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 we do not believe that the trie logic can
3052 handle case insensitive matching properly when the
3053 pattern is not unicode (thus forcing unicode semantics).
3055 If/when this is fixed the following define can be swapped
3056 in below to fully enable trie logic.
3058 XXX It may work if not UTF and/or /a (AT_LEAST_UNI_SEMANTICS) but perhaps
3061 #define TRIE_TYPE_IS_SAFE 1
3064 #define TRIE_TYPE_IS_SAFE ((UTF && UNI_SEMANTICS) || optype==EXACT)
3066 if ( last && TRIE_TYPE_IS_SAFE ) {
3067 make_trie( pRExC_state,
3068 startbranch, first, cur, tail, count,
3071 if ( PL_regkind[ OP( noper ) ] == EXACT
3073 && noper_next == tail
3078 optype = OP( noper );
3088 regprop(RExC_rx, mysv, cur);
3089 PerlIO_printf( Perl_debug_log,
3090 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
3091 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
3095 if ( last && TRIE_TYPE_IS_SAFE ) {
3096 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3097 #ifdef TRIE_STUDY_OPT
3098 if ( ((made == MADE_EXACT_TRIE &&
3099 startbranch == first)
3100 || ( first_non_open == first )) &&
3102 flags |= SCF_TRIE_RESTUDY;
3103 if ( startbranch == first
3106 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
3116 else if ( code == BRANCHJ ) { /* single branch is optimized. */
3117 scan = NEXTOPER(NEXTOPER(scan));
3118 } else /* single branch is optimized. */
3119 scan = NEXTOPER(scan);
3121 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
3122 scan_frame *newframe = NULL;
3127 if (OP(scan) != SUSPEND) {
3128 /* set the pointer */
3129 if (OP(scan) == GOSUB) {
3131 RExC_recurse[ARG2L(scan)] = scan;
3132 start = RExC_open_parens[paren-1];
3133 end = RExC_close_parens[paren-1];
3136 start = RExC_rxi->program + 1;
3140 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
3141 SAVEFREEPV(recursed);
3143 if (!PAREN_TEST(recursed,paren+1)) {
3144 PAREN_SET(recursed,paren+1);
3145 Newx(newframe,1,scan_frame);
3147 if (flags & SCF_DO_SUBSTR) {
3148 SCAN_COMMIT(pRExC_state,data,minlenp);
3149 data->longest = &(data->longest_float);
3151 is_inf = is_inf_internal = 1;
3152 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3153 cl_anything(pRExC_state, data->start_class);
3154 flags &= ~SCF_DO_STCLASS;
3157 Newx(newframe,1,scan_frame);
3160 end = regnext(scan);
3165 SAVEFREEPV(newframe);
3166 newframe->next = regnext(scan);
3167 newframe->last = last;
3168 newframe->stop = stopparen;
3169 newframe->prev = frame;
3179 else if (OP(scan) == EXACT) {
3180 I32 l = STR_LEN(scan);
3183 const U8 * const s = (U8*)STRING(scan);
3184 l = utf8_length(s, s + l);
3185 uc = utf8_to_uvchr(s, NULL);
3187 uc = *((U8*)STRING(scan));
3190 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3191 /* The code below prefers earlier match for fixed
3192 offset, later match for variable offset. */
3193 if (data->last_end == -1) { /* Update the start info. */
3194 data->last_start_min = data->pos_min;
3195 data->last_start_max = is_inf
3196 ? I32_MAX : data->pos_min + data->pos_delta;
3198 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3200 SvUTF8_on(data->last_found);
3202 SV * const sv = data->last_found;
3203 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3204 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3205 if (mg && mg->mg_len >= 0)
3206 mg->mg_len += utf8_length((U8*)STRING(scan),
3207 (U8*)STRING(scan)+STR_LEN(scan));
3209 data->last_end = data->pos_min + l;
3210 data->pos_min += l; /* As in the first entry. */
3211 data->flags &= ~SF_BEFORE_EOL;
3213 if (flags & SCF_DO_STCLASS_AND) {
3214 /* Check whether it is compatible with what we know already! */
3218 /* If compatible, we or it in below. It is compatible if is
3219 * in the bitmp and either 1) its bit or its fold is set, or 2)
3220 * it's for a locale. Even if there isn't unicode semantics
3221 * here, at runtime there may be because of matching against a
3222 * utf8 string, so accept a possible false positive for
3223 * latin1-range folds */
3225 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3226 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3227 && (!(data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD)
3228 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3233 ANYOF_CLASS_ZERO(data->start_class);
3234 ANYOF_BITMAP_ZERO(data->start_class);
3236 ANYOF_BITMAP_SET(data->start_class, uc);
3237 else if (uc >= 0x100) {
3240 /* Some Unicode code points fold to the Latin1 range; as
3241 * XXX temporary code, instead of figuring out if this is
3242 * one, just assume it is and set all the start class bits
3243 * that could be some such above 255 code point's fold
3244 * which will generate fals positives. As the code
3245 * elsewhere that does compute the fold settles down, it
3246 * can be extracted out and re-used here */
3247 for (i = 0; i < 256; i++){
3248 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3249 ANYOF_BITMAP_SET(data->start_class, i);
3253 data->start_class->flags &= ~ANYOF_EOS;
3255 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3257 else if (flags & SCF_DO_STCLASS_OR) {
3258 /* false positive possible if the class is case-folded */
3260 ANYOF_BITMAP_SET(data->start_class, uc);
3262 data->start_class->flags |= ANYOF_UNICODE_ALL;
3263 data->start_class->flags &= ~ANYOF_EOS;
3264 cl_and(data->start_class, and_withp);
3266 flags &= ~SCF_DO_STCLASS;
3268 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3269 I32 l = STR_LEN(scan);
3270 UV uc = *((U8*)STRING(scan));
3272 /* Search for fixed substrings supports EXACT only. */
3273 if (flags & SCF_DO_SUBSTR) {
3275 SCAN_COMMIT(pRExC_state, data, minlenp);
3278 const U8 * const s = (U8 *)STRING(scan);
3279 l = utf8_length(s, s + l);
3280 uc = utf8_to_uvchr(s, NULL);
3283 if (flags & SCF_DO_SUBSTR)
3285 if (flags & SCF_DO_STCLASS_AND) {
3286 /* Check whether it is compatible with what we know already! */
3289 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3290 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3291 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3295 ANYOF_CLASS_ZERO(data->start_class);
3296 ANYOF_BITMAP_ZERO(data->start_class);
3298 ANYOF_BITMAP_SET(data->start_class, uc);
3299 data->start_class->flags &= ~ANYOF_EOS;
3300 data->start_class->flags |= ANYOF_LOC_NONBITMAP_FOLD;
3301 if (OP(scan) == EXACTFL) {
3302 /* XXX This set is probably no longer necessary, and
3303 * probably wrong as LOCALE now is on in the initial
3305 data->start_class->flags |= ANYOF_LOCALE;
3309 /* Also set the other member of the fold pair. In case
3310 * that unicode semantics is called for at runtime, use
3311 * the full latin1 fold. (Can't do this for locale,
3312 * because not known until runtime */
3313 ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]);
3316 else if (uc >= 0x100) {
3318 for (i = 0; i < 256; i++){
3319 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3320 ANYOF_BITMAP_SET(data->start_class, i);
3325 else if (flags & SCF_DO_STCLASS_OR) {
3326 if (data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) {
3327 /* false positive possible if the class is case-folded.
3328 Assume that the locale settings are the same... */
3330 ANYOF_BITMAP_SET(data->start_class, uc);
3331 if (OP(scan) != EXACTFL) {
3333 /* And set the other member of the fold pair, but
3334 * can't do that in locale because not known until
3336 ANYOF_BITMAP_SET(data->start_class,
3337 PL_fold_latin1[uc]);
3340 data->start_class->flags &= ~ANYOF_EOS;
3342 cl_and(data->start_class, and_withp);
3344 flags &= ~SCF_DO_STCLASS;
3346 else if (REGNODE_VARIES(OP(scan))) {
3347 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3348 I32 f = flags, pos_before = 0;
3349 regnode * const oscan = scan;
3350 struct regnode_charclass_class this_class;
3351 struct regnode_charclass_class *oclass = NULL;
3352 I32 next_is_eval = 0;
3354 switch (PL_regkind[OP(scan)]) {
3355 case WHILEM: /* End of (?:...)* . */
3356 scan = NEXTOPER(scan);
3359 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3360 next = NEXTOPER(scan);
3361 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3363 maxcount = REG_INFTY;
3364 next = regnext(scan);
3365 scan = NEXTOPER(scan);
3369 if (flags & SCF_DO_SUBSTR)
3374 if (flags & SCF_DO_STCLASS) {
3376 maxcount = REG_INFTY;
3377 next = regnext(scan);
3378 scan = NEXTOPER(scan);
3381 is_inf = is_inf_internal = 1;
3382 scan = regnext(scan);
3383 if (flags & SCF_DO_SUBSTR) {
3384 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3385 data->longest = &(data->longest_float);
3387 goto optimize_curly_tail;
3389 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3390 && (scan->flags == stopparen))
3395 mincount = ARG1(scan);
3396 maxcount = ARG2(scan);
3398 next = regnext(scan);
3399 if (OP(scan) == CURLYX) {
3400 I32 lp = (data ? *(data->last_closep) : 0);
3401 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3403 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3404 next_is_eval = (OP(scan) == EVAL);
3406 if (flags & SCF_DO_SUBSTR) {
3407 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3408 pos_before = data->pos_min;
3412 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3414 data->flags |= SF_IS_INF;
3416 if (flags & SCF_DO_STCLASS) {
3417 cl_init(pRExC_state, &this_class);
3418 oclass = data->start_class;
3419 data->start_class = &this_class;
3420 f |= SCF_DO_STCLASS_AND;
3421 f &= ~SCF_DO_STCLASS_OR;
3423 /* Exclude from super-linear cache processing any {n,m}
3424 regops for which the combination of input pos and regex
3425 pos is not enough information to determine if a match
3428 For example, in the regex /foo(bar\s*){4,8}baz/ with the
3429 regex pos at the \s*, the prospects for a match depend not
3430 only on the input position but also on how many (bar\s*)
3431 repeats into the {4,8} we are. */
3432 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
3433 f &= ~SCF_WHILEM_VISITED_POS;
3435 /* This will finish on WHILEM, setting scan, or on NULL: */
3436 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3437 last, data, stopparen, recursed, NULL,
3439 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3441 if (flags & SCF_DO_STCLASS)
3442 data->start_class = oclass;
3443 if (mincount == 0 || minnext == 0) {
3444 if (flags & SCF_DO_STCLASS_OR) {
3445 cl_or(pRExC_state, data->start_class, &this_class);
3447 else if (flags & SCF_DO_STCLASS_AND) {
3448 /* Switch to OR mode: cache the old value of
3449 * data->start_class */
3451 StructCopy(data->start_class, and_withp,
3452 struct regnode_charclass_class);
3453 flags &= ~SCF_DO_STCLASS_AND;
3454 StructCopy(&this_class, data->start_class,
3455 struct regnode_charclass_class);
3456 flags |= SCF_DO_STCLASS_OR;
3457 data->start_class->flags |= ANYOF_EOS;
3459 } else { /* Non-zero len */
3460 if (flags & SCF_DO_STCLASS_OR) {
3461 cl_or(pRExC_state, data->start_class, &this_class);
3462 cl_and(data->start_class, and_withp);
3464 else if (flags & SCF_DO_STCLASS_AND)
3465 cl_and(data->start_class, &this_class);
3466 flags &= ~SCF_DO_STCLASS;
3468 if (!scan) /* It was not CURLYX, but CURLY. */
3470 if ( /* ? quantifier ok, except for (?{ ... }) */
3471 (next_is_eval || !(mincount == 0 && maxcount == 1))
3472 && (minnext == 0) && (deltanext == 0)
3473 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3474 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3476 ckWARNreg(RExC_parse,
3477 "Quantifier unexpected on zero-length expression");
3480 min += minnext * mincount;
3481 is_inf_internal |= ((maxcount == REG_INFTY
3482 && (minnext + deltanext) > 0)
3483 || deltanext == I32_MAX);
3484 is_inf |= is_inf_internal;
3485 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3487 /* Try powerful optimization CURLYX => CURLYN. */
3488 if ( OP(oscan) == CURLYX && data
3489 && data->flags & SF_IN_PAR
3490 && !(data->flags & SF_HAS_EVAL)
3491 && !deltanext && minnext == 1 ) {
3492 /* Try to optimize to CURLYN. */
3493 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3494 regnode * const nxt1 = nxt;
3501 if (!REGNODE_SIMPLE(OP(nxt))
3502 && !(PL_regkind[OP(nxt)] == EXACT
3503 && STR_LEN(nxt) == 1))
3509 if (OP(nxt) != CLOSE)
3511 if (RExC_open_parens) {
3512 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3513 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3515 /* Now we know that nxt2 is the only contents: */
3516 oscan->flags = (U8)ARG(nxt);
3518 OP(nxt1) = NOTHING; /* was OPEN. */
3521 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3522 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3523 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
3524 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3525 OP(nxt + 1) = OPTIMIZED; /* was count. */
3526 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
3531 /* Try optimization CURLYX => CURLYM. */
3532 if ( OP(oscan) == CURLYX && data
3533 && !(data->flags & SF_HAS_PAR)
3534 && !(data->flags & SF_HAS_EVAL)
3535 && !deltanext /* atom is fixed width */
3536 && minnext != 0 /* CURLYM can't handle zero width */
3538 /* XXXX How to optimize if data == 0? */
3539 /* Optimize to a simpler form. */
3540 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3544 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3545 && (OP(nxt2) != WHILEM))
3547 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3548 /* Need to optimize away parenths. */
3549 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
3550 /* Set the parenth number. */
3551 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3553 oscan->flags = (U8)ARG(nxt);
3554 if (RExC_open_parens) {
3555 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3556 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3558 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3559 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3562 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3563 OP(nxt + 1) = OPTIMIZED; /* was count. */
3564 NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */
3565 NEXT_OFF(nxt + 1) = 0; /* just for consistency. */
3568 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3569 regnode *nnxt = regnext(nxt1);
3571 if (reg_off_by_arg[OP(nxt1)])
3572 ARG_SET(nxt1, nxt2 - nxt1);
3573 else if (nxt2 - nxt1 < U16_MAX)
3574 NEXT_OFF(nxt1) = nxt2 - nxt1;
3576 OP(nxt) = NOTHING; /* Cannot beautify */
3581 /* Optimize again: */
3582 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3583 NULL, stopparen, recursed, NULL, 0,depth+1);
3588 else if ((OP(oscan) == CURLYX)
3589 && (flags & SCF_WHILEM_VISITED_POS)
3590 /* See the comment on a similar expression above.
3591 However, this time it's not a subexpression
3592 we care about, but the expression itself. */
3593 && (maxcount == REG_INFTY)
3594 && data && ++data->whilem_c < 16) {
3595 /* This stays as CURLYX, we can put the count/of pair. */
3596 /* Find WHILEM (as in regexec.c) */
3597 regnode *nxt = oscan + NEXT_OFF(oscan);
3599 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3601 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3602 | (RExC_whilem_seen << 4)); /* On WHILEM */
3604 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3606 if (flags & SCF_DO_SUBSTR) {
3607 SV *last_str = NULL;
3608 int counted = mincount != 0;
3610 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3611 #if defined(SPARC64_GCC_WORKAROUND)
3614 const char *s = NULL;
3617 if (pos_before >= data->last_start_min)
3620 b = data->last_start_min;
3623 s = SvPV_const(data->last_found, l);
3624 old = b - data->last_start_min;
3627 I32 b = pos_before >= data->last_start_min
3628 ? pos_before : data->last_start_min;
3630 const char * const s = SvPV_const(data->last_found, l);
3631 I32 old = b - data->last_start_min;
3635 old = utf8_hop((U8*)s, old) - (U8*)s;
3637 /* Get the added string: */
3638 last_str = newSVpvn_utf8(s + old, l, UTF);
3639 if (deltanext == 0 && pos_before == b) {
3640 /* What was added is a constant string */
3642 SvGROW(last_str, (mincount * l) + 1);
3643 repeatcpy(SvPVX(last_str) + l,
3644 SvPVX_const(last_str), l, mincount - 1);
3645 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3646 /* Add additional parts. */
3647 SvCUR_set(data->last_found,
3648 SvCUR(data->last_found) - l);
3649 sv_catsv(data->last_found, last_str);
3651 SV * sv = data->last_found;
3653 SvUTF8(sv) && SvMAGICAL(sv) ?
3654 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3655 if (mg && mg->mg_len >= 0)
3656 mg->mg_len += CHR_SVLEN(last_str) - l;
3658 data->last_end += l * (mincount - 1);
3661 /* start offset must point into the last copy */
3662 data->last_start_min += minnext * (mincount - 1);
3663 data->last_start_max += is_inf ? I32_MAX
3664 : (maxcount - 1) * (minnext + data->pos_delta);
3667 /* It is counted once already... */
3668 data->pos_min += minnext * (mincount - counted);
3669 data->pos_delta += - counted * deltanext +
3670 (minnext + deltanext) * maxcount - minnext * mincount;
3671 if (mincount != maxcount) {
3672 /* Cannot extend fixed substrings found inside
3674 SCAN_COMMIT(pRExC_state,data,minlenp);
3675 if (mincount && last_str) {
3676 SV * const sv = data->last_found;
3677 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3678 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3682 sv_setsv(sv, last_str);
3683 data->last_end = data->pos_min;
3684 data->last_start_min =
3685 data->pos_min - CHR_SVLEN(last_str);
3686 data->last_start_max = is_inf
3688 : data->pos_min + data->pos_delta
3689 - CHR_SVLEN(last_str);
3691 data->longest = &(data->longest_float);
3693 SvREFCNT_dec(last_str);
3695 if (data && (fl & SF_HAS_EVAL))
3696 data->flags |= SF_HAS_EVAL;
3697 optimize_curly_tail:
3698 if (OP(oscan) != CURLYX) {
3699 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3701 NEXT_OFF(oscan) += NEXT_OFF(next);
3704 default: /* REF, ANYOFV, and CLUMP only? */
3705 if (flags & SCF_DO_SUBSTR) {
3706 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3707 data->longest = &(data->longest_float);
3709 is_inf = is_inf_internal = 1;
3710 if (flags & SCF_DO_STCLASS_OR)
3711 cl_anything(pRExC_state, data->start_class);
3712 flags &= ~SCF_DO_STCLASS;
3716 else if (OP(scan) == LNBREAK) {
3717 if (flags & SCF_DO_STCLASS) {
3719 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3720 if (flags & SCF_DO_STCLASS_AND) {
3721 for (value = 0; value < 256; value++)
3722 if (!is_VERTWS_cp(value))
3723 ANYOF_BITMAP_CLEAR(data->start_class, value);
3726 for (value = 0; value < 256; value++)
3727 if (is_VERTWS_cp(value))
3728 ANYOF_BITMAP_SET(data->start_class, value);
3730 if (flags & SCF_DO_STCLASS_OR)
3731 cl_and(data->start_class, and_withp);
3732 flags &= ~SCF_DO_STCLASS;
3736 if (flags & SCF_DO_SUBSTR) {
3737 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3739 data->pos_delta += 1;
3740 data->longest = &(data->longest_float);
3743 else if (OP(scan) == FOLDCHAR) {
3744 int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2;
3745 flags &= ~SCF_DO_STCLASS;
3748 if (flags & SCF_DO_SUBSTR) {
3749 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3751 data->pos_delta += d;
3752 data->longest = &(data->longest_float);
3755 else if (REGNODE_SIMPLE(OP(scan))) {
3758 if (flags & SCF_DO_SUBSTR) {
3759 SCAN_COMMIT(pRExC_state,data,minlenp);
3763 if (flags & SCF_DO_STCLASS) {
3764 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3766 /* Some of the logic below assumes that switching
3767 locale on will only add false positives. */
3768 switch (PL_regkind[OP(scan)]) {
3772 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3773 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3774 cl_anything(pRExC_state, data->start_class);
3777 if (OP(scan) == SANY)
3779 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3780 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3781 || ANYOF_CLASS_TEST_ANY_SET(data->start_class));
3782 cl_anything(pRExC_state, data->start_class);
3784 if (flags & SCF_DO_STCLASS_AND || !value)
3785 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3788 if (flags & SCF_DO_STCLASS_AND)
3789 cl_and(data->start_class,
3790 (struct regnode_charclass_class*)scan);
3792 cl_or(pRExC_state, data->start_class,
3793 (struct regnode_charclass_class*)scan);
3796 if (flags & SCF_DO_STCLASS_AND) {
3797 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3798 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3799 if (OP(scan) == ALNUMU) {
3800 for (value = 0; value < 256; value++) {
3801 if (!isWORDCHAR_L1(value)) {
3802 ANYOF_BITMAP_CLEAR(data->start_class, value);
3806 for (value = 0; value < 256; value++) {
3807 if (!isALNUM(value)) {
3808 ANYOF_BITMAP_CLEAR(data->start_class, value);
3815 if (data->start_class->flags & ANYOF_LOCALE)
3816 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3818 /* Even if under locale, set the bits for non-locale
3819 * in case it isn't a true locale-node. This will
3820 * create false positives if it truly is locale */
3821 if (OP(scan) == ALNUMU) {
3822 for (value = 0; value < 256; value++) {
3823 if (isWORDCHAR_L1(value)) {
3824 ANYOF_BITMAP_SET(data->start_class, value);
3828 for (value = 0; value < 256; value++) {
3829 if (isALNUM(value)) {
3830 ANYOF_BITMAP_SET(data->start_class, value);
3837 if (flags & SCF_DO_STCLASS_AND) {
3838 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3839 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3840 if (OP(scan) == NALNUMU) {
3841 for (value = 0; value < 256; value++) {
3842 if (isWORDCHAR_L1(value)) {
3843 ANYOF_BITMAP_CLEAR(data->start_class, value);
3847 for (value = 0; value < 256; value++) {
3848 if (isALNUM(value)) {
3849 ANYOF_BITMAP_CLEAR(data->start_class, value);
3856 if (data->start_class->flags & ANYOF_LOCALE)
3857 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3859 /* Even if under locale, set the bits for non-locale in
3860 * case it isn't a true locale-node. This will create
3861 * false positives if it truly is locale */
3862 if (OP(scan) == NALNUMU) {
3863 for (value = 0; value < 256; value++) {
3864 if (! isWORDCHAR_L1(value)) {
3865 ANYOF_BITMAP_SET(data->start_class, value);
3869 for (value = 0; value < 256; value++) {
3870 if (! isALNUM(value)) {
3871 ANYOF_BITMAP_SET(data->start_class, value);
3878 if (flags & SCF_DO_STCLASS_AND) {
3879 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3880 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3881 if (OP(scan) == SPACEU) {
3882 for (value = 0; value < 256; value++) {
3883 if (!isSPACE_L1(value)) {
3884 ANYOF_BITMAP_CLEAR(data->start_class, value);
3888 for (value = 0; value < 256; value++) {
3889 if (!isSPACE(value)) {
3890 ANYOF_BITMAP_CLEAR(data->start_class, value);
3897 if (data->start_class->flags & ANYOF_LOCALE) {
3898 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3900 if (OP(scan) == SPACEU) {
3901 for (value = 0; value < 256; value++) {
3902 if (isSPACE_L1(value)) {
3903 ANYOF_BITMAP_SET(data->start_class, value);
3907 for (value = 0; value < 256; value++) {
3908 if (isSPACE(value)) {
3909 ANYOF_BITMAP_SET(data->start_class, value);
3916 if (flags & SCF_DO_STCLASS_AND) {
3917 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3918 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3919 if (OP(scan) == NSPACEU) {
3920 for (value = 0; value < 256; value++) {
3921 if (isSPACE_L1(value)) {
3922 ANYOF_BITMAP_CLEAR(data->start_class, value);
3926 for (value = 0; value < 256; value++) {
3927 if (isSPACE(value)) {
3928 ANYOF_BITMAP_CLEAR(data->start_class, value);
3935 if (data->start_class->flags & ANYOF_LOCALE)
3936 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3937 if (OP(scan) == NSPACEU) {
3938 for (value = 0; value < 256; value++) {
3939 if (!isSPACE_L1(value)) {
3940 ANYOF_BITMAP_SET(data->start_class, value);
3945 for (value = 0; value < 256; value++) {
3946 if (!isSPACE(value)) {
3947 ANYOF_BITMAP_SET(data->start_class, value);
3954 if (flags & SCF_DO_STCLASS_AND) {
3955 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3956 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3957 for (value = 0; value < 256; value++)
3958 if (!isDIGIT(value))
3959 ANYOF_BITMAP_CLEAR(data->start_class, value);
3963 if (data->start_class->flags & ANYOF_LOCALE)
3964 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3965 for (value = 0; value < 256; value++)
3967 ANYOF_BITMAP_SET(data->start_class, value);
3971 if (flags & SCF_DO_STCLASS_AND) {
3972 if (!(data->start_class->flags & ANYOF_LOCALE))
3973 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3974 for (value = 0; value < 256; value++)
3976 ANYOF_BITMAP_CLEAR(data->start_class, value);
3979 if (data->start_class->flags & ANYOF_LOCALE)
3980 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3981 for (value = 0; value < 256; value++)
3982 if (!isDIGIT(value))
3983 ANYOF_BITMAP_SET(data->start_class, value);
3986 CASE_SYNST_FNC(VERTWS);
3987 CASE_SYNST_FNC(HORIZWS);
3990 if (flags & SCF_DO_STCLASS_OR)
3991 cl_and(data->start_class, and_withp);
3992 flags &= ~SCF_DO_STCLASS;
3995 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3996 data->flags |= (OP(scan) == MEOL
4000 else if ( PL_regkind[OP(scan)] == BRANCHJ
4001 /* Lookbehind, or need to calculate parens/evals/stclass: */
4002 && (scan->flags || data || (flags & SCF_DO_STCLASS))
4003 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
4004 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4005 || OP(scan) == UNLESSM )
4007 /* Negative Lookahead/lookbehind
4008 In this case we can't do fixed string optimisation.
4011 I32 deltanext, minnext, fake = 0;
4013 struct regnode_charclass_class intrnl;
4016 data_fake.flags = 0;
4018 data_fake.whilem_c = data->whilem_c;
4019 data_fake.last_closep = data->last_closep;
4022 data_fake.last_closep = &fake;
4023 data_fake.pos_delta = delta;
4024 if ( flags & SCF_DO_STCLASS && !scan->flags
4025 && OP(scan) == IFMATCH ) { /* Lookahead */
4026 cl_init(pRExC_state, &intrnl);
4027 data_fake.start_class = &intrnl;
4028 f |= SCF_DO_STCLASS_AND;
4030 if (flags & SCF_WHILEM_VISITED_POS)
4031 f |= SCF_WHILEM_VISITED_POS;
4032 next = regnext(scan);
4033 nscan = NEXTOPER(NEXTOPER(scan));
4034 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
4035 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
4038 FAIL("Variable length lookbehind not implemented");
4040 else if (minnext > (I32)U8_MAX) {
4041 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4043 scan->flags = (U8)minnext;
4046 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4048 if (data_fake.flags & SF_HAS_EVAL)
4049 data->flags |= SF_HAS_EVAL;
4050 data->whilem_c = data_fake.whilem_c;
4052 if (f & SCF_DO_STCLASS_AND) {
4053 if (flags & SCF_DO_STCLASS_OR) {
4054 /* OR before, AND after: ideally we would recurse with
4055 * data_fake to get the AND applied by study of the
4056 * remainder of the pattern, and then derecurse;
4057 * *** HACK *** for now just treat as "no information".
4058 * See [perl #56690].
4060 cl_init(pRExC_state, data->start_class);
4062 /* AND before and after: combine and continue */
4063 const int was = (data->start_class->flags & ANYOF_EOS);
4065 cl_and(data->start_class, &intrnl);
4067 data->start_class->flags |= ANYOF_EOS;
4071 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4073 /* Positive Lookahead/lookbehind
4074 In this case we can do fixed string optimisation,
4075 but we must be careful about it. Note in the case of
4076 lookbehind the positions will be offset by the minimum
4077 length of the pattern, something we won't know about
4078 until after the recurse.
4080 I32 deltanext, fake = 0;
4082 struct regnode_charclass_class intrnl;
4084 /* We use SAVEFREEPV so that when the full compile
4085 is finished perl will clean up the allocated
4086 minlens when it's all done. This way we don't
4087 have to worry about freeing them when we know
4088 they wont be used, which would be a pain.
4091 Newx( minnextp, 1, I32 );
4092 SAVEFREEPV(minnextp);
4095 StructCopy(data, &data_fake, scan_data_t);
4096 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
4099 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
4100 data_fake.last_found=newSVsv(data->last_found);
4104 data_fake.last_closep = &fake;
4105 data_fake.flags = 0;
4106 data_fake.pos_delta = delta;
4108 data_fake.flags |= SF_IS_INF;
4109 if ( flags & SCF_DO_STCLASS && !scan->flags
4110 && OP(scan) == IFMATCH ) { /* Lookahead */
4111 cl_init(pRExC_state, &intrnl);
4112 data_fake.start_class = &intrnl;
4113 f |= SCF_DO_STCLASS_AND;
4115 if (flags & SCF_WHILEM_VISITED_POS)
4116 f |= SCF_WHILEM_VISITED_POS;
4117 next = regnext(scan);
4118 nscan = NEXTOPER(NEXTOPER(scan));
4120 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
4121 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
4124 FAIL("Variable length lookbehind not implemented");
4126 else if (*minnextp > (I32)U8_MAX) {
4127 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4129 scan->flags = (U8)*minnextp;
4134 if (f & SCF_DO_STCLASS_AND) {
4135 const int was = (data->start_class->flags & ANYOF_EOS);
4137 cl_and(data->start_class, &intrnl);
4139 data->start_class->flags |= ANYOF_EOS;
4142 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4144 if (data_fake.flags & SF_HAS_EVAL)
4145 data->flags |= SF_HAS_EVAL;
4146 data->whilem_c = data_fake.whilem_c;
4147 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
4148 if (RExC_rx->minlen<*minnextp)
4149 RExC_rx->minlen=*minnextp;
4150 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
4151 SvREFCNT_dec(data_fake.last_found);
4153 if ( data_fake.minlen_fixed != minlenp )
4155 data->offset_fixed= data_fake.offset_fixed;
4156 data->minlen_fixed= data_fake.minlen_fixed;
4157 data->lookbehind_fixed+= scan->flags;
4159 if ( data_fake.minlen_float != minlenp )
4161 data->minlen_float= data_fake.minlen_float;
4162 data->offset_float_min=data_fake.offset_float_min;
4163 data->offset_float_max=data_fake.offset_float_max;
4164 data->lookbehind_float+= scan->flags;
4173 else if (OP(scan) == OPEN) {
4174 if (stopparen != (I32)ARG(scan))
4177 else if (OP(scan) == CLOSE) {
4178 if (stopparen == (I32)ARG(scan)) {
4181 if ((I32)ARG(scan) == is_par) {
4182 next = regnext(scan);
4184 if ( next && (OP(next) != WHILEM) && next < last)
4185 is_par = 0; /* Disable optimization */
4188 *(data->last_closep) = ARG(scan);
4190 else if (OP(scan) == EVAL) {
4192 data->flags |= SF_HAS_EVAL;
4194 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
4195 if (flags & SCF_DO_SUBSTR) {
4196 SCAN_COMMIT(pRExC_state,data,minlenp);
4197 flags &= ~SCF_DO_SUBSTR;
4199 if (data && OP(scan)==ACCEPT) {
4200 data->flags |= SCF_SEEN_ACCEPT;
4205 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
4207 if (flags & SCF_DO_SUBSTR) {
4208 SCAN_COMMIT(pRExC_state,data,minlenp);
4209 data->longest = &(data->longest_float);
4211 is_inf = is_inf_internal = 1;
4212 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
4213 cl_anything(pRExC_state, data->start_class);
4214 flags &= ~SCF_DO_STCLASS;
4216 else if (OP(scan) == GPOS) {
4217 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
4218 !(delta || is_inf || (data && data->pos_delta)))
4220 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
4221 RExC_rx->extflags |= RXf_ANCH_GPOS;
4222 if (RExC_rx->gofs < (U32)min)
4223 RExC_rx->gofs = min;
4225 RExC_rx->extflags |= RXf_GPOS_FLOAT;
4229 #ifdef TRIE_STUDY_OPT
4230 #ifdef FULL_TRIE_STUDY
4231 else if (PL_regkind[OP(scan)] == TRIE) {
4232 /* NOTE - There is similar code to this block above for handling
4233 BRANCH nodes on the initial study. If you change stuff here
4235 regnode *trie_node= scan;
4236 regnode *tail= regnext(scan);
4237 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4238 I32 max1 = 0, min1 = I32_MAX;
4239 struct regnode_charclass_class accum;
4241 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
4242 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
4243 if (flags & SCF_DO_STCLASS)
4244 cl_init_zero(pRExC_state, &accum);
4250 const regnode *nextbranch= NULL;
4253 for ( word=1 ; word <= trie->wordcount ; word++)
4255 I32 deltanext=0, minnext=0, f = 0, fake;
4256 struct regnode_charclass_class this_class;
4258 data_fake.flags = 0;
4260 data_fake.whilem_c = data->whilem_c;
4261 data_fake.last_closep = data->last_closep;
4264 data_fake.last_closep = &fake;
4265 data_fake.pos_delta = delta;
4266 if (flags & SCF_DO_STCLASS) {
4267 cl_init(pRExC_state, &this_class);
4268 data_fake.start_class = &this_class;
4269 f = SCF_DO_STCLASS_AND;
4271 if (flags & SCF_WHILEM_VISITED_POS)
4272 f |= SCF_WHILEM_VISITED_POS;
4274 if (trie->jump[word]) {
4276 nextbranch = trie_node + trie->jump[0];
4277 scan= trie_node + trie->jump[word];
4278 /* We go from the jump point to the branch that follows
4279 it. Note this means we need the vestigal unused branches
4280 even though they arent otherwise used.
4282 minnext = study_chunk(pRExC_state, &scan, minlenp,
4283 &deltanext, (regnode *)nextbranch, &data_fake,
4284 stopparen, recursed, NULL, f,depth+1);
4286 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4287 nextbranch= regnext((regnode*)nextbranch);
4289 if (min1 > (I32)(minnext + trie->minlen))
4290 min1 = minnext + trie->minlen;
4291 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4292 max1 = minnext + deltanext + trie->maxlen;
4293 if (deltanext == I32_MAX)
4294 is_inf = is_inf_internal = 1;
4296 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4298 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4299 if ( stopmin > min + min1)
4300 stopmin = min + min1;
4301 flags &= ~SCF_DO_SUBSTR;
4303 data->flags |= SCF_SEEN_ACCEPT;
4306 if (data_fake.flags & SF_HAS_EVAL)
4307 data->flags |= SF_HAS_EVAL;
4308 data->whilem_c = data_fake.whilem_c;
4310 if (flags & SCF_DO_STCLASS)
4311 cl_or(pRExC_state, &accum, &this_class);
4314 if (flags & SCF_DO_SUBSTR) {
4315 data->pos_min += min1;
4316 data->pos_delta += max1 - min1;
4317 if (max1 != min1 || is_inf)
4318 data->longest = &(data->longest_float);
4321 delta += max1 - min1;
4322 if (flags & SCF_DO_STCLASS_OR) {
4323 cl_or(pRExC_state, data->start_class, &accum);
4325 cl_and(data->start_class, and_withp);
4326 flags &= ~SCF_DO_STCLASS;
4329 else if (flags & SCF_DO_STCLASS_AND) {
4331 cl_and(data->start_class, &accum);
4332 flags &= ~SCF_DO_STCLASS;
4335 /* Switch to OR mode: cache the old value of
4336 * data->start_class */
4338 StructCopy(data->start_class, and_withp,
4339 struct regnode_charclass_class);
4340 flags &= ~SCF_DO_STCLASS_AND;
4341 StructCopy(&accum, data->start_class,
4342 struct regnode_charclass_class);
4343 flags |= SCF_DO_STCLASS_OR;
4344 data->start_class->flags |= ANYOF_EOS;
4351 else if (PL_regkind[OP(scan)] == TRIE) {
4352 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4355 min += trie->minlen;
4356 delta += (trie->maxlen - trie->minlen);
4357 flags &= ~SCF_DO_STCLASS; /* xxx */
4358 if (flags & SCF_DO_SUBSTR) {
4359 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4360 data->pos_min += trie->minlen;
4361 data->pos_delta += (trie->maxlen - trie->minlen);
4362 if (trie->maxlen != trie->minlen)
4363 data->longest = &(data->longest_float);
4365 if (trie->jump) /* no more substrings -- for now /grr*/
4366 flags &= ~SCF_DO_SUBSTR;
4368 #endif /* old or new */
4369 #endif /* TRIE_STUDY_OPT */
4371 /* Else: zero-length, ignore. */
4372 scan = regnext(scan);
4377 stopparen = frame->stop;
4378 frame = frame->prev;
4379 goto fake_study_recurse;
4384 DEBUG_STUDYDATA("pre-fin:",data,depth);
4387 *deltap = is_inf_internal ? I32_MAX : delta;
4388 if (flags & SCF_DO_SUBSTR && is_inf)
4389 data->pos_delta = I32_MAX - data->pos_min;
4390 if (is_par > (I32)U8_MAX)
4392 if (is_par && pars==1 && data) {
4393 data->flags |= SF_IN_PAR;
4394 data->flags &= ~SF_HAS_PAR;
4396 else if (pars && data) {
4397 data->flags |= SF_HAS_PAR;
4398 data->flags &= ~SF_IN_PAR;
4400 if (flags & SCF_DO_STCLASS_OR)
4401 cl_and(data->start_class, and_withp);
4402 if (flags & SCF_TRIE_RESTUDY)
4403 data->flags |= SCF_TRIE_RESTUDY;
4405 DEBUG_STUDYDATA("post-fin:",data,depth);
4407 return min < stopmin ? min : stopmin;
4411 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4413 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4415 PERL_ARGS_ASSERT_ADD_DATA;
4417 Renewc(RExC_rxi->data,
4418 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4419 char, struct reg_data);
4421 Renew(RExC_rxi->data->what, count + n, U8);
4423 Newx(RExC_rxi->data->what, n, U8);
4424 RExC_rxi->data->count = count + n;
4425 Copy(s, RExC_rxi->data->what + count, n, U8);
4429 /*XXX: todo make this not included in a non debugging perl */
4430 #ifndef PERL_IN_XSUB_RE
4432 Perl_reginitcolors(pTHX)
4435 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4437 char *t = savepv(s);
4441 t = strchr(t, '\t');
4447 PL_colors[i] = t = (char *)"";
4452 PL_colors[i++] = (char *)"";
4459 #ifdef TRIE_STUDY_OPT
4460 #define CHECK_RESTUDY_GOTO \
4462 (data.flags & SCF_TRIE_RESTUDY) \
4466 #define CHECK_RESTUDY_GOTO
4470 - pregcomp - compile a regular expression into internal code
4472 * We can't allocate space until we know how big the compiled form will be,
4473 * but we can't compile it (and thus know how big it is) until we've got a
4474 * place to put the code. So we cheat: we compile it twice, once with code
4475 * generation turned off and size counting turned on, and once "for real".
4476 * This also means that we don't allocate space until we are sure that the
4477 * thing really will compile successfully, and we never have to move the
4478 * code and thus invalidate pointers into it. (Note that it has to be in
4479 * one piece because free() must be able to free it all.) [NB: not true in perl]
4481 * Beware that the optimization-preparation code in here knows about some
4482 * of the structure of the compiled regexp. [I'll say.]
4487 #ifndef PERL_IN_XSUB_RE
4488 #define RE_ENGINE_PTR &PL_core_reg_engine
4490 extern const struct regexp_engine my_reg_engine;
4491 #define RE_ENGINE_PTR &my_reg_engine
4494 #ifndef PERL_IN_XSUB_RE
4496 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4499 HV * const table = GvHV(PL_hintgv);
4501 PERL_ARGS_ASSERT_PREGCOMP;
4503 /* Dispatch a request to compile a regexp to correct
4506 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4507 GET_RE_DEBUG_FLAGS_DECL;
4508 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4509 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4511 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4514 return CALLREGCOMP_ENG(eng, pattern, flags);
4517 return Perl_re_compile(aTHX_ pattern, flags);
4522 Perl_re_compile(pTHX_ SV * const pattern, U32 orig_pm_flags)
4527 register regexp_internal *ri;
4536 /* these are all flags - maybe they should be turned
4537 * into a single int with different bit masks */
4538 I32 sawlookahead = 0;
4541 bool used_setjump = FALSE;
4542 regex_charset initial_charset = get_regex_charset(orig_pm_flags);
4547 RExC_state_t RExC_state;
4548 RExC_state_t * const pRExC_state = &RExC_state;
4549 #ifdef TRIE_STUDY_OPT
4551 RExC_state_t copyRExC_state;
4553 GET_RE_DEBUG_FLAGS_DECL;
4555 PERL_ARGS_ASSERT_RE_COMPILE;
4557 DEBUG_r(if (!PL_colorset) reginitcolors());
4559 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4560 RExC_uni_semantics = 0;
4561 RExC_contains_locale = 0;
4563 /****************** LONG JUMP TARGET HERE***********************/
4564 /* Longjmp back to here if have to switch in midstream to utf8 */
4565 if (! RExC_orig_utf8) {
4566 JMPENV_PUSH(jump_ret);
4567 used_setjump = TRUE;
4570 if (jump_ret == 0) { /* First time through */
4571 exp = SvPV(pattern, plen);
4573 /* ignore the utf8ness if the pattern is 0 length */
4575 RExC_utf8 = RExC_orig_utf8 = 0;
4579 SV *dsv= sv_newmortal();
4580 RE_PV_QUOTED_DECL(s, RExC_utf8,
4581 dsv, exp, plen, 60);
4582 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4583 PL_colors[4],PL_colors[5],s);
4586 else { /* longjumped back */
4589 /* If the cause for the longjmp was other than changing to utf8, pop
4590 * our own setjmp, and longjmp to the correct handler */
4591 if (jump_ret != UTF8_LONGJMP) {
4593 JMPENV_JUMP(jump_ret);
4598 /* It's possible to write a regexp in ascii that represents Unicode
4599 codepoints outside of the byte range, such as via \x{100}. If we
4600 detect such a sequence we have to convert the entire pattern to utf8
4601 and then recompile, as our sizing calculation will have been based
4602 on 1 byte == 1 character, but we will need to use utf8 to encode
4603 at least some part of the pattern, and therefore must convert the whole
4606 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4607 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4608 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4610 RExC_orig_utf8 = RExC_utf8 = 1;
4614 #ifdef TRIE_STUDY_OPT
4618 pm_flags = orig_pm_flags;
4620 if (initial_charset == REGEX_LOCALE_CHARSET) {
4621 RExC_contains_locale = 1;
4623 else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) {
4625 /* Set to use unicode semantics if the pattern is in utf8 and has the
4626 * 'depends' charset specified, as it means unicode when utf8 */
4627 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4631 RExC_flags = pm_flags;
4635 RExC_in_lookbehind = 0;
4636 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4637 RExC_seen_evals = 0;
4639 RExC_override_recoding = 0;
4641 /* First pass: determine size, legality. */
4649 RExC_emit = &PL_regdummy;
4650 RExC_whilem_seen = 0;
4651 RExC_open_parens = NULL;
4652 RExC_close_parens = NULL;
4654 RExC_paren_names = NULL;
4656 RExC_paren_name_list = NULL;
4658 RExC_recurse = NULL;
4659 RExC_recurse_count = 0;
4661 #if 0 /* REGC() is (currently) a NOP at the first pass.
4662 * Clever compilers notice this and complain. --jhi */
4663 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4665 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4666 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4667 RExC_precomp = NULL;
4671 /* Here, finished first pass. Get rid of any added setjmp */
4677 PerlIO_printf(Perl_debug_log,
4678 "Required size %"IVdf" nodes\n"
4679 "Starting second pass (creation)\n",
4682 RExC_lastparse=NULL;
4685 /* The first pass could have found things that force Unicode semantics */
4686 if ((RExC_utf8 || RExC_uni_semantics)
4687 && get_regex_charset(pm_flags) == REGEX_DEPENDS_CHARSET)
4689 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4692 /* Small enough for pointer-storage convention?
4693 If extralen==0, this means that we will not need long jumps. */
4694 if (RExC_size >= 0x10000L && RExC_extralen)
4695 RExC_size += RExC_extralen;
4698 if (RExC_whilem_seen > 15)
4699 RExC_whilem_seen = 15;
4701 /* Allocate space and zero-initialize. Note, the two step process
4702 of zeroing when in debug mode, thus anything assigned has to
4703 happen after that */
4704 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4705 r = (struct regexp*)SvANY(rx);
4706 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4707 char, regexp_internal);
4708 if ( r == NULL || ri == NULL )
4709 FAIL("Regexp out of space");
4711 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4712 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4714 /* bulk initialize base fields with 0. */
4715 Zero(ri, sizeof(regexp_internal), char);
4718 /* non-zero initialization begins here */
4720 r->engine= RE_ENGINE_PTR;
4721 r->extflags = pm_flags;
4723 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4724 bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET);
4726 /* The caret is output if there are any defaults: if not all the STD
4727 * flags are set, or if no character set specifier is needed */
4729 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4731 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4732 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4733 >> RXf_PMf_STD_PMMOD_SHIFT);
4734 const char *fptr = STD_PAT_MODS; /*"msix"*/
4736 /* Allocate for the worst case, which is all the std flags are turned
4737 * on. If more precision is desired, we could do a population count of
4738 * the flags set. This could be done with a small lookup table, or by
4739 * shifting, masking and adding, or even, when available, assembly
4740 * language for a machine-language population count.
4741 * We never output a minus, as all those are defaults, so are
4742 * covered by the caret */
4743 const STRLEN wraplen = plen + has_p + has_runon
4744 + has_default /* If needs a caret */
4746 /* If needs a character set specifier */
4747 + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0)
4748 + (sizeof(STD_PAT_MODS) - 1)
4749 + (sizeof("(?:)") - 1);
4751 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4753 SvFLAGS(rx) |= SvUTF8(pattern);
4756 /* If a default, cover it using the caret */
4758 *p++= DEFAULT_PAT_MOD;
4762 const char* const name = get_regex_charset_name(r->extflags, &len);
4763 Copy(name, p, len, char);
4767 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4770 while((ch = *fptr++)) {
4778 Copy(RExC_precomp, p, plen, char);
4779 assert ((RX_WRAPPED(rx) - p) < 16);
4780 r->pre_prefix = p - RX_WRAPPED(rx);
4786 SvCUR_set(rx, p - SvPVX_const(rx));
4790 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4792 if (RExC_seen & REG_SEEN_RECURSE) {
4793 Newxz(RExC_open_parens, RExC_npar,regnode *);
4794 SAVEFREEPV(RExC_open_parens);
4795 Newxz(RExC_close_parens,RExC_npar,regnode *);
4796 SAVEFREEPV(RExC_close_parens);
4799 /* Useful during FAIL. */
4800 #ifdef RE_TRACK_PATTERN_OFFSETS
4801 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4802 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4803 "%s %"UVuf" bytes for offset annotations.\n",
4804 ri->u.offsets ? "Got" : "Couldn't get",
4805 (UV)((2*RExC_size+1) * sizeof(U32))));
4807 SetProgLen(ri,RExC_size);
4811 REH_CALL_COMP_BEGIN_HOOK(pRExC_state->rx);
4813 /* Second pass: emit code. */
4814 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4819 RExC_emit_start = ri->program;
4820 RExC_emit = ri->program;
4821 RExC_emit_bound = ri->program + RExC_size + 1;
4823 /* Store the count of eval-groups for security checks: */
4824 RExC_rx->seen_evals = RExC_seen_evals;
4825 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4826 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4830 /* XXXX To minimize changes to RE engine we always allocate
4831 3-units-long substrs field. */
4832 Newx(r->substrs, 1, struct reg_substr_data);
4833 if (RExC_recurse_count) {
4834 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4835 SAVEFREEPV(RExC_recurse);
4839 r->minlen = minlen = sawlookahead = sawplus = sawopen = 0;
4840 Zero(r->substrs, 1, struct reg_substr_data);
4842 #ifdef TRIE_STUDY_OPT
4844 StructCopy(&zero_scan_data, &data, scan_data_t);
4845 copyRExC_state = RExC_state;
4848 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4850 RExC_state = copyRExC_state;
4851 if (seen & REG_TOP_LEVEL_BRANCHES)
4852 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4854 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4855 if (data.last_found) {
4856 SvREFCNT_dec(data.longest_fixed);
4857 SvREFCNT_dec(data.longest_float);
4858 SvREFCNT_dec(data.last_found);
4860 StructCopy(&zero_scan_data, &data, scan_data_t);
4863 StructCopy(&zero_scan_data, &data, scan_data_t);
4866 /* Dig out information for optimizations. */
4867 r->extflags = RExC_flags; /* was pm_op */
4868 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4871 SvUTF8_on(rx); /* Unicode in it? */
4872 ri->regstclass = NULL;
4873 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4874 r->intflags |= PREGf_NAUGHTY;
4875 scan = ri->program + 1; /* First BRANCH. */
4877 /* testing for BRANCH here tells us whether there is "must appear"
4878 data in the pattern. If there is then we can use it for optimisations */
4879 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4881 STRLEN longest_float_length, longest_fixed_length;
4882 struct regnode_charclass_class ch_class; /* pointed to by data */
4884 I32 last_close = 0; /* pointed to by data */
4885 regnode *first= scan;
4886 regnode *first_next= regnext(first);
4888 * Skip introductions and multiplicators >= 1
4889 * so that we can extract the 'meat' of the pattern that must
4890 * match in the large if() sequence following.
4891 * NOTE that EXACT is NOT covered here, as it is normally
4892 * picked up by the optimiser separately.
4894 * This is unfortunate as the optimiser isnt handling lookahead
4895 * properly currently.
4898 while ((OP(first) == OPEN && (sawopen = 1)) ||
4899 /* An OR of *one* alternative - should not happen now. */
4900 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4901 /* for now we can't handle lookbehind IFMATCH*/
4902 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
4903 (OP(first) == PLUS) ||
4904 (OP(first) == MINMOD) ||
4905 /* An {n,m} with n>0 */
4906 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4907 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4910 * the only op that could be a regnode is PLUS, all the rest
4911 * will be regnode_1 or regnode_2.
4914 if (OP(first) == PLUS)
4917 first += regarglen[OP(first)];
4919 first = NEXTOPER(first);
4920 first_next= regnext(first);
4923 /* Starting-point info. */
4925 DEBUG_PEEP("first:",first,0);
4926 /* Ignore EXACT as we deal with it later. */
4927 if (PL_regkind[OP(first)] == EXACT) {
4928 if (OP(first) == EXACT)
4929 NOOP; /* Empty, get anchored substr later. */
4931 ri->regstclass = first;
4934 else if (PL_regkind[OP(first)] == TRIE &&
4935 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4938 /* this can happen only on restudy */
4939 if ( OP(first) == TRIE ) {
4940 struct regnode_1 *trieop = (struct regnode_1 *)
4941 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4942 StructCopy(first,trieop,struct regnode_1);
4943 trie_op=(regnode *)trieop;
4945 struct regnode_charclass *trieop = (struct regnode_charclass *)
4946 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4947 StructCopy(first,trieop,struct regnode_charclass);
4948 trie_op=(regnode *)trieop;
4951 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4952 ri->regstclass = trie_op;
4955 else if (REGNODE_SIMPLE(OP(first)))
4956 ri->regstclass = first;
4957 else if (PL_regkind[OP(first)] == BOUND ||
4958 PL_regkind[OP(first)] == NBOUND)
4959 ri->regstclass = first;
4960 else if (PL_regkind[OP(first)] == BOL) {
4961 r->extflags |= (OP(first) == MBOL
4963 : (OP(first) == SBOL
4966 first = NEXTOPER(first);
4969 else if (OP(first) == GPOS) {
4970 r->extflags |= RXf_ANCH_GPOS;
4971 first = NEXTOPER(first);
4974 else if ((!sawopen || !RExC_sawback) &&
4975 (OP(first) == STAR &&
4976 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4977 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4979 /* turn .* into ^.* with an implied $*=1 */
4981 (OP(NEXTOPER(first)) == REG_ANY)
4984 r->extflags |= type;
4985 r->intflags |= PREGf_IMPLICIT;
4986 first = NEXTOPER(first);
4989 if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback)
4990 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4991 /* x+ must match at the 1st pos of run of x's */
4992 r->intflags |= PREGf_SKIP;
4994 /* Scan is after the zeroth branch, first is atomic matcher. */
4995 #ifdef TRIE_STUDY_OPT
4998 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4999 (IV)(first - scan + 1))
5003 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
5004 (IV)(first - scan + 1))
5010 * If there's something expensive in the r.e., find the
5011 * longest literal string that must appear and make it the
5012 * regmust. Resolve ties in favor of later strings, since
5013 * the regstart check works with the beginning of the r.e.
5014 * and avoiding duplication strengthens checking. Not a
5015 * strong reason, but sufficient in the absence of others.
5016 * [Now we resolve ties in favor of the earlier string if
5017 * it happens that c_offset_min has been invalidated, since the
5018 * earlier string may buy us something the later one won't.]
5021 data.longest_fixed = newSVpvs("");
5022 data.longest_float = newSVpvs("");
5023 data.last_found = newSVpvs("");
5024 data.longest = &(data.longest_fixed);
5026 if (!ri->regstclass) {
5027 cl_init(pRExC_state, &ch_class);
5028 data.start_class = &ch_class;
5029 stclass_flag = SCF_DO_STCLASS_AND;
5030 } else /* XXXX Check for BOUND? */
5032 data.last_closep = &last_close;
5034 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
5035 &data, -1, NULL, NULL,
5036 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
5042 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
5043 && data.last_start_min == 0 && data.last_end > 0
5044 && !RExC_seen_zerolen
5045 && !(RExC_seen & REG_SEEN_VERBARG)
5046 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
5047 r->extflags |= RXf_CHECK_ALL;
5048 scan_commit(pRExC_state, &data,&minlen,0);
5049 SvREFCNT_dec(data.last_found);
5051 /* Note that code very similar to this but for anchored string
5052 follows immediately below, changes may need to be made to both.
5055 longest_float_length = CHR_SVLEN(data.longest_float);
5056 if (longest_float_length
5057 || (data.flags & SF_FL_BEFORE_EOL
5058 && (!(data.flags & SF_FL_BEFORE_MEOL)
5059 || (RExC_flags & RXf_PMf_MULTILINE))))
5063 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
5064 && data.offset_fixed == data.offset_float_min
5065 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
5066 goto remove_float; /* As in (a)+. */
5068 /* copy the information about the longest float from the reg_scan_data
5069 over to the program. */
5070 if (SvUTF8(data.longest_float)) {
5071 r->float_utf8 = data.longest_float;
5072 r->float_substr = NULL;
5074 r->float_substr = data.longest_float;
5075 r->float_utf8 = NULL;
5077 /* float_end_shift is how many chars that must be matched that
5078 follow this item. We calculate it ahead of time as once the
5079 lookbehind offset is added in we lose the ability to correctly
5081 ml = data.minlen_float ? *(data.minlen_float)
5082 : (I32)longest_float_length;
5083 r->float_end_shift = ml - data.offset_float_min
5084 - longest_float_length + (SvTAIL(data.longest_float) != 0)
5085 + data.lookbehind_float;
5086 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
5087 r->float_max_offset = data.offset_float_max;
5088 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
5089 r->float_max_offset -= data.lookbehind_float;
5091 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
5092 && (!(data.flags & SF_FL_BEFORE_MEOL)
5093 || (RExC_flags & RXf_PMf_MULTILINE)));
5094 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
5098 r->float_substr = r->float_utf8 = NULL;
5099 SvREFCNT_dec(data.longest_float);
5100 longest_float_length = 0;
5103 /* Note that code very similar to this but for floating string
5104 is immediately above, changes may need to be made to both.
5107 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
5108 if (longest_fixed_length
5109 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
5110 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5111 || (RExC_flags & RXf_PMf_MULTILINE))))
5115 /* copy the information about the longest fixed
5116 from the reg_scan_data over to the program. */
5117 if (SvUTF8(data.longest_fixed)) {
5118 r->anchored_utf8 = data.longest_fixed;
5119 r->anchored_substr = NULL;
5121 r->anchored_substr = data.longest_fixed;
5122 r->anchored_utf8 = NULL;
5124 /* fixed_end_shift is how many chars that must be matched that
5125 follow this item. We calculate it ahead of time as once the
5126 lookbehind offset is added in we lose the ability to correctly
5128 ml = data.minlen_fixed ? *(data.minlen_fixed)
5129 : (I32)longest_fixed_length;
5130 r->anchored_end_shift = ml - data.offset_fixed
5131 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
5132 + data.lookbehind_fixed;
5133 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
5135 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
5136 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5137 || (RExC_flags & RXf_PMf_MULTILINE)));
5138 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
5141 r->anchored_substr = r->anchored_utf8 = NULL;
5142 SvREFCNT_dec(data.longest_fixed);
5143 longest_fixed_length = 0;
5146 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
5147 ri->regstclass = NULL;
5149 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
5151 && !(data.start_class->flags & ANYOF_EOS)
5152 && !cl_is_anything(data.start_class))
5154 const U32 n = add_data(pRExC_state, 1, "f");
5155 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5157 Newx(RExC_rxi->data->data[n], 1,
5158 struct regnode_charclass_class);
5159 StructCopy(data.start_class,
5160 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5161 struct regnode_charclass_class);
5162 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5163 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5164 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
5165 regprop(r, sv, (regnode*)data.start_class);
5166 PerlIO_printf(Perl_debug_log,
5167 "synthetic stclass \"%s\".\n",
5168 SvPVX_const(sv));});
5171 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
5172 if (longest_fixed_length > longest_float_length) {
5173 r->check_end_shift = r->anchored_end_shift;
5174 r->check_substr = r->anchored_substr;
5175 r->check_utf8 = r->anchored_utf8;
5176 r->check_offset_min = r->check_offset_max = r->anchored_offset;
5177 if (r->extflags & RXf_ANCH_SINGLE)
5178 r->extflags |= RXf_NOSCAN;
5181 r->check_end_shift = r->float_end_shift;
5182 r->check_substr = r->float_substr;
5183 r->check_utf8 = r->float_utf8;
5184 r->check_offset_min = r->float_min_offset;
5185 r->check_offset_max = r->float_max_offset;
5187 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
5188 This should be changed ASAP! */
5189 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
5190 r->extflags |= RXf_USE_INTUIT;
5191 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
5192 r->extflags |= RXf_INTUIT_TAIL;
5194 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
5195 if ( (STRLEN)minlen < longest_float_length )
5196 minlen= longest_float_length;
5197 if ( (STRLEN)minlen < longest_fixed_length )
5198 minlen= longest_fixed_length;
5202 /* Several toplevels. Best we can is to set minlen. */
5204 struct regnode_charclass_class ch_class;
5207 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
5209 scan = ri->program + 1;
5210 cl_init(pRExC_state, &ch_class);
5211 data.start_class = &ch_class;
5212 data.last_closep = &last_close;
5215 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
5216 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
5220 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
5221 = r->float_substr = r->float_utf8 = NULL;
5223 if (!(data.start_class->flags & ANYOF_EOS)
5224 && !cl_is_anything(data.start_class))
5226 const U32 n = add_data(pRExC_state, 1, "f");
5227 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5229 Newx(RExC_rxi->data->data[n], 1,
5230 struct regnode_charclass_class);
5231 StructCopy(data.start_class,
5232 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5233 struct regnode_charclass_class);
5234 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5235 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5236 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
5237 regprop(r, sv, (regnode*)data.start_class);
5238 PerlIO_printf(Perl_debug_log,
5239 "synthetic stclass \"%s\".\n",
5240 SvPVX_const(sv));});
5244 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
5245 the "real" pattern. */
5247 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
5248 (IV)minlen, (IV)r->minlen);
5250 r->minlenret = minlen;
5251 if (r->minlen < minlen)
5254 if (RExC_seen & REG_SEEN_GPOS)
5255 r->extflags |= RXf_GPOS_SEEN;
5256 if (RExC_seen & REG_SEEN_LOOKBEHIND)
5257 r->extflags |= RXf_LOOKBEHIND_SEEN;
5258 if (RExC_seen & REG_SEEN_EVAL)
5259 r->extflags |= RXf_EVAL_SEEN;
5260 if (RExC_seen & REG_SEEN_CANY)
5261 r->extflags |= RXf_CANY_SEEN;
5262 if (RExC_seen & REG_SEEN_VERBARG)
5263 r->intflags |= PREGf_VERBARG_SEEN;
5264 if (RExC_seen & REG_SEEN_CUTGROUP)
5265 r->intflags |= PREGf_CUTGROUP_SEEN;
5266 if (RExC_paren_names)
5267 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
5269 RXp_PAREN_NAMES(r) = NULL;
5271 #ifdef STUPID_PATTERN_CHECKS
5272 if (RX_PRELEN(rx) == 0)
5273 r->extflags |= RXf_NULL;
5274 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5275 /* XXX: this should happen BEFORE we compile */
5276 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5277 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
5278 r->extflags |= RXf_WHITE;
5279 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
5280 r->extflags |= RXf_START_ONLY;
5282 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5283 /* XXX: this should happen BEFORE we compile */
5284 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5286 regnode *first = ri->program + 1;
5289 if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END)
5290 r->extflags |= RXf_NULL;
5291 else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END)
5292 r->extflags |= RXf_START_ONLY;
5293 else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE
5294 && OP(regnext(first)) == END)
5295 r->extflags |= RXf_WHITE;
5299 if (RExC_paren_names) {
5300 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5301 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5304 ri->name_list_idx = 0;
5306 if (RExC_recurse_count) {
5307 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5308 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5309 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5312 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5313 /* assume we don't need to swap parens around before we match */
5316 PerlIO_printf(Perl_debug_log,"Final program:\n");
5319 #ifdef RE_TRACK_PATTERN_OFFSETS
5320 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5321 const U32 len = ri->u.offsets[0];
5323 GET_RE_DEBUG_FLAGS_DECL;
5324 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5325 for (i = 1; i <= len; i++) {
5326 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5327 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5328 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5330 PerlIO_printf(Perl_debug_log, "\n");
5336 #undef RE_ENGINE_PTR
5340 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5343 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5345 PERL_UNUSED_ARG(value);
5347 if (flags & RXapif_FETCH) {
5348 return reg_named_buff_fetch(rx, key, flags);
5349 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5350 Perl_croak_no_modify(aTHX);
5352 } else if (flags & RXapif_EXISTS) {
5353 return reg_named_buff_exists(rx, key, flags)
5356 } else if (flags & RXapif_REGNAMES) {
5357 return reg_named_buff_all(rx, flags);
5358 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5359 return reg_named_buff_scalar(rx, flags);
5361 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5367 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5370 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5371 PERL_UNUSED_ARG(lastkey);
5373 if (flags & RXapif_FIRSTKEY)
5374 return reg_named_buff_firstkey(rx, flags);
5375 else if (flags & RXapif_NEXTKEY)
5376 return reg_named_buff_nextkey(rx, flags);
5378 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5384 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5387 AV *retarray = NULL;
5389 struct regexp *const rx = (struct regexp *)SvANY(r);
5391 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5393 if (flags & RXapif_ALL)
5396 if (rx && RXp_PAREN_NAMES(rx)) {
5397 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5400 SV* sv_dat=HeVAL(he_str);
5401 I32 *nums=(I32*)SvPVX(sv_dat);
5402 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5403 if ((I32)(rx->nparens) >= nums[i]
5404 && rx->offs[nums[i]].start != -1
5405 && rx->offs[nums[i]].end != -1)
5408 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5412 ret = newSVsv(&PL_sv_undef);
5415 av_push(retarray, ret);
5418 return newRV_noinc(MUTABLE_SV(retarray));
5425 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5428 struct regexp *const rx = (struct regexp *)SvANY(r);
5430 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5432 if (rx && RXp_PAREN_NAMES(rx)) {
5433 if (flags & RXapif_ALL) {
5434 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5436 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5450 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5452 struct regexp *const rx = (struct regexp *)SvANY(r);
5454 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5456 if ( rx && RXp_PAREN_NAMES(rx) ) {
5457 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5459 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5466 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5468 struct regexp *const rx = (struct regexp *)SvANY(r);
5469 GET_RE_DEBUG_FLAGS_DECL;
5471 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5473 if (rx && RXp_PAREN_NAMES(rx)) {
5474 HV *hv = RXp_PAREN_NAMES(rx);
5476 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5479 SV* sv_dat = HeVAL(temphe);
5480 I32 *nums = (I32*)SvPVX(sv_dat);
5481 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5482 if ((I32)(rx->lastparen) >= nums[i] &&
5483 rx->offs[nums[i]].start != -1 &&
5484 rx->offs[nums[i]].end != -1)
5490 if (parno || flags & RXapif_ALL) {
5491 return newSVhek(HeKEY_hek(temphe));
5499 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5504 struct regexp *const rx = (struct regexp *)SvANY(r);
5506 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5508 if (rx && RXp_PAREN_NAMES(rx)) {
5509 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5510 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5511 } else if (flags & RXapif_ONE) {
5512 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5513 av = MUTABLE_AV(SvRV(ret));
5514 length = av_len(av);
5516 return newSViv(length + 1);
5518 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5522 return &PL_sv_undef;
5526 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5528 struct regexp *const rx = (struct regexp *)SvANY(r);
5531 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5533 if (rx && RXp_PAREN_NAMES(rx)) {
5534 HV *hv= RXp_PAREN_NAMES(rx);
5536 (void)hv_iterinit(hv);
5537 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5540 SV* sv_dat = HeVAL(temphe);
5541 I32 *nums = (I32*)SvPVX(sv_dat);
5542 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5543 if ((I32)(rx->lastparen) >= nums[i] &&
5544 rx->offs[nums[i]].start != -1 &&
5545 rx->offs[nums[i]].end != -1)
5551 if (parno || flags & RXapif_ALL) {
5552 av_push(av, newSVhek(HeKEY_hek(temphe)));
5557 return newRV_noinc(MUTABLE_SV(av));
5561 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5564 struct regexp *const rx = (struct regexp *)SvANY(r);
5569 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5572 sv_setsv(sv,&PL_sv_undef);
5576 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5578 i = rx->offs[0].start;
5582 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5584 s = rx->subbeg + rx->offs[0].end;
5585 i = rx->sublen - rx->offs[0].end;
5588 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5589 (s1 = rx->offs[paren].start) != -1 &&
5590 (t1 = rx->offs[paren].end) != -1)
5594 s = rx->subbeg + s1;
5596 sv_setsv(sv,&PL_sv_undef);
5599 assert(rx->sublen >= (s - rx->subbeg) + i );
5601 const int oldtainted = PL_tainted;
5603 sv_setpvn(sv, s, i);
5604 PL_tainted = oldtainted;
5605 if ( (rx->extflags & RXf_CANY_SEEN)
5606 ? (RXp_MATCH_UTF8(rx)
5607 && (!i || is_utf8_string((U8*)s, i)))
5608 : (RXp_MATCH_UTF8(rx)) )
5615 if (RXp_MATCH_TAINTED(rx)) {
5616 if (SvTYPE(sv) >= SVt_PVMG) {
5617 MAGIC* const mg = SvMAGIC(sv);
5620 SvMAGIC_set(sv, mg->mg_moremagic);
5622 if ((mgt = SvMAGIC(sv))) {
5623 mg->mg_moremagic = mgt;
5624 SvMAGIC_set(sv, mg);
5634 sv_setsv(sv,&PL_sv_undef);
5640 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5641 SV const * const value)
5643 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5645 PERL_UNUSED_ARG(rx);
5646 PERL_UNUSED_ARG(paren);
5647 PERL_UNUSED_ARG(value);
5650 Perl_croak_no_modify(aTHX);
5654 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5657 struct regexp *const rx = (struct regexp *)SvANY(r);
5661 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5663 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5665 /* $` / ${^PREMATCH} */
5666 case RX_BUFF_IDX_PREMATCH:
5667 if (rx->offs[0].start != -1) {
5668 i = rx->offs[0].start;
5676 /* $' / ${^POSTMATCH} */
5677 case RX_BUFF_IDX_POSTMATCH:
5678 if (rx->offs[0].end != -1) {
5679 i = rx->sublen - rx->offs[0].end;
5681 s1 = rx->offs[0].end;
5687 /* $& / ${^MATCH}, $1, $2, ... */
5689 if (paren <= (I32)rx->nparens &&
5690 (s1 = rx->offs[paren].start) != -1 &&
5691 (t1 = rx->offs[paren].end) != -1)
5696 if (ckWARN(WARN_UNINITIALIZED))
5697 report_uninit((const SV *)sv);
5702 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5703 const char * const s = rx->subbeg + s1;
5708 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5715 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5717 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5718 PERL_UNUSED_ARG(rx);
5722 return newSVpvs("Regexp");
5725 /* Scans the name of a named buffer from the pattern.
5726 * If flags is REG_RSN_RETURN_NULL returns null.
5727 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5728 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5729 * to the parsed name as looked up in the RExC_paren_names hash.
5730 * If there is an error throws a vFAIL().. type exception.
5733 #define REG_RSN_RETURN_NULL 0
5734 #define REG_RSN_RETURN_NAME 1
5735 #define REG_RSN_RETURN_DATA 2
5738 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5740 char *name_start = RExC_parse;
5742 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5744 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5745 /* skip IDFIRST by using do...while */
5748 RExC_parse += UTF8SKIP(RExC_parse);
5749 } while (isALNUM_utf8((U8*)RExC_parse));
5753 } while (isALNUM(*RExC_parse));
5758 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5759 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5760 if ( flags == REG_RSN_RETURN_NAME)
5762 else if (flags==REG_RSN_RETURN_DATA) {
5765 if ( ! sv_name ) /* should not happen*/
5766 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5767 if (RExC_paren_names)
5768 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5770 sv_dat = HeVAL(he_str);
5772 vFAIL("Reference to nonexistent named group");
5776 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5783 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5784 int rem=(int)(RExC_end - RExC_parse); \
5793 if (RExC_lastparse!=RExC_parse) \
5794 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5797 iscut ? "..." : "<" \
5800 PerlIO_printf(Perl_debug_log,"%16s",""); \
5803 num = RExC_size + 1; \
5805 num=REG_NODE_NUM(RExC_emit); \
5806 if (RExC_lastnum!=num) \
5807 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5809 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5810 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5811 (int)((depth*2)), "", \
5815 RExC_lastparse=RExC_parse; \
5820 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5821 DEBUG_PARSE_MSG((funcname)); \
5822 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5824 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5825 DEBUG_PARSE_MSG((funcname)); \
5826 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5829 /* This section of code defines the inversion list object and its methods. The
5830 * interfaces are highly subject to change, so as much as possible is static to
5831 * this file. An inversion list is here implemented as a malloc'd C array with
5832 * some added info. More will be coming when functionality is added later.
5834 * Some of the methods should always be private to the implementation, and some
5835 * should eventually be made public */
5837 #define INVLIST_INITIAL_LEN 10
5838 #define INVLIST_ARRAY_KEY "array"
5839 #define INVLIST_MAX_KEY "max"
5840 #define INVLIST_LEN_KEY "len"
5842 PERL_STATIC_INLINE UV*
5843 S_invlist_array(pTHX_ HV* const invlist)
5845 /* Returns the pointer to the inversion list's array. Every time the
5846 * length changes, this needs to be called in case malloc or realloc moved
5849 SV** list_ptr = hv_fetchs(invlist, INVLIST_ARRAY_KEY, FALSE);
5851 PERL_ARGS_ASSERT_INVLIST_ARRAY;
5853 if (list_ptr == NULL) {
5854 Perl_croak(aTHX_ "panic: inversion list without a '%s' element",
5858 return INT2PTR(UV *, SvUV(*list_ptr));
5861 PERL_STATIC_INLINE void
5862 S_invlist_set_array(pTHX_ HV* const invlist, const UV* const array)
5864 PERL_ARGS_ASSERT_INVLIST_SET_ARRAY;
5866 /* Sets the array stored in the inversion list to the memory beginning with
5869 if (hv_stores(invlist, INVLIST_ARRAY_KEY, newSVuv(PTR2UV(array))) == NULL) {
5870 Perl_croak(aTHX_ "panic: can't store '%s' entry in inversion list",
5875 PERL_STATIC_INLINE UV
5876 S_invlist_len(pTHX_ HV* const invlist)
5878 /* Returns the current number of elements in the inversion list's array */
5880 SV** len_ptr = hv_fetchs(invlist, INVLIST_LEN_KEY, FALSE);
5882 PERL_ARGS_ASSERT_INVLIST_LEN;
5884 if (len_ptr == NULL) {
5885 Perl_croak(aTHX_ "panic: inversion list without a '%s' element",
5889 return SvUV(*len_ptr);
5892 PERL_STATIC_INLINE UV
5893 S_invlist_max(pTHX_ HV* const invlist)
5895 /* Returns the maximum number of elements storable in the inversion list's
5896 * array, without having to realloc() */
5898 SV** max_ptr = hv_fetchs(invlist, INVLIST_MAX_KEY, FALSE);
5900 PERL_ARGS_ASSERT_INVLIST_MAX;
5902 if (max_ptr == NULL) {
5903 Perl_croak(aTHX_ "panic: inversion list without a '%s' element",
5907 return SvUV(*max_ptr);
5910 PERL_STATIC_INLINE void
5911 S_invlist_set_len(pTHX_ HV* const invlist, const UV len)
5913 /* Sets the current number of elements stored in the inversion list */
5915 PERL_ARGS_ASSERT_INVLIST_SET_LEN;
5917 if (len != 0 && len > invlist_max(invlist)) {
5918 Perl_croak(aTHX_ "panic: Can't make '%s=%"UVuf"' more than %s=%"UVuf" in inversion list", INVLIST_LEN_KEY, len, INVLIST_MAX_KEY, invlist_max(invlist));
5921 if (hv_stores(invlist, INVLIST_LEN_KEY, newSVuv(len)) == NULL) {
5922 Perl_croak(aTHX_ "panic: can't store '%s' entry in inversion list",
5927 PERL_STATIC_INLINE void
5928 S_invlist_set_max(pTHX_ HV* const invlist, const UV max)
5931 /* Sets the maximum number of elements storable in the inversion list
5932 * without having to realloc() */
5934 PERL_ARGS_ASSERT_INVLIST_SET_MAX;
5936 if (max < invlist_len(invlist)) {
5937 Perl_croak(aTHX_ "panic: Can't make '%s=%"UVuf"' less than %s=%"UVuf" in inversion list", INVLIST_MAX_KEY, invlist_len(invlist), INVLIST_LEN_KEY, invlist_max(invlist));
5940 if (hv_stores(invlist, INVLIST_MAX_KEY, newSVuv(max)) == NULL) {
5941 Perl_croak(aTHX_ "panic: can't store '%s' entry in inversion list",
5946 #ifndef PERL_IN_XSUB_RE
5948 Perl__new_invlist(pTHX_ IV initial_size)
5951 /* Return a pointer to a newly constructed inversion list, with enough
5952 * space to store 'initial_size' elements. If that number is negative, a
5953 * system default is used instead */
5955 HV* invlist = newHV();
5958 if (initial_size < 0) {
5959 initial_size = INVLIST_INITIAL_LEN;
5962 /* Allocate the initial space */
5963 Newx(list, initial_size, UV);
5964 invlist_set_array(invlist, list);
5966 /* set_len has to come before set_max, as the latter inspects the len */
5967 invlist_set_len(invlist, 0);
5968 invlist_set_max(invlist, initial_size);
5974 PERL_STATIC_INLINE void
5975 S_invlist_destroy(pTHX_ HV* const invlist)
5977 /* Inversion list destructor */
5979 SV** list_ptr = hv_fetchs(invlist, INVLIST_ARRAY_KEY, FALSE);
5981 PERL_ARGS_ASSERT_INVLIST_DESTROY;
5983 if (list_ptr != NULL) {
5984 UV *list = INT2PTR(UV *, SvUV(*list_ptr)); /* PERL_POISON needs lvalue */
5990 S_invlist_extend(pTHX_ HV* const invlist, const UV new_max)
5992 /* Change the maximum size of an inversion list (up or down) */
5996 const UV old_max = invlist_max(invlist);
5998 PERL_ARGS_ASSERT_INVLIST_EXTEND;
6000 if (old_max == new_max) { /* If a no-op */
6004 array = orig_array = invlist_array(invlist);
6005 Renew(array, new_max, UV);
6007 /* If the size change moved the list in memory, set the new one */
6008 if (array != orig_array) {
6009 invlist_set_array(invlist, array);
6012 invlist_set_max(invlist, new_max);
6016 PERL_STATIC_INLINE void
6017 S_invlist_trim(pTHX_ HV* const invlist)
6019 PERL_ARGS_ASSERT_INVLIST_TRIM;
6021 /* Change the length of the inversion list to how many entries it currently
6024 invlist_extend(invlist, invlist_len(invlist));
6027 /* An element is in an inversion list iff its index is even numbered: 0, 2, 4,
6030 #define ELEMENT_IN_INVLIST_SET(i) (! ((i) & 1))
6032 #ifndef PERL_IN_XSUB_RE
6034 Perl__append_range_to_invlist(pTHX_ HV* const invlist, const UV start, const UV end)
6036 /* Subject to change or removal. Append the range from 'start' to 'end' at
6037 * the end of the inversion list. The range must be above any existing
6040 UV* array = invlist_array(invlist);
6041 UV max = invlist_max(invlist);
6042 UV len = invlist_len(invlist);
6044 PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST;
6048 /* Here, the existing list is non-empty. The current max entry in the
6049 * list is generally the first value not in the set, except when the
6050 * set extends to the end of permissible values, in which case it is
6051 * the first entry in that final set, and so this call is an attempt to
6052 * append out-of-order */
6054 UV final_element = len - 1;
6055 if (array[final_element] > start
6056 || ELEMENT_IN_INVLIST_SET(final_element))
6058 Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list");
6061 /* Here, it is a legal append. If the new range begins with the first
6062 * value not in the set, it is extending the set, so the new first
6063 * value not in the set is one greater than the newly extended range.
6065 if (array[final_element] == start) {
6066 if (end != UV_MAX) {
6067 array[final_element] = end + 1;
6070 /* But if the end is the maximum representable on the machine,
6071 * just let the range that this would extend have no end */
6072 invlist_set_len(invlist, len - 1);
6078 /* Here the new range doesn't extend any existing set. Add it */
6080 len += 2; /* Includes an element each for the start and end of range */
6082 /* If overflows the existing space, extend, which may cause the array to be
6085 invlist_extend(invlist, len);
6086 array = invlist_array(invlist);
6089 invlist_set_len(invlist, len);
6091 /* The next item on the list starts the range, the one after that is
6092 * one past the new range. */
6093 array[len - 2] = start;
6094 if (end != UV_MAX) {
6095 array[len - 1] = end + 1;
6098 /* But if the end is the maximum representable on the machine, just let
6099 * the range have no end */
6100 invlist_set_len(invlist, len - 1);
6106 S_invlist_union(pTHX_ HV* const a, HV* const b)
6108 /* Return a new inversion list which is the union of two inversion lists.
6109 * The basis for this comes from "Unicode Demystified" Chapter 13 by
6110 * Richard Gillam, published by Addison-Wesley, and explained at some
6111 * length there. The preface says to incorporate its examples into your
6112 * code at your own risk.
6114 * The algorithm is like a merge sort.
6116 * XXX A potential performance improvement is to keep track as we go along
6117 * if only one of the inputs contributes to the result, meaning the other
6118 * is a subset of that one. In that case, we can skip the final copy and
6119 * return the larger of the input lists */
6121 UV* array_a = invlist_array(a); /* a's array */
6122 UV* array_b = invlist_array(b);
6123 UV len_a = invlist_len(a); /* length of a's array */
6124 UV len_b = invlist_len(b);
6126 HV* u; /* the resulting union */
6130 UV i_a = 0; /* current index into a's array */
6134 /* running count, as explained in the algorithm source book; items are
6135 * stopped accumulating and are output when the count changes to/from 0.
6136 * The count is incremented when we start a range that's in the set, and
6137 * decremented when we start a range that's not in the set. So its range
6138 * is 0 to 2. Only when the count is zero is something not in the set.
6142 PERL_ARGS_ASSERT_INVLIST_UNION;
6144 /* Size the union for the worst case: that the sets are completely
6146 u = _new_invlist(len_a + len_b);
6147 array_u = invlist_array(u);
6149 /* Go through each list item by item, stopping when exhausted one of
6151 while (i_a < len_a && i_b < len_b) {
6152 UV cp; /* The element to potentially add to the union's array */
6153 bool cp_in_set; /* is it in the the input list's set or not */
6155 /* We need to take one or the other of the two inputs for the union.
6156 * Since we are merging two sorted lists, we take the smaller of the
6157 * next items. In case of a tie, we take the one that is in its set
6158 * first. If we took one not in the set first, it would decrement the
6159 * count, possibly to 0 which would cause it to be output as ending the
6160 * range, and the next time through we would take the same number, and
6161 * output it again as beginning the next range. By doing it the
6162 * opposite way, there is no possibility that the count will be
6163 * momentarily decremented to 0, and thus the two adjoining ranges will
6164 * be seamlessly merged. (In a tie and both are in the set or both not
6165 * in the set, it doesn't matter which we take first.) */
6166 if (array_a[i_a] < array_b[i_b]
6167 || (array_a[i_a] == array_b[i_b] && ELEMENT_IN_INVLIST_SET(i_a)))
6169 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6173 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6177 /* Here, have chosen which of the two inputs to look at. Only output
6178 * if the running count changes to/from 0, which marks the
6179 * beginning/end of a range in that's in the set */
6182 array_u[i_u++] = cp;
6189 array_u[i_u++] = cp;
6194 /* Here, we are finished going through at least one of the lists, which
6195 * means there is something remaining in at most one. We check if the list
6196 * that hasn't been exhausted is positioned such that we are in the middle
6197 * of a range in its set or not. (We are in the set if the next item in
6198 * the array marks the beginning of something not in the set) If in the
6199 * set, we decrement 'count'; if 0, there is potentially more to output.
6200 * There are four cases:
6201 * 1) Both weren't in their sets, count is 0, and remains 0. What's left
6202 * in the union is entirely from the non-exhausted set.
6203 * 2) Both were in their sets, count is 2. Nothing further should
6204 * be output, as everything that remains will be in the exhausted
6205 * list's set, hence in the union; decrementing to 1 but not 0 insures
6207 * 3) the exhausted was in its set, non-exhausted isn't, count is 1.
6208 * Nothing further should be output because the union includes
6209 * everything from the exhausted set. Not decrementing insures that.
6210 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1;
6211 * decrementing to 0 insures that we look at the remainder of the
6212 * non-exhausted set */
6213 if ((i_a != len_a && ! ELEMENT_IN_INVLIST_SET(i_a))
6214 || (i_b != len_b && ! ELEMENT_IN_INVLIST_SET(i_b)))
6219 /* The final length is what we've output so far, plus what else is about to
6220 * be output. (If 'count' is non-zero, then the input list we exhausted
6221 * has everything remaining up to the machine's limit in its set, and hence
6222 * in the union, so there will be no further output. */
6225 /* At most one of the subexpressions will be non-zero */
6226 len_u += (len_a - i_a) + (len_b - i_b);
6229 /* Set result to final length, which can change the pointer to array_u, so
6231 if (len_u != invlist_len(u)) {
6232 invlist_set_len(u, len_u);
6234 array_u = invlist_array(u);
6237 /* When 'count' is 0, the list that was exhausted (if one was shorter than
6238 * the other) ended with everything above it not in its set. That means
6239 * that the remaining part of the union is precisely the same as the
6240 * non-exhausted list, so can just copy it unchanged. (If both list were
6241 * exhausted at the same time, then the operations below will be both 0.)
6244 IV copy_count; /* At most one will have a non-zero copy count */
6245 if ((copy_count = len_a - i_a) > 0) {
6246 Copy(array_a + i_a, array_u + i_u, copy_count, UV);
6248 else if ((copy_count = len_b - i_b) > 0) {
6249 Copy(array_b + i_b, array_u + i_u, copy_count, UV);
6257 S_invlist_intersection(pTHX_ HV* const a, HV* const b)
6259 /* Return the intersection of two inversion lists. The basis for this
6260 * comes from "Unicode Demystified" Chapter 13 by Richard Gillam, published
6261 * by Addison-Wesley, and explained at some length there. The preface says
6262 * to incorporate its examples into your code at your own risk.
6264 * The algorithm is like a merge sort, and is essentially the same as the
6268 UV* array_a = invlist_array(a); /* a's array */
6269 UV* array_b = invlist_array(b);
6270 UV len_a = invlist_len(a); /* length of a's array */
6271 UV len_b = invlist_len(b);
6273 HV* r; /* the resulting intersection */
6277 UV i_a = 0; /* current index into a's array */
6281 /* running count, as explained in the algorithm source book; items are
6282 * stopped accumulating and are output when the count changes to/from 2.
6283 * The count is incremented when we start a range that's in the set, and
6284 * decremented when we start a range that's not in the set. So its range
6285 * is 0 to 2. Only when the count is 2 is something in the intersection.
6289 PERL_ARGS_ASSERT_INVLIST_INTERSECTION;
6291 /* Size the intersection for the worst case: that the intersection ends up
6292 * fragmenting everything to be completely disjoint */
6293 r= _new_invlist(len_a + len_b);
6294 array_r = invlist_array(r);
6296 /* Go through each list item by item, stopping when exhausted one of
6298 while (i_a < len_a && i_b < len_b) {
6299 UV cp; /* The element to potentially add to the intersection's
6301 bool cp_in_set; /* Is it in the input list's set or not */
6303 /* We need to take one or the other of the two inputs for the union.
6304 * Since we are merging two sorted lists, we take the smaller of the
6305 * next items. In case of a tie, we take the one that is not in its
6306 * set first (a difference from the union algorithm). If we took one
6307 * in the set first, it would increment the count, possibly to 2 which
6308 * would cause it to be output as starting a range in the intersection,
6309 * and the next time through we would take that same number, and output
6310 * it again as ending the set. By doing it the opposite of this, we
6311 * there is no possibility that the count will be momentarily
6312 * incremented to 2. (In a tie and both are in the set or both not in
6313 * the set, it doesn't matter which we take first.) */
6314 if (array_a[i_a] < array_b[i_b]
6315 || (array_a[i_a] == array_b[i_b] && ! ELEMENT_IN_INVLIST_SET(i_a)))
6317 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6321 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6325 /* Here, have chosen which of the two inputs to look at. Only output
6326 * if the running count changes to/from 2, which marks the
6327 * beginning/end of a range that's in the intersection */
6331 array_r[i_r++] = cp;
6336 array_r[i_r++] = cp;
6342 /* Here, we are finished going through at least one of the sets, which
6343 * means there is something remaining in at most one. See the comments in
6345 if ((i_a != len_a && ! ELEMENT_IN_INVLIST_SET(i_a))
6346 || (i_b != len_b && ! ELEMENT_IN_INVLIST_SET(i_b)))
6351 /* The final length is what we've output so far plus what else is in the
6352 * intersection. Only one of the subexpressions below will be non-zero */
6355 len_r += (len_a - i_a) + (len_b - i_b);
6358 /* Set result to final length, which can change the pointer to array_r, so
6360 if (len_r != invlist_len(r)) {
6361 invlist_set_len(r, len_r);
6363 array_r = invlist_array(r);
6366 /* Finish outputting any remaining */
6367 if (count == 2) { /* Only one of will have a non-zero copy count */
6369 if ((copy_count = len_a - i_a) > 0) {
6370 Copy(array_a + i_a, array_r + i_r, copy_count, UV);
6372 else if ((copy_count = len_b - i_b) > 0) {
6373 Copy(array_b + i_b, array_r + i_r, copy_count, UV);
6381 S_add_range_to_invlist(pTHX_ HV* invlist, const UV start, const UV end)
6383 /* Add the range from 'start' to 'end' inclusive to the inversion list's
6384 * set. A pointer to the inversion list is returned. This may actually be
6385 * a new list, in which case the passed in one has been destroyed. The
6386 * passed in inversion list can be NULL, in which case a new one is created
6387 * with just the one range in it */
6393 if (invlist == NULL) {
6394 invlist = _new_invlist(2);
6398 len = invlist_len(invlist);
6401 /* If comes after the final entry, can just append it to the end */
6403 || start >= invlist_array(invlist)
6404 [invlist_len(invlist) - 1])
6406 _append_range_to_invlist(invlist, start, end);
6410 /* Here, can't just append things, create and return a new inversion list
6411 * which is the union of this range and the existing inversion list */
6412 range_invlist = _new_invlist(2);
6413 _append_range_to_invlist(range_invlist, start, end);
6415 added_invlist = invlist_union(invlist, range_invlist);
6417 /* The passed in list can be freed, as well as our temporary */
6418 invlist_destroy(range_invlist);
6419 if (invlist != added_invlist) {
6420 invlist_destroy(invlist);
6423 return added_invlist;
6426 PERL_STATIC_INLINE HV*
6427 S_add_cp_to_invlist(pTHX_ HV* invlist, const UV cp) {
6428 return add_range_to_invlist(invlist, cp, cp);
6431 /* End of inversion list object */
6434 - reg - regular expression, i.e. main body or parenthesized thing
6436 * Caller must absorb opening parenthesis.
6438 * Combining parenthesis handling with the base level of regular expression
6439 * is a trifle forced, but the need to tie the tails of the branches to what
6440 * follows makes it hard to avoid.
6442 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
6444 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
6446 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
6450 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
6451 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
6454 register regnode *ret; /* Will be the head of the group. */
6455 register regnode *br;
6456 register regnode *lastbr;
6457 register regnode *ender = NULL;
6458 register I32 parno = 0;
6460 U32 oregflags = RExC_flags;
6461 bool have_branch = 0;
6463 I32 freeze_paren = 0;
6464 I32 after_freeze = 0;
6466 /* for (?g), (?gc), and (?o) warnings; warning
6467 about (?c) will warn about (?g) -- japhy */
6469 #define WASTED_O 0x01
6470 #define WASTED_G 0x02
6471 #define WASTED_C 0x04
6472 #define WASTED_GC (0x02|0x04)
6473 I32 wastedflags = 0x00;
6475 char * parse_start = RExC_parse; /* MJD */
6476 char * const oregcomp_parse = RExC_parse;
6478 GET_RE_DEBUG_FLAGS_DECL;
6480 PERL_ARGS_ASSERT_REG;
6481 DEBUG_PARSE("reg ");
6483 *flagp = 0; /* Tentatively. */
6486 /* Make an OPEN node, if parenthesized. */
6488 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
6489 char *start_verb = RExC_parse;
6490 STRLEN verb_len = 0;
6491 char *start_arg = NULL;
6492 unsigned char op = 0;
6494 int internal_argval = 0; /* internal_argval is only useful if !argok */
6495 while ( *RExC_parse && *RExC_parse != ')' ) {
6496 if ( *RExC_parse == ':' ) {
6497 start_arg = RExC_parse + 1;
6503 verb_len = RExC_parse - start_verb;
6506 while ( *RExC_parse && *RExC_parse != ')' )
6508 if ( *RExC_parse != ')' )
6509 vFAIL("Unterminated verb pattern argument");
6510 if ( RExC_parse == start_arg )
6513 if ( *RExC_parse != ')' )
6514 vFAIL("Unterminated verb pattern");
6517 switch ( *start_verb ) {
6518 case 'A': /* (*ACCEPT) */
6519 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
6521 internal_argval = RExC_nestroot;
6524 case 'C': /* (*COMMIT) */
6525 if ( memEQs(start_verb,verb_len,"COMMIT") )
6528 case 'F': /* (*FAIL) */
6529 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
6534 case ':': /* (*:NAME) */
6535 case 'M': /* (*MARK:NAME) */
6536 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
6541 case 'P': /* (*PRUNE) */
6542 if ( memEQs(start_verb,verb_len,"PRUNE") )
6545 case 'S': /* (*SKIP) */
6546 if ( memEQs(start_verb,verb_len,"SKIP") )
6549 case 'T': /* (*THEN) */
6550 /* [19:06] <TimToady> :: is then */
6551 if ( memEQs(start_verb,verb_len,"THEN") ) {
6553 RExC_seen |= REG_SEEN_CUTGROUP;
6559 vFAIL3("Unknown verb pattern '%.*s'",
6560 verb_len, start_verb);
6563 if ( start_arg && internal_argval ) {
6564 vFAIL3("Verb pattern '%.*s' may not have an argument",
6565 verb_len, start_verb);
6566 } else if ( argok < 0 && !start_arg ) {
6567 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
6568 verb_len, start_verb);
6570 ret = reganode(pRExC_state, op, internal_argval);
6571 if ( ! internal_argval && ! SIZE_ONLY ) {
6573 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
6574 ARG(ret) = add_data( pRExC_state, 1, "S" );
6575 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
6582 if (!internal_argval)
6583 RExC_seen |= REG_SEEN_VERBARG;
6584 } else if ( start_arg ) {
6585 vFAIL3("Verb pattern '%.*s' may not have an argument",
6586 verb_len, start_verb);
6588 ret = reg_node(pRExC_state, op);
6590 nextchar(pRExC_state);
6593 if (*RExC_parse == '?') { /* (?...) */
6594 bool is_logical = 0;
6595 const char * const seqstart = RExC_parse;
6596 bool has_use_defaults = FALSE;
6599 paren = *RExC_parse++;
6600 ret = NULL; /* For look-ahead/behind. */
6603 case 'P': /* (?P...) variants for those used to PCRE/Python */
6604 paren = *RExC_parse++;
6605 if ( paren == '<') /* (?P<...>) named capture */
6607 else if (paren == '>') { /* (?P>name) named recursion */
6608 goto named_recursion;
6610 else if (paren == '=') { /* (?P=...) named backref */
6611 /* this pretty much dupes the code for \k<NAME> in regatom(), if
6612 you change this make sure you change that */
6613 char* name_start = RExC_parse;
6615 SV *sv_dat = reg_scan_name(pRExC_state,
6616 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6617 if (RExC_parse == name_start || *RExC_parse != ')')
6618 vFAIL2("Sequence %.3s... not terminated",parse_start);
6621 num = add_data( pRExC_state, 1, "S" );
6622 RExC_rxi->data->data[num]=(void*)sv_dat;
6623 SvREFCNT_inc_simple_void(sv_dat);
6626 ret = reganode(pRExC_state,
6629 : (MORE_ASCII_RESTRICTED)
6631 : (AT_LEAST_UNI_SEMANTICS)
6639 Set_Node_Offset(ret, parse_start+1);
6640 Set_Node_Cur_Length(ret); /* MJD */
6642 nextchar(pRExC_state);
6646 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6648 case '<': /* (?<...) */
6649 if (*RExC_parse == '!')
6651 else if (*RExC_parse != '=')
6657 case '\'': /* (?'...') */
6658 name_start= RExC_parse;
6659 svname = reg_scan_name(pRExC_state,
6660 SIZE_ONLY ? /* reverse test from the others */
6661 REG_RSN_RETURN_NAME :
6662 REG_RSN_RETURN_NULL);
6663 if (RExC_parse == name_start) {
6665 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6668 if (*RExC_parse != paren)
6669 vFAIL2("Sequence (?%c... not terminated",
6670 paren=='>' ? '<' : paren);
6674 if (!svname) /* shouldn't happen */
6676 "panic: reg_scan_name returned NULL");
6677 if (!RExC_paren_names) {
6678 RExC_paren_names= newHV();
6679 sv_2mortal(MUTABLE_SV(RExC_paren_names));
6681 RExC_paren_name_list= newAV();
6682 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
6685 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
6687 sv_dat = HeVAL(he_str);
6689 /* croak baby croak */
6691 "panic: paren_name hash element allocation failed");
6692 } else if ( SvPOK(sv_dat) ) {
6693 /* (?|...) can mean we have dupes so scan to check
6694 its already been stored. Maybe a flag indicating
6695 we are inside such a construct would be useful,
6696 but the arrays are likely to be quite small, so
6697 for now we punt -- dmq */
6698 IV count = SvIV(sv_dat);
6699 I32 *pv = (I32*)SvPVX(sv_dat);
6701 for ( i = 0 ; i < count ; i++ ) {
6702 if ( pv[i] == RExC_npar ) {
6708 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
6709 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
6710 pv[count] = RExC_npar;
6711 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
6714 (void)SvUPGRADE(sv_dat,SVt_PVNV);
6715 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
6717 SvIV_set(sv_dat, 1);
6720 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
6721 SvREFCNT_dec(svname);
6724 /*sv_dump(sv_dat);*/
6726 nextchar(pRExC_state);
6728 goto capturing_parens;
6730 RExC_seen |= REG_SEEN_LOOKBEHIND;
6731 RExC_in_lookbehind++;
6733 case '=': /* (?=...) */
6734 RExC_seen_zerolen++;
6736 case '!': /* (?!...) */
6737 RExC_seen_zerolen++;
6738 if (*RExC_parse == ')') {
6739 ret=reg_node(pRExC_state, OPFAIL);
6740 nextchar(pRExC_state);
6744 case '|': /* (?|...) */
6745 /* branch reset, behave like a (?:...) except that
6746 buffers in alternations share the same numbers */
6748 after_freeze = freeze_paren = RExC_npar;
6750 case ':': /* (?:...) */
6751 case '>': /* (?>...) */
6753 case '$': /* (?$...) */
6754 case '@': /* (?@...) */
6755 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
6757 case '#': /* (?#...) */
6758 while (*RExC_parse && *RExC_parse != ')')
6760 if (*RExC_parse != ')')
6761 FAIL("Sequence (?#... not terminated");
6762 nextchar(pRExC_state);
6765 case '0' : /* (?0) */
6766 case 'R' : /* (?R) */
6767 if (*RExC_parse != ')')
6768 FAIL("Sequence (?R) not terminated");
6769 ret = reg_node(pRExC_state, GOSTART);
6770 *flagp |= POSTPONED;
6771 nextchar(pRExC_state);
6774 { /* named and numeric backreferences */
6776 case '&': /* (?&NAME) */
6777 parse_start = RExC_parse - 1;
6780 SV *sv_dat = reg_scan_name(pRExC_state,
6781 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6782 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6784 goto gen_recurse_regop;
6787 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6789 vFAIL("Illegal pattern");
6791 goto parse_recursion;
6793 case '-': /* (?-1) */
6794 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6795 RExC_parse--; /* rewind to let it be handled later */
6799 case '1': case '2': case '3': case '4': /* (?1) */
6800 case '5': case '6': case '7': case '8': case '9':
6803 num = atoi(RExC_parse);
6804 parse_start = RExC_parse - 1; /* MJD */
6805 if (*RExC_parse == '-')
6807 while (isDIGIT(*RExC_parse))
6809 if (*RExC_parse!=')')
6810 vFAIL("Expecting close bracket");
6813 if ( paren == '-' ) {
6815 Diagram of capture buffer numbering.
6816 Top line is the normal capture buffer numbers
6817 Bottom line is the negative indexing as from
6821 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
6825 num = RExC_npar + num;
6828 vFAIL("Reference to nonexistent group");
6830 } else if ( paren == '+' ) {
6831 num = RExC_npar + num - 1;
6834 ret = reganode(pRExC_state, GOSUB, num);
6836 if (num > (I32)RExC_rx->nparens) {
6838 vFAIL("Reference to nonexistent group");
6840 ARG2L_SET( ret, RExC_recurse_count++);
6842 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6843 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
6847 RExC_seen |= REG_SEEN_RECURSE;
6848 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
6849 Set_Node_Offset(ret, parse_start); /* MJD */
6851 *flagp |= POSTPONED;
6852 nextchar(pRExC_state);
6854 } /* named and numeric backreferences */
6857 case '?': /* (??...) */
6859 if (*RExC_parse != '{') {
6861 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6864 *flagp |= POSTPONED;
6865 paren = *RExC_parse++;
6867 case '{': /* (?{...}) */
6872 char *s = RExC_parse;
6874 RExC_seen_zerolen++;
6875 RExC_seen |= REG_SEEN_EVAL;
6876 while (count && (c = *RExC_parse)) {
6887 if (*RExC_parse != ')') {
6889 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
6893 OP_4tree *sop, *rop;
6894 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
6897 Perl_save_re_context(aTHX);
6898 rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad);
6899 sop->op_private |= OPpREFCOUNTED;
6900 /* re_dup will OpREFCNT_inc */
6901 OpREFCNT_set(sop, 1);
6904 n = add_data(pRExC_state, 3, "nop");
6905 RExC_rxi->data->data[n] = (void*)rop;
6906 RExC_rxi->data->data[n+1] = (void*)sop;
6907 RExC_rxi->data->data[n+2] = (void*)pad;
6910 else { /* First pass */
6911 if (PL_reginterp_cnt < ++RExC_seen_evals
6913 /* No compiled RE interpolated, has runtime
6914 components ===> unsafe. */
6915 FAIL("Eval-group not allowed at runtime, use re 'eval'");
6916 if (PL_tainting && PL_tainted)
6917 FAIL("Eval-group in insecure regular expression");
6918 #if PERL_VERSION > 8
6919 if (IN_PERL_COMPILETIME)
6924 nextchar(pRExC_state);
6926 ret = reg_node(pRExC_state, LOGICAL);
6929 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
6930 /* deal with the length of this later - MJD */
6933 ret = reganode(pRExC_state, EVAL, n);
6934 Set_Node_Length(ret, RExC_parse - parse_start + 1);
6935 Set_Node_Offset(ret, parse_start);
6938 case '(': /* (?(?{...})...) and (?(?=...)...) */
6941 if (RExC_parse[0] == '?') { /* (?(?...)) */
6942 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
6943 || RExC_parse[1] == '<'
6944 || RExC_parse[1] == '{') { /* Lookahead or eval. */
6947 ret = reg_node(pRExC_state, LOGICAL);
6950 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
6954 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
6955 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
6957 char ch = RExC_parse[0] == '<' ? '>' : '\'';
6958 char *name_start= RExC_parse++;
6960 SV *sv_dat=reg_scan_name(pRExC_state,
6961 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6962 if (RExC_parse == name_start || *RExC_parse != ch)
6963 vFAIL2("Sequence (?(%c... not terminated",
6964 (ch == '>' ? '<' : ch));
6967 num = add_data( pRExC_state, 1, "S" );
6968 RExC_rxi->data->data[num]=(void*)sv_dat;
6969 SvREFCNT_inc_simple_void(sv_dat);
6971 ret = reganode(pRExC_state,NGROUPP,num);
6972 goto insert_if_check_paren;
6974 else if (RExC_parse[0] == 'D' &&
6975 RExC_parse[1] == 'E' &&
6976 RExC_parse[2] == 'F' &&
6977 RExC_parse[3] == 'I' &&
6978 RExC_parse[4] == 'N' &&
6979 RExC_parse[5] == 'E')
6981 ret = reganode(pRExC_state,DEFINEP,0);
6984 goto insert_if_check_paren;
6986 else if (RExC_parse[0] == 'R') {
6989 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
6990 parno = atoi(RExC_parse++);
6991 while (isDIGIT(*RExC_parse))
6993 } else if (RExC_parse[0] == '&') {
6996 sv_dat = reg_scan_name(pRExC_state,
6997 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6998 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
7000 ret = reganode(pRExC_state,INSUBP,parno);
7001 goto insert_if_check_paren;
7003 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
7006 parno = atoi(RExC_parse++);
7008 while (isDIGIT(*RExC_parse))
7010 ret = reganode(pRExC_state, GROUPP, parno);
7012 insert_if_check_paren:
7013 if ((c = *nextchar(pRExC_state)) != ')')
7014 vFAIL("Switch condition not recognized");
7016 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
7017 br = regbranch(pRExC_state, &flags, 1,depth+1);
7019 br = reganode(pRExC_state, LONGJMP, 0);
7021 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
7022 c = *nextchar(pRExC_state);
7027 vFAIL("(?(DEFINE)....) does not allow branches");
7028 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
7029 regbranch(pRExC_state, &flags, 1,depth+1);
7030 REGTAIL(pRExC_state, ret, lastbr);
7033 c = *nextchar(pRExC_state);
7038 vFAIL("Switch (?(condition)... contains too many branches");
7039 ender = reg_node(pRExC_state, TAIL);
7040 REGTAIL(pRExC_state, br, ender);
7042 REGTAIL(pRExC_state, lastbr, ender);
7043 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
7046 REGTAIL(pRExC_state, ret, ender);
7047 RExC_size++; /* XXX WHY do we need this?!!
7048 For large programs it seems to be required
7049 but I can't figure out why. -- dmq*/
7053 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
7057 RExC_parse--; /* for vFAIL to print correctly */
7058 vFAIL("Sequence (? incomplete");
7060 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
7062 has_use_defaults = TRUE;
7063 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
7064 set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics)
7065 ? REGEX_UNICODE_CHARSET
7066 : REGEX_DEPENDS_CHARSET);
7070 parse_flags: /* (?i) */
7072 U32 posflags = 0, negflags = 0;
7073 U32 *flagsp = &posflags;
7074 char has_charset_modifier = '\0';
7075 regex_charset cs = (RExC_utf8 || RExC_uni_semantics)
7076 ? REGEX_UNICODE_CHARSET
7077 : REGEX_DEPENDS_CHARSET;
7079 while (*RExC_parse) {
7080 /* && strchr("iogcmsx", *RExC_parse) */
7081 /* (?g), (?gc) and (?o) are useless here
7082 and must be globally applied -- japhy */
7083 switch (*RExC_parse) {
7084 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
7085 case LOCALE_PAT_MOD:
7086 if (has_charset_modifier) {
7087 goto excess_modifier;
7089 else if (flagsp == &negflags) {
7092 cs = REGEX_LOCALE_CHARSET;
7093 has_charset_modifier = LOCALE_PAT_MOD;
7094 RExC_contains_locale = 1;
7096 case UNICODE_PAT_MOD:
7097 if (has_charset_modifier) {
7098 goto excess_modifier;
7100 else if (flagsp == &negflags) {
7103 cs = REGEX_UNICODE_CHARSET;
7104 has_charset_modifier = UNICODE_PAT_MOD;
7106 case ASCII_RESTRICT_PAT_MOD:
7107 if (flagsp == &negflags) {
7110 if (has_charset_modifier) {
7111 if (cs != REGEX_ASCII_RESTRICTED_CHARSET) {
7112 goto excess_modifier;
7114 /* Doubled modifier implies more restricted */
7115 cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET;
7118 cs = REGEX_ASCII_RESTRICTED_CHARSET;
7120 has_charset_modifier = ASCII_RESTRICT_PAT_MOD;
7122 case DEPENDS_PAT_MOD:
7123 if (has_use_defaults) {
7124 goto fail_modifiers;
7126 else if (flagsp == &negflags) {
7129 else if (has_charset_modifier) {
7130 goto excess_modifier;
7133 /* The dual charset means unicode semantics if the
7134 * pattern (or target, not known until runtime) are
7135 * utf8, or something in the pattern indicates unicode
7137 cs = (RExC_utf8 || RExC_uni_semantics)
7138 ? REGEX_UNICODE_CHARSET
7139 : REGEX_DEPENDS_CHARSET;
7140 has_charset_modifier = DEPENDS_PAT_MOD;
7144 if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) {
7145 vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD);
7147 else if (has_charset_modifier == *(RExC_parse - 1)) {
7148 vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1));
7151 vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1));
7156 vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1));
7158 case ONCE_PAT_MOD: /* 'o' */
7159 case GLOBAL_PAT_MOD: /* 'g' */
7160 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7161 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
7162 if (! (wastedflags & wflagbit) ) {
7163 wastedflags |= wflagbit;
7166 "Useless (%s%c) - %suse /%c modifier",
7167 flagsp == &negflags ? "?-" : "?",
7169 flagsp == &negflags ? "don't " : "",
7176 case CONTINUE_PAT_MOD: /* 'c' */
7177 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7178 if (! (wastedflags & WASTED_C) ) {
7179 wastedflags |= WASTED_GC;
7182 "Useless (%sc) - %suse /gc modifier",
7183 flagsp == &negflags ? "?-" : "?",
7184 flagsp == &negflags ? "don't " : ""
7189 case KEEPCOPY_PAT_MOD: /* 'p' */
7190 if (flagsp == &negflags) {
7192 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
7194 *flagsp |= RXf_PMf_KEEPCOPY;
7198 /* A flag is a default iff it is following a minus, so
7199 * if there is a minus, it means will be trying to
7200 * re-specify a default which is an error */
7201 if (has_use_defaults || flagsp == &negflags) {
7204 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7208 wastedflags = 0; /* reset so (?g-c) warns twice */
7214 RExC_flags |= posflags;
7215 RExC_flags &= ~negflags;
7216 set_regex_charset(&RExC_flags, cs);
7218 oregflags |= posflags;
7219 oregflags &= ~negflags;
7220 set_regex_charset(&oregflags, cs);
7222 nextchar(pRExC_state);
7233 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7238 }} /* one for the default block, one for the switch */
7245 ret = reganode(pRExC_state, OPEN, parno);
7248 RExC_nestroot = parno;
7249 if (RExC_seen & REG_SEEN_RECURSE
7250 && !RExC_open_parens[parno-1])
7252 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7253 "Setting open paren #%"IVdf" to %d\n",
7254 (IV)parno, REG_NODE_NUM(ret)));
7255 RExC_open_parens[parno-1]= ret;
7258 Set_Node_Length(ret, 1); /* MJD */
7259 Set_Node_Offset(ret, RExC_parse); /* MJD */
7267 /* Pick up the branches, linking them together. */
7268 parse_start = RExC_parse; /* MJD */
7269 br = regbranch(pRExC_state, &flags, 1,depth+1);
7271 /* branch_len = (paren != 0); */
7275 if (*RExC_parse == '|') {
7276 if (!SIZE_ONLY && RExC_extralen) {
7277 reginsert(pRExC_state, BRANCHJ, br, depth+1);
7280 reginsert(pRExC_state, BRANCH, br, depth+1);
7281 Set_Node_Length(br, paren != 0);
7282 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
7286 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
7288 else if (paren == ':') {
7289 *flagp |= flags&SIMPLE;
7291 if (is_open) { /* Starts with OPEN. */
7292 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
7294 else if (paren != '?') /* Not Conditional */
7296 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7298 while (*RExC_parse == '|') {
7299 if (!SIZE_ONLY && RExC_extralen) {
7300 ender = reganode(pRExC_state, LONGJMP,0);
7301 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
7304 RExC_extralen += 2; /* Account for LONGJMP. */
7305 nextchar(pRExC_state);
7307 if (RExC_npar > after_freeze)
7308 after_freeze = RExC_npar;
7309 RExC_npar = freeze_paren;
7311 br = regbranch(pRExC_state, &flags, 0, depth+1);
7315 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
7317 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7320 if (have_branch || paren != ':') {
7321 /* Make a closing node, and hook it on the end. */
7324 ender = reg_node(pRExC_state, TAIL);
7327 ender = reganode(pRExC_state, CLOSE, parno);
7328 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
7329 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7330 "Setting close paren #%"IVdf" to %d\n",
7331 (IV)parno, REG_NODE_NUM(ender)));
7332 RExC_close_parens[parno-1]= ender;
7333 if (RExC_nestroot == parno)
7336 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
7337 Set_Node_Length(ender,1); /* MJD */
7343 *flagp &= ~HASWIDTH;
7346 ender = reg_node(pRExC_state, SUCCEED);
7349 ender = reg_node(pRExC_state, END);
7351 assert(!RExC_opend); /* there can only be one! */
7356 REGTAIL(pRExC_state, lastbr, ender);
7358 if (have_branch && !SIZE_ONLY) {
7360 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
7362 /* Hook the tails of the branches to the closing node. */
7363 for (br = ret; br; br = regnext(br)) {
7364 const U8 op = PL_regkind[OP(br)];
7366 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
7368 else if (op == BRANCHJ) {
7369 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
7377 static const char parens[] = "=!<,>";
7379 if (paren && (p = strchr(parens, paren))) {
7380 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
7381 int flag = (p - parens) > 1;
7384 node = SUSPEND, flag = 0;
7385 reginsert(pRExC_state, node,ret, depth+1);
7386 Set_Node_Cur_Length(ret);
7387 Set_Node_Offset(ret, parse_start + 1);
7389 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
7393 /* Check for proper termination. */
7395 RExC_flags = oregflags;
7396 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
7397 RExC_parse = oregcomp_parse;
7398 vFAIL("Unmatched (");
7401 else if (!paren && RExC_parse < RExC_end) {
7402 if (*RExC_parse == ')') {
7404 vFAIL("Unmatched )");
7407 FAIL("Junk on end of regexp"); /* "Can't happen". */
7411 if (RExC_in_lookbehind) {
7412 RExC_in_lookbehind--;
7414 if (after_freeze > RExC_npar)
7415 RExC_npar = after_freeze;
7420 - regbranch - one alternative of an | operator
7422 * Implements the concatenation operator.
7425 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
7428 register regnode *ret;
7429 register regnode *chain = NULL;
7430 register regnode *latest;
7431 I32 flags = 0, c = 0;
7432 GET_RE_DEBUG_FLAGS_DECL;
7434 PERL_ARGS_ASSERT_REGBRANCH;
7436 DEBUG_PARSE("brnc");
7441 if (!SIZE_ONLY && RExC_extralen)
7442 ret = reganode(pRExC_state, BRANCHJ,0);
7444 ret = reg_node(pRExC_state, BRANCH);
7445 Set_Node_Length(ret, 1);
7449 if (!first && SIZE_ONLY)
7450 RExC_extralen += 1; /* BRANCHJ */
7452 *flagp = WORST; /* Tentatively. */
7455 nextchar(pRExC_state);
7456 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
7458 latest = regpiece(pRExC_state, &flags,depth+1);
7459 if (latest == NULL) {
7460 if (flags & TRYAGAIN)
7464 else if (ret == NULL)
7466 *flagp |= flags&(HASWIDTH|POSTPONED);
7467 if (chain == NULL) /* First piece. */
7468 *flagp |= flags&SPSTART;
7471 REGTAIL(pRExC_state, chain, latest);
7476 if (chain == NULL) { /* Loop ran zero times. */
7477 chain = reg_node(pRExC_state, NOTHING);
7482 *flagp |= flags&SIMPLE;
7489 - regpiece - something followed by possible [*+?]
7491 * Note that the branching code sequences used for ? and the general cases
7492 * of * and + are somewhat optimized: they use the same NOTHING node as
7493 * both the endmarker for their branch list and the body of the last branch.
7494 * It might seem that this node could be dispensed with entirely, but the
7495 * endmarker role is not redundant.
7498 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7501 register regnode *ret;
7503 register char *next;
7505 const char * const origparse = RExC_parse;
7507 I32 max = REG_INFTY;
7509 const char *maxpos = NULL;
7510 GET_RE_DEBUG_FLAGS_DECL;
7512 PERL_ARGS_ASSERT_REGPIECE;
7514 DEBUG_PARSE("piec");
7516 ret = regatom(pRExC_state, &flags,depth+1);
7518 if (flags & TRYAGAIN)
7525 if (op == '{' && regcurly(RExC_parse)) {
7527 parse_start = RExC_parse; /* MJD */
7528 next = RExC_parse + 1;
7529 while (isDIGIT(*next) || *next == ',') {
7538 if (*next == '}') { /* got one */
7542 min = atoi(RExC_parse);
7546 maxpos = RExC_parse;
7548 if (!max && *maxpos != '0')
7549 max = REG_INFTY; /* meaning "infinity" */
7550 else if (max >= REG_INFTY)
7551 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
7553 nextchar(pRExC_state);
7556 if ((flags&SIMPLE)) {
7557 RExC_naughty += 2 + RExC_naughty / 2;
7558 reginsert(pRExC_state, CURLY, ret, depth+1);
7559 Set_Node_Offset(ret, parse_start+1); /* MJD */
7560 Set_Node_Cur_Length(ret);
7563 regnode * const w = reg_node(pRExC_state, WHILEM);
7566 REGTAIL(pRExC_state, ret, w);
7567 if (!SIZE_ONLY && RExC_extralen) {
7568 reginsert(pRExC_state, LONGJMP,ret, depth+1);
7569 reginsert(pRExC_state, NOTHING,ret, depth+1);
7570 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
7572 reginsert(pRExC_state, CURLYX,ret, depth+1);
7574 Set_Node_Offset(ret, parse_start+1);
7575 Set_Node_Length(ret,
7576 op == '{' ? (RExC_parse - parse_start) : 1);
7578 if (!SIZE_ONLY && RExC_extralen)
7579 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
7580 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
7582 RExC_whilem_seen++, RExC_extralen += 3;
7583 RExC_naughty += 4 + RExC_naughty; /* compound interest */
7592 vFAIL("Can't do {n,m} with n > m");
7594 ARG1_SET(ret, (U16)min);
7595 ARG2_SET(ret, (U16)max);
7607 #if 0 /* Now runtime fix should be reliable. */
7609 /* if this is reinstated, don't forget to put this back into perldiag:
7611 =item Regexp *+ operand could be empty at {#} in regex m/%s/
7613 (F) The part of the regexp subject to either the * or + quantifier
7614 could match an empty string. The {#} shows in the regular
7615 expression about where the problem was discovered.
7619 if (!(flags&HASWIDTH) && op != '?')
7620 vFAIL("Regexp *+ operand could be empty");
7623 parse_start = RExC_parse;
7624 nextchar(pRExC_state);
7626 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
7628 if (op == '*' && (flags&SIMPLE)) {
7629 reginsert(pRExC_state, STAR, ret, depth+1);
7633 else if (op == '*') {
7637 else if (op == '+' && (flags&SIMPLE)) {
7638 reginsert(pRExC_state, PLUS, ret, depth+1);
7642 else if (op == '+') {
7646 else if (op == '?') {
7651 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
7652 ckWARN3reg(RExC_parse,
7653 "%.*s matches null string many times",
7654 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
7658 if (RExC_parse < RExC_end && *RExC_parse == '?') {
7659 nextchar(pRExC_state);
7660 reginsert(pRExC_state, MINMOD, ret, depth+1);
7661 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
7663 #ifndef REG_ALLOW_MINMOD_SUSPEND
7666 if (RExC_parse < RExC_end && *RExC_parse == '+') {
7668 nextchar(pRExC_state);
7669 ender = reg_node(pRExC_state, SUCCEED);
7670 REGTAIL(pRExC_state, ret, ender);
7671 reginsert(pRExC_state, SUSPEND, ret, depth+1);
7673 ender = reg_node(pRExC_state, TAIL);
7674 REGTAIL(pRExC_state, ret, ender);
7678 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
7680 vFAIL("Nested quantifiers");
7687 /* reg_namedseq(pRExC_state,UVp, UV depth)
7689 This is expected to be called by a parser routine that has
7690 recognized '\N' and needs to handle the rest. RExC_parse is
7691 expected to point at the first char following the N at the time
7694 The \N may be inside (indicated by valuep not being NULL) or outside a
7697 \N may begin either a named sequence, or if outside a character class, mean
7698 to match a non-newline. For non single-quoted regexes, the tokenizer has
7699 attempted to decide which, and in the case of a named sequence converted it
7700 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
7701 where c1... are the characters in the sequence. For single-quoted regexes,
7702 the tokenizer passes the \N sequence through unchanged; this code will not
7703 attempt to determine this nor expand those. The net effect is that if the
7704 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
7705 signals that this \N occurrence means to match a non-newline.
7707 Only the \N{U+...} form should occur in a character class, for the same
7708 reason that '.' inside a character class means to just match a period: it
7709 just doesn't make sense.
7711 If valuep is non-null then it is assumed that we are parsing inside
7712 of a charclass definition and the first codepoint in the resolved
7713 string is returned via *valuep and the routine will return NULL.
7714 In this mode if a multichar string is returned from the charnames
7715 handler, a warning will be issued, and only the first char in the
7716 sequence will be examined. If the string returned is zero length
7717 then the value of *valuep is undefined and NON-NULL will
7718 be returned to indicate failure. (This will NOT be a valid pointer
7721 If valuep is null then it is assumed that we are parsing normal text and a
7722 new EXACT node is inserted into the program containing the resolved string,
7723 and a pointer to the new node is returned. But if the string is zero length
7724 a NOTHING node is emitted instead.
7726 On success RExC_parse is set to the char following the endbrace.
7727 Parsing failures will generate a fatal error via vFAIL(...)
7730 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp, U32 depth)
7732 char * endbrace; /* '}' following the name */
7733 regnode *ret = NULL;
7736 GET_RE_DEBUG_FLAGS_DECL;
7738 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
7742 /* The [^\n] meaning of \N ignores spaces and comments under the /x
7743 * modifier. The other meaning does not */
7744 p = (RExC_flags & RXf_PMf_EXTENDED)
7745 ? regwhite( pRExC_state, RExC_parse )
7748 /* Disambiguate between \N meaning a named character versus \N meaning
7749 * [^\n]. The former is assumed when it can't be the latter. */
7750 if (*p != '{' || regcurly(p)) {
7753 /* no bare \N in a charclass */
7754 vFAIL("\\N in a character class must be a named character: \\N{...}");
7756 nextchar(pRExC_state);
7757 ret = reg_node(pRExC_state, REG_ANY);
7758 *flagp |= HASWIDTH|SIMPLE;
7761 Set_Node_Length(ret, 1); /* MJD */
7765 /* Here, we have decided it should be a named sequence */
7767 /* The test above made sure that the next real character is a '{', but
7768 * under the /x modifier, it could be separated by space (or a comment and
7769 * \n) and this is not allowed (for consistency with \x{...} and the
7770 * tokenizer handling of \N{NAME}). */
7771 if (*RExC_parse != '{') {
7772 vFAIL("Missing braces on \\N{}");
7775 RExC_parse++; /* Skip past the '{' */
7777 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
7778 || ! (endbrace == RExC_parse /* nothing between the {} */
7779 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
7780 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
7782 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
7783 vFAIL("\\N{NAME} must be resolved by the lexer");
7786 if (endbrace == RExC_parse) { /* empty: \N{} */
7788 RExC_parse = endbrace + 1;
7789 return reg_node(pRExC_state,NOTHING);
7793 ckWARNreg(RExC_parse,
7794 "Ignoring zero length \\N{} in character class"
7796 RExC_parse = endbrace + 1;
7799 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
7802 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
7803 RExC_parse += 2; /* Skip past the 'U+' */
7805 if (valuep) { /* In a bracketed char class */
7806 /* We only pay attention to the first char of
7807 multichar strings being returned. I kinda wonder
7808 if this makes sense as it does change the behaviour
7809 from earlier versions, OTOH that behaviour was broken
7810 as well. XXX Solution is to recharacterize as
7811 [rest-of-class]|multi1|multi2... */
7813 STRLEN length_of_hex;
7814 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7815 | PERL_SCAN_DISALLOW_PREFIX
7816 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
7818 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
7819 if (endchar < endbrace) {
7820 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
7823 length_of_hex = (STRLEN)(endchar - RExC_parse);
7824 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
7826 /* The tokenizer should have guaranteed validity, but it's possible to
7827 * bypass it by using single quoting, so check */
7828 if (length_of_hex == 0
7829 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
7831 RExC_parse += length_of_hex; /* Includes all the valid */
7832 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
7833 ? UTF8SKIP(RExC_parse)
7835 /* Guard against malformed utf8 */
7836 if (RExC_parse >= endchar) RExC_parse = endchar;
7837 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7840 RExC_parse = endbrace + 1;
7841 if (endchar == endbrace) return NULL;
7843 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
7845 else { /* Not a char class */
7847 /* What is done here is to convert this to a sub-pattern of the form
7848 * (?:\x{char1}\x{char2}...)
7849 * and then call reg recursively. That way, it retains its atomicness,
7850 * while not having to worry about special handling that some code
7851 * points may have. toke.c has converted the original Unicode values
7852 * to native, so that we can just pass on the hex values unchanged. We
7853 * do have to set a flag to keep recoding from happening in the
7856 SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP);
7858 char *endchar; /* Points to '.' or '}' ending cur char in the input
7860 char *orig_end = RExC_end;
7862 while (RExC_parse < endbrace) {
7864 /* Code points are separated by dots. If none, there is only one
7865 * code point, and is terminated by the brace */
7866 endchar = RExC_parse + strcspn(RExC_parse, ".}");
7868 /* Convert to notation the rest of the code understands */
7869 sv_catpv(substitute_parse, "\\x{");
7870 sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse);
7871 sv_catpv(substitute_parse, "}");
7873 /* Point to the beginning of the next character in the sequence. */
7874 RExC_parse = endchar + 1;
7876 sv_catpv(substitute_parse, ")");
7878 RExC_parse = SvPV(substitute_parse, len);
7880 /* Don't allow empty number */
7882 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7884 RExC_end = RExC_parse + len;
7886 /* The values are Unicode, and therefore not subject to recoding */
7887 RExC_override_recoding = 1;
7889 ret = reg(pRExC_state, 1, flagp, depth+1);
7891 RExC_parse = endbrace;
7892 RExC_end = orig_end;
7893 RExC_override_recoding = 0;
7895 nextchar(pRExC_state);
7905 * It returns the code point in utf8 for the value in *encp.
7906 * value: a code value in the source encoding
7907 * encp: a pointer to an Encode object
7909 * If the result from Encode is not a single character,
7910 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
7913 S_reg_recode(pTHX_ const char value, SV **encp)
7916 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
7917 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
7918 const STRLEN newlen = SvCUR(sv);
7919 UV uv = UNICODE_REPLACEMENT;
7921 PERL_ARGS_ASSERT_REG_RECODE;
7925 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
7928 if (!newlen || numlen != newlen) {
7929 uv = UNICODE_REPLACEMENT;
7937 - regatom - the lowest level
7939 Try to identify anything special at the start of the pattern. If there
7940 is, then handle it as required. This may involve generating a single regop,
7941 such as for an assertion; or it may involve recursing, such as to
7942 handle a () structure.
7944 If the string doesn't start with something special then we gobble up
7945 as much literal text as we can.
7947 Once we have been able to handle whatever type of thing started the
7948 sequence, we return.
7950 Note: we have to be careful with escapes, as they can be both literal
7951 and special, and in the case of \10 and friends can either, depending
7952 on context. Specifically there are two separate switches for handling
7953 escape sequences, with the one for handling literal escapes requiring
7954 a dummy entry for all of the special escapes that are actually handled
7959 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7962 register regnode *ret = NULL;
7964 char *parse_start = RExC_parse;
7966 GET_RE_DEBUG_FLAGS_DECL;
7967 DEBUG_PARSE("atom");
7968 *flagp = WORST; /* Tentatively. */
7970 PERL_ARGS_ASSERT_REGATOM;
7973 switch ((U8)*RExC_parse) {
7975 RExC_seen_zerolen++;
7976 nextchar(pRExC_state);
7977 if (RExC_flags & RXf_PMf_MULTILINE)
7978 ret = reg_node(pRExC_state, MBOL);
7979 else if (RExC_flags & RXf_PMf_SINGLELINE)
7980 ret = reg_node(pRExC_state, SBOL);
7982 ret = reg_node(pRExC_state, BOL);
7983 Set_Node_Length(ret, 1); /* MJD */
7986 nextchar(pRExC_state);
7988 RExC_seen_zerolen++;
7989 if (RExC_flags & RXf_PMf_MULTILINE)
7990 ret = reg_node(pRExC_state, MEOL);
7991 else if (RExC_flags & RXf_PMf_SINGLELINE)
7992 ret = reg_node(pRExC_state, SEOL);
7994 ret = reg_node(pRExC_state, EOL);
7995 Set_Node_Length(ret, 1); /* MJD */
7998 nextchar(pRExC_state);
7999 if (RExC_flags & RXf_PMf_SINGLELINE)
8000 ret = reg_node(pRExC_state, SANY);
8002 ret = reg_node(pRExC_state, REG_ANY);
8003 *flagp |= HASWIDTH|SIMPLE;
8005 Set_Node_Length(ret, 1); /* MJD */
8009 char * const oregcomp_parse = ++RExC_parse;
8010 ret = regclass(pRExC_state,depth+1);
8011 if (*RExC_parse != ']') {
8012 RExC_parse = oregcomp_parse;
8013 vFAIL("Unmatched [");
8015 nextchar(pRExC_state);
8016 *flagp |= HASWIDTH|SIMPLE;
8017 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
8021 nextchar(pRExC_state);
8022 ret = reg(pRExC_state, 1, &flags,depth+1);
8024 if (flags & TRYAGAIN) {
8025 if (RExC_parse == RExC_end) {
8026 /* Make parent create an empty node if needed. */
8034 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
8038 if (flags & TRYAGAIN) {
8042 vFAIL("Internal urp");
8043 /* Supposed to be caught earlier. */
8046 if (!regcurly(RExC_parse)) {
8055 vFAIL("Quantifier follows nothing");
8060 This switch handles escape sequences that resolve to some kind
8061 of special regop and not to literal text. Escape sequnces that
8062 resolve to literal text are handled below in the switch marked
8065 Every entry in this switch *must* have a corresponding entry
8066 in the literal escape switch. However, the opposite is not
8067 required, as the default for this switch is to jump to the
8068 literal text handling code.
8070 switch ((U8)*++RExC_parse) {
8071 /* Special Escapes */
8073 RExC_seen_zerolen++;
8074 ret = reg_node(pRExC_state, SBOL);
8076 goto finish_meta_pat;
8078 ret = reg_node(pRExC_state, GPOS);
8079 RExC_seen |= REG_SEEN_GPOS;
8081 goto finish_meta_pat;
8083 RExC_seen_zerolen++;
8084 ret = reg_node(pRExC_state, KEEPS);
8086 /* XXX:dmq : disabling in-place substitution seems to
8087 * be necessary here to avoid cases of memory corruption, as
8088 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
8090 RExC_seen |= REG_SEEN_LOOKBEHIND;
8091 goto finish_meta_pat;
8093 ret = reg_node(pRExC_state, SEOL);
8095 RExC_seen_zerolen++; /* Do not optimize RE away */
8096 goto finish_meta_pat;
8098 ret = reg_node(pRExC_state, EOS);
8100 RExC_seen_zerolen++; /* Do not optimize RE away */
8101 goto finish_meta_pat;
8103 ret = reg_node(pRExC_state, CANY);
8104 RExC_seen |= REG_SEEN_CANY;
8105 *flagp |= HASWIDTH|SIMPLE;
8106 goto finish_meta_pat;
8108 ret = reg_node(pRExC_state, CLUMP);
8110 goto finish_meta_pat;
8112 switch (get_regex_charset(RExC_flags)) {
8113 case REGEX_LOCALE_CHARSET:
8116 case REGEX_UNICODE_CHARSET:
8119 case REGEX_ASCII_RESTRICTED_CHARSET:
8120 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8123 case REGEX_DEPENDS_CHARSET:
8129 ret = reg_node(pRExC_state, op);
8130 *flagp |= HASWIDTH|SIMPLE;
8131 goto finish_meta_pat;
8133 switch (get_regex_charset(RExC_flags)) {
8134 case REGEX_LOCALE_CHARSET:
8137 case REGEX_UNICODE_CHARSET:
8140 case REGEX_ASCII_RESTRICTED_CHARSET:
8141 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8144 case REGEX_DEPENDS_CHARSET:
8150 ret = reg_node(pRExC_state, op);
8151 *flagp |= HASWIDTH|SIMPLE;
8152 goto finish_meta_pat;
8154 RExC_seen_zerolen++;
8155 RExC_seen |= REG_SEEN_LOOKBEHIND;
8156 switch (get_regex_charset(RExC_flags)) {
8157 case REGEX_LOCALE_CHARSET:
8160 case REGEX_UNICODE_CHARSET:
8163 case REGEX_ASCII_RESTRICTED_CHARSET:
8164 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8167 case REGEX_DEPENDS_CHARSET:
8173 ret = reg_node(pRExC_state, op);
8174 FLAGS(ret) = get_regex_charset(RExC_flags);
8176 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8177 ckWARNregdep(RExC_parse, "\"\\b{\" is deprecated; use \"\\b\\{\" instead");
8179 goto finish_meta_pat;
8181 RExC_seen_zerolen++;
8182 RExC_seen |= REG_SEEN_LOOKBEHIND;
8183 switch (get_regex_charset(RExC_flags)) {
8184 case REGEX_LOCALE_CHARSET:
8187 case REGEX_UNICODE_CHARSET:
8190 case REGEX_ASCII_RESTRICTED_CHARSET:
8191 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8194 case REGEX_DEPENDS_CHARSET:
8200 ret = reg_node(pRExC_state, op);
8201 FLAGS(ret) = get_regex_charset(RExC_flags);
8203 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8204 ckWARNregdep(RExC_parse, "\"\\B{\" is deprecated; use \"\\B\\{\" instead");
8206 goto finish_meta_pat;
8208 switch (get_regex_charset(RExC_flags)) {
8209 case REGEX_LOCALE_CHARSET:
8212 case REGEX_UNICODE_CHARSET:
8215 case REGEX_ASCII_RESTRICTED_CHARSET:
8216 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8219 case REGEX_DEPENDS_CHARSET:
8225 ret = reg_node(pRExC_state, op);
8226 *flagp |= HASWIDTH|SIMPLE;
8227 goto finish_meta_pat;
8229 switch (get_regex_charset(RExC_flags)) {
8230 case REGEX_LOCALE_CHARSET:
8233 case REGEX_UNICODE_CHARSET:
8236 case REGEX_ASCII_RESTRICTED_CHARSET:
8237 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8240 case REGEX_DEPENDS_CHARSET:
8246 ret = reg_node(pRExC_state, op);
8247 *flagp |= HASWIDTH|SIMPLE;
8248 goto finish_meta_pat;
8250 switch (get_regex_charset(RExC_flags)) {
8251 case REGEX_LOCALE_CHARSET:
8254 case REGEX_ASCII_RESTRICTED_CHARSET:
8255 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8258 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8259 case REGEX_UNICODE_CHARSET:
8265 ret = reg_node(pRExC_state, op);
8266 *flagp |= HASWIDTH|SIMPLE;
8267 goto finish_meta_pat;
8269 switch (get_regex_charset(RExC_flags)) {
8270 case REGEX_LOCALE_CHARSET:
8273 case REGEX_ASCII_RESTRICTED_CHARSET:
8274 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8277 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8278 case REGEX_UNICODE_CHARSET:
8284 ret = reg_node(pRExC_state, op);
8285 *flagp |= HASWIDTH|SIMPLE;
8286 goto finish_meta_pat;
8288 ret = reg_node(pRExC_state, LNBREAK);
8289 *flagp |= HASWIDTH|SIMPLE;
8290 goto finish_meta_pat;
8292 ret = reg_node(pRExC_state, HORIZWS);
8293 *flagp |= HASWIDTH|SIMPLE;
8294 goto finish_meta_pat;
8296 ret = reg_node(pRExC_state, NHORIZWS);
8297 *flagp |= HASWIDTH|SIMPLE;
8298 goto finish_meta_pat;
8300 ret = reg_node(pRExC_state, VERTWS);
8301 *flagp |= HASWIDTH|SIMPLE;
8302 goto finish_meta_pat;
8304 ret = reg_node(pRExC_state, NVERTWS);
8305 *flagp |= HASWIDTH|SIMPLE;
8307 nextchar(pRExC_state);
8308 Set_Node_Length(ret, 2); /* MJD */
8313 char* const oldregxend = RExC_end;
8315 char* parse_start = RExC_parse - 2;
8318 if (RExC_parse[1] == '{') {
8319 /* a lovely hack--pretend we saw [\pX] instead */
8320 RExC_end = strchr(RExC_parse, '}');
8322 const U8 c = (U8)*RExC_parse;
8324 RExC_end = oldregxend;
8325 vFAIL2("Missing right brace on \\%c{}", c);
8330 RExC_end = RExC_parse + 2;
8331 if (RExC_end > oldregxend)
8332 RExC_end = oldregxend;
8336 ret = regclass(pRExC_state,depth+1);
8338 RExC_end = oldregxend;
8341 Set_Node_Offset(ret, parse_start + 2);
8342 Set_Node_Cur_Length(ret);
8343 nextchar(pRExC_state);
8344 *flagp |= HASWIDTH|SIMPLE;
8348 /* Handle \N and \N{NAME} here and not below because it can be
8349 multicharacter. join_exact() will join them up later on.
8350 Also this makes sure that things like /\N{BLAH}+/ and
8351 \N{BLAH} being multi char Just Happen. dmq*/
8353 ret= reg_namedseq(pRExC_state, NULL, flagp, depth);
8355 case 'k': /* Handle \k<NAME> and \k'NAME' */
8358 char ch= RExC_parse[1];
8359 if (ch != '<' && ch != '\'' && ch != '{') {
8361 vFAIL2("Sequence %.2s... not terminated",parse_start);
8363 /* this pretty much dupes the code for (?P=...) in reg(), if
8364 you change this make sure you change that */
8365 char* name_start = (RExC_parse += 2);
8367 SV *sv_dat = reg_scan_name(pRExC_state,
8368 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
8369 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
8370 if (RExC_parse == name_start || *RExC_parse != ch)
8371 vFAIL2("Sequence %.3s... not terminated",parse_start);
8374 num = add_data( pRExC_state, 1, "S" );
8375 RExC_rxi->data->data[num]=(void*)sv_dat;
8376 SvREFCNT_inc_simple_void(sv_dat);
8380 ret = reganode(pRExC_state,
8383 : (MORE_ASCII_RESTRICTED)
8385 : (AT_LEAST_UNI_SEMANTICS)
8393 /* override incorrect value set in reganode MJD */
8394 Set_Node_Offset(ret, parse_start+1);
8395 Set_Node_Cur_Length(ret); /* MJD */
8396 nextchar(pRExC_state);
8402 case '1': case '2': case '3': case '4':
8403 case '5': case '6': case '7': case '8': case '9':
8406 bool isg = *RExC_parse == 'g';
8411 if (*RExC_parse == '{') {
8415 if (*RExC_parse == '-') {
8419 if (hasbrace && !isDIGIT(*RExC_parse)) {
8420 if (isrel) RExC_parse--;
8422 goto parse_named_seq;
8424 num = atoi(RExC_parse);
8425 if (isg && num == 0)
8426 vFAIL("Reference to invalid group 0");
8428 num = RExC_npar - num;
8430 vFAIL("Reference to nonexistent or unclosed group");
8432 if (!isg && num > 9 && num >= RExC_npar)
8435 char * const parse_start = RExC_parse - 1; /* MJD */
8436 while (isDIGIT(*RExC_parse))
8438 if (parse_start == RExC_parse - 1)
8439 vFAIL("Unterminated \\g... pattern");
8441 if (*RExC_parse != '}')
8442 vFAIL("Unterminated \\g{...} pattern");
8446 if (num > (I32)RExC_rx->nparens)
8447 vFAIL("Reference to nonexistent group");
8450 ret = reganode(pRExC_state,
8453 : (MORE_ASCII_RESTRICTED)
8455 : (AT_LEAST_UNI_SEMANTICS)
8463 /* override incorrect value set in reganode MJD */
8464 Set_Node_Offset(ret, parse_start+1);
8465 Set_Node_Cur_Length(ret); /* MJD */
8467 nextchar(pRExC_state);
8472 if (RExC_parse >= RExC_end)
8473 FAIL("Trailing \\");
8476 /* Do not generate "unrecognized" warnings here, we fall
8477 back into the quick-grab loop below */
8484 if (RExC_flags & RXf_PMf_EXTENDED) {
8485 if ( reg_skipcomment( pRExC_state ) )
8492 parse_start = RExC_parse - 1;
8505 char_state latest_char_state = generic_char;
8506 register STRLEN len;
8511 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
8512 regnode * orig_emit;
8515 orig_emit = RExC_emit; /* Save the original output node position in
8516 case we need to output a different node
8518 ret = reg_node(pRExC_state,
8519 (U8) ((! FOLD) ? EXACT
8522 : (MORE_ASCII_RESTRICTED)
8524 : (AT_LEAST_UNI_SEMANTICS)
8529 for (len = 0, p = RExC_parse - 1;
8530 len < 127 && p < RExC_end;
8533 char * const oldp = p;
8535 if (RExC_flags & RXf_PMf_EXTENDED)
8536 p = regwhite( pRExC_state, p );
8547 /* Literal Escapes Switch
8549 This switch is meant to handle escape sequences that
8550 resolve to a literal character.
8552 Every escape sequence that represents something
8553 else, like an assertion or a char class, is handled
8554 in the switch marked 'Special Escapes' above in this
8555 routine, but also has an entry here as anything that
8556 isn't explicitly mentioned here will be treated as
8557 an unescaped equivalent literal.
8561 /* These are all the special escapes. */
8562 case 'A': /* Start assertion */
8563 case 'b': case 'B': /* Word-boundary assertion*/
8564 case 'C': /* Single char !DANGEROUS! */
8565 case 'd': case 'D': /* digit class */
8566 case 'g': case 'G': /* generic-backref, pos assertion */
8567 case 'h': case 'H': /* HORIZWS */
8568 case 'k': case 'K': /* named backref, keep marker */
8569 case 'N': /* named char sequence */
8570 case 'p': case 'P': /* Unicode property */
8571 case 'R': /* LNBREAK */
8572 case 's': case 'S': /* space class */
8573 case 'v': case 'V': /* VERTWS */
8574 case 'w': case 'W': /* word class */
8575 case 'X': /* eXtended Unicode "combining character sequence" */
8576 case 'z': case 'Z': /* End of line/string assertion */
8580 /* Anything after here is an escape that resolves to a
8581 literal. (Except digits, which may or may not)
8600 ender = ASCII_TO_NATIVE('\033');
8604 ender = ASCII_TO_NATIVE('\007');
8609 STRLEN brace_len = len;
8611 const char* error_msg;
8613 bool valid = grok_bslash_o(p,
8620 RExC_parse = p; /* going to die anyway; point
8621 to exact spot of failure */
8628 if (PL_encoding && ender < 0x100) {
8629 goto recode_encoding;
8638 char* const e = strchr(p, '}');
8642 vFAIL("Missing right brace on \\x{}");
8645 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8646 | PERL_SCAN_DISALLOW_PREFIX;
8647 STRLEN numlen = e - p - 1;
8648 ender = grok_hex(p + 1, &numlen, &flags, NULL);
8655 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8657 ender = grok_hex(p, &numlen, &flags, NULL);
8660 if (PL_encoding && ender < 0x100)
8661 goto recode_encoding;
8665 ender = grok_bslash_c(*p++, UTF, SIZE_ONLY);
8667 case '0': case '1': case '2': case '3':case '4':
8668 case '5': case '6': case '7': case '8':case '9':
8670 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
8672 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8674 ender = grok_oct(p, &numlen, &flags, NULL);
8684 if (PL_encoding && ender < 0x100)
8685 goto recode_encoding;
8688 if (! RExC_override_recoding) {
8689 SV* enc = PL_encoding;
8690 ender = reg_recode((const char)(U8)ender, &enc);
8691 if (!enc && SIZE_ONLY)
8692 ckWARNreg(p, "Invalid escape in the specified encoding");
8698 FAIL("Trailing \\");
8701 if (!SIZE_ONLY&& isALPHA(*p)) {
8702 /* Include any { following the alpha to emphasize
8703 * that it could be part of an escape at some point
8705 int len = (*(p + 1) == '{') ? 2 : 1;
8706 ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p);
8708 goto normal_default;
8713 if (UTF8_IS_START(*p) && UTF) {
8715 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
8716 &numlen, UTF8_ALLOW_DEFAULT);
8722 } /* End of switch on the literal */
8724 /* Certain characters are problematic because their folded
8725 * length is so different from their original length that it
8726 * isn't handleable by the optimizer. They are therefore not
8727 * placed in an EXACTish node; and are here handled specially.
8728 * (Even if the optimizer handled LATIN_SMALL_LETTER_SHARP_S,
8729 * putting it in a special node keeps regexec from having to
8730 * deal with a non-utf8 multi-char fold */
8732 && (ender > 255 || (! MORE_ASCII_RESTRICTED && ! LOC)))
8734 /* We look for either side of the fold. For example \xDF
8735 * folds to 'ss'. We look for both the single character
8736 * \xDF and the sequence 'ss'. When we find something that
8737 * could be one of those, we stop and flush whatever we
8738 * have output so far into the EXACTish node that was being
8739 * built. Then restore the input pointer to what it was.
8740 * regatom will return that EXACT node, and will be called
8741 * again, positioned so the first character is the one in
8742 * question, which we return in a different node type.
8743 * The multi-char folds are a sequence, so the occurrence
8744 * of the first character in that sequence doesn't
8745 * necessarily mean that what follows is the rest of the
8746 * sequence. We keep track of that with a state machine,
8747 * with the state being set to the latest character
8748 * processed before the current one. Most characters will
8749 * set the state to 0, but if one occurs that is part of a
8750 * potential tricky fold sequence, the state is set to that
8751 * character, and the next loop iteration sees if the state
8752 * should progress towards the final folded-from character,
8753 * or if it was a false alarm. If it turns out to be a
8754 * false alarm, the character(s) will be output in a new
8755 * EXACTish node, and join_exact() will later combine them.
8756 * In the case of the 'ss' sequence, which is more common
8757 * and more easily checked, some look-ahead is done to
8758 * save time by ruling-out some false alarms */
8761 latest_char_state = generic_char;
8765 if (AT_LEAST_UNI_SEMANTICS) {
8766 if (latest_char_state == char_s) { /* 'ss' */
8767 ender = LATIN_SMALL_LETTER_SHARP_S;
8770 else if (p < RExC_end) {
8772 /* Look-ahead at the next character. If it
8773 * is also an s, we handle as a sharp s
8774 * tricky regnode. */
8775 if (*p == 's' || *p == 'S') {
8777 /* But first flush anything in the
8778 * EXACTish buffer */
8783 p++; /* Account for swallowing this
8785 ender = LATIN_SMALL_LETTER_SHARP_S;
8788 /* Here, the next character is not a
8789 * literal 's', but still could
8790 * evaluate to one if part of a \o{},
8791 * \x or \OCTAL-DIGIT. The minimum
8792 * length required for that is 4, eg
8796 && (isDIGIT(*(p + 1))
8798 || *(p + 1) == 'o' ))
8801 /* Here, it could be an 's', too much
8802 * bother to figure it out here. Flush
8803 * the buffer if any; when come back
8804 * here, set the state so know that the
8805 * previous char was an 's' */
8807 latest_char_state = generic_char;
8811 latest_char_state = char_s;
8817 /* Here, can't be an 'ss' sequence, or at least not
8818 * one that could fold to/from the sharp ss */
8819 latest_char_state = generic_char;
8821 case 0x03C5: /* First char in upsilon series */
8822 if (p < RExC_end - 4) { /* Need >= 4 bytes left */
8823 latest_char_state = upsilon_1;
8830 latest_char_state = generic_char;
8833 case 0x03B9: /* First char in iota series */
8834 if (p < RExC_end - 4) {
8835 latest_char_state = iota_1;
8842 latest_char_state = generic_char;
8846 if (latest_char_state == upsilon_1) {
8847 latest_char_state = upsilon_2;
8849 else if (latest_char_state == iota_1) {
8850 latest_char_state = iota_2;
8853 latest_char_state = generic_char;
8857 if (latest_char_state == upsilon_2) {
8858 ender = GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS;
8861 else if (latest_char_state == iota_2) {
8862 ender = GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS;
8865 latest_char_state = generic_char;
8868 /* These are the tricky fold characters. Flush any
8870 case GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS:
8871 case GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS:
8872 case LATIN_SMALL_LETTER_SHARP_S:
8873 case LATIN_CAPITAL_LETTER_SHARP_S:
8882 char* const oldregxend = RExC_end;
8883 U8 tmpbuf[UTF8_MAXBYTES+1];
8885 /* Here, we know we need to generate a special
8886 * regnode, and 'ender' contains the tricky
8887 * character. What's done is to pretend it's in a
8888 * [bracketed] class, and let the code that deals
8889 * with those handle it, as that code has all the
8890 * intelligence necessary. First save the current
8891 * parse state, get rid of the already allocated
8892 * but empty EXACT node that the ANYOFV node will
8893 * replace, and point the parse to a buffer which
8894 * we fill with the character we want the regclass
8895 * code to think is being parsed */
8896 RExC_emit = orig_emit;
8897 RExC_parse = (char *) tmpbuf;
8899 U8 *d = uvchr_to_utf8(tmpbuf, ender);
8901 RExC_end = (char *) d;
8903 else { /* ender above 255 already excluded */
8904 tmpbuf[0] = (U8) ender;
8906 RExC_end = RExC_parse + 1;
8909 ret = regclass(pRExC_state,depth+1);
8911 /* Here, have parsed the buffer. Reset the parse to
8912 * the actual input, and return */
8913 RExC_end = oldregxend;
8916 Set_Node_Offset(ret, RExC_parse);
8917 Set_Node_Cur_Length(ret);
8918 nextchar(pRExC_state);
8919 *flagp |= HASWIDTH|SIMPLE;
8925 if ( RExC_flags & RXf_PMf_EXTENDED)
8926 p = regwhite( pRExC_state, p );
8928 /* Prime the casefolded buffer. Locale rules, which apply
8929 * only to code points < 256, aren't known until execution,
8930 * so for them, just output the original character using
8932 if (LOC && ender < 256) {
8933 if (UNI_IS_INVARIANT(ender)) {
8934 *tmpbuf = (U8) ender;
8937 *tmpbuf = UTF8_TWO_BYTE_HI(ender);
8938 *(tmpbuf + 1) = UTF8_TWO_BYTE_LO(ender);
8942 else if (isASCII(ender)) { /* Note: Here can't also be LOC
8944 ender = toLOWER(ender);
8945 *tmpbuf = (U8) ender;
8948 else if (! MORE_ASCII_RESTRICTED && ! LOC) {
8950 /* Locale and /aa require more selectivity about the
8951 * fold, so are handled below. Otherwise, here, just
8953 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
8956 /* Under locale rules or /aa we are not to mix,
8957 * respectively, ords < 256 or ASCII with non-. So
8958 * reject folds that mix them, using only the
8959 * non-folded code point. So do the fold to a
8960 * temporary, and inspect each character in it. */
8961 U8 trialbuf[UTF8_MAXBYTES_CASE+1];
8963 UV tmpender = toFOLD_uni(ender, trialbuf, &foldlen);
8964 U8* e = s + foldlen;
8965 bool fold_ok = TRUE;
8969 || (LOC && (UTF8_IS_INVARIANT(*s)
8970 || UTF8_IS_DOWNGRADEABLE_START(*s))))
8978 Copy(trialbuf, tmpbuf, foldlen, U8);
8982 uvuni_to_utf8(tmpbuf, ender);
8983 foldlen = UNISKIP(ender);
8987 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
8992 /* Emit all the Unicode characters. */
8994 for (foldbuf = tmpbuf;
8996 foldlen -= numlen) {
8997 ender = utf8_to_uvchr(foldbuf, &numlen);
8999 const STRLEN unilen = reguni(pRExC_state, ender, s);
9002 /* In EBCDIC the numlen
9003 * and unilen can differ. */
9005 if (numlen >= foldlen)
9009 break; /* "Can't happen." */
9013 const STRLEN unilen = reguni(pRExC_state, ender, s);
9022 REGC((char)ender, s++);
9028 /* Emit all the Unicode characters. */
9030 for (foldbuf = tmpbuf;
9032 foldlen -= numlen) {
9033 ender = utf8_to_uvchr(foldbuf, &numlen);
9035 const STRLEN unilen = reguni(pRExC_state, ender, s);
9038 /* In EBCDIC the numlen
9039 * and unilen can differ. */
9041 if (numlen >= foldlen)
9049 const STRLEN unilen = reguni(pRExC_state, ender, s);
9058 REGC((char)ender, s++);
9061 loopdone: /* Jumped to when encounters something that shouldn't be in
9064 Set_Node_Cur_Length(ret); /* MJD */
9065 nextchar(pRExC_state);
9067 /* len is STRLEN which is unsigned, need to copy to signed */
9070 vFAIL("Internal disaster");
9074 if (len == 1 && UNI_IS_INVARIANT(ender))
9078 RExC_size += STR_SZ(len);
9081 RExC_emit += STR_SZ(len);
9089 /* Jumped to when an unrecognized character set is encountered */
9091 Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags));
9096 S_regwhite( RExC_state_t *pRExC_state, char *p )
9098 const char *e = RExC_end;
9100 PERL_ARGS_ASSERT_REGWHITE;
9105 else if (*p == '#') {
9114 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
9122 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
9123 Character classes ([:foo:]) can also be negated ([:^foo:]).
9124 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
9125 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
9126 but trigger failures because they are currently unimplemented. */
9128 #define POSIXCC_DONE(c) ((c) == ':')
9129 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
9130 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
9133 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
9136 I32 namedclass = OOB_NAMEDCLASS;
9138 PERL_ARGS_ASSERT_REGPPOSIXCC;
9140 if (value == '[' && RExC_parse + 1 < RExC_end &&
9141 /* I smell either [: or [= or [. -- POSIX has been here, right? */
9142 POSIXCC(UCHARAT(RExC_parse))) {
9143 const char c = UCHARAT(RExC_parse);
9144 char* const s = RExC_parse++;
9146 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
9148 if (RExC_parse == RExC_end)
9149 /* Grandfather lone [:, [=, [. */
9152 const char* const t = RExC_parse++; /* skip over the c */
9155 if (UCHARAT(RExC_parse) == ']') {
9156 const char *posixcc = s + 1;
9157 RExC_parse++; /* skip over the ending ] */
9160 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
9161 const I32 skip = t - posixcc;
9163 /* Initially switch on the length of the name. */
9166 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
9167 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
9170 /* Names all of length 5. */
9171 /* alnum alpha ascii blank cntrl digit graph lower
9172 print punct space upper */
9173 /* Offset 4 gives the best switch position. */
9174 switch (posixcc[4]) {
9176 if (memEQ(posixcc, "alph", 4)) /* alpha */
9177 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
9180 if (memEQ(posixcc, "spac", 4)) /* space */
9181 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
9184 if (memEQ(posixcc, "grap", 4)) /* graph */
9185 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
9188 if (memEQ(posixcc, "asci", 4)) /* ascii */
9189 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
9192 if (memEQ(posixcc, "blan", 4)) /* blank */
9193 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
9196 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
9197 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
9200 if (memEQ(posixcc, "alnu", 4)) /* alnum */
9201 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
9204 if (memEQ(posixcc, "lowe", 4)) /* lower */
9205 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
9206 else if (memEQ(posixcc, "uppe", 4)) /* upper */
9207 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
9210 if (memEQ(posixcc, "digi", 4)) /* digit */
9211 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
9212 else if (memEQ(posixcc, "prin", 4)) /* print */
9213 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
9214 else if (memEQ(posixcc, "punc", 4)) /* punct */
9215 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
9220 if (memEQ(posixcc, "xdigit", 6))
9221 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
9225 if (namedclass == OOB_NAMEDCLASS)
9226 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
9228 assert (posixcc[skip] == ':');
9229 assert (posixcc[skip+1] == ']');
9230 } else if (!SIZE_ONLY) {
9231 /* [[=foo=]] and [[.foo.]] are still future. */
9233 /* adjust RExC_parse so the warning shows after
9235 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
9237 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9240 /* Maternal grandfather:
9241 * "[:" ending in ":" but not in ":]" */
9251 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
9255 PERL_ARGS_ASSERT_CHECKPOSIXCC;
9257 if (POSIXCC(UCHARAT(RExC_parse))) {
9258 const char *s = RExC_parse;
9259 const char c = *s++;
9263 if (*s && c == *s && s[1] == ']') {
9265 "POSIX syntax [%c %c] belongs inside character classes",
9268 /* [[=foo=]] and [[.foo.]] are still future. */
9269 if (POSIXCC_NOTYET(c)) {
9270 /* adjust RExC_parse so the error shows after
9272 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
9274 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9280 /* No locale test, and always Unicode semantics */
9281 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
9283 for (value = 0; value < 256; value++) \
9285 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9289 case ANYOF_N##NAME: \
9290 for (value = 0; value < 256; value++) \
9292 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9297 /* Like the above, but there are differences if we are in uni-8-bit or not, so
9298 * there are two tests passed in, to use depending on that. There aren't any
9299 * cases where the label is different from the name, so no need for that
9301 #define _C_C_T_(NAME, TEST_8, TEST_7, WORD) \
9303 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
9304 else if (UNI_SEMANTICS) { \
9305 for (value = 0; value < 256; value++) { \
9306 if (TEST_8(value)) stored += \
9307 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9311 for (value = 0; value < 128; value++) { \
9312 if (TEST_7(UNI_TO_NATIVE(value))) stored += \
9313 set_regclass_bit(pRExC_state, ret, \
9314 (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9320 case ANYOF_N##NAME: \
9321 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
9322 else if (UNI_SEMANTICS) { \
9323 for (value = 0; value < 256; value++) { \
9324 if (! TEST_8(value)) stored += \
9325 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9329 for (value = 0; value < 128; value++) { \
9330 if (! TEST_7(UNI_TO_NATIVE(value))) stored += set_regclass_bit( \
9331 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9333 if (AT_LEAST_ASCII_RESTRICTED) { \
9334 for (value = 128; value < 256; value++) { \
9335 stored += set_regclass_bit( \
9336 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9338 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; \
9341 /* For a non-ut8 target string with DEPENDS semantics, all above \
9342 * ASCII Latin1 code points match the complement of any of the \
9343 * classes. But in utf8, they have their Unicode semantics, so \
9344 * can't just set them in the bitmap, or else regexec.c will think \
9345 * they matched when they shouldn't. */ \
9346 ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; \
9354 S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr)
9357 /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes.
9358 * Locale folding is done at run-time, so this function should not be
9359 * called for nodes that are for locales.
9361 * This function sets the bit corresponding to the fold of the input
9362 * 'value', if not already set. The fold of 'f' is 'F', and the fold of
9365 * It also knows about the characters that are in the bitmap that have
9366 * folds that are matchable only outside it, and sets the appropriate lists
9369 * It returns the number of bits that actually changed from 0 to 1 */
9374 PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD;
9376 fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value]
9379 /* It assumes the bit for 'value' has already been set */
9380 if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) {
9381 ANYOF_BITMAP_SET(node, fold);
9384 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) {
9385 /* Certain Latin1 characters have matches outside the bitmap. To get
9386 * here, 'value' is one of those characters. None of these matches is
9387 * valid for ASCII characters under /aa, which have been excluded by
9388 * the 'if' above. The matches fall into three categories:
9389 * 1) They are singly folded-to or -from an above 255 character, as
9390 * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y
9392 * 2) They are part of a multi-char fold with another character in the
9393 * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill;
9394 * 3) They are part of a multi-char fold with a character not in the
9395 * bitmap, such as various ligatures.
9396 * We aren't dealing fully with multi-char folds, except we do deal
9397 * with the pattern containing a character that has a multi-char fold
9398 * (not so much the inverse).
9399 * For types 1) and 3), the matches only happen when the target string
9400 * is utf8; that's not true for 2), and we set a flag for it.
9402 * The code below adds to the passed in inversion list the single fold
9403 * closures for 'value'. The values are hard-coded here so that an
9404 * innocent-looking character class, like /[ks]/i won't have to go out
9405 * to disk to find the possible matches. XXX It would be better to
9406 * generate these via regen, in case a new version of the Unicode
9407 * standard adds new mappings, though that is not really likely. */
9412 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A);
9416 /* LATIN SMALL LETTER LONG S */
9417 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F);
9420 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9421 GREEK_SMALL_LETTER_MU);
9422 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9423 GREEK_CAPITAL_LETTER_MU);
9425 case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
9426 case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
9428 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B);
9429 if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */
9430 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9431 PL_fold_latin1[value]);
9434 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
9435 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9436 LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
9438 case LATIN_SMALL_LETTER_SHARP_S:
9439 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9440 LATIN_CAPITAL_LETTER_SHARP_S);
9442 /* Under /a, /d, and /u, this can match the two chars "ss" */
9443 if (! MORE_ASCII_RESTRICTED) {
9444 add_alternate(alternate_ptr, (U8 *) "ss", 2);
9446 /* And under /u or /a, it can match even if the target is
9448 if (AT_LEAST_UNI_SEMANTICS) {
9449 ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8;
9463 /* These all are targets of multi-character folds from code
9464 * points that require UTF8 to express, so they can't match
9465 * unless the target string is in UTF-8, so no action here is
9466 * necessary, as regexec.c properly handles the general case
9467 * for UTF-8 matching */
9470 /* Use deprecated warning to increase the chances of this
9472 ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value);
9476 else if (DEPENDS_SEMANTICS
9478 && PL_fold_latin1[value] != value)
9480 /* Under DEPENDS rules, non-ASCII Latin1 characters match their
9481 * folds only when the target string is in UTF-8. We add the fold
9482 * here to the list of things to match outside the bitmap, which
9483 * won't be looked at unless it is UTF8 (or else if something else
9484 * says to look even if not utf8, but those things better not happen
9485 * under DEPENDS semantics. */
9486 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]);
9493 PERL_STATIC_INLINE U8
9494 S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr)
9496 /* This inline function sets a bit in the bitmap if not already set, and if
9497 * appropriate, its fold, returning the number of bits that actually
9498 * changed from 0 to 1 */
9502 PERL_ARGS_ASSERT_SET_REGCLASS_BIT;
9504 if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */
9508 ANYOF_BITMAP_SET(node, value);
9511 if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */
9512 stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr);
9519 S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len)
9521 /* Adds input 'string' with length 'len' to the ANYOF node's unicode
9522 * alternate list, pointed to by 'alternate_ptr'. This is an array of
9523 * the multi-character folds of characters in the node */
9526 PERL_ARGS_ASSERT_ADD_ALTERNATE;
9528 if (! *alternate_ptr) {
9529 *alternate_ptr = newAV();
9531 sv = newSVpvn_utf8((char*)string, len, TRUE);
9532 av_push(*alternate_ptr, sv);
9537 parse a class specification and produce either an ANYOF node that
9538 matches the pattern or perhaps will be optimized into an EXACTish node
9539 instead. The node contains a bit map for the first 256 characters, with the
9540 corresponding bit set if that character is in the list. For characters
9541 above 255, a range list is used */
9544 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
9547 register UV nextvalue;
9548 register IV prevvalue = OOB_UNICODE;
9549 register IV range = 0;
9550 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
9551 register regnode *ret;
9554 char *rangebegin = NULL;
9555 bool need_class = 0;
9556 bool allow_full_fold = TRUE; /* Assume wants multi-char folding */
9558 STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more
9559 than just initialized. */
9562 /* code points this node matches that can't be stored in the bitmap */
9563 HV* nonbitmap = NULL;
9565 /* The items that are to match that aren't stored in the bitmap, but are a
9566 * result of things that are stored there. This is the fold closure of
9567 * such a character, either because it has DEPENDS semantics and shouldn't
9568 * be matched unless the target string is utf8, or is a code point that is
9569 * too large for the bit map, as for example, the fold of the MICRO SIGN is
9570 * above 255. This all is solely for performance reasons. By having this
9571 * code know the outside-the-bitmap folds that the bitmapped characters are
9572 * involved with, we don't have to go out to disk to find the list of
9573 * matches, unless the character class includes code points that aren't
9574 * storable in the bit map. That means that a character class with an 's'
9575 * in it, for example, doesn't need to go out to disk to find everything
9576 * that matches. A 2nd list is used so that the 'nonbitmap' list is kept
9577 * empty unless there is something whose fold we don't know about, and will
9578 * have to go out to the disk to find. */
9579 HV* l1_fold_invlist = NULL;
9581 /* List of multi-character folds that are matched by this node */
9582 AV* unicode_alternate = NULL;
9584 UV literal_endpoint = 0;
9586 UV stored = 0; /* how many chars stored in the bitmap */
9588 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
9589 case we need to change the emitted regop to an EXACT. */
9590 const char * orig_parse = RExC_parse;
9591 GET_RE_DEBUG_FLAGS_DECL;
9593 PERL_ARGS_ASSERT_REGCLASS;
9595 PERL_UNUSED_ARG(depth);
9598 DEBUG_PARSE("clas");
9600 /* Assume we are going to generate an ANYOF node. */
9601 ret = reganode(pRExC_state, ANYOF, 0);
9605 ANYOF_FLAGS(ret) = 0;
9608 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
9612 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
9614 /* We have decided to not allow multi-char folds in inverted character
9615 * classes, due to the confusion that can happen, even with classes
9616 * that are designed for a non-Unicode world: You have the peculiar
9618 "s s" =~ /^[^\xDF]+$/i => Y
9619 "ss" =~ /^[^\xDF]+$/i => N
9621 * See [perl #89750] */
9622 allow_full_fold = FALSE;
9626 RExC_size += ANYOF_SKIP;
9627 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
9630 RExC_emit += ANYOF_SKIP;
9632 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
9634 ANYOF_BITMAP_ZERO(ret);
9635 listsv = newSVpvs("# comment\n");
9636 initial_listsv_len = SvCUR(listsv);
9639 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9641 if (!SIZE_ONLY && POSIXCC(nextvalue))
9642 checkposixcc(pRExC_state);
9644 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
9645 if (UCHARAT(RExC_parse) == ']')
9649 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
9653 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
9656 rangebegin = RExC_parse;
9658 value = utf8n_to_uvchr((U8*)RExC_parse,
9659 RExC_end - RExC_parse,
9660 &numlen, UTF8_ALLOW_DEFAULT);
9661 RExC_parse += numlen;
9664 value = UCHARAT(RExC_parse++);
9666 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9667 if (value == '[' && POSIXCC(nextvalue))
9668 namedclass = regpposixcc(pRExC_state, value);
9669 else if (value == '\\') {
9671 value = utf8n_to_uvchr((U8*)RExC_parse,
9672 RExC_end - RExC_parse,
9673 &numlen, UTF8_ALLOW_DEFAULT);
9674 RExC_parse += numlen;
9677 value = UCHARAT(RExC_parse++);
9678 /* Some compilers cannot handle switching on 64-bit integer
9679 * values, therefore value cannot be an UV. Yes, this will
9680 * be a problem later if we want switch on Unicode.
9681 * A similar issue a little bit later when switching on
9682 * namedclass. --jhi */
9683 switch ((I32)value) {
9684 case 'w': namedclass = ANYOF_ALNUM; break;
9685 case 'W': namedclass = ANYOF_NALNUM; break;
9686 case 's': namedclass = ANYOF_SPACE; break;
9687 case 'S': namedclass = ANYOF_NSPACE; break;
9688 case 'd': namedclass = ANYOF_DIGIT; break;
9689 case 'D': namedclass = ANYOF_NDIGIT; break;
9690 case 'v': namedclass = ANYOF_VERTWS; break;
9691 case 'V': namedclass = ANYOF_NVERTWS; break;
9692 case 'h': namedclass = ANYOF_HORIZWS; break;
9693 case 'H': namedclass = ANYOF_NHORIZWS; break;
9694 case 'N': /* Handle \N{NAME} in class */
9696 /* We only pay attention to the first char of
9697 multichar strings being returned. I kinda wonder
9698 if this makes sense as it does change the behaviour
9699 from earlier versions, OTOH that behaviour was broken
9701 UV v; /* value is register so we cant & it /grrr */
9702 if (reg_namedseq(pRExC_state, &v, NULL, depth)) {
9712 if (RExC_parse >= RExC_end)
9713 vFAIL2("Empty \\%c{}", (U8)value);
9714 if (*RExC_parse == '{') {
9715 const U8 c = (U8)value;
9716 e = strchr(RExC_parse++, '}');
9718 vFAIL2("Missing right brace on \\%c{}", c);
9719 while (isSPACE(UCHARAT(RExC_parse)))
9721 if (e == RExC_parse)
9722 vFAIL2("Empty \\%c{}", c);
9724 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
9732 if (UCHARAT(RExC_parse) == '^') {
9735 value = value == 'p' ? 'P' : 'p'; /* toggle */
9736 while (isSPACE(UCHARAT(RExC_parse))) {
9742 /* Add the property name to the list. If /i matching, give
9743 * a different name which consists of the normal name
9744 * sandwiched between two underscores and '_i'. The design
9745 * is discussed in the commit message for this. */
9746 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s%.*s%s\n",
9747 (value=='p' ? '+' : '!'),
9756 /* The \p could match something in the Latin1 range, hence
9757 * something that isn't utf8 */
9758 ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8;
9759 namedclass = ANYOF_MAX; /* no official name, but it's named */
9761 /* \p means they want Unicode semantics */
9762 RExC_uni_semantics = 1;
9765 case 'n': value = '\n'; break;
9766 case 'r': value = '\r'; break;
9767 case 't': value = '\t'; break;
9768 case 'f': value = '\f'; break;
9769 case 'b': value = '\b'; break;
9770 case 'e': value = ASCII_TO_NATIVE('\033');break;
9771 case 'a': value = ASCII_TO_NATIVE('\007');break;
9773 RExC_parse--; /* function expects to be pointed at the 'o' */
9775 const char* error_msg;
9776 bool valid = grok_bslash_o(RExC_parse,
9781 RExC_parse += numlen;
9786 if (PL_encoding && value < 0x100) {
9787 goto recode_encoding;
9791 if (*RExC_parse == '{') {
9792 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
9793 | PERL_SCAN_DISALLOW_PREFIX;
9794 char * const e = strchr(RExC_parse++, '}');
9796 vFAIL("Missing right brace on \\x{}");
9798 numlen = e - RExC_parse;
9799 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
9803 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
9805 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
9806 RExC_parse += numlen;
9808 if (PL_encoding && value < 0x100)
9809 goto recode_encoding;
9812 value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY);
9814 case '0': case '1': case '2': case '3': case '4':
9815 case '5': case '6': case '7':
9817 /* Take 1-3 octal digits */
9818 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
9820 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
9821 RExC_parse += numlen;
9822 if (PL_encoding && value < 0x100)
9823 goto recode_encoding;
9827 if (! RExC_override_recoding) {
9828 SV* enc = PL_encoding;
9829 value = reg_recode((const char)(U8)value, &enc);
9830 if (!enc && SIZE_ONLY)
9831 ckWARNreg(RExC_parse,
9832 "Invalid escape in the specified encoding");
9836 /* Allow \_ to not give an error */
9837 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
9838 ckWARN2reg(RExC_parse,
9839 "Unrecognized escape \\%c in character class passed through",
9844 } /* end of \blah */
9850 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
9852 /* What matches in a locale is not known until runtime, so need to
9853 * (one time per class) allocate extra space to pass to regexec.
9854 * The space will contain a bit for each named class that is to be
9855 * matched against. This isn't needed for \p{} and pseudo-classes,
9856 * as they are not affected by locale, and hence are dealt with
9858 if (LOC && namedclass < ANYOF_MAX && ! need_class) {
9861 RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP;
9864 RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP;
9865 ANYOF_CLASS_ZERO(ret);
9867 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
9870 /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a
9871 * literal, as is the character that began the false range, i.e.
9872 * the 'a' in the examples */
9876 RExC_parse >= rangebegin ?
9877 RExC_parse - rangebegin : 0;
9878 ckWARN4reg(RExC_parse,
9879 "False [] range \"%*.*s\"",
9883 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
9884 if (prevvalue < 256) {
9886 set_regclass_bit(pRExC_state, ret, (U8) prevvalue, &l1_fold_invlist, &unicode_alternate);
9889 nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue);
9893 range = 0; /* this was not a true range */
9899 const char *what = NULL;
9902 /* Possible truncation here but in some 64-bit environments
9903 * the compiler gets heartburn about switch on 64-bit values.
9904 * A similar issue a little earlier when switching on value.
9906 switch ((I32)namedclass) {
9908 case _C_C_T_(ALNUMC, isALNUMC_L1, isALNUMC, "XPosixAlnum");
9909 case _C_C_T_(ALPHA, isALPHA_L1, isALPHA, "XPosixAlpha");
9910 case _C_C_T_(BLANK, isBLANK_L1, isBLANK, "XPosixBlank");
9911 case _C_C_T_(CNTRL, isCNTRL_L1, isCNTRL, "XPosixCntrl");
9912 case _C_C_T_(GRAPH, isGRAPH_L1, isGRAPH, "XPosixGraph");
9913 case _C_C_T_(LOWER, isLOWER_L1, isLOWER, "XPosixLower");
9914 case _C_C_T_(PRINT, isPRINT_L1, isPRINT, "XPosixPrint");
9915 case _C_C_T_(PSXSPC, isPSXSPC_L1, isPSXSPC, "XPosixSpace");
9916 case _C_C_T_(PUNCT, isPUNCT_L1, isPUNCT, "XPosixPunct");
9917 case _C_C_T_(UPPER, isUPPER_L1, isUPPER, "XPosixUpper");
9918 /* \s, \w match all unicode if utf8. */
9919 case _C_C_T_(SPACE, isSPACE_L1, isSPACE, "SpacePerl");
9920 case _C_C_T_(ALNUM, isWORDCHAR_L1, isALNUM, "Word");
9921 case _C_C_T_(XDIGIT, isXDIGIT_L1, isXDIGIT, "XPosixXDigit");
9922 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
9923 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
9926 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
9928 for (value = 0; value < 128; value++)
9930 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
9933 what = NULL; /* Doesn't match outside ascii, so
9934 don't want to add +utf8:: */
9938 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
9940 for (value = 128; value < 256; value++)
9942 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
9944 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
9950 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
9952 /* consecutive digits assumed */
9953 for (value = '0'; value <= '9'; value++)
9955 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9962 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
9964 /* consecutive digits assumed */
9965 for (value = 0; value < '0'; value++)
9967 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9968 for (value = '9' + 1; value < 256; value++)
9970 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9974 if (AT_LEAST_ASCII_RESTRICTED ) {
9975 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
9979 /* this is to handle \p and \P */
9982 vFAIL("Invalid [::] class");
9985 if (what && ! (AT_LEAST_ASCII_RESTRICTED)) {
9986 /* Strings such as "+utf8::isWord\n" */
9987 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
9992 } /* end of namedclass \blah */
9995 if (prevvalue > (IV)value) /* b-a */ {
9996 const int w = RExC_parse - rangebegin;
9997 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
9998 range = 0; /* not a valid range */
10002 prevvalue = value; /* save the beginning of the range */
10003 if (RExC_parse+1 < RExC_end
10004 && *RExC_parse == '-'
10005 && RExC_parse[1] != ']')
10009 /* a bad range like \w-, [:word:]- ? */
10010 if (namedclass > OOB_NAMEDCLASS) {
10011 if (ckWARN(WARN_REGEXP)) {
10013 RExC_parse >= rangebegin ?
10014 RExC_parse - rangebegin : 0;
10016 "False [] range \"%*.*s\"",
10021 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
10023 range = 1; /* yeah, it's a range! */
10024 continue; /* but do it the next time */
10028 /* non-Latin1 code point implies unicode semantics. Must be set in
10029 * pass1 so is there for the whole of pass 2 */
10031 RExC_uni_semantics = 1;
10034 /* now is the next time */
10036 if (prevvalue < 256) {
10037 const IV ceilvalue = value < 256 ? value : 255;
10040 /* In EBCDIC [\x89-\x91] should include
10041 * the \x8e but [i-j] should not. */
10042 if (literal_endpoint == 2 &&
10043 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
10044 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
10046 if (isLOWER(prevvalue)) {
10047 for (i = prevvalue; i <= ceilvalue; i++)
10048 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10050 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10053 for (i = prevvalue; i <= ceilvalue; i++)
10054 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10056 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10062 for (i = prevvalue; i <= ceilvalue; i++) {
10063 stored += set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10067 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
10068 const UV natvalue = NATIVE_TO_UNI(value);
10069 nonbitmap = add_range_to_invlist(nonbitmap, prevnatvalue, natvalue);
10072 literal_endpoint = 0;
10076 range = 0; /* this range (if it was one) is done now */
10083 /****** !SIZE_ONLY AFTER HERE *********/
10085 /* If folding and there are code points above 255, we calculate all
10086 * characters that could fold to or from the ones already on the list */
10087 if (FOLD && nonbitmap) {
10090 HV* fold_intersection;
10093 /* This is a list of all the characters that participate in folds
10094 * (except marks, etc in multi-char folds */
10095 if (! PL_utf8_foldable) {
10096 SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0);
10097 PL_utf8_foldable = _swash_to_invlist(swash);
10100 /* This is a hash that for a particular fold gives all characters
10101 * that are involved in it */
10102 if (! PL_utf8_foldclosures) {
10104 /* If we were unable to find any folds, then we likely won't be
10105 * able to find the closures. So just create an empty list.
10106 * Folding will effectively be restricted to the non-Unicode rules
10107 * hard-coded into Perl. (This case happens legitimately during
10108 * compilation of Perl itself before the Unicode tables are
10110 if (invlist_len(PL_utf8_foldable) == 0) {
10111 PL_utf8_foldclosures = _new_invlist(0);
10113 /* If the folds haven't been read in, call a fold function
10115 if (! PL_utf8_tofold) {
10116 U8 dummy[UTF8_MAXBYTES+1];
10118 to_utf8_fold((U8*) "A", dummy, &dummy_len);
10120 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
10124 /* Only the characters in this class that participate in folds need
10125 * be checked. Get the intersection of this class and all the
10126 * possible characters that are foldable. This can quickly narrow
10127 * down a large class */
10128 fold_intersection = invlist_intersection(PL_utf8_foldable, nonbitmap);
10130 /* Now look at the foldable characters in this class individually */
10131 fold_list = invlist_array(fold_intersection);
10132 for (i = 0; i < invlist_len(fold_intersection); i++) {
10135 /* The next entry is the beginning of the range that is in the
10137 UV start = fold_list[i++];
10140 /* The next entry is the beginning of the next range, which
10141 * isn't in the class, so the end of the current range is one
10142 * less than that */
10143 UV end = fold_list[i] - 1;
10145 /* Look at every character in the range */
10146 for (j = start; j <= end; j++) {
10149 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
10152 _to_uni_fold_flags(j, foldbuf, &foldlen, allow_full_fold);
10154 if (foldlen > (STRLEN)UNISKIP(f)) {
10156 /* Any multicharacter foldings (disallowed in
10157 * lookbehind patterns) require the following
10158 * transform: [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst) where
10159 * E folds into "pq" and F folds into "rst", all other
10160 * characters fold to single characters. We save away
10161 * these multicharacter foldings, to be later saved as
10162 * part of the additional "s" data. */
10163 if (! RExC_in_lookbehind) {
10165 U8* e = foldbuf + foldlen;
10167 /* If any of the folded characters of this are in
10168 * the Latin1 range, tell the regex engine that
10169 * this can match a non-utf8 target string. The
10170 * only multi-byte fold whose source is in the
10171 * Latin1 range (U+00DF) applies only when the
10172 * target string is utf8, or under unicode rules */
10173 if (j > 255 || AT_LEAST_UNI_SEMANTICS) {
10176 /* Can't mix ascii with non- under /aa */
10177 if (MORE_ASCII_RESTRICTED
10178 && (isASCII(*loc) != isASCII(j)))
10180 goto end_multi_fold;
10182 if (UTF8_IS_INVARIANT(*loc)
10183 || UTF8_IS_DOWNGRADEABLE_START(*loc))
10185 /* Can't mix above and below 256 under
10188 goto end_multi_fold;
10191 |= ANYOF_NONBITMAP_NON_UTF8;
10194 loc += UTF8SKIP(loc);
10198 add_alternate(&unicode_alternate, foldbuf, foldlen);
10202 /* This is special-cased, as it is the only letter which
10203 * has both a multi-fold and single-fold in Latin1. All
10204 * the other chars that have single and multi-folds are
10205 * always in utf8, and the utf8 folding algorithm catches
10207 if (! LOC && j == LATIN_CAPITAL_LETTER_SHARP_S) {
10208 stored += set_regclass_bit(pRExC_state,
10210 LATIN_SMALL_LETTER_SHARP_S,
10211 &l1_fold_invlist, &unicode_alternate);
10215 /* Single character fold. Add everything in its fold
10216 * closure to the list that this node should match */
10219 /* The fold closures data structure is a hash with the
10220 * keys being every character that is folded to, like
10221 * 'k', and the values each an array of everything that
10222 * folds to its key. e.g. [ 'k', 'K', KELVIN_SIGN ] */
10223 if ((listp = hv_fetch(PL_utf8_foldclosures,
10224 (char *) foldbuf, foldlen, FALSE)))
10226 AV* list = (AV*) *listp;
10228 for (k = 0; k <= av_len(list); k++) {
10229 SV** c_p = av_fetch(list, k, FALSE);
10232 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
10236 /* /aa doesn't allow folds between ASCII and
10237 * non-; /l doesn't allow them between above
10239 if ((MORE_ASCII_RESTRICTED
10240 && (isASCII(c) != isASCII(j)))
10241 || (LOC && ((c < 256) != (j < 256))))
10246 if (c < 256 && AT_LEAST_UNI_SEMANTICS) {
10247 stored += set_regclass_bit(pRExC_state,
10250 &l1_fold_invlist, &unicode_alternate);
10252 /* It may be that the code point is already
10253 * in this range or already in the bitmap,
10254 * in which case we need do nothing */
10255 else if ((c < start || c > end)
10257 || ! ANYOF_BITMAP_TEST(ret, c)))
10259 nonbitmap = add_cp_to_invlist(nonbitmap, c);
10266 invlist_destroy(fold_intersection);
10269 /* Combine the two lists into one. */
10270 if (l1_fold_invlist) {
10272 nonbitmap = invlist_union(nonbitmap, l1_fold_invlist);
10275 nonbitmap = l1_fold_invlist;
10279 /* Here, we have calculated what code points should be in the character
10280 * class. Now we can see about various optimizations. Fold calculation
10281 * needs to take place before inversion. Otherwise /[^k]/i would invert to
10282 * include K, which under /i would match k. */
10284 /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't
10285 * set the FOLD flag yet, so this this does optimize those. It doesn't
10286 * optimize locale. Doing so perhaps could be done as long as there is
10287 * nothing like \w in it; some thought also would have to be given to the
10288 * interaction with above 0x100 chars */
10290 && (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT
10291 && ! unicode_alternate
10293 && SvCUR(listsv) == initial_listsv_len)
10295 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
10296 ANYOF_BITMAP(ret)[value] ^= 0xFF;
10297 stored = 256 - stored;
10299 /* The inversion means that everything above 255 is matched; and at the
10300 * same time we clear the invert flag */
10301 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
10304 /* Folding in the bitmap is taken care of above, but not for locale (for
10305 * which we have to wait to see what folding is in effect at runtime), and
10306 * for things not in the bitmap. Set run-time fold flag for these */
10307 if (FOLD && (LOC || nonbitmap || unicode_alternate)) {
10308 ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD;
10311 /* A single character class can be "optimized" into an EXACTish node.
10312 * Note that since we don't currently count how many characters there are
10313 * outside the bitmap, we are XXX missing optimization possibilities for
10314 * them. This optimization can't happen unless this is a truly single
10315 * character class, which means that it can't be an inversion into a
10316 * many-character class, and there must be no possibility of there being
10317 * things outside the bitmap. 'stored' (only) for locales doesn't include
10318 * \w, etc, so have to make a special test that they aren't present
10320 * Similarly A 2-character class of the very special form like [bB] can be
10321 * optimized into an EXACTFish node, but only for non-locales, and for
10322 * characters which only have the two folds; so things like 'fF' and 'Ii'
10323 * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE
10326 && ! unicode_alternate
10327 && SvCUR(listsv) == initial_listsv_len
10328 && ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL))
10329 && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10330 || (! ANYOF_CLASS_TEST_ANY_SET(ret)))))
10331 || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10332 && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value))
10333 /* If the latest code point has a fold whose
10334 * bit is set, it must be the only other one */
10335 && ((prevvalue = PL_fold_latin1[value]) != (IV)value)
10336 && ANYOF_BITMAP_TEST(ret, prevvalue)))))
10338 /* Note that the information needed to decide to do this optimization
10339 * is not currently available until the 2nd pass, and that the actually
10340 * used EXACTish node takes less space than the calculated ANYOF node,
10341 * and hence the amount of space calculated in the first pass is larger
10342 * than actually used, so this optimization doesn't gain us any space.
10343 * But an EXACT node is faster than an ANYOF node, and can be combined
10344 * with any adjacent EXACT nodes later by the optimizer for further
10345 * gains. The speed of executing an EXACTF is similar to an ANYOF
10346 * node, so the optimization advantage comes from the ability to join
10347 * it to adjacent EXACT nodes */
10349 const char * cur_parse= RExC_parse;
10351 RExC_emit = (regnode *)orig_emit;
10352 RExC_parse = (char *)orig_parse;
10356 /* A locale node with one point can be folded; all the other cases
10357 * with folding will have two points, since we calculate them above
10359 if (ANYOF_FLAGS(ret) & ANYOF_LOC_NONBITMAP_FOLD) {
10365 } /* else 2 chars in the bit map: the folds of each other */
10366 else if (AT_LEAST_UNI_SEMANTICS || !isASCII(value)) {
10368 /* To join adjacent nodes, they must be the exact EXACTish type.
10369 * Try to use the most likely type, by using EXACTFU if the regex
10370 * calls for them, or is required because the character is
10374 else { /* Otherwise, more likely to be EXACTF type */
10378 ret = reg_node(pRExC_state, op);
10379 RExC_parse = (char *)cur_parse;
10380 if (UTF && ! NATIVE_IS_INVARIANT(value)) {
10381 *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value);
10382 *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value);
10384 RExC_emit += STR_SZ(2);
10387 *STRING(ret)= (char)value;
10389 RExC_emit += STR_SZ(1);
10391 SvREFCNT_dec(listsv);
10396 UV* nonbitmap_array = invlist_array(nonbitmap);
10397 UV nonbitmap_len = invlist_len(nonbitmap);
10400 /* Here have the full list of items to match that aren't in the
10401 * bitmap. Convert to the structure that the rest of the code is
10402 * expecting. XXX That rest of the code should convert to this
10404 for (i = 0; i < nonbitmap_len; i++) {
10406 /* The next entry is the beginning of the range that is in the
10408 UV start = nonbitmap_array[i++];
10411 /* The next entry is the beginning of the next range, which isn't
10412 * in the class, so the end of the current range is one less than
10413 * that. But if there is no next range, it means that the range
10414 * begun by 'start' extends to infinity, which for this platform
10415 * ends at UV_MAX */
10416 if (i == nonbitmap_len) {
10420 end = nonbitmap_array[i] - 1;
10423 if (start == end) {
10424 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", start);
10427 /* The \t sets the whole range */
10428 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
10433 invlist_destroy(nonbitmap);
10436 if (SvCUR(listsv) == initial_listsv_len && ! unicode_alternate) {
10437 ARG_SET(ret, ANYOF_NONBITMAP_EMPTY);
10438 SvREFCNT_dec(listsv);
10439 SvREFCNT_dec(unicode_alternate);
10443 AV * const av = newAV();
10445 /* The 0th element stores the character class description
10446 * in its textual form: used later (regexec.c:Perl_regclass_swash())
10447 * to initialize the appropriate swash (which gets stored in
10448 * the 1st element), and also useful for dumping the regnode.
10449 * The 2nd element stores the multicharacter foldings,
10450 * used later (regexec.c:S_reginclass()). */
10451 av_store(av, 0, listsv);
10452 av_store(av, 1, NULL);
10454 /* Store any computed multi-char folds only if we are allowing
10456 if (allow_full_fold) {
10457 av_store(av, 2, MUTABLE_SV(unicode_alternate));
10458 if (unicode_alternate) { /* This node is variable length */
10463 av_store(av, 2, NULL);
10465 rv = newRV_noinc(MUTABLE_SV(av));
10466 n = add_data(pRExC_state, 1, "s");
10467 RExC_rxi->data->data[n] = (void*)rv;
10475 /* reg_skipcomment()
10477 Absorbs an /x style # comments from the input stream.
10478 Returns true if there is more text remaining in the stream.
10479 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
10480 terminates the pattern without including a newline.
10482 Note its the callers responsibility to ensure that we are
10483 actually in /x mode
10488 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
10492 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
10494 while (RExC_parse < RExC_end)
10495 if (*RExC_parse++ == '\n') {
10500 /* we ran off the end of the pattern without ending
10501 the comment, so we have to add an \n when wrapping */
10502 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
10510 Advances the parse position, and optionally absorbs
10511 "whitespace" from the inputstream.
10513 Without /x "whitespace" means (?#...) style comments only,
10514 with /x this means (?#...) and # comments and whitespace proper.
10516 Returns the RExC_parse point from BEFORE the scan occurs.
10518 This is the /x friendly way of saying RExC_parse++.
10522 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
10524 char* const retval = RExC_parse++;
10526 PERL_ARGS_ASSERT_NEXTCHAR;
10529 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
10530 RExC_parse[2] == '#') {
10531 while (*RExC_parse != ')') {
10532 if (RExC_parse == RExC_end)
10533 FAIL("Sequence (?#... not terminated");
10539 if (RExC_flags & RXf_PMf_EXTENDED) {
10540 if (isSPACE(*RExC_parse)) {
10544 else if (*RExC_parse == '#') {
10545 if ( reg_skipcomment( pRExC_state ) )
10554 - reg_node - emit a node
10556 STATIC regnode * /* Location. */
10557 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
10560 register regnode *ptr;
10561 regnode * const ret = RExC_emit;
10562 GET_RE_DEBUG_FLAGS_DECL;
10564 PERL_ARGS_ASSERT_REG_NODE;
10567 SIZE_ALIGN(RExC_size);
10571 if (RExC_emit >= RExC_emit_bound)
10572 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10574 NODE_ALIGN_FILL(ret);
10576 FILL_ADVANCE_NODE(ptr, op);
10577 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 1);
10578 #ifdef RE_TRACK_PATTERN_OFFSETS
10579 if (RExC_offsets) { /* MJD */
10580 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
10581 "reg_node", __LINE__,
10583 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
10584 ? "Overwriting end of array!\n" : "OK",
10585 (UV)(RExC_emit - RExC_emit_start),
10586 (UV)(RExC_parse - RExC_start),
10587 (UV)RExC_offsets[0]));
10588 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
10596 - reganode - emit a node with an argument
10598 STATIC regnode * /* Location. */
10599 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
10602 register regnode *ptr;
10603 regnode * const ret = RExC_emit;
10604 GET_RE_DEBUG_FLAGS_DECL;
10606 PERL_ARGS_ASSERT_REGANODE;
10609 SIZE_ALIGN(RExC_size);
10614 assert(2==regarglen[op]+1);
10616 Anything larger than this has to allocate the extra amount.
10617 If we changed this to be:
10619 RExC_size += (1 + regarglen[op]);
10621 then it wouldn't matter. Its not clear what side effect
10622 might come from that so its not done so far.
10627 if (RExC_emit >= RExC_emit_bound)
10628 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10630 NODE_ALIGN_FILL(ret);
10632 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
10633 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (ptr) - 2);
10634 #ifdef RE_TRACK_PATTERN_OFFSETS
10635 if (RExC_offsets) { /* MJD */
10636 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10640 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
10641 "Overwriting end of array!\n" : "OK",
10642 (UV)(RExC_emit - RExC_emit_start),
10643 (UV)(RExC_parse - RExC_start),
10644 (UV)RExC_offsets[0]));
10645 Set_Cur_Node_Offset;
10653 - reguni - emit (if appropriate) a Unicode character
10656 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
10660 PERL_ARGS_ASSERT_REGUNI;
10662 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
10666 - reginsert - insert an operator in front of already-emitted operand
10668 * Means relocating the operand.
10671 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
10674 register regnode *src;
10675 register regnode *dst;
10676 register regnode *place;
10677 const int offset = regarglen[(U8)op];
10678 const int size = NODE_STEP_REGNODE + offset;
10679 GET_RE_DEBUG_FLAGS_DECL;
10681 PERL_ARGS_ASSERT_REGINSERT;
10682 PERL_UNUSED_ARG(depth);
10683 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
10684 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
10693 if (RExC_open_parens) {
10695 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
10696 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
10697 if ( RExC_open_parens[paren] >= opnd ) {
10698 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
10699 RExC_open_parens[paren] += size;
10701 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
10703 if ( RExC_close_parens[paren] >= opnd ) {
10704 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
10705 RExC_close_parens[paren] += size;
10707 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
10712 while (src > opnd) {
10713 StructCopy(--src, --dst, regnode);
10714 #ifdef RE_TRACK_PATTERN_OFFSETS
10715 if (RExC_offsets) { /* MJD 20010112 */
10716 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
10720 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
10721 ? "Overwriting end of array!\n" : "OK",
10722 (UV)(src - RExC_emit_start),
10723 (UV)(dst - RExC_emit_start),
10724 (UV)RExC_offsets[0]));
10725 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
10726 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
10732 place = opnd; /* Op node, where operand used to be. */
10733 #ifdef RE_TRACK_PATTERN_OFFSETS
10734 if (RExC_offsets) { /* MJD */
10735 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10739 (UV)(place - RExC_emit_start) > RExC_offsets[0]
10740 ? "Overwriting end of array!\n" : "OK",
10741 (UV)(place - RExC_emit_start),
10742 (UV)(RExC_parse - RExC_start),
10743 (UV)RExC_offsets[0]));
10744 Set_Node_Offset(place, RExC_parse);
10745 Set_Node_Length(place, 1);
10748 src = NEXTOPER(place);
10749 FILL_ADVANCE_NODE(place, op);
10750 REH_CALL_REGCOMP_HOOK(pRExC_state->rx, (place) - 1);
10751 Zero(src, offset, regnode);
10755 - regtail - set the next-pointer at the end of a node chain of p to val.
10756 - SEE ALSO: regtail_study
10758 /* TODO: All three parms should be const */
10760 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
10763 register regnode *scan;
10764 GET_RE_DEBUG_FLAGS_DECL;
10766 PERL_ARGS_ASSERT_REGTAIL;
10768 PERL_UNUSED_ARG(depth);
10774 /* Find last node. */
10777 regnode * const temp = regnext(scan);
10779 SV * const mysv=sv_newmortal();
10780 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
10781 regprop(RExC_rx, mysv, scan);
10782 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
10783 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
10784 (temp == NULL ? "->" : ""),
10785 (temp == NULL ? PL_reg_name[OP(val)] : "")
10793 if (reg_off_by_arg[OP(scan)]) {
10794 ARG_SET(scan, val - scan);
10797 NEXT_OFF(scan) = val - scan;
10803 - regtail_study - set the next-pointer at the end of a node chain of p to val.
10804 - Look for optimizable sequences at the same time.
10805 - currently only looks for EXACT chains.
10807 This is experimental code. The idea is to use this routine to perform
10808 in place optimizations on branches and groups as they are constructed,
10809 with the long term intention of removing optimization from study_chunk so
10810 that it is purely analytical.
10812 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
10813 to control which is which.
10816 /* TODO: All four parms should be const */
10819 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
10822 register regnode *scan;
10824 #ifdef EXPERIMENTAL_INPLACESCAN
10827 GET_RE_DEBUG_FLAGS_DECL;
10829 PERL_ARGS_ASSERT_REGTAIL_STUDY;
10835 /* Find last node. */
10839 regnode * const temp = regnext(scan);
10840 #ifdef EXPERIMENTAL_INPLACESCAN
10841 if (PL_regkind[OP(scan)] == EXACT)
10842 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
10846 switch (OP(scan)) {
10852 if( exact == PSEUDO )
10854 else if ( exact != OP(scan) )
10863 SV * const mysv=sv_newmortal();
10864 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
10865 regprop(RExC_rx, mysv, scan);
10866 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
10867 SvPV_nolen_const(mysv),
10868 REG_NODE_NUM(scan),
10869 PL_reg_name[exact]);
10876 SV * const mysv_val=sv_newmortal();
10877 DEBUG_PARSE_MSG("");
10878 regprop(RExC_rx, mysv_val, val);
10879 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
10880 SvPV_nolen_const(mysv_val),
10881 (IV)REG_NODE_NUM(val),
10885 if (reg_off_by_arg[OP(scan)]) {
10886 ARG_SET(scan, val - scan);
10889 NEXT_OFF(scan) = val - scan;
10897 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
10901 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
10907 for (bit=0; bit<32; bit++) {
10908 if (flags & (1<<bit)) {
10909 if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */
10912 if (!set++ && lead)
10913 PerlIO_printf(Perl_debug_log, "%s",lead);
10914 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
10917 if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) {
10918 if (!set++ && lead) {
10919 PerlIO_printf(Perl_debug_log, "%s",lead);
10922 case REGEX_UNICODE_CHARSET:
10923 PerlIO_printf(Perl_debug_log, "UNICODE");
10925 case REGEX_LOCALE_CHARSET:
10926 PerlIO_printf(Perl_debug_log, "LOCALE");
10928 case REGEX_ASCII_RESTRICTED_CHARSET:
10929 PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED");
10931 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
10932 PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED");
10935 PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET");
10941 PerlIO_printf(Perl_debug_log, "\n");
10943 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
10949 Perl_regdump(pTHX_ const regexp *r)
10953 SV * const sv = sv_newmortal();
10954 SV *dsv= sv_newmortal();
10955 RXi_GET_DECL(r,ri);
10956 GET_RE_DEBUG_FLAGS_DECL;
10958 PERL_ARGS_ASSERT_REGDUMP;
10960 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
10962 /* Header fields of interest. */
10963 if (r->anchored_substr) {
10964 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
10965 RE_SV_DUMPLEN(r->anchored_substr), 30);
10966 PerlIO_printf(Perl_debug_log,
10967 "anchored %s%s at %"IVdf" ",
10968 s, RE_SV_TAIL(r->anchored_substr),
10969 (IV)r->anchored_offset);
10970 } else if (r->anchored_utf8) {
10971 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
10972 RE_SV_DUMPLEN(r->anchored_utf8), 30);
10973 PerlIO_printf(Perl_debug_log,
10974 "anchored utf8 %s%s at %"IVdf" ",
10975 s, RE_SV_TAIL(r->anchored_utf8),
10976 (IV)r->anchored_offset);
10978 if (r->float_substr) {
10979 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
10980 RE_SV_DUMPLEN(r->float_substr), 30);
10981 PerlIO_printf(Perl_debug_log,
10982 "floating %s%s at %"IVdf"..%"UVuf" ",
10983 s, RE_SV_TAIL(r->float_substr),
10984 (IV)r->float_min_offset, (UV)r->float_max_offset);
10985 } else if (r->float_utf8) {
10986 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
10987 RE_SV_DUMPLEN(r->float_utf8), 30);
10988 PerlIO_printf(Perl_debug_log,
10989 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
10990 s, RE_SV_TAIL(r->float_utf8),
10991 (IV)r->float_min_offset, (UV)r->float_max_offset);
10993 if (r->check_substr || r->check_utf8)
10994 PerlIO_printf(Perl_debug_log,
10996 (r->check_substr == r->float_substr
10997 && r->check_utf8 == r->float_utf8
10998 ? "(checking floating" : "(checking anchored"));
10999 if (r->extflags & RXf_NOSCAN)
11000 PerlIO_printf(Perl_debug_log, " noscan");
11001 if (r->extflags & RXf_CHECK_ALL)
11002 PerlIO_printf(Perl_debug_log, " isall");
11003 if (r->check_substr || r->check_utf8)
11004 PerlIO_printf(Perl_debug_log, ") ");
11006 if (ri->regstclass) {
11007 regprop(r, sv, ri->regstclass);
11008 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
11010 if (r->extflags & RXf_ANCH) {
11011 PerlIO_printf(Perl_debug_log, "anchored");
11012 if (r->extflags & RXf_ANCH_BOL)
11013 PerlIO_printf(Perl_debug_log, "(BOL)");
11014 if (r->extflags & RXf_ANCH_MBOL)
11015 PerlIO_printf(Perl_debug_log, "(MBOL)");
11016 if (r->extflags & RXf_ANCH_SBOL)
11017 PerlIO_printf(Perl_debug_log, "(SBOL)");
11018 if (r->extflags & RXf_ANCH_GPOS)
11019 PerlIO_printf(Perl_debug_log, "(GPOS)");
11020 PerlIO_putc(Perl_debug_log, ' ');
11022 if (r->extflags & RXf_GPOS_SEEN)
11023 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
11024 if (r->intflags & PREGf_SKIP)
11025 PerlIO_printf(Perl_debug_log, "plus ");
11026 if (r->intflags & PREGf_IMPLICIT)
11027 PerlIO_printf(Perl_debug_log, "implicit ");
11028 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
11029 if (r->extflags & RXf_EVAL_SEEN)
11030 PerlIO_printf(Perl_debug_log, "with eval ");
11031 PerlIO_printf(Perl_debug_log, "\n");
11032 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
11034 PERL_ARGS_ASSERT_REGDUMP;
11035 PERL_UNUSED_CONTEXT;
11036 PERL_UNUSED_ARG(r);
11037 #endif /* DEBUGGING */
11041 - regprop - printable representation of opcode
11043 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
11046 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
11047 if (flags & ANYOF_INVERT) \
11048 /*make sure the invert info is in each */ \
11049 sv_catpvs(sv, "^"); \
11055 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
11060 RXi_GET_DECL(prog,progi);
11061 GET_RE_DEBUG_FLAGS_DECL;
11063 PERL_ARGS_ASSERT_REGPROP;
11067 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
11068 /* It would be nice to FAIL() here, but this may be called from
11069 regexec.c, and it would be hard to supply pRExC_state. */
11070 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
11071 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
11073 k = PL_regkind[OP(o)];
11076 sv_catpvs(sv, " ");
11077 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
11078 * is a crude hack but it may be the best for now since
11079 * we have no flag "this EXACTish node was UTF-8"
11081 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
11082 PERL_PV_ESCAPE_UNI_DETECT |
11083 PERL_PV_ESCAPE_NONASCII |
11084 PERL_PV_PRETTY_ELLIPSES |
11085 PERL_PV_PRETTY_LTGT |
11086 PERL_PV_PRETTY_NOCLEAR
11088 } else if (k == TRIE) {
11089 /* print the details of the trie in dumpuntil instead, as
11090 * progi->data isn't available here */
11091 const char op = OP(o);
11092 const U32 n = ARG(o);
11093 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
11094 (reg_ac_data *)progi->data->data[n] :
11096 const reg_trie_data * const trie
11097 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
11099 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
11100 DEBUG_TRIE_COMPILE_r(
11101 Perl_sv_catpvf(aTHX_ sv,
11102 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
11103 (UV)trie->startstate,
11104 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
11105 (UV)trie->wordcount,
11108 (UV)TRIE_CHARCOUNT(trie),
11109 (UV)trie->uniquecharcount
11112 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
11114 int rangestart = -1;
11115 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
11116 sv_catpvs(sv, "[");
11117 for (i = 0; i <= 256; i++) {
11118 if (i < 256 && BITMAP_TEST(bitmap,i)) {
11119 if (rangestart == -1)
11121 } else if (rangestart != -1) {
11122 if (i <= rangestart + 3)
11123 for (; rangestart < i; rangestart++)
11124 put_byte(sv, rangestart);
11126 put_byte(sv, rangestart);
11127 sv_catpvs(sv, "-");
11128 put_byte(sv, i - 1);
11133 sv_catpvs(sv, "]");
11136 } else if (k == CURLY) {
11137 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
11138 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
11139 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
11141 else if (k == WHILEM && o->flags) /* Ordinal/of */
11142 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
11143 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
11144 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
11145 if ( RXp_PAREN_NAMES(prog) ) {
11146 if ( k != REF || (OP(o) < NREF)) {
11147 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
11148 SV **name= av_fetch(list, ARG(o), 0 );
11150 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11153 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
11154 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
11155 I32 *nums=(I32*)SvPVX(sv_dat);
11156 SV **name= av_fetch(list, nums[0], 0 );
11159 for ( n=0; n<SvIVX(sv_dat); n++ ) {
11160 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
11161 (n ? "," : ""), (IV)nums[n]);
11163 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11167 } else if (k == GOSUB)
11168 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
11169 else if (k == VERB) {
11171 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
11172 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
11173 } else if (k == LOGICAL)
11174 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
11175 else if (k == FOLDCHAR)
11176 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
11177 else if (k == ANYOF) {
11178 int i, rangestart = -1;
11179 const U8 flags = ANYOF_FLAGS(o);
11182 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
11183 static const char * const anyofs[] = {
11216 if (flags & ANYOF_LOCALE)
11217 sv_catpvs(sv, "{loc}");
11218 if (flags & ANYOF_LOC_NONBITMAP_FOLD)
11219 sv_catpvs(sv, "{i}");
11220 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
11221 if (flags & ANYOF_INVERT)
11222 sv_catpvs(sv, "^");
11224 /* output what the standard cp 0-255 bitmap matches */
11225 for (i = 0; i <= 256; i++) {
11226 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
11227 if (rangestart == -1)
11229 } else if (rangestart != -1) {
11230 if (i <= rangestart + 3)
11231 for (; rangestart < i; rangestart++)
11232 put_byte(sv, rangestart);
11234 put_byte(sv, rangestart);
11235 sv_catpvs(sv, "-");
11236 put_byte(sv, i - 1);
11243 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11244 /* output any special charclass tests (used entirely under use locale) */
11245 if (ANYOF_CLASS_TEST_ANY_SET(o))
11246 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
11247 if (ANYOF_CLASS_TEST(o,i)) {
11248 sv_catpv(sv, anyofs[i]);
11252 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11254 if (flags & ANYOF_NON_UTF8_LATIN1_ALL) {
11255 sv_catpvs(sv, "{non-utf8-latin1-all}");
11258 /* output information about the unicode matching */
11259 if (flags & ANYOF_UNICODE_ALL)
11260 sv_catpvs(sv, "{unicode_all}");
11261 else if (ANYOF_NONBITMAP(o))
11262 sv_catpvs(sv, "{unicode}");
11263 if (flags & ANYOF_NONBITMAP_NON_UTF8)
11264 sv_catpvs(sv, "{outside bitmap}");
11266 if (ANYOF_NONBITMAP(o)) {
11268 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
11272 U8 s[UTF8_MAXBYTES_CASE+1];
11274 for (i = 0; i <= 256; i++) { /* just the first 256 */
11275 uvchr_to_utf8(s, i);
11277 if (i < 256 && swash_fetch(sw, s, TRUE)) {
11278 if (rangestart == -1)
11280 } else if (rangestart != -1) {
11281 if (i <= rangestart + 3)
11282 for (; rangestart < i; rangestart++) {
11283 const U8 * const e = uvchr_to_utf8(s,rangestart);
11285 for(p = s; p < e; p++)
11289 const U8 *e = uvchr_to_utf8(s,rangestart);
11291 for (p = s; p < e; p++)
11293 sv_catpvs(sv, "-");
11294 e = uvchr_to_utf8(s, i-1);
11295 for (p = s; p < e; p++)
11302 sv_catpvs(sv, "..."); /* et cetera */
11306 char *s = savesvpv(lv);
11307 char * const origs = s;
11309 while (*s && *s != '\n')
11313 const char * const t = ++s;
11331 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
11333 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
11334 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
11336 PERL_UNUSED_CONTEXT;
11337 PERL_UNUSED_ARG(sv);
11338 PERL_UNUSED_ARG(o);
11339 PERL_UNUSED_ARG(prog);
11340 #endif /* DEBUGGING */
11344 Perl_re_intuit_string(pTHX_ REGEXP * const r)
11345 { /* Assume that RE_INTUIT is set */
11347 struct regexp *const prog = (struct regexp *)SvANY(r);
11348 GET_RE_DEBUG_FLAGS_DECL;
11350 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
11351 PERL_UNUSED_CONTEXT;
11355 const char * const s = SvPV_nolen_const(prog->check_substr
11356 ? prog->check_substr : prog->check_utf8);
11358 if (!PL_colorset) reginitcolors();
11359 PerlIO_printf(Perl_debug_log,
11360 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
11362 prog->check_substr ? "" : "utf8 ",
11363 PL_colors[5],PL_colors[0],
11366 (strlen(s) > 60 ? "..." : ""));
11369 return prog->check_substr ? prog->check_substr : prog->check_utf8;
11375 handles refcounting and freeing the perl core regexp structure. When
11376 it is necessary to actually free the structure the first thing it
11377 does is call the 'free' method of the regexp_engine associated to
11378 the regexp, allowing the handling of the void *pprivate; member
11379 first. (This routine is not overridable by extensions, which is why
11380 the extensions free is called first.)
11382 See regdupe and regdupe_internal if you change anything here.
11384 #ifndef PERL_IN_XSUB_RE
11386 Perl_pregfree(pTHX_ REGEXP *r)
11392 Perl_pregfree2(pTHX_ REGEXP *rx)
11395 struct regexp *const r = (struct regexp *)SvANY(rx);
11396 GET_RE_DEBUG_FLAGS_DECL;
11398 PERL_ARGS_ASSERT_PREGFREE2;
11400 if (r->mother_re) {
11401 ReREFCNT_dec(r->mother_re);
11403 CALLREGFREE_PVT(rx); /* free the private data */
11404 SvREFCNT_dec(RXp_PAREN_NAMES(r));
11407 SvREFCNT_dec(r->anchored_substr);
11408 SvREFCNT_dec(r->anchored_utf8);
11409 SvREFCNT_dec(r->float_substr);
11410 SvREFCNT_dec(r->float_utf8);
11411 Safefree(r->substrs);
11413 RX_MATCH_COPY_FREE(rx);
11414 #ifdef PERL_OLD_COPY_ON_WRITE
11415 SvREFCNT_dec(r->saved_copy);
11422 This is a hacky workaround to the structural issue of match results
11423 being stored in the regexp structure which is in turn stored in
11424 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
11425 could be PL_curpm in multiple contexts, and could require multiple
11426 result sets being associated with the pattern simultaneously, such
11427 as when doing a recursive match with (??{$qr})
11429 The solution is to make a lightweight copy of the regexp structure
11430 when a qr// is returned from the code executed by (??{$qr}) this
11431 lightweight copy doesn't actually own any of its data except for
11432 the starp/end and the actual regexp structure itself.
11438 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
11440 struct regexp *ret;
11441 struct regexp *const r = (struct regexp *)SvANY(rx);
11442 register const I32 npar = r->nparens+1;
11444 PERL_ARGS_ASSERT_REG_TEMP_COPY;
11447 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
11448 ret = (struct regexp *)SvANY(ret_x);
11450 (void)ReREFCNT_inc(rx);
11451 /* We can take advantage of the existing "copied buffer" mechanism in SVs
11452 by pointing directly at the buffer, but flagging that the allocated
11453 space in the copy is zero. As we've just done a struct copy, it's now
11454 a case of zero-ing that, rather than copying the current length. */
11455 SvPV_set(ret_x, RX_WRAPPED(rx));
11456 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
11457 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
11458 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
11459 SvLEN_set(ret_x, 0);
11460 SvSTASH_set(ret_x, NULL);
11461 SvMAGIC_set(ret_x, NULL);
11462 Newx(ret->offs, npar, regexp_paren_pair);
11463 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11465 Newx(ret->substrs, 1, struct reg_substr_data);
11466 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11468 SvREFCNT_inc_void(ret->anchored_substr);
11469 SvREFCNT_inc_void(ret->anchored_utf8);
11470 SvREFCNT_inc_void(ret->float_substr);
11471 SvREFCNT_inc_void(ret->float_utf8);
11473 /* check_substr and check_utf8, if non-NULL, point to either their
11474 anchored or float namesakes, and don't hold a second reference. */
11476 RX_MATCH_COPIED_off(ret_x);
11477 #ifdef PERL_OLD_COPY_ON_WRITE
11478 ret->saved_copy = NULL;
11480 ret->mother_re = rx;
11486 /* regfree_internal()
11488 Free the private data in a regexp. This is overloadable by
11489 extensions. Perl takes care of the regexp structure in pregfree(),
11490 this covers the *pprivate pointer which technically perl doesn't
11491 know about, however of course we have to handle the
11492 regexp_internal structure when no extension is in use.
11494 Note this is called before freeing anything in the regexp
11499 Perl_regfree_internal(pTHX_ REGEXP * const rx)
11502 struct regexp *const r = (struct regexp *)SvANY(rx);
11503 RXi_GET_DECL(r,ri);
11504 GET_RE_DEBUG_FLAGS_DECL;
11506 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
11512 SV *dsv= sv_newmortal();
11513 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
11514 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
11515 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
11516 PL_colors[4],PL_colors[5],s);
11519 #ifdef RE_TRACK_PATTERN_OFFSETS
11521 Safefree(ri->u.offsets); /* 20010421 MJD */
11524 int n = ri->data->count;
11525 PAD* new_comppad = NULL;
11530 /* If you add a ->what type here, update the comment in regcomp.h */
11531 switch (ri->data->what[n]) {
11536 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
11539 Safefree(ri->data->data[n]);
11542 new_comppad = MUTABLE_AV(ri->data->data[n]);
11545 if (new_comppad == NULL)
11546 Perl_croak(aTHX_ "panic: pregfree comppad");
11547 PAD_SAVE_LOCAL(old_comppad,
11548 /* Watch out for global destruction's random ordering. */
11549 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
11552 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
11555 op_free((OP_4tree*)ri->data->data[n]);
11557 PAD_RESTORE_LOCAL(old_comppad);
11558 SvREFCNT_dec(MUTABLE_SV(new_comppad));
11559 new_comppad = NULL;
11564 { /* Aho Corasick add-on structure for a trie node.
11565 Used in stclass optimization only */
11567 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
11569 refcount = --aho->refcount;
11572 PerlMemShared_free(aho->states);
11573 PerlMemShared_free(aho->fail);
11574 /* do this last!!!! */
11575 PerlMemShared_free(ri->data->data[n]);
11576 PerlMemShared_free(ri->regstclass);
11582 /* trie structure. */
11584 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
11586 refcount = --trie->refcount;
11589 PerlMemShared_free(trie->charmap);
11590 PerlMemShared_free(trie->states);
11591 PerlMemShared_free(trie->trans);
11593 PerlMemShared_free(trie->bitmap);
11595 PerlMemShared_free(trie->jump);
11596 PerlMemShared_free(trie->wordinfo);
11597 /* do this last!!!! */
11598 PerlMemShared_free(ri->data->data[n]);
11603 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
11606 Safefree(ri->data->what);
11607 Safefree(ri->data);
11613 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11614 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11615 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11618 re_dup - duplicate a regexp.
11620 This routine is expected to clone a given regexp structure. It is only
11621 compiled under USE_ITHREADS.
11623 After all of the core data stored in struct regexp is duplicated
11624 the regexp_engine.dupe method is used to copy any private data
11625 stored in the *pprivate pointer. This allows extensions to handle
11626 any duplication it needs to do.
11628 See pregfree() and regfree_internal() if you change anything here.
11630 #if defined(USE_ITHREADS)
11631 #ifndef PERL_IN_XSUB_RE
11633 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
11637 const struct regexp *r = (const struct regexp *)SvANY(sstr);
11638 struct regexp *ret = (struct regexp *)SvANY(dstr);
11640 PERL_ARGS_ASSERT_RE_DUP_GUTS;
11642 npar = r->nparens+1;
11643 Newx(ret->offs, npar, regexp_paren_pair);
11644 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11646 /* no need to copy these */
11647 Newx(ret->swap, npar, regexp_paren_pair);
11650 if (ret->substrs) {
11651 /* Do it this way to avoid reading from *r after the StructCopy().
11652 That way, if any of the sv_dup_inc()s dislodge *r from the L1
11653 cache, it doesn't matter. */
11654 const bool anchored = r->check_substr
11655 ? r->check_substr == r->anchored_substr
11656 : r->check_utf8 == r->anchored_utf8;
11657 Newx(ret->substrs, 1, struct reg_substr_data);
11658 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11660 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
11661 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
11662 ret->float_substr = sv_dup_inc(ret->float_substr, param);
11663 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
11665 /* check_substr and check_utf8, if non-NULL, point to either their
11666 anchored or float namesakes, and don't hold a second reference. */
11668 if (ret->check_substr) {
11670 assert(r->check_utf8 == r->anchored_utf8);
11671 ret->check_substr = ret->anchored_substr;
11672 ret->check_utf8 = ret->anchored_utf8;
11674 assert(r->check_substr == r->float_substr);
11675 assert(r->check_utf8 == r->float_utf8);
11676 ret->check_substr = ret->float_substr;
11677 ret->check_utf8 = ret->float_utf8;
11679 } else if (ret->check_utf8) {
11681 ret->check_utf8 = ret->anchored_utf8;
11683 ret->check_utf8 = ret->float_utf8;
11688 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
11691 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
11693 if (RX_MATCH_COPIED(dstr))
11694 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
11696 ret->subbeg = NULL;
11697 #ifdef PERL_OLD_COPY_ON_WRITE
11698 ret->saved_copy = NULL;
11701 if (ret->mother_re) {
11702 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
11703 /* Our storage points directly to our mother regexp, but that's
11704 1: a buffer in a different thread
11705 2: something we no longer hold a reference on
11706 so we need to copy it locally. */
11707 /* Note we need to sue SvCUR() on our mother_re, because it, in
11708 turn, may well be pointing to its own mother_re. */
11709 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
11710 SvCUR(ret->mother_re)+1));
11711 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
11713 ret->mother_re = NULL;
11717 #endif /* PERL_IN_XSUB_RE */
11722 This is the internal complement to regdupe() which is used to copy
11723 the structure pointed to by the *pprivate pointer in the regexp.
11724 This is the core version of the extension overridable cloning hook.
11725 The regexp structure being duplicated will be copied by perl prior
11726 to this and will be provided as the regexp *r argument, however
11727 with the /old/ structures pprivate pointer value. Thus this routine
11728 may override any copying normally done by perl.
11730 It returns a pointer to the new regexp_internal structure.
11734 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
11737 struct regexp *const r = (struct regexp *)SvANY(rx);
11738 regexp_internal *reti;
11740 RXi_GET_DECL(r,ri);
11742 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
11744 npar = r->nparens+1;
11747 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
11748 Copy(ri->program, reti->program, len+1, regnode);
11751 reti->regstclass = NULL;
11754 struct reg_data *d;
11755 const int count = ri->data->count;
11758 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
11759 char, struct reg_data);
11760 Newx(d->what, count, U8);
11763 for (i = 0; i < count; i++) {
11764 d->what[i] = ri->data->what[i];
11765 switch (d->what[i]) {
11766 /* legal options are one of: sSfpontTua
11767 see also regcomp.h and pregfree() */
11768 case 'a': /* actually an AV, but the dup function is identical. */
11771 case 'p': /* actually an AV, but the dup function is identical. */
11772 case 'u': /* actually an HV, but the dup function is identical. */
11773 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
11776 /* This is cheating. */
11777 Newx(d->data[i], 1, struct regnode_charclass_class);
11778 StructCopy(ri->data->data[i], d->data[i],
11779 struct regnode_charclass_class);
11780 reti->regstclass = (regnode*)d->data[i];
11783 /* Compiled op trees are readonly and in shared memory,
11784 and can thus be shared without duplication. */
11786 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
11790 /* Trie stclasses are readonly and can thus be shared
11791 * without duplication. We free the stclass in pregfree
11792 * when the corresponding reg_ac_data struct is freed.
11794 reti->regstclass= ri->regstclass;
11798 ((reg_trie_data*)ri->data->data[i])->refcount++;
11802 d->data[i] = ri->data->data[i];
11805 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
11814 reti->name_list_idx = ri->name_list_idx;
11816 #ifdef RE_TRACK_PATTERN_OFFSETS
11817 if (ri->u.offsets) {
11818 Newx(reti->u.offsets, 2*len+1, U32);
11819 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
11822 SetProgLen(reti,len);
11825 return (void*)reti;
11828 #endif /* USE_ITHREADS */
11830 #ifndef PERL_IN_XSUB_RE
11833 - regnext - dig the "next" pointer out of a node
11836 Perl_regnext(pTHX_ register regnode *p)
11839 register I32 offset;
11844 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
11845 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
11848 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
11857 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
11860 STRLEN l1 = strlen(pat1);
11861 STRLEN l2 = strlen(pat2);
11864 const char *message;
11866 PERL_ARGS_ASSERT_RE_CROAK2;
11872 Copy(pat1, buf, l1 , char);
11873 Copy(pat2, buf + l1, l2 , char);
11874 buf[l1 + l2] = '\n';
11875 buf[l1 + l2 + 1] = '\0';
11877 /* ANSI variant takes additional second argument */
11878 va_start(args, pat2);
11882 msv = vmess(buf, &args);
11884 message = SvPV_const(msv,l1);
11887 Copy(message, buf, l1 , char);
11888 buf[l1-1] = '\0'; /* Overwrite \n */
11889 Perl_croak(aTHX_ "%s", buf);
11892 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
11894 #ifndef PERL_IN_XSUB_RE
11896 Perl_save_re_context(pTHX)
11900 struct re_save_state *state;
11902 SAVEVPTR(PL_curcop);
11903 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
11905 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
11906 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
11907 SSPUSHUV(SAVEt_RE_STATE);
11909 Copy(&PL_reg_state, state, 1, struct re_save_state);
11911 PL_reg_start_tmp = 0;
11912 PL_reg_start_tmpl = 0;
11913 PL_reg_oldsaved = NULL;
11914 PL_reg_oldsavedlen = 0;
11915 PL_reg_maxiter = 0;
11916 PL_reg_leftiter = 0;
11917 PL_reg_poscache = NULL;
11918 PL_reg_poscache_size = 0;
11919 #ifdef PERL_OLD_COPY_ON_WRITE
11923 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
11925 const REGEXP * const rx = PM_GETRE(PL_curpm);
11928 for (i = 1; i <= RX_NPARENS(rx); i++) {
11929 char digits[TYPE_CHARS(long)];
11930 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
11931 GV *const *const gvp
11932 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
11935 GV * const gv = *gvp;
11936 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
11946 clear_re(pTHX_ void *r)
11949 ReREFCNT_dec((REGEXP *)r);
11955 S_put_byte(pTHX_ SV *sv, int c)
11957 PERL_ARGS_ASSERT_PUT_BYTE;
11959 /* Our definition of isPRINT() ignores locales, so only bytes that are
11960 not part of UTF-8 are considered printable. I assume that the same
11961 holds for UTF-EBCDIC.
11962 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
11963 which Wikipedia says:
11965 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
11966 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
11967 identical, to the ASCII delete (DEL) or rubout control character.
11968 ) So the old condition can be simplified to !isPRINT(c) */
11971 Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c);
11974 Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c);
11978 const char string = c;
11979 if (c == '-' || c == ']' || c == '\\' || c == '^')
11980 sv_catpvs(sv, "\\");
11981 sv_catpvn(sv, &string, 1);
11986 #define CLEAR_OPTSTART \
11987 if (optstart) STMT_START { \
11988 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
11992 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
11994 STATIC const regnode *
11995 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
11996 const regnode *last, const regnode *plast,
11997 SV* sv, I32 indent, U32 depth)
12000 register U8 op = PSEUDO; /* Arbitrary non-END op. */
12001 register const regnode *next;
12002 const regnode *optstart= NULL;
12004 RXi_GET_DECL(r,ri);
12005 GET_RE_DEBUG_FLAGS_DECL;
12007 PERL_ARGS_ASSERT_DUMPUNTIL;
12009 #ifdef DEBUG_DUMPUNTIL
12010 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
12011 last ? last-start : 0,plast ? plast-start : 0);
12014 if (plast && plast < last)
12017 while (PL_regkind[op] != END && (!last || node < last)) {
12018 /* While that wasn't END last time... */
12021 if (op == CLOSE || op == WHILEM)
12023 next = regnext((regnode *)node);
12026 if (OP(node) == OPTIMIZED) {
12027 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
12034 regprop(r, sv, node);
12035 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
12036 (int)(2*indent + 1), "", SvPVX_const(sv));
12038 if (OP(node) != OPTIMIZED) {
12039 if (next == NULL) /* Next ptr. */
12040 PerlIO_printf(Perl_debug_log, " (0)");
12041 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
12042 PerlIO_printf(Perl_debug_log, " (FAIL)");
12044 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
12045 (void)PerlIO_putc(Perl_debug_log, '\n');
12049 if (PL_regkind[(U8)op] == BRANCHJ) {
12052 register const regnode *nnode = (OP(next) == LONGJMP
12053 ? regnext((regnode *)next)
12055 if (last && nnode > last)
12057 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
12060 else if (PL_regkind[(U8)op] == BRANCH) {
12062 DUMPUNTIL(NEXTOPER(node), next);
12064 else if ( PL_regkind[(U8)op] == TRIE ) {
12065 const regnode *this_trie = node;
12066 const char op = OP(node);
12067 const U32 n = ARG(node);
12068 const reg_ac_data * const ac = op>=AHOCORASICK ?
12069 (reg_ac_data *)ri->data->data[n] :
12071 const reg_trie_data * const trie =
12072 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
12074 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
12076 const regnode *nextbranch= NULL;
12079 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
12080 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
12082 PerlIO_printf(Perl_debug_log, "%*s%s ",
12083 (int)(2*(indent+3)), "",
12084 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
12085 PL_colors[0], PL_colors[1],
12086 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
12087 PERL_PV_PRETTY_ELLIPSES |
12088 PERL_PV_PRETTY_LTGT
12093 U16 dist= trie->jump[word_idx+1];
12094 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
12095 (UV)((dist ? this_trie + dist : next) - start));
12098 nextbranch= this_trie + trie->jump[0];
12099 DUMPUNTIL(this_trie + dist, nextbranch);
12101 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
12102 nextbranch= regnext((regnode *)nextbranch);
12104 PerlIO_printf(Perl_debug_log, "\n");
12107 if (last && next > last)
12112 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
12113 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
12114 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
12116 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
12118 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
12120 else if ( op == PLUS || op == STAR) {
12121 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
12123 else if (PL_regkind[(U8)op] == ANYOF) {
12124 /* arglen 1 + class block */
12125 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS)
12126 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
12127 node = NEXTOPER(node);
12129 else if (PL_regkind[(U8)op] == EXACT) {
12130 /* Literal string, where present. */
12131 node += NODE_SZ_STR(node) - 1;
12132 node = NEXTOPER(node);
12135 node = NEXTOPER(node);
12136 node += regarglen[(U8)op];
12138 if (op == CURLYX || op == OPEN)
12142 #ifdef DEBUG_DUMPUNTIL
12143 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
12148 #endif /* DEBUGGING */
12152 * c-indentation-style: bsd
12153 * c-basic-offset: 4
12154 * indent-tabs-mode: t
12157 * ex: set ts=8 sts=4 sw=4 noet: