This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Allow more EXACTFish nodes to be trieable
[perl5.git] / regcomp.c
CommitLineData
a0d0e21e
LW
1/* regcomp.c
2 */
3
4/*
4ac71550
TC
5 * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee
6 *
7 * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"]
a0d0e21e
LW
8 */
9
61296642
DM
10/* This file contains functions for compiling a regular expression. See
11 * also regexec.c which funnily enough, contains functions for executing
166f8a29 12 * a regular expression.
e4a054ea
DM
13 *
14 * This file is also copied at build time to ext/re/re_comp.c, where
15 * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT.
16 * This causes the main functions to be compiled under new names and with
17 * debugging support added, which makes "use re 'debug'" work.
166f8a29
DM
18 */
19
a687059c
LW
20/* NOTE: this is derived from Henry Spencer's regexp code, and should not
21 * confused with the original package (see point 3 below). Thanks, Henry!
22 */
23
24/* Additional note: this code is very heavily munged from Henry's version
25 * in places. In some spots I've traded clarity for efficiency, so don't
26 * blame Henry for some of the lack of readability.
27 */
28
e50aee73 29/* The names of the functions have been changed from regcomp and
3b753521 30 * regexec to pregcomp and pregexec in order to avoid conflicts
e50aee73
AD
31 * with the POSIX routines of the same names.
32*/
33
b9d5759e 34#ifdef PERL_EXT_RE_BUILD
54df2634 35#include "re_top.h"
b81d288d 36#endif
56953603 37
a687059c 38/*
e50aee73 39 * pregcomp and pregexec -- regsub and regerror are not used in perl
a687059c
LW
40 *
41 * Copyright (c) 1986 by University of Toronto.
42 * Written by Henry Spencer. Not derived from licensed software.
43 *
44 * Permission is granted to anyone to use this software for any
45 * purpose on any computer system, and to redistribute it freely,
46 * subject to the following restrictions:
47 *
48 * 1. The author is not responsible for the consequences of use of
49 * this software, no matter how awful, even if they arise
50 * from defects in it.
51 *
52 * 2. The origin of this software must not be misrepresented, either
53 * by explicit claim or by omission.
54 *
55 * 3. Altered versions must be plainly marked as such, and must not
56 * be misrepresented as being the original software.
57 *
58 *
59 **** Alterations to Henry's code are...
60 ****
4bb101f2 61 **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
1129b882
NC
62 **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
63 **** by Larry Wall and others
a687059c 64 ****
9ef589d8
LW
65 **** You may distribute under the terms of either the GNU General Public
66 **** License or the Artistic License, as specified in the README file.
67
a687059c
LW
68 *
69 * Beware that some of this code is subtly aware of the way operator
70 * precedence is structured in regular expressions. Serious changes in
71 * regular-expression syntax might require a total rethink.
72 */
73#include "EXTERN.h"
864dbfa3 74#define PERL_IN_REGCOMP_C
a687059c 75#include "perl.h"
d06ea78c 76
acfe0abc 77#ifndef PERL_IN_XSUB_RE
d06ea78c
GS
78# include "INTERN.h"
79#endif
c277df42
IZ
80
81#define REG_COMP_C
54df2634
NC
82#ifdef PERL_IN_XSUB_RE
83# include "re_comp.h"
afda64e8 84EXTERN_C const struct regexp_engine my_reg_engine;
54df2634
NC
85#else
86# include "regcomp.h"
87#endif
a687059c 88
f7e03a10 89#include "dquote_inline.h"
b992490d 90#include "invlist_inline.h"
1b0f46bf 91#include "unicode_constants.h"
04e98a4d 92
538e84ed
KW
93#define HAS_NONLATIN1_FOLD_CLOSURE(i) \
94 _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
f12c0118
KW
95#define HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(i) \
96 _HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)
26faadbd 97#define IS_NON_FINAL_FOLD(c) _IS_NON_FINAL_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
2c61f163 98#define IS_IN_SOME_FOLD_L1(c) _IS_IN_SOME_FOLD_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
94dc5c2d 99
a687059c
LW
100#ifndef STATIC
101#define STATIC static
102#endif
103
3f910e62
YO
104/* this is a chain of data about sub patterns we are processing that
105 need to be handled separately/specially in study_chunk. Its so
106 we can simulate recursion without losing state. */
107struct scan_frame;
108typedef struct scan_frame {
109 regnode *last_regnode; /* last node to process in this frame */
110 regnode *next_regnode; /* next node to process when last is reached */
111 U32 prev_recursed_depth;
112 I32 stopparen; /* what stopparen do we use */
3f910e62
YO
113
114 struct scan_frame *this_prev_frame; /* this previous frame */
115 struct scan_frame *prev_frame; /* previous frame */
116 struct scan_frame *next_frame; /* next frame */
117} scan_frame;
118
f4ae5a27
KW
119/* Certain characters are output as a sequence with the first being a
120 * backslash. */
eb7486de 121#define isBACKSLASHED_PUNCT(c) strchr("-[]\\^", c)
f4ae5a27
KW
122
123
09b2b2e6 124struct RExC_state_t {
514a91f1
DM
125 U32 flags; /* RXf_* are we folding, multilining? */
126 U32 pm_flags; /* PMf_* stuff from the calling PMOP */
830247a4 127 char *precomp; /* uncompiled string. */
711b303b 128 char *precomp_end; /* pointer to end of uncompiled string. */
288b8c02 129 REGEXP *rx_sv; /* The SV that is the regexp. */
f8fc2ecf 130 regexp *rx; /* perl core regexp structure */
538e84ed
KW
131 regexp_internal *rxi; /* internal data for regexp object
132 pprivate field */
fac92740 133 char *start; /* Start of input for compile */
830247a4
IZ
134 char *end; /* End of input for compile */
135 char *parse; /* Input-scan pointer. */
51684e3c
KW
136 char *copy_start; /* start of copy of input within
137 constructed parse string */
138 char *copy_start_in_input; /* Position in input string
139 corresponding to copy_start */
ea3daa5d 140 SSize_t whilem_seen; /* number of WHILEM in this expr */
fac92740 141 regnode *emit_start; /* Start of emitted-code area */
f55b7b26 142 regnode_offset emit; /* Code-emit pointer */
830247a4
IZ
143 I32 naughty; /* How bad is this pattern? */
144 I32 sawback; /* Did we see \1, ...? */
145 U32 seen;
88f063b4
KW
146 SSize_t size; /* Number of regnode equivalents in
147 pattern */
7c932d07
KW
148
149 /* position beyond 'precomp' of the warning message furthest away from
150 * 'precomp'. During the parse, no warnings are raised for any problems
151 * earlier in the parse than this position. This works if warnings are
152 * raised the first time a given spot is parsed, and if only one
153 * independent warning is raised for any given spot */
154 Size_t latest_warn_offset;
155
156 I32 npar; /* Capture buffer count so far in the
157 parse, (OPEN) plus one. ("par" 0 is
158 the whole pattern)*/
159 I32 total_par; /* During initial parse, is either 0,
160 or -1; the latter indicating a
161 reparse is needed. After that pass,
162 it is what 'npar' became after the
163 pass. Hence, it being > 0 indicates
164 we are in a reparse situation */
538e84ed
KW
165 I32 nestroot; /* root parens we are in - used by
166 accept */
830247a4 167 I32 seen_zerolen;
f55b7b26
KW
168 regnode_offset *open_parens; /* offsets to open parens */
169 regnode_offset *close_parens; /* offsets to close parens */
d5a00e4a 170 regnode *end_op; /* END node in program */
02daf0ab
YO
171 I32 utf8; /* whether the pattern is utf8 or not */
172 I32 orig_utf8; /* whether the pattern was originally in utf8 */
173 /* XXX use this for future optimisation of case
174 * where pattern must be upgraded to utf8. */
e40e74fe
KW
175 I32 uni_semantics; /* If a d charset modifier should use unicode
176 rules, even if the pattern is not in
177 utf8 */
81714fb9 178 HV *paren_names; /* Paren names */
538e84ed 179
40d049e4 180 regnode **recurse; /* Recurse regops */
88f063b4 181 I32 recurse_count; /* Number of recurse regops we have generated */
4286711a 182 U8 *study_chunk_recursed; /* bitmap of which subs we have moved
538e84ed 183 through */
09a65838 184 U32 study_chunk_recursed_bytes; /* bytes in bitmap */
b57e4118 185 I32 in_lookbehind;
4624b182 186 I32 contains_locale;
bb3f3ed2 187 I32 override_recoding;
b6d67071
KW
188#ifdef EBCDIC
189 I32 recode_x_to_native;
190#endif
9d53c457 191 I32 in_multi_char_class;
1acab4c5 192 struct reg_code_blocks *code_blocks;/* positions of literal (?{})
68e2671b 193 within pattern */
b1603ef8 194 int code_index; /* next code_blocks[] slot */
ee273784 195 SSize_t maxlen; /* mininum possible number of chars in string to match */
3f910e62
YO
196 scan_frame *frame_head;
197 scan_frame *frame_last;
198 U32 frame_count;
7eec73eb 199 AV *warn_text;
dc6d7f5c 200#ifdef ADD_TO_REGEXEC
830247a4
IZ
201 char *starttry; /* -Dr: where regtry was called. */
202#define RExC_starttry (pRExC_state->starttry)
203#endif
d24ca0c5 204 SV *runtime_code_qr; /* qr with the runtime code blocks */
3dab1dad 205#ifdef DEBUGGING
be8e71aa 206 const char *lastparse;
3dab1dad 207 I32 lastnum;
1f1031fe 208 AV *paren_name_list; /* idx -> name */
d9a72fcc 209 U32 study_chunk_recursed_count;
c9f0d54c
YO
210 SV *mysv1;
211 SV *mysv2;
88f063b4 212
3dab1dad
YO
213#define RExC_lastparse (pRExC_state->lastparse)
214#define RExC_lastnum (pRExC_state->lastnum)
1f1031fe 215#define RExC_paren_name_list (pRExC_state->paren_name_list)
d9a72fcc 216#define RExC_study_chunk_recursed_count (pRExC_state->study_chunk_recursed_count)
c9f0d54c
YO
217#define RExC_mysv (pRExC_state->mysv1)
218#define RExC_mysv1 (pRExC_state->mysv1)
219#define RExC_mysv2 (pRExC_state->mysv2)
220
3dab1dad 221#endif
bd0b62a6 222 bool seen_d_op;
911bd04e 223 bool strict;
da7cf1cc 224 bool study_started;
034602eb 225 bool in_script_run;
7c932d07 226 bool use_BRANCHJ;
09b2b2e6 227};
830247a4 228
e2509266 229#define RExC_flags (pRExC_state->flags)
514a91f1 230#define RExC_pm_flags (pRExC_state->pm_flags)
830247a4 231#define RExC_precomp (pRExC_state->precomp)
51684e3c
KW
232#define RExC_copy_start_in_input (pRExC_state->copy_start_in_input)
233#define RExC_copy_start_in_constructed (pRExC_state->copy_start)
711b303b 234#define RExC_precomp_end (pRExC_state->precomp_end)
288b8c02 235#define RExC_rx_sv (pRExC_state->rx_sv)
830247a4 236#define RExC_rx (pRExC_state->rx)
f8fc2ecf 237#define RExC_rxi (pRExC_state->rxi)
fac92740 238#define RExC_start (pRExC_state->start)
830247a4
IZ
239#define RExC_end (pRExC_state->end)
240#define RExC_parse (pRExC_state->parse)
7c932d07 241#define RExC_latest_warn_offset (pRExC_state->latest_warn_offset )
830247a4 242#define RExC_whilem_seen (pRExC_state->whilem_seen)
bd0b62a6
KW
243#define RExC_seen_d_op (pRExC_state->seen_d_op) /* Seen something that differs
244 under /d from /u ? */
512c0f5a 245
512c0f5a 246
7122b237 247#ifdef RE_TRACK_PATTERN_OFFSETS
44f36cd3 248# define RExC_offsets (RExC_rxi->u.offsets) /* I am not like the
538e84ed 249 others */
7122b237 250#endif
830247a4 251#define RExC_emit (pRExC_state->emit)
fac92740 252#define RExC_emit_start (pRExC_state->emit_start)
830247a4
IZ
253#define RExC_sawback (pRExC_state->sawback)
254#define RExC_seen (pRExC_state->seen)
255#define RExC_size (pRExC_state->size)
ee273784 256#define RExC_maxlen (pRExC_state->maxlen)
830247a4 257#define RExC_npar (pRExC_state->npar)
1533d1d2 258#define RExC_total_parens (pRExC_state->total_par)
e2e6a0f1 259#define RExC_nestroot (pRExC_state->nestroot)
830247a4 260#define RExC_seen_zerolen (pRExC_state->seen_zerolen)
1aa99e6b 261#define RExC_utf8 (pRExC_state->utf8)
e40e74fe 262#define RExC_uni_semantics (pRExC_state->uni_semantics)
02daf0ab 263#define RExC_orig_utf8 (pRExC_state->orig_utf8)
40d049e4
YO
264#define RExC_open_parens (pRExC_state->open_parens)
265#define RExC_close_parens (pRExC_state->close_parens)
5bd2d46e 266#define RExC_end_op (pRExC_state->end_op)
81714fb9 267#define RExC_paren_names (pRExC_state->paren_names)
40d049e4
YO
268#define RExC_recurse (pRExC_state->recurse)
269#define RExC_recurse_count (pRExC_state->recurse_count)
09a65838 270#define RExC_study_chunk_recursed (pRExC_state->study_chunk_recursed)
538e84ed
KW
271#define RExC_study_chunk_recursed_bytes \
272 (pRExC_state->study_chunk_recursed_bytes)
b57e4118 273#define RExC_in_lookbehind (pRExC_state->in_lookbehind)
4624b182 274#define RExC_contains_locale (pRExC_state->contains_locale)
b6d67071
KW
275#ifdef EBCDIC
276# define RExC_recode_x_to_native (pRExC_state->recode_x_to_native)
277#endif
9d53c457 278#define RExC_in_multi_char_class (pRExC_state->in_multi_char_class)
3f910e62
YO
279#define RExC_frame_head (pRExC_state->frame_head)
280#define RExC_frame_last (pRExC_state->frame_last)
281#define RExC_frame_count (pRExC_state->frame_count)
67cdf558 282#define RExC_strict (pRExC_state->strict)
da7cf1cc 283#define RExC_study_started (pRExC_state->study_started)
7eec73eb 284#define RExC_warn_text (pRExC_state->warn_text)
034602eb 285#define RExC_in_script_run (pRExC_state->in_script_run)
7c932d07 286#define RExC_use_BRANCHJ (pRExC_state->use_BRANCHJ)
830247a4 287
99807a43
HS
288/* Heuristic check on the complexity of the pattern: if TOO_NAUGHTY, we set
289 * a flag to disable back-off on the fixed/floating substrings - if it's
290 * a high complexity pattern we assume the benefit of avoiding a full match
291 * is worth the cost of checking for the substrings even if they rarely help.
292 */
293#define RExC_naughty (pRExC_state->naughty)
294#define TOO_NAUGHTY (10)
295#define MARK_NAUGHTY(add) \
296 if (RExC_naughty < TOO_NAUGHTY) \
297 RExC_naughty += (add)
298#define MARK_NAUGHTY_EXP(exp, add) \
299 if (RExC_naughty < TOO_NAUGHTY) \
300 RExC_naughty += RExC_naughty / (exp) + (add)
cde0cee5 301
a687059c
LW
302#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
303#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
412f55bb 304 ((*s) == '{' && regcurly(s)))
a687059c
LW
305
306/*
307 * Flags to be passed up and down.
308 */
a687059c 309#define WORST 0 /* Worst case. */
c4ead770
KW
310#define HASWIDTH 0x01 /* Known to not match null strings, could match
311 non-null ones. */
fda99bee 312
e64f369d 313/* Simple enough to be STAR/PLUS operand; in an EXACTish node must be a single
2fd92675
KW
314 * character. (There needs to be a case: in the switch statement in regexec.c
315 * for any node marked SIMPLE.) Note that this is not the same thing as
316 * REGNODE_SIMPLE */
fda99bee 317#define SIMPLE 0x02
e64f369d 318#define SPSTART 0x04 /* Starts with * or + */
8d9c2815
NC
319#define POSTPONED 0x08 /* (?1),(?&name), (??{...}) or similar */
320#define TRYAGAIN 0x10 /* Weeded out a declaration. */
c36361b2
KW
321#define RESTART_PARSE 0x20 /* Need to redo the parse */
322#define NEED_UTF8 0x40 /* In conjunction with RESTART_PARSE, need to
323 calcuate sizes as UTF-8 */
a687059c 324
3dab1dad
YO
325#define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1)
326
07be1b83
YO
327/* whether trie related optimizations are enabled */
328#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
329#define TRIE_STUDY_OPT
786e8c11 330#define FULL_TRIE_STUDY
07be1b83
YO
331#define TRIE_STCLASS
332#endif
1de06328
YO
333
334
40d049e4
YO
335
336#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
337#define PBITVAL(paren) (1 << ((paren) & 7))
338#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
339#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
340#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
341
82a6ada4 342#define REQUIRE_UTF8(flagp) STMT_START { \
8d9c2815 343 if (!UTF) { \
c36361b2 344 *flagp = RESTART_PARSE|NEED_UTF8; \
f55b7b26 345 return 0; \
8d9c2815 346 } \
82a6ada4 347 } STMT_END
40d049e4 348
d63d75f5
KW
349/* Change from /d into /u rules, and restart the parse. RExC_uni_semantics is
350 * a flag that indicates we've changed to /u during the parse. */
512c0f5a
KW
351#define REQUIRE_UNI_RULES(flagp, restart_retval) \
352 STMT_START { \
353 if (DEPENDS_SEMANTICS) { \
512c0f5a
KW
354 set_regex_charset(&RExC_flags, REGEX_UNICODE_CHARSET); \
355 RExC_uni_semantics = 1; \
bd0b62a6
KW
356 if (RExC_seen_d_op && LIKELY(RExC_total_parens >= 0)) { \
357 /* No need to restart the parse if we haven't seen \
358 * anything that differs between /u and /d, and no need \
359 * to restart immediately if we're going to reparse \
360 * anyway to count parens */ \
c36361b2 361 *flagp |= RESTART_PARSE; \
512c0f5a 362 return restart_retval; \
d1be211c 363 } \
512c0f5a
KW
364 } \
365 } STMT_END
366
7c932d07
KW
367#define BRANCH_MAX_OFFSET U16_MAX
368#define REQUIRE_BRANCHJ(flagp, restart_retval) \
369 STMT_START { \
370 RExC_use_BRANCHJ = 1; \
523a2924
KW
371 if (LIKELY(RExC_total_parens >= 0)) { \
372 /* No need to restart the parse immediately if we're \
373 * going to reparse anyway to count parens */ \
374 *flagp |= RESTART_PARSE; \
375 return restart_retval; \
376 } \
7c932d07
KW
377 } STMT_END
378
379#define REQUIRE_PARENS_PASS \
380 STMT_START { \
381 if (RExC_total_parens == 0) RExC_total_parens = -1; \
382 } STMT_END
383
f358228e
KW
384/* This is used to return failure (zero) early from the calling function if
385 * various flags in 'flags' are set. Two flags always cause a return:
386 * 'RESTART_PARSE' and 'NEED_UTF8'. 'extra' can be used to specify any
387 * additional flags that should cause a return; 0 if none. If the return will
388 * be done, '*flagp' is first set to be all of the flags that caused the
389 * return. */
390#define RETURN_FAIL_ON_RESTART_OR_FLAGS(flags,flagp,extra) \
fd7fc6ae 391 STMT_START { \
c36361b2
KW
392 if ((flags) & (RESTART_PARSE|NEED_UTF8|(extra))) { \
393 *(flagp) = (flags) & (RESTART_PARSE|NEED_UTF8|(extra)); \
f358228e 394 return 0; \
fd7fc6ae
YO
395 } \
396 } STMT_END
397
c36361b2 398#define MUST_RESTART(flags) ((flags) & (RESTART_PARSE))
fd7fc6ae 399
5d555bb8 400#define RETURN_FAIL_ON_RESTART(flags,flagp) \
f358228e 401 RETURN_FAIL_ON_RESTART_OR_FLAGS( flags, flagp, 0)
5d555bb8 402#define RETURN_FAIL_ON_RESTART_FLAGP(flagp) \
f358228e 403 if (MUST_RESTART(*(flagp))) return 0
fd7fc6ae 404
f19b1a63
KW
405/* This converts the named class defined in regcomp.h to its equivalent class
406 * number defined in handy.h. */
407#define namedclass_to_classnum(class) ((int) ((class) / 2))
408#define classnum_to_namedclass(classnum) ((classnum) * 2)
409
de92f5e6
KW
410#define _invlist_union_complement_2nd(a, b, output) \
411 _invlist_union_maybe_complement_2nd(a, b, TRUE, output)
412#define _invlist_intersection_complement_2nd(a, b, output) \
413 _invlist_intersection_maybe_complement_2nd(a, b, TRUE, output)
414
1de06328
YO
415/* About scan_data_t.
416
417 During optimisation we recurse through the regexp program performing
418 various inplace (keyhole style) optimisations. In addition study_chunk
419 and scan_commit populate this data structure with information about
538e84ed 420 what strings MUST appear in the pattern. We look for the longest
3b753521 421 string that must appear at a fixed location, and we look for the
1de06328
YO
422 longest string that may appear at a floating location. So for instance
423 in the pattern:
538e84ed 424
1de06328 425 /FOO[xX]A.*B[xX]BAR/
538e84ed 426
1de06328
YO
427 Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
428 strings (because they follow a .* construct). study_chunk will identify
429 both FOO and BAR as being the longest fixed and floating strings respectively.
538e84ed 430
1de06328 431 The strings can be composites, for instance
538e84ed 432
1de06328 433 /(f)(o)(o)/
538e84ed 434
1de06328 435 will result in a composite fixed substring 'foo'.
538e84ed 436
1de06328 437 For each string some basic information is maintained:
538e84ed 438
6ea77169 439 - min_offset
1de06328
YO
440 This is the position the string must appear at, or not before.
441 It also implicitly (when combined with minlenp) tells us how many
3b753521
FN
442 characters must match before the string we are searching for.
443 Likewise when combined with minlenp and the length of the string it
538e84ed 444 tells us how many characters must appear after the string we have
1de06328 445 found.
538e84ed 446
1de06328
YO
447 - max_offset
448 Only used for floating strings. This is the rightmost point that
ea3daa5d 449 the string can appear at. If set to SSize_t_MAX it indicates that the
1de06328 450 string can occur infinitely far to the right.
6e12f832 451 For fixed strings, it is equal to min_offset.
538e84ed 452
1de06328 453 - minlenp
2d608413
KW
454 A pointer to the minimum number of characters of the pattern that the
455 string was found inside. This is important as in the case of positive
538e84ed 456 lookahead or positive lookbehind we can have multiple patterns
1de06328 457 involved. Consider
538e84ed 458
1de06328 459 /(?=FOO).*F/
538e84ed 460
1de06328
YO
461 The minimum length of the pattern overall is 3, the minimum length
462 of the lookahead part is 3, but the minimum length of the part that
538e84ed 463 will actually match is 1. So 'FOO's minimum length is 3, but the
1de06328 464 minimum length for the F is 1. This is important as the minimum length
538e84ed 465 is used to determine offsets in front of and behind the string being
1de06328 466 looked for. Since strings can be composites this is the length of the
486ec47a 467 pattern at the time it was committed with a scan_commit. Note that
1de06328 468 the length is calculated by study_chunk, so that the minimum lengths
538e84ed 469 are not known until the full pattern has been compiled, thus the
1de06328 470 pointer to the value.
538e84ed 471
1de06328 472 - lookbehind
538e84ed 473
1de06328 474 In the case of lookbehind the string being searched for can be
538e84ed 475 offset past the start point of the final matching string.
1de06328
YO
476 If this value was just blithely removed from the min_offset it would
477 invalidate some of the calculations for how many chars must match
478 before or after (as they are derived from min_offset and minlen and
538e84ed 479 the length of the string being searched for).
1de06328
YO
480 When the final pattern is compiled and the data is moved from the
481 scan_data_t structure into the regexp structure the information
538e84ed
KW
482 about lookbehind is factored in, with the information that would
483 have been lost precalculated in the end_shift field for the
1de06328
YO
484 associated string.
485
486 The fields pos_min and pos_delta are used to store the minimum offset
538e84ed 487 and the delta to the maximum offset at the current point in the pattern.
1de06328
YO
488
489*/
2c2d71f5 490
cb6aef8b
DM
491struct scan_data_substrs {
492 SV *str; /* longest substring found in pattern */
493 SSize_t min_offset; /* earliest point in string it can appear */
494 SSize_t max_offset; /* latest point in string it can appear */
495 SSize_t *minlenp; /* pointer to the minlen relevant to the string */
496 SSize_t lookbehind; /* is the pos of the string modified by LB */
11683ecb 497 I32 flags; /* per substring SF_* and SCF_* flags */
cb6aef8b
DM
498};
499
2c2d71f5 500typedef struct scan_data_t {
1de06328
YO
501 /*I32 len_min; unused */
502 /*I32 len_delta; unused */
49f55535 503 SSize_t pos_min;
ea3daa5d 504 SSize_t pos_delta;
2c2d71f5 505 SV *last_found;
ea3daa5d 506 SSize_t last_end; /* min value, <0 unless valid. */
49f55535 507 SSize_t last_start_min;
ea3daa5d 508 SSize_t last_start_max;
2099df82
DM
509 U8 cur_is_floating; /* whether the last_* values should be set as
510 * the next fixed (0) or floating (1)
511 * substring */
6ea77169 512
2099df82
DM
513 /* [0] is longest fixed substring so far, [1] is longest float so far */
514 struct scan_data_substrs substrs[2];
6ea77169 515
11683ecb 516 I32 flags; /* common SF_* and SCF_* flags */
2c2d71f5 517 I32 whilem_c;
49f55535 518 SSize_t *last_closep;
b8f7bb16 519 regnode_ssc *start_class;
2c2d71f5
JH
520} scan_data_t;
521
a687059c 522/*
e50aee73 523 * Forward declarations for pregcomp()'s friends.
a687059c 524 */
a0d0e21e 525
6ea77169 526static const scan_data_t zero_scan_data = {
a2ca2017 527 0, 0, NULL, 0, 0, 0, 0,
6ea77169 528 {
11683ecb
DM
529 { NULL, 0, 0, 0, 0, 0 },
530 { NULL, 0, 0, 0, 0, 0 },
6ea77169
DM
531 },
532 0, 0, NULL, NULL
533};
c277df42 534
11683ecb
DM
535/* study flags */
536
07be1b83
YO
537#define SF_BEFORE_SEOL 0x0001
538#define SF_BEFORE_MEOL 0x0002
11683ecb 539#define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL)
c277df42 540
07be1b83
YO
541#define SF_IS_INF 0x0040
542#define SF_HAS_PAR 0x0080
543#define SF_IN_PAR 0x0100
544#define SF_HAS_EVAL 0x0200
bc604ad8
KW
545
546
547/* SCF_DO_SUBSTR is the flag that tells the regexp analyzer to track the
548 * longest substring in the pattern. When it is not set the optimiser keeps
549 * track of position, but does not keep track of the actual strings seen,
550 *
551 * So for instance /foo/ will be parsed with SCF_DO_SUBSTR being true, but
552 * /foo/i will not.
553 *
554 * Similarly, /foo.*(blah|erm|huh).*fnorble/ will have "foo" and "fnorble"
555 * parsed with SCF_DO_SUBSTR on, but while processing the (...) it will be
556 * turned off because of the alternation (BRANCH). */
07be1b83 557#define SCF_DO_SUBSTR 0x0400
bc604ad8 558
653099ff
GS
559#define SCF_DO_STCLASS_AND 0x0800
560#define SCF_DO_STCLASS_OR 0x1000
561#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
e1901655 562#define SCF_WHILEM_VISITED_POS 0x2000
c277df42 563
786e8c11 564#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
538e84ed 565#define SCF_SEEN_ACCEPT 0x8000
688e0391 566#define SCF_TRIE_DOING_RESTUDY 0x10000
4286711a
YO
567#define SCF_IN_DEFINE 0x20000
568
569
570
07be1b83 571
43fead97 572#define UTF cBOOL(RExC_utf8)
00b27cfc
KW
573
574/* The enums for all these are ordered so things work out correctly */
a62b1201 575#define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET)
538e84ed
KW
576#define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) \
577 == REGEX_DEPENDS_CHARSET)
1a26cbcb 578/* Use RExC_uni_semantics instead of this:
00b27cfc 579#define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET)
1a26cbcb 580*/
538e84ed
KW
581#define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) \
582 >= REGEX_UNICODE_CHARSET)
583#define ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
584 == REGEX_ASCII_RESTRICTED_CHARSET)
585#define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) \
586 >= REGEX_ASCII_RESTRICTED_CHARSET)
587#define ASCII_FOLD_RESTRICTED (get_regex_charset(RExC_flags) \
588 == REGEX_ASCII_MORE_RESTRICTED_CHARSET)
a62b1201 589
43fead97 590#define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD)
a0ed51b3 591
f2c2a6ab
KW
592/* For programs that want to be strictly Unicode compatible by dying if any
593 * attempt is made to match a non-Unicode code point against a Unicode
594 * property. */
595#define ALWAYS_WARN_SUPER ckDEAD(packWARN(WARN_NON_UNICODE))
596
93733859 597#define OOB_NAMEDCLASS -1
b8c5462f 598
8e661ac5
KW
599/* There is no code point that is out-of-bounds, so this is problematic. But
600 * its only current use is to initialize a variable that is always set before
601 * looked at. */
602#define OOB_UNICODE 0xDEADBEEF
603
a0ed51b3 604#define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv))
a0ed51b3 605
8615cb43 606
b45f050a
JF
607/* length of regex to show in messages that don't mark a position within */
608#define RegexLengthToShowInErrorMessages 127
609
610/*
611 * If MARKER[12] are adjusted, be sure to adjust the constants at the top
612 * of t/op/regmesg.t, the tests in t/op/re_tests, and those in
613 * op/pragma/warn/regcomp.
614 */
7253e4e3
RK
615#define MARKER1 "<-- HERE" /* marker as it appears in the description */
616#define MARKER2 " <-- HERE " /* marker as it appears within the regex */
b81d288d 617
538e84ed 618#define REPORT_LOCATION " in regex; marked by " MARKER1 \
147e3846 619 " in m/%" UTF8f MARKER2 "%" UTF8f "/"
b45f050a 620
285b5ca0
KW
621/* The code in this file in places uses one level of recursion with parsing
622 * rebased to an alternate string constructed by us in memory. This can take
623 * the form of something that is completely different from the input, or
624 * something that uses the input as part of the alternate. In the first case,
625 * there should be no possibility of an error, as we are in complete control of
51684e3c
KW
626 * the alternate string. But in the second case we don't completely control
627 * the input portion, so there may be errors in that. Here's an example:
285b5ca0
KW
628 * /[abc\x{DF}def]/ui
629 * is handled specially because \x{df} folds to a sequence of more than one
51684e3c 630 * character: 'ss'. What is done is to create and parse an alternate string,
285b5ca0
KW
631 * which looks like this:
632 * /(?:\x{DF}|[abc\x{DF}def])/ui
633 * where it uses the input unchanged in the middle of something it constructs,
634 * which is a branch for the DF outside the character class, and clustering
635 * parens around the whole thing. (It knows enough to skip the DF inside the
636 * class while in this substitute parse.) 'abc' and 'def' may have errors that
637 * need to be reported. The general situation looks like this:
638 *
51684e3c 639 * |<------- identical ------>|
285b5ca0 640 * sI tI xI eI
51684e3c 641 * Input: ---------------------------------------------------------------
285b5ca0
KW
642 * Constructed: ---------------------------------------------------
643 * sC tC xC eC EC
51684e3c 644 * |<------- identical ------>|
285b5ca0 645 *
51684e3c
KW
646 * sI..eI is the portion of the input pattern we are concerned with here.
647 * sC..EC is the constructed substitute parse string.
648 * sC..tC is constructed by us
649 * tC..eC is an exact duplicate of the portion of the input pattern tI..eI.
650 * In the diagram, these are vertically aligned.
651 * eC..EC is also constructed by us.
652 * xC is the position in the substitute parse string where we found a
653 * problem.
654 * xI is the position in the original pattern corresponding to xC.
285b5ca0 655 *
51684e3c
KW
656 * We want to display a message showing the real input string. Thus we need to
657 * translate from xC to xI. We know that xC >= tC, since the portion of the
658 * string sC..tC has been constructed by us, and so shouldn't have errors. We
659 * get:
660 * xI = tI + (xC - tC)
285b5ca0 661 *
51684e3c
KW
662 * When the substitute parse is constructed, the code needs to set:
663 * RExC_start (sC)
664 * RExC_end (eC)
665 * RExC_copy_start_in_input (tI)
666 * RExC_copy_start_in_constructed (tC)
667 * and restore them when done.
285b5ca0 668 *
51684e3c
KW
669 * During normal processing of the input pattern, both
670 * 'RExC_copy_start_in_input' and 'RExC_copy_start_in_constructed' are set to
671 * sI, so that xC equals xI.
285b5ca0
KW
672 */
673
51684e3c
KW
674#define sI RExC_precomp
675#define eI RExC_precomp_end
676#define sC RExC_start
677#define eC RExC_end
678#define tI RExC_copy_start_in_input
679#define tC RExC_copy_start_in_constructed
680#define xI(xC) (tI + (xC - tC))
681#define xI_offset(xC) (xI(xC) - sI)
285b5ca0
KW
682
683#define REPORT_LOCATION_ARGS(xC) \
684 UTF8fARG(UTF, \
51684e3c 685 (xI(xC) > eI) /* Don't run off end */ \
232b691f 686 ? eI - sI /* Length before the <--HERE */ \
bc929b61
KW
687 : ((xI_offset(xC) >= 0) \
688 ? xI_offset(xC) \
689 : (Perl_croak(aTHX_ "panic: %s: %d: negative offset: %" \
690 IVdf " trying to output message for " \
691 " pattern %.*s", \
51684e3c 692 __FILE__, __LINE__, (IV) xI_offset(xC), \
bc929b61 693 ((int) (eC - sC)), sC), 0)), \
51684e3c 694 sI), /* The input pattern printed up to the <--HERE */ \
285b5ca0 695 UTF8fARG(UTF, \
51684e3c
KW
696 (xI(xC) > eI) ? 0 : eI - xI(xC), /* Length after <--HERE */ \
697 (xI(xC) > eI) ? eI : xI(xC)) /* pattern after <--HERE */
c1d900c3 698
8a6d8ec6
HS
699/* Used to point after bad bytes for an error message, but avoid skipping
700 * past a nul byte. */
701#define SKIP_IF_CHAR(s) (!*(s) ? 0 : UTF ? UTF8SKIP(s) : 1)
702
ed68fe11
KW
703/* Set up to clean up after our imminent demise */
704#define PREPARE_TO_DIE \
705 STMT_START { \
706 if (RExC_rx_sv) \
707 SAVEFREESV(RExC_rx_sv); \
7c932d07
KW
708 if (RExC_open_parens) \
709 SAVEFREEPV(RExC_open_parens); \
710 if (RExC_close_parens) \
711 SAVEFREEPV(RExC_close_parens); \
ed68fe11
KW
712 } STMT_END
713
b45f050a
JF
714/*
715 * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given
716 * arg. Show regex, up to a maximum length. If it's too long, chop and add
717 * "...".
718 */
58e23c8d 719#define _FAIL(code) STMT_START { \
bfed75c6 720 const char *ellipses = ""; \
88f063b4 721 IV len = RExC_precomp_end - RExC_precomp; \
ccb2c380 722 \
ed68fe11 723 PREPARE_TO_DIE; \
ccb2c380
MP
724 if (len > RegexLengthToShowInErrorMessages) { \
725 /* chop 10 shorter than the max, to ensure meaning of "..." */ \
726 len = RegexLengthToShowInErrorMessages - 10; \
727 ellipses = "..."; \
728 } \
58e23c8d 729 code; \
ccb2c380 730} STMT_END
8615cb43 731
58e23c8d 732#define FAIL(msg) _FAIL( \
147e3846 733 Perl_croak(aTHX_ "%s in regex m/%" UTF8f "%s/", \
c1d900c3 734 msg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d
YO
735
736#define FAIL2(msg,arg) _FAIL( \
147e3846 737 Perl_croak(aTHX_ msg " in regex m/%" UTF8f "%s/", \
c1d900c3 738 arg, UTF8fARG(UTF, len, RExC_precomp), ellipses))
58e23c8d 739
b45f050a 740/*
b45f050a
JF
741 * Simple_vFAIL -- like FAIL, but marks the current location in the scan
742 */
ccb2c380 743#define Simple_vFAIL(m) STMT_START { \
ccb2c380 744 Perl_croak(aTHX_ "%s" REPORT_LOCATION, \
d528642a 745 m, REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 746} STMT_END
b45f050a
JF
747
748/*
749 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL()
750 */
ccb2c380 751#define vFAIL(m) STMT_START { \
ed68fe11 752 PREPARE_TO_DIE; \
ccb2c380
MP
753 Simple_vFAIL(m); \
754} STMT_END
b45f050a
JF
755
756/*
757 * Like Simple_vFAIL(), but accepts two arguments.
758 */
ccb2c380 759#define Simple_vFAIL2(m,a1) STMT_START { \
d528642a
KW
760 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
761 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 762} STMT_END
b45f050a
JF
763
764/*
765 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2().
766 */
ccb2c380 767#define vFAIL2(m,a1) STMT_START { \
ed68fe11 768 PREPARE_TO_DIE; \
ccb2c380
MP
769 Simple_vFAIL2(m, a1); \
770} STMT_END
b45f050a
JF
771
772
773/*
774 * Like Simple_vFAIL(), but accepts three arguments.
775 */
ccb2c380 776#define Simple_vFAIL3(m, a1, a2) STMT_START { \
c1d900c3 777 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 778 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 779} STMT_END
b45f050a
JF
780
781/*
782 * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3().
783 */
ccb2c380 784#define vFAIL3(m,a1,a2) STMT_START { \
ed68fe11 785 PREPARE_TO_DIE; \
ccb2c380
MP
786 Simple_vFAIL3(m, a1, a2); \
787} STMT_END
b45f050a
JF
788
789/*
790 * Like Simple_vFAIL(), but accepts four arguments.
791 */
ccb2c380 792#define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \
d528642a
KW
793 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, a3, \
794 REPORT_LOCATION_ARGS(RExC_parse)); \
ccb2c380 795} STMT_END
b45f050a 796
95db3ffa 797#define vFAIL4(m,a1,a2,a3) STMT_START { \
ed68fe11 798 PREPARE_TO_DIE; \
95db3ffa
KW
799 Simple_vFAIL4(m, a1, a2, a3); \
800} STMT_END
801
946095af 802/* A specialized version of vFAIL2 that works with UTF8f */
d528642a 803#define vFAIL2utf8f(m, a1) STMT_START { \
ed68fe11 804 PREPARE_TO_DIE; \
d528642a
KW
805 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, \
806 REPORT_LOCATION_ARGS(RExC_parse)); \
946095af
BF
807} STMT_END
808
3ba22297 809#define vFAIL3utf8f(m, a1, a2) STMT_START { \
ed68fe11 810 PREPARE_TO_DIE; \
3ba22297 811 S_re_croak2(aTHX_ UTF, m, REPORT_LOCATION, a1, a2, \
d528642a 812 REPORT_LOCATION_ARGS(RExC_parse)); \
3ba22297
KW
813} STMT_END
814
d62e749e
KW
815/* Setting this to NULL is a signal to not output warnings */
816#define TURN_OFF_WARNINGS_IN_SUBSTITUTE_PARSE RExC_copy_start_in_constructed = NULL
817#define RESTORE_WARNINGS RExC_copy_start_in_constructed = RExC_precomp
818
7c932d07
KW
819/* Since a warning can be generated multiple times as the input is reparsed, we
820 * output it the first time we come to that point in the parse, but suppress it
821 * otherwise. 'RExC_copy_start_in_constructed' being NULL is a flag to not
822 * generate any warnings */
55bcf857 823#define TO_OUTPUT_WARNINGS(loc) \
7c932d07
KW
824 ( RExC_copy_start_in_constructed \
825 && ((xI(loc)) - RExC_precomp) > (Ptrdiff_t) RExC_latest_warn_offset)
55bcf857 826
7c932d07
KW
827/* After we've emitted a warning, we save the position in the input so we don't
828 * output it again */
829#define UPDATE_WARNINGS_LOC(loc) \
830 STMT_START { \
831 if (TO_OUTPUT_WARNINGS(loc)) { \
832 RExC_latest_warn_offset = (xI(loc)) - RExC_precomp; \
833 } \
834 } STMT_END
77ccba3c 835
086f21bf 836/* 'warns' is the output of the packWARNx macro used in 'code' */
f8220fe2
KW
837#define _WARN_HELPER(loc, warns, code) \
838 STMT_START { \
d62e749e
KW
839 if (! RExC_copy_start_in_constructed) { \
840 Perl_croak( aTHX_ "panic! %s: %d: Tried to warn when none" \
841 " expected at '%s'", \
842 __FILE__, __LINE__, loc); \
843 } \
55bcf857 844 if (TO_OUTPUT_WARNINGS(loc)) { \
086f21bf
KW
845 if (ckDEAD(warns)) \
846 PREPARE_TO_DIE; \
338190a1 847 code; \
77ccba3c 848 UPDATE_WARNINGS_LOC(loc); \
338190a1 849 } \
f8220fe2 850 } STMT_END
946095af 851
5e0a247b 852/* m is not necessarily a "literal string", in this macro */
f8220fe2
KW
853#define reg_warn_non_literal_string(loc, m) \
854 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
855 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 856 "%s" REPORT_LOCATION, \
f8220fe2 857 m, REPORT_LOCATION_ARGS(loc)))
5e0a247b 858
f8220fe2
KW
859#define ckWARNreg(loc,m) \
860 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
861 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 862 m REPORT_LOCATION, \
f8220fe2 863 REPORT_LOCATION_ARGS(loc)))
ccb2c380 864
f8220fe2
KW
865#define vWARN(loc, m) \
866 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
867 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 868 m REPORT_LOCATION, \
f8220fe2 869 REPORT_LOCATION_ARGS(loc))) \
b927b7e9 870
f8220fe2
KW
871#define vWARN_dep(loc, m) \
872 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
873 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 874 m REPORT_LOCATION, \
f8220fe2 875 REPORT_LOCATION_ARGS(loc)))
0d6106aa 876
f8220fe2
KW
877#define ckWARNdep(loc,m) \
878 _WARN_HELPER(loc, packWARN(WARN_DEPRECATED), \
879 Perl_ck_warner_d(aTHX_ packWARN(WARN_DEPRECATED), \
d528642a 880 m REPORT_LOCATION, \
f8220fe2 881 REPORT_LOCATION_ARGS(loc)))
147508a2 882
f8220fe2
KW
883#define ckWARNregdep(loc,m) \
884 _WARN_HELPER(loc, packWARN2(WARN_DEPRECATED, WARN_REGEXP), \
885 Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, \
d528642a
KW
886 WARN_REGEXP), \
887 m REPORT_LOCATION, \
f8220fe2 888 REPORT_LOCATION_ARGS(loc)))
ccb2c380 889
f8220fe2
KW
890#define ckWARN2reg_d(loc,m, a1) \
891 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
892 Perl_ck_warner_d(aTHX_ packWARN(WARN_REGEXP), \
d528642a 893 m REPORT_LOCATION, \
f8220fe2 894 a1, REPORT_LOCATION_ARGS(loc)))
2335b3d3 895
f8220fe2
KW
896#define ckWARN2reg(loc, m, a1) \
897 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
898 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 899 m REPORT_LOCATION, \
f8220fe2 900 a1, REPORT_LOCATION_ARGS(loc)))
ccb2c380 901
f8220fe2
KW
902#define vWARN3(loc, m, a1, a2) \
903 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
904 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a 905 m REPORT_LOCATION, \
f8220fe2 906 a1, a2, REPORT_LOCATION_ARGS(loc)))
ccb2c380 907
f8220fe2
KW
908#define ckWARN3reg(loc, m, a1, a2) \
909 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
910 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
911 m REPORT_LOCATION, \
912 a1, a2, \
f8220fe2 913 REPORT_LOCATION_ARGS(loc)))
668c081a 914
f8220fe2
KW
915#define vWARN4(loc, m, a1, a2, a3) \
916 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
917 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
918 m REPORT_LOCATION, \
919 a1, a2, a3, \
f8220fe2 920 REPORT_LOCATION_ARGS(loc)))
ccb2c380 921
f8220fe2
KW
922#define ckWARN4reg(loc, m, a1, a2, a3) \
923 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
924 Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
925 m REPORT_LOCATION, \
926 a1, a2, a3, \
f8220fe2 927 REPORT_LOCATION_ARGS(loc)))
668c081a 928
f8220fe2
KW
929#define vWARN5(loc, m, a1, a2, a3, a4) \
930 _WARN_HELPER(loc, packWARN(WARN_REGEXP), \
931 Perl_warner(aTHX_ packWARN(WARN_REGEXP), \
d528642a
KW
932 m REPORT_LOCATION, \
933 a1, a2, a3, a4, \
f8220fe2 934 REPORT_LOCATION_ARGS(loc)))
9d1d55b5 935
aea5f181
KW
936#define ckWARNexperimental(loc, class, m) \
937 _WARN_HELPER(loc, packWARN(class), \
938 Perl_ck_warner_d(aTHX_ packWARN(class), \
939 m REPORT_LOCATION, \
940 REPORT_LOCATION_ARGS(loc)))
941
1a322bb4
KW
942/* Convert between a pointer to a node and its offset from the beginning of the
943 * program */
944#define REGNODE_p(offset) (RExC_emit_start + (offset))
945#define REGNODE_OFFSET(node) ((node) - RExC_emit_start)
946
538e84ed 947/* Macros for recording node offsets. 20001227 mjd@plover.com
fac92740
MJD
948 * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in
949 * element 2*n-1 of the array. Element #2n holds the byte length node #n.
950 * Element 0 holds the number n.
07be1b83 951 * Position is 1 indexed.
fac92740 952 */
7122b237 953#ifndef RE_TRACK_PATTERN_OFFSETS
d7c442ce 954#define Set_Node_Offset_To_R(offset,byte)
7122b237
YO
955#define Set_Node_Offset(node,byte)
956#define Set_Cur_Node_Offset
957#define Set_Node_Length_To_R(node,len)
958#define Set_Node_Length(node,len)
6a86c6ad 959#define Set_Node_Cur_Length(node,start)
538e84ed
KW
960#define Node_Offset(n)
961#define Node_Length(n)
7122b237
YO
962#define Set_Node_Offset_Length(node,offset,len)
963#define ProgLen(ri) ri->u.proglen
964#define SetProgLen(ri,x) ri->u.proglen = x
4cbeff97 965#define Track_Code(code)
7122b237
YO
966#else
967#define ProgLen(ri) ri->u.offsets[0]
968#define SetProgLen(ri,x) ri->u.offsets[0] = x
d7c442ce 969#define Set_Node_Offset_To_R(offset,byte) STMT_START { \
ccb2c380 970 MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \
d7c442ce
KW
971 __LINE__, (int)(offset), (int)(byte))); \
972 if((offset) < 0) { \
538e84ed 973 Perl_croak(aTHX_ "value of node is %d in Offset macro", \
d7c442ce 974 (int)(offset)); \
ccb2c380 975 } else { \
d7c442ce 976 RExC_offsets[2*(offset)-1] = (byte); \
ccb2c380 977 } \
ccb2c380
MP
978} STMT_END
979
88f063b4 980#define Set_Node_Offset(node,byte) \
1a322bb4 981 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (byte)-RExC_start)
ccb2c380
MP
982#define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse)
983
984#define Set_Node_Length_To_R(node,len) STMT_START { \
ccb2c380 985 MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \
551405c4 986 __LINE__, (int)(node), (int)(len))); \
ccb2c380 987 if((node) < 0) { \
538e84ed
KW
988 Perl_croak(aTHX_ "value of node is %d in Length macro", \
989 (int)(node)); \
ccb2c380
MP
990 } else { \
991 RExC_offsets[2*(node)] = (len); \
992 } \
ccb2c380
MP
993} STMT_END
994
995#define Set_Node_Length(node,len) \
1a322bb4 996 Set_Node_Length_To_R(REGNODE_OFFSET(node), len)
6a86c6ad
NC
997#define Set_Node_Cur_Length(node, start) \
998 Set_Node_Length(node, RExC_parse - start)
fac92740
MJD
999
1000/* Get offsets and lengths */
1a322bb4
KW
1001#define Node_Offset(n) (RExC_offsets[2*(REGNODE_OFFSET(n))-1])
1002#define Node_Length(n) (RExC_offsets[2*(REGNODE_OFFSET(n))])
fac92740 1003
07be1b83 1004#define Set_Node_Offset_Length(node,offset,len) STMT_START { \
1a322bb4
KW
1005 Set_Node_Offset_To_R(REGNODE_OFFSET(node), (offset)); \
1006 Set_Node_Length_To_R(REGNODE_OFFSET(node), (len)); \
07be1b83 1007} STMT_END
4cbeff97
TC
1008
1009#define Track_Code(code) STMT_START { code } STMT_END
7122b237 1010#endif
07be1b83
YO
1011
1012#if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS
1013#define EXPERIMENTAL_INPLACESCAN
f427392e 1014#endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/
07be1b83 1015
cb41e5d6
YO
1016#ifdef DEBUGGING
1017int
6ad9a8ab 1018Perl_re_printf(pTHX_ const char *fmt, ...)
cb41e5d6
YO
1019{
1020 va_list ap;
1021 int result;
1022 PerlIO *f= Perl_debug_log;
1023 PERL_ARGS_ASSERT_RE_PRINTF;
1024 va_start(ap, fmt);
1025 result = PerlIO_vprintf(f, fmt, ap);
1026 va_end(ap);
1027 return result;
1028}
1029
1030int
7b031478 1031Perl_re_indentf(pTHX_ const char *fmt, U32 depth, ...)
cb41e5d6
YO
1032{
1033 va_list ap;
1034 int result;
1035 PerlIO *f= Perl_debug_log;
1036 PERL_ARGS_ASSERT_RE_INDENTF;
1037 va_start(ap, depth);
daeb874b 1038 PerlIO_printf(f, "%*s", ( (int)depth % 20 ) * 2, "");
cb41e5d6
YO
1039 result = PerlIO_vprintf(f, fmt, ap);
1040 va_end(ap);
1041 return result;
1042}
1043#endif /* DEBUGGING */
1044
7b031478 1045#define DEBUG_RExC_seen() \
538e84ed 1046 DEBUG_OPTIMISE_MORE_r({ \
88f063b4 1047 Perl_re_printf( aTHX_ "RExC_seen: "); \
538e84ed 1048 \
e384d5c1 1049 if (RExC_seen & REG_ZERO_LEN_SEEN) \
88f063b4 1050 Perl_re_printf( aTHX_ "REG_ZERO_LEN_SEEN "); \
538e84ed 1051 \
e384d5c1 1052 if (RExC_seen & REG_LOOKBEHIND_SEEN) \
88f063b4 1053 Perl_re_printf( aTHX_ "REG_LOOKBEHIND_SEEN "); \
538e84ed 1054 \
e384d5c1 1055 if (RExC_seen & REG_GPOS_SEEN) \
88f063b4 1056 Perl_re_printf( aTHX_ "REG_GPOS_SEEN "); \
538e84ed 1057 \
e384d5c1 1058 if (RExC_seen & REG_RECURSE_SEEN) \
88f063b4 1059 Perl_re_printf( aTHX_ "REG_RECURSE_SEEN "); \
538e84ed 1060 \
7b031478 1061 if (RExC_seen & REG_TOP_LEVEL_BRANCHES_SEEN) \
88f063b4 1062 Perl_re_printf( aTHX_ "REG_TOP_LEVEL_BRANCHES_SEEN "); \
538e84ed 1063 \
e384d5c1 1064 if (RExC_seen & REG_VERBARG_SEEN) \
88f063b4 1065 Perl_re_printf( aTHX_ "REG_VERBARG_SEEN "); \
538e84ed 1066 \
e384d5c1 1067 if (RExC_seen & REG_CUTGROUP_SEEN) \
88f063b4 1068 Perl_re_printf( aTHX_ "REG_CUTGROUP_SEEN "); \
538e84ed 1069 \
e384d5c1 1070 if (RExC_seen & REG_RUN_ON_COMMENT_SEEN) \
88f063b4 1071 Perl_re_printf( aTHX_ "REG_RUN_ON_COMMENT_SEEN "); \
538e84ed 1072 \
e384d5c1 1073 if (RExC_seen & REG_UNFOLDED_MULTI_SEEN) \
88f063b4 1074 Perl_re_printf( aTHX_ "REG_UNFOLDED_MULTI_SEEN "); \
538e84ed 1075 \
7b031478 1076 if (RExC_seen & REG_UNBOUNDED_QUANTIFIER_SEEN) \
88f063b4 1077 Perl_re_printf( aTHX_ "REG_UNBOUNDED_QUANTIFIER_SEEN "); \
ee273784 1078 \
88f063b4 1079 Perl_re_printf( aTHX_ "\n"); \
9e9ecfdf
YO
1080 });
1081
fdfb4f21 1082#define DEBUG_SHOW_STUDY_FLAG(flags,flag) \
6ad9a8ab 1083 if ((flags) & flag) Perl_re_printf( aTHX_ "%s ", #flag)
fdfb4f21 1084
fdfb4f21 1085
f5a36d78
DM
1086#ifdef DEBUGGING
1087static void
1088S_debug_show_study_flags(pTHX_ U32 flags, const char *open_str,
1089 const char *close_str)
1090{
1091 if (!flags)
1092 return;
1093
1094 Perl_re_printf( aTHX_ "%s", open_str);
11683ecb
DM
1095 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_SEOL);
1096 DEBUG_SHOW_STUDY_FLAG(flags, SF_BEFORE_MEOL);
f5a36d78
DM
1097 DEBUG_SHOW_STUDY_FLAG(flags, SF_IS_INF);
1098 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_PAR);
1099 DEBUG_SHOW_STUDY_FLAG(flags, SF_IN_PAR);
1100 DEBUG_SHOW_STUDY_FLAG(flags, SF_HAS_EVAL);
1101 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_SUBSTR);
1102 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_AND);
1103 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS_OR);
1104 DEBUG_SHOW_STUDY_FLAG(flags, SCF_DO_STCLASS);
1105 DEBUG_SHOW_STUDY_FLAG(flags, SCF_WHILEM_VISITED_POS);
1106 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_RESTUDY);
1107 DEBUG_SHOW_STUDY_FLAG(flags, SCF_SEEN_ACCEPT);
1108 DEBUG_SHOW_STUDY_FLAG(flags, SCF_TRIE_DOING_RESTUDY);
1109 DEBUG_SHOW_STUDY_FLAG(flags, SCF_IN_DEFINE);
1110 Perl_re_printf( aTHX_ "%s", close_str);
1111}
1112
1113
1114static void
1115S_debug_studydata(pTHX_ const char *where, scan_data_t *data,
1116 U32 depth, int is_inf)
1117{
1118 GET_RE_DEBUG_FLAGS_DECL;
1119
1120 DEBUG_OPTIMISE_MORE_r({
1121 if (!data)
1122 return;
1123 Perl_re_indentf(aTHX_ "%s: Pos:%" IVdf "/%" IVdf " Flags: 0x%" UVXf,
1124 depth,
1125 where,
1126 (IV)data->pos_min,
1127 (IV)data->pos_delta,
1128 (UV)data->flags
1129 );
1130
1131 S_debug_show_study_flags(aTHX_ data->flags," [","]");
fdfb4f21 1132
f5a36d78
DM
1133 Perl_re_printf( aTHX_
1134 " Whilem_c: %" IVdf " Lcp: %" IVdf " %s",
1135 (IV)data->whilem_c,
1136 (IV)(data->last_closep ? *((data)->last_closep) : -1),
1137 is_inf ? "INF " : ""
1138 );
1139
11683ecb 1140 if (data->last_found) {
37b6262f 1141 int i;
11683ecb
DM
1142 Perl_re_printf(aTHX_
1143 "Last:'%s' %" IVdf ":%" IVdf "/%" IVdf,
1144 SvPVX_const(data->last_found),
1145 (IV)data->last_end,
1146 (IV)data->last_start_min,
1147 (IV)data->last_start_max
1148 );
1149
2099df82 1150 for (i = 0; i < 2; i++) {
37b6262f
DM
1151 Perl_re_printf(aTHX_
1152 " %s%s: '%s' @ %" IVdf "/%" IVdf,
2099df82
DM
1153 data->cur_is_floating == i ? "*" : "",
1154 i ? "Float" : "Fixed",
37b6262f
DM
1155 SvPVX_const(data->substrs[i].str),
1156 (IV)data->substrs[i].min_offset,
1157 (IV)data->substrs[i].max_offset
1158 );
1159 S_debug_show_study_flags(aTHX_ data->substrs[i].flags," [","]");
1160 }
11683ecb 1161 }
f5a36d78
DM
1162
1163 Perl_re_printf( aTHX_ "\n");
1164 });
1165}
1166
1167
1168static void
1169S_debug_peep(pTHX_ const char *str, const RExC_state_t *pRExC_state,
1170 regnode *scan, U32 depth, U32 flags)
1171{
1172 GET_RE_DEBUG_FLAGS_DECL;
1173
1174 DEBUG_OPTIMISE_r({
1175 regnode *Next;
1176
1177 if (!scan)
1178 return;
1179 Next = regnext(scan);
1180 regprop(RExC_rx, RExC_mysv, scan, NULL, pRExC_state);
1181 Perl_re_indentf( aTHX_ "%s>%3d: %s (%d)",
1182 depth,
1183 str,
1184 REG_NODE_NUM(scan), SvPV_nolen_const(RExC_mysv),
1185 Next ? (REG_NODE_NUM(Next)) : 0 );
1186 S_debug_show_study_flags(aTHX_ flags," [ ","]");
1187 Perl_re_printf( aTHX_ "\n");
1188 });
1189}
1190
1191
1192# define DEBUG_STUDYDATA(where, data, depth, is_inf) \
1193 S_debug_studydata(aTHX_ where, data, depth, is_inf)
1194
1195# define DEBUG_PEEP(str, scan, depth, flags) \
1196 S_debug_peep(aTHX_ str, pRExC_state, scan, depth, flags)
1197
1198#else
1199# define DEBUG_STUDYDATA(where, data, depth, is_inf) NOOP
1200# define DEBUG_PEEP(str, scan, depth, flags) NOOP
1201#endif
1de06328 1202
cb41e5d6 1203
c6871b76
KW
1204/* =========================================================
1205 * BEGIN edit_distance stuff.
1206 *
1207 * This calculates how many single character changes of any type are needed to
1208 * transform a string into another one. It is taken from version 3.1 of
1209 *
1210 * https://metacpan.org/pod/Text::Levenshtein::Damerau::XS
1211 */
1212
1213/* Our unsorted dictionary linked list. */
1214/* Note we use UVs, not chars. */
1215
1216struct dictionary{
1217 UV key;
1218 UV value;
1219 struct dictionary* next;
1220};
1221typedef struct dictionary item;
1222
1223
1224PERL_STATIC_INLINE item*
88f063b4 1225push(UV key, item* curr)
c6871b76
KW
1226{
1227 item* head;
d09f14bf 1228 Newx(head, 1, item);
c6871b76
KW
1229 head->key = key;
1230 head->value = 0;
1231 head->next = curr;
1232 return head;
1233}
1234
1235
1236PERL_STATIC_INLINE item*
1237find(item* head, UV key)
1238{
1239 item* iterator = head;
1240 while (iterator){
1241 if (iterator->key == key){
1242 return iterator;
1243 }
1244 iterator = iterator->next;
1245 }
1246
1247 return NULL;
1248}
1249
1250PERL_STATIC_INLINE item*
88f063b4 1251uniquePush(item* head, UV key)
c6871b76
KW
1252{
1253 item* iterator = head;
1254
1255 while (iterator){
1256 if (iterator->key == key) {
1257 return head;
1258 }
1259 iterator = iterator->next;
1260 }
1261
88f063b4 1262 return push(key, head);
c6871b76
KW
1263}
1264
1265PERL_STATIC_INLINE void
1266dict_free(item* head)
1267{
1268 item* iterator = head;
1269
1270 while (iterator) {
1271 item* temp = iterator;
1272 iterator = iterator->next;
1273 Safefree(temp);
1274 }
1275
1276 head = NULL;
1277}
1278
1279/* End of Dictionary Stuff */
1280
1281/* All calculations/work are done here */
1282STATIC int
1283S_edit_distance(const UV* src,
1284 const UV* tgt,
1285 const STRLEN x, /* length of src[] */
1286 const STRLEN y, /* length of tgt[] */
1287 const SSize_t maxDistance
1288)
1289{
1290 item *head = NULL;
88f063b4 1291 UV swapCount, swapScore, targetCharCount, i, j;
c6871b76
KW
1292 UV *scores;
1293 UV score_ceil = x + y;
1294
1295 PERL_ARGS_ASSERT_EDIT_DISTANCE;
1296
1297 /* intialize matrix start values */
d09f14bf 1298 Newx(scores, ( (x + 2) * (y + 2)), UV);
c6871b76
KW
1299 scores[0] = score_ceil;
1300 scores[1 * (y + 2) + 0] = score_ceil;
1301 scores[0 * (y + 2) + 1] = score_ceil;
1302 scores[1 * (y + 2) + 1] = 0;
88f063b4 1303 head = uniquePush(uniquePush(head, src[0]), tgt[0]);
c6871b76
KW
1304
1305 /* work loops */
1306 /* i = src index */
1307 /* j = tgt index */
1308 for (i=1;i<=x;i++) {
1309 if (i < x)
88f063b4 1310 head = uniquePush(head, src[i]);
c6871b76
KW
1311 scores[(i+1) * (y + 2) + 1] = i;
1312 scores[(i+1) * (y + 2) + 0] = score_ceil;
1313 swapCount = 0;
1314
1315 for (j=1;j<=y;j++) {
1316 if (i == 1) {
1317 if(j < y)
88f063b4 1318 head = uniquePush(head, tgt[j]);
c6871b76
KW
1319 scores[1 * (y + 2) + (j + 1)] = j;
1320 scores[0 * (y + 2) + (j + 1)] = score_ceil;
1321 }
1322
88f063b4 1323 targetCharCount = find(head, tgt[j-1])->value;
c6871b76
KW
1324 swapScore = scores[targetCharCount * (y + 2) + swapCount] + i - targetCharCount - 1 + j - swapCount;
1325
1326 if (src[i-1] != tgt[j-1]){
1327 scores[(i+1) * (y + 2) + (j + 1)] = MIN(swapScore,(MIN(scores[i * (y + 2) + j], MIN(scores[(i+1) * (y + 2) + j], scores[i * (y + 2) + (j + 1)])) + 1));
1328 }
1329 else {
1330 swapCount = j;
1331 scores[(i+1) * (y + 2) + (j + 1)] = MIN(scores[i * (y + 2) + j], swapScore);
1332 }
1333 }
1334
88f063b4 1335 find(head, src[i-1])->value = i;
c6871b76
KW
1336 }
1337
1338 {
1339 IV score = scores[(x+1) * (y + 2) + (y + 1)];
1340 dict_free(head);
1341 Safefree(scores);
1342 return (maxDistance != 0 && maxDistance < score)?(-1):score;
1343 }
1344}
1345
1346/* END of edit_distance() stuff
1347 * ========================================================= */
1348
8e35b056
KW
1349/* is c a control character for which we have a mnemonic? */
1350#define isMNEMONIC_CNTRL(c) _IS_MNEMONIC_CNTRL_ONLY_FOR_USE_BY_REGCOMP_DOT_C(c)
1351
549b4e78
KW
1352STATIC const char *
1353S_cntrl_to_mnemonic(const U8 c)
1354{
1355 /* Returns the mnemonic string that represents character 'c', if one
1356 * exists; NULL otherwise. The only ones that exist for the purposes of
1357 * this routine are a few control characters */
1358
1359 switch (c) {
1360 case '\a': return "\\a";
1361 case '\b': return "\\b";
1362 case ESC_NATIVE: return "\\e";
1363 case '\f': return "\\f";
1364 case '\n': return "\\n";
1365 case '\r': return "\\r";
1366 case '\t': return "\\t";
1367 }
1368
1369 return NULL;
1370}
1371
653099ff 1372/* Mark that we cannot extend a found fixed substring at this point.
37b6262f 1373 Update the longest found anchored substring or the longest found
653099ff
GS
1374 floating substrings if needed. */
1375
4327152a 1376STATIC void
ea3daa5d
FC
1377S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data,
1378 SSize_t *minlenp, int is_inf)
c277df42 1379{
e1ec3a88 1380 const STRLEN l = CHR_SVLEN(data->last_found);
2099df82 1381 SV * const longest_sv = data->substrs[data->cur_is_floating].str;
a2ca2017 1382 const STRLEN old_l = CHR_SVLEN(longest_sv);
1de06328 1383 GET_RE_DEBUG_FLAGS_DECL;
b81d288d 1384
7918f24d
NC
1385 PERL_ARGS_ASSERT_SCAN_COMMIT;
1386
c277df42 1387 if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
2099df82 1388 const U8 i = data->cur_is_floating;
a2ca2017 1389 SvSetMagicSV(longest_sv, data->last_found);
37b6262f
DM
1390 data->substrs[i].min_offset = l ? data->last_start_min : data->pos_min;
1391
2099df82
DM
1392 if (!i) /* fixed */
1393 data->substrs[0].max_offset = data->substrs[0].min_offset;
1394 else { /* float */
1395 data->substrs[1].max_offset = (l
646e8787
DM
1396 ? data->last_start_max
1397 : (data->pos_delta > SSize_t_MAX - data->pos_min
ea3daa5d
FC
1398 ? SSize_t_MAX
1399 : data->pos_min + data->pos_delta));
1400 if (is_inf
2099df82
DM
1401 || (STRLEN)data->substrs[1].max_offset > (STRLEN)SSize_t_MAX)
1402 data->substrs[1].max_offset = SSize_t_MAX;
37b6262f 1403 }
11683ecb 1404
37b6262f
DM
1405 if (data->flags & SF_BEFORE_EOL)
1406 data->substrs[i].flags |= (data->flags & SF_BEFORE_EOL);
1407 else
1408 data->substrs[i].flags &= ~SF_BEFORE_EOL;
1409 data->substrs[i].minlenp = minlenp;
1410 data->substrs[i].lookbehind = 0;
c277df42 1411 }
37b6262f 1412
c277df42 1413 SvCUR_set(data->last_found, 0);
0eda9292 1414 {
a28509cc 1415 SV * const sv = data->last_found;
097eb12c
AL
1416 if (SvUTF8(sv) && SvMAGICAL(sv)) {
1417 MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8);
1418 if (mg)
1419 mg->mg_len = 0;
1420 }
0eda9292 1421 }
c277df42
IZ
1422 data->last_end = -1;
1423 data->flags &= ~SF_BEFORE_EOL;
f5a36d78 1424 DEBUG_STUDYDATA("commit", data, 0, is_inf);
c277df42
IZ
1425}
1426
cdd87c1d
KW
1427/* An SSC is just a regnode_charclass_posix with an extra field: the inversion
1428 * list that describes which code points it matches */
1429
653099ff 1430STATIC void
3420edd7 1431S_ssc_anything(pTHX_ regnode_ssc *ssc)
653099ff 1432{
cdd87c1d
KW
1433 /* Set the SSC 'ssc' to match an empty string or any code point */
1434
557bd3fb 1435 PERL_ARGS_ASSERT_SSC_ANYTHING;
7918f24d 1436
71068078 1437 assert(is_ANYOF_SYNTHETIC(ssc));
3fffb88a 1438
0854ea0b
KW
1439 /* mortalize so won't leak */
1440 ssc->invlist = sv_2mortal(_add_range_to_invlist(NULL, 0, UV_MAX));
93e92956 1441 ANYOF_FLAGS(ssc) |= SSC_MATCHES_EMPTY_STRING; /* Plus matches empty */
653099ff
GS
1442}
1443
653099ff 1444STATIC int
dc3bf405 1445S_ssc_is_anything(const regnode_ssc *ssc)
653099ff 1446{
c144baaa
KW
1447 /* Returns TRUE if the SSC 'ssc' can match the empty string and any code
1448 * point; FALSE otherwise. Thus, this is used to see if using 'ssc' buys
1449 * us anything: if the function returns TRUE, 'ssc' hasn't been restricted
1450 * in any way, so there's no point in using it */
cdd87c1d
KW
1451
1452 UV start, end;
1453 bool ret;
653099ff 1454
557bd3fb 1455 PERL_ARGS_ASSERT_SSC_IS_ANYTHING;
7918f24d 1456
71068078 1457 assert(is_ANYOF_SYNTHETIC(ssc));
cdd87c1d 1458
93e92956 1459 if (! (ANYOF_FLAGS(ssc) & SSC_MATCHES_EMPTY_STRING)) {
cdd87c1d
KW
1460 return FALSE;
1461 }
1462
1463 /* See if the list consists solely of the range 0 - Infinity */
1464 invlist_iterinit(ssc->invlist);
1465 ret = invlist_iternext(ssc->invlist, &start, &end)
1466 && start == 0
1467 && end == UV_MAX;
1468
1469 invlist_iterfinish(ssc->invlist);
1470
1471 if (ret) {
1472 return TRUE;
1473 }
1474
1475 /* If e.g., both \w and \W are set, matches everything */
e0e1be5f 1476 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1477 int i;
1478 for (i = 0; i < ANYOF_POSIXL_MAX; i += 2) {
1479 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i+1)) {
1480 return TRUE;
1481 }
1482 }
1483 }
1484
1485 return FALSE;
653099ff
GS
1486}
1487
653099ff 1488STATIC void
cdd87c1d 1489S_ssc_init(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc)
653099ff 1490{
cdd87c1d
KW
1491 /* Initializes the SSC 'ssc'. This includes setting it to match an empty
1492 * string, any code point, or any posix class under locale */
1493
557bd3fb 1494 PERL_ARGS_ASSERT_SSC_INIT;
7918f24d 1495
557bd3fb 1496 Zero(ssc, 1, regnode_ssc);
71068078 1497 set_ANYOF_SYNTHETIC(ssc);
93e92956 1498 ARG_SET(ssc, ANYOF_ONLY_HAS_BITMAP);
3420edd7 1499 ssc_anything(ssc);
cdd87c1d 1500
2f306ab9
KW
1501 /* If any portion of the regex is to operate under locale rules that aren't
1502 * fully known at compile time, initialization includes it. The reason
1503 * this isn't done for all regexes is that the optimizer was written under
1504 * the assumption that locale was all-or-nothing. Given the complexity and
1505 * lack of documentation in the optimizer, and that there are inadequate
1506 * test cases for locale, many parts of it may not work properly, it is
1507 * safest to avoid locale unless necessary. */
cdd87c1d
KW
1508 if (RExC_contains_locale) {
1509 ANYOF_POSIXL_SETALL(ssc);
cdd87c1d
KW
1510 }
1511 else {
1512 ANYOF_POSIXL_ZERO(ssc);
1513 }
653099ff
GS
1514}
1515
b423522f 1516STATIC int
dc3bf405
BF
1517S_ssc_is_cp_posixl_init(const RExC_state_t *pRExC_state,
1518 const regnode_ssc *ssc)
b423522f
KW
1519{
1520 /* Returns TRUE if the SSC 'ssc' is in its initial state with regard only
1521 * to the list of code points matched, and locale posix classes; hence does
1522 * not check its flags) */
1523
1524 UV start, end;
1525 bool ret;
1526
1527 PERL_ARGS_ASSERT_SSC_IS_CP_POSIXL_INIT;
1528
71068078 1529 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1530
1531 invlist_iterinit(ssc->invlist);
1532 ret = invlist_iternext(ssc->invlist, &start, &end)
1533 && start == 0
1534 && end == UV_MAX;
1535
1536 invlist_iterfinish(ssc->invlist);
1537
1538 if (! ret) {
1539 return FALSE;
1540 }
1541
e0e1be5f 1542 if (RExC_contains_locale && ! ANYOF_POSIXL_SSC_TEST_ALL_SET(ssc)) {
31f05a37 1543 return FALSE;
b423522f
KW
1544 }
1545
1546 return TRUE;
1547}
1548
1549STATIC SV*
1550S_get_ANYOF_cp_list_for_ssc(pTHX_ const RExC_state_t *pRExC_state,
5c0f85ef 1551 const regnode_charclass* const node)
b423522f
KW
1552{
1553 /* Returns a mortal inversion list defining which code points are matched
1554 * by 'node', which is of type ANYOF. Handles complementing the result if
1555 * appropriate. If some code points aren't knowable at this time, the
31f05a37
KW
1556 * returned list must, and will, contain every code point that is a
1557 * possibility. */
b423522f 1558
e2506fa7 1559 SV* invlist = NULL;
1ee208c4 1560 SV* only_utf8_locale_invlist = NULL;
b423522f
KW
1561 unsigned int i;
1562 const U32 n = ARG(node);
31f05a37 1563 bool new_node_has_latin1 = FALSE;
b423522f
KW
1564
1565 PERL_ARGS_ASSERT_GET_ANYOF_CP_LIST_FOR_SSC;
1566
1567 /* Look at the data structure created by S_set_ANYOF_arg() */
93e92956 1568 if (n != ANYOF_ONLY_HAS_BITMAP) {
b423522f
KW
1569 SV * const rv = MUTABLE_SV(RExC_rxi->data->data[n]);
1570 AV * const av = MUTABLE_AV(SvRV(rv));
1571 SV **const ary = AvARRAY(av);
1572 assert(RExC_rxi->data->what[n] == 's');
1573
1574 if (ary[1] && ary[1] != &PL_sv_undef) { /* Has compile-time swash */
28118b9c 1575 invlist = sv_2mortal(invlist_clone(_get_swash_invlist(ary[1]), NULL));
b423522f
KW
1576 }
1577 else if (ary[0] && ary[0] != &PL_sv_undef) {
1578
1579 /* Here, no compile-time swash, and there are things that won't be
1580 * known until runtime -- we have to assume it could be anything */
e2506fa7 1581 invlist = sv_2mortal(_new_invlist(1));
b423522f
KW
1582 return _add_range_to_invlist(invlist, 0, UV_MAX);
1583 }
1ee208c4 1584 else if (ary[3] && ary[3] != &PL_sv_undef) {
b423522f
KW
1585
1586 /* Here no compile-time swash, and no run-time only data. Use the
1587 * node's inversion list */
28118b9c 1588 invlist = sv_2mortal(invlist_clone(ary[3], NULL));
1ee208c4
KW
1589 }
1590
1591 /* Get the code points valid only under UTF-8 locales */
037715a6 1592 if ((ANYOF_FLAGS(node) & ANYOFL_FOLD)
1ee208c4
KW
1593 && ary[2] && ary[2] != &PL_sv_undef)
1594 {
1595 only_utf8_locale_invlist = ary[2];
b423522f
KW
1596 }
1597 }
1598
e2506fa7
KW
1599 if (! invlist) {
1600 invlist = sv_2mortal(_new_invlist(0));
1601 }
1602
dcb20b36
KW
1603 /* An ANYOF node contains a bitmap for the first NUM_ANYOF_CODE_POINTS
1604 * code points, and an inversion list for the others, but if there are code
1605 * points that should match only conditionally on the target string being
1606 * UTF-8, those are placed in the inversion list, and not the bitmap.
1607 * Since there are circumstances under which they could match, they are
1608 * included in the SSC. But if the ANYOF node is to be inverted, we have
1609 * to exclude them here, so that when we invert below, the end result
1610 * actually does include them. (Think about "\xe0" =~ /[^\xc0]/di;). We
1611 * have to do this here before we add the unconditionally matched code
1612 * points */
b423522f
KW
1613 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1614 _invlist_intersection_complement_2nd(invlist,
1615 PL_UpperLatin1,
1616 &invlist);
1617 }
1618
1619 /* Add in the points from the bit map */
dcb20b36 1620 for (i = 0; i < NUM_ANYOF_CODE_POINTS; i++) {
b423522f 1621 if (ANYOF_BITMAP_TEST(node, i)) {
6f8848d5
TC
1622 unsigned int start = i++;
1623
1624 for (; i < NUM_ANYOF_CODE_POINTS && ANYOF_BITMAP_TEST(node, i); ++i) {
1625 /* empty */
1626 }
1627 invlist = _add_range_to_invlist(invlist, start, i-1);
31f05a37 1628 new_node_has_latin1 = TRUE;
b423522f
KW
1629 }
1630 }
1631
1632 /* If this can match all upper Latin1 code points, have to add them
ac33c516
KW
1633 * as well. But don't add them if inverting, as when that gets done below,
1634 * it would exclude all these characters, including the ones it shouldn't
1635 * that were added just above */
1636 if (! (ANYOF_FLAGS(node) & ANYOF_INVERT) && OP(node) == ANYOFD
f240c685
KW
1637 && (ANYOF_FLAGS(node) & ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER))
1638 {
b423522f
KW
1639 _invlist_union(invlist, PL_UpperLatin1, &invlist);
1640 }
1641
1642 /* Similarly for these */
93e92956 1643 if (ANYOF_FLAGS(node) & ANYOF_MATCHES_ALL_ABOVE_BITMAP) {
e0a1ff7a 1644 _invlist_union_complement_2nd(invlist, PL_InBitmap, &invlist);
b423522f
KW
1645 }
1646
1647 if (ANYOF_FLAGS(node) & ANYOF_INVERT) {
1648 _invlist_invert(invlist);
1649 }
037715a6 1650 else if (new_node_has_latin1 && ANYOF_FLAGS(node) & ANYOFL_FOLD) {
31f05a37
KW
1651
1652 /* Under /li, any 0-255 could fold to any other 0-255, depending on the
1653 * locale. We can skip this if there are no 0-255 at all. */
1654 _invlist_union(invlist, PL_Latin1, &invlist);
1655 }
1656
1ee208c4
KW
1657 /* Similarly add the UTF-8 locale possible matches. These have to be
1658 * deferred until after the non-UTF-8 locale ones are taken care of just
1659 * above, or it leads to wrong results under ANYOF_INVERT */
1660 if (only_utf8_locale_invlist) {
31f05a37 1661 _invlist_union_maybe_complement_2nd(invlist,
1ee208c4 1662 only_utf8_locale_invlist,
31f05a37
KW
1663 ANYOF_FLAGS(node) & ANYOF_INVERT,
1664 &invlist);
1665 }
b423522f
KW
1666
1667 return invlist;
1668}
1669
1051e1c4 1670/* These two functions currently do the exact same thing */
557bd3fb 1671#define ssc_init_zero ssc_init
653099ff 1672
cdd87c1d
KW
1673#define ssc_add_cp(ssc, cp) ssc_add_range((ssc), (cp), (cp))
1674#define ssc_match_all_cp(ssc) ssc_add_range(ssc, 0, UV_MAX)
1675
557bd3fb 1676/* 'AND' a given class with another one. Can create false positives. 'ssc'
93e92956
KW
1677 * should not be inverted. 'and_with->flags & ANYOF_MATCHES_POSIXL' should be
1678 * 0 if 'and_with' is a regnode_charclass instead of a regnode_ssc. */
cdd87c1d 1679
653099ff 1680STATIC void
b423522f 1681S_ssc_and(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1682 const regnode_charclass *and_with)
653099ff 1683{
cdd87c1d
KW
1684 /* Accumulate into SSC 'ssc' its 'AND' with 'and_with', which is either
1685 * another SSC or a regular ANYOF class. Can create false positives. */
40d049e4 1686
a0dd4231
KW
1687 SV* anded_cp_list;
1688 U8 anded_flags;
1e6ade67 1689
cdd87c1d 1690 PERL_ARGS_ASSERT_SSC_AND;
653099ff 1691
71068078 1692 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1693
1694 /* 'and_with' is used as-is if it too is an SSC; otherwise have to extract
1695 * the code point inversion list and just the relevant flags */
71068078 1696 if (is_ANYOF_SYNTHETIC(and_with)) {
7dcac5f6 1697 anded_cp_list = ((regnode_ssc *)and_with)->invlist;
a0dd4231 1698 anded_flags = ANYOF_FLAGS(and_with);
e9b08962
KW
1699
1700 /* XXX This is a kludge around what appears to be deficiencies in the
1701 * optimizer. If we make S_ssc_anything() add in the WARN_SUPER flag,
1702 * there are paths through the optimizer where it doesn't get weeded
1703 * out when it should. And if we don't make some extra provision for
1704 * it like the code just below, it doesn't get added when it should.
1705 * This solution is to add it only when AND'ing, which is here, and
1706 * only when what is being AND'ed is the pristine, original node
1707 * matching anything. Thus it is like adding it to ssc_anything() but
1708 * only when the result is to be AND'ed. Probably the same solution
1709 * could be adopted for the same problem we have with /l matching,
1710 * which is solved differently in S_ssc_init(), and that would lead to
1711 * fewer false positives than that solution has. But if this solution
1712 * creates bugs, the consequences are only that a warning isn't raised
1713 * that should be; while the consequences for having /l bugs is
1714 * incorrect matches */
7dcac5f6 1715 if (ssc_is_anything((regnode_ssc *)and_with)) {
f240c685 1716 anded_flags |= ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER;
e9b08962 1717 }
a0dd4231
KW
1718 }
1719 else {
5c0f85ef 1720 anded_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, and_with);
f240c685
KW
1721 if (OP(and_with) == ANYOFD) {
1722 anded_flags = ANYOF_FLAGS(and_with) & ANYOF_COMMON_FLAGS;
1723 }
1724 else {
1725 anded_flags = ANYOF_FLAGS(and_with)
1726 &( ANYOF_COMMON_FLAGS
108316fb
KW
1727 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1728 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1729 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(and_with))) {
1730 anded_flags &=
1731 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1732 }
f240c685 1733 }
a0dd4231
KW
1734 }
1735
1736 ANYOF_FLAGS(ssc) &= anded_flags;
cdd87c1d
KW
1737
1738 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1739 * C2 is the list of code points in 'and-with'; P2, its posix classes.
1740 * 'and_with' may be inverted. When not inverted, we have the situation of
1741 * computing:
1742 * (C1 | P1) & (C2 | P2)
1743 * = (C1 & (C2 | P2)) | (P1 & (C2 | P2))
1744 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1745 * <= ((C1 & C2) | P2)) | ( P1 | (P1 & P2))
1746 * <= ((C1 & C2) | P1 | P2)
1747 * Alternatively, the last few steps could be:
1748 * = ((C1 & C2) | (C1 & P2)) | ((P1 & C2) | (P1 & P2))
1749 * <= ((C1 & C2) | C1 ) | ( C2 | (P1 & P2))
1750 * <= (C1 | C2 | (P1 & P2))
1751 * We favor the second approach if either P1 or P2 is non-empty. This is
1752 * because these components are a barrier to doing optimizations, as what
1753 * they match cannot be known until the moment of matching as they are
1754 * dependent on the current locale, 'AND"ing them likely will reduce or
1755 * eliminate them.
1756 * But we can do better if we know that C1,P1 are in their initial state (a
1757 * frequent occurrence), each matching everything:
1758 * (<everything>) & (C2 | P2) = C2 | P2
1759 * Similarly, if C2,P2 are in their initial state (again a frequent
1760 * occurrence), the result is a no-op
1761 * (C1 | P1) & (<everything>) = C1 | P1
1762 *
1763 * Inverted, we have
1764 * (C1 | P1) & ~(C2 | P2) = (C1 | P1) & (~C2 & ~P2)
1765 * = (C1 & (~C2 & ~P2)) | (P1 & (~C2 & ~P2))
1766 * <= (C1 & ~C2) | (P1 & ~P2)
1767 * */
1aa99e6b 1768
a0dd4231 1769 if ((ANYOF_FLAGS(and_with) & ANYOF_INVERT)
71068078 1770 && ! is_ANYOF_SYNTHETIC(and_with))
a0dd4231 1771 {
cdd87c1d 1772 unsigned int i;
8951c461 1773
cdd87c1d
KW
1774 ssc_intersection(ssc,
1775 anded_cp_list,
1776 FALSE /* Has already been inverted */
1777 );
c6b76537 1778
cdd87c1d
KW
1779 /* If either P1 or P2 is empty, the intersection will be also; can skip
1780 * the loop */
93e92956 1781 if (! (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL)) {
cdd87c1d
KW
1782 ANYOF_POSIXL_ZERO(ssc);
1783 }
e0e1be5f 1784 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1785
1786 /* Note that the Posix class component P from 'and_with' actually
1787 * looks like:
1788 * P = Pa | Pb | ... | Pn
1789 * where each component is one posix class, such as in [\w\s].
1790 * Thus
1791 * ~P = ~(Pa | Pb | ... | Pn)
1792 * = ~Pa & ~Pb & ... & ~Pn
1793 * <= ~Pa | ~Pb | ... | ~Pn
1794 * The last is something we can easily calculate, but unfortunately
1795 * is likely to have many false positives. We could do better
1796 * in some (but certainly not all) instances if two classes in
1797 * P have known relationships. For example
1798 * :lower: <= :alpha: <= :alnum: <= \w <= :graph: <= :print:
1799 * So
1800 * :lower: & :print: = :lower:
1801 * And similarly for classes that must be disjoint. For example,
1802 * since \s and \w can have no elements in common based on rules in
1803 * the POSIX standard,
1804 * \w & ^\S = nothing
1805 * Unfortunately, some vendor locales do not meet the Posix
1806 * standard, in particular almost everything by Microsoft.
1807 * The loop below just changes e.g., \w into \W and vice versa */
1808
1ee208c4 1809 regnode_charclass_posixl temp;
cdd87c1d
KW
1810 int add = 1; /* To calculate the index of the complement */
1811
b1234259 1812 Zero(&temp, 1, regnode_charclass_posixl);
cdd87c1d
KW
1813 ANYOF_POSIXL_ZERO(&temp);
1814 for (i = 0; i < ANYOF_MAX; i++) {
1815 assert(i % 2 != 0
7dcac5f6
KW
1816 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)
1817 || ! ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i + 1));
cdd87c1d 1818
7dcac5f6 1819 if (ANYOF_POSIXL_TEST((regnode_charclass_posixl*) and_with, i)) {
cdd87c1d
KW
1820 ANYOF_POSIXL_SET(&temp, i + add);
1821 }
1822 add = 0 - add; /* 1 goes to -1; -1 goes to 1 */
1823 }
1824 ANYOF_POSIXL_AND(&temp, ssc);
c6b76537 1825
cdd87c1d
KW
1826 } /* else ssc already has no posixes */
1827 } /* else: Not inverted. This routine is a no-op if 'and_with' is an SSC
1828 in its initial state */
71068078 1829 else if (! is_ANYOF_SYNTHETIC(and_with)
7dcac5f6 1830 || ! ssc_is_cp_posixl_init(pRExC_state, (regnode_ssc *)and_with))
cdd87c1d
KW
1831 {
1832 /* But if 'ssc' is in its initial state, the result is just 'and_with';
1833 * copy it over 'ssc' */
1834 if (ssc_is_cp_posixl_init(pRExC_state, ssc)) {
71068078 1835 if (is_ANYOF_SYNTHETIC(and_with)) {
cdd87c1d
KW
1836 StructCopy(and_with, ssc, regnode_ssc);
1837 }
1838 else {
1839 ssc->invlist = anded_cp_list;
1840 ANYOF_POSIXL_ZERO(ssc);
93e92956 1841 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1842 ANYOF_POSIXL_OR((regnode_charclass_posixl*) and_with, ssc);
cdd87c1d
KW
1843 }
1844 }
1845 }
e0e1be5f 1846 else if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)
93e92956 1847 || (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL))
cdd87c1d
KW
1848 {
1849 /* One or the other of P1, P2 is non-empty. */
93e92956 1850 if (ANYOF_FLAGS(and_with) & ANYOF_MATCHES_POSIXL) {
1ea8b7fe
KW
1851 ANYOF_POSIXL_AND((regnode_charclass_posixl*) and_with, ssc);
1852 }
cdd87c1d
KW
1853 ssc_union(ssc, anded_cp_list, FALSE);
1854 }
1855 else { /* P1 = P2 = empty */
1856 ssc_intersection(ssc, anded_cp_list, FALSE);
1857 }
137165a6 1858 }
653099ff
GS
1859}
1860
653099ff 1861STATIC void
cdd87c1d 1862S_ssc_or(pTHX_ const RExC_state_t *pRExC_state, regnode_ssc *ssc,
7dcac5f6 1863 const regnode_charclass *or_with)
653099ff 1864{
cdd87c1d
KW
1865 /* Accumulate into SSC 'ssc' its 'OR' with 'or_with', which is either
1866 * another SSC or a regular ANYOF class. Can create false positives if
1867 * 'or_with' is to be inverted. */
7918f24d 1868
a0dd4231
KW
1869 SV* ored_cp_list;
1870 U8 ored_flags;
c6b76537 1871
cdd87c1d 1872 PERL_ARGS_ASSERT_SSC_OR;
c6b76537 1873
71068078 1874 assert(is_ANYOF_SYNTHETIC(ssc));
a0dd4231
KW
1875
1876 /* 'or_with' is used as-is if it too is an SSC; otherwise have to extract
1877 * the code point inversion list and just the relevant flags */
71068078 1878 if (is_ANYOF_SYNTHETIC(or_with)) {
7dcac5f6 1879 ored_cp_list = ((regnode_ssc*) or_with)->invlist;
a0dd4231
KW
1880 ored_flags = ANYOF_FLAGS(or_with);
1881 }
1882 else {
5c0f85ef 1883 ored_cp_list = get_ANYOF_cp_list_for_ssc(pRExC_state, or_with);
eff8b7dc 1884 ored_flags = ANYOF_FLAGS(or_with) & ANYOF_COMMON_FLAGS;
f240c685
KW
1885 if (OP(or_with) != ANYOFD) {
1886 ored_flags
1887 |= ANYOF_FLAGS(or_with)
108316fb
KW
1888 & ( ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
1889 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP);
d1c40ef5
KW
1890 if (ANYOFL_UTF8_LOCALE_REQD(ANYOF_FLAGS(or_with))) {
1891 ored_flags |=
1892 ANYOFL_SHARED_UTF8_LOCALE_fold_HAS_MATCHES_nonfold_REQD;
1893 }
f240c685 1894 }
a0dd4231
KW
1895 }
1896
1897 ANYOF_FLAGS(ssc) |= ored_flags;
cdd87c1d
KW
1898
1899 /* Below, C1 is the list of code points in 'ssc'; P1, its posix classes.
1900 * C2 is the list of code points in 'or-with'; P2, its posix classes.
1901 * 'or_with' may be inverted. When not inverted, we have the simple
1902 * situation of computing:
1903 * (C1 | P1) | (C2 | P2) = (C1 | C2) | (P1 | P2)
1904 * If P1|P2 yields a situation with both a class and its complement are
1905 * set, like having both \w and \W, this matches all code points, and we
1906 * can delete these from the P component of the ssc going forward. XXX We
1907 * might be able to delete all the P components, but I (khw) am not certain
1908 * about this, and it is better to be safe.
1909 *
1910 * Inverted, we have
1911 * (C1 | P1) | ~(C2 | P2) = (C1 | P1) | (~C2 & ~P2)
1912 * <= (C1 | P1) | ~C2
1913 * <= (C1 | ~C2) | P1
1914 * (which results in actually simpler code than the non-inverted case)
1915 * */
9826f543 1916
a0dd4231 1917 if ((ANYOF_FLAGS(or_with) & ANYOF_INVERT)
71068078 1918 && ! is_ANYOF_SYNTHETIC(or_with))
a0dd4231 1919 {
cdd87c1d 1920 /* We ignore P2, leaving P1 going forward */
1ea8b7fe 1921 } /* else Not inverted */
93e92956 1922 else if (ANYOF_FLAGS(or_with) & ANYOF_MATCHES_POSIXL) {
7dcac5f6 1923 ANYOF_POSIXL_OR((regnode_charclass_posixl*)or_with, ssc);
e0e1be5f 1924 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
cdd87c1d
KW
1925 unsigned int i;
1926 for (i = 0; i < ANYOF_MAX; i += 2) {
1927 if (ANYOF_POSIXL_TEST(ssc, i) && ANYOF_POSIXL_TEST(ssc, i + 1))
1928 {
1929 ssc_match_all_cp(ssc);
1930 ANYOF_POSIXL_CLEAR(ssc, i);
1931 ANYOF_POSIXL_CLEAR(ssc, i+1);
cdd87c1d
KW
1932 }
1933 }
1934 }
1aa99e6b 1935 }
cdd87c1d
KW
1936
1937 ssc_union(ssc,
1938 ored_cp_list,
1939 FALSE /* Already has been inverted */
1940 );
653099ff
GS
1941}
1942
b423522f
KW
1943PERL_STATIC_INLINE void
1944S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, const bool invert2nd)
1945{
1946 PERL_ARGS_ASSERT_SSC_UNION;
1947
71068078 1948 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1949
1950 _invlist_union_maybe_complement_2nd(ssc->invlist,
1951 invlist,
1952 invert2nd,
1953 &ssc->invlist);
1954}
1955
1956PERL_STATIC_INLINE void
1957S_ssc_intersection(pTHX_ regnode_ssc *ssc,
1958 SV* const invlist,
1959 const bool invert2nd)
1960{
1961 PERL_ARGS_ASSERT_SSC_INTERSECTION;
1962
71068078 1963 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1964
1965 _invlist_intersection_maybe_complement_2nd(ssc->invlist,
1966 invlist,
1967 invert2nd,
1968 &ssc->invlist);
1969}
1970
1971PERL_STATIC_INLINE void
1972S_ssc_add_range(pTHX_ regnode_ssc *ssc, const UV start, const UV end)
1973{
1974 PERL_ARGS_ASSERT_SSC_ADD_RANGE;
1975
71068078 1976 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1977
1978 ssc->invlist = _add_range_to_invlist(ssc->invlist, start, end);
1979}
1980
1981PERL_STATIC_INLINE void
1982S_ssc_cp_and(pTHX_ regnode_ssc *ssc, const UV cp)
1983{
1984 /* AND just the single code point 'cp' into the SSC 'ssc' */
1985
1986 SV* cp_list = _new_invlist(2);
1987
1988 PERL_ARGS_ASSERT_SSC_CP_AND;
1989
71068078 1990 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
1991
1992 cp_list = add_cp_to_invlist(cp_list, cp);
1993 ssc_intersection(ssc, cp_list,
1994 FALSE /* Not inverted */
1995 );
1996 SvREFCNT_dec_NN(cp_list);
1997}
1998
1999PERL_STATIC_INLINE void
dc3bf405 2000S_ssc_clear_locale(regnode_ssc *ssc)
b423522f
KW
2001{
2002 /* Set the SSC 'ssc' to not match any locale things */
b423522f
KW
2003 PERL_ARGS_ASSERT_SSC_CLEAR_LOCALE;
2004
71068078 2005 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f
KW
2006
2007 ANYOF_POSIXL_ZERO(ssc);
2008 ANYOF_FLAGS(ssc) &= ~ANYOF_LOCALE_FLAGS;
2009}
2010
b35552de
KW
2011#define NON_OTHER_COUNT NON_OTHER_COUNT_FOR_USE_ONLY_BY_REGCOMP_DOT_C
2012
2013STATIC bool
2014S_is_ssc_worth_it(const RExC_state_t * pRExC_state, const regnode_ssc * ssc)
2015{
2016 /* The synthetic start class is used to hopefully quickly winnow down
2017 * places where a pattern could start a match in the target string. If it
2018 * doesn't really narrow things down that much, there isn't much point to
2019 * having the overhead of using it. This function uses some very crude
2020 * heuristics to decide if to use the ssc or not.
2021 *
2022 * It returns TRUE if 'ssc' rules out more than half what it considers to
2023 * be the "likely" possible matches, but of course it doesn't know what the
2024 * actual things being matched are going to be; these are only guesses
2025 *
2026 * For /l matches, it assumes that the only likely matches are going to be
2027 * in the 0-255 range, uniformly distributed, so half of that is 127
2028 * For /a and /d matches, it assumes that the likely matches will be just
2029 * the ASCII range, so half of that is 63
2030 * For /u and there isn't anything matching above the Latin1 range, it
2031 * assumes that that is the only range likely to be matched, and uses
2032 * half that as the cut-off: 127. If anything matches above Latin1,
2033 * it assumes that all of Unicode could match (uniformly), except for
2034 * non-Unicode code points and things in the General Category "Other"
2035 * (unassigned, private use, surrogates, controls and formats). This
2036 * is a much large number. */
2037
b35552de
KW
2038 U32 count = 0; /* Running total of number of code points matched by
2039 'ssc' */
2040 UV start, end; /* Start and end points of current range in inversion
2041 list */
72400949
KW
2042 const U32 max_code_points = (LOC)
2043 ? 256
1a26cbcb
KW
2044 : (( ! RExC_uni_semantics
2045 || invlist_highest(ssc->invlist) < 256)
72400949
KW
2046 ? 128
2047 : NON_OTHER_COUNT);
2048 const U32 max_match = max_code_points / 2;
b35552de
KW
2049
2050 PERL_ARGS_ASSERT_IS_SSC_WORTH_IT;
2051
2052 invlist_iterinit(ssc->invlist);
2053 while (invlist_iternext(ssc->invlist, &start, &end)) {
72400949
KW
2054 if (start >= max_code_points) {
2055 break;
b35552de 2056 }
72400949 2057 end = MIN(end, max_code_points - 1);
b35552de 2058 count += end - start + 1;
72400949 2059 if (count >= max_match) {
b35552de
KW
2060 invlist_iterfinish(ssc->invlist);
2061 return FALSE;
2062 }
2063 }
2064
2065 return TRUE;
2066}
2067
2068
b423522f
KW
2069STATIC void
2070S_ssc_finalize(pTHX_ RExC_state_t *pRExC_state, regnode_ssc *ssc)
2071{
2072 /* The inversion list in the SSC is marked mortal; now we need a more
2073 * permanent copy, which is stored the same way that is done in a regular
dcb20b36
KW
2074 * ANYOF node, with the first NUM_ANYOF_CODE_POINTS code points in a bit
2075 * map */
b423522f 2076
28118b9c 2077 SV* invlist = invlist_clone(ssc->invlist, NULL);
b423522f
KW
2078
2079 PERL_ARGS_ASSERT_SSC_FINALIZE;
2080
71068078 2081 assert(is_ANYOF_SYNTHETIC(ssc));
b423522f 2082
a0dd4231 2083 /* The code in this file assumes that all but these flags aren't relevant
93e92956
KW
2084 * to the SSC, except SSC_MATCHES_EMPTY_STRING, which should be cleared
2085 * by the time we reach here */
f240c685
KW
2086 assert(! (ANYOF_FLAGS(ssc)
2087 & ~( ANYOF_COMMON_FLAGS
108316fb
KW
2088 |ANYOF_SHARED_d_MATCHES_ALL_NON_UTF8_NON_ASCII_non_d_WARN_SUPER
2089 |ANYOF_SHARED_d_UPPER_LATIN1_UTF8_STRING_MATCHES_non_d_RUNTIME_USER_PROP)));
a0dd4231 2090
b423522f
KW
2091 populate_ANYOF_from_invlist( (regnode *) ssc, &invlist);
2092
1ee208c4
KW
2093 set_ANYOF_arg(pRExC_state, (regnode *) ssc, invlist,
2094 NULL, NULL, NULL, FALSE);
b423522f 2095
85c8e306
KW
2096 /* Make sure is clone-safe */
2097 ssc->invlist = NULL;
2098
e0e1be5f 2099 if (ANYOF_POSIXL_SSC_TEST_ANY_SET(ssc)) {
93e92956 2100 ANYOF_FLAGS(ssc) |= ANYOF_MATCHES_POSIXL;
d156f5cb 2101 OP(ssc) = ANYOFPOSIXL;
e0e1be5f 2102 }
d156f5cb 2103 else if (RExC_contains_locale) {
b2e90ddf
KW
2104 OP(ssc) = ANYOFL;
2105 }
2106
1462525b 2107 assert(! (ANYOF_FLAGS(ssc) & ANYOF_LOCALE_FLAGS) || RExC_contains_locale);
b423522f
KW
2108}
2109
a3621e74
YO
2110#define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ]
2111#define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid )
2112#define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate )
538e84ed
KW
2113#define TRIE_LIST_USED(idx) ( trie->states[state].trans.list \
2114 ? (TRIE_LIST_CUR( idx ) - 1) \
2115 : 0 )
a3621e74 2116
3dab1dad
YO
2117
2118#ifdef DEBUGGING
07be1b83 2119/*
2b8b4781
NC
2120 dump_trie(trie,widecharmap,revcharmap)
2121 dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
2122 dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
3dab1dad
YO
2123
2124 These routines dump out a trie in a somewhat readable format.
07be1b83
YO
2125 The _interim_ variants are used for debugging the interim
2126 tables that are used to generate the final compressed
2127 representation which is what dump_trie expects.
2128
486ec47a 2129 Part of the reason for their existence is to provide a form
3dab1dad 2130 of documentation as to how the different representations function.
07be1b83
YO
2131
2132*/
3dab1dad
YO
2133
2134/*
3dab1dad
YO
2135 Dumps the final compressed table form of the trie to Perl_debug_log.
2136 Used for debugging make_trie().
2137*/
b9a59e08 2138
3dab1dad 2139STATIC void
2b8b4781
NC
2140S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
2141 AV *revcharmap, U32 depth)
3dab1dad
YO
2142{
2143 U32 state;
ab3bbdeb 2144 SV *sv=sv_newmortal();
55eed653 2145 int colwidth= widecharmap ? 6 : 4;
2e64971a 2146 U16 word;
3dab1dad
YO
2147 GET_RE_DEBUG_FLAGS_DECL;
2148
7918f24d 2149 PERL_ARGS_ASSERT_DUMP_TRIE;
ab3bbdeb 2150
6ad9a8ab 2151 Perl_re_indentf( aTHX_ "Char : %-6s%-6s%-4s ",
1e37780e 2152 depth+1, "Match","Base","Ofs" );
3dab1dad
YO
2153
2154 for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
2b8b4781 2155 SV ** const tmp = av_fetch( revcharmap, state, 0);
3dab1dad 2156 if ( tmp ) {
6ad9a8ab 2157 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2158 colwidth,
538e84ed 2159 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2160 PL_colors[0], PL_colors[1],
2161 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2162 PERL_PV_ESCAPE_FIRSTCHAR
2163 )
ab3bbdeb 2164 );
3dab1dad
YO
2165 }
2166 }
1e37780e
YO
2167 Perl_re_printf( aTHX_ "\n");
2168 Perl_re_indentf( aTHX_ "State|-----------------------", depth+1);
3dab1dad
YO
2169
2170 for( state = 0 ; state < trie->uniquecharcount ; state++ )
6ad9a8ab
YO
2171 Perl_re_printf( aTHX_ "%.*s", colwidth, "--------");
2172 Perl_re_printf( aTHX_ "\n");
3dab1dad 2173
1e2e3d02 2174 for( state = 1 ; state < trie->statecount ; state++ ) {
be8e71aa 2175 const U32 base = trie->states[ state ].trans.base;
3dab1dad 2176
147e3846 2177 Perl_re_indentf( aTHX_ "#%4" UVXf "|", depth+1, (UV)state);
3dab1dad
YO
2178
2179 if ( trie->states[ state ].wordnum ) {
1e37780e 2180 Perl_re_printf( aTHX_ " W%4X", trie->states[ state ].wordnum );
3dab1dad 2181 } else {
6ad9a8ab 2182 Perl_re_printf( aTHX_ "%6s", "" );
3dab1dad
YO
2183 }
2184
147e3846 2185 Perl_re_printf( aTHX_ " @%4" UVXf " ", (UV)base );
3dab1dad
YO
2186
2187 if ( base ) {
2188 U32 ofs = 0;
2189
2190 while( ( base + ofs < trie->uniquecharcount ) ||
2191 ( base + ofs - trie->uniquecharcount < trie->lasttrans
538e84ed
KW
2192 && trie->trans[ base + ofs - trie->uniquecharcount ].check
2193 != state))
3dab1dad
YO
2194 ofs++;
2195
147e3846 2196 Perl_re_printf( aTHX_ "+%2" UVXf "[ ", (UV)ofs);
3dab1dad
YO
2197
2198 for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) {
538e84ed
KW
2199 if ( ( base + ofs >= trie->uniquecharcount )
2200 && ( base + ofs - trie->uniquecharcount
2201 < trie->lasttrans )
2202 && trie->trans[ base + ofs
2203 - trie->uniquecharcount ].check == state )
3dab1dad 2204 {
147e3846 2205 Perl_re_printf( aTHX_ "%*" UVXf, colwidth,
1e37780e
YO
2206 (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next
2207 );
3dab1dad 2208 } else {
88f063b4 2209 Perl_re_printf( aTHX_ "%*s", colwidth," ." );
3dab1dad
YO
2210 }
2211 }
2212
6ad9a8ab 2213 Perl_re_printf( aTHX_ "]");
3dab1dad
YO
2214
2215 }
6ad9a8ab 2216 Perl_re_printf( aTHX_ "\n" );
3dab1dad 2217 }
6ad9a8ab 2218 Perl_re_indentf( aTHX_ "word_info N:(prev,len)=",
cb41e5d6 2219 depth);
2e64971a 2220 for (word=1; word <= trie->wordcount; word++) {
6ad9a8ab 2221 Perl_re_printf( aTHX_ " %d:(%d,%d)",
2e64971a
DM
2222 (int)word, (int)(trie->wordinfo[word].prev),
2223 (int)(trie->wordinfo[word].len));
2224 }
6ad9a8ab 2225 Perl_re_printf( aTHX_ "\n" );
538e84ed 2226}
3dab1dad 2227/*
3dab1dad 2228 Dumps a fully constructed but uncompressed trie in list form.
538e84ed 2229 List tries normally only are used for construction when the number of
3dab1dad
YO
2230 possible chars (trie->uniquecharcount) is very high.
2231 Used for debugging make_trie().
2232*/
2233STATIC void
55eed653 2234S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2235 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2236 U32 depth)
3dab1dad
YO
2237{
2238 U32 state;
ab3bbdeb 2239 SV *sv=sv_newmortal();
55eed653 2240 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2241 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2242
2243 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST;
2244
3dab1dad 2245 /* print out the table precompression. */
6ad9a8ab 2246 Perl_re_indentf( aTHX_ "State :Word | Transition Data\n",
cb41e5d6 2247 depth+1 );
6ad9a8ab 2248 Perl_re_indentf( aTHX_ "%s",
cb41e5d6 2249 depth+1, "------:-----+-----------------\n" );
538e84ed 2250
3dab1dad
YO
2251 for( state=1 ; state < next_alloc ; state ++ ) {
2252 U16 charid;
538e84ed 2253
147e3846 2254 Perl_re_indentf( aTHX_ " %4" UVXf " :",
cb41e5d6 2255 depth+1, (UV)state );
3dab1dad 2256 if ( ! trie->states[ state ].wordnum ) {
6ad9a8ab 2257 Perl_re_printf( aTHX_ "%5s| ","");
3dab1dad 2258 } else {
6ad9a8ab 2259 Perl_re_printf( aTHX_ "W%4x| ",
3dab1dad
YO
2260 trie->states[ state ].wordnum
2261 );
2262 }
2263 for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
538e84ed 2264 SV ** const tmp = av_fetch( revcharmap,
88f063b4 2265 TRIE_LIST_ITEM(state, charid).forid, 0);
ab3bbdeb 2266 if ( tmp ) {
147e3846 2267 Perl_re_printf( aTHX_ "%*s:%3X=%4" UVXf " | ",
ab3bbdeb 2268 colwidth,
538e84ed
KW
2269 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp),
2270 colwidth,
2271 PL_colors[0], PL_colors[1],
2272 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0)
2273 | PERL_PV_ESCAPE_FIRSTCHAR
ab3bbdeb 2274 ) ,
88f063b4
KW
2275 TRIE_LIST_ITEM(state, charid).forid,
2276 (UV)TRIE_LIST_ITEM(state, charid).newstate
1e2e3d02 2277 );
538e84ed 2278 if (!(charid % 10))
6ad9a8ab 2279 Perl_re_printf( aTHX_ "\n%*s| ",
664e119d 2280 (int)((depth * 2) + 14), "");
1e2e3d02 2281 }
ab3bbdeb 2282 }
6ad9a8ab 2283 Perl_re_printf( aTHX_ "\n");
3dab1dad 2284 }
538e84ed 2285}
3dab1dad
YO
2286
2287/*
3dab1dad 2288 Dumps a fully constructed but uncompressed trie in table form.
538e84ed
KW
2289 This is the normal DFA style state transition table, with a few
2290 twists to facilitate compression later.
3dab1dad
YO
2291 Used for debugging make_trie().
2292*/
2293STATIC void
55eed653 2294S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
2b8b4781
NC
2295 HV *widecharmap, AV *revcharmap, U32 next_alloc,
2296 U32 depth)
3dab1dad
YO
2297{
2298 U32 state;
2299 U16 charid;
ab3bbdeb 2300 SV *sv=sv_newmortal();
55eed653 2301 int colwidth= widecharmap ? 6 : 4;
3dab1dad 2302 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2303
2304 PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE;
538e84ed 2305
3dab1dad
YO
2306 /*
2307 print out the table precompression so that we can do a visual check
2308 that they are identical.
2309 */
538e84ed 2310
6ad9a8ab 2311 Perl_re_indentf( aTHX_ "Char : ", depth+1 );
3dab1dad
YO
2312
2313 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
2b8b4781 2314 SV ** const tmp = av_fetch( revcharmap, charid, 0);
3dab1dad 2315 if ( tmp ) {
6ad9a8ab 2316 Perl_re_printf( aTHX_ "%*s",
ab3bbdeb 2317 colwidth,
538e84ed 2318 pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
ab3bbdeb
YO
2319 PL_colors[0], PL_colors[1],
2320 (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
538e84ed
KW
2321 PERL_PV_ESCAPE_FIRSTCHAR
2322 )
ab3bbdeb 2323 );
3dab1dad
YO
2324 }
2325 }
2326
4aaafc03
YO
2327 Perl_re_printf( aTHX_ "\n");
2328 Perl_re_indentf( aTHX_ "State+-", depth+1 );
3dab1dad
YO
2329
2330 for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
6ad9a8ab 2331 Perl_re_printf( aTHX_ "%.*s", colwidth,"--------");
3dab1dad
YO
2332 }
2333
6ad9a8ab 2334 Perl_re_printf( aTHX_ "\n" );
3dab1dad
YO
2335
2336 for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) {
2337
147e3846 2338 Perl_re_indentf( aTHX_ "%4" UVXf " : ",
cb41e5d6 2339 depth+1,
3dab1dad
YO
2340 (UV)TRIE_NODENUM( state ) );
2341
2342 for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
ab3bbdeb
YO
2343 UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
2344 if (v)
147e3846 2345 Perl_re_printf( aTHX_ "%*" UVXf, colwidth, v );
ab3bbdeb 2346 else
6ad9a8ab 2347 Perl_re_printf( aTHX_ "%*s", colwidth, "." );
3dab1dad
YO
2348 }
2349 if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
147e3846 2350 Perl_re_printf( aTHX_ " (%4" UVXf ")\n",
538e84ed 2351 (UV)trie->trans[ state ].check );
3dab1dad 2352 } else {
147e3846 2353 Perl_re_printf( aTHX_ " (%4" UVXf ") W%4X\n",
538e84ed 2354 (UV)trie->trans[ state ].check,
3dab1dad
YO
2355 trie->states[ TRIE_NODENUM( state ) ].wordnum );
2356 }
2357 }
07be1b83 2358}
3dab1dad
YO
2359
2360#endif
2361
2e64971a 2362
786e8c11
YO
2363/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
2364 startbranch: the first branch in the whole branch sequence
2365 first : start branch of sequence of branch-exact nodes.
2366 May be the same as startbranch
2367 last : Thing following the last branch.
2368 May be the same as tail.
2369 tail : item following the branch sequence
2370 count : words in the sequence
a4525e78 2371 flags : currently the OP() type we will be building one of /EXACT(|F|FA|FU|FU_SS|L|FLU8)/
786e8c11 2372 depth : indent depth
3dab1dad 2373
786e8c11 2374Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
07be1b83 2375
786e8c11
YO
2376A trie is an N'ary tree where the branches are determined by digital
2377decomposition of the key. IE, at the root node you look up the 1st character and
2378follow that branch repeat until you find the end of the branches. Nodes can be
2379marked as "accepting" meaning they represent a complete word. Eg:
07be1b83 2380
786e8c11 2381 /he|she|his|hers/
72f13be8 2382
786e8c11
YO
2383would convert into the following structure. Numbers represent states, letters
2384following numbers represent valid transitions on the letter from that state, if
2385the number is in square brackets it represents an accepting state, otherwise it
2386will be in parenthesis.
07be1b83 2387
786e8c11
YO
2388 +-h->+-e->[3]-+-r->(8)-+-s->[9]
2389 | |
2390 | (2)
2391 | |
2392 (1) +-i->(6)-+-s->[7]
2393 |
2394 +-s->(3)-+-h->(4)-+-e->[5]
07be1b83 2395
786e8c11
YO
2396 Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers)
2397
2398This shows that when matching against the string 'hers' we will begin at state 1
2399read 'h' and move to state 2, read 'e' and move to state 3 which is accepting,
2400then read 'r' and go to state 8 followed by 's' which takes us to state 9 which
2401is also accepting. Thus we know that we can match both 'he' and 'hers' with a
2402single traverse. We store a mapping from accepting to state to which word was
2403matched, and then when we have multiple possibilities we try to complete the
b8fda935 2404rest of the regex in the order in which they occurred in the alternation.
786e8c11
YO
2405
2406The only prior NFA like behaviour that would be changed by the TRIE support is
2407the silent ignoring of duplicate alternations which are of the form:
2408
2409 / (DUPE|DUPE) X? (?{ ... }) Y /x
2410
4b714af6 2411Thus EVAL blocks following a trie may be called a different number of times with
786e8c11 2412and without the optimisation. With the optimisations dupes will be silently
486ec47a 2413ignored. This inconsistent behaviour of EVAL type nodes is well established as
786e8c11
YO
2414the following demonstrates:
2415
2416 'words'=~/(word|word|word)(?{ print $1 })[xyz]/
2417
2418which prints out 'word' three times, but
2419
2420 'words'=~/(word|word|word)(?{ print $1 })S/
2421
2422which doesnt print it out at all. This is due to other optimisations kicking in.
2423
2424Example of what happens on a structural level:
2425
486ec47a 2426The regexp /(ac|ad|ab)+/ will produce the following debug output:
786e8c11
YO
2427
2428 1: CURLYM[1] {1,32767}(18)
2429 5: BRANCH(8)
2430 6: EXACT <ac>(16)
2431 8: BRANCH(11)
2432 9: EXACT <ad>(16)
2433 11: BRANCH(14)
2434 12: EXACT <ab>(16)
2435 16: SUCCEED(0)
2436 17: NOTHING(18)
2437 18: END(0)
2438
2439This would be optimizable with startbranch=5, first=5, last=16, tail=16
2440and should turn into:
2441
2442 1: CURLYM[1] {1,32767}(18)
2443 5: TRIE(16)
2444 [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1]
2445 <ac>
2446 <ad>
2447 <ab>
2448 16: SUCCEED(0)
2449 17: NOTHING(18)
2450 18: END(0)
2451
2452Cases where tail != last would be like /(?foo|bar)baz/:
2453
2454 1: BRANCH(4)
2455 2: EXACT <foo>(8)
2456 4: BRANCH(7)
2457 5: EXACT <bar>(8)
2458 7: TAIL(8)
2459 8: EXACT <baz>(10)
2460 10: END(0)
2461
2462which would be optimizable with startbranch=1, first=1, last=7, tail=8
2463and would end up looking like:
2464
2465 1: TRIE(8)
2466 [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1]
2467 <foo>
2468 <bar>
2469 7: TAIL(8)
2470 8: EXACT <baz>(10)
2471 10: END(0)
2472
c80e42f3 2473 d = uvchr_to_utf8_flags(d, uv, 0);
786e8c11
YO
2474
2475is the recommended Unicode-aware way of saying
2476
2477 *(d++) = uv;
2478*/
2479
fab2782b 2480#define TRIE_STORE_REVCHAR(val) \
786e8c11 2481 STMT_START { \
73031816 2482 if (UTF) { \
668fcfea 2483 SV *zlopp = newSV(UTF8_MAXBYTES); \
88c9ea1e 2484 unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \
c80e42f3 2485 unsigned const char *const kapow = uvchr_to_utf8(flrbbbbb, val); \
73031816
NC
2486 SvCUR_set(zlopp, kapow - flrbbbbb); \
2487 SvPOK_on(zlopp); \
2488 SvUTF8_on(zlopp); \
2489 av_push(revcharmap, zlopp); \
2490 } else { \
fab2782b 2491 char ooooff = (char)val; \
73031816
NC
2492 av_push(revcharmap, newSVpvn(&ooooff, 1)); \
2493 } \
2494 } STMT_END
786e8c11 2495
914a25d5
KW
2496/* This gets the next character from the input, folding it if not already
2497 * folded. */
2498#define TRIE_READ_CHAR STMT_START { \
2499 wordlen++; \
2500 if ( UTF ) { \
2501 /* if it is UTF then it is either already folded, or does not need \
2502 * folding */ \
1c1d615a 2503 uvc = valid_utf8_to_uvchr( (const U8*) uc, &len); \
914a25d5
KW
2504 } \
2505 else if (folder == PL_fold_latin1) { \
7d006b13
KW
2506 /* This folder implies Unicode rules, which in the range expressible \
2507 * by not UTF is the lower case, with the two exceptions, one of \
2508 * which should have been taken care of before calling this */ \
2509 assert(*uc != LATIN_SMALL_LETTER_SHARP_S); \
2510 uvc = toLOWER_L1(*uc); \
2511 if (UNLIKELY(uvc == MICRO_SIGN)) uvc = GREEK_SMALL_LETTER_MU; \
2512 len = 1; \
914a25d5
KW
2513 } else { \
2514 /* raw data, will be folded later if needed */ \
2515 uvc = (U32)*uc; \
2516 len = 1; \
2517 } \
786e8c11
YO
2518} STMT_END
2519
2520
2521
2522#define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \
2523 if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
00195859 2524 U32 ging = TRIE_LIST_LEN( state ) * 2; \
f9003953 2525 Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \
00195859 2526 TRIE_LIST_LEN( state ) = ging; \
786e8c11
YO
2527 } \
2528 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \
2529 TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \
2530 TRIE_LIST_CUR( state )++; \
2531} STMT_END
07be1b83 2532
786e8c11 2533#define TRIE_LIST_NEW(state) STMT_START { \
d09f14bf 2534 Newx( trie->states[ state ].trans.list, \
786e8c11
YO
2535 4, reg_trie_trans_le ); \
2536 TRIE_LIST_CUR( state ) = 1; \
2537 TRIE_LIST_LEN( state ) = 4; \
2538} STMT_END
07be1b83 2539
786e8c11
YO
2540#define TRIE_HANDLE_WORD(state) STMT_START { \
2541 U16 dupe= trie->states[ state ].wordnum; \
2542 regnode * const noper_next = regnext( noper ); \
2543 \
786e8c11
YO
2544 DEBUG_r({ \
2545 /* store the word for dumping */ \
2546 SV* tmp; \
2547 if (OP(noper) != NOTHING) \
740cce10 2548 tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \
786e8c11 2549 else \
740cce10 2550 tmp = newSVpvn_utf8( "", 0, UTF ); \
2b8b4781 2551 av_push( trie_words, tmp ); \
786e8c11
YO
2552 }); \
2553 \
2554 curword++; \
2e64971a
DM
2555 trie->wordinfo[curword].prev = 0; \
2556 trie->wordinfo[curword].len = wordlen; \
2557 trie->wordinfo[curword].accept = state; \
786e8c11
YO
2558 \
2559 if ( noper_next < tail ) { \
2560 if (!trie->jump) \
538e84ed
KW
2561 trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, \
2562 sizeof(U16) ); \
7f69552c 2563 trie->jump[curword] = (U16)(noper_next - convert); \
786e8c11
YO
2564 if (!jumper) \
2565 jumper = noper_next; \
2566 if (!nextbranch) \
2567 nextbranch= regnext(cur); \
2568 } \
2569 \
2570 if ( dupe ) { \
2e64971a
DM
2571 /* It's a dupe. Pre-insert into the wordinfo[].prev */\
2572 /* chain, so that when the bits of chain are later */\
2573 /* linked together, the dups appear in the chain */\
2574 trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \
2575 trie->wordinfo[dupe].prev = curword; \
786e8c11
YO
2576 } else { \
2577 /* we haven't inserted this word yet. */ \
2578 trie->states[ state ].wordnum = curword; \
2579 } \
2580} STMT_END
07be1b83 2581
3dab1dad 2582
786e8c11
YO
2583#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
2584 ( ( base + charid >= ucharcount \
2585 && base + charid < ubound \
2586 && state == trie->trans[ base - ucharcount + charid ].check \
2587 && trie->trans[ base - ucharcount + charid ].next ) \
2588 ? trie->trans[ base - ucharcount + charid ].next \
2589 : ( state==1 ? special : 0 ) \
2590 )
3dab1dad 2591
8bcafbf4
YO
2592#define TRIE_BITMAP_SET_FOLDED(trie, uvc, folder) \
2593STMT_START { \
2594 TRIE_BITMAP_SET(trie, uvc); \
2595 /* store the folded codepoint */ \
2596 if ( folder ) \
2597 TRIE_BITMAP_SET(trie, folder[(U8) uvc ]); \
2598 \
2599 if ( !UTF ) { \
2600 /* store first byte of utf8 representation of */ \
2601 /* variant codepoints */ \
2602 if (! UVCHR_IS_INVARIANT(uvc)) { \
2603 TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); \
2604 } \
2605 } \
2606} STMT_END
786e8c11
YO
2607#define MADE_TRIE 1
2608#define MADE_JUMP_TRIE 2
2609#define MADE_EXACT_TRIE 4
3dab1dad 2610
a3621e74 2611STATIC I32
538e84ed
KW
2612S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch,
2613 regnode *first, regnode *last, regnode *tail,
2614 U32 word_count, U32 flags, U32 depth)
a3621e74
YO
2615{
2616 /* first pass, loop through and scan words */
2617 reg_trie_data *trie;
55eed653 2618 HV *widecharmap = NULL;
2b8b4781 2619 AV *revcharmap = newAV();
a3621e74 2620 regnode *cur;
a3621e74
YO
2621 STRLEN len = 0;
2622 UV uvc = 0;
2623 U16 curword = 0;
2624 U32 next_alloc = 0;
786e8c11
YO
2625 regnode *jumper = NULL;
2626 regnode *nextbranch = NULL;
7f69552c 2627 regnode *convert = NULL;
2e64971a 2628 U32 *prev_states; /* temp array mapping each state to previous one */
a3621e74 2629 /* we just use folder as a flag in utf8 */
1e696034 2630 const U8 * folder = NULL;
a3621e74 2631
3a611511
YO
2632 /* in the below add_data call we are storing either 'tu' or 'tuaa'
2633 * which stands for one trie structure, one hash, optionally followed
2634 * by two arrays */
2b8b4781 2635#ifdef DEBUGGING
3a611511 2636 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tuaa"));
2b8b4781
NC
2637 AV *trie_words = NULL;
2638 /* along with revcharmap, this only used during construction but both are
2639 * useful during debugging so we store them in the struct when debugging.
8e11feef 2640 */
2b8b4781 2641#else
cf78de0b 2642 const U32 data_slot = add_data( pRExC_state, STR_WITH_LEN("tu"));
3dab1dad 2643 STRLEN trie_charcount=0;
3dab1dad 2644#endif
2b8b4781 2645 SV *re_trie_maxbuff;
a3621e74 2646 GET_RE_DEBUG_FLAGS_DECL;
7918f24d
NC
2647
2648 PERL_ARGS_ASSERT_MAKE_TRIE;
72f13be8
YO
2649#ifndef DEBUGGING
2650 PERL_UNUSED_ARG(depth);
2651#endif
a3621e74 2652
1e696034 2653 switch (flags) {
f6b4b99d 2654 case EXACT: case EXACT_ONLY8: case EXACTL: break;
89829bb5 2655 case EXACTFAA:
fab2782b 2656 case EXACTFU_SS:
a4525e78
KW
2657 case EXACTFU:
2658 case EXACTFLU8: folder = PL_fold_latin1; break;
1e696034 2659 case EXACTF: folder = PL_fold; break;
fab2782b 2660 default: Perl_croak( aTHX_ "panic! In trie construction, unknown node type %u %s", (unsigned) flags, PL_reg_name[flags] );
1e696034
KW
2661 }
2662
c944940b 2663 trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
a3621e74 2664 trie->refcount = 1;
3dab1dad 2665 trie->startstate = 1;
786e8c11 2666 trie->wordcount = word_count;
f8fc2ecf 2667 RExC_rxi->data->data[ data_slot ] = (void*)trie;
c944940b 2668 trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) );
f6b4b99d 2669 if (flags == EXACT || flags == EXACT_ONLY8 || flags == EXACTL)
c944940b 2670 trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
2e64971a
DM
2671 trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc(
2672 trie->wordcount+1, sizeof(reg_trie_wordinfo));
2673
a3621e74 2674 DEBUG_r({
2b8b4781 2675 trie_words = newAV();
a3621e74 2676 });
a3621e74 2677
0111c4fd 2678 re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
316ebaf2 2679 assert(re_trie_maxbuff);
a3621e74 2680 if (!SvIOK(re_trie_maxbuff)) {
0111c4fd 2681 sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT);
a3621e74 2682 }
df826430 2683 DEBUG_TRIE_COMPILE_r({
6ad9a8ab 2684 Perl_re_indentf( aTHX_
cb41e5d6
YO
2685 "make_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
2686 depth+1,
88f063b4 2687 REG_NODE_NUM(startbranch), REG_NODE_NUM(first),
538e84ed 2688 REG_NODE_NUM(last), REG_NODE_NUM(tail), (int)depth);
3dab1dad 2689 });
538e84ed 2690
7f69552c
YO
2691 /* Find the node we are going to overwrite */
2692 if ( first == startbranch && OP( last ) != BRANCH ) {
2693 /* whole branch chain */
2694 convert = first;
2695 } else {
2696 /* branch sub-chain */
2697 convert = NEXTOPER( first );
2698 }
538e84ed 2699
a3621e74
YO
2700 /* -- First loop and Setup --
2701
2702 We first traverse the branches and scan each word to determine if it
2703 contains widechars, and how many unique chars there are, this is
2704 important as we have to build a table with at least as many columns as we
2705 have unique chars.
2706
2707 We use an array of integers to represent the character codes 0..255
538e84ed
KW
2708 (trie->charmap) and we use a an HV* to store Unicode characters. We use
2709 the native representation of the character value as the key and IV's for
2710 the coded index.
a3621e74
YO
2711
2712 *TODO* If we keep track of how many times each character is used we can
2713 remap the columns so that the table compression later on is more
3b753521 2714 efficient in terms of memory by ensuring the most common value is in the
a3621e74
YO
2715 middle and the least common are on the outside. IMO this would be better
2716 than a most to least common mapping as theres a decent chance the most
2717 common letter will share a node with the least common, meaning the node
486ec47a 2718 will not be compressible. With a middle is most common approach the worst
a3621e74
YO
2719 case is when we have the least common nodes twice.
2720
2721 */
2722
a3621e74 2723 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
df826430 2724 regnode *noper = NEXTOPER( cur );
944e05e3
YO
2725 const U8 *uc;
2726 const U8 *e;
bc031a7d 2727 int foldlen = 0;
07be1b83 2728 U32 wordlen = 0; /* required init */
bc031a7d
KW
2729 STRLEN minchars = 0;
2730 STRLEN maxchars = 0;
538e84ed
KW
2731 bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the
2732 bitmap?*/
a3621e74 2733
3dab1dad 2734 if (OP(noper) == NOTHING) {
20ed8c88
YO
2735 /* skip past a NOTHING at the start of an alternation
2736 * eg, /(?:)a|(?:b)/ should be the same as /a|b/
2737 */
df826430 2738 regnode *noper_next= regnext(noper);
944e05e3
YO
2739 if (noper_next < tail)
2740 noper= noper_next;
2741 }
2742
f6b4b99d
KW
2743 if ( noper < tail
2744 && ( OP(noper) == flags
2745 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac
KW
2746 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
2747 || OP(noper) == EXACTFU_SS))) )
f6b4b99d 2748 {
944e05e3
YO
2749 uc= (U8*)STRING(noper);
2750 e= uc + STR_LEN(noper);
2751 } else {
2752 trie->minlen= 0;
2753 continue;
3dab1dad 2754 }
df826430 2755
944e05e3 2756
fab2782b 2757 if ( set_bit ) { /* bitmap only alloced when !(UTF&&Folding) */
02daf0ab
YO
2758 TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte
2759 regardless of encoding */
fab2782b
YO
2760 if (OP( noper ) == EXACTFU_SS) {
2761 /* false positives are ok, so just set this */
0dc4a61d 2762 TRIE_BITMAP_SET(trie, LATIN_SMALL_LETTER_SHARP_S);
fab2782b
YO
2763 }
2764 }
dca5fc4c 2765
bc031a7d
KW
2766 for ( ; uc < e ; uc += len ) { /* Look at each char in the current
2767 branch */
3dab1dad 2768 TRIE_CHARCOUNT(trie)++;
a3621e74 2769 TRIE_READ_CHAR;
645de4ce 2770
bc031a7d
KW
2771 /* TRIE_READ_CHAR returns the current character, or its fold if /i
2772 * is in effect. Under /i, this character can match itself, or
2773 * anything that folds to it. If not under /i, it can match just
2774 * itself. Most folds are 1-1, for example k, K, and KELVIN SIGN
2775 * all fold to k, and all are single characters. But some folds
2776 * expand to more than one character, so for example LATIN SMALL
2777 * LIGATURE FFI folds to the three character sequence 'ffi'. If
2778 * the string beginning at 'uc' is 'ffi', it could be matched by
2779 * three characters, or just by the one ligature character. (It
2780 * could also be matched by two characters: LATIN SMALL LIGATURE FF
2781 * followed by 'i', or by 'f' followed by LATIN SMALL LIGATURE FI).
2782 * (Of course 'I' and/or 'F' instead of 'i' and 'f' can also
2783 * match.) The trie needs to know the minimum and maximum number
2784 * of characters that could match so that it can use size alone to
2785 * quickly reject many match attempts. The max is simple: it is
2786 * the number of folded characters in this branch (since a fold is
2787 * never shorter than what folds to it. */
2788
2789 maxchars++;
2790
2791 /* And the min is equal to the max if not under /i (indicated by
2792 * 'folder' being NULL), or there are no multi-character folds. If
2793 * there is a multi-character fold, the min is incremented just
2794 * once, for the character that folds to the sequence. Each
2795 * character in the sequence needs to be added to the list below of
2796 * characters in the trie, but we count only the first towards the
2797 * min number of characters needed. This is done through the
2798 * variable 'foldlen', which is returned by the macros that look
2799 * for these sequences as the number of bytes the sequence
2800 * occupies. Each time through the loop, we decrement 'foldlen' by
2801 * how many bytes the current char occupies. Only when it reaches
2802 * 0 do we increment 'minchars' or look for another multi-character
2803 * sequence. */
2804 if (folder == NULL) {
2805 minchars++;
2806 }
2807 else if (foldlen > 0) {
2808 foldlen -= (UTF) ? UTF8SKIP(uc) : 1;
645de4ce
KW
2809 }
2810 else {
bc031a7d
KW
2811 minchars++;
2812
2813 /* See if *uc is the beginning of a multi-character fold. If
2814 * so, we decrement the length remaining to look at, to account
2815 * for the current character this iteration. (We can use 'uc'
2816 * instead of the fold returned by TRIE_READ_CHAR because for
2817 * non-UTF, the latin1_safe macro is smart enough to account
2818 * for all the unfolded characters, and because for UTF, the
2819 * string will already have been folded earlier in the
2820 * compilation process */
2821 if (UTF) {
2822 if ((foldlen = is_MULTI_CHAR_FOLD_utf8_safe(uc, e))) {
2823 foldlen -= UTF8SKIP(uc);
645de4ce
KW
2824 }
2825 }
bc031a7d
KW
2826 else if ((foldlen = is_MULTI_CHAR_FOLD_latin1_safe(uc, e))) {
2827 foldlen--;
2828 }
645de4ce 2829 }
bc031a7d
KW
2830
2831 /* The current character (and any potential folds) should be added
2832 * to the possible matching characters for this position in this
2833 * branch */
a3621e74 2834 if ( uvc < 256 ) {
fab2782b
YO
2835 if ( folder ) {
2836 U8 folded= folder[ (U8) uvc ];
2837 if ( !trie->charmap[ folded ] ) {
2838 trie->charmap[ folded ]=( ++trie->uniquecharcount );
2839 TRIE_STORE_REVCHAR( folded );
2840 }
2841 }
a3621e74
YO
2842 if ( !trie->charmap[ uvc ] ) {
2843 trie->charmap[ uvc ]=( ++trie->uniquecharcount );
fab2782b 2844 TRIE_STORE_REVCHAR( uvc );
a3621e74 2845 }
02daf0ab 2846 if ( set_bit ) {
62012aee
KW
2847 /* store the codepoint in the bitmap, and its folded
2848 * equivalent. */
8bcafbf4 2849 TRIE_BITMAP_SET_FOLDED(trie, uvc, folder);
02daf0ab
YO
2850 set_bit = 0; /* We've done our bit :-) */
2851 }
a3621e74 2852 } else {
bc031a7d
KW
2853
2854 /* XXX We could come up with the list of code points that fold
2855 * to this using PL_utf8_foldclosures, except not for
2856 * multi-char folds, as there may be multiple combinations
2857 * there that could work, which needs to wait until runtime to
2858 * resolve (The comment about LIGATURE FFI above is such an
2859 * example */
2860
a3621e74 2861 SV** svpp;
55eed653
NC
2862 if ( !widecharmap )
2863 widecharmap = newHV();
a3621e74 2864
55eed653 2865 svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
a3621e74
YO
2866
2867 if ( !svpp )
147e3846 2868 Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%" UVXf, uvc );
a3621e74
YO
2869
2870 if ( !SvTRUE( *svpp ) ) {
2871 sv_setiv( *svpp, ++trie->uniquecharcount );
fab2782b 2872 TRIE_STORE_REVCHAR(uvc);
a3621e74
YO
2873 }
2874 }
bc031a7d
KW
2875 } /* end loop through characters in this branch of the trie */
2876
2877 /* We take the min and max for this branch and combine to find the min
2878 * and max for all branches processed so far */
3dab1dad 2879 if( cur == first ) {
bc031a7d
KW
2880 trie->minlen = minchars;
2881 trie->maxlen = maxchars;
2882 } else if (minchars < trie->minlen) {
2883 trie->minlen = minchars;
2884 } else if (maxchars > trie->maxlen) {
2885 trie->maxlen = maxchars;
fab2782b 2886 }
a3621e74
YO
2887 } /* end first pass */
2888 DEBUG_TRIE_COMPILE_r(
6ad9a8ab 2889 Perl_re_indentf( aTHX_
cb41e5d6
YO
2890 "TRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
2891 depth+1,
55eed653 2892 ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
be8e71aa
YO
2893 (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
2894 (int)trie->minlen, (int)trie->maxlen )
a3621e74 2895 );
a3621e74
YO
2896
2897 /*
2898 We now know what we are dealing with in terms of unique chars and
2899 string sizes so we can calculate how much memory a naive
0111c4fd
RGS
2900 representation using a flat table will take. If it's over a reasonable
2901 limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory
a3621e74
YO
2902 conservative but potentially much slower representation using an array
2903 of lists.
2904
2905 At the end we convert both representations into the same compressed
2906 form that will be used in regexec.c for matching with. The latter
2907 is a form that cannot be used to construct with but has memory
2908 properties similar to the list form and access properties similar
2909 to the table form making it both suitable for fast searches and
2910 small enough that its feasable to store for the duration of a program.
2911
2912 See the comment in the code where the compressed table is produced
2913 inplace from the flat tabe representation for an explanation of how
2914 the compression works.
2915
2916 */
2917
2918
2e64971a
DM
2919 Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32);
2920 prev_states[1] = 0;
2921
538e84ed
KW
2922 if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1)
2923 > SvIV(re_trie_maxbuff) )
2924 {
a3621e74
YO
2925 /*
2926 Second Pass -- Array Of Lists Representation
2927
2928 Each state will be represented by a list of charid:state records
2929 (reg_trie_trans_le) the first such element holds the CUR and LEN
2930 points of the allocated array. (See defines above).
2931
2932 We build the initial structure using the lists, and then convert
2933 it into the compressed table form which allows faster lookups
2934 (but cant be modified once converted).
a3621e74
YO
2935 */
2936
a3621e74
YO
2937 STRLEN transcount = 1;
2938
6ad9a8ab 2939 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using list compiler\n",
cb41e5d6 2940 depth+1));
686b73d4 2941
c944940b
JH
2942 trie->states = (reg_trie_state *)
2943 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
2944 sizeof(reg_trie_state) );
a3621e74
YO
2945 TRIE_LIST_NEW(1);
2946 next_alloc = 2;
2947
2948 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
2949
df826430 2950 regnode *noper = NEXTOPER( cur );
c445ea15
AL
2951 U32 state = 1; /* required init */
2952 U16 charid = 0; /* sanity init */
07be1b83 2953 U32 wordlen = 0; /* required init */
c445ea15 2954
df826430
YO
2955 if (OP(noper) == NOTHING) {
2956 regnode *noper_next= regnext(noper);
944e05e3
YO
2957 if (noper_next < tail)
2958 noper= noper_next;
df826430
YO
2959 }
2960
f6b4b99d
KW
2961 if ( noper < tail
2962 && ( OP(noper) == flags
2963 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac
KW
2964 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
2965 || OP(noper) == EXACTFU_SS))) )
f6b4b99d 2966 {
944e05e3
YO
2967 const U8 *uc= (U8*)STRING(noper);
2968 const U8 *e= uc + STR_LEN(noper);
2969
786e8c11 2970 for ( ; uc < e ; uc += len ) {
c445ea15 2971
786e8c11 2972 TRIE_READ_CHAR;
c445ea15 2973
786e8c11
YO
2974 if ( uvc < 256 ) {
2975 charid = trie->charmap[ uvc ];
c445ea15 2976 } else {
538e84ed
KW
2977 SV** const svpp = hv_fetch( widecharmap,
2978 (char*)&uvc,
2979 sizeof( UV ),
2980 0);
786e8c11
YO
2981 if ( !svpp ) {
2982 charid = 0;
2983 } else {
2984 charid=(U16)SvIV( *svpp );
2985 }
c445ea15 2986 }
538e84ed
KW
2987 /* charid is now 0 if we dont know the char read, or
2988 * nonzero if we do */
786e8c11 2989 if ( charid ) {
a3621e74 2990
786e8c11
YO
2991 U16 check;
2992 U32 newstate = 0;
a3621e74 2993
786e8c11
YO
2994 charid--;
2995 if ( !trie->states[ state ].trans.list ) {
2996 TRIE_LIST_NEW( state );
c445ea15 2997 }
538e84ed
KW
2998 for ( check = 1;
2999 check <= TRIE_LIST_USED( state );
3000 check++ )
3001 {
3002 if ( TRIE_LIST_ITEM( state, check ).forid
3003 == charid )
3004 {
786e8c11
YO
3005 newstate = TRIE_LIST_ITEM( state, check ).newstate;
3006 break;
3007 }
3008 }
3009 if ( ! newstate ) {
3010 newstate = next_alloc++;
2e64971a 3011 prev_states[newstate] = state;
786e8c11
YO
3012 TRIE_LIST_PUSH( state, charid, newstate );
3013 transcount++;
3014 }
3015 state = newstate;
3016 } else {
147e3846 3017 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
c445ea15 3018 }
a28509cc 3019 }
c445ea15 3020 }
3dab1dad 3021 TRIE_HANDLE_WORD(state);
a3621e74
YO
3022
3023 } /* end second pass */
3024
1e2e3d02 3025 /* next alloc is the NEXT state to be allocated */
538e84ed 3026 trie->statecount = next_alloc;
c944940b
JH
3027 trie->states = (reg_trie_state *)
3028 PerlMemShared_realloc( trie->states,
3029 next_alloc
3030 * sizeof(reg_trie_state) );
a3621e74 3031
3dab1dad 3032 /* and now dump it out before we compress it */
2b8b4781
NC
3033 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
3034 revcharmap, next_alloc,
3035 depth+1)
1e2e3d02 3036 );
a3621e74 3037
c944940b
JH
3038 trie->trans = (reg_trie_trans *)
3039 PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
a3621e74
YO
3040 {
3041 U32 state;
a3621e74
YO
3042 U32 tp = 0;
3043 U32 zp = 0;
3044
3045
3046 for( state=1 ; state < next_alloc ; state ++ ) {
3047 U32 base=0;
3048
3049 /*
3050 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3051 Perl_re_printf( aTHX_ "tp: %d zp: %d ",tp,zp)
a3621e74
YO
3052 );
3053 */
3054
3055 if (trie->states[state].trans.list) {
3056 U16 minid=TRIE_LIST_ITEM( state, 1).forid;
3057 U16 maxid=minid;
a28509cc 3058 U16 idx;
a3621e74
YO
3059
3060 for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
c445ea15
AL
3061 const U16 forid = TRIE_LIST_ITEM( state, idx).forid;
3062 if ( forid < minid ) {
3063 minid=forid;
3064 } else if ( forid > maxid ) {
3065 maxid=forid;
3066 }
a3621e74
YO
3067 }
3068 if ( transcount < tp + maxid - minid + 1) {
3069 transcount *= 2;
c944940b
JH
3070 trie->trans = (reg_trie_trans *)
3071 PerlMemShared_realloc( trie->trans,
446bd890
NC
3072 transcount
3073 * sizeof(reg_trie_trans) );
538e84ed
KW
3074 Zero( trie->trans + (transcount / 2),
3075 transcount / 2,
3076 reg_trie_trans );
a3621e74
YO
3077 }
3078 base = trie->uniquecharcount + tp - minid;
3079 if ( maxid == minid ) {
3080 U32 set = 0;
3081 for ( ; zp < tp ; zp++ ) {
3082 if ( ! trie->trans[ zp ].next ) {
3083 base = trie->uniquecharcount + zp - minid;
538e84ed
KW
3084 trie->trans[ zp ].next = TRIE_LIST_ITEM( state,
3085 1).newstate;
a3621e74
YO
3086 trie->trans[ zp ].check = state;
3087 set = 1;
3088 break;
3089 }
3090 }
3091 if ( !set ) {
538e84ed
KW
3092 trie->trans[ tp ].next = TRIE_LIST_ITEM( state,
3093 1).newstate;
a3621e74
YO
3094 trie->trans[ tp ].check = state;
3095 tp++;
3096 zp = tp;
3097 }
3098 } else {
3099 for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) {
538e84ed
KW
3100 const U32 tid = base
3101 - trie->uniquecharcount
3102 + TRIE_LIST_ITEM( state, idx ).forid;
3103 trie->trans[ tid ].next = TRIE_LIST_ITEM( state,
3104 idx ).newstate;
a3621e74
YO
3105 trie->trans[ tid ].check = state;
3106 }
3107 tp += ( maxid - minid + 1 );
3108 }
3109 Safefree(trie->states[ state ].trans.list);
3110 }
3111 /*
3112 DEBUG_TRIE_COMPILE_MORE_r(
6ad9a8ab 3113 Perl_re_printf( aTHX_ " base: %d\n",base);
a3621e74
YO
3114 );
3115 */
3116 trie->states[ state ].trans.base=base;
3117 }
cc601c31 3118 trie->lasttrans = tp + 1;
a3621e74
YO
3119 }
3120 } else {
3121 /*
3122 Second Pass -- Flat Table Representation.
3123
b423522f
KW
3124 we dont use the 0 slot of either trans[] or states[] so we add 1 to
3125 each. We know that we will need Charcount+1 trans at most to store
3126 the data (one row per char at worst case) So we preallocate both
3127 structures assuming worst case.
a3621e74
YO
3128
3129 We then construct the trie using only the .next slots of the entry
3130 structs.
3131
b423522f
KW
3132 We use the .check field of the first entry of the node temporarily
3133 to make compression both faster and easier by keeping track of how
3134 many non zero fields are in the node.
a3621e74
YO
3135
3136 Since trans are numbered from 1 any 0 pointer in the table is a FAIL
3137 transition.
3138
b423522f
KW
3139 There are two terms at use here: state as a TRIE_NODEIDX() which is
3140 a number representing the first entry of the node, and state as a
3141 TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1)
3142 and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3)
3143 if there are 2 entrys per node. eg:
a3621e74
YO
3144
3145 A B A B
3146 1. 2 4 1. 3 7
3147 2. 0 3 3. 0 5
3148 3. 0 0 5. 0 0
3149 4. 0 0 7. 0 0
3150
b423522f
KW
3151 The table is internally in the right hand, idx form. However as we
3152 also have to deal with the states array which is indexed by nodenum
3153 we have to use TRIE_NODENUM() to convert.
a3621e74
YO
3154
3155 */
6ad9a8ab 3156 DEBUG_TRIE_COMPILE_MORE_r( Perl_re_indentf( aTHX_ "Compiling trie using table compiler\n",
cb41e5d6 3157 depth+1));
3dab1dad 3158
c944940b
JH
3159 trie->trans = (reg_trie_trans *)
3160 PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
3161 * trie->uniquecharcount + 1,
3162 sizeof(reg_trie_trans) );
3163 trie->states = (reg_trie_state *)
3164 PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
3165 sizeof(reg_trie_state) );
a3621e74
YO
3166 next_alloc = trie->uniquecharcount + 1;
3167
3dab1dad 3168
a3621e74
YO
3169 for ( cur = first ; cur < last ; cur = regnext( cur ) ) {
3170
df826430 3171 regnode *noper = NEXTOPER( cur );
a3621e74
YO
3172
3173 U32 state = 1; /* required init */
3174
3175 U16 charid = 0; /* sanity init */
3176 U32 accept_state = 0; /* sanity init */
a3621e74 3177
07be1b83 3178 U32 wordlen = 0; /* required init */
a3621e74 3179
df826430
YO
3180 if (OP(noper) == NOTHING) {
3181 regnode *noper_next= regnext(noper);
944e05e3
YO
3182 if (noper_next < tail)
3183 noper= noper_next;
df826430 3184 }
fab2782b 3185
f6b4b99d
KW
3186 if ( noper < tail
3187 && ( OP(noper) == flags
3188 || (flags == EXACT && OP(noper) == EXACT_ONLY8)
a9f8c7ac
KW
3189 || (flags == EXACTFU && ( OP(noper) == EXACTFU_ONLY8
3190 || OP(noper) == EXACTFU_SS))) )
f6b4b99d 3191 {
944e05e3
YO
3192 const U8 *uc= (U8*)STRING(noper);
3193 const U8 *e= uc + STR_LEN(noper);
3194
786e8c11 3195 for ( ; uc < e ; uc += len ) {
a3621e74 3196
786e8c11 3197 TRIE_READ_CHAR;
a3621e74 3198
786e8c11
YO
3199 if ( uvc < 256 ) {
3200 charid = trie->charmap[ uvc ];
3201 } else {
538e84ed
KW
3202 SV* const * const svpp = hv_fetch( widecharmap,
3203 (char*)&uvc,
3204 sizeof( UV ),
3205 0);
786e8c11 3206 charid = svpp ? (U16)SvIV(*svpp) : 0;
a3621e74 3207 }
786e8c11
YO
3208 if ( charid ) {
3209 charid--;
3210 if ( !trie->trans[ state + charid ].next ) {
3211 trie->trans[ state + charid ].next = next_alloc;
3212 trie->trans[ state ].check++;
2e64971a
DM
3213 prev_states[TRIE_NODENUM(next_alloc)]
3214 = TRIE_NODENUM(state);
786e8c11
YO
3215 next_alloc += trie->uniquecharcount;
3216 }
3217 state = trie->trans[ state + charid ].next;
3218 } else {
147e3846 3219 Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %" IVdf, uvc );
786e8c11 3220 }
538e84ed
KW
3221 /* charid is now 0 if we dont know the char read, or
3222 * nonzero if we do */
a3621e74 3223 }
a3621e74 3224 }
3dab1dad
YO
3225 accept_state = TRIE_NODENUM( state );
3226 TRIE_HANDLE_WORD(accept_state);
a3621e74
YO
3227
3228 } /* end second pass */
3229
3dab1dad 3230 /* and now dump it out before we compress it */
2b8b4781
NC
3231 DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
3232 revcharmap,
3233 next_alloc, depth+1));
a3621e74 3234
a3621e74
YO
3235 {
3236 /*
3237 * Inplace compress the table.*
3238
3239 For sparse data sets the table constructed by the trie algorithm will
3240 be mostly 0/FAIL transitions or to put it another way mostly empty.
3241 (Note that leaf nodes will not contain any transitions.)
3242
3243 This algorithm compresses the tables by eliminating most such
3244 transitions, at the cost of a modest bit of extra work during lookup:
3245
3246 - Each states[] entry contains a .base field which indicates the
3247 index in the state[] array wheres its transition data is stored.
3248
3b753521 3249 - If .base is 0 there are no valid transitions from that node.
a3621e74
YO
3250
3251 - If .base is nonzero then charid is added to it to find an entry in
3252 the trans array.
3253
3254 -If trans[states[state].base+charid].check!=state then the
3255 transition is taken to be a 0/Fail transition. Thus if there are fail
3256 transitions at the front of the node then the .base offset will point
3257 somewhere inside the previous nodes data (or maybe even into a node
3258 even earlier), but the .check field determines if the transition is
3259 valid.
3260
786e8c11 3261 XXX - wrong maybe?
a3621e74 3262 The following process inplace converts the table to the compressed
3b753521 3263 table: We first do not compress the root node 1,and mark all its
a3621e74 3264 .check pointers as 1 and set its .base pointer as 1 as well. This
3b753521
FN
3265 allows us to do a DFA construction from the compressed table later,
3266 and ensures that any .base pointers we calculate later are greater
3267 than 0.
a3621e74
YO
3268
3269 - We set 'pos' to indicate the first entry of the second node.
3270
3271 - We then iterate over the columns of the node, finding the first and
3272 last used entry at l and m. We then copy l..m into pos..(pos+m-l),
3273 and set the .check pointers accordingly, and advance pos
3274 appropriately and repreat for the next node. Note that when we copy
3275 the next pointers we have to convert them from the original
3276 NODEIDX form to NODENUM form as the former is not valid post
3277 compression.
3278
3279 - If a node has no transitions used we mark its base as 0 and do not
3280 advance the pos pointer.
3281
3282 - If a node only has one transition we use a second pointer into the
3283 structure to fill in allocated fail transitions from other states.
3284 This pointer is independent of the main pointer and scans forward
3285 looking for null transitions that are allocated to a state. When it
3286 finds one it writes the single transition into the "hole". If the
786e8c11 3287 pointer doesnt find one the single transition is appended as normal.
a3621e74
YO
3288
3289 - Once compressed we can Renew/realloc the structures to release the
3290 excess space.
3291
3292 See "Table-Compression Methods" in sec 3.9 of the Red Dragon,
3293 specifically Fig 3.47 and the associated pseudocode.
3294
3295 demq
3296 */
a3b680e6 3297 const U32 laststate = TRIE_NODENUM( next_alloc );
a28509cc 3298 U32 state, charid;
a3621e74 3299 U32 pos = 0, zp=0;
1e2e3d02 3300 trie->statecount = laststate;
a3621e74
YO
3301
3302 for ( state = 1 ; state < laststate ; state++ ) {
3303 U8 flag = 0;
a28509cc
AL
3304 const U32 stateidx = TRIE_NODEIDX( state );
3305 const U32 o_used = trie->trans[ stateidx ].check;
3306 U32 used = trie->trans[ stateidx ].check;
a3621e74
YO
3307 trie->trans[ stateidx ].check = 0;
3308
538e84ed
KW
3309 for ( charid = 0;
3310 used && charid < trie->uniquecharcount;
3311 charid++ )
3312 {
a3621e74
YO
3313 if ( flag || trie->trans[ stateidx + charid ].next ) {
3314 if ( trie->trans[ stateidx + charid ].next ) {
3315 if (o_used == 1) {
3316 for ( ; zp < pos ; zp++ ) {
3317 if ( ! trie->trans[ zp ].next ) {
3318 break;
3319 }
3320 }
538e84ed
KW
3321 trie->states[ state ].trans.base
3322 = zp
3323 + trie->uniquecharcount
3324 - charid ;
3325 trie->trans[ zp ].next
3326 = SAFE_TRIE_NODENUM( trie->trans[ stateidx
3327 + charid ].next );
a3621e74
YO
3328 trie->trans[ zp ].check = state;
3329 if ( ++zp > pos ) pos = zp;
3330 break;
3331 }
3332 used--;
3333 }
3334 if ( !flag ) {
3335 flag = 1;
538e84ed
KW
3336 trie->states[ state ].trans.base
3337 = pos + trie->uniquecharcount - charid ;
a3621e74 3338 }
538e84ed
KW
3339 trie->trans[ pos ].next
3340 = SAFE_TRIE_NODENUM(
3341 trie->trans[ stateidx + charid ].next );
a3621e74
YO
3342 trie->trans[ pos ].check = state;
3343 pos++;
3344 }
3345 }
3346 }
cc601c31 3347 trie->lasttrans = pos + 1;
c944940b
JH
3348 trie->states = (reg_trie_state *)
3349 PerlMemShared_realloc( trie->states, laststate
3350 * sizeof(reg_trie_state) );
a3621e74 3351 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3352 Perl_re_indentf( aTHX_ "Alloc: %d Orig: %" IVdf " elements, Final:%" IVdf ". Savings of %%%5.2f\n",
cb41e5d6 3353 depth+1,
538e84ed
KW
3354 (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount
3355 + 1 ),
3356 (IV)next_alloc,
3357 (IV)pos,
3358 ( ( next_alloc - pos ) * 100 ) / (double)next_alloc );
a3621e74
YO
3359 );
3360
3361 } /* end table compress */
3362 }
1e2e3d02 3363 DEBUG_TRIE_COMPILE_MORE_r(
147e3846 3364 Perl_re_indentf( aTHX_ "Statecount:%" UVxf " Lasttrans:%" UVxf "\n",
cb41e5d6 3365 depth+1,
1e2e3d02
YO
3366 (UV)trie->statecount,
3367 (UV)trie->lasttrans)
3368 );
cc601c31 3369 /* resize the trans array to remove unused space */
c944940b
JH
3370 trie->trans = (reg_trie_trans *)
3371 PerlMemShared_realloc( trie->trans, trie->lasttrans
3372 * sizeof(reg_trie_trans) );
a3621e74 3373
538e84ed 3374 { /* Modify the program and insert the new TRIE node */
3dab1dad
YO
3375 U8 nodetype =(U8)(flags & 0xFF);
3376 char *str=NULL;
538e84ed 3377
07be1b83 3378#ifdef DEBUGGING
e62cc96a 3379 regnode *optimize = NULL;
7122b237
YO
3380#ifdef RE_TRACK_PATTERN_OFFSETS
3381
b57a0404
JH
3382 U32 mjd_offset = 0;
3383 U32 mjd_nodelen = 0;
7122b237
YO
3384#endif /* RE_TRACK_PATTERN_OFFSETS */
3385#endif /* DEBUGGING */
a3621e74 3386 /*
3dab1dad
YO
3387 This means we convert either the first branch or the first Exact,
3388 depending on whether the thing following (in 'last') is a branch
3389 or not and whther first is the startbranch (ie is it a sub part of
3390 the alternation or is it the whole thing.)
3b753521 3391 Assuming its a sub part we convert the EXACT otherwise we convert
3dab1dad 3392 the whole branch sequence, including the first.
a3621e74 3393 */
3dab1dad 3394 /* Find the node we are going to overwrite */
7f69552c 3395 if ( first != startbranch || OP( last ) == BRANCH ) {
07be1b83 3396 /* branch sub-chain */
3dab1dad 3397 NEXT_OFF( first ) = (U16)(last - first);
7122b237 3398#ifdef RE_TRACK_PATTERN_OFFSETS
07be1b83
YO
3399 DEBUG_r({
3400 mjd_offset= Node_Offset((convert));
3401 mjd_nodelen= Node_Length((convert));
3402 });
7122b237 3403#endif
7f69552c 3404 /* whole branch chain */
7122b237
YO
3405 }
3406#ifdef RE_TRACK_PATTERN_OFFSETS
3407 else {
7f69552c
YO
3408 DEBUG_r({
3409 const regnode *nop = NEXTOPER( convert );
3410 mjd_offset= Node_Offset((nop));
3411 mjd_nodelen= Node_Length((nop));
3412 });
07be1b83
YO
3413 }
3414 DEBUG_OPTIMISE_r(
147e3846 3415 Perl_re_indentf( aTHX_ "MJD offset:%" UVuf " MJD length:%" UVuf "\n",
cb41e5d6 3416 depth+1,
786e8c11 3417 (UV)mjd_offset, (UV)mjd_nodelen)
07be1b83 3418 );
7122b237 3419#endif
538e84ed 3420 /* But first we check to see if there is a common prefix we can
3dab1dad
YO
3421 split out as an EXACT and put in front of the TRIE node. */
3422 trie->startstate= 1;
55eed653 3423 if ( trie->bitmap && !widecharmap && !trie->jump ) {
5ee57374
YO
3424 /* we want to find the first state that has more than
3425 * one transition, if that state is not the first state
3426 * then we have a common prefix which we can remove.
3427 */
3dab1dad 3428 U32 state;
1e2e3d02 3429 for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
a3621e74