This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: small efficiency gain
[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);
653099ff
GS
766 if (!(and_with->flags & ANYOF_CLASS)
767 && !(cl->flags & ANYOF_CLASS)
768 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
769 && !(and_with->flags & ANYOF_FOLD)
770 && !(cl->flags & ANYOF_FOLD)) {
771 int i;
772
773 if (and_with->flags & ANYOF_INVERT)
774 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
775 cl->bitmap[i] &= ~and_with->bitmap[i];
776 else
777 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
778 cl->bitmap[i] &= and_with->bitmap[i];
779 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
780 if (!(and_with->flags & ANYOF_EOS))
781 cl->flags &= ~ANYOF_EOS;
1aa99e6b 782
8951c461
KW
783 if (!(and_with->flags & ANYOF_FOLD))
784 cl->flags &= ~ANYOF_FOLD;
785
3ff7ceb3 786 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_NONBITMAP &&
14ebb1a2 787 !(and_with->flags & ANYOF_INVERT)) {
1aa99e6b 788 cl->flags &= ~ANYOF_UNICODE_ALL;
ef87b810
KW
789 cl->flags |= and_with->flags & ANYOF_NONBITMAP; /* field is 2 bits; use
790 only the one(s)
791 actually set */
1aa99e6b
IH
792 ARG_SET(cl, ARG(and_with));
793 }
14ebb1a2
JH
794 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
795 !(and_with->flags & ANYOF_INVERT))
1aa99e6b 796 cl->flags &= ~ANYOF_UNICODE_ALL;
3ff7ceb3 797 if (!(and_with->flags & (ANYOF_NONBITMAP|ANYOF_UNICODE_ALL)) &&
14ebb1a2 798 !(and_with->flags & ANYOF_INVERT))
3ff7ceb3 799 cl->flags &= ~ANYOF_NONBITMAP;
653099ff
GS
800}
801
802/* 'OR' a given class with another one. Can create false positives */
803/* We assume that cl is not inverted */
804STATIC void
097eb12c 805S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
653099ff 806{
7918f24d
NC
807 PERL_ARGS_ASSERT_CL_OR;
808
653099ff
GS
809 if (or_with->flags & ANYOF_INVERT) {
810 /* We do not use
811 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
812 * <= (B1 | !B2) | (CL1 | !CL2)
813 * which is wasteful if CL2 is small, but we ignore CL2:
814 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
815 * XXXX Can we handle case-fold? Unclear:
816 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
817 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
818 */
819 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
820 && !(or_with->flags & ANYOF_FOLD)
821 && !(cl->flags & ANYOF_FOLD) ) {
822 int i;
823
824 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
825 cl->bitmap[i] |= ~or_with->bitmap[i];
826 } /* XXXX: logic is complicated otherwise */
827 else {
830247a4 828 cl_anything(pRExC_state, cl);
653099ff
GS
829 }
830 } else {
831 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
832 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
b81d288d 833 && (!(or_with->flags & ANYOF_FOLD)
653099ff
GS
834 || (cl->flags & ANYOF_FOLD)) ) {
835 int i;
836
837 /* OR char bitmap and class bitmap separately */
838 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
839 cl->bitmap[i] |= or_with->bitmap[i];
840 if (or_with->flags & ANYOF_CLASS) {
841 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
842 cl->classflags[i] |= or_with->classflags[i];
843 cl->flags |= ANYOF_CLASS;
844 }
845 }
846 else { /* XXXX: logic is complicated, leave it along for a moment. */
830247a4 847 cl_anything(pRExC_state, cl);
653099ff
GS
848 }
849 }
850 if (or_with->flags & ANYOF_EOS)
851 cl->flags |= ANYOF_EOS;
1aa99e6b 852
8951c461
KW
853 if (or_with->flags & ANYOF_FOLD)
854 cl->flags |= ANYOF_FOLD;
855
9826f543
KW
856 /* If both nodes match something outside the bitmap, but what they match
857 * outside is not the same pointer, and hence not easily compared, give up
858 * and allow the start class to match everything outside the bitmap */
3ff7ceb3 859 if (cl->flags & ANYOF_NONBITMAP && or_with->flags & ANYOF_NONBITMAP &&
1aa99e6b
IH
860 ARG(cl) != ARG(or_with)) {
861 cl->flags |= ANYOF_UNICODE_ALL;
1aa99e6b 862 }
9826f543 863
1aa99e6b
IH
864 if (or_with->flags & ANYOF_UNICODE_ALL) {
865 cl->flags |= ANYOF_UNICODE_ALL;
1aa99e6b 866 }
653099ff
GS
867}
868
a3621e74
YO
869#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
870#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
871#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
872#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
873
3dab1dad
YO
874
875#ifdef DEBUGGING
07be1b83 876/*
2b8b4781
NC
877 dump_trie(trie,widecharmap,revcharmap)
878 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
879 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
880
881 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
882 The _interim_ variants are used for debugging the interim
883 tables that are used to generate the final compressed
884 representation which is what dump_trie expects.
885
3dab1dad
YO
886 Part of the reason for their existance is to provide a form
887 of documentation as to how the different representations function.
07be1b83
YO
888
889*/
3dab1dad
YO
890
891/*
3dab1dad
YO
892 Dumps the final compressed table form of the trie to Perl_debug_log.
893 Used for debugging make_trie().
894*/
b9a59e08 895
3dab1dad 896STATIC void
2b8b4781
NC
897S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
898 AV *revcharmap, U32 depth)
3dab1dad
YO
899{
900 U32 state;
ab3bbdeb 901 SV *sv=sv_newmortal();
55eed653 902 int colwidth= widecharmap ? 6 : 4;
2e64971a 903 U16 word;
3dab1dad
YO
904 GET_RE_DEBUG_FLAGS_DECL;
905
7918f24d 906 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 907
3dab1dad
YO
908 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
909 (int)depth * 2 + 2,"",
910 "Match","Base","Ofs" );
911
912 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 913 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 914 if ( tmp ) {
ab3bbdeb
YO
915 PerlIO_printf( Perl_debug_log, "%*s",
916 colwidth,
ddc5bc0f 917 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
918 PL_colors[0], PL_colors[1],
919 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
920 PERL_PV_ESCAPE_FIRSTCHAR
921 )
922 );
3dab1dad
YO
923 }
924 }
925 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
926 (int)depth * 2 + 2,"");
927
928 for( state = 0 ; state < trie->uniquecharcount ; state++ )
ab3bbdeb 929 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
3dab1dad
YO
930 PerlIO_printf( Perl_debug_log, "\n");
931
1e2e3d02 932 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 933 const U32 base = trie->states[ state ].trans.base;
3dab1dad
YO
934
935 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
936
937 if ( trie->states[ state ].wordnum ) {
938 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
939 } else {
940 PerlIO_printf( Perl_debug_log, "%6s", "" );
941 }
942
943 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
944
945 if ( base ) {
946 U32 ofs = 0;
947
948 while( ( base + ofs < trie->uniquecharcount ) ||
949 ( base + ofs - trie->uniquecharcount < trie->lasttrans
950 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
951 ofs++;
952
953 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
954
955 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
956 if ( ( base + ofs >= trie->uniquecharcount ) &&
957 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
958 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
959 {
ab3bbdeb
YO
960 PerlIO_printf( Perl_debug_log, "%*"UVXf,
961 colwidth,
3dab1dad
YO
962 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
963 } else {
ab3bbdeb 964 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
3dab1dad
YO
965 }
966 }
967
968 PerlIO_printf( Perl_debug_log, "]");
969
970 }
971 PerlIO_printf( Perl_debug_log, "\n" );
972 }
2e64971a
DM
973 PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, "");
974 for (word=1; word <= trie->wordcount; word++) {
975 PerlIO_printf(Perl_debug_log, " %d:(%d,%d)",
976 (int)word, (int)(trie->wordinfo[word].prev),
977 (int)(trie->wordinfo[word].len));
978 }
979 PerlIO_printf(Perl_debug_log, "\n" );
3dab1dad
YO
980}
981/*
3dab1dad
YO
982 Dumps a fully constructed but uncompressed trie in list form.
983 List tries normally only are used for construction when the number of
984 possible chars (trie->uniquecharcount) is very high.
985 Used for debugging make_trie().
986*/
987STATIC void
55eed653 988S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
989 HV *widecharmap, AV *revcharmap, U32 next_alloc,
990 U32 depth)
3dab1dad
YO
991{
992 U32 state;
ab3bbdeb 993 SV *sv=sv_newmortal();
55eed653 994 int colwidth= widecharmap ? 6 : 4;
3dab1dad 995 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
996
997 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
998
3dab1dad 999 /* print out the table precompression. */
ab3bbdeb
YO
1000 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
1001 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
1002 "------:-----+-----------------\n" );
3dab1dad
YO
1003
1004 for( state=1 ; state < next_alloc ; state ++ ) {
1005 U16 charid;
1006
ab3bbdeb 1007 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
3dab1dad
YO
1008 (int)depth * 2 + 2,"", (UV)state );
1009 if ( ! trie->states[ state ].wordnum ) {
1010 PerlIO_printf( Perl_debug_log, "%5s| ","");
1011 } else {
1012 PerlIO_printf( Perl_debug_log, "W%4x| ",
1013 trie->states[ state ].wordnum
1014 );
1015 }
1016 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
2b8b4781 1017 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
ab3bbdeb
YO
1018 if ( tmp ) {
1019 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
1020 colwidth,
ddc5bc0f 1021 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
1022 PL_colors[0], PL_colors[1],
1023 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1024 PERL_PV_ESCAPE_FIRSTCHAR
1025 ) ,
1e2e3d02
YO
1026 TRIE_LIST_ITEM(state,charid).forid,
1027 (UV)TRIE_LIST_ITEM(state,charid).newstate
1028 );
1029 if (!(charid % 10))
664e119d
RGS
1030 PerlIO_printf(Perl_debug_log, "\n%*s| ",
1031 (int)((depth * 2) + 14), "");
1e2e3d02 1032 }
ab3bbdeb
YO
1033 }
1034 PerlIO_printf( Perl_debug_log, "\n");
3dab1dad
YO
1035 }
1036}
1037
1038/*
3dab1dad
YO
1039 Dumps a fully constructed but uncompressed trie in table form.
1040 This is the normal DFA style state transition table, with a few
1041 twists to facilitate compression later.
1042 Used for debugging make_trie().
1043*/
1044STATIC void
55eed653 1045S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
1046 HV *widecharmap, AV *revcharmap, U32 next_alloc,
1047 U32 depth)
3dab1dad
YO
1048{
1049 U32 state;
1050 U16 charid;
ab3bbdeb 1051 SV *sv=sv_newmortal();
55eed653 1052 int colwidth= widecharmap ? 6 : 4;
3dab1dad 1053 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
1054
1055 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
3dab1dad
YO
1056
1057 /*
1058 print out the table precompression so that we can do a visual check
1059 that they are identical.
1060 */
1061
1062 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1063
1064 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 1065 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 1066 if ( tmp ) {
ab3bbdeb
YO
1067 PerlIO_printf( Perl_debug_log, "%*s",
1068 colwidth,
ddc5bc0f 1069 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
1070 PL_colors[0], PL_colors[1],
1071 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1072 PERL_PV_ESCAPE_FIRSTCHAR
1073 )
1074 );
3dab1dad
YO
1075 }
1076 }
1077
1078 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1079
1080 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb 1081 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
3dab1dad
YO
1082 }
1083
1084 PerlIO_printf( Perl_debug_log, "\n" );
1085
1086 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1087
1088 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1089 (int)depth * 2 + 2,"",
1090 (UV)TRIE_NODENUM( state ) );
1091
1092 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
1093 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1094 if (v)
1095 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1096 else
1097 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
3dab1dad
YO
1098 }
1099 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1100 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1101 } else {
1102 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1103 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1104 }
1105 }
07be1b83 1106}
3dab1dad
YO
1107
1108#endif
1109
2e64971a 1110
786e8c11
YO
1111/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1112 startbranch: the first branch in the whole branch sequence
1113 first : start branch of sequence of branch-exact nodes.
1114 May be the same as startbranch
1115 last : Thing following the last branch.
1116 May be the same as tail.
1117 tail : item following the branch sequence
1118 count : words in the sequence
1119 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1120 depth : indent depth
3dab1dad 1121
786e8c11 1122Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 1123
786e8c11
YO
1124A trie is an N'ary tree where the branches are determined by digital
1125decomposition of the key. IE, at the root node you look up the 1st character and
1126follow that branch repeat until you find the end of the branches. Nodes can be
1127marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 1128
786e8c11 1129 /he|she|his|hers/
72f13be8 1130
786e8c11
YO
1131would convert into the following structure. Numbers represent states, letters
1132following numbers represent valid transitions on the letter from that state, if
1133the number is in square brackets it represents an accepting state, otherwise it
1134will be in parenthesis.
07be1b83 1135
786e8c11
YO
1136 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1137 | |
1138 | (2)
1139 | |
1140 (1) +-i->(6)-+-s->[7]
1141 |
1142 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 1143
786e8c11
YO
1144 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1145
1146This shows that when matching against the string 'hers' we will begin at state 1
1147read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1148then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1149is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1150single traverse. We store a mapping from accepting to state to which word was
1151matched, and then when we have multiple possibilities we try to complete the
1152rest of the regex in the order in which they occured in the alternation.
1153
1154The only prior NFA like behaviour that would be changed by the TRIE support is
1155the silent ignoring of duplicate alternations which are of the form:
1156
1157 / (DUPE|DUPE) X? (?{ ... }) Y /x
1158
4b714af6 1159Thus EVAL blocks following a trie may be called a different number of times with
786e8c11
YO
1160and without the optimisation. With the optimisations dupes will be silently
1161ignored. This inconsistant behaviour of EVAL type nodes is well established as
1162the following demonstrates:
1163
1164 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1165
1166which prints out 'word' three times, but
1167
1168 'words'=~/(word|word|word)(?{ print $1 })S/
1169
1170which doesnt print it out at all. This is due to other optimisations kicking in.
1171
1172Example of what happens on a structural level:
1173
1174The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1175
1176 1: CURLYM[1] {1,32767}(18)
1177 5: BRANCH(8)
1178 6: EXACT <ac>(16)
1179 8: BRANCH(11)
1180 9: EXACT <ad>(16)
1181 11: BRANCH(14)
1182 12: EXACT <ab>(16)
1183 16: SUCCEED(0)
1184 17: NOTHING(18)
1185 18: END(0)
1186
1187This would be optimizable with startbranch=5, first=5, last=16, tail=16
1188and should turn into:
1189
1190 1: CURLYM[1] {1,32767}(18)
1191 5: TRIE(16)
1192 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1193 <ac>
1194 <ad>
1195 <ab>
1196 16: SUCCEED(0)
1197 17: NOTHING(18)
1198 18: END(0)
1199
1200Cases where tail != last would be like /(?foo|bar)baz/:
1201
1202 1: BRANCH(4)
1203 2: EXACT <foo>(8)
1204 4: BRANCH(7)
1205 5: EXACT <bar>(8)
1206 7: TAIL(8)
1207 8: EXACT <baz>(10)
1208 10: END(0)
1209
1210which would be optimizable with startbranch=1, first=1, last=7, tail=8
1211and would end up looking like:
1212
1213 1: TRIE(8)
1214 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1215 <foo>
1216 <bar>
1217 7: TAIL(8)
1218 8: EXACT <baz>(10)
1219 10: END(0)
1220
1221 d = uvuni_to_utf8_flags(d, uv, 0);
1222
1223is the recommended Unicode-aware way of saying
1224
1225 *(d++) = uv;
1226*/
1227
1e2e3d02 1228#define TRIE_STORE_REVCHAR \
786e8c11 1229 STMT_START { \
73031816
NC
1230 if (UTF) { \
1231 SV *zlopp = newSV(2); \
88c9ea1e
CB
1232 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1233 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
73031816
NC
1234 SvCUR_set(zlopp, kapow - flrbbbbb); \
1235 SvPOK_on(zlopp); \
1236 SvUTF8_on(zlopp); \
1237 av_push(revcharmap, zlopp); \
1238 } else { \
6bdeddd2 1239 char ooooff = (char)uvc; \
73031816
NC
1240 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1241 } \
1242 } STMT_END
786e8c11
YO
1243
1244#define TRIE_READ_CHAR STMT_START { \
1245 wordlen++; \
1246 if ( UTF ) { \
1247 if ( folder ) { \
1248 if ( foldlen > 0 ) { \
1249 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1250 foldlen -= len; \
1251 scan += len; \
1252 len = 0; \
1253 } else { \
1254 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1255 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1256 foldlen -= UNISKIP( uvc ); \
1257 scan = foldbuf + UNISKIP( uvc ); \
1258 } \
1259 } else { \
1260 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1261 } \
1262 } else { \
1263 uvc = (U32)*uc; \
1264 len = 1; \
1265 } \
1266} STMT_END
1267
1268
1269
1270#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1271 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
f9003953
NC
1272 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1273 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
786e8c11
YO
1274 } \
1275 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1276 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1277 TRIE_LIST_CUR( state )++; \
1278} STMT_END
07be1b83 1279
786e8c11
YO
1280#define TRIE_LIST_NEW(state) STMT_START { \
1281 Newxz( trie->states[ state ].trans.list, \
1282 4, reg_trie_trans_le ); \
1283 TRIE_LIST_CUR( state ) = 1; \
1284 TRIE_LIST_LEN( state ) = 4; \
1285} STMT_END
07be1b83 1286
786e8c11
YO
1287#define TRIE_HANDLE_WORD(state) STMT_START { \
1288 U16 dupe= trie->states[ state ].wordnum; \
1289 regnode * const noper_next = regnext( noper ); \
1290 \
786e8c11
YO
1291 DEBUG_r({ \
1292 /* store the word for dumping */ \
1293 SV* tmp; \
1294 if (OP(noper) != NOTHING) \
740cce10 1295 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 1296 else \
740cce10 1297 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 1298 av_push( trie_words, tmp ); \
786e8c11
YO
1299 }); \
1300 \
1301 curword++; \
2e64971a
DM
1302 trie->wordinfo[curword].prev = 0; \
1303 trie->wordinfo[curword].len = wordlen; \
1304 trie->wordinfo[curword].accept = state; \
786e8c11
YO
1305 \
1306 if ( noper_next < tail ) { \
1307 if (!trie->jump) \
c944940b 1308 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
7f69552c 1309 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
1310 if (!jumper) \
1311 jumper = noper_next; \
1312 if (!nextbranch) \
1313 nextbranch= regnext(cur); \
1314 } \
1315 \
1316 if ( dupe ) { \
2e64971a
DM
1317 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
1318 /* chain, so that when the bits of chain are later */\
1319 /* linked together, the dups appear in the chain */\
1320 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
1321 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
1322 } else { \
1323 /* we haven't inserted this word yet. */ \
1324 trie->states[ state ].wordnum = curword; \
1325 } \
1326} STMT_END
07be1b83 1327
3dab1dad 1328
786e8c11
YO
1329#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1330 ( ( base + charid >= ucharcount \
1331 && base + charid < ubound \
1332 && state == trie->trans[ base - ucharcount + charid ].check \
1333 && trie->trans[ base - ucharcount + charid ].next ) \
1334 ? trie->trans[ base - ucharcount + charid ].next \
1335 : ( state==1 ? special : 0 ) \
1336 )
3dab1dad 1337
786e8c11
YO
1338#define MADE_TRIE 1
1339#define MADE_JUMP_TRIE 2
1340#define MADE_EXACT_TRIE 4
3dab1dad 1341
a3621e74 1342STATIC I32
786e8c11 1343S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
a3621e74 1344{
27da23d5 1345 dVAR;
a3621e74
YO
1346 /* first pass, loop through and scan words */
1347 reg_trie_data *trie;
55eed653 1348 HV *widecharmap = NULL;
2b8b4781 1349 AV *revcharmap = newAV();
a3621e74 1350 regnode *cur;
9f7f3913 1351 const U32 uniflags = UTF8_ALLOW_DEFAULT;
a3621e74
YO
1352 STRLEN len = 0;
1353 UV uvc = 0;
1354 U16 curword = 0;
1355 U32 next_alloc = 0;
786e8c11
YO
1356 regnode *jumper = NULL;
1357 regnode *nextbranch = NULL;
7f69552c 1358 regnode *convert = NULL;
2e64971a 1359 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 1360 /* we just use folder as a flag in utf8 */
1e696034 1361 const U8 * folder = NULL;
a3621e74 1362
2b8b4781
NC
1363#ifdef DEBUGGING
1364 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1365 AV *trie_words = NULL;
1366 /* along with revcharmap, this only used during construction but both are
1367 * useful during debugging so we store them in the struct when debugging.
8e11feef 1368 */
2b8b4781
NC
1369#else
1370 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
3dab1dad 1371 STRLEN trie_charcount=0;
3dab1dad 1372#endif
2b8b4781 1373 SV *re_trie_maxbuff;
a3621e74 1374 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
1375
1376 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
1377#ifndef DEBUGGING
1378 PERL_UNUSED_ARG(depth);
1379#endif
a3621e74 1380
1e696034
KW
1381 switch (flags) {
1382 case EXACTFU: folder = PL_fold_latin1; break;
1383 case EXACTF: folder = PL_fold; break;
1384 case EXACTFL: folder = PL_fold_locale; break;
1385 }
1386
c944940b 1387 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 1388 trie->refcount = 1;
3dab1dad 1389 trie->startstate = 1;
786e8c11 1390 trie->wordcount = word_count;
f8fc2ecf 1391 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 1392 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
3dab1dad 1393 if (!(UTF && folder))
c944940b 1394 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
1395 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
1396 trie->wordcount+1, sizeof(reg_trie_wordinfo));
1397
a3621e74 1398 DEBUG_r({
2b8b4781 1399 trie_words = newAV();
a3621e74 1400 });
a3621e74 1401
0111c4fd 1402 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
a3621e74 1403 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 1404 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 1405 }
3dab1dad
YO
1406 DEBUG_OPTIMISE_r({
1407 PerlIO_printf( Perl_debug_log,
786e8c11 1408 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
3dab1dad
YO
1409 (int)depth * 2 + 2, "",
1410 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
786e8c11 1411 REG_NODE_NUM(last), REG_NODE_NUM(tail),
85c3142d 1412 (int)depth);
3dab1dad 1413 });
7f69552c
YO
1414
1415 /* Find the node we are going to overwrite */
1416 if ( first == startbranch && OP( last ) != BRANCH ) {
1417 /* whole branch chain */
1418 convert = first;
1419 } else {
1420 /* branch sub-chain */
1421 convert = NEXTOPER( first );
1422 }
1423
a3621e74
YO
1424 /* -- First loop and Setup --
1425
1426 We first traverse the branches and scan each word to determine if it
1427 contains widechars, and how many unique chars there are, this is
1428 important as we have to build a table with at least as many columns as we
1429 have unique chars.
1430
1431 We use an array of integers to represent the character codes 0..255
38a44b82 1432 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
a3621e74
YO
1433 native representation of the character value as the key and IV's for the
1434 coded index.
1435
1436 *TODO* If we keep track of how many times each character is used we can
1437 remap the columns so that the table compression later on is more
3b753521 1438 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
1439 middle and the least common are on the outside. IMO this would be better
1440 than a most to least common mapping as theres a decent chance the most
1441 common letter will share a node with the least common, meaning the node
1442 will not be compressable. With a middle is most common approach the worst
1443 case is when we have the least common nodes twice.
1444
1445 */
1446
a3621e74 1447 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
c445ea15 1448 regnode * const noper = NEXTOPER( cur );
e1ec3a88 1449 const U8 *uc = (U8*)STRING( noper );
a28509cc 1450 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1451 STRLEN foldlen = 0;
1452 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2af232bd 1453 const U8 *scan = (U8*)NULL;
07be1b83 1454 U32 wordlen = 0; /* required init */
02daf0ab
YO
1455 STRLEN chars = 0;
1456 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
a3621e74 1457
3dab1dad
YO
1458 if (OP(noper) == NOTHING) {
1459 trie->minlen= 0;
1460 continue;
1461 }
02daf0ab
YO
1462 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1463 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1464 regardless of encoding */
1465
a3621e74 1466 for ( ; uc < e ; uc += len ) {
3dab1dad 1467 TRIE_CHARCOUNT(trie)++;
a3621e74 1468 TRIE_READ_CHAR;
3dab1dad 1469 chars++;
a3621e74
YO
1470 if ( uvc < 256 ) {
1471 if ( !trie->charmap[ uvc ] ) {
1472 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1473 if ( folder )
1474 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
3dab1dad 1475 TRIE_STORE_REVCHAR;
a3621e74 1476 }
02daf0ab 1477 if ( set_bit ) {
62012aee
KW
1478 /* store the codepoint in the bitmap, and its folded
1479 * equivalent. */
02daf0ab 1480 TRIE_BITMAP_SET(trie,uvc);
0921ee73
T
1481
1482 /* store the folded codepoint */
1483 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1484
1485 if ( !UTF ) {
1486 /* store first byte of utf8 representation of
1487 codepoints in the 127 < uvc < 256 range */
1488 if (127 < uvc && uvc < 192) {
1489 TRIE_BITMAP_SET(trie,194);
1490 } else if (191 < uvc ) {
1491 TRIE_BITMAP_SET(trie,195);
1492 /* && uvc < 256 -- we know uvc is < 256 already */
1493 }
1494 }
02daf0ab
YO
1495 set_bit = 0; /* We've done our bit :-) */
1496 }
a3621e74
YO
1497 } else {
1498 SV** svpp;
55eed653
NC
1499 if ( !widecharmap )
1500 widecharmap = newHV();
a3621e74 1501
55eed653 1502 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
1503
1504 if ( !svpp )
e4584336 1505 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
a3621e74
YO
1506
1507 if ( !SvTRUE( *svpp ) ) {
1508 sv_setiv( *svpp, ++trie->uniquecharcount );
3dab1dad 1509 TRIE_STORE_REVCHAR;
a3621e74
YO
1510 }
1511 }
1512 }
3dab1dad
YO
1513 if( cur == first ) {
1514 trie->minlen=chars;
1515 trie->maxlen=chars;
1516 } else if (chars < trie->minlen) {
1517 trie->minlen=chars;
1518 } else if (chars > trie->maxlen) {
1519 trie->maxlen=chars;
1520 }
1521
a3621e74
YO
1522 } /* end first pass */
1523 DEBUG_TRIE_COMPILE_r(
3dab1dad
YO
1524 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1525 (int)depth * 2 + 2,"",
55eed653 1526 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
1527 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1528 (int)trie->minlen, (int)trie->maxlen )
a3621e74 1529 );
a3621e74
YO
1530
1531 /*
1532 We now know what we are dealing with in terms of unique chars and
1533 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
1534 representation using a flat table will take. If it's over a reasonable
1535 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
1536 conservative but potentially much slower representation using an array
1537 of lists.
1538
1539 At the end we convert both representations into the same compressed
1540 form that will be used in regexec.c for matching with. The latter
1541 is a form that cannot be used to construct with but has memory
1542 properties similar to the list form and access properties similar
1543 to the table form making it both suitable for fast searches and
1544 small enough that its feasable to store for the duration of a program.
1545
1546 See the comment in the code where the compressed table is produced
1547 inplace from the flat tabe representation for an explanation of how
1548 the compression works.
1549
1550 */
1551
1552
2e64971a
DM
1553 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
1554 prev_states[1] = 0;
1555
3dab1dad 1556 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
a3621e74
YO
1557 /*
1558 Second Pass -- Array Of Lists Representation
1559
1560 Each state will be represented by a list of charid:state records
1561 (reg_trie_trans_le) the first such element holds the CUR and LEN
1562 points of the allocated array. (See defines above).
1563
1564 We build the initial structure using the lists, and then convert
1565 it into the compressed table form which allows faster lookups
1566 (but cant be modified once converted).
a3621e74
YO
1567 */
1568
a3621e74
YO
1569 STRLEN transcount = 1;
1570
1e2e3d02
YO
1571 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1572 "%*sCompiling trie using list compiler\n",
1573 (int)depth * 2 + 2, ""));
446bd890 1574
c944940b
JH
1575 trie->states = (reg_trie_state *)
1576 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1577 sizeof(reg_trie_state) );
a3621e74
YO
1578 TRIE_LIST_NEW(1);
1579 next_alloc = 2;
1580
1581 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1582
c445ea15
AL
1583 regnode * const noper = NEXTOPER( cur );
1584 U8 *uc = (U8*)STRING( noper );
1585 const U8 * const e = uc + STR_LEN( noper );
1586 U32 state = 1; /* required init */
1587 U16 charid = 0; /* sanity init */
1588 U8 *scan = (U8*)NULL; /* sanity init */
1589 STRLEN foldlen = 0; /* required init */
07be1b83 1590 U32 wordlen = 0; /* required init */
c445ea15
AL
1591 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1592
3dab1dad 1593 if (OP(noper) != NOTHING) {
786e8c11 1594 for ( ; uc < e ; uc += len ) {
c445ea15 1595
786e8c11 1596 TRIE_READ_CHAR;
c445ea15 1597
786e8c11
YO
1598 if ( uvc < 256 ) {
1599 charid = trie->charmap[ uvc ];
c445ea15 1600 } else {
55eed653 1601 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11
YO
1602 if ( !svpp ) {
1603 charid = 0;
1604 } else {
1605 charid=(U16)SvIV( *svpp );
1606 }
c445ea15 1607 }
786e8c11
YO
1608 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1609 if ( charid ) {
a3621e74 1610
786e8c11
YO
1611 U16 check;
1612 U32 newstate = 0;
a3621e74 1613
786e8c11
YO
1614 charid--;
1615 if ( !trie->states[ state ].trans.list ) {
1616 TRIE_LIST_NEW( state );
c445ea15 1617 }
786e8c11
YO
1618 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1619 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1620 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1621 break;
1622 }
1623 }
1624 if ( ! newstate ) {
1625 newstate = next_alloc++;
2e64971a 1626 prev_states[newstate] = state;
786e8c11
YO
1627 TRIE_LIST_PUSH( state, charid, newstate );
1628 transcount++;
1629 }
1630 state = newstate;
1631 } else {
1632 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
c445ea15 1633 }
a28509cc 1634 }
c445ea15 1635 }
3dab1dad 1636 TRIE_HANDLE_WORD(state);
a3621e74
YO
1637
1638 } /* end second pass */
1639
1e2e3d02
YO
1640 /* next alloc is the NEXT state to be allocated */
1641 trie->statecount = next_alloc;
c944940b
JH
1642 trie->states = (reg_trie_state *)
1643 PerlMemShared_realloc( trie->states,
1644 next_alloc
1645 * sizeof(reg_trie_state) );
a3621e74 1646
3dab1dad 1647 /* and now dump it out before we compress it */
2b8b4781
NC
1648 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1649 revcharmap, next_alloc,
1650 depth+1)
1e2e3d02 1651 );
a3621e74 1652
c944940b
JH
1653 trie->trans = (reg_trie_trans *)
1654 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
1655 {
1656 U32 state;
a3621e74
YO
1657 U32 tp = 0;
1658 U32 zp = 0;
1659
1660
1661 for( state=1 ; state < next_alloc ; state ++ ) {
1662 U32 base=0;
1663
1664 /*
1665 DEBUG_TRIE_COMPILE_MORE_r(
1666 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1667 );
1668 */
1669
1670 if (trie->states[state].trans.list) {
1671 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1672 U16 maxid=minid;
a28509cc 1673 U16 idx;
a3621e74
YO
1674
1675 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
1676 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1677 if ( forid < minid ) {
1678 minid=forid;
1679 } else if ( forid > maxid ) {
1680 maxid=forid;
1681 }
a3621e74
YO
1682 }
1683 if ( transcount < tp + maxid - minid + 1) {
1684 transcount *= 2;
c944940b
JH
1685 trie->trans = (reg_trie_trans *)
1686 PerlMemShared_realloc( trie->trans,
446bd890
NC
1687 transcount
1688 * sizeof(reg_trie_trans) );
a3621e74
YO
1689 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1690 }
1691 base = trie->uniquecharcount + tp - minid;
1692 if ( maxid == minid ) {
1693 U32 set = 0;
1694 for ( ; zp < tp ; zp++ ) {
1695 if ( ! trie->trans[ zp ].next ) {
1696 base = trie->uniquecharcount + zp - minid;
1697 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1698 trie->trans[ zp ].check = state;
1699 set = 1;
1700 break;
1701 }
1702 }
1703 if ( !set ) {
1704 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1705 trie->trans[ tp ].check = state;
1706 tp++;
1707 zp = tp;
1708 }
1709 } else {
1710 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15 1711 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
a3621e74
YO
1712 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1713 trie->trans[ tid ].check = state;
1714 }
1715 tp += ( maxid - minid + 1 );
1716 }
1717 Safefree(trie->states[ state ].trans.list);
1718 }
1719 /*
1720 DEBUG_TRIE_COMPILE_MORE_r(
1721 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1722 );
1723 */
1724 trie->states[ state ].trans.base=base;
1725 }
cc601c31 1726 trie->lasttrans = tp + 1;
a3621e74
YO
1727 }
1728 } else {
1729 /*
1730 Second Pass -- Flat Table Representation.
1731
1732 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1733 We know that we will need Charcount+1 trans at most to store the data
1734 (one row per char at worst case) So we preallocate both structures
1735 assuming worst case.
1736
1737 We then construct the trie using only the .next slots of the entry
1738 structs.
1739
3b753521 1740 We use the .check field of the first entry of the node temporarily to
a3621e74
YO
1741 make compression both faster and easier by keeping track of how many non
1742 zero fields are in the node.
1743
1744 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1745 transition.
1746
1747 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1748 number representing the first entry of the node, and state as a
1749 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1750 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1751 are 2 entrys per node. eg:
1752
1753 A B A B
1754 1. 2 4 1. 3 7
1755 2. 0 3 3. 0 5
1756 3. 0 0 5. 0 0
1757 4. 0 0 7. 0 0
1758
1759 The table is internally in the right hand, idx form. However as we also
1760 have to deal with the states array which is indexed by nodenum we have to
1761 use TRIE_NODENUM() to convert.
1762
1763 */
1e2e3d02
YO
1764 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1765 "%*sCompiling trie using table compiler\n",
1766 (int)depth * 2 + 2, ""));
3dab1dad 1767
c944940b
JH
1768 trie->trans = (reg_trie_trans *)
1769 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1770 * trie->uniquecharcount + 1,
1771 sizeof(reg_trie_trans) );
1772 trie->states = (reg_trie_state *)
1773 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1774 sizeof(reg_trie_state) );
a3621e74
YO
1775 next_alloc = trie->uniquecharcount + 1;
1776
3dab1dad 1777
a3621e74
YO
1778 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1779
c445ea15 1780 regnode * const noper = NEXTOPER( cur );
a28509cc
AL
1781 const U8 *uc = (U8*)STRING( noper );
1782 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1783
1784 U32 state = 1; /* required init */
1785
1786 U16 charid = 0; /* sanity init */
1787 U32 accept_state = 0; /* sanity init */
1788 U8 *scan = (U8*)NULL; /* sanity init */
1789
1790 STRLEN foldlen = 0; /* required init */
07be1b83 1791 U32 wordlen = 0; /* required init */
a3621e74
YO
1792 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1793
3dab1dad 1794 if ( OP(noper) != NOTHING ) {
786e8c11 1795 for ( ; uc < e ; uc += len ) {
a3621e74 1796
786e8c11 1797 TRIE_READ_CHAR;
a3621e74 1798
786e8c11
YO
1799 if ( uvc < 256 ) {
1800 charid = trie->charmap[ uvc ];
1801 } else {
55eed653 1802 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11 1803 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 1804 }
786e8c11
YO
1805 if ( charid ) {
1806 charid--;
1807 if ( !trie->trans[ state + charid ].next ) {
1808 trie->trans[ state + charid ].next = next_alloc;
1809 trie->trans[ state ].check++;
2e64971a
DM
1810 prev_states[TRIE_NODENUM(next_alloc)]
1811 = TRIE_NODENUM(state);
786e8c11
YO
1812 next_alloc += trie->uniquecharcount;
1813 }
1814 state = trie->trans[ state + charid ].next;
1815 } else {
1816 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1817 }
1818 /* charid is now 0 if we dont know the char read, or nonzero if we do */
a3621e74 1819 }
a3621e74 1820 }
3dab1dad
YO
1821 accept_state = TRIE_NODENUM( state );
1822 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
1823
1824 } /* end second pass */
1825
3dab1dad 1826 /* and now dump it out before we compress it */
2b8b4781
NC
1827 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1828 revcharmap,
1829 next_alloc, depth+1));
a3621e74 1830
a3621e74
YO
1831 {
1832 /*
1833 * Inplace compress the table.*
1834
1835 For sparse data sets the table constructed by the trie algorithm will
1836 be mostly 0/FAIL transitions or to put it another way mostly empty.
1837 (Note that leaf nodes will not contain any transitions.)
1838
1839 This algorithm compresses the tables by eliminating most such
1840 transitions, at the cost of a modest bit of extra work during lookup:
1841
1842 - Each states[] entry contains a .base field which indicates the
1843 index in the state[] array wheres its transition data is stored.
1844
3b753521 1845 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
1846
1847 - If .base is nonzero then charid is added to it to find an entry in
1848 the trans array.
1849
1850 -If trans[states[state].base+charid].check!=state then the
1851 transition is taken to be a 0/Fail transition. Thus if there are fail
1852 transitions at the front of the node then the .base offset will point
1853 somewhere inside the previous nodes data (or maybe even into a node
1854 even earlier), but the .check field determines if the transition is
1855 valid.
1856
786e8c11 1857 XXX - wrong maybe?
a3621e74 1858 The following process inplace converts the table to the compressed
3b753521 1859 table: We first do not compress the root node 1,and mark all its
a3621e74 1860 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
1861 allows us to do a DFA construction from the compressed table later,
1862 and ensures that any .base pointers we calculate later are greater
1863 than 0.
a3621e74
YO
1864
1865 - We set 'pos' to indicate the first entry of the second node.
1866
1867 - We then iterate over the columns of the node, finding the first and
1868 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1869 and set the .check pointers accordingly, and advance pos
1870 appropriately and repreat for the next node. Note that when we copy
1871 the next pointers we have to convert them from the original
1872 NODEIDX form to NODENUM form as the former is not valid post
1873 compression.
1874
1875 - If a node has no transitions used we mark its base as 0 and do not
1876 advance the pos pointer.
1877
1878 - If a node only has one transition we use a second pointer into the
1879 structure to fill in allocated fail transitions from other states.
1880 This pointer is independent of the main pointer and scans forward
1881 looking for null transitions that are allocated to a state. When it
1882 finds one it writes the single transition into the "hole". If the
786e8c11 1883 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
1884
1885 - Once compressed we can Renew/realloc the structures to release the
1886 excess space.
1887
1888 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1889 specifically Fig 3.47 and the associated pseudocode.
1890
1891 demq
1892 */
a3b680e6 1893 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 1894 U32 state, charid;
a3621e74 1895 U32 pos = 0, zp=0;
1e2e3d02 1896 trie->statecount = laststate;
a3621e74
YO
1897
1898 for ( state = 1 ; state < laststate ; state++ ) {
1899 U8 flag = 0;
a28509cc
AL
1900 const U32 stateidx = TRIE_NODEIDX( state );
1901 const U32 o_used = trie->trans[ stateidx ].check;
1902 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
1903 trie->trans[ stateidx ].check = 0;
1904
1905 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1906 if ( flag || trie->trans[ stateidx + charid ].next ) {
1907 if ( trie->trans[ stateidx + charid ].next ) {
1908 if (o_used == 1) {
1909 for ( ; zp < pos ; zp++ ) {
1910 if ( ! trie->trans[ zp ].next ) {
1911 break;
1912 }
1913 }
1914 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1915 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1916 trie->trans[ zp ].check = state;
1917 if ( ++zp > pos ) pos = zp;
1918 break;
1919 }
1920 used--;
1921 }
1922 if ( !flag ) {
1923 flag = 1;
1924 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1925 }
1926 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1927 trie->trans[ pos ].check = state;
1928 pos++;
1929 }
1930 }
1931 }
cc601c31 1932 trie->lasttrans = pos + 1;
c944940b
JH
1933 trie->states = (reg_trie_state *)
1934 PerlMemShared_realloc( trie->states, laststate
1935 * sizeof(reg_trie_state) );
a3621e74 1936 DEBUG_TRIE_COMPILE_MORE_r(
e4584336 1937 PerlIO_printf( Perl_debug_log,
3dab1dad
YO
1938 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1939 (int)depth * 2 + 2,"",
1940 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
5d7488b2
AL
1941 (IV)next_alloc,
1942 (IV)pos,
a3621e74
YO
1943 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1944 );
1945
1946 } /* end table compress */
1947 }
1e2e3d02
YO
1948 DEBUG_TRIE_COMPILE_MORE_r(
1949 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1950 (int)depth * 2 + 2, "",
1951 (UV)trie->statecount,
1952 (UV)trie->lasttrans)
1953 );
cc601c31 1954 /* resize the trans array to remove unused space */
c944940b
JH
1955 trie->trans = (reg_trie_trans *)
1956 PerlMemShared_realloc( trie->trans, trie->lasttrans
1957 * sizeof(reg_trie_trans) );
a3621e74 1958
3b753521 1959 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
1960 U8 nodetype =(U8)(flags & 0xFF);
1961 char *str=NULL;
786e8c11 1962
07be1b83 1963#ifdef DEBUGGING
e62cc96a 1964 regnode *optimize = NULL;
7122b237
YO
1965#ifdef RE_TRACK_PATTERN_OFFSETS
1966
b57a0404
JH
1967 U32 mjd_offset = 0;
1968 U32 mjd_nodelen = 0;
7122b237
YO
1969#endif /* RE_TRACK_PATTERN_OFFSETS */
1970#endif /* DEBUGGING */
a3621e74 1971 /*
3dab1dad
YO
1972 This means we convert either the first branch or the first Exact,
1973 depending on whether the thing following (in 'last') is a branch
1974 or not and whther first is the startbranch (ie is it a sub part of
1975 the alternation or is it the whole thing.)
3b753521 1976 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 1977 the whole branch sequence, including the first.
a3621e74 1978 */
3dab1dad 1979 /* Find the node we are going to overwrite */
7f69552c 1980 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 1981 /* branch sub-chain */
3dab1dad 1982 NEXT_OFF( first ) = (U16)(last - first);
7122b237 1983#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
1984 DEBUG_r({
1985 mjd_offset= Node_Offset((convert));
1986 mjd_nodelen= Node_Length((convert));
1987 });
7122b237 1988#endif
7f69552c 1989 /* whole branch chain */
7122b237
YO
1990 }
1991#ifdef RE_TRACK_PATTERN_OFFSETS
1992 else {
7f69552c
YO
1993 DEBUG_r({
1994 const regnode *nop = NEXTOPER( convert );
1995 mjd_offset= Node_Offset((nop));
1996 mjd_nodelen= Node_Length((nop));
1997 });
07be1b83
YO
1998 }
1999 DEBUG_OPTIMISE_r(
2000 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
2001 (int)depth * 2 + 2, "",
786e8c11 2002 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 2003 );
7122b237 2004#endif
3dab1dad
YO
2005 /* But first we check to see if there is a common prefix we can
2006 split out as an EXACT and put in front of the TRIE node. */
2007 trie->startstate= 1;
55eed653 2008 if ( trie->bitmap && !widecharmap && !trie->jump ) {
3dab1dad 2009 U32 state;
1e2e3d02 2010 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 2011 U32 ofs = 0;
8e11feef
RGS
2012 I32 idx = -1;
2013 U32 count = 0;
2014 const U32 base = trie->states[ state ].trans.base;
a3621e74 2015
3dab1dad 2016 if ( trie->states[state].wordnum )
8e11feef 2017 count = 1;
a3621e74 2018
8e11feef 2019 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
cc601c31
YO
2020 if ( ( base + ofs >= trie->uniquecharcount ) &&
2021 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
a3621e74
YO
2022 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
2023 {
3dab1dad 2024 if ( ++count > 1 ) {
2b8b4781 2025 SV **tmp = av_fetch( revcharmap, ofs, 0);
07be1b83 2026 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 2027 if ( state == 1 ) break;
3dab1dad
YO
2028 if ( count == 2 ) {
2029 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
2030 DEBUG_OPTIMISE_r(
8e11feef
RGS
2031 PerlIO_printf(Perl_debug_log,
2032 "%*sNew Start State=%"UVuf" Class: [",
2033 (int)depth * 2 + 2, "",
786e8c11 2034 (UV)state));
be8e71aa 2035 if (idx >= 0) {
2b8b4781 2036 SV ** const tmp = av_fetch( revcharmap, idx, 0);
be8e71aa 2037 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 2038
3dab1dad 2039 TRIE_BITMAP_SET(trie,*ch);
8e11feef
RGS
2040 if ( folder )
2041 TRIE_BITMAP_SET(trie, folder[ *ch ]);
3dab1dad 2042 DEBUG_OPTIMISE_r(
f1f66076 2043 PerlIO_printf(Perl_debug_log, "%s", (char*)ch)
3dab1dad 2044 );
8e11feef
RGS
2045 }
2046 }
2047 TRIE_BITMAP_SET(trie,*ch);
2048 if ( folder )
2049 TRIE_BITMAP_SET(trie,folder[ *ch ]);
2050 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
2051 }
2052 idx = ofs;
2053 }
3dab1dad
YO
2054 }
2055 if ( count == 1 ) {
2b8b4781 2056 SV **tmp = av_fetch( revcharmap, idx, 0);
c490c714
YO
2057 STRLEN len;
2058 char *ch = SvPV( *tmp, len );
de734bd5
A
2059 DEBUG_OPTIMISE_r({
2060 SV *sv=sv_newmortal();
8e11feef
RGS
2061 PerlIO_printf( Perl_debug_log,
2062 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
2063 (int)depth * 2 + 2, "",
de734bd5
A
2064 (UV)state, (UV)idx,
2065 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
2066 PL_colors[0], PL_colors[1],
2067 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2068 PERL_PV_ESCAPE_FIRSTCHAR
2069 )
2070 );
2071 });
3dab1dad
YO
2072 if ( state==1 ) {
2073 OP( convert ) = nodetype;
2074 str=STRING(convert);
2075 STR_LEN(convert)=0;
2076 }
c490c714
YO
2077 STR_LEN(convert) += len;
2078 while (len--)
de734bd5 2079 *str++ = *ch++;
8e11feef 2080 } else {
f9049ba1 2081#ifdef DEBUGGING
8e11feef
RGS
2082 if (state>1)
2083 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
f9049ba1 2084#endif
8e11feef
RGS
2085 break;
2086 }
2087 }
2e64971a 2088 trie->prefixlen = (state-1);
3dab1dad 2089 if (str) {
8e11feef 2090 regnode *n = convert+NODE_SZ_STR(convert);
07be1b83 2091 NEXT_OFF(convert) = NODE_SZ_STR(convert);
8e11feef 2092 trie->startstate = state;
07be1b83
YO
2093 trie->minlen -= (state - 1);
2094 trie->maxlen -= (state - 1);
33809eae
JH
2095#ifdef DEBUGGING
2096 /* At least the UNICOS C compiler choked on this
2097 * being argument to DEBUG_r(), so let's just have
2098 * it right here. */
2099 if (
2100#ifdef PERL_EXT_RE_BUILD
2101 1
2102#else
2103 DEBUG_r_TEST
2104#endif
2105 ) {
2106 regnode *fix = convert;
2107 U32 word = trie->wordcount;
2108 mjd_nodelen++;
2109 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2110 while( ++fix < n ) {
2111 Set_Node_Offset_Length(fix, 0, 0);
2112 }
2113 while (word--) {
2114 SV ** const tmp = av_fetch( trie_words, word, 0 );
2115 if (tmp) {
2116 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2117 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2118 else
2119 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2120 }
2121 }
2122 }
2123#endif
8e11feef
RGS
2124 if (trie->maxlen) {
2125 convert = n;
2126 } else {
3dab1dad 2127 NEXT_OFF(convert) = (U16)(tail - convert);
a5ca303d 2128 DEBUG_r(optimize= n);
3dab1dad
YO
2129 }
2130 }
2131 }
a5ca303d
YO
2132 if (!jumper)
2133 jumper = last;
3dab1dad 2134 if ( trie->maxlen ) {
8e11feef
RGS
2135 NEXT_OFF( convert ) = (U16)(tail - convert);
2136 ARG_SET( convert, data_slot );
786e8c11
YO
2137 /* Store the offset to the first unabsorbed branch in
2138 jump[0], which is otherwise unused by the jump logic.
2139 We use this when dumping a trie and during optimisation. */
2140 if (trie->jump)
7f69552c 2141 trie->jump[0] = (U16)(nextbranch - convert);
a5ca303d 2142
6c48061a
YO
2143 /* If the start state is not accepting (meaning there is no empty string/NOTHING)
2144 * and there is a bitmap
2145 * and the first "jump target" node we found leaves enough room
2146 * then convert the TRIE node into a TRIEC node, with the bitmap
2147 * embedded inline in the opcode - this is hypothetically faster.
2148 */
2149 if ( !trie->states[trie->startstate].wordnum
2150 && trie->bitmap
2151 && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
786e8c11
YO
2152 {
2153 OP( convert ) = TRIEC;
2154 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
446bd890 2155 PerlMemShared_free(trie->bitmap);
786e8c11
YO
2156 trie->bitmap= NULL;
2157 } else
2158 OP( convert ) = TRIE;
a3621e74 2159
3dab1dad
YO
2160 /* store the type in the flags */
2161 convert->flags = nodetype;
a5ca303d
YO
2162 DEBUG_r({
2163 optimize = convert
2164 + NODE_STEP_REGNODE
2165 + regarglen[ OP( convert ) ];
2166 });
2167 /* XXX We really should free up the resource in trie now,
2168 as we won't use them - (which resources?) dmq */
3dab1dad 2169 }
a3621e74 2170 /* needed for dumping*/
e62cc96a 2171 DEBUG_r(if (optimize) {
07be1b83 2172 regnode *opt = convert;
bcdf7404 2173
e62cc96a 2174 while ( ++opt < optimize) {
07be1b83
YO
2175 Set_Node_Offset_Length(opt,0,0);
2176 }
786e8c11
YO
2177 /*
2178 Try to clean up some of the debris left after the
2179 optimisation.
a3621e74 2180 */
786e8c11 2181 while( optimize < jumper ) {
07be1b83 2182 mjd_nodelen += Node_Length((optimize));
a3621e74 2183 OP( optimize ) = OPTIMIZED;
07be1b83 2184 Set_Node_Offset_Length(optimize,0,0);
a3621e74
YO
2185 optimize++;
2186 }
07be1b83 2187 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
a3621e74
YO
2188 });
2189 } /* end node insert */
2e64971a
DM
2190
2191 /* Finish populating the prev field of the wordinfo array. Walk back
2192 * from each accept state until we find another accept state, and if
2193 * so, point the first word's .prev field at the second word. If the
2194 * second already has a .prev field set, stop now. This will be the
2195 * case either if we've already processed that word's accept state,
3b753521
FN
2196 * or that state had multiple words, and the overspill words were
2197 * already linked up earlier.
2e64971a
DM
2198 */
2199 {
2200 U16 word;
2201 U32 state;
2202 U16 prev;
2203
2204 for (word=1; word <= trie->wordcount; word++) {
2205 prev = 0;
2206 if (trie->wordinfo[word].prev)
2207 continue;
2208 state = trie->wordinfo[word].accept;
2209 while (state) {
2210 state = prev_states[state];
2211 if (!state)
2212 break;
2213 prev = trie->states[state].wordnum;
2214 if (prev)
2215 break;
2216 }
2217 trie->wordinfo[word].prev = prev;
2218 }
2219 Safefree(prev_states);
2220 }
2221
2222
2223 /* and now dump out the compressed format */
2224 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
2225
55eed653 2226 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2b8b4781
NC
2227#ifdef DEBUGGING
2228 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2229 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2230#else
2231 SvREFCNT_dec(revcharmap);
07be1b83 2232#endif
786e8c11
YO
2233 return trie->jump
2234 ? MADE_JUMP_TRIE
2235 : trie->startstate>1
2236 ? MADE_EXACT_TRIE
2237 : MADE_TRIE;
2238}
2239
2240STATIC void
2241S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2242{
3b753521 2243/* The Trie is constructed and compressed now so we can build a fail array if it's needed
786e8c11
YO
2244
2245 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2246 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2247 ISBN 0-201-10088-6
2248
2249 We find the fail state for each state in the trie, this state is the longest proper
3b753521
FN
2250 suffix of the current state's 'word' that is also a proper prefix of another word in our
2251 trie. State 1 represents the word '' and is thus the default fail state. This allows
786e8c11
YO
2252 the DFA not to have to restart after its tried and failed a word at a given point, it
2253 simply continues as though it had been matching the other word in the first place.
2254 Consider
2255 'abcdgu'=~/abcdefg|cdgu/
2256 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
3b753521
FN
2257 fail, which would bring us to the state representing 'd' in the second word where we would
2258 try 'g' and succeed, proceeding to match 'cdgu'.
786e8c11
YO
2259 */
2260 /* add a fail transition */
3251b653
NC
2261 const U32 trie_offset = ARG(source);
2262 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
786e8c11
YO
2263 U32 *q;
2264 const U32 ucharcount = trie->uniquecharcount;
1e2e3d02 2265 const U32 numstates = trie->statecount;
786e8c11
YO
2266 const U32 ubound = trie->lasttrans + ucharcount;
2267 U32 q_read = 0;
2268 U32 q_write = 0;
2269 U32 charid;
2270 U32 base = trie->states[ 1 ].trans.base;
2271 U32 *fail;
2272 reg_ac_data *aho;
2273 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2274 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2275
2276 PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE;
786e8c11
YO
2277#ifndef DEBUGGING
2278 PERL_UNUSED_ARG(depth);
2279#endif
2280
2281
2282 ARG_SET( stclass, data_slot );
c944940b 2283 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
f8fc2ecf 2284 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3251b653 2285 aho->trie=trie_offset;
446bd890
NC
2286 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2287 Copy( trie->states, aho->states, numstates, reg_trie_state );
786e8c11 2288 Newxz( q, numstates, U32);
c944940b 2289 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
786e8c11
YO
2290 aho->refcount = 1;
2291 fail = aho->fail;
2292 /* initialize fail[0..1] to be 1 so that we always have
2293 a valid final fail state */
2294 fail[ 0 ] = fail[ 1 ] = 1;
2295
2296 for ( charid = 0; charid < ucharcount ; charid++ ) {
2297 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2298 if ( newstate ) {
2299 q[ q_write ] = newstate;
2300 /* set to point at the root */
2301 fail[ q[ q_write++ ] ]=1;
2302 }
2303 }
2304 while ( q_read < q_write) {
2305 const U32 cur = q[ q_read++ % numstates ];
2306 base = trie->states[ cur ].trans.base;
2307
2308 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2309 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2310 if (ch_state) {
2311 U32 fail_state = cur;
2312 U32 fail_base;
2313 do {
2314 fail_state = fail[ fail_state ];
2315 fail_base = aho->states[ fail_state ].trans.base;
2316 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2317
2318 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2319 fail[ ch_state ] = fail_state;
2320 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2321 {
2322 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2323 }
2324 q[ q_write++ % numstates] = ch_state;
2325 }
2326 }
2327 }
2328 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2329 when we fail in state 1, this allows us to use the
2330 charclass scan to find a valid start char. This is based on the principle
2331 that theres a good chance the string being searched contains lots of stuff
2332 that cant be a start char.
2333 */
2334 fail[ 0 ] = fail[ 1 ] = 0;
2335 DEBUG_TRIE_COMPILE_r({
6d99fb9b
JH
2336 PerlIO_printf(Perl_debug_log,
2337 "%*sStclass Failtable (%"UVuf" states): 0",
2338 (int)(depth * 2), "", (UV)numstates
1e2e3d02 2339 );
786e8c11
YO
2340 for( q_read=1; q_read<numstates; q_read++ ) {
2341 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2342 }
2343 PerlIO_printf(Perl_debug_log, "\n");
2344 });
2345 Safefree(q);
2346 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
a3621e74
YO
2347}
2348
786e8c11 2349
a3621e74 2350/*
5d1c421c
JH
2351 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2352 * These need to be revisited when a newer toolchain becomes available.
2353 */
2354#if defined(__sparc64__) && defined(__GNUC__)
2355# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2356# undef SPARC64_GCC_WORKAROUND
2357# define SPARC64_GCC_WORKAROUND 1
2358# endif
2359#endif
2360
07be1b83 2361#define DEBUG_PEEP(str,scan,depth) \
b515a41d 2362 DEBUG_OPTIMISE_r({if (scan){ \
07be1b83
YO
2363 SV * const mysv=sv_newmortal(); \
2364 regnode *Next = regnext(scan); \
2365 regprop(RExC_rx, mysv, scan); \
7f69552c 2366 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
07be1b83
YO
2367 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2368 Next ? (REG_NODE_NUM(Next)) : 0 ); \
b515a41d 2369 }});
07be1b83 2370
1de06328
YO
2371
2372
2373
2374
07be1b83
YO
2375#define JOIN_EXACT(scan,min,flags) \
2376 if (PL_regkind[OP(scan)] == EXACT) \
2377 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2378
be8e71aa 2379STATIC U32
07be1b83
YO
2380S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2381 /* Merge several consecutive EXACTish nodes into one. */
2382 regnode *n = regnext(scan);
2383 U32 stringok = 1;
2384 regnode *next = scan + NODE_SZ_STR(scan);
2385 U32 merged = 0;
2386 U32 stopnow = 0;
2387#ifdef DEBUGGING
2388 regnode *stop = scan;
72f13be8 2389 GET_RE_DEBUG_FLAGS_DECL;
f9049ba1 2390#else
d47053eb
RGS
2391 PERL_UNUSED_ARG(depth);
2392#endif
7918f24d
NC
2393
2394 PERL_ARGS_ASSERT_JOIN_EXACT;
d47053eb 2395#ifndef EXPERIMENTAL_INPLACESCAN
f9049ba1
SP
2396 PERL_UNUSED_ARG(flags);
2397 PERL_UNUSED_ARG(val);
07be1b83 2398#endif
07be1b83
YO
2399 DEBUG_PEEP("join",scan,depth);
2400
2401 /* Skip NOTHING, merge EXACT*. */
2402 while (n &&
2403 ( PL_regkind[OP(n)] == NOTHING ||
2404 (stringok && (OP(n) == OP(scan))))
2405 && NEXT_OFF(n)
2406 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2407
2408 if (OP(n) == TAIL || n > next)
2409 stringok = 0;
2410 if (PL_regkind[OP(n)] == NOTHING) {
07be1b83
YO
2411 DEBUG_PEEP("skip:",n,depth);
2412 NEXT_OFF(scan) += NEXT_OFF(n);
2413 next = n + NODE_STEP_REGNODE;
2414#ifdef DEBUGGING
2415 if (stringok)
2416 stop = n;
2417#endif
2418 n = regnext(n);
2419 }
2420 else if (stringok) {
786e8c11 2421 const unsigned int oldl = STR_LEN(scan);
07be1b83
YO
2422 regnode * const nnext = regnext(n);
2423
2424 DEBUG_PEEP("merg",n,depth);
2425
2426 merged++;
2427 if (oldl + STR_LEN(n) > U8_MAX)
2428 break;
2429 NEXT_OFF(scan) += NEXT_OFF(n);
2430 STR_LEN(scan) += STR_LEN(n);
2431 next = n + NODE_SZ_STR(n);
2432 /* Now we can overwrite *n : */
2433 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2434#ifdef DEBUGGING
2435 stop = next - 1;
2436#endif
2437 n = nnext;
2438 if (stopnow) break;
2439 }
2440
d47053eb
RGS
2441#ifdef EXPERIMENTAL_INPLACESCAN
2442 if (flags && !NEXT_OFF(n)) {
2443 DEBUG_PEEP("atch", val, depth);
2444 if (reg_off_by_arg[OP(n)]) {
2445 ARG_SET(n, val - n);
2446 }
2447 else {
2448 NEXT_OFF(n) = val - n;
2449 }
2450 stopnow = 1;
2451 }
07be1b83
YO
2452#endif
2453 }
ced7f090
KW
2454#define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390
2455#define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS
2456#define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0
2457#define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS
2c2b7f86
KW
2458
2459 if (UTF
2460 && ( OP(scan) == EXACTF || OP(scan) == EXACTFU)
2461 && ( STR_LEN(scan) >= 6 ) )
2462 {
07be1b83
YO
2463 /*
2464 Two problematic code points in Unicode casefolding of EXACT nodes:
2465
2466 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2467 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2468
2469 which casefold to
2470
2471 Unicode UTF-8
2472
2473 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2474 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2475
2476 This means that in case-insensitive matching (or "loose matching",
2477 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2478 length of the above casefolded versions) can match a target string
2479 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2480 This would rather mess up the minimum length computation.
2481
2482 What we'll do is to look for the tail four bytes, and then peek
2483 at the preceding two bytes to see whether we need to decrease
2484 the minimum length by four (six minus two).
2485
2486 Thanks to the design of UTF-8, there cannot be false matches:
2487 A sequence of valid UTF-8 bytes cannot be a subsequence of
2488 another valid sequence of UTF-8 bytes.
2489
2490 */
2491 char * const s0 = STRING(scan), *s, *t;
2492 char * const s1 = s0 + STR_LEN(scan) - 1;
2493 char * const s2 = s1 - 4;
e294cc5d
JH
2494#ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2495 const char t0[] = "\xaf\x49\xaf\x42";
2496#else
07be1b83 2497 const char t0[] = "\xcc\x88\xcc\x81";
e294cc5d 2498#endif
07be1b83
YO
2499 const char * const t1 = t0 + 3;
2500
2501 for (s = s0 + 2;
2502 s < s2 && (t = ninstr(s, s1, t0, t1));
2503 s = t + 4) {
e294cc5d
JH
2504#ifdef EBCDIC
2505 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2506 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2507#else
07be1b83
YO
2508 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2509 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
e294cc5d 2510#endif
07be1b83
YO
2511 *min -= 4;
2512 }
2513 }
2514
2515#ifdef DEBUGGING
2516 /* Allow dumping */
2517 n = scan + NODE_SZ_STR(scan);
2518 while (n <= stop) {
2519 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2520 OP(n) = OPTIMIZED;
2521 NEXT_OFF(n) = 0;
2522 }
2523 n++;
2524 }
2525#endif
2526 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2527 return stopnow;
2528}
2529
653099ff
GS
2530/* REx optimizer. Converts nodes into quickier variants "in place".
2531 Finds fixed substrings. */
2532
a0288114 2533/* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
c277df42
IZ
2534 to the position after last scanned or to NULL. */
2535
40d049e4
YO
2536#define INIT_AND_WITHP \
2537 assert(!and_withp); \
2538 Newx(and_withp,1,struct regnode_charclass_class); \
2539 SAVEFREEPV(and_withp)
07be1b83 2540
b515a41d
YO
2541/* this is a chain of data about sub patterns we are processing that
2542 need to be handled seperately/specially in study_chunk. Its so
2543 we can simulate recursion without losing state. */
2544struct scan_frame;
2545typedef struct scan_frame {
2546 regnode *last; /* last node to process in this frame */
2547 regnode *next; /* next node to process when last is reached */
2548 struct scan_frame *prev; /*previous frame*/
2549 I32 stop; /* what stopparen do we use */
2550} scan_frame;
2551
304ee84b
YO
2552
2553#define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2554
e1d1eefb
YO
2555#define CASE_SYNST_FNC(nAmE) \
2556case nAmE: \
2557 if (flags & SCF_DO_STCLASS_AND) { \
2558 for (value = 0; value < 256; value++) \
2559 if (!is_ ## nAmE ## _cp(value)) \
2560 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2561 } \
2562 else { \
2563 for (value = 0; value < 256; value++) \
2564 if (is_ ## nAmE ## _cp(value)) \
2565 ANYOF_BITMAP_SET(data->start_class, value); \
2566 } \
2567 break; \
2568case N ## nAmE: \
2569 if (flags & SCF_DO_STCLASS_AND) { \
2570 for (value = 0; value < 256; value++) \
2571 if (is_ ## nAmE ## _cp(value)) \
2572 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2573 } \
2574 else { \
2575 for (value = 0; value < 256; value++) \
2576 if (!is_ ## nAmE ## _cp(value)) \
2577 ANYOF_BITMAP_SET(data->start_class, value); \
2578 } \
2579 break
2580
2581
2582
76e3520e 2583STATIC I32
40d049e4 2584S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
1de06328 2585 I32 *minlenp, I32 *deltap,
40d049e4
YO
2586 regnode *last,
2587 scan_data_t *data,
2588 I32 stopparen,
2589 U8* recursed,
2590 struct regnode_charclass_class *and_withp,
2591 U32 flags, U32 depth)
c277df42
IZ
2592 /* scanp: Start here (read-write). */
2593 /* deltap: Write maxlen-minlen here. */
2594 /* last: Stop before this one. */
40d049e4
YO
2595 /* data: string data about the pattern */
2596 /* stopparen: treat close N as END */
2597 /* recursed: which subroutines have we recursed into */
2598 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
c277df42 2599{
97aff369 2600 dVAR;
c277df42
IZ
2601 I32 min = 0, pars = 0, code;
2602 regnode *scan = *scanp, *next;
2603 I32 delta = 0;
2604 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
aca2d497 2605 int is_inf_internal = 0; /* The studied chunk is infinite */
c277df42
IZ
2606 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2607 scan_data_t data_fake;
a3621e74 2608 SV *re_trie_maxbuff = NULL;
786e8c11 2609 regnode *first_non_open = scan;
e2e6a0f1 2610 I32 stopmin = I32_MAX;
8aa23a47 2611 scan_frame *frame = NULL;
a3621e74 2612 GET_RE_DEBUG_FLAGS_DECL;
8aa23a47 2613
7918f24d
NC
2614 PERL_ARGS_ASSERT_STUDY_CHUNK;
2615
13a24bad 2616#ifdef DEBUGGING
40d049e4 2617 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
13a24bad 2618#endif
40d049e4 2619
786e8c11 2620 if ( depth == 0 ) {
40d049e4 2621 while (first_non_open && OP(first_non_open) == OPEN)
786e8c11
YO
2622 first_non_open=regnext(first_non_open);
2623 }
2624
b81d288d 2625
8aa23a47
YO
2626 fake_study_recurse:
2627 while ( scan && OP(scan) != END && scan < last ){
2628 /* Peephole optimizer: */
304ee84b 2629 DEBUG_STUDYDATA("Peep:", data,depth);
8aa23a47
YO
2630 DEBUG_PEEP("Peep",scan,depth);
2631 JOIN_EXACT(scan,&min,0);
2632
2633 /* Follow the next-chain of the current node and optimize
2634 away all the NOTHINGs from it. */
2635 if (OP(scan) != CURLYX) {
2636 const int max = (reg_off_by_arg[OP(scan)]
2637 ? I32_MAX
2638 /* I32 may be smaller than U16 on CRAYs! */
2639 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2640 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2641 int noff;
2642 regnode *n = scan;
2643
2644 /* Skip NOTHING and LONGJMP. */
2645 while ((n = regnext(n))
2646 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2647 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2648 && off + noff < max)
2649 off += noff;
2650 if (reg_off_by_arg[OP(scan)])
2651 ARG(scan) = off;
2652 else
2653 NEXT_OFF(scan) = off;
2654 }
a3621e74 2655
c277df42 2656
8aa23a47
YO
2657
2658 /* The principal pseudo-switch. Cannot be a switch, since we
2659 look into several different things. */
2660 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2661 || OP(scan) == IFTHEN) {
2662 next = regnext(scan);
2663 code = OP(scan);
2664 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2665
2666 if (OP(next) == code || code == IFTHEN) {
2667 /* NOTE - There is similar code to this block below for handling
2668 TRIE nodes on a re-study. If you change stuff here check there
2669 too. */
2670 I32 max1 = 0, min1 = I32_MAX, num = 0;
2671 struct regnode_charclass_class accum;
2672 regnode * const startbranch=scan;
2673
2674 if (flags & SCF_DO_SUBSTR)
304ee84b 2675 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
8aa23a47
YO
2676 if (flags & SCF_DO_STCLASS)
2677 cl_init_zero(pRExC_state, &accum);
2678
2679 while (OP(scan) == code) {
2680 I32 deltanext, minnext, f = 0, fake;
2681 struct regnode_charclass_class this_class;
2682
2683 num++;
2684 data_fake.flags = 0;
2685 if (data) {
2686 data_fake.whilem_c = data->whilem_c;
2687 data_fake.last_closep = data->last_closep;
2688 }
2689 else
2690 data_fake.last_closep = &fake;
58e23c8d
YO
2691
2692 data_fake.pos_delta = delta;
8aa23a47
YO
2693 next = regnext(scan);
2694 scan = NEXTOPER(scan);
2695 if (code != BRANCH)
c277df42 2696 scan = NEXTOPER(scan);
8aa23a47
YO
2697 if (flags & SCF_DO_STCLASS) {
2698 cl_init(pRExC_state, &this_class);
2699 data_fake.start_class = &this_class;
2700 f = SCF_DO_STCLASS_AND;
58e23c8d 2701 }
8aa23a47
YO
2702 if (flags & SCF_WHILEM_VISITED_POS)
2703 f |= SCF_WHILEM_VISITED_POS;
2704
2705 /* we suppose the run is continuous, last=next...*/
2706 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2707 next, &data_fake,
2708 stopparen, recursed, NULL, f,depth+1);
2709 if (min1 > minnext)
2710 min1 = minnext;
2711 if (max1 < minnext + deltanext)
2712 max1 = minnext + deltanext;
2713 if (deltanext == I32_MAX)
2714 is_inf = is_inf_internal = 1;
2715 scan = next;
2716 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2717 pars++;
2718 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2719 if ( stopmin > minnext)
2720 stopmin = min + min1;
2721 flags &= ~SCF_DO_SUBSTR;
2722 if (data)
2723 data->flags |= SCF_SEEN_ACCEPT;
2724 }
2725 if (data) {
2726 if (data_fake.flags & SF_HAS_EVAL)
2727 data->flags |= SF_HAS_EVAL;
2728 data->whilem_c = data_fake.whilem_c;
3dab1dad 2729 }
8aa23a47
YO
2730 if (flags & SCF_DO_STCLASS)
2731 cl_or(pRExC_state, &accum, &this_class);
2732 }
2733 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2734 min1 = 0;
2735 if (flags & SCF_DO_SUBSTR) {
2736 data->pos_min += min1;
2737 data->pos_delta += max1 - min1;
2738 if (max1 != min1 || is_inf)
2739 data->longest = &(data->longest_float);
2740 }
2741 min += min1;
2742 delta += max1 - min1;
2743 if (flags & SCF_DO_STCLASS_OR) {
2744 cl_or(pRExC_state, data->start_class, &accum);
2745 if (min1) {
2746 cl_and(data->start_class, and_withp);
2747 flags &= ~SCF_DO_STCLASS;
653099ff 2748 }
8aa23a47
YO
2749 }
2750 else if (flags & SCF_DO_STCLASS_AND) {
2751 if (min1) {
2752 cl_and(data->start_class, &accum);
2753 flags &= ~SCF_DO_STCLASS;
de0c8cb8 2754 }
8aa23a47
YO
2755 else {
2756 /* Switch to OR mode: cache the old value of
2757 * data->start_class */
2758 INIT_AND_WITHP;
2759 StructCopy(data->start_class, and_withp,
2760 struct regnode_charclass_class);
2761 flags &= ~SCF_DO_STCLASS_AND;
2762 StructCopy(&accum, data->start_class,
2763 struct regnode_charclass_class);
2764 flags |= SCF_DO_STCLASS_OR;
2765 data->start_class->flags |= ANYOF_EOS;
de0c8cb8 2766 }
8aa23a47 2767 }
a3621e74 2768
8aa23a47
YO
2769 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2770 /* demq.
a3621e74 2771
8aa23a47
YO
2772 Assuming this was/is a branch we are dealing with: 'scan' now
2773 points at the item that follows the branch sequence, whatever
2774 it is. We now start at the beginning of the sequence and look
2775 for subsequences of
a3621e74 2776
8aa23a47
YO
2777 BRANCH->EXACT=>x1
2778 BRANCH->EXACT=>x2
2779 tail
a3621e74 2780
8aa23a47 2781 which would be constructed from a pattern like /A|LIST|OF|WORDS/
a3621e74 2782
8aa23a47
YO
2783 If we can find such a subseqence we need to turn the first
2784 element into a trie and then add the subsequent branch exact
2785 strings to the trie.
a3621e74 2786
8aa23a47 2787 We have two cases
a3621e74 2788
3b753521 2789 1. patterns where the whole set of branches can be converted.
a3621e74 2790
8aa23a47 2791 2. patterns where only a subset can be converted.
a3621e74 2792
8aa23a47
YO
2793 In case 1 we can replace the whole set with a single regop
2794 for the trie. In case 2 we need to keep the start and end
3b753521 2795 branches so
a3621e74 2796
8aa23a47
YO
2797 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2798 becomes BRANCH TRIE; BRANCH X;
786e8c11 2799
8aa23a47
YO
2800 There is an additional case, that being where there is a
2801 common prefix, which gets split out into an EXACT like node
2802 preceding the TRIE node.
a3621e74 2803
8aa23a47
YO
2804 If x(1..n)==tail then we can do a simple trie, if not we make
2805 a "jump" trie, such that when we match the appropriate word
2806 we "jump" to the appopriate tail node. Essentailly we turn
2807 a nested if into a case structure of sorts.
b515a41d 2808
8aa23a47
YO
2809 */
2810
2811 int made=0;
2812 if (!re_trie_maxbuff) {
2813 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2814 if (!SvIOK(re_trie_maxbuff))
2815 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2816 }
2817 if ( SvIV(re_trie_maxbuff)>=0 ) {
2818 regnode *cur;
2819 regnode *first = (regnode *)NULL;
2820 regnode *last = (regnode *)NULL;
2821 regnode *tail = scan;
2822 U8 optype = 0;
2823 U32 count=0;
a3621e74
YO
2824
2825#ifdef DEBUGGING
8aa23a47 2826 SV * const mysv = sv_newmortal(); /* for dumping */
a3621e74 2827#endif
8aa23a47
YO
2828 /* var tail is used because there may be a TAIL
2829 regop in the way. Ie, the exacts will point to the
2830 thing following the TAIL, but the last branch will
2831 point at the TAIL. So we advance tail. If we
2832 have nested (?:) we may have to move through several
2833 tails.
2834 */
2835
2836 while ( OP( tail ) == TAIL ) {
2837 /* this is the TAIL generated by (?:) */
2838 tail = regnext( tail );
2839 }
a3621e74 2840
8aa23a47
YO
2841
2842 DEBUG_OPTIMISE_r({
2843 regprop(RExC_rx, mysv, tail );
2844 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2845 (int)depth * 2 + 2, "",
2846 "Looking for TRIE'able sequences. Tail node is: ",
2847 SvPV_nolen_const( mysv )
2848 );
2849 });
2850
2851 /*
2852
2853 step through the branches, cur represents each
2854 branch, noper is the first thing to be matched
2855 as part of that branch and noper_next is the
2856 regnext() of that node. if noper is an EXACT
2857 and noper_next is the same as scan (our current
2858 position in the regex) then the EXACT branch is
2859 a possible optimization target. Once we have
2860 two or more consequetive such branches we can
2861 create a trie of the EXACT's contents and stich
2862 it in place. If the sequence represents all of
2863 the branches we eliminate the whole thing and
2864 replace it with a single TRIE. If it is a
2865 subsequence then we need to stitch it in. This
2866 means the first branch has to remain, and needs
2867 to be repointed at the item on the branch chain
2868 following the last branch optimized. This could
2869 be either a BRANCH, in which case the
2870 subsequence is internal, or it could be the
2871 item following the branch sequence in which
2872 case the subsequence is at the end.
2873
2874 */
2875
2876 /* dont use tail as the end marker for this traverse */
2877 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2878 regnode * const noper = NEXTOPER( cur );
b515a41d 2879#if defined(DEBUGGING) || defined(NOJUMPTRIE)
8aa23a47 2880 regnode * const noper_next = regnext( noper );
b515a41d
YO
2881#endif
2882
8aa23a47
YO
2883 DEBUG_OPTIMISE_r({
2884 regprop(RExC_rx, mysv, cur);
2885 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2886 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2887
2888 regprop(RExC_rx, mysv, noper);
2889 PerlIO_printf( Perl_debug_log, " -> %s",
2890 SvPV_nolen_const(mysv));
2891
2892 if ( noper_next ) {
2893 regprop(RExC_rx, mysv, noper_next );
2894 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2895 SvPV_nolen_const(mysv));
2896 }
2897 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2898 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2899 });
2900 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2901 : PL_regkind[ OP( noper ) ] == EXACT )
2902 || OP(noper) == NOTHING )
786e8c11 2903#ifdef NOJUMPTRIE
8aa23a47 2904 && noper_next == tail
786e8c11 2905#endif
8aa23a47
YO
2906 && count < U16_MAX)
2907 {
2908 count++;
2909 if ( !first || optype == NOTHING ) {
2910 if (!first) first = cur;
2911 optype = OP( noper );
2912 } else {
2913 last = cur;
2914 }
2915 } else {
a0a388a1 2916/*
0abd0d78
YO
2917 Currently we do not believe that the trie logic can
2918 handle case insensitive matching properly when the
2919 pattern is not unicode (thus forcing unicode semantics).
2920
2921 If/when this is fixed the following define can be swapped
2922 in below to fully enable trie logic.
2923
a0a388a1 2924#define TRIE_TYPE_IS_SAFE 1
0abd0d78
YO
2925
2926*/
2927#define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2928
a0a388a1 2929 if ( last && TRIE_TYPE_IS_SAFE ) {
8aa23a47
YO
2930 make_trie( pRExC_state,
2931 startbranch, first, cur, tail, count,
2932 optype, depth+1 );
2933 }
2934 if ( PL_regkind[ OP( noper ) ] == EXACT
786e8c11 2935#ifdef NOJUMPTRIE
8aa23a47 2936 && noper_next == tail
786e8c11 2937#endif
8aa23a47
YO
2938 ){
2939 count = 1;
2940 first = cur;
2941 optype = OP( noper );
2942 } else {
2943 count = 0;
2944 first = NULL;
2945 optype = 0;
2946 }
2947 last = NULL;
2948 }
2949 }
2950 DEBUG_OPTIMISE_r({
2951 regprop(RExC_rx, mysv, cur);
2952 PerlIO_printf( Perl_debug_log,
2953 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2954 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2955
2956 });
a0a388a1
YO
2957
2958 if ( last && TRIE_TYPE_IS_SAFE ) {
8aa23a47 2959 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3dab1dad 2960#ifdef TRIE_STUDY_OPT
8aa23a47
YO
2961 if ( ((made == MADE_EXACT_TRIE &&
2962 startbranch == first)
2963 || ( first_non_open == first )) &&
2964 depth==0 ) {
2965 flags |= SCF_TRIE_RESTUDY;
2966 if ( startbranch == first
2967 && scan == tail )
2968 {
2969 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2970 }
2971 }
3dab1dad 2972#endif
8aa23a47
YO
2973 }
2974 }
2975
2976 } /* do trie */
2977
653099ff 2978 }
8aa23a47
YO
2979 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2980 scan = NEXTOPER(NEXTOPER(scan));
2981 } else /* single branch is optimized. */
2982 scan = NEXTOPER(scan);
2983 continue;
2984 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2985 scan_frame *newframe = NULL;
2986 I32 paren;
2987 regnode *start;
2988 regnode *end;
2989
2990 if (OP(scan) != SUSPEND) {
2991 /* set the pointer */
2992 if (OP(scan) == GOSUB) {
2993 paren = ARG(scan);
2994 RExC_recurse[ARG2L(scan)] = scan;
2995 start = RExC_open_parens[paren-1];
2996 end = RExC_close_parens[paren-1];
2997 } else {
2998 paren = 0;
f8fc2ecf 2999 start = RExC_rxi->program + 1;
8aa23a47
YO
3000 end = RExC_opend;
3001 }
3002 if (!recursed) {
3003 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
3004 SAVEFREEPV(recursed);
3005 }
3006 if (!PAREN_TEST(recursed,paren+1)) {
3007 PAREN_SET(recursed,paren+1);
3008 Newx(newframe,1,scan_frame);
3009 } else {
3010 if (flags & SCF_DO_SUBSTR) {
304ee84b 3011 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3012 data->longest = &(data->longest_float);
3013 }
3014 is_inf = is_inf_internal = 1;
3015 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3016 cl_anything(pRExC_state, data->start_class);
3017 flags &= ~SCF_DO_STCLASS;
3018 }
3019 } else {
3020 Newx(newframe,1,scan_frame);
3021 paren = stopparen;
3022 start = scan+2;
3023 end = regnext(scan);
3024 }
3025 if (newframe) {
3026 assert(start);
3027 assert(end);
3028 SAVEFREEPV(newframe);
3029 newframe->next = regnext(scan);
3030 newframe->last = last;
3031 newframe->stop = stopparen;
3032 newframe->prev = frame;
3033
3034 frame = newframe;
3035 scan = start;
3036 stopparen = paren;
3037 last = end;
3038
3039 continue;
3040 }
3041 }
3042 else if (OP(scan) == EXACT) {
3043 I32 l = STR_LEN(scan);
3044 UV uc;
3045 if (UTF) {
3046 const U8 * const s = (U8*)STRING(scan);
3047 l = utf8_length(s, s + l);
3048 uc = utf8_to_uvchr(s, NULL);
3049 } else {
3050 uc = *((U8*)STRING(scan));
3051 }
3052 min += l;
3053 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
3054 /* The code below prefers earlier match for fixed
3055 offset, later match for variable offset. */
3056 if (data->last_end == -1) { /* Update the start info. */
3057 data->last_start_min = data->pos_min;
3058 data->last_start_max = is_inf
3059 ? I32_MAX : data->pos_min + data->pos_delta;
b515a41d 3060 }
8aa23a47
YO
3061 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
3062 if (UTF)
3063 SvUTF8_on(data->last_found);
3064 {
3065 SV * const sv = data->last_found;
3066 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3067 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3068 if (mg && mg->mg_len >= 0)
3069 mg->mg_len += utf8_length((U8*)STRING(scan),
3070 (U8*)STRING(scan)+STR_LEN(scan));
b515a41d 3071 }
8aa23a47
YO
3072 data->last_end = data->pos_min + l;
3073 data->pos_min += l; /* As in the first entry. */
3074 data->flags &= ~SF_BEFORE_EOL;
3075 }
3076 if (flags & SCF_DO_STCLASS_AND) {
3077 /* Check whether it is compatible with what we know already! */
3078 int compat = 1;
3079
54251c2e
KW
3080
3081 /* If compatibile, we or it in below. It is compatible if is
3082 * in the bitmp and either 1) its bit or its fold is set, or 2)
3083 * it's for a locale. Even if there isn't unicode semantics
3084 * here, at runtime there may be because of matching against a
3085 * utf8 string, so accept a possible false positive for
3086 * latin1-range folds */
8aa23a47
YO
3087 if (uc >= 0x100 ||
3088 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3089 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3090 && (!(data->start_class->flags & ANYOF_FOLD)
54251c2e 3091 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
8aa23a47
YO
3092 )
3093 compat = 0;
3094 ANYOF_CLASS_ZERO(data->start_class);
3095 ANYOF_BITMAP_ZERO(data->start_class);
3096 if (compat)
3097 ANYOF_BITMAP_SET(data->start_class, uc);
3098 data->start_class->flags &= ~ANYOF_EOS;
3099 if (uc < 0x100)
3100 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
3101 }
3102 else if (flags & SCF_DO_STCLASS_OR) {
3103 /* false positive possible if the class is case-folded */
3104 if (uc < 0x100)
3105 ANYOF_BITMAP_SET(data->start_class, uc);
3106 else
3107 data->start_class->flags |= ANYOF_UNICODE_ALL;
3108 data->start_class->flags &= ~ANYOF_EOS;
3109 cl_and(data->start_class, and_withp);
3110 }
3111 flags &= ~SCF_DO_STCLASS;
3112 }
3113 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
3114 I32 l = STR_LEN(scan);
3115 UV uc = *((U8*)STRING(scan));
3116
3117 /* Search for fixed substrings supports EXACT only. */
3118 if (flags & SCF_DO_SUBSTR) {
3119 assert(data);
304ee84b 3120 SCAN_COMMIT(pRExC_state, data, minlenp);
8aa23a47
YO
3121 }
3122 if (UTF) {
3123 const U8 * const s = (U8 *)STRING(scan);
3124 l = utf8_length(s, s + l);
3125 uc = utf8_to_uvchr(s, NULL);
3126 }
3127 min += l;
3128 if (flags & SCF_DO_SUBSTR)
3129 data->pos_min += l;
3130 if (flags & SCF_DO_STCLASS_AND) {
3131 /* Check whether it is compatible with what we know already! */
3132 int compat = 1;
8aa23a47 3133 if (uc >= 0x100 ||
54251c2e
KW
3134 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3135 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3136 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc])))
3137 {
8aa23a47 3138 compat = 0;
54251c2e 3139 }
8aa23a47
YO
3140 ANYOF_CLASS_ZERO(data->start_class);
3141 ANYOF_BITMAP_ZERO(data->start_class);
3142 if (compat) {
3143 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff 3144 data->start_class->flags &= ~ANYOF_EOS;
8aa23a47 3145 data->start_class->flags |= ANYOF_FOLD;
970c8436 3146 if (OP(scan) == EXACTFL) {
8aa23a47 3147 data->start_class->flags |= ANYOF_LOCALE;
970c8436
KW
3148 }
3149 else {
3150
54251c2e
KW
3151 /* Also set the other member of the fold pair. In case
3152 * that unicode semantics is called for at runtime, use
3153 * the full latin1 fold. (Can't do this for locale,
3154 * because not known until runtime */
3155 ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]);
970c8436 3156 }
653099ff 3157 }
8aa23a47
YO
3158 }
3159 else if (flags & SCF_DO_STCLASS_OR) {
3160 if (data->start_class->flags & ANYOF_FOLD) {
3161 /* false positive possible if the class is case-folded.
3162 Assume that the locale settings are the same... */
970c8436 3163 if (uc < 0x100) {
1aa99e6b 3164 ANYOF_BITMAP_SET(data->start_class, uc);
970c8436
KW
3165 if (OP(scan) != EXACTFL) {
3166
3167 /* And set the other member of the fold pair, but
3168 * can't do that in locale because not known until
3169 * run-time */
3170 ANYOF_BITMAP_SET(data->start_class,
54251c2e 3171 PL_fold_latin1[uc]);
970c8436
KW
3172 }
3173 }
653099ff
GS
3174 data->start_class->flags &= ~ANYOF_EOS;
3175 }
8aa23a47 3176 cl_and(data->start_class, and_withp);
653099ff 3177 }
8aa23a47
YO
3178 flags &= ~SCF_DO_STCLASS;
3179 }
e52fc539 3180 else if (REGNODE_VARIES(OP(scan))) {
8aa23a47
YO
3181 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3182 I32 f = flags, pos_before = 0;
3183 regnode * const oscan = scan;
3184 struct regnode_charclass_class this_class;
3185 struct regnode_charclass_class *oclass = NULL;
3186 I32 next_is_eval = 0;
3187
3188 switch (PL_regkind[OP(scan)]) {
3189 case WHILEM: /* End of (?:...)* . */
3190 scan = NEXTOPER(scan);
3191 goto finish;
3192 case PLUS:
3193 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3194 next = NEXTOPER(scan);
3195 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3196 mincount = 1;
3197 maxcount = REG_INFTY;
3198 next = regnext(scan);
3199 scan = NEXTOPER(scan);
3200 goto do_curly;
3201 }
3202 }
3203 if (flags & SCF_DO_SUBSTR)
3204 data->pos_min++;
3205 min++;
3206 /* Fall through. */
3207 case STAR:
3208 if (flags & SCF_DO_STCLASS) {
3209 mincount = 0;
3210 maxcount = REG_INFTY;
3211 next = regnext(scan);
3212 scan = NEXTOPER(scan);
3213 goto do_curly;
3214 }
3215 is_inf = is_inf_internal = 1;
3216 scan = regnext(scan);
c277df42 3217 if (flags & SCF_DO_SUBSTR) {
304ee84b 3218 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
8aa23a47 3219 data->longest = &(data->longest_float);
c277df42 3220 }
8aa23a47
YO
3221 goto optimize_curly_tail;
3222 case CURLY:
3223 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3224 && (scan->flags == stopparen))
3225 {
3226 mincount = 1;
3227 maxcount = 1;
3228 } else {
3229 mincount = ARG1(scan);
3230 maxcount = ARG2(scan);
653099ff 3231 }
8aa23a47
YO
3232 next = regnext(scan);
3233 if (OP(scan) == CURLYX) {
3234 I32 lp = (data ? *(data->last_closep) : 0);
3235 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
653099ff 3236 }
8aa23a47
YO
3237 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3238 next_is_eval = (OP(scan) == EVAL);
3239 do_curly:
3240 if (flags & SCF_DO_SUBSTR) {
304ee84b 3241 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
8aa23a47 3242 pos_before = data->pos_min;
b45f050a 3243 }
8aa23a47
YO
3244 if (data) {
3245 fl = data->flags;
3246 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3247 if (is_inf)
3248 data->flags |= SF_IS_INF;
3249 }
3250 if (flags & SCF_DO_STCLASS) {
3251 cl_init(pRExC_state, &this_class);
3252 oclass = data->start_class;
3253 data->start_class = &this_class;
3254 f |= SCF_DO_STCLASS_AND;
3255 f &= ~SCF_DO_STCLASS_OR;
3256 }
779bcb7d
NC
3257 /* Exclude from super-linear cache processing any {n,m}
3258 regops for which the combination of input pos and regex
3259 pos is not enough information to determine if a match
3260 will be possible.
3261
3262 For example, in the regex /foo(bar\s*){4,8}baz/ with the
3263 regex pos at the \s*, the prospects for a match depend not
3264 only on the input position but also on how many (bar\s*)
3265 repeats into the {4,8} we are. */
3266 if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY))
8aa23a47 3267 f &= ~SCF_WHILEM_VISITED_POS;
b45f050a 3268
8aa23a47
YO
3269 /* This will finish on WHILEM, setting scan, or on NULL: */
3270 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3271 last, data, stopparen, recursed, NULL,
3272 (mincount == 0
3273 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
b515a41d 3274
8aa23a47
YO
3275 if (flags & SCF_DO_STCLASS)
3276 data->start_class = oclass;
3277 if (mincount == 0 || minnext == 0) {
3278 if (flags & SCF_DO_STCLASS_OR) {
3279 cl_or(pRExC_state, data->start_class, &this_class);
3280 }
3281 else if (flags & SCF_DO_STCLASS_AND) {
3282 /* Switch to OR mode: cache the old value of
3283 * data->start_class */
3284 INIT_AND_WITHP;
3285 StructCopy(data->start_class, and_withp,
3286 struct regnode_charclass_class);
3287 flags &= ~SCF_DO_STCLASS_AND;
3288 StructCopy(&this_class, data->start_class,
3289 struct regnode_charclass_class);
3290 flags |= SCF_DO_STCLASS_OR;
3291 data->start_class->flags |= ANYOF_EOS;
3292 }
3293 } else { /* Non-zero len */
3294 if (flags & SCF_DO_STCLASS_OR) {
3295 cl_or(pRExC_state, data->start_class, &this_class);
3296 cl_and(data->start_class, and_withp);
3297 }
3298 else if (flags & SCF_DO_STCLASS_AND)
3299 cl_and(data->start_class, &this_class);
3300 flags &= ~SCF_DO_STCLASS;
3301 }
3302 if (!scan) /* It was not CURLYX, but CURLY. */
3303 scan = next;
3304 if ( /* ? quantifier ok, except for (?{ ... }) */
3305 (next_is_eval || !(mincount == 0 && maxcount == 1))
3306 && (minnext == 0) && (deltanext == 0)
3307 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
668c081a 3308 && maxcount <= REG_INFTY/3) /* Complement check for big count */
8aa23a47 3309 {
668c081a
NC
3310 ckWARNreg(RExC_parse,
3311 "Quantifier unexpected on zero-length expression");
8aa23a47
YO
3312 }
3313
3314 min += minnext * mincount;
3315 is_inf_internal |= ((maxcount == REG_INFTY
3316 && (minnext + deltanext) > 0)
3317 || deltanext == I32_MAX);
3318 is_inf |= is_inf_internal;
3319 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3320
3321 /* Try powerful optimization CURLYX => CURLYN. */
3322 if ( OP(oscan) == CURLYX && data
3323 && data->flags & SF_IN_PAR
3324 && !(data->flags & SF_HAS_EVAL)
3325 && !deltanext && minnext == 1 ) {
3326 /* Try to optimize to CURLYN. */
3327 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3328 regnode * const nxt1 = nxt;
497b47a8 3329#ifdef DEBUGGING
8aa23a47 3330 regnode *nxt2;
497b47a8 3331#endif
c277df42 3332
8aa23a47
YO
3333 /* Skip open. */
3334 nxt = regnext(nxt);
e52fc539 3335 if (!REGNODE_SIMPLE(OP(nxt))
8aa23a47
YO
3336 && !(PL_regkind[OP(nxt)] == EXACT
3337 && STR_LEN(nxt) == 1))
3338 goto nogo;
497b47a8 3339#ifdef DEBUGGING
8aa23a47 3340 nxt2 = nxt;
497b47a8 3341#endif
8aa23a47
YO
3342 nxt = regnext(nxt);
3343 if (OP(nxt) != CLOSE)
3344 goto nogo;
3345 if (RExC_open_parens) {
3346 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3347 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3348 }
3349 /* Now we know that nxt2 is the only contents: */
3350 oscan->flags = (U8)ARG(nxt);
3351 OP(oscan) = CURLYN;
3352 OP(nxt1) = NOTHING; /* was OPEN. */
40d049e4 3353
c277df42 3354#ifdef DEBUGGING
8aa23a47 3355 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
fda99bee
KW
3356 NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */
3357 NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */
8aa23a47
YO
3358 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3359 OP(nxt + 1) = OPTIMIZED; /* was count. */
fda99bee 3360 NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */
b81d288d 3361#endif
8aa23a47
YO
3362 }
3363 nogo:
3364
3365 /* Try optimization CURLYX => CURLYM. */
3366 if ( OP(oscan) == CURLYX && data
3367 && !(data->flags & SF_HAS_PAR)
3368 && !(data->flags & SF_HAS_EVAL)
3369 && !deltanext /* atom is fixed width */
3370 && minnext != 0 /* CURLYM can't handle zero width */
3371 ) {
3372 /* XXXX How to optimize if data == 0? */
3373 /* Optimize to a simpler form. */
3374 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3375 regnode *nxt2;
3376
3377 OP(oscan) = CURLYM;
3378 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3379 && (OP(nxt2) != WHILEM))
3380 nxt = nxt2;
3381 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3382 /* Need to optimize away parenths. */
b3c0965f 3383 if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) {
8aa23a47
YO
3384 /* Set the parenth number. */
3385 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3386
8aa23a47
YO
3387 oscan->flags = (U8)ARG(nxt);
3388 if (RExC_open_parens) {
3389 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3390 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
40d049e4 3391 }
8aa23a47
YO
3392 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3393 OP(nxt) = OPTIMIZED; /* was CLOSE. */
40d049e4 3394
c277df42 3395#ifdef DEBUGGING
8aa23a47
YO
3396 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3397 OP(nxt + 1) = OPTIMIZED; /* was count. */
3398 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3399 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
b81d288d 3400#endif
c277df42 3401#if 0
8aa23a47
YO
3402 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3403 regnode *nnxt = regnext(nxt1);
8aa23a47
YO
3404 if (nnxt == nxt) {
3405 if (reg_off_by_arg[OP(nxt1)])
3406 ARG_SET(nxt1, nxt2 - nxt1);
3407 else if (nxt2 - nxt1 < U16_MAX)
3408 NEXT_OFF(nxt1) = nxt2 - nxt1;
3409 else
3410 OP(nxt) = NOTHING; /* Cannot beautify */
c277df42 3411 }
8aa23a47 3412 nxt1 = nnxt;
c277df42 3413 }
5d1c421c 3414#endif
8aa23a47
YO
3415 /* Optimize again: */
3416 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3417 NULL, stopparen, recursed, NULL, 0,depth+1);
3418 }
3419 else
3420 oscan->flags = 0;
3421 }
3422 else if ((OP(oscan) == CURLYX)
3423 && (flags & SCF_WHILEM_VISITED_POS)
3424 /* See the comment on a similar expression above.
3b753521 3425 However, this time it's not a subexpression
8aa23a47
YO
3426 we care about, but the expression itself. */
3427 && (maxcount == REG_INFTY)
3428 && data && ++data->whilem_c < 16) {
3429 /* This stays as CURLYX, we can put the count/of pair. */
3430 /* Find WHILEM (as in regexec.c) */
3431 regnode *nxt = oscan + NEXT_OFF(oscan);
3432
3433 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3434 nxt += ARG(nxt);
3435 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3436 | (RExC_whilem_seen << 4)); /* On WHILEM */
3437 }
3438 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3439 pars++;
3440 if (flags & SCF_DO_SUBSTR) {
3441 SV *last_str = NULL;
3442 int counted = mincount != 0;
a0ed51b3 3443
8aa23a47
YO
3444 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3445#if defined(SPARC64_GCC_WORKAROUND)
3446 I32 b = 0;
3447 STRLEN l = 0;
3448 const char *s = NULL;
3449 I32 old = 0;
b515a41d 3450
8aa23a47
YO
3451 if (pos_before >= data->last_start_min)
3452 b = pos_before;
3453 else
3454 b = data->last_start_min;
b515a41d 3455
8aa23a47
YO
3456 l = 0;
3457 s = SvPV_const(data->last_found, l);
3458 old = b - data->last_start_min;
3459
3460#else
3461 I32 b = pos_before >= data->last_start_min
3462 ? pos_before : data->last_start_min;
3463 STRLEN l;
3464 const char * const s = SvPV_const(data->last_found, l);
3465 I32 old = b - data->last_start_min;
3466#endif
3467
3468 if (UTF)
3469 old = utf8_hop((U8*)s, old) - (U8*)s;
8aa23a47
YO
3470 l -= old;
3471 /* Get the added string: */
740cce10 3472 last_str = newSVpvn_utf8(s + old, l, UTF);
8aa23a47
YO
3473 if (deltanext == 0 && pos_before == b) {
3474 /* What was added is a constant string */
3475 if (mincount > 1) {
3476 SvGROW(last_str, (mincount * l) + 1);
3477 repeatcpy(SvPVX(last_str) + l,
3478 SvPVX_const(last_str), l, mincount - 1);
3479 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3480 /* Add additional parts. */
3481 SvCUR_set(data->last_found,
3482 SvCUR(data->last_found) - l);
3483 sv_catsv(data->last_found, last_str);
3484 {
3485 SV * sv = data->last_found;
3486 MAGIC *mg =
3487 SvUTF8(sv) && SvMAGICAL(sv) ?
3488 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3489 if (mg && mg->mg_len >= 0)
bd94e887 3490 mg->mg_len += CHR_SVLEN(last_str) - l;
b515a41d 3491 }
8aa23a47 3492 data->last_end += l * (mincount - 1);
b515a41d 3493 }
8aa23a47
YO
3494 } else {
3495 /* start offset must point into the last copy */
3496 data->last_start_min += minnext * (mincount - 1);
3497 data->last_start_max += is_inf ? I32_MAX
3498 : (maxcount - 1) * (minnext + data->pos_delta);
3499 }
c277df42 3500 }
8aa23a47
YO
3501 /* It is counted once already... */
3502 data->pos_min += minnext * (mincount - counted);
3503 data->pos_delta += - counted * deltanext +
3504 (minnext + deltanext) * maxcount - minnext * mincount;
3505 if (mincount != maxcount) {
3506 /* Cannot extend fixed substrings found inside
3507 the group. */
304ee84b 3508 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3509 if (mincount && last_str) {
3510 SV * const sv = data->last_found;
3511 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3512 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3513
3514 if (mg)
3515 mg->mg_len = -1;
3516 sv_setsv(sv, last_str);
3517 data->last_end = data->pos_min;
3518 data->last_start_min =
3519 data->pos_min - CHR_SVLEN(last_str);
3520 data->last_start_max = is_inf
3521 ? I32_MAX
3522 : data->pos_min + data->pos_delta
3523 - CHR_SVLEN(last_str);
3524 }
3525 data->longest = &(data->longest_float);
3526 }
3527 SvREFCNT_dec(last_str);
c277df42 3528 }
8aa23a47
YO
3529 if (data && (fl & SF_HAS_EVAL))
3530 data->flags |= SF_HAS_EVAL;
3531 optimize_curly_tail:
3532 if (OP(oscan) != CURLYX) {
3533 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3534 && NEXT_OFF(next))
3535 NEXT_OFF(oscan) += NEXT_OFF(next);
3536 }
3537 continue;
3538 default: /* REF and CLUMP only? */
3539 if (flags & SCF_DO_SUBSTR) {
304ee84b 3540 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
8aa23a47
YO
3541 data->longest = &(data->longest_float);
3542 }
3543 is_inf = is_inf_internal = 1;
3544 if (flags & SCF_DO_STCLASS_OR)
3545 cl_anything(pRExC_state, data->start_class);
3546 flags &= ~SCF_DO_STCLASS;
3547 break;
c277df42 3548 }
8aa23a47 3549 }
e1d1eefb
YO
3550 else if (OP(scan) == LNBREAK) {
3551 if (flags & SCF_DO_STCLASS) {
3552 int value = 0;
3553 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3554 if (flags & SCF_DO_STCLASS_AND) {
3555 for (value = 0; value < 256; value++)
e64b1bd1 3556 if (!is_VERTWS_cp(value))
b9a59e08
KW
3557 ANYOF_BITMAP_CLEAR(data->start_class, value);
3558 }
3559 else {
e1d1eefb 3560 for (value = 0; value < 256; value++)
e64b1bd1 3561 if (is_VERTWS_cp(value))
b9a59e08
KW
3562 ANYOF_BITMAP_SET(data->start_class, value);
3563 }
e1d1eefb
YO
3564 if (flags & SCF_DO_STCLASS_OR)
3565 cl_and(data->start_class, and_withp);
3566 flags &= ~SCF_DO_STCLASS;
3567 }
3568 min += 1;
f9a79580 3569 delta += 1;
e1d1eefb
YO
3570 if (flags & SCF_DO_SUBSTR) {
3571 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3572 data->pos_min += 1;
f9a79580 3573 data->pos_delta += 1;
e1d1eefb
YO
3574 data->longest = &(data->longest_float);
3575 }
e1d1eefb 3576 }
f9a79580 3577 else if (OP(scan) == FOLDCHAR) {
ced7f090 3578 int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2;
f9a79580
RGS
3579 flags &= ~SCF_DO_STCLASS;
3580 min += 1;
3581 delta += d;
3582 if (flags & SCF_DO_SUBSTR) {
3583 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3584 data->pos_min += 1;
3585 data->pos_delta += d;
3586 data->longest = &(data->longest_float);
3587 }
3588 }
e52fc539 3589 else if (REGNODE_SIMPLE(OP(scan))) {
8aa23a47 3590 int value = 0;
653099ff 3591
8aa23a47 3592 if (flags & SCF_DO_SUBSTR) {
304ee84b 3593 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3594 data->pos_min++;
3595 }
3596 min++;
3597 if (flags & SCF_DO_STCLASS) {
3598 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
b515a41d 3599
8aa23a47
YO
3600 /* Some of the logic below assumes that switching
3601 locale on will only add false positives. */
3602 switch (PL_regkind[OP(scan)]) {
3603 case SANY:
3604 default:
3605 do_default:
3606 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3607 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3608 cl_anything(pRExC_state, data->start_class);
3609 break;
3610 case REG_ANY:
3611 if (OP(scan) == SANY)
3612 goto do_default;
3613 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3614 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
cefafd73
KW
3615 || ((data->start_class->flags & ANYOF_CLASS)
3616 && ANYOF_CLASS_TEST_ANY_SET(data->start_class)));
8aa23a47 3617 cl_anything(pRExC_state, data->start_class);
653099ff 3618 }
8aa23a47
YO
3619 if (flags & SCF_DO_STCLASS_AND || !value)
3620 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3621 break;
3622 case ANYOF:
3623 if (flags & SCF_DO_STCLASS_AND)
3624 cl_and(data->start_class,
3625 (struct regnode_charclass_class*)scan);
653099ff 3626 else
8aa23a47
YO
3627 cl_or(pRExC_state, data->start_class,
3628 (struct regnode_charclass_class*)scan);
3629 break;
3630 case ALNUM:
3631 if (flags & SCF_DO_STCLASS_AND) {
3632 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3633 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
a12cf05f
KW
3634 if (FLAGS(scan) & USE_UNI) {
3635 for (value = 0; value < 256; value++) {
3636 if (!isWORDCHAR_L1(value)) {
3637 ANYOF_BITMAP_CLEAR(data->start_class, value);
3638 }
3639 }
3640 } else {
3641 for (value = 0; value < 256; value++) {
3642 if (!isALNUM(value)) {
3643 ANYOF_BITMAP_CLEAR(data->start_class, value);
3644 }
3645 }
3646 }
8aa23a47 3647 }
653099ff 3648 }
8aa23a47
YO
3649 else {
3650 if (data->start_class->flags & ANYOF_LOCALE)
3651 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
a12cf05f
KW
3652 else if (FLAGS(scan) & USE_UNI) {
3653 for (value = 0; value < 256; value++) {
3654 if (isWORDCHAR_L1(value)) {
3655 ANYOF_BITMAP_SET(data->start_class, value);
3656 }
3657 }
3658 } else {
3659 for (value = 0; value < 256; value++) {
3660 if (isALNUM(value)) {
3661 ANYOF_BITMAP_SET(data->start_class, value);
3662 }
3663 }
3664 }
8aa23a47
YO
3665 }
3666 break;
3667 case ALNUML:
3668 if (flags & SCF_DO_STCLASS_AND) {
3669 if (data->start_class->flags & ANYOF_LOCALE)
3670 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3671 }
3672 else {
3673 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3674 data->start_class->flags |= ANYOF_LOCALE;
3675 }
3676 break;
3677 case NALNUM:
3678 if (flags & SCF_DO_STCLASS_AND) {
3679 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3680 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
a12cf05f
KW
3681 if (FLAGS(scan) & USE_UNI) {
3682 for (value = 0; value < 256; value++) {
3683 if (isWORDCHAR_L1(value)) {
3684 ANYOF_BITMAP_CLEAR(data->start_class, value);
3685 }
3686 }
3687 } else {
3688 for (value = 0; value < 256; value++) {
3689 if (isALNUM(value)) {
3690 ANYOF_BITMAP_CLEAR(data->start_class, value);
3691 }
3692 }
3693 }
653099ff
GS
3694 }
3695 }
8aa23a47
YO
3696 else {
3697 if (data->start_class->flags & ANYOF_LOCALE)
3698 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3699 else {
3700 for (value = 0; value < 256; value++)
3701 if (!isALNUM(value))
b9a59e08 3702 ANYOF_BITMAP_SET(data->start_class, value);
8aa23a47 3703 }
653099ff 3704 }
8aa23a47
YO
3705 break;
3706 case NALNUML:
3707 if (flags & SCF_DO_STCLASS_AND) {
3708 if (data->start_class->flags & ANYOF_LOCALE)
3709 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
653099ff 3710 }
8aa23a47
YO
3711 else {
3712 data->start_class->flags |= ANYOF_LOCALE;
3713 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3714 }
3715 break;
3716 case SPACE:
3717 if (flags & SCF_DO_STCLASS_AND) {
3718 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3719 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
a12cf05f
KW
3720 if (FLAGS(scan) & USE_UNI) {