This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
API spelling patch, by Jerry D. Hedden
[perl5.git] / regcomp.c
CommitLineData
a0d0e21e
LW
1/* regcomp.c
2 */
3
4/*
5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
6 */
7
61296642
DM
8/* This file contains functions for compiling a regular expression. See
9 * also regexec.c which funnily enough, contains functions for executing
166f8a29 10 * a regular expression.
e4a054ea
DM
11 *
12 * This file is also copied at build time to ext/re/re_comp.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM
16 */
17
a687059c
LW
18/* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
20 */
21
22/* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
25 */
26
e50aee73
AD
27/* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
30*/
31
b9d5759e 32#ifdef PERL_EXT_RE_BUILD
54df2634 33#include "re_top.h"
b81d288d 34#endif
56953603 35
a687059c 36/*
e50aee73 37 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
38 *
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
41 *
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
45 *
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
48 * from defects in it.
49 *
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
52 *
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
55 *
56 *
57 **** Alterations to Henry's code are...
58 ****
4bb101f2 59 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
fdf8c088 60 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 by Larry Wall and others
a687059c 61 ****
9ef589d8
LW
62 **** You may distribute under the terms of either the GNU General Public
63 **** License or the Artistic License, as specified in the README file.
64
a687059c
LW
65 *
66 * Beware that some of this code is subtly aware of the way operator
67 * precedence is structured in regular expressions. Serious changes in
68 * regular-expression syntax might require a total rethink.
69 */
70#include "EXTERN.h"
864dbfa3 71#define PERL_IN_REGCOMP_C
a687059c 72#include "perl.h"
d06ea78c 73
acfe0abc 74#ifndef PERL_IN_XSUB_RE
d06ea78c
GS
75# include "INTERN.h"
76#endif
c277df42
IZ
77
78#define REG_COMP_C
54df2634
NC
79#ifdef PERL_IN_XSUB_RE
80# include "re_comp.h"
81#else
82# include "regcomp.h"
83#endif
a687059c 84
d4cce5f1 85#ifdef op
11343788 86#undef op
d4cce5f1 87#endif /* op */
11343788 88
fe14fcc3 89#ifdef MSDOS
7e4e8c89 90# if defined(BUGGY_MSC6)
fe14fcc3 91 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
7e4e8c89 92# pragma optimize("a",off)
fe14fcc3 93 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
7e4e8c89
NC
94# pragma optimize("w",on )
95# endif /* BUGGY_MSC6 */
fe14fcc3
LW
96#endif /* MSDOS */
97
a687059c
LW
98#ifndef STATIC
99#define STATIC static
100#endif
101
830247a4 102typedef struct RExC_state_t {
e2509266 103 U32 flags; /* are we folding, multilining? */
830247a4 104 char *precomp; /* uncompiled string. */
f8fc2ecf
YO
105 regexp *rx; /* perl core regexp structure */
106 regexp_internal *rxi; /* internal data for regexp object pprivate field */
fac92740 107 char *start; /* Start of input for compile */
830247a4
IZ
108 char *end; /* End of input for compile */
109 char *parse; /* Input-scan pointer. */
110 I32 whilem_seen; /* number of WHILEM in this expr */
fac92740 111 regnode *emit_start; /* Start of emitted-code area */
3b57cd43 112 regnode *emit_bound; /* First regnode outside of the allocated space */
ffc61ed2 113 regnode *emit; /* Code-emit pointer; &regdummy = don't = compiling */
830247a4
IZ
114 I32 naughty; /* How bad is this pattern? */
115 I32 sawback; /* Did we see \1, ...? */
116 U32 seen;
117 I32 size; /* Code size. */
c74340f9
YO
118 I32 npar; /* Capture buffer count, (OPEN). */
119 I32 cpar; /* Capture buffer count, (CLOSE). */
e2e6a0f1 120 I32 nestroot; /* root parens we are in - used by accept */
830247a4
IZ
121 I32 extralen;
122 I32 seen_zerolen;
123 I32 seen_evals;
40d049e4
YO
124 regnode **open_parens; /* pointers to open parens */
125 regnode **close_parens; /* pointers to close parens */
126 regnode *opend; /* END node in program */
02daf0ab
YO
127 I32 utf8; /* whether the pattern is utf8 or not */
128 I32 orig_utf8; /* whether the pattern was originally in utf8 */
129 /* XXX use this for future optimisation of case
130 * where pattern must be upgraded to utf8. */
6bda09f9 131 HV *charnames; /* cache of named sequences */
81714fb9 132 HV *paren_names; /* Paren names */
1f1031fe 133
40d049e4
YO
134 regnode **recurse; /* Recurse regops */
135 I32 recurse_count; /* Number of recurse regops */
830247a4
IZ
136#if ADD_TO_REGEXEC
137 char *starttry; /* -Dr: where regtry was called. */
138#define RExC_starttry (pRExC_state->starttry)
139#endif
3dab1dad 140#ifdef DEBUGGING
be8e71aa 141 const char *lastparse;
3dab1dad 142 I32 lastnum;
1f1031fe 143 AV *paren_name_list; /* idx -> name */
3dab1dad
YO
144#define RExC_lastparse (pRExC_state->lastparse)
145#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 146#define RExC_paren_name_list (pRExC_state->paren_name_list)
3dab1dad 147#endif
830247a4
IZ
148} RExC_state_t;
149
e2509266 150#define RExC_flags (pRExC_state->flags)
830247a4
IZ
151#define RExC_precomp (pRExC_state->precomp)
152#define RExC_rx (pRExC_state->rx)
f8fc2ecf 153#define RExC_rxi (pRExC_state->rxi)
fac92740 154#define RExC_start (pRExC_state->start)
830247a4
IZ
155#define RExC_end (pRExC_state->end)
156#define RExC_parse (pRExC_state->parse)
157#define RExC_whilem_seen (pRExC_state->whilem_seen)
7122b237
YO
158#ifdef RE_TRACK_PATTERN_OFFSETS
159#define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
160#endif
830247a4 161#define RExC_emit (pRExC_state->emit)
fac92740 162#define RExC_emit_start (pRExC_state->emit_start)
3b57cd43 163#define RExC_emit_bound (pRExC_state->emit_bound)
830247a4
IZ
164#define RExC_naughty (pRExC_state->naughty)
165#define RExC_sawback (pRExC_state->sawback)
166#define RExC_seen (pRExC_state->seen)
167#define RExC_size (pRExC_state->size)
168#define RExC_npar (pRExC_state->npar)
e2e6a0f1 169#define RExC_nestroot (pRExC_state->nestroot)
830247a4
IZ
170#define RExC_extralen (pRExC_state->extralen)
171#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
172#define RExC_seen_evals (pRExC_state->seen_evals)
1aa99e6b 173#define RExC_utf8 (pRExC_state->utf8)
02daf0ab 174#define RExC_orig_utf8 (pRExC_state->orig_utf8)
fc8cd66c 175#define RExC_charnames (pRExC_state->charnames)
40d049e4
YO
176#define RExC_open_parens (pRExC_state->open_parens)
177#define RExC_close_parens (pRExC_state->close_parens)
178#define RExC_opend (pRExC_state->opend)
81714fb9 179#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
180#define RExC_recurse (pRExC_state->recurse)
181#define RExC_recurse_count (pRExC_state->recurse_count)
830247a4 182
cde0cee5 183
a687059c
LW
184#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
185#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
186 ((*s) == '{' && regcurly(s)))
a687059c 187
35c8bce7
LW
188#ifdef SPSTART
189#undef SPSTART /* dratted cpp namespace... */
190#endif
a687059c
LW
191/*
192 * Flags to be passed up and down.
193 */
a687059c 194#define WORST 0 /* Worst case. */
a3b492c3
YO
195#define HASWIDTH 0x01 /* Known to match non-null strings. */
196#define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
197#define SPSTART 0x04 /* Starts with * or +. */
198#define TRYAGAIN 0x08 /* Weeded out a declaration. */
199#define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
a687059c 200
3dab1dad
YO
201#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
202
07be1b83
YO
203/* whether trie related optimizations are enabled */
204#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
205#define TRIE_STUDY_OPT
786e8c11 206#define FULL_TRIE_STUDY
07be1b83
YO
207#define TRIE_STCLASS
208#endif
1de06328
YO
209
210
40d049e4
YO
211
212#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
213#define PBITVAL(paren) (1 << ((paren) & 7))
214#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
215#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
216#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
217
218
1de06328
YO
219/* About scan_data_t.
220
221 During optimisation we recurse through the regexp program performing
222 various inplace (keyhole style) optimisations. In addition study_chunk
223 and scan_commit populate this data structure with information about
224 what strings MUST appear in the pattern. We look for the longest
225 string that must appear for at a fixed location, and we look for the
226 longest string that may appear at a floating location. So for instance
227 in the pattern:
228
229 /FOO[xX]A.*B[xX]BAR/
230
231 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
232 strings (because they follow a .* construct). study_chunk will identify
233 both FOO and BAR as being the longest fixed and floating strings respectively.
234
235 The strings can be composites, for instance
236
237 /(f)(o)(o)/
238
239 will result in a composite fixed substring 'foo'.
240
241 For each string some basic information is maintained:
242
243 - offset or min_offset
244 This is the position the string must appear at, or not before.
245 It also implicitly (when combined with minlenp) tells us how many
246 character must match before the string we are searching.
247 Likewise when combined with minlenp and the length of the string
248 tells us how many characters must appear after the string we have
249 found.
250
251 - max_offset
252 Only used for floating strings. This is the rightmost point that
253 the string can appear at. Ifset to I32 max it indicates that the
254 string can occur infinitely far to the right.
255
256 - minlenp
257 A pointer to the minimum length of the pattern that the string
258 was found inside. This is important as in the case of positive
259 lookahead or positive lookbehind we can have multiple patterns
260 involved. Consider
261
262 /(?=FOO).*F/
263
264 The minimum length of the pattern overall is 3, the minimum length
265 of the lookahead part is 3, but the minimum length of the part that
266 will actually match is 1. So 'FOO's minimum length is 3, but the
267 minimum length for the F is 1. This is important as the minimum length
268 is used to determine offsets in front of and behind the string being
269 looked for. Since strings can be composites this is the length of the
270 pattern at the time it was commited with a scan_commit. Note that
271 the length is calculated by study_chunk, so that the minimum lengths
272 are not known until the full pattern has been compiled, thus the
273 pointer to the value.
274
275 - lookbehind
276
277 In the case of lookbehind the string being searched for can be
278 offset past the start point of the final matching string.
279 If this value was just blithely removed from the min_offset it would
280 invalidate some of the calculations for how many chars must match
281 before or after (as they are derived from min_offset and minlen and
282 the length of the string being searched for).
283 When the final pattern is compiled and the data is moved from the
284 scan_data_t structure into the regexp structure the information
285 about lookbehind is factored in, with the information that would
286 have been lost precalculated in the end_shift field for the
287 associated string.
288
289 The fields pos_min and pos_delta are used to store the minimum offset
290 and the delta to the maximum offset at the current point in the pattern.
291
292*/
2c2d71f5
JH
293
294typedef struct scan_data_t {
1de06328
YO
295 /*I32 len_min; unused */
296 /*I32 len_delta; unused */
2c2d71f5
JH
297 I32 pos_min;
298 I32 pos_delta;
299 SV *last_found;
1de06328 300 I32 last_end; /* min value, <0 unless valid. */
2c2d71f5
JH
301 I32 last_start_min;
302 I32 last_start_max;
1de06328
YO
303 SV **longest; /* Either &l_fixed, or &l_float. */
304 SV *longest_fixed; /* longest fixed string found in pattern */
305 I32 offset_fixed; /* offset where it starts */
306 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
307 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
308 SV *longest_float; /* longest floating string found in pattern */
309 I32 offset_float_min; /* earliest point in string it can appear */
310 I32 offset_float_max; /* latest point in string it can appear */
311 I32 *minlen_float; /* pointer to the minlen relevent to the string */
312 I32 lookbehind_float; /* is the position of the string modified by LB */
2c2d71f5
JH
313 I32 flags;
314 I32 whilem_c;
cb434fcc 315 I32 *last_closep;
653099ff 316 struct regnode_charclass_class *start_class;
2c2d71f5
JH
317} scan_data_t;
318
a687059c 319/*
e50aee73 320 * Forward declarations for pregcomp()'s friends.
a687059c 321 */
a0d0e21e 322
27da23d5 323static const scan_data_t zero_scan_data =
1de06328 324 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
c277df42
IZ
325
326#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
07be1b83
YO
327#define SF_BEFORE_SEOL 0x0001
328#define SF_BEFORE_MEOL 0x0002
c277df42
IZ
329#define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
330#define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
331
09b7f37c
CB
332#ifdef NO_UNARY_PLUS
333# define SF_FIX_SHIFT_EOL (0+2)
334# define SF_FL_SHIFT_EOL (0+4)
335#else
336# define SF_FIX_SHIFT_EOL (+2)
337# define SF_FL_SHIFT_EOL (+4)
338#endif
c277df42
IZ
339
340#define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
341#define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
342
343#define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
344#define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
07be1b83
YO
345#define SF_IS_INF 0x0040
346#define SF_HAS_PAR 0x0080
347#define SF_IN_PAR 0x0100
348#define SF_HAS_EVAL 0x0200
349#define SCF_DO_SUBSTR 0x0400
653099ff
GS
350#define SCF_DO_STCLASS_AND 0x0800
351#define SCF_DO_STCLASS_OR 0x1000
352#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 353#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 354
786e8c11 355#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
e2e6a0f1 356#define SCF_SEEN_ACCEPT 0x8000
07be1b83 357
eb160463 358#define UTF (RExC_utf8 != 0)
bbe252da
YO
359#define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
360#define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
a0ed51b3 361
ffc61ed2 362#define OOB_UNICODE 12345678
93733859 363#define OOB_NAMEDCLASS -1
b8c5462f 364
a0ed51b3
LW
365#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
366#define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
367
8615cb43 368
b45f050a
JF
369/* length of regex to show in messages that don't mark a position within */
370#define RegexLengthToShowInErrorMessages 127
371
372/*
373 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
374 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
375 * op/pragma/warn/regcomp.
376 */
7253e4e3
RK
377#define MARKER1 "<-- HERE" /* marker as it appears in the description */
378#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 379
7253e4e3 380#define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
b45f050a
JF
381
382/*
383 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
384 * arg. Show regex, up to a maximum length. If it's too long, chop and add
385 * "...".
386 */
58e23c8d 387#define _FAIL(code) STMT_START { \
bfed75c6 388 const char *ellipses = ""; \
ccb2c380
MP
389 IV len = RExC_end - RExC_precomp; \
390 \
391 if (!SIZE_ONLY) \
392 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
393 if (len > RegexLengthToShowInErrorMessages) { \
394 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
395 len = RegexLengthToShowInErrorMessages - 10; \
396 ellipses = "..."; \
397 } \
58e23c8d 398 code; \
ccb2c380 399} STMT_END
8615cb43 400
58e23c8d
YO
401#define FAIL(msg) _FAIL( \
402 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
403 msg, (int)len, RExC_precomp, ellipses))
404
405#define FAIL2(msg,arg) _FAIL( \
406 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
407 arg, (int)len, RExC_precomp, ellipses))
408
b45f050a 409/*
b45f050a
JF
410 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
411 */
ccb2c380 412#define Simple_vFAIL(m) STMT_START { \
a28509cc 413 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
414 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
415 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
416} STMT_END
b45f050a
JF
417
418/*
419 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
420 */
ccb2c380
MP
421#define vFAIL(m) STMT_START { \
422 if (!SIZE_ONLY) \
423 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
424 Simple_vFAIL(m); \
425} STMT_END
b45f050a
JF
426
427/*
428 * Like Simple_vFAIL(), but accepts two arguments.
429 */
ccb2c380 430#define Simple_vFAIL2(m,a1) STMT_START { \
a28509cc 431 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
432 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
433 (int)offset, RExC_precomp, RExC_precomp + offset); \
434} STMT_END
b45f050a
JF
435
436/*
437 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
438 */
ccb2c380
MP
439#define vFAIL2(m,a1) STMT_START { \
440 if (!SIZE_ONLY) \
441 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
442 Simple_vFAIL2(m, a1); \
443} STMT_END
b45f050a
JF
444
445
446/*
447 * Like Simple_vFAIL(), but accepts three arguments.
448 */
ccb2c380 449#define Simple_vFAIL3(m, a1, a2) STMT_START { \
a28509cc 450 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
451 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
452 (int)offset, RExC_precomp, RExC_precomp + offset); \
453} STMT_END
b45f050a
JF
454
455/*
456 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
457 */
ccb2c380
MP
458#define vFAIL3(m,a1,a2) STMT_START { \
459 if (!SIZE_ONLY) \
460 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
461 Simple_vFAIL3(m, a1, a2); \
462} STMT_END
b45f050a
JF
463
464/*
465 * Like Simple_vFAIL(), but accepts four arguments.
466 */
ccb2c380 467#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
a28509cc 468 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
469 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
470 (int)offset, RExC_precomp, RExC_precomp + offset); \
471} STMT_END
b45f050a 472
ccb2c380 473#define vWARN(loc,m) STMT_START { \
a28509cc 474 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
475 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
476 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
477} STMT_END
478
479#define vWARNdep(loc,m) STMT_START { \
a28509cc 480 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
481 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
482 "%s" REPORT_LOCATION, \
483 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
484} STMT_END
485
486
487#define vWARN2(loc, m, a1) STMT_START { \
a28509cc 488 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
489 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
491} STMT_END
492
493#define vWARN3(loc, m, a1, a2) STMT_START { \
a28509cc 494 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
495 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
496 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
497} STMT_END
498
499#define vWARN4(loc, m, a1, a2, a3) STMT_START { \
a28509cc 500 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
501 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
502 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
503} STMT_END
504
505#define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
a28509cc 506 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
507 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
508 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
509} STMT_END
9d1d55b5 510
8615cb43 511
cd439c50 512/* Allow for side effects in s */
ccb2c380
MP
513#define REGC(c,s) STMT_START { \
514 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
515} STMT_END
cd439c50 516
fac92740
MJD
517/* Macros for recording node offsets. 20001227 mjd@plover.com
518 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
519 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
520 * Element 0 holds the number n.
07be1b83 521 * Position is 1 indexed.
fac92740 522 */
7122b237
YO
523#ifndef RE_TRACK_PATTERN_OFFSETS
524#define Set_Node_Offset_To_R(node,byte)
525#define Set_Node_Offset(node,byte)
526#define Set_Cur_Node_Offset
527#define Set_Node_Length_To_R(node,len)
528#define Set_Node_Length(node,len)
529#define Set_Node_Cur_Length(node)
530#define Node_Offset(n)
531#define Node_Length(n)
532#define Set_Node_Offset_Length(node,offset,len)
533#define ProgLen(ri) ri->u.proglen
534#define SetProgLen(ri,x) ri->u.proglen = x
535#else
536#define ProgLen(ri) ri->u.offsets[0]
537#define SetProgLen(ri,x) ri->u.offsets[0] = x
ccb2c380
MP
538#define Set_Node_Offset_To_R(node,byte) STMT_START { \
539 if (! SIZE_ONLY) { \
540 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
2a49f0f5 541 __LINE__, (int)(node), (int)(byte))); \
ccb2c380 542 if((node) < 0) { \
551405c4 543 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
ccb2c380
MP
544 } else { \
545 RExC_offsets[2*(node)-1] = (byte); \
546 } \
547 } \
548} STMT_END
549
550#define Set_Node_Offset(node,byte) \
551 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
552#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
553
554#define Set_Node_Length_To_R(node,len) STMT_START { \
555 if (! SIZE_ONLY) { \
556 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 557 __LINE__, (int)(node), (int)(len))); \
ccb2c380 558 if((node) < 0) { \
551405c4 559 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
ccb2c380
MP
560 } else { \
561 RExC_offsets[2*(node)] = (len); \
562 } \
563 } \
564} STMT_END
565
566#define Set_Node_Length(node,len) \
567 Set_Node_Length_To_R((node)-RExC_emit_start, len)
568#define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
569#define Set_Node_Cur_Length(node) \
570 Set_Node_Length(node, RExC_parse - parse_start)
fac92740
MJD
571
572/* Get offsets and lengths */
573#define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
574#define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
575
07be1b83
YO
576#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
577 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
578 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
579} STMT_END
7122b237 580#endif
07be1b83
YO
581
582#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
583#define EXPERIMENTAL_INPLACESCAN
7122b237 584#endif /*RE_TRACK_PATTERN_OFFSETS*/
07be1b83 585
304ee84b
YO
586#define DEBUG_STUDYDATA(str,data,depth) \
587DEBUG_OPTIMISE_MORE_r(if(data){ \
1de06328 588 PerlIO_printf(Perl_debug_log, \
304ee84b
YO
589 "%*s" str "Pos:%"IVdf"/%"IVdf \
590 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
1de06328
YO
591 (int)(depth)*2, "", \
592 (IV)((data)->pos_min), \
593 (IV)((data)->pos_delta), \
304ee84b 594 (UV)((data)->flags), \
1de06328 595 (IV)((data)->whilem_c), \
304ee84b
YO
596 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
597 is_inf ? "INF " : "" \
1de06328
YO
598 ); \
599 if ((data)->last_found) \
600 PerlIO_printf(Perl_debug_log, \
601 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
602 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
603 SvPVX_const((data)->last_found), \
604 (IV)((data)->last_end), \
605 (IV)((data)->last_start_min), \
606 (IV)((data)->last_start_max), \
607 ((data)->longest && \
608 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
609 SvPVX_const((data)->longest_fixed), \
610 (IV)((data)->offset_fixed), \
611 ((data)->longest && \
612 (data)->longest==&((data)->longest_float)) ? "*" : "", \
613 SvPVX_const((data)->longest_float), \
614 (IV)((data)->offset_float_min), \
615 (IV)((data)->offset_float_max) \
616 ); \
617 PerlIO_printf(Perl_debug_log,"\n"); \
618});
619
acfe0abc 620static void clear_re(pTHX_ void *r);
4327152a 621
653099ff 622/* Mark that we cannot extend a found fixed substring at this point.
786e8c11 623 Update the longest found anchored substring and the longest found
653099ff
GS
624 floating substrings if needed. */
625
4327152a 626STATIC void
304ee84b 627S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
c277df42 628{
e1ec3a88
AL
629 const STRLEN l = CHR_SVLEN(data->last_found);
630 const STRLEN old_l = CHR_SVLEN(*data->longest);
1de06328 631 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 632
c277df42 633 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
6b43b216 634 SvSetMagicSV(*data->longest, data->last_found);
c277df42
IZ
635 if (*data->longest == data->longest_fixed) {
636 data->offset_fixed = l ? data->last_start_min : data->pos_min;
637 if (data->flags & SF_BEFORE_EOL)
b81d288d 638 data->flags
c277df42
IZ
639 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
640 else
641 data->flags &= ~SF_FIX_BEFORE_EOL;
1de06328
YO
642 data->minlen_fixed=minlenp;
643 data->lookbehind_fixed=0;
a0ed51b3 644 }
304ee84b 645 else { /* *data->longest == data->longest_float */
c277df42 646 data->offset_float_min = l ? data->last_start_min : data->pos_min;
b81d288d
AB
647 data->offset_float_max = (l
648 ? data->last_start_max
c277df42 649 : data->pos_min + data->pos_delta);
304ee84b 650 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
9051bda5 651 data->offset_float_max = I32_MAX;
c277df42 652 if (data->flags & SF_BEFORE_EOL)
b81d288d 653 data->flags
c277df42
IZ
654 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
655 else
656 data->flags &= ~SF_FL_BEFORE_EOL;
1de06328
YO
657 data->minlen_float=minlenp;
658 data->lookbehind_float=0;
c277df42
IZ
659 }
660 }
661 SvCUR_set(data->last_found, 0);
0eda9292 662 {
a28509cc 663 SV * const sv = data->last_found;
097eb12c
AL
664 if (SvUTF8(sv) && SvMAGICAL(sv)) {
665 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
666 if (mg)
667 mg->mg_len = 0;
668 }
0eda9292 669 }
c277df42
IZ
670 data->last_end = -1;
671 data->flags &= ~SF_BEFORE_EOL;
bcdf7404 672 DEBUG_STUDYDATA("commit: ",data,0);
c277df42
IZ
673}
674
653099ff
GS
675/* Can match anything (initialization) */
676STATIC void
097eb12c 677S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 678{
653099ff 679 ANYOF_CLASS_ZERO(cl);
f8bef550 680 ANYOF_BITMAP_SETALL(cl);
1aa99e6b 681 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
653099ff
GS
682 if (LOC)
683 cl->flags |= ANYOF_LOCALE;
684}
685
686/* Can match anything (initialization) */
687STATIC int
5f66b61c 688S_cl_is_anything(const struct regnode_charclass_class *cl)
653099ff
GS
689{
690 int value;
691
aaa51d5e 692 for (value = 0; value <= ANYOF_MAX; value += 2)
653099ff
GS
693 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
694 return 1;
1aa99e6b
IH
695 if (!(cl->flags & ANYOF_UNICODE_ALL))
696 return 0;
10edeb5d 697 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
f8bef550 698 return 0;
653099ff
GS
699 return 1;
700}
701
702/* Can match anything (initialization) */
703STATIC void
097eb12c 704S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 705{
8ecf7187 706 Zero(cl, 1, struct regnode_charclass_class);
653099ff 707 cl->type = ANYOF;
830247a4 708 cl_anything(pRExC_state, cl);
653099ff
GS
709}
710
711STATIC void
097eb12c 712S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 713{
8ecf7187 714 Zero(cl, 1, struct regnode_charclass_class);
653099ff 715 cl->type = ANYOF;
830247a4 716 cl_anything(pRExC_state, cl);
653099ff
GS
717 if (LOC)
718 cl->flags |= ANYOF_LOCALE;
719}
720
721/* 'And' a given class with another one. Can create false positives */
722/* We assume that cl is not inverted */
723STATIC void
5f66b61c 724S_cl_and(struct regnode_charclass_class *cl,
a28509cc 725 const struct regnode_charclass_class *and_with)
653099ff 726{
40d049e4
YO
727
728 assert(and_with->type == ANYOF);
653099ff
GS
729 if (!(and_with->flags & ANYOF_CLASS)
730 && !(cl->flags & ANYOF_CLASS)
731 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
732 && !(and_with->flags & ANYOF_FOLD)
733 && !(cl->flags & ANYOF_FOLD)) {
734 int i;
735
736 if (and_with->flags & ANYOF_INVERT)
737 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
738 cl->bitmap[i] &= ~and_with->bitmap[i];
739 else
740 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
741 cl->bitmap[i] &= and_with->bitmap[i];
742 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
743 if (!(and_with->flags & ANYOF_EOS))
744 cl->flags &= ~ANYOF_EOS;
1aa99e6b 745
14ebb1a2
JH
746 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
747 !(and_with->flags & ANYOF_INVERT)) {
1aa99e6b
IH
748 cl->flags &= ~ANYOF_UNICODE_ALL;
749 cl->flags |= ANYOF_UNICODE;
750 ARG_SET(cl, ARG(and_with));
751 }
14ebb1a2
JH
752 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
753 !(and_with->flags & ANYOF_INVERT))
1aa99e6b 754 cl->flags &= ~ANYOF_UNICODE_ALL;
14ebb1a2
JH
755 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
756 !(and_with->flags & ANYOF_INVERT))
1aa99e6b 757 cl->flags &= ~ANYOF_UNICODE;
653099ff
GS
758}
759
760/* 'OR' a given class with another one. Can create false positives */
761/* We assume that cl is not inverted */
762STATIC void
097eb12c 763S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
653099ff 764{
653099ff
GS
765 if (or_with->flags & ANYOF_INVERT) {
766 /* We do not use
767 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
768 * <= (B1 | !B2) | (CL1 | !CL2)
769 * which is wasteful if CL2 is small, but we ignore CL2:
770 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
771 * XXXX Can we handle case-fold? Unclear:
772 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
773 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
774 */
775 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
776 && !(or_with->flags & ANYOF_FOLD)
777 && !(cl->flags & ANYOF_FOLD) ) {
778 int i;
779
780 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
781 cl->bitmap[i] |= ~or_with->bitmap[i];
782 } /* XXXX: logic is complicated otherwise */
783 else {
830247a4 784 cl_anything(pRExC_state, cl);
653099ff
GS
785 }
786 } else {
787 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
788 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
b81d288d 789 && (!(or_with->flags & ANYOF_FOLD)
653099ff
GS
790 || (cl->flags & ANYOF_FOLD)) ) {
791 int i;
792
793 /* OR char bitmap and class bitmap separately */
794 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
795 cl->bitmap[i] |= or_with->bitmap[i];
796 if (or_with->flags & ANYOF_CLASS) {
797 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
798 cl->classflags[i] |= or_with->classflags[i];
799 cl->flags |= ANYOF_CLASS;
800 }
801 }
802 else { /* XXXX: logic is complicated, leave it along for a moment. */
830247a4 803 cl_anything(pRExC_state, cl);
653099ff
GS
804 }
805 }
806 if (or_with->flags & ANYOF_EOS)
807 cl->flags |= ANYOF_EOS;
1aa99e6b
IH
808
809 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
810 ARG(cl) != ARG(or_with)) {
811 cl->flags |= ANYOF_UNICODE_ALL;
812 cl->flags &= ~ANYOF_UNICODE;
813 }
814 if (or_with->flags & ANYOF_UNICODE_ALL) {
815 cl->flags |= ANYOF_UNICODE_ALL;
816 cl->flags &= ~ANYOF_UNICODE;
817 }
653099ff
GS
818}
819
a3621e74
YO
820#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
821#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
822#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
823#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
824
3dab1dad
YO
825
826#ifdef DEBUGGING
07be1b83 827/*
2b8b4781
NC
828 dump_trie(trie,widecharmap,revcharmap)
829 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
830 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
831
832 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
833 The _interim_ variants are used for debugging the interim
834 tables that are used to generate the final compressed
835 representation which is what dump_trie expects.
836
3dab1dad
YO
837 Part of the reason for their existance is to provide a form
838 of documentation as to how the different representations function.
07be1b83
YO
839
840*/
3dab1dad
YO
841
842/*
3dab1dad
YO
843 Dumps the final compressed table form of the trie to Perl_debug_log.
844 Used for debugging make_trie().
845*/
846
847STATIC void
2b8b4781
NC
848S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
849 AV *revcharmap, U32 depth)
3dab1dad
YO
850{
851 U32 state;
ab3bbdeb 852 SV *sv=sv_newmortal();
55eed653 853 int colwidth= widecharmap ? 6 : 4;
3dab1dad
YO
854 GET_RE_DEBUG_FLAGS_DECL;
855
ab3bbdeb 856
3dab1dad
YO
857 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
858 (int)depth * 2 + 2,"",
859 "Match","Base","Ofs" );
860
861 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 862 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 863 if ( tmp ) {
ab3bbdeb
YO
864 PerlIO_printf( Perl_debug_log, "%*s",
865 colwidth,
ddc5bc0f 866 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
867 PL_colors[0], PL_colors[1],
868 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
869 PERL_PV_ESCAPE_FIRSTCHAR
870 )
871 );
3dab1dad
YO
872 }
873 }
874 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
875 (int)depth * 2 + 2,"");
876
877 for( state = 0 ; state < trie->uniquecharcount ; state++ )
ab3bbdeb 878 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
3dab1dad
YO
879 PerlIO_printf( Perl_debug_log, "\n");
880
1e2e3d02 881 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 882 const U32 base = trie->states[ state ].trans.base;
3dab1dad
YO
883
884 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
885
886 if ( trie->states[ state ].wordnum ) {
887 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
888 } else {
889 PerlIO_printf( Perl_debug_log, "%6s", "" );
890 }
891
892 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
893
894 if ( base ) {
895 U32 ofs = 0;
896
897 while( ( base + ofs < trie->uniquecharcount ) ||
898 ( base + ofs - trie->uniquecharcount < trie->lasttrans
899 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
900 ofs++;
901
902 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
903
904 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
905 if ( ( base + ofs >= trie->uniquecharcount ) &&
906 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
907 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
908 {
ab3bbdeb
YO
909 PerlIO_printf( Perl_debug_log, "%*"UVXf,
910 colwidth,
3dab1dad
YO
911 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
912 } else {
ab3bbdeb 913 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
3dab1dad
YO
914 }
915 }
916
917 PerlIO_printf( Perl_debug_log, "]");
918
919 }
920 PerlIO_printf( Perl_debug_log, "\n" );
921 }
922}
923/*
3dab1dad
YO
924 Dumps a fully constructed but uncompressed trie in list form.
925 List tries normally only are used for construction when the number of
926 possible chars (trie->uniquecharcount) is very high.
927 Used for debugging make_trie().
928*/
929STATIC void
55eed653 930S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
931 HV *widecharmap, AV *revcharmap, U32 next_alloc,
932 U32 depth)
3dab1dad
YO
933{
934 U32 state;
ab3bbdeb 935 SV *sv=sv_newmortal();
55eed653 936 int colwidth= widecharmap ? 6 : 4;
3dab1dad
YO
937 GET_RE_DEBUG_FLAGS_DECL;
938 /* print out the table precompression. */
ab3bbdeb
YO
939 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
940 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
941 "------:-----+-----------------\n" );
3dab1dad
YO
942
943 for( state=1 ; state < next_alloc ; state ++ ) {
944 U16 charid;
945
ab3bbdeb 946 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
3dab1dad
YO
947 (int)depth * 2 + 2,"", (UV)state );
948 if ( ! trie->states[ state ].wordnum ) {
949 PerlIO_printf( Perl_debug_log, "%5s| ","");
950 } else {
951 PerlIO_printf( Perl_debug_log, "W%4x| ",
952 trie->states[ state ].wordnum
953 );
954 }
955 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
2b8b4781 956 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
ab3bbdeb
YO
957 if ( tmp ) {
958 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
959 colwidth,
ddc5bc0f 960 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
961 PL_colors[0], PL_colors[1],
962 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
963 PERL_PV_ESCAPE_FIRSTCHAR
964 ) ,
1e2e3d02
YO
965 TRIE_LIST_ITEM(state,charid).forid,
966 (UV)TRIE_LIST_ITEM(state,charid).newstate
967 );
968 if (!(charid % 10))
664e119d
RGS
969 PerlIO_printf(Perl_debug_log, "\n%*s| ",
970 (int)((depth * 2) + 14), "");
1e2e3d02 971 }
ab3bbdeb
YO
972 }
973 PerlIO_printf( Perl_debug_log, "\n");
3dab1dad
YO
974 }
975}
976
977/*
3dab1dad
YO
978 Dumps a fully constructed but uncompressed trie in table form.
979 This is the normal DFA style state transition table, with a few
980 twists to facilitate compression later.
981 Used for debugging make_trie().
982*/
983STATIC void
55eed653 984S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
985 HV *widecharmap, AV *revcharmap, U32 next_alloc,
986 U32 depth)
3dab1dad
YO
987{
988 U32 state;
989 U16 charid;
ab3bbdeb 990 SV *sv=sv_newmortal();
55eed653 991 int colwidth= widecharmap ? 6 : 4;
3dab1dad
YO
992 GET_RE_DEBUG_FLAGS_DECL;
993
994 /*
995 print out the table precompression so that we can do a visual check
996 that they are identical.
997 */
998
999 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1000
1001 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 1002 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 1003 if ( tmp ) {
ab3bbdeb
YO
1004 PerlIO_printf( Perl_debug_log, "%*s",
1005 colwidth,
ddc5bc0f 1006 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
1007 PL_colors[0], PL_colors[1],
1008 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1009 PERL_PV_ESCAPE_FIRSTCHAR
1010 )
1011 );
3dab1dad
YO
1012 }
1013 }
1014
1015 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1016
1017 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb 1018 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
3dab1dad
YO
1019 }
1020
1021 PerlIO_printf( Perl_debug_log, "\n" );
1022
1023 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1024
1025 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1026 (int)depth * 2 + 2,"",
1027 (UV)TRIE_NODENUM( state ) );
1028
1029 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
1030 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1031 if (v)
1032 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1033 else
1034 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
3dab1dad
YO
1035 }
1036 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1037 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1038 } else {
1039 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1040 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1041 }
1042 }
07be1b83 1043}
3dab1dad
YO
1044
1045#endif
1046
786e8c11
YO
1047/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1048 startbranch: the first branch in the whole branch sequence
1049 first : start branch of sequence of branch-exact nodes.
1050 May be the same as startbranch
1051 last : Thing following the last branch.
1052 May be the same as tail.
1053 tail : item following the branch sequence
1054 count : words in the sequence
1055 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1056 depth : indent depth
3dab1dad 1057
786e8c11 1058Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 1059
786e8c11
YO
1060A trie is an N'ary tree where the branches are determined by digital
1061decomposition of the key. IE, at the root node you look up the 1st character and
1062follow that branch repeat until you find the end of the branches. Nodes can be
1063marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 1064
786e8c11 1065 /he|she|his|hers/
72f13be8 1066
786e8c11
YO
1067would convert into the following structure. Numbers represent states, letters
1068following numbers represent valid transitions on the letter from that state, if
1069the number is in square brackets it represents an accepting state, otherwise it
1070will be in parenthesis.
07be1b83 1071
786e8c11
YO
1072 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1073 | |
1074 | (2)
1075 | |
1076 (1) +-i->(6)-+-s->[7]
1077 |
1078 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 1079
786e8c11
YO
1080 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1081
1082This shows that when matching against the string 'hers' we will begin at state 1
1083read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1084then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1085is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1086single traverse. We store a mapping from accepting to state to which word was
1087matched, and then when we have multiple possibilities we try to complete the
1088rest of the regex in the order in which they occured in the alternation.
1089
1090The only prior NFA like behaviour that would be changed by the TRIE support is
1091the silent ignoring of duplicate alternations which are of the form:
1092
1093 / (DUPE|DUPE) X? (?{ ... }) Y /x
1094
1095Thus EVAL blocks follwing a trie may be called a different number of times with
1096and without the optimisation. With the optimisations dupes will be silently
1097ignored. This inconsistant behaviour of EVAL type nodes is well established as
1098the following demonstrates:
1099
1100 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1101
1102which prints out 'word' three times, but
1103
1104 'words'=~/(word|word|word)(?{ print $1 })S/
1105
1106which doesnt print it out at all. This is due to other optimisations kicking in.
1107
1108Example of what happens on a structural level:
1109
1110The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1111
1112 1: CURLYM[1] {1,32767}(18)
1113 5: BRANCH(8)
1114 6: EXACT <ac>(16)
1115 8: BRANCH(11)
1116 9: EXACT <ad>(16)
1117 11: BRANCH(14)
1118 12: EXACT <ab>(16)
1119 16: SUCCEED(0)
1120 17: NOTHING(18)
1121 18: END(0)
1122
1123This would be optimizable with startbranch=5, first=5, last=16, tail=16
1124and should turn into:
1125
1126 1: CURLYM[1] {1,32767}(18)
1127 5: TRIE(16)
1128 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1129 <ac>
1130 <ad>
1131 <ab>
1132 16: SUCCEED(0)
1133 17: NOTHING(18)
1134 18: END(0)
1135
1136Cases where tail != last would be like /(?foo|bar)baz/:
1137
1138 1: BRANCH(4)
1139 2: EXACT <foo>(8)
1140 4: BRANCH(7)
1141 5: EXACT <bar>(8)
1142 7: TAIL(8)
1143 8: EXACT <baz>(10)
1144 10: END(0)
1145
1146which would be optimizable with startbranch=1, first=1, last=7, tail=8
1147and would end up looking like:
1148
1149 1: TRIE(8)
1150 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1151 <foo>
1152 <bar>
1153 7: TAIL(8)
1154 8: EXACT <baz>(10)
1155 10: END(0)
1156
1157 d = uvuni_to_utf8_flags(d, uv, 0);
1158
1159is the recommended Unicode-aware way of saying
1160
1161 *(d++) = uv;
1162*/
1163
1e2e3d02 1164#define TRIE_STORE_REVCHAR \
786e8c11 1165 STMT_START { \
1e2e3d02 1166 SV *tmp = newSVpvs(""); \
786e8c11 1167 if (UTF) SvUTF8_on(tmp); \
1e2e3d02 1168 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
2b8b4781 1169 av_push( revcharmap, tmp ); \
786e8c11
YO
1170 } STMT_END
1171
1172#define TRIE_READ_CHAR STMT_START { \
1173 wordlen++; \
1174 if ( UTF ) { \
1175 if ( folder ) { \
1176 if ( foldlen > 0 ) { \
1177 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1178 foldlen -= len; \
1179 scan += len; \
1180 len = 0; \
1181 } else { \
1182 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1183 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1184 foldlen -= UNISKIP( uvc ); \
1185 scan = foldbuf + UNISKIP( uvc ); \
1186 } \
1187 } else { \
1188 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1189 } \
1190 } else { \
1191 uvc = (U32)*uc; \
1192 len = 1; \
1193 } \
1194} STMT_END
1195
1196
1197
1198#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1199 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
f9003953
NC
1200 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1201 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
786e8c11
YO
1202 } \
1203 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1204 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1205 TRIE_LIST_CUR( state )++; \
1206} STMT_END
07be1b83 1207
786e8c11
YO
1208#define TRIE_LIST_NEW(state) STMT_START { \
1209 Newxz( trie->states[ state ].trans.list, \
1210 4, reg_trie_trans_le ); \
1211 TRIE_LIST_CUR( state ) = 1; \
1212 TRIE_LIST_LEN( state ) = 4; \
1213} STMT_END
07be1b83 1214
786e8c11
YO
1215#define TRIE_HANDLE_WORD(state) STMT_START { \
1216 U16 dupe= trie->states[ state ].wordnum; \
1217 regnode * const noper_next = regnext( noper ); \
1218 \
1219 if (trie->wordlen) \
1220 trie->wordlen[ curword ] = wordlen; \
1221 DEBUG_r({ \
1222 /* store the word for dumping */ \
1223 SV* tmp; \
1224 if (OP(noper) != NOTHING) \
1225 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1226 else \
1227 tmp = newSVpvn( "", 0 ); \
1228 if ( UTF ) SvUTF8_on( tmp ); \
2b8b4781 1229 av_push( trie_words, tmp ); \
786e8c11
YO
1230 }); \
1231 \
1232 curword++; \
1233 \
1234 if ( noper_next < tail ) { \
1235 if (!trie->jump) \
c944940b 1236 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
7f69552c 1237 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
1238 if (!jumper) \
1239 jumper = noper_next; \
1240 if (!nextbranch) \
1241 nextbranch= regnext(cur); \
1242 } \
1243 \
1244 if ( dupe ) { \
1245 /* So it's a dupe. This means we need to maintain a */\
1246 /* linked-list from the first to the next. */\
1247 /* we only allocate the nextword buffer when there */\
1248 /* a dupe, so first time we have to do the allocation */\
1249 if (!trie->nextword) \
c944940b 1250 trie->nextword = (U16 *) \
446bd890 1251 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
786e8c11
YO
1252 while ( trie->nextword[dupe] ) \
1253 dupe= trie->nextword[dupe]; \
1254 trie->nextword[dupe]= curword; \
1255 } else { \
1256 /* we haven't inserted this word yet. */ \
1257 trie->states[ state ].wordnum = curword; \
1258 } \
1259} STMT_END
07be1b83 1260
3dab1dad 1261
786e8c11
YO
1262#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1263 ( ( base + charid >= ucharcount \
1264 && base + charid < ubound \
1265 && state == trie->trans[ base - ucharcount + charid ].check \
1266 && trie->trans[ base - ucharcount + charid ].next ) \
1267 ? trie->trans[ base - ucharcount + charid ].next \
1268 : ( state==1 ? special : 0 ) \
1269 )
3dab1dad 1270
786e8c11
YO
1271#define MADE_TRIE 1
1272#define MADE_JUMP_TRIE 2
1273#define MADE_EXACT_TRIE 4
3dab1dad 1274
a3621e74 1275STATIC I32
786e8c11 1276S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
a3621e74 1277{
27da23d5 1278 dVAR;
a3621e74
YO
1279 /* first pass, loop through and scan words */
1280 reg_trie_data *trie;
55eed653 1281 HV *widecharmap = NULL;
2b8b4781 1282 AV *revcharmap = newAV();
a3621e74 1283 regnode *cur;
9f7f3913 1284 const U32 uniflags = UTF8_ALLOW_DEFAULT;
a3621e74
YO
1285 STRLEN len = 0;
1286 UV uvc = 0;
1287 U16 curword = 0;
1288 U32 next_alloc = 0;
786e8c11
YO
1289 regnode *jumper = NULL;
1290 regnode *nextbranch = NULL;
7f69552c 1291 regnode *convert = NULL;
a3621e74 1292 /* we just use folder as a flag in utf8 */
e1ec3a88 1293 const U8 * const folder = ( flags == EXACTF
a3621e74
YO
1294 ? PL_fold
1295 : ( flags == EXACTFL
1296 ? PL_fold_locale
1297 : NULL
1298 )
1299 );
1300
2b8b4781
NC
1301#ifdef DEBUGGING
1302 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1303 AV *trie_words = NULL;
1304 /* along with revcharmap, this only used during construction but both are
1305 * useful during debugging so we store them in the struct when debugging.
8e11feef 1306 */
2b8b4781
NC
1307#else
1308 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
3dab1dad 1309 STRLEN trie_charcount=0;
3dab1dad 1310#endif
2b8b4781 1311 SV *re_trie_maxbuff;
a3621e74 1312 GET_RE_DEBUG_FLAGS_DECL;
72f13be8
YO
1313#ifndef DEBUGGING
1314 PERL_UNUSED_ARG(depth);
1315#endif
a3621e74 1316
c944940b 1317 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 1318 trie->refcount = 1;
3dab1dad 1319 trie->startstate = 1;
786e8c11 1320 trie->wordcount = word_count;
f8fc2ecf 1321 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 1322 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
3dab1dad 1323 if (!(UTF && folder))
c944940b 1324 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
a3621e74 1325 DEBUG_r({
2b8b4781 1326 trie_words = newAV();
a3621e74 1327 });
a3621e74 1328
0111c4fd 1329 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
a3621e74 1330 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 1331 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 1332 }
3dab1dad
YO
1333 DEBUG_OPTIMISE_r({
1334 PerlIO_printf( Perl_debug_log,
786e8c11 1335 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
3dab1dad
YO
1336 (int)depth * 2 + 2, "",
1337 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
786e8c11 1338 REG_NODE_NUM(last), REG_NODE_NUM(tail),
85c3142d 1339 (int)depth);
3dab1dad 1340 });
7f69552c
YO
1341
1342 /* Find the node we are going to overwrite */
1343 if ( first == startbranch && OP( last ) != BRANCH ) {
1344 /* whole branch chain */
1345 convert = first;
1346 } else {
1347 /* branch sub-chain */
1348 convert = NEXTOPER( first );
1349 }
1350
a3621e74
YO
1351 /* -- First loop and Setup --
1352
1353 We first traverse the branches and scan each word to determine if it
1354 contains widechars, and how many unique chars there are, this is
1355 important as we have to build a table with at least as many columns as we
1356 have unique chars.
1357
1358 We use an array of integers to represent the character codes 0..255
38a44b82 1359 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
a3621e74
YO
1360 native representation of the character value as the key and IV's for the
1361 coded index.
1362
1363 *TODO* If we keep track of how many times each character is used we can
1364 remap the columns so that the table compression later on is more
1365 efficient in terms of memory by ensuring most common value is in the
1366 middle and the least common are on the outside. IMO this would be better
1367 than a most to least common mapping as theres a decent chance the most
1368 common letter will share a node with the least common, meaning the node
1369 will not be compressable. With a middle is most common approach the worst
1370 case is when we have the least common nodes twice.
1371
1372 */
1373
a3621e74 1374 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
c445ea15 1375 regnode * const noper = NEXTOPER( cur );
e1ec3a88 1376 const U8 *uc = (U8*)STRING( noper );
a28509cc 1377 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1378 STRLEN foldlen = 0;
1379 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2af232bd 1380 const U8 *scan = (U8*)NULL;
07be1b83 1381 U32 wordlen = 0; /* required init */
02daf0ab
YO
1382 STRLEN chars = 0;
1383 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
a3621e74 1384
3dab1dad
YO
1385 if (OP(noper) == NOTHING) {
1386 trie->minlen= 0;
1387 continue;
1388 }
02daf0ab
YO
1389 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1390 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1391 regardless of encoding */
1392
a3621e74 1393 for ( ; uc < e ; uc += len ) {
3dab1dad 1394 TRIE_CHARCOUNT(trie)++;
a3621e74 1395 TRIE_READ_CHAR;
3dab1dad 1396 chars++;
a3621e74
YO
1397 if ( uvc < 256 ) {
1398 if ( !trie->charmap[ uvc ] ) {
1399 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1400 if ( folder )
1401 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
3dab1dad 1402 TRIE_STORE_REVCHAR;
a3621e74 1403 }
02daf0ab
YO
1404 if ( set_bit ) {
1405 /* store the codepoint in the bitmap, and if its ascii
1406 also store its folded equivelent. */
1407 TRIE_BITMAP_SET(trie,uvc);
0921ee73
T
1408
1409 /* store the folded codepoint */
1410 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1411
1412 if ( !UTF ) {
1413 /* store first byte of utf8 representation of
1414 codepoints in the 127 < uvc < 256 range */
1415 if (127 < uvc && uvc < 192) {
1416 TRIE_BITMAP_SET(trie,194);
1417 } else if (191 < uvc ) {
1418 TRIE_BITMAP_SET(trie,195);
1419 /* && uvc < 256 -- we know uvc is < 256 already */
1420 }
1421 }
02daf0ab
YO
1422 set_bit = 0; /* We've done our bit :-) */
1423 }
a3621e74
YO
1424 } else {
1425 SV** svpp;
55eed653
NC
1426 if ( !widecharmap )
1427 widecharmap = newHV();
a3621e74 1428
55eed653 1429 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
1430
1431 if ( !svpp )
e4584336 1432 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
a3621e74
YO
1433
1434 if ( !SvTRUE( *svpp ) ) {
1435 sv_setiv( *svpp, ++trie->uniquecharcount );
3dab1dad 1436 TRIE_STORE_REVCHAR;
a3621e74
YO
1437 }
1438 }
1439 }
3dab1dad
YO
1440 if( cur == first ) {
1441 trie->minlen=chars;
1442 trie->maxlen=chars;
1443 } else if (chars < trie->minlen) {
1444 trie->minlen=chars;
1445 } else if (chars > trie->maxlen) {
1446 trie->maxlen=chars;
1447 }
1448
a3621e74
YO
1449 } /* end first pass */
1450 DEBUG_TRIE_COMPILE_r(
3dab1dad
YO
1451 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1452 (int)depth * 2 + 2,"",
55eed653 1453 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
1454 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1455 (int)trie->minlen, (int)trie->maxlen )
a3621e74 1456 );
c944940b 1457 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
a3621e74
YO
1458
1459 /*
1460 We now know what we are dealing with in terms of unique chars and
1461 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
1462 representation using a flat table will take. If it's over a reasonable
1463 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
1464 conservative but potentially much slower representation using an array
1465 of lists.
1466
1467 At the end we convert both representations into the same compressed
1468 form that will be used in regexec.c for matching with. The latter
1469 is a form that cannot be used to construct with but has memory
1470 properties similar to the list form and access properties similar
1471 to the table form making it both suitable for fast searches and
1472 small enough that its feasable to store for the duration of a program.
1473
1474 See the comment in the code where the compressed table is produced
1475 inplace from the flat tabe representation for an explanation of how
1476 the compression works.
1477
1478 */
1479
1480
3dab1dad 1481 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
a3621e74
YO
1482 /*
1483 Second Pass -- Array Of Lists Representation
1484
1485 Each state will be represented by a list of charid:state records
1486 (reg_trie_trans_le) the first such element holds the CUR and LEN
1487 points of the allocated array. (See defines above).
1488
1489 We build the initial structure using the lists, and then convert
1490 it into the compressed table form which allows faster lookups
1491 (but cant be modified once converted).
a3621e74
YO
1492 */
1493
a3621e74
YO
1494 STRLEN transcount = 1;
1495
1e2e3d02
YO
1496 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1497 "%*sCompiling trie using list compiler\n",
1498 (int)depth * 2 + 2, ""));
446bd890 1499
c944940b
JH
1500 trie->states = (reg_trie_state *)
1501 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1502 sizeof(reg_trie_state) );
a3621e74
YO
1503 TRIE_LIST_NEW(1);
1504 next_alloc = 2;
1505
1506 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1507
c445ea15
AL
1508 regnode * const noper = NEXTOPER( cur );
1509 U8 *uc = (U8*)STRING( noper );
1510 const U8 * const e = uc + STR_LEN( noper );
1511 U32 state = 1; /* required init */
1512 U16 charid = 0; /* sanity init */
1513 U8 *scan = (U8*)NULL; /* sanity init */
1514 STRLEN foldlen = 0; /* required init */
07be1b83 1515 U32 wordlen = 0; /* required init */
c445ea15
AL
1516 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1517
3dab1dad 1518 if (OP(noper) != NOTHING) {
786e8c11 1519 for ( ; uc < e ; uc += len ) {
c445ea15 1520
786e8c11 1521 TRIE_READ_CHAR;
c445ea15 1522
786e8c11
YO
1523 if ( uvc < 256 ) {
1524 charid = trie->charmap[ uvc ];
c445ea15 1525 } else {
55eed653 1526 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11
YO
1527 if ( !svpp ) {
1528 charid = 0;
1529 } else {
1530 charid=(U16)SvIV( *svpp );
1531 }
c445ea15 1532 }
786e8c11
YO
1533 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1534 if ( charid ) {
a3621e74 1535
786e8c11
YO
1536 U16 check;
1537 U32 newstate = 0;
a3621e74 1538
786e8c11
YO
1539 charid--;
1540 if ( !trie->states[ state ].trans.list ) {
1541 TRIE_LIST_NEW( state );
c445ea15 1542 }
786e8c11
YO
1543 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1544 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1545 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1546 break;
1547 }
1548 }
1549 if ( ! newstate ) {
1550 newstate = next_alloc++;
1551 TRIE_LIST_PUSH( state, charid, newstate );
1552 transcount++;
1553 }
1554 state = newstate;
1555 } else {
1556 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
c445ea15 1557 }
a28509cc 1558 }
c445ea15 1559 }
3dab1dad 1560 TRIE_HANDLE_WORD(state);
a3621e74
YO
1561
1562 } /* end second pass */
1563
1e2e3d02
YO
1564 /* next alloc is the NEXT state to be allocated */
1565 trie->statecount = next_alloc;
c944940b
JH
1566 trie->states = (reg_trie_state *)
1567 PerlMemShared_realloc( trie->states,
1568 next_alloc
1569 * sizeof(reg_trie_state) );
a3621e74 1570
3dab1dad 1571 /* and now dump it out before we compress it */
2b8b4781
NC
1572 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1573 revcharmap, next_alloc,
1574 depth+1)
1e2e3d02 1575 );
a3621e74 1576
c944940b
JH
1577 trie->trans = (reg_trie_trans *)
1578 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
1579 {
1580 U32 state;
a3621e74
YO
1581 U32 tp = 0;
1582 U32 zp = 0;
1583
1584
1585 for( state=1 ; state < next_alloc ; state ++ ) {
1586 U32 base=0;
1587
1588 /*
1589 DEBUG_TRIE_COMPILE_MORE_r(
1590 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1591 );
1592 */
1593
1594 if (trie->states[state].trans.list) {
1595 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1596 U16 maxid=minid;
a28509cc 1597 U16 idx;
a3621e74
YO
1598
1599 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
1600 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1601 if ( forid < minid ) {
1602 minid=forid;
1603 } else if ( forid > maxid ) {
1604 maxid=forid;
1605 }
a3621e74
YO
1606 }
1607 if ( transcount < tp + maxid - minid + 1) {
1608 transcount *= 2;
c944940b
JH
1609 trie->trans = (reg_trie_trans *)
1610 PerlMemShared_realloc( trie->trans,
446bd890
NC
1611 transcount
1612 * sizeof(reg_trie_trans) );
a3621e74
YO
1613 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1614 }
1615 base = trie->uniquecharcount + tp - minid;
1616 if ( maxid == minid ) {
1617 U32 set = 0;
1618 for ( ; zp < tp ; zp++ ) {
1619 if ( ! trie->trans[ zp ].next ) {
1620 base = trie->uniquecharcount + zp - minid;
1621 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1622 trie->trans[ zp ].check = state;
1623 set = 1;
1624 break;
1625 }
1626 }
1627 if ( !set ) {
1628 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1629 trie->trans[ tp ].check = state;
1630 tp++;
1631 zp = tp;
1632 }
1633 } else {
1634 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15 1635 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
a3621e74
YO
1636 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1637 trie->trans[ tid ].check = state;
1638 }
1639 tp += ( maxid - minid + 1 );
1640 }
1641 Safefree(trie->states[ state ].trans.list);
1642 }
1643 /*
1644 DEBUG_TRIE_COMPILE_MORE_r(
1645 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1646 );
1647 */
1648 trie->states[ state ].trans.base=base;
1649 }
cc601c31 1650 trie->lasttrans = tp + 1;
a3621e74
YO
1651 }
1652 } else {
1653 /*
1654 Second Pass -- Flat Table Representation.
1655
1656 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1657 We know that we will need Charcount+1 trans at most to store the data
1658 (one row per char at worst case) So we preallocate both structures
1659 assuming worst case.
1660
1661 We then construct the trie using only the .next slots of the entry
1662 structs.
1663
1664 We use the .check field of the first entry of the node temporarily to
1665 make compression both faster and easier by keeping track of how many non
1666 zero fields are in the node.
1667
1668 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1669 transition.
1670
1671 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1672 number representing the first entry of the node, and state as a
1673 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1674 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1675 are 2 entrys per node. eg:
1676
1677 A B A B
1678 1. 2 4 1. 3 7
1679 2. 0 3 3. 0 5
1680 3. 0 0 5. 0 0
1681 4. 0 0 7. 0 0
1682
1683 The table is internally in the right hand, idx form. However as we also
1684 have to deal with the states array which is indexed by nodenum we have to
1685 use TRIE_NODENUM() to convert.
1686
1687 */
1e2e3d02
YO
1688 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1689 "%*sCompiling trie using table compiler\n",
1690 (int)depth * 2 + 2, ""));
3dab1dad 1691
c944940b
JH
1692 trie->trans = (reg_trie_trans *)
1693 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1694 * trie->uniquecharcount + 1,
1695 sizeof(reg_trie_trans) );
1696 trie->states = (reg_trie_state *)
1697 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1698 sizeof(reg_trie_state) );
a3621e74
YO
1699 next_alloc = trie->uniquecharcount + 1;
1700
3dab1dad 1701
a3621e74
YO
1702 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1703
c445ea15 1704 regnode * const noper = NEXTOPER( cur );
a28509cc
AL
1705 const U8 *uc = (U8*)STRING( noper );
1706 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1707
1708 U32 state = 1; /* required init */
1709
1710 U16 charid = 0; /* sanity init */
1711 U32 accept_state = 0; /* sanity init */
1712 U8 *scan = (U8*)NULL; /* sanity init */
1713
1714 STRLEN foldlen = 0; /* required init */
07be1b83 1715 U32 wordlen = 0; /* required init */
a3621e74
YO
1716 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1717
3dab1dad 1718 if ( OP(noper) != NOTHING ) {
786e8c11 1719 for ( ; uc < e ; uc += len ) {
a3621e74 1720
786e8c11 1721 TRIE_READ_CHAR;
a3621e74 1722
786e8c11
YO
1723 if ( uvc < 256 ) {
1724 charid = trie->charmap[ uvc ];
1725 } else {
55eed653 1726 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11 1727 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 1728 }
786e8c11
YO
1729 if ( charid ) {
1730 charid--;
1731 if ( !trie->trans[ state + charid ].next ) {
1732 trie->trans[ state + charid ].next = next_alloc;
1733 trie->trans[ state ].check++;
1734 next_alloc += trie->uniquecharcount;
1735 }
1736 state = trie->trans[ state + charid ].next;
1737 } else {
1738 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1739 }
1740 /* charid is now 0 if we dont know the char read, or nonzero if we do */
a3621e74 1741 }
a3621e74 1742 }
3dab1dad
YO
1743 accept_state = TRIE_NODENUM( state );
1744 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
1745
1746 } /* end second pass */
1747
3dab1dad 1748 /* and now dump it out before we compress it */
2b8b4781
NC
1749 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1750 revcharmap,
1751 next_alloc, depth+1));
a3621e74 1752
a3621e74
YO
1753 {
1754 /*
1755 * Inplace compress the table.*
1756
1757 For sparse data sets the table constructed by the trie algorithm will
1758 be mostly 0/FAIL transitions or to put it another way mostly empty.
1759 (Note that leaf nodes will not contain any transitions.)
1760
1761 This algorithm compresses the tables by eliminating most such
1762 transitions, at the cost of a modest bit of extra work during lookup:
1763
1764 - Each states[] entry contains a .base field which indicates the
1765 index in the state[] array wheres its transition data is stored.
1766
1767 - If .base is 0 there are no valid transitions from that node.
1768
1769 - If .base is nonzero then charid is added to it to find an entry in
1770 the trans array.
1771
1772 -If trans[states[state].base+charid].check!=state then the
1773 transition is taken to be a 0/Fail transition. Thus if there are fail
1774 transitions at the front of the node then the .base offset will point
1775 somewhere inside the previous nodes data (or maybe even into a node
1776 even earlier), but the .check field determines if the transition is
1777 valid.
1778
786e8c11 1779 XXX - wrong maybe?
a3621e74
YO
1780 The following process inplace converts the table to the compressed
1781 table: We first do not compress the root node 1,and mark its all its
1782 .check pointers as 1 and set its .base pointer as 1 as well. This
1783 allows to do a DFA construction from the compressed table later, and
1784 ensures that any .base pointers we calculate later are greater than
1785 0.
1786
1787 - We set 'pos' to indicate the first entry of the second node.
1788
1789 - We then iterate over the columns of the node, finding the first and
1790 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1791 and set the .check pointers accordingly, and advance pos
1792 appropriately and repreat for the next node. Note that when we copy
1793 the next pointers we have to convert them from the original
1794 NODEIDX form to NODENUM form as the former is not valid post
1795 compression.
1796
1797 - If a node has no transitions used we mark its base as 0 and do not
1798 advance the pos pointer.
1799
1800 - If a node only has one transition we use a second pointer into the
1801 structure to fill in allocated fail transitions from other states.
1802 This pointer is independent of the main pointer and scans forward
1803 looking for null transitions that are allocated to a state. When it
1804 finds one it writes the single transition into the "hole". If the
786e8c11 1805 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
1806
1807 - Once compressed we can Renew/realloc the structures to release the
1808 excess space.
1809
1810 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1811 specifically Fig 3.47 and the associated pseudocode.
1812
1813 demq
1814 */
a3b680e6 1815 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 1816 U32 state, charid;
a3621e74 1817 U32 pos = 0, zp=0;
1e2e3d02 1818 trie->statecount = laststate;
a3621e74
YO
1819
1820 for ( state = 1 ; state < laststate ; state++ ) {
1821 U8 flag = 0;
a28509cc
AL
1822 const U32 stateidx = TRIE_NODEIDX( state );
1823 const U32 o_used = trie->trans[ stateidx ].check;
1824 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
1825 trie->trans[ stateidx ].check = 0;
1826
1827 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1828 if ( flag || trie->trans[ stateidx + charid ].next ) {
1829 if ( trie->trans[ stateidx + charid ].next ) {
1830 if (o_used == 1) {
1831 for ( ; zp < pos ; zp++ ) {
1832 if ( ! trie->trans[ zp ].next ) {
1833 break;
1834 }
1835 }
1836 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1837 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1838 trie->trans[ zp ].check = state;
1839 if ( ++zp > pos ) pos = zp;
1840 break;
1841 }
1842 used--;
1843 }
1844 if ( !flag ) {
1845 flag = 1;
1846 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1847 }
1848 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1849 trie->trans[ pos ].check = state;
1850 pos++;
1851 }
1852 }
1853 }
cc601c31 1854 trie->lasttrans = pos + 1;
c944940b
JH
1855 trie->states = (reg_trie_state *)
1856 PerlMemShared_realloc( trie->states, laststate
1857 * sizeof(reg_trie_state) );
a3621e74 1858 DEBUG_TRIE_COMPILE_MORE_r(
e4584336 1859 PerlIO_printf( Perl_debug_log,
3dab1dad
YO
1860 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1861 (int)depth * 2 + 2,"",
1862 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
5d7488b2
AL
1863 (IV)next_alloc,
1864 (IV)pos,
a3621e74
YO
1865 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1866 );
1867
1868 } /* end table compress */
1869 }
1e2e3d02
YO
1870 DEBUG_TRIE_COMPILE_MORE_r(
1871 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1872 (int)depth * 2 + 2, "",
1873 (UV)trie->statecount,
1874 (UV)trie->lasttrans)
1875 );
cc601c31 1876 /* resize the trans array to remove unused space */
c944940b
JH
1877 trie->trans = (reg_trie_trans *)
1878 PerlMemShared_realloc( trie->trans, trie->lasttrans
1879 * sizeof(reg_trie_trans) );
a3621e74 1880
3dab1dad 1881 /* and now dump out the compressed format */
2b8b4781 1882 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
07be1b83 1883
3dab1dad 1884 { /* Modify the program and insert the new TRIE node*/
3dab1dad
YO
1885 U8 nodetype =(U8)(flags & 0xFF);
1886 char *str=NULL;
786e8c11 1887
07be1b83 1888#ifdef DEBUGGING
e62cc96a 1889 regnode *optimize = NULL;
7122b237
YO
1890#ifdef RE_TRACK_PATTERN_OFFSETS
1891
b57a0404
JH
1892 U32 mjd_offset = 0;
1893 U32 mjd_nodelen = 0;
7122b237
YO
1894#endif /* RE_TRACK_PATTERN_OFFSETS */
1895#endif /* DEBUGGING */
a3621e74 1896 /*
3dab1dad
YO
1897 This means we convert either the first branch or the first Exact,
1898 depending on whether the thing following (in 'last') is a branch
1899 or not and whther first is the startbranch (ie is it a sub part of
1900 the alternation or is it the whole thing.)
1901 Assuming its a sub part we conver the EXACT otherwise we convert
1902 the whole branch sequence, including the first.
a3621e74 1903 */
3dab1dad 1904 /* Find the node we are going to overwrite */
7f69552c 1905 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 1906 /* branch sub-chain */
3dab1dad 1907 NEXT_OFF( first ) = (U16)(last - first);
7122b237 1908#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
1909 DEBUG_r({
1910 mjd_offset= Node_Offset((convert));
1911 mjd_nodelen= Node_Length((convert));
1912 });
7122b237 1913#endif
7f69552c 1914 /* whole branch chain */
7122b237
YO
1915 }
1916#ifdef RE_TRACK_PATTERN_OFFSETS
1917 else {
7f69552c
YO
1918 DEBUG_r({
1919 const regnode *nop = NEXTOPER( convert );
1920 mjd_offset= Node_Offset((nop));
1921 mjd_nodelen= Node_Length((nop));
1922 });
07be1b83
YO
1923 }
1924 DEBUG_OPTIMISE_r(
1925 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1926 (int)depth * 2 + 2, "",
786e8c11 1927 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 1928 );
7122b237 1929#endif
3dab1dad
YO
1930 /* But first we check to see if there is a common prefix we can
1931 split out as an EXACT and put in front of the TRIE node. */
1932 trie->startstate= 1;
55eed653 1933 if ( trie->bitmap && !widecharmap && !trie->jump ) {
3dab1dad 1934 U32 state;
1e2e3d02 1935 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 1936 U32 ofs = 0;
8e11feef
RGS
1937 I32 idx = -1;
1938 U32 count = 0;
1939 const U32 base = trie->states[ state ].trans.base;
a3621e74 1940
3dab1dad 1941 if ( trie->states[state].wordnum )
8e11feef 1942 count = 1;
a3621e74 1943
8e11feef 1944 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
cc601c31
YO
1945 if ( ( base + ofs >= trie->uniquecharcount ) &&
1946 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
a3621e74
YO
1947 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1948 {
3dab1dad 1949 if ( ++count > 1 ) {
2b8b4781 1950 SV **tmp = av_fetch( revcharmap, ofs, 0);
07be1b83 1951 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 1952 if ( state == 1 ) break;
3dab1dad
YO
1953 if ( count == 2 ) {
1954 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1955 DEBUG_OPTIMISE_r(
8e11feef
RGS
1956 PerlIO_printf(Perl_debug_log,
1957 "%*sNew Start State=%"UVuf" Class: [",
1958 (int)depth * 2 + 2, "",
786e8c11 1959 (UV)state));
be8e71aa 1960 if (idx >= 0) {
2b8b4781 1961 SV ** const tmp = av_fetch( revcharmap, idx, 0);
be8e71aa 1962 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 1963
3dab1dad 1964 TRIE_BITMAP_SET(trie,*ch);
8e11feef
RGS
1965 if ( folder )
1966 TRIE_BITMAP_SET(trie, folder[ *ch ]);
3dab1dad 1967 DEBUG_OPTIMISE_r(
07be1b83 1968 PerlIO_printf(Perl_debug_log, (char*)ch)
3dab1dad 1969 );
8e11feef
RGS
1970 }
1971 }
1972 TRIE_BITMAP_SET(trie,*ch);
1973 if ( folder )
1974 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1975 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1976 }
1977 idx = ofs;
1978 }
3dab1dad
YO
1979 }
1980 if ( count == 1 ) {
2b8b4781 1981 SV **tmp = av_fetch( revcharmap, idx, 0);
c490c714
YO
1982 STRLEN len;
1983 char *ch = SvPV( *tmp, len );
de734bd5
A
1984 DEBUG_OPTIMISE_r({
1985 SV *sv=sv_newmortal();
8e11feef
RGS
1986 PerlIO_printf( Perl_debug_log,
1987 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1988 (int)depth * 2 + 2, "",
de734bd5
A
1989 (UV)state, (UV)idx,
1990 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1991 PL_colors[0], PL_colors[1],
1992 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1993 PERL_PV_ESCAPE_FIRSTCHAR
1994 )
1995 );
1996 });
3dab1dad
YO
1997 if ( state==1 ) {
1998 OP( convert ) = nodetype;
1999 str=STRING(convert);
2000 STR_LEN(convert)=0;
2001 }
c490c714
YO
2002 STR_LEN(convert) += len;
2003 while (len--)
de734bd5 2004 *str++ = *ch++;
8e11feef 2005 } else {
f9049ba1 2006#ifdef DEBUGGING
8e11feef
RGS
2007 if (state>1)
2008 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
f9049ba1 2009#endif
8e11feef
RGS
2010 break;
2011 }
2012 }
3dab1dad 2013 if (str) {
8e11feef 2014 regnode *n = convert+NODE_SZ_STR(convert);
07be1b83 2015 NEXT_OFF(convert) = NODE_SZ_STR(convert);
8e11feef 2016 trie->startstate = state;
07be1b83
YO
2017 trie->minlen -= (state - 1);
2018 trie->maxlen -= (state - 1);
2019 DEBUG_r({
2020 regnode *fix = convert;
de734bd5 2021 U32 word = trie->wordcount;
07be1b83
YO
2022 mjd_nodelen++;
2023 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2024 while( ++fix < n ) {
2025 Set_Node_Offset_Length(fix, 0, 0);
2026 }
de734bd5 2027 while (word--) {
2b8b4781 2028 SV ** const tmp = av_fetch( trie_words, word, 0 );
de734bd5
A
2029 if (tmp) {
2030 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2031 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2032 else
2033 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2034 }
2035 }
07be1b83 2036 });
8e11feef
RGS
2037 if (trie->maxlen) {
2038 convert = n;
2039 } else {
3dab1dad 2040 NEXT_OFF(convert) = (U16)(tail - convert);
a5ca303d 2041 DEBUG_r(optimize= n);
3dab1dad
YO
2042 }
2043 }
2044 }
a5ca303d
YO
2045 if (!jumper)
2046 jumper = last;
3dab1dad 2047 if ( trie->maxlen ) {
8e11feef
RGS
2048 NEXT_OFF( convert ) = (U16)(tail - convert);
2049 ARG_SET( convert, data_slot );
786e8c11
YO
2050 /* Store the offset to the first unabsorbed branch in
2051 jump[0], which is otherwise unused by the jump logic.
2052 We use this when dumping a trie and during optimisation. */
2053 if (trie->jump)
7f69552c 2054 trie->jump[0] = (U16)(nextbranch - convert);
a5ca303d 2055
786e8c11
YO
2056 /* XXXX */
2057 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
1de06328 2058 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
786e8c11
YO
2059 {
2060 OP( convert ) = TRIEC;
2061 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
446bd890 2062 PerlMemShared_free(trie->bitmap);
786e8c11
YO
2063 trie->bitmap= NULL;
2064 } else
2065 OP( convert ) = TRIE;
a3621e74 2066
3dab1dad
YO
2067 /* store the type in the flags */
2068 convert->flags = nodetype;
a5ca303d
YO
2069 DEBUG_r({
2070 optimize = convert
2071 + NODE_STEP_REGNODE
2072 + regarglen[ OP( convert ) ];
2073 });
2074 /* XXX We really should free up the resource in trie now,
2075 as we won't use them - (which resources?) dmq */
3dab1dad 2076 }
a3621e74 2077 /* needed for dumping*/
e62cc96a 2078 DEBUG_r(if (optimize) {
07be1b83 2079 regnode *opt = convert;
bcdf7404 2080
e62cc96a 2081 while ( ++opt < optimize) {
07be1b83
YO
2082 Set_Node_Offset_Length(opt,0,0);
2083 }
786e8c11
YO
2084 /*
2085 Try to clean up some of the debris left after the
2086 optimisation.
a3621e74 2087 */
786e8c11 2088 while( optimize < jumper ) {
07be1b83 2089 mjd_nodelen += Node_Length((optimize));
a3621e74 2090 OP( optimize ) = OPTIMIZED;
07be1b83 2091 Set_Node_Offset_Length(optimize,0,0);
a3621e74
YO
2092 optimize++;
2093 }
07be1b83 2094 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
a3621e74
YO
2095 });
2096 } /* end node insert */
55eed653 2097 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2b8b4781
NC
2098#ifdef DEBUGGING
2099 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2100 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2101#else
2102 SvREFCNT_dec(revcharmap);
07be1b83 2103#endif
786e8c11
YO
2104 return trie->jump
2105 ? MADE_JUMP_TRIE
2106 : trie->startstate>1
2107 ? MADE_EXACT_TRIE
2108 : MADE_TRIE;
2109}
2110
2111STATIC void
2112S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2113{
2114/* The Trie is constructed and compressed now so we can build a fail array now if its needed
2115
2116 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2117 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2118 ISBN 0-201-10088-6
2119
2120 We find the fail state for each state in the trie, this state is the longest proper
2121 suffix of the current states 'word' that is also a proper prefix of another word in our
2122 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2123 the DFA not to have to restart after its tried and failed a word at a given point, it
2124 simply continues as though it had been matching the other word in the first place.
2125 Consider
2126 'abcdgu'=~/abcdefg|cdgu/
2127 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2128 fail, which would bring use to the state representing 'd' in the second word where we would
2129 try 'g' and succeed, prodceding to match 'cdgu'.
2130 */
2131 /* add a fail transition */
3251b653
NC
2132 const U32 trie_offset = ARG(source);
2133 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
786e8c11
YO
2134 U32 *q;
2135 const U32 ucharcount = trie->uniquecharcount;
1e2e3d02 2136 const U32 numstates = trie->statecount;
786e8c11
YO
2137 const U32 ubound = trie->lasttrans + ucharcount;
2138 U32 q_read = 0;
2139 U32 q_write = 0;
2140 U32 charid;
2141 U32 base = trie->states[ 1 ].trans.base;
2142 U32 *fail;
2143 reg_ac_data *aho;
2144 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2145 GET_RE_DEBUG_FLAGS_DECL;
2146#ifndef DEBUGGING
2147 PERL_UNUSED_ARG(depth);
2148#endif
2149
2150
2151 ARG_SET( stclass, data_slot );
c944940b 2152 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
f8fc2ecf 2153 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3251b653 2154 aho->trie=trie_offset;
446bd890
NC
2155 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2156 Copy( trie->states, aho->states, numstates, reg_trie_state );
786e8c11 2157 Newxz( q, numstates, U32);
c944940b 2158 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
786e8c11
YO
2159 aho->refcount = 1;
2160 fail = aho->fail;
2161 /* initialize fail[0..1] to be 1 so that we always have
2162 a valid final fail state */
2163 fail[ 0 ] = fail[ 1 ] = 1;
2164
2165 for ( charid = 0; charid < ucharcount ; charid++ ) {
2166 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2167 if ( newstate ) {
2168 q[ q_write ] = newstate;
2169 /* set to point at the root */
2170 fail[ q[ q_write++ ] ]=1;
2171 }
2172 }
2173 while ( q_read < q_write) {
2174 const U32 cur = q[ q_read++ % numstates ];
2175 base = trie->states[ cur ].trans.base;
2176
2177 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2178 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2179 if (ch_state) {
2180 U32 fail_state = cur;
2181 U32 fail_base;
2182 do {
2183 fail_state = fail[ fail_state ];
2184 fail_base = aho->states[ fail_state ].trans.base;
2185 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2186
2187 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2188 fail[ ch_state ] = fail_state;
2189 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2190 {
2191 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2192 }
2193 q[ q_write++ % numstates] = ch_state;
2194 }
2195 }
2196 }
2197 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2198 when we fail in state 1, this allows us to use the
2199 charclass scan to find a valid start char. This is based on the principle
2200 that theres a good chance the string being searched contains lots of stuff
2201 that cant be a start char.
2202 */
2203 fail[ 0 ] = fail[ 1 ] = 0;
2204 DEBUG_TRIE_COMPILE_r({
6d99fb9b
JH
2205 PerlIO_printf(Perl_debug_log,
2206 "%*sStclass Failtable (%"UVuf" states): 0",
2207 (int)(depth * 2), "", (UV)numstates
1e2e3d02 2208 );
786e8c11
YO
2209 for( q_read=1; q_read<numstates; q_read++ ) {
2210 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2211 }
2212 PerlIO_printf(Perl_debug_log, "\n");
2213 });
2214 Safefree(q);
2215 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
a3621e74
YO
2216}
2217
786e8c11 2218
a3621e74 2219/*
5d1c421c
JH
2220 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2221 * These need to be revisited when a newer toolchain becomes available.
2222 */
2223#if defined(__sparc64__) && defined(__GNUC__)
2224# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2225# undef SPARC64_GCC_WORKAROUND
2226# define SPARC64_GCC_WORKAROUND 1
2227# endif
2228#endif
2229
07be1b83 2230#define DEBUG_PEEP(str,scan,depth) \
b515a41d 2231 DEBUG_OPTIMISE_r({if (scan){ \
07be1b83
YO
2232 SV * const mysv=sv_newmortal(); \
2233 regnode *Next = regnext(scan); \
2234 regprop(RExC_rx, mysv, scan); \
7f69552c 2235 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
07be1b83
YO
2236 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2237 Next ? (REG_NODE_NUM(Next)) : 0 ); \
b515a41d 2238 }});
07be1b83 2239
1de06328
YO
2240
2241
2242
2243
07be1b83
YO
2244#define JOIN_EXACT(scan,min,flags) \
2245 if (PL_regkind[OP(scan)] == EXACT) \
2246 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2247
be8e71aa 2248STATIC U32
07be1b83
YO
2249S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2250 /* Merge several consecutive EXACTish nodes into one. */
2251 regnode *n = regnext(scan);
2252 U32 stringok = 1;
2253 regnode *next = scan + NODE_SZ_STR(scan);
2254 U32 merged = 0;
2255 U32 stopnow = 0;
2256#ifdef DEBUGGING
2257 regnode *stop = scan;
72f13be8 2258 GET_RE_DEBUG_FLAGS_DECL;
f9049ba1 2259#else
d47053eb
RGS
2260 PERL_UNUSED_ARG(depth);
2261#endif
2262#ifndef EXPERIMENTAL_INPLACESCAN
f9049ba1
SP
2263 PERL_UNUSED_ARG(flags);
2264 PERL_UNUSED_ARG(val);
07be1b83 2265#endif
07be1b83
YO
2266 DEBUG_PEEP("join",scan,depth);
2267
2268 /* Skip NOTHING, merge EXACT*. */
2269 while (n &&
2270 ( PL_regkind[OP(n)] == NOTHING ||
2271 (stringok && (OP(n) == OP(scan))))
2272 && NEXT_OFF(n)
2273 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2274
2275 if (OP(n) == TAIL || n > next)
2276 stringok = 0;
2277 if (PL_regkind[OP(n)] == NOTHING) {
07be1b83
YO
2278 DEBUG_PEEP("skip:",n,depth);
2279 NEXT_OFF(scan) += NEXT_OFF(n);
2280 next = n + NODE_STEP_REGNODE;
2281#ifdef DEBUGGING
2282 if (stringok)
2283 stop = n;
2284#endif
2285 n = regnext(n);
2286 }
2287 else if (stringok) {
786e8c11 2288 const unsigned int oldl = STR_LEN(scan);
07be1b83
YO
2289 regnode * const nnext = regnext(n);
2290
2291 DEBUG_PEEP("merg",n,depth);
2292
2293 merged++;
2294 if (oldl + STR_LEN(n) > U8_MAX)
2295 break;
2296 NEXT_OFF(scan) += NEXT_OFF(n);
2297 STR_LEN(scan) += STR_LEN(n);
2298 next = n + NODE_SZ_STR(n);
2299 /* Now we can overwrite *n : */
2300 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2301#ifdef DEBUGGING
2302 stop = next - 1;
2303#endif
2304 n = nnext;
2305 if (stopnow) break;
2306 }
2307
d47053eb
RGS
2308#ifdef EXPERIMENTAL_INPLACESCAN
2309 if (flags && !NEXT_OFF(n)) {
2310 DEBUG_PEEP("atch", val, depth);
2311 if (reg_off_by_arg[OP(n)]) {
2312 ARG_SET(n, val - n);
2313 }
2314 else {
2315 NEXT_OFF(n) = val - n;
2316 }
2317 stopnow = 1;
2318 }
07be1b83
YO
2319#endif
2320 }
2321
2322 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2323 /*
2324 Two problematic code points in Unicode casefolding of EXACT nodes:
2325
2326 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2327 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2328
2329 which casefold to
2330
2331 Unicode UTF-8
2332
2333 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2334 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2335
2336 This means that in case-insensitive matching (or "loose matching",
2337 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2338 length of the above casefolded versions) can match a target string
2339 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2340 This would rather mess up the minimum length computation.
2341
2342 What we'll do is to look for the tail four bytes, and then peek
2343 at the preceding two bytes to see whether we need to decrease
2344 the minimum length by four (six minus two).
2345
2346 Thanks to the design of UTF-8, there cannot be false matches:
2347 A sequence of valid UTF-8 bytes cannot be a subsequence of
2348 another valid sequence of UTF-8 bytes.
2349
2350 */
2351 char * const s0 = STRING(scan), *s, *t;
2352 char * const s1 = s0 + STR_LEN(scan) - 1;
2353 char * const s2 = s1 - 4;
e294cc5d
JH
2354#ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2355 const char t0[] = "\xaf\x49\xaf\x42";
2356#else
07be1b83 2357 const char t0[] = "\xcc\x88\xcc\x81";
e294cc5d 2358#endif
07be1b83
YO
2359 const char * const t1 = t0 + 3;
2360
2361 for (s = s0 + 2;
2362 s < s2 && (t = ninstr(s, s1, t0, t1));
2363 s = t + 4) {
e294cc5d
JH
2364#ifdef EBCDIC
2365 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2366 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2367#else
07be1b83
YO
2368 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2369 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
e294cc5d 2370#endif
07be1b83
YO
2371 *min -= 4;
2372 }
2373 }
2374
2375#ifdef DEBUGGING
2376 /* Allow dumping */
2377 n = scan + NODE_SZ_STR(scan);
2378 while (n <= stop) {
2379 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2380 OP(n) = OPTIMIZED;
2381 NEXT_OFF(n) = 0;
2382 }
2383 n++;
2384 }
2385#endif
2386 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2387 return stopnow;
2388}
2389
653099ff
GS
2390/* REx optimizer. Converts nodes into quickier variants "in place".
2391 Finds fixed substrings. */
2392
a0288114 2393/* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
c277df42
IZ
2394 to the position after last scanned or to NULL. */
2395
40d049e4
YO
2396#define INIT_AND_WITHP \
2397 assert(!and_withp); \
2398 Newx(and_withp,1,struct regnode_charclass_class); \
2399 SAVEFREEPV(and_withp)
07be1b83 2400
b515a41d
YO
2401/* this is a chain of data about sub patterns we are processing that
2402 need to be handled seperately/specially in study_chunk. Its so
2403 we can simulate recursion without losing state. */
2404struct scan_frame;
2405typedef struct scan_frame {
2406 regnode *last; /* last node to process in this frame */
2407 regnode *next; /* next node to process when last is reached */
2408 struct scan_frame *prev; /*previous frame*/
2409 I32 stop; /* what stopparen do we use */
2410} scan_frame;
2411
304ee84b
YO
2412
2413#define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2414
e1d1eefb
YO
2415#define CASE_SYNST_FNC(nAmE) \
2416case nAmE: \
2417 if (flags & SCF_DO_STCLASS_AND) { \
2418 for (value = 0; value < 256; value++) \
2419 if (!is_ ## nAmE ## _cp(value)) \
2420 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2421 } \
2422 else { \
2423 for (value = 0; value < 256; value++) \
2424 if (is_ ## nAmE ## _cp(value)) \
2425 ANYOF_BITMAP_SET(data->start_class, value); \
2426 } \
2427 break; \
2428case N ## nAmE: \
2429 if (flags & SCF_DO_STCLASS_AND) { \
2430 for (value = 0; value < 256; value++) \
2431 if (is_ ## nAmE ## _cp(value)) \
2432 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2433 } \
2434 else { \
2435 for (value = 0; value < 256; value++) \
2436 if (!is_ ## nAmE ## _cp(value)) \
2437 ANYOF_BITMAP_SET(data->start_class, value); \
2438 } \
2439 break
2440
2441
2442
76e3520e 2443STATIC I32
40d049e4 2444S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
1de06328 2445 I32 *minlenp, I32 *deltap,
40d049e4
YO
2446 regnode *last,
2447 scan_data_t *data,
2448 I32 stopparen,
2449 U8* recursed,
2450 struct regnode_charclass_class *and_withp,
2451 U32 flags, U32 depth)
c277df42
IZ
2452 /* scanp: Start here (read-write). */
2453 /* deltap: Write maxlen-minlen here. */
2454 /* last: Stop before this one. */
40d049e4
YO
2455 /* data: string data about the pattern */
2456 /* stopparen: treat close N as END */
2457 /* recursed: which subroutines have we recursed into */
2458 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
c277df42 2459{
97aff369 2460 dVAR;
c277df42
IZ
2461 I32 min = 0, pars = 0, code;
2462 regnode *scan = *scanp, *next;
2463 I32 delta = 0;
2464 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
aca2d497 2465 int is_inf_internal = 0; /* The studied chunk is infinite */
c277df42
IZ
2466 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2467 scan_data_t data_fake;
a3621e74 2468 SV *re_trie_maxbuff = NULL;
786e8c11 2469 regnode *first_non_open = scan;
e2e6a0f1 2470 I32 stopmin = I32_MAX;
8aa23a47
YO
2471 scan_frame *frame = NULL;
2472
a3621e74 2473 GET_RE_DEBUG_FLAGS_DECL;
8aa23a47 2474
13a24bad 2475#ifdef DEBUGGING
40d049e4 2476 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
13a24bad 2477#endif
40d049e4 2478
786e8c11 2479 if ( depth == 0 ) {
40d049e4 2480 while (first_non_open && OP(first_non_open) == OPEN)
786e8c11
YO
2481 first_non_open=regnext(first_non_open);
2482 }
2483
b81d288d 2484
8aa23a47
YO
2485 fake_study_recurse:
2486 while ( scan && OP(scan) != END && scan < last ){
2487 /* Peephole optimizer: */
304ee84b 2488 DEBUG_STUDYDATA("Peep:", data,depth);
8aa23a47
YO
2489 DEBUG_PEEP("Peep",scan,depth);
2490 JOIN_EXACT(scan,&min,0);
2491
2492 /* Follow the next-chain of the current node and optimize
2493 away all the NOTHINGs from it. */
2494 if (OP(scan) != CURLYX) {
2495 const int max = (reg_off_by_arg[OP(scan)]
2496 ? I32_MAX
2497 /* I32 may be smaller than U16 on CRAYs! */
2498 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2499 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2500 int noff;
2501 regnode *n = scan;
2502
2503 /* Skip NOTHING and LONGJMP. */
2504 while ((n = regnext(n))
2505 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2506 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2507 && off + noff < max)
2508 off += noff;
2509 if (reg_off_by_arg[OP(scan)])
2510 ARG(scan) = off;
2511 else
2512 NEXT_OFF(scan) = off;
2513 }
a3621e74 2514
c277df42 2515
8aa23a47
YO
2516
2517 /* The principal pseudo-switch. Cannot be a switch, since we
2518 look into several different things. */
2519 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2520 || OP(scan) == IFTHEN) {
2521 next = regnext(scan);
2522 code = OP(scan);
2523 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2524
2525 if (OP(next) == code || code == IFTHEN) {
2526 /* NOTE - There is similar code to this block below for handling
2527 TRIE nodes on a re-study. If you change stuff here check there
2528 too. */
2529 I32 max1 = 0, min1 = I32_MAX, num = 0;
2530 struct regnode_charclass_class accum;
2531 regnode * const startbranch=scan;
2532
2533 if (flags & SCF_DO_SUBSTR)
304ee84b 2534 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
8aa23a47
YO
2535 if (flags & SCF_DO_STCLASS)
2536 cl_init_zero(pRExC_state, &accum);
2537
2538 while (OP(scan) == code) {
2539 I32 deltanext, minnext, f = 0, fake;
2540 struct regnode_charclass_class this_class;
2541
2542 num++;
2543 data_fake.flags = 0;
2544 if (data) {
2545 data_fake.whilem_c = data->whilem_c;
2546 data_fake.last_closep = data->last_closep;
2547 }
2548 else
2549 data_fake.last_closep = &fake;
58e23c8d
YO
2550
2551 data_fake.pos_delta = delta;
8aa23a47
YO
2552 next = regnext(scan);
2553 scan = NEXTOPER(scan);
2554 if (code != BRANCH)
c277df42 2555 scan = NEXTOPER(scan);
8aa23a47
YO
2556 if (flags & SCF_DO_STCLASS) {
2557 cl_init(pRExC_state, &this_class);
2558 data_fake.start_class = &this_class;
2559 f = SCF_DO_STCLASS_AND;
58e23c8d 2560 }
8aa23a47
YO
2561 if (flags & SCF_WHILEM_VISITED_POS)
2562 f |= SCF_WHILEM_VISITED_POS;
2563
2564 /* we suppose the run is continuous, last=next...*/
2565 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2566 next, &data_fake,
2567 stopparen, recursed, NULL, f,depth+1);
2568 if (min1 > minnext)
2569 min1 = minnext;
2570 if (max1 < minnext + deltanext)
2571 max1 = minnext + deltanext;
2572 if (deltanext == I32_MAX)
2573 is_inf = is_inf_internal = 1;
2574 scan = next;
2575 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2576 pars++;
2577 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2578 if ( stopmin > minnext)
2579 stopmin = min + min1;
2580 flags &= ~SCF_DO_SUBSTR;
2581 if (data)
2582 data->flags |= SCF_SEEN_ACCEPT;
2583 }
2584 if (data) {
2585 if (data_fake.flags & SF_HAS_EVAL)
2586 data->flags |= SF_HAS_EVAL;
2587 data->whilem_c = data_fake.whilem_c;
3dab1dad 2588 }
8aa23a47
YO
2589 if (flags & SCF_DO_STCLASS)
2590 cl_or(pRExC_state, &accum, &this_class);
2591 }
2592 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2593 min1 = 0;
2594 if (flags & SCF_DO_SUBSTR) {
2595 data->pos_min += min1;
2596 data->pos_delta += max1 - min1;
2597 if (max1 != min1 || is_inf)
2598 data->longest = &(data->longest_float);
2599 }
2600 min += min1;
2601 delta += max1 - min1;
2602 if (flags & SCF_DO_STCLASS_OR) {
2603 cl_or(pRExC_state, data->start_class, &accum);
2604 if (min1) {
2605 cl_and(data->start_class, and_withp);
2606 flags &= ~SCF_DO_STCLASS;
653099ff 2607 }
8aa23a47
YO
2608 }
2609 else if (flags & SCF_DO_STCLASS_AND) {
2610 if (min1) {
2611 cl_and(data->start_class, &accum);
2612 flags &= ~SCF_DO_STCLASS;
de0c8cb8 2613 }
8aa23a47
YO
2614 else {
2615 /* Switch to OR mode: cache the old value of
2616 * data->start_class */
2617 INIT_AND_WITHP;
2618 StructCopy(data->start_class, and_withp,
2619 struct regnode_charclass_class);
2620 flags &= ~SCF_DO_STCLASS_AND;
2621 StructCopy(&accum, data->start_class,
2622 struct regnode_charclass_class);
2623 flags |= SCF_DO_STCLASS_OR;
2624 data->start_class->flags |= ANYOF_EOS;
de0c8cb8 2625 }
8aa23a47 2626 }
a3621e74 2627
8aa23a47
YO
2628 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2629 /* demq.
a3621e74 2630
8aa23a47
YO
2631 Assuming this was/is a branch we are dealing with: 'scan' now
2632 points at the item that follows the branch sequence, whatever
2633 it is. We now start at the beginning of the sequence and look
2634 for subsequences of
a3621e74 2635
8aa23a47
YO
2636 BRANCH->EXACT=>x1
2637 BRANCH->EXACT=>x2
2638 tail
a3621e74 2639
8aa23a47 2640 which would be constructed from a pattern like /A|LIST|OF|WORDS/
a3621e74 2641
8aa23a47
YO
2642 If we can find such a subseqence we need to turn the first
2643 element into a trie and then add the subsequent branch exact
2644 strings to the trie.
a3621e74 2645
8aa23a47 2646 We have two cases
a3621e74 2647
8aa23a47 2648 1. patterns where the whole set of branch can be converted.
a3621e74 2649
8aa23a47 2650 2. patterns where only a subset can be converted.
a3621e74 2651
8aa23a47
YO
2652 In case 1 we can replace the whole set with a single regop
2653 for the trie. In case 2 we need to keep the start and end
2654 branchs so
a3621e74 2655
8aa23a47
YO
2656 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2657 becomes BRANCH TRIE; BRANCH X;
786e8c11 2658
8aa23a47
YO
2659 There is an additional case, that being where there is a
2660 common prefix, which gets split out into an EXACT like node
2661 preceding the TRIE node.
a3621e74 2662
8aa23a47
YO
2663 If x(1..n)==tail then we can do a simple trie, if not we make
2664 a "jump" trie, such that when we match the appropriate word
2665 we "jump" to the appopriate tail node. Essentailly we turn
2666 a nested if into a case structure of sorts.
b515a41d 2667
8aa23a47
YO
2668 */
2669
2670 int made=0;
2671 if (!re_trie_maxbuff) {
2672 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2673 if (!SvIOK(re_trie_maxbuff))
2674 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2675 }
2676 if ( SvIV(re_trie_maxbuff)>=0 ) {
2677 regnode *cur;
2678 regnode *first = (regnode *)NULL;
2679 regnode *last = (regnode *)NULL;
2680 regnode *tail = scan;
2681 U8 optype = 0;
2682 U32 count=0;
a3621e74
YO
2683
2684#ifdef DEBUGGING
8aa23a47 2685 SV * const mysv = sv_newmortal(); /* for dumping */
a3621e74 2686#endif
8aa23a47
YO
2687 /* var tail is used because there may be a TAIL
2688 regop in the way. Ie, the exacts will point to the
2689 thing following the TAIL, but the last branch will
2690 point at the TAIL. So we advance tail. If we
2691 have nested (?:) we may have to move through several
2692 tails.
2693 */
2694
2695 while ( OP( tail ) == TAIL ) {
2696 /* this is the TAIL generated by (?:) */
2697 tail = regnext( tail );
2698 }
a3621e74 2699
8aa23a47
YO
2700
2701 DEBUG_OPTIMISE_r({
2702 regprop(RExC_rx, mysv, tail );
2703 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2704 (int)depth * 2 + 2, "",
2705 "Looking for TRIE'able sequences. Tail node is: ",
2706 SvPV_nolen_const( mysv )
2707 );
2708 });
2709
2710 /*
2711
2712 step through the branches, cur represents each
2713 branch, noper is the first thing to be matched
2714 as part of that branch and noper_next is the
2715 regnext() of that node. if noper is an EXACT
2716 and noper_next is the same as scan (our current
2717 position in the regex) then the EXACT branch is
2718 a possible optimization target. Once we have
2719 two or more consequetive such branches we can
2720 create a trie of the EXACT's contents and stich
2721 it in place. If the sequence represents all of
2722 the branches we eliminate the whole thing and
2723 replace it with a single TRIE. If it is a
2724 subsequence then we need to stitch it in. This
2725 means the first branch has to remain, and needs
2726 to be repointed at the item on the branch chain
2727 following the last branch optimized. This could
2728 be either a BRANCH, in which case the
2729 subsequence is internal, or it could be the
2730 item following the branch sequence in which
2731 case the subsequence is at the end.
2732
2733 */
2734
2735 /* dont use tail as the end marker for this traverse */
2736 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2737 regnode * const noper = NEXTOPER( cur );
b515a41d 2738#if defined(DEBUGGING) || defined(NOJUMPTRIE)
8aa23a47 2739 regnode * const noper_next = regnext( noper );
b515a41d
YO
2740#endif
2741
8aa23a47
YO
2742 DEBUG_OPTIMISE_r({
2743 regprop(RExC_rx, mysv, cur);
2744 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2745 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2746
2747 regprop(RExC_rx, mysv, noper);
2748 PerlIO_printf( Perl_debug_log, " -> %s",
2749 SvPV_nolen_const(mysv));
2750
2751 if ( noper_next ) {
2752 regprop(RExC_rx, mysv, noper_next );
2753 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2754 SvPV_nolen_const(mysv));
2755 }
2756 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2757 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2758 });
2759 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2760 : PL_regkind[ OP( noper ) ] == EXACT )
2761 || OP(noper) == NOTHING )
786e8c11 2762#ifdef NOJUMPTRIE
8aa23a47 2763 && noper_next == tail
786e8c11 2764#endif
8aa23a47
YO
2765 && count < U16_MAX)
2766 {
2767 count++;
2768 if ( !first || optype == NOTHING ) {
2769 if (!first) first = cur;
2770 optype = OP( noper );
2771 } else {
2772 last = cur;
2773 }
2774 } else {
2775 if ( last ) {
2776 make_trie( pRExC_state,
2777 startbranch, first, cur, tail, count,
2778 optype, depth+1 );
2779 }
2780 if ( PL_regkind[ OP( noper ) ] == EXACT
786e8c11 2781#ifdef NOJUMPTRIE
8aa23a47 2782 && noper_next == tail
786e8c11 2783#endif
8aa23a47
YO
2784 ){
2785 count = 1;
2786 first = cur;
2787 optype = OP( noper );
2788 } else {
2789 count = 0;
2790 first = NULL;
2791 optype = 0;
2792 }
2793 last = NULL;
2794 }
2795 }
2796 DEBUG_OPTIMISE_r({
2797 regprop(RExC_rx, mysv, cur);
2798 PerlIO_printf( Perl_debug_log,
2799 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2800 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2801
2802 });
2803 if ( last ) {
2804 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3dab1dad 2805#ifdef TRIE_STUDY_OPT
8aa23a47
YO
2806 if ( ((made == MADE_EXACT_TRIE &&
2807 startbranch == first)
2808 || ( first_non_open == first )) &&
2809 depth==0 ) {
2810 flags |= SCF_TRIE_RESTUDY;
2811 if ( startbranch == first
2812 && scan == tail )
2813 {
2814 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2815 }
2816 }
3dab1dad 2817#endif
8aa23a47
YO
2818 }
2819 }
2820
2821 } /* do trie */
2822
653099ff 2823 }
8aa23a47
YO
2824 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2825 scan = NEXTOPER(NEXTOPER(scan));
2826 } else /* single branch is optimized. */
2827 scan = NEXTOPER(scan);
2828 continue;
2829 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2830 scan_frame *newframe = NULL;
2831 I32 paren;
2832 regnode *start;
2833 regnode *end;
2834
2835 if (OP(scan) != SUSPEND) {
2836 /* set the pointer */
2837 if (OP(scan) == GOSUB) {
2838 paren = ARG(scan);
2839 RExC_recurse[ARG2L(scan)] = scan;
2840 start = RExC_open_parens[paren-1];
2841 end = RExC_close_parens[paren-1];
2842 } else {
2843 paren = 0;
f8fc2ecf 2844 start = RExC_rxi->program + 1;
8aa23a47
YO
2845 end = RExC_opend;
2846 }
2847 if (!recursed) {
2848 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2849 SAVEFREEPV(recursed);
2850 }
2851 if (!PAREN_TEST(recursed,paren+1)) {
2852 PAREN_SET(recursed,paren+1);
2853 Newx(newframe,1,scan_frame);
2854 } else {
2855 if (flags & SCF_DO_SUBSTR) {
304ee84b 2856 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
2857 data->longest = &(data->longest_float);
2858 }
2859 is_inf = is_inf_internal = 1;
2860 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2861 cl_anything(pRExC_state, data->start_class);
2862 flags &= ~SCF_DO_STCLASS;
2863 }
2864 } else {
2865 Newx(newframe,1,scan_frame);
2866 paren = stopparen;
2867 start = scan+2;
2868 end = regnext(scan);
2869 }
2870 if (newframe) {
2871 assert(start);
2872 assert(end);
2873 SAVEFREEPV(newframe);
2874 newframe->next = regnext(scan);
2875 newframe->last = last;
2876 newframe->stop = stopparen;
2877 newframe->prev = frame;
2878
2879 frame = newframe;
2880 scan = start;
2881 stopparen = paren;
2882 last = end;
2883
2884 continue;
2885 }
2886 }
2887 else if (OP(scan) == EXACT) {
2888 I32 l = STR_LEN(scan);
2889 UV uc;
2890 if (UTF) {
2891 const U8 * const s = (U8*)STRING(scan);
2892 l = utf8_length(s, s + l);
2893 uc = utf8_to_uvchr(s, NULL);
2894 } else {
2895 uc = *((U8*)STRING(scan));
2896 }
2897 min += l;
2898 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2899 /* The code below prefers earlier match for fixed
2900 offset, later match for variable offset. */
2901 if (data->last_end == -1) { /* Update the start info. */
2902 data->last_start_min = data->pos_min;
2903 data->last_start_max = is_inf
2904 ? I32_MAX : data->pos_min + data->pos_delta;
b515a41d 2905 }
8aa23a47
YO
2906 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2907 if (UTF)
2908 SvUTF8_on(data->last_found);
2909 {
2910 SV * const sv = data->last_found;
2911 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2912 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2913 if (mg && mg->mg_len >= 0)
2914 mg->mg_len += utf8_length((U8*)STRING(scan),
2915 (U8*)STRING(scan)+STR_LEN(scan));
b515a41d 2916 }
8aa23a47
YO
2917 data->last_end = data->pos_min + l;
2918 data->pos_min += l; /* As in the first entry. */
2919 data->flags &= ~SF_BEFORE_EOL;
2920 }
2921 if (flags & SCF_DO_STCLASS_AND) {
2922 /* Check whether it is compatible with what we know already! */
2923 int compat = 1;
2924
2925 if (uc >= 0x100 ||
2926 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2927 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2928 && (!(data->start_class->flags & ANYOF_FOLD)
2929 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2930 )
2931 compat = 0;
2932 ANYOF_CLASS_ZERO(data->start_class);
2933 ANYOF_BITMAP_ZERO(data->start_class);
2934 if (compat)
2935 ANYOF_BITMAP_SET(data->start_class, uc);
2936 data->start_class->flags &= ~ANYOF_EOS;
2937 if (uc < 0x100)
2938 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2939 }
2940 else if (flags & SCF_DO_STCLASS_OR) {
2941 /* false positive possible if the class is case-folded */
2942 if (uc < 0x100)
2943 ANYOF_BITMAP_SET(data->start_class, uc);
2944 else
2945 data->start_class->flags |= ANYOF_UNICODE_ALL;
2946 data->start_class->flags &= ~ANYOF_EOS;
2947 cl_and(data->start_class, and_withp);
2948 }
2949 flags &= ~SCF_DO_STCLASS;
2950 }
2951 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2952 I32 l = STR_LEN(scan);
2953 UV uc = *((U8*)STRING(scan));
2954
2955 /* Search for fixed substrings supports EXACT only. */
2956 if (flags & SCF_DO_SUBSTR) {
2957 assert(data);
304ee84b 2958 SCAN_COMMIT(pRExC_state, data, minlenp);
8aa23a47
YO
2959 }
2960 if (UTF) {
2961 const U8 * const s = (U8 *)STRING(scan);
2962 l = utf8_length(s, s + l);
2963 uc = utf8_to_uvchr(s, NULL);
2964 }
2965 min += l;
2966 if (flags & SCF_DO_SUBSTR)
2967 data->pos_min += l;
2968 if (flags & SCF_DO_STCLASS_AND) {
2969 /* Check whether it is compatible with what we know already! */
2970 int compat = 1;
2971
2972 if (uc >= 0x100 ||
2973 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2974 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2975 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2976 compat = 0;
2977 ANYOF_CLASS_ZERO(data->start_class);
2978 ANYOF_BITMAP_ZERO(data->start_class);
2979 if (compat) {
2980 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff 2981 data->start_class->flags &= ~ANYOF_EOS;
8aa23a47
YO
2982 data->start_class->flags |= ANYOF_FOLD;
2983 if (OP(scan) == EXACTFL)
2984 data->start_class->flags |= ANYOF_LOCALE;
653099ff 2985 }
8aa23a47
YO
2986 }
2987 else if (flags & SCF_DO_STCLASS_OR) {
2988 if (data->start_class->flags & ANYOF_FOLD) {
2989 /* false positive possible if the class is case-folded.
2990 Assume that the locale settings are the same... */
1aa99e6b
IH
2991 if (uc < 0x100)
2992 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff
GS
2993 data->start_class->flags &= ~ANYOF_EOS;
2994 }
8aa23a47 2995 cl_and(data->start_class, and_withp);
653099ff 2996 }
8aa23a47
YO
2997 flags &= ~SCF_DO_STCLASS;
2998 }
2999 else if (strchr((const char*)PL_varies,OP(scan))) {
3000 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3001 I32 f = flags, pos_before = 0;
3002 regnode * const oscan = scan;
3003 struct regnode_charclass_class this_class;
3004 struct regnode_charclass_class *oclass = NULL;
3005 I32 next_is_eval = 0;
3006
3007 switch (PL_regkind[OP(scan)]) {
3008 case WHILEM: /* End of (?:...)* . */
3009 scan = NEXTOPER(scan);
3010 goto finish;
3011 case PLUS:
3012 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3013 next = NEXTOPER(scan);
3014 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3015 mincount = 1;
3016 maxcount = REG_INFTY;
3017 next = regnext(scan);
3018 scan = NEXTOPER(scan);
3019 goto do_curly;
3020 }
3021 }
3022 if (flags & SCF_DO_SUBSTR)
3023 data->pos_min++;
3024 min++;
3025 /* Fall through. */
3026 case STAR:
3027 if (flags & SCF_DO_STCLASS) {
3028 mincount = 0;
3029 maxcount = REG_INFTY;
3030 next = regnext(scan);
3031 scan = NEXTOPER(scan);
3032 goto do_curly;
3033 }
3034 is_inf = is_inf_internal = 1;
3035 scan = regnext(scan);
c277df42 3036 if (flags & SCF_DO_SUBSTR) {
304ee84b 3037 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
8aa23a47 3038 data->longest = &(data->longest_float);
c277df42 3039 }
8aa23a47
YO
3040 goto optimize_curly_tail;
3041 case CURLY:
3042 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3043 && (scan->flags == stopparen))
3044 {
3045 mincount = 1;
3046 maxcount = 1;
3047 } else {
3048 mincount = ARG1(scan);
3049 maxcount = ARG2(scan);
653099ff 3050 }
8aa23a47
YO
3051 next = regnext(scan);
3052 if (OP(scan) == CURLYX) {
3053 I32 lp = (data ? *(data->last_closep) : 0);
3054 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
653099ff 3055 }
8aa23a47
YO
3056 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3057 next_is_eval = (OP(scan) == EVAL);
3058 do_curly:
3059 if (flags & SCF_DO_SUBSTR) {
304ee84b 3060 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
8aa23a47 3061 pos_before = data->pos_min;
b45f050a 3062 }
8aa23a47
YO
3063 if (data) {
3064 fl = data->flags;
3065 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3066 if (is_inf)
3067 data->flags |= SF_IS_INF;
3068 }
3069 if (flags & SCF_DO_STCLASS) {
3070 cl_init(pRExC_state, &this_class);
3071 oclass = data->start_class;
3072 data->start_class = &this_class;
3073 f |= SCF_DO_STCLASS_AND;
3074 f &= ~SCF_DO_STCLASS_OR;
3075 }
3076 /* These are the cases when once a subexpression
3077 fails at a particular position, it cannot succeed
3078 even after backtracking at the enclosing scope.
3079
3080 XXXX what if minimal match and we are at the
3081 initial run of {n,m}? */
3082 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3083 f &= ~SCF_WHILEM_VISITED_POS;
b45f050a 3084
8aa23a47
YO
3085 /* This will finish on WHILEM, setting scan, or on NULL: */
3086 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3087 last, data, stopparen, recursed, NULL,
3088 (mincount == 0
3089 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
b515a41d 3090
8aa23a47
YO
3091 if (flags & SCF_DO_STCLASS)
3092 data->start_class = oclass;
3093 if (mincount == 0 || minnext == 0) {
3094 if (flags & SCF_DO_STCLASS_OR) {
3095 cl_or(pRExC_state, data->start_class, &this_class);
3096 }
3097 else if (flags & SCF_DO_STCLASS_AND) {
3098 /* Switch to OR mode: cache the old value of
3099 * data->start_class */
3100 INIT_AND_WITHP;
3101 StructCopy(data->start_class, and_withp,
3102 struct regnode_charclass_class);
3103 flags &= ~SCF_DO_STCLASS_AND;
3104 StructCopy(&this_class, data->start_class,
3105 struct regnode_charclass_class);
3106 flags |= SCF_DO_STCLASS_OR;
3107 data->start_class->flags |= ANYOF_EOS;
3108 }
3109 } else { /* Non-zero len */
3110 if (flags & SCF_DO_STCLASS_OR) {
3111 cl_or(pRExC_state, data->start_class, &this_class);
3112 cl_and(data->start_class, and_withp);
3113 }
3114 else if (flags & SCF_DO_STCLASS_AND)
3115 cl_and(data->start_class, &this_class);
3116 flags &= ~SCF_DO_STCLASS;
3117 }
3118 if (!scan) /* It was not CURLYX, but CURLY. */
3119 scan = next;
3120 if ( /* ? quantifier ok, except for (?{ ... }) */
3121 (next_is_eval || !(mincount == 0 && maxcount == 1))
3122 && (minnext == 0) && (deltanext == 0)
3123 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3124 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3125 && ckWARN(WARN_REGEXP))
3126 {
3127 vWARN(RExC_parse,
3128 "Quantifier unexpected on zero-length expression");
3129 }
3130
3131 min += minnext * mincount;
3132 is_inf_internal |= ((maxcount == REG_INFTY
3133 && (minnext + deltanext) > 0)
3134 || deltanext == I32_MAX);
3135 is_inf |= is_inf_internal;
3136 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3137
3138 /* Try powerful optimization CURLYX => CURLYN. */
3139 if ( OP(oscan) == CURLYX && data
3140 && data->flags & SF_IN_PAR
3141 && !(data->flags & SF_HAS_EVAL)
3142 && !deltanext && minnext == 1 ) {
3143 /* Try to optimize to CURLYN. */
3144 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3145 regnode * const nxt1 = nxt;
497b47a8 3146#ifdef DEBUGGING
8aa23a47 3147 regnode *nxt2;
497b47a8 3148#endif
c277df42 3149
8aa23a47
YO
3150 /* Skip open. */
3151 nxt = regnext(nxt);
3152 if (!strchr((const char*)PL_simple,OP(nxt))
3153 && !(PL_regkind[OP(nxt)] == EXACT
3154 && STR_LEN(nxt) == 1))
3155 goto nogo;
497b47a8 3156#ifdef DEBUGGING
8aa23a47 3157 nxt2 = nxt;
497b47a8 3158#endif
8aa23a47
YO
3159 nxt = regnext(nxt);
3160 if (OP(nxt) != CLOSE)
3161 goto nogo;
3162 if (RExC_open_parens) {
3163 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3164 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3165 }
3166 /* Now we know that nxt2 is the only contents: */
3167 oscan->flags = (U8)ARG(nxt);
3168 OP(oscan) = CURLYN;
3169 OP(nxt1) = NOTHING; /* was OPEN. */
40d049e4 3170
c277df42 3171#ifdef DEBUGGING
8aa23a47
YO
3172 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3173 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3174 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3175 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3176 OP(nxt + 1) = OPTIMIZED; /* was count. */
3177 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
b81d288d 3178#endif
8aa23a47
YO
3179 }
3180 nogo:
3181
3182 /* Try optimization CURLYX => CURLYM. */
3183 if ( OP(oscan) == CURLYX && data
3184 && !(data->flags & SF_HAS_PAR)
3185 && !(data->flags & SF_HAS_EVAL)
3186 && !deltanext /* atom is fixed width */
3187 && minnext != 0 /* CURLYM can't handle zero width */
3188 ) {
3189 /* XXXX How to optimize if data == 0? */
3190 /* Optimize to a simpler form. */
3191 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3192 regnode *nxt2;
3193
3194 OP(oscan) = CURLYM;
3195 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3196 && (OP(nxt2) != WHILEM))
3197 nxt = nxt2;
3198 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3199 /* Need to optimize away parenths. */
3200 if (data->flags & SF_IN_PAR) {
3201 /* Set the parenth number. */
3202 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3203
3204 if (OP(nxt) != CLOSE)
3205 FAIL("Panic opt close");
3206 oscan->flags = (U8)ARG(nxt);
3207 if (RExC_open_parens) {
3208 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3209 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
40d049e4 3210 }
8aa23a47
YO
3211 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3212 OP(nxt) = OPTIMIZED; /* was CLOSE. */
40d049e4 3213
c277df42 3214#ifdef DEBUGGING
8aa23a47
YO
3215 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3216 OP(nxt + 1) = OPTIMIZED; /* was count. */
3217 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3218 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
b81d288d 3219#endif
c277df42 3220#if 0
8aa23a47
YO
3221 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3222 regnode *nnxt = regnext(nxt1);
3223
3224 if (nnxt == nxt) {
3225 if (reg_off_by_arg[OP(nxt1)])
3226 ARG_SET(nxt1, nxt2 - nxt1);
3227 else if (nxt2 - nxt1 < U16_MAX)
3228 NEXT_OFF(nxt1) = nxt2 - nxt1;
3229 else
3230 OP(nxt) = NOTHING; /* Cannot beautify */
c277df42 3231 }
8aa23a47 3232 nxt1 = nnxt;
c277df42 3233 }
5d1c421c 3234#endif
8aa23a47
YO
3235 /* Optimize again: */
3236 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3237 NULL, stopparen, recursed, NULL, 0,depth+1);
3238 }
3239 else
3240 oscan->flags = 0;
3241 }
3242 else if ((OP(oscan) == CURLYX)
3243 && (flags & SCF_WHILEM_VISITED_POS)
3244 /* See the comment on a similar expression above.
3245 However, this time it not a subexpression
3246 we care about, but the expression itself. */
3247 && (maxcount == REG_INFTY)
3248 && data && ++data->whilem_c < 16) {
3249 /* This stays as CURLYX, we can put the count/of pair. */
3250 /* Find WHILEM (as in regexec.c) */
3251 regnode *nxt = oscan + NEXT_OFF(oscan);
3252
3253 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3254 nxt += ARG(nxt);
3255 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3256 | (RExC_whilem_seen << 4)); /* On WHILEM */
3257 }
3258 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3259 pars++;
3260 if (flags & SCF_DO_SUBSTR) {
3261 SV *last_str = NULL;
3262 int counted = mincount != 0;
a0ed51b3 3263
8aa23a47
YO
3264 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3265#if defined(SPARC64_GCC_WORKAROUND)
3266 I32 b = 0;
3267 STRLEN l = 0;
3268 const char *s = NULL;
3269 I32 old = 0;
b515a41d 3270
8aa23a47
YO
3271 if (pos_before >= data->last_start_min)
3272 b = pos_before;
3273 else
3274 b = data->last_start_min;
b515a41d 3275
8aa23a47
YO
3276 l = 0;
3277 s = SvPV_const(data->last_found, l);
3278 old = b - data->last_start_min;
3279
3280#else
3281 I32 b = pos_before >= data->last_start_min
3282 ? pos_before : data->last_start_min;
3283 STRLEN l;
3284 const char * const s = SvPV_const(data->last_found, l);
3285 I32 old = b - data->last_start_min;
3286#endif
3287
3288 if (UTF)
3289 old = utf8_hop((U8*)s, old) - (U8*)s;
3290
3291 l -= old;
3292 /* Get the added string: */
3293 last_str = newSVpvn(s + old, l);
3294 if (UTF)
3295 SvUTF8_on(last_str);
3296 if (deltanext == 0 && pos_before == b) {
3297 /* What was added is a constant string */
3298 if (mincount > 1) {
3299 SvGROW(last_str, (mincount * l) + 1);
3300 repeatcpy(SvPVX(last_str) + l,
3301 SvPVX_const(last_str), l, mincount - 1);
3302 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3303 /* Add additional parts. */
3304 SvCUR_set(data->last_found,
3305 SvCUR(data->last_found) - l);
3306 sv_catsv(data->last_found, last_str);
3307 {
3308 SV * sv = data->last_found;
3309 MAGIC *mg =
3310 SvUTF8(sv) && SvMAGICAL(sv) ?
3311 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3312 if (mg && mg->mg_len >= 0)
bd94e887 3313 mg->mg_len += CHR_SVLEN(last_str) - l;
b515a41d 3314 }
8aa23a47 3315 data->last_end += l * (mincount - 1);
b515a41d 3316 }
8aa23a47
YO
3317 } else {
3318 /* start offset must point into the last copy */
3319 data->last_start_min += minnext * (mincount - 1);
3320 data->last_start_max += is_inf ? I32_MAX
3321 : (maxcount - 1) * (minnext + data->pos_delta);
3322 }
c277df42 3323 }
8aa23a47
YO
3324 /* It is counted once already... */
3325 data->pos_min += minnext * (mincount - counted);
3326 data->pos_delta += - counted * deltanext +
3327 (minnext + deltanext) * maxcount - minnext * mincount;
3328 if (mincount != maxcount) {
3329 /* Cannot extend fixed substrings found inside
3330 the group. */
304ee84b 3331 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3332 if (mincount && last_str) {
3333 SV * const sv = data->last_found;
3334 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3335 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3336
3337 if (mg)
3338 mg->mg_len = -1;
3339 sv_setsv(sv, last_str);
3340 data->last_end = data->pos_min;
3341 data->last_start_min =
3342 data->pos_min - CHR_SVLEN(last_str);
3343 data->last_start_max = is_inf
3344 ? I32_MAX
3345 : data->pos_min + data->pos_delta
3346 - CHR_SVLEN(last_str);
3347 }
3348 data->longest = &(data->longest_float);
3349 }
3350 SvREFCNT_dec(last_str);
c277df42 3351 }
8aa23a47
YO
3352 if (data && (fl & SF_HAS_EVAL))
3353 data->flags |= SF_HAS_EVAL;
3354 optimize_curly_tail:
3355 if (OP(oscan) != CURLYX) {
3356 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3357 && NEXT_OFF(next))
3358 NEXT_OFF(oscan) += NEXT_OFF(next);
3359 }
3360 continue;
3361 default: /* REF and CLUMP only? */
3362 if (flags & SCF_DO_SUBSTR) {
304ee84b 3363 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
8aa23a47
YO
3364 data->longest = &(data->longest_float);
3365 }
3366 is_inf = is_inf_internal = 1;
3367 if (flags & SCF_DO_STCLASS_OR)
3368 cl_anything(pRExC_state, data->start_class);
3369 flags &= ~SCF_DO_STCLASS;
3370 break;
c277df42 3371 }
8aa23a47 3372 }
e1d1eefb
YO
3373 else if (OP(scan) == LNBREAK) {
3374 if (flags & SCF_DO_STCLASS) {
3375 int value = 0;
3376 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3377 if (flags & SCF_DO_STCLASS_AND) {
3378 for (value = 0; value < 256; value++)
e64b1bd1 3379 if (!is_VERTWS_cp(value))
e1d1eefb
YO
3380 ANYOF_BITMAP_CLEAR(data->start_class, value);
3381 }
3382 else {
3383 for (value = 0; value < 256; value++)
e64b1bd1 3384 if (is_VERTWS_cp(value))
e1d1eefb
YO
3385 ANYOF_BITMAP_SET(data->start_class, value);
3386 }
3387 if (flags & SCF_DO_STCLASS_OR)
3388 cl_and(data->start_class, and_withp);
3389 flags &= ~SCF_DO_STCLASS;
3390 }
3391 min += 1;
f9a79580 3392 delta += 1;
e1d1eefb
YO
3393 if (flags & SCF_DO_SUBSTR) {
3394 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3395 data->pos_min += 1;
f9a79580 3396 data->pos_delta += 1;
e1d1eefb
YO
3397 data->longest = &(data->longest_float);
3398 }
3399
3400 }
f9a79580
RGS
3401 else if (OP(scan) == FOLDCHAR) {
3402 int d = ARG(scan)==0xDF ? 1 : 2;
3403 flags &= ~SCF_DO_STCLASS;
3404 min += 1;
3405 delta += d;
3406 if (flags & SCF_DO_SUBSTR) {
3407 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3408 data->pos_min += 1;
3409 data->pos_delta += d;
3410 data->longest = &(data->longest_float);
3411 }
3412 }
8aa23a47
YO
3413 else if (strchr((const char*)PL_simple,OP(scan))) {
3414 int value = 0;
653099ff 3415
8aa23a47 3416 if (flags & SCF_DO_SUBSTR) {
304ee84b 3417 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3418 data->pos_min++;
3419 }
3420 min++;
3421 if (flags & SCF_DO_STCLASS) {
3422 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
b515a41d 3423
8aa23a47
YO
3424 /* Some of the logic below assumes that switching
3425 locale on will only add false positives. */
3426 switch (PL_regkind[OP(scan)]) {
3427 case SANY:
3428 default:
3429 do_default:
3430 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3431 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3432 cl_anything(pRExC_state, data->start_class);
3433 break;
3434 case REG_ANY:
3435 if (OP(scan) == SANY)
3436 goto do_default;
3437 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3438 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3439 || (data->start_class->flags & ANYOF_CLASS));
3440 cl_anything(pRExC_state, data->start_class);
653099ff 3441 }
8aa23a47
YO
3442 if (flags & SCF_DO_STCLASS_AND || !value)
3443 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3444 break;
3445 case ANYOF:
3446 if (flags & SCF_DO_STCLASS_AND)
3447 cl_and(data->start_class,
3448 (struct regnode_charclass_class*)scan);
653099ff 3449 else
8aa23a47
YO
3450 cl_or(pRExC_state, data->start_class,
3451 (struct regnode_charclass_class*)scan);
3452 break;
3453 case ALNUM:
3454 if (flags & SCF_DO_STCLASS_AND) {
3455 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3456 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3457 for (value = 0; value < 256; value++)
3458 if (!isALNUM(value))
3459 ANYOF_BITMAP_CLEAR(data->start_class, value);
3460 }
653099ff 3461 }
8aa23a47
YO
3462 else {
3463 if (data->start_class->flags & ANYOF_LOCALE)
3464 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3465 else {
3466 for (value = 0; value < 256; value++)
3467 if (isALNUM(value))
3468 ANYOF_BITMAP_SET(data->start_class, value);
653099ff 3469 }
8aa23a47
YO
3470 }
3471 break;
3472 case ALNUML:
3473 if (flags & SCF_DO_STCLASS_AND) {
3474 if (data->start_class->flags & ANYOF_LOCALE)
3475 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3476 }
3477 else {
3478 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3479 data->start_class->flags |= ANYOF_LOCALE;
3480 }
3481 break;
3482 case NALNUM:
3483 if (flags & SCF_DO_STCLASS_AND) {
3484 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3485 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3486 for (value = 0; value < 256; value++)
3487 if (isALNUM(value))
3488 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff
GS
3489 }
3490 }
8aa23a47
YO
3491 else {
3492 if (data->start_class->flags & ANYOF_LOCALE)
3493 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3494 else {
3495 for (value = 0; value < 256; value++)
3496 if (!isALNUM(value))
3497 ANYOF_BITMAP_SET(data->start_class, value);
3498 }
653099ff 3499 }
8aa23a47
YO
3500 break;
3501 case NALNUML:
3502 if (flags & SCF_DO_STCLASS_AND) {
3503 if (data->start_class->flags & ANYOF_LOCALE)
3504 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
653099ff 3505 }
8aa23a47
YO
3506 else {
3507 data->start_class->flags |= ANYOF_LOCALE;
3508 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3509 }
3510 break;
3511 case SPACE:
3512 if (flags & SCF_DO_STCLASS_AND) {
3513 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3514 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3515 for (value = 0; value < 256; value++)
3516 if (!isSPACE(value))
3517 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff
GS
3518 }
3519 }
8aa23a47
YO
3520 else {
3521 if (data->start_class->flags & ANYOF_LOCALE)
3522 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3523 else {
3524 for (value = 0; value < 256; value++)
3525 if (isSPACE(value))
3526 ANYOF_BITMAP_SET(data->start_class, value);
3527 }
653099ff 3528 }
8aa23a47
YO
3529 break;
3530 case SPACEL:
3531 if (flags & SCF_DO_STCLASS_AND) {
3532 if (data->start_class->flags & ANYOF_LOCALE)
3533 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3534 }
3535 else {
3536 data->start_class->flags |= ANYOF_LOCALE;
3537 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3538 }
3539 break;
3540 case NSPACE:
3541 if (flags & SCF_DO_STCLASS_AND) {
3542 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3543 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3544 for (value = 0; value < 256; value++)
3545 if (isSPACE(value))
3546 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff 3547 }
8aa23a47
YO
3548 }
3549 else {
3550 if (data->start_class->flags & ANYOF_LOCALE)
3551 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3552 else {
3553 for (value = 0; value < 256; value++)
3554 if (!isSPACE(value))
3555 ANYOF_BITMAP_SET(data->start_class, value);
653099ff
GS
3556 }
3557 }
8aa23a47
YO
3558 break;
3559 case NSPACEL:
3560 if (flags & SCF_DO_STCLASS_AND) {
3561 if (data->start_class->flags & ANYOF_LOCALE) {
3562 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3563 for (value = 0; value < 256; value++)
3564 if (!isSPACE(value))
3565 ANYOF_BITMAP_CLEAR(data->start_class, value);
3566 }
653099ff 3567 }
8aa23a47
YO
3568 else {
3569 data->start_class->flags |= ANYOF_LOCALE;
3570 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3571 }
3572 break;
3573 case DIGIT:
3574 if (flags & SCF_DO_STCLASS_AND) {
3575 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3576 for (value = 0; value < 256; value++)
3577 if (!isDIGIT(value))
3578 ANYOF_BITMAP_CLEAR(data->start_class, value);
3579 }
3580 else {
3581 if (data->start_class->flags & ANYOF_LOCALE)
3582 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3583 else {
3584 for (value = 0; value < 256; value++)
3585 if (isDIGIT(value))
3586 ANYOF_BITMAP_SET(data->start_class, value);
3587 }
3588 }
3589 break;
3590 case NDIGIT:
3591 if (flags & SCF_DO_STCLASS_AND) {
3592 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3593 for (value = 0; value < 256; value++)
3594 if (isDIGIT(value))
3595 ANYOF_BITMAP_CLEAR(data->start_class, value);
3596 }
3597 else {
3598 if (data->start_class->flags & ANYOF_LOCALE)
3599 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3600 else {
3601 for (value = 0; value < 256; value++)
3602 if (!isDIGIT(value))
3603 ANYOF_BITMAP_SET(data->start_class, value);
653099ff
GS
3604 }
3605 }
8aa23a47 3606 break;
e1d1eefb
YO
3607 CASE_SYNST_FNC(VERTWS);
3608 CASE_SYNST_FNC(HORIZWS);
3609
8aa23a47
YO
3610 }
3611 if (flags & SCF_DO_STCLASS_OR)
3612 cl_and(data->start_class, and_withp);
3613 flags &= ~SCF_DO_STCLASS;
3614 }
3615 }
3616 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3617 data->flags |= (OP(scan) == MEOL
3618 ? SF_BEFORE_MEOL
3619 : SF_BEFORE_SEOL);
3620 }
3621 else if ( PL_regkind[OP(scan)] == BRANCHJ
3622 /* Lookbehind, or need to calculate parens/evals/stclass: */
3623 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3624 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3625 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3626 || OP(scan) == UNLESSM )
3627 {
3628 /* Negative Lookahead/lookbehind
3629 In this case we can't do fixed string optimisation.
3630 */
1de06328 3631
8aa23a47
YO
3632 I32 deltanext, minnext, fake = 0;
3633 regnode *nscan;
3634 struct regnode_charclass_class intrnl;
3635 int f = 0;
1de06328 3636
8aa23a47
YO
3637 data_fake.flags = 0;
3638 if (data) {
3639 data_fake.whilem_c = data->whilem_c;
3640 data_fake.last_closep = data->last_closep;
c277df42 3641 }
8aa23a47
YO
3642 else
3643 data_fake.last_closep = &fake;
58e23c8d 3644 data_fake.pos_delta = delta;
8aa23a47
YO
3645 if ( flags & SCF_DO_STCLASS && !scan->flags
3646 && OP(scan) == IFMATCH ) { /* Lookahead */
3647 cl_init(pRExC_state, &intrnl);
3648 data_fake.start_class = &intrnl;
3649 f |= SCF_DO_STCLASS_AND;
3650 }
3651 if (flags & SCF_WHILEM_VISITED_POS)
3652 f |= SCF_WHILEM_VISITED_POS;
3653 next = regnext(scan);
3654 nscan = NEXTOPER(NEXTOPER(scan));
3655 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3656 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3657 if (scan->flags) {
3658 if (deltanext) {
58e23c8d 3659 FAIL("Variable length lookbehind not implemented");
8aa23a47
YO
3660 }
3661 else if (minnext > (I32)U8_MAX) {
58e23c8d 3662 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
8aa23a47
YO
3663 }
3664 scan->flags = (U8)minnext;
3665 }
3666 if (data) {
3667 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3668 pars++;
3669 if (data_fake.flags & SF_HAS_EVAL)
3670 data->flags |= SF_HAS_EVAL;
3671 data->whilem_c = data_fake.whilem_c;
3672 }
3673 if (f & SCF_DO_STCLASS_AND) {
3674 const int was = (data->start_class->flags & ANYOF_EOS);
3675
3676 cl_and(data->start_class, &intrnl);
3677 if (was)
3678 data->start_class->flags |= ANYOF_EOS;
3679 }
cb434fcc 3680 }
8aa23a47
YO
3681#if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3682 else {
3683 /* Positive Lookahead/lookbehind
3684 In this case we can do fixed string optimisation,
3685 but we must be careful about it. Note in the case of
3686 lookbehind the positions will be offset by the minimum
3687 length of the pattern, something we won't know about
3688 until after the recurse.
3689 */
3690 I32 deltanext, fake = 0;
3691 regnode *nscan;
3692 struct regnode_charclass_class intrnl;
3693 int f = 0;
3694 /* We use SAVEFREEPV so that when the full compile
3695 is finished perl will clean up the allocated
3696 minlens when its all done. This was we don't
3697 have to worry about freeing them when we know
3698 they wont be used, which would be a pain.
3699 */
3700 I32 *minnextp;
3701 Newx( minnextp, 1, I32 );
3702 SAVEFREEPV(minnextp);
3703
3704 if (data) {
3705 StructCopy(data, &data_fake, scan_data_t);
3706 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3707 f |= SCF_DO_SUBSTR;
3708 if (scan->flags)
304ee84b 3709 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
8aa23a47
YO
3710 data_fake.last_found=newSVsv(data->last_found);
3711 }
3712 }
3713 else
3714 data_fake.last_closep = &fake;
3715 data_fake.flags = 0;
58e23c8d 3716 data_fake.pos_delta = delta;
8aa23a47
YO
3717 if (is_inf)
3718 data_fake.flags |= SF_IS_INF;
3719 if ( flags & SCF_DO_STCLASS && !scan->flags
3720 && OP(scan) == IFMATCH ) { /* Lookahead */
3721 cl_init(pRExC_state, &intrnl);
3722 data_fake.start_class = &intrnl;
3723 f |= SCF_DO_STCLASS_AND;
3724 }
3725 if (flags & SCF_WHILEM_VISITED_POS)
3726 f |= SCF_WHILEM_VISITED_POS;
3727 next = regnext(scan);
3728 nscan = NEXTOPER(NEXTOPER(scan));
3729
3730 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3731 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3732 if (scan->flags) {
3733 if (deltanext) {
58e23c8d 3734 FAIL("Variable length lookbehind not implemented");
8aa23a47
YO
3735 }
3736 else if (*minnextp > (I32)U8_MAX) {
58e23c8d 3737 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
8aa23a47
YO
3738 }
3739 scan->flags = (U8)*minnextp;
3740 }
3741
3742 *minnextp += min;
3743
3744 if (f & SCF_DO_STCLASS_AND) {
3745 const int was = (data->start_class->flags & ANYOF_EOS);
3746
3747 cl_and(data->start_class, &intrnl);
3748 if (was)
3749 data->start_class->flags |= ANYOF_EOS;
3750 }
3751 if (data) {
3752 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3753 pars++;
3754 if (data_fake.flags & SF_HAS_EVAL)
3755 data->flags |= SF_HAS_EVAL;
3756 data->whilem_c = data_fake.whilem_c;
3757 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3758 if (RExC_rx->minlen<*minnextp)
3759 RExC_rx->minlen=*minnextp;
304ee84b 3760 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
8aa23a47
YO
3761 SvREFCNT_dec(data_fake.last_found);
3762
3763 if ( data_fake.minlen_fixed != minlenp )
3764 {
3765 data->offset_fixed= data_fake.offset_fixed;
3766 data->minlen_fixed= data_fake.minlen_fixed;
3767 data->lookbehind_fixed+= scan->flags;
3768 }
3769 if ( data_fake.minlen_float != minlenp )
3770 {
3771 data->minlen_float= data_fake.minlen_float;
3772 data->offset_float_min=data_fake.offset_float_min;
3773 data->offset_float_max=data_fake.offset_float_max;
3774 data->lookbehind_float+= scan->flags;
3775 }
3776 }
3777 }
3778
3779
40d049e4 3780 }
8aa23a47
YO
3781#endif
3782 }
3783 else if (OP(scan) == OPEN) {
3784 if (stopparen != (I32)ARG(scan))
3785 pars++;
3786 }
3787 else if (OP(scan) == CLOSE) {
3788 if (stopparen == (I32)ARG(scan)) {
3789 break;
3790 }
3791 if ((I32)ARG(scan) == is_par) {
3792 next = regnext(scan);
b515a41d 3793
8aa23a47
YO
3794 if ( next && (OP(next) != WHILEM) && next < last)
3795 is_par = 0; /* Disable optimization */
40d049e4 3796 }
8aa23a47
YO
3797 if (data)
3798 *(data->last_closep) = ARG(scan);
3799 }
3800 else if (OP(scan) == EVAL) {
c277df42
IZ
3801 if (data)
3802 data->flags |= SF_HAS_EVAL;
8aa23a47
YO
3803 }
3804 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3805 if (flags & SCF_DO_SUBSTR) {
304ee84b 3806 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47 3807 flags &= ~SCF_DO_SUBSTR;
40d049e4 3808 }
8aa23a47
YO
3809 if (data && OP(scan)==ACCEPT) {