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 len = UTF8SKIP(uc);\
1391 uvc = to_utf8_fold( uc, 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 exp = SvPV(pattern, plen);
4558 if (plen == 0) { /* ignore the utf8ness if the pattern is 0 length */
4559 RExC_utf8 = RExC_orig_utf8 = 0;
4562 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4564 RExC_uni_semantics = 0;
4565 RExC_contains_locale = 0;
4567 /****************** LONG JUMP TARGET HERE***********************/
4568 /* Longjmp back to here if have to switch in midstream to utf8 */
4569 if (! RExC_orig_utf8) {
4570 JMPENV_PUSH(jump_ret);
4571 used_setjump = TRUE;
4574 if (jump_ret == 0) { /* First time through */
4578 SV *dsv= sv_newmortal();
4579 RE_PV_QUOTED_DECL(s, RExC_utf8,
4580 dsv, exp, plen, 60);
4581 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4582 PL_colors[4],PL_colors[5],s);
4585 else { /* longjumped back */
4588 /* If the cause for the longjmp was other than changing to utf8, pop
4589 * our own setjmp, and longjmp to the correct handler */
4590 if (jump_ret != UTF8_LONGJMP) {
4592 JMPENV_JUMP(jump_ret);
4597 /* It's possible to write a regexp in ascii that represents Unicode
4598 codepoints outside of the byte range, such as via \x{100}. If we
4599 detect such a sequence we have to convert the entire pattern to utf8
4600 and then recompile, as our sizing calculation will have been based
4601 on 1 byte == 1 character, but we will need to use utf8 to encode
4602 at least some part of the pattern, and therefore must convert the whole
4605 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4606 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4607 exp = (char*)Perl_bytes_to_utf8(aTHX_
4608 (U8*)SvPV_nomg(pattern, plen),
4611 RExC_orig_utf8 = RExC_utf8 = 1;
4615 #ifdef TRIE_STUDY_OPT
4619 pm_flags = orig_pm_flags;
4621 if (initial_charset == REGEX_LOCALE_CHARSET) {
4622 RExC_contains_locale = 1;
4624 else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) {
4626 /* Set to use unicode semantics if the pattern is in utf8 and has the
4627 * 'depends' charset specified, as it means unicode when utf8 */
4628 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4632 RExC_flags = pm_flags;
4636 RExC_in_lookbehind = 0;
4637 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4638 RExC_seen_evals = 0;
4640 RExC_override_recoding = 0;
4642 /* First pass: determine size, legality. */
4650 RExC_emit = &PL_regdummy;
4651 RExC_whilem_seen = 0;
4652 RExC_open_parens = NULL;
4653 RExC_close_parens = NULL;
4655 RExC_paren_names = NULL;
4657 RExC_paren_name_list = NULL;
4659 RExC_recurse = NULL;
4660 RExC_recurse_count = 0;
4662 #if 0 /* REGC() is (currently) a NOP at the first pass.
4663 * Clever compilers notice this and complain. --jhi */
4664 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4666 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4667 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4668 RExC_precomp = NULL;
4672 /* Here, finished first pass. Get rid of any added setjmp */
4678 PerlIO_printf(Perl_debug_log,
4679 "Required size %"IVdf" nodes\n"
4680 "Starting second pass (creation)\n",
4683 RExC_lastparse=NULL;
4686 /* The first pass could have found things that force Unicode semantics */
4687 if ((RExC_utf8 || RExC_uni_semantics)
4688 && get_regex_charset(pm_flags) == REGEX_DEPENDS_CHARSET)
4690 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4693 /* Small enough for pointer-storage convention?
4694 If extralen==0, this means that we will not need long jumps. */
4695 if (RExC_size >= 0x10000L && RExC_extralen)
4696 RExC_size += RExC_extralen;
4699 if (RExC_whilem_seen > 15)
4700 RExC_whilem_seen = 15;
4702 /* Allocate space and zero-initialize. Note, the two step process
4703 of zeroing when in debug mode, thus anything assigned has to
4704 happen after that */
4705 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4706 r = (struct regexp*)SvANY(rx);
4707 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4708 char, regexp_internal);
4709 if ( r == NULL || ri == NULL )
4710 FAIL("Regexp out of space");
4712 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4713 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4715 /* bulk initialize base fields with 0. */
4716 Zero(ri, sizeof(regexp_internal), char);
4719 /* non-zero initialization begins here */
4721 r->engine= RE_ENGINE_PTR;
4722 r->extflags = pm_flags;
4724 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4725 bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET);
4727 /* The caret is output if there are any defaults: if not all the STD
4728 * flags are set, or if no character set specifier is needed */
4730 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4732 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4733 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4734 >> RXf_PMf_STD_PMMOD_SHIFT);
4735 const char *fptr = STD_PAT_MODS; /*"msix"*/
4737 /* Allocate for the worst case, which is all the std flags are turned
4738 * on. If more precision is desired, we could do a population count of
4739 * the flags set. This could be done with a small lookup table, or by
4740 * shifting, masking and adding, or even, when available, assembly
4741 * language for a machine-language population count.
4742 * We never output a minus, as all those are defaults, so are
4743 * covered by the caret */
4744 const STRLEN wraplen = plen + has_p + has_runon
4745 + has_default /* If needs a caret */
4747 /* If needs a character set specifier */
4748 + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0)
4749 + (sizeof(STD_PAT_MODS) - 1)
4750 + (sizeof("(?:)") - 1);
4752 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4754 SvFLAGS(rx) |= SvUTF8(pattern);
4757 /* If a default, cover it using the caret */
4759 *p++= DEFAULT_PAT_MOD;
4763 const char* const name = get_regex_charset_name(r->extflags, &len);
4764 Copy(name, p, len, char);
4768 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4771 while((ch = *fptr++)) {
4779 Copy(RExC_precomp, p, plen, char);
4780 assert ((RX_WRAPPED(rx) - p) < 16);
4781 r->pre_prefix = p - RX_WRAPPED(rx);
4787 SvCUR_set(rx, p - SvPVX_const(rx));
4791 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4793 if (RExC_seen & REG_SEEN_RECURSE) {
4794 Newxz(RExC_open_parens, RExC_npar,regnode *);
4795 SAVEFREEPV(RExC_open_parens);
4796 Newxz(RExC_close_parens,RExC_npar,regnode *);
4797 SAVEFREEPV(RExC_close_parens);
4800 /* Useful during FAIL. */
4801 #ifdef RE_TRACK_PATTERN_OFFSETS
4802 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4803 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4804 "%s %"UVuf" bytes for offset annotations.\n",
4805 ri->u.offsets ? "Got" : "Couldn't get",
4806 (UV)((2*RExC_size+1) * sizeof(U32))));
4808 SetProgLen(ri,RExC_size);
4813 /* Second pass: emit code. */
4814 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4819 RExC_emit_start = ri->program;
4820 RExC_emit = ri->program;
4821 RExC_emit_bound = ri->program + RExC_size + 1;
4823 /* Store the count of eval-groups for security checks: */
4824 RExC_rx->seen_evals = RExC_seen_evals;
4825 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4826 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4830 /* XXXX To minimize changes to RE engine we always allocate
4831 3-units-long substrs field. */
4832 Newx(r->substrs, 1, struct reg_substr_data);
4833 if (RExC_recurse_count) {
4834 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4835 SAVEFREEPV(RExC_recurse);
4839 r->minlen = minlen = sawlookahead = sawplus = sawopen = 0;
4840 Zero(r->substrs, 1, struct reg_substr_data);
4842 #ifdef TRIE_STUDY_OPT
4844 StructCopy(&zero_scan_data, &data, scan_data_t);
4845 copyRExC_state = RExC_state;
4848 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4850 RExC_state = copyRExC_state;
4851 if (seen & REG_TOP_LEVEL_BRANCHES)
4852 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4854 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4855 if (data.last_found) {
4856 SvREFCNT_dec(data.longest_fixed);
4857 SvREFCNT_dec(data.longest_float);
4858 SvREFCNT_dec(data.last_found);
4860 StructCopy(&zero_scan_data, &data, scan_data_t);
4863 StructCopy(&zero_scan_data, &data, scan_data_t);
4866 /* Dig out information for optimizations. */
4867 r->extflags = RExC_flags; /* was pm_op */
4868 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4871 SvUTF8_on(rx); /* Unicode in it? */
4872 ri->regstclass = NULL;
4873 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4874 r->intflags |= PREGf_NAUGHTY;
4875 scan = ri->program + 1; /* First BRANCH. */
4877 /* testing for BRANCH here tells us whether there is "must appear"
4878 data in the pattern. If there is then we can use it for optimisations */
4879 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4881 STRLEN longest_float_length, longest_fixed_length;
4882 struct regnode_charclass_class ch_class; /* pointed to by data */
4884 I32 last_close = 0; /* pointed to by data */
4885 regnode *first= scan;
4886 regnode *first_next= regnext(first);
4888 * Skip introductions and multiplicators >= 1
4889 * so that we can extract the 'meat' of the pattern that must
4890 * match in the large if() sequence following.
4891 * NOTE that EXACT is NOT covered here, as it is normally
4892 * picked up by the optimiser separately.
4894 * This is unfortunate as the optimiser isnt handling lookahead
4895 * properly currently.
4898 while ((OP(first) == OPEN && (sawopen = 1)) ||
4899 /* An OR of *one* alternative - should not happen now. */
4900 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4901 /* for now we can't handle lookbehind IFMATCH*/
4902 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
4903 (OP(first) == PLUS) ||
4904 (OP(first) == MINMOD) ||
4905 /* An {n,m} with n>0 */
4906 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4907 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4910 * the only op that could be a regnode is PLUS, all the rest
4911 * will be regnode_1 or regnode_2.
4914 if (OP(first) == PLUS)
4917 first += regarglen[OP(first)];
4919 first = NEXTOPER(first);
4920 first_next= regnext(first);
4923 /* Starting-point info. */
4925 DEBUG_PEEP("first:",first,0);
4926 /* Ignore EXACT as we deal with it later. */
4927 if (PL_regkind[OP(first)] == EXACT) {
4928 if (OP(first) == EXACT)
4929 NOOP; /* Empty, get anchored substr later. */
4931 ri->regstclass = first;
4934 else if (PL_regkind[OP(first)] == TRIE &&
4935 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4938 /* this can happen only on restudy */
4939 if ( OP(first) == TRIE ) {
4940 struct regnode_1 *trieop = (struct regnode_1 *)
4941 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4942 StructCopy(first,trieop,struct regnode_1);
4943 trie_op=(regnode *)trieop;
4945 struct regnode_charclass *trieop = (struct regnode_charclass *)
4946 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4947 StructCopy(first,trieop,struct regnode_charclass);
4948 trie_op=(regnode *)trieop;
4951 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4952 ri->regstclass = trie_op;
4955 else if (REGNODE_SIMPLE(OP(first)))
4956 ri->regstclass = first;
4957 else if (PL_regkind[OP(first)] == BOUND ||
4958 PL_regkind[OP(first)] == NBOUND)
4959 ri->regstclass = first;
4960 else if (PL_regkind[OP(first)] == BOL) {
4961 r->extflags |= (OP(first) == MBOL
4963 : (OP(first) == SBOL
4966 first = NEXTOPER(first);
4969 else if (OP(first) == GPOS) {
4970 r->extflags |= RXf_ANCH_GPOS;
4971 first = NEXTOPER(first);
4974 else if ((!sawopen || !RExC_sawback) &&
4975 (OP(first) == STAR &&
4976 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4977 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4979 /* turn .* into ^.* with an implied $*=1 */
4981 (OP(NEXTOPER(first)) == REG_ANY)
4984 r->extflags |= type;
4985 r->intflags |= PREGf_IMPLICIT;
4986 first = NEXTOPER(first);
4989 if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback)
4990 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4991 /* x+ must match at the 1st pos of run of x's */
4992 r->intflags |= PREGf_SKIP;
4994 /* Scan is after the zeroth branch, first is atomic matcher. */
4995 #ifdef TRIE_STUDY_OPT
4998 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4999 (IV)(first - scan + 1))
5003 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
5004 (IV)(first - scan + 1))
5010 * If there's something expensive in the r.e., find the
5011 * longest literal string that must appear and make it the
5012 * regmust. Resolve ties in favor of later strings, since
5013 * the regstart check works with the beginning of the r.e.
5014 * and avoiding duplication strengthens checking. Not a
5015 * strong reason, but sufficient in the absence of others.
5016 * [Now we resolve ties in favor of the earlier string if
5017 * it happens that c_offset_min has been invalidated, since the
5018 * earlier string may buy us something the later one won't.]
5021 data.longest_fixed = newSVpvs("");
5022 data.longest_float = newSVpvs("");
5023 data.last_found = newSVpvs("");
5024 data.longest = &(data.longest_fixed);
5026 if (!ri->regstclass) {
5027 cl_init(pRExC_state, &ch_class);
5028 data.start_class = &ch_class;
5029 stclass_flag = SCF_DO_STCLASS_AND;
5030 } else /* XXXX Check for BOUND? */
5032 data.last_closep = &last_close;
5034 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
5035 &data, -1, NULL, NULL,
5036 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
5042 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
5043 && data.last_start_min == 0 && data.last_end > 0
5044 && !RExC_seen_zerolen
5045 && !(RExC_seen & REG_SEEN_VERBARG)
5046 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
5047 r->extflags |= RXf_CHECK_ALL;
5048 scan_commit(pRExC_state, &data,&minlen,0);
5049 SvREFCNT_dec(data.last_found);
5051 /* Note that code very similar to this but for anchored string
5052 follows immediately below, changes may need to be made to both.
5055 longest_float_length = CHR_SVLEN(data.longest_float);
5056 if (longest_float_length
5057 || (data.flags & SF_FL_BEFORE_EOL
5058 && (!(data.flags & SF_FL_BEFORE_MEOL)
5059 || (RExC_flags & RXf_PMf_MULTILINE))))
5063 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
5064 && data.offset_fixed == data.offset_float_min
5065 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
5066 goto remove_float; /* As in (a)+. */
5068 /* copy the information about the longest float from the reg_scan_data
5069 over to the program. */
5070 if (SvUTF8(data.longest_float)) {
5071 r->float_utf8 = data.longest_float;
5072 r->float_substr = NULL;
5074 r->float_substr = data.longest_float;
5075 r->float_utf8 = NULL;
5077 /* float_end_shift is how many chars that must be matched that
5078 follow this item. We calculate it ahead of time as once the
5079 lookbehind offset is added in we lose the ability to correctly
5081 ml = data.minlen_float ? *(data.minlen_float)
5082 : (I32)longest_float_length;
5083 r->float_end_shift = ml - data.offset_float_min
5084 - longest_float_length + (SvTAIL(data.longest_float) != 0)
5085 + data.lookbehind_float;
5086 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
5087 r->float_max_offset = data.offset_float_max;
5088 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
5089 r->float_max_offset -= data.lookbehind_float;
5091 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
5092 && (!(data.flags & SF_FL_BEFORE_MEOL)
5093 || (RExC_flags & RXf_PMf_MULTILINE)));
5094 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
5098 r->float_substr = r->float_utf8 = NULL;
5099 SvREFCNT_dec(data.longest_float);
5100 longest_float_length = 0;
5103 /* Note that code very similar to this but for floating string
5104 is immediately above, changes may need to be made to both.
5107 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
5108 if (longest_fixed_length
5109 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
5110 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5111 || (RExC_flags & RXf_PMf_MULTILINE))))
5115 /* copy the information about the longest fixed
5116 from the reg_scan_data over to the program. */
5117 if (SvUTF8(data.longest_fixed)) {
5118 r->anchored_utf8 = data.longest_fixed;
5119 r->anchored_substr = NULL;
5121 r->anchored_substr = data.longest_fixed;
5122 r->anchored_utf8 = NULL;
5124 /* fixed_end_shift is how many chars that must be matched that
5125 follow this item. We calculate it ahead of time as once the
5126 lookbehind offset is added in we lose the ability to correctly
5128 ml = data.minlen_fixed ? *(data.minlen_fixed)
5129 : (I32)longest_fixed_length;
5130 r->anchored_end_shift = ml - data.offset_fixed
5131 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
5132 + data.lookbehind_fixed;
5133 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
5135 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
5136 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5137 || (RExC_flags & RXf_PMf_MULTILINE)));
5138 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
5141 r->anchored_substr = r->anchored_utf8 = NULL;
5142 SvREFCNT_dec(data.longest_fixed);
5143 longest_fixed_length = 0;
5146 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
5147 ri->regstclass = NULL;
5149 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
5151 && !(data.start_class->flags & ANYOF_EOS)
5152 && !cl_is_anything(data.start_class))
5154 const U32 n = add_data(pRExC_state, 1, "f");
5155 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5157 Newx(RExC_rxi->data->data[n], 1,
5158 struct regnode_charclass_class);
5159 StructCopy(data.start_class,
5160 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5161 struct regnode_charclass_class);
5162 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5163 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5164 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
5165 regprop(r, sv, (regnode*)data.start_class);
5166 PerlIO_printf(Perl_debug_log,
5167 "synthetic stclass \"%s\".\n",
5168 SvPVX_const(sv));});
5171 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
5172 if (longest_fixed_length > longest_float_length) {
5173 r->check_end_shift = r->anchored_end_shift;
5174 r->check_substr = r->anchored_substr;
5175 r->check_utf8 = r->anchored_utf8;
5176 r->check_offset_min = r->check_offset_max = r->anchored_offset;
5177 if (r->extflags & RXf_ANCH_SINGLE)
5178 r->extflags |= RXf_NOSCAN;
5181 r->check_end_shift = r->float_end_shift;
5182 r->check_substr = r->float_substr;
5183 r->check_utf8 = r->float_utf8;
5184 r->check_offset_min = r->float_min_offset;
5185 r->check_offset_max = r->float_max_offset;
5187 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
5188 This should be changed ASAP! */
5189 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
5190 r->extflags |= RXf_USE_INTUIT;
5191 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
5192 r->extflags |= RXf_INTUIT_TAIL;
5194 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
5195 if ( (STRLEN)minlen < longest_float_length )
5196 minlen= longest_float_length;
5197 if ( (STRLEN)minlen < longest_fixed_length )
5198 minlen= longest_fixed_length;
5202 /* Several toplevels. Best we can is to set minlen. */
5204 struct regnode_charclass_class ch_class;
5207 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
5209 scan = ri->program + 1;
5210 cl_init(pRExC_state, &ch_class);
5211 data.start_class = &ch_class;
5212 data.last_closep = &last_close;
5215 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
5216 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
5220 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
5221 = r->float_substr = r->float_utf8 = NULL;
5223 if (!(data.start_class->flags & ANYOF_EOS)
5224 && !cl_is_anything(data.start_class))
5226 const U32 n = add_data(pRExC_state, 1, "f");
5227 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5229 Newx(RExC_rxi->data->data[n], 1,
5230 struct regnode_charclass_class);
5231 StructCopy(data.start_class,
5232 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5233 struct regnode_charclass_class);
5234 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5235 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5236 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
5237 regprop(r, sv, (regnode*)data.start_class);
5238 PerlIO_printf(Perl_debug_log,
5239 "synthetic stclass \"%s\".\n",
5240 SvPVX_const(sv));});
5244 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
5245 the "real" pattern. */
5247 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
5248 (IV)minlen, (IV)r->minlen);
5250 r->minlenret = minlen;
5251 if (r->minlen < minlen)
5254 if (RExC_seen & REG_SEEN_GPOS)
5255 r->extflags |= RXf_GPOS_SEEN;
5256 if (RExC_seen & REG_SEEN_LOOKBEHIND)
5257 r->extflags |= RXf_LOOKBEHIND_SEEN;
5258 if (RExC_seen & REG_SEEN_EVAL)
5259 r->extflags |= RXf_EVAL_SEEN;
5260 if (RExC_seen & REG_SEEN_CANY)
5261 r->extflags |= RXf_CANY_SEEN;
5262 if (RExC_seen & REG_SEEN_VERBARG)
5263 r->intflags |= PREGf_VERBARG_SEEN;
5264 if (RExC_seen & REG_SEEN_CUTGROUP)
5265 r->intflags |= PREGf_CUTGROUP_SEEN;
5266 if (RExC_paren_names)
5267 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
5269 RXp_PAREN_NAMES(r) = NULL;
5271 #ifdef STUPID_PATTERN_CHECKS
5272 if (RX_PRELEN(rx) == 0)
5273 r->extflags |= RXf_NULL;
5274 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5275 /* XXX: this should happen BEFORE we compile */
5276 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5277 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
5278 r->extflags |= RXf_WHITE;
5279 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
5280 r->extflags |= RXf_START_ONLY;
5282 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5283 /* XXX: this should happen BEFORE we compile */
5284 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5286 regnode *first = ri->program + 1;
5289 if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END)
5290 r->extflags |= RXf_NULL;
5291 else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END)
5292 r->extflags |= RXf_START_ONLY;
5293 else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE
5294 && OP(regnext(first)) == END)
5295 r->extflags |= RXf_WHITE;
5299 if (RExC_paren_names) {
5300 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5301 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5304 ri->name_list_idx = 0;
5306 if (RExC_recurse_count) {
5307 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5308 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5309 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5312 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5313 /* assume we don't need to swap parens around before we match */
5316 PerlIO_printf(Perl_debug_log,"Final program:\n");
5319 #ifdef RE_TRACK_PATTERN_OFFSETS
5320 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5321 const U32 len = ri->u.offsets[0];
5323 GET_RE_DEBUG_FLAGS_DECL;
5324 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5325 for (i = 1; i <= len; i++) {
5326 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5327 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5328 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5330 PerlIO_printf(Perl_debug_log, "\n");
5336 #undef RE_ENGINE_PTR
5340 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5343 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5345 PERL_UNUSED_ARG(value);
5347 if (flags & RXapif_FETCH) {
5348 return reg_named_buff_fetch(rx, key, flags);
5349 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5350 Perl_croak_no_modify(aTHX);
5352 } else if (flags & RXapif_EXISTS) {
5353 return reg_named_buff_exists(rx, key, flags)
5356 } else if (flags & RXapif_REGNAMES) {
5357 return reg_named_buff_all(rx, flags);
5358 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5359 return reg_named_buff_scalar(rx, flags);
5361 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5367 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5370 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5371 PERL_UNUSED_ARG(lastkey);
5373 if (flags & RXapif_FIRSTKEY)
5374 return reg_named_buff_firstkey(rx, flags);
5375 else if (flags & RXapif_NEXTKEY)
5376 return reg_named_buff_nextkey(rx, flags);
5378 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5384 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5387 AV *retarray = NULL;
5389 struct regexp *const rx = (struct regexp *)SvANY(r);
5391 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5393 if (flags & RXapif_ALL)
5396 if (rx && RXp_PAREN_NAMES(rx)) {
5397 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5400 SV* sv_dat=HeVAL(he_str);
5401 I32 *nums=(I32*)SvPVX(sv_dat);
5402 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5403 if ((I32)(rx->nparens) >= nums[i]
5404 && rx->offs[nums[i]].start != -1
5405 && rx->offs[nums[i]].end != -1)
5408 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5412 ret = newSVsv(&PL_sv_undef);
5415 av_push(retarray, ret);
5418 return newRV_noinc(MUTABLE_SV(retarray));
5425 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5428 struct regexp *const rx = (struct regexp *)SvANY(r);
5430 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5432 if (rx && RXp_PAREN_NAMES(rx)) {
5433 if (flags & RXapif_ALL) {
5434 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5436 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5450 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5452 struct regexp *const rx = (struct regexp *)SvANY(r);
5454 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5456 if ( rx && RXp_PAREN_NAMES(rx) ) {
5457 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5459 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5466 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5468 struct regexp *const rx = (struct regexp *)SvANY(r);
5469 GET_RE_DEBUG_FLAGS_DECL;
5471 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5473 if (rx && RXp_PAREN_NAMES(rx)) {
5474 HV *hv = RXp_PAREN_NAMES(rx);
5476 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5479 SV* sv_dat = HeVAL(temphe);
5480 I32 *nums = (I32*)SvPVX(sv_dat);
5481 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5482 if ((I32)(rx->lastparen) >= nums[i] &&
5483 rx->offs[nums[i]].start != -1 &&
5484 rx->offs[nums[i]].end != -1)
5490 if (parno || flags & RXapif_ALL) {
5491 return newSVhek(HeKEY_hek(temphe));
5499 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5504 struct regexp *const rx = (struct regexp *)SvANY(r);
5506 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5508 if (rx && RXp_PAREN_NAMES(rx)) {
5509 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5510 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5511 } else if (flags & RXapif_ONE) {
5512 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5513 av = MUTABLE_AV(SvRV(ret));
5514 length = av_len(av);
5516 return newSViv(length + 1);
5518 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5522 return &PL_sv_undef;
5526 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5528 struct regexp *const rx = (struct regexp *)SvANY(r);
5531 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5533 if (rx && RXp_PAREN_NAMES(rx)) {
5534 HV *hv= RXp_PAREN_NAMES(rx);
5536 (void)hv_iterinit(hv);
5537 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5540 SV* sv_dat = HeVAL(temphe);
5541 I32 *nums = (I32*)SvPVX(sv_dat);
5542 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5543 if ((I32)(rx->lastparen) >= nums[i] &&
5544 rx->offs[nums[i]].start != -1 &&
5545 rx->offs[nums[i]].end != -1)
5551 if (parno || flags & RXapif_ALL) {
5552 av_push(av, newSVhek(HeKEY_hek(temphe)));
5557 return newRV_noinc(MUTABLE_SV(av));
5561 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5564 struct regexp *const rx = (struct regexp *)SvANY(r);
5569 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5572 sv_setsv(sv,&PL_sv_undef);
5576 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5578 i = rx->offs[0].start;
5582 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5584 s = rx->subbeg + rx->offs[0].end;
5585 i = rx->sublen - rx->offs[0].end;
5588 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5589 (s1 = rx->offs[paren].start) != -1 &&
5590 (t1 = rx->offs[paren].end) != -1)
5594 s = rx->subbeg + s1;
5596 sv_setsv(sv,&PL_sv_undef);
5599 assert(rx->sublen >= (s - rx->subbeg) + i );
5601 const int oldtainted = PL_tainted;
5603 sv_setpvn(sv, s, i);
5604 PL_tainted = oldtainted;
5605 if ( (rx->extflags & RXf_CANY_SEEN)
5606 ? (RXp_MATCH_UTF8(rx)
5607 && (!i || is_utf8_string((U8*)s, i)))
5608 : (RXp_MATCH_UTF8(rx)) )
5615 if (RXp_MATCH_TAINTED(rx)) {
5616 if (SvTYPE(sv) >= SVt_PVMG) {
5617 MAGIC* const mg = SvMAGIC(sv);
5620 SvMAGIC_set(sv, mg->mg_moremagic);
5622 if ((mgt = SvMAGIC(sv))) {
5623 mg->mg_moremagic = mgt;
5624 SvMAGIC_set(sv, mg);
5634 sv_setsv(sv,&PL_sv_undef);
5640 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5641 SV const * const value)
5643 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5645 PERL_UNUSED_ARG(rx);
5646 PERL_UNUSED_ARG(paren);
5647 PERL_UNUSED_ARG(value);
5650 Perl_croak_no_modify(aTHX);
5654 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5657 struct regexp *const rx = (struct regexp *)SvANY(r);
5661 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5663 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5665 /* $` / ${^PREMATCH} */
5666 case RX_BUFF_IDX_PREMATCH:
5667 if (rx->offs[0].start != -1) {
5668 i = rx->offs[0].start;
5676 /* $' / ${^POSTMATCH} */
5677 case RX_BUFF_IDX_POSTMATCH:
5678 if (rx->offs[0].end != -1) {
5679 i = rx->sublen - rx->offs[0].end;
5681 s1 = rx->offs[0].end;
5687 /* $& / ${^MATCH}, $1, $2, ... */
5689 if (paren <= (I32)rx->nparens &&
5690 (s1 = rx->offs[paren].start) != -1 &&
5691 (t1 = rx->offs[paren].end) != -1)
5696 if (ckWARN(WARN_UNINITIALIZED))
5697 report_uninit((const SV *)sv);
5702 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5703 const char * const s = rx->subbeg + s1;
5708 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5715 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5717 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5718 PERL_UNUSED_ARG(rx);
5722 return newSVpvs("Regexp");
5725 /* Scans the name of a named buffer from the pattern.
5726 * If flags is REG_RSN_RETURN_NULL returns null.
5727 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5728 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5729 * to the parsed name as looked up in the RExC_paren_names hash.
5730 * If there is an error throws a vFAIL().. type exception.
5733 #define REG_RSN_RETURN_NULL 0
5734 #define REG_RSN_RETURN_NAME 1
5735 #define REG_RSN_RETURN_DATA 2
5738 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5740 char *name_start = RExC_parse;
5742 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5744 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5745 /* skip IDFIRST by using do...while */
5748 RExC_parse += UTF8SKIP(RExC_parse);
5749 } while (isALNUM_utf8((U8*)RExC_parse));
5753 } while (isALNUM(*RExC_parse));
5758 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5759 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5760 if ( flags == REG_RSN_RETURN_NAME)
5762 else if (flags==REG_RSN_RETURN_DATA) {
5765 if ( ! sv_name ) /* should not happen*/
5766 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5767 if (RExC_paren_names)
5768 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5770 sv_dat = HeVAL(he_str);
5772 vFAIL("Reference to nonexistent named group");
5776 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5783 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5784 int rem=(int)(RExC_end - RExC_parse); \
5793 if (RExC_lastparse!=RExC_parse) \
5794 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5797 iscut ? "..." : "<" \
5800 PerlIO_printf(Perl_debug_log,"%16s",""); \
5803 num = RExC_size + 1; \
5805 num=REG_NODE_NUM(RExC_emit); \
5806 if (RExC_lastnum!=num) \
5807 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5809 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5810 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5811 (int)((depth*2)), "", \
5815 RExC_lastparse=RExC_parse; \
5820 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5821 DEBUG_PARSE_MSG((funcname)); \
5822 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5824 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5825 DEBUG_PARSE_MSG((funcname)); \
5826 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5829 /* This section of code defines the inversion list object and its methods. The
5830 * interfaces are highly subject to change, so as much as possible is static to
5831 * this file. An inversion list is here implemented as a malloc'd C UV array
5832 * with some added info that is placed as UVs at the beginning in a header
5833 * portion. An inversion list for Unicode is an array of code points, sorted
5834 * by ordinal number. The zeroth element is the first code point in the list.
5835 * The 1th element is the first element beyond that not in the list. In other
5836 * words, the first range is
5837 * invlist[0]..(invlist[1]-1)
5838 * The other ranges follow. Thus every element that is divisible by two marks
5839 * the beginning of a range that is in the list, and every element not
5840 * divisible by two marks the beginning of a range not in the list. A single
5841 * element inversion list that contains the single code point N generally
5842 * consists of two elements
5845 * (The exception is when N is the highest representable value on the
5846 * machine, in which case the list containing just it would be a single
5847 * element, itself. By extension, if the last range in the list extends to
5848 * infinity, then the first element of that range will be in the inversion list
5849 * at a position that is divisible by two, and is the final element in the
5851 * Taking the complement (inverting) an inversion list is quite simple, if the
5852 * first element is 0, remove it; otherwise add a 0 element at the beginning.
5853 * This implementation reserves an element at the beginning of each inversion list
5854 * to contain 0 when the list contains 0, and contains 1 otherwise. The actual
5855 * beginning of the list is either that element if 0, or the next one if 1.
5857 * More about inversion lists can be found in "Unicode Demystified"
5858 * Chapter 13 by Richard Gillam, published by Addison-Wesley.
5859 * More will be coming when functionality is added later.
5861 * The inversion list data structure is currently implemented as an SV pointing
5862 * to an array of UVs that the SV thinks are bytes. This allows us to have an
5863 * array of UV whose memory management is automatically handled by the existing
5864 * facilities for SV's.
5866 * Some of the methods should always be private to the implementation, and some
5867 * should eventually be made public */
5869 #define INVLIST_LEN_OFFSET 0 /* Number of elements in the inversion list */
5870 #define INVLIST_ITER_OFFSET 1 /* Current iteration position */
5872 #define INVLIST_ZERO_OFFSET 2 /* 0 or 1; must be last element in header */
5873 /* The UV at position ZERO contains either 0 or 1. If 0, the inversion list
5874 * contains the code point U+00000, and begins here. If 1, the inversion list
5875 * doesn't contain U+0000, and it begins at the next UV in the array.
5876 * Inverting an inversion list consists of adding or removing the 0 at the
5877 * beginning of it. By reserving a space for that 0, inversion can be made
5880 #define HEADER_LENGTH (INVLIST_ZERO_OFFSET + 1)
5882 /* Internally things are UVs */
5883 #define TO_INTERNAL_SIZE(x) ((x + HEADER_LENGTH) * sizeof(UV))
5884 #define FROM_INTERNAL_SIZE(x) ((x / sizeof(UV)) - HEADER_LENGTH)
5886 #define INVLIST_INITIAL_LEN 10
5888 PERL_STATIC_INLINE UV*
5889 S__invlist_array_init(pTHX_ SV* const invlist, const bool will_have_0)
5891 /* Returns a pointer to the first element in the inversion list's array.
5892 * This is called upon initialization of an inversion list. Where the
5893 * array begins depends on whether the list has the code point U+0000
5894 * in it or not. The other parameter tells it whether the code that
5895 * follows this call is about to put a 0 in the inversion list or not.
5896 * The first element is either the element with 0, if 0, or the next one,
5899 UV* zero = get_invlist_zero_addr(invlist);
5901 PERL_ARGS_ASSERT__INVLIST_ARRAY_INIT;
5904 assert(! *get_invlist_len_addr(invlist));
5906 /* 1^1 = 0; 1^0 = 1 */
5907 *zero = 1 ^ will_have_0;
5908 return zero + *zero;
5911 PERL_STATIC_INLINE UV*
5912 S_invlist_array(pTHX_ SV* const invlist)
5914 /* Returns the pointer to the inversion list's array. Every time the
5915 * length changes, this needs to be called in case malloc or realloc moved
5918 PERL_ARGS_ASSERT_INVLIST_ARRAY;
5920 /* Must not be empty */
5921 assert(*get_invlist_len_addr(invlist));
5922 assert(*get_invlist_zero_addr(invlist) == 0
5923 || *get_invlist_zero_addr(invlist) == 1);
5925 /* The array begins either at the element reserved for zero if the
5926 * list contains 0 (that element will be set to 0), or otherwise the next
5927 * element (in which case the reserved element will be set to 1). */
5928 return (UV *) (get_invlist_zero_addr(invlist)
5929 + *get_invlist_zero_addr(invlist));
5932 PERL_STATIC_INLINE UV*
5933 S_get_invlist_len_addr(pTHX_ SV* invlist)
5935 /* Return the address of the UV that contains the current number
5936 * of used elements in the inversion list */
5938 PERL_ARGS_ASSERT_GET_INVLIST_LEN_ADDR;
5940 return (UV *) (SvPVX(invlist) + (INVLIST_LEN_OFFSET * sizeof (UV)));
5943 PERL_STATIC_INLINE UV
5944 S_invlist_len(pTHX_ SV* const invlist)
5946 /* Returns the current number of elements in the inversion list's array */
5948 PERL_ARGS_ASSERT_INVLIST_LEN;
5950 return *get_invlist_len_addr(invlist);
5953 PERL_STATIC_INLINE void
5954 S_invlist_set_len(pTHX_ SV* const invlist, const UV len)
5956 /* Sets the current number of elements stored in the inversion list */
5958 PERL_ARGS_ASSERT_INVLIST_SET_LEN;
5960 *get_invlist_len_addr(invlist) = len;
5962 assert(len <= SvLEN(invlist));
5964 SvCUR_set(invlist, TO_INTERNAL_SIZE(len));
5965 /* If the list contains U+0000, that element is part of the header,
5966 * and should not be counted as part of the array. It will contain
5967 * 0 in that case, and 1 otherwise. So we could flop 0=>1, 1=>0 and
5969 * SvCUR_set(invlist,
5970 * TO_INTERNAL_SIZE(len
5971 * - (*get_invlist_zero_addr(inv_list) ^ 1)));
5972 * But, this is only valid if len is not 0. The consequences of not doing
5973 * this is that the memory allocation code may think that 1 more UV is
5974 * being used than actually is, and so might do an unnecessary grow. That
5975 * seems worth not bothering to make this the precise amount.
5977 * Note that when inverting, SvCUR shouldn't change */
5980 PERL_STATIC_INLINE UV
5981 S_invlist_max(pTHX_ SV* const invlist)
5983 /* Returns the maximum number of elements storable in the inversion list's
5984 * array, without having to realloc() */
5986 PERL_ARGS_ASSERT_INVLIST_MAX;
5988 return FROM_INTERNAL_SIZE(SvLEN(invlist));
5991 PERL_STATIC_INLINE UV*
5992 S_get_invlist_zero_addr(pTHX_ SV* invlist)
5994 /* Return the address of the UV that is reserved to hold 0 if the inversion
5995 * list contains 0. This has to be the last element of the heading, as the
5996 * list proper starts with either it if 0, or the next element if not.
5997 * (But we force it to contain either 0 or 1) */
5999 PERL_ARGS_ASSERT_GET_INVLIST_ZERO_ADDR;
6001 return (UV *) (SvPVX(invlist) + (INVLIST_ZERO_OFFSET * sizeof (UV)));
6004 #ifndef PERL_IN_XSUB_RE
6006 Perl__new_invlist(pTHX_ IV initial_size)
6009 /* Return a pointer to a newly constructed inversion list, with enough
6010 * space to store 'initial_size' elements. If that number is negative, a
6011 * system default is used instead */
6015 if (initial_size < 0) {
6016 initial_size = INVLIST_INITIAL_LEN;
6019 /* Allocate the initial space */
6020 new_list = newSV(TO_INTERNAL_SIZE(initial_size));
6021 invlist_set_len(new_list, 0);
6023 /* Force iterinit() to be used to get iteration to work */
6024 *get_invlist_iter_addr(new_list) = UV_MAX;
6026 /* This should force a segfault if a method doesn't initialize this
6028 *get_invlist_zero_addr(new_list) = UV_MAX;
6035 S_invlist_extend(pTHX_ SV* const invlist, const UV new_max)
6037 /* Grow the maximum size of an inversion list */
6039 PERL_ARGS_ASSERT_INVLIST_EXTEND;
6041 SvGROW((SV *)invlist, TO_INTERNAL_SIZE(new_max));
6044 PERL_STATIC_INLINE void
6045 S_invlist_trim(pTHX_ SV* const invlist)
6047 PERL_ARGS_ASSERT_INVLIST_TRIM;
6049 /* Change the length of the inversion list to how many entries it currently
6052 SvPV_shrink_to_cur((SV *) invlist);
6055 /* An element is in an inversion list iff its index is even numbered: 0, 2, 4,
6058 #define ELEMENT_IN_INVLIST_SET(i) (! ((i) & 1))
6059 #define PREV_ELEMENT_IN_INVLIST_SET(i) (! ELEMENT_IN_INVLIST_SET(i))
6061 #ifndef PERL_IN_XSUB_RE
6063 Perl__append_range_to_invlist(pTHX_ SV* const invlist, const UV start, const UV end)
6065 /* Subject to change or removal. Append the range from 'start' to 'end' at
6066 * the end of the inversion list. The range must be above any existing
6070 UV max = invlist_max(invlist);
6071 UV len = invlist_len(invlist);
6073 PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST;
6075 if (len == 0) { /* Empty lists must be initialized */
6076 array = _invlist_array_init(invlist, start == 0);
6079 /* Here, the existing list is non-empty. The current max entry in the
6080 * list is generally the first value not in the set, except when the
6081 * set extends to the end of permissible values, in which case it is
6082 * the first entry in that final set, and so this call is an attempt to
6083 * append out-of-order */
6085 UV final_element = len - 1;
6086 array = invlist_array(invlist);
6087 if (array[final_element] > start
6088 || ELEMENT_IN_INVLIST_SET(final_element))
6090 Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list");
6093 /* Here, it is a legal append. If the new range begins with the first
6094 * value not in the set, it is extending the set, so the new first
6095 * value not in the set is one greater than the newly extended range.
6097 if (array[final_element] == start) {
6098 if (end != UV_MAX) {
6099 array[final_element] = end + 1;
6102 /* But if the end is the maximum representable on the machine,
6103 * just let the range that this would extend have no end */
6104 invlist_set_len(invlist, len - 1);
6110 /* Here the new range doesn't extend any existing set. Add it */
6112 len += 2; /* Includes an element each for the start and end of range */
6114 /* If overflows the existing space, extend, which may cause the array to be
6117 invlist_extend(invlist, len);
6118 invlist_set_len(invlist, len); /* Have to set len here to avoid assert
6119 failure in invlist_array() */
6120 array = invlist_array(invlist);
6123 invlist_set_len(invlist, len);
6126 /* The next item on the list starts the range, the one after that is
6127 * one past the new range. */
6128 array[len - 2] = start;
6129 if (end != UV_MAX) {
6130 array[len - 1] = end + 1;
6133 /* But if the end is the maximum representable on the machine, just let
6134 * the range have no end */
6135 invlist_set_len(invlist, len - 1);
6140 Perl__invlist_union(pTHX_ SV* const a, SV* const b, SV** output)
6142 /* Take the union of two inversion lists and point 'result' to it. If
6143 * 'result' on input points to one of the two lists, the reference count to
6144 * that list will be decremented.
6145 * The basis for this comes from "Unicode Demystified" Chapter 13 by
6146 * Richard Gillam, published by Addison-Wesley, and explained at some
6147 * length there. The preface says to incorporate its examples into your
6148 * code at your own risk.
6150 * The algorithm is like a merge sort.
6152 * XXX A potential performance improvement is to keep track as we go along
6153 * if only one of the inputs contributes to the result, meaning the other
6154 * is a subset of that one. In that case, we can skip the final copy and
6155 * return the larger of the input lists, but then outside code might need
6156 * to keep track of whether to free the input list or not */
6158 UV* array_a; /* a's array */
6160 UV len_a; /* length of a's array */
6163 SV* u; /* the resulting union */
6167 UV i_a = 0; /* current index into a's array */
6171 /* running count, as explained in the algorithm source book; items are
6172 * stopped accumulating and are output when the count changes to/from 0.
6173 * The count is incremented when we start a range that's in the set, and
6174 * decremented when we start a range that's not in the set. So its range
6175 * is 0 to 2. Only when the count is zero is something not in the set.
6179 PERL_ARGS_ASSERT__INVLIST_UNION;
6181 /* If either one is empty, the union is the other one */
6182 len_a = invlist_len(a);
6187 else if (output != &b) {
6188 *output = invlist_clone(b);
6190 /* else *output already = b; */
6193 else if ((len_b = invlist_len(b)) == 0) {
6197 else if (output != &a) {
6198 *output = invlist_clone(a);
6200 /* else *output already = a; */
6204 /* Here both lists exist and are non-empty */
6205 array_a = invlist_array(a);
6206 array_b = invlist_array(b);
6208 /* Size the union for the worst case: that the sets are completely
6210 u = _new_invlist(len_a + len_b);
6212 /* Will contain U+0000 if either component does */
6213 array_u = _invlist_array_init(u, (len_a > 0 && array_a[0] == 0)
6214 || (len_b > 0 && array_b[0] == 0));
6216 /* Go through each list item by item, stopping when exhausted one of
6218 while (i_a < len_a && i_b < len_b) {
6219 UV cp; /* The element to potentially add to the union's array */
6220 bool cp_in_set; /* is it in the the input list's set or not */
6222 /* We need to take one or the other of the two inputs for the union.
6223 * Since we are merging two sorted lists, we take the smaller of the
6224 * next items. In case of a tie, we take the one that is in its set
6225 * first. If we took one not in the set first, it would decrement the
6226 * count, possibly to 0 which would cause it to be output as ending the
6227 * range, and the next time through we would take the same number, and
6228 * output it again as beginning the next range. By doing it the
6229 * opposite way, there is no possibility that the count will be
6230 * momentarily decremented to 0, and thus the two adjoining ranges will
6231 * be seamlessly merged. (In a tie and both are in the set or both not
6232 * in the set, it doesn't matter which we take first.) */
6233 if (array_a[i_a] < array_b[i_b]
6234 || (array_a[i_a] == array_b[i_b] && ELEMENT_IN_INVLIST_SET(i_a)))
6236 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6240 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6244 /* Here, have chosen which of the two inputs to look at. Only output
6245 * if the running count changes to/from 0, which marks the
6246 * beginning/end of a range in that's in the set */
6249 array_u[i_u++] = cp;
6256 array_u[i_u++] = cp;
6261 /* Here, we are finished going through at least one of the lists, which
6262 * means there is something remaining in at most one. We check if the list
6263 * that hasn't been exhausted is positioned such that we are in the middle
6264 * of a range in its set or not. (i_a and i_b point to the element beyond
6265 * the one we care about.) If in the set, we decrement 'count'; if 0, there
6266 * is potentially more to output.
6267 * There are four cases:
6268 * 1) Both weren't in their sets, count is 0, and remains 0. What's left
6269 * in the union is entirely from the non-exhausted set.
6270 * 2) Both were in their sets, count is 2. Nothing further should
6271 * be output, as everything that remains will be in the exhausted
6272 * list's set, hence in the union; decrementing to 1 but not 0 insures
6274 * 3) the exhausted was in its set, non-exhausted isn't, count is 1.
6275 * Nothing further should be output because the union includes
6276 * everything from the exhausted set. Not decrementing ensures that.
6277 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1;
6278 * decrementing to 0 insures that we look at the remainder of the
6279 * non-exhausted set */
6280 if ((i_a != len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6281 || (i_b != len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6286 /* The final length is what we've output so far, plus what else is about to
6287 * be output. (If 'count' is non-zero, then the input list we exhausted
6288 * has everything remaining up to the machine's limit in its set, and hence
6289 * in the union, so there will be no further output. */
6292 /* At most one of the subexpressions will be non-zero */
6293 len_u += (len_a - i_a) + (len_b - i_b);
6296 /* Set result to final length, which can change the pointer to array_u, so
6298 if (len_u != invlist_len(u)) {
6299 invlist_set_len(u, len_u);
6301 array_u = invlist_array(u);
6304 /* When 'count' is 0, the list that was exhausted (if one was shorter than
6305 * the other) ended with everything above it not in its set. That means
6306 * that the remaining part of the union is precisely the same as the
6307 * non-exhausted list, so can just copy it unchanged. (If both list were
6308 * exhausted at the same time, then the operations below will be both 0.)
6311 IV copy_count; /* At most one will have a non-zero copy count */
6312 if ((copy_count = len_a - i_a) > 0) {
6313 Copy(array_a + i_a, array_u + i_u, copy_count, UV);
6315 else if ((copy_count = len_b - i_b) > 0) {
6316 Copy(array_b + i_b, array_u + i_u, copy_count, UV);
6320 /* We may be removing a reference to one of the inputs */
6321 if (&a == output || &b == output) {
6322 SvREFCNT_dec(*output);
6330 Perl__invlist_intersection(pTHX_ SV* const a, SV* const b, SV** i)
6332 /* Take the intersection of two inversion lists and point 'i' to it. If
6333 * 'i' on input points to one of the two lists, the reference count to that
6334 * list will be decremented.
6335 * The basis for this comes from "Unicode Demystified" Chapter 13 by
6336 * Richard Gillam, published by Addison-Wesley, and explained at some
6337 * length there. The preface says to incorporate its examples into your
6338 * code at your own risk. In fact, it had bugs
6340 * The algorithm is like a merge sort, and is essentially the same as the
6344 UV* array_a; /* a's array */
6346 UV len_a; /* length of a's array */
6349 SV* r; /* the resulting intersection */
6353 UV i_a = 0; /* current index into a's array */
6357 /* running count, as explained in the algorithm source book; items are
6358 * stopped accumulating and are output when the count changes to/from 2.
6359 * The count is incremented when we start a range that's in the set, and
6360 * decremented when we start a range that's not in the set. So its range
6361 * is 0 to 2. Only when the count is 2 is something in the intersection.
6365 PERL_ARGS_ASSERT__INVLIST_INTERSECTION;
6367 /* If either one is empty, the intersection is null */
6368 len_a = invlist_len(a);
6369 if ((len_a == 0) || ((len_b = invlist_len(b)) == 0)) {
6370 *i = _new_invlist(0);
6372 /* If the result is the same as one of the inputs, the input is being
6383 /* Here both lists exist and are non-empty */
6384 array_a = invlist_array(a);
6385 array_b = invlist_array(b);
6387 /* Size the intersection for the worst case: that the intersection ends up
6388 * fragmenting everything to be completely disjoint */
6389 r= _new_invlist(len_a + len_b);
6391 /* Will contain U+0000 iff both components do */
6392 array_r = _invlist_array_init(r, len_a > 0 && array_a[0] == 0
6393 && len_b > 0 && array_b[0] == 0);
6395 /* Go through each list item by item, stopping when exhausted one of
6397 while (i_a < len_a && i_b < len_b) {
6398 UV cp; /* The element to potentially add to the intersection's
6400 bool cp_in_set; /* Is it in the input list's set or not */
6402 /* We need to take one or the other of the two inputs for the
6403 * intersection. Since we are merging two sorted lists, we take the
6404 * smaller of the next items. In case of a tie, we take the one that
6405 * is not in its set first (a difference from the union algorithm). If
6406 * we took one in the set first, it would increment the count, possibly
6407 * to 2 which would cause it to be output as starting a range in the
6408 * intersection, and the next time through we would take that same
6409 * number, and output it again as ending the set. By doing it the
6410 * opposite of this, there is no possibility that the count will be
6411 * momentarily incremented to 2. (In a tie and both are in the set or
6412 * both not in the set, it doesn't matter which we take first.) */
6413 if (array_a[i_a] < array_b[i_b]
6414 || (array_a[i_a] == array_b[i_b] && ! ELEMENT_IN_INVLIST_SET(i_a)))
6416 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6420 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6424 /* Here, have chosen which of the two inputs to look at. Only output
6425 * if the running count changes to/from 2, which marks the
6426 * beginning/end of a range that's in the intersection */
6430 array_r[i_r++] = cp;
6435 array_r[i_r++] = cp;
6441 /* Here, we are finished going through at least one of the lists, which
6442 * means there is something remaining in at most one. We check if the list
6443 * that has been exhausted is positioned such that we are in the middle
6444 * of a range in its set or not. (i_a and i_b point to elements 1 beyond
6445 * the ones we care about.) There are four cases:
6446 * 1) Both weren't in their sets, count is 0, and remains 0. There's
6447 * nothing left in the intersection.
6448 * 2) Both were in their sets, count is 2 and perhaps is incremented to
6449 * above 2. What should be output is exactly that which is in the
6450 * non-exhausted set, as everything it has is also in the intersection
6451 * set, and everything it doesn't have can't be in the intersection
6452 * 3) The exhausted was in its set, non-exhausted isn't, count is 1, and
6453 * gets incremented to 2. Like the previous case, the intersection is
6454 * everything that remains in the non-exhausted set.
6455 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1, and
6456 * remains 1. And the intersection has nothing more. */
6457 if ((i_a == len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6458 || (i_b == len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6463 /* The final length is what we've output so far plus what else is in the
6464 * intersection. At most one of the subexpressions below will be non-zero */
6467 len_r += (len_a - i_a) + (len_b - i_b);
6470 /* Set result to final length, which can change the pointer to array_r, so
6472 if (len_r != invlist_len(r)) {
6473 invlist_set_len(r, len_r);
6475 array_r = invlist_array(r);
6478 /* Finish outputting any remaining */
6479 if (count >= 2) { /* At most one will have a non-zero copy count */
6481 if ((copy_count = len_a - i_a) > 0) {
6482 Copy(array_a + i_a, array_r + i_r, copy_count, UV);
6484 else if ((copy_count = len_b - i_b) > 0) {
6485 Copy(array_b + i_b, array_r + i_r, copy_count, UV);
6489 /* We may be removing a reference to one of the inputs */
6490 if (&a == i || &b == i) {
6501 S_add_range_to_invlist(pTHX_ SV* invlist, const UV start, const UV end)
6503 /* Add the range from 'start' to 'end' inclusive to the inversion list's
6504 * set. A pointer to the inversion list is returned. This may actually be
6505 * a new list, in which case the passed in one has been destroyed. The
6506 * passed in inversion list can be NULL, in which case a new one is created
6507 * with just the one range in it */
6512 if (invlist == NULL) {
6513 invlist = _new_invlist(2);
6517 len = invlist_len(invlist);
6520 /* If comes after the final entry, can just append it to the end */
6522 || start >= invlist_array(invlist)
6523 [invlist_len(invlist) - 1])
6525 _append_range_to_invlist(invlist, start, end);
6529 /* Here, can't just append things, create and return a new inversion list
6530 * which is the union of this range and the existing inversion list */
6531 range_invlist = _new_invlist(2);
6532 _append_range_to_invlist(range_invlist, start, end);
6534 _invlist_union(invlist, range_invlist, &invlist);
6536 /* The temporary can be freed */
6537 SvREFCNT_dec(range_invlist);
6542 PERL_STATIC_INLINE SV*
6543 S_add_cp_to_invlist(pTHX_ SV* invlist, const UV cp) {
6544 return add_range_to_invlist(invlist, cp, cp);
6547 #ifndef PERL_IN_XSUB_RE
6549 Perl__invlist_invert(pTHX_ SV* const invlist)
6551 /* Complement the input inversion list. This adds a 0 if the list didn't
6552 * have a zero; removes it otherwise. As described above, the data
6553 * structure is set up so that this is very efficient */
6555 UV* len_pos = get_invlist_len_addr(invlist);
6557 PERL_ARGS_ASSERT__INVLIST_INVERT;
6559 /* The inverse of matching nothing is matching everything */
6560 if (*len_pos == 0) {
6561 _append_range_to_invlist(invlist, 0, UV_MAX);
6565 /* The exclusive or complents 0 to 1; and 1 to 0. If the result is 1, the
6566 * zero element was a 0, so it is being removed, so the length decrements
6567 * by 1; and vice-versa. SvCUR is unaffected */
6568 if (*get_invlist_zero_addr(invlist) ^= 1) {
6577 Perl__invlist_invert_prop(pTHX_ SV* const invlist)
6579 /* Complement the input inversion list (which must be a Unicode property,
6580 * all of which don't match above the Unicode maximum code point.) And
6581 * Perl has chosen to not have the inversion match above that either. This
6582 * adds a 0x110000 if the list didn't end with it, and removes it if it did
6588 PERL_ARGS_ASSERT__INVLIST_INVERT_PROP;
6590 _invlist_invert(invlist);
6592 len = invlist_len(invlist);
6594 if (len != 0) { /* If empty do nothing */
6595 array = invlist_array(invlist);
6596 if (array[len - 1] != PERL_UNICODE_MAX + 1) {
6597 /* Add 0x110000. First, grow if necessary */
6599 if (invlist_max(invlist) < len) {
6600 invlist_extend(invlist, len);
6601 array = invlist_array(invlist);
6603 invlist_set_len(invlist, len);
6604 array[len - 1] = PERL_UNICODE_MAX + 1;
6606 else { /* Remove the 0x110000 */
6607 invlist_set_len(invlist, len - 1);
6615 PERL_STATIC_INLINE SV*
6616 S_invlist_clone(pTHX_ SV* const invlist)
6619 /* Return a new inversion list that is a copy of the input one, which is
6622 SV* new_invlist = _new_invlist(SvCUR(invlist));
6624 PERL_ARGS_ASSERT_INVLIST_CLONE;
6626 Copy(SvPVX(invlist), SvPVX(new_invlist), SvCUR(invlist), char);
6630 #ifndef PERL_IN_XSUB_RE
6632 Perl__invlist_subtract(pTHX_ SV* const a, SV* const b, SV** result)
6634 /* Point result to an inversion list which consists of all elements in 'a'
6635 * that aren't also in 'b' */
6637 PERL_ARGS_ASSERT__INVLIST_SUBTRACT;
6639 /* Subtracting nothing retains the original */
6640 if (invlist_len(b) == 0) {
6642 /* If the result is not to be the same variable as the original, create
6645 *result = invlist_clone(a);
6648 SV *b_copy = invlist_clone(b);
6649 _invlist_invert(b_copy); /* Everything not in 'b' */
6650 _invlist_intersection(a, b_copy, result); /* Everything in 'a' not in
6652 SvREFCNT_dec(b_copy);
6663 PERL_STATIC_INLINE UV*
6664 S_get_invlist_iter_addr(pTHX_ SV* invlist)
6666 /* Return the address of the UV that contains the current iteration
6669 PERL_ARGS_ASSERT_GET_INVLIST_ITER_ADDR;
6671 return (UV *) (SvPVX(invlist) + (INVLIST_ITER_OFFSET * sizeof (UV)));
6674 PERL_STATIC_INLINE void
6675 S_invlist_iterinit(pTHX_ SV* invlist) /* Initialize iterator for invlist */
6677 PERL_ARGS_ASSERT_INVLIST_ITERINIT;
6679 *get_invlist_iter_addr(invlist) = 0;
6683 S_invlist_iternext(pTHX_ SV* invlist, UV* start, UV* end)
6685 UV* pos = get_invlist_iter_addr(invlist);
6686 UV len = invlist_len(invlist);
6689 PERL_ARGS_ASSERT_INVLIST_ITERNEXT;
6692 *pos = UV_MAX; /* Force iternit() to be required next time */
6696 array = invlist_array(invlist);
6698 *start = array[(*pos)++];
6704 *end = array[(*pos)++] - 1;
6712 S_invlist_dump(pTHX_ SV* const invlist, const char * const header)
6714 /* Dumps out the ranges in an inversion list. The string 'header'
6715 * if present is output on a line before the first range */
6719 if (header && strlen(header)) {
6720 PerlIO_printf(Perl_debug_log, "%s\n", header);
6722 invlist_iterinit(invlist);
6723 while (invlist_iternext(invlist, &start, &end)) {
6724 if (end == UV_MAX) {
6725 PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. INFINITY\n", start);
6728 PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. 0x%04"UVXf"\n", start, end);
6734 #undef HEADER_LENGTH
6735 #undef INVLIST_INITIAL_LENGTH
6736 #undef TO_INTERNAL_SIZE
6737 #undef FROM_INTERNAL_SIZE
6738 #undef INVLIST_LEN_OFFSET
6739 #undef INVLIST_ZERO_OFFSET
6740 #undef INVLIST_ITER_OFFSET
6742 /* End of inversion list object */
6745 - reg - regular expression, i.e. main body or parenthesized thing
6747 * Caller must absorb opening parenthesis.
6749 * Combining parenthesis handling with the base level of regular expression
6750 * is a trifle forced, but the need to tie the tails of the branches to what
6751 * follows makes it hard to avoid.
6753 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
6755 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
6757 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
6761 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
6762 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
6765 register regnode *ret; /* Will be the head of the group. */
6766 register regnode *br;
6767 register regnode *lastbr;
6768 register regnode *ender = NULL;
6769 register I32 parno = 0;
6771 U32 oregflags = RExC_flags;
6772 bool have_branch = 0;
6774 I32 freeze_paren = 0;
6775 I32 after_freeze = 0;
6777 /* for (?g), (?gc), and (?o) warnings; warning
6778 about (?c) will warn about (?g) -- japhy */
6780 #define WASTED_O 0x01
6781 #define WASTED_G 0x02
6782 #define WASTED_C 0x04
6783 #define WASTED_GC (0x02|0x04)
6784 I32 wastedflags = 0x00;
6786 char * parse_start = RExC_parse; /* MJD */
6787 char * const oregcomp_parse = RExC_parse;
6789 GET_RE_DEBUG_FLAGS_DECL;
6791 PERL_ARGS_ASSERT_REG;
6792 DEBUG_PARSE("reg ");
6794 *flagp = 0; /* Tentatively. */
6797 /* Make an OPEN node, if parenthesized. */
6799 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
6800 char *start_verb = RExC_parse;
6801 STRLEN verb_len = 0;
6802 char *start_arg = NULL;
6803 unsigned char op = 0;
6805 int internal_argval = 0; /* internal_argval is only useful if !argok */
6806 while ( *RExC_parse && *RExC_parse != ')' ) {
6807 if ( *RExC_parse == ':' ) {
6808 start_arg = RExC_parse + 1;
6814 verb_len = RExC_parse - start_verb;
6817 while ( *RExC_parse && *RExC_parse != ')' )
6819 if ( *RExC_parse != ')' )
6820 vFAIL("Unterminated verb pattern argument");
6821 if ( RExC_parse == start_arg )
6824 if ( *RExC_parse != ')' )
6825 vFAIL("Unterminated verb pattern");
6828 switch ( *start_verb ) {
6829 case 'A': /* (*ACCEPT) */
6830 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
6832 internal_argval = RExC_nestroot;
6835 case 'C': /* (*COMMIT) */
6836 if ( memEQs(start_verb,verb_len,"COMMIT") )
6839 case 'F': /* (*FAIL) */
6840 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
6845 case ':': /* (*:NAME) */
6846 case 'M': /* (*MARK:NAME) */
6847 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
6852 case 'P': /* (*PRUNE) */
6853 if ( memEQs(start_verb,verb_len,"PRUNE") )
6856 case 'S': /* (*SKIP) */
6857 if ( memEQs(start_verb,verb_len,"SKIP") )
6860 case 'T': /* (*THEN) */
6861 /* [19:06] <TimToady> :: is then */
6862 if ( memEQs(start_verb,verb_len,"THEN") ) {
6864 RExC_seen |= REG_SEEN_CUTGROUP;
6870 vFAIL3("Unknown verb pattern '%.*s'",
6871 verb_len, start_verb);
6874 if ( start_arg && internal_argval ) {
6875 vFAIL3("Verb pattern '%.*s' may not have an argument",
6876 verb_len, start_verb);
6877 } else if ( argok < 0 && !start_arg ) {
6878 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
6879 verb_len, start_verb);
6881 ret = reganode(pRExC_state, op, internal_argval);
6882 if ( ! internal_argval && ! SIZE_ONLY ) {
6884 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
6885 ARG(ret) = add_data( pRExC_state, 1, "S" );
6886 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
6893 if (!internal_argval)
6894 RExC_seen |= REG_SEEN_VERBARG;
6895 } else if ( start_arg ) {
6896 vFAIL3("Verb pattern '%.*s' may not have an argument",
6897 verb_len, start_verb);
6899 ret = reg_node(pRExC_state, op);
6901 nextchar(pRExC_state);
6904 if (*RExC_parse == '?') { /* (?...) */
6905 bool is_logical = 0;
6906 const char * const seqstart = RExC_parse;
6907 bool has_use_defaults = FALSE;
6910 paren = *RExC_parse++;
6911 ret = NULL; /* For look-ahead/behind. */
6914 case 'P': /* (?P...) variants for those used to PCRE/Python */
6915 paren = *RExC_parse++;
6916 if ( paren == '<') /* (?P<...>) named capture */
6918 else if (paren == '>') { /* (?P>name) named recursion */
6919 goto named_recursion;
6921 else if (paren == '=') { /* (?P=...) named backref */
6922 /* this pretty much dupes the code for \k<NAME> in regatom(), if
6923 you change this make sure you change that */
6924 char* name_start = RExC_parse;
6926 SV *sv_dat = reg_scan_name(pRExC_state,
6927 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6928 if (RExC_parse == name_start || *RExC_parse != ')')
6929 vFAIL2("Sequence %.3s... not terminated",parse_start);
6932 num = add_data( pRExC_state, 1, "S" );
6933 RExC_rxi->data->data[num]=(void*)sv_dat;
6934 SvREFCNT_inc_simple_void(sv_dat);
6937 ret = reganode(pRExC_state,
6940 : (MORE_ASCII_RESTRICTED)
6942 : (AT_LEAST_UNI_SEMANTICS)
6950 Set_Node_Offset(ret, parse_start+1);
6951 Set_Node_Cur_Length(ret); /* MJD */
6953 nextchar(pRExC_state);
6957 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6959 case '<': /* (?<...) */
6960 if (*RExC_parse == '!')
6962 else if (*RExC_parse != '=')
6968 case '\'': /* (?'...') */
6969 name_start= RExC_parse;
6970 svname = reg_scan_name(pRExC_state,
6971 SIZE_ONLY ? /* reverse test from the others */
6972 REG_RSN_RETURN_NAME :
6973 REG_RSN_RETURN_NULL);
6974 if (RExC_parse == name_start) {
6976 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6979 if (*RExC_parse != paren)
6980 vFAIL2("Sequence (?%c... not terminated",
6981 paren=='>' ? '<' : paren);
6985 if (!svname) /* shouldn't happen */
6987 "panic: reg_scan_name returned NULL");
6988 if (!RExC_paren_names) {
6989 RExC_paren_names= newHV();
6990 sv_2mortal(MUTABLE_SV(RExC_paren_names));
6992 RExC_paren_name_list= newAV();
6993 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
6996 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
6998 sv_dat = HeVAL(he_str);
7000 /* croak baby croak */
7002 "panic: paren_name hash element allocation failed");
7003 } else if ( SvPOK(sv_dat) ) {
7004 /* (?|...) can mean we have dupes so scan to check
7005 its already been stored. Maybe a flag indicating
7006 we are inside such a construct would be useful,
7007 but the arrays are likely to be quite small, so
7008 for now we punt -- dmq */
7009 IV count = SvIV(sv_dat);
7010 I32 *pv = (I32*)SvPVX(sv_dat);
7012 for ( i = 0 ; i < count ; i++ ) {
7013 if ( pv[i] == RExC_npar ) {
7019 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
7020 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
7021 pv[count] = RExC_npar;
7022 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
7025 (void)SvUPGRADE(sv_dat,SVt_PVNV);
7026 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
7028 SvIV_set(sv_dat, 1);
7031 /* Yes this does cause a memory leak in debugging Perls */
7032 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
7033 SvREFCNT_dec(svname);
7036 /*sv_dump(sv_dat);*/
7038 nextchar(pRExC_state);
7040 goto capturing_parens;
7042 RExC_seen |= REG_SEEN_LOOKBEHIND;
7043 RExC_in_lookbehind++;
7045 case '=': /* (?=...) */
7046 RExC_seen_zerolen++;
7048 case '!': /* (?!...) */
7049 RExC_seen_zerolen++;
7050 if (*RExC_parse == ')') {
7051 ret=reg_node(pRExC_state, OPFAIL);
7052 nextchar(pRExC_state);
7056 case '|': /* (?|...) */
7057 /* branch reset, behave like a (?:...) except that
7058 buffers in alternations share the same numbers */
7060 after_freeze = freeze_paren = RExC_npar;
7062 case ':': /* (?:...) */
7063 case '>': /* (?>...) */
7065 case '$': /* (?$...) */
7066 case '@': /* (?@...) */
7067 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
7069 case '#': /* (?#...) */
7070 while (*RExC_parse && *RExC_parse != ')')
7072 if (*RExC_parse != ')')
7073 FAIL("Sequence (?#... not terminated");
7074 nextchar(pRExC_state);
7077 case '0' : /* (?0) */
7078 case 'R' : /* (?R) */
7079 if (*RExC_parse != ')')
7080 FAIL("Sequence (?R) not terminated");
7081 ret = reg_node(pRExC_state, GOSTART);
7082 *flagp |= POSTPONED;
7083 nextchar(pRExC_state);
7086 { /* named and numeric backreferences */
7088 case '&': /* (?&NAME) */
7089 parse_start = RExC_parse - 1;
7092 SV *sv_dat = reg_scan_name(pRExC_state,
7093 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7094 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
7096 goto gen_recurse_regop;
7099 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
7101 vFAIL("Illegal pattern");
7103 goto parse_recursion;
7105 case '-': /* (?-1) */
7106 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
7107 RExC_parse--; /* rewind to let it be handled later */
7111 case '1': case '2': case '3': case '4': /* (?1) */
7112 case '5': case '6': case '7': case '8': case '9':
7115 num = atoi(RExC_parse);
7116 parse_start = RExC_parse - 1; /* MJD */
7117 if (*RExC_parse == '-')
7119 while (isDIGIT(*RExC_parse))
7121 if (*RExC_parse!=')')
7122 vFAIL("Expecting close bracket");
7125 if ( paren == '-' ) {
7127 Diagram of capture buffer numbering.
7128 Top line is the normal capture buffer numbers
7129 Bottom line is the negative indexing as from
7133 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
7137 num = RExC_npar + num;
7140 vFAIL("Reference to nonexistent group");
7142 } else if ( paren == '+' ) {
7143 num = RExC_npar + num - 1;
7146 ret = reganode(pRExC_state, GOSUB, num);
7148 if (num > (I32)RExC_rx->nparens) {
7150 vFAIL("Reference to nonexistent group");
7152 ARG2L_SET( ret, RExC_recurse_count++);
7154 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7155 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
7159 RExC_seen |= REG_SEEN_RECURSE;
7160 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
7161 Set_Node_Offset(ret, parse_start); /* MJD */
7163 *flagp |= POSTPONED;
7164 nextchar(pRExC_state);
7166 } /* named and numeric backreferences */
7169 case '?': /* (??...) */
7171 if (*RExC_parse != '{') {
7173 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7176 *flagp |= POSTPONED;
7177 paren = *RExC_parse++;
7179 case '{': /* (?{...}) */
7184 char *s = RExC_parse;
7186 RExC_seen_zerolen++;
7187 RExC_seen |= REG_SEEN_EVAL;
7188 while (count && (c = *RExC_parse)) {
7199 if (*RExC_parse != ')') {
7201 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
7205 OP_4tree *sop, *rop;
7206 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
7209 Perl_save_re_context(aTHX);
7210 rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad);
7211 sop->op_private |= OPpREFCOUNTED;
7212 /* re_dup will OpREFCNT_inc */
7213 OpREFCNT_set(sop, 1);
7216 n = add_data(pRExC_state, 3, "nop");
7217 RExC_rxi->data->data[n] = (void*)rop;
7218 RExC_rxi->data->data[n+1] = (void*)sop;
7219 RExC_rxi->data->data[n+2] = (void*)pad;
7222 else { /* First pass */
7223 if (PL_reginterp_cnt < ++RExC_seen_evals
7225 /* No compiled RE interpolated, has runtime
7226 components ===> unsafe. */
7227 FAIL("Eval-group not allowed at runtime, use re 'eval'");
7228 if (PL_tainting && PL_tainted)
7229 FAIL("Eval-group in insecure regular expression");
7230 #if PERL_VERSION > 8
7231 if (IN_PERL_COMPILETIME)
7236 nextchar(pRExC_state);
7238 ret = reg_node(pRExC_state, LOGICAL);
7241 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
7242 /* deal with the length of this later - MJD */
7245 ret = reganode(pRExC_state, EVAL, n);
7246 Set_Node_Length(ret, RExC_parse - parse_start + 1);
7247 Set_Node_Offset(ret, parse_start);
7250 case '(': /* (?(?{...})...) and (?(?=...)...) */
7253 if (RExC_parse[0] == '?') { /* (?(?...)) */
7254 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
7255 || RExC_parse[1] == '<'
7256 || RExC_parse[1] == '{') { /* Lookahead or eval. */
7259 ret = reg_node(pRExC_state, LOGICAL);
7262 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
7266 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
7267 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
7269 char ch = RExC_parse[0] == '<' ? '>' : '\'';
7270 char *name_start= RExC_parse++;
7272 SV *sv_dat=reg_scan_name(pRExC_state,
7273 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7274 if (RExC_parse == name_start || *RExC_parse != ch)
7275 vFAIL2("Sequence (?(%c... not terminated",
7276 (ch == '>' ? '<' : ch));
7279 num = add_data( pRExC_state, 1, "S" );
7280 RExC_rxi->data->data[num]=(void*)sv_dat;
7281 SvREFCNT_inc_simple_void(sv_dat);
7283 ret = reganode(pRExC_state,NGROUPP,num);
7284 goto insert_if_check_paren;
7286 else if (RExC_parse[0] == 'D' &&
7287 RExC_parse[1] == 'E' &&
7288 RExC_parse[2] == 'F' &&
7289 RExC_parse[3] == 'I' &&
7290 RExC_parse[4] == 'N' &&
7291 RExC_parse[5] == 'E')
7293 ret = reganode(pRExC_state,DEFINEP,0);
7296 goto insert_if_check_paren;
7298 else if (RExC_parse[0] == 'R') {
7301 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
7302 parno = atoi(RExC_parse++);
7303 while (isDIGIT(*RExC_parse))
7305 } else if (RExC_parse[0] == '&') {
7308 sv_dat = reg_scan_name(pRExC_state,
7309 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7310 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
7312 ret = reganode(pRExC_state,INSUBP,parno);
7313 goto insert_if_check_paren;
7315 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
7318 parno = atoi(RExC_parse++);
7320 while (isDIGIT(*RExC_parse))
7322 ret = reganode(pRExC_state, GROUPP, parno);
7324 insert_if_check_paren:
7325 if ((c = *nextchar(pRExC_state)) != ')')
7326 vFAIL("Switch condition not recognized");
7328 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
7329 br = regbranch(pRExC_state, &flags, 1,depth+1);
7331 br = reganode(pRExC_state, LONGJMP, 0);
7333 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
7334 c = *nextchar(pRExC_state);
7339 vFAIL("(?(DEFINE)....) does not allow branches");
7340 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
7341 regbranch(pRExC_state, &flags, 1,depth+1);
7342 REGTAIL(pRExC_state, ret, lastbr);
7345 c = *nextchar(pRExC_state);
7350 vFAIL("Switch (?(condition)... contains too many branches");
7351 ender = reg_node(pRExC_state, TAIL);
7352 REGTAIL(pRExC_state, br, ender);
7354 REGTAIL(pRExC_state, lastbr, ender);
7355 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
7358 REGTAIL(pRExC_state, ret, ender);
7359 RExC_size++; /* XXX WHY do we need this?!!
7360 For large programs it seems to be required
7361 but I can't figure out why. -- dmq*/
7365 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
7369 RExC_parse--; /* for vFAIL to print correctly */
7370 vFAIL("Sequence (? incomplete");
7372 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
7374 has_use_defaults = TRUE;
7375 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
7376 set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics)
7377 ? REGEX_UNICODE_CHARSET
7378 : REGEX_DEPENDS_CHARSET);
7382 parse_flags: /* (?i) */
7384 U32 posflags = 0, negflags = 0;
7385 U32 *flagsp = &posflags;
7386 char has_charset_modifier = '\0';
7387 regex_charset cs = (RExC_utf8 || RExC_uni_semantics)
7388 ? REGEX_UNICODE_CHARSET
7389 : REGEX_DEPENDS_CHARSET;
7391 while (*RExC_parse) {
7392 /* && strchr("iogcmsx", *RExC_parse) */
7393 /* (?g), (?gc) and (?o) are useless here
7394 and must be globally applied -- japhy */
7395 switch (*RExC_parse) {
7396 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
7397 case LOCALE_PAT_MOD:
7398 if (has_charset_modifier) {
7399 goto excess_modifier;
7401 else if (flagsp == &negflags) {
7404 cs = REGEX_LOCALE_CHARSET;
7405 has_charset_modifier = LOCALE_PAT_MOD;
7406 RExC_contains_locale = 1;
7408 case UNICODE_PAT_MOD:
7409 if (has_charset_modifier) {
7410 goto excess_modifier;
7412 else if (flagsp == &negflags) {
7415 cs = REGEX_UNICODE_CHARSET;
7416 has_charset_modifier = UNICODE_PAT_MOD;
7418 case ASCII_RESTRICT_PAT_MOD:
7419 if (flagsp == &negflags) {
7422 if (has_charset_modifier) {
7423 if (cs != REGEX_ASCII_RESTRICTED_CHARSET) {
7424 goto excess_modifier;
7426 /* Doubled modifier implies more restricted */
7427 cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET;
7430 cs = REGEX_ASCII_RESTRICTED_CHARSET;
7432 has_charset_modifier = ASCII_RESTRICT_PAT_MOD;
7434 case DEPENDS_PAT_MOD:
7435 if (has_use_defaults) {
7436 goto fail_modifiers;
7438 else if (flagsp == &negflags) {
7441 else if (has_charset_modifier) {
7442 goto excess_modifier;
7445 /* The dual charset means unicode semantics if the
7446 * pattern (or target, not known until runtime) are
7447 * utf8, or something in the pattern indicates unicode
7449 cs = (RExC_utf8 || RExC_uni_semantics)
7450 ? REGEX_UNICODE_CHARSET
7451 : REGEX_DEPENDS_CHARSET;
7452 has_charset_modifier = DEPENDS_PAT_MOD;
7456 if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) {
7457 vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD);
7459 else if (has_charset_modifier == *(RExC_parse - 1)) {
7460 vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1));
7463 vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1));
7468 vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1));
7470 case ONCE_PAT_MOD: /* 'o' */
7471 case GLOBAL_PAT_MOD: /* 'g' */
7472 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7473 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
7474 if (! (wastedflags & wflagbit) ) {
7475 wastedflags |= wflagbit;
7478 "Useless (%s%c) - %suse /%c modifier",
7479 flagsp == &negflags ? "?-" : "?",
7481 flagsp == &negflags ? "don't " : "",
7488 case CONTINUE_PAT_MOD: /* 'c' */
7489 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7490 if (! (wastedflags & WASTED_C) ) {
7491 wastedflags |= WASTED_GC;
7494 "Useless (%sc) - %suse /gc modifier",
7495 flagsp == &negflags ? "?-" : "?",
7496 flagsp == &negflags ? "don't " : ""
7501 case KEEPCOPY_PAT_MOD: /* 'p' */
7502 if (flagsp == &negflags) {
7504 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
7506 *flagsp |= RXf_PMf_KEEPCOPY;
7510 /* A flag is a default iff it is following a minus, so
7511 * if there is a minus, it means will be trying to
7512 * re-specify a default which is an error */
7513 if (has_use_defaults || flagsp == &negflags) {
7516 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7520 wastedflags = 0; /* reset so (?g-c) warns twice */
7526 RExC_flags |= posflags;
7527 RExC_flags &= ~negflags;
7528 set_regex_charset(&RExC_flags, cs);
7530 oregflags |= posflags;
7531 oregflags &= ~negflags;
7532 set_regex_charset(&oregflags, cs);
7534 nextchar(pRExC_state);
7545 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7550 }} /* one for the default block, one for the switch */
7557 ret = reganode(pRExC_state, OPEN, parno);
7560 RExC_nestroot = parno;
7561 if (RExC_seen & REG_SEEN_RECURSE
7562 && !RExC_open_parens[parno-1])
7564 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7565 "Setting open paren #%"IVdf" to %d\n",
7566 (IV)parno, REG_NODE_NUM(ret)));
7567 RExC_open_parens[parno-1]= ret;
7570 Set_Node_Length(ret, 1); /* MJD */
7571 Set_Node_Offset(ret, RExC_parse); /* MJD */
7579 /* Pick up the branches, linking them together. */
7580 parse_start = RExC_parse; /* MJD */
7581 br = regbranch(pRExC_state, &flags, 1,depth+1);
7583 /* branch_len = (paren != 0); */
7587 if (*RExC_parse == '|') {
7588 if (!SIZE_ONLY && RExC_extralen) {
7589 reginsert(pRExC_state, BRANCHJ, br, depth+1);
7592 reginsert(pRExC_state, BRANCH, br, depth+1);
7593 Set_Node_Length(br, paren != 0);
7594 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
7598 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
7600 else if (paren == ':') {
7601 *flagp |= flags&SIMPLE;
7603 if (is_open) { /* Starts with OPEN. */
7604 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
7606 else if (paren != '?') /* Not Conditional */
7608 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7610 while (*RExC_parse == '|') {
7611 if (!SIZE_ONLY && RExC_extralen) {
7612 ender = reganode(pRExC_state, LONGJMP,0);
7613 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
7616 RExC_extralen += 2; /* Account for LONGJMP. */
7617 nextchar(pRExC_state);
7619 if (RExC_npar > after_freeze)
7620 after_freeze = RExC_npar;
7621 RExC_npar = freeze_paren;
7623 br = regbranch(pRExC_state, &flags, 0, depth+1);
7627 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
7629 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7632 if (have_branch || paren != ':') {
7633 /* Make a closing node, and hook it on the end. */
7636 ender = reg_node(pRExC_state, TAIL);
7639 ender = reganode(pRExC_state, CLOSE, parno);
7640 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
7641 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7642 "Setting close paren #%"IVdf" to %d\n",
7643 (IV)parno, REG_NODE_NUM(ender)));
7644 RExC_close_parens[parno-1]= ender;
7645 if (RExC_nestroot == parno)
7648 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
7649 Set_Node_Length(ender,1); /* MJD */
7655 *flagp &= ~HASWIDTH;
7658 ender = reg_node(pRExC_state, SUCCEED);
7661 ender = reg_node(pRExC_state, END);
7663 assert(!RExC_opend); /* there can only be one! */
7668 REGTAIL(pRExC_state, lastbr, ender);
7670 if (have_branch && !SIZE_ONLY) {
7672 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
7674 /* Hook the tails of the branches to the closing node. */
7675 for (br = ret; br; br = regnext(br)) {
7676 const U8 op = PL_regkind[OP(br)];
7678 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
7680 else if (op == BRANCHJ) {
7681 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
7689 static const char parens[] = "=!<,>";
7691 if (paren && (p = strchr(parens, paren))) {
7692 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
7693 int flag = (p - parens) > 1;
7696 node = SUSPEND, flag = 0;
7697 reginsert(pRExC_state, node,ret, depth+1);
7698 Set_Node_Cur_Length(ret);
7699 Set_Node_Offset(ret, parse_start + 1);
7701 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
7705 /* Check for proper termination. */
7707 RExC_flags = oregflags;
7708 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
7709 RExC_parse = oregcomp_parse;
7710 vFAIL("Unmatched (");
7713 else if (!paren && RExC_parse < RExC_end) {
7714 if (*RExC_parse == ')') {
7716 vFAIL("Unmatched )");
7719 FAIL("Junk on end of regexp"); /* "Can't happen". */
7723 if (RExC_in_lookbehind) {
7724 RExC_in_lookbehind--;
7726 if (after_freeze > RExC_npar)
7727 RExC_npar = after_freeze;
7732 - regbranch - one alternative of an | operator
7734 * Implements the concatenation operator.
7737 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
7740 register regnode *ret;
7741 register regnode *chain = NULL;
7742 register regnode *latest;
7743 I32 flags = 0, c = 0;
7744 GET_RE_DEBUG_FLAGS_DECL;
7746 PERL_ARGS_ASSERT_REGBRANCH;
7748 DEBUG_PARSE("brnc");
7753 if (!SIZE_ONLY && RExC_extralen)
7754 ret = reganode(pRExC_state, BRANCHJ,0);
7756 ret = reg_node(pRExC_state, BRANCH);
7757 Set_Node_Length(ret, 1);
7761 if (!first && SIZE_ONLY)
7762 RExC_extralen += 1; /* BRANCHJ */
7764 *flagp = WORST; /* Tentatively. */
7767 nextchar(pRExC_state);
7768 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
7770 latest = regpiece(pRExC_state, &flags,depth+1);
7771 if (latest == NULL) {
7772 if (flags & TRYAGAIN)
7776 else if (ret == NULL)
7778 *flagp |= flags&(HASWIDTH|POSTPONED);
7779 if (chain == NULL) /* First piece. */
7780 *flagp |= flags&SPSTART;
7783 REGTAIL(pRExC_state, chain, latest);
7788 if (chain == NULL) { /* Loop ran zero times. */
7789 chain = reg_node(pRExC_state, NOTHING);
7794 *flagp |= flags&SIMPLE;
7801 - regpiece - something followed by possible [*+?]
7803 * Note that the branching code sequences used for ? and the general cases
7804 * of * and + are somewhat optimized: they use the same NOTHING node as
7805 * both the endmarker for their branch list and the body of the last branch.
7806 * It might seem that this node could be dispensed with entirely, but the
7807 * endmarker role is not redundant.
7810 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7813 register regnode *ret;
7815 register char *next;
7817 const char * const origparse = RExC_parse;
7819 I32 max = REG_INFTY;
7820 #ifdef RE_TRACK_PATTERN_OFFSETS
7823 const char *maxpos = NULL;
7824 GET_RE_DEBUG_FLAGS_DECL;
7826 PERL_ARGS_ASSERT_REGPIECE;
7828 DEBUG_PARSE("piec");
7830 ret = regatom(pRExC_state, &flags,depth+1);
7832 if (flags & TRYAGAIN)
7839 if (op == '{' && regcurly(RExC_parse)) {
7841 #ifdef RE_TRACK_PATTERN_OFFSETS
7842 parse_start = RExC_parse; /* MJD */
7844 next = RExC_parse + 1;
7845 while (isDIGIT(*next) || *next == ',') {
7854 if (*next == '}') { /* got one */
7858 min = atoi(RExC_parse);
7862 maxpos = RExC_parse;
7864 if (!max && *maxpos != '0')
7865 max = REG_INFTY; /* meaning "infinity" */
7866 else if (max >= REG_INFTY)
7867 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
7869 nextchar(pRExC_state);
7872 if ((flags&SIMPLE)) {
7873 RExC_naughty += 2 + RExC_naughty / 2;
7874 reginsert(pRExC_state, CURLY, ret, depth+1);
7875 Set_Node_Offset(ret, parse_start+1); /* MJD */
7876 Set_Node_Cur_Length(ret);
7879 regnode * const w = reg_node(pRExC_state, WHILEM);
7882 REGTAIL(pRExC_state, ret, w);
7883 if (!SIZE_ONLY && RExC_extralen) {
7884 reginsert(pRExC_state, LONGJMP,ret, depth+1);
7885 reginsert(pRExC_state, NOTHING,ret, depth+1);
7886 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
7888 reginsert(pRExC_state, CURLYX,ret, depth+1);
7890 Set_Node_Offset(ret, parse_start+1);
7891 Set_Node_Length(ret,
7892 op == '{' ? (RExC_parse - parse_start) : 1);
7894 if (!SIZE_ONLY && RExC_extralen)
7895 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
7896 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
7898 RExC_whilem_seen++, RExC_extralen += 3;
7899 RExC_naughty += 4 + RExC_naughty; /* compound interest */
7908 vFAIL("Can't do {n,m} with n > m");
7910 ARG1_SET(ret, (U16)min);
7911 ARG2_SET(ret, (U16)max);
7923 #if 0 /* Now runtime fix should be reliable. */
7925 /* if this is reinstated, don't forget to put this back into perldiag:
7927 =item Regexp *+ operand could be empty at {#} in regex m/%s/
7929 (F) The part of the regexp subject to either the * or + quantifier
7930 could match an empty string. The {#} shows in the regular
7931 expression about where the problem was discovered.
7935 if (!(flags&HASWIDTH) && op != '?')
7936 vFAIL("Regexp *+ operand could be empty");
7939 #ifdef RE_TRACK_PATTERN_OFFSETS
7940 parse_start = RExC_parse;
7942 nextchar(pRExC_state);
7944 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
7946 if (op == '*' && (flags&SIMPLE)) {
7947 reginsert(pRExC_state, STAR, ret, depth+1);
7951 else if (op == '*') {
7955 else if (op == '+' && (flags&SIMPLE)) {
7956 reginsert(pRExC_state, PLUS, ret, depth+1);
7960 else if (op == '+') {
7964 else if (op == '?') {
7969 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
7970 ckWARN3reg(RExC_parse,
7971 "%.*s matches null string many times",
7972 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
7976 if (RExC_parse < RExC_end && *RExC_parse == '?') {
7977 nextchar(pRExC_state);
7978 reginsert(pRExC_state, MINMOD, ret, depth+1);
7979 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
7981 #ifndef REG_ALLOW_MINMOD_SUSPEND
7984 if (RExC_parse < RExC_end && *RExC_parse == '+') {
7986 nextchar(pRExC_state);
7987 ender = reg_node(pRExC_state, SUCCEED);
7988 REGTAIL(pRExC_state, ret, ender);
7989 reginsert(pRExC_state, SUSPEND, ret, depth+1);
7991 ender = reg_node(pRExC_state, TAIL);
7992 REGTAIL(pRExC_state, ret, ender);
7996 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
7998 vFAIL("Nested quantifiers");
8005 /* reg_namedseq(pRExC_state,UVp, UV depth)
8007 This is expected to be called by a parser routine that has
8008 recognized '\N' and needs to handle the rest. RExC_parse is
8009 expected to point at the first char following the N at the time
8012 The \N may be inside (indicated by valuep not being NULL) or outside a
8015 \N may begin either a named sequence, or if outside a character class, mean
8016 to match a non-newline. For non single-quoted regexes, the tokenizer has
8017 attempted to decide which, and in the case of a named sequence converted it
8018 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
8019 where c1... are the characters in the sequence. For single-quoted regexes,
8020 the tokenizer passes the \N sequence through unchanged; this code will not
8021 attempt to determine this nor expand those. The net effect is that if the
8022 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
8023 signals that this \N occurrence means to match a non-newline.
8025 Only the \N{U+...} form should occur in a character class, for the same
8026 reason that '.' inside a character class means to just match a period: it
8027 just doesn't make sense.
8029 If valuep is non-null then it is assumed that we are parsing inside
8030 of a charclass definition and the first codepoint in the resolved
8031 string is returned via *valuep and the routine will return NULL.
8032 In this mode if a multichar string is returned from the charnames
8033 handler, a warning will be issued, and only the first char in the
8034 sequence will be examined. If the string returned is zero length
8035 then the value of *valuep is undefined and NON-NULL will
8036 be returned to indicate failure. (This will NOT be a valid pointer
8039 If valuep is null then it is assumed that we are parsing normal text and a
8040 new EXACT node is inserted into the program containing the resolved string,
8041 and a pointer to the new node is returned. But if the string is zero length
8042 a NOTHING node is emitted instead.
8044 On success RExC_parse is set to the char following the endbrace.
8045 Parsing failures will generate a fatal error via vFAIL(...)
8048 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp, U32 depth)
8050 char * endbrace; /* '}' following the name */
8051 regnode *ret = NULL;
8054 GET_RE_DEBUG_FLAGS_DECL;
8056 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
8060 /* The [^\n] meaning of \N ignores spaces and comments under the /x
8061 * modifier. The other meaning does not */
8062 p = (RExC_flags & RXf_PMf_EXTENDED)
8063 ? regwhite( pRExC_state, RExC_parse )
8066 /* Disambiguate between \N meaning a named character versus \N meaning
8067 * [^\n]. The former is assumed when it can't be the latter. */
8068 if (*p != '{' || regcurly(p)) {
8071 /* no bare \N in a charclass */
8072 vFAIL("\\N in a character class must be a named character: \\N{...}");
8074 nextchar(pRExC_state);
8075 ret = reg_node(pRExC_state, REG_ANY);
8076 *flagp |= HASWIDTH|SIMPLE;
8079 Set_Node_Length(ret, 1); /* MJD */
8083 /* Here, we have decided it should be a named sequence */
8085 /* The test above made sure that the next real character is a '{', but
8086 * under the /x modifier, it could be separated by space (or a comment and
8087 * \n) and this is not allowed (for consistency with \x{...} and the
8088 * tokenizer handling of \N{NAME}). */
8089 if (*RExC_parse != '{') {
8090 vFAIL("Missing braces on \\N{}");
8093 RExC_parse++; /* Skip past the '{' */
8095 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
8096 || ! (endbrace == RExC_parse /* nothing between the {} */
8097 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
8098 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
8100 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
8101 vFAIL("\\N{NAME} must be resolved by the lexer");
8104 if (endbrace == RExC_parse) { /* empty: \N{} */
8106 RExC_parse = endbrace + 1;
8107 return reg_node(pRExC_state,NOTHING);
8111 ckWARNreg(RExC_parse,
8112 "Ignoring zero length \\N{} in character class"
8114 RExC_parse = endbrace + 1;
8117 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
8120 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
8121 RExC_parse += 2; /* Skip past the 'U+' */
8123 if (valuep) { /* In a bracketed char class */
8124 /* We only pay attention to the first char of
8125 multichar strings being returned. I kinda wonder
8126 if this makes sense as it does change the behaviour
8127 from earlier versions, OTOH that behaviour was broken
8128 as well. XXX Solution is to recharacterize as
8129 [rest-of-class]|multi1|multi2... */
8131 STRLEN length_of_hex;
8132 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8133 | PERL_SCAN_DISALLOW_PREFIX
8134 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
8136 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
8137 if (endchar < endbrace) {
8138 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
8141 length_of_hex = (STRLEN)(endchar - RExC_parse);
8142 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
8144 /* The tokenizer should have guaranteed validity, but it's possible to
8145 * bypass it by using single quoting, so check */
8146 if (length_of_hex == 0
8147 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
8149 RExC_parse += length_of_hex; /* Includes all the valid */
8150 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
8151 ? UTF8SKIP(RExC_parse)
8153 /* Guard against malformed utf8 */
8154 if (RExC_parse >= endchar) RExC_parse = endchar;
8155 vFAIL("Invalid hexadecimal number in \\N{U+...}");
8158 RExC_parse = endbrace + 1;
8159 if (endchar == endbrace) return NULL;
8161 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
8163 else { /* Not a char class */
8165 /* What is done here is to convert this to a sub-pattern of the form
8166 * (?:\x{char1}\x{char2}...)
8167 * and then call reg recursively. That way, it retains its atomicness,
8168 * while not having to worry about special handling that some code
8169 * points may have. toke.c has converted the original Unicode values
8170 * to native, so that we can just pass on the hex values unchanged. We
8171 * do have to set a flag to keep recoding from happening in the
8174 SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP);
8176 char *endchar; /* Points to '.' or '}' ending cur char in the input
8178 char *orig_end = RExC_end;
8180 while (RExC_parse < endbrace) {
8182 /* Code points are separated by dots. If none, there is only one
8183 * code point, and is terminated by the brace */
8184 endchar = RExC_parse + strcspn(RExC_parse, ".}");
8186 /* Convert to notation the rest of the code understands */
8187 sv_catpv(substitute_parse, "\\x{");
8188 sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse);
8189 sv_catpv(substitute_parse, "}");
8191 /* Point to the beginning of the next character in the sequence. */
8192 RExC_parse = endchar + 1;
8194 sv_catpv(substitute_parse, ")");
8196 RExC_parse = SvPV(substitute_parse, len);
8198 /* Don't allow empty number */
8200 vFAIL("Invalid hexadecimal number in \\N{U+...}");
8202 RExC_end = RExC_parse + len;
8204 /* The values are Unicode, and therefore not subject to recoding */
8205 RExC_override_recoding = 1;
8207 ret = reg(pRExC_state, 1, flagp, depth+1);
8209 RExC_parse = endbrace;
8210 RExC_end = orig_end;
8211 RExC_override_recoding = 0;
8213 nextchar(pRExC_state);
8223 * It returns the code point in utf8 for the value in *encp.
8224 * value: a code value in the source encoding
8225 * encp: a pointer to an Encode object
8227 * If the result from Encode is not a single character,
8228 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
8231 S_reg_recode(pTHX_ const char value, SV **encp)
8234 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
8235 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
8236 const STRLEN newlen = SvCUR(sv);
8237 UV uv = UNICODE_REPLACEMENT;
8239 PERL_ARGS_ASSERT_REG_RECODE;
8243 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
8246 if (!newlen || numlen != newlen) {
8247 uv = UNICODE_REPLACEMENT;
8255 - regatom - the lowest level
8257 Try to identify anything special at the start of the pattern. If there
8258 is, then handle it as required. This may involve generating a single regop,
8259 such as for an assertion; or it may involve recursing, such as to
8260 handle a () structure.
8262 If the string doesn't start with something special then we gobble up
8263 as much literal text as we can.
8265 Once we have been able to handle whatever type of thing started the
8266 sequence, we return.
8268 Note: we have to be careful with escapes, as they can be both literal
8269 and special, and in the case of \10 and friends can either, depending
8270 on context. Specifically there are two separate switches for handling
8271 escape sequences, with the one for handling literal escapes requiring
8272 a dummy entry for all of the special escapes that are actually handled
8277 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
8280 register regnode *ret = NULL;
8282 char *parse_start = RExC_parse;
8284 GET_RE_DEBUG_FLAGS_DECL;
8285 DEBUG_PARSE("atom");
8286 *flagp = WORST; /* Tentatively. */
8288 PERL_ARGS_ASSERT_REGATOM;
8291 switch ((U8)*RExC_parse) {
8293 RExC_seen_zerolen++;
8294 nextchar(pRExC_state);
8295 if (RExC_flags & RXf_PMf_MULTILINE)
8296 ret = reg_node(pRExC_state, MBOL);
8297 else if (RExC_flags & RXf_PMf_SINGLELINE)
8298 ret = reg_node(pRExC_state, SBOL);
8300 ret = reg_node(pRExC_state, BOL);
8301 Set_Node_Length(ret, 1); /* MJD */
8304 nextchar(pRExC_state);
8306 RExC_seen_zerolen++;
8307 if (RExC_flags & RXf_PMf_MULTILINE)
8308 ret = reg_node(pRExC_state, MEOL);
8309 else if (RExC_flags & RXf_PMf_SINGLELINE)
8310 ret = reg_node(pRExC_state, SEOL);
8312 ret = reg_node(pRExC_state, EOL);
8313 Set_Node_Length(ret, 1); /* MJD */
8316 nextchar(pRExC_state);
8317 if (RExC_flags & RXf_PMf_SINGLELINE)
8318 ret = reg_node(pRExC_state, SANY);
8320 ret = reg_node(pRExC_state, REG_ANY);
8321 *flagp |= HASWIDTH|SIMPLE;
8323 Set_Node_Length(ret, 1); /* MJD */
8327 char * const oregcomp_parse = ++RExC_parse;
8328 ret = regclass(pRExC_state,depth+1);
8329 if (*RExC_parse != ']') {
8330 RExC_parse = oregcomp_parse;
8331 vFAIL("Unmatched [");
8333 nextchar(pRExC_state);
8334 *flagp |= HASWIDTH|SIMPLE;
8335 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
8339 nextchar(pRExC_state);
8340 ret = reg(pRExC_state, 1, &flags,depth+1);
8342 if (flags & TRYAGAIN) {
8343 if (RExC_parse == RExC_end) {
8344 /* Make parent create an empty node if needed. */
8352 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
8356 if (flags & TRYAGAIN) {
8360 vFAIL("Internal urp");
8361 /* Supposed to be caught earlier. */
8364 if (!regcurly(RExC_parse)) {
8373 vFAIL("Quantifier follows nothing");
8378 This switch handles escape sequences that resolve to some kind
8379 of special regop and not to literal text. Escape sequnces that
8380 resolve to literal text are handled below in the switch marked
8383 Every entry in this switch *must* have a corresponding entry
8384 in the literal escape switch. However, the opposite is not
8385 required, as the default for this switch is to jump to the
8386 literal text handling code.
8388 switch ((U8)*++RExC_parse) {
8389 /* Special Escapes */
8391 RExC_seen_zerolen++;
8392 ret = reg_node(pRExC_state, SBOL);
8394 goto finish_meta_pat;
8396 ret = reg_node(pRExC_state, GPOS);
8397 RExC_seen |= REG_SEEN_GPOS;
8399 goto finish_meta_pat;
8401 RExC_seen_zerolen++;
8402 ret = reg_node(pRExC_state, KEEPS);
8404 /* XXX:dmq : disabling in-place substitution seems to
8405 * be necessary here to avoid cases of memory corruption, as
8406 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
8408 RExC_seen |= REG_SEEN_LOOKBEHIND;
8409 goto finish_meta_pat;
8411 ret = reg_node(pRExC_state, SEOL);
8413 RExC_seen_zerolen++; /* Do not optimize RE away */
8414 goto finish_meta_pat;
8416 ret = reg_node(pRExC_state, EOS);
8418 RExC_seen_zerolen++; /* Do not optimize RE away */
8419 goto finish_meta_pat;
8421 ret = reg_node(pRExC_state, CANY);
8422 RExC_seen |= REG_SEEN_CANY;
8423 *flagp |= HASWIDTH|SIMPLE;
8424 goto finish_meta_pat;
8426 ret = reg_node(pRExC_state, CLUMP);
8428 goto finish_meta_pat;
8430 switch (get_regex_charset(RExC_flags)) {
8431 case REGEX_LOCALE_CHARSET:
8434 case REGEX_UNICODE_CHARSET:
8437 case REGEX_ASCII_RESTRICTED_CHARSET:
8438 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8441 case REGEX_DEPENDS_CHARSET:
8447 ret = reg_node(pRExC_state, op);
8448 *flagp |= HASWIDTH|SIMPLE;
8449 goto finish_meta_pat;
8451 switch (get_regex_charset(RExC_flags)) {
8452 case REGEX_LOCALE_CHARSET:
8455 case REGEX_UNICODE_CHARSET:
8458 case REGEX_ASCII_RESTRICTED_CHARSET:
8459 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8462 case REGEX_DEPENDS_CHARSET:
8468 ret = reg_node(pRExC_state, op);
8469 *flagp |= HASWIDTH|SIMPLE;
8470 goto finish_meta_pat;
8472 RExC_seen_zerolen++;
8473 RExC_seen |= REG_SEEN_LOOKBEHIND;
8474 switch (get_regex_charset(RExC_flags)) {
8475 case REGEX_LOCALE_CHARSET:
8478 case REGEX_UNICODE_CHARSET:
8481 case REGEX_ASCII_RESTRICTED_CHARSET:
8482 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8485 case REGEX_DEPENDS_CHARSET:
8491 ret = reg_node(pRExC_state, op);
8492 FLAGS(ret) = get_regex_charset(RExC_flags);
8494 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8495 ckWARNregdep(RExC_parse, "\"\\b{\" is deprecated; use \"\\b\\{\" instead");
8497 goto finish_meta_pat;
8499 RExC_seen_zerolen++;
8500 RExC_seen |= REG_SEEN_LOOKBEHIND;
8501 switch (get_regex_charset(RExC_flags)) {
8502 case REGEX_LOCALE_CHARSET:
8505 case REGEX_UNICODE_CHARSET:
8508 case REGEX_ASCII_RESTRICTED_CHARSET:
8509 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8512 case REGEX_DEPENDS_CHARSET:
8518 ret = reg_node(pRExC_state, op);
8519 FLAGS(ret) = get_regex_charset(RExC_flags);
8521 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8522 ckWARNregdep(RExC_parse, "\"\\B{\" is deprecated; use \"\\B\\{\" instead");
8524 goto finish_meta_pat;
8526 switch (get_regex_charset(RExC_flags)) {
8527 case REGEX_LOCALE_CHARSET:
8530 case REGEX_UNICODE_CHARSET:
8533 case REGEX_ASCII_RESTRICTED_CHARSET:
8534 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8537 case REGEX_DEPENDS_CHARSET:
8543 ret = reg_node(pRExC_state, op);
8544 *flagp |= HASWIDTH|SIMPLE;
8545 goto finish_meta_pat;
8547 switch (get_regex_charset(RExC_flags)) {
8548 case REGEX_LOCALE_CHARSET:
8551 case REGEX_UNICODE_CHARSET:
8554 case REGEX_ASCII_RESTRICTED_CHARSET:
8555 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8558 case REGEX_DEPENDS_CHARSET:
8564 ret = reg_node(pRExC_state, op);
8565 *flagp |= HASWIDTH|SIMPLE;
8566 goto finish_meta_pat;
8568 switch (get_regex_charset(RExC_flags)) {
8569 case REGEX_LOCALE_CHARSET:
8572 case REGEX_ASCII_RESTRICTED_CHARSET:
8573 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8576 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8577 case REGEX_UNICODE_CHARSET:
8583 ret = reg_node(pRExC_state, op);
8584 *flagp |= HASWIDTH|SIMPLE;
8585 goto finish_meta_pat;
8587 switch (get_regex_charset(RExC_flags)) {
8588 case REGEX_LOCALE_CHARSET:
8591 case REGEX_ASCII_RESTRICTED_CHARSET:
8592 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8595 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8596 case REGEX_UNICODE_CHARSET:
8602 ret = reg_node(pRExC_state, op);
8603 *flagp |= HASWIDTH|SIMPLE;
8604 goto finish_meta_pat;
8606 ret = reg_node(pRExC_state, LNBREAK);
8607 *flagp |= HASWIDTH|SIMPLE;
8608 goto finish_meta_pat;
8610 ret = reg_node(pRExC_state, HORIZWS);
8611 *flagp |= HASWIDTH|SIMPLE;
8612 goto finish_meta_pat;
8614 ret = reg_node(pRExC_state, NHORIZWS);
8615 *flagp |= HASWIDTH|SIMPLE;
8616 goto finish_meta_pat;
8618 ret = reg_node(pRExC_state, VERTWS);
8619 *flagp |= HASWIDTH|SIMPLE;
8620 goto finish_meta_pat;
8622 ret = reg_node(pRExC_state, NVERTWS);
8623 *flagp |= HASWIDTH|SIMPLE;
8625 nextchar(pRExC_state);
8626 Set_Node_Length(ret, 2); /* MJD */
8631 char* const oldregxend = RExC_end;
8633 char* parse_start = RExC_parse - 2;
8636 if (RExC_parse[1] == '{') {
8637 /* a lovely hack--pretend we saw [\pX] instead */
8638 RExC_end = strchr(RExC_parse, '}');
8640 const U8 c = (U8)*RExC_parse;
8642 RExC_end = oldregxend;
8643 vFAIL2("Missing right brace on \\%c{}", c);
8648 RExC_end = RExC_parse + 2;
8649 if (RExC_end > oldregxend)
8650 RExC_end = oldregxend;
8654 ret = regclass(pRExC_state,depth+1);
8656 RExC_end = oldregxend;
8659 Set_Node_Offset(ret, parse_start + 2);
8660 Set_Node_Cur_Length(ret);
8661 nextchar(pRExC_state);
8662 *flagp |= HASWIDTH|SIMPLE;
8666 /* Handle \N and \N{NAME} here and not below because it can be
8667 multicharacter. join_exact() will join them up later on.
8668 Also this makes sure that things like /\N{BLAH}+/ and
8669 \N{BLAH} being multi char Just Happen. dmq*/
8671 ret= reg_namedseq(pRExC_state, NULL, flagp, depth);
8673 case 'k': /* Handle \k<NAME> and \k'NAME' */
8676 char ch= RExC_parse[1];
8677 if (ch != '<' && ch != '\'' && ch != '{') {
8679 vFAIL2("Sequence %.2s... not terminated",parse_start);
8681 /* this pretty much dupes the code for (?P=...) in reg(), if
8682 you change this make sure you change that */
8683 char* name_start = (RExC_parse += 2);
8685 SV *sv_dat = reg_scan_name(pRExC_state,
8686 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
8687 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
8688 if (RExC_parse == name_start || *RExC_parse != ch)
8689 vFAIL2("Sequence %.3s... not terminated",parse_start);
8692 num = add_data( pRExC_state, 1, "S" );
8693 RExC_rxi->data->data[num]=(void*)sv_dat;
8694 SvREFCNT_inc_simple_void(sv_dat);
8698 ret = reganode(pRExC_state,
8701 : (MORE_ASCII_RESTRICTED)
8703 : (AT_LEAST_UNI_SEMANTICS)
8711 /* override incorrect value set in reganode MJD */
8712 Set_Node_Offset(ret, parse_start+1);
8713 Set_Node_Cur_Length(ret); /* MJD */
8714 nextchar(pRExC_state);
8720 case '1': case '2': case '3': case '4':
8721 case '5': case '6': case '7': case '8': case '9':
8724 bool isg = *RExC_parse == 'g';
8729 if (*RExC_parse == '{') {
8733 if (*RExC_parse == '-') {
8737 if (hasbrace && !isDIGIT(*RExC_parse)) {
8738 if (isrel) RExC_parse--;
8740 goto parse_named_seq;
8742 num = atoi(RExC_parse);
8743 if (isg && num == 0)
8744 vFAIL("Reference to invalid group 0");
8746 num = RExC_npar - num;
8748 vFAIL("Reference to nonexistent or unclosed group");
8750 if (!isg && num > 9 && num >= RExC_npar)
8753 char * const parse_start = RExC_parse - 1; /* MJD */
8754 while (isDIGIT(*RExC_parse))
8756 if (parse_start == RExC_parse - 1)
8757 vFAIL("Unterminated \\g... pattern");
8759 if (*RExC_parse != '}')
8760 vFAIL("Unterminated \\g{...} pattern");
8764 if (num > (I32)RExC_rx->nparens)
8765 vFAIL("Reference to nonexistent group");
8768 ret = reganode(pRExC_state,
8771 : (MORE_ASCII_RESTRICTED)
8773 : (AT_LEAST_UNI_SEMANTICS)
8781 /* override incorrect value set in reganode MJD */
8782 Set_Node_Offset(ret, parse_start+1);
8783 Set_Node_Cur_Length(ret); /* MJD */
8785 nextchar(pRExC_state);
8790 if (RExC_parse >= RExC_end)
8791 FAIL("Trailing \\");
8794 /* Do not generate "unrecognized" warnings here, we fall
8795 back into the quick-grab loop below */
8802 if (RExC_flags & RXf_PMf_EXTENDED) {
8803 if ( reg_skipcomment( pRExC_state ) )
8810 parse_start = RExC_parse - 1;
8823 char_state latest_char_state = generic_char;
8824 register STRLEN len;
8829 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
8830 regnode * orig_emit;
8833 orig_emit = RExC_emit; /* Save the original output node position in
8834 case we need to output a different node
8836 ret = reg_node(pRExC_state,
8837 (U8) ((! FOLD) ? EXACT
8840 : (MORE_ASCII_RESTRICTED)
8842 : (AT_LEAST_UNI_SEMANTICS)
8847 for (len = 0, p = RExC_parse - 1;
8848 len < 127 && p < RExC_end;
8851 char * const oldp = p;
8853 if (RExC_flags & RXf_PMf_EXTENDED)
8854 p = regwhite( pRExC_state, p );
8865 /* Literal Escapes Switch
8867 This switch is meant to handle escape sequences that
8868 resolve to a literal character.
8870 Every escape sequence that represents something
8871 else, like an assertion or a char class, is handled
8872 in the switch marked 'Special Escapes' above in this
8873 routine, but also has an entry here as anything that
8874 isn't explicitly mentioned here will be treated as
8875 an unescaped equivalent literal.
8879 /* These are all the special escapes. */
8880 case 'A': /* Start assertion */
8881 case 'b': case 'B': /* Word-boundary assertion*/
8882 case 'C': /* Single char !DANGEROUS! */
8883 case 'd': case 'D': /* digit class */
8884 case 'g': case 'G': /* generic-backref, pos assertion */
8885 case 'h': case 'H': /* HORIZWS */
8886 case 'k': case 'K': /* named backref, keep marker */
8887 case 'N': /* named char sequence */
8888 case 'p': case 'P': /* Unicode property */
8889 case 'R': /* LNBREAK */
8890 case 's': case 'S': /* space class */
8891 case 'v': case 'V': /* VERTWS */
8892 case 'w': case 'W': /* word class */
8893 case 'X': /* eXtended Unicode "combining character sequence" */
8894 case 'z': case 'Z': /* End of line/string assertion */
8898 /* Anything after here is an escape that resolves to a
8899 literal. (Except digits, which may or may not)
8918 ender = ASCII_TO_NATIVE('\033');
8922 ender = ASCII_TO_NATIVE('\007');
8927 STRLEN brace_len = len;
8929 const char* error_msg;
8931 bool valid = grok_bslash_o(p,
8938 RExC_parse = p; /* going to die anyway; point
8939 to exact spot of failure */
8946 if (PL_encoding && ender < 0x100) {
8947 goto recode_encoding;
8956 char* const e = strchr(p, '}');
8960 vFAIL("Missing right brace on \\x{}");
8963 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8964 | PERL_SCAN_DISALLOW_PREFIX;
8965 STRLEN numlen = e - p - 1;
8966 ender = grok_hex(p + 1, &numlen, &flags, NULL);
8973 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8975 ender = grok_hex(p, &numlen, &flags, NULL);
8978 if (PL_encoding && ender < 0x100)
8979 goto recode_encoding;
8983 ender = grok_bslash_c(*p++, UTF, SIZE_ONLY);
8985 case '0': case '1': case '2': case '3':case '4':
8986 case '5': case '6': case '7': case '8':case '9':
8988 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
8990 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8992 ender = grok_oct(p, &numlen, &flags, NULL);
9002 if (PL_encoding && ender < 0x100)
9003 goto recode_encoding;
9006 if (! RExC_override_recoding) {
9007 SV* enc = PL_encoding;
9008 ender = reg_recode((const char)(U8)ender, &enc);
9009 if (!enc && SIZE_ONLY)
9010 ckWARNreg(p, "Invalid escape in the specified encoding");
9016 FAIL("Trailing \\");
9019 if (!SIZE_ONLY&& isALPHA(*p)) {
9020 /* Include any { following the alpha to emphasize
9021 * that it could be part of an escape at some point
9023 int len = (*(p + 1) == '{') ? 2 : 1;
9024 ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p);
9026 goto normal_default;
9031 if (UTF8_IS_START(*p) && UTF) {
9033 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
9034 &numlen, UTF8_ALLOW_DEFAULT);
9040 } /* End of switch on the literal */
9042 /* Certain characters are problematic because their folded
9043 * length is so different from their original length that it
9044 * isn't handleable by the optimizer. They are therefore not
9045 * placed in an EXACTish node; and are here handled specially.
9046 * (Even if the optimizer handled LATIN_SMALL_LETTER_SHARP_S,
9047 * putting it in a special node keeps regexec from having to
9048 * deal with a non-utf8 multi-char fold */
9050 && (ender > 255 || (! MORE_ASCII_RESTRICTED && ! LOC)))
9052 /* We look for either side of the fold. For example \xDF
9053 * folds to 'ss'. We look for both the single character
9054 * \xDF and the sequence 'ss'. When we find something that
9055 * could be one of those, we stop and flush whatever we
9056 * have output so far into the EXACTish node that was being
9057 * built. Then restore the input pointer to what it was.
9058 * regatom will return that EXACT node, and will be called
9059 * again, positioned so the first character is the one in
9060 * question, which we return in a different node type.
9061 * The multi-char folds are a sequence, so the occurrence
9062 * of the first character in that sequence doesn't
9063 * necessarily mean that what follows is the rest of the
9064 * sequence. We keep track of that with a state machine,
9065 * with the state being set to the latest character
9066 * processed before the current one. Most characters will
9067 * set the state to 0, but if one occurs that is part of a
9068 * potential tricky fold sequence, the state is set to that
9069 * character, and the next loop iteration sees if the state
9070 * should progress towards the final folded-from character,
9071 * or if it was a false alarm. If it turns out to be a
9072 * false alarm, the character(s) will be output in a new
9073 * EXACTish node, and join_exact() will later combine them.
9074 * In the case of the 'ss' sequence, which is more common
9075 * and more easily checked, some look-ahead is done to
9076 * save time by ruling-out some false alarms */
9079 latest_char_state = generic_char;
9083 case 0x17F: /* LATIN SMALL LETTER LONG S */
9084 if (AT_LEAST_UNI_SEMANTICS) {
9085 if (latest_char_state == char_s) { /* 'ss' */
9086 ender = LATIN_SMALL_LETTER_SHARP_S;
9089 else if (p < RExC_end) {
9091 /* Look-ahead at the next character. If it
9092 * is also an s, we handle as a sharp s
9093 * tricky regnode. */
9094 if (*p == 's' || *p == 'S') {
9096 /* But first flush anything in the
9097 * EXACTish buffer */
9102 p++; /* Account for swallowing this
9104 ender = LATIN_SMALL_LETTER_SHARP_S;
9107 /* Here, the next character is not a
9108 * literal 's', but still could
9109 * evaluate to one if part of a \o{},
9110 * \x or \OCTAL-DIGIT. The minimum
9111 * length required for that is 4, eg
9115 && (isDIGIT(*(p + 1))
9117 || *(p + 1) == 'o' ))
9120 /* Here, it could be an 's', too much
9121 * bother to figure it out here. Flush
9122 * the buffer if any; when come back
9123 * here, set the state so know that the
9124 * previous char was an 's' */
9126 latest_char_state = generic_char;
9130 latest_char_state = char_s;
9136 /* Here, can't be an 'ss' sequence, or at least not
9137 * one that could fold to/from the sharp ss */
9138 latest_char_state = generic_char;
9140 case 0x03C5: /* First char in upsilon series */
9141 case 0x03A5: /* Also capital UPSILON, which folds to
9142 03C5, and hence exhibits the same
9144 if (p < RExC_end - 4) { /* Need >= 4 bytes left */
9145 latest_char_state = upsilon_1;
9152 latest_char_state = generic_char;
9155 case 0x03B9: /* First char in iota series */
9156 case 0x0399: /* Also capital IOTA */
9157 case 0x1FBE: /* GREEK PROSGEGRAMMENI folds to 3B9 */
9158 case 0x0345: /* COMBINING GREEK YPOGEGRAMMENI folds
9160 if (p < RExC_end - 4) {
9161 latest_char_state = iota_1;
9168 latest_char_state = generic_char;
9172 if (latest_char_state == upsilon_1) {
9173 latest_char_state = upsilon_2;
9175 else if (latest_char_state == iota_1) {
9176 latest_char_state = iota_2;
9179 latest_char_state = generic_char;
9183 if (latest_char_state == upsilon_2) {
9184 ender = GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS;
9187 else if (latest_char_state == iota_2) {
9188 ender = GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS;
9191 latest_char_state = generic_char;
9194 /* These are the tricky fold characters. Flush any
9195 * buffer first. (When adding to this list, also should
9196 * add them to fold_grind.t to make sure get tested) */
9197 case GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS:
9198 case GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS:
9199 case LATIN_SMALL_LETTER_SHARP_S:
9200 case LATIN_CAPITAL_LETTER_SHARP_S:
9201 case 0x1FD3: /* GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA */
9202 case 0x1FE3: /* GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA */
9209 char* const oldregxend = RExC_end;
9210 U8 tmpbuf[UTF8_MAXBYTES+1];
9212 /* Here, we know we need to generate a special
9213 * regnode, and 'ender' contains the tricky
9214 * character. What's done is to pretend it's in a
9215 * [bracketed] class, and let the code that deals
9216 * with those handle it, as that code has all the
9217 * intelligence necessary. First save the current
9218 * parse state, get rid of the already allocated
9219 * but empty EXACT node that the ANYOFV node will
9220 * replace, and point the parse to a buffer which
9221 * we fill with the character we want the regclass
9222 * code to think is being parsed */
9223 RExC_emit = orig_emit;
9224 RExC_parse = (char *) tmpbuf;
9226 U8 *d = uvchr_to_utf8(tmpbuf, ender);
9228 RExC_end = (char *) d;
9230 else { /* ender above 255 already excluded */
9231 tmpbuf[0] = (U8) ender;
9233 RExC_end = RExC_parse + 1;
9236 ret = regclass(pRExC_state,depth+1);
9238 /* Here, have parsed the buffer. Reset the parse to
9239 * the actual input, and return */
9240 RExC_end = oldregxend;
9243 Set_Node_Offset(ret, RExC_parse);
9244 Set_Node_Cur_Length(ret);
9245 nextchar(pRExC_state);
9246 *flagp |= HASWIDTH|SIMPLE;
9252 if ( RExC_flags & RXf_PMf_EXTENDED)
9253 p = regwhite( pRExC_state, p );
9255 /* Prime the casefolded buffer. Locale rules, which apply
9256 * only to code points < 256, aren't known until execution,
9257 * so for them, just output the original character using
9259 if (LOC && ender < 256) {
9260 if (UNI_IS_INVARIANT(ender)) {
9261 *tmpbuf = (U8) ender;
9264 *tmpbuf = UTF8_TWO_BYTE_HI(ender);
9265 *(tmpbuf + 1) = UTF8_TWO_BYTE_LO(ender);
9269 else if (isASCII(ender)) { /* Note: Here can't also be LOC
9271 ender = toLOWER(ender);
9272 *tmpbuf = (U8) ender;
9275 else if (! MORE_ASCII_RESTRICTED && ! LOC) {
9277 /* Locale and /aa require more selectivity about the
9278 * fold, so are handled below. Otherwise, here, just
9280 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
9283 /* Under locale rules or /aa we are not to mix,
9284 * respectively, ords < 256 or ASCII with non-. So
9285 * reject folds that mix them, using only the
9286 * non-folded code point. So do the fold to a
9287 * temporary, and inspect each character in it. */
9288 U8 trialbuf[UTF8_MAXBYTES_CASE+1];
9290 UV tmpender = toFOLD_uni(ender, trialbuf, &foldlen);
9291 U8* e = s + foldlen;
9292 bool fold_ok = TRUE;
9296 || (LOC && (UTF8_IS_INVARIANT(*s)
9297 || UTF8_IS_DOWNGRADEABLE_START(*s))))
9305 Copy(trialbuf, tmpbuf, foldlen, U8);
9309 uvuni_to_utf8(tmpbuf, ender);
9310 foldlen = UNISKIP(ender);
9314 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
9319 /* Emit all the Unicode characters. */
9321 for (foldbuf = tmpbuf;
9323 foldlen -= numlen) {
9324 ender = utf8_to_uvchr(foldbuf, &numlen);
9326 const STRLEN unilen = reguni(pRExC_state, ender, s);
9329 /* In EBCDIC the numlen
9330 * and unilen can differ. */
9332 if (numlen >= foldlen)
9336 break; /* "Can't happen." */
9340 const STRLEN unilen = reguni(pRExC_state, ender, s);
9349 REGC((char)ender, s++);
9355 /* Emit all the Unicode characters. */
9357 for (foldbuf = tmpbuf;
9359 foldlen -= numlen) {
9360 ender = utf8_to_uvchr(foldbuf, &numlen);
9362 const STRLEN unilen = reguni(pRExC_state, ender, s);
9365 /* In EBCDIC the numlen
9366 * and unilen can differ. */
9368 if (numlen >= foldlen)
9376 const STRLEN unilen = reguni(pRExC_state, ender, s);
9385 REGC((char)ender, s++);
9388 loopdone: /* Jumped to when encounters something that shouldn't be in
9391 Set_Node_Cur_Length(ret); /* MJD */
9392 nextchar(pRExC_state);
9394 /* len is STRLEN which is unsigned, need to copy to signed */
9397 vFAIL("Internal disaster");
9401 if (len == 1 && UNI_IS_INVARIANT(ender))
9405 RExC_size += STR_SZ(len);
9408 RExC_emit += STR_SZ(len);
9416 /* Jumped to when an unrecognized character set is encountered */
9418 Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags));
9423 S_regwhite( RExC_state_t *pRExC_state, char *p )
9425 const char *e = RExC_end;
9427 PERL_ARGS_ASSERT_REGWHITE;
9432 else if (*p == '#') {
9441 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
9449 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
9450 Character classes ([:foo:]) can also be negated ([:^foo:]).
9451 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
9452 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
9453 but trigger failures because they are currently unimplemented. */
9455 #define POSIXCC_DONE(c) ((c) == ':')
9456 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
9457 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
9460 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
9463 I32 namedclass = OOB_NAMEDCLASS;
9465 PERL_ARGS_ASSERT_REGPPOSIXCC;
9467 if (value == '[' && RExC_parse + 1 < RExC_end &&
9468 /* I smell either [: or [= or [. -- POSIX has been here, right? */
9469 POSIXCC(UCHARAT(RExC_parse))) {
9470 const char c = UCHARAT(RExC_parse);
9471 char* const s = RExC_parse++;
9473 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
9475 if (RExC_parse == RExC_end)
9476 /* Grandfather lone [:, [=, [. */
9479 const char* const t = RExC_parse++; /* skip over the c */
9482 if (UCHARAT(RExC_parse) == ']') {
9483 const char *posixcc = s + 1;
9484 RExC_parse++; /* skip over the ending ] */
9487 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
9488 const I32 skip = t - posixcc;
9490 /* Initially switch on the length of the name. */
9493 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
9494 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
9497 /* Names all of length 5. */
9498 /* alnum alpha ascii blank cntrl digit graph lower
9499 print punct space upper */
9500 /* Offset 4 gives the best switch position. */
9501 switch (posixcc[4]) {
9503 if (memEQ(posixcc, "alph", 4)) /* alpha */
9504 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
9507 if (memEQ(posixcc, "spac", 4)) /* space */
9508 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
9511 if (memEQ(posixcc, "grap", 4)) /* graph */
9512 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
9515 if (memEQ(posixcc, "asci", 4)) /* ascii */
9516 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
9519 if (memEQ(posixcc, "blan", 4)) /* blank */
9520 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
9523 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
9524 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
9527 if (memEQ(posixcc, "alnu", 4)) /* alnum */
9528 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
9531 if (memEQ(posixcc, "lowe", 4)) /* lower */
9532 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
9533 else if (memEQ(posixcc, "uppe", 4)) /* upper */
9534 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
9537 if (memEQ(posixcc, "digi", 4)) /* digit */
9538 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
9539 else if (memEQ(posixcc, "prin", 4)) /* print */
9540 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
9541 else if (memEQ(posixcc, "punc", 4)) /* punct */
9542 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
9547 if (memEQ(posixcc, "xdigit", 6))
9548 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
9552 if (namedclass == OOB_NAMEDCLASS)
9553 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
9555 assert (posixcc[skip] == ':');
9556 assert (posixcc[skip+1] == ']');
9557 } else if (!SIZE_ONLY) {
9558 /* [[=foo=]] and [[.foo.]] are still future. */
9560 /* adjust RExC_parse so the warning shows after
9562 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
9564 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9567 /* Maternal grandfather:
9568 * "[:" ending in ":" but not in ":]" */
9578 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
9582 PERL_ARGS_ASSERT_CHECKPOSIXCC;
9584 if (POSIXCC(UCHARAT(RExC_parse))) {
9585 const char *s = RExC_parse;
9586 const char c = *s++;
9590 if (*s && c == *s && s[1] == ']') {
9592 "POSIX syntax [%c %c] belongs inside character classes",
9595 /* [[=foo=]] and [[.foo.]] are still future. */
9596 if (POSIXCC_NOTYET(c)) {
9597 /* adjust RExC_parse so the error shows after
9599 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
9601 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9607 /* No locale test, and always Unicode semantics, no ignore-case differences */
9608 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
9610 for (value = 0; value < 256; value++) \
9612 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9616 case ANYOF_N##NAME: \
9617 for (value = 0; value < 256; value++) \
9619 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9624 /* Like the above, but there are differences if we are in uni-8-bit or not, so
9625 * there are two tests passed in, to use depending on that. There aren't any
9626 * cases where the label is different from the name, so no need for that
9628 * Sets 'what' to WORD which is the property name for non-bitmap code points;
9629 * But, uses FOLD_WORD instead if /i has been selected, to allow a different
9631 #define _C_C_T_(NAME, TEST_8, TEST_7, WORD, FOLD_WORD) \
9633 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
9634 else if (UNI_SEMANTICS) { \
9635 for (value = 0; value < 256; value++) { \
9636 if (TEST_8(value)) stored += \
9637 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9641 for (value = 0; value < 128; value++) { \
9642 if (TEST_7(UNI_TO_NATIVE(value))) stored += \
9643 set_regclass_bit(pRExC_state, ret, \
9644 (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9655 case ANYOF_N##NAME: \
9656 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
9657 else if (UNI_SEMANTICS) { \
9658 for (value = 0; value < 256; value++) { \
9659 if (! TEST_8(value)) stored += \
9660 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9664 for (value = 0; value < 128; value++) { \
9665 if (! TEST_7(UNI_TO_NATIVE(value))) stored += set_regclass_bit( \
9666 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9668 if (AT_LEAST_ASCII_RESTRICTED) { \
9669 for (value = 128; value < 256; value++) { \
9670 stored += set_regclass_bit( \
9671 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9673 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; \
9676 /* For a non-ut8 target string with DEPENDS semantics, all above \
9677 * ASCII Latin1 code points match the complement of any of the \
9678 * classes. But in utf8, they have their Unicode semantics, so \
9679 * can't just set them in the bitmap, or else regexec.c will think \
9680 * they matched when they shouldn't. */ \
9681 ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; \
9694 S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr)
9697 /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes.
9698 * Locale folding is done at run-time, so this function should not be
9699 * called for nodes that are for locales.
9701 * This function sets the bit corresponding to the fold of the input
9702 * 'value', if not already set. The fold of 'f' is 'F', and the fold of
9705 * It also knows about the characters that are in the bitmap that have
9706 * folds that are matchable only outside it, and sets the appropriate lists
9709 * It returns the number of bits that actually changed from 0 to 1 */
9714 PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD;
9716 fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value]
9719 /* It assumes the bit for 'value' has already been set */
9720 if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) {
9721 ANYOF_BITMAP_SET(node, fold);
9724 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) {
9725 /* Certain Latin1 characters have matches outside the bitmap. To get
9726 * here, 'value' is one of those characters. None of these matches is
9727 * valid for ASCII characters under /aa, which have been excluded by
9728 * the 'if' above. The matches fall into three categories:
9729 * 1) They are singly folded-to or -from an above 255 character, as
9730 * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y
9732 * 2) They are part of a multi-char fold with another character in the
9733 * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill;
9734 * 3) They are part of a multi-char fold with a character not in the
9735 * bitmap, such as various ligatures.
9736 * We aren't dealing fully with multi-char folds, except we do deal
9737 * with the pattern containing a character that has a multi-char fold
9738 * (not so much the inverse).
9739 * For types 1) and 3), the matches only happen when the target string
9740 * is utf8; that's not true for 2), and we set a flag for it.
9742 * The code below adds to the passed in inversion list the single fold
9743 * closures for 'value'. The values are hard-coded here so that an
9744 * innocent-looking character class, like /[ks]/i won't have to go out
9745 * to disk to find the possible matches. XXX It would be better to
9746 * generate these via regen, in case a new version of the Unicode
9747 * standard adds new mappings, though that is not really likely. */
9752 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A);
9756 /* LATIN SMALL LETTER LONG S */
9757 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F);
9760 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9761 GREEK_SMALL_LETTER_MU);
9762 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9763 GREEK_CAPITAL_LETTER_MU);
9765 case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
9766 case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
9768 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B);
9769 if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */
9770 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9771 PL_fold_latin1[value]);
9774 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
9775 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9776 LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
9778 case LATIN_SMALL_LETTER_SHARP_S:
9779 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9780 LATIN_CAPITAL_LETTER_SHARP_S);
9782 /* Under /a, /d, and /u, this can match the two chars "ss" */
9783 if (! MORE_ASCII_RESTRICTED) {
9784 add_alternate(alternate_ptr, (U8 *) "ss", 2);
9786 /* And under /u or /a, it can match even if the target is
9788 if (AT_LEAST_UNI_SEMANTICS) {
9789 ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8;
9803 /* These all are targets of multi-character folds from code
9804 * points that require UTF8 to express, so they can't match
9805 * unless the target string is in UTF-8, so no action here is
9806 * necessary, as regexec.c properly handles the general case
9807 * for UTF-8 matching */
9810 /* Use deprecated warning to increase the chances of this
9812 ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value);
9816 else if (DEPENDS_SEMANTICS
9818 && PL_fold_latin1[value] != value)
9820 /* Under DEPENDS rules, non-ASCII Latin1 characters match their
9821 * folds only when the target string is in UTF-8. We add the fold
9822 * here to the list of things to match outside the bitmap, which
9823 * won't be looked at unless it is UTF8 (or else if something else
9824 * says to look even if not utf8, but those things better not happen
9825 * under DEPENDS semantics. */
9826 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]);
9833 PERL_STATIC_INLINE U8
9834 S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr)
9836 /* This inline function sets a bit in the bitmap if not already set, and if
9837 * appropriate, its fold, returning the number of bits that actually
9838 * changed from 0 to 1 */
9842 PERL_ARGS_ASSERT_SET_REGCLASS_BIT;
9844 if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */
9848 ANYOF_BITMAP_SET(node, value);
9851 if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */
9852 stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr);
9859 S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len)
9861 /* Adds input 'string' with length 'len' to the ANYOF node's unicode
9862 * alternate list, pointed to by 'alternate_ptr'. This is an array of
9863 * the multi-character folds of characters in the node */
9866 PERL_ARGS_ASSERT_ADD_ALTERNATE;
9868 if (! *alternate_ptr) {
9869 *alternate_ptr = newAV();
9871 sv = newSVpvn_utf8((char*)string, len, TRUE);
9872 av_push(*alternate_ptr, sv);
9877 parse a class specification and produce either an ANYOF node that
9878 matches the pattern or perhaps will be optimized into an EXACTish node
9879 instead. The node contains a bit map for the first 256 characters, with the
9880 corresponding bit set if that character is in the list. For characters
9881 above 255, a range list is used */
9884 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
9887 register UV nextvalue;
9888 register IV prevvalue = OOB_UNICODE;
9889 register IV range = 0;
9890 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
9891 register regnode *ret;
9894 char *rangebegin = NULL;
9895 bool need_class = 0;
9896 bool allow_full_fold = TRUE; /* Assume wants multi-char folding */
9898 STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more
9899 than just initialized. */
9902 /* code points this node matches that can't be stored in the bitmap */
9903 SV* nonbitmap = NULL;
9905 /* The items that are to match that aren't stored in the bitmap, but are a
9906 * result of things that are stored there. This is the fold closure of
9907 * such a character, either because it has DEPENDS semantics and shouldn't
9908 * be matched unless the target string is utf8, or is a code point that is
9909 * too large for the bit map, as for example, the fold of the MICRO SIGN is
9910 * above 255. This all is solely for performance reasons. By having this
9911 * code know the outside-the-bitmap folds that the bitmapped characters are
9912 * involved with, we don't have to go out to disk to find the list of
9913 * matches, unless the character class includes code points that aren't
9914 * storable in the bit map. That means that a character class with an 's'
9915 * in it, for example, doesn't need to go out to disk to find everything
9916 * that matches. A 2nd list is used so that the 'nonbitmap' list is kept
9917 * empty unless there is something whose fold we don't know about, and will
9918 * have to go out to the disk to find. */
9919 SV* l1_fold_invlist = NULL;
9921 /* List of multi-character folds that are matched by this node */
9922 AV* unicode_alternate = NULL;
9924 UV literal_endpoint = 0;
9926 UV stored = 0; /* how many chars stored in the bitmap */
9928 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
9929 case we need to change the emitted regop to an EXACT. */
9930 const char * orig_parse = RExC_parse;
9931 GET_RE_DEBUG_FLAGS_DECL;
9933 PERL_ARGS_ASSERT_REGCLASS;
9935 PERL_UNUSED_ARG(depth);
9938 DEBUG_PARSE("clas");
9940 /* Assume we are going to generate an ANYOF node. */
9941 ret = reganode(pRExC_state, ANYOF, 0);
9945 ANYOF_FLAGS(ret) = 0;
9948 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
9952 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
9954 /* We have decided to not allow multi-char folds in inverted character
9955 * classes, due to the confusion that can happen, especially with
9956 * classes that are designed for a non-Unicode world: You have the
9957 * peculiar case that:
9958 "s s" =~ /^[^\xDF]+$/i => Y
9959 "ss" =~ /^[^\xDF]+$/i => N
9961 * See [perl #89750] */
9962 allow_full_fold = FALSE;
9966 RExC_size += ANYOF_SKIP;
9967 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
9970 RExC_emit += ANYOF_SKIP;
9972 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
9974 ANYOF_BITMAP_ZERO(ret);
9975 listsv = newSVpvs("# comment\n");
9976 initial_listsv_len = SvCUR(listsv);
9979 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9981 if (!SIZE_ONLY && POSIXCC(nextvalue))
9982 checkposixcc(pRExC_state);
9984 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
9985 if (UCHARAT(RExC_parse) == ']')
9989 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
9993 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
9996 rangebegin = RExC_parse;
9998 value = utf8n_to_uvchr((U8*)RExC_parse,
9999 RExC_end - RExC_parse,
10000 &numlen, UTF8_ALLOW_DEFAULT);
10001 RExC_parse += numlen;
10004 value = UCHARAT(RExC_parse++);
10006 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
10007 if (value == '[' && POSIXCC(nextvalue))
10008 namedclass = regpposixcc(pRExC_state, value);
10009 else if (value == '\\') {
10011 value = utf8n_to_uvchr((U8*)RExC_parse,
10012 RExC_end - RExC_parse,
10013 &numlen, UTF8_ALLOW_DEFAULT);
10014 RExC_parse += numlen;
10017 value = UCHARAT(RExC_parse++);
10018 /* Some compilers cannot handle switching on 64-bit integer
10019 * values, therefore value cannot be an UV. Yes, this will
10020 * be a problem later if we want switch on Unicode.
10021 * A similar issue a little bit later when switching on
10022 * namedclass. --jhi */
10023 switch ((I32)value) {
10024 case 'w': namedclass = ANYOF_ALNUM; break;
10025 case 'W': namedclass = ANYOF_NALNUM; break;
10026 case 's': namedclass = ANYOF_SPACE; break;
10027 case 'S': namedclass = ANYOF_NSPACE; break;
10028 case 'd': namedclass = ANYOF_DIGIT; break;
10029 case 'D': namedclass = ANYOF_NDIGIT; break;
10030 case 'v': namedclass = ANYOF_VERTWS; break;
10031 case 'V': namedclass = ANYOF_NVERTWS; break;
10032 case 'h': namedclass = ANYOF_HORIZWS; break;
10033 case 'H': namedclass = ANYOF_NHORIZWS; break;
10034 case 'N': /* Handle \N{NAME} in class */
10036 /* We only pay attention to the first char of
10037 multichar strings being returned. I kinda wonder
10038 if this makes sense as it does change the behaviour
10039 from earlier versions, OTOH that behaviour was broken
10041 UV v; /* value is register so we cant & it /grrr */
10042 if (reg_namedseq(pRExC_state, &v, NULL, depth)) {
10052 if (RExC_parse >= RExC_end)
10053 vFAIL2("Empty \\%c{}", (U8)value);
10054 if (*RExC_parse == '{') {
10055 const U8 c = (U8)value;
10056 e = strchr(RExC_parse++, '}');
10058 vFAIL2("Missing right brace on \\%c{}", c);
10059 while (isSPACE(UCHARAT(RExC_parse)))
10061 if (e == RExC_parse)
10062 vFAIL2("Empty \\%c{}", c);
10063 n = e - RExC_parse;
10064 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
10072 if (UCHARAT(RExC_parse) == '^') {
10075 value = value == 'p' ? 'P' : 'p'; /* toggle */
10076 while (isSPACE(UCHARAT(RExC_parse))) {
10082 /* Add the property name to the list. If /i matching, give
10083 * a different name which consists of the normal name
10084 * sandwiched between two underscores and '_i'. The design
10085 * is discussed in the commit message for this. */
10086 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s%.*s%s\n",
10087 (value=='p' ? '+' : '!'),
10088 (FOLD) ? "__" : "",
10094 RExC_parse = e + 1;
10096 /* The \p could match something in the Latin1 range, hence
10097 * something that isn't utf8 */
10098 ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8;
10099 namedclass = ANYOF_MAX; /* no official name, but it's named */
10101 /* \p means they want Unicode semantics */
10102 RExC_uni_semantics = 1;
10105 case 'n': value = '\n'; break;
10106 case 'r': value = '\r'; break;
10107 case 't': value = '\t'; break;
10108 case 'f': value = '\f'; break;
10109 case 'b': value = '\b'; break;
10110 case 'e': value = ASCII_TO_NATIVE('\033');break;
10111 case 'a': value = ASCII_TO_NATIVE('\007');break;
10113 RExC_parse--; /* function expects to be pointed at the 'o' */
10115 const char* error_msg;
10116 bool valid = grok_bslash_o(RExC_parse,
10121 RExC_parse += numlen;
10126 if (PL_encoding && value < 0x100) {
10127 goto recode_encoding;
10131 if (*RExC_parse == '{') {
10132 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
10133 | PERL_SCAN_DISALLOW_PREFIX;
10134 char * const e = strchr(RExC_parse++, '}');
10136 vFAIL("Missing right brace on \\x{}");
10138 numlen = e - RExC_parse;
10139 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
10140 RExC_parse = e + 1;
10143 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
10145 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
10146 RExC_parse += numlen;
10148 if (PL_encoding && value < 0x100)
10149 goto recode_encoding;
10152 value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY);
10154 case '0': case '1': case '2': case '3': case '4':
10155 case '5': case '6': case '7':
10157 /* Take 1-3 octal digits */
10158 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
10160 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
10161 RExC_parse += numlen;
10162 if (PL_encoding && value < 0x100)
10163 goto recode_encoding;
10167 if (! RExC_override_recoding) {
10168 SV* enc = PL_encoding;
10169 value = reg_recode((const char)(U8)value, &enc);
10170 if (!enc && SIZE_ONLY)
10171 ckWARNreg(RExC_parse,
10172 "Invalid escape in the specified encoding");
10176 /* Allow \_ to not give an error */
10177 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
10178 ckWARN2reg(RExC_parse,
10179 "Unrecognized escape \\%c in character class passed through",
10184 } /* end of \blah */
10187 literal_endpoint++;
10190 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
10192 /* What matches in a locale is not known until runtime, so need to
10193 * (one time per class) allocate extra space to pass to regexec.
10194 * The space will contain a bit for each named class that is to be
10195 * matched against. This isn't needed for \p{} and pseudo-classes,
10196 * as they are not affected by locale, and hence are dealt with
10198 if (LOC && namedclass < ANYOF_MAX && ! need_class) {
10201 RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP;
10204 RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP;
10205 ANYOF_CLASS_ZERO(ret);
10207 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
10210 /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a
10211 * literal, as is the character that began the false range, i.e.
10212 * the 'a' in the examples */
10216 RExC_parse >= rangebegin ?
10217 RExC_parse - rangebegin : 0;
10218 ckWARN4reg(RExC_parse,
10219 "False [] range \"%*.*s\"",
10223 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
10224 if (prevvalue < 256) {
10226 set_regclass_bit(pRExC_state, ret, (U8) prevvalue, &l1_fold_invlist, &unicode_alternate);
10229 nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue);
10233 range = 0; /* this was not a true range */
10239 const char *what = NULL;
10242 /* Possible truncation here but in some 64-bit environments
10243 * the compiler gets heartburn about switch on 64-bit values.
10244 * A similar issue a little earlier when switching on value.
10246 switch ((I32)namedclass) {
10248 case _C_C_T_(ALNUMC, isALNUMC_L1, isALNUMC, "XPosixAlnum", "XPosixAlnum");
10249 case _C_C_T_(ALPHA, isALPHA_L1, isALPHA, "XPosixAlpha", "XPosixAlpha");
10250 case _C_C_T_(BLANK, isBLANK_L1, isBLANK, "XPosixBlank", "XPosixBlank");
10251 case _C_C_T_(CNTRL, isCNTRL_L1, isCNTRL, "XPosixCntrl", "XPosixCntrl");
10252 case _C_C_T_(GRAPH, isGRAPH_L1, isGRAPH, "XPosixGraph", "XPosixGraph");
10253 case _C_C_T_(LOWER, isLOWER_L1, isLOWER, "XPosixLower", "__XPosixLower_i");
10254 case _C_C_T_(PRINT, isPRINT_L1, isPRINT, "XPosixPrint", "XPosixPrint");
10255 case _C_C_T_(PSXSPC, isPSXSPC_L1, isPSXSPC, "XPosixSpace", "XPosixSpace");
10256 case _C_C_T_(PUNCT, isPUNCT_L1, isPUNCT, "XPosixPunct", "XPosixPunct");
10257 case _C_C_T_(UPPER, isUPPER_L1, isUPPER, "XPosixUpper", "__XPosixUpper_i");
10258 /* \s, \w match all unicode if utf8. */
10259 case _C_C_T_(SPACE, isSPACE_L1, isSPACE, "SpacePerl", "SpacePerl");
10260 case _C_C_T_(ALNUM, isWORDCHAR_L1, isALNUM, "Word", "Word");
10261 case _C_C_T_(XDIGIT, isXDIGIT_L1, isXDIGIT, "XPosixXDigit", "XPosixXDigit");
10262 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
10263 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
10266 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
10268 for (value = 0; value < 128; value++)
10270 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
10273 what = NULL; /* Doesn't match outside ascii, so
10274 don't want to add +utf8:: */
10278 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
10280 for (value = 128; value < 256; value++)
10282 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
10284 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
10290 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
10292 /* consecutive digits assumed */
10293 for (value = '0'; value <= '9'; value++)
10295 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
10302 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
10304 /* consecutive digits assumed */
10305 for (value = 0; value < '0'; value++)
10307 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
10308 for (value = '9' + 1; value < 256; value++)
10310 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
10314 if (AT_LEAST_ASCII_RESTRICTED ) {
10315 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
10319 /* this is to handle \p and \P */
10322 vFAIL("Invalid [::] class");
10325 if (what && ! (AT_LEAST_ASCII_RESTRICTED)) {
10326 /* Strings such as "+utf8::isWord\n" */
10327 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s\n", yesno, what);
10332 } /* end of namedclass \blah */
10335 if (prevvalue > (IV)value) /* b-a */ {
10336 const int w = RExC_parse - rangebegin;
10337 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
10338 range = 0; /* not a valid range */
10342 prevvalue = value; /* save the beginning of the range */
10343 if (RExC_parse+1 < RExC_end
10344 && *RExC_parse == '-'
10345 && RExC_parse[1] != ']')
10349 /* a bad range like \w-, [:word:]- ? */
10350 if (namedclass > OOB_NAMEDCLASS) {
10351 if (ckWARN(WARN_REGEXP)) {
10353 RExC_parse >= rangebegin ?
10354 RExC_parse - rangebegin : 0;
10356 "False [] range \"%*.*s\"",
10361 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
10363 range = 1; /* yeah, it's a range! */
10364 continue; /* but do it the next time */
10368 /* non-Latin1 code point implies unicode semantics. Must be set in
10369 * pass1 so is there for the whole of pass 2 */
10371 RExC_uni_semantics = 1;
10374 /* now is the next time */
10376 if (prevvalue < 256) {
10377 const IV ceilvalue = value < 256 ? value : 255;
10380 /* In EBCDIC [\x89-\x91] should include
10381 * the \x8e but [i-j] should not. */
10382 if (literal_endpoint == 2 &&
10383 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
10384 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
10386 if (isLOWER(prevvalue)) {
10387 for (i = prevvalue; i <= ceilvalue; i++)
10388 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10390 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10393 for (i = prevvalue; i <= ceilvalue; i++)
10394 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10396 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10402 for (i = prevvalue; i <= ceilvalue; i++) {
10403 stored += set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10407 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
10408 const UV natvalue = NATIVE_TO_UNI(value);
10409 nonbitmap = add_range_to_invlist(nonbitmap, prevnatvalue, natvalue);
10412 literal_endpoint = 0;
10416 range = 0; /* this range (if it was one) is done now */
10423 /****** !SIZE_ONLY AFTER HERE *********/
10425 /* If folding and there are code points above 255, we calculate all
10426 * characters that could fold to or from the ones already on the list */
10427 if (FOLD && nonbitmap) {
10428 UV start, end; /* End points of code point ranges */
10430 SV* fold_intersection;
10432 /* This is a list of all the characters that participate in folds
10433 * (except marks, etc in multi-char folds */
10434 if (! PL_utf8_foldable) {
10435 SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0);
10436 PL_utf8_foldable = _swash_to_invlist(swash);
10439 /* This is a hash that for a particular fold gives all characters
10440 * that are involved in it */
10441 if (! PL_utf8_foldclosures) {
10443 /* If we were unable to find any folds, then we likely won't be
10444 * able to find the closures. So just create an empty list.
10445 * Folding will effectively be restricted to the non-Unicode rules
10446 * hard-coded into Perl. (This case happens legitimately during
10447 * compilation of Perl itself before the Unicode tables are
10449 if (invlist_len(PL_utf8_foldable) == 0) {
10450 PL_utf8_foldclosures = newHV();
10452 /* If the folds haven't been read in, call a fold function
10454 if (! PL_utf8_tofold) {
10455 U8 dummy[UTF8_MAXBYTES+1];
10458 /* This particular string is above \xff in both UTF-8 and
10460 to_utf8_fold((U8*) "\xC8\x80", dummy, &dummy_len);
10461 assert(PL_utf8_tofold); /* Verify that worked */
10463 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
10467 /* Only the characters in this class that participate in folds need
10468 * be checked. Get the intersection of this class and all the
10469 * possible characters that are foldable. This can quickly narrow
10470 * down a large class */
10471 _invlist_intersection(PL_utf8_foldable, nonbitmap, &fold_intersection);
10473 /* Now look at the foldable characters in this class individually */
10474 invlist_iterinit(fold_intersection);
10475 while (invlist_iternext(fold_intersection, &start, &end)) {
10478 /* Look at every character in the range */
10479 for (j = start; j <= end; j++) {
10482 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
10485 _to_uni_fold_flags(j, foldbuf, &foldlen, allow_full_fold);
10487 if (foldlen > (STRLEN)UNISKIP(f)) {
10489 /* Any multicharacter foldings (disallowed in
10490 * lookbehind patterns) require the following
10491 * transform: [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst) where
10492 * E folds into "pq" and F folds into "rst", all other
10493 * characters fold to single characters. We save away
10494 * these multicharacter foldings, to be later saved as
10495 * part of the additional "s" data. */
10496 if (! RExC_in_lookbehind) {
10498 U8* e = foldbuf + foldlen;
10500 /* If any of the folded characters of this are in
10501 * the Latin1 range, tell the regex engine that
10502 * this can match a non-utf8 target string. The
10503 * only multi-byte fold whose source is in the
10504 * Latin1 range (U+00DF) applies only when the
10505 * target string is utf8, or under unicode rules */
10506 if (j > 255 || AT_LEAST_UNI_SEMANTICS) {
10509 /* Can't mix ascii with non- under /aa */
10510 if (MORE_ASCII_RESTRICTED
10511 && (isASCII(*loc) != isASCII(j)))
10513 goto end_multi_fold;
10515 if (UTF8_IS_INVARIANT(*loc)
10516 || UTF8_IS_DOWNGRADEABLE_START(*loc))
10518 /* Can't mix above and below 256 under
10521 goto end_multi_fold;
10524 |= ANYOF_NONBITMAP_NON_UTF8;
10527 loc += UTF8SKIP(loc);
10531 add_alternate(&unicode_alternate, foldbuf, foldlen);
10535 /* This is special-cased, as it is the only letter which
10536 * has both a multi-fold and single-fold in Latin1. All
10537 * the other chars that have single and multi-folds are
10538 * always in utf8, and the utf8 folding algorithm catches
10540 if (! LOC && j == LATIN_CAPITAL_LETTER_SHARP_S) {
10541 stored += set_regclass_bit(pRExC_state,
10543 LATIN_SMALL_LETTER_SHARP_S,
10544 &l1_fold_invlist, &unicode_alternate);
10548 /* Single character fold. Add everything in its fold
10549 * closure to the list that this node should match */
10552 /* The fold closures data structure is a hash with the
10553 * keys being every character that is folded to, like
10554 * 'k', and the values each an array of everything that
10555 * folds to its key. e.g. [ 'k', 'K', KELVIN_SIGN ] */
10556 if ((listp = hv_fetch(PL_utf8_foldclosures,
10557 (char *) foldbuf, foldlen, FALSE)))
10559 AV* list = (AV*) *listp;
10561 for (k = 0; k <= av_len(list); k++) {
10562 SV** c_p = av_fetch(list, k, FALSE);
10565 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
10569 /* /aa doesn't allow folds between ASCII and
10570 * non-; /l doesn't allow them between above
10572 if ((MORE_ASCII_RESTRICTED
10573 && (isASCII(c) != isASCII(j)))
10574 || (LOC && ((c < 256) != (j < 256))))
10579 if (c < 256 && AT_LEAST_UNI_SEMANTICS) {
10580 stored += set_regclass_bit(pRExC_state,
10583 &l1_fold_invlist, &unicode_alternate);
10585 /* It may be that the code point is already
10586 * in this range or already in the bitmap,
10587 * in which case we need do nothing */
10588 else if ((c < start || c > end)
10590 || ! ANYOF_BITMAP_TEST(ret, c)))
10592 nonbitmap = add_cp_to_invlist(nonbitmap, c);
10599 SvREFCNT_dec(fold_intersection);
10602 /* Combine the two lists into one. */
10603 if (l1_fold_invlist) {
10605 _invlist_union(nonbitmap, l1_fold_invlist, &nonbitmap);
10606 SvREFCNT_dec(l1_fold_invlist);
10609 nonbitmap = l1_fold_invlist;
10613 /* Here, we have calculated what code points should be in the character
10614 * class. Now we can see about various optimizations. Fold calculation
10615 * needs to take place before inversion. Otherwise /[^k]/i would invert to
10616 * include K, which under /i would match k. */
10618 /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't
10619 * set the FOLD flag yet, so this this does optimize those. It doesn't
10620 * optimize locale. Doing so perhaps could be done as long as there is
10621 * nothing like \w in it; some thought also would have to be given to the
10622 * interaction with above 0x100 chars */
10624 && (ANYOF_FLAGS(ret) & ANYOF_INVERT)
10625 && ! unicode_alternate
10626 /* In case of /d, there are some things that should match only when in
10627 * not in the bitmap, i.e., they require UTF8 to match. These are
10628 * listed in nonbitmap. */
10630 || ! DEPENDS_SEMANTICS
10631 || (ANYOF_FLAGS(ret) & ANYOF_NONBITMAP_NON_UTF8))
10632 && SvCUR(listsv) == initial_listsv_len)
10635 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
10636 ANYOF_BITMAP(ret)[value] ^= 0xFF;
10637 /* The inversion means that everything above 255 is matched */
10638 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
10641 /* Here, also has things outside the bitmap. Go through each bit
10642 * individually and add it to the list to get rid of from those
10643 * things not in the bitmap */
10644 SV *remove_list = _new_invlist(2);
10645 _invlist_invert(nonbitmap);
10646 for (value = 0; value < 256; ++value) {
10647 if (ANYOF_BITMAP_TEST(ret, value)) {
10648 ANYOF_BITMAP_CLEAR(ret, value);
10649 remove_list = add_cp_to_invlist(remove_list, value);
10652 ANYOF_BITMAP_SET(ret, value);
10655 _invlist_subtract(nonbitmap, remove_list, &nonbitmap);
10656 SvREFCNT_dec(remove_list);
10659 stored = 256 - stored;
10661 /* Clear the invert flag since have just done it here */
10662 ANYOF_FLAGS(ret) &= ~ANYOF_INVERT;
10665 /* Folding in the bitmap is taken care of above, but not for locale (for
10666 * which we have to wait to see what folding is in effect at runtime), and
10667 * for things not in the bitmap. Set run-time fold flag for these */
10668 if (FOLD && (LOC || nonbitmap || unicode_alternate)) {
10669 ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD;
10672 /* A single character class can be "optimized" into an EXACTish node.
10673 * Note that since we don't currently count how many characters there are
10674 * outside the bitmap, we are XXX missing optimization possibilities for
10675 * them. This optimization can't happen unless this is a truly single
10676 * character class, which means that it can't be an inversion into a
10677 * many-character class, and there must be no possibility of there being
10678 * things outside the bitmap. 'stored' (only) for locales doesn't include
10679 * \w, etc, so have to make a special test that they aren't present
10681 * Similarly A 2-character class of the very special form like [bB] can be
10682 * optimized into an EXACTFish node, but only for non-locales, and for
10683 * characters which only have the two folds; so things like 'fF' and 'Ii'
10684 * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE
10687 && ! unicode_alternate
10688 && SvCUR(listsv) == initial_listsv_len
10689 && ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL))
10690 && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10691 || (! ANYOF_CLASS_TEST_ANY_SET(ret)))))
10692 || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10693 && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value))
10694 /* If the latest code point has a fold whose
10695 * bit is set, it must be the only other one */
10696 && ((prevvalue = PL_fold_latin1[value]) != (IV)value)
10697 && ANYOF_BITMAP_TEST(ret, prevvalue)))))
10699 /* Note that the information needed to decide to do this optimization
10700 * is not currently available until the 2nd pass, and that the actually
10701 * used EXACTish node takes less space than the calculated ANYOF node,
10702 * and hence the amount of space calculated in the first pass is larger
10703 * than actually used, so this optimization doesn't gain us any space.
10704 * But an EXACT node is faster than an ANYOF node, and can be combined
10705 * with any adjacent EXACT nodes later by the optimizer for further
10706 * gains. The speed of executing an EXACTF is similar to an ANYOF
10707 * node, so the optimization advantage comes from the ability to join
10708 * it to adjacent EXACT nodes */
10710 const char * cur_parse= RExC_parse;
10712 RExC_emit = (regnode *)orig_emit;
10713 RExC_parse = (char *)orig_parse;
10717 /* A locale node with one point can be folded; all the other cases
10718 * with folding will have two points, since we calculate them above
10720 if (ANYOF_FLAGS(ret) & ANYOF_LOC_NONBITMAP_FOLD) {
10727 else { /* else 2 chars in the bit map: the folds of each other */
10729 /* Use the folded value, which for the cases where we get here,
10730 * is just the lower case of the current one (which may resolve to
10731 * itself, or to the other one */
10732 value = toLOWER_LATIN1(value);
10733 if (AT_LEAST_UNI_SEMANTICS || !isASCII(value)) {
10735 /* To join adjacent nodes, they must be the exact EXACTish
10736 * type. Try to use the most likely type, by using EXACTFU if
10737 * the regex calls for them, or is required because the
10738 * character is non-ASCII */
10741 else { /* Otherwise, more likely to be EXACTF type */
10746 ret = reg_node(pRExC_state, op);
10747 RExC_parse = (char *)cur_parse;
10748 if (UTF && ! NATIVE_IS_INVARIANT(value)) {
10749 *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value);
10750 *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value);
10752 RExC_emit += STR_SZ(2);
10755 *STRING(ret)= (char)value;
10757 RExC_emit += STR_SZ(1);
10759 SvREFCNT_dec(listsv);
10765 invlist_iterinit(nonbitmap);
10766 while (invlist_iternext(nonbitmap, &start, &end)) {
10767 if (start == end) {
10768 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", start);
10771 /* The \t sets the whole range */
10772 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
10777 SvREFCNT_dec(nonbitmap);
10780 if (SvCUR(listsv) == initial_listsv_len && ! unicode_alternate) {
10781 ARG_SET(ret, ANYOF_NONBITMAP_EMPTY);
10782 SvREFCNT_dec(listsv);
10783 SvREFCNT_dec(unicode_alternate);
10787 AV * const av = newAV();
10789 /* The 0th element stores the character class description
10790 * in its textual form: used later (regexec.c:Perl_regclass_swash())
10791 * to initialize the appropriate swash (which gets stored in
10792 * the 1st element), and also useful for dumping the regnode.
10793 * The 2nd element stores the multicharacter foldings,
10794 * used later (regexec.c:S_reginclass()). */
10795 av_store(av, 0, listsv);
10796 av_store(av, 1, NULL);
10798 /* Store any computed multi-char folds only if we are allowing
10800 if (allow_full_fold) {
10801 av_store(av, 2, MUTABLE_SV(unicode_alternate));
10802 if (unicode_alternate) { /* This node is variable length */
10807 av_store(av, 2, NULL);
10809 rv = newRV_noinc(MUTABLE_SV(av));
10810 n = add_data(pRExC_state, 1, "s");
10811 RExC_rxi->data->data[n] = (void*)rv;
10819 /* reg_skipcomment()
10821 Absorbs an /x style # comments from the input stream.
10822 Returns true if there is more text remaining in the stream.
10823 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
10824 terminates the pattern without including a newline.
10826 Note its the callers responsibility to ensure that we are
10827 actually in /x mode
10832 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
10836 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
10838 while (RExC_parse < RExC_end)
10839 if (*RExC_parse++ == '\n') {
10844 /* we ran off the end of the pattern without ending
10845 the comment, so we have to add an \n when wrapping */
10846 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
10854 Advances the parse position, and optionally absorbs
10855 "whitespace" from the inputstream.
10857 Without /x "whitespace" means (?#...) style comments only,
10858 with /x this means (?#...) and # comments and whitespace proper.
10860 Returns the RExC_parse point from BEFORE the scan occurs.
10862 This is the /x friendly way of saying RExC_parse++.
10866 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
10868 char* const retval = RExC_parse++;
10870 PERL_ARGS_ASSERT_NEXTCHAR;
10873 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
10874 RExC_parse[2] == '#') {
10875 while (*RExC_parse != ')') {
10876 if (RExC_parse == RExC_end)
10877 FAIL("Sequence (?#... not terminated");
10883 if (RExC_flags & RXf_PMf_EXTENDED) {
10884 if (isSPACE(*RExC_parse)) {
10888 else if (*RExC_parse == '#') {
10889 if ( reg_skipcomment( pRExC_state ) )
10898 - reg_node - emit a node
10900 STATIC regnode * /* Location. */
10901 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
10904 register regnode *ptr;
10905 regnode * const ret = RExC_emit;
10906 GET_RE_DEBUG_FLAGS_DECL;
10908 PERL_ARGS_ASSERT_REG_NODE;
10911 SIZE_ALIGN(RExC_size);
10915 if (RExC_emit >= RExC_emit_bound)
10916 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10918 NODE_ALIGN_FILL(ret);
10920 FILL_ADVANCE_NODE(ptr, op);
10921 #ifdef RE_TRACK_PATTERN_OFFSETS
10922 if (RExC_offsets) { /* MJD */
10923 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
10924 "reg_node", __LINE__,
10926 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
10927 ? "Overwriting end of array!\n" : "OK",
10928 (UV)(RExC_emit - RExC_emit_start),
10929 (UV)(RExC_parse - RExC_start),
10930 (UV)RExC_offsets[0]));
10931 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
10939 - reganode - emit a node with an argument
10941 STATIC regnode * /* Location. */
10942 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
10945 register regnode *ptr;
10946 regnode * const ret = RExC_emit;
10947 GET_RE_DEBUG_FLAGS_DECL;
10949 PERL_ARGS_ASSERT_REGANODE;
10952 SIZE_ALIGN(RExC_size);
10957 assert(2==regarglen[op]+1);
10959 Anything larger than this has to allocate the extra amount.
10960 If we changed this to be:
10962 RExC_size += (1 + regarglen[op]);
10964 then it wouldn't matter. Its not clear what side effect
10965 might come from that so its not done so far.
10970 if (RExC_emit >= RExC_emit_bound)
10971 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10973 NODE_ALIGN_FILL(ret);
10975 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
10976 #ifdef RE_TRACK_PATTERN_OFFSETS
10977 if (RExC_offsets) { /* MJD */
10978 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10982 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
10983 "Overwriting end of array!\n" : "OK",
10984 (UV)(RExC_emit - RExC_emit_start),
10985 (UV)(RExC_parse - RExC_start),
10986 (UV)RExC_offsets[0]));
10987 Set_Cur_Node_Offset;
10995 - reguni - emit (if appropriate) a Unicode character
10998 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
11002 PERL_ARGS_ASSERT_REGUNI;
11004 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
11008 - reginsert - insert an operator in front of already-emitted operand
11010 * Means relocating the operand.
11013 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
11016 register regnode *src;
11017 register regnode *dst;
11018 register regnode *place;
11019 const int offset = regarglen[(U8)op];
11020 const int size = NODE_STEP_REGNODE + offset;
11021 GET_RE_DEBUG_FLAGS_DECL;
11023 PERL_ARGS_ASSERT_REGINSERT;
11024 PERL_UNUSED_ARG(depth);
11025 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
11026 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
11035 if (RExC_open_parens) {
11037 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
11038 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
11039 if ( RExC_open_parens[paren] >= opnd ) {
11040 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
11041 RExC_open_parens[paren] += size;
11043 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
11045 if ( RExC_close_parens[paren] >= opnd ) {
11046 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
11047 RExC_close_parens[paren] += size;
11049 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
11054 while (src > opnd) {
11055 StructCopy(--src, --dst, regnode);
11056 #ifdef RE_TRACK_PATTERN_OFFSETS
11057 if (RExC_offsets) { /* MJD 20010112 */
11058 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
11062 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
11063 ? "Overwriting end of array!\n" : "OK",
11064 (UV)(src - RExC_emit_start),
11065 (UV)(dst - RExC_emit_start),
11066 (UV)RExC_offsets[0]));
11067 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
11068 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
11074 place = opnd; /* Op node, where operand used to be. */
11075 #ifdef RE_TRACK_PATTERN_OFFSETS
11076 if (RExC_offsets) { /* MJD */
11077 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
11081 (UV)(place - RExC_emit_start) > RExC_offsets[0]
11082 ? "Overwriting end of array!\n" : "OK",
11083 (UV)(place - RExC_emit_start),
11084 (UV)(RExC_parse - RExC_start),
11085 (UV)RExC_offsets[0]));
11086 Set_Node_Offset(place, RExC_parse);
11087 Set_Node_Length(place, 1);
11090 src = NEXTOPER(place);
11091 FILL_ADVANCE_NODE(place, op);
11092 Zero(src, offset, regnode);
11096 - regtail - set the next-pointer at the end of a node chain of p to val.
11097 - SEE ALSO: regtail_study
11099 /* TODO: All three parms should be const */
11101 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
11104 register regnode *scan;
11105 GET_RE_DEBUG_FLAGS_DECL;
11107 PERL_ARGS_ASSERT_REGTAIL;
11109 PERL_UNUSED_ARG(depth);
11115 /* Find last node. */
11118 regnode * const temp = regnext(scan);
11120 SV * const mysv=sv_newmortal();
11121 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
11122 regprop(RExC_rx, mysv, scan);
11123 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
11124 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
11125 (temp == NULL ? "->" : ""),
11126 (temp == NULL ? PL_reg_name[OP(val)] : "")
11134 if (reg_off_by_arg[OP(scan)]) {
11135 ARG_SET(scan, val - scan);
11138 NEXT_OFF(scan) = val - scan;
11144 - regtail_study - set the next-pointer at the end of a node chain of p to val.
11145 - Look for optimizable sequences at the same time.
11146 - currently only looks for EXACT chains.
11148 This is experimental code. The idea is to use this routine to perform
11149 in place optimizations on branches and groups as they are constructed,
11150 with the long term intention of removing optimization from study_chunk so
11151 that it is purely analytical.
11153 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
11154 to control which is which.
11157 /* TODO: All four parms should be const */
11160 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
11163 register regnode *scan;
11165 #ifdef EXPERIMENTAL_INPLACESCAN
11168 GET_RE_DEBUG_FLAGS_DECL;
11170 PERL_ARGS_ASSERT_REGTAIL_STUDY;
11176 /* Find last node. */
11180 regnode * const temp = regnext(scan);
11181 #ifdef EXPERIMENTAL_INPLACESCAN
11182 if (PL_regkind[OP(scan)] == EXACT)
11183 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
11187 switch (OP(scan)) {
11193 if( exact == PSEUDO )
11195 else if ( exact != OP(scan) )
11204 SV * const mysv=sv_newmortal();
11205 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
11206 regprop(RExC_rx, mysv, scan);
11207 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
11208 SvPV_nolen_const(mysv),
11209 REG_NODE_NUM(scan),
11210 PL_reg_name[exact]);
11217 SV * const mysv_val=sv_newmortal();
11218 DEBUG_PARSE_MSG("");
11219 regprop(RExC_rx, mysv_val, val);
11220 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
11221 SvPV_nolen_const(mysv_val),
11222 (IV)REG_NODE_NUM(val),
11226 if (reg_off_by_arg[OP(scan)]) {
11227 ARG_SET(scan, val - scan);
11230 NEXT_OFF(scan) = val - scan;
11238 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
11242 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
11248 for (bit=0; bit<32; bit++) {
11249 if (flags & (1<<bit)) {
11250 if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */
11253 if (!set++ && lead)
11254 PerlIO_printf(Perl_debug_log, "%s",lead);
11255 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
11258 if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) {
11259 if (!set++ && lead) {
11260 PerlIO_printf(Perl_debug_log, "%s",lead);
11263 case REGEX_UNICODE_CHARSET:
11264 PerlIO_printf(Perl_debug_log, "UNICODE");
11266 case REGEX_LOCALE_CHARSET:
11267 PerlIO_printf(Perl_debug_log, "LOCALE");
11269 case REGEX_ASCII_RESTRICTED_CHARSET:
11270 PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED");
11272 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
11273 PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED");
11276 PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET");
11282 PerlIO_printf(Perl_debug_log, "\n");
11284 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
11290 Perl_regdump(pTHX_ const regexp *r)
11294 SV * const sv = sv_newmortal();
11295 SV *dsv= sv_newmortal();
11296 RXi_GET_DECL(r,ri);
11297 GET_RE_DEBUG_FLAGS_DECL;
11299 PERL_ARGS_ASSERT_REGDUMP;
11301 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
11303 /* Header fields of interest. */
11304 if (r->anchored_substr) {
11305 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
11306 RE_SV_DUMPLEN(r->anchored_substr), 30);
11307 PerlIO_printf(Perl_debug_log,
11308 "anchored %s%s at %"IVdf" ",
11309 s, RE_SV_TAIL(r->anchored_substr),
11310 (IV)r->anchored_offset);
11311 } else if (r->anchored_utf8) {
11312 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
11313 RE_SV_DUMPLEN(r->anchored_utf8), 30);
11314 PerlIO_printf(Perl_debug_log,
11315 "anchored utf8 %s%s at %"IVdf" ",
11316 s, RE_SV_TAIL(r->anchored_utf8),
11317 (IV)r->anchored_offset);
11319 if (r->float_substr) {
11320 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
11321 RE_SV_DUMPLEN(r->float_substr), 30);
11322 PerlIO_printf(Perl_debug_log,
11323 "floating %s%s at %"IVdf"..%"UVuf" ",
11324 s, RE_SV_TAIL(r->float_substr),
11325 (IV)r->float_min_offset, (UV)r->float_max_offset);
11326 } else if (r->float_utf8) {
11327 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
11328 RE_SV_DUMPLEN(r->float_utf8), 30);
11329 PerlIO_printf(Perl_debug_log,
11330 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
11331 s, RE_SV_TAIL(r->float_utf8),
11332 (IV)r->float_min_offset, (UV)r->float_max_offset);
11334 if (r->check_substr || r->check_utf8)
11335 PerlIO_printf(Perl_debug_log,
11337 (r->check_substr == r->float_substr
11338 && r->check_utf8 == r->float_utf8
11339 ? "(checking floating" : "(checking anchored"));
11340 if (r->extflags & RXf_NOSCAN)
11341 PerlIO_printf(Perl_debug_log, " noscan");
11342 if (r->extflags & RXf_CHECK_ALL)
11343 PerlIO_printf(Perl_debug_log, " isall");
11344 if (r->check_substr || r->check_utf8)
11345 PerlIO_printf(Perl_debug_log, ") ");
11347 if (ri->regstclass) {
11348 regprop(r, sv, ri->regstclass);
11349 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
11351 if (r->extflags & RXf_ANCH) {
11352 PerlIO_printf(Perl_debug_log, "anchored");
11353 if (r->extflags & RXf_ANCH_BOL)
11354 PerlIO_printf(Perl_debug_log, "(BOL)");
11355 if (r->extflags & RXf_ANCH_MBOL)
11356 PerlIO_printf(Perl_debug_log, "(MBOL)");
11357 if (r->extflags & RXf_ANCH_SBOL)
11358 PerlIO_printf(Perl_debug_log, "(SBOL)");
11359 if (r->extflags & RXf_ANCH_GPOS)
11360 PerlIO_printf(Perl_debug_log, "(GPOS)");
11361 PerlIO_putc(Perl_debug_log, ' ');
11363 if (r->extflags & RXf_GPOS_SEEN)
11364 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
11365 if (r->intflags & PREGf_SKIP)
11366 PerlIO_printf(Perl_debug_log, "plus ");
11367 if (r->intflags & PREGf_IMPLICIT)
11368 PerlIO_printf(Perl_debug_log, "implicit ");
11369 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
11370 if (r->extflags & RXf_EVAL_SEEN)
11371 PerlIO_printf(Perl_debug_log, "with eval ");
11372 PerlIO_printf(Perl_debug_log, "\n");
11373 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
11375 PERL_ARGS_ASSERT_REGDUMP;
11376 PERL_UNUSED_CONTEXT;
11377 PERL_UNUSED_ARG(r);
11378 #endif /* DEBUGGING */
11382 - regprop - printable representation of opcode
11384 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
11387 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
11388 if (flags & ANYOF_INVERT) \
11389 /*make sure the invert info is in each */ \
11390 sv_catpvs(sv, "^"); \
11396 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
11401 RXi_GET_DECL(prog,progi);
11402 GET_RE_DEBUG_FLAGS_DECL;
11404 PERL_ARGS_ASSERT_REGPROP;
11408 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
11409 /* It would be nice to FAIL() here, but this may be called from
11410 regexec.c, and it would be hard to supply pRExC_state. */
11411 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
11412 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
11414 k = PL_regkind[OP(o)];
11417 sv_catpvs(sv, " ");
11418 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
11419 * is a crude hack but it may be the best for now since
11420 * we have no flag "this EXACTish node was UTF-8"
11422 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
11423 PERL_PV_ESCAPE_UNI_DETECT |
11424 PERL_PV_ESCAPE_NONASCII |
11425 PERL_PV_PRETTY_ELLIPSES |
11426 PERL_PV_PRETTY_LTGT |
11427 PERL_PV_PRETTY_NOCLEAR
11429 } else if (k == TRIE) {
11430 /* print the details of the trie in dumpuntil instead, as
11431 * progi->data isn't available here */
11432 const char op = OP(o);
11433 const U32 n = ARG(o);
11434 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
11435 (reg_ac_data *)progi->data->data[n] :
11437 const reg_trie_data * const trie
11438 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
11440 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
11441 DEBUG_TRIE_COMPILE_r(
11442 Perl_sv_catpvf(aTHX_ sv,
11443 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
11444 (UV)trie->startstate,
11445 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
11446 (UV)trie->wordcount,
11449 (UV)TRIE_CHARCOUNT(trie),
11450 (UV)trie->uniquecharcount
11453 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
11455 int rangestart = -1;
11456 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
11457 sv_catpvs(sv, "[");
11458 for (i = 0; i <= 256; i++) {
11459 if (i < 256 && BITMAP_TEST(bitmap,i)) {
11460 if (rangestart == -1)
11462 } else if (rangestart != -1) {
11463 if (i <= rangestart + 3)
11464 for (; rangestart < i; rangestart++)
11465 put_byte(sv, rangestart);
11467 put_byte(sv, rangestart);
11468 sv_catpvs(sv, "-");
11469 put_byte(sv, i - 1);
11474 sv_catpvs(sv, "]");
11477 } else if (k == CURLY) {
11478 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
11479 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
11480 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
11482 else if (k == WHILEM && o->flags) /* Ordinal/of */
11483 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
11484 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
11485 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
11486 if ( RXp_PAREN_NAMES(prog) ) {
11487 if ( k != REF || (OP(o) < NREF)) {
11488 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
11489 SV **name= av_fetch(list, ARG(o), 0 );
11491 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11494 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
11495 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
11496 I32 *nums=(I32*)SvPVX(sv_dat);
11497 SV **name= av_fetch(list, nums[0], 0 );
11500 for ( n=0; n<SvIVX(sv_dat); n++ ) {
11501 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
11502 (n ? "," : ""), (IV)nums[n]);
11504 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11508 } else if (k == GOSUB)
11509 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
11510 else if (k == VERB) {
11512 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
11513 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
11514 } else if (k == LOGICAL)
11515 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
11516 else if (k == FOLDCHAR)
11517 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
11518 else if (k == ANYOF) {
11519 int i, rangestart = -1;
11520 const U8 flags = ANYOF_FLAGS(o);
11523 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
11524 static const char * const anyofs[] = {
11557 if (flags & ANYOF_LOCALE)
11558 sv_catpvs(sv, "{loc}");
11559 if (flags & ANYOF_LOC_NONBITMAP_FOLD)
11560 sv_catpvs(sv, "{i}");
11561 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
11562 if (flags & ANYOF_INVERT)
11563 sv_catpvs(sv, "^");
11565 /* output what the standard cp 0-255 bitmap matches */
11566 for (i = 0; i <= 256; i++) {
11567 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
11568 if (rangestart == -1)
11570 } else if (rangestart != -1) {
11571 if (i <= rangestart + 3)
11572 for (; rangestart < i; rangestart++)
11573 put_byte(sv, rangestart);
11575 put_byte(sv, rangestart);
11576 sv_catpvs(sv, "-");
11577 put_byte(sv, i - 1);
11584 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11585 /* output any special charclass tests (used entirely under use locale) */
11586 if (ANYOF_CLASS_TEST_ANY_SET(o))
11587 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
11588 if (ANYOF_CLASS_TEST(o,i)) {
11589 sv_catpv(sv, anyofs[i]);
11593 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11595 if (flags & ANYOF_NON_UTF8_LATIN1_ALL) {
11596 sv_catpvs(sv, "{non-utf8-latin1-all}");
11599 /* output information about the unicode matching */
11600 if (flags & ANYOF_UNICODE_ALL)
11601 sv_catpvs(sv, "{unicode_all}");
11602 else if (ANYOF_NONBITMAP(o))
11603 sv_catpvs(sv, "{unicode}");
11604 if (flags & ANYOF_NONBITMAP_NON_UTF8)
11605 sv_catpvs(sv, "{outside bitmap}");
11607 if (ANYOF_NONBITMAP(o)) {
11609 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
11613 U8 s[UTF8_MAXBYTES_CASE+1];
11615 for (i = 0; i <= 256; i++) { /* just the first 256 */
11616 uvchr_to_utf8(s, i);
11618 if (i < 256 && swash_fetch(sw, s, TRUE)) {
11619 if (rangestart == -1)
11621 } else if (rangestart != -1) {
11622 if (i <= rangestart + 3)
11623 for (; rangestart < i; rangestart++) {
11624 const U8 * const e = uvchr_to_utf8(s,rangestart);
11626 for(p = s; p < e; p++)
11630 const U8 *e = uvchr_to_utf8(s,rangestart);
11632 for (p = s; p < e; p++)
11634 sv_catpvs(sv, "-");
11635 e = uvchr_to_utf8(s, i-1);
11636 for (p = s; p < e; p++)
11643 sv_catpvs(sv, "..."); /* et cetera */
11647 char *s = savesvpv(lv);
11648 char * const origs = s;
11650 while (*s && *s != '\n')
11654 const char * const t = ++s;
11672 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
11674 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
11675 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
11677 PERL_UNUSED_CONTEXT;
11678 PERL_UNUSED_ARG(sv);
11679 PERL_UNUSED_ARG(o);
11680 PERL_UNUSED_ARG(prog);
11681 #endif /* DEBUGGING */
11685 Perl_re_intuit_string(pTHX_ REGEXP * const r)
11686 { /* Assume that RE_INTUIT is set */
11688 struct regexp *const prog = (struct regexp *)SvANY(r);
11689 GET_RE_DEBUG_FLAGS_DECL;
11691 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
11692 PERL_UNUSED_CONTEXT;
11696 const char * const s = SvPV_nolen_const(prog->check_substr
11697 ? prog->check_substr : prog->check_utf8);
11699 if (!PL_colorset) reginitcolors();
11700 PerlIO_printf(Perl_debug_log,
11701 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
11703 prog->check_substr ? "" : "utf8 ",
11704 PL_colors[5],PL_colors[0],
11707 (strlen(s) > 60 ? "..." : ""));
11710 return prog->check_substr ? prog->check_substr : prog->check_utf8;
11716 handles refcounting and freeing the perl core regexp structure. When
11717 it is necessary to actually free the structure the first thing it
11718 does is call the 'free' method of the regexp_engine associated to
11719 the regexp, allowing the handling of the void *pprivate; member
11720 first. (This routine is not overridable by extensions, which is why
11721 the extensions free is called first.)
11723 See regdupe and regdupe_internal if you change anything here.
11725 #ifndef PERL_IN_XSUB_RE
11727 Perl_pregfree(pTHX_ REGEXP *r)
11733 Perl_pregfree2(pTHX_ REGEXP *rx)
11736 struct regexp *const r = (struct regexp *)SvANY(rx);
11737 GET_RE_DEBUG_FLAGS_DECL;
11739 PERL_ARGS_ASSERT_PREGFREE2;
11741 if (r->mother_re) {
11742 ReREFCNT_dec(r->mother_re);
11744 CALLREGFREE_PVT(rx); /* free the private data */
11745 SvREFCNT_dec(RXp_PAREN_NAMES(r));
11748 SvREFCNT_dec(r->anchored_substr);
11749 SvREFCNT_dec(r->anchored_utf8);
11750 SvREFCNT_dec(r->float_substr);
11751 SvREFCNT_dec(r->float_utf8);
11752 Safefree(r->substrs);
11754 RX_MATCH_COPY_FREE(rx);
11755 #ifdef PERL_OLD_COPY_ON_WRITE
11756 SvREFCNT_dec(r->saved_copy);
11763 This is a hacky workaround to the structural issue of match results
11764 being stored in the regexp structure which is in turn stored in
11765 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
11766 could be PL_curpm in multiple contexts, and could require multiple
11767 result sets being associated with the pattern simultaneously, such
11768 as when doing a recursive match with (??{$qr})
11770 The solution is to make a lightweight copy of the regexp structure
11771 when a qr// is returned from the code executed by (??{$qr}) this
11772 lightweight copy doesn't actually own any of its data except for
11773 the starp/end and the actual regexp structure itself.
11779 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
11781 struct regexp *ret;
11782 struct regexp *const r = (struct regexp *)SvANY(rx);
11783 register const I32 npar = r->nparens+1;
11785 PERL_ARGS_ASSERT_REG_TEMP_COPY;
11788 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
11789 ret = (struct regexp *)SvANY(ret_x);
11791 (void)ReREFCNT_inc(rx);
11792 /* We can take advantage of the existing "copied buffer" mechanism in SVs
11793 by pointing directly at the buffer, but flagging that the allocated
11794 space in the copy is zero. As we've just done a struct copy, it's now
11795 a case of zero-ing that, rather than copying the current length. */
11796 SvPV_set(ret_x, RX_WRAPPED(rx));
11797 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
11798 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
11799 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
11800 SvLEN_set(ret_x, 0);
11801 SvSTASH_set(ret_x, NULL);
11802 SvMAGIC_set(ret_x, NULL);
11803 Newx(ret->offs, npar, regexp_paren_pair);
11804 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11806 Newx(ret->substrs, 1, struct reg_substr_data);
11807 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11809 SvREFCNT_inc_void(ret->anchored_substr);
11810 SvREFCNT_inc_void(ret->anchored_utf8);
11811 SvREFCNT_inc_void(ret->float_substr);
11812 SvREFCNT_inc_void(ret->float_utf8);
11814 /* check_substr and check_utf8, if non-NULL, point to either their
11815 anchored or float namesakes, and don't hold a second reference. */
11817 RX_MATCH_COPIED_off(ret_x);
11818 #ifdef PERL_OLD_COPY_ON_WRITE
11819 ret->saved_copy = NULL;
11821 ret->mother_re = rx;
11827 /* regfree_internal()
11829 Free the private data in a regexp. This is overloadable by
11830 extensions. Perl takes care of the regexp structure in pregfree(),
11831 this covers the *pprivate pointer which technically perl doesn't
11832 know about, however of course we have to handle the
11833 regexp_internal structure when no extension is in use.
11835 Note this is called before freeing anything in the regexp
11840 Perl_regfree_internal(pTHX_ REGEXP * const rx)
11843 struct regexp *const r = (struct regexp *)SvANY(rx);
11844 RXi_GET_DECL(r,ri);
11845 GET_RE_DEBUG_FLAGS_DECL;
11847 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
11853 SV *dsv= sv_newmortal();
11854 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
11855 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
11856 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
11857 PL_colors[4],PL_colors[5],s);
11860 #ifdef RE_TRACK_PATTERN_OFFSETS
11862 Safefree(ri->u.offsets); /* 20010421 MJD */
11865 int n = ri->data->count;
11866 PAD* new_comppad = NULL;
11871 /* If you add a ->what type here, update the comment in regcomp.h */
11872 switch (ri->data->what[n]) {
11877 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
11880 Safefree(ri->data->data[n]);
11883 new_comppad = MUTABLE_AV(ri->data->data[n]);
11886 if (new_comppad == NULL)
11887 Perl_croak(aTHX_ "panic: pregfree comppad");
11888 PAD_SAVE_LOCAL(old_comppad,
11889 /* Watch out for global destruction's random ordering. */
11890 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
11893 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
11896 op_free((OP_4tree*)ri->data->data[n]);
11898 PAD_RESTORE_LOCAL(old_comppad);
11899 SvREFCNT_dec(MUTABLE_SV(new_comppad));
11900 new_comppad = NULL;
11905 { /* Aho Corasick add-on structure for a trie node.
11906 Used in stclass optimization only */
11908 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
11910 refcount = --aho->refcount;
11913 PerlMemShared_free(aho->states);
11914 PerlMemShared_free(aho->fail);
11915 /* do this last!!!! */
11916 PerlMemShared_free(ri->data->data[n]);
11917 PerlMemShared_free(ri->regstclass);
11923 /* trie structure. */
11925 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
11927 refcount = --trie->refcount;
11930 PerlMemShared_free(trie->charmap);
11931 PerlMemShared_free(trie->states);
11932 PerlMemShared_free(trie->trans);
11934 PerlMemShared_free(trie->bitmap);
11936 PerlMemShared_free(trie->jump);
11937 PerlMemShared_free(trie->wordinfo);
11938 /* do this last!!!! */
11939 PerlMemShared_free(ri->data->data[n]);
11944 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
11947 Safefree(ri->data->what);
11948 Safefree(ri->data);
11954 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11955 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11956 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11959 re_dup - duplicate a regexp.
11961 This routine is expected to clone a given regexp structure. It is only
11962 compiled under USE_ITHREADS.
11964 After all of the core data stored in struct regexp is duplicated
11965 the regexp_engine.dupe method is used to copy any private data
11966 stored in the *pprivate pointer. This allows extensions to handle
11967 any duplication it needs to do.
11969 See pregfree() and regfree_internal() if you change anything here.
11971 #if defined(USE_ITHREADS)
11972 #ifndef PERL_IN_XSUB_RE
11974 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
11978 const struct regexp *r = (const struct regexp *)SvANY(sstr);
11979 struct regexp *ret = (struct regexp *)SvANY(dstr);
11981 PERL_ARGS_ASSERT_RE_DUP_GUTS;
11983 npar = r->nparens+1;
11984 Newx(ret->offs, npar, regexp_paren_pair);
11985 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11987 /* no need to copy these */
11988 Newx(ret->swap, npar, regexp_paren_pair);
11991 if (ret->substrs) {
11992 /* Do it this way to avoid reading from *r after the StructCopy().
11993 That way, if any of the sv_dup_inc()s dislodge *r from the L1
11994 cache, it doesn't matter. */
11995 const bool anchored = r->check_substr
11996 ? r->check_substr == r->anchored_substr
11997 : r->check_utf8 == r->anchored_utf8;
11998 Newx(ret->substrs, 1, struct reg_substr_data);
11999 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
12001 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
12002 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
12003 ret->float_substr = sv_dup_inc(ret->float_substr, param);
12004 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
12006 /* check_substr and check_utf8, if non-NULL, point to either their
12007 anchored or float namesakes, and don't hold a second reference. */
12009 if (ret->check_substr) {
12011 assert(r->check_utf8 == r->anchored_utf8);
12012 ret->check_substr = ret->anchored_substr;
12013 ret->check_utf8 = ret->anchored_utf8;
12015 assert(r->check_substr == r->float_substr);
12016 assert(r->check_utf8 == r->float_utf8);
12017 ret->check_substr = ret->float_substr;
12018 ret->check_utf8 = ret->float_utf8;
12020 } else if (ret->check_utf8) {
12022 ret->check_utf8 = ret->anchored_utf8;
12024 ret->check_utf8 = ret->float_utf8;
12029 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
12032 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
12034 if (RX_MATCH_COPIED(dstr))
12035 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
12037 ret->subbeg = NULL;
12038 #ifdef PERL_OLD_COPY_ON_WRITE
12039 ret->saved_copy = NULL;
12042 if (ret->mother_re) {
12043 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
12044 /* Our storage points directly to our mother regexp, but that's
12045 1: a buffer in a different thread
12046 2: something we no longer hold a reference on
12047 so we need to copy it locally. */
12048 /* Note we need to sue SvCUR() on our mother_re, because it, in
12049 turn, may well be pointing to its own mother_re. */
12050 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
12051 SvCUR(ret->mother_re)+1));
12052 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
12054 ret->mother_re = NULL;
12058 #endif /* PERL_IN_XSUB_RE */
12063 This is the internal complement to regdupe() which is used to copy
12064 the structure pointed to by the *pprivate pointer in the regexp.
12065 This is the core version of the extension overridable cloning hook.
12066 The regexp structure being duplicated will be copied by perl prior
12067 to this and will be provided as the regexp *r argument, however
12068 with the /old/ structures pprivate pointer value. Thus this routine
12069 may override any copying normally done by perl.
12071 It returns a pointer to the new regexp_internal structure.
12075 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
12078 struct regexp *const r = (struct regexp *)SvANY(rx);
12079 regexp_internal *reti;
12081 RXi_GET_DECL(r,ri);
12083 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
12087 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
12088 Copy(ri->program, reti->program, len+1, regnode);
12091 reti->regstclass = NULL;
12094 struct reg_data *d;
12095 const int count = ri->data->count;
12098 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
12099 char, struct reg_data);
12100 Newx(d->what, count, U8);
12103 for (i = 0; i < count; i++) {
12104 d->what[i] = ri->data->what[i];
12105 switch (d->what[i]) {
12106 /* legal options are one of: sSfpontTua
12107 see also regcomp.h and pregfree() */
12108 case 'a': /* actually an AV, but the dup function is identical. */
12111 case 'p': /* actually an AV, but the dup function is identical. */
12112 case 'u': /* actually an HV, but the dup function is identical. */
12113 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
12116 /* This is cheating. */
12117 Newx(d->data[i], 1, struct regnode_charclass_class);
12118 StructCopy(ri->data->data[i], d->data[i],
12119 struct regnode_charclass_class);
12120 reti->regstclass = (regnode*)d->data[i];
12123 /* Compiled op trees are readonly and in shared memory,
12124 and can thus be shared without duplication. */
12126 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
12130 /* Trie stclasses are readonly and can thus be shared
12131 * without duplication. We free the stclass in pregfree
12132 * when the corresponding reg_ac_data struct is freed.
12134 reti->regstclass= ri->regstclass;
12138 ((reg_trie_data*)ri->data->data[i])->refcount++;
12142 d->data[i] = ri->data->data[i];
12145 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
12154 reti->name_list_idx = ri->name_list_idx;
12156 #ifdef RE_TRACK_PATTERN_OFFSETS
12157 if (ri->u.offsets) {
12158 Newx(reti->u.offsets, 2*len+1, U32);
12159 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
12162 SetProgLen(reti,len);
12165 return (void*)reti;
12168 #endif /* USE_ITHREADS */
12170 #ifndef PERL_IN_XSUB_RE
12173 - regnext - dig the "next" pointer out of a node
12176 Perl_regnext(pTHX_ register regnode *p)
12179 register I32 offset;
12184 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
12185 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
12188 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
12197 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
12200 STRLEN l1 = strlen(pat1);
12201 STRLEN l2 = strlen(pat2);
12204 const char *message;
12206 PERL_ARGS_ASSERT_RE_CROAK2;
12212 Copy(pat1, buf, l1 , char);
12213 Copy(pat2, buf + l1, l2 , char);
12214 buf[l1 + l2] = '\n';
12215 buf[l1 + l2 + 1] = '\0';
12217 /* ANSI variant takes additional second argument */
12218 va_start(args, pat2);
12222 msv = vmess(buf, &args);
12224 message = SvPV_const(msv,l1);
12227 Copy(message, buf, l1 , char);
12228 buf[l1-1] = '\0'; /* Overwrite \n */
12229 Perl_croak(aTHX_ "%s", buf);
12232 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
12234 #ifndef PERL_IN_XSUB_RE
12236 Perl_save_re_context(pTHX)
12240 struct re_save_state *state;
12242 SAVEVPTR(PL_curcop);
12243 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
12245 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
12246 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
12247 SSPUSHUV(SAVEt_RE_STATE);
12249 Copy(&PL_reg_state, state, 1, struct re_save_state);
12251 PL_reg_start_tmp = 0;
12252 PL_reg_start_tmpl = 0;
12253 PL_reg_oldsaved = NULL;
12254 PL_reg_oldsavedlen = 0;
12255 PL_reg_maxiter = 0;
12256 PL_reg_leftiter = 0;
12257 PL_reg_poscache = NULL;
12258 PL_reg_poscache_size = 0;
12259 #ifdef PERL_OLD_COPY_ON_WRITE
12263 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
12265 const REGEXP * const rx = PM_GETRE(PL_curpm);
12268 for (i = 1; i <= RX_NPARENS(rx); i++) {
12269 char digits[TYPE_CHARS(long)];
12270 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
12271 GV *const *const gvp
12272 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
12275 GV * const gv = *gvp;
12276 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
12286 clear_re(pTHX_ void *r)
12289 ReREFCNT_dec((REGEXP *)r);
12295 S_put_byte(pTHX_ SV *sv, int c)
12297 PERL_ARGS_ASSERT_PUT_BYTE;
12299 /* Our definition of isPRINT() ignores locales, so only bytes that are
12300 not part of UTF-8 are considered printable. I assume that the same
12301 holds for UTF-EBCDIC.
12302 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
12303 which Wikipedia says:
12305 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
12306 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
12307 identical, to the ASCII delete (DEL) or rubout control character.
12308 ) So the old condition can be simplified to !isPRINT(c) */
12311 Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c);
12314 Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c);
12318 const char string = c;
12319 if (c == '-' || c == ']' || c == '\\' || c == '^')
12320 sv_catpvs(sv, "\\");
12321 sv_catpvn(sv, &string, 1);
12326 #define CLEAR_OPTSTART \
12327 if (optstart) STMT_START { \
12328 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
12332 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
12334 STATIC const regnode *
12335 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
12336 const regnode *last, const regnode *plast,
12337 SV* sv, I32 indent, U32 depth)
12340 register U8 op = PSEUDO; /* Arbitrary non-END op. */
12341 register const regnode *next;
12342 const regnode *optstart= NULL;
12344 RXi_GET_DECL(r,ri);
12345 GET_RE_DEBUG_FLAGS_DECL;
12347 PERL_ARGS_ASSERT_DUMPUNTIL;
12349 #ifdef DEBUG_DUMPUNTIL
12350 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
12351 last ? last-start : 0,plast ? plast-start : 0);
12354 if (plast && plast < last)
12357 while (PL_regkind[op] != END && (!last || node < last)) {
12358 /* While that wasn't END last time... */
12361 if (op == CLOSE || op == WHILEM)
12363 next = regnext((regnode *)node);
12366 if (OP(node) == OPTIMIZED) {
12367 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
12374 regprop(r, sv, node);
12375 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
12376 (int)(2*indent + 1), "", SvPVX_const(sv));
12378 if (OP(node) != OPTIMIZED) {
12379 if (next == NULL) /* Next ptr. */
12380 PerlIO_printf(Perl_debug_log, " (0)");
12381 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
12382 PerlIO_printf(Perl_debug_log, " (FAIL)");
12384 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
12385 (void)PerlIO_putc(Perl_debug_log, '\n');
12389 if (PL_regkind[(U8)op] == BRANCHJ) {
12392 register const regnode *nnode = (OP(next) == LONGJMP
12393 ? regnext((regnode *)next)
12395 if (last && nnode > last)
12397 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
12400 else if (PL_regkind[(U8)op] == BRANCH) {
12402 DUMPUNTIL(NEXTOPER(node), next);
12404 else if ( PL_regkind[(U8)op] == TRIE ) {
12405 const regnode *this_trie = node;
12406 const char op = OP(node);
12407 const U32 n = ARG(node);
12408 const reg_ac_data * const ac = op>=AHOCORASICK ?
12409 (reg_ac_data *)ri->data->data[n] :
12411 const reg_trie_data * const trie =
12412 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
12414 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
12416 const regnode *nextbranch= NULL;
12419 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
12420 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
12422 PerlIO_printf(Perl_debug_log, "%*s%s ",
12423 (int)(2*(indent+3)), "",
12424 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
12425 PL_colors[0], PL_colors[1],
12426 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
12427 PERL_PV_PRETTY_ELLIPSES |
12428 PERL_PV_PRETTY_LTGT
12433 U16 dist= trie->jump[word_idx+1];
12434 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
12435 (UV)((dist ? this_trie + dist : next) - start));
12438 nextbranch= this_trie + trie->jump[0];
12439 DUMPUNTIL(this_trie + dist, nextbranch);
12441 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
12442 nextbranch= regnext((regnode *)nextbranch);
12444 PerlIO_printf(Perl_debug_log, "\n");
12447 if (last && next > last)
12452 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
12453 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
12454 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
12456 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
12458 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
12460 else if ( op == PLUS || op == STAR) {
12461 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
12463 else if (PL_regkind[(U8)op] == ANYOF) {
12464 /* arglen 1 + class block */
12465 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS)
12466 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
12467 node = NEXTOPER(node);
12469 else if (PL_regkind[(U8)op] == EXACT) {
12470 /* Literal string, where present. */
12471 node += NODE_SZ_STR(node) - 1;
12472 node = NEXTOPER(node);
12475 node = NEXTOPER(node);
12476 node += regarglen[(U8)op];
12478 if (op == CURLYX || op == OPEN)
12482 #ifdef DEBUG_DUMPUNTIL
12483 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
12488 #endif /* DEBUGGING */
12492 * c-indentation-style: bsd
12493 * c-basic-offset: 4
12494 * indent-tabs-mode: t
12497 * ex: set ts=8 sts=4 sw=4 noet: