5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
10 /* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
12 * a regular expression.
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
20 /* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
24 /* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
29 /* The names of the functions have been changed from regcomp and
30 * regexec to pregcomp and pregexec in order to avoid conflicts
31 * with the POSIX routines of the same names.
34 #ifdef PERL_EXT_RE_BUILD
39 * pregcomp and pregexec -- regsub and regerror are not used in perl
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
59 **** Alterations to Henry's code are...
61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
74 #define PERL_IN_REGCOMP_C
77 #ifndef PERL_IN_XSUB_RE
82 #ifdef PERL_IN_XSUB_RE
88 #include "dquote_static.c"
95 # if defined(BUGGY_MSC6)
96 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
97 # pragma optimize("a",off)
98 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
99 # pragma optimize("w",on )
100 # endif /* BUGGY_MSC6 */
104 #define STATIC static
107 typedef struct RExC_state_t {
108 U32 flags; /* are we folding, multilining? */
109 char *precomp; /* uncompiled string. */
110 REGEXP *rx_sv; /* The SV that is the regexp. */
111 regexp *rx; /* perl core regexp structure */
112 regexp_internal *rxi; /* internal data for regexp object pprivate field */
113 char *start; /* Start of input for compile */
114 char *end; /* End of input for compile */
115 char *parse; /* Input-scan pointer. */
116 I32 whilem_seen; /* number of WHILEM in this expr */
117 regnode *emit_start; /* Start of emitted-code area */
118 regnode *emit_bound; /* First regnode outside of the allocated space */
119 regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */
120 I32 naughty; /* How bad is this pattern? */
121 I32 sawback; /* Did we see \1, ...? */
123 I32 size; /* Code size. */
124 I32 npar; /* Capture buffer count, (OPEN). */
125 I32 cpar; /* Capture buffer count, (CLOSE). */
126 I32 nestroot; /* root parens we are in - used by accept */
130 regnode **open_parens; /* pointers to open parens */
131 regnode **close_parens; /* pointers to close parens */
132 regnode *opend; /* END node in program */
133 I32 utf8; /* whether the pattern is utf8 or not */
134 I32 orig_utf8; /* whether the pattern was originally in utf8 */
135 /* XXX use this for future optimisation of case
136 * where pattern must be upgraded to utf8. */
137 I32 uni_semantics; /* If a d charset modifier should use unicode
138 rules, even if the pattern is not in
140 HV *paren_names; /* Paren names */
142 regnode **recurse; /* Recurse regops */
143 I32 recurse_count; /* Number of recurse regops */
146 I32 override_recoding;
148 char *starttry; /* -Dr: where regtry was called. */
149 #define RExC_starttry (pRExC_state->starttry)
152 const char *lastparse;
154 AV *paren_name_list; /* idx -> name */
155 #define RExC_lastparse (pRExC_state->lastparse)
156 #define RExC_lastnum (pRExC_state->lastnum)
157 #define RExC_paren_name_list (pRExC_state->paren_name_list)
161 #define RExC_flags (pRExC_state->flags)
162 #define RExC_precomp (pRExC_state->precomp)
163 #define RExC_rx_sv (pRExC_state->rx_sv)
164 #define RExC_rx (pRExC_state->rx)
165 #define RExC_rxi (pRExC_state->rxi)
166 #define RExC_start (pRExC_state->start)
167 #define RExC_end (pRExC_state->end)
168 #define RExC_parse (pRExC_state->parse)
169 #define RExC_whilem_seen (pRExC_state->whilem_seen)
170 #ifdef RE_TRACK_PATTERN_OFFSETS
171 #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
173 #define RExC_emit (pRExC_state->emit)
174 #define RExC_emit_start (pRExC_state->emit_start)
175 #define RExC_emit_bound (pRExC_state->emit_bound)
176 #define RExC_naughty (pRExC_state->naughty)
177 #define RExC_sawback (pRExC_state->sawback)
178 #define RExC_seen (pRExC_state->seen)
179 #define RExC_size (pRExC_state->size)
180 #define RExC_npar (pRExC_state->npar)
181 #define RExC_nestroot (pRExC_state->nestroot)
182 #define RExC_extralen (pRExC_state->extralen)
183 #define RExC_seen_zerolen (pRExC_state->seen_zerolen)
184 #define RExC_seen_evals (pRExC_state->seen_evals)
185 #define RExC_utf8 (pRExC_state->utf8)
186 #define RExC_uni_semantics (pRExC_state->uni_semantics)
187 #define RExC_orig_utf8 (pRExC_state->orig_utf8)
188 #define RExC_open_parens (pRExC_state->open_parens)
189 #define RExC_close_parens (pRExC_state->close_parens)
190 #define RExC_opend (pRExC_state->opend)
191 #define RExC_paren_names (pRExC_state->paren_names)
192 #define RExC_recurse (pRExC_state->recurse)
193 #define RExC_recurse_count (pRExC_state->recurse_count)
194 #define RExC_in_lookbehind (pRExC_state->in_lookbehind)
195 #define RExC_contains_locale (pRExC_state->contains_locale)
196 #define RExC_override_recoding (pRExC_state->override_recoding)
199 #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
200 #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
201 ((*s) == '{' && regcurly(s)))
204 #undef SPSTART /* dratted cpp namespace... */
207 * Flags to be passed up and down.
209 #define WORST 0 /* Worst case. */
210 #define HASWIDTH 0x01 /* Known to match non-null strings. */
212 /* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single
213 * character, and if utf8, must be invariant. Note that this is not the same thing as REGNODE_SIMPLE */
215 #define SPSTART 0x04 /* Starts with * or +. */
216 #define TRYAGAIN 0x08 /* Weeded out a declaration. */
217 #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
219 #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
221 /* whether trie related optimizations are enabled */
222 #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
223 #define TRIE_STUDY_OPT
224 #define FULL_TRIE_STUDY
230 #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
231 #define PBITVAL(paren) (1 << ((paren) & 7))
232 #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
233 #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
234 #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
236 /* If not already in utf8, do a longjmp back to the beginning */
237 #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */
238 #define REQUIRE_UTF8 STMT_START { \
239 if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \
242 /* About scan_data_t.
244 During optimisation we recurse through the regexp program performing
245 various inplace (keyhole style) optimisations. In addition study_chunk
246 and scan_commit populate this data structure with information about
247 what strings MUST appear in the pattern. We look for the longest
248 string that must appear at a fixed location, and we look for the
249 longest string that may appear at a floating location. So for instance
254 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
255 strings (because they follow a .* construct). study_chunk will identify
256 both FOO and BAR as being the longest fixed and floating strings respectively.
258 The strings can be composites, for instance
262 will result in a composite fixed substring 'foo'.
264 For each string some basic information is maintained:
266 - offset or min_offset
267 This is the position the string must appear at, or not before.
268 It also implicitly (when combined with minlenp) tells us how many
269 characters must match before the string we are searching for.
270 Likewise when combined with minlenp and the length of the string it
271 tells us how many characters must appear after the string we have
275 Only used for floating strings. This is the rightmost point that
276 the string can appear at. If set to I32 max it indicates that the
277 string can occur infinitely far to the right.
280 A pointer to the minimum length of the pattern that the string
281 was found inside. This is important as in the case of positive
282 lookahead or positive lookbehind we can have multiple patterns
287 The minimum length of the pattern overall is 3, the minimum length
288 of the lookahead part is 3, but the minimum length of the part that
289 will actually match is 1. So 'FOO's minimum length is 3, but the
290 minimum length for the F is 1. This is important as the minimum length
291 is used to determine offsets in front of and behind the string being
292 looked for. Since strings can be composites this is the length of the
293 pattern at the time it was committed with a scan_commit. Note that
294 the length is calculated by study_chunk, so that the minimum lengths
295 are not known until the full pattern has been compiled, thus the
296 pointer to the value.
300 In the case of lookbehind the string being searched for can be
301 offset past the start point of the final matching string.
302 If this value was just blithely removed from the min_offset it would
303 invalidate some of the calculations for how many chars must match
304 before or after (as they are derived from min_offset and minlen and
305 the length of the string being searched for).
306 When the final pattern is compiled and the data is moved from the
307 scan_data_t structure into the regexp structure the information
308 about lookbehind is factored in, with the information that would
309 have been lost precalculated in the end_shift field for the
312 The fields pos_min and pos_delta are used to store the minimum offset
313 and the delta to the maximum offset at the current point in the pattern.
317 typedef struct scan_data_t {
318 /*I32 len_min; unused */
319 /*I32 len_delta; unused */
323 I32 last_end; /* min value, <0 unless valid. */
326 SV **longest; /* Either &l_fixed, or &l_float. */
327 SV *longest_fixed; /* longest fixed string found in pattern */
328 I32 offset_fixed; /* offset where it starts */
329 I32 *minlen_fixed; /* pointer to the minlen relevant to the string */
330 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
331 SV *longest_float; /* longest floating string found in pattern */
332 I32 offset_float_min; /* earliest point in string it can appear */
333 I32 offset_float_max; /* latest point in string it can appear */
334 I32 *minlen_float; /* pointer to the minlen relevant to the string */
335 I32 lookbehind_float; /* is the position of the string modified by LB */
339 struct regnode_charclass_class *start_class;
343 * Forward declarations for pregcomp()'s friends.
346 static const scan_data_t zero_scan_data =
347 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
349 #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
350 #define SF_BEFORE_SEOL 0x0001
351 #define SF_BEFORE_MEOL 0x0002
352 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
353 #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
356 # define SF_FIX_SHIFT_EOL (0+2)
357 # define SF_FL_SHIFT_EOL (0+4)
359 # define SF_FIX_SHIFT_EOL (+2)
360 # define SF_FL_SHIFT_EOL (+4)
363 #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
364 #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
366 #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
367 #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
368 #define SF_IS_INF 0x0040
369 #define SF_HAS_PAR 0x0080
370 #define SF_IN_PAR 0x0100
371 #define SF_HAS_EVAL 0x0200
372 #define SCF_DO_SUBSTR 0x0400
373 #define SCF_DO_STCLASS_AND 0x0800
374 #define SCF_DO_STCLASS_OR 0x1000
375 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
376 #define SCF_WHILEM_VISITED_POS 0x2000
378 #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
379 #define SCF_SEEN_ACCEPT 0x8000
381 #define UTF cBOOL(RExC_utf8)
382 #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
383 #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
384 #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_DEPENDS_CHARSET)
385 #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) >= REGEX_UNICODE_CHARSET)
386 #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_RESTRICTED_CHARSET)
387 #define MORE_ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
388 #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) >= REGEX_ASCII_RESTRICTED_CHARSET)
390 #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
392 #define OOB_UNICODE 12345678
393 #define OOB_NAMEDCLASS -1
395 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
396 #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
399 /* length of regex to show in messages that don't mark a position within */
400 #define RegexLengthToShowInErrorMessages 127
403 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
404 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
405 * op/pragma/warn/regcomp.
407 #define MARKER1 "<-- HERE" /* marker as it appears in the description */
408 #define MARKER2 " <-- HERE " /* marker as it appears within the regex */
410 #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
413 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
414 * arg. Show regex, up to a maximum length. If it's too long, chop and add
417 #define _FAIL(code) STMT_START { \
418 const char *ellipses = ""; \
419 IV len = RExC_end - RExC_precomp; \
422 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
423 if (len > RegexLengthToShowInErrorMessages) { \
424 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
425 len = RegexLengthToShowInErrorMessages - 10; \
431 #define FAIL(msg) _FAIL( \
432 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
433 msg, (int)len, RExC_precomp, ellipses))
435 #define FAIL2(msg,arg) _FAIL( \
436 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
437 arg, (int)len, RExC_precomp, ellipses))
440 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
442 #define Simple_vFAIL(m) STMT_START { \
443 const IV offset = RExC_parse - RExC_precomp; \
444 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
445 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
449 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
451 #define vFAIL(m) STMT_START { \
453 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
458 * Like Simple_vFAIL(), but accepts two arguments.
460 #define Simple_vFAIL2(m,a1) STMT_START { \
461 const IV offset = RExC_parse - RExC_precomp; \
462 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
463 (int)offset, RExC_precomp, RExC_precomp + offset); \
467 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
469 #define vFAIL2(m,a1) STMT_START { \
471 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
472 Simple_vFAIL2(m, a1); \
477 * Like Simple_vFAIL(), but accepts three arguments.
479 #define Simple_vFAIL3(m, a1, a2) STMT_START { \
480 const IV offset = RExC_parse - RExC_precomp; \
481 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
482 (int)offset, RExC_precomp, RExC_precomp + offset); \
486 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
488 #define vFAIL3(m,a1,a2) STMT_START { \
490 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
491 Simple_vFAIL3(m, a1, a2); \
495 * Like Simple_vFAIL(), but accepts four arguments.
497 #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
498 const IV offset = RExC_parse - RExC_precomp; \
499 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
500 (int)offset, RExC_precomp, RExC_precomp + offset); \
503 #define ckWARNreg(loc,m) STMT_START { \
504 const IV offset = loc - RExC_precomp; \
505 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
506 (int)offset, RExC_precomp, RExC_precomp + offset); \
509 #define ckWARNregdep(loc,m) STMT_START { \
510 const IV offset = loc - RExC_precomp; \
511 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
513 (int)offset, RExC_precomp, RExC_precomp + offset); \
516 #define ckWARN2regdep(loc,m, a1) STMT_START { \
517 const IV offset = loc - RExC_precomp; \
518 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
520 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
523 #define ckWARN2reg(loc, m, a1) STMT_START { \
524 const IV offset = loc - RExC_precomp; \
525 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
526 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
529 #define vWARN3(loc, m, a1, a2) STMT_START { \
530 const IV offset = loc - RExC_precomp; \
531 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
532 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
535 #define ckWARN3reg(loc, m, a1, a2) STMT_START { \
536 const IV offset = loc - RExC_precomp; \
537 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
538 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
541 #define vWARN4(loc, m, a1, a2, a3) STMT_START { \
542 const IV offset = loc - RExC_precomp; \
543 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
544 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
547 #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
548 const IV offset = loc - RExC_precomp; \
549 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
550 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
553 #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
554 const IV offset = loc - RExC_precomp; \
555 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
556 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
560 /* Allow for side effects in s */
561 #define REGC(c,s) STMT_START { \
562 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
565 /* Macros for recording node offsets. 20001227 mjd@plover.com
566 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
567 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
568 * Element 0 holds the number n.
569 * Position is 1 indexed.
571 #ifndef RE_TRACK_PATTERN_OFFSETS
572 #define Set_Node_Offset_To_R(node,byte)
573 #define Set_Node_Offset(node,byte)
574 #define Set_Cur_Node_Offset
575 #define Set_Node_Length_To_R(node,len)
576 #define Set_Node_Length(node,len)
577 #define Set_Node_Cur_Length(node)
578 #define Node_Offset(n)
579 #define Node_Length(n)
580 #define Set_Node_Offset_Length(node,offset,len)
581 #define ProgLen(ri) ri->u.proglen
582 #define SetProgLen(ri,x) ri->u.proglen = x
584 #define ProgLen(ri) ri->u.offsets[0]
585 #define SetProgLen(ri,x) ri->u.offsets[0] = x
586 #define Set_Node_Offset_To_R(node,byte) STMT_START { \
588 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
589 __LINE__, (int)(node), (int)(byte))); \
591 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
593 RExC_offsets[2*(node)-1] = (byte); \
598 #define Set_Node_Offset(node,byte) \
599 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
600 #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
602 #define Set_Node_Length_To_R(node,len) STMT_START { \
604 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
605 __LINE__, (int)(node), (int)(len))); \
607 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
609 RExC_offsets[2*(node)] = (len); \
614 #define Set_Node_Length(node,len) \
615 Set_Node_Length_To_R((node)-RExC_emit_start, len)
616 #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
617 #define Set_Node_Cur_Length(node) \
618 Set_Node_Length(node, RExC_parse - parse_start)
620 /* Get offsets and lengths */
621 #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
622 #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
624 #define Set_Node_Offset_Length(node,offset,len) STMT_START { \
625 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
626 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
630 #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
631 #define EXPERIMENTAL_INPLACESCAN
632 #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
634 #define DEBUG_STUDYDATA(str,data,depth) \
635 DEBUG_OPTIMISE_MORE_r(if(data){ \
636 PerlIO_printf(Perl_debug_log, \
637 "%*s" str "Pos:%"IVdf"/%"IVdf \
638 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
639 (int)(depth)*2, "", \
640 (IV)((data)->pos_min), \
641 (IV)((data)->pos_delta), \
642 (UV)((data)->flags), \
643 (IV)((data)->whilem_c), \
644 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
645 is_inf ? "INF " : "" \
647 if ((data)->last_found) \
648 PerlIO_printf(Perl_debug_log, \
649 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
650 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
651 SvPVX_const((data)->last_found), \
652 (IV)((data)->last_end), \
653 (IV)((data)->last_start_min), \
654 (IV)((data)->last_start_max), \
655 ((data)->longest && \
656 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
657 SvPVX_const((data)->longest_fixed), \
658 (IV)((data)->offset_fixed), \
659 ((data)->longest && \
660 (data)->longest==&((data)->longest_float)) ? "*" : "", \
661 SvPVX_const((data)->longest_float), \
662 (IV)((data)->offset_float_min), \
663 (IV)((data)->offset_float_max) \
665 PerlIO_printf(Perl_debug_log,"\n"); \
668 static void clear_re(pTHX_ void *r);
670 /* Mark that we cannot extend a found fixed substring at this point.
671 Update the longest found anchored substring and the longest found
672 floating substrings if needed. */
675 S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
677 const STRLEN l = CHR_SVLEN(data->last_found);
678 const STRLEN old_l = CHR_SVLEN(*data->longest);
679 GET_RE_DEBUG_FLAGS_DECL;
681 PERL_ARGS_ASSERT_SCAN_COMMIT;
683 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
684 SvSetMagicSV(*data->longest, data->last_found);
685 if (*data->longest == data->longest_fixed) {
686 data->offset_fixed = l ? data->last_start_min : data->pos_min;
687 if (data->flags & SF_BEFORE_EOL)
689 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
691 data->flags &= ~SF_FIX_BEFORE_EOL;
692 data->minlen_fixed=minlenp;
693 data->lookbehind_fixed=0;
695 else { /* *data->longest == data->longest_float */
696 data->offset_float_min = l ? data->last_start_min : data->pos_min;
697 data->offset_float_max = (l
698 ? data->last_start_max
699 : data->pos_min + data->pos_delta);
700 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
701 data->offset_float_max = I32_MAX;
702 if (data->flags & SF_BEFORE_EOL)
704 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
706 data->flags &= ~SF_FL_BEFORE_EOL;
707 data->minlen_float=minlenp;
708 data->lookbehind_float=0;
711 SvCUR_set(data->last_found, 0);
713 SV * const sv = data->last_found;
714 if (SvUTF8(sv) && SvMAGICAL(sv)) {
715 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
721 data->flags &= ~SF_BEFORE_EOL;
722 DEBUG_STUDYDATA("commit: ",data,0);
725 /* Can match anything (initialization) */
727 S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
729 PERL_ARGS_ASSERT_CL_ANYTHING;
731 ANYOF_BITMAP_SETALL(cl);
732 cl->flags = ANYOF_CLASS|ANYOF_EOS|ANYOF_UNICODE_ALL
733 |ANYOF_LOC_NONBITMAP_FOLD|ANYOF_NON_UTF8_LATIN1_ALL;
735 /* If any portion of the regex is to operate under locale rules,
736 * initialization includes it. The reason this isn't done for all regexes
737 * is that the optimizer was written under the assumption that locale was
738 * all-or-nothing. Given the complexity and lack of documentation in the
739 * optimizer, and that there are inadequate test cases for locale, so many
740 * parts of it may not work properly, it is safest to avoid locale unless
742 if (RExC_contains_locale) {
743 ANYOF_CLASS_SETALL(cl); /* /l uses class */
744 cl->flags |= ANYOF_LOCALE;
747 ANYOF_CLASS_ZERO(cl); /* Only /l uses class now */
751 /* Can match anything (initialization) */
753 S_cl_is_anything(const struct regnode_charclass_class *cl)
757 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
759 for (value = 0; value <= ANYOF_MAX; value += 2)
760 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
762 if (!(cl->flags & ANYOF_UNICODE_ALL))
764 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
769 /* Can match anything (initialization) */
771 S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
773 PERL_ARGS_ASSERT_CL_INIT;
775 Zero(cl, 1, struct regnode_charclass_class);
777 cl_anything(pRExC_state, cl);
778 ARG_SET(cl, ANYOF_NONBITMAP_EMPTY);
781 /* These two functions currently do the exact same thing */
782 #define cl_init_zero S_cl_init
784 /* 'AND' a given class with another one. Can create false positives. 'cl'
785 * should not be inverted. 'and_with->flags & ANYOF_CLASS' should be 0 if
786 * 'and_with' is a regnode_charclass instead of a regnode_charclass_class. */
788 S_cl_and(struct regnode_charclass_class *cl,
789 const struct regnode_charclass_class *and_with)
791 PERL_ARGS_ASSERT_CL_AND;
793 assert(and_with->type == ANYOF);
795 /* I (khw) am not sure all these restrictions are necessary XXX */
796 if (!(ANYOF_CLASS_TEST_ANY_SET(and_with))
797 && !(ANYOF_CLASS_TEST_ANY_SET(cl))
798 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
799 && !(and_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
800 && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) {
803 if (and_with->flags & ANYOF_INVERT)
804 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
805 cl->bitmap[i] &= ~and_with->bitmap[i];
807 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
808 cl->bitmap[i] &= and_with->bitmap[i];
809 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
811 if (and_with->flags & ANYOF_INVERT) {
813 /* Here, the and'ed node is inverted. Get the AND of the flags that
814 * aren't affected by the inversion. Those that are affected are
815 * handled individually below */
816 U8 affected_flags = cl->flags & ~INVERSION_UNAFFECTED_FLAGS;
817 cl->flags &= (and_with->flags & INVERSION_UNAFFECTED_FLAGS);
818 cl->flags |= affected_flags;
820 /* We currently don't know how to deal with things that aren't in the
821 * bitmap, but we know that the intersection is no greater than what
822 * is already in cl, so let there be false positives that get sorted
823 * out after the synthetic start class succeeds, and the node is
824 * matched for real. */
826 /* The inversion of these two flags indicate that the resulting
827 * intersection doesn't have them */
828 if (and_with->flags & ANYOF_UNICODE_ALL) {
829 cl->flags &= ~ANYOF_UNICODE_ALL;
831 if (and_with->flags & ANYOF_NON_UTF8_LATIN1_ALL) {
832 cl->flags &= ~ANYOF_NON_UTF8_LATIN1_ALL;
835 else { /* and'd node is not inverted */
836 U8 outside_bitmap_but_not_utf8; /* Temp variable */
838 if (! ANYOF_NONBITMAP(and_with)) {
840 /* Here 'and_with' doesn't match anything outside the bitmap
841 * (except possibly ANYOF_UNICODE_ALL), which means the
842 * intersection can't either, except for ANYOF_UNICODE_ALL, in
843 * which case we don't know what the intersection is, but it's no
844 * greater than what cl already has, so can just leave it alone,
845 * with possible false positives */
846 if (! (and_with->flags & ANYOF_UNICODE_ALL)) {
847 ARG_SET(cl, ANYOF_NONBITMAP_EMPTY);
848 cl->flags &= ~ANYOF_NONBITMAP_NON_UTF8;
851 else if (! ANYOF_NONBITMAP(cl)) {
853 /* Here, 'and_with' does match something outside the bitmap, and cl
854 * doesn't have a list of things to match outside the bitmap. If
855 * cl can match all code points above 255, the intersection will
856 * be those above-255 code points that 'and_with' matches. If cl
857 * can't match all Unicode code points, it means that it can't
858 * match anything outside the bitmap (since the 'if' that got us
859 * into this block tested for that), so we leave the bitmap empty.
861 if (cl->flags & ANYOF_UNICODE_ALL) {
862 ARG_SET(cl, ARG(and_with));
864 /* and_with's ARG may match things that don't require UTF8.
865 * And now cl's will too, in spite of this being an 'and'. See
866 * the comments below about the kludge */
867 cl->flags |= and_with->flags & ANYOF_NONBITMAP_NON_UTF8;
871 /* Here, both 'and_with' and cl match something outside the
872 * bitmap. Currently we do not do the intersection, so just match
873 * whatever cl had at the beginning. */
877 /* Take the intersection of the two sets of flags. However, the
878 * ANYOF_NONBITMAP_NON_UTF8 flag is treated as an 'or'. This is a
879 * kludge around the fact that this flag is not treated like the others
880 * which are initialized in cl_anything(). The way the optimizer works
881 * is that the synthetic start class (SSC) is initialized to match
882 * anything, and then the first time a real node is encountered, its
883 * values are AND'd with the SSC's with the result being the values of
884 * the real node. However, there are paths through the optimizer where
885 * the AND never gets called, so those initialized bits are set
886 * inappropriately, which is not usually a big deal, as they just cause
887 * false positives in the SSC, which will just mean a probably
888 * imperceptible slow down in execution. However this bit has a
889 * higher false positive consequence in that it can cause utf8.pm,
890 * utf8_heavy.pl ... to be loaded when not necessary, which is a much
891 * bigger slowdown and also causes significant extra memory to be used.
892 * In order to prevent this, the code now takes a different tack. The
893 * bit isn't set unless some part of the regular expression needs it,
894 * but once set it won't get cleared. This means that these extra
895 * modules won't get loaded unless there was some path through the
896 * pattern that would have required them anyway, and so any false
897 * positives that occur by not ANDing them out when they could be
898 * aren't as severe as they would be if we treated this bit like all
900 outside_bitmap_but_not_utf8 = (cl->flags | and_with->flags)
901 & ANYOF_NONBITMAP_NON_UTF8;
902 cl->flags &= and_with->flags;
903 cl->flags |= outside_bitmap_but_not_utf8;
907 /* 'OR' a given class with another one. Can create false positives. 'cl'
908 * should not be inverted. 'or_with->flags & ANYOF_CLASS' should be 0 if
909 * 'or_with' is a regnode_charclass instead of a regnode_charclass_class. */
911 S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
913 PERL_ARGS_ASSERT_CL_OR;
915 if (or_with->flags & ANYOF_INVERT) {
917 /* Here, the or'd node is to be inverted. This means we take the
918 * complement of everything not in the bitmap, but currently we don't
919 * know what that is, so give up and match anything */
920 if (ANYOF_NONBITMAP(or_with)) {
921 cl_anything(pRExC_state, cl);
924 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
925 * <= (B1 | !B2) | (CL1 | !CL2)
926 * which is wasteful if CL2 is small, but we ignore CL2:
927 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
928 * XXXX Can we handle case-fold? Unclear:
929 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
930 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
932 else if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
933 && !(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
934 && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD) ) {
937 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
938 cl->bitmap[i] |= ~or_with->bitmap[i];
939 } /* XXXX: logic is complicated otherwise */
941 cl_anything(pRExC_state, cl);
944 /* And, we can just take the union of the flags that aren't affected
945 * by the inversion */
946 cl->flags |= or_with->flags & INVERSION_UNAFFECTED_FLAGS;
948 /* For the remaining flags:
949 ANYOF_UNICODE_ALL and inverted means to not match anything above
950 255, which means that the union with cl should just be
951 what cl has in it, so can ignore this flag
952 ANYOF_NON_UTF8_LATIN1_ALL and inverted means if not utf8 and ord
953 is 127-255 to match them, but then invert that, so the
954 union with cl should just be what cl has in it, so can
957 } else { /* 'or_with' is not inverted */
958 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
959 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
960 && (!(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD)
961 || (cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) ) {
964 /* OR char bitmap and class bitmap separately */
965 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
966 cl->bitmap[i] |= or_with->bitmap[i];
967 if (ANYOF_CLASS_TEST_ANY_SET(or_with)) {
968 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
969 cl->classflags[i] |= or_with->classflags[i];
970 cl->flags |= ANYOF_CLASS;
973 else { /* XXXX: logic is complicated, leave it along for a moment. */
974 cl_anything(pRExC_state, cl);
977 if (ANYOF_NONBITMAP(or_with)) {
979 /* Use the added node's outside-the-bit-map match if there isn't a
980 * conflict. If there is a conflict (both nodes match something
981 * outside the bitmap, but what they match outside is not the same
982 * pointer, and hence not easily compared until XXX we extend
983 * inversion lists this far), give up and allow the start class to
984 * match everything outside the bitmap. If that stuff is all above
985 * 255, can just set UNICODE_ALL, otherwise caould be anything. */
986 if (! ANYOF_NONBITMAP(cl)) {
987 ARG_SET(cl, ARG(or_with));
989 else if (ARG(cl) != ARG(or_with)) {
991 if ((or_with->flags & ANYOF_NONBITMAP_NON_UTF8)) {
992 cl_anything(pRExC_state, cl);
995 cl->flags |= ANYOF_UNICODE_ALL;
1000 /* Take the union */
1001 cl->flags |= or_with->flags;
1005 #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
1006 #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
1007 #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
1008 #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
1013 dump_trie(trie,widecharmap,revcharmap)
1014 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
1015 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
1017 These routines dump out a trie in a somewhat readable format.
1018 The _interim_ variants are used for debugging the interim
1019 tables that are used to generate the final compressed
1020 representation which is what dump_trie expects.
1022 Part of the reason for their existence is to provide a form
1023 of documentation as to how the different representations function.
1028 Dumps the final compressed table form of the trie to Perl_debug_log.
1029 Used for debugging make_trie().
1033 S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
1034 AV *revcharmap, U32 depth)
1037 SV *sv=sv_newmortal();
1038 int colwidth= widecharmap ? 6 : 4;
1040 GET_RE_DEBUG_FLAGS_DECL;
1042 PERL_ARGS_ASSERT_DUMP_TRIE;
1044 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
1045 (int)depth * 2 + 2,"",
1046 "Match","Base","Ofs" );
1048 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
1049 SV ** const tmp = av_fetch( revcharmap, state, 0);
1051 PerlIO_printf( Perl_debug_log, "%*s",
1053 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1054 PL_colors[0], PL_colors[1],
1055 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1056 PERL_PV_ESCAPE_FIRSTCHAR
1061 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
1062 (int)depth * 2 + 2,"");
1064 for( state = 0 ; state < trie->uniquecharcount ; state++ )
1065 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
1066 PerlIO_printf( Perl_debug_log, "\n");
1068 for( state = 1 ; state < trie->statecount ; state++ ) {
1069 const U32 base = trie->states[ state ].trans.base;
1071 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
1073 if ( trie->states[ state ].wordnum ) {
1074 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
1076 PerlIO_printf( Perl_debug_log, "%6s", "" );
1079 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
1084 while( ( base + ofs < trie->uniquecharcount ) ||
1085 ( base + ofs - trie->uniquecharcount < trie->lasttrans
1086 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
1089 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
1091 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
1092 if ( ( base + ofs >= trie->uniquecharcount ) &&
1093 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
1094 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1096 PerlIO_printf( Perl_debug_log, "%*"UVXf,
1098 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
1100 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
1104 PerlIO_printf( Perl_debug_log, "]");
1107 PerlIO_printf( Perl_debug_log, "\n" );
1109 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
1110 for (word=1; word <= trie->wordcount; word++) {
1111 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
1112 (int)word, (int)(trie->wordinfo[word].prev),
1113 (int)(trie->wordinfo[word].len));
1115 PerlIO_printf(Perl_debug_log, "\n" );
1118 Dumps a fully constructed but uncompressed trie in list form.
1119 List tries normally only are used for construction when the number of
1120 possible chars (trie->uniquecharcount) is very high.
1121 Used for debugging make_trie().
1124 S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
1125 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1129 SV *sv=sv_newmortal();
1130 int colwidth= widecharmap ? 6 : 4;
1131 GET_RE_DEBUG_FLAGS_DECL;
1133 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
1135 /* print out the table precompression. */
1136 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1137 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1138 "------:-----+-----------------\n" );
1140 for( state=1 ; state < next_alloc ; state ++ ) {
1143 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
1144 (int)depth * 2 + 2,"", (UV)state );
1145 if ( ! trie->states[ state ].wordnum ) {
1146 PerlIO_printf( Perl_debug_log, "%5s| ","");
1148 PerlIO_printf( Perl_debug_log, "W%4x| ",
1149 trie->states[ state ].wordnum
1152 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
1153 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
1155 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1157 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1158 PL_colors[0], PL_colors[1],
1159 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1160 PERL_PV_ESCAPE_FIRSTCHAR
1162 TRIE_LIST_ITEM(state,charid).forid,
1163 (UV)TRIE_LIST_ITEM(state,charid).newstate
1166 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1167 (int)((depth * 2) + 14), "");
1170 PerlIO_printf( Perl_debug_log, "\n");
1175 Dumps a fully constructed but uncompressed trie in table form.
1176 This is the normal DFA style state transition table, with a few
1177 twists to facilitate compression later.
1178 Used for debugging make_trie().
1181 S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
1182 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1187 SV *sv=sv_newmortal();
1188 int colwidth= widecharmap ? 6 : 4;
1189 GET_RE_DEBUG_FLAGS_DECL;
1191 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
1194 print out the table precompression so that we can do a visual check
1195 that they are identical.
1198 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1200 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1201 SV ** const tmp = av_fetch( revcharmap, charid, 0);
1203 PerlIO_printf( Perl_debug_log, "%*s",
1205 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
1206 PL_colors[0], PL_colors[1],
1207 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1208 PERL_PV_ESCAPE_FIRSTCHAR
1214 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1216 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
1217 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
1220 PerlIO_printf( Perl_debug_log, "\n" );
1222 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1224 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1225 (int)depth * 2 + 2,"",
1226 (UV)TRIE_NODENUM( state ) );
1228 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
1229 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1231 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1233 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
1235 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1236 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1238 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1239 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1247 /* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1248 startbranch: the first branch in the whole branch sequence
1249 first : start branch of sequence of branch-exact nodes.
1250 May be the same as startbranch
1251 last : Thing following the last branch.
1252 May be the same as tail.
1253 tail : item following the branch sequence
1254 count : words in the sequence
1255 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1256 depth : indent depth
1258 Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
1260 A trie is an N'ary tree where the branches are determined by digital
1261 decomposition of the key. IE, at the root node you look up the 1st character and
1262 follow that branch repeat until you find the end of the branches. Nodes can be
1263 marked as "accepting" meaning they represent a complete word. Eg:
1267 would convert into the following structure. Numbers represent states, letters
1268 following numbers represent valid transitions on the letter from that state, if
1269 the number is in square brackets it represents an accepting state, otherwise it
1270 will be in parenthesis.
1272 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1276 (1) +-i->(6)-+-s->[7]
1278 +-s->(3)-+-h->(4)-+-e->[5]
1280 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1282 This shows that when matching against the string 'hers' we will begin at state 1
1283 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1284 then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1285 is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1286 single traverse. We store a mapping from accepting to state to which word was
1287 matched, and then when we have multiple possibilities we try to complete the
1288 rest of the regex in the order in which they occured in the alternation.
1290 The only prior NFA like behaviour that would be changed by the TRIE support is
1291 the silent ignoring of duplicate alternations which are of the form:
1293 / (DUPE|DUPE) X? (?{ ... }) Y /x
1295 Thus EVAL blocks following a trie may be called a different number of times with
1296 and without the optimisation. With the optimisations dupes will be silently
1297 ignored. This inconsistent behaviour of EVAL type nodes is well established as
1298 the following demonstrates:
1300 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1302 which prints out 'word' three times, but
1304 'words'=~/(word|word|word)(?{ print $1 })S/
1306 which doesnt print it out at all. This is due to other optimisations kicking in.
1308 Example of what happens on a structural level:
1310 The regexp /(ac|ad|ab)+/ will produce the following debug output:
1312 1: CURLYM[1] {1,32767}(18)
1323 This would be optimizable with startbranch=5, first=5, last=16, tail=16
1324 and should turn into:
1326 1: CURLYM[1] {1,32767}(18)
1328 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1336 Cases where tail != last would be like /(?foo|bar)baz/:
1346 which would be optimizable with startbranch=1, first=1, last=7, tail=8
1347 and would end up looking like:
1350 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1357 d = uvuni_to_utf8_flags(d, uv, 0);
1359 is the recommended Unicode-aware way of saying
1364 #define TRIE_STORE_REVCHAR \
1367 SV *zlopp = newSV(2); \
1368 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1369 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
1370 SvCUR_set(zlopp, kapow - flrbbbbb); \
1373 av_push(revcharmap, zlopp); \
1375 char ooooff = (char)uvc; \
1376 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1380 #define TRIE_READ_CHAR STMT_START { \
1384 if ( foldlen > 0 ) { \
1385 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1390 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1391 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1392 foldlen -= UNISKIP( uvc ); \
1393 scan = foldbuf + UNISKIP( uvc ); \
1396 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1406 #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1407 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
1408 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1409 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
1411 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1412 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1413 TRIE_LIST_CUR( state )++; \
1416 #define TRIE_LIST_NEW(state) STMT_START { \
1417 Newxz( trie->states[ state ].trans.list, \
1418 4, reg_trie_trans_le ); \
1419 TRIE_LIST_CUR( state ) = 1; \
1420 TRIE_LIST_LEN( state ) = 4; \
1423 #define TRIE_HANDLE_WORD(state) STMT_START { \
1424 U16 dupe= trie->states[ state ].wordnum; \
1425 regnode * const noper_next = regnext( noper ); \
1428 /* store the word for dumping */ \
1430 if (OP(noper) != NOTHING) \
1431 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
1433 tmp = newSVpvn_utf8( "", 0, UTF ); \
1434 av_push( trie_words, tmp ); \
1438 trie->wordinfo[curword].prev = 0; \
1439 trie->wordinfo[curword].len = wordlen; \
1440 trie->wordinfo[curword].accept = state; \
1442 if ( noper_next < tail ) { \
1444 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
1445 trie->jump[curword] = (U16)(noper_next - convert); \
1447 jumper = noper_next; \
1449 nextbranch= regnext(cur); \
1453 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1454 /* chain, so that when the bits of chain are later */\
1455 /* linked together, the dups appear in the chain */\
1456 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1457 trie->wordinfo[dupe].prev = curword; \
1459 /* we haven't inserted this word yet. */ \
1460 trie->states[ state ].wordnum = curword; \
1465 #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1466 ( ( base + charid >= ucharcount \
1467 && base + charid < ubound \
1468 && state == trie->trans[ base - ucharcount + charid ].check \
1469 && trie->trans[ base - ucharcount + charid ].next ) \
1470 ? trie->trans[ base - ucharcount + charid ].next \
1471 : ( state==1 ? special : 0 ) \
1475 #define MADE_JUMP_TRIE 2
1476 #define MADE_EXACT_TRIE 4
1479 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
1482 /* first pass, loop through and scan words */
1483 reg_trie_data *trie;
1484 HV *widecharmap = NULL;
1485 AV *revcharmap = newAV();
1487 const U32 uniflags = UTF8_ALLOW_DEFAULT;
1492 regnode *jumper = NULL;
1493 regnode *nextbranch = NULL;
1494 regnode *convert = NULL;
1495 U32 *prev_states; /* temp array mapping each state to previous one */
1496 /* we just use folder as a flag in utf8 */
1497 const U8 * folder = NULL;
1500 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1501 AV *trie_words = NULL;
1502 /* along with revcharmap, this only used during construction but both are
1503 * useful during debugging so we store them in the struct when debugging.
1506 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
1507 STRLEN trie_charcount=0;
1509 SV *re_trie_maxbuff;
1510 GET_RE_DEBUG_FLAGS_DECL;
1512 PERL_ARGS_ASSERT_MAKE_TRIE;
1514 PERL_UNUSED_ARG(depth);
1519 case EXACTFU: folder = PL_fold_latin1; break;
1520 case EXACTF: folder = PL_fold; break;
1521 case EXACTFL: folder = PL_fold_locale; break;
1524 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
1526 trie->startstate = 1;
1527 trie->wordcount = word_count;
1528 RExC_rxi->data->data[ data_slot ] = (void*)trie;
1529 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
1530 if (!(UTF && folder))
1531 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
1532 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1533 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1536 trie_words = newAV();
1539 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
1540 if (!SvIOK(re_trie_maxbuff)) {
1541 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
1544 PerlIO_printf( Perl_debug_log,
1545 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
1546 (int)depth * 2 + 2, "",
1547 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
1548 REG_NODE_NUM(last), REG_NODE_NUM(tail),
1552 /* Find the node we are going to overwrite */
1553 if ( first == startbranch && OP( last ) != BRANCH ) {
1554 /* whole branch chain */
1557 /* branch sub-chain */
1558 convert = NEXTOPER( first );
1561 /* -- First loop and Setup --
1563 We first traverse the branches and scan each word to determine if it
1564 contains widechars, and how many unique chars there are, this is
1565 important as we have to build a table with at least as many columns as we
1568 We use an array of integers to represent the character codes 0..255
1569 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
1570 native representation of the character value as the key and IV's for the
1573 *TODO* If we keep track of how many times each character is used we can
1574 remap the columns so that the table compression later on is more
1575 efficient in terms of memory by ensuring the most common value is in the
1576 middle and the least common are on the outside. IMO this would be better
1577 than a most to least common mapping as theres a decent chance the most
1578 common letter will share a node with the least common, meaning the node
1579 will not be compressible. With a middle is most common approach the worst
1580 case is when we have the least common nodes twice.
1584 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1585 regnode * const noper = NEXTOPER( cur );
1586 const U8 *uc = (U8*)STRING( noper );
1587 const U8 * const e = uc + STR_LEN( noper );
1589 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1590 const U8 *scan = (U8*)NULL;
1591 U32 wordlen = 0; /* required init */
1593 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
1595 if (OP(noper) == NOTHING) {
1599 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1600 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1601 regardless of encoding */
1603 for ( ; uc < e ; uc += len ) {
1604 TRIE_CHARCOUNT(trie)++;
1608 if ( !trie->charmap[ uvc ] ) {
1609 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1611 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
1615 /* store the codepoint in the bitmap, and its folded
1617 TRIE_BITMAP_SET(trie,uvc);
1619 /* store the folded codepoint */
1620 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1623 /* store first byte of utf8 representation of
1624 variant codepoints */
1625 if (! UNI_IS_INVARIANT(uvc)) {
1626 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
1629 set_bit = 0; /* We've done our bit :-) */
1634 widecharmap = newHV();
1636 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
1639 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
1641 if ( !SvTRUE( *svpp ) ) {
1642 sv_setiv( *svpp, ++trie->uniquecharcount );
1647 if( cur == first ) {
1650 } else if (chars < trie->minlen) {
1652 } else if (chars > trie->maxlen) {
1656 } /* end first pass */
1657 DEBUG_TRIE_COMPILE_r(
1658 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1659 (int)depth * 2 + 2,"",
1660 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
1661 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1662 (int)trie->minlen, (int)trie->maxlen )
1666 We now know what we are dealing with in terms of unique chars and
1667 string sizes so we can calculate how much memory a naive
1668 representation using a flat table will take. If it's over a reasonable
1669 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
1670 conservative but potentially much slower representation using an array
1673 At the end we convert both representations into the same compressed
1674 form that will be used in regexec.c for matching with. The latter
1675 is a form that cannot be used to construct with but has memory
1676 properties similar to the list form and access properties similar
1677 to the table form making it both suitable for fast searches and
1678 small enough that its feasable to store for the duration of a program.
1680 See the comment in the code where the compressed table is produced
1681 inplace from the flat tabe representation for an explanation of how
1682 the compression works.
1687 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1690 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
1692 Second Pass -- Array Of Lists Representation
1694 Each state will be represented by a list of charid:state records
1695 (reg_trie_trans_le) the first such element holds the CUR and LEN
1696 points of the allocated array. (See defines above).
1698 We build the initial structure using the lists, and then convert
1699 it into the compressed table form which allows faster lookups
1700 (but cant be modified once converted).
1703 STRLEN transcount = 1;
1705 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1706 "%*sCompiling trie using list compiler\n",
1707 (int)depth * 2 + 2, ""));
1709 trie->states = (reg_trie_state *)
1710 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1711 sizeof(reg_trie_state) );
1715 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1717 regnode * const noper = NEXTOPER( cur );
1718 U8 *uc = (U8*)STRING( noper );
1719 const U8 * const e = uc + STR_LEN( noper );
1720 U32 state = 1; /* required init */
1721 U16 charid = 0; /* sanity init */
1722 U8 *scan = (U8*)NULL; /* sanity init */
1723 STRLEN foldlen = 0; /* required init */
1724 U32 wordlen = 0; /* required init */
1725 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1727 if (OP(noper) != NOTHING) {
1728 for ( ; uc < e ; uc += len ) {
1733 charid = trie->charmap[ uvc ];
1735 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1739 charid=(U16)SvIV( *svpp );
1742 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1749 if ( !trie->states[ state ].trans.list ) {
1750 TRIE_LIST_NEW( state );
1752 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1753 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1754 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1759 newstate = next_alloc++;
1760 prev_states[newstate] = state;
1761 TRIE_LIST_PUSH( state, charid, newstate );
1766 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1770 TRIE_HANDLE_WORD(state);
1772 } /* end second pass */
1774 /* next alloc is the NEXT state to be allocated */
1775 trie->statecount = next_alloc;
1776 trie->states = (reg_trie_state *)
1777 PerlMemShared_realloc( trie->states,
1779 * sizeof(reg_trie_state) );
1781 /* and now dump it out before we compress it */
1782 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1783 revcharmap, next_alloc,
1787 trie->trans = (reg_trie_trans *)
1788 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
1795 for( state=1 ; state < next_alloc ; state ++ ) {
1799 DEBUG_TRIE_COMPILE_MORE_r(
1800 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1804 if (trie->states[state].trans.list) {
1805 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1809 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1810 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1811 if ( forid < minid ) {
1813 } else if ( forid > maxid ) {
1817 if ( transcount < tp + maxid - minid + 1) {
1819 trie->trans = (reg_trie_trans *)
1820 PerlMemShared_realloc( trie->trans,
1822 * sizeof(reg_trie_trans) );
1823 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1825 base = trie->uniquecharcount + tp - minid;
1826 if ( maxid == minid ) {
1828 for ( ; zp < tp ; zp++ ) {
1829 if ( ! trie->trans[ zp ].next ) {
1830 base = trie->uniquecharcount + zp - minid;
1831 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1832 trie->trans[ zp ].check = state;
1838 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1839 trie->trans[ tp ].check = state;
1844 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
1845 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
1846 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1847 trie->trans[ tid ].check = state;
1849 tp += ( maxid - minid + 1 );
1851 Safefree(trie->states[ state ].trans.list);
1854 DEBUG_TRIE_COMPILE_MORE_r(
1855 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1858 trie->states[ state ].trans.base=base;
1860 trie->lasttrans = tp + 1;
1864 Second Pass -- Flat Table Representation.
1866 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1867 We know that we will need Charcount+1 trans at most to store the data
1868 (one row per char at worst case) So we preallocate both structures
1869 assuming worst case.
1871 We then construct the trie using only the .next slots of the entry
1874 We use the .check field of the first entry of the node temporarily to
1875 make compression both faster and easier by keeping track of how many non
1876 zero fields are in the node.
1878 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1881 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1882 number representing the first entry of the node, and state as a
1883 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1884 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1885 are 2 entrys per node. eg:
1893 The table is internally in the right hand, idx form. However as we also
1894 have to deal with the states array which is indexed by nodenum we have to
1895 use TRIE_NODENUM() to convert.
1898 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1899 "%*sCompiling trie using table compiler\n",
1900 (int)depth * 2 + 2, ""));
1902 trie->trans = (reg_trie_trans *)
1903 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1904 * trie->uniquecharcount + 1,
1905 sizeof(reg_trie_trans) );
1906 trie->states = (reg_trie_state *)
1907 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1908 sizeof(reg_trie_state) );
1909 next_alloc = trie->uniquecharcount + 1;
1912 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1914 regnode * const noper = NEXTOPER( cur );
1915 const U8 *uc = (U8*)STRING( noper );
1916 const U8 * const e = uc + STR_LEN( noper );
1918 U32 state = 1; /* required init */
1920 U16 charid = 0; /* sanity init */
1921 U32 accept_state = 0; /* sanity init */
1922 U8 *scan = (U8*)NULL; /* sanity init */
1924 STRLEN foldlen = 0; /* required init */
1925 U32 wordlen = 0; /* required init */
1926 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1928 if ( OP(noper) != NOTHING ) {
1929 for ( ; uc < e ; uc += len ) {
1934 charid = trie->charmap[ uvc ];
1936 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
1937 charid = svpp ? (U16)SvIV(*svpp) : 0;
1941 if ( !trie->trans[ state + charid ].next ) {
1942 trie->trans[ state + charid ].next = next_alloc;
1943 trie->trans[ state ].check++;
1944 prev_states[TRIE_NODENUM(next_alloc)]
1945 = TRIE_NODENUM(state);
1946 next_alloc += trie->uniquecharcount;
1948 state = trie->trans[ state + charid ].next;
1950 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1952 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1955 accept_state = TRIE_NODENUM( state );
1956 TRIE_HANDLE_WORD(accept_state);
1958 } /* end second pass */
1960 /* and now dump it out before we compress it */
1961 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1963 next_alloc, depth+1));
1967 * Inplace compress the table.*
1969 For sparse data sets the table constructed by the trie algorithm will
1970 be mostly 0/FAIL transitions or to put it another way mostly empty.
1971 (Note that leaf nodes will not contain any transitions.)
1973 This algorithm compresses the tables by eliminating most such
1974 transitions, at the cost of a modest bit of extra work during lookup:
1976 - Each states[] entry contains a .base field which indicates the
1977 index in the state[] array wheres its transition data is stored.
1979 - If .base is 0 there are no valid transitions from that node.
1981 - If .base is nonzero then charid is added to it to find an entry in
1984 -If trans[states[state].base+charid].check!=state then the
1985 transition is taken to be a 0/Fail transition. Thus if there are fail
1986 transitions at the front of the node then the .base offset will point
1987 somewhere inside the previous nodes data (or maybe even into a node
1988 even earlier), but the .check field determines if the transition is
1992 The following process inplace converts the table to the compressed
1993 table: We first do not compress the root node 1,and mark all its
1994 .check pointers as 1 and set its .base pointer as 1 as well. This
1995 allows us to do a DFA construction from the compressed table later,
1996 and ensures that any .base pointers we calculate later are greater
1999 - We set 'pos' to indicate the first entry of the second node.
2001 - We then iterate over the columns of the node, finding the first and
2002 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
2003 and set the .check pointers accordingly, and advance pos
2004 appropriately and repreat for the next node. Note that when we copy
2005 the next pointers we have to convert them from the original
2006 NODEIDX form to NODENUM form as the former is not valid post
2009 - If a node has no transitions used we mark its base as 0 and do not
2010 advance the pos pointer.
2012 - If a node only has one transition we use a second pointer into the
2013 structure to fill in allocated fail transitions from other states.
2014 This pointer is independent of the main pointer and scans forward
2015 looking for null transitions that are allocated to a state. When it
2016 finds one it writes the single transition into the "hole". If the
2017 pointer doesnt find one the single transition is appended as normal.
2019 - Once compressed we can Renew/realloc the structures to release the
2022 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
2023 specifically Fig 3.47 and the associated pseudocode.
2027 const U32 laststate = TRIE_NODENUM( next_alloc );
2030 trie->statecount = laststate;
2032 for ( state = 1 ; state < laststate ; state++ ) {
2034 const U32 stateidx = TRIE_NODEIDX( state );
2035 const U32 o_used = trie->trans[ stateidx ].check;
2036 U32 used = trie->trans[ stateidx ].check;
2037 trie->trans[ stateidx ].check = 0;
2039 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
2040 if ( flag || trie->trans[ stateidx + charid ].next ) {
2041 if ( trie->trans[ stateidx + charid ].next ) {
2043 for ( ; zp < pos ; zp++ ) {
2044 if ( ! trie->trans[ zp ].next ) {
2048 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
2049 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
2050 trie->trans[ zp ].check = state;
2051 if ( ++zp > pos ) pos = zp;
2058 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
2060 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
2061 trie->trans[ pos ].check = state;
2066 trie->lasttrans = pos + 1;
2067 trie->states = (reg_trie_state *)
2068 PerlMemShared_realloc( trie->states, laststate
2069 * sizeof(reg_trie_state) );
2070 DEBUG_TRIE_COMPILE_MORE_r(
2071 PerlIO_printf( Perl_debug_log,
2072 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
2073 (int)depth * 2 + 2,"",
2074 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
2077 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
2080 } /* end table compress */
2082 DEBUG_TRIE_COMPILE_MORE_r(
2083 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
2084 (int)depth * 2 + 2, "",
2085 (UV)trie->statecount,
2086 (UV)trie->lasttrans)
2088 /* resize the trans array to remove unused space */
2089 trie->trans = (reg_trie_trans *)
2090 PerlMemShared_realloc( trie->trans, trie->lasttrans
2091 * sizeof(reg_trie_trans) );
2093 { /* Modify the program and insert the new TRIE node */
2094 U8 nodetype =(U8)(flags & 0xFF);
2098 regnode *optimize = NULL;
2099 #ifdef RE_TRACK_PATTERN_OFFSETS
2102 U32 mjd_nodelen = 0;
2103 #endif /* RE_TRACK_PATTERN_OFFSETS */
2104 #endif /* DEBUGGING */
2106 This means we convert either the first branch or the first Exact,
2107 depending on whether the thing following (in 'last') is a branch
2108 or not and whther first is the startbranch (ie is it a sub part of
2109 the alternation or is it the whole thing.)
2110 Assuming its a sub part we convert the EXACT otherwise we convert
2111 the whole branch sequence, including the first.
2113 /* Find the node we are going to overwrite */
2114 if ( first != startbranch || OP( last ) == BRANCH ) {
2115 /* branch sub-chain */
2116 NEXT_OFF( first ) = (U16)(last - first);
2117 #ifdef RE_TRACK_PATTERN_OFFSETS
2119 mjd_offset= Node_Offset((convert));
2120 mjd_nodelen= Node_Length((convert));
2123 /* whole branch chain */
2125 #ifdef RE_TRACK_PATTERN_OFFSETS
2128 const regnode *nop = NEXTOPER( convert );
2129 mjd_offset= Node_Offset((nop));
2130 mjd_nodelen= Node_Length((nop));
2134 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2135 (int)depth * 2 + 2, "",
2136 (UV)mjd_offset, (UV)mjd_nodelen)
2139 /* But first we check to see if there is a common prefix we can
2140 split out as an EXACT and put in front of the TRIE node. */
2141 trie->startstate= 1;
2142 if ( trie->bitmap && !widecharmap && !trie->jump ) {
2144 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
2148 const U32 base = trie->states[ state ].trans.base;
2150 if ( trie->states[state].wordnum )
2153 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
2154 if ( ( base + ofs >= trie->uniquecharcount ) &&
2155 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
2156 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2158 if ( ++count > 1 ) {
2159 SV **tmp = av_fetch( revcharmap, ofs, 0);
2160 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
2161 if ( state == 1 ) break;
2163 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2165 PerlIO_printf(Perl_debug_log,
2166 "%*sNew Start State=%"UVuf" Class: [",
2167 (int)depth * 2 + 2, "",
2170 SV ** const tmp = av_fetch( revcharmap, idx, 0);
2171 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
2173 TRIE_BITMAP_SET(trie,*ch);
2175 TRIE_BITMAP_SET(trie, folder[ *ch ]);
2177 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
2181 TRIE_BITMAP_SET(trie,*ch);
2183 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2184 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2190 SV **tmp = av_fetch( revcharmap, idx, 0);
2192 char *ch = SvPV( *tmp, len );
2194 SV *sv=sv_newmortal();
2195 PerlIO_printf( Perl_debug_log,
2196 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2197 (int)depth * 2 + 2, "",
2199 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2200 PL_colors[0], PL_colors[1],
2201 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2202 PERL_PV_ESCAPE_FIRSTCHAR
2207 OP( convert ) = nodetype;
2208 str=STRING(convert);
2211 STR_LEN(convert) += len;
2217 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
2222 trie->prefixlen = (state-1);
2224 regnode *n = convert+NODE_SZ_STR(convert);
2225 NEXT_OFF(convert) = NODE_SZ_STR(convert);
2226 trie->startstate = state;
2227 trie->minlen -= (state - 1);
2228 trie->maxlen -= (state - 1);
2230 /* At least the UNICOS C compiler choked on this
2231 * being argument to DEBUG_r(), so let's just have
2234 #ifdef PERL_EXT_RE_BUILD
2240 regnode *fix = convert;
2241 U32 word = trie->wordcount;
2243 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2244 while( ++fix < n ) {
2245 Set_Node_Offset_Length(fix, 0, 0);
2248 SV ** const tmp = av_fetch( trie_words, word, 0 );
2250 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2251 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2253 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2261 NEXT_OFF(convert) = (U16)(tail - convert);
2262 DEBUG_r(optimize= n);
2268 if ( trie->maxlen ) {
2269 NEXT_OFF( convert ) = (U16)(tail - convert);
2270 ARG_SET( convert, data_slot );
2271 /* Store the offset to the first unabsorbed branch in
2272 jump[0], which is otherwise unused by the jump logic.
2273 We use this when dumping a trie and during optimisation. */
2275 trie->jump[0] = (U16)(nextbranch - convert);
2277 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2278 * and there is a bitmap
2279 * and the first "jump target" node we found leaves enough room
2280 * then convert the TRIE node into a TRIEC node, with the bitmap
2281 * embedded inline in the opcode - this is hypothetically faster.
2283 if ( !trie->states[trie->startstate].wordnum
2285 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
2287 OP( convert ) = TRIEC;
2288 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
2289 PerlMemShared_free(trie->bitmap);
2292 OP( convert ) = TRIE;
2294 /* store the type in the flags */
2295 convert->flags = nodetype;
2299 + regarglen[ OP( convert ) ];
2301 /* XXX We really should free up the resource in trie now,
2302 as we won't use them - (which resources?) dmq */
2304 /* needed for dumping*/
2305 DEBUG_r(if (optimize) {
2306 regnode *opt = convert;
2308 while ( ++opt < optimize) {
2309 Set_Node_Offset_Length(opt,0,0);
2312 Try to clean up some of the debris left after the
2315 while( optimize < jumper ) {
2316 mjd_nodelen += Node_Length((optimize));
2317 OP( optimize ) = OPTIMIZED;
2318 Set_Node_Offset_Length(optimize,0,0);
2321 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
2323 } /* end node insert */
2325 /* Finish populating the prev field of the wordinfo array. Walk back
2326 * from each accept state until we find another accept state, and if
2327 * so, point the first word's .prev field at the second word. If the
2328 * second already has a .prev field set, stop now. This will be the
2329 * case either if we've already processed that word's accept state,
2330 * or that state had multiple words, and the overspill words were
2331 * already linked up earlier.
2338 for (word=1; word <= trie->wordcount; word++) {
2340 if (trie->wordinfo[word].prev)
2342 state = trie->wordinfo[word].accept;
2344 state = prev_states[state];
2347 prev = trie->states[state].wordnum;
2351 trie->wordinfo[word].prev = prev;
2353 Safefree(prev_states);
2357 /* and now dump out the compressed format */
2358 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2360 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2362 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2363 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2365 SvREFCNT_dec(revcharmap);
2369 : trie->startstate>1
2375 S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2377 /* The Trie is constructed and compressed now so we can build a fail array if it's needed
2379 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2380 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2383 We find the fail state for each state in the trie, this state is the longest proper
2384 suffix of the current state's 'word' that is also a proper prefix of another word in our
2385 trie. State 1 represents the word '' and is thus the default fail state. This allows
2386 the DFA not to have to restart after its tried and failed a word at a given point, it
2387 simply continues as though it had been matching the other word in the first place.
2389 'abcdgu'=~/abcdefg|cdgu/
2390 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2391 fail, which would bring us to the state representing 'd' in the second word where we would
2392 try 'g' and succeed, proceeding to match 'cdgu'.
2394 /* add a fail transition */
2395 const U32 trie_offset = ARG(source);
2396 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
2398 const U32 ucharcount = trie->uniquecharcount;
2399 const U32 numstates = trie->statecount;
2400 const U32 ubound = trie->lasttrans + ucharcount;
2404 U32 base = trie->states[ 1 ].trans.base;
2407 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2408 GET_RE_DEBUG_FLAGS_DECL;
2410 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
2412 PERL_UNUSED_ARG(depth);
2416 ARG_SET( stclass, data_slot );
2417 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
2418 RExC_rxi->data->data[ data_slot ] = (void*)aho;
2419 aho->trie=trie_offset;
2420 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2421 Copy( trie->states, aho->states, numstates, reg_trie_state );
2422 Newxz( q, numstates, U32);
2423 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
2426 /* initialize fail[0..1] to be 1 so that we always have
2427 a valid final fail state */
2428 fail[ 0 ] = fail[ 1 ] = 1;
2430 for ( charid = 0; charid < ucharcount ; charid++ ) {
2431 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2433 q[ q_write ] = newstate;
2434 /* set to point at the root */
2435 fail[ q[ q_write++ ] ]=1;
2438 while ( q_read < q_write) {
2439 const U32 cur = q[ q_read++ % numstates ];
2440 base = trie->states[ cur ].trans.base;
2442 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2443 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2445 U32 fail_state = cur;
2448 fail_state = fail[ fail_state ];
2449 fail_base = aho->states[ fail_state ].trans.base;
2450 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2452 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2453 fail[ ch_state ] = fail_state;
2454 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2456 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2458 q[ q_write++ % numstates] = ch_state;
2462 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2463 when we fail in state 1, this allows us to use the
2464 charclass scan to find a valid start char. This is based on the principle
2465 that theres a good chance the string being searched contains lots of stuff
2466 that cant be a start char.
2468 fail[ 0 ] = fail[ 1 ] = 0;
2469 DEBUG_TRIE_COMPILE_r({
2470 PerlIO_printf(Perl_debug_log,
2471 "%*sStclass Failtable (%"UVuf" states): 0",
2472 (int)(depth * 2), "", (UV)numstates
2474 for( q_read=1; q_read<numstates; q_read++ ) {
2475 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2477 PerlIO_printf(Perl_debug_log, "\n");
2480 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
2485 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2486 * These need to be revisited when a newer toolchain becomes available.
2488 #if defined(__sparc64__) && defined(__GNUC__)
2489 # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2490 # undef SPARC64_GCC_WORKAROUND
2491 # define SPARC64_GCC_WORKAROUND 1
2495 #define DEBUG_PEEP(str,scan,depth) \
2496 DEBUG_OPTIMISE_r({if (scan){ \
2497 SV * const mysv=sv_newmortal(); \
2498 regnode *Next = regnext(scan); \
2499 regprop(RExC_rx, mysv, scan); \
2500 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
2501 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2502 Next ? (REG_NODE_NUM(Next)) : 0 ); \
2509 #define JOIN_EXACT(scan,min,flags) \
2510 if (PL_regkind[OP(scan)] == EXACT) \
2511 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2514 S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2515 /* Merge several consecutive EXACTish nodes into one. */
2516 regnode *n = regnext(scan);
2518 regnode *next = scan + NODE_SZ_STR(scan);
2522 regnode *stop = scan;
2523 GET_RE_DEBUG_FLAGS_DECL;
2525 PERL_UNUSED_ARG(depth);
2528 PERL_ARGS_ASSERT_JOIN_EXACT;
2529 #ifndef EXPERIMENTAL_INPLACESCAN
2530 PERL_UNUSED_ARG(flags);
2531 PERL_UNUSED_ARG(val);
2533 DEBUG_PEEP("join",scan,depth);
2535 /* Skip NOTHING, merge EXACT*. */
2537 ( PL_regkind[OP(n)] == NOTHING ||
2538 (stringok && (OP(n) == OP(scan))))
2540 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2542 if (OP(n) == TAIL || n > next)
2544 if (PL_regkind[OP(n)] == NOTHING) {
2545 DEBUG_PEEP("skip:",n,depth);
2546 NEXT_OFF(scan) += NEXT_OFF(n);
2547 next = n + NODE_STEP_REGNODE;
2554 else if (stringok) {
2555 const unsigned int oldl = STR_LEN(scan);
2556 regnode * const nnext = regnext(n);
2558 DEBUG_PEEP("merg",n,depth);
2561 if (oldl + STR_LEN(n) > U8_MAX)
2563 NEXT_OFF(scan) += NEXT_OFF(n);
2564 STR_LEN(scan) += STR_LEN(n);
2565 next = n + NODE_SZ_STR(n);
2566 /* Now we can overwrite *n : */
2567 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2575 #ifdef EXPERIMENTAL_INPLACESCAN
2576 if (flags && !NEXT_OFF(n)) {
2577 DEBUG_PEEP("atch", val, depth);
2578 if (reg_off_by_arg[OP(n)]) {
2579 ARG_SET(n, val - n);
2582 NEXT_OFF(n) = val - n;
2588 #define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390
2589 #define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS
2590 #define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0
2591 #define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS
2594 && ( OP(scan) == EXACTF || OP(scan) == EXACTFU || OP(scan) == EXACTFA)
2595 && ( STR_LEN(scan) >= 6 ) )
2598 Two problematic code points in Unicode casefolding of EXACT nodes:
2600 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2601 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2607 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2608 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2610 This means that in case-insensitive matching (or "loose matching",
2611 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2612 length of the above casefolded versions) can match a target string
2613 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2614 This would rather mess up the minimum length computation.
2616 What we'll do is to look for the tail four bytes, and then peek
2617 at the preceding two bytes to see whether we need to decrease
2618 the minimum length by four (six minus two).
2620 Thanks to the design of UTF-8, there cannot be false matches:
2621 A sequence of valid UTF-8 bytes cannot be a subsequence of
2622 another valid sequence of UTF-8 bytes.
2625 char * const s0 = STRING(scan), *s, *t;
2626 char * const s1 = s0 + STR_LEN(scan) - 1;
2627 char * const s2 = s1 - 4;
2628 #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2629 const char t0[] = "\xaf\x49\xaf\x42";
2631 const char t0[] = "\xcc\x88\xcc\x81";
2633 const char * const t1 = t0 + 3;
2636 s < s2 && (t = ninstr(s, s1, t0, t1));
2639 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2640 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2642 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2643 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
2651 n = scan + NODE_SZ_STR(scan);
2653 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2660 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2664 /* REx optimizer. Converts nodes into quicker variants "in place".
2665 Finds fixed substrings. */
2667 /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
2668 to the position after last scanned or to NULL. */
2670 #define INIT_AND_WITHP \
2671 assert(!and_withp); \
2672 Newx(and_withp,1,struct regnode_charclass_class); \
2673 SAVEFREEPV(and_withp)
2675 /* this is a chain of data about sub patterns we are processing that
2676 need to be handled separately/specially in study_chunk. Its so
2677 we can simulate recursion without losing state. */
2679 typedef struct scan_frame {
2680 regnode *last; /* last node to process in this frame */
2681 regnode *next; /* next node to process when last is reached */
2682 struct scan_frame *prev; /*previous frame*/
2683 I32 stop; /* what stopparen do we use */
2687 #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2689 #define CASE_SYNST_FNC(nAmE) \
2691 if (flags & SCF_DO_STCLASS_AND) { \
2692 for (value = 0; value < 256; value++) \
2693 if (!is_ ## nAmE ## _cp(value)) \
2694 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2697 for (value = 0; value < 256; value++) \
2698 if (is_ ## nAmE ## _cp(value)) \
2699 ANYOF_BITMAP_SET(data->start_class, value); \
2703 if (flags & SCF_DO_STCLASS_AND) { \
2704 for (value = 0; value < 256; value++) \
2705 if (is_ ## nAmE ## _cp(value)) \
2706 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2709 for (value = 0; value < 256; value++) \
2710 if (!is_ ## nAmE ## _cp(value)) \
2711 ANYOF_BITMAP_SET(data->start_class, value); \
2718 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
2719 I32 *minlenp, I32 *deltap,
2724 struct regnode_charclass_class *and_withp,
2725 U32 flags, U32 depth)
2726 /* scanp: Start here (read-write). */
2727 /* deltap: Write maxlen-minlen here. */
2728 /* last: Stop before this one. */
2729 /* data: string data about the pattern */
2730 /* stopparen: treat close N as END */
2731 /* recursed: which subroutines have we recursed into */
2732 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
2735 I32 min = 0, pars = 0, code;
2736 regnode *scan = *scanp, *next;
2738 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
2739 int is_inf_internal = 0; /* The studied chunk is infinite */
2740 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2741 scan_data_t data_fake;
2742 SV *re_trie_maxbuff = NULL;
2743 regnode *first_non_open = scan;
2744 I32 stopmin = I32_MAX;
2745 scan_frame *frame = NULL;
2746 GET_RE_DEBUG_FLAGS_DECL;
2748 PERL_ARGS_ASSERT_STUDY_CHUNK;
2751 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
2755 while (first_non_open && OP(first_non_open) == OPEN)
2756 first_non_open=regnext(first_non_open);
2761 while ( scan && OP(scan) != END && scan < last ){
2762 /* Peephole optimizer: */
2763 DEBUG_STUDYDATA("Peep:", data,depth);
2764 DEBUG_PEEP("Peep",scan,depth);
2765 JOIN_EXACT(scan,&min,0);
2767 /* Follow the next-chain of the current node and optimize
2768 away all the NOTHINGs from it. */
2769 if (OP(scan) != CURLYX) {
2770 const int max = (reg_off_by_arg[OP(scan)]
2772 /* I32 may be smaller than U16 on CRAYs! */
2773 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2774 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2778 /* Skip NOTHING and LONGJMP. */
2779 while ((n = regnext(n))
2780 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2781 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2782 && off + noff < max)
2784 if (reg_off_by_arg[OP(scan)])
2787 NEXT_OFF(scan) = off;
2792 /* The principal pseudo-switch. Cannot be a switch, since we
2793 look into several different things. */
2794 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2795 || OP(scan) == IFTHEN) {
2796 next = regnext(scan);
2798 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2800 if (OP(next) == code || code == IFTHEN) {
2801 /* NOTE - There is similar code to this block below for handling
2802 TRIE nodes on a re-study. If you change stuff here check there
2804 I32 max1 = 0, min1 = I32_MAX, num = 0;
2805 struct regnode_charclass_class accum;
2806 regnode * const startbranch=scan;
2808 if (flags & SCF_DO_SUBSTR)
2809 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
2810 if (flags & SCF_DO_STCLASS)
2811 cl_init_zero(pRExC_state, &accum);
2813 while (OP(scan) == code) {
2814 I32 deltanext, minnext, f = 0, fake;
2815 struct regnode_charclass_class this_class;
2818 data_fake.flags = 0;
2820 data_fake.whilem_c = data->whilem_c;
2821 data_fake.last_closep = data->last_closep;
2824 data_fake.last_closep = &fake;
2826 data_fake.pos_delta = delta;
2827 next = regnext(scan);
2828 scan = NEXTOPER(scan);
2830 scan = NEXTOPER(scan);
2831 if (flags & SCF_DO_STCLASS) {
2832 cl_init(pRExC_state, &this_class);
2833 data_fake.start_class = &this_class;
2834 f = SCF_DO_STCLASS_AND;
2836 if (flags & SCF_WHILEM_VISITED_POS)
2837 f |= SCF_WHILEM_VISITED_POS;
2839 /* we suppose the run is continuous, last=next...*/
2840 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2842 stopparen, recursed, NULL, f,depth+1);
2845 if (max1 < minnext + deltanext)
2846 max1 = minnext + deltanext;
2847 if (deltanext == I32_MAX)
2848 is_inf = is_inf_internal = 1;
2850 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2852 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2853 if ( stopmin > minnext)
2854 stopmin = min + min1;
2855 flags &= ~SCF_DO_SUBSTR;
2857 data->flags |= SCF_SEEN_ACCEPT;
2860 if (data_fake.flags & SF_HAS_EVAL)
2861 data->flags |= SF_HAS_EVAL;
2862 data->whilem_c = data_fake.whilem_c;
2864 if (flags & SCF_DO_STCLASS)
2865 cl_or(pRExC_state, &accum, &this_class);
2867 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2869 if (flags & SCF_DO_SUBSTR) {
2870 data->pos_min += min1;
2871 data->pos_delta += max1 - min1;
2872 if (max1 != min1 || is_inf)
2873 data->longest = &(data->longest_float);
2876 delta += max1 - min1;
2877 if (flags & SCF_DO_STCLASS_OR) {
2878 cl_or(pRExC_state, data->start_class, &accum);
2880 cl_and(data->start_class, and_withp);
2881 flags &= ~SCF_DO_STCLASS;
2884 else if (flags & SCF_DO_STCLASS_AND) {
2886 cl_and(data->start_class, &accum);
2887 flags &= ~SCF_DO_STCLASS;
2890 /* Switch to OR mode: cache the old value of
2891 * data->start_class */
2893 StructCopy(data->start_class, and_withp,
2894 struct regnode_charclass_class);
2895 flags &= ~SCF_DO_STCLASS_AND;
2896 StructCopy(&accum, data->start_class,
2897 struct regnode_charclass_class);
2898 flags |= SCF_DO_STCLASS_OR;
2899 data->start_class->flags |= ANYOF_EOS;
2903 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2906 Assuming this was/is a branch we are dealing with: 'scan' now
2907 points at the item that follows the branch sequence, whatever
2908 it is. We now start at the beginning of the sequence and look
2915 which would be constructed from a pattern like /A|LIST|OF|WORDS/
2917 If we can find such a subsequence we need to turn the first
2918 element into a trie and then add the subsequent branch exact
2919 strings to the trie.
2923 1. patterns where the whole set of branches can be converted.
2925 2. patterns where only a subset can be converted.
2927 In case 1 we can replace the whole set with a single regop
2928 for the trie. In case 2 we need to keep the start and end
2931 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2932 becomes BRANCH TRIE; BRANCH X;
2934 There is an additional case, that being where there is a
2935 common prefix, which gets split out into an EXACT like node
2936 preceding the TRIE node.
2938 If x(1..n)==tail then we can do a simple trie, if not we make
2939 a "jump" trie, such that when we match the appropriate word
2940 we "jump" to the appropriate tail node. Essentially we turn
2941 a nested if into a case structure of sorts.
2946 if (!re_trie_maxbuff) {
2947 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2948 if (!SvIOK(re_trie_maxbuff))
2949 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2951 if ( SvIV(re_trie_maxbuff)>=0 ) {
2953 regnode *first = (regnode *)NULL;
2954 regnode *last = (regnode *)NULL;
2955 regnode *tail = scan;
2960 SV * const mysv = sv_newmortal(); /* for dumping */
2962 /* var tail is used because there may be a TAIL
2963 regop in the way. Ie, the exacts will point to the
2964 thing following the TAIL, but the last branch will
2965 point at the TAIL. So we advance tail. If we
2966 have nested (?:) we may have to move through several
2970 while ( OP( tail ) == TAIL ) {
2971 /* this is the TAIL generated by (?:) */
2972 tail = regnext( tail );
2977 regprop(RExC_rx, mysv, tail );
2978 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2979 (int)depth * 2 + 2, "",
2980 "Looking for TRIE'able sequences. Tail node is: ",
2981 SvPV_nolen_const( mysv )
2987 step through the branches, cur represents each
2988 branch, noper is the first thing to be matched
2989 as part of that branch and noper_next is the
2990 regnext() of that node. if noper is an EXACT
2991 and noper_next is the same as scan (our current
2992 position in the regex) then the EXACT branch is
2993 a possible optimization target. Once we have
2994 two or more consecutive such branches we can
2995 create a trie of the EXACT's contents and stich
2996 it in place. If the sequence represents all of
2997 the branches we eliminate the whole thing and
2998 replace it with a single TRIE. If it is a
2999 subsequence then we need to stitch it in. This
3000 means the first branch has to remain, and needs
3001 to be repointed at the item on the branch chain
3002 following the last branch optimized. This could
3003 be either a BRANCH, in which case the
3004 subsequence is internal, or it could be the
3005 item following the branch sequence in which
3006 case the subsequence is at the end.
3010 /* dont use tail as the end marker for this traverse */
3011 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
3012 regnode * const noper = NEXTOPER( cur );
3013 #if defined(DEBUGGING) || defined(NOJUMPTRIE)
3014 regnode * const noper_next = regnext( noper );
3018 regprop(RExC_rx, mysv, cur);
3019 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
3020 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
3022 regprop(RExC_rx, mysv, noper);
3023 PerlIO_printf( Perl_debug_log, " -> %s",
3024 SvPV_nolen_const(mysv));
3027 regprop(RExC_rx, mysv, noper_next );
3028 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
3029 SvPV_nolen_const(mysv));
3031 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
3032 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
3034 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
3035 : PL_regkind[ OP( noper ) ] == EXACT )
3036 || OP(noper) == NOTHING )
3038 && noper_next == tail
3043 if ( !first || optype == NOTHING ) {
3044 if (!first) first = cur;
3045 optype = OP( noper );
3051 Currently the trie logic handles case insensitive matching properly only
3052 when the pattern is UTF-8 and the node is EXACTFU (thus forcing unicode
3055 If/when this is fixed the following define can be swapped
3056 in below to fully enable trie logic.
3058 #define TRIE_TYPE_IS_SAFE 1
3061 #define TRIE_TYPE_IS_SAFE ((UTF && optype == EXACTFU) || optype==EXACT)
3063 if ( last && TRIE_TYPE_IS_SAFE ) {
3064 make_trie( pRExC_state,
3065 startbranch, first, cur, tail, count,
3068 if ( PL_regkind[ OP( noper ) ] == EXACT
3070 && noper_next == tail
3075 optype = OP( noper );
3085 regprop(RExC_rx, mysv, cur);
3086 PerlIO_printf( Perl_debug_log,
3087 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
3088 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
3092 if ( last && TRIE_TYPE_IS_SAFE ) {
3093 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3094 #ifdef TRIE_STUDY_OPT
3095 if ( ((made == MADE_EXACT_TRIE &&
3096 startbranch == first)
3097 || ( first_non_open == first )) &&
3099 flags |= SCF_TRIE_RESTUDY;
3100 if ( startbranch == first
3103 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
3113 else if ( code == BRANCHJ ) { /* single branch is optimized. */
3114 scan = NEXTOPER(NEXTOPER(scan));
3115 } else /* single branch is optimized. */
3116 scan = NEXTOPER(scan);
3118 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
3119 scan_frame *newframe = NULL;
3124 if (OP(scan) != SUSPEND) {
3125 /* set the pointer */
3126 if (OP(scan) == GOSUB) {
3128 RExC_recurse[ARG2L(scan)] = scan;
3129 start = RExC_open_parens[paren-1];
3130 end = RExC_close_parens[paren-1];
3133 start = RExC_rxi->program + 1;
3137 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
3138 SAVEFREEPV(recursed);
3140 if (!PAREN_TEST(recursed,paren+1)) {
3141 PAREN_SET(recursed,paren+1);
3142 Newx(newframe,1,scan_frame);
3144 if (flags & SCF_DO_SUBSTR) {
3145 SCAN_COMMIT(pRExC_state,data,minlenp);
3146 data->longest = &(data->longest_float);
3148 is_inf = is_inf_internal = 1;
3149 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3150 cl_anything(pRExC_state, data->start_class);
3151 flags &= ~SCF_DO_STCLASS;
3154 Newx(newframe,1,scan_frame);
3157 end = regnext(scan);
3162 SAVEFREEPV(newframe);
3163 newframe->next = regnext(scan);
3164 newframe->last = last;
3165 newframe->stop = stopparen;
3166 newframe->prev = frame;
3176 else if (OP(scan) == EXACT) {
3177 I32 l = STR_LEN(scan);
3180 const U8 * const s = (U8*)STRING(scan);
3181 l = utf8_length(s, s + l);
3182 uc = utf8_to_uvchr(s, NULL);
3184 uc = *((U8*)STRING(scan));
3187 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3188 /* The code below prefers earlier match for fixed
3189 offset, later match for variable offset. */
3190 if (data->last_end == -1) { /* Update the start info. */
3191 data->last_start_min = data->pos_min;
3192 data->last_start_max = is_inf
3193 ? I32_MAX : data->pos_min + data->pos_delta;
3195 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3197 SvUTF8_on(data->last_found);
3199 SV * const sv = data->last_found;
3200 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3201 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3202 if (mg && mg->mg_len >= 0)
3203 mg->mg_len += utf8_length((U8*)STRING(scan),
3204 (U8*)STRING(scan)+STR_LEN(scan));
3206 data->last_end = data->pos_min + l;
3207 data->pos_min += l; /* As in the first entry. */
3208 data->flags &= ~SF_BEFORE_EOL;
3210 if (flags & SCF_DO_STCLASS_AND) {
3211 /* Check whether it is compatible with what we know already! */
3215 /* If compatible, we or it in below. It is compatible if is
3216 * in the bitmp and either 1) its bit or its fold is set, or 2)
3217 * it's for a locale. Even if there isn't unicode semantics
3218 * here, at runtime there may be because of matching against a
3219 * utf8 string, so accept a possible false positive for
3220 * latin1-range folds */
3222 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3223 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3224 && (!(data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD)
3225 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3230 ANYOF_CLASS_ZERO(data->start_class);
3231 ANYOF_BITMAP_ZERO(data->start_class);
3233 ANYOF_BITMAP_SET(data->start_class, uc);
3234 else if (uc >= 0x100) {
3237 /* Some Unicode code points fold to the Latin1 range; as
3238 * XXX temporary code, instead of figuring out if this is
3239 * one, just assume it is and set all the start class bits
3240 * that could be some such above 255 code point's fold
3241 * which will generate fals positives. As the code
3242 * elsewhere that does compute the fold settles down, it
3243 * can be extracted out and re-used here */
3244 for (i = 0; i < 256; i++){
3245 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3246 ANYOF_BITMAP_SET(data->start_class, i);
3250 data->start_class->flags &= ~ANYOF_EOS;
3252 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3254 else if (flags & SCF_DO_STCLASS_OR) {
3255 /* false positive possible if the class is case-folded */
3257 ANYOF_BITMAP_SET(data->start_class, uc);
3259 data->start_class->flags |= ANYOF_UNICODE_ALL;
3260 data->start_class->flags &= ~ANYOF_EOS;
3261 cl_and(data->start_class, and_withp);
3263 flags &= ~SCF_DO_STCLASS;
3265 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3266 I32 l = STR_LEN(scan);
3267 UV uc = *((U8*)STRING(scan));
3269 /* Search for fixed substrings supports EXACT only. */
3270 if (flags & SCF_DO_SUBSTR) {
3272 SCAN_COMMIT(pRExC_state, data, minlenp);
3275 const U8 * const s = (U8 *)STRING(scan);
3276 l = utf8_length(s, s + l);
3277 uc = utf8_to_uvchr(s, NULL);
3280 if (flags & SCF_DO_SUBSTR)
3282 if (flags & SCF_DO_STCLASS_AND) {
3283 /* Check whether it is compatible with what we know already! */
3286 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3287 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3288 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3292 ANYOF_CLASS_ZERO(data->start_class);
3293 ANYOF_BITMAP_ZERO(data->start_class);
3295 ANYOF_BITMAP_SET(data->start_class, uc);
3296 data->start_class->flags &= ~ANYOF_EOS;
3297 data->start_class->flags |= ANYOF_LOC_NONBITMAP_FOLD;
3298 if (OP(scan) == EXACTFL) {
3299 /* XXX This set is probably no longer necessary, and
3300 * probably wrong as LOCALE now is on in the initial
3302 data->start_class->flags |= ANYOF_LOCALE;
3306 /* Also set the other member of the fold pair. In case
3307 * that unicode semantics is called for at runtime, use
3308 * the full latin1 fold. (Can't do this for locale,
3309 * because not known until runtime */
3310 ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]);
3313 else if (uc >= 0x100) {
3315 for (i = 0; i < 256; i++){
3316 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) {
3317 ANYOF_BITMAP_SET(data->start_class, i);
3322 else if (flags & SCF_DO_STCLASS_OR) {
3323 if (data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) {
3324 /* false positive possible if the class is case-folded.
3325 Assume that the locale settings are the same... */
3327 ANYOF_BITMAP_SET(data->start_class, uc);
3328 if (OP(scan) != EXACTFL) {
3330 /* And set the other member of the fold pair, but
3331 * can't do that in locale because not known until
3333 ANYOF_BITMAP_SET(data->start_class,
3334 PL_fold_latin1[uc]);
3337 data->start_class->flags &= ~ANYOF_EOS;
3339 cl_and(data->start_class, and_withp);
3341 flags &= ~SCF_DO_STCLASS;
3343 else if (REGNODE_VARIES(OP(scan))) {
3344 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3345 I32 f = flags, pos_before = 0;
3346 regnode * const oscan = scan;
3347 struct regnode_charclass_class this_class;
3348 struct regnode_charclass_class *oclass = NULL;
3349 I32 next_is_eval = 0;
3351 switch (PL_regkind[OP(scan)]) {
3352 case WHILEM: /* End of (?:...)* . */
3353 scan = NEXTOPER(scan);
3356 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3357 next = NEXTOPER(scan);
3358 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3360 maxcount = REG_INFTY;
3361 next = regnext(scan);
3362 scan = NEXTOPER(scan);
3366 if (flags & SCF_DO_SUBSTR)
3371 if (flags & SCF_DO_STCLASS) {
3373 maxcount = REG_INFTY;
3374 next = regnext(scan);
3375 scan = NEXTOPER(scan);
3378 is_inf = is_inf_internal = 1;
3379 scan = regnext(scan);
3380 if (flags & SCF_DO_SUBSTR) {
3381 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
3382 data->longest = &(data->longest_float);
3384 goto optimize_curly_tail;
3386 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3387 && (scan->flags == stopparen))
3392 mincount = ARG1(scan);
3393 maxcount = ARG2(scan);
3395 next = regnext(scan);
3396 if (OP(scan) == CURLYX) {
3397 I32 lp = (data ? *(data->last_closep) : 0);
3398 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
3400 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3401 next_is_eval = (OP(scan) == EVAL);
3403 if (flags & SCF_DO_SUBSTR) {
3404 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
3405 pos_before = data->pos_min;
3409 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3411 data->flags |= SF_IS_INF;
3413 if (flags & SCF_DO_STCLASS) {
3414 cl_init(pRExC_state, &this_class);
3415 oclass = data->start_class;
3416 data->start_class = &this_class;
3417 f |= SCF_DO_STCLASS_AND;
3418 f &= ~SCF_DO_STCLASS_OR;
3420 /* Exclude from super-linear cache processing any {n,m}
3421 regops for which the combination of input pos and regex
3422 pos is not enough information to determine if a match
3425 For example, in the regex /foo(bar\s*){4,8}baz/ with the
3426 regex pos at the \s*, the prospects for a match depend not
3427 only on the input position but also on how many (bar\s*)
3428 repeats into the {4,8} we are. */
3429 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
3430 f &= ~SCF_WHILEM_VISITED_POS;
3432 /* This will finish on WHILEM, setting scan, or on NULL: */
3433 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3434 last, data, stopparen, recursed, NULL,
3436 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
3438 if (flags & SCF_DO_STCLASS)
3439 data->start_class = oclass;
3440 if (mincount == 0 || minnext == 0) {
3441 if (flags & SCF_DO_STCLASS_OR) {
3442 cl_or(pRExC_state, data->start_class, &this_class);
3444 else if (flags & SCF_DO_STCLASS_AND) {
3445 /* Switch to OR mode: cache the old value of
3446 * data->start_class */
3448 StructCopy(data->start_class, and_withp,
3449 struct regnode_charclass_class);
3450 flags &= ~SCF_DO_STCLASS_AND;
3451 StructCopy(&this_class, data->start_class,
3452 struct regnode_charclass_class);
3453 flags |= SCF_DO_STCLASS_OR;
3454 data->start_class->flags |= ANYOF_EOS;
3456 } else { /* Non-zero len */
3457 if (flags & SCF_DO_STCLASS_OR) {
3458 cl_or(pRExC_state, data->start_class, &this_class);
3459 cl_and(data->start_class, and_withp);
3461 else if (flags & SCF_DO_STCLASS_AND)
3462 cl_and(data->start_class, &this_class);
3463 flags &= ~SCF_DO_STCLASS;
3465 if (!scan) /* It was not CURLYX, but CURLY. */
3467 if ( /* ? quantifier ok, except for (?{ ... }) */
3468 (next_is_eval || !(mincount == 0 && maxcount == 1))
3469 && (minnext == 0) && (deltanext == 0)
3470 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3471 && maxcount <= REG_INFTY/3) /* Complement check for big count */
3473 ckWARNreg(RExC_parse,
3474 "Quantifier unexpected on zero-length expression");
3477 min += minnext * mincount;
3478 is_inf_internal |= ((maxcount == REG_INFTY
3479 && (minnext + deltanext) > 0)
3480 || deltanext == I32_MAX);
3481 is_inf |= is_inf_internal;
3482 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3484 /* Try powerful optimization CURLYX => CURLYN. */
3485 if ( OP(oscan) == CURLYX && data
3486 && data->flags & SF_IN_PAR
3487 && !(data->flags & SF_HAS_EVAL)
3488 && !deltanext && minnext == 1 ) {
3489 /* Try to optimize to CURLYN. */
3490 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3491 regnode * const nxt1 = nxt;
3498 if (!REGNODE_SIMPLE(OP(nxt))
3499 && !(PL_regkind[OP(nxt)] == EXACT
3500 && STR_LEN(nxt) == 1))
3506 if (OP(nxt) != CLOSE)
3508 if (RExC_open_parens) {
3509 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3510 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3512 /* Now we know that nxt2 is the only contents: */
3513 oscan->flags = (U8)ARG(nxt);
3515 OP(nxt1) = NOTHING; /* was OPEN. */
3518 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3519 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3520 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
3521 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3522 OP(nxt + 1) = OPTIMIZED; /* was count. */
3523 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
3528 /* Try optimization CURLYX => CURLYM. */
3529 if ( OP(oscan) == CURLYX && data
3530 && !(data->flags & SF_HAS_PAR)
3531 && !(data->flags & SF_HAS_EVAL)
3532 && !deltanext /* atom is fixed width */
3533 && minnext != 0 /* CURLYM can't handle zero width */
3535 /* XXXX How to optimize if data == 0? */
3536 /* Optimize to a simpler form. */
3537 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3541 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3542 && (OP(nxt2) != WHILEM))
3544 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3545 /* Need to optimize away parenths. */
3546 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
3547 /* Set the parenth number. */
3548 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3550 oscan->flags = (U8)ARG(nxt);
3551 if (RExC_open_parens) {
3552 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3553 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
3555 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3556 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3559 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3560 OP(nxt + 1) = OPTIMIZED; /* was count. */
3561 NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */
3562 NEXT_OFF(nxt + 1) = 0; /* just for consistency. */
3565 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3566 regnode *nnxt = regnext(nxt1);
3568 if (reg_off_by_arg[OP(nxt1)])
3569 ARG_SET(nxt1, nxt2 - nxt1);
3570 else if (nxt2 - nxt1 < U16_MAX)
3571 NEXT_OFF(nxt1) = nxt2 - nxt1;
3573 OP(nxt) = NOTHING; /* Cannot beautify */
3578 /* Optimize again: */
3579 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3580 NULL, stopparen, recursed, NULL, 0,depth+1);
3585 else if ((OP(oscan) == CURLYX)
3586 && (flags & SCF_WHILEM_VISITED_POS)
3587 /* See the comment on a similar expression above.
3588 However, this time it's not a subexpression
3589 we care about, but the expression itself. */
3590 && (maxcount == REG_INFTY)
3591 && data && ++data->whilem_c < 16) {
3592 /* This stays as CURLYX, we can put the count/of pair. */
3593 /* Find WHILEM (as in regexec.c) */
3594 regnode *nxt = oscan + NEXT_OFF(oscan);
3596 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3598 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3599 | (RExC_whilem_seen << 4)); /* On WHILEM */
3601 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3603 if (flags & SCF_DO_SUBSTR) {
3604 SV *last_str = NULL;
3605 int counted = mincount != 0;
3607 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3608 #if defined(SPARC64_GCC_WORKAROUND)
3611 const char *s = NULL;
3614 if (pos_before >= data->last_start_min)
3617 b = data->last_start_min;
3620 s = SvPV_const(data->last_found, l);
3621 old = b - data->last_start_min;
3624 I32 b = pos_before >= data->last_start_min
3625 ? pos_before : data->last_start_min;
3627 const char * const s = SvPV_const(data->last_found, l);
3628 I32 old = b - data->last_start_min;
3632 old = utf8_hop((U8*)s, old) - (U8*)s;
3634 /* Get the added string: */
3635 last_str = newSVpvn_utf8(s + old, l, UTF);
3636 if (deltanext == 0 && pos_before == b) {
3637 /* What was added is a constant string */
3639 SvGROW(last_str, (mincount * l) + 1);
3640 repeatcpy(SvPVX(last_str) + l,
3641 SvPVX_const(last_str), l, mincount - 1);
3642 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3643 /* Add additional parts. */
3644 SvCUR_set(data->last_found,
3645 SvCUR(data->last_found) - l);
3646 sv_catsv(data->last_found, last_str);
3648 SV * sv = data->last_found;
3650 SvUTF8(sv) && SvMAGICAL(sv) ?
3651 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3652 if (mg && mg->mg_len >= 0)
3653 mg->mg_len += CHR_SVLEN(last_str) - l;
3655 data->last_end += l * (mincount - 1);
3658 /* start offset must point into the last copy */
3659 data->last_start_min += minnext * (mincount - 1);
3660 data->last_start_max += is_inf ? I32_MAX
3661 : (maxcount - 1) * (minnext + data->pos_delta);
3664 /* It is counted once already... */
3665 data->pos_min += minnext * (mincount - counted);
3666 data->pos_delta += - counted * deltanext +
3667 (minnext + deltanext) * maxcount - minnext * mincount;
3668 if (mincount != maxcount) {
3669 /* Cannot extend fixed substrings found inside
3671 SCAN_COMMIT(pRExC_state,data,minlenp);
3672 if (mincount && last_str) {
3673 SV * const sv = data->last_found;
3674 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3675 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3679 sv_setsv(sv, last_str);
3680 data->last_end = data->pos_min;
3681 data->last_start_min =
3682 data->pos_min - CHR_SVLEN(last_str);
3683 data->last_start_max = is_inf
3685 : data->pos_min + data->pos_delta
3686 - CHR_SVLEN(last_str);
3688 data->longest = &(data->longest_float);
3690 SvREFCNT_dec(last_str);
3692 if (data && (fl & SF_HAS_EVAL))
3693 data->flags |= SF_HAS_EVAL;
3694 optimize_curly_tail:
3695 if (OP(oscan) != CURLYX) {
3696 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3698 NEXT_OFF(oscan) += NEXT_OFF(next);
3701 default: /* REF, ANYOFV, and CLUMP only? */
3702 if (flags & SCF_DO_SUBSTR) {
3703 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3704 data->longest = &(data->longest_float);
3706 is_inf = is_inf_internal = 1;
3707 if (flags & SCF_DO_STCLASS_OR)
3708 cl_anything(pRExC_state, data->start_class);
3709 flags &= ~SCF_DO_STCLASS;
3713 else if (OP(scan) == LNBREAK) {
3714 if (flags & SCF_DO_STCLASS) {
3716 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3717 if (flags & SCF_DO_STCLASS_AND) {
3718 for (value = 0; value < 256; value++)
3719 if (!is_VERTWS_cp(value))
3720 ANYOF_BITMAP_CLEAR(data->start_class, value);
3723 for (value = 0; value < 256; value++)
3724 if (is_VERTWS_cp(value))
3725 ANYOF_BITMAP_SET(data->start_class, value);
3727 if (flags & SCF_DO_STCLASS_OR)
3728 cl_and(data->start_class, and_withp);
3729 flags &= ~SCF_DO_STCLASS;
3733 if (flags & SCF_DO_SUBSTR) {
3734 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3736 data->pos_delta += 1;
3737 data->longest = &(data->longest_float);
3740 else if (OP(scan) == FOLDCHAR) {
3741 int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2;
3742 flags &= ~SCF_DO_STCLASS;
3745 if (flags & SCF_DO_SUBSTR) {
3746 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3748 data->pos_delta += d;
3749 data->longest = &(data->longest_float);
3752 else if (REGNODE_SIMPLE(OP(scan))) {
3755 if (flags & SCF_DO_SUBSTR) {
3756 SCAN_COMMIT(pRExC_state,data,minlenp);
3760 if (flags & SCF_DO_STCLASS) {
3761 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3763 /* Some of the logic below assumes that switching
3764 locale on will only add false positives. */
3765 switch (PL_regkind[OP(scan)]) {
3769 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3770 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3771 cl_anything(pRExC_state, data->start_class);
3774 if (OP(scan) == SANY)
3776 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3777 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3778 || ANYOF_CLASS_TEST_ANY_SET(data->start_class));
3779 cl_anything(pRExC_state, data->start_class);
3781 if (flags & SCF_DO_STCLASS_AND || !value)
3782 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3785 if (flags & SCF_DO_STCLASS_AND)
3786 cl_and(data->start_class,
3787 (struct regnode_charclass_class*)scan);
3789 cl_or(pRExC_state, data->start_class,
3790 (struct regnode_charclass_class*)scan);
3793 if (flags & SCF_DO_STCLASS_AND) {
3794 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3795 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3796 if (OP(scan) == ALNUMU) {
3797 for (value = 0; value < 256; value++) {
3798 if (!isWORDCHAR_L1(value)) {
3799 ANYOF_BITMAP_CLEAR(data->start_class, value);
3803 for (value = 0; value < 256; value++) {
3804 if (!isALNUM(value)) {
3805 ANYOF_BITMAP_CLEAR(data->start_class, value);
3812 if (data->start_class->flags & ANYOF_LOCALE)
3813 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3815 /* Even if under locale, set the bits for non-locale
3816 * in case it isn't a true locale-node. This will
3817 * create false positives if it truly is locale */
3818 if (OP(scan) == ALNUMU) {
3819 for (value = 0; value < 256; value++) {
3820 if (isWORDCHAR_L1(value)) {
3821 ANYOF_BITMAP_SET(data->start_class, value);
3825 for (value = 0; value < 256; value++) {
3826 if (isALNUM(value)) {
3827 ANYOF_BITMAP_SET(data->start_class, value);
3834 if (flags & SCF_DO_STCLASS_AND) {
3835 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3836 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3837 if (OP(scan) == NALNUMU) {
3838 for (value = 0; value < 256; value++) {
3839 if (isWORDCHAR_L1(value)) {
3840 ANYOF_BITMAP_CLEAR(data->start_class, value);
3844 for (value = 0; value < 256; value++) {
3845 if (isALNUM(value)) {
3846 ANYOF_BITMAP_CLEAR(data->start_class, value);
3853 if (data->start_class->flags & ANYOF_LOCALE)
3854 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3856 /* Even if under locale, set the bits for non-locale in
3857 * case it isn't a true locale-node. This will create
3858 * false positives if it truly is locale */
3859 if (OP(scan) == NALNUMU) {
3860 for (value = 0; value < 256; value++) {
3861 if (! isWORDCHAR_L1(value)) {
3862 ANYOF_BITMAP_SET(data->start_class, value);
3866 for (value = 0; value < 256; value++) {
3867 if (! isALNUM(value)) {
3868 ANYOF_BITMAP_SET(data->start_class, value);
3875 if (flags & SCF_DO_STCLASS_AND) {
3876 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3877 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3878 if (OP(scan) == SPACEU) {
3879 for (value = 0; value < 256; value++) {
3880 if (!isSPACE_L1(value)) {
3881 ANYOF_BITMAP_CLEAR(data->start_class, value);
3885 for (value = 0; value < 256; value++) {
3886 if (!isSPACE(value)) {
3887 ANYOF_BITMAP_CLEAR(data->start_class, value);
3894 if (data->start_class->flags & ANYOF_LOCALE) {
3895 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3897 if (OP(scan) == SPACEU) {
3898 for (value = 0; value < 256; value++) {
3899 if (isSPACE_L1(value)) {
3900 ANYOF_BITMAP_SET(data->start_class, value);
3904 for (value = 0; value < 256; value++) {
3905 if (isSPACE(value)) {
3906 ANYOF_BITMAP_SET(data->start_class, value);
3913 if (flags & SCF_DO_STCLASS_AND) {
3914 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3915 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3916 if (OP(scan) == NSPACEU) {
3917 for (value = 0; value < 256; value++) {
3918 if (isSPACE_L1(value)) {
3919 ANYOF_BITMAP_CLEAR(data->start_class, value);
3923 for (value = 0; value < 256; value++) {
3924 if (isSPACE(value)) {
3925 ANYOF_BITMAP_CLEAR(data->start_class, value);
3932 if (data->start_class->flags & ANYOF_LOCALE)
3933 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3934 if (OP(scan) == NSPACEU) {
3935 for (value = 0; value < 256; value++) {
3936 if (!isSPACE_L1(value)) {
3937 ANYOF_BITMAP_SET(data->start_class, value);
3942 for (value = 0; value < 256; value++) {
3943 if (!isSPACE(value)) {
3944 ANYOF_BITMAP_SET(data->start_class, value);
3951 if (flags & SCF_DO_STCLASS_AND) {
3952 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3953 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3954 for (value = 0; value < 256; value++)
3955 if (!isDIGIT(value))
3956 ANYOF_BITMAP_CLEAR(data->start_class, value);
3960 if (data->start_class->flags & ANYOF_LOCALE)
3961 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3962 for (value = 0; value < 256; value++)
3964 ANYOF_BITMAP_SET(data->start_class, value);
3968 if (flags & SCF_DO_STCLASS_AND) {
3969 if (!(data->start_class->flags & ANYOF_LOCALE))
3970 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3971 for (value = 0; value < 256; value++)
3973 ANYOF_BITMAP_CLEAR(data->start_class, value);
3976 if (data->start_class->flags & ANYOF_LOCALE)
3977 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3978 for (value = 0; value < 256; value++)
3979 if (!isDIGIT(value))
3980 ANYOF_BITMAP_SET(data->start_class, value);
3983 CASE_SYNST_FNC(VERTWS);
3984 CASE_SYNST_FNC(HORIZWS);
3987 if (flags & SCF_DO_STCLASS_OR)
3988 cl_and(data->start_class, and_withp);
3989 flags &= ~SCF_DO_STCLASS;
3992 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3993 data->flags |= (OP(scan) == MEOL
3997 else if ( PL_regkind[OP(scan)] == BRANCHJ
3998 /* Lookbehind, or need to calculate parens/evals/stclass: */
3999 && (scan->flags || data || (flags & SCF_DO_STCLASS))
4000 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
4001 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4002 || OP(scan) == UNLESSM )
4004 /* Negative Lookahead/lookbehind
4005 In this case we can't do fixed string optimisation.
4008 I32 deltanext, minnext, fake = 0;
4010 struct regnode_charclass_class intrnl;
4013 data_fake.flags = 0;
4015 data_fake.whilem_c = data->whilem_c;
4016 data_fake.last_closep = data->last_closep;
4019 data_fake.last_closep = &fake;
4020 data_fake.pos_delta = delta;
4021 if ( flags & SCF_DO_STCLASS && !scan->flags
4022 && OP(scan) == IFMATCH ) { /* Lookahead */
4023 cl_init(pRExC_state, &intrnl);
4024 data_fake.start_class = &intrnl;
4025 f |= SCF_DO_STCLASS_AND;
4027 if (flags & SCF_WHILEM_VISITED_POS)
4028 f |= SCF_WHILEM_VISITED_POS;
4029 next = regnext(scan);
4030 nscan = NEXTOPER(NEXTOPER(scan));
4031 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
4032 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
4035 FAIL("Variable length lookbehind not implemented");
4037 else if (minnext > (I32)U8_MAX) {
4038 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4040 scan->flags = (U8)minnext;
4043 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4045 if (data_fake.flags & SF_HAS_EVAL)
4046 data->flags |= SF_HAS_EVAL;
4047 data->whilem_c = data_fake.whilem_c;
4049 if (f & SCF_DO_STCLASS_AND) {
4050 if (flags & SCF_DO_STCLASS_OR) {
4051 /* OR before, AND after: ideally we would recurse with
4052 * data_fake to get the AND applied by study of the
4053 * remainder of the pattern, and then derecurse;
4054 * *** HACK *** for now just treat as "no information".
4055 * See [perl #56690].
4057 cl_init(pRExC_state, data->start_class);
4059 /* AND before and after: combine and continue */
4060 const int was = (data->start_class->flags & ANYOF_EOS);
4062 cl_and(data->start_class, &intrnl);
4064 data->start_class->flags |= ANYOF_EOS;
4068 #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
4070 /* Positive Lookahead/lookbehind
4071 In this case we can do fixed string optimisation,
4072 but we must be careful about it. Note in the case of
4073 lookbehind the positions will be offset by the minimum
4074 length of the pattern, something we won't know about
4075 until after the recurse.
4077 I32 deltanext, fake = 0;
4079 struct regnode_charclass_class intrnl;
4081 /* We use SAVEFREEPV so that when the full compile
4082 is finished perl will clean up the allocated
4083 minlens when it's all done. This way we don't
4084 have to worry about freeing them when we know
4085 they wont be used, which would be a pain.
4088 Newx( minnextp, 1, I32 );
4089 SAVEFREEPV(minnextp);
4092 StructCopy(data, &data_fake, scan_data_t);
4093 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
4096 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
4097 data_fake.last_found=newSVsv(data->last_found);
4101 data_fake.last_closep = &fake;
4102 data_fake.flags = 0;
4103 data_fake.pos_delta = delta;
4105 data_fake.flags |= SF_IS_INF;
4106 if ( flags & SCF_DO_STCLASS && !scan->flags
4107 && OP(scan) == IFMATCH ) { /* Lookahead */
4108 cl_init(pRExC_state, &intrnl);
4109 data_fake.start_class = &intrnl;
4110 f |= SCF_DO_STCLASS_AND;
4112 if (flags & SCF_WHILEM_VISITED_POS)
4113 f |= SCF_WHILEM_VISITED_POS;
4114 next = regnext(scan);
4115 nscan = NEXTOPER(NEXTOPER(scan));
4117 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
4118 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
4121 FAIL("Variable length lookbehind not implemented");
4123 else if (*minnextp > (I32)U8_MAX) {
4124 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
4126 scan->flags = (U8)*minnextp;
4131 if (f & SCF_DO_STCLASS_AND) {
4132 const int was = (data->start_class->flags & ANYOF_EOS);
4134 cl_and(data->start_class, &intrnl);
4136 data->start_class->flags |= ANYOF_EOS;
4139 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4141 if (data_fake.flags & SF_HAS_EVAL)
4142 data->flags |= SF_HAS_EVAL;
4143 data->whilem_c = data_fake.whilem_c;
4144 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
4145 if (RExC_rx->minlen<*minnextp)
4146 RExC_rx->minlen=*minnextp;
4147 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
4148 SvREFCNT_dec(data_fake.last_found);
4150 if ( data_fake.minlen_fixed != minlenp )
4152 data->offset_fixed= data_fake.offset_fixed;
4153 data->minlen_fixed= data_fake.minlen_fixed;
4154 data->lookbehind_fixed+= scan->flags;
4156 if ( data_fake.minlen_float != minlenp )
4158 data->minlen_float= data_fake.minlen_float;
4159 data->offset_float_min=data_fake.offset_float_min;
4160 data->offset_float_max=data_fake.offset_float_max;
4161 data->lookbehind_float+= scan->flags;
4170 else if (OP(scan) == OPEN) {
4171 if (stopparen != (I32)ARG(scan))
4174 else if (OP(scan) == CLOSE) {
4175 if (stopparen == (I32)ARG(scan)) {
4178 if ((I32)ARG(scan) == is_par) {
4179 next = regnext(scan);
4181 if ( next && (OP(next) != WHILEM) && next < last)
4182 is_par = 0; /* Disable optimization */
4185 *(data->last_closep) = ARG(scan);
4187 else if (OP(scan) == EVAL) {
4189 data->flags |= SF_HAS_EVAL;
4191 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
4192 if (flags & SCF_DO_SUBSTR) {
4193 SCAN_COMMIT(pRExC_state,data,minlenp);
4194 flags &= ~SCF_DO_SUBSTR;
4196 if (data && OP(scan)==ACCEPT) {
4197 data->flags |= SCF_SEEN_ACCEPT;
4202 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
4204 if (flags & SCF_DO_SUBSTR) {
4205 SCAN_COMMIT(pRExC_state,data,minlenp);
4206 data->longest = &(data->longest_float);
4208 is_inf = is_inf_internal = 1;
4209 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
4210 cl_anything(pRExC_state, data->start_class);
4211 flags &= ~SCF_DO_STCLASS;
4213 else if (OP(scan) == GPOS) {
4214 if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
4215 !(delta || is_inf || (data && data->pos_delta)))
4217 if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
4218 RExC_rx->extflags |= RXf_ANCH_GPOS;
4219 if (RExC_rx->gofs < (U32)min)
4220 RExC_rx->gofs = min;
4222 RExC_rx->extflags |= RXf_GPOS_FLOAT;
4226 #ifdef TRIE_STUDY_OPT
4227 #ifdef FULL_TRIE_STUDY
4228 else if (PL_regkind[OP(scan)] == TRIE) {
4229 /* NOTE - There is similar code to this block above for handling
4230 BRANCH nodes on the initial study. If you change stuff here
4232 regnode *trie_node= scan;
4233 regnode *tail= regnext(scan);
4234 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4235 I32 max1 = 0, min1 = I32_MAX;
4236 struct regnode_charclass_class accum;
4238 if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
4239 SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
4240 if (flags & SCF_DO_STCLASS)
4241 cl_init_zero(pRExC_state, &accum);
4247 const regnode *nextbranch= NULL;
4250 for ( word=1 ; word <= trie->wordcount ; word++)
4252 I32 deltanext=0, minnext=0, f = 0, fake;
4253 struct regnode_charclass_class this_class;
4255 data_fake.flags = 0;
4257 data_fake.whilem_c = data->whilem_c;
4258 data_fake.last_closep = data->last_closep;
4261 data_fake.last_closep = &fake;
4262 data_fake.pos_delta = delta;
4263 if (flags & SCF_DO_STCLASS) {
4264 cl_init(pRExC_state, &this_class);
4265 data_fake.start_class = &this_class;
4266 f = SCF_DO_STCLASS_AND;
4268 if (flags & SCF_WHILEM_VISITED_POS)
4269 f |= SCF_WHILEM_VISITED_POS;
4271 if (trie->jump[word]) {
4273 nextbranch = trie_node + trie->jump[0];
4274 scan= trie_node + trie->jump[word];
4275 /* We go from the jump point to the branch that follows
4276 it. Note this means we need the vestigal unused branches
4277 even though they arent otherwise used.
4279 minnext = study_chunk(pRExC_state, &scan, minlenp,
4280 &deltanext, (regnode *)nextbranch, &data_fake,
4281 stopparen, recursed, NULL, f,depth+1);
4283 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
4284 nextbranch= regnext((regnode*)nextbranch);
4286 if (min1 > (I32)(minnext + trie->minlen))
4287 min1 = minnext + trie->minlen;
4288 if (max1 < (I32)(minnext + deltanext + trie->maxlen))
4289 max1 = minnext + deltanext + trie->maxlen;
4290 if (deltanext == I32_MAX)
4291 is_inf = is_inf_internal = 1;
4293 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
4295 if (data_fake.flags & SCF_SEEN_ACCEPT) {
4296 if ( stopmin > min + min1)
4297 stopmin = min + min1;
4298 flags &= ~SCF_DO_SUBSTR;
4300 data->flags |= SCF_SEEN_ACCEPT;
4303 if (data_fake.flags & SF_HAS_EVAL)
4304 data->flags |= SF_HAS_EVAL;
4305 data->whilem_c = data_fake.whilem_c;
4307 if (flags & SCF_DO_STCLASS)
4308 cl_or(pRExC_state, &accum, &this_class);
4311 if (flags & SCF_DO_SUBSTR) {
4312 data->pos_min += min1;
4313 data->pos_delta += max1 - min1;
4314 if (max1 != min1 || is_inf)
4315 data->longest = &(data->longest_float);
4318 delta += max1 - min1;
4319 if (flags & SCF_DO_STCLASS_OR) {
4320 cl_or(pRExC_state, data->start_class, &accum);
4322 cl_and(data->start_class, and_withp);
4323 flags &= ~SCF_DO_STCLASS;
4326 else if (flags & SCF_DO_STCLASS_AND) {
4328 cl_and(data->start_class, &accum);
4329 flags &= ~SCF_DO_STCLASS;
4332 /* Switch to OR mode: cache the old value of
4333 * data->start_class */
4335 StructCopy(data->start_class, and_withp,
4336 struct regnode_charclass_class);
4337 flags &= ~SCF_DO_STCLASS_AND;
4338 StructCopy(&accum, data->start_class,
4339 struct regnode_charclass_class);
4340 flags |= SCF_DO_STCLASS_OR;
4341 data->start_class->flags |= ANYOF_EOS;
4348 else if (PL_regkind[OP(scan)] == TRIE) {
4349 reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
4352 min += trie->minlen;
4353 delta += (trie->maxlen - trie->minlen);
4354 flags &= ~SCF_DO_STCLASS; /* xxx */
4355 if (flags & SCF_DO_SUBSTR) {
4356 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
4357 data->pos_min += trie->minlen;
4358 data->pos_delta += (trie->maxlen - trie->minlen);
4359 if (trie->maxlen != trie->minlen)
4360 data->longest = &(data->longest_float);
4362 if (trie->jump) /* no more substrings -- for now /grr*/
4363 flags &= ~SCF_DO_SUBSTR;
4365 #endif /* old or new */
4366 #endif /* TRIE_STUDY_OPT */
4368 /* Else: zero-length, ignore. */
4369 scan = regnext(scan);
4374 stopparen = frame->stop;
4375 frame = frame->prev;
4376 goto fake_study_recurse;
4381 DEBUG_STUDYDATA("pre-fin:",data,depth);
4384 *deltap = is_inf_internal ? I32_MAX : delta;
4385 if (flags & SCF_DO_SUBSTR && is_inf)
4386 data->pos_delta = I32_MAX - data->pos_min;
4387 if (is_par > (I32)U8_MAX)
4389 if (is_par && pars==1 && data) {
4390 data->flags |= SF_IN_PAR;
4391 data->flags &= ~SF_HAS_PAR;
4393 else if (pars && data) {
4394 data->flags |= SF_HAS_PAR;
4395 data->flags &= ~SF_IN_PAR;
4397 if (flags & SCF_DO_STCLASS_OR)
4398 cl_and(data->start_class, and_withp);
4399 if (flags & SCF_TRIE_RESTUDY)
4400 data->flags |= SCF_TRIE_RESTUDY;
4402 DEBUG_STUDYDATA("post-fin:",data,depth);
4404 return min < stopmin ? min : stopmin;
4408 S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s)
4410 U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
4412 PERL_ARGS_ASSERT_ADD_DATA;
4414 Renewc(RExC_rxi->data,
4415 sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
4416 char, struct reg_data);
4418 Renew(RExC_rxi->data->what, count + n, U8);
4420 Newx(RExC_rxi->data->what, n, U8);
4421 RExC_rxi->data->count = count + n;
4422 Copy(s, RExC_rxi->data->what + count, n, U8);
4426 /*XXX: todo make this not included in a non debugging perl */
4427 #ifndef PERL_IN_XSUB_RE
4429 Perl_reginitcolors(pTHX)
4432 const char * const s = PerlEnv_getenv("PERL_RE_COLORS");
4434 char *t = savepv(s);
4438 t = strchr(t, '\t');
4444 PL_colors[i] = t = (char *)"";
4449 PL_colors[i++] = (char *)"";
4456 #ifdef TRIE_STUDY_OPT
4457 #define CHECK_RESTUDY_GOTO \
4459 (data.flags & SCF_TRIE_RESTUDY) \
4463 #define CHECK_RESTUDY_GOTO
4467 - pregcomp - compile a regular expression into internal code
4469 * We can't allocate space until we know how big the compiled form will be,
4470 * but we can't compile it (and thus know how big it is) until we've got a
4471 * place to put the code. So we cheat: we compile it twice, once with code
4472 * generation turned off and size counting turned on, and once "for real".
4473 * This also means that we don't allocate space until we are sure that the
4474 * thing really will compile successfully, and we never have to move the
4475 * code and thus invalidate pointers into it. (Note that it has to be in
4476 * one piece because free() must be able to free it all.) [NB: not true in perl]
4478 * Beware that the optimization-preparation code in here knows about some
4479 * of the structure of the compiled regexp. [I'll say.]
4484 #ifndef PERL_IN_XSUB_RE
4485 #define RE_ENGINE_PTR &PL_core_reg_engine
4487 extern const struct regexp_engine my_reg_engine;
4488 #define RE_ENGINE_PTR &my_reg_engine
4491 #ifndef PERL_IN_XSUB_RE
4493 Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags)
4496 HV * const table = GvHV(PL_hintgv);
4498 PERL_ARGS_ASSERT_PREGCOMP;
4500 /* Dispatch a request to compile a regexp to correct
4503 SV **ptr= hv_fetchs(table, "regcomp", FALSE);
4504 GET_RE_DEBUG_FLAGS_DECL;
4505 if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
4506 const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
4508 PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
4511 return CALLREGCOMP_ENG(eng, pattern, flags);
4514 return Perl_re_compile(aTHX_ pattern, flags);
4519 Perl_re_compile(pTHX_ SV * const pattern, U32 orig_pm_flags)
4524 register regexp_internal *ri;
4533 /* these are all flags - maybe they should be turned
4534 * into a single int with different bit masks */
4535 I32 sawlookahead = 0;
4538 bool used_setjump = FALSE;
4539 regex_charset initial_charset = get_regex_charset(orig_pm_flags);
4544 RExC_state_t RExC_state;
4545 RExC_state_t * const pRExC_state = &RExC_state;
4546 #ifdef TRIE_STUDY_OPT
4548 RExC_state_t copyRExC_state;
4550 GET_RE_DEBUG_FLAGS_DECL;
4552 PERL_ARGS_ASSERT_RE_COMPILE;
4554 DEBUG_r(if (!PL_colorset) reginitcolors());
4556 RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern);
4557 RExC_uni_semantics = 0;
4558 RExC_contains_locale = 0;
4560 /****************** LONG JUMP TARGET HERE***********************/
4561 /* Longjmp back to here if have to switch in midstream to utf8 */
4562 if (! RExC_orig_utf8) {
4563 JMPENV_PUSH(jump_ret);
4564 used_setjump = TRUE;
4567 if (jump_ret == 0) { /* First time through */
4568 exp = SvPV(pattern, plen);
4570 /* ignore the utf8ness if the pattern is 0 length */
4572 RExC_utf8 = RExC_orig_utf8 = 0;
4576 SV *dsv= sv_newmortal();
4577 RE_PV_QUOTED_DECL(s, RExC_utf8,
4578 dsv, exp, plen, 60);
4579 PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
4580 PL_colors[4],PL_colors[5],s);
4583 else { /* longjumped back */
4586 /* If the cause for the longjmp was other than changing to utf8, pop
4587 * our own setjmp, and longjmp to the correct handler */
4588 if (jump_ret != UTF8_LONGJMP) {
4590 JMPENV_JUMP(jump_ret);
4595 /* It's possible to write a regexp in ascii that represents Unicode
4596 codepoints outside of the byte range, such as via \x{100}. If we
4597 detect such a sequence we have to convert the entire pattern to utf8
4598 and then recompile, as our sizing calculation will have been based
4599 on 1 byte == 1 character, but we will need to use utf8 to encode
4600 at least some part of the pattern, and therefore must convert the whole
4603 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log,
4604 "UTF8 mismatch! Converting to utf8 for resizing and compile\n"));
4605 exp = (char*)Perl_bytes_to_utf8(aTHX_ (U8*)SvPV(pattern, plen), &len);
4607 RExC_orig_utf8 = RExC_utf8 = 1;
4611 #ifdef TRIE_STUDY_OPT
4615 pm_flags = orig_pm_flags;
4617 if (initial_charset == REGEX_LOCALE_CHARSET) {
4618 RExC_contains_locale = 1;
4620 else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) {
4622 /* Set to use unicode semantics if the pattern is in utf8 and has the
4623 * 'depends' charset specified, as it means unicode when utf8 */
4624 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4628 RExC_flags = pm_flags;
4632 RExC_in_lookbehind = 0;
4633 RExC_seen_zerolen = *exp == '^' ? -1 : 0;
4634 RExC_seen_evals = 0;
4636 RExC_override_recoding = 0;
4638 /* First pass: determine size, legality. */
4646 RExC_emit = &PL_regdummy;
4647 RExC_whilem_seen = 0;
4648 RExC_open_parens = NULL;
4649 RExC_close_parens = NULL;
4651 RExC_paren_names = NULL;
4653 RExC_paren_name_list = NULL;
4655 RExC_recurse = NULL;
4656 RExC_recurse_count = 0;
4658 #if 0 /* REGC() is (currently) a NOP at the first pass.
4659 * Clever compilers notice this and complain. --jhi */
4660 REGC((U8)REG_MAGIC, (char*)RExC_emit);
4662 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"));
4663 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4664 RExC_precomp = NULL;
4668 /* Here, finished first pass. Get rid of any added setjmp */
4674 PerlIO_printf(Perl_debug_log,
4675 "Required size %"IVdf" nodes\n"
4676 "Starting second pass (creation)\n",
4679 RExC_lastparse=NULL;
4682 /* The first pass could have found things that force Unicode semantics */
4683 if ((RExC_utf8 || RExC_uni_semantics)
4684 && get_regex_charset(pm_flags) == REGEX_DEPENDS_CHARSET)
4686 set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET);
4689 /* Small enough for pointer-storage convention?
4690 If extralen==0, this means that we will not need long jumps. */
4691 if (RExC_size >= 0x10000L && RExC_extralen)
4692 RExC_size += RExC_extralen;
4695 if (RExC_whilem_seen > 15)
4696 RExC_whilem_seen = 15;
4698 /* Allocate space and zero-initialize. Note, the two step process
4699 of zeroing when in debug mode, thus anything assigned has to
4700 happen after that */
4701 rx = (REGEXP*) newSV_type(SVt_REGEXP);
4702 r = (struct regexp*)SvANY(rx);
4703 Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
4704 char, regexp_internal);
4705 if ( r == NULL || ri == NULL )
4706 FAIL("Regexp out of space");
4708 /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */
4709 Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
4711 /* bulk initialize base fields with 0. */
4712 Zero(ri, sizeof(regexp_internal), char);
4715 /* non-zero initialization begins here */
4717 r->engine= RE_ENGINE_PTR;
4718 r->extflags = pm_flags;
4720 bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY);
4721 bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET);
4723 /* The caret is output if there are any defaults: if not all the STD
4724 * flags are set, or if no character set specifier is needed */
4726 (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD)
4728 bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT);
4729 U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD)
4730 >> RXf_PMf_STD_PMMOD_SHIFT);
4731 const char *fptr = STD_PAT_MODS; /*"msix"*/
4733 /* Allocate for the worst case, which is all the std flags are turned
4734 * on. If more precision is desired, we could do a population count of
4735 * the flags set. This could be done with a small lookup table, or by
4736 * shifting, masking and adding, or even, when available, assembly
4737 * language for a machine-language population count.
4738 * We never output a minus, as all those are defaults, so are
4739 * covered by the caret */
4740 const STRLEN wraplen = plen + has_p + has_runon
4741 + has_default /* If needs a caret */
4743 /* If needs a character set specifier */
4744 + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0)
4745 + (sizeof(STD_PAT_MODS) - 1)
4746 + (sizeof("(?:)") - 1);
4748 p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */
4750 SvFLAGS(rx) |= SvUTF8(pattern);
4753 /* If a default, cover it using the caret */
4755 *p++= DEFAULT_PAT_MOD;
4759 const char* const name = get_regex_charset_name(r->extflags, &len);
4760 Copy(name, p, len, char);
4764 *p++ = KEEPCOPY_PAT_MOD; /*'p'*/
4767 while((ch = *fptr++)) {
4775 Copy(RExC_precomp, p, plen, char);
4776 assert ((RX_WRAPPED(rx) - p) < 16);
4777 r->pre_prefix = p - RX_WRAPPED(rx);
4783 SvCUR_set(rx, p - SvPVX_const(rx));
4787 r->nparens = RExC_npar - 1; /* set early to validate backrefs */
4789 if (RExC_seen & REG_SEEN_RECURSE) {
4790 Newxz(RExC_open_parens, RExC_npar,regnode *);
4791 SAVEFREEPV(RExC_open_parens);
4792 Newxz(RExC_close_parens,RExC_npar,regnode *);
4793 SAVEFREEPV(RExC_close_parens);
4796 /* Useful during FAIL. */
4797 #ifdef RE_TRACK_PATTERN_OFFSETS
4798 Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
4799 DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log,
4800 "%s %"UVuf" bytes for offset annotations.\n",
4801 ri->u.offsets ? "Got" : "Couldn't get",
4802 (UV)((2*RExC_size+1) * sizeof(U32))));
4804 SetProgLen(ri,RExC_size);
4809 /* Second pass: emit code. */
4810 RExC_flags = pm_flags; /* don't let top level (?i) bleed */
4815 RExC_emit_start = ri->program;
4816 RExC_emit = ri->program;
4817 RExC_emit_bound = ri->program + RExC_size + 1;
4819 /* Store the count of eval-groups for security checks: */
4820 RExC_rx->seen_evals = RExC_seen_evals;
4821 REGC((U8)REG_MAGIC, (char*) RExC_emit++);
4822 if (reg(pRExC_state, 0, &flags,1) == NULL) {
4826 /* XXXX To minimize changes to RE engine we always allocate
4827 3-units-long substrs field. */
4828 Newx(r->substrs, 1, struct reg_substr_data);
4829 if (RExC_recurse_count) {
4830 Newxz(RExC_recurse,RExC_recurse_count,regnode *);
4831 SAVEFREEPV(RExC_recurse);
4835 r->minlen = minlen = sawlookahead = sawplus = sawopen = 0;
4836 Zero(r->substrs, 1, struct reg_substr_data);
4838 #ifdef TRIE_STUDY_OPT
4840 StructCopy(&zero_scan_data, &data, scan_data_t);
4841 copyRExC_state = RExC_state;
4844 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
4846 RExC_state = copyRExC_state;
4847 if (seen & REG_TOP_LEVEL_BRANCHES)
4848 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
4850 RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
4851 if (data.last_found) {
4852 SvREFCNT_dec(data.longest_fixed);
4853 SvREFCNT_dec(data.longest_float);
4854 SvREFCNT_dec(data.last_found);
4856 StructCopy(&zero_scan_data, &data, scan_data_t);
4859 StructCopy(&zero_scan_data, &data, scan_data_t);
4862 /* Dig out information for optimizations. */
4863 r->extflags = RExC_flags; /* was pm_op */
4864 /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */
4867 SvUTF8_on(rx); /* Unicode in it? */
4868 ri->regstclass = NULL;
4869 if (RExC_naughty >= 10) /* Probably an expensive pattern. */
4870 r->intflags |= PREGf_NAUGHTY;
4871 scan = ri->program + 1; /* First BRANCH. */
4873 /* testing for BRANCH here tells us whether there is "must appear"
4874 data in the pattern. If there is then we can use it for optimisations */
4875 if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
4877 STRLEN longest_float_length, longest_fixed_length;
4878 struct regnode_charclass_class ch_class; /* pointed to by data */
4880 I32 last_close = 0; /* pointed to by data */
4881 regnode *first= scan;
4882 regnode *first_next= regnext(first);
4884 * Skip introductions and multiplicators >= 1
4885 * so that we can extract the 'meat' of the pattern that must
4886 * match in the large if() sequence following.
4887 * NOTE that EXACT is NOT covered here, as it is normally
4888 * picked up by the optimiser separately.
4890 * This is unfortunate as the optimiser isnt handling lookahead
4891 * properly currently.
4894 while ((OP(first) == OPEN && (sawopen = 1)) ||
4895 /* An OR of *one* alternative - should not happen now. */
4896 (OP(first) == BRANCH && OP(first_next) != BRANCH) ||
4897 /* for now we can't handle lookbehind IFMATCH*/
4898 (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) ||
4899 (OP(first) == PLUS) ||
4900 (OP(first) == MINMOD) ||
4901 /* An {n,m} with n>0 */
4902 (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) ||
4903 (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END ))
4906 * the only op that could be a regnode is PLUS, all the rest
4907 * will be regnode_1 or regnode_2.
4910 if (OP(first) == PLUS)
4913 first += regarglen[OP(first)];
4915 first = NEXTOPER(first);
4916 first_next= regnext(first);
4919 /* Starting-point info. */
4921 DEBUG_PEEP("first:",first,0);
4922 /* Ignore EXACT as we deal with it later. */
4923 if (PL_regkind[OP(first)] == EXACT) {
4924 if (OP(first) == EXACT)
4925 NOOP; /* Empty, get anchored substr later. */
4927 ri->regstclass = first;
4930 else if (PL_regkind[OP(first)] == TRIE &&
4931 ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
4934 /* this can happen only on restudy */
4935 if ( OP(first) == TRIE ) {
4936 struct regnode_1 *trieop = (struct regnode_1 *)
4937 PerlMemShared_calloc(1, sizeof(struct regnode_1));
4938 StructCopy(first,trieop,struct regnode_1);
4939 trie_op=(regnode *)trieop;
4941 struct regnode_charclass *trieop = (struct regnode_charclass *)
4942 PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
4943 StructCopy(first,trieop,struct regnode_charclass);
4944 trie_op=(regnode *)trieop;
4947 make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
4948 ri->regstclass = trie_op;
4951 else if (REGNODE_SIMPLE(OP(first)))
4952 ri->regstclass = first;
4953 else if (PL_regkind[OP(first)] == BOUND ||
4954 PL_regkind[OP(first)] == NBOUND)
4955 ri->regstclass = first;
4956 else if (PL_regkind[OP(first)] == BOL) {
4957 r->extflags |= (OP(first) == MBOL
4959 : (OP(first) == SBOL
4962 first = NEXTOPER(first);
4965 else if (OP(first) == GPOS) {
4966 r->extflags |= RXf_ANCH_GPOS;
4967 first = NEXTOPER(first);
4970 else if ((!sawopen || !RExC_sawback) &&
4971 (OP(first) == STAR &&
4972 PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
4973 !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
4975 /* turn .* into ^.* with an implied $*=1 */
4977 (OP(NEXTOPER(first)) == REG_ANY)
4980 r->extflags |= type;
4981 r->intflags |= PREGf_IMPLICIT;
4982 first = NEXTOPER(first);
4985 if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback)
4986 && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */
4987 /* x+ must match at the 1st pos of run of x's */
4988 r->intflags |= PREGf_SKIP;
4990 /* Scan is after the zeroth branch, first is atomic matcher. */
4991 #ifdef TRIE_STUDY_OPT
4994 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
4995 (IV)(first - scan + 1))
4999 PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
5000 (IV)(first - scan + 1))
5006 * If there's something expensive in the r.e., find the
5007 * longest literal string that must appear and make it the
5008 * regmust. Resolve ties in favor of later strings, since
5009 * the regstart check works with the beginning of the r.e.
5010 * and avoiding duplication strengthens checking. Not a
5011 * strong reason, but sufficient in the absence of others.
5012 * [Now we resolve ties in favor of the earlier string if
5013 * it happens that c_offset_min has been invalidated, since the
5014 * earlier string may buy us something the later one won't.]
5017 data.longest_fixed = newSVpvs("");
5018 data.longest_float = newSVpvs("");
5019 data.last_found = newSVpvs("");
5020 data.longest = &(data.longest_fixed);
5022 if (!ri->regstclass) {
5023 cl_init(pRExC_state, &ch_class);
5024 data.start_class = &ch_class;
5025 stclass_flag = SCF_DO_STCLASS_AND;
5026 } else /* XXXX Check for BOUND? */
5028 data.last_closep = &last_close;
5030 minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */
5031 &data, -1, NULL, NULL,
5032 SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
5038 if ( RExC_npar == 1 && data.longest == &(data.longest_fixed)
5039 && data.last_start_min == 0 && data.last_end > 0
5040 && !RExC_seen_zerolen
5041 && !(RExC_seen & REG_SEEN_VERBARG)
5042 && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
5043 r->extflags |= RXf_CHECK_ALL;
5044 scan_commit(pRExC_state, &data,&minlen,0);
5045 SvREFCNT_dec(data.last_found);
5047 /* Note that code very similar to this but for anchored string
5048 follows immediately below, changes may need to be made to both.
5051 longest_float_length = CHR_SVLEN(data.longest_float);
5052 if (longest_float_length
5053 || (data.flags & SF_FL_BEFORE_EOL
5054 && (!(data.flags & SF_FL_BEFORE_MEOL)
5055 || (RExC_flags & RXf_PMf_MULTILINE))))
5059 if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
5060 && data.offset_fixed == data.offset_float_min
5061 && SvCUR(data.longest_fixed) == SvCUR(data.longest_float))
5062 goto remove_float; /* As in (a)+. */
5064 /* copy the information about the longest float from the reg_scan_data
5065 over to the program. */
5066 if (SvUTF8(data.longest_float)) {
5067 r->float_utf8 = data.longest_float;
5068 r->float_substr = NULL;
5070 r->float_substr = data.longest_float;
5071 r->float_utf8 = NULL;
5073 /* float_end_shift is how many chars that must be matched that
5074 follow this item. We calculate it ahead of time as once the
5075 lookbehind offset is added in we lose the ability to correctly
5077 ml = data.minlen_float ? *(data.minlen_float)
5078 : (I32)longest_float_length;
5079 r->float_end_shift = ml - data.offset_float_min
5080 - longest_float_length + (SvTAIL(data.longest_float) != 0)
5081 + data.lookbehind_float;
5082 r->float_min_offset = data.offset_float_min - data.lookbehind_float;
5083 r->float_max_offset = data.offset_float_max;
5084 if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
5085 r->float_max_offset -= data.lookbehind_float;
5087 t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
5088 && (!(data.flags & SF_FL_BEFORE_MEOL)
5089 || (RExC_flags & RXf_PMf_MULTILINE)));
5090 fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
5094 r->float_substr = r->float_utf8 = NULL;
5095 SvREFCNT_dec(data.longest_float);
5096 longest_float_length = 0;
5099 /* Note that code very similar to this but for floating string
5100 is immediately above, changes may need to be made to both.
5103 longest_fixed_length = CHR_SVLEN(data.longest_fixed);
5104 if (longest_fixed_length
5105 || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */
5106 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5107 || (RExC_flags & RXf_PMf_MULTILINE))))
5111 /* copy the information about the longest fixed
5112 from the reg_scan_data over to the program. */
5113 if (SvUTF8(data.longest_fixed)) {
5114 r->anchored_utf8 = data.longest_fixed;
5115 r->anchored_substr = NULL;
5117 r->anchored_substr = data.longest_fixed;
5118 r->anchored_utf8 = NULL;
5120 /* fixed_end_shift is how many chars that must be matched that
5121 follow this item. We calculate it ahead of time as once the
5122 lookbehind offset is added in we lose the ability to correctly
5124 ml = data.minlen_fixed ? *(data.minlen_fixed)
5125 : (I32)longest_fixed_length;
5126 r->anchored_end_shift = ml - data.offset_fixed
5127 - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
5128 + data.lookbehind_fixed;
5129 r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
5131 t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
5132 && (!(data.flags & SF_FIX_BEFORE_MEOL)
5133 || (RExC_flags & RXf_PMf_MULTILINE)));
5134 fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
5137 r->anchored_substr = r->anchored_utf8 = NULL;
5138 SvREFCNT_dec(data.longest_fixed);
5139 longest_fixed_length = 0;
5142 && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
5143 ri->regstclass = NULL;
5145 if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
5147 && !(data.start_class->flags & ANYOF_EOS)
5148 && !cl_is_anything(data.start_class))
5150 const U32 n = add_data(pRExC_state, 1, "f");
5151 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5153 Newx(RExC_rxi->data->data[n], 1,
5154 struct regnode_charclass_class);
5155 StructCopy(data.start_class,
5156 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5157 struct regnode_charclass_class);
5158 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5159 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5160 DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
5161 regprop(r, sv, (regnode*)data.start_class);
5162 PerlIO_printf(Perl_debug_log,
5163 "synthetic stclass \"%s\".\n",
5164 SvPVX_const(sv));});
5167 /* A temporary algorithm prefers floated substr to fixed one to dig more info. */
5168 if (longest_fixed_length > longest_float_length) {
5169 r->check_end_shift = r->anchored_end_shift;
5170 r->check_substr = r->anchored_substr;
5171 r->check_utf8 = r->anchored_utf8;
5172 r->check_offset_min = r->check_offset_max = r->anchored_offset;
5173 if (r->extflags & RXf_ANCH_SINGLE)
5174 r->extflags |= RXf_NOSCAN;
5177 r->check_end_shift = r->float_end_shift;
5178 r->check_substr = r->float_substr;
5179 r->check_utf8 = r->float_utf8;
5180 r->check_offset_min = r->float_min_offset;
5181 r->check_offset_max = r->float_max_offset;
5183 /* XXXX Currently intuiting is not compatible with ANCH_GPOS.
5184 This should be changed ASAP! */
5185 if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
5186 r->extflags |= RXf_USE_INTUIT;
5187 if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
5188 r->extflags |= RXf_INTUIT_TAIL;
5190 /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
5191 if ( (STRLEN)minlen < longest_float_length )
5192 minlen= longest_float_length;
5193 if ( (STRLEN)minlen < longest_fixed_length )
5194 minlen= longest_fixed_length;
5198 /* Several toplevels. Best we can is to set minlen. */
5200 struct regnode_charclass_class ch_class;
5203 DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
5205 scan = ri->program + 1;
5206 cl_init(pRExC_state, &ch_class);
5207 data.start_class = &ch_class;
5208 data.last_closep = &last_close;
5211 minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
5212 &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
5216 r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
5217 = r->float_substr = r->float_utf8 = NULL;
5219 if (!(data.start_class->flags & ANYOF_EOS)
5220 && !cl_is_anything(data.start_class))
5222 const U32 n = add_data(pRExC_state, 1, "f");
5223 data.start_class->flags |= ANYOF_IS_SYNTHETIC;
5225 Newx(RExC_rxi->data->data[n], 1,
5226 struct regnode_charclass_class);
5227 StructCopy(data.start_class,
5228 (struct regnode_charclass_class*)RExC_rxi->data->data[n],
5229 struct regnode_charclass_class);
5230 ri->regstclass = (regnode*)RExC_rxi->data->data[n];
5231 r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
5232 DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
5233 regprop(r, sv, (regnode*)data.start_class);
5234 PerlIO_printf(Perl_debug_log,
5235 "synthetic stclass \"%s\".\n",
5236 SvPVX_const(sv));});
5240 /* Guard against an embedded (?=) or (?<=) with a longer minlen than
5241 the "real" pattern. */
5243 PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
5244 (IV)minlen, (IV)r->minlen);
5246 r->minlenret = minlen;
5247 if (r->minlen < minlen)
5250 if (RExC_seen & REG_SEEN_GPOS)
5251 r->extflags |= RXf_GPOS_SEEN;
5252 if (RExC_seen & REG_SEEN_LOOKBEHIND)
5253 r->extflags |= RXf_LOOKBEHIND_SEEN;
5254 if (RExC_seen & REG_SEEN_EVAL)
5255 r->extflags |= RXf_EVAL_SEEN;
5256 if (RExC_seen & REG_SEEN_CANY)
5257 r->extflags |= RXf_CANY_SEEN;
5258 if (RExC_seen & REG_SEEN_VERBARG)
5259 r->intflags |= PREGf_VERBARG_SEEN;
5260 if (RExC_seen & REG_SEEN_CUTGROUP)
5261 r->intflags |= PREGf_CUTGROUP_SEEN;
5262 if (RExC_paren_names)
5263 RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names));
5265 RXp_PAREN_NAMES(r) = NULL;
5267 #ifdef STUPID_PATTERN_CHECKS
5268 if (RX_PRELEN(rx) == 0)
5269 r->extflags |= RXf_NULL;
5270 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5271 /* XXX: this should happen BEFORE we compile */
5272 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5273 else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3))
5274 r->extflags |= RXf_WHITE;
5275 else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^')
5276 r->extflags |= RXf_START_ONLY;
5278 if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ')
5279 /* XXX: this should happen BEFORE we compile */
5280 r->extflags |= (RXf_SKIPWHITE|RXf_WHITE);
5282 regnode *first = ri->program + 1;
5285 if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END)
5286 r->extflags |= RXf_NULL;
5287 else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END)
5288 r->extflags |= RXf_START_ONLY;
5289 else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE
5290 && OP(regnext(first)) == END)
5291 r->extflags |= RXf_WHITE;
5295 if (RExC_paren_names) {
5296 ri->name_list_idx = add_data( pRExC_state, 1, "a" );
5297 ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list);
5300 ri->name_list_idx = 0;
5302 if (RExC_recurse_count) {
5303 for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
5304 const regnode *scan = RExC_recurse[RExC_recurse_count-1];
5305 ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
5308 Newxz(r->offs, RExC_npar, regexp_paren_pair);
5309 /* assume we don't need to swap parens around before we match */
5312 PerlIO_printf(Perl_debug_log,"Final program:\n");
5315 #ifdef RE_TRACK_PATTERN_OFFSETS
5316 DEBUG_OFFSETS_r(if (ri->u.offsets) {
5317 const U32 len = ri->u.offsets[0];
5319 GET_RE_DEBUG_FLAGS_DECL;
5320 PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]);
5321 for (i = 1; i <= len; i++) {
5322 if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2])
5323 PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
5324 (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]);
5326 PerlIO_printf(Perl_debug_log, "\n");
5332 #undef RE_ENGINE_PTR
5336 Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value,
5339 PERL_ARGS_ASSERT_REG_NAMED_BUFF;
5341 PERL_UNUSED_ARG(value);
5343 if (flags & RXapif_FETCH) {
5344 return reg_named_buff_fetch(rx, key, flags);
5345 } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) {
5346 Perl_croak_no_modify(aTHX);
5348 } else if (flags & RXapif_EXISTS) {
5349 return reg_named_buff_exists(rx, key, flags)
5352 } else if (flags & RXapif_REGNAMES) {
5353 return reg_named_buff_all(rx, flags);
5354 } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) {
5355 return reg_named_buff_scalar(rx, flags);
5357 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags);
5363 Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey,
5366 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER;
5367 PERL_UNUSED_ARG(lastkey);
5369 if (flags & RXapif_FIRSTKEY)
5370 return reg_named_buff_firstkey(rx, flags);
5371 else if (flags & RXapif_NEXTKEY)
5372 return reg_named_buff_nextkey(rx, flags);
5374 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags);
5380 Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv,
5383 AV *retarray = NULL;
5385 struct regexp *const rx = (struct regexp *)SvANY(r);
5387 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH;
5389 if (flags & RXapif_ALL)
5392 if (rx && RXp_PAREN_NAMES(rx)) {
5393 HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 );
5396 SV* sv_dat=HeVAL(he_str);
5397 I32 *nums=(I32*)SvPVX(sv_dat);
5398 for ( i=0; i<SvIVX(sv_dat); i++ ) {
5399 if ((I32)(rx->nparens) >= nums[i]
5400 && rx->offs[nums[i]].start != -1
5401 && rx->offs[nums[i]].end != -1)
5404 CALLREG_NUMBUF_FETCH(r,nums[i],ret);
5408 ret = newSVsv(&PL_sv_undef);
5411 av_push(retarray, ret);
5414 return newRV_noinc(MUTABLE_SV(retarray));
5421 Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key,
5424 struct regexp *const rx = (struct regexp *)SvANY(r);
5426 PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS;
5428 if (rx && RXp_PAREN_NAMES(rx)) {
5429 if (flags & RXapif_ALL) {
5430 return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0);
5432 SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags);
5446 Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags)
5448 struct regexp *const rx = (struct regexp *)SvANY(r);
5450 PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY;
5452 if ( rx && RXp_PAREN_NAMES(rx) ) {
5453 (void)hv_iterinit(RXp_PAREN_NAMES(rx));
5455 return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY);
5462 Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags)
5464 struct regexp *const rx = (struct regexp *)SvANY(r);
5465 GET_RE_DEBUG_FLAGS_DECL;
5467 PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY;
5469 if (rx && RXp_PAREN_NAMES(rx)) {
5470 HV *hv = RXp_PAREN_NAMES(rx);
5472 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5475 SV* sv_dat = HeVAL(temphe);
5476 I32 *nums = (I32*)SvPVX(sv_dat);
5477 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5478 if ((I32)(rx->lastparen) >= nums[i] &&
5479 rx->offs[nums[i]].start != -1 &&
5480 rx->offs[nums[i]].end != -1)
5486 if (parno || flags & RXapif_ALL) {
5487 return newSVhek(HeKEY_hek(temphe));
5495 Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags)
5500 struct regexp *const rx = (struct regexp *)SvANY(r);
5502 PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR;
5504 if (rx && RXp_PAREN_NAMES(rx)) {
5505 if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) {
5506 return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx)));
5507 } else if (flags & RXapif_ONE) {
5508 ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES));
5509 av = MUTABLE_AV(SvRV(ret));
5510 length = av_len(av);
5512 return newSViv(length + 1);
5514 Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags);
5518 return &PL_sv_undef;
5522 Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags)
5524 struct regexp *const rx = (struct regexp *)SvANY(r);
5527 PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL;
5529 if (rx && RXp_PAREN_NAMES(rx)) {
5530 HV *hv= RXp_PAREN_NAMES(rx);
5532 (void)hv_iterinit(hv);
5533 while ( (temphe = hv_iternext_flags(hv,0)) ) {
5536 SV* sv_dat = HeVAL(temphe);
5537 I32 *nums = (I32*)SvPVX(sv_dat);
5538 for ( i = 0; i < SvIVX(sv_dat); i++ ) {
5539 if ((I32)(rx->lastparen) >= nums[i] &&
5540 rx->offs[nums[i]].start != -1 &&
5541 rx->offs[nums[i]].end != -1)
5547 if (parno || flags & RXapif_ALL) {
5548 av_push(av, newSVhek(HeKEY_hek(temphe)));
5553 return newRV_noinc(MUTABLE_SV(av));
5557 Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren,
5560 struct regexp *const rx = (struct regexp *)SvANY(r);
5565 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH;
5568 sv_setsv(sv,&PL_sv_undef);
5572 if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) {
5574 i = rx->offs[0].start;
5578 if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) {
5580 s = rx->subbeg + rx->offs[0].end;
5581 i = rx->sublen - rx->offs[0].end;
5584 if ( 0 <= paren && paren <= (I32)rx->nparens &&
5585 (s1 = rx->offs[paren].start) != -1 &&
5586 (t1 = rx->offs[paren].end) != -1)
5590 s = rx->subbeg + s1;
5592 sv_setsv(sv,&PL_sv_undef);
5595 assert(rx->sublen >= (s - rx->subbeg) + i );
5597 const int oldtainted = PL_tainted;
5599 sv_setpvn(sv, s, i);
5600 PL_tainted = oldtainted;
5601 if ( (rx->extflags & RXf_CANY_SEEN)
5602 ? (RXp_MATCH_UTF8(rx)
5603 && (!i || is_utf8_string((U8*)s, i)))
5604 : (RXp_MATCH_UTF8(rx)) )
5611 if (RXp_MATCH_TAINTED(rx)) {
5612 if (SvTYPE(sv) >= SVt_PVMG) {
5613 MAGIC* const mg = SvMAGIC(sv);
5616 SvMAGIC_set(sv, mg->mg_moremagic);
5618 if ((mgt = SvMAGIC(sv))) {
5619 mg->mg_moremagic = mgt;
5620 SvMAGIC_set(sv, mg);
5630 sv_setsv(sv,&PL_sv_undef);
5636 Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren,
5637 SV const * const value)
5639 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE;
5641 PERL_UNUSED_ARG(rx);
5642 PERL_UNUSED_ARG(paren);
5643 PERL_UNUSED_ARG(value);
5646 Perl_croak_no_modify(aTHX);
5650 Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv,
5653 struct regexp *const rx = (struct regexp *)SvANY(r);
5657 PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH;
5659 /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */
5661 /* $` / ${^PREMATCH} */
5662 case RX_BUFF_IDX_PREMATCH:
5663 if (rx->offs[0].start != -1) {
5664 i = rx->offs[0].start;
5672 /* $' / ${^POSTMATCH} */
5673 case RX_BUFF_IDX_POSTMATCH:
5674 if (rx->offs[0].end != -1) {
5675 i = rx->sublen - rx->offs[0].end;
5677 s1 = rx->offs[0].end;
5683 /* $& / ${^MATCH}, $1, $2, ... */
5685 if (paren <= (I32)rx->nparens &&
5686 (s1 = rx->offs[paren].start) != -1 &&
5687 (t1 = rx->offs[paren].end) != -1)
5692 if (ckWARN(WARN_UNINITIALIZED))
5693 report_uninit((const SV *)sv);
5698 if (i > 0 && RXp_MATCH_UTF8(rx)) {
5699 const char * const s = rx->subbeg + s1;
5704 if (is_utf8_string_loclen((U8*)s, i, &ep, &el))
5711 Perl_reg_qr_package(pTHX_ REGEXP * const rx)
5713 PERL_ARGS_ASSERT_REG_QR_PACKAGE;
5714 PERL_UNUSED_ARG(rx);
5718 return newSVpvs("Regexp");
5721 /* Scans the name of a named buffer from the pattern.
5722 * If flags is REG_RSN_RETURN_NULL returns null.
5723 * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
5724 * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
5725 * to the parsed name as looked up in the RExC_paren_names hash.
5726 * If there is an error throws a vFAIL().. type exception.
5729 #define REG_RSN_RETURN_NULL 0
5730 #define REG_RSN_RETURN_NAME 1
5731 #define REG_RSN_RETURN_DATA 2
5734 S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags)
5736 char *name_start = RExC_parse;
5738 PERL_ARGS_ASSERT_REG_SCAN_NAME;
5740 if (isIDFIRST_lazy_if(RExC_parse, UTF)) {
5741 /* skip IDFIRST by using do...while */
5744 RExC_parse += UTF8SKIP(RExC_parse);
5745 } while (isALNUM_utf8((U8*)RExC_parse));
5749 } while (isALNUM(*RExC_parse));
5754 = newSVpvn_flags(name_start, (int)(RExC_parse - name_start),
5755 SVs_TEMP | (UTF ? SVf_UTF8 : 0));
5756 if ( flags == REG_RSN_RETURN_NAME)
5758 else if (flags==REG_RSN_RETURN_DATA) {
5761 if ( ! sv_name ) /* should not happen*/
5762 Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
5763 if (RExC_paren_names)
5764 he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
5766 sv_dat = HeVAL(he_str);
5768 vFAIL("Reference to nonexistent named group");
5772 Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
5779 #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
5780 int rem=(int)(RExC_end - RExC_parse); \
5789 if (RExC_lastparse!=RExC_parse) \
5790 PerlIO_printf(Perl_debug_log," >%.*s%-*s", \
5793 iscut ? "..." : "<" \
5796 PerlIO_printf(Perl_debug_log,"%16s",""); \
5799 num = RExC_size + 1; \
5801 num=REG_NODE_NUM(RExC_emit); \
5802 if (RExC_lastnum!=num) \
5803 PerlIO_printf(Perl_debug_log,"|%4d",num); \
5805 PerlIO_printf(Perl_debug_log,"|%4s",""); \
5806 PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
5807 (int)((depth*2)), "", \
5811 RExC_lastparse=RExC_parse; \
5816 #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \
5817 DEBUG_PARSE_MSG((funcname)); \
5818 PerlIO_printf(Perl_debug_log,"%4s","\n"); \
5820 #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
5821 DEBUG_PARSE_MSG((funcname)); \
5822 PerlIO_printf(Perl_debug_log,fmt "\n",args); \
5825 /* This section of code defines the inversion list object and its methods. The
5826 * interfaces are highly subject to change, so as much as possible is static to
5827 * this file. An inversion list is here implemented as a malloc'd C array with
5828 * some added info. More will be coming when functionality is added later.
5830 * Some of the methods should always be private to the implementation, and some
5831 * should eventually be made public */
5833 #define INVLIST_INITIAL_LEN 10
5834 #define INVLIST_ARRAY_KEY "array"
5835 #define INVLIST_MAX_KEY "max"
5836 #define INVLIST_LEN_KEY "len"
5838 PERL_STATIC_INLINE UV*
5839 S_invlist_array(pTHX_ HV* const invlist)
5841 /* Returns the pointer to the inversion list's array. Every time the
5842 * length changes, this needs to be called in case malloc or realloc moved
5845 SV** list_ptr = hv_fetchs(invlist, INVLIST_ARRAY_KEY, FALSE);
5847 PERL_ARGS_ASSERT_INVLIST_ARRAY;
5849 if (list_ptr == NULL) {
5850 Perl_croak(aTHX_ "panic: inversion list without a '%s' element",
5854 return INT2PTR(UV *, SvUV(*list_ptr));
5857 PERL_STATIC_INLINE void
5858 S_invlist_set_array(pTHX_ HV* const invlist, const UV* const array)
5860 PERL_ARGS_ASSERT_INVLIST_SET_ARRAY;
5862 /* Sets the array stored in the inversion list to the memory beginning with
5865 if (hv_stores(invlist, INVLIST_ARRAY_KEY, newSVuv(PTR2UV(array))) == NULL) {
5866 Perl_croak(aTHX_ "panic: can't store '%s' entry in inversion list",
5871 PERL_STATIC_INLINE UV
5872 S_invlist_len(pTHX_ HV* const invlist)
5874 /* Returns the current number of elements in the inversion list's array */
5876 SV** len_ptr = hv_fetchs(invlist, INVLIST_LEN_KEY, FALSE);
5878 PERL_ARGS_ASSERT_INVLIST_LEN;
5880 if (len_ptr == NULL) {
5881 Perl_croak(aTHX_ "panic: inversion list without a '%s' element",
5885 return SvUV(*len_ptr);
5888 PERL_STATIC_INLINE UV
5889 S_invlist_max(pTHX_ HV* const invlist)
5891 /* Returns the maximum number of elements storable in the inversion list's
5892 * array, without having to realloc() */
5894 SV** max_ptr = hv_fetchs(invlist, INVLIST_MAX_KEY, FALSE);
5896 PERL_ARGS_ASSERT_INVLIST_MAX;
5898 if (max_ptr == NULL) {
5899 Perl_croak(aTHX_ "panic: inversion list without a '%s' element",
5903 return SvUV(*max_ptr);
5906 PERL_STATIC_INLINE void
5907 S_invlist_set_len(pTHX_ HV* const invlist, const UV len)
5909 /* Sets the current number of elements stored in the inversion list */
5911 PERL_ARGS_ASSERT_INVLIST_SET_LEN;
5913 if (len != 0 && len > invlist_max(invlist)) {
5914 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));
5917 if (hv_stores(invlist, INVLIST_LEN_KEY, newSVuv(len)) == NULL) {
5918 Perl_croak(aTHX_ "panic: can't store '%s' entry in inversion list",
5923 PERL_STATIC_INLINE void
5924 S_invlist_set_max(pTHX_ HV* const invlist, const UV max)
5927 /* Sets the maximum number of elements storable in the inversion list
5928 * without having to realloc() */
5930 PERL_ARGS_ASSERT_INVLIST_SET_MAX;
5932 if (max < invlist_len(invlist)) {
5933 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));
5936 if (hv_stores(invlist, INVLIST_MAX_KEY, newSVuv(max)) == NULL) {
5937 Perl_croak(aTHX_ "panic: can't store '%s' entry in inversion list",
5942 #ifndef PERL_IN_XSUB_RE
5944 Perl__new_invlist(pTHX_ IV initial_size)
5947 /* Return a pointer to a newly constructed inversion list, with enough
5948 * space to store 'initial_size' elements. If that number is negative, a
5949 * system default is used instead */
5951 HV* invlist = newHV();
5954 if (initial_size < 0) {
5955 initial_size = INVLIST_INITIAL_LEN;
5958 /* Allocate the initial space */
5959 Newx(list, initial_size, UV);
5960 invlist_set_array(invlist, list);
5962 /* set_len has to come before set_max, as the latter inspects the len */
5963 invlist_set_len(invlist, 0);
5964 invlist_set_max(invlist, initial_size);
5970 PERL_STATIC_INLINE void
5971 S_invlist_destroy(pTHX_ HV* const invlist)
5973 /* Inversion list destructor */
5975 SV** list_ptr = hv_fetchs(invlist, INVLIST_ARRAY_KEY, FALSE);
5977 PERL_ARGS_ASSERT_INVLIST_DESTROY;
5979 if (list_ptr != NULL) {
5980 UV *list = INT2PTR(UV *, SvUV(*list_ptr)); /* PERL_POISON needs lvalue */
5983 SvREFCNT_dec(invlist);
5987 S_invlist_extend(pTHX_ HV* const invlist, const UV new_max)
5989 /* Change the maximum size of an inversion list (up or down) */
5993 const UV old_max = invlist_max(invlist);
5995 PERL_ARGS_ASSERT_INVLIST_EXTEND;
5997 if (old_max == new_max) { /* If a no-op */
6001 array = orig_array = invlist_array(invlist);
6002 Renew(array, new_max, UV);
6004 /* If the size change moved the list in memory, set the new one */
6005 if (array != orig_array) {
6006 invlist_set_array(invlist, array);
6009 invlist_set_max(invlist, new_max);
6013 PERL_STATIC_INLINE void
6014 S_invlist_trim(pTHX_ HV* const invlist)
6016 PERL_ARGS_ASSERT_INVLIST_TRIM;
6018 /* Change the length of the inversion list to how many entries it currently
6021 invlist_extend(invlist, invlist_len(invlist));
6024 /* An element is in an inversion list iff its index is even numbered: 0, 2, 4,
6027 #define ELEMENT_IN_INVLIST_SET(i) (! ((i) & 1))
6028 #define PREV_ELEMENT_IN_INVLIST_SET(i) ! ELEMENT_IN_INVLIST_SET(i)
6030 #ifndef PERL_IN_XSUB_RE
6032 Perl__append_range_to_invlist(pTHX_ HV* const invlist, const UV start, const UV end)
6034 /* Subject to change or removal. Append the range from 'start' to 'end' at
6035 * the end of the inversion list. The range must be above any existing
6038 UV* array = invlist_array(invlist);
6039 UV max = invlist_max(invlist);
6040 UV len = invlist_len(invlist);
6042 PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST;
6046 /* Here, the existing list is non-empty. The current max entry in the
6047 * list is generally the first value not in the set, except when the
6048 * set extends to the end of permissible values, in which case it is
6049 * the first entry in that final set, and so this call is an attempt to
6050 * append out-of-order */
6052 UV final_element = len - 1;
6053 if (array[final_element] > start
6054 || ELEMENT_IN_INVLIST_SET(final_element))
6056 Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list");
6059 /* Here, it is a legal append. If the new range begins with the first
6060 * value not in the set, it is extending the set, so the new first
6061 * value not in the set is one greater than the newly extended range.
6063 if (array[final_element] == start) {
6064 if (end != UV_MAX) {
6065 array[final_element] = end + 1;
6068 /* But if the end is the maximum representable on the machine,
6069 * just let the range that this would extend have no end */
6070 invlist_set_len(invlist, len - 1);
6076 /* Here the new range doesn't extend any existing set. Add it */
6078 len += 2; /* Includes an element each for the start and end of range */
6080 /* If overflows the existing space, extend, which may cause the array to be
6083 invlist_extend(invlist, len);
6084 array = invlist_array(invlist);
6087 invlist_set_len(invlist, len);
6089 /* The next item on the list starts the range, the one after that is
6090 * one past the new range. */
6091 array[len - 2] = start;
6092 if (end != UV_MAX) {
6093 array[len - 1] = end + 1;
6096 /* But if the end is the maximum representable on the machine, just let
6097 * the range have no end */
6098 invlist_set_len(invlist, len - 1);
6104 S_invlist_union(pTHX_ HV* const a, HV* const b)
6106 /* Return a new inversion list which is the union of two inversion lists.
6107 * The basis for this comes from "Unicode Demystified" Chapter 13 by
6108 * Richard Gillam, published by Addison-Wesley, and explained at some
6109 * length there. The preface says to incorporate its examples into your
6110 * code at your own risk.
6112 * The algorithm is like a merge sort.
6114 * XXX A potential performance improvement is to keep track as we go along
6115 * if only one of the inputs contributes to the result, meaning the other
6116 * is a subset of that one. In that case, we can skip the final copy and
6117 * return the larger of the input lists */
6119 UV* array_a = invlist_array(a); /* a's array */
6120 UV* array_b = invlist_array(b);
6121 UV len_a = invlist_len(a); /* length of a's array */
6122 UV len_b = invlist_len(b);
6124 HV* u; /* the resulting union */
6128 UV i_a = 0; /* current index into a's array */
6132 /* running count, as explained in the algorithm source book; items are
6133 * stopped accumulating and are output when the count changes to/from 0.
6134 * The count is incremented when we start a range that's in the set, and
6135 * decremented when we start a range that's not in the set. So its range
6136 * is 0 to 2. Only when the count is zero is something not in the set.
6140 PERL_ARGS_ASSERT_INVLIST_UNION;
6142 /* Size the union for the worst case: that the sets are completely
6144 u = _new_invlist(len_a + len_b);
6145 array_u = invlist_array(u);
6147 /* Go through each list item by item, stopping when exhausted one of
6149 while (i_a < len_a && i_b < len_b) {
6150 UV cp; /* The element to potentially add to the union's array */
6151 bool cp_in_set; /* is it in the the input list's set or not */
6153 /* We need to take one or the other of the two inputs for the union.
6154 * Since we are merging two sorted lists, we take the smaller of the
6155 * next items. In case of a tie, we take the one that is in its set
6156 * first. If we took one not in the set first, it would decrement the
6157 * count, possibly to 0 which would cause it to be output as ending the
6158 * range, and the next time through we would take the same number, and
6159 * output it again as beginning the next range. By doing it the
6160 * opposite way, there is no possibility that the count will be
6161 * momentarily decremented to 0, and thus the two adjoining ranges will
6162 * be seamlessly merged. (In a tie and both are in the set or both not
6163 * in the set, it doesn't matter which we take first.) */
6164 if (array_a[i_a] < array_b[i_b]
6165 || (array_a[i_a] == array_b[i_b] && ELEMENT_IN_INVLIST_SET(i_a)))
6167 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6171 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6175 /* Here, have chosen which of the two inputs to look at. Only output
6176 * if the running count changes to/from 0, which marks the
6177 * beginning/end of a range in that's in the set */
6180 array_u[i_u++] = cp;
6187 array_u[i_u++] = cp;
6192 /* Here, we are finished going through at least one of the lists, which
6193 * means there is something remaining in at most one. We check if the list
6194 * that hasn't been exhausted is positioned such that we are in the middle
6195 * of a range in its set or not. (i_a and i_b point to the element beyond
6196 * the one we care about.) If in the set, we decrement 'count'; if 0, there
6197 * is potentially more to output.
6198 * There are four cases:
6199 * 1) Both weren't in their sets, count is 0, and remains 0. What's left
6200 * in the union is entirely from the non-exhausted set.
6201 * 2) Both were in their sets, count is 2. Nothing further should
6202 * be output, as everything that remains will be in the exhausted
6203 * list's set, hence in the union; decrementing to 1 but not 0 insures
6205 * 3) the exhausted was in its set, non-exhausted isn't, count is 1.
6206 * Nothing further should be output because the union includes
6207 * everything from the exhausted set. Not decrementing ensures that.
6208 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1;
6209 * decrementing to 0 insures that we look at the remainder of the
6210 * non-exhausted set */
6211 if ((i_a != len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6212 || (i_b != len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6217 /* The final length is what we've output so far, plus what else is about to
6218 * be output. (If 'count' is non-zero, then the input list we exhausted
6219 * has everything remaining up to the machine's limit in its set, and hence
6220 * in the union, so there will be no further output. */
6223 /* At most one of the subexpressions will be non-zero */
6224 len_u += (len_a - i_a) + (len_b - i_b);
6227 /* Set result to final length, which can change the pointer to array_u, so
6229 if (len_u != invlist_len(u)) {
6230 invlist_set_len(u, len_u);
6232 array_u = invlist_array(u);
6235 /* When 'count' is 0, the list that was exhausted (if one was shorter than
6236 * the other) ended with everything above it not in its set. That means
6237 * that the remaining part of the union is precisely the same as the
6238 * non-exhausted list, so can just copy it unchanged. (If both list were
6239 * exhausted at the same time, then the operations below will be both 0.)
6242 IV copy_count; /* At most one will have a non-zero copy count */
6243 if ((copy_count = len_a - i_a) > 0) {
6244 Copy(array_a + i_a, array_u + i_u, copy_count, UV);
6246 else if ((copy_count = len_b - i_b) > 0) {
6247 Copy(array_b + i_b, array_u + i_u, copy_count, UV);
6255 S_invlist_intersection(pTHX_ HV* const a, HV* const b)
6257 /* Return the intersection of two inversion lists. The basis for this
6258 * comes from "Unicode Demystified" Chapter 13 by Richard Gillam, published
6259 * by Addison-Wesley, and explained at some length there. The preface says
6260 * to incorporate its examples into your code at your own risk. In fact,
6263 * The algorithm is like a merge sort, and is essentially the same as the
6267 UV* array_a = invlist_array(a); /* a's array */
6268 UV* array_b = invlist_array(b);
6269 UV len_a = invlist_len(a); /* length of a's array */
6270 UV len_b = invlist_len(b);
6272 HV* r; /* the resulting intersection */
6276 UV i_a = 0; /* current index into a's array */
6280 /* running count, as explained in the algorithm source book; items are
6281 * stopped accumulating and are output when the count changes to/from 2.
6282 * The count is incremented when we start a range that's in the set, and
6283 * decremented when we start a range that's not in the set. So its range
6284 * is 0 to 2. Only when the count is 2 is something in the intersection.
6288 PERL_ARGS_ASSERT_INVLIST_INTERSECTION;
6290 /* Size the intersection for the worst case: that the intersection ends up
6291 * fragmenting everything to be completely disjoint */
6292 r= _new_invlist(len_a + len_b);
6293 array_r = invlist_array(r);
6295 /* Go through each list item by item, stopping when exhausted one of
6297 while (i_a < len_a && i_b < len_b) {
6298 UV cp; /* The element to potentially add to the intersection's
6300 bool cp_in_set; /* Is it in the input list's set or not */
6302 /* We need to take one or the other of the two inputs for the
6303 * intersection. Since we are merging two sorted lists, we take the
6304 * smaller of the next items. In case of a tie, we take the one that
6305 * is not in its set first (a difference from the union algorithm). If
6306 * we took one in the set first, it would increment the count, possibly
6307 * to 2 which would cause it to be output as starting a range in the
6308 * intersection, and the next time through we would take that same
6309 * number, and output it again as ending the set. By doing it the
6310 * opposite of this, there is no possibility that the count will be
6311 * momentarily incremented to 2. (In a tie and both are in the set or
6312 * both not in the set, it doesn't matter which we take first.) */
6313 if (array_a[i_a] < array_b[i_b]
6314 || (array_a[i_a] == array_b[i_b] && ! ELEMENT_IN_INVLIST_SET(i_a)))
6316 cp_in_set = ELEMENT_IN_INVLIST_SET(i_a);
6320 cp_in_set = ELEMENT_IN_INVLIST_SET(i_b);
6324 /* Here, have chosen which of the two inputs to look at. Only output
6325 * if the running count changes to/from 2, which marks the
6326 * beginning/end of a range that's in the intersection */
6330 array_r[i_r++] = cp;
6335 array_r[i_r++] = cp;
6341 /* Here, we are finished going through at least one of the lists, which
6342 * means there is something remaining in at most one. We check if the list
6343 * that has been exhausted is positioned such that we are in the middle
6344 * of a range in its set or not. (i_a and i_b point to elements 1 beyond
6345 * the ones we care about.) There are four cases:
6346 * 1) Both weren't in their sets, count is 0, and remains 0. There's
6347 * nothing left in the intersection.
6348 * 2) Both were in their sets, count is 2 and perhaps is incremented to
6349 * above 2. What should be output is exactly that which is in the
6350 * non-exhausted set, as everything it has is also in the intersection
6351 * set, and everything it doesn't have can't be in the intersection
6352 * 3) The exhausted was in its set, non-exhausted isn't, count is 1, and
6353 * gets incremented to 2. Like the previous case, the intersection is
6354 * everything that remains in the non-exhausted set.
6355 * 4) the exhausted wasn't in its set, non-exhausted is, count is 1, and
6356 * remains 1. And the intersection has nothing more. */
6357 if ((i_a == len_a && PREV_ELEMENT_IN_INVLIST_SET(i_a))
6358 || (i_b == len_b && PREV_ELEMENT_IN_INVLIST_SET(i_b)))
6363 /* The final length is what we've output so far plus what else is in the
6364 * intersection. At most one of the subexpressions below will be non-zero */
6367 len_r += (len_a - i_a) + (len_b - i_b);
6370 /* Set result to final length, which can change the pointer to array_r, so
6372 if (len_r != invlist_len(r)) {
6373 invlist_set_len(r, len_r);
6375 array_r = invlist_array(r);
6378 /* Finish outputting any remaining */
6379 if (count >= 2) { /* At most one will have a non-zero copy count */
6381 if ((copy_count = len_a - i_a) > 0) {
6382 Copy(array_a + i_a, array_r + i_r, copy_count, UV);
6384 else if ((copy_count = len_b - i_b) > 0) {
6385 Copy(array_b + i_b, array_r + i_r, copy_count, UV);
6393 S_add_range_to_invlist(pTHX_ HV* invlist, const UV start, const UV end)
6395 /* Add the range from 'start' to 'end' inclusive to the inversion list's
6396 * set. A pointer to the inversion list is returned. This may actually be
6397 * a new list, in which case the passed in one has been destroyed. The
6398 * passed in inversion list can be NULL, in which case a new one is created
6399 * with just the one range in it */
6405 if (invlist == NULL) {
6406 invlist = _new_invlist(2);
6410 len = invlist_len(invlist);
6413 /* If comes after the final entry, can just append it to the end */
6415 || start >= invlist_array(invlist)
6416 [invlist_len(invlist) - 1])
6418 _append_range_to_invlist(invlist, start, end);
6422 /* Here, can't just append things, create and return a new inversion list
6423 * which is the union of this range and the existing inversion list */
6424 range_invlist = _new_invlist(2);
6425 _append_range_to_invlist(range_invlist, start, end);
6427 added_invlist = invlist_union(invlist, range_invlist);
6429 /* The passed in list can be freed, as well as our temporary */
6430 invlist_destroy(range_invlist);
6431 if (invlist != added_invlist) {
6432 invlist_destroy(invlist);
6435 return added_invlist;
6438 PERL_STATIC_INLINE HV*
6439 S_add_cp_to_invlist(pTHX_ HV* invlist, const UV cp) {
6440 return add_range_to_invlist(invlist, cp, cp);
6443 /* End of inversion list object */
6446 - reg - regular expression, i.e. main body or parenthesized thing
6448 * Caller must absorb opening parenthesis.
6450 * Combining parenthesis handling with the base level of regular expression
6451 * is a trifle forced, but the need to tie the tails of the branches to what
6452 * follows makes it hard to avoid.
6454 #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1)
6456 #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1)
6458 #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
6462 S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
6463 /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
6466 register regnode *ret; /* Will be the head of the group. */
6467 register regnode *br;
6468 register regnode *lastbr;
6469 register regnode *ender = NULL;
6470 register I32 parno = 0;
6472 U32 oregflags = RExC_flags;
6473 bool have_branch = 0;
6475 I32 freeze_paren = 0;
6476 I32 after_freeze = 0;
6478 /* for (?g), (?gc), and (?o) warnings; warning
6479 about (?c) will warn about (?g) -- japhy */
6481 #define WASTED_O 0x01
6482 #define WASTED_G 0x02
6483 #define WASTED_C 0x04
6484 #define WASTED_GC (0x02|0x04)
6485 I32 wastedflags = 0x00;
6487 char * parse_start = RExC_parse; /* MJD */
6488 char * const oregcomp_parse = RExC_parse;
6490 GET_RE_DEBUG_FLAGS_DECL;
6492 PERL_ARGS_ASSERT_REG;
6493 DEBUG_PARSE("reg ");
6495 *flagp = 0; /* Tentatively. */
6498 /* Make an OPEN node, if parenthesized. */
6500 if ( *RExC_parse == '*') { /* (*VERB:ARG) */
6501 char *start_verb = RExC_parse;
6502 STRLEN verb_len = 0;
6503 char *start_arg = NULL;
6504 unsigned char op = 0;
6506 int internal_argval = 0; /* internal_argval is only useful if !argok */
6507 while ( *RExC_parse && *RExC_parse != ')' ) {
6508 if ( *RExC_parse == ':' ) {
6509 start_arg = RExC_parse + 1;
6515 verb_len = RExC_parse - start_verb;
6518 while ( *RExC_parse && *RExC_parse != ')' )
6520 if ( *RExC_parse != ')' )
6521 vFAIL("Unterminated verb pattern argument");
6522 if ( RExC_parse == start_arg )
6525 if ( *RExC_parse != ')' )
6526 vFAIL("Unterminated verb pattern");
6529 switch ( *start_verb ) {
6530 case 'A': /* (*ACCEPT) */
6531 if ( memEQs(start_verb,verb_len,"ACCEPT") ) {
6533 internal_argval = RExC_nestroot;
6536 case 'C': /* (*COMMIT) */
6537 if ( memEQs(start_verb,verb_len,"COMMIT") )
6540 case 'F': /* (*FAIL) */
6541 if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) {
6546 case ':': /* (*:NAME) */
6547 case 'M': /* (*MARK:NAME) */
6548 if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) {
6553 case 'P': /* (*PRUNE) */
6554 if ( memEQs(start_verb,verb_len,"PRUNE") )
6557 case 'S': /* (*SKIP) */
6558 if ( memEQs(start_verb,verb_len,"SKIP") )
6561 case 'T': /* (*THEN) */
6562 /* [19:06] <TimToady> :: is then */
6563 if ( memEQs(start_verb,verb_len,"THEN") ) {
6565 RExC_seen |= REG_SEEN_CUTGROUP;
6571 vFAIL3("Unknown verb pattern '%.*s'",
6572 verb_len, start_verb);
6575 if ( start_arg && internal_argval ) {
6576 vFAIL3("Verb pattern '%.*s' may not have an argument",
6577 verb_len, start_verb);
6578 } else if ( argok < 0 && !start_arg ) {
6579 vFAIL3("Verb pattern '%.*s' has a mandatory argument",
6580 verb_len, start_verb);
6582 ret = reganode(pRExC_state, op, internal_argval);
6583 if ( ! internal_argval && ! SIZE_ONLY ) {
6585 SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
6586 ARG(ret) = add_data( pRExC_state, 1, "S" );
6587 RExC_rxi->data->data[ARG(ret)]=(void*)sv;
6594 if (!internal_argval)
6595 RExC_seen |= REG_SEEN_VERBARG;
6596 } else if ( start_arg ) {
6597 vFAIL3("Verb pattern '%.*s' may not have an argument",
6598 verb_len, start_verb);
6600 ret = reg_node(pRExC_state, op);
6602 nextchar(pRExC_state);
6605 if (*RExC_parse == '?') { /* (?...) */
6606 bool is_logical = 0;
6607 const char * const seqstart = RExC_parse;
6608 bool has_use_defaults = FALSE;
6611 paren = *RExC_parse++;
6612 ret = NULL; /* For look-ahead/behind. */
6615 case 'P': /* (?P...) variants for those used to PCRE/Python */
6616 paren = *RExC_parse++;
6617 if ( paren == '<') /* (?P<...>) named capture */
6619 else if (paren == '>') { /* (?P>name) named recursion */
6620 goto named_recursion;
6622 else if (paren == '=') { /* (?P=...) named backref */
6623 /* this pretty much dupes the code for \k<NAME> in regatom(), if
6624 you change this make sure you change that */
6625 char* name_start = RExC_parse;
6627 SV *sv_dat = reg_scan_name(pRExC_state,
6628 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6629 if (RExC_parse == name_start || *RExC_parse != ')')
6630 vFAIL2("Sequence %.3s... not terminated",parse_start);
6633 num = add_data( pRExC_state, 1, "S" );
6634 RExC_rxi->data->data[num]=(void*)sv_dat;
6635 SvREFCNT_inc_simple_void(sv_dat);
6638 ret = reganode(pRExC_state,
6641 : (MORE_ASCII_RESTRICTED)
6643 : (AT_LEAST_UNI_SEMANTICS)
6651 Set_Node_Offset(ret, parse_start+1);
6652 Set_Node_Cur_Length(ret); /* MJD */
6654 nextchar(pRExC_state);
6658 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6660 case '<': /* (?<...) */
6661 if (*RExC_parse == '!')
6663 else if (*RExC_parse != '=')
6669 case '\'': /* (?'...') */
6670 name_start= RExC_parse;
6671 svname = reg_scan_name(pRExC_state,
6672 SIZE_ONLY ? /* reverse test from the others */
6673 REG_RSN_RETURN_NAME :
6674 REG_RSN_RETURN_NULL);
6675 if (RExC_parse == name_start) {
6677 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6680 if (*RExC_parse != paren)
6681 vFAIL2("Sequence (?%c... not terminated",
6682 paren=='>' ? '<' : paren);
6686 if (!svname) /* shouldn't happen */
6688 "panic: reg_scan_name returned NULL");
6689 if (!RExC_paren_names) {
6690 RExC_paren_names= newHV();
6691 sv_2mortal(MUTABLE_SV(RExC_paren_names));
6693 RExC_paren_name_list= newAV();
6694 sv_2mortal(MUTABLE_SV(RExC_paren_name_list));
6697 he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
6699 sv_dat = HeVAL(he_str);
6701 /* croak baby croak */
6703 "panic: paren_name hash element allocation failed");
6704 } else if ( SvPOK(sv_dat) ) {
6705 /* (?|...) can mean we have dupes so scan to check
6706 its already been stored. Maybe a flag indicating
6707 we are inside such a construct would be useful,
6708 but the arrays are likely to be quite small, so
6709 for now we punt -- dmq */
6710 IV count = SvIV(sv_dat);
6711 I32 *pv = (I32*)SvPVX(sv_dat);
6713 for ( i = 0 ; i < count ; i++ ) {
6714 if ( pv[i] == RExC_npar ) {
6720 pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1);
6721 SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32));
6722 pv[count] = RExC_npar;
6723 SvIV_set(sv_dat, SvIVX(sv_dat) + 1);
6726 (void)SvUPGRADE(sv_dat,SVt_PVNV);
6727 sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
6729 SvIV_set(sv_dat, 1);
6732 if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname)))
6733 SvREFCNT_dec(svname);
6736 /*sv_dump(sv_dat);*/
6738 nextchar(pRExC_state);
6740 goto capturing_parens;
6742 RExC_seen |= REG_SEEN_LOOKBEHIND;
6743 RExC_in_lookbehind++;
6745 case '=': /* (?=...) */
6746 RExC_seen_zerolen++;
6748 case '!': /* (?!...) */
6749 RExC_seen_zerolen++;
6750 if (*RExC_parse == ')') {
6751 ret=reg_node(pRExC_state, OPFAIL);
6752 nextchar(pRExC_state);
6756 case '|': /* (?|...) */
6757 /* branch reset, behave like a (?:...) except that
6758 buffers in alternations share the same numbers */
6760 after_freeze = freeze_paren = RExC_npar;
6762 case ':': /* (?:...) */
6763 case '>': /* (?>...) */
6765 case '$': /* (?$...) */
6766 case '@': /* (?@...) */
6767 vFAIL2("Sequence (?%c...) not implemented", (int)paren);
6769 case '#': /* (?#...) */
6770 while (*RExC_parse && *RExC_parse != ')')
6772 if (*RExC_parse != ')')
6773 FAIL("Sequence (?#... not terminated");
6774 nextchar(pRExC_state);
6777 case '0' : /* (?0) */
6778 case 'R' : /* (?R) */
6779 if (*RExC_parse != ')')
6780 FAIL("Sequence (?R) not terminated");
6781 ret = reg_node(pRExC_state, GOSTART);
6782 *flagp |= POSTPONED;
6783 nextchar(pRExC_state);
6786 { /* named and numeric backreferences */
6788 case '&': /* (?&NAME) */
6789 parse_start = RExC_parse - 1;
6792 SV *sv_dat = reg_scan_name(pRExC_state,
6793 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6794 num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
6796 goto gen_recurse_regop;
6799 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6801 vFAIL("Illegal pattern");
6803 goto parse_recursion;
6805 case '-': /* (?-1) */
6806 if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
6807 RExC_parse--; /* rewind to let it be handled later */
6811 case '1': case '2': case '3': case '4': /* (?1) */
6812 case '5': case '6': case '7': case '8': case '9':
6815 num = atoi(RExC_parse);
6816 parse_start = RExC_parse - 1; /* MJD */
6817 if (*RExC_parse == '-')
6819 while (isDIGIT(*RExC_parse))
6821 if (*RExC_parse!=')')
6822 vFAIL("Expecting close bracket");
6825 if ( paren == '-' ) {
6827 Diagram of capture buffer numbering.
6828 Top line is the normal capture buffer numbers
6829 Bottom line is the negative indexing as from
6833 /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
6837 num = RExC_npar + num;
6840 vFAIL("Reference to nonexistent group");
6842 } else if ( paren == '+' ) {
6843 num = RExC_npar + num - 1;
6846 ret = reganode(pRExC_state, GOSUB, num);
6848 if (num > (I32)RExC_rx->nparens) {
6850 vFAIL("Reference to nonexistent group");
6852 ARG2L_SET( ret, RExC_recurse_count++);
6854 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
6855 "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
6859 RExC_seen |= REG_SEEN_RECURSE;
6860 Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
6861 Set_Node_Offset(ret, parse_start); /* MJD */
6863 *flagp |= POSTPONED;
6864 nextchar(pRExC_state);
6866 } /* named and numeric backreferences */
6869 case '?': /* (??...) */
6871 if (*RExC_parse != '{') {
6873 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
6876 *flagp |= POSTPONED;
6877 paren = *RExC_parse++;
6879 case '{': /* (?{...}) */
6884 char *s = RExC_parse;
6886 RExC_seen_zerolen++;
6887 RExC_seen |= REG_SEEN_EVAL;
6888 while (count && (c = *RExC_parse)) {
6899 if (*RExC_parse != ')') {
6901 vFAIL("Sequence (?{...}) not terminated or not {}-balanced");
6905 OP_4tree *sop, *rop;
6906 SV * const sv = newSVpvn(s, RExC_parse - 1 - s);
6909 Perl_save_re_context(aTHX);
6910 rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad);
6911 sop->op_private |= OPpREFCOUNTED;
6912 /* re_dup will OpREFCNT_inc */
6913 OpREFCNT_set(sop, 1);
6916 n = add_data(pRExC_state, 3, "nop");
6917 RExC_rxi->data->data[n] = (void*)rop;
6918 RExC_rxi->data->data[n+1] = (void*)sop;
6919 RExC_rxi->data->data[n+2] = (void*)pad;
6922 else { /* First pass */
6923 if (PL_reginterp_cnt < ++RExC_seen_evals
6925 /* No compiled RE interpolated, has runtime
6926 components ===> unsafe. */
6927 FAIL("Eval-group not allowed at runtime, use re 'eval'");
6928 if (PL_tainting && PL_tainted)
6929 FAIL("Eval-group in insecure regular expression");
6930 #if PERL_VERSION > 8
6931 if (IN_PERL_COMPILETIME)
6936 nextchar(pRExC_state);
6938 ret = reg_node(pRExC_state, LOGICAL);
6941 REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n));
6942 /* deal with the length of this later - MJD */
6945 ret = reganode(pRExC_state, EVAL, n);
6946 Set_Node_Length(ret, RExC_parse - parse_start + 1);
6947 Set_Node_Offset(ret, parse_start);
6950 case '(': /* (?(?{...})...) and (?(?=...)...) */
6953 if (RExC_parse[0] == '?') { /* (?(?...)) */
6954 if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
6955 || RExC_parse[1] == '<'
6956 || RExC_parse[1] == '{') { /* Lookahead or eval. */
6959 ret = reg_node(pRExC_state, LOGICAL);
6962 REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1));
6966 else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
6967 || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
6969 char ch = RExC_parse[0] == '<' ? '>' : '\'';
6970 char *name_start= RExC_parse++;
6972 SV *sv_dat=reg_scan_name(pRExC_state,
6973 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
6974 if (RExC_parse == name_start || *RExC_parse != ch)
6975 vFAIL2("Sequence (?(%c... not terminated",
6976 (ch == '>' ? '<' : ch));
6979 num = add_data( pRExC_state, 1, "S" );
6980 RExC_rxi->data->data[num]=(void*)sv_dat;
6981 SvREFCNT_inc_simple_void(sv_dat);
6983 ret = reganode(pRExC_state,NGROUPP,num);
6984 goto insert_if_check_paren;
6986 else if (RExC_parse[0] == 'D' &&
6987 RExC_parse[1] == 'E' &&
6988 RExC_parse[2] == 'F' &&
6989 RExC_parse[3] == 'I' &&
6990 RExC_parse[4] == 'N' &&
6991 RExC_parse[5] == 'E')
6993 ret = reganode(pRExC_state,DEFINEP,0);
6996 goto insert_if_check_paren;
6998 else if (RExC_parse[0] == 'R') {
7001 if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
7002 parno = atoi(RExC_parse++);
7003 while (isDIGIT(*RExC_parse))
7005 } else if (RExC_parse[0] == '&') {
7008 sv_dat = reg_scan_name(pRExC_state,
7009 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
7010 parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
7012 ret = reganode(pRExC_state,INSUBP,parno);
7013 goto insert_if_check_paren;
7015 else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
7018 parno = atoi(RExC_parse++);
7020 while (isDIGIT(*RExC_parse))
7022 ret = reganode(pRExC_state, GROUPP, parno);
7024 insert_if_check_paren:
7025 if ((c = *nextchar(pRExC_state)) != ')')
7026 vFAIL("Switch condition not recognized");
7028 REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0));
7029 br = regbranch(pRExC_state, &flags, 1,depth+1);
7031 br = reganode(pRExC_state, LONGJMP, 0);
7033 REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0));
7034 c = *nextchar(pRExC_state);
7039 vFAIL("(?(DEFINE)....) does not allow branches");
7040 lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
7041 regbranch(pRExC_state, &flags, 1,depth+1);
7042 REGTAIL(pRExC_state, ret, lastbr);
7045 c = *nextchar(pRExC_state);
7050 vFAIL("Switch (?(condition)... contains too many branches");
7051 ender = reg_node(pRExC_state, TAIL);
7052 REGTAIL(pRExC_state, br, ender);
7054 REGTAIL(pRExC_state, lastbr, ender);
7055 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender);
7058 REGTAIL(pRExC_state, ret, ender);
7059 RExC_size++; /* XXX WHY do we need this?!!
7060 For large programs it seems to be required
7061 but I can't figure out why. -- dmq*/
7065 vFAIL2("Unknown switch condition (?(%.2s", RExC_parse);
7069 RExC_parse--; /* for vFAIL to print correctly */
7070 vFAIL("Sequence (? incomplete");
7072 case DEFAULT_PAT_MOD: /* Use default flags with the exceptions
7074 has_use_defaults = TRUE;
7075 STD_PMMOD_FLAGS_CLEAR(&RExC_flags);
7076 set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics)
7077 ? REGEX_UNICODE_CHARSET
7078 : REGEX_DEPENDS_CHARSET);
7082 parse_flags: /* (?i) */
7084 U32 posflags = 0, negflags = 0;
7085 U32 *flagsp = &posflags;
7086 char has_charset_modifier = '\0';
7087 regex_charset cs = (RExC_utf8 || RExC_uni_semantics)
7088 ? REGEX_UNICODE_CHARSET
7089 : REGEX_DEPENDS_CHARSET;
7091 while (*RExC_parse) {
7092 /* && strchr("iogcmsx", *RExC_parse) */
7093 /* (?g), (?gc) and (?o) are useless here
7094 and must be globally applied -- japhy */
7095 switch (*RExC_parse) {
7096 CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp);
7097 case LOCALE_PAT_MOD:
7098 if (has_charset_modifier) {
7099 goto excess_modifier;
7101 else if (flagsp == &negflags) {
7104 cs = REGEX_LOCALE_CHARSET;
7105 has_charset_modifier = LOCALE_PAT_MOD;
7106 RExC_contains_locale = 1;
7108 case UNICODE_PAT_MOD:
7109 if (has_charset_modifier) {
7110 goto excess_modifier;
7112 else if (flagsp == &negflags) {
7115 cs = REGEX_UNICODE_CHARSET;
7116 has_charset_modifier = UNICODE_PAT_MOD;
7118 case ASCII_RESTRICT_PAT_MOD:
7119 if (flagsp == &negflags) {
7122 if (has_charset_modifier) {
7123 if (cs != REGEX_ASCII_RESTRICTED_CHARSET) {
7124 goto excess_modifier;
7126 /* Doubled modifier implies more restricted */
7127 cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET;
7130 cs = REGEX_ASCII_RESTRICTED_CHARSET;
7132 has_charset_modifier = ASCII_RESTRICT_PAT_MOD;
7134 case DEPENDS_PAT_MOD:
7135 if (has_use_defaults) {
7136 goto fail_modifiers;
7138 else if (flagsp == &negflags) {
7141 else if (has_charset_modifier) {
7142 goto excess_modifier;
7145 /* The dual charset means unicode semantics if the
7146 * pattern (or target, not known until runtime) are
7147 * utf8, or something in the pattern indicates unicode
7149 cs = (RExC_utf8 || RExC_uni_semantics)
7150 ? REGEX_UNICODE_CHARSET
7151 : REGEX_DEPENDS_CHARSET;
7152 has_charset_modifier = DEPENDS_PAT_MOD;
7156 if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) {
7157 vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD);
7159 else if (has_charset_modifier == *(RExC_parse - 1)) {
7160 vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1));
7163 vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1));
7168 vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1));
7170 case ONCE_PAT_MOD: /* 'o' */
7171 case GLOBAL_PAT_MOD: /* 'g' */
7172 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7173 const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G;
7174 if (! (wastedflags & wflagbit) ) {
7175 wastedflags |= wflagbit;
7178 "Useless (%s%c) - %suse /%c modifier",
7179 flagsp == &negflags ? "?-" : "?",
7181 flagsp == &negflags ? "don't " : "",
7188 case CONTINUE_PAT_MOD: /* 'c' */
7189 if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
7190 if (! (wastedflags & WASTED_C) ) {
7191 wastedflags |= WASTED_GC;
7194 "Useless (%sc) - %suse /gc modifier",
7195 flagsp == &negflags ? "?-" : "?",
7196 flagsp == &negflags ? "don't " : ""
7201 case KEEPCOPY_PAT_MOD: /* 'p' */
7202 if (flagsp == &negflags) {
7204 ckWARNreg(RExC_parse + 1,"Useless use of (?-p)");
7206 *flagsp |= RXf_PMf_KEEPCOPY;
7210 /* A flag is a default iff it is following a minus, so
7211 * if there is a minus, it means will be trying to
7212 * re-specify a default which is an error */
7213 if (has_use_defaults || flagsp == &negflags) {
7216 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7220 wastedflags = 0; /* reset so (?g-c) warns twice */
7226 RExC_flags |= posflags;
7227 RExC_flags &= ~negflags;
7228 set_regex_charset(&RExC_flags, cs);
7230 oregflags |= posflags;
7231 oregflags &= ~negflags;
7232 set_regex_charset(&oregflags, cs);
7234 nextchar(pRExC_state);
7245 vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart);
7250 }} /* one for the default block, one for the switch */
7257 ret = reganode(pRExC_state, OPEN, parno);
7260 RExC_nestroot = parno;
7261 if (RExC_seen & REG_SEEN_RECURSE
7262 && !RExC_open_parens[parno-1])
7264 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7265 "Setting open paren #%"IVdf" to %d\n",
7266 (IV)parno, REG_NODE_NUM(ret)));
7267 RExC_open_parens[parno-1]= ret;
7270 Set_Node_Length(ret, 1); /* MJD */
7271 Set_Node_Offset(ret, RExC_parse); /* MJD */
7279 /* Pick up the branches, linking them together. */
7280 parse_start = RExC_parse; /* MJD */
7281 br = regbranch(pRExC_state, &flags, 1,depth+1);
7283 /* branch_len = (paren != 0); */
7287 if (*RExC_parse == '|') {
7288 if (!SIZE_ONLY && RExC_extralen) {
7289 reginsert(pRExC_state, BRANCHJ, br, depth+1);
7292 reginsert(pRExC_state, BRANCH, br, depth+1);
7293 Set_Node_Length(br, paren != 0);
7294 Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
7298 RExC_extralen += 1; /* For BRANCHJ-BRANCH. */
7300 else if (paren == ':') {
7301 *flagp |= flags&SIMPLE;
7303 if (is_open) { /* Starts with OPEN. */
7304 REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */
7306 else if (paren != '?') /* Not Conditional */
7308 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7310 while (*RExC_parse == '|') {
7311 if (!SIZE_ONLY && RExC_extralen) {
7312 ender = reganode(pRExC_state, LONGJMP,0);
7313 REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */
7316 RExC_extralen += 2; /* Account for LONGJMP. */
7317 nextchar(pRExC_state);
7319 if (RExC_npar > after_freeze)
7320 after_freeze = RExC_npar;
7321 RExC_npar = freeze_paren;
7323 br = regbranch(pRExC_state, &flags, 0, depth+1);
7327 REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */
7329 *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED);
7332 if (have_branch || paren != ':') {
7333 /* Make a closing node, and hook it on the end. */
7336 ender = reg_node(pRExC_state, TAIL);
7339 ender = reganode(pRExC_state, CLOSE, parno);
7340 if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
7341 DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
7342 "Setting close paren #%"IVdf" to %d\n",
7343 (IV)parno, REG_NODE_NUM(ender)));
7344 RExC_close_parens[parno-1]= ender;
7345 if (RExC_nestroot == parno)
7348 Set_Node_Offset(ender,RExC_parse+1); /* MJD */
7349 Set_Node_Length(ender,1); /* MJD */
7355 *flagp &= ~HASWIDTH;
7358 ender = reg_node(pRExC_state, SUCCEED);
7361 ender = reg_node(pRExC_state, END);
7363 assert(!RExC_opend); /* there can only be one! */
7368 REGTAIL(pRExC_state, lastbr, ender);
7370 if (have_branch && !SIZE_ONLY) {
7372 RExC_seen |= REG_TOP_LEVEL_BRANCHES;
7374 /* Hook the tails of the branches to the closing node. */
7375 for (br = ret; br; br = regnext(br)) {
7376 const U8 op = PL_regkind[OP(br)];
7378 REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender);
7380 else if (op == BRANCHJ) {
7381 REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender);
7389 static const char parens[] = "=!<,>";
7391 if (paren && (p = strchr(parens, paren))) {
7392 U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH;
7393 int flag = (p - parens) > 1;
7396 node = SUSPEND, flag = 0;
7397 reginsert(pRExC_state, node,ret, depth+1);
7398 Set_Node_Cur_Length(ret);
7399 Set_Node_Offset(ret, parse_start + 1);
7401 REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL));
7405 /* Check for proper termination. */
7407 RExC_flags = oregflags;
7408 if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') {
7409 RExC_parse = oregcomp_parse;
7410 vFAIL("Unmatched (");
7413 else if (!paren && RExC_parse < RExC_end) {
7414 if (*RExC_parse == ')') {
7416 vFAIL("Unmatched )");
7419 FAIL("Junk on end of regexp"); /* "Can't happen". */
7423 if (RExC_in_lookbehind) {
7424 RExC_in_lookbehind--;
7426 if (after_freeze > RExC_npar)
7427 RExC_npar = after_freeze;
7432 - regbranch - one alternative of an | operator
7434 * Implements the concatenation operator.
7437 S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth)
7440 register regnode *ret;
7441 register regnode *chain = NULL;
7442 register regnode *latest;
7443 I32 flags = 0, c = 0;
7444 GET_RE_DEBUG_FLAGS_DECL;
7446 PERL_ARGS_ASSERT_REGBRANCH;
7448 DEBUG_PARSE("brnc");
7453 if (!SIZE_ONLY && RExC_extralen)
7454 ret = reganode(pRExC_state, BRANCHJ,0);
7456 ret = reg_node(pRExC_state, BRANCH);
7457 Set_Node_Length(ret, 1);
7461 if (!first && SIZE_ONLY)
7462 RExC_extralen += 1; /* BRANCHJ */
7464 *flagp = WORST; /* Tentatively. */
7467 nextchar(pRExC_state);
7468 while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') {
7470 latest = regpiece(pRExC_state, &flags,depth+1);
7471 if (latest == NULL) {
7472 if (flags & TRYAGAIN)
7476 else if (ret == NULL)
7478 *flagp |= flags&(HASWIDTH|POSTPONED);
7479 if (chain == NULL) /* First piece. */
7480 *flagp |= flags&SPSTART;
7483 REGTAIL(pRExC_state, chain, latest);
7488 if (chain == NULL) { /* Loop ran zero times. */
7489 chain = reg_node(pRExC_state, NOTHING);
7494 *flagp |= flags&SIMPLE;
7501 - regpiece - something followed by possible [*+?]
7503 * Note that the branching code sequences used for ? and the general cases
7504 * of * and + are somewhat optimized: they use the same NOTHING node as
7505 * both the endmarker for their branch list and the body of the last branch.
7506 * It might seem that this node could be dispensed with entirely, but the
7507 * endmarker role is not redundant.
7510 S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7513 register regnode *ret;
7515 register char *next;
7517 const char * const origparse = RExC_parse;
7519 I32 max = REG_INFTY;
7520 #ifdef RE_TRACK_PATTERN_OFFSETS
7523 const char *maxpos = NULL;
7524 GET_RE_DEBUG_FLAGS_DECL;
7526 PERL_ARGS_ASSERT_REGPIECE;
7528 DEBUG_PARSE("piec");
7530 ret = regatom(pRExC_state, &flags,depth+1);
7532 if (flags & TRYAGAIN)
7539 if (op == '{' && regcurly(RExC_parse)) {
7541 #ifdef RE_TRACK_PATTERN_OFFSETS
7542 parse_start = RExC_parse; /* MJD */
7544 next = RExC_parse + 1;
7545 while (isDIGIT(*next) || *next == ',') {
7554 if (*next == '}') { /* got one */
7558 min = atoi(RExC_parse);
7562 maxpos = RExC_parse;
7564 if (!max && *maxpos != '0')
7565 max = REG_INFTY; /* meaning "infinity" */
7566 else if (max >= REG_INFTY)
7567 vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1);
7569 nextchar(pRExC_state);
7572 if ((flags&SIMPLE)) {
7573 RExC_naughty += 2 + RExC_naughty / 2;
7574 reginsert(pRExC_state, CURLY, ret, depth+1);
7575 Set_Node_Offset(ret, parse_start+1); /* MJD */
7576 Set_Node_Cur_Length(ret);
7579 regnode * const w = reg_node(pRExC_state, WHILEM);
7582 REGTAIL(pRExC_state, ret, w);
7583 if (!SIZE_ONLY && RExC_extralen) {
7584 reginsert(pRExC_state, LONGJMP,ret, depth+1);
7585 reginsert(pRExC_state, NOTHING,ret, depth+1);
7586 NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
7588 reginsert(pRExC_state, CURLYX,ret, depth+1);
7590 Set_Node_Offset(ret, parse_start+1);
7591 Set_Node_Length(ret,
7592 op == '{' ? (RExC_parse - parse_start) : 1);
7594 if (!SIZE_ONLY && RExC_extralen)
7595 NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */
7596 REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING));
7598 RExC_whilem_seen++, RExC_extralen += 3;
7599 RExC_naughty += 4 + RExC_naughty; /* compound interest */
7608 vFAIL("Can't do {n,m} with n > m");
7610 ARG1_SET(ret, (U16)min);
7611 ARG2_SET(ret, (U16)max);
7623 #if 0 /* Now runtime fix should be reliable. */
7625 /* if this is reinstated, don't forget to put this back into perldiag:
7627 =item Regexp *+ operand could be empty at {#} in regex m/%s/
7629 (F) The part of the regexp subject to either the * or + quantifier
7630 could match an empty string. The {#} shows in the regular
7631 expression about where the problem was discovered.
7635 if (!(flags&HASWIDTH) && op != '?')
7636 vFAIL("Regexp *+ operand could be empty");
7639 #ifdef RE_TRACK_PATTERN_OFFSETS
7640 parse_start = RExC_parse;
7642 nextchar(pRExC_state);
7644 *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
7646 if (op == '*' && (flags&SIMPLE)) {
7647 reginsert(pRExC_state, STAR, ret, depth+1);
7651 else if (op == '*') {
7655 else if (op == '+' && (flags&SIMPLE)) {
7656 reginsert(pRExC_state, PLUS, ret, depth+1);
7660 else if (op == '+') {
7664 else if (op == '?') {
7669 if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) {
7670 ckWARN3reg(RExC_parse,
7671 "%.*s matches null string many times",
7672 (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0),
7676 if (RExC_parse < RExC_end && *RExC_parse == '?') {
7677 nextchar(pRExC_state);
7678 reginsert(pRExC_state, MINMOD, ret, depth+1);
7679 REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
7681 #ifndef REG_ALLOW_MINMOD_SUSPEND
7684 if (RExC_parse < RExC_end && *RExC_parse == '+') {
7686 nextchar(pRExC_state);
7687 ender = reg_node(pRExC_state, SUCCEED);
7688 REGTAIL(pRExC_state, ret, ender);
7689 reginsert(pRExC_state, SUSPEND, ret, depth+1);
7691 ender = reg_node(pRExC_state, TAIL);
7692 REGTAIL(pRExC_state, ret, ender);
7696 if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
7698 vFAIL("Nested quantifiers");
7705 /* reg_namedseq(pRExC_state,UVp, UV depth)
7707 This is expected to be called by a parser routine that has
7708 recognized '\N' and needs to handle the rest. RExC_parse is
7709 expected to point at the first char following the N at the time
7712 The \N may be inside (indicated by valuep not being NULL) or outside a
7715 \N may begin either a named sequence, or if outside a character class, mean
7716 to match a non-newline. For non single-quoted regexes, the tokenizer has
7717 attempted to decide which, and in the case of a named sequence converted it
7718 into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...},
7719 where c1... are the characters in the sequence. For single-quoted regexes,
7720 the tokenizer passes the \N sequence through unchanged; this code will not
7721 attempt to determine this nor expand those. The net effect is that if the
7722 beginning of the passed-in pattern isn't '{U+' or there is no '}', it
7723 signals that this \N occurrence means to match a non-newline.
7725 Only the \N{U+...} form should occur in a character class, for the same
7726 reason that '.' inside a character class means to just match a period: it
7727 just doesn't make sense.
7729 If valuep is non-null then it is assumed that we are parsing inside
7730 of a charclass definition and the first codepoint in the resolved
7731 string is returned via *valuep and the routine will return NULL.
7732 In this mode if a multichar string is returned from the charnames
7733 handler, a warning will be issued, and only the first char in the
7734 sequence will be examined. If the string returned is zero length
7735 then the value of *valuep is undefined and NON-NULL will
7736 be returned to indicate failure. (This will NOT be a valid pointer
7739 If valuep is null then it is assumed that we are parsing normal text and a
7740 new EXACT node is inserted into the program containing the resolved string,
7741 and a pointer to the new node is returned. But if the string is zero length
7742 a NOTHING node is emitted instead.
7744 On success RExC_parse is set to the char following the endbrace.
7745 Parsing failures will generate a fatal error via vFAIL(...)
7748 S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp, U32 depth)
7750 char * endbrace; /* '}' following the name */
7751 regnode *ret = NULL;
7754 GET_RE_DEBUG_FLAGS_DECL;
7756 PERL_ARGS_ASSERT_REG_NAMEDSEQ;
7760 /* The [^\n] meaning of \N ignores spaces and comments under the /x
7761 * modifier. The other meaning does not */
7762 p = (RExC_flags & RXf_PMf_EXTENDED)
7763 ? regwhite( pRExC_state, RExC_parse )
7766 /* Disambiguate between \N meaning a named character versus \N meaning
7767 * [^\n]. The former is assumed when it can't be the latter. */
7768 if (*p != '{' || regcurly(p)) {
7771 /* no bare \N in a charclass */
7772 vFAIL("\\N in a character class must be a named character: \\N{...}");
7774 nextchar(pRExC_state);
7775 ret = reg_node(pRExC_state, REG_ANY);
7776 *flagp |= HASWIDTH|SIMPLE;
7779 Set_Node_Length(ret, 1); /* MJD */
7783 /* Here, we have decided it should be a named sequence */
7785 /* The test above made sure that the next real character is a '{', but
7786 * under the /x modifier, it could be separated by space (or a comment and
7787 * \n) and this is not allowed (for consistency with \x{...} and the
7788 * tokenizer handling of \N{NAME}). */
7789 if (*RExC_parse != '{') {
7790 vFAIL("Missing braces on \\N{}");
7793 RExC_parse++; /* Skip past the '{' */
7795 if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */
7796 || ! (endbrace == RExC_parse /* nothing between the {} */
7797 || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */
7798 && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */
7800 if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */
7801 vFAIL("\\N{NAME} must be resolved by the lexer");
7804 if (endbrace == RExC_parse) { /* empty: \N{} */
7806 RExC_parse = endbrace + 1;
7807 return reg_node(pRExC_state,NOTHING);
7811 ckWARNreg(RExC_parse,
7812 "Ignoring zero length \\N{} in character class"
7814 RExC_parse = endbrace + 1;
7817 return (regnode *) &RExC_parse; /* Invalid regnode pointer */
7820 REQUIRE_UTF8; /* named sequences imply Unicode semantics */
7821 RExC_parse += 2; /* Skip past the 'U+' */
7823 if (valuep) { /* In a bracketed char class */
7824 /* We only pay attention to the first char of
7825 multichar strings being returned. I kinda wonder
7826 if this makes sense as it does change the behaviour
7827 from earlier versions, OTOH that behaviour was broken
7828 as well. XXX Solution is to recharacterize as
7829 [rest-of-class]|multi1|multi2... */
7831 STRLEN length_of_hex;
7832 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
7833 | PERL_SCAN_DISALLOW_PREFIX
7834 | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
7836 char * endchar = RExC_parse + strcspn(RExC_parse, ".}");
7837 if (endchar < endbrace) {
7838 ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class");
7841 length_of_hex = (STRLEN)(endchar - RExC_parse);
7842 *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL);
7844 /* The tokenizer should have guaranteed validity, but it's possible to
7845 * bypass it by using single quoting, so check */
7846 if (length_of_hex == 0
7847 || length_of_hex != (STRLEN)(endchar - RExC_parse) )
7849 RExC_parse += length_of_hex; /* Includes all the valid */
7850 RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */
7851 ? UTF8SKIP(RExC_parse)
7853 /* Guard against malformed utf8 */
7854 if (RExC_parse >= endchar) RExC_parse = endchar;
7855 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7858 RExC_parse = endbrace + 1;
7859 if (endchar == endbrace) return NULL;
7861 ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */
7863 else { /* Not a char class */
7865 /* What is done here is to convert this to a sub-pattern of the form
7866 * (?:\x{char1}\x{char2}...)
7867 * and then call reg recursively. That way, it retains its atomicness,
7868 * while not having to worry about special handling that some code
7869 * points may have. toke.c has converted the original Unicode values
7870 * to native, so that we can just pass on the hex values unchanged. We
7871 * do have to set a flag to keep recoding from happening in the
7874 SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP);
7876 char *endchar; /* Points to '.' or '}' ending cur char in the input
7878 char *orig_end = RExC_end;
7880 while (RExC_parse < endbrace) {
7882 /* Code points are separated by dots. If none, there is only one
7883 * code point, and is terminated by the brace */
7884 endchar = RExC_parse + strcspn(RExC_parse, ".}");
7886 /* Convert to notation the rest of the code understands */
7887 sv_catpv(substitute_parse, "\\x{");
7888 sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse);
7889 sv_catpv(substitute_parse, "}");
7891 /* Point to the beginning of the next character in the sequence. */
7892 RExC_parse = endchar + 1;
7894 sv_catpv(substitute_parse, ")");
7896 RExC_parse = SvPV(substitute_parse, len);
7898 /* Don't allow empty number */
7900 vFAIL("Invalid hexadecimal number in \\N{U+...}");
7902 RExC_end = RExC_parse + len;
7904 /* The values are Unicode, and therefore not subject to recoding */
7905 RExC_override_recoding = 1;
7907 ret = reg(pRExC_state, 1, flagp, depth+1);
7909 RExC_parse = endbrace;
7910 RExC_end = orig_end;
7911 RExC_override_recoding = 0;
7913 nextchar(pRExC_state);
7923 * It returns the code point in utf8 for the value in *encp.
7924 * value: a code value in the source encoding
7925 * encp: a pointer to an Encode object
7927 * If the result from Encode is not a single character,
7928 * it returns U+FFFD (Replacement character) and sets *encp to NULL.
7931 S_reg_recode(pTHX_ const char value, SV **encp)
7934 SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP);
7935 const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv);
7936 const STRLEN newlen = SvCUR(sv);
7937 UV uv = UNICODE_REPLACEMENT;
7939 PERL_ARGS_ASSERT_REG_RECODE;
7943 ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
7946 if (!newlen || numlen != newlen) {
7947 uv = UNICODE_REPLACEMENT;
7955 - regatom - the lowest level
7957 Try to identify anything special at the start of the pattern. If there
7958 is, then handle it as required. This may involve generating a single regop,
7959 such as for an assertion; or it may involve recursing, such as to
7960 handle a () structure.
7962 If the string doesn't start with something special then we gobble up
7963 as much literal text as we can.
7965 Once we have been able to handle whatever type of thing started the
7966 sequence, we return.
7968 Note: we have to be careful with escapes, as they can be both literal
7969 and special, and in the case of \10 and friends can either, depending
7970 on context. Specifically there are two separate switches for handling
7971 escape sequences, with the one for handling literal escapes requiring
7972 a dummy entry for all of the special escapes that are actually handled
7977 S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth)
7980 register regnode *ret = NULL;
7982 char *parse_start = RExC_parse;
7984 GET_RE_DEBUG_FLAGS_DECL;
7985 DEBUG_PARSE("atom");
7986 *flagp = WORST; /* Tentatively. */
7988 PERL_ARGS_ASSERT_REGATOM;
7991 switch ((U8)*RExC_parse) {
7993 RExC_seen_zerolen++;
7994 nextchar(pRExC_state);
7995 if (RExC_flags & RXf_PMf_MULTILINE)
7996 ret = reg_node(pRExC_state, MBOL);
7997 else if (RExC_flags & RXf_PMf_SINGLELINE)
7998 ret = reg_node(pRExC_state, SBOL);
8000 ret = reg_node(pRExC_state, BOL);
8001 Set_Node_Length(ret, 1); /* MJD */
8004 nextchar(pRExC_state);
8006 RExC_seen_zerolen++;
8007 if (RExC_flags & RXf_PMf_MULTILINE)
8008 ret = reg_node(pRExC_state, MEOL);
8009 else if (RExC_flags & RXf_PMf_SINGLELINE)
8010 ret = reg_node(pRExC_state, SEOL);
8012 ret = reg_node(pRExC_state, EOL);
8013 Set_Node_Length(ret, 1); /* MJD */
8016 nextchar(pRExC_state);
8017 if (RExC_flags & RXf_PMf_SINGLELINE)
8018 ret = reg_node(pRExC_state, SANY);
8020 ret = reg_node(pRExC_state, REG_ANY);
8021 *flagp |= HASWIDTH|SIMPLE;
8023 Set_Node_Length(ret, 1); /* MJD */
8027 char * const oregcomp_parse = ++RExC_parse;
8028 ret = regclass(pRExC_state,depth+1);
8029 if (*RExC_parse != ']') {
8030 RExC_parse = oregcomp_parse;
8031 vFAIL("Unmatched [");
8033 nextchar(pRExC_state);
8034 *flagp |= HASWIDTH|SIMPLE;
8035 Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */
8039 nextchar(pRExC_state);
8040 ret = reg(pRExC_state, 1, &flags,depth+1);
8042 if (flags & TRYAGAIN) {
8043 if (RExC_parse == RExC_end) {
8044 /* Make parent create an empty node if needed. */
8052 *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED);
8056 if (flags & TRYAGAIN) {
8060 vFAIL("Internal urp");
8061 /* Supposed to be caught earlier. */
8064 if (!regcurly(RExC_parse)) {
8073 vFAIL("Quantifier follows nothing");
8078 This switch handles escape sequences that resolve to some kind
8079 of special regop and not to literal text. Escape sequnces that
8080 resolve to literal text are handled below in the switch marked
8083 Every entry in this switch *must* have a corresponding entry
8084 in the literal escape switch. However, the opposite is not
8085 required, as the default for this switch is to jump to the
8086 literal text handling code.
8088 switch ((U8)*++RExC_parse) {
8089 /* Special Escapes */
8091 RExC_seen_zerolen++;
8092 ret = reg_node(pRExC_state, SBOL);
8094 goto finish_meta_pat;
8096 ret = reg_node(pRExC_state, GPOS);
8097 RExC_seen |= REG_SEEN_GPOS;
8099 goto finish_meta_pat;
8101 RExC_seen_zerolen++;
8102 ret = reg_node(pRExC_state, KEEPS);
8104 /* XXX:dmq : disabling in-place substitution seems to
8105 * be necessary here to avoid cases of memory corruption, as
8106 * with: C<$_="x" x 80; s/x\K/y/> -- rgs
8108 RExC_seen |= REG_SEEN_LOOKBEHIND;
8109 goto finish_meta_pat;
8111 ret = reg_node(pRExC_state, SEOL);
8113 RExC_seen_zerolen++; /* Do not optimize RE away */
8114 goto finish_meta_pat;
8116 ret = reg_node(pRExC_state, EOS);
8118 RExC_seen_zerolen++; /* Do not optimize RE away */
8119 goto finish_meta_pat;
8121 ret = reg_node(pRExC_state, CANY);
8122 RExC_seen |= REG_SEEN_CANY;
8123 *flagp |= HASWIDTH|SIMPLE;
8124 goto finish_meta_pat;
8126 ret = reg_node(pRExC_state, CLUMP);
8128 goto finish_meta_pat;
8130 switch (get_regex_charset(RExC_flags)) {
8131 case REGEX_LOCALE_CHARSET:
8134 case REGEX_UNICODE_CHARSET:
8137 case REGEX_ASCII_RESTRICTED_CHARSET:
8138 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8141 case REGEX_DEPENDS_CHARSET:
8147 ret = reg_node(pRExC_state, op);
8148 *flagp |= HASWIDTH|SIMPLE;
8149 goto finish_meta_pat;
8151 switch (get_regex_charset(RExC_flags)) {
8152 case REGEX_LOCALE_CHARSET:
8155 case REGEX_UNICODE_CHARSET:
8158 case REGEX_ASCII_RESTRICTED_CHARSET:
8159 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8162 case REGEX_DEPENDS_CHARSET:
8168 ret = reg_node(pRExC_state, op);
8169 *flagp |= HASWIDTH|SIMPLE;
8170 goto finish_meta_pat;
8172 RExC_seen_zerolen++;
8173 RExC_seen |= REG_SEEN_LOOKBEHIND;
8174 switch (get_regex_charset(RExC_flags)) {
8175 case REGEX_LOCALE_CHARSET:
8178 case REGEX_UNICODE_CHARSET:
8181 case REGEX_ASCII_RESTRICTED_CHARSET:
8182 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8185 case REGEX_DEPENDS_CHARSET:
8191 ret = reg_node(pRExC_state, op);
8192 FLAGS(ret) = get_regex_charset(RExC_flags);
8194 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8195 ckWARNregdep(RExC_parse, "\"\\b{\" is deprecated; use \"\\b\\{\" instead");
8197 goto finish_meta_pat;
8199 RExC_seen_zerolen++;
8200 RExC_seen |= REG_SEEN_LOOKBEHIND;
8201 switch (get_regex_charset(RExC_flags)) {
8202 case REGEX_LOCALE_CHARSET:
8205 case REGEX_UNICODE_CHARSET:
8208 case REGEX_ASCII_RESTRICTED_CHARSET:
8209 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8212 case REGEX_DEPENDS_CHARSET:
8218 ret = reg_node(pRExC_state, op);
8219 FLAGS(ret) = get_regex_charset(RExC_flags);
8221 if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') {
8222 ckWARNregdep(RExC_parse, "\"\\B{\" is deprecated; use \"\\B\\{\" instead");
8224 goto finish_meta_pat;
8226 switch (get_regex_charset(RExC_flags)) {
8227 case REGEX_LOCALE_CHARSET:
8230 case REGEX_UNICODE_CHARSET:
8233 case REGEX_ASCII_RESTRICTED_CHARSET:
8234 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8237 case REGEX_DEPENDS_CHARSET:
8243 ret = reg_node(pRExC_state, op);
8244 *flagp |= HASWIDTH|SIMPLE;
8245 goto finish_meta_pat;
8247 switch (get_regex_charset(RExC_flags)) {
8248 case REGEX_LOCALE_CHARSET:
8251 case REGEX_UNICODE_CHARSET:
8254 case REGEX_ASCII_RESTRICTED_CHARSET:
8255 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8258 case REGEX_DEPENDS_CHARSET:
8264 ret = reg_node(pRExC_state, op);
8265 *flagp |= HASWIDTH|SIMPLE;
8266 goto finish_meta_pat;
8268 switch (get_regex_charset(RExC_flags)) {
8269 case REGEX_LOCALE_CHARSET:
8272 case REGEX_ASCII_RESTRICTED_CHARSET:
8273 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8276 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8277 case REGEX_UNICODE_CHARSET:
8283 ret = reg_node(pRExC_state, op);
8284 *flagp |= HASWIDTH|SIMPLE;
8285 goto finish_meta_pat;
8287 switch (get_regex_charset(RExC_flags)) {
8288 case REGEX_LOCALE_CHARSET:
8291 case REGEX_ASCII_RESTRICTED_CHARSET:
8292 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
8295 case REGEX_DEPENDS_CHARSET: /* No difference between these */
8296 case REGEX_UNICODE_CHARSET:
8302 ret = reg_node(pRExC_state, op);
8303 *flagp |= HASWIDTH|SIMPLE;
8304 goto finish_meta_pat;
8306 ret = reg_node(pRExC_state, LNBREAK);
8307 *flagp |= HASWIDTH|SIMPLE;
8308 goto finish_meta_pat;
8310 ret = reg_node(pRExC_state, HORIZWS);
8311 *flagp |= HASWIDTH|SIMPLE;
8312 goto finish_meta_pat;
8314 ret = reg_node(pRExC_state, NHORIZWS);
8315 *flagp |= HASWIDTH|SIMPLE;
8316 goto finish_meta_pat;
8318 ret = reg_node(pRExC_state, VERTWS);
8319 *flagp |= HASWIDTH|SIMPLE;
8320 goto finish_meta_pat;
8322 ret = reg_node(pRExC_state, NVERTWS);
8323 *flagp |= HASWIDTH|SIMPLE;
8325 nextchar(pRExC_state);
8326 Set_Node_Length(ret, 2); /* MJD */
8331 char* const oldregxend = RExC_end;
8333 char* parse_start = RExC_parse - 2;
8336 if (RExC_parse[1] == '{') {
8337 /* a lovely hack--pretend we saw [\pX] instead */
8338 RExC_end = strchr(RExC_parse, '}');
8340 const U8 c = (U8)*RExC_parse;
8342 RExC_end = oldregxend;
8343 vFAIL2("Missing right brace on \\%c{}", c);
8348 RExC_end = RExC_parse + 2;
8349 if (RExC_end > oldregxend)
8350 RExC_end = oldregxend;
8354 ret = regclass(pRExC_state,depth+1);
8356 RExC_end = oldregxend;
8359 Set_Node_Offset(ret, parse_start + 2);
8360 Set_Node_Cur_Length(ret);
8361 nextchar(pRExC_state);
8362 *flagp |= HASWIDTH|SIMPLE;
8366 /* Handle \N and \N{NAME} here and not below because it can be
8367 multicharacter. join_exact() will join them up later on.
8368 Also this makes sure that things like /\N{BLAH}+/ and
8369 \N{BLAH} being multi char Just Happen. dmq*/
8371 ret= reg_namedseq(pRExC_state, NULL, flagp, depth);
8373 case 'k': /* Handle \k<NAME> and \k'NAME' */
8376 char ch= RExC_parse[1];
8377 if (ch != '<' && ch != '\'' && ch != '{') {
8379 vFAIL2("Sequence %.2s... not terminated",parse_start);
8381 /* this pretty much dupes the code for (?P=...) in reg(), if
8382 you change this make sure you change that */
8383 char* name_start = (RExC_parse += 2);
8385 SV *sv_dat = reg_scan_name(pRExC_state,
8386 SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
8387 ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\'';
8388 if (RExC_parse == name_start || *RExC_parse != ch)
8389 vFAIL2("Sequence %.3s... not terminated",parse_start);
8392 num = add_data( pRExC_state, 1, "S" );
8393 RExC_rxi->data->data[num]=(void*)sv_dat;
8394 SvREFCNT_inc_simple_void(sv_dat);
8398 ret = reganode(pRExC_state,
8401 : (MORE_ASCII_RESTRICTED)
8403 : (AT_LEAST_UNI_SEMANTICS)
8411 /* override incorrect value set in reganode MJD */
8412 Set_Node_Offset(ret, parse_start+1);
8413 Set_Node_Cur_Length(ret); /* MJD */
8414 nextchar(pRExC_state);
8420 case '1': case '2': case '3': case '4':
8421 case '5': case '6': case '7': case '8': case '9':
8424 bool isg = *RExC_parse == 'g';
8429 if (*RExC_parse == '{') {
8433 if (*RExC_parse == '-') {
8437 if (hasbrace && !isDIGIT(*RExC_parse)) {
8438 if (isrel) RExC_parse--;
8440 goto parse_named_seq;
8442 num = atoi(RExC_parse);
8443 if (isg && num == 0)
8444 vFAIL("Reference to invalid group 0");
8446 num = RExC_npar - num;
8448 vFAIL("Reference to nonexistent or unclosed group");
8450 if (!isg && num > 9 && num >= RExC_npar)
8453 char * const parse_start = RExC_parse - 1; /* MJD */
8454 while (isDIGIT(*RExC_parse))
8456 if (parse_start == RExC_parse - 1)
8457 vFAIL("Unterminated \\g... pattern");
8459 if (*RExC_parse != '}')
8460 vFAIL("Unterminated \\g{...} pattern");
8464 if (num > (I32)RExC_rx->nparens)
8465 vFAIL("Reference to nonexistent group");
8468 ret = reganode(pRExC_state,
8471 : (MORE_ASCII_RESTRICTED)
8473 : (AT_LEAST_UNI_SEMANTICS)
8481 /* override incorrect value set in reganode MJD */
8482 Set_Node_Offset(ret, parse_start+1);
8483 Set_Node_Cur_Length(ret); /* MJD */
8485 nextchar(pRExC_state);
8490 if (RExC_parse >= RExC_end)
8491 FAIL("Trailing \\");
8494 /* Do not generate "unrecognized" warnings here, we fall
8495 back into the quick-grab loop below */
8502 if (RExC_flags & RXf_PMf_EXTENDED) {
8503 if ( reg_skipcomment( pRExC_state ) )
8510 parse_start = RExC_parse - 1;
8523 char_state latest_char_state = generic_char;
8524 register STRLEN len;
8529 U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
8530 regnode * orig_emit;
8533 orig_emit = RExC_emit; /* Save the original output node position in
8534 case we need to output a different node
8536 ret = reg_node(pRExC_state,
8537 (U8) ((! FOLD) ? EXACT
8540 : (MORE_ASCII_RESTRICTED)
8542 : (AT_LEAST_UNI_SEMANTICS)
8547 for (len = 0, p = RExC_parse - 1;
8548 len < 127 && p < RExC_end;
8551 char * const oldp = p;
8553 if (RExC_flags & RXf_PMf_EXTENDED)
8554 p = regwhite( pRExC_state, p );
8565 /* Literal Escapes Switch
8567 This switch is meant to handle escape sequences that
8568 resolve to a literal character.
8570 Every escape sequence that represents something
8571 else, like an assertion or a char class, is handled
8572 in the switch marked 'Special Escapes' above in this
8573 routine, but also has an entry here as anything that
8574 isn't explicitly mentioned here will be treated as
8575 an unescaped equivalent literal.
8579 /* These are all the special escapes. */
8580 case 'A': /* Start assertion */
8581 case 'b': case 'B': /* Word-boundary assertion*/
8582 case 'C': /* Single char !DANGEROUS! */
8583 case 'd': case 'D': /* digit class */
8584 case 'g': case 'G': /* generic-backref, pos assertion */
8585 case 'h': case 'H': /* HORIZWS */
8586 case 'k': case 'K': /* named backref, keep marker */
8587 case 'N': /* named char sequence */
8588 case 'p': case 'P': /* Unicode property */
8589 case 'R': /* LNBREAK */
8590 case 's': case 'S': /* space class */
8591 case 'v': case 'V': /* VERTWS */
8592 case 'w': case 'W': /* word class */
8593 case 'X': /* eXtended Unicode "combining character sequence" */
8594 case 'z': case 'Z': /* End of line/string assertion */
8598 /* Anything after here is an escape that resolves to a
8599 literal. (Except digits, which may or may not)
8618 ender = ASCII_TO_NATIVE('\033');
8622 ender = ASCII_TO_NATIVE('\007');
8627 STRLEN brace_len = len;
8629 const char* error_msg;
8631 bool valid = grok_bslash_o(p,
8638 RExC_parse = p; /* going to die anyway; point
8639 to exact spot of failure */
8646 if (PL_encoding && ender < 0x100) {
8647 goto recode_encoding;
8656 char* const e = strchr(p, '}');
8660 vFAIL("Missing right brace on \\x{}");
8663 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
8664 | PERL_SCAN_DISALLOW_PREFIX;
8665 STRLEN numlen = e - p - 1;
8666 ender = grok_hex(p + 1, &numlen, &flags, NULL);
8673 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
8675 ender = grok_hex(p, &numlen, &flags, NULL);
8678 if (PL_encoding && ender < 0x100)
8679 goto recode_encoding;
8683 ender = grok_bslash_c(*p++, UTF, SIZE_ONLY);
8685 case '0': case '1': case '2': case '3':case '4':
8686 case '5': case '6': case '7': case '8':case '9':
8688 (isDIGIT(p[1]) && atoi(p) >= RExC_npar))
8690 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
8692 ender = grok_oct(p, &numlen, &flags, NULL);
8702 if (PL_encoding && ender < 0x100)
8703 goto recode_encoding;
8706 if (! RExC_override_recoding) {
8707 SV* enc = PL_encoding;
8708 ender = reg_recode((const char)(U8)ender, &enc);
8709 if (!enc && SIZE_ONLY)
8710 ckWARNreg(p, "Invalid escape in the specified encoding");
8716 FAIL("Trailing \\");
8719 if (!SIZE_ONLY&& isALPHA(*p)) {
8720 /* Include any { following the alpha to emphasize
8721 * that it could be part of an escape at some point
8723 int len = (*(p + 1) == '{') ? 2 : 1;
8724 ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p);
8726 goto normal_default;
8731 if (UTF8_IS_START(*p) && UTF) {
8733 ender = utf8n_to_uvchr((U8*)p, RExC_end - p,
8734 &numlen, UTF8_ALLOW_DEFAULT);
8740 } /* End of switch on the literal */
8742 /* Certain characters are problematic because their folded
8743 * length is so different from their original length that it
8744 * isn't handleable by the optimizer. They are therefore not
8745 * placed in an EXACTish node; and are here handled specially.
8746 * (Even if the optimizer handled LATIN_SMALL_LETTER_SHARP_S,
8747 * putting it in a special node keeps regexec from having to
8748 * deal with a non-utf8 multi-char fold */
8750 && (ender > 255 || (! MORE_ASCII_RESTRICTED && ! LOC)))
8752 /* We look for either side of the fold. For example \xDF
8753 * folds to 'ss'. We look for both the single character
8754 * \xDF and the sequence 'ss'. When we find something that
8755 * could be one of those, we stop and flush whatever we
8756 * have output so far into the EXACTish node that was being
8757 * built. Then restore the input pointer to what it was.
8758 * regatom will return that EXACT node, and will be called
8759 * again, positioned so the first character is the one in
8760 * question, which we return in a different node type.
8761 * The multi-char folds are a sequence, so the occurrence
8762 * of the first character in that sequence doesn't
8763 * necessarily mean that what follows is the rest of the
8764 * sequence. We keep track of that with a state machine,
8765 * with the state being set to the latest character
8766 * processed before the current one. Most characters will
8767 * set the state to 0, but if one occurs that is part of a
8768 * potential tricky fold sequence, the state is set to that
8769 * character, and the next loop iteration sees if the state
8770 * should progress towards the final folded-from character,
8771 * or if it was a false alarm. If it turns out to be a
8772 * false alarm, the character(s) will be output in a new
8773 * EXACTish node, and join_exact() will later combine them.
8774 * In the case of the 'ss' sequence, which is more common
8775 * and more easily checked, some look-ahead is done to
8776 * save time by ruling-out some false alarms */
8779 latest_char_state = generic_char;
8783 case 0x17F: /* LATIN SMALL LETTER LONG S */
8784 if (AT_LEAST_UNI_SEMANTICS) {
8785 if (latest_char_state == char_s) { /* 'ss' */
8786 ender = LATIN_SMALL_LETTER_SHARP_S;
8789 else if (p < RExC_end) {
8791 /* Look-ahead at the next character. If it
8792 * is also an s, we handle as a sharp s
8793 * tricky regnode. */
8794 if (*p == 's' || *p == 'S') {
8796 /* But first flush anything in the
8797 * EXACTish buffer */
8802 p++; /* Account for swallowing this
8804 ender = LATIN_SMALL_LETTER_SHARP_S;
8807 /* Here, the next character is not a
8808 * literal 's', but still could
8809 * evaluate to one if part of a \o{},
8810 * \x or \OCTAL-DIGIT. The minimum
8811 * length required for that is 4, eg
8815 && (isDIGIT(*(p + 1))
8817 || *(p + 1) == 'o' ))
8820 /* Here, it could be an 's', too much
8821 * bother to figure it out here. Flush
8822 * the buffer if any; when come back
8823 * here, set the state so know that the
8824 * previous char was an 's' */
8826 latest_char_state = generic_char;
8830 latest_char_state = char_s;
8836 /* Here, can't be an 'ss' sequence, or at least not
8837 * one that could fold to/from the sharp ss */
8838 latest_char_state = generic_char;
8840 case 0x03C5: /* First char in upsilon series */
8841 case 0x03A5: /* Also capital UPSILON, which folds to
8842 03C5, and hence exhibits the same
8844 if (p < RExC_end - 4) { /* Need >= 4 bytes left */
8845 latest_char_state = upsilon_1;
8852 latest_char_state = generic_char;
8855 case 0x03B9: /* First char in iota series */
8856 case 0x0399: /* Also capital IOTA */
8857 case 0x1FBE: /* GREEK PROSGEGRAMMENI folds to 3B9 */
8858 case 0x0345: /* COMBINING GREEK YPOGEGRAMMENI folds
8860 if (p < RExC_end - 4) {
8861 latest_char_state = iota_1;
8868 latest_char_state = generic_char;
8872 if (latest_char_state == upsilon_1) {
8873 latest_char_state = upsilon_2;
8875 else if (latest_char_state == iota_1) {
8876 latest_char_state = iota_2;
8879 latest_char_state = generic_char;
8883 if (latest_char_state == upsilon_2) {
8884 ender = GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS;
8887 else if (latest_char_state == iota_2) {
8888 ender = GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS;
8891 latest_char_state = generic_char;
8894 /* These are the tricky fold characters. Flush any
8895 * buffer first. (When adding to this list, also should
8896 * add them to fold_grind.t to make sure get tested) */
8897 case GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS:
8898 case GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS:
8899 case LATIN_SMALL_LETTER_SHARP_S:
8900 case LATIN_CAPITAL_LETTER_SHARP_S:
8901 case 0x1FD3: /* GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA */
8902 case 0x1FE3: /* GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA */
8909 char* const oldregxend = RExC_end;
8910 U8 tmpbuf[UTF8_MAXBYTES+1];
8912 /* Here, we know we need to generate a special
8913 * regnode, and 'ender' contains the tricky
8914 * character. What's done is to pretend it's in a
8915 * [bracketed] class, and let the code that deals
8916 * with those handle it, as that code has all the
8917 * intelligence necessary. First save the current
8918 * parse state, get rid of the already allocated
8919 * but empty EXACT node that the ANYOFV node will
8920 * replace, and point the parse to a buffer which
8921 * we fill with the character we want the regclass
8922 * code to think is being parsed */
8923 RExC_emit = orig_emit;
8924 RExC_parse = (char *) tmpbuf;
8926 U8 *d = uvchr_to_utf8(tmpbuf, ender);
8928 RExC_end = (char *) d;
8930 else { /* ender above 255 already excluded */
8931 tmpbuf[0] = (U8) ender;
8933 RExC_end = RExC_parse + 1;
8936 ret = regclass(pRExC_state,depth+1);
8938 /* Here, have parsed the buffer. Reset the parse to
8939 * the actual input, and return */
8940 RExC_end = oldregxend;
8943 Set_Node_Offset(ret, RExC_parse);
8944 Set_Node_Cur_Length(ret);
8945 nextchar(pRExC_state);
8946 *flagp |= HASWIDTH|SIMPLE;
8952 if ( RExC_flags & RXf_PMf_EXTENDED)
8953 p = regwhite( pRExC_state, p );
8955 /* Prime the casefolded buffer. Locale rules, which apply
8956 * only to code points < 256, aren't known until execution,
8957 * so for them, just output the original character using
8959 if (LOC && ender < 256) {
8960 if (UNI_IS_INVARIANT(ender)) {
8961 *tmpbuf = (U8) ender;
8964 *tmpbuf = UTF8_TWO_BYTE_HI(ender);
8965 *(tmpbuf + 1) = UTF8_TWO_BYTE_LO(ender);
8969 else if (isASCII(ender)) { /* Note: Here can't also be LOC
8971 ender = toLOWER(ender);
8972 *tmpbuf = (U8) ender;
8975 else if (! MORE_ASCII_RESTRICTED && ! LOC) {
8977 /* Locale and /aa require more selectivity about the
8978 * fold, so are handled below. Otherwise, here, just
8980 ender = toFOLD_uni(ender, tmpbuf, &foldlen);
8983 /* Under locale rules or /aa we are not to mix,
8984 * respectively, ords < 256 or ASCII with non-. So
8985 * reject folds that mix them, using only the
8986 * non-folded code point. So do the fold to a
8987 * temporary, and inspect each character in it. */
8988 U8 trialbuf[UTF8_MAXBYTES_CASE+1];
8990 UV tmpender = toFOLD_uni(ender, trialbuf, &foldlen);
8991 U8* e = s + foldlen;
8992 bool fold_ok = TRUE;
8996 || (LOC && (UTF8_IS_INVARIANT(*s)
8997 || UTF8_IS_DOWNGRADEABLE_START(*s))))
9005 Copy(trialbuf, tmpbuf, foldlen, U8);
9009 uvuni_to_utf8(tmpbuf, ender);
9010 foldlen = UNISKIP(ender);
9014 if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */
9019 /* Emit all the Unicode characters. */
9021 for (foldbuf = tmpbuf;
9023 foldlen -= numlen) {
9024 ender = utf8_to_uvchr(foldbuf, &numlen);
9026 const STRLEN unilen = reguni(pRExC_state, ender, s);
9029 /* In EBCDIC the numlen
9030 * and unilen can differ. */
9032 if (numlen >= foldlen)
9036 break; /* "Can't happen." */
9040 const STRLEN unilen = reguni(pRExC_state, ender, s);
9049 REGC((char)ender, s++);
9055 /* Emit all the Unicode characters. */
9057 for (foldbuf = tmpbuf;
9059 foldlen -= numlen) {
9060 ender = utf8_to_uvchr(foldbuf, &numlen);
9062 const STRLEN unilen = reguni(pRExC_state, ender, s);
9065 /* In EBCDIC the numlen
9066 * and unilen can differ. */
9068 if (numlen >= foldlen)
9076 const STRLEN unilen = reguni(pRExC_state, ender, s);
9085 REGC((char)ender, s++);
9088 loopdone: /* Jumped to when encounters something that shouldn't be in
9091 Set_Node_Cur_Length(ret); /* MJD */
9092 nextchar(pRExC_state);
9094 /* len is STRLEN which is unsigned, need to copy to signed */
9097 vFAIL("Internal disaster");
9101 if (len == 1 && UNI_IS_INVARIANT(ender))
9105 RExC_size += STR_SZ(len);
9108 RExC_emit += STR_SZ(len);
9116 /* Jumped to when an unrecognized character set is encountered */
9118 Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags));
9123 S_regwhite( RExC_state_t *pRExC_state, char *p )
9125 const char *e = RExC_end;
9127 PERL_ARGS_ASSERT_REGWHITE;
9132 else if (*p == '#') {
9141 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
9149 /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]].
9150 Character classes ([:foo:]) can also be negated ([:^foo:]).
9151 Returns a named class id (ANYOF_XXX) if successful, -1 otherwise.
9152 Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed,
9153 but trigger failures because they are currently unimplemented. */
9155 #define POSIXCC_DONE(c) ((c) == ':')
9156 #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.')
9157 #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c))
9160 S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value)
9163 I32 namedclass = OOB_NAMEDCLASS;
9165 PERL_ARGS_ASSERT_REGPPOSIXCC;
9167 if (value == '[' && RExC_parse + 1 < RExC_end &&
9168 /* I smell either [: or [= or [. -- POSIX has been here, right? */
9169 POSIXCC(UCHARAT(RExC_parse))) {
9170 const char c = UCHARAT(RExC_parse);
9171 char* const s = RExC_parse++;
9173 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c)
9175 if (RExC_parse == RExC_end)
9176 /* Grandfather lone [:, [=, [. */
9179 const char* const t = RExC_parse++; /* skip over the c */
9182 if (UCHARAT(RExC_parse) == ']') {
9183 const char *posixcc = s + 1;
9184 RExC_parse++; /* skip over the ending ] */
9187 const I32 complement = *posixcc == '^' ? *posixcc++ : 0;
9188 const I32 skip = t - posixcc;
9190 /* Initially switch on the length of the name. */
9193 if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */
9194 namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM;
9197 /* Names all of length 5. */
9198 /* alnum alpha ascii blank cntrl digit graph lower
9199 print punct space upper */
9200 /* Offset 4 gives the best switch position. */
9201 switch (posixcc[4]) {
9203 if (memEQ(posixcc, "alph", 4)) /* alpha */
9204 namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA;
9207 if (memEQ(posixcc, "spac", 4)) /* space */
9208 namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC;
9211 if (memEQ(posixcc, "grap", 4)) /* graph */
9212 namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH;
9215 if (memEQ(posixcc, "asci", 4)) /* ascii */
9216 namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII;
9219 if (memEQ(posixcc, "blan", 4)) /* blank */
9220 namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK;
9223 if (memEQ(posixcc, "cntr", 4)) /* cntrl */
9224 namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL;
9227 if (memEQ(posixcc, "alnu", 4)) /* alnum */
9228 namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC;
9231 if (memEQ(posixcc, "lowe", 4)) /* lower */
9232 namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER;
9233 else if (memEQ(posixcc, "uppe", 4)) /* upper */
9234 namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER;
9237 if (memEQ(posixcc, "digi", 4)) /* digit */
9238 namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT;
9239 else if (memEQ(posixcc, "prin", 4)) /* print */
9240 namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT;
9241 else if (memEQ(posixcc, "punc", 4)) /* punct */
9242 namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT;
9247 if (memEQ(posixcc, "xdigit", 6))
9248 namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT;
9252 if (namedclass == OOB_NAMEDCLASS)
9253 Simple_vFAIL3("POSIX class [:%.*s:] unknown",
9255 assert (posixcc[skip] == ':');
9256 assert (posixcc[skip+1] == ']');
9257 } else if (!SIZE_ONLY) {
9258 /* [[=foo=]] and [[.foo.]] are still future. */
9260 /* adjust RExC_parse so the warning shows after
9262 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']')
9264 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9267 /* Maternal grandfather:
9268 * "[:" ending in ":" but not in ":]" */
9278 S_checkposixcc(pTHX_ RExC_state_t *pRExC_state)
9282 PERL_ARGS_ASSERT_CHECKPOSIXCC;
9284 if (POSIXCC(UCHARAT(RExC_parse))) {
9285 const char *s = RExC_parse;
9286 const char c = *s++;
9290 if (*s && c == *s && s[1] == ']') {
9292 "POSIX syntax [%c %c] belongs inside character classes",
9295 /* [[=foo=]] and [[.foo.]] are still future. */
9296 if (POSIXCC_NOTYET(c)) {
9297 /* adjust RExC_parse so the error shows after
9299 while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']')
9301 Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c);
9307 /* No locale test, and always Unicode semantics */
9308 #define _C_C_T_NOLOC_(NAME,TEST,WORD) \
9310 for (value = 0; value < 256; value++) \
9312 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9316 case ANYOF_N##NAME: \
9317 for (value = 0; value < 256; value++) \
9319 stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9324 /* Like the above, but there are differences if we are in uni-8-bit or not, so
9325 * there are two tests passed in, to use depending on that. There aren't any
9326 * cases where the label is different from the name, so no need for that
9328 #define _C_C_T_(NAME, TEST_8, TEST_7, WORD) \
9330 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \
9331 else if (UNI_SEMANTICS) { \
9332 for (value = 0; value < 256; value++) { \
9333 if (TEST_8(value)) stored += \
9334 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9338 for (value = 0; value < 128; value++) { \
9339 if (TEST_7(UNI_TO_NATIVE(value))) stored += \
9340 set_regclass_bit(pRExC_state, ret, \
9341 (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9347 case ANYOF_N##NAME: \
9348 if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \
9349 else if (UNI_SEMANTICS) { \
9350 for (value = 0; value < 256; value++) { \
9351 if (! TEST_8(value)) stored += \
9352 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \
9356 for (value = 0; value < 128; value++) { \
9357 if (! TEST_7(UNI_TO_NATIVE(value))) stored += set_regclass_bit( \
9358 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9360 if (AT_LEAST_ASCII_RESTRICTED) { \
9361 for (value = 128; value < 256; value++) { \
9362 stored += set_regclass_bit( \
9363 pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \
9365 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; \
9368 /* For a non-ut8 target string with DEPENDS semantics, all above \
9369 * ASCII Latin1 code points match the complement of any of the \
9370 * classes. But in utf8, they have their Unicode semantics, so \
9371 * can't just set them in the bitmap, or else regexec.c will think \
9372 * they matched when they shouldn't. */ \
9373 ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; \
9381 S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr)
9384 /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes.
9385 * Locale folding is done at run-time, so this function should not be
9386 * called for nodes that are for locales.
9388 * This function sets the bit corresponding to the fold of the input
9389 * 'value', if not already set. The fold of 'f' is 'F', and the fold of
9392 * It also knows about the characters that are in the bitmap that have
9393 * folds that are matchable only outside it, and sets the appropriate lists
9396 * It returns the number of bits that actually changed from 0 to 1 */
9401 PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD;
9403 fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value]
9406 /* It assumes the bit for 'value' has already been set */
9407 if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) {
9408 ANYOF_BITMAP_SET(node, fold);
9411 if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) {
9412 /* Certain Latin1 characters have matches outside the bitmap. To get
9413 * here, 'value' is one of those characters. None of these matches is
9414 * valid for ASCII characters under /aa, which have been excluded by
9415 * the 'if' above. The matches fall into three categories:
9416 * 1) They are singly folded-to or -from an above 255 character, as
9417 * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y
9419 * 2) They are part of a multi-char fold with another character in the
9420 * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill;
9421 * 3) They are part of a multi-char fold with a character not in the
9422 * bitmap, such as various ligatures.
9423 * We aren't dealing fully with multi-char folds, except we do deal
9424 * with the pattern containing a character that has a multi-char fold
9425 * (not so much the inverse).
9426 * For types 1) and 3), the matches only happen when the target string
9427 * is utf8; that's not true for 2), and we set a flag for it.
9429 * The code below adds to the passed in inversion list the single fold
9430 * closures for 'value'. The values are hard-coded here so that an
9431 * innocent-looking character class, like /[ks]/i won't have to go out
9432 * to disk to find the possible matches. XXX It would be better to
9433 * generate these via regen, in case a new version of the Unicode
9434 * standard adds new mappings, though that is not really likely. */
9439 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A);
9443 /* LATIN SMALL LETTER LONG S */
9444 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F);
9447 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9448 GREEK_SMALL_LETTER_MU);
9449 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9450 GREEK_CAPITAL_LETTER_MU);
9452 case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE:
9453 case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE:
9455 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B);
9456 if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */
9457 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9458 PL_fold_latin1[value]);
9461 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
9462 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9463 LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS);
9465 case LATIN_SMALL_LETTER_SHARP_S:
9466 *invlist_ptr = add_cp_to_invlist(*invlist_ptr,
9467 LATIN_CAPITAL_LETTER_SHARP_S);
9469 /* Under /a, /d, and /u, this can match the two chars "ss" */
9470 if (! MORE_ASCII_RESTRICTED) {
9471 add_alternate(alternate_ptr, (U8 *) "ss", 2);
9473 /* And under /u or /a, it can match even if the target is
9475 if (AT_LEAST_UNI_SEMANTICS) {
9476 ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8;
9490 /* These all are targets of multi-character folds from code
9491 * points that require UTF8 to express, so they can't match
9492 * unless the target string is in UTF-8, so no action here is
9493 * necessary, as regexec.c properly handles the general case
9494 * for UTF-8 matching */
9497 /* Use deprecated warning to increase the chances of this
9499 ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value);
9503 else if (DEPENDS_SEMANTICS
9505 && PL_fold_latin1[value] != value)
9507 /* Under DEPENDS rules, non-ASCII Latin1 characters match their
9508 * folds only when the target string is in UTF-8. We add the fold
9509 * here to the list of things to match outside the bitmap, which
9510 * won't be looked at unless it is UTF8 (or else if something else
9511 * says to look even if not utf8, but those things better not happen
9512 * under DEPENDS semantics. */
9513 *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]);
9520 PERL_STATIC_INLINE U8
9521 S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, HV** invlist_ptr, AV** alternate_ptr)
9523 /* This inline function sets a bit in the bitmap if not already set, and if
9524 * appropriate, its fold, returning the number of bits that actually
9525 * changed from 0 to 1 */
9529 PERL_ARGS_ASSERT_SET_REGCLASS_BIT;
9531 if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */
9535 ANYOF_BITMAP_SET(node, value);
9538 if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */
9539 stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr);
9546 S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len)
9548 /* Adds input 'string' with length 'len' to the ANYOF node's unicode
9549 * alternate list, pointed to by 'alternate_ptr'. This is an array of
9550 * the multi-character folds of characters in the node */
9553 PERL_ARGS_ASSERT_ADD_ALTERNATE;
9555 if (! *alternate_ptr) {
9556 *alternate_ptr = newAV();
9558 sv = newSVpvn_utf8((char*)string, len, TRUE);
9559 av_push(*alternate_ptr, sv);
9564 parse a class specification and produce either an ANYOF node that
9565 matches the pattern or perhaps will be optimized into an EXACTish node
9566 instead. The node contains a bit map for the first 256 characters, with the
9567 corresponding bit set if that character is in the list. For characters
9568 above 255, a range list is used */
9571 S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
9574 register UV nextvalue;
9575 register IV prevvalue = OOB_UNICODE;
9576 register IV range = 0;
9577 UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */
9578 register regnode *ret;
9581 char *rangebegin = NULL;
9582 bool need_class = 0;
9583 bool allow_full_fold = TRUE; /* Assume wants multi-char folding */
9585 STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more
9586 than just initialized. */
9589 /* code points this node matches that can't be stored in the bitmap */
9590 HV* nonbitmap = NULL;
9592 /* The items that are to match that aren't stored in the bitmap, but are a
9593 * result of things that are stored there. This is the fold closure of
9594 * such a character, either because it has DEPENDS semantics and shouldn't
9595 * be matched unless the target string is utf8, or is a code point that is
9596 * too large for the bit map, as for example, the fold of the MICRO SIGN is
9597 * above 255. This all is solely for performance reasons. By having this
9598 * code know the outside-the-bitmap folds that the bitmapped characters are
9599 * involved with, we don't have to go out to disk to find the list of
9600 * matches, unless the character class includes code points that aren't
9601 * storable in the bit map. That means that a character class with an 's'
9602 * in it, for example, doesn't need to go out to disk to find everything
9603 * that matches. A 2nd list is used so that the 'nonbitmap' list is kept
9604 * empty unless there is something whose fold we don't know about, and will
9605 * have to go out to the disk to find. */
9606 HV* l1_fold_invlist = NULL;
9608 /* List of multi-character folds that are matched by this node */
9609 AV* unicode_alternate = NULL;
9611 UV literal_endpoint = 0;
9613 UV stored = 0; /* how many chars stored in the bitmap */
9615 regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in
9616 case we need to change the emitted regop to an EXACT. */
9617 const char * orig_parse = RExC_parse;
9618 GET_RE_DEBUG_FLAGS_DECL;
9620 PERL_ARGS_ASSERT_REGCLASS;
9622 PERL_UNUSED_ARG(depth);
9625 DEBUG_PARSE("clas");
9627 /* Assume we are going to generate an ANYOF node. */
9628 ret = reganode(pRExC_state, ANYOF, 0);
9632 ANYOF_FLAGS(ret) = 0;
9635 if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */
9639 ANYOF_FLAGS(ret) |= ANYOF_INVERT;
9641 /* We have decided to not allow multi-char folds in inverted character
9642 * classes, due to the confusion that can happen, especially with
9643 * classes that are designed for a non-Unicode world: You have the
9644 * peculiar case that:
9645 "s s" =~ /^[^\xDF]+$/i => Y
9646 "ss" =~ /^[^\xDF]+$/i => N
9648 * See [perl #89750] */
9649 allow_full_fold = FALSE;
9653 RExC_size += ANYOF_SKIP;
9654 listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */
9657 RExC_emit += ANYOF_SKIP;
9659 ANYOF_FLAGS(ret) |= ANYOF_LOCALE;
9661 ANYOF_BITMAP_ZERO(ret);
9662 listsv = newSVpvs("# comment\n");
9663 initial_listsv_len = SvCUR(listsv);
9666 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9668 if (!SIZE_ONLY && POSIXCC(nextvalue))
9669 checkposixcc(pRExC_state);
9671 /* allow 1st char to be ] (allowing it to be - is dealt with later) */
9672 if (UCHARAT(RExC_parse) == ']')
9676 while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
9680 namedclass = OOB_NAMEDCLASS; /* initialize as illegal */
9683 rangebegin = RExC_parse;
9685 value = utf8n_to_uvchr((U8*)RExC_parse,
9686 RExC_end - RExC_parse,
9687 &numlen, UTF8_ALLOW_DEFAULT);
9688 RExC_parse += numlen;
9691 value = UCHARAT(RExC_parse++);
9693 nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0;
9694 if (value == '[' && POSIXCC(nextvalue))
9695 namedclass = regpposixcc(pRExC_state, value);
9696 else if (value == '\\') {
9698 value = utf8n_to_uvchr((U8*)RExC_parse,
9699 RExC_end - RExC_parse,
9700 &numlen, UTF8_ALLOW_DEFAULT);
9701 RExC_parse += numlen;
9704 value = UCHARAT(RExC_parse++);
9705 /* Some compilers cannot handle switching on 64-bit integer
9706 * values, therefore value cannot be an UV. Yes, this will
9707 * be a problem later if we want switch on Unicode.
9708 * A similar issue a little bit later when switching on
9709 * namedclass. --jhi */
9710 switch ((I32)value) {
9711 case 'w': namedclass = ANYOF_ALNUM; break;
9712 case 'W': namedclass = ANYOF_NALNUM; break;
9713 case 's': namedclass = ANYOF_SPACE; break;
9714 case 'S': namedclass = ANYOF_NSPACE; break;
9715 case 'd': namedclass = ANYOF_DIGIT; break;
9716 case 'D': namedclass = ANYOF_NDIGIT; break;
9717 case 'v': namedclass = ANYOF_VERTWS; break;
9718 case 'V': namedclass = ANYOF_NVERTWS; break;
9719 case 'h': namedclass = ANYOF_HORIZWS; break;
9720 case 'H': namedclass = ANYOF_NHORIZWS; break;
9721 case 'N': /* Handle \N{NAME} in class */
9723 /* We only pay attention to the first char of
9724 multichar strings being returned. I kinda wonder
9725 if this makes sense as it does change the behaviour
9726 from earlier versions, OTOH that behaviour was broken
9728 UV v; /* value is register so we cant & it /grrr */
9729 if (reg_namedseq(pRExC_state, &v, NULL, depth)) {
9739 if (RExC_parse >= RExC_end)
9740 vFAIL2("Empty \\%c{}", (U8)value);
9741 if (*RExC_parse == '{') {
9742 const U8 c = (U8)value;
9743 e = strchr(RExC_parse++, '}');
9745 vFAIL2("Missing right brace on \\%c{}", c);
9746 while (isSPACE(UCHARAT(RExC_parse)))
9748 if (e == RExC_parse)
9749 vFAIL2("Empty \\%c{}", c);
9751 while (isSPACE(UCHARAT(RExC_parse + n - 1)))
9759 if (UCHARAT(RExC_parse) == '^') {
9762 value = value == 'p' ? 'P' : 'p'; /* toggle */
9763 while (isSPACE(UCHARAT(RExC_parse))) {
9769 /* Add the property name to the list. If /i matching, give
9770 * a different name which consists of the normal name
9771 * sandwiched between two underscores and '_i'. The design
9772 * is discussed in the commit message for this. */
9773 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s%.*s%s\n",
9774 (value=='p' ? '+' : '!'),
9783 /* The \p could match something in the Latin1 range, hence
9784 * something that isn't utf8 */
9785 ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8;
9786 namedclass = ANYOF_MAX; /* no official name, but it's named */
9788 /* \p means they want Unicode semantics */
9789 RExC_uni_semantics = 1;
9792 case 'n': value = '\n'; break;
9793 case 'r': value = '\r'; break;
9794 case 't': value = '\t'; break;
9795 case 'f': value = '\f'; break;
9796 case 'b': value = '\b'; break;
9797 case 'e': value = ASCII_TO_NATIVE('\033');break;
9798 case 'a': value = ASCII_TO_NATIVE('\007');break;
9800 RExC_parse--; /* function expects to be pointed at the 'o' */
9802 const char* error_msg;
9803 bool valid = grok_bslash_o(RExC_parse,
9808 RExC_parse += numlen;
9813 if (PL_encoding && value < 0x100) {
9814 goto recode_encoding;
9818 if (*RExC_parse == '{') {
9819 I32 flags = PERL_SCAN_ALLOW_UNDERSCORES
9820 | PERL_SCAN_DISALLOW_PREFIX;
9821 char * const e = strchr(RExC_parse++, '}');
9823 vFAIL("Missing right brace on \\x{}");
9825 numlen = e - RExC_parse;
9826 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
9830 I32 flags = PERL_SCAN_DISALLOW_PREFIX;
9832 value = grok_hex(RExC_parse, &numlen, &flags, NULL);
9833 RExC_parse += numlen;
9835 if (PL_encoding && value < 0x100)
9836 goto recode_encoding;
9839 value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY);
9841 case '0': case '1': case '2': case '3': case '4':
9842 case '5': case '6': case '7':
9844 /* Take 1-3 octal digits */
9845 I32 flags = PERL_SCAN_SILENT_ILLDIGIT;
9847 value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
9848 RExC_parse += numlen;
9849 if (PL_encoding && value < 0x100)
9850 goto recode_encoding;
9854 if (! RExC_override_recoding) {
9855 SV* enc = PL_encoding;
9856 value = reg_recode((const char)(U8)value, &enc);
9857 if (!enc && SIZE_ONLY)
9858 ckWARNreg(RExC_parse,
9859 "Invalid escape in the specified encoding");
9863 /* Allow \_ to not give an error */
9864 if (!SIZE_ONLY && isALNUM(value) && value != '_') {
9865 ckWARN2reg(RExC_parse,
9866 "Unrecognized escape \\%c in character class passed through",
9871 } /* end of \blah */
9877 if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */
9879 /* What matches in a locale is not known until runtime, so need to
9880 * (one time per class) allocate extra space to pass to regexec.
9881 * The space will contain a bit for each named class that is to be
9882 * matched against. This isn't needed for \p{} and pseudo-classes,
9883 * as they are not affected by locale, and hence are dealt with
9885 if (LOC && namedclass < ANYOF_MAX && ! need_class) {
9888 RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP;
9891 RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP;
9892 ANYOF_CLASS_ZERO(ret);
9894 ANYOF_FLAGS(ret) |= ANYOF_CLASS;
9897 /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a
9898 * literal, as is the character that began the false range, i.e.
9899 * the 'a' in the examples */
9903 RExC_parse >= rangebegin ?
9904 RExC_parse - rangebegin : 0;
9905 ckWARN4reg(RExC_parse,
9906 "False [] range \"%*.*s\"",
9910 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
9911 if (prevvalue < 256) {
9913 set_regclass_bit(pRExC_state, ret, (U8) prevvalue, &l1_fold_invlist, &unicode_alternate);
9916 nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue);
9920 range = 0; /* this was not a true range */
9926 const char *what = NULL;
9929 /* Possible truncation here but in some 64-bit environments
9930 * the compiler gets heartburn about switch on 64-bit values.
9931 * A similar issue a little earlier when switching on value.
9933 switch ((I32)namedclass) {
9935 case _C_C_T_(ALNUMC, isALNUMC_L1, isALNUMC, "XPosixAlnum");
9936 case _C_C_T_(ALPHA, isALPHA_L1, isALPHA, "XPosixAlpha");
9937 case _C_C_T_(BLANK, isBLANK_L1, isBLANK, "XPosixBlank");
9938 case _C_C_T_(CNTRL, isCNTRL_L1, isCNTRL, "XPosixCntrl");
9939 case _C_C_T_(GRAPH, isGRAPH_L1, isGRAPH, "XPosixGraph");
9940 case _C_C_T_(LOWER, isLOWER_L1, isLOWER, "XPosixLower");
9941 case _C_C_T_(PRINT, isPRINT_L1, isPRINT, "XPosixPrint");
9942 case _C_C_T_(PSXSPC, isPSXSPC_L1, isPSXSPC, "XPosixSpace");
9943 case _C_C_T_(PUNCT, isPUNCT_L1, isPUNCT, "XPosixPunct");
9944 case _C_C_T_(UPPER, isUPPER_L1, isUPPER, "XPosixUpper");
9945 /* \s, \w match all unicode if utf8. */
9946 case _C_C_T_(SPACE, isSPACE_L1, isSPACE, "SpacePerl");
9947 case _C_C_T_(ALNUM, isWORDCHAR_L1, isALNUM, "Word");
9948 case _C_C_T_(XDIGIT, isXDIGIT_L1, isXDIGIT, "XPosixXDigit");
9949 case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace");
9950 case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace");
9953 ANYOF_CLASS_SET(ret, ANYOF_ASCII);
9955 for (value = 0; value < 128; value++)
9957 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
9960 what = NULL; /* Doesn't match outside ascii, so
9961 don't want to add +utf8:: */
9965 ANYOF_CLASS_SET(ret, ANYOF_NASCII);
9967 for (value = 128; value < 256; value++)
9969 set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate);
9971 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
9977 ANYOF_CLASS_SET(ret, ANYOF_DIGIT);
9979 /* consecutive digits assumed */
9980 for (value = '0'; value <= '9'; value++)
9982 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9989 ANYOF_CLASS_SET(ret, ANYOF_NDIGIT);
9991 /* consecutive digits assumed */
9992 for (value = 0; value < '0'; value++)
9994 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
9995 for (value = '9' + 1; value < 256; value++)
9997 set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate);
10001 if (AT_LEAST_ASCII_RESTRICTED ) {
10002 ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL;
10006 /* this is to handle \p and \P */
10009 vFAIL("Invalid [::] class");
10012 if (what && ! (AT_LEAST_ASCII_RESTRICTED)) {
10013 /* Strings such as "+utf8::isWord\n" */
10014 Perl_sv_catpvf(aTHX_ listsv, "%cutf8::Is%s\n", yesno, what);
10019 } /* end of namedclass \blah */
10022 if (prevvalue > (IV)value) /* b-a */ {
10023 const int w = RExC_parse - rangebegin;
10024 Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin);
10025 range = 0; /* not a valid range */
10029 prevvalue = value; /* save the beginning of the range */
10030 if (RExC_parse+1 < RExC_end
10031 && *RExC_parse == '-'
10032 && RExC_parse[1] != ']')
10036 /* a bad range like \w-, [:word:]- ? */
10037 if (namedclass > OOB_NAMEDCLASS) {
10038 if (ckWARN(WARN_REGEXP)) {
10040 RExC_parse >= rangebegin ?
10041 RExC_parse - rangebegin : 0;
10043 "False [] range \"%*.*s\"",
10048 set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate);
10050 range = 1; /* yeah, it's a range! */
10051 continue; /* but do it the next time */
10055 /* non-Latin1 code point implies unicode semantics. Must be set in
10056 * pass1 so is there for the whole of pass 2 */
10058 RExC_uni_semantics = 1;
10061 /* now is the next time */
10063 if (prevvalue < 256) {
10064 const IV ceilvalue = value < 256 ? value : 255;
10067 /* In EBCDIC [\x89-\x91] should include
10068 * the \x8e but [i-j] should not. */
10069 if (literal_endpoint == 2 &&
10070 ((isLOWER(prevvalue) && isLOWER(ceilvalue)) ||
10071 (isUPPER(prevvalue) && isUPPER(ceilvalue))))
10073 if (isLOWER(prevvalue)) {
10074 for (i = prevvalue; i <= ceilvalue; i++)
10075 if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10077 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10080 for (i = prevvalue; i <= ceilvalue; i++)
10081 if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) {
10083 set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10089 for (i = prevvalue; i <= ceilvalue; i++) {
10090 stored += set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate);
10094 const UV prevnatvalue = NATIVE_TO_UNI(prevvalue);
10095 const UV natvalue = NATIVE_TO_UNI(value);
10096 nonbitmap = add_range_to_invlist(nonbitmap, prevnatvalue, natvalue);
10099 literal_endpoint = 0;
10103 range = 0; /* this range (if it was one) is done now */
10110 /****** !SIZE_ONLY AFTER HERE *********/
10112 /* If folding and there are code points above 255, we calculate all
10113 * characters that could fold to or from the ones already on the list */
10114 if (FOLD && nonbitmap) {
10117 HV* fold_intersection;
10120 /* This is a list of all the characters that participate in folds
10121 * (except marks, etc in multi-char folds */
10122 if (! PL_utf8_foldable) {
10123 SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0);
10124 PL_utf8_foldable = _swash_to_invlist(swash);
10127 /* This is a hash that for a particular fold gives all characters
10128 * that are involved in it */
10129 if (! PL_utf8_foldclosures) {
10131 /* If we were unable to find any folds, then we likely won't be
10132 * able to find the closures. So just create an empty list.
10133 * Folding will effectively be restricted to the non-Unicode rules
10134 * hard-coded into Perl. (This case happens legitimately during
10135 * compilation of Perl itself before the Unicode tables are
10137 if (invlist_len(PL_utf8_foldable) == 0) {
10138 PL_utf8_foldclosures = _new_invlist(0);
10140 /* If the folds haven't been read in, call a fold function
10142 if (! PL_utf8_tofold) {
10143 U8 dummy[UTF8_MAXBYTES+1];
10145 to_utf8_fold((U8*) "A", dummy, &dummy_len);
10147 PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold);
10151 /* Only the characters in this class that participate in folds need
10152 * be checked. Get the intersection of this class and all the
10153 * possible characters that are foldable. This can quickly narrow
10154 * down a large class */
10155 fold_intersection = invlist_intersection(PL_utf8_foldable, nonbitmap);
10157 /* Now look at the foldable characters in this class individually */
10158 fold_list = invlist_array(fold_intersection);
10159 for (i = 0; i < invlist_len(fold_intersection); i++) {
10162 /* The next entry is the beginning of the range that is in the
10164 UV start = fold_list[i++];
10167 /* The next entry is the beginning of the next range, which
10168 * isn't in the class, so the end of the current range is one
10169 * less than that */
10170 UV end = fold_list[i] - 1;
10172 /* Look at every character in the range */
10173 for (j = start; j <= end; j++) {
10176 U8 foldbuf[UTF8_MAXBYTES_CASE+1];
10179 _to_uni_fold_flags(j, foldbuf, &foldlen, allow_full_fold);
10181 if (foldlen > (STRLEN)UNISKIP(f)) {
10183 /* Any multicharacter foldings (disallowed in
10184 * lookbehind patterns) require the following
10185 * transform: [ABCDEF] -> (?:[ABCabcDEFd]|pq|rst) where
10186 * E folds into "pq" and F folds into "rst", all other
10187 * characters fold to single characters. We save away
10188 * these multicharacter foldings, to be later saved as
10189 * part of the additional "s" data. */
10190 if (! RExC_in_lookbehind) {
10192 U8* e = foldbuf + foldlen;
10194 /* If any of the folded characters of this are in
10195 * the Latin1 range, tell the regex engine that
10196 * this can match a non-utf8 target string. The
10197 * only multi-byte fold whose source is in the
10198 * Latin1 range (U+00DF) applies only when the
10199 * target string is utf8, or under unicode rules */
10200 if (j > 255 || AT_LEAST_UNI_SEMANTICS) {
10203 /* Can't mix ascii with non- under /aa */
10204 if (MORE_ASCII_RESTRICTED
10205 && (isASCII(*loc) != isASCII(j)))
10207 goto end_multi_fold;
10209 if (UTF8_IS_INVARIANT(*loc)
10210 || UTF8_IS_DOWNGRADEABLE_START(*loc))
10212 /* Can't mix above and below 256 under
10215 goto end_multi_fold;
10218 |= ANYOF_NONBITMAP_NON_UTF8;
10221 loc += UTF8SKIP(loc);
10225 add_alternate(&unicode_alternate, foldbuf, foldlen);
10229 /* This is special-cased, as it is the only letter which
10230 * has both a multi-fold and single-fold in Latin1. All
10231 * the other chars that have single and multi-folds are
10232 * always in utf8, and the utf8 folding algorithm catches
10234 if (! LOC && j == LATIN_CAPITAL_LETTER_SHARP_S) {
10235 stored += set_regclass_bit(pRExC_state,
10237 LATIN_SMALL_LETTER_SHARP_S,
10238 &l1_fold_invlist, &unicode_alternate);
10242 /* Single character fold. Add everything in its fold
10243 * closure to the list that this node should match */
10246 /* The fold closures data structure is a hash with the
10247 * keys being every character that is folded to, like
10248 * 'k', and the values each an array of everything that
10249 * folds to its key. e.g. [ 'k', 'K', KELVIN_SIGN ] */
10250 if ((listp = hv_fetch(PL_utf8_foldclosures,
10251 (char *) foldbuf, foldlen, FALSE)))
10253 AV* list = (AV*) *listp;
10255 for (k = 0; k <= av_len(list); k++) {
10256 SV** c_p = av_fetch(list, k, FALSE);
10259 Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure");
10263 /* /aa doesn't allow folds between ASCII and
10264 * non-; /l doesn't allow them between above
10266 if ((MORE_ASCII_RESTRICTED
10267 && (isASCII(c) != isASCII(j)))
10268 || (LOC && ((c < 256) != (j < 256))))
10273 if (c < 256 && AT_LEAST_UNI_SEMANTICS) {
10274 stored += set_regclass_bit(pRExC_state,
10277 &l1_fold_invlist, &unicode_alternate);
10279 /* It may be that the code point is already
10280 * in this range or already in the bitmap,
10281 * in which case we need do nothing */
10282 else if ((c < start || c > end)
10284 || ! ANYOF_BITMAP_TEST(ret, c)))
10286 nonbitmap = add_cp_to_invlist(nonbitmap, c);
10293 invlist_destroy(fold_intersection);
10296 /* Combine the two lists into one. */
10297 if (l1_fold_invlist) {
10299 HV* temp = invlist_union(nonbitmap, l1_fold_invlist);
10300 invlist_destroy(nonbitmap);
10302 invlist_destroy(l1_fold_invlist);
10305 nonbitmap = l1_fold_invlist;
10309 /* Here, we have calculated what code points should be in the character
10310 * class. Now we can see about various optimizations. Fold calculation
10311 * needs to take place before inversion. Otherwise /[^k]/i would invert to
10312 * include K, which under /i would match k. */
10314 /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't
10315 * set the FOLD flag yet, so this this does optimize those. It doesn't
10316 * optimize locale. Doing so perhaps could be done as long as there is
10317 * nothing like \w in it; some thought also would have to be given to the
10318 * interaction with above 0x100 chars */
10320 && (ANYOF_FLAGS(ret) & ANYOF_FLAGS_ALL) == ANYOF_INVERT
10321 && ! unicode_alternate
10323 && SvCUR(listsv) == initial_listsv_len)
10325 for (value = 0; value < ANYOF_BITMAP_SIZE; ++value)
10326 ANYOF_BITMAP(ret)[value] ^= 0xFF;
10327 stored = 256 - stored;
10329 /* The inversion means that everything above 255 is matched; and at the
10330 * same time we clear the invert flag */
10331 ANYOF_FLAGS(ret) = ANYOF_UNICODE_ALL;
10334 /* Folding in the bitmap is taken care of above, but not for locale (for
10335 * which we have to wait to see what folding is in effect at runtime), and
10336 * for things not in the bitmap. Set run-time fold flag for these */
10337 if (FOLD && (LOC || nonbitmap || unicode_alternate)) {
10338 ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD;
10341 /* A single character class can be "optimized" into an EXACTish node.
10342 * Note that since we don't currently count how many characters there are
10343 * outside the bitmap, we are XXX missing optimization possibilities for
10344 * them. This optimization can't happen unless this is a truly single
10345 * character class, which means that it can't be an inversion into a
10346 * many-character class, and there must be no possibility of there being
10347 * things outside the bitmap. 'stored' (only) for locales doesn't include
10348 * \w, etc, so have to make a special test that they aren't present
10350 * Similarly A 2-character class of the very special form like [bB] can be
10351 * optimized into an EXACTFish node, but only for non-locales, and for
10352 * characters which only have the two folds; so things like 'fF' and 'Ii'
10353 * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE
10356 && ! unicode_alternate
10357 && SvCUR(listsv) == initial_listsv_len
10358 && ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL))
10359 && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10360 || (! ANYOF_CLASS_TEST_ANY_SET(ret)))))
10361 || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE))
10362 && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value))
10363 /* If the latest code point has a fold whose
10364 * bit is set, it must be the only other one */
10365 && ((prevvalue = PL_fold_latin1[value]) != (IV)value)
10366 && ANYOF_BITMAP_TEST(ret, prevvalue)))))
10368 /* Note that the information needed to decide to do this optimization
10369 * is not currently available until the 2nd pass, and that the actually
10370 * used EXACTish node takes less space than the calculated ANYOF node,
10371 * and hence the amount of space calculated in the first pass is larger
10372 * than actually used, so this optimization doesn't gain us any space.
10373 * But an EXACT node is faster than an ANYOF node, and can be combined
10374 * with any adjacent EXACT nodes later by the optimizer for further
10375 * gains. The speed of executing an EXACTF is similar to an ANYOF
10376 * node, so the optimization advantage comes from the ability to join
10377 * it to adjacent EXACT nodes */
10379 const char * cur_parse= RExC_parse;
10381 RExC_emit = (regnode *)orig_emit;
10382 RExC_parse = (char *)orig_parse;
10386 /* A locale node with one point can be folded; all the other cases
10387 * with folding will have two points, since we calculate them above
10389 if (ANYOF_FLAGS(ret) & ANYOF_LOC_NONBITMAP_FOLD) {
10395 } /* else 2 chars in the bit map: the folds of each other */
10396 else if (AT_LEAST_UNI_SEMANTICS || !isASCII(value)) {
10398 /* To join adjacent nodes, they must be the exact EXACTish type.
10399 * Try to use the most likely type, by using EXACTFU if the regex
10400 * calls for them, or is required because the character is
10404 else { /* Otherwise, more likely to be EXACTF type */
10408 ret = reg_node(pRExC_state, op);
10409 RExC_parse = (char *)cur_parse;
10410 if (UTF && ! NATIVE_IS_INVARIANT(value)) {
10411 *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value);
10412 *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value);
10414 RExC_emit += STR_SZ(2);
10417 *STRING(ret)= (char)value;
10419 RExC_emit += STR_SZ(1);
10421 SvREFCNT_dec(listsv);
10426 UV* nonbitmap_array = invlist_array(nonbitmap);
10427 UV nonbitmap_len = invlist_len(nonbitmap);
10430 /* Here have the full list of items to match that aren't in the
10431 * bitmap. Convert to the structure that the rest of the code is
10432 * expecting. XXX That rest of the code should convert to this
10434 for (i = 0; i < nonbitmap_len; i++) {
10436 /* The next entry is the beginning of the range that is in the
10438 UV start = nonbitmap_array[i++];
10441 /* The next entry is the beginning of the next range, which isn't
10442 * in the class, so the end of the current range is one less than
10443 * that. But if there is no next range, it means that the range
10444 * begun by 'start' extends to infinity, which for this platform
10445 * ends at UV_MAX */
10446 if (i == nonbitmap_len) {
10450 end = nonbitmap_array[i] - 1;
10453 if (start == end) {
10454 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\n", start);
10457 /* The \t sets the whole range */
10458 Perl_sv_catpvf(aTHX_ listsv, "%04"UVxf"\t%04"UVxf"\n",
10463 invlist_destroy(nonbitmap);
10466 if (SvCUR(listsv) == initial_listsv_len && ! unicode_alternate) {
10467 ARG_SET(ret, ANYOF_NONBITMAP_EMPTY);
10468 SvREFCNT_dec(listsv);
10469 SvREFCNT_dec(unicode_alternate);
10473 AV * const av = newAV();
10475 /* The 0th element stores the character class description
10476 * in its textual form: used later (regexec.c:Perl_regclass_swash())
10477 * to initialize the appropriate swash (which gets stored in
10478 * the 1st element), and also useful for dumping the regnode.
10479 * The 2nd element stores the multicharacter foldings,
10480 * used later (regexec.c:S_reginclass()). */
10481 av_store(av, 0, listsv);
10482 av_store(av, 1, NULL);
10484 /* Store any computed multi-char folds only if we are allowing
10486 if (allow_full_fold) {
10487 av_store(av, 2, MUTABLE_SV(unicode_alternate));
10488 if (unicode_alternate) { /* This node is variable length */
10493 av_store(av, 2, NULL);
10495 rv = newRV_noinc(MUTABLE_SV(av));
10496 n = add_data(pRExC_state, 1, "s");
10497 RExC_rxi->data->data[n] = (void*)rv;
10505 /* reg_skipcomment()
10507 Absorbs an /x style # comments from the input stream.
10508 Returns true if there is more text remaining in the stream.
10509 Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment
10510 terminates the pattern without including a newline.
10512 Note its the callers responsibility to ensure that we are
10513 actually in /x mode
10518 S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state)
10522 PERL_ARGS_ASSERT_REG_SKIPCOMMENT;
10524 while (RExC_parse < RExC_end)
10525 if (*RExC_parse++ == '\n') {
10530 /* we ran off the end of the pattern without ending
10531 the comment, so we have to add an \n when wrapping */
10532 RExC_seen |= REG_SEEN_RUN_ON_COMMENT;
10540 Advances the parse position, and optionally absorbs
10541 "whitespace" from the inputstream.
10543 Without /x "whitespace" means (?#...) style comments only,
10544 with /x this means (?#...) and # comments and whitespace proper.
10546 Returns the RExC_parse point from BEFORE the scan occurs.
10548 This is the /x friendly way of saying RExC_parse++.
10552 S_nextchar(pTHX_ RExC_state_t *pRExC_state)
10554 char* const retval = RExC_parse++;
10556 PERL_ARGS_ASSERT_NEXTCHAR;
10559 if (*RExC_parse == '(' && RExC_parse[1] == '?' &&
10560 RExC_parse[2] == '#') {
10561 while (*RExC_parse != ')') {
10562 if (RExC_parse == RExC_end)
10563 FAIL("Sequence (?#... not terminated");
10569 if (RExC_flags & RXf_PMf_EXTENDED) {
10570 if (isSPACE(*RExC_parse)) {
10574 else if (*RExC_parse == '#') {
10575 if ( reg_skipcomment( pRExC_state ) )
10584 - reg_node - emit a node
10586 STATIC regnode * /* Location. */
10587 S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op)
10590 register regnode *ptr;
10591 regnode * const ret = RExC_emit;
10592 GET_RE_DEBUG_FLAGS_DECL;
10594 PERL_ARGS_ASSERT_REG_NODE;
10597 SIZE_ALIGN(RExC_size);
10601 if (RExC_emit >= RExC_emit_bound)
10602 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10604 NODE_ALIGN_FILL(ret);
10606 FILL_ADVANCE_NODE(ptr, op);
10607 #ifdef RE_TRACK_PATTERN_OFFSETS
10608 if (RExC_offsets) { /* MJD */
10609 MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n",
10610 "reg_node", __LINE__,
10612 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0]
10613 ? "Overwriting end of array!\n" : "OK",
10614 (UV)(RExC_emit - RExC_emit_start),
10615 (UV)(RExC_parse - RExC_start),
10616 (UV)RExC_offsets[0]));
10617 Set_Node_Offset(RExC_emit, RExC_parse + (op == END));
10625 - reganode - emit a node with an argument
10627 STATIC regnode * /* Location. */
10628 S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg)
10631 register regnode *ptr;
10632 regnode * const ret = RExC_emit;
10633 GET_RE_DEBUG_FLAGS_DECL;
10635 PERL_ARGS_ASSERT_REGANODE;
10638 SIZE_ALIGN(RExC_size);
10643 assert(2==regarglen[op]+1);
10645 Anything larger than this has to allocate the extra amount.
10646 If we changed this to be:
10648 RExC_size += (1 + regarglen[op]);
10650 then it wouldn't matter. Its not clear what side effect
10651 might come from that so its not done so far.
10656 if (RExC_emit >= RExC_emit_bound)
10657 Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op);
10659 NODE_ALIGN_FILL(ret);
10661 FILL_ADVANCE_NODE_ARG(ptr, op, arg);
10662 #ifdef RE_TRACK_PATTERN_OFFSETS
10663 if (RExC_offsets) { /* MJD */
10664 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10668 (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ?
10669 "Overwriting end of array!\n" : "OK",
10670 (UV)(RExC_emit - RExC_emit_start),
10671 (UV)(RExC_parse - RExC_start),
10672 (UV)RExC_offsets[0]));
10673 Set_Cur_Node_Offset;
10681 - reguni - emit (if appropriate) a Unicode character
10684 S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s)
10688 PERL_ARGS_ASSERT_REGUNI;
10690 return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s);
10694 - reginsert - insert an operator in front of already-emitted operand
10696 * Means relocating the operand.
10699 S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
10702 register regnode *src;
10703 register regnode *dst;
10704 register regnode *place;
10705 const int offset = regarglen[(U8)op];
10706 const int size = NODE_STEP_REGNODE + offset;
10707 GET_RE_DEBUG_FLAGS_DECL;
10709 PERL_ARGS_ASSERT_REGINSERT;
10710 PERL_UNUSED_ARG(depth);
10711 /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
10712 DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]);
10721 if (RExC_open_parens) {
10723 /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/
10724 for ( paren=0 ; paren < RExC_npar ; paren++ ) {
10725 if ( RExC_open_parens[paren] >= opnd ) {
10726 /*DEBUG_PARSE_FMT("open"," - %d",size);*/
10727 RExC_open_parens[paren] += size;
10729 /*DEBUG_PARSE_FMT("open"," - %s","ok");*/
10731 if ( RExC_close_parens[paren] >= opnd ) {
10732 /*DEBUG_PARSE_FMT("close"," - %d",size);*/
10733 RExC_close_parens[paren] += size;
10735 /*DEBUG_PARSE_FMT("close"," - %s","ok");*/
10740 while (src > opnd) {
10741 StructCopy(--src, --dst, regnode);
10742 #ifdef RE_TRACK_PATTERN_OFFSETS
10743 if (RExC_offsets) { /* MJD 20010112 */
10744 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n",
10748 (UV)(dst - RExC_emit_start) > RExC_offsets[0]
10749 ? "Overwriting end of array!\n" : "OK",
10750 (UV)(src - RExC_emit_start),
10751 (UV)(dst - RExC_emit_start),
10752 (UV)RExC_offsets[0]));
10753 Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src));
10754 Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src));
10760 place = opnd; /* Op node, where operand used to be. */
10761 #ifdef RE_TRACK_PATTERN_OFFSETS
10762 if (RExC_offsets) { /* MJD */
10763 MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n",
10767 (UV)(place - RExC_emit_start) > RExC_offsets[0]
10768 ? "Overwriting end of array!\n" : "OK",
10769 (UV)(place - RExC_emit_start),
10770 (UV)(RExC_parse - RExC_start),
10771 (UV)RExC_offsets[0]));
10772 Set_Node_Offset(place, RExC_parse);
10773 Set_Node_Length(place, 1);
10776 src = NEXTOPER(place);
10777 FILL_ADVANCE_NODE(place, op);
10778 Zero(src, offset, regnode);
10782 - regtail - set the next-pointer at the end of a node chain of p to val.
10783 - SEE ALSO: regtail_study
10785 /* TODO: All three parms should be const */
10787 S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
10790 register regnode *scan;
10791 GET_RE_DEBUG_FLAGS_DECL;
10793 PERL_ARGS_ASSERT_REGTAIL;
10795 PERL_UNUSED_ARG(depth);
10801 /* Find last node. */
10804 regnode * const temp = regnext(scan);
10806 SV * const mysv=sv_newmortal();
10807 DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
10808 regprop(RExC_rx, mysv, scan);
10809 PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
10810 SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
10811 (temp == NULL ? "->" : ""),
10812 (temp == NULL ? PL_reg_name[OP(val)] : "")
10820 if (reg_off_by_arg[OP(scan)]) {
10821 ARG_SET(scan, val - scan);
10824 NEXT_OFF(scan) = val - scan;
10830 - regtail_study - set the next-pointer at the end of a node chain of p to val.
10831 - Look for optimizable sequences at the same time.
10832 - currently only looks for EXACT chains.
10834 This is experimental code. The idea is to use this routine to perform
10835 in place optimizations on branches and groups as they are constructed,
10836 with the long term intention of removing optimization from study_chunk so
10837 that it is purely analytical.
10839 Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used
10840 to control which is which.
10843 /* TODO: All four parms should be const */
10846 S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth)
10849 register regnode *scan;
10851 #ifdef EXPERIMENTAL_INPLACESCAN
10854 GET_RE_DEBUG_FLAGS_DECL;
10856 PERL_ARGS_ASSERT_REGTAIL_STUDY;
10862 /* Find last node. */
10866 regnode * const temp = regnext(scan);
10867 #ifdef EXPERIMENTAL_INPLACESCAN
10868 if (PL_regkind[OP(scan)] == EXACT)
10869 if (join_exact(pRExC_state,scan,&min,1,val,depth+1))
10873 switch (OP(scan)) {
10879 if( exact == PSEUDO )
10881 else if ( exact != OP(scan) )
10890 SV * const mysv=sv_newmortal();
10891 DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
10892 regprop(RExC_rx, mysv, scan);
10893 PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
10894 SvPV_nolen_const(mysv),
10895 REG_NODE_NUM(scan),
10896 PL_reg_name[exact]);
10903 SV * const mysv_val=sv_newmortal();
10904 DEBUG_PARSE_MSG("");
10905 regprop(RExC_rx, mysv_val, val);
10906 PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
10907 SvPV_nolen_const(mysv_val),
10908 (IV)REG_NODE_NUM(val),
10912 if (reg_off_by_arg[OP(scan)]) {
10913 ARG_SET(scan, val - scan);
10916 NEXT_OFF(scan) = val - scan;
10924 - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form
10928 S_regdump_extflags(pTHX_ const char *lead, const U32 flags)
10934 for (bit=0; bit<32; bit++) {
10935 if (flags & (1<<bit)) {
10936 if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */
10939 if (!set++ && lead)
10940 PerlIO_printf(Perl_debug_log, "%s",lead);
10941 PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]);
10944 if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) {
10945 if (!set++ && lead) {
10946 PerlIO_printf(Perl_debug_log, "%s",lead);
10949 case REGEX_UNICODE_CHARSET:
10950 PerlIO_printf(Perl_debug_log, "UNICODE");
10952 case REGEX_LOCALE_CHARSET:
10953 PerlIO_printf(Perl_debug_log, "LOCALE");
10955 case REGEX_ASCII_RESTRICTED_CHARSET:
10956 PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED");
10958 case REGEX_ASCII_MORE_RESTRICTED_CHARSET:
10959 PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED");
10962 PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET");
10968 PerlIO_printf(Perl_debug_log, "\n");
10970 PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead);
10976 Perl_regdump(pTHX_ const regexp *r)
10980 SV * const sv = sv_newmortal();
10981 SV *dsv= sv_newmortal();
10982 RXi_GET_DECL(r,ri);
10983 GET_RE_DEBUG_FLAGS_DECL;
10985 PERL_ARGS_ASSERT_REGDUMP;
10987 (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
10989 /* Header fields of interest. */
10990 if (r->anchored_substr) {
10991 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
10992 RE_SV_DUMPLEN(r->anchored_substr), 30);
10993 PerlIO_printf(Perl_debug_log,
10994 "anchored %s%s at %"IVdf" ",
10995 s, RE_SV_TAIL(r->anchored_substr),
10996 (IV)r->anchored_offset);
10997 } else if (r->anchored_utf8) {
10998 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
10999 RE_SV_DUMPLEN(r->anchored_utf8), 30);
11000 PerlIO_printf(Perl_debug_log,
11001 "anchored utf8 %s%s at %"IVdf" ",
11002 s, RE_SV_TAIL(r->anchored_utf8),
11003 (IV)r->anchored_offset);
11005 if (r->float_substr) {
11006 RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
11007 RE_SV_DUMPLEN(r->float_substr), 30);
11008 PerlIO_printf(Perl_debug_log,
11009 "floating %s%s at %"IVdf"..%"UVuf" ",
11010 s, RE_SV_TAIL(r->float_substr),
11011 (IV)r->float_min_offset, (UV)r->float_max_offset);
11012 } else if (r->float_utf8) {
11013 RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
11014 RE_SV_DUMPLEN(r->float_utf8), 30);
11015 PerlIO_printf(Perl_debug_log,
11016 "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
11017 s, RE_SV_TAIL(r->float_utf8),
11018 (IV)r->float_min_offset, (UV)r->float_max_offset);
11020 if (r->check_substr || r->check_utf8)
11021 PerlIO_printf(Perl_debug_log,
11023 (r->check_substr == r->float_substr
11024 && r->check_utf8 == r->float_utf8
11025 ? "(checking floating" : "(checking anchored"));
11026 if (r->extflags & RXf_NOSCAN)
11027 PerlIO_printf(Perl_debug_log, " noscan");
11028 if (r->extflags & RXf_CHECK_ALL)
11029 PerlIO_printf(Perl_debug_log, " isall");
11030 if (r->check_substr || r->check_utf8)
11031 PerlIO_printf(Perl_debug_log, ") ");
11033 if (ri->regstclass) {
11034 regprop(r, sv, ri->regstclass);
11035 PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
11037 if (r->extflags & RXf_ANCH) {
11038 PerlIO_printf(Perl_debug_log, "anchored");
11039 if (r->extflags & RXf_ANCH_BOL)
11040 PerlIO_printf(Perl_debug_log, "(BOL)");
11041 if (r->extflags & RXf_ANCH_MBOL)
11042 PerlIO_printf(Perl_debug_log, "(MBOL)");
11043 if (r->extflags & RXf_ANCH_SBOL)
11044 PerlIO_printf(Perl_debug_log, "(SBOL)");
11045 if (r->extflags & RXf_ANCH_GPOS)
11046 PerlIO_printf(Perl_debug_log, "(GPOS)");
11047 PerlIO_putc(Perl_debug_log, ' ');
11049 if (r->extflags & RXf_GPOS_SEEN)
11050 PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
11051 if (r->intflags & PREGf_SKIP)
11052 PerlIO_printf(Perl_debug_log, "plus ");
11053 if (r->intflags & PREGf_IMPLICIT)
11054 PerlIO_printf(Perl_debug_log, "implicit ");
11055 PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
11056 if (r->extflags & RXf_EVAL_SEEN)
11057 PerlIO_printf(Perl_debug_log, "with eval ");
11058 PerlIO_printf(Perl_debug_log, "\n");
11059 DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags));
11061 PERL_ARGS_ASSERT_REGDUMP;
11062 PERL_UNUSED_CONTEXT;
11063 PERL_UNUSED_ARG(r);
11064 #endif /* DEBUGGING */
11068 - regprop - printable representation of opcode
11070 #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \
11073 Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \
11074 if (flags & ANYOF_INVERT) \
11075 /*make sure the invert info is in each */ \
11076 sv_catpvs(sv, "^"); \
11082 Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o)
11087 RXi_GET_DECL(prog,progi);
11088 GET_RE_DEBUG_FLAGS_DECL;
11090 PERL_ARGS_ASSERT_REGPROP;
11094 if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
11095 /* It would be nice to FAIL() here, but this may be called from
11096 regexec.c, and it would be hard to supply pRExC_state. */
11097 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
11098 sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */
11100 k = PL_regkind[OP(o)];
11103 sv_catpvs(sv, " ");
11104 /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
11105 * is a crude hack but it may be the best for now since
11106 * we have no flag "this EXACTish node was UTF-8"
11108 pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1],
11109 PERL_PV_ESCAPE_UNI_DETECT |
11110 PERL_PV_ESCAPE_NONASCII |
11111 PERL_PV_PRETTY_ELLIPSES |
11112 PERL_PV_PRETTY_LTGT |
11113 PERL_PV_PRETTY_NOCLEAR
11115 } else if (k == TRIE) {
11116 /* print the details of the trie in dumpuntil instead, as
11117 * progi->data isn't available here */
11118 const char op = OP(o);
11119 const U32 n = ARG(o);
11120 const reg_ac_data * const ac = IS_TRIE_AC(op) ?
11121 (reg_ac_data *)progi->data->data[n] :
11123 const reg_trie_data * const trie
11124 = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
11126 Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]);
11127 DEBUG_TRIE_COMPILE_r(
11128 Perl_sv_catpvf(aTHX_ sv,
11129 "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
11130 (UV)trie->startstate,
11131 (IV)trie->statecount-1, /* -1 because of the unused 0 element */
11132 (UV)trie->wordcount,
11135 (UV)TRIE_CHARCOUNT(trie),
11136 (UV)trie->uniquecharcount
11139 if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
11141 int rangestart = -1;
11142 U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
11143 sv_catpvs(sv, "[");
11144 for (i = 0; i <= 256; i++) {
11145 if (i < 256 && BITMAP_TEST(bitmap,i)) {
11146 if (rangestart == -1)
11148 } else if (rangestart != -1) {
11149 if (i <= rangestart + 3)
11150 for (; rangestart < i; rangestart++)
11151 put_byte(sv, rangestart);
11153 put_byte(sv, rangestart);
11154 sv_catpvs(sv, "-");
11155 put_byte(sv, i - 1);
11160 sv_catpvs(sv, "]");
11163 } else if (k == CURLY) {
11164 if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
11165 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
11166 Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o));
11168 else if (k == WHILEM && o->flags) /* Ordinal/of */
11169 Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
11170 else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) {
11171 Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
11172 if ( RXp_PAREN_NAMES(prog) ) {
11173 if ( k != REF || (OP(o) < NREF)) {
11174 AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]);
11175 SV **name= av_fetch(list, ARG(o), 0 );
11177 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11180 AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]);
11181 SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]);
11182 I32 *nums=(I32*)SvPVX(sv_dat);
11183 SV **name= av_fetch(list, nums[0], 0 );
11186 for ( n=0; n<SvIVX(sv_dat); n++ ) {
11187 Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf,
11188 (n ? "," : ""), (IV)nums[n]);
11190 Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name));
11194 } else if (k == GOSUB)
11195 Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
11196 else if (k == VERB) {
11198 Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
11199 SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ]))));
11200 } else if (k == LOGICAL)
11201 Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
11202 else if (k == FOLDCHAR)
11203 Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) );
11204 else if (k == ANYOF) {
11205 int i, rangestart = -1;
11206 const U8 flags = ANYOF_FLAGS(o);
11209 /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */
11210 static const char * const anyofs[] = {
11243 if (flags & ANYOF_LOCALE)
11244 sv_catpvs(sv, "{loc}");
11245 if (flags & ANYOF_LOC_NONBITMAP_FOLD)
11246 sv_catpvs(sv, "{i}");
11247 Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]);
11248 if (flags & ANYOF_INVERT)
11249 sv_catpvs(sv, "^");
11251 /* output what the standard cp 0-255 bitmap matches */
11252 for (i = 0; i <= 256; i++) {
11253 if (i < 256 && ANYOF_BITMAP_TEST(o,i)) {
11254 if (rangestart == -1)
11256 } else if (rangestart != -1) {
11257 if (i <= rangestart + 3)
11258 for (; rangestart < i; rangestart++)
11259 put_byte(sv, rangestart);
11261 put_byte(sv, rangestart);
11262 sv_catpvs(sv, "-");
11263 put_byte(sv, i - 1);
11270 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11271 /* output any special charclass tests (used entirely under use locale) */
11272 if (ANYOF_CLASS_TEST_ANY_SET(o))
11273 for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++)
11274 if (ANYOF_CLASS_TEST(o,i)) {
11275 sv_catpv(sv, anyofs[i]);
11279 EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags);
11281 if (flags & ANYOF_NON_UTF8_LATIN1_ALL) {
11282 sv_catpvs(sv, "{non-utf8-latin1-all}");
11285 /* output information about the unicode matching */
11286 if (flags & ANYOF_UNICODE_ALL)
11287 sv_catpvs(sv, "{unicode_all}");
11288 else if (ANYOF_NONBITMAP(o))
11289 sv_catpvs(sv, "{unicode}");
11290 if (flags & ANYOF_NONBITMAP_NON_UTF8)
11291 sv_catpvs(sv, "{outside bitmap}");
11293 if (ANYOF_NONBITMAP(o)) {
11295 SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0);
11299 U8 s[UTF8_MAXBYTES_CASE+1];
11301 for (i = 0; i <= 256; i++) { /* just the first 256 */
11302 uvchr_to_utf8(s, i);
11304 if (i < 256 && swash_fetch(sw, s, TRUE)) {
11305 if (rangestart == -1)
11307 } else if (rangestart != -1) {
11308 if (i <= rangestart + 3)
11309 for (; rangestart < i; rangestart++) {
11310 const U8 * const e = uvchr_to_utf8(s,rangestart);
11312 for(p = s; p < e; p++)
11316 const U8 *e = uvchr_to_utf8(s,rangestart);
11318 for (p = s; p < e; p++)
11320 sv_catpvs(sv, "-");
11321 e = uvchr_to_utf8(s, i-1);
11322 for (p = s; p < e; p++)
11329 sv_catpvs(sv, "..."); /* et cetera */
11333 char *s = savesvpv(lv);
11334 char * const origs = s;
11336 while (*s && *s != '\n')
11340 const char * const t = ++s;
11358 Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]);
11360 else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH))
11361 Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags));
11363 PERL_UNUSED_CONTEXT;
11364 PERL_UNUSED_ARG(sv);
11365 PERL_UNUSED_ARG(o);
11366 PERL_UNUSED_ARG(prog);
11367 #endif /* DEBUGGING */
11371 Perl_re_intuit_string(pTHX_ REGEXP * const r)
11372 { /* Assume that RE_INTUIT is set */
11374 struct regexp *const prog = (struct regexp *)SvANY(r);
11375 GET_RE_DEBUG_FLAGS_DECL;
11377 PERL_ARGS_ASSERT_RE_INTUIT_STRING;
11378 PERL_UNUSED_CONTEXT;
11382 const char * const s = SvPV_nolen_const(prog->check_substr
11383 ? prog->check_substr : prog->check_utf8);
11385 if (!PL_colorset) reginitcolors();
11386 PerlIO_printf(Perl_debug_log,
11387 "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n",
11389 prog->check_substr ? "" : "utf8 ",
11390 PL_colors[5],PL_colors[0],
11393 (strlen(s) > 60 ? "..." : ""));
11396 return prog->check_substr ? prog->check_substr : prog->check_utf8;
11402 handles refcounting and freeing the perl core regexp structure. When
11403 it is necessary to actually free the structure the first thing it
11404 does is call the 'free' method of the regexp_engine associated to
11405 the regexp, allowing the handling of the void *pprivate; member
11406 first. (This routine is not overridable by extensions, which is why
11407 the extensions free is called first.)
11409 See regdupe and regdupe_internal if you change anything here.
11411 #ifndef PERL_IN_XSUB_RE
11413 Perl_pregfree(pTHX_ REGEXP *r)
11419 Perl_pregfree2(pTHX_ REGEXP *rx)
11422 struct regexp *const r = (struct regexp *)SvANY(rx);
11423 GET_RE_DEBUG_FLAGS_DECL;
11425 PERL_ARGS_ASSERT_PREGFREE2;
11427 if (r->mother_re) {
11428 ReREFCNT_dec(r->mother_re);
11430 CALLREGFREE_PVT(rx); /* free the private data */
11431 SvREFCNT_dec(RXp_PAREN_NAMES(r));
11434 SvREFCNT_dec(r->anchored_substr);
11435 SvREFCNT_dec(r->anchored_utf8);
11436 SvREFCNT_dec(r->float_substr);
11437 SvREFCNT_dec(r->float_utf8);
11438 Safefree(r->substrs);
11440 RX_MATCH_COPY_FREE(rx);
11441 #ifdef PERL_OLD_COPY_ON_WRITE
11442 SvREFCNT_dec(r->saved_copy);
11449 This is a hacky workaround to the structural issue of match results
11450 being stored in the regexp structure which is in turn stored in
11451 PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern
11452 could be PL_curpm in multiple contexts, and could require multiple
11453 result sets being associated with the pattern simultaneously, such
11454 as when doing a recursive match with (??{$qr})
11456 The solution is to make a lightweight copy of the regexp structure
11457 when a qr// is returned from the code executed by (??{$qr}) this
11458 lightweight copy doesn't actually own any of its data except for
11459 the starp/end and the actual regexp structure itself.
11465 Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx)
11467 struct regexp *ret;
11468 struct regexp *const r = (struct regexp *)SvANY(rx);
11469 register const I32 npar = r->nparens+1;
11471 PERL_ARGS_ASSERT_REG_TEMP_COPY;
11474 ret_x = (REGEXP*) newSV_type(SVt_REGEXP);
11475 ret = (struct regexp *)SvANY(ret_x);
11477 (void)ReREFCNT_inc(rx);
11478 /* We can take advantage of the existing "copied buffer" mechanism in SVs
11479 by pointing directly at the buffer, but flagging that the allocated
11480 space in the copy is zero. As we've just done a struct copy, it's now
11481 a case of zero-ing that, rather than copying the current length. */
11482 SvPV_set(ret_x, RX_WRAPPED(rx));
11483 SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8);
11484 memcpy(&(ret->xpv_cur), &(r->xpv_cur),
11485 sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur));
11486 SvLEN_set(ret_x, 0);
11487 SvSTASH_set(ret_x, NULL);
11488 SvMAGIC_set(ret_x, NULL);
11489 Newx(ret->offs, npar, regexp_paren_pair);
11490 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11492 Newx(ret->substrs, 1, struct reg_substr_data);
11493 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11495 SvREFCNT_inc_void(ret->anchored_substr);
11496 SvREFCNT_inc_void(ret->anchored_utf8);
11497 SvREFCNT_inc_void(ret->float_substr);
11498 SvREFCNT_inc_void(ret->float_utf8);
11500 /* check_substr and check_utf8, if non-NULL, point to either their
11501 anchored or float namesakes, and don't hold a second reference. */
11503 RX_MATCH_COPIED_off(ret_x);
11504 #ifdef PERL_OLD_COPY_ON_WRITE
11505 ret->saved_copy = NULL;
11507 ret->mother_re = rx;
11513 /* regfree_internal()
11515 Free the private data in a regexp. This is overloadable by
11516 extensions. Perl takes care of the regexp structure in pregfree(),
11517 this covers the *pprivate pointer which technically perl doesn't
11518 know about, however of course we have to handle the
11519 regexp_internal structure when no extension is in use.
11521 Note this is called before freeing anything in the regexp
11526 Perl_regfree_internal(pTHX_ REGEXP * const rx)
11529 struct regexp *const r = (struct regexp *)SvANY(rx);
11530 RXi_GET_DECL(r,ri);
11531 GET_RE_DEBUG_FLAGS_DECL;
11533 PERL_ARGS_ASSERT_REGFREE_INTERNAL;
11539 SV *dsv= sv_newmortal();
11540 RE_PV_QUOTED_DECL(s, RX_UTF8(rx),
11541 dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60);
11542 PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
11543 PL_colors[4],PL_colors[5],s);
11546 #ifdef RE_TRACK_PATTERN_OFFSETS
11548 Safefree(ri->u.offsets); /* 20010421 MJD */
11551 int n = ri->data->count;
11552 PAD* new_comppad = NULL;
11557 /* If you add a ->what type here, update the comment in regcomp.h */
11558 switch (ri->data->what[n]) {
11563 SvREFCNT_dec(MUTABLE_SV(ri->data->data[n]));
11566 Safefree(ri->data->data[n]);
11569 new_comppad = MUTABLE_AV(ri->data->data[n]);
11572 if (new_comppad == NULL)
11573 Perl_croak(aTHX_ "panic: pregfree comppad");
11574 PAD_SAVE_LOCAL(old_comppad,
11575 /* Watch out for global destruction's random ordering. */
11576 (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
11579 refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
11582 op_free((OP_4tree*)ri->data->data[n]);
11584 PAD_RESTORE_LOCAL(old_comppad);
11585 SvREFCNT_dec(MUTABLE_SV(new_comppad));
11586 new_comppad = NULL;
11591 { /* Aho Corasick add-on structure for a trie node.
11592 Used in stclass optimization only */
11594 reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
11596 refcount = --aho->refcount;
11599 PerlMemShared_free(aho->states);
11600 PerlMemShared_free(aho->fail);
11601 /* do this last!!!! */
11602 PerlMemShared_free(ri->data->data[n]);
11603 PerlMemShared_free(ri->regstclass);
11609 /* trie structure. */
11611 reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
11613 refcount = --trie->refcount;
11616 PerlMemShared_free(trie->charmap);
11617 PerlMemShared_free(trie->states);
11618 PerlMemShared_free(trie->trans);
11620 PerlMemShared_free(trie->bitmap);
11622 PerlMemShared_free(trie->jump);
11623 PerlMemShared_free(trie->wordinfo);
11624 /* do this last!!!! */
11625 PerlMemShared_free(ri->data->data[n]);
11630 Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
11633 Safefree(ri->data->what);
11634 Safefree(ri->data);
11640 #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t))
11641 #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t))
11642 #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
11645 re_dup - duplicate a regexp.
11647 This routine is expected to clone a given regexp structure. It is only
11648 compiled under USE_ITHREADS.
11650 After all of the core data stored in struct regexp is duplicated
11651 the regexp_engine.dupe method is used to copy any private data
11652 stored in the *pprivate pointer. This allows extensions to handle
11653 any duplication it needs to do.
11655 See pregfree() and regfree_internal() if you change anything here.
11657 #if defined(USE_ITHREADS)
11658 #ifndef PERL_IN_XSUB_RE
11660 Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param)
11664 const struct regexp *r = (const struct regexp *)SvANY(sstr);
11665 struct regexp *ret = (struct regexp *)SvANY(dstr);
11667 PERL_ARGS_ASSERT_RE_DUP_GUTS;
11669 npar = r->nparens+1;
11670 Newx(ret->offs, npar, regexp_paren_pair);
11671 Copy(r->offs, ret->offs, npar, regexp_paren_pair);
11673 /* no need to copy these */
11674 Newx(ret->swap, npar, regexp_paren_pair);
11677 if (ret->substrs) {
11678 /* Do it this way to avoid reading from *r after the StructCopy().
11679 That way, if any of the sv_dup_inc()s dislodge *r from the L1
11680 cache, it doesn't matter. */
11681 const bool anchored = r->check_substr
11682 ? r->check_substr == r->anchored_substr
11683 : r->check_utf8 == r->anchored_utf8;
11684 Newx(ret->substrs, 1, struct reg_substr_data);
11685 StructCopy(r->substrs, ret->substrs, struct reg_substr_data);
11687 ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param);
11688 ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param);
11689 ret->float_substr = sv_dup_inc(ret->float_substr, param);
11690 ret->float_utf8 = sv_dup_inc(ret->float_utf8, param);
11692 /* check_substr and check_utf8, if non-NULL, point to either their
11693 anchored or float namesakes, and don't hold a second reference. */
11695 if (ret->check_substr) {
11697 assert(r->check_utf8 == r->anchored_utf8);
11698 ret->check_substr = ret->anchored_substr;
11699 ret->check_utf8 = ret->anchored_utf8;
11701 assert(r->check_substr == r->float_substr);
11702 assert(r->check_utf8 == r->float_utf8);
11703 ret->check_substr = ret->float_substr;
11704 ret->check_utf8 = ret->float_utf8;
11706 } else if (ret->check_utf8) {
11708 ret->check_utf8 = ret->anchored_utf8;
11710 ret->check_utf8 = ret->float_utf8;
11715 RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param);
11718 RXi_SET(ret,CALLREGDUPE_PVT(dstr,param));
11720 if (RX_MATCH_COPIED(dstr))
11721 ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen);
11723 ret->subbeg = NULL;
11724 #ifdef PERL_OLD_COPY_ON_WRITE
11725 ret->saved_copy = NULL;
11728 if (ret->mother_re) {
11729 if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) {
11730 /* Our storage points directly to our mother regexp, but that's
11731 1: a buffer in a different thread
11732 2: something we no longer hold a reference on
11733 so we need to copy it locally. */
11734 /* Note we need to sue SvCUR() on our mother_re, because it, in
11735 turn, may well be pointing to its own mother_re. */
11736 SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re),
11737 SvCUR(ret->mother_re)+1));
11738 SvLEN_set(dstr, SvCUR(ret->mother_re)+1);
11740 ret->mother_re = NULL;
11744 #endif /* PERL_IN_XSUB_RE */
11749 This is the internal complement to regdupe() which is used to copy
11750 the structure pointed to by the *pprivate pointer in the regexp.
11751 This is the core version of the extension overridable cloning hook.
11752 The regexp structure being duplicated will be copied by perl prior
11753 to this and will be provided as the regexp *r argument, however
11754 with the /old/ structures pprivate pointer value. Thus this routine
11755 may override any copying normally done by perl.
11757 It returns a pointer to the new regexp_internal structure.
11761 Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param)
11764 struct regexp *const r = (struct regexp *)SvANY(rx);
11765 regexp_internal *reti;
11767 RXi_GET_DECL(r,ri);
11769 PERL_ARGS_ASSERT_REGDUPE_INTERNAL;
11773 Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal);
11774 Copy(ri->program, reti->program, len+1, regnode);
11777 reti->regstclass = NULL;
11780 struct reg_data *d;
11781 const int count = ri->data->count;
11784 Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
11785 char, struct reg_data);
11786 Newx(d->what, count, U8);
11789 for (i = 0; i < count; i++) {
11790 d->what[i] = ri->data->what[i];
11791 switch (d->what[i]) {
11792 /* legal options are one of: sSfpontTua
11793 see also regcomp.h and pregfree() */
11794 case 'a': /* actually an AV, but the dup function is identical. */
11797 case 'p': /* actually an AV, but the dup function is identical. */
11798 case 'u': /* actually an HV, but the dup function is identical. */
11799 d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param);
11802 /* This is cheating. */
11803 Newx(d->data[i], 1, struct regnode_charclass_class);
11804 StructCopy(ri->data->data[i], d->data[i],
11805 struct regnode_charclass_class);
11806 reti->regstclass = (regnode*)d->data[i];
11809 /* Compiled op trees are readonly and in shared memory,
11810 and can thus be shared without duplication. */
11812 d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
11816 /* Trie stclasses are readonly and can thus be shared
11817 * without duplication. We free the stclass in pregfree
11818 * when the corresponding reg_ac_data struct is freed.
11820 reti->regstclass= ri->regstclass;
11824 ((reg_trie_data*)ri->data->data[i])->refcount++;
11828 d->data[i] = ri->data->data[i];
11831 Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
11840 reti->name_list_idx = ri->name_list_idx;
11842 #ifdef RE_TRACK_PATTERN_OFFSETS
11843 if (ri->u.offsets) {
11844 Newx(reti->u.offsets, 2*len+1, U32);
11845 Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32);
11848 SetProgLen(reti,len);
11851 return (void*)reti;
11854 #endif /* USE_ITHREADS */
11856 #ifndef PERL_IN_XSUB_RE
11859 - regnext - dig the "next" pointer out of a node
11862 Perl_regnext(pTHX_ register regnode *p)
11865 register I32 offset;
11870 if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */
11871 Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX);
11874 offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
11883 S_re_croak2(pTHX_ const char* pat1,const char* pat2,...)
11886 STRLEN l1 = strlen(pat1);
11887 STRLEN l2 = strlen(pat2);
11890 const char *message;
11892 PERL_ARGS_ASSERT_RE_CROAK2;
11898 Copy(pat1, buf, l1 , char);
11899 Copy(pat2, buf + l1, l2 , char);
11900 buf[l1 + l2] = '\n';
11901 buf[l1 + l2 + 1] = '\0';
11903 /* ANSI variant takes additional second argument */
11904 va_start(args, pat2);
11908 msv = vmess(buf, &args);
11910 message = SvPV_const(msv,l1);
11913 Copy(message, buf, l1 , char);
11914 buf[l1-1] = '\0'; /* Overwrite \n */
11915 Perl_croak(aTHX_ "%s", buf);
11918 /* XXX Here's a total kludge. But we need to re-enter for swash routines. */
11920 #ifndef PERL_IN_XSUB_RE
11922 Perl_save_re_context(pTHX)
11926 struct re_save_state *state;
11928 SAVEVPTR(PL_curcop);
11929 SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1);
11931 state = (struct re_save_state *)(PL_savestack + PL_savestack_ix);
11932 PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE;
11933 SSPUSHUV(SAVEt_RE_STATE);
11935 Copy(&PL_reg_state, state, 1, struct re_save_state);
11937 PL_reg_start_tmp = 0;
11938 PL_reg_start_tmpl = 0;
11939 PL_reg_oldsaved = NULL;
11940 PL_reg_oldsavedlen = 0;
11941 PL_reg_maxiter = 0;
11942 PL_reg_leftiter = 0;
11943 PL_reg_poscache = NULL;
11944 PL_reg_poscache_size = 0;
11945 #ifdef PERL_OLD_COPY_ON_WRITE
11949 /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */
11951 const REGEXP * const rx = PM_GETRE(PL_curpm);
11954 for (i = 1; i <= RX_NPARENS(rx); i++) {
11955 char digits[TYPE_CHARS(long)];
11956 const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i);
11957 GV *const *const gvp
11958 = (GV**)hv_fetch(PL_defstash, digits, len, 0);
11961 GV * const gv = *gvp;
11962 if (SvTYPE(gv) == SVt_PVGV && GvSV(gv))
11972 clear_re(pTHX_ void *r)
11975 ReREFCNT_dec((REGEXP *)r);
11981 S_put_byte(pTHX_ SV *sv, int c)
11983 PERL_ARGS_ASSERT_PUT_BYTE;
11985 /* Our definition of isPRINT() ignores locales, so only bytes that are
11986 not part of UTF-8 are considered printable. I assume that the same
11987 holds for UTF-EBCDIC.
11988 Also, code point 255 is not printable in either (it's E0 in EBCDIC,
11989 which Wikipedia says:
11991 EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all
11992 ones (binary 1111 1111, hexadecimal FF). It is similar, but not
11993 identical, to the ASCII delete (DEL) or rubout control character.
11994 ) So the old condition can be simplified to !isPRINT(c) */
11997 Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c);
12000 Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c);
12004 const char string = c;
12005 if (c == '-' || c == ']' || c == '\\' || c == '^')
12006 sv_catpvs(sv, "\\");
12007 sv_catpvn(sv, &string, 1);
12012 #define CLEAR_OPTSTART \
12013 if (optstart) STMT_START { \
12014 DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
12018 #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
12020 STATIC const regnode *
12021 S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
12022 const regnode *last, const regnode *plast,
12023 SV* sv, I32 indent, U32 depth)
12026 register U8 op = PSEUDO; /* Arbitrary non-END op. */
12027 register const regnode *next;
12028 const regnode *optstart= NULL;
12030 RXi_GET_DECL(r,ri);
12031 GET_RE_DEBUG_FLAGS_DECL;
12033 PERL_ARGS_ASSERT_DUMPUNTIL;
12035 #ifdef DEBUG_DUMPUNTIL
12036 PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
12037 last ? last-start : 0,plast ? plast-start : 0);
12040 if (plast && plast < last)
12043 while (PL_regkind[op] != END && (!last || node < last)) {
12044 /* While that wasn't END last time... */
12047 if (op == CLOSE || op == WHILEM)
12049 next = regnext((regnode *)node);
12052 if (OP(node) == OPTIMIZED) {
12053 if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
12060 regprop(r, sv, node);
12061 PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
12062 (int)(2*indent + 1), "", SvPVX_const(sv));
12064 if (OP(node) != OPTIMIZED) {
12065 if (next == NULL) /* Next ptr. */
12066 PerlIO_printf(Perl_debug_log, " (0)");
12067 else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
12068 PerlIO_printf(Perl_debug_log, " (FAIL)");
12070 PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start));
12071 (void)PerlIO_putc(Perl_debug_log, '\n');
12075 if (PL_regkind[(U8)op] == BRANCHJ) {
12078 register const regnode *nnode = (OP(next) == LONGJMP
12079 ? regnext((regnode *)next)
12081 if (last && nnode > last)
12083 DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
12086 else if (PL_regkind[(U8)op] == BRANCH) {
12088 DUMPUNTIL(NEXTOPER(node), next);
12090 else if ( PL_regkind[(U8)op] == TRIE ) {
12091 const regnode *this_trie = node;
12092 const char op = OP(node);
12093 const U32 n = ARG(node);
12094 const reg_ac_data * const ac = op>=AHOCORASICK ?
12095 (reg_ac_data *)ri->data->data[n] :
12097 const reg_trie_data * const trie =
12098 (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
12100 AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]);
12102 const regnode *nextbranch= NULL;
12105 for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
12106 SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
12108 PerlIO_printf(Perl_debug_log, "%*s%s ",
12109 (int)(2*(indent+3)), "",
12110 elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
12111 PL_colors[0], PL_colors[1],
12112 (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
12113 PERL_PV_PRETTY_ELLIPSES |
12114 PERL_PV_PRETTY_LTGT
12119 U16 dist= trie->jump[word_idx+1];
12120 PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
12121 (UV)((dist ? this_trie + dist : next) - start));
12124 nextbranch= this_trie + trie->jump[0];
12125 DUMPUNTIL(this_trie + dist, nextbranch);
12127 if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
12128 nextbranch= regnext((regnode *)nextbranch);
12130 PerlIO_printf(Perl_debug_log, "\n");
12133 if (last && next > last)
12138 else if ( op == CURLY ) { /* "next" might be very big: optimizer */
12139 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
12140 NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
12142 else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
12144 DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
12146 else if ( op == PLUS || op == STAR) {
12147 DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
12149 else if (PL_regkind[(U8)op] == ANYOF) {
12150 /* arglen 1 + class block */
12151 node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS)
12152 ? ANYOF_CLASS_SKIP : ANYOF_SKIP);
12153 node = NEXTOPER(node);
12155 else if (PL_regkind[(U8)op] == EXACT) {
12156 /* Literal string, where present. */
12157 node += NODE_SZ_STR(node) - 1;
12158 node = NEXTOPER(node);
12161 node = NEXTOPER(node);
12162 node += regarglen[(U8)op];
12164 if (op == CURLYX || op == OPEN)
12168 #ifdef DEBUG_DUMPUNTIL
12169 PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
12174 #endif /* DEBUGGING */
12178 * c-indentation-style: bsd
12179 * c-basic-offset: 4
12180 * indent-tabs-mode: t
12183 * ex: set ts=8 sts=4 sw=4 noet: