 if (*next == '}') { /* got one */
7855 min = atoi(RExC_parse);
7859 maxpos = RExC_parse;
7861 if (!max && *maxpos != '0')
7862 max = REG_INFTY; /* meaning "infinity" */
7863 else if (max >= REG_INFTY)
7864 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
7866 nextchar(pRExC_state);
7869 if ((flags&SIMPLE)) {
7870 RExC_naughty += 2 + RExC_naughty / 2;
7871 reginsert(pRExC_state, CURLY, ret, depth+1);
7872 Set_Node_Offset(ret, parse_start+1); /* MJD */
7873 Set_Node_Cur_Length(ret);
7876 regnode * const w = reg_node(pRExC_state, WHILEM);
7879 REGTAIL(pRExC_state, ret, w);
7880 if (!SIZE_ONLY && RExC_extralen) {
7881 reginsert(pRExC_state, LONGJMP,ret, depth+1);
7882 reginsert(pRExC_state, NOTHING,ret, depth+1);
7883 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
7885 reginsert(pRExC_state, CURLYX,ret, depth+1);
7887 Set_Node_Offset(ret, parse_start+1);
7888 Set_Node_Length(ret,
7889 op == '{' ? (RExC_parse - parse_start) : 1);
7891 if (!SIZE_ONLY && RExC_extralen)
7892 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
7893 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
7895 RExC_whilem_seen++, RExC_extralen += 3;
7896 RExC_naughty += 4 + RExC_naughty; /* compound interest */
7905 vFAIL("Can't do {n,m} with n > m");
7907 ARG1_SET(ret, (U16)min);
7908 ARG2_SET(ret, (U16)max);
7920 #if 0 /* Now runtime fix should be reliable. */
7922 /* if this is reinstated, don't forget to put this back into perldiag:
7924 =item Regexp *+ operand could be empty at {#} in regex m/%s/
7926 (F) The part of the regexp subject to either the * or + quantifier
7927 could match an empty string. The {#} shows in the regular
7928 expression about where the problem was discovered.
7932 if (!(flags&HASWIDTH) && op != '?')
7933 vFAIL("Regexp *+ operand could be empty");
7936 #ifdef RE_TRACK_PATTERN_OFFSETS
7937 parse_start = RExC_parse;
7939 nextchar(pRExC_state);
7941 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
7943 if (op == '*' && (flags&SIMPLE)) {
7944 reginsert(pRExC_state, STAR, ret, depth+1);
7948 else if (op == '*') {
7952 else if (op == '+' && (flags&SIMPLE)) {
7953 reginsert(pRExC_state, PLUS, ret, depth+1);
7957 else if (op == '+') {
7961 else if (op == '?') {
7966 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
7967 ckWARN3reg(RExC_parse,
7968 "%.*s matches null string many times",
7969 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
7973 if (RExC_parse < RExC_end && *RExC_parse == '?') {
7974 nextchar(pRExC_state);
7975 reginsert(pRExC_state, MINMOD, ret, depth+1);
7976 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
7978 #ifndef REG_ALLOW_MINMOD_SUSPEND
7981 if (RExC_parse < RExC_end && *RExC_parse == '+') {
7983 nextchar(pRExC_state);
7984 ender = reg_node(pRExC_state, SUCCEED);
7985 REGTAIL(pRExC_state, ret, ender);
7986 reginsert(pRExC_state, SUSPEND, ret, depth+1);
7988 ender = reg_node(pRExC_state, TAIL);
7989 REGTAIL(pRExC_state, ret, ender);
7993 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
7995 vFAIL("Nested quantifiers");
8002 /* reg_namedseq(pRExC_state,UVp, UV depth)
8004 This is expected to be called by a parser routine that has
8005 recognized '\N' and needs to handle the rest. RExC_parse is
8006 expected to point at the first char following the N at the time
8009 The \N may be inside (indicated by valuep not being NULL) or outside a
8012 \N may begin either a named sequence, or if outside a character class, mean
8013 to match a non-newline. For non single-quoted regexes, the tokenizer has
8014 attempted to decide which, and in the case of a named sequence converted it
8015 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
8016 where c1... are the characters in the sequence. For single-quoted regexes,
8017 the tokenizer passes the \N sequence through unchanged; this code will not
8018 attempt to determine this nor expand those. The net effect is that if the
8019 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
8020 signals that this \N occurrence means to match a non-newline.
8022 Only the \N{U+...} form should occur in a character class, for the same
8023 reason that '.' inside a character class means to just match a period: it
8024 just doesn't make sense.
8026 If valuep is non-null then it is assumed that we are parsing inside
8027 of a charclass definition and the first codepoint in the resolved
8028 string is returned via *valuep and the routine will return NULL.
8029 In this mode if a multichar string is returned from the charnames
8030 handler, a warning will be issued, and only the first char in the
8031 sequence will be examined. If the string returned is zero length
8032 then the value of *valuep is undefined and NON-NULL will
8033 be returned to indicate failure. (This will NOT be a valid pointer
8036 If valuep is null then it is assumed that we are parsing normal text and a
8037 new EXACT node is inserted into the program containing the resolved string,
8038 and a pointer to the new node is returned. But if the string is zero length
8039 a NOTHING node is emitted instead.
8041 On success RExC_parse is set to the char following the endbrace.
8042 Parsing failures will generate a fatal error via vFAIL(...)
8045 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp, U32 depth)
8047 char * endbrace; /* '}' following the name */
8048 regnode *ret = NULL;
8051 GET_RE_DEBUG_FLAGS_DECL;
8053 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
8057 /* The [^\n] meaning of \N ignores spaces and comments under the /x
8058 * modifier. The other meaning does not */
8059 p = (RExC_flags & RXf_PMf_EXTENDED)
8060 ? regwhite( pRExC_state, RExC_parse )
8063 /* Disambiguate between \N meaning a named character versus \N meaning
8064 * [^\n]. The former is assumed when it can't be the latter. */
8065 if (*p != '{' || regcurly(p)) {
8068 /* no bare \N in a charclass */
8069 vFAIL("\\N in a character class must be a named character: \\N{...}");
8071 nextchar(pRExC_state);
8072 ret = reg_node(pRExC_state, REG_ANY);
8073 *flagp |= HASWIDTH|SIMPLE;
8076 Set_Node_Length(ret, 1); /* MJD */
8080 /* Here, we have decided it should be a named sequence */
8082 /* The test above made sure that the next real character is a '{', but
8083 * under the /x modifier, it could be separated by space (or a comment and
8084 * \n) and this is not allowed (for consistency with \x{...} and the
8085 * tokenizer handling of \N{NAME}). */
8086 if (*RExC_parse != '{') {
8087 vFAIL("Missing braces on \\N{}");
8090 RExC_parse++; /* Skip past the '{' */
8092 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
8093 || ! (endbrace == RExC_parse /* nothing between the {} */
8094 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
8095 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
8097 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
8098 vFAIL("\\N{NAME} must be resolved by the lexer");
8101 if (endbrace == RExC_parse) { /* empty: \N{} */
8103 RExC_parse = endbrace + 1;
8104 return reg_node(pRExC_state,NOTHING);
8108 ckWARNreg(RExC_parse,
8109 "Ignoring zero length \\N{} in character class"
8111 RExC_parse = endbrace + 1;
8114 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
8117 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
8118 RExC_parse += 2; /* Skip past the 'U+' */
8120 if (valuep) { /* In a bracketed char class */
8121 /* We only pay attention to the first char of
8122 multichar strings being returned. I kinda wonder
8123 if this makes sense as it does change the behaviour
8124 from earlier versions, OTOH that behaviour was broken
8125 as well. XXX Solution is to recharacterize as
8126 [rest-of-class]|multi1|multi2... */
8128 STRLEN length_of_hex;
8129 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8130 | PERL_SCAN_DISALLOW_PREFIX
8131 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
8133 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
8134 if (endchar < endbrace) {
8135 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
8138 length_of_hex = (STRLEN)(endchar - RExC_parse);
8139 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
8141 /* The tokenizer should have guaranteed validity, but it's possible to
8142 * bypass it by using single quoting, so check */
8143 if (length_of_hex == 0
8144 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
8146 RExC_parse += length_of_hex; /* Includes all the valid */
8147 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
8148 ? UTF8SKIP(RExC_parse)
8150 /* Guard against malformed utf8 */
8151 if (RExC_parse >= endchar) RExC_parse = endchar;
8152 vFAIL("Invalid hexadecimal number in \\N{U+...}");
8155 RExC_parse = endbrace + 1;
8156 if (endchar == endbrace) return NULL;
8158 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
8160 else { /* Not a char class */
8162 /* What is done here is to convert this to a sub-pattern of the form
8163 * (?:\x{char1}\x{char2}...)
8164 * and then call reg recursively. That way, it retains its atomicness,
8165 * while not having to worry about special handling that some code
8166 * points may have. toke.c has converted the original Unicode values
8167 * to native, so that we can just pass on the hex values unchanged. We
8168 * do have to set a flag to keep recoding from happening in the
8171 SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP);
8173 char *endchar; /* Points to '.' or '}' ending cur char in the input
8175 char *orig_end = RExC_end;
8177 while (RExC_parse < endbrace) {
8179 /* Code points are separated by dots. If none, there is only one
8180 * code point, and is terminated by the brace */
8181 endchar = RExC_parse + strcspn(RExC_parse, ".}");
8183 /* Convert to notation the rest of the code understands */
8184 sv_catpv(substitute_parse, "\\x{");
8185 sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse);
8186 sv_catpv(substitute_parse, "}");
8188 /* Point to the beginning of the next character in the sequence. */
8189 RExC_parse = endchar + 1;
8191 sv_catpv(substitute_parse, ")");
8193 RExC_parse = SvPV(substitute_parse, len);
8195 /* Don't allow empty number */
8197 vFAIL("Invalid hexadecimal number in \\N{U+...}");
8199 RExC_end = RExC_parse + len;
8201 /* The values are Unicode, and therefore not subject to recoding */
8202 RExC_override_recoding = 1;
8204 ret = reg(pRExC_state, 1, flagp, depth+1);
8206 RExC_parse = endbrace;
8207 RExC_end = orig_end;
8208 RExC_override_recoding = 0;
8210 nextchar(pRExC_state);
8220 * It returns the code point in utf8 for the value in *encp.
8221 * value: a code value in the source encoding
8222 * encp: a pointer to an Encode object
8224 * If the result from Encode is not a single character,
8225 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
8228 S_reg_recode(pTHX_ const char value, SV **encp)
8231 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
8232 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
8233 const STRLEN newlen = SvCUR(sv);
8234 UV uv = UNICODE_REPLACEMENT;
8236 PERL_ARGS_ASSERT_REG_RECODE;
8240 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
8243 if (!newlen || numlen != newlen) {
8244 uv = UNICODE_REPLACEMENT;
8252 - regatom - the lowest level
8254 Try to identify anything special at the start of the pattern. If there
8255 is, then handle it as required. This may involve generating a single regop,
8256 such as for an assertion; or it may involve recursing, such as to
8257 handle a () structure.
8259 If the string doesn't start with something special then we gobble up
8260 as much literal text as we can.
8262 Once we have been able to handle whatever type of thing started the
8263 sequence, we return.
8265 Note: we have to be careful with escapes, as they can be both literal
8266 and special, and in the case of \10 and friends can either, depending
8267 on context. Specifically there are two separate switches for handling
8268 escape sequences, with the one for handling literal escapes requiring
8269 a dummy entry for all of the special escapes that are actually handled
8274 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
8277 register regnode *ret = NULL;
8279 char *parse_start = RExC_parse;
8281 GET_RE_DEBUG_FLAGS_DECL;
8282 DEBUG_PARSE("atom");
8283 *flagp = WORST; /* Tentatively. */
8285 PERL_ARGS_ASSERT_REGATOM;
8288 switch ((U8)*RExC_parse) {
8290 RExC_seen_zerolen++;
8291 nextchar(pRExC_state);
8292 if (RExC_flags & RXf_PMf_MULTILINE)
8293 ret = reg_node(pRExC_state, MBOL);
8294 else if (RExC_flags & RXf_PMf_SINGLELINE)
8295 ret = reg_node(pRExC_state, SBOL);
8297 ret = reg_node(pRExC_state, BOL);
8298 Set_Node_Length(ret, 1); /* MJD */
8301 nextchar(pRExC_state);
8303 RExC_seen_zerolen++;
8304 if (RExC_flags & RXf_PMf_MULTILINE)
8305 ret = reg_node(pRExC_state, MEOL);
8306 else if (RExC_flags & RXf_PMf_SINGLELINE)
8307 ret = reg_node(pRExC_state, SEOL);
8309 ret = reg_node(pRExC_state, EOL);
8310 Set_Node_Length(ret, 1); /* MJD */
8313 nextchar(pRExC_state);
8314 if (RExC_flags & RXf_PMf_SINGLELINE)
8315 ret = reg_node(pRExC_state, SANY);
8317 ret = reg_node(pRExC_state, REG_ANY);
8318 *flagp |= HASWIDTH|SIMPLE;
8320 Set_Node_Length(ret, 1); /* MJD */
8324 char * const oregcomp_parse = ++RExC_parse;
8325 ret = regclass(pRExC_state,depth+1);
8326 if (*RExC_parse != ']') {
8327 RExC_parse = oregcomp_parse;
8328 vFAIL("Unmatched [");
8330 nextchar(pRExC_state);
8331 *flagp |= HASWIDTH|SIMPLE;
8332 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
8336 nextchar(pRExC_state);
8337 ret = reg(pRExC_state, 1, &flags,depth+1);
8339 if (flags & TRYAGAIN) {
8340 if (RExC_parse == RExC_end) {
8341 /* Make parent create an empty node if needed. */
8349 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
8353 if (flags & TRYAGAIN) {
8357 vFAIL("Internal urp");
8358 /* Supposed to be caught earlier. */
8361 if (!regcurly(RExC_parse)) {
8370 vFAIL("Quantifier follows nothing");
8375 This switch handles escape sequences that resolve to some kind
8376 of special regop and not to literal text. Escape sequnces that
8377 resolve to literal text are handled below in the switch marked
8380 Every entry in this switch *must* have a corresponding entry
8381 in the literal escape switch. However, the opposite is not
8382 required, as the default for this switch is to jump to the
8383 literal text handling code.
8385 switch ((U8)*++RExC_parse) {
8386 /* Special Escapes */
8388 RExC_seen_zerolen++;
8389 ret = reg_node(pRExC_state, SBOL);
8391 goto finish_meta_pat;
8393 ret = reg_node(pRExC_state, GPOS);
8394 RExC_seen |= REG_SEEN_GPOS;
8396 goto finish_meta_pat;
8398 RExC_seen_zerolen++;
8399 ret = reg_node(pRExC_state, KEEPS);
8401 /* XXX:dmq : disabling in-place substitution seems to
8402 * be necessary here to avoid cases of memory corruption, as
8403 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
8405 RExC_seen |= REG_SEEN_LOOKBEHIND;
8406 goto finish_meta_pat;
8408 ret = reg_node(pRExC_state, SEOL);
8410 RExC_seen_zerolen++; /* Do not optimize RE away */
8411 goto finish_meta_pat;
8413 ret = reg_node(pRExC_state, EOS);
8415 RExC_seen_zerolen++; /* Do not optimize RE away */
8416 goto finish_meta_pat;
8418 ret = reg_node(pRExC_state, CANY);
8419 RExC_seen |= REG_SEEN_CANY;
8420 *flagp |= HASWIDTH|SIMPLE;
8421 goto finish_meta_pat;
8423 ret = reg_node(pRExC_state, CLUMP);
8425 goto finish_meta_pat;
8427 switch (get_regex_charset(RExC_flags)) {
8428 case REGEX_LOCALE_CHARSET:
8431 case REGEX_UNICODE_CHARSET:
8434 case REGEX_ASCII_RESTRICTED_CHARSET:
8435 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8438 case REGEX_DEPENDS_CHARSET:
8444 ret = reg_node(pRExC_state, op);
8445 *flagp |= HASWIDTH|SIMPLE;
8446 goto finish_meta_pat;
8448 switch (get_regex_charset(RExC_flags)) {
8449 case REGEX_LOCALE_CHARSET:
8452 case REGEX_UNICODE_CHARSET:
8455 case REGEX_ASCII_RESTRICTED_CHARSET:
8456 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8459 case REGEX_DEPENDS_CHARSET:
8465 ret = reg_node(pRExC_state, op);
8466 *flagp |= HASWIDTH|SIMPLE;
8467 goto finish_meta_pat;
8469 RExC_seen_zerolen++;
8470 RExC_seen |= REG_SEEN_LOOKBEHIND;
8471 switch (get_regex_charset(RExC_flags)) {
8472 case REGEX_LOCALE_CHARSET:
8475 case REGEX_UNICODE_CHARSET:
8478 case REGEX_ASCII_RESTRICTED_CHARSET:
8479 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8482 case REGEX_DEPENDS_CHARSET:
8488 ret = reg_node(pRExC_state, op);
8489 FLAGS(ret) = get_regex_charset(RExC_flags);
8491 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8492 ckWARNregdep(RExC_parse, "\"\\b{\" is deprecated; use \"\\b\\{\" instead");
8494 goto finish_meta_pat;
8496 RExC_seen_zerolen++;
8497 RExC_seen |= REG_SEEN_LOOKBEHIND;
8498 switch (get_regex_charset(RExC_flags)) {
8499 case REGEX_LOCALE_CHARSET:
8502 case REGEX_UNICODE_CHARSET:
8505 case REGEX_ASCII_RESTRICTED_CHARSET:
8506 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8509 case REGEX_DEPENDS_CHARSET:
8515 ret = reg_node(pRExC_state, op);
8516 FLAGS(ret) = get_regex_charset(RExC_flags);
8518 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8519 ckWARNregdep(RExC_parse, "\"\\B{\" is deprecated; use \"\\B\\{\" instead");
8521 goto finish_meta_pat;
8523 switch (get_regex_charset(RExC_flags)) {
8524 case REGEX_LOCALE_CHARSET:
8527 case REGEX_UNICODE_CHARSET:
8530 case REGEX_ASCII_RESTRICTED_CHARSET:
8531 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8534 case REGEX_DEPENDS_CHARSET:
8540 ret = reg_node(pRExC_state, op);
8541 *flagp |= HASWIDTH|SIMPLE;
8542 goto finish_meta_pat;
8544 switch (get_regex_charset(RExC_flags)) {
8545 case REGEX_LOCALE_CHARSET:
8548 case REGEX_UNICODE_CHARSET:
8551 case REGEX_ASCII_RESTRICTED_CHARSET:
8552 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8555 case REGEX_DEPENDS_CHARSET:
8561 ret = reg_node(pRExC_state, op);
8562 *flagp |= HASWIDTH|SIMPLE;
8563 goto finish_meta_pat;
8565 switch (get_regex_charset(RExC_flags)) {
8566 case REGEX_LOCALE_CHARSET:
8569 case REGEX_ASCII_RESTRICTED_CHARSET:
8570 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8573 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8574 case REGEX_UNICODE_CHARSET:
8580 ret = reg_node(pRExC_state, op);
8581 *flagp |= HASWIDTH|SIMPLE;
8582 goto finish_meta_pat;
8584 switch (get_regex_charset(RExC_flags)) {
8585 case REGEX_LOCALE_CHARSET:
8588 case REGEX_ASCII_RESTRICTED_CHARSET:
8589 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8592 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8593 case REGEX_UNICODE_CHARSET:
8599 ret = reg_node(pRExC_state, op);
8600 *flagp |= HASWIDTH|SIMPLE;
8601 goto finish_meta_pat;
8603 ret = reg_node(pRExC_state, LNBREAK);
8604 *flagp |= HASWIDTH|SIMPLE;
8605 goto finish_meta_pat;
8607 ret = reg_node(pRExC_state, HORIZWS);
8608 *flagp |= HASWIDTH|SIMPLE;
8609 goto finish_meta_pat;
8611 ret = reg_node(pRExC_state, NHORIZWS);
8612 *flagp |= HASWIDTH|SIMPLE;
8613 goto finish_meta_pat;
8615 ret = reg_node(pRExC_state, VERTWS);
8616 *flagp |= HASWIDTH|SIMPLE;
8617 goto finish_meta_pat;
8619 ret = reg_node(pRExC_state, NVERTWS);
8620 *flagp |= HASWIDTH|SIMPLE;
8622 nextchar(pRExC_state);
8623 Set_Node_Length(ret, 2); /* MJD */
8628 char* const oldregxend = RExC_end;
8630 char* parse_start = RExC_parse - 2;
8633 if (RExC_parse[1] == '{') {
8634 /* a lovely hack--pretend we saw [\pX] instead */
8635 RExC_end = strchr(RExC_parse, '}');
8637 const U8 c = (U8)*RExC_parse;
8639 RExC_end = oldregxend;
8640 vFAIL2("Missing right brace on \\%c{}", c);
8645 RExC_end = RExC_parse + 2;
8646 if (RExC_end > oldregxend)
8647 RExC_end = oldregxend;
8651 ret = regclass(pRExC_state,depth+1);
8653 RExC_end = oldregxend;
8656 Set_Node_Offset(ret, parse_start + 2);
8657 Set_Node_Cur_Length(ret);
8658 nextchar(pRExC_state);
8659 *flagp |= HASWIDTH|SIMPLE;
8663 /* Handle \N and \N{NAME} here and not below because it can be
8664 multicharacter. join_exact() will join them up later on.
8665 Also this makes sure that things like /\N{BLAH}+/ and
8666 \N{BLAH} being multi char Just Happen. dmq*/
8668 ret= reg_namedseq(pRExC_state, NULL, flagp, depth);
8670 case 'k': /* Handle \k<NAME> and \k'NAME' */
8673 char ch= RExC_parse[1];
8674 if (ch != '<' && ch != '\'' && ch != '{') {
8676 vFAIL2("Sequence %.2s... not terminated",parse_start);
8678 /* this pretty much dupes the code for (?P=...) in reg(), if
8679 you change this make sure you change that */
8680 char* name_start = (RExC_parse += 2);
8682 SV *sv_dat = reg_scan_name(pRExC_state,
8683 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
8684 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
8685 if (RExC_parse == name_start || *RExC_parse != ch)
8686 vFAIL2("Sequence %.3s... not terminated",parse_start);
8689 num = add_data( pRExC_state, 1, "S" );
8690 RExC_rxi->data->data[num]=(void*)sv_dat;
8691 SvREFCNT_inc_simple_void(sv_dat);
8695 ret = reganode(pRExC_state,
8698 : (MORE_ASCII_RESTRICTED)
8700 : (AT_LEAST_UNI_SEMANTICS)
8708 /* override incorrect value set in reganode MJD */
8709 Set_Node_Offset(ret, parse_start+1);
8710 Set_Node_Cur_Length(ret); /* MJD */
8711 nextchar(pRExC_state);
8717 case '1': case '2': case '3': case '4':
8718 case '5': case '6': case '7': case '8': case '9':
8721 bool isg = *RExC_parse == 'g';
8726 if (*RExC_parse == '{') {
8730 if (*RExC_parse == '-') {
8734 if (hasbrace && !isDIGIT(*RExC_parse)) {
8735 if (isrel) RExC_parse--;
8737 goto parse_named_seq;
8739 num = atoi(RExC_parse);
8740 if (isg && num == 0)
8741 vFAIL("Reference to invalid group 0");
8743 num = RExC_npar - num;
8745 vFAIL("Reference to nonexistent or unclosed group");
8747 if (!isg && num > 9 && num >= RExC_npar)
8750 char * const parse_start = RExC_parse - 1; /* MJD */
8751 while (isDIGIT(*RExC_parse))
8753 if (parse_start == RExC_parse - 1)
8754 vFAIL("Unterminated \\g... pattern");
8756 if (*RExC_parse != '}')
8757 vFAIL("Unterminated \\g{...} pattern");
8761 if (num > (I32)RExC_rx->nparens)
8762 vFAIL("Reference to nonexistent group");
8765 ret = reganode(pRExC_state,
8768 : (MORE_ASCII_RESTRICTED)
8770 : (AT_LEAST_UNI_SEMANTICS)
8778 /* override incorrect value set in reganode MJD */
8779 Set_Node_Offset(ret, parse_start+1);
8780 Set_Node_Cur_Length(ret); /* MJD */
8782 nextchar(pRExC_state);
8787 if (RExC_parse >= RExC_end)
8788 FAIL("Trailing \\");
8791 /* Do not generate "unrecognized" warnings here, we fall
8792 back into the quick-grab loop below */
8799 if (RExC_flags & RXf_PMf_EXTENDED) {
8800 if ( reg_skipcomment( pRExC_state ) )
8807 parse_start = RExC_parse - 1;
8820 char_state latest_char_state = generic_char;
8821 register STRLEN len;
8826 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
8827 regnode * orig_emit;
8830 orig_emit = RExC_emit; /* Save the original output node position in
8831 case we need to output a different node
8833 ret = reg_node(pRExC_state,
8834 (U8) ((! FOLD) ? EXACT
8837 : (MORE_ASCII_RESTRICTED)
8839 : (AT_LEAST_UNI_SEMANTICS)
8844 for (len = 0, p = RExC_parse - 1;
8845 len < 127 && p < RExC_end;
8848 char * const oldp = p;
8850 if (RExC_flags & RXf_PMf_EXTENDED)
8851 p = regwhite( pRExC_state, p );
8862 /* Literal Escapes Switch
8864 This switch is meant to handle escape sequences that
8865 resolve to a literal character.
8867 Every escape sequence that represents something
8868 else, like an assertion or a char class, is handled
8869 in the switch marked 'Special Escapes' above in this
8870 routine, but also has an entry here as anything that
8871 isn't explicitly mentioned here will be treated as
8872 an unescaped equivalent literal.
8876 /* These are all the special escapes. */
8877 case 'A': /* Start assertion */
8878 case 'b': case 'B': /* Word-boundary assertion*/
8879 case 'C': /* Single char !DANGEROUS! */
8880 case 'd': case 'D': /* digit class */
8881 case 'g': case 'G': /* generic-backref, pos assertion */
8882 case 'h': case 'H': /* HORIZWS */
8883 case 'k': case 'K': /* named backref, keep marker */
8884 case 'N': /* named char sequence */
8885 case 'p': case 'P': /* Unicode property */
8886 case 'R': /* LNBREAK */
8887 case 's': case 'S': /* space class */
8888 case 'v': case 'V': /* VERTWS */
8889 case 'w': case 'W': /* word class */
8890 case 'X': /* eXtended Unicode "combining character sequence" */
8891 case 'z': case 'Z': /* End of line/string assertion */
8895 /* Anything after here is an escape that resolves to a
8896 literal. (Except digits, which may or may not)
8915 ender = ASCII_TO_NATIVE('\033');
8919 ender = ASCII_TO_NATIVE('\007');
8924 STRLEN brace_len = len;
8926 const char* error_msg;
8928 bool valid = grok_bslash_o(p,
8935 RExC_parse = p; /* going to die anyway; point
8936 to exact spot of failure */
8943 if (PL_encoding && ender < 0x100) {
8944 goto recode_encoding;
8953 char* const e = strchr(p, '}');
8957 vFAIL("Missing right brace on \\x{}");
8960 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8961 | PERL_SCAN_DISALLOW_PREFIX;
8962 STRLEN numlen = e - p - 1;
8963 ender = grok_hex(p + 1, &numlen, &flags, NULL);
8970 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8972 ender = grok_hex(p, &numlen, &flags, NULL);
8975 if (PL_encoding && ender < 0x100)
8976 goto recode_encoding;
8980 ender = grok_bslash_c(*p++, UTF, SIZE_ONLY);
8982 case '0': case '1': case '2': case '3':case '4':
8983 case '5': case '6': case '7': case '8':case '9':
8985 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
8987 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8989 ender = grok_oct(p, &numlen, &flags, NULL);
8999 if (PL_encoding && ender < 0x100)
9000 goto recode_encoding;
9003 if (! RExC_override_recoding) {
9004 SV* enc = PL_encoding;
9005 ender = reg_recode((const char)(U8)ender, &enc);
9006 if (!enc && SIZE_ONLY)
9007 ckWARNreg(p, "Invalid escape in the specified encoding");
9013 FAIL("Trailing \\");
9016 if (!SIZE_ONLY&& isALPHA(*p)) {
9017 /* Include any { following the alpha to emphasize
9018 * that it could be part of an escape at some point
9020 int len = (*(p + 1) == '{') ? 2 : 1;
9021 ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p);
9023 goto normal_default;
9028 if (UTF8_IS_START(*p) && UTF) {
9030 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
9031 &numlen, UTF8_ALLOW_DEFAULT);
9037 } /* End of switch on the literal */
9039 /* Certain characters are problematic because their folded
9040 * length is so different from their original length that it
9041 * isn't handleable by the optimizer. They are therefore not
9042 * placed in an EXACTish node; and are here handled specially.
9043 * (Even if the optimizer handled LATIN_SMALL_LETTER_SHARP_S,
9044 * putting it in a special node keeps regexec from having to
9045 * deal with a non-utf8 multi-char fold */
9047 && (ender > 255 || (! MORE_ASCII_RESTRICTED && ! LOC)))
9049 /* We look for either side of the fold. For example \xDF
9050 * folds to 'ss'. We look for both the single character
9051 * \xDF and the sequence 'ss'. When we find something that
9052 * could be one of those, we stop and flush whatever we
9053 * have output so far into the EXACTish node that was being
9054 * built. Then restore the input pointer to what it was.
9055 * regatom will return that EXACT node, and will be called
9056 * again, positioned so the first character is the one in
9057 * question, which we return in a different node type.
9058 * The multi-char folds are a sequence, so the occurrence
9059 * of the first character in that sequence doesn't
9060 * necessarily mean that what follows is the rest of the
9061 * sequence. We keep track of that with a state machine,
9062 * with the state being set to the latest character
9063 * processed before the current one. Most characters will
9064 * set the state to 0, but if one occurs that is part of a
9065 * potential tricky fold sequence, the state is set to that
9066 * character, and the next loop iteration sees if the state
9067 * should progress towards the final folded-from character,
9068 * or if it was a false alarm. If it turns out to be a
9069 * false alarm, the character(s) will be output in a new
9070 * EXACTish node, and join_exact() will later combine them.
9071 * In the case of the 'ss' sequence, which is more common
9072 * and more easily checked, some look-ahead is done to
9073 * save time by ruling-out some false alarms */
9076 latest_char_state = generic_char;
9080 case 0x17F: /* LATIN SMALL LETTER LONG S */
9081 if (AT_LEAST_UNI_SEMANTICS) {
9082 if (latest_char_state == char_s) { /* 'ss' */
9083 ender = LATIN_SMALL_LETTER_SHARP_S;
9086 else if (p < RExC_end) {
9088 /* Look-ahead at the next character. If it
9089 * is also an s, we handle as a sharp s
9090 * tricky regnode. */
9091 if (*p == 's' || *p == 'S') {
9093 /* But first flush anything in the
9094 * EXACTish buffer */
9099 p++; /* Account for swallowing this
9101 ender = LATIN_SMALL_LETTER_SHARP_S;
9104 /* Here, the next character is not a
9105 * literal 's', but still could
9106 * evaluate to one if part of a \o{},
9107 * \x or \OCTAL-DIGIT. The minimum
9108 * length required for that is 4, eg
9112 && (isDIGIT(*(p + 1))
9114 || *(p + 1) == 'o' ))
9117 /* Here, it could be an 's', too much
9118 * bother to figure it out here. Flush
9119 * the buffer if any; when come back
9120 * here, set the state so know that the
9121 * previous char was an 's' */
9123 latest_char_state = generic_char;
9127 latest_char_state = char_s;
9133 /* Here, can't be an 'ss' sequence, or at least not
9134 * one that could fold to/from the sharp ss */
9135 latest_char_state = generic_char;
9137 case 0x03C5: /* First char in upsilon series */
9138 case 0x03A5: /* Also capital UPSILON, which folds to
9139 03C5, and hence exhibits the same
9141 if (p < RExC_end - 4) { /* Need >= 4 bytes left */
9142 latest_char_state = upsilon_1;
9149 latest_char_state = generic_char;
9152 case 0x03B9: /* First char in iota series */
9153 case 0x0399: /* Also capital IOTA */
9154 case 0x1FBE: /* GREEK PROSGEGRAMMENI folds to 3B9 */
9155 case 0x0345: /* COMBINING GREEK YPOGEGRAMMENI folds
9157 if (p < RExC_end - 4) {
9158 latest_char_state = iota_1;
9165 latest_char_state = generic_char;
9169 if (latest_char_state == upsilon_1) {
9170 latest_char_state = upsilon_2;
9172 else if (latest_char_state == iota_1) {
9173 latest_char_state = iota_2;
9176 latest_char_state = generic_char;
9180 if (latest_char_state == upsilon_2) {
9181 ender = GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS;
9184 else if (latest_char_state == iota_2) {
9185 ender = GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS;
9188 latest_char_state = generic_char;
9191 /* These are the tricky fold characters. Flush any
9192 * buffer first. (When adding to this list, also should
9193 * add them to fold_grind.t to make sure get tested) */
9194 case GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS:
9195 case GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS:
9196 case LATIN_SMALL_LETTER_SHARP_S:
9197 case LATIN_CAPITAL_LETTER_SHARP_S:
9198 case 0x1FD3: /* GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA */
9199 case 0x1FE3: /* GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA */
9206 char* const oldregxend = RExC_end;
9207 U8 tmpbuf[UTF8_MAXBYTES+1];
9209 /* Here, we know we need to generate a special
9210 * regnode, and 'ender' contains the tricky
9211 * character. What's done is to pretend it's in a
9212 * [bracketed] class, and let the code that deals
9213 * with those handle it, as that code has all the
9214 * intelligence necessary. First save the current
9215 * parse state, get rid of the already allocated
9216 * but empty EXACT node that the ANYOFV node will
9217 * replace, and point the parse to a buffer which
9218 * we fill with the character we want the regclass
9219 * code to think is being parsed */
9220 RExC_emit = orig_emit;
9221 RExC_parse = (char *) tmpbuf;
9223 U8 *d = uvchr_to_utf8(tmpbuf, ender);
9225 RExC_end = (char *) d;
9227 else { /* ender above 255 already excluded */
9228 tmpbuf[0] = (U8) ender;
9230 RExC_end = RExC_parse + 1;
9233 ret = regclass(pRExC_state,depth+1);
9235 /* Here, have parsed the buffer. Reset the parse to
9236 * the actual input, and return */
9237 RExC_end = oldregxend;
9240 Set_Node_Offset(ret, RExC_parse);
9241 Set_Node_Cur_Length(ret);
9242 nextchar(pRExC_state);
9243 *flagp |= HASWIDTH|SIMPLE;
9249 if ( RExC_flags & RXf_PMf_EXTENDED)
9250 p = regwhite( pRExC_state, p );
9252 /* Prime the casefolded buffer. Locale rules, which apply
9253 * only to code points < 256, aren't known until execution,
9254 * so for them, just output the original character using
9256 if (LOC && ender < 256) {
9257 if (UNI_IS_INVARIANT(ender)) {
9258 *tmpbuf = (U8) ender;
9261 *tmpbuf = UTF8_TWO_BYTE_HI(ender);
9262 *(tmpbuf + 1) = UTF8_TWO_BYTE_LO(ender);
9266 else if (isASCII(ender)) { /* Note: Here can't also be LOC
9268 ender = toLOWER(ender);
9269 *tmpbuf = (U8) ender;
9272 else if (! MORE_ASCII_RESTRICTED && ! LOC) {
9274 /* Locale and /aa require more selectivity about the
9275 * fold, so are handled below. Otherwise, here, just
9277 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
9280 /* Under locale rules or /aa we are not to mix,
9281 * respectively, ords < 256 or ASCII with non-. So
9282 * reject folds that mix them, using only the
9283 * non-folded code point. So do the fold to a
9284 * temporary, and inspect each character in it. */
9285 U8 trialbuf[UTF8_MAXBYTES_CASE+1];
9287 UV tmpender = toFOLD_uni(ender, trialbuf, &foldlen);
9288 U8* e = s + foldlen;
9289 bool fold_ok = TRUE;
9293 || (LOC && (UTF8_IS_INVARIANT(*s)
9294 || UTF8_IS_DOWNGRADEABLE_START(*s))))
9302 Copy(trialbuf, tmpbuf, foldlen, U8);
9306 uvuni_to_utf8(tmpbuf, ender);
9307 foldlen = UNISKIP(ender);
9311 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
9316 /* Emit all the Unicode characters. */
9318 for (foldbuf = tmpbuf;
9320 foldlen -= numlen) {
9321 ender = utf8_to_uvchr(foldbuf, &numlen);
9323 const STRLEN unilen = reguni(pRExC_state, ender, s);
9326 /* In EBCDIC the numlen
9327 * and unilen can differ. */
9329 if (numlen >= foldlen)
9333 break; /* "Can't happen." */
9337 const STRLEN unilen = reguni(pRExC_state, ender, s);
9346 REGC((char)ender, s++);
9352 /* Emit all the Unicode characters. */
9354 for (foldbuf = tmpbuf;
9356 foldlen -= numlen) {
9357 ender = utf8_to_uvchr(foldbuf, &numlen);
9359 const STRLEN unilen = reguni(pRExC_state, ender, s);
9362 /* In EBCDIC the numlen
9363 * and unilen can differ. */
9365 if (numlen >= foldlen)
9373 const STRLEN unilen = reguni(pRExC_state, ender, s);
9382 REGC((char)ender, s++);
9385 loopdone: /* Jumped to when encounters something that shouldn't be in
9388 Set_Node_Cur_Length(ret); /* MJD */
9389 nextchar(pRExC_state);
9391 /* len is STRLEN which is unsigned, need to copy to signed */
9394 vFAIL("Internal disaster");
9398 if (len == 1 && UNI_IS_INVARIANT(ender))
9402 RExC_size += STR_SZ(len);
9405 RExC_emit += STR_SZ(len);
9413 /* Jumped to when an unrecognized character set is encountered */
9415 Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags));
9420 S_regwhite( RExC_state_t *pRExC_state, char *p )
9422 const char *e = RExC_end;
9424 PERL_ARGS_ASSERT_REGWHITE;
9429 else if (*p == '#') {
9438 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
9446 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
9447 Character classes ([:foo:]) can also be negated ([:^foo:]).
9448 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
9449 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
9450 but trigger failures because they are currently unimplemented. */
9452 #define POSIXCC_DONE(c) ((c) == ':')
9453 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
9454 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
9457 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
9460 I32 namedclass = OOB_NAMEDCLASS;
9462 PERL_ARGS_ASSERT_REGPPOSIXCC;
9464 if (value == '[' && RExC_parse + 1 < RExC_end &&
9465 /* I smell either [: or [= or [. -- POSIX has been here, right? */
9466 POSIXCC(UCHARAT(RExC_parse))) {
9467 const char c = UCHARAT(RExC_parse);
9468 char* const s = RExC_parse++;
9470 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
9472 if (RExC_parse == RExC_end)
9473 /* Grandfather lone [:, [=, [. */
9476 const char* const t = RExC_parse++; /* skip over the c */
9479 if (UCHARAT(RExC_parse) == ']') {
9480 const char *posixcc = s + 1;
9481 RExC_parse++; /* skip over the ending ] */
9484 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
9485 const I32 skip = t - posixcc;
9487 /* Initially switch on the length of the name. */
9490 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
9491 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
9494 /* Names all of length 5. */
9495 /* alnum alpha ascii blank cntrl digit graph lower
9496 print punct space upper */
9497 /* Offset 4 gives the best switch position. */
9498 switch (posixcc[4]) {
9500 if (memEQ(posixcc, "alph", 4)) /* alpha */
9501 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
9504 if (memEQ(posixcc, "spac", 4)) /* space */
9505 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
9508 if (memEQ(posixcc, "grap", 4)) /* graph */
9509 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
9512 if (memEQ(posixcc, "asci", 4)) /* ascii */
9513 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
9516 if (memEQ(posixcc, "blan", 4)) /* blank */
9517 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
9520 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
9521 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
9524 if (memEQ(posixcc, "alnu", 4)) /* alnum */
9525 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
9528 if (memEQ(posixcc, "lowe", 4)) /* lower */
9529 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
9530 else if (memEQ(posixcc, "uppe", 4)) /* upper */
9531 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
9534 if (memEQ(posixcc, "digi", 4)) /* digit */
9535 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
9536 else if (memEQ(posixcc, "prin", 4)) /* print */
9537 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
9538 else if (memEQ(posixcc, "punc", 4)) /* punct */
9539 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
9544 if (memEQ(posixcc, "xdigit", 6))
9545 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
9549 if (namedclass == OOB_NAMEDCLASS)
9550 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
9552 assert (posixcc[skip] == ':');
9553 assert (posixcc[skip+1] == ']');
9554 } else if (!SIZE_ONLY) {
9555 /* [[=foo=]] and [[.foo.]] are still future. */
9557 /* adjust RExC_parse so the warning shows after
9559 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
9561 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9564 /* Maternal grandfather:
9565 * "[:" ending in ":" but not in ":]" */
9575 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
9579 PERL_ARGS_ASSERT_CHECKPOSIXCC;
9581 if (POSIXCC(UCHARAT(RExC_parse))) {
9582 const char *s = RExC_parse;
9583 const char c = *s++;
9587 if (*s && c == *s && s[1] == ']') {
9589 "POSIX syntax [%c %c] belongs inside character classes",
9592 /* [[=foo=]] and [[.foo.]] are still future. */
9593 if (POSIXCC_NOTYET(c)) {
9594 /* adjust RExC_parse so the error shows after
9596 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
9598 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9604 /* No locale test, and always Unicode semantics */
9605 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
9607 for (value = 0; value < 256; value++) \
9609 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9613 case ANYOF_N##NAME: \
9614 for (value = 0; value < 256; value++) \
9616 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9621 /* Like the above, but there are differences if we are in uni-8-bit or not, so
9622 * there are two tests passed in, to use depending on that. There aren't any
9623 * cases where the label is different from the name, so no need for that
9625 #define _C_C_T_(NAME, TEST_8, TEST_7, WORD) \
9627 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
9628 else if (UNI_SEMANTICS) { \
9629 for (value = 0; value < 256; value++) { \
9630 if (TEST_8(value)) stored += \
9631 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9635 for (value = 0; value < 128; value++) { \
9636 if (TEST_7(UNI_TO_NATIVE(value))) stored += \
9637 set_regclass_bit(pRExC_state, ret, \
9638 (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9644 case ANYOF_N##NAME: \
9645 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
9646 else if (UNI_SEMANTICS) { \
9647 for (value = 0; value < 256; value++) { \
9648 if (! TEST_8(value)) stored += \
9649 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9653 for (value = 0; value < 128; value++) { \
9654 if (! TEST_7(UNI_TO_NATIVE(value))) stored += set_regclass_bit( \
9655 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9657 if (AT_LEAST_ASCII_RESTRICTED) { \
9658 for (value = 128; value < 256; value++) { \
9659 stored += set_regclass_bit( \
9660 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9662 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; \
9665 /* For a non-ut8 target string with DEPENDS semantics, all above \
9666 * ASCII Latin1 code points match the complement of any of the \
9667 * classes. But in utf8, they have their Unicode semantics, so \
9668 * can't just set them in the bitmap, or else regexec.c will think \
9669 * they matched when they shouldn't. */ \
9670 ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; \
9678 S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr)
9681 /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes.
9682 * Locale folding is done at run-time, so this function should not be
9683 * called for nodes that are for locales.
9685 * This function sets the bit corresponding to the fold of the input
9686 * 'value', if not already set. The fold of 'f' is 'F', and the fold of
9689 * It also knows about the characters that are in the bitmap that have
9690 * folds that are matchable only outside it, and sets the appropriate lists
9693 * It returns the number of bits that actually changed from 0 to 1 */
9698 PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD;
9700 fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value]
9703 /* It assumes the bit for 'value' has already been set */
9704 if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) {
9705 ANYOF_BITMAP_SET(node, fold);
9708 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) {
9709 /* Certain Latin1 characters have matches outside the bitmap. To get
9710 * here, 'value' is one of those characters. None of these matches is
9711 * valid for ASCII characters under /aa, which have been excluded by
9712 * the 'if' above. The matches fall into three categories:
9713 * 1) They are singly folded-to or -from an above 255 character, as
9714 * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y
9716 * 2) They are part of a multi-char fold with another character in the
9717 * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill;
9718 * 3) They are part of a multi-char fold with a character not in the
9719 * bitmap, such as various ligatures.
9720 * We aren't dealing fully with multi-char folds, except we do deal
9721 * with the pattern containing a character that has a multi-char fold
9722 * (not so much the inverse).
9723 * For types 1) and 3), the matches only happen when the target string
9724 * is utf8; that's not true for 2), and we set a flag for it.
9726 * The code below adds to the passed in inversion list the single fold
9727 * closures for 'value'. The values are hard-coded here so that an
9728 * innocent-looking character class, like /[ks]/i won't have to go out
9729 * to disk to find the possible matches. XXX It would be better to
9730 * generate these via regen, in case a new version of the Unicode
9731 * standard adds new mappings, though that is not really likely. */
9736 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A);
9740 /* LATIN SMALL LETTER LONG S */
9741 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F);
9744 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9745 GREEK_SMALL_LETTER_MU);
9746 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9747 GREEK_CAPITAL_LETTER_MU);
9749 case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
9750 case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
9752 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B);
9753 if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */
9754 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9755 PL_fold_latin1[value]);
9758 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
9759 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9760 LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
9762 case LATIN_SMALL_LETTER_SHARP_S:
9763 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9764 LATIN_CAPITAL_LETTER_SHARP_S);
9766 /* Under /a, /d, and /u, this can match the two chars "ss" */
9767 if (! MORE_ASCII_RESTRICTED) {
9768 add_alternate(alternate_ptr, (U8 *) "ss", 2);
9770 /* And under /u or /a, it can match even if the target is
9772 if (AT_LEAST_UNI_SEMANTICS) {
9773 ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8;
9787 /* These all are targets of multi-character folds from code
9788 * points that require UTF8 to express, so they can't match
9789 * unless the target string is in UTF-8, so no action here is
9790 * necessary, as regexec.c properly handles the general case
9791 * for UTF-8 matching */
9794 /* Use deprecated warning to increase the chances of this
9796 ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value);
9800 else if (DEPENDS_SEMANTICS
9802 && PL_fold_latin1[value] != value)
9804 /* Under DEPENDS rules, non-ASCII Latin1 characters match their
9805 * folds only when the target string is in UTF-8. We add the fold
9806 * here to the list of things to match outside the bitmap, which
9807 * won't be looked at unless it is UTF8 (or else if something else
9808 * says to look even if not utf8, but those things better not happen
9809 * under DEPENDS semantics. */
9810 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]);
9817 PERL_STATIC_INLINE U8
9818 S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr)
9820 /* This inline function sets a bit in the bitmap if not already set, and if
9821 * appropriate, its fold, returning the number of bits that actually
9822 * changed from 0 to 1 */
9826 PERL_ARGS_ASSERT_SET_REGCLASS_BIT;
9828 if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */
9832 ANYOF_BITMAP_SET(node, value);
9835 if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */
9836 stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr);
9843 S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len)
9845 /* Adds input 'string' with length 'len' to the ANYOF node's unicode
9846 * alternate list, pointed to by 'alternate_ptr'. This is an array of
9847 * the multi-character folds of characters in the node */
9850 PERL_ARGS_ASSERT_ADD_ALTERNATE;
9852 if (! *alternate_ptr) {
9853 *alternate_ptr = newAV();
9855 sv = newSVpvn_utf8((char*)string, len, TRUE);
9856 av_push(*alternate_ptr, sv);
9861 parse a class specification and produce either an ANYOF node that
9862 matches the pattern or perhaps will be optimized into an EXACTish node
9863 instead. The node contains a bit map for the first 256 characters, with the
9864 corresponding bit set if that character is in the list. For characters
9865 above 255, a range list is used */
9868 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
9871 register UV nextvalue;
9872 register IV prevvalue = OOB_UNICODE;
9873 register IV range = 0;
9874 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
9875 register regnode *ret;
9878 char *rangebegin = NULL;
9879 bool need_class = 0;
9880 bool allow_full_fold = TRUE; /* Assume wants multi-char folding */
9882 STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more
9883 than just initialized. */
9886 /* code points this node matches that can't be stored in the bitmap */
9887 SV* nonbitmap = NULL;
9889 /* The items that are to match that aren't stored in the bitmap, but are a
9890 * result of things that are stored there. This is the fold closure of
9891 * such a character, either because it has DEPENDS semantics and shouldn't
9892 * be matched unless the target string is utf8, or is a code point that is
9893 * too large for the bit map, as for example, the fold of the MICRO SIGN is
9894 * above 255. This all is solely for performance reasons. By having this
9895 * code know the outside-the-bitmap folds that the bitmapped characters are
9896 * involved with, we don't have to go out to disk to find the list of
9897 * matches, unless the character class includes code points that aren't
9898 * storable in the bit map. That means that a character class with an 's'
9899 * in it, for example, doesn't need to go out to disk to find everything
9900 * that matches. A 2nd list is used so that the 'nonbitmap' list is kept
9901 * empty unless there is something whose fold we don't know about, and will
9902 * have to go out to the disk to find. */
9903 SV* l1_fold_invlist = NULL;
9905 /* List of multi-character folds that are matched by this node */
9906 AV* unicode_alternate = NULL;
9908 UV literal_endpoint = 0;
9910 UV stored = 0; /* how many chars stored in the bitmap */
9912 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
9913 case we need to change the emitted regop to an EXACT. */
9914 const char * orig_parse = RExC_parse;
9915 GET_RE_DEBUG_FLAGS_DECL;
9917 PERL_ARGS_ASSERT_REGCLASS;
9919 PERL_UNUSED_ARG(depth);
9922 DEBUG_PARSE("clas");
9924 /* Assume we are going to generate an ANYOF node. */
9925 ret = reganode(pRExC_state, ANYOF, 0);
9929 ANYOF_FLAGS(ret) = 0;
9932 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
9936 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
9938 /* We have decided to not allow multi-char folds in inverted character
9939 * classes, due to the confusion that can happen, especially with
9940 * classes that are designed for a non-Unicode world: You have the
9941 * peculiar case that:
9942 "s s" =~ /^[^\xDF]+$/i => Y
9943 "ss" =~ /^[^\xDF]+$/i => N
9945 * See [perl #89750] */
9946 allow_full_fold = FALSE;
9950 RExC_size += ANYOF_SKIP;
9951 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
9954 RExC_emit += ANYOF_SKIP;
9956 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
9958 ANYOF_BITMAP_ZERO(ret);
9959 listsv = newSVpvs("# comment\n");
9960 initial_listsv_len = SvCUR(listsv);
9963 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9965 if (!SIZE_ONLY && POSIXCC(nextvalue))
9966 checkposixcc(pRExC_state);
9968 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
9969 if (UCHARAT(RExC_parse) == ']')
9973 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
9977 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
9980 rangebegin = RExC_parse;
9982 value = utf8n_to_uvchr((U8*)RExC_parse,
9983 RExC_end - RExC_parse,
9984 &numlen, UTF8_ALLOW_DEFAULT);
9985 RExC_parse += numlen;
9988 value = UCHARAT(RExC_parse++);
9990 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9991 if (value == '[' && POSIXCC(nextvalue))
9992 namedclass = regpposixcc(pRExC_state, value);
9993 else if (value == '\\') {
9995 value = utf8n_to_uvchr((U8*)RExC_parse,
9996 RExC_end - RExC_parse,
9997 &numlen, UTF8_ALLOW_DEFAULT);
9998 RExC_parse += numlen;
10001 value = UCHARAT(RExC_parse++);
10002 /* Some compilers cannot handle switching on 64-bit integer
10003 * values, therefore value cannot be an UV. Yes, this will
10004 * be a problem later if we want switch on Unicode.
10005 * A similar issue a little bit later when switching on
10006 * namedclass. --jhi */
10007 switch ((I32)value) {
10008 case 'w': namedclass = ANYOF_ALNUM; break;
10009 case 'W': namedclass = ANYOF_NALNUM; break;
10010 case 's': namedclass = ANYOF_SPACE; break;
10011 case 'S': namedclass = ANYOF_NSPACE; break;
10012 case 'd': namedclass = ANYOF_DIGIT; break;
10013 case 'D': namedclass = ANYOF_NDIGIT; break;
10014 case 'v': namedclass = ANYOF_VERTWS; break;
10015 case 'V': namedclass = ANYOF_NVERTWS; break;
10016 case 'h': namedclass = ANYOF_HORIZWS; break;
10017 case 'H': namedclass = ANYOF_NHORIZWS; break;
10018 case 'N': /* Handle \N{NAME} in class */
10020 /* We only pay attention to the first char of
10021 multichar strings being returned. I kinda wonder
10022 if this makes sense as it does change the behaviour
10023 from earlier versions, OTOH that behaviour was broken
10025 UV v; /* value is register so we cant & it /grrr */
10026 if (reg_namedseq(pRExC_state, &v, NULL, depth)) {
10036 if (RExC_parse >= RExC_end)
10037 vFAIL2("Empty \\%c{}", (U8)value);
10038 if (*RExC_parse == '{') {
10039 const U8 c = (U8)value;
10040 e = strchr(RExC_parse++, '}');
10042 vFAIL2("Missing right brace on \\%c{}", c);
10043 while (isSPACE(UCHARAT(RExC_parse)))
10045 if (e == RExC_parse)
10046 vFAIL2("Empty \\%c{}", c);
10047 n = e - RExC_parse;
10048 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
10056 if (UCHARAT(RExC_parse) == '^') {
10059 value = value == 'p' ? 'P' : 'p'; /* toggle */
10060 while (isSPACE(UCHARAT(RExC_parse))) {
10066 /* Add the property name to the list. If /i matching, give
10067 * a different name which consists of the normal name
10068 * sandwiched between two underscores and '_i'. The design
10069 * is discussed in the commit message for this. */
10070 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s%.*s%s\n",
10071 (value=='p' ? '+' : '!'),
10072 (FOLD) ? "__" : "",
10078 RExC_parse = e + 1;
10080 /* The \p could match something in the Latin1 range, hence
10081 * something that isn't utf8 */
10082 ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8;
10083 namedclass = ANYOF_MAX; /* no official name, but it's named */
10085 /* \p means they want Unicode semantics */
10086 RExC_uni_semantics = 1;
10089 case 'n': value = '\n'; break;
10090 case 'r': value = '\r'; break;
10091 case 't': value = '\t'; break;
10092 case 'f': value = '\f'; break;
10093 case 'b': value = '\b'; break;
10094 case 'e': value = ASCII_TO_NATIVE('\033');break;
10095 case 'a': value = ASCII_TO_NATIVE('\007');break;
10097 RExC_parse--; /* function expects to be pointed at the 'o' */
10099 const char* error_msg;
10100 bool valid = grok_bslash_o(RExC_parse,
10105 RExC_parse += numlen;
10110 if (PL_encoding && value < 0x100) {
10111 goto recode_encoding;
10115 if (*RExC_parse == '{') {
10116 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
10117 | PERL_SCAN_DISALLOW_PREFIX;
10118 char * const e = strchr(RExC_parse++, '}');
10120 vFAIL("Missing right brace on \\x{}");
10122 numlen = e - RExC_parse;
10123 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
10124 RExC_parse = e + 1;
10127 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
10129 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
10130 RExC_parse += numlen;
10132 if (PL_encoding && value < 0x100)
10133 goto recode_encoding;
10136 value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY);
10138 case '0': case '1': case '2': case '3': case '4':
10139 case '5': case '6': case '7':
10141 /* Take 1-3 octal digits */
10142 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
10144 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
10145 RExC_parse += numlen;
10146 if (PL_encoding && value < 0x100)
10147 goto recode_encoding;
10151 if (! RExC_override_recoding) {
10152 SV* enc = PL_encoding;
10153 value = reg_recode((const char)(U8)value, &enc);
10154 if (!enc && SIZE_ONLY)
10155 ckWARNreg(RExC_parse,
10156 "Invalid escape in the specified encoding");
10160 /* Allow \_ to not give an error */
10161 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
10162 ckWARN2reg(RExC_parse,
10163 "Unrecognized escape \\%c in character class passed through",
10168 } /* end of \blah */
10171 literal_endpoint++;
10174 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
10176 /* What matches in a locale is not known until runtime, so need to
10177 * (one time per class) allocate extra space to pass to regexec.
10178 * The space will contain a bit for each named class that is to be
10179 * matched against. This isn't needed for \p{} and pseudo-classes,
10180 * as they are not affected by locale, and hence are dealt with
10182 if (LOC && namedclass < ANYOF_MAX && ! need_class) {
10185 RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP;
10188 RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP;
10189 ANYOF_CLASS_ZERO(ret);
10191 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
10194 /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a
10195 * literal, as is the character that began the false range, i.e.
10196 * the 'a' in the examples */
10200 RExC_parse >= rangebegin ?
10201 RExC_parse - rangebegin : 0;
10202 ckWARN4reg(RExC_parse,
10203 "False [] range \"%*.*s\"",
10207 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
10208 if (prevvalue < 256) {
10210 set_regclass_bit(pRExC_state, ret, (U8) prevvalue, &l1_fold_invlist, &unicode_alternate);
10213 nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue);
10217 range = 0; /* this was not a true range */
10223 const char *what = NULL;
10226 /* Possible truncation here but in some 64-bit environments
10227 * the compiler gets heartburn about switch on 64-bit values.
10228 * A similar issue a little earlier when switching on value.
10230 switch ((I32)namedclass) {
10232 case _C_C_T_(ALNUMC, isALNUMC_L1, isALNUMC, "XPosixAlnum");
10233 case _C_C_T_(ALPHA, isALPHA_L1, isALPHA, "XPosixAlpha");
10234 case _C_C_T_(BLANK, isBLANK_L1, isBLANK, "XPosixBlank");
10235 case _C_C_T_(CNTRL, isCNTRL_L1, isCNTRL, "XPosixCntrl");
10236 case _C_C_T_(GRAPH, isGRAPH_L1, isGRAPH, "XPosixGraph");
10237 case _C_C_T_(LOWER, isLOWER_L1, isLOWER, "XPosixLower");
10238 case _C_C_T_(PRINT, isPRINT_L1, isPRINT, "XPosixPrint");
10239 case _C_C_T_(PSXSPC, isPSXSPC_L1, isPSXSPC, "XPosixSpace");
10240 case _C_C_T_(PUNCT, isPUNCT_L1, isPUNCT, "XPosixPunct");
10241 case _C_C_T_(UPPER, isUPPER_L1, isUPPER, "XPosixUpper");
10242 /* \s, \w match all unicode if utf8. */
10243 case _C_C_T_(SPACE, isSPACE_L1, isSPACE, "SpacePerl");
10244 case _C_C_T_(ALNUM, isWORDCHAR_L1, isALNUM, "Word");
10245 case _C_C_T_(XDIGIT, isXDIGIT_L1, isXDIGIT, "XPosixXDigit");
10246 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
10247 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
10250 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
10252 for (value = 0; value < 128; value++)
10254 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
10257 what = NULL; /* Doesn't match outside ascii, so
10258 don't want to add +utf8:: */
10262 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
10264 for (value = 128; value < 256; value++)
10266 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
10268 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
10274 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
10276 /* consecutive digits assumed */
10277 for (value = '0'; value <= '9'; value++)
10279 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
10286 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
10288 /* consecutive digits assumed */
10289 for (value = 0; value < '0'; value++)
10291 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
10292 for (value = '9' + 1; value < 256; value++)
10294 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
10298 if (AT_LEAST_ASCII_RESTRICTED ) {
10299 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
10303 /* this is to handle \p and \P */
10306 vFAIL("Invalid [::] class");
10309 if (what && ! (AT_LEAST_ASCII_RESTRICTED)) {
10310 /* Strings such as "+utf8::isWord\n" */
10311 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
10316 } /* end of namedclass \blah */
10319 if (prevvalue > (IV)value) /* b-a */ {
10320 const int w = RExC_parse - rangebegin;
10321 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
10322 range = 0; /* not a valid range */
10326 prevvalue = value; /* save the beginning of the range */
10327 if (RExC_parse+1 < RExC_end
10328 && *RExC_parse == '-'
10329 && RExC_parse[1] != ']')
10333 /* a bad range like \w-, [:word:]- ? */
10334 if (namedclass > OOB_NAMEDCLASS) {
10335 if (ckWARN(WARN_REGEXP)) {
10337 RExC_parse >= rangebegin ?
10338 RExC_parse - rangebegin : 0;
10340 "False [] range \"%*.*s\"",
10345 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
10347 range = 1; /* yeah, it's a range! */
10348 continue; /* but do it the next time */
10352 /* non-Latin1 code point implies unicode semantics. Must be set in
10353 * pass1 so is there for the whole of pass 2 */
10355 RExC_uni_semantics = 1;
10358 /* now is the next time */
10360 if (prevvalue < 256) {
10361 const IV ceilvalue = value < 256 ? value : 255;
10364 /* In EBCDIC [\x89-\x91] should include
10365 * the \x8e but [i-j] should not. */
10366 if (literal_endpoint == 2 &&
10367 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
10368 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
10370 if (isLOWER(prevvalue)) {
10371 for (i = prevvalue; i <= ceilvalue; i++)
10372 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10374 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10377 for (i = prevvalue; i <= ceilvalue; i++)
10378 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10380 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10386 for (i = prevvalue; i <= ceilvalue; i++) {
10387 stored += set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10391 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
10392 const UV natvalue = NATIVE_TO_UNI(value);
10393 nonbitmap = add_range_to_invlist(nonbitmap, prevnatvalue, natvalue);
10396 literal_endpoint = 0;
10400 range = 0; /* this range (if it was one) is done now */
10407 /****** !SIZE_ONLY AFTER HERE *********/
10409 /* If folding and there are code points above 255, we calculate all
10410 * characters that could fold to or from the ones already on the list */
10411 if (FOLD && nonbitmap) {
10412 UV start, end; /* End points of code point ranges */
10414 SV* fold_intersection;
10416 /* This is a list of all the characters that participate in folds
10417 * (except marks, etc in multi-char folds */
10418 if (! PL_utf8_foldable) {
10419 SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0);
10420 PL_utf8_foldable = _swash_to_invlist(swash);
10423 /* This is a hash that for a particular fold gives all characters
10424 * that are involved in it */
10425 if (! PL_utf8_foldclosures) {
10427 /* If we were unable to find any folds, then we likely won't be
10428 * able to find the closures. So just create an empty list.
10429 * Folding will effectively be restricted to the non-Unicode rules
10430 * hard-coded into Perl. (This case happens legitimately during
10431 * compilation of Perl itself before the Unicode tables are
10433 if (invlist_len(PL_utf8_foldable) == 0) {
10434 PL_utf8_foldclosures = newHV();
10436 /* If the folds haven't been read in, call a fold function
10438 if (! PL_utf8_tofold) {
10439 U8 dummy[UTF8_MAXBYTES+1];
10441 to_utf8_fold((U8*) "A", dummy, &dummy_len);
10443 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
10447 /* Only the characters in this class that participate in folds need
10448 * be checked. Get the intersection of this class and all the
10449 * possible characters that are foldable. This can quickly narrow
10450 * down a large class */
10451 _invlist_intersection(PL_utf8_foldable, nonbitmap, &fold_intersection);
10453 /* Now look at the foldable characters in this class individually */
10454 invlist_iterinit(fold_intersection);
10455 while (invlist_iternext(fold_intersection, &start, &end)) {
10458 /* Look at every character in the range */
10459 for (j = start; j <= end; j++) {
10462 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
10465 _to_uni_fold_flags(j, foldbuf, &foldlen, allow_full_fold);
10467 if (foldlen > (STRLEN)UNISKIP(f)) {
10469 /* Any multicharacter foldings (disallowed in
10470 * lookbehind patterns) require the following
10471 * transform: [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst) where
10472 * E folds into "pq" and F folds into "rst", all other
10473 * characters fold to single characters. We save away
10474 * these multicharacter foldings, to be later saved as
10475 * part of the additional "s" data. */
10476 if (! RExC_in_lookbehind) {
10478 U8* e = foldbuf + foldlen;
10480 /* If any of the folded characters of this are in
10481 * the Latin1 range, tell the regex engine that
10482 * this can match a non-utf8 target string. The
10483 * only multi-byte fold whose source is in the
10484 * Latin1 range (U+00DF) applies only when the
10485 * target string is utf8, or under unicode rules */
10486 if (j > 255 || AT_LEAST_UNI_SEMANTICS) {
10489 /* Can't mix ascii with non- under /aa */
10490 if (MORE_ASCII_RESTRICTED
10491 && (isASCII(*loc) != isASCII(j)))
10493 goto end_multi_fold;
10495 if (UTF8_IS_INVARIANT(*loc)
10496 || UTF8_IS_DOWNGRADEABLE_START(*loc))
10498 /* Can't mix above and below 256 under
10501 goto end_multi_fold;
10504 |= ANYOF_NONBITMAP_NON_UTF8;
10507 loc += UTF8SKIP(loc);
10511 add_alternate(&unicode_alternate, foldbuf, foldlen);
10515 /* This is special-cased, as it is the only letter which
10516 * has both a multi-fold and single-fold in Latin1. All
10517 * the other chars that have single and multi-folds are
10518 * always in utf8, and the utf8 folding algorithm catches
10520 if (! LOC && j == LATIN_CAPITAL_LETTER_SHARP_S) {
10521 stored += set_regclass_bit(pRExC_state,
10523 LATIN_SMALL_LETTER_SHARP_S,
10524 &l1_fold_invlist, &unicode_alternate);
10528 /* Single character fold. Add everything in its fold
10529 * closure to the list that this node should match */
10532 /* The fold closures data structure is a hash with the
10533 * keys being every character that is folded to, like
10534 * 'k', and the values each an array of everything that
10535 * folds to its key. e.g. [ 'k', 'K', KELVIN_SIGN ] */
10536 if ((listp = hv_fetch(PL_utf8_foldclosures,
10537 (char *) foldbuf, foldlen, FALSE)))
10539 AV* list = (AV*) *listp;
10541 for (k = 0; k <= av_len(list); k++) {
10542 SV** c_p = av_fetch(list, k, FALSE);
10545 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
10549 /* /aa doesn't allow folds between ASCII and
10550 * non-; /l doesn't allow them between above
10552 if ((MORE_ASCII_RESTRICTED
10553 && (isASCII(c) != isASCII(j)))
10554 || (LOC && ((c < 256) != (j < 256))))
10559 if (c < 256 && AT_LEAST_UNI_SEMANTICS) {
10560 stored += set_regclass_bit(pRExC_state,
10563 &l1_fold_invlist, &unicode_alternate);
10565 /* It may be that the code point is already
10566 * in this range or already in the bitmap,
10567 * in which case we need do nothing */
10568 else if ((c < start || c > end)
10570 || ! ANYOF_BITMAP_TEST(ret, c)))
10572 nonbitmap = add_cp_to_invlist(nonbitmap, c);
10579 SvREFCNT_dec(fold_intersection);
10582 /* Combine the two lists into one. */
10583 if (l1_fold_invlist) {
10585 _invlist_union(nonbitmap, l1_fold_invlist, &nonbitmap);
10586 SvREFCNT_dec(l1_fold_invlist);
10589 nonbitmap = l1_fold_invlist;
10593 /* Here, we have calculated what code points should be in the character
10594 * class. Now we can see about various optimizations. Fold calculation
10595 * needs to take place before inversion. Otherwise /[^k]/i would invert to
10596 * include K, which under /i would match k. */
10598 /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't
10599 * set the FOLD flag yet, so this this does optimize those. It doesn't
10600 * optimize locale. Doing so perhaps could be done as long as there is
10601 * nothing like \w in it; some thought also would have to be given to the
10602 * interaction with above 0x100 chars */
10604 && (ANYOF_FLAGS(ret) & ANYOF_INVERT)
10605 && ! unicode_alternate
10606 /* In case of /d, there are some things that should match only when in
10607 * not in the bitmap, i.e., they require UTF8 to match. These are
10608 * listed in nonbitmap. */
10610 || ! DEPENDS_SEMANTICS
10611 || (ANYOF_FLAGS(ret) & ANYOF_NONBITMAP_NON_UTF8))
10612 && SvCUR(listsv) == initial_listsv_len)
10615 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
10616 ANYOF_BITMAP(ret)[value] ^= 0xFF;
10617 /* The inversion means that everything above 255 is matched */
10618 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
10621 /* Here, also has things outside the bitmap. Go through each bit
10622 * individually and add it to the list to get rid of from those
10623 * things not in the bitmap */
10624 SV *remove_list = _new_invlist(2);
10625 _invlist_invert(nonbitmap);
10626 for (value = 0; value < 256; ++value) {
10627 if (ANYOF_BITMAP_TEST(ret, value)) {
10628 ANYOF_BITMAP_CLEAR(ret, value);
10629 remove_list = add_cp_to_invlist(remove_list, value);
10632 ANYOF_BITMAP_SET(ret, value);
10635 _invlist_subtract(nonbitmap, remove_list, &nonbitmap);
10636 SvREFCNT_dec(remove_list);
10639 stored = 256 - stored;
10641 /* Clear the invert flag since have just done it here */
10642 ANYOF_FLAGS(ret) &= ~ANYOF_INVERT;
10645 /* Folding in the bitmap is taken care of above, but not for locale (for
10646 * which we have to wait to see what folding is in effect at runtime), and
10647 * for things not in the bitmap. Set run-time fold flag for these */
10648 if (FOLD && (LOC || nonbitmap || unicode_alternate)) {
10649 ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD;
10652 /* A single character class can be "optimized" into an EXACTish node.
10653 * Note that since we don't currently count how many characters there are
10654 * outside the bitmap, we are XXX missing optimization possibilities for
10655 * them. This optimization can't happen unless this is a truly single
10656 * character class, which means that it can't be an inversion into a
10657 * many-character class, and there must be no possibility of there being
10658 * things outside the bitmap. 'stored' (only) for locales doesn't include
10659 * \w, etc, so have to make a special test that they aren't present
10661 * Similarly A 2-character class of the very special form like [bB] can be
10662 * optimized into an EXACTFish node, but only for non-locales, and for
10663 * characters which only have the two folds; so things like 'fF' and 'Ii'
10664 * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE
10667 && ! unicode_alternate
10668 && SvCUR(listsv) == initial_listsv_len
10669 && ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL))
10670 && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10671 || (! ANYOF_CLASS_TEST_ANY_SET(ret)))))
10672 || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10673 && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value))
10674 /* If the latest code point has a fold whose
10675 * bit is set, it must be the only other one */
10676 && ((prevvalue = PL_fold_latin1[value]) != (IV)value)
10677 && ANYOF_BITMAP_TEST(ret, prevvalue)))))
10679 /* Note that the information needed to decide to do this optimization
10680 * is not currently available until the 2nd pass, and that the actually
10681 * used EXACTish node takes less space than the calculated ANYOF node,
10682 * and hence the amount of space calculated in the first pass is larger
10683 * than actually used, so this optimization doesn't gain us any space.
10684 * But an EXACT node is faster than an ANYOF node, and can be combined
10685 * with any adjacent EXACT nodes later by the optimizer for further
10686 * gains. The speed of executing an EXACTF is similar to an ANYOF
10687 * node, so the optimization advantage comes from the ability to join
10688 * it to adjacent EXACT nodes */
10690 const char * cur_parse= RExC_parse;
10692 RExC_emit = (regnode *)orig_emit;
10693 RExC_parse = (char *)orig_parse;
10697 /* A locale node with one point can be folded; all the other cases
10698 * with folding will have two points, since we calculate them above
10700 if (ANYOF_FLAGS(ret) & ANYOF_LOC_NONBITMAP_FOLD) {
10707 else { /* else 2 chars in the bit map: the folds of each other */
10709 /* Use the folded value, which for the cases where we get here,
10710 * is just the lower case of the current one (which may resolve to
10711 * itself, or to the other one */
10712 value = toLOWER_LATIN1(value);
10713 if (AT_LEAST_UNI_SEMANTICS || !isASCII(value)) {
10715 /* To join adjacent nodes, they must be the exact EXACTish
10716 * type. Try to use the most likely type, by using EXACTFU if
10717 * the regex calls for them, or is required because the
10718 * character is non-ASCII */
10721 else { /* Otherwise, more likely to be EXACTF type */
10726 ret = reg_node(pRExC_state, op);
10727 RExC_parse = (char *)cur_parse;
10728 if (UTF && ! NATIVE_IS_INVARIANT(value)) {
10729 *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value);
10730 *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value);
10732 RExC_emit += STR_SZ(2);
10735 *STRING(ret)= (char)value;
10737 RExC_emit += STR_SZ(1);
10739 SvREFCNT_dec(listsv);
10745 invlist_iterinit(nonbitmap);
10746 while (invlist_iternext(nonbitmap, &start, &end)) {
10747 if (start == end) {
10748 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", start);
10751 /* The \t sets the whole range */
10752 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
10757 SvREFCNT_dec(nonbitmap);
10760 if (SvCUR(listsv) == initial_listsv_len && ! unicode_alternate) {
10761 ARG_SET(ret, ANYOF_NONBITMAP_EMPTY);
10762 SvREFCNT_dec(listsv);
10763 SvREFCNT_dec(unicode_alternate);
10767 AV * const av = newAV();
10769 /* The 0th element stores the character class description
10770 * in its textual form: used later (regexec.c:Perl_regclass_swash())
10771 * to initialize the appropriate swash (which gets stored in
10772 * the 1st element), and also useful for dumping the regnode.
10773 * The 2nd element stores the multicharacter foldings,
10774 * used later (regexec.c:S_reginclass()). */
10775 av_store(av, 0, listsv);
10776 av_store(av, 1, NULL);
10778 /* Store any computed multi-char folds only if we are allowing
10780 if (allow_full_fold) {
10781 av_store(av, 2, MUTABLE_SV(unicode_alternate));
10782 if (unicode_alternate) { /* This node is variable length */
10787 av_store(av, 2, NULL);
10789 rv = newRV_noinc(MUTABLE_SV(av));
10790 n = add_data(pRExC_state, 1, "s");
10791 RExC_rxi->data->data[n] = (void*)rv;
10799 /* reg_skipcomment()
10801 Absorbs an /x style # comments from the input stream.
10802 Returns true if there is more text remaining in the stream.
10803 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
10804 terminates the pattern without including a newline.
10806 Note its the callers responsibility to ensure that we are
10807 actually in /x mode
10812 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
10816 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
10818 while (RExC_parse < RExC_end)
10819 if (*RExC_parse++ == '\n') {
10824 /* we ran off the end of the pattern without ending
10825 the comment, so we have to add an \n when wrapping */
10826 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
10834 Advances the parse position, and optionally absorbs
10835 "whitespace" from the inputstream.
10837 Without /x "whitespace" means (?#...) style comments only,
10838 with /x this means (?#...) and # comments and whitespace proper.
10840 Returns the RExC_parse point from BEFORE the scan occurs.
10842 This is the /x friendly way of saying RExC_parse++.
10846 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
10848 char* const retval = RExC_parse++;
10850 PERL_ARGS_ASSERT_NEXTCHAR;
10853 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
10854 RExC_parse[2] == '#') {
10855 while (*RExC_parse != ')') {
10856 if (RExC_parse == RExC_end)
10857 FAIL("Sequence (?#... not terminated");
10863 if (RExC_flags & RXf_PMf_EXTENDED) {
10864 if (isSPACE(*RExC_parse)) {
10868 else if (*RExC_parse == '#') {
10869 if ( reg_skipcomment( pRExC_state ) )
10878 - reg_node - emit a node
10880 STATIC regnode * /* Location. */
10881 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
10884 register regnode *ptr;
10885 regnode * const ret = RExC_emit;
10886 GET_RE_DEBUG_FLAGS_DECL;
10888 PERL_ARGS_ASSERT_REG_NODE;
10891 SIZE_ALIGN(RExC_size);
10895 if (RExC_emit >= RExC_emit_bound)
10896 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10898 NODE_ALIGN_FILL(ret);
10900 FILL_ADVANCE_NODE(ptr, op);
10901 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 1);
10902 #ifdef RE_TRACK_PATTERN_OFFSETS
10903 if (RExC_offsets) { /* MJD */
10904 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
10905 "reg_node", __LINE__,
10907 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
10908 ? "Overwriting end of array!\n" : "OK",
10909 (UV)(RExC_emit - RExC_emit_start),
10910 (UV)(RExC_parse - RExC_start),
10911 (UV)RExC_offsets[0]));
10912 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
10920 - reganode - emit a node with an argument
10922 STATIC regnode * /* Location. */
10923 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
10926 register regnode *ptr;
10927 regnode * const ret = RExC_emit;
10928 GET_RE_DEBUG_FLAGS_DECL;
10930 PERL_ARGS_ASSERT_REGANODE;
10933 SIZE_ALIGN(RExC_size);
10938 assert(2==regarglen[op]+1);
10940 Anything larger than this has to allocate the extra amount.
10941 If we changed this to be:
10943 RExC_size += (1 + regarglen[op]);
10945 then it wouldn't matter. Its not clear what side effect
10946 might come from that so its not done so far.
10951 if (RExC_emit >= RExC_emit_bound)
10952 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10954 NODE_ALIGN_FILL(ret);
10956 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
10957 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 2);
10958 #ifdef RE_TRACK_PATTERN_OFFSETS
10959 if (RExC_offsets) { /* MJD */
10960 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10964 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
10965 "Overwriting end of array!\n" : "OK",
10966 (UV)(RExC_emit - RExC_emit_start),
10967 (UV)(RExC_parse - RExC_start),
10968 (UV)RExC_offsets[0]));
10969 Set_Cur_Node_Offset;
10977 - reguni - emit (if appropriate) a Unicode character
10980 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
10984 PERL_ARGS_ASSERT_REGUNI;
10986 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
10990 - reginsert - insert an operator in front of already-emitted operand
10992 * Means relocating the operand.
10995 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
10998 register regnode *src;
10999 register regnode *dst;
11000 register regnode *place;
11001 const int offset = regarglen[(U8)op];
11002 const int size = NODE_STEP_REGNODE + offset;
11003 GET_RE_DEBUG_FLAGS_DECL;
11005 PERL_ARGS_ASSERT_REGINSERT;
11006 PERL_UNUSED_ARG(depth);
11007 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
11008 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
11017 if (RExC_open_parens) {
11019 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
11020 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
11021 if ( RExC_open_parens[paren] >= opnd ) {
11022 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
11023 RExC_open_parens[paren] += size;
11025 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
11027 if ( RExC_close_parens[paren] >= opnd ) {
11028 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
11029 RExC_close_parens[paren] += size;
11031 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
11036 while (src > opnd) {
11037 StructCopy(--src, --dst, regnode);
11038 #ifdef RE_TRACK_PATTERN_OFFSETS
11039 if (RExC_offsets) { /* MJD 20010112 */
11040 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
11044 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
11045 ? "Overwriting end of array!\n" : "OK",
11046 (UV)(src - RExC_emit_start),
11047 (UV)(dst - RExC_emit_start),
11048 (UV)RExC_offsets[0]));
11049 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
11050 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
11056 place = opnd; /* Op node, where operand used to be. */
11057 #ifdef RE_TRACK_PATTERN_OFFSETS
11058 if (RExC_offsets) { /* MJD */
11059 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
11063 (UV)(place - RExC_emit_start) > RExC_offsets[0]
11064 ? "Overwriting end of array!\n" : "OK",
11065 (UV)(place - RExC_emit_start),
11066 (UV)(RExC_parse - RExC_start),
11067 (UV)RExC_offsets[0]));
11068 Set_Node_Offset(place, RExC_parse);
11069 Set_Node_Length(place, 1);
11072 src = NEXTOPER(place);
11073 FILL_ADVANCE_NODE(place, op);
11074 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (place) - 1);
11075 Zero(src, offset, regnode);
11079 - regtail - set the next-pointer at the end of a node chain of p to val.
11080 - SEE ALSO: regtail_study
11082 /* TODO: All three parms should be const */
11084 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
11087 register regnode *scan;
11088 GET_RE_DEBUG_FLAGS_DECL;
11090 PERL_ARGS_ASSERT_REGTAIL;
11092 PERL_UNUSED_ARG(depth);
11098 /* Find last node. */
11101 regnode * const temp = regnext(scan);
11103 SV * const mysv=sv_newmortal();
11104 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
11105 regprop(RExC_rx, mysv, scan);
11106 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
11107 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
11108 (temp == NULL ? "->" : ""),
11109 (temp == NULL ? PL_reg_name[OP(val)] : "")
11117 if (reg_off_by_arg[OP(scan)]) {
11118 ARG_SET(scan, val - scan);
11121 NEXT_OFF(scan) = val - scan;
11127 - regtail_study - set the next-pointer at the end of a node chain of p to val.
11128 - Look for optimizable sequences at the same time.
11129 - currently only looks for EXACT chains.
11131 This is experimental code. The idea is to use this routine to perform
11132 in place optimizations on branches and groups as they are constructed,
11133 with the long term intention of removing optimization from study_chunk so
11134 that it is purely analytical.
11136 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
11137 to control which is which.
11140 /* TODO: All four parms should be const */
11143 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
11146 register regnode *scan;
11148 #ifdef EXPERIMENTAL_INPLACESCAN
11151 GET_RE_DEBUG_FLAGS_DECL;
11153 PERL_ARGS_ASSERT_REGTAIL_STUDY;
11159 /* Find last node. */
11163 regnode * const temp = regnext(scan);
11164 #ifdef EXPERIMENTAL_INPLACESCAN
11165 if (PL_regkind[OP(scan)] == EXACT)
11166 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
11170 switch (OP(scan)) {
11176 if( exact == PSEUDO )
11178 else if ( exact != OP(scan) )
11187 SV * const mysv=sv_newmortal();
11188 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
11189 regprop(RExC_rx, mysv, scan);
11190 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
11191 SvPV_nolen_const(mysv),
11192 REG_NODE_NUM(scan),
11193 PL_reg_name[exact]);
11200 SV * const mysv_val=sv_newmortal();
11201 DEBUG_PARSE_MSG("");
11202 regprop(RExC_rx, mysv_val, val);
11203 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
11204 SvPV_nolen_const(mysv_val),
11205 (IV)REG_NODE_NUM(val),
11209 if (reg_off_by_arg[OP(scan)]) {
11210 ARG_SET(scan, val - scan);
11213 NEXT_OFF(scan) = val - scan;
11221 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
11225 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
11231 for (bit=0; bit<32; bit++) {
11232 if (flags & (1<<bit)) {
11233 if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */
11236 if (!set++ && lead)
11237 PerlIO_printf(Perl_debug_log, "%s",lead);
11238 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
11241 if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) {
11242 if (!set++ && lead) {
11243 PerlIO_printf(Perl_debug_log, "%s",lead);
11246 case REGEX_UNICODE_CHARSET:
11247 PerlIO_printf(Perl_debug_log, "UNICODE");
11249 case REGEX_LOCALE_CHARSET:
11250 PerlIO_printf(Perl_debug_log, "LOCALE");
11252 case REGEX_ASCII_RESTRICTED_CHARSET:
11253 PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED");
11255 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
11256 PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED");
11259 PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET");
11265 PerlIO_printf(Perl_debug_log, "\n");
11267 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
11273 Perl_regdump(pTHX_ const regexp *r)
11277 SV * const sv = sv_newmortal();
11278 SV *dsv= sv_newmortal();
11279 RXi_GET_DECL(r,ri);
11280 GET_RE_DEBUG_FLAGS_DECL;
11282 PERL_ARGS_ASSERT_REGDUMP;
11284 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
11286 /* Header fields of interest. */
11287 if (r->anchored_substr) {
11288 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
11289 RE_SV_DUMPLEN(r->anchored_substr), 30);
11290 PerlIO_printf(Perl_debug_log,
11291 "anchored %s%s at %"IVdf" ",
11292 s, RE_SV_TAIL(r->anchored_substr),
11293 (IV)r->anchored_offset);
11294 } else if (r->anchored_utf8) {
11295 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
11296 RE_SV_DUMPLEN(r->anchored_utf8), 30);
11297 PerlIO_printf(Perl_debug_log,
11298 "anchored utf8 %s%s at %"IVdf" ",
11299 s, RE_SV_TAIL(r->anchored_utf8),
11300 (IV)r->anchored_offset);
11302 if (r->float_substr) {
11303 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
11304 RE_SV_DUMPLEN(r->float_substr), 30);
11305 PerlIO_printf(Perl_debug_log,
11306 "floating %s%s at %"IVdf"..%"UVuf" ",
11307 s, RE_SV_TAIL(r->float_substr),
11308 (IV)r->float_min_offset, (UV)r->float_max_offset);
11309 } else if (r->float_utf8) {
11310 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
11311 RE_SV_DUMPLEN(r->float_utf8), 30);
11312 PerlIO_printf(Perl_debug_log,
11313 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
11314 s, RE_SV_TAIL(r->float_utf8),
11315 (IV)r->float_min_offset, (UV)r->float_max_offset);
11317 if (r->check_substr || r->check_utf8)
11318 PerlIO_printf(Perl_debug_log,
11320 (r->check_substr == r->float_substr
11321 && r->check_utf8 == r->float_utf8
11322 ? "(checking floating" : "(checking anchored"));
11323 if (r->extflags & RXf_NOSCAN)
11324 PerlIO_printf(Perl_debug_log, " noscan");
11325 if (r->extflags & RXf_CHECK_ALL)
11326 PerlIO_printf(Perl_debug_log, " isall");
11327 if (r->check_substr || r->check_utf8)
11328 PerlIO_printf(Perl_debug_log, ") ");
11330 if (ri->regstclass) {
11331 regprop(r, sv, ri->regstclass);
11332 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
11334 if (r->extflags & RXf_ANCH) {
11335 PerlIO_printf(Perl_debug_log, "anchored");
11336 if (r->extflags & RXf_ANCH_BOL)
11337 PerlIO_printf(Perl_debug_log, "(BOL)");
11338 if (r->extflags & RXf_ANCH_MBOL)
11339 PerlIO_printf(Perl_debug_log, "(MBOL)");
11340 if (r->extflags & RXf_ANCH_SBOL)
11341 PerlIO_printf(Perl_debug_log, "(SBOL)");
11342 if (r->extflags & RXf_ANCH_GPOS)
11343 PerlIO_printf(Perl_debug_log, "(GPOS)");
11344 PerlIO_putc(Perl_debug_log, ' ');
11346 if (r->extflags & RXf_GPOS_SEEN)
11347 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
11348 if (r->intflags & PREGf_SKIP)
11349 PerlIO_printf(Perl_debug_log, "plus ");
11350 if (r->intflags & PREGf_IMPLICIT)
11351 PerlIO_printf(Perl_debug_log, "implicit ");
11352 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
11353 if (r->extflags & RXf_EVAL_SEEN)
11354 PerlIO_printf(Perl_debug_log, "with eval ");
11355 PerlIO_printf(Perl_debug_log, "\n");
11356 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
11358 PERL_ARGS_ASSERT_REGDUMP;
11359 PERL_UNUSED_CONTEXT;
11360 PERL_UNUSED_ARG(r);
11361 #endif /* DEBUGGING */
11365 - regprop - printable representation of opcode
11367 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
11370 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
11371 if (flags & ANYOF_INVERT) \
11372 /*make sure the invert info is in each */ \
11373 sv_catpvs(sv, "^"); \
11379 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
11384 RXi_GET_DECL(prog,progi);
11385 GET_RE_DEBUG_FLAGS_DECL;
11387 PERL_ARGS_ASSERT_REGPROP;
11391 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
11392 /* It would be nice to FAIL() here, but this may be called from
11393 regexec.c, and it would be hard to supply pRExC_state. */
11394 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
11395 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
11397 k = PL_regkind[OP(o)];
11400 sv_catpvs(sv, " ");
11401 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
11402 * is a crude hack but it may be the best for now since
11403 * we have no flag "this EXACTish node was UTF-8"
11405 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
11406 PERL_PV_ESCAPE_UNI_DETECT |
11407 PERL_PV_ESCAPE_NONASCII |
11408 PERL_PV_PRETTY_ELLIPSES |
11409 PERL_PV_PRETTY_LTGT |
11410 PERL_PV_PRETTY_NOCLEAR
11412 } else if (k == TRIE) {
11413 /* print the details of the trie in dumpuntil instead, as
11414 * progi->data isn't available here */
11415 const char op = OP(o);
11416 const U32 n = ARG(o);
11417 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
11418 (reg_ac_data *)progi->data->data[n] :
11420 const reg_trie_data * const trie
11421 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
11423 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
11424 DEBUG_TRIE_COMPILE_r(
11425 Perl_sv_catpvf(aTHX_ sv,
11426 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
11427 (UV)trie->startstate,
11428 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
11429 (UV)trie->wordcount,
11432 (UV)TRIE_CHARCOUNT(trie),
11433 (UV)trie->uniquecharcount
11436 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
11438 int rangestart = -1;
11439 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
11440 sv_catpvs(sv, "[");
11441 for (i = 0; i <= 256; i++) {
11442 if (i < 256 && BITMAP_TEST(bitmap,i)) {
11443 if (rangestart == -1)
11445 } else if (rangestart != -1) {
11446 if (i <= rangestart + 3)
11447 for (; rangestart < i; rangestart++)
11448 put_byte(sv, rangestart);
11450 put_byte(sv, rangestart);
11451 sv_catpvs(sv, "-");
11452 put_byte(sv, i - 1);
11457 sv_catpvs(sv, "]");
11460 } else if (k == CURLY) {
11461 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
11462 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
11463 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
11465 else if (k == WHILEM && o->flags) /* Ordinal/of */
11466 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
11467 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
11468 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
11469 if ( RXp_PAREN_NAMES(prog) ) {
11470 if ( k != REF || (OP(o) < NREF)) {
11471 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
11472 SV **name= av_fetch(list, ARG(o), 0 );
11474 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11477 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
11478 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
11479 I32 *nums=(I32*)SvPVX(sv_dat);
11480 SV **name= av_fetch(list, nums[0], 0 );
11483 for ( n=0; n<SvIVX(sv_dat); n++ ) {
11484 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
11485 (n ? "," : ""), (IV)nums[n]);
11487 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11491 } else if (k == GOSUB)
11492 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
11493 else if (k == VERB) {
11495 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
11496 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
11497 } else if (k == LOGICAL)
11498 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
11499 else if (k == FOLDCHAR)
11500 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
11501 else if (k == ANYOF) {
11502 int i, rangestart = -1;
11503 const U8 flags = ANYOF_FLAGS(o);
11506 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
11507 static const char * const anyofs[] = {
11540 if (flags & ANYOF_LOCALE)
11541 sv_catpvs(sv, "{loc}");
11542 if (flags & ANYOF_LOC_NONBITMAP_FOLD)
11543 sv_catpvs(sv, "{i}");
11544 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
11545 if (flags & ANYOF_INVERT)
11546 sv_catpvs(sv, "^");
11548 /* output what the standard cp 0-255 bitmap matches */
11549 for (i = 0; i <= 256; i++) {
11550 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
11551 if (rangestart == -1)
11553 } else if (rangestart != -1) {
11554 if (i <= rangestart + 3)
11555 for (; rangestart < i; rangestart++)
11556 put_byte(sv, rangestart);
11558 put_byte(sv, rangestart);
11559 sv_catpvs(sv, "-");
11560 put_byte(sv, i - 1);
11567 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11568 /* output any special charclass tests (used entirely under use locale) */
11569 if (ANYOF_CLASS_TEST_ANY_SET(o))
11570 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
11571 if (ANYOF_CLASS_TEST(o,i)) {
11572 sv_catpv(sv, anyofs[i]);
11576 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11578 if (flags & ANYOF_NON_UTF8_LATIN1_ALL) {
11579 sv_catpvs(sv, "{non-utf8-latin1-all}");
11582 /* output information about the unicode matching */
11583 if (flags & ANYOF_UNICODE_ALL)
11584 sv_catpvs(sv, "{unicode_all}");
11585 else if (ANYOF_NONBITMAP(o))
11586 sv_catpvs(sv, "{unicode}");
11587 if (flags & ANYOF_NONBITMAP_NON_UTF8)
11588 sv_catpvs(sv, "{outside bitmap}");
11590 if (ANYOF_NONBITMAP(o)) {
11592 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
11596 U8 s[UTF8_MAXBYTES_CASE+1];
11598 for (i = 0; i <= 256; i++) { /* just the first 256 */
11599 uvchr_to_utf8(s, i);
11601 if (i < 256 && swash_fetch(sw, s, TRUE)) {
11602 if (rangestart == -1)
11604 } else if (rangestart != -1) {
11605 if (i <= rangestart + 3)
11606 for (; rangestart < i; rangestart++) {
11607 const U8 * const e = uvchr_to_utf8(s,rangestart);
11609 for(p = s; p < e; p++)
11613 const U8 *e = uvchr_to_utf8(s,rangestart);
11615 for (p = s; p < e; p++)
11617 sv_catpvs(sv, "-");
11618 e = uvchr_to_utf8(s, i-1);
11619 for (p = s; p < e; p++)
11626 sv_catpvs(sv, "..."); /* et cetera */
11630 char *s = savesvpv(lv);
11631 char * const origs = s;
11633 while (*s && *s != '\n')
11637 const char * const t = ++s;
11655 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
11657 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
11658 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
11660 PERL_UNUSED_CONTEXT;
11661 PERL_UNUSED_ARG(sv);
11662 PERL_UNUSED_ARG(o);
11663 PERL_UNUSED_ARG(prog);
11664 #endif /* DEBUGGING */
11668 Perl_re_intuit_string(pTHX_ REGEXP * const r)
11669 { /* Assume that RE_INTUIT is set */
11671 struct regexp *const prog = (struct regexp *)SvANY(r);
11672 GET_RE_DEBUG_FLAGS_DECL;
11674 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
11675 PERL_UNUSED_CONTEXT;
11679 const char * const s = SvPV_nolen_const(prog->check_substr
11680 ? prog->check_substr : prog->check_utf8);
11682 if (!PL_colorset) reginitcolors();
11683 PerlIO_printf(Perl_debug_log,
11684 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
11686 prog->check_substr ? "" : "utf8 ",
11687 PL_colors[5],PL_colors[0],
11690 (strlen(s) > 60 ? "..." : ""));
11693 return prog->check_substr ? prog->check_substr : prog->check_utf8;
11699 handles refcounting and freeing the perl core regexp structure. When
11700 it is necessary to actually free the structure the first thing it
11701 does is call the 'free' method of the regexp_engine associated to
11702 the regexp, allowing the handling of the void *pprivate; member
11703 first. (This routine is not overridable by extensions, which is why
11704 the extensions free is called first.)
11706 See regdupe and regdupe_internal if you change anything here.
11708 #ifndef PERL_IN_XSUB_RE
11710 Perl_pregfree(pTHX_ REGEXP *r)
11716 Perl_pregfree2(pTHX_ REGEXP *rx)
11719 struct regexp *const r = (struct regexp *)SvANY(rx);
11720 GET_RE_DEBUG_FLAGS_DECL;
11722 PERL_ARGS_ASSERT_PREGFREE2;
11724 if (r->mother_re) {
11725 ReREFCNT_dec(r->mother_re);
11727 CALLREGFREE_PVT(rx); /* free the private data */
11728 SvREFCNT_dec(RXp_PAREN_NAMES(r));
11731 SvREFCNT_dec(r->anchored_substr);
11732 SvREFCNT_dec(r->anchored_utf8);
11733 SvREFCNT_dec(r->float_substr);
11734 SvREFCNT_dec(r->float_utf8);
11735 Safefree(r->substrs);
11737 RX_MATCH_COPY_FREE(rx);
11738 #ifdef PERL_OLD_COPY_ON_WRITE
11739 SvREFCNT_dec(r->saved_copy);
11746 This is a hacky workaround to the structural issue of match results
11747 being stored in the regexp structure which is in turn stored in
11748 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
11749 could be PL_curpm in multiple contexts, and could require multiple
11750 result sets being associated with the pattern simultaneously, such
11751 as when doing a recursive match with (??{$qr})
11753 The solution is to make a lightweight copy of the regexp structure
11754 when a qr// is returned from the code executed by (??{$qr}) this
11755 lightweight copy doesn't actually own any of its data except for
11756 the starp/end and the actual regexp structure itself.
11762 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
11764 struct regexp *ret;
11765 struct regexp *const r = (struct regexp *)SvANY(rx);
11766 register const I32 npar = r->nparens+1;
11768 PERL_ARGS_ASSERT_REG_TEMP_COPY;
11771 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
11772 ret = (struct regexp *)SvANY(ret_x);
11774 (void)ReREFCNT_inc(rx);
11775 /* We can take advantage of the existing "copied buffer" mechanism in SVs
11776 by pointing directly at the buffer, but flagging that the allocated
11777 space in the copy is zero. As we've just done a struct copy, it's now
11778 a case of zero-ing that, rather than copying the current length. */
11779 SvPV_set(ret_x, RX_WRAPPED(rx));
11780 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
11781 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
11782 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
11783 SvLEN_set(ret_x, 0);
11784 SvSTASH_set(ret_x, NULL);
11785 SvMAGIC_set(ret_x, NULL);
11786 Newx(ret->offs, npar, regexp_paren_pair);
11787 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11789 Newx(ret->substrs, 1, struct reg_substr_data);
11790 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11792 SvREFCNT_inc_void(ret->anchored_substr);
11793 SvREFCNT_inc_void(ret->anchored_utf8);
11794 SvREFCNT_inc_void(ret->float_substr);
11795 SvREFCNT_inc_void(ret->float_utf8);
11797 /* check_substr and check_utf8, if non-NULL, point to either their
11798 anchored or float namesakes, and don't hold a second reference. */
11800 RX_MATCH_COPIED_off(ret_x);
11801 #ifdef PERL_OLD_COPY_ON_WRITE
11802 ret->saved_copy = NULL;
11804 ret->mother_re = rx;
11810 /* regfree_internal()
11812 Free the private data in a regexp. This is overloadable by
11813 extensions. Perl takes care of the regexp structure in pregfree(),
11814 this covers the *pprivate pointer which technically perl doesn't
11815 know about, however of course we have to handle the
11816 regexp_internal structure when no extension is in use.
11818 Note this is called before freeing anything in the regexp
11823 Perl_regfree_internal(pTHX_ REGEXP * const rx)
11826 struct regexp *const r = (struct regexp *)SvANY(rx);
11827 RXi_GET_DECL(r,ri);
11828 GET_RE_DEBUG_FLAGS_DECL;
11830 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
11836 SV *dsv= sv_newmortal();
11837 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
11838 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
11839 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
11840 PL_colors[4],PL_colors[5],s);
11843 #ifdef RE_TRACK_PATTERN_OFFSETS
11845 Safefree(ri->u.offsets); /* 20010421 MJD */
11848 int n = ri->data->count;
11849 PAD* new_comppad = NULL;
11854 /* If you add a ->what type here, update the comment in regcomp.h */
11855 switch (ri->data->what[n]) {
11860 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
11863 Safefree(ri->data->data[n]);
11866 new_comppad = MUTABLE_AV(ri->data->data[n]);
11869 if (new_comppad == NULL)
11870 Perl_croak(aTHX_ "panic: pregfree comppad");
11871 PAD_SAVE_LOCAL(old_comppad,
11872 /* Watch out for global destruction's random ordering. */
11873 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
11876 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
11879 op_free((OP_4tree*)ri->data->data[n]);
11881 PAD_RESTORE_LOCAL(old_comppad);
11882 SvREFCNT_dec(MUTABLE_SV(new_comppad));
11883 new_comppad = NULL;
11888 { /* Aho Corasick add-on structure for a trie node.
11889 Used in stclass optimization only */
11891 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
11893 refcount = --aho->refcount;
11896 PerlMemShared_free(aho->states);
11897 PerlMemShared_free(aho->fail);
11898 /* do this last!!!! */
11899 PerlMemShared_free(ri->data->data[n]);
11900 PerlMemShared_free(ri->regstclass);
11906 /* trie structure. */
11908 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
11910 refcount = --trie->refcount;
11913 PerlMemShared_free(trie->charmap);
11914 PerlMemShared_free(trie->states);
11915 PerlMemShared_free(trie->trans);
11917 PerlMemShared_free(trie->bitmap);
11919 PerlMemShared_free(trie->jump);
11920 PerlMemShared_free(trie->wordinfo);
11921 /* do this last!!!! */
11922 PerlMemShared_free(ri->data->data[n]);
11927 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
11930 Safefree(ri->data->what);
11931 Safefree(ri->data);
11937 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11938 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11939 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11942 re_dup - duplicate a regexp.
11944 This routine is expected to clone a given regexp structure. It is only
11945 compiled under USE_ITHREADS.
11947 After all of the core data stored in struct regexp is duplicated
11948 the regexp_engine.dupe method is used to copy any private data
11949 stored in the *pprivate pointer. This allows extensions to handle
11950 any duplication it needs to do.
11952 See pregfree() and regfree_internal() if you change anything here.
11954 #if defined(USE_ITHREADS)
11955 #ifndef PERL_IN_XSUB_RE
11957 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
11961 const struct regexp *r = (const struct regexp *)SvANY(sstr);
11962 struct regexp *ret = (struct regexp *)SvANY(dstr);
11964 PERL_ARGS_ASSERT_RE_DUP_GUTS;
11966 npar = r->nparens+1;
11967 Newx(ret->offs, npar, regexp_paren_pair);
11968 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11970 /* no need to copy these */
11971 Newx(ret->swap, npar, regexp_paren_pair);
11974 if (ret->substrs) {
11975 /* Do it this way to avoid reading from *r after the StructCopy().
11976 That way, if any of the sv_dup_inc()s dislodge *r from the L1
11977 cache, it doesn't matter. */
11978 const bool anchored = r->check_substr
11979 ? r->check_substr == r->anchored_substr
11980 : r->check_utf8 == r->anchored_utf8;
11981 Newx(ret->substrs, 1, struct reg_substr_data);
11982 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11984 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
11985 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
11986 ret->float_substr = sv_dup_inc(ret->float_substr, param);
11987 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
11989 /* check_substr and check_utf8, if non-NULL, point to either their
11990 anchored or float namesakes, and don't hold a second reference. */
11992 if (ret->check_substr) {
11994 assert(r->check_utf8 == r->anchored_utf8);
11995 ret->check_substr = ret->anchored_substr;
11996 ret->check_utf8 = ret->anchored_utf8;
11998 assert(r->check_substr == r->float_substr);
11999 assert(r->check_utf8 == r->float_utf8);
12000 ret->check_substr = ret->float_substr;
12001 ret->check_utf8 = ret->float_utf8;
12003 } else if (ret->check_utf8) {
12005 ret->check_utf8 = ret->anchored_utf8;
12007 ret->check_utf8 = ret->float_utf8;
12012 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
12015 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
12017 if (RX_MATCH_COPIED(dstr))
12018 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
12020 ret->subbeg = NULL;
12021 #ifdef PERL_OLD_COPY_ON_WRITE
12022 ret->saved_copy = NULL;
12025 if (ret->mother_re) {
12026 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
12027 /* Our storage points directly to our mother regexp, but that's
12028 1: a buffer in a different thread
12029 2: something we no longer hold a reference on
12030 so we need to copy it locally. */
12031 /* Note we need to sue SvCUR() on our mother_re, because it, in
12032 turn, may well be pointing to its own mother_re. */
12033 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
12034 SvCUR(ret->mother_re)+1));
12035 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
12037 ret->mother_re = NULL;
12041 #endif /* PERL_IN_XSUB_RE */
12046 This is the internal complement to regdupe() which is used to copy
12047 the structure pointed to by the *pprivate pointer in the regexp.
12048 This is the core version of the extension overridable cloning hook.
12049 The regexp structure being duplicated will be copied by perl prior
12050 to this and will be provided as the regexp *r argument, however
12051 with the /old/ structures pprivate pointer value. Thus this routine
12052 may override any copying normally done by perl.
12054 It returns a pointer to the new regexp_internal structure.
12058 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
12061 struct regexp *const r = (struct regexp *)SvANY(rx);
12062 regexp_internal *reti;
12064 RXi_GET_DECL(r,ri);
12066 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
12070 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
12071 Copy(ri->program, reti->program, len+1, regnode);
12074 reti->regstclass = NULL;
12077 struct reg_data *d;
12078 const int count = ri->data->count;
12081 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
12082 char, struct reg_data);
12083 Newx(d->what, count, U8);
12086 for (i = 0; i < count; i++) {
12087 d->what[i] = ri->data->what[i];
12088 switch (d->what[i]) {
12089 /* legal options are one of: sSfpontTua
12090 see also regcomp.h and pregfree() */
12091 case 'a': /* actually an AV, but the dup function is identical. */
12094 case 'p': /* actually an AV, but the dup function is identical. */
12095 case 'u': /* actually an HV, but the dup function is identical. */
12096 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
12099 /* This is cheating. */
12100 Newx(d->data[i], 1, struct regnode_charclass_class);
12101 StructCopy(ri->data->data[i], d->data[i],
12102 struct regnode_charclass_class);
12103 reti->regstclass = (regnode*)d->data[i];
12106 /* Compiled op trees are readonly and in shared memory,
12107 and can thus be shared without duplication. */
12109 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
12113 /* Trie stclasses are readonly and can thus be shared
12114 * without duplication. We free the stclass in pregfree
12115 * when the corresponding reg_ac_data struct is freed.
12117 reti->regstclass= ri->regstclass;
12121 ((reg_trie_data*)ri->data->data[i])->refcount++;
12125 d->data[i] = ri->data->data[i];
12128 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
12137 reti->name_list_idx = ri->name_list_idx;
12139 #ifdef RE_TRACK_PATTERN_OFFSETS
12140 if (ri->u.offsets) {
12141 Newx(reti->u.offsets, 2*len+1, U32);
12142 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
12145 SetProgLen(reti,len);
12148 return (void*)reti;
12151 #endif /* USE_ITHREADS */
12153 #ifndef PERL_IN_XSUB_RE
12156 - regnext - dig the "next" pointer out of a node
12159 Perl_regnext(pTHX_ register regnode *p)
12162 register I32 offset;
12167 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
12168 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
12171 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
12180 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
12183 STRLEN l1 = strlen(pat1);
12184 STRLEN l2 = strlen(pat2);
12187 const char *message;
12189 PERL_ARGS_ASSERT_RE_CROAK2;
12195 Copy(pat1, buf, l1 , char);
12196 Copy(pat2, buf + l1, l2 , char);
12197 buf[l1 + l2] = '\n';
12198 buf[l1 + l2 + 1] = '\0';
12200 /* ANSI variant takes additional second argument */
12201 va_start(args, pat2);
12205 msv = vmess(buf, &args);
12207 message = SvPV_const(msv,l1);
12210 Copy(message, buf, l1 , char);
12211 buf[l1-1] = '\0'; /* Overwrite \n */
12212 Perl_croak(aTHX_ "%s", buf);
12215 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
12217 #ifndef PERL_IN_XSUB_RE
12219 Perl_save_re_context(pTHX)
12223 struct re_save_state *state;
12225 SAVEVPTR(PL_curcop);
12226 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
12228 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
12229 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
12230 SSPUSHUV(SAVEt_RE_STATE);
12232 Copy(&PL_reg_state, state, 1, struct re_save_state);
12234 PL_reg_start_tmp = 0;
12235 PL_reg_start_tmpl = 0;
12236 PL_reg_oldsaved = NULL;
12237 PL_reg_oldsavedlen = 0;
12238 PL_reg_maxiter = 0;
12239 PL_reg_leftiter = 0;
12240 PL_reg_poscache = NULL;
12241 PL_reg_poscache_size = 0;
12242 #ifdef PERL_OLD_COPY_ON_WRITE
12246 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
12248 const REGEXP * const rx = PM_GETRE(PL_curpm);
12251 for (i = 1; i <= RX_NPARENS(rx); i++) {
12252 char digits[TYPE_CHARS(long)];
12253 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
12254 GV *const *const gvp
12255 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
12258 GV * const gv = *gvp;
12259 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
12269 clear_re(pTHX_ void *r)
12272 ReREFCNT_dec((REGEXP *)r);
12278 S_put_byte(pTHX_ SV *sv, int c)
12280 PERL_ARGS_ASSERT_PUT_BYTE;
12282 /* Our definition of isPRINT() ignores locales, so only bytes that are
12283 not part of UTF-8 are considered printable. I assume that the same
12284 holds for UTF-EBCDIC.
12285 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
12286 which Wikipedia says:
12288 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
12289 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
12290 identical, to the ASCII delete (DEL) or rubout control character.
12291 ) So the old condition can be simplified to !isPRINT(c) */
12294 Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c);
12297 Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c);
12301 const char string = c;
12302 if (c == '-' || c == ']' || c == '\\' || c == '^')
12303 sv_catpvs(sv, "\\");
12304 sv_catpvn(sv, &string, 1);
12309 #define CLEAR_OPTSTART \
12310 if (optstart) STMT_START { \
12311 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
12315 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
12317 STATIC const regnode *
12318 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
12319 const regnode *last, const regnode *plast,
12320 SV* sv, I32 indent, U32 depth)
12323 register U8 op = PSEUDO; /* Arbitrary non-END op. */
12324 register const regnode *next;
12325 const regnode *optstart= NULL;
12327 RXi_GET_DECL(r,ri);
12328 GET_RE_DEBUG_FLAGS_DECL;
12330 PERL_ARGS_ASSERT_DUMPUNTIL;
12332 #ifdef DEBUG_DUMPUNTIL
12333 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
12334 last ? last-start : 0,plast ? plast-start : 0);
12337 if (plast && plast < last)
12340 while (PL_regkind[op] != END && (!last || node < last)) {
12341 /* While that wasn't END last time... */
12344 if (op == CLOSE || op == WHILEM)
12346 next = regnext((regnode *)node);
12349 if (OP(node) == OPTIMIZED) {
12350 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
12357 regprop(r, sv, node);
12358 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
12359 (int)(2*indent + 1), "", SvPVX_const(sv));
12361 if (OP(node) != OPTIMIZED) {
12362 if (next == NULL) /* Next ptr. */
12363 PerlIO_printf(Perl_debug_log, " (0)");
12364 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
12365 PerlIO_printf(Perl_debug_log, " (FAIL)");
12367 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
12368 (void)PerlIO_putc(Perl_debug_log, '\n');
12372 if (PL_regkind[(U8)op] == BRANCHJ) {
12375 register const regnode *nnode = (OP(next) == LONGJMP
12376 ? regnext((regnode *)next)
12378 if (last && nnode > last)
12380 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
12383 else if (PL_regkind[(U8)op] == BRANCH) {
12385 DUMPUNTIL(NEXTOPER(node), next);
12387 else if ( PL_regkind[(U8)op] == TRIE ) {
12388 const regnode *this_trie = node;
12389 const char op = OP(node);
12390 const U32 n = ARG(node);
12391 const reg_ac_data * const ac = op>=AHOCORASICK ?
12392 (reg_ac_data *)ri->data->data[n] :
12394 const reg_trie_data * const trie =
12395 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
12397 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
12399 const regnode *nextbranch= NULL;
12402 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
12403 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
12405 PerlIO_printf(Perl_debug_log, "%*s%s ",
12406 (int)(2*(indent+3)), "",
12407 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
12408 PL_colors[0], PL_colors[1],
12409 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
12410 PERL_PV_PRETTY_ELLIPSES |
12411 PERL_PV_PRETTY_LTGT
12416 U16 dist= trie->jump[word_idx+1];
12417 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
12418 (UV)((dist ? this_trie + dist : next) - start));
12421 nextbranch= this_trie + trie->jump[0];
12422 DUMPUNTIL(this_trie + dist, nextbranch);
12424 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
12425 nextbranch= regnext((regnode *)nextbranch);
12427 PerlIO_printf(Perl_debug_log, "\n");
12430 if (last && next > last)
12435 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
12436 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
12437 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
12439 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
12441 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
12443 else if ( op == PLUS || op == STAR) {
12444 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
12446 else if (PL_regkind[(U8)op] == ANYOF) {
12447 /* arglen 1 + class block */
12448 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS)
12449 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
12450 node = NEXTOPER(node);
12452 else if (PL_regkind[(U8)op] == EXACT) {
12453 /* Literal string, where present. */
12454 node += NODE_SZ_STR(node) - 1;
12455 node = NEXTOPER(node);
12458 node = NEXTOPER(node);
12459 node += regarglen[(U8)op];
12461 if (op == CURLYX || op == OPEN)
12465 #ifdef DEBUG_DUMPUNTIL
12466 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
12471 #endif /* DEBUGGING */
12475 * c-indentation-style: bsd
12476 * c-basic-offset: 4
12477 * indent-tabs-mode: t
12480 * ex: set ts=8 sts=4 sw=4 noet: