This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
First class regexps.
[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 { \
73031816
NC
1166 if (UTF) { \
1167 SV *zlopp = newSV(2); \
88c9ea1e
CB
1168 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
1169 unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \
73031816
NC
1170 SvCUR_set(zlopp, kapow - flrbbbbb); \
1171 SvPOK_on(zlopp); \
1172 SvUTF8_on(zlopp); \
1173 av_push(revcharmap, zlopp); \
1174 } else { \
6bdeddd2 1175 char ooooff = (char)uvc; \
73031816
NC
1176 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
1177 } \
1178 } STMT_END
786e8c11
YO
1179
1180#define TRIE_READ_CHAR STMT_START { \
1181 wordlen++; \
1182 if ( UTF ) { \
1183 if ( folder ) { \
1184 if ( foldlen > 0 ) { \
1185 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1186 foldlen -= len; \
1187 scan += len; \
1188 len = 0; \
1189 } else { \
1190 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1191 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1192 foldlen -= UNISKIP( uvc ); \
1193 scan = foldbuf + UNISKIP( uvc ); \
1194 } \
1195 } else { \
1196 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1197 } \
1198 } else { \
1199 uvc = (U32)*uc; \
1200 len = 1; \
1201 } \
1202} STMT_END
1203
1204
1205
1206#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1207 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
f9003953
NC
1208 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1209 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
786e8c11
YO
1210 } \
1211 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1212 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1213 TRIE_LIST_CUR( state )++; \
1214} STMT_END
07be1b83 1215
786e8c11
YO
1216#define TRIE_LIST_NEW(state) STMT_START { \
1217 Newxz( trie->states[ state ].trans.list, \
1218 4, reg_trie_trans_le ); \
1219 TRIE_LIST_CUR( state ) = 1; \
1220 TRIE_LIST_LEN( state ) = 4; \
1221} STMT_END
07be1b83 1222
786e8c11
YO
1223#define TRIE_HANDLE_WORD(state) STMT_START { \
1224 U16 dupe= trie->states[ state ].wordnum; \
1225 regnode * const noper_next = regnext( noper ); \
1226 \
1227 if (trie->wordlen) \
1228 trie->wordlen[ curword ] = wordlen; \
1229 DEBUG_r({ \
1230 /* store the word for dumping */ \
1231 SV* tmp; \
1232 if (OP(noper) != NOTHING) \
1233 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1234 else \
1235 tmp = newSVpvn( "", 0 ); \
1236 if ( UTF ) SvUTF8_on( tmp ); \
2b8b4781 1237 av_push( trie_words, tmp ); \
786e8c11
YO
1238 }); \
1239 \
1240 curword++; \
1241 \
1242 if ( noper_next < tail ) { \
1243 if (!trie->jump) \
c944940b 1244 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
7f69552c 1245 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
1246 if (!jumper) \
1247 jumper = noper_next; \
1248 if (!nextbranch) \
1249 nextbranch= regnext(cur); \
1250 } \
1251 \
1252 if ( dupe ) { \
1253 /* So it's a dupe. This means we need to maintain a */\
1254 /* linked-list from the first to the next. */\
1255 /* we only allocate the nextword buffer when there */\
1256 /* a dupe, so first time we have to do the allocation */\
1257 if (!trie->nextword) \
c944940b 1258 trie->nextword = (U16 *) \
446bd890 1259 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
786e8c11
YO
1260 while ( trie->nextword[dupe] ) \
1261 dupe= trie->nextword[dupe]; \
1262 trie->nextword[dupe]= curword; \
1263 } else { \
1264 /* we haven't inserted this word yet. */ \
1265 trie->states[ state ].wordnum = curword; \
1266 } \
1267} STMT_END
07be1b83 1268
3dab1dad 1269
786e8c11
YO
1270#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1271 ( ( base + charid >= ucharcount \
1272 && base + charid < ubound \
1273 && state == trie->trans[ base - ucharcount + charid ].check \
1274 && trie->trans[ base - ucharcount + charid ].next ) \
1275 ? trie->trans[ base - ucharcount + charid ].next \
1276 : ( state==1 ? special : 0 ) \
1277 )
3dab1dad 1278
786e8c11
YO
1279#define MADE_TRIE 1
1280#define MADE_JUMP_TRIE 2
1281#define MADE_EXACT_TRIE 4
3dab1dad 1282
a3621e74 1283STATIC I32
786e8c11 1284S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
a3621e74 1285{
27da23d5 1286 dVAR;
a3621e74
YO
1287 /* first pass, loop through and scan words */
1288 reg_trie_data *trie;
55eed653 1289 HV *widecharmap = NULL;
2b8b4781 1290 AV *revcharmap = newAV();
a3621e74 1291 regnode *cur;
9f7f3913 1292 const U32 uniflags = UTF8_ALLOW_DEFAULT;
a3621e74
YO
1293 STRLEN len = 0;
1294 UV uvc = 0;
1295 U16 curword = 0;
1296 U32 next_alloc = 0;
786e8c11
YO
1297 regnode *jumper = NULL;
1298 regnode *nextbranch = NULL;
7f69552c 1299 regnode *convert = NULL;
a3621e74 1300 /* we just use folder as a flag in utf8 */
e1ec3a88 1301 const U8 * const folder = ( flags == EXACTF
a3621e74
YO
1302 ? PL_fold
1303 : ( flags == EXACTFL
1304 ? PL_fold_locale
1305 : NULL
1306 )
1307 );
1308
2b8b4781
NC
1309#ifdef DEBUGGING
1310 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1311 AV *trie_words = NULL;
1312 /* along with revcharmap, this only used during construction but both are
1313 * useful during debugging so we store them in the struct when debugging.
8e11feef 1314 */
2b8b4781
NC
1315#else
1316 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
3dab1dad 1317 STRLEN trie_charcount=0;
3dab1dad 1318#endif
2b8b4781 1319 SV *re_trie_maxbuff;
a3621e74 1320 GET_RE_DEBUG_FLAGS_DECL;
72f13be8
YO
1321#ifndef DEBUGGING
1322 PERL_UNUSED_ARG(depth);
1323#endif
a3621e74 1324
c944940b 1325 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 1326 trie->refcount = 1;
3dab1dad 1327 trie->startstate = 1;
786e8c11 1328 trie->wordcount = word_count;
f8fc2ecf 1329 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 1330 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
3dab1dad 1331 if (!(UTF && folder))
c944940b 1332 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
a3621e74 1333 DEBUG_r({
2b8b4781 1334 trie_words = newAV();
a3621e74 1335 });
a3621e74 1336
0111c4fd 1337 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
a3621e74 1338 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 1339 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 1340 }
3dab1dad
YO
1341 DEBUG_OPTIMISE_r({
1342 PerlIO_printf( Perl_debug_log,
786e8c11 1343 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
3dab1dad
YO
1344 (int)depth * 2 + 2, "",
1345 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
786e8c11 1346 REG_NODE_NUM(last), REG_NODE_NUM(tail),
85c3142d 1347 (int)depth);
3dab1dad 1348 });
7f69552c
YO
1349
1350 /* Find the node we are going to overwrite */
1351 if ( first == startbranch && OP( last ) != BRANCH ) {
1352 /* whole branch chain */
1353 convert = first;
1354 } else {
1355 /* branch sub-chain */
1356 convert = NEXTOPER( first );
1357 }
1358
a3621e74
YO
1359 /* -- First loop and Setup --
1360
1361 We first traverse the branches and scan each word to determine if it
1362 contains widechars, and how many unique chars there are, this is
1363 important as we have to build a table with at least as many columns as we
1364 have unique chars.
1365
1366 We use an array of integers to represent the character codes 0..255
38a44b82 1367 (trie->charmap) and we use a an HV* to store Unicode characters. We use the
a3621e74
YO
1368 native representation of the character value as the key and IV's for the
1369 coded index.
1370
1371 *TODO* If we keep track of how many times each character is used we can
1372 remap the columns so that the table compression later on is more
1373 efficient in terms of memory by ensuring most common value is in the
1374 middle and the least common are on the outside. IMO this would be better
1375 than a most to least common mapping as theres a decent chance the most
1376 common letter will share a node with the least common, meaning the node
1377 will not be compressable. With a middle is most common approach the worst
1378 case is when we have the least common nodes twice.
1379
1380 */
1381
a3621e74 1382 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
c445ea15 1383 regnode * const noper = NEXTOPER( cur );
e1ec3a88 1384 const U8 *uc = (U8*)STRING( noper );
a28509cc 1385 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1386 STRLEN foldlen = 0;
1387 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2af232bd 1388 const U8 *scan = (U8*)NULL;
07be1b83 1389 U32 wordlen = 0; /* required init */
02daf0ab
YO
1390 STRLEN chars = 0;
1391 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
a3621e74 1392
3dab1dad
YO
1393 if (OP(noper) == NOTHING) {
1394 trie->minlen= 0;
1395 continue;
1396 }
02daf0ab
YO
1397 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1398 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1399 regardless of encoding */
1400
a3621e74 1401 for ( ; uc < e ; uc += len ) {
3dab1dad 1402 TRIE_CHARCOUNT(trie)++;
a3621e74 1403 TRIE_READ_CHAR;
3dab1dad 1404 chars++;
a3621e74
YO
1405 if ( uvc < 256 ) {
1406 if ( !trie->charmap[ uvc ] ) {
1407 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1408 if ( folder )
1409 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
3dab1dad 1410 TRIE_STORE_REVCHAR;
a3621e74 1411 }
02daf0ab
YO
1412 if ( set_bit ) {
1413 /* store the codepoint in the bitmap, and if its ascii
1414 also store its folded equivelent. */
1415 TRIE_BITMAP_SET(trie,uvc);
0921ee73
T
1416
1417 /* store the folded codepoint */
1418 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1419
1420 if ( !UTF ) {
1421 /* store first byte of utf8 representation of
1422 codepoints in the 127 < uvc < 256 range */
1423 if (127 < uvc && uvc < 192) {
1424 TRIE_BITMAP_SET(trie,194);
1425 } else if (191 < uvc ) {
1426 TRIE_BITMAP_SET(trie,195);
1427 /* && uvc < 256 -- we know uvc is < 256 already */
1428 }
1429 }
02daf0ab
YO
1430 set_bit = 0; /* We've done our bit :-) */
1431 }
a3621e74
YO
1432 } else {
1433 SV** svpp;
55eed653
NC
1434 if ( !widecharmap )
1435 widecharmap = newHV();
a3621e74 1436
55eed653 1437 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
1438
1439 if ( !svpp )
e4584336 1440 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
a3621e74
YO
1441
1442 if ( !SvTRUE( *svpp ) ) {
1443 sv_setiv( *svpp, ++trie->uniquecharcount );
3dab1dad 1444 TRIE_STORE_REVCHAR;
a3621e74
YO
1445 }
1446 }
1447 }
3dab1dad
YO
1448 if( cur == first ) {
1449 trie->minlen=chars;
1450 trie->maxlen=chars;
1451 } else if (chars < trie->minlen) {
1452 trie->minlen=chars;
1453 } else if (chars > trie->maxlen) {
1454 trie->maxlen=chars;
1455 }
1456
a3621e74
YO
1457 } /* end first pass */
1458 DEBUG_TRIE_COMPILE_r(
3dab1dad
YO
1459 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1460 (int)depth * 2 + 2,"",
55eed653 1461 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
1462 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1463 (int)trie->minlen, (int)trie->maxlen )
a3621e74 1464 );
c944940b 1465 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
a3621e74
YO
1466
1467 /*
1468 We now know what we are dealing with in terms of unique chars and
1469 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
1470 representation using a flat table will take. If it's over a reasonable
1471 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
1472 conservative but potentially much slower representation using an array
1473 of lists.
1474
1475 At the end we convert both representations into the same compressed
1476 form that will be used in regexec.c for matching with. The latter
1477 is a form that cannot be used to construct with but has memory
1478 properties similar to the list form and access properties similar
1479 to the table form making it both suitable for fast searches and
1480 small enough that its feasable to store for the duration of a program.
1481
1482 See the comment in the code where the compressed table is produced
1483 inplace from the flat tabe representation for an explanation of how
1484 the compression works.
1485
1486 */
1487
1488
3dab1dad 1489 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
a3621e74
YO
1490 /*
1491 Second Pass -- Array Of Lists Representation
1492
1493 Each state will be represented by a list of charid:state records
1494 (reg_trie_trans_le) the first such element holds the CUR and LEN
1495 points of the allocated array. (See defines above).
1496
1497 We build the initial structure using the lists, and then convert
1498 it into the compressed table form which allows faster lookups
1499 (but cant be modified once converted).
a3621e74
YO
1500 */
1501
a3621e74
YO
1502 STRLEN transcount = 1;
1503
1e2e3d02
YO
1504 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1505 "%*sCompiling trie using list compiler\n",
1506 (int)depth * 2 + 2, ""));
446bd890 1507
c944940b
JH
1508 trie->states = (reg_trie_state *)
1509 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1510 sizeof(reg_trie_state) );
a3621e74
YO
1511 TRIE_LIST_NEW(1);
1512 next_alloc = 2;
1513
1514 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1515
c445ea15
AL
1516 regnode * const noper = NEXTOPER( cur );
1517 U8 *uc = (U8*)STRING( noper );
1518 const U8 * const e = uc + STR_LEN( noper );
1519 U32 state = 1; /* required init */
1520 U16 charid = 0; /* sanity init */
1521 U8 *scan = (U8*)NULL; /* sanity init */
1522 STRLEN foldlen = 0; /* required init */
07be1b83 1523 U32 wordlen = 0; /* required init */
c445ea15
AL
1524 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1525
3dab1dad 1526 if (OP(noper) != NOTHING) {
786e8c11 1527 for ( ; uc < e ; uc += len ) {
c445ea15 1528
786e8c11 1529 TRIE_READ_CHAR;
c445ea15 1530
786e8c11
YO
1531 if ( uvc < 256 ) {
1532 charid = trie->charmap[ uvc ];
c445ea15 1533 } else {
55eed653 1534 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11
YO
1535 if ( !svpp ) {
1536 charid = 0;
1537 } else {
1538 charid=(U16)SvIV( *svpp );
1539 }
c445ea15 1540 }
786e8c11
YO
1541 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1542 if ( charid ) {
a3621e74 1543
786e8c11
YO
1544 U16 check;
1545 U32 newstate = 0;
a3621e74 1546
786e8c11
YO
1547 charid--;
1548 if ( !trie->states[ state ].trans.list ) {
1549 TRIE_LIST_NEW( state );
c445ea15 1550 }
786e8c11
YO
1551 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1552 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1553 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1554 break;
1555 }
1556 }
1557 if ( ! newstate ) {
1558 newstate = next_alloc++;
1559 TRIE_LIST_PUSH( state, charid, newstate );
1560 transcount++;
1561 }
1562 state = newstate;
1563 } else {
1564 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
c445ea15 1565 }
a28509cc 1566 }
c445ea15 1567 }
3dab1dad 1568 TRIE_HANDLE_WORD(state);
a3621e74
YO
1569
1570 } /* end second pass */
1571
1e2e3d02
YO
1572 /* next alloc is the NEXT state to be allocated */
1573 trie->statecount = next_alloc;
c944940b
JH
1574 trie->states = (reg_trie_state *)
1575 PerlMemShared_realloc( trie->states,
1576 next_alloc
1577 * sizeof(reg_trie_state) );
a3621e74 1578
3dab1dad 1579 /* and now dump it out before we compress it */
2b8b4781
NC
1580 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1581 revcharmap, next_alloc,
1582 depth+1)
1e2e3d02 1583 );
a3621e74 1584
c944940b
JH
1585 trie->trans = (reg_trie_trans *)
1586 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
1587 {
1588 U32 state;
a3621e74
YO
1589 U32 tp = 0;
1590 U32 zp = 0;
1591
1592
1593 for( state=1 ; state < next_alloc ; state ++ ) {
1594 U32 base=0;
1595
1596 /*
1597 DEBUG_TRIE_COMPILE_MORE_r(
1598 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1599 );
1600 */
1601
1602 if (trie->states[state].trans.list) {
1603 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1604 U16 maxid=minid;
a28509cc 1605 U16 idx;
a3621e74
YO
1606
1607 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
1608 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1609 if ( forid < minid ) {
1610 minid=forid;
1611 } else if ( forid > maxid ) {
1612 maxid=forid;
1613 }
a3621e74
YO
1614 }
1615 if ( transcount < tp + maxid - minid + 1) {
1616 transcount *= 2;
c944940b
JH
1617 trie->trans = (reg_trie_trans *)
1618 PerlMemShared_realloc( trie->trans,
446bd890
NC
1619 transcount
1620 * sizeof(reg_trie_trans) );
a3621e74
YO
1621 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1622 }
1623 base = trie->uniquecharcount + tp - minid;
1624 if ( maxid == minid ) {
1625 U32 set = 0;
1626 for ( ; zp < tp ; zp++ ) {
1627 if ( ! trie->trans[ zp ].next ) {
1628 base = trie->uniquecharcount + zp - minid;
1629 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1630 trie->trans[ zp ].check = state;
1631 set = 1;
1632 break;
1633 }
1634 }
1635 if ( !set ) {
1636 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1637 trie->trans[ tp ].check = state;
1638 tp++;
1639 zp = tp;
1640 }
1641 } else {
1642 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15 1643 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
a3621e74
YO
1644 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1645 trie->trans[ tid ].check = state;
1646 }
1647 tp += ( maxid - minid + 1 );
1648 }
1649 Safefree(trie->states[ state ].trans.list);
1650 }
1651 /*
1652 DEBUG_TRIE_COMPILE_MORE_r(
1653 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1654 );
1655 */
1656 trie->states[ state ].trans.base=base;
1657 }
cc601c31 1658 trie->lasttrans = tp + 1;
a3621e74
YO
1659 }
1660 } else {
1661 /*
1662 Second Pass -- Flat Table Representation.
1663
1664 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1665 We know that we will need Charcount+1 trans at most to store the data
1666 (one row per char at worst case) So we preallocate both structures
1667 assuming worst case.
1668
1669 We then construct the trie using only the .next slots of the entry
1670 structs.
1671
1672 We use the .check field of the first entry of the node temporarily to
1673 make compression both faster and easier by keeping track of how many non
1674 zero fields are in the node.
1675
1676 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1677 transition.
1678
1679 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1680 number representing the first entry of the node, and state as a
1681 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1682 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1683 are 2 entrys per node. eg:
1684
1685 A B A B
1686 1. 2 4 1. 3 7
1687 2. 0 3 3. 0 5
1688 3. 0 0 5. 0 0
1689 4. 0 0 7. 0 0
1690
1691 The table is internally in the right hand, idx form. However as we also
1692 have to deal with the states array which is indexed by nodenum we have to
1693 use TRIE_NODENUM() to convert.
1694
1695 */
1e2e3d02
YO
1696 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1697 "%*sCompiling trie using table compiler\n",
1698 (int)depth * 2 + 2, ""));
3dab1dad 1699
c944940b
JH
1700 trie->trans = (reg_trie_trans *)
1701 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1702 * trie->uniquecharcount + 1,
1703 sizeof(reg_trie_trans) );
1704 trie->states = (reg_trie_state *)
1705 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1706 sizeof(reg_trie_state) );
a3621e74
YO
1707 next_alloc = trie->uniquecharcount + 1;
1708
3dab1dad 1709
a3621e74
YO
1710 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1711
c445ea15 1712 regnode * const noper = NEXTOPER( cur );
a28509cc
AL
1713 const U8 *uc = (U8*)STRING( noper );
1714 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1715
1716 U32 state = 1; /* required init */
1717
1718 U16 charid = 0; /* sanity init */
1719 U32 accept_state = 0; /* sanity init */
1720 U8 *scan = (U8*)NULL; /* sanity init */
1721
1722 STRLEN foldlen = 0; /* required init */
07be1b83 1723 U32 wordlen = 0; /* required init */
a3621e74
YO
1724 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1725
3dab1dad 1726 if ( OP(noper) != NOTHING ) {
786e8c11 1727 for ( ; uc < e ; uc += len ) {
a3621e74 1728
786e8c11 1729 TRIE_READ_CHAR;
a3621e74 1730
786e8c11
YO
1731 if ( uvc < 256 ) {
1732 charid = trie->charmap[ uvc ];
1733 } else {
55eed653 1734 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11 1735 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 1736 }
786e8c11
YO
1737 if ( charid ) {
1738 charid--;
1739 if ( !trie->trans[ state + charid ].next ) {
1740 trie->trans[ state + charid ].next = next_alloc;
1741 trie->trans[ state ].check++;
1742 next_alloc += trie->uniquecharcount;
1743 }
1744 state = trie->trans[ state + charid ].next;
1745 } else {
1746 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1747 }
1748 /* charid is now 0 if we dont know the char read, or nonzero if we do */
a3621e74 1749 }
a3621e74 1750 }
3dab1dad
YO
1751 accept_state = TRIE_NODENUM( state );
1752 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
1753
1754 } /* end second pass */
1755
3dab1dad 1756 /* and now dump it out before we compress it */
2b8b4781
NC
1757 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1758 revcharmap,
1759 next_alloc, depth+1));
a3621e74 1760
a3621e74
YO
1761 {
1762 /*
1763 * Inplace compress the table.*
1764
1765 For sparse data sets the table constructed by the trie algorithm will
1766 be mostly 0/FAIL transitions or to put it another way mostly empty.
1767 (Note that leaf nodes will not contain any transitions.)
1768
1769 This algorithm compresses the tables by eliminating most such
1770 transitions, at the cost of a modest bit of extra work during lookup:
1771
1772 - Each states[] entry contains a .base field which indicates the
1773 index in the state[] array wheres its transition data is stored.
1774
1775 - If .base is 0 there are no valid transitions from that node.
1776
1777 - If .base is nonzero then charid is added to it to find an entry in
1778 the trans array.
1779
1780 -If trans[states[state].base+charid].check!=state then the
1781 transition is taken to be a 0/Fail transition. Thus if there are fail
1782 transitions at the front of the node then the .base offset will point
1783 somewhere inside the previous nodes data (or maybe even into a node
1784 even earlier), but the .check field determines if the transition is
1785 valid.
1786
786e8c11 1787 XXX - wrong maybe?
a3621e74
YO
1788 The following process inplace converts the table to the compressed
1789 table: We first do not compress the root node 1,and mark its all its
1790 .check pointers as 1 and set its .base pointer as 1 as well. This
1791 allows to do a DFA construction from the compressed table later, and
1792 ensures that any .base pointers we calculate later are greater than
1793 0.
1794
1795 - We set 'pos' to indicate the first entry of the second node.
1796
1797 - We then iterate over the columns of the node, finding the first and
1798 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1799 and set the .check pointers accordingly, and advance pos
1800 appropriately and repreat for the next node. Note that when we copy
1801 the next pointers we have to convert them from the original
1802 NODEIDX form to NODENUM form as the former is not valid post
1803 compression.
1804
1805 - If a node has no transitions used we mark its base as 0 and do not
1806 advance the pos pointer.
1807
1808 - If a node only has one transition we use a second pointer into the
1809 structure to fill in allocated fail transitions from other states.
1810 This pointer is independent of the main pointer and scans forward
1811 looking for null transitions that are allocated to a state. When it
1812 finds one it writes the single transition into the "hole". If the
786e8c11 1813 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
1814
1815 - Once compressed we can Renew/realloc the structures to release the
1816 excess space.
1817
1818 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1819 specifically Fig 3.47 and the associated pseudocode.
1820
1821 demq
1822 */
a3b680e6 1823 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 1824 U32 state, charid;
a3621e74 1825 U32 pos = 0, zp=0;
1e2e3d02 1826 trie->statecount = laststate;
a3621e74
YO
1827
1828 for ( state = 1 ; state < laststate ; state++ ) {
1829 U8 flag = 0;
a28509cc
AL
1830 const U32 stateidx = TRIE_NODEIDX( state );
1831 const U32 o_used = trie->trans[ stateidx ].check;
1832 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
1833 trie->trans[ stateidx ].check = 0;
1834
1835 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1836 if ( flag || trie->trans[ stateidx + charid ].next ) {
1837 if ( trie->trans[ stateidx + charid ].next ) {
1838 if (o_used == 1) {
1839 for ( ; zp < pos ; zp++ ) {
1840 if ( ! trie->trans[ zp ].next ) {
1841 break;
1842 }
1843 }
1844 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1845 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1846 trie->trans[ zp ].check = state;
1847 if ( ++zp > pos ) pos = zp;
1848 break;
1849 }
1850 used--;
1851 }
1852 if ( !flag ) {
1853 flag = 1;
1854 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1855 }
1856 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1857 trie->trans[ pos ].check = state;
1858 pos++;
1859 }
1860 }
1861 }
cc601c31 1862 trie->lasttrans = pos + 1;
c944940b
JH
1863 trie->states = (reg_trie_state *)
1864 PerlMemShared_realloc( trie->states, laststate
1865 * sizeof(reg_trie_state) );
a3621e74 1866 DEBUG_TRIE_COMPILE_MORE_r(
e4584336 1867 PerlIO_printf( Perl_debug_log,
3dab1dad
YO
1868 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1869 (int)depth * 2 + 2,"",
1870 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
5d7488b2
AL
1871 (IV)next_alloc,
1872 (IV)pos,
a3621e74
YO
1873 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1874 );
1875
1876 } /* end table compress */
1877 }
1e2e3d02
YO
1878 DEBUG_TRIE_COMPILE_MORE_r(
1879 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1880 (int)depth * 2 + 2, "",
1881 (UV)trie->statecount,
1882 (UV)trie->lasttrans)
1883 );
cc601c31 1884 /* resize the trans array to remove unused space */
c944940b
JH
1885 trie->trans = (reg_trie_trans *)
1886 PerlMemShared_realloc( trie->trans, trie->lasttrans
1887 * sizeof(reg_trie_trans) );
a3621e74 1888
3dab1dad 1889 /* and now dump out the compressed format */
2b8b4781 1890 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
07be1b83 1891
3dab1dad 1892 { /* Modify the program and insert the new TRIE node*/
3dab1dad
YO
1893 U8 nodetype =(U8)(flags & 0xFF);
1894 char *str=NULL;
786e8c11 1895
07be1b83 1896#ifdef DEBUGGING
e62cc96a 1897 regnode *optimize = NULL;
7122b237
YO
1898#ifdef RE_TRACK_PATTERN_OFFSETS
1899
b57a0404
JH
1900 U32 mjd_offset = 0;
1901 U32 mjd_nodelen = 0;
7122b237
YO
1902#endif /* RE_TRACK_PATTERN_OFFSETS */
1903#endif /* DEBUGGING */
a3621e74 1904 /*
3dab1dad
YO
1905 This means we convert either the first branch or the first Exact,
1906 depending on whether the thing following (in 'last') is a branch
1907 or not and whther first is the startbranch (ie is it a sub part of
1908 the alternation or is it the whole thing.)
1909 Assuming its a sub part we conver the EXACT otherwise we convert
1910 the whole branch sequence, including the first.
a3621e74 1911 */
3dab1dad 1912 /* Find the node we are going to overwrite */
7f69552c 1913 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 1914 /* branch sub-chain */
3dab1dad 1915 NEXT_OFF( first ) = (U16)(last - first);
7122b237 1916#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
1917 DEBUG_r({
1918 mjd_offset= Node_Offset((convert));
1919 mjd_nodelen= Node_Length((convert));
1920 });
7122b237 1921#endif
7f69552c 1922 /* whole branch chain */
7122b237
YO
1923 }
1924#ifdef RE_TRACK_PATTERN_OFFSETS
1925 else {
7f69552c
YO
1926 DEBUG_r({
1927 const regnode *nop = NEXTOPER( convert );
1928 mjd_offset= Node_Offset((nop));
1929 mjd_nodelen= Node_Length((nop));
1930 });
07be1b83
YO
1931 }
1932 DEBUG_OPTIMISE_r(
1933 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1934 (int)depth * 2 + 2, "",
786e8c11 1935 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 1936 );
7122b237 1937#endif
3dab1dad
YO
1938 /* But first we check to see if there is a common prefix we can
1939 split out as an EXACT and put in front of the TRIE node. */
1940 trie->startstate= 1;
55eed653 1941 if ( trie->bitmap && !widecharmap && !trie->jump ) {
3dab1dad 1942 U32 state;
1e2e3d02 1943 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 1944 U32 ofs = 0;
8e11feef
RGS
1945 I32 idx = -1;
1946 U32 count = 0;
1947 const U32 base = trie->states[ state ].trans.base;
a3621e74 1948
3dab1dad 1949 if ( trie->states[state].wordnum )
8e11feef 1950 count = 1;
a3621e74 1951
8e11feef 1952 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
cc601c31
YO
1953 if ( ( base + ofs >= trie->uniquecharcount ) &&
1954 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
a3621e74
YO
1955 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1956 {
3dab1dad 1957 if ( ++count > 1 ) {
2b8b4781 1958 SV **tmp = av_fetch( revcharmap, ofs, 0);
07be1b83 1959 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 1960 if ( state == 1 ) break;
3dab1dad
YO
1961 if ( count == 2 ) {
1962 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1963 DEBUG_OPTIMISE_r(
8e11feef
RGS
1964 PerlIO_printf(Perl_debug_log,
1965 "%*sNew Start State=%"UVuf" Class: [",
1966 (int)depth * 2 + 2, "",
786e8c11 1967 (UV)state));
be8e71aa 1968 if (idx >= 0) {
2b8b4781 1969 SV ** const tmp = av_fetch( revcharmap, idx, 0);
be8e71aa 1970 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 1971
3dab1dad 1972 TRIE_BITMAP_SET(trie,*ch);
8e11feef
RGS
1973 if ( folder )
1974 TRIE_BITMAP_SET(trie, folder[ *ch ]);
3dab1dad 1975 DEBUG_OPTIMISE_r(
07be1b83 1976 PerlIO_printf(Perl_debug_log, (char*)ch)
3dab1dad 1977 );
8e11feef
RGS
1978 }
1979 }
1980 TRIE_BITMAP_SET(trie,*ch);
1981 if ( folder )
1982 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1983 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1984 }
1985 idx = ofs;
1986 }
3dab1dad
YO
1987 }
1988 if ( count == 1 ) {
2b8b4781 1989 SV **tmp = av_fetch( revcharmap, idx, 0);
c490c714
YO
1990 STRLEN len;
1991 char *ch = SvPV( *tmp, len );
de734bd5
A
1992 DEBUG_OPTIMISE_r({
1993 SV *sv=sv_newmortal();
8e11feef
RGS
1994 PerlIO_printf( Perl_debug_log,
1995 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1996 (int)depth * 2 + 2, "",
de734bd5
A
1997 (UV)state, (UV)idx,
1998 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1999 PL_colors[0], PL_colors[1],
2000 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
2001 PERL_PV_ESCAPE_FIRSTCHAR
2002 )
2003 );
2004 });
3dab1dad
YO
2005 if ( state==1 ) {
2006 OP( convert ) = nodetype;
2007 str=STRING(convert);
2008 STR_LEN(convert)=0;
2009 }
c490c714
YO
2010 STR_LEN(convert) += len;
2011 while (len--)
de734bd5 2012 *str++ = *ch++;
8e11feef 2013 } else {
f9049ba1 2014#ifdef DEBUGGING
8e11feef
RGS
2015 if (state>1)
2016 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
f9049ba1 2017#endif
8e11feef
RGS
2018 break;
2019 }
2020 }
3dab1dad 2021 if (str) {
8e11feef 2022 regnode *n = convert+NODE_SZ_STR(convert);
07be1b83 2023 NEXT_OFF(convert) = NODE_SZ_STR(convert);
8e11feef 2024 trie->startstate = state;
07be1b83
YO
2025 trie->minlen -= (state - 1);
2026 trie->maxlen -= (state - 1);
33809eae
JH
2027#ifdef DEBUGGING
2028 /* At least the UNICOS C compiler choked on this
2029 * being argument to DEBUG_r(), so let's just have
2030 * it right here. */
2031 if (
2032#ifdef PERL_EXT_RE_BUILD
2033 1
2034#else
2035 DEBUG_r_TEST
2036#endif
2037 ) {
2038 regnode *fix = convert;
2039 U32 word = trie->wordcount;
2040 mjd_nodelen++;
2041 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2042 while( ++fix < n ) {
2043 Set_Node_Offset_Length(fix, 0, 0);
2044 }
2045 while (word--) {
2046 SV ** const tmp = av_fetch( trie_words, word, 0 );
2047 if (tmp) {
2048 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2049 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2050 else
2051 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2052 }
2053 }
2054 }
2055#endif
8e11feef
RGS
2056 if (trie->maxlen) {
2057 convert = n;
2058 } else {
3dab1dad 2059 NEXT_OFF(convert) = (U16)(tail - convert);
a5ca303d 2060 DEBUG_r(optimize= n);
3dab1dad
YO
2061 }
2062 }
2063 }
a5ca303d
YO
2064 if (!jumper)
2065 jumper = last;
3dab1dad 2066 if ( trie->maxlen ) {
8e11feef
RGS
2067 NEXT_OFF( convert ) = (U16)(tail - convert);
2068 ARG_SET( convert, data_slot );
786e8c11
YO
2069 /* Store the offset to the first unabsorbed branch in
2070 jump[0], which is otherwise unused by the jump logic.
2071 We use this when dumping a trie and during optimisation. */
2072 if (trie->jump)
7f69552c 2073 trie->jump[0] = (U16)(nextbranch - convert);
a5ca303d 2074
786e8c11
YO
2075 /* XXXX */
2076 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
1de06328 2077 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
786e8c11
YO
2078 {
2079 OP( convert ) = TRIEC;
2080 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
446bd890 2081 PerlMemShared_free(trie->bitmap);
786e8c11
YO
2082 trie->bitmap= NULL;
2083 } else
2084 OP( convert ) = TRIE;
a3621e74 2085
3dab1dad
YO
2086 /* store the type in the flags */
2087 convert->flags = nodetype;
a5ca303d
YO
2088 DEBUG_r({
2089 optimize = convert
2090 + NODE_STEP_REGNODE
2091 + regarglen[ OP( convert ) ];
2092 });
2093 /* XXX We really should free up the resource in trie now,
2094 as we won't use them - (which resources?) dmq */
3dab1dad 2095 }
a3621e74 2096 /* needed for dumping*/
e62cc96a 2097 DEBUG_r(if (optimize) {
07be1b83 2098 regnode *opt = convert;
bcdf7404 2099
e62cc96a 2100 while ( ++opt < optimize) {
07be1b83
YO
2101 Set_Node_Offset_Length(opt,0,0);
2102 }
786e8c11
YO
2103 /*
2104 Try to clean up some of the debris left after the
2105 optimisation.
a3621e74 2106 */
786e8c11 2107 while( optimize < jumper ) {
07be1b83 2108 mjd_nodelen += Node_Length((optimize));
a3621e74 2109 OP( optimize ) = OPTIMIZED;
07be1b83 2110 Set_Node_Offset_Length(optimize,0,0);
a3621e74
YO
2111 optimize++;
2112 }
07be1b83 2113 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
a3621e74
YO
2114 });
2115 } /* end node insert */
55eed653 2116 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2b8b4781
NC
2117#ifdef DEBUGGING
2118 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2119 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2120#else
2121 SvREFCNT_dec(revcharmap);
07be1b83 2122#endif
786e8c11
YO
2123 return trie->jump
2124 ? MADE_JUMP_TRIE
2125 : trie->startstate>1
2126 ? MADE_EXACT_TRIE
2127 : MADE_TRIE;
2128}
2129
2130STATIC void
2131S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2132{
2133/* The Trie is constructed and compressed now so we can build a fail array now if its needed
2134
2135 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2136 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2137 ISBN 0-201-10088-6
2138
2139 We find the fail state for each state in the trie, this state is the longest proper
2140 suffix of the current states 'word' that is also a proper prefix of another word in our
2141 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2142 the DFA not to have to restart after its tried and failed a word at a given point, it
2143 simply continues as though it had been matching the other word in the first place.
2144 Consider
2145 'abcdgu'=~/abcdefg|cdgu/
2146 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2147 fail, which would bring use to the state representing 'd' in the second word where we would
2148 try 'g' and succeed, prodceding to match 'cdgu'.
2149 */
2150 /* add a fail transition */
3251b653
NC
2151 const U32 trie_offset = ARG(source);
2152 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
786e8c11
YO
2153 U32 *q;
2154 const U32 ucharcount = trie->uniquecharcount;
1e2e3d02 2155 const U32 numstates = trie->statecount;
786e8c11
YO
2156 const U32 ubound = trie->lasttrans + ucharcount;
2157 U32 q_read = 0;
2158 U32 q_write = 0;
2159 U32 charid;
2160 U32 base = trie->states[ 1 ].trans.base;
2161 U32 *fail;
2162 reg_ac_data *aho;
2163 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2164 GET_RE_DEBUG_FLAGS_DECL;
2165#ifndef DEBUGGING
2166 PERL_UNUSED_ARG(depth);
2167#endif
2168
2169
2170 ARG_SET( stclass, data_slot );
c944940b 2171 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
f8fc2ecf 2172 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3251b653 2173 aho->trie=trie_offset;
446bd890
NC
2174 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2175 Copy( trie->states, aho->states, numstates, reg_trie_state );
786e8c11 2176 Newxz( q, numstates, U32);
c944940b 2177 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
786e8c11
YO
2178 aho->refcount = 1;
2179 fail = aho->fail;
2180 /* initialize fail[0..1] to be 1 so that we always have
2181 a valid final fail state */
2182 fail[ 0 ] = fail[ 1 ] = 1;
2183
2184 for ( charid = 0; charid < ucharcount ; charid++ ) {
2185 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2186 if ( newstate ) {
2187 q[ q_write ] = newstate;
2188 /* set to point at the root */
2189 fail[ q[ q_write++ ] ]=1;
2190 }
2191 }
2192 while ( q_read < q_write) {
2193 const U32 cur = q[ q_read++ % numstates ];
2194 base = trie->states[ cur ].trans.base;
2195
2196 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2197 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2198 if (ch_state) {
2199 U32 fail_state = cur;
2200 U32 fail_base;
2201 do {
2202 fail_state = fail[ fail_state ];
2203 fail_base = aho->states[ fail_state ].trans.base;
2204 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2205
2206 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2207 fail[ ch_state ] = fail_state;
2208 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2209 {
2210 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2211 }
2212 q[ q_write++ % numstates] = ch_state;
2213 }
2214 }
2215 }
2216 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2217 when we fail in state 1, this allows us to use the
2218 charclass scan to find a valid start char. This is based on the principle
2219 that theres a good chance the string being searched contains lots of stuff
2220 that cant be a start char.
2221 */
2222 fail[ 0 ] = fail[ 1 ] = 0;
2223 DEBUG_TRIE_COMPILE_r({
6d99fb9b
JH
2224 PerlIO_printf(Perl_debug_log,
2225 "%*sStclass Failtable (%"UVuf" states): 0",
2226 (int)(depth * 2), "", (UV)numstates
1e2e3d02 2227 );
786e8c11
YO
2228 for( q_read=1; q_read<numstates; q_read++ ) {
2229 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2230 }
2231 PerlIO_printf(Perl_debug_log, "\n");
2232 });
2233 Safefree(q);
2234 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
a3621e74
YO
2235}
2236
786e8c11 2237
a3621e74 2238/*
5d1c421c
JH
2239 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2240 * These need to be revisited when a newer toolchain becomes available.
2241 */
2242#if defined(__sparc64__) && defined(__GNUC__)
2243# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2244# undef SPARC64_GCC_WORKAROUND
2245# define SPARC64_GCC_WORKAROUND 1
2246# endif
2247#endif
2248
07be1b83 2249#define DEBUG_PEEP(str,scan,depth) \
b515a41d 2250 DEBUG_OPTIMISE_r({if (scan){ \
07be1b83
YO
2251 SV * const mysv=sv_newmortal(); \
2252 regnode *Next = regnext(scan); \
2253 regprop(RExC_rx, mysv, scan); \
7f69552c 2254 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
07be1b83
YO
2255 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2256 Next ? (REG_NODE_NUM(Next)) : 0 ); \
b515a41d 2257 }});
07be1b83 2258
1de06328
YO
2259
2260
2261
2262
07be1b83
YO
2263#define JOIN_EXACT(scan,min,flags) \
2264 if (PL_regkind[OP(scan)] == EXACT) \
2265 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2266
be8e71aa 2267STATIC U32
07be1b83
YO
2268S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2269 /* Merge several consecutive EXACTish nodes into one. */
2270 regnode *n = regnext(scan);
2271 U32 stringok = 1;
2272 regnode *next = scan + NODE_SZ_STR(scan);
2273 U32 merged = 0;
2274 U32 stopnow = 0;
2275#ifdef DEBUGGING
2276 regnode *stop = scan;
72f13be8 2277 GET_RE_DEBUG_FLAGS_DECL;
f9049ba1 2278#else
d47053eb
RGS
2279 PERL_UNUSED_ARG(depth);
2280#endif
2281#ifndef EXPERIMENTAL_INPLACESCAN
f9049ba1
SP
2282 PERL_UNUSED_ARG(flags);
2283 PERL_UNUSED_ARG(val);
07be1b83 2284#endif
07be1b83
YO
2285 DEBUG_PEEP("join",scan,depth);
2286
2287 /* Skip NOTHING, merge EXACT*. */
2288 while (n &&
2289 ( PL_regkind[OP(n)] == NOTHING ||
2290 (stringok && (OP(n) == OP(scan))))
2291 && NEXT_OFF(n)
2292 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2293
2294 if (OP(n) == TAIL || n > next)
2295 stringok = 0;
2296 if (PL_regkind[OP(n)] == NOTHING) {
07be1b83
YO
2297 DEBUG_PEEP("skip:",n,depth);
2298 NEXT_OFF(scan) += NEXT_OFF(n);
2299 next = n + NODE_STEP_REGNODE;
2300#ifdef DEBUGGING
2301 if (stringok)
2302 stop = n;
2303#endif
2304 n = regnext(n);
2305 }
2306 else if (stringok) {
786e8c11 2307 const unsigned int oldl = STR_LEN(scan);
07be1b83
YO
2308 regnode * const nnext = regnext(n);
2309
2310 DEBUG_PEEP("merg",n,depth);
2311
2312 merged++;
2313 if (oldl + STR_LEN(n) > U8_MAX)
2314 break;
2315 NEXT_OFF(scan) += NEXT_OFF(n);
2316 STR_LEN(scan) += STR_LEN(n);
2317 next = n + NODE_SZ_STR(n);
2318 /* Now we can overwrite *n : */
2319 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2320#ifdef DEBUGGING
2321 stop = next - 1;
2322#endif
2323 n = nnext;
2324 if (stopnow) break;
2325 }
2326
d47053eb
RGS
2327#ifdef EXPERIMENTAL_INPLACESCAN
2328 if (flags && !NEXT_OFF(n)) {
2329 DEBUG_PEEP("atch", val, depth);
2330 if (reg_off_by_arg[OP(n)]) {
2331 ARG_SET(n, val - n);
2332 }
2333 else {
2334 NEXT_OFF(n) = val - n;
2335 }
2336 stopnow = 1;
2337 }
07be1b83
YO
2338#endif
2339 }
2340
2341 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2342 /*
2343 Two problematic code points in Unicode casefolding of EXACT nodes:
2344
2345 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2346 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2347
2348 which casefold to
2349
2350 Unicode UTF-8
2351
2352 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2353 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2354
2355 This means that in case-insensitive matching (or "loose matching",
2356 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2357 length of the above casefolded versions) can match a target string
2358 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2359 This would rather mess up the minimum length computation.
2360
2361 What we'll do is to look for the tail four bytes, and then peek
2362 at the preceding two bytes to see whether we need to decrease
2363 the minimum length by four (six minus two).
2364
2365 Thanks to the design of UTF-8, there cannot be false matches:
2366 A sequence of valid UTF-8 bytes cannot be a subsequence of
2367 another valid sequence of UTF-8 bytes.
2368
2369 */
2370 char * const s0 = STRING(scan), *s, *t;
2371 char * const s1 = s0 + STR_LEN(scan) - 1;
2372 char * const s2 = s1 - 4;
e294cc5d
JH
2373#ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2374 const char t0[] = "\xaf\x49\xaf\x42";
2375#else
07be1b83 2376 const char t0[] = "\xcc\x88\xcc\x81";
e294cc5d 2377#endif
07be1b83
YO
2378 const char * const t1 = t0 + 3;
2379
2380 for (s = s0 + 2;
2381 s < s2 && (t = ninstr(s, s1, t0, t1));
2382 s = t + 4) {
e294cc5d
JH
2383#ifdef EBCDIC
2384 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2385 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2386#else
07be1b83
YO
2387 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2388 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
e294cc5d 2389#endif
07be1b83
YO
2390 *min -= 4;
2391 }
2392 }
2393
2394#ifdef DEBUGGING
2395 /* Allow dumping */
2396 n = scan + NODE_SZ_STR(scan);
2397 while (n <= stop) {
2398 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2399 OP(n) = OPTIMIZED;
2400 NEXT_OFF(n) = 0;
2401 }
2402 n++;
2403 }
2404#endif
2405 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2406 return stopnow;
2407}
2408
653099ff
GS
2409/* REx optimizer. Converts nodes into quickier variants "in place".
2410 Finds fixed substrings. */
2411
a0288114 2412/* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
c277df42
IZ
2413 to the position after last scanned or to NULL. */
2414
40d049e4
YO
2415#define INIT_AND_WITHP \
2416 assert(!and_withp); \
2417 Newx(and_withp,1,struct regnode_charclass_class); \
2418 SAVEFREEPV(and_withp)
07be1b83 2419
b515a41d
YO
2420/* this is a chain of data about sub patterns we are processing that
2421 need to be handled seperately/specially in study_chunk. Its so
2422 we can simulate recursion without losing state. */
2423struct scan_frame;
2424typedef struct scan_frame {
2425 regnode *last; /* last node to process in this frame */
2426 regnode *next; /* next node to process when last is reached */
2427 struct scan_frame *prev; /*previous frame*/
2428 I32 stop; /* what stopparen do we use */
2429} scan_frame;
2430
304ee84b
YO
2431
2432#define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2433
e1d1eefb
YO
2434#define CASE_SYNST_FNC(nAmE) \
2435case nAmE: \
2436 if (flags & SCF_DO_STCLASS_AND) { \
2437 for (value = 0; value < 256; value++) \
2438 if (!is_ ## nAmE ## _cp(value)) \
2439 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2440 } \
2441 else { \
2442 for (value = 0; value < 256; value++) \
2443 if (is_ ## nAmE ## _cp(value)) \
2444 ANYOF_BITMAP_SET(data->start_class, value); \
2445 } \
2446 break; \
2447case N ## nAmE: \
2448 if (flags & SCF_DO_STCLASS_AND) { \
2449 for (value = 0; value < 256; value++) \
2450 if (is_ ## nAmE ## _cp(value)) \
2451 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2452 } \
2453 else { \
2454 for (value = 0; value < 256; value++) \
2455 if (!is_ ## nAmE ## _cp(value)) \
2456 ANYOF_BITMAP_SET(data->start_class, value); \
2457 } \
2458 break
2459
2460
2461
76e3520e 2462STATIC I32
40d049e4 2463S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
1de06328 2464 I32 *minlenp, I32 *deltap,
40d049e4
YO
2465 regnode *last,
2466 scan_data_t *data,
2467 I32 stopparen,
2468 U8* recursed,
2469 struct regnode_charclass_class *and_withp,
2470 U32 flags, U32 depth)
c277df42
IZ
2471 /* scanp: Start here (read-write). */
2472 /* deltap: Write maxlen-minlen here. */
2473 /* last: Stop before this one. */
40d049e4
YO
2474 /* data: string data about the pattern */
2475 /* stopparen: treat close N as END */
2476 /* recursed: which subroutines have we recursed into */
2477 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
c277df42 2478{
97aff369 2479 dVAR;
c277df42
IZ
2480 I32 min = 0, pars = 0, code;
2481 regnode *scan = *scanp, *next;
2482 I32 delta = 0;
2483 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
aca2d497 2484 int is_inf_internal = 0; /* The studied chunk is infinite */
c277df42
IZ
2485 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2486 scan_data_t data_fake;
a3621e74 2487 SV *re_trie_maxbuff = NULL;
786e8c11 2488 regnode *first_non_open = scan;
e2e6a0f1 2489 I32 stopmin = I32_MAX;
8aa23a47
YO
2490 scan_frame *frame = NULL;
2491
a3621e74 2492 GET_RE_DEBUG_FLAGS_DECL;
8aa23a47 2493
13a24bad 2494#ifdef DEBUGGING
40d049e4 2495 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
13a24bad 2496#endif
40d049e4 2497
786e8c11 2498 if ( depth == 0 ) {
40d049e4 2499 while (first_non_open && OP(first_non_open) == OPEN)
786e8c11
YO
2500 first_non_open=regnext(first_non_open);
2501 }
2502
b81d288d 2503
8aa23a47
YO
2504 fake_study_recurse:
2505 while ( scan && OP(scan) != END && scan < last ){
2506 /* Peephole optimizer: */
304ee84b 2507 DEBUG_STUDYDATA("Peep:", data,depth);
8aa23a47
YO
2508 DEBUG_PEEP("Peep",scan,depth);
2509 JOIN_EXACT(scan,&min,0);
2510
2511 /* Follow the next-chain of the current node and optimize
2512 away all the NOTHINGs from it. */
2513 if (OP(scan) != CURLYX) {
2514 const int max = (reg_off_by_arg[OP(scan)]
2515 ? I32_MAX
2516 /* I32 may be smaller than U16 on CRAYs! */
2517 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2518 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2519 int noff;
2520 regnode *n = scan;
2521
2522 /* Skip NOTHING and LONGJMP. */
2523 while ((n = regnext(n))
2524 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2525 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2526 && off + noff < max)
2527 off += noff;
2528 if (reg_off_by_arg[OP(scan)])
2529 ARG(scan) = off;
2530 else
2531 NEXT_OFF(scan) = off;
2532 }
a3621e74 2533
c277df42 2534
8aa23a47
YO
2535
2536 /* The principal pseudo-switch. Cannot be a switch, since we
2537 look into several different things. */
2538 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2539 || OP(scan) == IFTHEN) {
2540 next = regnext(scan);
2541 code = OP(scan);
2542 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2543
2544 if (OP(next) == code || code == IFTHEN) {
2545 /* NOTE - There is similar code to this block below for handling
2546 TRIE nodes on a re-study. If you change stuff here check there
2547 too. */
2548 I32 max1 = 0, min1 = I32_MAX, num = 0;
2549 struct regnode_charclass_class accum;
2550 regnode * const startbranch=scan;
2551
2552 if (flags & SCF_DO_SUBSTR)
304ee84b 2553 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
8aa23a47
YO
2554 if (flags & SCF_DO_STCLASS)
2555 cl_init_zero(pRExC_state, &accum);
2556
2557 while (OP(scan) == code) {
2558 I32 deltanext, minnext, f = 0, fake;
2559 struct regnode_charclass_class this_class;
2560
2561 num++;
2562 data_fake.flags = 0;
2563 if (data) {
2564 data_fake.whilem_c = data->whilem_c;
2565 data_fake.last_closep = data->last_closep;
2566 }
2567 else
2568 data_fake.last_closep = &fake;
58e23c8d
YO
2569
2570 data_fake.pos_delta = delta;
8aa23a47
YO
2571 next = regnext(scan);
2572 scan = NEXTOPER(scan);
2573 if (code != BRANCH)
c277df42 2574 scan = NEXTOPER(scan);
8aa23a47
YO
2575 if (flags & SCF_DO_STCLASS) {
2576 cl_init(pRExC_state, &this_class);
2577 data_fake.start_class = &this_class;
2578 f = SCF_DO_STCLASS_AND;
58e23c8d 2579 }
8aa23a47
YO
2580 if (flags & SCF_WHILEM_VISITED_POS)
2581 f |= SCF_WHILEM_VISITED_POS;
2582
2583 /* we suppose the run is continuous, last=next...*/
2584 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2585 next, &data_fake,
2586 stopparen, recursed, NULL, f,depth+1);
2587 if (min1 > minnext)
2588 min1 = minnext;
2589 if (max1 < minnext + deltanext)
2590 max1 = minnext + deltanext;
2591 if (deltanext == I32_MAX)
2592 is_inf = is_inf_internal = 1;
2593 scan = next;
2594 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2595 pars++;
2596 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2597 if ( stopmin > minnext)
2598 stopmin = min + min1;
2599 flags &= ~SCF_DO_SUBSTR;
2600 if (data)
2601 data->flags |= SCF_SEEN_ACCEPT;
2602 }
2603 if (data) {
2604 if (data_fake.flags & SF_HAS_EVAL)
2605 data->flags |= SF_HAS_EVAL;
2606 data->whilem_c = data_fake.whilem_c;
3dab1dad 2607 }
8aa23a47
YO
2608 if (flags & SCF_DO_STCLASS)
2609 cl_or(pRExC_state, &accum, &this_class);
2610 }
2611 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2612 min1 = 0;
2613 if (flags & SCF_DO_SUBSTR) {
2614 data->pos_min += min1;
2615 data->pos_delta += max1 - min1;
2616 if (max1 != min1 || is_inf)
2617 data->longest = &(data->longest_float);
2618 }
2619 min += min1;
2620 delta += max1 - min1;
2621 if (flags & SCF_DO_STCLASS_OR) {
2622 cl_or(pRExC_state, data->start_class, &accum);
2623 if (min1) {
2624 cl_and(data->start_class, and_withp);
2625 flags &= ~SCF_DO_STCLASS;
653099ff 2626 }
8aa23a47
YO
2627 }
2628 else if (flags & SCF_DO_STCLASS_AND) {
2629 if (min1) {
2630 cl_and(data->start_class, &accum);
2631 flags &= ~SCF_DO_STCLASS;
de0c8cb8 2632 }
8aa23a47
YO
2633 else {
2634 /* Switch to OR mode: cache the old value of
2635 * data->start_class */
2636 INIT_AND_WITHP;
2637 StructCopy(data->start_class, and_withp,
2638 struct regnode_charclass_class);
2639 flags &= ~SCF_DO_STCLASS_AND;
2640 StructCopy(&accum, data->start_class,
2641 struct regnode_charclass_class);
2642 flags |= SCF_DO_STCLASS_OR;
2643 data->start_class->flags |= ANYOF_EOS;
de0c8cb8 2644 }
8aa23a47 2645 }
a3621e74 2646
8aa23a47
YO
2647 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2648 /* demq.
a3621e74 2649
8aa23a47
YO
2650 Assuming this was/is a branch we are dealing with: 'scan' now
2651 points at the item that follows the branch sequence, whatever
2652 it is. We now start at the beginning of the sequence and look
2653 for subsequences of
a3621e74 2654
8aa23a47
YO
2655 BRANCH->EXACT=>x1
2656 BRANCH->EXACT=>x2
2657 tail
a3621e74 2658
8aa23a47 2659 which would be constructed from a pattern like /A|LIST|OF|WORDS/
a3621e74 2660
8aa23a47
YO
2661 If we can find such a subseqence we need to turn the first
2662 element into a trie and then add the subsequent branch exact
2663 strings to the trie.
a3621e74 2664
8aa23a47 2665 We have two cases
a3621e74 2666
8aa23a47 2667 1. patterns where the whole set of branch can be converted.
a3621e74 2668
8aa23a47 2669 2. patterns where only a subset can be converted.
a3621e74 2670
8aa23a47
YO
2671 In case 1 we can replace the whole set with a single regop
2672 for the trie. In case 2 we need to keep the start and end
2673 branchs so
a3621e74 2674
8aa23a47
YO
2675 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2676 becomes BRANCH TRIE; BRANCH X;
786e8c11 2677
8aa23a47
YO
2678 There is an additional case, that being where there is a
2679 common prefix, which gets split out into an EXACT like node
2680 preceding the TRIE node.
a3621e74 2681
8aa23a47
YO
2682 If x(1..n)==tail then we can do a simple trie, if not we make
2683 a "jump" trie, such that when we match the appropriate word
2684 we "jump" to the appopriate tail node. Essentailly we turn
2685 a nested if into a case structure of sorts.
b515a41d 2686
8aa23a47
YO
2687 */
2688
2689 int made=0;
2690 if (!re_trie_maxbuff) {
2691 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2692 if (!SvIOK(re_trie_maxbuff))
2693 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2694 }
2695 if ( SvIV(re_trie_maxbuff)>=0 ) {
2696 regnode *cur;
2697 regnode *first = (regnode *)NULL;
2698 regnode *last = (regnode *)NULL;
2699 regnode *tail = scan;
2700 U8 optype = 0;
2701 U32 count=0;
a3621e74
YO
2702
2703#ifdef DEBUGGING
8aa23a47 2704 SV * const mysv = sv_newmortal(); /* for dumping */
a3621e74 2705#endif
8aa23a47
YO
2706 /* var tail is used because there may be a TAIL
2707 regop in the way. Ie, the exacts will point to the
2708 thing following the TAIL, but the last branch will
2709 point at the TAIL. So we advance tail. If we
2710 have nested (?:) we may have to move through several
2711 tails.
2712 */
2713
2714 while ( OP( tail ) == TAIL ) {
2715 /* this is the TAIL generated by (?:) */
2716 tail = regnext( tail );
2717 }
a3621e74 2718
8aa23a47
YO
2719
2720 DEBUG_OPTIMISE_r({
2721 regprop(RExC_rx, mysv, tail );
2722 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2723 (int)depth * 2 + 2, "",
2724 "Looking for TRIE'able sequences. Tail node is: ",
2725 SvPV_nolen_const( mysv )
2726 );
2727 });
2728
2729 /*
2730
2731 step through the branches, cur represents each
2732 branch, noper is the first thing to be matched
2733 as part of that branch and noper_next is the
2734 regnext() of that node. if noper is an EXACT
2735 and noper_next is the same as scan (our current
2736 position in the regex) then the EXACT branch is
2737 a possible optimization target. Once we have
2738 two or more consequetive such branches we can
2739 create a trie of the EXACT's contents and stich
2740 it in place. If the sequence represents all of
2741 the branches we eliminate the whole thing and
2742 replace it with a single TRIE. If it is a
2743 subsequence then we need to stitch it in. This
2744 means the first branch has to remain, and needs
2745 to be repointed at the item on the branch chain
2746 following the last branch optimized. This could
2747 be either a BRANCH, in which case the
2748 subsequence is internal, or it could be the
2749 item following the branch sequence in which
2750 case the subsequence is at the end.
2751
2752 */
2753
2754 /* dont use tail as the end marker for this traverse */
2755 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2756 regnode * const noper = NEXTOPER( cur );
b515a41d 2757#if defined(DEBUGGING) || defined(NOJUMPTRIE)
8aa23a47 2758 regnode * const noper_next = regnext( noper );
b515a41d
YO
2759#endif
2760
8aa23a47
YO
2761 DEBUG_OPTIMISE_r({
2762 regprop(RExC_rx, mysv, cur);
2763 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2764 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2765
2766 regprop(RExC_rx, mysv, noper);
2767 PerlIO_printf( Perl_debug_log, " -> %s",
2768 SvPV_nolen_const(mysv));
2769
2770 if ( noper_next ) {
2771 regprop(RExC_rx, mysv, noper_next );
2772 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2773 SvPV_nolen_const(mysv));
2774 }
2775 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2776 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2777 });
2778 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2779 : PL_regkind[ OP( noper ) ] == EXACT )
2780 || OP(noper) == NOTHING )
786e8c11 2781#ifdef NOJUMPTRIE
8aa23a47 2782 && noper_next == tail
786e8c11 2783#endif
8aa23a47
YO
2784 && count < U16_MAX)
2785 {
2786 count++;
2787 if ( !first || optype == NOTHING ) {
2788 if (!first) first = cur;
2789 optype = OP( noper );
2790 } else {
2791 last = cur;
2792 }
2793 } else {
a0a388a1
YO
2794/*
2795 Currently we assume that the trie can handle unicode and ascii
2796 matches fold cased matches. If this proves true then the following
2797 define will prevent tries in this situation.
2798
2799 #define TRIE_TYPE_IS_SAFE (UTF || optype==EXACT)
2800*/
2801#define TRIE_TYPE_IS_SAFE 1
2802 if ( last && TRIE_TYPE_IS_SAFE ) {
8aa23a47
YO
2803 make_trie( pRExC_state,
2804 startbranch, first, cur, tail, count,
2805 optype, depth+1 );
2806 }
2807 if ( PL_regkind[ OP( noper ) ] == EXACT
786e8c11 2808#ifdef NOJUMPTRIE
8aa23a47 2809 && noper_next == tail
786e8c11 2810#endif
8aa23a47
YO
2811 ){
2812 count = 1;
2813 first = cur;
2814 optype = OP( noper );
2815 } else {
2816 count = 0;
2817 first = NULL;
2818 optype = 0;
2819 }
2820 last = NULL;
2821 }
2822 }
2823 DEBUG_OPTIMISE_r({
2824 regprop(RExC_rx, mysv, cur);
2825 PerlIO_printf( Perl_debug_log,
2826 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2827 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2828
2829 });
a0a388a1
YO
2830
2831 if ( last && TRIE_TYPE_IS_SAFE ) {
8aa23a47 2832 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3dab1dad 2833#ifdef TRIE_STUDY_OPT
8aa23a47
YO
2834 if ( ((made == MADE_EXACT_TRIE &&
2835 startbranch == first)
2836 || ( first_non_open == first )) &&
2837 depth==0 ) {
2838 flags |= SCF_TRIE_RESTUDY;
2839 if ( startbranch == first
2840 && scan == tail )
2841 {
2842 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2843 }
2844 }
3dab1dad 2845#endif
8aa23a47
YO
2846 }
2847 }
2848
2849 } /* do trie */
2850
653099ff 2851 }
8aa23a47
YO
2852 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2853 scan = NEXTOPER(NEXTOPER(scan));
2854 } else /* single branch is optimized. */
2855 scan = NEXTOPER(scan);
2856 continue;
2857 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2858 scan_frame *newframe = NULL;
2859 I32 paren;
2860 regnode *start;
2861 regnode *end;
2862
2863 if (OP(scan) != SUSPEND) {
2864 /* set the pointer */
2865 if (OP(scan) == GOSUB) {
2866 paren = ARG(scan);
2867 RExC_recurse[ARG2L(scan)] = scan;
2868 start = RExC_open_parens[paren-1];
2869 end = RExC_close_parens[paren-1];
2870 } else {
2871 paren = 0;
f8fc2ecf 2872 start = RExC_rxi->program + 1;
8aa23a47
YO
2873 end = RExC_opend;
2874 }
2875 if (!recursed) {
2876 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2877 SAVEFREEPV(recursed);
2878 }
2879 if (!PAREN_TEST(recursed,paren+1)) {
2880 PAREN_SET(recursed,paren+1);
2881 Newx(newframe,1,scan_frame);
2882 } else {
2883 if (flags & SCF_DO_SUBSTR) {
304ee84b 2884 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
2885 data->longest = &(data->longest_float);
2886 }
2887 is_inf = is_inf_internal = 1;
2888 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2889 cl_anything(pRExC_state, data->start_class);
2890 flags &= ~SCF_DO_STCLASS;
2891 }
2892 } else {
2893 Newx(newframe,1,scan_frame);
2894 paren = stopparen;
2895 start = scan+2;
2896 end = regnext(scan);
2897 }
2898 if (newframe) {
2899 assert(start);
2900 assert(end);
2901 SAVEFREEPV(newframe);
2902 newframe->next = regnext(scan);
2903 newframe->last = last;
2904 newframe->stop = stopparen;
2905 newframe->prev = frame;
2906
2907 frame = newframe;
2908 scan = start;
2909 stopparen = paren;
2910 last = end;
2911
2912 continue;
2913 }
2914 }
2915 else if (OP(scan) == EXACT) {
2916 I32 l = STR_LEN(scan);
2917 UV uc;
2918 if (UTF) {
2919 const U8 * const s = (U8*)STRING(scan);
2920 l = utf8_length(s, s + l);
2921 uc = utf8_to_uvchr(s, NULL);
2922 } else {
2923 uc = *((U8*)STRING(scan));
2924 }
2925 min += l;
2926 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2927 /* The code below prefers earlier match for fixed
2928 offset, later match for variable offset. */
2929 if (data->last_end == -1) { /* Update the start info. */
2930 data->last_start_min = data->pos_min;
2931 data->last_start_max = is_inf
2932 ? I32_MAX : data->pos_min + data->pos_delta;
b515a41d 2933 }
8aa23a47
YO
2934 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2935 if (UTF)
2936 SvUTF8_on(data->last_found);
2937 {
2938 SV * const sv = data->last_found;
2939 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2940 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2941 if (mg && mg->mg_len >= 0)
2942 mg->mg_len += utf8_length((U8*)STRING(scan),
2943 (U8*)STRING(scan)+STR_LEN(scan));
b515a41d 2944 }
8aa23a47
YO
2945 data->last_end = data->pos_min + l;
2946 data->pos_min += l; /* As in the first entry. */
2947 data->flags &= ~SF_BEFORE_EOL;
2948 }
2949 if (flags & SCF_DO_STCLASS_AND) {
2950 /* Check whether it is compatible with what we know already! */
2951 int compat = 1;
2952
2953 if (uc >= 0x100 ||
2954 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2955 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2956 && (!(data->start_class->flags & ANYOF_FOLD)
2957 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2958 )
2959 compat = 0;
2960 ANYOF_CLASS_ZERO(data->start_class);
2961 ANYOF_BITMAP_ZERO(data->start_class);
2962 if (compat)
2963 ANYOF_BITMAP_SET(data->start_class, uc);
2964 data->start_class->flags &= ~ANYOF_EOS;
2965 if (uc < 0x100)
2966 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2967 }
2968 else if (flags & SCF_DO_STCLASS_OR) {
2969 /* false positive possible if the class is case-folded */
2970 if (uc < 0x100)
2971 ANYOF_BITMAP_SET(data->start_class, uc);
2972 else
2973 data->start_class->flags |= ANYOF_UNICODE_ALL;
2974 data->start_class->flags &= ~ANYOF_EOS;
2975 cl_and(data->start_class, and_withp);
2976 }
2977 flags &= ~SCF_DO_STCLASS;
2978 }
2979 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2980 I32 l = STR_LEN(scan);
2981 UV uc = *((U8*)STRING(scan));
2982
2983 /* Search for fixed substrings supports EXACT only. */
2984 if (flags & SCF_DO_SUBSTR) {
2985 assert(data);
304ee84b 2986 SCAN_COMMIT(pRExC_state, data, minlenp);
8aa23a47
YO
2987 }
2988 if (UTF) {
2989 const U8 * const s = (U8 *)STRING(scan);
2990 l = utf8_length(s, s + l);
2991 uc = utf8_to_uvchr(s, NULL);
2992 }
2993 min += l;
2994 if (flags & SCF_DO_SUBSTR)
2995 data->pos_min += l;
2996 if (flags & SCF_DO_STCLASS_AND) {
2997 /* Check whether it is compatible with what we know already! */
2998 int compat = 1;
2999
3000 if (uc >= 0x100 ||
3001 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
3002 && !ANYOF_BITMAP_TEST(data->start_class, uc)
3003 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
3004 compat = 0;
3005 ANYOF_CLASS_ZERO(data->start_class);
3006 ANYOF_BITMAP_ZERO(data->start_class);
3007 if (compat) {
3008 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff 3009 data->start_class->flags &= ~ANYOF_EOS;
8aa23a47
YO
3010 data->start_class->flags |= ANYOF_FOLD;
3011 if (OP(scan) == EXACTFL)
3012 data->start_class->flags |= ANYOF_LOCALE;
653099ff 3013 }
8aa23a47
YO
3014 }
3015 else if (flags & SCF_DO_STCLASS_OR) {
3016 if (data->start_class->flags & ANYOF_FOLD) {
3017 /* false positive possible if the class is case-folded.
3018 Assume that the locale settings are the same... */
1aa99e6b
IH
3019 if (uc < 0x100)
3020 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff
GS
3021 data->start_class->flags &= ~ANYOF_EOS;
3022 }
8aa23a47 3023 cl_and(data->start_class, and_withp);
653099ff 3024 }
8aa23a47
YO
3025 flags &= ~SCF_DO_STCLASS;
3026 }
3027 else if (strchr((const char*)PL_varies,OP(scan))) {
3028 I32 mincount, maxcount, minnext, deltanext, fl = 0;
3029 I32 f = flags, pos_before = 0;
3030 regnode * const oscan = scan;
3031 struct regnode_charclass_class this_class;
3032 struct regnode_charclass_class *oclass = NULL;
3033 I32 next_is_eval = 0;
3034
3035 switch (PL_regkind[OP(scan)]) {
3036 case WHILEM: /* End of (?:...)* . */
3037 scan = NEXTOPER(scan);
3038 goto finish;
3039 case PLUS:
3040 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3041 next = NEXTOPER(scan);
3042 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3043 mincount = 1;
3044 maxcount = REG_INFTY;
3045 next = regnext(scan);
3046 scan = NEXTOPER(scan);
3047 goto do_curly;
3048 }
3049 }
3050 if (flags & SCF_DO_SUBSTR)
3051 data->pos_min++;
3052 min++;
3053 /* Fall through. */
3054 case STAR:
3055 if (flags & SCF_DO_STCLASS) {
3056 mincount = 0;
3057 maxcount = REG_INFTY;
3058 next = regnext(scan);
3059 scan = NEXTOPER(scan);
3060 goto do_curly;
3061 }
3062 is_inf = is_inf_internal = 1;
3063 scan = regnext(scan);
c277df42 3064 if (flags & SCF_DO_SUBSTR) {
304ee84b 3065 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
8aa23a47 3066 data->longest = &(data->longest_float);
c277df42 3067 }
8aa23a47
YO
3068 goto optimize_curly_tail;
3069 case CURLY:
3070 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3071 && (scan->flags == stopparen))
3072 {
3073 mincount = 1;
3074 maxcount = 1;
3075 } else {
3076 mincount = ARG1(scan);
3077 maxcount = ARG2(scan);
653099ff 3078 }
8aa23a47
YO
3079 next = regnext(scan);
3080 if (OP(scan) == CURLYX) {
3081 I32 lp = (data ? *(data->last_closep) : 0);
3082 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
653099ff 3083 }
8aa23a47
YO
3084 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3085 next_is_eval = (OP(scan) == EVAL);
3086 do_curly:
3087 if (flags & SCF_DO_SUBSTR) {
304ee84b 3088 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
8aa23a47 3089 pos_before = data->pos_min;
b45f050a 3090 }
8aa23a47
YO
3091 if (data) {
3092 fl = data->flags;
3093 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3094 if (is_inf)
3095 data->flags |= SF_IS_INF;
3096 }
3097 if (flags & SCF_DO_STCLASS) {
3098 cl_init(pRExC_state, &this_class);
3099 oclass = data->start_class;
3100 data->start_class = &this_class;
3101 f |= SCF_DO_STCLASS_AND;
3102 f &= ~SCF_DO_STCLASS_OR;
3103 }
3104 /* These are the cases when once a subexpression
3105 fails at a particular position, it cannot succeed
3106 even after backtracking at the enclosing scope.
3107
3108 XXXX what if minimal match and we are at the
3109 initial run of {n,m}? */
3110 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3111 f &= ~SCF_WHILEM_VISITED_POS;
b45f050a 3112
8aa23a47
YO
3113 /* This will finish on WHILEM, setting scan, or on NULL: */
3114 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3115 last, data, stopparen, recursed, NULL,
3116 (mincount == 0
3117 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
b515a41d 3118
8aa23a47
YO
3119 if (flags & SCF_DO_STCLASS)
3120 data->start_class = oclass;
3121 if (mincount == 0 || minnext == 0) {
3122 if (flags & SCF_DO_STCLASS_OR) {
3123 cl_or(pRExC_state, data->start_class, &this_class);
3124 }
3125 else if (flags & SCF_DO_STCLASS_AND) {
3126 /* Switch to OR mode: cache the old value of
3127 * data->start_class */
3128 INIT_AND_WITHP;
3129 StructCopy(data->start_class, and_withp,
3130 struct regnode_charclass_class);
3131 flags &= ~SCF_DO_STCLASS_AND;
3132 StructCopy(&this_class, data->start_class,
3133 struct regnode_charclass_class);
3134 flags |= SCF_DO_STCLASS_OR;
3135 data->start_class->flags |= ANYOF_EOS;
3136 }
3137 } else { /* Non-zero len */
3138 if (flags & SCF_DO_STCLASS_OR) {
3139 cl_or(pRExC_state, data->start_class, &this_class);
3140 cl_and(data->start_class, and_withp);
3141 }
3142 else if (flags & SCF_DO_STCLASS_AND)
3143 cl_and(data->start_class, &this_class);
3144 flags &= ~SCF_DO_STCLASS;
3145 }
3146 if (!scan) /* It was not CURLYX, but CURLY. */
3147 scan = next;
3148 if ( /* ? quantifier ok, except for (?{ ... }) */
3149 (next_is_eval || !(mincount == 0 && maxcount == 1))
3150 && (minnext == 0) && (deltanext == 0)
3151 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3152 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3153 && ckWARN(WARN_REGEXP))
3154 {
3155 vWARN(RExC_parse,
3156 "Quantifier unexpected on zero-length expression");
3157 }
3158
3159 min += minnext * mincount;
3160 is_inf_internal |= ((maxcount == REG_INFTY
3161 && (minnext + deltanext) > 0)
3162 || deltanext == I32_MAX);
3163 is_inf |= is_inf_internal;
3164 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3165
3166 /* Try powerful optimization CURLYX => CURLYN. */
3167 if ( OP(oscan) == CURLYX && data
3168 && data->flags & SF_IN_PAR
3169 && !(data->flags & SF_HAS_EVAL)
3170 && !deltanext && minnext == 1 ) {
3171 /* Try to optimize to CURLYN. */
3172 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3173 regnode * const nxt1 = nxt;
497b47a8 3174#ifdef DEBUGGING
8aa23a47 3175 regnode *nxt2;
497b47a8 3176#endif
c277df42 3177
8aa23a47
YO
3178 /* Skip open. */
3179 nxt = regnext(nxt);
3180 if (!strchr((const char*)PL_simple,OP(nxt))
3181 && !(PL_regkind[OP(nxt)] == EXACT
3182 && STR_LEN(nxt) == 1))
3183 goto nogo;
497b47a8 3184#ifdef DEBUGGING
8aa23a47 3185 nxt2 = nxt;
497b47a8 3186#endif
8aa23a47
YO
3187 nxt = regnext(nxt);
3188 if (OP(nxt) != CLOSE)
3189 goto nogo;
3190 if (RExC_open_parens) {
3191 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3192 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3193 }
3194 /* Now we know that nxt2 is the only contents: */
3195 oscan->flags = (U8)ARG(nxt);
3196 OP(oscan) = CURLYN;
3197 OP(nxt1) = NOTHING; /* was OPEN. */
40d049e4 3198
c277df42 3199#ifdef DEBUGGING
8aa23a47
YO
3200 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3201 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3202 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3203 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3204 OP(nxt + 1) = OPTIMIZED; /* was count. */
3205 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
b81d288d 3206#endif
8aa23a47
YO
3207 }
3208 nogo:
3209
3210 /* Try optimization CURLYX => CURLYM. */
3211 if ( OP(oscan) == CURLYX && data
3212 && !(data->flags & SF_HAS_PAR)
3213 && !(data->flags & SF_HAS_EVAL)
3214 && !deltanext /* atom is fixed width */
3215 && minnext != 0 /* CURLYM can't handle zero width */
3216 ) {
3217 /* XXXX How to optimize if data == 0? */
3218 /* Optimize to a simpler form. */
3219 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3220 regnode *nxt2;
3221
3222 OP(oscan) = CURLYM;
3223 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3224 && (OP(nxt2) != WHILEM))
3225 nxt = nxt2;
3226 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3227 /* Need to optimize away parenths. */
3228 if (data->flags & SF_IN_PAR) {
3229 /* Set the parenth number. */
3230 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3231
3232 if (OP(nxt) != CLOSE)
3233 FAIL("Panic opt close");
3234 oscan->flags = (U8)ARG(nxt);
3235 if (RExC_open_parens) {
3236 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3237 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
40d049e4 3238 }
8aa23a47
YO
3239 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3240 OP(nxt) = OPTIMIZED; /* was CLOSE. */
40d049e4 3241
c277df42 3242#ifdef DEBUGGING
8aa23a47
YO
3243 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3244 OP(nxt + 1) = OPTIMIZED; /* was count. */
3245 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3246 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
b81d288d 3247#endif
c277df42 3248#if 0
8aa23a47
YO
3249 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3250 regnode *nnxt = regnext(nxt1);
3251
3252 if (nnxt == nxt) {
3253 if (reg_off_by_arg[OP(nxt1)])
3254 ARG_SET(nxt1, nxt2 - nxt1);
3255 else if (nxt2 - nxt1 < U16_MAX)
3256 NEXT_OFF(nxt1) = nxt2 - nxt1;
3257 else
3258 OP(nxt) = NOTHING; /* Cannot beautify */
c277df42 3259 }
8aa23a47 3260 nxt1 = nnxt;
c277df42 3261 }
5d1c421c 3262#endif
8aa23a47
YO
3263 /* Optimize again: */
3264 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3265 NULL, stopparen, recursed, NULL, 0,depth+1);
3266 }
3267 else
3268 oscan->flags = 0;
3269 }
3270 else if ((OP(oscan) == CURLYX)
3271 && (flags & SCF_WHILEM_VISITED_POS)
3272 /* See the comment on a similar expression above.
3273 However, this time it not a subexpression
3274 we care about, but the expression itself. */
3275 && (maxcount == REG_INFTY)
3276 && data && ++data->whilem_c < 16) {
3277 /* This stays as CURLYX, we can put the count/of pair. */
3278 /* Find WHILEM (as in regexec.c) */
3279 regnode *nxt = oscan + NEXT_OFF(oscan);
3280
3281 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3282 nxt += ARG(nxt);
3283 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3284 | (RExC_whilem_seen << 4)); /* On WHILEM */
3285 }
3286 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3287 pars++;
3288 if (flags & SCF_DO_SUBSTR) {
3289 SV *last_str = NULL;
3290 int counted = mincount != 0;
a0ed51b3 3291
8aa23a47
YO
3292 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3293#if defined(SPARC64_GCC_WORKAROUND)
3294 I32 b = 0;
3295 STRLEN l = 0;
3296 const char *s = NULL;
3297 I32 old = 0;
b515a41d 3298
8aa23a47
YO
3299 if (pos_before >= data->last_start_min)
3300 b = pos_before;
3301 else
3302 b = data->last_start_min;
b515a41d 3303
8aa23a47
YO
3304 l = 0;
3305 s = SvPV_const(data->last_found, l);
3306 old = b - data->last_start_min;
3307
3308#else
3309 I32 b = pos_before >= data->last_start_min
3310 ? pos_before : data->last_start_min;
3311 STRLEN l;
3312 const char * const s = SvPV_const(data->last_found, l);
3313 I32 old = b - data->last_start_min;
3314#endif
3315
3316 if (UTF)
3317 old = utf8_hop((U8*)s, old) - (U8*)s;
3318
3319 l -= old;
3320 /* Get the added string: */
3321 last_str = newSVpvn(s + old, l);
3322 if (UTF)
3323 SvUTF8_on(last_str);
3324 if (deltanext == 0 && pos_before == b) {
3325 /* What was added is a constant string */
3326 if (mincount > 1) {
3327 SvGROW(last_str, (mincount * l) + 1);
3328 repeatcpy(SvPVX(last_str) + l,
3329 SvPVX_const(last_str), l, mincount - 1);
3330 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3331 /* Add additional parts. */
3332 SvCUR_set(data->last_found,
3333 SvCUR(data->last_found) - l);
3334 sv_catsv(data->last_found, last_str);
3335 {
3336 SV * sv = data->last_found;
3337 MAGIC *mg =
3338 SvUTF8(sv) && SvMAGICAL(sv) ?
3339 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3340 if (mg && mg->mg_len >= 0)
bd94e887 3341 mg->mg_len += CHR_SVLEN(last_str) - l;
b515a41d 3342 }
8aa23a47 3343 data->last_end += l * (mincount - 1);
b515a41d 3344 }
8aa23a47
YO
3345 } else {
3346 /* start offset must point into the last copy */
3347 data->last_start_min += minnext * (mincount - 1);
3348 data->last_start_max += is_inf ? I32_MAX
3349 : (maxcount - 1) * (minnext + data->pos_delta);
3350 }
c277df42 3351 }
8aa23a47
YO
3352 /* It is counted once already... */
3353 data->pos_min += minnext * (mincount - counted);
3354 data->pos_delta += - counted * deltanext +
3355 (minnext + deltanext) * maxcount - minnext * mincount;
3356 if (mincount != maxcount) {
3357 /* Cannot extend fixed substrings found inside
3358 the group. */
304ee84b 3359 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3360 if (mincount && last_str) {
3361 SV * const sv = data->last_found;
3362 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3363 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3364
3365 if (mg)
3366 mg->mg_len = -1;
3367 sv_setsv(sv, last_str);
3368 data->last_end = data->pos_min;
3369 data->last_start_min =
3370 data->pos_min - CHR_SVLEN(last_str);
3371 data->last_start_max = is_inf
3372 ? I32_MAX
3373 : data->pos_min + data->pos_delta
3374 - CHR_SVLEN(last_str);
3375 }
3376 data->longest = &(data->longest_float);
3377 }
3378 SvREFCNT_dec(last_str);
c277df42 3379 }
8aa23a47
YO
3380 if (data && (fl & SF_HAS_EVAL))
3381 data->flags |= SF_HAS_EVAL;
3382 optimize_curly_tail:
3383 if (OP(oscan) != CURLYX) {
3384 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3385 && NEXT_OFF(next))
3386 NEXT_OFF(oscan) += NEXT_OFF(next);
3387 }
3388 continue;
3389 default: /* REF and CLUMP only? */
3390 if (flags & SCF_DO_SUBSTR) {
304ee84b 3391 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
8aa23a47
YO
3392 data->longest = &(data->longest_float);
3393 }
3394 is_inf = is_inf_internal = 1;
3395 if (flags & SCF_DO_STCLASS_OR)
3396 cl_anything(pRExC_state, data->start_class);
3397 flags &= ~SCF_DO_STCLASS;
3398 break;
c277df42 3399 }
8aa23a47 3400 }
e1d1eefb
YO
3401 else if (OP(scan) == LNBREAK) {
3402 if (flags & SCF_DO_STCLASS) {
3403 int value = 0;
3404 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3405 if (flags & SCF_DO_STCLASS_AND) {
3406 for (value = 0; value < 256; value++)
e64b1bd1 3407 if (!is_VERTWS_cp(value))
e1d1eefb
YO
3408 ANYOF_BITMAP_CLEAR(data->start_class, value);
3409 }
3410 else {
3411 for (value = 0; value < 256; value++)
e64b1bd1 3412 if (is_VERTWS_cp(value))
e1d1eefb
YO
3413 ANYOF_BITMAP_SET(data->start_class, value);
3414 }
3415 if (flags & SCF_DO_STCLASS_OR)
3416 cl_and(data->start_class, and_withp);
3417 flags &= ~SCF_DO_STCLASS;
3418 }
3419 min += 1;
f9a79580 3420 delta += 1;
e1d1eefb
YO
3421 if (flags & SCF_DO_SUBSTR) {
3422 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3423 data->pos_min += 1;
f9a79580 3424 data->pos_delta += 1;
e1d1eefb
YO
3425 data->longest = &(data->longest_float);
3426 }
3427
3428 }
f9a79580
RGS
3429 else if (OP(scan) == FOLDCHAR) {
3430 int d = ARG(scan)==0xDF ? 1 : 2;
3431 flags &= ~SCF_DO_STCLASS;
3432 min += 1;
3433 delta += d;
3434 if (flags & SCF_DO_SUBSTR) {
3435 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3436 data->pos_min += 1;
3437 data->pos_delta += d;
3438 data->longest = &(data->longest_float);
3439 }
3440 }
8aa23a47
YO
3441 else if (strchr((const char*)PL_simple,OP(scan))) {
3442 int value = 0;
653099ff 3443
8aa23a47 3444 if (flags & SCF_DO_SUBSTR) {
304ee84b 3445 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3446 data->pos_min++;
3447 }
3448 min++;
3449 if (flags & SCF_DO_STCLASS) {
3450 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
b515a41d 3451
8aa23a47
YO
3452 /* Some of the logic below assumes that switching
3453 locale on will only add false positives. */
3454 switch (PL_regkind[OP(scan)]) {
3455 case SANY:
3456 default:
3457 do_default:
3458 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3459 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3460 cl_anything(pRExC_state, data->start_class);
3461 break;
3462 case REG_ANY:
3463 if (OP(scan) == SANY)
3464 goto do_default;
3465 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3466 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3467 || (data->start_class->flags & ANYOF_CLASS));
3468 cl_anything(pRExC_state, data->start_class);
653099ff 3469 }
8aa23a47
YO
3470 if (flags & SCF_DO_STCLASS_AND || !value)
3471 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3472 break;
3473 case ANYOF:
3474 if (flags & SCF_DO_STCLASS_AND)
3475 cl_and(data->start_class,
3476 (struct regnode_charclass_class*)scan);
653099ff 3477 else
8aa23a47
YO
3478 cl_or(pRExC_state, data->start_class,
3479 (struct regnode_charclass_class*)scan);
3480 break;
3481 case ALNUM:
3482 if (flags & SCF_DO_STCLASS_AND) {
3483 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3484 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3485 for (value = 0; value < 256; value++)
3486 if (!isALNUM(value))
3487 ANYOF_BITMAP_CLEAR(data->start_class, value);
3488 }
653099ff 3489 }
8aa23a47
YO
3490 else {
3491 if (data->start_class->flags & ANYOF_LOCALE)
3492 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3493 else {
3494 for (value = 0; value < 256; value++)
3495 if (isALNUM(value))
3496 ANYOF_BITMAP_SET(data->start_class, value);
653099ff 3497 }
8aa23a47
YO
3498 }
3499 break;
3500 case ALNUML:
3501 if (flags & SCF_DO_STCLASS_AND) {
3502 if (data->start_class->flags & ANYOF_LOCALE)
3503 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3504 }
3505 else {
3506 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3507 data->start_class->flags |= ANYOF_LOCALE;
3508 }
3509 break;
3510 case NALNUM:
3511 if (flags & SCF_DO_STCLASS_AND) {
3512 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3513 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3514 for (value = 0; value < 256; value++)
3515 if (isALNUM(value))
3516 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff
GS
3517 }
3518 }
8aa23a47
YO
3519 else {
3520 if (data->start_class->flags & ANYOF_LOCALE)
3521 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3522 else {
3523 for (value = 0; value < 256; value++)
3524 if (!isALNUM(value))
3525 ANYOF_BITMAP_SET(data->start_class, value);
3526 }
653099ff 3527 }
8aa23a47
YO
3528 break;
3529 case NALNUML:
3530 if (flags & SCF_DO_STCLASS_AND) {
3531 if (data->start_class->flags & ANYOF_LOCALE)
3532 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
653099ff 3533 }
8aa23a47
YO
3534 else {
3535 data->start_class->flags |= ANYOF_LOCALE;
3536 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3537 }
3538 break;
3539 case SPACE:
3540 if (flags & SCF_DO_STCLASS_AND) {
3541 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3542 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3543 for (value = 0; value < 256; value++)
3544 if (!isSPACE(value))
3545 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff
GS
3546 }
3547 }
8aa23a47
YO
3548 else {
3549 if (data->start_class->flags & ANYOF_LOCALE)
3550 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3551 else {
3552 for (value = 0; value < 256; value++)
3553 if (isSPACE(value))
3554 ANYOF_BITMAP_SET(data->start_class, value);
3555 }
653099ff 3556 }
8aa23a47
YO
3557 break;
3558 case SPACEL:
3559 if (flags & SCF_DO_STCLASS_AND) {
3560 if (data->start_class->flags & ANYOF_LOCALE)
3561 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3562 }
3563 else {
3564 data->start_class->flags |= ANYOF_LOCALE;
3565 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3566 }
3567 break;
3568 case NSPACE:
3569 if (flags & SCF_DO_STCLASS_AND) {
3570 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3571 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3572 for (value = 0; value < 256; value++)
3573 if (isSPACE(value))
3574 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff 3575 }
8aa23a47
YO
3576 }
3577 else {
3578 if (data->start_class->flags & ANYOF_LOCALE)
3579 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3580 else {
3581 for (value = 0; value < 256; value++)
3582 if (!isSPACE(value))
3583 ANYOF_BITMAP_SET(data->start_class, value);
653099ff
GS
3584 }
3585 }
8aa23a47
YO
3586 break;
3587 case NSPACEL:
3588 if (flags & SCF_DO_STCLASS_AND) {
3589 if (data->start_class->flags & ANYOF_LOCALE) {
3590 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3591 for (value = 0; value < 256; value++)
3592 if (!isSPACE(value))
3593 ANYOF_BITMAP_CLEAR(data->start_class, value);
3594 }
653099ff 3595 }
8aa23a47
YO
3596 else {
3597 data->start_class->flags |= ANYOF_LOCALE;
3598 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3599 }
3600 break;
3601 case DIGIT:
3602 if (flags & SCF_DO_STCLASS_AND) {
3603 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3604 for (value = 0; value < 256; value++)
3605 if (!isDIGIT(value))
3606 ANYOF_BITMAP_CLEAR(data->start_class, value);
3607 }
3608 else {
3609 if (data->start_class->flags & ANYOF_LOCALE)
3610 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3611 else {
3612 for (value = 0; value < 256; value++)
3613 if (isDIGIT(value))
3614 ANYOF_BITMAP_SET(data->start_class, value);
3615 }
3616 }
3617 break;
3618 case NDIGIT:
3619 if (flags & SCF_DO_STCLASS_AND) {
3620 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3621 for (value = 0; value < 256; value++)
3622 if (isDIGIT(value))
3623 ANYOF_BITMAP_CLEAR(data->start_class, value);
3624 }
3625 else {
3626 if (data->start_class->flags & ANYOF_LOCALE)
3627 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3628 else {
3629 for (value = 0; value < 256; value++)
3630 if (!isDIGIT(value))
3631 ANYOF_BITMAP_SET(data->start_class, value);
653099ff
GS
3632 }
3633 }
8aa23a47 3634 break;
e1d1eefb
YO
3635 CASE_SYNST_FNC(VERTWS);
3636 CASE_SYNST_FNC(HORIZWS);
3637
8aa23a47
YO
3638 }
3639 if (flags & SCF_DO_STCLASS_OR)
3640 cl_and(data->start_class, and_withp);
3641 flags &= ~SCF_DO_STCLASS;
3642 }
3643 }
3644 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3645 data->flags |= (OP(scan) == MEOL
3646 ? SF_BEFORE_MEOL
3647 : SF_BEFORE_SEOL);
3648 }
3649 else if ( PL_regkind[OP(scan)] == BRANCHJ
3650 /* Lookbehind, or need to calculate parens/evals/stclass: */
3651 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3652 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3653 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3654 || OP(scan) == UNLESSM )
3655 {
3656 /* Negative Lookahead/lookbehind
3657 In this case we can't do fixed string optimisation.
3658 */
1de06328 3659
8aa23a47
YO
3660 I32 deltanext, minnext, fake = 0;
3661 regnode *nscan;
3662 struct regnode_charclass_class intrnl;
3663 int f = 0;
1de06328 3664
8aa23a47
YO
3665 data_fake.flags = 0;
3666 if (data) {
3667 data_fake.whilem_c = data->whilem_c;
3668 data_fake.last_closep = data->last_closep;
c277df42 3669 }
8aa23a47
YO
3670 else
3671 data_fake.last_closep = &fake;
58e23c8d 3672 data_fake.pos_delta = delta;
8aa23a47
YO
3673 if ( flags & SCF_DO_STCLASS && !scan->flags
3674 && OP(scan) == IFMATCH ) { /* Lookahead */
3675 cl_init(pRExC_state, &intrnl);
3676 data_fake.start_class = &intrnl;
3677 f |= SCF_DO_STCLASS_AND;
3678 }
3679 if (flags & SCF_WHILEM_VISITED_POS)
3680 f |= SCF_WHILEM_VISITED_POS;
3681 next = regnext(scan);
3682 nscan = NEXTOPER(NEXTOPER(scan));
3683 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3684 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3685 if (scan->flags) {
3686 if (deltanext) {
58e23c8d 3687 FAIL("Variable length lookbehind not implemented");
8aa23a47
YO
3688 }
3689 else if (minnext > (I32)U8_MAX) {
58e23c8d 3690 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
8aa23a47
YO
3691 }
3692 scan->flags = (U8)minnext;
3693 }
3694 if (data) {
3695 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3696 pars++;
3697 if (data_fake.flags & SF_HAS_EVAL)
3698 data->flags |= SF_HAS_EVAL;
3699 data->whilem_c = data_fake.whilem_c;
3700 }
3701 if (f & SCF_DO_STCLASS_AND) {
3702 const int was = (data->start_class->flags & ANYOF_EOS);
3703
3704 cl_and(data->start_class, &intrnl);
3705 if (was)
3706 data->start_class->flags |= ANYOF_EOS;
3707 }
cb434fcc 3708 }
8aa23a47
YO
3709#if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3710 else {
3711 /* Positive Lookahead/lookbehind
3712 In this case we can do fixed string optimisation,
3713 but we must be careful about it. Note in the case of
3714 lookbehind the positions will be offset by the minimum
3715 length of the pattern, something we won't know about
3716 until after the recurse.
3717 */
3718 I32 deltanext, fake = 0;
3719 regnode *nscan;
3720 struct regnode_charclass_class intrnl;
3721 int f = 0;
3722 /* We use SAVEFREEPV so that when the full compile
3723 is finished perl will clean up the allocated
3724 minlens when its all done. This was we don't
3725 have to worry about freeing them when we know
3726 they wont be used, which would be a pain.
3727 */
3728 I32 *minnextp;
3729 Newx( minnextp, 1, I32 );
3730 SAVEFREEPV(minnextp);
3731
3732 if (data) {
3733 StructCopy(data, &data_fake, scan_data_t);
3734 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3735 f |= SCF_DO_SUBSTR;
3736 if (scan->flags)
304ee84b 3737 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
8aa23a47
YO
3738 data_fake.last_found=newSVsv(data->last_found);
3739 }
3740 }
3741 else
3742 data_fake.last_closep = &fake;
3743 data_fake.flags = 0;
58e23c8d 3744 data_fake.pos_delta = delta;
8aa23a47
YO
3745 if (is_inf)
3746 data_fake.flags |= SF_IS_INF;
3747 if ( flags & SCF_DO_STCLASS && !scan->flags
3748 && OP(scan) == IFMATCH ) { /* Lookahead */
3749 cl_init(pRExC_state, &intrnl);
3750 data_fake.start_class = &intrnl;
3751 f |= SCF_DO_STCLASS_AND;
3752 }
3753 if (flags & SCF_WHILEM_VISITED_POS)
3754 f |= SCF_WHILEM_VISITED_POS;
3755 next = regnext(scan);
3756 nscan = NEXTOPER(NEXTOPER(scan));
3757
3758 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3759 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3760 if (scan->flags) {
3761 if (deltanext) {
58e23c8d 3762 FAIL("Variable length lookbehind not implemented");
8aa23a47
YO
3763 }
3764 else if (*minnextp > (I32)U8_MAX) {
58e23c8d 3765 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
8aa23a47
YO
3766 }
3767 scan->flags = (U8)*minnextp;
3768 }
3769
3770 *minnextp += min;
3771
3772 if (f & SCF_DO_STCLASS_AND) {
3773 const int was = (data->start_class->flags & ANYOF_EOS);
3774
3775 cl_and(data->start_class, &intrnl);
3776 if (was)
3777 data->start_class->flags |= ANYOF_EOS;
3778 }
3779 if (data) {
3780 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3781 pars++;
3782 if (data_fake.flags & SF_HAS_EVAL)
3783 data->flags |= SF_HAS_EVAL;
3784 data->whilem_c = data_fake.whilem_c;
3785 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3786 if (RExC_rx->minlen<*minnextp)
3787 RExC_rx->minlen=*minnextp;
304ee84b 3788 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
8aa23a47
YO
3789 SvREFCNT_dec(data_fake.last_found);
3790
3791 if ( data_fake.minlen_fixed != minlenp )
3792 {
3793 data->offset_fixed= data_fake.offset_fixed;
3794 data->minlen_fixed= data_fake.minlen_fixed;
3795 data->lookbehind_fixed+= scan->flags;
3796 }
3797 if ( data_fake.minlen_float != minlenp )
3798 {
3799 data->minlen_float= data_fake.minlen_float;
3800 data->offset_float_min=data_fake.offset_float_min;
3801 data->offset_float_max=data_fake.offset_float_max;
3802 data->lookbehind_float+= scan->flags;
3803 }
3804 }
3805 }
3806
3807
40d049e4 3808 }
8aa23a47
YO
3809#endif
3810 }
3811 else if (OP(scan) == OPEN) {
3812 if (stopparen != (I32)ARG(scan))
3813 pars++;
3814 }
3815 else if (OP(scan) == CLOSE) {
3816 if (stopparen == (I32)ARG(scan)) {