5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
88 #include "dquote_static.c"
95 # if defined(BUGGY_MSC6)
96 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
97 # pragma optimize("a",off)
98 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
99 # pragma optimize("w",on )
100 # endif /* BUGGY_MSC6 */
104 #define STATIC static
107 typedef struct RExC_state_t {
108 U32 flags; /* are we folding, multilining? */
109 char *precomp; /* uncompiled string. */
110 REGEXP *rx_sv; /* The SV that is the regexp. */
111 regexp *rx; /* perl core regexp structure */
112 regexp_internal *rxi; /* internal data for regexp object pprivate field */
113 char *start; /* Start of input for compile */
114 char *end; /* End of input for compile */
115 char *parse; /* Input-scan pointer. */
116 I32 whilem_seen; /* number of WHILEM in this expr */
117 regnode *emit_start; /* Start of emitted-code area */
118 regnode *emit_bound; /* First regnode outside of the allocated space */
119 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
120 I32 naughty; /* How bad is this pattern? */
121 I32 sawback; /* Did we see \1, ...? */
123 I32 size; /* Code size. */
124 I32 npar; /* Capture buffer count, (OPEN). */
125 I32 cpar; /* Capture buffer count, (CLOSE). */
126 I32 nestroot; /* root parens we are in - used by accept */
130 regnode **open_parens; /* pointers to open parens */
131 regnode **close_parens; /* pointers to close parens */
132 regnode *opend; /* END node in program */
133 I32 utf8; /* whether the pattern is utf8 or not */
134 I32 orig_utf8; /* whether the pattern was originally in utf8 */
135 /* XXX use this for future optimisation of case
136 * where pattern must be upgraded to utf8. */
137 I32 uni_semantics; /* If a d charset modifier should use unicode
138 rules, even if the pattern is not in
140 HV *paren_names; /* Paren names */
142 regnode **recurse; /* Recurse regops */
143 I32 recurse_count; /* Number of recurse regops */
146 I32 override_recoding;
148 char *starttry; /* -Dr: where regtry was called. */
149 #define RExC_starttry (pRExC_state->starttry)
152 const char *lastparse;
154 AV *paren_name_list; /* idx -> name */
155 #define RExC_lastparse (pRExC_state->lastparse)
156 #define RExC_lastnum (pRExC_state->lastnum)
157 #define RExC_paren_name_list (pRExC_state->paren_name_list)
161 #define RExC_flags (pRExC_state->flags)
162 #define RExC_precomp (pRExC_state->precomp)
163 #define RExC_rx_sv (pRExC_state->rx_sv)
164 #define RExC_rx (pRExC_state->rx)
165 #define RExC_rxi (pRExC_state->rxi)
166 #define RExC_start (pRExC_state->start)
167 #define RExC_end (pRExC_state->end)
168 #define RExC_parse (pRExC_state->parse)
169 #define RExC_whilem_seen (pRExC_state->whilem_seen)
170 #ifdef RE_TRACK_PATTERN_OFFSETS
171 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
173 #define RExC_emit (pRExC_state->emit)
174 #define RExC_emit_start (pRExC_state->emit_start)
175 #define RExC_emit_bound (pRExC_state->emit_bound)
176 #define RExC_naughty (pRExC_state->naughty)
177 #define RExC_sawback (pRExC_state->sawback)
178 #define RExC_seen (pRExC_state->seen)
179 #define RExC_size (pRExC_state->size)
180 #define RExC_npar (pRExC_state->npar)
181 #define RExC_nestroot (pRExC_state->nestroot)
182 #define RExC_extralen (pRExC_state->extralen)
183 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
184 #define RExC_seen_evals (pRExC_state->seen_evals)
185 #define RExC_utf8 (pRExC_state->utf8)
186 #define RExC_uni_semantics (pRExC_state->uni_semantics)
187 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
188 #define RExC_open_parens (pRExC_state->open_parens)
189 #define RExC_close_parens (pRExC_state->close_parens)
190 #define RExC_opend (pRExC_state->opend)
191 #define RExC_paren_names (pRExC_state->paren_names)
192 #define RExC_recurse (pRExC_state->recurse)
193 #define RExC_recurse_count (pRExC_state->recurse_count)
194 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
195 #define RExC_contains_locale (pRExC_state->contains_locale)
196 #define RExC_override_recoding (pRExC_state->override_recoding)
199 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
200 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
201 ((*s) == '{' && regcurly(s)))
204 #undef SPSTART /* dratted cpp namespace... */
207 * Flags to be passed up and down.
209 #define WORST 0 /* Worst case. */
210 #define HASWIDTH 0x01 /* Known to match non-null strings. */
212 /* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single
213 * character, and if utf8, must be invariant. Note that this is not the same thing as REGNODE_SIMPLE */
215 #define SPSTART 0x04 /* Starts with * or +. */
216 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
217 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
219 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
221 /* whether trie related optimizations are enabled */
222 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
223 #define TRIE_STUDY_OPT
224 #define FULL_TRIE_STUDY
230 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
231 #define PBITVAL(paren) (1 << ((paren) & 7))
232 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
233 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
234 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
236 /* If not already in utf8, do a longjmp back to the beginning */
237 #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */
238 #define REQUIRE_UTF8 STMT_START { \
239 if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \
242 /* About scan_data_t.
244 During optimisation we recurse through the regexp program performing
245 various inplace (keyhole style) optimisations. In addition study_chunk
246 and scan_commit populate this data structure with information about
247 what strings MUST appear in the pattern. We look for the longest
248 string that must appear at a fixed location, and we look for the
249 longest string that may appear at a floating location. So for instance
254 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
255 strings (because they follow a .* construct). study_chunk will identify
256 both FOO and BAR as being the longest fixed and floating strings respectively.
258 The strings can be composites, for instance
262 will result in a composite fixed substring 'foo'.
264 For each string some basic information is maintained:
266 - offset or min_offset
267 This is the position the string must appear at, or not before.
268 It also implicitly (when combined with minlenp) tells us how many
269 characters must match before the string we are searching for.
270 Likewise when combined with minlenp and the length of the string it
271 tells us how many characters must appear after the string we have
275 Only used for floating strings. This is the rightmost point that
276 the string can appear at. If set to I32 max it indicates that the
277 string can occur infinitely far to the right.
280 A pointer to the minimum length of the pattern that the string
281 was found inside. This is important as in the case of positive
282 lookahead or positive lookbehind we can have multiple patterns
287 The minimum length of the pattern overall is 3, the minimum length
288 of the lookahead part is 3, but the minimum length of the part that
289 will actually match is 1. So 'FOO's minimum length is 3, but the
290 minimum length for the F is 1. This is important as the minimum length
291 is used to determine offsets in front of and behind the string being
292 looked for. Since strings can be composites this is the length of the
293 pattern at the time it was committed with a scan_commit. Note that
294 the length is calculated by study_chunk, so that the minimum lengths
295 are not known until the full pattern has been compiled, thus the
296 pointer to the value.
300 In the case of lookbehind the string being searched for can be
301 offset past the start point of the final matching string.
302 If this value was just blithely removed from the min_offset it would
303 invalidate some of the calculations for how many chars must match
304 before or after (as they are derived from min_offset and minlen and
305 the length of the string being searched for).
306 When the final pattern is compiled and the data is moved from the
307 scan_data_t structure into the regexp structure the information
308 about lookbehind is factored in, with the information that would
309 have been lost precalculated in the end_shift field for the
312 The fields pos_min and pos_delta are used to store the minimum offset
313 and the delta to the maximum offset at the current point in the pattern.
317 typedef struct scan_data_t {
318 /*I32 len_min; unused */
319 /*I32 len_delta; unused */
323 I32 last_end; /* min value, <0 unless valid. */
326 SV **longest; /* Either &l_fixed, or &l_float. */
327 SV *longest_fixed; /* longest fixed string found in pattern */
328 I32 offset_fixed; /* offset where it starts */
329 I32 *minlen_fixed; /* pointer to the minlen relevant to the string */
330 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
331 SV *longest_float; /* longest floating string found in pattern */
332 I32 offset_float_min; /* earliest point in string it can appear */
333 I32 offset_float_max; /* latest point in string it can appear */
334 I32 *minlen_float; /* pointer to the minlen relevant to the string */
335 I32 lookbehind_float; /* is the position of the string modified by LB */
339 struct regnode_charclass_class *start_class;
343 * Forward declarations for pregcomp()'s friends.
346 static const scan_data_t zero_scan_data =
347 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
349 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
350 #define SF_BEFORE_SEOL 0x0001
351 #define SF_BEFORE_MEOL 0x0002
352 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
353 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
356 # define SF_FIX_SHIFT_EOL (0+2)
357 # define SF_FL_SHIFT_EOL (0+4)
359 # define SF_FIX_SHIFT_EOL (+2)
360 # define SF_FL_SHIFT_EOL (+4)
363 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
364 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
366 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
367 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
368 #define SF_IS_INF 0x0040
369 #define SF_HAS_PAR 0x0080
370 #define SF_IN_PAR 0x0100
371 #define SF_HAS_EVAL 0x0200
372 #define SCF_DO_SUBSTR 0x0400
373 #define SCF_DO_STCLASS_AND 0x0800
374 #define SCF_DO_STCLASS_OR 0x1000
375 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
376 #define SCF_WHILEM_VISITED_POS 0x2000
378 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
379 #define SCF_SEEN_ACCEPT 0x8000
381 #define UTF cBOOL(RExC_utf8)
382 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
383 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
384 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_DEPENDS_CHARSET)
385 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) >= REGEX_UNICODE_CHARSET)
386 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_RESTRICTED_CHARSET)
387 #define MORE_ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
388 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) >= REGEX_ASCII_RESTRICTED_CHARSET)
390 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
392 #define OOB_UNICODE 12345678
393 #define OOB_NAMEDCLASS -1
395 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
396 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
399 /* length of regex to show in messages that don't mark a position within */
400 #define RegexLengthToShowInErrorMessages 127
403 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
404 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
405 * op/pragma/warn/regcomp.
407 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
408 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
410 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
413 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
414 * arg. Show regex, up to a maximum length. If it's too long, chop and add
417 #define _FAIL(code) STMT_START { \
418 const char *ellipses = ""; \
419 IV len = RExC_end - RExC_precomp; \
422 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
423 if (len > RegexLengthToShowInErrorMessages) { \
424 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
425 len = RegexLengthToShowInErrorMessages - 10; \
431 #define FAIL(msg) _FAIL( \
432 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
433 msg, (int)len, RExC_precomp, ellipses))
435 #define FAIL2(msg,arg) _FAIL( \
436 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
437 arg, (int)len, RExC_precomp, ellipses))
440 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
442 #define Simple_vFAIL(m) STMT_START { \
443 const IV offset = RExC_parse - RExC_precomp; \
444 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
445 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
449 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
451 #define vFAIL(m) STMT_START { \
453 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
458 * Like Simple_vFAIL(), but accepts two arguments.
460 #define Simple_vFAIL2(m,a1) STMT_START { \
461 const IV offset = RExC_parse - RExC_precomp; \
462 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
463 (int)offset, RExC_precomp, RExC_precomp + offset); \
467 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
469 #define vFAIL2(m,a1) STMT_START { \
471 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
472 Simple_vFAIL2(m, a1); \
477 * Like Simple_vFAIL(), but accepts three arguments.
479 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
480 const IV offset = RExC_parse - RExC_precomp; \
481 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
482 (int)offset, RExC_precomp, RExC_precomp + offset); \
486 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
488 #define vFAIL3(m,a1,a2) STMT_START { \
490 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
491 Simple_vFAIL3(m, a1, a2); \
495 * Like Simple_vFAIL(), but accepts four arguments.
497 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
498 const IV offset = RExC_parse - RExC_precomp; \
499 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
500 (int)offset, RExC_precomp, RExC_precomp + offset); \
503 #define ckWARNreg(loc,m) STMT_START { \
504 const IV offset = loc - RExC_precomp; \
505 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
506 (int)offset, RExC_precomp, RExC_precomp + offset); \
509 #define ckWARNregdep(loc,m) STMT_START { \
510 const IV offset = loc - RExC_precomp; \
511 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
513 (int)offset, RExC_precomp, RExC_precomp + offset); \
516 #define ckWARN2regdep(loc,m, a1) STMT_START { \
517 const IV offset = loc - RExC_precomp; \
518 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
520 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
523 #define ckWARN2reg(loc, m, a1) STMT_START { \
524 const IV offset = loc - RExC_precomp; \
525 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
526 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
529 #define vWARN3(loc, m, a1, a2) STMT_START { \
530 const IV offset = loc - RExC_precomp; \
531 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
532 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
535 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
536 const IV offset = loc - RExC_precomp; \
537 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
538 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
541 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
542 const IV offset = loc - RExC_precomp; \
543 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
544 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
547 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
548 const IV offset = loc - RExC_precomp; \
549 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
550 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
553 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
554 const IV offset = loc - RExC_precomp; \
555 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
556 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
560 /* Allow for side effects in s */
561 #define REGC(c,s) STMT_START { \
562 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
565 /* Macros for recording node offsets. 20001227 mjd@plover.com
566 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
567 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
568 * Element 0 holds the number n.
569 * Position is 1 indexed.
571 #ifndef RE_TRACK_PATTERN_OFFSETS
572 #define Set_Node_Offset_To_R(node,byte)
573 #define Set_Node_Offset(node,byte)
574 #define Set_Cur_Node_Offset
575 #define Set_Node_Length_To_R(node,len)
576 #define Set_Node_Length(node,len)
577 #define Set_Node_Cur_Length(node)
578 #define Node_Offset(n)
579 #define Node_Length(n)
580 #define Set_Node_Offset_Length(node,offset,len)
581 #define ProgLen(ri) ri->u.proglen
582 #define SetProgLen(ri,x) ri->u.proglen = x
584 #define ProgLen(ri) ri->u.offsets[0]
585 #define SetProgLen(ri,x) ri->u.offsets[0] = x
586 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
588 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
589 __LINE__, (int)(node), (int)(byte))); \
591 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
593 RExC_offsets[2*(node)-1] = (byte); \
598 #define Set_Node_Offset(node,byte) \
599 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
600 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
602 #define Set_Node_Length_To_R(node,len) STMT_START { \
604 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
605 __LINE__, (int)(node), (int)(len))); \
607 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
609 RExC_offsets[2*(node)] = (len); \
614 #define Set_Node_Length(node,len) \
615 Set_Node_Length_To_R((node)-RExC_emit_start, len)
616 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
617 #define Set_Node_Cur_Length(node) \
618 Set_Node_Length(node, RExC_parse - parse_start)
620 /* Get offsets and lengths */
621 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
622 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
624 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
625 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
626 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
630 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
631 #define EXPERIMENTAL_INPLACESCAN
632 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
634 #define DEBUG_STUDYDATA(str,data,depth) \
635 DEBUG_OPTIMISE_MORE_r(if(data){ \
636 PerlIO_printf(Perl_debug_log, \
637 "%*s" str "Pos:%"IVdf"/%"IVdf \
638 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
639 (int)(depth)*2, "", \
640 (IV)((data)->pos_min), \
641 (IV)((data)->pos_delta), \
642 (UV)((data)->flags), \
643 (IV)((data)->whilem_c), \
644 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
645 is_inf ? "INF " : "" \
647 if ((data)->last_found) \
648 PerlIO_printf(Perl_debug_log, \
649 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
650 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
651 SvPVX_const((data)->last_found), \
652 (IV)((data)->last_end), \
653 (IV)((data)->last_start_min), \
654 (IV)((data)->last_start_max), \
655 ((data)->longest && \
656 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
657 SvPVX_const((data)->longest_fixed), \
658 (IV)((data)->offset_fixed), \
659 ((data)->longest && \
660 (data)->longest==&((data)->longest_float)) ? "*" : "", \
661 SvPVX_const((data)->longest_float), \
662 (IV)((data)->offset_float_min), \
663 (IV)((data)->offset_float_max) \
665 PerlIO_printf(Perl_debug_log,"\n"); \
668 static void clear_re(pTHX_ void *r);
670 /* Mark that we cannot extend a found fixed substring at this point.
671 Update the longest found anchored substring and the longest found
672 floating substrings if needed. */
675 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
677 const STRLEN l = CHR_SVLEN(data->last_found);
678 const STRLEN old_l = CHR_SVLEN(*data->longest);
679 GET_RE_DEBUG_FLAGS_DECL;
681 PERL_ARGS_ASSERT_SCAN_COMMIT;
683 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
684 SvSetMagicSV(*data->longest, data->last_found);
685 if (*data->longest == data->longest_fixed) {
686 data->offset_fixed = l ? data->last_start_min : data->pos_min;
687 if (data->flags & SF_BEFORE_EOL)
689 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
691 data->flags &= ~SF_FIX_BEFORE_EOL;
692 data->minlen_fixed=minlenp;
693 data->lookbehind_fixed=0;
695 else { /* *data->longest == data->longest_float */
696 data->offset_float_min = l ? data->last_start_min : data->pos_min;
697 data->offset_float_max = (l
698 ? data->last_start_max
699 : data->pos_min + data->pos_delta);
700 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
701 data->offset_float_max = I32_MAX;
702 if (data->flags & SF_BEFORE_EOL)
704 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
706 data->flags &= ~SF_FL_BEFORE_EOL;
707 data->minlen_float=minlenp;
708 data->lookbehind_float=0;
711 SvCUR_set(data->last_found, 0);
713 SV * const sv = data->last_found;
714 if (SvUTF8(sv) && SvMAGICAL(sv)) {
715 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
721 data->flags &= ~SF_BEFORE_EOL;
722 DEBUG_STUDYDATA("commit: ",data,0);
725 /* Can match anything (initialization) */
727 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
729 PERL_ARGS_ASSERT_CL_ANYTHING;
731 ANYOF_BITMAP_SETALL(cl);
732 cl->flags = ANYOF_CLASS|ANYOF_EOS|ANYOF_UNICODE_ALL
733 |ANYOF_LOC_NONBITMAP_FOLD|ANYOF_NON_UTF8_LATIN1_ALL;
735 /* If any portion of the regex is to operate under locale rules,
736 * initialization includes it. The reason this isn't done for all regexes
737 * is that the optimizer was written under the assumption that locale was
738 * all-or-nothing. Given the complexity and lack of documentation in the
739 * optimizer, and that there are inadequate test cases for locale, so many
740 * parts of it may not work properly, it is safest to avoid locale unless
742 if (RExC_contains_locale) {
743 ANYOF_CLASS_SETALL(cl); /* /l uses class */
744 cl->flags |= ANYOF_LOCALE;
747 ANYOF_CLASS_ZERO(cl); /* Only /l uses class now */
751 /* Can match anything (initialization) */
753 S_cl_is_anything(const struct regnode_charclass_class *cl)
757 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
759 for (value = 0; value <= ANYOF_MAX; value += 2)
760 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
762 if (!(cl->flags & ANYOF_UNICODE_ALL))
764 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
769 /* Can match anything (initialization) */
771 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
773 PERL_ARGS_ASSERT_CL_INIT;
775 Zero(cl, 1, struct regnode_charclass_class);
777 cl_anything(pRExC_state, cl);
778 ARG_SET(cl, ANYOF_NONBITMAP_EMPTY);
781 /* These two functions currently do the exact same thing */
782 #define cl_init_zero S_cl_init
784 /* 'AND' a given class with another one. Can create false positives. 'cl'
785 * should not be inverted. 'and_with->flags & ANYOF_CLASS' should be 0 if
786 * 'and_with' is a regnode_charclass instead of a regnode_charclass_class. */
788 S_cl_and(struct regnode_charclass_class *cl,
789 const struct regnode_charclass_class *and_with)
791 PERL_ARGS_ASSERT_CL_AND;
793 assert(and_with->type == ANYOF);
795 /* I (khw) am not sure all these restrictions are necessary XXX */
796 if (!(ANYOF_CLASS_TEST_ANY_SET(and_with))
797 && !(ANYOF_CLASS_TEST_ANY_SET(cl))
798 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
799 && !(and_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
800 && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) {
803 if (and_with->flags & ANYOF_INVERT)
804 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
805 cl->bitmap[i] &= ~and_with->bitmap[i];
807 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
808 cl->bitmap[i] &= and_with->bitmap[i];
809 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
811 if (and_with->flags & ANYOF_INVERT) {
813 /* Here, the and'ed node is inverted. Get the AND of the flags that
814 * aren't affected by the inversion. Those that are affected are
815 * handled individually below */
816 U8 affected_flags = cl->flags & ~INVERSION_UNAFFECTED_FLAGS;
817 cl->flags &= (and_with->flags & INVERSION_UNAFFECTED_FLAGS);
818 cl->flags |= affected_flags;
820 /* We currently don't know how to deal with things that aren't in the
821 * bitmap, but we know that the intersection is no greater than what
822 * is already in cl, so let there be false positives that get sorted
823 * out after the synthetic start class succeeds, and the node is
824 * matched for real. */
826 /* The inversion of these two flags indicate that the resulting
827 * intersection doesn't have them */
828 if (and_with->flags & ANYOF_UNICODE_ALL) {
829 cl->flags &= ~ANYOF_UNICODE_ALL;
831 if (and_with->flags & ANYOF_NON_UTF8_LATIN1_ALL) {
832 cl->flags &= ~ANYOF_NON_UTF8_LATIN1_ALL;
835 else { /* and'd node is not inverted */
836 U8 outside_bitmap_but_not_utf8; /* Temp variable */
838 if (! ANYOF_NONBITMAP(and_with)) {
840 /* Here 'and_with' doesn't match anything outside the bitmap
841 * (except possibly ANYOF_UNICODE_ALL), which means the
842 * intersection can't either, except for ANYOF_UNICODE_ALL, in
843 * which case we don't know what the intersection is, but it's no
844 * greater than what cl already has, so can just leave it alone,
845 * with possible false positives */
846 if (! (and_with->flags & ANYOF_UNICODE_ALL)) {
847 ARG_SET(cl, ANYOF_NONBITMAP_EMPTY);
848 cl->flags &= ~ANYOF_NONBITMAP_NON_UTF8;
851 else if (! ANYOF_NONBITMAP(cl)) {
853 /* Here, 'and_with' does match something outside the bitmap, and cl
854 * doesn't have a list of things to match outside the bitmap. If
855 * cl can match all code points above 255, the intersection will
856 * be those above-255 code points that 'and_with' matches. If cl
857 * can't match all Unicode code points, it means that it can't
858 * match anything outside the bitmap (since the 'if' that got us
859 * into this block tested for that), so we leave the bitmap empty.
861 if (cl->flags & ANYOF_UNICODE_ALL) {
862 ARG_SET(cl, ARG(and_with));
864 /* and_with's ARG may match things that don't require UTF8.
865 * And now cl's will too, in spite of this being an 'and'. See
866 * the comments below about the kludge */
867 cl->flags |= and_with->flags & ANYOF_NONBITMAP_NON_UTF8;
871 /* Here, both 'and_with' and cl match something outside the
872 * bitmap. Currently we do not do the intersection, so just match
873 * whatever cl had at the beginning. */
877 /* Take the intersection of the two sets of flags. However, the
878 * ANYOF_NONBITMAP_NON_UTF8 flag is treated as an 'or'. This is a
879 * kludge around the fact that this flag is not treated like the others
880 * which are initialized in cl_anything(). The way the optimizer works
881 * is that the synthetic start class (SSC) is initialized to match
882 * anything, and then the first time a real node is encountered, its
883 * values are AND'd with the SSC's with the result being the values of
884 * the real node. However, there are paths through the optimizer where
885 * the AND never gets called, so those initialized bits are set
886 * inappropriately, which is not usually a big deal, as they just cause
887 * false positives in the SSC, which will just mean a probably
888 * imperceptible slow down in execution. However this bit has a
889 * higher false positive consequence in that it can cause utf8.pm,
890 * utf8_heavy.pl ... to be loaded when not necessary, which is a much
891 * bigger slowdown and also causes significant extra memory to be used.
892 * In order to prevent this, the code now takes a different tack. The
893 * bit isn't set unless some part of the regular expression needs it,
894 * but once set it won't get cleared. This means that these extra
895 * modules won't get loaded unless there was some path through the
896 * pattern that would have required them anyway, and so any false
897 * positives that occur by not ANDing them out when they could be
898 * aren't as severe as they would be if we treated this bit like all
900 outside_bitmap_but_not_utf8 = (cl->flags | and_with->flags)
901 & ANYOF_NONBITMAP_NON_UTF8;
902 cl->flags &= and_with->flags;
903 cl->flags |= outside_bitmap_but_not_utf8;
907 /* 'OR' a given class with another one. Can create false positives. 'cl'
908 * should not be inverted. 'or_with->flags & ANYOF_CLASS' should be 0 if
909 * 'or_with' is a regnode_charclass instead of a regnode_charclass_class. */
911 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
913 PERL_ARGS_ASSERT_CL_OR;
915 if (or_with->flags & ANYOF_INVERT) {
917 /* Here, the or'd node is to be inverted. This means we take the
918 * complement of everything not in the bitmap, but currently we don't
919 * know what that is, so give up and match anything */
920 if (ANYOF_NONBITMAP(or_with)) {
921 cl_anything(pRExC_state, cl);
924 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
925 * <= (B1 | !B2) | (CL1 | !CL2)
926 * which is wasteful if CL2 is small, but we ignore CL2:
927 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
928 * XXXX Can we handle case-fold? Unclear:
929 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
930 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
932 else if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
933 && !(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
934 && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD) ) {
937 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
938 cl->bitmap[i] |= ~or_with->bitmap[i];
939 } /* XXXX: logic is complicated otherwise */
941 cl_anything(pRExC_state, cl);
944 /* And, we can just take the union of the flags that aren't affected
945 * by the inversion */
946 cl->flags |= or_with->flags & INVERSION_UNAFFECTED_FLAGS;
948 /* For the remaining flags:
949 ANYOF_UNICODE_ALL and inverted means to not match anything above
950 255, which means that the union with cl should just be
951 what cl has in it, so can ignore this flag
952 ANYOF_NON_UTF8_LATIN1_ALL and inverted means if not utf8 and ord
953 is 127-255 to match them, but then invert that, so the
954 union with cl should just be what cl has in it, so can
957 } else { /* 'or_with' is not inverted */
958 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
959 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
960 && (!(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
961 || (cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) ) {
964 /* OR char bitmap and class bitmap separately */
965 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
966 cl->bitmap[i] |= or_with->bitmap[i];
967 if (ANYOF_CLASS_TEST_ANY_SET(or_with)) {
968 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
969 cl->classflags[i] |= or_with->classflags[i];
970 cl->flags |= ANYOF_CLASS;
973 else { /* XXXX: logic is complicated, leave it along for a moment. */
974 cl_anything(pRExC_state, cl);
977 if (ANYOF_NONBITMAP(or_with)) {
979 /* Use the added node's outside-the-bit-map match if there isn't a
980 * conflict. If there is a conflict (both nodes match something
981 * outside the bitmap, but what they match outside is not the same
982 * pointer, and hence not easily compared until XXX we extend
983 * inversion lists this far), give up and allow the start class to
984 * match everything outside the bitmap. If that stuff is all above
985 * 255, can just set UNICODE_ALL, otherwise caould be anything. */
986 if (! ANYOF_NONBITMAP(cl)) {
987 ARG_SET(cl, ARG(or_with));
989 else if (ARG(cl) != ARG(or_with)) {
991 if ((or_with->flags & ANYOF_NONBITMAP_NON_UTF8)) {
992 cl_anything(pRExC_state, cl);
995 cl->flags |= ANYOF_UNICODE_ALL;
1000 /* Take the union */
1001 cl->flags |= or_with->flags;
1005 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
1006 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
1007 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
1008 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
1013 dump_trie(trie,widecharmap,revcharmap)
1014 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
1015 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
1017 These routines dump out a trie in a somewhat readable format.
1018 The _interim_ variants are used for debugging the interim
1019 tables that are used to generate the final compressed
1020 representation which is what dump_trie expects.
1022 Part of the reason for their existence is to provide a form
1023 of documentation as to how the different representations function.
1028 Dumps the final compressed table form of the trie to Perl_debug_log.
1029 Used for debugging make_trie().
1033 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
1034 AV *revcharmap, U32 depth)
1037 SV *sv=sv_newmortal();
1038 int colwidth= widecharmap ? 6 : 4;
1040 GET_RE_DEBUG_FLAGS_DECL;
1042 PERL_ARGS_ASSERT_DUMP_TRIE;
1044 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
1045 (int)depth * 2 + 2,"",
1046 "Match","Base","Ofs" );
1048 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
1049 SV ** const tmp = av_fetch( revcharmap, state, 0);
1051 PerlIO_printf( Perl_debug_log, "%*s",
1053 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1054 PL_colors[0], PL_colors[1],
1055 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1056 PERL_PV_ESCAPE_FIRSTCHAR
1061 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
1062 (int)depth * 2 + 2,"");
1064 for( state = 0 ; state < trie->uniquecharcount ; state++ )
1065 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
1066 PerlIO_printf( Perl_debug_log, "\n");
1068 for( state = 1 ; state < trie->statecount ; state++ ) {
1069 const U32 base = trie->states[ state ].trans.base;
1071 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
1073 if ( trie->states[ state ].wordnum ) {
1074 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
1076 PerlIO_printf( Perl_debug_log, "%6s", "" );
1079 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
1084 while( ( base + ofs < trie->uniquecharcount ) ||
1085 ( base + ofs - trie->uniquecharcount < trie->lasttrans
1086 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
1089 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
1091 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1092 if ( ( base + ofs >= trie->uniquecharcount ) &&
1093 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1094 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1096 PerlIO_printf( Perl_debug_log, "%*"UVXf,
1098 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
1100 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
1104 PerlIO_printf( Perl_debug_log, "]");
1107 PerlIO_printf( Perl_debug_log, "\n" );
1109 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
1110 for (word=1; word <= trie->wordcount; word++) {
1111 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
1112 (int)word, (int)(trie->wordinfo[word].prev),
1113 (int)(trie->wordinfo[word].len));
1115 PerlIO_printf(Perl_debug_log, "\n" );
1118 Dumps a fully constructed but uncompressed trie in list form.
1119 List tries normally only are used for construction when the number of
1120 possible chars (trie->uniquecharcount) is very high.
1121 Used for debugging make_trie().
1124 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
1125 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1129 SV *sv=sv_newmortal();
1130 int colwidth= widecharmap ? 6 : 4;
1131 GET_RE_DEBUG_FLAGS_DECL;
1133 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1135 /* print out the table precompression. */
1136 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1137 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1138 "------:-----+-----------------\n" );
1140 for( state=1 ; state < next_alloc ; state ++ ) {
1143 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1144 (int)depth * 2 + 2,"", (UV)state );
1145 if ( ! trie->states[ state ].wordnum ) {
1146 PerlIO_printf( Perl_debug_log, "%5s| ","");
1148 PerlIO_printf( Perl_debug_log, "W%4x| ",
1149 trie->states[ state ].wordnum
1152 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1153 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
1155 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1157 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1158 PL_colors[0], PL_colors[1],
1159 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1160 PERL_PV_ESCAPE_FIRSTCHAR
1162 TRIE_LIST_ITEM(state,charid).forid,
1163 (UV)TRIE_LIST_ITEM(state,charid).newstate
1166 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1167 (int)((depth * 2) + 14), "");
1170 PerlIO_printf( Perl_debug_log, "\n");
1175 Dumps a fully constructed but uncompressed trie in table form.
1176 This is the normal DFA style state transition table, with a few
1177 twists to facilitate compression later.
1178 Used for debugging make_trie().
1181 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1182 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1187 SV *sv=sv_newmortal();
1188 int colwidth= widecharmap ? 6 : 4;
1189 GET_RE_DEBUG_FLAGS_DECL;
1191 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1194 print out the table precompression so that we can do a visual check
1195 that they are identical.
1198 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1200 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1201 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1203 PerlIO_printf( Perl_debug_log, "%*s",
1205 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1206 PL_colors[0], PL_colors[1],
1207 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1208 PERL_PV_ESCAPE_FIRSTCHAR
1214 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1216 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1217 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1220 PerlIO_printf( Perl_debug_log, "\n" );
1222 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1224 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1225 (int)depth * 2 + 2,"",
1226 (UV)TRIE_NODENUM( state ) );
1228 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1229 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1231 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1233 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1235 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1236 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1238 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1239 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1247 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1248 startbranch: the first branch in the whole branch sequence
1249 first : start branch of sequence of branch-exact nodes.
1250 May be the same as startbranch
1251 last : Thing following the last branch.
1252 May be the same as tail.
1253 tail : item following the branch sequence
1254 count : words in the sequence
1255 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1256 depth : indent depth
1258 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1260 A trie is an N'ary tree where the branches are determined by digital
1261 decomposition of the key. IE, at the root node you look up the 1st character and
1262 follow that branch repeat until you find the end of the branches. Nodes can be
1263 marked as "accepting" meaning they represent a complete word. Eg:
1267 would convert into the following structure. Numbers represent states, letters
1268 following numbers represent valid transitions on the letter from that state, if
1269 the number is in square brackets it represents an accepting state, otherwise it
1270 will be in parenthesis.
1272 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1276 (1) +-i->(6)-+-s->[7]
1278 +-s->(3)-+-h->(4)-+-e->[5]
1280 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1282 This shows that when matching against the string 'hers' we will begin at state 1
1283 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1284 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1285 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1286 single traverse. We store a mapping from accepting to state to which word was
1287 matched, and then when we have multiple possibilities we try to complete the
1288 rest of the regex in the order in which they occured in the alternation.
1290 The only prior NFA like behaviour that would be changed by the TRIE support is
1291 the silent ignoring of duplicate alternations which are of the form:
1293 / (DUPE|DUPE) X? (?{ ... }) Y /x
1295 Thus EVAL blocks following a trie may be called a different number of times with
1296 and without the optimisation. With the optimisations dupes will be silently
1297 ignored. This inconsistent behaviour of EVAL type nodes is well established as
1298 the following demonstrates:
1300 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1302 which prints out 'word' three times, but
1304 'words'=~/(word|word|word)(?{ print $1 })S/
1306 which doesnt print it out at all. This is due to other optimisations kicking in.
1308 Example of what happens on a structural level:
1310 The regexp /(ac|ad|ab)+/ will produce the following debug output:
1312 1: CURLYM[1] {1,32767}(18)
1323 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1324 and should turn into:
1326 1: CURLYM[1] {1,32767}(18)
1328 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1336 Cases where tail != last would be like /(?foo|bar)baz/:
1346 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1347 and would end up looking like:
1350 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1357 d = uvuni_to_utf8_flags(d, uv, 0);
1359 is the recommended Unicode-aware way of saying
1364 #define TRIE_STORE_REVCHAR \
1367 SV *zlopp = newSV(2); \
1368 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1369 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1370 SvCUR_set(zlopp, kapow - flrbbbbb); \
1373 av_push(revcharmap, zlopp); \
1375 char ooooff = (char)uvc; \
1376 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1380 #define TRIE_READ_CHAR STMT_START { \
1384 if ( foldlen > 0 ) { \
1385 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1390 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1391 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1392 foldlen -= UNISKIP( uvc ); \
1393 scan = foldbuf + UNISKIP( uvc ); \
1396 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1406 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1407 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1408 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1409 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1411 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1412 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1413 TRIE_LIST_CUR( state )++; \
1416 #define TRIE_LIST_NEW(state) STMT_START { \
1417 Newxz( trie->states[ state ].trans.list, \
1418 4, reg_trie_trans_le ); \
1419 TRIE_LIST_CUR( state ) = 1; \
1420 TRIE_LIST_LEN( state ) = 4; \
1423 #define TRIE_HANDLE_WORD(state) STMT_START { \
1424 U16 dupe= trie->states[ state ].wordnum; \
1425 regnode * const noper_next = regnext( noper ); \
1428 /* store the word for dumping */ \
1430 if (OP(noper) != NOTHING) \
1431 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1433 tmp = newSVpvn_utf8( "", 0, UTF ); \
1434 av_push( trie_words, tmp ); \
1438 trie->wordinfo[curword].prev = 0; \
1439 trie->wordinfo[curword].len = wordlen; \
1440 trie->wordinfo[curword].accept = state; \
1442 if ( noper_next < tail ) { \
1444 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1445 trie->jump[curword] = (U16)(noper_next - convert); \
1447 jumper = noper_next; \
1449 nextbranch= regnext(cur); \
1453 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1454 /* chain, so that when the bits of chain are later */\
1455 /* linked together, the dups appear in the chain */\
1456 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1457 trie->wordinfo[dupe].prev = curword; \
1459 /* we haven't inserted this word yet. */ \
1460 trie->states[ state ].wordnum = curword; \
1465 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1466 ( ( base + charid >= ucharcount \
1467 && base + charid < ubound \
1468 && state == trie->trans[ base - ucharcount + charid ].check \
1469 && trie->trans[ base - ucharcount + charid ].next ) \
1470 ? trie->trans[ base - ucharcount + charid ].next \
1471 : ( state==1 ? special : 0 ) \
1475 #define MADE_JUMP_TRIE 2
1476 #define MADE_EXACT_TRIE 4
1479 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1482 /* first pass, loop through and scan words */
1483 reg_trie_data *trie;
1484 HV *widecharmap = NULL;
1485 AV *revcharmap = newAV();
1487 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1492 regnode *jumper = NULL;
1493 regnode *nextbranch = NULL;
1494 regnode *convert = NULL;
1495 U32 *prev_states; /* temp array mapping each state to previous one */
1496 /* we just use folder as a flag in utf8 */
1497 const U8 * folder = NULL;
1500 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1501 AV *trie_words = NULL;
1502 /* along with revcharmap, this only used during construction but both are
1503 * useful during debugging so we store them in the struct when debugging.
1506 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1507 STRLEN trie_charcount=0;
1509 SV *re_trie_maxbuff;
1510 GET_RE_DEBUG_FLAGS_DECL;
1512 PERL_ARGS_ASSERT_MAKE_TRIE;
1514 PERL_UNUSED_ARG(depth);
1519 case EXACTFU: folder = PL_fold_latin1; break;
1520 case EXACTF: folder = PL_fold; break;
1521 case EXACTFL: folder = PL_fold_locale; break;
1524 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1526 trie->startstate = 1;
1527 trie->wordcount = word_count;
1528 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1529 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1530 if (!(UTF && folder))
1531 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1532 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1533 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1536 trie_words = newAV();
1539 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1540 if (!SvIOK(re_trie_maxbuff)) {
1541 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1544 PerlIO_printf( Perl_debug_log,
1545 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1546 (int)depth * 2 + 2, "",
1547 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1548 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1552 /* Find the node we are going to overwrite */
1553 if ( first == startbranch && OP( last ) != BRANCH ) {
1554 /* whole branch chain */
1557 /* branch sub-chain */
1558 convert = NEXTOPER( first );
1561 /* -- First loop and Setup --
1563 We first traverse the branches and scan each word to determine if it
1564 contains widechars, and how many unique chars there are, this is
1565 important as we have to build a table with at least as many columns as we
1568 We use an array of integers to represent the character codes 0..255
1569 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1570 native representation of the character value as the key and IV's for the
1573 *TODO* If we keep track of how many times each character is used we can
1574 remap the columns so that the table compression later on is more
1575 efficient in terms of memory by ensuring the most common value is in the
1576 middle and the least common are on the outside. IMO this would be better
1577 than a most to least common mapping as theres a decent chance the most
1578 common letter will share a node with the least common, meaning the node
1579 will not be compressible. With a middle is most common approach the worst
1580 case is when we have the least common nodes twice.
1584 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1585 regnode * const noper = NEXTOPER( cur );
1586 const U8 *uc = (U8*)STRING( noper );
1587 const U8 * const e = uc + STR_LEN( noper );
1589 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1590 const U8 *scan = (U8*)NULL;
1591 U32 wordlen = 0; /* required init */
1593 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1595 if (OP(noper) == NOTHING) {
1599 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1600 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1601 regardless of encoding */
1603 for ( ; uc < e ; uc += len ) {
1604 TRIE_CHARCOUNT(trie)++;
1608 if ( !trie->charmap[ uvc ] ) {
1609 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1611 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1615 /* store the codepoint in the bitmap, and its folded
1617 TRIE_BITMAP_SET(trie,uvc);
1619 /* store the folded codepoint */
1620 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1623 /* store first byte of utf8 representation of
1624 variant codepoints */
1625 if (! UNI_IS_INVARIANT(uvc)) {
1626 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
1629 set_bit = 0; /* We've done our bit :-) */
1634 widecharmap = newHV();
1636 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1639 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1641 if ( !SvTRUE( *svpp ) ) {
1642 sv_setiv( *svpp, ++trie->uniquecharcount );
1647 if( cur == first ) {
1650 } else if (chars < trie->minlen) {
1652 } else if (chars > trie->maxlen) {
1656 } /* end first pass */
1657 DEBUG_TRIE_COMPILE_r(
1658 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1659 (int)depth * 2 + 2,"",
1660 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1661 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1662 (int)trie->minlen, (int)trie->maxlen )
1666 We now know what we are dealing with in terms of unique chars and
1667 string sizes so we can calculate how much memory a naive
1668 representation using a flat table will take. If it's over a reasonable
1669 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1670 conservative but potentially much slower representation using an array
1673 At the end we convert both representations into the same compressed
1674 form that will be used in regexec.c for matching with. The latter
1675 is a form that cannot be used to construct with but has memory
1676 properties similar to the list form and access properties similar
1677 to the table form making it both suitable for fast searches and
1678 small enough that its feasable to store for the duration of a program.
1680 See the comment in the code where the compressed table is produced
1681 inplace from the flat tabe representation for an explanation of how
1682 the compression works.
1687 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1690 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1692 Second Pass -- Array Of Lists Representation
1694 Each state will be represented by a list of charid:state records
1695 (reg_trie_trans_le) the first such element holds the CUR and LEN
1696 points of the allocated array. (See defines above).
1698 We build the initial structure using the lists, and then convert
1699 it into the compressed table form which allows faster lookups
1700 (but cant be modified once converted).
1703 STRLEN transcount = 1;
1705 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1706 "%*sCompiling trie using list compiler\n",
1707 (int)depth * 2 + 2, ""));
1709 trie->states = (reg_trie_state *)
1710 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1711 sizeof(reg_trie_state) );
1715 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1717 regnode * const noper = NEXTOPER( cur );
1718 U8 *uc = (U8*)STRING( noper );
1719 const U8 * const e = uc + STR_LEN( noper );
1720 U32 state = 1; /* required init */
1721 U16 charid = 0; /* sanity init */
1722 U8 *scan = (U8*)NULL; /* sanity init */
1723 STRLEN foldlen = 0; /* required init */
1724 U32 wordlen = 0; /* required init */
1725 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1727 if (OP(noper) != NOTHING) {
1728 for ( ; uc < e ; uc += len ) {
1733 charid = trie->charmap[ uvc ];
1735 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1739 charid=(U16)SvIV( *svpp );
1742 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1749 if ( !trie->states[ state ].trans.list ) {
1750 TRIE_LIST_NEW( state );
1752 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1753 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1754 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1759 newstate = next_alloc++;
1760 prev_states[newstate] = state;
1761 TRIE_LIST_PUSH( state, charid, newstate );
1766 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1770 TRIE_HANDLE_WORD(state);
1772 } /* end second pass */
1774 /* next alloc is the NEXT state to be allocated */
1775 trie->statecount = next_alloc;
1776 trie->states = (reg_trie_state *)
1777 PerlMemShared_realloc( trie->states,
1779 * sizeof(reg_trie_state) );
1781 /* and now dump it out before we compress it */
1782 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1783 revcharmap, next_alloc,
1787 trie->trans = (reg_trie_trans *)
1788 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1795 for( state=1 ; state < next_alloc ; state ++ ) {
1799 DEBUG_TRIE_COMPILE_MORE_r(
1800 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1804 if (trie->states[state].trans.list) {
1805 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1809 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1810 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1811 if ( forid < minid ) {
1813 } else if ( forid > maxid ) {
1817 if ( transcount < tp + maxid - minid + 1) {
1819 trie->trans = (reg_trie_trans *)
1820 PerlMemShared_realloc( trie->trans,
1822 * sizeof(reg_trie_trans) );
1823 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1825 base = trie->uniquecharcount + tp - minid;
1826 if ( maxid == minid ) {
1828 for ( ; zp < tp ; zp++ ) {
1829 if ( ! trie->trans[ zp ].next ) {
1830 base = trie->uniquecharcount + zp - minid;
1831 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1832 trie->trans[ zp ].check = state;
1838 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1839 trie->trans[ tp ].check = state;
1844 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1845 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1846 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1847 trie->trans[ tid ].check = state;
1849 tp += ( maxid - minid + 1 );
1851 Safefree(trie->states[ state ].trans.list);
1854 DEBUG_TRIE_COMPILE_MORE_r(
1855 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1858 trie->states[ state ].trans.base=base;
1860 trie->lasttrans = tp + 1;
1864 Second Pass -- Flat Table Representation.
1866 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1867 We know that we will need Charcount+1 trans at most to store the data
1868 (one row per char at worst case) So we preallocate both structures
1869 assuming worst case.
1871 We then construct the trie using only the .next slots of the entry
1874 We use the .check field of the first entry of the node temporarily to
1875 make compression both faster and easier by keeping track of how many non
1876 zero fields are in the node.
1878 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1881 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1882 number representing the first entry of the node, and state as a
1883 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1884 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1885 are 2 entrys per node. eg:
1893 The table is internally in the right hand, idx form. However as we also
1894 have to deal with the states array which is indexed by nodenum we have to
1895 use TRIE_NODENUM() to convert.
1898 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1899 "%*sCompiling trie using table compiler\n",
1900 (int)depth * 2 + 2, ""));
1902 trie->trans = (reg_trie_trans *)
1903 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1904 * trie->uniquecharcount + 1,
1905 sizeof(reg_trie_trans) );
1906 trie->states = (reg_trie_state *)
1907 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1908 sizeof(reg_trie_state) );
1909 next_alloc = trie->uniquecharcount + 1;
1912 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1914 regnode * const noper = NEXTOPER( cur );
1915 const U8 *uc = (U8*)STRING( noper );
1916 const U8 * const e = uc + STR_LEN( noper );
1918 U32 state = 1; /* required init */
1920 U16 charid = 0; /* sanity init */
1921 U32 accept_state = 0; /* sanity init */
1922 U8 *scan = (U8*)NULL; /* sanity init */
1924 STRLEN foldlen = 0; /* required init */
1925 U32 wordlen = 0; /* required init */
1926 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1928 if ( OP(noper) != NOTHING ) {
1929 for ( ; uc < e ; uc += len ) {
1934 charid = trie->charmap[ uvc ];
1936 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1937 charid = svpp ? (U16)SvIV(*svpp) : 0;
1941 if ( !trie->trans[ state + charid ].next ) {
1942 trie->trans[ state + charid ].next = next_alloc;
1943 trie->trans[ state ].check++;
1944 prev_states[TRIE_NODENUM(next_alloc)]
1945 = TRIE_NODENUM(state);
1946 next_alloc += trie->uniquecharcount;
1948 state = trie->trans[ state + charid ].next;
1950 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1952 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1955 accept_state = TRIE_NODENUM( state );
1956 TRIE_HANDLE_WORD(accept_state);
1958 } /* end second pass */
1960 /* and now dump it out before we compress it */
1961 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1963 next_alloc, depth+1));
1967 * Inplace compress the table.*
1969 For sparse data sets the table constructed by the trie algorithm will
1970 be mostly 0/FAIL transitions or to put it another way mostly empty.
1971 (Note that leaf nodes will not contain any transitions.)
1973 This algorithm compresses the tables by eliminating most such
1974 transitions, at the cost of a modest bit of extra work during lookup:
1976 - Each states[] entry contains a .base field which indicates the
1977 index in the state[] array wheres its transition data is stored.
1979 - If .base is 0 there are no valid transitions from that node.
1981 - If .base is nonzero then charid is added to it to find an entry in
1984 -If trans[states[state].base+charid].check!=state then the
1985 transition is taken to be a 0/Fail transition. Thus if there are fail
1986 transitions at the front of the node then the .base offset will point
1987 somewhere inside the previous nodes data (or maybe even into a node
1988 even earlier), but the .check field determines if the transition is
1992 The following process inplace converts the table to the compressed
1993 table: We first do not compress the root node 1,and mark all its
1994 .check pointers as 1 and set its .base pointer as 1 as well. This
1995 allows us to do a DFA construction from the compressed table later,
1996 and ensures that any .base pointers we calculate later are greater
1999 - We set 'pos' to indicate the first entry of the second node.
2001 - We then iterate over the columns of the node, finding the first and
2002 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
2003 and set the .check pointers accordingly, and advance pos
2004 appropriately and repreat for the next node. Note that when we copy
2005 the next pointers we have to convert them from the original
2006 NODEIDX form to NODENUM form as the former is not valid post
2009 - If a node has no transitions used we mark its base as 0 and do not
2010 advance the pos pointer.
2012 - If a node only has one transition we use a second pointer into the
2013 structure to fill in allocated fail transitions from other states.
2014 This pointer is independent of the main pointer and scans forward
2015 looking for null transitions that are allocated to a state. When it
2016 finds one it writes the single transition into the "hole". If the
2017 pointer doesnt find one the single transition is appended as normal.
2019 - Once compressed we can Renew/realloc the structures to release the
2022 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
2023 specifically Fig 3.47 and the associated pseudocode.
2027 const U32 laststate = TRIE_NODENUM( next_alloc );
2030 trie->statecount = laststate;
2032 for ( state = 1 ; state < laststate ; state++ ) {
2034 const U32 stateidx = TRIE_NODEIDX( state );
2035 const U32 o_used = trie->trans[ stateidx ].check;
2036 U32 used = trie->trans[ stateidx ].check;
2037 trie->trans[ stateidx ].check = 0;
2039 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
2040 if ( flag || trie->trans[ stateidx + charid ].next ) {
2041 if ( trie->trans[ stateidx + charid ].next ) {
2043 for ( ; zp < pos ; zp++ ) {
2044 if ( ! trie->trans[ zp ].next ) {
2048 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
2049 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
2050 trie->trans[ zp ].check = state;
2051 if ( ++zp > pos ) pos = zp;
2058 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
2060 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
2061 trie->trans[ pos ].check = state;
2066 trie->lasttrans = pos + 1;
2067 trie->states = (reg_trie_state *)
2068 PerlMemShared_realloc( trie->states, laststate
2069 * sizeof(reg_trie_state) );
2070 DEBUG_TRIE_COMPILE_MORE_r(
2071 PerlIO_printf( Perl_debug_log,
2072 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
2073 (int)depth * 2 + 2,"",
2074 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
2077 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
2080 } /* end table compress */
2082 DEBUG_TRIE_COMPILE_MORE_r(
2083 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
2084 (int)depth * 2 + 2, "",
2085 (UV)trie->statecount,
2086 (UV)trie->lasttrans)
2088 /* resize the trans array to remove unused space */
2089 trie->trans = (reg_trie_trans *)
2090 PerlMemShared_realloc( trie->trans, trie->lasttrans
2091 * sizeof(reg_trie_trans) );
2093 { /* Modify the program and insert the new TRIE node */
2094 U8 nodetype =(U8)(flags & 0xFF);
2098 regnode *optimize = NULL;
2099 #ifdef RE_TRACK_PATTERN_OFFSETS
2102 U32 mjd_nodelen = 0;
2103 #endif /* RE_TRACK_PATTERN_OFFSETS */
2104 #endif /* DEBUGGING */
2106 This means we convert either the first branch or the first Exact,
2107 depending on whether the thing following (in 'last') is a branch
2108 or not and whther first is the startbranch (ie is it a sub part of
2109 the alternation or is it the whole thing.)
2110 Assuming its a sub part we convert the EXACT otherwise we convert
2111 the whole branch sequence, including the first.
2113 /* Find the node we are going to overwrite */
2114 if ( first != startbranch || OP( last ) == BRANCH ) {
2115 /* branch sub-chain */
2116 NEXT_OFF( first ) = (U16)(last - first);
2117 #ifdef RE_TRACK_PATTERN_OFFSETS
2119 mjd_offset= Node_Offset((convert));
2120 mjd_nodelen= Node_Length((convert));
2123 /* whole branch chain */
2125 #ifdef RE_TRACK_PATTERN_OFFSETS
2128 const regnode *nop = NEXTOPER( convert );
2129 mjd_offset= Node_Offset((nop));
2130 mjd_nodelen= Node_Length((nop));
2134 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2135 (int)depth * 2 + 2, "",
2136 (UV)mjd_offset, (UV)mjd_nodelen)
2139 /* But first we check to see if there is a common prefix we can
2140 split out as an EXACT and put in front of the TRIE node. */
2141 trie->startstate= 1;
2142 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2144 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2148 const U32 base = trie->states[ state ].trans.base;
2150 if ( trie->states[state].wordnum )
2153 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2154 if ( ( base + ofs >= trie->uniquecharcount ) &&
2155 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2156 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2158 if ( ++count > 1 ) {
2159 SV **tmp = av_fetch( revcharmap, ofs, 0);
2160 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2161 if ( state == 1 ) break;
2163 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2165 PerlIO_printf(Perl_debug_log,
2166 "%*sNew Start State=%"UVuf" Class: [",
2167 (int)depth * 2 + 2, "",
2170 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2171 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2173 TRIE_BITMAP_SET(trie,*ch);
2175 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2177 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2181 TRIE_BITMAP_SET(trie,*ch);
2183 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2184 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2190 SV **tmp = av_fetch( revcharmap, idx, 0);
2192 char *ch = SvPV( *tmp, len );
2194 SV *sv=sv_newmortal();
2195 PerlIO_printf( Perl_debug_log,
2196 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2197 (int)depth * 2 + 2, "",
2199 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2200 PL_colors[0], PL_colors[1],
2201 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2202 PERL_PV_ESCAPE_FIRSTCHAR
2207 OP( convert ) = nodetype;
2208 str=STRING(convert);
2211 STR_LEN(convert) += len;
2217 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2222 trie->prefixlen = (state-1);
2224 regnode *n = convert+NODE_SZ_STR(convert);
2225 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2226 trie->startstate = state;
2227 trie->minlen -= (state - 1);
2228 trie->maxlen -= (state - 1);
2230 /* At least the UNICOS C compiler choked on this
2231 * being argument to DEBUG_r(), so let's just have
2234 #ifdef PERL_EXT_RE_BUILD
2240 regnode *fix = convert;
2241 U32 word = trie->wordcount;
2243 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2244 while( ++fix < n ) {
2245 Set_Node_Offset_Length(fix, 0, 0);
2248 SV ** const tmp = av_fetch( trie_words, word, 0 );
2250 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2251 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2253 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2261 NEXT_OFF(convert) = (U16)(tail - convert);
2262 DEBUG_r(optimize= n);
2268 if ( trie->maxlen ) {
2269 NEXT_OFF( convert ) = (U16)(tail - convert);
2270 ARG_SET( convert, data_slot );
2271 /* Store the offset to the first unabsorbed branch in
2272 jump[0], which is otherwise unused by the jump logic.
2273 We use this when dumping a trie and during optimisation. */
2275 trie->jump[0] = (U16)(nextbranch - convert);
2277 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2278 * and there is a bitmap
2279 * and the first "jump target" node we found leaves enough room
2280 * then convert the TRIE node into a TRIEC node, with the bitmap
2281 * embedded inline in the opcode - this is hypothetically faster.
2283 if ( !trie->states[trie->startstate].wordnum
2285 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2287 OP( convert ) = TRIEC;
2288 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2289 PerlMemShared_free(trie->bitmap);
2292 OP( convert ) = TRIE;
2294 /* store the type in the flags */
2295 convert->flags = nodetype;
2299 + regarglen[ OP( convert ) ];
2301 /* XXX We really should free up the resource in trie now,
2302 as we won't use them - (which resources?) dmq */
2304 /* needed for dumping*/
2305 DEBUG_r(if (optimize) {
2306 regnode *opt = convert;
2308 while ( ++opt < optimize) {
2309 Set_Node_Offset_Length(opt,0,0);
2312 Try to clean up some of the debris left after the
2315 while( optimize < jumper ) {
2316 mjd_nodelen += Node_Length((optimize));
2317 OP( optimize ) = OPTIMIZED;
2318 Set_Node_Offset_Length(optimize,0,0);
2321 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2323 } /* end node insert */
2325 /* Finish populating the prev field of the wordinfo array. Walk back
2326 * from each accept state until we find another accept state, and if
2327 * so, point the first word's .prev field at the second word. If the
2328 * second already has a .prev field set, stop now. This will be the
2329 * case either if we've already processed that word's accept state,
2330 * or that state had multiple words, and the overspill words were
2331 * already linked up earlier.
2338 for (word=1; word <= trie->wordcount; word++) {
2340 if (trie->wordinfo[word].prev)
2342 state = trie->wordinfo[word].accept;
2344 state = prev_states[state];
2347 prev = trie->states[state].wordnum;
2351 trie->wordinfo[word].prev = prev;
2353 Safefree(prev_states);
2357 /* and now dump out the compressed format */
2358 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2360 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2362 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2363 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2365 SvREFCNT_dec(revcharmap);
2369 : trie->startstate>1
2375 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2377 /* The Trie is constructed and compressed now so we can build a fail array if it's needed
2379 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2380 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2383 We find the fail state for each state in the trie, this state is the longest proper
2384 suffix of the current state's 'word' that is also a proper prefix of another word in our
2385 trie. State 1 represents the word '' and is thus the default fail state. This allows
2386 the DFA not to have to restart after its tried and failed a word at a given point, it
2387 simply continues as though it had been matching the other word in the first place.
2389 'abcdgu'=~/abcdefg|cdgu/
2390 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2391 fail, which would bring us to the state representing 'd' in the second word where we would
2392 try 'g' and succeed, proceeding to match 'cdgu'.
2394 /* add a fail transition */
2395 const U32 trie_offset = ARG(source);
2396 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2398 const U32 ucharcount = trie->uniquecharcount;
2399 const U32 numstates = trie->statecount;
2400 const U32 ubound = trie->lasttrans + ucharcount;
2404 U32 base = trie->states[ 1 ].trans.base;
2407 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2408 GET_RE_DEBUG_FLAGS_DECL;
2410 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2412 PERL_UNUSED_ARG(depth);
2416 ARG_SET( stclass, data_slot );
2417 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2418 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2419 aho->trie=trie_offset;
2420 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2421 Copy( trie->states, aho->states, numstates, reg_trie_state );
2422 Newxz( q, numstates, U32);
2423 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2426 /* initialize fail[0..1] to be 1 so that we always have
2427 a valid final fail state */
2428 fail[ 0 ] = fail[ 1 ] = 1;
2430 for ( charid = 0; charid < ucharcount ; charid++ ) {
2431 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2433 q[ q_write ] = newstate;
2434 /* set to point at the root */
2435 fail[ q[ q_write++ ] ]=1;
2438 while ( q_read < q_write) {
2439 const U32 cur = q[ q_read++ % numstates ];
2440 base = trie->states[ cur ].trans.base;
2442 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2443 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2445 U32 fail_state = cur;
2448 fail_state = fail[ fail_state ];
2449 fail_base = aho->states[ fail_state ].trans.base;
2450 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2452 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2453 fail[ ch_state ] = fail_state;
2454 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2456 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2458 q[ q_write++ % numstates] = ch_state;
2462 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2463 when we fail in state 1, this allows us to use the
2464 charclass scan to find a valid start char. This is based on the principle
2465 that theres a good chance the string being searched contains lots of stuff
2466 that cant be a start char.
2468 fail[ 0 ] = fail[ 1 ] = 0;
2469 DEBUG_TRIE_COMPILE_r({
2470 PerlIO_printf(Perl_debug_log,
2471 "%*sStclass Failtable (%"UVuf" states): 0",
2472 (int)(depth * 2), "", (UV)numstates
2474 for( q_read=1; q_read<numstates; q_read++ ) {
2475 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2477 PerlIO_printf(Perl_debug_log, "\n");
2480 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2485 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2486 * These need to be revisited when a newer toolchain becomes available.
2488 #if defined(__sparc64__) && defined(__GNUC__)
2489 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2490 # undef SPARC64_GCC_WORKAROUND
2491 # define SPARC64_GCC_WORKAROUND 1
2495 #define DEBUG_PEEP(str,scan,depth) \
2496 DEBUG_OPTIMISE_r({if (scan){ \
2497 SV * const mysv=sv_newmortal(); \
2498 regnode *Next = regnext(scan); \
2499 regprop(RExC_rx, mysv, scan); \
2500 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2501 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2502 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2509 #define JOIN_EXACT(scan,min,flags) \
2510 if (PL_regkind[OP(scan)] == EXACT) \
2511 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2514 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2515 /* Merge several consecutive EXACTish nodes into one. */
2516 regnode *n = regnext(scan);
2518 regnode *next = scan + NODE_SZ_STR(scan);
2522 regnode *stop = scan;
2523 GET_RE_DEBUG_FLAGS_DECL;
2525 PERL_UNUSED_ARG(depth);
2528 PERL_ARGS_ASSERT_JOIN_EXACT;
2529 #ifndef EXPERIMENTAL_INPLACESCAN
2530 PERL_UNUSED_ARG(flags);
2531 PERL_UNUSED_ARG(val);
2533 DEBUG_PEEP("join",scan,depth);
2535 /* Skip NOTHING, merge EXACT*. */
2537 ( PL_regkind[OP(n)] == NOTHING ||
2538 (stringok && (OP(n) == OP(scan))))
2540 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2542 if (OP(n) == TAIL || n > next)
2544 if (PL_regkind[OP(n)] == NOTHING) {
2545 DEBUG_PEEP("skip:",n,depth);
2546 NEXT_OFF(scan) += NEXT_OFF(n);
2547 next = n + NODE_STEP_REGNODE;
2554 else if (stringok) {
2555 const unsigned int oldl = STR_LEN(scan);
2556 regnode * const nnext = regnext(n);
2558 DEBUG_PEEP("merg",n,depth);
2561 if (oldl + STR_LEN(n) > U8_MAX)
2563 NEXT_OFF(scan) += NEXT_OFF(n);
2564 STR_LEN(scan) += STR_LEN(n);
2565 next = n + NODE_SZ_STR(n);
2566 /* Now we can overwrite *n : */
2567 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2575 #ifdef EXPERIMENTAL_INPLACESCAN
2576 if (flags && !NEXT_OFF(n)) {
2577 DEBUG_PEEP("atch", val, depth);
2578 if (reg_off_by_arg[OP(n)]) {
2579 ARG_SET(n, val - n);
2582 NEXT_OFF(n) = val - n;
2588 #define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390
2589 #define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS
2590 #define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0
2591 #define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS
2594 && ( OP(scan) == EXACTF || OP(scan) == EXACTFU || OP(scan) == EXACTFA)
2595 && ( STR_LEN(scan) >= 6 ) )
2598 Two problematic code points in Unicode casefolding of EXACT nodes:
2600 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2601 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2607 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2608 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2610 This means that in case-insensitive matching (or "loose matching",
2611 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2612 length of the above casefolded versions) can match a target string
2613 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2614 This would rather mess up the minimum length computation.
2616 What we'll do is to look for the tail four bytes, and then peek
2617 at the preceding two bytes to see whether we need to decrease
2618 the minimum length by four (six minus two).
2620 Thanks to the design of UTF-8, there cannot be false matches:
2621 A sequence of valid UTF-8 bytes cannot be a subsequence of
2622 another valid sequence of UTF-8 bytes.
2625 char * const s0 = STRING(scan), *s, *t;
2626 char * const s1 = s0 + STR_LEN(scan) - 1;
2627 char * const s2 = s1 - 4;
2628 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2629 const char t0[] = "\xaf\x49\xaf\x42";
2631 const char t0[] = "\xcc\x88\xcc\x81";
2633 const char * const t1 = t0 + 3;
2636 s < s2 && (t = ninstr(s, s1, t0, t1));
2639 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2640 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2642 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2643 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2651 n = scan + NODE_SZ_STR(scan);
2653 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2660 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2664 /* REx optimizer. Converts nodes into quicker variants "in place".
2665 Finds fixed substrings. */
2667 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2668 to the position after last scanned or to NULL. */
2670 #define INIT_AND_WITHP \
2671 assert(!and_withp); \
2672 Newx(and_withp,1,struct regnode_charclass_class); \
2673 SAVEFREEPV(and_withp)
2675 /* this is a chain of data about sub patterns we are processing that
2676 need to be handled separately/specially in study_chunk. Its so
2677 we can simulate recursion without losing state. */
2679 typedef struct scan_frame {
2680 regnode *last; /* last node to process in this frame */
2681 regnode *next; /* next node to process when last is reached */
2682 struct scan_frame *prev; /*previous frame*/
2683 I32 stop; /* what stopparen do we use */
2687 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2689 #define CASE_SYNST_FNC(nAmE) \
2691 if (flags & SCF_DO_STCLASS_AND) { \
2692 for (value = 0; value < 256; value++) \
2693 if (!is_ ## nAmE ## _cp(value)) \
2694 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2697 for (value = 0; value < 256; value++) \
2698 if (is_ ## nAmE ## _cp(value)) \
2699 ANYOF_BITMAP_SET(data->start_class, value); \
2703 if (flags & SCF_DO_STCLASS_AND) { \
2704 for (value = 0; value < 256; value++) \
2705 if (is_ ## nAmE ## _cp(value)) \
2706 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2709 for (value = 0; value < 256; value++) \
2710 if (!is_ ## nAmE ## _cp(value)) \
2711 ANYOF_BITMAP_SET(data->start_class, value); \
2718 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2719 I32 *minlenp, I32 *deltap,
2724 struct regnode_charclass_class *and_withp,
2725 U32 flags, U32 depth)
2726 /* scanp: Start here (read-write). */
2727 /* deltap: Write maxlen-minlen here. */
2728 /* last: Stop before this one. */
2729 /* data: string data about the pattern */
2730 /* stopparen: treat close N as END */
2731 /* recursed: which subroutines have we recursed into */
2732 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2735 I32 min = 0, pars = 0, code;
2736 regnode *scan = *scanp, *next;
2738 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2739 int is_inf_internal = 0; /* The studied chunk is infinite */
2740 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2741 scan_data_t data_fake;
2742 SV *re_trie_maxbuff = NULL;
2743 regnode *first_non_open = scan;
2744 I32 stopmin = I32_MAX;
2745 scan_frame *frame = NULL;
2746 GET_RE_DEBUG_FLAGS_DECL;
2748 PERL_ARGS_ASSERT_STUDY_CHUNK;
2751 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2755 while (first_non_open && OP(first_non_open) == OPEN)
2756 first_non_open=regnext(first_non_open);
2761 while ( scan && OP(scan) != END && scan < last ){
2762 /* Peephole optimizer: */
2763 DEBUG_STUDYDATA("Peep:", data,depth);
2764 DEBUG_PEEP("Peep",scan,depth);
2765 JOIN_EXACT(scan,&min,0);
2767 /* Follow the next-chain of the current node and optimize
2768 away all the NOTHINGs from it. */
2769 if (OP(scan) != CURLYX) {
2770 const int max = (reg_off_by_arg[OP(scan)]
2772 /* I32 may be smaller than U16 on CRAYs! */
2773 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2774 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2778 /* Skip NOTHING and LONGJMP. */
2779 while ((n = regnext(n))
2780 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2781 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2782 && off + noff < max)
2784 if (reg_off_by_arg[OP(scan)])
2787 NEXT_OFF(scan) = off;
2792 /* The principal pseudo-switch. Cannot be a switch, since we
2793 look into several different things. */
2794 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2795 || OP(scan) == IFTHEN) {
2796 next = regnext(scan);
2798 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2800 if (OP(next) == code || code == IFTHEN) {
2801 /* NOTE - There is similar code to this block below for handling
2802 TRIE nodes on a re-study. If you change stuff here check there
2804 I32 max1 = 0, min1 = I32_MAX, num = 0;
2805 struct regnode_charclass_class accum;
2806 regnode * const startbranch=scan;
2808 if (flags & SCF_DO_SUBSTR)
2809 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2810 if (flags & SCF_DO_STCLASS)
2811 cl_init_zero(pRExC_state, &accum);
2813 while (OP(scan) == code) {
2814 I32 deltanext, minnext, f = 0, fake;
2815 struct regnode_charclass_class this_class;
2818 data_fake.flags = 0;
2820 data_fake.whilem_c = data->whilem_c;
2821 data_fake.last_closep = data->last_closep;
2824 data_fake.last_closep = &fake;
2826 data_fake.pos_delta = delta;
2827 next = regnext(scan);
2828 scan = NEXTOPER(scan);
2830 scan = NEXTOPER(scan);
2831 if (flags & SCF_DO_STCLASS) {
2832 cl_init(pRExC_state, &this_class);
2833 data_fake.start_class = &this_class;
2834 f = SCF_DO_STCLASS_AND;
2836 if (flags & SCF_WHILEM_VISITED_POS)
2837 f |= SCF_WHILEM_VISITED_POS;
2839 /* we suppose the run is continuous, last=next...*/
2840 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2842 stopparen, recursed, NULL, f,depth+1);
2845 if (max1 < minnext + deltanext)
2846 max1 = minnext + deltanext;
2847 if (deltanext == I32_MAX)
2848 is_inf = is_inf_internal = 1;
2850 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2852 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2853 if ( stopmin > minnext)
2854 stopmin = min + min1;
2855 flags &= ~SCF_DO_SUBSTR;
2857 data->flags |= SCF_SEEN_ACCEPT;
2860 if (data_fake.flags & SF_HAS_EVAL)
2861 data->flags |= SF_HAS_EVAL;
2862 data->whilem_c = data_fake.whilem_c;
2864 if (flags & SCF_DO_STCLASS)
2865 cl_or(pRExC_state, &accum, &this_class);
2867 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2869 if (flags & SCF_DO_SUBSTR) {
2870 data->pos_min += min1;
2871 data->pos_delta += max1 - min1;
2872 if (max1 != min1 || is_inf)
2873 data->longest = &(data->longest_float);
2876 delta += max1 - min1;
2877 if (flags & SCF_DO_STCLASS_OR) {
2878 cl_or(pRExC_state, data->start_class, &accum);
2880 cl_and(data->start_class, and_withp);
2881 flags &= ~SCF_DO_STCLASS;
2884 else if (flags & SCF_DO_STCLASS_AND) {
2886 cl_and(data->start_class, &accum);
2887 flags &= ~SCF_DO_STCLASS;
2890 /* Switch to OR mode: cache the old value of
2891 * data->start_class */
2893 StructCopy(data->start_class, and_withp,
2894 struct regnode_charclass_class);
2895 flags &= ~SCF_DO_STCLASS_AND;
2896 StructCopy(&accum, data->start_class,
2897 struct regnode_charclass_class);
2898 flags |= SCF_DO_STCLASS_OR;
2899 data->start_class->flags |= ANYOF_EOS;
2903 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2906 Assuming this was/is a branch we are dealing with: 'scan' now
2907 points at the item that follows the branch sequence, whatever
2908 it is. We now start at the beginning of the sequence and look
2915 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2917 If we can find such a subsequence we need to turn the first
2918 element into a trie and then add the subsequent branch exact
2919 strings to the trie.
2923 1. patterns where the whole set of branches can be converted.
2925 2. patterns where only a subset can be converted.
2927 In case 1 we can replace the whole set with a single regop
2928 for the trie. In case 2 we need to keep the start and end
2931 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2932 becomes BRANCH TRIE; BRANCH X;
2934 There is an additional case, that being where there is a
2935 common prefix, which gets split out into an EXACT like node
2936 preceding the TRIE node.
2938 If x(1..n)==tail then we can do a simple trie, if not we make
2939 a "jump" trie, such that when we match the appropriate word
2940 we "jump" to the appropriate tail node. Essentially we turn
2941 a nested if into a case structure of sorts.
2946 if (!re_trie_maxbuff) {
2947 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2948 if (!SvIOK(re_trie_maxbuff))
2949 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2951 if ( SvIV(re_trie_maxbuff)>=0 ) {
2953 regnode *first = (regnode *)NULL;
2954 regnode *last = (regnode *)NULL;
2955 regnode *tail = scan;
2960 SV * const mysv = sv_newmortal(); /* for dumping */
2962 /* var tail is used because there may be a TAIL
2963 regop in the way. Ie, the exacts will point to the
2964 thing following the TAIL, but the last branch will
2965 point at the TAIL. So we advance tail. If we
2966 have nested (?:) we may have to move through several
2970 while ( OP( tail ) == TAIL ) {
2971 /* this is the TAIL generated by (?:) */
2972 tail = regnext( tail );
2977 regprop(RExC_rx, mysv, tail );
2978 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2979 (int)depth * 2 + 2, "",
2980 "Looking for TRIE'able sequences. Tail node is: ",
2981 SvPV_nolen_const( mysv )
2987 step through the branches, cur represents each
2988 branch, noper is the first thing to be matched
2989 as part of that branch and noper_next is the
2990 regnext() of that node. if noper is an EXACT
2991 and noper_next is the same as scan (our current
2992 position in the regex) then the EXACT branch is
2993 a possible optimization target. Once we have
2994 two or more consecutive such branches we can
2995 create a trie of the EXACT's contents and stich
2996 it in place. If the sequence represents all of
2997 the branches we eliminate the whole thing and
2998 replace it with a single TRIE. If it is a
2999 subsequence then we need to stitch it in. This
3000 means the first branch has to remain, and needs
3001 to be repointed at the item on the branch chain
3002 following the last branch optimized. This could
3003 be either a BRANCH, in which case the
3004 subsequence is internal, or it could be the
3005 item following the branch sequence in which
3006 case the subsequence is at the end.
3010 /* dont use tail as the end marker for this traverse */
3011 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
3012 regnode * const noper = NEXTOPER( cur );
3013 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
3014 regnode * const noper_next = regnext( noper );
3018 regprop(RExC_rx, mysv, cur);
3019 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
3020 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
3022 regprop(RExC_rx, mysv, noper);
3023 PerlIO_printf( Perl_debug_log, " -> %s",
3024 SvPV_nolen_const(mysv));
3027 regprop(RExC_rx, mysv, noper_next );
3028 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
3029 SvPV_nolen_const(mysv));
3031 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
3032 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
3034 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
3035 : PL_regkind[ OP( noper ) ] == EXACT )
3036 || OP(noper) == NOTHING )
3038 && noper_next == tail
3043 if ( !first || optype == NOTHING ) {
3044 if (!first) first = cur;
3045 optype = OP( noper );
3051 Currently the trie logic handles case insensitive matching properly only
3052 when the pattern is UTF-8 and the node is EXACTFU (thus forcing unicode
3055 If/when this is fixed the following define can be swapped
3056 in below to fully enable trie logic.
3058 #define TRIE_TYPE_IS_SAFE 1
3061 #define TRIE_TYPE_IS_SAFE ((UTF && optype == EXACTFU) || optype==EXACT)
3063 if ( last && TRIE_TYPE_IS_SAFE ) {
3064 make_trie( pRExC_state,
3065 startbranch, first, cur, tail, count,
3068 if ( PL_regkind[ OP( noper ) ] == EXACT
3070 && noper_next == tail
3075 optype = OP( noper );
3085 regprop(RExC_rx, mysv, cur);
3086 PerlIO_printf( Perl_debug_log,
3087 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
3088 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
3092 if ( last && TRIE_TYPE_IS_SAFE ) {
3093 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3094 #ifdef TRIE_STUDY_OPT
3095 if ( ((made == MADE_EXACT_TRIE &&
3096 startbranch == first)
3097 || ( first_non_open == first )) &&
3099 flags |= SCF_TRIE_RESTUDY;
3100 if ( startbranch == first
3103 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
3113 else if ( code == BRANCHJ ) { /* single branch is optimized. */
3114 scan = NEXTOPER(NEXTOPER(scan));
3115 } else /* single branch is optimized. */
3116 scan = NEXTOPER(scan);
3118 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
3119 scan_frame *newframe = NULL;
3124 if (OP(scan) != SUSPEND) {
3125 /* set the pointer */
3126 if (OP(scan) == GOSUB) {
3128 RExC_recurse[ARG2L(scan)] = scan;
3129 start = RExC_open_parens[paren-1];
3130 end = RExC_close_parens[paren-1];
3133 start = RExC_rxi->program + 1;
3137 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
3138 SAVEFREEPV(recursed);
3140 if (!PAREN_TEST(recursed,paren+1)) {
3141 PAREN_SET(recursed,paren+1);
3142 Newx(newframe,1,scan_frame);
3144 if (flags & SCF_DO_SUBSTR) {
3145 SCAN_COMMIT(pRExC_state,data,minlenp);
3146 data->longest = &(data->longest_float);
3148 is_inf = is_inf_internal = 1;
3149 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3150 cl_anything(pRExC_state, data->start_class);
3151 flags &= ~SCF_DO_STCLASS;
3154 Newx(newframe,1,scan_frame);
3157 end = regnext(scan);
3162 SAVEFREEPV(newframe);
3163 newframe->next = regnext(scan);
3164 newframe->last = last;
3165 newframe->stop = stopparen;
3166 newframe->prev = frame;
3176 else if (OP(scan) == EXACT) {
3177 I32 l = STR_LEN(scan);
3180 const U8 * const s = (U8*)STRING(scan);
3181 l = utf8_length(s, s + l);
3182 uc = utf8_to_uvchr(s, NULL);
3184 uc = *((U8*)STRING(scan));
3187 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3188 /* The code below prefers earlier match for fixed
3189 offset, later match for variable offset. */
3190 if (data->last_end == -1) { /* Update the start info. */
3191 data->last_start_min = data->pos_min;
3192 data->last_start_max = is_inf
3193 ? I32_MAX : data->pos_min + data->pos_delta;
3195 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3197 SvUTF8_on(data->last_found);
3199 SV * const sv = data->last_found;
3200 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3201 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3202 if (mg && mg->mg_len >= 0)
3203 mg->mg_len += utf8_length((U8*)STRING(scan),
3204 (U8*)STRING(scan)+STR_LEN(scan));
3206 data->last_end = data->pos_min + l;
3207 data->pos_min += l; /* As in the first entry. */
3208 data->flags &= ~SF_BEFORE_EOL;
3210 if (flags & SCF_DO_STCLASS_AND) {
3211 /* Check whether it is compatible with what we know already! */
3215 /* If compatible, we or it in below. It is compatible if is
3216 * in the bitmp and either 1) its bit or its fold is set, or 2)
3217 * it's for a locale. Even if there isn't unicode semantics
3218 * here, at runtime there may be because of matching against a
3219 * utf8 string, so accept a possible false positive for
3220 * latin1-range folds */
3222 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3223 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3224 && (!(data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD)
3225 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3230 ANYOF_CLASS_ZERO(data->start_class);
3231 ANYOF_BITMAP_ZERO(data->start_class);
3233 ANYOF_BITMAP_SET(data->start_class, uc);
3234 else if (uc >= 0x100) {
3237 /* Some Unicode code points fold to the Latin1 range; as
3238 * XXX temporary code, instead of figuring out if this is
3239 * one, just assume it is and set all the start class bits
3240 * that could be some such above 255 code point's fold
3241 * which will generate fals positives. As the code
3242 * elsewhere that does compute the fold settles down, it
3243 * can be extracted out and re-used here */
3244 for (i = 0; i < 256; i++){
3245 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3246 ANYOF_BITMAP_SET(data->start_class, i);
3250 data->start_class->flags &= ~ANYOF_EOS;
3252 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3254 else if (flags & SCF_DO_STCLASS_OR) {
3255 /* false positive possible if the class is case-folded */
3257 ANYOF_BITMAP_SET(data->start_class, uc);
3259 data->start_class->flags |= ANYOF_UNICODE_ALL;
3260 data->start_class->flags &= ~ANYOF_EOS;
3261 cl_and(data->start_class, and_withp);
3263 flags &= ~SCF_DO_STCLASS;
3265 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3266 I32 l = STR_LEN(scan);
3267 UV uc = *((U8*)STRING(scan));
3269 /* Search for fixed substrings supports EXACT only. */
3270 if (flags & SCF_DO_SUBSTR) {
3272 SCAN_COMMIT(pRExC_state, data, minlenp);
3275 const U8 * const s = (U8 *)STRING(scan);
3276 l = utf8_length(s, s + l);
3277 uc = utf8_to_uvchr(s, NULL);
3280 if (flags & SCF_DO_SUBSTR)
3282 if (flags & SCF_DO_STCLASS_AND) {
3283 /* Check whether it is compatible with what we know already! */
3286 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3287 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3288 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3292 ANYOF_CLASS_ZERO(data->start_class);
3293 ANYOF_BITMAP_ZERO(data->start_class);
3295 ANYOF_BITMAP_SET(data->start_class, uc);
3296 data->start_class->flags &= ~ANYOF_EOS;
3297 data->start_class->flags |= ANYOF_LOC_NONBITMAP_FOLD;
3298 if (OP(scan) == EXACTFL) {
3299 /* XXX This set is probably no longer necessary, and
3300 * probably wrong as LOCALE now is on in the initial
3302 data->start_class->flags |= ANYOF_LOCALE;
3306 /* Also set the other member of the fold pair. In case
3307 * that unicode semantics is called for at runtime, use
3308 * the full latin1 fold. (Can't do this for locale,
3309 * because not known until runtime */
3310 ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]);
3313 else if (uc >= 0x100) {
3315 for (i = 0; i < 256; i++){
3316 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3317 ANYOF_BITMAP_SET(data->start_class, i);
3322 else if (flags & SCF_DO_STCLASS_OR) {
3323 if (data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) {
3324 /* false positive possible if the class is case-folded.
3325 Assume that the locale settings are the same... */
3327 ANYOF_BITMAP_SET(data->start_class, uc);
3328 if (OP(scan) != EXACTFL) {
3330 /* And set the other member of the fold pair, but
3331 * can't do that in locale because not known until
3333 ANYOF_BITMAP_SET(data->start_class,
3334 PL_fold_latin1[uc]);
3337 data->start_class->flags &= ~ANYOF_EOS;
3339 cl_and(data->start_class, and_withp);
3341 flags &= ~SCF_DO_STCLASS;
3343 else if (REGNODE_VARIES(OP(scan))) {
3344 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3345 I32 f = flags, pos_before = 0;
3346 regnode * const oscan = scan;
3347 struct regnode_charclass_class this_class;
3348 struct regnode_charclass_class *oclass = NULL;
3349 I32 next_is_eval = 0;
3351 switch (PL_regkind[OP(scan)]) {
3352 case WHILEM: /* End of (?:...)* . */
3353 scan = NEXTOPER(scan);
3356 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3357 next = NEXTOPER(scan);
3358 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3360 maxcount = REG_INFTY;
3361 next = regnext(scan);
3362 scan = NEXTOPER(scan);
3366 if (flags & SCF_DO_SUBSTR)
3371 if (flags & SCF_DO_STCLASS) {
3373 maxcount = REG_INFTY;
3374 next = regnext(scan);
3375 scan = NEXTOPER(scan);
3378 is_inf = is_inf_internal = 1;
3379 scan = regnext(scan);
3380 if (flags & SCF_DO_SUBSTR) {
3381 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3382 data->longest = &(data->longest_float);
3384 goto optimize_curly_tail;
3386 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3387 && (scan->flags == stopparen))
3392 mincount = ARG1(scan);
3393 maxcount = ARG2(scan);
3395 next = regnext(scan);
3396 if (OP(scan) == CURLYX) {
3397 I32 lp = (data ? *(data->last_closep) : 0);
3398 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3400 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3401 next_is_eval = (OP(scan) == EVAL);
3403 if (flags & SCF_DO_SUBSTR) {
3404 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3405 pos_before = data->pos_min;
3409 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3411 data->flags |= SF_IS_INF;
3413 if (flags & SCF_DO_STCLASS) {
3414 cl_init(pRExC_state, &this_class);
3415 oclass = data->start_class;
3416 data->start_class = &this_class;
3417 f |= SCF_DO_STCLASS_AND;
3418 f &= ~SCF_DO_STCLASS_OR;
3420 /* Exclude from super-linear cache processing any {n,m}
3421 regops for which the combination of input pos and regex
3422 pos is not enough information to determine if a match
3425 For example, in the regex /foo(bar\s*){4,8}baz/ with the
3426 regex pos at the \s*, the prospects for a match depend not
3427 only on the input position but also on how many (bar\s*)
3428 repeats into the {4,8} we are. */
3429 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
3430 f &= ~SCF_WHILEM_VISITED_POS;
3432 /* This will finish on WHILEM, setting scan, or on NULL: */
3433 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3434 last, data, stopparen, recursed, NULL,
3436 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3438 if (flags & SCF_DO_STCLASS)
3439 data->start_class = oclass;
3440 if (mincount == 0 || minnext == 0) {
3441 if (flags & SCF_DO_STCLASS_OR) {
3442 cl_or(pRExC_state, data->start_class, &this_class);
3444 else if (flags & SCF_DO_STCLASS_AND) {
3445 /* Switch to OR mode: cache the old value of
3446 * data->start_class */
3448 StructCopy(data->start_class, and_withp,
3449 struct regnode_charclass_class);
3450 flags &= ~SCF_DO_STCLASS_AND;
3451 StructCopy(&this_class, data->start_class,
3452 struct regnode_charclass_class);
3453 flags |= SCF_DO_STCLASS_OR;
3454 data->start_class->flags |= ANYOF_EOS;
3456 } else { /* Non-zero len */
3457 if (flags & SCF_DO_STCLASS_OR) {
3458 cl_or(pRExC_state, data->start_class, &this_class);
3459 cl_and(data->start_class, and_withp);
3461 else if (flags & SCF_DO_STCLASS_AND)
3462 cl_and(data->start_class, &this_class);
3463 flags &= ~SCF_DO_STCLASS;
3465 if (!scan) /* It was not CURLYX, but CURLY. */
3467 if ( /* ? quantifier ok, except for (?{ ... }) */
3468 (next_is_eval || !(mincount == 0 && maxcount == 1))
3469 && (minnext == 0) && (deltanext == 0)
3470 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3471 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3473 ckWARNreg(RExC_parse,
3474 "Quantifier unexpected on zero-length expression");
3477 min += minnext * mincount;
3478 is_inf_internal |= ((maxcount == REG_INFTY
3479 && (minnext + deltanext) > 0)
3480 || deltanext == I32_MAX);
3481 is_inf |= is_inf_internal;
3482 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3484 /* Try powerful optimization CURLYX => CURLYN. */
3485 if ( OP(oscan) == CURLYX && data
3486 && data->flags & SF_IN_PAR
3487 && !(data->flags & SF_HAS_EVAL)
3488 && !deltanext && minnext == 1 ) {
3489 /* Try to optimize to CURLYN. */
3490 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3491 regnode * const nxt1 = nxt;
3498 if (!REGNODE_SIMPLE(OP(nxt))
3499 && !(PL_regkind[OP(nxt)] == EXACT
3500 && STR_LEN(nxt) == 1))
3506 if (OP(nxt) != CLOSE)
3508 if (RExC_open_parens) {
3509 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3510 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3512 /* Now we know that nxt2 is the only contents: */
3513 oscan->flags = (U8)ARG(nxt);
3515 OP(nxt1) = NOTHING; /* was OPEN. */
3518 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3519 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3520 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
3521 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3522 OP(nxt + 1) = OPTIMIZED; /* was count. */
3523 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
3528 /* Try optimization CURLYX => CURLYM. */
3529 if ( OP(oscan) == CURLYX && data
3530 && !(data->flags & SF_HAS_PAR)
3531 && !(data->flags & SF_HAS_EVAL)
3532 && !deltanext /* atom is fixed width */
3533 && minnext != 0 /* CURLYM can't handle zero width */
3535 /* XXXX How to optimize if data == 0? */
3536 /* Optimize to a simpler form. */
3537 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3541 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3542 && (OP(nxt2) != WHILEM))
3544 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3545 /* Need to optimize away parenths. */
3546 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
3547 /* Set the parenth number. */
3548 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3550 oscan->flags = (U8)ARG(nxt);
3551 if (RExC_open_parens) {
3552 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3553 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3555 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3556 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3559 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3560 OP(nxt + 1) = OPTIMIZED; /* was count. */
3561 NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */
3562 NEXT_OFF(nxt + 1) = 0; /* just for consistency. */
3565 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3566 regnode *nnxt = regnext(nxt1);
3568 if (reg_off_by_arg[OP(nxt1)])
3569 ARG_SET(nxt1, nxt2 - nxt1);
3570 else if (nxt2 - nxt1 < U16_MAX)
3571 NEXT_OFF(nxt1) = nxt2 - nxt1;
3573 OP(nxt) = NOTHING; /* Cannot beautify */
3578 /* Optimize again: */
3579 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3580 NULL, stopparen, recursed, NULL, 0,depth+1);
3585 else if ((OP(oscan) == CURLYX)
3586 && (flags & SCF_WHILEM_VISITED_POS)
3587 /* See the comment on a similar expression above.
3588 However, this time it's not a subexpression
3589 we care about, but the expression itself. */
3590 && (maxcount == REG_INFTY)
3591 && data && ++data->whilem_c < 16) {
3592 /* This stays as CURLYX, we can put the count/of pair. */
3593 /* Find WHILEM (as in regexec.c) */
3594 regnode *nxt = oscan + NEXT_OFF(oscan);
3596 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3598 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3599 | (RExC_whilem_seen << 4)); /* On WHILEM */
3601 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3603 if (flags & SCF_DO_SUBSTR) {
3604 SV *last_str = NULL;
3605 int counted = mincount != 0;
3607 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3608 #if defined(SPARC64_GCC_WORKAROUND)
3611 const char *s = NULL;
3614 if (pos_before >= data->last_start_min)
3617 b = data->last_start_min;
3620 s = SvPV_const(data->last_found, l);
3621 old = b - data->last_start_min;
3624 I32 b = pos_before >= data->last_start_min
3625 ? pos_before : data->last_start_min;
3627 const char * const s = SvPV_const(data->last_found, l);
3628 I32 old = b - data->last_start_min;
3632 old = utf8_hop((U8*)s, old) - (U8*)s;
3634 /* Get the added string: */
3635 last_str = newSVpvn_utf8(s + old, l, UTF);
3636 if (deltanext == 0 && pos_before == b) {
3637 /* What was added is a constant string */
3639 SvGROW(last_str, (mincount * l) + 1);
3640 repeatcpy(SvPVX(last_str) + l,
3641 SvPVX_const(last_str), l, mincount - 1);
3642 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3643 /* Add additional parts. */
3644 SvCUR_set(data->last_found,
3645 SvCUR(data->last_found) - l);
3646 sv_catsv(data->last_found, last_str);
3648 SV * sv = data->last_found;
3650 SvUTF8(sv) && SvMAGICAL(sv) ?
3651 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3652 if (mg && mg->mg_len >= 0)
3653 mg->mg_len += CHR_SVLEN(last_str) - l;
3655 data->last_end += l * (mincount - 1);
3658 /* start offset must point into the last copy */
3659 data->last_start_min += minnext * (mincount - 1);
3660 data->last_start_max += is_inf ? I32_MAX
3661 : (maxcount - 1) * (minnext + data->pos_delta);
3664 /* It is counted once already... */
3665 data->pos_min += minnext * (mincount - counted);
3666 data->pos_delta += - counted * deltanext +
3667 (minnext + deltanext) * maxcount - minnext * mincount;
3668 if (mincount != maxcount) {
3669 /* Cannot extend fixed substrings found inside
3671 SCAN_COMMIT(pRExC_state,data,minlenp);
3672 if (mincount && last_str) {
3673 SV * const sv = data->last_found;
3674 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3675 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3679 sv_setsv(sv, last_str);
3680 data->last_end = data->pos_min;
3681 data->last_start_min =
3682 data->pos_min - CHR_SVLEN(last_str);
3683 data->last_start_max = is_inf
3685 : data->pos_min + data->pos_delta
3686 - CHR_SVLEN(last_str);
3688 data->longest = &(data->longest_float);
3690 SvREFCNT_dec(last_str);
3692 if (data && (fl & SF_HAS_EVAL))
3693 data->flags |= SF_HAS_EVAL;
3694 optimize_curly_tail:
3695 if (OP(oscan) != CURLYX) {
3696 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3698 NEXT_OFF(oscan) += NEXT_OFF(next);
3701 default: /* REF, ANYOFV, and CLUMP only? */
3702 if (flags & SCF_DO_SUBSTR) {
3703 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3704 data->longest = &(data->longest_float);
3706 is_inf = is_inf_internal = 1;
3707 if (flags & SCF_DO_STCLASS_OR)
3708 cl_anything(pRExC_state, data->start_class);
3709 flags &= ~SCF_DO_STCLASS;
3713 else if (OP(scan) == LNBREAK) {
3714 if (flags & SCF_DO_STCLASS) {
3716 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3717 if (flags & SCF_DO_STCLASS_AND) {
3718 for (value = 0; value < 256; value++)
3719 if (!is_VERTWS_cp(value))
3720 ANYOF_BITMAP_CLEAR(data->start_class, value);
3723 for (value = 0; value < 256; value++)
3724 if (is_VERTWS_cp(value))
3725 ANYOF_BITMAP_SET(data->start_class, value);
3727 if (flags & SCF_DO_STCLASS_OR)
3728 cl_and(data->start_class, and_withp);
3729 flags &= ~SCF_DO_STCLASS;
3733 if (flags & SCF_DO_SUBSTR) {
3734 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3736 data->pos_delta += 1;
3737 data->longest = &(data->longest_float);
3740 else if (OP(scan) == FOLDCHAR) {
3741 int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2;
3742 flags &= ~SCF_DO_STCLASS;
3745 if (flags & SCF_DO_SUBSTR) {
3746 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3748 data->pos_delta += d;
3749 data->longest = &(data->longest_float);
3752 else if (REGNODE_SIMPLE(OP(scan))) {
3755 if (flags & SCF_DO_SUBSTR) {
3756 SCAN_COMMIT(pRExC_state,data,minlenp);
3760 if (flags & SCF_DO_STCLASS) {
3761 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3763 /* Some of the logic below assumes that switching
3764 locale on will only add false positives. */
3765 switch (PL_regkind[OP(scan)]) {
3769 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3770 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3771 cl_anything(pRExC_state, data->start_class);
3774 if (OP(scan) == SANY)
3776 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3777 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3778 || ANYOF_CLASS_TEST_ANY_SET(data->start_class));
3779 cl_anything(pRExC_state, data->start_class);
3781 if (flags & SCF_DO_STCLASS_AND || !value)
3782 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3785 if (flags & SCF_DO_STCLASS_AND)
3786 cl_and(data->start_class,
3787 (struct regnode_charclass_class*)scan);
3789 cl_or(pRExC_state, data->start_class,
3790 (struct regnode_charclass_class*)scan);
3793 if (flags & SCF_DO_STCLASS_AND) {
3794 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3795 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3796 if (OP(scan) == ALNUMU) {
3797 for (value = 0; value < 256; value++) {
3798 if (!isWORDCHAR_L1(value)) {
3799 ANYOF_BITMAP_CLEAR(data->start_class, value);
3803 for (value = 0; value < 256; value++) {
3804 if (!isALNUM(value)) {
3805 ANYOF_BITMAP_CLEAR(data->start_class, value);
3812 if (data->start_class->flags & ANYOF_LOCALE)
3813 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3815 /* Even if under locale, set the bits for non-locale
3816 * in case it isn't a true locale-node. This will
3817 * create false positives if it truly is locale */
3818 if (OP(scan) == ALNUMU) {
3819 for (value = 0; value < 256; value++) {
3820 if (isWORDCHAR_L1(value)) {
3821 ANYOF_BITMAP_SET(data->start_class, value);
3825 for (value = 0; value < 256; value++) {
3826 if (isALNUM(value)) {
3827 ANYOF_BITMAP_SET(data->start_class, value);
3834 if (flags & SCF_DO_STCLASS_AND) {
3835 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3836 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3837 if (OP(scan) == NALNUMU) {
3838 for (value = 0; value < 256; value++) {
3839 if (isWORDCHAR_L1(value)) {
3840 ANYOF_BITMAP_CLEAR(data->start_class, value);
3844 for (value = 0; value < 256; value++) {
3845 if (isALNUM(value)) {
3846 ANYOF_BITMAP_CLEAR(data->start_class, value);
3853 if (data->start_class->flags & ANYOF_LOCALE)
3854 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3856 /* Even if under locale, set the bits for non-locale in
3857 * case it isn't a true locale-node. This will create
3858 * false positives if it truly is locale */
3859 if (OP(scan) == NALNUMU) {
3860 for (value = 0; value < 256; value++) {
3861 if (! isWORDCHAR_L1(value)) {
3862 ANYOF_BITMAP_SET(data->start_class, value);
3866 for (value = 0; value < 256; value++) {
3867 if (! isALNUM(value)) {
3868 ANYOF_BITMAP_SET(data->start_class, value);
3875 if (flags & SCF_DO_STCLASS_AND) {
3876 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3877 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3878 if (OP(scan) == SPACEU) {
3879 for (value = 0; value < 256; value++) {
3880 if (!isSPACE_L1(value)) {
3881 ANYOF_BITMAP_CLEAR(data->start_class, value);
3885 for (value = 0; value < 256; value++) {
3886 if (!isSPACE(value)) {
3887 ANYOF_BITMAP_CLEAR(data->start_class, value);
3894 if (data->start_class->flags & ANYOF_LOCALE) {
3895 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3897 if (OP(scan) == SPACEU) {
3898 for (value = 0; value < 256; value++) {
3899 if (isSPACE_L1(value)) {
3900 ANYOF_BITMAP_SET(data->start_class, value);
3904 for (value = 0; value < 256; value++) {
3905 if (isSPACE(value)) {
3906 ANYOF_BITMAP_SET(data->start_class, value);
3913 if (flags & SCF_DO_STCLASS_AND) {
3914 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3915 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3916 if (OP(scan) == NSPACEU) {
3917 for (value = 0; value < 256; value++) {
3918 if (isSPACE_L1(value)) {
3919 ANYOF_BITMAP_CLEAR(data->start_class, value);
3923 for (value = 0; value < 256; value++) {
3924 if (isSPACE(value)) {
3925 ANYOF_BITMAP_CLEAR(data->start_class, value);
3932 if (data->start_class->flags & ANYOF_LOCALE)
3933 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3934 if (OP(scan) == NSPACEU) {
3935 for (value = 0; value < 256; value++) {
3936 if (!isSPACE_L1(value)) {
3937 ANYOF_BITMAP_SET(data->start_class, value);
3942 for (value = 0; value < 256; value++) {
3943 if (!isSPACE(value)) {
3944 ANYOF_BITMAP_SET(data->start_class, value);
3951 if (flags & SCF_DO_STCLASS_AND) {
3952 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3953 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3954 for (value = 0; value < 256; value++)
3955 if (!isDIGIT(value))
3956 ANYOF_BITMAP_CLEAR(data->start_class, value);
3960 if (data->start_class->flags & ANYOF_LOCALE)
3961 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3962 for (value = 0; value < 256; value++)
3964 ANYOF_BITMAP_SET(data->start_class, value);
3968 if (flags & SCF_DO_STCLASS_AND) {
3969 if (!(data->start_class->flags & ANYOF_LOCALE))
3970 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3971 for (value = 0; value < 256; value++)
3973 ANYOF_BITMAP_CLEAR(data->start_class, value);
3976 if (data->start_class->flags & ANYOF_LOCALE)
3977 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3978 for (value = 0; value < 256; value++)
3979 if (!isDIGIT(value))
3980 ANYOF_BITMAP_SET(data->start_class, value);
3983 CASE_SYNST_FNC(VERTWS);
3984 CASE_SYNST_FNC(HORIZWS);
3987 if (flags & SCF_DO_STCLASS_OR)
3988 cl_and(data->start_class, and_withp);
3989 flags &= ~SCF_DO_STCLASS;
3992 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3993 data->flags |= (OP(scan) == MEOL
3997 else if ( PL_regkind[OP(scan)] == BRANCHJ
3998 /* Lookbehind, or need to calculate parens/evals/stclass: */
3999 && (scan->flags || data || (flags & SCF_DO_STCLASS))
4000 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
4001 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4002 || OP(scan) == UNLESSM )
4004 /* Negative Lookahead/lookbehind
4005 In this case we can't do fixed string optimisation.
4008 I32 deltanext, minnext, fake = 0;
4010 struct regnode_charclass_class intrnl;
4013 data_fake.flags = 0;
4015 data_fake.whilem_c = data->whilem_c;
4016 data_fake.last_closep = data->last_closep;
4019 data_fake.last_closep = &fake;
4020 data_fake.pos_delta = delta;
4021 if ( flags & SCF_DO_STCLASS && !scan->flags
4022 && OP(scan) == IFMATCH ) { /* Lookahead */
4023 cl_init(pRExC_state, &intrnl);
4024 data_fake.start_class = &intrnl;
4025 f |= SCF_DO_STCLASS_AND;
4027 if (flags & SCF_WHILEM_VISITED_POS)
4028 f |= SCF_WHILEM_VISITED_POS;
4029 next = regnext(scan);
4030 nscan = NEXTOPER(NEXTOPER(scan));
4031 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
4032 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
4035 FAIL("Variable length lookbehind not implemented");
4037 else if (minnext > (I32)U8_MAX) {
4038 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4040 scan->flags = (U8)minnext;
4043 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4045 if (data_fake.flags & SF_HAS_EVAL)
4046 data->flags |= SF_HAS_EVAL;
4047 data->whilem_c = data_fake.whilem_c;
4049 if (f & SCF_DO_STCLASS_AND) {
4050 if (flags & SCF_DO_STCLASS_OR) {
4051 /* OR before, AND after: ideally we would recurse with
4052 * data_fake to get the AND applied by study of the
4053 * remainder of the pattern, and then derecurse;
4054 * *** HACK *** for now just treat as "no information".
4055 * See [perl #56690].
4057 cl_init(pRExC_state, data->start_class);
4059 /* AND before and after: combine and continue */
4060 const int was = (data->start_class->flags & ANYOF_EOS);
4062 cl_and(data->start_class, &intrnl);
4064 data->start_class->flags |= ANYOF_EOS;
4068 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4070 /* Positive Lookahead/lookbehind
4071 In this case we can do fixed string optimisation,
4072 but we must be careful about it. Note in the case of
4073 lookbehind the positions will be offset by the minimum
4074 length of the pattern, something we won't know about
4075 until after the recurse.
4077 I32 deltanext, fake = 0;
4079 struct regnode_charclass_class intrnl;
4081 /* We use SAVEFREEPV so that when the full compile
4082 is finished perl will clean up the allocated
4083 minlens when it's all done. This way we don't
4084 have to worry about freeing them when we know
4085 they wont be used, which would be a pain.
4088 Newx( minnextp, 1, I32 );
4089 SAVEFREEPV(minnextp);
4092 StructCopy(data, &data_fake, scan_data_t);
4093 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
4096 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
4097 data_fake.last_found=newSVsv(data->last_found);
4101 data_fake.last_closep = &fake;
4102 data_fake.flags = 0;
4103 data_fake.pos_delta = delta;
4105 data_fake.flags |= SF_IS_INF;
4106 if ( flags & SCF_DO_STCLASS && !scan->flags
4107 && OP(scan) == IFMATCH ) { /* Lookahead */
4108 cl_init(pRExC_state, &intrnl);
4109 data_fake.start_class = &intrnl;
4110 f |= SCF_DO_STCLASS_AND;
4112 if (flags & SCF_WHILEM_VISITED_POS)
4113 f |= SCF_WHILEM_VISITED_POS;
4114 next = regnext(scan);
4115 nscan = NEXTOPER(NEXTOPER(scan));
4117 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
4118 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
4121 FAIL("Variable length lookbehind not implemented");
4123 else if (*minnextp > (I32)U8_MAX) {
4124 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4126 scan->flags = (U8)*minnextp;
4131 if (f & SCF_DO_STCLASS_AND) {
4132 const int was = (data->start_class->flags & ANYOF_EOS);
4134 cl_and(data->start_class, &intrnl);
4136 data->start_class->flags |= ANYOF_EOS;
4139 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4141 if (data_fake.flags & SF_HAS_EVAL)
4142 data->flags |= SF_HAS_EVAL;
4143 data->whilem_c = data_fake.whilem_c;
4144 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
4145 if (RExC_rx->minlen<*minnextp)
4146 RExC_rx->minlen=*minnextp;
4147 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
4148 SvREFCNT_dec(data_fake.last_found);
4150 if ( data_fake.minlen_fixed != minlenp )
4152 data->offset_fixed= data_fake.offset_fixed;
4153 data->minlen_fixed= data_fake.minlen_fixed;
4154 data->lookbehind_fixed+= scan->flags;
4156 if ( data_fake.minlen_float != minlenp )
4158 data->minlen_float= data_fake.minlen_float;
4159 data->offset_float_min=data_fake.offset_float_min;
4160 data->offset_float_max=data_fake.offset_float_max;
4161 data->lookbehind_float+= scan->flags;
4170 else if (OP(scan) == OPEN) {
4171 if (stopparen != (I32)ARG(scan))
4174 else if (OP(scan) == CLOSE) {
4175 if (stopparen == (I32)ARG(scan)) {
4178 if ((I32)ARG(scan) == is_par) {
4179 next = regnext(scan);
4181 if ( next && (OP(next) != WHILEM) && next < last)
4182 is_par = 0; /* Disable optimization */
4185 *(data->last_closep) = ARG(scan);
4187 else if (OP(scan) == EVAL) {
4189 data->flags |= SF_HAS_EVAL;
4191 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
4192 if (flags & SCF_DO_SUBSTR) {
4193 SCAN_COMMIT(pRExC_state,data,minlenp);
4194 flags &= ~SCF_DO_SUBSTR;
4196 if (data && OP(scan)==ACCEPT) {
4197 data->flags |= SCF_SEEN_ACCEPT;
4202 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
4204 if (flags & SCF_DO_SUBSTR) {
4205 SCAN_COMMIT(pRExC_state,data,minlenp);
4206 data->longest = &(data->longest_float);
4208 is_inf = is_inf_internal = 1;
4209 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
4210 cl_anything(pRExC_state, data->start_class);
4211 flags &= ~SCF_DO_STCLASS;
4213 else if (OP(scan) == GPOS) {
4214 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
4215 !(delta || is_inf || (data && data->pos_delta)))
4217 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
4218 RExC_rx->extflags |= RXf_ANCH_GPOS;
4219 if (RExC_rx->gofs < (U32)min)
4220 RExC_rx->gofs = min;
4222 RExC_rx->extflags |= RXf_GPOS_FLOAT;
4226 #ifdef TRIE_STUDY_OPT
4227 #ifdef FULL_TRIE_STUDY
4228 else if (PL_regkind[OP(scan)] == TRIE) {
4229 /* NOTE - There is similar code to this block above for handling
4230 BRANCH nodes on the initial study. If you change stuff here
4232 regnode *trie_node= scan;
4233 regnode *tail= regnext(scan);
4234 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4235 I32 max1 = 0, min1 = I32_MAX;
4236 struct regnode_charclass_class accum;
4238 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
4239 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
4240 if (flags & SCF_DO_STCLASS)
4241 cl_init_zero(pRExC_state, &accum);
4247 const regnode *nextbranch= NULL;
4250 for ( word=1 ; word <= trie->wordcount ; word++)
4252 I32 deltanext=0, minnext=0, f = 0, fake;
4253 struct regnode_charclass_class this_class;
4255 data_fake.flags = 0;
4257 data_fake.whilem_c = data->whilem_c;
4258 data_fake.last_closep = data->last_closep;
4261 data_fake.last_closep = &fake;
4262 data_fake.pos_delta = delta;
4263 if (flags & SCF_DO_STCLASS) {
4264 cl_init(pRExC_state, &this_class);
4265 data_fake.start_class = &this_class;
4266 f = SCF_DO_STCLASS_AND;
4268 if (flags & SCF_WHILEM_VISITED_POS)
4269 f |= SCF_WHILEM_VISITED_POS;
4271 if (trie->jump[word]) {
4273 nextbranch = trie_node + trie->jump[0];
4274 scan= trie_node + trie->jump[word];
4275 /* We go from the jump point to the branch that follows
4276 it. Note this means we need the vestigal unused branches
4277 even though they arent otherwise used.
4279 minnext = study_chunk(pRExC_state, &scan, minlenp,
4280 &deltanext, (regnode *)nextbranch, &data_fake,
4281 stopparen, recursed, NULL, f,depth+1);
4283 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4284 nextbranch= regnext((regnode*)nextbranch);
4286 if (min1 > (I32)(minnext + trie->minlen))
4287 min1 = minnext + trie->minlen;
4288 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4289 max1 = minnext + deltanext + trie->maxlen;
4290 if (deltanext == I32_MAX)
4291 is_inf = is_inf_internal = 1;
4293 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4295 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4296 if ( stopmin > min + min1)
4297 stopmin = min + min1;
4298 flags &= ~SCF_DO_SUBSTR;
4300 data->flags |= SCF_SEEN_ACCEPT;
4303 if (data_fake.flags & SF_HAS_EVAL)
4304 data->flags |= SF_HAS_EVAL;
4305 data->whilem_c = data_fake.whilem_c;
4307 if (flags & SCF_DO_STCLASS)
4308 cl_or(pRExC_state, &accum, &this_class);
4311 if (flags & SCF_DO_SUBSTR) {
4312 data->pos_min += min1;
4313 data->pos_delta += max1 - min1;
4314 if (max1 != min1 || is_inf)
4315 data->longest = &(data->longest_float);
4318 delta += max1 - min1;
4319 if (flags & SCF_DO_STCLASS_OR) {
4320 cl_or(pRExC_state, data->start_class, &accum);
4322 cl_and(data->start_class, and_withp);
4323 flags &= ~SCF_DO_STCLASS;
4326 else if (flags & SCF_DO_STCLASS_AND) {
4328 cl_and(data->start_class, &accum);
4329 flags &= ~SCF_DO_STCLASS;
4332 /* Switch to OR mode: cache the old value of
4333 * data->start_class */
4335 StructCopy(data->start_class, and_withp,
4336 struct regnode_charclass_class);
4337 flags &= ~SCF_DO_STCLASS_AND;
4338 StructCopy(&accum, data->start_class,
4339 struct regnode_charclass_class);
4340 flags |= SCF_DO_STCLASS_OR;
4341 data->start_class->flags |= ANYOF_EOS;
4348 else if (PL_regkind[OP(scan)] == TRIE) {
4349 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4352 min += trie->minlen;
4353 delta += (trie->maxlen - trie->minlen);
4354 flags &= ~SCF_DO_STCLASS; /* xxx */
4355 if (flags & SCF_DO_SUBSTR) {
4356 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4357 data->pos_min += trie->minlen;
4358 data->pos_delta += (trie->maxlen - trie->minlen);
4359 if (trie->maxlen != trie->minlen)
4360 data->longest = &(data->longest_float);
4362 if (trie->jump) /* no more substrings -- for now /grr*/
4363 flags &= ~SCF_DO_SUBSTR;
4365 #endif /* old or new */
4366 #endif /* TRIE_STUDY_OPT */
4368 /* Else: zero-length, ignore. */
4369 scan = regnext(scan);
4374 stopparen = frame->stop;
4375 frame = frame->prev;
4376 goto fake_study_recurse;
4381 DEBUG_STUDYDATA("pre-fin:",data,depth);
4384 *deltap = is_inf_internal ? I32_MAX : delta;
4385 if (flags & SCF_DO_SUBSTR && is_inf)
4386 data->pos_delta = I32_MAX - data->pos_min;
4387 if (is_par > (I32)U8_MAX)
4389 if (is_par && pars==1 && data) {
4390 data->flags |= SF_IN_PAR;
4391 data->flags &= ~SF_HAS_PAR;
4393 else if (pars && data) {
4394 data->flags |= SF_HAS_PAR;
4395 data->flags &= ~SF_IN_PAR;
4397 if (flags & SCF_DO_STCLASS_OR)
4398 cl_and(data->start_class, and_withp);
4399 if (flags & SCF_TRIE_RESTUDY)
4400 data->flags |= SCF_TRIE_RESTUDY;
4402 DEBUG_STUDYDATA("post-fin:",data,depth);
4404 return min < stopmin ? min : stopmin;
4408 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4410 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4412 PERL_ARGS_ASSERT_ADD_DATA;
4414 Renewc(RExC_rxi->data,
4415 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4416 char, struct reg_data);
4418 Renew(RExC_rxi->data->what, count + n, U8);
4420 Newx(RExC_rxi->data->what, n, U8);
4421 RExC_rxi->data->count = count + n;
4422 Copy(s, RExC_rxi->data->what + count, n, U8);
4426 /*XXX: todo make this not included in a non debugging perl */
4427 #ifndef PERL_IN_XSUB_RE
4429 Perl_reginitcolors(pTHX)
4432 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4434 char *t = savepv(s);
4438 t = strchr(t, '\t');
4444 PL_colors[i] = t = (char *)"";
4449 PL_colors[i++] = (char *)"";
4456 #ifdef TRIE_STUDY_OPT
4457 #define CHECK_RESTUDY_GOTO \
4459 (data.flags & SCF_TRIE_RESTUDY) \
4463 #define CHECK_RESTUDY_GOTO
4467 - pregcomp - compile a regular expression into internal code
4469 * We can't allocate space until we know how big the compiled form will be,
4470 * but we can't compile it (and thus know how big it is) until we've got a
4471 * place to put the code. So we cheat: we compile it twice, once with code
4472 * generation turned off and size counting turned on, and once "for real".
4473 * This also means that we don't allocate space until we are sure that the
4474 * thing really will compile successfully, and we never have to move the
4475 * code and thus invalidate pointers into it. (Note that it has to be in
4476 * one piece because free() must be able to free it all.) [NB: not true in perl]
4478 * Beware that the optimization-preparation code in here knows about some
4479 * of the structure of the compiled regexp. [I'll say.]
4484 #ifndef PERL_IN_XSUB_RE
4485 #define RE_ENGINE_PTR &PL_core_reg_engine
4487 extern const struct regexp_engine my_reg_engine;
4488 #define RE_ENGINE_PTR &my_reg_engine
4491 #ifndef PERL_IN_XSUB_RE
4493 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4496 HV * const table = GvHV(PL_hintgv);
4498 PERL_ARGS_ASSERT_PREGCOMP;
4500 /* Dispatch a request to compile a regexp to correct
4503 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4504 GET_RE_DEBUG_FLAGS_DECL;
4505 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4506 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4508 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4511 return CALLREGCOMP_ENG(eng, pattern, flags);
4514 return Perl_re_compile(aTHX_ pattern, flags);
4519 Perl_re_compile(pTHX_ SV * const pattern, U32 orig_pm_flags)
4524 register regexp_internal *ri;
4533 /* these are all flags - maybe they should be turned
4534 * into a single int with different bit masks */
4535 I32 sawlookahead = 0;
4538 bool used_setjump = FALSE;
4539 regex_charset initial_charset = get_regex_charset(orig_pm_flags);
4544 RExC_state_t RExC_state;
4545 RExC_state_t * const pRExC_state = &RExC_state;
4546 #ifdef TRIE_STUDY_OPT
4548 RExC_state_t copyRExC_state;
4550 GET_RE_DEBUG_FLAGS_DECL;
4552 PERL_ARGS_ASSERT_RE_COMPILE;
4554 DEBUG_r(if (!PL_colorset) reginitcolors());
4556 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4557 RExC_uni_semantics = 0;
4558 RExC_contains_locale = 0;
4560 /****************** LONG JUMP TARGET HERE***********************/
4561 /* Longjmp back to here if have to switch in midstream to utf8 */
4562 if (! RExC_orig_utf8) {
4563 JMPENV_PUSH(jump_ret);
4564 used_setjump = TRUE;
4567 if (jump_ret == 0) { /* First time through */
4568 exp = SvPV(pattern, plen);
4570 /* ignore the utf8ness if the pattern is 0 length */
4572 RExC_utf8 = RExC_orig_utf8 = 0;
4576 SV *dsv= sv_newmortal();
4577 RE_PV_QUOTED_DECL(s, RExC_utf8,
4578 dsv, exp, plen, 60);
4579 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4580 PL_colors[4],PL_colors[5],s);
4583 else { /* longjumped back */
4586 /* If the cause for the longjmp was other than changing to utf8, pop
4587 * our own setjmp, and longjmp to the correct handler */
4588 if (jump_ret != UTF8_LONGJMP) {
4590 JMPENV_JUMP(jump_ret);
4595 /* It's possible to write a regexp in ascii that represents Unicode
4596 codepoints outside of the byte range, such as via \x{100}. If we
4597 detect such a sequence we have to convert the entire pattern to utf8
4598 and then recompile, as our sizing calculation will have been based
4599 on 1 byte == 1 character, but we will need to use utf8 to encode
4600 at least some part of the pattern, and therefore must convert the whole
4603 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4604 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4605 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4607 RExC_orig_utf8 = RExC_utf8 = 1;
4611 #ifdef TRIE_STUDY_OPT
4615 pm_flags = orig_pm_flags;
4617 if (initial_charset == REGEX_LOCALE_CHARSET) {
4618 RExC_contains_locale = 1;
4620 else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) {
4622 /* Set to use unicode semantics if the pattern is in utf8 and has the
4623 * 'depends' charset specified, as it means unicode when utf8 */
4624 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4628 RExC_flags = pm_flags;
4632 RExC_in_lookbehind = 0;
4633 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4634 RExC_seen_evals = 0;
4636 RExC_override_recoding = 0;
4638 /* First pass: determine size, legality. */
4646 RExC_emit = &PL_regdummy;
4647 RExC_whilem_seen = 0;
4648 RExC_open_parens = NULL;
4649 RExC_close_parens = NULL;
4651 RExC_paren_names = NULL;
4653 RExC_paren_name_list = NULL;
4655 RExC_recurse = NULL;
4656 RExC_recurse_count = 0;
4658 #if 0 /* REGC() is (currently) a NOP at the first pass.
4659 * Clever compilers notice this and complain. --jhi */
4660 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4662 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4663 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4664 RExC_precomp = NULL;
4668 /* Here, finished first pass. Get rid of any added setjmp */
4674 PerlIO_printf(Perl_debug_log,
4675 "Required size %"IVdf" nodes\n"
4676 "Starting second pass (creation)\n",
4679 RExC_lastparse=NULL;
4682 /* The first pass could have found things that force Unicode semantics */
4683 if ((RExC_utf8 || RExC_uni_semantics)
4684 && get_regex_charset(pm_flags) == REGEX_DEPENDS_CHARSET)
4686 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4689 /* Small enough for pointer-storage convention?
4690 If extralen==0, this means that we will not need long jumps. */
4691 if (RExC_size >= 0x10000L && RExC_extralen)
4692 RExC_size += RExC_extralen;
4695 if (RExC_whilem_seen > 15)
4696 RExC_whilem_seen = 15;
4698 /* Allocate space and zero-initialize. Note, the two step process
4699 of zeroing when in debug mode, thus anything assigned has to
4700 happen after that */
4701 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4702 r = (struct regexp*)SvANY(rx);
4703 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4704 char, regexp_internal);
4705 if ( r == NULL || ri == NULL )
4706 FAIL("Regexp out of space");
4708 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4709 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4711 /* bulk initialize base fields with 0. */
4712 Zero(ri, sizeof(regexp_internal), char);
4715 /* non-zero initialization begins here */
4717 r->engine= RE_ENGINE_PTR;
4718 r->extflags = pm_flags;
4720 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4721 bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET);
4723 /* The caret is output if there are any defaults: if not all the STD
4724 * flags are set, or if no character set specifier is needed */
4726 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4728 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4729 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4730 >> RXf_PMf_STD_PMMOD_SHIFT);
4731 const char *fptr = STD_PAT_MODS; /*"msix"*/
4733 /* Allocate for the worst case, which is all the std flags are turned
4734 * on. If more precision is desired, we could do a population count of
4735 * the flags set. This could be done with a small lookup table, or by
4736 * shifting, masking and adding, or even, when available, assembly
4737 * language for a machine-language population count.
4738 * We never output a minus, as all those are defaults, so are
4739 * covered by the caret */
4740 const STRLEN wraplen = plen + has_p + has_runon
4741 + has_default /* If needs a caret */
4743 /* If needs a character set specifier */
4744 + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0)
4745 + (sizeof(STD_PAT_MODS) - 1)
4746 + (sizeof("(?:)") - 1);
4748 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4750 SvFLAGS(rx) |= SvUTF8(pattern);
4753 /* If a default, cover it using the caret */
4755 *p++= DEFAULT_PAT_MOD;
4759 const char* const name = get_regex_charset_name(r->extflags, &len);
4760 Copy(name, p, len, char);
4764 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4767 while((ch = *fptr++)) {
4775 Copy(RExC_precomp, p, plen, char);
4776 assert ((RX_WRAPPED(rx) - p) < 16);
4777 r->pre_prefix = p - RX_WRAPPED(rx);
4783 SvCUR_set(rx, p - SvPVX_const(rx));
4787 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4789 if (RExC_seen & REG_SEEN_RECURSE) {
4790 Newxz(RExC_open_parens, RExC_npar,regnode *);
4791 SAVEFREEPV(RExC_open_parens);
4792 Newxz(RExC_close_parens,RExC_npar,regnode *);
4793 SAVEFREEPV(RExC_close_parens);
4796 /* Useful during FAIL. */
4797 #ifdef RE_TRACK_PATTERN_OFFSETS
4798 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4799 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4800 "%s %"UVuf" bytes for offset annotations.\n",
4801 ri->u.offsets ? "Got" : "Couldn't get",
4802 (UV)((2*RExC_size+1) * sizeof(U32))));
4804 SetProgLen(ri,RExC_size);
4809 /* Second pass: emit code. */
4810 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4815 RExC_emit_start = ri->program;
4816 RExC_emit = ri->program;
4817 RExC_emit_bound = ri->program + RExC_size + 1;
4819 /* Store the count of eval-groups for security checks: */
4820 RExC_rx->seen_evals = RExC_seen_evals;
4821 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4822 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4826 /* XXXX To minimize changes to RE engine we always allocate
4827 3-units-long substrs field. */
4828 Newx(r->substrs, 1, struct reg_substr_data);
4829 if (RExC_recurse_count) {
4830 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4831 SAVEFREEPV(RExC_recurse);
4835 r->minlen = minlen = sawlookahead = sawplus = sawopen = 0;
4836 Zero(r->substrs, 1, struct reg_substr_data);
4838 #ifdef TRIE_STUDY_OPT
4840 StructCopy(&zero_scan_data, &data, scan_data_t);
4841 copyRExC_state = RExC_state;
4844 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4846 RExC_state = copyRExC_state;
4847 if (seen & REG_TOP_LEVEL_BRANCHES)
4848 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4850 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4851 if (data.last_found) {
4852 SvREFCNT_dec(data.longest_fixed);
4853 SvREFCNT_dec(data.longest_float);
4854 SvREFCNT_dec(data.last_found);
4856 StructCopy(&zero_scan_data, &data, scan_data_t);
4859 StructCopy(&zero_scan_data, &data, scan_data_t);
4862 /* Dig out information for optimizations. */
4863 r->extflags = RExC_flags; /* was pm_op */
4864 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4867 SvUTF8_on(rx); /* Unicode in it? */
4868 ri->regstclass = NULL;
4869 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4870 r->intflags |= PREGf_NAUGHTY;
4871 scan = ri->program + 1; /* First BRANCH. */
4873 /* testing for BRANCH here tells us whether there is "must appear"
4874 data in the pattern. If there is then we can use it for optimisations */
4875 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4877 STRLEN longest_float_length, longest_fixed_length;
4878 struct regnode_charclass_class ch_class; /* pointed to by data */
4880 I32 last_close = 0; /* pointed to by data */
4881 regnode *first= scan;
4882 regnode *first_next= regnext(first);
4884 * Skip introductions and multiplicators >= 1
4885 * so that we can extract the 'meat' of the pattern that must
4886 * match in the large if() sequence following.
4887 * NOTE that EXACT is NOT covered here, as it is normally
4888 * picked up by the optimiser separately.
4890 * This is unfortunate as the optimiser isnt handling lookahead
4891 * properly currently.
4894 while ((OP(first) == OPEN && (sawopen = 1)) ||
4895 /* An OR of *one* alternative - should not happen now. */
4896 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4897 /* for now we can't handle lookbehind IFMATCH*/
4898 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
4899 (OP(first) == PLUS) ||
4900 (OP(first) == MINMOD) ||
4901 /* An {n,m} with n>0 */
4902 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4903 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4906 * the only op that could be a regnode is PLUS, all the rest
4907 * will be regnode_1 or regnode_2.
4910 if (OP(first) == PLUS)
4913 first += regarglen[OP(first)];
4915 first = NEXTOPER(first);
4916 first_next= regnext(first);
4919 /* Starting-point info. */
4921 DEBUG_PEEP("first:",first,0);
4922 /* Ignore EXACT as we deal with it later. */
4923 if (PL_regkind[OP(first)] == EXACT) {
4924 if (OP(first) == EXACT)
4925 NOOP; /* Empty, get anchored substr later. */
4927 ri->regstclass = first;
4930 else if (PL_regkind[OP(first)] == TRIE &&
4931 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4934 /* this can happen only on restudy */
4935 if ( OP(first) == TRIE ) {
4936 struct regnode_1 *trieop = (struct regnode_1 *)
4937 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4938 StructCopy(first,trieop,struct regnode_1);
4939 trie_op=(regnode *)trieop;
4941 struct regnode_charclass *trieop = (struct regnode_charclass *)
4942 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4943 StructCopy(first,trieop,struct regnode_charclass);
4944 trie_op=(regnode *)trieop;
4947 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4948 ri->regstclass = trie_op;
4951 else if (REGNODE_SIMPLE(OP(first)))
4952 ri->regstclass = first;
4953 else if (PL_regkind[OP(first)] == BOUND ||
4954 PL_regkind[OP(first)] == NBOUND)
4955 ri->regstclass = first;
4956 else if (PL_regkind[OP(first)] == BOL) {
4957 r->extflags |= (OP(first) == MBOL
4959 : (OP(first) == SBOL
4962 first = NEXTOPER(first);
4965 else if (OP(first) == GPOS) {
4966 r->extflags |= RXf_ANCH_GPOS;
4967 first = NEXTOPER(first);
4970 else if ((!sawopen || !RExC_sawback) &&
4971 (OP(first) == STAR &&
4972 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4973 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4975 /* turn .* into ^.* with an implied $*=1 */
4977 (OP(NEXTOPER(first)) == REG_ANY)
4980 r->extflags |= type;
4981 r->intflags |= PREGf_IMPLICIT;
4982 first = NEXTOPER(first);
4985 if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback)
4986 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4987 /* x+ must match at the 1st pos of run of x's */
4988 r->intflags |= PREGf_SKIP;
4990 /* Scan is after the zeroth branch, first is atomic matcher. */
4991 #ifdef TRIE_STUDY_OPT
4994 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4995 (IV)(first - scan + 1))
4999 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
5000 (IV)(first - scan + 1))
5006 * If there's something expensive in the r.e., find the
5007 * longest literal string that must appear and make it the
5008 * regmust. Resolve ties in favor of later strings, since
5009 * the regstart check works with the beginning of the r.e.
5010 * and avoiding duplication strengthens checking. Not a
5011 * strong reason, but sufficient in the absence of others.
5012 * [Now we resolve ties in favor of the earlier string if
5013 * it happens that c_offset_min has been invalidated, since the
5014 * earlier string may buy us something the later one won't.]
5017 data.longest_fixed = newSVpvs("");
5018 data.longest_float = newSVpvs("");
5019 data.last_found = newSVpvs("");
5020 data.longest = &(data.longest_fixed);
5022 if (!ri->regstclass) {
5023 cl_init(pRExC_state, &ch_class);
5024 data.start_class = &ch_class;
5025 stclass_flag = SCF_DO_STCLASS_AND;
5026 } else /* XXXX Check for BOUND? */
5028 data.last_closep = &last_close;
5030 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
5031 &data, -1, NULL, NULL,
5032 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
5038 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
5039 && data.last_start_min == 0 && data.last_end > 0
5040 && !RExC_seen_zerolen
5041 && !(RExC_seen & REG_SEEN_VERBARG)
5042 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
5043 r->extflags |= RXf_CHECK_ALL;
5044 scan_commit(pRExC_state, &data,&minlen,0);
5045 SvREFCNT_dec(data.last_found);
5047 /* Note that code very similar to this but for anchored string
5048 follows immediately below, changes may need to be made to both.
5051 longest_float_length = CHR_SVLEN(data.longest_float);
5052 if (longest_float_length
5053 || (data.flags & SF_FL_BEFORE_EOL
5054 && (!(data.flags & SF_FL_BEFORE_MEOL)
5055 || (RExC_flags & RXf_PMf_MULTILINE))))
5059 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
5060 && data.offset_fixed == data.offset_float_min
5061 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
5062 goto remove_float; /* As in (a)+. */
5064 /* copy the information about the longest float from the reg_scan_data
5065 over to the program. */
5066 if (SvUTF8(data.longest_float)) {
5067 r->float_utf8 = data.longest_float;
5068 r->float_substr = NULL;
5070 r->float_substr = data.longest_float;
5071 r->float_utf8 = NULL;
5073 /* float_end_shift is how many chars that must be matched that
5074 follow this item. We calculate it ahead of time as once the
5075 lookbehind offset is added in we lose the ability to correctly
5077 ml = data.minlen_float ? *(data.minlen_float)
5078 : (I32)longest_float_length;
5079 r->float_end_shift = ml - data.offset_float_min
5080 - longest_float_length + (SvTAIL(data.longest_float) != 0)
5081 + data.lookbehind_float;
5082 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
5083 r->float_max_offset = data.offset_float_max;
5084 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
5085 r->float_max_offset -= data.lookbehind_float;
5087 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
5088 && (!(data.flags & SF_FL_BEFORE_MEOL)
5089 || (RExC_flags & RXf_PMf_MULTILINE)));
5090 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
5094 r->float_substr = r->float_utf8 = NULL;
5095 SvREFCNT_dec(data.longest_float);
5096 longest_float_length = 0;
5099 /* Note that code very similar to this but for floating string
5100 is immediately above, changes may need to be made to both.
5103 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
5104 if (longest_fixed_length
5105 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
5106 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5107 || (RExC_flags & RXf_PMf_MULTILINE))))
5111 /* copy the information about the longest fixed
5112 from the reg_scan_data over to the program. */
5113 if (SvUTF8(data.longest_fixed)) {
5114 r->anchored_utf8 = data.longest_fixed;
5115 r->anchored_substr = NULL;
5117 r->anchored_substr = data.longest_fixed;
5118 r->anchored_utf8 = NULL;
5120 /* fixed_end_shift is how many chars that must be matched that
5121 follow this item. We calculate it ahead of time as once the
5122 lookbehind offset is added in we lose the ability to correctly
5124 ml = data.minlen_fixed ? *(data.minlen_fixed)
5125 : (I32)longest_fixed_length;
5126 r->anchored_end_shift = ml - data.offset_fixed
5127 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
5128 + data.lookbehind_fixed;
5129 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
5131 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
5132 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5133 || (RExC_flags & RXf_PMf_MULTILINE)));
5134 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
5137 r->anchored_substr = r->anchored_utf8 = NULL;
5138 SvREFCNT_dec(data.longest_fixed);
5139 longest_fixed_length = 0;
5142 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
5143 ri->regstclass = NULL;
5145 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
5147 && !(data.start_class->flags & ANYOF_EOS)
5148 && !cl_is_anything(data.start_class))
5150 const U32 n = add_data(pRExC_state, 1, "f");
5151 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5153 Newx(RExC_rxi->data->data[n], 1,
5154 struct regnode_charclass_class);
5155 StructCopy(data.start_class,
5156 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5157 struct regnode_charclass_class);
5158 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5159 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5160 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
5161 regprop(r, sv, (regnode*)data.start_class);
5162 PerlIO_printf(Perl_debug_log,
5163 "synthetic stclass \"%s\".\n",
5164 SvPVX_const(sv));});
5167 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
5168 if (longest_fixed_length > longest_float_length) {
5169 r->check_end_shift = r->anchored_end_shift;
5170 r->check_substr = r->anchored_substr;
5171 r->check_utf8 = r->anchored_utf8;
5172 r->check_offset_min = r->check_offset_max = r->anchored_offset;
5173 if (r->extflags & RXf_ANCH_SINGLE)
5174 r->extflags |= RXf_NOSCAN;
5177 r->check_end_shift = r->float_end_shift;
5178 r->check_substr = r->float_substr;
5179 r->check_utf8 = r->float_utf8;
5180 r->check_offset_min = r->float_min_offset;
5181 r->check_offset_max = r->float_max_offset;
5183 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
5184 This should be changed ASAP! */
5185 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
5186 r->extflags |= RXf_USE_INTUIT;
5187 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
5188 r->extflags |= RXf_INTUIT_TAIL;
5190 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
5191 if ( (STRLEN)minlen < longest_float_length )
5192 minlen= longest_float_length;
5193 if ( (STRLEN)minlen < longest_fixed_length )
5194 minlen= longest_fixed_length;
5198 /* Several toplevels. Best we can is to set minlen. */
5200 struct regnode_charclass_class ch_class;
5203 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
5205 scan = ri->program + 1;
5206 cl_init(pRExC_state, &ch_class);
5207 data.start_class = &ch_class;
5208 data.last_closep = &last_close;
5211 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
5212 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
5216 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
5217 = r->float_substr = r->float_utf8 = NULL;
5219 if (!(data.start_class->flags & ANYOF_EOS)
5220 && !cl_is_anything(data.start_class))
5222 const U32 n = add_data(pRExC_state, 1, "f");
5223 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5225 Newx(RExC_rxi->data->data[n], 1,
5226 struct regnode_charclass_class);
5227 StructCopy(data.start_class,
5228 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5229 struct regnode_charclass_class);
5230 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5231 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5232 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
5233 regprop(r, sv, (regnode*)data.start_class);
5234 PerlIO_printf(Perl_debug_log,
5235 "synthetic stclass \"%s\".\n",
5236 SvPVX_const(sv));});
5240 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
5241 the "real" pattern. */
5243 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
5244 (IV)minlen, (IV)r->minlen);
5246 r->minlenret = minlen;
5247 if (r->minlen < minlen)
5250 if (RExC_seen & REG_SEEN_GPOS)
5251 r->extflags |= RXf_GPOS_SEEN;
5252 if (RExC_seen & REG_SEEN_LOOKBEHIND)
5253 r->extflags |= RXf_LOOKBEHIND_SEEN;
5254 if (RExC_seen & REG_SEEN_EVAL)
5255 r->extflags |= RXf_EVAL_SEEN;
5256 if (RExC_seen & REG_SEEN_CANY)
5257 r->extflags |= RXf_CANY_SEEN;
5258 if (RExC_seen & REG_SEEN_VERBARG)
5259 r->intflags |= PREGf_VERBARG_SEEN;
5260 if (RExC_seen & REG_SEEN_CUTGROUP)
5261 r->intflags |= PREGf_CUTGROUP_SEEN;
5262 if (RExC_paren_names)
5263 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
5265 RXp_PAREN_NAMES(r) = NULL;
5267 #ifdef STUPID_PATTERN_CHECKS
5268 if (RX_PRELEN(rx) == 0)
5269 r->extflags |= RXf_NULL;
5270 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5271 /* XXX: this should happen BEFORE we compile */
5272 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5273 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
5274 r->extflags |= RXf_WHITE;
5275 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
5276 r->extflags |= RXf_START_ONLY;
5278 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5279 /* XXX: this should happen BEFORE we compile */
5280 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5282 regnode *first = ri->program + 1;
5285 if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END)
5286 r->extflags |= RXf_NULL;
5287 else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END)
5288 r->extflags |= RXf_START_ONLY;
5289 else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE
5290 && OP(regnext(first)) == END)
5291 r->extflags |= RXf_WHITE;
5295 if (RExC_paren_names) {
5296 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5297 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5300 ri->name_list_idx = 0;
5302 if (RExC_recurse_count) {
5303 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5304 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5305 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5308 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5309 /* assume we don't need to swap parens around before we match */
5312 PerlIO_printf(Perl_debug_log,"Final program:\n");
5315 #ifdef RE_TRACK_PATTERN_OFFSETS
5316 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5317 const U32 len = ri->u.offsets[0];
5319 GET_RE_DEBUG_FLAGS_DECL;
5320 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5321 for (i = 1; i <= len; i++) {
5322 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5323 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5324 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5326 PerlIO_printf(Perl_debug_log, "\n");
5332 #undef RE_ENGINE_PTR
5336 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5339 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5341 PERL_UNUSED_ARG(value);
5343 if (flags & RXapif_FETCH) {
5344 return reg_named_buff_fetch(rx, key, flags);
5345 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5346 Perl_croak_no_modify(aTHX);
5348 } else if (flags & RXapif_EXISTS) {
5349 return reg_named_buff_exists(rx, key, flags)
5352 } else if (flags & RXapif_REGNAMES) {
5353 return reg_named_buff_all(rx, flags);
5354 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5355 return reg_named_buff_scalar(rx, flags);
5357 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5363 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5366 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5367 PERL_UNUSED_ARG(lastkey);
5369 if (flags & RXapif_FIRSTKEY)
5370 return reg_named_buff_firstkey(rx, flags);
5371 else if (flags & RXapif_NEXTKEY)
5372 return reg_named_buff_nextkey(rx, flags);
5374 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5380 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5383 AV *retarray = NULL;
5385 struct regexp *const rx = (struct regexp *)SvANY(r);
5387 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5389 if (flags & RXapif_ALL)
5392 if (rx && RXp_PAREN_NAMES(rx)) {
5393 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5396 SV* sv_dat=HeVAL(he_str);
5397 I32 *nums=(I32*)SvPVX(sv_dat);
5398 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5399 if ((I32)(rx->nparens) >= nums[i]
5400 && rx->offs[nums[i]].start != -1
5401 && rx->offs[nums[i]].end != -1)
5404 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5408 ret = newSVsv(&PL_sv_undef);
5411 av_push(retarray, ret);
5414 return newRV_noinc(MUTABLE_SV(retarray));
5421 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5424 struct regexp *const rx = (struct regexp *)SvANY(r);
5426 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5428 if (rx && RXp_PAREN_NAMES(rx)) {
5429 if (flags & RXapif_ALL) {
5430 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5432 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5446 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5448 struct regexp *const rx = (struct regexp *)SvANY(r);
5450 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5452 if ( rx && RXp_PAREN_NAMES(rx) ) {
5453 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5455 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5462 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5464 struct regexp *const rx = (struct regexp *)SvANY(r);
5465 GET_RE_DEBUG_FLAGS_DECL;
5467 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5469 if (rx && RXp_PAREN_NAMES(rx)) {
5470 HV *hv = RXp_PAREN_NAMES(rx);
5472 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5475 SV* sv_dat = HeVAL(temphe);
5476 I32 *nums = (I32*)SvPVX(sv_dat);
5477 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5478 if ((I32)(rx->lastparen) >= nums[i] &&
5479 rx->offs[nums[i]].start != -1 &&
5480 rx->offs[nums[i]].end != -1)
5486 if (parno || flags & RXapif_ALL) {
5487 return newSVhek(HeKEY_hek(temphe));
5495 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5500 struct regexp *const rx = (struct regexp *)SvANY(r);
5502 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5504 if (rx && RXp_PAREN_NAMES(rx)) {
5505 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5506 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5507 } else if (flags & RXapif_ONE) {
5508 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5509 av = MUTABLE_AV(SvRV(ret));
5510 length = av_len(av);
5512 return newSViv(length + 1);
5514 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5518 return &PL_sv_undef;
5522 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5524 struct regexp *const rx = (struct regexp *)SvANY(r);
5527 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5529 if (rx && RXp_PAREN_NAMES(rx)) {
5530 HV *hv= RXp_PAREN_NAMES(rx);
5532 (void)hv_iterinit(hv);
5533 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5536 SV* sv_dat = HeVAL(temphe);
5537 I32 *nums = (I32*)SvPVX(sv_dat);
5538 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5539 if ((I32)(rx->lastparen) >= nums[i] &&
5540 rx->offs[nums[i]].start != -1 &&
5541 rx->offs[nums[i]].end != -1)
5547 if (parno || flags & RXapif_ALL) {
5548 av_push(av, newSVhek(HeKEY_hek(temphe)));
5553 return newRV_noinc(MUTABLE_SV(av));
5557 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5560 struct regexp *const rx = (struct regexp *)SvANY(r);
5565 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5568 sv_setsv(sv,&PL_sv_undef);
5572 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5574 i = rx->offs[0].start;
5578 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5580 s = rx->subbeg + rx->offs[0].end;
5581 i = rx->sublen - rx->offs[0].end;
5584 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5585 (s1 = rx->offs[paren].start) != -1 &&
5586 (t1 = rx->offs[paren].end) != -1)
5590 s = rx->subbeg + s1;
5592 sv_setsv(sv,&PL_sv_undef);
5595 assert(rx->sublen >= (s - rx->subbeg) + i );
5597 const int oldtainted = PL_tainted;
5599 sv_setpvn(sv, s, i);
5600 PL_tainted = oldtainted;
5601 if ( (rx->extflags & RXf_CANY_SEEN)
5602 ? (RXp_MATCH_UTF8(rx)
5603 && (!i || is_utf8_string((U8*)s, i)))
5604 : (RXp_MATCH_UTF8(rx)) )
5611 if (RXp_MATCH_TAINTED(rx)) {
5612 if (SvTYPE(sv) >= SVt_PVMG) {
5613 MAGIC* const mg = SvMAGIC(sv);
5616 SvMAGIC_set(sv, mg->mg_moremagic);
5618 if ((mgt = SvMAGIC(sv))) {
5619 mg->mg_moremagic = mgt;
5620 SvMAGIC_set(sv, mg);
5630 sv_setsv(sv,&PL_sv_undef);
5636 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5637 SV const * const value)
5639 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5641 PERL_UNUSED_ARG(rx);
5642 PERL_UNUSED_ARG(paren);
5643 PERL_UNUSED_ARG(value);
5646 Perl_croak_no_modify(aTHX);
5650 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5653 struct regexp *const rx = (struct regexp *)SvANY(r);
5657 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5659 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5661 /* $` / ${^PREMATCH} */
5662 case RX_BUFF_IDX_PREMATCH:
5663 if (rx->offs[0].start != -1) {
5664 i = rx->offs[0].start;
5672 /* $' / ${^POSTMATCH} */
5673 case RX_BUFF_IDX_POSTMATCH:
5674 if (rx->offs[0].end != -1) {
5675 i = rx->sublen - rx->offs[0].end;
5677 s1 = rx->offs[0].end;
5683 /* $& / ${^MATCH}, $1, $2, ... */
5685 if (paren <= (I32)rx->nparens &&
5686 (s1 = rx->offs[paren].start) != -1 &&
5687 (t1 = rx->offs[paren].end) != -1)
5692 if (ckWARN(WARN_UNINITIALIZED))
5693 report_uninit((const SV *)sv);
5698 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5699 const char * const s = rx->subbeg + s1;
5704 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5711 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5713 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5714 PERL_UNUSED_ARG(rx);
5718 return newSVpvs("Regexp");
5721 /* Scans the name of a named buffer from the pattern.
5722 * If flags is REG_RSN_RETURN_NULL returns null.
5723 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5724 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5725 * to the parsed name as looked up in the RExC_paren_names hash.
5726 * If there is an error throws a vFAIL().. type exception.
5729 #define REG_RSN_RETURN_NULL 0
5730 #define REG_RSN_RETURN_NAME 1
5731 #define REG_RSN_RETURN_DATA 2
5734 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5736 char *name_start = RExC_parse;
5738 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5740 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5741 /* skip IDFIRST by using do...while */
5744 RExC_parse += UTF8SKIP(RExC_parse);
5745 } while (isALNUM_utf8((U8*)RExC_parse));
5749 } while (isALNUM(*RExC_parse));
5754 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5755 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5756 if ( flags == REG_RSN_RETURN_NAME)
5758 else if (flags==REG_RSN_RETURN_DATA) {
5761 if ( ! sv_name ) /* should not happen*/
5762 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5763 if (RExC_paren_names)
5764 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5766 sv_dat = HeVAL(he_str);
5768 vFAIL("Reference to nonexistent named group");
5772 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5779 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5780 int rem=(int)(RExC_end - RExC_parse); \
5789 if (RExC_lastparse!=RExC_parse) \
5790 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5793 iscut ? "..." : "<" \
5796 PerlIO_printf(Perl_debug_log,"%16s",""); \
5799 num = RExC_size + 1; \
5801 num=REG_NODE_NUM(RExC_emit); \
5802 if (RExC_lastnum!=num) \
5803 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5805 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5806 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5807 (int)((depth*2)), "", \
5811 RExC_lastparse=RExC_parse; \
5816 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5817 DEBUG_PARSE_MSG((funcname)); \
5818 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5820 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5821 DEBUG_PARSE_MSG((funcname)); \
5822 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5825 /* This section of code defines the inversion list object and its methods. The
5826 * interfaces are highly subject to change, so as much as possible is static to
5827 * this file. An inversion list is here implemented as a malloc'd C array with
5828 * some added info. More will be coming when functionality is added later.
5830 * It is currently implemented as an HV to the outside world, but is actually
5831 * an SV pointing to an array of UVs that the SV thinks are bytes. This allows
5832 * us to have an array of UV whose memory management is automatically handled
5833 * by the existing facilities for SV's.
5835 * Some of the methods should always be private to the implementation, and some
5836 * should eventually be made public */
5838 #define INVLIST_INITIAL_LEN 10
5840 PERL_STATIC_INLINE UV*
5841 S_invlist_array(pTHX_ HV* const invlist)
5843 /* Returns the pointer to the inversion list's array. Every time the
5844 * length changes, this needs to be called in case malloc or realloc moved
5847 PERL_ARGS_ASSERT_INVLIST_ARRAY;
5849 return (UV *) SvPVX(invlist);
5852 PERL_STATIC_INLINE UV
5853 S_invlist_len(pTHX_ HV* const invlist)
5855 /* Returns the current number of elements in the inversion list's array */
5857 PERL_ARGS_ASSERT_INVLIST_LEN;
5859 return SvCUR(invlist) / sizeof(UV);
5862 PERL_STATIC_INLINE UV
5863 S_invlist_max(pTHX_ HV* const invlist)
5865 /* Returns the maximum number of elements storable in the inversion list's
5866 * array, without having to realloc() */
5868 PERL_ARGS_ASSERT_INVLIST_MAX;
5870 return SvLEN(invlist) / sizeof(UV);
5873 PERL_STATIC_INLINE void
5874 S_invlist_set_len(pTHX_ HV* const invlist, const UV len)
5876 /* Sets the current number of elements stored in the inversion list */
5878 PERL_ARGS_ASSERT_INVLIST_SET_LEN;
5880 SvCUR_set(invlist, len * sizeof(UV));
5883 PERL_STATIC_INLINE void
5884 S_invlist_set_max(pTHX_ HV* const invlist, const UV max)
5887 /* Sets the maximum number of elements storable in the inversion list
5888 * without having to realloc() */
5890 PERL_ARGS_ASSERT_INVLIST_SET_MAX;
5892 if (max < invlist_len(invlist)) {
5893 Perl_croak(aTHX_ "panic: Can't make max size '%"UVuf"' less than current length %"UVuf" in inversion list", invlist_max(invlist), invlist_len(invlist));
5896 SvLEN_set(invlist, max * sizeof(UV));
5899 #ifndef PERL_IN_XSUB_RE
5901 Perl__new_invlist(pTHX_ IV initial_size)
5904 /* Return a pointer to a newly constructed inversion list, with enough
5905 * space to store 'initial_size' elements. If that number is negative, a
5906 * system default is used instead */
5908 if (initial_size < 0) {
5909 initial_size = INVLIST_INITIAL_LEN;
5912 /* Allocate the initial space */
5913 return (HV *) newSV(initial_size * sizeof(UV));
5917 PERL_STATIC_INLINE void
5918 S_invlist_destroy(pTHX_ HV* const invlist)
5920 /* Inversion list destructor */
5922 PERL_ARGS_ASSERT_INVLIST_DESTROY;
5924 SvREFCNT_dec(invlist);
5928 S_invlist_extend(pTHX_ HV* const invlist, const UV new_max)
5930 /* Grow the maximum size of an inversion list */
5932 PERL_ARGS_ASSERT_INVLIST_EXTEND;
5934 SvGROW((SV *)invlist, new_max * sizeof(UV));
5937 PERL_STATIC_INLINE void
5938 S_invlist_trim(pTHX_ HV* const invlist)
5940 PERL_ARGS_ASSERT_INVLIST_TRIM;
5942 /* Change the length of the inversion list to how many entries it currently
5945 SvPV_shrink_to_cur((SV *) invlist);
5948 /* An element is in an inversion list iff its index is even numbered: 0, 2, 4,
5951 #define ELEMENT_IN_INVLIST_SET(i) (! ((i) & 1))
5952 #define PREV_ELEMENT_IN_INVLIST_SET(i) ! ELEMENT_IN_INVLIST_SET(i)
5954 #ifndef PERL_IN_XSUB_RE
5956 Perl__append_range_to_invlist(pTHX_ HV* const invlist, const UV start, const UV end)
5958 /* Subject to change or removal. Append the range from 'start' to 'end' at
5959 * the end of the inversion list. The range must be above any existing
5962 UV* array = invlist_array(invlist);
5963 UV max = invlist_max(invlist);
5964 UV len = invlist_len(invlist);
5966 PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST;
5970 /* Here, the existing list is non-empty. The current max entry in the
5971 * list is generally the first value not in the set, except when the
5972 * set extends to the end of permissible values, in which case it is
5973 * the first entry in that final set, and so this call is an attempt to
5974 * append out-of-order */
5976 UV final_element = len - 1;
5977 if (array[final_element] > start
5978 || ELEMENT_IN_INVLIST_SET(final_element))
5980 Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list");
5983 /* Here, it is a legal append. If the new range begins with the first
5984 * value not in the set, it is extending the set, so the new first
5985 * value not in the set is one greater than the newly extended range.
5987 if (array[final_element] == start) {
5988 if (end != UV_MAX) {
5989 array[final_element] = end + 1;
5992 /* But if the end is the maximum representable on the machine,
5993 * just let the range that this would extend have no end */
5994 invlist_set_len(invlist, len - 1);
6000 /* Here the new range doesn't extend any existing set. Add it */
6002 len += 2; /* Includes an element each for the start and end of range */
6004 /* If overflows the existing space, extend, which may cause the array to be
6007 invlist_extend(invlist, len);
6008 array = invlist_array(invlist);
6011 invlist_set_len(invlist, len);
6013 /* The next item on the list starts the range, the one after that is
6014 * one past the new range. */
6015 array[len - 2] = start;
6016 if (end != UV_MAX) {
6017 array[len - 1] = end + 1;
6020 /* But if the end is the maximum representable on the machine, just let
6021 * the range have no end */
6022 invlist_set_len(invlist, len - 1);
6028 S_invlist_union(pTHX_ HV* const a, HV* const b)
6030 /* Return a new inversion list which is the union of two inversion lists.
6031 * The basis for this comes from "Unicode Demystified" Chapter 13 by
6032 * Richard Gillam, published by Addison-Wesley, and explained at some
6033 * length there. The preface says to incorporate its examples into your
6034 * code at your own risk.
6036 * The algorithm is like a merge sort.
6038 * XXX A potential performance improvement is to keep track as we go along
6039 * if only one of the inputs contributes to the result, meaning the other
6040 * is a subset of that one. In that case, we can skip the final copy and
6041 * return the larger of the input lists */
6043 UV* array_a = invlist_array(a); /* a's array */
6044 UV* array_b = invlist_array(b);
6045 UV len_a = invlist_len(a); /* length of a's array */
6046 UV len_b = invlist_len(b);
6048 HV* u; /* the resulting union */
6052 UV i_a = 0; /* current index into a's array */
6056 /* running count, as explained in the algorithm source book; items are
6057 * stopped accumulating and are output when the count changes to/from 0.
6058 * The count is incremented when we start a range that's in the set, and
6059 * decremented when we start a range that's not in the set. So its range
6060 * is 0 to 2. Only when the count is zero is something not in the set.
6064 PERL_ARGS_ASSERT_INVLIST_UNION;
6066 /* Size the union for the worst case: that the sets are completely
6068 u = _new_invlist(len_a + len_b);
6069 array_u = invlist_array(u);
6071 /* Go through each list item by item, stopping when exhausted one of
6073 while (i_a < len_a && i_b < len_b) {
6074 UV cp; /* The element to potentially add to the union's array */
6075 bool cp_in_set; /* is it in the the input list's set or not */
6077 /* We need to take one or the other of the two inputs for the union.
6078 * Since we are merging two sorted lists, we take the smaller of the
6079 * next items. In case of a tie, we take the one that is in its set
6080 * first. If we took one not in the set first, it would decrement the
6081 * count, possibly to 0 which would cause it to be output as ending the
6082 * range, and the next time through we would take the same number, and
6083 * output it again as beginning the next range. By doing it the
6084 * opposite way, there is no possibility that the count will be
6085 * momentarily decremented to 0, and thus the two adjoining ranges will
6086 * be seamlessly merged. (In a tie and both are in the set or both not
6087 * in the set, it doesn't matter which we take first.) */
6088 if (array_a[i_a] < array_b[i_b]
6089 || (array_a[i_a] == array_b[i_b] && ELEMENT_IN_INVLIST_SET(i_a)))
6091 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6095 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6099 /* Here, have chosen which of the two inputs to look at. Only output
6100 * if the running count changes to/from 0, which marks the
6101 * beginning/end of a range in that's in the set */
6104 array_u[i_u++] = cp;
6111 array_u[i_u++] = cp;
6116 /* Here, we are finished going through at least one of the lists, which
6117 * means there is something remaining in at most one. We check if the list
6118 * that hasn't been exhausted is positioned such that we are in the middle
6119 * of a range in its set or not. (i_a and i_b point to the element beyond
6120 * the one we care about.) If in the set, we decrement 'count'; if 0, there
6121 * is potentially more to output.
6122 * There are four cases:
6123 * 1) Both weren't in their sets, count is 0, and remains 0. What's left
6124 * in the union is entirely from the non-exhausted set.
6125 * 2) Both were in their sets, count is 2. Nothing further should
6126 * be output, as everything that remains will be in the exhausted
6127 * list's set, hence in the union; decrementing to 1 but not 0 insures
6129 * 3) the exhausted was in its set, non-exhausted isn't, count is 1.
6130 * Nothing further should be output because the union includes
6131 * everything from the exhausted set. Not decrementing ensures that.
6132 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1;
6133 * decrementing to 0 insures that we look at the remainder of the
6134 * non-exhausted set */
6135 if ((i_a != len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6136 || (i_b != len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6141 /* The final length is what we've output so far, plus what else is about to
6142 * be output. (If 'count' is non-zero, then the input list we exhausted
6143 * has everything remaining up to the machine's limit in its set, and hence
6144 * in the union, so there will be no further output. */
6147 /* At most one of the subexpressions will be non-zero */
6148 len_u += (len_a - i_a) + (len_b - i_b);
6151 /* Set result to final length, which can change the pointer to array_u, so
6153 if (len_u != invlist_len(u)) {
6154 invlist_set_len(u, len_u);
6156 array_u = invlist_array(u);
6159 /* When 'count' is 0, the list that was exhausted (if one was shorter than
6160 * the other) ended with everything above it not in its set. That means
6161 * that the remaining part of the union is precisely the same as the
6162 * non-exhausted list, so can just copy it unchanged. (If both list were
6163 * exhausted at the same time, then the operations below will be both 0.)
6166 IV copy_count; /* At most one will have a non-zero copy count */
6167 if ((copy_count = len_a - i_a) > 0) {
6168 Copy(array_a + i_a, array_u + i_u, copy_count, UV);
6170 else if ((copy_count = len_b - i_b) > 0) {
6171 Copy(array_b + i_b, array_u + i_u, copy_count, UV);
6179 S_invlist_intersection(pTHX_ HV* const a, HV* const b)
6181 /* Return the intersection of two inversion lists. The basis for this
6182 * comes from "Unicode Demystified" Chapter 13 by Richard Gillam, published
6183 * by Addison-Wesley, and explained at some length there. The preface says
6184 * to incorporate its examples into your code at your own risk. In fact,
6187 * The algorithm is like a merge sort, and is essentially the same as the
6191 UV* array_a = invlist_array(a); /* a's array */
6192 UV* array_b = invlist_array(b);
6193 UV len_a = invlist_len(a); /* length of a's array */
6194 UV len_b = invlist_len(b);
6196 HV* r; /* the resulting intersection */
6200 UV i_a = 0; /* current index into a's array */
6204 /* running count, as explained in the algorithm source book; items are
6205 * stopped accumulating and are output when the count changes to/from 2.
6206 * The count is incremented when we start a range that's in the set, and
6207 * decremented when we start a range that's not in the set. So its range
6208 * is 0 to 2. Only when the count is 2 is something in the intersection.
6212 PERL_ARGS_ASSERT_INVLIST_INTERSECTION;
6214 /* Size the intersection for the worst case: that the intersection ends up
6215 * fragmenting everything to be completely disjoint */
6216 r= _new_invlist(len_a + len_b);
6217 array_r = invlist_array(r);
6219 /* Go through each list item by item, stopping when exhausted one of
6221 while (i_a < len_a && i_b < len_b) {
6222 UV cp; /* The element to potentially add to the intersection's
6224 bool cp_in_set; /* Is it in the input list's set or not */
6226 /* We need to take one or the other of the two inputs for the
6227 * intersection. Since we are merging two sorted lists, we take the
6228 * smaller of the next items. In case of a tie, we take the one that
6229 * is not in its set first (a difference from the union algorithm). If
6230 * we took one in the set first, it would increment the count, possibly
6231 * to 2 which would cause it to be output as starting a range in the
6232 * intersection, and the next time through we would take that same
6233 * number, and output it again as ending the set. By doing it the
6234 * opposite of this, there is no possibility that the count will be
6235 * momentarily incremented to 2. (In a tie and both are in the set or
6236 * both not in the set, it doesn't matter which we take first.) */
6237 if (array_a[i_a] < array_b[i_b]
6238 || (array_a[i_a] == array_b[i_b] && ! ELEMENT_IN_INVLIST_SET(i_a)))
6240 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6244 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6248 /* Here, have chosen which of the two inputs to look at. Only output
6249 * if the running count changes to/from 2, which marks the
6250 * beginning/end of a range that's in the intersection */
6254 array_r[i_r++] = cp;
6259 array_r[i_r++] = cp;
6265 /* Here, we are finished going through at least one of the lists, which
6266 * means there is something remaining in at most one. We check if the list
6267 * that has been exhausted is positioned such that we are in the middle
6268 * of a range in its set or not. (i_a and i_b point to elements 1 beyond
6269 * the ones we care about.) There are four cases:
6270 * 1) Both weren't in their sets, count is 0, and remains 0. There's
6271 * nothing left in the intersection.
6272 * 2) Both were in their sets, count is 2 and perhaps is incremented to
6273 * above 2. What should be output is exactly that which is in the
6274 * non-exhausted set, as everything it has is also in the intersection
6275 * set, and everything it doesn't have can't be in the intersection
6276 * 3) The exhausted was in its set, non-exhausted isn't, count is 1, and
6277 * gets incremented to 2. Like the previous case, the intersection is
6278 * everything that remains in the non-exhausted set.
6279 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1, and
6280 * remains 1. And the intersection has nothing more. */
6281 if ((i_a == len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6282 || (i_b == len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6287 /* The final length is what we've output so far plus what else is in the
6288 * intersection. At most one of the subexpressions below will be non-zero */
6291 len_r += (len_a - i_a) + (len_b - i_b);
6294 /* Set result to final length, which can change the pointer to array_r, so
6296 if (len_r != invlist_len(r)) {
6297 invlist_set_len(r, len_r);
6299 array_r = invlist_array(r);
6302 /* Finish outputting any remaining */
6303 if (count >= 2) { /* At most one will have a non-zero copy count */
6305 if ((copy_count = len_a - i_a) > 0) {
6306 Copy(array_a + i_a, array_r + i_r, copy_count, UV);
6308 else if ((copy_count = len_b - i_b) > 0) {
6309 Copy(array_b + i_b, array_r + i_r, copy_count, UV);
6317 S_add_range_to_invlist(pTHX_ HV* invlist, const UV start, const UV end)
6319 /* Add the range from 'start' to 'end' inclusive to the inversion list's
6320 * set. A pointer to the inversion list is returned. This may actually be
6321 * a new list, in which case the passed in one has been destroyed. The
6322 * passed in inversion list can be NULL, in which case a new one is created
6323 * with just the one range in it */
6329 if (invlist == NULL) {
6330 invlist = _new_invlist(2);
6334 len = invlist_len(invlist);
6337 /* If comes after the final entry, can just append it to the end */
6339 || start >= invlist_array(invlist)
6340 [invlist_len(invlist) - 1])
6342 _append_range_to_invlist(invlist, start, end);
6346 /* Here, can't just append things, create and return a new inversion list
6347 * which is the union of this range and the existing inversion list */
6348 range_invlist = _new_invlist(2);
6349 _append_range_to_invlist(range_invlist, start, end);
6351 added_invlist = invlist_union(invlist, range_invlist);
6353 /* The passed in list can be freed, as well as our temporary */
6354 invlist_destroy(range_invlist);
6355 if (invlist != added_invlist) {
6356 invlist_destroy(invlist);
6359 return added_invlist;
6362 PERL_STATIC_INLINE HV*
6363 S_add_cp_to_invlist(pTHX_ HV* invlist, const UV cp) {
6364 return add_range_to_invlist(invlist, cp, cp);
6367 /* End of inversion list object */
6370 - reg - regular expression, i.e. main body or parenthesized thing
6372 * Caller must absorb opening parenthesis.
6374 * Combining parenthesis handling with the base level of regular expression
6375 * is a trifle forced, but the need to tie the tails of the branches to what
6376 * follows makes it hard to avoid.
6378 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
6380 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
6382 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
6386 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
6387 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
6390 register regnode *ret; /* Will be the head of the group. */
6391 register regnode *br;
6392 register regnode *lastbr;
6393 register regnode *ender = NULL;
6394 register I32 parno = 0;
6396 U32 oregflags = RExC_flags;
6397 bool have_branch = 0;
6399 I32 freeze_paren = 0;
6400 I32 after_freeze = 0;
6402 /* for (?g), (?gc), and (?o) warnings; warning
6403 about (?c) will warn about (?g) -- japhy */
6405 #define WASTED_O 0x01
6406 #define WASTED_G 0x02
6407 #define WASTED_C 0x04
6408 #define WASTED_GC (0x02|0x04)
6409 I32 wastedflags = 0x00;
6411 char * parse_start = RExC_parse; /* MJD */
6412 char * const oregcomp_parse = RExC_parse;
6414 GET_RE_DEBUG_FLAGS_DECL;
6416 PERL_ARGS_ASSERT_REG;
6417 DEBUG_PARSE("reg ");
6419 *flagp = 0; /* Tentatively. */
6422 /* Make an OPEN node, if parenthesized. */
6424 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
6425 char *start_verb = RExC_parse;
6426 STRLEN verb_len = 0;
6427 char *start_arg = NULL;
6428 unsigned char op = 0;
6430 int internal_argval = 0; /* internal_argval is only useful if !argok */
6431 while ( *RExC_parse && *RExC_parse != ')' ) {
6432 if ( *RExC_parse == ':' ) {
6433 start_arg = RExC_parse + 1;
6439 verb_len = RExC_parse - start_verb;
6442 while ( *RExC_parse && *RExC_parse != ')' )
6444 if ( *RExC_parse != ')' )
6445 vFAIL("Unterminated verb pattern argument");
6446 if ( RExC_parse == start_arg )
6449 if ( *RExC_parse != ')' )
6450 vFAIL("Unterminated verb pattern");
6453 switch ( *start_verb ) {
6454 case 'A': /* (*ACCEPT) */
6455 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
6457 internal_argval = RExC_nestroot;
6460 case 'C': /* (*COMMIT) */
6461 if ( memEQs(start_verb,verb_len,"COMMIT") )
6464 case 'F': /* (*FAIL) */
6465 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
6470 case ':': /* (*:NAME) */
6471 case 'M': /* (*MARK:NAME) */
6472 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
6477 case 'P': /* (*PRUNE) */
6478 if ( memEQs(start_verb,verb_len,"PRUNE") )
6481 case 'S': /* (*SKIP) */
6482 if ( memEQs(start_verb,verb_len,"SKIP") )
6485 case 'T': /* (*THEN) */
6486 /* [19:06] <TimToady> :: is then */
6487 if ( memEQs(start_verb,verb_len,"THEN") ) {
6489 RExC_seen |= REG_SEEN_CUTGROUP;
6495 vFAIL3("Unknown verb pattern '%.*s'",
6496 verb_len, start_verb);
6499 if ( start_arg && internal_argval ) {
6500 vFAIL3("Verb pattern '%.*s' may not have an argument",
6501 verb_len, start_verb);
6502 } else if ( argok < 0 && !start_arg ) {
6503 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
6504 verb_len, start_verb);
6506 ret = reganode(pRExC_state, op, internal_argval);
6507 if ( ! internal_argval && ! SIZE_ONLY ) {
6509 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
6510 ARG(ret) = add_data( pRExC_state, 1, "S" );
6511 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
6518 if (!internal_argval)
6519 RExC_seen |= REG_SEEN_VERBARG;
6520 } else if ( start_arg ) {
6521 vFAIL3("Verb pattern '%.*s' may not have an argument",
6522 verb_len, start_verb);
6524 ret = reg_node(pRExC_state, op);
6526 nextchar(pRExC_state);
6529 if (*RExC_parse == '?') { /* (?...) */
6530 bool is_logical = 0;
6531 const char * const seqstart = RExC_parse;
6532 bool has_use_defaults = FALSE;
6535 paren = *RExC_parse++;
6536 ret = NULL; /* For look-ahead/behind. */
6539 case 'P': /* (?P...) variants for those used to PCRE/Python */
6540 paren = *RExC_parse++;
6541 if ( paren == '<') /* (?P<...>) named capture */
6543 else if (paren == '>') { /* (?P>name) named recursion */
6544 goto named_recursion;
6546 else if (paren == '=') { /* (?P=...) named backref */
6547 /* this pretty much dupes the code for \k<NAME> in regatom(), if
6548 you change this make sure you change that */
6549 char* name_start = RExC_parse;
6551 SV *sv_dat = reg_scan_name(pRExC_state,
6552 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6553 if (RExC_parse == name_start || *RExC_parse != ')')
6554 vFAIL2("Sequence %.3s... not terminated",parse_start);
6557 num = add_data( pRExC_state, 1, "S" );
6558 RExC_rxi->data->data[num]=(void*)sv_dat;
6559 SvREFCNT_inc_simple_void(sv_dat);
6562 ret = reganode(pRExC_state,
6565 : (MORE_ASCII_RESTRICTED)
6567 : (AT_LEAST_UNI_SEMANTICS)
6575 Set_Node_Offset(ret, parse_start+1);
6576 Set_Node_Cur_Length(ret); /* MJD */
6578 nextchar(pRExC_state);
6582 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6584 case '<': /* (?<...) */
6585 if (*RExC_parse == '!')
6587 else if (*RExC_parse != '=')
6593 case '\'': /* (?'...') */
6594 name_start= RExC_parse;
6595 svname = reg_scan_name(pRExC_state,
6596 SIZE_ONLY ? /* reverse test from the others */
6597 REG_RSN_RETURN_NAME :
6598 REG_RSN_RETURN_NULL);
6599 if (RExC_parse == name_start) {
6601 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6604 if (*RExC_parse != paren)
6605 vFAIL2("Sequence (?%c... not terminated",
6606 paren=='>' ? '<' : paren);
6610 if (!svname) /* shouldn't happen */
6612 "panic: reg_scan_name returned NULL");
6613 if (!RExC_paren_names) {
6614 RExC_paren_names= newHV();
6615 sv_2mortal(MUTABLE_SV(RExC_paren_names));
6617 RExC_paren_name_list= newAV();
6618 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
6621 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
6623 sv_dat = HeVAL(he_str);
6625 /* croak baby croak */
6627 "panic: paren_name hash element allocation failed");
6628 } else if ( SvPOK(sv_dat) ) {
6629 /* (?|...) can mean we have dupes so scan to check
6630 its already been stored. Maybe a flag indicating
6631 we are inside such a construct would be useful,
6632 but the arrays are likely to be quite small, so
6633 for now we punt -- dmq */
6634 IV count = SvIV(sv_dat);
6635 I32 *pv = (I32*)SvPVX(sv_dat);
6637 for ( i = 0 ; i < count ; i++ ) {
6638 if ( pv[i] == RExC_npar ) {
6644 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
6645 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
6646 pv[count] = RExC_npar;
6647 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
6650 (void)SvUPGRADE(sv_dat,SVt_PVNV);
6651 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
6653 SvIV_set(sv_dat, 1);
6656 /* Yes this does cause a memory leak in debugging Perls */
6657 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
6658 SvREFCNT_dec(svname);
6661 /*sv_dump(sv_dat);*/
6663 nextchar(pRExC_state);
6665 goto capturing_parens;
6667 RExC_seen |= REG_SEEN_LOOKBEHIND;
6668 RExC_in_lookbehind++;
6670 case '=': /* (?=...) */
6671 RExC_seen_zerolen++;
6673 case '!': /* (?!...) */
6674 RExC_seen_zerolen++;
6675 if (*RExC_parse == ')') {
6676 ret=reg_node(pRExC_state, OPFAIL);
6677 nextchar(pRExC_state);
6681 case '|': /* (?|...) */
6682 /* branch reset, behave like a (?:...) except that
6683 buffers in alternations share the same numbers */
6685 after_freeze = freeze_paren = RExC_npar;
6687 case ':': /* (?:...) */
6688 case '>': /* (?>...) */
6690 case '$': /* (?$...) */
6691 case '@': /* (?@...) */
6692 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
6694 case '#': /* (?#...) */
6695 while (*RExC_parse && *RExC_parse != ')')
6697 if (*RExC_parse != ')')
6698 FAIL("Sequence (?#... not terminated");
6699 nextchar(pRExC_state);
6702 case '0' : /* (?0) */
6703 case 'R' : /* (?R) */
6704 if (*RExC_parse != ')')
6705 FAIL("Sequence (?R) not terminated");
6706 ret = reg_node(pRExC_state, GOSTART);
6707 *flagp |= POSTPONED;
6708 nextchar(pRExC_state);
6711 { /* named and numeric backreferences */
6713 case '&': /* (?&NAME) */
6714 parse_start = RExC_parse - 1;
6717 SV *sv_dat = reg_scan_name(pRExC_state,
6718 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6719 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6721 goto gen_recurse_regop;
6724 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6726 vFAIL("Illegal pattern");
6728 goto parse_recursion;
6730 case '-': /* (?-1) */
6731 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6732 RExC_parse--; /* rewind to let it be handled later */
6736 case '1': case '2': case '3': case '4': /* (?1) */
6737 case '5': case '6': case '7': case '8': case '9':
6740 num = atoi(RExC_parse);
6741 parse_start = RExC_parse - 1; /* MJD */
6742 if (*RExC_parse == '-')
6744 while (isDIGIT(*RExC_parse))
6746 if (*RExC_parse!=')')
6747 vFAIL("Expecting close bracket");
6750 if ( paren == '-' ) {
6752 Diagram of capture buffer numbering.
6753 Top line is the normal capture buffer numbers
6754 Bottom line is the negative indexing as from
6758 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
6762 num = RExC_npar + num;
6765 vFAIL("Reference to nonexistent group");
6767 } else if ( paren == '+' ) {
6768 num = RExC_npar + num - 1;
6771 ret = reganode(pRExC_state, GOSUB, num);
6773 if (num > (I32)RExC_rx->nparens) {
6775 vFAIL("Reference to nonexistent group");
6777 ARG2L_SET( ret, RExC_recurse_count++);
6779 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6780 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
6784 RExC_seen |= REG_SEEN_RECURSE;
6785 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
6786 Set_Node_Offset(ret, parse_start); /* MJD */
6788 *flagp |= POSTPONED;
6789 nextchar(pRExC_state);
6791 } /* named and numeric backreferences */
6794 case '?': /* (??...) */
6796 if (*RExC_parse != '{') {
6798 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6801 *flagp |= POSTPONED;
6802 paren = *RExC_parse++;
6804 case '{': /* (?{...}) */
6809 char *s = RExC_parse;
6811 RExC_seen_zerolen++;
6812 RExC_seen |= REG_SEEN_EVAL;
6813 while (count && (c = *RExC_parse)) {
6824 if (*RExC_parse != ')') {
6826 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
6830 OP_4tree *sop, *rop;
6831 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
6834 Perl_save_re_context(aTHX);
6835 rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad);
6836 sop->op_private |= OPpREFCOUNTED;
6837 /* re_dup will OpREFCNT_inc */
6838 OpREFCNT_set(sop, 1);
6841 n = add_data(pRExC_state, 3, "nop");
6842 RExC_rxi->data->data[n] = (void*)rop;
6843 RExC_rxi->data->data[n+1] = (void*)sop;
6844 RExC_rxi->data->data[n+2] = (void*)pad;
6847 else { /* First pass */
6848 if (PL_reginterp_cnt < ++RExC_seen_evals
6850 /* No compiled RE interpolated, has runtime
6851 components ===> unsafe. */
6852 FAIL("Eval-group not allowed at runtime, use re 'eval'");
6853 if (PL_tainting && PL_tainted)
6854 FAIL("Eval-group in insecure regular expression");
6855 #if PERL_VERSION > 8
6856 if (IN_PERL_COMPILETIME)
6861 nextchar(pRExC_state);
6863 ret = reg_node(pRExC_state, LOGICAL);
6866 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
6867 /* deal with the length of this later - MJD */
6870 ret = reganode(pRExC_state, EVAL, n);
6871 Set_Node_Length(ret, RExC_parse - parse_start + 1);
6872 Set_Node_Offset(ret, parse_start);
6875 case '(': /* (?(?{...})...) and (?(?=...)...) */
6878 if (RExC_parse[0] == '?') { /* (?(?...)) */
6879 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
6880 || RExC_parse[1] == '<'
6881 || RExC_parse[1] == '{') { /* Lookahead or eval. */
6884 ret = reg_node(pRExC_state, LOGICAL);
6887 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
6891 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
6892 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
6894 char ch = RExC_parse[0] == '<' ? '>' : '\'';
6895 char *name_start= RExC_parse++;
6897 SV *sv_dat=reg_scan_name(pRExC_state,
6898 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6899 if (RExC_parse == name_start || *RExC_parse != ch)
6900 vFAIL2("Sequence (?(%c... not terminated",
6901 (ch == '>' ? '<' : ch));
6904 num = add_data( pRExC_state, 1, "S" );
6905 RExC_rxi->data->data[num]=(void*)sv_dat;
6906 SvREFCNT_inc_simple_void(sv_dat);
6908 ret = reganode(pRExC_state,NGROUPP,num);
6909 goto insert_if_check_paren;
6911 else if (RExC_parse[0] == 'D' &&
6912 RExC_parse[1] == 'E' &&
6913 RExC_parse[2] == 'F' &&
6914 RExC_parse[3] == 'I' &&
6915 RExC_parse[4] == 'N' &&
6916 RExC_parse[5] == 'E')
6918 ret = reganode(pRExC_state,DEFINEP,0);
6921 goto insert_if_check_paren;
6923 else if (RExC_parse[0] == 'R') {
6926 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6927 parno = atoi(RExC_parse++);
6928 while (isDIGIT(*RExC_parse))
6930 } else if (RExC_parse[0] == '&') {
6933 sv_dat = reg_scan_name(pRExC_state,
6934 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6935 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6937 ret = reganode(pRExC_state,INSUBP,parno);
6938 goto insert_if_check_paren;
6940 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6943 parno = atoi(RExC_parse++);
6945 while (isDIGIT(*RExC_parse))
6947 ret = reganode(pRExC_state, GROUPP, parno);
6949 insert_if_check_paren:
6950 if ((c = *nextchar(pRExC_state)) != ')')
6951 vFAIL("Switch condition not recognized");
6953 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
6954 br = regbranch(pRExC_state, &flags, 1,depth+1);
6956 br = reganode(pRExC_state, LONGJMP, 0);
6958 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
6959 c = *nextchar(pRExC_state);
6964 vFAIL("(?(DEFINE)....) does not allow branches");
6965 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
6966 regbranch(pRExC_state, &flags, 1,depth+1);
6967 REGTAIL(pRExC_state, ret, lastbr);
6970 c = *nextchar(pRExC_state);
6975 vFAIL("Switch (?(condition)... contains too many branches");
6976 ender = reg_node(pRExC_state, TAIL);
6977 REGTAIL(pRExC_state, br, ender);
6979 REGTAIL(pRExC_state, lastbr, ender);
6980 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
6983 REGTAIL(pRExC_state, ret, ender);
6984 RExC_size++; /* XXX WHY do we need this?!!
6985 For large programs it seems to be required
6986 but I can't figure out why. -- dmq*/
6990 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
6994 RExC_parse--; /* for vFAIL to print correctly */
6995 vFAIL("Sequence (? incomplete");
6997 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
6999 has_use_defaults = TRUE;
7000 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
7001 set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics)
7002 ? REGEX_UNICODE_CHARSET
7003 : REGEX_DEPENDS_CHARSET);
7007 parse_flags: /* (?i) */
7009 U32 posflags = 0, negflags = 0;
7010 U32 *flagsp = &posflags;
7011 char has_charset_modifier = '\0';
7012 regex_charset cs = (RExC_utf8 || RExC_uni_semantics)
7013 ? REGEX_UNICODE_CHARSET
7014 : REGEX_DEPENDS_CHARSET;
7016 while (*RExC_parse) {
7017 /* && strchr("iogcmsx", *RExC_parse) */
7018 /* (?g), (?gc) and (?o) are useless here
7019 and must be globally applied -- japhy */
7020 switch (*RExC_parse) {
7021 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
7022 case LOCALE_PAT_MOD:
7023 if (has_charset_modifier) {
7024 goto excess_modifier;
7026 else if (flagsp == &negflags) {
7029 cs = REGEX_LOCALE_CHARSET;
7030 has_charset_modifier = LOCALE_PAT_MOD;
7031 RExC_contains_locale = 1;
7033 case UNICODE_PAT_MOD:
7034 if (has_charset_modifier) {
7035 goto excess_modifier;
7037 else if (flagsp == &negflags) {
7040 cs = REGEX_UNICODE_CHARSET;
7041 has_charset_modifier = UNICODE_PAT_MOD;
7043 case ASCII_RESTRICT_PAT_MOD:
7044 if (flagsp == &negflags) {
7047 if (has_charset_modifier) {
7048 if (cs != REGEX_ASCII_RESTRICTED_CHARSET) {
7049 goto excess_modifier;
7051 /* Doubled modifier implies more restricted */
7052 cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET;
7055 cs = REGEX_ASCII_RESTRICTED_CHARSET;
7057 has_charset_modifier = ASCII_RESTRICT_PAT_MOD;
7059 case DEPENDS_PAT_MOD:
7060 if (has_use_defaults) {
7061 goto fail_modifiers;
7063 else if (flagsp == &negflags) {
7066 else if (has_charset_modifier) {
7067 goto excess_modifier;
7070 /* The dual charset means unicode semantics if the
7071 * pattern (or target, not known until runtime) are
7072 * utf8, or something in the pattern indicates unicode
7074 cs = (RExC_utf8 || RExC_uni_semantics)
7075 ? REGEX_UNICODE_CHARSET
7076 : REGEX_DEPENDS_CHARSET;
7077 has_charset_modifier = DEPENDS_PAT_MOD;
7081 if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) {
7082 vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD);
7084 else if (has_charset_modifier == *(RExC_parse - 1)) {
7085 vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1));
7088 vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1));
7093 vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1));
7095 case ONCE_PAT_MOD: /* 'o' */
7096 case GLOBAL_PAT_MOD: /* 'g' */
7097 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7098 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
7099 if (! (wastedflags & wflagbit) ) {
7100 wastedflags |= wflagbit;
7103 "Useless (%s%c) - %suse /%c modifier",
7104 flagsp == &negflags ? "?-" : "?",
7106 flagsp == &negflags ? "don't " : "",
7113 case CONTINUE_PAT_MOD: /* 'c' */
7114 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7115 if (! (wastedflags & WASTED_C) ) {
7116 wastedflags |= WASTED_GC;
7119 "Useless (%sc) - %suse /gc modifier",
7120 flagsp == &negflags ? "?-" : "?",
7121 flagsp == &negflags ? "don't " : ""
7126 case KEEPCOPY_PAT_MOD: /* 'p' */
7127 if (flagsp == &negflags) {
7129 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
7131 *flagsp |= RXf_PMf_KEEPCOPY;
7135 /* A flag is a default iff it is following a minus, so
7136 * if there is a minus, it means will be trying to
7137 * re-specify a default which is an error */
7138 if (has_use_defaults || flagsp == &negflags) {
7141 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7145 wastedflags = 0; /* reset so (?g-c) warns twice */
7151 RExC_flags |= posflags;
7152 RExC_flags &= ~negflags;
7153 set_regex_charset(&RExC_flags, cs);
7155 oregflags |= posflags;
7156 oregflags &= ~negflags;
7157 set_regex_charset(&oregflags, cs);
7159 nextchar(pRExC_state);
7170 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7175 }} /* one for the default block, one for the switch */
7182 ret = reganode(pRExC_state, OPEN, parno);
7185 RExC_nestroot = parno;
7186 if (RExC_seen & REG_SEEN_RECURSE
7187 && !RExC_open_parens[parno-1])
7189 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7190 "Setting open paren #%"IVdf" to %d\n",
7191 (IV)parno, REG_NODE_NUM(ret)));
7192 RExC_open_parens[parno-1]= ret;
7195 Set_Node_Length(ret, 1); /* MJD */
7196 Set_Node_Offset(ret, RExC_parse); /* MJD */
7204 /* Pick up the branches, linking them together. */
7205 parse_start = RExC_parse; /* MJD */
7206 br = regbranch(pRExC_state, &flags, 1,depth+1);
7208 /* branch_len = (paren != 0); */
7212 if (*RExC_parse == '|') {
7213 if (!SIZE_ONLY && RExC_extralen) {
7214 reginsert(pRExC_state, BRANCHJ, br, depth+1);
7217 reginsert(pRExC_state, BRANCH, br, depth+1);
7218 Set_Node_Length(br, paren != 0);
7219 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
7223 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
7225 else if (paren == ':') {
7226 *flagp |= flags&SIMPLE;
7228 if (is_open) { /* Starts with OPEN. */
7229 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
7231 else if (paren != '?') /* Not Conditional */
7233 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7235 while (*RExC_parse == '|') {
7236 if (!SIZE_ONLY && RExC_extralen) {
7237 ender = reganode(pRExC_state, LONGJMP,0);
7238 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
7241 RExC_extralen += 2; /* Account for LONGJMP. */
7242 nextchar(pRExC_state);
7244 if (RExC_npar > after_freeze)
7245 after_freeze = RExC_npar;
7246 RExC_npar = freeze_paren;
7248 br = regbranch(pRExC_state, &flags, 0, depth+1);
7252 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
7254 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7257 if (have_branch || paren != ':') {
7258 /* Make a closing node, and hook it on the end. */
7261 ender = reg_node(pRExC_state, TAIL);
7264 ender = reganode(pRExC_state, CLOSE, parno);
7265 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
7266 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7267 "Setting close paren #%"IVdf" to %d\n",
7268 (IV)parno, REG_NODE_NUM(ender)));
7269 RExC_close_parens[parno-1]= ender;
7270 if (RExC_nestroot == parno)
7273 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
7274 Set_Node_Length(ender,1); /* MJD */
7280 *flagp &= ~HASWIDTH;
7283 ender = reg_node(pRExC_state, SUCCEED);
7286 ender = reg_node(pRExC_state, END);
7288 assert(!RExC_opend); /* there can only be one! */
7293 REGTAIL(pRExC_state, lastbr, ender);
7295 if (have_branch && !SIZE_ONLY) {
7297 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
7299 /* Hook the tails of the branches to the closing node. */
7300 for (br = ret; br; br = regnext(br)) {
7301 const U8 op = PL_regkind[OP(br)];
7303 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
7305 else if (op == BRANCHJ) {
7306 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
7314 static const char parens[] = "=!<,>";
7316 if (paren && (p = strchr(parens, paren))) {
7317 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
7318 int flag = (p - parens) > 1;
7321 node = SUSPEND, flag = 0;
7322 reginsert(pRExC_state, node,ret, depth+1);
7323 Set_Node_Cur_Length(ret);
7324 Set_Node_Offset(ret, parse_start + 1);
7326 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
7330 /* Check for proper termination. */
7332 RExC_flags = oregflags;
7333 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
7334 RExC_parse = oregcomp_parse;
7335 vFAIL("Unmatched (");
7338 else if (!paren && RExC_parse < RExC_end) {
7339 if (*RExC_parse == ')') {
7341 vFAIL("Unmatched )");
7344 FAIL("Junk on end of regexp"); /* "Can't happen". */
7348 if (RExC_in_lookbehind) {
7349 RExC_in_lookbehind--;
7351 if (after_freeze > RExC_npar)
7352 RExC_npar = after_freeze;
7357 - regbranch - one alternative of an | operator
7359 * Implements the concatenation operator.
7362 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
7365 register regnode *ret;
7366 register regnode *chain = NULL;
7367 register regnode *latest;
7368 I32 flags = 0, c = 0;
7369 GET_RE_DEBUG_FLAGS_DECL;
7371 PERL_ARGS_ASSERT_REGBRANCH;
7373 DEBUG_PARSE("brnc");
7378 if (!SIZE_ONLY && RExC_extralen)
7379 ret = reganode(pRExC_state, BRANCHJ,0);
7381 ret = reg_node(pRExC_state, BRANCH);
7382 Set_Node_Length(ret, 1);
7386 if (!first && SIZE_ONLY)
7387 RExC_extralen += 1; /* BRANCHJ */
7389 *flagp = WORST; /* Tentatively. */
7392 nextchar(pRExC_state);
7393 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
7395 latest = regpiece(pRExC_state, &flags,depth+1);
7396 if (latest == NULL) {
7397 if (flags & TRYAGAIN)
7401 else if (ret == NULL)
7403 *flagp |= flags&(HASWIDTH|POSTPONED);
7404 if (chain == NULL) /* First piece. */
7405 *flagp |= flags&SPSTART;
7408 REGTAIL(pRExC_state, chain, latest);
7413 if (chain == NULL) { /* Loop ran zero times. */
7414 chain = reg_node(pRExC_state, NOTHING);
7419 *flagp |= flags&SIMPLE;
7426 - regpiece - something followed by possible [*+?]
7428 * Note that the branching code sequences used for ? and the general cases
7429 * of * and + are somewhat optimized: they use the same NOTHING node as
7430 * both the endmarker for their branch list and the body of the last branch.
7431 * It might seem that this node could be dispensed with entirely, but the
7432 * endmarker role is not redundant.
7435 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7438 register regnode *ret;
7440 register char *next;
7442 const char * const origparse = RExC_parse;
7444 I32 max = REG_INFTY;
7445 #ifdef RE_TRACK_PATTERN_OFFSETS
7448 const char *maxpos = NULL;
7449 GET_RE_DEBUG_FLAGS_DECL;
7451 PERL_ARGS_ASSERT_REGPIECE;
7453 DEBUG_PARSE("piec");
7455 ret = regatom(pRExC_state, &flags,depth+1);
7457 if (flags & TRYAGAIN)
7464 if (op == '{' && regcurly(RExC_parse)) {
7466 #ifdef RE_TRACK_PATTERN_OFFSETS
7467 parse_start = RExC_parse; /* MJD */
7469 next = RExC_parse + 1;
7470 while (isDIGIT(*next) || *next == ',') {
7479 if (*next == '}') { /* got one */
7483 min = atoi(RExC_parse);
7487 maxpos = RExC_parse;
7489 if (!max && *maxpos != '0')
7490 max = REG_INFTY; /* meaning "infinity" */
7491 else if (max >= REG_INFTY)
7492 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
7494 nextchar(pRExC_state);
7497 if ((flags&SIMPLE)) {
7498 RExC_naughty += 2 + RExC_naughty / 2;
7499 reginsert(pRExC_state, CURLY, ret, depth+1);
7500 Set_Node_Offset(ret, parse_start+1); /* MJD */
7501 Set_Node_Cur_Length(ret);
7504 regnode * const w = reg_node(pRExC_state, WHILEM);
7507 REGTAIL(pRExC_state, ret, w);
7508 if (!SIZE_ONLY && RExC_extralen) {
7509 reginsert(pRExC_state, LONGJMP,ret, depth+1);
7510 reginsert(pRExC_state, NOTHING,ret, depth+1);
7511 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
7513 reginsert(pRExC_state, CURLYX,ret, depth+1);
7515 Set_Node_Offset(ret, parse_start+1);
7516 Set_Node_Length(ret,
7517 op == '{' ? (RExC_parse - parse_start) : 1);
7519 if (!SIZE_ONLY && RExC_extralen)
7520 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
7521 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
7523 RExC_whilem_seen++, RExC_extralen += 3;
7524 RExC_naughty += 4 + RExC_naughty; /* compound interest */
7533 vFAIL("Can't do {n,m} with n > m");
7535 ARG1_SET(ret, (U16)min);
7536 ARG2_SET(ret, (U16)max);
7548 #if 0 /* Now runtime fix should be reliable. */
7550 /* if this is reinstated, don't forget to put this back into perldiag:
7552 =item Regexp *+ operand could be empty at {#} in regex m/%s/
7554 (F) The part of the regexp subject to either the * or + quantifier
7555 could match an empty string. The {#} shows in the regular
7556 expression about where the problem was discovered.
7560 if (!(flags&HASWIDTH) && op != '?')
7561 vFAIL("Regexp *+ operand could be empty");
7564 #ifdef RE_TRACK_PATTERN_OFFSETS
7565 parse_start = RExC_parse;
7567 nextchar(pRExC_state);
7569 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
7571 if (op == '*' && (flags&SIMPLE)) {
7572 reginsert(pRExC_state, STAR, ret, depth+1);
7576 else if (op == '*') {
7580 else if (op == '+' && (flags&SIMPLE)) {
7581 reginsert(pRExC_state, PLUS, ret, depth+1);
7585 else if (op == '+') {
7589 else if (op == '?') {
7594 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
7595 ckWARN3reg(RExC_parse,
7596 "%.*s matches null string many times",
7597 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
7601 if (RExC_parse < RExC_end && *RExC_parse == '?') {
7602 nextchar(pRExC_state);
7603 reginsert(pRExC_state, MINMOD, ret, depth+1);
7604 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
7606 #ifndef REG_ALLOW_MINMOD_SUSPEND
7609 if (RExC_parse < RExC_end && *RExC_parse == '+') {
7611 nextchar(pRExC_state);
7612 ender = reg_node(pRExC_state, SUCCEED);
7613 REGTAIL(pRExC_state, ret, ender);
7614 reginsert(pRExC_state, SUSPEND, ret, depth+1);
7616 ender = reg_node(pRExC_state, TAIL);
7617 REGTAIL(pRExC_state, ret, ender);
7621 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
7623 vFAIL("Nested quantifiers");
7630 /* reg_namedseq(pRExC_state,UVp, UV depth)
7632 This is expected to be called by a parser routine that has
7633 recognized '\N' and needs to handle the rest. RExC_parse is
7634 expected to point at the first char following the N at the time
7637 The \N may be inside (indicated by valuep not being NULL) or outside a
7640 \N may begin either a named sequence, or if outside a character class, mean
7641 to match a non-newline. For non single-quoted regexes, the tokenizer has
7642 attempted to decide which, and in the case of a named sequence converted it
7643 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
7644 where c1... are the characters in the sequence. For single-quoted regexes,
7645 the tokenizer passes the \N sequence through unchanged; this code will not
7646 attempt to determine this nor expand those. The net effect is that if the
7647 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
7648 signals that this \N occurrence means to match a non-newline.
7650 Only the \N{U+...} form should occur in a character class, for the same
7651 reason that '.' inside a character class means to just match a period: it
7652 just doesn't make sense.
7654 If valuep is non-null then it is assumed that we are parsing inside
7655 of a charclass definition and the first codepoint in the resolved
7656 string is returned via *valuep and the routine will return NULL.
7657 In this mode if a multichar string is returned from the charnames
7658 handler, a warning will be issued, and only the first char in the
7659 sequence will be examined. If the string returned is zero length
7660 then the value of *valuep is undefined and NON-NULL will
7661 be returned to indicate failure. (This will NOT be a valid pointer
7664 If valuep is null then it is assumed that we are parsing normal text and a
7665 new EXACT node is inserted into the program containing the resolved string,
7666 and a pointer to the new node is returned. But if the string is zero length
7667 a NOTHING node is emitted instead.
7669 On success RExC_parse is set to the char following the endbrace.
7670 Parsing failures will generate a fatal error via vFAIL(...)
7673 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp, U32 depth)
7675 char * endbrace; /* '}' following the name */
7676 regnode *ret = NULL;
7679 GET_RE_DEBUG_FLAGS_DECL;
7681 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
7685 /* The [^\n] meaning of \N ignores spaces and comments under the /x
7686 * modifier. The other meaning does not */
7687 p = (RExC_flags & RXf_PMf_EXTENDED)
7688 ? regwhite( pRExC_state, RExC_parse )
7691 /* Disambiguate between \N meaning a named character versus \N meaning
7692 * [^\n]. The former is assumed when it can't be the latter. */
7693 if (*p != '{' || regcurly(p)) {
7696 /* no bare \N in a charclass */
7697 vFAIL("\\N in a character class must be a named character: \\N{...}");
7699 nextchar(pRExC_state);
7700 ret = reg_node(pRExC_state, REG_ANY);
7701 *flagp |= HASWIDTH|SIMPLE;
7704 Set_Node_Length(ret, 1); /* MJD */
7708 /* Here, we have decided it should be a named sequence */
7710 /* The test above made sure that the next real character is a '{', but
7711 * under the /x modifier, it could be separated by space (or a comment and
7712 * \n) and this is not allowed (for consistency with \x{...} and the
7713 * tokenizer handling of \N{NAME}). */
7714 if (*RExC_parse != '{') {
7715 vFAIL("Missing braces on \\N{}");
7718 RExC_parse++; /* Skip past the '{' */
7720 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
7721 || ! (endbrace == RExC_parse /* nothing between the {} */
7722 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
7723 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
7725 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
7726 vFAIL("\\N{NAME} must be resolved by the lexer");
7729 if (endbrace == RExC_parse) { /* empty: \N{} */
7731 RExC_parse = endbrace + 1;
7732 return reg_node(pRExC_state,NOTHING);
7736 ckWARNreg(RExC_parse,
7737 "Ignoring zero length \\N{} in character class"
7739 RExC_parse = endbrace + 1;
7742 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
7745 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
7746 RExC_parse += 2; /* Skip past the 'U+' */
7748 if (valuep) { /* In a bracketed char class */
7749 /* We only pay attention to the first char of
7750 multichar strings being returned. I kinda wonder
7751 if this makes sense as it does change the behaviour
7752 from earlier versions, OTOH that behaviour was broken
7753 as well. XXX Solution is to recharacterize as
7754 [rest-of-class]|multi1|multi2... */
7756 STRLEN length_of_hex;
7757 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7758 | PERL_SCAN_DISALLOW_PREFIX
7759 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
7761 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
7762 if (endchar < endbrace) {
7763 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
7766 length_of_hex = (STRLEN)(endchar - RExC_parse);
7767 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
7769 /* The tokenizer should have guaranteed validity, but it's possible to
7770 * bypass it by using single quoting, so check */
7771 if (length_of_hex == 0
7772 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
7774 RExC_parse += length_of_hex; /* Includes all the valid */
7775 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
7776 ? UTF8SKIP(RExC_parse)
7778 /* Guard against malformed utf8 */
7779 if (RExC_parse >= endchar) RExC_parse = endchar;
7780 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7783 RExC_parse = endbrace + 1;
7784 if (endchar == endbrace) return NULL;
7786 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
7788 else { /* Not a char class */
7790 /* What is done here is to convert this to a sub-pattern of the form
7791 * (?:\x{char1}\x{char2}...)
7792 * and then call reg recursively. That way, it retains its atomicness,
7793 * while not having to worry about special handling that some code
7794 * points may have. toke.c has converted the original Unicode values
7795 * to native, so that we can just pass on the hex values unchanged. We
7796 * do have to set a flag to keep recoding from happening in the
7799 SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP);
7801 char *endchar; /* Points to '.' or '}' ending cur char in the input
7803 char *orig_end = RExC_end;
7805 while (RExC_parse < endbrace) {
7807 /* Code points are separated by dots. If none, there is only one
7808 * code point, and is terminated by the brace */
7809 endchar = RExC_parse + strcspn(RExC_parse, ".}");
7811 /* Convert to notation the rest of the code understands */
7812 sv_catpv(substitute_parse, "\\x{");
7813 sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse);
7814 sv_catpv(substitute_parse, "}");
7816 /* Point to the beginning of the next character in the sequence. */
7817 RExC_parse = endchar + 1;
7819 sv_catpv(substitute_parse, ")");
7821 RExC_parse = SvPV(substitute_parse, len);
7823 /* Don't allow empty number */
7825 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7827 RExC_end = RExC_parse + len;
7829 /* The values are Unicode, and therefore not subject to recoding */
7830 RExC_override_recoding = 1;
7832 ret = reg(pRExC_state, 1, flagp, depth+1);
7834 RExC_parse = endbrace;
7835 RExC_end = orig_end;
7836 RExC_override_recoding = 0;
7838 nextchar(pRExC_state);
7848 * It returns the code point in utf8 for the value in *encp.
7849 * value: a code value in the source encoding
7850 * encp: a pointer to an Encode object
7852 * If the result from Encode is not a single character,
7853 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
7856 S_reg_recode(pTHX_ const char value, SV **encp)
7859 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
7860 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
7861 const STRLEN newlen = SvCUR(sv);
7862 UV uv = UNICODE_REPLACEMENT;
7864 PERL_ARGS_ASSERT_REG_RECODE;
7868 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
7871 if (!newlen || numlen != newlen) {
7872 uv = UNICODE_REPLACEMENT;
7880 - regatom - the lowest level
7882 Try to identify anything special at the start of the pattern. If there
7883 is, then handle it as required. This may involve generating a single regop,
7884 such as for an assertion; or it may involve recursing, such as to
7885 handle a () structure.
7887 If the string doesn't start with something special then we gobble up
7888 as much literal text as we can.
7890 Once we have been able to handle whatever type of thing started the
7891 sequence, we return.
7893 Note: we have to be careful with escapes, as they can be both literal
7894 and special, and in the case of \10 and friends can either, depending
7895 on context. Specifically there are two separate switches for handling
7896 escape sequences, with the one for handling literal escapes requiring
7897 a dummy entry for all of the special escapes that are actually handled
7902 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7905 register regnode *ret = NULL;
7907 char *parse_start = RExC_parse;
7909 GET_RE_DEBUG_FLAGS_DECL;
7910 DEBUG_PARSE("atom");
7911 *flagp = WORST; /* Tentatively. */
7913 PERL_ARGS_ASSERT_REGATOM;
7916 switch ((U8)*RExC_parse) {
7918 RExC_seen_zerolen++;
7919 nextchar(pRExC_state);
7920 if (RExC_flags & RXf_PMf_MULTILINE)
7921 ret = reg_node(pRExC_state, MBOL);
7922 else if (RExC_flags & RXf_PMf_SINGLELINE)
7923 ret = reg_node(pRExC_state, SBOL);
7925 ret = reg_node(pRExC_state, BOL);
7926 Set_Node_Length(ret, 1); /* MJD */
7929 nextchar(pRExC_state);
7931 RExC_seen_zerolen++;
7932 if (RExC_flags & RXf_PMf_MULTILINE)
7933 ret = reg_node(pRExC_state, MEOL);
7934 else if (RExC_flags & RXf_PMf_SINGLELINE)
7935 ret = reg_node(pRExC_state, SEOL);
7937 ret = reg_node(pRExC_state, EOL);
7938 Set_Node_Length(ret, 1); /* MJD */
7941 nextchar(pRExC_state);
7942 if (RExC_flags & RXf_PMf_SINGLELINE)
7943 ret = reg_node(pRExC_state, SANY);
7945 ret = reg_node(pRExC_state, REG_ANY);
7946 *flagp |= HASWIDTH|SIMPLE;
7948 Set_Node_Length(ret, 1); /* MJD */
7952 char * const oregcomp_parse = ++RExC_parse;
7953 ret = regclass(pRExC_state,depth+1);
7954 if (*RExC_parse != ']') {
7955 RExC_parse = oregcomp_parse;
7956 vFAIL("Unmatched [");
7958 nextchar(pRExC_state);
7959 *flagp |= HASWIDTH|SIMPLE;
7960 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
7964 nextchar(pRExC_state);
7965 ret = reg(pRExC_state, 1, &flags,depth+1);
7967 if (flags & TRYAGAIN) {
7968 if (RExC_parse == RExC_end) {
7969 /* Make parent create an empty node if needed. */
7977 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
7981 if (flags & TRYAGAIN) {
7985 vFAIL("Internal urp");
7986 /* Supposed to be caught earlier. */
7989 if (!regcurly(RExC_parse)) {
7998 vFAIL("Quantifier follows nothing");
8003 This switch handles escape sequences that resolve to some kind
8004 of special regop and not to literal text. Escape sequnces that
8005 resolve to literal text are handled below in the switch marked
8008 Every entry in this switch *must* have a corresponding entry
8009 in the literal escape switch. However, the opposite is not
8010 required, as the default for this switch is to jump to the
8011 literal text handling code.
8013 switch ((U8)*++RExC_parse) {
8014 /* Special Escapes */
8016 RExC_seen_zerolen++;
8017 ret = reg_node(pRExC_state, SBOL);
8019 goto finish_meta_pat;
8021 ret = reg_node(pRExC_state, GPOS);
8022 RExC_seen |= REG_SEEN_GPOS;
8024 goto finish_meta_pat;
8026 RExC_seen_zerolen++;
8027 ret = reg_node(pRExC_state, KEEPS);
8029 /* XXX:dmq : disabling in-place substitution seems to
8030 * be necessary here to avoid cases of memory corruption, as
8031 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
8033 RExC_seen |= REG_SEEN_LOOKBEHIND;
8034 goto finish_meta_pat;
8036 ret = reg_node(pRExC_state, SEOL);
8038 RExC_seen_zerolen++; /* Do not optimize RE away */
8039 goto finish_meta_pat;
8041 ret = reg_node(pRExC_state, EOS);
8043 RExC_seen_zerolen++; /* Do not optimize RE away */
8044 goto finish_meta_pat;
8046 ret = reg_node(pRExC_state, CANY);
8047 RExC_seen |= REG_SEEN_CANY;
8048 *flagp |= HASWIDTH|SIMPLE;
8049 goto finish_meta_pat;
8051 ret = reg_node(pRExC_state, CLUMP);
8053 goto finish_meta_pat;
8055 switch (get_regex_charset(RExC_flags)) {
8056 case REGEX_LOCALE_CHARSET:
8059 case REGEX_UNICODE_CHARSET:
8062 case REGEX_ASCII_RESTRICTED_CHARSET:
8063 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8066 case REGEX_DEPENDS_CHARSET:
8072 ret = reg_node(pRExC_state, op);
8073 *flagp |= HASWIDTH|SIMPLE;
8074 goto finish_meta_pat;
8076 switch (get_regex_charset(RExC_flags)) {
8077 case REGEX_LOCALE_CHARSET:
8080 case REGEX_UNICODE_CHARSET:
8083 case REGEX_ASCII_RESTRICTED_CHARSET:
8084 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8087 case REGEX_DEPENDS_CHARSET:
8093 ret = reg_node(pRExC_state, op);
8094 *flagp |= HASWIDTH|SIMPLE;
8095 goto finish_meta_pat;
8097 RExC_seen_zerolen++;
8098 RExC_seen |= REG_SEEN_LOOKBEHIND;
8099 switch (get_regex_charset(RExC_flags)) {
8100 case REGEX_LOCALE_CHARSET:
8103 case REGEX_UNICODE_CHARSET:
8106 case REGEX_ASCII_RESTRICTED_CHARSET:
8107 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8110 case REGEX_DEPENDS_CHARSET:
8116 ret = reg_node(pRExC_state, op);
8117 FLAGS(ret) = get_regex_charset(RExC_flags);
8119 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8120 ckWARNregdep(RExC_parse, "\"\\b{\" is deprecated; use \"\\b\\{\" instead");
8122 goto finish_meta_pat;
8124 RExC_seen_zerolen++;
8125 RExC_seen |= REG_SEEN_LOOKBEHIND;
8126 switch (get_regex_charset(RExC_flags)) {
8127 case REGEX_LOCALE_CHARSET:
8130 case REGEX_UNICODE_CHARSET:
8133 case REGEX_ASCII_RESTRICTED_CHARSET:
8134 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8137 case REGEX_DEPENDS_CHARSET:
8143 ret = reg_node(pRExC_state, op);
8144 FLAGS(ret) = get_regex_charset(RExC_flags);
8146 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8147 ckWARNregdep(RExC_parse, "\"\\B{\" is deprecated; use \"\\B\\{\" instead");
8149 goto finish_meta_pat;
8151 switch (get_regex_charset(RExC_flags)) {
8152 case REGEX_LOCALE_CHARSET:
8155 case REGEX_UNICODE_CHARSET:
8158 case REGEX_ASCII_RESTRICTED_CHARSET:
8159 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8162 case REGEX_DEPENDS_CHARSET:
8168 ret = reg_node(pRExC_state, op);
8169 *flagp |= HASWIDTH|SIMPLE;
8170 goto finish_meta_pat;
8172 switch (get_regex_charset(RExC_flags)) {
8173 case REGEX_LOCALE_CHARSET:
8176 case REGEX_UNICODE_CHARSET:
8179 case REGEX_ASCII_RESTRICTED_CHARSET:
8180 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8183 case REGEX_DEPENDS_CHARSET:
8189 ret = reg_node(pRExC_state, op);
8190 *flagp |= HASWIDTH|SIMPLE;
8191 goto finish_meta_pat;
8193 switch (get_regex_charset(RExC_flags)) {
8194 case REGEX_LOCALE_CHARSET:
8197 case REGEX_ASCII_RESTRICTED_CHARSET:
8198 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8201 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8202 case REGEX_UNICODE_CHARSET:
8208 ret = reg_node(pRExC_state, op);
8209 *flagp |= HASWIDTH|SIMPLE;
8210 goto finish_meta_pat;
8212 switch (get_regex_charset(RExC_flags)) {
8213 case REGEX_LOCALE_CHARSET:
8216 case REGEX_ASCII_RESTRICTED_CHARSET:
8217 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8220 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8221 case REGEX_UNICODE_CHARSET:
8227 ret = reg_node(pRExC_state, op);
8228 *flagp |= HASWIDTH|SIMPLE;
8229 goto finish_meta_pat;
8231 ret = reg_node(pRExC_state, LNBREAK);
8232 *flagp |= HASWIDTH|SIMPLE;
8233 goto finish_meta_pat;
8235 ret = reg_node(pRExC_state, HORIZWS);
8236 *flagp |= HASWIDTH|SIMPLE;
8237 goto finish_meta_pat;
8239 ret = reg_node(pRExC_state, NHORIZWS);
8240 *flagp |= HASWIDTH|SIMPLE;
8241 goto finish_meta_pat;
8243 ret = reg_node(pRExC_state, VERTWS);
8244 *flagp |= HASWIDTH|SIMPLE;
8245 goto finish_meta_pat;
8247 ret = reg_node(pRExC_state, NVERTWS);
8248 *flagp |= HASWIDTH|SIMPLE;
8250 nextchar(pRExC_state);
8251 Set_Node_Length(ret, 2); /* MJD */
8256 char* const oldregxend = RExC_end;
8258 char* parse_start = RExC_parse - 2;
8261 if (RExC_parse[1] == '{') {
8262 /* a lovely hack--pretend we saw [\pX] instead */
8263 RExC_end = strchr(RExC_parse, '}');
8265 const U8 c = (U8)*RExC_parse;
8267 RExC_end = oldregxend;
8268 vFAIL2("Missing right brace on \\%c{}", c);
8273 RExC_end = RExC_parse + 2;
8274 if (RExC_end > oldregxend)
8275 RExC_end = oldregxend;
8279 ret = regclass(pRExC_state,depth+1);
8281 RExC_end = oldregxend;
8284 Set_Node_Offset(ret, parse_start + 2);
8285 Set_Node_Cur_Length(ret);
8286 nextchar(pRExC_state);
8287 *flagp |= HASWIDTH|SIMPLE;
8291 /* Handle \N and \N{NAME} here and not below because it can be
8292 multicharacter. join_exact() will join them up later on.
8293 Also this makes sure that things like /\N{BLAH}+/ and
8294 \N{BLAH} being multi char Just Happen. dmq*/
8296 ret= reg_namedseq(pRExC_state, NULL, flagp, depth);
8298 case 'k': /* Handle \k<NAME> and \k'NAME' */
8301 char ch= RExC_parse[1];
8302 if (ch != '<' && ch != '\'' && ch != '{') {
8304 vFAIL2("Sequence %.2s... not terminated",parse_start);
8306 /* this pretty much dupes the code for (?P=...) in reg(), if
8307 you change this make sure you change that */
8308 char* name_start = (RExC_parse += 2);
8310 SV *sv_dat = reg_scan_name(pRExC_state,
8311 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
8312 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
8313 if (RExC_parse == name_start || *RExC_parse != ch)
8314 vFAIL2("Sequence %.3s... not terminated",parse_start);
8317 num = add_data( pRExC_state, 1, "S" );
8318 RExC_rxi->data->data[num]=(void*)sv_dat;
8319 SvREFCNT_inc_simple_void(sv_dat);
8323 ret = reganode(pRExC_state,
8326 : (MORE_ASCII_RESTRICTED)
8328 : (AT_LEAST_UNI_SEMANTICS)
8336 /* override incorrect value set in reganode MJD */
8337 Set_Node_Offset(ret, parse_start+1);
8338 Set_Node_Cur_Length(ret); /* MJD */
8339 nextchar(pRExC_state);
8345 case '1': case '2': case '3': case '4':
8346 case '5': case '6': case '7': case '8': case '9':
8349 bool isg = *RExC_parse == 'g';
8354 if (*RExC_parse == '{') {
8358 if (*RExC_parse == '-') {
8362 if (hasbrace && !isDIGIT(*RExC_parse)) {
8363 if (isrel) RExC_parse--;
8365 goto parse_named_seq;
8367 num = atoi(RExC_parse);
8368 if (isg && num == 0)
8369 vFAIL("Reference to invalid group 0");
8371 num = RExC_npar - num;
8373 vFAIL("Reference to nonexistent or unclosed group");
8375 if (!isg && num > 9 && num >= RExC_npar)
8378 char * const parse_start = RExC_parse - 1; /* MJD */
8379 while (isDIGIT(*RExC_parse))
8381 if (parse_start == RExC_parse - 1)
8382 vFAIL("Unterminated \\g... pattern");
8384 if (*RExC_parse != '}')
8385 vFAIL("Unterminated \\g{...} pattern");
8389 if (num > (I32)RExC_rx->nparens)
8390 vFAIL("Reference to nonexistent group");
8393 ret = reganode(pRExC_state,
8396 : (MORE_ASCII_RESTRICTED)
8398 : (AT_LEAST_UNI_SEMANTICS)
8406 /* override incorrect value set in reganode MJD */
8407 Set_Node_Offset(ret, parse_start+1);
8408 Set_Node_Cur_Length(ret); /* MJD */
8410 nextchar(pRExC_state);
8415 if (RExC_parse >= RExC_end)
8416 FAIL("Trailing \\");
8419 /* Do not generate "unrecognized" warnings here, we fall
8420 back into the quick-grab loop below */
8427 if (RExC_flags & RXf_PMf_EXTENDED) {
8428 if ( reg_skipcomment( pRExC_state ) )
8435 parse_start = RExC_parse - 1;
8448 char_state latest_char_state = generic_char;
8449 register STRLEN len;
8454 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
8455 regnode * orig_emit;
8458 orig_emit = RExC_emit; /* Save the original output node position in
8459 case we need to output a different node
8461 ret = reg_node(pRExC_state,
8462 (U8) ((! FOLD) ? EXACT
8465 : (MORE_ASCII_RESTRICTED)
8467 : (AT_LEAST_UNI_SEMANTICS)
8472 for (len = 0, p = RExC_parse - 1;
8473 len < 127 && p < RExC_end;
8476 char * const oldp = p;
8478 if (RExC_flags & RXf_PMf_EXTENDED)
8479 p = regwhite( pRExC_state, p );
8490 /* Literal Escapes Switch
8492 This switch is meant to handle escape sequences that
8493 resolve to a literal character.
8495 Every escape sequence that represents something
8496 else, like an assertion or a char class, is handled
8497 in the switch marked 'Special Escapes' above in this
8498 routine, but also has an entry here as anything that
8499 isn't explicitly mentioned here will be treated as
8500 an unescaped equivalent literal.
8504 /* These are all the special escapes. */
8505 case 'A': /* Start assertion */
8506 case 'b': case 'B': /* Word-boundary assertion*/
8507 case 'C': /* Single char !DANGEROUS! */
8508 case 'd': case 'D': /* digit class */
8509 case 'g': case 'G': /* generic-backref, pos assertion */
8510 case 'h': case 'H': /* HORIZWS */
8511 case 'k': case 'K': /* named backref, keep marker */
8512 case 'N': /* named char sequence */
8513 case 'p': case 'P': /* Unicode property */
8514 case 'R': /* LNBREAK */
8515 case 's': case 'S': /* space class */
8516 case 'v': case 'V': /* VERTWS */
8517 case 'w': case 'W': /* word class */
8518 case 'X': /* eXtended Unicode "combining character sequence" */
8519 case 'z': case 'Z': /* End of line/string assertion */
8523 /* Anything after here is an escape that resolves to a
8524 literal. (Except digits, which may or may not)
8543 ender = ASCII_TO_NATIVE('\033');
8547 ender = ASCII_TO_NATIVE('\007');
8552 STRLEN brace_len = len;
8554 const char* error_msg;
8556 bool valid = grok_bslash_o(p,
8563 RExC_parse = p; /* going to die anyway; point
8564 to exact spot of failure */
8571 if (PL_encoding && ender < 0x100) {
8572 goto recode_encoding;
8581 char* const e = strchr(p, '}');
8585 vFAIL("Missing right brace on \\x{}");
8588 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8589 | PERL_SCAN_DISALLOW_PREFIX;
8590 STRLEN numlen = e - p - 1;
8591 ender = grok_hex(p + 1, &numlen, &flags, NULL);
8598 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8600 ender = grok_hex(p, &numlen, &flags, NULL);
8603 if (PL_encoding && ender < 0x100)
8604 goto recode_encoding;
8608 ender = grok_bslash_c(*p++, UTF, SIZE_ONLY);
8610 case '0': case '1': case '2': case '3':case '4':
8611 case '5': case '6': case '7': case '8':case '9':
8613 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
8615 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8617 ender = grok_oct(p, &numlen, &flags, NULL);
8627 if (PL_encoding && ender < 0x100)
8628 goto recode_encoding;
8631 if (! RExC_override_recoding) {
8632 SV* enc = PL_encoding;
8633 ender = reg_recode((const char)(U8)ender, &enc);
8634 if (!enc && SIZE_ONLY)
8635 ckWARNreg(p, "Invalid escape in the specified encoding");
8641 FAIL("Trailing \\");
8644 if (!SIZE_ONLY&& isALPHA(*p)) {
8645 /* Include any { following the alpha to emphasize
8646 * that it could be part of an escape at some point
8648 int len = (*(p + 1) == '{') ? 2 : 1;
8649 ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p);
8651 goto normal_default;
8656 if (UTF8_IS_START(*p) && UTF) {
8658 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
8659 &numlen, UTF8_ALLOW_DEFAULT);
8665 } /* End of switch on the literal */
8667 /* Certain characters are problematic because their folded
8668 * length is so different from their original length that it
8669 * isn't handleable by the optimizer. They are therefore not
8670 * placed in an EXACTish node; and are here handled specially.
8671 * (Even if the optimizer handled LATIN_SMALL_LETTER_SHARP_S,
8672 * putting it in a special node keeps regexec from having to
8673 * deal with a non-utf8 multi-char fold */
8675 && (ender > 255 || (! MORE_ASCII_RESTRICTED && ! LOC)))
8677 /* We look for either side of the fold. For example \xDF
8678 * folds to 'ss'. We look for both the single character
8679 * \xDF and the sequence 'ss'. When we find something that
8680 * could be one of those, we stop and flush whatever we
8681 * have output so far into the EXACTish node that was being
8682 * built. Then restore the input pointer to what it was.
8683 * regatom will return that EXACT node, and will be called
8684 * again, positioned so the first character is the one in
8685 * question, which we return in a different node type.
8686 * The multi-char folds are a sequence, so the occurrence
8687 * of the first character in that sequence doesn't
8688 * necessarily mean that what follows is the rest of the
8689 * sequence. We keep track of that with a state machine,
8690 * with the state being set to the latest character
8691 * processed before the current one. Most characters will
8692 * set the state to 0, but if one occurs that is part of a
8693 * potential tricky fold sequence, the state is set to that
8694 * character, and the next loop iteration sees if the state
8695 * should progress towards the final folded-from character,
8696 * or if it was a false alarm. If it turns out to be a
8697 * false alarm, the character(s) will be output in a new
8698 * EXACTish node, and join_exact() will later combine them.
8699 * In the case of the 'ss' sequence, which is more common
8700 * and more easily checked, some look-ahead is done to
8701 * save time by ruling-out some false alarms */
8704 latest_char_state = generic_char;
8708 case 0x17F: /* LATIN SMALL LETTER LONG S */
8709 if (AT_LEAST_UNI_SEMANTICS) {
8710 if (latest_char_state == char_s) { /* 'ss' */
8711 ender = LATIN_SMALL_LETTER_SHARP_S;
8714 else if (p < RExC_end) {
8716 /* Look-ahead at the next character. If it
8717 * is also an s, we handle as a sharp s
8718 * tricky regnode. */
8719 if (*p == 's' || *p == 'S') {
8721 /* But first flush anything in the
8722 * EXACTish buffer */
8727 p++; /* Account for swallowing this
8729 ender = LATIN_SMALL_LETTER_SHARP_S;
8732 /* Here, the next character is not a
8733 * literal 's', but still could
8734 * evaluate to one if part of a \o{},
8735 * \x or \OCTAL-DIGIT. The minimum
8736 * length required for that is 4, eg
8740 && (isDIGIT(*(p + 1))
8742 || *(p + 1) == 'o' ))
8745 /* Here, it could be an 's', too much
8746 * bother to figure it out here. Flush
8747 * the buffer if any; when come back
8748 * here, set the state so know that the
8749 * previous char was an 's' */
8751 latest_char_state = generic_char;
8755 latest_char_state = char_s;
8761 /* Here, can't be an 'ss' sequence, or at least not
8762 * one that could fold to/from the sharp ss */
8763 latest_char_state = generic_char;
8765 case 0x03C5: /* First char in upsilon series */
8766 case 0x03A5: /* Also capital UPSILON, which folds to
8767 03C5, and hence exhibits the same
8769 if (p < RExC_end - 4) { /* Need >= 4 bytes left */
8770 latest_char_state = upsilon_1;
8777 latest_char_state = generic_char;
8780 case 0x03B9: /* First char in iota series */
8781 case 0x0399: /* Also capital IOTA */
8782 case 0x1FBE: /* GREEK PROSGEGRAMMENI folds to 3B9 */
8783 case 0x0345: /* COMBINING GREEK YPOGEGRAMMENI folds
8785 if (p < RExC_end - 4) {
8786 latest_char_state = iota_1;
8793 latest_char_state = generic_char;
8797 if (latest_char_state == upsilon_1) {
8798 latest_char_state = upsilon_2;
8800 else if (latest_char_state == iota_1) {
8801 latest_char_state = iota_2;
8804 latest_char_state = generic_char;
8808 if (latest_char_state == upsilon_2) {
8809 ender = GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS;
8812 else if (latest_char_state == iota_2) {
8813 ender = GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS;
8816 latest_char_state = generic_char;
8819 /* These are the tricky fold characters. Flush any
8820 * buffer first. (When adding to this list, also should
8821 * add them to fold_grind.t to make sure get tested) */
8822 case GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS:
8823 case GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS:
8824 case LATIN_SMALL_LETTER_SHARP_S:
8825 case LATIN_CAPITAL_LETTER_SHARP_S:
8826 case 0x1FD3: /* GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA */
8827 case 0x1FE3: /* GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA */
8834 char* const oldregxend = RExC_end;
8835 U8 tmpbuf[UTF8_MAXBYTES+1];
8837 /* Here, we know we need to generate a special
8838 * regnode, and 'ender' contains the tricky
8839 * character. What's done is to pretend it's in a
8840 * [bracketed] class, and let the code that deals
8841 * with those handle it, as that code has all the
8842 * intelligence necessary. First save the current
8843 * parse state, get rid of the already allocated
8844 * but empty EXACT node that the ANYOFV node will
8845 * replace, and point the parse to a buffer which
8846 * we fill with the character we want the regclass
8847 * code to think is being parsed */
8848 RExC_emit = orig_emit;
8849 RExC_parse = (char *) tmpbuf;
8851 U8 *d = uvchr_to_utf8(tmpbuf, ender);
8853 RExC_end = (char *) d;
8855 else { /* ender above 255 already excluded */
8856 tmpbuf[0] = (U8) ender;
8858 RExC_end = RExC_parse + 1;
8861 ret = regclass(pRExC_state,depth+1);
8863 /* Here, have parsed the buffer. Reset the parse to
8864 * the actual input, and return */
8865 RExC_end = oldregxend;
8868 Set_Node_Offset(ret, RExC_parse);
8869 Set_Node_Cur_Length(ret);
8870 nextchar(pRExC_state);
8871 *flagp |= HASWIDTH|SIMPLE;
8877 if ( RExC_flags & RXf_PMf_EXTENDED)
8878 p = regwhite( pRExC_state, p );
8880 /* Prime the casefolded buffer. Locale rules, which apply
8881 * only to code points < 256, aren't known until execution,
8882 * so for them, just output the original character using
8884 if (LOC && ender < 256) {
8885 if (UNI_IS_INVARIANT(ender)) {
8886 *tmpbuf = (U8) ender;
8889 *tmpbuf = UTF8_TWO_BYTE_HI(ender);
8890 *(tmpbuf + 1) = UTF8_TWO_BYTE_LO(ender);
8894 else if (isASCII(ender)) { /* Note: Here can't also be LOC
8896 ender = toLOWER(ender);
8897 *tmpbuf = (U8) ender;
8900 else if (! MORE_ASCII_RESTRICTED && ! LOC) {
8902 /* Locale and /aa require more selectivity about the
8903 * fold, so are handled below. Otherwise, here, just
8905 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
8908 /* Under locale rules or /aa we are not to mix,
8909 * respectively, ords < 256 or ASCII with non-. So
8910 * reject folds that mix them, using only the
8911 * non-folded code point. So do the fold to a
8912 * temporary, and inspect each character in it. */
8913 U8 trialbuf[UTF8_MAXBYTES_CASE+1];
8915 UV tmpender = toFOLD_uni(ender, trialbuf, &foldlen);
8916 U8* e = s + foldlen;
8917 bool fold_ok = TRUE;
8921 || (LOC && (UTF8_IS_INVARIANT(*s)
8922 || UTF8_IS_DOWNGRADEABLE_START(*s))))
8930 Copy(trialbuf, tmpbuf, foldlen, U8);
8934 uvuni_to_utf8(tmpbuf, ender);
8935 foldlen = UNISKIP(ender);
8939 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
8944 /* Emit all the Unicode characters. */
8946 for (foldbuf = tmpbuf;
8948 foldlen -= numlen) {
8949 ender = utf8_to_uvchr(foldbuf, &numlen);
8951 const STRLEN unilen = reguni(pRExC_state, ender, s);
8954 /* In EBCDIC the numlen
8955 * and unilen can differ. */
8957 if (numlen >= foldlen)
8961 break; /* "Can't happen." */
8965 const STRLEN unilen = reguni(pRExC_state, ender, s);
8974 REGC((char)ender, s++);
8980 /* Emit all the Unicode characters. */
8982 for (foldbuf = tmpbuf;
8984 foldlen -= numlen) {
8985 ender = utf8_to_uvchr(foldbuf, &numlen);
8987 const STRLEN unilen = reguni(pRExC_state, ender, s);
8990 /* In EBCDIC the numlen
8991 * and unilen can differ. */
8993 if (numlen >= foldlen)
9001 const STRLEN unilen = reguni(pRExC_state, ender, s);
9010 REGC((char)ender, s++);
9013 loopdone: /* Jumped to when encounters something that shouldn't be in
9016 Set_Node_Cur_Length(ret); /* MJD */
9017 nextchar(pRExC_state);
9019 /* len is STRLEN which is unsigned, need to copy to signed */
9022 vFAIL("Internal disaster");
9026 if (len == 1 && UNI_IS_INVARIANT(ender))
9030 RExC_size += STR_SZ(len);
9033 RExC_emit += STR_SZ(len);
9041 /* Jumped to when an unrecognized character set is encountered */
9043 Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags));
9048 S_regwhite( RExC_state_t *pRExC_state, char *p )
9050 const char *e = RExC_end;
9052 PERL_ARGS_ASSERT_REGWHITE;
9057 else if (*p == '#') {
9066 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
9074 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
9075 Character classes ([:foo:]) can also be negated ([:^foo:]).
9076 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
9077 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
9078 but trigger failures because they are currently unimplemented. */
9080 #define POSIXCC_DONE(c) ((c) == ':')
9081 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
9082 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
9085 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
9088 I32 namedclass = OOB_NAMEDCLASS;
9090 PERL_ARGS_ASSERT_REGPPOSIXCC;
9092 if (value == '[' && RExC_parse + 1 < RExC_end &&
9093 /* I smell either [: or [= or [. -- POSIX has been here, right? */
9094 POSIXCC(UCHARAT(RExC_parse))) {
9095 const char c = UCHARAT(RExC_parse);
9096 char* const s = RExC_parse++;
9098 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
9100 if (RExC_parse == RExC_end)
9101 /* Grandfather lone [:, [=, [. */
9104 const char* const t = RExC_parse++; /* skip over the c */
9107 if (UCHARAT(RExC_parse) == ']') {
9108 const char *posixcc = s + 1;
9109 RExC_parse++; /* skip over the ending ] */
9112 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
9113 const I32 skip = t - posixcc;
9115 /* Initially switch on the length of the name. */
9118 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
9119 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
9122 /* Names all of length 5. */
9123 /* alnum alpha ascii blank cntrl digit graph lower
9124 print punct space upper */
9125 /* Offset 4 gives the best switch position. */
9126 switch (posixcc[4]) {
9128 if (memEQ(posixcc, "alph", 4)) /* alpha */
9129 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
9132 if (memEQ(posixcc, "spac", 4)) /* space */
9133 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
9136 if (memEQ(posixcc, "grap", 4)) /* graph */
9137 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
9140 if (memEQ(posixcc, "asci", 4)) /* ascii */
9141 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
9144 if (memEQ(posixcc, "blan", 4)) /* blank */
9145 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
9148 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
9149 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
9152 if (memEQ(posixcc, "alnu", 4)) /* alnum */
9153 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
9156 if (memEQ(posixcc, "lowe", 4)) /* lower */
9157 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
9158 else if (memEQ(posixcc, "uppe", 4)) /* upper */
9159 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
9162 if (memEQ(posixcc, "digi", 4)) /* digit */
9163 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
9164 else if (memEQ(posixcc, "prin", 4)) /* print */
9165 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
9166 else if (memEQ(posixcc, "punc", 4)) /* punct */
9167 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
9172 if (memEQ(posixcc, "xdigit", 6))
9173 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
9177 if (namedclass == OOB_NAMEDCLASS)
9178 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
9180 assert (posixcc[skip] == ':');
9181 assert (posixcc[skip+1] == ']');
9182 } else if (!SIZE_ONLY) {
9183 /* [[=foo=]] and [[.foo.]] are still future. */
9185 /* adjust RExC_parse so the warning shows after
9187 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
9189 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9192 /* Maternal grandfather:
9193 * "[:" ending in ":" but not in ":]" */
9203 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
9207 PERL_ARGS_ASSERT_CHECKPOSIXCC;
9209 if (POSIXCC(UCHARAT(RExC_parse))) {
9210 const char *s = RExC_parse;
9211 const char c = *s++;
9215 if (*s && c == *s && s[1] == ']') {
9217 "POSIX syntax [%c %c] belongs inside character classes",
9220 /* [[=foo=]] and [[.foo.]] are still future. */
9221 if (POSIXCC_NOTYET(c)) {
9222 /* adjust RExC_parse so the error shows after
9224 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
9226 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9232 /* No locale test, and always Unicode semantics */
9233 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
9235 for (value = 0; value < 256; value++) \
9237 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9241 case ANYOF_N##NAME: \
9242 for (value = 0; value < 256; value++) \
9244 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9249 /* Like the above, but there are differences if we are in uni-8-bit or not, so
9250 * there are two tests passed in, to use depending on that. There aren't any
9251 * cases where the label is different from the name, so no need for that
9253 #define _C_C_T_(NAME, TEST_8, TEST_7, WORD) \
9255 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
9256 else if (UNI_SEMANTICS) { \
9257 for (value = 0; value < 256; value++) { \
9258 if (TEST_8(value)) stored += \
9259 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9263 for (value = 0; value < 128; value++) { \
9264 if (TEST_7(UNI_TO_NATIVE(value))) stored += \
9265 set_regclass_bit(pRExC_state, ret, \
9266 (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9272 case ANYOF_N##NAME: \
9273 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
9274 else if (UNI_SEMANTICS) { \
9275 for (value = 0; value < 256; value++) { \
9276 if (! TEST_8(value)) stored += \
9277 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9281 for (value = 0; value < 128; value++) { \
9282 if (! TEST_7(UNI_TO_NATIVE(value))) stored += set_regclass_bit( \
9283 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9285 if (AT_LEAST_ASCII_RESTRICTED) { \
9286 for (value = 128; value < 256; value++) { \
9287 stored += set_regclass_bit( \
9288 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9290 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; \
9293 /* For a non-ut8 target string with DEPENDS semantics, all above \
9294 * ASCII Latin1 code points match the complement of any of the \
9295 * classes. But in utf8, they have their Unicode semantics, so \
9296 * can't just set them in the bitmap, or else regexec.c will think \
9297 * they matched when they shouldn't. */ \
9298 ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; \
9306 S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr)
9309 /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes.
9310 * Locale folding is done at run-time, so this function should not be
9311 * called for nodes that are for locales.
9313 * This function sets the bit corresponding to the fold of the input
9314 * 'value', if not already set. The fold of 'f' is 'F', and the fold of
9317 * It also knows about the characters that are in the bitmap that have
9318 * folds that are matchable only outside it, and sets the appropriate lists
9321 * It returns the number of bits that actually changed from 0 to 1 */
9326 PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD;
9328 fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value]
9331 /* It assumes the bit for 'value' has already been set */
9332 if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) {
9333 ANYOF_BITMAP_SET(node, fold);
9336 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) {
9337 /* Certain Latin1 characters have matches outside the bitmap. To get
9338 * here, 'value' is one of those characters. None of these matches is
9339 * valid for ASCII characters under /aa, which have been excluded by
9340 * the 'if' above. The matches fall into three categories:
9341 * 1) They are singly folded-to or -from an above 255 character, as
9342 * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y
9344 * 2) They are part of a multi-char fold with another character in the
9345 * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill;
9346 * 3) They are part of a multi-char fold with a character not in the
9347 * bitmap, such as various ligatures.
9348 * We aren't dealing fully with multi-char folds, except we do deal
9349 * with the pattern containing a character that has a multi-char fold
9350 * (not so much the inverse).
9351 * For types 1) and 3), the matches only happen when the target string
9352 * is utf8; that's not true for 2), and we set a flag for it.
9354 * The code below adds to the passed in inversion list the single fold
9355 * closures for 'value'. The values are hard-coded here so that an
9356 * innocent-looking character class, like /[ks]/i won't have to go out
9357 * to disk to find the possible matches. XXX It would be better to
9358 * generate these via regen, in case a new version of the Unicode
9359 * standard adds new mappings, though that is not really likely. */
9364 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A);
9368 /* LATIN SMALL LETTER LONG S */
9369 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F);
9372 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9373 GREEK_SMALL_LETTER_MU);
9374 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9375 GREEK_CAPITAL_LETTER_MU);
9377 case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
9378 case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
9380 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B);
9381 if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */
9382 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9383 PL_fold_latin1[value]);
9386 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
9387 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9388 LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
9390 case LATIN_SMALL_LETTER_SHARP_S:
9391 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9392 LATIN_CAPITAL_LETTER_SHARP_S);
9394 /* Under /a, /d, and /u, this can match the two chars "ss" */
9395 if (! MORE_ASCII_RESTRICTED) {
9396 add_alternate(alternate_ptr, (U8 *) "ss", 2);
9398 /* And under /u or /a, it can match even if the target is
9400 if (AT_LEAST_UNI_SEMANTICS) {
9401 ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8;
9415 /* These all are targets of multi-character folds from code
9416 * points that require UTF8 to express, so they can't match
9417 * unless the target string is in UTF-8, so no action here is
9418 * necessary, as regexec.c properly handles the general case
9419 * for UTF-8 matching */
9422 /* Use deprecated warning to increase the chances of this
9424 ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value);
9428 else if (DEPENDS_SEMANTICS
9430 && PL_fold_latin1[value] != value)
9432 /* Under DEPENDS rules, non-ASCII Latin1 characters match their
9433 * folds only when the target string is in UTF-8. We add the fold
9434 * here to the list of things to match outside the bitmap, which
9435 * won't be looked at unless it is UTF8 (or else if something else
9436 * says to look even if not utf8, but those things better not happen
9437 * under DEPENDS semantics. */
9438 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]);
9445 PERL_STATIC_INLINE U8
9446 S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr)
9448 /* This inline function sets a bit in the bitmap if not already set, and if
9449 * appropriate, its fold, returning the number of bits that actually
9450 * changed from 0 to 1 */
9454 PERL_ARGS_ASSERT_SET_REGCLASS_BIT;
9456 if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */
9460 ANYOF_BITMAP_SET(node, value);
9463 if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */
9464 stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr);
9471 S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len)
9473 /* Adds input 'string' with length 'len' to the ANYOF node's unicode
9474 * alternate list, pointed to by 'alternate_ptr'. This is an array of
9475 * the multi-character folds of characters in the node */
9478 PERL_ARGS_ASSERT_ADD_ALTERNATE;
9480 if (! *alternate_ptr) {
9481 *alternate_ptr = newAV();
9483 sv = newSVpvn_utf8((char*)string, len, TRUE);
9484 av_push(*alternate_ptr, sv);
9489 parse a class specification and produce either an ANYOF node that
9490 matches the pattern or perhaps will be optimized into an EXACTish node
9491 instead. The node contains a bit map for the first 256 characters, with the
9492 corresponding bit set if that character is in the list. For characters
9493 above 255, a range list is used */
9496 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
9499 register UV nextvalue;
9500 register IV prevvalue = OOB_UNICODE;
9501 register IV range = 0;
9502 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
9503 register regnode *ret;
9506 char *rangebegin = NULL;
9507 bool need_class = 0;
9508 bool allow_full_fold = TRUE; /* Assume wants multi-char folding */
9510 STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more
9511 than just initialized. */
9514 /* code points this node matches that can't be stored in the bitmap */
9515 HV* nonbitmap = NULL;
9517 /* The items that are to match that aren't stored in the bitmap, but are a
9518 * result of things that are stored there. This is the fold closure of
9519 * such a character, either because it has DEPENDS semantics and shouldn't
9520 * be matched unless the target string is utf8, or is a code point that is
9521 * too large for the bit map, as for example, the fold of the MICRO SIGN is
9522 * above 255. This all is solely for performance reasons. By having this
9523 * code know the outside-the-bitmap folds that the bitmapped characters are
9524 * involved with, we don't have to go out to disk to find the list of
9525 * matches, unless the character class includes code points that aren't
9526 * storable in the bit map. That means that a character class with an 's'
9527 * in it, for example, doesn't need to go out to disk to find everything
9528 * that matches. A 2nd list is used so that the 'nonbitmap' list is kept
9529 * empty unless there is something whose fold we don't know about, and will
9530 * have to go out to the disk to find. */
9531 HV* l1_fold_invlist = NULL;
9533 /* List of multi-character folds that are matched by this node */
9534 AV* unicode_alternate = NULL;
9536 UV literal_endpoint = 0;
9538 UV stored = 0; /* how many chars stored in the bitmap */
9540 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
9541 case we need to change the emitted regop to an EXACT. */
9542 const char * orig_parse = RExC_parse;
9543 GET_RE_DEBUG_FLAGS_DECL;
9545 PERL_ARGS_ASSERT_REGCLASS;
9547 PERL_UNUSED_ARG(depth);
9550 DEBUG_PARSE("clas");
9552 /* Assume we are going to generate an ANYOF node. */
9553 ret = reganode(pRExC_state, ANYOF, 0);
9557 ANYOF_FLAGS(ret) = 0;
9560 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
9564 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
9566 /* We have decided to not allow multi-char folds in inverted character
9567 * classes, due to the confusion that can happen, especially with
9568 * classes that are designed for a non-Unicode world: You have the
9569 * peculiar case that:
9570 "s s" =~ /^[^\xDF]+$/i => Y
9571 "ss" =~ /^[^\xDF]+$/i => N
9573 * See [perl #89750] */
9574 allow_full_fold = FALSE;
9578 RExC_size += ANYOF_SKIP;
9579 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
9582 RExC_emit += ANYOF_SKIP;
9584 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
9586 ANYOF_BITMAP_ZERO(ret);
9587 listsv = newSVpvs("# comment\n");
9588 initial_listsv_len = SvCUR(listsv);
9591 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9593 if (!SIZE_ONLY && POSIXCC(nextvalue))
9594 checkposixcc(pRExC_state);
9596 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
9597 if (UCHARAT(RExC_parse) == ']')
9601 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
9605 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
9608 rangebegin = RExC_parse;
9610 value = utf8n_to_uvchr((U8*)RExC_parse,
9611 RExC_end - RExC_parse,
9612 &numlen, UTF8_ALLOW_DEFAULT);
9613 RExC_parse += numlen;
9616 value = UCHARAT(RExC_parse++);
9618 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9619 if (value == '[' && POSIXCC(nextvalue))
9620 namedclass = regpposixcc(pRExC_state, value);
9621 else if (value == '\\') {
9623 value = utf8n_to_uvchr((U8*)RExC_parse,
9624 RExC_end - RExC_parse,
9625 &numlen, UTF8_ALLOW_DEFAULT);
9626 RExC_parse += numlen;
9629 value = UCHARAT(RExC_parse++);
9630 /* Some compilers cannot handle switching on 64-bit integer
9631 * values, therefore value cannot be an UV. Yes, this will
9632 * be a problem later if we want switch on Unicode.
9633 * A similar issue a little bit later when switching on
9634 * namedclass. --jhi */
9635 switch ((I32)value) {
9636 case 'w': namedclass = ANYOF_ALNUM; break;
9637 case 'W': namedclass = ANYOF_NALNUM; break;
9638 case 's': namedclass = ANYOF_SPACE; break;
9639 case 'S': namedclass = ANYOF_NSPACE; break;
9640 case 'd': namedclass = ANYOF_DIGIT; break;
9641 case 'D': namedclass = ANYOF_NDIGIT; break;
9642 case 'v': namedclass = ANYOF_VERTWS; break;
9643 case 'V': namedclass = ANYOF_NVERTWS; break;
9644 case 'h': namedclass = ANYOF_HORIZWS; break;
9645 case 'H': namedclass = ANYOF_NHORIZWS; break;
9646 case 'N': /* Handle \N{NAME} in class */
9648 /* We only pay attention to the first char of
9649 multichar strings being returned. I kinda wonder
9650 if this makes sense as it does change the behaviour
9651 from earlier versions, OTOH that behaviour was broken
9653 UV v; /* value is register so we cant & it /grrr */
9654 if (reg_namedseq(pRExC_state, &v, NULL, depth)) {
9664 if (RExC_parse >= RExC_end)
9665 vFAIL2("Empty \\%c{}", (U8)value);
9666 if (*RExC_parse == '{') {
9667 const U8 c = (U8)value;
9668 e = strchr(RExC_parse++, '}');
9670 vFAIL2("Missing right brace on \\%c{}", c);
9671 while (isSPACE(UCHARAT(RExC_parse)))
9673 if (e == RExC_parse)
9674 vFAIL2("Empty \\%c{}", c);
9676 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
9684 if (UCHARAT(RExC_parse) == '^') {
9687 value = value == 'p' ? 'P' : 'p'; /* toggle */
9688 while (isSPACE(UCHARAT(RExC_parse))) {
9694 /* Add the property name to the list. If /i matching, give
9695 * a different name which consists of the normal name
9696 * sandwiched between two underscores and '_i'. The design
9697 * is discussed in the commit message for this. */
9698 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s%.*s%s\n",
9699 (value=='p' ? '+' : '!'),
9708 /* The \p could match something in the Latin1 range, hence
9709 * something that isn't utf8 */
9710 ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8;
9711 namedclass = ANYOF_MAX; /* no official name, but it's named */
9713 /* \p means they want Unicode semantics */
9714 RExC_uni_semantics = 1;
9717 case 'n': value = '\n'; break;
9718 case 'r': value = '\r'; break;
9719 case 't': value = '\t'; break;
9720 case 'f': value = '\f'; break;
9721 case 'b': value = '\b'; break;
9722 case 'e': value = ASCII_TO_NATIVE('\033');break;
9723 case 'a': value = ASCII_TO_NATIVE('\007');break;
9725 RExC_parse--; /* function expects to be pointed at the 'o' */
9727 const char* error_msg;
9728 bool valid = grok_bslash_o(RExC_parse,
9733 RExC_parse += numlen;
9738 if (PL_encoding && value < 0x100) {
9739 goto recode_encoding;
9743 if (*RExC_parse == '{') {
9744 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
9745 | PERL_SCAN_DISALLOW_PREFIX;
9746 char * const e = strchr(RExC_parse++, '}');
9748 vFAIL("Missing right brace on \\x{}");
9750 numlen = e - RExC_parse;
9751 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
9755 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
9757 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
9758 RExC_parse += numlen;
9760 if (PL_encoding && value < 0x100)
9761 goto recode_encoding;
9764 value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY);
9766 case '0': case '1': case '2': case '3': case '4':
9767 case '5': case '6': case '7':
9769 /* Take 1-3 octal digits */
9770 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
9772 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
9773 RExC_parse += numlen;
9774 if (PL_encoding && value < 0x100)
9775 goto recode_encoding;
9779 if (! RExC_override_recoding) {
9780 SV* enc = PL_encoding;
9781 value = reg_recode((const char)(U8)value, &enc);
9782 if (!enc && SIZE_ONLY)
9783 ckWARNreg(RExC_parse,
9784 "Invalid escape in the specified encoding");
9788 /* Allow \_ to not give an error */
9789 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
9790 ckWARN2reg(RExC_parse,
9791 "Unrecognized escape \\%c in character class passed through",
9796 } /* end of \blah */
9802 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
9804 /* What matches in a locale is not known until runtime, so need to
9805 * (one time per class) allocate extra space to pass to regexec.
9806 * The space will contain a bit for each named class that is to be
9807 * matched against. This isn't needed for \p{} and pseudo-classes,
9808 * as they are not affected by locale, and hence are dealt with
9810 if (LOC && namedclass < ANYOF_MAX && ! need_class) {
9813 RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP;
9816 RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP;
9817 ANYOF_CLASS_ZERO(ret);
9819 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
9822 /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a
9823 * literal, as is the character that began the false range, i.e.
9824 * the 'a' in the examples */
9828 RExC_parse >= rangebegin ?
9829 RExC_parse - rangebegin : 0;
9830 ckWARN4reg(RExC_parse,
9831 "False [] range \"%*.*s\"",
9835 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
9836 if (prevvalue < 256) {
9838 set_regclass_bit(pRExC_state, ret, (U8) prevvalue, &l1_fold_invlist, &unicode_alternate);
9841 nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue);
9845 range = 0; /* this was not a true range */
9851 const char *what = NULL;
9854 /* Possible truncation here but in some 64-bit environments
9855 * the compiler gets heartburn about switch on 64-bit values.
9856 * A similar issue a little earlier when switching on value.
9858 switch ((I32)namedclass) {
9860 case _C_C_T_(ALNUMC, isALNUMC_L1, isALNUMC, "XPosixAlnum");
9861 case _C_C_T_(ALPHA, isALPHA_L1, isALPHA, "XPosixAlpha");
9862 case _C_C_T_(BLANK, isBLANK_L1, isBLANK, "XPosixBlank");
9863 case _C_C_T_(CNTRL, isCNTRL_L1, isCNTRL, "XPosixCntrl");
9864 case _C_C_T_(GRAPH, isGRAPH_L1, isGRAPH, "XPosixGraph");
9865 case _C_C_T_(LOWER, isLOWER_L1, isLOWER, "XPosixLower");
9866 case _C_C_T_(PRINT, isPRINT_L1, isPRINT, "XPosixPrint");
9867 case _C_C_T_(PSXSPC, isPSXSPC_L1, isPSXSPC, "XPosixSpace");
9868 case _C_C_T_(PUNCT, isPUNCT_L1, isPUNCT, "XPosixPunct");
9869 case _C_C_T_(UPPER, isUPPER_L1, isUPPER, "XPosixUpper");
9870 /* \s, \w match all unicode if utf8. */
9871 case _C_C_T_(SPACE, isSPACE_L1, isSPACE, "SpacePerl");
9872 case _C_C_T_(ALNUM, isWORDCHAR_L1, isALNUM, "Word");
9873 case _C_C_T_(XDIGIT, isXDIGIT_L1, isXDIGIT, "XPosixXDigit");
9874 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
9875 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
9878 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
9880 for (value = 0; value < 128; value++)
9882 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
9885 what = NULL; /* Doesn't match outside ascii, so
9886 don't want to add +utf8:: */
9890 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
9892 for (value = 128; value < 256; value++)
9894 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
9896 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
9902 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
9904 /* consecutive digits assumed */
9905 for (value = '0'; value <= '9'; value++)
9907 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9914 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
9916 /* consecutive digits assumed */
9917 for (value = 0; value < '0'; value++)
9919 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9920 for (value = '9' + 1; value < 256; value++)
9922 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9926 if (AT_LEAST_ASCII_RESTRICTED ) {
9927 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
9931 /* this is to handle \p and \P */
9934 vFAIL("Invalid [::] class");
9937 if (what && ! (AT_LEAST_ASCII_RESTRICTED)) {
9938 /* Strings such as "+utf8::isWord\n" */
9939 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
9944 } /* end of namedclass \blah */
9947 if (prevvalue > (IV)value) /* b-a */ {
9948 const int w = RExC_parse - rangebegin;
9949 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
9950 range = 0; /* not a valid range */
9954 prevvalue = value; /* save the beginning of the range */
9955 if (RExC_parse+1 < RExC_end
9956 && *RExC_parse == '-'
9957 && RExC_parse[1] != ']')
9961 /* a bad range like \w-, [:word:]- ? */
9962 if (namedclass > OOB_NAMEDCLASS) {
9963 if (ckWARN(WARN_REGEXP)) {
9965 RExC_parse >= rangebegin ?
9966 RExC_parse - rangebegin : 0;
9968 "False [] range \"%*.*s\"",
9973 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
9975 range = 1; /* yeah, it's a range! */
9976 continue; /* but do it the next time */
9980 /* non-Latin1 code point implies unicode semantics. Must be set in
9981 * pass1 so is there for the whole of pass 2 */
9983 RExC_uni_semantics = 1;
9986 /* now is the next time */
9988 if (prevvalue < 256) {
9989 const IV ceilvalue = value < 256 ? value : 255;
9992 /* In EBCDIC [\x89-\x91] should include
9993 * the \x8e but [i-j] should not. */
9994 if (literal_endpoint == 2 &&
9995 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
9996 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
9998 if (isLOWER(prevvalue)) {
9999 for (i = prevvalue; i <= ceilvalue; i++)
10000 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10002 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10005 for (i = prevvalue; i <= ceilvalue; i++)
10006 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10008 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10014 for (i = prevvalue; i <= ceilvalue; i++) {
10015 stored += set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10019 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
10020 const UV natvalue = NATIVE_TO_UNI(value);
10021 nonbitmap = add_range_to_invlist(nonbitmap, prevnatvalue, natvalue);
10024 literal_endpoint = 0;
10028 range = 0; /* this range (if it was one) is done now */
10035 /****** !SIZE_ONLY AFTER HERE *********/
10037 /* If folding and there are code points above 255, we calculate all
10038 * characters that could fold to or from the ones already on the list */
10039 if (FOLD && nonbitmap) {
10042 HV* fold_intersection;
10045 /* This is a list of all the characters that participate in folds
10046 * (except marks, etc in multi-char folds */
10047 if (! PL_utf8_foldable) {
10048 SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0);
10049 PL_utf8_foldable = _swash_to_invlist(swash);
10052 /* This is a hash that for a particular fold gives all characters
10053 * that are involved in it */
10054 if (! PL_utf8_foldclosures) {
10056 /* If we were unable to find any folds, then we likely won't be
10057 * able to find the closures. So just create an empty list.
10058 * Folding will effectively be restricted to the non-Unicode rules
10059 * hard-coded into Perl. (This case happens legitimately during
10060 * compilation of Perl itself before the Unicode tables are
10062 if (invlist_len(PL_utf8_foldable) == 0) {
10063 PL_utf8_foldclosures = _new_invlist(0);
10065 /* If the folds haven't been read in, call a fold function
10067 if (! PL_utf8_tofold) {
10068 U8 dummy[UTF8_MAXBYTES+1];
10070 to_utf8_fold((U8*) "A", dummy, &dummy_len);
10072 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
10076 /* Only the characters in this class that participate in folds need
10077 * be checked. Get the intersection of this class and all the
10078 * possible characters that are foldable. This can quickly narrow
10079 * down a large class */
10080 fold_intersection = invlist_intersection(PL_utf8_foldable, nonbitmap);
10082 /* Now look at the foldable characters in this class individually */
10083 fold_list = invlist_array(fold_intersection);
10084 for (i = 0; i < invlist_len(fold_intersection); i++) {
10087 /* The next entry is the beginning of the range that is in the
10089 UV start = fold_list[i++];
10092 /* The next entry is the beginning of the next range, which
10093 * isn't in the class, so the end of the current range is one
10094 * less than that */
10095 UV end = fold_list[i] - 1;
10097 /* Look at every character in the range */
10098 for (j = start; j <= end; j++) {
10101 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
10104 _to_uni_fold_flags(j, foldbuf, &foldlen, allow_full_fold);
10106 if (foldlen > (STRLEN)UNISKIP(f)) {
10108 /* Any multicharacter foldings (disallowed in
10109 * lookbehind patterns) require the following
10110 * transform: [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst) where
10111 * E folds into "pq" and F folds into "rst", all other
10112 * characters fold to single characters. We save away
10113 * these multicharacter foldings, to be later saved as
10114 * part of the additional "s" data. */
10115 if (! RExC_in_lookbehind) {
10117 U8* e = foldbuf + foldlen;
10119 /* If any of the folded characters of this are in
10120 * the Latin1 range, tell the regex engine that
10121 * this can match a non-utf8 target string. The
10122 * only multi-byte fold whose source is in the
10123 * Latin1 range (U+00DF) applies only when the
10124 * target string is utf8, or under unicode rules */
10125 if (j > 255 || AT_LEAST_UNI_SEMANTICS) {
10128 /* Can't mix ascii with non- under /aa */
10129 if (MORE_ASCII_RESTRICTED
10130 && (isASCII(*loc) != isASCII(j)))
10132 goto end_multi_fold;
10134 if (UTF8_IS_INVARIANT(*loc)
10135 || UTF8_IS_DOWNGRADEABLE_START(*loc))
10137 /* Can't mix above and below 256 under
10140 goto end_multi_fold;
10143 |= ANYOF_NONBITMAP_NON_UTF8;
10146 loc += UTF8SKIP(loc);
10150 add_alternate(&unicode_alternate, foldbuf, foldlen);
10154 /* This is special-cased, as it is the only letter which
10155 * has both a multi-fold and single-fold in Latin1. All
10156 * the other chars that have single and multi-folds are
10157 * always in utf8, and the utf8 folding algorithm catches
10159 if (! LOC && j == LATIN_CAPITAL_LETTER_SHARP_S) {
10160 stored += set_regclass_bit(pRExC_state,
10162 LATIN_SMALL_LETTER_SHARP_S,
10163 &l1_fold_invlist, &unicode_alternate);
10167 /* Single character fold. Add everything in its fold
10168 * closure to the list that this node should match */
10171 /* The fold closures data structure is a hash with the
10172 * keys being every character that is folded to, like
10173 * 'k', and the values each an array of everything that
10174 * folds to its key. e.g. [ 'k', 'K', KELVIN_SIGN ] */
10175 if ((listp = hv_fetch(PL_utf8_foldclosures,
10176 (char *) foldbuf, foldlen, FALSE)))
10178 AV* list = (AV*) *listp;
10180 for (k = 0; k <= av_len(list); k++) {
10181 SV** c_p = av_fetch(list, k, FALSE);
10184 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
10188 /* /aa doesn't allow folds between ASCII and
10189 * non-; /l doesn't allow them between above
10191 if ((MORE_ASCII_RESTRICTED
10192 && (isASCII(c) != isASCII(j)))
10193 || (LOC && ((c < 256) != (j < 256))))
10198 if (c < 256 && AT_LEAST_UNI_SEMANTICS) {
10199 stored += set_regclass_bit(pRExC_state,
10202 &l1_fold_invlist, &unicode_alternate);
10204 /* It may be that the code point is already
10205 * in this range or already in the bitmap,
10206 * in which case we need do nothing */
10207 else if ((c < start || c > end)
10209 || ! ANYOF_BITMAP_TEST(ret, c)))
10211 nonbitmap = add_cp_to_invlist(nonbitmap, c);
10218 invlist_destroy(fold_intersection);
10221 /* Combine the two lists into one. */
10222 if (l1_fold_invlist) {
10224 HV* temp = invlist_union(nonbitmap, l1_fold_invlist);
10225 invlist_destroy(nonbitmap);
10227 invlist_destroy(l1_fold_invlist);
10230 nonbitmap = l1_fold_invlist;
10234 /* Here, we have calculated what code points should be in the character
10235 * class. Now we can see about various optimizations. Fold calculation
10236 * needs to take place before inversion. Otherwise /[^k]/i would invert to
10237 * include K, which under /i would match k. */
10239 /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't
10240 * set the FOLD flag yet, so this this does optimize those. It doesn't
10241 * optimize locale. Doing so perhaps could be done as long as there is
10242 * nothing like \w in it; some thought also would have to be given to the
10243 * interaction with above 0x100 chars */
10245 && (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT
10246 && ! unicode_alternate
10248 && SvCUR(listsv) == initial_listsv_len)
10250 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
10251 ANYOF_BITMAP(ret)[value] ^= 0xFF;
10252 stored = 256 - stored;
10254 /* The inversion means that everything above 255 is matched; and at the
10255 * same time we clear the invert flag */
10256 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
10259 /* Folding in the bitmap is taken care of above, but not for locale (for
10260 * which we have to wait to see what folding is in effect at runtime), and
10261 * for things not in the bitmap. Set run-time fold flag for these */
10262 if (FOLD && (LOC || nonbitmap || unicode_alternate)) {
10263 ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD;
10266 /* A single character class can be "optimized" into an EXACTish node.
10267 * Note that since we don't currently count how many characters there are
10268 * outside the bitmap, we are XXX missing optimization possibilities for
10269 * them. This optimization can't happen unless this is a truly single
10270 * character class, which means that it can't be an inversion into a
10271 * many-character class, and there must be no possibility of there being
10272 * things outside the bitmap. 'stored' (only) for locales doesn't include
10273 * \w, etc, so have to make a special test that they aren't present
10275 * Similarly A 2-character class of the very special form like [bB] can be
10276 * optimized into an EXACTFish node, but only for non-locales, and for
10277 * characters which only have the two folds; so things like 'fF' and 'Ii'
10278 * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE
10281 && ! unicode_alternate
10282 && SvCUR(listsv) == initial_listsv_len
10283 && ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL))
10284 && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10285 || (! ANYOF_CLASS_TEST_ANY_SET(ret)))))
10286 || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10287 && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value))
10288 /* If the latest code point has a fold whose
10289 * bit is set, it must be the only other one */
10290 && ((prevvalue = PL_fold_latin1[value]) != (IV)value)
10291 && ANYOF_BITMAP_TEST(ret, prevvalue)))))
10293 /* Note that the information needed to decide to do this optimization
10294 * is not currently available until the 2nd pass, and that the actually
10295 * used EXACTish node takes less space than the calculated ANYOF node,
10296 * and hence the amount of space calculated in the first pass is larger
10297 * than actually used, so this optimization doesn't gain us any space.
10298 * But an EXACT node is faster than an ANYOF node, and can be combined
10299 * with any adjacent EXACT nodes later by the optimizer for further
10300 * gains. The speed of executing an EXACTF is similar to an ANYOF
10301 * node, so the optimization advantage comes from the ability to join
10302 * it to adjacent EXACT nodes */
10304 const char * cur_parse= RExC_parse;
10306 RExC_emit = (regnode *)orig_emit;
10307 RExC_parse = (char *)orig_parse;
10311 /* A locale node with one point can be folded; all the other cases
10312 * with folding will have two points, since we calculate them above
10314 if (ANYOF_FLAGS(ret) & ANYOF_LOC_NONBITMAP_FOLD) {
10320 } /* else 2 chars in the bit map: the folds of each other */
10321 else if (AT_LEAST_UNI_SEMANTICS || !isASCII(value)) {
10323 /* To join adjacent nodes, they must be the exact EXACTish type.
10324 * Try to use the most likely type, by using EXACTFU if the regex
10325 * calls for them, or is required because the character is
10329 else { /* Otherwise, more likely to be EXACTF type */
10333 ret = reg_node(pRExC_state, op);
10334 RExC_parse = (char *)cur_parse;
10335 if (UTF && ! NATIVE_IS_INVARIANT(value)) {
10336 *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value);
10337 *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value);
10339 RExC_emit += STR_SZ(2);
10342 *STRING(ret)= (char)value;
10344 RExC_emit += STR_SZ(1);
10346 SvREFCNT_dec(listsv);
10351 UV* nonbitmap_array = invlist_array(nonbitmap);
10352 UV nonbitmap_len = invlist_len(nonbitmap);
10355 /* Here have the full list of items to match that aren't in the
10356 * bitmap. Convert to the structure that the rest of the code is
10357 * expecting. XXX That rest of the code should convert to this
10359 for (i = 0; i < nonbitmap_len; i++) {
10361 /* The next entry is the beginning of the range that is in the
10363 UV start = nonbitmap_array[i++];
10366 /* The next entry is the beginning of the next range, which isn't
10367 * in the class, so the end of the current range is one less than
10368 * that. But if there is no next range, it means that the range
10369 * begun by 'start' extends to infinity, which for this platform
10370 * ends at UV_MAX */
10371 if (i == nonbitmap_len) {
10375 end = nonbitmap_array[i] - 1;
10378 if (start == end) {
10379 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", start);
10382 /* The \t sets the whole range */
10383 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
10388 invlist_destroy(nonbitmap);
10391 if (SvCUR(listsv) == initial_listsv_len && ! unicode_alternate) {
10392 ARG_SET(ret, ANYOF_NONBITMAP_EMPTY);
10393 SvREFCNT_dec(listsv);
10394 SvREFCNT_dec(unicode_alternate);
10398 AV * const av = newAV();
10400 /* The 0th element stores the character class description
10401 * in its textual form: used later (regexec.c:Perl_regclass_swash())
10402 * to initialize the appropriate swash (which gets stored in
10403 * the 1st element), and also useful for dumping the regnode.
10404 * The 2nd element stores the multicharacter foldings,
10405 * used later (regexec.c:S_reginclass()). */
10406 av_store(av, 0, listsv);
10407 av_store(av, 1, NULL);
10409 /* Store any computed multi-char folds only if we are allowing
10411 if (allow_full_fold) {
10412 av_store(av, 2, MUTABLE_SV(unicode_alternate));
10413 if (unicode_alternate) { /* This node is variable length */
10418 av_store(av, 2, NULL);
10420 rv = newRV_noinc(MUTABLE_SV(av));
10421 n = add_data(pRExC_state, 1, "s");
10422 RExC_rxi->data->data[n] = (void*)rv;
10430 /* reg_skipcomment()
10432 Absorbs an /x style # comments from the input stream.
10433 Returns true if there is more text remaining in the stream.
10434 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
10435 terminates the pattern without including a newline.
10437 Note its the callers responsibility to ensure that we are
10438 actually in /x mode
10443 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
10447 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
10449 while (RExC_parse < RExC_end)
10450 if (*RExC_parse++ == '\n') {
10455 /* we ran off the end of the pattern without ending
10456 the comment, so we have to add an \n when wrapping */
10457 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
10465 Advances the parse position, and optionally absorbs
10466 "whitespace" from the inputstream.
10468 Without /x "whitespace" means (?#...) style comments only,
10469 with /x this means (?#...) and # comments and whitespace proper.
10471 Returns the RExC_parse point from BEFORE the scan occurs.
10473 This is the /x friendly way of saying RExC_parse++.
10477 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
10479 char* const retval = RExC_parse++;
10481 PERL_ARGS_ASSERT_NEXTCHAR;
10484 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
10485 RExC_parse[2] == '#') {
10486 while (*RExC_parse != ')') {
10487 if (RExC_parse == RExC_end)
10488 FAIL("Sequence (?#... not terminated");
10494 if (RExC_flags & RXf_PMf_EXTENDED) {
10495 if (isSPACE(*RExC_parse)) {
10499 else if (*RExC_parse == '#') {
10500 if ( reg_skipcomment( pRExC_state ) )
10509 - reg_node - emit a node
10511 STATIC regnode * /* Location. */
10512 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
10515 register regnode *ptr;
10516 regnode * const ret = RExC_emit;
10517 GET_RE_DEBUG_FLAGS_DECL;
10519 PERL_ARGS_ASSERT_REG_NODE;
10522 SIZE_ALIGN(RExC_size);
10526 if (RExC_emit >= RExC_emit_bound)
10527 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10529 NODE_ALIGN_FILL(ret);
10531 FILL_ADVANCE_NODE(ptr, op);
10532 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 1);
10533 #ifdef RE_TRACK_PATTERN_OFFSETS
10534 if (RExC_offsets) { /* MJD */
10535 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
10536 "reg_node", __LINE__,
10538 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
10539 ? "Overwriting end of array!\n" : "OK",
10540 (UV)(RExC_emit - RExC_emit_start),
10541 (UV)(RExC_parse - RExC_start),
10542 (UV)RExC_offsets[0]));
10543 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
10551 - reganode - emit a node with an argument
10553 STATIC regnode * /* Location. */
10554 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
10557 register regnode *ptr;
10558 regnode * const ret = RExC_emit;
10559 GET_RE_DEBUG_FLAGS_DECL;
10561 PERL_ARGS_ASSERT_REGANODE;
10564 SIZE_ALIGN(RExC_size);
10569 assert(2==regarglen[op]+1);
10571 Anything larger than this has to allocate the extra amount.
10572 If we changed this to be:
10574 RExC_size += (1 + regarglen[op]);
10576 then it wouldn't matter. Its not clear what side effect
10577 might come from that so its not done so far.
10582 if (RExC_emit >= RExC_emit_bound)
10583 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10585 NODE_ALIGN_FILL(ret);
10587 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
10588 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 2);
10589 #ifdef RE_TRACK_PATTERN_OFFSETS
10590 if (RExC_offsets) { /* MJD */
10591 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10595 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
10596 "Overwriting end of array!\n" : "OK",
10597 (UV)(RExC_emit - RExC_emit_start),
10598 (UV)(RExC_parse - RExC_start),
10599 (UV)RExC_offsets[0]));
10600 Set_Cur_Node_Offset;
10608 - reguni - emit (if appropriate) a Unicode character
10611 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
10615 PERL_ARGS_ASSERT_REGUNI;
10617 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
10621 - reginsert - insert an operator in front of already-emitted operand
10623 * Means relocating the operand.
10626 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
10629 register regnode *src;
10630 register regnode *dst;
10631 register regnode *place;
10632 const int offset = regarglen[(U8)op];
10633 const int size = NODE_STEP_REGNODE + offset;
10634 GET_RE_DEBUG_FLAGS_DECL;
10636 PERL_ARGS_ASSERT_REGINSERT;
10637 PERL_UNUSED_ARG(depth);
10638 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
10639 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
10648 if (RExC_open_parens) {
10650 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
10651 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
10652 if ( RExC_open_parens[paren] >= opnd ) {
10653 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
10654 RExC_open_parens[paren] += size;
10656 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
10658 if ( RExC_close_parens[paren] >= opnd ) {
10659 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
10660 RExC_close_parens[paren] += size;
10662 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
10667 while (src > opnd) {
10668 StructCopy(--src, --dst, regnode);
10669 #ifdef RE_TRACK_PATTERN_OFFSETS
10670 if (RExC_offsets) { /* MJD 20010112 */
10671 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
10675 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
10676 ? "Overwriting end of array!\n" : "OK",
10677 (UV)(src - RExC_emit_start),
10678 (UV)(dst - RExC_emit_start),
10679 (UV)RExC_offsets[0]));
10680 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
10681 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
10687 place = opnd; /* Op node, where operand used to be. */
10688 #ifdef RE_TRACK_PATTERN_OFFSETS
10689 if (RExC_offsets) { /* MJD */
10690 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10694 (UV)(place - RExC_emit_start) > RExC_offsets[0]
10695 ? "Overwriting end of array!\n" : "OK",
10696 (UV)(place - RExC_emit_start),
10697 (UV)(RExC_parse - RExC_start),
10698 (UV)RExC_offsets[0]));
10699 Set_Node_Offset(place, RExC_parse);
10700 Set_Node_Length(place, 1);
10703 src = NEXTOPER(place);
10704 FILL_ADVANCE_NODE(place, op);
10705 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (place) - 1);
10706 Zero(src, offset, regnode);
10710 - regtail - set the next-pointer at the end of a node chain of p to val.
10711 - SEE ALSO: regtail_study
10713 /* TODO: All three parms should be const */
10715 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
10718 register regnode *scan;
10719 GET_RE_DEBUG_FLAGS_DECL;
10721 PERL_ARGS_ASSERT_REGTAIL;
10723 PERL_UNUSED_ARG(depth);
10729 /* Find last node. */
10732 regnode * const temp = regnext(scan);
10734 SV * const mysv=sv_newmortal();
10735 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
10736 regprop(RExC_rx, mysv, scan);
10737 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
10738 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
10739 (temp == NULL ? "->" : ""),
10740 (temp == NULL ? PL_reg_name[OP(val)] : "")
10748 if (reg_off_by_arg[OP(scan)]) {
10749 ARG_SET(scan, val - scan);
10752 NEXT_OFF(scan) = val - scan;
10758 - regtail_study - set the next-pointer at the end of a node chain of p to val.
10759 - Look for optimizable sequences at the same time.
10760 - currently only looks for EXACT chains.
10762 This is experimental code. The idea is to use this routine to perform
10763 in place optimizations on branches and groups as they are constructed,
10764 with the long term intention of removing optimization from study_chunk so
10765 that it is purely analytical.
10767 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
10768 to control which is which.
10771 /* TODO: All four parms should be const */
10774 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
10777 register regnode *scan;
10779 #ifdef EXPERIMENTAL_INPLACESCAN
10782 GET_RE_DEBUG_FLAGS_DECL;
10784 PERL_ARGS_ASSERT_REGTAIL_STUDY;
10790 /* Find last node. */
10794 regnode * const temp = regnext(scan);
10795 #ifdef EXPERIMENTAL_INPLACESCAN
10796 if (PL_regkind[OP(scan)] == EXACT)
10797 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
10801 switch (OP(scan)) {
10807 if( exact == PSEUDO )
10809 else if ( exact != OP(scan) )
10818 SV * const mysv=sv_newmortal();
10819 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
10820 regprop(RExC_rx, mysv, scan);
10821 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
10822 SvPV_nolen_const(mysv),
10823 REG_NODE_NUM(scan),
10824 PL_reg_name[exact]);
10831 SV * const mysv_val=sv_newmortal();
10832 DEBUG_PARSE_MSG("");
10833 regprop(RExC_rx, mysv_val, val);
10834 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
10835 SvPV_nolen_const(mysv_val),
10836 (IV)REG_NODE_NUM(val),
10840 if (reg_off_by_arg[OP(scan)]) {
10841 ARG_SET(scan, val - scan);
10844 NEXT_OFF(scan) = val - scan;
10852 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
10856 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
10862 for (bit=0; bit<32; bit++) {
10863 if (flags & (1<<bit)) {
10864 if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */
10867 if (!set++ && lead)
10868 PerlIO_printf(Perl_debug_log, "%s",lead);
10869 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
10872 if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) {
10873 if (!set++ && lead) {
10874 PerlIO_printf(Perl_debug_log, "%s",lead);
10877 case REGEX_UNICODE_CHARSET:
10878 PerlIO_printf(Perl_debug_log, "UNICODE");
10880 case REGEX_LOCALE_CHARSET:
10881 PerlIO_printf(Perl_debug_log, "LOCALE");
10883 case REGEX_ASCII_RESTRICTED_CHARSET:
10884 PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED");
10886 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
10887 PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED");
10890 PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET");
10896 PerlIO_printf(Perl_debug_log, "\n");
10898 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
10904 Perl_regdump(pTHX_ const regexp *r)
10908 SV * const sv = sv_newmortal();
10909 SV *dsv= sv_newmortal();
10910 RXi_GET_DECL(r,ri);
10911 GET_RE_DEBUG_FLAGS_DECL;
10913 PERL_ARGS_ASSERT_REGDUMP;
10915 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
10917 /* Header fields of interest. */
10918 if (r->anchored_substr) {
10919 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
10920 RE_SV_DUMPLEN(r->anchored_substr), 30);
10921 PerlIO_printf(Perl_debug_log,
10922 "anchored %s%s at %"IVdf" ",
10923 s, RE_SV_TAIL(r->anchored_substr),
10924 (IV)r->anchored_offset);
10925 } else if (r->anchored_utf8) {
10926 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
10927 RE_SV_DUMPLEN(r->anchored_utf8), 30);
10928 PerlIO_printf(Perl_debug_log,
10929 "anchored utf8 %s%s at %"IVdf" ",
10930 s, RE_SV_TAIL(r->anchored_utf8),
10931 (IV)r->anchored_offset);
10933 if (r->float_substr) {
10934 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
10935 RE_SV_DUMPLEN(r->float_substr), 30);
10936 PerlIO_printf(Perl_debug_log,
10937 "floating %s%s at %"IVdf"..%"UVuf" ",
10938 s, RE_SV_TAIL(r->float_substr),
10939 (IV)r->float_min_offset, (UV)r->float_max_offset);
10940 } else if (r->float_utf8) {
10941 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
10942 RE_SV_DUMPLEN(r->float_utf8), 30);
10943 PerlIO_printf(Perl_debug_log,
10944 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
10945 s, RE_SV_TAIL(r->float_utf8),
10946 (IV)r->float_min_offset, (UV)r->float_max_offset);
10948 if (r->check_substr || r->check_utf8)
10949 PerlIO_printf(Perl_debug_log,
10951 (r->check_substr == r->float_substr
10952 && r->check_utf8 == r->float_utf8
10953 ? "(checking floating" : "(checking anchored"));
10954 if (r->extflags & RXf_NOSCAN)
10955 PerlIO_printf(Perl_debug_log, " noscan");
10956 if (r->extflags & RXf_CHECK_ALL)
10957 PerlIO_printf(Perl_debug_log, " isall");
10958 if (r->check_substr || r->check_utf8)
10959 PerlIO_printf(Perl_debug_log, ") ");
10961 if (ri->regstclass) {
10962 regprop(r, sv, ri->regstclass);
10963 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
10965 if (r->extflags & RXf_ANCH) {
10966 PerlIO_printf(Perl_debug_log, "anchored");
10967 if (r->extflags & RXf_ANCH_BOL)
10968 PerlIO_printf(Perl_debug_log, "(BOL)");
10969 if (r->extflags & RXf_ANCH_MBOL)
10970 PerlIO_printf(Perl_debug_log, "(MBOL)");
10971 if (r->extflags & RXf_ANCH_SBOL)
10972 PerlIO_printf(Perl_debug_log, "(SBOL)");
10973 if (r->extflags & RXf_ANCH_GPOS)
10974 PerlIO_printf(Perl_debug_log, "(GPOS)");
10975 PerlIO_putc(Perl_debug_log, ' ');
10977 if (r->extflags & RXf_GPOS_SEEN)
10978 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
10979 if (r->intflags & PREGf_SKIP)
10980 PerlIO_printf(Perl_debug_log, "plus ");
10981 if (r->intflags & PREGf_IMPLICIT)
10982 PerlIO_printf(Perl_debug_log, "implicit ");
10983 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
10984 if (r->extflags & RXf_EVAL_SEEN)
10985 PerlIO_printf(Perl_debug_log, "with eval ");
10986 PerlIO_printf(Perl_debug_log, "\n");
10987 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
10989 PERL_ARGS_ASSERT_REGDUMP;
10990 PERL_UNUSED_CONTEXT;
10991 PERL_UNUSED_ARG(r);
10992 #endif /* DEBUGGING */
10996 - regprop - printable representation of opcode
10998 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
11001 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
11002 if (flags & ANYOF_INVERT) \
11003 /*make sure the invert info is in each */ \
11004 sv_catpvs(sv, "^"); \
11010 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
11015 RXi_GET_DECL(prog,progi);
11016 GET_RE_DEBUG_FLAGS_DECL;
11018 PERL_ARGS_ASSERT_REGPROP;
11022 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
11023 /* It would be nice to FAIL() here, but this may be called from
11024 regexec.c, and it would be hard to supply pRExC_state. */
11025 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
11026 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
11028 k = PL_regkind[OP(o)];
11031 sv_catpvs(sv, " ");
11032 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
11033 * is a crude hack but it may be the best for now since
11034 * we have no flag "this EXACTish node was UTF-8"
11036 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
11037 PERL_PV_ESCAPE_UNI_DETECT |
11038 PERL_PV_ESCAPE_NONASCII |
11039 PERL_PV_PRETTY_ELLIPSES |
11040 PERL_PV_PRETTY_LTGT |
11041 PERL_PV_PRETTY_NOCLEAR
11043 } else if (k == TRIE) {
11044 /* print the details of the trie in dumpuntil instead, as
11045 * progi->data isn't available here */
11046 const char op = OP(o);
11047 const U32 n = ARG(o);
11048 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
11049 (reg_ac_data *)progi->data->data[n] :
11051 const reg_trie_data * const trie
11052 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
11054 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
11055 DEBUG_TRIE_COMPILE_r(
11056 Perl_sv_catpvf(aTHX_ sv,
11057 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
11058 (UV)trie->startstate,
11059 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
11060 (UV)trie->wordcount,
11063 (UV)TRIE_CHARCOUNT(trie),
11064 (UV)trie->uniquecharcount
11067 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
11069 int rangestart = -1;
11070 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
11071 sv_catpvs(sv, "[");
11072 for (i = 0; i <= 256; i++) {
11073 if (i < 256 && BITMAP_TEST(bitmap,i)) {
11074 if (rangestart == -1)
11076 } else if (rangestart != -1) {
11077 if (i <= rangestart + 3)
11078 for (; rangestart < i; rangestart++)
11079 put_byte(sv, rangestart);
11081 put_byte(sv, rangestart);
11082 sv_catpvs(sv, "-");
11083 put_byte(sv, i - 1);
11088 sv_catpvs(sv, "]");
11091 } else if (k == CURLY) {
11092 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
11093 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
11094 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
11096 else if (k == WHILEM && o->flags) /* Ordinal/of */
11097 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
11098 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
11099 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
11100 if ( RXp_PAREN_NAMES(prog) ) {
11101 if ( k != REF || (OP(o) < NREF)) {
11102 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
11103 SV **name= av_fetch(list, ARG(o), 0 );
11105 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11108 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
11109 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
11110 I32 *nums=(I32*)SvPVX(sv_dat);
11111 SV **name= av_fetch(list, nums[0], 0 );
11114 for ( n=0; n<SvIVX(sv_dat); n++ ) {
11115 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
11116 (n ? "," : ""), (IV)nums[n]);
11118 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11122 } else if (k == GOSUB)
11123 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
11124 else if (k == VERB) {
11126 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
11127 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
11128 } else if (k == LOGICAL)
11129 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
11130 else if (k == FOLDCHAR)
11131 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
11132 else if (k == ANYOF) {
11133 int i, rangestart = -1;
11134 const U8 flags = ANYOF_FLAGS(o);
11137 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
11138 static const char * const anyofs[] = {
11171 if (flags & ANYOF_LOCALE)
11172 sv_catpvs(sv, "{loc}");
11173 if (flags & ANYOF_LOC_NONBITMAP_FOLD)
11174 sv_catpvs(sv, "{i}");
11175 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
11176 if (flags & ANYOF_INVERT)
11177 sv_catpvs(sv, "^");
11179 /* output what the standard cp 0-255 bitmap matches */
11180 for (i = 0; i <= 256; i++) {
11181 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
11182 if (rangestart == -1)
11184 } else if (rangestart != -1) {
11185 if (i <= rangestart + 3)
11186 for (; rangestart < i; rangestart++)
11187 put_byte(sv, rangestart);
11189 put_byte(sv, rangestart);
11190 sv_catpvs(sv, "-");
11191 put_byte(sv, i - 1);
11198 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11199 /* output any special charclass tests (used entirely under use locale) */
11200 if (ANYOF_CLASS_TEST_ANY_SET(o))
11201 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
11202 if (ANYOF_CLASS_TEST(o,i)) {
11203 sv_catpv(sv, anyofs[i]);
11207 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11209 if (flags & ANYOF_NON_UTF8_LATIN1_ALL) {
11210 sv_catpvs(sv, "{non-utf8-latin1-all}");
11213 /* output information about the unicode matching */
11214 if (flags & ANYOF_UNICODE_ALL)
11215 sv_catpvs(sv, "{unicode_all}");
11216 else if (ANYOF_NONBITMAP(o))
11217 sv_catpvs(sv, "{unicode}");
11218 if (flags & ANYOF_NONBITMAP_NON_UTF8)
11219 sv_catpvs(sv, "{outside bitmap}");
11221 if (ANYOF_NONBITMAP(o)) {
11223 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
11227 U8 s[UTF8_MAXBYTES_CASE+1];
11229 for (i = 0; i <= 256; i++) { /* just the first 256 */
11230 uvchr_to_utf8(s, i);
11232 if (i < 256 && swash_fetch(sw, s, TRUE)) {
11233 if (rangestart == -1)
11235 } else if (rangestart != -1) {
11236 if (i <= rangestart + 3)
11237 for (; rangestart < i; rangestart++) {
11238 const U8 * const e = uvchr_to_utf8(s,rangestart);
11240 for(p = s; p < e; p++)
11244 const U8 *e = uvchr_to_utf8(s,rangestart);
11246 for (p = s; p < e; p++)
11248 sv_catpvs(sv, "-");
11249 e = uvchr_to_utf8(s, i-1);
11250 for (p = s; p < e; p++)
11257 sv_catpvs(sv, "..."); /* et cetera */
11261 char *s = savesvpv(lv);
11262 char * const origs = s;
11264 while (*s && *s != '\n')
11268 const char * const t = ++s;
11286 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
11288 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
11289 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
11291 PERL_UNUSED_CONTEXT;
11292 PERL_UNUSED_ARG(sv);
11293 PERL_UNUSED_ARG(o);
11294 PERL_UNUSED_ARG(prog);
11295 #endif /* DEBUGGING */
11299 Perl_re_intuit_string(pTHX_ REGEXP * const r)
11300 { /* Assume that RE_INTUIT is set */
11302 struct regexp *const prog = (struct regexp *)SvANY(r);
11303 GET_RE_DEBUG_FLAGS_DECL;
11305 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
11306 PERL_UNUSED_CONTEXT;
11310 const char * const s = SvPV_nolen_const(prog->check_substr
11311 ? prog->check_substr : prog->check_utf8);
11313 if (!PL_colorset) reginitcolors();
11314 PerlIO_printf(Perl_debug_log,
11315 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
11317 prog->check_substr ? "" : "utf8 ",
11318 PL_colors[5],PL_colors[0],
11321 (strlen(s) > 60 ? "..." : ""));
11324 return prog->check_substr ? prog->check_substr : prog->check_utf8;
11330 handles refcounting and freeing the perl core regexp structure. When
11331 it is necessary to actually free the structure the first thing it
11332 does is call the 'free' method of the regexp_engine associated to
11333 the regexp, allowing the handling of the void *pprivate; member
11334 first. (This routine is not overridable by extensions, which is why
11335 the extensions free is called first.)
11337 See regdupe and regdupe_internal if you change anything here.
11339 #ifndef PERL_IN_XSUB_RE
11341 Perl_pregfree(pTHX_ REGEXP *r)
11347 Perl_pregfree2(pTHX_ REGEXP *rx)
11350 struct regexp *const r = (struct regexp *)SvANY(rx);
11351 GET_RE_DEBUG_FLAGS_DECL;
11353 PERL_ARGS_ASSERT_PREGFREE2;
11355 if (r->mother_re) {
11356 ReREFCNT_dec(r->mother_re);
11358 CALLREGFREE_PVT(rx); /* free the private data */
11359 SvREFCNT_dec(RXp_PAREN_NAMES(r));
11362 SvREFCNT_dec(r->anchored_substr);
11363 SvREFCNT_dec(r->anchored_utf8);
11364 SvREFCNT_dec(r->float_substr);
11365 SvREFCNT_dec(r->float_utf8);
11366 Safefree(r->substrs);
11368 RX_MATCH_COPY_FREE(rx);
11369 #ifdef PERL_OLD_COPY_ON_WRITE
11370 SvREFCNT_dec(r->saved_copy);
11377 This is a hacky workaround to the structural issue of match results
11378 being stored in the regexp structure which is in turn stored in
11379 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
11380 could be PL_curpm in multiple contexts, and could require multiple
11381 result sets being associated with the pattern simultaneously, such
11382 as when doing a recursive match with (??{$qr})
11384 The solution is to make a lightweight copy of the regexp structure
11385 when a qr// is returned from the code executed by (??{$qr}) this
11386 lightweight copy doesn't actually own any of its data except for
11387 the starp/end and the actual regexp structure itself.
11393 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
11395 struct regexp *ret;
11396 struct regexp *const r = (struct regexp *)SvANY(rx);
11397 register const I32 npar = r->nparens+1;
11399 PERL_ARGS_ASSERT_REG_TEMP_COPY;
11402 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
11403 ret = (struct regexp *)SvANY(ret_x);
11405 (void)ReREFCNT_inc(rx);
11406 /* We can take advantage of the existing "copied buffer" mechanism in SVs
11407 by pointing directly at the buffer, but flagging that the allocated
11408 space in the copy is zero. As we've just done a struct copy, it's now
11409 a case of zero-ing that, rather than copying the current length. */
11410 SvPV_set(ret_x, RX_WRAPPED(rx));
11411 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
11412 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
11413 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
11414 SvLEN_set(ret_x, 0);
11415 SvSTASH_set(ret_x, NULL);
11416 SvMAGIC_set(ret_x, NULL);
11417 Newx(ret->offs, npar, regexp_paren_pair);
11418 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11420 Newx(ret->substrs, 1, struct reg_substr_data);
11421 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11423 SvREFCNT_inc_void(ret->anchored_substr);
11424 SvREFCNT_inc_void(ret->anchored_utf8);
11425 SvREFCNT_inc_void(ret->float_substr);
11426 SvREFCNT_inc_void(ret->float_utf8);
11428 /* check_substr and check_utf8, if non-NULL, point to either their
11429 anchored or float namesakes, and don't hold a second reference. */
11431 RX_MATCH_COPIED_off(ret_x);
11432 #ifdef PERL_OLD_COPY_ON_WRITE
11433 ret->saved_copy = NULL;
11435 ret->mother_re = rx;
11441 /* regfree_internal()
11443 Free the private data in a regexp. This is overloadable by
11444 extensions. Perl takes care of the regexp structure in pregfree(),
11445 this covers the *pprivate pointer which technically perl doesn't
11446 know about, however of course we have to handle the
11447 regexp_internal structure when no extension is in use.
11449 Note this is called before freeing anything in the regexp
11454 Perl_regfree_internal(pTHX_ REGEXP * const rx)
11457 struct regexp *const r = (struct regexp *)SvANY(rx);
11458 RXi_GET_DECL(r,ri);
11459 GET_RE_DEBUG_FLAGS_DECL;
11461 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
11467 SV *dsv= sv_newmortal();
11468 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
11469 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
11470 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
11471 PL_colors[4],PL_colors[5],s);
11474 #ifdef RE_TRACK_PATTERN_OFFSETS
11476 Safefree(ri->u.offsets); /* 20010421 MJD */
11479 int n = ri->data->count;
11480 PAD* new_comppad = NULL;
11485 /* If you add a ->what type here, update the comment in regcomp.h */
11486 switch (ri->data->what[n]) {
11491 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
11494 Safefree(ri->data->data[n]);
11497 new_comppad = MUTABLE_AV(ri->data->data[n]);
11500 if (new_comppad == NULL)
11501 Perl_croak(aTHX_ "panic: pregfree comppad");
11502 PAD_SAVE_LOCAL(old_comppad,
11503 /* Watch out for global destruction's random ordering. */
11504 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
11507 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
11510 op_free((OP_4tree*)ri->data->data[n]);
11512 PAD_RESTORE_LOCAL(old_comppad);
11513 SvREFCNT_dec(MUTABLE_SV(new_comppad));
11514 new_comppad = NULL;
11519 { /* Aho Corasick add-on structure for a trie node.
11520 Used in stclass optimization only */
11522 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
11524 refcount = --aho->refcount;
11527 PerlMemShared_free(aho->states);
11528 PerlMemShared_free(aho->fail);
11529 /* do this last!!!! */
11530 PerlMemShared_free(ri->data->data[n]);
11531 PerlMemShared_free(ri->regstclass);
11537 /* trie structure. */
11539 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
11541 refcount = --trie->refcount;
11544 PerlMemShared_free(trie->charmap);
11545 PerlMemShared_free(trie->states);
11546 PerlMemShared_free(trie->trans);
11548 PerlMemShared_free(trie->bitmap);
11550 PerlMemShared_free(trie->jump);
11551 PerlMemShared_free(trie->wordinfo);
11552 /* do this last!!!! */
11553 PerlMemShared_free(ri->data->data[n]);
11558 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
11561 Safefree(ri->data->what);
11562 Safefree(ri->data);
11568 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11569 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11570 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11573 re_dup - duplicate a regexp.
11575 This routine is expected to clone a given regexp structure. It is only
11576 compiled under USE_ITHREADS.
11578 After all of the core data stored in struct regexp is duplicated
11579 the regexp_engine.dupe method is used to copy any private data
11580 stored in the *pprivate pointer. This allows extensions to handle
11581 any duplication it needs to do.
11583 See pregfree() and regfree_internal() if you change anything here.
11585 #if defined(USE_ITHREADS)
11586 #ifndef PERL_IN_XSUB_RE
11588 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
11592 const struct regexp *r = (const struct regexp *)SvANY(sstr);
11593 struct regexp *ret = (struct regexp *)SvANY(dstr);
11595 PERL_ARGS_ASSERT_RE_DUP_GUTS;
11597 npar = r->nparens+1;
11598 Newx(ret->offs, npar, regexp_paren_pair);
11599 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11601 /* no need to copy these */
11602 Newx(ret->swap, npar, regexp_paren_pair);
11605 if (ret->substrs) {
11606 /* Do it this way to avoid reading from *r after the StructCopy().
11607 That way, if any of the sv_dup_inc()s dislodge *r from the L1
11608 cache, it doesn't matter. */
11609 const bool anchored = r->check_substr
11610 ? r->check_substr == r->anchored_substr
11611 : r->check_utf8 == r->anchored_utf8;
11612 Newx(ret->substrs, 1, struct reg_substr_data);
11613 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11615 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
11616 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
11617 ret->float_substr = sv_dup_inc(ret->float_substr, param);
11618 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
11620 /* check_substr and check_utf8, if non-NULL, point to either their
11621 anchored or float namesakes, and don't hold a second reference. */
11623 if (ret->check_substr) {
11625 assert(r->check_utf8 == r->anchored_utf8);
11626 ret->check_substr = ret->anchored_substr;
11627 ret->check_utf8 = ret->anchored_utf8;
11629 assert(r->check_substr == r->float_substr);
11630 assert(r->check_utf8 == r->float_utf8);
11631 ret->check_substr = ret->float_substr;
11632 ret->check_utf8 = ret->float_utf8;
11634 } else if (ret->check_utf8) {
11636 ret->check_utf8 = ret->anchored_utf8;
11638 ret->check_utf8 = ret->float_utf8;
11643 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
11646 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
11648 if (RX_MATCH_COPIED(dstr))
11649 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
11651 ret->subbeg = NULL;
11652 #ifdef PERL_OLD_COPY_ON_WRITE
11653 ret->saved_copy = NULL;
11656 if (ret->mother_re) {
11657 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
11658 /* Our storage points directly to our mother regexp, but that's
11659 1: a buffer in a different thread
11660 2: something we no longer hold a reference on
11661 so we need to copy it locally. */
11662 /* Note we need to sue SvCUR() on our mother_re, because it, in
11663 turn, may well be pointing to its own mother_re. */
11664 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
11665 SvCUR(ret->mother_re)+1));
11666 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
11668 ret->mother_re = NULL;
11672 #endif /* PERL_IN_XSUB_RE */
11677 This is the internal complement to regdupe() which is used to copy
11678 the structure pointed to by the *pprivate pointer in the regexp.
11679 This is the core version of the extension overridable cloning hook.
11680 The regexp structure being duplicated will be copied by perl prior
11681 to this and will be provided as the regexp *r argument, however
11682 with the /old/ structures pprivate pointer value. Thus this routine
11683 may override any copying normally done by perl.
11685 It returns a pointer to the new regexp_internal structure.
11689 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
11692 struct regexp *const r = (struct regexp *)SvANY(rx);
11693 regexp_internal *reti;
11695 RXi_GET_DECL(r,ri);
11697 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
11701 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
11702 Copy(ri->program, reti->program, len+1, regnode);
11705 reti->regstclass = NULL;
11708 struct reg_data *d;
11709 const int count = ri->data->count;
11712 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
11713 char, struct reg_data);
11714 Newx(d->what, count, U8);
11717 for (i = 0; i < count; i++) {
11718 d->what[i] = ri->data->what[i];
11719 switch (d->what[i]) {
11720 /* legal options are one of: sSfpontTua
11721 see also regcomp.h and pregfree() */
11722 case 'a': /* actually an AV, but the dup function is identical. */
11725 case 'p': /* actually an AV, but the dup function is identical. */
11726 case 'u': /* actually an HV, but the dup function is identical. */
11727 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
11730 /* This is cheating. */
11731 Newx(d->data[i], 1, struct regnode_charclass_class);
11732 StructCopy(ri->data->data[i], d->data[i],
11733 struct regnode_charclass_class);
11734 reti->regstclass = (regnode*)d->data[i];
11737 /* Compiled op trees are readonly and in shared memory,
11738 and can thus be shared without duplication. */
11740 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
11744 /* Trie stclasses are readonly and can thus be shared
11745 * without duplication. We free the stclass in pregfree
11746 * when the corresponding reg_ac_data struct is freed.
11748 reti->regstclass= ri->regstclass;
11752 ((reg_trie_data*)ri->data->data[i])->refcount++;
11756 d->data[i] = ri->data->data[i];
11759 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
11768 reti->name_list_idx = ri->name_list_idx;
11770 #ifdef RE_TRACK_PATTERN_OFFSETS
11771 if (ri->u.offsets) {
11772 Newx(reti->u.offsets, 2*len+1, U32);
11773 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
11776 SetProgLen(reti,len);
11779 return (void*)reti;
11782 #endif /* USE_ITHREADS */
11784 #ifndef PERL_IN_XSUB_RE
11787 - regnext - dig the "next" pointer out of a node
11790 Perl_regnext(pTHX_ register regnode *p)
11793 register I32 offset;
11798 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
11799 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
11802 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
11811 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
11814 STRLEN l1 = strlen(pat1);
11815 STRLEN l2 = strlen(pat2);
11818 const char *message;
11820 PERL_ARGS_ASSERT_RE_CROAK2;
11826 Copy(pat1, buf, l1 , char);
11827 Copy(pat2, buf + l1, l2 , char);
11828 buf[l1 + l2] = '\n';
11829 buf[l1 + l2 + 1] = '\0';
11831 /* ANSI variant takes additional second argument */
11832 va_start(args, pat2);
11836 msv = vmess(buf, &args);
11838 message = SvPV_const(msv,l1);
11841 Copy(message, buf, l1 , char);
11842 buf[l1-1] = '\0'; /* Overwrite \n */
11843 Perl_croak(aTHX_ "%s", buf);
11846 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
11848 #ifndef PERL_IN_XSUB_RE
11850 Perl_save_re_context(pTHX)
11854 struct re_save_state *state;
11856 SAVEVPTR(PL_curcop);
11857 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
11859 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
11860 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
11861 SSPUSHUV(SAVEt_RE_STATE);
11863 Copy(&PL_reg_state, state, 1, struct re_save_state);
11865 PL_reg_start_tmp = 0;
11866 PL_reg_start_tmpl = 0;
11867 PL_reg_oldsaved = NULL;
11868 PL_reg_oldsavedlen = 0;
11869 PL_reg_maxiter = 0;
11870 PL_reg_leftiter = 0;
11871 PL_reg_poscache = NULL;
11872 PL_reg_poscache_size = 0;
11873 #ifdef PERL_OLD_COPY_ON_WRITE
11877 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
11879 const REGEXP * const rx = PM_GETRE(PL_curpm);
11882 for (i = 1; i <= RX_NPARENS(rx); i++) {
11883 char digits[TYPE_CHARS(long)];
11884 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
11885 GV *const *const gvp
11886 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
11889 GV * const gv = *gvp;
11890 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
11900 clear_re(pTHX_ void *r)
11903 ReREFCNT_dec((REGEXP *)r);
11909 S_put_byte(pTHX_ SV *sv, int c)
11911 PERL_ARGS_ASSERT_PUT_BYTE;
11913 /* Our definition of isPRINT() ignores locales, so only bytes that are
11914 not part of UTF-8 are considered printable. I assume that the same
11915 holds for UTF-EBCDIC.
11916 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
11917 which Wikipedia says:
11919 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
11920 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
11921 identical, to the ASCII delete (DEL) or rubout control character.
11922 ) So the old condition can be simplified to !isPRINT(c) */
11925 Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c);
11928 Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c);
11932 const char string = c;
11933 if (c == '-' || c == ']' || c == '\\' || c == '^')
11934 sv_catpvs(sv, "\\");
11935 sv_catpvn(sv, &string, 1);
11940 #define CLEAR_OPTSTART \
11941 if (optstart) STMT_START { \
11942 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
11946 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
11948 STATIC const regnode *
11949 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
11950 const regnode *last, const regnode *plast,
11951 SV* sv, I32 indent, U32 depth)
11954 register U8 op = PSEUDO; /* Arbitrary non-END op. */
11955 register const regnode *next;
11956 const regnode *optstart= NULL;
11958 RXi_GET_DECL(r,ri);
11959 GET_RE_DEBUG_FLAGS_DECL;
11961 PERL_ARGS_ASSERT_DUMPUNTIL;
11963 #ifdef DEBUG_DUMPUNTIL
11964 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
11965 last ? last-start : 0,plast ? plast-start : 0);
11968 if (plast && plast < last)
11971 while (PL_regkind[op] != END && (!last || node < last)) {
11972 /* While that wasn't END last time... */
11975 if (op == CLOSE || op == WHILEM)
11977 next = regnext((regnode *)node);
11980 if (OP(node) == OPTIMIZED) {
11981 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
11988 regprop(r, sv, node);
11989 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
11990 (int)(2*indent + 1), "", SvPVX_const(sv));
11992 if (OP(node) != OPTIMIZED) {
11993 if (next == NULL) /* Next ptr. */
11994 PerlIO_printf(Perl_debug_log, " (0)");
11995 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
11996 PerlIO_printf(Perl_debug_log, " (FAIL)");
11998 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
11999 (void)PerlIO_putc(Perl_debug_log, '\n');
12003 if (PL_regkind[(U8)op] == BRANCHJ) {
12006 register const regnode *nnode = (OP(next) == LONGJMP
12007 ? regnext((regnode *)next)
12009 if (last && nnode > last)
12011 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
12014 else if (PL_regkind[(U8)op] == BRANCH) {
12016 DUMPUNTIL(NEXTOPER(node), next);
12018 else if ( PL_regkind[(U8)op] == TRIE ) {
12019 const regnode *this_trie = node;
12020 const char op = OP(node);
12021 const U32 n = ARG(node);
12022 const reg_ac_data * const ac = op>=AHOCORASICK ?
12023 (reg_ac_data *)ri->data->data[n] :
12025 const reg_trie_data * const trie =
12026 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
12028 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
12030 const regnode *nextbranch= NULL;
12033 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
12034 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
12036 PerlIO_printf(Perl_debug_log, "%*s%s ",
12037 (int)(2*(indent+3)), "",
12038 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
12039 PL_colors[0], PL_colors[1],
12040 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
12041 PERL_PV_PRETTY_ELLIPSES |
12042 PERL_PV_PRETTY_LTGT
12047 U16 dist= trie->jump[word_idx+1];
12048 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
12049 (UV)((dist ? this_trie + dist : next) - start));
12052 nextbranch= this_trie + trie->jump[0];
12053 DUMPUNTIL(this_trie + dist, nextbranch);
12055 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
12056 nextbranch= regnext((regnode *)nextbranch);
12058 PerlIO_printf(Perl_debug_log, "\n");
12061 if (last && next > last)
12066 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
12067 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
12068 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
12070 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
12072 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
12074 else if ( op == PLUS || op == STAR) {
12075 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
12077 else if (PL_regkind[(U8)op] == ANYOF) {
12078 /* arglen 1 + class block */
12079 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS)
12080 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
12081 node = NEXTOPER(node);
12083 else if (PL_regkind[(U8)op] == EXACT) {
12084 /* Literal string, where present. */
12085 node += NODE_SZ_STR(node) - 1;
12086 node = NEXTOPER(node);
12089 node = NEXTOPER(node);
12090 node += regarglen[(U8)op];
12092 if (op == CURLYX || op == OPEN)
12096 #ifdef DEBUG_DUMPUNTIL
12097 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
12102 #endif /* DEBUGGING */
12106 * c-indentation-style: bsd
12107 * c-basic-offset: 4
12108 * indent-tabs-mode: t
12111 * ex: set ts=8 sts=4 sw=4 noet:
projects / perl / modules / re-engine-Hooks.git / blob