Commit | Line | Data |
---|---|---|
a0d0e21e LW |
1 | /* regcomp.c |
2 | */ | |
3 | ||
4 | /* | |
4ac71550 TC |
5 | * 'A fair jaw-cracker dwarf-language must be.' --Samwise Gamgee |
6 | * | |
7 | * [p.285 of _The Lord of the Rings_, II/iii: "The Ring Goes South"] | |
a0d0e21e LW |
8 | */ |
9 | ||
61296642 DM |
10 | /* This file contains functions for compiling a regular expression. See |
11 | * also regexec.c which funnily enough, contains functions for executing | |
166f8a29 | 12 | * a regular expression. |
e4a054ea DM |
13 | * |
14 | * This file is also copied at build time to ext/re/re_comp.c, where | |
15 | * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT. | |
16 | * This causes the main functions to be compiled under new names and with | |
17 | * debugging support added, which makes "use re 'debug'" work. | |
166f8a29 DM |
18 | */ |
19 | ||
a687059c LW |
20 | /* NOTE: this is derived from Henry Spencer's regexp code, and should not |
21 | * confused with the original package (see point 3 below). Thanks, Henry! | |
22 | */ | |
23 | ||
24 | /* Additional note: this code is very heavily munged from Henry's version | |
25 | * in places. In some spots I've traded clarity for efficiency, so don't | |
26 | * blame Henry for some of the lack of readability. | |
27 | */ | |
28 | ||
e50aee73 | 29 | /* The names of the functions have been changed from regcomp and |
3b753521 | 30 | * regexec to pregcomp and pregexec in order to avoid conflicts |
e50aee73 AD |
31 | * with the POSIX routines of the same names. |
32 | */ | |
33 | ||
b9d5759e | 34 | #ifdef PERL_EXT_RE_BUILD |
54df2634 | 35 | #include "re_top.h" |
b81d288d | 36 | #endif |
56953603 | 37 | |
a687059c | 38 | /* |
e50aee73 | 39 | * pregcomp and pregexec -- regsub and regerror are not used in perl |
a687059c LW |
40 | * |
41 | * Copyright (c) 1986 by University of Toronto. | |
42 | * Written by Henry Spencer. Not derived from licensed software. | |
43 | * | |
44 | * Permission is granted to anyone to use this software for any | |
45 | * purpose on any computer system, and to redistribute it freely, | |
46 | * subject to the following restrictions: | |
47 | * | |
48 | * 1. The author is not responsible for the consequences of use of | |
49 | * this software, no matter how awful, even if they arise | |
50 | * from defects in it. | |
51 | * | |
52 | * 2. The origin of this software must not be misrepresented, either | |
53 | * by explicit claim or by omission. | |
54 | * | |
55 | * 3. Altered versions must be plainly marked as such, and must not | |
56 | * be misrepresented as being the original software. | |
57 | * | |
58 | * | |
59 | **** Alterations to Henry's code are... | |
60 | **** | |
4bb101f2 | 61 | **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, |
1129b882 NC |
62 | **** 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008 |
63 | **** by Larry Wall and others | |
a687059c | 64 | **** |
9ef589d8 LW |
65 | **** You may distribute under the terms of either the GNU General Public |
66 | **** License or the Artistic License, as specified in the README file. | |
67 | ||
a687059c LW |
68 | * |
69 | * Beware that some of this code is subtly aware of the way operator | |
70 | * precedence is structured in regular expressions. Serious changes in | |
71 | * regular-expression syntax might require a total rethink. | |
72 | */ | |
73 | #include "EXTERN.h" | |
864dbfa3 | 74 | #define PERL_IN_REGCOMP_C |
a687059c | 75 | #include "perl.h" |
d06ea78c | 76 | |
acfe0abc | 77 | #ifndef PERL_IN_XSUB_RE |
d06ea78c GS |
78 | # include "INTERN.h" |
79 | #endif | |
c277df42 IZ |
80 | |
81 | #define REG_COMP_C | |
54df2634 NC |
82 | #ifdef PERL_IN_XSUB_RE |
83 | # include "re_comp.h" | |
84 | #else | |
85 | # include "regcomp.h" | |
86 | #endif | |
a687059c | 87 | |
04e98a4d AD |
88 | #include "dquote_static.c" |
89 | ||
d4cce5f1 | 90 | #ifdef op |
11343788 | 91 | #undef op |
d4cce5f1 | 92 | #endif /* op */ |
11343788 | 93 | |
fe14fcc3 | 94 | #ifdef MSDOS |
7e4e8c89 | 95 | # if defined(BUGGY_MSC6) |
fe14fcc3 | 96 | /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */ |
7e4e8c89 | 97 | # pragma optimize("a",off) |
fe14fcc3 | 98 | /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/ |
7e4e8c89 NC |
99 | # pragma optimize("w",on ) |
100 | # endif /* BUGGY_MSC6 */ | |
fe14fcc3 LW |
101 | #endif /* MSDOS */ |
102 | ||
a687059c LW |
103 | #ifndef STATIC |
104 | #define STATIC static | |
105 | #endif | |
106 | ||
830247a4 | 107 | typedef struct RExC_state_t { |
e2509266 | 108 | U32 flags; /* are we folding, multilining? */ |
830247a4 | 109 | char *precomp; /* uncompiled string. */ |
288b8c02 | 110 | REGEXP *rx_sv; /* The SV that is the regexp. */ |
f8fc2ecf YO |
111 | regexp *rx; /* perl core regexp structure */ |
112 | regexp_internal *rxi; /* internal data for regexp object pprivate field */ | |
fac92740 | 113 | char *start; /* Start of input for compile */ |
830247a4 IZ |
114 | char *end; /* End of input for compile */ |
115 | char *parse; /* Input-scan pointer. */ | |
116 | I32 whilem_seen; /* number of WHILEM in this expr */ | |
fac92740 | 117 | regnode *emit_start; /* Start of emitted-code area */ |
3b57cd43 | 118 | regnode *emit_bound; /* First regnode outside of the allocated space */ |
ffc61ed2 | 119 | regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */ |
830247a4 IZ |
120 | I32 naughty; /* How bad is this pattern? */ |
121 | I32 sawback; /* Did we see \1, ...? */ | |
122 | U32 seen; | |
123 | I32 size; /* Code size. */ | |
c74340f9 YO |
124 | I32 npar; /* Capture buffer count, (OPEN). */ |
125 | I32 cpar; /* Capture buffer count, (CLOSE). */ | |
e2e6a0f1 | 126 | I32 nestroot; /* root parens we are in - used by accept */ |
830247a4 IZ |
127 | I32 extralen; |
128 | I32 seen_zerolen; | |
129 | I32 seen_evals; | |
40d049e4 YO |
130 | regnode **open_parens; /* pointers to open parens */ |
131 | regnode **close_parens; /* pointers to close parens */ | |
132 | regnode *opend; /* END node in program */ | |
02daf0ab YO |
133 | I32 utf8; /* whether the pattern is utf8 or not */ |
134 | I32 orig_utf8; /* whether the pattern was originally in utf8 */ | |
135 | /* XXX use this for future optimisation of case | |
136 | * where pattern must be upgraded to utf8. */ | |
e40e74fe KW |
137 | I32 uni_semantics; /* If a d charset modifier should use unicode |
138 | rules, even if the pattern is not in | |
139 | utf8 */ | |
81714fb9 | 140 | HV *paren_names; /* Paren names */ |
1f1031fe | 141 | |
40d049e4 YO |
142 | regnode **recurse; /* Recurse regops */ |
143 | I32 recurse_count; /* Number of recurse regops */ | |
b57e4118 | 144 | I32 in_lookbehind; |
4624b182 | 145 | I32 contains_locale; |
bb3f3ed2 | 146 | I32 override_recoding; |
830247a4 IZ |
147 | #if ADD_TO_REGEXEC |
148 | char *starttry; /* -Dr: where regtry was called. */ | |
149 | #define RExC_starttry (pRExC_state->starttry) | |
150 | #endif | |
3dab1dad | 151 | #ifdef DEBUGGING |
be8e71aa | 152 | const char *lastparse; |
3dab1dad | 153 | I32 lastnum; |
1f1031fe | 154 | AV *paren_name_list; /* idx -> name */ |
3dab1dad YO |
155 | #define RExC_lastparse (pRExC_state->lastparse) |
156 | #define RExC_lastnum (pRExC_state->lastnum) | |
1f1031fe | 157 | #define RExC_paren_name_list (pRExC_state->paren_name_list) |
3dab1dad | 158 | #endif |
830247a4 IZ |
159 | } RExC_state_t; |
160 | ||
e2509266 | 161 | #define RExC_flags (pRExC_state->flags) |
830247a4 | 162 | #define RExC_precomp (pRExC_state->precomp) |
288b8c02 | 163 | #define RExC_rx_sv (pRExC_state->rx_sv) |
830247a4 | 164 | #define RExC_rx (pRExC_state->rx) |
f8fc2ecf | 165 | #define RExC_rxi (pRExC_state->rxi) |
fac92740 | 166 | #define RExC_start (pRExC_state->start) |
830247a4 IZ |
167 | #define RExC_end (pRExC_state->end) |
168 | #define RExC_parse (pRExC_state->parse) | |
169 | #define RExC_whilem_seen (pRExC_state->whilem_seen) | |
7122b237 YO |
170 | #ifdef RE_TRACK_PATTERN_OFFSETS |
171 | #define RExC_offsets (pRExC_state->rxi->u.offsets) /* I am not like the others */ | |
172 | #endif | |
830247a4 | 173 | #define RExC_emit (pRExC_state->emit) |
fac92740 | 174 | #define RExC_emit_start (pRExC_state->emit_start) |
3b57cd43 | 175 | #define RExC_emit_bound (pRExC_state->emit_bound) |
830247a4 IZ |
176 | #define RExC_naughty (pRExC_state->naughty) |
177 | #define RExC_sawback (pRExC_state->sawback) | |
178 | #define RExC_seen (pRExC_state->seen) | |
179 | #define RExC_size (pRExC_state->size) | |
180 | #define RExC_npar (pRExC_state->npar) | |
e2e6a0f1 | 181 | #define RExC_nestroot (pRExC_state->nestroot) |
830247a4 IZ |
182 | #define RExC_extralen (pRExC_state->extralen) |
183 | #define RExC_seen_zerolen (pRExC_state->seen_zerolen) | |
184 | #define RExC_seen_evals (pRExC_state->seen_evals) | |
1aa99e6b | 185 | #define RExC_utf8 (pRExC_state->utf8) |
e40e74fe | 186 | #define RExC_uni_semantics (pRExC_state->uni_semantics) |
02daf0ab | 187 | #define RExC_orig_utf8 (pRExC_state->orig_utf8) |
40d049e4 YO |
188 | #define RExC_open_parens (pRExC_state->open_parens) |
189 | #define RExC_close_parens (pRExC_state->close_parens) | |
190 | #define RExC_opend (pRExC_state->opend) | |
81714fb9 | 191 | #define RExC_paren_names (pRExC_state->paren_names) |
40d049e4 YO |
192 | #define RExC_recurse (pRExC_state->recurse) |
193 | #define RExC_recurse_count (pRExC_state->recurse_count) | |
b57e4118 | 194 | #define RExC_in_lookbehind (pRExC_state->in_lookbehind) |
4624b182 | 195 | #define RExC_contains_locale (pRExC_state->contains_locale) |
bb3f3ed2 | 196 | #define RExC_override_recoding (pRExC_state->override_recoding) |
830247a4 | 197 | |
cde0cee5 | 198 | |
a687059c LW |
199 | #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?') |
200 | #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \ | |
201 | ((*s) == '{' && regcurly(s))) | |
a687059c | 202 | |
35c8bce7 LW |
203 | #ifdef SPSTART |
204 | #undef SPSTART /* dratted cpp namespace... */ | |
205 | #endif | |
a687059c LW |
206 | /* |
207 | * Flags to be passed up and down. | |
208 | */ | |
a687059c | 209 | #define WORST 0 /* Worst case. */ |
a3b492c3 | 210 | #define HASWIDTH 0x01 /* Known to match non-null strings. */ |
fda99bee KW |
211 | |
212 | /* Simple enough to be STAR/PLUS operand, in an EXACT node must be a single | |
d7b56a3c | 213 | * character, and if utf8, must be invariant. Note that this is not the same thing as REGNODE_SIMPLE */ |
fda99bee | 214 | #define SIMPLE 0x02 |
a3b492c3 YO |
215 | #define SPSTART 0x04 /* Starts with * or +. */ |
216 | #define TRYAGAIN 0x08 /* Weeded out a declaration. */ | |
217 | #define POSTPONED 0x10 /* (?1),(?&name), (??{...}) or similar */ | |
a687059c | 218 | |
3dab1dad YO |
219 | #define REG_NODE_NUM(x) ((x) ? (int)((x)-RExC_emit_start) : -1) |
220 | ||
07be1b83 YO |
221 | /* whether trie related optimizations are enabled */ |
222 | #if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION | |
223 | #define TRIE_STUDY_OPT | |
786e8c11 | 224 | #define FULL_TRIE_STUDY |
07be1b83 YO |
225 | #define TRIE_STCLASS |
226 | #endif | |
1de06328 YO |
227 | |
228 | ||
40d049e4 YO |
229 | |
230 | #define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3] | |
231 | #define PBITVAL(paren) (1 << ((paren) & 7)) | |
232 | #define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren)) | |
233 | #define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren) | |
234 | #define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren)) | |
235 | ||
bbd61b5f KW |
236 | /* If not already in utf8, do a longjmp back to the beginning */ |
237 | #define UTF8_LONGJMP 42 /* Choose a value not likely to ever conflict */ | |
238 | #define REQUIRE_UTF8 STMT_START { \ | |
239 | if (! UTF) JMPENV_JUMP(UTF8_LONGJMP); \ | |
240 | } STMT_END | |
40d049e4 | 241 | |
1de06328 YO |
242 | /* About scan_data_t. |
243 | ||
244 | During optimisation we recurse through the regexp program performing | |
245 | various inplace (keyhole style) optimisations. In addition study_chunk | |
246 | and scan_commit populate this data structure with information about | |
247 | what strings MUST appear in the pattern. We look for the longest | |
3b753521 | 248 | string that must appear at a fixed location, and we look for the |
1de06328 YO |
249 | longest string that may appear at a floating location. So for instance |
250 | in the pattern: | |
251 | ||
252 | /FOO[xX]A.*B[xX]BAR/ | |
253 | ||
254 | Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating | |
255 | strings (because they follow a .* construct). study_chunk will identify | |
256 | both FOO and BAR as being the longest fixed and floating strings respectively. | |
257 | ||
258 | The strings can be composites, for instance | |
259 | ||
260 | /(f)(o)(o)/ | |
261 | ||
262 | will result in a composite fixed substring 'foo'. | |
263 | ||
264 | For each string some basic information is maintained: | |
265 | ||
266 | - offset or min_offset | |
267 | This is the position the string must appear at, or not before. | |
268 | It also implicitly (when combined with minlenp) tells us how many | |
3b753521 FN |
269 | characters must match before the string we are searching for. |
270 | Likewise when combined with minlenp and the length of the string it | |
1de06328 YO |
271 | tells us how many characters must appear after the string we have |
272 | found. | |
273 | ||
274 | - max_offset | |
275 | Only used for floating strings. This is the rightmost point that | |
3b753521 | 276 | the string can appear at. If set to I32 max it indicates that the |
1de06328 YO |
277 | string can occur infinitely far to the right. |
278 | ||
279 | - minlenp | |
280 | A pointer to the minimum length of the pattern that the string | |
281 | was found inside. This is important as in the case of positive | |
282 | lookahead or positive lookbehind we can have multiple patterns | |
283 | involved. Consider | |
284 | ||
285 | /(?=FOO).*F/ | |
286 | ||
287 | The minimum length of the pattern overall is 3, the minimum length | |
288 | of the lookahead part is 3, but the minimum length of the part that | |
289 | will actually match is 1. So 'FOO's minimum length is 3, but the | |
290 | minimum length for the F is 1. This is important as the minimum length | |
291 | is used to determine offsets in front of and behind the string being | |
292 | looked for. Since strings can be composites this is the length of the | |
486ec47a | 293 | pattern at the time it was committed with a scan_commit. Note that |
1de06328 YO |
294 | the length is calculated by study_chunk, so that the minimum lengths |
295 | are not known until the full pattern has been compiled, thus the | |
296 | pointer to the value. | |
297 | ||
298 | - lookbehind | |
299 | ||
300 | In the case of lookbehind the string being searched for can be | |
301 | offset past the start point of the final matching string. | |
302 | If this value was just blithely removed from the min_offset it would | |
303 | invalidate some of the calculations for how many chars must match | |
304 | before or after (as they are derived from min_offset and minlen and | |
305 | the length of the string being searched for). | |
306 | When the final pattern is compiled and the data is moved from the | |
307 | scan_data_t structure into the regexp structure the information | |
308 | about lookbehind is factored in, with the information that would | |
309 | have been lost precalculated in the end_shift field for the | |
310 | associated string. | |
311 | ||
312 | The fields pos_min and pos_delta are used to store the minimum offset | |
313 | and the delta to the maximum offset at the current point in the pattern. | |
314 | ||
315 | */ | |
2c2d71f5 JH |
316 | |
317 | typedef struct scan_data_t { | |
1de06328 YO |
318 | /*I32 len_min; unused */ |
319 | /*I32 len_delta; unused */ | |
2c2d71f5 JH |
320 | I32 pos_min; |
321 | I32 pos_delta; | |
322 | SV *last_found; | |
1de06328 | 323 | I32 last_end; /* min value, <0 unless valid. */ |
2c2d71f5 JH |
324 | I32 last_start_min; |
325 | I32 last_start_max; | |
1de06328 YO |
326 | SV **longest; /* Either &l_fixed, or &l_float. */ |
327 | SV *longest_fixed; /* longest fixed string found in pattern */ | |
328 | I32 offset_fixed; /* offset where it starts */ | |
486ec47a | 329 | I32 *minlen_fixed; /* pointer to the minlen relevant to the string */ |
1de06328 YO |
330 | I32 lookbehind_fixed; /* is the position of the string modfied by LB */ |
331 | SV *longest_float; /* longest floating string found in pattern */ | |
332 | I32 offset_float_min; /* earliest point in string it can appear */ | |
333 | I32 offset_float_max; /* latest point in string it can appear */ | |
486ec47a | 334 | I32 *minlen_float; /* pointer to the minlen relevant to the string */ |
1de06328 | 335 | I32 lookbehind_float; /* is the position of the string modified by LB */ |
2c2d71f5 JH |
336 | I32 flags; |
337 | I32 whilem_c; | |
cb434fcc | 338 | I32 *last_closep; |
653099ff | 339 | struct regnode_charclass_class *start_class; |
2c2d71f5 JH |
340 | } scan_data_t; |
341 | ||
a687059c | 342 | /* |
e50aee73 | 343 | * Forward declarations for pregcomp()'s friends. |
a687059c | 344 | */ |
a0d0e21e | 345 | |
27da23d5 | 346 | static const scan_data_t zero_scan_data = |
1de06328 | 347 | { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 ,0}; |
c277df42 IZ |
348 | |
349 | #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL) | |
07be1b83 YO |
350 | #define SF_BEFORE_SEOL 0x0001 |
351 | #define SF_BEFORE_MEOL 0x0002 | |
c277df42 IZ |
352 | #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL) |
353 | #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL) | |
354 | ||
09b7f37c CB |
355 | #ifdef NO_UNARY_PLUS |
356 | # define SF_FIX_SHIFT_EOL (0+2) | |
357 | # define SF_FL_SHIFT_EOL (0+4) | |
358 | #else | |
359 | # define SF_FIX_SHIFT_EOL (+2) | |
360 | # define SF_FL_SHIFT_EOL (+4) | |
361 | #endif | |
c277df42 IZ |
362 | |
363 | #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL) | |
364 | #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL) | |
365 | ||
366 | #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL) | |
367 | #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */ | |
07be1b83 YO |
368 | #define SF_IS_INF 0x0040 |
369 | #define SF_HAS_PAR 0x0080 | |
370 | #define SF_IN_PAR 0x0100 | |
371 | #define SF_HAS_EVAL 0x0200 | |
372 | #define SCF_DO_SUBSTR 0x0400 | |
653099ff GS |
373 | #define SCF_DO_STCLASS_AND 0x0800 |
374 | #define SCF_DO_STCLASS_OR 0x1000 | |
375 | #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR) | |
e1901655 | 376 | #define SCF_WHILEM_VISITED_POS 0x2000 |
c277df42 | 377 | |
786e8c11 | 378 | #define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */ |
e2e6a0f1 | 379 | #define SCF_SEEN_ACCEPT 0x8000 |
07be1b83 | 380 | |
43fead97 | 381 | #define UTF cBOOL(RExC_utf8) |
a62b1201 KW |
382 | #define LOC (get_regex_charset(RExC_flags) == REGEX_LOCALE_CHARSET) |
383 | #define UNI_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_UNICODE_CHARSET) | |
cfaf538b KW |
384 | #define DEPENDS_SEMANTICS (get_regex_charset(RExC_flags) == REGEX_DEPENDS_CHARSET) |
385 | #define AT_LEAST_UNI_SEMANTICS (get_regex_charset(RExC_flags) >= REGEX_UNICODE_CHARSET) | |
386 | #define ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_RESTRICTED_CHARSET) | |
2f7f8cb1 KW |
387 | #define MORE_ASCII_RESTRICTED (get_regex_charset(RExC_flags) == REGEX_ASCII_MORE_RESTRICTED_CHARSET) |
388 | #define AT_LEAST_ASCII_RESTRICTED (get_regex_charset(RExC_flags) >= REGEX_ASCII_RESTRICTED_CHARSET) | |
a62b1201 | 389 | |
43fead97 | 390 | #define FOLD cBOOL(RExC_flags & RXf_PMf_FOLD) |
a0ed51b3 | 391 | |
ffc61ed2 | 392 | #define OOB_UNICODE 12345678 |
93733859 | 393 | #define OOB_NAMEDCLASS -1 |
b8c5462f | 394 | |
a0ed51b3 LW |
395 | #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv)) |
396 | #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b) | |
397 | ||
8615cb43 | 398 | |
b45f050a JF |
399 | /* length of regex to show in messages that don't mark a position within */ |
400 | #define RegexLengthToShowInErrorMessages 127 | |
401 | ||
402 | /* | |
403 | * If MARKER[12] are adjusted, be sure to adjust the constants at the top | |
404 | * of t/op/regmesg.t, the tests in t/op/re_tests, and those in | |
405 | * op/pragma/warn/regcomp. | |
406 | */ | |
7253e4e3 RK |
407 | #define MARKER1 "<-- HERE" /* marker as it appears in the description */ |
408 | #define MARKER2 " <-- HERE " /* marker as it appears within the regex */ | |
b81d288d | 409 | |
7253e4e3 | 410 | #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/" |
b45f050a JF |
411 | |
412 | /* | |
413 | * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given | |
414 | * arg. Show regex, up to a maximum length. If it's too long, chop and add | |
415 | * "...". | |
416 | */ | |
58e23c8d | 417 | #define _FAIL(code) STMT_START { \ |
bfed75c6 | 418 | const char *ellipses = ""; \ |
ccb2c380 MP |
419 | IV len = RExC_end - RExC_precomp; \ |
420 | \ | |
421 | if (!SIZE_ONLY) \ | |
288b8c02 | 422 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
423 | if (len > RegexLengthToShowInErrorMessages) { \ |
424 | /* chop 10 shorter than the max, to ensure meaning of "..." */ \ | |
425 | len = RegexLengthToShowInErrorMessages - 10; \ | |
426 | ellipses = "..."; \ | |
427 | } \ | |
58e23c8d | 428 | code; \ |
ccb2c380 | 429 | } STMT_END |
8615cb43 | 430 | |
58e23c8d YO |
431 | #define FAIL(msg) _FAIL( \ |
432 | Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \ | |
433 | msg, (int)len, RExC_precomp, ellipses)) | |
434 | ||
435 | #define FAIL2(msg,arg) _FAIL( \ | |
436 | Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \ | |
437 | arg, (int)len, RExC_precomp, ellipses)) | |
438 | ||
b45f050a | 439 | /* |
b45f050a JF |
440 | * Simple_vFAIL -- like FAIL, but marks the current location in the scan |
441 | */ | |
ccb2c380 | 442 | #define Simple_vFAIL(m) STMT_START { \ |
a28509cc | 443 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
444 | Perl_croak(aTHX_ "%s" REPORT_LOCATION, \ |
445 | m, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
446 | } STMT_END | |
b45f050a JF |
447 | |
448 | /* | |
449 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL() | |
450 | */ | |
ccb2c380 MP |
451 | #define vFAIL(m) STMT_START { \ |
452 | if (!SIZE_ONLY) \ | |
288b8c02 | 453 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
454 | Simple_vFAIL(m); \ |
455 | } STMT_END | |
b45f050a JF |
456 | |
457 | /* | |
458 | * Like Simple_vFAIL(), but accepts two arguments. | |
459 | */ | |
ccb2c380 | 460 | #define Simple_vFAIL2(m,a1) STMT_START { \ |
a28509cc | 461 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
462 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \ |
463 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
464 | } STMT_END | |
b45f050a JF |
465 | |
466 | /* | |
467 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2(). | |
468 | */ | |
ccb2c380 MP |
469 | #define vFAIL2(m,a1) STMT_START { \ |
470 | if (!SIZE_ONLY) \ | |
288b8c02 | 471 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
472 | Simple_vFAIL2(m, a1); \ |
473 | } STMT_END | |
b45f050a JF |
474 | |
475 | ||
476 | /* | |
477 | * Like Simple_vFAIL(), but accepts three arguments. | |
478 | */ | |
ccb2c380 | 479 | #define Simple_vFAIL3(m, a1, a2) STMT_START { \ |
a28509cc | 480 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
481 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \ |
482 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
483 | } STMT_END | |
b45f050a JF |
484 | |
485 | /* | |
486 | * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3(). | |
487 | */ | |
ccb2c380 MP |
488 | #define vFAIL3(m,a1,a2) STMT_START { \ |
489 | if (!SIZE_ONLY) \ | |
288b8c02 | 490 | SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx_sv); \ |
ccb2c380 MP |
491 | Simple_vFAIL3(m, a1, a2); \ |
492 | } STMT_END | |
b45f050a JF |
493 | |
494 | /* | |
495 | * Like Simple_vFAIL(), but accepts four arguments. | |
496 | */ | |
ccb2c380 | 497 | #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \ |
a28509cc | 498 | const IV offset = RExC_parse - RExC_precomp; \ |
ccb2c380 MP |
499 | S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \ |
500 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
501 | } STMT_END | |
b45f050a | 502 | |
668c081a | 503 | #define ckWARNreg(loc,m) STMT_START { \ |
a28509cc | 504 | const IV offset = loc - RExC_precomp; \ |
f10f4c18 NC |
505 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
506 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
ccb2c380 MP |
507 | } STMT_END |
508 | ||
668c081a | 509 | #define ckWARNregdep(loc,m) STMT_START { \ |
a28509cc | 510 | const IV offset = loc - RExC_precomp; \ |
d1d15184 | 511 | Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ |
f10f4c18 NC |
512 | m REPORT_LOCATION, \ |
513 | (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
ccb2c380 MP |
514 | } STMT_END |
515 | ||
2335b3d3 KW |
516 | #define ckWARN2regdep(loc,m, a1) STMT_START { \ |
517 | const IV offset = loc - RExC_precomp; \ | |
518 | Perl_ck_warner_d(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ | |
519 | m REPORT_LOCATION, \ | |
520 | a1, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
521 | } STMT_END | |
522 | ||
668c081a | 523 | #define ckWARN2reg(loc, m, a1) STMT_START { \ |
a28509cc | 524 | const IV offset = loc - RExC_precomp; \ |
668c081a | 525 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
ccb2c380 MP |
526 | a1, (int)offset, RExC_precomp, RExC_precomp + offset); \ |
527 | } STMT_END | |
528 | ||
529 | #define vWARN3(loc, m, a1, a2) STMT_START { \ | |
a28509cc | 530 | const IV offset = loc - RExC_precomp; \ |
ccb2c380 MP |
531 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
532 | a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
533 | } STMT_END | |
534 | ||
668c081a NC |
535 | #define ckWARN3reg(loc, m, a1, a2) STMT_START { \ |
536 | const IV offset = loc - RExC_precomp; \ | |
537 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ | |
538 | a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
539 | } STMT_END | |
540 | ||
ccb2c380 | 541 | #define vWARN4(loc, m, a1, a2, a3) STMT_START { \ |
a28509cc | 542 | const IV offset = loc - RExC_precomp; \ |
ccb2c380 MP |
543 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
544 | a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
545 | } STMT_END | |
546 | ||
668c081a NC |
547 | #define ckWARN4reg(loc, m, a1, a2, a3) STMT_START { \ |
548 | const IV offset = loc - RExC_precomp; \ | |
549 | Perl_ck_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ | |
550 | a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
551 | } STMT_END | |
552 | ||
ccb2c380 | 553 | #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \ |
a28509cc | 554 | const IV offset = loc - RExC_precomp; \ |
ccb2c380 MP |
555 | Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ |
556 | a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \ | |
557 | } STMT_END | |
9d1d55b5 | 558 | |
8615cb43 | 559 | |
cd439c50 | 560 | /* Allow for side effects in s */ |
ccb2c380 MP |
561 | #define REGC(c,s) STMT_START { \ |
562 | if (!SIZE_ONLY) *(s) = (c); else (void)(s); \ | |
563 | } STMT_END | |
cd439c50 | 564 | |
fac92740 MJD |
565 | /* Macros for recording node offsets. 20001227 mjd@plover.com |
566 | * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in | |
567 | * element 2*n-1 of the array. Element #2n holds the byte length node #n. | |
568 | * Element 0 holds the number n. | |
07be1b83 | 569 | * Position is 1 indexed. |
fac92740 | 570 | */ |
7122b237 YO |
571 | #ifndef RE_TRACK_PATTERN_OFFSETS |
572 | #define Set_Node_Offset_To_R(node,byte) | |
573 | #define Set_Node_Offset(node,byte) | |
574 | #define Set_Cur_Node_Offset | |
575 | #define Set_Node_Length_To_R(node,len) | |
576 | #define Set_Node_Length(node,len) | |
577 | #define Set_Node_Cur_Length(node) | |
578 | #define Node_Offset(n) | |
579 | #define Node_Length(n) | |
580 | #define Set_Node_Offset_Length(node,offset,len) | |
581 | #define ProgLen(ri) ri->u.proglen | |
582 | #define SetProgLen(ri,x) ri->u.proglen = x | |
583 | #else | |
584 | #define ProgLen(ri) ri->u.offsets[0] | |
585 | #define SetProgLen(ri,x) ri->u.offsets[0] = x | |
ccb2c380 MP |
586 | #define Set_Node_Offset_To_R(node,byte) STMT_START { \ |
587 | if (! SIZE_ONLY) { \ | |
588 | MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \ | |
2a49f0f5 | 589 | __LINE__, (int)(node), (int)(byte))); \ |
ccb2c380 | 590 | if((node) < 0) { \ |
551405c4 | 591 | Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \ |
ccb2c380 MP |
592 | } else { \ |
593 | RExC_offsets[2*(node)-1] = (byte); \ | |
594 | } \ | |
595 | } \ | |
596 | } STMT_END | |
597 | ||
598 | #define Set_Node_Offset(node,byte) \ | |
599 | Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start) | |
600 | #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse) | |
601 | ||
602 | #define Set_Node_Length_To_R(node,len) STMT_START { \ | |
603 | if (! SIZE_ONLY) { \ | |
604 | MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \ | |
551405c4 | 605 | __LINE__, (int)(node), (int)(len))); \ |
ccb2c380 | 606 | if((node) < 0) { \ |
551405c4 | 607 | Perl_croak(aTHX_ "value of node is %d in Length macro", (int)(node)); \ |
ccb2c380 MP |
608 | } else { \ |
609 | RExC_offsets[2*(node)] = (len); \ | |
610 | } \ | |
611 | } \ | |
612 | } STMT_END | |
613 | ||
614 | #define Set_Node_Length(node,len) \ | |
615 | Set_Node_Length_To_R((node)-RExC_emit_start, len) | |
616 | #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len) | |
617 | #define Set_Node_Cur_Length(node) \ | |
618 | Set_Node_Length(node, RExC_parse - parse_start) | |
fac92740 MJD |
619 | |
620 | /* Get offsets and lengths */ | |
621 | #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1]) | |
622 | #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)]) | |
623 | ||
07be1b83 YO |
624 | #define Set_Node_Offset_Length(node,offset,len) STMT_START { \ |
625 | Set_Node_Offset_To_R((node)-RExC_emit_start, (offset)); \ | |
626 | Set_Node_Length_To_R((node)-RExC_emit_start, (len)); \ | |
627 | } STMT_END | |
7122b237 | 628 | #endif |
07be1b83 YO |
629 | |
630 | #if PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS | |
631 | #define EXPERIMENTAL_INPLACESCAN | |
f427392e | 632 | #endif /*PERL_ENABLE_EXPERIMENTAL_REGEX_OPTIMISATIONS*/ |
07be1b83 | 633 | |
304ee84b YO |
634 | #define DEBUG_STUDYDATA(str,data,depth) \ |
635 | DEBUG_OPTIMISE_MORE_r(if(data){ \ | |
1de06328 | 636 | PerlIO_printf(Perl_debug_log, \ |
304ee84b YO |
637 | "%*s" str "Pos:%"IVdf"/%"IVdf \ |
638 | " Flags: 0x%"UVXf" Whilem_c: %"IVdf" Lcp: %"IVdf" %s", \ | |
1de06328 YO |
639 | (int)(depth)*2, "", \ |
640 | (IV)((data)->pos_min), \ | |
641 | (IV)((data)->pos_delta), \ | |
304ee84b | 642 | (UV)((data)->flags), \ |
1de06328 | 643 | (IV)((data)->whilem_c), \ |
304ee84b YO |
644 | (IV)((data)->last_closep ? *((data)->last_closep) : -1), \ |
645 | is_inf ? "INF " : "" \ | |
1de06328 YO |
646 | ); \ |
647 | if ((data)->last_found) \ | |
648 | PerlIO_printf(Perl_debug_log, \ | |
649 | "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \ | |
650 | " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \ | |
651 | SvPVX_const((data)->last_found), \ | |
652 | (IV)((data)->last_end), \ | |
653 | (IV)((data)->last_start_min), \ | |
654 | (IV)((data)->last_start_max), \ | |
655 | ((data)->longest && \ | |
656 | (data)->longest==&((data)->longest_fixed)) ? "*" : "", \ | |
657 | SvPVX_const((data)->longest_fixed), \ | |
658 | (IV)((data)->offset_fixed), \ | |
659 | ((data)->longest && \ | |
660 | (data)->longest==&((data)->longest_float)) ? "*" : "", \ | |
661 | SvPVX_const((data)->longest_float), \ | |
662 | (IV)((data)->offset_float_min), \ | |
663 | (IV)((data)->offset_float_max) \ | |
664 | ); \ | |
665 | PerlIO_printf(Perl_debug_log,"\n"); \ | |
666 | }); | |
667 | ||
acfe0abc | 668 | static void clear_re(pTHX_ void *r); |
4327152a | 669 | |
653099ff | 670 | /* Mark that we cannot extend a found fixed substring at this point. |
786e8c11 | 671 | Update the longest found anchored substring and the longest found |
653099ff GS |
672 | floating substrings if needed. */ |
673 | ||
4327152a | 674 | STATIC void |
304ee84b | 675 | S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp, int is_inf) |
c277df42 | 676 | { |
e1ec3a88 AL |
677 | const STRLEN l = CHR_SVLEN(data->last_found); |
678 | const STRLEN old_l = CHR_SVLEN(*data->longest); | |
1de06328 | 679 | GET_RE_DEBUG_FLAGS_DECL; |
b81d288d | 680 | |
7918f24d NC |
681 | PERL_ARGS_ASSERT_SCAN_COMMIT; |
682 | ||
c277df42 | 683 | if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) { |
6b43b216 | 684 | SvSetMagicSV(*data->longest, data->last_found); |
c277df42 IZ |
685 | if (*data->longest == data->longest_fixed) { |
686 | data->offset_fixed = l ? data->last_start_min : data->pos_min; | |
687 | if (data->flags & SF_BEFORE_EOL) | |
b81d288d | 688 | data->flags |
c277df42 IZ |
689 | |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL); |
690 | else | |
691 | data->flags &= ~SF_FIX_BEFORE_EOL; | |
686b73d4 | 692 | data->minlen_fixed=minlenp; |
1de06328 | 693 | data->lookbehind_fixed=0; |
a0ed51b3 | 694 | } |
304ee84b | 695 | else { /* *data->longest == data->longest_float */ |
c277df42 | 696 | data->offset_float_min = l ? data->last_start_min : data->pos_min; |
b81d288d AB |
697 | data->offset_float_max = (l |
698 | ? data->last_start_max | |
c277df42 | 699 | : data->pos_min + data->pos_delta); |
304ee84b | 700 | if (is_inf || (U32)data->offset_float_max > (U32)I32_MAX) |
9051bda5 | 701 | data->offset_float_max = I32_MAX; |
c277df42 | 702 | if (data->flags & SF_BEFORE_EOL) |
b81d288d | 703 | data->flags |
c277df42 IZ |
704 | |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL); |
705 | else | |
706 | data->flags &= ~SF_FL_BEFORE_EOL; | |
1de06328 YO |
707 | data->minlen_float=minlenp; |
708 | data->lookbehind_float=0; | |
c277df42 IZ |
709 | } |
710 | } | |
711 | SvCUR_set(data->last_found, 0); | |
0eda9292 | 712 | { |
a28509cc | 713 | SV * const sv = data->last_found; |
097eb12c AL |
714 | if (SvUTF8(sv) && SvMAGICAL(sv)) { |
715 | MAGIC * const mg = mg_find(sv, PERL_MAGIC_utf8); | |
716 | if (mg) | |
717 | mg->mg_len = 0; | |
718 | } | |
0eda9292 | 719 | } |
c277df42 IZ |
720 | data->last_end = -1; |
721 | data->flags &= ~SF_BEFORE_EOL; | |
bcdf7404 | 722 | DEBUG_STUDYDATA("commit: ",data,0); |
c277df42 IZ |
723 | } |
724 | ||
653099ff GS |
725 | /* Can match anything (initialization) */ |
726 | STATIC void | |
3fffb88a | 727 | S_cl_anything(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) |
653099ff | 728 | { |
7918f24d NC |
729 | PERL_ARGS_ASSERT_CL_ANYTHING; |
730 | ||
f8bef550 | 731 | ANYOF_BITMAP_SETALL(cl); |
dd58aee1 | 732 | cl->flags = ANYOF_CLASS|ANYOF_EOS|ANYOF_UNICODE_ALL |
3ad98780 | 733 | |ANYOF_LOC_NONBITMAP_FOLD|ANYOF_NON_UTF8_LATIN1_ALL; |
3fffb88a KW |
734 | |
735 | /* If any portion of the regex is to operate under locale rules, | |
736 | * initialization includes it. The reason this isn't done for all regexes | |
737 | * is that the optimizer was written under the assumption that locale was | |
738 | * all-or-nothing. Given the complexity and lack of documentation in the | |
739 | * optimizer, and that there are inadequate test cases for locale, so many | |
740 | * parts of it may not work properly, it is safest to avoid locale unless | |
741 | * necessary. */ | |
742 | if (RExC_contains_locale) { | |
9d7a1e63 | 743 | ANYOF_CLASS_SETALL(cl); /* /l uses class */ |
3fffb88a KW |
744 | cl->flags |= ANYOF_LOCALE; |
745 | } | |
9d7a1e63 KW |
746 | else { |
747 | ANYOF_CLASS_ZERO(cl); /* Only /l uses class now */ | |
748 | } | |
653099ff GS |
749 | } |
750 | ||
751 | /* Can match anything (initialization) */ | |
752 | STATIC int | |
5f66b61c | 753 | S_cl_is_anything(const struct regnode_charclass_class *cl) |
653099ff GS |
754 | { |
755 | int value; | |
756 | ||
7918f24d NC |
757 | PERL_ARGS_ASSERT_CL_IS_ANYTHING; |
758 | ||
aaa51d5e | 759 | for (value = 0; value <= ANYOF_MAX; value += 2) |
653099ff GS |
760 | if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1)) |
761 | return 1; | |
1aa99e6b IH |
762 | if (!(cl->flags & ANYOF_UNICODE_ALL)) |
763 | return 0; | |
10edeb5d | 764 | if (!ANYOF_BITMAP_TESTALLSET((const void*)cl)) |
f8bef550 | 765 | return 0; |
653099ff GS |
766 | return 1; |
767 | } | |
768 | ||
769 | /* Can match anything (initialization) */ | |
770 | STATIC void | |
e755fd73 | 771 | S_cl_init(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) |
653099ff | 772 | { |
7918f24d NC |
773 | PERL_ARGS_ASSERT_CL_INIT; |
774 | ||
8ecf7187 | 775 | Zero(cl, 1, struct regnode_charclass_class); |
653099ff | 776 | cl->type = ANYOF; |
3fffb88a | 777 | cl_anything(pRExC_state, cl); |
1411dba4 | 778 | ARG_SET(cl, ANYOF_NONBITMAP_EMPTY); |
653099ff GS |
779 | } |
780 | ||
1051e1c4 KW |
781 | /* These two functions currently do the exact same thing */ |
782 | #define cl_init_zero S_cl_init | |
653099ff | 783 | |
dd58aee1 KW |
784 | /* 'AND' a given class with another one. Can create false positives. 'cl' |
785 | * should not be inverted. 'and_with->flags & ANYOF_CLASS' should be 0 if | |
786 | * 'and_with' is a regnode_charclass instead of a regnode_charclass_class. */ | |
653099ff | 787 | STATIC void |
5f66b61c | 788 | S_cl_and(struct regnode_charclass_class *cl, |
a28509cc | 789 | const struct regnode_charclass_class *and_with) |
653099ff | 790 | { |
7918f24d | 791 | PERL_ARGS_ASSERT_CL_AND; |
40d049e4 YO |
792 | |
793 | assert(and_with->type == ANYOF); | |
1e6ade67 | 794 | |
c6b76537 | 795 | /* I (khw) am not sure all these restrictions are necessary XXX */ |
1e6ade67 KW |
796 | if (!(ANYOF_CLASS_TEST_ANY_SET(and_with)) |
797 | && !(ANYOF_CLASS_TEST_ANY_SET(cl)) | |
653099ff | 798 | && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) |
39065660 KW |
799 | && !(and_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
800 | && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) { | |
653099ff GS |
801 | int i; |
802 | ||
803 | if (and_with->flags & ANYOF_INVERT) | |
804 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
805 | cl->bitmap[i] &= ~and_with->bitmap[i]; | |
806 | else | |
807 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
808 | cl->bitmap[i] &= and_with->bitmap[i]; | |
809 | } /* XXXX: logic is complicated otherwise, leave it along for a moment. */ | |
1aa99e6b | 810 | |
c6b76537 | 811 | if (and_with->flags & ANYOF_INVERT) { |
8951c461 | 812 | |
c6b76537 KW |
813 | /* Here, the and'ed node is inverted. Get the AND of the flags that |
814 | * aren't affected by the inversion. Those that are affected are | |
815 | * handled individually below */ | |
816 | U8 affected_flags = cl->flags & ~INVERSION_UNAFFECTED_FLAGS; | |
817 | cl->flags &= (and_with->flags & INVERSION_UNAFFECTED_FLAGS); | |
818 | cl->flags |= affected_flags; | |
819 | ||
820 | /* We currently don't know how to deal with things that aren't in the | |
821 | * bitmap, but we know that the intersection is no greater than what | |
822 | * is already in cl, so let there be false positives that get sorted | |
823 | * out after the synthetic start class succeeds, and the node is | |
824 | * matched for real. */ | |
825 | ||
826 | /* The inversion of these two flags indicate that the resulting | |
827 | * intersection doesn't have them */ | |
828 | if (and_with->flags & ANYOF_UNICODE_ALL) { | |
4713bfe1 KW |
829 | cl->flags &= ~ANYOF_UNICODE_ALL; |
830 | } | |
c6b76537 KW |
831 | if (and_with->flags & ANYOF_NON_UTF8_LATIN1_ALL) { |
832 | cl->flags &= ~ANYOF_NON_UTF8_LATIN1_ALL; | |
137165a6 | 833 | } |
1aa99e6b | 834 | } |
c6b76537 | 835 | else { /* and'd node is not inverted */ |
3ad98780 KW |
836 | U8 outside_bitmap_but_not_utf8; /* Temp variable */ |
837 | ||
137165a6 | 838 | if (! ANYOF_NONBITMAP(and_with)) { |
c6b76537 KW |
839 | |
840 | /* Here 'and_with' doesn't match anything outside the bitmap | |
841 | * (except possibly ANYOF_UNICODE_ALL), which means the | |
842 | * intersection can't either, except for ANYOF_UNICODE_ALL, in | |
843 | * which case we don't know what the intersection is, but it's no | |
844 | * greater than what cl already has, so can just leave it alone, | |
845 | * with possible false positives */ | |
846 | if (! (and_with->flags & ANYOF_UNICODE_ALL)) { | |
847 | ARG_SET(cl, ANYOF_NONBITMAP_EMPTY); | |
871d0d1a | 848 | cl->flags &= ~ANYOF_NONBITMAP_NON_UTF8; |
c6b76537 | 849 | } |
137165a6 | 850 | } |
c6b76537 KW |
851 | else if (! ANYOF_NONBITMAP(cl)) { |
852 | ||
853 | /* Here, 'and_with' does match something outside the bitmap, and cl | |
854 | * doesn't have a list of things to match outside the bitmap. If | |
855 | * cl can match all code points above 255, the intersection will | |
3ad98780 KW |
856 | * be those above-255 code points that 'and_with' matches. If cl |
857 | * can't match all Unicode code points, it means that it can't | |
858 | * match anything outside the bitmap (since the 'if' that got us | |
859 | * into this block tested for that), so we leave the bitmap empty. | |
860 | */ | |
c6b76537 KW |
861 | if (cl->flags & ANYOF_UNICODE_ALL) { |
862 | ARG_SET(cl, ARG(and_with)); | |
3ad98780 KW |
863 | |
864 | /* and_with's ARG may match things that don't require UTF8. | |
865 | * And now cl's will too, in spite of this being an 'and'. See | |
866 | * the comments below about the kludge */ | |
867 | cl->flags |= and_with->flags & ANYOF_NONBITMAP_NON_UTF8; | |
c6b76537 KW |
868 | } |
869 | } | |
870 | else { | |
871 | /* Here, both 'and_with' and cl match something outside the | |
872 | * bitmap. Currently we do not do the intersection, so just match | |
873 | * whatever cl had at the beginning. */ | |
874 | } | |
875 | ||
876 | ||
3ad98780 KW |
877 | /* Take the intersection of the two sets of flags. However, the |
878 | * ANYOF_NONBITMAP_NON_UTF8 flag is treated as an 'or'. This is a | |
879 | * kludge around the fact that this flag is not treated like the others | |
880 | * which are initialized in cl_anything(). The way the optimizer works | |
881 | * is that the synthetic start class (SSC) is initialized to match | |
882 | * anything, and then the first time a real node is encountered, its | |
883 | * values are AND'd with the SSC's with the result being the values of | |
884 | * the real node. However, there are paths through the optimizer where | |
885 | * the AND never gets called, so those initialized bits are set | |
886 | * inappropriately, which is not usually a big deal, as they just cause | |
887 | * false positives in the SSC, which will just mean a probably | |
888 | * imperceptible slow down in execution. However this bit has a | |
889 | * higher false positive consequence in that it can cause utf8.pm, | |
890 | * utf8_heavy.pl ... to be loaded when not necessary, which is a much | |
891 | * bigger slowdown and also causes significant extra memory to be used. | |
892 | * In order to prevent this, the code now takes a different tack. The | |
893 | * bit isn't set unless some part of the regular expression needs it, | |
894 | * but once set it won't get cleared. This means that these extra | |
895 | * modules won't get loaded unless there was some path through the | |
896 | * pattern that would have required them anyway, and so any false | |
897 | * positives that occur by not ANDing them out when they could be | |
898 | * aren't as severe as they would be if we treated this bit like all | |
899 | * the others */ | |
900 | outside_bitmap_but_not_utf8 = (cl->flags | and_with->flags) | |
901 | & ANYOF_NONBITMAP_NON_UTF8; | |
c6b76537 | 902 | cl->flags &= and_with->flags; |
3ad98780 | 903 | cl->flags |= outside_bitmap_but_not_utf8; |
137165a6 | 904 | } |
653099ff GS |
905 | } |
906 | ||
dd58aee1 KW |
907 | /* 'OR' a given class with another one. Can create false positives. 'cl' |
908 | * should not be inverted. 'or_with->flags & ANYOF_CLASS' should be 0 if | |
909 | * 'or_with' is a regnode_charclass instead of a regnode_charclass_class. */ | |
653099ff | 910 | STATIC void |
3fffb88a | 911 | S_cl_or(const RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with) |
653099ff | 912 | { |
7918f24d NC |
913 | PERL_ARGS_ASSERT_CL_OR; |
914 | ||
653099ff | 915 | if (or_with->flags & ANYOF_INVERT) { |
c6b76537 KW |
916 | |
917 | /* Here, the or'd node is to be inverted. This means we take the | |
918 | * complement of everything not in the bitmap, but currently we don't | |
919 | * know what that is, so give up and match anything */ | |
920 | if (ANYOF_NONBITMAP(or_with)) { | |
3fffb88a | 921 | cl_anything(pRExC_state, cl); |
c6b76537 | 922 | } |
653099ff GS |
923 | /* We do not use |
924 | * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2)) | |
925 | * <= (B1 | !B2) | (CL1 | !CL2) | |
926 | * which is wasteful if CL2 is small, but we ignore CL2: | |
927 | * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1 | |
928 | * XXXX Can we handle case-fold? Unclear: | |
929 | * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) = | |
930 | * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i')) | |
931 | */ | |
c6b76537 | 932 | else if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) |
39065660 KW |
933 | && !(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
934 | && !(cl->flags & ANYOF_LOC_NONBITMAP_FOLD) ) { | |
653099ff GS |
935 | int i; |
936 | ||
937 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
938 | cl->bitmap[i] |= ~or_with->bitmap[i]; | |
939 | } /* XXXX: logic is complicated otherwise */ | |
940 | else { | |
3fffb88a | 941 | cl_anything(pRExC_state, cl); |
653099ff | 942 | } |
c6b76537 KW |
943 | |
944 | /* And, we can just take the union of the flags that aren't affected | |
945 | * by the inversion */ | |
946 | cl->flags |= or_with->flags & INVERSION_UNAFFECTED_FLAGS; | |
947 | ||
948 | /* For the remaining flags: | |
949 | ANYOF_UNICODE_ALL and inverted means to not match anything above | |
950 | 255, which means that the union with cl should just be | |
951 | what cl has in it, so can ignore this flag | |
952 | ANYOF_NON_UTF8_LATIN1_ALL and inverted means if not utf8 and ord | |
953 | is 127-255 to match them, but then invert that, so the | |
954 | union with cl should just be what cl has in it, so can | |
955 | ignore this flag | |
956 | */ | |
957 | } else { /* 'or_with' is not inverted */ | |
653099ff GS |
958 | /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */ |
959 | if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) | |
39065660 KW |
960 | && (!(or_with->flags & ANYOF_LOC_NONBITMAP_FOLD) |
961 | || (cl->flags & ANYOF_LOC_NONBITMAP_FOLD)) ) { | |
653099ff GS |
962 | int i; |
963 | ||
964 | /* OR char bitmap and class bitmap separately */ | |
965 | for (i = 0; i < ANYOF_BITMAP_SIZE; i++) | |
966 | cl->bitmap[i] |= or_with->bitmap[i]; | |
1e6ade67 | 967 | if (ANYOF_CLASS_TEST_ANY_SET(or_with)) { |
653099ff GS |
968 | for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++) |
969 | cl->classflags[i] |= or_with->classflags[i]; | |
970 | cl->flags |= ANYOF_CLASS; | |
971 | } | |
972 | } | |
973 | else { /* XXXX: logic is complicated, leave it along for a moment. */ | |
3fffb88a | 974 | cl_anything(pRExC_state, cl); |
653099ff | 975 | } |
9826f543 | 976 | |
c6b76537 KW |
977 | if (ANYOF_NONBITMAP(or_with)) { |
978 | ||
979 | /* Use the added node's outside-the-bit-map match if there isn't a | |
980 | * conflict. If there is a conflict (both nodes match something | |
981 | * outside the bitmap, but what they match outside is not the same | |
982 | * pointer, and hence not easily compared until XXX we extend | |
983 | * inversion lists this far), give up and allow the start class to | |
d94b1d13 KW |
984 | * match everything outside the bitmap. If that stuff is all above |
985 | * 255, can just set UNICODE_ALL, otherwise caould be anything. */ | |
c6b76537 KW |
986 | if (! ANYOF_NONBITMAP(cl)) { |
987 | ARG_SET(cl, ARG(or_with)); | |
988 | } | |
989 | else if (ARG(cl) != ARG(or_with)) { | |
d94b1d13 KW |
990 | |
991 | if ((or_with->flags & ANYOF_NONBITMAP_NON_UTF8)) { | |
992 | cl_anything(pRExC_state, cl); | |
993 | } | |
994 | else { | |
995 | cl->flags |= ANYOF_UNICODE_ALL; | |
996 | } | |
c6b76537 | 997 | } |
4c34a693 | 998 | } |
0b9668ee KW |
999 | |
1000 | /* Take the union */ | |
1001 | cl->flags |= or_with->flags; | |
1aa99e6b | 1002 | } |
653099ff GS |
1003 | } |
1004 | ||
a3621e74 YO |
1005 | #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ] |
1006 | #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid ) | |
1007 | #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate ) | |
1008 | #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 ) | |
1009 | ||
3dab1dad YO |
1010 | |
1011 | #ifdef DEBUGGING | |
07be1b83 | 1012 | /* |
2b8b4781 NC |
1013 | dump_trie(trie,widecharmap,revcharmap) |
1014 | dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc) | |
1015 | dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc) | |
3dab1dad YO |
1016 | |
1017 | These routines dump out a trie in a somewhat readable format. | |
07be1b83 YO |
1018 | The _interim_ variants are used for debugging the interim |
1019 | tables that are used to generate the final compressed | |
1020 | representation which is what dump_trie expects. | |
1021 | ||
486ec47a | 1022 | Part of the reason for their existence is to provide a form |
3dab1dad | 1023 | of documentation as to how the different representations function. |
07be1b83 YO |
1024 | |
1025 | */ | |
3dab1dad YO |
1026 | |
1027 | /* | |
3dab1dad YO |
1028 | Dumps the final compressed table form of the trie to Perl_debug_log. |
1029 | Used for debugging make_trie(). | |
1030 | */ | |
b9a59e08 | 1031 | |
3dab1dad | 1032 | STATIC void |
2b8b4781 NC |
1033 | S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap, |
1034 | AV *revcharmap, U32 depth) | |
3dab1dad YO |
1035 | { |
1036 | U32 state; | |
ab3bbdeb | 1037 | SV *sv=sv_newmortal(); |
55eed653 | 1038 | int colwidth= widecharmap ? 6 : 4; |
2e64971a | 1039 | U16 word; |
3dab1dad YO |
1040 | GET_RE_DEBUG_FLAGS_DECL; |
1041 | ||
7918f24d | 1042 | PERL_ARGS_ASSERT_DUMP_TRIE; |
ab3bbdeb | 1043 | |
3dab1dad YO |
1044 | PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ", |
1045 | (int)depth * 2 + 2,"", | |
1046 | "Match","Base","Ofs" ); | |
1047 | ||
1048 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) { | |
2b8b4781 | 1049 | SV ** const tmp = av_fetch( revcharmap, state, 0); |
3dab1dad | 1050 | if ( tmp ) { |
ab3bbdeb YO |
1051 | PerlIO_printf( Perl_debug_log, "%*s", |
1052 | colwidth, | |
ddc5bc0f | 1053 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
ab3bbdeb YO |
1054 | PL_colors[0], PL_colors[1], |
1055 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
1056 | PERL_PV_ESCAPE_FIRSTCHAR | |
1057 | ) | |
1058 | ); | |
3dab1dad YO |
1059 | } |
1060 | } | |
1061 | PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------", | |
1062 | (int)depth * 2 + 2,""); | |
1063 | ||
1064 | for( state = 0 ; state < trie->uniquecharcount ; state++ ) | |
ab3bbdeb | 1065 | PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------"); |
3dab1dad YO |
1066 | PerlIO_printf( Perl_debug_log, "\n"); |
1067 | ||
1e2e3d02 | 1068 | for( state = 1 ; state < trie->statecount ; state++ ) { |
be8e71aa | 1069 | const U32 base = trie->states[ state ].trans.base; |
3dab1dad YO |
1070 | |
1071 | PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state); | |
1072 | ||
1073 | if ( trie->states[ state ].wordnum ) { | |
1074 | PerlIO_printf( Perl_debug_log, " W%4X", trie->states[ state ].wordnum ); | |
1075 | } else { | |
1076 | PerlIO_printf( Perl_debug_log, "%6s", "" ); | |
1077 | } | |
1078 | ||
1079 | PerlIO_printf( Perl_debug_log, " @%4"UVXf" ", (UV)base ); | |
1080 | ||
1081 | if ( base ) { | |
1082 | U32 ofs = 0; | |
1083 | ||
1084 | while( ( base + ofs < trie->uniquecharcount ) || | |
1085 | ( base + ofs - trie->uniquecharcount < trie->lasttrans | |
1086 | && trie->trans[ base + ofs - trie->uniquecharcount ].check != state)) | |
1087 | ofs++; | |
1088 | ||
1089 | PerlIO_printf( Perl_debug_log, "+%2"UVXf"[ ", (UV)ofs); | |
1090 | ||
1091 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { | |
1092 | if ( ( base + ofs >= trie->uniquecharcount ) && | |
1093 | ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && | |
1094 | trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) | |
1095 | { | |
ab3bbdeb YO |
1096 | PerlIO_printf( Perl_debug_log, "%*"UVXf, |
1097 | colwidth, | |
3dab1dad YO |
1098 | (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next ); |
1099 | } else { | |
ab3bbdeb | 1100 | PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." ); |
3dab1dad YO |
1101 | } |
1102 | } | |
1103 | ||
1104 | PerlIO_printf( Perl_debug_log, "]"); | |
1105 | ||
1106 | } | |
1107 | PerlIO_printf( Perl_debug_log, "\n" ); | |
1108 | } | |
2e64971a DM |
1109 | PerlIO_printf(Perl_debug_log, "%*sword_info N:(prev,len)=", (int)depth*2, ""); |
1110 | for (word=1; word <= trie->wordcount; word++) { | |
1111 | PerlIO_printf(Perl_debug_log, " %d:(%d,%d)", | |
1112 | (int)word, (int)(trie->wordinfo[word].prev), | |
1113 | (int)(trie->wordinfo[word].len)); | |
1114 | } | |
1115 | PerlIO_printf(Perl_debug_log, "\n" ); | |
3dab1dad YO |
1116 | } |
1117 | /* | |
3dab1dad YO |
1118 | Dumps a fully constructed but uncompressed trie in list form. |
1119 | List tries normally only are used for construction when the number of | |
1120 | possible chars (trie->uniquecharcount) is very high. | |
1121 | Used for debugging make_trie(). | |
1122 | */ | |
1123 | STATIC void | |
55eed653 | 1124 | S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie, |
2b8b4781 NC |
1125 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
1126 | U32 depth) | |
3dab1dad YO |
1127 | { |
1128 | U32 state; | |
ab3bbdeb | 1129 | SV *sv=sv_newmortal(); |
55eed653 | 1130 | int colwidth= widecharmap ? 6 : 4; |
3dab1dad | 1131 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
1132 | |
1133 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_LIST; | |
1134 | ||
3dab1dad | 1135 | /* print out the table precompression. */ |
ab3bbdeb YO |
1136 | PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s", |
1137 | (int)depth * 2 + 2,"", (int)depth * 2 + 2,"", | |
1138 | "------:-----+-----------------\n" ); | |
3dab1dad YO |
1139 | |
1140 | for( state=1 ; state < next_alloc ; state ++ ) { | |
1141 | U16 charid; | |
1142 | ||
ab3bbdeb | 1143 | PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :", |
3dab1dad YO |
1144 | (int)depth * 2 + 2,"", (UV)state ); |
1145 | if ( ! trie->states[ state ].wordnum ) { | |
1146 | PerlIO_printf( Perl_debug_log, "%5s| ",""); | |
1147 | } else { | |
1148 | PerlIO_printf( Perl_debug_log, "W%4x| ", | |
1149 | trie->states[ state ].wordnum | |
1150 | ); | |
1151 | } | |
1152 | for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) { | |
2b8b4781 | 1153 | SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0); |
ab3bbdeb YO |
1154 | if ( tmp ) { |
1155 | PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ", | |
1156 | colwidth, | |
ddc5bc0f | 1157 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
ab3bbdeb YO |
1158 | PL_colors[0], PL_colors[1], |
1159 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
1160 | PERL_PV_ESCAPE_FIRSTCHAR | |
1161 | ) , | |
1e2e3d02 YO |
1162 | TRIE_LIST_ITEM(state,charid).forid, |
1163 | (UV)TRIE_LIST_ITEM(state,charid).newstate | |
1164 | ); | |
1165 | if (!(charid % 10)) | |
664e119d RGS |
1166 | PerlIO_printf(Perl_debug_log, "\n%*s| ", |
1167 | (int)((depth * 2) + 14), ""); | |
1e2e3d02 | 1168 | } |
ab3bbdeb YO |
1169 | } |
1170 | PerlIO_printf( Perl_debug_log, "\n"); | |
3dab1dad YO |
1171 | } |
1172 | } | |
1173 | ||
1174 | /* | |
3dab1dad YO |
1175 | Dumps a fully constructed but uncompressed trie in table form. |
1176 | This is the normal DFA style state transition table, with a few | |
1177 | twists to facilitate compression later. | |
1178 | Used for debugging make_trie(). | |
1179 | */ | |
1180 | STATIC void | |
55eed653 | 1181 | S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie, |
2b8b4781 NC |
1182 | HV *widecharmap, AV *revcharmap, U32 next_alloc, |
1183 | U32 depth) | |
3dab1dad YO |
1184 | { |
1185 | U32 state; | |
1186 | U16 charid; | |
ab3bbdeb | 1187 | SV *sv=sv_newmortal(); |
55eed653 | 1188 | int colwidth= widecharmap ? 6 : 4; |
3dab1dad | 1189 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
1190 | |
1191 | PERL_ARGS_ASSERT_DUMP_TRIE_INTERIM_TABLE; | |
3dab1dad YO |
1192 | |
1193 | /* | |
1194 | print out the table precompression so that we can do a visual check | |
1195 | that they are identical. | |
1196 | */ | |
1197 | ||
1198 | PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" ); | |
1199 | ||
1200 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { | |
2b8b4781 | 1201 | SV ** const tmp = av_fetch( revcharmap, charid, 0); |
3dab1dad | 1202 | if ( tmp ) { |
ab3bbdeb YO |
1203 | PerlIO_printf( Perl_debug_log, "%*s", |
1204 | colwidth, | |
ddc5bc0f | 1205 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth, |
ab3bbdeb YO |
1206 | PL_colors[0], PL_colors[1], |
1207 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
1208 | PERL_PV_ESCAPE_FIRSTCHAR | |
1209 | ) | |
1210 | ); | |
3dab1dad YO |
1211 | } |
1212 | } | |
1213 | ||
1214 | PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" ); | |
1215 | ||
1216 | for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) { | |
ab3bbdeb | 1217 | PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------"); |
3dab1dad YO |
1218 | } |
1219 | ||
1220 | PerlIO_printf( Perl_debug_log, "\n" ); | |
1221 | ||
1222 | for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) { | |
1223 | ||
1224 | PerlIO_printf( Perl_debug_log, "%*s%4"UVXf" : ", | |
1225 | (int)depth * 2 + 2,"", | |
1226 | (UV)TRIE_NODENUM( state ) ); | |
1227 | ||
1228 | for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { | |
ab3bbdeb YO |
1229 | UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next ); |
1230 | if (v) | |
1231 | PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v ); | |
1232 | else | |
1233 | PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." ); | |
3dab1dad YO |
1234 | } |
1235 | if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) { | |
1236 | PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check ); | |
1237 | } else { | |
1238 | PerlIO_printf( Perl_debug_log, " (%4"UVXf") W%4X\n", (UV)trie->trans[ state ].check, | |
1239 | trie->states[ TRIE_NODENUM( state ) ].wordnum ); | |
1240 | } | |
1241 | } | |
07be1b83 | 1242 | } |
3dab1dad YO |
1243 | |
1244 | #endif | |
1245 | ||
2e64971a | 1246 | |
786e8c11 YO |
1247 | /* make_trie(startbranch,first,last,tail,word_count,flags,depth) |
1248 | startbranch: the first branch in the whole branch sequence | |
1249 | first : start branch of sequence of branch-exact nodes. | |
1250 | May be the same as startbranch | |
1251 | last : Thing following the last branch. | |
1252 | May be the same as tail. | |
1253 | tail : item following the branch sequence | |
1254 | count : words in the sequence | |
1255 | flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/ | |
1256 | depth : indent depth | |
3dab1dad | 1257 | |
786e8c11 | 1258 | Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node. |
07be1b83 | 1259 | |
786e8c11 YO |
1260 | A trie is an N'ary tree where the branches are determined by digital |
1261 | decomposition of the key. IE, at the root node you look up the 1st character and | |
1262 | follow that branch repeat until you find the end of the branches. Nodes can be | |
1263 | marked as "accepting" meaning they represent a complete word. Eg: | |
07be1b83 | 1264 | |
786e8c11 | 1265 | /he|she|his|hers/ |
72f13be8 | 1266 | |
786e8c11 YO |
1267 | would convert into the following structure. Numbers represent states, letters |
1268 | following numbers represent valid transitions on the letter from that state, if | |
1269 | the number is in square brackets it represents an accepting state, otherwise it | |
1270 | will be in parenthesis. | |
07be1b83 | 1271 | |
786e8c11 YO |
1272 | +-h->+-e->[3]-+-r->(8)-+-s->[9] |
1273 | | | | |
1274 | | (2) | |
1275 | | | | |
1276 | (1) +-i->(6)-+-s->[7] | |
1277 | | | |
1278 | +-s->(3)-+-h->(4)-+-e->[5] | |
07be1b83 | 1279 | |
786e8c11 YO |
1280 | Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers) |
1281 | ||
1282 | This shows that when matching against the string 'hers' we will begin at state 1 | |
1283 | read 'h' and move to state 2, read 'e' and move to state 3 which is accepting, | |
1284 | then read 'r' and go to state 8 followed by 's' which takes us to state 9 which | |
1285 | is also accepting. Thus we know that we can match both 'he' and 'hers' with a | |
1286 | single traverse. We store a mapping from accepting to state to which word was | |
1287 | matched, and then when we have multiple possibilities we try to complete the | |
1288 | rest of the regex in the order in which they occured in the alternation. | |
1289 | ||
1290 | The only prior NFA like behaviour that would be changed by the TRIE support is | |
1291 | the silent ignoring of duplicate alternations which are of the form: | |
1292 | ||
1293 | / (DUPE|DUPE) X? (?{ ... }) Y /x | |
1294 | ||
4b714af6 | 1295 | Thus EVAL blocks following a trie may be called a different number of times with |
786e8c11 | 1296 | and without the optimisation. With the optimisations dupes will be silently |
486ec47a | 1297 | ignored. This inconsistent behaviour of EVAL type nodes is well established as |
786e8c11 YO |
1298 | the following demonstrates: |
1299 | ||
1300 | 'words'=~/(word|word|word)(?{ print $1 })[xyz]/ | |
1301 | ||
1302 | which prints out 'word' three times, but | |
1303 | ||
1304 | 'words'=~/(word|word|word)(?{ print $1 })S/ | |
1305 | ||
1306 | which doesnt print it out at all. This is due to other optimisations kicking in. | |
1307 | ||
1308 | Example of what happens on a structural level: | |
1309 | ||
486ec47a | 1310 | The regexp /(ac|ad|ab)+/ will produce the following debug output: |
786e8c11 YO |
1311 | |
1312 | 1: CURLYM[1] {1,32767}(18) | |
1313 | 5: BRANCH(8) | |
1314 | 6: EXACT <ac>(16) | |
1315 | 8: BRANCH(11) | |
1316 | 9: EXACT <ad>(16) | |
1317 | 11: BRANCH(14) | |
1318 | 12: EXACT <ab>(16) | |
1319 | 16: SUCCEED(0) | |
1320 | 17: NOTHING(18) | |
1321 | 18: END(0) | |
1322 | ||
1323 | This would be optimizable with startbranch=5, first=5, last=16, tail=16 | |
1324 | and should turn into: | |
1325 | ||
1326 | 1: CURLYM[1] {1,32767}(18) | |
1327 | 5: TRIE(16) | |
1328 | [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1] | |
1329 | <ac> | |
1330 | <ad> | |
1331 | <ab> | |
1332 | 16: SUCCEED(0) | |
1333 | 17: NOTHING(18) | |
1334 | 18: END(0) | |
1335 | ||
1336 | Cases where tail != last would be like /(?foo|bar)baz/: | |
1337 | ||
1338 | 1: BRANCH(4) | |
1339 | 2: EXACT <foo>(8) | |
1340 | 4: BRANCH(7) | |
1341 | 5: EXACT <bar>(8) | |
1342 | 7: TAIL(8) | |
1343 | 8: EXACT <baz>(10) | |
1344 | 10: END(0) | |
1345 | ||
1346 | which would be optimizable with startbranch=1, first=1, last=7, tail=8 | |
1347 | and would end up looking like: | |
1348 | ||
1349 | 1: TRIE(8) | |
1350 | [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1] | |
1351 | <foo> | |
1352 | <bar> | |
1353 | 7: TAIL(8) | |
1354 | 8: EXACT <baz>(10) | |
1355 | 10: END(0) | |
1356 | ||
1357 | d = uvuni_to_utf8_flags(d, uv, 0); | |
1358 | ||
1359 | is the recommended Unicode-aware way of saying | |
1360 | ||
1361 | *(d++) = uv; | |
1362 | */ | |
1363 | ||
1e2e3d02 | 1364 | #define TRIE_STORE_REVCHAR \ |
786e8c11 | 1365 | STMT_START { \ |
73031816 NC |
1366 | if (UTF) { \ |
1367 | SV *zlopp = newSV(2); \ | |
88c9ea1e CB |
1368 | unsigned char *flrbbbbb = (unsigned char *) SvPVX(zlopp); \ |
1369 | unsigned const char *const kapow = uvuni_to_utf8(flrbbbbb, uvc & 0xFF); \ | |
73031816 NC |
1370 | SvCUR_set(zlopp, kapow - flrbbbbb); \ |
1371 | SvPOK_on(zlopp); \ | |
1372 | SvUTF8_on(zlopp); \ | |
1373 | av_push(revcharmap, zlopp); \ | |
1374 | } else { \ | |
6bdeddd2 | 1375 | char ooooff = (char)uvc; \ |
73031816 NC |
1376 | av_push(revcharmap, newSVpvn(&ooooff, 1)); \ |
1377 | } \ | |
1378 | } STMT_END | |
786e8c11 YO |
1379 | |
1380 | #define TRIE_READ_CHAR STMT_START { \ | |
1381 | wordlen++; \ | |
1382 | if ( UTF ) { \ | |
1383 | if ( folder ) { \ | |
1384 | if ( foldlen > 0 ) { \ | |
1385 | uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \ | |
1386 | foldlen -= len; \ | |
1387 | scan += len; \ | |
1388 | len = 0; \ | |
1389 | } else { \ | |
c81f2f9e KW |
1390 | len = UTF8SKIP(uc);\ |
1391 | uvc = to_utf8_fold( uc, foldbuf, &foldlen); \ | |
786e8c11 YO |
1392 | foldlen -= UNISKIP( uvc ); \ |
1393 | scan = foldbuf + UNISKIP( uvc ); \ | |
1394 | } \ | |
1395 | } else { \ | |
1396 | uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\ | |
1397 | } \ | |
1398 | } else { \ | |
1399 | uvc = (U32)*uc; \ | |
1400 | len = 1; \ | |
1401 | } \ | |
1402 | } STMT_END | |
1403 | ||
1404 | ||
1405 | ||
1406 | #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \ | |
1407 | if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \ | |
f9003953 NC |
1408 | U32 ging = TRIE_LIST_LEN( state ) *= 2; \ |
1409 | Renew( trie->states[ state ].trans.list, ging, reg_trie_trans_le ); \ | |
786e8c11 YO |
1410 | } \ |
1411 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \ | |
1412 | TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \ | |
1413 | TRIE_LIST_CUR( state )++; \ | |
1414 | } STMT_END | |
07be1b83 | 1415 | |
786e8c11 YO |
1416 | #define TRIE_LIST_NEW(state) STMT_START { \ |
1417 | Newxz( trie->states[ state ].trans.list, \ | |
1418 | 4, reg_trie_trans_le ); \ | |
1419 | TRIE_LIST_CUR( state ) = 1; \ | |
1420 | TRIE_LIST_LEN( state ) = 4; \ | |
1421 | } STMT_END | |
07be1b83 | 1422 | |
786e8c11 YO |
1423 | #define TRIE_HANDLE_WORD(state) STMT_START { \ |
1424 | U16 dupe= trie->states[ state ].wordnum; \ | |
1425 | regnode * const noper_next = regnext( noper ); \ | |
1426 | \ | |
786e8c11 YO |
1427 | DEBUG_r({ \ |
1428 | /* store the word for dumping */ \ | |
1429 | SV* tmp; \ | |
1430 | if (OP(noper) != NOTHING) \ | |
740cce10 | 1431 | tmp = newSVpvn_utf8(STRING(noper), STR_LEN(noper), UTF); \ |
786e8c11 | 1432 | else \ |
740cce10 | 1433 | tmp = newSVpvn_utf8( "", 0, UTF ); \ |
2b8b4781 | 1434 | av_push( trie_words, tmp ); \ |
786e8c11 YO |
1435 | }); \ |
1436 | \ | |
1437 | curword++; \ | |
2e64971a DM |
1438 | trie->wordinfo[curword].prev = 0; \ |
1439 | trie->wordinfo[curword].len = wordlen; \ | |
1440 | trie->wordinfo[curword].accept = state; \ | |
786e8c11 YO |
1441 | \ |
1442 | if ( noper_next < tail ) { \ | |
1443 | if (!trie->jump) \ | |
c944940b | 1444 | trie->jump = (U16 *) PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \ |
7f69552c | 1445 | trie->jump[curword] = (U16)(noper_next - convert); \ |
786e8c11 YO |
1446 | if (!jumper) \ |
1447 | jumper = noper_next; \ | |
1448 | if (!nextbranch) \ | |
1449 | nextbranch= regnext(cur); \ | |
1450 | } \ | |
1451 | \ | |
1452 | if ( dupe ) { \ | |
2e64971a DM |
1453 | /* It's a dupe. Pre-insert into the wordinfo[].prev */\ |
1454 | /* chain, so that when the bits of chain are later */\ | |
1455 | /* linked together, the dups appear in the chain */\ | |
1456 | trie->wordinfo[curword].prev = trie->wordinfo[dupe].prev; \ | |
1457 | trie->wordinfo[dupe].prev = curword; \ | |
786e8c11 YO |
1458 | } else { \ |
1459 | /* we haven't inserted this word yet. */ \ | |
1460 | trie->states[ state ].wordnum = curword; \ | |
1461 | } \ | |
1462 | } STMT_END | |
07be1b83 | 1463 | |
3dab1dad | 1464 | |
786e8c11 YO |
1465 | #define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \ |
1466 | ( ( base + charid >= ucharcount \ | |
1467 | && base + charid < ubound \ | |
1468 | && state == trie->trans[ base - ucharcount + charid ].check \ | |
1469 | && trie->trans[ base - ucharcount + charid ].next ) \ | |
1470 | ? trie->trans[ base - ucharcount + charid ].next \ | |
1471 | : ( state==1 ? special : 0 ) \ | |
1472 | ) | |
3dab1dad | 1473 | |
786e8c11 YO |
1474 | #define MADE_TRIE 1 |
1475 | #define MADE_JUMP_TRIE 2 | |
1476 | #define MADE_EXACT_TRIE 4 | |
3dab1dad | 1477 | |
a3621e74 | 1478 | STATIC I32 |
786e8c11 | 1479 | S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth) |
a3621e74 | 1480 | { |
27da23d5 | 1481 | dVAR; |
a3621e74 YO |
1482 | /* first pass, loop through and scan words */ |
1483 | reg_trie_data *trie; | |
55eed653 | 1484 | HV *widecharmap = NULL; |
2b8b4781 | 1485 | AV *revcharmap = newAV(); |
a3621e74 | 1486 | regnode *cur; |
9f7f3913 | 1487 | const U32 uniflags = UTF8_ALLOW_DEFAULT; |
a3621e74 YO |
1488 | STRLEN len = 0; |
1489 | UV uvc = 0; | |
1490 | U16 curword = 0; | |
1491 | U32 next_alloc = 0; | |
786e8c11 YO |
1492 | regnode *jumper = NULL; |
1493 | regnode *nextbranch = NULL; | |
7f69552c | 1494 | regnode *convert = NULL; |
2e64971a | 1495 | U32 *prev_states; /* temp array mapping each state to previous one */ |
a3621e74 | 1496 | /* we just use folder as a flag in utf8 */ |
1e696034 | 1497 | const U8 * folder = NULL; |
a3621e74 | 1498 | |
2b8b4781 NC |
1499 | #ifdef DEBUGGING |
1500 | const U32 data_slot = add_data( pRExC_state, 4, "tuuu" ); | |
1501 | AV *trie_words = NULL; | |
1502 | /* along with revcharmap, this only used during construction but both are | |
1503 | * useful during debugging so we store them in the struct when debugging. | |
8e11feef | 1504 | */ |
2b8b4781 NC |
1505 | #else |
1506 | const U32 data_slot = add_data( pRExC_state, 2, "tu" ); | |
3dab1dad | 1507 | STRLEN trie_charcount=0; |
3dab1dad | 1508 | #endif |
2b8b4781 | 1509 | SV *re_trie_maxbuff; |
a3621e74 | 1510 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
1511 | |
1512 | PERL_ARGS_ASSERT_MAKE_TRIE; | |
72f13be8 YO |
1513 | #ifndef DEBUGGING |
1514 | PERL_UNUSED_ARG(depth); | |
1515 | #endif | |
a3621e74 | 1516 | |
1e696034 | 1517 | switch (flags) { |
2f7f8cb1 | 1518 | case EXACTFA: |
1e696034 KW |
1519 | case EXACTFU: folder = PL_fold_latin1; break; |
1520 | case EXACTF: folder = PL_fold; break; | |
1521 | case EXACTFL: folder = PL_fold_locale; break; | |
1522 | } | |
1523 | ||
c944940b | 1524 | trie = (reg_trie_data *) PerlMemShared_calloc( 1, sizeof(reg_trie_data) ); |
a3621e74 | 1525 | trie->refcount = 1; |
3dab1dad | 1526 | trie->startstate = 1; |
786e8c11 | 1527 | trie->wordcount = word_count; |
f8fc2ecf | 1528 | RExC_rxi->data->data[ data_slot ] = (void*)trie; |
c944940b | 1529 | trie->charmap = (U16 *) PerlMemShared_calloc( 256, sizeof(U16) ); |
3dab1dad | 1530 | if (!(UTF && folder)) |
c944940b | 1531 | trie->bitmap = (char *) PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 ); |
2e64971a DM |
1532 | trie->wordinfo = (reg_trie_wordinfo *) PerlMemShared_calloc( |
1533 | trie->wordcount+1, sizeof(reg_trie_wordinfo)); | |
1534 | ||
a3621e74 | 1535 | DEBUG_r({ |
2b8b4781 | 1536 | trie_words = newAV(); |
a3621e74 | 1537 | }); |
a3621e74 | 1538 | |
0111c4fd | 1539 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); |
a3621e74 | 1540 | if (!SvIOK(re_trie_maxbuff)) { |
0111c4fd | 1541 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); |
a3621e74 | 1542 | } |
3dab1dad YO |
1543 | DEBUG_OPTIMISE_r({ |
1544 | PerlIO_printf( Perl_debug_log, | |
786e8c11 | 1545 | "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n", |
3dab1dad YO |
1546 | (int)depth * 2 + 2, "", |
1547 | REG_NODE_NUM(startbranch),REG_NODE_NUM(first), | |
786e8c11 | 1548 | REG_NODE_NUM(last), REG_NODE_NUM(tail), |
85c3142d | 1549 | (int)depth); |
3dab1dad | 1550 | }); |
7f69552c YO |
1551 | |
1552 | /* Find the node we are going to overwrite */ | |
1553 | if ( first == startbranch && OP( last ) != BRANCH ) { | |
1554 | /* whole branch chain */ | |
1555 | convert = first; | |
1556 | } else { | |
1557 | /* branch sub-chain */ | |
1558 | convert = NEXTOPER( first ); | |
1559 | } | |
1560 | ||
a3621e74 YO |
1561 | /* -- First loop and Setup -- |
1562 | ||
1563 | We first traverse the branches and scan each word to determine if it | |
1564 | contains widechars, and how many unique chars there are, this is | |
1565 | important as we have to build a table with at least as many columns as we | |
1566 | have unique chars. | |
1567 | ||
1568 | We use an array of integers to represent the character codes 0..255 | |
38a44b82 | 1569 | (trie->charmap) and we use a an HV* to store Unicode characters. We use the |
a3621e74 YO |
1570 | native representation of the character value as the key and IV's for the |
1571 | coded index. | |
1572 | ||
1573 | *TODO* If we keep track of how many times each character is used we can | |
1574 | remap the columns so that the table compression later on is more | |
3b753521 | 1575 | efficient in terms of memory by ensuring the most common value is in the |
a3621e74 YO |
1576 | middle and the least common are on the outside. IMO this would be better |
1577 | than a most to least common mapping as theres a decent chance the most | |
1578 | common letter will share a node with the least common, meaning the node | |
486ec47a | 1579 | will not be compressible. With a middle is most common approach the worst |
a3621e74 YO |
1580 | case is when we have the least common nodes twice. |
1581 | ||
1582 | */ | |
1583 | ||
a3621e74 | 1584 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
c445ea15 | 1585 | regnode * const noper = NEXTOPER( cur ); |
e1ec3a88 | 1586 | const U8 *uc = (U8*)STRING( noper ); |
a28509cc | 1587 | const U8 * const e = uc + STR_LEN( noper ); |
a3621e74 YO |
1588 | STRLEN foldlen = 0; |
1589 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; | |
2af232bd | 1590 | const U8 *scan = (U8*)NULL; |
07be1b83 | 1591 | U32 wordlen = 0; /* required init */ |
02daf0ab YO |
1592 | STRLEN chars = 0; |
1593 | bool set_bit = trie->bitmap ? 1 : 0; /*store the first char in the bitmap?*/ | |
a3621e74 | 1594 | |
3dab1dad YO |
1595 | if (OP(noper) == NOTHING) { |
1596 | trie->minlen= 0; | |
1597 | continue; | |
1598 | } | |
02daf0ab YO |
1599 | if ( set_bit ) /* bitmap only alloced when !(UTF&&Folding) */ |
1600 | TRIE_BITMAP_SET(trie,*uc); /* store the raw first byte | |
1601 | regardless of encoding */ | |
1602 | ||
a3621e74 | 1603 | for ( ; uc < e ; uc += len ) { |
3dab1dad | 1604 | TRIE_CHARCOUNT(trie)++; |
a3621e74 | 1605 | TRIE_READ_CHAR; |
3dab1dad | 1606 | chars++; |
a3621e74 YO |
1607 | if ( uvc < 256 ) { |
1608 | if ( !trie->charmap[ uvc ] ) { | |
1609 | trie->charmap[ uvc ]=( ++trie->uniquecharcount ); | |
1610 | if ( folder ) | |
1611 | trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ]; | |
3dab1dad | 1612 | TRIE_STORE_REVCHAR; |
a3621e74 | 1613 | } |
02daf0ab | 1614 | if ( set_bit ) { |
62012aee KW |
1615 | /* store the codepoint in the bitmap, and its folded |
1616 | * equivalent. */ | |
02daf0ab | 1617 | TRIE_BITMAP_SET(trie,uvc); |
0921ee73 T |
1618 | |
1619 | /* store the folded codepoint */ | |
1620 | if ( folder ) TRIE_BITMAP_SET(trie,folder[ uvc ]); | |
1621 | ||
1622 | if ( !UTF ) { | |
1623 | /* store first byte of utf8 representation of | |
acdf4139 KW |
1624 | variant codepoints */ |
1625 | if (! UNI_IS_INVARIANT(uvc)) { | |
1626 | TRIE_BITMAP_SET(trie, UTF8_TWO_BYTE_HI(uvc)); | |
0921ee73 T |
1627 | } |
1628 | } | |
02daf0ab YO |
1629 | set_bit = 0; /* We've done our bit :-) */ |
1630 | } | |
a3621e74 YO |
1631 | } else { |
1632 | SV** svpp; | |
55eed653 NC |
1633 | if ( !widecharmap ) |
1634 | widecharmap = newHV(); | |
a3621e74 | 1635 | |
55eed653 | 1636 | svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 ); |
a3621e74 YO |
1637 | |
1638 | if ( !svpp ) | |
e4584336 | 1639 | Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc ); |
a3621e74 YO |
1640 | |
1641 | if ( !SvTRUE( *svpp ) ) { | |
1642 | sv_setiv( *svpp, ++trie->uniquecharcount ); | |
3dab1dad | 1643 | TRIE_STORE_REVCHAR; |
a3621e74 YO |
1644 | } |
1645 | } | |
1646 | } | |
3dab1dad YO |
1647 | if( cur == first ) { |
1648 | trie->minlen=chars; | |
1649 | trie->maxlen=chars; | |
1650 | } else if (chars < trie->minlen) { | |
1651 | trie->minlen=chars; | |
1652 | } else if (chars > trie->maxlen) { | |
1653 | trie->maxlen=chars; | |
1654 | } | |
1655 | ||
a3621e74 YO |
1656 | } /* end first pass */ |
1657 | DEBUG_TRIE_COMPILE_r( | |
3dab1dad YO |
1658 | PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n", |
1659 | (int)depth * 2 + 2,"", | |
55eed653 | 1660 | ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count, |
be8e71aa YO |
1661 | (int)TRIE_CHARCOUNT(trie), trie->uniquecharcount, |
1662 | (int)trie->minlen, (int)trie->maxlen ) | |
a3621e74 | 1663 | ); |
a3621e74 YO |
1664 | |
1665 | /* | |
1666 | We now know what we are dealing with in terms of unique chars and | |
1667 | string sizes so we can calculate how much memory a naive | |
0111c4fd RGS |
1668 | representation using a flat table will take. If it's over a reasonable |
1669 | limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory | |
a3621e74 YO |
1670 | conservative but potentially much slower representation using an array |
1671 | of lists. | |
1672 | ||
1673 | At the end we convert both representations into the same compressed | |
1674 | form that will be used in regexec.c for matching with. The latter | |
1675 | is a form that cannot be used to construct with but has memory | |
1676 | properties similar to the list form and access properties similar | |
1677 | to the table form making it both suitable for fast searches and | |
1678 | small enough that its feasable to store for the duration of a program. | |
1679 | ||
1680 | See the comment in the code where the compressed table is produced | |
1681 | inplace from the flat tabe representation for an explanation of how | |
1682 | the compression works. | |
1683 | ||
1684 | */ | |
1685 | ||
1686 | ||
2e64971a DM |
1687 | Newx(prev_states, TRIE_CHARCOUNT(trie) + 2, U32); |
1688 | prev_states[1] = 0; | |
1689 | ||
3dab1dad | 1690 | if ( (IV)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) { |
a3621e74 YO |
1691 | /* |
1692 | Second Pass -- Array Of Lists Representation | |
1693 | ||
1694 | Each state will be represented by a list of charid:state records | |
1695 | (reg_trie_trans_le) the first such element holds the CUR and LEN | |
1696 | points of the allocated array. (See defines above). | |
1697 | ||
1698 | We build the initial structure using the lists, and then convert | |
1699 | it into the compressed table form which allows faster lookups | |
1700 | (but cant be modified once converted). | |
a3621e74 YO |
1701 | */ |
1702 | ||
a3621e74 YO |
1703 | STRLEN transcount = 1; |
1704 | ||
1e2e3d02 YO |
1705 | DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, |
1706 | "%*sCompiling trie using list compiler\n", | |
1707 | (int)depth * 2 + 2, "")); | |
686b73d4 | 1708 | |
c944940b JH |
1709 | trie->states = (reg_trie_state *) |
1710 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, | |
1711 | sizeof(reg_trie_state) ); | |
a3621e74 YO |
1712 | TRIE_LIST_NEW(1); |
1713 | next_alloc = 2; | |
1714 | ||
1715 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { | |
1716 | ||
c445ea15 AL |
1717 | regnode * const noper = NEXTOPER( cur ); |
1718 | U8 *uc = (U8*)STRING( noper ); | |
1719 | const U8 * const e = uc + STR_LEN( noper ); | |
1720 | U32 state = 1; /* required init */ | |
1721 | U16 charid = 0; /* sanity init */ | |
1722 | U8 *scan = (U8*)NULL; /* sanity init */ | |
1723 | STRLEN foldlen = 0; /* required init */ | |
07be1b83 | 1724 | U32 wordlen = 0; /* required init */ |
c445ea15 AL |
1725 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
1726 | ||
3dab1dad | 1727 | if (OP(noper) != NOTHING) { |
786e8c11 | 1728 | for ( ; uc < e ; uc += len ) { |
c445ea15 | 1729 | |
786e8c11 | 1730 | TRIE_READ_CHAR; |
c445ea15 | 1731 | |
786e8c11 YO |
1732 | if ( uvc < 256 ) { |
1733 | charid = trie->charmap[ uvc ]; | |
c445ea15 | 1734 | } else { |
55eed653 | 1735 | SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0); |
786e8c11 YO |
1736 | if ( !svpp ) { |
1737 | charid = 0; | |
1738 | } else { | |
1739 | charid=(U16)SvIV( *svpp ); | |
1740 | } | |
c445ea15 | 1741 | } |
786e8c11 YO |
1742 | /* charid is now 0 if we dont know the char read, or nonzero if we do */ |
1743 | if ( charid ) { | |
a3621e74 | 1744 | |
786e8c11 YO |
1745 | U16 check; |
1746 | U32 newstate = 0; | |
a3621e74 | 1747 | |
786e8c11 YO |
1748 | charid--; |
1749 | if ( !trie->states[ state ].trans.list ) { | |
1750 | TRIE_LIST_NEW( state ); | |
c445ea15 | 1751 | } |
786e8c11 YO |
1752 | for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) { |
1753 | if ( TRIE_LIST_ITEM( state, check ).forid == charid ) { | |
1754 | newstate = TRIE_LIST_ITEM( state, check ).newstate; | |
1755 | break; | |
1756 | } | |
1757 | } | |
1758 | if ( ! newstate ) { | |
1759 | newstate = next_alloc++; | |
2e64971a | 1760 | prev_states[newstate] = state; |
786e8c11 YO |
1761 | TRIE_LIST_PUSH( state, charid, newstate ); |
1762 | transcount++; | |
1763 | } | |
1764 | state = newstate; | |
1765 | } else { | |
1766 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); | |
c445ea15 | 1767 | } |
a28509cc | 1768 | } |
c445ea15 | 1769 | } |
3dab1dad | 1770 | TRIE_HANDLE_WORD(state); |
a3621e74 YO |
1771 | |
1772 | } /* end second pass */ | |
1773 | ||
1e2e3d02 YO |
1774 | /* next alloc is the NEXT state to be allocated */ |
1775 | trie->statecount = next_alloc; | |
c944940b JH |
1776 | trie->states = (reg_trie_state *) |
1777 | PerlMemShared_realloc( trie->states, | |
1778 | next_alloc | |
1779 | * sizeof(reg_trie_state) ); | |
a3621e74 | 1780 | |
3dab1dad | 1781 | /* and now dump it out before we compress it */ |
2b8b4781 NC |
1782 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap, |
1783 | revcharmap, next_alloc, | |
1784 | depth+1) | |
1e2e3d02 | 1785 | ); |
a3621e74 | 1786 | |
c944940b JH |
1787 | trie->trans = (reg_trie_trans *) |
1788 | PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) ); | |
a3621e74 YO |
1789 | { |
1790 | U32 state; | |
a3621e74 YO |
1791 | U32 tp = 0; |
1792 | U32 zp = 0; | |
1793 | ||
1794 | ||
1795 | for( state=1 ; state < next_alloc ; state ++ ) { | |
1796 | U32 base=0; | |
1797 | ||
1798 | /* | |
1799 | DEBUG_TRIE_COMPILE_MORE_r( | |
1800 | PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp) | |
1801 | ); | |
1802 | */ | |
1803 | ||
1804 | if (trie->states[state].trans.list) { | |
1805 | U16 minid=TRIE_LIST_ITEM( state, 1).forid; | |
1806 | U16 maxid=minid; | |
a28509cc | 1807 | U16 idx; |
a3621e74 YO |
1808 | |
1809 | for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) { | |
c445ea15 AL |
1810 | const U16 forid = TRIE_LIST_ITEM( state, idx).forid; |
1811 | if ( forid < minid ) { | |
1812 | minid=forid; | |
1813 | } else if ( forid > maxid ) { | |
1814 | maxid=forid; | |
1815 | } | |
a3621e74 YO |
1816 | } |
1817 | if ( transcount < tp + maxid - minid + 1) { | |
1818 | transcount *= 2; | |
c944940b JH |
1819 | trie->trans = (reg_trie_trans *) |
1820 | PerlMemShared_realloc( trie->trans, | |
446bd890 NC |
1821 | transcount |
1822 | * sizeof(reg_trie_trans) ); | |
a3621e74 YO |
1823 | Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans ); |
1824 | } | |
1825 | base = trie->uniquecharcount + tp - minid; | |
1826 | if ( maxid == minid ) { | |
1827 | U32 set = 0; | |
1828 | for ( ; zp < tp ; zp++ ) { | |
1829 | if ( ! trie->trans[ zp ].next ) { | |
1830 | base = trie->uniquecharcount + zp - minid; | |
1831 | trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate; | |
1832 | trie->trans[ zp ].check = state; | |
1833 | set = 1; | |
1834 | break; | |
1835 | } | |
1836 | } | |
1837 | if ( !set ) { | |
1838 | trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate; | |
1839 | trie->trans[ tp ].check = state; | |
1840 | tp++; | |
1841 | zp = tp; | |
1842 | } | |
1843 | } else { | |
1844 | for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) { | |
c445ea15 | 1845 | const U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid; |
a3621e74 YO |
1846 | trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate; |
1847 | trie->trans[ tid ].check = state; | |
1848 | } | |
1849 | tp += ( maxid - minid + 1 ); | |
1850 | } | |
1851 | Safefree(trie->states[ state ].trans.list); | |
1852 | } | |
1853 | /* | |
1854 | DEBUG_TRIE_COMPILE_MORE_r( | |
1855 | PerlIO_printf( Perl_debug_log, " base: %d\n",base); | |
1856 | ); | |
1857 | */ | |
1858 | trie->states[ state ].trans.base=base; | |
1859 | } | |
cc601c31 | 1860 | trie->lasttrans = tp + 1; |
a3621e74 YO |
1861 | } |
1862 | } else { | |
1863 | /* | |
1864 | Second Pass -- Flat Table Representation. | |
1865 | ||
1866 | we dont use the 0 slot of either trans[] or states[] so we add 1 to each. | |
1867 | We know that we will need Charcount+1 trans at most to store the data | |
1868 | (one row per char at worst case) So we preallocate both structures | |
1869 | assuming worst case. | |
1870 | ||
1871 | We then construct the trie using only the .next slots of the entry | |
1872 | structs. | |
1873 | ||
3b753521 | 1874 | We use the .check field of the first entry of the node temporarily to |
a3621e74 YO |
1875 | make compression both faster and easier by keeping track of how many non |
1876 | zero fields are in the node. | |
1877 | ||
1878 | Since trans are numbered from 1 any 0 pointer in the table is a FAIL | |
1879 | transition. | |
1880 | ||
1881 | There are two terms at use here: state as a TRIE_NODEIDX() which is a | |
1882 | number representing the first entry of the node, and state as a | |
1883 | TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and | |
1884 | TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there | |
1885 | are 2 entrys per node. eg: | |
1886 | ||
1887 | A B A B | |
1888 | 1. 2 4 1. 3 7 | |
1889 | 2. 0 3 3. 0 5 | |
1890 | 3. 0 0 5. 0 0 | |
1891 | 4. 0 0 7. 0 0 | |
1892 | ||
1893 | The table is internally in the right hand, idx form. However as we also | |
1894 | have to deal with the states array which is indexed by nodenum we have to | |
1895 | use TRIE_NODENUM() to convert. | |
1896 | ||
1897 | */ | |
1e2e3d02 YO |
1898 | DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, |
1899 | "%*sCompiling trie using table compiler\n", | |
1900 | (int)depth * 2 + 2, "")); | |
3dab1dad | 1901 | |
c944940b JH |
1902 | trie->trans = (reg_trie_trans *) |
1903 | PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 ) | |
1904 | * trie->uniquecharcount + 1, | |
1905 | sizeof(reg_trie_trans) ); | |
1906 | trie->states = (reg_trie_state *) | |
1907 | PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2, | |
1908 | sizeof(reg_trie_state) ); | |
a3621e74 YO |
1909 | next_alloc = trie->uniquecharcount + 1; |
1910 | ||
3dab1dad | 1911 | |
a3621e74 YO |
1912 | for ( cur = first ; cur < last ; cur = regnext( cur ) ) { |
1913 | ||
c445ea15 | 1914 | regnode * const noper = NEXTOPER( cur ); |
a28509cc AL |
1915 | const U8 *uc = (U8*)STRING( noper ); |
1916 | const U8 * const e = uc + STR_LEN( noper ); | |
a3621e74 YO |
1917 | |
1918 | U32 state = 1; /* required init */ | |
1919 | ||
1920 | U16 charid = 0; /* sanity init */ | |
1921 | U32 accept_state = 0; /* sanity init */ | |
1922 | U8 *scan = (U8*)NULL; /* sanity init */ | |
1923 | ||
1924 | STRLEN foldlen = 0; /* required init */ | |
07be1b83 | 1925 | U32 wordlen = 0; /* required init */ |
a3621e74 YO |
1926 | U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; |
1927 | ||
3dab1dad | 1928 | if ( OP(noper) != NOTHING ) { |
786e8c11 | 1929 | for ( ; uc < e ; uc += len ) { |
a3621e74 | 1930 | |
786e8c11 | 1931 | TRIE_READ_CHAR; |
a3621e74 | 1932 | |
786e8c11 YO |
1933 | if ( uvc < 256 ) { |
1934 | charid = trie->charmap[ uvc ]; | |
1935 | } else { | |
55eed653 | 1936 | SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0); |
786e8c11 | 1937 | charid = svpp ? (U16)SvIV(*svpp) : 0; |
a3621e74 | 1938 | } |
786e8c11 YO |
1939 | if ( charid ) { |
1940 | charid--; | |
1941 | if ( !trie->trans[ state + charid ].next ) { | |
1942 | trie->trans[ state + charid ].next = next_alloc; | |
1943 | trie->trans[ state ].check++; | |
2e64971a DM |
1944 | prev_states[TRIE_NODENUM(next_alloc)] |
1945 | = TRIE_NODENUM(state); | |
786e8c11 YO |
1946 | next_alloc += trie->uniquecharcount; |
1947 | } | |
1948 | state = trie->trans[ state + charid ].next; | |
1949 | } else { | |
1950 | Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); | |
1951 | } | |
1952 | /* charid is now 0 if we dont know the char read, or nonzero if we do */ | |
a3621e74 | 1953 | } |
a3621e74 | 1954 | } |
3dab1dad YO |
1955 | accept_state = TRIE_NODENUM( state ); |
1956 | TRIE_HANDLE_WORD(accept_state); | |
a3621e74 YO |
1957 | |
1958 | } /* end second pass */ | |
1959 | ||
3dab1dad | 1960 | /* and now dump it out before we compress it */ |
2b8b4781 NC |
1961 | DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap, |
1962 | revcharmap, | |
1963 | next_alloc, depth+1)); | |
a3621e74 | 1964 | |
a3621e74 YO |
1965 | { |
1966 | /* | |
1967 | * Inplace compress the table.* | |
1968 | ||
1969 | For sparse data sets the table constructed by the trie algorithm will | |
1970 | be mostly 0/FAIL transitions or to put it another way mostly empty. | |
1971 | (Note that leaf nodes will not contain any transitions.) | |
1972 | ||
1973 | This algorithm compresses the tables by eliminating most such | |
1974 | transitions, at the cost of a modest bit of extra work during lookup: | |
1975 | ||
1976 | - Each states[] entry contains a .base field which indicates the | |
1977 | index in the state[] array wheres its transition data is stored. | |
1978 | ||
3b753521 | 1979 | - If .base is 0 there are no valid transitions from that node. |
a3621e74 YO |
1980 | |
1981 | - If .base is nonzero then charid is added to it to find an entry in | |
1982 | the trans array. | |
1983 | ||
1984 | -If trans[states[state].base+charid].check!=state then the | |
1985 | transition is taken to be a 0/Fail transition. Thus if there are fail | |
1986 | transitions at the front of the node then the .base offset will point | |
1987 | somewhere inside the previous nodes data (or maybe even into a node | |
1988 | even earlier), but the .check field determines if the transition is | |
1989 | valid. | |
1990 | ||
786e8c11 | 1991 | XXX - wrong maybe? |
a3621e74 | 1992 | The following process inplace converts the table to the compressed |
3b753521 | 1993 | table: We first do not compress the root node 1,and mark all its |
a3621e74 | 1994 | .check pointers as 1 and set its .base pointer as 1 as well. This |
3b753521 FN |
1995 | allows us to do a DFA construction from the compressed table later, |
1996 | and ensures that any .base pointers we calculate later are greater | |
1997 | than 0. | |
a3621e74 YO |
1998 | |
1999 | - We set 'pos' to indicate the first entry of the second node. | |
2000 | ||
2001 | - We then iterate over the columns of the node, finding the first and | |
2002 | last used entry at l and m. We then copy l..m into pos..(pos+m-l), | |
2003 | and set the .check pointers accordingly, and advance pos | |
2004 | appropriately and repreat for the next node. Note that when we copy | |
2005 | the next pointers we have to convert them from the original | |
2006 | NODEIDX form to NODENUM form as the former is not valid post | |
2007 | compression. | |
2008 | ||
2009 | - If a node has no transitions used we mark its base as 0 and do not | |
2010 | advance the pos pointer. | |
2011 | ||
2012 | - If a node only has one transition we use a second pointer into the | |
2013 | structure to fill in allocated fail transitions from other states. | |
2014 | This pointer is independent of the main pointer and scans forward | |
2015 | looking for null transitions that are allocated to a state. When it | |
2016 | finds one it writes the single transition into the "hole". If the | |
786e8c11 | 2017 | pointer doesnt find one the single transition is appended as normal. |
a3621e74 YO |
2018 | |
2019 | - Once compressed we can Renew/realloc the structures to release the | |
2020 | excess space. | |
2021 | ||
2022 | See "Table-Compression Methods" in sec 3.9 of the Red Dragon, | |
2023 | specifically Fig 3.47 and the associated pseudocode. | |
2024 | ||
2025 | demq | |
2026 | */ | |
a3b680e6 | 2027 | const U32 laststate = TRIE_NODENUM( next_alloc ); |
a28509cc | 2028 | U32 state, charid; |
a3621e74 | 2029 | U32 pos = 0, zp=0; |
1e2e3d02 | 2030 | trie->statecount = laststate; |
a3621e74 YO |
2031 | |
2032 | for ( state = 1 ; state < laststate ; state++ ) { | |
2033 | U8 flag = 0; | |
a28509cc AL |
2034 | const U32 stateidx = TRIE_NODEIDX( state ); |
2035 | const U32 o_used = trie->trans[ stateidx ].check; | |
2036 | U32 used = trie->trans[ stateidx ].check; | |
a3621e74 YO |
2037 | trie->trans[ stateidx ].check = 0; |
2038 | ||
2039 | for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) { | |
2040 | if ( flag || trie->trans[ stateidx + charid ].next ) { | |
2041 | if ( trie->trans[ stateidx + charid ].next ) { | |
2042 | if (o_used == 1) { | |
2043 | for ( ; zp < pos ; zp++ ) { | |
2044 | if ( ! trie->trans[ zp ].next ) { | |
2045 | break; | |
2046 | } | |
2047 | } | |
2048 | trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ; | |
2049 | trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); | |
2050 | trie->trans[ zp ].check = state; | |
2051 | if ( ++zp > pos ) pos = zp; | |
2052 | break; | |
2053 | } | |
2054 | used--; | |
2055 | } | |
2056 | if ( !flag ) { | |
2057 | flag = 1; | |
2058 | trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ; | |
2059 | } | |
2060 | trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); | |
2061 | trie->trans[ pos ].check = state; | |
2062 | pos++; | |
2063 | } | |
2064 | } | |
2065 | } | |
cc601c31 | 2066 | trie->lasttrans = pos + 1; |
c944940b JH |
2067 | trie->states = (reg_trie_state *) |
2068 | PerlMemShared_realloc( trie->states, laststate | |
2069 | * sizeof(reg_trie_state) ); | |
a3621e74 | 2070 | DEBUG_TRIE_COMPILE_MORE_r( |
e4584336 | 2071 | PerlIO_printf( Perl_debug_log, |
3dab1dad YO |
2072 | "%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n", |
2073 | (int)depth * 2 + 2,"", | |
2074 | (int)( ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1 ), | |
5d7488b2 AL |
2075 | (IV)next_alloc, |
2076 | (IV)pos, | |
a3621e74 YO |
2077 | ( ( next_alloc - pos ) * 100 ) / (double)next_alloc ); |
2078 | ); | |
2079 | ||
2080 | } /* end table compress */ | |
2081 | } | |
1e2e3d02 YO |
2082 | DEBUG_TRIE_COMPILE_MORE_r( |
2083 | PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n", | |
2084 | (int)depth * 2 + 2, "", | |
2085 | (UV)trie->statecount, | |
2086 | (UV)trie->lasttrans) | |
2087 | ); | |
cc601c31 | 2088 | /* resize the trans array to remove unused space */ |
c944940b JH |
2089 | trie->trans = (reg_trie_trans *) |
2090 | PerlMemShared_realloc( trie->trans, trie->lasttrans | |
2091 | * sizeof(reg_trie_trans) ); | |
a3621e74 | 2092 | |
3b753521 | 2093 | { /* Modify the program and insert the new TRIE node */ |
3dab1dad YO |
2094 | U8 nodetype =(U8)(flags & 0xFF); |
2095 | char *str=NULL; | |
786e8c11 | 2096 | |
07be1b83 | 2097 | #ifdef DEBUGGING |
e62cc96a | 2098 | regnode *optimize = NULL; |
7122b237 YO |
2099 | #ifdef RE_TRACK_PATTERN_OFFSETS |
2100 | ||
b57a0404 JH |
2101 | U32 mjd_offset = 0; |
2102 | U32 mjd_nodelen = 0; | |
7122b237 YO |
2103 | #endif /* RE_TRACK_PATTERN_OFFSETS */ |
2104 | #endif /* DEBUGGING */ | |
a3621e74 | 2105 | /* |
3dab1dad YO |
2106 | This means we convert either the first branch or the first Exact, |
2107 | depending on whether the thing following (in 'last') is a branch | |
2108 | or not and whther first is the startbranch (ie is it a sub part of | |
2109 | the alternation or is it the whole thing.) | |
3b753521 | 2110 | Assuming its a sub part we convert the EXACT otherwise we convert |
3dab1dad | 2111 | the whole branch sequence, including the first. |
a3621e74 | 2112 | */ |
3dab1dad | 2113 | /* Find the node we are going to overwrite */ |
7f69552c | 2114 | if ( first != startbranch || OP( last ) == BRANCH ) { |
07be1b83 | 2115 | /* branch sub-chain */ |
3dab1dad | 2116 | NEXT_OFF( first ) = (U16)(last - first); |
7122b237 | 2117 | #ifdef RE_TRACK_PATTERN_OFFSETS |
07be1b83 YO |
2118 | DEBUG_r({ |
2119 | mjd_offset= Node_Offset((convert)); | |
2120 | mjd_nodelen= Node_Length((convert)); | |
2121 | }); | |
7122b237 | 2122 | #endif |
7f69552c | 2123 | /* whole branch chain */ |
7122b237 YO |
2124 | } |
2125 | #ifdef RE_TRACK_PATTERN_OFFSETS | |
2126 | else { | |
7f69552c YO |
2127 | DEBUG_r({ |
2128 | const regnode *nop = NEXTOPER( convert ); | |
2129 | mjd_offset= Node_Offset((nop)); | |
2130 | mjd_nodelen= Node_Length((nop)); | |
2131 | }); | |
07be1b83 YO |
2132 | } |
2133 | DEBUG_OPTIMISE_r( | |
2134 | PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n", | |
2135 | (int)depth * 2 + 2, "", | |
786e8c11 | 2136 | (UV)mjd_offset, (UV)mjd_nodelen) |
07be1b83 | 2137 | ); |
7122b237 | 2138 | #endif |
3dab1dad YO |
2139 | /* But first we check to see if there is a common prefix we can |
2140 | split out as an EXACT and put in front of the TRIE node. */ | |
2141 | trie->startstate= 1; | |
55eed653 | 2142 | if ( trie->bitmap && !widecharmap && !trie->jump ) { |
3dab1dad | 2143 | U32 state; |
1e2e3d02 | 2144 | for ( state = 1 ; state < trie->statecount-1 ; state++ ) { |
a3621e74 | 2145 | U32 ofs = 0; |
8e11feef RGS |
2146 | I32 idx = -1; |
2147 | U32 count = 0; | |
2148 | const U32 base = trie->states[ state ].trans.base; | |
a3621e74 | 2149 | |
3dab1dad | 2150 | if ( trie->states[state].wordnum ) |
8e11feef | 2151 | count = 1; |
a3621e74 | 2152 | |
8e11feef | 2153 | for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { |
cc601c31 YO |
2154 | if ( ( base + ofs >= trie->uniquecharcount ) && |
2155 | ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && | |
a3621e74 YO |
2156 | trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) |
2157 | { | |
3dab1dad | 2158 | if ( ++count > 1 ) { |
2b8b4781 | 2159 | SV **tmp = av_fetch( revcharmap, ofs, 0); |
07be1b83 | 2160 | const U8 *ch = (U8*)SvPV_nolen_const( *tmp ); |
8e11feef | 2161 | if ( state == 1 ) break; |
3dab1dad YO |
2162 | if ( count == 2 ) { |
2163 | Zero(trie->bitmap, ANYOF_BITMAP_SIZE, char); | |
2164 | DEBUG_OPTIMISE_r( | |
8e11feef RGS |
2165 | PerlIO_printf(Perl_debug_log, |
2166 | "%*sNew Start State=%"UVuf" Class: [", | |
2167 | (int)depth * 2 + 2, "", | |
786e8c11 | 2168 | (UV)state)); |
be8e71aa | 2169 | if (idx >= 0) { |
2b8b4781 | 2170 | SV ** const tmp = av_fetch( revcharmap, idx, 0); |
be8e71aa | 2171 | const U8 * const ch = (U8*)SvPV_nolen_const( *tmp ); |
8e11feef | 2172 | |
3dab1dad | 2173 | TRIE_BITMAP_SET(trie,*ch); |
8e11feef RGS |
2174 | if ( folder ) |
2175 | TRIE_BITMAP_SET(trie, folder[ *ch ]); | |
3dab1dad | 2176 | DEBUG_OPTIMISE_r( |
f1f66076 | 2177 | PerlIO_printf(Perl_debug_log, "%s", (char*)ch) |
3dab1dad | 2178 | ); |
8e11feef RGS |
2179 | } |
2180 | } | |
2181 | TRIE_BITMAP_SET(trie,*ch); | |
2182 | if ( folder ) | |
2183 | TRIE_BITMAP_SET(trie,folder[ *ch ]); | |
2184 | DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"%s", ch)); | |
2185 | } | |
2186 | idx = ofs; | |
2187 | } | |
3dab1dad YO |
2188 | } |
2189 | if ( count == 1 ) { | |
2b8b4781 | 2190 | SV **tmp = av_fetch( revcharmap, idx, 0); |
c490c714 YO |
2191 | STRLEN len; |
2192 | char *ch = SvPV( *tmp, len ); | |
de734bd5 A |
2193 | DEBUG_OPTIMISE_r({ |
2194 | SV *sv=sv_newmortal(); | |
8e11feef RGS |
2195 | PerlIO_printf( Perl_debug_log, |
2196 | "%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n", | |
2197 | (int)depth * 2 + 2, "", | |
de734bd5 A |
2198 | (UV)state, (UV)idx, |
2199 | pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6, | |
2200 | PL_colors[0], PL_colors[1], | |
2201 | (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) | | |
2202 | PERL_PV_ESCAPE_FIRSTCHAR | |
2203 | ) | |
2204 | ); | |
2205 | }); | |
3dab1dad YO |
2206 | if ( state==1 ) { |
2207 | OP( convert ) = nodetype; | |
2208 | str=STRING(convert); | |
2209 | STR_LEN(convert)=0; | |
2210 | } | |
c490c714 YO |
2211 | STR_LEN(convert) += len; |
2212 | while (len--) | |
de734bd5 | 2213 | *str++ = *ch++; |
8e11feef | 2214 | } else { |
f9049ba1 | 2215 | #ifdef DEBUGGING |
8e11feef RGS |
2216 | if (state>1) |
2217 | DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n")); | |
f9049ba1 | 2218 | #endif |
8e11feef RGS |
2219 | break; |
2220 | } | |
2221 | } | |
2e64971a | 2222 | trie->prefixlen = (state-1); |
3dab1dad | 2223 | if (str) { |
8e11feef | 2224 | regnode *n = convert+NODE_SZ_STR(convert); |
07be1b83 | 2225 | NEXT_OFF(convert) = NODE_SZ_STR(convert); |
8e11feef | 2226 | trie->startstate = state; |
07be1b83 YO |
2227 | trie->minlen -= (state - 1); |
2228 | trie->maxlen -= (state - 1); | |
33809eae JH |
2229 | #ifdef DEBUGGING |
2230 | /* At least the UNICOS C compiler choked on this | |
2231 | * being argument to DEBUG_r(), so let's just have | |
2232 | * it right here. */ | |
2233 | if ( | |
2234 | #ifdef PERL_EXT_RE_BUILD | |
2235 | 1 | |
2236 | #else | |
2237 | DEBUG_r_TEST | |
2238 | #endif | |
2239 | ) { | |
2240 | regnode *fix = convert; | |
2241 | U32 word = trie->wordcount; | |
2242 | mjd_nodelen++; | |
2243 | Set_Node_Offset_Length(convert, mjd_offset, state - 1); | |
2244 | while( ++fix < n ) { | |
2245 | Set_Node_Offset_Length(fix, 0, 0); | |
2246 | } | |
2247 | while (word--) { | |
2248 | SV ** const tmp = av_fetch( trie_words, word, 0 ); | |
2249 | if (tmp) { | |
2250 | if ( STR_LEN(convert) <= SvCUR(*tmp) ) | |
2251 | sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert)); | |
2252 | else | |
2253 | sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp)); | |
2254 | } | |
2255 | } | |
2256 | } | |
2257 | #endif | |
8e11feef RGS |
2258 | if (trie->maxlen) { |
2259 | convert = n; | |
2260 | } else { | |
3dab1dad | 2261 | NEXT_OFF(convert) = (U16)(tail - convert); |
a5ca303d | 2262 | DEBUG_r(optimize= n); |
3dab1dad YO |
2263 | } |
2264 | } | |
2265 | } | |
a5ca303d YO |
2266 | if (!jumper) |
2267 | jumper = last; | |
3dab1dad | 2268 | if ( trie->maxlen ) { |
8e11feef RGS |
2269 | NEXT_OFF( convert ) = (U16)(tail - convert); |
2270 | ARG_SET( convert, data_slot ); | |
786e8c11 YO |
2271 | /* Store the offset to the first unabsorbed branch in |
2272 | jump[0], which is otherwise unused by the jump logic. | |
2273 | We use this when dumping a trie and during optimisation. */ | |
2274 | if (trie->jump) | |
7f69552c | 2275 | trie->jump[0] = (U16)(nextbranch - convert); |
a5ca303d | 2276 | |
6c48061a YO |
2277 | /* If the start state is not accepting (meaning there is no empty string/NOTHING) |
2278 | * and there is a bitmap | |
2279 | * and the first "jump target" node we found leaves enough room | |
2280 | * then convert the TRIE node into a TRIEC node, with the bitmap | |
2281 | * embedded inline in the opcode - this is hypothetically faster. | |
2282 | */ | |
2283 | if ( !trie->states[trie->startstate].wordnum | |
2284 | && trie->bitmap | |
2285 | && ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) ) | |
786e8c11 YO |
2286 | { |
2287 | OP( convert ) = TRIEC; | |
2288 | Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char); | |
446bd890 | 2289 | PerlMemShared_free(trie->bitmap); |
786e8c11 YO |
2290 | trie->bitmap= NULL; |
2291 | } else | |
2292 | OP( convert ) = TRIE; | |
a3621e74 | 2293 | |
3dab1dad YO |
2294 | /* store the type in the flags */ |
2295 | convert->flags = nodetype; | |
a5ca303d YO |
2296 | DEBUG_r({ |
2297 | optimize = convert | |
2298 | + NODE_STEP_REGNODE | |
2299 | + regarglen[ OP( convert ) ]; | |
2300 | }); | |
2301 | /* XXX We really should free up the resource in trie now, | |
2302 | as we won't use them - (which resources?) dmq */ | |
3dab1dad | 2303 | } |
a3621e74 | 2304 | /* needed for dumping*/ |
e62cc96a | 2305 | DEBUG_r(if (optimize) { |
07be1b83 | 2306 | regnode *opt = convert; |
bcdf7404 | 2307 | |
e62cc96a | 2308 | while ( ++opt < optimize) { |
07be1b83 YO |
2309 | Set_Node_Offset_Length(opt,0,0); |
2310 | } | |
786e8c11 YO |
2311 | /* |
2312 | Try to clean up some of the debris left after the | |
2313 | optimisation. | |
a3621e74 | 2314 | */ |
786e8c11 | 2315 | while( optimize < jumper ) { |
07be1b83 | 2316 | mjd_nodelen += Node_Length((optimize)); |
a3621e74 | 2317 | OP( optimize ) = OPTIMIZED; |
07be1b83 | 2318 | Set_Node_Offset_Length(optimize,0,0); |
a3621e74 YO |
2319 | optimize++; |
2320 | } | |
07be1b83 | 2321 | Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen); |
a3621e74 YO |
2322 | }); |
2323 | } /* end node insert */ | |
2e64971a DM |
2324 | |
2325 | /* Finish populating the prev field of the wordinfo array. Walk back | |
2326 | * from each accept state until we find another accept state, and if | |
2327 | * so, point the first word's .prev field at the second word. If the | |
2328 | * second already has a .prev field set, stop now. This will be the | |
2329 | * case either if we've already processed that word's accept state, | |
3b753521 FN |
2330 | * or that state had multiple words, and the overspill words were |
2331 | * already linked up earlier. | |
2e64971a DM |
2332 | */ |
2333 | { | |
2334 | U16 word; | |
2335 | U32 state; | |
2336 | U16 prev; | |
2337 | ||
2338 | for (word=1; word <= trie->wordcount; word++) { | |
2339 | prev = 0; | |
2340 | if (trie->wordinfo[word].prev) | |
2341 | continue; | |
2342 | state = trie->wordinfo[word].accept; | |
2343 | while (state) { | |
2344 | state = prev_states[state]; | |
2345 | if (!state) | |
2346 | break; | |
2347 | prev = trie->states[state].wordnum; | |
2348 | if (prev) | |
2349 | break; | |
2350 | } | |
2351 | trie->wordinfo[word].prev = prev; | |
2352 | } | |
2353 | Safefree(prev_states); | |
2354 | } | |
2355 | ||
2356 | ||
2357 | /* and now dump out the compressed format */ | |
2358 | DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1)); | |
2359 | ||
55eed653 | 2360 | RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap; |
2b8b4781 NC |
2361 | #ifdef DEBUGGING |
2362 | RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words; | |
2363 | RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap; | |
2364 | #else | |
2365 | SvREFCNT_dec(revcharmap); | |
07be1b83 | 2366 | #endif |
786e8c11 YO |
2367 | return trie->jump |
2368 | ? MADE_JUMP_TRIE | |
2369 | : trie->startstate>1 | |
2370 | ? MADE_EXACT_TRIE | |
2371 | : MADE_TRIE; | |
2372 | } | |
2373 | ||
2374 | STATIC void | |
2375 | S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth) | |
2376 | { | |
3b753521 | 2377 | /* The Trie is constructed and compressed now so we can build a fail array if it's needed |
786e8c11 YO |
2378 | |
2379 | This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the | |
2380 | "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88 | |
2381 | ISBN 0-201-10088-6 | |
2382 | ||
2383 | We find the fail state for each state in the trie, this state is the longest proper | |
3b753521 FN |
2384 | suffix of the current state's 'word' that is also a proper prefix of another word in our |
2385 | trie. State 1 represents the word '' and is thus the default fail state. This allows | |
786e8c11 YO |
2386 | the DFA not to have to restart after its tried and failed a word at a given point, it |
2387 | simply continues as though it had been matching the other word in the first place. | |
2388 | Consider | |
2389 | 'abcdgu'=~/abcdefg|cdgu/ | |
2390 | When we get to 'd' we are still matching the first word, we would encounter 'g' which would | |
3b753521 FN |
2391 | fail, which would bring us to the state representing 'd' in the second word where we would |
2392 | try 'g' and succeed, proceeding to match 'cdgu'. | |
786e8c11 YO |
2393 | */ |
2394 | /* add a fail transition */ | |
3251b653 NC |
2395 | const U32 trie_offset = ARG(source); |
2396 | reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset]; | |
786e8c11 YO |
2397 | U32 *q; |
2398 | const U32 ucharcount = trie->uniquecharcount; | |
1e2e3d02 | 2399 | const U32 numstates = trie->statecount; |
786e8c11 YO |
2400 | const U32 ubound = trie->lasttrans + ucharcount; |
2401 | U32 q_read = 0; | |
2402 | U32 q_write = 0; | |
2403 | U32 charid; | |
2404 | U32 base = trie->states[ 1 ].trans.base; | |
2405 | U32 *fail; | |
2406 | reg_ac_data *aho; | |
2407 | const U32 data_slot = add_data( pRExC_state, 1, "T" ); | |
2408 | GET_RE_DEBUG_FLAGS_DECL; | |
7918f24d NC |
2409 | |
2410 | PERL_ARGS_ASSERT_MAKE_TRIE_FAILTABLE; | |
786e8c11 YO |
2411 | #ifndef DEBUGGING |
2412 | PERL_UNUSED_ARG(depth); | |
2413 | #endif | |
2414 | ||
2415 | ||
2416 | ARG_SET( stclass, data_slot ); | |
c944940b | 2417 | aho = (reg_ac_data *) PerlMemShared_calloc( 1, sizeof(reg_ac_data) ); |
f8fc2ecf | 2418 | RExC_rxi->data->data[ data_slot ] = (void*)aho; |
3251b653 | 2419 | aho->trie=trie_offset; |
446bd890 NC |
2420 | aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) ); |
2421 | Copy( trie->states, aho->states, numstates, reg_trie_state ); | |
786e8c11 | 2422 | Newxz( q, numstates, U32); |
c944940b | 2423 | aho->fail = (U32 *) PerlMemShared_calloc( numstates, sizeof(U32) ); |
786e8c11 YO |
2424 | aho->refcount = 1; |
2425 | fail = aho->fail; | |
2426 | /* initialize fail[0..1] to be 1 so that we always have | |
2427 | a valid final fail state */ | |
2428 | fail[ 0 ] = fail[ 1 ] = 1; | |
2429 | ||
2430 | for ( charid = 0; charid < ucharcount ; charid++ ) { | |
2431 | const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 ); | |
2432 | if ( newstate ) { | |
2433 | q[ q_write ] = newstate; | |
2434 | /* set to point at the root */ | |
2435 | fail[ q[ q_write++ ] ]=1; | |
2436 | } | |
2437 | } | |
2438 | while ( q_read < q_write) { | |
2439 | const U32 cur = q[ q_read++ % numstates ]; | |
2440 | base = trie->states[ cur ].trans.base; | |
2441 | ||
2442 | for ( charid = 0 ; charid < ucharcount ; charid++ ) { | |
2443 | const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 ); | |
2444 | if (ch_state) { | |
2445 | U32 fail_state = cur; | |
2446 | U32 fail_base; | |
2447 | do { | |
2448 | fail_state = fail[ fail_state ]; | |
2449 | fail_base = aho->states[ fail_state ].trans.base; | |
2450 | } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) ); | |
2451 | ||
2452 | fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ); | |
2453 | fail[ ch_state ] = fail_state; | |
2454 | if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum ) | |
2455 | { | |
2456 | aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum; | |
2457 | } | |
2458 | q[ q_write++ % numstates] = ch_state; | |
2459 | } | |
2460 | } | |
2461 | } | |
2462 | /* restore fail[0..1] to 0 so that we "fall out" of the AC loop | |
2463 | when we fail in state 1, this allows us to use the | |
2464 | charclass scan to find a valid start char. This is based on the principle | |
2465 | that theres a good chance the string being searched contains lots of stuff | |
2466 | that cant be a start char. | |
2467 | */ | |
2468 | fail[ 0 ] = fail[ 1 ] = 0; | |
2469 | DEBUG_TRIE_COMPILE_r({ | |
6d99fb9b JH |
2470 | PerlIO_printf(Perl_debug_log, |
2471 | "%*sStclass Failtable (%"UVuf" states): 0", | |
2472 | (int)(depth * 2), "", (UV)numstates | |
1e2e3d02 | 2473 | ); |
786e8c11 YO |
2474 | for( q_read=1; q_read<numstates; q_read++ ) { |
2475 | PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]); | |
2476 | } | |
2477 | PerlIO_printf(Perl_debug_log, "\n"); | |
2478 | }); | |
2479 | Safefree(q); | |
2480 | /*RExC_seen |= REG_SEEN_TRIEDFA;*/ | |
a3621e74 YO |
2481 | } |
2482 | ||
786e8c11 | 2483 | |
a3621e74 | 2484 | /* |
5d1c421c JH |
2485 | * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2. |
2486 | * These need to be revisited when a newer toolchain becomes available. | |
2487 | */ | |
2488 | #if defined(__sparc64__) && defined(__GNUC__) | |
2489 | # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96) | |
2490 | # undef SPARC64_GCC_WORKAROUND | |
2491 | # define SPARC64_GCC_WORKAROUND 1 | |
2492 | # endif | |
2493 | #endif | |
2494 | ||
07be1b83 | 2495 | #define DEBUG_PEEP(str,scan,depth) \ |
b515a41d | 2496 | DEBUG_OPTIMISE_r({if (scan){ \ |
07be1b83 YO |
2497 | SV * const mysv=sv_newmortal(); \ |
2498 | regnode *Next = regnext(scan); \ | |
2499 | regprop(RExC_rx, mysv, scan); \ | |
7f69552c | 2500 | PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \ |
07be1b83 YO |
2501 | (int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\ |
2502 | Next ? (REG_NODE_NUM(Next)) : 0 ); \ | |
b515a41d | 2503 | }}); |
07be1b83 | 2504 | |
1de06328 YO |
2505 | |
2506 | ||
2507 | ||
2508 | ||
07be1b83 YO |
2509 | #define JOIN_EXACT(scan,min,flags) \ |
2510 | if (PL_regkind[OP(scan)] == EXACT) \ | |
2511 | join_exact(pRExC_state,(scan),(min),(flags),NULL,depth+1) | |
2512 | ||
be8e71aa | 2513 | STATIC U32 |
07be1b83 YO |
2514 | S_join_exact(pTHX_ RExC_state_t *pRExC_state, regnode *scan, I32 *min, U32 flags,regnode *val, U32 depth) { |
2515 | /* Merge several consecutive EXACTish nodes into one. */ | |
2516 | regnode *n = regnext(scan); | |
2517 | U32 stringok = 1; | |
2518 | regnode *next = scan + NODE_SZ_STR(scan); | |
2519 | U32 merged = 0; | |
2520 | U32 stopnow = 0; | |
2521 | #ifdef DEBUGGING | |
2522 | regnode *stop = scan; | |
72f13be8 | 2523 | GET_RE_DEBUG_FLAGS_DECL; |
f9049ba1 | 2524 | #else |
d47053eb RGS |
2525 | PERL_UNUSED_ARG(depth); |
2526 | #endif | |
7918f24d NC |
2527 | |
2528 | PERL_ARGS_ASSERT_JOIN_EXACT; | |
d47053eb | 2529 | #ifndef EXPERIMENTAL_INPLACESCAN |
f9049ba1 SP |
2530 | PERL_UNUSED_ARG(flags); |
2531 | PERL_UNUSED_ARG(val); | |
07be1b83 | 2532 | #endif |
07be1b83 YO |
2533 | DEBUG_PEEP("join",scan,depth); |
2534 | ||
2535 | /* Skip NOTHING, merge EXACT*. */ | |
2536 | while (n && | |
2537 | ( PL_regkind[OP(n)] == NOTHING || | |
2538 | (stringok && (OP(n) == OP(scan)))) | |
2539 | && NEXT_OFF(n) | |
2540 | && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) { | |
2541 | ||
2542 | if (OP(n) == TAIL || n > next) | |
2543 | stringok = 0; | |
2544 | if (PL_regkind[OP(n)] == NOTHING) { | |
07be1b83 YO |
2545 | DEBUG_PEEP("skip:",n,depth); |
2546 | NEXT_OFF(scan) += NEXT_OFF(n); | |
2547 | next = n + NODE_STEP_REGNODE; | |
2548 | #ifdef DEBUGGING | |
2549 | if (stringok) | |
2550 | stop = n; | |
2551 | #endif | |
2552 | n = regnext(n); | |
2553 | } | |
2554 | else if (stringok) { | |
786e8c11 | 2555 | const unsigned int oldl = STR_LEN(scan); |
07be1b83 YO |
2556 | regnode * const nnext = regnext(n); |
2557 | ||
2558 | DEBUG_PEEP("merg",n,depth); | |
2559 | ||
2560 | merged++; | |
2561 | if (oldl + STR_LEN(n) > U8_MAX) | |
2562 | break; | |
2563 | NEXT_OFF(scan) += NEXT_OFF(n); | |
2564 | STR_LEN(scan) += STR_LEN(n); | |
2565 | next = n + NODE_SZ_STR(n); | |
2566 | /* Now we can overwrite *n : */ | |
2567 | Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char); | |
2568 | #ifdef DEBUGGING | |
2569 | stop = next - 1; | |
2570 | #endif | |
2571 | n = nnext; | |
2572 | if (stopnow) break; | |
2573 | } | |
2574 | ||
d47053eb RGS |
2575 | #ifdef EXPERIMENTAL_INPLACESCAN |
2576 | if (flags && !NEXT_OFF(n)) { | |
2577 | DEBUG_PEEP("atch", val, depth); | |
2578 | if (reg_off_by_arg[OP(n)]) { | |
2579 | ARG_SET(n, val - n); | |
2580 | } | |
2581 | else { | |
2582 | NEXT_OFF(n) = val - n; | |
2583 | } | |
2584 | stopnow = 1; | |
2585 | } | |
07be1b83 YO |
2586 | #endif |
2587 | } | |
ced7f090 KW |
2588 | #define GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS 0x0390 |
2589 | #define IOTA_D_T GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS | |
2590 | #define GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS 0x03B0 | |
2591 | #define UPSILON_D_T GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS | |
2c2b7f86 KW |
2592 | |
2593 | if (UTF | |
2f7f8cb1 | 2594 | && ( OP(scan) == EXACTF || OP(scan) == EXACTFU || OP(scan) == EXACTFA) |
2c2b7f86 KW |
2595 | && ( STR_LEN(scan) >= 6 ) ) |
2596 | { | |
07be1b83 YO |
2597 | /* |
2598 | Two problematic code points in Unicode casefolding of EXACT nodes: | |
2599 | ||
2600 | U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS | |
2601 | U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS | |
2602 | ||
2603 | which casefold to | |
2604 | ||
2605 | Unicode UTF-8 | |
2606 | ||
2607 | U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81 | |
2608 | U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81 | |
2609 | ||
2610 | This means that in case-insensitive matching (or "loose matching", | |
2611 | as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte | |
2612 | length of the above casefolded versions) can match a target string | |
2613 | of length two (the byte length of UTF-8 encoded U+0390 or U+03B0). | |
2614 | This would rather mess up the minimum length computation. | |
2615 | ||
2616 | What we'll do is to look for the tail four bytes, and then peek | |
2617 | at the preceding two bytes to see whether we need to decrease | |
2618 | the minimum length by four (six minus two). | |
2619 | ||
2620 | Thanks to the design of UTF-8, there cannot be false matches: | |
2621 | A sequence of valid UTF-8 bytes cannot be a subsequence of | |
2622 | another valid sequence of UTF-8 bytes. | |
2623 | ||
2624 | */ | |
2625 | char * const s0 = STRING(scan), *s, *t; | |
2626 | char * const s1 = s0 + STR_LEN(scan) - 1; | |
2627 | char * const s2 = s1 - 4; | |
e294cc5d JH |
2628 | #ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */ |
2629 | const char t0[] = "\xaf\x49\xaf\x42"; | |
2630 | #else | |
07be1b83 | 2631 | const char t0[] = "\xcc\x88\xcc\x81"; |
e294cc5d | 2632 | #endif |
07be1b83 YO |
2633 | const char * const t1 = t0 + 3; |
2634 | ||
2635 | for (s = s0 + 2; | |
2636 | s < s2 && (t = ninstr(s, s1, t0, t1)); | |
2637 | s = t + 4) { | |
e294cc5d JH |
2638 | #ifdef EBCDIC |
2639 | if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) || | |
2640 | ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5)) | |
2641 | #else | |
07be1b83 YO |
2642 | if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) || |
2643 | ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF)) | |
e294cc5d | 2644 | #endif |
07be1b83 YO |
2645 | *min -= 4; |
2646 | } | |
2647 | } | |
2648 | ||
2649 | #ifdef DEBUGGING | |
bb789b09 DM |
2650 | /* Allow dumping but overwriting the collection of skipped |
2651 | * ops and/or strings with fake optimized ops */ | |
07be1b83 YO |
2652 | n = scan + NODE_SZ_STR(scan); |
2653 | while (n <= stop) { | |
bb789b09 DM |
2654 | OP(n) = OPTIMIZED; |
2655 | FLAGS(n) = 0; | |
2656 | NEXT_OFF(n) = 0; | |
07be1b83 YO |
2657 | n++; |
2658 | } | |
2659 | #endif | |
2660 | DEBUG_OPTIMISE_r(if (merged){DEBUG_PEEP("finl",scan,depth)}); | |
2661 | return stopnow; | |
2662 | } | |
2663 | ||
486ec47a | 2664 | /* REx optimizer. Converts nodes into quicker variants "in place". |
653099ff GS |
2665 | Finds fixed substrings. */ |
2666 | ||
a0288114 | 2667 | /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set |
c277df42 IZ |
2668 | to the position after last scanned or to NULL. */ |
2669 | ||
40d049e4 YO |
2670 | #define INIT_AND_WITHP \ |
2671 | assert(!and_withp); \ | |
2672 | Newx(and_withp,1,struct regnode_charclass_class); \ | |
2673 | SAVEFREEPV(and_withp) | |
07be1b83 | 2674 | |
b515a41d | 2675 | /* this is a chain of data about sub patterns we are processing that |
486ec47a | 2676 | need to be handled separately/specially in study_chunk. Its so |
b515a41d YO |
2677 | we can simulate recursion without losing state. */ |
2678 | struct scan_frame; | |
2679 | typedef struct scan_frame { | |
2680 | regnode *last; /* last node to process in this frame */ | |
2681 | regnode *next; /* next node to process when last is reached */ | |
2682 | struct scan_frame *prev; /*previous frame*/ | |
2683 | I32 stop; /* what stopparen do we use */ | |
2684 | } scan_frame; | |
2685 | ||
304ee84b YO |
2686 | |
2687 | #define SCAN_COMMIT(s, data, m) scan_commit(s, data, m, is_inf) | |
2688 | ||
e1d1eefb YO |
2689 | #define CASE_SYNST_FNC(nAmE) \ |
2690 | case nAmE: \ | |
2691 | if (flags & SCF_DO_STCLASS_AND) { \ | |
2692 | for (value = 0; value < 256; value++) \ | |
2693 | if (!is_ ## nAmE ## _cp(value)) \ | |
2694 | ANYOF_BITMAP_CLEAR(data->start_class, value); \ | |
2695 | } \ | |
2696 | else { \ | |
2697 | for (value = 0; value < 256; value++) \ | |
2698 | if (is_ ## nAmE ## _cp(value)) \ | |
2699 | ANYOF_BITMAP_SET(data->start_class, value); \ | |
2700 | } \ | |
2701 | break; \ | |
2702 | case N ## nAmE: \ | |
2703 | if (flags & SCF_DO_STCLASS_AND) { \ | |
2704 | for (value = 0; value < 256; value++) \ | |
2705 | if (is_ ## nAmE ## _cp(value)) \ | |
2706 | ANYOF_BITMAP_CLEAR(data->start_class, value); \ | |
2707 | } \ | |
2708 | else { \ | |
2709 | for (value = 0; value < 256; value++) \ | |
2710 | if (!is_ ## nAmE ## _cp(value)) \ | |
2711 | ANYOF_BITMAP_SET(data->start_class, value); \ | |
2712 | } \ | |
2713 | break | |
2714 | ||
2715 | ||
2716 | ||
76e3520e | 2717 | STATIC I32 |
40d049e4 | 2718 | S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, |
1de06328 | 2719 | I32 *minlenp, I32 *deltap, |
40d049e4 YO |
2720 | regnode *last, |
2721 | scan_data_t *data, | |
2722 | I32 stopparen, | |
2723 | U8* recursed, | |
2724 | struct regnode_charclass_class *and_withp, | |
2725 | U32 flags, U32 depth) | |
c277df42 IZ |
2726 | /* scanp: Start here (read-write). */ |
2727 | /* deltap: Write maxlen-minlen here. */ | |
2728 | /* last: Stop before this one. */ | |
40d049e4 YO |
2729 | /* data: string data about the pattern */ |
2730 | /* stopparen: treat close N as END */ | |
2731 | /* recursed: which subroutines have we recursed into */ | |
2732 | /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */ | |
c277df42 | 2733 | { |
97aff369 | 2734 | dVAR; |
c277df42 IZ |
2735 | I32 min = 0, pars = 0, code; |
2736 | regnode *scan = *scanp, *next; | |
2737 | I32 delta = 0; | |
2738 | int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF); | |
aca2d497 | 2739 | int is_inf_internal = 0; /* The studied chunk is infinite */ |
c277df42 IZ |
2740 | I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0; |
2741 | scan_data_t data_fake; | |
a3621e74 | 2742 | SV *re_trie_maxbuff = NULL; |
786e8c11 | 2743 | regnode *first_non_open = scan; |
e2e6a0f1 | 2744 | I32 stopmin = I32_MAX; |
8aa23a47 | 2745 | scan_frame *frame = NULL; |
a3621e74 | 2746 | GET_RE_DEBUG_FLAGS_DECL; |
8aa23a47 | 2747 | |
7918f24d NC |
2748 | PERL_ARGS_ASSERT_STUDY_CHUNK; |
2749 | ||
13a24bad | 2750 | #ifdef DEBUGGING |
40d049e4 | 2751 | StructCopy(&zero_scan_data, &data_fake, scan_data_t); |
13a24bad | 2752 | #endif |
40d049e4 | 2753 | |
786e8c11 | 2754 | if ( depth == 0 ) { |
40d049e4 | 2755 | while (first_non_open && OP(first_non_open) == OPEN) |
786e8c11 YO |
2756 | first_non_open=regnext(first_non_open); |
2757 | } | |
2758 | ||
b81d288d | 2759 | |
8aa23a47 YO |
2760 | fake_study_recurse: |
2761 | while ( scan && OP(scan) != END && scan < last ){ | |
2762 | /* Peephole optimizer: */ | |
304ee84b | 2763 | DEBUG_STUDYDATA("Peep:", data,depth); |
8aa23a47 YO |
2764 | DEBUG_PEEP("Peep",scan,depth); |
2765 | JOIN_EXACT(scan,&min,0); | |
2766 | ||
2767 | /* Follow the next-chain of the current node and optimize | |
2768 | away all the NOTHINGs from it. */ | |
2769 | if (OP(scan) != CURLYX) { | |
2770 | const int max = (reg_off_by_arg[OP(scan)] | |
2771 | ? I32_MAX | |
2772 | /* I32 may be smaller than U16 on CRAYs! */ | |
2773 | : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX)); | |
2774 | int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan)); | |
2775 | int noff; | |
2776 | regnode *n = scan; | |
686b73d4 | 2777 | |
8aa23a47 YO |
2778 | /* Skip NOTHING and LONGJMP. */ |
2779 | while ((n = regnext(n)) | |
2780 | && ((PL_regkind[OP(n)] == NOTHING && (noff = NEXT_OFF(n))) | |
2781 | || ((OP(n) == LONGJMP) && (noff = ARG(n)))) | |
2782 | && off + noff < max) | |
2783 | off += noff; | |
2784 | if (reg_off_by_arg[OP(scan)]) | |
2785 | ARG(scan) = off; | |
2786 | else | |
2787 | NEXT_OFF(scan) = off; | |
2788 | } | |
a3621e74 | 2789 | |
c277df42 | 2790 | |
8aa23a47 YO |
2791 | |
2792 | /* The principal pseudo-switch. Cannot be a switch, since we | |
2793 | look into several different things. */ | |
2794 | if (OP(scan) == BRANCH || OP(scan) == BRANCHJ | |
2795 | || OP(scan) == IFTHEN) { | |
2796 | next = regnext(scan); | |
2797 | code = OP(scan); | |
2798 | /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */ | |
686b73d4 | 2799 | |
8aa23a47 YO |
2800 | if (OP(next) == code || code == IFTHEN) { |
2801 | /* NOTE - There is similar code to this block below for handling | |
2802 | TRIE nodes on a re-study. If you change stuff here check there | |
2803 | too. */ | |
2804 | I32 max1 = 0, min1 = I32_MAX, num = 0; | |
2805 | struct regnode_charclass_class accum; | |
2806 | regnode * const startbranch=scan; | |
686b73d4 | 2807 | |
8aa23a47 | 2808 | if (flags & SCF_DO_SUBSTR) |
304ee84b | 2809 | SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot merge strings after this. */ |
8aa23a47 | 2810 | if (flags & SCF_DO_STCLASS) |
e755fd73 | 2811 | cl_init_zero(pRExC_state, &accum); |
8aa23a47 YO |
2812 | |
2813 | while (OP(scan) == code) { | |
2814 | I32 deltanext, minnext, f = 0, fake; | |
2815 | struct regnode_charclass_class this_class; | |
2816 | ||
2817 | num++; | |
2818 | data_fake.flags = 0; | |
2819 | if (data) { | |
2820 | data_fake.whilem_c = data->whilem_c; | |
2821 | data_fake.last_closep = data->last_closep; | |
2822 | } | |
2823 | else | |
2824 | data_fake.last_closep = &fake; | |
58e23c8d YO |
2825 | |
2826 | data_fake.pos_delta = delta; | |
8aa23a47 YO |
2827 | next = regnext(scan); |
2828 | scan = NEXTOPER(scan); | |
2829 | if (code != BRANCH) | |
c277df42 | 2830 | scan = NEXTOPER(scan); |
8aa23a47 | 2831 | if (flags & SCF_DO_STCLASS) { |
e755fd73 | 2832 | cl_init(pRExC_state, &this_class); |
8aa23a47 YO |
2833 | data_fake.start_class = &this_class; |
2834 | f = SCF_DO_STCLASS_AND; | |
58e23c8d | 2835 | } |
8aa23a47 YO |
2836 | if (flags & SCF_WHILEM_VISITED_POS) |
2837 | f |= SCF_WHILEM_VISITED_POS; | |
2838 | ||
2839 | /* we suppose the run is continuous, last=next...*/ | |
2840 | minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, | |
2841 | next, &data_fake, | |
2842 | stopparen, recursed, NULL, f,depth+1); | |
2843 | if (min1 > minnext) | |
2844 | min1 = minnext; | |
2845 | if (max1 < minnext + deltanext) | |
2846 | max1 = minnext + deltanext; | |
2847 | if (deltanext == I32_MAX) | |
2848 | is_inf = is_inf_internal = 1; | |
2849 | scan = next; | |
2850 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
2851 | pars++; | |
2852 | if (data_fake.flags & SCF_SEEN_ACCEPT) { | |
2853 | if ( stopmin > minnext) | |
2854 | stopmin = min + min1; | |
2855 | flags &= ~SCF_DO_SUBSTR; | |
2856 | if (data) | |
2857 | data->flags |= SCF_SEEN_ACCEPT; | |
2858 | } | |
2859 | if (data) { | |
2860 | if (data_fake.flags & SF_HAS_EVAL) | |
2861 | data->flags |= SF_HAS_EVAL; | |
2862 | data->whilem_c = data_fake.whilem_c; | |
3dab1dad | 2863 | } |
8aa23a47 | 2864 | if (flags & SCF_DO_STCLASS) |
3fffb88a | 2865 | cl_or(pRExC_state, &accum, &this_class); |
8aa23a47 YO |
2866 | } |
2867 | if (code == IFTHEN && num < 2) /* Empty ELSE branch */ | |
2868 | min1 = 0; | |
2869 | if (flags & SCF_DO_SUBSTR) { | |
2870 | data->pos_min += min1; | |
2871 | data->pos_delta += max1 - min1; | |
2872 | if (max1 != min1 || is_inf) | |
2873 | data->longest = &(data->longest_float); | |
2874 | } | |
2875 | min += min1; | |
2876 | delta += max1 - min1; | |
2877 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 2878 | cl_or(pRExC_state, data->start_class, &accum); |
8aa23a47 YO |
2879 | if (min1) { |
2880 | cl_and(data->start_class, and_withp); | |
2881 | flags &= ~SCF_DO_STCLASS; | |
653099ff | 2882 | } |
8aa23a47 YO |
2883 | } |
2884 | else if (flags & SCF_DO_STCLASS_AND) { | |
2885 | if (min1) { | |
2886 | cl_and(data->start_class, &accum); | |
2887 | flags &= ~SCF_DO_STCLASS; | |
de0c8cb8 | 2888 | } |
8aa23a47 YO |
2889 | else { |
2890 | /* Switch to OR mode: cache the old value of | |
2891 | * data->start_class */ | |
2892 | INIT_AND_WITHP; | |
2893 | StructCopy(data->start_class, and_withp, | |
2894 | struct regnode_charclass_class); | |
2895 | flags &= ~SCF_DO_STCLASS_AND; | |
2896 | StructCopy(&accum, data->start_class, | |
2897 | struct regnode_charclass_class); | |
2898 | flags |= SCF_DO_STCLASS_OR; | |
2899 | data->start_class->flags |= ANYOF_EOS; | |
de0c8cb8 | 2900 | } |
8aa23a47 | 2901 | } |
a3621e74 | 2902 | |
8aa23a47 YO |
2903 | if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) { |
2904 | /* demq. | |
a3621e74 | 2905 | |
8aa23a47 YO |
2906 | Assuming this was/is a branch we are dealing with: 'scan' now |
2907 | points at the item that follows the branch sequence, whatever | |
2908 | it is. We now start at the beginning of the sequence and look | |
2909 | for subsequences of | |
a3621e74 | 2910 | |
8aa23a47 YO |
2911 | BRANCH->EXACT=>x1 |
2912 | BRANCH->EXACT=>x2 | |
2913 | tail | |
a3621e74 | 2914 | |
8aa23a47 | 2915 | which would be constructed from a pattern like /A|LIST|OF|WORDS/ |
a3621e74 | 2916 | |
486ec47a | 2917 | If we can find such a subsequence we need to turn the first |
8aa23a47 YO |
2918 | element into a trie and then add the subsequent branch exact |
2919 | strings to the trie. | |
a3621e74 | 2920 | |
8aa23a47 | 2921 | We have two cases |
a3621e74 | 2922 | |
3b753521 | 2923 | 1. patterns where the whole set of branches can be converted. |
a3621e74 | 2924 | |
8aa23a47 | 2925 | 2. patterns where only a subset can be converted. |
a3621e74 | 2926 | |
8aa23a47 YO |
2927 | In case 1 we can replace the whole set with a single regop |
2928 | for the trie. In case 2 we need to keep the start and end | |
3b753521 | 2929 | branches so |
a3621e74 | 2930 | |
8aa23a47 YO |
2931 | 'BRANCH EXACT; BRANCH EXACT; BRANCH X' |
2932 | becomes BRANCH TRIE; BRANCH X; | |
786e8c11 | 2933 | |
8aa23a47 YO |
2934 | There is an additional case, that being where there is a |
2935 | common prefix, which gets split out into an EXACT like node | |
2936 | preceding the TRIE node. | |
a3621e74 | 2937 | |
8aa23a47 YO |
2938 | If x(1..n)==tail then we can do a simple trie, if not we make |
2939 | a "jump" trie, such that when we match the appropriate word | |
486ec47a | 2940 | we "jump" to the appropriate tail node. Essentially we turn |
8aa23a47 | 2941 | a nested if into a case structure of sorts. |
b515a41d | 2942 | |
8aa23a47 | 2943 | */ |
686b73d4 | 2944 | |
8aa23a47 YO |
2945 | int made=0; |
2946 | if (!re_trie_maxbuff) { | |
2947 | re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); | |
2948 | if (!SvIOK(re_trie_maxbuff)) | |
2949 | sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); | |
2950 | } | |
2951 | if ( SvIV(re_trie_maxbuff)>=0 ) { | |
2952 | regnode *cur; | |
2953 | regnode *first = (regnode *)NULL; | |
2954 | regnode *last = (regnode *)NULL; | |
2955 | regnode *tail = scan; | |
2956 | U8 optype = 0; | |
2957 | U32 count=0; | |
a3621e74 YO |
2958 | |
2959 | #ifdef DEBUGGING | |
8aa23a47 | 2960 | SV * const mysv = sv_newmortal(); /* for dumping */ |
a3621e74 | 2961 | #endif |
8aa23a47 YO |
2962 | /* var tail is used because there may be a TAIL |
2963 | regop in the way. Ie, the exacts will point to the | |
2964 | thing following the TAIL, but the last branch will | |
2965 | point at the TAIL. So we advance tail. If we | |
2966 | have nested (?:) we may have to move through several | |
2967 | tails. | |
2968 | */ | |
2969 | ||
2970 | while ( OP( tail ) == TAIL ) { | |
2971 | /* this is the TAIL generated by (?:) */ | |
2972 | tail = regnext( tail ); | |
2973 | } | |
a3621e74 | 2974 | |
8aa23a47 YO |
2975 | |
2976 | DEBUG_OPTIMISE_r({ | |
2977 | regprop(RExC_rx, mysv, tail ); | |
2978 | PerlIO_printf( Perl_debug_log, "%*s%s%s\n", | |
2979 | (int)depth * 2 + 2, "", | |
2980 | "Looking for TRIE'able sequences. Tail node is: ", | |
2981 | SvPV_nolen_const( mysv ) | |
2982 | ); | |
2983 | }); | |
2984 | ||
2985 | /* | |
2986 | ||
2987 | step through the branches, cur represents each | |
2988 | branch, noper is the first thing to be matched | |
2989 | as part of that branch and noper_next is the | |
2990 | regnext() of that node. if noper is an EXACT | |
2991 | and noper_next is the same as scan (our current | |
2992 | position in the regex) then the EXACT branch is | |
2993 | a possible optimization target. Once we have | |
486ec47a | 2994 | two or more consecutive such branches we can |
8aa23a47 YO |
2995 | create a trie of the EXACT's contents and stich |
2996 | it in place. If the sequence represents all of | |
2997 | the branches we eliminate the whole thing and | |
2998 | replace it with a single TRIE. If it is a | |
2999 | subsequence then we need to stitch it in. This | |
3000 | means the first branch has to remain, and needs | |
3001 | to be repointed at the item on the branch chain | |
3002 | following the last branch optimized. This could | |
3003 | be either a BRANCH, in which case the | |
3004 | subsequence is internal, or it could be the | |
3005 | item following the branch sequence in which | |
3006 | case the subsequence is at the end. | |
3007 | ||
3008 | */ | |
3009 | ||
3010 | /* dont use tail as the end marker for this traverse */ | |
3011 | for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) { | |
3012 | regnode * const noper = NEXTOPER( cur ); | |
b515a41d | 3013 | #if defined(DEBUGGING) || defined(NOJUMPTRIE) |
8aa23a47 | 3014 | regnode * const noper_next = regnext( noper ); |
b515a41d YO |
3015 | #endif |
3016 | ||
8aa23a47 YO |
3017 | DEBUG_OPTIMISE_r({ |
3018 | regprop(RExC_rx, mysv, cur); | |
3019 | PerlIO_printf( Perl_debug_log, "%*s- %s (%d)", | |
3020 | (int)depth * 2 + 2,"", SvPV_nolen_const( mysv ), REG_NODE_NUM(cur) ); | |
3021 | ||
3022 | regprop(RExC_rx, mysv, noper); | |
3023 | PerlIO_printf( Perl_debug_log, " -> %s", | |
3024 | SvPV_nolen_const(mysv)); | |
3025 | ||
3026 | if ( noper_next ) { | |
3027 | regprop(RExC_rx, mysv, noper_next ); | |
3028 | PerlIO_printf( Perl_debug_log,"\t=> %s\t", | |
3029 | SvPV_nolen_const(mysv)); | |
3030 | } | |
3031 | PerlIO_printf( Perl_debug_log, "(First==%d,Last==%d,Cur==%d)\n", | |
3032 | REG_NODE_NUM(first), REG_NODE_NUM(last), REG_NODE_NUM(cur) ); | |
3033 | }); | |
3034 | if ( (((first && optype!=NOTHING) ? OP( noper ) == optype | |
3035 | : PL_regkind[ OP( noper ) ] == EXACT ) | |
3036 | || OP(noper) == NOTHING ) | |
786e8c11 | 3037 | #ifdef NOJUMPTRIE |
8aa23a47 | 3038 | && noper_next == tail |
786e8c11 | 3039 | #endif |
8aa23a47 YO |
3040 | && count < U16_MAX) |
3041 | { | |
3042 | count++; | |
3043 | if ( !first || optype == NOTHING ) { | |
3044 | if (!first) first = cur; | |
3045 | optype = OP( noper ); | |
3046 | } else { | |
3047 | last = cur; | |
3048 | } | |
3049 | } else { | |
a0a388a1 | 3050 | /* |
fbebf34e KW |
3051 | Currently the trie logic handles case insensitive matching properly only |
3052 | when the pattern is UTF-8 and the node is EXACTFU (thus forcing unicode | |
3053 | semantics). | |
0abd0d78 YO |
3054 | |
3055 | If/when this is fixed the following define can be swapped | |
3056 | in below to fully enable trie logic. | |
3057 | ||
a0a388a1 | 3058 | #define TRIE_TYPE_IS_SAFE 1 |
0abd0d78 YO |
3059 | |
3060 | */ | |
fbebf34e | 3061 | #define TRIE_TYPE_IS_SAFE ((UTF && optype == EXACTFU) || optype==EXACT) |
0abd0d78 | 3062 | |
a0a388a1 | 3063 | if ( last && TRIE_TYPE_IS_SAFE ) { |
8aa23a47 YO |
3064 | make_trie( pRExC_state, |
3065 | startbranch, first, cur, tail, count, | |
3066 | optype, depth+1 ); | |
3067 | } | |
3068 | if ( PL_regkind[ OP( noper ) ] == EXACT | |
786e8c11 | 3069 | #ifdef NOJUMPTRIE |
8aa23a47 | 3070 | && noper_next == tail |
786e8c11 | 3071 | #endif |
8aa23a47 YO |
3072 | ){ |
3073 | count = 1; | |
3074 | first = cur; | |
3075 | optype = OP( noper ); | |
3076 | } else { | |
3077 | count = 0; | |
3078 | first = NULL; | |
3079 | optype = 0; | |
3080 | } | |
3081 | last = NULL; | |
3082 | } | |
3083 | } | |
3084 | DEBUG_OPTIMISE_r({ | |
3085 | regprop(RExC_rx, mysv, cur); | |
3086 | PerlIO_printf( Perl_debug_log, | |
3087 | "%*s- %s (%d) <SCAN FINISHED>\n", (int)depth * 2 + 2, | |
3088 | "", SvPV_nolen_const( mysv ),REG_NODE_NUM(cur)); | |
3089 | ||
3090 | }); | |
a0a388a1 YO |
3091 | |
3092 | if ( last && TRIE_TYPE_IS_SAFE ) { | |
8aa23a47 | 3093 | made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 ); |
686b73d4 | 3094 | #ifdef TRIE_STUDY_OPT |
8aa23a47 YO |
3095 | if ( ((made == MADE_EXACT_TRIE && |
3096 | startbranch == first) | |
3097 | || ( first_non_open == first )) && | |
3098 | depth==0 ) { | |
3099 | flags |= SCF_TRIE_RESTUDY; | |
3100 | if ( startbranch == first | |
3101 | && scan == tail ) | |
3102 | { | |
3103 | RExC_seen &=~REG_TOP_LEVEL_BRANCHES; | |
3104 | } | |
3105 | } | |
3dab1dad | 3106 | #endif |
8aa23a47 YO |
3107 | } |
3108 | } | |
3109 | ||
3110 | } /* do trie */ | |
3111 | ||
653099ff | 3112 | } |
8aa23a47 YO |
3113 | else if ( code == BRANCHJ ) { /* single branch is optimized. */ |
3114 | scan = NEXTOPER(NEXTOPER(scan)); | |
3115 | } else /* single branch is optimized. */ | |
3116 | scan = NEXTOPER(scan); | |
3117 | continue; | |
3118 | } else if (OP(scan) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) { | |
3119 | scan_frame *newframe = NULL; | |
3120 | I32 paren; | |
3121 | regnode *start; | |
3122 | regnode *end; | |
3123 | ||
3124 | if (OP(scan) != SUSPEND) { | |
3125 | /* set the pointer */ | |
3126 | if (OP(scan) == GOSUB) { | |
3127 | paren = ARG(scan); | |
3128 | RExC_recurse[ARG2L(scan)] = scan; | |
3129 | start = RExC_open_parens[paren-1]; | |
3130 | end = RExC_close_parens[paren-1]; | |
3131 | } else { | |
3132 | paren = 0; | |
f8fc2ecf | 3133 | start = RExC_rxi->program + 1; |
8aa23a47 YO |
3134 | end = RExC_opend; |
3135 | } | |
3136 | if (!recursed) { | |
3137 | Newxz(recursed, (((RExC_npar)>>3) +1), U8); | |
3138 | SAVEFREEPV(recursed); | |
3139 | } | |
3140 | if (!PAREN_TEST(recursed,paren+1)) { | |
3141 | PAREN_SET(recursed,paren+1); | |
3142 | Newx(newframe,1,scan_frame); | |
3143 | } else { | |
3144 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 3145 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 YO |
3146 | data->longest = &(data->longest_float); |
3147 | } | |
3148 | is_inf = is_inf_internal = 1; | |
3149 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ | |
3fffb88a | 3150 | cl_anything(pRExC_state, data->start_class); |
8aa23a47 YO |
3151 | flags &= ~SCF_DO_STCLASS; |
3152 | } | |
3153 | } else { | |
3154 | Newx(newframe,1,scan_frame); | |
3155 | paren = stopparen; | |
3156 | start = scan+2; | |
3157 | end = regnext(scan); | |
3158 | } | |
3159 | if (newframe) { | |
3160 | assert(start); | |
3161 | assert(end); | |
3162 | SAVEFREEPV(newframe); | |
3163 | newframe->next = regnext(scan); | |
3164 | newframe->last = last; | |
3165 | newframe->stop = stopparen; | |
3166 | newframe->prev = frame; | |
3167 | ||
3168 | frame = newframe; | |
3169 | scan = start; | |
3170 | stopparen = paren; | |
3171 | last = end; | |
3172 | ||
3173 | continue; | |
3174 | } | |
3175 | } | |
3176 | else if (OP(scan) == EXACT) { | |
3177 | I32 l = STR_LEN(scan); | |
3178 | UV uc; | |
3179 | if (UTF) { | |
3180 | const U8 * const s = (U8*)STRING(scan); | |
3181 | l = utf8_length(s, s + l); | |
3182 | uc = utf8_to_uvchr(s, NULL); | |
3183 | } else { | |
3184 | uc = *((U8*)STRING(scan)); | |
3185 | } | |
3186 | min += l; | |
3187 | if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */ | |
3188 | /* The code below prefers earlier match for fixed | |
3189 | offset, later match for variable offset. */ | |
3190 | if (data->last_end == -1) { /* Update the start info. */ | |
3191 | data->last_start_min = data->pos_min; | |
3192 | data->last_start_max = is_inf | |
3193 | ? I32_MAX : data->pos_min + data->pos_delta; | |
b515a41d | 3194 | } |
8aa23a47 YO |
3195 | sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan)); |
3196 | if (UTF) | |
3197 | SvUTF8_on(data->last_found); | |
3198 | { | |
3199 | SV * const sv = data->last_found; | |
3200 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? | |
3201 | mg_find(sv, PERL_MAGIC_utf8) : NULL; | |
3202 | if (mg && mg->mg_len >= 0) | |
3203 | mg->mg_len += utf8_length((U8*)STRING(scan), | |
3204 | (U8*)STRING(scan)+STR_LEN(scan)); | |
b515a41d | 3205 | } |
8aa23a47 YO |
3206 | data->last_end = data->pos_min + l; |
3207 | data->pos_min += l; /* As in the first entry. */ | |
3208 | data->flags &= ~SF_BEFORE_EOL; | |
3209 | } | |
3210 | if (flags & SCF_DO_STCLASS_AND) { | |
3211 | /* Check whether it is compatible with what we know already! */ | |
3212 | int compat = 1; | |
3213 | ||
54251c2e | 3214 | |
486ec47a | 3215 | /* If compatible, we or it in below. It is compatible if is |
54251c2e KW |
3216 | * in the bitmp and either 1) its bit or its fold is set, or 2) |
3217 | * it's for a locale. Even if there isn't unicode semantics | |
3218 | * here, at runtime there may be because of matching against a | |
3219 | * utf8 string, so accept a possible false positive for | |
3220 | * latin1-range folds */ | |
8aa23a47 YO |
3221 | if (uc >= 0x100 || |
3222 | (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) | |
3223 | && !ANYOF_BITMAP_TEST(data->start_class, uc) | |
39065660 | 3224 | && (!(data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) |
54251c2e | 3225 | || !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc]))) |
8aa23a47 | 3226 | ) |
d18bf9dc | 3227 | { |
8aa23a47 | 3228 | compat = 0; |
d18bf9dc | 3229 | } |
8aa23a47 YO |
3230 | ANYOF_CLASS_ZERO(data->start_class); |
3231 | ANYOF_BITMAP_ZERO(data->start_class); | |
3232 | if (compat) | |
3233 | ANYOF_BITMAP_SET(data->start_class, uc); | |
d18bf9dc KW |
3234 | else if (uc >= 0x100) { |
3235 | int i; | |
3236 | ||
3237 | /* Some Unicode code points fold to the Latin1 range; as | |
3238 | * XXX temporary code, instead of figuring out if this is | |
3239 | * one, just assume it is and set all the start class bits | |
3240 | * that could be some such above 255 code point's fold | |
3241 | * which will generate fals positives. As the code | |
3242 | * elsewhere that does compute the fold settles down, it | |
3243 | * can be extracted out and re-used here */ | |
3244 | for (i = 0; i < 256; i++){ | |
3245 | if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) { | |
3246 | ANYOF_BITMAP_SET(data->start_class, i); | |
3247 | } | |
3248 | } | |
3249 | } | |
8aa23a47 YO |
3250 | data->start_class->flags &= ~ANYOF_EOS; |
3251 | if (uc < 0x100) | |
3252 | data->start_class->flags &= ~ANYOF_UNICODE_ALL; | |
3253 | } | |
3254 | else if (flags & SCF_DO_STCLASS_OR) { | |
3255 | /* false positive possible if the class is case-folded */ | |
3256 | if (uc < 0x100) | |
3257 | ANYOF_BITMAP_SET(data->start_class, uc); | |
3258 | else | |
3259 | data->start_class->flags |= ANYOF_UNICODE_ALL; | |
3260 | data->start_class->flags &= ~ANYOF_EOS; | |
3261 | cl_and(data->start_class, and_withp); | |
3262 | } | |
3263 | flags &= ~SCF_DO_STCLASS; | |
3264 | } | |
3265 | else if (PL_regkind[OP(scan)] == EXACT) { /* But OP != EXACT! */ | |
3266 | I32 l = STR_LEN(scan); | |
3267 | UV uc = *((U8*)STRING(scan)); | |
3268 | ||
3269 | /* Search for fixed substrings supports EXACT only. */ | |
3270 | if (flags & SCF_DO_SUBSTR) { | |
3271 | assert(data); | |
304ee84b | 3272 | SCAN_COMMIT(pRExC_state, data, minlenp); |
8aa23a47 YO |
3273 | } |
3274 | if (UTF) { | |
3275 | const U8 * const s = (U8 *)STRING(scan); | |
3276 | l = utf8_length(s, s + l); | |
3277 | uc = utf8_to_uvchr(s, NULL); | |
3278 | } | |
3279 | min += l; | |
3280 | if (flags & SCF_DO_SUBSTR) | |
3281 | data->pos_min += l; | |
3282 | if (flags & SCF_DO_STCLASS_AND) { | |
3283 | /* Check whether it is compatible with what we know already! */ | |
3284 | int compat = 1; | |
8aa23a47 | 3285 | if (uc >= 0x100 || |
54251c2e KW |
3286 | (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) |
3287 | && !ANYOF_BITMAP_TEST(data->start_class, uc) | |
3288 | && !ANYOF_BITMAP_TEST(data->start_class, PL_fold_latin1[uc]))) | |
3289 | { | |
8aa23a47 | 3290 | compat = 0; |
54251c2e | 3291 | } |
8aa23a47 YO |
3292 | ANYOF_CLASS_ZERO(data->start_class); |
3293 | ANYOF_BITMAP_ZERO(data->start_class); | |
3294 | if (compat) { | |
3295 | ANYOF_BITMAP_SET(data->start_class, uc); | |
653099ff | 3296 | data->start_class->flags &= ~ANYOF_EOS; |
39065660 | 3297 | data->start_class->flags |= ANYOF_LOC_NONBITMAP_FOLD; |
970c8436 | 3298 | if (OP(scan) == EXACTFL) { |
af302e7f KW |
3299 | /* XXX This set is probably no longer necessary, and |
3300 | * probably wrong as LOCALE now is on in the initial | |
3301 | * state */ | |
8aa23a47 | 3302 | data->start_class->flags |= ANYOF_LOCALE; |
970c8436 KW |
3303 | } |
3304 | else { | |
3305 | ||
54251c2e KW |
3306 | /* Also set the other member of the fold pair. In case |
3307 | * that unicode semantics is called for at runtime, use | |
3308 | * the full latin1 fold. (Can't do this for locale, | |
3309 | * because not known until runtime */ | |
3310 | ANYOF_BITMAP_SET(data->start_class, PL_fold_latin1[uc]); | |
970c8436 | 3311 | } |
653099ff | 3312 | } |
d18bf9dc KW |
3313 | else if (uc >= 0x100) { |
3314 | int i; | |
3315 | for (i = 0; i < 256; i++){ | |
3316 | if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(i)) { | |
3317 | ANYOF_BITMAP_SET(data->start_class, i); | |
3318 | } | |
3319 | } | |
3320 | } | |
8aa23a47 YO |
3321 | } |
3322 | else if (flags & SCF_DO_STCLASS_OR) { | |
39065660 | 3323 | if (data->start_class->flags & ANYOF_LOC_NONBITMAP_FOLD) { |
8aa23a47 YO |
3324 | /* false positive possible if the class is case-folded. |
3325 | Assume that the locale settings are the same... */ | |
970c8436 | 3326 | if (uc < 0x100) { |
1aa99e6b | 3327 | ANYOF_BITMAP_SET(data->start_class, uc); |
970c8436 KW |
3328 | if (OP(scan) != EXACTFL) { |
3329 | ||
3330 | /* And set the other member of the fold pair, but | |
3331 | * can't do that in locale because not known until | |
3332 | * run-time */ | |
3333 | ANYOF_BITMAP_SET(data->start_class, | |
54251c2e | 3334 | PL_fold_latin1[uc]); |
970c8436 KW |
3335 | } |
3336 | } | |
653099ff GS |
3337 | data->start_class->flags &= ~ANYOF_EOS; |
3338 | } | |
8aa23a47 | 3339 | cl_and(data->start_class, and_withp); |
653099ff | 3340 | } |
8aa23a47 YO |
3341 | flags &= ~SCF_DO_STCLASS; |
3342 | } | |
e52fc539 | 3343 | else if (REGNODE_VARIES(OP(scan))) { |
8aa23a47 YO |
3344 | I32 mincount, maxcount, minnext, deltanext, fl = 0; |
3345 | I32 f = flags, pos_before = 0; | |
3346 | regnode * const oscan = scan; | |
3347 | struct regnode_charclass_class this_class; | |
3348 | struct regnode_charclass_class *oclass = NULL; | |
3349 | I32 next_is_eval = 0; | |
3350 | ||
3351 | switch (PL_regkind[OP(scan)]) { | |
3352 | case WHILEM: /* End of (?:...)* . */ | |
3353 | scan = NEXTOPER(scan); | |
3354 | goto finish; | |
3355 | case PLUS: | |
3356 | if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) { | |
3357 | next = NEXTOPER(scan); | |
3358 | if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) { | |
3359 | mincount = 1; | |
3360 | maxcount = REG_INFTY; | |
3361 | next = regnext(scan); | |
3362 | scan = NEXTOPER(scan); | |
3363 | goto do_curly; | |
3364 | } | |
3365 | } | |
3366 | if (flags & SCF_DO_SUBSTR) | |
3367 | data->pos_min++; | |
3368 | min++; | |
3369 | /* Fall through. */ | |
3370 | case STAR: | |
3371 | if (flags & SCF_DO_STCLASS) { | |
3372 | mincount = 0; | |
3373 | maxcount = REG_INFTY; | |
3374 | next = regnext(scan); | |
3375 | scan = NEXTOPER(scan); | |
3376 | goto do_curly; | |
3377 | } | |
3378 | is_inf = is_inf_internal = 1; | |
3379 | scan = regnext(scan); | |
c277df42 | 3380 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 3381 | SCAN_COMMIT(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */ |
8aa23a47 | 3382 | data->longest = &(data->longest_float); |
c277df42 | 3383 | } |
8aa23a47 YO |
3384 | goto optimize_curly_tail; |
3385 | case CURLY: | |
3386 | if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM) | |
3387 | && (scan->flags == stopparen)) | |
3388 | { | |
3389 | mincount = 1; | |
3390 | maxcount = 1; | |
3391 | } else { | |
3392 | mincount = ARG1(scan); | |
3393 | maxcount = ARG2(scan); | |
653099ff | 3394 | } |
8aa23a47 YO |
3395 | next = regnext(scan); |
3396 | if (OP(scan) == CURLYX) { | |
3397 | I32 lp = (data ? *(data->last_closep) : 0); | |
3398 | scan->flags = ((lp <= (I32)U8_MAX) ? (U8)lp : U8_MAX); | |
653099ff | 3399 | } |
8aa23a47 YO |
3400 | scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; |
3401 | next_is_eval = (OP(scan) == EVAL); | |
3402 | do_curly: | |
3403 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 3404 | if (mincount == 0) SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */ |
8aa23a47 | 3405 | pos_before = data->pos_min; |
b45f050a | 3406 | } |
8aa23a47 YO |
3407 | if (data) { |
3408 | fl = data->flags; | |
3409 | data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL); | |
3410 | if (is_inf) | |
3411 | data->flags |= SF_IS_INF; | |
3412 | } | |
3413 | if (flags & SCF_DO_STCLASS) { | |
e755fd73 | 3414 | cl_init(pRExC_state, &this_class); |
8aa23a47 YO |
3415 | oclass = data->start_class; |
3416 | data->start_class = &this_class; | |
3417 | f |= SCF_DO_STCLASS_AND; | |
3418 | f &= ~SCF_DO_STCLASS_OR; | |
3419 | } | |
779bcb7d NC |
3420 | /* Exclude from super-linear cache processing any {n,m} |
3421 | regops for which the combination of input pos and regex | |
3422 | pos is not enough information to determine if a match | |
3423 | will be possible. | |
3424 | ||
3425 | For example, in the regex /foo(bar\s*){4,8}baz/ with the | |
3426 | regex pos at the \s*, the prospects for a match depend not | |
3427 | only on the input position but also on how many (bar\s*) | |
3428 | repeats into the {4,8} we are. */ | |
3429 | if ((mincount > 1) || (maxcount > 1 && maxcount != REG_INFTY)) | |
8aa23a47 | 3430 | f &= ~SCF_WHILEM_VISITED_POS; |
b45f050a | 3431 | |
8aa23a47 YO |
3432 | /* This will finish on WHILEM, setting scan, or on NULL: */ |
3433 | minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, | |
3434 | last, data, stopparen, recursed, NULL, | |
3435 | (mincount == 0 | |
3436 | ? (f & ~SCF_DO_SUBSTR) : f),depth+1); | |
b515a41d | 3437 | |
8aa23a47 YO |
3438 | if (flags & SCF_DO_STCLASS) |
3439 | data->start_class = oclass; | |
3440 | if (mincount == 0 || minnext == 0) { | |
3441 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 3442 | cl_or(pRExC_state, data->start_class, &this_class); |
8aa23a47 YO |
3443 | } |
3444 | else if (flags & SCF_DO_STCLASS_AND) { | |
3445 | /* Switch to OR mode: cache the old value of | |
3446 | * data->start_class */ | |
3447 | INIT_AND_WITHP; | |
3448 | StructCopy(data->start_class, and_withp, | |
3449 | struct regnode_charclass_class); | |
3450 | flags &= ~SCF_DO_STCLASS_AND; | |
3451 | StructCopy(&this_class, data->start_class, | |
3452 | struct regnode_charclass_class); | |
3453 | flags |= SCF_DO_STCLASS_OR; | |
3454 | data->start_class->flags |= ANYOF_EOS; | |
3455 | } | |
3456 | } else { /* Non-zero len */ | |
3457 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 3458 | cl_or(pRExC_state, data->start_class, &this_class); |
8aa23a47 YO |
3459 | cl_and(data->start_class, and_withp); |
3460 | } | |
3461 | else if (flags & SCF_DO_STCLASS_AND) | |
3462 | cl_and(data->start_class, &this_class); | |
3463 | flags &= ~SCF_DO_STCLASS; | |
3464 | } | |
3465 | if (!scan) /* It was not CURLYX, but CURLY. */ | |
3466 | scan = next; | |
3467 | if ( /* ? quantifier ok, except for (?{ ... }) */ | |
3468 | (next_is_eval || !(mincount == 0 && maxcount == 1)) | |
3469 | && (minnext == 0) && (deltanext == 0) | |
3470 | && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR)) | |
668c081a | 3471 | && maxcount <= REG_INFTY/3) /* Complement check for big count */ |
8aa23a47 | 3472 | { |
668c081a NC |
3473 | ckWARNreg(RExC_parse, |
3474 | "Quantifier unexpected on zero-length expression"); | |
8aa23a47 YO |
3475 | } |
3476 | ||
3477 | min += minnext * mincount; | |
3478 | is_inf_internal |= ((maxcount == REG_INFTY | |
3479 | && (minnext + deltanext) > 0) | |
3480 | || deltanext == I32_MAX); | |
3481 | is_inf |= is_inf_internal; | |
3482 | delta += (minnext + deltanext) * maxcount - minnext * mincount; | |
3483 | ||
3484 | /* Try powerful optimization CURLYX => CURLYN. */ | |
3485 | if ( OP(oscan) == CURLYX && data | |
3486 | && data->flags & SF_IN_PAR | |
3487 | && !(data->flags & SF_HAS_EVAL) | |
3488 | && !deltanext && minnext == 1 ) { | |
3489 | /* Try to optimize to CURLYN. */ | |
3490 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; | |
3491 | regnode * const nxt1 = nxt; | |
497b47a8 | 3492 | #ifdef DEBUGGING |
8aa23a47 | 3493 | regnode *nxt2; |
497b47a8 | 3494 | #endif |
c277df42 | 3495 | |
8aa23a47 YO |
3496 | /* Skip open. */ |
3497 | nxt = regnext(nxt); | |
e52fc539 | 3498 | if (!REGNODE_SIMPLE(OP(nxt)) |
8aa23a47 YO |
3499 | && !(PL_regkind[OP(nxt)] == EXACT |
3500 | && STR_LEN(nxt) == 1)) | |
3501 | goto nogo; | |
497b47a8 | 3502 | #ifdef DEBUGGING |
8aa23a47 | 3503 | nxt2 = nxt; |
497b47a8 | 3504 | #endif |
8aa23a47 YO |
3505 | nxt = regnext(nxt); |
3506 | if (OP(nxt) != CLOSE) | |
3507 | goto nogo; | |
3508 | if (RExC_open_parens) { | |
3509 | RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/ | |
3510 | RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/ | |
3511 | } | |
3512 | /* Now we know that nxt2 is the only contents: */ | |
3513 | oscan->flags = (U8)ARG(nxt); | |
3514 | OP(oscan) = CURLYN; | |
3515 | OP(nxt1) = NOTHING; /* was OPEN. */ | |
40d049e4 | 3516 | |
c277df42 | 3517 | #ifdef DEBUGGING |
8aa23a47 | 3518 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
fda99bee KW |
3519 | NEXT_OFF(nxt1+ 1) = 0; /* just for consistency. */ |
3520 | NEXT_OFF(nxt2) = 0; /* just for consistency with CURLY. */ | |
8aa23a47 YO |
3521 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ |
3522 | OP(nxt + 1) = OPTIMIZED; /* was count. */ | |
fda99bee | 3523 | NEXT_OFF(nxt+ 1) = 0; /* just for consistency. */ |
b81d288d | 3524 | #endif |
8aa23a47 YO |
3525 | } |
3526 | nogo: | |
3527 | ||
3528 | /* Try optimization CURLYX => CURLYM. */ | |
3529 | if ( OP(oscan) == CURLYX && data | |
3530 | && !(data->flags & SF_HAS_PAR) | |
3531 | && !(data->flags & SF_HAS_EVAL) | |
3532 | && !deltanext /* atom is fixed width */ | |
3533 | && minnext != 0 /* CURLYM can't handle zero width */ | |
3534 | ) { | |
3535 | /* XXXX How to optimize if data == 0? */ | |
3536 | /* Optimize to a simpler form. */ | |
3537 | regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */ | |
3538 | regnode *nxt2; | |
3539 | ||
3540 | OP(oscan) = CURLYM; | |
3541 | while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/ | |
3542 | && (OP(nxt2) != WHILEM)) | |
3543 | nxt = nxt2; | |
3544 | OP(nxt2) = SUCCEED; /* Whas WHILEM */ | |
3545 | /* Need to optimize away parenths. */ | |
b3c0965f | 3546 | if ((data->flags & SF_IN_PAR) && OP(nxt) == CLOSE) { |
8aa23a47 YO |
3547 | /* Set the parenth number. */ |
3548 | regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/ | |
3549 | ||
8aa23a47 YO |
3550 | oscan->flags = (U8)ARG(nxt); |
3551 | if (RExC_open_parens) { | |
3552 | RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/ | |
3553 | RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/ | |
40d049e4 | 3554 | } |
8aa23a47 YO |
3555 | OP(nxt1) = OPTIMIZED; /* was OPEN. */ |
3556 | OP(nxt) = OPTIMIZED; /* was CLOSE. */ | |
40d049e4 | 3557 | |
c277df42 | 3558 | #ifdef DEBUGGING |
8aa23a47 YO |
3559 | OP(nxt1 + 1) = OPTIMIZED; /* was count. */ |
3560 | OP(nxt + 1) = OPTIMIZED; /* was count. */ | |
486ec47a PA |
3561 | NEXT_OFF(nxt1 + 1) = 0; /* just for consistency. */ |
3562 | NEXT_OFF(nxt + 1) = 0; /* just for consistency. */ | |
b81d288d | 3563 | #endif |
c277df42 | 3564 | #if 0 |
8aa23a47 YO |
3565 | while ( nxt1 && (OP(nxt1) != WHILEM)) { |
3566 | regnode *nnxt = regnext(nxt1); | |
8aa23a47 YO |
3567 | if (nnxt == nxt) { |
3568 | if (reg_off_by_arg[OP(nxt1)]) | |
3569 | ARG_SET(nxt1, nxt2 - nxt1); | |
3570 | else if (nxt2 - nxt1 < U16_MAX) | |
3571 | NEXT_OFF(nxt1) = nxt2 - nxt1; | |
3572 | else | |
3573 | OP(nxt) = NOTHING; /* Cannot beautify */ | |
c277df42 | 3574 | } |
8aa23a47 | 3575 | nxt1 = nnxt; |
c277df42 | 3576 | } |
5d1c421c | 3577 | #endif |
8aa23a47 YO |
3578 | /* Optimize again: */ |
3579 | study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt, | |
3580 | NULL, stopparen, recursed, NULL, 0,depth+1); | |
3581 | } | |
3582 | else | |
3583 | oscan->flags = 0; | |
3584 | } | |
3585 | else if ((OP(oscan) == CURLYX) | |
3586 | && (flags & SCF_WHILEM_VISITED_POS) | |
3587 | /* See the comment on a similar expression above. | |
3b753521 | 3588 | However, this time it's not a subexpression |
8aa23a47 YO |
3589 | we care about, but the expression itself. */ |
3590 | && (maxcount == REG_INFTY) | |
3591 | && data && ++data->whilem_c < 16) { | |
3592 | /* This stays as CURLYX, we can put the count/of pair. */ | |
3593 | /* Find WHILEM (as in regexec.c) */ | |
3594 | regnode *nxt = oscan + NEXT_OFF(oscan); | |
3595 | ||
3596 | if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */ | |
3597 | nxt += ARG(nxt); | |
3598 | PREVOPER(nxt)->flags = (U8)(data->whilem_c | |
3599 | | (RExC_whilem_seen << 4)); /* On WHILEM */ | |
3600 | } | |
3601 | if (data && fl & (SF_HAS_PAR|SF_IN_PAR)) | |
3602 | pars++; | |
3603 | if (flags & SCF_DO_SUBSTR) { | |
3604 | SV *last_str = NULL; | |
3605 | int counted = mincount != 0; | |
a0ed51b3 | 3606 | |
8aa23a47 YO |
3607 | if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */ |
3608 | #if defined(SPARC64_GCC_WORKAROUND) | |
3609 | I32 b = 0; | |
3610 | STRLEN l = 0; | |
3611 | const char *s = NULL; | |
3612 | I32 old = 0; | |
b515a41d | 3613 | |
8aa23a47 YO |
3614 | if (pos_before >= data->last_start_min) |
3615 | b = pos_before; | |
3616 | else | |
3617 | b = data->last_start_min; | |
b515a41d | 3618 | |
8aa23a47 YO |
3619 | l = 0; |
3620 | s = SvPV_const(data->last_found, l); | |
3621 | old = b - data->last_start_min; | |
3622 | ||
3623 | #else | |
3624 | I32 b = pos_before >= data->last_start_min | |
3625 | ? pos_before : data->last_start_min; | |
3626 | STRLEN l; | |
3627 | const char * const s = SvPV_const(data->last_found, l); | |
3628 | I32 old = b - data->last_start_min; | |
3629 | #endif | |
3630 | ||
3631 | if (UTF) | |
3632 | old = utf8_hop((U8*)s, old) - (U8*)s; | |
8aa23a47 YO |
3633 | l -= old; |
3634 | /* Get the added string: */ | |
740cce10 | 3635 | last_str = newSVpvn_utf8(s + old, l, UTF); |
8aa23a47 YO |
3636 | if (deltanext == 0 && pos_before == b) { |
3637 | /* What was added is a constant string */ | |
3638 | if (mincount > 1) { | |
3639 | SvGROW(last_str, (mincount * l) + 1); | |
3640 | repeatcpy(SvPVX(last_str) + l, | |
3641 | SvPVX_const(last_str), l, mincount - 1); | |
3642 | SvCUR_set(last_str, SvCUR(last_str) * mincount); | |
3643 | /* Add additional parts. */ | |
3644 | SvCUR_set(data->last_found, | |
3645 | SvCUR(data->last_found) - l); | |
3646 | sv_catsv(data->last_found, last_str); | |
3647 | { | |
3648 | SV * sv = data->last_found; | |
3649 | MAGIC *mg = | |
3650 | SvUTF8(sv) && SvMAGICAL(sv) ? | |
3651 | mg_find(sv, PERL_MAGIC_utf8) : NULL; | |
3652 | if (mg && mg->mg_len >= 0) | |
bd94e887 | 3653 | mg->mg_len += CHR_SVLEN(last_str) - l; |
b515a41d | 3654 | } |
8aa23a47 | 3655 | data->last_end += l * (mincount - 1); |
b515a41d | 3656 | } |
8aa23a47 YO |
3657 | } else { |
3658 | /* start offset must point into the last copy */ | |
3659 | data->last_start_min += minnext * (mincount - 1); | |
3660 | data->last_start_max += is_inf ? I32_MAX | |
3661 | : (maxcount - 1) * (minnext + data->pos_delta); | |
3662 | } | |
c277df42 | 3663 | } |
8aa23a47 YO |
3664 | /* It is counted once already... */ |
3665 | data->pos_min += minnext * (mincount - counted); | |
3666 | data->pos_delta += - counted * deltanext + | |
3667 | (minnext + deltanext) * maxcount - minnext * mincount; | |
3668 | if (mincount != maxcount) { | |
3669 | /* Cannot extend fixed substrings found inside | |
3670 | the group. */ | |
304ee84b | 3671 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 YO |
3672 | if (mincount && last_str) { |
3673 | SV * const sv = data->last_found; | |
3674 | MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? | |
3675 | mg_find(sv, PERL_MAGIC_utf8) : NULL; | |
3676 | ||
3677 | if (mg) | |
3678 | mg->mg_len = -1; | |
3679 | sv_setsv(sv, last_str); | |
3680 | data->last_end = data->pos_min; | |
3681 | data->last_start_min = | |
3682 | data->pos_min - CHR_SVLEN(last_str); | |
3683 | data->last_start_max = is_inf | |
3684 | ? I32_MAX | |
3685 | : data->pos_min + data->pos_delta | |
3686 | - CHR_SVLEN(last_str); | |
3687 | } | |
3688 | data->longest = &(data->longest_float); | |
3689 | } | |
3690 | SvREFCNT_dec(last_str); | |
c277df42 | 3691 | } |
8aa23a47 YO |
3692 | if (data && (fl & SF_HAS_EVAL)) |
3693 | data->flags |= SF_HAS_EVAL; | |
3694 | optimize_curly_tail: | |
3695 | if (OP(oscan) != CURLYX) { | |
3696 | while (PL_regkind[OP(next = regnext(oscan))] == NOTHING | |
3697 | && NEXT_OFF(next)) | |
3698 | NEXT_OFF(oscan) += NEXT_OFF(next); | |
3699 | } | |
3700 | continue; | |
f56b6394 | 3701 | default: /* REF, ANYOFV, and CLUMP only? */ |
8aa23a47 | 3702 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 3703 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
8aa23a47 YO |
3704 | data->longest = &(data->longest_float); |
3705 | } | |
3706 | is_inf = is_inf_internal = 1; | |
3707 | if (flags & SCF_DO_STCLASS_OR) | |
3fffb88a | 3708 | cl_anything(pRExC_state, data->start_class); |
8aa23a47 YO |
3709 | flags &= ~SCF_DO_STCLASS; |
3710 | break; | |
c277df42 | 3711 | } |
8aa23a47 | 3712 | } |
e1d1eefb YO |
3713 | else if (OP(scan) == LNBREAK) { |
3714 | if (flags & SCF_DO_STCLASS) { | |
3715 | int value = 0; | |
3716 | data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */ | |
3717 | if (flags & SCF_DO_STCLASS_AND) { | |
3718 | for (value = 0; value < 256; value++) | |
e64b1bd1 | 3719 | if (!is_VERTWS_cp(value)) |
b9a59e08 KW |
3720 | ANYOF_BITMAP_CLEAR(data->start_class, value); |
3721 | } | |
3722 | else { | |
e1d1eefb | 3723 | for (value = 0; value < 256; value++) |
e64b1bd1 | 3724 | if (is_VERTWS_cp(value)) |
b9a59e08 KW |
3725 | ANYOF_BITMAP_SET(data->start_class, value); |
3726 | } | |
e1d1eefb YO |
3727 | if (flags & SCF_DO_STCLASS_OR) |
3728 | cl_and(data->start_class, and_withp); | |
3729 | flags &= ~SCF_DO_STCLASS; | |
3730 | } | |
3731 | min += 1; | |
f9a79580 | 3732 | delta += 1; |
e1d1eefb YO |
3733 | if (flags & SCF_DO_SUBSTR) { |
3734 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ | |
3735 | data->pos_min += 1; | |
f9a79580 | 3736 | data->pos_delta += 1; |
e1d1eefb YO |
3737 | data->longest = &(data->longest_float); |
3738 | } | |
e1d1eefb | 3739 | } |
f9a79580 | 3740 | else if (OP(scan) == FOLDCHAR) { |
ced7f090 | 3741 | int d = ARG(scan) == LATIN_SMALL_LETTER_SHARP_S ? 1 : 2; |
f9a79580 RGS |
3742 | flags &= ~SCF_DO_STCLASS; |
3743 | min += 1; | |
3744 | delta += d; | |
3745 | if (flags & SCF_DO_SUBSTR) { | |
3746 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ | |
3747 | data->pos_min += 1; | |
3748 | data->pos_delta += d; | |
3749 | data->longest = &(data->longest_float); | |
3750 | } | |
3751 | } | |
e52fc539 | 3752 | else if (REGNODE_SIMPLE(OP(scan))) { |
8aa23a47 | 3753 | int value = 0; |
653099ff | 3754 | |
8aa23a47 | 3755 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 3756 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 YO |
3757 | data->pos_min++; |
3758 | } | |
3759 | min++; | |
3760 | if (flags & SCF_DO_STCLASS) { | |
3761 | data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */ | |
b515a41d | 3762 | |
8aa23a47 YO |
3763 | /* Some of the logic below assumes that switching |
3764 | locale on will only add false positives. */ | |
3765 | switch (PL_regkind[OP(scan)]) { | |
3766 | case SANY: | |
3767 | default: | |
3768 | do_default: | |
3769 | /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */ | |
3770 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ | |
3fffb88a | 3771 | cl_anything(pRExC_state, data->start_class); |
8aa23a47 YO |
3772 | break; |
3773 | case REG_ANY: | |
3774 | if (OP(scan) == SANY) | |
3775 | goto do_default; | |
3776 | if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */ | |
3777 | value = (ANYOF_BITMAP_TEST(data->start_class,'\n') | |
3a15e693 | 3778 | || ANYOF_CLASS_TEST_ANY_SET(data->start_class)); |
3fffb88a | 3779 | cl_anything(pRExC_state, data->start_class); |
653099ff | 3780 | } |
8aa23a47 YO |
3781 | if (flags & SCF_DO_STCLASS_AND || !value) |
3782 | ANYOF_BITMAP_CLEAR(data->start_class,'\n'); | |
3783 | break; | |
3784 | case ANYOF: | |
3785 | if (flags & SCF_DO_STCLASS_AND) | |
3786 | cl_and(data->start_class, | |
3787 | (struct regnode_charclass_class*)scan); | |
653099ff | 3788 | else |
3fffb88a | 3789 | cl_or(pRExC_state, data->start_class, |
8aa23a47 YO |
3790 | (struct regnode_charclass_class*)scan); |
3791 | break; | |
3792 | case ALNUM: | |
3793 | if (flags & SCF_DO_STCLASS_AND) { | |
3794 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
3795 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM); | |
980866de | 3796 | if (OP(scan) == ALNUMU) { |
a12cf05f KW |
3797 | for (value = 0; value < 256; value++) { |
3798 | if (!isWORDCHAR_L1(value)) { | |
3799 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3800 | } | |
3801 | } | |
3802 | } else { | |
3803 | for (value = 0; value < 256; value++) { | |
3804 | if (!isALNUM(value)) { | |
3805 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3806 | } | |
3807 | } | |
3808 | } | |
8aa23a47 | 3809 | } |
653099ff | 3810 | } |
8aa23a47 YO |
3811 | else { |
3812 | if (data->start_class->flags & ANYOF_LOCALE) | |
3813 | ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM); | |
af302e7f KW |
3814 | |
3815 | /* Even if under locale, set the bits for non-locale | |
3816 | * in case it isn't a true locale-node. This will | |
3817 | * create false positives if it truly is locale */ | |
3818 | if (OP(scan) == ALNUMU) { | |
a12cf05f KW |
3819 | for (value = 0; value < 256; value++) { |
3820 | if (isWORDCHAR_L1(value)) { | |
3821 | ANYOF_BITMAP_SET(data->start_class, value); | |
3822 | } | |
3823 | } | |
3824 | } else { | |
3825 | for (value = 0; value < 256; value++) { | |
3826 | if (isALNUM(value)) { | |
3827 | ANYOF_BITMAP_SET(data->start_class, value); | |
3828 | } | |
3829 | } | |
3830 | } | |
8aa23a47 YO |
3831 | } |
3832 | break; | |
8aa23a47 YO |
3833 | case NALNUM: |
3834 | if (flags & SCF_DO_STCLASS_AND) { | |
3835 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
3836 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM); | |
980866de | 3837 | if (OP(scan) == NALNUMU) { |
a12cf05f KW |
3838 | for (value = 0; value < 256; value++) { |
3839 | if (isWORDCHAR_L1(value)) { | |
3840 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3841 | } | |
3842 | } | |
3843 | } else { | |
3844 | for (value = 0; value < 256; value++) { | |
3845 | if (isALNUM(value)) { | |
3846 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3847 | } | |
3848 | } | |
3849 | } | |
653099ff GS |
3850 | } |
3851 | } | |
8aa23a47 YO |
3852 | else { |
3853 | if (data->start_class->flags & ANYOF_LOCALE) | |
3854 | ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM); | |
af302e7f | 3855 | |
75950e1c KW |
3856 | /* Even if under locale, set the bits for non-locale in |
3857 | * case it isn't a true locale-node. This will create | |
3858 | * false positives if it truly is locale */ | |
3859 | if (OP(scan) == NALNUMU) { | |
3860 | for (value = 0; value < 256; value++) { | |
3861 | if (! isWORDCHAR_L1(value)) { | |
3862 | ANYOF_BITMAP_SET(data->start_class, value); | |
3863 | } | |
e9a9c1bc | 3864 | } |
75950e1c KW |
3865 | } else { |
3866 | for (value = 0; value < 256; value++) { | |
3867 | if (! isALNUM(value)) { | |
3868 | ANYOF_BITMAP_SET(data->start_class, value); | |
3869 | } | |
3870 | } | |
3871 | } | |
653099ff | 3872 | } |
8aa23a47 | 3873 | break; |
8aa23a47 YO |
3874 | case SPACE: |
3875 | if (flags & SCF_DO_STCLASS_AND) { | |
3876 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
3877 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE); | |
980866de | 3878 | if (OP(scan) == SPACEU) { |
a12cf05f KW |
3879 | for (value = 0; value < 256; value++) { |
3880 | if (!isSPACE_L1(value)) { | |
3881 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3882 | } | |
3883 | } | |
3884 | } else { | |
3885 | for (value = 0; value < 256; value++) { | |
3886 | if (!isSPACE(value)) { | |
3887 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3888 | } | |
3889 | } | |
3890 | } | |
653099ff GS |
3891 | } |
3892 | } | |
8aa23a47 | 3893 | else { |
a12cf05f | 3894 | if (data->start_class->flags & ANYOF_LOCALE) { |
8aa23a47 | 3895 | ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE); |
a12cf05f | 3896 | } |
af302e7f | 3897 | if (OP(scan) == SPACEU) { |
a12cf05f KW |
3898 | for (value = 0; value < 256; value++) { |
3899 | if (isSPACE_L1(value)) { | |
3900 | ANYOF_BITMAP_SET(data->start_class, value); | |
3901 | } | |
3902 | } | |
3903 | } else { | |
3904 | for (value = 0; value < 256; value++) { | |
3905 | if (isSPACE(value)) { | |
3906 | ANYOF_BITMAP_SET(data->start_class, value); | |
3907 | } | |
3908 | } | |
8aa23a47 | 3909 | } |
653099ff | 3910 | } |
8aa23a47 | 3911 | break; |
8aa23a47 YO |
3912 | case NSPACE: |
3913 | if (flags & SCF_DO_STCLASS_AND) { | |
3914 | if (!(data->start_class->flags & ANYOF_LOCALE)) { | |
3915 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE); | |
980866de | 3916 | if (OP(scan) == NSPACEU) { |
a12cf05f KW |
3917 | for (value = 0; value < 256; value++) { |
3918 | if (isSPACE_L1(value)) { | |
3919 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3920 | } | |
3921 | } | |
3922 | } else { | |
3923 | for (value = 0; value < 256; value++) { | |
3924 | if (isSPACE(value)) { | |
3925 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3926 | } | |
3927 | } | |
3928 | } | |
653099ff | 3929 | } |
8aa23a47 YO |
3930 | } |
3931 | else { | |
3932 | if (data->start_class->flags & ANYOF_LOCALE) | |
3933 | ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE); | |
af302e7f | 3934 | if (OP(scan) == NSPACEU) { |
a12cf05f KW |
3935 | for (value = 0; value < 256; value++) { |
3936 | if (!isSPACE_L1(value)) { | |
3937 | ANYOF_BITMAP_SET(data->start_class, value); | |
3938 | } | |
3939 | } | |
3940 | } | |
3941 | else { | |
3942 | for (value = 0; value < 256; value++) { | |
3943 | if (!isSPACE(value)) { | |
3944 | ANYOF_BITMAP_SET(data->start_class, value); | |
3945 | } | |
3946 | } | |
3947 | } | |
653099ff | 3948 | } |
8aa23a47 | 3949 | break; |
8aa23a47 YO |
3950 | case DIGIT: |
3951 | if (flags & SCF_DO_STCLASS_AND) { | |
bcc0256f | 3952 | if (!(data->start_class->flags & ANYOF_LOCALE)) { |
bf3c5c06 KW |
3953 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT); |
3954 | for (value = 0; value < 256; value++) | |
3955 | if (!isDIGIT(value)) | |
3956 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
bcc0256f | 3957 | } |
8aa23a47 YO |
3958 | } |
3959 | else { | |
3960 | if (data->start_class->flags & ANYOF_LOCALE) | |
3961 | ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT); | |
75950e1c KW |
3962 | for (value = 0; value < 256; value++) |
3963 | if (isDIGIT(value)) | |
3964 | ANYOF_BITMAP_SET(data->start_class, value); | |
8aa23a47 YO |
3965 | } |
3966 | break; | |
3967 | case NDIGIT: | |
3968 | if (flags & SCF_DO_STCLASS_AND) { | |
bcc0256f | 3969 | if (!(data->start_class->flags & ANYOF_LOCALE)) |
bf3c5c06 | 3970 | ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT); |
8aa23a47 YO |
3971 | for (value = 0; value < 256; value++) |
3972 | if (isDIGIT(value)) | |
3973 | ANYOF_BITMAP_CLEAR(data->start_class, value); | |
3974 | } | |
3975 | else { | |
3976 | if (data->start_class->flags & ANYOF_LOCALE) | |
3977 | ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT); | |
75950e1c KW |
3978 | for (value = 0; value < 256; value++) |
3979 | if (!isDIGIT(value)) | |
3980 | ANYOF_BITMAP_SET(data->start_class, value); | |
653099ff | 3981 | } |
8aa23a47 | 3982 | break; |
e1d1eefb YO |
3983 | CASE_SYNST_FNC(VERTWS); |
3984 | CASE_SYNST_FNC(HORIZWS); | |
686b73d4 | 3985 | |
8aa23a47 YO |
3986 | } |
3987 | if (flags & SCF_DO_STCLASS_OR) | |
3988 | cl_and(data->start_class, and_withp); | |
3989 | flags &= ~SCF_DO_STCLASS; | |
3990 | } | |
3991 | } | |
3992 | else if (PL_regkind[OP(scan)] == EOL && flags & SCF_DO_SUBSTR) { | |
3993 | data->flags |= (OP(scan) == MEOL | |
3994 | ? SF_BEFORE_MEOL | |
3995 | : SF_BEFORE_SEOL); | |
3996 | } | |
3997 | else if ( PL_regkind[OP(scan)] == BRANCHJ | |
3998 | /* Lookbehind, or need to calculate parens/evals/stclass: */ | |
3999 | && (scan->flags || data || (flags & SCF_DO_STCLASS)) | |
4000 | && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) { | |
4001 | if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY | |
4002 | || OP(scan) == UNLESSM ) | |
4003 | { | |
4004 | /* Negative Lookahead/lookbehind | |
4005 | In this case we can't do fixed string optimisation. | |
4006 | */ | |
1de06328 | 4007 | |
8aa23a47 YO |
4008 | I32 deltanext, minnext, fake = 0; |
4009 | regnode *nscan; | |
4010 | struct regnode_charclass_class intrnl; | |
4011 | int f = 0; | |
1de06328 | 4012 | |
8aa23a47 YO |
4013 | data_fake.flags = 0; |
4014 | if (data) { | |
4015 | data_fake.whilem_c = data->whilem_c; | |
4016 | data_fake.last_closep = data->last_closep; | |
c277df42 | 4017 | } |
8aa23a47 YO |
4018 | else |
4019 | data_fake.last_closep = &fake; | |
58e23c8d | 4020 | data_fake.pos_delta = delta; |
8aa23a47 YO |
4021 | if ( flags & SCF_DO_STCLASS && !scan->flags |
4022 | && OP(scan) == IFMATCH ) { /* Lookahead */ | |
e755fd73 | 4023 | cl_init(pRExC_state, &intrnl); |
8aa23a47 YO |
4024 | data_fake.start_class = &intrnl; |
4025 | f |= SCF_DO_STCLASS_AND; | |
4026 | } | |
4027 | if (flags & SCF_WHILEM_VISITED_POS) | |
4028 | f |= SCF_WHILEM_VISITED_POS; | |
4029 | next = regnext(scan); | |
4030 | nscan = NEXTOPER(NEXTOPER(scan)); | |
4031 | minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, | |
4032 | last, &data_fake, stopparen, recursed, NULL, f, depth+1); | |
4033 | if (scan->flags) { | |
4034 | if (deltanext) { | |
58e23c8d | 4035 | FAIL("Variable length lookbehind not implemented"); |
8aa23a47 YO |
4036 | } |
4037 | else if (minnext > (I32)U8_MAX) { | |
58e23c8d | 4038 | FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX); |
8aa23a47 YO |
4039 | } |
4040 | scan->flags = (U8)minnext; | |
4041 | } | |
4042 | if (data) { | |
4043 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
4044 | pars++; | |
4045 | if (data_fake.flags & SF_HAS_EVAL) | |
4046 | data->flags |= SF_HAS_EVAL; | |
4047 | data->whilem_c = data_fake.whilem_c; | |
4048 | } | |
4049 | if (f & SCF_DO_STCLASS_AND) { | |
906cdd2b HS |
4050 | if (flags & SCF_DO_STCLASS_OR) { |
4051 | /* OR before, AND after: ideally we would recurse with | |
4052 | * data_fake to get the AND applied by study of the | |
4053 | * remainder of the pattern, and then derecurse; | |
4054 | * *** HACK *** for now just treat as "no information". | |
4055 | * See [perl #56690]. | |
4056 | */ | |
e755fd73 | 4057 | cl_init(pRExC_state, data->start_class); |
906cdd2b HS |
4058 | } else { |
4059 | /* AND before and after: combine and continue */ | |
4060 | const int was = (data->start_class->flags & ANYOF_EOS); | |
4061 | ||
4062 | cl_and(data->start_class, &intrnl); | |
4063 | if (was) | |
4064 | data->start_class->flags |= ANYOF_EOS; | |
4065 | } | |
8aa23a47 | 4066 | } |
cb434fcc | 4067 | } |
8aa23a47 YO |
4068 | #if PERL_ENABLE_POSITIVE_ASSERTION_STUDY |
4069 | else { | |
4070 | /* Positive Lookahead/lookbehind | |
4071 | In this case we can do fixed string optimisation, | |
4072 | but we must be careful about it. Note in the case of | |
4073 | lookbehind the positions will be offset by the minimum | |
4074 | length of the pattern, something we won't know about | |
4075 | until after the recurse. | |
4076 | */ | |
4077 | I32 deltanext, fake = 0; | |
4078 | regnode *nscan; | |
4079 | struct regnode_charclass_class intrnl; | |
4080 | int f = 0; | |
4081 | /* We use SAVEFREEPV so that when the full compile | |
4082 | is finished perl will clean up the allocated | |
3b753521 | 4083 | minlens when it's all done. This way we don't |
8aa23a47 YO |
4084 | have to worry about freeing them when we know |
4085 | they wont be used, which would be a pain. | |
4086 | */ | |
4087 | I32 *minnextp; | |
4088 | Newx( minnextp, 1, I32 ); | |
4089 | SAVEFREEPV(minnextp); | |
4090 | ||
4091 | if (data) { | |
4092 | StructCopy(data, &data_fake, scan_data_t); | |
4093 | if ((flags & SCF_DO_SUBSTR) && data->last_found) { | |
4094 | f |= SCF_DO_SUBSTR; | |
4095 | if (scan->flags) | |
304ee84b | 4096 | SCAN_COMMIT(pRExC_state, &data_fake,minlenp); |
8aa23a47 YO |
4097 | data_fake.last_found=newSVsv(data->last_found); |
4098 | } | |
4099 | } | |
4100 | else | |
4101 | data_fake.last_closep = &fake; | |
4102 | data_fake.flags = 0; | |
58e23c8d | 4103 | data_fake.pos_delta = delta; |
8aa23a47 YO |
4104 | if (is_inf) |
4105 | data_fake.flags |= SF_IS_INF; | |
4106 | if ( flags & SCF_DO_STCLASS && !scan->flags | |
4107 | && OP(scan) == IFMATCH ) { /* Lookahead */ | |
e755fd73 | 4108 | cl_init(pRExC_state, &intrnl); |
8aa23a47 YO |
4109 | data_fake.start_class = &intrnl; |
4110 | f |= SCF_DO_STCLASS_AND; | |
4111 | } | |
4112 | if (flags & SCF_WHILEM_VISITED_POS) | |
4113 | f |= SCF_WHILEM_VISITED_POS; | |
4114 | next = regnext(scan); | |
4115 | nscan = NEXTOPER(NEXTOPER(scan)); | |
4116 | ||
4117 | *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext, | |
4118 | last, &data_fake, stopparen, recursed, NULL, f,depth+1); | |
4119 | if (scan->flags) { | |
4120 | if (deltanext) { | |
58e23c8d | 4121 | FAIL("Variable length lookbehind not implemented"); |
8aa23a47 YO |
4122 | } |
4123 | else if (*minnextp > (I32)U8_MAX) { | |
58e23c8d | 4124 | FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX); |
8aa23a47 YO |
4125 | } |
4126 | scan->flags = (U8)*minnextp; | |
4127 | } | |
4128 | ||
4129 | *minnextp += min; | |
4130 | ||
4131 | if (f & SCF_DO_STCLASS_AND) { | |
4132 | const int was = (data->start_class->flags & ANYOF_EOS); | |
4133 | ||
4134 | cl_and(data->start_class, &intrnl); | |
4135 | if (was) | |
4136 | data->start_class->flags |= ANYOF_EOS; | |
4137 | } | |
4138 | if (data) { | |
4139 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
4140 | pars++; | |
4141 | if (data_fake.flags & SF_HAS_EVAL) | |
4142 | data->flags |= SF_HAS_EVAL; | |
4143 | data->whilem_c = data_fake.whilem_c; | |
4144 | if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) { | |
4145 | if (RExC_rx->minlen<*minnextp) | |
4146 | RExC_rx->minlen=*minnextp; | |
304ee84b | 4147 | SCAN_COMMIT(pRExC_state, &data_fake, minnextp); |
8aa23a47 YO |
4148 | SvREFCNT_dec(data_fake.last_found); |
4149 | ||
4150 | if ( data_fake.minlen_fixed != minlenp ) | |
4151 | { | |
4152 | data->offset_fixed= data_fake.offset_fixed; | |
4153 | data->minlen_fixed= data_fake.minlen_fixed; | |
4154 | data->lookbehind_fixed+= scan->flags; | |
4155 | } | |
4156 | if ( data_fake.minlen_float != minlenp ) | |
4157 | { | |
4158 | data->minlen_float= data_fake.minlen_float; | |
4159 | data->offset_float_min=data_fake.offset_float_min; | |
4160 | data->offset_float_max=data_fake.offset_float_max; | |
4161 | data->lookbehind_float+= scan->flags; | |
4162 | } | |
4163 | } | |
4164 | } | |
4165 | ||
4166 | ||
40d049e4 | 4167 | } |
8aa23a47 YO |
4168 | #endif |
4169 | } | |
4170 | else if (OP(scan) == OPEN) { | |
4171 | if (stopparen != (I32)ARG(scan)) | |
4172 | pars++; | |
4173 | } | |
4174 | else if (OP(scan) == CLOSE) { | |
4175 | if (stopparen == (I32)ARG(scan)) { | |
4176 | break; | |
4177 | } | |
4178 | if ((I32)ARG(scan) == is_par) { | |
4179 | next = regnext(scan); | |
b515a41d | 4180 | |
8aa23a47 YO |
4181 | if ( next && (OP(next) != WHILEM) && next < last) |
4182 | is_par = 0; /* Disable optimization */ | |
40d049e4 | 4183 | } |
8aa23a47 YO |
4184 | if (data) |
4185 | *(data->last_closep) = ARG(scan); | |
4186 | } | |
4187 | else if (OP(scan) == EVAL) { | |
c277df42 IZ |
4188 | if (data) |
4189 | data->flags |= SF_HAS_EVAL; | |
8aa23a47 YO |
4190 | } |
4191 | else if ( PL_regkind[OP(scan)] == ENDLIKE ) { | |
4192 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 4193 | SCAN_COMMIT(pRExC_state,data,minlenp); |
8aa23a47 | 4194 | flags &= ~SCF_DO_SUBSTR; |
40d049e4 | 4195 | } |
8aa23a47 YO |
4196 | if (data && OP(scan)==ACCEPT) { |
4197 | data->flags |= SCF_SEEN_ACCEPT; | |
4198 | if (stopmin > min) | |
4199 | stopmin = min; | |
e2e6a0f1 | 4200 | } |
8aa23a47 YO |
4201 | } |
4202 | else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */ | |
4203 | { | |
0f5d15d6 | 4204 | if (flags & SCF_DO_SUBSTR) { |
304ee84b | 4205 | SCAN_COMMIT(pRExC_state,data,minlenp); |
0f5d15d6 IZ |
4206 | data->longest = &(data->longest_float); |
4207 | } | |
4208 | is_inf = is_inf_internal = 1; | |
653099ff | 4209 | if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ |
3fffb88a | 4210 | cl_anything(pRExC_state, data->start_class); |
96776eda | 4211 | flags &= ~SCF_DO_STCLASS; |
8aa23a47 | 4212 | } |
58e23c8d | 4213 | else if (OP(scan) == GPOS) { |
bbe252da | 4214 | if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) && |
58e23c8d YO |
4215 | !(delta || is_inf || (data && data->pos_delta))) |
4216 | { | |
bbe252da YO |
4217 | if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR)) |
4218 | RExC_rx->extflags |= RXf_ANCH_GPOS; | |
58e23c8d YO |
4219 | if (RExC_rx->gofs < (U32)min) |
4220 | RExC_rx->gofs = min; | |
4221 | } else { | |
bbe252da | 4222 | RExC_rx->extflags |= RXf_GPOS_FLOAT; |
58e23c8d YO |
4223 | RExC_rx->gofs = 0; |
4224 | } | |
4225 | } | |
786e8c11 | 4226 | #ifdef TRIE_STUDY_OPT |
40d049e4 | 4227 | #ifdef FULL_TRIE_STUDY |
8aa23a47 YO |
4228 | else if (PL_regkind[OP(scan)] == TRIE) { |
4229 | /* NOTE - There is similar code to this block above for handling | |
4230 | BRANCH nodes on the initial study. If you change stuff here | |
4231 | check there too. */ | |
4232 | regnode *trie_node= scan; | |
4233 | regnode *tail= regnext(scan); | |
f8fc2ecf | 4234 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
8aa23a47 YO |
4235 | I32 max1 = 0, min1 = I32_MAX; |
4236 | struct regnode_charclass_class accum; | |
4237 | ||
4238 | if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */ | |
304ee84b | 4239 | SCAN_COMMIT(pRExC_state, data,minlenp); /* Cannot merge strings after this. */ |
8aa23a47 | 4240 | if (flags & SCF_DO_STCLASS) |
e755fd73 | 4241 | cl_init_zero(pRExC_state, &accum); |
8aa23a47 YO |
4242 | |
4243 | if (!trie->jump) { | |
4244 | min1= trie->minlen; | |
4245 | max1= trie->maxlen; | |
4246 | } else { | |
4247 | const regnode *nextbranch= NULL; | |
4248 | U32 word; | |
4249 | ||
4250 | for ( word=1 ; word <= trie->wordcount ; word++) | |
4251 | { | |
4252 | I32 deltanext=0, minnext=0, f = 0, fake; | |
4253 | struct regnode_charclass_class this_class; | |
4254 | ||
4255 | data_fake.flags = 0; | |
4256 | if (data) { | |
4257 | data_fake.whilem_c = data->whilem_c; | |
4258 | data_fake.last_closep = data->last_closep; | |
4259 | } | |
4260 | else | |
4261 | data_fake.last_closep = &fake; | |
58e23c8d | 4262 | data_fake.pos_delta = delta; |
8aa23a47 | 4263 | if (flags & SCF_DO_STCLASS) { |
e755fd73 | 4264 | cl_init(pRExC_state, &this_class); |
8aa23a47 YO |
4265 | data_fake.start_class = &this_class; |
4266 | f = SCF_DO_STCLASS_AND; | |
4267 | } | |
4268 | if (flags & SCF_WHILEM_VISITED_POS) | |
4269 | f |= SCF_WHILEM_VISITED_POS; | |
4270 | ||
4271 | if (trie->jump[word]) { | |
4272 | if (!nextbranch) | |
4273 | nextbranch = trie_node + trie->jump[0]; | |
4274 | scan= trie_node + trie->jump[word]; | |
4275 | /* We go from the jump point to the branch that follows | |
4276 | it. Note this means we need the vestigal unused branches | |
4277 | even though they arent otherwise used. | |
4278 | */ | |
4279 | minnext = study_chunk(pRExC_state, &scan, minlenp, | |
4280 | &deltanext, (regnode *)nextbranch, &data_fake, | |
4281 | stopparen, recursed, NULL, f,depth+1); | |
4282 | } | |
4283 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) | |
4284 | nextbranch= regnext((regnode*)nextbranch); | |
4285 | ||
4286 | if (min1 > (I32)(minnext + trie->minlen)) | |
4287 | min1 = minnext + trie->minlen; | |
4288 | if (max1 < (I32)(minnext + deltanext + trie->maxlen)) | |
4289 | max1 = minnext + deltanext + trie->maxlen; | |
4290 | if (deltanext == I32_MAX) | |
4291 | is_inf = is_inf_internal = 1; | |
4292 | ||
4293 | if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) | |
4294 | pars++; | |
4295 | if (data_fake.flags & SCF_SEEN_ACCEPT) { | |
4296 | if ( stopmin > min + min1) | |
4297 | stopmin = min + min1; | |
4298 | flags &= ~SCF_DO_SUBSTR; | |
4299 | if (data) | |
4300 | data->flags |= SCF_SEEN_ACCEPT; | |
4301 | } | |
4302 | if (data) { | |
4303 | if (data_fake.flags & SF_HAS_EVAL) | |
4304 | data->flags |= SF_HAS_EVAL; | |
4305 | data->whilem_c = data_fake.whilem_c; | |
4306 | } | |
4307 | if (flags & SCF_DO_STCLASS) | |
3fffb88a | 4308 | cl_or(pRExC_state, &accum, &this_class); |
8aa23a47 YO |
4309 | } |
4310 | } | |
4311 | if (flags & SCF_DO_SUBSTR) { | |
4312 | data->pos_min += min1; | |
4313 | data->pos_delta += max1 - min1; | |
4314 | if (max1 != min1 || is_inf) | |
4315 | data->longest = &(data->longest_float); | |
4316 | } | |
4317 | min += min1; | |
4318 | delta += max1 - min1; | |
4319 | if (flags & SCF_DO_STCLASS_OR) { | |
3fffb88a | 4320 | cl_or(pRExC_state, data->start_class, &accum); |
8aa23a47 YO |
4321 | if (min1) { |
4322 | cl_and(data->start_class, and_withp); | |
4323 | flags &= ~SCF_DO_STCLASS; | |
4324 | } | |
4325 | } | |
4326 | else if (flags & SCF_DO_STCLASS_AND) { | |
4327 | if (min1) { | |
4328 | cl_and(data->start_class, &accum); | |
4329 | flags &= ~SCF_DO_STCLASS; | |
4330 | } | |
4331 | else { | |
4332 | /* Switch to OR mode: cache the old value of | |
4333 | * data->start_class */ | |
4334 | INIT_AND_WITHP; | |
4335 | StructCopy(data->start_class, and_withp, | |
4336 | struct regnode_charclass_class); | |
4337 | flags &= ~SCF_DO_STCLASS_AND; | |
4338 | StructCopy(&accum, data->start_class, | |
4339 | struct regnode_charclass_class); | |
4340 | flags |= SCF_DO_STCLASS_OR; | |
4341 | data->start_class->flags |= ANYOF_EOS; | |
4342 | } | |
4343 | } | |
4344 | scan= tail; | |
4345 | continue; | |
4346 | } | |
786e8c11 | 4347 | #else |
8aa23a47 | 4348 | else if (PL_regkind[OP(scan)] == TRIE) { |
f8fc2ecf | 4349 | reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ]; |
8aa23a47 YO |
4350 | U8*bang=NULL; |
4351 | ||
4352 | min += trie->minlen; | |
4353 | delta += (trie->maxlen - trie->minlen); | |
4354 | flags &= ~SCF_DO_STCLASS; /* xxx */ | |
4355 | if (flags & SCF_DO_SUBSTR) { | |
304ee84b | 4356 | SCAN_COMMIT(pRExC_state,data,minlenp); /* Cannot expect anything... */ |
8aa23a47 YO |
4357 | data->pos_min += trie->minlen; |
4358 | data->pos_delta += (trie->maxlen - trie->minlen); | |
4359 | if (trie->maxlen != trie->minlen) | |
4360 | data->longest = &(data->longest_float); | |
4361 | } | |
4362 | if (trie->jump) /* no more substrings -- for now /grr*/ | |
4363 | flags &= ~SCF_DO_SUBSTR; | |
b515a41d | 4364 | } |
8aa23a47 | 4365 | #endif /* old or new */ |
686b73d4 | 4366 | #endif /* TRIE_STUDY_OPT */ |
e1d1eefb | 4367 | |
8aa23a47 YO |
4368 | /* Else: zero-length, ignore. */ |
4369 | scan = regnext(scan); | |
4370 | } | |
4371 | if (frame) { | |
4372 | last = frame->last; | |
4373 | scan = frame->next; | |
4374 | stopparen = frame->stop; | |
4375 | frame = frame->prev; | |
4376 | goto fake_study_recurse; | |
c277df42 IZ |
4377 | } |
4378 | ||
4379 | finish: | |
8aa23a47 | 4380 | assert(!frame); |
304ee84b | 4381 | DEBUG_STUDYDATA("pre-fin:",data,depth); |
8aa23a47 | 4382 | |
c277df42 | 4383 | *scanp = scan; |
aca2d497 | 4384 | *deltap = is_inf_internal ? I32_MAX : delta; |
b81d288d | 4385 | if (flags & SCF_DO_SUBSTR && is_inf) |
c277df42 | 4386 | data->pos_delta = I32_MAX - data->pos_min; |
786e8c11 | 4387 | if (is_par > (I32)U8_MAX) |
c277df42 IZ |
4388 | is_par = 0; |
4389 | if (is_par && pars==1 && data) { | |
4390 | data->flags |= SF_IN_PAR; | |
4391 | data->flags &= ~SF_HAS_PAR; | |
a0ed51b3 LW |
4392 | } |
4393 | else if (pars && data) { | |
c277df42 IZ |
4394 | data->flags |= SF_HAS_PAR; |
4395 | data->flags &= ~SF_IN_PAR; | |
4396 | } | |
653099ff | 4397 | if (flags & SCF_DO_STCLASS_OR) |
40d049e4 | 4398 | cl_and(data->start_class, and_withp); |
786e8c11 YO |
4399 | if (flags & SCF_TRIE_RESTUDY) |
4400 | data->flags |= SCF_TRIE_RESTUDY; | |
1de06328 | 4401 | |
304ee84b | 4402 | DEBUG_STUDYDATA("post-fin:",data,depth); |
1de06328 | 4403 | |
e2e6a0f1 | 4404 | return min < stopmin ? min : stopmin; |
c277df42 IZ |
4405 | } |
4406 | ||
2eccd3b2 NC |
4407 | STATIC U32 |
4408 | S_add_data(RExC_state_t *pRExC_state, U32 n, const char *s) | |
c277df42 | 4409 | { |
4a4e7719 NC |
4410 | U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0; |
4411 | ||
7918f24d NC |
4412 | PERL_ARGS_ASSERT_ADD_DATA; |
4413 | ||
4a4e7719 NC |
4414 | Renewc(RExC_rxi->data, |
4415 | sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1), | |
4416 | char, struct reg_data); | |
4417 | if(count) | |
f8fc2ecf | 4418 | Renew(RExC_rxi->data->what, count + n, U8); |
4a4e7719 | 4419 | else |
f8fc2ecf | 4420 | Newx(RExC_rxi->data->what, n, U8); |
4a4e7719 NC |
4421 | RExC_rxi->data->count = count + n; |
4422 | Copy(s, RExC_rxi->data->what + count, n, U8); | |
4423 | return count; | |
c277df42 IZ |
4424 | } |
4425 | ||
f8149455 | 4426 | /*XXX: todo make this not included in a non debugging perl */ |
76234dfb | 4427 | #ifndef PERL_IN_XSUB_RE |
d88dccdf | 4428 | void |
864dbfa3 | 4429 | Perl_reginitcolors(pTHX) |
d88dccdf | 4430 | { |
97aff369 | 4431 | dVAR; |
1df70142 | 4432 | const char * const s = PerlEnv_getenv("PERL_RE_COLORS"); |
d88dccdf | 4433 | if (s) { |
1df70142 AL |
4434 | char *t = savepv(s); |
4435 | int i = 0; | |
4436 | PL_colors[0] = t; | |
d88dccdf | 4437 | while (++i < 6) { |
1df70142 AL |
4438 | t = strchr(t, '\t'); |
4439 | if (t) { | |
4440 | *t = '\0'; | |
4441 | PL_colors[i] = ++t; | |
d88dccdf IZ |
4442 | } |
4443 | else | |
1df70142 | 4444 | PL_colors[i] = t = (char *)""; |
d88dccdf IZ |
4445 | } |
4446 | } else { | |
1df70142 | 4447 | int i = 0; |
b81d288d | 4448 | while (i < 6) |
06b5626a | 4449 | PL_colors[i++] = (char *)""; |
d88dccdf IZ |
4450 | } |
4451 | PL_colorset = 1; | |
4452 | } | |
76234dfb | 4453 | #endif |
8615cb43 | 4454 | |
07be1b83 | 4455 | |
786e8c11 YO |
4456 | #ifdef TRIE_STUDY_OPT |
4457 | #define CHECK_RESTUDY_GOTO \ | |
4458 | if ( \ | |
4459 | (data.flags & SCF_TRIE_RESTUDY) \ | |
4460 | && ! restudied++ \ | |
4461 | ) goto reStudy | |
4462 | #else | |
4463 | #define CHECK_RESTUDY_GOTO | |
4464 | #endif | |
f9f4320a | 4465 | |
a687059c | 4466 | /* |
e50aee73 | 4467 | - pregcomp - compile a regular expression into internal code |
a687059c LW |
4468 | * |
4469 | * We can't allocate space until we know how big the compiled form will be, | |
4470 | * but we can't compile it (and thus know how big it is) until we've got a | |
4471 | * place to put the code. So we cheat: we compile it twice, once with code | |
4472 | * generation turned off and size counting turned on, and once "for real". | |
4473 | * This also means that we don't allocate space until we are sure that the | |
4474 | * thing really will compile successfully, and we never have to move the | |
4475 | * code and thus invalidate pointers into it. (Note that it has to be in | |
4476 | * one piece because free() must be able to free it all.) [NB: not true in perl] | |
4477 | * | |
4478 | * Beware that the optimization-preparation code in here knows about some | |
4479 | * of the structure of the compiled regexp. [I'll say.] | |
4480 | */ | |
b9b4dddf YO |
4481 | |
4482 | ||
4483 | ||
f9f4320a | 4484 | #ifndef PERL_IN_XSUB_RE |
f9f4320a YO |
4485 | #define RE_ENGINE_PTR &PL_core_reg_engine |
4486 | #else | |
f9f4320a YO |
4487 | extern const struct regexp_engine my_reg_engine; |
4488 | #define RE_ENGINE_PTR &my_reg_engine | |
4489 | #endif | |
6d5c990f RGS |
4490 | |
4491 | #ifndef PERL_IN_XSUB_RE | |
3ab4a224 | 4492 | REGEXP * |
1593ad57 | 4493 | Perl_pregcomp(pTHX_ SV * const pattern, const U32 flags) |
a687059c | 4494 | { |
97aff369 | 4495 | dVAR; |
6d5c990f | 4496 | HV * const table = GvHV(PL_hintgv); |
7918f24d NC |
4497 | |
4498 | PERL_ARGS_ASSERT_PREGCOMP; | |
4499 | ||
f9f4320a YO |
4500 | /* Dispatch a request to compile a regexp to correct |
4501 | regexp engine. */ | |
f9f4320a YO |
4502 | if (table) { |
4503 | SV **ptr= hv_fetchs(table, "regcomp", FALSE); | |
6d5c990f | 4504 | GET_RE_DEBUG_FLAGS_DECL; |
1e2e3d02 | 4505 | if (ptr && SvIOK(*ptr) && SvIV(*ptr)) { |
f9f4320a YO |
4506 | const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr)); |
4507 | DEBUG_COMPILE_r({ | |
8d8756e7 | 4508 | PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n", |
f9f4320a YO |
4509 | SvIV(*ptr)); |
4510 | }); | |
3ab4a224 | 4511 | return CALLREGCOMP_ENG(eng, pattern, flags); |
f9f4320a | 4512 | } |
b9b4dddf | 4513 | } |
3ab4a224 | 4514 | return Perl_re_compile(aTHX_ pattern, flags); |
2a5d9b1d | 4515 | } |
6d5c990f | 4516 | #endif |
2a5d9b1d | 4517 | |
3ab4a224 | 4518 | REGEXP * |
29b09c41 | 4519 | Perl_re_compile(pTHX_ SV * const pattern, U32 orig_pm_flags) |
2a5d9b1d RGS |
4520 | { |
4521 | dVAR; | |
288b8c02 NC |
4522 | REGEXP *rx; |
4523 | struct regexp *r; | |
f8fc2ecf | 4524 | register regexp_internal *ri; |
3ab4a224 | 4525 | STRLEN plen; |
4d6b2893 | 4526 | char* VOL exp; |
5d51ce98 | 4527 | char* xend; |
c277df42 | 4528 | regnode *scan; |
a0d0e21e | 4529 | I32 flags; |
a0d0e21e | 4530 | I32 minlen = 0; |
29b09c41 | 4531 | U32 pm_flags; |
e7f38d0f YO |
4532 | |
4533 | /* these are all flags - maybe they should be turned | |
4534 | * into a single int with different bit masks */ | |
4535 | I32 sawlookahead = 0; | |
a0d0e21e LW |
4536 | I32 sawplus = 0; |
4537 | I32 sawopen = 0; | |
29b09c41 | 4538 | bool used_setjump = FALSE; |
4624b182 | 4539 | regex_charset initial_charset = get_regex_charset(orig_pm_flags); |
e7f38d0f | 4540 | |
bbd61b5f KW |
4541 | U8 jump_ret = 0; |
4542 | dJMPENV; | |
2c2d71f5 | 4543 | scan_data_t data; |
830247a4 | 4544 | RExC_state_t RExC_state; |
be8e71aa | 4545 | RExC_state_t * const pRExC_state = &RExC_state; |
07be1b83 | 4546 | #ifdef TRIE_STUDY_OPT |
5d51ce98 | 4547 | int restudied; |
07be1b83 YO |
4548 | RExC_state_t copyRExC_state; |
4549 | #endif | |
2a5d9b1d | 4550 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
4551 | |
4552 | PERL_ARGS_ASSERT_RE_COMPILE; | |
4553 | ||
6d5c990f | 4554 | DEBUG_r(if (!PL_colorset) reginitcolors()); |
a0d0e21e | 4555 | |
11951bcb KW |
4556 | exp = SvPV(pattern, plen); |
4557 | ||
4558 | if (plen == 0) { /* ignore the utf8ness if the pattern is 0 length */ | |
4559 | RExC_utf8 = RExC_orig_utf8 = 0; | |
4560 | } | |
4561 | else { | |
4562 | RExC_utf8 = RExC_orig_utf8 = SvUTF8(pattern); | |
4563 | } | |
e40e74fe | 4564 | RExC_uni_semantics = 0; |
4624b182 | 4565 | RExC_contains_locale = 0; |
7b597bb8 | 4566 | |
d6bd454d | 4567 | /****************** LONG JUMP TARGET HERE***********************/ |
bbd61b5f KW |
4568 | /* Longjmp back to here if have to switch in midstream to utf8 */ |
4569 | if (! RExC_orig_utf8) { | |
4570 | JMPENV_PUSH(jump_ret); | |
29b09c41 | 4571 | used_setjump = TRUE; |
bbd61b5f KW |
4572 | } |
4573 | ||
5d51ce98 | 4574 | if (jump_ret == 0) { /* First time through */ |
29b09c41 | 4575 | xend = exp + plen; |
29b09c41 | 4576 | |
5d51ce98 KW |
4577 | DEBUG_COMPILE_r({ |
4578 | SV *dsv= sv_newmortal(); | |
4579 | RE_PV_QUOTED_DECL(s, RExC_utf8, | |
4580 | dsv, exp, plen, 60); | |
4581 | PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n", | |
4582 | PL_colors[4],PL_colors[5],s); | |
4583 | }); | |
4584 | } | |
4585 | else { /* longjumped back */ | |
bbd61b5f KW |
4586 | STRLEN len = plen; |
4587 | ||
5d51ce98 KW |
4588 | /* If the cause for the longjmp was other than changing to utf8, pop |
4589 | * our own setjmp, and longjmp to the correct handler */ | |
bbd61b5f KW |
4590 | if (jump_ret != UTF8_LONGJMP) { |
4591 | JMPENV_POP; | |
4592 | JMPENV_JUMP(jump_ret); | |
4593 | } | |
4594 | ||
595598ee KW |
4595 | GET_RE_DEBUG_FLAGS; |
4596 | ||
bbd61b5f KW |
4597 | /* It's possible to write a regexp in ascii that represents Unicode |
4598 | codepoints outside of the byte range, such as via \x{100}. If we | |
4599 | detect such a sequence we have to convert the entire pattern to utf8 | |
4600 | and then recompile, as our sizing calculation will have been based | |
4601 | on 1 byte == 1 character, but we will need to use utf8 to encode | |
4602 | at least some part of the pattern, and therefore must convert the whole | |
4603 | thing. | |
4604 | -- dmq */ | |
4605 | DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, | |
4606 | "UTF8 mismatch! Converting to utf8 for resizing and compile\n")); | |
3e0b93e8 KW |
4607 | exp = (char*)Perl_bytes_to_utf8(aTHX_ |
4608 | (U8*)SvPV_nomg(pattern, plen), | |
4609 | &len); | |
bbd61b5f KW |
4610 | xend = exp + len; |
4611 | RExC_orig_utf8 = RExC_utf8 = 1; | |
4612 | SAVEFREEPV(exp); | |
4613 | } | |
4614 | ||
5d51ce98 KW |
4615 | #ifdef TRIE_STUDY_OPT |
4616 | restudied = 0; | |
4617 | #endif | |
4618 | ||
29b09c41 | 4619 | pm_flags = orig_pm_flags; |
a62b1201 | 4620 | |
4624b182 KW |
4621 | if (initial_charset == REGEX_LOCALE_CHARSET) { |
4622 | RExC_contains_locale = 1; | |
4623 | } | |
4624 | else if (RExC_utf8 && initial_charset == REGEX_DEPENDS_CHARSET) { | |
4625 | ||
4626 | /* Set to use unicode semantics if the pattern is in utf8 and has the | |
4627 | * 'depends' charset specified, as it means unicode when utf8 */ | |
a62b1201 | 4628 | set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET); |
29b09c41 KW |
4629 | } |
4630 | ||
02daf0ab | 4631 | RExC_precomp = exp; |
c737faaf | 4632 | RExC_flags = pm_flags; |
830247a4 | 4633 | RExC_sawback = 0; |
bbce6d69 | 4634 | |
830247a4 | 4635 | RExC_seen = 0; |
b57e4118 | 4636 | RExC_in_lookbehind = 0; |
830247a4 IZ |
4637 | RExC_seen_zerolen = *exp == '^' ? -1 : 0; |
4638 | RExC_seen_evals = 0; | |
4639 | RExC_extralen = 0; | |
e2a7e165 | 4640 | RExC_override_recoding = 0; |
c277df42 | 4641 | |
bbce6d69 | 4642 | /* First pass: determine size, legality. */ |
830247a4 | 4643 | RExC_parse = exp; |
fac92740 | 4644 | RExC_start = exp; |
830247a4 IZ |
4645 | RExC_end = xend; |
4646 | RExC_naughty = 0; | |
4647 | RExC_npar = 1; | |
e2e6a0f1 | 4648 | RExC_nestroot = 0; |
830247a4 IZ |
4649 | RExC_size = 0L; |
4650 | RExC_emit = &PL_regdummy; | |
4651 | RExC_whilem_seen = 0; | |
40d049e4 YO |
4652 | RExC_open_parens = NULL; |
4653 | RExC_close_parens = NULL; | |
4654 | RExC_opend = NULL; | |
81714fb9 | 4655 | RExC_paren_names = NULL; |
1f1031fe YO |
4656 | #ifdef DEBUGGING |
4657 | RExC_paren_name_list = NULL; | |
4658 | #endif | |
40d049e4 YO |
4659 | RExC_recurse = NULL; |
4660 | RExC_recurse_count = 0; | |
81714fb9 | 4661 | |
85ddcde9 JH |
4662 | #if 0 /* REGC() is (currently) a NOP at the first pass. |
4663 | * Clever compilers notice this and complain. --jhi */ | |
830247a4 | 4664 | REGC((U8)REG_MAGIC, (char*)RExC_emit); |
85ddcde9 | 4665 | #endif |
44bed856 KW |
4666 | DEBUG_PARSE_r( |
4667 | PerlIO_printf(Perl_debug_log, "Starting first pass (sizing)\n"); | |
4668 | RExC_lastnum=0; | |
4669 | RExC_lastparse=NULL; | |
4670 | ); | |
3dab1dad | 4671 | if (reg(pRExC_state, 0, &flags,1) == NULL) { |
c445ea15 | 4672 | RExC_precomp = NULL; |
a0d0e21e LW |
4673 | return(NULL); |
4674 | } | |
bbd61b5f | 4675 | |
29b09c41 KW |
4676 | /* Here, finished first pass. Get rid of any added setjmp */ |
4677 | if (used_setjump) { | |
bbd61b5f | 4678 | JMPENV_POP; |
02daf0ab | 4679 | } |
e40e74fe | 4680 | |
07be1b83 | 4681 | DEBUG_PARSE_r({ |
81714fb9 YO |
4682 | PerlIO_printf(Perl_debug_log, |
4683 | "Required size %"IVdf" nodes\n" | |
4684 | "Starting second pass (creation)\n", | |
4685 | (IV)RExC_size); | |
07be1b83 YO |
4686 | RExC_lastnum=0; |
4687 | RExC_lastparse=NULL; | |
4688 | }); | |
e40e74fe KW |
4689 | |
4690 | /* The first pass could have found things that force Unicode semantics */ | |
4691 | if ((RExC_utf8 || RExC_uni_semantics) | |
4692 | && get_regex_charset(pm_flags) == REGEX_DEPENDS_CHARSET) | |
4693 | { | |
4694 | set_regex_charset(&pm_flags, REGEX_UNICODE_CHARSET); | |
4695 | } | |
4696 | ||
c277df42 IZ |
4697 | /* Small enough for pointer-storage convention? |
4698 | If extralen==0, this means that we will not need long jumps. */ | |
830247a4 IZ |
4699 | if (RExC_size >= 0x10000L && RExC_extralen) |
4700 | RExC_size += RExC_extralen; | |
c277df42 | 4701 | else |
830247a4 IZ |
4702 | RExC_extralen = 0; |
4703 | if (RExC_whilem_seen > 15) | |
4704 | RExC_whilem_seen = 15; | |
a0d0e21e | 4705 | |
f9f4320a YO |
4706 | /* Allocate space and zero-initialize. Note, the two step process |
4707 | of zeroing when in debug mode, thus anything assigned has to | |
4708 | happen after that */ | |
d2f13c59 | 4709 | rx = (REGEXP*) newSV_type(SVt_REGEXP); |
288b8c02 | 4710 | r = (struct regexp*)SvANY(rx); |
f8fc2ecf YO |
4711 | Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), |
4712 | char, regexp_internal); | |
4713 | if ( r == NULL || ri == NULL ) | |
b45f050a | 4714 | FAIL("Regexp out of space"); |
0f79a09d GS |
4715 | #ifdef DEBUGGING |
4716 | /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */ | |
f8fc2ecf | 4717 | Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char); |
58e23c8d | 4718 | #else |
f8fc2ecf YO |
4719 | /* bulk initialize base fields with 0. */ |
4720 | Zero(ri, sizeof(regexp_internal), char); | |
0f79a09d | 4721 | #endif |
58e23c8d YO |
4722 | |
4723 | /* non-zero initialization begins here */ | |
f8fc2ecf | 4724 | RXi_SET( r, ri ); |
f9f4320a | 4725 | r->engine= RE_ENGINE_PTR; |
c737faaf | 4726 | r->extflags = pm_flags; |
bcdf7404 | 4727 | { |
f7819f85 | 4728 | bool has_p = ((r->extflags & RXf_PMf_KEEPCOPY) == RXf_PMf_KEEPCOPY); |
a62b1201 | 4729 | bool has_charset = (get_regex_charset(r->extflags) != REGEX_DEPENDS_CHARSET); |
c5ea2ffa KW |
4730 | |
4731 | /* The caret is output if there are any defaults: if not all the STD | |
4732 | * flags are set, or if no character set specifier is needed */ | |
4733 | bool has_default = | |
4734 | (((r->extflags & RXf_PMf_STD_PMMOD) != RXf_PMf_STD_PMMOD) | |
4735 | || ! has_charset); | |
bcdf7404 | 4736 | bool has_runon = ((RExC_seen & REG_SEEN_RUN_ON_COMMENT)==REG_SEEN_RUN_ON_COMMENT); |
14f3b9f2 NC |
4737 | U16 reganch = (U16)((r->extflags & RXf_PMf_STD_PMMOD) |
4738 | >> RXf_PMf_STD_PMMOD_SHIFT); | |
bcdf7404 YO |
4739 | const char *fptr = STD_PAT_MODS; /*"msix"*/ |
4740 | char *p; | |
fb85c044 | 4741 | /* Allocate for the worst case, which is all the std flags are turned |
c5ea2ffa KW |
4742 | * on. If more precision is desired, we could do a population count of |
4743 | * the flags set. This could be done with a small lookup table, or by | |
4744 | * shifting, masking and adding, or even, when available, assembly | |
4745 | * language for a machine-language population count. | |
4746 | * We never output a minus, as all those are defaults, so are | |
4747 | * covered by the caret */ | |
fb85c044 | 4748 | const STRLEN wraplen = plen + has_p + has_runon |
c5ea2ffa | 4749 | + has_default /* If needs a caret */ |
a62b1201 KW |
4750 | |
4751 | /* If needs a character set specifier */ | |
4752 | + ((has_charset) ? MAX_CHARSET_NAME_LENGTH : 0) | |
bcdf7404 YO |
4753 | + (sizeof(STD_PAT_MODS) - 1) |
4754 | + (sizeof("(?:)") - 1); | |
4755 | ||
c5ea2ffa | 4756 | p = sv_grow(MUTABLE_SV(rx), wraplen + 1); /* +1 for the ending NUL */ |
f7c278bf | 4757 | SvPOK_on(rx); |
8f6ae13c | 4758 | SvFLAGS(rx) |= SvUTF8(pattern); |
bcdf7404 | 4759 | *p++='('; *p++='?'; |
9de15fec KW |
4760 | |
4761 | /* If a default, cover it using the caret */ | |
c5ea2ffa | 4762 | if (has_default) { |
85508812 | 4763 | *p++= DEFAULT_PAT_MOD; |
fb85c044 | 4764 | } |
c5ea2ffa | 4765 | if (has_charset) { |
a62b1201 KW |
4766 | STRLEN len; |
4767 | const char* const name = get_regex_charset_name(r->extflags, &len); | |
4768 | Copy(name, p, len, char); | |
4769 | p += len; | |
9de15fec | 4770 | } |
f7819f85 A |
4771 | if (has_p) |
4772 | *p++ = KEEPCOPY_PAT_MOD; /*'p'*/ | |
bcdf7404 | 4773 | { |
bcdf7404 | 4774 | char ch; |
bcdf7404 YO |
4775 | while((ch = *fptr++)) { |
4776 | if(reganch & 1) | |
4777 | *p++ = ch; | |
bcdf7404 YO |
4778 | reganch >>= 1; |
4779 | } | |
bcdf7404 YO |
4780 | } |
4781 | ||
28d8d7f4 | 4782 | *p++ = ':'; |
bb661a58 | 4783 | Copy(RExC_precomp, p, plen, char); |
efd26800 NC |
4784 | assert ((RX_WRAPPED(rx) - p) < 16); |
4785 | r->pre_prefix = p - RX_WRAPPED(rx); | |
bb661a58 | 4786 | p += plen; |
bcdf7404 | 4787 | if (has_runon) |
28d8d7f4 YO |
4788 | *p++ = '\n'; |
4789 | *p++ = ')'; | |
4790 | *p = 0; | |
fb85c044 | 4791 | SvCUR_set(rx, p - SvPVX_const(rx)); |
bcdf7404 YO |
4792 | } |
4793 | ||
bbe252da | 4794 | r->intflags = 0; |
830247a4 | 4795 | r->nparens = RExC_npar - 1; /* set early to validate backrefs */ |
81714fb9 | 4796 | |
6bda09f9 | 4797 | if (RExC_seen & REG_SEEN_RECURSE) { |
40d049e4 YO |
4798 | Newxz(RExC_open_parens, RExC_npar,regnode *); |
4799 | SAVEFREEPV(RExC_open_parens); | |
4800 | Newxz(RExC_close_parens,RExC_npar,regnode *); | |
4801 | SAVEFREEPV(RExC_close_parens); | |
6bda09f9 YO |
4802 | } |
4803 | ||
4804 | /* Useful during FAIL. */ | |
7122b237 YO |
4805 | #ifdef RE_TRACK_PATTERN_OFFSETS |
4806 | Newxz(ri->u.offsets, 2*RExC_size+1, U32); /* MJD 20001228 */ | |
a3621e74 | 4807 | DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log, |
2af232bd | 4808 | "%s %"UVuf" bytes for offset annotations.\n", |
7122b237 | 4809 | ri->u.offsets ? "Got" : "Couldn't get", |
392fbf5d | 4810 | (UV)((2*RExC_size+1) * sizeof(U32)))); |
7122b237 YO |
4811 | #endif |
4812 | SetProgLen(ri,RExC_size); | |
288b8c02 | 4813 | RExC_rx_sv = rx; |
830247a4 | 4814 | RExC_rx = r; |
f8fc2ecf | 4815 | RExC_rxi = ri; |
bbce6d69 | 4816 | |
4817 | /* Second pass: emit code. */ | |
c737faaf | 4818 | RExC_flags = pm_flags; /* don't let top level (?i) bleed */ |
830247a4 IZ |
4819 | RExC_parse = exp; |
4820 | RExC_end = xend; | |
4821 | RExC_naughty = 0; | |
4822 | RExC_npar = 1; | |
f8fc2ecf YO |
4823 | RExC_emit_start = ri->program; |
4824 | RExC_emit = ri->program; | |
3b57cd43 YO |
4825 | RExC_emit_bound = ri->program + RExC_size + 1; |
4826 | ||
2cd61cdb | 4827 | /* Store the count of eval-groups for security checks: */ |
f8149455 | 4828 | RExC_rx->seen_evals = RExC_seen_evals; |
830247a4 | 4829 | REGC((U8)REG_MAGIC, (char*) RExC_emit++); |
80757612 | 4830 | if (reg(pRExC_state, 0, &flags,1) == NULL) { |
288b8c02 | 4831 | ReREFCNT_dec(rx); |
a0d0e21e | 4832 | return(NULL); |
80757612 | 4833 | } |
07be1b83 YO |
4834 | /* XXXX To minimize changes to RE engine we always allocate |
4835 | 3-units-long substrs field. */ | |
4836 | Newx(r->substrs, 1, struct reg_substr_data); | |
40d049e4 YO |
4837 | if (RExC_recurse_count) { |
4838 | Newxz(RExC_recurse,RExC_recurse_count,regnode *); | |
4839 | SAVEFREEPV(RExC_recurse); | |
4840 | } | |
a0d0e21e | 4841 | |
07be1b83 | 4842 | reStudy: |
e7f38d0f | 4843 | r->minlen = minlen = sawlookahead = sawplus = sawopen = 0; |
07be1b83 | 4844 | Zero(r->substrs, 1, struct reg_substr_data); |
a3621e74 | 4845 | |
07be1b83 | 4846 | #ifdef TRIE_STUDY_OPT |
0934c9d9 SH |
4847 | if (!restudied) { |
4848 | StructCopy(&zero_scan_data, &data, scan_data_t); | |
4849 | copyRExC_state = RExC_state; | |
4850 | } else { | |
5d458dd8 | 4851 | U32 seen=RExC_seen; |
07be1b83 | 4852 | DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n")); |
5d458dd8 YO |
4853 | |
4854 | RExC_state = copyRExC_state; | |
4855 | if (seen & REG_TOP_LEVEL_BRANCHES) | |
4856 | RExC_seen |= REG_TOP_LEVEL_BRANCHES; | |
4857 | else | |
4858 | RExC_seen &= ~REG_TOP_LEVEL_BRANCHES; | |
1de06328 | 4859 | if (data.last_found) { |
07be1b83 | 4860 | SvREFCNT_dec(data.longest_fixed); |
07be1b83 | 4861 | SvREFCNT_dec(data.longest_float); |
07be1b83 | 4862 | SvREFCNT_dec(data.last_found); |
1de06328 | 4863 | } |
40d049e4 | 4864 | StructCopy(&zero_scan_data, &data, scan_data_t); |
07be1b83 | 4865 | } |
40d049e4 YO |
4866 | #else |
4867 | StructCopy(&zero_scan_data, &data, scan_data_t); | |
07be1b83 | 4868 | #endif |
fc8cd66c | 4869 | |
a0d0e21e | 4870 | /* Dig out information for optimizations. */ |
f7819f85 | 4871 | r->extflags = RExC_flags; /* was pm_op */ |
c737faaf YO |
4872 | /*dmq: removed as part of de-PMOP: pm->op_pmflags = RExC_flags; */ |
4873 | ||
a0ed51b3 | 4874 | if (UTF) |
8f6ae13c | 4875 | SvUTF8_on(rx); /* Unicode in it? */ |
f8fc2ecf | 4876 | ri->regstclass = NULL; |
830247a4 | 4877 | if (RExC_naughty >= 10) /* Probably an expensive pattern. */ |
bbe252da | 4878 | r->intflags |= PREGf_NAUGHTY; |
f8fc2ecf | 4879 | scan = ri->program + 1; /* First BRANCH. */ |
2779dcf1 | 4880 | |
1de06328 YO |
4881 | /* testing for BRANCH here tells us whether there is "must appear" |
4882 | data in the pattern. If there is then we can use it for optimisations */ | |
eaf3ca90 | 4883 | if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */ |
c277df42 | 4884 | I32 fake; |
c5254dd6 | 4885 | STRLEN longest_float_length, longest_fixed_length; |
07be1b83 | 4886 | struct regnode_charclass_class ch_class; /* pointed to by data */ |
653099ff | 4887 | int stclass_flag; |
07be1b83 | 4888 | I32 last_close = 0; /* pointed to by data */ |
5339e136 YO |
4889 | regnode *first= scan; |
4890 | regnode *first_next= regnext(first); | |
639081d6 YO |
4891 | /* |
4892 | * Skip introductions and multiplicators >= 1 | |
4893 | * so that we can extract the 'meat' of the pattern that must | |
4894 | * match in the large if() sequence following. | |
4895 | * NOTE that EXACT is NOT covered here, as it is normally | |
4896 | * picked up by the optimiser separately. | |
4897 | * | |
4898 | * This is unfortunate as the optimiser isnt handling lookahead | |
4899 | * properly currently. | |
4900 | * | |
4901 | */ | |
a0d0e21e | 4902 | while ((OP(first) == OPEN && (sawopen = 1)) || |
653099ff | 4903 | /* An OR of *one* alternative - should not happen now. */ |
5339e136 | 4904 | (OP(first) == BRANCH && OP(first_next) != BRANCH) || |
07be1b83 | 4905 | /* for now we can't handle lookbehind IFMATCH*/ |
e7f38d0f | 4906 | (OP(first) == IFMATCH && !first->flags && (sawlookahead = 1)) || |
a0d0e21e LW |
4907 | (OP(first) == PLUS) || |
4908 | (OP(first) == MINMOD) || | |
653099ff | 4909 | /* An {n,m} with n>0 */ |
5339e136 YO |
4910 | (PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) || |
4911 | (OP(first) == NOTHING && PL_regkind[OP(first_next)] != END )) | |
07be1b83 | 4912 | { |
639081d6 YO |
4913 | /* |
4914 | * the only op that could be a regnode is PLUS, all the rest | |
4915 | * will be regnode_1 or regnode_2. | |
4916 | * | |
4917 | */ | |
a0d0e21e LW |
4918 | if (OP(first) == PLUS) |
4919 | sawplus = 1; | |
4920 | else | |
3dab1dad | 4921 | first += regarglen[OP(first)]; |
686b73d4 | 4922 | |
639081d6 | 4923 | first = NEXTOPER(first); |
5339e136 | 4924 | first_next= regnext(first); |
a687059c LW |
4925 | } |
4926 | ||
a0d0e21e LW |
4927 | /* Starting-point info. */ |
4928 | again: | |
786e8c11 | 4929 | DEBUG_PEEP("first:",first,0); |
07be1b83 | 4930 | /* Ignore EXACT as we deal with it later. */ |
3dab1dad | 4931 | if (PL_regkind[OP(first)] == EXACT) { |
1aa99e6b | 4932 | if (OP(first) == EXACT) |
6f207bd3 | 4933 | NOOP; /* Empty, get anchored substr later. */ |
e5fbd0ff | 4934 | else |
f8fc2ecf | 4935 | ri->regstclass = first; |
b3c9acc1 | 4936 | } |
686b73d4 | 4937 | #ifdef TRIE_STCLASS |
786e8c11 | 4938 | else if (PL_regkind[OP(first)] == TRIE && |
f8fc2ecf | 4939 | ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0) |
07be1b83 | 4940 | { |
786e8c11 | 4941 | regnode *trie_op; |
07be1b83 | 4942 | /* this can happen only on restudy */ |
786e8c11 | 4943 | if ( OP(first) == TRIE ) { |
c944940b | 4944 | struct regnode_1 *trieop = (struct regnode_1 *) |
446bd890 | 4945 | PerlMemShared_calloc(1, sizeof(struct regnode_1)); |
786e8c11 YO |
4946 | StructCopy(first,trieop,struct regnode_1); |
4947 | trie_op=(regnode *)trieop; | |
4948 | } else { | |
c944940b | 4949 | struct regnode_charclass *trieop = (struct regnode_charclass *) |
446bd890 | 4950 | PerlMemShared_calloc(1, sizeof(struct regnode_charclass)); |
786e8c11 YO |
4951 | StructCopy(first,trieop,struct regnode_charclass); |
4952 | trie_op=(regnode *)trieop; | |
4953 | } | |
1de06328 | 4954 | OP(trie_op)+=2; |
786e8c11 | 4955 | make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0); |
f8fc2ecf | 4956 | ri->regstclass = trie_op; |
07be1b83 | 4957 | } |
686b73d4 | 4958 | #endif |
e52fc539 | 4959 | else if (REGNODE_SIMPLE(OP(first))) |
f8fc2ecf | 4960 | ri->regstclass = first; |
3dab1dad YO |
4961 | else if (PL_regkind[OP(first)] == BOUND || |
4962 | PL_regkind[OP(first)] == NBOUND) | |
f8fc2ecf | 4963 | ri->regstclass = first; |
3dab1dad | 4964 | else if (PL_regkind[OP(first)] == BOL) { |
bbe252da YO |
4965 | r->extflags |= (OP(first) == MBOL |
4966 | ? RXf_ANCH_MBOL | |
cad2e5aa | 4967 | : (OP(first) == SBOL |
bbe252da YO |
4968 | ? RXf_ANCH_SBOL |
4969 | : RXf_ANCH_BOL)); | |
a0d0e21e | 4970 | first = NEXTOPER(first); |
774d564b | 4971 | goto again; |
4972 | } | |
4973 | else if (OP(first) == GPOS) { | |
bbe252da | 4974 | r->extflags |= RXf_ANCH_GPOS; |
774d564b | 4975 | first = NEXTOPER(first); |
4976 | goto again; | |
a0d0e21e | 4977 | } |
cf2a2b69 YO |
4978 | else if ((!sawopen || !RExC_sawback) && |
4979 | (OP(first) == STAR && | |
3dab1dad | 4980 | PL_regkind[OP(NEXTOPER(first))] == REG_ANY) && |
bbe252da | 4981 | !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL)) |
a0d0e21e LW |
4982 | { |
4983 | /* turn .* into ^.* with an implied $*=1 */ | |
1df70142 AL |
4984 | const int type = |
4985 | (OP(NEXTOPER(first)) == REG_ANY) | |
bbe252da YO |
4986 | ? RXf_ANCH_MBOL |
4987 | : RXf_ANCH_SBOL; | |
4988 | r->extflags |= type; | |
4989 | r->intflags |= PREGf_IMPLICIT; | |
a0d0e21e | 4990 | first = NEXTOPER(first); |
774d564b | 4991 | goto again; |
a0d0e21e | 4992 | } |
e7f38d0f | 4993 | if (sawplus && !sawlookahead && (!sawopen || !RExC_sawback) |
830247a4 | 4994 | && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */ |
cad2e5aa | 4995 | /* x+ must match at the 1st pos of run of x's */ |
bbe252da | 4996 | r->intflags |= PREGf_SKIP; |
a0d0e21e | 4997 | |
c277df42 | 4998 | /* Scan is after the zeroth branch, first is atomic matcher. */ |
be8e71aa | 4999 | #ifdef TRIE_STUDY_OPT |
81714fb9 | 5000 | DEBUG_PARSE_r( |
be8e71aa YO |
5001 | if (!restudied) |
5002 | PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n", | |
5003 | (IV)(first - scan + 1)) | |
5004 | ); | |
5005 | #else | |
81714fb9 | 5006 | DEBUG_PARSE_r( |
be8e71aa YO |
5007 | PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n", |
5008 | (IV)(first - scan + 1)) | |
5009 | ); | |
5010 | #endif | |
5011 | ||
5012 | ||
a0d0e21e LW |
5013 | /* |
5014 | * If there's something expensive in the r.e., find the | |
5015 | * longest literal string that must appear and make it the | |
5016 | * regmust. Resolve ties in favor of later strings, since | |
5017 | * the regstart check works with the beginning of the r.e. | |
5018 | * and avoiding duplication strengthens checking. Not a | |
5019 | * strong reason, but sufficient in the absence of others. | |
5020 | * [Now we resolve ties in favor of the earlier string if | |
c277df42 | 5021 | * it happens that c_offset_min has been invalidated, since the |
a0d0e21e LW |
5022 | * earlier string may buy us something the later one won't.] |
5023 | */ | |
686b73d4 | 5024 | |
396482e1 GA |
5025 | data.longest_fixed = newSVpvs(""); |
5026 | data.longest_float = newSVpvs(""); | |
5027 | data.last_found = newSVpvs(""); | |
c277df42 IZ |
5028 | data.longest = &(data.longest_fixed); |
5029 | first = scan; | |
f8fc2ecf | 5030 | if (!ri->regstclass) { |
e755fd73 | 5031 | cl_init(pRExC_state, &ch_class); |
653099ff GS |
5032 | data.start_class = &ch_class; |
5033 | stclass_flag = SCF_DO_STCLASS_AND; | |
5034 | } else /* XXXX Check for BOUND? */ | |
5035 | stclass_flag = 0; | |
cb434fcc | 5036 | data.last_closep = &last_close; |
de8c5301 | 5037 | |
1de06328 | 5038 | minlen = study_chunk(pRExC_state, &first, &minlen, &fake, scan + RExC_size, /* Up to end */ |
40d049e4 YO |
5039 | &data, -1, NULL, NULL, |
5040 | SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0); | |
07be1b83 | 5041 | |
686b73d4 | 5042 | |
786e8c11 YO |
5043 | CHECK_RESTUDY_GOTO; |
5044 | ||
5045 | ||
830247a4 | 5046 | if ( RExC_npar == 1 && data.longest == &(data.longest_fixed) |
b81d288d | 5047 | && data.last_start_min == 0 && data.last_end > 0 |
830247a4 | 5048 | && !RExC_seen_zerolen |
2bf803e2 | 5049 | && !(RExC_seen & REG_SEEN_VERBARG) |
bbe252da YO |
5050 | && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS))) |
5051 | r->extflags |= RXf_CHECK_ALL; | |
304ee84b | 5052 | scan_commit(pRExC_state, &data,&minlen,0); |
c277df42 IZ |
5053 | SvREFCNT_dec(data.last_found); |
5054 | ||
1de06328 YO |
5055 | /* Note that code very similar to this but for anchored string |
5056 | follows immediately below, changes may need to be made to both. | |
5057 | Be careful. | |
5058 | */ | |
a0ed51b3 | 5059 | longest_float_length = CHR_SVLEN(data.longest_float); |
c5254dd6 | 5060 | if (longest_float_length |
c277df42 IZ |
5061 | || (data.flags & SF_FL_BEFORE_EOL |
5062 | && (!(data.flags & SF_FL_BEFORE_MEOL) | |
bbe252da | 5063 | || (RExC_flags & RXf_PMf_MULTILINE)))) |
1de06328 | 5064 | { |
1182767e | 5065 | I32 t,ml; |
cf93c79d | 5066 | |
1de06328 | 5067 | if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */ |
aca2d497 IZ |
5068 | && data.offset_fixed == data.offset_float_min |
5069 | && SvCUR(data.longest_fixed) == SvCUR(data.longest_float)) | |
5070 | goto remove_float; /* As in (a)+. */ | |
5071 | ||
1de06328 YO |
5072 | /* copy the information about the longest float from the reg_scan_data |
5073 | over to the program. */ | |
33b8afdf JH |
5074 | if (SvUTF8(data.longest_float)) { |
5075 | r->float_utf8 = data.longest_float; | |
c445ea15 | 5076 | r->float_substr = NULL; |
33b8afdf JH |
5077 | } else { |
5078 | r->float_substr = data.longest_float; | |
c445ea15 | 5079 | r->float_utf8 = NULL; |
33b8afdf | 5080 | } |
1de06328 YO |
5081 | /* float_end_shift is how many chars that must be matched that |
5082 | follow this item. We calculate it ahead of time as once the | |
5083 | lookbehind offset is added in we lose the ability to correctly | |
5084 | calculate it.*/ | |
5085 | ml = data.minlen_float ? *(data.minlen_float) | |
1182767e | 5086 | : (I32)longest_float_length; |
1de06328 YO |
5087 | r->float_end_shift = ml - data.offset_float_min |
5088 | - longest_float_length + (SvTAIL(data.longest_float) != 0) | |
5089 | + data.lookbehind_float; | |
5090 | r->float_min_offset = data.offset_float_min - data.lookbehind_float; | |
c277df42 | 5091 | r->float_max_offset = data.offset_float_max; |
1182767e | 5092 | if (data.offset_float_max < I32_MAX) /* Don't offset infinity */ |
1de06328 YO |
5093 | r->float_max_offset -= data.lookbehind_float; |
5094 | ||
cf93c79d IZ |
5095 | t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */ |
5096 | && (!(data.flags & SF_FL_BEFORE_MEOL) | |
bbe252da | 5097 | || (RExC_flags & RXf_PMf_MULTILINE))); |
33b8afdf | 5098 | fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0); |
a0ed51b3 LW |
5099 | } |
5100 | else { | |
aca2d497 | 5101 | remove_float: |
c445ea15 | 5102 | r->float_substr = r->float_utf8 = NULL; |
c277df42 | 5103 | SvREFCNT_dec(data.longest_float); |
c5254dd6 | 5104 | longest_float_length = 0; |
a0d0e21e | 5105 | } |
c277df42 | 5106 | |
1de06328 YO |
5107 | /* Note that code very similar to this but for floating string |
5108 | is immediately above, changes may need to be made to both. | |
5109 | Be careful. | |
5110 | */ | |
a0ed51b3 | 5111 | longest_fixed_length = CHR_SVLEN(data.longest_fixed); |
c5254dd6 | 5112 | if (longest_fixed_length |
c277df42 IZ |
5113 | || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */ |
5114 | && (!(data.flags & SF_FIX_BEFORE_MEOL) | |
bbe252da | 5115 | || (RExC_flags & RXf_PMf_MULTILINE)))) |
1de06328 | 5116 | { |
1182767e | 5117 | I32 t,ml; |
cf93c79d | 5118 | |
1de06328 YO |
5119 | /* copy the information about the longest fixed |
5120 | from the reg_scan_data over to the program. */ | |
33b8afdf JH |
5121 | if (SvUTF8(data.longest_fixed)) { |
5122 | r->anchored_utf8 = data.longest_fixed; | |
c445ea15 | 5123 | r->anchored_substr = NULL; |
33b8afdf JH |
5124 | } else { |
5125 | r->anchored_substr = data.longest_fixed; | |
c445ea15 | 5126 | r->anchored_utf8 = NULL; |
33b8afdf | 5127 | } |
1de06328 YO |
5128 | /* fixed_end_shift is how many chars that must be matched that |
5129 | follow this item. We calculate it ahead of time as once the | |
5130 | lookbehind offset is added in we lose the ability to correctly | |
5131 | calculate it.*/ | |
5132 | ml = data.minlen_fixed ? *(data.minlen_fixed) | |
1182767e | 5133 | : (I32)longest_fixed_length; |
1de06328 YO |
5134 | r->anchored_end_shift = ml - data.offset_fixed |
5135 | - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0) | |
5136 | + data.lookbehind_fixed; | |
5137 | r->anchored_offset = data.offset_fixed - data.lookbehind_fixed; | |
5138 | ||
cf93c79d IZ |
5139 | t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */ |
5140 | && (!(data.flags & SF_FIX_BEFORE_MEOL) | |
bbe252da | 5141 | || (RExC_flags & RXf_PMf_MULTILINE))); |
33b8afdf | 5142 | fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0); |
a0ed51b3 LW |
5143 | } |
5144 | else { | |
c445ea15 | 5145 | r->anchored_substr = r->anchored_utf8 = NULL; |
c277df42 | 5146 | SvREFCNT_dec(data.longest_fixed); |
c5254dd6 | 5147 | longest_fixed_length = 0; |
a0d0e21e | 5148 | } |
f8fc2ecf YO |
5149 | if (ri->regstclass |
5150 | && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY)) | |
5151 | ri->regstclass = NULL; | |
f4244008 | 5152 | |
33b8afdf JH |
5153 | if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset) |
5154 | && stclass_flag | |
653099ff | 5155 | && !(data.start_class->flags & ANYOF_EOS) |
eb160463 GS |
5156 | && !cl_is_anything(data.start_class)) |
5157 | { | |
2eccd3b2 | 5158 | const U32 n = add_data(pRExC_state, 1, "f"); |
c613755a | 5159 | data.start_class->flags |= ANYOF_IS_SYNTHETIC; |
653099ff | 5160 | |
f8fc2ecf | 5161 | Newx(RExC_rxi->data->data[n], 1, |
653099ff GS |
5162 | struct regnode_charclass_class); |
5163 | StructCopy(data.start_class, | |
f8fc2ecf | 5164 | (struct regnode_charclass_class*)RExC_rxi->data->data[n], |
653099ff | 5165 | struct regnode_charclass_class); |
f8fc2ecf | 5166 | ri->regstclass = (regnode*)RExC_rxi->data->data[n]; |
bbe252da | 5167 | r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
a3621e74 | 5168 | DEBUG_COMPILE_r({ SV *sv = sv_newmortal(); |
32fc9b6a | 5169 | regprop(r, sv, (regnode*)data.start_class); |
9c5ffd7c | 5170 | PerlIO_printf(Perl_debug_log, |
a0288114 | 5171 | "synthetic stclass \"%s\".\n", |
3f7c398e | 5172 | SvPVX_const(sv));}); |
653099ff | 5173 | } |
c277df42 IZ |
5174 | |
5175 | /* A temporary algorithm prefers floated substr to fixed one to dig more info. */ | |
c5254dd6 | 5176 | if (longest_fixed_length > longest_float_length) { |
1de06328 | 5177 | r->check_end_shift = r->anchored_end_shift; |
c277df42 | 5178 | r->check_substr = r->anchored_substr; |
33b8afdf | 5179 | r->check_utf8 = r->anchored_utf8; |
c277df42 | 5180 | r->check_offset_min = r->check_offset_max = r->anchored_offset; |
bbe252da YO |
5181 | if (r->extflags & RXf_ANCH_SINGLE) |
5182 | r->extflags |= RXf_NOSCAN; | |
a0ed51b3 LW |
5183 | } |
5184 | else { | |
1de06328 | 5185 | r->check_end_shift = r->float_end_shift; |
c277df42 | 5186 | r->check_substr = r->float_substr; |
33b8afdf | 5187 | r->check_utf8 = r->float_utf8; |
1de06328 YO |
5188 | r->check_offset_min = r->float_min_offset; |
5189 | r->check_offset_max = r->float_max_offset; | |
a0d0e21e | 5190 | } |
30382c73 IZ |
5191 | /* XXXX Currently intuiting is not compatible with ANCH_GPOS. |
5192 | This should be changed ASAP! */ | |
bbe252da YO |
5193 | if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) { |
5194 | r->extflags |= RXf_USE_INTUIT; | |
33b8afdf | 5195 | if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8)) |
bbe252da | 5196 | r->extflags |= RXf_INTUIT_TAIL; |
cad2e5aa | 5197 | } |
1de06328 YO |
5198 | /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere) |
5199 | if ( (STRLEN)minlen < longest_float_length ) | |
5200 | minlen= longest_float_length; | |
5201 | if ( (STRLEN)minlen < longest_fixed_length ) | |
5202 | minlen= longest_fixed_length; | |
5203 | */ | |
a0ed51b3 LW |
5204 | } |
5205 | else { | |
c277df42 IZ |
5206 | /* Several toplevels. Best we can is to set minlen. */ |
5207 | I32 fake; | |
653099ff | 5208 | struct regnode_charclass_class ch_class; |
cb434fcc | 5209 | I32 last_close = 0; |
686b73d4 | 5210 | |
5d458dd8 | 5211 | DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n")); |
07be1b83 | 5212 | |
f8fc2ecf | 5213 | scan = ri->program + 1; |
e755fd73 | 5214 | cl_init(pRExC_state, &ch_class); |
653099ff | 5215 | data.start_class = &ch_class; |
cb434fcc | 5216 | data.last_closep = &last_close; |
07be1b83 | 5217 | |
de8c5301 | 5218 | |
1de06328 | 5219 | minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size, |
40d049e4 | 5220 | &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0); |
de8c5301 | 5221 | |
786e8c11 | 5222 | CHECK_RESTUDY_GOTO; |
07be1b83 | 5223 | |
33b8afdf | 5224 | r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8 |
c445ea15 | 5225 | = r->float_substr = r->float_utf8 = NULL; |
f4244008 | 5226 | |
653099ff | 5227 | if (!(data.start_class->flags & ANYOF_EOS) |
eb160463 GS |
5228 | && !cl_is_anything(data.start_class)) |
5229 | { | |
2eccd3b2 | 5230 | const U32 n = add_data(pRExC_state, 1, "f"); |
c613755a | 5231 | data.start_class->flags |= ANYOF_IS_SYNTHETIC; |
653099ff | 5232 | |
f8fc2ecf | 5233 | Newx(RExC_rxi->data->data[n], 1, |
653099ff GS |
5234 | struct regnode_charclass_class); |
5235 | StructCopy(data.start_class, | |
f8fc2ecf | 5236 | (struct regnode_charclass_class*)RExC_rxi->data->data[n], |
653099ff | 5237 | struct regnode_charclass_class); |
f8fc2ecf | 5238 | ri->regstclass = (regnode*)RExC_rxi->data->data[n]; |
bbe252da | 5239 | r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */ |
a3621e74 | 5240 | DEBUG_COMPILE_r({ SV* sv = sv_newmortal(); |
32fc9b6a | 5241 | regprop(r, sv, (regnode*)data.start_class); |
9c5ffd7c | 5242 | PerlIO_printf(Perl_debug_log, |
a0288114 | 5243 | "synthetic stclass \"%s\".\n", |
3f7c398e | 5244 | SvPVX_const(sv));}); |
653099ff | 5245 | } |
a0d0e21e LW |
5246 | } |
5247 | ||
1de06328 YO |
5248 | /* Guard against an embedded (?=) or (?<=) with a longer minlen than |
5249 | the "real" pattern. */ | |
cf9788e3 RGS |
5250 | DEBUG_OPTIMISE_r({ |
5251 | PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n", | |
70685ca0 | 5252 | (IV)minlen, (IV)r->minlen); |
cf9788e3 | 5253 | }); |
de8c5301 | 5254 | r->minlenret = minlen; |
1de06328 YO |
5255 | if (r->minlen < minlen) |
5256 | r->minlen = minlen; | |
5257 | ||
b81d288d | 5258 | if (RExC_seen & REG_SEEN_GPOS) |
bbe252da | 5259 | r->extflags |= RXf_GPOS_SEEN; |
830247a4 | 5260 | if (RExC_seen & REG_SEEN_LOOKBEHIND) |
bbe252da | 5261 | r->extflags |= RXf_LOOKBEHIND_SEEN; |
830247a4 | 5262 | if (RExC_seen & REG_SEEN_EVAL) |
bbe252da | 5263 | r->extflags |= RXf_EVAL_SEEN; |
f33976b4 | 5264 | if (RExC_seen & REG_SEEN_CANY) |
bbe252da | 5265 | r->extflags |= RXf_CANY_SEEN; |
e2e6a0f1 | 5266 | if (RExC_seen & REG_SEEN_VERBARG) |
bbe252da | 5267 | r->intflags |= PREGf_VERBARG_SEEN; |
5d458dd8 | 5268 | if (RExC_seen & REG_SEEN_CUTGROUP) |
bbe252da | 5269 | r->intflags |= PREGf_CUTGROUP_SEEN; |
81714fb9 | 5270 | if (RExC_paren_names) |
85fbaab2 | 5271 | RXp_PAREN_NAMES(r) = MUTABLE_HV(SvREFCNT_inc(RExC_paren_names)); |
81714fb9 | 5272 | else |
5daac39c | 5273 | RXp_PAREN_NAMES(r) = NULL; |
0ac6acae | 5274 | |
7bd1e614 | 5275 | #ifdef STUPID_PATTERN_CHECKS |
5509d87a | 5276 | if (RX_PRELEN(rx) == 0) |
640f820d | 5277 | r->extflags |= RXf_NULL; |
5509d87a | 5278 | if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ') |
0ac6acae AB |
5279 | /* XXX: this should happen BEFORE we compile */ |
5280 | r->extflags |= (RXf_SKIPWHITE|RXf_WHITE); | |
5509d87a | 5281 | else if (RX_PRELEN(rx) == 3 && memEQ("\\s+", RX_PRECOMP(rx), 3)) |
0ac6acae | 5282 | r->extflags |= RXf_WHITE; |
5509d87a | 5283 | else if (RX_PRELEN(rx) == 1 && RXp_PRECOMP(rx)[0] == '^') |
e357fc67 | 5284 | r->extflags |= RXf_START_ONLY; |
f1b875a0 | 5285 | #else |
5509d87a | 5286 | if (r->extflags & RXf_SPLIT && RX_PRELEN(rx) == 1 && RX_PRECOMP(rx)[0] == ' ') |
7bd1e614 YO |
5287 | /* XXX: this should happen BEFORE we compile */ |
5288 | r->extflags |= (RXf_SKIPWHITE|RXf_WHITE); | |
5289 | else { | |
5290 | regnode *first = ri->program + 1; | |
39aa8307 | 5291 | U8 fop = OP(first); |
f6d9469c DM |
5292 | |
5293 | if (PL_regkind[fop] == NOTHING && OP(NEXTOPER(first)) == END) | |
640f820d | 5294 | r->extflags |= RXf_NULL; |
f6d9469c | 5295 | else if (PL_regkind[fop] == BOL && OP(NEXTOPER(first)) == END) |
7bd1e614 | 5296 | r->extflags |= RXf_START_ONLY; |
f6d9469c DM |
5297 | else if (fop == PLUS && OP(NEXTOPER(first)) == SPACE |
5298 | && OP(regnext(first)) == END) | |
7bd1e614 YO |
5299 | r->extflags |= RXf_WHITE; |
5300 | } | |
f1b875a0 | 5301 | #endif |
1f1031fe YO |
5302 | #ifdef DEBUGGING |
5303 | if (RExC_paren_names) { | |
af534a04 | 5304 | ri->name_list_idx = add_data( pRExC_state, 1, "a" ); |
1f1031fe YO |
5305 | ri->data->data[ri->name_list_idx] = (void*)SvREFCNT_inc(RExC_paren_name_list); |
5306 | } else | |
1f1031fe | 5307 | #endif |
cde0cee5 | 5308 | ri->name_list_idx = 0; |
1f1031fe | 5309 | |
40d049e4 YO |
5310 | if (RExC_recurse_count) { |
5311 | for ( ; RExC_recurse_count ; RExC_recurse_count-- ) { | |
5312 | const regnode *scan = RExC_recurse[RExC_recurse_count-1]; | |
5313 | ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan ); | |
5314 | } | |
5315 | } | |
f0ab9afb | 5316 | Newxz(r->offs, RExC_npar, regexp_paren_pair); |
c74340f9 YO |
5317 | /* assume we don't need to swap parens around before we match */ |
5318 | ||
be8e71aa YO |
5319 | DEBUG_DUMP_r({ |
5320 | PerlIO_printf(Perl_debug_log,"Final program:\n"); | |
3dab1dad YO |
5321 | regdump(r); |
5322 | }); | |
7122b237 YO |
5323 | #ifdef RE_TRACK_PATTERN_OFFSETS |
5324 | DEBUG_OFFSETS_r(if (ri->u.offsets) { | |
5325 | const U32 len = ri->u.offsets[0]; | |
8e9a8a48 YO |
5326 | U32 i; |
5327 | GET_RE_DEBUG_FLAGS_DECL; | |
7122b237 | 5328 | PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->u.offsets[0]); |
8e9a8a48 | 5329 | for (i = 1; i <= len; i++) { |
7122b237 | 5330 | if (ri->u.offsets[i*2-1] || ri->u.offsets[i*2]) |
8e9a8a48 | 5331 | PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ", |
7122b237 | 5332 | (UV)i, (UV)ri->u.offsets[i*2-1], (UV)ri->u.offsets[i*2]); |
8e9a8a48 YO |
5333 | } |
5334 | PerlIO_printf(Perl_debug_log, "\n"); | |
5335 | }); | |
7122b237 | 5336 | #endif |
288b8c02 | 5337 | return rx; |
a687059c LW |
5338 | } |
5339 | ||
f9f4320a | 5340 | #undef RE_ENGINE_PTR |
3dab1dad | 5341 | |
93b32b6d | 5342 | |
81714fb9 | 5343 | SV* |
192b9cd1 AB |
5344 | Perl_reg_named_buff(pTHX_ REGEXP * const rx, SV * const key, SV * const value, |
5345 | const U32 flags) | |
5346 | { | |
7918f24d NC |
5347 | PERL_ARGS_ASSERT_REG_NAMED_BUFF; |
5348 | ||
192b9cd1 AB |
5349 | PERL_UNUSED_ARG(value); |
5350 | ||
f1b875a0 | 5351 | if (flags & RXapif_FETCH) { |
192b9cd1 | 5352 | return reg_named_buff_fetch(rx, key, flags); |
f1b875a0 | 5353 | } else if (flags & (RXapif_STORE | RXapif_DELETE | RXapif_CLEAR)) { |
6ad8f254 | 5354 | Perl_croak_no_modify(aTHX); |
192b9cd1 | 5355 | return NULL; |
f1b875a0 | 5356 | } else if (flags & RXapif_EXISTS) { |
192b9cd1 AB |
5357 | return reg_named_buff_exists(rx, key, flags) |
5358 | ? &PL_sv_yes | |
5359 | : &PL_sv_no; | |
f1b875a0 | 5360 | } else if (flags & RXapif_REGNAMES) { |
192b9cd1 | 5361 | return reg_named_buff_all(rx, flags); |
f1b875a0 | 5362 | } else if (flags & (RXapif_SCALAR | RXapif_REGNAMES_COUNT)) { |
192b9cd1 AB |
5363 | return reg_named_buff_scalar(rx, flags); |
5364 | } else { | |
5365 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff", (int)flags); | |
5366 | return NULL; | |
5367 | } | |
5368 | } | |
5369 | ||
5370 | SV* | |
5371 | Perl_reg_named_buff_iter(pTHX_ REGEXP * const rx, const SV * const lastkey, | |
5372 | const U32 flags) | |
5373 | { | |
7918f24d | 5374 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ITER; |
192b9cd1 AB |
5375 | PERL_UNUSED_ARG(lastkey); |
5376 | ||
f1b875a0 | 5377 | if (flags & RXapif_FIRSTKEY) |
192b9cd1 | 5378 | return reg_named_buff_firstkey(rx, flags); |
f1b875a0 | 5379 | else if (flags & RXapif_NEXTKEY) |
192b9cd1 AB |
5380 | return reg_named_buff_nextkey(rx, flags); |
5381 | else { | |
5382 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_iter", (int)flags); | |
5383 | return NULL; | |
5384 | } | |
5385 | } | |
5386 | ||
5387 | SV* | |
288b8c02 NC |
5388 | Perl_reg_named_buff_fetch(pTHX_ REGEXP * const r, SV * const namesv, |
5389 | const U32 flags) | |
81714fb9 | 5390 | { |
44a2ac75 YO |
5391 | AV *retarray = NULL; |
5392 | SV *ret; | |
288b8c02 | 5393 | struct regexp *const rx = (struct regexp *)SvANY(r); |
7918f24d NC |
5394 | |
5395 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FETCH; | |
5396 | ||
f1b875a0 | 5397 | if (flags & RXapif_ALL) |
44a2ac75 | 5398 | retarray=newAV(); |
93b32b6d | 5399 | |
5daac39c NC |
5400 | if (rx && RXp_PAREN_NAMES(rx)) { |
5401 | HE *he_str = hv_fetch_ent( RXp_PAREN_NAMES(rx), namesv, 0, 0 ); | |
93b32b6d YO |
5402 | if (he_str) { |
5403 | IV i; | |
5404 | SV* sv_dat=HeVAL(he_str); | |
5405 | I32 *nums=(I32*)SvPVX(sv_dat); | |
5406 | for ( i=0; i<SvIVX(sv_dat); i++ ) { | |
192b9cd1 AB |
5407 | if ((I32)(rx->nparens) >= nums[i] |
5408 | && rx->offs[nums[i]].start != -1 | |
5409 | && rx->offs[nums[i]].end != -1) | |
93b32b6d | 5410 | { |
49d7dfbc | 5411 | ret = newSVpvs(""); |
288b8c02 | 5412 | CALLREG_NUMBUF_FETCH(r,nums[i],ret); |
93b32b6d YO |
5413 | if (!retarray) |
5414 | return ret; | |
5415 | } else { | |
7402016d AB |
5416 | if (retarray) |
5417 | ret = newSVsv(&PL_sv_undef); | |
93b32b6d | 5418 | } |
ec83ea38 | 5419 | if (retarray) |
93b32b6d | 5420 | av_push(retarray, ret); |
81714fb9 | 5421 | } |
93b32b6d | 5422 | if (retarray) |
ad64d0ec | 5423 | return newRV_noinc(MUTABLE_SV(retarray)); |
192b9cd1 AB |
5424 | } |
5425 | } | |
5426 | return NULL; | |
5427 | } | |
5428 | ||
5429 | bool | |
288b8c02 | 5430 | Perl_reg_named_buff_exists(pTHX_ REGEXP * const r, SV * const key, |
192b9cd1 AB |
5431 | const U32 flags) |
5432 | { | |
288b8c02 | 5433 | struct regexp *const rx = (struct regexp *)SvANY(r); |
7918f24d NC |
5434 | |
5435 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_EXISTS; | |
5436 | ||
5daac39c | 5437 | if (rx && RXp_PAREN_NAMES(rx)) { |
f1b875a0 | 5438 | if (flags & RXapif_ALL) { |
5daac39c | 5439 | return hv_exists_ent(RXp_PAREN_NAMES(rx), key, 0); |
192b9cd1 | 5440 | } else { |
288b8c02 | 5441 | SV *sv = CALLREG_NAMED_BUFF_FETCH(r, key, flags); |
6499cc01 RGS |
5442 | if (sv) { |
5443 | SvREFCNT_dec(sv); | |
192b9cd1 AB |
5444 | return TRUE; |
5445 | } else { | |
5446 | return FALSE; | |
5447 | } | |
5448 | } | |
5449 | } else { | |
5450 | return FALSE; | |
5451 | } | |
5452 | } | |
5453 | ||
5454 | SV* | |
288b8c02 | 5455 | Perl_reg_named_buff_firstkey(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 | 5456 | { |
288b8c02 | 5457 | struct regexp *const rx = (struct regexp *)SvANY(r); |
7918f24d NC |
5458 | |
5459 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_FIRSTKEY; | |
5460 | ||
5daac39c NC |
5461 | if ( rx && RXp_PAREN_NAMES(rx) ) { |
5462 | (void)hv_iterinit(RXp_PAREN_NAMES(rx)); | |
192b9cd1 | 5463 | |
288b8c02 | 5464 | return CALLREG_NAMED_BUFF_NEXTKEY(r, NULL, flags & ~RXapif_FIRSTKEY); |
1e1d4b91 JJ |
5465 | } else { |
5466 | return FALSE; | |
5467 | } | |
192b9cd1 AB |
5468 | } |
5469 | ||
5470 | SV* | |
288b8c02 | 5471 | Perl_reg_named_buff_nextkey(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 | 5472 | { |
288b8c02 | 5473 | struct regexp *const rx = (struct regexp *)SvANY(r); |
250257bb | 5474 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
5475 | |
5476 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_NEXTKEY; | |
5477 | ||
5daac39c NC |
5478 | if (rx && RXp_PAREN_NAMES(rx)) { |
5479 | HV *hv = RXp_PAREN_NAMES(rx); | |
192b9cd1 AB |
5480 | HE *temphe; |
5481 | while ( (temphe = hv_iternext_flags(hv,0)) ) { | |
5482 | IV i; | |
5483 | IV parno = 0; | |
5484 | SV* sv_dat = HeVAL(temphe); | |
5485 | I32 *nums = (I32*)SvPVX(sv_dat); | |
5486 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { | |
250257bb | 5487 | if ((I32)(rx->lastparen) >= nums[i] && |
192b9cd1 AB |
5488 | rx->offs[nums[i]].start != -1 && |
5489 | rx->offs[nums[i]].end != -1) | |
5490 | { | |
5491 | parno = nums[i]; | |
5492 | break; | |
5493 | } | |
5494 | } | |
f1b875a0 | 5495 | if (parno || flags & RXapif_ALL) { |
a663657d | 5496 | return newSVhek(HeKEY_hek(temphe)); |
192b9cd1 | 5497 | } |
81714fb9 YO |
5498 | } |
5499 | } | |
44a2ac75 YO |
5500 | return NULL; |
5501 | } | |
5502 | ||
192b9cd1 | 5503 | SV* |
288b8c02 | 5504 | Perl_reg_named_buff_scalar(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 AB |
5505 | { |
5506 | SV *ret; | |
5507 | AV *av; | |
5508 | I32 length; | |
288b8c02 | 5509 | struct regexp *const rx = (struct regexp *)SvANY(r); |
192b9cd1 | 5510 | |
7918f24d NC |
5511 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_SCALAR; |
5512 | ||
5daac39c | 5513 | if (rx && RXp_PAREN_NAMES(rx)) { |
f1b875a0 | 5514 | if (flags & (RXapif_ALL | RXapif_REGNAMES_COUNT)) { |
5daac39c | 5515 | return newSViv(HvTOTALKEYS(RXp_PAREN_NAMES(rx))); |
f1b875a0 | 5516 | } else if (flags & RXapif_ONE) { |
288b8c02 | 5517 | ret = CALLREG_NAMED_BUFF_ALL(r, (flags | RXapif_REGNAMES)); |
502c6561 | 5518 | av = MUTABLE_AV(SvRV(ret)); |
192b9cd1 | 5519 | length = av_len(av); |
ec83ea38 | 5520 | SvREFCNT_dec(ret); |
192b9cd1 AB |
5521 | return newSViv(length + 1); |
5522 | } else { | |
5523 | Perl_croak(aTHX_ "panic: Unknown flags %d in named_buff_scalar", (int)flags); | |
5524 | return NULL; | |
5525 | } | |
5526 | } | |
5527 | return &PL_sv_undef; | |
5528 | } | |
5529 | ||
5530 | SV* | |
288b8c02 | 5531 | Perl_reg_named_buff_all(pTHX_ REGEXP * const r, const U32 flags) |
192b9cd1 | 5532 | { |
288b8c02 | 5533 | struct regexp *const rx = (struct regexp *)SvANY(r); |
192b9cd1 AB |
5534 | AV *av = newAV(); |
5535 | ||
7918f24d NC |
5536 | PERL_ARGS_ASSERT_REG_NAMED_BUFF_ALL; |
5537 | ||
5daac39c NC |
5538 | if (rx && RXp_PAREN_NAMES(rx)) { |
5539 | HV *hv= RXp_PAREN_NAMES(rx); | |
192b9cd1 AB |
5540 | HE *temphe; |
5541 | (void)hv_iterinit(hv); | |
5542 | while ( (temphe = hv_iternext_flags(hv,0)) ) { | |
5543 | IV i; | |
5544 | IV parno = 0; | |
5545 | SV* sv_dat = HeVAL(temphe); | |
5546 | I32 *nums = (I32*)SvPVX(sv_dat); | |
5547 | for ( i = 0; i < SvIVX(sv_dat); i++ ) { | |
250257bb | 5548 | if ((I32)(rx->lastparen) >= nums[i] && |
192b9cd1 AB |
5549 | rx->offs[nums[i]].start != -1 && |
5550 | rx->offs[nums[i]].end != -1) | |
5551 | { | |
5552 | parno = nums[i]; | |
5553 | break; | |
5554 | } | |
5555 | } | |
f1b875a0 | 5556 | if (parno || flags & RXapif_ALL) { |
a663657d | 5557 | av_push(av, newSVhek(HeKEY_hek(temphe))); |
192b9cd1 AB |
5558 | } |
5559 | } | |
5560 | } | |
5561 | ||
ad64d0ec | 5562 | return newRV_noinc(MUTABLE_SV(av)); |
192b9cd1 AB |
5563 | } |
5564 | ||
49d7dfbc | 5565 | void |
288b8c02 NC |
5566 | Perl_reg_numbered_buff_fetch(pTHX_ REGEXP * const r, const I32 paren, |
5567 | SV * const sv) | |
44a2ac75 | 5568 | { |
288b8c02 | 5569 | struct regexp *const rx = (struct regexp *)SvANY(r); |
44a2ac75 | 5570 | char *s = NULL; |
a9d504c3 | 5571 | I32 i = 0; |
44a2ac75 | 5572 | I32 s1, t1; |
7918f24d NC |
5573 | |
5574 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_FETCH; | |
44a2ac75 | 5575 | |
cde0cee5 YO |
5576 | if (!rx->subbeg) { |
5577 | sv_setsv(sv,&PL_sv_undef); | |
49d7dfbc | 5578 | return; |
cde0cee5 YO |
5579 | } |
5580 | else | |
f1b875a0 | 5581 | if (paren == RX_BUFF_IDX_PREMATCH && rx->offs[0].start != -1) { |
44a2ac75 | 5582 | /* $` */ |
f0ab9afb | 5583 | i = rx->offs[0].start; |
cde0cee5 | 5584 | s = rx->subbeg; |
44a2ac75 YO |
5585 | } |
5586 | else | |
f1b875a0 | 5587 | if (paren == RX_BUFF_IDX_POSTMATCH && rx->offs[0].end != -1) { |
44a2ac75 | 5588 | /* $' */ |
f0ab9afb NC |
5589 | s = rx->subbeg + rx->offs[0].end; |
5590 | i = rx->sublen - rx->offs[0].end; | |
44a2ac75 YO |
5591 | } |
5592 | else | |
5593 | if ( 0 <= paren && paren <= (I32)rx->nparens && | |
f0ab9afb NC |
5594 | (s1 = rx->offs[paren].start) != -1 && |
5595 | (t1 = rx->offs[paren].end) != -1) | |
44a2ac75 YO |
5596 | { |
5597 | /* $& $1 ... */ | |
5598 | i = t1 - s1; | |
5599 | s = rx->subbeg + s1; | |
cde0cee5 YO |
5600 | } else { |
5601 | sv_setsv(sv,&PL_sv_undef); | |
49d7dfbc | 5602 | return; |
cde0cee5 YO |
5603 | } |
5604 | assert(rx->sublen >= (s - rx->subbeg) + i ); | |
5605 | if (i >= 0) { | |
5606 | const int oldtainted = PL_tainted; | |
5607 | TAINT_NOT; | |
5608 | sv_setpvn(sv, s, i); | |
5609 | PL_tainted = oldtainted; | |
5610 | if ( (rx->extflags & RXf_CANY_SEEN) | |
07bc277f | 5611 | ? (RXp_MATCH_UTF8(rx) |
cde0cee5 | 5612 | && (!i || is_utf8_string((U8*)s, i))) |
07bc277f | 5613 | : (RXp_MATCH_UTF8(rx)) ) |
cde0cee5 YO |
5614 | { |
5615 | SvUTF8_on(sv); | |
5616 | } | |
5617 | else | |
5618 | SvUTF8_off(sv); | |
5619 | if (PL_tainting) { | |
07bc277f | 5620 | if (RXp_MATCH_TAINTED(rx)) { |
cde0cee5 YO |
5621 | if (SvTYPE(sv) >= SVt_PVMG) { |
5622 | MAGIC* const mg = SvMAGIC(sv); | |
5623 | MAGIC* mgt; | |
5624 | PL_tainted = 1; | |
5625 | SvMAGIC_set(sv, mg->mg_moremagic); | |
5626 | SvTAINT(sv); | |
5627 | if ((mgt = SvMAGIC(sv))) { | |
5628 | mg->mg_moremagic = mgt; | |
5629 | SvMAGIC_set(sv, mg); | |
44a2ac75 | 5630 | } |
cde0cee5 YO |
5631 | } else { |
5632 | PL_tainted = 1; | |
5633 | SvTAINT(sv); | |
5634 | } | |
5635 | } else | |
5636 | SvTAINTED_off(sv); | |
44a2ac75 | 5637 | } |
81714fb9 | 5638 | } else { |
44a2ac75 | 5639 | sv_setsv(sv,&PL_sv_undef); |
49d7dfbc | 5640 | return; |
81714fb9 YO |
5641 | } |
5642 | } | |
93b32b6d | 5643 | |
2fdbfb4d AB |
5644 | void |
5645 | Perl_reg_numbered_buff_store(pTHX_ REGEXP * const rx, const I32 paren, | |
5646 | SV const * const value) | |
5647 | { | |
7918f24d NC |
5648 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_STORE; |
5649 | ||
2fdbfb4d AB |
5650 | PERL_UNUSED_ARG(rx); |
5651 | PERL_UNUSED_ARG(paren); | |
5652 | PERL_UNUSED_ARG(value); | |
5653 | ||
5654 | if (!PL_localizing) | |
6ad8f254 | 5655 | Perl_croak_no_modify(aTHX); |
2fdbfb4d AB |
5656 | } |
5657 | ||
5658 | I32 | |
288b8c02 | 5659 | Perl_reg_numbered_buff_length(pTHX_ REGEXP * const r, const SV * const sv, |
2fdbfb4d AB |
5660 | const I32 paren) |
5661 | { | |
288b8c02 | 5662 | struct regexp *const rx = (struct regexp *)SvANY(r); |
2fdbfb4d AB |
5663 | I32 i; |
5664 | I32 s1, t1; | |
5665 | ||
7918f24d NC |
5666 | PERL_ARGS_ASSERT_REG_NUMBERED_BUFF_LENGTH; |
5667 | ||
2fdbfb4d AB |
5668 | /* Some of this code was originally in C<Perl_magic_len> in F<mg.c> */ |
5669 | switch (paren) { | |
192b9cd1 | 5670 | /* $` / ${^PREMATCH} */ |
f1b875a0 | 5671 | case RX_BUFF_IDX_PREMATCH: |
2fdbfb4d AB |
5672 | if (rx->offs[0].start != -1) { |
5673 | i = rx->offs[0].start; | |
5674 | if (i > 0) { | |
5675 | s1 = 0; | |
5676 | t1 = i; | |
5677 | goto getlen; | |
5678 | } | |
5679 | } | |
5680 | return 0; | |
192b9cd1 | 5681 | /* $' / ${^POSTMATCH} */ |
f1b875a0 | 5682 | case RX_BUFF_IDX_POSTMATCH: |
2fdbfb4d AB |
5683 | if (rx->offs[0].end != -1) { |
5684 | i = rx->sublen - rx->offs[0].end; | |
5685 | if (i > 0) { | |
5686 | s1 = rx->offs[0].end; | |
5687 | t1 = rx->sublen; | |
5688 | goto getlen; | |
5689 | } | |
5690 | } | |
5691 | return 0; | |
192b9cd1 AB |
5692 | /* $& / ${^MATCH}, $1, $2, ... */ |
5693 | default: | |
2fdbfb4d AB |
5694 | if (paren <= (I32)rx->nparens && |
5695 | (s1 = rx->offs[paren].start) != -1 && | |
5696 | (t1 = rx->offs[paren].end) != -1) | |
5697 | { | |
5698 | i = t1 - s1; | |
5699 | goto getlen; | |
5700 | } else { | |
5701 | if (ckWARN(WARN_UNINITIALIZED)) | |
ad64d0ec | 5702 | report_uninit((const SV *)sv); |
2fdbfb4d AB |
5703 | return 0; |
5704 | } | |
5705 | } | |
5706 | getlen: | |
07bc277f | 5707 | if (i > 0 && RXp_MATCH_UTF8(rx)) { |
2fdbfb4d AB |
5708 | const char * const s = rx->subbeg + s1; |
5709 | const U8 *ep; | |
5710 | STRLEN el; | |
5711 | ||
5712 | i = t1 - s1; | |
5713 | if (is_utf8_string_loclen((U8*)s, i, &ep, &el)) | |
5714 | i = el; | |
5715 | } | |
5716 | return i; | |
5717 | } | |
5718 | ||
fe578d7f | 5719 | SV* |
49d7dfbc | 5720 | Perl_reg_qr_package(pTHX_ REGEXP * const rx) |
fe578d7f | 5721 | { |
7918f24d | 5722 | PERL_ARGS_ASSERT_REG_QR_PACKAGE; |
fe578d7f | 5723 | PERL_UNUSED_ARG(rx); |
0fc92fc6 YO |
5724 | if (0) |
5725 | return NULL; | |
5726 | else | |
5727 | return newSVpvs("Regexp"); | |
fe578d7f | 5728 | } |
0a4db386 | 5729 | |
894be9b7 | 5730 | /* Scans the name of a named buffer from the pattern. |
0a4db386 YO |
5731 | * If flags is REG_RSN_RETURN_NULL returns null. |
5732 | * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name | |
5733 | * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding | |
5734 | * to the parsed name as looked up in the RExC_paren_names hash. | |
5735 | * If there is an error throws a vFAIL().. type exception. | |
894be9b7 | 5736 | */ |
0a4db386 YO |
5737 | |
5738 | #define REG_RSN_RETURN_NULL 0 | |
5739 | #define REG_RSN_RETURN_NAME 1 | |
5740 | #define REG_RSN_RETURN_DATA 2 | |
5741 | ||
894be9b7 | 5742 | STATIC SV* |
7918f24d NC |
5743 | S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) |
5744 | { | |
894be9b7 | 5745 | char *name_start = RExC_parse; |
1f1031fe | 5746 | |
7918f24d NC |
5747 | PERL_ARGS_ASSERT_REG_SCAN_NAME; |
5748 | ||
1f1031fe YO |
5749 | if (isIDFIRST_lazy_if(RExC_parse, UTF)) { |
5750 | /* skip IDFIRST by using do...while */ | |
5751 | if (UTF) | |
5752 | do { | |
5753 | RExC_parse += UTF8SKIP(RExC_parse); | |
5754 | } while (isALNUM_utf8((U8*)RExC_parse)); | |
5755 | else | |
5756 | do { | |
5757 | RExC_parse++; | |
5758 | } while (isALNUM(*RExC_parse)); | |
894be9b7 | 5759 | } |
1f1031fe | 5760 | |
0a4db386 | 5761 | if ( flags ) { |
59cd0e26 NC |
5762 | SV* sv_name |
5763 | = newSVpvn_flags(name_start, (int)(RExC_parse - name_start), | |
5764 | SVs_TEMP | (UTF ? SVf_UTF8 : 0)); | |
0a4db386 YO |
5765 | if ( flags == REG_RSN_RETURN_NAME) |
5766 | return sv_name; | |
5767 | else if (flags==REG_RSN_RETURN_DATA) { | |
5768 | HE *he_str = NULL; | |
5769 | SV *sv_dat = NULL; | |
5770 | if ( ! sv_name ) /* should not happen*/ | |
5771 | Perl_croak(aTHX_ "panic: no svname in reg_scan_name"); | |
5772 | if (RExC_paren_names) | |
5773 | he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 ); | |
5774 | if ( he_str ) | |
5775 | sv_dat = HeVAL(he_str); | |
5776 | if ( ! sv_dat ) | |
5777 | vFAIL("Reference to nonexistent named group"); | |
5778 | return sv_dat; | |
5779 | } | |
5780 | else { | |
5781 | Perl_croak(aTHX_ "panic: bad flag in reg_scan_name"); | |
5782 | } | |
5783 | /* NOT REACHED */ | |
894be9b7 | 5784 | } |
0a4db386 | 5785 | return NULL; |
894be9b7 YO |
5786 | } |
5787 | ||
3dab1dad YO |
5788 | #define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \ |
5789 | int rem=(int)(RExC_end - RExC_parse); \ | |
5790 | int cut; \ | |
5791 | int num; \ | |
5792 | int iscut=0; \ | |
5793 | if (rem>10) { \ | |
5794 | rem=10; \ | |
5795 | iscut=1; \ | |
5796 | } \ | |
5797 | cut=10-rem; \ | |
5798 | if (RExC_lastparse!=RExC_parse) \ | |
5799 | PerlIO_printf(Perl_debug_log," >%.*s%-*s", \ | |
5800 | rem, RExC_parse, \ | |
5801 | cut + 4, \ | |
5802 | iscut ? "..." : "<" \ | |
5803 | ); \ | |
5804 | else \ | |
5805 | PerlIO_printf(Perl_debug_log,"%16s",""); \ | |
5806 | \ | |
5807 | if (SIZE_ONLY) \ | |
3b57cd43 | 5808 | num = RExC_size + 1; \ |
3dab1dad YO |
5809 | else \ |
5810 | num=REG_NODE_NUM(RExC_emit); \ | |
5811 | if (RExC_lastnum!=num) \ | |
0a4db386 | 5812 | PerlIO_printf(Perl_debug_log,"|%4d",num); \ |
3dab1dad | 5813 | else \ |
0a4db386 | 5814 | PerlIO_printf(Perl_debug_log,"|%4s",""); \ |
be8e71aa YO |
5815 | PerlIO_printf(Perl_debug_log,"|%*s%-4s", \ |
5816 | (int)((depth*2)), "", \ | |
3dab1dad YO |
5817 | (funcname) \ |
5818 | ); \ | |
5819 | RExC_lastnum=num; \ | |
5820 | RExC_lastparse=RExC_parse; \ | |
5821 | }) | |
5822 | ||
07be1b83 YO |
5823 | |
5824 | ||
3dab1dad YO |
5825 | #define DEBUG_PARSE(funcname) DEBUG_PARSE_r({ \ |
5826 | DEBUG_PARSE_MSG((funcname)); \ | |
5827 | PerlIO_printf(Perl_debug_log,"%4s","\n"); \ | |
5828 | }) | |
6bda09f9 YO |
5829 | #define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \ |
5830 | DEBUG_PARSE_MSG((funcname)); \ | |
5831 | PerlIO_printf(Perl_debug_log,fmt "\n",args); \ | |
5832 | }) | |
d764b54e KW |
5833 | |
5834 | /* This section of code defines the inversion list object and its methods. The | |
5835 | * interfaces are highly subject to change, so as much as possible is static to | |
fa2d2a23 KW |
5836 | * this file. An inversion list is here implemented as a malloc'd C UV array |
5837 | * with some added info that is placed as UVs at the beginning in a header | |
5838 | * portion. An inversion list for Unicode is an array of code points, sorted | |
5839 | * by ordinal number. The zeroth element is the first code point in the list. | |
5840 | * The 1th element is the first element beyond that not in the list. In other | |
5841 | * words, the first range is | |
5842 | * invlist[0]..(invlist[1]-1) | |
dbe7a391 KW |
5843 | * The other ranges follow. Thus every element whose index is divisible by two |
5844 | * marks the beginning of a range that is in the list, and every element not | |
fa2d2a23 KW |
5845 | * divisible by two marks the beginning of a range not in the list. A single |
5846 | * element inversion list that contains the single code point N generally | |
5847 | * consists of two elements | |
5848 | * invlist[0] == N | |
5849 | * invlist[1] == N+1 | |
5850 | * (The exception is when N is the highest representable value on the | |
5851 | * machine, in which case the list containing just it would be a single | |
5852 | * element, itself. By extension, if the last range in the list extends to | |
5853 | * infinity, then the first element of that range will be in the inversion list | |
5854 | * at a position that is divisible by two, and is the final element in the | |
5855 | * list.) | |
f1b67122 KW |
5856 | * Taking the complement (inverting) an inversion list is quite simple, if the |
5857 | * first element is 0, remove it; otherwise add a 0 element at the beginning. | |
5858 | * This implementation reserves an element at the beginning of each inversion list | |
5859 | * to contain 0 when the list contains 0, and contains 1 otherwise. The actual | |
5860 | * beginning of the list is either that element if 0, or the next one if 1. | |
5861 | * | |
fa2d2a23 KW |
5862 | * More about inversion lists can be found in "Unicode Demystified" |
5863 | * Chapter 13 by Richard Gillam, published by Addison-Wesley. | |
97b14ce7 | 5864 | * More will be coming when functionality is added later. |
d764b54e | 5865 | * |
fa2d2a23 KW |
5866 | * The inversion list data structure is currently implemented as an SV pointing |
5867 | * to an array of UVs that the SV thinks are bytes. This allows us to have an | |
5868 | * array of UV whose memory management is automatically handled by the existing | |
5869 | * facilities for SV's. | |
62672576 | 5870 | * |
d764b54e KW |
5871 | * Some of the methods should always be private to the implementation, and some |
5872 | * should eventually be made public */ | |
5873 | ||
fa2d2a23 KW |
5874 | #define INVLIST_LEN_OFFSET 0 /* Number of elements in the inversion list */ |
5875 | #define INVLIST_ITER_OFFSET 1 /* Current iteration position */ | |
5876 | ||
f1b67122 KW |
5877 | #define INVLIST_ZERO_OFFSET 2 /* 0 or 1; must be last element in header */ |
5878 | /* The UV at position ZERO contains either 0 or 1. If 0, the inversion list | |
5879 | * contains the code point U+00000, and begins here. If 1, the inversion list | |
5880 | * doesn't contain U+0000, and it begins at the next UV in the array. | |
5881 | * Inverting an inversion list consists of adding or removing the 0 at the | |
5882 | * beginning of it. By reserving a space for that 0, inversion can be made | |
5883 | * very fast */ | |
5884 | ||
5885 | #define HEADER_LENGTH (INVLIST_ZERO_OFFSET + 1) | |
97b14ce7 KW |
5886 | |
5887 | /* Internally things are UVs */ | |
5888 | #define TO_INTERNAL_SIZE(x) ((x + HEADER_LENGTH) * sizeof(UV)) | |
5889 | #define FROM_INTERNAL_SIZE(x) ((x / sizeof(UV)) - HEADER_LENGTH) | |
5890 | ||
d764b54e | 5891 | #define INVLIST_INITIAL_LEN 10 |
d764b54e KW |
5892 | |
5893 | PERL_STATIC_INLINE UV* | |
f1b67122 KW |
5894 | S__invlist_array_init(pTHX_ SV* const invlist, const bool will_have_0) |
5895 | { | |
5896 | /* Returns a pointer to the first element in the inversion list's array. | |
5897 | * This is called upon initialization of an inversion list. Where the | |
5898 | * array begins depends on whether the list has the code point U+0000 | |
5899 | * in it or not. The other parameter tells it whether the code that | |
5900 | * follows this call is about to put a 0 in the inversion list or not. | |
5901 | * The first element is either the element with 0, if 0, or the next one, | |
5902 | * if 1 */ | |
5903 | ||
5904 | UV* zero = get_invlist_zero_addr(invlist); | |
5905 | ||
5906 | PERL_ARGS_ASSERT__INVLIST_ARRAY_INIT; | |
5907 | ||
5908 | /* Must be empty */ | |
5909 | assert(! *get_invlist_len_addr(invlist)); | |
5910 | ||
5911 | /* 1^1 = 0; 1^0 = 1 */ | |
5912 | *zero = 1 ^ will_have_0; | |
5913 | return zero + *zero; | |
5914 | } | |
5915 | ||
5916 | PERL_STATIC_INLINE UV* | |
a25abddc | 5917 | S_invlist_array(pTHX_ SV* const invlist) |
d764b54e KW |
5918 | { |
5919 | /* Returns the pointer to the inversion list's array. Every time the | |
5920 | * length changes, this needs to be called in case malloc or realloc moved | |
5921 | * it */ | |
5922 | ||
d764b54e KW |
5923 | PERL_ARGS_ASSERT_INVLIST_ARRAY; |
5924 | ||
dbe7a391 KW |
5925 | /* Must not be empty. If these fail, you probably didn't check for <len> |
5926 | * being non-zero before trying to get the array */ | |
f1b67122 KW |
5927 | assert(*get_invlist_len_addr(invlist)); |
5928 | assert(*get_invlist_zero_addr(invlist) == 0 | |
5929 | || *get_invlist_zero_addr(invlist) == 1); | |
5930 | ||
5931 | /* The array begins either at the element reserved for zero if the | |
5932 | * list contains 0 (that element will be set to 0), or otherwise the next | |
5933 | * element (in which case the reserved element will be set to 1). */ | |
5934 | return (UV *) (get_invlist_zero_addr(invlist) | |
5935 | + *get_invlist_zero_addr(invlist)); | |
d764b54e KW |
5936 | } |
5937 | ||
61bdbf38 KW |
5938 | PERL_STATIC_INLINE UV* |
5939 | S_get_invlist_len_addr(pTHX_ SV* invlist) | |
5940 | { | |
5941 | /* Return the address of the UV that contains the current number | |
5942 | * of used elements in the inversion list */ | |
5943 | ||
5944 | PERL_ARGS_ASSERT_GET_INVLIST_LEN_ADDR; | |
5945 | ||
5946 | return (UV *) (SvPVX(invlist) + (INVLIST_LEN_OFFSET * sizeof (UV))); | |
5947 | } | |
5948 | ||
d764b54e | 5949 | PERL_STATIC_INLINE UV |
a25abddc | 5950 | S_invlist_len(pTHX_ SV* const invlist) |
d764b54e | 5951 | { |
dbe7a391 KW |
5952 | /* Returns the current number of elements stored in the inversion list's |
5953 | * array */ | |
d764b54e | 5954 | |
d764b54e KW |
5955 | PERL_ARGS_ASSERT_INVLIST_LEN; |
5956 | ||
61bdbf38 | 5957 | return *get_invlist_len_addr(invlist); |
d764b54e KW |
5958 | } |
5959 | ||
c56a880b KW |
5960 | PERL_STATIC_INLINE void |
5961 | S_invlist_set_len(pTHX_ SV* const invlist, const UV len) | |
5962 | { | |
5963 | /* Sets the current number of elements stored in the inversion list */ | |
5964 | ||
5965 | PERL_ARGS_ASSERT_INVLIST_SET_LEN; | |
5966 | ||
c56a880b | 5967 | *get_invlist_len_addr(invlist) = len; |
f1b67122 | 5968 | |
32f89ef6 KW |
5969 | assert(len <= SvLEN(invlist)); |
5970 | ||
f1b67122 KW |
5971 | SvCUR_set(invlist, TO_INTERNAL_SIZE(len)); |
5972 | /* If the list contains U+0000, that element is part of the header, | |
5973 | * and should not be counted as part of the array. It will contain | |
5974 | * 0 in that case, and 1 otherwise. So we could flop 0=>1, 1=>0 and | |
5975 | * subtract: | |
5976 | * SvCUR_set(invlist, | |
5977 | * TO_INTERNAL_SIZE(len | |
5978 | * - (*get_invlist_zero_addr(inv_list) ^ 1))); | |
5979 | * But, this is only valid if len is not 0. The consequences of not doing | |
9479a769 KW |
5980 | * this is that the memory allocation code may think that 1 more UV is |
5981 | * being used than actually is, and so might do an unnecessary grow. That | |
5982 | * seems worth not bothering to make this the precise amount. | |
25e94a65 KW |
5983 | * |
5984 | * Note that when inverting, SvCUR shouldn't change */ | |
c56a880b KW |
5985 | } |
5986 | ||
d764b54e | 5987 | PERL_STATIC_INLINE UV |
a25abddc | 5988 | S_invlist_max(pTHX_ SV* const invlist) |
d764b54e KW |
5989 | { |
5990 | /* Returns the maximum number of elements storable in the inversion list's | |
5991 | * array, without having to realloc() */ | |
5992 | ||
d764b54e KW |
5993 | PERL_ARGS_ASSERT_INVLIST_MAX; |
5994 | ||
005b65ed | 5995 | return FROM_INTERNAL_SIZE(SvLEN(invlist)); |
d764b54e KW |
5996 | } |
5997 | ||
f1b67122 KW |
5998 | PERL_STATIC_INLINE UV* |
5999 | S_get_invlist_zero_addr(pTHX_ SV* invlist) | |
6000 | { | |
6001 | /* Return the address of the UV that is reserved to hold 0 if the inversion | |
6002 | * list contains 0. This has to be the last element of the heading, as the | |
6003 | * list proper starts with either it if 0, or the next element if not. | |
6004 | * (But we force it to contain either 0 or 1) */ | |
6005 | ||
6006 | PERL_ARGS_ASSERT_GET_INVLIST_ZERO_ADDR; | |
6007 | ||
6008 | return (UV *) (SvPVX(invlist) + (INVLIST_ZERO_OFFSET * sizeof (UV))); | |
6009 | } | |
d764b54e | 6010 | |
8d69a883 | 6011 | #ifndef PERL_IN_XSUB_RE |
a25abddc | 6012 | SV* |
d764b54e KW |
6013 | Perl__new_invlist(pTHX_ IV initial_size) |
6014 | { | |
6015 | ||
6016 | /* Return a pointer to a newly constructed inversion list, with enough | |
6017 | * space to store 'initial_size' elements. If that number is negative, a | |
6018 | * system default is used instead */ | |
6019 | ||
97b14ce7 KW |
6020 | SV* new_list; |
6021 | ||
d764b54e KW |
6022 | if (initial_size < 0) { |
6023 | initial_size = INVLIST_INITIAL_LEN; | |
6024 | } | |
6025 | ||
6026 | /* Allocate the initial space */ | |
97b14ce7 KW |
6027 | new_list = newSV(TO_INTERNAL_SIZE(initial_size)); |
6028 | invlist_set_len(new_list, 0); | |
6029 | ||
f3dc70d1 KW |
6030 | /* Force iterinit() to be used to get iteration to work */ |
6031 | *get_invlist_iter_addr(new_list) = UV_MAX; | |
6032 | ||
f1b67122 KW |
6033 | /* This should force a segfault if a method doesn't initialize this |
6034 | * properly */ | |
6035 | *get_invlist_zero_addr(new_list) = UV_MAX; | |
6036 | ||
97b14ce7 | 6037 | return new_list; |
d764b54e | 6038 | } |
8d69a883 | 6039 | #endif |
d764b54e | 6040 | |
d764b54e | 6041 | STATIC void |
a25abddc | 6042 | S_invlist_extend(pTHX_ SV* const invlist, const UV new_max) |
d764b54e | 6043 | { |
62672576 | 6044 | /* Grow the maximum size of an inversion list */ |
d764b54e KW |
6045 | |
6046 | PERL_ARGS_ASSERT_INVLIST_EXTEND; | |
6047 | ||
005b65ed | 6048 | SvGROW((SV *)invlist, TO_INTERNAL_SIZE(new_max)); |
d764b54e KW |
6049 | } |
6050 | ||
6051 | PERL_STATIC_INLINE void | |
a25abddc | 6052 | S_invlist_trim(pTHX_ SV* const invlist) |
d764b54e KW |
6053 | { |
6054 | PERL_ARGS_ASSERT_INVLIST_TRIM; | |
6055 | ||
6056 | /* Change the length of the inversion list to how many entries it currently | |
6057 | * has */ | |
6058 | ||
62672576 | 6059 | SvPV_shrink_to_cur((SV *) invlist); |
d764b54e KW |
6060 | } |
6061 | ||
6062 | /* An element is in an inversion list iff its index is even numbered: 0, 2, 4, | |
6063 | * etc */ | |
4096c37b KW |
6064 | #define ELEMENT_RANGE_MATCHES_INVLIST(i) (! ((i) & 1)) |
6065 | #define PREV_RANGE_MATCHES_INVLIST(i) (! ELEMENT_RANGE_MATCHES_INVLIST(i)) | |
d764b54e | 6066 | |
8d69a883 | 6067 | #ifndef PERL_IN_XSUB_RE |
d764b54e | 6068 | void |
a25abddc | 6069 | Perl__append_range_to_invlist(pTHX_ SV* const invlist, const UV start, const UV end) |
d764b54e KW |
6070 | { |
6071 | /* Subject to change or removal. Append the range from 'start' to 'end' at | |
6072 | * the end of the inversion list. The range must be above any existing | |
6073 | * ones. */ | |
6074 | ||
f1b67122 | 6075 | UV* array; |
d764b54e KW |
6076 | UV max = invlist_max(invlist); |
6077 | UV len = invlist_len(invlist); | |
6078 | ||
6079 | PERL_ARGS_ASSERT__APPEND_RANGE_TO_INVLIST; | |
6080 | ||
f1b67122 KW |
6081 | if (len == 0) { /* Empty lists must be initialized */ |
6082 | array = _invlist_array_init(invlist, start == 0); | |
6083 | } | |
6084 | else { | |
d764b54e KW |
6085 | /* Here, the existing list is non-empty. The current max entry in the |
6086 | * list is generally the first value not in the set, except when the | |
6087 | * set extends to the end of permissible values, in which case it is | |
6088 | * the first entry in that final set, and so this call is an attempt to | |
6089 | * append out-of-order */ | |
6090 | ||
6091 | UV final_element = len - 1; | |
f1b67122 | 6092 | array = invlist_array(invlist); |
d764b54e | 6093 | if (array[final_element] > start |
4096c37b | 6094 | || ELEMENT_RANGE_MATCHES_INVLIST(final_element)) |
d764b54e KW |
6095 | { |
6096 | Perl_croak(aTHX_ "panic: attempting to append to an inversion list, but wasn't at the end of the list"); | |
6097 | } | |
6098 | ||
6099 | /* Here, it is a legal append. If the new range begins with the first | |
6100 | * value not in the set, it is extending the set, so the new first | |
6101 | * value not in the set is one greater than the newly extended range. | |
6102 | * */ | |
6103 | if (array[final_element] == start) { | |
6104 | if (end != UV_MAX) { | |
6105 | array[final_element] = end + 1; | |
6106 | } | |
6107 | else { | |
6108 | /* But if the end is the maximum representable on the machine, | |
dbe7a391 | 6109 | * just let the range that this would extend to have no end */ |
d764b54e KW |
6110 | invlist_set_len(invlist, len - 1); |
6111 | } | |
6112 | return; | |
6113 | } | |
6114 | } | |
6115 | ||
6116 | /* Here the new range doesn't extend any existing set. Add it */ | |
6117 | ||
6118 | len += 2; /* Includes an element each for the start and end of range */ | |
6119 | ||
6120 | /* If overflows the existing space, extend, which may cause the array to be | |
6121 | * moved */ | |
6122 | if (max < len) { | |
6123 | invlist_extend(invlist, len); | |
f1b67122 KW |
6124 | invlist_set_len(invlist, len); /* Have to set len here to avoid assert |
6125 | failure in invlist_array() */ | |
d764b54e KW |
6126 | array = invlist_array(invlist); |
6127 | } | |
f1b67122 KW |
6128 | else { |
6129 | invlist_set_len(invlist, len); | |
6130 | } | |
d764b54e KW |
6131 | |
6132 | /* The next item on the list starts the range, the one after that is | |
6133 | * one past the new range. */ | |
6134 | array[len - 2] = start; | |
6135 | if (end != UV_MAX) { | |
6136 | array[len - 1] = end + 1; | |
6137 | } | |
6138 | else { | |
6139 | /* But if the end is the maximum representable on the machine, just let | |
6140 | * the range have no end */ | |
6141 | invlist_set_len(invlist, len - 1); | |
6142 | } | |
6143 | } | |
6144 | ||
d5e82ecc KW |
6145 | STATIC IV |
6146 | S_invlist_search(pTHX_ SV* const invlist, const UV cp) | |
6147 | { | |
6148 | /* Searches the inversion list for the entry that contains the input code | |
6149 | * point <cp>. If <cp> is not in the list, -1 is returned. Otherwise, the | |
6150 | * return value is the index into the list's array of the range that | |
6151 | * contains <cp> */ | |
6152 | ||
6153 | IV low = 0; | |
6154 | IV high = invlist_len(invlist); | |
6155 | const UV * const array = invlist_array(invlist); | |
6156 | ||
6157 | PERL_ARGS_ASSERT_INVLIST_SEARCH; | |
6158 | ||
6159 | /* If list is empty or the code point is before the first element, return | |
6160 | * failure. */ | |
6161 | if (high == 0 || cp < array[0]) { | |
6162 | return -1; | |
6163 | } | |
6164 | ||
6165 | /* Binary search. What we are looking for is <i> such that | |
6166 | * array[i] <= cp < array[i+1] | |
6167 | * The loop below converges on the i+1. */ | |
6168 | while (low < high) { | |
6169 | IV mid = (low + high) / 2; | |
6170 | if (array[mid] <= cp) { | |
6171 | low = mid + 1; | |
6172 | ||
6173 | /* We could do this extra test to exit the loop early. | |
6174 | if (cp < array[low]) { | |
6175 | return mid; | |
6176 | } | |
6177 | */ | |
6178 | } | |
6179 | else { /* cp < array[mid] */ | |
6180 | high = mid; | |
6181 | } | |
6182 | } | |
6183 | ||
6184 | return high - 1; | |
6185 | } | |
6186 | ||
86f766ab | 6187 | void |
b6a0ff33 KW |
6188 | Perl__invlist_populate_swatch(pTHX_ SV* const invlist, const UV start, const UV end, U8* swatch) |
6189 | { | |
6190 | /* populates a swatch of a swash the same way swatch_get() does in utf8.c, | |
6191 | * but is used when the swash has an inversion list. This makes this much | |
6192 | * faster, as it uses a binary search instead of a linear one. This is | |
6193 | * intimately tied to that function, and perhaps should be in utf8.c, | |
6194 | * except it is intimately tied to inversion lists as well. It assumes | |
6195 | * that <swatch> is all 0's on input */ | |
6196 | ||
6197 | UV current = start; | |
6198 | const IV len = invlist_len(invlist); | |
6199 | IV i; | |
6200 | const UV * array; | |
6201 | ||
6202 | PERL_ARGS_ASSERT__INVLIST_POPULATE_SWATCH; | |
6203 | ||
6204 | if (len == 0) { /* Empty inversion list */ | |
6205 | return; | |
6206 | } | |
6207 | ||
6208 | array = invlist_array(invlist); | |
6209 | ||
6210 | /* Find which element it is */ | |
6211 | i = invlist_search(invlist, start); | |
6212 | ||
6213 | /* We populate from <start> to <end> */ | |
6214 | while (current < end) { | |
6215 | UV upper; | |
6216 | ||
6217 | /* The inversion list gives the results for every possible code point | |
6218 | * after the first one in the list. Only those ranges whose index is | |
6219 | * even are ones that the inversion list matches. For the odd ones, | |
6220 | * and if the initial code point is not in the list, we have to skip | |
6221 | * forward to the next element */ | |
6222 | if (i == -1 || ! ELEMENT_RANGE_MATCHES_INVLIST(i)) { | |
6223 | i++; | |
6224 | if (i >= len) { /* Finished if beyond the end of the array */ | |
6225 | return; | |
6226 | } | |
6227 | current = array[i]; | |
6228 | if (current >= end) { /* Finished if beyond the end of what we | |
6229 | are populating */ | |
6230 | return; | |
6231 | } | |
6232 | } | |
6233 | assert(current >= start); | |
6234 | ||
6235 | /* The current range ends one below the next one, except don't go past | |
6236 | * <end> */ | |
6237 | i++; | |
6238 | upper = (i < len && array[i] < end) ? array[i] : end; | |
6239 | ||
6240 | /* Here we are in a range that matches. Populate a bit in the 3-bit U8 | |
6241 | * for each code point in it */ | |
6242 | for (; current < upper; current++) { | |
6243 | const STRLEN offset = (STRLEN)(current - start); | |
6244 | swatch[offset >> 3] |= 1 << (offset & 7); | |
6245 | } | |
6246 | ||
6247 | /* Quit if at the end of the list */ | |
6248 | if (i >= len) { | |
6249 | ||
6250 | /* But first, have to deal with the highest possible code point on | |
6251 | * the platform. The previous code assumes that <end> is one | |
6252 | * beyond where we want to populate, but that is impossible at the | |
6253 | * platform's infinity, so have to handle it specially */ | |
6254 | if (UNLIKELY(end == UV_MAX && ELEMENT_RANGE_MATCHES_INVLIST(len-1))) | |
6255 | { | |
6256 | const STRLEN offset = (STRLEN)(end - start); | |
6257 | swatch[offset >> 3] |= 1 << (offset & 7); | |
6258 | } | |
6259 | return; | |
6260 | } | |
6261 | ||
6262 | /* Advance to the next range, which will be for code points not in the | |
6263 | * inversion list */ | |
6264 | current = array[i]; | |
6265 | } | |
6266 | ||
6267 | return; | |
6268 | } | |
6269 | ||
6270 | void | |
86f766ab | 6271 | Perl__invlist_union(pTHX_ SV* const a, SV* const b, SV** output) |
d764b54e | 6272 | { |
4065ba03 KW |
6273 | /* Take the union of two inversion lists and point <output> to it. *output |
6274 | * should be defined upon input, and if it points to one of the two lists, | |
6275 | * the reference count to that list will be decremented. | |
d764b54e KW |
6276 | * The basis for this comes from "Unicode Demystified" Chapter 13 by |
6277 | * Richard Gillam, published by Addison-Wesley, and explained at some | |
6278 | * length there. The preface says to incorporate its examples into your | |
6279 | * code at your own risk. | |
6280 | * | |
6281 | * The algorithm is like a merge sort. | |
6282 | * | |
6283 | * XXX A potential performance improvement is to keep track as we go along | |
6284 | * if only one of the inputs contributes to the result, meaning the other | |
6285 | * is a subset of that one. In that case, we can skip the final copy and | |
a2995b7f KW |
6286 | * return the larger of the input lists, but then outside code might need |
6287 | * to keep track of whether to free the input list or not */ | |
d764b54e | 6288 | |
f1b67122 KW |
6289 | UV* array_a; /* a's array */ |
6290 | UV* array_b; | |
6291 | UV len_a; /* length of a's array */ | |
6292 | UV len_b; | |
d764b54e | 6293 | |
a25abddc | 6294 | SV* u; /* the resulting union */ |
d764b54e KW |
6295 | UV* array_u; |
6296 | UV len_u; | |
6297 | ||
6298 | UV i_a = 0; /* current index into a's array */ | |
6299 | UV i_b = 0; | |
6300 | UV i_u = 0; | |
6301 | ||
6302 | /* running count, as explained in the algorithm source book; items are | |
6303 | * stopped accumulating and are output when the count changes to/from 0. | |
6304 | * The count is incremented when we start a range that's in the set, and | |
6305 | * decremented when we start a range that's not in the set. So its range | |
6306 | * is 0 to 2. Only when the count is zero is something not in the set. | |
6307 | */ | |
6308 | UV count = 0; | |
6309 | ||
37e85ffe | 6310 | PERL_ARGS_ASSERT__INVLIST_UNION; |
60825692 | 6311 | assert(a != b); |
d764b54e | 6312 | |
f1b67122 KW |
6313 | /* If either one is empty, the union is the other one */ |
6314 | len_a = invlist_len(a); | |
6315 | if (len_a == 0) { | |
4065ba03 | 6316 | if (*output == a) { |
f1b67122 KW |
6317 | SvREFCNT_dec(a); |
6318 | } | |
4065ba03 | 6319 | if (*output != b) { |
f1b67122 | 6320 | *output = invlist_clone(b); |
dbe7a391 | 6321 | } /* else *output already = b; */ |
f1b67122 KW |
6322 | return; |
6323 | } | |
6324 | else if ((len_b = invlist_len(b)) == 0) { | |
4065ba03 | 6325 | if (*output == b) { |
f1b67122 KW |
6326 | SvREFCNT_dec(b); |
6327 | } | |
4065ba03 | 6328 | if (*output != a) { |
f1b67122 KW |
6329 | *output = invlist_clone(a); |
6330 | } | |
6331 | /* else *output already = a; */ | |
6332 | return; | |
6333 | } | |
6334 | ||
6335 | /* Here both lists exist and are non-empty */ | |
6336 | array_a = invlist_array(a); | |
6337 | array_b = invlist_array(b); | |
6338 | ||
d764b54e KW |
6339 | /* Size the union for the worst case: that the sets are completely |
6340 | * disjoint */ | |
6341 | u = _new_invlist(len_a + len_b); | |
f1b67122 KW |
6342 | |
6343 | /* Will contain U+0000 if either component does */ | |
6344 | array_u = _invlist_array_init(u, (len_a > 0 && array_a[0] == 0) | |
6345 | || (len_b > 0 && array_b[0] == 0)); | |
d764b54e KW |
6346 | |
6347 | /* Go through each list item by item, stopping when exhausted one of | |
6348 | * them */ | |
6349 | while (i_a < len_a && i_b < len_b) { | |
6350 | UV cp; /* The element to potentially add to the union's array */ | |
6351 | bool cp_in_set; /* is it in the the input list's set or not */ | |
6352 | ||
6353 | /* We need to take one or the other of the two inputs for the union. | |
6354 | * Since we are merging two sorted lists, we take the smaller of the | |
6355 | * next items. In case of a tie, we take the one that is in its set | |
6356 | * first. If we took one not in the set first, it would decrement the | |
6357 | * count, possibly to 0 which would cause it to be output as ending the | |
6358 | * range, and the next time through we would take the same number, and | |
6359 | * output it again as beginning the next range. By doing it the | |
6360 | * opposite way, there is no possibility that the count will be | |
6361 | * momentarily decremented to 0, and thus the two adjoining ranges will | |
6362 | * be seamlessly merged. (In a tie and both are in the set or both not | |
6363 | * in the set, it doesn't matter which we take first.) */ | |
6364 | if (array_a[i_a] < array_b[i_b] | |
4096c37b KW |
6365 | || (array_a[i_a] == array_b[i_b] |
6366 | && ELEMENT_RANGE_MATCHES_INVLIST(i_a))) | |
d764b54e | 6367 | { |
4096c37b | 6368 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a); |
d764b54e KW |
6369 | cp= array_a[i_a++]; |
6370 | } | |
6371 | else { | |
4096c37b | 6372 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b); |
d764b54e KW |
6373 | cp= array_b[i_b++]; |
6374 | } | |
6375 | ||
6376 | /* Here, have chosen which of the two inputs to look at. Only output | |
6377 | * if the running count changes to/from 0, which marks the | |
6378 | * beginning/end of a range in that's in the set */ | |
6379 | if (cp_in_set) { | |
6380 | if (count == 0) { | |
6381 | array_u[i_u++] = cp; | |
6382 | } | |
6383 | count++; | |
6384 | } | |
6385 | else { | |
6386 | count--; | |
6387 | if (count == 0) { | |
6388 | array_u[i_u++] = cp; | |
6389 | } | |
6390 | } | |
6391 | } | |
6392 | ||
6393 | /* Here, we are finished going through at least one of the lists, which | |
6394 | * means there is something remaining in at most one. We check if the list | |
6395 | * that hasn't been exhausted is positioned such that we are in the middle | |
bac5f0ae KW |
6396 | * of a range in its set or not. (i_a and i_b point to the element beyond |
6397 | * the one we care about.) If in the set, we decrement 'count'; if 0, there | |
6398 | * is potentially more to output. | |
d764b54e KW |
6399 | * There are four cases: |
6400 | * 1) Both weren't in their sets, count is 0, and remains 0. What's left | |
6401 | * in the union is entirely from the non-exhausted set. | |
6402 | * 2) Both were in their sets, count is 2. Nothing further should | |
6403 | * be output, as everything that remains will be in the exhausted | |
6404 | * list's set, hence in the union; decrementing to 1 but not 0 insures | |
6405 | * that | |
6406 | * 3) the exhausted was in its set, non-exhausted isn't, count is 1. | |
6407 | * Nothing further should be output because the union includes | |
bac5f0ae | 6408 | * everything from the exhausted set. Not decrementing ensures that. |
d764b54e KW |
6409 | * 4) the exhausted wasn't in its set, non-exhausted is, count is 1; |
6410 | * decrementing to 0 insures that we look at the remainder of the | |
6411 | * non-exhausted set */ | |
4096c37b KW |
6412 | if ((i_a != len_a && PREV_RANGE_MATCHES_INVLIST(i_a)) |
6413 | || (i_b != len_b && PREV_RANGE_MATCHES_INVLIST(i_b))) | |
d764b54e KW |
6414 | { |
6415 | count--; | |
6416 | } | |
6417 | ||
6418 | /* The final length is what we've output so far, plus what else is about to | |
6419 | * be output. (If 'count' is non-zero, then the input list we exhausted | |
6420 | * has everything remaining up to the machine's limit in its set, and hence | |
6421 | * in the union, so there will be no further output. */ | |
6422 | len_u = i_u; | |
6423 | if (count == 0) { | |
6424 | /* At most one of the subexpressions will be non-zero */ | |
6425 | len_u += (len_a - i_a) + (len_b - i_b); | |
6426 | } | |
6427 | ||
6428 | /* Set result to final length, which can change the pointer to array_u, so | |
6429 | * re-find it */ | |
6430 | if (len_u != invlist_len(u)) { | |
6431 | invlist_set_len(u, len_u); | |
6432 | invlist_trim(u); | |
6433 | array_u = invlist_array(u); | |
6434 | } | |
6435 | ||
6436 | /* When 'count' is 0, the list that was exhausted (if one was shorter than | |
6437 | * the other) ended with everything above it not in its set. That means | |
6438 | * that the remaining part of the union is precisely the same as the | |
6439 | * non-exhausted list, so can just copy it unchanged. (If both list were | |
6440 | * exhausted at the same time, then the operations below will be both 0.) | |
6441 | */ | |
6442 | if (count == 0) { | |
6443 | IV copy_count; /* At most one will have a non-zero copy count */ | |
6444 | if ((copy_count = len_a - i_a) > 0) { | |
6445 | Copy(array_a + i_a, array_u + i_u, copy_count, UV); | |
6446 | } | |
6447 | else if ((copy_count = len_b - i_b) > 0) { | |
6448 | Copy(array_b + i_b, array_u + i_u, copy_count, UV); | |
6449 | } | |
6450 | } | |
6451 | ||
a2995b7f | 6452 | /* We may be removing a reference to one of the inputs */ |
4065ba03 | 6453 | if (a == *output || b == *output) { |
a2995b7f KW |
6454 | SvREFCNT_dec(*output); |
6455 | } | |
6456 | ||
6457 | *output = u; | |
6458 | return; | |
d764b54e KW |
6459 | } |
6460 | ||
86f766ab KW |
6461 | void |
6462 | Perl__invlist_intersection(pTHX_ SV* const a, SV* const b, SV** i) | |
d764b54e | 6463 | { |
4065ba03 KW |
6464 | /* Take the intersection of two inversion lists and point <i> to it. *i |
6465 | * should be defined upon input, and if it points to one of the two lists, | |
6466 | * the reference count to that list will be decremented. | |
a2995b7f KW |
6467 | * The basis for this comes from "Unicode Demystified" Chapter 13 by |
6468 | * Richard Gillam, published by Addison-Wesley, and explained at some | |
6469 | * length there. The preface says to incorporate its examples into your | |
6470 | * code at your own risk. In fact, it had bugs | |
d764b54e KW |
6471 | * |
6472 | * The algorithm is like a merge sort, and is essentially the same as the | |
6473 | * union above | |
6474 | */ | |
6475 | ||
f1b67122 KW |
6476 | UV* array_a; /* a's array */ |
6477 | UV* array_b; | |
6478 | UV len_a; /* length of a's array */ | |
6479 | UV len_b; | |
d764b54e | 6480 | |
a25abddc | 6481 | SV* r; /* the resulting intersection */ |
d764b54e KW |
6482 | UV* array_r; |
6483 | UV len_r; | |
6484 | ||
6485 | UV i_a = 0; /* current index into a's array */ | |
6486 | UV i_b = 0; | |
6487 | UV i_r = 0; | |
6488 | ||
6489 | /* running count, as explained in the algorithm source book; items are | |
6490 | * stopped accumulating and are output when the count changes to/from 2. | |
6491 | * The count is incremented when we start a range that's in the set, and | |
6492 | * decremented when we start a range that's not in the set. So its range | |
6493 | * is 0 to 2. Only when the count is 2 is something in the intersection. | |
6494 | */ | |
6495 | UV count = 0; | |
6496 | ||
37e85ffe | 6497 | PERL_ARGS_ASSERT__INVLIST_INTERSECTION; |
60825692 | 6498 | assert(a != b); |
d764b54e | 6499 | |
f1b67122 KW |
6500 | /* If either one is empty, the intersection is null */ |
6501 | len_a = invlist_len(a); | |
6502 | if ((len_a == 0) || ((len_b = invlist_len(b)) == 0)) { | |
f1b67122 KW |
6503 | |
6504 | /* If the result is the same as one of the inputs, the input is being | |
6505 | * overwritten */ | |
4065ba03 | 6506 | if (*i == a) { |
f1b67122 KW |
6507 | SvREFCNT_dec(a); |
6508 | } | |
4065ba03 | 6509 | else if (*i == b) { |
f1b67122 KW |
6510 | SvREFCNT_dec(b); |
6511 | } | |
2ea86699 KW |
6512 | |
6513 | *i = _new_invlist(0); | |
f1b67122 KW |
6514 | return; |
6515 | } | |
6516 | ||
6517 | /* Here both lists exist and are non-empty */ | |
6518 | array_a = invlist_array(a); | |
6519 | array_b = invlist_array(b); | |
6520 | ||
d764b54e KW |
6521 | /* Size the intersection for the worst case: that the intersection ends up |
6522 | * fragmenting everything to be completely disjoint */ | |
6523 | r= _new_invlist(len_a + len_b); | |
f1b67122 KW |
6524 | |
6525 | /* Will contain U+0000 iff both components do */ | |
6526 | array_r = _invlist_array_init(r, len_a > 0 && array_a[0] == 0 | |
6527 | && len_b > 0 && array_b[0] == 0); | |
d764b54e KW |
6528 | |
6529 | /* Go through each list item by item, stopping when exhausted one of | |
6530 | * them */ | |
6531 | while (i_a < len_a && i_b < len_b) { | |
6532 | UV cp; /* The element to potentially add to the intersection's | |
6533 | array */ | |
6534 | bool cp_in_set; /* Is it in the input list's set or not */ | |
6535 | ||
c4a30257 KW |
6536 | /* We need to take one or the other of the two inputs for the |
6537 | * intersection. Since we are merging two sorted lists, we take the | |
6538 | * smaller of the next items. In case of a tie, we take the one that | |
6539 | * is not in its set first (a difference from the union algorithm). If | |
6540 | * we took one in the set first, it would increment the count, possibly | |
6541 | * to 2 which would cause it to be output as starting a range in the | |
6542 | * intersection, and the next time through we would take that same | |
6543 | * number, and output it again as ending the set. By doing it the | |
6544 | * opposite of this, there is no possibility that the count will be | |
6545 | * momentarily incremented to 2. (In a tie and both are in the set or | |
6546 | * both not in the set, it doesn't matter which we take first.) */ | |
d764b54e | 6547 | if (array_a[i_a] < array_b[i_b] |
4096c37b KW |
6548 | || (array_a[i_a] == array_b[i_b] |
6549 | && ! ELEMENT_RANGE_MATCHES_INVLIST(i_a))) | |
d764b54e | 6550 | { |
4096c37b | 6551 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_a); |
d764b54e KW |
6552 | cp= array_a[i_a++]; |
6553 | } | |
6554 | else { | |
4096c37b | 6555 | cp_in_set = ELEMENT_RANGE_MATCHES_INVLIST(i_b); |
d764b54e KW |
6556 | cp= array_b[i_b++]; |
6557 | } | |
6558 | ||
6559 | /* Here, have chosen which of the two inputs to look at. Only output | |
6560 | * if the running count changes to/from 2, which marks the | |
6561 | * beginning/end of a range that's in the intersection */ | |
6562 | if (cp_in_set) { | |
6563 | count++; | |
6564 | if (count == 2) { | |
6565 | array_r[i_r++] = cp; | |
6566 | } | |
6567 | } | |
6568 | else { | |
6569 | if (count == 2) { | |
6570 | array_r[i_r++] = cp; | |
6571 | } | |
6572 | count--; | |
6573 | } | |
6574 | } | |
6575 | ||
c4a30257 KW |
6576 | /* Here, we are finished going through at least one of the lists, which |
6577 | * means there is something remaining in at most one. We check if the list | |
6578 | * that has been exhausted is positioned such that we are in the middle | |
6579 | * of a range in its set or not. (i_a and i_b point to elements 1 beyond | |
6580 | * the ones we care about.) There are four cases: | |
6581 | * 1) Both weren't in their sets, count is 0, and remains 0. There's | |
6582 | * nothing left in the intersection. | |
6583 | * 2) Both were in their sets, count is 2 and perhaps is incremented to | |
6584 | * above 2. What should be output is exactly that which is in the | |
6585 | * non-exhausted set, as everything it has is also in the intersection | |
6586 | * set, and everything it doesn't have can't be in the intersection | |
6587 | * 3) The exhausted was in its set, non-exhausted isn't, count is 1, and | |
6588 | * gets incremented to 2. Like the previous case, the intersection is | |
6589 | * everything that remains in the non-exhausted set. | |
6590 | * 4) the exhausted wasn't in its set, non-exhausted is, count is 1, and | |
6591 | * remains 1. And the intersection has nothing more. */ | |
4096c37b KW |
6592 | if ((i_a == len_a && PREV_RANGE_MATCHES_INVLIST(i_a)) |
6593 | || (i_b == len_b && PREV_RANGE_MATCHES_INVLIST(i_b))) | |
d764b54e | 6594 | { |
c4a30257 | 6595 | count++; |
d764b54e KW |
6596 | } |
6597 | ||
6598 | /* The final length is what we've output so far plus what else is in the | |
c4a30257 | 6599 | * intersection. At most one of the subexpressions below will be non-zero */ |
d764b54e | 6600 | len_r = i_r; |
c4a30257 | 6601 | if (count >= 2) { |
d764b54e KW |
6602 | len_r += (len_a - i_a) + (len_b - i_b); |
6603 | } | |
6604 | ||
6605 | /* Set result to final length, which can change the pointer to array_r, so | |
6606 | * re-find it */ | |
6607 | if (len_r != invlist_len(r)) { | |
6608 | invlist_set_len(r, len_r); | |
6609 | invlist_trim(r); | |
6610 | array_r = invlist_array(r); | |
6611 | } | |
6612 | ||
6613 | /* Finish outputting any remaining */ | |
c4a30257 | 6614 | if (count >= 2) { /* At most one will have a non-zero copy count */ |
d764b54e KW |
6615 | IV copy_count; |
6616 | if ((copy_count = len_a - i_a) > 0) { | |
6617 | Copy(array_a + i_a, array_r + i_r, copy_count, UV); | |
6618 | } | |
6619 | else if ((copy_count = len_b - i_b) > 0) { | |
6620 | Copy(array_b + i_b, array_r + i_r, copy_count, UV); | |
6621 | } | |
6622 | } | |
6623 | ||
a2995b7f | 6624 | /* We may be removing a reference to one of the inputs */ |
4065ba03 | 6625 | if (a == *i || b == *i) { |
a2995b7f KW |
6626 | SvREFCNT_dec(*i); |
6627 | } | |
6628 | ||
6629 | *i = r; | |
6630 | return; | |
d764b54e KW |
6631 | } |
6632 | ||
3c234b35 TC |
6633 | #endif |
6634 | ||
a25abddc KW |
6635 | STATIC SV* |
6636 | S_add_range_to_invlist(pTHX_ SV* invlist, const UV start, const UV end) | |
d764b54e KW |
6637 | { |
6638 | /* Add the range from 'start' to 'end' inclusive to the inversion list's | |
6639 | * set. A pointer to the inversion list is returned. This may actually be | |
c52a3e71 KW |
6640 | * a new list, in which case the passed in one has been destroyed. The |
6641 | * passed in inversion list can be NULL, in which case a new one is created | |
6642 | * with just the one range in it */ | |
d764b54e | 6643 | |
a25abddc | 6644 | SV* range_invlist; |
c52a3e71 | 6645 | UV len; |
d764b54e | 6646 | |
c52a3e71 KW |
6647 | if (invlist == NULL) { |
6648 | invlist = _new_invlist(2); | |
6649 | len = 0; | |
6650 | } | |
6651 | else { | |
6652 | len = invlist_len(invlist); | |
6653 | } | |
d764b54e KW |
6654 | |
6655 | /* If comes after the final entry, can just append it to the end */ | |
6656 | if (len == 0 | |
6657 | || start >= invlist_array(invlist) | |
6658 | [invlist_len(invlist) - 1]) | |
6659 | { | |
6660 | _append_range_to_invlist(invlist, start, end); | |
6661 | return invlist; | |
6662 | } | |
6663 | ||
6664 | /* Here, can't just append things, create and return a new inversion list | |
6665 | * which is the union of this range and the existing inversion list */ | |
6666 | range_invlist = _new_invlist(2); | |
6667 | _append_range_to_invlist(range_invlist, start, end); | |
6668 | ||
37e85ffe | 6669 | _invlist_union(invlist, range_invlist, &invlist); |
d764b54e | 6670 | |
0a89af2f | 6671 | /* The temporary can be freed */ |
318c430e | 6672 | SvREFCNT_dec(range_invlist); |
d764b54e | 6673 | |
6d63a9fb | 6674 | return invlist; |
d764b54e KW |
6675 | } |
6676 | ||
a25abddc KW |
6677 | PERL_STATIC_INLINE SV* |
6678 | S_add_cp_to_invlist(pTHX_ SV* invlist, const UV cp) { | |
c229b64c KW |
6679 | return add_range_to_invlist(invlist, cp, cp); |
6680 | } | |
6681 | ||
3c234b35 | 6682 | #ifndef PERL_IN_XSUB_RE |
86f766ab KW |
6683 | void |
6684 | Perl__invlist_invert(pTHX_ SV* const invlist) | |
25e94a65 KW |
6685 | { |
6686 | /* Complement the input inversion list. This adds a 0 if the list didn't | |
6687 | * have a zero; removes it otherwise. As described above, the data | |
6688 | * structure is set up so that this is very efficient */ | |
6689 | ||
6690 | UV* len_pos = get_invlist_len_addr(invlist); | |
6691 | ||
37e85ffe | 6692 | PERL_ARGS_ASSERT__INVLIST_INVERT; |
25e94a65 KW |
6693 | |
6694 | /* The inverse of matching nothing is matching everything */ | |
6695 | if (*len_pos == 0) { | |
6696 | _append_range_to_invlist(invlist, 0, UV_MAX); | |
6697 | return; | |
6698 | } | |
6699 | ||
6700 | /* The exclusive or complents 0 to 1; and 1 to 0. If the result is 1, the | |
6701 | * zero element was a 0, so it is being removed, so the length decrements | |
6702 | * by 1; and vice-versa. SvCUR is unaffected */ | |
6703 | if (*get_invlist_zero_addr(invlist) ^= 1) { | |
6704 | (*len_pos)--; | |
6705 | } | |
6706 | else { | |
6707 | (*len_pos)++; | |
6708 | } | |
6709 | } | |
89302fc2 KW |
6710 | |
6711 | void | |
6712 | Perl__invlist_invert_prop(pTHX_ SV* const invlist) | |
6713 | { | |
6714 | /* Complement the input inversion list (which must be a Unicode property, | |
6715 | * all of which don't match above the Unicode maximum code point.) And | |
6716 | * Perl has chosen to not have the inversion match above that either. This | |
6717 | * adds a 0x110000 if the list didn't end with it, and removes it if it did | |
6718 | */ | |
6719 | ||
6720 | UV len; | |
6721 | UV* array; | |
6722 | ||
6723 | PERL_ARGS_ASSERT__INVLIST_INVERT_PROP; | |
6724 | ||
6725 | _invlist_invert(invlist); | |
6726 | ||
6727 | len = invlist_len(invlist); | |
6728 | ||
6729 | if (len != 0) { /* If empty do nothing */ | |
6730 | array = invlist_array(invlist); | |
6731 | if (array[len - 1] != PERL_UNICODE_MAX + 1) { | |
6732 | /* Add 0x110000. First, grow if necessary */ | |
6733 | len++; | |
6734 | if (invlist_max(invlist) < len) { | |
6735 | invlist_extend(invlist, len); | |
6736 | array = invlist_array(invlist); | |
6737 | } | |
6738 | invlist_set_len(invlist, len); | |
6739 | array[len - 1] = PERL_UNICODE_MAX + 1; | |
6740 | } | |
6741 | else { /* Remove the 0x110000 */ | |
6742 | invlist_set_len(invlist, len - 1); | |
6743 | } | |
6744 | } | |
6745 | ||
6746 | return; | |
6747 | } | |
3c234b35 | 6748 | #endif |
25e94a65 KW |
6749 | |
6750 | PERL_STATIC_INLINE SV* | |
6751 | S_invlist_clone(pTHX_ SV* const invlist) | |
6752 | { | |
6753 | ||
6754 | /* Return a new inversion list that is a copy of the input one, which is | |
6755 | * unchanged */ | |
6756 | ||
6c6c83ac KW |
6757 | /* Need to allocate extra space to accommodate Perl's addition of a |
6758 | * trailing NUL to SvPV's, since it thinks they are always strings */ | |
6759 | SV* new_invlist = _new_invlist(invlist_len(invlist) + 1); | |
6d47fb3d | 6760 | STRLEN length = SvCUR(invlist); |
25e94a65 KW |
6761 | |
6762 | PERL_ARGS_ASSERT_INVLIST_CLONE; | |
6763 | ||
6d47fb3d KW |
6764 | SvCUR_set(new_invlist, length); /* This isn't done automatically */ |
6765 | Copy(SvPVX(invlist), SvPVX(new_invlist), length, char); | |
6766 | ||
25e94a65 KW |
6767 | return new_invlist; |
6768 | } | |
6769 | ||
3c234b35 | 6770 | #ifndef PERL_IN_XSUB_RE |
86f766ab KW |
6771 | void |
6772 | Perl__invlist_subtract(pTHX_ SV* const a, SV* const b, SV** result) | |
25e94a65 | 6773 | { |
dbe7a391 | 6774 | /* Point <result> to an inversion list which consists of all elements in |
4065ba03 KW |
6775 | * <a> that aren't also in <b>. *result should be defined upon input, and |
6776 | * if it points to C<b> its reference count will be decremented. */ | |
25e94a65 | 6777 | |
37e85ffe | 6778 | PERL_ARGS_ASSERT__INVLIST_SUBTRACT; |
60825692 | 6779 | assert(a != b); |
25e94a65 KW |
6780 | |
6781 | /* Subtracting nothing retains the original */ | |
6782 | if (invlist_len(b) == 0) { | |
6783 | ||
2ea86699 KW |
6784 | if (*result == b) { |
6785 | SvREFCNT_dec(b); | |
6786 | } | |
6787 | ||
25e94a65 KW |
6788 | /* If the result is not to be the same variable as the original, create |
6789 | * a copy */ | |
4065ba03 | 6790 | if (*result != a) { |
25e94a65 KW |
6791 | *result = invlist_clone(a); |
6792 | } | |
6793 | } else { | |
6794 | SV *b_copy = invlist_clone(b); | |
37e85ffe | 6795 | _invlist_invert(b_copy); /* Everything not in 'b' */ |
2ea86699 KW |
6796 | |
6797 | if (*result == b) { | |
6798 | SvREFCNT_dec(b); | |
6799 | } | |
6800 | ||
37e85ffe | 6801 | _invlist_intersection(a, b_copy, result); /* Everything in 'a' not in |
25e94a65 KW |
6802 | 'b' */ |
6803 | SvREFCNT_dec(b_copy); | |
6804 | } | |
6805 | ||
25e94a65 KW |
6806 | return; |
6807 | } | |
3c234b35 | 6808 | #endif |
25e94a65 | 6809 | |
f3dc70d1 KW |
6810 | PERL_STATIC_INLINE UV* |
6811 | S_get_invlist_iter_addr(pTHX_ SV* invlist) | |
6812 | { | |
6813 | /* Return the address of the UV that contains the current iteration | |
6814 | * position */ | |
6815 | ||
6816 | PERL_ARGS_ASSERT_GET_INVLIST_ITER_ADDR; | |
6817 | ||
6818 | return (UV *) (SvPVX(invlist) + (INVLIST_ITER_OFFSET * sizeof (UV))); | |
6819 | } | |
6820 | ||
6821 | PERL_STATIC_INLINE void | |
6822 | S_invlist_iterinit(pTHX_ SV* invlist) /* Initialize iterator for invlist */ | |
6823 | { | |
6824 | PERL_ARGS_ASSERT_INVLIST_ITERINIT; | |
6825 | ||
6826 | *get_invlist_iter_addr(invlist) = 0; | |
6827 | } | |
6828 | ||
6829 | STATIC bool | |
6830 | S_invlist_iternext(pTHX_ SV* invlist, UV* start, UV* end) | |
6831 | { | |
dbe7a391 KW |
6832 | /* An C<invlist_iterinit> call on <invlist> must be used to set this up. |
6833 | * This call sets in <*start> and <*end>, the next range in <invlist>. | |
6834 | * Returns <TRUE> if successful and the next call will return the next | |
6835 | * range; <FALSE> if was already at the end of the list. If the latter, | |
6836 | * <*start> and <*end> are unchanged, and the next call to this function | |
6837 | * will start over at the beginning of the list */ | |
6838 | ||
f3dc70d1 KW |
6839 | UV* pos = get_invlist_iter_addr(invlist); |
6840 | UV len = invlist_len(invlist); | |
6841 | UV *array; | |
6842 | ||
6843 | PERL_ARGS_ASSERT_INVLIST_ITERNEXT; | |
6844 | ||
6845 | if (*pos >= len) { | |
6846 | *pos = UV_MAX; /* Force iternit() to be required next time */ | |
6847 | return FALSE; | |
6848 | } | |
6849 | ||
6850 | array = invlist_array(invlist); | |
6851 | ||
6852 | *start = array[(*pos)++]; | |
6853 | ||
6854 | if (*pos >= len) { | |
6855 | *end = UV_MAX; | |
6856 | } | |
6857 | else { | |
6858 | *end = array[(*pos)++] - 1; | |
6859 | } | |
6860 | ||
6861 | return TRUE; | |
6862 | } | |
6863 | ||
b2b97e77 KW |
6864 | #ifndef PERL_IN_XSUB_RE |
6865 | SV * | |
6866 | Perl__invlist_contents(pTHX_ SV* const invlist) | |
6867 | { | |
6868 | /* Get the contents of an inversion list into a string SV so that they can | |
6869 | * be printed out. It uses the format traditionally done for debug tracing | |
6870 | */ | |
6871 | ||
6872 | UV start, end; | |
6873 | SV* output = newSVpvs("\n"); | |
6874 | ||
6875 | PERL_ARGS_ASSERT__INVLIST_CONTENTS; | |
6876 | ||
6877 | invlist_iterinit(invlist); | |
6878 | while (invlist_iternext(invlist, &start, &end)) { | |
6879 | if (end == UV_MAX) { | |
6880 | Perl_sv_catpvf(aTHX_ output, "%04"UVXf"\tINFINITY\n", start); | |
6881 | } | |
6882 | else if (end != start) { | |
6883 | Perl_sv_catpvf(aTHX_ output, "%04"UVXf"\t%04"UVXf"\n", | |
6884 | start, end); | |
6885 | } | |
6886 | else { | |
6887 | Perl_sv_catpvf(aTHX_ output, "%04"UVXf"\n", start); | |
6888 | } | |
6889 | } | |
6890 | ||
6891 | return output; | |
6892 | } | |
6893 | #endif | |
6894 | ||
768318b8 KW |
6895 | #if 0 |
6896 | void | |
6897 | S_invlist_dump(pTHX_ SV* const invlist, const char * const header) | |
6898 | { | |
6899 | /* Dumps out the ranges in an inversion list. The string 'header' | |
6900 | * if present is output on a line before the first range */ | |
6901 | ||
6902 | UV start, end; | |
6903 | ||
6904 | if (header && strlen(header)) { | |
6905 | PerlIO_printf(Perl_debug_log, "%s\n", header); | |
6906 | } | |
6907 | invlist_iterinit(invlist); | |
6908 | while (invlist_iternext(invlist, &start, &end)) { | |
6909 | if (end == UV_MAX) { | |
6910 | PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. INFINITY\n", start); | |
6911 | } | |
6912 | else { | |
6913 | PerlIO_printf(Perl_debug_log, "0x%04"UVXf" .. 0x%04"UVXf"\n", start, end); | |
6914 | } | |
6915 | } | |
6916 | } | |
6917 | #endif | |
6918 | ||
97b14ce7 | 6919 | #undef HEADER_LENGTH |
060b7a35 | 6920 | #undef INVLIST_INITIAL_LENGTH |
005b65ed KW |
6921 | #undef TO_INTERNAL_SIZE |
6922 | #undef FROM_INTERNAL_SIZE | |
f1b67122 KW |
6923 | #undef INVLIST_LEN_OFFSET |
6924 | #undef INVLIST_ZERO_OFFSET | |
f3dc70d1 | 6925 | #undef INVLIST_ITER_OFFSET |
060b7a35 | 6926 | |
d764b54e KW |
6927 | /* End of inversion list object */ |
6928 | ||
a687059c LW |
6929 | /* |
6930 | - reg - regular expression, i.e. main body or parenthesized thing | |
6931 | * | |
6932 | * Caller must absorb opening parenthesis. | |
6933 | * | |
6934 | * Combining parenthesis handling with the base level of regular expression | |
6935 | * is a trifle forced, but the need to tie the tails of the branches to what | |
6936 | * follows makes it hard to avoid. | |
6937 | */ | |
07be1b83 YO |
6938 | #define REGTAIL(x,y,z) regtail((x),(y),(z),depth+1) |
6939 | #ifdef DEBUGGING | |
6940 | #define REGTAIL_STUDY(x,y,z) regtail_study((x),(y),(z),depth+1) | |
6941 | #else | |
6942 | #define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1) | |
6943 | #endif | |
3dab1dad | 6944 | |
76e3520e | 6945 | STATIC regnode * |
3dab1dad | 6946 | S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth) |
c277df42 | 6947 | /* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */ |
a687059c | 6948 | { |
27da23d5 | 6949 | dVAR; |
c277df42 IZ |
6950 | register regnode *ret; /* Will be the head of the group. */ |
6951 | register regnode *br; | |
6952 | register regnode *lastbr; | |
cbbf8932 | 6953 | register regnode *ender = NULL; |
a0d0e21e | 6954 | register I32 parno = 0; |
cbbf8932 | 6955 | I32 flags; |
f7819f85 | 6956 | U32 oregflags = RExC_flags; |
6136c704 AL |
6957 | bool have_branch = 0; |
6958 | bool is_open = 0; | |
594d7033 YO |
6959 | I32 freeze_paren = 0; |
6960 | I32 after_freeze = 0; | |
9d1d55b5 JP |
6961 | |
6962 | /* for (?g), (?gc), and (?o) warnings; warning | |
6963 | about (?c) will warn about (?g) -- japhy */ | |
6964 | ||
6136c704 AL |
6965 | #define WASTED_O 0x01 |
6966 | #define WASTED_G 0x02 | |
6967 | #define WASTED_C 0x04 | |
6968 | #define WASTED_GC (0x02|0x04) | |
cbbf8932 | 6969 | I32 wastedflags = 0x00; |
9d1d55b5 | 6970 | |
fac92740 | 6971 | char * parse_start = RExC_parse; /* MJD */ |
a28509cc | 6972 | char * const oregcomp_parse = RExC_parse; |
a0d0e21e | 6973 | |
3dab1dad | 6974 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
6975 | |
6976 | PERL_ARGS_ASSERT_REG; | |
3dab1dad YO |
6977 | DEBUG_PARSE("reg "); |
6978 | ||
821b33a5 | 6979 | *flagp = 0; /* Tentatively. */ |
a0d0e21e | 6980 | |
9d1d55b5 | 6981 | |
a0d0e21e LW |
6982 | /* Make an OPEN node, if parenthesized. */ |
6983 | if (paren) { | |
e2e6a0f1 YO |
6984 | if ( *RExC_parse == '*') { /* (*VERB:ARG) */ |
6985 | char *start_verb = RExC_parse; | |
6986 | STRLEN verb_len = 0; | |
6987 | char *start_arg = NULL; | |
6988 | unsigned char op = 0; | |
6989 | int argok = 1; | |
6990 | int internal_argval = 0; /* internal_argval is only useful if !argok */ | |
6991 | while ( *RExC_parse && *RExC_parse != ')' ) { | |
6992 | if ( *RExC_parse == ':' ) { | |
6993 | start_arg = RExC_parse + 1; | |
6994 | break; | |
6995 | } | |
6996 | RExC_parse++; | |
6997 | } | |
6998 | ++start_verb; | |
6999 | verb_len = RExC_parse - start_verb; | |
7000 | if ( start_arg ) { | |
7001 | RExC_parse++; | |
7002 | while ( *RExC_parse && *RExC_parse != ')' ) | |
7003 | RExC_parse++; | |
7004 | if ( *RExC_parse != ')' ) | |
7005 | vFAIL("Unterminated verb pattern argument"); | |
7006 | if ( RExC_parse == start_arg ) | |
7007 | start_arg = NULL; | |
7008 | } else { | |
7009 | if ( *RExC_parse != ')' ) | |
7010 | vFAIL("Unterminated verb pattern"); | |
7011 | } | |
5d458dd8 | 7012 | |
e2e6a0f1 YO |
7013 | switch ( *start_verb ) { |
7014 | case 'A': /* (*ACCEPT) */ | |
568a785a | 7015 | if ( memEQs(start_verb,verb_len,"ACCEPT") ) { |
e2e6a0f1 YO |
7016 | op = ACCEPT; |
7017 | internal_argval = RExC_nestroot; | |
7018 | } | |
7019 | break; | |
7020 | case 'C': /* (*COMMIT) */ | |
568a785a | 7021 | if ( memEQs(start_verb,verb_len,"COMMIT") ) |
e2e6a0f1 | 7022 | op = COMMIT; |
e2e6a0f1 YO |
7023 | break; |
7024 | case 'F': /* (*FAIL) */ | |
568a785a | 7025 | if ( verb_len==1 || memEQs(start_verb,verb_len,"FAIL") ) { |
e2e6a0f1 YO |
7026 | op = OPFAIL; |
7027 | argok = 0; | |
7028 | } | |
7029 | break; | |
5d458dd8 YO |
7030 | case ':': /* (*:NAME) */ |
7031 | case 'M': /* (*MARK:NAME) */ | |
568a785a | 7032 | if ( verb_len==0 || memEQs(start_verb,verb_len,"MARK") ) { |
e2e6a0f1 | 7033 | op = MARKPOINT; |
5d458dd8 YO |
7034 | argok = -1; |
7035 | } | |
7036 | break; | |
7037 | case 'P': /* (*PRUNE) */ | |
568a785a | 7038 | if ( memEQs(start_verb,verb_len,"PRUNE") ) |
5d458dd8 | 7039 | op = PRUNE; |
e2e6a0f1 | 7040 | break; |
5d458dd8 | 7041 | case 'S': /* (*SKIP) */ |
568a785a | 7042 | if ( memEQs(start_verb,verb_len,"SKIP") ) |
5d458dd8 YO |
7043 | op = SKIP; |
7044 | break; | |
7045 | case 'T': /* (*THEN) */ | |
7046 | /* [19:06] <TimToady> :: is then */ | |
568a785a | 7047 | if ( memEQs(start_verb,verb_len,"THEN") ) { |
5d458dd8 YO |
7048 | op = CUTGROUP; |
7049 | RExC_seen |= REG_SEEN_CUTGROUP; | |
7050 | } | |
e2e6a0f1 YO |
7051 | break; |
7052 | } | |
7053 | if ( ! op ) { | |
7054 | RExC_parse++; | |
7055 | vFAIL3("Unknown verb pattern '%.*s'", | |
7056 | verb_len, start_verb); | |
7057 | } | |
7058 | if ( argok ) { | |
7059 | if ( start_arg && internal_argval ) { | |
7060 | vFAIL3("Verb pattern '%.*s' may not have an argument", | |
7061 | verb_len, start_verb); | |
7062 | } else if ( argok < 0 && !start_arg ) { | |
7063 | vFAIL3("Verb pattern '%.*s' has a mandatory argument", | |
7064 | verb_len, start_verb); | |
7065 | } else { | |
7066 | ret = reganode(pRExC_state, op, internal_argval); | |
7067 | if ( ! internal_argval && ! SIZE_ONLY ) { | |
7068 | if (start_arg) { | |
7069 | SV *sv = newSVpvn( start_arg, RExC_parse - start_arg); | |
7070 | ARG(ret) = add_data( pRExC_state, 1, "S" ); | |
f8fc2ecf | 7071 | RExC_rxi->data->data[ARG(ret)]=(void*)sv; |
e2e6a0f1 YO |
7072 | ret->flags = 0; |
7073 | } else { | |
7074 | ret->flags = 1; | |
7075 | } | |
7076 | } | |
7077 | } | |
7078 | if (!internal_argval) | |
7079 | RExC_seen |= REG_SEEN_VERBARG; | |
7080 | } else if ( start_arg ) { | |
7081 | vFAIL3("Verb pattern '%.*s' may not have an argument", | |
7082 | verb_len, start_verb); | |
7083 | } else { | |
7084 | ret = reg_node(pRExC_state, op); | |
7085 | } | |
7086 | nextchar(pRExC_state); | |
7087 | return ret; | |
7088 | } else | |
fac92740 | 7089 | if (*RExC_parse == '?') { /* (?...) */ |
6136c704 | 7090 | bool is_logical = 0; |
a28509cc | 7091 | const char * const seqstart = RExC_parse; |
fb85c044 | 7092 | bool has_use_defaults = FALSE; |
ca9dfc88 | 7093 | |
830247a4 IZ |
7094 | RExC_parse++; |
7095 | paren = *RExC_parse++; | |
c277df42 | 7096 | ret = NULL; /* For look-ahead/behind. */ |
a0d0e21e | 7097 | switch (paren) { |
894be9b7 | 7098 | |
1f1031fe YO |
7099 | case 'P': /* (?P...) variants for those used to PCRE/Python */ |
7100 | paren = *RExC_parse++; | |
7101 | if ( paren == '<') /* (?P<...>) named capture */ | |
7102 | goto named_capture; | |
7103 | else if (paren == '>') { /* (?P>name) named recursion */ | |
7104 | goto named_recursion; | |
7105 | } | |
7106 | else if (paren == '=') { /* (?P=...) named backref */ | |
7107 | /* this pretty much dupes the code for \k<NAME> in regatom(), if | |
7108 | you change this make sure you change that */ | |
7109 | char* name_start = RExC_parse; | |
7110 | U32 num = 0; | |
7111 | SV *sv_dat = reg_scan_name(pRExC_state, | |
7112 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
7113 | if (RExC_parse == name_start || *RExC_parse != ')') | |
7114 | vFAIL2("Sequence %.3s... not terminated",parse_start); | |
7115 | ||
7116 | if (!SIZE_ONLY) { | |
7117 | num = add_data( pRExC_state, 1, "S" ); | |
7118 | RExC_rxi->data->data[num]=(void*)sv_dat; | |
5a5094bd | 7119 | SvREFCNT_inc_simple_void(sv_dat); |
1f1031fe YO |
7120 | } |
7121 | RExC_sawback = 1; | |
4444fd9f KW |
7122 | ret = reganode(pRExC_state, |
7123 | ((! FOLD) | |
7124 | ? NREF | |
2f7f8cb1 KW |
7125 | : (MORE_ASCII_RESTRICTED) |
7126 | ? NREFFA | |
7127 | : (AT_LEAST_UNI_SEMANTICS) | |
7128 | ? NREFFU | |
7129 | : (LOC) | |
7130 | ? NREFFL | |
7131 | : NREFF), | |
4444fd9f | 7132 | num); |
1f1031fe YO |
7133 | *flagp |= HASWIDTH; |
7134 | ||
7135 | Set_Node_Offset(ret, parse_start+1); | |
7136 | Set_Node_Cur_Length(ret); /* MJD */ | |
7137 | ||
7138 | nextchar(pRExC_state); | |
7139 | return ret; | |
7140 | } | |
57b84237 YO |
7141 | RExC_parse++; |
7142 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
7143 | /*NOTREACHED*/ | |
7144 | case '<': /* (?<...) */ | |
b81d288d | 7145 | if (*RExC_parse == '!') |
c277df42 | 7146 | paren = ','; |
0a4db386 | 7147 | else if (*RExC_parse != '=') |
1f1031fe | 7148 | named_capture: |
0a4db386 | 7149 | { /* (?<...>) */ |
81714fb9 | 7150 | char *name_start; |
894be9b7 | 7151 | SV *svname; |
81714fb9 YO |
7152 | paren= '>'; |
7153 | case '\'': /* (?'...') */ | |
7154 | name_start= RExC_parse; | |
0a4db386 YO |
7155 | svname = reg_scan_name(pRExC_state, |
7156 | SIZE_ONLY ? /* reverse test from the others */ | |
7157 | REG_RSN_RETURN_NAME : | |
7158 | REG_RSN_RETURN_NULL); | |
57b84237 YO |
7159 | if (RExC_parse == name_start) { |
7160 | RExC_parse++; | |
7161 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
7162 | /*NOTREACHED*/ | |
7163 | } | |
81714fb9 YO |
7164 | if (*RExC_parse != paren) |
7165 | vFAIL2("Sequence (?%c... not terminated", | |
7166 | paren=='>' ? '<' : paren); | |
7167 | if (SIZE_ONLY) { | |
e62cc96a YO |
7168 | HE *he_str; |
7169 | SV *sv_dat = NULL; | |
486ec47a | 7170 | if (!svname) /* shouldn't happen */ |
894be9b7 YO |
7171 | Perl_croak(aTHX_ |
7172 | "panic: reg_scan_name returned NULL"); | |
81714fb9 YO |
7173 | if (!RExC_paren_names) { |
7174 | RExC_paren_names= newHV(); | |
ad64d0ec | 7175 | sv_2mortal(MUTABLE_SV(RExC_paren_names)); |
1f1031fe YO |
7176 | #ifdef DEBUGGING |
7177 | RExC_paren_name_list= newAV(); | |
ad64d0ec | 7178 | sv_2mortal(MUTABLE_SV(RExC_paren_name_list)); |
1f1031fe | 7179 | #endif |
81714fb9 YO |
7180 | } |
7181 | he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 ); | |
e62cc96a | 7182 | if ( he_str ) |
81714fb9 | 7183 | sv_dat = HeVAL(he_str); |
e62cc96a | 7184 | if ( ! sv_dat ) { |
81714fb9 | 7185 | /* croak baby croak */ |
e62cc96a YO |
7186 | Perl_croak(aTHX_ |
7187 | "panic: paren_name hash element allocation failed"); | |
7188 | } else if ( SvPOK(sv_dat) ) { | |
76a476f9 YO |
7189 | /* (?|...) can mean we have dupes so scan to check |
7190 | its already been stored. Maybe a flag indicating | |
7191 | we are inside such a construct would be useful, | |
7192 | but the arrays are likely to be quite small, so | |
7193 | for now we punt -- dmq */ | |
7194 | IV count = SvIV(sv_dat); | |
7195 | I32 *pv = (I32*)SvPVX(sv_dat); | |
7196 | IV i; | |
7197 | for ( i = 0 ; i < count ; i++ ) { | |
7198 | if ( pv[i] == RExC_npar ) { | |
7199 | count = 0; | |
7200 | break; | |
7201 | } | |
7202 | } | |
7203 | if ( count ) { | |
7204 | pv = (I32*)SvGROW(sv_dat, SvCUR(sv_dat) + sizeof(I32)+1); | |
7205 | SvCUR_set(sv_dat, SvCUR(sv_dat) + sizeof(I32)); | |
7206 | pv[count] = RExC_npar; | |
3a92e6ae | 7207 | SvIV_set(sv_dat, SvIVX(sv_dat) + 1); |
76a476f9 | 7208 | } |
81714fb9 YO |
7209 | } else { |
7210 | (void)SvUPGRADE(sv_dat,SVt_PVNV); | |
7211 | sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32)); | |
7212 | SvIOK_on(sv_dat); | |
3ec35e0f | 7213 | SvIV_set(sv_dat, 1); |
e62cc96a | 7214 | } |
1f1031fe | 7215 | #ifdef DEBUGGING |
17a3c617 | 7216 | /* Yes this does cause a memory leak in debugging Perls */ |
1f1031fe YO |
7217 | if (!av_store(RExC_paren_name_list, RExC_npar, SvREFCNT_inc(svname))) |
7218 | SvREFCNT_dec(svname); | |
7219 | #endif | |
e62cc96a | 7220 | |
81714fb9 YO |
7221 | /*sv_dump(sv_dat);*/ |
7222 | } | |
7223 | nextchar(pRExC_state); | |
7224 | paren = 1; | |
7225 | goto capturing_parens; | |
7226 | } | |
7227 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
b57e4118 | 7228 | RExC_in_lookbehind++; |
830247a4 | 7229 | RExC_parse++; |
fac92740 | 7230 | case '=': /* (?=...) */ |
89c6a13e | 7231 | RExC_seen_zerolen++; |
5c3fa2e7 | 7232 | break; |
fac92740 | 7233 | case '!': /* (?!...) */ |
830247a4 | 7234 | RExC_seen_zerolen++; |
e2e6a0f1 YO |
7235 | if (*RExC_parse == ')') { |
7236 | ret=reg_node(pRExC_state, OPFAIL); | |
7237 | nextchar(pRExC_state); | |
7238 | return ret; | |
7239 | } | |
594d7033 YO |
7240 | break; |
7241 | case '|': /* (?|...) */ | |
7242 | /* branch reset, behave like a (?:...) except that | |
7243 | buffers in alternations share the same numbers */ | |
7244 | paren = ':'; | |
7245 | after_freeze = freeze_paren = RExC_npar; | |
7246 | break; | |
fac92740 MJD |
7247 | case ':': /* (?:...) */ |
7248 | case '>': /* (?>...) */ | |
a0d0e21e | 7249 | break; |
fac92740 MJD |
7250 | case '$': /* (?$...) */ |
7251 | case '@': /* (?@...) */ | |
8615cb43 | 7252 | vFAIL2("Sequence (?%c...) not implemented", (int)paren); |
a0d0e21e | 7253 | break; |
fac92740 | 7254 | case '#': /* (?#...) */ |
830247a4 IZ |
7255 | while (*RExC_parse && *RExC_parse != ')') |
7256 | RExC_parse++; | |
7257 | if (*RExC_parse != ')') | |
c277df42 | 7258 | FAIL("Sequence (?#... not terminated"); |
830247a4 | 7259 | nextchar(pRExC_state); |
a0d0e21e LW |
7260 | *flagp = TRYAGAIN; |
7261 | return NULL; | |
894be9b7 YO |
7262 | case '0' : /* (?0) */ |
7263 | case 'R' : /* (?R) */ | |
7264 | if (*RExC_parse != ')') | |
6bda09f9 | 7265 | FAIL("Sequence (?R) not terminated"); |
1a147d38 | 7266 | ret = reg_node(pRExC_state, GOSTART); |
a3b492c3 | 7267 | *flagp |= POSTPONED; |
7f69552c YO |
7268 | nextchar(pRExC_state); |
7269 | return ret; | |
7270 | /*notreached*/ | |
894be9b7 YO |
7271 | { /* named and numeric backreferences */ |
7272 | I32 num; | |
894be9b7 YO |
7273 | case '&': /* (?&NAME) */ |
7274 | parse_start = RExC_parse - 1; | |
1f1031fe | 7275 | named_recursion: |
894be9b7 | 7276 | { |
0a4db386 YO |
7277 | SV *sv_dat = reg_scan_name(pRExC_state, |
7278 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
7279 | num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0; | |
894be9b7 YO |
7280 | } |
7281 | goto gen_recurse_regop; | |
7282 | /* NOT REACHED */ | |
542fa716 YO |
7283 | case '+': |
7284 | if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) { | |
7285 | RExC_parse++; | |
7286 | vFAIL("Illegal pattern"); | |
7287 | } | |
7288 | goto parse_recursion; | |
7289 | /* NOT REACHED*/ | |
7290 | case '-': /* (?-1) */ | |
7291 | if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) { | |
7292 | RExC_parse--; /* rewind to let it be handled later */ | |
7293 | goto parse_flags; | |
7294 | } | |
7295 | /*FALLTHROUGH */ | |
6bda09f9 YO |
7296 | case '1': case '2': case '3': case '4': /* (?1) */ |
7297 | case '5': case '6': case '7': case '8': case '9': | |
7298 | RExC_parse--; | |
542fa716 | 7299 | parse_recursion: |
894be9b7 YO |
7300 | num = atoi(RExC_parse); |
7301 | parse_start = RExC_parse - 1; /* MJD */ | |
542fa716 YO |
7302 | if (*RExC_parse == '-') |
7303 | RExC_parse++; | |
6bda09f9 YO |
7304 | while (isDIGIT(*RExC_parse)) |
7305 | RExC_parse++; | |
7306 | if (*RExC_parse!=')') | |
7307 | vFAIL("Expecting close bracket"); | |
686b73d4 | 7308 | |
894be9b7 | 7309 | gen_recurse_regop: |
542fa716 YO |
7310 | if ( paren == '-' ) { |
7311 | /* | |
7312 | Diagram of capture buffer numbering. | |
7313 | Top line is the normal capture buffer numbers | |
3b753521 | 7314 | Bottom line is the negative indexing as from |
542fa716 YO |
7315 | the X (the (?-2)) |
7316 | ||
7317 | + 1 2 3 4 5 X 6 7 | |
7318 | /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/ | |
7319 | - 5 4 3 2 1 X x x | |
7320 | ||
7321 | */ | |
7322 | num = RExC_npar + num; | |
7323 | if (num < 1) { | |
7324 | RExC_parse++; | |
7325 | vFAIL("Reference to nonexistent group"); | |
7326 | } | |
7327 | } else if ( paren == '+' ) { | |
7328 | num = RExC_npar + num - 1; | |
7329 | } | |
7330 | ||
1a147d38 | 7331 | ret = reganode(pRExC_state, GOSUB, num); |
6bda09f9 YO |
7332 | if (!SIZE_ONLY) { |
7333 | if (num > (I32)RExC_rx->nparens) { | |
7334 | RExC_parse++; | |
7335 | vFAIL("Reference to nonexistent group"); | |
7336 | } | |
40d049e4 | 7337 | ARG2L_SET( ret, RExC_recurse_count++); |
6bda09f9 | 7338 | RExC_emit++; |
226de585 | 7339 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, |
acff02b8 | 7340 | "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret))); |
894be9b7 | 7341 | } else { |
6bda09f9 | 7342 | RExC_size++; |
6bda09f9 | 7343 | } |
0a4db386 | 7344 | RExC_seen |= REG_SEEN_RECURSE; |
6bda09f9 | 7345 | Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */ |
58663417 RGS |
7346 | Set_Node_Offset(ret, parse_start); /* MJD */ |
7347 | ||
a3b492c3 | 7348 | *flagp |= POSTPONED; |
6bda09f9 YO |
7349 | nextchar(pRExC_state); |
7350 | return ret; | |
894be9b7 YO |
7351 | } /* named and numeric backreferences */ |
7352 | /* NOT REACHED */ | |
7353 | ||
fac92740 | 7354 | case '?': /* (??...) */ |
6136c704 | 7355 | is_logical = 1; |
57b84237 YO |
7356 | if (*RExC_parse != '{') { |
7357 | RExC_parse++; | |
7358 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
7359 | /*NOTREACHED*/ | |
7360 | } | |
a3b492c3 | 7361 | *flagp |= POSTPONED; |
830247a4 | 7362 | paren = *RExC_parse++; |
0f5d15d6 | 7363 | /* FALL THROUGH */ |
fac92740 | 7364 | case '{': /* (?{...}) */ |
c277df42 | 7365 | { |
2eccd3b2 NC |
7366 | I32 count = 1; |
7367 | U32 n = 0; | |
c277df42 | 7368 | char c; |
830247a4 | 7369 | char *s = RExC_parse; |
c277df42 | 7370 | |
830247a4 IZ |
7371 | RExC_seen_zerolen++; |
7372 | RExC_seen |= REG_SEEN_EVAL; | |
7373 | while (count && (c = *RExC_parse)) { | |
6136c704 AL |
7374 | if (c == '\\') { |
7375 | if (RExC_parse[1]) | |
7376 | RExC_parse++; | |
7377 | } | |
b81d288d | 7378 | else if (c == '{') |
c277df42 | 7379 | count++; |
b81d288d | 7380 | else if (c == '}') |
c277df42 | 7381 | count--; |
830247a4 | 7382 | RExC_parse++; |
c277df42 | 7383 | } |
6136c704 | 7384 | if (*RExC_parse != ')') { |
686b73d4 | 7385 | RExC_parse = s; |
b45f050a JF |
7386 | vFAIL("Sequence (?{...}) not terminated or not {}-balanced"); |
7387 | } | |
c277df42 | 7388 | if (!SIZE_ONLY) { |
f3548bdc | 7389 | PAD *pad; |
6136c704 AL |
7390 | OP_4tree *sop, *rop; |
7391 | SV * const sv = newSVpvn(s, RExC_parse - 1 - s); | |
c277df42 | 7392 | |
569233ed SB |
7393 | ENTER; |
7394 | Perl_save_re_context(aTHX); | |
d59a8b3e | 7395 | rop = Perl_sv_compile_2op_is_broken(aTHX_ sv, &sop, "re", &pad); |
9b978d73 DM |
7396 | sop->op_private |= OPpREFCOUNTED; |
7397 | /* re_dup will OpREFCNT_inc */ | |
7398 | OpREFCNT_set(sop, 1); | |
569233ed | 7399 | LEAVE; |
c277df42 | 7400 | |
830247a4 | 7401 | n = add_data(pRExC_state, 3, "nop"); |
f8fc2ecf YO |
7402 | RExC_rxi->data->data[n] = (void*)rop; |
7403 | RExC_rxi->data->data[n+1] = (void*)sop; | |
7404 | RExC_rxi->data->data[n+2] = (void*)pad; | |
c277df42 | 7405 | SvREFCNT_dec(sv); |
a0ed51b3 | 7406 | } |
e24b16f9 | 7407 | else { /* First pass */ |
830247a4 | 7408 | if (PL_reginterp_cnt < ++RExC_seen_evals |
923e4eb5 | 7409 | && IN_PERL_RUNTIME) |
2cd61cdb IZ |
7410 | /* No compiled RE interpolated, has runtime |
7411 | components ===> unsafe. */ | |
7412 | FAIL("Eval-group not allowed at runtime, use re 'eval'"); | |
5b61d3f7 | 7413 | if (PL_tainting && PL_tainted) |
cc6b7395 | 7414 | FAIL("Eval-group in insecure regular expression"); |
54df2634 | 7415 | #if PERL_VERSION > 8 |
923e4eb5 | 7416 | if (IN_PERL_COMPILETIME) |
b5c19bd7 | 7417 | PL_cv_has_eval = 1; |
54df2634 | 7418 | #endif |
c277df42 | 7419 | } |
b5c19bd7 | 7420 | |
830247a4 | 7421 | nextchar(pRExC_state); |
6136c704 | 7422 | if (is_logical) { |
830247a4 | 7423 | ret = reg_node(pRExC_state, LOGICAL); |
0f5d15d6 IZ |
7424 | if (!SIZE_ONLY) |
7425 | ret->flags = 2; | |
3dab1dad | 7426 | REGTAIL(pRExC_state, ret, reganode(pRExC_state, EVAL, n)); |
fac92740 | 7427 | /* deal with the length of this later - MJD */ |
0f5d15d6 IZ |
7428 | return ret; |
7429 | } | |
ccb2c380 MP |
7430 | ret = reganode(pRExC_state, EVAL, n); |
7431 | Set_Node_Length(ret, RExC_parse - parse_start + 1); | |
7432 | Set_Node_Offset(ret, parse_start); | |
7433 | return ret; | |
c277df42 | 7434 | } |
fac92740 | 7435 | case '(': /* (?(?{...})...) and (?(?=...)...) */ |
c277df42 | 7436 | { |
0a4db386 | 7437 | int is_define= 0; |
fac92740 | 7438 | if (RExC_parse[0] == '?') { /* (?(?...)) */ |
b81d288d AB |
7439 | if (RExC_parse[1] == '=' || RExC_parse[1] == '!' |
7440 | || RExC_parse[1] == '<' | |
830247a4 | 7441 | || RExC_parse[1] == '{') { /* Lookahead or eval. */ |
c277df42 | 7442 | I32 flag; |
686b73d4 | 7443 | |
830247a4 | 7444 | ret = reg_node(pRExC_state, LOGICAL); |
0f5d15d6 IZ |
7445 | if (!SIZE_ONLY) |
7446 | ret->flags = 1; | |
3dab1dad | 7447 | REGTAIL(pRExC_state, ret, reg(pRExC_state, 1, &flag,depth+1)); |
c277df42 | 7448 | goto insert_if; |
b81d288d | 7449 | } |
a0ed51b3 | 7450 | } |
0a4db386 YO |
7451 | else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */ |
7452 | || RExC_parse[0] == '\'' ) /* (?('NAME')...) */ | |
7453 | { | |
7454 | char ch = RExC_parse[0] == '<' ? '>' : '\''; | |
7455 | char *name_start= RExC_parse++; | |
2eccd3b2 | 7456 | U32 num = 0; |
0a4db386 YO |
7457 | SV *sv_dat=reg_scan_name(pRExC_state, |
7458 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
7459 | if (RExC_parse == name_start || *RExC_parse != ch) | |
7460 | vFAIL2("Sequence (?(%c... not terminated", | |
7461 | (ch == '>' ? '<' : ch)); | |
7462 | RExC_parse++; | |
7463 | if (!SIZE_ONLY) { | |
7464 | num = add_data( pRExC_state, 1, "S" ); | |
f8fc2ecf | 7465 | RExC_rxi->data->data[num]=(void*)sv_dat; |
5a5094bd | 7466 | SvREFCNT_inc_simple_void(sv_dat); |
0a4db386 YO |
7467 | } |
7468 | ret = reganode(pRExC_state,NGROUPP,num); | |
7469 | goto insert_if_check_paren; | |
7470 | } | |
7471 | else if (RExC_parse[0] == 'D' && | |
7472 | RExC_parse[1] == 'E' && | |
7473 | RExC_parse[2] == 'F' && | |
7474 | RExC_parse[3] == 'I' && | |
7475 | RExC_parse[4] == 'N' && | |
7476 | RExC_parse[5] == 'E') | |
7477 | { | |
7478 | ret = reganode(pRExC_state,DEFINEP,0); | |
7479 | RExC_parse +=6 ; | |
7480 | is_define = 1; | |
7481 | goto insert_if_check_paren; | |
7482 | } | |
7483 | else if (RExC_parse[0] == 'R') { | |
7484 | RExC_parse++; | |
7485 | parno = 0; | |
7486 | if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) { | |
7487 | parno = atoi(RExC_parse++); | |
7488 | while (isDIGIT(*RExC_parse)) | |
7489 | RExC_parse++; | |
7490 | } else if (RExC_parse[0] == '&') { | |
7491 | SV *sv_dat; | |
7492 | RExC_parse++; | |
7493 | sv_dat = reg_scan_name(pRExC_state, | |
7494 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
7495 | parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0; | |
7496 | } | |
1a147d38 | 7497 | ret = reganode(pRExC_state,INSUBP,parno); |
0a4db386 YO |
7498 | goto insert_if_check_paren; |
7499 | } | |
830247a4 | 7500 | else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) { |
fac92740 | 7501 | /* (?(1)...) */ |
6136c704 | 7502 | char c; |
830247a4 | 7503 | parno = atoi(RExC_parse++); |
c277df42 | 7504 | |
830247a4 IZ |
7505 | while (isDIGIT(*RExC_parse)) |
7506 | RExC_parse++; | |
fac92740 | 7507 | ret = reganode(pRExC_state, GROUPP, parno); |
2af232bd | 7508 | |
0a4db386 | 7509 | insert_if_check_paren: |
830247a4 | 7510 | if ((c = *nextchar(pRExC_state)) != ')') |
b45f050a | 7511 | vFAIL("Switch condition not recognized"); |
c277df42 | 7512 | insert_if: |
3dab1dad YO |
7513 | REGTAIL(pRExC_state, ret, reganode(pRExC_state, IFTHEN, 0)); |
7514 | br = regbranch(pRExC_state, &flags, 1,depth+1); | |
c277df42 | 7515 | if (br == NULL) |
830247a4 | 7516 | br = reganode(pRExC_state, LONGJMP, 0); |
c277df42 | 7517 | else |
3dab1dad | 7518 | REGTAIL(pRExC_state, br, reganode(pRExC_state, LONGJMP, 0)); |
830247a4 | 7519 | c = *nextchar(pRExC_state); |
d1b80229 IZ |
7520 | if (flags&HASWIDTH) |
7521 | *flagp |= HASWIDTH; | |
c277df42 | 7522 | if (c == '|') { |
0a4db386 YO |
7523 | if (is_define) |
7524 | vFAIL("(?(DEFINE)....) does not allow branches"); | |
830247a4 | 7525 | lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */ |
3dab1dad YO |
7526 | regbranch(pRExC_state, &flags, 1,depth+1); |
7527 | REGTAIL(pRExC_state, ret, lastbr); | |
d1b80229 IZ |
7528 | if (flags&HASWIDTH) |
7529 | *flagp |= HASWIDTH; | |
830247a4 | 7530 | c = *nextchar(pRExC_state); |
a0ed51b3 LW |
7531 | } |
7532 | else | |
c277df42 IZ |
7533 | lastbr = NULL; |
7534 | if (c != ')') | |
8615cb43 | 7535 | vFAIL("Switch (?(condition)... contains too many branches"); |
830247a4 | 7536 | ender = reg_node(pRExC_state, TAIL); |
3dab1dad | 7537 | REGTAIL(pRExC_state, br, ender); |
c277df42 | 7538 | if (lastbr) { |
3dab1dad YO |
7539 | REGTAIL(pRExC_state, lastbr, ender); |
7540 | REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); | |
a0ed51b3 LW |
7541 | } |
7542 | else | |
3dab1dad | 7543 | REGTAIL(pRExC_state, ret, ender); |
3b57cd43 YO |
7544 | RExC_size++; /* XXX WHY do we need this?!! |
7545 | For large programs it seems to be required | |
7546 | but I can't figure out why. -- dmq*/ | |
c277df42 | 7547 | return ret; |
a0ed51b3 LW |
7548 | } |
7549 | else { | |
830247a4 | 7550 | vFAIL2("Unknown switch condition (?(%.2s", RExC_parse); |
c277df42 IZ |
7551 | } |
7552 | } | |
1b1626e4 | 7553 | case 0: |
830247a4 | 7554 | RExC_parse--; /* for vFAIL to print correctly */ |
8615cb43 | 7555 | vFAIL("Sequence (? incomplete"); |
1b1626e4 | 7556 | break; |
85508812 KW |
7557 | case DEFAULT_PAT_MOD: /* Use default flags with the exceptions |
7558 | that follow */ | |
fb85c044 KW |
7559 | has_use_defaults = TRUE; |
7560 | STD_PMMOD_FLAGS_CLEAR(&RExC_flags); | |
e40e74fe KW |
7561 | set_regex_charset(&RExC_flags, (RExC_utf8 || RExC_uni_semantics) |
7562 | ? REGEX_UNICODE_CHARSET | |
7563 | : REGEX_DEPENDS_CHARSET); | |
fb85c044 | 7564 | goto parse_flags; |
a0d0e21e | 7565 | default: |
cde0cee5 YO |
7566 | --RExC_parse; |
7567 | parse_flags: /* (?i) */ | |
7568 | { | |
7569 | U32 posflags = 0, negflags = 0; | |
7570 | U32 *flagsp = &posflags; | |
f6a766d5 | 7571 | char has_charset_modifier = '\0'; |
295c2f7d KW |
7572 | regex_charset cs = (RExC_utf8 || RExC_uni_semantics) |
7573 | ? REGEX_UNICODE_CHARSET | |
7574 | : REGEX_DEPENDS_CHARSET; | |
cde0cee5 YO |
7575 | |
7576 | while (*RExC_parse) { | |
7577 | /* && strchr("iogcmsx", *RExC_parse) */ | |
9d1d55b5 JP |
7578 | /* (?g), (?gc) and (?o) are useless here |
7579 | and must be globally applied -- japhy */ | |
cde0cee5 YO |
7580 | switch (*RExC_parse) { |
7581 | CASE_STD_PMMOD_FLAGS_PARSE_SET(flagsp); | |
9de15fec | 7582 | case LOCALE_PAT_MOD: |
f6a766d5 KW |
7583 | if (has_charset_modifier) { |
7584 | goto excess_modifier; | |
7585 | } | |
7586 | else if (flagsp == &negflags) { | |
9442e3b8 | 7587 | goto neg_modifier; |
9de15fec | 7588 | } |
a62b1201 | 7589 | cs = REGEX_LOCALE_CHARSET; |
f6a766d5 | 7590 | has_charset_modifier = LOCALE_PAT_MOD; |
4624b182 | 7591 | RExC_contains_locale = 1; |
9de15fec KW |
7592 | break; |
7593 | case UNICODE_PAT_MOD: | |
f6a766d5 KW |
7594 | if (has_charset_modifier) { |
7595 | goto excess_modifier; | |
7596 | } | |
7597 | else if (flagsp == &negflags) { | |
9442e3b8 | 7598 | goto neg_modifier; |
9de15fec | 7599 | } |
a62b1201 | 7600 | cs = REGEX_UNICODE_CHARSET; |
f6a766d5 | 7601 | has_charset_modifier = UNICODE_PAT_MOD; |
9de15fec | 7602 | break; |
cfaf538b | 7603 | case ASCII_RESTRICT_PAT_MOD: |
f6a766d5 | 7604 | if (flagsp == &negflags) { |
9442e3b8 | 7605 | goto neg_modifier; |
cfaf538b | 7606 | } |
f6a766d5 KW |
7607 | if (has_charset_modifier) { |
7608 | if (cs != REGEX_ASCII_RESTRICTED_CHARSET) { | |
7609 | goto excess_modifier; | |
7610 | } | |
2f7f8cb1 | 7611 | /* Doubled modifier implies more restricted */ |
f6a766d5 KW |
7612 | cs = REGEX_ASCII_MORE_RESTRICTED_CHARSET; |
7613 | } | |
2f7f8cb1 KW |
7614 | else { |
7615 | cs = REGEX_ASCII_RESTRICTED_CHARSET; | |
7616 | } | |
f6a766d5 | 7617 | has_charset_modifier = ASCII_RESTRICT_PAT_MOD; |
cfaf538b | 7618 | break; |
50e91148 | 7619 | case DEPENDS_PAT_MOD: |
9442e3b8 | 7620 | if (has_use_defaults) { |
9de15fec | 7621 | goto fail_modifiers; |
f6a766d5 | 7622 | } |
9442e3b8 KW |
7623 | else if (flagsp == &negflags) { |
7624 | goto neg_modifier; | |
7625 | } | |
f6a766d5 KW |
7626 | else if (has_charset_modifier) { |
7627 | goto excess_modifier; | |
9de15fec | 7628 | } |
7b98bc43 KW |
7629 | |
7630 | /* The dual charset means unicode semantics if the | |
7631 | * pattern (or target, not known until runtime) are | |
e40e74fe KW |
7632 | * utf8, or something in the pattern indicates unicode |
7633 | * semantics */ | |
7634 | cs = (RExC_utf8 || RExC_uni_semantics) | |
a62b1201 KW |
7635 | ? REGEX_UNICODE_CHARSET |
7636 | : REGEX_DEPENDS_CHARSET; | |
f6a766d5 | 7637 | has_charset_modifier = DEPENDS_PAT_MOD; |
9de15fec | 7638 | break; |
f6a766d5 KW |
7639 | excess_modifier: |
7640 | RExC_parse++; | |
7641 | if (has_charset_modifier == ASCII_RESTRICT_PAT_MOD) { | |
0c96c706 | 7642 | vFAIL2("Regexp modifier \"%c\" may appear a maximum of twice", ASCII_RESTRICT_PAT_MOD); |
f6a766d5 KW |
7643 | } |
7644 | else if (has_charset_modifier == *(RExC_parse - 1)) { | |
0c96c706 | 7645 | vFAIL2("Regexp modifier \"%c\" may not appear twice", *(RExC_parse - 1)); |
f6a766d5 KW |
7646 | } |
7647 | else { | |
0c96c706 | 7648 | vFAIL3("Regexp modifiers \"%c\" and \"%c\" are mutually exclusive", has_charset_modifier, *(RExC_parse - 1)); |
f6a766d5 KW |
7649 | } |
7650 | /*NOTREACHED*/ | |
9442e3b8 KW |
7651 | neg_modifier: |
7652 | RExC_parse++; | |
7653 | vFAIL2("Regexp modifier \"%c\" may not appear after the \"-\"", *(RExC_parse - 1)); | |
7654 | /*NOTREACHED*/ | |
f7819f85 A |
7655 | case ONCE_PAT_MOD: /* 'o' */ |
7656 | case GLOBAL_PAT_MOD: /* 'g' */ | |
9d1d55b5 | 7657 | if (SIZE_ONLY && ckWARN(WARN_REGEXP)) { |
6136c704 | 7658 | const I32 wflagbit = *RExC_parse == 'o' ? WASTED_O : WASTED_G; |
9d1d55b5 JP |
7659 | if (! (wastedflags & wflagbit) ) { |
7660 | wastedflags |= wflagbit; | |
7661 | vWARN5( | |
7662 | RExC_parse + 1, | |
7663 | "Useless (%s%c) - %suse /%c modifier", | |
7664 | flagsp == &negflags ? "?-" : "?", | |
7665 | *RExC_parse, | |
7666 | flagsp == &negflags ? "don't " : "", | |
7667 | *RExC_parse | |
7668 | ); | |
7669 | } | |
7670 | } | |
cde0cee5 YO |
7671 | break; |
7672 | ||
f7819f85 | 7673 | case CONTINUE_PAT_MOD: /* 'c' */ |
9d1d55b5 | 7674 | if (SIZE_ONLY && ckWARN(WARN_REGEXP)) { |
6136c704 AL |
7675 | if (! (wastedflags & WASTED_C) ) { |
7676 | wastedflags |= WASTED_GC; | |
9d1d55b5 JP |
7677 | vWARN3( |
7678 | RExC_parse + 1, | |
7679 | "Useless (%sc) - %suse /gc modifier", | |
7680 | flagsp == &negflags ? "?-" : "?", | |
7681 | flagsp == &negflags ? "don't " : "" | |
7682 | ); | |
7683 | } | |
7684 | } | |
cde0cee5 | 7685 | break; |
f7819f85 | 7686 | case KEEPCOPY_PAT_MOD: /* 'p' */ |
cde0cee5 | 7687 | if (flagsp == &negflags) { |
668c081a NC |
7688 | if (SIZE_ONLY) |
7689 | ckWARNreg(RExC_parse + 1,"Useless use of (?-p)"); | |
cde0cee5 YO |
7690 | } else { |
7691 | *flagsp |= RXf_PMf_KEEPCOPY; | |
7692 | } | |
7693 | break; | |
7694 | case '-': | |
3b753521 | 7695 | /* A flag is a default iff it is following a minus, so |
fb85c044 KW |
7696 | * if there is a minus, it means will be trying to |
7697 | * re-specify a default which is an error */ | |
7698 | if (has_use_defaults || flagsp == &negflags) { | |
9de15fec | 7699 | fail_modifiers: |
57b84237 YO |
7700 | RExC_parse++; |
7701 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
7702 | /*NOTREACHED*/ | |
7703 | } | |
cde0cee5 YO |
7704 | flagsp = &negflags; |
7705 | wastedflags = 0; /* reset so (?g-c) warns twice */ | |
7706 | break; | |
7707 | case ':': | |
7708 | paren = ':'; | |
7709 | /*FALLTHROUGH*/ | |
7710 | case ')': | |
7711 | RExC_flags |= posflags; | |
7712 | RExC_flags &= ~negflags; | |
a62b1201 | 7713 | set_regex_charset(&RExC_flags, cs); |
f7819f85 A |
7714 | if (paren != ':') { |
7715 | oregflags |= posflags; | |
7716 | oregflags &= ~negflags; | |
a62b1201 | 7717 | set_regex_charset(&oregflags, cs); |
f7819f85 | 7718 | } |
cde0cee5 YO |
7719 | nextchar(pRExC_state); |
7720 | if (paren != ':') { | |
7721 | *flagp = TRYAGAIN; | |
7722 | return NULL; | |
7723 | } else { | |
7724 | ret = NULL; | |
7725 | goto parse_rest; | |
7726 | } | |
7727 | /*NOTREACHED*/ | |
7728 | default: | |
cde0cee5 YO |
7729 | RExC_parse++; |
7730 | vFAIL3("Sequence (%.*s...) not recognized", RExC_parse-seqstart, seqstart); | |
7731 | /*NOTREACHED*/ | |
7732 | } | |
830247a4 | 7733 | ++RExC_parse; |
48c036b1 | 7734 | } |
cde0cee5 | 7735 | }} /* one for the default block, one for the switch */ |
a0d0e21e | 7736 | } |
fac92740 | 7737 | else { /* (...) */ |
81714fb9 | 7738 | capturing_parens: |
830247a4 IZ |
7739 | parno = RExC_npar; |
7740 | RExC_npar++; | |
e2e6a0f1 | 7741 | |
830247a4 | 7742 | ret = reganode(pRExC_state, OPEN, parno); |
e2e6a0f1 YO |
7743 | if (!SIZE_ONLY ){ |
7744 | if (!RExC_nestroot) | |
7745 | RExC_nestroot = parno; | |
c009da3d YO |
7746 | if (RExC_seen & REG_SEEN_RECURSE |
7747 | && !RExC_open_parens[parno-1]) | |
7748 | { | |
e2e6a0f1 | 7749 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, |
40d049e4 YO |
7750 | "Setting open paren #%"IVdf" to %d\n", |
7751 | (IV)parno, REG_NODE_NUM(ret))); | |
e2e6a0f1 YO |
7752 | RExC_open_parens[parno-1]= ret; |
7753 | } | |
6bda09f9 | 7754 | } |
fac92740 MJD |
7755 | Set_Node_Length(ret, 1); /* MJD */ |
7756 | Set_Node_Offset(ret, RExC_parse); /* MJD */ | |
6136c704 | 7757 | is_open = 1; |
a0d0e21e | 7758 | } |
a0ed51b3 | 7759 | } |
fac92740 | 7760 | else /* ! paren */ |
a0d0e21e | 7761 | ret = NULL; |
cde0cee5 YO |
7762 | |
7763 | parse_rest: | |
a0d0e21e | 7764 | /* Pick up the branches, linking them together. */ |
fac92740 | 7765 | parse_start = RExC_parse; /* MJD */ |
3dab1dad | 7766 | br = regbranch(pRExC_state, &flags, 1,depth+1); |
ee91d26e | 7767 | |
fac92740 | 7768 | /* branch_len = (paren != 0); */ |
2af232bd | 7769 | |
a0d0e21e LW |
7770 | if (br == NULL) |
7771 | return(NULL); | |
830247a4 IZ |
7772 | if (*RExC_parse == '|') { |
7773 | if (!SIZE_ONLY && RExC_extralen) { | |
6bda09f9 | 7774 | reginsert(pRExC_state, BRANCHJ, br, depth+1); |
a0ed51b3 | 7775 | } |
fac92740 | 7776 | else { /* MJD */ |
6bda09f9 | 7777 | reginsert(pRExC_state, BRANCH, br, depth+1); |
fac92740 MJD |
7778 | Set_Node_Length(br, paren != 0); |
7779 | Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start); | |
7780 | } | |
c277df42 IZ |
7781 | have_branch = 1; |
7782 | if (SIZE_ONLY) | |
830247a4 | 7783 | RExC_extralen += 1; /* For BRANCHJ-BRANCH. */ |
a0ed51b3 LW |
7784 | } |
7785 | else if (paren == ':') { | |
c277df42 IZ |
7786 | *flagp |= flags&SIMPLE; |
7787 | } | |
6136c704 | 7788 | if (is_open) { /* Starts with OPEN. */ |
3dab1dad | 7789 | REGTAIL(pRExC_state, ret, br); /* OPEN -> first. */ |
a0ed51b3 LW |
7790 | } |
7791 | else if (paren != '?') /* Not Conditional */ | |
a0d0e21e | 7792 | ret = br; |
8ae10a67 | 7793 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
c277df42 | 7794 | lastbr = br; |
830247a4 IZ |
7795 | while (*RExC_parse == '|') { |
7796 | if (!SIZE_ONLY && RExC_extralen) { | |
7797 | ender = reganode(pRExC_state, LONGJMP,0); | |
3dab1dad | 7798 | REGTAIL(pRExC_state, NEXTOPER(NEXTOPER(lastbr)), ender); /* Append to the previous. */ |
c277df42 IZ |
7799 | } |
7800 | if (SIZE_ONLY) | |
830247a4 IZ |
7801 | RExC_extralen += 2; /* Account for LONGJMP. */ |
7802 | nextchar(pRExC_state); | |
594d7033 YO |
7803 | if (freeze_paren) { |
7804 | if (RExC_npar > after_freeze) | |
7805 | after_freeze = RExC_npar; | |
7806 | RExC_npar = freeze_paren; | |
7807 | } | |
3dab1dad | 7808 | br = regbranch(pRExC_state, &flags, 0, depth+1); |
2af232bd | 7809 | |
a687059c | 7810 | if (br == NULL) |
a0d0e21e | 7811 | return(NULL); |
3dab1dad | 7812 | REGTAIL(pRExC_state, lastbr, br); /* BRANCH -> BRANCH. */ |
c277df42 | 7813 | lastbr = br; |
8ae10a67 | 7814 | *flagp |= flags & (SPSTART | HASWIDTH | POSTPONED); |
a0d0e21e LW |
7815 | } |
7816 | ||
c277df42 IZ |
7817 | if (have_branch || paren != ':') { |
7818 | /* Make a closing node, and hook it on the end. */ | |
7819 | switch (paren) { | |
7820 | case ':': | |
830247a4 | 7821 | ender = reg_node(pRExC_state, TAIL); |
c277df42 IZ |
7822 | break; |
7823 | case 1: | |
830247a4 | 7824 | ender = reganode(pRExC_state, CLOSE, parno); |
40d049e4 YO |
7825 | if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) { |
7826 | DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log, | |
7827 | "Setting close paren #%"IVdf" to %d\n", | |
7828 | (IV)parno, REG_NODE_NUM(ender))); | |
7829 | RExC_close_parens[parno-1]= ender; | |
e2e6a0f1 YO |
7830 | if (RExC_nestroot == parno) |
7831 | RExC_nestroot = 0; | |
40d049e4 | 7832 | } |
fac92740 MJD |
7833 | Set_Node_Offset(ender,RExC_parse+1); /* MJD */ |
7834 | Set_Node_Length(ender,1); /* MJD */ | |
c277df42 IZ |
7835 | break; |
7836 | case '<': | |
c277df42 IZ |
7837 | case ',': |
7838 | case '=': | |
7839 | case '!': | |
c277df42 | 7840 | *flagp &= ~HASWIDTH; |
821b33a5 IZ |
7841 | /* FALL THROUGH */ |
7842 | case '>': | |
830247a4 | 7843 | ender = reg_node(pRExC_state, SUCCEED); |
c277df42 IZ |
7844 | break; |
7845 | case 0: | |
830247a4 | 7846 | ender = reg_node(pRExC_state, END); |
40d049e4 YO |
7847 | if (!SIZE_ONLY) { |
7848 | assert(!RExC_opend); /* there can only be one! */ | |
7849 | RExC_opend = ender; | |
7850 | } | |
c277df42 IZ |
7851 | break; |
7852 | } | |
eaf3ca90 | 7853 | REGTAIL(pRExC_state, lastbr, ender); |
a0d0e21e | 7854 | |
9674d46a | 7855 | if (have_branch && !SIZE_ONLY) { |
eaf3ca90 YO |
7856 | if (depth==1) |
7857 | RExC_seen |= REG_TOP_LEVEL_BRANCHES; | |
7858 | ||
c277df42 | 7859 | /* Hook the tails of the branches to the closing node. */ |
9674d46a AL |
7860 | for (br = ret; br; br = regnext(br)) { |
7861 | const U8 op = PL_regkind[OP(br)]; | |
7862 | if (op == BRANCH) { | |
07be1b83 | 7863 | REGTAIL_STUDY(pRExC_state, NEXTOPER(br), ender); |
9674d46a AL |
7864 | } |
7865 | else if (op == BRANCHJ) { | |
07be1b83 | 7866 | REGTAIL_STUDY(pRExC_state, NEXTOPER(NEXTOPER(br)), ender); |
9674d46a | 7867 | } |
c277df42 IZ |
7868 | } |
7869 | } | |
a0d0e21e | 7870 | } |
c277df42 IZ |
7871 | |
7872 | { | |
e1ec3a88 AL |
7873 | const char *p; |
7874 | static const char parens[] = "=!<,>"; | |
c277df42 IZ |
7875 | |
7876 | if (paren && (p = strchr(parens, paren))) { | |
eb160463 | 7877 | U8 node = ((p - parens) % 2) ? UNLESSM : IFMATCH; |
c277df42 IZ |
7878 | int flag = (p - parens) > 1; |
7879 | ||
7880 | if (paren == '>') | |
7881 | node = SUSPEND, flag = 0; | |
6bda09f9 | 7882 | reginsert(pRExC_state, node,ret, depth+1); |
45948336 EP |
7883 | Set_Node_Cur_Length(ret); |
7884 | Set_Node_Offset(ret, parse_start + 1); | |
c277df42 | 7885 | ret->flags = flag; |
07be1b83 | 7886 | REGTAIL_STUDY(pRExC_state, ret, reg_node(pRExC_state, TAIL)); |
c277df42 | 7887 | } |
a0d0e21e LW |
7888 | } |
7889 | ||
7890 | /* Check for proper termination. */ | |
ce3e6498 | 7891 | if (paren) { |
e2509266 | 7892 | RExC_flags = oregflags; |
830247a4 IZ |
7893 | if (RExC_parse >= RExC_end || *nextchar(pRExC_state) != ')') { |
7894 | RExC_parse = oregcomp_parse; | |
380a0633 | 7895 | vFAIL("Unmatched ("); |
ce3e6498 | 7896 | } |
a0ed51b3 | 7897 | } |
830247a4 IZ |
7898 | else if (!paren && RExC_parse < RExC_end) { |
7899 | if (*RExC_parse == ')') { | |
7900 | RExC_parse++; | |
380a0633 | 7901 | vFAIL("Unmatched )"); |
a0ed51b3 LW |
7902 | } |
7903 | else | |
b45f050a | 7904 | FAIL("Junk on end of regexp"); /* "Can't happen". */ |
a0d0e21e LW |
7905 | /* NOTREACHED */ |
7906 | } | |
b57e4118 KW |
7907 | |
7908 | if (RExC_in_lookbehind) { | |
7909 | RExC_in_lookbehind--; | |
7910 | } | |
fd4be6f0 | 7911 | if (after_freeze > RExC_npar) |
594d7033 | 7912 | RExC_npar = after_freeze; |
a0d0e21e | 7913 | return(ret); |
a687059c LW |
7914 | } |
7915 | ||
7916 | /* | |
7917 | - regbranch - one alternative of an | operator | |
7918 | * | |
7919 | * Implements the concatenation operator. | |
7920 | */ | |
76e3520e | 7921 | STATIC regnode * |
3dab1dad | 7922 | S_regbranch(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, I32 first, U32 depth) |
a687059c | 7923 | { |
97aff369 | 7924 | dVAR; |
c277df42 IZ |
7925 | register regnode *ret; |
7926 | register regnode *chain = NULL; | |
7927 | register regnode *latest; | |
7928 | I32 flags = 0, c = 0; | |
3dab1dad | 7929 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
7930 | |
7931 | PERL_ARGS_ASSERT_REGBRANCH; | |
7932 | ||
3dab1dad | 7933 | DEBUG_PARSE("brnc"); |
02daf0ab | 7934 | |
b81d288d | 7935 | if (first) |
c277df42 IZ |
7936 | ret = NULL; |
7937 | else { | |
b81d288d | 7938 | if (!SIZE_ONLY && RExC_extralen) |
830247a4 | 7939 | ret = reganode(pRExC_state, BRANCHJ,0); |
fac92740 | 7940 | else { |
830247a4 | 7941 | ret = reg_node(pRExC_state, BRANCH); |
fac92740 MJD |
7942 | Set_Node_Length(ret, 1); |
7943 | } | |
c277df42 | 7944 | } |
686b73d4 | 7945 | |
b81d288d | 7946 | if (!first && SIZE_ONLY) |
830247a4 | 7947 | RExC_extralen += 1; /* BRANCHJ */ |
b81d288d | 7948 | |
c277df42 | 7949 | *flagp = WORST; /* Tentatively. */ |
a0d0e21e | 7950 | |
830247a4 IZ |
7951 | RExC_parse--; |
7952 | nextchar(pRExC_state); | |
7953 | while (RExC_parse < RExC_end && *RExC_parse != '|' && *RExC_parse != ')') { | |
a0d0e21e | 7954 | flags &= ~TRYAGAIN; |
3dab1dad | 7955 | latest = regpiece(pRExC_state, &flags,depth+1); |
a0d0e21e LW |
7956 | if (latest == NULL) { |
7957 | if (flags & TRYAGAIN) | |
7958 | continue; | |
7959 | return(NULL); | |
a0ed51b3 LW |
7960 | } |
7961 | else if (ret == NULL) | |
c277df42 | 7962 | ret = latest; |
8ae10a67 | 7963 | *flagp |= flags&(HASWIDTH|POSTPONED); |
c277df42 | 7964 | if (chain == NULL) /* First piece. */ |
a0d0e21e LW |
7965 | *flagp |= flags&SPSTART; |
7966 | else { | |
830247a4 | 7967 | RExC_naughty++; |
3dab1dad | 7968 | REGTAIL(pRExC_state, chain, latest); |
a687059c | 7969 | } |
a0d0e21e | 7970 | chain = latest; |
c277df42 IZ |
7971 | c++; |
7972 | } | |
7973 | if (chain == NULL) { /* Loop ran zero times. */ | |
830247a4 | 7974 | chain = reg_node(pRExC_state, NOTHING); |
c277df42 IZ |
7975 | if (ret == NULL) |
7976 | ret = chain; | |
7977 | } | |
7978 | if (c == 1) { | |
7979 | *flagp |= flags&SIMPLE; | |
a0d0e21e | 7980 | } |
a687059c | 7981 | |
d4c19fe8 | 7982 | return ret; |
a687059c LW |
7983 | } |
7984 | ||
7985 | /* | |
7986 | - regpiece - something followed by possible [*+?] | |
7987 | * | |
7988 | * Note that the branching code sequences used for ? and the general cases | |
7989 | * of * and + are somewhat optimized: they use the same NOTHING node as | |
7990 | * both the endmarker for their branch list and the body of the last branch. | |
7991 | * It might seem that this node could be dispensed with entirely, but the | |
7992 | * endmarker role is not redundant. | |
7993 | */ | |
76e3520e | 7994 | STATIC regnode * |
3dab1dad | 7995 | S_regpiece(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
a687059c | 7996 | { |
97aff369 | 7997 | dVAR; |
c277df42 | 7998 | register regnode *ret; |
a0d0e21e LW |
7999 | register char op; |
8000 | register char *next; | |
8001 | I32 flags; | |
1df70142 | 8002 | const char * const origparse = RExC_parse; |
a0d0e21e | 8003 | I32 min; |
c277df42 | 8004 | I32 max = REG_INFTY; |
f19a8d85 | 8005 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 8006 | char *parse_start; |
f19a8d85 | 8007 | #endif |
10edeb5d | 8008 | const char *maxpos = NULL; |
3dab1dad | 8009 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
8010 | |
8011 | PERL_ARGS_ASSERT_REGPIECE; | |
8012 | ||
3dab1dad | 8013 | DEBUG_PARSE("piec"); |
a0d0e21e | 8014 | |
3dab1dad | 8015 | ret = regatom(pRExC_state, &flags,depth+1); |
a0d0e21e LW |
8016 | if (ret == NULL) { |
8017 | if (flags & TRYAGAIN) | |
8018 | *flagp |= TRYAGAIN; | |
8019 | return(NULL); | |
8020 | } | |
8021 | ||
830247a4 | 8022 | op = *RExC_parse; |
a0d0e21e | 8023 | |
830247a4 | 8024 | if (op == '{' && regcurly(RExC_parse)) { |
10edeb5d | 8025 | maxpos = NULL; |
f19a8d85 | 8026 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 8027 | parse_start = RExC_parse; /* MJD */ |
f19a8d85 | 8028 | #endif |
830247a4 | 8029 | next = RExC_parse + 1; |
a0d0e21e LW |
8030 | while (isDIGIT(*next) || *next == ',') { |
8031 | if (*next == ',') { | |
8032 | if (maxpos) | |
8033 | break; | |
8034 | else | |
8035 | maxpos = next; | |
a687059c | 8036 | } |
a0d0e21e LW |
8037 | next++; |
8038 | } | |
8039 | if (*next == '}') { /* got one */ | |
8040 | if (!maxpos) | |
8041 | maxpos = next; | |
830247a4 IZ |
8042 | RExC_parse++; |
8043 | min = atoi(RExC_parse); | |
a0d0e21e LW |
8044 | if (*maxpos == ',') |
8045 | maxpos++; | |
8046 | else | |
830247a4 | 8047 | maxpos = RExC_parse; |
a0d0e21e LW |
8048 | max = atoi(maxpos); |
8049 | if (!max && *maxpos != '0') | |
c277df42 IZ |
8050 | max = REG_INFTY; /* meaning "infinity" */ |
8051 | else if (max >= REG_INFTY) | |
8615cb43 | 8052 | vFAIL2("Quantifier in {,} bigger than %d", REG_INFTY - 1); |
830247a4 IZ |
8053 | RExC_parse = next; |
8054 | nextchar(pRExC_state); | |
a0d0e21e LW |
8055 | |
8056 | do_curly: | |
8057 | if ((flags&SIMPLE)) { | |
830247a4 | 8058 | RExC_naughty += 2 + RExC_naughty / 2; |
6bda09f9 | 8059 | reginsert(pRExC_state, CURLY, ret, depth+1); |
fac92740 MJD |
8060 | Set_Node_Offset(ret, parse_start+1); /* MJD */ |
8061 | Set_Node_Cur_Length(ret); | |
a0d0e21e LW |
8062 | } |
8063 | else { | |
3dab1dad | 8064 | regnode * const w = reg_node(pRExC_state, WHILEM); |
2c2d71f5 JH |
8065 | |
8066 | w->flags = 0; | |
3dab1dad | 8067 | REGTAIL(pRExC_state, ret, w); |
830247a4 | 8068 | if (!SIZE_ONLY && RExC_extralen) { |
6bda09f9 YO |
8069 | reginsert(pRExC_state, LONGJMP,ret, depth+1); |
8070 | reginsert(pRExC_state, NOTHING,ret, depth+1); | |
c277df42 IZ |
8071 | NEXT_OFF(ret) = 3; /* Go over LONGJMP. */ |
8072 | } | |
6bda09f9 | 8073 | reginsert(pRExC_state, CURLYX,ret, depth+1); |
fac92740 MJD |
8074 | /* MJD hk */ |
8075 | Set_Node_Offset(ret, parse_start+1); | |
2af232bd | 8076 | Set_Node_Length(ret, |
fac92740 | 8077 | op == '{' ? (RExC_parse - parse_start) : 1); |
2af232bd | 8078 | |
830247a4 | 8079 | if (!SIZE_ONLY && RExC_extralen) |
c277df42 | 8080 | NEXT_OFF(ret) = 3; /* Go over NOTHING to LONGJMP. */ |
3dab1dad | 8081 | REGTAIL(pRExC_state, ret, reg_node(pRExC_state, NOTHING)); |
c277df42 | 8082 | if (SIZE_ONLY) |
830247a4 IZ |
8083 | RExC_whilem_seen++, RExC_extralen += 3; |
8084 | RExC_naughty += 4 + RExC_naughty; /* compound interest */ | |
a0d0e21e | 8085 | } |
c277df42 | 8086 | ret->flags = 0; |
a0d0e21e LW |
8087 | |
8088 | if (min > 0) | |
821b33a5 IZ |
8089 | *flagp = WORST; |
8090 | if (max > 0) | |
8091 | *flagp |= HASWIDTH; | |
8fa23287 | 8092 | if (max < min) |
8615cb43 | 8093 | vFAIL("Can't do {n,m} with n > m"); |
c277df42 | 8094 | if (!SIZE_ONLY) { |
eb160463 GS |
8095 | ARG1_SET(ret, (U16)min); |
8096 | ARG2_SET(ret, (U16)max); | |
a687059c | 8097 | } |
a687059c | 8098 | |
a0d0e21e | 8099 | goto nest_check; |
a687059c | 8100 | } |
a0d0e21e | 8101 | } |
a687059c | 8102 | |
a0d0e21e LW |
8103 | if (!ISMULT1(op)) { |
8104 | *flagp = flags; | |
a687059c | 8105 | return(ret); |
a0d0e21e | 8106 | } |
bb20fd44 | 8107 | |
c277df42 | 8108 | #if 0 /* Now runtime fix should be reliable. */ |
b45f050a JF |
8109 | |
8110 | /* if this is reinstated, don't forget to put this back into perldiag: | |
8111 | ||
8112 | =item Regexp *+ operand could be empty at {#} in regex m/%s/ | |
8113 | ||
8114 | (F) The part of the regexp subject to either the * or + quantifier | |
8115 | could match an empty string. The {#} shows in the regular | |
8116 | expression about where the problem was discovered. | |
8117 | ||
8118 | */ | |
8119 | ||
bb20fd44 | 8120 | if (!(flags&HASWIDTH) && op != '?') |
b45f050a | 8121 | vFAIL("Regexp *+ operand could be empty"); |
b81d288d | 8122 | #endif |
bb20fd44 | 8123 | |
f19a8d85 | 8124 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 8125 | parse_start = RExC_parse; |
f19a8d85 | 8126 | #endif |
830247a4 | 8127 | nextchar(pRExC_state); |
a0d0e21e | 8128 | |
821b33a5 | 8129 | *flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH); |
a0d0e21e LW |
8130 | |
8131 | if (op == '*' && (flags&SIMPLE)) { | |
6bda09f9 | 8132 | reginsert(pRExC_state, STAR, ret, depth+1); |
c277df42 | 8133 | ret->flags = 0; |
830247a4 | 8134 | RExC_naughty += 4; |
a0d0e21e LW |
8135 | } |
8136 | else if (op == '*') { | |
8137 | min = 0; | |
8138 | goto do_curly; | |
a0ed51b3 LW |
8139 | } |
8140 | else if (op == '+' && (flags&SIMPLE)) { | |
6bda09f9 | 8141 | reginsert(pRExC_state, PLUS, ret, depth+1); |
c277df42 | 8142 | ret->flags = 0; |
830247a4 | 8143 | RExC_naughty += 3; |
a0d0e21e LW |
8144 | } |
8145 | else if (op == '+') { | |
8146 | min = 1; | |
8147 | goto do_curly; | |
a0ed51b3 LW |
8148 | } |
8149 | else if (op == '?') { | |
a0d0e21e LW |
8150 | min = 0; max = 1; |
8151 | goto do_curly; | |
8152 | } | |
8153 | nest_check: | |
668c081a NC |
8154 | if (!SIZE_ONLY && !(flags&(HASWIDTH|POSTPONED)) && max > REG_INFTY/3) { |
8155 | ckWARN3reg(RExC_parse, | |
8156 | "%.*s matches null string many times", | |
8157 | (int)(RExC_parse >= origparse ? RExC_parse - origparse : 0), | |
8158 | origparse); | |
a0d0e21e LW |
8159 | } |
8160 | ||
b9b4dddf | 8161 | if (RExC_parse < RExC_end && *RExC_parse == '?') { |
830247a4 | 8162 | nextchar(pRExC_state); |
6bda09f9 | 8163 | reginsert(pRExC_state, MINMOD, ret, depth+1); |
3dab1dad | 8164 | REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE); |
a0d0e21e | 8165 | } |
b9b4dddf YO |
8166 | #ifndef REG_ALLOW_MINMOD_SUSPEND |
8167 | else | |
8168 | #endif | |
8169 | if (RExC_parse < RExC_end && *RExC_parse == '+') { | |
8170 | regnode *ender; | |
8171 | nextchar(pRExC_state); | |
8172 | ender = reg_node(pRExC_state, SUCCEED); | |
8173 | REGTAIL(pRExC_state, ret, ender); | |
8174 | reginsert(pRExC_state, SUSPEND, ret, depth+1); | |
8175 | ret->flags = 0; | |
8176 | ender = reg_node(pRExC_state, TAIL); | |
8177 | REGTAIL(pRExC_state, ret, ender); | |
8178 | /*ret= ender;*/ | |
8179 | } | |
8180 | ||
8181 | if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) { | |
830247a4 | 8182 | RExC_parse++; |
b45f050a JF |
8183 | vFAIL("Nested quantifiers"); |
8184 | } | |
a0d0e21e LW |
8185 | |
8186 | return(ret); | |
a687059c LW |
8187 | } |
8188 | ||
fc8cd66c | 8189 | |
9d64099b | 8190 | /* reg_namedseq(pRExC_state,UVp, UV depth) |
fc8cd66c YO |
8191 | |
8192 | This is expected to be called by a parser routine that has | |
afefe6bf | 8193 | recognized '\N' and needs to handle the rest. RExC_parse is |
fc8cd66c YO |
8194 | expected to point at the first char following the N at the time |
8195 | of the call. | |
ff3f963a KW |
8196 | |
8197 | The \N may be inside (indicated by valuep not being NULL) or outside a | |
8198 | character class. | |
8199 | ||
8200 | \N may begin either a named sequence, or if outside a character class, mean | |
8201 | to match a non-newline. For non single-quoted regexes, the tokenizer has | |
8202 | attempted to decide which, and in the case of a named sequence converted it | |
8203 | into one of the forms: \N{} (if the sequence is null), or \N{U+c1.c2...}, | |
8204 | where c1... are the characters in the sequence. For single-quoted regexes, | |
8205 | the tokenizer passes the \N sequence through unchanged; this code will not | |
8206 | attempt to determine this nor expand those. The net effect is that if the | |
8207 | beginning of the passed-in pattern isn't '{U+' or there is no '}', it | |
8208 | signals that this \N occurrence means to match a non-newline. | |
8209 | ||
8210 | Only the \N{U+...} form should occur in a character class, for the same | |
8211 | reason that '.' inside a character class means to just match a period: it | |
8212 | just doesn't make sense. | |
fc8cd66c YO |
8213 | |
8214 | If valuep is non-null then it is assumed that we are parsing inside | |
8215 | of a charclass definition and the first codepoint in the resolved | |
8216 | string is returned via *valuep and the routine will return NULL. | |
8217 | In this mode if a multichar string is returned from the charnames | |
ff3f963a | 8218 | handler, a warning will be issued, and only the first char in the |
fc8cd66c YO |
8219 | sequence will be examined. If the string returned is zero length |
8220 | then the value of *valuep is undefined and NON-NULL will | |
8221 | be returned to indicate failure. (This will NOT be a valid pointer | |
8222 | to a regnode.) | |
8223 | ||
ff3f963a KW |
8224 | If valuep is null then it is assumed that we are parsing normal text and a |
8225 | new EXACT node is inserted into the program containing the resolved string, | |
8226 | and a pointer to the new node is returned. But if the string is zero length | |
8227 | a NOTHING node is emitted instead. | |
afefe6bf | 8228 | |
fc8cd66c | 8229 | On success RExC_parse is set to the char following the endbrace. |
ff3f963a | 8230 | Parsing failures will generate a fatal error via vFAIL(...) |
fc8cd66c YO |
8231 | */ |
8232 | STATIC regnode * | |
9d64099b | 8233 | S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep, I32 *flagp, U32 depth) |
fc8cd66c | 8234 | { |
c3c41406 | 8235 | char * endbrace; /* '}' following the name */ |
fc8cd66c | 8236 | regnode *ret = NULL; |
c3c41406 | 8237 | char* p; |
ff3f963a KW |
8238 | |
8239 | GET_RE_DEBUG_FLAGS_DECL; | |
7918f24d NC |
8240 | |
8241 | PERL_ARGS_ASSERT_REG_NAMEDSEQ; | |
ff3f963a KW |
8242 | |
8243 | GET_RE_DEBUG_FLAGS; | |
c3c41406 KW |
8244 | |
8245 | /* The [^\n] meaning of \N ignores spaces and comments under the /x | |
8246 | * modifier. The other meaning does not */ | |
8247 | p = (RExC_flags & RXf_PMf_EXTENDED) | |
8248 | ? regwhite( pRExC_state, RExC_parse ) | |
8249 | : RExC_parse; | |
7918f24d | 8250 | |
ff3f963a | 8251 | /* Disambiguate between \N meaning a named character versus \N meaning |
c3c41406 KW |
8252 | * [^\n]. The former is assumed when it can't be the latter. */ |
8253 | if (*p != '{' || regcurly(p)) { | |
8254 | RExC_parse = p; | |
ff3f963a | 8255 | if (valuep) { |
afefe6bf | 8256 | /* no bare \N in a charclass */ |
ff3f963a KW |
8257 | vFAIL("\\N in a character class must be a named character: \\N{...}"); |
8258 | } | |
afefe6bf RGS |
8259 | nextchar(pRExC_state); |
8260 | ret = reg_node(pRExC_state, REG_ANY); | |
8261 | *flagp |= HASWIDTH|SIMPLE; | |
8262 | RExC_naughty++; | |
8263 | RExC_parse--; | |
8264 | Set_Node_Length(ret, 1); /* MJD */ | |
8265 | return ret; | |
fc8cd66c | 8266 | } |
a4893424 | 8267 | |
c3c41406 KW |
8268 | /* Here, we have decided it should be a named sequence */ |
8269 | ||
8270 | /* The test above made sure that the next real character is a '{', but | |
8271 | * under the /x modifier, it could be separated by space (or a comment and | |
8272 | * \n) and this is not allowed (for consistency with \x{...} and the | |
8273 | * tokenizer handling of \N{NAME}). */ | |
8274 | if (*RExC_parse != '{') { | |
8275 | vFAIL("Missing braces on \\N{}"); | |
8276 | } | |
8277 | ||
ff3f963a | 8278 | RExC_parse++; /* Skip past the '{' */ |
c3c41406 KW |
8279 | |
8280 | if (! (endbrace = strchr(RExC_parse, '}')) /* no trailing brace */ | |
8281 | || ! (endbrace == RExC_parse /* nothing between the {} */ | |
8282 | || (endbrace - RExC_parse >= 2 /* U+ (bad hex is checked below */ | |
8283 | && strnEQ(RExC_parse, "U+", 2)))) /* for a better error msg) */ | |
8284 | { | |
8285 | if (endbrace) RExC_parse = endbrace; /* position msg's '<--HERE' */ | |
8286 | vFAIL("\\N{NAME} must be resolved by the lexer"); | |
8287 | } | |
8288 | ||
ff3f963a KW |
8289 | if (endbrace == RExC_parse) { /* empty: \N{} */ |
8290 | if (! valuep) { | |
8291 | RExC_parse = endbrace + 1; | |
8292 | return reg_node(pRExC_state,NOTHING); | |
a4893424 | 8293 | } |
fc8cd66c | 8294 | |
ff3f963a KW |
8295 | if (SIZE_ONLY) { |
8296 | ckWARNreg(RExC_parse, | |
8297 | "Ignoring zero length \\N{} in character class" | |
8298 | ); | |
8299 | RExC_parse = endbrace + 1; | |
8300 | } | |
8301 | *valuep = 0; | |
8302 | return (regnode *) &RExC_parse; /* Invalid regnode pointer */ | |
fc8cd66c | 8303 | } |
ff3f963a | 8304 | |
62fed28b | 8305 | REQUIRE_UTF8; /* named sequences imply Unicode semantics */ |
ff3f963a KW |
8306 | RExC_parse += 2; /* Skip past the 'U+' */ |
8307 | ||
8308 | if (valuep) { /* In a bracketed char class */ | |
8309 | /* We only pay attention to the first char of | |
8310 | multichar strings being returned. I kinda wonder | |
8311 | if this makes sense as it does change the behaviour | |
8312 | from earlier versions, OTOH that behaviour was broken | |
8313 | as well. XXX Solution is to recharacterize as | |
8314 | [rest-of-class]|multi1|multi2... */ | |
8315 | ||
8316 | STRLEN length_of_hex; | |
8317 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES | |
8318 | | PERL_SCAN_DISALLOW_PREFIX | |
8319 | | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0); | |
8320 | ||
37820adc KW |
8321 | char * endchar = RExC_parse + strcspn(RExC_parse, ".}"); |
8322 | if (endchar < endbrace) { | |
ff3f963a KW |
8323 | ckWARNreg(endchar, "Using just the first character returned by \\N{} in character class"); |
8324 | } | |
ff3f963a KW |
8325 | |
8326 | length_of_hex = (STRLEN)(endchar - RExC_parse); | |
8327 | *valuep = grok_hex(RExC_parse, &length_of_hex, &flags, NULL); | |
8328 | ||
8329 | /* The tokenizer should have guaranteed validity, but it's possible to | |
8330 | * bypass it by using single quoting, so check */ | |
c3c41406 KW |
8331 | if (length_of_hex == 0 |
8332 | || length_of_hex != (STRLEN)(endchar - RExC_parse) ) | |
8333 | { | |
8334 | RExC_parse += length_of_hex; /* Includes all the valid */ | |
8335 | RExC_parse += (RExC_orig_utf8) /* point to after 1st invalid */ | |
8336 | ? UTF8SKIP(RExC_parse) | |
8337 | : 1; | |
8338 | /* Guard against malformed utf8 */ | |
8339 | if (RExC_parse >= endchar) RExC_parse = endchar; | |
8340 | vFAIL("Invalid hexadecimal number in \\N{U+...}"); | |
ff3f963a KW |
8341 | } |
8342 | ||
8343 | RExC_parse = endbrace + 1; | |
8344 | if (endchar == endbrace) return NULL; | |
8345 | ||
8346 | ret = (regnode *) &RExC_parse; /* Invalid regnode pointer */ | |
fc8cd66c | 8347 | } |
ff3f963a | 8348 | else { /* Not a char class */ |
e2a7e165 KW |
8349 | |
8350 | /* What is done here is to convert this to a sub-pattern of the form | |
8351 | * (?:\x{char1}\x{char2}...) | |
8352 | * and then call reg recursively. That way, it retains its atomicness, | |
8353 | * while not having to worry about special handling that some code | |
8354 | * points may have. toke.c has converted the original Unicode values | |
8355 | * to native, so that we can just pass on the hex values unchanged. We | |
8356 | * do have to set a flag to keep recoding from happening in the | |
8357 | * recursion */ | |
8358 | ||
8359 | SV * substitute_parse = newSVpvn_flags("?:", 2, SVf_UTF8|SVs_TEMP); | |
8360 | STRLEN len; | |
ff3f963a KW |
8361 | char *endchar; /* Points to '.' or '}' ending cur char in the input |
8362 | stream */ | |
e2a7e165 KW |
8363 | char *orig_end = RExC_end; |
8364 | ||
8365 | while (RExC_parse < endbrace) { | |
ff3f963a KW |
8366 | |
8367 | /* Code points are separated by dots. If none, there is only one | |
8368 | * code point, and is terminated by the brace */ | |
37820adc | 8369 | endchar = RExC_parse + strcspn(RExC_parse, ".}"); |
ff3f963a | 8370 | |
e2a7e165 KW |
8371 | /* Convert to notation the rest of the code understands */ |
8372 | sv_catpv(substitute_parse, "\\x{"); | |
8373 | sv_catpvn(substitute_parse, RExC_parse, endchar - RExC_parse); | |
8374 | sv_catpv(substitute_parse, "}"); | |
ff3f963a KW |
8375 | |
8376 | /* Point to the beginning of the next character in the sequence. */ | |
8377 | RExC_parse = endchar + 1; | |
ff3f963a | 8378 | } |
e2a7e165 | 8379 | sv_catpv(substitute_parse, ")"); |
ff3f963a | 8380 | |
e2a7e165 | 8381 | RExC_parse = SvPV(substitute_parse, len); |
ff3f963a | 8382 | |
e2a7e165 KW |
8383 | /* Don't allow empty number */ |
8384 | if (len < 8) { | |
8385 | vFAIL("Invalid hexadecimal number in \\N{U+...}"); | |
ff3f963a | 8386 | } |
e2a7e165 | 8387 | RExC_end = RExC_parse + len; |
ff3f963a | 8388 | |
e2a7e165 KW |
8389 | /* The values are Unicode, and therefore not subject to recoding */ |
8390 | RExC_override_recoding = 1; | |
8391 | ||
8392 | ret = reg(pRExC_state, 1, flagp, depth+1); | |
8393 | ||
8394 | RExC_parse = endbrace; | |
8395 | RExC_end = orig_end; | |
8396 | RExC_override_recoding = 0; | |
ff3f963a | 8397 | |
ff3f963a KW |
8398 | nextchar(pRExC_state); |
8399 | } | |
8400 | ||
8401 | return ret; | |
fc8cd66c YO |
8402 | } |
8403 | ||
8404 | ||
9e08bc66 TS |
8405 | /* |
8406 | * reg_recode | |
8407 | * | |
8408 | * It returns the code point in utf8 for the value in *encp. | |
8409 | * value: a code value in the source encoding | |
8410 | * encp: a pointer to an Encode object | |
8411 | * | |
8412 | * If the result from Encode is not a single character, | |
8413 | * it returns U+FFFD (Replacement character) and sets *encp to NULL. | |
8414 | */ | |
8415 | STATIC UV | |
8416 | S_reg_recode(pTHX_ const char value, SV **encp) | |
8417 | { | |
8418 | STRLEN numlen = 1; | |
59cd0e26 | 8419 | SV * const sv = newSVpvn_flags(&value, numlen, SVs_TEMP); |
c86f7df5 | 8420 | const char * const s = *encp ? sv_recode_to_utf8(sv, *encp) : SvPVX(sv); |
9e08bc66 TS |
8421 | const STRLEN newlen = SvCUR(sv); |
8422 | UV uv = UNICODE_REPLACEMENT; | |
8423 | ||
7918f24d NC |
8424 | PERL_ARGS_ASSERT_REG_RECODE; |
8425 | ||
9e08bc66 TS |
8426 | if (newlen) |
8427 | uv = SvUTF8(sv) | |
8428 | ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT) | |
8429 | : *(U8*)s; | |
8430 | ||
8431 | if (!newlen || numlen != newlen) { | |
8432 | uv = UNICODE_REPLACEMENT; | |
c86f7df5 | 8433 | *encp = NULL; |
9e08bc66 TS |
8434 | } |
8435 | return uv; | |
8436 | } | |
8437 | ||
fc8cd66c | 8438 | |
a687059c LW |
8439 | /* |
8440 | - regatom - the lowest level | |
ee9b8eae YO |
8441 | |
8442 | Try to identify anything special at the start of the pattern. If there | |
8443 | is, then handle it as required. This may involve generating a single regop, | |
8444 | such as for an assertion; or it may involve recursing, such as to | |
8445 | handle a () structure. | |
8446 | ||
8447 | If the string doesn't start with something special then we gobble up | |
8448 | as much literal text as we can. | |
8449 | ||
8450 | Once we have been able to handle whatever type of thing started the | |
8451 | sequence, we return. | |
8452 | ||
8453 | Note: we have to be careful with escapes, as they can be both literal | |
8454 | and special, and in the case of \10 and friends can either, depending | |
486ec47a | 8455 | on context. Specifically there are two separate switches for handling |
ee9b8eae YO |
8456 | escape sequences, with the one for handling literal escapes requiring |
8457 | a dummy entry for all of the special escapes that are actually handled | |
8458 | by the other. | |
8459 | */ | |
8460 | ||
76e3520e | 8461 | STATIC regnode * |
3dab1dad | 8462 | S_regatom(pTHX_ RExC_state_t *pRExC_state, I32 *flagp, U32 depth) |
a687059c | 8463 | { |
97aff369 | 8464 | dVAR; |
cbbf8932 | 8465 | register regnode *ret = NULL; |
a0d0e21e | 8466 | I32 flags; |
45948336 | 8467 | char *parse_start = RExC_parse; |
980866de | 8468 | U8 op; |
3dab1dad YO |
8469 | GET_RE_DEBUG_FLAGS_DECL; |
8470 | DEBUG_PARSE("atom"); | |
a0d0e21e LW |
8471 | *flagp = WORST; /* Tentatively. */ |
8472 | ||
7918f24d | 8473 | PERL_ARGS_ASSERT_REGATOM; |
ee9b8eae | 8474 | |
a0d0e21e | 8475 | tryagain: |
f9a79580 | 8476 | switch ((U8)*RExC_parse) { |
a0d0e21e | 8477 | case '^': |
830247a4 IZ |
8478 | RExC_seen_zerolen++; |
8479 | nextchar(pRExC_state); | |
bbe252da | 8480 | if (RExC_flags & RXf_PMf_MULTILINE) |
830247a4 | 8481 | ret = reg_node(pRExC_state, MBOL); |
bbe252da | 8482 | else if (RExC_flags & RXf_PMf_SINGLELINE) |
830247a4 | 8483 | ret = reg_node(pRExC_state, SBOL); |
a0d0e21e | 8484 | else |
830247a4 | 8485 | ret = reg_node(pRExC_state, BOL); |
fac92740 | 8486 | Set_Node_Length(ret, 1); /* MJD */ |
a0d0e21e LW |
8487 | break; |
8488 | case '$': | |
830247a4 | 8489 | nextchar(pRExC_state); |
b81d288d | 8490 | if (*RExC_parse) |
830247a4 | 8491 | RExC_seen_zerolen++; |
bbe252da | 8492 | if (RExC_flags & RXf_PMf_MULTILINE) |
830247a4 | 8493 | ret = reg_node(pRExC_state, MEOL); |
bbe252da | 8494 | else if (RExC_flags & RXf_PMf_SINGLELINE) |
830247a4 | 8495 | ret = reg_node(pRExC_state, SEOL); |
a0d0e21e | 8496 | else |
830247a4 | 8497 | ret = reg_node(pRExC_state, EOL); |
fac92740 | 8498 | Set_Node_Length(ret, 1); /* MJD */ |
a0d0e21e LW |
8499 | break; |
8500 | case '.': | |
830247a4 | 8501 | nextchar(pRExC_state); |
bbe252da | 8502 | if (RExC_flags & RXf_PMf_SINGLELINE) |
ffc61ed2 JH |
8503 | ret = reg_node(pRExC_state, SANY); |
8504 | else | |
8505 | ret = reg_node(pRExC_state, REG_ANY); | |
8506 | *flagp |= HASWIDTH|SIMPLE; | |
830247a4 | 8507 | RExC_naughty++; |
fac92740 | 8508 | Set_Node_Length(ret, 1); /* MJD */ |
a0d0e21e LW |
8509 | break; |
8510 | case '[': | |
b45f050a | 8511 | { |
3dab1dad YO |
8512 | char * const oregcomp_parse = ++RExC_parse; |
8513 | ret = regclass(pRExC_state,depth+1); | |
830247a4 IZ |
8514 | if (*RExC_parse != ']') { |
8515 | RExC_parse = oregcomp_parse; | |
b45f050a JF |
8516 | vFAIL("Unmatched ["); |
8517 | } | |
830247a4 | 8518 | nextchar(pRExC_state); |
a0d0e21e | 8519 | *flagp |= HASWIDTH|SIMPLE; |
fac92740 | 8520 | Set_Node_Length(ret, RExC_parse - oregcomp_parse + 1); /* MJD */ |
a0d0e21e | 8521 | break; |
b45f050a | 8522 | } |
a0d0e21e | 8523 | case '(': |
830247a4 | 8524 | nextchar(pRExC_state); |
3dab1dad | 8525 | ret = reg(pRExC_state, 1, &flags,depth+1); |
a0d0e21e | 8526 | if (ret == NULL) { |
bf93d4cc | 8527 | if (flags & TRYAGAIN) { |
830247a4 | 8528 | if (RExC_parse == RExC_end) { |
bf93d4cc GS |
8529 | /* Make parent create an empty node if needed. */ |
8530 | *flagp |= TRYAGAIN; | |
8531 | return(NULL); | |
8532 | } | |
a0d0e21e | 8533 | goto tryagain; |
bf93d4cc | 8534 | } |
a0d0e21e LW |
8535 | return(NULL); |
8536 | } | |
a3b492c3 | 8537 | *flagp |= flags&(HASWIDTH|SPSTART|SIMPLE|POSTPONED); |
a0d0e21e LW |
8538 | break; |
8539 | case '|': | |
8540 | case ')': | |
8541 | if (flags & TRYAGAIN) { | |
8542 | *flagp |= TRYAGAIN; | |
8543 | return NULL; | |
8544 | } | |
b45f050a | 8545 | vFAIL("Internal urp"); |
a0d0e21e LW |
8546 | /* Supposed to be caught earlier. */ |
8547 | break; | |
85afd4ae | 8548 | case '{': |
830247a4 IZ |
8549 | if (!regcurly(RExC_parse)) { |
8550 | RExC_parse++; | |
85afd4ae CS |
8551 | goto defchar; |
8552 | } | |
8553 | /* FALL THROUGH */ | |
a0d0e21e LW |
8554 | case '?': |
8555 | case '+': | |
8556 | case '*': | |
830247a4 | 8557 | RExC_parse++; |
b45f050a | 8558 | vFAIL("Quantifier follows nothing"); |
a0d0e21e LW |
8559 | break; |
8560 | case '\\': | |
ee9b8eae YO |
8561 | /* Special Escapes |
8562 | ||
8563 | This switch handles escape sequences that resolve to some kind | |
8564 | of special regop and not to literal text. Escape sequnces that | |
8565 | resolve to literal text are handled below in the switch marked | |
8566 | "Literal Escapes". | |
8567 | ||
8568 | Every entry in this switch *must* have a corresponding entry | |
8569 | in the literal escape switch. However, the opposite is not | |
8570 | required, as the default for this switch is to jump to the | |
8571 | literal text handling code. | |
8572 | */ | |
a0a388a1 | 8573 | switch ((U8)*++RExC_parse) { |
ee9b8eae | 8574 | /* Special Escapes */ |
a0d0e21e | 8575 | case 'A': |
830247a4 IZ |
8576 | RExC_seen_zerolen++; |
8577 | ret = reg_node(pRExC_state, SBOL); | |
a0d0e21e | 8578 | *flagp |= SIMPLE; |
ee9b8eae | 8579 | goto finish_meta_pat; |
a0d0e21e | 8580 | case 'G': |
830247a4 IZ |
8581 | ret = reg_node(pRExC_state, GPOS); |
8582 | RExC_seen |= REG_SEEN_GPOS; | |
a0d0e21e | 8583 | *flagp |= SIMPLE; |
ee9b8eae YO |
8584 | goto finish_meta_pat; |
8585 | case 'K': | |
8586 | RExC_seen_zerolen++; | |
8587 | ret = reg_node(pRExC_state, KEEPS); | |
8588 | *flagp |= SIMPLE; | |
37923168 RGS |
8589 | /* XXX:dmq : disabling in-place substitution seems to |
8590 | * be necessary here to avoid cases of memory corruption, as | |
8591 | * with: C<$_="x" x 80; s/x\K/y/> -- rgs | |
8592 | */ | |
8593 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
ee9b8eae | 8594 | goto finish_meta_pat; |
a0d0e21e | 8595 | case 'Z': |
830247a4 | 8596 | ret = reg_node(pRExC_state, SEOL); |
a0d0e21e | 8597 | *flagp |= SIMPLE; |
a1917ab9 | 8598 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
ee9b8eae | 8599 | goto finish_meta_pat; |
b85d18e9 | 8600 | case 'z': |
830247a4 | 8601 | ret = reg_node(pRExC_state, EOS); |
b85d18e9 | 8602 | *flagp |= SIMPLE; |
830247a4 | 8603 | RExC_seen_zerolen++; /* Do not optimize RE away */ |
ee9b8eae | 8604 | goto finish_meta_pat; |
4a2d328f | 8605 | case 'C': |
f33976b4 DB |
8606 | ret = reg_node(pRExC_state, CANY); |
8607 | RExC_seen |= REG_SEEN_CANY; | |
a0ed51b3 | 8608 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 8609 | goto finish_meta_pat; |
a0ed51b3 | 8610 | case 'X': |
830247a4 | 8611 | ret = reg_node(pRExC_state, CLUMP); |
a0ed51b3 | 8612 | *flagp |= HASWIDTH; |
ee9b8eae | 8613 | goto finish_meta_pat; |
a0d0e21e | 8614 | case 'w': |
980866de KW |
8615 | switch (get_regex_charset(RExC_flags)) { |
8616 | case REGEX_LOCALE_CHARSET: | |
8617 | op = ALNUML; | |
8618 | break; | |
8619 | case REGEX_UNICODE_CHARSET: | |
8620 | op = ALNUMU; | |
8621 | break; | |
cfaf538b | 8622 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8623 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8624 | op = ALNUMA; |
8625 | break; | |
980866de KW |
8626 | case REGEX_DEPENDS_CHARSET: |
8627 | op = ALNUM; | |
8628 | break; | |
8629 | default: | |
8630 | goto bad_charset; | |
a12cf05f | 8631 | } |
980866de | 8632 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 8633 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 8634 | goto finish_meta_pat; |
a0d0e21e | 8635 | case 'W': |
980866de KW |
8636 | switch (get_regex_charset(RExC_flags)) { |
8637 | case REGEX_LOCALE_CHARSET: | |
8638 | op = NALNUML; | |
8639 | break; | |
8640 | case REGEX_UNICODE_CHARSET: | |
8641 | op = NALNUMU; | |
8642 | break; | |
cfaf538b | 8643 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8644 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8645 | op = NALNUMA; |
8646 | break; | |
980866de KW |
8647 | case REGEX_DEPENDS_CHARSET: |
8648 | op = NALNUM; | |
8649 | break; | |
8650 | default: | |
8651 | goto bad_charset; | |
a12cf05f | 8652 | } |
980866de | 8653 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 8654 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 8655 | goto finish_meta_pat; |
a0d0e21e | 8656 | case 'b': |
830247a4 IZ |
8657 | RExC_seen_zerolen++; |
8658 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
63ac0dad KW |
8659 | switch (get_regex_charset(RExC_flags)) { |
8660 | case REGEX_LOCALE_CHARSET: | |
8661 | op = BOUNDL; | |
8662 | break; | |
8663 | case REGEX_UNICODE_CHARSET: | |
8664 | op = BOUNDU; | |
8665 | break; | |
cfaf538b | 8666 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8667 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8668 | op = BOUNDA; |
8669 | break; | |
63ac0dad KW |
8670 | case REGEX_DEPENDS_CHARSET: |
8671 | op = BOUND; | |
8672 | break; | |
8673 | default: | |
8674 | goto bad_charset; | |
a12cf05f | 8675 | } |
63ac0dad | 8676 | ret = reg_node(pRExC_state, op); |
b988e673 | 8677 | FLAGS(ret) = get_regex_charset(RExC_flags); |
a0d0e21e | 8678 | *flagp |= SIMPLE; |
5024bc2d KW |
8679 | if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') { |
8680 | ckWARNregdep(RExC_parse, "\"\\b{\" is deprecated; use \"\\b\\{\" instead"); | |
8681 | } | |
ee9b8eae | 8682 | goto finish_meta_pat; |
a0d0e21e | 8683 | case 'B': |
830247a4 IZ |
8684 | RExC_seen_zerolen++; |
8685 | RExC_seen |= REG_SEEN_LOOKBEHIND; | |
63ac0dad KW |
8686 | switch (get_regex_charset(RExC_flags)) { |
8687 | case REGEX_LOCALE_CHARSET: | |
8688 | op = NBOUNDL; | |
8689 | break; | |
8690 | case REGEX_UNICODE_CHARSET: | |
8691 | op = NBOUNDU; | |
8692 | break; | |
cfaf538b | 8693 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8694 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8695 | op = NBOUNDA; |
8696 | break; | |
63ac0dad KW |
8697 | case REGEX_DEPENDS_CHARSET: |
8698 | op = NBOUND; | |
8699 | break; | |
8700 | default: | |
8701 | goto bad_charset; | |
a12cf05f | 8702 | } |
63ac0dad | 8703 | ret = reg_node(pRExC_state, op); |
b988e673 | 8704 | FLAGS(ret) = get_regex_charset(RExC_flags); |
a0d0e21e | 8705 | *flagp |= SIMPLE; |
5024bc2d KW |
8706 | if (! SIZE_ONLY && (U8) *(RExC_parse + 1) == '{') { |
8707 | ckWARNregdep(RExC_parse, "\"\\B{\" is deprecated; use \"\\B\\{\" instead"); | |
8708 | } | |
ee9b8eae | 8709 | goto finish_meta_pat; |
a0d0e21e | 8710 | case 's': |
980866de KW |
8711 | switch (get_regex_charset(RExC_flags)) { |
8712 | case REGEX_LOCALE_CHARSET: | |
8713 | op = SPACEL; | |
8714 | break; | |
8715 | case REGEX_UNICODE_CHARSET: | |
8716 | op = SPACEU; | |
8717 | break; | |
cfaf538b | 8718 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8719 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8720 | op = SPACEA; |
8721 | break; | |
980866de KW |
8722 | case REGEX_DEPENDS_CHARSET: |
8723 | op = SPACE; | |
8724 | break; | |
8725 | default: | |
8726 | goto bad_charset; | |
a12cf05f | 8727 | } |
980866de | 8728 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 8729 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 8730 | goto finish_meta_pat; |
a0d0e21e | 8731 | case 'S': |
980866de KW |
8732 | switch (get_regex_charset(RExC_flags)) { |
8733 | case REGEX_LOCALE_CHARSET: | |
8734 | op = NSPACEL; | |
8735 | break; | |
8736 | case REGEX_UNICODE_CHARSET: | |
8737 | op = NSPACEU; | |
8738 | break; | |
cfaf538b | 8739 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8740 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8741 | op = NSPACEA; |
8742 | break; | |
980866de KW |
8743 | case REGEX_DEPENDS_CHARSET: |
8744 | op = NSPACE; | |
8745 | break; | |
8746 | default: | |
8747 | goto bad_charset; | |
a12cf05f | 8748 | } |
980866de | 8749 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 8750 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 8751 | goto finish_meta_pat; |
a0d0e21e | 8752 | case 'd': |
56ae17b4 KW |
8753 | switch (get_regex_charset(RExC_flags)) { |
8754 | case REGEX_LOCALE_CHARSET: | |
8755 | op = DIGITL; | |
8756 | break; | |
cfaf538b | 8757 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8758 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8759 | op = DIGITA; |
8760 | break; | |
56ae17b4 KW |
8761 | case REGEX_DEPENDS_CHARSET: /* No difference between these */ |
8762 | case REGEX_UNICODE_CHARSET: | |
8763 | op = DIGIT; | |
8764 | break; | |
8765 | default: | |
8766 | goto bad_charset; | |
6ab9ea91 | 8767 | } |
56ae17b4 | 8768 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 8769 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 8770 | goto finish_meta_pat; |
a0d0e21e | 8771 | case 'D': |
56ae17b4 KW |
8772 | switch (get_regex_charset(RExC_flags)) { |
8773 | case REGEX_LOCALE_CHARSET: | |
8774 | op = NDIGITL; | |
8775 | break; | |
cfaf538b | 8776 | case REGEX_ASCII_RESTRICTED_CHARSET: |
2f7f8cb1 | 8777 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
cfaf538b KW |
8778 | op = NDIGITA; |
8779 | break; | |
56ae17b4 KW |
8780 | case REGEX_DEPENDS_CHARSET: /* No difference between these */ |
8781 | case REGEX_UNICODE_CHARSET: | |
8782 | op = NDIGIT; | |
8783 | break; | |
8784 | default: | |
8785 | goto bad_charset; | |
6ab9ea91 | 8786 | } |
56ae17b4 | 8787 | ret = reg_node(pRExC_state, op); |
a0d0e21e | 8788 | *flagp |= HASWIDTH|SIMPLE; |
ee9b8eae | 8789 | goto finish_meta_pat; |
e1d1eefb YO |
8790 | case 'R': |
8791 | ret = reg_node(pRExC_state, LNBREAK); | |
8792 | *flagp |= HASWIDTH|SIMPLE; | |
8793 | goto finish_meta_pat; | |
8794 | case 'h': | |
8795 | ret = reg_node(pRExC_state, HORIZWS); | |
8796 | *flagp |= HASWIDTH|SIMPLE; | |
8797 | goto finish_meta_pat; | |
8798 | case 'H': | |
8799 | ret = reg_node(pRExC_state, NHORIZWS); | |
8800 | *flagp |= HASWIDTH|SIMPLE; | |
8801 | goto finish_meta_pat; | |
ee9b8eae | 8802 | case 'v': |
e1d1eefb YO |
8803 | ret = reg_node(pRExC_state, VERTWS); |
8804 | *flagp |= HASWIDTH|SIMPLE; | |
ee9b8eae YO |
8805 | goto finish_meta_pat; |
8806 | case 'V': | |
e1d1eefb YO |
8807 | ret = reg_node(pRExC_state, NVERTWS); |
8808 | *flagp |= HASWIDTH|SIMPLE; | |
ee9b8eae | 8809 | finish_meta_pat: |
830247a4 | 8810 | nextchar(pRExC_state); |
fac92740 | 8811 | Set_Node_Length(ret, 2); /* MJD */ |
ee9b8eae | 8812 | break; |
a14b48bc LW |
8813 | case 'p': |
8814 | case 'P': | |
686b73d4 | 8815 | { |
3dab1dad | 8816 | char* const oldregxend = RExC_end; |
d008bc60 | 8817 | #ifdef DEBUGGING |
ccb2c380 | 8818 | char* parse_start = RExC_parse - 2; |
d008bc60 | 8819 | #endif |
a14b48bc | 8820 | |
830247a4 | 8821 | if (RExC_parse[1] == '{') { |
3568d838 | 8822 | /* a lovely hack--pretend we saw [\pX] instead */ |
830247a4 IZ |
8823 | RExC_end = strchr(RExC_parse, '}'); |
8824 | if (!RExC_end) { | |
3dab1dad | 8825 | const U8 c = (U8)*RExC_parse; |
830247a4 IZ |
8826 | RExC_parse += 2; |
8827 | RExC_end = oldregxend; | |
0da60cf5 | 8828 | vFAIL2("Missing right brace on \\%c{}", c); |
b45f050a | 8829 | } |
830247a4 | 8830 | RExC_end++; |
a14b48bc | 8831 | } |
af6f566e | 8832 | else { |
830247a4 | 8833 | RExC_end = RExC_parse + 2; |
af6f566e HS |
8834 | if (RExC_end > oldregxend) |
8835 | RExC_end = oldregxend; | |
8836 | } | |
830247a4 | 8837 | RExC_parse--; |
a14b48bc | 8838 | |
3dab1dad | 8839 | ret = regclass(pRExC_state,depth+1); |
a14b48bc | 8840 | |
830247a4 IZ |
8841 | RExC_end = oldregxend; |
8842 | RExC_parse--; | |
ccb2c380 MP |
8843 | |
8844 | Set_Node_Offset(ret, parse_start + 2); | |
8845 | Set_Node_Cur_Length(ret); | |
830247a4 | 8846 | nextchar(pRExC_state); |
a14b48bc LW |
8847 | *flagp |= HASWIDTH|SIMPLE; |
8848 | } | |
8849 | break; | |
fc8cd66c | 8850 | case 'N': |
afefe6bf | 8851 | /* Handle \N and \N{NAME} here and not below because it can be |
fc8cd66c YO |
8852 | multicharacter. join_exact() will join them up later on. |
8853 | Also this makes sure that things like /\N{BLAH}+/ and | |
8854 | \N{BLAH} being multi char Just Happen. dmq*/ | |
8855 | ++RExC_parse; | |
9d64099b | 8856 | ret= reg_namedseq(pRExC_state, NULL, flagp, depth); |
fc8cd66c | 8857 | break; |
0a4db386 | 8858 | case 'k': /* Handle \k<NAME> and \k'NAME' */ |
1f1031fe | 8859 | parse_named_seq: |
81714fb9 YO |
8860 | { |
8861 | char ch= RExC_parse[1]; | |
1f1031fe YO |
8862 | if (ch != '<' && ch != '\'' && ch != '{') { |
8863 | RExC_parse++; | |
8864 | vFAIL2("Sequence %.2s... not terminated",parse_start); | |
81714fb9 | 8865 | } else { |
1f1031fe YO |
8866 | /* this pretty much dupes the code for (?P=...) in reg(), if |
8867 | you change this make sure you change that */ | |
81714fb9 | 8868 | char* name_start = (RExC_parse += 2); |
2eccd3b2 | 8869 | U32 num = 0; |
0a4db386 YO |
8870 | SV *sv_dat = reg_scan_name(pRExC_state, |
8871 | SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA); | |
1f1031fe | 8872 | ch= (ch == '<') ? '>' : (ch == '{') ? '}' : '\''; |
81714fb9 | 8873 | if (RExC_parse == name_start || *RExC_parse != ch) |
1f1031fe YO |
8874 | vFAIL2("Sequence %.3s... not terminated",parse_start); |
8875 | ||
8876 | if (!SIZE_ONLY) { | |
8877 | num = add_data( pRExC_state, 1, "S" ); | |
8878 | RExC_rxi->data->data[num]=(void*)sv_dat; | |
5a5094bd | 8879 | SvREFCNT_inc_simple_void(sv_dat); |
1f1031fe YO |
8880 | } |
8881 | ||
81714fb9 YO |
8882 | RExC_sawback = 1; |
8883 | ret = reganode(pRExC_state, | |
4444fd9f KW |
8884 | ((! FOLD) |
8885 | ? NREF | |
2f7f8cb1 KW |
8886 | : (MORE_ASCII_RESTRICTED) |
8887 | ? NREFFA | |
8888 | : (AT_LEAST_UNI_SEMANTICS) | |
8889 | ? NREFFU | |
8890 | : (LOC) | |
8891 | ? NREFFL | |
8892 | : NREFF), | |
4444fd9f | 8893 | num); |
81714fb9 | 8894 | *flagp |= HASWIDTH; |
1f1031fe | 8895 | |
81714fb9 YO |
8896 | /* override incorrect value set in reganode MJD */ |
8897 | Set_Node_Offset(ret, parse_start+1); | |
8898 | Set_Node_Cur_Length(ret); /* MJD */ | |
8899 | nextchar(pRExC_state); | |
1f1031fe | 8900 | |
81714fb9 YO |
8901 | } |
8902 | break; | |
1f1031fe | 8903 | } |
2bf803e2 | 8904 | case 'g': |
a0d0e21e LW |
8905 | case '1': case '2': case '3': case '4': |
8906 | case '5': case '6': case '7': case '8': case '9': | |
8907 | { | |
c74340f9 | 8908 | I32 num; |
2bf803e2 YO |
8909 | bool isg = *RExC_parse == 'g'; |
8910 | bool isrel = 0; | |
8911 | bool hasbrace = 0; | |
8912 | if (isg) { | |
c74340f9 | 8913 | RExC_parse++; |
2bf803e2 YO |
8914 | if (*RExC_parse == '{') { |
8915 | RExC_parse++; | |
8916 | hasbrace = 1; | |
8917 | } | |
8918 | if (*RExC_parse == '-') { | |
8919 | RExC_parse++; | |
8920 | isrel = 1; | |
8921 | } | |
1f1031fe YO |
8922 | if (hasbrace && !isDIGIT(*RExC_parse)) { |
8923 | if (isrel) RExC_parse--; | |
8924 | RExC_parse -= 2; | |
8925 | goto parse_named_seq; | |
8926 | } } | |
c74340f9 | 8927 | num = atoi(RExC_parse); |
b72d83b2 RGS |
8928 | if (isg && num == 0) |
8929 | vFAIL("Reference to invalid group 0"); | |
c74340f9 | 8930 | if (isrel) { |
5624f11d | 8931 | num = RExC_npar - num; |
c74340f9 YO |
8932 | if (num < 1) |
8933 | vFAIL("Reference to nonexistent or unclosed group"); | |
8934 | } | |
2bf803e2 | 8935 | if (!isg && num > 9 && num >= RExC_npar) |
a0d0e21e LW |
8936 | goto defchar; |
8937 | else { | |
3dab1dad | 8938 | char * const parse_start = RExC_parse - 1; /* MJD */ |
830247a4 IZ |
8939 | while (isDIGIT(*RExC_parse)) |
8940 | RExC_parse++; | |
1f1031fe YO |
8941 | if (parse_start == RExC_parse - 1) |
8942 | vFAIL("Unterminated \\g... pattern"); | |
2bf803e2 YO |
8943 | if (hasbrace) { |
8944 | if (*RExC_parse != '}') | |
8945 | vFAIL("Unterminated \\g{...} pattern"); | |
8946 | RExC_parse++; | |
8947 | } | |
c74340f9 YO |
8948 | if (!SIZE_ONLY) { |
8949 | if (num > (I32)RExC_rx->nparens) | |
8950 | vFAIL("Reference to nonexistent group"); | |
c74340f9 | 8951 | } |
830247a4 | 8952 | RExC_sawback = 1; |
eb160463 | 8953 | ret = reganode(pRExC_state, |
4444fd9f KW |
8954 | ((! FOLD) |
8955 | ? REF | |
2f7f8cb1 KW |
8956 | : (MORE_ASCII_RESTRICTED) |
8957 | ? REFFA | |
8958 | : (AT_LEAST_UNI_SEMANTICS) | |
8959 | ? REFFU | |
8960 | : (LOC) | |
8961 | ? REFFL | |
8962 | : REFF), | |
4444fd9f | 8963 | num); |
a0d0e21e | 8964 | *flagp |= HASWIDTH; |
2af232bd | 8965 | |
fac92740 | 8966 | /* override incorrect value set in reganode MJD */ |
2af232bd | 8967 | Set_Node_Offset(ret, parse_start+1); |
fac92740 | 8968 | Set_Node_Cur_Length(ret); /* MJD */ |
830247a4 IZ |
8969 | RExC_parse--; |
8970 | nextchar(pRExC_state); | |
a0d0e21e LW |
8971 | } |
8972 | } | |
8973 | break; | |
8974 | case '\0': | |
830247a4 | 8975 | if (RExC_parse >= RExC_end) |
b45f050a | 8976 | FAIL("Trailing \\"); |
a0d0e21e LW |
8977 | /* FALL THROUGH */ |
8978 | default: | |
a0288114 | 8979 | /* Do not generate "unrecognized" warnings here, we fall |
c9f97d15 | 8980 | back into the quick-grab loop below */ |
45948336 | 8981 | parse_start--; |
a0d0e21e LW |
8982 | goto defchar; |
8983 | } | |
8984 | break; | |
4633a7c4 LW |
8985 | |
8986 | case '#': | |
bbe252da | 8987 | if (RExC_flags & RXf_PMf_EXTENDED) { |
bcdf7404 | 8988 | if ( reg_skipcomment( pRExC_state ) ) |
4633a7c4 LW |
8989 | goto tryagain; |
8990 | } | |
8991 | /* FALL THROUGH */ | |
8992 | ||
f9a79580 | 8993 | default: |
561784a5 KW |
8994 | |
8995 | parse_start = RExC_parse - 1; | |
8996 | ||
8997 | RExC_parse++; | |
8998 | ||
8999 | defchar: { | |
d669c36c | 9000 | typedef enum { |
9bed422d KW |
9001 | generic_char = 0, |
9002 | char_s, | |
d669c36c KW |
9003 | upsilon_1, |
9004 | upsilon_2, | |
9005 | iota_1, | |
9006 | iota_2, | |
9007 | } char_state; | |
9bed422d | 9008 | char_state latest_char_state = generic_char; |
ba210ebe | 9009 | register STRLEN len; |
58ae7d3f | 9010 | register UV ender; |
a0d0e21e | 9011 | register char *p; |
3dab1dad | 9012 | char *s; |
80aecb99 | 9013 | STRLEN foldlen; |
89ebb4a3 | 9014 | U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf; |
6e326e84 | 9015 | regnode * orig_emit; |
f06dbbb7 | 9016 | |
58ae7d3f | 9017 | ender = 0; |
6e326e84 KW |
9018 | orig_emit = RExC_emit; /* Save the original output node position in |
9019 | case we need to output a different node | |
9020 | type */ | |
eb160463 | 9021 | ret = reg_node(pRExC_state, |
2c2b7f86 KW |
9022 | (U8) ((! FOLD) ? EXACT |
9023 | : (LOC) | |
9024 | ? EXACTFL | |
2f7f8cb1 KW |
9025 | : (MORE_ASCII_RESTRICTED) |
9026 | ? EXACTFA | |
9027 | : (AT_LEAST_UNI_SEMANTICS) | |
9028 | ? EXACTFU | |
9029 | : EXACTF) | |
2c2b7f86 | 9030 | ); |
cd439c50 | 9031 | s = STRING(ret); |
830247a4 IZ |
9032 | for (len = 0, p = RExC_parse - 1; |
9033 | len < 127 && p < RExC_end; | |
a0d0e21e LW |
9034 | len++) |
9035 | { | |
3dab1dad | 9036 | char * const oldp = p; |
5b5a24f7 | 9037 | |
bbe252da | 9038 | if (RExC_flags & RXf_PMf_EXTENDED) |
bcdf7404 | 9039 | p = regwhite( pRExC_state, p ); |
f9a79580 | 9040 | switch ((U8)*p) { |
a0d0e21e LW |
9041 | case '^': |
9042 | case '$': | |
9043 | case '.': | |
9044 | case '[': | |
9045 | case '(': | |
9046 | case ')': | |
9047 | case '|': | |
9048 | goto loopdone; | |
9049 | case '\\': | |
ee9b8eae YO |
9050 | /* Literal Escapes Switch |
9051 | ||
9052 | This switch is meant to handle escape sequences that | |
9053 | resolve to a literal character. | |
9054 | ||
9055 | Every escape sequence that represents something | |
9056 | else, like an assertion or a char class, is handled | |
9057 | in the switch marked 'Special Escapes' above in this | |
9058 | routine, but also has an entry here as anything that | |
9059 | isn't explicitly mentioned here will be treated as | |
9060 | an unescaped equivalent literal. | |
9061 | */ | |
9062 | ||
a0a388a1 | 9063 | switch ((U8)*++p) { |
ee9b8eae YO |
9064 | /* These are all the special escapes. */ |
9065 | case 'A': /* Start assertion */ | |
9066 | case 'b': case 'B': /* Word-boundary assertion*/ | |
9067 | case 'C': /* Single char !DANGEROUS! */ | |
9068 | case 'd': case 'D': /* digit class */ | |
9069 | case 'g': case 'G': /* generic-backref, pos assertion */ | |
e1d1eefb | 9070 | case 'h': case 'H': /* HORIZWS */ |
ee9b8eae YO |
9071 | case 'k': case 'K': /* named backref, keep marker */ |
9072 | case 'N': /* named char sequence */ | |
38a44b82 | 9073 | case 'p': case 'P': /* Unicode property */ |
e1d1eefb | 9074 | case 'R': /* LNBREAK */ |
ee9b8eae | 9075 | case 's': case 'S': /* space class */ |
e1d1eefb | 9076 | case 'v': case 'V': /* VERTWS */ |
ee9b8eae YO |
9077 | case 'w': case 'W': /* word class */ |
9078 | case 'X': /* eXtended Unicode "combining character sequence" */ | |
9079 | case 'z': case 'Z': /* End of line/string assertion */ | |
a0d0e21e LW |
9080 | --p; |
9081 | goto loopdone; | |
ee9b8eae YO |
9082 | |
9083 | /* Anything after here is an escape that resolves to a | |
9084 | literal. (Except digits, which may or may not) | |
9085 | */ | |
a0d0e21e LW |
9086 | case 'n': |
9087 | ender = '\n'; | |
9088 | p++; | |
a687059c | 9089 | break; |
a0d0e21e LW |
9090 | case 'r': |
9091 | ender = '\r'; | |
9092 | p++; | |
a687059c | 9093 | break; |
a0d0e21e LW |
9094 | case 't': |
9095 | ender = '\t'; | |
9096 | p++; | |
a687059c | 9097 | break; |
a0d0e21e LW |
9098 | case 'f': |
9099 | ender = '\f'; | |
9100 | p++; | |
a687059c | 9101 | break; |
a0d0e21e | 9102 | case 'e': |
c7f1f016 | 9103 | ender = ASCII_TO_NATIVE('\033'); |
a0d0e21e | 9104 | p++; |
a687059c | 9105 | break; |
a0d0e21e | 9106 | case 'a': |
c7f1f016 | 9107 | ender = ASCII_TO_NATIVE('\007'); |
a0d0e21e | 9108 | p++; |
a687059c | 9109 | break; |
f0a2b745 KW |
9110 | case 'o': |
9111 | { | |
9112 | STRLEN brace_len = len; | |
00c0cb6d | 9113 | UV result; |
454155d9 KW |
9114 | const char* error_msg; |
9115 | ||
9116 | bool valid = grok_bslash_o(p, | |
9117 | &result, | |
9118 | &brace_len, | |
9119 | &error_msg, | |
9120 | 1); | |
9121 | p += brace_len; | |
9122 | if (! valid) { | |
9123 | RExC_parse = p; /* going to die anyway; point | |
9124 | to exact spot of failure */ | |
f0a2b745 KW |
9125 | vFAIL(error_msg); |
9126 | } | |
00c0cb6d DG |
9127 | else |
9128 | { | |
9129 | ender = result; | |
9130 | } | |
f0a2b745 KW |
9131 | if (PL_encoding && ender < 0x100) { |
9132 | goto recode_encoding; | |
9133 | } | |
9134 | if (ender > 0xff) { | |
62fed28b | 9135 | REQUIRE_UTF8; |
f0a2b745 KW |
9136 | } |
9137 | break; | |
9138 | } | |
a0d0e21e | 9139 | case 'x': |
a0ed51b3 | 9140 | if (*++p == '{') { |
1df70142 | 9141 | char* const e = strchr(p, '}'); |
686b73d4 | 9142 | |
b45f050a | 9143 | if (!e) { |
830247a4 | 9144 | RExC_parse = p + 1; |
b45f050a JF |
9145 | vFAIL("Missing right brace on \\x{}"); |
9146 | } | |
de5f0749 | 9147 | else { |
a4c04bdc NC |
9148 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES |
9149 | | PERL_SCAN_DISALLOW_PREFIX; | |
1df70142 | 9150 | STRLEN numlen = e - p - 1; |
53305cf1 | 9151 | ender = grok_hex(p + 1, &numlen, &flags, NULL); |
aaa80028 | 9152 | if (ender > 0xff) |
62fed28b | 9153 | REQUIRE_UTF8; |
a0ed51b3 LW |
9154 | p = e + 1; |
9155 | } | |
a0ed51b3 LW |
9156 | } |
9157 | else { | |
a4c04bdc | 9158 | I32 flags = PERL_SCAN_DISALLOW_PREFIX; |
1df70142 | 9159 | STRLEN numlen = 2; |
53305cf1 | 9160 | ender = grok_hex(p, &numlen, &flags, NULL); |
a0ed51b3 LW |
9161 | p += numlen; |
9162 | } | |
9e08bc66 TS |
9163 | if (PL_encoding && ender < 0x100) |
9164 | goto recode_encoding; | |
a687059c | 9165 | break; |
a0d0e21e LW |
9166 | case 'c': |
9167 | p++; | |
17a3df4c | 9168 | ender = grok_bslash_c(*p++, UTF, SIZE_ONLY); |
a687059c | 9169 | break; |
a0d0e21e LW |
9170 | case '0': case '1': case '2': case '3':case '4': |
9171 | case '5': case '6': case '7': case '8':case '9': | |
9172 | if (*p == '0' || | |
ca67da41 | 9173 | (isDIGIT(p[1]) && atoi(p) >= RExC_npar)) |
c99e91e9 KW |
9174 | { |
9175 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT; | |
1df70142 | 9176 | STRLEN numlen = 3; |
53305cf1 | 9177 | ender = grok_oct(p, &numlen, &flags, NULL); |
fa1639c5 | 9178 | if (ender > 0xff) { |
62fed28b | 9179 | REQUIRE_UTF8; |
609122bd | 9180 | } |
a0d0e21e LW |
9181 | p += numlen; |
9182 | } | |
9183 | else { | |
9184 | --p; | |
9185 | goto loopdone; | |
a687059c | 9186 | } |
9e08bc66 TS |
9187 | if (PL_encoding && ender < 0x100) |
9188 | goto recode_encoding; | |
9189 | break; | |
9190 | recode_encoding: | |
e2a7e165 | 9191 | if (! RExC_override_recoding) { |
9e08bc66 TS |
9192 | SV* enc = PL_encoding; |
9193 | ender = reg_recode((const char)(U8)ender, &enc); | |
668c081a NC |
9194 | if (!enc && SIZE_ONLY) |
9195 | ckWARNreg(p, "Invalid escape in the specified encoding"); | |
62fed28b | 9196 | REQUIRE_UTF8; |
9e08bc66 | 9197 | } |
a687059c | 9198 | break; |
a0d0e21e | 9199 | case '\0': |
830247a4 | 9200 | if (p >= RExC_end) |
b45f050a | 9201 | FAIL("Trailing \\"); |
a687059c | 9202 | /* FALL THROUGH */ |
a0d0e21e | 9203 | default: |
216bfc0a KW |
9204 | if (!SIZE_ONLY&& isALPHA(*p)) { |
9205 | /* Include any { following the alpha to emphasize | |
9206 | * that it could be part of an escape at some point | |
9207 | * in the future */ | |
9208 | int len = (*(p + 1) == '{') ? 2 : 1; | |
9209 | ckWARN3reg(p + len, "Unrecognized escape \\%.*s passed through", len, p); | |
9210 | } | |
a0ed51b3 | 9211 | goto normal_default; |
a0d0e21e LW |
9212 | } |
9213 | break; | |
a687059c | 9214 | default: |
a0ed51b3 | 9215 | normal_default: |
fd400ab9 | 9216 | if (UTF8_IS_START(*p) && UTF) { |
1df70142 | 9217 | STRLEN numlen; |
5e12f4fb | 9218 | ender = utf8n_to_uvchr((U8*)p, RExC_end - p, |
9f7f3913 | 9219 | &numlen, UTF8_ALLOW_DEFAULT); |
a0ed51b3 LW |
9220 | p += numlen; |
9221 | } | |
9222 | else | |
5b67c30a | 9223 | ender = (U8) *p++; |
a0d0e21e | 9224 | break; |
7e2509c1 KW |
9225 | } /* End of switch on the literal */ |
9226 | ||
6e326e84 KW |
9227 | /* Certain characters are problematic because their folded |
9228 | * length is so different from their original length that it | |
9229 | * isn't handleable by the optimizer. They are therefore not | |
9230 | * placed in an EXACTish node; and are here handled specially. | |
9231 | * (Even if the optimizer handled LATIN_SMALL_LETTER_SHARP_S, | |
9232 | * putting it in a special node keeps regexec from having to | |
9233 | * deal with a non-utf8 multi-char fold */ | |
2f7f8cb1 | 9234 | if (FOLD |
d669c36c | 9235 | && (ender > 255 || (! MORE_ASCII_RESTRICTED && ! LOC))) |
2f7f8cb1 | 9236 | { |
d669c36c KW |
9237 | /* We look for either side of the fold. For example \xDF |
9238 | * folds to 'ss'. We look for both the single character | |
9239 | * \xDF and the sequence 'ss'. When we find something that | |
9240 | * could be one of those, we stop and flush whatever we | |
9241 | * have output so far into the EXACTish node that was being | |
9242 | * built. Then restore the input pointer to what it was. | |
9243 | * regatom will return that EXACT node, and will be called | |
9244 | * again, positioned so the first character is the one in | |
9245 | * question, which we return in a different node type. | |
9246 | * The multi-char folds are a sequence, so the occurrence | |
9247 | * of the first character in that sequence doesn't | |
9248 | * necessarily mean that what follows is the rest of the | |
9249 | * sequence. We keep track of that with a state machine, | |
9250 | * with the state being set to the latest character | |
9251 | * processed before the current one. Most characters will | |
9252 | * set the state to 0, but if one occurs that is part of a | |
9253 | * potential tricky fold sequence, the state is set to that | |
9254 | * character, and the next loop iteration sees if the state | |
9255 | * should progress towards the final folded-from character, | |
9256 | * or if it was a false alarm. If it turns out to be a | |
9257 | * false alarm, the character(s) will be output in a new | |
9258 | * EXACTish node, and join_exact() will later combine them. | |
9259 | * In the case of the 'ss' sequence, which is more common | |
9260 | * and more easily checked, some look-ahead is done to | |
9261 | * save time by ruling-out some false alarms */ | |
9262 | switch (ender) { | |
9263 | default: | |
9bed422d | 9264 | latest_char_state = generic_char; |
d669c36c KW |
9265 | break; |
9266 | case 's': | |
9267 | case 'S': | |
fe5a0c86 | 9268 | case 0x17F: /* LATIN SMALL LETTER LONG S */ |
d669c36c KW |
9269 | if (AT_LEAST_UNI_SEMANTICS) { |
9270 | if (latest_char_state == char_s) { /* 'ss' */ | |
9271 | ender = LATIN_SMALL_LETTER_SHARP_S; | |
9272 | goto do_tricky; | |
9273 | } | |
9274 | else if (p < RExC_end) { | |
9275 | ||
9276 | /* Look-ahead at the next character. If it | |
9277 | * is also an s, we handle as a sharp s | |
9278 | * tricky regnode. */ | |
9279 | if (*p == 's' || *p == 'S') { | |
9280 | ||
9281 | /* But first flush anything in the | |
9282 | * EXACTish buffer */ | |
9283 | if (len != 0) { | |
9284 | p = oldp; | |
9285 | goto loopdone; | |
9286 | } | |
9287 | p++; /* Account for swallowing this | |
9288 | 's' up */ | |
9289 | ender = LATIN_SMALL_LETTER_SHARP_S; | |
9290 | goto do_tricky; | |
9291 | } | |
9292 | /* Here, the next character is not a | |
9293 | * literal 's', but still could | |
9294 | * evaluate to one if part of a \o{}, | |
9295 | * \x or \OCTAL-DIGIT. The minimum | |
9296 | * length required for that is 4, eg | |
9297 | * \x53 or \123 */ | |
9298 | else if (*p == '\\' | |
9299 | && p < RExC_end - 4 | |
9300 | && (isDIGIT(*(p + 1)) | |
9301 | || *(p + 1) == 'x' | |
9302 | || *(p + 1) == 'o' )) | |
9303 | { | |
9304 | ||
9305 | /* Here, it could be an 's', too much | |
9306 | * bother to figure it out here. Flush | |
9307 | * the buffer if any; when come back | |
9308 | * here, set the state so know that the | |
9309 | * previous char was an 's' */ | |
9310 | if (len != 0) { | |
9bed422d | 9311 | latest_char_state = generic_char; |
d669c36c KW |
9312 | p = oldp; |
9313 | goto loopdone; | |
9314 | } | |
9315 | latest_char_state = char_s; | |
9316 | break; | |
9317 | } | |
9318 | } | |
9319 | } | |
6e326e84 | 9320 | |
d669c36c KW |
9321 | /* Here, can't be an 'ss' sequence, or at least not |
9322 | * one that could fold to/from the sharp ss */ | |
9bed422d | 9323 | latest_char_state = generic_char; |
d669c36c KW |
9324 | break; |
9325 | case 0x03C5: /* First char in upsilon series */ | |
13d34caa KW |
9326 | case 0x03A5: /* Also capital UPSILON, which folds to |
9327 | 03C5, and hence exhibits the same | |
9328 | problem */ | |
d669c36c KW |
9329 | if (p < RExC_end - 4) { /* Need >= 4 bytes left */ |
9330 | latest_char_state = upsilon_1; | |
9331 | if (len != 0) { | |
9332 | p = oldp; | |
9333 | goto loopdone; | |
9334 | } | |
9335 | } | |
9336 | else { | |
9bed422d | 9337 | latest_char_state = generic_char; |
d669c36c KW |
9338 | } |
9339 | break; | |
9340 | case 0x03B9: /* First char in iota series */ | |
13d34caa | 9341 | case 0x0399: /* Also capital IOTA */ |
2d74afe1 KW |
9342 | case 0x1FBE: /* GREEK PROSGEGRAMMENI folds to 3B9 */ |
9343 | case 0x0345: /* COMBINING GREEK YPOGEGRAMMENI folds | |
9344 | to 3B9 */ | |
d669c36c KW |
9345 | if (p < RExC_end - 4) { |
9346 | latest_char_state = iota_1; | |
9347 | if (len != 0) { | |
9348 | p = oldp; | |
9349 | goto loopdone; | |
9350 | } | |
9351 | } | |
9352 | else { | |
9bed422d | 9353 | latest_char_state = generic_char; |
d669c36c KW |
9354 | } |
9355 | break; | |
9356 | case 0x0308: | |
9357 | if (latest_char_state == upsilon_1) { | |
9358 | latest_char_state = upsilon_2; | |
9359 | } | |
9360 | else if (latest_char_state == iota_1) { | |
9361 | latest_char_state = iota_2; | |
9362 | } | |
9363 | else { | |
9bed422d | 9364 | latest_char_state = generic_char; |
d669c36c KW |
9365 | } |
9366 | break; | |
9367 | case 0x301: | |
9368 | if (latest_char_state == upsilon_2) { | |
9369 | ender = GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS; | |
9370 | goto do_tricky; | |
9371 | } | |
9372 | else if (latest_char_state == iota_2) { | |
9373 | ender = GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS; | |
9374 | goto do_tricky; | |
9375 | } | |
9bed422d | 9376 | latest_char_state = generic_char; |
d669c36c | 9377 | break; |
6e326e84 | 9378 | |
d669c36c | 9379 | /* These are the tricky fold characters. Flush any |
dddecdc7 KW |
9380 | * buffer first. (When adding to this list, also should |
9381 | * add them to fold_grind.t to make sure get tested) */ | |
d669c36c KW |
9382 | case GREEK_SMALL_LETTER_UPSILON_WITH_DIALYTIKA_AND_TONOS: |
9383 | case GREEK_SMALL_LETTER_IOTA_WITH_DIALYTIKA_AND_TONOS: | |
9384 | case LATIN_SMALL_LETTER_SHARP_S: | |
9385 | case LATIN_CAPITAL_LETTER_SHARP_S: | |
dddecdc7 KW |
9386 | case 0x1FD3: /* GREEK SMALL LETTER IOTA WITH DIALYTIKA AND OXIA */ |
9387 | case 0x1FE3: /* GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND OXIA */ | |
d669c36c KW |
9388 | if (len != 0) { |
9389 | p = oldp; | |
9390 | goto loopdone; | |
9391 | } | |
9392 | /* FALL THROUGH */ | |
9393 | do_tricky: { | |
9394 | char* const oldregxend = RExC_end; | |
9395 | U8 tmpbuf[UTF8_MAXBYTES+1]; | |
9396 | ||
9397 | /* Here, we know we need to generate a special | |
9398 | * regnode, and 'ender' contains the tricky | |
9399 | * character. What's done is to pretend it's in a | |
9400 | * [bracketed] class, and let the code that deals | |
9401 | * with those handle it, as that code has all the | |
9402 | * intelligence necessary. First save the current | |
9403 | * parse state, get rid of the already allocated | |
9404 | * but empty EXACT node that the ANYOFV node will | |
9405 | * replace, and point the parse to a buffer which | |
9406 | * we fill with the character we want the regclass | |
9407 | * code to think is being parsed */ | |
9408 | RExC_emit = orig_emit; | |
9409 | RExC_parse = (char *) tmpbuf; | |
9410 | if (UTF) { | |
9411 | U8 *d = uvchr_to_utf8(tmpbuf, ender); | |
9412 | *d = '\0'; | |
9413 | RExC_end = (char *) d; | |
9414 | } | |
0169d36e KW |
9415 | else { /* ender above 255 already excluded */ |
9416 | tmpbuf[0] = (U8) ender; | |
d669c36c KW |
9417 | tmpbuf[1] = '\0'; |
9418 | RExC_end = RExC_parse + 1; | |
9419 | } | |
6e326e84 | 9420 | |
d669c36c KW |
9421 | ret = regclass(pRExC_state,depth+1); |
9422 | ||
9423 | /* Here, have parsed the buffer. Reset the parse to | |
9424 | * the actual input, and return */ | |
9425 | RExC_end = oldregxend; | |
9426 | RExC_parse = p - 1; | |
6e326e84 | 9427 | |
d669c36c KW |
9428 | Set_Node_Offset(ret, RExC_parse); |
9429 | Set_Node_Cur_Length(ret); | |
9430 | nextchar(pRExC_state); | |
9431 | *flagp |= HASWIDTH|SIMPLE; | |
9432 | return ret; | |
9433 | } | |
6e326e84 KW |
9434 | } |
9435 | } | |
9436 | ||
bcdf7404 YO |
9437 | if ( RExC_flags & RXf_PMf_EXTENDED) |
9438 | p = regwhite( pRExC_state, p ); | |
60a8b682 | 9439 | if (UTF && FOLD) { |
17580e7a KW |
9440 | /* Prime the casefolded buffer. Locale rules, which apply |
9441 | * only to code points < 256, aren't known until execution, | |
9442 | * so for them, just output the original character using | |
9443 | * utf8 */ | |
9444 | if (LOC && ender < 256) { | |
9445 | if (UNI_IS_INVARIANT(ender)) { | |
9446 | *tmpbuf = (U8) ender; | |
9447 | foldlen = 1; | |
9448 | } else { | |
9449 | *tmpbuf = UTF8_TWO_BYTE_HI(ender); | |
9450 | *(tmpbuf + 1) = UTF8_TWO_BYTE_LO(ender); | |
9451 | foldlen = 2; | |
9452 | } | |
9453 | } | |
9454 | else if (isASCII(ender)) { /* Note: Here can't also be LOC | |
9455 | */ | |
2f7f8cb1 | 9456 | ender = toLOWER(ender); |
cd64649c | 9457 | *tmpbuf = (U8) ender; |
2f7f8cb1 KW |
9458 | foldlen = 1; |
9459 | } | |
17580e7a KW |
9460 | else if (! MORE_ASCII_RESTRICTED && ! LOC) { |
9461 | ||
9462 | /* Locale and /aa require more selectivity about the | |
9463 | * fold, so are handled below. Otherwise, here, just | |
9464 | * use the fold */ | |
2f7f8cb1 KW |
9465 | ender = toFOLD_uni(ender, tmpbuf, &foldlen); |
9466 | } | |
9467 | else { | |
17580e7a KW |
9468 | /* Under locale rules or /aa we are not to mix, |
9469 | * respectively, ords < 256 or ASCII with non-. So | |
9470 | * reject folds that mix them, using only the | |
9471 | * non-folded code point. So do the fold to a | |
9472 | * temporary, and inspect each character in it. */ | |
2f7f8cb1 KW |
9473 | U8 trialbuf[UTF8_MAXBYTES_CASE+1]; |
9474 | U8* s = trialbuf; | |
9475 | UV tmpender = toFOLD_uni(ender, trialbuf, &foldlen); | |
9476 | U8* e = s + foldlen; | |
9477 | bool fold_ok = TRUE; | |
9478 | ||
9479 | while (s < e) { | |
17580e7a KW |
9480 | if (isASCII(*s) |
9481 | || (LOC && (UTF8_IS_INVARIANT(*s) | |
9482 | || UTF8_IS_DOWNGRADEABLE_START(*s)))) | |
9483 | { | |
2f7f8cb1 KW |
9484 | fold_ok = FALSE; |
9485 | break; | |
9486 | } | |
9487 | s += UTF8SKIP(s); | |
9488 | } | |
9489 | if (fold_ok) { | |
9490 | Copy(trialbuf, tmpbuf, foldlen, U8); | |
9491 | ender = tmpender; | |
9492 | } | |
9493 | else { | |
9494 | uvuni_to_utf8(tmpbuf, ender); | |
9495 | foldlen = UNISKIP(ender); | |
9496 | } | |
9497 | } | |
60a8b682 | 9498 | } |
bcdf7404 | 9499 | if (p < RExC_end && ISMULT2(p)) { /* Back off on ?+*. */ |
a0d0e21e LW |
9500 | if (len) |
9501 | p = oldp; | |
16ea2a2e | 9502 | else if (UTF) { |
80aecb99 | 9503 | if (FOLD) { |
60a8b682 | 9504 | /* Emit all the Unicode characters. */ |
1df70142 | 9505 | STRLEN numlen; |
80aecb99 JH |
9506 | for (foldbuf = tmpbuf; |
9507 | foldlen; | |
9508 | foldlen -= numlen) { | |
9509 | ender = utf8_to_uvchr(foldbuf, &numlen); | |
9dc45d57 | 9510 | if (numlen > 0) { |
71207a34 | 9511 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
0ebc6274 JH |
9512 | s += unilen; |
9513 | len += unilen; | |
9514 | /* In EBCDIC the numlen | |
9515 | * and unilen can differ. */ | |
9dc45d57 | 9516 | foldbuf += numlen; |
47654450 JH |
9517 | if (numlen >= foldlen) |
9518 | break; | |
9dc45d57 JH |
9519 | } |
9520 | else | |
9521 | break; /* "Can't happen." */ | |
80aecb99 JH |
9522 | } |
9523 | } | |
9524 | else { | |
71207a34 | 9525 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
9ede7db1 | 9526 | if (unilen > 0) { |
0ebc6274 JH |
9527 | s += unilen; |
9528 | len += unilen; | |
9dc45d57 | 9529 | } |
80aecb99 | 9530 | } |
a0ed51b3 | 9531 | } |
a0d0e21e LW |
9532 | else { |
9533 | len++; | |
eb160463 | 9534 | REGC((char)ender, s++); |
a0d0e21e LW |
9535 | } |
9536 | break; | |
a687059c | 9537 | } |
16ea2a2e | 9538 | if (UTF) { |
80aecb99 | 9539 | if (FOLD) { |
60a8b682 | 9540 | /* Emit all the Unicode characters. */ |
1df70142 | 9541 | STRLEN numlen; |
80aecb99 JH |
9542 | for (foldbuf = tmpbuf; |
9543 | foldlen; | |
9544 | foldlen -= numlen) { | |
9545 | ender = utf8_to_uvchr(foldbuf, &numlen); | |
9dc45d57 | 9546 | if (numlen > 0) { |
71207a34 | 9547 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
0ebc6274 JH |
9548 | len += unilen; |
9549 | s += unilen; | |
9550 | /* In EBCDIC the numlen | |
9551 | * and unilen can differ. */ | |
9dc45d57 | 9552 | foldbuf += numlen; |
47654450 JH |
9553 | if (numlen >= foldlen) |
9554 | break; | |
9dc45d57 JH |
9555 | } |
9556 | else | |
9557 | break; | |
80aecb99 JH |
9558 | } |
9559 | } | |
9560 | else { | |
71207a34 | 9561 | const STRLEN unilen = reguni(pRExC_state, ender, s); |
9ede7db1 | 9562 | if (unilen > 0) { |
0ebc6274 JH |
9563 | s += unilen; |
9564 | len += unilen; | |
9dc45d57 | 9565 | } |
80aecb99 JH |
9566 | } |
9567 | len--; | |
a0ed51b3 | 9568 | } |
d669c36c | 9569 | else { |
eb160463 | 9570 | REGC((char)ender, s++); |
d669c36c | 9571 | } |
a0d0e21e | 9572 | } |
7e2509c1 KW |
9573 | loopdone: /* Jumped to when encounters something that shouldn't be in |
9574 | the node */ | |
830247a4 | 9575 | RExC_parse = p - 1; |
fac92740 | 9576 | Set_Node_Cur_Length(ret); /* MJD */ |
830247a4 | 9577 | nextchar(pRExC_state); |
793db0cb JH |
9578 | { |
9579 | /* len is STRLEN which is unsigned, need to copy to signed */ | |
9580 | IV iv = len; | |
9581 | if (iv < 0) | |
9582 | vFAIL("Internal disaster"); | |
9583 | } | |
a0d0e21e LW |
9584 | if (len > 0) |
9585 | *flagp |= HASWIDTH; | |
090f7165 | 9586 | if (len == 1 && UNI_IS_INVARIANT(ender)) |
a0d0e21e | 9587 | *flagp |= SIMPLE; |
686b73d4 | 9588 | |
cd439c50 | 9589 | if (SIZE_ONLY) |
830247a4 | 9590 | RExC_size += STR_SZ(len); |
3dab1dad YO |
9591 | else { |
9592 | STR_LEN(ret) = len; | |
830247a4 | 9593 | RExC_emit += STR_SZ(len); |
07be1b83 | 9594 | } |
3dab1dad | 9595 | } |
a0d0e21e LW |
9596 | break; |
9597 | } | |
a687059c | 9598 | |
a0d0e21e | 9599 | return(ret); |
980866de KW |
9600 | |
9601 | /* Jumped to when an unrecognized character set is encountered */ | |
9602 | bad_charset: | |
9603 | Perl_croak(aTHX_ "panic: Unknown regex character set encoding: %u", get_regex_charset(RExC_flags)); | |
9604 | return(NULL); | |
a687059c LW |
9605 | } |
9606 | ||
873ef191 | 9607 | STATIC char * |
bcdf7404 | 9608 | S_regwhite( RExC_state_t *pRExC_state, char *p ) |
5b5a24f7 | 9609 | { |
bcdf7404 | 9610 | const char *e = RExC_end; |
7918f24d NC |
9611 | |
9612 | PERL_ARGS_ASSERT_REGWHITE; | |
9613 | ||
5b5a24f7 CS |
9614 | while (p < e) { |
9615 | if (isSPACE(*p)) | |
9616 | ++p; | |
9617 | else if (*p == '#') { | |
bcdf7404 | 9618 | bool ended = 0; |
5b5a24f7 | 9619 | do { |
bcdf7404 YO |
9620 | if (*p++ == '\n') { |
9621 | ended = 1; | |
9622 | break; | |
9623 | } | |
9624 | } while (p < e); | |
9625 | if (!ended) | |
9626 | RExC_seen |= REG_SEEN_RUN_ON_COMMENT; | |
5b5a24f7 CS |
9627 | } |
9628 | else | |
9629 | break; | |
9630 | } | |
9631 | return p; | |
9632 | } | |
9633 | ||
b8c5462f JH |
9634 | /* Parse POSIX character classes: [[:foo:]], [[=foo=]], [[.foo.]]. |
9635 | Character classes ([:foo:]) can also be negated ([:^foo:]). | |
9636 | Returns a named class id (ANYOF_XXX) if successful, -1 otherwise. | |
9637 | Equivalence classes ([=foo=]) and composites ([.foo.]) are parsed, | |
beeb77fc | 9638 | but trigger failures because they are currently unimplemented. */ |
9a86a77b JH |
9639 | |
9640 | #define POSIXCC_DONE(c) ((c) == ':') | |
9641 | #define POSIXCC_NOTYET(c) ((c) == '=' || (c) == '.') | |
9642 | #define POSIXCC(c) (POSIXCC_DONE(c) || POSIXCC_NOTYET(c)) | |
9643 | ||
b8c5462f | 9644 | STATIC I32 |
830247a4 | 9645 | S_regpposixcc(pTHX_ RExC_state_t *pRExC_state, I32 value) |
620e46c5 | 9646 | { |
97aff369 | 9647 | dVAR; |
936ed897 | 9648 | I32 namedclass = OOB_NAMEDCLASS; |
620e46c5 | 9649 | |
7918f24d NC |
9650 | PERL_ARGS_ASSERT_REGPPOSIXCC; |
9651 | ||
830247a4 | 9652 | if (value == '[' && RExC_parse + 1 < RExC_end && |
620e46c5 | 9653 | /* I smell either [: or [= or [. -- POSIX has been here, right? */ |
9a86a77b | 9654 | POSIXCC(UCHARAT(RExC_parse))) { |
1df70142 | 9655 | const char c = UCHARAT(RExC_parse); |
097eb12c | 9656 | char* const s = RExC_parse++; |
686b73d4 | 9657 | |
9a86a77b | 9658 | while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != c) |
830247a4 IZ |
9659 | RExC_parse++; |
9660 | if (RExC_parse == RExC_end) | |
620e46c5 | 9661 | /* Grandfather lone [:, [=, [. */ |
830247a4 | 9662 | RExC_parse = s; |
620e46c5 | 9663 | else { |
3dab1dad | 9664 | const char* const t = RExC_parse++; /* skip over the c */ |
80916619 NC |
9665 | assert(*t == c); |
9666 | ||
9a86a77b | 9667 | if (UCHARAT(RExC_parse) == ']') { |
3dab1dad | 9668 | const char *posixcc = s + 1; |
830247a4 | 9669 | RExC_parse++; /* skip over the ending ] */ |
3dab1dad | 9670 | |
b8c5462f | 9671 | if (*s == ':') { |
1df70142 AL |
9672 | const I32 complement = *posixcc == '^' ? *posixcc++ : 0; |
9673 | const I32 skip = t - posixcc; | |
80916619 NC |
9674 | |
9675 | /* Initially switch on the length of the name. */ | |
9676 | switch (skip) { | |
9677 | case 4: | |
3dab1dad YO |
9678 | if (memEQ(posixcc, "word", 4)) /* this is not POSIX, this is the Perl \w */ |
9679 | namedclass = complement ? ANYOF_NALNUM : ANYOF_ALNUM; | |
cc4319de | 9680 | break; |
80916619 NC |
9681 | case 5: |
9682 | /* Names all of length 5. */ | |
9683 | /* alnum alpha ascii blank cntrl digit graph lower | |
9684 | print punct space upper */ | |
9685 | /* Offset 4 gives the best switch position. */ | |
9686 | switch (posixcc[4]) { | |
9687 | case 'a': | |
3dab1dad YO |
9688 | if (memEQ(posixcc, "alph", 4)) /* alpha */ |
9689 | namedclass = complement ? ANYOF_NALPHA : ANYOF_ALPHA; | |
80916619 NC |
9690 | break; |
9691 | case 'e': | |
3dab1dad YO |
9692 | if (memEQ(posixcc, "spac", 4)) /* space */ |
9693 | namedclass = complement ? ANYOF_NPSXSPC : ANYOF_PSXSPC; | |
80916619 NC |
9694 | break; |
9695 | case 'h': | |
3dab1dad YO |
9696 | if (memEQ(posixcc, "grap", 4)) /* graph */ |
9697 | namedclass = complement ? ANYOF_NGRAPH : ANYOF_GRAPH; | |
80916619 NC |
9698 | break; |
9699 | case 'i': | |
3dab1dad YO |
9700 | if (memEQ(posixcc, "asci", 4)) /* ascii */ |
9701 | namedclass = complement ? ANYOF_NASCII : ANYOF_ASCII; | |
80916619 NC |
9702 | break; |
9703 | case 'k': | |
3dab1dad YO |
9704 | if (memEQ(posixcc, "blan", 4)) /* blank */ |
9705 | namedclass = complement ? ANYOF_NBLANK : ANYOF_BLANK; | |
80916619 NC |
9706 | break; |
9707 | case 'l': | |
3dab1dad YO |
9708 | if (memEQ(posixcc, "cntr", 4)) /* cntrl */ |
9709 | namedclass = complement ? ANYOF_NCNTRL : ANYOF_CNTRL; | |
80916619 NC |
9710 | break; |
9711 | case 'm': | |
3dab1dad YO |
9712 | if (memEQ(posixcc, "alnu", 4)) /* alnum */ |
9713 | namedclass = complement ? ANYOF_NALNUMC : ANYOF_ALNUMC; | |
80916619 NC |
9714 | break; |
9715 | case 'r': | |
3dab1dad YO |
9716 | if (memEQ(posixcc, "lowe", 4)) /* lower */ |
9717 | namedclass = complement ? ANYOF_NLOWER : ANYOF_LOWER; | |
9718 | else if (memEQ(posixcc, "uppe", 4)) /* upper */ | |
9719 | namedclass = complement ? ANYOF_NUPPER : ANYOF_UPPER; | |
80916619 NC |
9720 | break; |
9721 | case 't': | |
3dab1dad YO |
9722 | if (memEQ(posixcc, "digi", 4)) /* digit */ |
9723 | namedclass = complement ? ANYOF_NDIGIT : ANYOF_DIGIT; | |
9724 | else if (memEQ(posixcc, "prin", 4)) /* print */ | |
9725 | namedclass = complement ? ANYOF_NPRINT : ANYOF_PRINT; | |
9726 | else if (memEQ(posixcc, "punc", 4)) /* punct */ | |
9727 | namedclass = complement ? ANYOF_NPUNCT : ANYOF_PUNCT; | |
80916619 | 9728 | break; |
b8c5462f JH |
9729 | } |
9730 | break; | |
80916619 | 9731 | case 6: |
3dab1dad YO |
9732 | if (memEQ(posixcc, "xdigit", 6)) |
9733 | namedclass = complement ? ANYOF_NXDIGIT : ANYOF_XDIGIT; | |
b8c5462f JH |
9734 | break; |
9735 | } | |
80916619 NC |
9736 | |
9737 | if (namedclass == OOB_NAMEDCLASS) | |
b45f050a JF |
9738 | Simple_vFAIL3("POSIX class [:%.*s:] unknown", |
9739 | t - s - 1, s + 1); | |
80916619 NC |
9740 | assert (posixcc[skip] == ':'); |
9741 | assert (posixcc[skip+1] == ']'); | |
b45f050a | 9742 | } else if (!SIZE_ONLY) { |
b8c5462f | 9743 | /* [[=foo=]] and [[.foo.]] are still future. */ |
b45f050a | 9744 | |
830247a4 | 9745 | /* adjust RExC_parse so the warning shows after |
b45f050a | 9746 | the class closes */ |
9a86a77b | 9747 | while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse) != ']') |
830247a4 | 9748 | RExC_parse++; |
b45f050a JF |
9749 | Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c); |
9750 | } | |
b8c5462f JH |
9751 | } else { |
9752 | /* Maternal grandfather: | |
9753 | * "[:" ending in ":" but not in ":]" */ | |
830247a4 | 9754 | RExC_parse = s; |
767d463e | 9755 | } |
620e46c5 JH |
9756 | } |
9757 | } | |
9758 | ||
b8c5462f JH |
9759 | return namedclass; |
9760 | } | |
9761 | ||
9762 | STATIC void | |
830247a4 | 9763 | S_checkposixcc(pTHX_ RExC_state_t *pRExC_state) |
b8c5462f | 9764 | { |
97aff369 | 9765 | dVAR; |
7918f24d NC |
9766 | |
9767 | PERL_ARGS_ASSERT_CHECKPOSIXCC; | |
9768 | ||
3dab1dad | 9769 | if (POSIXCC(UCHARAT(RExC_parse))) { |
1df70142 AL |
9770 | const char *s = RExC_parse; |
9771 | const char c = *s++; | |
b8c5462f | 9772 | |
3dab1dad | 9773 | while (isALNUM(*s)) |
b8c5462f JH |
9774 | s++; |
9775 | if (*s && c == *s && s[1] == ']') { | |
668c081a NC |
9776 | ckWARN3reg(s+2, |
9777 | "POSIX syntax [%c %c] belongs inside character classes", | |
9778 | c, c); | |
b45f050a JF |
9779 | |
9780 | /* [[=foo=]] and [[.foo.]] are still future. */ | |
9a86a77b | 9781 | if (POSIXCC_NOTYET(c)) { |
830247a4 | 9782 | /* adjust RExC_parse so the error shows after |
b45f050a | 9783 | the class closes */ |
9a86a77b | 9784 | while (UCHARAT(RExC_parse) && UCHARAT(RExC_parse++) != ']') |
3dab1dad | 9785 | NOOP; |
b45f050a JF |
9786 | Simple_vFAIL3("POSIX syntax [%c %c] is reserved for future extensions", c, c); |
9787 | } | |
b8c5462f JH |
9788 | } |
9789 | } | |
620e46c5 JH |
9790 | } |
9791 | ||
4f03b4b6 | 9792 | /* No locale test, and always Unicode semantics, no ignore-case differences */ |
003331de KW |
9793 | #define _C_C_T_NOLOC_(NAME,TEST,WORD) \ |
9794 | ANYOF_##NAME: \ | |
9795 | for (value = 0; value < 256; value++) \ | |
9796 | if (TEST) \ | |
5bfec14d | 9797 | stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
003331de KW |
9798 | yesno = '+'; \ |
9799 | what = WORD; \ | |
9800 | break; \ | |
9801 | case ANYOF_N##NAME: \ | |
9802 | for (value = 0; value < 256; value++) \ | |
9803 | if (!TEST) \ | |
5bfec14d | 9804 | stored += set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
003331de KW |
9805 | yesno = '!'; \ |
9806 | what = WORD; \ | |
e1d1eefb | 9807 | break |
89836f1f | 9808 | |
a12cf05f KW |
9809 | /* Like the above, but there are differences if we are in uni-8-bit or not, so |
9810 | * there are two tests passed in, to use depending on that. There aren't any | |
9811 | * cases where the label is different from the name, so no need for that | |
4f03b4b6 KW |
9812 | * parameter. |
9813 | * Sets 'what' to WORD which is the property name for non-bitmap code points; | |
9814 | * But, uses FOLD_WORD instead if /i has been selected, to allow a different | |
9815 | * property name */ | |
9816 | #define _C_C_T_(NAME, TEST_8, TEST_7, WORD, FOLD_WORD) \ | |
003331de KW |
9817 | ANYOF_##NAME: \ |
9818 | if (LOC) ANYOF_CLASS_SET(ret, ANYOF_##NAME); \ | |
9819 | else if (UNI_SEMANTICS) { \ | |
9820 | for (value = 0; value < 256; value++) { \ | |
f952827c | 9821 | if (TEST_8(value)) stored += \ |
5bfec14d | 9822 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
003331de KW |
9823 | } \ |
9824 | } \ | |
9825 | else { \ | |
9826 | for (value = 0; value < 128; value++) { \ | |
f952827c | 9827 | if (TEST_7(UNI_TO_NATIVE(value))) stored += \ |
4c9daa0a | 9828 | set_regclass_bit(pRExC_state, ret, \ |
5bfec14d | 9829 | (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \ |
003331de KW |
9830 | } \ |
9831 | } \ | |
9832 | yesno = '+'; \ | |
4f03b4b6 KW |
9833 | if (FOLD) { \ |
9834 | what = FOLD_WORD; \ | |
9835 | } \ | |
9836 | else { \ | |
9837 | what = WORD; \ | |
9838 | } \ | |
003331de KW |
9839 | break; \ |
9840 | case ANYOF_N##NAME: \ | |
9841 | if (LOC) ANYOF_CLASS_SET(ret, ANYOF_N##NAME); \ | |
9842 | else if (UNI_SEMANTICS) { \ | |
9843 | for (value = 0; value < 256; value++) { \ | |
f952827c | 9844 | if (! TEST_8(value)) stored += \ |
5bfec14d | 9845 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); \ |
003331de KW |
9846 | } \ |
9847 | } \ | |
9848 | else { \ | |
9849 | for (value = 0; value < 128; value++) { \ | |
4c9daa0a | 9850 | if (! TEST_7(UNI_TO_NATIVE(value))) stored += set_regclass_bit( \ |
5bfec14d | 9851 | pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \ |
003331de | 9852 | } \ |
2f7f8cb1 | 9853 | if (AT_LEAST_ASCII_RESTRICTED) { \ |
cfaf538b | 9854 | for (value = 128; value < 256; value++) { \ |
4c9daa0a | 9855 | stored += set_regclass_bit( \ |
5bfec14d | 9856 | pRExC_state, ret, (U8) UNI_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); \ |
cfaf538b | 9857 | } \ |
137165a6 | 9858 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; \ |
cfaf538b KW |
9859 | } \ |
9860 | else { \ | |
9861 | /* For a non-ut8 target string with DEPENDS semantics, all above \ | |
9862 | * ASCII Latin1 code points match the complement of any of the \ | |
9863 | * classes. But in utf8, they have their Unicode semantics, so \ | |
9864 | * can't just set them in the bitmap, or else regexec.c will think \ | |
9865 | * they matched when they shouldn't. */ \ | |
137165a6 | 9866 | ANYOF_FLAGS(ret) |= ANYOF_NON_UTF8_LATIN1_ALL; \ |
cfaf538b | 9867 | } \ |
003331de KW |
9868 | } \ |
9869 | yesno = '!'; \ | |
4f03b4b6 KW |
9870 | if (FOLD) { \ |
9871 | what = FOLD_WORD; \ | |
9872 | } \ | |
9873 | else { \ | |
9874 | what = WORD; \ | |
9875 | } \ | |
a12cf05f KW |
9876 | break |
9877 | ||
2283d326 | 9878 | STATIC U8 |
a25abddc | 9879 | S_set_regclass_bit_fold(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr) |
2283d326 KW |
9880 | { |
9881 | ||
9882 | /* Handle the setting of folds in the bitmap for non-locale ANYOF nodes. | |
9883 | * Locale folding is done at run-time, so this function should not be | |
9884 | * called for nodes that are for locales. | |
9885 | * | |
d50a4f90 | 9886 | * This function sets the bit corresponding to the fold of the input |
2283d326 KW |
9887 | * 'value', if not already set. The fold of 'f' is 'F', and the fold of |
9888 | * 'F' is 'f'. | |
9889 | * | |
d50a4f90 KW |
9890 | * It also knows about the characters that are in the bitmap that have |
9891 | * folds that are matchable only outside it, and sets the appropriate lists | |
9892 | * and flags. | |
9893 | * | |
9894 | * It returns the number of bits that actually changed from 0 to 1 */ | |
2283d326 KW |
9895 | |
9896 | U8 stored = 0; | |
2283d326 KW |
9897 | U8 fold; |
9898 | ||
4c9daa0a KW |
9899 | PERL_ARGS_ASSERT_SET_REGCLASS_BIT_FOLD; |
9900 | ||
cfaf538b | 9901 | fold = (AT_LEAST_UNI_SEMANTICS) ? PL_fold_latin1[value] |
2f7f8cb1 | 9902 | : PL_fold[value]; |
2283d326 KW |
9903 | |
9904 | /* It assumes the bit for 'value' has already been set */ | |
9905 | if (fold != value && ! ANYOF_BITMAP_TEST(node, fold)) { | |
9906 | ANYOF_BITMAP_SET(node, fold); | |
9907 | stored++; | |
9908 | } | |
d50a4f90 KW |
9909 | if (_HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value) && (! isASCII(value) || ! MORE_ASCII_RESTRICTED)) { |
9910 | /* Certain Latin1 characters have matches outside the bitmap. To get | |
9911 | * here, 'value' is one of those characters. None of these matches is | |
9912 | * valid for ASCII characters under /aa, which have been excluded by | |
9913 | * the 'if' above. The matches fall into three categories: | |
9914 | * 1) They are singly folded-to or -from an above 255 character, as | |
9915 | * LATIN SMALL LETTER Y WITH DIAERESIS and LATIN CAPITAL LETTER Y | |
9916 | * WITH DIAERESIS; | |
9917 | * 2) They are part of a multi-char fold with another character in the | |
9918 | * bitmap, only LATIN SMALL LETTER SHARP S => "ss" fits that bill; | |
9919 | * 3) They are part of a multi-char fold with a character not in the | |
9920 | * bitmap, such as various ligatures. | |
9921 | * We aren't dealing fully with multi-char folds, except we do deal | |
9922 | * with the pattern containing a character that has a multi-char fold | |
9923 | * (not so much the inverse). | |
9924 | * For types 1) and 3), the matches only happen when the target string | |
9925 | * is utf8; that's not true for 2), and we set a flag for it. | |
9926 | * | |
9927 | * The code below adds to the passed in inversion list the single fold | |
9928 | * closures for 'value'. The values are hard-coded here so that an | |
9929 | * innocent-looking character class, like /[ks]/i won't have to go out | |
9930 | * to disk to find the possible matches. XXX It would be better to | |
9931 | * generate these via regen, in case a new version of the Unicode | |
9932 | * standard adds new mappings, though that is not really likely. */ | |
9933 | switch (value) { | |
9934 | case 'k': | |
9935 | case 'K': | |
9936 | /* KELVIN SIGN */ | |
9937 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212A); | |
9938 | break; | |
9939 | case 's': | |
9940 | case 'S': | |
9941 | /* LATIN SMALL LETTER LONG S */ | |
9942 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x017F); | |
9943 | break; | |
9944 | case MICRO_SIGN: | |
9945 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, | |
9946 | GREEK_SMALL_LETTER_MU); | |
9947 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, | |
9948 | GREEK_CAPITAL_LETTER_MU); | |
9949 | break; | |
9950 | case LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE: | |
9951 | case LATIN_SMALL_LETTER_A_WITH_RING_ABOVE: | |
9952 | /* ANGSTROM SIGN */ | |
9953 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, 0x212B); | |
9954 | if (DEPENDS_SEMANTICS) { /* See DEPENDS comment below */ | |
9955 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, | |
9956 | PL_fold_latin1[value]); | |
9957 | } | |
9958 | break; | |
9959 | case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS: | |
9960 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, | |
9961 | LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS); | |
9962 | break; | |
9963 | case LATIN_SMALL_LETTER_SHARP_S: | |
1d4120df KW |
9964 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, |
9965 | LATIN_CAPITAL_LETTER_SHARP_S); | |
d50a4f90 | 9966 | |
419d8974 | 9967 | /* Under /a, /d, and /u, this can match the two chars "ss" */ |
d50a4f90 KW |
9968 | if (! MORE_ASCII_RESTRICTED) { |
9969 | add_alternate(alternate_ptr, (U8 *) "ss", 2); | |
9970 | ||
419d8974 KW |
9971 | /* And under /u or /a, it can match even if the target is |
9972 | * not utf8 */ | |
9973 | if (AT_LEAST_UNI_SEMANTICS) { | |
d50a4f90 KW |
9974 | ANYOF_FLAGS(node) |= ANYOF_NONBITMAP_NON_UTF8; |
9975 | } | |
9976 | } | |
9977 | break; | |
9978 | case 'F': case 'f': | |
9979 | case 'I': case 'i': | |
9980 | case 'L': case 'l': | |
9981 | case 'T': case 't': | |
d50a4f90 KW |
9982 | case 'A': case 'a': |
9983 | case 'H': case 'h': | |
9984 | case 'J': case 'j': | |
9985 | case 'N': case 'n': | |
9986 | case 'W': case 'w': | |
9987 | case 'Y': case 'y': | |
f580a93d KW |
9988 | /* These all are targets of multi-character folds from code |
9989 | * points that require UTF8 to express, so they can't match | |
9990 | * unless the target string is in UTF-8, so no action here is | |
9991 | * necessary, as regexec.c properly handles the general case | |
9992 | * for UTF-8 matching */ | |
d50a4f90 KW |
9993 | break; |
9994 | default: | |
9995 | /* Use deprecated warning to increase the chances of this | |
9996 | * being output */ | |
9997 | ckWARN2regdep(RExC_parse, "Perl folding rules are not up-to-date for 0x%x; please use the perlbug utility to report;", value); | |
9998 | break; | |
9999 | } | |
10000 | } | |
10001 | else if (DEPENDS_SEMANTICS | |
f56b6394 | 10002 | && ! isASCII(value) |
d50a4f90 KW |
10003 | && PL_fold_latin1[value] != value) |
10004 | { | |
10005 | /* Under DEPENDS rules, non-ASCII Latin1 characters match their | |
10006 | * folds only when the target string is in UTF-8. We add the fold | |
10007 | * here to the list of things to match outside the bitmap, which | |
10008 | * won't be looked at unless it is UTF8 (or else if something else | |
10009 | * says to look even if not utf8, but those things better not happen | |
10010 | * under DEPENDS semantics. */ | |
10011 | *invlist_ptr = add_cp_to_invlist(*invlist_ptr, PL_fold_latin1[value]); | |
2283d326 KW |
10012 | } |
10013 | ||
10014 | return stored; | |
10015 | } | |
10016 | ||
10017 | ||
10018 | PERL_STATIC_INLINE U8 | |
a25abddc | 10019 | S_set_regclass_bit(pTHX_ RExC_state_t *pRExC_state, regnode* node, const U8 value, SV** invlist_ptr, AV** alternate_ptr) |
2283d326 KW |
10020 | { |
10021 | /* This inline function sets a bit in the bitmap if not already set, and if | |
10022 | * appropriate, its fold, returning the number of bits that actually | |
10023 | * changed from 0 to 1 */ | |
10024 | ||
10025 | U8 stored; | |
10026 | ||
4c9daa0a KW |
10027 | PERL_ARGS_ASSERT_SET_REGCLASS_BIT; |
10028 | ||
2283d326 KW |
10029 | if (ANYOF_BITMAP_TEST(node, value)) { /* Already set */ |
10030 | return 0; | |
10031 | } | |
10032 | ||
10033 | ANYOF_BITMAP_SET(node, value); | |
10034 | stored = 1; | |
10035 | ||
10036 | if (FOLD && ! LOC) { /* Locale folds aren't known until runtime */ | |
2c6aa593 | 10037 | stored += set_regclass_bit_fold(pRExC_state, node, value, invlist_ptr, alternate_ptr); |
2283d326 KW |
10038 | } |
10039 | ||
10040 | return stored; | |
10041 | } | |
10042 | ||
c8453963 KW |
10043 | STATIC void |
10044 | S_add_alternate(pTHX_ AV** alternate_ptr, U8* string, STRLEN len) | |
10045 | { | |
10046 | /* Adds input 'string' with length 'len' to the ANYOF node's unicode | |
10047 | * alternate list, pointed to by 'alternate_ptr'. This is an array of | |
10048 | * the multi-character folds of characters in the node */ | |
10049 | SV *sv; | |
10050 | ||
10051 | PERL_ARGS_ASSERT_ADD_ALTERNATE; | |
10052 | ||
10053 | if (! *alternate_ptr) { | |
10054 | *alternate_ptr = newAV(); | |
10055 | } | |
10056 | sv = newSVpvn_utf8((char*)string, len, TRUE); | |
10057 | av_push(*alternate_ptr, sv); | |
10058 | return; | |
10059 | } | |
10060 | ||
7f6f358c YO |
10061 | /* |
10062 | parse a class specification and produce either an ANYOF node that | |
ddad5e0b | 10063 | matches the pattern or perhaps will be optimized into an EXACTish node |
679d1424 KW |
10064 | instead. The node contains a bit map for the first 256 characters, with the |
10065 | corresponding bit set if that character is in the list. For characters | |
10066 | above 255, a range list is used */ | |
89836f1f | 10067 | |
76e3520e | 10068 | STATIC regnode * |
3dab1dad | 10069 | S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth) |
a687059c | 10070 | { |
97aff369 | 10071 | dVAR; |
9a86a77b | 10072 | register UV nextvalue; |
3568d838 | 10073 | register IV prevvalue = OOB_UNICODE; |
ffc61ed2 | 10074 | register IV range = 0; |
e1d1eefb | 10075 | UV value = 0; /* XXX:dmq: needs to be referenceable (unfortunately) */ |
c277df42 | 10076 | register regnode *ret; |
ba210ebe | 10077 | STRLEN numlen; |
ffc61ed2 | 10078 | IV namedclass; |
cbbf8932 | 10079 | char *rangebegin = NULL; |
936ed897 | 10080 | bool need_class = 0; |
827f5bb8 | 10081 | bool allow_full_fold = TRUE; /* Assume wants multi-char folding */ |
c445ea15 | 10082 | SV *listsv = NULL; |
137165a6 KW |
10083 | STRLEN initial_listsv_len = 0; /* Kind of a kludge to see if it is more |
10084 | than just initialized. */ | |
dc415b4a KW |
10085 | SV* properties = NULL; /* Code points that match \p{} \P{} */ |
10086 | UV element_count = 0; /* Number of distinct elements in the class. | |
10087 | Optimizations may be possible if this is tiny */ | |
ffc61ed2 | 10088 | UV n; |
53742956 | 10089 | |
dc415b4a | 10090 | /* Unicode properties are stored in a swash; this holds the current one |
88d45d28 KW |
10091 | * being parsed. If this swash is the only above-latin1 component of the |
10092 | * character class, an optimization is to pass it directly on to the | |
10093 | * execution engine. Otherwise, it is set to NULL to indicate that there | |
10094 | * are other things in the class that have to be dealt with at execution | |
10095 | * time */ | |
dc415b4a KW |
10096 | SV* swash = NULL; /* Code points that match \p{} \P{} */ |
10097 | ||
10098 | /* Set if a component of this character class is user-defined; just passed | |
10099 | * on to the engine */ | |
10100 | UV has_user_defined_property = 0; | |
10101 | ||
53742956 | 10102 | /* code points this node matches that can't be stored in the bitmap */ |
a25abddc | 10103 | SV* nonbitmap = NULL; |
53742956 KW |
10104 | |
10105 | /* The items that are to match that aren't stored in the bitmap, but are a | |
10106 | * result of things that are stored there. This is the fold closure of | |
10107 | * such a character, either because it has DEPENDS semantics and shouldn't | |
10108 | * be matched unless the target string is utf8, or is a code point that is | |
10109 | * too large for the bit map, as for example, the fold of the MICRO SIGN is | |
10110 | * above 255. This all is solely for performance reasons. By having this | |
10111 | * code know the outside-the-bitmap folds that the bitmapped characters are | |
10112 | * involved with, we don't have to go out to disk to find the list of | |
10113 | * matches, unless the character class includes code points that aren't | |
10114 | * storable in the bit map. That means that a character class with an 's' | |
10115 | * in it, for example, doesn't need to go out to disk to find everything | |
10116 | * that matches. A 2nd list is used so that the 'nonbitmap' list is kept | |
10117 | * empty unless there is something whose fold we don't know about, and will | |
10118 | * have to go out to the disk to find. */ | |
a25abddc | 10119 | SV* l1_fold_invlist = NULL; |
53742956 KW |
10120 | |
10121 | /* List of multi-character folds that are matched by this node */ | |
cbbf8932 | 10122 | AV* unicode_alternate = NULL; |
1b2d223b JH |
10123 | #ifdef EBCDIC |
10124 | UV literal_endpoint = 0; | |
10125 | #endif | |
ffc130aa | 10126 | UV stored = 0; /* how many chars stored in the bitmap */ |
ffc61ed2 | 10127 | |
3dab1dad | 10128 | regnode * const orig_emit = RExC_emit; /* Save the original RExC_emit in |
7f6f358c | 10129 | case we need to change the emitted regop to an EXACT. */ |
07be1b83 | 10130 | const char * orig_parse = RExC_parse; |
72f13be8 | 10131 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
10132 | |
10133 | PERL_ARGS_ASSERT_REGCLASS; | |
76e84362 SH |
10134 | #ifndef DEBUGGING |
10135 | PERL_UNUSED_ARG(depth); | |
10136 | #endif | |
72f13be8 | 10137 | |
3dab1dad | 10138 | DEBUG_PARSE("clas"); |
7f6f358c YO |
10139 | |
10140 | /* Assume we are going to generate an ANYOF node. */ | |
ffc61ed2 JH |
10141 | ret = reganode(pRExC_state, ANYOF, 0); |
10142 | ||
56ca34ca KW |
10143 | |
10144 | if (!SIZE_ONLY) { | |
ffc61ed2 | 10145 | ANYOF_FLAGS(ret) = 0; |
56ca34ca | 10146 | } |
ffc61ed2 | 10147 | |
9a86a77b | 10148 | if (UCHARAT(RExC_parse) == '^') { /* Complement of range. */ |
ffc61ed2 JH |
10149 | RExC_naughty++; |
10150 | RExC_parse++; | |
10151 | if (!SIZE_ONLY) | |
10152 | ANYOF_FLAGS(ret) |= ANYOF_INVERT; | |
827f5bb8 KW |
10153 | |
10154 | /* We have decided to not allow multi-char folds in inverted character | |
ac455f4c KW |
10155 | * classes, due to the confusion that can happen, especially with |
10156 | * classes that are designed for a non-Unicode world: You have the | |
10157 | * peculiar case that: | |
827f5bb8 KW |
10158 | "s s" =~ /^[^\xDF]+$/i => Y |
10159 | "ss" =~ /^[^\xDF]+$/i => N | |
10160 | * | |
10161 | * See [perl #89750] */ | |
10162 | allow_full_fold = FALSE; | |
ffc61ed2 | 10163 | } |
a0d0e21e | 10164 | |
73060fc4 | 10165 | if (SIZE_ONLY) { |
830247a4 | 10166 | RExC_size += ANYOF_SKIP; |
73060fc4 JH |
10167 | listsv = &PL_sv_undef; /* For code scanners: listsv always non-NULL. */ |
10168 | } | |
936ed897 | 10169 | else { |
830247a4 | 10170 | RExC_emit += ANYOF_SKIP; |
3a15e693 | 10171 | if (LOC) { |
936ed897 | 10172 | ANYOF_FLAGS(ret) |= ANYOF_LOCALE; |
3a15e693 | 10173 | } |
ffc61ed2 | 10174 | ANYOF_BITMAP_ZERO(ret); |
396482e1 | 10175 | listsv = newSVpvs("# comment\n"); |
137165a6 | 10176 | initial_listsv_len = SvCUR(listsv); |
a0d0e21e | 10177 | } |
b8c5462f | 10178 | |
9a86a77b JH |
10179 | nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0; |
10180 | ||
b938889d | 10181 | if (!SIZE_ONLY && POSIXCC(nextvalue)) |
830247a4 | 10182 | checkposixcc(pRExC_state); |
b8c5462f | 10183 | |
f064b6ad HS |
10184 | /* allow 1st char to be ] (allowing it to be - is dealt with later) */ |
10185 | if (UCHARAT(RExC_parse) == ']') | |
10186 | goto charclassloop; | |
ffc61ed2 | 10187 | |
fc8cd66c | 10188 | parseit: |
9a86a77b | 10189 | while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') { |
ffc61ed2 JH |
10190 | |
10191 | charclassloop: | |
10192 | ||
10193 | namedclass = OOB_NAMEDCLASS; /* initialize as illegal */ | |
10194 | ||
dc415b4a | 10195 | if (!range) { |
830247a4 | 10196 | rangebegin = RExC_parse; |
dc415b4a KW |
10197 | element_count++; |
10198 | } | |
ffc61ed2 | 10199 | if (UTF) { |
5e12f4fb | 10200 | value = utf8n_to_uvchr((U8*)RExC_parse, |
3568d838 | 10201 | RExC_end - RExC_parse, |
9f7f3913 | 10202 | &numlen, UTF8_ALLOW_DEFAULT); |
ffc61ed2 JH |
10203 | RExC_parse += numlen; |
10204 | } | |
10205 | else | |
10206 | value = UCHARAT(RExC_parse++); | |
7f6f358c | 10207 | |
9a86a77b JH |
10208 | nextvalue = RExC_parse < RExC_end ? UCHARAT(RExC_parse) : 0; |
10209 | if (value == '[' && POSIXCC(nextvalue)) | |
830247a4 | 10210 | namedclass = regpposixcc(pRExC_state, value); |
620e46c5 | 10211 | else if (value == '\\') { |
ffc61ed2 | 10212 | if (UTF) { |
5e12f4fb | 10213 | value = utf8n_to_uvchr((U8*)RExC_parse, |
ffc61ed2 | 10214 | RExC_end - RExC_parse, |
9f7f3913 | 10215 | &numlen, UTF8_ALLOW_DEFAULT); |
ffc61ed2 JH |
10216 | RExC_parse += numlen; |
10217 | } | |
10218 | else | |
10219 | value = UCHARAT(RExC_parse++); | |
470c3474 | 10220 | /* Some compilers cannot handle switching on 64-bit integer |
ffc61ed2 | 10221 | * values, therefore value cannot be an UV. Yes, this will |
e2962f66 JH |
10222 | * be a problem later if we want switch on Unicode. |
10223 | * A similar issue a little bit later when switching on | |
10224 | * namedclass. --jhi */ | |
ffc61ed2 | 10225 | switch ((I32)value) { |
b8c5462f JH |
10226 | case 'w': namedclass = ANYOF_ALNUM; break; |
10227 | case 'W': namedclass = ANYOF_NALNUM; break; | |
10228 | case 's': namedclass = ANYOF_SPACE; break; | |
10229 | case 'S': namedclass = ANYOF_NSPACE; break; | |
10230 | case 'd': namedclass = ANYOF_DIGIT; break; | |
10231 | case 'D': namedclass = ANYOF_NDIGIT; break; | |
e1d1eefb YO |
10232 | case 'v': namedclass = ANYOF_VERTWS; break; |
10233 | case 'V': namedclass = ANYOF_NVERTWS; break; | |
10234 | case 'h': namedclass = ANYOF_HORIZWS; break; | |
10235 | case 'H': namedclass = ANYOF_NHORIZWS; break; | |
fc8cd66c YO |
10236 | case 'N': /* Handle \N{NAME} in class */ |
10237 | { | |
10238 | /* We only pay attention to the first char of | |
10239 | multichar strings being returned. I kinda wonder | |
10240 | if this makes sense as it does change the behaviour | |
10241 | from earlier versions, OTOH that behaviour was broken | |
10242 | as well. */ | |
10243 | UV v; /* value is register so we cant & it /grrr */ | |
9d64099b | 10244 | if (reg_namedseq(pRExC_state, &v, NULL, depth)) { |
fc8cd66c YO |
10245 | goto parseit; |
10246 | } | |
10247 | value= v; | |
10248 | } | |
10249 | break; | |
ffc61ed2 JH |
10250 | case 'p': |
10251 | case 'P': | |
3dab1dad YO |
10252 | { |
10253 | char *e; | |
af6f566e | 10254 | if (RExC_parse >= RExC_end) |
2a4859cd | 10255 | vFAIL2("Empty \\%c{}", (U8)value); |
ffc61ed2 | 10256 | if (*RExC_parse == '{') { |
1df70142 | 10257 | const U8 c = (U8)value; |
ffc61ed2 JH |
10258 | e = strchr(RExC_parse++, '}'); |
10259 | if (!e) | |
0da60cf5 | 10260 | vFAIL2("Missing right brace on \\%c{}", c); |
ab13f0c7 JH |
10261 | while (isSPACE(UCHARAT(RExC_parse))) |
10262 | RExC_parse++; | |
10263 | if (e == RExC_parse) | |
0da60cf5 | 10264 | vFAIL2("Empty \\%c{}", c); |
ffc61ed2 | 10265 | n = e - RExC_parse; |
ab13f0c7 JH |
10266 | while (isSPACE(UCHARAT(RExC_parse + n - 1))) |
10267 | n--; | |
ffc61ed2 JH |
10268 | } |
10269 | else { | |
10270 | e = RExC_parse; | |
10271 | n = 1; | |
10272 | } | |
ee410026 | 10273 | if (!SIZE_ONLY) { |
dc415b4a KW |
10274 | SV** invlistsvp; |
10275 | SV* invlist; | |
10276 | char* name; | |
ab13f0c7 JH |
10277 | if (UCHARAT(RExC_parse) == '^') { |
10278 | RExC_parse++; | |
10279 | n--; | |
10280 | value = value == 'p' ? 'P' : 'p'; /* toggle */ | |
10281 | while (isSPACE(UCHARAT(RExC_parse))) { | |
10282 | RExC_parse++; | |
10283 | n--; | |
10284 | } | |
10285 | } | |
dc415b4a KW |
10286 | /* Try to get the definition of the property into |
10287 | * <invlist>. If /i is in effect, the effective property | |
10288 | * will have its name be <__NAME_i>. The design is | |
10289 | * discussed in commit | |
10290 | * 2f833f5208e26b208886e51e09e2c072b5eabb46 */ | |
10291 | Newx(name, n + sizeof("_i__\n"), char); | |
10292 | ||
10293 | sprintf(name, "%s%.*s%s\n", | |
10294 | (FOLD) ? "__" : "", | |
10295 | (int)n, | |
10296 | RExC_parse, | |
10297 | (FOLD) ? "_i" : "" | |
10298 | ); | |
10299 | ||
10300 | /* Look up the property name, and get its swash and | |
10301 | * inversion list, if the property is found */ | |
10302 | if (! (ANYOF_FLAGS(ret) & ANYOF_INVERT)) { | |
10303 | if (swash) { | |
10304 | SvREFCNT_dec(swash); | |
10305 | } | |
10306 | swash = _core_swash_init("utf8", name, &PL_sv_undef, | |
10307 | 1, /* binary */ | |
10308 | 0, /* not tr/// */ | |
10309 | TRUE, /* this routine will handle | |
10310 | undefined properties */ | |
10311 | NULL, FALSE /* No inversion list */ | |
10312 | ); | |
10313 | } | |
10314 | ||
10315 | if ( ANYOF_FLAGS(ret) & ANYOF_INVERT | |
10316 | || ! swash | |
10317 | || ! SvROK(swash) | |
10318 | || ! SvTYPE(SvRV(swash)) == SVt_PVHV | |
10319 | || ! (invlistsvp = | |
10320 | hv_fetchs(MUTABLE_HV(SvRV(swash)), | |
10321 | "INVLIST", FALSE)) | |
10322 | || ! (invlist = *invlistsvp)) | |
10323 | { | |
10324 | if (swash) { | |
10325 | SvREFCNT_dec(swash); | |
10326 | swash = NULL; | |
10327 | } | |
10328 | ||
10329 | /* Here didn't find it. It could be a user-defined | |
10330 | * property that will be available at run-time. Add it | |
10331 | * to the list to look up then */ | |
10332 | Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s\n", | |
10333 | (value == 'p' ? '+' : '!'), | |
10334 | name); | |
10335 | has_user_defined_property = 1; | |
10336 | ||
10337 | /* We don't know yet, so have to assume that the | |
10338 | * property could match something in the Latin1 range, | |
10339 | * hence something that isn't utf8 */ | |
10340 | ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8; | |
10341 | } | |
10342 | else { | |
10343 | ||
10344 | /* Here, did get the swash and its inversion list. If | |
10345 | * the swash is from a user-defined property, then this | |
10346 | * whole character class should be regarded as such */ | |
10347 | SV** user_defined_svp = | |
10348 | hv_fetchs(MUTABLE_HV(SvRV(swash)), | |
10349 | "USER_DEFINED", FALSE); | |
10350 | if (user_defined_svp) { | |
10351 | has_user_defined_property | |
10352 | |= SvUV(*user_defined_svp); | |
10353 | } | |
10354 | ||
10355 | /* Invert if asking for the complement */ | |
10356 | if (value == 'P') { | |
10357 | ||
10358 | /* Add to any existing list */ | |
10359 | if (! properties) { | |
10360 | properties = invlist_clone(invlist); | |
10361 | _invlist_invert(properties); | |
10362 | } | |
10363 | else { | |
10364 | invlist = invlist_clone(invlist); | |
10365 | _invlist_invert(invlist); | |
10366 | _invlist_union(properties, invlist, &properties); | |
10367 | SvREFCNT_dec(invlist); | |
10368 | } | |
2f833f52 | 10369 | |
dc415b4a KW |
10370 | /* The swash can't be used as-is, because we've |
10371 | * inverted things; delay removing it to here after | |
10372 | * have copied its invlist above */ | |
10373 | SvREFCNT_dec(swash); | |
10374 | swash = NULL; | |
10375 | } | |
10376 | else { | |
10377 | if (! properties) { | |
10378 | properties = invlist_clone(invlist); | |
10379 | } | |
10380 | else { | |
10381 | _invlist_union(properties, invlist, &properties); | |
10382 | } | |
10383 | } | |
10384 | } | |
10385 | Safefree(name); | |
ffc61ed2 JH |
10386 | } |
10387 | RExC_parse = e + 1; | |
08fc12dd KW |
10388 | |
10389 | /* The \p could match something in the Latin1 range, hence | |
10390 | * something that isn't utf8 */ | |
db8c82dd | 10391 | ANYOF_FLAGS(ret) |= ANYOF_NONBITMAP_NON_UTF8; |
f81125e2 | 10392 | namedclass = ANYOF_MAX; /* no official name, but it's named */ |
e40e74fe KW |
10393 | |
10394 | /* \p means they want Unicode semantics */ | |
10395 | RExC_uni_semantics = 1; | |
3dab1dad | 10396 | } |
f81125e2 | 10397 | break; |
b8c5462f JH |
10398 | case 'n': value = '\n'; break; |
10399 | case 'r': value = '\r'; break; | |
10400 | case 't': value = '\t'; break; | |
10401 | case 'f': value = '\f'; break; | |
10402 | case 'b': value = '\b'; break; | |
c7f1f016 NIS |
10403 | case 'e': value = ASCII_TO_NATIVE('\033');break; |
10404 | case 'a': value = ASCII_TO_NATIVE('\007');break; | |
f0a2b745 KW |
10405 | case 'o': |
10406 | RExC_parse--; /* function expects to be pointed at the 'o' */ | |
454155d9 KW |
10407 | { |
10408 | const char* error_msg; | |
10409 | bool valid = grok_bslash_o(RExC_parse, | |
f0a2b745 KW |
10410 | &value, |
10411 | &numlen, | |
454155d9 KW |
10412 | &error_msg, |
10413 | SIZE_ONLY); | |
10414 | RExC_parse += numlen; | |
10415 | if (! valid) { | |
10416 | vFAIL(error_msg); | |
10417 | } | |
f0a2b745 | 10418 | } |
f0a2b745 KW |
10419 | if (PL_encoding && value < 0x100) { |
10420 | goto recode_encoding; | |
10421 | } | |
10422 | break; | |
b8c5462f | 10423 | case 'x': |
ffc61ed2 | 10424 | if (*RExC_parse == '{') { |
a4c04bdc NC |
10425 | I32 flags = PERL_SCAN_ALLOW_UNDERSCORES |
10426 | | PERL_SCAN_DISALLOW_PREFIX; | |
3dab1dad | 10427 | char * const e = strchr(RExC_parse++, '}'); |
b81d288d | 10428 | if (!e) |
ffc61ed2 | 10429 | vFAIL("Missing right brace on \\x{}"); |
53305cf1 NC |
10430 | |
10431 | numlen = e - RExC_parse; | |
10432 | value = grok_hex(RExC_parse, &numlen, &flags, NULL); | |
ffc61ed2 JH |
10433 | RExC_parse = e + 1; |
10434 | } | |
10435 | else { | |
a4c04bdc | 10436 | I32 flags = PERL_SCAN_DISALLOW_PREFIX; |
53305cf1 NC |
10437 | numlen = 2; |
10438 | value = grok_hex(RExC_parse, &numlen, &flags, NULL); | |
ffc61ed2 JH |
10439 | RExC_parse += numlen; |
10440 | } | |
9e08bc66 TS |
10441 | if (PL_encoding && value < 0x100) |
10442 | goto recode_encoding; | |
b8c5462f JH |
10443 | break; |
10444 | case 'c': | |
17a3df4c | 10445 | value = grok_bslash_c(*RExC_parse++, UTF, SIZE_ONLY); |
b8c5462f JH |
10446 | break; |
10447 | case '0': case '1': case '2': case '3': case '4': | |
c99e91e9 | 10448 | case '5': case '6': case '7': |
9e08bc66 | 10449 | { |
c99e91e9 KW |
10450 | /* Take 1-3 octal digits */ |
10451 | I32 flags = PERL_SCAN_SILENT_ILLDIGIT; | |
9e08bc66 TS |
10452 | numlen = 3; |
10453 | value = grok_oct(--RExC_parse, &numlen, &flags, NULL); | |
10454 | RExC_parse += numlen; | |
10455 | if (PL_encoding && value < 0x100) | |
10456 | goto recode_encoding; | |
10457 | break; | |
10458 | } | |
10459 | recode_encoding: | |
e2a7e165 | 10460 | if (! RExC_override_recoding) { |
9e08bc66 TS |
10461 | SV* enc = PL_encoding; |
10462 | value = reg_recode((const char)(U8)value, &enc); | |
668c081a NC |
10463 | if (!enc && SIZE_ONLY) |
10464 | ckWARNreg(RExC_parse, | |
10465 | "Invalid escape in the specified encoding"); | |
9e08bc66 TS |
10466 | break; |
10467 | } | |
1028017a | 10468 | default: |
c99e91e9 KW |
10469 | /* Allow \_ to not give an error */ |
10470 | if (!SIZE_ONLY && isALNUM(value) && value != '_') { | |
668c081a NC |
10471 | ckWARN2reg(RExC_parse, |
10472 | "Unrecognized escape \\%c in character class passed through", | |
10473 | (int)value); | |
c99e91e9 | 10474 | } |
1028017a | 10475 | break; |
b8c5462f | 10476 | } |
ffc61ed2 | 10477 | } /* end of \blah */ |
1b2d223b JH |
10478 | #ifdef EBCDIC |
10479 | else | |
10480 | literal_endpoint++; | |
10481 | #endif | |
ffc61ed2 JH |
10482 | |
10483 | if (namedclass > OOB_NAMEDCLASS) { /* this is a named class \blah */ | |
10484 | ||
2c63ecad KW |
10485 | /* What matches in a locale is not known until runtime, so need to |
10486 | * (one time per class) allocate extra space to pass to regexec. | |
10487 | * The space will contain a bit for each named class that is to be | |
10488 | * matched against. This isn't needed for \p{} and pseudo-classes, | |
10489 | * as they are not affected by locale, and hence are dealt with | |
10490 | * separately */ | |
10491 | if (LOC && namedclass < ANYOF_MAX && ! need_class) { | |
10492 | need_class = 1; | |
10493 | if (SIZE_ONLY) { | |
dd58aee1 | 10494 | RExC_size += ANYOF_CLASS_SKIP - ANYOF_SKIP; |
2c63ecad KW |
10495 | } |
10496 | else { | |
dd58aee1 | 10497 | RExC_emit += ANYOF_CLASS_SKIP - ANYOF_SKIP; |
2c63ecad KW |
10498 | ANYOF_CLASS_ZERO(ret); |
10499 | } | |
9051cfd9 | 10500 | ANYOF_FLAGS(ret) |= ANYOF_CLASS; |
2c63ecad | 10501 | } |
ffc61ed2 | 10502 | |
d5788240 | 10503 | /* a bad range like a-\d, a-[:digit:]. The '-' is taken as a |
1d791ab2 KW |
10504 | * literal, as is the character that began the false range, i.e. |
10505 | * the 'a' in the examples */ | |
ffc61ed2 | 10506 | if (range) { |
73b437c8 | 10507 | if (!SIZE_ONLY) { |
668c081a NC |
10508 | const int w = |
10509 | RExC_parse >= rangebegin ? | |
10510 | RExC_parse - rangebegin : 0; | |
10511 | ckWARN4reg(RExC_parse, | |
b45f050a | 10512 | "False [] range \"%*.*s\"", |
097eb12c | 10513 | w, w, rangebegin); |
668c081a | 10514 | |
1d791ab2 | 10515 | stored += |
5bfec14d | 10516 | set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate); |
3568d838 | 10517 | if (prevvalue < 256) { |
2283d326 | 10518 | stored += |
5bfec14d | 10519 | set_regclass_bit(pRExC_state, ret, (U8) prevvalue, &l1_fold_invlist, &unicode_alternate); |
ffc61ed2 JH |
10520 | } |
10521 | else { | |
1d791ab2 | 10522 | nonbitmap = add_cp_to_invlist(nonbitmap, prevvalue); |
ffc61ed2 | 10523 | } |
b8c5462f | 10524 | } |
ffc61ed2 JH |
10525 | |
10526 | range = 0; /* this was not a true range */ | |
73b437c8 | 10527 | } |
ffc61ed2 | 10528 | |
73b437c8 | 10529 | if (!SIZE_ONLY) { |
c49a72a9 NC |
10530 | const char *what = NULL; |
10531 | char yesno = 0; | |
10532 | ||
e2962f66 JH |
10533 | /* Possible truncation here but in some 64-bit environments |
10534 | * the compiler gets heartburn about switch on 64-bit values. | |
10535 | * A similar issue a little earlier when switching on value. | |
98f323fa | 10536 | * --jhi */ |
e2962f66 | 10537 | switch ((I32)namedclass) { |
da7fcca4 | 10538 | |
4f03b4b6 KW |
10539 | case _C_C_T_(ALNUMC, isALNUMC_L1, isALNUMC, "XPosixAlnum", "XPosixAlnum"); |
10540 | case _C_C_T_(ALPHA, isALPHA_L1, isALPHA, "XPosixAlpha", "XPosixAlpha"); | |
10541 | case _C_C_T_(BLANK, isBLANK_L1, isBLANK, "XPosixBlank", "XPosixBlank"); | |
10542 | case _C_C_T_(CNTRL, isCNTRL_L1, isCNTRL, "XPosixCntrl", "XPosixCntrl"); | |
10543 | case _C_C_T_(GRAPH, isGRAPH_L1, isGRAPH, "XPosixGraph", "XPosixGraph"); | |
10544 | case _C_C_T_(LOWER, isLOWER_L1, isLOWER, "XPosixLower", "__XPosixLower_i"); | |
10545 | case _C_C_T_(PRINT, isPRINT_L1, isPRINT, "XPosixPrint", "XPosixPrint"); | |
10546 | case _C_C_T_(PSXSPC, isPSXSPC_L1, isPSXSPC, "XPosixSpace", "XPosixSpace"); | |
10547 | case _C_C_T_(PUNCT, isPUNCT_L1, isPUNCT, "XPosixPunct", "XPosixPunct"); | |
10548 | case _C_C_T_(UPPER, isUPPER_L1, isUPPER, "XPosixUpper", "__XPosixUpper_i"); | |
a12cf05f | 10549 | /* \s, \w match all unicode if utf8. */ |
4f03b4b6 KW |
10550 | case _C_C_T_(SPACE, isSPACE_L1, isSPACE, "SpacePerl", "SpacePerl"); |
10551 | case _C_C_T_(ALNUM, isWORDCHAR_L1, isALNUM, "Word", "Word"); | |
10552 | case _C_C_T_(XDIGIT, isXDIGIT_L1, isXDIGIT, "XPosixXDigit", "XPosixXDigit"); | |
e1d1eefb YO |
10553 | case _C_C_T_NOLOC_(VERTWS, is_VERTWS_latin1(&value), "VertSpace"); |
10554 | case _C_C_T_NOLOC_(HORIZWS, is_HORIZWS_latin1(&value), "HorizSpace"); | |
73b437c8 JH |
10555 | case ANYOF_ASCII: |
10556 | if (LOC) | |
936ed897 | 10557 | ANYOF_CLASS_SET(ret, ANYOF_ASCII); |
73b437c8 | 10558 | else { |
1ba5c669 | 10559 | for (value = 0; value < 128; value++) |
2283d326 | 10560 | stored += |
5bfec14d | 10561 | set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); |
73b437c8 | 10562 | } |
c49a72a9 | 10563 | yesno = '+'; |
ce1c68b2 KW |
10564 | what = NULL; /* Doesn't match outside ascii, so |
10565 | don't want to add +utf8:: */ | |
73b437c8 JH |
10566 | break; |
10567 | case ANYOF_NASCII: | |
10568 | if (LOC) | |
936ed897 | 10569 | ANYOF_CLASS_SET(ret, ANYOF_NASCII); |
73b437c8 | 10570 | else { |
1ba5c669 | 10571 | for (value = 128; value < 256; value++) |
2283d326 | 10572 | stored += |
5bfec14d | 10573 | set_regclass_bit(pRExC_state, ret, (U8) ASCII_TO_NATIVE(value), &l1_fold_invlist, &unicode_alternate); |
73b437c8 | 10574 | } |
cfaf538b | 10575 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; |
c49a72a9 NC |
10576 | yesno = '!'; |
10577 | what = "ASCII"; | |
89836f1f | 10578 | break; |
ffc61ed2 JH |
10579 | case ANYOF_DIGIT: |
10580 | if (LOC) | |
10581 | ANYOF_CLASS_SET(ret, ANYOF_DIGIT); | |
10582 | else { | |
10583 | /* consecutive digits assumed */ | |
10584 | for (value = '0'; value <= '9'; value++) | |
2283d326 | 10585 | stored += |
5bfec14d | 10586 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); |
ffc61ed2 | 10587 | } |
c49a72a9 | 10588 | yesno = '+'; |
779d7b58 | 10589 | what = "Digit"; |
ffc61ed2 JH |
10590 | break; |
10591 | case ANYOF_NDIGIT: | |
10592 | if (LOC) | |
10593 | ANYOF_CLASS_SET(ret, ANYOF_NDIGIT); | |
10594 | else { | |
10595 | /* consecutive digits assumed */ | |
10596 | for (value = 0; value < '0'; value++) | |
2283d326 | 10597 | stored += |
5bfec14d | 10598 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); |
ffc61ed2 | 10599 | for (value = '9' + 1; value < 256; value++) |
2283d326 | 10600 | stored += |
5bfec14d | 10601 | set_regclass_bit(pRExC_state, ret, (U8) value, &l1_fold_invlist, &unicode_alternate); |
ffc61ed2 | 10602 | } |
c49a72a9 | 10603 | yesno = '!'; |
779d7b58 | 10604 | what = "Digit"; |
2f7f8cb1 | 10605 | if (AT_LEAST_ASCII_RESTRICTED ) { |
cfaf538b KW |
10606 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; |
10607 | } | |
89836f1f | 10608 | break; |
f81125e2 JP |
10609 | case ANYOF_MAX: |
10610 | /* this is to handle \p and \P */ | |
10611 | break; | |
73b437c8 | 10612 | default: |
b45f050a | 10613 | vFAIL("Invalid [::] class"); |
73b437c8 | 10614 | break; |
b8c5462f | 10615 | } |
2f7f8cb1 | 10616 | if (what && ! (AT_LEAST_ASCII_RESTRICTED)) { |
c49a72a9 | 10617 | /* Strings such as "+utf8::isWord\n" */ |
9ca75586 | 10618 | Perl_sv_catpvf(aTHX_ listsv, "%cutf8::%s\n", yesno, what); |
ef87b810 | 10619 | } |
ce1c68b2 | 10620 | |
73b437c8 | 10621 | continue; |
a0d0e21e | 10622 | } |
ffc61ed2 JH |
10623 | } /* end of namedclass \blah */ |
10624 | ||
a0d0e21e | 10625 | if (range) { |
eb160463 | 10626 | if (prevvalue > (IV)value) /* b-a */ { |
d4c19fe8 AL |
10627 | const int w = RExC_parse - rangebegin; |
10628 | Simple_vFAIL4("Invalid [] range \"%*.*s\"", w, w, rangebegin); | |
3568d838 | 10629 | range = 0; /* not a valid range */ |
73b437c8 | 10630 | } |
a0d0e21e LW |
10631 | } |
10632 | else { | |
3568d838 | 10633 | prevvalue = value; /* save the beginning of the range */ |
646253b5 KW |
10634 | if (RExC_parse+1 < RExC_end |
10635 | && *RExC_parse == '-' | |
10636 | && RExC_parse[1] != ']') | |
10637 | { | |
830247a4 | 10638 | RExC_parse++; |
ffc61ed2 JH |
10639 | |
10640 | /* a bad range like \w-, [:word:]- ? */ | |
10641 | if (namedclass > OOB_NAMEDCLASS) { | |
afd78fd5 | 10642 | if (ckWARN(WARN_REGEXP)) { |
d4c19fe8 | 10643 | const int w = |
afd78fd5 JH |
10644 | RExC_parse >= rangebegin ? |
10645 | RExC_parse - rangebegin : 0; | |
830247a4 | 10646 | vWARN4(RExC_parse, |
b45f050a | 10647 | "False [] range \"%*.*s\"", |
097eb12c | 10648 | w, w, rangebegin); |
afd78fd5 | 10649 | } |
73b437c8 | 10650 | if (!SIZE_ONLY) |
2283d326 | 10651 | stored += |
5bfec14d | 10652 | set_regclass_bit(pRExC_state, ret, '-', &l1_fold_invlist, &unicode_alternate); |
73b437c8 | 10653 | } else |
ffc61ed2 JH |
10654 | range = 1; /* yeah, it's a range! */ |
10655 | continue; /* but do it the next time */ | |
a0d0e21e | 10656 | } |
a687059c | 10657 | } |
ffc61ed2 | 10658 | |
046c4055 KW |
10659 | /* non-Latin1 code point implies unicode semantics. Must be set in |
10660 | * pass1 so is there for the whole of pass 2 */ | |
56ca34ca KW |
10661 | if (value > 255) { |
10662 | RExC_uni_semantics = 1; | |
10663 | } | |
10664 | ||
93733859 | 10665 | /* now is the next time */ |
ae5c130c | 10666 | if (!SIZE_ONLY) { |
3568d838 | 10667 | if (prevvalue < 256) { |
1df70142 | 10668 | const IV ceilvalue = value < 256 ? value : 255; |
3dab1dad | 10669 | IV i; |
3568d838 | 10670 | #ifdef EBCDIC |
1b2d223b JH |
10671 | /* In EBCDIC [\x89-\x91] should include |
10672 | * the \x8e but [i-j] should not. */ | |
10673 | if (literal_endpoint == 2 && | |
10674 | ((isLOWER(prevvalue) && isLOWER(ceilvalue)) || | |
10675 | (isUPPER(prevvalue) && isUPPER(ceilvalue)))) | |
ffc61ed2 | 10676 | { |
3568d838 JH |
10677 | if (isLOWER(prevvalue)) { |
10678 | for (i = prevvalue; i <= ceilvalue; i++) | |
2670d666 | 10679 | if (isLOWER(i) && !ANYOF_BITMAP_TEST(ret,i)) { |
2283d326 | 10680 | stored += |
5bfec14d | 10681 | set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate); |
2670d666 | 10682 | } |
ffc61ed2 | 10683 | } else { |
3568d838 | 10684 | for (i = prevvalue; i <= ceilvalue; i++) |
2670d666 | 10685 | if (isUPPER(i) && !ANYOF_BITMAP_TEST(ret,i)) { |
2283d326 | 10686 | stored += |
5bfec14d | 10687 | set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate); |
2670d666 | 10688 | } |
ffc61ed2 | 10689 | } |
8ada0baa | 10690 | } |
ffc61ed2 | 10691 | else |
8ada0baa | 10692 | #endif |
07be1b83 | 10693 | for (i = prevvalue; i <= ceilvalue; i++) { |
5bfec14d | 10694 | stored += set_regclass_bit(pRExC_state, ret, (U8) i, &l1_fold_invlist, &unicode_alternate); |
07be1b83 | 10695 | } |
3568d838 | 10696 | } |
56ca34ca KW |
10697 | if (value > 255) { |
10698 | const UV prevnatvalue = NATIVE_TO_UNI(prevvalue); | |
10699 | const UV natvalue = NATIVE_TO_UNI(value); | |
56ca34ca | 10700 | nonbitmap = add_range_to_invlist(nonbitmap, prevnatvalue, natvalue); |
56ca34ca | 10701 | } |
1b2d223b JH |
10702 | #ifdef EBCDIC |
10703 | literal_endpoint = 0; | |
10704 | #endif | |
8ada0baa | 10705 | } |
ffc61ed2 JH |
10706 | |
10707 | range = 0; /* this range (if it was one) is done now */ | |
a0d0e21e | 10708 | } |
ffc61ed2 | 10709 | |
ffc61ed2 | 10710 | |
7f6f358c YO |
10711 | |
10712 | if (SIZE_ONLY) | |
10713 | return ret; | |
10714 | /****** !SIZE_ONLY AFTER HERE *********/ | |
10715 | ||
0c6e4288 KW |
10716 | /* If folding and there are code points above 255, we calculate all |
10717 | * characters that could fold to or from the ones already on the list */ | |
10718 | if (FOLD && nonbitmap) { | |
0d527bf8 | 10719 | UV start, end; /* End points of code point ranges */ |
56ca34ca | 10720 | |
4065ba03 | 10721 | SV* fold_intersection = NULL; |
93e5bb1c KW |
10722 | |
10723 | /* This is a list of all the characters that participate in folds | |
10724 | * (except marks, etc in multi-char folds */ | |
10725 | if (! PL_utf8_foldable) { | |
10726 | SV* swash = swash_init("utf8", "Cased", &PL_sv_undef, 1, 0); | |
10727 | PL_utf8_foldable = _swash_to_invlist(swash); | |
10728 | } | |
56ca34ca | 10729 | |
93e5bb1c KW |
10730 | /* This is a hash that for a particular fold gives all characters |
10731 | * that are involved in it */ | |
10732 | if (! PL_utf8_foldclosures) { | |
10733 | ||
10734 | /* If we were unable to find any folds, then we likely won't be | |
10735 | * able to find the closures. So just create an empty list. | |
10736 | * Folding will effectively be restricted to the non-Unicode rules | |
10737 | * hard-coded into Perl. (This case happens legitimately during | |
10738 | * compilation of Perl itself before the Unicode tables are | |
10739 | * generated) */ | |
10740 | if (invlist_len(PL_utf8_foldable) == 0) { | |
ddc1cd80 | 10741 | PL_utf8_foldclosures = newHV(); |
93e5bb1c KW |
10742 | } else { |
10743 | /* If the folds haven't been read in, call a fold function | |
10744 | * to force that */ | |
10745 | if (! PL_utf8_tofold) { | |
10746 | U8 dummy[UTF8_MAXBYTES+1]; | |
10747 | STRLEN dummy_len; | |
f26f1b9c KW |
10748 | |
10749 | /* This particular string is above \xff in both UTF-8 and | |
10750 | * UTFEBCDIC */ | |
10751 | to_utf8_fold((U8*) "\xC8\x80", dummy, &dummy_len); | |
686c8a98 | 10752 | assert(PL_utf8_tofold); /* Verify that worked */ |
56ca34ca | 10753 | } |
93e5bb1c | 10754 | PL_utf8_foldclosures = _swash_inversion_hash(PL_utf8_tofold); |
56ca34ca | 10755 | } |
93e5bb1c KW |
10756 | } |
10757 | ||
dbe7a391 KW |
10758 | /* Only the characters in this class that participate in folds need be |
10759 | * checked. Get the intersection of this class and all the possible | |
10760 | * characters that are foldable. This can quickly narrow down a large | |
10761 | * class */ | |
37e85ffe | 10762 | _invlist_intersection(PL_utf8_foldable, nonbitmap, &fold_intersection); |
93e5bb1c KW |
10763 | |
10764 | /* Now look at the foldable characters in this class individually */ | |
0d527bf8 KW |
10765 | invlist_iterinit(fold_intersection); |
10766 | while (invlist_iternext(fold_intersection, &start, &end)) { | |
93e5bb1c KW |
10767 | UV j; |
10768 | ||
93e5bb1c KW |
10769 | /* Look at every character in the range */ |
10770 | for (j = start; j <= end; j++) { | |
10771 | ||
10772 | /* Get its fold */ | |
10773 | U8 foldbuf[UTF8_MAXBYTES_CASE+1]; | |
10774 | STRLEN foldlen; | |
827f5bb8 KW |
10775 | const UV f = |
10776 | _to_uni_fold_flags(j, foldbuf, &foldlen, allow_full_fold); | |
93e5bb1c KW |
10777 | |
10778 | if (foldlen > (STRLEN)UNISKIP(f)) { | |
10779 | ||
dbe7a391 KW |
10780 | /* Any multicharacter foldings (disallowed in lookbehind |
10781 | * patterns) require the following transform: [ABCDEF] -> | |
10782 | * (?:[ABCabcDEFd]|pq|rst) where E folds into "pq" and F | |
10783 | * folds into "rst", all other characters fold to single | |
10784 | * characters. We save away these multicharacter foldings, | |
10785 | * to be later saved as part of the additional "s" data. */ | |
93e5bb1c KW |
10786 | if (! RExC_in_lookbehind) { |
10787 | U8* loc = foldbuf; | |
10788 | U8* e = foldbuf + foldlen; | |
10789 | ||
dbe7a391 KW |
10790 | /* If any of the folded characters of this are in the |
10791 | * Latin1 range, tell the regex engine that this can | |
10792 | * match a non-utf8 target string. The only multi-byte | |
10793 | * fold whose source is in the Latin1 range (U+00DF) | |
10794 | * applies only when the target string is utf8, or | |
10795 | * under unicode rules */ | |
93e5bb1c KW |
10796 | if (j > 255 || AT_LEAST_UNI_SEMANTICS) { |
10797 | while (loc < e) { | |
10798 | ||
10799 | /* Can't mix ascii with non- under /aa */ | |
10800 | if (MORE_ASCII_RESTRICTED | |
10801 | && (isASCII(*loc) != isASCII(j))) | |
10802 | { | |
10803 | goto end_multi_fold; | |
10804 | } | |
10805 | if (UTF8_IS_INVARIANT(*loc) | |
10806 | || UTF8_IS_DOWNGRADEABLE_START(*loc)) | |
10807 | { | |
dbe7a391 KW |
10808 | /* Can't mix above and below 256 under LOC |
10809 | */ | |
93e5bb1c | 10810 | if (LOC) { |
2f7f8cb1 KW |
10811 | goto end_multi_fold; |
10812 | } | |
93e5bb1c KW |
10813 | ANYOF_FLAGS(ret) |
10814 | |= ANYOF_NONBITMAP_NON_UTF8; | |
10815 | break; | |
8e3094e5 | 10816 | } |
93e5bb1c | 10817 | loc += UTF8SKIP(loc); |
8e3094e5 | 10818 | } |
56ca34ca | 10819 | } |
17580e7a | 10820 | |
93e5bb1c KW |
10821 | add_alternate(&unicode_alternate, foldbuf, foldlen); |
10822 | end_multi_fold: ; | |
10823 | } | |
14e30abc KW |
10824 | |
10825 | /* This is special-cased, as it is the only letter which | |
10826 | * has both a multi-fold and single-fold in Latin1. All | |
10827 | * the other chars that have single and multi-folds are | |
10828 | * always in utf8, and the utf8 folding algorithm catches | |
10829 | * them */ | |
10830 | if (! LOC && j == LATIN_CAPITAL_LETTER_SHARP_S) { | |
10831 | stored += set_regclass_bit(pRExC_state, | |
10832 | ret, | |
10833 | LATIN_SMALL_LETTER_SHARP_S, | |
10834 | &l1_fold_invlist, &unicode_alternate); | |
10835 | } | |
93e5bb1c KW |
10836 | } |
10837 | else { | |
10838 | /* Single character fold. Add everything in its fold | |
dbe7a391 | 10839 | * closure to the list that this node should match */ |
93e5bb1c KW |
10840 | SV** listp; |
10841 | ||
dbe7a391 KW |
10842 | /* The fold closures data structure is a hash with the keys |
10843 | * being every character that is folded to, like 'k', and | |
10844 | * the values each an array of everything that folds to its | |
10845 | * key. e.g. [ 'k', 'K', KELVIN_SIGN ] */ | |
93e5bb1c KW |
10846 | if ((listp = hv_fetch(PL_utf8_foldclosures, |
10847 | (char *) foldbuf, foldlen, FALSE))) | |
10848 | { | |
10849 | AV* list = (AV*) *listp; | |
10850 | IV k; | |
10851 | for (k = 0; k <= av_len(list); k++) { | |
10852 | SV** c_p = av_fetch(list, k, FALSE); | |
10853 | UV c; | |
10854 | if (c_p == NULL) { | |
10855 | Perl_croak(aTHX_ "panic: invalid PL_utf8_foldclosures structure"); | |
10856 | } | |
10857 | c = SvUV(*c_p); | |
10858 | ||
dbe7a391 KW |
10859 | /* /aa doesn't allow folds between ASCII and non-; |
10860 | * /l doesn't allow them between above and below | |
10861 | * 256 */ | |
93e5bb1c KW |
10862 | if ((MORE_ASCII_RESTRICTED |
10863 | && (isASCII(c) != isASCII(j))) | |
10864 | || (LOC && ((c < 256) != (j < 256)))) | |
10865 | { | |
10866 | continue; | |
10867 | } | |
56ca34ca | 10868 | |
93e5bb1c KW |
10869 | if (c < 256 && AT_LEAST_UNI_SEMANTICS) { |
10870 | stored += set_regclass_bit(pRExC_state, | |
10871 | ret, | |
10872 | (U8) c, | |
10873 | &l1_fold_invlist, &unicode_alternate); | |
10874 | } | |
dbe7a391 KW |
10875 | /* It may be that the code point is already in |
10876 | * this range or already in the bitmap, in | |
10877 | * which case we need do nothing */ | |
93e5bb1c KW |
10878 | else if ((c < start || c > end) |
10879 | && (c > 255 | |
10880 | || ! ANYOF_BITMAP_TEST(ret, c))) | |
10881 | { | |
10882 | nonbitmap = add_cp_to_invlist(nonbitmap, c); | |
56ca34ca KW |
10883 | } |
10884 | } | |
10885 | } | |
10886 | } | |
10887 | } | |
93e5bb1c | 10888 | } |
318c430e | 10889 | SvREFCNT_dec(fold_intersection); |
56ca34ca KW |
10890 | } |
10891 | ||
53742956 KW |
10892 | /* Combine the two lists into one. */ |
10893 | if (l1_fold_invlist) { | |
10894 | if (nonbitmap) { | |
37e85ffe | 10895 | _invlist_union(nonbitmap, l1_fold_invlist, &nonbitmap); |
318c430e | 10896 | SvREFCNT_dec(l1_fold_invlist); |
53742956 KW |
10897 | } |
10898 | else { | |
10899 | nonbitmap = l1_fold_invlist; | |
10900 | } | |
10901 | } | |
10902 | ||
dc415b4a KW |
10903 | /* And combine the result (if any) with any inversion list from properties. |
10904 | * The lists are kept separate up to now because we don't want to fold the | |
10905 | * properties */ | |
10906 | if (properties) { | |
10907 | if (nonbitmap) { | |
10908 | _invlist_union(nonbitmap, properties, &nonbitmap); | |
10909 | SvREFCNT_dec(properties); | |
10910 | } | |
10911 | else { | |
10912 | nonbitmap = properties; | |
10913 | } | |
10914 | } | |
10915 | ||
10916 | ||
fb9bfbf7 | 10917 | /* Here, we have calculated what code points should be in the character |
dbe7a391 KW |
10918 | * class. |
10919 | * | |
10920 | * Now we can see about various optimizations. Fold calculation (which we | |
10921 | * did above) needs to take place before inversion. Otherwise /[^k]/i | |
10922 | * would invert to include K, which under /i would match k, which it | |
10923 | * shouldn't. */ | |
fb9bfbf7 | 10924 | |
f56b6394 | 10925 | /* Optimize inverted simple patterns (e.g. [^a-z]). Note that we haven't |
dbe7a391 | 10926 | * set the FOLD flag yet, so this does optimize those. It doesn't |
40c78556 KW |
10927 | * optimize locale. Doing so perhaps could be done as long as there is |
10928 | * nothing like \w in it; some thought also would have to be given to the | |
10929 | * interaction with above 0x100 chars */ | |
dbe7a391 KW |
10930 | if ((ANYOF_FLAGS(ret) & ANYOF_INVERT) |
10931 | && ! LOC | |
137165a6 | 10932 | && ! unicode_alternate |
2fde50e1 KW |
10933 | /* In case of /d, there are some things that should match only when in |
10934 | * not in the bitmap, i.e., they require UTF8 to match. These are | |
dbe7a391 KW |
10935 | * listed in nonbitmap, but if ANYOF_NONBITMAP_NON_UTF8 is set in this |
10936 | * case, they don't require UTF8, so can invert here */ | |
2fde50e1 KW |
10937 | && (! nonbitmap |
10938 | || ! DEPENDS_SEMANTICS | |
10939 | || (ANYOF_FLAGS(ret) & ANYOF_NONBITMAP_NON_UTF8)) | |
137165a6 KW |
10940 | && SvCUR(listsv) == initial_listsv_len) |
10941 | { | |
2fde50e1 | 10942 | if (! nonbitmap) { |
641b49c8 KW |
10943 | for (value = 0; value < ANYOF_BITMAP_SIZE; ++value) |
10944 | ANYOF_BITMAP(ret)[value] ^= 0xFF; | |
2fde50e1 KW |
10945 | /* The inversion means that everything above 255 is matched */ |
10946 | ANYOF_FLAGS(ret) |= ANYOF_UNICODE_ALL; | |
10947 | } | |
10948 | else { | |
10949 | /* Here, also has things outside the bitmap. Go through each bit | |
10950 | * individually and add it to the list to get rid of from those | |
10951 | * things not in the bitmap */ | |
10952 | SV *remove_list = _new_invlist(2); | |
37e85ffe | 10953 | _invlist_invert(nonbitmap); |
2fde50e1 KW |
10954 | for (value = 0; value < 256; ++value) { |
10955 | if (ANYOF_BITMAP_TEST(ret, value)) { | |
10956 | ANYOF_BITMAP_CLEAR(ret, value); | |
10957 | remove_list = add_cp_to_invlist(remove_list, value); | |
10958 | } | |
10959 | else { | |
10960 | ANYOF_BITMAP_SET(ret, value); | |
10961 | } | |
10962 | } | |
dbe7a391 KW |
10963 | |
10964 | /* And do the removal */ | |
37e85ffe | 10965 | _invlist_subtract(nonbitmap, remove_list, &nonbitmap); |
2fde50e1 KW |
10966 | SvREFCNT_dec(remove_list); |
10967 | } | |
10968 | ||
40c78556 KW |
10969 | stored = 256 - stored; |
10970 | ||
2fde50e1 KW |
10971 | /* Clear the invert flag since have just done it here */ |
10972 | ANYOF_FLAGS(ret) &= ~ANYOF_INVERT; | |
40c78556 KW |
10973 | } |
10974 | ||
0222889f KW |
10975 | /* Folding in the bitmap is taken care of above, but not for locale (for |
10976 | * which we have to wait to see what folding is in effect at runtime), and | |
10977 | * for things not in the bitmap. Set run-time fold flag for these */ | |
53742956 | 10978 | if (FOLD && (LOC || nonbitmap || unicode_alternate)) { |
0222889f | 10979 | ANYOF_FLAGS(ret) |= ANYOF_LOC_NONBITMAP_FOLD; |
f56b6394 KW |
10980 | } |
10981 | ||
2786be71 KW |
10982 | /* A single character class can be "optimized" into an EXACTish node. |
10983 | * Note that since we don't currently count how many characters there are | |
10984 | * outside the bitmap, we are XXX missing optimization possibilities for | |
10985 | * them. This optimization can't happen unless this is a truly single | |
10986 | * character class, which means that it can't be an inversion into a | |
10987 | * many-character class, and there must be no possibility of there being | |
10988 | * things outside the bitmap. 'stored' (only) for locales doesn't include | |
6da63e10 KW |
10989 | * \w, etc, so have to make a special test that they aren't present |
10990 | * | |
10991 | * Similarly A 2-character class of the very special form like [bB] can be | |
10992 | * optimized into an EXACTFish node, but only for non-locales, and for | |
10993 | * characters which only have the two folds; so things like 'fF' and 'Ii' | |
10994 | * wouldn't work because they are part of the fold of 'LATIN SMALL LIGATURE | |
10995 | * FI'. */ | |
137165a6 | 10996 | if (! nonbitmap |
53742956 | 10997 | && ! unicode_alternate |
137165a6 KW |
10998 | && SvCUR(listsv) == initial_listsv_len |
10999 | && ! (ANYOF_FLAGS(ret) & (ANYOF_INVERT|ANYOF_UNICODE_ALL)) | |
6da63e10 KW |
11000 | && (((stored == 1 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE)) |
11001 | || (! ANYOF_CLASS_TEST_ANY_SET(ret))))) | |
11002 | || (stored == 2 && ((! (ANYOF_FLAGS(ret) & ANYOF_LOCALE)) | |
11003 | && (! _HAS_NONLATIN1_FOLD_CLOSURE_ONLY_FOR_USE_BY_REGCOMP_DOT_C_AND_REGEXEC_DOT_C(value)) | |
11004 | /* If the latest code point has a fold whose | |
11005 | * bit is set, it must be the only other one */ | |
2dcac756 | 11006 | && ((prevvalue = PL_fold_latin1[value]) != (IV)value) |
6da63e10 | 11007 | && ANYOF_BITMAP_TEST(ret, prevvalue))))) |
2786be71 KW |
11008 | { |
11009 | /* Note that the information needed to decide to do this optimization | |
11010 | * is not currently available until the 2nd pass, and that the actually | |
6da63e10 KW |
11011 | * used EXACTish node takes less space than the calculated ANYOF node, |
11012 | * and hence the amount of space calculated in the first pass is larger | |
2786be71 KW |
11013 | * than actually used, so this optimization doesn't gain us any space. |
11014 | * But an EXACT node is faster than an ANYOF node, and can be combined | |
11015 | * with any adjacent EXACT nodes later by the optimizer for further | |
6da63e10 KW |
11016 | * gains. The speed of executing an EXACTF is similar to an ANYOF |
11017 | * node, so the optimization advantage comes from the ability to join | |
11018 | * it to adjacent EXACT nodes */ | |
2786be71 | 11019 | |
07be1b83 | 11020 | const char * cur_parse= RExC_parse; |
6da63e10 | 11021 | U8 op; |
07be1b83 YO |
11022 | RExC_emit = (regnode *)orig_emit; |
11023 | RExC_parse = (char *)orig_parse; | |
2786be71 | 11024 | |
6da63e10 KW |
11025 | if (stored == 1) { |
11026 | ||
11027 | /* A locale node with one point can be folded; all the other cases | |
11028 | * with folding will have two points, since we calculate them above | |
11029 | */ | |
39065660 | 11030 | if (ANYOF_FLAGS(ret) & ANYOF_LOC_NONBITMAP_FOLD) { |
6da63e10 KW |
11031 | op = EXACTFL; |
11032 | } | |
11033 | else { | |
11034 | op = EXACT; | |
11035 | } | |
b36527fc KW |
11036 | } |
11037 | else { /* else 2 chars in the bit map: the folds of each other */ | |
11038 | ||
11039 | /* Use the folded value, which for the cases where we get here, | |
11040 | * is just the lower case of the current one (which may resolve to | |
11041 | * itself, or to the other one */ | |
11042 | value = toLOWER_LATIN1(value); | |
11043 | if (AT_LEAST_UNI_SEMANTICS || !isASCII(value)) { | |
6da63e10 | 11044 | |
e62862f6 KW |
11045 | /* To join adjacent nodes, they must be the exact EXACTish |
11046 | * type. Try to use the most likely type, by using EXACTFU if | |
11047 | * the regex calls for them, or is required because the | |
11048 | * character is non-ASCII */ | |
11049 | op = EXACTFU; | |
11050 | } | |
11051 | else { /* Otherwise, more likely to be EXACTF type */ | |
11052 | op = EXACTF; | |
11053 | } | |
b36527fc | 11054 | } |
6da63e10 KW |
11055 | |
11056 | ret = reg_node(pRExC_state, op); | |
07be1b83 | 11057 | RExC_parse = (char *)cur_parse; |
2786be71 KW |
11058 | if (UTF && ! NATIVE_IS_INVARIANT(value)) { |
11059 | *STRING(ret)= UTF8_EIGHT_BIT_HI((U8) value); | |
11060 | *(STRING(ret) + 1)= UTF8_EIGHT_BIT_LO((U8) value); | |
11061 | STR_LEN(ret)= 2; | |
11062 | RExC_emit += STR_SZ(2); | |
11063 | } | |
11064 | else { | |
11065 | *STRING(ret)= (char)value; | |
11066 | STR_LEN(ret)= 1; | |
11067 | RExC_emit += STR_SZ(1); | |
11068 | } | |
ef8d46e8 | 11069 | SvREFCNT_dec(listsv); |
7f6f358c YO |
11070 | return ret; |
11071 | } | |
ffc61ed2 | 11072 | |
dc415b4a KW |
11073 | /* If there is a swash and more than one element, we can't use the swash in |
11074 | * the optimization below. */ | |
11075 | if (swash && element_count > 1) { | |
11076 | SvREFCNT_dec(swash); | |
11077 | swash = NULL; | |
11078 | } | |
c16787fd KW |
11079 | if (! nonbitmap |
11080 | && SvCUR(listsv) == initial_listsv_len | |
11081 | && ! unicode_alternate) | |
11082 | { | |
137165a6 KW |
11083 | ARG_SET(ret, ANYOF_NONBITMAP_EMPTY); |
11084 | SvREFCNT_dec(listsv); | |
11085 | SvREFCNT_dec(unicode_alternate); | |
11086 | } | |
11087 | else { | |
11088 | ||
097eb12c | 11089 | AV * const av = newAV(); |
ffc61ed2 | 11090 | SV *rv; |
9e55ce06 | 11091 | /* The 0th element stores the character class description |
6a0407ee | 11092 | * in its textual form: used later (regexec.c:Perl_regclass_swash()) |
9e55ce06 | 11093 | * to initialize the appropriate swash (which gets stored in |
dbe7a391 KW |
11094 | * element [1]), and also useful for dumping the regnode. |
11095 | * Element [2] stores the multicharacter foldings, | |
c16787fd KW |
11096 | * used later (regexec.c:S_reginclass()). |
11097 | * Element [3] stores the nonbitmap inversion list for use in addition | |
11098 | * or instead of element [0]. | |
dc415b4a KW |
11099 | * Element [4] is set if any component of the class is from a |
11100 | * user-defined property */ | |
c16787fd KW |
11101 | av_store(av, 0, (SvCUR(listsv) == initial_listsv_len) |
11102 | ? &PL_sv_undef | |
11103 | : listsv); | |
88d45d28 KW |
11104 | if (swash) { |
11105 | av_store(av, 1, swash); | |
11106 | SvREFCNT_dec(nonbitmap); | |
11107 | } | |
11108 | else { | |
11109 | av_store(av, 1, NULL); | |
c16787fd KW |
11110 | if (nonbitmap) { |
11111 | av_store(av, 3, nonbitmap); | |
dc415b4a | 11112 | av_store(av, 4, newSVuv(has_user_defined_property)); |
c16787fd | 11113 | } |
88d45d28 | 11114 | } |
827f5bb8 KW |
11115 | |
11116 | /* Store any computed multi-char folds only if we are allowing | |
11117 | * them */ | |
11118 | if (allow_full_fold) { | |
7b4a7e58 KW |
11119 | av_store(av, 2, MUTABLE_SV(unicode_alternate)); |
11120 | if (unicode_alternate) { /* This node is variable length */ | |
11121 | OP(ret) = ANYOFV; | |
11122 | } | |
827f5bb8 KW |
11123 | } |
11124 | else { | |
11125 | av_store(av, 2, NULL); | |
11126 | } | |
ad64d0ec | 11127 | rv = newRV_noinc(MUTABLE_SV(av)); |
19860706 | 11128 | n = add_data(pRExC_state, 1, "s"); |
f8fc2ecf | 11129 | RExC_rxi->data->data[n] = (void*)rv; |
ffc61ed2 | 11130 | ARG_SET(ret, n); |
a0ed51b3 | 11131 | } |
a0ed51b3 LW |
11132 | return ret; |
11133 | } | |
89836f1f YO |
11134 | #undef _C_C_T_ |
11135 | ||
a0ed51b3 | 11136 | |
bcdf7404 YO |
11137 | /* reg_skipcomment() |
11138 | ||
11139 | Absorbs an /x style # comments from the input stream. | |
11140 | Returns true if there is more text remaining in the stream. | |
11141 | Will set the REG_SEEN_RUN_ON_COMMENT flag if the comment | |
11142 | terminates the pattern without including a newline. | |
11143 | ||
11144 | Note its the callers responsibility to ensure that we are | |
11145 | actually in /x mode | |
11146 | ||
11147 | */ | |
11148 | ||
11149 | STATIC bool | |
11150 | S_reg_skipcomment(pTHX_ RExC_state_t *pRExC_state) | |
11151 | { | |
11152 | bool ended = 0; | |
7918f24d NC |
11153 | |
11154 | PERL_ARGS_ASSERT_REG_SKIPCOMMENT; | |
11155 | ||
bcdf7404 YO |
11156 | while (RExC_parse < RExC_end) |
11157 | if (*RExC_parse++ == '\n') { | |
11158 | ended = 1; | |
11159 | break; | |
11160 | } | |
11161 | if (!ended) { | |
11162 | /* we ran off the end of the pattern without ending | |
11163 | the comment, so we have to add an \n when wrapping */ | |
11164 | RExC_seen |= REG_SEEN_RUN_ON_COMMENT; | |
11165 | return 0; | |
11166 | } else | |
11167 | return 1; | |
11168 | } | |
11169 | ||
11170 | /* nextchar() | |
11171 | ||
3b753521 | 11172 | Advances the parse position, and optionally absorbs |
bcdf7404 YO |
11173 | "whitespace" from the inputstream. |
11174 | ||
11175 | Without /x "whitespace" means (?#...) style comments only, | |
11176 | with /x this means (?#...) and # comments and whitespace proper. | |
11177 | ||
11178 | Returns the RExC_parse point from BEFORE the scan occurs. | |
11179 | ||
11180 | This is the /x friendly way of saying RExC_parse++. | |
11181 | */ | |
11182 | ||
76e3520e | 11183 | STATIC char* |
830247a4 | 11184 | S_nextchar(pTHX_ RExC_state_t *pRExC_state) |
a0d0e21e | 11185 | { |
097eb12c | 11186 | char* const retval = RExC_parse++; |
a0d0e21e | 11187 | |
7918f24d NC |
11188 | PERL_ARGS_ASSERT_NEXTCHAR; |
11189 | ||
4633a7c4 | 11190 | for (;;) { |
d224c965 KW |
11191 | if (RExC_end - RExC_parse >= 3 |
11192 | && *RExC_parse == '(' | |
11193 | && RExC_parse[1] == '?' | |
11194 | && RExC_parse[2] == '#') | |
11195 | { | |
e994fd66 AE |
11196 | while (*RExC_parse != ')') { |
11197 | if (RExC_parse == RExC_end) | |
11198 | FAIL("Sequence (?#... not terminated"); | |
830247a4 | 11199 | RExC_parse++; |
e994fd66 | 11200 | } |
830247a4 | 11201 | RExC_parse++; |
4633a7c4 LW |
11202 | continue; |
11203 | } | |
bbe252da | 11204 | if (RExC_flags & RXf_PMf_EXTENDED) { |
830247a4 IZ |
11205 | if (isSPACE(*RExC_parse)) { |
11206 | RExC_parse++; | |
748a9306 LW |
11207 | continue; |
11208 | } | |
830247a4 | 11209 | else if (*RExC_parse == '#') { |
bcdf7404 YO |
11210 | if ( reg_skipcomment( pRExC_state ) ) |
11211 | continue; | |
748a9306 | 11212 | } |
748a9306 | 11213 | } |
4633a7c4 | 11214 | return retval; |
a0d0e21e | 11215 | } |
a687059c LW |
11216 | } |
11217 | ||
11218 | /* | |
c277df42 | 11219 | - reg_node - emit a node |
a0d0e21e | 11220 | */ |
76e3520e | 11221 | STATIC regnode * /* Location. */ |
830247a4 | 11222 | S_reg_node(pTHX_ RExC_state_t *pRExC_state, U8 op) |
a687059c | 11223 | { |
97aff369 | 11224 | dVAR; |
c277df42 | 11225 | register regnode *ptr; |
504618e9 | 11226 | regnode * const ret = RExC_emit; |
07be1b83 | 11227 | GET_RE_DEBUG_FLAGS_DECL; |
a687059c | 11228 | |
7918f24d NC |
11229 | PERL_ARGS_ASSERT_REG_NODE; |
11230 | ||
c277df42 | 11231 | if (SIZE_ONLY) { |
830247a4 IZ |
11232 | SIZE_ALIGN(RExC_size); |
11233 | RExC_size += 1; | |
a0d0e21e LW |
11234 | return(ret); |
11235 | } | |
3b57cd43 YO |
11236 | if (RExC_emit >= RExC_emit_bound) |
11237 | Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op); | |
11238 | ||
c277df42 | 11239 | NODE_ALIGN_FILL(ret); |
a0d0e21e | 11240 | ptr = ret; |
c277df42 | 11241 | FILL_ADVANCE_NODE(ptr, op); |
7122b237 | 11242 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 11243 | if (RExC_offsets) { /* MJD */ |
07be1b83 | 11244 | MJD_OFFSET_DEBUG(("%s:%d: (op %s) %s %"UVuf" (len %"UVuf") (max %"UVuf").\n", |
fac92740 | 11245 | "reg_node", __LINE__, |
13d6edb4 | 11246 | PL_reg_name[op], |
07be1b83 YO |
11247 | (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] |
11248 | ? "Overwriting end of array!\n" : "OK", | |
11249 | (UV)(RExC_emit - RExC_emit_start), | |
11250 | (UV)(RExC_parse - RExC_start), | |
11251 | (UV)RExC_offsets[0])); | |
ccb2c380 | 11252 | Set_Node_Offset(RExC_emit, RExC_parse + (op == END)); |
fac92740 | 11253 | } |
7122b237 | 11254 | #endif |
830247a4 | 11255 | RExC_emit = ptr; |
a0d0e21e | 11256 | return(ret); |
a687059c LW |
11257 | } |
11258 | ||
11259 | /* | |
a0d0e21e LW |
11260 | - reganode - emit a node with an argument |
11261 | */ | |
76e3520e | 11262 | STATIC regnode * /* Location. */ |
830247a4 | 11263 | S_reganode(pTHX_ RExC_state_t *pRExC_state, U8 op, U32 arg) |
fe14fcc3 | 11264 | { |
97aff369 | 11265 | dVAR; |
c277df42 | 11266 | register regnode *ptr; |
504618e9 | 11267 | regnode * const ret = RExC_emit; |
07be1b83 | 11268 | GET_RE_DEBUG_FLAGS_DECL; |
fe14fcc3 | 11269 | |
7918f24d NC |
11270 | PERL_ARGS_ASSERT_REGANODE; |
11271 | ||
c277df42 | 11272 | if (SIZE_ONLY) { |
830247a4 IZ |
11273 | SIZE_ALIGN(RExC_size); |
11274 | RExC_size += 2; | |
6bda09f9 YO |
11275 | /* |
11276 | We can't do this: | |
11277 | ||
11278 | assert(2==regarglen[op]+1); | |
686b73d4 | 11279 | |
6bda09f9 YO |
11280 | Anything larger than this has to allocate the extra amount. |
11281 | If we changed this to be: | |
11282 | ||
11283 | RExC_size += (1 + regarglen[op]); | |
11284 | ||
11285 | then it wouldn't matter. Its not clear what side effect | |
11286 | might come from that so its not done so far. | |
11287 | -- dmq | |
11288 | */ | |
a0d0e21e LW |
11289 | return(ret); |
11290 | } | |
3b57cd43 YO |
11291 | if (RExC_emit >= RExC_emit_bound) |
11292 | Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %d", op); | |
11293 | ||
c277df42 | 11294 | NODE_ALIGN_FILL(ret); |
a0d0e21e | 11295 | ptr = ret; |
c277df42 | 11296 | FILL_ADVANCE_NODE_ARG(ptr, op, arg); |
7122b237 | 11297 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 11298 | if (RExC_offsets) { /* MJD */ |
07be1b83 | 11299 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n", |
fac92740 | 11300 | "reganode", |
ccb2c380 | 11301 | __LINE__, |
13d6edb4 | 11302 | PL_reg_name[op], |
07be1b83 | 11303 | (UV)(RExC_emit - RExC_emit_start) > RExC_offsets[0] ? |
fac92740 | 11304 | "Overwriting end of array!\n" : "OK", |
07be1b83 YO |
11305 | (UV)(RExC_emit - RExC_emit_start), |
11306 | (UV)(RExC_parse - RExC_start), | |
11307 | (UV)RExC_offsets[0])); | |
ccb2c380 | 11308 | Set_Cur_Node_Offset; |
fac92740 | 11309 | } |
7122b237 | 11310 | #endif |
830247a4 | 11311 | RExC_emit = ptr; |
a0d0e21e | 11312 | return(ret); |
fe14fcc3 LW |
11313 | } |
11314 | ||
11315 | /* | |
cd439c50 | 11316 | - reguni - emit (if appropriate) a Unicode character |
a0ed51b3 | 11317 | */ |
71207a34 AL |
11318 | STATIC STRLEN |
11319 | S_reguni(pTHX_ const RExC_state_t *pRExC_state, UV uv, char* s) | |
a0ed51b3 | 11320 | { |
97aff369 | 11321 | dVAR; |
7918f24d NC |
11322 | |
11323 | PERL_ARGS_ASSERT_REGUNI; | |
11324 | ||
71207a34 | 11325 | return SIZE_ONLY ? UNISKIP(uv) : (uvchr_to_utf8((U8*)s, uv) - (U8*)s); |
a0ed51b3 LW |
11326 | } |
11327 | ||
11328 | /* | |
a0d0e21e LW |
11329 | - reginsert - insert an operator in front of already-emitted operand |
11330 | * | |
11331 | * Means relocating the operand. | |
11332 | */ | |
76e3520e | 11333 | STATIC void |
6bda09f9 | 11334 | S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth) |
a687059c | 11335 | { |
97aff369 | 11336 | dVAR; |
c277df42 IZ |
11337 | register regnode *src; |
11338 | register regnode *dst; | |
11339 | register regnode *place; | |
504618e9 | 11340 | const int offset = regarglen[(U8)op]; |
6bda09f9 | 11341 | const int size = NODE_STEP_REGNODE + offset; |
07be1b83 | 11342 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
11343 | |
11344 | PERL_ARGS_ASSERT_REGINSERT; | |
def51078 | 11345 | PERL_UNUSED_ARG(depth); |
22c35a8c | 11346 | /* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */ |
13d6edb4 | 11347 | DEBUG_PARSE_FMT("inst"," - %s",PL_reg_name[op]); |
c277df42 | 11348 | if (SIZE_ONLY) { |
6bda09f9 | 11349 | RExC_size += size; |
a0d0e21e LW |
11350 | return; |
11351 | } | |
a687059c | 11352 | |
830247a4 | 11353 | src = RExC_emit; |
6bda09f9 | 11354 | RExC_emit += size; |
830247a4 | 11355 | dst = RExC_emit; |
40d049e4 | 11356 | if (RExC_open_parens) { |
6bda09f9 | 11357 | int paren; |
3b57cd43 | 11358 | /*DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);*/ |
6bda09f9 | 11359 | for ( paren=0 ; paren < RExC_npar ; paren++ ) { |
40d049e4 | 11360 | if ( RExC_open_parens[paren] >= opnd ) { |
3b57cd43 | 11361 | /*DEBUG_PARSE_FMT("open"," - %d",size);*/ |
40d049e4 YO |
11362 | RExC_open_parens[paren] += size; |
11363 | } else { | |
3b57cd43 | 11364 | /*DEBUG_PARSE_FMT("open"," - %s","ok");*/ |
40d049e4 YO |
11365 | } |
11366 | if ( RExC_close_parens[paren] >= opnd ) { | |
3b57cd43 | 11367 | /*DEBUG_PARSE_FMT("close"," - %d",size);*/ |
40d049e4 YO |
11368 | RExC_close_parens[paren] += size; |
11369 | } else { | |
3b57cd43 | 11370 | /*DEBUG_PARSE_FMT("close"," - %s","ok");*/ |
40d049e4 YO |
11371 | } |
11372 | } | |
6bda09f9 | 11373 | } |
40d049e4 | 11374 | |
fac92740 | 11375 | while (src > opnd) { |
c277df42 | 11376 | StructCopy(--src, --dst, regnode); |
7122b237 | 11377 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 11378 | if (RExC_offsets) { /* MJD 20010112 */ |
07be1b83 | 11379 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s copy %"UVuf" -> %"UVuf" (max %"UVuf").\n", |
fac92740 | 11380 | "reg_insert", |
ccb2c380 | 11381 | __LINE__, |
13d6edb4 | 11382 | PL_reg_name[op], |
07be1b83 YO |
11383 | (UV)(dst - RExC_emit_start) > RExC_offsets[0] |
11384 | ? "Overwriting end of array!\n" : "OK", | |
11385 | (UV)(src - RExC_emit_start), | |
11386 | (UV)(dst - RExC_emit_start), | |
11387 | (UV)RExC_offsets[0])); | |
ccb2c380 MP |
11388 | Set_Node_Offset_To_R(dst-RExC_emit_start, Node_Offset(src)); |
11389 | Set_Node_Length_To_R(dst-RExC_emit_start, Node_Length(src)); | |
fac92740 | 11390 | } |
7122b237 | 11391 | #endif |
fac92740 MJD |
11392 | } |
11393 | ||
a0d0e21e LW |
11394 | |
11395 | place = opnd; /* Op node, where operand used to be. */ | |
7122b237 | 11396 | #ifdef RE_TRACK_PATTERN_OFFSETS |
fac92740 | 11397 | if (RExC_offsets) { /* MJD */ |
07be1b83 | 11398 | MJD_OFFSET_DEBUG(("%s(%d): (op %s) %s %"UVuf" <- %"UVuf" (max %"UVuf").\n", |
fac92740 | 11399 | "reginsert", |
ccb2c380 | 11400 | __LINE__, |
13d6edb4 | 11401 | PL_reg_name[op], |
07be1b83 | 11402 | (UV)(place - RExC_emit_start) > RExC_offsets[0] |
fac92740 | 11403 | ? "Overwriting end of array!\n" : "OK", |
07be1b83 YO |
11404 | (UV)(place - RExC_emit_start), |
11405 | (UV)(RExC_parse - RExC_start), | |
786e8c11 | 11406 | (UV)RExC_offsets[0])); |
ccb2c380 | 11407 | Set_Node_Offset(place, RExC_parse); |
45948336 | 11408 | Set_Node_Length(place, 1); |
fac92740 | 11409 | } |
7122b237 | 11410 | #endif |
c277df42 IZ |
11411 | src = NEXTOPER(place); |
11412 | FILL_ADVANCE_NODE(place, op); | |
11413 | Zero(src, offset, regnode); | |
a687059c LW |
11414 | } |
11415 | ||
11416 | /* | |
c277df42 | 11417 | - regtail - set the next-pointer at the end of a node chain of p to val. |
3dab1dad | 11418 | - SEE ALSO: regtail_study |
a0d0e21e | 11419 | */ |
097eb12c | 11420 | /* TODO: All three parms should be const */ |
76e3520e | 11421 | STATIC void |
3dab1dad | 11422 | S_regtail(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth) |
a687059c | 11423 | { |
97aff369 | 11424 | dVAR; |
c277df42 | 11425 | register regnode *scan; |
72f13be8 | 11426 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
11427 | |
11428 | PERL_ARGS_ASSERT_REGTAIL; | |
f9049ba1 SP |
11429 | #ifndef DEBUGGING |
11430 | PERL_UNUSED_ARG(depth); | |
11431 | #endif | |
a0d0e21e | 11432 | |
c277df42 | 11433 | if (SIZE_ONLY) |
a0d0e21e LW |
11434 | return; |
11435 | ||
11436 | /* Find last node. */ | |
11437 | scan = p; | |
11438 | for (;;) { | |
504618e9 | 11439 | regnode * const temp = regnext(scan); |
3dab1dad YO |
11440 | DEBUG_PARSE_r({ |
11441 | SV * const mysv=sv_newmortal(); | |
11442 | DEBUG_PARSE_MSG((scan==p ? "tail" : "")); | |
11443 | regprop(RExC_rx, mysv, scan); | |
eaf3ca90 YO |
11444 | PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n", |
11445 | SvPV_nolen_const(mysv), REG_NODE_NUM(scan), | |
11446 | (temp == NULL ? "->" : ""), | |
13d6edb4 | 11447 | (temp == NULL ? PL_reg_name[OP(val)] : "") |
eaf3ca90 | 11448 | ); |
3dab1dad YO |
11449 | }); |
11450 | if (temp == NULL) | |
11451 | break; | |
11452 | scan = temp; | |
11453 | } | |
11454 | ||
11455 | if (reg_off_by_arg[OP(scan)]) { | |
11456 | ARG_SET(scan, val - scan); | |
11457 | } | |
11458 | else { | |
11459 | NEXT_OFF(scan) = val - scan; | |
11460 | } | |
11461 | } | |
11462 | ||
07be1b83 | 11463 | #ifdef DEBUGGING |
3dab1dad YO |
11464 | /* |
11465 | - regtail_study - set the next-pointer at the end of a node chain of p to val. | |
11466 | - Look for optimizable sequences at the same time. | |
11467 | - currently only looks for EXACT chains. | |
07be1b83 | 11468 | |
486ec47a | 11469 | This is experimental code. The idea is to use this routine to perform |
07be1b83 YO |
11470 | in place optimizations on branches and groups as they are constructed, |
11471 | with the long term intention of removing optimization from study_chunk so | |
11472 | that it is purely analytical. | |
11473 | ||
11474 | Currently only used when in DEBUG mode. The macro REGTAIL_STUDY() is used | |
11475 | to control which is which. | |
11476 | ||
3dab1dad YO |
11477 | */ |
11478 | /* TODO: All four parms should be const */ | |
07be1b83 | 11479 | |
3dab1dad YO |
11480 | STATIC U8 |
11481 | S_regtail_study(pTHX_ RExC_state_t *pRExC_state, regnode *p, const regnode *val,U32 depth) | |
11482 | { | |
11483 | dVAR; | |
11484 | register regnode *scan; | |
07be1b83 YO |
11485 | U8 exact = PSEUDO; |
11486 | #ifdef EXPERIMENTAL_INPLACESCAN | |
11487 | I32 min = 0; | |
11488 | #endif | |
3dab1dad YO |
11489 | GET_RE_DEBUG_FLAGS_DECL; |
11490 | ||
7918f24d NC |
11491 | PERL_ARGS_ASSERT_REGTAIL_STUDY; |
11492 | ||
07be1b83 | 11493 | |
3dab1dad YO |
11494 | if (SIZE_ONLY) |
11495 | return exact; | |
11496 | ||
11497 | /* Find last node. */ | |
11498 | ||
11499 | scan = p; | |
11500 | for (;;) { | |
11501 | regnode * const temp = regnext(scan); | |
07be1b83 YO |
11502 | #ifdef EXPERIMENTAL_INPLACESCAN |
11503 | if (PL_regkind[OP(scan)] == EXACT) | |
11504 | if (join_exact(pRExC_state,scan,&min,1,val,depth+1)) | |
11505 | return EXACT; | |
11506 | #endif | |
3dab1dad YO |
11507 | if ( exact ) { |
11508 | switch (OP(scan)) { | |
11509 | case EXACT: | |
11510 | case EXACTF: | |
2f7f8cb1 | 11511 | case EXACTFA: |
2c2b7f86 | 11512 | case EXACTFU: |
3dab1dad YO |
11513 | case EXACTFL: |
11514 | if( exact == PSEUDO ) | |
11515 | exact= OP(scan); | |
07be1b83 YO |
11516 | else if ( exact != OP(scan) ) |
11517 | exact= 0; | |
3dab1dad YO |
11518 | case NOTHING: |
11519 | break; | |
11520 | default: | |
11521 | exact= 0; | |
11522 | } | |
11523 | } | |
11524 | DEBUG_PARSE_r({ | |
11525 | SV * const mysv=sv_newmortal(); | |
11526 | DEBUG_PARSE_MSG((scan==p ? "tsdy" : "")); | |
11527 | regprop(RExC_rx, mysv, scan); | |
eaf3ca90 | 11528 | PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n", |
3dab1dad | 11529 | SvPV_nolen_const(mysv), |
eaf3ca90 | 11530 | REG_NODE_NUM(scan), |
13d6edb4 | 11531 | PL_reg_name[exact]); |
3dab1dad | 11532 | }); |
a0d0e21e LW |
11533 | if (temp == NULL) |
11534 | break; | |
11535 | scan = temp; | |
11536 | } | |
07be1b83 YO |
11537 | DEBUG_PARSE_r({ |
11538 | SV * const mysv_val=sv_newmortal(); | |
11539 | DEBUG_PARSE_MSG(""); | |
11540 | regprop(RExC_rx, mysv_val, val); | |
70685ca0 JH |
11541 | PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n", |
11542 | SvPV_nolen_const(mysv_val), | |
11543 | (IV)REG_NODE_NUM(val), | |
11544 | (IV)(val - scan) | |
07be1b83 YO |
11545 | ); |
11546 | }); | |
c277df42 IZ |
11547 | if (reg_off_by_arg[OP(scan)]) { |
11548 | ARG_SET(scan, val - scan); | |
a0ed51b3 LW |
11549 | } |
11550 | else { | |
c277df42 IZ |
11551 | NEXT_OFF(scan) = val - scan; |
11552 | } | |
3dab1dad YO |
11553 | |
11554 | return exact; | |
a687059c | 11555 | } |
07be1b83 | 11556 | #endif |
a687059c LW |
11557 | |
11558 | /* | |
fd181c75 | 11559 | - regdump - dump a regexp onto Perl_debug_log in vaguely comprehensible form |
a687059c | 11560 | */ |
f7819f85 | 11561 | #ifdef DEBUGGING |
c33269f7 | 11562 | static void |
7918f24d NC |
11563 | S_regdump_extflags(pTHX_ const char *lead, const U32 flags) |
11564 | { | |
f7819f85 A |
11565 | int bit; |
11566 | int set=0; | |
a62b1201 | 11567 | regex_charset cs; |
7918f24d | 11568 | |
f7819f85 A |
11569 | for (bit=0; bit<32; bit++) { |
11570 | if (flags & (1<<bit)) { | |
a62b1201 KW |
11571 | if ((1<<bit) & RXf_PMf_CHARSET) { /* Output separately, below */ |
11572 | continue; | |
11573 | } | |
f7819f85 A |
11574 | if (!set++ && lead) |
11575 | PerlIO_printf(Perl_debug_log, "%s",lead); | |
11576 | PerlIO_printf(Perl_debug_log, "%s ",PL_reg_extflags_name[bit]); | |
11577 | } | |
11578 | } | |
a62b1201 KW |
11579 | if ((cs = get_regex_charset(flags)) != REGEX_DEPENDS_CHARSET) { |
11580 | if (!set++ && lead) { | |
11581 | PerlIO_printf(Perl_debug_log, "%s",lead); | |
11582 | } | |
11583 | switch (cs) { | |
11584 | case REGEX_UNICODE_CHARSET: | |
11585 | PerlIO_printf(Perl_debug_log, "UNICODE"); | |
11586 | break; | |
11587 | case REGEX_LOCALE_CHARSET: | |
11588 | PerlIO_printf(Perl_debug_log, "LOCALE"); | |
11589 | break; | |
cfaf538b KW |
11590 | case REGEX_ASCII_RESTRICTED_CHARSET: |
11591 | PerlIO_printf(Perl_debug_log, "ASCII-RESTRICTED"); | |
11592 | break; | |
2f7f8cb1 KW |
11593 | case REGEX_ASCII_MORE_RESTRICTED_CHARSET: |
11594 | PerlIO_printf(Perl_debug_log, "ASCII-MORE_RESTRICTED"); | |
11595 | break; | |
a62b1201 KW |
11596 | default: |
11597 | PerlIO_printf(Perl_debug_log, "UNKNOWN CHARACTER SET"); | |
11598 | break; | |
11599 | } | |
11600 | } | |
f7819f85 A |
11601 | if (lead) { |
11602 | if (set) | |
11603 | PerlIO_printf(Perl_debug_log, "\n"); | |
11604 | else | |
11605 | PerlIO_printf(Perl_debug_log, "%s[none-set]\n",lead); | |
11606 | } | |
11607 | } | |
11608 | #endif | |
11609 | ||
a687059c | 11610 | void |
097eb12c | 11611 | Perl_regdump(pTHX_ const regexp *r) |
a687059c | 11612 | { |
35ff7856 | 11613 | #ifdef DEBUGGING |
97aff369 | 11614 | dVAR; |
c445ea15 | 11615 | SV * const sv = sv_newmortal(); |
ab3bbdeb | 11616 | SV *dsv= sv_newmortal(); |
f8fc2ecf | 11617 | RXi_GET_DECL(r,ri); |
f7819f85 | 11618 | GET_RE_DEBUG_FLAGS_DECL; |
a687059c | 11619 | |
7918f24d NC |
11620 | PERL_ARGS_ASSERT_REGDUMP; |
11621 | ||
f8fc2ecf | 11622 | (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0); |
a0d0e21e LW |
11623 | |
11624 | /* Header fields of interest. */ | |
ab3bbdeb YO |
11625 | if (r->anchored_substr) { |
11626 | RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr), | |
11627 | RE_SV_DUMPLEN(r->anchored_substr), 30); | |
7b0972df | 11628 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
11629 | "anchored %s%s at %"IVdf" ", |
11630 | s, RE_SV_TAIL(r->anchored_substr), | |
7b0972df | 11631 | (IV)r->anchored_offset); |
ab3bbdeb YO |
11632 | } else if (r->anchored_utf8) { |
11633 | RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8), | |
11634 | RE_SV_DUMPLEN(r->anchored_utf8), 30); | |
33b8afdf | 11635 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
11636 | "anchored utf8 %s%s at %"IVdf" ", |
11637 | s, RE_SV_TAIL(r->anchored_utf8), | |
33b8afdf | 11638 | (IV)r->anchored_offset); |
ab3bbdeb YO |
11639 | } |
11640 | if (r->float_substr) { | |
11641 | RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr), | |
11642 | RE_SV_DUMPLEN(r->float_substr), 30); | |
7b0972df | 11643 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
11644 | "floating %s%s at %"IVdf"..%"UVuf" ", |
11645 | s, RE_SV_TAIL(r->float_substr), | |
7b0972df | 11646 | (IV)r->float_min_offset, (UV)r->float_max_offset); |
ab3bbdeb YO |
11647 | } else if (r->float_utf8) { |
11648 | RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8), | |
11649 | RE_SV_DUMPLEN(r->float_utf8), 30); | |
33b8afdf | 11650 | PerlIO_printf(Perl_debug_log, |
ab3bbdeb YO |
11651 | "floating utf8 %s%s at %"IVdf"..%"UVuf" ", |
11652 | s, RE_SV_TAIL(r->float_utf8), | |
33b8afdf | 11653 | (IV)r->float_min_offset, (UV)r->float_max_offset); |
ab3bbdeb | 11654 | } |
33b8afdf | 11655 | if (r->check_substr || r->check_utf8) |
b81d288d | 11656 | PerlIO_printf(Perl_debug_log, |
10edeb5d JH |
11657 | (const char *) |
11658 | (r->check_substr == r->float_substr | |
11659 | && r->check_utf8 == r->float_utf8 | |
11660 | ? "(checking floating" : "(checking anchored")); | |
bbe252da | 11661 | if (r->extflags & RXf_NOSCAN) |
c277df42 | 11662 | PerlIO_printf(Perl_debug_log, " noscan"); |
bbe252da | 11663 | if (r->extflags & RXf_CHECK_ALL) |
c277df42 | 11664 | PerlIO_printf(Perl_debug_log, " isall"); |
33b8afdf | 11665 | if (r->check_substr || r->check_utf8) |
c277df42 IZ |
11666 | PerlIO_printf(Perl_debug_log, ") "); |
11667 | ||
f8fc2ecf YO |
11668 | if (ri->regstclass) { |
11669 | regprop(r, sv, ri->regstclass); | |
1de06328 | 11670 | PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv)); |
46fc3d4c | 11671 | } |
bbe252da | 11672 | if (r->extflags & RXf_ANCH) { |
774d564b | 11673 | PerlIO_printf(Perl_debug_log, "anchored"); |
bbe252da | 11674 | if (r->extflags & RXf_ANCH_BOL) |
774d564b | 11675 | PerlIO_printf(Perl_debug_log, "(BOL)"); |
bbe252da | 11676 | if (r->extflags & RXf_ANCH_MBOL) |
c277df42 | 11677 | PerlIO_printf(Perl_debug_log, "(MBOL)"); |
bbe252da | 11678 | if (r->extflags & RXf_ANCH_SBOL) |
cad2e5aa | 11679 | PerlIO_printf(Perl_debug_log, "(SBOL)"); |
bbe252da | 11680 | if (r->extflags & RXf_ANCH_GPOS) |
774d564b | 11681 | PerlIO_printf(Perl_debug_log, "(GPOS)"); |
11682 | PerlIO_putc(Perl_debug_log, ' '); | |
11683 | } | |
bbe252da | 11684 | if (r->extflags & RXf_GPOS_SEEN) |
70685ca0 | 11685 | PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs); |
bbe252da | 11686 | if (r->intflags & PREGf_SKIP) |
760ac839 | 11687 | PerlIO_printf(Perl_debug_log, "plus "); |
bbe252da | 11688 | if (r->intflags & PREGf_IMPLICIT) |
760ac839 | 11689 | PerlIO_printf(Perl_debug_log, "implicit "); |
70685ca0 | 11690 | PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen); |
bbe252da | 11691 | if (r->extflags & RXf_EVAL_SEEN) |
ce862d02 | 11692 | PerlIO_printf(Perl_debug_log, "with eval "); |
760ac839 | 11693 | PerlIO_printf(Perl_debug_log, "\n"); |
f7819f85 | 11694 | DEBUG_FLAGS_r(regdump_extflags("r->extflags: ",r->extflags)); |
65e66c80 | 11695 | #else |
7918f24d | 11696 | PERL_ARGS_ASSERT_REGDUMP; |
96a5add6 | 11697 | PERL_UNUSED_CONTEXT; |
65e66c80 | 11698 | PERL_UNUSED_ARG(r); |
17c3b450 | 11699 | #endif /* DEBUGGING */ |
a687059c LW |
11700 | } |
11701 | ||
11702 | /* | |
a0d0e21e LW |
11703 | - regprop - printable representation of opcode |
11704 | */ | |
3339dfd8 YO |
11705 | #define EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags) \ |
11706 | STMT_START { \ | |
11707 | if (do_sep) { \ | |
11708 | Perl_sv_catpvf(aTHX_ sv,"%s][%s",PL_colors[1],PL_colors[0]); \ | |
11709 | if (flags & ANYOF_INVERT) \ | |
11710 | /*make sure the invert info is in each */ \ | |
11711 | sv_catpvs(sv, "^"); \ | |
11712 | do_sep = 0; \ | |
11713 | } \ | |
11714 | } STMT_END | |
11715 | ||
46fc3d4c | 11716 | void |
32fc9b6a | 11717 | Perl_regprop(pTHX_ const regexp *prog, SV *sv, const regnode *o) |
a687059c | 11718 | { |
35ff7856 | 11719 | #ifdef DEBUGGING |
97aff369 | 11720 | dVAR; |
9b155405 | 11721 | register int k; |
f8fc2ecf | 11722 | RXi_GET_DECL(prog,progi); |
1de06328 | 11723 | GET_RE_DEBUG_FLAGS_DECL; |
f8fc2ecf | 11724 | |
7918f24d | 11725 | PERL_ARGS_ASSERT_REGPROP; |
a0d0e21e | 11726 | |
76f68e9b | 11727 | sv_setpvs(sv, ""); |
8aa23a47 | 11728 | |
03363afd | 11729 | if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */ |
830247a4 IZ |
11730 | /* It would be nice to FAIL() here, but this may be called from |
11731 | regexec.c, and it would be hard to supply pRExC_state. */ | |
a5ca303d | 11732 | Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX); |
13d6edb4 | 11733 | sv_catpv(sv, PL_reg_name[OP(o)]); /* Take off const! */ |
9b155405 | 11734 | |
3dab1dad | 11735 | k = PL_regkind[OP(o)]; |
9b155405 | 11736 | |
2a782b5b | 11737 | if (k == EXACT) { |
f92a2122 | 11738 | sv_catpvs(sv, " "); |
ab3bbdeb YO |
11739 | /* Using is_utf8_string() (via PERL_PV_UNI_DETECT) |
11740 | * is a crude hack but it may be the best for now since | |
11741 | * we have no flag "this EXACTish node was UTF-8" | |
11742 | * --jhi */ | |
f92a2122 NC |
11743 | pv_pretty(sv, STRING(o), STR_LEN(o), 60, PL_colors[0], PL_colors[1], |
11744 | PERL_PV_ESCAPE_UNI_DETECT | | |
c89df6cf | 11745 | PERL_PV_ESCAPE_NONASCII | |
f92a2122 NC |
11746 | PERL_PV_PRETTY_ELLIPSES | |
11747 | PERL_PV_PRETTY_LTGT | | |
11748 | PERL_PV_PRETTY_NOCLEAR | |
11749 | ); | |
bb263b4e | 11750 | } else if (k == TRIE) { |
3dab1dad | 11751 | /* print the details of the trie in dumpuntil instead, as |
f8fc2ecf | 11752 | * progi->data isn't available here */ |
1de06328 | 11753 | const char op = OP(o); |
647f639f | 11754 | const U32 n = ARG(o); |
1de06328 | 11755 | const reg_ac_data * const ac = IS_TRIE_AC(op) ? |
f8fc2ecf | 11756 | (reg_ac_data *)progi->data->data[n] : |
1de06328 | 11757 | NULL; |
3251b653 NC |
11758 | const reg_trie_data * const trie |
11759 | = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie]; | |
1de06328 | 11760 | |
13d6edb4 | 11761 | Perl_sv_catpvf(aTHX_ sv, "-%s",PL_reg_name[o->flags]); |
1de06328 YO |
11762 | DEBUG_TRIE_COMPILE_r( |
11763 | Perl_sv_catpvf(aTHX_ sv, | |
11764 | "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">", | |
11765 | (UV)trie->startstate, | |
1e2e3d02 | 11766 | (IV)trie->statecount-1, /* -1 because of the unused 0 element */ |
1de06328 YO |
11767 | (UV)trie->wordcount, |
11768 | (UV)trie->minlen, | |
11769 | (UV)trie->maxlen, | |
11770 | (UV)TRIE_CHARCOUNT(trie), | |
11771 | (UV)trie->uniquecharcount | |
11772 | ) | |
11773 | ); | |
11774 | if ( IS_ANYOF_TRIE(op) || trie->bitmap ) { | |
11775 | int i; | |
11776 | int rangestart = -1; | |
f46cb337 | 11777 | U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie); |
f3a2811a | 11778 | sv_catpvs(sv, "["); |
1de06328 YO |
11779 | for (i = 0; i <= 256; i++) { |
11780 | if (i < 256 && BITMAP_TEST(bitmap,i)) { | |
11781 | if (rangestart == -1) | |
11782 | rangestart = i; | |
11783 | } else if (rangestart != -1) { | |
11784 | if (i <= rangestart + 3) | |
11785 | for (; rangestart < i; rangestart++) | |
11786 | put_byte(sv, rangestart); | |
11787 | else { | |
11788 | put_byte(sv, rangestart); | |
11789 | sv_catpvs(sv, "-"); | |
11790 | put_byte(sv, i - 1); | |
11791 | } | |
11792 | rangestart = -1; | |
11793 | } | |
11794 | } | |
f3a2811a | 11795 | sv_catpvs(sv, "]"); |
1de06328 YO |
11796 | } |
11797 | ||
a3621e74 | 11798 | } else if (k == CURLY) { |
cb434fcc | 11799 | if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX) |
cea2e8a9 GS |
11800 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */ |
11801 | Perl_sv_catpvf(aTHX_ sv, " {%d,%d}", ARG1(o), ARG2(o)); | |
a0d0e21e | 11802 | } |
2c2d71f5 JH |
11803 | else if (k == WHILEM && o->flags) /* Ordinal/of */ |
11804 | Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4); | |
1f1031fe | 11805 | else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT) { |
894356b3 | 11806 | Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */ |
5daac39c | 11807 | if ( RXp_PAREN_NAMES(prog) ) { |
9d6ecd7a | 11808 | if ( k != REF || (OP(o) < NREF)) { |
502c6561 | 11809 | AV *list= MUTABLE_AV(progi->data->data[progi->name_list_idx]); |
ee9b8eae YO |
11810 | SV **name= av_fetch(list, ARG(o), 0 ); |
11811 | if (name) | |
11812 | Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name)); | |
11813 | } | |
11814 | else { | |
502c6561 | 11815 | AV *list= MUTABLE_AV(progi->data->data[ progi->name_list_idx ]); |
ad64d0ec | 11816 | SV *sv_dat= MUTABLE_SV(progi->data->data[ ARG( o ) ]); |
ee9b8eae YO |
11817 | I32 *nums=(I32*)SvPVX(sv_dat); |
11818 | SV **name= av_fetch(list, nums[0], 0 ); | |
11819 | I32 n; | |
11820 | if (name) { | |
11821 | for ( n=0; n<SvIVX(sv_dat); n++ ) { | |
11822 | Perl_sv_catpvf(aTHX_ sv, "%s%"IVdf, | |
11823 | (n ? "," : ""), (IV)nums[n]); | |
11824 | } | |
11825 | Perl_sv_catpvf(aTHX_ sv, " '%"SVf"'", SVfARG(*name)); | |
1f1031fe | 11826 | } |
1f1031fe | 11827 | } |
ee9b8eae | 11828 | } |
1f1031fe | 11829 | } else if (k == GOSUB) |
6bda09f9 | 11830 | Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */ |
e2e6a0f1 YO |
11831 | else if (k == VERB) { |
11832 | if (!o->flags) | |
11833 | Perl_sv_catpvf(aTHX_ sv, ":%"SVf, | |
ad64d0ec | 11834 | SVfARG((MUTABLE_SV(progi->data->data[ ARG( o ) ])))); |
e2e6a0f1 | 11835 | } else if (k == LOGICAL) |
04ebc1ab | 11836 | Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */ |
f9a79580 | 11837 | else if (k == FOLDCHAR) |
df44d732 | 11838 | Perl_sv_catpvf(aTHX_ sv, "[0x%"UVXf"]", PTR2UV(ARG(o)) ); |
653099ff GS |
11839 | else if (k == ANYOF) { |
11840 | int i, rangestart = -1; | |
2d03de9c | 11841 | const U8 flags = ANYOF_FLAGS(o); |
24d786f4 | 11842 | int do_sep = 0; |
0bd48802 AL |
11843 | |
11844 | /* Should be synchronized with * ANYOF_ #xdefines in regcomp.h */ | |
11845 | static const char * const anyofs[] = { | |
653099ff GS |
11846 | "\\w", |
11847 | "\\W", | |
11848 | "\\s", | |
11849 | "\\S", | |
11850 | "\\d", | |
11851 | "\\D", | |
11852 | "[:alnum:]", | |
11853 | "[:^alnum:]", | |
11854 | "[:alpha:]", | |
11855 | "[:^alpha:]", | |
11856 | "[:ascii:]", | |
11857 | "[:^ascii:]", | |
24d786f4 YO |
11858 | "[:cntrl:]", |
11859 | "[:^cntrl:]", | |
653099ff GS |
11860 | "[:graph:]", |
11861 | "[:^graph:]", | |
11862 | "[:lower:]", | |
11863 | "[:^lower:]", | |
11864 | "[:print:]", | |
11865 | "[:^print:]", | |
11866 | "[:punct:]", | |
11867 | "[:^punct:]", | |
11868 | "[:upper:]", | |
aaa51d5e | 11869 | "[:^upper:]", |
653099ff | 11870 | "[:xdigit:]", |
aaa51d5e JF |
11871 | "[:^xdigit:]", |
11872 | "[:space:]", | |
11873 | "[:^space:]", | |
11874 | "[:blank:]", | |
11875 | "[:^blank:]" | |
653099ff GS |
11876 | }; |
11877 | ||
19860706 | 11878 | if (flags & ANYOF_LOCALE) |
396482e1 | 11879 | sv_catpvs(sv, "{loc}"); |
39065660 | 11880 | if (flags & ANYOF_LOC_NONBITMAP_FOLD) |
396482e1 | 11881 | sv_catpvs(sv, "{i}"); |
653099ff | 11882 | Perl_sv_catpvf(aTHX_ sv, "[%s", PL_colors[0]); |
19860706 | 11883 | if (flags & ANYOF_INVERT) |
396482e1 | 11884 | sv_catpvs(sv, "^"); |
686b73d4 | 11885 | |
3339dfd8 | 11886 | /* output what the standard cp 0-255 bitmap matches */ |
ffc61ed2 JH |
11887 | for (i = 0; i <= 256; i++) { |
11888 | if (i < 256 && ANYOF_BITMAP_TEST(o,i)) { | |
11889 | if (rangestart == -1) | |
11890 | rangestart = i; | |
11891 | } else if (rangestart != -1) { | |
11892 | if (i <= rangestart + 3) | |
11893 | for (; rangestart < i; rangestart++) | |
653099ff | 11894 | put_byte(sv, rangestart); |
ffc61ed2 JH |
11895 | else { |
11896 | put_byte(sv, rangestart); | |
396482e1 | 11897 | sv_catpvs(sv, "-"); |
ffc61ed2 | 11898 | put_byte(sv, i - 1); |
653099ff | 11899 | } |
24d786f4 | 11900 | do_sep = 1; |
ffc61ed2 | 11901 | rangestart = -1; |
653099ff | 11902 | } |
847a199f | 11903 | } |
3339dfd8 YO |
11904 | |
11905 | EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags); | |
3a15e693 KW |
11906 | /* output any special charclass tests (used entirely under use locale) */ |
11907 | if (ANYOF_CLASS_TEST_ANY_SET(o)) | |
bb7a0f54 | 11908 | for (i = 0; i < (int)(sizeof(anyofs)/sizeof(char*)); i++) |
24d786f4 | 11909 | if (ANYOF_CLASS_TEST(o,i)) { |
ffc61ed2 | 11910 | sv_catpv(sv, anyofs[i]); |
24d786f4 YO |
11911 | do_sep = 1; |
11912 | } | |
11913 | ||
3339dfd8 YO |
11914 | EMIT_ANYOF_TEST_SEPARATOR(do_sep,sv,flags); |
11915 | ||
11454c59 KW |
11916 | if (flags & ANYOF_NON_UTF8_LATIN1_ALL) { |
11917 | sv_catpvs(sv, "{non-utf8-latin1-all}"); | |
11918 | } | |
11919 | ||
3339dfd8 | 11920 | /* output information about the unicode matching */ |
ef87b810 | 11921 | if (flags & ANYOF_UNICODE_ALL) |
396482e1 | 11922 | sv_catpvs(sv, "{unicode_all}"); |
137165a6 | 11923 | else if (ANYOF_NONBITMAP(o)) |
ef87b810 | 11924 | sv_catpvs(sv, "{unicode}"); |
f5ecd18d | 11925 | if (flags & ANYOF_NONBITMAP_NON_UTF8) |
ef87b810 | 11926 | sv_catpvs(sv, "{outside bitmap}"); |
ffc61ed2 | 11927 | |
1aa9930e | 11928 | if (ANYOF_NONBITMAP(o)) { |
dbe7a391 | 11929 | SV *lv; /* Set if there is something outside the bit map */ |
32fc9b6a | 11930 | SV * const sw = regclass_swash(prog, o, FALSE, &lv, 0); |
f1114c30 KW |
11931 | bool byte_output = FALSE; /* If something in the bitmap has been |
11932 | output */ | |
686b73d4 | 11933 | |
c16787fd | 11934 | if (lv && lv != &PL_sv_undef) { |
ffc61ed2 | 11935 | if (sw) { |
89ebb4a3 | 11936 | U8 s[UTF8_MAXBYTES_CASE+1]; |
24d786f4 | 11937 | |
dbe7a391 | 11938 | for (i = 0; i <= 256; i++) { /* Look at chars in bitmap */ |
1df70142 | 11939 | uvchr_to_utf8(s, i); |
686b73d4 | 11940 | |
dcf8909a KW |
11941 | if (i < 256 |
11942 | && ! ANYOF_BITMAP_TEST(o, i) /* Don't duplicate | |
11943 | things already | |
11944 | output as part | |
11945 | of the bitmap */ | |
11946 | && swash_fetch(sw, s, TRUE)) | |
11947 | { | |
ffc61ed2 JH |
11948 | if (rangestart == -1) |
11949 | rangestart = i; | |
11950 | } else if (rangestart != -1) { | |
f1114c30 | 11951 | byte_output = TRUE; |
ffc61ed2 JH |
11952 | if (i <= rangestart + 3) |
11953 | for (; rangestart < i; rangestart++) { | |
7128c099 | 11954 | put_byte(sv, rangestart); |
ffc61ed2 JH |
11955 | } |
11956 | else { | |
7128c099 | 11957 | put_byte(sv, rangestart); |
396482e1 | 11958 | sv_catpvs(sv, "-"); |
7128c099 | 11959 | put_byte(sv, i-1); |
ffc61ed2 | 11960 | } |
e87973a9 | 11961 | rangestart = -1; |
19860706 | 11962 | } |
e87973a9 | 11963 | } |
19860706 | 11964 | } |
fde631ed | 11965 | |
ffc61ed2 | 11966 | { |
2e0de35c | 11967 | char *s = savesvpv(lv); |
c445ea15 | 11968 | char * const origs = s; |
686b73d4 | 11969 | |
3dab1dad YO |
11970 | while (*s && *s != '\n') |
11971 | s++; | |
686b73d4 | 11972 | |
ffc61ed2 | 11973 | if (*s == '\n') { |
2d03de9c | 11974 | const char * const t = ++s; |
686b73d4 | 11975 | |
f1114c30 KW |
11976 | if (byte_output) { |
11977 | sv_catpvs(sv, " "); | |
11978 | } | |
11979 | ||
ffc61ed2 JH |
11980 | while (*s) { |
11981 | if (*s == '\n') | |
11982 | *s = ' '; | |
1a9c8476 KW |
11983 | } |
11984 | else if (*s == '\t') { | |
11985 | *s = '-'; | |
11986 | } | |
ffc61ed2 JH |
11987 | s++; |
11988 | } | |
11989 | if (s[-1] == ' ') | |
11990 | s[-1] = 0; | |
686b73d4 | 11991 | |
ffc61ed2 | 11992 | sv_catpv(sv, t); |
fde631ed | 11993 | } |
686b73d4 | 11994 | |
ffc61ed2 | 11995 | Safefree(origs); |
fde631ed | 11996 | } |
c16787fd | 11997 | SvREFCNT_dec(lv); |
fde631ed | 11998 | } |
653099ff | 11999 | } |
ffc61ed2 | 12000 | |
653099ff GS |
12001 | Perl_sv_catpvf(aTHX_ sv, "%s]", PL_colors[1]); |
12002 | } | |
9b155405 | 12003 | else if (k == BRANCHJ && (OP(o) == UNLESSM || OP(o) == IFMATCH)) |
07be1b83 | 12004 | Perl_sv_catpvf(aTHX_ sv, "[%d]", -(o->flags)); |
65e66c80 | 12005 | #else |
96a5add6 | 12006 | PERL_UNUSED_CONTEXT; |
65e66c80 SP |
12007 | PERL_UNUSED_ARG(sv); |
12008 | PERL_UNUSED_ARG(o); | |
f9049ba1 | 12009 | PERL_UNUSED_ARG(prog); |
17c3b450 | 12010 | #endif /* DEBUGGING */ |
35ff7856 | 12011 | } |
a687059c | 12012 | |
cad2e5aa | 12013 | SV * |
288b8c02 | 12014 | Perl_re_intuit_string(pTHX_ REGEXP * const r) |
cad2e5aa | 12015 | { /* Assume that RE_INTUIT is set */ |
97aff369 | 12016 | dVAR; |
288b8c02 | 12017 | struct regexp *const prog = (struct regexp *)SvANY(r); |
a3621e74 | 12018 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
12019 | |
12020 | PERL_ARGS_ASSERT_RE_INTUIT_STRING; | |
96a5add6 AL |
12021 | PERL_UNUSED_CONTEXT; |
12022 | ||
a3621e74 | 12023 | DEBUG_COMPILE_r( |
cfd0369c | 12024 | { |
2d03de9c | 12025 | const char * const s = SvPV_nolen_const(prog->check_substr |
cfd0369c | 12026 | ? prog->check_substr : prog->check_utf8); |
cad2e5aa JH |
12027 | |
12028 | if (!PL_colorset) reginitcolors(); | |
12029 | PerlIO_printf(Perl_debug_log, | |
a0288114 | 12030 | "%sUsing REx %ssubstr:%s \"%s%.60s%s%s\"\n", |
33b8afdf JH |
12031 | PL_colors[4], |
12032 | prog->check_substr ? "" : "utf8 ", | |
12033 | PL_colors[5],PL_colors[0], | |
cad2e5aa JH |
12034 | s, |
12035 | PL_colors[1], | |
12036 | (strlen(s) > 60 ? "..." : "")); | |
12037 | } ); | |
12038 | ||
33b8afdf | 12039 | return prog->check_substr ? prog->check_substr : prog->check_utf8; |
cad2e5aa JH |
12040 | } |
12041 | ||
84da74a7 | 12042 | /* |
f8149455 | 12043 | pregfree() |
84da74a7 | 12044 | |
f8149455 YO |
12045 | handles refcounting and freeing the perl core regexp structure. When |
12046 | it is necessary to actually free the structure the first thing it | |
3b753521 | 12047 | does is call the 'free' method of the regexp_engine associated to |
f8149455 YO |
12048 | the regexp, allowing the handling of the void *pprivate; member |
12049 | first. (This routine is not overridable by extensions, which is why | |
12050 | the extensions free is called first.) | |
12051 | ||
12052 | See regdupe and regdupe_internal if you change anything here. | |
84da74a7 | 12053 | */ |
f8149455 | 12054 | #ifndef PERL_IN_XSUB_RE |
2b69d0c2 | 12055 | void |
84679df5 | 12056 | Perl_pregfree(pTHX_ REGEXP *r) |
a687059c | 12057 | { |
288b8c02 NC |
12058 | SvREFCNT_dec(r); |
12059 | } | |
12060 | ||
12061 | void | |
12062 | Perl_pregfree2(pTHX_ REGEXP *rx) | |
12063 | { | |
27da23d5 | 12064 | dVAR; |
288b8c02 | 12065 | struct regexp *const r = (struct regexp *)SvANY(rx); |
fc32ee4a | 12066 | GET_RE_DEBUG_FLAGS_DECL; |
a3621e74 | 12067 | |
7918f24d NC |
12068 | PERL_ARGS_ASSERT_PREGFREE2; |
12069 | ||
28d8d7f4 YO |
12070 | if (r->mother_re) { |
12071 | ReREFCNT_dec(r->mother_re); | |
12072 | } else { | |
288b8c02 | 12073 | CALLREGFREE_PVT(rx); /* free the private data */ |
ef8d46e8 | 12074 | SvREFCNT_dec(RXp_PAREN_NAMES(r)); |
28d8d7f4 YO |
12075 | } |
12076 | if (r->substrs) { | |
ef8d46e8 VP |
12077 | SvREFCNT_dec(r->anchored_substr); |
12078 | SvREFCNT_dec(r->anchored_utf8); | |
12079 | SvREFCNT_dec(r->float_substr); | |
12080 | SvREFCNT_dec(r->float_utf8); | |
28d8d7f4 YO |
12081 | Safefree(r->substrs); |
12082 | } | |
288b8c02 | 12083 | RX_MATCH_COPY_FREE(rx); |
f8c7b90f | 12084 | #ifdef PERL_OLD_COPY_ON_WRITE |
ef8d46e8 | 12085 | SvREFCNT_dec(r->saved_copy); |
ed252734 | 12086 | #endif |
f0ab9afb | 12087 | Safefree(r->offs); |
f8149455 | 12088 | } |
28d8d7f4 YO |
12089 | |
12090 | /* reg_temp_copy() | |
12091 | ||
12092 | This is a hacky workaround to the structural issue of match results | |
12093 | being stored in the regexp structure which is in turn stored in | |
12094 | PL_curpm/PL_reg_curpm. The problem is that due to qr// the pattern | |
12095 | could be PL_curpm in multiple contexts, and could require multiple | |
12096 | result sets being associated with the pattern simultaneously, such | |
12097 | as when doing a recursive match with (??{$qr}) | |
12098 | ||
12099 | The solution is to make a lightweight copy of the regexp structure | |
12100 | when a qr// is returned from the code executed by (??{$qr}) this | |
486ec47a | 12101 | lightweight copy doesn't actually own any of its data except for |
28d8d7f4 YO |
12102 | the starp/end and the actual regexp structure itself. |
12103 | ||
12104 | */ | |
12105 | ||
12106 | ||
84679df5 | 12107 | REGEXP * |
f0826785 | 12108 | Perl_reg_temp_copy (pTHX_ REGEXP *ret_x, REGEXP *rx) |
7918f24d | 12109 | { |
f0826785 | 12110 | struct regexp *ret; |
288b8c02 | 12111 | struct regexp *const r = (struct regexp *)SvANY(rx); |
28d8d7f4 | 12112 | register const I32 npar = r->nparens+1; |
7918f24d NC |
12113 | |
12114 | PERL_ARGS_ASSERT_REG_TEMP_COPY; | |
12115 | ||
f0826785 BM |
12116 | if (!ret_x) |
12117 | ret_x = (REGEXP*) newSV_type(SVt_REGEXP); | |
12118 | ret = (struct regexp *)SvANY(ret_x); | |
12119 | ||
288b8c02 | 12120 | (void)ReREFCNT_inc(rx); |
f7c278bf NC |
12121 | /* We can take advantage of the existing "copied buffer" mechanism in SVs |
12122 | by pointing directly at the buffer, but flagging that the allocated | |
12123 | space in the copy is zero. As we've just done a struct copy, it's now | |
12124 | a case of zero-ing that, rather than copying the current length. */ | |
12125 | SvPV_set(ret_x, RX_WRAPPED(rx)); | |
8f6ae13c | 12126 | SvFLAGS(ret_x) |= SvFLAGS(rx) & (SVf_POK|SVp_POK|SVf_UTF8); |
b6f60916 NC |
12127 | memcpy(&(ret->xpv_cur), &(r->xpv_cur), |
12128 | sizeof(regexp) - STRUCT_OFFSET(regexp, xpv_cur)); | |
f7c278bf | 12129 | SvLEN_set(ret_x, 0); |
b9ad13ac | 12130 | SvSTASH_set(ret_x, NULL); |
703c388d | 12131 | SvMAGIC_set(ret_x, NULL); |
f0ab9afb NC |
12132 | Newx(ret->offs, npar, regexp_paren_pair); |
12133 | Copy(r->offs, ret->offs, npar, regexp_paren_pair); | |
28d8d7f4 | 12134 | if (r->substrs) { |
28d8d7f4 | 12135 | Newx(ret->substrs, 1, struct reg_substr_data); |
6ab65676 NC |
12136 | StructCopy(r->substrs, ret->substrs, struct reg_substr_data); |
12137 | ||
12138 | SvREFCNT_inc_void(ret->anchored_substr); | |
12139 | SvREFCNT_inc_void(ret->anchored_utf8); | |
12140 | SvREFCNT_inc_void(ret->float_substr); | |
12141 | SvREFCNT_inc_void(ret->float_utf8); | |
12142 | ||
12143 | /* check_substr and check_utf8, if non-NULL, point to either their | |
12144 | anchored or float namesakes, and don't hold a second reference. */ | |
486913e4 | 12145 | } |
288b8c02 | 12146 | RX_MATCH_COPIED_off(ret_x); |
28d8d7f4 | 12147 | #ifdef PERL_OLD_COPY_ON_WRITE |
b89b0c6f | 12148 | ret->saved_copy = NULL; |
28d8d7f4 | 12149 | #endif |
288b8c02 | 12150 | ret->mother_re = rx; |
28d8d7f4 | 12151 | |
288b8c02 | 12152 | return ret_x; |
28d8d7f4 | 12153 | } |
f8149455 YO |
12154 | #endif |
12155 | ||
12156 | /* regfree_internal() | |
12157 | ||
12158 | Free the private data in a regexp. This is overloadable by | |
12159 | extensions. Perl takes care of the regexp structure in pregfree(), | |
3b753521 | 12160 | this covers the *pprivate pointer which technically perl doesn't |
f8149455 YO |
12161 | know about, however of course we have to handle the |
12162 | regexp_internal structure when no extension is in use. | |
12163 | ||
12164 | Note this is called before freeing anything in the regexp | |
12165 | structure. | |
12166 | */ | |
12167 | ||
12168 | void | |
288b8c02 | 12169 | Perl_regfree_internal(pTHX_ REGEXP * const rx) |
f8149455 YO |
12170 | { |
12171 | dVAR; | |
288b8c02 | 12172 | struct regexp *const r = (struct regexp *)SvANY(rx); |
f8149455 YO |
12173 | RXi_GET_DECL(r,ri); |
12174 | GET_RE_DEBUG_FLAGS_DECL; | |
7918f24d NC |
12175 | |
12176 | PERL_ARGS_ASSERT_REGFREE_INTERNAL; | |
12177 | ||
f8149455 YO |
12178 | DEBUG_COMPILE_r({ |
12179 | if (!PL_colorset) | |
12180 | reginitcolors(); | |
12181 | { | |
12182 | SV *dsv= sv_newmortal(); | |
3c8556c3 | 12183 | RE_PV_QUOTED_DECL(s, RX_UTF8(rx), |
5509d87a | 12184 | dsv, RX_PRECOMP(rx), RX_PRELEN(rx), 60); |
f8149455 YO |
12185 | PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n", |
12186 | PL_colors[4],PL_colors[5],s); | |
12187 | } | |
12188 | }); | |
7122b237 YO |
12189 | #ifdef RE_TRACK_PATTERN_OFFSETS |
12190 | if (ri->u.offsets) | |
12191 | Safefree(ri->u.offsets); /* 20010421 MJD */ | |
12192 | #endif | |
f8fc2ecf YO |
12193 | if (ri->data) { |
12194 | int n = ri->data->count; | |
f3548bdc DM |
12195 | PAD* new_comppad = NULL; |
12196 | PAD* old_comppad; | |
4026c95a | 12197 | PADOFFSET refcnt; |
dfad63ad | 12198 | |
c277df42 | 12199 | while (--n >= 0) { |
261faec3 | 12200 | /* If you add a ->what type here, update the comment in regcomp.h */ |
f8fc2ecf | 12201 | switch (ri->data->what[n]) { |
af534a04 | 12202 | case 'a': |
c277df42 | 12203 | case 's': |
81714fb9 | 12204 | case 'S': |
55eed653 | 12205 | case 'u': |
ad64d0ec | 12206 | SvREFCNT_dec(MUTABLE_SV(ri->data->data[n])); |
c277df42 | 12207 | break; |
653099ff | 12208 | case 'f': |
f8fc2ecf | 12209 | Safefree(ri->data->data[n]); |
653099ff | 12210 | break; |
dfad63ad | 12211 | case 'p': |
502c6561 | 12212 | new_comppad = MUTABLE_AV(ri->data->data[n]); |
dfad63ad | 12213 | break; |
c277df42 | 12214 | case 'o': |
dfad63ad | 12215 | if (new_comppad == NULL) |
cea2e8a9 | 12216 | Perl_croak(aTHX_ "panic: pregfree comppad"); |
f3548bdc DM |
12217 | PAD_SAVE_LOCAL(old_comppad, |
12218 | /* Watch out for global destruction's random ordering. */ | |
c445ea15 | 12219 | (SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL |
f3548bdc | 12220 | ); |
b34c0dd4 | 12221 | OP_REFCNT_LOCK; |
f8fc2ecf | 12222 | refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]); |
4026c95a SH |
12223 | OP_REFCNT_UNLOCK; |
12224 | if (!refcnt) | |
f8fc2ecf | 12225 | op_free((OP_4tree*)ri->data->data[n]); |
9b978d73 | 12226 | |
f3548bdc | 12227 | PAD_RESTORE_LOCAL(old_comppad); |
ad64d0ec | 12228 | SvREFCNT_dec(MUTABLE_SV(new_comppad)); |
dfad63ad | 12229 | new_comppad = NULL; |
c277df42 IZ |
12230 | break; |
12231 | case 'n': | |
9e55ce06 | 12232 | break; |
07be1b83 | 12233 | case 'T': |
be8e71aa YO |
12234 | { /* Aho Corasick add-on structure for a trie node. |
12235 | Used in stclass optimization only */ | |
07be1b83 | 12236 | U32 refcount; |
f8fc2ecf | 12237 | reg_ac_data *aho=(reg_ac_data*)ri->data->data[n]; |
07be1b83 YO |
12238 | OP_REFCNT_LOCK; |
12239 | refcount = --aho->refcount; | |
12240 | OP_REFCNT_UNLOCK; | |
12241 | if ( !refcount ) { | |
446bd890 NC |
12242 | PerlMemShared_free(aho->states); |
12243 | PerlMemShared_free(aho->fail); | |
446bd890 NC |
12244 | /* do this last!!!! */ |
12245 | PerlMemShared_free(ri->data->data[n]); | |
12246 | PerlMemShared_free(ri->regstclass); | |
07be1b83 YO |
12247 | } |
12248 | } | |
12249 | break; | |
a3621e74 | 12250 | case 't': |
07be1b83 | 12251 | { |
be8e71aa | 12252 | /* trie structure. */ |
07be1b83 | 12253 | U32 refcount; |
f8fc2ecf | 12254 | reg_trie_data *trie=(reg_trie_data*)ri->data->data[n]; |
07be1b83 YO |
12255 | OP_REFCNT_LOCK; |
12256 | refcount = --trie->refcount; | |
12257 | OP_REFCNT_UNLOCK; | |
12258 | if ( !refcount ) { | |
446bd890 | 12259 | PerlMemShared_free(trie->charmap); |
446bd890 NC |
12260 | PerlMemShared_free(trie->states); |
12261 | PerlMemShared_free(trie->trans); | |
07be1b83 | 12262 | if (trie->bitmap) |
446bd890 | 12263 | PerlMemShared_free(trie->bitmap); |
786e8c11 | 12264 | if (trie->jump) |
446bd890 | 12265 | PerlMemShared_free(trie->jump); |
2e64971a | 12266 | PerlMemShared_free(trie->wordinfo); |
446bd890 NC |
12267 | /* do this last!!!! */ |
12268 | PerlMemShared_free(ri->data->data[n]); | |
a3621e74 | 12269 | } |
07be1b83 YO |
12270 | } |
12271 | break; | |
c277df42 | 12272 | default: |
f8fc2ecf | 12273 | Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]); |
c277df42 IZ |
12274 | } |
12275 | } | |
f8fc2ecf YO |
12276 | Safefree(ri->data->what); |
12277 | Safefree(ri->data); | |
a0d0e21e | 12278 | } |
28d8d7f4 | 12279 | |
f8fc2ecf | 12280 | Safefree(ri); |
a687059c | 12281 | } |
c277df42 | 12282 | |
a09252eb NC |
12283 | #define av_dup_inc(s,t) MUTABLE_AV(sv_dup_inc((const SV *)s,t)) |
12284 | #define hv_dup_inc(s,t) MUTABLE_HV(sv_dup_inc((const SV *)s,t)) | |
84da74a7 YO |
12285 | #define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL) |
12286 | ||
12287 | /* | |
32cd70f6 | 12288 | re_dup - duplicate a regexp. |
84da74a7 | 12289 | |
8233f606 DM |
12290 | This routine is expected to clone a given regexp structure. It is only |
12291 | compiled under USE_ITHREADS. | |
32cd70f6 | 12292 | |
f8149455 YO |
12293 | After all of the core data stored in struct regexp is duplicated |
12294 | the regexp_engine.dupe method is used to copy any private data | |
12295 | stored in the *pprivate pointer. This allows extensions to handle | |
12296 | any duplication it needs to do. | |
12297 | ||
12298 | See pregfree() and regfree_internal() if you change anything here. | |
84da74a7 | 12299 | */ |
a3c0e9ca | 12300 | #if defined(USE_ITHREADS) |
f8149455 | 12301 | #ifndef PERL_IN_XSUB_RE |
288b8c02 NC |
12302 | void |
12303 | Perl_re_dup_guts(pTHX_ const REGEXP *sstr, REGEXP *dstr, CLONE_PARAMS *param) | |
84da74a7 | 12304 | { |
84da74a7 | 12305 | dVAR; |
a86a1ca7 | 12306 | I32 npar; |
288b8c02 NC |
12307 | const struct regexp *r = (const struct regexp *)SvANY(sstr); |
12308 | struct regexp *ret = (struct regexp *)SvANY(dstr); | |
f8149455 | 12309 | |
7918f24d NC |
12310 | PERL_ARGS_ASSERT_RE_DUP_GUTS; |
12311 | ||
84da74a7 | 12312 | npar = r->nparens+1; |
f0ab9afb NC |
12313 | Newx(ret->offs, npar, regexp_paren_pair); |
12314 | Copy(r->offs, ret->offs, npar, regexp_paren_pair); | |
6057429f | 12315 | if(ret->swap) { |
28d8d7f4 | 12316 | /* no need to copy these */ |
f0ab9afb | 12317 | Newx(ret->swap, npar, regexp_paren_pair); |
28d8d7f4 | 12318 | } |
84da74a7 | 12319 | |
6057429f | 12320 | if (ret->substrs) { |
32cd70f6 NC |
12321 | /* Do it this way to avoid reading from *r after the StructCopy(). |
12322 | That way, if any of the sv_dup_inc()s dislodge *r from the L1 | |
12323 | cache, it doesn't matter. */ | |
66b1de87 NC |
12324 | const bool anchored = r->check_substr |
12325 | ? r->check_substr == r->anchored_substr | |
12326 | : r->check_utf8 == r->anchored_utf8; | |
785a26d5 | 12327 | Newx(ret->substrs, 1, struct reg_substr_data); |
a86a1ca7 NC |
12328 | StructCopy(r->substrs, ret->substrs, struct reg_substr_data); |
12329 | ||
32cd70f6 NC |
12330 | ret->anchored_substr = sv_dup_inc(ret->anchored_substr, param); |
12331 | ret->anchored_utf8 = sv_dup_inc(ret->anchored_utf8, param); | |
12332 | ret->float_substr = sv_dup_inc(ret->float_substr, param); | |
12333 | ret->float_utf8 = sv_dup_inc(ret->float_utf8, param); | |
a86a1ca7 | 12334 | |
32cd70f6 NC |
12335 | /* check_substr and check_utf8, if non-NULL, point to either their |
12336 | anchored or float namesakes, and don't hold a second reference. */ | |
12337 | ||
12338 | if (ret->check_substr) { | |
12339 | if (anchored) { | |
12340 | assert(r->check_utf8 == r->anchored_utf8); | |
12341 | ret->check_substr = ret->anchored_substr; | |
12342 | ret->check_utf8 = ret->anchored_utf8; | |
12343 | } else { | |
12344 | assert(r->check_substr == r->float_substr); | |
12345 | assert(r->check_utf8 == r->float_utf8); | |
12346 | ret->check_substr = ret->float_substr; | |
12347 | ret->check_utf8 = ret->float_utf8; | |
12348 | } | |
66b1de87 NC |
12349 | } else if (ret->check_utf8) { |
12350 | if (anchored) { | |
12351 | ret->check_utf8 = ret->anchored_utf8; | |
12352 | } else { | |
12353 | ret->check_utf8 = ret->float_utf8; | |
12354 | } | |
32cd70f6 | 12355 | } |
6057429f | 12356 | } |
f8149455 | 12357 | |
5daac39c | 12358 | RXp_PAREN_NAMES(ret) = hv_dup_inc(RXp_PAREN_NAMES(ret), param); |
bcdf7404 | 12359 | |
6057429f | 12360 | if (ret->pprivate) |
288b8c02 | 12361 | RXi_SET(ret,CALLREGDUPE_PVT(dstr,param)); |
f8149455 | 12362 | |
288b8c02 | 12363 | if (RX_MATCH_COPIED(dstr)) |
6057429f | 12364 | ret->subbeg = SAVEPVN(ret->subbeg, ret->sublen); |
f8149455 YO |
12365 | else |
12366 | ret->subbeg = NULL; | |
12367 | #ifdef PERL_OLD_COPY_ON_WRITE | |
12368 | ret->saved_copy = NULL; | |
12369 | #endif | |
6057429f | 12370 | |
c2123ae3 NC |
12371 | if (ret->mother_re) { |
12372 | if (SvPVX_const(dstr) == SvPVX_const(ret->mother_re)) { | |
12373 | /* Our storage points directly to our mother regexp, but that's | |
12374 | 1: a buffer in a different thread | |
12375 | 2: something we no longer hold a reference on | |
12376 | so we need to copy it locally. */ | |
d5aafdca FC |
12377 | /* Note we need to use SvCUR(), rather than |
12378 | SvLEN(), on our mother_re, because it, in | |
c2123ae3 NC |
12379 | turn, may well be pointing to its own mother_re. */ |
12380 | SvPV_set(dstr, SAVEPVN(SvPVX_const(ret->mother_re), | |
12381 | SvCUR(ret->mother_re)+1)); | |
12382 | SvLEN_set(dstr, SvCUR(ret->mother_re)+1); | |
12383 | } | |
12384 | ret->mother_re = NULL; | |
12385 | } | |
6057429f | 12386 | ret->gofs = 0; |
f8149455 YO |
12387 | } |
12388 | #endif /* PERL_IN_XSUB_RE */ | |
12389 | ||
12390 | /* | |
12391 | regdupe_internal() | |
12392 | ||
12393 | This is the internal complement to regdupe() which is used to copy | |
12394 | the structure pointed to by the *pprivate pointer in the regexp. | |
12395 | This is the core version of the extension overridable cloning hook. | |
12396 | The regexp structure being duplicated will be copied by perl prior | |
12397 | to this and will be provided as the regexp *r argument, however | |
12398 | with the /old/ structures pprivate pointer value. Thus this routine | |
12399 | may override any copying normally done by perl. | |
12400 | ||
12401 | It returns a pointer to the new regexp_internal structure. | |
12402 | */ | |
12403 | ||
12404 | void * | |
288b8c02 | 12405 | Perl_regdupe_internal(pTHX_ REGEXP * const rx, CLONE_PARAMS *param) |
f8149455 YO |
12406 | { |
12407 | dVAR; | |
288b8c02 | 12408 | struct regexp *const r = (struct regexp *)SvANY(rx); |
f8149455 | 12409 | regexp_internal *reti; |
0780bc72 | 12410 | int len; |
f8149455 | 12411 | RXi_GET_DECL(r,ri); |
7918f24d NC |
12412 | |
12413 | PERL_ARGS_ASSERT_REGDUPE_INTERNAL; | |
f8149455 | 12414 | |
7122b237 | 12415 | len = ProgLen(ri); |
f8149455 | 12416 | |
45cf4570 | 12417 | Newxc(reti, sizeof(regexp_internal) + len*sizeof(regnode), char, regexp_internal); |
f8149455 YO |
12418 | Copy(ri->program, reti->program, len+1, regnode); |
12419 | ||
f8149455 | 12420 | |
f8fc2ecf | 12421 | reti->regstclass = NULL; |
bcdf7404 | 12422 | |
f8fc2ecf | 12423 | if (ri->data) { |
84da74a7 | 12424 | struct reg_data *d; |
f8fc2ecf | 12425 | const int count = ri->data->count; |
84da74a7 YO |
12426 | int i; |
12427 | ||
12428 | Newxc(d, sizeof(struct reg_data) + count*sizeof(void *), | |
12429 | char, struct reg_data); | |
12430 | Newx(d->what, count, U8); | |
12431 | ||
12432 | d->count = count; | |
12433 | for (i = 0; i < count; i++) { | |
f8fc2ecf | 12434 | d->what[i] = ri->data->what[i]; |
84da74a7 | 12435 | switch (d->what[i]) { |
af534a04 | 12436 | /* legal options are one of: sSfpontTua |
84da74a7 | 12437 | see also regcomp.h and pregfree() */ |
af534a04 | 12438 | case 'a': /* actually an AV, but the dup function is identical. */ |
84da74a7 | 12439 | case 's': |
81714fb9 | 12440 | case 'S': |
0536c0a7 | 12441 | case 'p': /* actually an AV, but the dup function is identical. */ |
55eed653 | 12442 | case 'u': /* actually an HV, but the dup function is identical. */ |
ad64d0ec | 12443 | d->data[i] = sv_dup_inc((const SV *)ri->data->data[i], param); |
84da74a7 | 12444 | break; |
84da74a7 YO |
12445 | case 'f': |
12446 | /* This is cheating. */ | |
12447 | Newx(d->data[i], 1, struct regnode_charclass_class); | |
f8fc2ecf | 12448 | StructCopy(ri->data->data[i], d->data[i], |
84da74a7 | 12449 | struct regnode_charclass_class); |
f8fc2ecf | 12450 | reti->regstclass = (regnode*)d->data[i]; |
84da74a7 YO |
12451 | break; |
12452 | case 'o': | |
bbe252da YO |
12453 | /* Compiled op trees are readonly and in shared memory, |
12454 | and can thus be shared without duplication. */ | |
84da74a7 | 12455 | OP_REFCNT_LOCK; |
f8fc2ecf | 12456 | d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]); |
84da74a7 YO |
12457 | OP_REFCNT_UNLOCK; |
12458 | break; | |
23eab42c NC |
12459 | case 'T': |
12460 | /* Trie stclasses are readonly and can thus be shared | |
12461 | * without duplication. We free the stclass in pregfree | |
12462 | * when the corresponding reg_ac_data struct is freed. | |
12463 | */ | |
12464 | reti->regstclass= ri->regstclass; | |
12465 | /* Fall through */ | |
84da74a7 | 12466 | case 't': |
84da74a7 | 12467 | OP_REFCNT_LOCK; |
0536c0a7 | 12468 | ((reg_trie_data*)ri->data->data[i])->refcount++; |
84da74a7 | 12469 | OP_REFCNT_UNLOCK; |
0536c0a7 NC |
12470 | /* Fall through */ |
12471 | case 'n': | |
12472 | d->data[i] = ri->data->data[i]; | |
84da74a7 | 12473 | break; |
84da74a7 | 12474 | default: |
f8fc2ecf | 12475 | Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]); |
84da74a7 YO |
12476 | } |
12477 | } | |
12478 | ||
f8fc2ecf | 12479 | reti->data = d; |
84da74a7 YO |
12480 | } |
12481 | else | |
f8fc2ecf | 12482 | reti->data = NULL; |
84da74a7 | 12483 | |
cde0cee5 YO |
12484 | reti->name_list_idx = ri->name_list_idx; |
12485 | ||
7122b237 YO |
12486 | #ifdef RE_TRACK_PATTERN_OFFSETS |
12487 | if (ri->u.offsets) { | |
12488 | Newx(reti->u.offsets, 2*len+1, U32); | |
12489 | Copy(ri->u.offsets, reti->u.offsets, 2*len+1, U32); | |
12490 | } | |
12491 | #else | |
12492 | SetProgLen(reti,len); | |
12493 | #endif | |
12494 | ||
f8149455 | 12495 | return (void*)reti; |
84da74a7 | 12496 | } |
f8149455 YO |
12497 | |
12498 | #endif /* USE_ITHREADS */ | |
84da74a7 | 12499 | |
f8149455 | 12500 | #ifndef PERL_IN_XSUB_RE |
bcdf7404 | 12501 | |
c277df42 IZ |
12502 | /* |
12503 | - regnext - dig the "next" pointer out of a node | |
c277df42 IZ |
12504 | */ |
12505 | regnode * | |
864dbfa3 | 12506 | Perl_regnext(pTHX_ register regnode *p) |
c277df42 | 12507 | { |
97aff369 | 12508 | dVAR; |
c277df42 IZ |
12509 | register I32 offset; |
12510 | ||
f8fc2ecf | 12511 | if (!p) |
c277df42 IZ |
12512 | return(NULL); |
12513 | ||
35db910f KW |
12514 | if (OP(p) > REGNODE_MAX) { /* regnode.type is unsigned */ |
12515 | Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(p), (int)REGNODE_MAX); | |
12516 | } | |
12517 | ||
c277df42 IZ |
12518 | offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p)); |
12519 | if (offset == 0) | |
12520 | return(NULL); | |
12521 | ||
c277df42 | 12522 | return(p+offset); |
c277df42 | 12523 | } |
76234dfb | 12524 | #endif |
c277df42 | 12525 | |
686b73d4 | 12526 | STATIC void |
cea2e8a9 | 12527 | S_re_croak2(pTHX_ const char* pat1,const char* pat2,...) |
c277df42 IZ |
12528 | { |
12529 | va_list args; | |
12530 | STRLEN l1 = strlen(pat1); | |
12531 | STRLEN l2 = strlen(pat2); | |
12532 | char buf[512]; | |
06bf62c7 | 12533 | SV *msv; |
73d840c0 | 12534 | const char *message; |
c277df42 | 12535 | |
7918f24d NC |
12536 | PERL_ARGS_ASSERT_RE_CROAK2; |
12537 | ||
c277df42 IZ |
12538 | if (l1 > 510) |
12539 | l1 = 510; | |
12540 | if (l1 + l2 > 510) | |
12541 | l2 = 510 - l1; | |
12542 | Copy(pat1, buf, l1 , char); | |
12543 | Copy(pat2, buf + l1, l2 , char); | |
3b818b81 GS |
12544 | buf[l1 + l2] = '\n'; |
12545 | buf[l1 + l2 + 1] = '\0'; | |
8736538c AS |
12546 | #ifdef I_STDARG |
12547 | /* ANSI variant takes additional second argument */ | |
c277df42 | 12548 | va_start(args, pat2); |
8736538c AS |
12549 | #else |
12550 | va_start(args); | |
12551 | #endif | |
5a844595 | 12552 | msv = vmess(buf, &args); |
c277df42 | 12553 | va_end(args); |
cfd0369c | 12554 | message = SvPV_const(msv,l1); |
c277df42 IZ |
12555 | if (l1 > 512) |
12556 | l1 = 512; | |
12557 | Copy(message, buf, l1 , char); | |
197cf9b9 | 12558 | buf[l1-1] = '\0'; /* Overwrite \n */ |
cea2e8a9 | 12559 | Perl_croak(aTHX_ "%s", buf); |
c277df42 | 12560 | } |
a0ed51b3 LW |
12561 | |
12562 | /* XXX Here's a total kludge. But we need to re-enter for swash routines. */ | |
12563 | ||
76234dfb | 12564 | #ifndef PERL_IN_XSUB_RE |
a0ed51b3 | 12565 | void |
864dbfa3 | 12566 | Perl_save_re_context(pTHX) |
b81d288d | 12567 | { |
97aff369 | 12568 | dVAR; |
1ade1aa1 NC |
12569 | |
12570 | struct re_save_state *state; | |
12571 | ||
12572 | SAVEVPTR(PL_curcop); | |
12573 | SSGROW(SAVESTACK_ALLOC_FOR_RE_SAVE_STATE + 1); | |
12574 | ||
12575 | state = (struct re_save_state *)(PL_savestack + PL_savestack_ix); | |
12576 | PL_savestack_ix += SAVESTACK_ALLOC_FOR_RE_SAVE_STATE; | |
c6bf6a65 | 12577 | SSPUSHUV(SAVEt_RE_STATE); |
1ade1aa1 | 12578 | |
46ab3289 | 12579 | Copy(&PL_reg_state, state, 1, struct re_save_state); |
1ade1aa1 | 12580 | |
a0ed51b3 | 12581 | PL_reg_start_tmp = 0; |
a0ed51b3 | 12582 | PL_reg_start_tmpl = 0; |
c445ea15 | 12583 | PL_reg_oldsaved = NULL; |
a5db57d6 | 12584 | PL_reg_oldsavedlen = 0; |
a5db57d6 | 12585 | PL_reg_maxiter = 0; |
a5db57d6 | 12586 | PL_reg_leftiter = 0; |
c445ea15 | 12587 | PL_reg_poscache = NULL; |
a5db57d6 | 12588 | PL_reg_poscache_size = 0; |
1ade1aa1 NC |
12589 | #ifdef PERL_OLD_COPY_ON_WRITE |
12590 | PL_nrs = NULL; | |
12591 | #endif | |
ada6e8a9 | 12592 | |
c445ea15 AL |
12593 | /* Save $1..$n (#18107: UTF-8 s/(\w+)/uc($1)/e); AMS 20021106. */ |
12594 | if (PL_curpm) { | |
12595 | const REGEXP * const rx = PM_GETRE(PL_curpm); | |
12596 | if (rx) { | |
1df70142 | 12597 | U32 i; |
07bc277f | 12598 | for (i = 1; i <= RX_NPARENS(rx); i++) { |
1df70142 | 12599 | char digits[TYPE_CHARS(long)]; |
d9fad198 | 12600 | const STRLEN len = my_snprintf(digits, sizeof(digits), "%lu", (long)i); |
49f27e4b NC |
12601 | GV *const *const gvp |
12602 | = (GV**)hv_fetch(PL_defstash, digits, len, 0); | |
12603 | ||
b37c2d43 AL |
12604 | if (gvp) { |
12605 | GV * const gv = *gvp; | |
12606 | if (SvTYPE(gv) == SVt_PVGV && GvSV(gv)) | |
12607 | save_scalar(gv); | |
49f27e4b | 12608 | } |
ada6e8a9 AMS |
12609 | } |
12610 | } | |
12611 | } | |
a0ed51b3 | 12612 | } |
76234dfb | 12613 | #endif |
51371543 | 12614 | |
51371543 | 12615 | static void |
acfe0abc | 12616 | clear_re(pTHX_ void *r) |
51371543 | 12617 | { |
97aff369 | 12618 | dVAR; |
84679df5 | 12619 | ReREFCNT_dec((REGEXP *)r); |
51371543 | 12620 | } |
ffbc6a93 | 12621 | |
a28509cc AL |
12622 | #ifdef DEBUGGING |
12623 | ||
12624 | STATIC void | |
12625 | S_put_byte(pTHX_ SV *sv, int c) | |
12626 | { | |
7918f24d NC |
12627 | PERL_ARGS_ASSERT_PUT_BYTE; |
12628 | ||
7fddd944 NC |
12629 | /* Our definition of isPRINT() ignores locales, so only bytes that are |
12630 | not part of UTF-8 are considered printable. I assume that the same | |
12631 | holds for UTF-EBCDIC. | |
12632 | Also, code point 255 is not printable in either (it's E0 in EBCDIC, | |
12633 | which Wikipedia says: | |
12634 | ||
12635 | EO, or Eight Ones, is an 8-bit EBCDIC character code represented as all | |
12636 | ones (binary 1111 1111, hexadecimal FF). It is similar, but not | |
12637 | identical, to the ASCII delete (DEL) or rubout control character. | |
12638 | ) So the old condition can be simplified to !isPRINT(c) */ | |
9ce2357e KW |
12639 | if (!isPRINT(c)) { |
12640 | if (c < 256) { | |
12641 | Perl_sv_catpvf(aTHX_ sv, "\\x%02x", c); | |
12642 | } | |
12643 | else { | |
12644 | Perl_sv_catpvf(aTHX_ sv, "\\x{%x}", c); | |
12645 | } | |
12646 | } | |
5e7aa789 | 12647 | else { |
88c9ea1e | 12648 | const char string = c; |
5e7aa789 NC |
12649 | if (c == '-' || c == ']' || c == '\\' || c == '^') |
12650 | sv_catpvs(sv, "\\"); | |
12651 | sv_catpvn(sv, &string, 1); | |
12652 | } | |
a28509cc AL |
12653 | } |
12654 | ||
786e8c11 | 12655 | |
3dab1dad YO |
12656 | #define CLEAR_OPTSTART \ |
12657 | if (optstart) STMT_START { \ | |
70685ca0 | 12658 | DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \ |
3dab1dad YO |
12659 | optstart=NULL; \ |
12660 | } STMT_END | |
12661 | ||
786e8c11 | 12662 | #define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1); |
3dab1dad | 12663 | |
b5a2f8d8 NC |
12664 | STATIC const regnode * |
12665 | S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node, | |
786e8c11 YO |
12666 | const regnode *last, const regnode *plast, |
12667 | SV* sv, I32 indent, U32 depth) | |
a28509cc | 12668 | { |
97aff369 | 12669 | dVAR; |
786e8c11 | 12670 | register U8 op = PSEUDO; /* Arbitrary non-END op. */ |
b5a2f8d8 | 12671 | register const regnode *next; |
3dab1dad | 12672 | const regnode *optstart= NULL; |
1f1031fe | 12673 | |
f8fc2ecf | 12674 | RXi_GET_DECL(r,ri); |
3dab1dad | 12675 | GET_RE_DEBUG_FLAGS_DECL; |
7918f24d NC |
12676 | |
12677 | PERL_ARGS_ASSERT_DUMPUNTIL; | |
12678 | ||
786e8c11 YO |
12679 | #ifdef DEBUG_DUMPUNTIL |
12680 | PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start, | |
12681 | last ? last-start : 0,plast ? plast-start : 0); | |
12682 | #endif | |
12683 | ||
12684 | if (plast && plast < last) | |
12685 | last= plast; | |
12686 | ||
12687 | while (PL_regkind[op] != END && (!last || node < last)) { | |
a28509cc | 12688 | /* While that wasn't END last time... */ |
a28509cc AL |
12689 | NODE_ALIGN(node); |
12690 | op = OP(node); | |
de734bd5 | 12691 | if (op == CLOSE || op == WHILEM) |
786e8c11 | 12692 | indent--; |
b5a2f8d8 | 12693 | next = regnext((regnode *)node); |
1f1031fe | 12694 | |
a28509cc | 12695 | /* Where, what. */ |
8e11feef | 12696 | if (OP(node) == OPTIMIZED) { |
e68ec53f | 12697 | if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE)) |
8e11feef | 12698 | optstart = node; |
3dab1dad | 12699 | else |
8e11feef | 12700 | goto after_print; |
3dab1dad YO |
12701 | } else |
12702 | CLEAR_OPTSTART; | |
686b73d4 | 12703 | |
32fc9b6a | 12704 | regprop(r, sv, node); |
a28509cc | 12705 | PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start), |
786e8c11 | 12706 | (int)(2*indent + 1), "", SvPVX_const(sv)); |
1f1031fe YO |
12707 | |
12708 | if (OP(node) != OPTIMIZED) { | |
12709 | if (next == NULL) /* Next ptr. */ | |
12710 | PerlIO_printf(Perl_debug_log, " (0)"); | |
12711 | else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH ) | |
12712 | PerlIO_printf(Perl_debug_log, " (FAIL)"); | |
12713 | else | |
12714 | PerlIO_printf(Perl_debug_log, " (%"IVdf")", (IV)(next - start)); | |
12715 | (void)PerlIO_putc(Perl_debug_log, '\n'); | |
12716 | } | |
12717 | ||
a28509cc AL |
12718 | after_print: |
12719 | if (PL_regkind[(U8)op] == BRANCHJ) { | |
be8e71aa YO |
12720 | assert(next); |
12721 | { | |
12722 | register const regnode *nnode = (OP(next) == LONGJMP | |
b5a2f8d8 NC |
12723 | ? regnext((regnode *)next) |
12724 | : next); | |
be8e71aa YO |
12725 | if (last && nnode > last) |
12726 | nnode = last; | |
786e8c11 | 12727 | DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode); |
be8e71aa | 12728 | } |
a28509cc AL |
12729 | } |
12730 | else if (PL_regkind[(U8)op] == BRANCH) { | |
be8e71aa | 12731 | assert(next); |
786e8c11 | 12732 | DUMPUNTIL(NEXTOPER(node), next); |
a28509cc AL |
12733 | } |
12734 | else if ( PL_regkind[(U8)op] == TRIE ) { | |
7f69552c | 12735 | const regnode *this_trie = node; |
1de06328 | 12736 | const char op = OP(node); |
647f639f | 12737 | const U32 n = ARG(node); |
1de06328 | 12738 | const reg_ac_data * const ac = op>=AHOCORASICK ? |
f8fc2ecf | 12739 | (reg_ac_data *)ri->data->data[n] : |
1de06328 | 12740 | NULL; |
3251b653 NC |
12741 | const reg_trie_data * const trie = |
12742 | (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie]; | |
2b8b4781 | 12743 | #ifdef DEBUGGING |
502c6561 | 12744 | AV *const trie_words = MUTABLE_AV(ri->data->data[n + TRIE_WORDS_OFFSET]); |
2b8b4781 | 12745 | #endif |
786e8c11 | 12746 | const regnode *nextbranch= NULL; |
a28509cc | 12747 | I32 word_idx; |
76f68e9b | 12748 | sv_setpvs(sv, ""); |
786e8c11 | 12749 | for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) { |
2b8b4781 | 12750 | SV ** const elem_ptr = av_fetch(trie_words,word_idx,0); |
686b73d4 | 12751 | |
786e8c11 YO |
12752 | PerlIO_printf(Perl_debug_log, "%*s%s ", |
12753 | (int)(2*(indent+3)), "", | |
12754 | elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60, | |
ab3bbdeb YO |
12755 | PL_colors[0], PL_colors[1], |
12756 | (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) | | |
95b611b0 | 12757 | PERL_PV_PRETTY_ELLIPSES | |
7f69552c | 12758 | PERL_PV_PRETTY_LTGT |
786e8c11 YO |
12759 | ) |
12760 | : "???" | |
12761 | ); | |
12762 | if (trie->jump) { | |
40d049e4 | 12763 | U16 dist= trie->jump[word_idx+1]; |
70685ca0 JH |
12764 | PerlIO_printf(Perl_debug_log, "(%"UVuf")\n", |
12765 | (UV)((dist ? this_trie + dist : next) - start)); | |
786e8c11 YO |
12766 | if (dist) { |
12767 | if (!nextbranch) | |
24b23f37 | 12768 | nextbranch= this_trie + trie->jump[0]; |
7f69552c YO |
12769 | DUMPUNTIL(this_trie + dist, nextbranch); |
12770 | } | |
786e8c11 YO |
12771 | if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) |
12772 | nextbranch= regnext((regnode *)nextbranch); | |
12773 | } else { | |
12774 | PerlIO_printf(Perl_debug_log, "\n"); | |
a28509cc | 12775 | } |
786e8c11 YO |
12776 | } |
12777 | if (last && next > last) | |
12778 | node= last; | |
12779 | else | |
12780 | node= next; | |
a28509cc | 12781 | } |
786e8c11 YO |
12782 | else if ( op == CURLY ) { /* "next" might be very big: optimizer */ |
12783 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, | |
12784 | NEXTOPER(node) + EXTRA_STEP_2ARGS + 1); | |
a28509cc AL |
12785 | } |
12786 | else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) { | |
be8e71aa | 12787 | assert(next); |
786e8c11 | 12788 | DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next); |
a28509cc AL |
12789 | } |
12790 | else if ( op == PLUS || op == STAR) { | |
786e8c11 | 12791 | DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1); |
a28509cc | 12792 | } |
f56b6394 | 12793 | else if (PL_regkind[(U8)op] == ANYOF) { |
a28509cc | 12794 | /* arglen 1 + class block */ |
4a3ee7a8 | 12795 | node += 1 + ((ANYOF_FLAGS(node) & ANYOF_CLASS) |
a28509cc AL |
12796 | ? ANYOF_CLASS_SKIP : ANYOF_SKIP); |
12797 | node = NEXTOPER(node); | |
12798 | } | |
12799 | else if (PL_regkind[(U8)op] == EXACT) { | |
12800 | /* Literal string, where present. */ | |
12801 | node += NODE_SZ_STR(node) - 1; | |
12802 | node = NEXTOPER(node); | |
12803 | } | |
12804 | else { | |
12805 | node = NEXTOPER(node); | |
12806 | node += regarglen[(U8)op]; | |
12807 | } | |
12808 | if (op == CURLYX || op == OPEN) | |
786e8c11 | 12809 | indent++; |
a28509cc | 12810 | } |
3dab1dad | 12811 | CLEAR_OPTSTART; |
786e8c11 | 12812 | #ifdef DEBUG_DUMPUNTIL |
70685ca0 | 12813 | PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent); |
786e8c11 | 12814 | #endif |
1de06328 | 12815 | return node; |
a28509cc AL |
12816 | } |
12817 | ||
12818 | #endif /* DEBUGGING */ | |
12819 | ||
241d1a3b NC |
12820 | /* |
12821 | * Local variables: | |
12822 | * c-indentation-style: bsd | |
12823 | * c-basic-offset: 4 | |
12824 | * indent-tabs-mode: t | |
12825 | * End: | |
12826 | * | |
37442d52 RGS |
12827 | * ex: set ts=8 sts=4 sw=4 noet: |
12828 | */ |