This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
The revenge of the consts
[perl5.git] / regcomp.c
CommitLineData
a0d0e21e
LW
1/* regcomp.c
2 */
3
4/*
5 * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee
6 */
7
61296642
DM
8/* This file contains functions for compiling a regular expression. See
9 * also regexec.c which funnily enough, contains functions for executing
166f8a29 10 * a regular expression.
e4a054ea
DM
11 *
12 * This file is also copied at build time to ext/re/re_comp.c, where
13 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
14 * This causes the main functions to be compiled under new names and with
15 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM
16 */
17
a687059c
LW
18/* NOTE: this is derived from Henry Spencer's regexp code, and should not
19 * confused with the original package (see point 3 below). Thanks, Henry!
20 */
21
22/* Additional note: this code is very heavily munged from Henry's version
23 * in places. In some spots I've traded clarity for efficiency, so don't
24 * blame Henry for some of the lack of readability.
25 */
26
e50aee73
AD
27/* The names of the functions have been changed from regcomp and
28 * regexec to pregcomp and pregexec in order to avoid conflicts
29 * with the POSIX routines of the same names.
30*/
31
b9d5759e 32#ifdef PERL_EXT_RE_BUILD
54df2634 33#include "re_top.h"
b81d288d 34#endif
56953603 35
a687059c 36/*
e50aee73 37 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
38 *
39 * Copyright (c) 1986 by University of Toronto.
40 * Written by Henry Spencer. Not derived from licensed software.
41 *
42 * Permission is granted to anyone to use this software for any
43 * purpose on any computer system, and to redistribute it freely,
44 * subject to the following restrictions:
45 *
46 * 1. The author is not responsible for the consequences of use of
47 * this software, no matter how awful, even if they arise
48 * from defects in it.
49 *
50 * 2. The origin of this software must not be misrepresented, either
51 * by explicit claim or by omission.
52 *
53 * 3. Altered versions must be plainly marked as such, and must not
54 * be misrepresented as being the original software.
55 *
56 *
57 **** Alterations to Henry's code are...
58 ****
4bb101f2 59 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
fdf8c088 60 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007 by Larry Wall and others
a687059c 61 ****
9ef589d8
LW
62 **** You may distribute under the terms of either the GNU General Public
63 **** License or the Artistic License, as specified in the README file.
64
a687059c
LW
65 *
66 * Beware that some of this code is subtly aware of the way operator
67 * precedence is structured in regular expressions. Serious changes in
68 * regular-expression syntax might require a total rethink.
69 */
70#include "EXTERN.h"
864dbfa3 71#define PERL_IN_REGCOMP_C
a687059c 72#include "perl.h"
d06ea78c 73
acfe0abc 74#ifndef PERL_IN_XSUB_RE
d06ea78c
GS
75# include "INTERN.h"
76#endif
c277df42
IZ
77
78#define REG_COMP_C
54df2634
NC
79#ifdef PERL_IN_XSUB_RE
80# include "re_comp.h"
81#else
82# include "regcomp.h"
83#endif
a687059c 84
d4cce5f1 85#ifdef op
11343788 86#undef op
d4cce5f1 87#endif /* op */
11343788 88
fe14fcc3 89#ifdef MSDOS
7e4e8c89 90# if defined(BUGGY_MSC6)
fe14fcc3 91 /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */
7e4e8c89 92# pragma optimize("a",off)
fe14fcc3 93 /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/
7e4e8c89
NC
94# pragma optimize("w",on )
95# endif /* BUGGY_MSC6 */
fe14fcc3
LW
96#endif /* MSDOS */
97
a687059c
LW
98#ifndef STATIC
99#define STATIC static
100#endif
101
830247a4 102typedef struct RExC_state_t {
e2509266 103 U32 flags; /* are we folding, multilining? */
830247a4 104 char *precomp; /* uncompiled string. */
f8fc2ecf
YO
105 regexp *rx; /* perl core regexp structure */
106 regexp_internal *rxi; /* internal data for regexp object pprivate field */
fac92740 107 char *start; /* Start of input for compile */
830247a4
IZ
108 char *end; /* End of input for compile */
109 char *parse; /* Input-scan pointer. */
110 I32 whilem_seen; /* number of WHILEM in this expr */
fac92740 111 regnode *emit_start; /* Start of emitted-code area */
3b57cd43 112 regnode *emit_bound; /* First regnode outside of the allocated space */
ffc61ed2 113 regnode *emit; /* Code-emit pointer; &regdummy = don't = compiling */
830247a4
IZ
114 I32 naughty; /* How bad is this pattern? */
115 I32 sawback; /* Did we see \1, ...? */
116 U32 seen;
117 I32 size; /* Code size. */
c74340f9
YO
118 I32 npar; /* Capture buffer count, (OPEN). */
119 I32 cpar; /* Capture buffer count, (CLOSE). */
e2e6a0f1 120 I32 nestroot; /* root parens we are in - used by accept */
830247a4
IZ
121 I32 extralen;
122 I32 seen_zerolen;
123 I32 seen_evals;
40d049e4
YO
124 regnode **open_parens; /* pointers to open parens */
125 regnode **close_parens; /* pointers to close parens */
126 regnode *opend; /* END node in program */
02daf0ab
YO
127 I32 utf8; /* whether the pattern is utf8 or not */
128 I32 orig_utf8; /* whether the pattern was originally in utf8 */
129 /* XXX use this for future optimisation of case
130 * where pattern must be upgraded to utf8. */
6bda09f9 131 HV *charnames; /* cache of named sequences */
81714fb9 132 HV *paren_names; /* Paren names */
1f1031fe 133
40d049e4
YO
134 regnode **recurse; /* Recurse regops */
135 I32 recurse_count; /* Number of recurse regops */
830247a4
IZ
136#if ADD_TO_REGEXEC
137 char *starttry; /* -Dr: where regtry was called. */
138#define RExC_starttry (pRExC_state->starttry)
139#endif
3dab1dad 140#ifdef DEBUGGING
be8e71aa 141 const char *lastparse;
3dab1dad 142 I32 lastnum;
1f1031fe 143 AV *paren_name_list; /* idx -> name */
3dab1dad
YO
144#define RExC_lastparse (pRExC_state->lastparse)
145#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 146#define RExC_paren_name_list (pRExC_state->paren_name_list)
3dab1dad 147#endif
830247a4
IZ
148} RExC_state_t;
149
e2509266 150#define RExC_flags (pRExC_state->flags)
830247a4
IZ
151#define RExC_precomp (pRExC_state->precomp)
152#define RExC_rx (pRExC_state->rx)
f8fc2ecf 153#define RExC_rxi (pRExC_state->rxi)
fac92740 154#define RExC_start (pRExC_state->start)
830247a4
IZ
155#define RExC_end (pRExC_state->end)
156#define RExC_parse (pRExC_state->parse)
157#define RExC_whilem_seen (pRExC_state->whilem_seen)
7122b237
YO
158#ifdef RE_TRACK_PATTERN_OFFSETS
159#define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */
160#endif
830247a4 161#define RExC_emit (pRExC_state->emit)
fac92740 162#define RExC_emit_start (pRExC_state->emit_start)
3b57cd43 163#define RExC_emit_bound (pRExC_state->emit_bound)
830247a4
IZ
164#define RExC_naughty (pRExC_state->naughty)
165#define RExC_sawback (pRExC_state->sawback)
166#define RExC_seen (pRExC_state->seen)
167#define RExC_size (pRExC_state->size)
168#define RExC_npar (pRExC_state->npar)
e2e6a0f1 169#define RExC_nestroot (pRExC_state->nestroot)
830247a4
IZ
170#define RExC_extralen (pRExC_state->extralen)
171#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
172#define RExC_seen_evals (pRExC_state->seen_evals)
1aa99e6b 173#define RExC_utf8 (pRExC_state->utf8)
02daf0ab 174#define RExC_orig_utf8 (pRExC_state->orig_utf8)
fc8cd66c 175#define RExC_charnames (pRExC_state->charnames)
40d049e4
YO
176#define RExC_open_parens (pRExC_state->open_parens)
177#define RExC_close_parens (pRExC_state->close_parens)
178#define RExC_opend (pRExC_state->opend)
81714fb9 179#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
180#define RExC_recurse (pRExC_state->recurse)
181#define RExC_recurse_count (pRExC_state->recurse_count)
830247a4 182
cde0cee5 183
a687059c
LW
184#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
185#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
186 ((*s) == '{' && regcurly(s)))
a687059c 187
35c8bce7
LW
188#ifdef SPSTART
189#undef SPSTART /* dratted cpp namespace... */
190#endif
a687059c
LW
191/*
192 * Flags to be passed up and down.
193 */
a687059c 194#define WORST 0 /* Worst case. */
a3b492c3
YO
195#define HASWIDTH 0x01 /* Known to match non-null strings. */
196#define SIMPLE 0x02 /* Simple enough to be STAR/PLUS operand. */
197#define SPSTART 0x04 /* Starts with * or +. */
198#define TRYAGAIN 0x08 /* Weeded out a declaration. */
199#define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */
a687059c 200
3dab1dad
YO
201#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
202
07be1b83
YO
203/* whether trie related optimizations are enabled */
204#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
205#define TRIE_STUDY_OPT
786e8c11 206#define FULL_TRIE_STUDY
07be1b83
YO
207#define TRIE_STCLASS
208#endif
1de06328
YO
209
210
40d049e4
YO
211
212#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
213#define PBITVAL(paren) (1 << ((paren) & 7))
214#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
215#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
216#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
217
218
1de06328
YO
219/* About scan_data_t.
220
221 During optimisation we recurse through the regexp program performing
222 various inplace (keyhole style) optimisations. In addition study_chunk
223 and scan_commit populate this data structure with information about
224 what strings MUST appear in the pattern. We look for the longest
225 string that must appear for at a fixed location, and we look for the
226 longest string that may appear at a floating location. So for instance
227 in the pattern:
228
229 /FOO[xX]A.*B[xX]BAR/
230
231 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
232 strings (because they follow a .* construct). study_chunk will identify
233 both FOO and BAR as being the longest fixed and floating strings respectively.
234
235 The strings can be composites, for instance
236
237 /(f)(o)(o)/
238
239 will result in a composite fixed substring 'foo'.
240
241 For each string some basic information is maintained:
242
243 - offset or min_offset
244 This is the position the string must appear at, or not before.
245 It also implicitly (when combined with minlenp) tells us how many
246 character must match before the string we are searching.
247 Likewise when combined with minlenp and the length of the string
248 tells us how many characters must appear after the string we have
249 found.
250
251 - max_offset
252 Only used for floating strings. This is the rightmost point that
253 the string can appear at. Ifset to I32 max it indicates that the
254 string can occur infinitely far to the right.
255
256 - minlenp
257 A pointer to the minimum length of the pattern that the string
258 was found inside. This is important as in the case of positive
259 lookahead or positive lookbehind we can have multiple patterns
260 involved. Consider
261
262 /(?=FOO).*F/
263
264 The minimum length of the pattern overall is 3, the minimum length
265 of the lookahead part is 3, but the minimum length of the part that
266 will actually match is 1. So 'FOO's minimum length is 3, but the
267 minimum length for the F is 1. This is important as the minimum length
268 is used to determine offsets in front of and behind the string being
269 looked for. Since strings can be composites this is the length of the
270 pattern at the time it was commited with a scan_commit. Note that
271 the length is calculated by study_chunk, so that the minimum lengths
272 are not known until the full pattern has been compiled, thus the
273 pointer to the value.
274
275 - lookbehind
276
277 In the case of lookbehind the string being searched for can be
278 offset past the start point of the final matching string.
279 If this value was just blithely removed from the min_offset it would
280 invalidate some of the calculations for how many chars must match
281 before or after (as they are derived from min_offset and minlen and
282 the length of the string being searched for).
283 When the final pattern is compiled and the data is moved from the
284 scan_data_t structure into the regexp structure the information
285 about lookbehind is factored in, with the information that would
286 have been lost precalculated in the end_shift field for the
287 associated string.
288
289 The fields pos_min and pos_delta are used to store the minimum offset
290 and the delta to the maximum offset at the current point in the pattern.
291
292*/
2c2d71f5
JH
293
294typedef struct scan_data_t {
1de06328
YO
295 /*I32 len_min; unused */
296 /*I32 len_delta; unused */
2c2d71f5
JH
297 I32 pos_min;
298 I32 pos_delta;
299 SV *last_found;
1de06328 300 I32 last_end; /* min value, <0 unless valid. */
2c2d71f5
JH
301 I32 last_start_min;
302 I32 last_start_max;
1de06328
YO
303 SV **longest; /* Either &l_fixed, or &l_float. */
304 SV *longest_fixed; /* longest fixed string found in pattern */
305 I32 offset_fixed; /* offset where it starts */
306 I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
307 I32 lookbehind_fixed; /* is the position of the string modfied by LB */
308 SV *longest_float; /* longest floating string found in pattern */
309 I32 offset_float_min; /* earliest point in string it can appear */
310 I32 offset_float_max; /* latest point in string it can appear */
311 I32 *minlen_float; /* pointer to the minlen relevent to the string */
312 I32 lookbehind_float; /* is the position of the string modified by LB */
2c2d71f5
JH
313 I32 flags;
314 I32 whilem_c;
cb434fcc 315 I32 *last_closep;
653099ff 316 struct regnode_charclass_class *start_class;
2c2d71f5
JH
317} scan_data_t;
318
a687059c 319/*
e50aee73 320 * Forward declarations for pregcomp()'s friends.
a687059c 321 */
a0d0e21e 322
27da23d5 323static const scan_data_t zero_scan_data =
1de06328 324 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0};
c277df42
IZ
325
326#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
07be1b83
YO
327#define SF_BEFORE_SEOL 0x0001
328#define SF_BEFORE_MEOL 0x0002
c277df42
IZ
329#define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL)
330#define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL)
331
09b7f37c
CB
332#ifdef NO_UNARY_PLUS
333# define SF_FIX_SHIFT_EOL (0+2)
334# define SF_FL_SHIFT_EOL (0+4)
335#else
336# define SF_FIX_SHIFT_EOL (+2)
337# define SF_FL_SHIFT_EOL (+4)
338#endif
c277df42
IZ
339
340#define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL)
341#define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL)
342
343#define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL)
344#define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */
07be1b83
YO
345#define SF_IS_INF 0x0040
346#define SF_HAS_PAR 0x0080
347#define SF_IN_PAR 0x0100
348#define SF_HAS_EVAL 0x0200
349#define SCF_DO_SUBSTR 0x0400
653099ff
GS
350#define SCF_DO_STCLASS_AND 0x0800
351#define SCF_DO_STCLASS_OR 0x1000
352#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 353#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 354
786e8c11 355#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
e2e6a0f1 356#define SCF_SEEN_ACCEPT 0x8000
07be1b83 357
eb160463 358#define UTF (RExC_utf8 != 0)
bbe252da
YO
359#define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
360#define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
a0ed51b3 361
ffc61ed2 362#define OOB_UNICODE 12345678
93733859 363#define OOB_NAMEDCLASS -1
b8c5462f 364
a0ed51b3
LW
365#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
366#define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b)
367
8615cb43 368
b45f050a
JF
369/* length of regex to show in messages that don't mark a position within */
370#define RegexLengthToShowInErrorMessages 127
371
372/*
373 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
374 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
375 * op/pragma/warn/regcomp.
376 */
7253e4e3
RK
377#define MARKER1 "<-- HERE" /* marker as it appears in the description */
378#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 379
7253e4e3 380#define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/"
b45f050a
JF
381
382/*
383 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
384 * arg. Show regex, up to a maximum length. If it's too long, chop and add
385 * "...".
386 */
58e23c8d 387#define _FAIL(code) STMT_START { \
bfed75c6 388 const char *ellipses = ""; \
ccb2c380
MP
389 IV len = RExC_end - RExC_precomp; \
390 \
391 if (!SIZE_ONLY) \
392 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
393 if (len > RegexLengthToShowInErrorMessages) { \
394 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
395 len = RegexLengthToShowInErrorMessages - 10; \
396 ellipses = "..."; \
397 } \
58e23c8d 398 code; \
ccb2c380 399} STMT_END
8615cb43 400
58e23c8d
YO
401#define FAIL(msg) _FAIL( \
402 Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
403 msg, (int)len, RExC_precomp, ellipses))
404
405#define FAIL2(msg,arg) _FAIL( \
406 Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
407 arg, (int)len, RExC_precomp, ellipses))
408
b45f050a 409/*
b45f050a
JF
410 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
411 */
ccb2c380 412#define Simple_vFAIL(m) STMT_START { \
a28509cc 413 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
414 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
415 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
416} STMT_END
b45f050a
JF
417
418/*
419 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
420 */
ccb2c380
MP
421#define vFAIL(m) STMT_START { \
422 if (!SIZE_ONLY) \
423 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
424 Simple_vFAIL(m); \
425} STMT_END
b45f050a
JF
426
427/*
428 * Like Simple_vFAIL(), but accepts two arguments.
429 */
ccb2c380 430#define Simple_vFAIL2(m,a1) STMT_START { \
a28509cc 431 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
432 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \
433 (int)offset, RExC_precomp, RExC_precomp + offset); \
434} STMT_END
b45f050a
JF
435
436/*
437 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
438 */
ccb2c380
MP
439#define vFAIL2(m,a1) STMT_START { \
440 if (!SIZE_ONLY) \
441 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
442 Simple_vFAIL2(m, a1); \
443} STMT_END
b45f050a
JF
444
445
446/*
447 * Like Simple_vFAIL(), but accepts three arguments.
448 */
ccb2c380 449#define Simple_vFAIL3(m, a1, a2) STMT_START { \
a28509cc 450 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
451 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \
452 (int)offset, RExC_precomp, RExC_precomp + offset); \
453} STMT_END
b45f050a
JF
454
455/*
456 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
457 */
ccb2c380
MP
458#define vFAIL3(m,a1,a2) STMT_START { \
459 if (!SIZE_ONLY) \
460 SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \
461 Simple_vFAIL3(m, a1, a2); \
462} STMT_END
b45f050a
JF
463
464/*
465 * Like Simple_vFAIL(), but accepts four arguments.
466 */
ccb2c380 467#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
a28509cc 468 const IV offset = RExC_parse - RExC_precomp; \
ccb2c380
MP
469 S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \
470 (int)offset, RExC_precomp, RExC_precomp + offset); \
471} STMT_END
b45f050a 472
ccb2c380 473#define vWARN(loc,m) STMT_START { \
a28509cc 474 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
475 Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \
476 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
477} STMT_END
478
479#define vWARNdep(loc,m) STMT_START { \
a28509cc 480 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
481 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
482 "%s" REPORT_LOCATION, \
483 m, (int)offset, RExC_precomp, RExC_precomp + offset); \
484} STMT_END
485
486
487#define vWARN2(loc, m, a1) STMT_START { \
a28509cc 488 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
489 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
490 a1, (int)offset, RExC_precomp, RExC_precomp + offset); \
491} STMT_END
492
493#define vWARN3(loc, m, a1, a2) STMT_START { \
a28509cc 494 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
495 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
496 a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \
497} STMT_END
498
499#define vWARN4(loc, m, a1, a2, a3) STMT_START { \
a28509cc 500 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
501 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
502 a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \
503} STMT_END
504
505#define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \
a28509cc 506 const IV offset = loc - RExC_precomp; \
ccb2c380
MP
507 Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \
508 a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \
509} STMT_END
9d1d55b5 510
8615cb43 511
cd439c50 512/* Allow for side effects in s */
ccb2c380
MP
513#define REGC(c,s) STMT_START { \
514 if (!SIZE_ONLY) *(s) = (c); else (void)(s); \
515} STMT_END
cd439c50 516
fac92740
MJD
517/* Macros for recording node offsets. 20001227 mjd@plover.com
518 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
519 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
520 * Element 0 holds the number n.
07be1b83 521 * Position is 1 indexed.
fac92740 522 */
7122b237
YO
523#ifndef RE_TRACK_PATTERN_OFFSETS
524#define Set_Node_Offset_To_R(node,byte)
525#define Set_Node_Offset(node,byte)
526#define Set_Cur_Node_Offset
527#define Set_Node_Length_To_R(node,len)
528#define Set_Node_Length(node,len)
529#define Set_Node_Cur_Length(node)
530#define Node_Offset(n)
531#define Node_Length(n)
532#define Set_Node_Offset_Length(node,offset,len)
533#define ProgLen(ri) ri->u.proglen
534#define SetProgLen(ri,x) ri->u.proglen = x
535#else
536#define ProgLen(ri) ri->u.offsets[0]
537#define SetProgLen(ri,x) ri->u.offsets[0] = x
ccb2c380
MP
538#define Set_Node_Offset_To_R(node,byte) STMT_START { \
539 if (! SIZE_ONLY) { \
540 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
2a49f0f5 541 __LINE__, (int)(node), (int)(byte))); \
ccb2c380 542 if((node) < 0) { \
551405c4 543 Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
ccb2c380
MP
544 } else { \
545 RExC_offsets[2*(node)-1] = (byte); \
546 } \
547 } \
548} STMT_END
549
550#define Set_Node_Offset(node,byte) \
551 Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start)
552#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
553
554#define Set_Node_Length_To_R(node,len) STMT_START { \
555 if (! SIZE_ONLY) { \
556 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 557 __LINE__, (int)(node), (int)(len))); \
ccb2c380 558 if((node) < 0) { \
551405c4 559 Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \
ccb2c380
MP
560 } else { \
561 RExC_offsets[2*(node)] = (len); \
562 } \
563 } \
564} STMT_END
565
566#define Set_Node_Length(node,len) \
567 Set_Node_Length_To_R((node)-RExC_emit_start, len)
568#define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len)
569#define Set_Node_Cur_Length(node) \
570 Set_Node_Length(node, RExC_parse - parse_start)
fac92740
MJD
571
572/* Get offsets and lengths */
573#define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1])
574#define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)])
575
07be1b83
YO
576#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
577 Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \
578 Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \
579} STMT_END
7122b237 580#endif
07be1b83
YO
581
582#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
583#define EXPERIMENTAL_INPLACESCAN
7122b237 584#endif /*RE_TRACK_PATTERN_OFFSETS*/
07be1b83 585
304ee84b
YO
586#define DEBUG_STUDYDATA(str,data,depth) \
587DEBUG_OPTIMISE_MORE_r(if(data){ \
1de06328 588 PerlIO_printf(Perl_debug_log, \
304ee84b
YO
589 "%*s" str "Pos:%"IVdf"/%"IVdf \
590 " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \
1de06328
YO
591 (int)(depth)*2, "", \
592 (IV)((data)->pos_min), \
593 (IV)((data)->pos_delta), \
304ee84b 594 (UV)((data)->flags), \
1de06328 595 (IV)((data)->whilem_c), \
304ee84b
YO
596 (IV)((data)->last_closep ? *((data)->last_closep) : -1), \
597 is_inf ? "INF " : "" \
1de06328
YO
598 ); \
599 if ((data)->last_found) \
600 PerlIO_printf(Perl_debug_log, \
601 "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
602 " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
603 SvPVX_const((data)->last_found), \
604 (IV)((data)->last_end), \
605 (IV)((data)->last_start_min), \
606 (IV)((data)->last_start_max), \
607 ((data)->longest && \
608 (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
609 SvPVX_const((data)->longest_fixed), \
610 (IV)((data)->offset_fixed), \
611 ((data)->longest && \
612 (data)->longest==&((data)->longest_float)) ? "*" : "", \
613 SvPVX_const((data)->longest_float), \
614 (IV)((data)->offset_float_min), \
615 (IV)((data)->offset_float_max) \
616 ); \
617 PerlIO_printf(Perl_debug_log,"\n"); \
618});
619
acfe0abc 620static void clear_re(pTHX_ void *r);
4327152a 621
653099ff 622/* Mark that we cannot extend a found fixed substring at this point.
786e8c11 623 Update the longest found anchored substring and the longest found
653099ff
GS
624 floating substrings if needed. */
625
4327152a 626STATIC void
304ee84b 627S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf)
c277df42 628{
e1ec3a88
AL
629 const STRLEN l = CHR_SVLEN(data->last_found);
630 const STRLEN old_l = CHR_SVLEN(*data->longest);
1de06328 631 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 632
c277df42 633 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
6b43b216 634 SvSetMagicSV(*data->longest, data->last_found);
c277df42
IZ
635 if (*data->longest == data->longest_fixed) {
636 data->offset_fixed = l ? data->last_start_min : data->pos_min;
637 if (data->flags & SF_BEFORE_EOL)
b81d288d 638 data->flags
c277df42
IZ
639 |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
640 else
641 data->flags &= ~SF_FIX_BEFORE_EOL;
1de06328
YO
642 data->minlen_fixed=minlenp;
643 data->lookbehind_fixed=0;
a0ed51b3 644 }
304ee84b 645 else { /* *data->longest == data->longest_float */
c277df42 646 data->offset_float_min = l ? data->last_start_min : data->pos_min;
b81d288d
AB
647 data->offset_float_max = (l
648 ? data->last_start_max
c277df42 649 : data->pos_min + data->pos_delta);
304ee84b 650 if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX)
9051bda5 651 data->offset_float_max = I32_MAX;
c277df42 652 if (data->flags & SF_BEFORE_EOL)
b81d288d 653 data->flags
c277df42
IZ
654 |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
655 else
656 data->flags &= ~SF_FL_BEFORE_EOL;
1de06328
YO
657 data->minlen_float=minlenp;
658 data->lookbehind_float=0;
c277df42
IZ
659 }
660 }
661 SvCUR_set(data->last_found, 0);
0eda9292 662 {
a28509cc 663 SV * const sv = data->last_found;
097eb12c
AL
664 if (SvUTF8(sv) && SvMAGICAL(sv)) {
665 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
666 if (mg)
667 mg->mg_len = 0;
668 }
0eda9292 669 }
c277df42
IZ
670 data->last_end = -1;
671 data->flags &= ~SF_BEFORE_EOL;
bcdf7404 672 DEBUG_STUDYDATA("commit: ",data,0);
c277df42
IZ
673}
674
653099ff
GS
675/* Can match anything (initialization) */
676STATIC void
097eb12c 677S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 678{
653099ff 679 ANYOF_CLASS_ZERO(cl);
f8bef550 680 ANYOF_BITMAP_SETALL(cl);
1aa99e6b 681 cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL;
653099ff
GS
682 if (LOC)
683 cl->flags |= ANYOF_LOCALE;
684}
685
686/* Can match anything (initialization) */
687STATIC int
5f66b61c 688S_cl_is_anything(const struct regnode_charclass_class *cl)
653099ff
GS
689{
690 int value;
691
aaa51d5e 692 for (value = 0; value <= ANYOF_MAX; value += 2)
653099ff
GS
693 if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1))
694 return 1;
1aa99e6b
IH
695 if (!(cl->flags & ANYOF_UNICODE_ALL))
696 return 0;
10edeb5d 697 if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
f8bef550 698 return 0;
653099ff
GS
699 return 1;
700}
701
702/* Can match anything (initialization) */
703STATIC void
097eb12c 704S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 705{
8ecf7187 706 Zero(cl, 1, struct regnode_charclass_class);
653099ff 707 cl->type = ANYOF;
830247a4 708 cl_anything(pRExC_state, cl);
653099ff
GS
709}
710
711STATIC void
097eb12c 712S_cl_init_zero(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl)
653099ff 713{
8ecf7187 714 Zero(cl, 1, struct regnode_charclass_class);
653099ff 715 cl->type = ANYOF;
830247a4 716 cl_anything(pRExC_state, cl);
653099ff
GS
717 if (LOC)
718 cl->flags |= ANYOF_LOCALE;
719}
720
721/* 'And' a given class with another one. Can create false positives */
722/* We assume that cl is not inverted */
723STATIC void
5f66b61c 724S_cl_and(struct regnode_charclass_class *cl,
a28509cc 725 const struct regnode_charclass_class *and_with)
653099ff 726{
40d049e4
YO
727
728 assert(and_with->type == ANYOF);
653099ff
GS
729 if (!(and_with->flags & ANYOF_CLASS)
730 && !(cl->flags & ANYOF_CLASS)
731 && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
732 && !(and_with->flags & ANYOF_FOLD)
733 && !(cl->flags & ANYOF_FOLD)) {
734 int i;
735
736 if (and_with->flags & ANYOF_INVERT)
737 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
738 cl->bitmap[i] &= ~and_with->bitmap[i];
739 else
740 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
741 cl->bitmap[i] &= and_with->bitmap[i];
742 } /* XXXX: logic is complicated otherwise, leave it along for a moment. */
743 if (!(and_with->flags & ANYOF_EOS))
744 cl->flags &= ~ANYOF_EOS;
1aa99e6b 745
14ebb1a2
JH
746 if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE &&
747 !(and_with->flags & ANYOF_INVERT)) {
1aa99e6b
IH
748 cl->flags &= ~ANYOF_UNICODE_ALL;
749 cl->flags |= ANYOF_UNICODE;
750 ARG_SET(cl, ARG(and_with));
751 }
14ebb1a2
JH
752 if (!(and_with->flags & ANYOF_UNICODE_ALL) &&
753 !(and_with->flags & ANYOF_INVERT))
1aa99e6b 754 cl->flags &= ~ANYOF_UNICODE_ALL;
14ebb1a2
JH
755 if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) &&
756 !(and_with->flags & ANYOF_INVERT))
1aa99e6b 757 cl->flags &= ~ANYOF_UNICODE;
653099ff
GS
758}
759
760/* 'OR' a given class with another one. Can create false positives */
761/* We assume that cl is not inverted */
762STATIC void
097eb12c 763S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with)
653099ff 764{
653099ff
GS
765 if (or_with->flags & ANYOF_INVERT) {
766 /* We do not use
767 * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2))
768 * <= (B1 | !B2) | (CL1 | !CL2)
769 * which is wasteful if CL2 is small, but we ignore CL2:
770 * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1
771 * XXXX Can we handle case-fold? Unclear:
772 * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) =
773 * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i'))
774 */
775 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
776 && !(or_with->flags & ANYOF_FOLD)
777 && !(cl->flags & ANYOF_FOLD) ) {
778 int i;
779
780 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
781 cl->bitmap[i] |= ~or_with->bitmap[i];
782 } /* XXXX: logic is complicated otherwise */
783 else {
830247a4 784 cl_anything(pRExC_state, cl);
653099ff
GS
785 }
786 } else {
787 /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */
788 if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
b81d288d 789 && (!(or_with->flags & ANYOF_FOLD)
653099ff
GS
790 || (cl->flags & ANYOF_FOLD)) ) {
791 int i;
792
793 /* OR char bitmap and class bitmap separately */
794 for (i = 0; i < ANYOF_BITMAP_SIZE; i++)
795 cl->bitmap[i] |= or_with->bitmap[i];
796 if (or_with->flags & ANYOF_CLASS) {
797 for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++)
798 cl->classflags[i] |= or_with->classflags[i];
799 cl->flags |= ANYOF_CLASS;
800 }
801 }
802 else { /* XXXX: logic is complicated, leave it along for a moment. */
830247a4 803 cl_anything(pRExC_state, cl);
653099ff
GS
804 }
805 }
806 if (or_with->flags & ANYOF_EOS)
807 cl->flags |= ANYOF_EOS;
1aa99e6b
IH
808
809 if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE &&
810 ARG(cl) != ARG(or_with)) {
811 cl->flags |= ANYOF_UNICODE_ALL;
812 cl->flags &= ~ANYOF_UNICODE;
813 }
814 if (or_with->flags & ANYOF_UNICODE_ALL) {
815 cl->flags |= ANYOF_UNICODE_ALL;
816 cl->flags &= ~ANYOF_UNICODE;
817 }
653099ff
GS
818}
819
a3621e74
YO
820#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
821#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
822#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
823#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 )
824
3dab1dad
YO
825
826#ifdef DEBUGGING
07be1b83 827/*
2b8b4781
NC
828 dump_trie(trie,widecharmap,revcharmap)
829 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
830 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
831
832 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
833 The _interim_ variants are used for debugging the interim
834 tables that are used to generate the final compressed
835 representation which is what dump_trie expects.
836
3dab1dad
YO
837 Part of the reason for their existance is to provide a form
838 of documentation as to how the different representations function.
07be1b83
YO
839
840*/
3dab1dad
YO
841
842/*
3dab1dad
YO
843 Dumps the final compressed table form of the trie to Perl_debug_log.
844 Used for debugging make_trie().
845*/
846
847STATIC void
2b8b4781
NC
848S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
849 AV *revcharmap, U32 depth)
3dab1dad
YO
850{
851 U32 state;
ab3bbdeb 852 SV *sv=sv_newmortal();
55eed653 853 int colwidth= widecharmap ? 6 : 4;
3dab1dad
YO
854 GET_RE_DEBUG_FLAGS_DECL;
855
ab3bbdeb 856
3dab1dad
YO
857 PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
858 (int)depth * 2 + 2,"",
859 "Match","Base","Ofs" );
860
861 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 862 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 863 if ( tmp ) {
ab3bbdeb
YO
864 PerlIO_printf( Perl_debug_log, "%*s",
865 colwidth,
ddc5bc0f 866 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
867 PL_colors[0], PL_colors[1],
868 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
869 PERL_PV_ESCAPE_FIRSTCHAR
870 )
871 );
3dab1dad
YO
872 }
873 }
874 PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
875 (int)depth * 2 + 2,"");
876
877 for( state = 0 ; state < trie->uniquecharcount ; state++ )
ab3bbdeb 878 PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
3dab1dad
YO
879 PerlIO_printf( Perl_debug_log, "\n");
880
1e2e3d02 881 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 882 const U32 base = trie->states[ state ].trans.base;
3dab1dad
YO
883
884 PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
885
886 if ( trie->states[ state ].wordnum ) {
887 PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum );
888 } else {
889 PerlIO_printf( Perl_debug_log, "%6s", "" );
890 }
891
892 PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base );
893
894 if ( base ) {
895 U32 ofs = 0;
896
897 while( ( base + ofs < trie->uniquecharcount ) ||
898 ( base + ofs - trie->uniquecharcount < trie->lasttrans
899 && trie->trans[ base + ofs - trie->uniquecharcount ].check != state))
900 ofs++;
901
902 PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs);
903
904 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
905 if ( ( base + ofs >= trie->uniquecharcount ) &&
906 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
907 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
908 {
ab3bbdeb
YO
909 PerlIO_printf( Perl_debug_log, "%*"UVXf,
910 colwidth,
3dab1dad
YO
911 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
912 } else {
ab3bbdeb 913 PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
3dab1dad
YO
914 }
915 }
916
917 PerlIO_printf( Perl_debug_log, "]");
918
919 }
920 PerlIO_printf( Perl_debug_log, "\n" );
921 }
922}
923/*
3dab1dad
YO
924 Dumps a fully constructed but uncompressed trie in list form.
925 List tries normally only are used for construction when the number of
926 possible chars (trie->uniquecharcount) is very high.
927 Used for debugging make_trie().
928*/
929STATIC void
55eed653 930S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
931 HV *widecharmap, AV *revcharmap, U32 next_alloc,
932 U32 depth)
3dab1dad
YO
933{
934 U32 state;
ab3bbdeb 935 SV *sv=sv_newmortal();
55eed653 936 int colwidth= widecharmap ? 6 : 4;
3dab1dad
YO
937 GET_RE_DEBUG_FLAGS_DECL;
938 /* print out the table precompression. */
ab3bbdeb
YO
939 PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
940 (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
941 "------:-----+-----------------\n" );
3dab1dad
YO
942
943 for( state=1 ; state < next_alloc ; state ++ ) {
944 U16 charid;
945
ab3bbdeb 946 PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
3dab1dad
YO
947 (int)depth * 2 + 2,"", (UV)state );
948 if ( ! trie->states[ state ].wordnum ) {
949 PerlIO_printf( Perl_debug_log, "%5s| ","");
950 } else {
951 PerlIO_printf( Perl_debug_log, "W%4x| ",
952 trie->states[ state ].wordnum
953 );
954 }
955 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
2b8b4781 956 SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
ab3bbdeb
YO
957 if ( tmp ) {
958 PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
959 colwidth,
ddc5bc0f 960 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
961 PL_colors[0], PL_colors[1],
962 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
963 PERL_PV_ESCAPE_FIRSTCHAR
964 ) ,
1e2e3d02
YO
965 TRIE_LIST_ITEM(state,charid).forid,
966 (UV)TRIE_LIST_ITEM(state,charid).newstate
967 );
968 if (!(charid % 10))
664e119d
RGS
969 PerlIO_printf(Perl_debug_log, "\n%*s| ",
970 (int)((depth * 2) + 14), "");
1e2e3d02 971 }
ab3bbdeb
YO
972 }
973 PerlIO_printf( Perl_debug_log, "\n");
3dab1dad
YO
974 }
975}
976
977/*
3dab1dad
YO
978 Dumps a fully constructed but uncompressed trie in table form.
979 This is the normal DFA style state transition table, with a few
980 twists to facilitate compression later.
981 Used for debugging make_trie().
982*/
983STATIC void
55eed653 984S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
985 HV *widecharmap, AV *revcharmap, U32 next_alloc,
986 U32 depth)
3dab1dad
YO
987{
988 U32 state;
989 U16 charid;
ab3bbdeb 990 SV *sv=sv_newmortal();
55eed653 991 int colwidth= widecharmap ? 6 : 4;
3dab1dad
YO
992 GET_RE_DEBUG_FLAGS_DECL;
993
994 /*
995 print out the table precompression so that we can do a visual check
996 that they are identical.
997 */
998
999 PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
1000
1001 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 1002 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 1003 if ( tmp ) {
ab3bbdeb
YO
1004 PerlIO_printf( Perl_debug_log, "%*s",
1005 colwidth,
ddc5bc0f 1006 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
1007 PL_colors[0], PL_colors[1],
1008 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1009 PERL_PV_ESCAPE_FIRSTCHAR
1010 )
1011 );
3dab1dad
YO
1012 }
1013 }
1014
1015 PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
1016
1017 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb 1018 PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
3dab1dad
YO
1019 }
1020
1021 PerlIO_printf( Perl_debug_log, "\n" );
1022
1023 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
1024
1025 PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ",
1026 (int)depth * 2 + 2,"",
1027 (UV)TRIE_NODENUM( state ) );
1028
1029 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
1030 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
1031 if (v)
1032 PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
1033 else
1034 PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
3dab1dad
YO
1035 }
1036 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
1037 PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
1038 } else {
1039 PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check,
1040 trie->states[ TRIE_NODENUM( state ) ].wordnum );
1041 }
1042 }
07be1b83 1043}
3dab1dad
YO
1044
1045#endif
1046
786e8c11
YO
1047/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
1048 startbranch: the first branch in the whole branch sequence
1049 first : start branch of sequence of branch-exact nodes.
1050 May be the same as startbranch
1051 last : Thing following the last branch.
1052 May be the same as tail.
1053 tail : item following the branch sequence
1054 count : words in the sequence
1055 flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
1056 depth : indent depth
3dab1dad 1057
786e8c11 1058Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 1059
786e8c11
YO
1060A trie is an N'ary tree where the branches are determined by digital
1061decomposition of the key. IE, at the root node you look up the 1st character and
1062follow that branch repeat until you find the end of the branches. Nodes can be
1063marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 1064
786e8c11 1065 /he|she|his|hers/
72f13be8 1066
786e8c11
YO
1067would convert into the following structure. Numbers represent states, letters
1068following numbers represent valid transitions on the letter from that state, if
1069the number is in square brackets it represents an accepting state, otherwise it
1070will be in parenthesis.
07be1b83 1071
786e8c11
YO
1072 +-h->+-e->[3]-+-r->(8)-+-s->[9]
1073 | |
1074 | (2)
1075 | |
1076 (1) +-i->(6)-+-s->[7]
1077 |
1078 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 1079
786e8c11
YO
1080 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
1081
1082This shows that when matching against the string 'hers' we will begin at state 1
1083read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
1084then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
1085is also accepting. Thus we know that we can match both 'he' and 'hers' with a
1086single traverse. We store a mapping from accepting to state to which word was
1087matched, and then when we have multiple possibilities we try to complete the
1088rest of the regex in the order in which they occured in the alternation.
1089
1090The only prior NFA like behaviour that would be changed by the TRIE support is
1091the silent ignoring of duplicate alternations which are of the form:
1092
1093 / (DUPE|DUPE) X? (?{ ... }) Y /x
1094
1095Thus EVAL blocks follwing a trie may be called a different number of times with
1096and without the optimisation. With the optimisations dupes will be silently
1097ignored. This inconsistant behaviour of EVAL type nodes is well established as
1098the following demonstrates:
1099
1100 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
1101
1102which prints out 'word' three times, but
1103
1104 'words'=~/(word|word|word)(?{ print $1 })S/
1105
1106which doesnt print it out at all. This is due to other optimisations kicking in.
1107
1108Example of what happens on a structural level:
1109
1110The regexp /(ac|ad|ab)+/ will produce the folowing debug output:
1111
1112 1: CURLYM[1] {1,32767}(18)
1113 5: BRANCH(8)
1114 6: EXACT <ac>(16)
1115 8: BRANCH(11)
1116 9: EXACT <ad>(16)
1117 11: BRANCH(14)
1118 12: EXACT <ab>(16)
1119 16: SUCCEED(0)
1120 17: NOTHING(18)
1121 18: END(0)
1122
1123This would be optimizable with startbranch=5, first=5, last=16, tail=16
1124and should turn into:
1125
1126 1: CURLYM[1] {1,32767}(18)
1127 5: TRIE(16)
1128 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
1129 <ac>
1130 <ad>
1131 <ab>
1132 16: SUCCEED(0)
1133 17: NOTHING(18)
1134 18: END(0)
1135
1136Cases where tail != last would be like /(?foo|bar)baz/:
1137
1138 1: BRANCH(4)
1139 2: EXACT <foo>(8)
1140 4: BRANCH(7)
1141 5: EXACT <bar>(8)
1142 7: TAIL(8)
1143 8: EXACT <baz>(10)
1144 10: END(0)
1145
1146which would be optimizable with startbranch=1, first=1, last=7, tail=8
1147and would end up looking like:
1148
1149 1: TRIE(8)
1150 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
1151 <foo>
1152 <bar>
1153 7: TAIL(8)
1154 8: EXACT <baz>(10)
1155 10: END(0)
1156
1157 d = uvuni_to_utf8_flags(d, uv, 0);
1158
1159is the recommended Unicode-aware way of saying
1160
1161 *(d++) = uv;
1162*/
1163
1e2e3d02 1164#define TRIE_STORE_REVCHAR \
786e8c11 1165 STMT_START { \
1e2e3d02 1166 SV *tmp = newSVpvs(""); \
786e8c11 1167 if (UTF) SvUTF8_on(tmp); \
1e2e3d02 1168 Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
2b8b4781 1169 av_push( revcharmap, tmp ); \
786e8c11
YO
1170 } STMT_END
1171
1172#define TRIE_READ_CHAR STMT_START { \
1173 wordlen++; \
1174 if ( UTF ) { \
1175 if ( folder ) { \
1176 if ( foldlen > 0 ) { \
1177 uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \
1178 foldlen -= len; \
1179 scan += len; \
1180 len = 0; \
1181 } else { \
1182 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1183 uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \
1184 foldlen -= UNISKIP( uvc ); \
1185 scan = foldbuf + UNISKIP( uvc ); \
1186 } \
1187 } else { \
1188 uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\
1189 } \
1190 } else { \
1191 uvc = (U32)*uc; \
1192 len = 1; \
1193 } \
1194} STMT_END
1195
1196
1197
1198#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
1199 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
f9003953
NC
1200 U32 ging = TRIE_LIST_LEN( state ) *= 2; \
1201 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
786e8c11
YO
1202 } \
1203 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
1204 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
1205 TRIE_LIST_CUR( state )++; \
1206} STMT_END
07be1b83 1207
786e8c11
YO
1208#define TRIE_LIST_NEW(state) STMT_START { \
1209 Newxz( trie->states[ state ].trans.list, \
1210 4, reg_trie_trans_le ); \
1211 TRIE_LIST_CUR( state ) = 1; \
1212 TRIE_LIST_LEN( state ) = 4; \
1213} STMT_END
07be1b83 1214
786e8c11
YO
1215#define TRIE_HANDLE_WORD(state) STMT_START { \
1216 U16 dupe= trie->states[ state ].wordnum; \
1217 regnode * const noper_next = regnext( noper ); \
1218 \
1219 if (trie->wordlen) \
1220 trie->wordlen[ curword ] = wordlen; \
1221 DEBUG_r({ \
1222 /* store the word for dumping */ \
1223 SV* tmp; \
1224 if (OP(noper) != NOTHING) \
1225 tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
1226 else \
1227 tmp = newSVpvn( "", 0 ); \
1228 if ( UTF ) SvUTF8_on( tmp ); \
2b8b4781 1229 av_push( trie_words, tmp ); \
786e8c11
YO
1230 }); \
1231 \
1232 curword++; \
1233 \
1234 if ( noper_next < tail ) { \
1235 if (!trie->jump) \
c944940b 1236 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
7f69552c 1237 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
1238 if (!jumper) \
1239 jumper = noper_next; \
1240 if (!nextbranch) \
1241 nextbranch= regnext(cur); \
1242 } \
1243 \
1244 if ( dupe ) { \
1245 /* So it's a dupe. This means we need to maintain a */\
1246 /* linked-list from the first to the next. */\
1247 /* we only allocate the nextword buffer when there */\
1248 /* a dupe, so first time we have to do the allocation */\
1249 if (!trie->nextword) \
c944940b 1250 trie->nextword = (U16 *) \
446bd890 1251 PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
786e8c11
YO
1252 while ( trie->nextword[dupe] ) \
1253 dupe= trie->nextword[dupe]; \
1254 trie->nextword[dupe]= curword; \
1255 } else { \
1256 /* we haven't inserted this word yet. */ \
1257 trie->states[ state ].wordnum = curword; \
1258 } \
1259} STMT_END
07be1b83 1260
3dab1dad 1261
786e8c11
YO
1262#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
1263 ( ( base + charid >= ucharcount \
1264 && base + charid < ubound \
1265 && state == trie->trans[ base - ucharcount + charid ].check \
1266 && trie->trans[ base - ucharcount + charid ].next ) \
1267 ? trie->trans[ base - ucharcount + charid ].next \
1268 : ( state==1 ? special : 0 ) \
1269 )
3dab1dad 1270
786e8c11
YO
1271#define MADE_TRIE 1
1272#define MADE_JUMP_TRIE 2
1273#define MADE_EXACT_TRIE 4
3dab1dad 1274
a3621e74 1275STATIC I32
786e8c11 1276S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
a3621e74 1277{
27da23d5 1278 dVAR;
a3621e74
YO
1279 /* first pass, loop through and scan words */
1280 reg_trie_data *trie;
55eed653 1281 HV *widecharmap = NULL;
2b8b4781 1282 AV *revcharmap = newAV();
a3621e74 1283 regnode *cur;
9f7f3913 1284 const U32 uniflags = UTF8_ALLOW_DEFAULT;
a3621e74
YO
1285 STRLEN len = 0;
1286 UV uvc = 0;
1287 U16 curword = 0;
1288 U32 next_alloc = 0;
786e8c11
YO
1289 regnode *jumper = NULL;
1290 regnode *nextbranch = NULL;
7f69552c 1291 regnode *convert = NULL;
a3621e74 1292 /* we just use folder as a flag in utf8 */
e1ec3a88 1293 const U8 * const folder = ( flags == EXACTF
a3621e74
YO
1294 ? PL_fold
1295 : ( flags == EXACTFL
1296 ? PL_fold_locale
1297 : NULL
1298 )
1299 );
1300
2b8b4781
NC
1301#ifdef DEBUGGING
1302 const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
1303 AV *trie_words = NULL;
1304 /* along with revcharmap, this only used during construction but both are
1305 * useful during debugging so we store them in the struct when debugging.
8e11feef 1306 */
2b8b4781
NC
1307#else
1308 const U32 data_slot = add_data( pRExC_state, 2, "tu" );
3dab1dad 1309 STRLEN trie_charcount=0;
3dab1dad 1310#endif
2b8b4781 1311 SV *re_trie_maxbuff;
a3621e74 1312 GET_RE_DEBUG_FLAGS_DECL;
72f13be8
YO
1313#ifndef DEBUGGING
1314 PERL_UNUSED_ARG(depth);
1315#endif
a3621e74 1316
c944940b 1317 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 1318 trie->refcount = 1;
3dab1dad 1319 trie->startstate = 1;
786e8c11 1320 trie->wordcount = word_count;
f8fc2ecf 1321 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 1322 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
3dab1dad 1323 if (!(UTF && folder))
c944940b 1324 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
a3621e74 1325 DEBUG_r({
2b8b4781 1326 trie_words = newAV();
a3621e74 1327 });
a3621e74 1328
0111c4fd 1329 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
a3621e74 1330 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 1331 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 1332 }
3dab1dad
YO
1333 DEBUG_OPTIMISE_r({
1334 PerlIO_printf( Perl_debug_log,
786e8c11 1335 "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
3dab1dad
YO
1336 (int)depth * 2 + 2, "",
1337 REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
786e8c11 1338 REG_NODE_NUM(last), REG_NODE_NUM(tail),
85c3142d 1339 (int)depth);
3dab1dad 1340 });
7f69552c
YO
1341
1342 /* Find the node we are going to overwrite */
1343 if ( first == startbranch && OP( last ) != BRANCH ) {
1344 /* whole branch chain */
1345 convert = first;
1346 } else {
1347 /* branch sub-chain */
1348 convert = NEXTOPER( first );
1349 }
1350
a3621e74
YO
1351 /* -- First loop and Setup --
1352
1353 We first traverse the branches and scan each word to determine if it
1354 contains widechars, and how many unique chars there are, this is
1355 important as we have to build a table with at least as many columns as we
1356 have unique chars.
1357
1358 We use an array of integers to represent the character codes 0..255
1359 (trie->charmap) and we use a an HV* to store unicode characters. We use the
1360 native representation of the character value as the key and IV's for the
1361 coded index.
1362
1363 *TODO* If we keep track of how many times each character is used we can
1364 remap the columns so that the table compression later on is more
1365 efficient in terms of memory by ensuring most common value is in the
1366 middle and the least common are on the outside. IMO this would be better
1367 than a most to least common mapping as theres a decent chance the most
1368 common letter will share a node with the least common, meaning the node
1369 will not be compressable. With a middle is most common approach the worst
1370 case is when we have the least common nodes twice.
1371
1372 */
1373
a3621e74 1374 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
c445ea15 1375 regnode * const noper = NEXTOPER( cur );
e1ec3a88 1376 const U8 *uc = (U8*)STRING( noper );
a28509cc 1377 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1378 STRLEN foldlen = 0;
1379 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
2af232bd 1380 const U8 *scan = (U8*)NULL;
07be1b83 1381 U32 wordlen = 0; /* required init */
02daf0ab
YO
1382 STRLEN chars = 0;
1383 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/
a3621e74 1384
3dab1dad
YO
1385 if (OP(noper) == NOTHING) {
1386 trie->minlen= 0;
1387 continue;
1388 }
02daf0ab
YO
1389 if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */
1390 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
1391 regardless of encoding */
1392
a3621e74 1393 for ( ; uc < e ; uc += len ) {
3dab1dad 1394 TRIE_CHARCOUNT(trie)++;
a3621e74 1395 TRIE_READ_CHAR;
3dab1dad 1396 chars++;
a3621e74
YO
1397 if ( uvc < 256 ) {
1398 if ( !trie->charmap[ uvc ] ) {
1399 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
1400 if ( folder )
1401 trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ];
3dab1dad 1402 TRIE_STORE_REVCHAR;
a3621e74 1403 }
02daf0ab
YO
1404 if ( set_bit ) {
1405 /* store the codepoint in the bitmap, and if its ascii
1406 also store its folded equivelent. */
1407 TRIE_BITMAP_SET(trie,uvc);
1408 if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]);
1409 set_bit = 0; /* We've done our bit :-) */
1410 }
a3621e74
YO
1411 } else {
1412 SV** svpp;
55eed653
NC
1413 if ( !widecharmap )
1414 widecharmap = newHV();
a3621e74 1415
55eed653 1416 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
1417
1418 if ( !svpp )
e4584336 1419 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
a3621e74
YO
1420
1421 if ( !SvTRUE( *svpp ) ) {
1422 sv_setiv( *svpp, ++trie->uniquecharcount );
3dab1dad 1423 TRIE_STORE_REVCHAR;
a3621e74
YO
1424 }
1425 }
1426 }
3dab1dad
YO
1427 if( cur == first ) {
1428 trie->minlen=chars;
1429 trie->maxlen=chars;
1430 } else if (chars < trie->minlen) {
1431 trie->minlen=chars;
1432 } else if (chars > trie->maxlen) {
1433 trie->maxlen=chars;
1434 }
1435
a3621e74
YO
1436 } /* end first pass */
1437 DEBUG_TRIE_COMPILE_r(
3dab1dad
YO
1438 PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
1439 (int)depth * 2 + 2,"",
55eed653 1440 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
1441 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
1442 (int)trie->minlen, (int)trie->maxlen )
a3621e74 1443 );
c944940b 1444 trie->wordlen = (U32 *) PerlMemShared_calloc( word_count, sizeof(U32) );
a3621e74
YO
1445
1446 /*
1447 We now know what we are dealing with in terms of unique chars and
1448 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
1449 representation using a flat table will take. If it's over a reasonable
1450 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
1451 conservative but potentially much slower representation using an array
1452 of lists.
1453
1454 At the end we convert both representations into the same compressed
1455 form that will be used in regexec.c for matching with. The latter
1456 is a form that cannot be used to construct with but has memory
1457 properties similar to the list form and access properties similar
1458 to the table form making it both suitable for fast searches and
1459 small enough that its feasable to store for the duration of a program.
1460
1461 See the comment in the code where the compressed table is produced
1462 inplace from the flat tabe representation for an explanation of how
1463 the compression works.
1464
1465 */
1466
1467
3dab1dad 1468 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) {
a3621e74
YO
1469 /*
1470 Second Pass -- Array Of Lists Representation
1471
1472 Each state will be represented by a list of charid:state records
1473 (reg_trie_trans_le) the first such element holds the CUR and LEN
1474 points of the allocated array. (See defines above).
1475
1476 We build the initial structure using the lists, and then convert
1477 it into the compressed table form which allows faster lookups
1478 (but cant be modified once converted).
a3621e74
YO
1479 */
1480
a3621e74
YO
1481 STRLEN transcount = 1;
1482
1e2e3d02
YO
1483 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1484 "%*sCompiling trie using list compiler\n",
1485 (int)depth * 2 + 2, ""));
446bd890 1486
c944940b
JH
1487 trie->states = (reg_trie_state *)
1488 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1489 sizeof(reg_trie_state) );
a3621e74
YO
1490 TRIE_LIST_NEW(1);
1491 next_alloc = 2;
1492
1493 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1494
c445ea15
AL
1495 regnode * const noper = NEXTOPER( cur );
1496 U8 *uc = (U8*)STRING( noper );
1497 const U8 * const e = uc + STR_LEN( noper );
1498 U32 state = 1; /* required init */
1499 U16 charid = 0; /* sanity init */
1500 U8 *scan = (U8*)NULL; /* sanity init */
1501 STRLEN foldlen = 0; /* required init */
07be1b83 1502 U32 wordlen = 0; /* required init */
c445ea15
AL
1503 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1504
3dab1dad 1505 if (OP(noper) != NOTHING) {
786e8c11 1506 for ( ; uc < e ; uc += len ) {
c445ea15 1507
786e8c11 1508 TRIE_READ_CHAR;
c445ea15 1509
786e8c11
YO
1510 if ( uvc < 256 ) {
1511 charid = trie->charmap[ uvc ];
c445ea15 1512 } else {
55eed653 1513 SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11
YO
1514 if ( !svpp ) {
1515 charid = 0;
1516 } else {
1517 charid=(U16)SvIV( *svpp );
1518 }
c445ea15 1519 }
786e8c11
YO
1520 /* charid is now 0 if we dont know the char read, or nonzero if we do */
1521 if ( charid ) {
a3621e74 1522
786e8c11
YO
1523 U16 check;
1524 U32 newstate = 0;
a3621e74 1525
786e8c11
YO
1526 charid--;
1527 if ( !trie->states[ state ].trans.list ) {
1528 TRIE_LIST_NEW( state );
c445ea15 1529 }
786e8c11
YO
1530 for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) {
1531 if ( TRIE_LIST_ITEM( state, check ).forid == charid ) {
1532 newstate = TRIE_LIST_ITEM( state, check ).newstate;
1533 break;
1534 }
1535 }
1536 if ( ! newstate ) {
1537 newstate = next_alloc++;
1538 TRIE_LIST_PUSH( state, charid, newstate );
1539 transcount++;
1540 }
1541 state = newstate;
1542 } else {
1543 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
c445ea15 1544 }
a28509cc 1545 }
c445ea15 1546 }
3dab1dad 1547 TRIE_HANDLE_WORD(state);
a3621e74
YO
1548
1549 } /* end second pass */
1550
1e2e3d02
YO
1551 /* next alloc is the NEXT state to be allocated */
1552 trie->statecount = next_alloc;
c944940b
JH
1553 trie->states = (reg_trie_state *)
1554 PerlMemShared_realloc( trie->states,
1555 next_alloc
1556 * sizeof(reg_trie_state) );
a3621e74 1557
3dab1dad 1558 /* and now dump it out before we compress it */
2b8b4781
NC
1559 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
1560 revcharmap, next_alloc,
1561 depth+1)
1e2e3d02 1562 );
a3621e74 1563
c944940b
JH
1564 trie->trans = (reg_trie_trans *)
1565 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
1566 {
1567 U32 state;
a3621e74
YO
1568 U32 tp = 0;
1569 U32 zp = 0;
1570
1571
1572 for( state=1 ; state < next_alloc ; state ++ ) {
1573 U32 base=0;
1574
1575 /*
1576 DEBUG_TRIE_COMPILE_MORE_r(
1577 PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp)
1578 );
1579 */
1580
1581 if (trie->states[state].trans.list) {
1582 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
1583 U16 maxid=minid;
a28509cc 1584 U16 idx;
a3621e74
YO
1585
1586 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
1587 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
1588 if ( forid < minid ) {
1589 minid=forid;
1590 } else if ( forid > maxid ) {
1591 maxid=forid;
1592 }
a3621e74
YO
1593 }
1594 if ( transcount < tp + maxid - minid + 1) {
1595 transcount *= 2;
c944940b
JH
1596 trie->trans = (reg_trie_trans *)
1597 PerlMemShared_realloc( trie->trans,
446bd890
NC
1598 transcount
1599 * sizeof(reg_trie_trans) );
a3621e74
YO
1600 Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
1601 }
1602 base = trie->uniquecharcount + tp - minid;
1603 if ( maxid == minid ) {
1604 U32 set = 0;
1605 for ( ; zp < tp ; zp++ ) {
1606 if ( ! trie->trans[ zp ].next ) {
1607 base = trie->uniquecharcount + zp - minid;
1608 trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1609 trie->trans[ zp ].check = state;
1610 set = 1;
1611 break;
1612 }
1613 }
1614 if ( !set ) {
1615 trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate;
1616 trie->trans[ tp ].check = state;
1617 tp++;
1618 zp = tp;
1619 }
1620 } else {
1621 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15 1622 const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid;
a3621e74
YO
1623 trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate;
1624 trie->trans[ tid ].check = state;
1625 }
1626 tp += ( maxid - minid + 1 );
1627 }
1628 Safefree(trie->states[ state ].trans.list);
1629 }
1630 /*
1631 DEBUG_TRIE_COMPILE_MORE_r(
1632 PerlIO_printf( Perl_debug_log, " base: %d\n",base);
1633 );
1634 */
1635 trie->states[ state ].trans.base=base;
1636 }
cc601c31 1637 trie->lasttrans = tp + 1;
a3621e74
YO
1638 }
1639 } else {
1640 /*
1641 Second Pass -- Flat Table Representation.
1642
1643 we dont use the 0 slot of either trans[] or states[] so we add 1 to each.
1644 We know that we will need Charcount+1 trans at most to store the data
1645 (one row per char at worst case) So we preallocate both structures
1646 assuming worst case.
1647
1648 We then construct the trie using only the .next slots of the entry
1649 structs.
1650
1651 We use the .check field of the first entry of the node temporarily to
1652 make compression both faster and easier by keeping track of how many non
1653 zero fields are in the node.
1654
1655 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
1656 transition.
1657
1658 There are two terms at use here: state as a TRIE_NODEIDX() which is a
1659 number representing the first entry of the node, and state as a
1660 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and
1661 TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there
1662 are 2 entrys per node. eg:
1663
1664 A B A B
1665 1. 2 4 1. 3 7
1666 2. 0 3 3. 0 5
1667 3. 0 0 5. 0 0
1668 4. 0 0 7. 0 0
1669
1670 The table is internally in the right hand, idx form. However as we also
1671 have to deal with the states array which is indexed by nodenum we have to
1672 use TRIE_NODENUM() to convert.
1673
1674 */
1e2e3d02
YO
1675 DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
1676 "%*sCompiling trie using table compiler\n",
1677 (int)depth * 2 + 2, ""));
3dab1dad 1678
c944940b
JH
1679 trie->trans = (reg_trie_trans *)
1680 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
1681 * trie->uniquecharcount + 1,
1682 sizeof(reg_trie_trans) );
1683 trie->states = (reg_trie_state *)
1684 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
1685 sizeof(reg_trie_state) );
a3621e74
YO
1686 next_alloc = trie->uniquecharcount + 1;
1687
3dab1dad 1688
a3621e74
YO
1689 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
1690
c445ea15 1691 regnode * const noper = NEXTOPER( cur );
a28509cc
AL
1692 const U8 *uc = (U8*)STRING( noper );
1693 const U8 * const e = uc + STR_LEN( noper );
a3621e74
YO
1694
1695 U32 state = 1; /* required init */
1696
1697 U16 charid = 0; /* sanity init */
1698 U32 accept_state = 0; /* sanity init */
1699 U8 *scan = (U8*)NULL; /* sanity init */
1700
1701 STRLEN foldlen = 0; /* required init */
07be1b83 1702 U32 wordlen = 0; /* required init */
a3621e74
YO
1703 U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
1704
3dab1dad 1705 if ( OP(noper) != NOTHING ) {
786e8c11 1706 for ( ; uc < e ; uc += len ) {
a3621e74 1707
786e8c11 1708 TRIE_READ_CHAR;
a3621e74 1709
786e8c11
YO
1710 if ( uvc < 256 ) {
1711 charid = trie->charmap[ uvc ];
1712 } else {
55eed653 1713 SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
786e8c11 1714 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 1715 }
786e8c11
YO
1716 if ( charid ) {
1717 charid--;
1718 if ( !trie->trans[ state + charid ].next ) {
1719 trie->trans[ state + charid ].next = next_alloc;
1720 trie->trans[ state ].check++;
1721 next_alloc += trie->uniquecharcount;
1722 }
1723 state = trie->trans[ state + charid ].next;
1724 } else {
1725 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
1726 }
1727 /* charid is now 0 if we dont know the char read, or nonzero if we do */
a3621e74 1728 }
a3621e74 1729 }
3dab1dad
YO
1730 accept_state = TRIE_NODENUM( state );
1731 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
1732
1733 } /* end second pass */
1734
3dab1dad 1735 /* and now dump it out before we compress it */
2b8b4781
NC
1736 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
1737 revcharmap,
1738 next_alloc, depth+1));
a3621e74 1739
a3621e74
YO
1740 {
1741 /*
1742 * Inplace compress the table.*
1743
1744 For sparse data sets the table constructed by the trie algorithm will
1745 be mostly 0/FAIL transitions or to put it another way mostly empty.
1746 (Note that leaf nodes will not contain any transitions.)
1747
1748 This algorithm compresses the tables by eliminating most such
1749 transitions, at the cost of a modest bit of extra work during lookup:
1750
1751 - Each states[] entry contains a .base field which indicates the
1752 index in the state[] array wheres its transition data is stored.
1753
1754 - If .base is 0 there are no valid transitions from that node.
1755
1756 - If .base is nonzero then charid is added to it to find an entry in
1757 the trans array.
1758
1759 -If trans[states[state].base+charid].check!=state then the
1760 transition is taken to be a 0/Fail transition. Thus if there are fail
1761 transitions at the front of the node then the .base offset will point
1762 somewhere inside the previous nodes data (or maybe even into a node
1763 even earlier), but the .check field determines if the transition is
1764 valid.
1765
786e8c11 1766 XXX - wrong maybe?
a3621e74
YO
1767 The following process inplace converts the table to the compressed
1768 table: We first do not compress the root node 1,and mark its all its
1769 .check pointers as 1 and set its .base pointer as 1 as well. This
1770 allows to do a DFA construction from the compressed table later, and
1771 ensures that any .base pointers we calculate later are greater than
1772 0.
1773
1774 - We set 'pos' to indicate the first entry of the second node.
1775
1776 - We then iterate over the columns of the node, finding the first and
1777 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
1778 and set the .check pointers accordingly, and advance pos
1779 appropriately and repreat for the next node. Note that when we copy
1780 the next pointers we have to convert them from the original
1781 NODEIDX form to NODENUM form as the former is not valid post
1782 compression.
1783
1784 - If a node has no transitions used we mark its base as 0 and do not
1785 advance the pos pointer.
1786
1787 - If a node only has one transition we use a second pointer into the
1788 structure to fill in allocated fail transitions from other states.
1789 This pointer is independent of the main pointer and scans forward
1790 looking for null transitions that are allocated to a state. When it
1791 finds one it writes the single transition into the "hole". If the
786e8c11 1792 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
1793
1794 - Once compressed we can Renew/realloc the structures to release the
1795 excess space.
1796
1797 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
1798 specifically Fig 3.47 and the associated pseudocode.
1799
1800 demq
1801 */
a3b680e6 1802 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 1803 U32 state, charid;
a3621e74 1804 U32 pos = 0, zp=0;
1e2e3d02 1805 trie->statecount = laststate;
a3621e74
YO
1806
1807 for ( state = 1 ; state < laststate ; state++ ) {
1808 U8 flag = 0;
a28509cc
AL
1809 const U32 stateidx = TRIE_NODEIDX( state );
1810 const U32 o_used = trie->trans[ stateidx ].check;
1811 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
1812 trie->trans[ stateidx ].check = 0;
1813
1814 for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) {
1815 if ( flag || trie->trans[ stateidx + charid ].next ) {
1816 if ( trie->trans[ stateidx + charid ].next ) {
1817 if (o_used == 1) {
1818 for ( ; zp < pos ; zp++ ) {
1819 if ( ! trie->trans[ zp ].next ) {
1820 break;
1821 }
1822 }
1823 trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ;
1824 trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1825 trie->trans[ zp ].check = state;
1826 if ( ++zp > pos ) pos = zp;
1827 break;
1828 }
1829 used--;
1830 }
1831 if ( !flag ) {
1832 flag = 1;
1833 trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ;
1834 }
1835 trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next );
1836 trie->trans[ pos ].check = state;
1837 pos++;
1838 }
1839 }
1840 }
cc601c31 1841 trie->lasttrans = pos + 1;
c944940b
JH
1842 trie->states = (reg_trie_state *)
1843 PerlMemShared_realloc( trie->states, laststate
1844 * sizeof(reg_trie_state) );
a3621e74 1845 DEBUG_TRIE_COMPILE_MORE_r(
e4584336 1846 PerlIO_printf( Perl_debug_log,
3dab1dad
YO
1847 "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
1848 (int)depth * 2 + 2,"",
1849 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ),
5d7488b2
AL
1850 (IV)next_alloc,
1851 (IV)pos,
a3621e74
YO
1852 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
1853 );
1854
1855 } /* end table compress */
1856 }
1e2e3d02
YO
1857 DEBUG_TRIE_COMPILE_MORE_r(
1858 PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
1859 (int)depth * 2 + 2, "",
1860 (UV)trie->statecount,
1861 (UV)trie->lasttrans)
1862 );
cc601c31 1863 /* resize the trans array to remove unused space */
c944940b
JH
1864 trie->trans = (reg_trie_trans *)
1865 PerlMemShared_realloc( trie->trans, trie->lasttrans
1866 * sizeof(reg_trie_trans) );
a3621e74 1867
3dab1dad 1868 /* and now dump out the compressed format */
2b8b4781 1869 DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
07be1b83 1870
3dab1dad 1871 { /* Modify the program and insert the new TRIE node*/
3dab1dad
YO
1872 U8 nodetype =(U8)(flags & 0xFF);
1873 char *str=NULL;
786e8c11 1874
07be1b83 1875#ifdef DEBUGGING
e62cc96a 1876 regnode *optimize = NULL;
7122b237
YO
1877#ifdef RE_TRACK_PATTERN_OFFSETS
1878
b57a0404
JH
1879 U32 mjd_offset = 0;
1880 U32 mjd_nodelen = 0;
7122b237
YO
1881#endif /* RE_TRACK_PATTERN_OFFSETS */
1882#endif /* DEBUGGING */
a3621e74 1883 /*
3dab1dad
YO
1884 This means we convert either the first branch or the first Exact,
1885 depending on whether the thing following (in 'last') is a branch
1886 or not and whther first is the startbranch (ie is it a sub part of
1887 the alternation or is it the whole thing.)
1888 Assuming its a sub part we conver the EXACT otherwise we convert
1889 the whole branch sequence, including the first.
a3621e74 1890 */
3dab1dad 1891 /* Find the node we are going to overwrite */
7f69552c 1892 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 1893 /* branch sub-chain */
3dab1dad 1894 NEXT_OFF( first ) = (U16)(last - first);
7122b237 1895#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
1896 DEBUG_r({
1897 mjd_offset= Node_Offset((convert));
1898 mjd_nodelen= Node_Length((convert));
1899 });
7122b237 1900#endif
7f69552c 1901 /* whole branch chain */
7122b237
YO
1902 }
1903#ifdef RE_TRACK_PATTERN_OFFSETS
1904 else {
7f69552c
YO
1905 DEBUG_r({
1906 const regnode *nop = NEXTOPER( convert );
1907 mjd_offset= Node_Offset((nop));
1908 mjd_nodelen= Node_Length((nop));
1909 });
07be1b83
YO
1910 }
1911 DEBUG_OPTIMISE_r(
1912 PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
1913 (int)depth * 2 + 2, "",
786e8c11 1914 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 1915 );
7122b237 1916#endif
3dab1dad
YO
1917 /* But first we check to see if there is a common prefix we can
1918 split out as an EXACT and put in front of the TRIE node. */
1919 trie->startstate= 1;
55eed653 1920 if ( trie->bitmap && !widecharmap && !trie->jump ) {
3dab1dad 1921 U32 state;
1e2e3d02 1922 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74 1923 U32 ofs = 0;
8e11feef
RGS
1924 I32 idx = -1;
1925 U32 count = 0;
1926 const U32 base = trie->states[ state ].trans.base;
a3621e74 1927
3dab1dad 1928 if ( trie->states[state].wordnum )
8e11feef 1929 count = 1;
a3621e74 1930
8e11feef 1931 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
cc601c31
YO
1932 if ( ( base + ofs >= trie->uniquecharcount ) &&
1933 ( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
a3621e74
YO
1934 trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
1935 {
3dab1dad 1936 if ( ++count > 1 ) {
2b8b4781 1937 SV **tmp = av_fetch( revcharmap, ofs, 0);
07be1b83 1938 const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 1939 if ( state == 1 ) break;
3dab1dad
YO
1940 if ( count == 2 ) {
1941 Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char);
1942 DEBUG_OPTIMISE_r(
8e11feef
RGS
1943 PerlIO_printf(Perl_debug_log,
1944 "%*sNew Start State=%"UVuf" Class: [",
1945 (int)depth * 2 + 2, "",
786e8c11 1946 (UV)state));
be8e71aa 1947 if (idx >= 0) {
2b8b4781 1948 SV ** const tmp = av_fetch( revcharmap, idx, 0);
be8e71aa 1949 const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
8e11feef 1950
3dab1dad 1951 TRIE_BITMAP_SET(trie,*ch);
8e11feef
RGS
1952 if ( folder )
1953 TRIE_BITMAP_SET(trie, folder[ *ch ]);
3dab1dad 1954 DEBUG_OPTIMISE_r(
07be1b83 1955 PerlIO_printf(Perl_debug_log, (char*)ch)
3dab1dad 1956 );
8e11feef
RGS
1957 }
1958 }
1959 TRIE_BITMAP_SET(trie,*ch);
1960 if ( folder )
1961 TRIE_BITMAP_SET(trie,folder[ *ch ]);
1962 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch));
1963 }
1964 idx = ofs;
1965 }
3dab1dad
YO
1966 }
1967 if ( count == 1 ) {
2b8b4781 1968 SV **tmp = av_fetch( revcharmap, idx, 0);
de734bd5
A
1969 char *ch = SvPV_nolen( *tmp );
1970 DEBUG_OPTIMISE_r({
1971 SV *sv=sv_newmortal();
8e11feef
RGS
1972 PerlIO_printf( Perl_debug_log,
1973 "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
1974 (int)depth * 2 + 2, "",
de734bd5
A
1975 (UV)state, (UV)idx,
1976 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
1977 PL_colors[0], PL_colors[1],
1978 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
1979 PERL_PV_ESCAPE_FIRSTCHAR
1980 )
1981 );
1982 });
3dab1dad
YO
1983 if ( state==1 ) {
1984 OP( convert ) = nodetype;
1985 str=STRING(convert);
1986 STR_LEN(convert)=0;
1987 }
de734bd5
A
1988 while (*ch) {
1989 *str++ = *ch++;
1990 STR_LEN(convert)++;
1991 }
1992
8e11feef 1993 } else {
f9049ba1 1994#ifdef DEBUGGING
8e11feef
RGS
1995 if (state>1)
1996 DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
f9049ba1 1997#endif
8e11feef
RGS
1998 break;
1999 }
2000 }
3dab1dad 2001 if (str) {
8e11feef 2002 regnode *n = convert+NODE_SZ_STR(convert);
07be1b83 2003 NEXT_OFF(convert) = NODE_SZ_STR(convert);
8e11feef 2004 trie->startstate = state;
07be1b83
YO
2005 trie->minlen -= (state - 1);
2006 trie->maxlen -= (state - 1);
2007 DEBUG_r({
2008 regnode *fix = convert;
de734bd5 2009 U32 word = trie->wordcount;
07be1b83
YO
2010 mjd_nodelen++;
2011 Set_Node_Offset_Length(convert, mjd_offset, state - 1);
2012 while( ++fix < n ) {
2013 Set_Node_Offset_Length(fix, 0, 0);
2014 }
de734bd5 2015 while (word--) {
2b8b4781 2016 SV ** const tmp = av_fetch( trie_words, word, 0 );
de734bd5
A
2017 if (tmp) {
2018 if ( STR_LEN(convert) <= SvCUR(*tmp) )
2019 sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
2020 else
2021 sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
2022 }
2023 }
07be1b83 2024 });
8e11feef
RGS
2025 if (trie->maxlen) {
2026 convert = n;
2027 } else {
3dab1dad 2028 NEXT_OFF(convert) = (U16)(tail - convert);
a5ca303d 2029 DEBUG_r(optimize= n);
3dab1dad
YO
2030 }
2031 }
2032 }
a5ca303d
YO
2033 if (!jumper)
2034 jumper = last;
3dab1dad 2035 if ( trie->maxlen ) {
8e11feef
RGS
2036 NEXT_OFF( convert ) = (U16)(tail - convert);
2037 ARG_SET( convert, data_slot );
786e8c11
YO
2038 /* Store the offset to the first unabsorbed branch in
2039 jump[0], which is otherwise unused by the jump logic.
2040 We use this when dumping a trie and during optimisation. */
2041 if (trie->jump)
7f69552c 2042 trie->jump[0] = (U16)(nextbranch - convert);
a5ca303d 2043
786e8c11
YO
2044 /* XXXX */
2045 if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
1de06328 2046 ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
786e8c11
YO
2047 {
2048 OP( convert ) = TRIEC;
2049 Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
446bd890 2050 PerlMemShared_free(trie->bitmap);
786e8c11
YO
2051 trie->bitmap= NULL;
2052 } else
2053 OP( convert ) = TRIE;
a3621e74 2054
3dab1dad
YO
2055 /* store the type in the flags */
2056 convert->flags = nodetype;
a5ca303d
YO
2057 DEBUG_r({
2058 optimize = convert
2059 + NODE_STEP_REGNODE
2060 + regarglen[ OP( convert ) ];
2061 });
2062 /* XXX We really should free up the resource in trie now,
2063 as we won't use them - (which resources?) dmq */
3dab1dad 2064 }
a3621e74 2065 /* needed for dumping*/
e62cc96a 2066 DEBUG_r(if (optimize) {
07be1b83 2067 regnode *opt = convert;
bcdf7404 2068
e62cc96a 2069 while ( ++opt < optimize) {
07be1b83
YO
2070 Set_Node_Offset_Length(opt,0,0);
2071 }
786e8c11
YO
2072 /*
2073 Try to clean up some of the debris left after the
2074 optimisation.
a3621e74 2075 */
786e8c11 2076 while( optimize < jumper ) {
07be1b83 2077 mjd_nodelen += Node_Length((optimize));
a3621e74 2078 OP( optimize ) = OPTIMIZED;
07be1b83 2079 Set_Node_Offset_Length(optimize,0,0);
a3621e74
YO
2080 optimize++;
2081 }
07be1b83 2082 Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
a3621e74
YO
2083 });
2084 } /* end node insert */
55eed653 2085 RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
2b8b4781
NC
2086#ifdef DEBUGGING
2087 RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
2088 RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
2089#else
2090 SvREFCNT_dec(revcharmap);
07be1b83 2091#endif
786e8c11
YO
2092 return trie->jump
2093 ? MADE_JUMP_TRIE
2094 : trie->startstate>1
2095 ? MADE_EXACT_TRIE
2096 : MADE_TRIE;
2097}
2098
2099STATIC void
2100S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
2101{
2102/* The Trie is constructed and compressed now so we can build a fail array now if its needed
2103
2104 This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
2105 "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
2106 ISBN 0-201-10088-6
2107
2108 We find the fail state for each state in the trie, this state is the longest proper
2109 suffix of the current states 'word' that is also a proper prefix of another word in our
2110 trie. State 1 represents the word '' and is the thus the default fail state. This allows
2111 the DFA not to have to restart after its tried and failed a word at a given point, it
2112 simply continues as though it had been matching the other word in the first place.
2113 Consider
2114 'abcdgu'=~/abcdefg|cdgu/
2115 When we get to 'd' we are still matching the first word, we would encounter 'g' which would
2116 fail, which would bring use to the state representing 'd' in the second word where we would
2117 try 'g' and succeed, prodceding to match 'cdgu'.
2118 */
2119 /* add a fail transition */
3251b653
NC
2120 const U32 trie_offset = ARG(source);
2121 reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
786e8c11
YO
2122 U32 *q;
2123 const U32 ucharcount = trie->uniquecharcount;
1e2e3d02 2124 const U32 numstates = trie->statecount;
786e8c11
YO
2125 const U32 ubound = trie->lasttrans + ucharcount;
2126 U32 q_read = 0;
2127 U32 q_write = 0;
2128 U32 charid;
2129 U32 base = trie->states[ 1 ].trans.base;
2130 U32 *fail;
2131 reg_ac_data *aho;
2132 const U32 data_slot = add_data( pRExC_state, 1, "T" );
2133 GET_RE_DEBUG_FLAGS_DECL;
2134#ifndef DEBUGGING
2135 PERL_UNUSED_ARG(depth);
2136#endif
2137
2138
2139 ARG_SET( stclass, data_slot );
c944940b 2140 aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
f8fc2ecf 2141 RExC_rxi->data->data[ data_slot ] = (void*)aho;
3251b653 2142 aho->trie=trie_offset;
446bd890
NC
2143 aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
2144 Copy( trie->states, aho->states, numstates, reg_trie_state );
786e8c11 2145 Newxz( q, numstates, U32);
c944940b 2146 aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) );
786e8c11
YO
2147 aho->refcount = 1;
2148 fail = aho->fail;
2149 /* initialize fail[0..1] to be 1 so that we always have
2150 a valid final fail state */
2151 fail[ 0 ] = fail[ 1 ] = 1;
2152
2153 for ( charid = 0; charid < ucharcount ; charid++ ) {
2154 const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
2155 if ( newstate ) {
2156 q[ q_write ] = newstate;
2157 /* set to point at the root */
2158 fail[ q[ q_write++ ] ]=1;
2159 }
2160 }
2161 while ( q_read < q_write) {
2162 const U32 cur = q[ q_read++ % numstates ];
2163 base = trie->states[ cur ].trans.base;
2164
2165 for ( charid = 0 ; charid < ucharcount ; charid++ ) {
2166 const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
2167 if (ch_state) {
2168 U32 fail_state = cur;
2169 U32 fail_base;
2170 do {
2171 fail_state = fail[ fail_state ];
2172 fail_base = aho->states[ fail_state ].trans.base;
2173 } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
2174
2175 fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
2176 fail[ ch_state ] = fail_state;
2177 if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
2178 {
2179 aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
2180 }
2181 q[ q_write++ % numstates] = ch_state;
2182 }
2183 }
2184 }
2185 /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
2186 when we fail in state 1, this allows us to use the
2187 charclass scan to find a valid start char. This is based on the principle
2188 that theres a good chance the string being searched contains lots of stuff
2189 that cant be a start char.
2190 */
2191 fail[ 0 ] = fail[ 1 ] = 0;
2192 DEBUG_TRIE_COMPILE_r({
6d99fb9b
JH
2193 PerlIO_printf(Perl_debug_log,
2194 "%*sStclass Failtable (%"UVuf" states): 0",
2195 (int)(depth * 2), "", (UV)numstates
1e2e3d02 2196 );
786e8c11
YO
2197 for( q_read=1; q_read<numstates; q_read++ ) {
2198 PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
2199 }
2200 PerlIO_printf(Perl_debug_log, "\n");
2201 });
2202 Safefree(q);
2203 /*RExC_seen |= REG_SEEN_TRIEDFA;*/
a3621e74
YO
2204}
2205
786e8c11 2206
a3621e74 2207/*
5d1c421c
JH
2208 * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2.
2209 * These need to be revisited when a newer toolchain becomes available.
2210 */
2211#if defined(__sparc64__) && defined(__GNUC__)
2212# if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96)
2213# undef SPARC64_GCC_WORKAROUND
2214# define SPARC64_GCC_WORKAROUND 1
2215# endif
2216#endif
2217
07be1b83 2218#define DEBUG_PEEP(str,scan,depth) \
b515a41d 2219 DEBUG_OPTIMISE_r({if (scan){ \
07be1b83
YO
2220 SV * const mysv=sv_newmortal(); \
2221 regnode *Next = regnext(scan); \
2222 regprop(RExC_rx, mysv, scan); \
7f69552c 2223 PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
07be1b83
YO
2224 (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
2225 Next ? (REG_NODE_NUM(Next)) : 0 ); \
b515a41d 2226 }});
07be1b83 2227
1de06328
YO
2228
2229
2230
2231
07be1b83
YO
2232#define JOIN_EXACT(scan,min,flags) \
2233 if (PL_regkind[OP(scan)] == EXACT) \
2234 join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1)
2235
be8e71aa 2236STATIC U32
07be1b83
YO
2237S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) {
2238 /* Merge several consecutive EXACTish nodes into one. */
2239 regnode *n = regnext(scan);
2240 U32 stringok = 1;
2241 regnode *next = scan + NODE_SZ_STR(scan);
2242 U32 merged = 0;
2243 U32 stopnow = 0;
2244#ifdef DEBUGGING
2245 regnode *stop = scan;
72f13be8 2246 GET_RE_DEBUG_FLAGS_DECL;
f9049ba1 2247#else
d47053eb
RGS
2248 PERL_UNUSED_ARG(depth);
2249#endif
2250#ifndef EXPERIMENTAL_INPLACESCAN
f9049ba1
SP
2251 PERL_UNUSED_ARG(flags);
2252 PERL_UNUSED_ARG(val);
07be1b83 2253#endif
07be1b83
YO
2254 DEBUG_PEEP("join",scan,depth);
2255
2256 /* Skip NOTHING, merge EXACT*. */
2257 while (n &&
2258 ( PL_regkind[OP(n)] == NOTHING ||
2259 (stringok && (OP(n) == OP(scan))))
2260 && NEXT_OFF(n)
2261 && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) {
2262
2263 if (OP(n) == TAIL || n > next)
2264 stringok = 0;
2265 if (PL_regkind[OP(n)] == NOTHING) {
07be1b83
YO
2266 DEBUG_PEEP("skip:",n,depth);
2267 NEXT_OFF(scan) += NEXT_OFF(n);
2268 next = n + NODE_STEP_REGNODE;
2269#ifdef DEBUGGING
2270 if (stringok)
2271 stop = n;
2272#endif
2273 n = regnext(n);
2274 }
2275 else if (stringok) {
786e8c11 2276 const unsigned int oldl = STR_LEN(scan);
07be1b83
YO
2277 regnode * const nnext = regnext(n);
2278
2279 DEBUG_PEEP("merg",n,depth);
2280
2281 merged++;
2282 if (oldl + STR_LEN(n) > U8_MAX)
2283 break;
2284 NEXT_OFF(scan) += NEXT_OFF(n);
2285 STR_LEN(scan) += STR_LEN(n);
2286 next = n + NODE_SZ_STR(n);
2287 /* Now we can overwrite *n : */
2288 Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char);
2289#ifdef DEBUGGING
2290 stop = next - 1;
2291#endif
2292 n = nnext;
2293 if (stopnow) break;
2294 }
2295
d47053eb
RGS
2296#ifdef EXPERIMENTAL_INPLACESCAN
2297 if (flags && !NEXT_OFF(n)) {
2298 DEBUG_PEEP("atch", val, depth);
2299 if (reg_off_by_arg[OP(n)]) {
2300 ARG_SET(n, val - n);
2301 }
2302 else {
2303 NEXT_OFF(n) = val - n;
2304 }
2305 stopnow = 1;
2306 }
07be1b83
YO
2307#endif
2308 }
2309
2310 if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) {
2311 /*
2312 Two problematic code points in Unicode casefolding of EXACT nodes:
2313
2314 U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS
2315 U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS
2316
2317 which casefold to
2318
2319 Unicode UTF-8
2320
2321 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81
2322 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81
2323
2324 This means that in case-insensitive matching (or "loose matching",
2325 as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte
2326 length of the above casefolded versions) can match a target string
2327 of length two (the byte length of UTF-8 encoded U+0390 or U+03B0).
2328 This would rather mess up the minimum length computation.
2329
2330 What we'll do is to look for the tail four bytes, and then peek
2331 at the preceding two bytes to see whether we need to decrease
2332 the minimum length by four (six minus two).
2333
2334 Thanks to the design of UTF-8, there cannot be false matches:
2335 A sequence of valid UTF-8 bytes cannot be a subsequence of
2336 another valid sequence of UTF-8 bytes.
2337
2338 */
2339 char * const s0 = STRING(scan), *s, *t;
2340 char * const s1 = s0 + STR_LEN(scan) - 1;
2341 char * const s2 = s1 - 4;
e294cc5d
JH
2342#ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
2343 const char t0[] = "\xaf\x49\xaf\x42";
2344#else
07be1b83 2345 const char t0[] = "\xcc\x88\xcc\x81";
e294cc5d 2346#endif
07be1b83
YO
2347 const char * const t1 = t0 + 3;
2348
2349 for (s = s0 + 2;
2350 s < s2 && (t = ninstr(s, s1, t0, t1));
2351 s = t + 4) {
e294cc5d
JH
2352#ifdef EBCDIC
2353 if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
2354 ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
2355#else
07be1b83
YO
2356 if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
2357 ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
e294cc5d 2358#endif
07be1b83
YO
2359 *min -= 4;
2360 }
2361 }
2362
2363#ifdef DEBUGGING
2364 /* Allow dumping */
2365 n = scan + NODE_SZ_STR(scan);
2366 while (n <= stop) {
2367 if (PL_regkind[OP(n)] != NOTHING || OP(n) == NOTHING) {
2368 OP(n) = OPTIMIZED;
2369 NEXT_OFF(n) = 0;
2370 }
2371 n++;
2372 }
2373#endif
2374 DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)});
2375 return stopnow;
2376}
2377
653099ff
GS
2378/* REx optimizer. Converts nodes into quickier variants "in place".
2379 Finds fixed substrings. */
2380
a0288114 2381/* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
c277df42
IZ
2382 to the position after last scanned or to NULL. */
2383
40d049e4
YO
2384#define INIT_AND_WITHP \
2385 assert(!and_withp); \
2386 Newx(and_withp,1,struct regnode_charclass_class); \
2387 SAVEFREEPV(and_withp)
07be1b83 2388
b515a41d
YO
2389/* this is a chain of data about sub patterns we are processing that
2390 need to be handled seperately/specially in study_chunk. Its so
2391 we can simulate recursion without losing state. */
2392struct scan_frame;
2393typedef struct scan_frame {
2394 regnode *last; /* last node to process in this frame */
2395 regnode *next; /* next node to process when last is reached */
2396 struct scan_frame *prev; /*previous frame*/
2397 I32 stop; /* what stopparen do we use */
2398} scan_frame;
2399
304ee84b
YO
2400
2401#define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf)
2402
e1d1eefb
YO
2403#define CASE_SYNST_FNC(nAmE) \
2404case nAmE: \
2405 if (flags & SCF_DO_STCLASS_AND) { \
2406 for (value = 0; value < 256; value++) \
2407 if (!is_ ## nAmE ## _cp(value)) \
2408 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2409 } \
2410 else { \
2411 for (value = 0; value < 256; value++) \
2412 if (is_ ## nAmE ## _cp(value)) \
2413 ANYOF_BITMAP_SET(data->start_class, value); \
2414 } \
2415 break; \
2416case N ## nAmE: \
2417 if (flags & SCF_DO_STCLASS_AND) { \
2418 for (value = 0; value < 256; value++) \
2419 if (is_ ## nAmE ## _cp(value)) \
2420 ANYOF_BITMAP_CLEAR(data->start_class, value); \
2421 } \
2422 else { \
2423 for (value = 0; value < 256; value++) \
2424 if (!is_ ## nAmE ## _cp(value)) \
2425 ANYOF_BITMAP_SET(data->start_class, value); \
2426 } \
2427 break
2428
2429
2430
76e3520e 2431STATIC I32
40d049e4 2432S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
1de06328 2433 I32 *minlenp, I32 *deltap,
40d049e4
YO
2434 regnode *last,
2435 scan_data_t *data,
2436 I32 stopparen,
2437 U8* recursed,
2438 struct regnode_charclass_class *and_withp,
2439 U32 flags, U32 depth)
c277df42
IZ
2440 /* scanp: Start here (read-write). */
2441 /* deltap: Write maxlen-minlen here. */
2442 /* last: Stop before this one. */
40d049e4
YO
2443 /* data: string data about the pattern */
2444 /* stopparen: treat close N as END */
2445 /* recursed: which subroutines have we recursed into */
2446 /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
c277df42 2447{
97aff369 2448 dVAR;
c277df42
IZ
2449 I32 min = 0, pars = 0, code;
2450 regnode *scan = *scanp, *next;
2451 I32 delta = 0;
2452 int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF);
aca2d497 2453 int is_inf_internal = 0; /* The studied chunk is infinite */
c277df42
IZ
2454 I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0;
2455 scan_data_t data_fake;
a3621e74 2456 SV *re_trie_maxbuff = NULL;
786e8c11 2457 regnode *first_non_open = scan;
e2e6a0f1 2458 I32 stopmin = I32_MAX;
8aa23a47
YO
2459 scan_frame *frame = NULL;
2460
a3621e74 2461 GET_RE_DEBUG_FLAGS_DECL;
8aa23a47 2462
13a24bad 2463#ifdef DEBUGGING
40d049e4 2464 StructCopy(&zero_scan_data, &data_fake, scan_data_t);
13a24bad 2465#endif
40d049e4 2466
786e8c11 2467 if ( depth == 0 ) {
40d049e4 2468 while (first_non_open && OP(first_non_open) == OPEN)
786e8c11
YO
2469 first_non_open=regnext(first_non_open);
2470 }
2471
b81d288d 2472
8aa23a47
YO
2473 fake_study_recurse:
2474 while ( scan && OP(scan) != END && scan < last ){
2475 /* Peephole optimizer: */
304ee84b 2476 DEBUG_STUDYDATA("Peep:", data,depth);
8aa23a47
YO
2477 DEBUG_PEEP("Peep",scan,depth);
2478 JOIN_EXACT(scan,&min,0);
2479
2480 /* Follow the next-chain of the current node and optimize
2481 away all the NOTHINGs from it. */
2482 if (OP(scan) != CURLYX) {
2483 const int max = (reg_off_by_arg[OP(scan)]
2484 ? I32_MAX
2485 /* I32 may be smaller than U16 on CRAYs! */
2486 : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX));
2487 int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan));
2488 int noff;
2489 regnode *n = scan;
2490
2491 /* Skip NOTHING and LONGJMP. */
2492 while ((n = regnext(n))
2493 && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n)))
2494 || ((OP(n) == LONGJMP) && (noff = ARG(n))))
2495 && off + noff < max)
2496 off += noff;
2497 if (reg_off_by_arg[OP(scan)])
2498 ARG(scan) = off;
2499 else
2500 NEXT_OFF(scan) = off;
2501 }
a3621e74 2502
c277df42 2503
8aa23a47
YO
2504
2505 /* The principal pseudo-switch. Cannot be a switch, since we
2506 look into several different things. */
2507 if (OP(scan) == BRANCH || OP(scan) == BRANCHJ
2508 || OP(scan) == IFTHEN) {
2509 next = regnext(scan);
2510 code = OP(scan);
2511 /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */
2512
2513 if (OP(next) == code || code == IFTHEN) {
2514 /* NOTE - There is similar code to this block below for handling
2515 TRIE nodes on a re-study. If you change stuff here check there
2516 too. */
2517 I32 max1 = 0, min1 = I32_MAX, num = 0;
2518 struct regnode_charclass_class accum;
2519 regnode * const startbranch=scan;
2520
2521 if (flags & SCF_DO_SUBSTR)
304ee84b 2522 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
8aa23a47
YO
2523 if (flags & SCF_DO_STCLASS)
2524 cl_init_zero(pRExC_state, &accum);
2525
2526 while (OP(scan) == code) {
2527 I32 deltanext, minnext, f = 0, fake;
2528 struct regnode_charclass_class this_class;
2529
2530 num++;
2531 data_fake.flags = 0;
2532 if (data) {
2533 data_fake.whilem_c = data->whilem_c;
2534 data_fake.last_closep = data->last_closep;
2535 }
2536 else
2537 data_fake.last_closep = &fake;
58e23c8d
YO
2538
2539 data_fake.pos_delta = delta;
8aa23a47
YO
2540 next = regnext(scan);
2541 scan = NEXTOPER(scan);
2542 if (code != BRANCH)
c277df42 2543 scan = NEXTOPER(scan);
8aa23a47
YO
2544 if (flags & SCF_DO_STCLASS) {
2545 cl_init(pRExC_state, &this_class);
2546 data_fake.start_class = &this_class;
2547 f = SCF_DO_STCLASS_AND;
58e23c8d 2548 }
8aa23a47
YO
2549 if (flags & SCF_WHILEM_VISITED_POS)
2550 f |= SCF_WHILEM_VISITED_POS;
2551
2552 /* we suppose the run is continuous, last=next...*/
2553 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
2554 next, &data_fake,
2555 stopparen, recursed, NULL, f,depth+1);
2556 if (min1 > minnext)
2557 min1 = minnext;
2558 if (max1 < minnext + deltanext)
2559 max1 = minnext + deltanext;
2560 if (deltanext == I32_MAX)
2561 is_inf = is_inf_internal = 1;
2562 scan = next;
2563 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
2564 pars++;
2565 if (data_fake.flags & SCF_SEEN_ACCEPT) {
2566 if ( stopmin > minnext)
2567 stopmin = min + min1;
2568 flags &= ~SCF_DO_SUBSTR;
2569 if (data)
2570 data->flags |= SCF_SEEN_ACCEPT;
2571 }
2572 if (data) {
2573 if (data_fake.flags & SF_HAS_EVAL)
2574 data->flags |= SF_HAS_EVAL;
2575 data->whilem_c = data_fake.whilem_c;
3dab1dad 2576 }
8aa23a47
YO
2577 if (flags & SCF_DO_STCLASS)
2578 cl_or(pRExC_state, &accum, &this_class);
2579 }
2580 if (code == IFTHEN && num < 2) /* Empty ELSE branch */
2581 min1 = 0;
2582 if (flags & SCF_DO_SUBSTR) {
2583 data->pos_min += min1;
2584 data->pos_delta += max1 - min1;
2585 if (max1 != min1 || is_inf)
2586 data->longest = &(data->longest_float);
2587 }
2588 min += min1;
2589 delta += max1 - min1;
2590 if (flags & SCF_DO_STCLASS_OR) {
2591 cl_or(pRExC_state, data->start_class, &accum);
2592 if (min1) {
2593 cl_and(data->start_class, and_withp);
2594 flags &= ~SCF_DO_STCLASS;
653099ff 2595 }
8aa23a47
YO
2596 }
2597 else if (flags & SCF_DO_STCLASS_AND) {
2598 if (min1) {
2599 cl_and(data->start_class, &accum);
2600 flags &= ~SCF_DO_STCLASS;
de0c8cb8 2601 }
8aa23a47
YO
2602 else {
2603 /* Switch to OR mode: cache the old value of
2604 * data->start_class */
2605 INIT_AND_WITHP;
2606 StructCopy(data->start_class, and_withp,
2607 struct regnode_charclass_class);
2608 flags &= ~SCF_DO_STCLASS_AND;
2609 StructCopy(&accum, data->start_class,
2610 struct regnode_charclass_class);
2611 flags |= SCF_DO_STCLASS_OR;
2612 data->start_class->flags |= ANYOF_EOS;
de0c8cb8 2613 }
8aa23a47 2614 }
a3621e74 2615
8aa23a47
YO
2616 if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
2617 /* demq.
a3621e74 2618
8aa23a47
YO
2619 Assuming this was/is a branch we are dealing with: 'scan' now
2620 points at the item that follows the branch sequence, whatever
2621 it is. We now start at the beginning of the sequence and look
2622 for subsequences of
a3621e74 2623
8aa23a47
YO
2624 BRANCH->EXACT=>x1
2625 BRANCH->EXACT=>x2
2626 tail
a3621e74 2627
8aa23a47 2628 which would be constructed from a pattern like /A|LIST|OF|WORDS/
a3621e74 2629
8aa23a47
YO
2630 If we can find such a subseqence we need to turn the first
2631 element into a trie and then add the subsequent branch exact
2632 strings to the trie.
a3621e74 2633
8aa23a47 2634 We have two cases
a3621e74 2635
8aa23a47 2636 1. patterns where the whole set of branch can be converted.
a3621e74 2637
8aa23a47 2638 2. patterns where only a subset can be converted.
a3621e74 2639
8aa23a47
YO
2640 In case 1 we can replace the whole set with a single regop
2641 for the trie. In case 2 we need to keep the start and end
2642 branchs so
a3621e74 2643
8aa23a47
YO
2644 'BRANCH EXACT; BRANCH EXACT; BRANCH X'
2645 becomes BRANCH TRIE; BRANCH X;
786e8c11 2646
8aa23a47
YO
2647 There is an additional case, that being where there is a
2648 common prefix, which gets split out into an EXACT like node
2649 preceding the TRIE node.
a3621e74 2650
8aa23a47
YO
2651 If x(1..n)==tail then we can do a simple trie, if not we make
2652 a "jump" trie, such that when we match the appropriate word
2653 we "jump" to the appopriate tail node. Essentailly we turn
2654 a nested if into a case structure of sorts.
b515a41d 2655
8aa23a47
YO
2656 */
2657
2658 int made=0;
2659 if (!re_trie_maxbuff) {
2660 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
2661 if (!SvIOK(re_trie_maxbuff))
2662 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
2663 }
2664 if ( SvIV(re_trie_maxbuff)>=0 ) {
2665 regnode *cur;
2666 regnode *first = (regnode *)NULL;
2667 regnode *last = (regnode *)NULL;
2668 regnode *tail = scan;
2669 U8 optype = 0;
2670 U32 count=0;
a3621e74
YO
2671
2672#ifdef DEBUGGING
8aa23a47 2673 SV * const mysv = sv_newmortal(); /* for dumping */
a3621e74 2674#endif
8aa23a47
YO
2675 /* var tail is used because there may be a TAIL
2676 regop in the way. Ie, the exacts will point to the
2677 thing following the TAIL, but the last branch will
2678 point at the TAIL. So we advance tail. If we
2679 have nested (?:) we may have to move through several
2680 tails.
2681 */
2682
2683 while ( OP( tail ) == TAIL ) {
2684 /* this is the TAIL generated by (?:) */
2685 tail = regnext( tail );
2686 }
a3621e74 2687
8aa23a47
YO
2688
2689 DEBUG_OPTIMISE_r({
2690 regprop(RExC_rx, mysv, tail );
2691 PerlIO_printf( Perl_debug_log, "%*s%s%s\n",
2692 (int)depth * 2 + 2, "",
2693 "Looking for TRIE'able sequences. Tail node is: ",
2694 SvPV_nolen_const( mysv )
2695 );
2696 });
2697
2698 /*
2699
2700 step through the branches, cur represents each
2701 branch, noper is the first thing to be matched
2702 as part of that branch and noper_next is the
2703 regnext() of that node. if noper is an EXACT
2704 and noper_next is the same as scan (our current
2705 position in the regex) then the EXACT branch is
2706 a possible optimization target. Once we have
2707 two or more consequetive such branches we can
2708 create a trie of the EXACT's contents and stich
2709 it in place. If the sequence represents all of
2710 the branches we eliminate the whole thing and
2711 replace it with a single TRIE. If it is a
2712 subsequence then we need to stitch it in. This
2713 means the first branch has to remain, and needs
2714 to be repointed at the item on the branch chain
2715 following the last branch optimized. This could
2716 be either a BRANCH, in which case the
2717 subsequence is internal, or it could be the
2718 item following the branch sequence in which
2719 case the subsequence is at the end.
2720
2721 */
2722
2723 /* dont use tail as the end marker for this traverse */
2724 for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
2725 regnode * const noper = NEXTOPER( cur );
b515a41d 2726#if defined(DEBUGGING) || defined(NOJUMPTRIE)
8aa23a47 2727 regnode * const noper_next = regnext( noper );
b515a41d
YO
2728#endif
2729
8aa23a47
YO
2730 DEBUG_OPTIMISE_r({
2731 regprop(RExC_rx, mysv, cur);
2732 PerlIO_printf( Perl_debug_log, "%*s- %s (%d)",
2733 (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) );
2734
2735 regprop(RExC_rx, mysv, noper);
2736 PerlIO_printf( Perl_debug_log, " -> %s",
2737 SvPV_nolen_const(mysv));
2738
2739 if ( noper_next ) {
2740 regprop(RExC_rx, mysv, noper_next );
2741 PerlIO_printf( Perl_debug_log,"\t=> %s\t",
2742 SvPV_nolen_const(mysv));
2743 }
2744 PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n",
2745 REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) );
2746 });
2747 if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
2748 : PL_regkind[ OP( noper ) ] == EXACT )
2749 || OP(noper) == NOTHING )
786e8c11 2750#ifdef NOJUMPTRIE
8aa23a47 2751 && noper_next == tail
786e8c11 2752#endif
8aa23a47
YO
2753 && count < U16_MAX)
2754 {
2755 count++;
2756 if ( !first || optype == NOTHING ) {
2757 if (!first) first = cur;
2758 optype = OP( noper );
2759 } else {
2760 last = cur;
2761 }
2762 } else {
2763 if ( last ) {
2764 make_trie( pRExC_state,
2765 startbranch, first, cur, tail, count,
2766 optype, depth+1 );
2767 }
2768 if ( PL_regkind[ OP( noper ) ] == EXACT
786e8c11 2769#ifdef NOJUMPTRIE
8aa23a47 2770 && noper_next == tail
786e8c11 2771#endif
8aa23a47
YO
2772 ){
2773 count = 1;
2774 first = cur;
2775 optype = OP( noper );
2776 } else {
2777 count = 0;
2778 first = NULL;
2779 optype = 0;
2780 }
2781 last = NULL;
2782 }
2783 }
2784 DEBUG_OPTIMISE_r({
2785 regprop(RExC_rx, mysv, cur);
2786 PerlIO_printf( Perl_debug_log,
2787 "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2,
2788 "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur));
2789
2790 });
2791 if ( last ) {
2792 made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
3dab1dad 2793#ifdef TRIE_STUDY_OPT
8aa23a47
YO
2794 if ( ((made == MADE_EXACT_TRIE &&
2795 startbranch == first)
2796 || ( first_non_open == first )) &&
2797 depth==0 ) {
2798 flags |= SCF_TRIE_RESTUDY;
2799 if ( startbranch == first
2800 && scan == tail )
2801 {
2802 RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
2803 }
2804 }
3dab1dad 2805#endif
8aa23a47
YO
2806 }
2807 }
2808
2809 } /* do trie */
2810
653099ff 2811 }
8aa23a47
YO
2812 else if ( code == BRANCHJ ) { /* single branch is optimized. */
2813 scan = NEXTOPER(NEXTOPER(scan));
2814 } else /* single branch is optimized. */
2815 scan = NEXTOPER(scan);
2816 continue;
2817 } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
2818 scan_frame *newframe = NULL;
2819 I32 paren;
2820 regnode *start;
2821 regnode *end;
2822
2823 if (OP(scan) != SUSPEND) {
2824 /* set the pointer */
2825 if (OP(scan) == GOSUB) {
2826 paren = ARG(scan);
2827 RExC_recurse[ARG2L(scan)] = scan;
2828 start = RExC_open_parens[paren-1];
2829 end = RExC_close_parens[paren-1];
2830 } else {
2831 paren = 0;
f8fc2ecf 2832 start = RExC_rxi->program + 1;
8aa23a47
YO
2833 end = RExC_opend;
2834 }
2835 if (!recursed) {
2836 Newxz(recursed, (((RExC_npar)>>3) +1), U8);
2837 SAVEFREEPV(recursed);
2838 }
2839 if (!PAREN_TEST(recursed,paren+1)) {
2840 PAREN_SET(recursed,paren+1);
2841 Newx(newframe,1,scan_frame);
2842 } else {
2843 if (flags & SCF_DO_SUBSTR) {
304ee84b 2844 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
2845 data->longest = &(data->longest_float);
2846 }
2847 is_inf = is_inf_internal = 1;
2848 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
2849 cl_anything(pRExC_state, data->start_class);
2850 flags &= ~SCF_DO_STCLASS;
2851 }
2852 } else {
2853 Newx(newframe,1,scan_frame);
2854 paren = stopparen;
2855 start = scan+2;
2856 end = regnext(scan);
2857 }
2858 if (newframe) {
2859 assert(start);
2860 assert(end);
2861 SAVEFREEPV(newframe);
2862 newframe->next = regnext(scan);
2863 newframe->last = last;
2864 newframe->stop = stopparen;
2865 newframe->prev = frame;
2866
2867 frame = newframe;
2868 scan = start;
2869 stopparen = paren;
2870 last = end;
2871
2872 continue;
2873 }
2874 }
2875 else if (OP(scan) == EXACT) {
2876 I32 l = STR_LEN(scan);
2877 UV uc;
2878 if (UTF) {
2879 const U8 * const s = (U8*)STRING(scan);
2880 l = utf8_length(s, s + l);
2881 uc = utf8_to_uvchr(s, NULL);
2882 } else {
2883 uc = *((U8*)STRING(scan));
2884 }
2885 min += l;
2886 if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */
2887 /* The code below prefers earlier match for fixed
2888 offset, later match for variable offset. */
2889 if (data->last_end == -1) { /* Update the start info. */
2890 data->last_start_min = data->pos_min;
2891 data->last_start_max = is_inf
2892 ? I32_MAX : data->pos_min + data->pos_delta;
b515a41d 2893 }
8aa23a47
YO
2894 sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan));
2895 if (UTF)
2896 SvUTF8_on(data->last_found);
2897 {
2898 SV * const sv = data->last_found;
2899 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
2900 mg_find(sv, PERL_MAGIC_utf8) : NULL;
2901 if (mg && mg->mg_len >= 0)
2902 mg->mg_len += utf8_length((U8*)STRING(scan),
2903 (U8*)STRING(scan)+STR_LEN(scan));
b515a41d 2904 }
8aa23a47
YO
2905 data->last_end = data->pos_min + l;
2906 data->pos_min += l; /* As in the first entry. */
2907 data->flags &= ~SF_BEFORE_EOL;
2908 }
2909 if (flags & SCF_DO_STCLASS_AND) {
2910 /* Check whether it is compatible with what we know already! */
2911 int compat = 1;
2912
2913 if (uc >= 0x100 ||
2914 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2915 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2916 && (!(data->start_class->flags & ANYOF_FOLD)
2917 || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2918 )
2919 compat = 0;
2920 ANYOF_CLASS_ZERO(data->start_class);
2921 ANYOF_BITMAP_ZERO(data->start_class);
2922 if (compat)
2923 ANYOF_BITMAP_SET(data->start_class, uc);
2924 data->start_class->flags &= ~ANYOF_EOS;
2925 if (uc < 0x100)
2926 data->start_class->flags &= ~ANYOF_UNICODE_ALL;
2927 }
2928 else if (flags & SCF_DO_STCLASS_OR) {
2929 /* false positive possible if the class is case-folded */
2930 if (uc < 0x100)
2931 ANYOF_BITMAP_SET(data->start_class, uc);
2932 else
2933 data->start_class->flags |= ANYOF_UNICODE_ALL;
2934 data->start_class->flags &= ~ANYOF_EOS;
2935 cl_and(data->start_class, and_withp);
2936 }
2937 flags &= ~SCF_DO_STCLASS;
2938 }
2939 else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */
2940 I32 l = STR_LEN(scan);
2941 UV uc = *((U8*)STRING(scan));
2942
2943 /* Search for fixed substrings supports EXACT only. */
2944 if (flags & SCF_DO_SUBSTR) {
2945 assert(data);
304ee84b 2946 SCAN_COMMIT(pRExC_state, data, minlenp);
8aa23a47
YO
2947 }
2948 if (UTF) {
2949 const U8 * const s = (U8 *)STRING(scan);
2950 l = utf8_length(s, s + l);
2951 uc = utf8_to_uvchr(s, NULL);
2952 }
2953 min += l;
2954 if (flags & SCF_DO_SUBSTR)
2955 data->pos_min += l;
2956 if (flags & SCF_DO_STCLASS_AND) {
2957 /* Check whether it is compatible with what we know already! */
2958 int compat = 1;
2959
2960 if (uc >= 0x100 ||
2961 (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE))
2962 && !ANYOF_BITMAP_TEST(data->start_class, uc)
2963 && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc])))
2964 compat = 0;
2965 ANYOF_CLASS_ZERO(data->start_class);
2966 ANYOF_BITMAP_ZERO(data->start_class);
2967 if (compat) {
2968 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff 2969 data->start_class->flags &= ~ANYOF_EOS;
8aa23a47
YO
2970 data->start_class->flags |= ANYOF_FOLD;
2971 if (OP(scan) == EXACTFL)
2972 data->start_class->flags |= ANYOF_LOCALE;
653099ff 2973 }
8aa23a47
YO
2974 }
2975 else if (flags & SCF_DO_STCLASS_OR) {
2976 if (data->start_class->flags & ANYOF_FOLD) {
2977 /* false positive possible if the class is case-folded.
2978 Assume that the locale settings are the same... */
1aa99e6b
IH
2979 if (uc < 0x100)
2980 ANYOF_BITMAP_SET(data->start_class, uc);
653099ff
GS
2981 data->start_class->flags &= ~ANYOF_EOS;
2982 }
8aa23a47 2983 cl_and(data->start_class, and_withp);
653099ff 2984 }
8aa23a47
YO
2985 flags &= ~SCF_DO_STCLASS;
2986 }
2987 else if (strchr((const char*)PL_varies,OP(scan))) {
2988 I32 mincount, maxcount, minnext, deltanext, fl = 0;
2989 I32 f = flags, pos_before = 0;
2990 regnode * const oscan = scan;
2991 struct regnode_charclass_class this_class;
2992 struct regnode_charclass_class *oclass = NULL;
2993 I32 next_is_eval = 0;
2994
2995 switch (PL_regkind[OP(scan)]) {
2996 case WHILEM: /* End of (?:...)* . */
2997 scan = NEXTOPER(scan);
2998 goto finish;
2999 case PLUS:
3000 if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) {
3001 next = NEXTOPER(scan);
3002 if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) {
3003 mincount = 1;
3004 maxcount = REG_INFTY;
3005 next = regnext(scan);
3006 scan = NEXTOPER(scan);
3007 goto do_curly;
3008 }
3009 }
3010 if (flags & SCF_DO_SUBSTR)
3011 data->pos_min++;
3012 min++;
3013 /* Fall through. */
3014 case STAR:
3015 if (flags & SCF_DO_STCLASS) {
3016 mincount = 0;
3017 maxcount = REG_INFTY;
3018 next = regnext(scan);
3019 scan = NEXTOPER(scan);
3020 goto do_curly;
3021 }
3022 is_inf = is_inf_internal = 1;
3023 scan = regnext(scan);
c277df42 3024 if (flags & SCF_DO_SUBSTR) {
304ee84b 3025 SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
8aa23a47 3026 data->longest = &(data->longest_float);
c277df42 3027 }
8aa23a47
YO
3028 goto optimize_curly_tail;
3029 case CURLY:
3030 if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
3031 && (scan->flags == stopparen))
3032 {
3033 mincount = 1;
3034 maxcount = 1;
3035 } else {
3036 mincount = ARG1(scan);
3037 maxcount = ARG2(scan);
653099ff 3038 }
8aa23a47
YO
3039 next = regnext(scan);
3040 if (OP(scan) == CURLYX) {
3041 I32 lp = (data ? *(data->last_closep) : 0);
3042 scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX);
653099ff 3043 }
8aa23a47
YO
3044 scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS;
3045 next_is_eval = (OP(scan) == EVAL);
3046 do_curly:
3047 if (flags & SCF_DO_SUBSTR) {
304ee84b 3048 if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
8aa23a47 3049 pos_before = data->pos_min;
b45f050a 3050 }
8aa23a47
YO
3051 if (data) {
3052 fl = data->flags;
3053 data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL);
3054 if (is_inf)
3055 data->flags |= SF_IS_INF;
3056 }
3057 if (flags & SCF_DO_STCLASS) {
3058 cl_init(pRExC_state, &this_class);
3059 oclass = data->start_class;
3060 data->start_class = &this_class;
3061 f |= SCF_DO_STCLASS_AND;
3062 f &= ~SCF_DO_STCLASS_OR;
3063 }
3064 /* These are the cases when once a subexpression
3065 fails at a particular position, it cannot succeed
3066 even after backtracking at the enclosing scope.
3067
3068 XXXX what if minimal match and we are at the
3069 initial run of {n,m}? */
3070 if ((mincount != maxcount - 1) && (maxcount != REG_INFTY))
3071 f &= ~SCF_WHILEM_VISITED_POS;
b45f050a 3072
8aa23a47
YO
3073 /* This will finish on WHILEM, setting scan, or on NULL: */
3074 minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
3075 last, data, stopparen, recursed, NULL,
3076 (mincount == 0
3077 ? (f & ~SCF_DO_SUBSTR) : f),depth+1);
b515a41d 3078
8aa23a47
YO
3079 if (flags & SCF_DO_STCLASS)
3080 data->start_class = oclass;
3081 if (mincount == 0 || minnext == 0) {
3082 if (flags & SCF_DO_STCLASS_OR) {
3083 cl_or(pRExC_state, data->start_class, &this_class);
3084 }
3085 else if (flags & SCF_DO_STCLASS_AND) {
3086 /* Switch to OR mode: cache the old value of
3087 * data->start_class */
3088 INIT_AND_WITHP;
3089 StructCopy(data->start_class, and_withp,
3090 struct regnode_charclass_class);
3091 flags &= ~SCF_DO_STCLASS_AND;
3092 StructCopy(&this_class, data->start_class,
3093 struct regnode_charclass_class);
3094 flags |= SCF_DO_STCLASS_OR;
3095 data->start_class->flags |= ANYOF_EOS;
3096 }
3097 } else { /* Non-zero len */
3098 if (flags & SCF_DO_STCLASS_OR) {
3099 cl_or(pRExC_state, data->start_class, &this_class);
3100 cl_and(data->start_class, and_withp);
3101 }
3102 else if (flags & SCF_DO_STCLASS_AND)
3103 cl_and(data->start_class, &this_class);
3104 flags &= ~SCF_DO_STCLASS;
3105 }
3106 if (!scan) /* It was not CURLYX, but CURLY. */
3107 scan = next;
3108 if ( /* ? quantifier ok, except for (?{ ... }) */
3109 (next_is_eval || !(mincount == 0 && maxcount == 1))
3110 && (minnext == 0) && (deltanext == 0)
3111 && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR))
3112 && maxcount <= REG_INFTY/3 /* Complement check for big count */
3113 && ckWARN(WARN_REGEXP))
3114 {
3115 vWARN(RExC_parse,
3116 "Quantifier unexpected on zero-length expression");
3117 }
3118
3119 min += minnext * mincount;
3120 is_inf_internal |= ((maxcount == REG_INFTY
3121 && (minnext + deltanext) > 0)
3122 || deltanext == I32_MAX);
3123 is_inf |= is_inf_internal;
3124 delta += (minnext + deltanext) * maxcount - minnext * mincount;
3125
3126 /* Try powerful optimization CURLYX => CURLYN. */
3127 if ( OP(oscan) == CURLYX && data
3128 && data->flags & SF_IN_PAR
3129 && !(data->flags & SF_HAS_EVAL)
3130 && !deltanext && minnext == 1 ) {
3131 /* Try to optimize to CURLYN. */
3132 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS;
3133 regnode * const nxt1 = nxt;
497b47a8 3134#ifdef DEBUGGING
8aa23a47 3135 regnode *nxt2;
497b47a8 3136#endif
c277df42 3137
8aa23a47
YO
3138 /* Skip open. */
3139 nxt = regnext(nxt);
3140 if (!strchr((const char*)PL_simple,OP(nxt))
3141 && !(PL_regkind[OP(nxt)] == EXACT
3142 && STR_LEN(nxt) == 1))
3143 goto nogo;
497b47a8 3144#ifdef DEBUGGING
8aa23a47 3145 nxt2 = nxt;
497b47a8 3146#endif
8aa23a47
YO
3147 nxt = regnext(nxt);
3148 if (OP(nxt) != CLOSE)
3149 goto nogo;
3150 if (RExC_open_parens) {
3151 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3152 RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
3153 }
3154 /* Now we know that nxt2 is the only contents: */
3155 oscan->flags = (U8)ARG(nxt);
3156 OP(oscan) = CURLYN;
3157 OP(nxt1) = NOTHING; /* was OPEN. */
40d049e4 3158
c277df42 3159#ifdef DEBUGGING
8aa23a47
YO
3160 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3161 NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */
3162 NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */
3163 OP(nxt) = OPTIMIZED; /* was CLOSE. */
3164 OP(nxt + 1) = OPTIMIZED; /* was count. */
3165 NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */
b81d288d 3166#endif
8aa23a47
YO
3167 }
3168 nogo:
3169
3170 /* Try optimization CURLYX => CURLYM. */
3171 if ( OP(oscan) == CURLYX && data
3172 && !(data->flags & SF_HAS_PAR)
3173 && !(data->flags & SF_HAS_EVAL)
3174 && !deltanext /* atom is fixed width */
3175 && minnext != 0 /* CURLYM can't handle zero width */
3176 ) {
3177 /* XXXX How to optimize if data == 0? */
3178 /* Optimize to a simpler form. */
3179 regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */
3180 regnode *nxt2;
3181
3182 OP(oscan) = CURLYM;
3183 while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/
3184 && (OP(nxt2) != WHILEM))
3185 nxt = nxt2;
3186 OP(nxt2) = SUCCEED; /* Whas WHILEM */
3187 /* Need to optimize away parenths. */
3188 if (data->flags & SF_IN_PAR) {
3189 /* Set the parenth number. */
3190 regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/
3191
3192 if (OP(nxt) != CLOSE)
3193 FAIL("Panic opt close");
3194 oscan->flags = (U8)ARG(nxt);
3195 if (RExC_open_parens) {
3196 RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
3197 RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
40d049e4 3198 }
8aa23a47
YO
3199 OP(nxt1) = OPTIMIZED; /* was OPEN. */
3200 OP(nxt) = OPTIMIZED; /* was CLOSE. */
40d049e4 3201
c277df42 3202#ifdef DEBUGGING
8aa23a47
YO
3203 OP(nxt1 + 1) = OPTIMIZED; /* was count. */
3204 OP(nxt + 1) = OPTIMIZED; /* was count. */
3205 NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */
3206 NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */
b81d288d 3207#endif
c277df42 3208#if 0
8aa23a47
YO
3209 while ( nxt1 && (OP(nxt1) != WHILEM)) {
3210 regnode *nnxt = regnext(nxt1);
3211
3212 if (nnxt == nxt) {
3213 if (reg_off_by_arg[OP(nxt1)])
3214 ARG_SET(nxt1, nxt2 - nxt1);
3215 else if (nxt2 - nxt1 < U16_MAX)
3216 NEXT_OFF(nxt1) = nxt2 - nxt1;
3217 else
3218 OP(nxt) = NOTHING; /* Cannot beautify */
c277df42 3219 }
8aa23a47 3220 nxt1 = nnxt;
c277df42 3221 }
5d1c421c 3222#endif
8aa23a47
YO
3223 /* Optimize again: */
3224 study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
3225 NULL, stopparen, recursed, NULL, 0,depth+1);
3226 }
3227 else
3228 oscan->flags = 0;
3229 }
3230 else if ((OP(oscan) == CURLYX)
3231 && (flags & SCF_WHILEM_VISITED_POS)
3232 /* See the comment on a similar expression above.
3233 However, this time it not a subexpression
3234 we care about, but the expression itself. */
3235 && (maxcount == REG_INFTY)
3236 && data && ++data->whilem_c < 16) {
3237 /* This stays as CURLYX, we can put the count/of pair. */
3238 /* Find WHILEM (as in regexec.c) */
3239 regnode *nxt = oscan + NEXT_OFF(oscan);
3240
3241 if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */
3242 nxt += ARG(nxt);
3243 PREVOPER(nxt)->flags = (U8)(data->whilem_c
3244 | (RExC_whilem_seen << 4)); /* On WHILEM */
3245 }
3246 if (data && fl & (SF_HAS_PAR|SF_IN_PAR))
3247 pars++;
3248 if (flags & SCF_DO_SUBSTR) {
3249 SV *last_str = NULL;
3250 int counted = mincount != 0;
a0ed51b3 3251
8aa23a47
YO
3252 if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */
3253#if defined(SPARC64_GCC_WORKAROUND)
3254 I32 b = 0;
3255 STRLEN l = 0;
3256 const char *s = NULL;
3257 I32 old = 0;
b515a41d 3258
8aa23a47
YO
3259 if (pos_before >= data->last_start_min)
3260 b = pos_before;
3261 else
3262 b = data->last_start_min;
b515a41d 3263
8aa23a47
YO
3264 l = 0;
3265 s = SvPV_const(data->last_found, l);
3266 old = b - data->last_start_min;
3267
3268#else
3269 I32 b = pos_before >= data->last_start_min
3270 ? pos_before : data->last_start_min;
3271 STRLEN l;
3272 const char * const s = SvPV_const(data->last_found, l);
3273 I32 old = b - data->last_start_min;
3274#endif
3275
3276 if (UTF)
3277 old = utf8_hop((U8*)s, old) - (U8*)s;
3278
3279 l -= old;
3280 /* Get the added string: */
3281 last_str = newSVpvn(s + old, l);
3282 if (UTF)
3283 SvUTF8_on(last_str);
3284 if (deltanext == 0 && pos_before == b) {
3285 /* What was added is a constant string */
3286 if (mincount > 1) {
3287 SvGROW(last_str, (mincount * l) + 1);
3288 repeatcpy(SvPVX(last_str) + l,
3289 SvPVX_const(last_str), l, mincount - 1);
3290 SvCUR_set(last_str, SvCUR(last_str) * mincount);
3291 /* Add additional parts. */
3292 SvCUR_set(data->last_found,
3293 SvCUR(data->last_found) - l);
3294 sv_catsv(data->last_found, last_str);
3295 {
3296 SV * sv = data->last_found;
3297 MAGIC *mg =
3298 SvUTF8(sv) && SvMAGICAL(sv) ?
3299 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3300 if (mg && mg->mg_len >= 0)
3301 mg->mg_len += CHR_SVLEN(last_str);
b515a41d 3302 }
8aa23a47 3303 data->last_end += l * (mincount - 1);
b515a41d 3304 }
8aa23a47
YO
3305 } else {
3306 /* start offset must point into the last copy */
3307 data->last_start_min += minnext * (mincount - 1);
3308 data->last_start_max += is_inf ? I32_MAX
3309 : (maxcount - 1) * (minnext + data->pos_delta);
3310 }
c277df42 3311 }
8aa23a47
YO
3312 /* It is counted once already... */
3313 data->pos_min += minnext * (mincount - counted);
3314 data->pos_delta += - counted * deltanext +
3315 (minnext + deltanext) * maxcount - minnext * mincount;
3316 if (mincount != maxcount) {
3317 /* Cannot extend fixed substrings found inside
3318 the group. */
304ee84b 3319 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3320 if (mincount && last_str) {
3321 SV * const sv = data->last_found;
3322 MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
3323 mg_find(sv, PERL_MAGIC_utf8) : NULL;
3324
3325 if (mg)
3326 mg->mg_len = -1;
3327 sv_setsv(sv, last_str);
3328 data->last_end = data->pos_min;
3329 data->last_start_min =
3330 data->pos_min - CHR_SVLEN(last_str);
3331 data->last_start_max = is_inf
3332 ? I32_MAX
3333 : data->pos_min + data->pos_delta
3334 - CHR_SVLEN(last_str);
3335 }
3336 data->longest = &(data->longest_float);
3337 }
3338 SvREFCNT_dec(last_str);
c277df42 3339 }
8aa23a47
YO
3340 if (data && (fl & SF_HAS_EVAL))
3341 data->flags |= SF_HAS_EVAL;
3342 optimize_curly_tail:
3343 if (OP(oscan) != CURLYX) {
3344 while (PL_regkind[OP(next = regnext(oscan))] == NOTHING
3345 && NEXT_OFF(next))
3346 NEXT_OFF(oscan) += NEXT_OFF(next);
3347 }
3348 continue;
3349 default: /* REF and CLUMP only? */
3350 if (flags & SCF_DO_SUBSTR) {
304ee84b 3351 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
8aa23a47
YO
3352 data->longest = &(data->longest_float);
3353 }
3354 is_inf = is_inf_internal = 1;
3355 if (flags & SCF_DO_STCLASS_OR)
3356 cl_anything(pRExC_state, data->start_class);
3357 flags &= ~SCF_DO_STCLASS;
3358 break;
c277df42 3359 }
8aa23a47 3360 }
e1d1eefb
YO
3361 else if (OP(scan) == LNBREAK) {
3362 if (flags & SCF_DO_STCLASS) {
3363 int value = 0;
3364 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
3365 if (flags & SCF_DO_STCLASS_AND) {
3366 for (value = 0; value < 256; value++)
e64b1bd1 3367 if (!is_VERTWS_cp(value))
e1d1eefb
YO
3368 ANYOF_BITMAP_CLEAR(data->start_class, value);
3369 }
3370 else {
3371 for (value = 0; value < 256; value++)
e64b1bd1 3372 if (is_VERTWS_cp(value))
e1d1eefb
YO
3373 ANYOF_BITMAP_SET(data->start_class, value);
3374 }
3375 if (flags & SCF_DO_STCLASS_OR)
3376 cl_and(data->start_class, and_withp);
3377 flags &= ~SCF_DO_STCLASS;
3378 }
3379 min += 1;
f9a79580 3380 delta += 1;
e1d1eefb
YO
3381 if (flags & SCF_DO_SUBSTR) {
3382 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3383 data->pos_min += 1;
f9a79580 3384 data->pos_delta += 1;
e1d1eefb
YO
3385 data->longest = &(data->longest_float);
3386 }
3387
3388 }
f9a79580
RGS
3389 else if (OP(scan) == FOLDCHAR) {
3390 int d = ARG(scan)==0xDF ? 1 : 2;
3391 flags &= ~SCF_DO_STCLASS;
3392 min += 1;
3393 delta += d;
3394 if (flags & SCF_DO_SUBSTR) {
3395 SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */
3396 data->pos_min += 1;
3397 data->pos_delta += d;
3398 data->longest = &(data->longest_float);
3399 }
3400 }
8aa23a47
YO
3401 else if (strchr((const char*)PL_simple,OP(scan))) {
3402 int value = 0;
653099ff 3403
8aa23a47 3404 if (flags & SCF_DO_SUBSTR) {
304ee84b 3405 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47
YO
3406 data->pos_min++;
3407 }
3408 min++;
3409 if (flags & SCF_DO_STCLASS) {
3410 data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */
b515a41d 3411
8aa23a47
YO
3412 /* Some of the logic below assumes that switching
3413 locale on will only add false positives. */
3414 switch (PL_regkind[OP(scan)]) {
3415 case SANY:
3416 default:
3417 do_default:
3418 /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */
3419 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
3420 cl_anything(pRExC_state, data->start_class);
3421 break;
3422 case REG_ANY:
3423 if (OP(scan) == SANY)
3424 goto do_default;
3425 if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */
3426 value = (ANYOF_BITMAP_TEST(data->start_class,'\n')
3427 || (data->start_class->flags & ANYOF_CLASS));
3428 cl_anything(pRExC_state, data->start_class);
653099ff 3429 }
8aa23a47
YO
3430 if (flags & SCF_DO_STCLASS_AND || !value)
3431 ANYOF_BITMAP_CLEAR(data->start_class,'\n');
3432 break;
3433 case ANYOF:
3434 if (flags & SCF_DO_STCLASS_AND)
3435 cl_and(data->start_class,
3436 (struct regnode_charclass_class*)scan);
653099ff 3437 else
8aa23a47
YO
3438 cl_or(pRExC_state, data->start_class,
3439 (struct regnode_charclass_class*)scan);
3440 break;
3441 case ALNUM:
3442 if (flags & SCF_DO_STCLASS_AND) {
3443 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3444 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3445 for (value = 0; value < 256; value++)
3446 if (!isALNUM(value))
3447 ANYOF_BITMAP_CLEAR(data->start_class, value);
3448 }
653099ff 3449 }
8aa23a47
YO
3450 else {
3451 if (data->start_class->flags & ANYOF_LOCALE)
3452 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3453 else {
3454 for (value = 0; value < 256; value++)
3455 if (isALNUM(value))
3456 ANYOF_BITMAP_SET(data->start_class, value);
653099ff 3457 }
8aa23a47
YO
3458 }
3459 break;
3460 case ALNUML:
3461 if (flags & SCF_DO_STCLASS_AND) {
3462 if (data->start_class->flags & ANYOF_LOCALE)
3463 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM);
3464 }
3465 else {
3466 ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM);
3467 data->start_class->flags |= ANYOF_LOCALE;
3468 }
3469 break;
3470 case NALNUM:
3471 if (flags & SCF_DO_STCLASS_AND) {
3472 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3473 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
3474 for (value = 0; value < 256; value++)
3475 if (isALNUM(value))
3476 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff
GS
3477 }
3478 }
8aa23a47
YO
3479 else {
3480 if (data->start_class->flags & ANYOF_LOCALE)
3481 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3482 else {
3483 for (value = 0; value < 256; value++)
3484 if (!isALNUM(value))
3485 ANYOF_BITMAP_SET(data->start_class, value);
3486 }
653099ff 3487 }
8aa23a47
YO
3488 break;
3489 case NALNUML:
3490 if (flags & SCF_DO_STCLASS_AND) {
3491 if (data->start_class->flags & ANYOF_LOCALE)
3492 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM);
653099ff 3493 }
8aa23a47
YO
3494 else {
3495 data->start_class->flags |= ANYOF_LOCALE;
3496 ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM);
3497 }
3498 break;
3499 case SPACE:
3500 if (flags & SCF_DO_STCLASS_AND) {
3501 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3502 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3503 for (value = 0; value < 256; value++)
3504 if (!isSPACE(value))
3505 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff
GS
3506 }
3507 }
8aa23a47
YO
3508 else {
3509 if (data->start_class->flags & ANYOF_LOCALE)
3510 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3511 else {
3512 for (value = 0; value < 256; value++)
3513 if (isSPACE(value))
3514 ANYOF_BITMAP_SET(data->start_class, value);
3515 }
653099ff 3516 }
8aa23a47
YO
3517 break;
3518 case SPACEL:
3519 if (flags & SCF_DO_STCLASS_AND) {
3520 if (data->start_class->flags & ANYOF_LOCALE)
3521 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE);
3522 }
3523 else {
3524 data->start_class->flags |= ANYOF_LOCALE;
3525 ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE);
3526 }
3527 break;
3528 case NSPACE:
3529 if (flags & SCF_DO_STCLASS_AND) {
3530 if (!(data->start_class->flags & ANYOF_LOCALE)) {
3531 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3532 for (value = 0; value < 256; value++)
3533 if (isSPACE(value))
3534 ANYOF_BITMAP_CLEAR(data->start_class, value);
653099ff 3535 }
8aa23a47
YO
3536 }
3537 else {
3538 if (data->start_class->flags & ANYOF_LOCALE)
3539 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3540 else {
3541 for (value = 0; value < 256; value++)
3542 if (!isSPACE(value))
3543 ANYOF_BITMAP_SET(data->start_class, value);
653099ff
GS
3544 }
3545 }
8aa23a47
YO
3546 break;
3547 case NSPACEL:
3548 if (flags & SCF_DO_STCLASS_AND) {
3549 if (data->start_class->flags & ANYOF_LOCALE) {
3550 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE);
3551 for (value = 0; value < 256; value++)
3552 if (!isSPACE(value))
3553 ANYOF_BITMAP_CLEAR(data->start_class, value);
3554 }
653099ff 3555 }
8aa23a47
YO
3556 else {
3557 data->start_class->flags |= ANYOF_LOCALE;
3558 ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE);
3559 }
3560 break;
3561 case DIGIT:
3562 if (flags & SCF_DO_STCLASS_AND) {
3563 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT);
3564 for (value = 0; value < 256; value++)
3565 if (!isDIGIT(value))
3566 ANYOF_BITMAP_CLEAR(data->start_class, value);
3567 }
3568 else {
3569 if (data->start_class->flags & ANYOF_LOCALE)
3570 ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT);
3571 else {
3572 for (value = 0; value < 256; value++)
3573 if (isDIGIT(value))
3574 ANYOF_BITMAP_SET(data->start_class, value);
3575 }
3576 }
3577 break;
3578 case NDIGIT:
3579 if (flags & SCF_DO_STCLASS_AND) {
3580 ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT);
3581 for (value = 0; value < 256; value++)
3582 if (isDIGIT(value))
3583 ANYOF_BITMAP_CLEAR(data->start_class, value);
3584 }
3585 else {
3586 if (data->start_class->flags & ANYOF_LOCALE)
3587 ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT);
3588 else {
3589 for (value = 0; value < 256; value++)
3590 if (!isDIGIT(value))
3591 ANYOF_BITMAP_SET(data->start_class, value);
653099ff
GS
3592 }
3593 }
8aa23a47 3594 break;
e1d1eefb
YO
3595 CASE_SYNST_FNC(VERTWS);
3596 CASE_SYNST_FNC(HORIZWS);
3597
8aa23a47
YO
3598 }
3599 if (flags & SCF_DO_STCLASS_OR)
3600 cl_and(data->start_class, and_withp);
3601 flags &= ~SCF_DO_STCLASS;
3602 }
3603 }
3604 else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) {
3605 data->flags |= (OP(scan) == MEOL
3606 ? SF_BEFORE_MEOL
3607 : SF_BEFORE_SEOL);
3608 }
3609 else if ( PL_regkind[OP(scan)] == BRANCHJ
3610 /* Lookbehind, or need to calculate parens/evals/stclass: */
3611 && (scan->flags || data || (flags & SCF_DO_STCLASS))
3612 && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) {
3613 if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3614 || OP(scan) == UNLESSM )
3615 {
3616 /* Negative Lookahead/lookbehind
3617 In this case we can't do fixed string optimisation.
3618 */
1de06328 3619
8aa23a47
YO
3620 I32 deltanext, minnext, fake = 0;
3621 regnode *nscan;
3622 struct regnode_charclass_class intrnl;
3623 int f = 0;
1de06328 3624
8aa23a47
YO
3625 data_fake.flags = 0;
3626 if (data) {
3627 data_fake.whilem_c = data->whilem_c;
3628 data_fake.last_closep = data->last_closep;
c277df42 3629 }
8aa23a47
YO
3630 else
3631 data_fake.last_closep = &fake;
58e23c8d 3632 data_fake.pos_delta = delta;
8aa23a47
YO
3633 if ( flags & SCF_DO_STCLASS && !scan->flags
3634 && OP(scan) == IFMATCH ) { /* Lookahead */
3635 cl_init(pRExC_state, &intrnl);
3636 data_fake.start_class = &intrnl;
3637 f |= SCF_DO_STCLASS_AND;
3638 }
3639 if (flags & SCF_WHILEM_VISITED_POS)
3640 f |= SCF_WHILEM_VISITED_POS;
3641 next = regnext(scan);
3642 nscan = NEXTOPER(NEXTOPER(scan));
3643 minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
3644 last, &data_fake, stopparen, recursed, NULL, f, depth+1);
3645 if (scan->flags) {
3646 if (deltanext) {
58e23c8d 3647 FAIL("Variable length lookbehind not implemented");
8aa23a47
YO
3648 }
3649 else if (minnext > (I32)U8_MAX) {
58e23c8d 3650 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
8aa23a47
YO
3651 }
3652 scan->flags = (U8)minnext;
3653 }
3654 if (data) {
3655 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3656 pars++;
3657 if (data_fake.flags & SF_HAS_EVAL)
3658 data->flags |= SF_HAS_EVAL;
3659 data->whilem_c = data_fake.whilem_c;
3660 }
3661 if (f & SCF_DO_STCLASS_AND) {
3662 const int was = (data->start_class->flags & ANYOF_EOS);
3663
3664 cl_and(data->start_class, &intrnl);
3665 if (was)
3666 data->start_class->flags |= ANYOF_EOS;
3667 }
cb434fcc 3668 }
8aa23a47
YO
3669#if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
3670 else {
3671 /* Positive Lookahead/lookbehind
3672 In this case we can do fixed string optimisation,
3673 but we must be careful about it. Note in the case of
3674 lookbehind the positions will be offset by the minimum
3675 length of the pattern, something we won't know about
3676 until after the recurse.
3677 */
3678 I32 deltanext, fake = 0;
3679 regnode *nscan;
3680 struct regnode_charclass_class intrnl;
3681 int f = 0;
3682 /* We use SAVEFREEPV so that when the full compile
3683 is finished perl will clean up the allocated
3684 minlens when its all done. This was we don't
3685 have to worry about freeing them when we know
3686 they wont be used, which would be a pain.
3687 */
3688 I32 *minnextp;
3689 Newx( minnextp, 1, I32 );
3690 SAVEFREEPV(minnextp);
3691
3692 if (data) {
3693 StructCopy(data, &data_fake, scan_data_t);
3694 if ((flags & SCF_DO_SUBSTR) && data->last_found) {
3695 f |= SCF_DO_SUBSTR;
3696 if (scan->flags)
304ee84b 3697 SCAN_COMMIT(pRExC_state, &data_fake,minlenp);
8aa23a47
YO
3698 data_fake.last_found=newSVsv(data->last_found);
3699 }
3700 }
3701 else
3702 data_fake.last_closep = &fake;
3703 data_fake.flags = 0;
58e23c8d 3704 data_fake.pos_delta = delta;
8aa23a47
YO
3705 if (is_inf)
3706 data_fake.flags |= SF_IS_INF;
3707 if ( flags & SCF_DO_STCLASS && !scan->flags
3708 && OP(scan) == IFMATCH ) { /* Lookahead */
3709 cl_init(pRExC_state, &intrnl);
3710 data_fake.start_class = &intrnl;
3711 f |= SCF_DO_STCLASS_AND;
3712 }
3713 if (flags & SCF_WHILEM_VISITED_POS)
3714 f |= SCF_WHILEM_VISITED_POS;
3715 next = regnext(scan);
3716 nscan = NEXTOPER(NEXTOPER(scan));
3717
3718 *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
3719 last, &data_fake, stopparen, recursed, NULL, f,depth+1);
3720 if (scan->flags) {
3721 if (deltanext) {
58e23c8d 3722 FAIL("Variable length lookbehind not implemented");
8aa23a47
YO
3723 }
3724 else if (*minnextp > (I32)U8_MAX) {
58e23c8d 3725 FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
8aa23a47
YO
3726 }
3727 scan->flags = (U8)*minnextp;
3728 }
3729
3730 *minnextp += min;
3731
3732 if (f & SCF_DO_STCLASS_AND) {
3733 const int was = (data->start_class->flags & ANYOF_EOS);
3734
3735 cl_and(data->start_class, &intrnl);
3736 if (was)
3737 data->start_class->flags |= ANYOF_EOS;
3738 }
3739 if (data) {
3740 if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
3741 pars++;
3742 if (data_fake.flags & SF_HAS_EVAL)
3743 data->flags |= SF_HAS_EVAL;
3744 data->whilem_c = data_fake.whilem_c;
3745 if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
3746 if (RExC_rx->minlen<*minnextp)
3747 RExC_rx->minlen=*minnextp;
304ee84b 3748 SCAN_COMMIT(pRExC_state, &data_fake, minnextp);
8aa23a47
YO
3749 SvREFCNT_dec(data_fake.last_found);
3750
3751 if ( data_fake.minlen_fixed != minlenp )
3752 {
3753 data->offset_fixed= data_fake.offset_fixed;
3754 data->minlen_fixed= data_fake.minlen_fixed;
3755 data->lookbehind_fixed+= scan->flags;
3756 }
3757 if ( data_fake.minlen_float != minlenp )
3758 {
3759 data->minlen_float= data_fake.minlen_float;
3760 data->offset_float_min=data_fake.offset_float_min;
3761 data->offset_float_max=data_fake.offset_float_max;
3762 data->lookbehind_float+= scan->flags;
3763 }
3764 }
3765 }
3766
3767
40d049e4 3768 }
8aa23a47
YO
3769#endif
3770 }
3771 else if (OP(scan) == OPEN) {
3772 if (stopparen != (I32)ARG(scan))
3773 pars++;
3774 }
3775 else if (OP(scan) == CLOSE) {
3776 if (stopparen == (I32)ARG(scan)) {
3777 break;
3778 }
3779 if ((I32)ARG(scan) == is_par) {
3780 next = regnext(scan);
b515a41d 3781
8aa23a47
YO
3782 if ( next && (OP(next) != WHILEM) && next < last)
3783 is_par = 0; /* Disable optimization */
40d049e4 3784 }
8aa23a47
YO
3785 if (data)
3786 *(data->last_closep) = ARG(scan);
3787 }
3788 else if (OP(scan) == EVAL) {
c277df42
IZ
3789 if (data)
3790 data->flags |= SF_HAS_EVAL;
8aa23a47
YO
3791 }
3792 else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
3793 if (flags & SCF_DO_SUBSTR) {
304ee84b 3794 SCAN_COMMIT(pRExC_state,data,minlenp);
8aa23a47 3795 flags &= ~SCF_DO_SUBSTR;
40d049e4 3796 }
8aa23a47
YO
3797 if (data && OP(scan)==ACCEPT) {
3798 data->flags |= SCF_SEEN_ACCEPT;
3799 if (stopmin > min)
3800 stopmin = min;
e2e6a0f1 3801 }
8aa23a47
YO
3802 }
3803 else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
3804 {
0f5d15d6 3805 if (flags & SCF_DO_SUBSTR) {
304ee84b 3806 SCAN_COMMIT(pRExC_state,data,minlenp);
0f5d15d6
IZ
3807 data->longest = &(data->longest_float);
3808 }
3809 is_inf = is_inf_internal = 1;
653099ff 3810 if (flags & SCF_DO_STCLASS_OR) /* Allow everything */
830247a4