This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: More work on ANYOF_CLASS
[perl5.git] / regcomp.c
CommitLineData
a0d0e21e
LW
1/* regcomp.c
2 */
3
4/*
4ac71550
TC
5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
6 *
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
a0d0e21e
LW
8 */
9
61296642
DM
10/* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
166f8a29 12 * a regular expression.
e4a054ea
DM
13 *
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.
166f8a29
DM
18 */
19
a687059c
LW
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!
22 */
23
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.
27 */
28
e50aee73 29/* The names of the functions have been changed from regcomp and
3b753521 30 * regexec to pregcomp and pregexec in order to avoid conflicts
e50aee73
AD
31 * with the POSIX routines of the same names.
32*/
33
b9d5759e 34#ifdef PERL_EXT_RE_BUILD
54df2634 35#include "re_top.h"
b81d288d 36#endif
56953603 37
a687059c 38/*
e50aee73 39 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
40 *
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
43 *
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:
47 *
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
50 * from defects in it.
51 *
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
54 *
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
57 *
58 *
59 **** Alterations to Henry's code are...
60 ****
4bb101f2 61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
a687059c 64 ****
9ef589d8
LW
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.
67
a687059c
LW
68 *
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.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGCOMP_C
a687059c 75#include "perl.h"
d06ea78c 76
acfe0abc 77#ifndef PERL_IN_XSUB_RE
d06ea78c
GS
78# include "INTERN.h"
79#endif
c277df42
IZ
80
81#define REG_COMP_C
54df2634
NC
82#ifdef PERL_IN_XSUB_RE
83# include "re_comp.h"
84#else
85# include "regcomp.h"
86#endif
a687059c 87
04e98a4d
AD
88#include "dquote_static.c"
89
d4cce5f1 90#ifdef op
11343788 91#undef op
d4cce5f1 92#endif /* op */
11343788 93
fe14fcc3 94#ifdef MSDOS
7e4e8c89 95# if defined(BUGGY_MSC6)
fe14fcc3 96 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
7e4e8c89 97# pragma optimize("a",off)
fe14fcc3 98 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
7e4e8c89
NC
99# pragma optimize("w",on )
100# endif /* BUGGY_MSC6 */
fe14fcc3
LW
101#endif /* MSDOS */
102
a687059c
LW
103#ifndef STATIC
104#define STATIC static
105#endif
106
830247a4 107typedef struct RExC_state_t {
e2509266 108 U32 flags; /* are we folding, multilining? */
830247a4 109 char *precomp; /* uncompiled string. */
288b8c02 110 REGEXP *rx_sv; /* The SV that is the regexp. */
f8fc2ecf
YO
111 regexp *rx; /* perl core regexp structure */
112 regexp_internal *rxi; /* internal data for regexp object pprivate field */
fac92740 113 char *start; /* Start of input for compile */
830247a4
IZ
114 char *end; /* End of input for compile */
115 char *parse; /* Input-scan pointer. */
116 I32 whilem_seen; /* number of WHILEM in this expr */
fac92740 117 regnode *emit_start; /* Start of emitted-code area */
3b57cd43 118 regnode *emit_bound; /* First regnode outside of the allocated space */
ffc61ed2 119 regnode *emit; /* Code-emit pointer; &regdummy = don't = compiling */
830247a4
IZ
120 I32 naughty; /* How bad is this pattern? */
121 I32 sawback; /* Did we see \1, ...? */
122 U32 seen;
123 I32 size; /* Code size. */
c74340f9
YO
124 I32 npar; /* Capture buffer count, (OPEN). */
125 I32 cpar; /* Capture buffer count, (CLOSE). */
e2e6a0f1 126 I32 nestroot; /* root parens we are in - used by accept */
830247a4
IZ
127 I32 extralen;
128 I32 seen_zerolen;
129 I32 seen_evals;
40d049e4
YO
130 regnode **open_parens; /* pointers to open parens */
131 regnode **close_parens; /* pointers to close parens */
132 regnode *opend; /* END node in program */
02daf0ab
YO
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. */
81714fb9 137 HV *paren_names; /* Paren names */
1f1031fe 138
40d049e4
YO
139 regnode **recurse; /* Recurse regops */
140 I32 recurse_count; /* Number of recurse regops */
830247a4
IZ
141#if ADD_TO_REGEXEC
142 char *starttry; /* -Dr: where regtry was called. */
143#define RExC_starttry (pRExC_state->starttry)
144#endif
3dab1dad 145#ifdef DEBUGGING
be8e71aa 146 const char *lastparse;
3dab1dad 147 I32 lastnum;
1f1031fe 148 AV *paren_name_list; /* idx -> name */
3dab1dad
YO
149#define RExC_lastparse (pRExC_state->lastparse)
150#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 151#define RExC_paren_name_list (pRExC_state->paren_name_list)
3dab1dad 152#endif
830247a4
IZ
153} RExC_state_t;
154
e2509266 155#define RExC_flags (pRExC_state->flags)
830247a4 156#define RExC_precomp (pRExC_state->precomp)
288b8c02 157#define RExC_rx_sv (pRExC_state->rx_sv)
830247a4 158#define RExC_rx (pRExC_state->rx)
f8fc2ecf 159#define RExC_rxi (pRExC_state->rxi)
fac92740 160#define RExC_start (pRExC_state->start)
830247a4
IZ
161#define RExC_end (pRExC_state->end)
162#define RExC_parse (pRExC_state->parse)
163#define RExC_whilem_seen (pRExC_state->whilem_seen)
7122b237
YO
164#ifdef RE_TRACK_PATTERN_OFFSETS
165#define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
166#endif
830247a4 167#define RExC_emit (pRExC_state->emit)
fac92740 168#define RExC_emit_start (pRExC_state->emit_start)
3b57cd43 169#define RExC_emit_bound (pRExC_state->emit_bound)
830247a4
IZ
170#define RExC_naughty (pRExC_state->naughty)
171#define RExC_sawback (pRExC_state->sawback)
172#define RExC_seen (pRExC_state->seen)
173#define RExC_size (pRExC_state->size)
174#define RExC_npar (pRExC_state->npar)
e2e6a0f1 175#define RExC_nestroot (pRExC_state->nestroot)
830247a4
IZ
176#define RExC_extralen (pRExC_state->extralen)
177#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
178#define RExC_seen_evals (pRExC_state->seen_evals)
1aa99e6b 179#define RExC_utf8 (pRExC_state->utf8)
02daf0ab 180#define RExC_orig_utf8 (pRExC_state->orig_utf8)
40d049e4
YO
181#define RExC_open_parens (pRExC_state->open_parens)
182#define RExC_close_parens (pRExC_state->close_parens)
183#define RExC_opend (pRExC_state->opend)
81714fb9 184#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
185#define RExC_recurse (pRExC_state->recurse)
186#define RExC_recurse_count (pRExC_state->recurse_count)
830247a4 187
cde0cee5 188
a687059c
LW
189#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
190#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
191 ((*s) == '{' && regcurly(s)))
a687059c 192
35c8bce7
LW
193#ifdef SPSTART
194#undef SPSTART /* dratted cpp namespace... */
195#endif
a687059c
LW
196/*
197 * Flags to be passed up and down.
198 */
a687059c 199#define WORST 0 /* Worst case. */
a3b492c3 200#define HASWIDTH 0x01 /* Known to match non-null strings. */
fda99bee
KW
201
202/* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single
203 * character, and if utf8, must be invariant. */
204#define SIMPLE 0x02
a3b492c3
YO
205#define SPSTART 0x04 /* Starts with * or +. */
206#define TRYAGAIN 0x08 /* Weeded out a declaration. */
207#define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
a687059c 208
3dab1dad
YO
209#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
210
07be1b83
YO
211/* whether trie related optimizations are enabled */
212#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
213#define TRIE_STUDY_OPT
786e8c11 214#define FULL_TRIE_STUDY
07be1b83
YO
215#define TRIE_STCLASS
216#endif
1de06328
YO
217
218
40d049e4
YO
219
220#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
221#define PBITVAL(paren) (1 << ((paren) & 7))
222#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
223#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
224#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
225
bbd61b5f
KW
226/* If not already in utf8, do a longjmp back to the beginning */
227#define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */
228#define REQUIRE_UTF8 STMT_START { \
229 if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \
230 } STMT_END
40d049e4 231
1de06328
YO
232/* About scan_data_t.
233
234 During optimisation we recurse through the regexp program performing
235 various inplace (keyhole style) optimisations. In addition study_chunk
236 and scan_commit populate this data structure with information about
237 what strings MUST appear in the pattern. We look for the longest
3b753521 238 string that must appear at a fixed location, and we look for the
1de06328
YO
239 longest string that may appear at a floating location. So for instance
240 in the pattern:
241
242 /FOO[xX]A.*B[xX]BAR/
243
244 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
245 strings (because they follow a .* construct). study_chunk will identify
246 both FOO and BAR as being the longest fixed and floating strings respectively.
247
248 The strings can be composites, for instance
249
250 /(f)(o)(o)/
251
252 will result in a composite fixed substring 'foo'.
253
254 For each string some basic information is maintained:
255
256 - offset or min_offset
257 This is the position the string must appear at, or not before.
258 It also implicitly (when combined with minlenp) tells us how many
3b753521
FN
259 characters must match before the string we are searching for.
260 Likewise when combined with minlenp and the length of the string it
1de06328
YO
261 tells us how many characters must appear after the string we have
262 found.
263
264 - max_offset
265 Only used for floating strings. This is the rightmost point that
3b753521 266 the string can appear at. If set to I32 max it indicates that the
1de06328
YO
267 string can occur infinitely far to the right.
268
269 - minlenp
270 A pointer to the minimum length of the pattern that the string
271 was found inside. This is important as in the case of positive
272 lookahead or positive lookbehind we can have multiple patterns
273 involved. Consider
274
275 /(?=FOO).*F/
276
277 The minimum length of the pattern overall is 3, the minimum length
278 of the lookahead part is 3, but the minimum length of the part that
279 will actually match is 1. So 'FOO's minimum length is 3, but the
280 minimum length for the F is 1. This is important as the minimum length
281 is used to determine offsets in front of and behind the string being
282 looked for. Since strings can be composites this is the length of the
283 pattern at the time it was commited with a scan_commit. Note that
284 the length is calculated by study_chunk, so that the minimum lengths
285 are not known until the full pattern has been compiled, thus the
286 pointer to the value.
287
288 - lookbehind
289
290 In the case of lookbehind the string being searched for can be
291 offset past the start point of the final matching string.
292 If this value was just blithely removed from the min_offset it would
293 invalidate some of the calculations for how many chars must match
294 before or after (as they are derived from min_offset and minlen and
295 the length of the string being searched for).
296 When the final pattern is compiled and the data is moved from the
297 scan_data_t structure into the regexp structure the information
298 about lookbehind is factored in, with the information that would
299 have been lost precalculated in the end_shift field for the
300 associated string.
301
302 The fields pos_min and pos_delta are used to store the minimum offset
303 and the delta to the maximum offset at the current point in the pattern.
304
305*/
2c2d71f5
JH
306
307typedef struct scan_data_t {
1de06328
YO
308 /*I32 len_min; unused */
309 /*I32 len_delta; unused */
2c2d71f5
JH
310 I32 pos_min;
311 I32 pos_delta;
312 SV *last_found;
1de06328 313 I32 last_end; /* min value, <0 unless valid. */
2c2d71f5
JH
314 I32 last_start_min;
315 I32 last_start_max;
1de06328
YO
316 SV **longest; /* Either &l_fixed, or &l_float. */
317 SV *longest_fixed; /* longest fixed string found in pattern */
318 I32 offset_fixed; /* offset where it starts */
319 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
320 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
321 SV *longest_float; /* longest floating string found in pattern */
322 I32 offset_float_min; /* earliest point in string it can appear */
323 I32 offset_float_max; /* latest point in string it can appear */
324 I32 *minlen_float; /* pointer to the minlen relevent to the string */
325 I32 lookbehind_float; /* is the position of the string modified by LB */
2c2d71f5
JH
326 I32 flags;
327 I32 whilem_c;
cb434fcc 328 I32 *last_closep;
653099ff 329 struct regnode_charclass_class *start_class;
2c2d71f5
JH
330} scan_data_t;
331
a687059c 332/*
e50aee73 333 * Forward declarations for pregcomp()'s friends.
a687059c 334 */
a0d0e21e 335
27da23d5 336static const scan_data_t zero_scan_data =
1de06328 337 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
c277df42
IZ
338
339#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
07be1b83
YO
340#define SF_BEFORE_SEOL 0x0001
341#define SF_BEFORE_MEOL 0x0002
c277df42
IZ
342#define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
343#define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
344
09b7f37c
CB
345#ifdef NO_UNARY_PLUS
346# define SF_FIX_SHIFT_EOL (0+2)
347# define SF_FL_SHIFT_EOL (0+4)
348#else
349# define SF_FIX_SHIFT_EOL (+2)
350# define SF_FL_SHIFT_EOL (+4)
351#endif
c277df42
IZ
352
353#define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
354#define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
355
356#define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
357#define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
07be1b83
YO
358#define SF_IS_INF 0x0040
359#define SF_HAS_PAR 0x0080
360#define SF_IN_PAR 0x0100
361#define SF_HAS_EVAL 0x0200
362#define SCF_DO_SUBSTR 0x0400
653099ff
GS
363#define SCF_DO_STCLASS_AND 0x0800
364#define SCF_DO_STCLASS_OR 0x1000
365#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 366#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 367
786e8c11 368#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
e2e6a0f1 369#define SCF_SEEN_ACCEPT 0x8000
07be1b83 370
43fead97
KW
371#define UTF cBOOL(RExC_utf8)
372#define LOC cBOOL(RExC_flags & RXf_PMf_LOCALE)
373#define UNI_SEMANTICS cBOOL(RExC_flags & RXf_PMf_UNICODE)
374#define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
a0ed51b3 375
ffc61ed2 376#define OOB_UNICODE 12345678
93733859 377#define OOB_NAMEDCLASS -1
b8c5462f 378
a0ed51b3
LW
379#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
380#define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
381
8615cb43 382
b45f050a
JF
383/* length of regex to show in messages that don't mark a position within */
384#define RegexLengthToShowInErrorMessages 127
385
386/*
387 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
388 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
389 * op/pragma/warn/regcomp.
390 */
7253e4e3
RK
391#define MARKER1 "<-- HERE" /* marker as it appears in the description */
392#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 393
7253e4e3 394#define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
b45f050a
JF
395
396/*
397 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
398 * arg. Show regex, up to a maximum length. If it's too long, chop and add
399 * "...".
400 */
58e23c8d 401#define _FAIL(code) STMT_START { \
bfed75c6 402 const char *ellipses = ""; \
ccb2c380
MP
403 IV len = RExC_end - RExC_precomp; \
404 \
405 if (!SIZE_ONLY) \
288b8c02 406 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
ccb2c380
MP
407 if (len > RegexLengthToShowInErrorMessages) { \
408 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
409 len = RegexLengthToShowInErrorMessages - 10; \
410 ellipses = "..."; \
411 } \
58e23c8d 412 code; \
ccb2c380 413} STMT_END
8615cb43 414
58e23c8d
YO
415#define FAIL(msg) _FAIL( \
416 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
417 msg, (int)len, RExC_precomp, ellipses))
418
419#define FAIL2(msg,arg) _FAIL( \
420 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
421 arg, (int)len, RExC_precomp, ellipses))
422
b45f050a 423/*
b45f050a
JF
424 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
425 */
ccb2c380 426#define Simple_vFAIL(m) STMT_START { \
a28509cc 427 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
428 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
429 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
430} STMT_END
b45f050a
JF
431
432/*
433 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
434 */
ccb2c380
MP
435#define vFAIL(m) STMT_START { \
436 if (!SIZE_ONLY) \
288b8c02 437 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
ccb2c380
MP
438 Simple_vFAIL(m); \
439} STMT_END
b45f050a
JF
440
441/*
442 * Like Simple_vFAIL(), but accepts two arguments.
443 */
ccb2c380 444#define Simple_vFAIL2(m,a1) STMT_START { \
a28509cc 445 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
446 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
447 (int)offset, RExC_precomp, RExC_precomp + offset); \
448} STMT_END
b45f050a
JF
449
450/*
451 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
452 */
ccb2c380
MP
453#define vFAIL2(m,a1) STMT_START { \
454 if (!SIZE_ONLY) \
288b8c02 455 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
ccb2c380
MP
456 Simple_vFAIL2(m, a1); \
457} STMT_END
b45f050a
JF
458
459
460/*
461 * Like Simple_vFAIL(), but accepts three arguments.
462 */
ccb2c380 463#define Simple_vFAIL3(m, a1, a2) STMT_START { \
a28509cc 464 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
465 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
466 (int)offset, RExC_precomp, RExC_precomp + offset); \
467} STMT_END
b45f050a
JF
468
469/*
470 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
471 */
ccb2c380
MP
472#define vFAIL3(m,a1,a2) STMT_START { \
473 if (!SIZE_ONLY) \
288b8c02 474 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \
ccb2c380
MP
475 Simple_vFAIL3(m, a1, a2); \
476} STMT_END
b45f050a
JF
477
478/*
479 * Like Simple_vFAIL(), but accepts four arguments.
480 */
ccb2c380 481#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
a28509cc 482 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
483 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
484 (int)offset, RExC_precomp, RExC_precomp + offset); \
485} STMT_END
b45f050a 486
668c081a 487#define ckWARNreg(loc,m) STMT_START { \
a28509cc 488 const IV offset = loc - RExC_precomp; \
f10f4c18
NC
489 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 (int)offset, RExC_precomp, RExC_precomp + offset); \
ccb2c380
MP
491} STMT_END
492
668c081a 493#define ckWARNregdep(loc,m) STMT_START { \
a28509cc 494 const IV offset = loc - RExC_precomp; \
d1d15184 495 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
f10f4c18
NC
496 m REPORT_LOCATION, \
497 (int)offset, RExC_precomp, RExC_precomp + offset); \
ccb2c380
MP
498} STMT_END
499
668c081a 500#define ckWARN2reg(loc, m, a1) STMT_START { \
a28509cc 501 const IV offset = loc - RExC_precomp; \
668c081a 502 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
ccb2c380
MP
503 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
504} STMT_END
505
506#define vWARN3(loc, m, a1, a2) STMT_START { \
a28509cc 507 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
508 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
509 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
510} STMT_END
511
668c081a
NC
512#define ckWARN3reg(loc, m, a1, a2) STMT_START { \
513 const IV offset = loc - RExC_precomp; \
514 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
515 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
516} STMT_END
517
ccb2c380 518#define vWARN4(loc, m, a1, a2, a3) STMT_START { \
a28509cc 519 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
520 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
521 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
522} STMT_END
523
668c081a
NC
524#define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \
525 const IV offset = loc - RExC_precomp; \
526 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
527 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
528} STMT_END
529
ccb2c380 530#define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
a28509cc 531 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
532 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
533 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
534} STMT_END
9d1d55b5 535
8615cb43 536
cd439c50 537/* Allow for side effects in s */
ccb2c380
MP
538#define REGC(c,s) STMT_START { \
539 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
540} STMT_END
cd439c50 541
fac92740
MJD
542/* Macros for recording node offsets. 20001227 mjd@plover.com
543 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
544 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
545 * Element 0 holds the number n.
07be1b83 546 * Position is 1 indexed.
fac92740 547 */
7122b237
YO
548#ifndef RE_TRACK_PATTERN_OFFSETS
549#define Set_Node_Offset_To_R(node,byte)
550#define Set_Node_Offset(node,byte)
551#define Set_Cur_Node_Offset
552#define Set_Node_Length_To_R(node,len)
553#define Set_Node_Length(node,len)
554#define Set_Node_Cur_Length(node)
555#define Node_Offset(n)
556#define Node_Length(n)
557#define Set_Node_Offset_Length(node,offset,len)
558#define ProgLen(ri) ri->u.proglen
559#define SetProgLen(ri,x) ri->u.proglen = x
560#else
561#define ProgLen(ri) ri->u.offsets[0]
562#define SetProgLen(ri,x) ri->u.offsets[0] = x
ccb2c380
MP
563#define Set_Node_Offset_To_R(node,byte) STMT_START { \
564 if (! SIZE_ONLY) { \
565 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
2a49f0f5 566 __LINE__, (int)(node), (int)(byte))); \
ccb2c380 567 if((node) < 0) { \
551405c4 568 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
ccb2c380
MP
569 } else { \
570 RExC_offsets[2*(node)-1] = (byte); \
571 } \
572 } \
573} STMT_END
574
575#define Set_Node_Offset(node,byte) \
576 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
577#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
578
579#define Set_Node_Length_To_R(node,len) STMT_START { \
580 if (! SIZE_ONLY) { \
581 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 582 __LINE__, (int)(node), (int)(len))); \
ccb2c380 583 if((node) < 0) { \
551405c4 584 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
ccb2c380
MP
585 } else { \
586 RExC_offsets[2*(node)] = (len); \
587 } \
588 } \
589} STMT_END
590
591#define Set_Node_Length(node,len) \
592 Set_Node_Length_To_R((node)-RExC_emit_start, len)
593#define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
594#define Set_Node_Cur_Length(node) \
595 Set_Node_Length(node, RExC_parse - parse_start)
fac92740
MJD
596
597/* Get offsets and lengths */
598#define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
599#define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
600
07be1b83
YO
601#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
602 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
603 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
604} STMT_END
7122b237 605#endif
07be1b83
YO
606
607#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
608#define EXPERIMENTAL_INPLACESCAN
f427392e 609#endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
07be1b83 610
304ee84b
YO
611#define DEBUG_STUDYDATA(str,data,depth) \
612DEBUG_OPTIMISE_MORE_r(if(data){ \
1de06328 613 PerlIO_printf(Perl_debug_log, \
304ee84b
YO
614 "%*s" str "Pos:%"IVdf"/%"IVdf \
615 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
1de06328
YO
616 (int)(depth)*2, "", \
617 (IV)((data)->pos_min), \
618 (IV)((data)->pos_delta), \
304ee84b 619 (UV)((data)->flags), \
1de06328 620 (IV)((data)->whilem_c), \
304ee84b
YO
621 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
622 is_inf ? "INF " : "" \
1de06328
YO
623 ); \
624 if ((data)->last_found) \
625 PerlIO_printf(Perl_debug_log, \
626 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
627 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
628 SvPVX_const((data)->last_found), \
629 (IV)((data)->last_end), \
630 (IV)((data)->last_start_min), \
631 (IV)((data)->last_start_max), \
632 ((data)->longest && \
633 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
634 SvPVX_const((data)->longest_fixed), \
635 (IV)((data)->offset_fixed), \
636 ((data)->longest && \
637 (data)->longest==&((data)->longest_float)) ? "*" : "", \
638 SvPVX_const((data)->longest_float), \
639 (IV)((data)->offset_float_min), \
640 (IV)((data)->offset_float_max) \
641 ); \
642 PerlIO_printf(Perl_debug_log,"\n"); \
643});
644
acfe0abc 645static void clear_re(pTHX_ void *r);
4327152a 646
653099ff 647/* Mark that we cannot extend a found fixed substring at this point.
786e8c11 648 Update the longest found anchored substring and the longest found
653099ff
GS
649 floating substrings if needed. */
650
4327152a 651STATIC void
304ee84b 652S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
c277df42 653{
e1ec3a88
AL
654 const STRLEN l = CHR_SVLEN(data->last_found);
655 const STRLEN old_l = CHR_SVLEN(*data->longest);
1de06328 656 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 657
7918f24d
NC
658 PERL_ARGS_ASSERT_SCAN_COMMIT;
659
c277df42 660 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
6b43b216 661 SvSetMagicSV(*data->longest, data->last_found);
c277df42
IZ
662 if (*data->longest == data->longest_fixed) {
663 data->offset_fixed = l ? data->last_start_min : data->pos_min;
664 if (data->flags & SF_BEFORE_EOL)
b81d288d 665 data->flags
c277df42
IZ
666 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
667 else
668 data->flags &= ~SF_FIX_BEFORE_EOL;
1de06328
YO
669 data->minlen_fixed=minlenp;
670 data->lookbehind_fixed=0;
a0ed51b3 671 }
304ee84b 672 else { /* *data->longest == data->longest_float */
c277df42 673 data->offset_float_min = l ? data->last_start_min : data->pos_min;
b81d288d
AB
674 data->offset_float_max = (l
675 ? data->last_start_max
c277df42 676 : data->pos_min + data->pos_delta);
304ee84b 677 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
9051bda5 678 data->offset_float_max = I32_MAX;
c277df42 679 if (data->flags & SF_BEFORE_EOL)
b81d288d 680 data->flags
c277df42
IZ
681 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
682 else
683 data->flags &= ~SF_FL_BEFORE_EOL;
1de06328
YO
684 data->minlen_float=minlenp;
685 data->lookbehind_float=0;
c277df42
IZ
686 }
687 }
688 SvCUR_set(data->last_found, 0);
0eda9292 689 {
a28509cc 690 SV * const sv = data->last_found;
097eb12c
AL
691 if (SvUTF8(sv) && SvMAGICAL(sv)) {
692 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
693 if (mg)
694 mg->mg_len = 0;
695 }
0eda9292 696 }
c277df42
IZ
697 data->last_end = -1;
698 data->flags &= ~SF_BEFORE_EOL;
bcdf7404 699 DEBUG_STUDYDATA("commit: ",data,0);
c277df42
IZ
700}
701
653099ff
GS
702/* Can match anything (initialization) */
703STATIC void
097eb12c 704S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 705{
7918f24d
NC
706 PERL_ARGS_ASSERT_CL_ANYTHING;
707
653099ff 708 ANYOF_CLASS_ZERO(cl);
f8bef550 709 ANYOF_BITMAP_SETALL(cl);
1aa99e6b 710 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
653099ff
GS
711 if (LOC)
712 cl->flags |= ANYOF_LOCALE;
8951c461 713 cl->flags |= ANYOF_FOLD;
653099ff
GS
714}
715
716/* Can match anything (initialization) */
717STATIC int
5f66b61c 718S_cl_is_anything(const struct regnode_charclass_class *cl)
653099ff
GS
719{
720 int value;
721
7918f24d
NC
722 PERL_ARGS_ASSERT_CL_IS_ANYTHING;
723
aaa51d5e 724 for (value = 0; value <= ANYOF_MAX; value += 2)
653099ff
GS
725 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
726 return 1;
1aa99e6b
IH
727 if (!(cl->flags & ANYOF_UNICODE_ALL))
728 return 0;
10edeb5d 729 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
f8bef550 730 return 0;
653099ff
GS
731 return 1;
732}
733
734/* Can match anything (initialization) */
735STATIC void
097eb12c 736S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 737{
7918f24d
NC
738 PERL_ARGS_ASSERT_CL_INIT;
739
8ecf7187 740 Zero(cl, 1, struct regnode_charclass_class);
653099ff 741 cl->type = ANYOF;
830247a4 742 cl_anything(pRExC_state, cl);
653099ff
GS
743}
744
745STATIC void
097eb12c 746S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 747{
7918f24d
NC
748 PERL_ARGS_ASSERT_CL_INIT_ZERO;
749
8ecf7187 750 Zero(cl, 1, struct regnode_charclass_class);
653099ff 751 cl->type = ANYOF;
830247a4 752 cl_anything(pRExC_state, cl);
653099ff
GS
753 if (LOC)
754 cl->flags |= ANYOF_LOCALE;
755}
756
757/* 'And' a given class with another one. Can create false positives */
758/* We assume that cl is not inverted */
759STATIC void
5f66b61c 760S_cl_and(struct regnode_charclass_class *cl,
a28509cc 761 const struct regnode_charclass_class *and_with)
653099ff 762{
7918f24d 763 PERL_ARGS_ASSERT_CL_AND;
40d049e4
YO
764
765 assert(and_with->type == ANYOF);
1e6ade67
KW
766
767 if (!(ANYOF_CLASS_TEST_ANY_SET(and_with))
768 && !(ANYOF_CLASS_TEST_ANY_SET(cl))
653099ff
GS
769 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
770 && !(and_with->flags & ANYOF_FOLD)
771 && !(cl->flags & ANYOF_FOLD)) {
772 int i;
773
774 if (and_with->flags & ANYOF_INVERT)
775 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
776 cl->bitmap[i] &= ~and_with->bitmap[i];
777 else
778 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
779 cl->bitmap[i] &= and_with->bitmap[i];
780 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
781 if (!(and_with->flags & ANYOF_EOS))
782 cl->flags &= ~ANYOF_EOS;
1aa99e6b 783
8951c461
KW
784 if (!(and_with->flags & ANYOF_FOLD))
785 cl->flags &= ~ANYOF_FOLD;
786
3ff7ceb3 787 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_NONBITMAP &&
14ebb1a2 788 !(and_with->flags & ANYOF_INVERT)) {
1aa99e6b 789 cl->flags &= ~ANYOF_UNICODE_ALL;
ef87b810
KW
790 cl->flags |= and_with->flags & ANYOF_NONBITMAP; /* field is 2 bits; use
791 only the one(s)
792 actually set */
1aa99e6b
IH
793 ARG_SET(cl, ARG(and_with));
794 }
14ebb1a2
JH
795 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
796 !(and_with->flags & ANYOF_INVERT))
1aa99e6b 797 cl->flags &= ~ANYOF_UNICODE_ALL;
3ff7ceb3 798 if (!(and_with->flags & (ANYOF_NONBITMAP|ANYOF_UNICODE_ALL)) &&
14ebb1a2 799 !(and_with->flags & ANYOF_INVERT))
3ff7ceb3 800 cl->flags &= ~ANYOF_NONBITMAP;
653099ff
GS
801}
802
803/* 'OR' a given class with another one. Can create false positives */
804/* We assume that cl is not inverted */
805STATIC void
097eb12c 806S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
653099ff 807{
7918f24d
NC
808 PERL_ARGS_ASSERT_CL_OR;
809
653099ff
GS
810 if (or_with->flags & ANYOF_INVERT) {
811 /* We do not use
812 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
813 * <= (B1 | !B2) | (CL1 | !CL2)
814 * which is wasteful if CL2 is small, but we ignore CL2:
815 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
816 * XXXX Can we handle case-fold? Unclear:
817 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
818 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
819 */
820 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
821 && !(or_with->flags & ANYOF_FOLD)
822 && !(cl->flags & ANYOF_FOLD) ) {
823 int i;
824
825 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
826 cl->bitmap[i] |= ~or_with->bitmap[i];
827 } /* XXXX: logic is complicated otherwise */
828 else {
830247a4 829 cl_anything(pRExC_state, cl);
653099ff
GS
830 }
831 } else {
832 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
833 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
b81d288d 834 && (!(or_with->flags & ANYOF_FOLD)
653099ff
GS
835 || (cl->flags & ANYOF_FOLD)) ) {
836 int i;
837
838 /* OR char bitmap and class bitmap separately */
839 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
840 cl->bitmap[i] |= or_with->bitmap[i];
1e6ade67 841 if (ANYOF_CLASS_TEST_ANY_SET(or_with)) {
653099ff
GS
842 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
843 cl->classflags[i] |= or_with->classflags[i];
844 cl->flags |= ANYOF_CLASS;
845 }
846 }
847 else { /* XXXX: logic is complicated, leave it along for a moment. */
830247a4 848 cl_anything(pRExC_state, cl);
653099ff
GS
849 }
850 }
851 if (or_with->flags & ANYOF_EOS)
852 cl->flags |= ANYOF_EOS;
1aa99e6b 853
8951c461
KW
854 if (or_with->flags & ANYOF_FOLD)
855 cl->flags |= ANYOF_FOLD;
856
9826f543
KW
857 /* If both nodes match something outside the bitmap, but what they match
858 * outside is not the same pointer, and hence not easily compared, give up
859 * and allow the start class to match everything outside the bitmap */
3ff7ceb3 860 if (cl->flags & ANYOF_NONBITMAP && or_with->flags & ANYOF_NONBITMAP &&
1aa99e6b
IH
861 ARG(cl) != ARG(or_with)) {
862 cl->flags |= ANYOF_UNICODE_ALL;
1aa99e6b 863 }
9826f543 864
1aa99e6b
IH
865 if (or_with->flags & ANYOF_UNICODE_ALL) {
866 cl->flags |= ANYOF_UNICODE_ALL;
1aa99e6b 867 }
653099ff
GS
868}
869
a3621e74
YO
870#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
871#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
872#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
873#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
874
3dab1dad
YO
875
876#ifdef DEBUGGING
07be1b83 877/*
2b8b4781
NC
878 dump_trie(trie,widecharmap,revcharmap)
879 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
880 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
881
882 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
883 The _interim_ variants are used for debugging the interim
884 tables that are used to generate the final compressed
885 representation which is what dump_trie expects.
886
3dab1dad
YO
887 Part of the reason for their existance is to provide a form
888 of documentation as to how the different representations function.
07be1b83
YO
889
890*/
3dab1dad
YO
891
892/*
3dab1dad
YO
893 Dumps the final compressed table form of the trie to Perl_debug_log.
894 Used for debugging make_trie().
895*/
b9a59e08 896
3dab1dad 897STATIC void
2b8b4781
NC
898S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
899 AV *revcharmap, U32 depth)
3dab1dad
YO
900{
901 U32 state;
ab3bbdeb 902 SV *sv=sv_newmortal();
55eed653 903 int colwidth= widecharmap ? 6 : 4;
2e64971a 904 U16 word;
3dab1dad
YO
905 GET_RE_DEBUG_FLAGS_DECL;
906
7918f24d 907 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 908
3dab1dad
YO
909 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
910 (int)depth * 2 + 2,"",
911 "Match","Base","Ofs" );
912
913 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 914 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 915 if ( tmp ) {
ab3bbdeb
YO
916 PerlIO_printf( Perl_debug_log, "%*s",
917 colwidth,
ddc5bc0f 918 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
919 PL_colors[0], PL_colors[1],
920 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
921 PERL_PV_ESCAPE_FIRSTCHAR
922 )
923 );
3dab1dad
YO
924 }
925 }
926 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
927 (int)depth * 2 + 2,"");
928
929 for( state = 0 ; state < trie->uniquecharcount ; state++ )
ab3bbdeb 930 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
3dab1dad
YO
931 PerlIO_printf( Perl_debug_log, "\n");
932
1e2e3d02 933 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 934 const U32 base = trie->states[ state ].trans.base;
3dab1dad
YO
935
936 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
937
938 if ( trie->states[ state ].wordnum ) {
939 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
940 } else {
941 PerlIO_printf( Perl_debug_log, "%6s", "" );
942 }
943
944 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
945
946 if ( base ) {
947 U32 ofs = 0;
948
949 while( ( base + ofs < trie->uniquecharcount ) ||
950 ( base + ofs - trie->uniquecharcount < trie->lasttrans
951 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
952 ofs++;
953
954 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
955
956 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
957 if ( ( base + ofs >= trie->uniquecharcount ) &&
958 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
959 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
960 {
ab3bbdeb
YO
961 PerlIO_printf( Perl_debug_log, "%*"UVXf,
962 colwidth,
3dab1dad
YO
963 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
964 } else {
ab3bbdeb 965 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
3dab1dad
YO
966 }
967 }
968
969 PerlIO_printf( Perl_debug_log, "]");
970
971 }
972 PerlIO_printf( Perl_debug_log, "\n" );
973 }
2e64971a
DM
974 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
975 for (word=1; word <= trie->wordcount; word++) {
976 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
977 (int)word, (int)(trie->wordinfo[word].prev),
978 (int)(trie->wordinfo[word].len));
979 }
980 PerlIO_printf(Perl_debug_log, "\n" );
3dab1dad
YO
981}
982/*
3dab1dad
YO
983 Dumps a fully constructed but uncompressed trie in list form.
984 List tries normally only are used for construction when the number of
985 possible chars (trie->uniquecharcount) is very high.
986 Used for debugging make_trie().
987*/
988STATIC void
55eed653 989S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
990 HV *widecharmap, AV *revcharmap, U32 next_alloc,
991 U32 depth)
3dab1dad
YO
992{
993 U32 state;
ab3bbdeb 994 SV *sv=sv_newmortal();
55eed653 995 int colwidth= widecharmap ? 6 : 4;
3dab1dad 996 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
997
998 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
999
3dab1dad 1000 /* print out the table precompression. */
ab3bbdeb
YO
1001 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1002 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1003 "------:-----+-----------------\n" );
3dab1dad
YO
1004
1005 for( state=1 ; state < next_alloc ; state ++ ) {
1006 U16 charid;
1007
ab3bbdeb 1008 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
3dab1dad
YO
1009 (int)depth * 2 + 2,"", (UV)state );
1010 if ( ! trie->states[ state ].wordnum ) {
1011 PerlIO_printf( Perl_debug_log, "%5s| ","");
1012 } else {
1013 PerlIO_printf( Perl_debug_log, "W%4x| ",
1014 trie->states[ state ].wordnum
1015 );
1016 }
1017 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
2b8b4781 1018 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
ab3bbdeb
YO
1019 if ( tmp ) {
1020 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1021 colwidth,
ddc5bc0f 1022 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
1023 PL_colors[0], PL_colors[1],
1024 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1025 PERL_PV_ESCAPE_FIRSTCHAR
1026 ) ,
1e2e3d02
YO
1027 TRIE_LIST_ITEM(state,charid).forid,
1028 (UV)TRIE_LIST_ITEM(state,charid).newstate
1029 );
1030 if (!(charid % 10))
664e119d
RGS
1031 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1032 (int)((depth * 2) + 14), "");
1e2e3d02 1033 }
ab3bbdeb
YO
1034 }
1035 PerlIO_printf( Perl_debug_log, "\n");
3dab1dad
YO
1036 }
1037}
1038
1039/*
3dab1dad
YO
1040 Dumps a fully constructed but uncompressed trie in table form.
1041 This is the normal DFA style state transition table, with a few
1042 twists to facilitate compression later.
1043 Used for debugging make_trie().
1044*/
1045STATIC void
55eed653 1046S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
1047 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1048 U32 depth)
3dab1dad
YO
1049{
1050 U32 state;
1051 U16 charid;
ab3bbdeb 1052 SV *sv=sv_newmortal();
55eed653 1053 int colwidth= widecharmap ? 6 : 4;
3dab1dad 1054 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
1055
1056 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
3dab1dad
YO
1057
1058 /*
1059 print out the table precompression so that we can do a visual check
1060 that they are identical.
1061 */
1062
1063 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1064
1065 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 1066 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 1067 if ( tmp ) {
ab3bbdeb
YO
1068 PerlIO_printf( Perl_debug_log, "%*s",
1069 colwidth,
ddc5bc0f 1070 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
1071 PL_colors[0], PL_colors[1],
1072 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1073 PERL_PV_ESCAPE_FIRSTCHAR
1074 )
1075 );
3dab1dad
YO
1076 }
1077 }
1078
1079 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1080
1081 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb 1082 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
3dab1dad
YO
1083 }
1084
1085 PerlIO_printf( Perl_debug_log, "\n" );
1086
1087 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1088
1089 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1090 (int)depth * 2 + 2,"",
1091 (UV)TRIE_NODENUM( state ) );
1092
1093 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
1094 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1095 if (v)
1096 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1097 else
1098 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
3dab1dad
YO
1099 }
1100 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1101 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1102 } else {
1103 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1104 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1105 }
1106 }
07be1b83 1107}
3dab1dad
YO
1108
1109#endif
1110
2e64971a 1111
786e8c11
YO
1112/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1113 startbranch: the first branch in the whole branch sequence
1114 first : start branch of sequence of branch-exact nodes.
1115 May be the same as startbranch
1116 last : Thing following the last branch.
1117 May be the same as tail.
1118 tail : item following the branch sequence
1119 count : words in the sequence
1120 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1121 depth : indent depth
3dab1dad 1122
786e8c11 1123Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 1124
786e8c11
YO
1125A trie is an N'ary tree where the branches are determined by digital
1126decomposition of the key. IE, at the root node you look up the 1st character and
1127follow that branch repeat until you find the end of the branches. Nodes can be
1128marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 1129
786e8c11 1130 /he|she|his|hers/
72f13be8 1131
786e8c11
YO
1132would convert into the following structure. Numbers represent states, letters
1133following numbers represent valid transitions on the letter from that state, if
1134the number is in square brackets it represents an accepting state, otherwise it
1135will be in parenthesis.
07be1b83 1136
786e8c11
YO
1137 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1138 | |
1139 | (2)
1140 | |
1141 (1) +-i->(6)-+-s->[7]
1142 |
1143 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 1144
786e8c11
YO
1145 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1146
1147This shows that when matching against the string 'hers' we will begin at state 1
1148read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1149then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1150is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1151single traverse. We store a mapping from accepting to state to which word was
1152matched, and then when we have multiple possibilities we try to complete the
1153rest of the regex in the order in which they occured in the alternation.
1154
1155The only prior NFA like behaviour that would be changed by the TRIE support is
1156the silent ignoring of duplicate alternations which are of the form:
1157
1158 / (DUPE|DUPE) X? (?{ ... }) Y /x
1159
4b714af6 1160Thus EVAL blocks following a trie may be called a different number of times with
786e8c11
YO
1161and without the optimisation. With the optimisations dupes will be silently
1162ignored. This inconsistant behaviour of EVAL type nodes is well established as
1163the following demonstrates:
1164
1165 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1166
1167which prints out 'word' three times, but
1168
1169 'words'=~/(word|word|word)(?{ print $1 })S/
1170
1171which doesnt print it out at all. This is due to other optimisations kicking in.
1172
1173Example of what happens on a structural level:
1174
1175The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1176
1177 1: CURLYM[1] {1,32767}(18)
1178 5: BRANCH(8)
1179 6: EXACT <ac>(16)
1180 8: BRANCH(11)
1181 9: EXACT <ad>(16)
1182 11: BRANCH(14)
1183 12: EXACT <ab>(16)
1184 16: SUCCEED(0)
1185 17: NOTHING(18)
1186 18: END(0)
1187
1188This would be optimizable with startbranch=5, first=5, last=16, tail=16
1189and should turn into:
1190
1191 1: CURLYM[1] {1,32767}(18)
1192 5: TRIE(16)
1193 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1194 <ac>
1195 <ad>
1196 <ab>
1197 16: SUCCEED(0)
1198 17: NOTHING(18)
1199 18: END(0)
1200
1201Cases where tail != last would be like /(?foo|bar)baz/:
1202
1203 1: BRANCH(4)
1204 2: EXACT <foo>(8)
1205 4: BRANCH(7)
1206 5: EXACT <bar>(8)
1207 7: TAIL(8)
1208 8: EXACT <baz>(10)
1209 10: END(0)
1210
1211which would be optimizable with startbranch=1, first=1, last=7, tail=8
1212and would end up looking like:
1213
1214 1: TRIE(8)
1215 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1216 <foo>
1217 <bar>
1218 7: TAIL(8)
1219 8: EXACT <baz>(10)
1220 10: END(0)
1221
1222 d = uvuni_to_utf8_flags(d, uv, 0);
1223
1224is the recommended Unicode-aware way of saying
1225
1226 *(d++) = uv;
1227*/
1228
1e2e3d02 1229#define TRIE_STORE_REVCHAR \
786e8c11 1230 STMT_START { \
73031816
NC
1231 if (UTF) { \
1232 SV *zlopp = newSV(2); \
88c9ea1e
CB
1233 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1234 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
73031816
NC
1235 SvCUR_set(zlopp, kapow - flrbbbbb); \
1236 SvPOK_on(zlopp); \
1237 SvUTF8_on(zlopp); \
1238 av_push(revcharmap, zlopp); \
1239 } else { \
6bdeddd2 1240 char ooooff = (char)uvc; \
73031816
NC
1241 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1242 } \
1243 } STMT_END
786e8c11
YO
1244
1245#define TRIE_READ_CHAR STMT_START { \
1246 wordlen++; \
1247 if ( UTF ) { \
1248 if ( folder ) { \
1249 if ( foldlen > 0 ) { \
1250 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1251 foldlen -= len; \
1252 scan += len; \
1253 len = 0; \
1254 } else { \
1255 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1256 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1257 foldlen -= UNISKIP( uvc ); \
1258 scan = foldbuf + UNISKIP( uvc ); \
1259 } \
1260 } else { \
1261 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1262 } \
1263 } else { \
1264 uvc = (U32)*uc; \
1265 len = 1; \
1266 } \
1267} STMT_END
1268
1269
1270
1271#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1272 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
f9003953
NC
1273 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1274 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
786e8c11
YO
1275 } \
1276 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1277 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1278 TRIE_LIST_CUR( state )++; \
1279} STMT_END
07be1b83 1280
786e8c11
YO
1281#define TRIE_LIST_NEW(state) STMT_START { \
1282 Newxz( trie->states[ state ].trans.list, \
1283 4, reg_trie_trans_le ); \
1284 TRIE_LIST_CUR( state ) = 1; \
1285 TRIE_LIST_LEN( state ) = 4; \
1286} STMT_END
07be1b83 1287
786e8c11
YO
1288#define TRIE_HANDLE_WORD(state) STMT_START { \
1289 U16 dupe= trie->states[ state ].wordnum; \
1290 regnode * const noper_next = regnext( noper ); \
1291 \
786e8c11
YO
1292 DEBUG_r({ \
1293 /* store the word for dumping */ \
1294 SV* tmp; \
1295 if (OP(noper) != NOTHING) \
740cce10 1296 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 1297 else \
740cce10 1298 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 1299 av_push( trie_words, tmp ); \
786e8c11
YO
1300 }); \
1301 \
1302 curword++; \
2e64971a
DM
1303 trie->wordinfo[curword].prev = 0; \
1304 trie->wordinfo[curword].len = wordlen; \
1305 trie->wordinfo[curword].accept = state; \
786e8c11
YO
1306 \
1307 if ( noper_next < tail ) { \
1308 if (!trie->jump) \
c944940b 1309 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
7f69552c 1310 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
1311 if (!jumper) \
1312 jumper = noper_next; \
1313 if (!nextbranch) \
1314 nextbranch= regnext(cur); \
1315 } \
1316 \
1317 if ( dupe ) { \
2e64971a
DM
1318 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1319 /* chain, so that when the bits of chain are later */\
1320 /* linked together, the dups appear in the chain */\
1321 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1322 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
1323 } else { \
1324 /* we haven't inserted this word yet. */ \
1325 trie->states[ state ].wordnum = curword; \
1326 } \
1327} STMT_END
07be1b83 1328
3dab1dad 1329
786e8c11
YO
1330#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1331 ( ( base + charid >= ucharcount \
1332 && base + charid < ubound \
1333 && state == trie->trans[ base - ucharcount + charid ].check \
1334 && trie->trans[ base - ucharcount + charid ].next ) \
1335 ? trie->trans[ base - ucharcount + charid ].next \
1336 : ( state==1 ? special : 0 ) \
1337 )
3dab1dad 1338
786e8c11
YO
1339#define MADE_TRIE 1
1340#define MADE_JUMP_TRIE 2
1341#define MADE_EXACT_TRIE 4
3dab1dad 1342
a3621e74 1343STATIC I32
786e8c11 1344S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
a3621e74 1345{
27da23d5 1346 dVAR;
a3621e74
YO
1347 /* first pass, loop through and scan words */
1348 reg_trie_data *trie;
55eed653 1349 HV *widecharmap = NULL;
2b8b4781 1350 AV *revcharmap = newAV();
a3621e74 1351 regnode *cur;
9f7f3913 1352 const U32 uniflags = UTF8_ALLOW_DEFAULT;
a3621e74
YO
1353 STRLEN len = 0;
1354 UV uvc = 0;
1355 U16 curword = 0;
1356 U32 next_alloc = 0;
786e8c11
YO
1357 regnode *jumper = NULL;
1358 regnode *nextbranch = NULL;
7f69552c 1359 regnode *convert = NULL;
2e64971a 1360 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 1361 /* we just use folder as a flag in utf8 */
1e696034 1362 const U8 * folder = NULL;
a3621e74 1363
2b8b4781
NC
1364#ifdef DEBUGGING
1365 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1366 AV *trie_words = NULL;
1367 /* along with revcharmap, this only used during construction but both are
1368 * useful during debugging so we store them in the struct when debugging.
8e11feef 1369 */
2b8b4781
NC
1370#else
1371 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
3dab1dad 1372 STRLEN trie_charcount=0;
3dab1dad 1373#endif
2b8b4781 1374 SV *re_trie_maxbuff;
a3621e74 1375 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
1376
1377 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
1378#ifndef DEBUGGING
1379 PERL_UNUSED_ARG(depth);
1380#endif
a3621e74 1381
1e696034
KW
1382 switch (flags) {
1383 case EXACTFU: folder = PL_fold_latin1; break;
1384 case EXACTF: folder = PL_fold; break;
1385 case EXACTFL: folder = PL_fold_locale; break;
1386 }
1387
c944940b 1388 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 1389 trie->refcount = 1;
3dab1dad 1390 trie->startstate = 1;
786e8c11 1391 trie->wordcount = word_count;
f8fc2ecf 1392 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 1393 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
3dab1dad 1394 if (!(UTF && folder))
c944940b 1395 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
1396 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1397 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1398
a3621e74 1399 DEBUG_r({
2b8b4781 1400 trie_words = newAV();
a3621e74 1401 });
a3621e74 1402
0111c4fd 1403 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
a3621e74 1404 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 1405 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 1406 }
3dab1dad
YO
1407 DEBUG_OPTIMISE_r({
1408 PerlIO_printf( Perl_debug_log,
786e8c11 1409 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
3dab1dad
YO
1410 (int)depth * 2 + 2, "",
1411 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
786e8c11 1412 REG_NODE_NUM(last), REG_NODE_NUM(tail),
85c3142d 1413 (int)depth);
3dab1dad 1414 });
7f69552c
YO
1415
1416 /* Find the node we are going to overwrite */
1417 if ( first == startbranch && OP( last ) != BRANCH ) {
1418 /* whole branch chain */
1419 convert = first;
1420 } else {
1421 /* branch sub-chain */
1422 convert = NEXTOPER( first );
1423 }
1424
a3621e74
YO
1425 /* -- First loop and Setup --
1426
1427 We first traverse the branches and scan each word to determine if it
1428 contains widechars, and how many unique chars there are, this is
1429 important as we have to build a table with at least as many columns as we
1430 have unique chars.
1431
1432 We use an array of integers to represent the character codes 0..255
38a44b82 1433 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
a3621e74
YO
1434 native representation of the character value as the key and IV's for the
1435 coded index.
1436
1437 *TODO* If we keep track of how many times each character is used we can
1438 remap the columns so that the table compression later on is more
3b753521 1439 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
1440 middle and the least common are on the outside. IMO this would be better
1441 than a most to least common mapping as theres a decent chance the most
1442 common letter will share a node with the least common, meaning the node
1443 will not be compressable. With a middle is most common approach the worst
1444 case is when we have the least common nodes twice.
1445
1446 */
1447
a3621e74 1448 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
c445ea15 1449 regnode * const noper = NEXTOPER( cur );
e1ec3a88 1450 const U8 *uc = (U8*)STRING( noper );
a28509cc 1451 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1452 STRLEN foldlen = 0;
1453 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2af232bd 1454 const U8 *scan = (U8*)NULL;
07be1b83 1455 U32 wordlen = 0; /* required init */
02daf0ab
YO
1456 STRLEN chars = 0;
1457 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
a3621e74 1458
3dab1dad
YO
1459 if (OP(noper) == NOTHING) {
1460 trie->minlen= 0;
1461 continue;
1462 }
02daf0ab
YO
1463 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1464 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1465 regardless of encoding */
1466
a3621e74 1467 for ( ; uc < e ; uc += len ) {
3dab1dad 1468 TRIE_CHARCOUNT(trie)++;
a3621e74 1469 TRIE_READ_CHAR;
3dab1dad 1470 chars++;
a3621e74
YO
1471 if ( uvc < 256 ) {
1472 if ( !trie->charmap[ uvc ] ) {
1473 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1474 if ( folder )
1475 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
3dab1dad 1476 TRIE_STORE_REVCHAR;
a3621e74 1477 }
02daf0ab 1478 if ( set_bit ) {
62012aee
KW
1479 /* store the codepoint in the bitmap, and its folded
1480 * equivalent. */
02daf0ab 1481 TRIE_BITMAP_SET(trie,uvc);
0921ee73
T
1482
1483 /* store the folded codepoint */
1484 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1485
1486 if ( !UTF ) {
1487 /* store first byte of utf8 representation of
acdf4139
KW
1488 variant codepoints */
1489 if (! UNI_IS_INVARIANT(uvc)) {
1490 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc));
0921ee73
T
1491 }
1492 }
02daf0ab
YO
1493 set_bit = 0; /* We've done our bit :-) */
1494 }
a3621e74
YO
1495 } else {
1496 SV** svpp;
55eed653
NC
1497 if ( !widecharmap )
1498 widecharmap = newHV();
a3621e74 1499
55eed653 1500 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
1501
1502 if ( !svpp )
e4584336 1503 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
a3621e74
YO
1504
1505 if ( !SvTRUE( *svpp ) ) {
1506 sv_setiv( *svpp, ++trie->uniquecharcount );
3dab1dad 1507 TRIE_STORE_REVCHAR;
a3621e74
YO
1508 }
1509 }
1510 }
3dab1dad
YO
1511 if( cur == first ) {
1512 trie->minlen=chars;
1513 trie->maxlen=chars;
1514 } else if (chars < trie->minlen) {
1515 trie->minlen=chars;
1516 } else if (chars > trie->maxlen) {
1517 trie->maxlen=chars;
1518 }
1519
a3621e74
YO
1520 } /* end first pass */
1521 DEBUG_TRIE_COMPILE_r(
3dab1dad
YO
1522 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1523 (int)depth * 2 + 2,"",
55eed653 1524 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
1525 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1526 (int)trie->minlen, (int)trie->maxlen )
a3621e74 1527 );
a3621e74
YO
1528
1529 /*
1530 We now know what we are dealing with in terms of unique chars and
1531 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
1532 representation using a flat table will take. If it's over a reasonable
1533 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
1534 conservative but potentially much slower representation using an array
1535 of lists.
1536
1537 At the end we convert both representations into the same compressed
1538 form that will be used in regexec.c for matching with. The latter
1539 is a form that cannot be used to construct with but has memory
1540 properties similar to the list form and access properties similar
1541 to the table form making it both suitable for fast searches and
1542 small enough that its feasable to store for the duration of a program.
1543
1544 See the comment in the code where the compressed table is produced
1545 inplace from the flat tabe representation for an explanation of how
1546 the compression works.
1547
1548 */
1549
1550
2e64971a
DM
1551 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1552 prev_states[1] = 0;
1553
3dab1dad 1554 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
a3621e74
YO
1555 /*
1556 Second Pass -- Array Of Lists Representation
1557
1558 Each state will be represented by a list of charid:state records
1559 (reg_trie_trans_le) the first such element holds the CUR and LEN
1560 points of the allocated array. (See defines above).
1561
1562 We build the initial structure using the lists, and then convert
1563 it into the compressed table form which allows faster lookups
1564 (but cant be modified once converted).
a3621e74
YO
1565 */
1566
a3621e74
YO
1567 STRLEN transcount = 1;
1568
1e2e3d02
YO
1569 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1570 "%*sCompiling trie using list compiler\n",
1571 (int)depth * 2 + 2, ""));
446bd890 1572
c944940b
JH
1573 trie->states = (reg_trie_state *)
1574 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1575 sizeof(reg_trie_state) );
a3621e74
YO
1576 TRIE_LIST_NEW(1);
1577 next_alloc = 2;
1578
1579 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1580
c445ea15
AL
1581 regnode * const noper = NEXTOPER( cur );
1582 U8 *uc = (U8*)STRING( noper );
1583 const U8 * const e = uc + STR_LEN( noper );
1584 U32 state = 1; /* required init */
1585 U16 charid = 0; /* sanity init */
1586 U8 *scan = (U8*)NULL; /* sanity init */
1587 STRLEN foldlen = 0; /* required init */
07be1b83 1588 U32 wordlen = 0; /* required init */
c445ea15
AL
1589 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1590
3dab1dad 1591 if (OP(noper) != NOTHING) {
786e8c11 1592 for ( ; uc < e ; uc += len ) {
c445ea15 1593
786e8c11 1594 TRIE_READ_CHAR;
c445ea15 1595
786e8c11
YO
1596 if ( uvc < 256 ) {
1597 charid = trie->charmap[ uvc ];
c445ea15 1598 } else {
55eed653 1599 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11
YO
1600 if ( !svpp ) {
1601 charid = 0;
1602 } else {
1603 charid=(U16)SvIV( *svpp );
1604 }
c445ea15 1605 }
786e8c11
YO
1606 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1607 if ( charid ) {
a3621e74 1608
786e8c11
YO
1609 U16 check;
1610 U32 newstate = 0;
a3621e74 1611
786e8c11
YO
1612 charid--;
1613 if ( !trie->states[ state ].trans.list ) {
1614 TRIE_LIST_NEW( state );
c445ea15 1615 }
786e8c11
YO
1616 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1617 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1618 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1619 break;
1620 }
1621 }
1622 if ( ! newstate ) {
1623 newstate = next_alloc++;
2e64971a 1624 prev_states[newstate] = state;
786e8c11
YO
1625 TRIE_LIST_PUSH( state, charid, newstate );
1626 transcount++;
1627 }
1628 state = newstate;
1629 } else {
1630 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
c445ea15 1631 }
a28509cc 1632 }
c445ea15 1633 }
3dab1dad 1634 TRIE_HANDLE_WORD(state);
a3621e74
YO
1635
1636 } /* end second pass */
1637
1e2e3d02
YO
1638 /* next alloc is the NEXT state to be allocated */
1639 trie->statecount = next_alloc;
c944940b
JH
1640 trie->states = (reg_trie_state *)
1641 PerlMemShared_realloc( trie->states,
1642 next_alloc
1643 * sizeof(reg_trie_state) );
a3621e74 1644
3dab1dad 1645 /* and now dump it out before we compress it */
2b8b4781
NC
1646 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1647 revcharmap, next_alloc,
1648 depth+1)
1e2e3d02 1649 );
a3621e74 1650
c944940b
JH
1651 trie->trans = (reg_trie_trans *)
1652 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
1653 {
1654 U32 state;
a3621e74
YO
1655 U32 tp = 0;
1656 U32 zp = 0;
1657
1658
1659 for( state=1 ; state < next_alloc ; state ++ ) {
1660 U32 base=0;
1661
1662 /*
1663 DEBUG_TRIE_COMPILE_MORE_r(
1664 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1665 );
1666 */
1667
1668 if (trie->states[state].trans.list) {
1669 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1670 U16 maxid=minid;
a28509cc 1671 U16 idx;
a3621e74
YO
1672
1673 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
1674 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1675 if ( forid < minid ) {
1676 minid=forid;
1677 } else if ( forid > maxid ) {
1678 maxid=forid;
1679 }
a3621e74
YO
1680 }
1681 if ( transcount < tp + maxid - minid + 1) {
1682 transcount *= 2;
c944940b
JH
1683 trie->trans = (reg_trie_trans *)
1684 PerlMemShared_realloc( trie->trans,
446bd890
NC
1685 transcount
1686 * sizeof(reg_trie_trans) );
a3621e74
YO
1687 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1688 }
1689 base = trie->uniquecharcount + tp - minid;
1690 if ( maxid == minid ) {
1691 U32 set = 0;
1692 for ( ; zp < tp ; zp++ ) {
1693 if ( ! trie->trans[ zp ].next ) {
1694 base = trie->uniquecharcount + zp - minid;
1695 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1696 trie->trans[ zp ].check = state;
1697 set = 1;
1698 break;
1699 }
1700 }
1701 if ( !set ) {
1702 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1703 trie->trans[ tp ].check = state;
1704 tp++;
1705 zp = tp;
1706 }
1707 } else {
1708 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15 1709 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
a3621e74
YO
1710 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1711 trie->trans[ tid ].check = state;
1712 }
1713 tp += ( maxid - minid + 1 );
1714 }
1715 Safefree(trie->states[ state ].trans.list);
1716 }
1717 /*
1718 DEBUG_TRIE_COMPILE_MORE_r(
1719 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1720 );
1721 */
1722 trie->states[ state ].trans.base=base;
1723 }
cc601c31 1724 trie->lasttrans = tp + 1;
a3621e74
YO
1725 }
1726 } else {
1727 /*
1728 Second Pass -- Flat Table Representation.
1729
1730 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1731 We know that we will need Charcount+1 trans at most to store the data
1732 (one row per char at worst case) So we preallocate both structures
1733 assuming worst case.
1734
1735 We then construct the trie using only the .next slots of the entry
1736 structs.
1737
3b753521 1738 We use the .check field of the first entry of the node temporarily to
a3621e74
YO
1739 make compression both faster and easier by keeping track of how many non
1740 zero fields are in the node.
1741
1742 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1743 transition.
1744
1745 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1746 number representing the first entry of the node, and state as a
1747 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1748 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1749 are 2 entrys per node. eg:
1750
1751 A B A B
1752 1. 2 4 1. 3 7
1753 2. 0 3 3. 0 5
1754 3. 0 0 5. 0 0
1755 4. 0 0 7. 0 0
1756
1757 The table is internally in the right hand, idx form. However as we also
1758 have to deal with the states array which is indexed by nodenum we have to
1759 use TRIE_NODENUM() to convert.
1760
1761 */
1e2e3d02
YO
1762 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1763 "%*sCompiling trie using table compiler\n",
1764 (int)depth * 2 + 2, ""));
3dab1dad 1765
c944940b
JH
1766 trie->trans = (reg_trie_trans *)
1767 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1768 * trie->uniquecharcount + 1,
1769 sizeof(reg_trie_trans) );
1770 trie->states = (reg_trie_state *)
1771 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1772 sizeof(reg_trie_state) );
a3621e74
YO
1773 next_alloc = trie->uniquecharcount + 1;
1774
3dab1dad 1775
a3621e74
YO
1776 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1777
c445ea15 1778 regnode * const noper = NEXTOPER( cur );
a28509cc
AL
1779 const U8 *uc = (U8*)STRING( noper );
1780 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1781
1782 U32 state = 1; /* required init */
1783
1784 U16 charid = 0; /* sanity init */
1785 U32 accept_state = 0; /* sanity init */
1786 U8 *scan = (U8*)NULL; /* sanity init */
1787
1788 STRLEN foldlen = 0; /* required init */
07be1b83 1789 U32 wordlen = 0; /* required init */
a3621e74
YO
1790 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1791
3dab1dad 1792 if ( OP(noper) != NOTHING ) {
786e8c11 1793 for ( ; uc < e ; uc += len ) {
a3621e74 1794
786e8c11 1795 TRIE_READ_CHAR;
a3621e74 1796
786e8c11
YO
1797 if ( uvc < 256 ) {
1798 charid = trie->charmap[ uvc ];
1799 } else {
55eed653 1800 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11 1801 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 1802 }
786e8c11
YO
1803 if ( charid ) {
1804 charid--;
1805 if ( !trie->trans[ state + charid ].next ) {
1806 trie->trans[ state + charid ].next = next_alloc;
1807 trie->trans[ state ].check++;
2e64971a
DM
1808 prev_states[TRIE_NODENUM(next_alloc)]
1809 = TRIE_NODENUM(state);
786e8c11
YO
1810 next_alloc += trie->uniquecharcount;
1811 }
1812 state = trie->trans[ state + charid ].next;
1813 } else {
1814 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1815 }
1816 /* charid is now 0 if we dont know the char read, or nonzero if we do */
a3621e74 1817 }
a3621e74 1818 }
3dab1dad
YO
1819 accept_state = TRIE_NODENUM( state );
1820 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
1821
1822 } /* end second pass */
1823
3dab1dad 1824 /* and now dump it out before we compress it */
2b8b4781
NC
1825 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1826 revcharmap,
1827 next_alloc, depth+1));
a3621e74 1828
a3621e74
YO
1829 {
1830 /*
1831 * Inplace compress the table.*
1832
1833 For sparse data sets the table constructed by the trie algorithm will
1834 be mostly 0/FAIL transitions or to put it another way mostly empty.
1835 (Note that leaf nodes will not contain any transitions.)
1836
1837 This algorithm compresses the tables by eliminating most such
1838 transitions, at the cost of a modest bit of extra work during lookup:
1839
1840 - Each states[] entry contains a .base field which indicates the
1841 index in the state[] array wheres its transition data is stored.
1842
3b753521 1843 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
1844
1845 - If .base is nonzero then charid is added to it to find an entry in
1846 the trans array.
1847
1848 -If trans[states[state].base+charid].check!=state then the
1849 transition is taken to be a 0/Fail transition. Thus if there are fail
1850 transitions at the front of the node then the .base offset will point
1851 somewhere inside the previous nodes data (or maybe even into a node
1852 even earlier), but the .check field determines if the transition is
1853 valid.
1854
786e8c11 1855 XXX - wrong maybe?
a3621e74 1856 The following process inplace converts the table to the compressed
3b753521 1857 table: We first do not compress the root node 1,and mark all its
a3621e74 1858 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
1859 allows us to do a DFA construction from the compressed table later,
1860 and ensures that any .base pointers we calculate later are greater
1861 than 0.
a3621e74
YO
1862
1863 - We set 'pos' to indicate the first entry of the second node.
1864
1865 - We then iterate over the columns of the node, finding the first and
1866 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1867 and set the .check pointers accordingly, and advance pos
1868 appropriately and repreat for the next node. Note that when we copy
1869 the next pointers we have to convert them from the original
1870 NODEIDX form to NODENUM form as the former is not valid post
1871 compression.
1872
1873 - If a node has no transitions used we mark its base as 0 and do not
1874 advance the pos pointer.
1875
1876 - If a node only has one transition we use a second pointer into the
1877 structure to fill in allocated fail transitions from other states.
1878 This pointer is independent of the main pointer and scans forward
1879 looking for null transitions that are allocated to a state. When it
1880 finds one it writes the single transition into the "hole". If the
786e8c11 1881 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
1882
1883 - Once compressed we can Renew/realloc the structures to release the
1884 excess space.
1885
1886 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1887 specifically Fig 3.47 and the associated pseudocode.
1888
1889 demq
1890 */
a3b680e6 1891 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 1892 U32 state, charid;
a3621e74 1893 U32 pos = 0, zp=0;
1e2e3d02 1894 trie->statecount = laststate;
a3621e74
YO
1895
1896 for ( state = 1 ; state < laststate ; state++ ) {
1897 U8 flag = 0;
a28509cc
AL
1898 const U32 stateidx = TRIE_NODEIDX( state );
1899 const U32 o_used = trie->trans[ stateidx ].check;
1900 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
1901 trie->trans[ stateidx ].check = 0;
1902
1903 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1904 if ( flag || trie->trans[ stateidx + charid ].next ) {
1905 if ( trie->trans[ stateidx + charid ].next ) {
1906 if (o_used == 1) {
1907 for ( ; zp < pos ; zp++ ) {
1908 if ( ! trie->trans[ zp ].next ) {
1909 break;
1910 }
1911 }
1912 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1913 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1914 trie->trans[ zp ].check = state;
1915 if ( ++zp > pos ) pos = zp;
1916 break;
1917 }
1918 used--;
1919 }
1920 if ( !flag ) {
1921 flag = 1;
1922 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1923 }
1924 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1925 trie->trans[ pos ].check = state;
1926 pos++;
1927 }
1928 }
1929 }
cc601c31 1930 trie->lasttrans = pos + 1;
c944940b
JH
1931 trie->states = (reg_trie_state *)
1932 PerlMemShared_realloc( trie->states, laststate
1933 * sizeof(reg_trie_state) );
a3621e74 1934 DEBUG_TRIE_COMPILE_MORE_r(
e4584336 1935 PerlIO_printf( Perl_debug_log,
3dab1dad
YO
1936 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1937 (int)depth * 2 + 2,"",
1938 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
5d7488b2
AL
1939 (IV)next_alloc,
1940 (IV)pos,
a3621e74
YO
1941 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1942 );
1943
1944 } /* end table compress */
1945 }
1e2e3d02
YO
1946 DEBUG_TRIE_COMPILE_MORE_r(
1947 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1948 (int)depth * 2 + 2, "",
1949 (UV)trie->statecount,
1950 (UV)trie->lasttrans)
1951 );
cc601c31 1952 /* resize the trans array to remove unused space */
c944940b
JH
1953 trie->trans = (reg_trie_trans *)
1954 PerlMemShared_realloc( trie->trans, trie->lasttrans
1955 * sizeof(reg_trie_trans) );
a3621e74 1956
3b753521 1957 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
1958 U8 nodetype =(U8)(flags & 0xFF);
1959 char *str=NULL;
786e8c11 1960
07be1b83 1961#ifdef DEBUGGING
e62cc96a 1962 regnode *optimize = NULL;
7122b237
YO
1963#ifdef RE_TRACK_PATTERN_OFFSETS
1964
b57a0404
JH
1965 U32 mjd_offset = 0;
1966 U32 mjd_nodelen = 0;
7122b237
YO
1967#endif /* RE_TRACK_PATTERN_OFFSETS */
1968#endif /* DEBUGGING */
a3621e74 1969 /*
3dab1dad
YO
1970 This means we convert either the first branch or the first Exact,
1971 depending on whether the thing following (in 'last') is a branch
1972 or not and whther first is the startbranch (ie is it a sub part of
1973 the alternation or is it the whole thing.)
3b753521 1974 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 1975 the whole branch sequence, including the first.
a3621e74 1976 */
3dab1dad 1977 /* Find the node we are going to overwrite */
7f69552c 1978 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 1979 /* branch sub-chain */
3dab1dad 1980 NEXT_OFF( first ) = (U16)(last - first);
7122b237 1981#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
1982 DEBUG_r({
1983 mjd_offset= Node_Offset((convert));
1984 mjd_nodelen= Node_Length((convert));
1985 });
7122b237 1986#endif
7f69552c 1987 /* whole branch chain */
7122b237
YO
1988 }
1989#ifdef RE_TRACK_PATTERN_OFFSETS
1990 else {
7f69552c
YO
1991 DEBUG_r({
1992 const regnode *nop = NEXTOPER( convert );
1993 mjd_offset= Node_Offset((nop));
1994 mjd_nodelen= Node_Length((nop));
1995 });
07be1b83
YO
1996 }
1997 DEBUG_OPTIMISE_r(
1998 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1999 (int)depth * 2 + 2, "",
786e8c11 2000 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 2001 );
7122b237 2002#endif
3dab1dad
YO
2003 /* But first we check to see if there is a common prefix we can
2004 split out as an EXACT and put in front of the TRIE node. */
2005 trie->startstate= 1;
55eed653 2006 if ( trie->bitmap && !widecharmap && !trie->jump ) {
3dab1dad 2007 U32 state;
1e2e3d02 2008 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 2009 U32 ofs = 0;
8e11feef
RGS
2010 I32 idx = -1;
2011 U32 count = 0;
2012 const U32 base = trie->states[ state ].trans.base;
a3621e74 2013
3dab1dad 2014 if ( trie->states[state].wordnum )
8e11feef 2015 count = 1;
a3621e74 2016
8e11feef 2017 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
cc601c31
YO
2018 if ( ( base + ofs >= trie->uniquecharcount ) &&
2019 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
a3621e74
YO
2020 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2021 {
3dab1dad 2022 if ( ++count > 1 ) {
2b8b4781 2023 SV **tmp = av_fetch( revcharmap, ofs, 0);
07be1b83 2024 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 2025 if ( state == 1 ) break;
3dab1dad
YO
2026 if ( count == 2 ) {
2027 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2028 DEBUG_OPTIMISE_r(
8e11feef
RGS
2029 PerlIO_printf(Perl_debug_log,
2030 "%*sNew Start State=%"UVuf" Class: [",
2031 (int)depth * 2 + 2, "",
786e8c11 2032 (UV)state));
be8e71aa 2033 if (idx >= 0) {
2b8b4781 2034 SV ** const tmp = av_fetch( revcharmap, idx, 0);
be8e71aa 2035 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 2036
3dab1dad 2037 TRIE_BITMAP_SET(trie,*ch);
8e11feef
RGS
2038 if ( folder )
2039 TRIE_BITMAP_SET(trie, folder[ *ch ]);
3dab1dad 2040 DEBUG_OPTIMISE_r(
f1f66076 2041 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
3dab1dad 2042 );
8e11feef
RGS
2043 }
2044 }
2045 TRIE_BITMAP_SET(trie,*ch);
2046 if ( folder )
2047 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2048 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2049 }
2050 idx = ofs;
2051 }
3dab1dad
YO
2052 }
2053 if ( count == 1 ) {
2b8b4781 2054 SV **tmp = av_fetch( revcharmap, idx, 0);
c490c714
YO
2055 STRLEN len;
2056 char *ch = SvPV( *tmp, len );
de734bd5
A
2057 DEBUG_OPTIMISE_r({
2058 SV *sv=sv_newmortal();
8e11feef
RGS
2059 PerlIO_printf( Perl_debug_log,
2060 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2061 (int)depth * 2 + 2, "",
de734bd5
A
2062 (UV)state, (UV)idx,
2063 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2064 PL_colors[0], PL_colors[1],
2065 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2066 PERL_PV_ESCAPE_FIRSTCHAR
2067 )
2068 );
2069 });
3dab1dad
YO
2070 if ( state==1 ) {
2071 OP( convert ) = nodetype;
2072 str=STRING(convert);
2073 STR_LEN(convert)=0;
2074 }
c490c714
YO
2075 STR_LEN(convert) += len;
2076 while (len--)
de734bd5 2077 *str++ = *ch++;
8e11feef 2078 } else {
f9049ba1 2079#ifdef DEBUGGING
8e11feef
RGS
2080 if (state>1)
2081 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
f9049ba1 2082#endif
8e11feef
RGS
2083 break;
2084 }
2085 }
2e64971a 2086 trie->prefixlen = (state-1);
3dab1dad 2087 if (str) {
8e11feef 2088 regnode *n = convert+NODE_SZ_STR(convert);
07be1b83 2089 NEXT_OFF(convert) = NODE_SZ_STR(convert);
8e11feef 2090 trie->startstate = state;
07be1b83
YO
2091 trie->minlen -= (state - 1);
2092 trie->maxlen -= (state - 1);
33809eae
JH
2093#ifdef DEBUGGING
2094 /* At least the UNICOS C compiler choked on this
2095 * being argument to DEBUG_r(), so let's just have
2096 * it right here. */
2097 if (
2098#ifdef PERL_EXT_RE_BUILD
2099 1
2100#else
2101 DEBUG_r_TEST
2102#endif
2103 ) {
2104 regnode *fix = convert;
2105 U32 word = trie->wordcount;
2106 mjd_nodelen++;
2107 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2108 while( ++fix < n ) {
2109 Set_Node_Offset_Length(fix, 0, 0);
2110 }
2111 while (word--) {
2112 SV ** const tmp = av_fetch( trie_words, word, 0 );
2113 if (tmp) {
2114 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2115 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2116 else
2117 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2118 }
2119 }
2120 }
2121#endif
8e11feef
RGS
2122 if (trie->maxlen) {
2123 convert = n;
2124 } else {
3dab1dad 2125 NEXT_OFF(convert) = (U16)(tail - convert);
a5ca303d 2126 DEBUG_r(optimize= n);
3dab1dad
YO
2127 }
2128 }
2129 }
a5ca303d
YO
2130 if (!jumper)
2131 jumper = last;
3dab1dad 2132 if ( trie->maxlen ) {
8e11feef
RGS
2133 NEXT_OFF( convert ) = (U16)(tail - convert);
2134 ARG_SET( convert, data_slot );
786e8c11
YO
2135 /* Store the offset to the first unabsorbed branch in
2136 jump[0], which is otherwise unused by the jump logic.
2137 We use this when dumping a trie and during optimisation. */
2138 if (trie->jump)
7f69552c 2139 trie->jump[0] = (U16)(nextbranch - convert);
a5ca303d 2140
6c48061a
YO
2141 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2142 * and there is a bitmap
2143 * and the first "jump target" node we found leaves enough room
2144 * then convert the TRIE node into a TRIEC node, with the bitmap
2145 * embedded inline in the opcode - this is hypothetically faster.
2146 */
2147 if ( !trie->states[trie->startstate].wordnum
2148 && trie->bitmap
2149 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
786e8c11
YO
2150 {
2151 OP( convert ) = TRIEC;
2152 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
446bd890 2153 PerlMemShared_free(trie->bitmap);
786e8c11
YO
2154 trie->bitmap= NULL;
2155 } else
2156 OP( convert ) = TRIE;
a3621e74 2157
3dab1dad
YO
2158 /* store the type in the flags */
2159 convert->flags = nodetype;
a5ca303d
YO
2160 DEBUG_r({
2161 optimize = convert
2162 + NODE_STEP_REGNODE
2163 + regarglen[ OP( convert ) ];
2164 });
2165 /* XXX We really should free up the resource in trie now,
2166 as we won't use them - (which resources?) dmq */
3dab1dad 2167 }
a3621e74 2168 /* needed for dumping*/
e62cc96a 2169 DEBUG_r(if (optimize) {
07be1b83 2170 regnode *opt = convert;
bcdf7404 2171
e62cc96a 2172 while ( ++opt < optimize) {
07be1b83
YO
2173 Set_Node_Offset_Length(opt,0,0);
2174 }
786e8c11
YO
2175 /*
2176 Try to clean up some of the debris left after the
2177 optimisation.
a3621e74 2178 */
786e8c11 2179 while( optimize < jumper ) {
07be1b83 2180 mjd_nodelen += Node_Length((optimize));
a3621e74 2181 OP( optimize ) = OPTIMIZED;
07be1b83 2182 Set_Node_Offset_Length(optimize,0,0);
a3621e74
YO
2183 optimize++;
2184 }
07be1b83 2185 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
a3621e74
YO
2186 });
2187 } /* end node insert */
2e64971a
DM
2188
2189 /* Finish populating the prev field of the wordinfo array. Walk back
2190 * from each accept state until we find another accept state, and if
2191 * so, point the first word's .prev field at the second word. If the
2192 * second already has a .prev field set, stop now. This will be the
2193 * case either if we've already processed that word's accept state,
3b753521
FN
2194 * or that state had multiple words, and the overspill words were
2195 * already linked up earlier.
2e64971a
DM
2196 */
2197 {
2198 U16 word;
2199 U32 state;
2200 U16 prev;
2201
2202 for (word=1; word <= trie->wordcount; word++) {
2203 prev = 0;
2204 if (trie->wordinfo[word].prev)
2205 continue;
2206 state = trie->wordinfo[word].accept;
2207 while (state) {
2208 state = prev_states[state];
2209 if (!state)
2210 break;
2211 prev = trie->states[state].wordnum;
2212 if (prev)
2213 break;
2214 }
2215 trie->wordinfo[word].prev = prev;
2216 }
2217 Safefree(prev_states);
2218 }
2219
2220
2221 /* and now dump out the compressed format */
2222 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2223
55eed653 2224 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2b8b4781
NC
2225#ifdef DEBUGGING
2226 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2227 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2228#else
2229 SvREFCNT_dec(revcharmap);
07be1b83 2230#endif
786e8c11
YO
2231 return trie->jump
2232 ? MADE_JUMP_TRIE
2233 : trie->startstate>1
2234 ? MADE_EXACT_TRIE
2235 : MADE_TRIE;
2236}
2237
2238STATIC void
2239S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2240{
3b753521 2241/* The Trie is constructed and compressed now so we can build a fail array if it's needed
786e8c11
YO
2242
2243 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2244 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2245 ISBN 0-201-10088-6
2246
2247 We find the fail state for each state in the trie, this state is the longest proper
3b753521
FN
2248 suffix of the current state's 'word' that is also a proper prefix of another word in our
2249 trie. State 1 represents the word '' and is thus the default fail state. This allows
786e8c11
YO
2250 the DFA not to have to restart after its tried and failed a word at a given point, it
2251 simply continues as though it had been matching the other word in the first place.
2252 Consider
2253 'abcdgu'=~/abcdefg|cdgu/
2254 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
3b753521
FN
2255 fail, which would bring us to the state representing 'd' in the second word where we would
2256 try 'g' and succeed, proceeding to match 'cdgu'.
786e8c11
YO
2257 */
2258 /* add a fail transition */
3251b653
NC
2259 const U32 trie_offset = ARG(source);
2260 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
786e8c11
YO
2261 U32 *q;
2262 const U32 ucharcount = trie->uniquecharcount;
1e2e3d02 2263 const U32 numstates = trie->statecount;
786e8c11
YO
2264 const U32 ubound = trie->lasttrans + ucharcount;
2265 U32 q_read = 0;
2266 U32 q_write = 0;
2267 U32 charid;
2268 U32 base = trie->states[ 1 ].trans.base;
2269 U32 *fail;
2270 reg_ac_data *aho;
2271 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2272 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2273
2274 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
786e8c11
YO
2275#ifndef DEBUGGING
2276 PERL_UNUSED_ARG(depth);
2277#endif
2278
2279
2280 ARG_SET( stclass, data_slot );
c944940b 2281 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
f8fc2ecf 2282 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3251b653 2283 aho->trie=trie_offset;
446bd890
NC
2284 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2285 Copy( trie->states, aho->states, numstates, reg_trie_state );
786e8c11 2286 Newxz( q, numstates, U32);
c944940b 2287 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
786e8c11
YO
2288 aho->refcount = 1;
2289 fail = aho->fail;
2290 /* initialize fail[0..1] to be 1 so that we always have
2291 a valid final fail state */
2292 fail[ 0 ] = fail[ 1 ] = 1;
2293
2294 for ( charid = 0; charid < ucharcount ; charid++ ) {
2295 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2296 if ( newstate ) {
2297 q[ q_write ] = newstate;
2298 /* set to point at the root */
2299 fail[ q[ q_write++ ] ]=1;
2300 }
2301 }
2302 while ( q_read < q_write) {
2303 const U32 cur = q[ q_read++ % numstates ];
2304 base = trie->states[ cur ].trans.base;
2305
2306 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2307 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2308 if (ch_state) {
2309 U32 fail_state = cur;
2310 U32 fail_base;
2311 do {
2312 fail_state = fail[ fail_state ];
2313 fail_base = aho->states[ fail_state ].trans.base;
2314 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2315
2316 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2317 fail[ ch_state ] = fail_state;
2318 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2319 {
2320 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2321 }
2322 q[ q_write++ % numstates] = ch_state;
2323 }
2324 }
2325 }
2326 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2327 when we fail in state 1, this allows us to use the
2328 charclass scan to find a valid start char. This is based on the principle
2329 that theres a good chance the string being searched contains lots of stuff
2330 that cant be a start char.
2331 */
2332 fail[ 0 ] = fail[ 1 ] = 0;
2333 DEBUG_TRIE_COMPILE_r({
6d99fb9b
JH
2334 PerlIO_printf(Perl_debug_log,
2335 "%*sStclass Failtable (%"UVuf" states): 0",
2336 (int)(depth * 2), "", (UV)numstates
1e2e3d02 2337 );
786e8c11
YO
2338 for( q_read=1; q_read<numstates; q_read++ ) {
2339 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2340 }
2341 PerlIO_printf(Perl_debug_log, "\n");
2342 });
2343 Safefree(q);
2344 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
a3621e74
YO
2345}
2346
786e8c11 2347
a3621e74 2348/*
5d1c421c
JH
2349 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2350 * These need to be revisited when a newer toolchain becomes available.
2351 */
2352#if defined(__sparc64__) && defined(__GNUC__)
2353# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2354# undef SPARC64_GCC_WORKAROUND
2355# define SPARC64_GCC_WORKAROUND 1
2356# endif
2357#endif
2358
07be1b83 2359#define DEBUG_PEEP(str,scan,depth) \
b515a41d 2360 DEBUG_OPTIMISE_r({if (scan){ \
07be1b83
YO
2361 SV * const mysv=sv_newmortal(); \
2362 regnode *Next = regnext(scan); \
2363 regprop(RExC_rx, mysv, scan); \
7f69552c 2364 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
07be1b83
YO
2365 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2366 Next ? (REG_NODE_NUM(Next)) : 0 ); \
b515a41d 2367 }});
07be1b83 2368
1de06328
YO
2369
2370
2371
2372
07be1b83
YO
2373#define JOIN_EXACT(scan,min,flags) \
2374 if (PL_regkind[OP(scan)] == EXACT) \
2375 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2376
be8e71aa 2377STATIC U32
07be1b83
YO
2378S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2379 /* Merge several consecutive EXACTish nodes into one. */
2380 regnode *n = regnext(scan);
2381 U32 stringok = 1;
2382 regnode *next = scan + NODE_SZ_STR(scan);
2383 U32 merged = 0;
2384 U32 stopnow = 0;
2385#ifdef DEBUGGING
2386 regnode *stop = scan;
72f13be8 2387 GET_RE_DEBUG_FLAGS_DECL;
f9049ba1 2388#else
d47053eb
RGS
2389 PERL_UNUSED_ARG(depth);
2390#endif
7918f24d
NC
2391
2392 PERL_ARGS_ASSERT_JOIN_EXACT;
d47053eb 2393#ifndef EXPERIMENTAL_INPLACESCAN
f9049ba1
SP
2394 PERL_UNUSED_ARG(flags);
2395 PERL_UNUSED_ARG(val);
07be1b83 2396#endif
07be1b83
YO
2397 DEBUG_PEEP("join",scan,depth);
2398
2399 /* Skip NOTHING, merge EXACT*. */
2400 while (n &&
2401 ( PL_regkind[OP(n)] == NOTHING ||
2402 (stringok && (OP(n) == OP(scan))))
2403 && NEXT_OFF(n)
2404 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2405
2406 if (OP(n) == TAIL || n > next)
2407 stringok = 0;
2408 if (PL_regkind[OP(n)] == NOTHING) {
07be1b83
YO
2409 DEBUG_PEEP("skip:",n,depth);
2410 NEXT_OFF(scan) += NEXT_OFF(n);
2411 next = n + NODE_STEP_REGNODE;
2412#ifdef DEBUGGING
2413 if (stringok)
2414 stop = n;
2415#endif
2416 n = regnext(n);
2417 }
2418 else if (stringok) {
786e8c11 2419 const unsigned int oldl = STR_LEN(scan);
07be1b83
YO
2420 regnode * const nnext = regnext(n);
2421
2422 DEBUG_PEEP("merg",n,depth);
2423
2424 merged++;
2425 if (oldl + STR_LEN(n) > U8_MAX)
2426 break;
2427 NEXT_OFF(scan) += NEXT_OFF(n);
2428 STR_LEN(scan) += STR_LEN(n);
2429 next = n + NODE_SZ_STR(n);
2430 /* Now we can overwrite *n : */
2431 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2432#ifdef DEBUGGING
2433 stop = next - 1;
2434#endif
2435 n = nnext;
2436 if (stopnow) break;
2437 }
2438
d47053eb
RGS
2439#ifdef EXPERIMENTAL_INPLACESCAN
2440 if (flags && !NEXT_OFF(n)) {
2441 DEBUG_PEEP("atch", val, depth);
2442 if (reg_off_by_arg[OP(n)]) {
2443 ARG_SET(n, val - n);
2444 }
2445 else {
2446 NEXT_OFF(n) = val - n;
2447 }
2448 stopnow = 1;
2449 }
07be1b83
YO
2450#endif
2451 }
ced7f090
KW
2452#define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390
2453#define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS
2454#define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0
2455#define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS
2c2b7f86
KW
2456
2457 if (UTF
2458 && ( OP(scan) == EXACTF || OP(scan) == EXACTFU)
2459 && ( STR_LEN(scan) >= 6 ) )
2460 {
07be1b83
YO
2461 /*
2462 Two problematic code points in Unicode casefolding of EXACT nodes:
2463
2464 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2465 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2466
2467 which casefold to
2468
2469 Unicode UTF-8
2470
2471 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2472 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2473
2474 This means that in case-insensitive matching (or "loose matching",
2475 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2476 length of the above casefolded versions) can match a target string
2477 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2478 This would rather mess up the minimum length computation.
2479
2480 What we'll do is to look for the tail four bytes, and then peek
2481 at the preceding two bytes to see whether we need to decrease
2482 the minimum length by four (six minus two).
2483
2484 Thanks to the design of UTF-8, there cannot be false matches:
2485 A sequence of valid UTF-8 bytes cannot be a subsequence of
2486 another valid sequence of UTF-8 bytes.
2487
2488 */
2489 char * const s0 = STRING(scan), *s, *t;
2490 char * const s1 = s0 + STR_LEN(scan) - 1;
2491 char * const s2 = s1 - 4;
e294cc5d
JH
2492#ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2493 const char t0[] = "\xaf\x49\xaf\x42";
2494#else
07be1b83 2495 const char t0[] = "\xcc\x88\xcc\x81";
e294cc5d 2496#endif
07be1b83
YO
2497 const char * const t1 = t0 + 3;
2498
2499 for (s = s0 + 2;
2500 s < s2 && (t = ninstr(s, s1, t0, t1));
2501 s = t + 4) {
e294cc5d
JH
2502#ifdef EBCDIC
2503 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2504 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2505#else
07be1b83
YO
2506 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2507 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
e294cc5d 2508#endif
07be1b83
YO
2509 *min -= 4;
2510 }
2511 }
2512
2513#ifdef DEBUGGING
2514 /* Allow dumping */
2515 n = scan + NODE_SZ_STR(scan);
2516 while (n <= stop) {
2517 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2518 OP(n) = OPTIMIZED;
2519 NEXT_OFF(n) = 0;
2520 }
2521 n++;
2522 }
2523#endif
2524 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2525 return stopnow;
2526}
2527
653099ff
GS
2528/* REx optimizer. Converts nodes into quickier variants "in place".
2529 Finds fixed substrings. */
2530
a0288114 2531/* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
c277df42
IZ
2532 to the position after last scanned or to NULL. */
2533
40d049e4
YO
2534#define INIT_AND_WITHP \
2535 assert(!and_withp); \
2536 Newx(and_withp,1,struct regnode_charclass_class); \
2537 SAVEFREEPV(and_withp)
07be1b83 2538
b515a41d
YO
2539/* this is a chain of data about sub patterns we are processing that
2540 need to be handled seperately/specially in study_chunk. Its so
2541 we can simulate recursion without losing state. */
2542struct scan_frame;
2543typedef struct scan_frame {
2544 regnode *last; /* last node to process in this frame */
2545 regnode *next; /* next node to process when last is reached */
2546 struct scan_frame *prev; /*previous frame*/
2547 I32 stop; /* what stopparen do we use */
2548} scan_frame;
2549
304ee84b
YO
2550
2551#define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2552
e1d1eefb
YO
2553#define CASE_SYNST_FNC(nAmE) \
2554case nAmE: \
2555 if (flags & SCF_DO_STCLASS_AND) { \
2556 for (value = 0; value < 256; value++) \
2557 if (!is_ ## nAmE ## _cp(value)) \
2558 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2559 } \
2560 else { \
2561 for (value = 0; value < 256; value++) \
2562 if (is_ ## nAmE ## _cp(value)) \
2563 ANYOF_BITMAP_SET(data->start_class, value); \
2564 } \
2565 break; \
2566case N ## nAmE: \
2567 if (flags & SCF_DO_STCLASS_AND) { \
2568 for (value = 0; value < 256; value++) \
2569 if (is_ ## nAmE ## _cp(value)) \
2570 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2571 } \
2572 else { \
2573 for (value = 0; value < 256; value++) \
2574 if (!is_ ## nAmE ## _cp(value)) \
2575 ANYOF_BITMAP_SET(data->start_class, value); \
2576 } \
2577 break
2578
2579
2580
76e3520e 2581STATIC I32
40d049e4 2582S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
1de06328 2583 I32 *minlenp, I32 *deltap,
40d049e4
YO
2584 regnode *last,
2585 scan_data_t *data,
2586 I32 stopparen,
2587 U8* recursed,
2588 struct regnode_charclass_class *and_withp,
2589 U32 flags, U32 depth)
c277df42
IZ
2590 /* scanp: Start here (read-write). */
2591 /* deltap: Write maxlen-minlen here. */
2592 /* last: Stop before this one. */
40d049e4
YO
2593 /* data: string data about the pattern */
2594 /* stopparen: treat close N as END */
2595 /* recursed: which subroutines have we recursed into */
2596 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
c277df42 2597{
97aff369 2598 dVAR;
c277df42
IZ
2599 I32 min = 0, pars = 0, code;
2600 regnode *scan = *scanp, *next;
2601 I32 delta = 0;
2602 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
aca2d497 2603 int is_inf_internal = 0; /* The studied chunk is infinite */
c277df42
IZ
2604 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2605 scan_data_t data_fake;
a3621e74 2606 SV *re_trie_maxbuff = NULL;
786e8c11 2607 regnode *first_non_open = scan;
e2e6a0f1 2608 I32 stopmin = I32_MAX;
8aa23a47 2609 scan_frame *frame = NULL;
a3621e74 2610 GET_RE_DEBUG_FLAGS_DECL;
8aa23a47 2611
7918f24d
NC
2612 PERL_ARGS_ASSERT_STUDY_CHUNK;
2613
13a24bad 2614#ifdef DEBUGGING
40d049e4 2615 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
13a24bad 2616#endif
40d049e4 2617
786e8c11 2618 if ( depth == 0 ) {
40d049e4 2619 while (first_non_open && OP(first_non_open) == OPEN)
786e8c11
YO
2620 first_non_open=regnext(first_non_open);
2621 }
2622
b81d288d 2623
8aa23a47
YO
2624 fake_study_recurse:
2625 while ( scan && OP(scan) != END && scan < last ){
2626 /* Peephole optimizer: */
304ee84b 2627 DEBUG_STUDYDATA("Peep:", data,depth);
8aa23a47
YO
2628 DEBUG_PEEP("Peep",scan,depth);
2629 JOIN_EXACT(scan,&min,0);
2630
2631 /* Follow the next-chain of the current node and optimize
2632 away all the NOTHINGs from it. */
2633 if (OP(scan) != CURLYX) {
2634 const int max = (reg_off_by_arg[OP(scan)]
2635 ? I32_MAX
2636 /* I32 may be smaller than U16 on CRAYs! */
2637 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2638 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2639 int noff;
2640 regnode *n = scan;
2641
2642 /* Skip NOTHING and LONGJMP. */
2643 while ((n = regnext(n))
2644 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2645 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2646 && off + noff < max)
2647 off += noff;
2648 if (reg_off_by_arg[OP(scan)])
2649 ARG(scan) = off;
2650 else
2651 NEXT_OFF(scan) = off;
2652 }
a3621e74 2653
c277df42 2654
8aa23a47
YO
2655
2656 /* The principal pseudo-switch. Cannot be a switch, since we
2657 look into several different things. */
2658 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2659 || OP(scan) == IFTHEN) {
2660 next = regnext(scan);
2661 code = OP(scan);
2662 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2663
2664 if (OP(next) == code || code == IFTHEN) {
2665 /* NOTE - There is similar code to this block below for handling
2666 TRIE nodes on a re-study. If you change stuff here check there
2667 too. */
2668 I32 max1 = 0, min1 = I32_MAX, num = 0;
2669 struct regnode_charclass_class accum;
2670 regnode * const startbranch=scan;
2671
2672 if (flags & SCF_DO_SUBSTR)
304ee84b 2673 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
8aa23a47
YO
2674 if (flags & SCF_DO_STCLASS)
2675 cl_init_zero(pRExC_state, &accum);
2676
2677 while (OP(scan) == code) {
2678 I32 deltanext, minnext, f = 0, fake;
2679 struct regnode_charclass_class this_class;
2680
2681 num++;
2682 data_fake.flags = 0;
2683 if (data) {
2684 data_fake.whilem_c = data->whilem_c;
2685 data_fake.last_closep = data->last_closep;
2686 }
2687 else
2688 data_fake.last_closep = &fake;
58e23c8d
YO
2689
2690 data_fake.pos_delta = delta;
8aa23a47
YO
2691 next = regnext(scan);
2692 scan = NEXTOPER(scan);
2693 if (code != BRANCH)
c277df42 2694 scan = NEXTOPER(scan);
8aa23a47
YO
2695 if (flags & SCF_DO_STCLASS) {
2696 cl_init(pRExC_state, &this_class);
2697 data_fake.start_class = &this_class;
2698 f = SCF_DO_STCLASS_AND;
58e23c8d 2699 }
8aa23a47
YO
2700 if (flags & SCF_WHILEM_VISITED_POS)
2701 f |= SCF_WHILEM_VISITED_POS;
2702
2703 /* we suppose the run is continuous, last=next...*/
2704 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2705 next, &data_fake,
2706 stopparen, recursed, NULL, f,depth+1);
2707 if (min1 > minnext)
2708 min1 = minnext;
2709 if (max1 < minnext + deltanext)
2710 max1 = minnext + deltanext;
2711 if (deltanext == I32_MAX)
2712 is_inf = is_inf_internal = 1;
2713 scan = next;
2714 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2715 pars++;
2716 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2717 if ( stopmin > minnext)
2718 stopmin = min + min1;
2719 flags &= ~SCF_DO_SUBSTR;
2720 if (data)
2721 data->flags |= SCF_SEEN_ACCEPT;
2722 }
2723 if (data) {
2724 if (data_fake.flags & SF_HAS_EVAL)
2725 data->flags |= SF_HAS_EVAL;
2726 data->whilem_c = data_fake.whilem_c;
3dab1dad 2727 }
8aa23a47
YO
2728 if (flags & SCF_DO_STCLASS)
2729 cl_or(pRExC_state, &accum, &this_class);
2730 }
2731 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2732 min1 = 0;
2733 if (flags & SCF_DO_SUBSTR) {
2734 data->pos_min += min1;
2735 data->pos_delta += max1 - min1;
2736 if (max1 != min1 || is_inf)
2737 data->longest = &(data->longest_float);
2738 }
2739 min += min1;
2740 delta += max1 - min1;
2741 if (flags & SCF_DO_STCLASS_OR) {
2742 cl_or(pRExC_state, data->start_class, &accum);
2743 if (min1) {
2744 cl_and(data->start_class, and_withp);
2745 flags &= ~SCF_DO_STCLASS;
653099ff 2746 }
8aa23a47
YO
2747 }
2748 else if (flags & SCF_DO_STCLASS_AND) {
2749 if (min1) {
2750 cl_and(data->start_class, &accum);
2751 flags &= ~SCF_DO_STCLASS;
de0c8cb8 2752 }
8aa23a47
YO
2753 else {
2754 /* Switch to OR mode: cache the old value of
2755 * data->start_class */
2756 INIT_AND_WITHP;
2757 StructCopy(data->start_class, and_withp,
2758 struct regnode_charclass_class);
2759 flags &= ~SCF_DO_STCLASS_AND;
2760 StructCopy(&accum, data->start_class,
2761 struct regnode_charclass_class);
2762 flags |= SCF_DO_STCLASS_OR;
2763 data->start_class->flags |= ANYOF_EOS;
de0c8cb8 2764 }
8aa23a47 2765 }
a3621e74 2766
8aa23a47
YO
2767 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2768 /* demq.
a3621e74 2769
8aa23a47
YO
2770 Assuming this was/is a branch we are dealing with: 'scan' now
2771 points at the item that follows the branch sequence, whatever
2772 it is. We now start at the beginning of the sequence and look
2773 for subsequences of
a3621e74 2774
8aa23a47
YO
2775 BRANCH->EXACT=>x1
2776 BRANCH->EXACT=>x2
2777 tail
a3621e74 2778
8aa23a47 2779 which would be constructed from a pattern like /A|LIST|OF|WORDS/
a3621e74 2780
8aa23a47
YO
2781 If we can find such a subseqence we need to turn the first
2782 element into a trie and then add the subsequent branch exact
2783 strings to the trie.
a3621e74 2784
8aa23a47 2785 We have two cases
a3621e74 2786
3b753521 2787 1. patterns where the whole set of branches can be converted.
a3621e74 2788
8aa23a47 2789 2. patterns where only a subset can be converted.
a3621e74 2790
8aa23a47
YO
2791 In case 1 we can replace the whole set with a single regop
2792 for the trie. In case 2 we need to keep the start and end
3b753521 2793 branches so
a3621e74 2794
8aa23a47
YO
2795 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2796 becomes BRANCH TRIE; BRANCH X;
786e8c11 2797
8aa23a47
YO
2798 There is an additional case, that being where there is a
2799 common prefix, which gets split out into an EXACT like node
2800 preceding the TRIE node.
a3621e74 2801
8aa23a47
YO
2802 If x(1..n)==tail then we can do a simple trie, if not we make
2803 a "jump" trie, such that when we match the appropriate word
2804 we "jump" to the appopriate tail node. Essentailly we turn
2805 a nested if into a case structure of sorts.
b515a41d 2806
8aa23a47
YO
2807 */
2808
2809 int made=0;
2810 if (!re_trie_maxbuff) {
2811 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2812 if (!SvIOK(re_trie_maxbuff))
2813 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2814 }
2815 if ( SvIV(re_trie_maxbuff)>=0 ) {
2816 regnode *cur;
2817 regnode *first = (regnode *)NULL;
2818 regnode *last = (regnode *)NULL;
2819 regnode *tail = scan;
2820 U8 optype = 0;
2821 U32 count=0;
a3621e74
YO
2822
2823#ifdef DEBUGGING
8aa23a47 2824 SV * const mysv = sv_newmortal(); /* for dumping */
a3621e74 2825#endif
8aa23a47
YO
2826 /* var tail is used because there may be a TAIL
2827 regop in the way. Ie, the exacts will point to the
2828 thing following the TAIL, but the last branch will
2829 point at the TAIL. So we advance tail. If we
2830 have nested (?:) we may have to move through several
2831 tails.
2832 */
2833
2834 while ( OP( tail ) == TAIL ) {
2835 /* this is the TAIL generated by (?:) */
2836 tail = regnext( tail );
2837 }
a3621e74 2838
8aa23a47
YO
2839
2840 DEBUG_OPTIMISE_r({
2841 regprop(RExC_rx, mysv, tail );
2842 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2843 (int)depth * 2 + 2, "",
2844 "Looking for TRIE'able sequences. Tail node is: ",
2845 SvPV_nolen_const( mysv )
2846 );
2847 });
2848
2849 /*
2850
2851 step through the branches, cur represents each
2852 branch, noper is the first thing to be matched
2853 as part of that branch and noper_next is the
2854 regnext() of that node. if noper is an EXACT
2855 and noper_next is the same as scan (our current
2856 position in the regex) then the EXACT branch is
2857 a possible optimization target. Once we have
2858 two or more consequetive such branches we can
2859 create a trie of the EXACT's contents and stich
2860 it in place. If the sequence represents all of
2861 the branches we eliminate the whole thing and
2862 replace it with a single TRIE. If it is a
2863 subsequence then we need to stitch it in. This
2864 means the first branch has to remain, and needs
2865 to be repointed at the item on the branch chain
2866 following the last branch optimized. This could
2867 be either a BRANCH, in which case the
2868 subsequence is internal, or it could be the
2869 item following the branch sequence in which
2870 case the subsequence is at the end.
2871
2872 */
2873
2874 /* dont use tail as the end marker for this traverse */
2875 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2876 regnode * const noper = NEXTOPER( cur );
b515a41d 2877#if defined(DEBUGGING) || defined(NOJUMPTRIE)
8aa23a47 2878 regnode * const noper_next = regnext( noper );
b515a41d
YO
2879#endif
2880
8aa23a47
YO
2881 DEBUG_OPTIMISE_r({
2882 regprop(RExC_rx, mysv, cur);
2883 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2884 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2885
2886 regprop(RExC_rx, mysv, noper);
2887 PerlIO_printf( Perl_debug_log, " -> %s",
2888 SvPV_nolen_const(mysv));
2889
2890 if ( noper_next ) {
2891 regprop(RExC_rx, mysv, noper_next );
2892 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2893 SvPV_nolen_const(mysv));
2894 }
2895 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2896 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2897 });
2898 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2899 : PL_regkind[ OP( noper ) ] == EXACT )
2900 || OP(noper) == NOTHING )
786e8c11 2901#ifdef NOJUMPTRIE
8aa23a47 2902 && noper_next == tail
786e8c11 2903#endif
8aa23a47
YO
2904 && count < U16_MAX)
2905 {
2906 count++;
2907 if ( !first || optype == NOTHING ) {
2908 if (!first) first = cur;
2909 optype = OP( noper );
2910 } else {
2911 last = cur;
2912 }
2913 } else {
a0a388a1 2914/*
0abd0d78
YO
2915 Currently we do not believe that the trie logic can
2916 handle case insensitive matching properly when the
2917 pattern is not unicode (thus forcing unicode semantics).
2918
2919 If/when this is fixed the following define can be swapped
2920 in below to fully enable trie logic.
2921
a0a388a1 2922#define TRIE_TYPE_IS_SAFE 1
0abd0d78
YO
2923
2924*/
2925#define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2926
a0a388a1 2927 if ( last && TRIE_TYPE_IS_SAFE ) {
8aa23a47
YO
2928 make_trie( pRExC_state,
2929 startbranch, first, cur, tail, count,
2930 optype, depth+1 );
2931 }
2932 if ( PL_regkind[ OP( noper ) ] == EXACT
786e8c11 2933#ifdef NOJUMPTRIE
8aa23a47 2934 && noper_next == tail
786e8c11 2935#endif
8aa23a47
YO
2936 ){
2937 count = 1;
2938 first = cur;
2939 optype = OP( noper );
2940 } else {
2941 count = 0;
2942 first = NULL;
2943 optype = 0;
2944 }
2945 last = NULL;
2946 }
2947 }
2948 DEBUG_OPTIMISE_r({
2949 regprop(RExC_rx, mysv, cur);
2950 PerlIO_printf( Perl_debug_log,
2951 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2952 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2953
2954 });
a0a388a1
YO
2955
2956 if ( last && TRIE_TYPE_IS_SAFE ) {
8aa23a47 2957 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3dab1dad 2958#ifdef TRIE_STUDY_OPT
8aa23a47
YO
2959 if ( ((made == MADE_EXACT_TRIE &&
2960 startbranch == first)
2961 || ( first_non_open == first )) &&
2962 depth==0 ) {
2963 flags |= SCF_TRIE_RESTUDY;
2964 if ( startbranch == first
2965 && scan == tail )
2966 {
2967 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2968 }
2969 }
3dab1dad 2970#endif
8aa23a47
YO
2971 }
2972 }
2973
2974 } /* do trie */
2975
653099ff 2976 }
8aa23a47
YO
2977 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2978 scan = NEXTOPER(NEXTOPER(scan));
2979 } else /* single branch is optimized. */
2980 scan = NEXTOPER(scan);
2981 continue;
2982 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2983 scan_frame *newframe = NULL;
2984 I32 paren;
2985 regnode *start;
2986 regnode *end;
2987
2988 if (OP(scan) != SUSPEND) {
2989 /* set the pointer */
2990 if (OP(scan) == GOSUB) {
2991 paren = ARG(scan);
2992 RExC_recurse[ARG2L(scan)] = scan;
2993 start = RExC_open_parens[paren-1];
2994 end = RExC_close_parens[paren-1];
2995 } else {
2996 paren = 0;
f8fc2ecf 2997 start = RExC_rxi->program + 1;
8aa23a47
YO
2998 end = RExC_opend;
2999 }
3000 if (!recursed) {
3001 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
3002 SAVEFREEPV(recursed);
3003 }
3004 if (!PAREN_TEST(recursed,paren+1)) {
3005 PAREN_SET(recursed,paren+1);
3006 Newx(newframe,1,scan_frame);
3007 } else {
3008 if (flags & SCF_DO_SUBSTR) {
304ee84b 3009 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3010 data->longest = &(data->longest_float);
3011 }
3012 is_inf = is_inf_internal = 1;
3013 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3014 cl_anything(pRExC_state, data->start_class);
3015 flags &= ~SCF_DO_STCLASS;
3016 }
3017 } else {
3018 Newx(newframe,1,scan_frame);
3019 paren = stopparen;
3020 start = scan+2;
3021 end = regnext(scan);
3022 }
3023 if (newframe) {
3024 assert(start);
3025 assert(end);
3026 SAVEFREEPV(newframe);
3027 newframe->next = regnext(scan);
3028 newframe->last = last;
3029 newframe->stop = stopparen;
3030 newframe->prev = frame;
3031
3032 frame = newframe;
3033 scan = start;
3034 stopparen = paren;
3035 last = end;
3036
3037 continue;
3038 }
3039 }
3040 else if (OP(scan) == EXACT) {
3041 I32 l = STR_LEN(scan);
3042 UV uc;
3043 if (UTF) {
3044 const U8 * const s = (U8*)STRING(scan);
3045 l = utf8_length(s, s + l);
3046 uc = utf8_to_uvchr(s, NULL);
3047 } else {
3048 uc = *((U8*)STRING(scan));
3049 }
3050 min += l;
3051 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3052 /* The code below prefers earlier match for fixed
3053 offset, later match for variable offset. */
3054 if (data->last_end == -1) { /* Update the start info. */
3055 data->last_start_min = data->pos_min;
3056 data->last_start_max = is_inf
3057 ? I32_MAX : data->pos_min + data->pos_delta;
b515a41d 3058 }
8aa23a47
YO
3059 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3060 if (UTF)
3061 SvUTF8_on(data->last_found);
3062 {
3063 SV * const sv = data->last_found;
3064 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3065 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3066 if (mg && mg->mg_len >= 0)
3067 mg->mg_len += utf8_length((U8*)STRING(scan),
3068 (U8*)STRING(scan)+STR_LEN(scan));
b515a41d 3069 }
8aa23a47
YO
3070 data->last_end = data->pos_min + l;
3071 data->pos_min += l; /* As in the first entry. */
3072 data->flags &= ~SF_BEFORE_EOL;
3073 }
3074 if (flags & SCF_DO_STCLASS_AND) {
3075 /* Check whether it is compatible with what we know already! */
3076 int compat = 1;
3077
54251c2e
KW
3078
3079 /* If compatibile, we or it in below. It is compatible if is
3080 * in the bitmp and either 1) its bit or its fold is set, or 2)
3081 * it's for a locale. Even if there isn't unicode semantics
3082 * here, at runtime there may be because of matching against a
3083 * utf8 string, so accept a possible false positive for
3084 * latin1-range folds */
8aa23a47
YO
3085 if (uc >= 0x100 ||
3086 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3087 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3088 && (!(data->start_class->flags & ANYOF_FOLD)
54251c2e 3089 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
8aa23a47
YO
3090 )
3091 compat = 0;
3092 ANYOF_CLASS_ZERO(data->start_class);
3093 ANYOF_BITMAP_ZERO(data->start_class);
3094 if (compat)
3095 ANYOF_BITMAP_SET(data->start_class, uc);
3096 data->start_class->flags &= ~ANYOF_EOS;
3097 if (uc < 0x100)
3098 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3099 }
3100 else if (flags & SCF_DO_STCLASS_OR) {
3101 /* false positive possible if the class is case-folded */
3102 if (uc < 0x100)
3103 ANYOF_BITMAP_SET(data->start_class, uc);
3104 else
3105 data->start_class->flags |= ANYOF_UNICODE_ALL;
3106 data->start_class->flags &= ~ANYOF_EOS;
3107 cl_and(data->start_class, and_withp);
3108 }
3109 flags &= ~SCF_DO_STCLASS;
3110 }
3111 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3112 I32 l = STR_LEN(scan);
3113 UV uc = *((U8*)STRING(scan));
3114
3115 /* Search for fixed substrings supports EXACT only. */
3116 if (flags & SCF_DO_SUBSTR) {
3117 assert(data);
304ee84b 3118 SCAN_COMMIT(pRExC_state, data, minlenp);
8aa23a47
YO
3119 }
3120 if (UTF) {
3121 const U8 * const s = (U8 *)STRING(scan);
3122 l = utf8_length(s, s + l);
3123 uc = utf8_to_uvchr(s, NULL);
3124 }
3125 min += l;
3126 if (flags & SCF_DO_SUBSTR)
3127 data->pos_min += l;
3128 if (flags & SCF_DO_STCLASS_AND) {
3129 /* Check whether it is compatible with what we know already! */
3130 int compat = 1;
8aa23a47 3131 if (uc >= 0x100 ||
54251c2e
KW
3132 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3133 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3134 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3135 {
8aa23a47 3136 compat = 0;
54251c2e 3137 }
8aa23a47
YO
3138 ANYOF_CLASS_ZERO(data->start_class);
3139 ANYOF_BITMAP_ZERO(data->start_class);
3140 if (compat) {
3141 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff 3142 data->start_class->flags &= ~ANYOF_EOS;
8aa23a47 3143 data->start_class->flags |= ANYOF_FOLD;
970c8436 3144 if (OP(scan) == EXACTFL) {
8aa23a47 3145 data->start_class->flags |= ANYOF_LOCALE;
970c8436
KW
3146 }
3147 else {
3148
54251c2e
KW
3149 /* Also set the other member of the fold pair. In case
3150 * that unicode semantics is called for at runtime, use
3151 * the full latin1 fold. (Can't do this for locale,
3152 * because not known until runtime */
3153 ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]);
970c8436 3154 }
653099ff 3155 }
8aa23a47
YO
3156 }
3157 else if (flags & SCF_DO_STCLASS_OR) {
3158 if (data->start_class->flags & ANYOF_FOLD) {
3159 /* false positive possible if the class is case-folded.
3160 Assume that the locale settings are the same... */
970c8436 3161 if (uc < 0x100) {
1aa99e6b 3162 ANYOF_BITMAP_SET(data->start_class, uc);
970c8436
KW
3163 if (OP(scan) != EXACTFL) {
3164
3165 /* And set the other member of the fold pair, but
3166 * can't do that in locale because not known until
3167 * run-time */
3168 ANYOF_BITMAP_SET(data->start_class,
54251c2e 3169 PL_fold_latin1[uc]);
970c8436
KW
3170 }
3171 }
653099ff
GS
3172 data->start_class->flags &= ~ANYOF_EOS;
3173 }
8aa23a47 3174 cl_and(data->start_class, and_withp);
653099ff 3175 }
8aa23a47
YO
3176 flags &= ~SCF_DO_STCLASS;
3177 }
e52fc539 3178 else if (REGNODE_VARIES(OP(scan))) {
8aa23a47
YO
3179 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3180 I32 f = flags, pos_before = 0;
3181 regnode * const oscan = scan;
3182 struct regnode_charclass_class this_class;
3183 struct regnode_charclass_class *oclass = NULL;
3184 I32 next_is_eval = 0;
3185
3186 switch (PL_regkind[OP(scan)]) {
3187 case WHILEM: /* End of (?:...)* . */
3188 scan = NEXTOPER(scan);
3189 goto finish;
3190 case PLUS:
3191 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3192 next = NEXTOPER(scan);
3193 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3194 mincount = 1;
3195 maxcount = REG_INFTY;
3196 next = regnext(scan);
3197 scan = NEXTOPER(scan);
3198 goto do_curly;
3199 }
3200 }
3201 if (flags & SCF_DO_SUBSTR)
3202 data->pos_min++;
3203 min++;
3204 /* Fall through. */
3205 case STAR:
3206 if (flags & SCF_DO_STCLASS) {
3207 mincount = 0;
3208 maxcount = REG_INFTY;
3209 next = regnext(scan);
3210 scan = NEXTOPER(scan);
3211 goto do_curly;
3212 }
3213 is_inf = is_inf_internal = 1;
3214 scan = regnext(scan);
c277df42 3215 if (flags & SCF_DO_SUBSTR) {
304ee84b 3216 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
8aa23a47 3217 data->longest = &(data->longest_float);
c277df42 3218 }
8aa23a47
YO
3219 goto optimize_curly_tail;
3220 case CURLY:
3221 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3222 && (scan->flags == stopparen))
3223 {
3224 mincount = 1;
3225 maxcount = 1;
3226 } else {
3227 mincount = ARG1(scan);
3228 maxcount = ARG2(scan);
653099ff 3229 }
8aa23a47
YO
3230 next = regnext(scan);
3231 if (OP(scan) == CURLYX) {
3232 I32 lp = (data ? *(data->last_closep) : 0);
3233 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
653099ff 3234 }
8aa23a47
YO
3235 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3236 next_is_eval = (OP(scan) == EVAL);
3237 do_curly:
3238 if (flags & SCF_DO_SUBSTR) {
304ee84b 3239 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
8aa23a47 3240 pos_before = data->pos_min;
b45f050a 3241 }
8aa23a47
YO
3242 if (data) {
3243 fl = data->flags;
3244 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3245 if (is_inf)
3246 data->flags |= SF_IS_INF;
3247 }
3248 if (flags & SCF_DO_STCLASS) {
3249 cl_init(pRExC_state, &this_class);
3250 oclass = data->start_class;
3251 data->start_class = &this_class;
3252 f |= SCF_DO_STCLASS_AND;
3253 f &= ~SCF_DO_STCLASS_OR;
3254 }
779bcb7d
NC
3255 /* Exclude from super-linear cache processing any {n,m}
3256 regops for which the combination of input pos and regex
3257 pos is not enough information to determine if a match
3258 will be possible.
3259
3260 For example, in the regex /foo(bar\s*){4,8}baz/ with the
3261 regex pos at the \s*, the prospects for a match depend not
3262 only on the input position but also on how many (bar\s*)
3263 repeats into the {4,8} we are. */
3264 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
8aa23a47 3265 f &= ~SCF_WHILEM_VISITED_POS;
b45f050a 3266
8aa23a47
YO
3267 /* This will finish on WHILEM, setting scan, or on NULL: */
3268 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3269 last, data, stopparen, recursed, NULL,
3270 (mincount == 0
3271 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
b515a41d 3272
8aa23a47
YO
3273 if (flags & SCF_DO_STCLASS)
3274 data->start_class = oclass;
3275 if (mincount == 0 || minnext == 0) {
3276 if (flags & SCF_DO_STCLASS_OR) {
3277 cl_or(pRExC_state, data->start_class, &this_class);
3278 }
3279 else if (flags & SCF_DO_STCLASS_AND) {
3280 /* Switch to OR mode: cache the old value of
3281 * data->start_class */
3282 INIT_AND_WITHP;
3283 StructCopy(data->start_class, and_withp,
3284 struct regnode_charclass_class);
3285 flags &= ~SCF_DO_STCLASS_AND;
3286 StructCopy(&this_class, data->start_class,
3287 struct regnode_charclass_class);
3288 flags |= SCF_DO_STCLASS_OR;
3289 data->start_class->flags |= ANYOF_EOS;
3290 }
3291 } else { /* Non-zero len */
3292 if (flags & SCF_DO_STCLASS_OR) {
3293 cl_or(pRExC_state, data->start_class, &this_class);
3294 cl_and(data->start_class, and_withp);
3295 }
3296 else if (flags & SCF_DO_STCLASS_AND)
3297 cl_and(data->start_class, &this_class);
3298 flags &= ~SCF_DO_STCLASS;
3299 }
3300 if (!scan) /* It was not CURLYX, but CURLY. */
3301 scan = next;
3302 if ( /* ? quantifier ok, except for (?{ ... }) */
3303 (next_is_eval || !(mincount == 0 && maxcount == 1))
3304 && (minnext == 0) && (deltanext == 0)
3305 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
668c081a 3306 && maxcount <= REG_INFTY/3) /* Complement check for big count */
8aa23a47 3307 {
668c081a
NC
3308 ckWARNreg(RExC_parse,
3309 "Quantifier unexpected on zero-length expression");
8aa23a47
YO
3310 }
3311
3312 min += minnext * mincount;
3313 is_inf_internal |= ((maxcount == REG_INFTY
3314 && (minnext + deltanext) > 0)
3315 || deltanext == I32_MAX);
3316 is_inf |= is_inf_internal;
3317 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3318
3319 /* Try powerful optimization CURLYX => CURLYN. */
3320 if ( OP(oscan) == CURLYX && data
3321 && data->flags & SF_IN_PAR
3322 && !(data->flags & SF_HAS_EVAL)
3323 && !deltanext && minnext == 1 ) {
3324 /* Try to optimize to CURLYN. */
3325 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3326 regnode * const nxt1 = nxt;
497b47a8 3327#ifdef DEBUGGING
8aa23a47 3328 regnode *nxt2;
497b47a8 3329#endif
c277df42 3330
8aa23a47
YO
3331 /* Skip open. */
3332 nxt = regnext(nxt);
e52fc539 3333 if (!REGNODE_SIMPLE(OP(nxt))
8aa23a47
YO
3334 && !(PL_regkind[OP(nxt)] == EXACT
3335 && STR_LEN(nxt) == 1))
3336 goto nogo;
497b47a8 3337#ifdef DEBUGGING
8aa23a47 3338 nxt2 = nxt;
497b47a8 3339#endif
8aa23a47
YO
3340 nxt = regnext(nxt);
3341 if (OP(nxt) != CLOSE)
3342 goto nogo;
3343 if (RExC_open_parens) {
3344 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3345 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3346 }
3347 /* Now we know that nxt2 is the only contents: */
3348 oscan->flags = (U8)ARG(nxt);
3349 OP(oscan) = CURLYN;
3350 OP(nxt1) = NOTHING; /* was OPEN. */
40d049e4 3351
c277df42 3352#ifdef DEBUGGING
8aa23a47 3353 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
fda99bee
KW
3354 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3355 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
8aa23a47
YO
3356 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3357 OP(nxt + 1) = OPTIMIZED; /* was count. */
fda99bee 3358 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
b81d288d 3359#endif
8aa23a47
YO
3360 }
3361 nogo:
3362
3363 /* Try optimization CURLYX => CURLYM. */
3364 if ( OP(oscan) == CURLYX && data
3365 && !(data->flags & SF_HAS_PAR)
3366 && !(data->flags & SF_HAS_EVAL)
3367 && !deltanext /* atom is fixed width */
3368 && minnext != 0 /* CURLYM can't handle zero width */
3369 ) {
3370 /* XXXX How to optimize if data == 0? */
3371 /* Optimize to a simpler form. */
3372 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3373 regnode *nxt2;
3374
3375 OP(oscan) = CURLYM;
3376 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3377 && (OP(nxt2) != WHILEM))
3378 nxt = nxt2;
3379 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3380 /* Need to optimize away parenths. */
b3c0965f 3381 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
8aa23a47
YO
3382 /* Set the parenth number. */
3383 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3384
8aa23a47
YO
3385 oscan->flags = (U8)ARG(nxt);
3386 if (RExC_open_parens) {
3387 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3388 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
40d049e4 3389 }
8aa23a47
YO
3390 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3391 OP(nxt) = OPTIMIZED; /* was CLOSE. */
40d049e4 3392
c277df42 3393#ifdef DEBUGGING
8aa23a47
YO
3394 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3395 OP(nxt + 1) = OPTIMIZED; /* was count. */
3396 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3397 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
b81d288d 3398#endif
c277df42 3399#if 0
8aa23a47
YO
3400 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3401 regnode *nnxt = regnext(nxt1);
8aa23a47
YO
3402 if (nnxt == nxt) {
3403 if (reg_off_by_arg[OP(nxt1)])
3404 ARG_SET(nxt1, nxt2 - nxt1);
3405 else if (nxt2 - nxt1 < U16_MAX)
3406 NEXT_OFF(nxt1) = nxt2 - nxt1;
3407 else
3408 OP(nxt) = NOTHING; /* Cannot beautify */
c277df42 3409 }
8aa23a47 3410 nxt1 = nnxt;
c277df42 3411 }
5d1c421c 3412#endif
8aa23a47
YO
3413 /* Optimize again: */
3414 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3415 NULL, stopparen, recursed, NULL, 0,depth+1);
3416 }
3417 else
3418 oscan->flags = 0;
3419 }
3420 else if ((OP(oscan) == CURLYX)
3421 && (flags & SCF_WHILEM_VISITED_POS)
3422 /* See the comment on a similar expression above.
3b753521 3423 However, this time it's not a subexpression
8aa23a47
YO
3424 we care about, but the expression itself. */
3425 && (maxcount == REG_INFTY)
3426 && data && ++data->whilem_c < 16) {
3427 /* This stays as CURLYX, we can put the count/of pair. */
3428 /* Find WHILEM (as in regexec.c) */
3429 regnode *nxt = oscan + NEXT_OFF(oscan);
3430
3431 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3432 nxt += ARG(nxt);
3433 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3434 | (RExC_whilem_seen << 4)); /* On WHILEM */
3435 }
3436 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3437 pars++;
3438 if (flags & SCF_DO_SUBSTR) {
3439 SV *last_str = NULL;
3440 int counted = mincount != 0;
a0ed51b3 3441
8aa23a47
YO
3442 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3443#if defined(SPARC64_GCC_WORKAROUND)
3444 I32 b = 0;
3445 STRLEN l = 0;
3446 const char *s = NULL;
3447 I32 old = 0;
b515a41d 3448
8aa23a47
YO
3449 if (pos_before >= data->last_start_min)
3450 b = pos_before;
3451 else
3452 b = data->last_start_min;
b515a41d 3453
8aa23a47
YO
3454 l = 0;
3455 s = SvPV_const(data->last_found, l);
3456 old = b - data->last_start_min;
3457
3458#else
3459 I32 b = pos_before >= data->last_start_min
3460 ? pos_before : data->last_start_min;
3461 STRLEN l;
3462 const char * const s = SvPV_const(data->last_found, l);
3463 I32 old = b - data->last_start_min;
3464#endif
3465
3466 if (UTF)
3467 old = utf8_hop((U8*)s, old) - (U8*)s;
8aa23a47
YO
3468 l -= old;
3469 /* Get the added string: */
740cce10 3470 last_str = newSVpvn_utf8(s + old, l, UTF);
8aa23a47
YO
3471 if (deltanext == 0 && pos_before == b) {
3472 /* What was added is a constant string */
3473 if (mincount > 1) {
3474 SvGROW(last_str, (mincount * l) + 1);
3475 repeatcpy(SvPVX(last_str) + l,
3476 SvPVX_const(last_str), l, mincount - 1);
3477 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3478 /* Add additional parts. */
3479 SvCUR_set(data->last_found,
3480 SvCUR(data->last_found) - l);
3481 sv_catsv(data->last_found, last_str);
3482 {
3483 SV * sv = data->last_found;
3484 MAGIC *mg =
3485 SvUTF8(sv) && SvMAGICAL(sv) ?
3486 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3487 if (mg && mg->mg_len >= 0)
bd94e887 3488 mg->mg_len += CHR_SVLEN(last_str) - l;
b515a41d 3489 }
8aa23a47 3490 data->last_end += l * (mincount - 1);
b515a41d 3491 }
8aa23a47
YO
3492 } else {
3493 /* start offset must point into the last copy */
3494 data->last_start_min += minnext * (mincount - 1);
3495 data->last_start_max += is_inf ? I32_MAX
3496 : (maxcount - 1) * (minnext + data->pos_delta);
3497 }
c277df42 3498 }
8aa23a47
YO
3499 /* It is counted once already... */
3500 data->pos_min += minnext * (mincount - counted);
3501 data->pos_delta += - counted * deltanext +
3502 (minnext + deltanext) * maxcount - minnext * mincount;
3503 if (mincount != maxcount) {
3504 /* Cannot extend fixed substrings found inside
3505 the group. */
304ee84b 3506 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3507 if (mincount && last_str) {
3508 SV * const sv = data->last_found;
3509 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3510 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3511
3512 if (mg)
3513 mg->mg_len = -1;
3514 sv_setsv(sv, last_str);
3515 data->last_end = data->pos_min;
3516 data->last_start_min =
3517 data->pos_min - CHR_SVLEN(last_str);
3518 data->last_start_max = is_inf
3519 ? I32_MAX
3520 : data->pos_min + data->pos_delta
3521 - CHR_SVLEN(last_str);
3522 }
3523 data->longest = &(data->longest_float);
3524 }
3525 SvREFCNT_dec(last_str);
c277df42 3526 }
8aa23a47
YO
3527 if (data && (fl & SF_HAS_EVAL))
3528 data->flags |= SF_HAS_EVAL;
3529 optimize_curly_tail:
3530 if (OP(oscan) != CURLYX) {
3531 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3532 && NEXT_OFF(next))
3533 NEXT_OFF(oscan) += NEXT_OFF(next);
3534 }
3535 continue;
3536 default: /* REF and CLUMP only? */
3537 if (flags & SCF_DO_SUBSTR) {
304ee84b 3538 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
8aa23a47
YO
3539 data->longest = &(data->longest_float);
3540 }
3541 is_inf = is_inf_internal = 1;
3542 if (flags & SCF_DO_STCLASS_OR)
3543 cl_anything(pRExC_state, data->start_class);
3544 flags &= ~SCF_DO_STCLASS;
3545 break;
c277df42 3546 }
8aa23a47 3547 }
e1d1eefb
YO
3548 else if (OP(scan) == LNBREAK) {
3549 if (flags & SCF_DO_STCLASS) {
3550 int value = 0;
3551 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3552 if (flags & SCF_DO_STCLASS_AND) {
3553 for (value = 0; value < 256; value++)
e64b1bd1 3554 if (!is_VERTWS_cp(value))
b9a59e08
KW
3555 ANYOF_BITMAP_CLEAR(data->start_class, value);
3556 }
3557 else {
e1d1eefb 3558 for (value = 0; value < 256; value++)
e64b1bd1 3559 if (is_VERTWS_cp(value))
b9a59e08
KW
3560 ANYOF_BITMAP_SET(data->start_class, value);
3561 }
e1d1eefb
YO
3562 if (flags & SCF_DO_STCLASS_OR)
3563 cl_and(data->start_class, and_withp);
3564 flags &= ~SCF_DO_STCLASS;
3565 }
3566 min += 1;
f9a79580 3567 delta += 1;
e1d1eefb
YO
3568 if (flags & SCF_DO_SUBSTR) {
3569 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3570 data->pos_min += 1;
f9a79580 3571 data->pos_delta += 1;
e1d1eefb
YO
3572 data->longest = &(data->longest_float);
3573 }
e1d1eefb 3574 }
f9a79580 3575 else if (OP(scan) == FOLDCHAR) {
ced7f090 3576 int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2;
f9a79580
RGS
3577 flags &= ~SCF_DO_STCLASS;
3578 min += 1;
3579 delta += d;
3580 if (flags & SCF_DO_SUBSTR) {
3581 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3582 data->pos_min += 1;
3583 data->pos_delta += d;
3584 data->longest = &(data->longest_float);
3585 }
3586 }
e52fc539 3587 else if (REGNODE_SIMPLE(OP(scan))) {
8aa23a47 3588 int value = 0;
653099ff 3589
8aa23a47 3590 if (flags & SCF_DO_SUBSTR) {
304ee84b 3591 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3592 data->pos_min++;
3593 }
3594 min++;
3595 if (flags & SCF_DO_STCLASS) {
3596 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
b515a41d 3597
8aa23a47
YO
3598 /* Some of the logic below assumes that switching
3599 locale on will only add false positives. */
3600 switch (PL_regkind[OP(scan)]) {
3601 case SANY:
3602 default:
3603 do_default:
3604 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3605 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3606 cl_anything(pRExC_state, data->start_class);
3607 break;
3608 case REG_ANY:
3609 if (OP(scan) == SANY)
3610 goto do_default;
3611 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3612 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3a15e693 3613 || ANYOF_CLASS_TEST_ANY_SET(data->start_class));
8aa23a47 3614 cl_anything(pRExC_state, data->start_class);
653099ff 3615 }
8aa23a47
YO
3616 if (flags & SCF_DO_STCLASS_AND || !value)
3617 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3618 break;
3619 case ANYOF:
3620 if (flags & SCF_DO_STCLASS_AND)
3621 cl_and(data->start_class,
3622 (struct regnode_charclass_class*)scan);
653099ff 3623 else
8aa23a47
YO
3624 cl_or(pRExC_state, data->start_class,
3625 (struct regnode_charclass_class*)scan);
3626 break;
3627 case ALNUM:
3628 if (flags & SCF_DO_STCLASS_AND) {
3629 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3630 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
a12cf05f
KW
3631 if (FLAGS(scan) & USE_UNI) {
3632 for (value = 0; value < 256; value++) {
3633 if (!isWORDCHAR_L1(value)) {
3634 ANYOF_BITMAP_CLEAR(data->start_class, value);
3635 }
3636 }
3637 } else {
3638 for (value = 0; value < 256; value++) {
3639 if (!isALNUM(value)) {
3640 ANYOF_BITMAP_CLEAR(data->start_class, value);
3641 }
3642 }
3643 }
8aa23a47 3644 }
653099ff 3645 }
8aa23a47
YO
3646 else {
3647 if (data->start_class->flags & ANYOF_LOCALE)
3648 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
a12cf05f
KW
3649 else if (FLAGS(scan) & USE_UNI) {
3650 for (value = 0; value < 256; value++) {
3651 if (isWORDCHAR_L1(value)) {
3652 ANYOF_BITMAP_SET(data->start_class, value);
3653 }
3654 }
3655 } else {
3656 for (value = 0; value < 256; value++) {
3657 if (isALNUM(value)) {
3658 ANYOF_BITMAP_SET(data->start_class, value);
3659 }
3660 }
3661 }
8aa23a47
YO
3662 }
3663 break;
3664 case ALNUML:
3665 if (flags & SCF_DO_STCLASS_AND) {
3666 if (data->start_class->flags & ANYOF_LOCALE)
3667 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3668 }
3669 else {
3670 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3671 data->start_class->flags |= ANYOF_LOCALE;
3672 }
3673 break;
3674 case NALNUM:
3675 if (flags & SCF_DO_STCLASS_AND) {
3676 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3677 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
a12cf05f
KW
3678 if (FLAGS(scan) & USE_UNI) {
3679 for (value = 0; value < 256; value++) {
3680 if (isWORDCHAR_L1(value)) {
3681 ANYOF_BITMAP_CLEAR(data->start_class, value);
3682 }
3683 }
3684 } else {
3685 for (value = 0; value < 256; value++) {
3686 if (isALNUM(value)) {
3687 ANYOF_BITMAP_CLEAR(data->start_class, value);
3688 }
3689 }
3690 }
653099ff
GS
3691 }
3692 }
8aa23a47
YO
3693 else {
3694 if (data->start_class->flags & ANYOF_LOCALE)
3695 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3696 else {
3697 for (value = 0; value < 256; value++)
3698 if (!isALNUM(value))
b9a59e08 3699 ANYOF_BITMAP_SET(data->start_class, value);
8aa23a47 3700 }
653099ff 3701 }
8aa23a47
YO
3702 break;
3703 case NALNUML:
3704 if (flags & SCF_DO_STCLASS_AND) {
3705 if (data->start_class->flags & ANYOF_LOCALE)
3706 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
653099ff 3707 }
8aa23a47
YO
3708 else {
3709 data->start_class->flags |= ANYOF_LOCALE;
3710 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3711 }
3712 break;
3713 case SPACE:
3714 if (flags & SCF_DO_STCLASS_AND) {
3715 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3716 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);