1 | /* Extended regular expression matching and search library, |
2 | version 0.12. |
3 | (Implements POSIX draft P1003.2/D11.2, except for some of the |
4 | internationalization features.) |
5 | |
6 | Copyright (C) 1993-2024 Free Software Foundation, Inc. |
7 | This file is part of the GNU C Library. |
8 | |
9 | The GNU C Library is free software; you can redistribute it and/or |
10 | modify it under the terms of the GNU Lesser General Public |
11 | License as published by the Free Software Foundation; either |
12 | version 2.1 of the License, or (at your option) any later version. |
13 | |
14 | The GNU C Library is distributed in the hope that it will be useful, |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
17 | Lesser General Public License for more details. |
18 | |
19 | You should have received a copy of the GNU Lesser General Public |
20 | License along with the GNU C Library; if not, write to the Free |
21 | Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA |
22 | 02110-1301 USA. */ |
23 | |
24 | /* This file has been modified for usage in libiberty. It includes "xregex.h" |
25 | instead of <regex.h>. The "xregex.h" header file renames all external |
26 | routines with an "x" prefix so they do not collide with the native regex |
27 | routines or with other components regex routines. */ |
28 | /* AIX requires this to be the first thing in the file. */ |
29 | #if defined _AIX && !defined __GNUC__ && !defined REGEX_MALLOC |
30 | #pragma alloca |
31 | #endif |
32 | |
33 | #if __GNUC__ >= 12 |
34 | # pragma GCC diagnostic ignored "-Wuse-after-free" |
35 | #endif |
36 | |
37 | #undef _GNU_SOURCE |
38 | #define _GNU_SOURCE |
39 | |
40 | #ifndef INSIDE_RECURSION |
41 | # ifdef HAVE_CONFIG_H |
42 | # include <config.h> |
43 | # endif |
44 | #endif |
45 | |
46 | #include <ansidecl.h> |
47 | |
48 | #ifndef INSIDE_RECURSION |
49 | |
50 | # if defined STDC_HEADERS && !defined emacs |
51 | # include <stddef.h> |
52 | # define PTR_INT_TYPE ptrdiff_t |
53 | # else |
54 | /* We need this for `regex.h', and perhaps for the Emacs include files. */ |
55 | # include <sys/types.h> |
56 | # define PTR_INT_TYPE long |
57 | # endif |
58 | |
59 | # define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC) |
60 | |
61 | /* For platform which support the ISO C amendement 1 functionality we |
62 | support user defined character classes. */ |
63 | # if defined _LIBC || WIDE_CHAR_SUPPORT |
64 | /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */ |
65 | # include <wchar.h> |
66 | # include <wctype.h> |
67 | # endif |
68 | |
69 | # ifdef _LIBC |
70 | /* We have to keep the namespace clean. */ |
71 | # define regfree(preg) __regfree (preg) |
72 | # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef) |
73 | # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags) |
74 | # define regerror(errcode, preg, errbuf, errbuf_size) \ |
75 | __regerror(errcode, preg, errbuf, errbuf_size) |
76 | # define re_set_registers(bu, re, nu, st, en) \ |
77 | __re_set_registers (bu, re, nu, st, en) |
78 | # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \ |
79 | __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop) |
80 | # define re_match(bufp, string, size, pos, regs) \ |
81 | __re_match (bufp, string, size, pos, regs) |
82 | # define re_search(bufp, string, size, startpos, range, regs) \ |
83 | __re_search (bufp, string, size, startpos, range, regs) |
84 | # define re_compile_pattern(pattern, length, bufp) \ |
85 | __re_compile_pattern (pattern, length, bufp) |
86 | # define re_set_syntax(syntax) __re_set_syntax (syntax) |
87 | # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \ |
88 | __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop) |
89 | # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp) |
90 | |
91 | # define btowc __btowc |
92 | |
93 | /* We are also using some library internals. */ |
94 | # include <locale/localeinfo.h> |
95 | # include <locale/elem-hash.h> |
96 | # include <langinfo.h> |
97 | # include <locale/coll-lookup.h> |
98 | # endif |
99 | |
100 | /* This is for other GNU distributions with internationalized messages. */ |
101 | # if (HAVE_LIBINTL_H && ENABLE_NLS) || defined _LIBC |
102 | # include <libintl.h> |
103 | # ifdef _LIBC |
104 | # undef gettext |
105 | # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES) |
106 | # endif |
107 | # else |
108 | # define gettext(msgid) (msgid) |
109 | # endif |
110 | |
111 | # ifndef gettext_noop |
112 | /* This define is so xgettext can find the internationalizable |
113 | strings. */ |
114 | # define gettext_noop(String) String |
115 | # endif |
116 | |
117 | /* The `emacs' switch turns on certain matching commands |
118 | that make sense only in Emacs. */ |
119 | # ifdef emacs |
120 | |
121 | # include "lisp.h" |
122 | # include "buffer.h" |
123 | # include "syntax.h" |
124 | |
125 | # else /* not emacs */ |
126 | |
127 | /* If we are not linking with Emacs proper, |
128 | we can't use the relocating allocator |
129 | even if config.h says that we can. */ |
130 | # undef REL_ALLOC |
131 | |
132 | # if defined STDC_HEADERS || defined _LIBC |
133 | # include <stdlib.h> |
134 | # else |
135 | char *malloc (); |
136 | char *realloc (); |
137 | # endif |
138 | |
139 | /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow. |
140 | If nothing else has been done, use the method below. */ |
141 | # ifdef INHIBIT_STRING_HEADER |
142 | # if !(defined HAVE_BZERO && defined HAVE_BCOPY) |
143 | # if !defined bzero && !defined bcopy |
144 | # undef INHIBIT_STRING_HEADER |
145 | # endif |
146 | # endif |
147 | # endif |
148 | |
149 | /* This is the normal way of making sure we have a bcopy and a bzero. |
150 | This is used in most programs--a few other programs avoid this |
151 | by defining INHIBIT_STRING_HEADER. */ |
152 | # ifndef INHIBIT_STRING_HEADER |
153 | # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC |
154 | # include <string.h> |
155 | # ifndef bzero |
156 | # ifndef _LIBC |
157 | # define bzero(s, n) ((void) memset (s, '\0', n)) |
158 | # else |
159 | # define bzero(s, n) __bzero (s, n) |
160 | # endif |
161 | # endif |
162 | # else |
163 | # include <strings.h> |
164 | # ifndef memcmp |
165 | # define memcmp(s1, s2, n) bcmp (s1, s2, n) |
166 | # endif |
167 | # ifndef memcpy |
168 | # define memcpy(d, s, n) (bcopy (s, d, n), (d)) |
169 | # endif |
170 | # endif |
171 | # endif |
172 | |
173 | /* Define the syntax stuff for \<, \>, etc. */ |
174 | |
175 | /* This must be nonzero for the wordchar and notwordchar pattern |
176 | commands in re_match_2. */ |
177 | # ifndef Sword |
178 | # define Sword 1 |
179 | # endif |
180 | |
181 | # ifdef SWITCH_ENUM_BUG |
182 | # define SWITCH_ENUM_CAST(x) ((int)(x)) |
183 | # else |
184 | # define SWITCH_ENUM_CAST(x) (x) |
185 | # endif |
186 | |
187 | # endif /* not emacs */ |
188 | |
189 | # if defined _LIBC || HAVE_LIMITS_H |
190 | # include <limits.h> |
191 | # endif |
192 | |
193 | # ifndef MB_LEN_MAX |
194 | # define MB_LEN_MAX 1 |
195 | # endif |
196 | |
197 | /* Get the interface, including the syntax bits. */ |
198 | # include "xregex.h" /* change for libiberty */ |
199 | |
200 | /* isalpha etc. are used for the character classes. */ |
201 | # include <ctype.h> |
202 | |
203 | /* Jim Meyering writes: |
204 | |
205 | "... Some ctype macros are valid only for character codes that |
206 | isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when |
207 | using /bin/cc or gcc but without giving an ansi option). So, all |
208 | ctype uses should be through macros like ISPRINT... If |
209 | STDC_HEADERS is defined, then autoconf has verified that the ctype |
210 | macros don't need to be guarded with references to isascii. ... |
211 | Defining isascii to 1 should let any compiler worth its salt |
212 | eliminate the && through constant folding." |
213 | Solaris defines some of these symbols so we must undefine them first. */ |
214 | |
215 | # undef ISASCII |
216 | # if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII) |
217 | # define ISASCII(c) 1 |
218 | # else |
219 | # define ISASCII(c) isascii(c) |
220 | # endif |
221 | |
222 | # ifdef isblank |
223 | # define ISBLANK(c) (ISASCII (c) && isblank (c)) |
224 | # else |
225 | # define ISBLANK(c) ((c) == ' ' || (c) == '\t') |
226 | # endif |
227 | # ifdef isgraph |
228 | # define ISGRAPH(c) (ISASCII (c) && isgraph (c)) |
229 | # else |
230 | # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c)) |
231 | # endif |
232 | |
233 | # undef ISPRINT |
234 | # define ISPRINT(c) (ISASCII (c) && isprint (c)) |
235 | # define ISDIGIT(c) (ISASCII (c) && isdigit (c)) |
236 | # define ISALNUM(c) (ISASCII (c) && isalnum (c)) |
237 | # define ISALPHA(c) (ISASCII (c) && isalpha (c)) |
238 | # define ISCNTRL(c) (ISASCII (c) && iscntrl (c)) |
239 | # define ISLOWER(c) (ISASCII (c) && islower (c)) |
240 | # define ISPUNCT(c) (ISASCII (c) && ispunct (c)) |
241 | # define ISSPACE(c) (ISASCII (c) && isspace (c)) |
242 | # define ISUPPER(c) (ISASCII (c) && isupper (c)) |
243 | # define ISXDIGIT(c) (ISASCII (c) && isxdigit (c)) |
244 | |
245 | # ifdef _tolower |
246 | # define TOLOWER(c) _tolower(c) |
247 | # else |
248 | # define TOLOWER(c) tolower(c) |
249 | # endif |
250 | |
251 | # ifndef NULL |
252 | # define NULL (void *)0 |
253 | # endif |
254 | |
255 | /* We remove any previous definition of `SIGN_EXTEND_CHAR', |
256 | since ours (we hope) works properly with all combinations of |
257 | machines, compilers, `char' and `unsigned char' argument types. |
258 | (Per Bothner suggested the basic approach.) */ |
259 | # undef SIGN_EXTEND_CHAR |
260 | # if __STDC__ |
261 | # define SIGN_EXTEND_CHAR(c) ((signed char) (c)) |
262 | # else /* not __STDC__ */ |
263 | /* As in Harbison and Steele. */ |
264 | # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128) |
265 | # endif |
266 | |
267 | # ifndef emacs |
268 | /* How many characters in the character set. */ |
269 | # define CHAR_SET_SIZE 256 |
270 | |
271 | # ifdef SYNTAX_TABLE |
272 | |
273 | extern char *re_syntax_table; |
274 | |
275 | # else /* not SYNTAX_TABLE */ |
276 | |
277 | static char re_syntax_table[CHAR_SET_SIZE]; |
278 | |
279 | static void init_syntax_once (void); |
280 | |
281 | static void |
282 | init_syntax_once (void) |
283 | { |
284 | register int c; |
285 | static int done = 0; |
286 | |
287 | if (done) |
288 | return; |
289 | bzero (re_syntax_table, sizeof re_syntax_table); |
290 | |
291 | for (c = 0; c < CHAR_SET_SIZE; ++c) |
292 | if (ISALNUM (c)) |
293 | re_syntax_table[c] = Sword; |
294 | |
295 | re_syntax_table['_'] = Sword; |
296 | |
297 | done = 1; |
298 | } |
299 | |
300 | # endif /* not SYNTAX_TABLE */ |
301 | |
302 | # define SYNTAX(c) re_syntax_table[(unsigned char) (c)] |
303 | |
304 | # endif /* emacs */ |
305 | |
306 | /* Integer type for pointers. */ |
307 | # if !defined _LIBC && !defined HAVE_UINTPTR_T |
308 | typedef unsigned long int uintptr_t; |
309 | # endif |
310 | |
311 | /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we |
312 | use `alloca' instead of `malloc'. This is because using malloc in |
313 | re_search* or re_match* could cause memory leaks when C-g is used in |
314 | Emacs; also, malloc is slower and causes storage fragmentation. On |
315 | the other hand, malloc is more portable, and easier to debug. |
316 | |
317 | Because we sometimes use alloca, some routines have to be macros, |
318 | not functions -- `alloca'-allocated space disappears at the end of the |
319 | function it is called in. */ |
320 | |
321 | # ifdef REGEX_MALLOC |
322 | |
323 | # define REGEX_ALLOCATE malloc |
324 | # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize) |
325 | # define REGEX_FREE free |
326 | |
327 | # else /* not REGEX_MALLOC */ |
328 | |
329 | /* Emacs already defines alloca, sometimes. */ |
330 | # ifndef alloca |
331 | |
332 | /* Make alloca work the best possible way. */ |
333 | # ifdef __GNUC__ |
334 | # define alloca __builtin_alloca |
335 | # else /* not __GNUC__ */ |
336 | # if HAVE_ALLOCA_H |
337 | # include <alloca.h> |
338 | # endif /* HAVE_ALLOCA_H */ |
339 | # endif /* not __GNUC__ */ |
340 | |
341 | # endif /* not alloca */ |
342 | |
343 | # define REGEX_ALLOCATE alloca |
344 | |
345 | /* Assumes a `char *destination' variable. */ |
346 | # define REGEX_REALLOCATE(source, osize, nsize) \ |
347 | (destination = (char *) alloca (nsize), \ |
348 | memcpy (destination, source, osize)) |
349 | |
350 | /* No need to do anything to free, after alloca. */ |
351 | # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */ |
352 | |
353 | # endif /* not REGEX_MALLOC */ |
354 | |
355 | /* Define how to allocate the failure stack. */ |
356 | |
357 | # if defined REL_ALLOC && defined REGEX_MALLOC |
358 | |
359 | # define REGEX_ALLOCATE_STACK(size) \ |
360 | r_alloc (&failure_stack_ptr, (size)) |
361 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
362 | r_re_alloc (&failure_stack_ptr, (nsize)) |
363 | # define REGEX_FREE_STACK(ptr) \ |
364 | r_alloc_free (&failure_stack_ptr) |
365 | |
366 | # else /* not using relocating allocator */ |
367 | |
368 | # ifdef REGEX_MALLOC |
369 | |
370 | # define REGEX_ALLOCATE_STACK malloc |
371 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize) |
372 | # define REGEX_FREE_STACK free |
373 | |
374 | # else /* not REGEX_MALLOC */ |
375 | |
376 | # define REGEX_ALLOCATE_STACK alloca |
377 | |
378 | # define REGEX_REALLOCATE_STACK(source, osize, nsize) \ |
379 | REGEX_REALLOCATE (source, osize, nsize) |
380 | /* No need to explicitly free anything. */ |
381 | # define REGEX_FREE_STACK(arg) |
382 | |
383 | # endif /* not REGEX_MALLOC */ |
384 | # endif /* not using relocating allocator */ |
385 | |
386 | |
387 | /* True if `size1' is non-NULL and PTR is pointing anywhere inside |
388 | `string1' or just past its end. This works if PTR is NULL, which is |
389 | a good thing. */ |
390 | # define FIRST_STRING_P(ptr) \ |
391 | (size1 && string1 <= (ptr) && (ptr) <= string1 + size1) |
392 | |
393 | /* (Re)Allocate N items of type T using malloc, or fail. */ |
394 | # define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t))) |
395 | # define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t))) |
396 | # define RETALLOC_IF(addr, n, t) \ |
397 | if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t) |
398 | # define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t))) |
399 | |
400 | # define BYTEWIDTH 8 /* In bits. */ |
401 | |
402 | # define STREQ(s1, s2) ((strcmp (s1, s2) == 0)) |
403 | |
404 | # undef MAX |
405 | # undef MIN |
406 | # define MAX(a, b) ((a) > (b) ? (a) : (b)) |
407 | # define MIN(a, b) ((a) < (b) ? (a) : (b)) |
408 | |
409 | typedef char boolean; |
410 | # define false 0 |
411 | # define true 1 |
412 | |
413 | static reg_errcode_t byte_regex_compile (const char *pattern, size_t size, |
414 | reg_syntax_t syntax, |
415 | struct re_pattern_buffer *bufp); |
416 | |
417 | static int byte_re_match_2_internal (struct re_pattern_buffer *bufp, |
418 | const char *string1, int size1, |
419 | const char *string2, int size2, |
420 | int pos, |
421 | struct re_registers *regs, |
422 | int stop); |
423 | static int byte_re_search_2 (struct re_pattern_buffer *bufp, |
424 | const char *string1, int size1, |
425 | const char *string2, int size2, |
426 | int startpos, int range, |
427 | struct re_registers *regs, int stop); |
428 | static int byte_re_compile_fastmap (struct re_pattern_buffer *bufp); |
429 | |
430 | #ifdef MBS_SUPPORT |
431 | static reg_errcode_t wcs_regex_compile (const char *pattern, size_t size, |
432 | reg_syntax_t syntax, |
433 | struct re_pattern_buffer *bufp); |
434 | |
435 | |
436 | static int wcs_re_match_2_internal (struct re_pattern_buffer *bufp, |
437 | const char *cstring1, int csize1, |
438 | const char *cstring2, int csize2, |
439 | int pos, |
440 | struct re_registers *regs, |
441 | int stop, |
442 | wchar_t *string1, int size1, |
443 | wchar_t *string2, int size2, |
444 | int *mbs_offset1, int *mbs_offset2); |
445 | static int wcs_re_search_2 (struct re_pattern_buffer *bufp, |
446 | const char *string1, int size1, |
447 | const char *string2, int size2, |
448 | int startpos, int range, |
449 | struct re_registers *regs, int stop); |
450 | static int wcs_re_compile_fastmap (struct re_pattern_buffer *bufp); |
451 | #endif |
452 | |
453 | /* These are the command codes that appear in compiled regular |
454 | expressions. Some opcodes are followed by argument bytes. A |
455 | command code can specify any interpretation whatsoever for its |
456 | arguments. Zero bytes may appear in the compiled regular expression. */ |
457 | |
458 | typedef enum |
459 | { |
460 | no_op = 0, |
461 | |
462 | /* Succeed right away--no more backtracking. */ |
463 | succeed, |
464 | |
465 | /* Followed by one byte giving n, then by n literal bytes. */ |
466 | exactn, |
467 | |
468 | # ifdef MBS_SUPPORT |
469 | /* Same as exactn, but contains binary data. */ |
470 | exactn_bin, |
471 | # endif |
472 | |
473 | /* Matches any (more or less) character. */ |
474 | anychar, |
475 | |
476 | /* Matches any one char belonging to specified set. First |
477 | following byte is number of bitmap bytes. Then come bytes |
478 | for a bitmap saying which chars are in. Bits in each byte |
479 | are ordered low-bit-first. A character is in the set if its |
480 | bit is 1. A character too large to have a bit in the map is |
481 | automatically not in the set. */ |
482 | /* ifdef MBS_SUPPORT, following element is length of character |
483 | classes, length of collating symbols, length of equivalence |
484 | classes, length of character ranges, and length of characters. |
485 | Next, character class element, collating symbols elements, |
486 | equivalence class elements, range elements, and character |
487 | elements follow. |
488 | See regex_compile function. */ |
489 | charset, |
490 | |
491 | /* Same parameters as charset, but match any character that is |
492 | not one of those specified. */ |
493 | charset_not, |
494 | |
495 | /* Start remembering the text that is matched, for storing in a |
496 | register. Followed by one byte with the register number, in |
497 | the range 0 to one less than the pattern buffer's re_nsub |
498 | field. Then followed by one byte with the number of groups |
499 | inner to this one. (This last has to be part of the |
500 | start_memory only because we need it in the on_failure_jump |
501 | of re_match_2.) */ |
502 | start_memory, |
503 | |
504 | /* Stop remembering the text that is matched and store it in a |
505 | memory register. Followed by one byte with the register |
506 | number, in the range 0 to one less than `re_nsub' in the |
507 | pattern buffer, and one byte with the number of inner groups, |
508 | just like `start_memory'. (We need the number of inner |
509 | groups here because we don't have any easy way of finding the |
510 | corresponding start_memory when we're at a stop_memory.) */ |
511 | stop_memory, |
512 | |
513 | /* Match a duplicate of something remembered. Followed by one |
514 | byte containing the register number. */ |
515 | duplicate, |
516 | |
517 | /* Fail unless at beginning of line. */ |
518 | begline, |
519 | |
520 | /* Fail unless at end of line. */ |
521 | endline, |
522 | |
523 | /* Succeeds if at beginning of buffer (if emacs) or at beginning |
524 | of string to be matched (if not). */ |
525 | begbuf, |
526 | |
527 | /* Analogously, for end of buffer/string. */ |
528 | endbuf, |
529 | |
530 | /* Followed by two byte relative address to which to jump. */ |
531 | jump, |
532 | |
533 | /* Same as jump, but marks the end of an alternative. */ |
534 | jump_past_alt, |
535 | |
536 | /* Followed by two-byte relative address of place to resume at |
537 | in case of failure. */ |
538 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
539 | on_failure_jump, |
540 | |
541 | /* Like on_failure_jump, but pushes a placeholder instead of the |
542 | current string position when executed. */ |
543 | on_failure_keep_string_jump, |
544 | |
545 | /* Throw away latest failure point and then jump to following |
546 | two-byte relative address. */ |
547 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
548 | pop_failure_jump, |
549 | |
550 | /* Change to pop_failure_jump if know won't have to backtrack to |
551 | match; otherwise change to jump. This is used to jump |
552 | back to the beginning of a repeat. If what follows this jump |
553 | clearly won't match what the repeat does, such that we can be |
554 | sure that there is no use backtracking out of repetitions |
555 | already matched, then we change it to a pop_failure_jump. |
556 | Followed by two-byte address. */ |
557 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
558 | maybe_pop_jump, |
559 | |
560 | /* Jump to following two-byte address, and push a dummy failure |
561 | point. This failure point will be thrown away if an attempt |
562 | is made to use it for a failure. A `+' construct makes this |
563 | before the first repeat. Also used as an intermediary kind |
564 | of jump when compiling an alternative. */ |
565 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
566 | dummy_failure_jump, |
567 | |
568 | /* Push a dummy failure point and continue. Used at the end of |
569 | alternatives. */ |
570 | push_dummy_failure, |
571 | |
572 | /* Followed by two-byte relative address and two-byte number n. |
573 | After matching N times, jump to the address upon failure. */ |
574 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
575 | succeed_n, |
576 | |
577 | /* Followed by two-byte relative address, and two-byte number n. |
578 | Jump to the address N times, then fail. */ |
579 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
580 | jump_n, |
581 | |
582 | /* Set the following two-byte relative address to the |
583 | subsequent two-byte number. The address *includes* the two |
584 | bytes of number. */ |
585 | /* ifdef MBS_SUPPORT, the size of address is 1. */ |
586 | set_number_at, |
587 | |
588 | wordchar, /* Matches any word-constituent character. */ |
589 | notwordchar, /* Matches any char that is not a word-constituent. */ |
590 | |
591 | wordbeg, /* Succeeds if at word beginning. */ |
592 | wordend, /* Succeeds if at word end. */ |
593 | |
594 | wordbound, /* Succeeds if at a word boundary. */ |
595 | notwordbound /* Succeeds if not at a word boundary. */ |
596 | |
597 | # ifdef emacs |
598 | ,before_dot, /* Succeeds if before point. */ |
599 | at_dot, /* Succeeds if at point. */ |
600 | after_dot, /* Succeeds if after point. */ |
601 | |
602 | /* Matches any character whose syntax is specified. Followed by |
603 | a byte which contains a syntax code, e.g., Sword. */ |
604 | syntaxspec, |
605 | |
606 | /* Matches any character whose syntax is not that specified. */ |
607 | notsyntaxspec |
608 | # endif /* emacs */ |
609 | } re_opcode_t; |
610 | #endif /* not INSIDE_RECURSION */ |
611 | |
612 | |
613 | #ifdef BYTE |
614 | # define CHAR_T char |
615 | # define UCHAR_T unsigned char |
616 | # define COMPILED_BUFFER_VAR bufp->buffer |
617 | # define OFFSET_ADDRESS_SIZE 2 |
618 | # define PREFIX(name) byte_##name |
619 | # define ARG_PREFIX(name) name |
620 | # define PUT_CHAR(c) putchar (c) |
621 | #else |
622 | # ifdef WCHAR |
623 | # define CHAR_T wchar_t |
624 | # define UCHAR_T wchar_t |
625 | # define COMPILED_BUFFER_VAR wc_buffer |
626 | # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */ |
627 | # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_T)+1) |
628 | # define PREFIX(name) wcs_##name |
629 | # define ARG_PREFIX(name) c##name |
630 | /* Should we use wide stream?? */ |
631 | # define PUT_CHAR(c) printf ("%C", c); |
632 | # define TRUE 1 |
633 | # define FALSE 0 |
634 | # else |
635 | # ifdef MBS_SUPPORT |
636 | # define WCHAR |
637 | # define INSIDE_RECURSION |
638 | # include "regex.c" |
639 | # undef INSIDE_RECURSION |
640 | # endif |
641 | # define BYTE |
642 | # define INSIDE_RECURSION |
643 | # include "regex.c" |
644 | # undef INSIDE_RECURSION |
645 | # endif |
646 | #endif |
647 | |
648 | #ifdef INSIDE_RECURSION |
649 | /* Common operations on the compiled pattern. */ |
650 | |
651 | /* Store NUMBER in two contiguous bytes starting at DESTINATION. */ |
652 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ |
653 | |
654 | # ifdef WCHAR |
655 | # define STORE_NUMBER(destination, number) \ |
656 | do { \ |
657 | *(destination) = (UCHAR_T)(number); \ |
658 | } while (0) |
659 | # else /* BYTE */ |
660 | # define STORE_NUMBER(destination, number) \ |
661 | do { \ |
662 | (destination)[0] = (number) & 0377; \ |
663 | (destination)[1] = (number) >> 8; \ |
664 | } while (0) |
665 | # endif /* WCHAR */ |
666 | |
667 | /* Same as STORE_NUMBER, except increment DESTINATION to |
668 | the byte after where the number is stored. Therefore, DESTINATION |
669 | must be an lvalue. */ |
670 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ |
671 | |
672 | # define STORE_NUMBER_AND_INCR(destination, number) \ |
673 | do { \ |
674 | STORE_NUMBER (destination, number); \ |
675 | (destination) += OFFSET_ADDRESS_SIZE; \ |
676 | } while (0) |
677 | |
678 | /* Put into DESTINATION a number stored in two contiguous bytes starting |
679 | at SOURCE. */ |
680 | /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */ |
681 | |
682 | # ifdef WCHAR |
683 | # define EXTRACT_NUMBER(destination, source) \ |
684 | do { \ |
685 | (destination) = *(source); \ |
686 | } while (0) |
687 | # else /* BYTE */ |
688 | # define EXTRACT_NUMBER(destination, source) \ |
689 | do { \ |
690 | (destination) = *(source) & 0377; \ |
691 | (destination) += ((unsigned) SIGN_EXTEND_CHAR (*((source) + 1))) << 8; \ |
692 | } while (0) |
693 | # endif |
694 | |
695 | # ifdef DEBUG |
696 | static void PREFIX(extract_number) (int *dest, UCHAR_T *source); |
697 | static void |
698 | PREFIX(extract_number) (int *dest, UCHAR_T *source) |
699 | { |
700 | # ifdef WCHAR |
701 | *dest = *source; |
702 | # else /* BYTE */ |
703 | int temp = SIGN_EXTEND_CHAR (*(source + 1)); |
704 | *dest = *source & 0377; |
705 | *dest += temp << 8; |
706 | # endif |
707 | } |
708 | |
709 | # ifndef EXTRACT_MACROS /* To debug the macros. */ |
710 | # undef EXTRACT_NUMBER |
711 | # define EXTRACT_NUMBER(dest, src) PREFIX(extract_number) (&dest, src) |
712 | # endif /* not EXTRACT_MACROS */ |
713 | |
714 | # endif /* DEBUG */ |
715 | |
716 | /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number. |
717 | SOURCE must be an lvalue. */ |
718 | |
719 | # define EXTRACT_NUMBER_AND_INCR(destination, source) \ |
720 | do { \ |
721 | EXTRACT_NUMBER (destination, source); \ |
722 | (source) += OFFSET_ADDRESS_SIZE; \ |
723 | } while (0) |
724 | |
725 | # ifdef DEBUG |
726 | static void PREFIX(extract_number_and_incr) (int *destination, |
727 | UCHAR_T **source); |
728 | static void |
729 | PREFIX(extract_number_and_incr) (int *destination, UCHAR_T **source) |
730 | { |
731 | PREFIX(extract_number) (destination, *source); |
732 | *source += OFFSET_ADDRESS_SIZE; |
733 | } |
734 | |
735 | # ifndef EXTRACT_MACROS |
736 | # undef EXTRACT_NUMBER_AND_INCR |
737 | # define EXTRACT_NUMBER_AND_INCR(dest, src) \ |
738 | PREFIX(extract_number_and_incr) (&dest, &src) |
739 | # endif /* not EXTRACT_MACROS */ |
740 | |
741 | # endif /* DEBUG */ |
742 | |
743 | |
744 | |
745 | /* If DEBUG is defined, Regex prints many voluminous messages about what |
746 | it is doing (if the variable `debug' is nonzero). If linked with the |
747 | main program in `iregex.c', you can enter patterns and strings |
748 | interactively. And if linked with the main program in `main.c' and |
749 | the other test files, you can run the already-written tests. */ |
750 | |
751 | # ifdef DEBUG |
752 | |
753 | # ifndef DEFINED_ONCE |
754 | |
755 | /* We use standard I/O for debugging. */ |
756 | # include <stdio.h> |
757 | |
758 | /* It is useful to test things that ``must'' be true when debugging. */ |
759 | # include <assert.h> |
760 | |
761 | static int debug; |
762 | |
763 | # define DEBUG_STATEMENT(e) e |
764 | # define DEBUG_PRINT1(x) if (debug) printf (x) |
765 | # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2) |
766 | # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3) |
767 | # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4) |
768 | # endif /* not DEFINED_ONCE */ |
769 | |
770 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \ |
771 | if (debug) PREFIX(print_partial_compiled_pattern) (s, e) |
772 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \ |
773 | if (debug) PREFIX(print_double_string) (w, s1, sz1, s2, sz2) |
774 | |
775 | |
776 | /* Print the fastmap in human-readable form. */ |
777 | |
778 | # ifndef DEFINED_ONCE |
779 | void |
780 | print_fastmap (char *fastmap) |
781 | { |
782 | unsigned was_a_range = 0; |
783 | unsigned i = 0; |
784 | |
785 | while (i < (1 << BYTEWIDTH)) |
786 | { |
787 | if (fastmap[i++]) |
788 | { |
789 | was_a_range = 0; |
790 | putchar (i - 1); |
791 | while (i < (1 << BYTEWIDTH) && fastmap[i]) |
792 | { |
793 | was_a_range = 1; |
794 | i++; |
795 | } |
796 | if (was_a_range) |
797 | { |
798 | printf ("-" ); |
799 | putchar (i - 1); |
800 | } |
801 | } |
802 | } |
803 | putchar ('\n'); |
804 | } |
805 | # endif /* not DEFINED_ONCE */ |
806 | |
807 | |
808 | /* Print a compiled pattern string in human-readable form, starting at |
809 | the START pointer into it and ending just before the pointer END. */ |
810 | |
811 | void |
812 | PREFIX(print_partial_compiled_pattern) (UCHAR_T *start, UCHAR_T *end) |
813 | { |
814 | int mcnt, mcnt2; |
815 | UCHAR_T *p1; |
816 | UCHAR_T *p = start; |
817 | UCHAR_T *pend = end; |
818 | |
819 | if (start == NULL) |
820 | { |
821 | printf ("(null)\n" ); |
822 | return; |
823 | } |
824 | |
825 | /* Loop over pattern commands. */ |
826 | while (p < pend) |
827 | { |
828 | # ifdef _LIBC |
829 | printf ("%td:\t" , p - start); |
830 | # else |
831 | printf ("%ld:\t" , (long int) (p - start)); |
832 | # endif |
833 | |
834 | switch ((re_opcode_t) *p++) |
835 | { |
836 | case no_op: |
837 | printf ("/no_op" ); |
838 | break; |
839 | |
840 | case exactn: |
841 | mcnt = *p++; |
842 | printf ("/exactn/%d" , mcnt); |
843 | do |
844 | { |
845 | putchar ('/'); |
846 | PUT_CHAR (*p++); |
847 | } |
848 | while (--mcnt); |
849 | break; |
850 | |
851 | # ifdef MBS_SUPPORT |
852 | case exactn_bin: |
853 | mcnt = *p++; |
854 | printf ("/exactn_bin/%d" , mcnt); |
855 | do |
856 | { |
857 | printf("/%lx" , (long int) *p++); |
858 | } |
859 | while (--mcnt); |
860 | break; |
861 | # endif /* MBS_SUPPORT */ |
862 | |
863 | case start_memory: |
864 | mcnt = *p++; |
865 | printf ("/start_memory/%d/%ld" , mcnt, (long int) *p++); |
866 | break; |
867 | |
868 | case stop_memory: |
869 | mcnt = *p++; |
870 | printf ("/stop_memory/%d/%ld" , mcnt, (long int) *p++); |
871 | break; |
872 | |
873 | case duplicate: |
874 | printf ("/duplicate/%ld" , (long int) *p++); |
875 | break; |
876 | |
877 | case anychar: |
878 | printf ("/anychar" ); |
879 | break; |
880 | |
881 | case charset: |
882 | case charset_not: |
883 | { |
884 | # ifdef WCHAR |
885 | int i, length; |
886 | wchar_t *workp = p; |
887 | printf ("/charset [%s" , |
888 | (re_opcode_t) *(workp - 1) == charset_not ? "^" : "" ); |
889 | p += 5; |
890 | length = *workp++; /* the length of char_classes */ |
891 | for (i=0 ; i<length ; i++) |
892 | printf("[:%lx:]" , (long int) *p++); |
893 | length = *workp++; /* the length of collating_symbol */ |
894 | for (i=0 ; i<length ;) |
895 | { |
896 | printf("[." ); |
897 | while(*p != 0) |
898 | PUT_CHAR((i++,*p++)); |
899 | i++,p++; |
900 | printf(".]" ); |
901 | } |
902 | length = *workp++; /* the length of equivalence_class */ |
903 | for (i=0 ; i<length ;) |
904 | { |
905 | printf("[=" ); |
906 | while(*p != 0) |
907 | PUT_CHAR((i++,*p++)); |
908 | i++,p++; |
909 | printf("=]" ); |
910 | } |
911 | length = *workp++; /* the length of char_range */ |
912 | for (i=0 ; i<length ; i++) |
913 | { |
914 | wchar_t range_start = *p++; |
915 | wchar_t range_end = *p++; |
916 | printf("%C-%C" , range_start, range_end); |
917 | } |
918 | length = *workp++; /* the length of char */ |
919 | for (i=0 ; i<length ; i++) |
920 | printf("%C" , *p++); |
921 | putchar (']'); |
922 | # else |
923 | register int c, last = -100; |
924 | register int in_range = 0; |
925 | |
926 | printf ("/charset [%s" , |
927 | (re_opcode_t) *(p - 1) == charset_not ? "^" : "" ); |
928 | |
929 | assert (p + *p < pend); |
930 | |
931 | for (c = 0; c < 256; c++) |
932 | if (c / 8 < *p |
933 | && (p[1 + (c/8)] & (1 << (c % 8)))) |
934 | { |
935 | /* Are we starting a range? */ |
936 | if (last + 1 == c && ! in_range) |
937 | { |
938 | putchar ('-'); |
939 | in_range = 1; |
940 | } |
941 | /* Have we broken a range? */ |
942 | else if (last + 1 != c && in_range) |
943 | { |
944 | putchar (last); |
945 | in_range = 0; |
946 | } |
947 | |
948 | if (! in_range) |
949 | putchar (c); |
950 | |
951 | last = c; |
952 | } |
953 | |
954 | if (in_range) |
955 | putchar (last); |
956 | |
957 | putchar (']'); |
958 | |
959 | p += 1 + *p; |
960 | # endif /* WCHAR */ |
961 | } |
962 | break; |
963 | |
964 | case begline: |
965 | printf ("/begline" ); |
966 | break; |
967 | |
968 | case endline: |
969 | printf ("/endline" ); |
970 | break; |
971 | |
972 | case on_failure_jump: |
973 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
974 | # ifdef _LIBC |
975 | printf ("/on_failure_jump to %td" , p + mcnt - start); |
976 | # else |
977 | printf ("/on_failure_jump to %ld" , (long int) (p + mcnt - start)); |
978 | # endif |
979 | break; |
980 | |
981 | case on_failure_keep_string_jump: |
982 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
983 | # ifdef _LIBC |
984 | printf ("/on_failure_keep_string_jump to %td" , p + mcnt - start); |
985 | # else |
986 | printf ("/on_failure_keep_string_jump to %ld" , |
987 | (long int) (p + mcnt - start)); |
988 | # endif |
989 | break; |
990 | |
991 | case dummy_failure_jump: |
992 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
993 | # ifdef _LIBC |
994 | printf ("/dummy_failure_jump to %td" , p + mcnt - start); |
995 | # else |
996 | printf ("/dummy_failure_jump to %ld" , (long int) (p + mcnt - start)); |
997 | # endif |
998 | break; |
999 | |
1000 | case push_dummy_failure: |
1001 | printf ("/push_dummy_failure" ); |
1002 | break; |
1003 | |
1004 | case maybe_pop_jump: |
1005 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
1006 | # ifdef _LIBC |
1007 | printf ("/maybe_pop_jump to %td" , p + mcnt - start); |
1008 | # else |
1009 | printf ("/maybe_pop_jump to %ld" , (long int) (p + mcnt - start)); |
1010 | # endif |
1011 | break; |
1012 | |
1013 | case pop_failure_jump: |
1014 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
1015 | # ifdef _LIBC |
1016 | printf ("/pop_failure_jump to %td" , p + mcnt - start); |
1017 | # else |
1018 | printf ("/pop_failure_jump to %ld" , (long int) (p + mcnt - start)); |
1019 | # endif |
1020 | break; |
1021 | |
1022 | case jump_past_alt: |
1023 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
1024 | # ifdef _LIBC |
1025 | printf ("/jump_past_alt to %td" , p + mcnt - start); |
1026 | # else |
1027 | printf ("/jump_past_alt to %ld" , (long int) (p + mcnt - start)); |
1028 | # endif |
1029 | break; |
1030 | |
1031 | case jump: |
1032 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
1033 | # ifdef _LIBC |
1034 | printf ("/jump to %td" , p + mcnt - start); |
1035 | # else |
1036 | printf ("/jump to %ld" , (long int) (p + mcnt - start)); |
1037 | # endif |
1038 | break; |
1039 | |
1040 | case succeed_n: |
1041 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
1042 | p1 = p + mcnt; |
1043 | PREFIX(extract_number_and_incr) (&mcnt2, &p); |
1044 | # ifdef _LIBC |
1045 | printf ("/succeed_n to %td, %d times" , p1 - start, mcnt2); |
1046 | # else |
1047 | printf ("/succeed_n to %ld, %d times" , |
1048 | (long int) (p1 - start), mcnt2); |
1049 | # endif |
1050 | break; |
1051 | |
1052 | case jump_n: |
1053 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
1054 | p1 = p + mcnt; |
1055 | PREFIX(extract_number_and_incr) (&mcnt2, &p); |
1056 | printf ("/jump_n to %d, %d times" , p1 - start, mcnt2); |
1057 | break; |
1058 | |
1059 | case set_number_at: |
1060 | PREFIX(extract_number_and_incr) (&mcnt, &p); |
1061 | p1 = p + mcnt; |
1062 | PREFIX(extract_number_and_incr) (&mcnt2, &p); |
1063 | # ifdef _LIBC |
1064 | printf ("/set_number_at location %td to %d" , p1 - start, mcnt2); |
1065 | # else |
1066 | printf ("/set_number_at location %ld to %d" , |
1067 | (long int) (p1 - start), mcnt2); |
1068 | # endif |
1069 | break; |
1070 | |
1071 | case wordbound: |
1072 | printf ("/wordbound" ); |
1073 | break; |
1074 | |
1075 | case notwordbound: |
1076 | printf ("/notwordbound" ); |
1077 | break; |
1078 | |
1079 | case wordbeg: |
1080 | printf ("/wordbeg" ); |
1081 | break; |
1082 | |
1083 | case wordend: |
1084 | printf ("/wordend" ); |
1085 | break; |
1086 | |
1087 | # ifdef emacs |
1088 | case before_dot: |
1089 | printf ("/before_dot" ); |
1090 | break; |
1091 | |
1092 | case at_dot: |
1093 | printf ("/at_dot" ); |
1094 | break; |
1095 | |
1096 | case after_dot: |
1097 | printf ("/after_dot" ); |
1098 | break; |
1099 | |
1100 | case syntaxspec: |
1101 | printf ("/syntaxspec" ); |
1102 | mcnt = *p++; |
1103 | printf ("/%d" , mcnt); |
1104 | break; |
1105 | |
1106 | case notsyntaxspec: |
1107 | printf ("/notsyntaxspec" ); |
1108 | mcnt = *p++; |
1109 | printf ("/%d" , mcnt); |
1110 | break; |
1111 | # endif /* emacs */ |
1112 | |
1113 | case wordchar: |
1114 | printf ("/wordchar" ); |
1115 | break; |
1116 | |
1117 | case notwordchar: |
1118 | printf ("/notwordchar" ); |
1119 | break; |
1120 | |
1121 | case begbuf: |
1122 | printf ("/begbuf" ); |
1123 | break; |
1124 | |
1125 | case endbuf: |
1126 | printf ("/endbuf" ); |
1127 | break; |
1128 | |
1129 | default: |
1130 | printf ("?%ld" , (long int) *(p-1)); |
1131 | } |
1132 | |
1133 | putchar ('\n'); |
1134 | } |
1135 | |
1136 | # ifdef _LIBC |
1137 | printf ("%td:\tend of pattern.\n" , p - start); |
1138 | # else |
1139 | printf ("%ld:\tend of pattern.\n" , (long int) (p - start)); |
1140 | # endif |
1141 | } |
1142 | |
1143 | |
1144 | void |
1145 | PREFIX(print_compiled_pattern) (struct re_pattern_buffer *bufp) |
1146 | { |
1147 | UCHAR_T *buffer = (UCHAR_T*) bufp->buffer; |
1148 | |
1149 | PREFIX(print_partial_compiled_pattern) (buffer, buffer |
1150 | + bufp->used / sizeof(UCHAR_T)); |
1151 | printf ("%ld bytes used/%ld bytes allocated.\n" , |
1152 | bufp->used, bufp->allocated); |
1153 | |
1154 | if (bufp->fastmap_accurate && bufp->fastmap) |
1155 | { |
1156 | printf ("fastmap: " ); |
1157 | print_fastmap (bufp->fastmap); |
1158 | } |
1159 | |
1160 | # ifdef _LIBC |
1161 | printf ("re_nsub: %Zd\t" , bufp->re_nsub); |
1162 | # else |
1163 | printf ("re_nsub: %ld\t" , (long int) bufp->re_nsub); |
1164 | # endif |
1165 | printf ("regs_alloc: %d\t" , bufp->regs_allocated); |
1166 | printf ("can_be_null: %d\t" , bufp->can_be_null); |
1167 | printf ("newline_anchor: %d\n" , bufp->newline_anchor); |
1168 | printf ("no_sub: %d\t" , bufp->no_sub); |
1169 | printf ("not_bol: %d\t" , bufp->not_bol); |
1170 | printf ("not_eol: %d\t" , bufp->not_eol); |
1171 | printf ("syntax: %lx\n" , bufp->syntax); |
1172 | /* Perhaps we should print the translate table? */ |
1173 | } |
1174 | |
1175 | |
1176 | void |
1177 | PREFIX(print_double_string) (const CHAR_T *where, const CHAR_T *string1, |
1178 | int size1, const CHAR_T *string2, int size2) |
1179 | { |
1180 | int this_char; |
1181 | |
1182 | if (where == NULL) |
1183 | printf ("(null)" ); |
1184 | else |
1185 | { |
1186 | int cnt; |
1187 | |
1188 | if (FIRST_STRING_P (where)) |
1189 | { |
1190 | for (this_char = where - string1; this_char < size1; this_char++) |
1191 | PUT_CHAR (string1[this_char]); |
1192 | |
1193 | where = string2; |
1194 | } |
1195 | |
1196 | cnt = 0; |
1197 | for (this_char = where - string2; this_char < size2; this_char++) |
1198 | { |
1199 | PUT_CHAR (string2[this_char]); |
1200 | if (++cnt > 100) |
1201 | { |
1202 | fputs ("..." , stdout); |
1203 | break; |
1204 | } |
1205 | } |
1206 | } |
1207 | } |
1208 | |
1209 | # ifndef DEFINED_ONCE |
1210 | void |
1211 | printchar (int c) |
1212 | { |
1213 | putc (c, stderr); |
1214 | } |
1215 | # endif |
1216 | |
1217 | # else /* not DEBUG */ |
1218 | |
1219 | # ifndef DEFINED_ONCE |
1220 | # undef assert |
1221 | # define assert(e) |
1222 | |
1223 | # define DEBUG_STATEMENT(e) |
1224 | # define DEBUG_PRINT1(x) |
1225 | # define DEBUG_PRINT2(x1, x2) |
1226 | # define DEBUG_PRINT3(x1, x2, x3) |
1227 | # define DEBUG_PRINT4(x1, x2, x3, x4) |
1228 | # endif /* not DEFINED_ONCE */ |
1229 | # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) |
1230 | # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) |
1231 | |
1232 | # endif /* not DEBUG */ |
1233 | |
1234 | |
1235 | |
1236 | # ifdef WCHAR |
1237 | /* This convert a multibyte string to a wide character string. |
1238 | And write their correspondances to offset_buffer(see below) |
1239 | and write whether each wchar_t is binary data to is_binary. |
1240 | This assume invalid multibyte sequences as binary data. |
1241 | We assume offset_buffer and is_binary is already allocated |
1242 | enough space. */ |
1243 | |
1244 | static size_t convert_mbs_to_wcs (CHAR_T *dest, const unsigned char* src, |
1245 | size_t len, int *offset_buffer, |
1246 | char *is_binary); |
1247 | static size_t |
1248 | convert_mbs_to_wcs (CHAR_T *dest, const unsigned char*src, size_t len, |
1249 | int *offset_buffer, char *is_binary) |
1250 | /* It hold correspondances between src(char string) and |
1251 | dest(wchar_t string) for optimization. |
1252 | e.g. src = "xxxyzz" |
1253 | dest = {'X', 'Y', 'Z'} |
1254 | (each "xxx", "y" and "zz" represent one multibyte character |
1255 | corresponding to 'X', 'Y' and 'Z'.) |
1256 | offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")} |
1257 | = {0, 3, 4, 6} |
1258 | */ |
1259 | { |
1260 | wchar_t *pdest = dest; |
1261 | const unsigned char *psrc = src; |
1262 | size_t wc_count = 0; |
1263 | |
1264 | mbstate_t mbs; |
1265 | int i, consumed; |
1266 | size_t mb_remain = len; |
1267 | size_t mb_count = 0; |
1268 | |
1269 | /* Initialize the conversion state. */ |
1270 | memset (&mbs, 0, sizeof (mbstate_t)); |
1271 | |
1272 | offset_buffer[0] = 0; |
1273 | for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed, |
1274 | psrc += consumed) |
1275 | { |
1276 | #ifdef _LIBC |
1277 | consumed = __mbrtowc (pdest, psrc, mb_remain, &mbs); |
1278 | #else |
1279 | consumed = mbrtowc (pdest, psrc, mb_remain, &mbs); |
1280 | #endif |
1281 | |
1282 | if (consumed <= 0) |
1283 | /* failed to convert. maybe src contains binary data. |
1284 | So we consume 1 byte manualy. */ |
1285 | { |
1286 | *pdest = *psrc; |
1287 | consumed = 1; |
1288 | is_binary[wc_count] = TRUE; |
1289 | } |
1290 | else |
1291 | is_binary[wc_count] = FALSE; |
1292 | /* In sjis encoding, we use yen sign as escape character in |
1293 | place of reverse solidus. So we convert 0x5c(yen sign in |
1294 | sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse |
1295 | solidus in UCS2). */ |
1296 | if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5) |
1297 | *pdest = (wchar_t) *psrc; |
1298 | |
1299 | offset_buffer[wc_count + 1] = mb_count += consumed; |
1300 | } |
1301 | |
1302 | /* Fill remain of the buffer with sentinel. */ |
1303 | for (i = wc_count + 1 ; i <= len ; i++) |
1304 | offset_buffer[i] = mb_count + 1; |
1305 | |
1306 | return wc_count; |
1307 | } |
1308 | |
1309 | # endif /* WCHAR */ |
1310 | |
1311 | #else /* not INSIDE_RECURSION */ |
1312 | |
1313 | /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can |
1314 | also be assigned to arbitrarily: each pattern buffer stores its own |
1315 | syntax, so it can be changed between regex compilations. */ |
1316 | /* This has no initializer because initialized variables in Emacs |
1317 | become read-only after dumping. */ |
1318 | reg_syntax_t re_syntax_options; |
1319 | |
1320 | |
1321 | /* Specify the precise syntax of regexps for compilation. This provides |
1322 | for compatibility for various utilities which historically have |
1323 | different, incompatible syntaxes. |
1324 | |
1325 | The argument SYNTAX is a bit mask comprised of the various bits |
1326 | defined in regex.h. We return the old syntax. */ |
1327 | |
1328 | reg_syntax_t |
1329 | re_set_syntax (reg_syntax_t syntax) |
1330 | { |
1331 | reg_syntax_t ret = re_syntax_options; |
1332 | |
1333 | re_syntax_options = syntax; |
1334 | # ifdef DEBUG |
1335 | if (syntax & RE_DEBUG) |
1336 | debug = 1; |
1337 | else if (debug) /* was on but now is not */ |
1338 | debug = 0; |
1339 | # endif /* DEBUG */ |
1340 | return ret; |
1341 | } |
1342 | # ifdef _LIBC |
1343 | weak_alias (__re_set_syntax, re_set_syntax) |
1344 | # endif |
1345 | |
1346 | /* This table gives an error message for each of the error codes listed |
1347 | in regex.h. Obviously the order here has to be same as there. |
1348 | POSIX doesn't require that we do anything for REG_NOERROR, |
1349 | but why not be nice? */ |
1350 | |
1351 | static const char *re_error_msgid[] = |
1352 | { |
1353 | gettext_noop ("Success" ), /* REG_NOERROR */ |
1354 | gettext_noop ("No match" ), /* REG_NOMATCH */ |
1355 | gettext_noop ("Invalid regular expression" ), /* REG_BADPAT */ |
1356 | gettext_noop ("Invalid collation character" ), /* REG_ECOLLATE */ |
1357 | gettext_noop ("Invalid character class name" ), /* REG_ECTYPE */ |
1358 | gettext_noop ("Trailing backslash" ), /* REG_EESCAPE */ |
1359 | gettext_noop ("Invalid back reference" ), /* REG_ESUBREG */ |
1360 | gettext_noop ("Unmatched [ or [^" ), /* REG_EBRACK */ |
1361 | gettext_noop ("Unmatched ( or \\(" ), /* REG_EPAREN */ |
1362 | gettext_noop ("Unmatched \\{" ), /* REG_EBRACE */ |
1363 | gettext_noop ("Invalid content of \\{\\}" ), /* REG_BADBR */ |
1364 | gettext_noop ("Invalid range end" ), /* REG_ERANGE */ |
1365 | gettext_noop ("Memory exhausted" ), /* REG_ESPACE */ |
1366 | gettext_noop ("Invalid preceding regular expression" ), /* REG_BADRPT */ |
1367 | gettext_noop ("Premature end of regular expression" ), /* REG_EEND */ |
1368 | gettext_noop ("Regular expression too big" ), /* REG_ESIZE */ |
1369 | gettext_noop ("Unmatched ) or \\)" ) /* REG_ERPAREN */ |
1370 | }; |
1371 | |
1372 | #endif /* INSIDE_RECURSION */ |
1373 | |
1374 | #ifndef DEFINED_ONCE |
1375 | /* Avoiding alloca during matching, to placate r_alloc. */ |
1376 | |
1377 | /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the |
1378 | searching and matching functions should not call alloca. On some |
1379 | systems, alloca is implemented in terms of malloc, and if we're |
1380 | using the relocating allocator routines, then malloc could cause a |
1381 | relocation, which might (if the strings being searched are in the |
1382 | ralloc heap) shift the data out from underneath the regexp |
1383 | routines. |
1384 | |
1385 | Here's another reason to avoid allocation: Emacs |
1386 | processes input from X in a signal handler; processing X input may |
1387 | call malloc; if input arrives while a matching routine is calling |
1388 | malloc, then we're scrod. But Emacs can't just block input while |
1389 | calling matching routines; then we don't notice interrupts when |
1390 | they come in. So, Emacs blocks input around all regexp calls |
1391 | except the matching calls, which it leaves unprotected, in the |
1392 | faith that they will not malloc. */ |
1393 | |
1394 | /* Normally, this is fine. */ |
1395 | # define MATCH_MAY_ALLOCATE |
1396 | |
1397 | /* When using GNU C, we are not REALLY using the C alloca, no matter |
1398 | what config.h may say. So don't take precautions for it. */ |
1399 | # ifdef __GNUC__ |
1400 | # undef C_ALLOCA |
1401 | # endif |
1402 | |
1403 | /* The match routines may not allocate if (1) they would do it with malloc |
1404 | and (2) it's not safe for them to use malloc. |
1405 | Note that if REL_ALLOC is defined, matching would not use malloc for the |
1406 | failure stack, but we would still use it for the register vectors; |
1407 | so REL_ALLOC should not affect this. */ |
1408 | # if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs |
1409 | # undef MATCH_MAY_ALLOCATE |
1410 | # endif |
1411 | #endif /* not DEFINED_ONCE */ |
1412 | |
1413 | #ifdef INSIDE_RECURSION |
1414 | /* Failure stack declarations and macros; both re_compile_fastmap and |
1415 | re_match_2 use a failure stack. These have to be macros because of |
1416 | REGEX_ALLOCATE_STACK. */ |
1417 | |
1418 | |
1419 | /* Number of failure points for which to initially allocate space |
1420 | when matching. If this number is exceeded, we allocate more |
1421 | space, so it is not a hard limit. */ |
1422 | # ifndef INIT_FAILURE_ALLOC |
1423 | # define INIT_FAILURE_ALLOC 5 |
1424 | # endif |
1425 | |
1426 | /* Roughly the maximum number of failure points on the stack. Would be |
1427 | exactly that if always used MAX_FAILURE_ITEMS items each time we failed. |
1428 | This is a variable only so users of regex can assign to it; we never |
1429 | change it ourselves. */ |
1430 | |
1431 | # ifdef INT_IS_16BIT |
1432 | |
1433 | # ifndef DEFINED_ONCE |
1434 | # if defined MATCH_MAY_ALLOCATE |
1435 | /* 4400 was enough to cause a crash on Alpha OSF/1, |
1436 | whose default stack limit is 2mb. */ |
1437 | long int re_max_failures = 4000; |
1438 | # else |
1439 | long int re_max_failures = 2000; |
1440 | # endif |
1441 | # endif |
1442 | |
1443 | union PREFIX(fail_stack_elt) |
1444 | { |
1445 | UCHAR_T *pointer; |
1446 | long int integer; |
1447 | }; |
1448 | |
1449 | typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); |
1450 | |
1451 | typedef struct |
1452 | { |
1453 | PREFIX(fail_stack_elt_t) *stack; |
1454 | unsigned long int size; |
1455 | unsigned long int avail; /* Offset of next open position. */ |
1456 | } PREFIX(fail_stack_type); |
1457 | |
1458 | # else /* not INT_IS_16BIT */ |
1459 | |
1460 | # ifndef DEFINED_ONCE |
1461 | # if defined MATCH_MAY_ALLOCATE |
1462 | /* 4400 was enough to cause a crash on Alpha OSF/1, |
1463 | whose default stack limit is 2mb. */ |
1464 | int re_max_failures = 4000; |
1465 | # else |
1466 | int re_max_failures = 2000; |
1467 | # endif |
1468 | # endif |
1469 | |
1470 | union PREFIX(fail_stack_elt) |
1471 | { |
1472 | UCHAR_T *pointer; |
1473 | int integer; |
1474 | }; |
1475 | |
1476 | typedef union PREFIX(fail_stack_elt) PREFIX(fail_stack_elt_t); |
1477 | |
1478 | typedef struct |
1479 | { |
1480 | PREFIX(fail_stack_elt_t) *stack; |
1481 | unsigned size; |
1482 | unsigned avail; /* Offset of next open position. */ |
1483 | } PREFIX(fail_stack_type); |
1484 | |
1485 | # endif /* INT_IS_16BIT */ |
1486 | |
1487 | # ifndef DEFINED_ONCE |
1488 | # define FAIL_STACK_EMPTY() (fail_stack.avail == 0) |
1489 | # define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0) |
1490 | # define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size) |
1491 | # endif |
1492 | |
1493 | |
1494 | /* Define macros to initialize and free the failure stack. |
1495 | Do `return -2' if the alloc fails. */ |
1496 | |
1497 | # ifdef MATCH_MAY_ALLOCATE |
1498 | # define INIT_FAIL_STACK() \ |
1499 | do { \ |
1500 | fail_stack.stack = (PREFIX(fail_stack_elt_t) *) \ |
1501 | REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (PREFIX(fail_stack_elt_t))); \ |
1502 | \ |
1503 | if (fail_stack.stack == NULL) \ |
1504 | return -2; \ |
1505 | \ |
1506 | fail_stack.size = INIT_FAILURE_ALLOC; \ |
1507 | fail_stack.avail = 0; \ |
1508 | } while (0) |
1509 | |
1510 | # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack) |
1511 | # else |
1512 | # define INIT_FAIL_STACK() \ |
1513 | do { \ |
1514 | fail_stack.avail = 0; \ |
1515 | } while (0) |
1516 | |
1517 | # define RESET_FAIL_STACK() |
1518 | # endif |
1519 | |
1520 | |
1521 | /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items. |
1522 | |
1523 | Return 1 if succeeds, and 0 if either ran out of memory |
1524 | allocating space for it or it was already too large. |
1525 | |
1526 | REGEX_REALLOCATE_STACK requires `destination' be declared. */ |
1527 | |
1528 | # define DOUBLE_FAIL_STACK(fail_stack) \ |
1529 | ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \ |
1530 | ? 0 \ |
1531 | : ((fail_stack).stack = (PREFIX(fail_stack_elt_t) *) \ |
1532 | REGEX_REALLOCATE_STACK ((fail_stack).stack, \ |
1533 | (fail_stack).size * sizeof (PREFIX(fail_stack_elt_t)), \ |
1534 | ((fail_stack).size << 1) * sizeof (PREFIX(fail_stack_elt_t))),\ |
1535 | \ |
1536 | (fail_stack).stack == NULL \ |
1537 | ? 0 \ |
1538 | : ((fail_stack).size <<= 1, \ |
1539 | 1))) |
1540 | |
1541 | |
1542 | /* Push pointer POINTER on FAIL_STACK. |
1543 | Return 1 if was able to do so and 0 if ran out of memory allocating |
1544 | space to do so. */ |
1545 | # define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \ |
1546 | ((FAIL_STACK_FULL () \ |
1547 | && !DOUBLE_FAIL_STACK (FAIL_STACK)) \ |
1548 | ? 0 \ |
1549 | : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \ |
1550 | 1)) |
1551 | |
1552 | /* Push a pointer value onto the failure stack. |
1553 | Assumes the variable `fail_stack'. Probably should only |
1554 | be called from within `PUSH_FAILURE_POINT'. */ |
1555 | # define PUSH_FAILURE_POINTER(item) \ |
1556 | fail_stack.stack[fail_stack.avail++].pointer = (UCHAR_T *) (item) |
1557 | |
1558 | /* This pushes an integer-valued item onto the failure stack. |
1559 | Assumes the variable `fail_stack'. Probably should only |
1560 | be called from within `PUSH_FAILURE_POINT'. */ |
1561 | # define PUSH_FAILURE_INT(item) \ |
1562 | fail_stack.stack[fail_stack.avail++].integer = (item) |
1563 | |
1564 | /* Push a fail_stack_elt_t value onto the failure stack. |
1565 | Assumes the variable `fail_stack'. Probably should only |
1566 | be called from within `PUSH_FAILURE_POINT'. */ |
1567 | # define PUSH_FAILURE_ELT(item) \ |
1568 | fail_stack.stack[fail_stack.avail++] = (item) |
1569 | |
1570 | /* These three POP... operations complement the three PUSH... operations. |
1571 | All assume that `fail_stack' is nonempty. */ |
1572 | # define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer |
1573 | # define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer |
1574 | # define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail] |
1575 | |
1576 | /* Used to omit pushing failure point id's when we're not debugging. */ |
1577 | # ifdef DEBUG |
1578 | # define DEBUG_PUSH PUSH_FAILURE_INT |
1579 | # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT () |
1580 | # else |
1581 | # define DEBUG_PUSH(item) |
1582 | # define DEBUG_POP(item_addr) |
1583 | # endif |
1584 | |
1585 | |
1586 | /* Push the information about the state we will need |
1587 | if we ever fail back to it. |
1588 | |
1589 | Requires variables fail_stack, regstart, regend, reg_info, and |
1590 | num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination' |
1591 | be declared. |
1592 | |
1593 | Does `return FAILURE_CODE' if runs out of memory. */ |
1594 | |
1595 | # define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \ |
1596 | do { \ |
1597 | char *destination; \ |
1598 | /* Must be int, so when we don't save any registers, the arithmetic \ |
1599 | of 0 + -1 isn't done as unsigned. */ \ |
1600 | /* Can't be int, since there is not a shred of a guarantee that int \ |
1601 | is wide enough to hold a value of something to which pointer can \ |
1602 | be assigned */ \ |
1603 | active_reg_t this_reg; \ |
1604 | \ |
1605 | DEBUG_STATEMENT (failure_id++); \ |
1606 | DEBUG_STATEMENT (nfailure_points_pushed++); \ |
1607 | DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \ |
1608 | DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\ |
1609 | DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\ |
1610 | \ |
1611 | DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \ |
1612 | DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \ |
1613 | \ |
1614 | /* Ensure we have enough space allocated for what we will push. */ \ |
1615 | while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \ |
1616 | { \ |
1617 | if (!DOUBLE_FAIL_STACK (fail_stack)) \ |
1618 | return failure_code; \ |
1619 | \ |
1620 | DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \ |
1621 | (fail_stack).size); \ |
1622 | DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\ |
1623 | } \ |
1624 | \ |
1625 | /* Push the info, starting with the registers. */ \ |
1626 | DEBUG_PRINT1 ("\n"); \ |
1627 | \ |
1628 | if (1) \ |
1629 | for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \ |
1630 | this_reg++) \ |
1631 | { \ |
1632 | DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \ |
1633 | DEBUG_STATEMENT (num_regs_pushed++); \ |
1634 | \ |
1635 | DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ |
1636 | PUSH_FAILURE_POINTER (regstart[this_reg]); \ |
1637 | \ |
1638 | DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ |
1639 | PUSH_FAILURE_POINTER (regend[this_reg]); \ |
1640 | \ |
1641 | DEBUG_PRINT2 (" info: %p\n ", \ |
1642 | reg_info[this_reg].word.pointer); \ |
1643 | DEBUG_PRINT2 (" match_null=%d", \ |
1644 | REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \ |
1645 | DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \ |
1646 | DEBUG_PRINT2 (" matched_something=%d", \ |
1647 | MATCHED_SOMETHING (reg_info[this_reg])); \ |
1648 | DEBUG_PRINT2 (" ever_matched=%d", \ |
1649 | EVER_MATCHED_SOMETHING (reg_info[this_reg])); \ |
1650 | DEBUG_PRINT1 ("\n"); \ |
1651 | PUSH_FAILURE_ELT (reg_info[this_reg].word); \ |
1652 | } \ |
1653 | \ |
1654 | DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\ |
1655 | PUSH_FAILURE_INT (lowest_active_reg); \ |
1656 | \ |
1657 | DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\ |
1658 | PUSH_FAILURE_INT (highest_active_reg); \ |
1659 | \ |
1660 | DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \ |
1661 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \ |
1662 | PUSH_FAILURE_POINTER (pattern_place); \ |
1663 | \ |
1664 | DEBUG_PRINT2 (" Pushing string %p: `", string_place); \ |
1665 | DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \ |
1666 | size2); \ |
1667 | DEBUG_PRINT1 ("'\n"); \ |
1668 | PUSH_FAILURE_POINTER (string_place); \ |
1669 | \ |
1670 | DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \ |
1671 | DEBUG_PUSH (failure_id); \ |
1672 | } while (0) |
1673 | |
1674 | # ifndef DEFINED_ONCE |
1675 | /* This is the number of items that are pushed and popped on the stack |
1676 | for each register. */ |
1677 | # define NUM_REG_ITEMS 3 |
1678 | |
1679 | /* Individual items aside from the registers. */ |
1680 | # ifdef DEBUG |
1681 | # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */ |
1682 | # else |
1683 | # define NUM_NONREG_ITEMS 4 |
1684 | # endif |
1685 | |
1686 | /* We push at most this many items on the stack. */ |
1687 | /* We used to use (num_regs - 1), which is the number of registers |
1688 | this regexp will save; but that was changed to 5 |
1689 | to avoid stack overflow for a regexp with lots of parens. */ |
1690 | # define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS) |
1691 | |
1692 | /* We actually push this many items. */ |
1693 | # define NUM_FAILURE_ITEMS \ |
1694 | (((0 \ |
1695 | ? 0 : highest_active_reg - lowest_active_reg + 1) \ |
1696 | * NUM_REG_ITEMS) \ |
1697 | + NUM_NONREG_ITEMS) |
1698 | |
1699 | /* How many items can still be added to the stack without overflowing it. */ |
1700 | # define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail) |
1701 | # endif /* not DEFINED_ONCE */ |
1702 | |
1703 | |
1704 | /* Pops what PUSH_FAIL_STACK pushes. |
1705 | |
1706 | We restore into the parameters, all of which should be lvalues: |
1707 | STR -- the saved data position. |
1708 | PAT -- the saved pattern position. |
1709 | LOW_REG, HIGH_REG -- the highest and lowest active registers. |
1710 | REGSTART, REGEND -- arrays of string positions. |
1711 | REG_INFO -- array of information about each subexpression. |
1712 | |
1713 | Also assumes the variables `fail_stack' and (if debugging), `bufp', |
1714 | `pend', `string1', `size1', `string2', and `size2'. */ |
1715 | # define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\ |
1716 | { \ |
1717 | DEBUG_STATEMENT (unsigned failure_id;) \ |
1718 | active_reg_t this_reg; \ |
1719 | const UCHAR_T *string_temp; \ |
1720 | \ |
1721 | assert (!FAIL_STACK_EMPTY ()); \ |
1722 | \ |
1723 | /* Remove failure points and point to how many regs pushed. */ \ |
1724 | DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \ |
1725 | DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \ |
1726 | DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \ |
1727 | \ |
1728 | assert (fail_stack.avail >= NUM_NONREG_ITEMS); \ |
1729 | \ |
1730 | DEBUG_POP (&failure_id); \ |
1731 | DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \ |
1732 | \ |
1733 | /* If the saved string location is NULL, it came from an \ |
1734 | on_failure_keep_string_jump opcode, and we want to throw away the \ |
1735 | saved NULL, thus retaining our current position in the string. */ \ |
1736 | string_temp = POP_FAILURE_POINTER (); \ |
1737 | if (string_temp != NULL) \ |
1738 | str = (const CHAR_T *) string_temp; \ |
1739 | \ |
1740 | DEBUG_PRINT2 (" Popping string %p: `", str); \ |
1741 | DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \ |
1742 | DEBUG_PRINT1 ("'\n"); \ |
1743 | \ |
1744 | pat = (UCHAR_T *) POP_FAILURE_POINTER (); \ |
1745 | DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \ |
1746 | DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \ |
1747 | \ |
1748 | /* Restore register info. */ \ |
1749 | high_reg = (active_reg_t) POP_FAILURE_INT (); \ |
1750 | DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \ |
1751 | \ |
1752 | low_reg = (active_reg_t) POP_FAILURE_INT (); \ |
1753 | DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \ |
1754 | \ |
1755 | if (1) \ |
1756 | for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \ |
1757 | { \ |
1758 | DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \ |
1759 | \ |
1760 | reg_info[this_reg].word = POP_FAILURE_ELT (); \ |
1761 | DEBUG_PRINT2 (" info: %p\n", \ |
1762 | reg_info[this_reg].word.pointer); \ |
1763 | \ |
1764 | regend[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ |
1765 | DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \ |
1766 | \ |
1767 | regstart[this_reg] = (const CHAR_T *) POP_FAILURE_POINTER (); \ |
1768 | DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \ |
1769 | } \ |
1770 | else \ |
1771 | { \ |
1772 | for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \ |
1773 | { \ |
1774 | reg_info[this_reg].word.integer = 0; \ |
1775 | regend[this_reg] = 0; \ |
1776 | regstart[this_reg] = 0; \ |
1777 | } \ |
1778 | highest_active_reg = high_reg; \ |
1779 | } \ |
1780 | \ |
1781 | set_regs_matched_done = 0; \ |
1782 | DEBUG_STATEMENT (nfailure_points_popped++); \ |
1783 | } /* POP_FAILURE_POINT */ |
1784 | |
1785 | /* Structure for per-register (a.k.a. per-group) information. |
1786 | Other register information, such as the |
1787 | starting and ending positions (which are addresses), and the list of |
1788 | inner groups (which is a bits list) are maintained in separate |
1789 | variables. |
1790 | |
1791 | We are making a (strictly speaking) nonportable assumption here: that |
1792 | the compiler will pack our bit fields into something that fits into |
1793 | the type of `word', i.e., is something that fits into one item on the |
1794 | failure stack. */ |
1795 | |
1796 | |
1797 | /* Declarations and macros for re_match_2. */ |
1798 | |
1799 | typedef union |
1800 | { |
1801 | PREFIX(fail_stack_elt_t) word; |
1802 | struct |
1803 | { |
1804 | /* This field is one if this group can match the empty string, |
1805 | zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */ |
1806 | # define MATCH_NULL_UNSET_VALUE 3 |
1807 | unsigned match_null_string_p : 2; |
1808 | unsigned is_active : 1; |
1809 | unsigned matched_something : 1; |
1810 | unsigned ever_matched_something : 1; |
1811 | } bits; |
1812 | } PREFIX(register_info_type); |
1813 | |
1814 | # ifndef DEFINED_ONCE |
1815 | # define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p) |
1816 | # define IS_ACTIVE(R) ((R).bits.is_active) |
1817 | # define MATCHED_SOMETHING(R) ((R).bits.matched_something) |
1818 | # define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something) |
1819 | |
1820 | |
1821 | /* Call this when have matched a real character; it sets `matched' flags |
1822 | for the subexpressions which we are currently inside. Also records |
1823 | that those subexprs have matched. */ |
1824 | # define SET_REGS_MATCHED() \ |
1825 | do \ |
1826 | { \ |
1827 | if (!set_regs_matched_done) \ |
1828 | { \ |
1829 | active_reg_t r; \ |
1830 | set_regs_matched_done = 1; \ |
1831 | for (r = lowest_active_reg; r <= highest_active_reg; r++) \ |
1832 | { \ |
1833 | MATCHED_SOMETHING (reg_info[r]) \ |
1834 | = EVER_MATCHED_SOMETHING (reg_info[r]) \ |
1835 | = 1; \ |
1836 | } \ |
1837 | } \ |
1838 | } \ |
1839 | while (0) |
1840 | # endif /* not DEFINED_ONCE */ |
1841 | |
1842 | /* Registers are set to a sentinel when they haven't yet matched. */ |
1843 | static CHAR_T PREFIX(reg_unset_dummy); |
1844 | # define REG_UNSET_VALUE (&PREFIX(reg_unset_dummy)) |
1845 | # define REG_UNSET(e) ((e) == REG_UNSET_VALUE) |
1846 | |
1847 | /* Subroutine declarations and macros for regex_compile. */ |
1848 | static void PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg); |
1849 | static void PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, |
1850 | int arg1, int arg2); |
1851 | static void PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, |
1852 | int arg, UCHAR_T *end); |
1853 | static void PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, |
1854 | int arg1, int arg2, UCHAR_T *end); |
1855 | static boolean PREFIX(at_begline_loc_p) (const CHAR_T *pattern, |
1856 | const CHAR_T *p, |
1857 | reg_syntax_t syntax); |
1858 | static boolean PREFIX(at_endline_loc_p) (const CHAR_T *p, |
1859 | const CHAR_T *pend, |
1860 | reg_syntax_t syntax); |
1861 | # ifdef WCHAR |
1862 | static reg_errcode_t wcs_compile_range (CHAR_T range_start, |
1863 | const CHAR_T **p_ptr, |
1864 | const CHAR_T *pend, |
1865 | char *translate, |
1866 | reg_syntax_t syntax, |
1867 | UCHAR_T *b, |
1868 | CHAR_T *char_set); |
1869 | static void insert_space (int num, CHAR_T *loc, CHAR_T *end); |
1870 | # else /* BYTE */ |
1871 | static reg_errcode_t byte_compile_range (unsigned int range_start, |
1872 | const char **p_ptr, |
1873 | const char *pend, |
1874 | char *translate, |
1875 | reg_syntax_t syntax, |
1876 | unsigned char *b); |
1877 | # endif /* WCHAR */ |
1878 | |
1879 | /* Fetch the next character in the uncompiled pattern---translating it |
1880 | if necessary. Also cast from a signed character in the constant |
1881 | string passed to us by the user to an unsigned char that we can use |
1882 | as an array index (in, e.g., `translate'). */ |
1883 | /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, |
1884 | because it is impossible to allocate 4GB array for some encodings |
1885 | which have 4 byte character_set like UCS4. */ |
1886 | # ifndef PATFETCH |
1887 | # ifdef WCHAR |
1888 | # define PATFETCH(c) \ |
1889 | do {if (p == pend) return REG_EEND; \ |
1890 | c = (UCHAR_T) *p++; \ |
1891 | if (translate && (c <= 0xff)) c = (UCHAR_T) translate[c]; \ |
1892 | } while (0) |
1893 | # else /* BYTE */ |
1894 | # define PATFETCH(c) \ |
1895 | do {if (p == pend) return REG_EEND; \ |
1896 | c = (unsigned char) *p++; \ |
1897 | if (translate) c = (unsigned char) translate[c]; \ |
1898 | } while (0) |
1899 | # endif /* WCHAR */ |
1900 | # endif |
1901 | |
1902 | /* Fetch the next character in the uncompiled pattern, with no |
1903 | translation. */ |
1904 | # define PATFETCH_RAW(c) \ |
1905 | do {if (p == pend) return REG_EEND; \ |
1906 | c = (UCHAR_T) *p++; \ |
1907 | } while (0) |
1908 | |
1909 | /* Go backwards one character in the pattern. */ |
1910 | # define PATUNFETCH p-- |
1911 | |
1912 | |
1913 | /* If `translate' is non-null, return translate[D], else just D. We |
1914 | cast the subscript to translate because some data is declared as |
1915 | `char *', to avoid warnings when a string constant is passed. But |
1916 | when we use a character as a subscript we must make it unsigned. */ |
1917 | /* ifdef MBS_SUPPORT, we translate only if character <= 0xff, |
1918 | because it is impossible to allocate 4GB array for some encodings |
1919 | which have 4 byte character_set like UCS4. */ |
1920 | |
1921 | # ifndef TRANSLATE |
1922 | # ifdef WCHAR |
1923 | # define TRANSLATE(d) \ |
1924 | ((translate && ((UCHAR_T) (d)) <= 0xff) \ |
1925 | ? (char) translate[(unsigned char) (d)] : (d)) |
1926 | # else /* BYTE */ |
1927 | # define TRANSLATE(d) \ |
1928 | (translate ? (char) translate[(unsigned char) (d)] : (char) (d)) |
1929 | # endif /* WCHAR */ |
1930 | # endif |
1931 | |
1932 | |
1933 | /* Macros for outputting the compiled pattern into `buffer'. */ |
1934 | |
1935 | /* If the buffer isn't allocated when it comes in, use this. */ |
1936 | # define INIT_BUF_SIZE (32 * sizeof(UCHAR_T)) |
1937 | |
1938 | /* Make sure we have at least N more bytes of space in buffer. */ |
1939 | # ifdef WCHAR |
1940 | # define GET_BUFFER_SPACE(n) \ |
1941 | while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \ |
1942 | + (n)*sizeof(CHAR_T)) > bufp->allocated) \ |
1943 | EXTEND_BUFFER () |
1944 | # else /* BYTE */ |
1945 | # define GET_BUFFER_SPACE(n) \ |
1946 | while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \ |
1947 | EXTEND_BUFFER () |
1948 | # endif /* WCHAR */ |
1949 | |
1950 | /* Make sure we have one more byte of buffer space and then add C to it. */ |
1951 | # define BUF_PUSH(c) \ |
1952 | do { \ |
1953 | GET_BUFFER_SPACE (1); \ |
1954 | *b++ = (UCHAR_T) (c); \ |
1955 | } while (0) |
1956 | |
1957 | |
1958 | /* Ensure we have two more bytes of buffer space and then append C1 and C2. */ |
1959 | # define BUF_PUSH_2(c1, c2) \ |
1960 | do { \ |
1961 | GET_BUFFER_SPACE (2); \ |
1962 | *b++ = (UCHAR_T) (c1); \ |
1963 | *b++ = (UCHAR_T) (c2); \ |
1964 | } while (0) |
1965 | |
1966 | |
1967 | /* As with BUF_PUSH_2, except for three bytes. */ |
1968 | # define BUF_PUSH_3(c1, c2, c3) \ |
1969 | do { \ |
1970 | GET_BUFFER_SPACE (3); \ |
1971 | *b++ = (UCHAR_T) (c1); \ |
1972 | *b++ = (UCHAR_T) (c2); \ |
1973 | *b++ = (UCHAR_T) (c3); \ |
1974 | } while (0) |
1975 | |
1976 | /* Store a jump with opcode OP at LOC to location TO. We store a |
1977 | relative address offset by the three bytes the jump itself occupies. */ |
1978 | # define STORE_JUMP(op, loc, to) \ |
1979 | PREFIX(store_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE))) |
1980 | |
1981 | /* Likewise, for a two-argument jump. */ |
1982 | # define STORE_JUMP2(op, loc, to, arg) \ |
1983 | PREFIX(store_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg) |
1984 | |
1985 | /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */ |
1986 | # define INSERT_JUMP(op, loc, to) \ |
1987 | PREFIX(insert_op1) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b) |
1988 | |
1989 | /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */ |
1990 | # define INSERT_JUMP2(op, loc, to, arg) \ |
1991 | PREFIX(insert_op2) (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\ |
1992 | arg, b) |
1993 | |
1994 | /* This is not an arbitrary limit: the arguments which represent offsets |
1995 | into the pattern are two bytes long. So if 2^16 bytes turns out to |
1996 | be too small, many things would have to change. */ |
1997 | /* Any other compiler which, like MSC, has allocation limit below 2^16 |
1998 | bytes will have to use approach similar to what was done below for |
1999 | MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up |
2000 | reallocating to 0 bytes. Such thing is not going to work too well. |
2001 | You have been warned!! */ |
2002 | # ifndef DEFINED_ONCE |
2003 | # if defined _MSC_VER && !defined WIN32 |
2004 | /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes. |
2005 | The REALLOC define eliminates a flurry of conversion warnings, |
2006 | but is not required. */ |
2007 | # define MAX_BUF_SIZE 65500L |
2008 | # define REALLOC(p,s) realloc ((p), (size_t) (s)) |
2009 | # else |
2010 | # define MAX_BUF_SIZE (1L << 16) |
2011 | # define REALLOC(p,s) realloc ((p), (s)) |
2012 | # endif |
2013 | |
2014 | /* Extend the buffer by twice its current size via realloc and |
2015 | reset the pointers that pointed into the old block to point to the |
2016 | correct places in the new one. If extending the buffer results in it |
2017 | being larger than MAX_BUF_SIZE, then flag memory exhausted. */ |
2018 | # if __BOUNDED_POINTERS__ |
2019 | # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated) |
2020 | # define MOVE_BUFFER_POINTER(P) \ |
2021 | (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr) |
2022 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
2023 | else \ |
2024 | { \ |
2025 | SET_HIGH_BOUND (b); \ |
2026 | SET_HIGH_BOUND (begalt); \ |
2027 | if (fixup_alt_jump) \ |
2028 | SET_HIGH_BOUND (fixup_alt_jump); \ |
2029 | if (laststart) \ |
2030 | SET_HIGH_BOUND (laststart); \ |
2031 | if (pending_exact) \ |
2032 | SET_HIGH_BOUND (pending_exact); \ |
2033 | } |
2034 | # else |
2035 | # define MOVE_BUFFER_POINTER(P) (P) += incr |
2036 | # define ELSE_EXTEND_BUFFER_HIGH_BOUND |
2037 | # endif |
2038 | # endif /* not DEFINED_ONCE */ |
2039 | |
2040 | # ifdef WCHAR |
2041 | # define EXTEND_BUFFER() \ |
2042 | do { \ |
2043 | UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ |
2044 | int wchar_count; \ |
2045 | if (bufp->allocated + sizeof(UCHAR_T) > MAX_BUF_SIZE) \ |
2046 | return REG_ESIZE; \ |
2047 | bufp->allocated <<= 1; \ |
2048 | if (bufp->allocated > MAX_BUF_SIZE) \ |
2049 | bufp->allocated = MAX_BUF_SIZE; \ |
2050 | /* How many characters the new buffer can have? */ \ |
2051 | wchar_count = bufp->allocated / sizeof(UCHAR_T); \ |
2052 | if (wchar_count == 0) wchar_count = 1; \ |
2053 | /* Truncate the buffer to CHAR_T align. */ \ |
2054 | bufp->allocated = wchar_count * sizeof(UCHAR_T); \ |
2055 | RETALLOC (COMPILED_BUFFER_VAR, wchar_count, UCHAR_T); \ |
2056 | bufp->buffer = (char*)COMPILED_BUFFER_VAR; \ |
2057 | if (COMPILED_BUFFER_VAR == NULL) \ |
2058 | return REG_ESPACE; \ |
2059 | /* If the buffer moved, move all the pointers into it. */ \ |
2060 | if (old_buffer != COMPILED_BUFFER_VAR) \ |
2061 | { \ |
2062 | PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \ |
2063 | MOVE_BUFFER_POINTER (b); \ |
2064 | MOVE_BUFFER_POINTER (begalt); \ |
2065 | if (fixup_alt_jump) \ |
2066 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
2067 | if (laststart) \ |
2068 | MOVE_BUFFER_POINTER (laststart); \ |
2069 | if (pending_exact) \ |
2070 | MOVE_BUFFER_POINTER (pending_exact); \ |
2071 | } \ |
2072 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
2073 | } while (0) |
2074 | # else /* BYTE */ |
2075 | # define EXTEND_BUFFER() \ |
2076 | do { \ |
2077 | UCHAR_T *old_buffer = COMPILED_BUFFER_VAR; \ |
2078 | if (bufp->allocated == MAX_BUF_SIZE) \ |
2079 | return REG_ESIZE; \ |
2080 | bufp->allocated <<= 1; \ |
2081 | if (bufp->allocated > MAX_BUF_SIZE) \ |
2082 | bufp->allocated = MAX_BUF_SIZE; \ |
2083 | bufp->buffer = (UCHAR_T *) REALLOC (COMPILED_BUFFER_VAR, \ |
2084 | bufp->allocated); \ |
2085 | if (COMPILED_BUFFER_VAR == NULL) \ |
2086 | return REG_ESPACE; \ |
2087 | /* If the buffer moved, move all the pointers into it. */ \ |
2088 | if (old_buffer != COMPILED_BUFFER_VAR) \ |
2089 | { \ |
2090 | PTR_INT_TYPE incr = COMPILED_BUFFER_VAR - old_buffer; \ |
2091 | MOVE_BUFFER_POINTER (b); \ |
2092 | MOVE_BUFFER_POINTER (begalt); \ |
2093 | if (fixup_alt_jump) \ |
2094 | MOVE_BUFFER_POINTER (fixup_alt_jump); \ |
2095 | if (laststart) \ |
2096 | MOVE_BUFFER_POINTER (laststart); \ |
2097 | if (pending_exact) \ |
2098 | MOVE_BUFFER_POINTER (pending_exact); \ |
2099 | } \ |
2100 | ELSE_EXTEND_BUFFER_HIGH_BOUND \ |
2101 | } while (0) |
2102 | # endif /* WCHAR */ |
2103 | |
2104 | # ifndef DEFINED_ONCE |
2105 | /* Since we have one byte reserved for the register number argument to |
2106 | {start,stop}_memory, the maximum number of groups we can report |
2107 | things about is what fits in that byte. */ |
2108 | # define MAX_REGNUM 255 |
2109 | |
2110 | /* But patterns can have more than `MAX_REGNUM' registers. We just |
2111 | ignore the excess. */ |
2112 | typedef unsigned regnum_t; |
2113 | |
2114 | |
2115 | /* Macros for the compile stack. */ |
2116 | |
2117 | /* Since offsets can go either forwards or backwards, this type needs to |
2118 | be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */ |
2119 | /* int may be not enough when sizeof(int) == 2. */ |
2120 | typedef long pattern_offset_t; |
2121 | |
2122 | typedef struct |
2123 | { |
2124 | pattern_offset_t begalt_offset; |
2125 | pattern_offset_t fixup_alt_jump; |
2126 | pattern_offset_t inner_group_offset; |
2127 | pattern_offset_t laststart_offset; |
2128 | regnum_t regnum; |
2129 | } compile_stack_elt_t; |
2130 | |
2131 | |
2132 | typedef struct |
2133 | { |
2134 | compile_stack_elt_t *stack; |
2135 | unsigned size; |
2136 | unsigned avail; /* Offset of next open position. */ |
2137 | } compile_stack_type; |
2138 | |
2139 | |
2140 | # define INIT_COMPILE_STACK_SIZE 32 |
2141 | |
2142 | # define COMPILE_STACK_EMPTY (compile_stack.avail == 0) |
2143 | # define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size) |
2144 | |
2145 | /* The next available element. */ |
2146 | # define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail]) |
2147 | |
2148 | # endif /* not DEFINED_ONCE */ |
2149 | |
2150 | /* Set the bit for character C in a list. */ |
2151 | # ifndef DEFINED_ONCE |
2152 | # define SET_LIST_BIT(c) \ |
2153 | (b[((unsigned char) (c)) / BYTEWIDTH] \ |
2154 | |= 1 << (((unsigned char) c) % BYTEWIDTH)) |
2155 | # endif /* DEFINED_ONCE */ |
2156 | |
2157 | /* Get the next unsigned number in the uncompiled pattern. */ |
2158 | # define GET_UNSIGNED_NUMBER(num) \ |
2159 | { \ |
2160 | while (p != pend) \ |
2161 | { \ |
2162 | PATFETCH (c); \ |
2163 | if (c < '0' || c > '9') \ |
2164 | break; \ |
2165 | if (num <= RE_DUP_MAX) \ |
2166 | { \ |
2167 | if (num < 0) \ |
2168 | num = 0; \ |
2169 | num = num * 10 + c - '0'; \ |
2170 | } \ |
2171 | } \ |
2172 | } |
2173 | |
2174 | # ifndef DEFINED_ONCE |
2175 | # if defined _LIBC || WIDE_CHAR_SUPPORT |
2176 | /* The GNU C library provides support for user-defined character classes |
2177 | and the functions from ISO C amendement 1. */ |
2178 | # ifdef CHARCLASS_NAME_MAX |
2179 | # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX |
2180 | # else |
2181 | /* This shouldn't happen but some implementation might still have this |
2182 | problem. Use a reasonable default value. */ |
2183 | # define CHAR_CLASS_MAX_LENGTH 256 |
2184 | # endif |
2185 | |
2186 | # ifdef _LIBC |
2187 | # define IS_CHAR_CLASS(string) __wctype (string) |
2188 | # else |
2189 | # define IS_CHAR_CLASS(string) wctype (string) |
2190 | # endif |
2191 | # else |
2192 | # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */ |
2193 | |
2194 | # define IS_CHAR_CLASS(string) \ |
2195 | (STREQ (string, "alpha") || STREQ (string, "upper") \ |
2196 | || STREQ (string, "lower") || STREQ (string, "digit") \ |
2197 | || STREQ (string, "alnum") || STREQ (string, "xdigit") \ |
2198 | || STREQ (string, "space") || STREQ (string, "print") \ |
2199 | || STREQ (string, "punct") || STREQ (string, "graph") \ |
2200 | || STREQ (string, "cntrl") || STREQ (string, "blank")) |
2201 | # endif |
2202 | # endif /* DEFINED_ONCE */ |
2203 | |
2204 | # ifndef MATCH_MAY_ALLOCATE |
2205 | |
2206 | /* If we cannot allocate large objects within re_match_2_internal, |
2207 | we make the fail stack and register vectors global. |
2208 | The fail stack, we grow to the maximum size when a regexp |
2209 | is compiled. |
2210 | The register vectors, we adjust in size each time we |
2211 | compile a regexp, according to the number of registers it needs. */ |
2212 | |
2213 | static PREFIX(fail_stack_type) fail_stack; |
2214 | |
2215 | /* Size with which the following vectors are currently allocated. |
2216 | That is so we can make them bigger as needed, |
2217 | but never make them smaller. */ |
2218 | # ifdef DEFINED_ONCE |
2219 | static int regs_allocated_size; |
2220 | |
2221 | static const char ** regstart, ** regend; |
2222 | static const char ** old_regstart, ** old_regend; |
2223 | static const char **best_regstart, **best_regend; |
2224 | static const char **reg_dummy; |
2225 | # endif /* DEFINED_ONCE */ |
2226 | |
2227 | static PREFIX(register_info_type) *PREFIX(reg_info); |
2228 | static PREFIX(register_info_type) *PREFIX(reg_info_dummy); |
2229 | |
2230 | /* Make the register vectors big enough for NUM_REGS registers, |
2231 | but don't make them smaller. */ |
2232 | |
2233 | static void |
2234 | PREFIX(regex_grow_registers) (int num_regs) |
2235 | { |
2236 | if (num_regs > regs_allocated_size) |
2237 | { |
2238 | RETALLOC_IF (regstart, num_regs, const char *); |
2239 | RETALLOC_IF (regend, num_regs, const char *); |
2240 | RETALLOC_IF (old_regstart, num_regs, const char *); |
2241 | RETALLOC_IF (old_regend, num_regs, const char *); |
2242 | RETALLOC_IF (best_regstart, num_regs, const char *); |
2243 | RETALLOC_IF (best_regend, num_regs, const char *); |
2244 | RETALLOC_IF (PREFIX(reg_info), num_regs, PREFIX(register_info_type)); |
2245 | RETALLOC_IF (reg_dummy, num_regs, const char *); |
2246 | RETALLOC_IF (PREFIX(reg_info_dummy), num_regs, PREFIX(register_info_type)); |
2247 | |
2248 | regs_allocated_size = num_regs; |
2249 | } |
2250 | } |
2251 | |
2252 | # endif /* not MATCH_MAY_ALLOCATE */ |
2253 | |
2254 | # ifndef DEFINED_ONCE |
2255 | static boolean group_in_compile_stack (compile_stack_type compile_stack, |
2256 | regnum_t regnum); |
2257 | # endif /* not DEFINED_ONCE */ |
2258 | |
2259 | /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. |
2260 | Returns one of error codes defined in `regex.h', or zero for success. |
2261 | |
2262 | Assumes the `allocated' (and perhaps `buffer') and `translate' |
2263 | fields are set in BUFP on entry. |
2264 | |
2265 | If it succeeds, results are put in BUFP (if it returns an error, the |
2266 | contents of BUFP are undefined): |
2267 | `buffer' is the compiled pattern; |
2268 | `syntax' is set to SYNTAX; |
2269 | `used' is set to the length of the compiled pattern; |
2270 | `fastmap_accurate' is zero; |
2271 | `re_nsub' is the number of subexpressions in PATTERN; |
2272 | `not_bol' and `not_eol' are zero; |
2273 | |
2274 | The `fastmap' and `newline_anchor' fields are neither |
2275 | examined nor set. */ |
2276 | |
2277 | /* Return, freeing storage we allocated. */ |
2278 | # ifdef WCHAR |
2279 | # define FREE_STACK_RETURN(value) \ |
2280 | return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value) |
2281 | # else |
2282 | # define FREE_STACK_RETURN(value) \ |
2283 | return (free (compile_stack.stack), value) |
2284 | # endif /* WCHAR */ |
2285 | |
2286 | static reg_errcode_t |
2287 | PREFIX(regex_compile) (const char *ARG_PREFIX(pattern), |
2288 | size_t ARG_PREFIX(size), reg_syntax_t syntax, |
2289 | struct re_pattern_buffer *bufp) |
2290 | { |
2291 | /* We fetch characters from PATTERN here. Even though PATTERN is |
2292 | `char *' (i.e., signed), we declare these variables as unsigned, so |
2293 | they can be reliably used as array indices. */ |
2294 | register UCHAR_T c, c1; |
2295 | |
2296 | #ifdef WCHAR |
2297 | /* A temporary space to keep wchar_t pattern and compiled pattern. */ |
2298 | CHAR_T *pattern, *COMPILED_BUFFER_VAR; |
2299 | size_t size; |
2300 | /* offset buffer for optimization. See convert_mbs_to_wc. */ |
2301 | int *mbs_offset = NULL; |
2302 | /* It hold whether each wchar_t is binary data or not. */ |
2303 | char *is_binary = NULL; |
2304 | /* A flag whether exactn is handling binary data or not. */ |
2305 | char is_exactn_bin = FALSE; |
2306 | #endif /* WCHAR */ |
2307 | |
2308 | /* A random temporary spot in PATTERN. */ |
2309 | const CHAR_T *p1; |
2310 | |
2311 | /* Points to the end of the buffer, where we should append. */ |
2312 | register UCHAR_T *b; |
2313 | |
2314 | /* Keeps track of unclosed groups. */ |
2315 | compile_stack_type compile_stack; |
2316 | |
2317 | /* Points to the current (ending) position in the pattern. */ |
2318 | #ifdef WCHAR |
2319 | const CHAR_T *p; |
2320 | const CHAR_T *pend; |
2321 | #else /* BYTE */ |
2322 | const CHAR_T *p = pattern; |
2323 | const CHAR_T *pend = pattern + size; |
2324 | #endif /* WCHAR */ |
2325 | |
2326 | /* How to translate the characters in the pattern. */ |
2327 | RE_TRANSLATE_TYPE translate = bufp->translate; |
2328 | |
2329 | /* Address of the count-byte of the most recently inserted `exactn' |
2330 | command. This makes it possible to tell if a new exact-match |
2331 | character can be added to that command or if the character requires |
2332 | a new `exactn' command. */ |
2333 | UCHAR_T *pending_exact = 0; |
2334 | |
2335 | /* Address of start of the most recently finished expression. |
2336 | This tells, e.g., postfix * where to find the start of its |
2337 | operand. Reset at the beginning of groups and alternatives. */ |
2338 | UCHAR_T *laststart = 0; |
2339 | |
2340 | /* Address of beginning of regexp, or inside of last group. */ |
2341 | UCHAR_T *begalt; |
2342 | |
2343 | /* Address of the place where a forward jump should go to the end of |
2344 | the containing expression. Each alternative of an `or' -- except the |
2345 | last -- ends with a forward jump of this sort. */ |
2346 | UCHAR_T *fixup_alt_jump = 0; |
2347 | |
2348 | /* Counts open-groups as they are encountered. Remembered for the |
2349 | matching close-group on the compile stack, so the same register |
2350 | number is put in the stop_memory as the start_memory. */ |
2351 | regnum_t regnum = 0; |
2352 | |
2353 | #ifdef WCHAR |
2354 | /* Initialize the wchar_t PATTERN and offset_buffer. */ |
2355 | p = pend = pattern = TALLOC(csize + 1, CHAR_T); |
2356 | mbs_offset = TALLOC(csize + 1, int); |
2357 | is_binary = TALLOC(csize + 1, char); |
2358 | if (pattern == NULL || mbs_offset == NULL || is_binary == NULL) |
2359 | { |
2360 | free(pattern); |
2361 | free(mbs_offset); |
2362 | free(is_binary); |
2363 | return REG_ESPACE; |
2364 | } |
2365 | pattern[csize] = L'\0'; /* sentinel */ |
2366 | size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary); |
2367 | pend = p + size; |
2368 | if (size < 0) |
2369 | { |
2370 | free(pattern); |
2371 | free(mbs_offset); |
2372 | free(is_binary); |
2373 | return REG_BADPAT; |
2374 | } |
2375 | #endif |
2376 | |
2377 | #ifdef DEBUG |
2378 | DEBUG_PRINT1 ("\nCompiling pattern: " ); |
2379 | if (debug) |
2380 | { |
2381 | unsigned debug_count; |
2382 | |
2383 | for (debug_count = 0; debug_count < size; debug_count++) |
2384 | PUT_CHAR (pattern[debug_count]); |
2385 | putchar ('\n'); |
2386 | } |
2387 | #endif /* DEBUG */ |
2388 | |
2389 | /* Initialize the compile stack. */ |
2390 | compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); |
2391 | if (compile_stack.stack == NULL) |
2392 | { |
2393 | #ifdef WCHAR |
2394 | free(pattern); |
2395 | free(mbs_offset); |
2396 | free(is_binary); |
2397 | #endif |
2398 | return REG_ESPACE; |
2399 | } |
2400 | |
2401 | compile_stack.size = INIT_COMPILE_STACK_SIZE; |
2402 | compile_stack.avail = 0; |
2403 | |
2404 | /* Initialize the pattern buffer. */ |
2405 | bufp->syntax = syntax; |
2406 | bufp->fastmap_accurate = 0; |
2407 | bufp->not_bol = bufp->not_eol = 0; |
2408 | |
2409 | /* Set `used' to zero, so that if we return an error, the pattern |
2410 | printer (for debugging) will think there's no pattern. We reset it |
2411 | at the end. */ |
2412 | bufp->used = 0; |
2413 | |
2414 | /* Always count groups, whether or not bufp->no_sub is set. */ |
2415 | bufp->re_nsub = 0; |
2416 | |
2417 | #if !defined emacs && !defined SYNTAX_TABLE |
2418 | /* Initialize the syntax table. */ |
2419 | init_syntax_once (); |
2420 | #endif |
2421 | |
2422 | if (bufp->allocated == 0) |
2423 | { |
2424 | if (bufp->buffer) |
2425 | { /* If zero allocated, but buffer is non-null, try to realloc |
2426 | enough space. This loses if buffer's address is bogus, but |
2427 | that is the user's responsibility. */ |
2428 | #ifdef WCHAR |
2429 | /* Free bufp->buffer and allocate an array for wchar_t pattern |
2430 | buffer. */ |
2431 | free(bufp->buffer); |
2432 | COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(UCHAR_T), |
2433 | UCHAR_T); |
2434 | #else |
2435 | RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, UCHAR_T); |
2436 | #endif /* WCHAR */ |
2437 | } |
2438 | else |
2439 | { /* Caller did not allocate a buffer. Do it for them. */ |
2440 | COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(UCHAR_T), |
2441 | UCHAR_T); |
2442 | } |
2443 | |
2444 | if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE); |
2445 | #ifdef WCHAR |
2446 | bufp->buffer = (char*)COMPILED_BUFFER_VAR; |
2447 | #endif /* WCHAR */ |
2448 | bufp->allocated = INIT_BUF_SIZE; |
2449 | } |
2450 | #ifdef WCHAR |
2451 | else |
2452 | COMPILED_BUFFER_VAR = (UCHAR_T*) bufp->buffer; |
2453 | #endif |
2454 | |
2455 | begalt = b = COMPILED_BUFFER_VAR; |
2456 | |
2457 | /* Loop through the uncompiled pattern until we're at the end. */ |
2458 | while (p != pend) |
2459 | { |
2460 | PATFETCH (c); |
2461 | |
2462 | switch (c) |
2463 | { |
2464 | case '^': |
2465 | { |
2466 | if ( /* If at start of pattern, it's an operator. */ |
2467 | p == pattern + 1 |
2468 | /* If context independent, it's an operator. */ |
2469 | || syntax & RE_CONTEXT_INDEP_ANCHORS |
2470 | /* Otherwise, depends on what's come before. */ |
2471 | || PREFIX(at_begline_loc_p) (pattern, p, syntax)) |
2472 | BUF_PUSH (begline); |
2473 | else |
2474 | goto normal_char; |
2475 | } |
2476 | break; |
2477 | |
2478 | |
2479 | case '$': |
2480 | { |
2481 | if ( /* If at end of pattern, it's an operator. */ |
2482 | p == pend |
2483 | /* If context independent, it's an operator. */ |
2484 | || syntax & RE_CONTEXT_INDEP_ANCHORS |
2485 | /* Otherwise, depends on what's next. */ |
2486 | || PREFIX(at_endline_loc_p) (p, pend, syntax)) |
2487 | BUF_PUSH (endline); |
2488 | else |
2489 | goto normal_char; |
2490 | } |
2491 | break; |
2492 | |
2493 | |
2494 | case '+': |
2495 | case '?': |
2496 | if ((syntax & RE_BK_PLUS_QM) |
2497 | || (syntax & RE_LIMITED_OPS)) |
2498 | goto normal_char; |
2499 | /* Fall through. */ |
2500 | handle_plus: |
2501 | case '*': |
2502 | /* If there is no previous pattern... */ |
2503 | if (!laststart) |
2504 | { |
2505 | if (syntax & RE_CONTEXT_INVALID_OPS) |
2506 | FREE_STACK_RETURN (REG_BADRPT); |
2507 | else if (!(syntax & RE_CONTEXT_INDEP_OPS)) |
2508 | goto normal_char; |
2509 | } |
2510 | |
2511 | { |
2512 | /* Are we optimizing this jump? */ |
2513 | boolean keep_string_p = false; |
2514 | |
2515 | /* 1 means zero (many) matches is allowed. */ |
2516 | char zero_times_ok = 0, many_times_ok = 0; |
2517 | |
2518 | /* If there is a sequence of repetition chars, collapse it |
2519 | down to just one (the right one). We can't combine |
2520 | interval operators with these because of, e.g., `a{2}*', |
2521 | which should only match an even number of `a's. */ |
2522 | |
2523 | for (;;) |
2524 | { |
2525 | zero_times_ok |= c != '+'; |
2526 | many_times_ok |= c != '?'; |
2527 | |
2528 | if (p == pend) |
2529 | break; |
2530 | |
2531 | PATFETCH (c); |
2532 | |
2533 | if (c == '*' |
2534 | || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?'))) |
2535 | ; |
2536 | |
2537 | else if (syntax & RE_BK_PLUS_QM && c == '\\') |
2538 | { |
2539 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
2540 | |
2541 | PATFETCH (c1); |
2542 | if (!(c1 == '+' || c1 == '?')) |
2543 | { |
2544 | PATUNFETCH; |
2545 | PATUNFETCH; |
2546 | break; |
2547 | } |
2548 | |
2549 | c = c1; |
2550 | } |
2551 | else |
2552 | { |
2553 | PATUNFETCH; |
2554 | break; |
2555 | } |
2556 | |
2557 | /* If we get here, we found another repeat character. */ |
2558 | } |
2559 | |
2560 | /* Star, etc. applied to an empty pattern is equivalent |
2561 | to an empty pattern. */ |
2562 | if (!laststart) |
2563 | break; |
2564 | |
2565 | /* Now we know whether or not zero matches is allowed |
2566 | and also whether or not two or more matches is allowed. */ |
2567 | if (many_times_ok) |
2568 | { /* More than one repetition is allowed, so put in at the |
2569 | end a backward relative jump from `b' to before the next |
2570 | jump we're going to put in below (which jumps from |
2571 | laststart to after this jump). |
2572 | |
2573 | But if we are at the `*' in the exact sequence `.*\n', |
2574 | insert an unconditional jump backwards to the ., |
2575 | instead of the beginning of the loop. This way we only |
2576 | push a failure point once, instead of every time |
2577 | through the loop. */ |
2578 | assert (p - 1 > pattern); |
2579 | |
2580 | /* Allocate the space for the jump. */ |
2581 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
2582 | |
2583 | /* We know we are not at the first character of the pattern, |
2584 | because laststart was nonzero. And we've already |
2585 | incremented `p', by the way, to be the character after |
2586 | the `*'. Do we have to do something analogous here |
2587 | for null bytes, because of RE_DOT_NOT_NULL? */ |
2588 | if (TRANSLATE (*(p - 2)) == TRANSLATE ('.') |
2589 | && zero_times_ok |
2590 | && p < pend && TRANSLATE (*p) == TRANSLATE ('\n') |
2591 | && !(syntax & RE_DOT_NEWLINE)) |
2592 | { /* We have .*\n. */ |
2593 | STORE_JUMP (jump, b, laststart); |
2594 | keep_string_p = true; |
2595 | } |
2596 | else |
2597 | /* Anything else. */ |
2598 | STORE_JUMP (maybe_pop_jump, b, laststart - |
2599 | (1 + OFFSET_ADDRESS_SIZE)); |
2600 | |
2601 | /* We've added more stuff to the buffer. */ |
2602 | b += 1 + OFFSET_ADDRESS_SIZE; |
2603 | } |
2604 | |
2605 | /* On failure, jump from laststart to b + 3, which will be the |
2606 | end of the buffer after this jump is inserted. */ |
2607 | /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of |
2608 | 'b + 3'. */ |
2609 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
2610 | INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump |
2611 | : on_failure_jump, |
2612 | laststart, b + 1 + OFFSET_ADDRESS_SIZE); |
2613 | pending_exact = 0; |
2614 | b += 1 + OFFSET_ADDRESS_SIZE; |
2615 | |
2616 | if (!zero_times_ok) |
2617 | { |
2618 | /* At least one repetition is required, so insert a |
2619 | `dummy_failure_jump' before the initial |
2620 | `on_failure_jump' instruction of the loop. This |
2621 | effects a skip over that instruction the first time |
2622 | we hit that loop. */ |
2623 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
2624 | INSERT_JUMP (dummy_failure_jump, laststart, laststart + |
2625 | 2 + 2 * OFFSET_ADDRESS_SIZE); |
2626 | b += 1 + OFFSET_ADDRESS_SIZE; |
2627 | } |
2628 | } |
2629 | break; |
2630 | |
2631 | |
2632 | case '.': |
2633 | laststart = b; |
2634 | BUF_PUSH (anychar); |
2635 | break; |
2636 | |
2637 | |
2638 | case '[': |
2639 | { |
2640 | boolean had_char_class = false; |
2641 | #ifdef WCHAR |
2642 | CHAR_T range_start = 0xffffffff; |
2643 | #else |
2644 | unsigned int range_start = 0xffffffff; |
2645 | #endif |
2646 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
2647 | |
2648 | #ifdef WCHAR |
2649 | /* We assume a charset(_not) structure as a wchar_t array. |
2650 | charset[0] = (re_opcode_t) charset(_not) |
2651 | charset[1] = l (= length of char_classes) |
2652 | charset[2] = m (= length of collating_symbols) |
2653 | charset[3] = n (= length of equivalence_classes) |
2654 | charset[4] = o (= length of char_ranges) |
2655 | charset[5] = p (= length of chars) |
2656 | |
2657 | charset[6] = char_class (wctype_t) |
2658 | charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t) |
2659 | ... |
2660 | charset[l+5] = char_class (wctype_t) |
2661 | |
2662 | charset[l+6] = collating_symbol (wchar_t) |
2663 | ... |
2664 | charset[l+m+5] = collating_symbol (wchar_t) |
2665 | ifdef _LIBC we use the index if |
2666 | _NL_COLLATE_SYMB_EXTRAMB instead of |
2667 | wchar_t string. |
2668 | |
2669 | charset[l+m+6] = equivalence_classes (wchar_t) |
2670 | ... |
2671 | charset[l+m+n+5] = equivalence_classes (wchar_t) |
2672 | ifdef _LIBC we use the index in |
2673 | _NL_COLLATE_WEIGHT instead of |
2674 | wchar_t string. |
2675 | |
2676 | charset[l+m+n+6] = range_start |
2677 | charset[l+m+n+7] = range_end |
2678 | ... |
2679 | charset[l+m+n+2o+4] = range_start |
2680 | charset[l+m+n+2o+5] = range_end |
2681 | ifdef _LIBC we use the value looked up |
2682 | in _NL_COLLATE_COLLSEQ instead of |
2683 | wchar_t character. |
2684 | |
2685 | charset[l+m+n+2o+6] = char |
2686 | ... |
2687 | charset[l+m+n+2o+p+5] = char |
2688 | |
2689 | */ |
2690 | |
2691 | /* We need at least 6 spaces: the opcode, the length of |
2692 | char_classes, the length of collating_symbols, the length of |
2693 | equivalence_classes, the length of char_ranges, the length of |
2694 | chars. */ |
2695 | GET_BUFFER_SPACE (6); |
2696 | |
2697 | /* Save b as laststart. And We use laststart as the pointer |
2698 | to the first element of the charset here. |
2699 | In other words, laststart[i] indicates charset[i]. */ |
2700 | laststart = b; |
2701 | |
2702 | /* We test `*p == '^' twice, instead of using an if |
2703 | statement, so we only need one BUF_PUSH. */ |
2704 | BUF_PUSH (*p == '^' ? charset_not : charset); |
2705 | if (*p == '^') |
2706 | p++; |
2707 | |
2708 | /* Push the length of char_classes, the length of |
2709 | collating_symbols, the length of equivalence_classes, the |
2710 | length of char_ranges and the length of chars. */ |
2711 | BUF_PUSH_3 (0, 0, 0); |
2712 | BUF_PUSH_2 (0, 0); |
2713 | |
2714 | /* Remember the first position in the bracket expression. */ |
2715 | p1 = p; |
2716 | |
2717 | /* charset_not matches newline according to a syntax bit. */ |
2718 | if ((re_opcode_t) b[-6] == charset_not |
2719 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) |
2720 | { |
2721 | BUF_PUSH('\n'); |
2722 | laststart[5]++; /* Update the length of characters */ |
2723 | } |
2724 | |
2725 | /* Read in characters and ranges, setting map bits. */ |
2726 | for (;;) |
2727 | { |
2728 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
2729 | |
2730 | PATFETCH (c); |
2731 | |
2732 | /* \ might escape characters inside [...] and [^...]. */ |
2733 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') |
2734 | { |
2735 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
2736 | |
2737 | PATFETCH (c1); |
2738 | BUF_PUSH(c1); |
2739 | laststart[5]++; /* Update the length of chars */ |
2740 | range_start = c1; |
2741 | continue; |
2742 | } |
2743 | |
2744 | /* Could be the end of the bracket expression. If it's |
2745 | not (i.e., when the bracket expression is `[]' so |
2746 | far), the ']' character bit gets set way below. */ |
2747 | if (c == ']' && p != p1 + 1) |
2748 | break; |
2749 | |
2750 | /* Look ahead to see if it's a range when the last thing |
2751 | was a character class. */ |
2752 | if (had_char_class && c == '-' && *p != ']') |
2753 | FREE_STACK_RETURN (REG_ERANGE); |
2754 | |
2755 | /* Look ahead to see if it's a range when the last thing |
2756 | was a character: if this is a hyphen not at the |
2757 | beginning or the end of a list, then it's the range |
2758 | operator. */ |
2759 | if (c == '-' |
2760 | && !(p - 2 >= pattern && p[-2] == '[') |
2761 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') |
2762 | && *p != ']') |
2763 | { |
2764 | reg_errcode_t ret; |
2765 | /* Allocate the space for range_start and range_end. */ |
2766 | GET_BUFFER_SPACE (2); |
2767 | /* Update the pointer to indicate end of buffer. */ |
2768 | b += 2; |
2769 | ret = wcs_compile_range (range_start, &p, pend, translate, |
2770 | syntax, b, laststart); |
2771 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
2772 | range_start = 0xffffffff; |
2773 | } |
2774 | else if (p[0] == '-' && p[1] != ']') |
2775 | { /* This handles ranges made up of characters only. */ |
2776 | reg_errcode_t ret; |
2777 | |
2778 | /* Move past the `-'. */ |
2779 | PATFETCH (c1); |
2780 | /* Allocate the space for range_start and range_end. */ |
2781 | GET_BUFFER_SPACE (2); |
2782 | /* Update the pointer to indicate end of buffer. */ |
2783 | b += 2; |
2784 | ret = wcs_compile_range (c, &p, pend, translate, syntax, b, |
2785 | laststart); |
2786 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
2787 | range_start = 0xffffffff; |
2788 | } |
2789 | |
2790 | /* See if we're at the beginning of a possible character |
2791 | class. */ |
2792 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') |
2793 | { /* Leave room for the null. */ |
2794 | char str[CHAR_CLASS_MAX_LENGTH + 1]; |
2795 | |
2796 | PATFETCH (c); |
2797 | c1 = 0; |
2798 | |
2799 | /* If pattern is `[[:'. */ |
2800 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
2801 | |
2802 | for (;;) |
2803 | { |
2804 | PATFETCH (c); |
2805 | if ((c == ':' && *p == ']') || p == pend) |
2806 | break; |
2807 | if (c1 < CHAR_CLASS_MAX_LENGTH) |
2808 | str[c1++] = c; |
2809 | else |
2810 | /* This is in any case an invalid class name. */ |
2811 | str[0] = '\0'; |
2812 | } |
2813 | str[c1] = '\0'; |
2814 | |
2815 | /* If isn't a word bracketed by `[:' and `:]': |
2816 | undo the ending character, the letters, and leave |
2817 | the leading `:' and `[' (but store them as character). */ |
2818 | if (c == ':' && *p == ']') |
2819 | { |
2820 | wctype_t wt; |
2821 | uintptr_t alignedp; |
2822 | |
2823 | /* Query the character class as wctype_t. */ |
2824 | wt = IS_CHAR_CLASS (str); |
2825 | if (wt == 0) |
2826 | FREE_STACK_RETURN (REG_ECTYPE); |
2827 | |
2828 | /* Throw away the ] at the end of the character |
2829 | class. */ |
2830 | PATFETCH (c); |
2831 | |
2832 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
2833 | |
2834 | /* Allocate the space for character class. */ |
2835 | GET_BUFFER_SPACE(CHAR_CLASS_SIZE); |
2836 | /* Update the pointer to indicate end of buffer. */ |
2837 | b += CHAR_CLASS_SIZE; |
2838 | /* Move data which follow character classes |
2839 | not to violate the data. */ |
2840 | insert_space(CHAR_CLASS_SIZE, |
2841 | laststart + 6 + laststart[1], |
2842 | b - 1); |
2843 | alignedp = ((uintptr_t)(laststart + 6 + laststart[1]) |
2844 | + __alignof__(wctype_t) - 1) |
2845 | & ~(uintptr_t)(__alignof__(wctype_t) - 1); |
2846 | /* Store the character class. */ |
2847 | *((wctype_t*)alignedp) = wt; |
2848 | /* Update length of char_classes */ |
2849 | laststart[1] += CHAR_CLASS_SIZE; |
2850 | |
2851 | had_char_class = true; |
2852 | } |
2853 | else |
2854 | { |
2855 | c1++; |
2856 | while (c1--) |
2857 | PATUNFETCH; |
2858 | BUF_PUSH ('['); |
2859 | BUF_PUSH (':'); |
2860 | laststart[5] += 2; /* Update the length of characters */ |
2861 | range_start = ':'; |
2862 | had_char_class = false; |
2863 | } |
2864 | } |
2865 | else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '=' |
2866 | || *p == '.')) |
2867 | { |
2868 | CHAR_T str[128]; /* Should be large enough. */ |
2869 | CHAR_T delim = *p; /* '=' or '.' */ |
2870 | # ifdef _LIBC |
2871 | uint32_t nrules = |
2872 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); |
2873 | # endif |
2874 | PATFETCH (c); |
2875 | c1 = 0; |
2876 | |
2877 | /* If pattern is `[[=' or '[[.'. */ |
2878 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
2879 | |
2880 | for (;;) |
2881 | { |
2882 | PATFETCH (c); |
2883 | if ((c == delim && *p == ']') || p == pend) |
2884 | break; |
2885 | if (c1 < sizeof (str) - 1) |
2886 | str[c1++] = c; |
2887 | else |
2888 | /* This is in any case an invalid class name. */ |
2889 | str[0] = '\0'; |
2890 | } |
2891 | str[c1] = '\0'; |
2892 | |
2893 | if (c == delim && *p == ']' && str[0] != '\0') |
2894 | { |
2895 | unsigned int i, offset; |
2896 | /* If we have no collation data we use the default |
2897 | collation in which each character is in a class |
2898 | by itself. It also means that ASCII is the |
2899 | character set and therefore we cannot have character |
2900 | with more than one byte in the multibyte |
2901 | representation. */ |
2902 | |
2903 | /* If not defined _LIBC, we push the name and |
2904 | `\0' for the sake of matching performance. */ |
2905 | int datasize = c1 + 1; |
2906 | |
2907 | # ifdef _LIBC |
2908 | int32_t idx = 0; |
2909 | if (nrules == 0) |
2910 | # endif |
2911 | { |
2912 | if (c1 != 1) |
2913 | FREE_STACK_RETURN (REG_ECOLLATE); |
2914 | } |
2915 | # ifdef _LIBC |
2916 | else |
2917 | { |
2918 | const int32_t *table; |
2919 | const int32_t *weights; |
2920 | const int32_t *extra; |
2921 | const int32_t *indirect; |
2922 | wint_t *cp; |
2923 | |
2924 | /* This #include defines a local function! */ |
2925 | # include <locale/weightwc.h> |
2926 | |
2927 | if(delim == '=') |
2928 | { |
2929 | /* We push the index for equivalence class. */ |
2930 | cp = (wint_t*)str; |
2931 | |
2932 | table = (const int32_t *) |
2933 | _NL_CURRENT (LC_COLLATE, |
2934 | _NL_COLLATE_TABLEWC); |
2935 | weights = (const int32_t *) |
2936 | _NL_CURRENT (LC_COLLATE, |
2937 | _NL_COLLATE_WEIGHTWC); |
2938 | extra = (const int32_t *) |
2939 | _NL_CURRENT (LC_COLLATE, |
2940 | _NL_COLLATE_EXTRAWC); |
2941 | indirect = (const int32_t *) |
2942 | _NL_CURRENT (LC_COLLATE, |
2943 | _NL_COLLATE_INDIRECTWC); |
2944 | |
2945 | idx = findidx ((const wint_t**)&cp); |
2946 | if (idx == 0 || cp < (wint_t*) str + c1) |
2947 | /* This is no valid character. */ |
2948 | FREE_STACK_RETURN (REG_ECOLLATE); |
2949 | |
2950 | str[0] = (wchar_t)idx; |
2951 | } |
2952 | else /* delim == '.' */ |
2953 | { |
2954 | /* We push collation sequence value |
2955 | for collating symbol. */ |
2956 | int32_t table_size; |
2957 | const int32_t *symb_table; |
2958 | const unsigned char *extra; |
2959 | int32_t idx; |
2960 | int32_t elem; |
2961 | int32_t second; |
2962 | int32_t hash; |
2963 | char char_str[c1]; |
2964 | |
2965 | /* We have to convert the name to a single-byte |
2966 | string. This is possible since the names |
2967 | consist of ASCII characters and the internal |
2968 | representation is UCS4. */ |
2969 | for (i = 0; i < c1; ++i) |
2970 | char_str[i] = str[i]; |
2971 | |
2972 | table_size = |
2973 | _NL_CURRENT_WORD (LC_COLLATE, |
2974 | _NL_COLLATE_SYMB_HASH_SIZEMB); |
2975 | symb_table = (const int32_t *) |
2976 | _NL_CURRENT (LC_COLLATE, |
2977 | _NL_COLLATE_SYMB_TABLEMB); |
2978 | extra = (const unsigned char *) |
2979 | _NL_CURRENT (LC_COLLATE, |
2980 | _NL_COLLATE_SYMB_EXTRAMB); |
2981 | |
2982 | /* Locate the character in the hashing table. */ |
2983 | hash = elem_hash (char_str, c1); |
2984 | |
2985 | idx = 0; |
2986 | elem = hash % table_size; |
2987 | second = hash % (table_size - 2); |
2988 | while (symb_table[2 * elem] != 0) |
2989 | { |
2990 | /* First compare the hashing value. */ |
2991 | if (symb_table[2 * elem] == hash |
2992 | && c1 == extra[symb_table[2 * elem + 1]] |
2993 | && memcmp (char_str, |
2994 | &extra[symb_table[2 * elem + 1] |
2995 | + 1], c1) == 0) |
2996 | { |
2997 | /* Yep, this is the entry. */ |
2998 | idx = symb_table[2 * elem + 1]; |
2999 | idx += 1 + extra[idx]; |
3000 | break; |
3001 | } |
3002 | |
3003 | /* Next entry. */ |
3004 | elem += second; |
3005 | } |
3006 | |
3007 | if (symb_table[2 * elem] != 0) |
3008 | { |
3009 | /* Compute the index of the byte sequence |
3010 | in the table. */ |
3011 | idx += 1 + extra[idx]; |
3012 | /* Adjust for the alignment. */ |
3013 | idx = (idx + 3) & ~3; |
3014 | |
3015 | str[0] = (wchar_t) idx + 4; |
3016 | } |
3017 | else if (symb_table[2 * elem] == 0 && c1 == 1) |
3018 | { |
3019 | /* No valid character. Match it as a |
3020 | single byte character. */ |
3021 | had_char_class = false; |
3022 | BUF_PUSH(str[0]); |
3023 | /* Update the length of characters */ |
3024 | laststart[5]++; |
3025 | range_start = str[0]; |
3026 | |
3027 | /* Throw away the ] at the end of the |
3028 | collating symbol. */ |
3029 | PATFETCH (c); |
3030 | /* exit from the switch block. */ |
3031 | continue; |
3032 | } |
3033 | else |
3034 | FREE_STACK_RETURN (REG_ECOLLATE); |
3035 | } |
3036 | datasize = 1; |
3037 | } |
3038 | # endif |
3039 | /* Throw away the ] at the end of the equivalence |
3040 | class (or collating symbol). */ |
3041 | PATFETCH (c); |
3042 | |
3043 | /* Allocate the space for the equivalence class |
3044 | (or collating symbol) (and '\0' if needed). */ |
3045 | GET_BUFFER_SPACE(datasize); |
3046 | /* Update the pointer to indicate end of buffer. */ |
3047 | b += datasize; |
3048 | |
3049 | if (delim == '=') |
3050 | { /* equivalence class */ |
3051 | /* Calculate the offset of char_ranges, |
3052 | which is next to equivalence_classes. */ |
3053 | offset = laststart[1] + laststart[2] |
3054 | + laststart[3] +6; |
3055 | /* Insert space. */ |
3056 | insert_space(datasize, laststart + offset, b - 1); |
3057 | |
3058 | /* Write the equivalence_class and \0. */ |
3059 | for (i = 0 ; i < datasize ; i++) |
3060 | laststart[offset + i] = str[i]; |
3061 | |
3062 | /* Update the length of equivalence_classes. */ |
3063 | laststart[3] += datasize; |
3064 | had_char_class = true; |
3065 | } |
3066 | else /* delim == '.' */ |
3067 | { /* collating symbol */ |
3068 | /* Calculate the offset of the equivalence_classes, |
3069 | which is next to collating_symbols. */ |
3070 | offset = laststart[1] + laststart[2] + 6; |
3071 | /* Insert space and write the collationg_symbol |
3072 | and \0. */ |
3073 | insert_space(datasize, laststart + offset, b-1); |
3074 | for (i = 0 ; i < datasize ; i++) |
3075 | laststart[offset + i] = str[i]; |
3076 | |
3077 | /* In re_match_2_internal if range_start < -1, we |
3078 | assume -range_start is the offset of the |
3079 | collating symbol which is specified as |
3080 | the character of the range start. So we assign |
3081 | -(laststart[1] + laststart[2] + 6) to |
3082 | range_start. */ |
3083 | range_start = -(laststart[1] + laststart[2] + 6); |
3084 | /* Update the length of collating_symbol. */ |
3085 | laststart[2] += datasize; |
3086 | had_char_class = false; |
3087 | } |
3088 | } |
3089 | else |
3090 | { |
3091 | c1++; |
3092 | while (c1--) |
3093 | PATUNFETCH; |
3094 | BUF_PUSH ('['); |
3095 | BUF_PUSH (delim); |
3096 | laststart[5] += 2; /* Update the length of characters */ |
3097 | range_start = delim; |
3098 | had_char_class = false; |
3099 | } |
3100 | } |
3101 | else |
3102 | { |
3103 | had_char_class = false; |
3104 | BUF_PUSH(c); |
3105 | laststart[5]++; /* Update the length of characters */ |
3106 | range_start = c; |
3107 | } |
3108 | } |
3109 | |
3110 | #else /* BYTE */ |
3111 | /* Ensure that we have enough space to push a charset: the |
3112 | opcode, the length count, and the bitset; 34 bytes in all. */ |
3113 | GET_BUFFER_SPACE (34); |
3114 | |
3115 | laststart = b; |
3116 | |
3117 | /* We test `*p == '^' twice, instead of using an if |
3118 | statement, so we only need one BUF_PUSH. */ |
3119 | BUF_PUSH (*p == '^' ? charset_not : charset); |
3120 | if (*p == '^') |
3121 | p++; |
3122 | |
3123 | /* Remember the first position in the bracket expression. */ |
3124 | p1 = p; |
3125 | |
3126 | /* Push the number of bytes in the bitmap. */ |
3127 | BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH); |
3128 | |
3129 | /* Clear the whole map. */ |
3130 | bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH); |
3131 | |
3132 | /* charset_not matches newline according to a syntax bit. */ |
3133 | if ((re_opcode_t) b[-2] == charset_not |
3134 | && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) |
3135 | SET_LIST_BIT ('\n'); |
3136 | |
3137 | /* Read in characters and ranges, setting map bits. */ |
3138 | for (;;) |
3139 | { |
3140 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
3141 | |
3142 | PATFETCH (c); |
3143 | |
3144 | /* \ might escape characters inside [...] and [^...]. */ |
3145 | if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') |
3146 | { |
3147 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
3148 | |
3149 | PATFETCH (c1); |
3150 | SET_LIST_BIT (c1); |
3151 | range_start = c1; |
3152 | continue; |
3153 | } |
3154 | |
3155 | /* Could be the end of the bracket expression. If it's |
3156 | not (i.e., when the bracket expression is `[]' so |
3157 | far), the ']' character bit gets set way below. */ |
3158 | if (c == ']' && p != p1 + 1) |
3159 | break; |
3160 | |
3161 | /* Look ahead to see if it's a range when the last thing |
3162 | was a character class. */ |
3163 | if (had_char_class && c == '-' && *p != ']') |
3164 | FREE_STACK_RETURN (REG_ERANGE); |
3165 | |
3166 | /* Look ahead to see if it's a range when the last thing |
3167 | was a character: if this is a hyphen not at the |
3168 | beginning or the end of a list, then it's the range |
3169 | operator. */ |
3170 | if (c == '-' |
3171 | && !(p - 2 >= pattern && p[-2] == '[') |
3172 | && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') |
3173 | && *p != ']') |
3174 | { |
3175 | reg_errcode_t ret |
3176 | = byte_compile_range (range_start, &p, pend, translate, |
3177 | syntax, b); |
3178 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
3179 | range_start = 0xffffffff; |
3180 | } |
3181 | |
3182 | else if (p[0] == '-' && p[1] != ']') |
3183 | { /* This handles ranges made up of characters only. */ |
3184 | reg_errcode_t ret; |
3185 | |
3186 | /* Move past the `-'. */ |
3187 | PATFETCH (c1); |
3188 | |
3189 | ret = byte_compile_range (c, &p, pend, translate, syntax, b); |
3190 | if (ret != REG_NOERROR) FREE_STACK_RETURN (ret); |
3191 | range_start = 0xffffffff; |
3192 | } |
3193 | |
3194 | /* See if we're at the beginning of a possible character |
3195 | class. */ |
3196 | |
3197 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') |
3198 | { /* Leave room for the null. */ |
3199 | char str[CHAR_CLASS_MAX_LENGTH + 1]; |
3200 | |
3201 | PATFETCH (c); |
3202 | c1 = 0; |
3203 | |
3204 | /* If pattern is `[[:'. */ |
3205 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
3206 | |
3207 | for (;;) |
3208 | { |
3209 | PATFETCH (c); |
3210 | if ((c == ':' && *p == ']') || p == pend) |
3211 | break; |
3212 | if (c1 < CHAR_CLASS_MAX_LENGTH) |
3213 | str[c1++] = c; |
3214 | else |
3215 | /* This is in any case an invalid class name. */ |
3216 | str[0] = '\0'; |
3217 | } |
3218 | str[c1] = '\0'; |
3219 | |
3220 | /* If isn't a word bracketed by `[:' and `:]': |
3221 | undo the ending character, the letters, and leave |
3222 | the leading `:' and `[' (but set bits for them). */ |
3223 | if (c == ':' && *p == ']') |
3224 | { |
3225 | # if defined _LIBC || WIDE_CHAR_SUPPORT |
3226 | boolean is_lower = STREQ (str, "lower" ); |
3227 | boolean is_upper = STREQ (str, "upper" ); |
3228 | wctype_t wt; |
3229 | int ch; |
3230 | |
3231 | wt = IS_CHAR_CLASS (str); |
3232 | if (wt == 0) |
3233 | FREE_STACK_RETURN (REG_ECTYPE); |
3234 | |
3235 | /* Throw away the ] at the end of the character |
3236 | class. */ |
3237 | PATFETCH (c); |
3238 | |
3239 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
3240 | |
3241 | for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) |
3242 | { |
3243 | # ifdef _LIBC |
3244 | if (__iswctype (__btowc (ch), wt)) |
3245 | SET_LIST_BIT (ch); |
3246 | # else |
3247 | if (iswctype (btowc (ch), wt)) |
3248 | SET_LIST_BIT (ch); |
3249 | # endif |
3250 | |
3251 | if (translate && (is_upper || is_lower) |
3252 | && (ISUPPER (ch) || ISLOWER (ch))) |
3253 | SET_LIST_BIT (ch); |
3254 | } |
3255 | |
3256 | had_char_class = true; |
3257 | # else |
3258 | int ch; |
3259 | boolean is_alnum = STREQ (str, "alnum" ); |
3260 | boolean is_alpha = STREQ (str, "alpha" ); |
3261 | boolean is_blank = STREQ (str, "blank" ); |
3262 | boolean is_cntrl = STREQ (str, "cntrl" ); |
3263 | boolean is_digit = STREQ (str, "digit" ); |
3264 | boolean is_graph = STREQ (str, "graph" ); |
3265 | boolean is_lower = STREQ (str, "lower" ); |
3266 | boolean is_print = STREQ (str, "print" ); |
3267 | boolean is_punct = STREQ (str, "punct" ); |
3268 | boolean is_space = STREQ (str, "space" ); |
3269 | boolean is_upper = STREQ (str, "upper" ); |
3270 | boolean is_xdigit = STREQ (str, "xdigit" ); |
3271 | |
3272 | if (!IS_CHAR_CLASS (str)) |
3273 | FREE_STACK_RETURN (REG_ECTYPE); |
3274 | |
3275 | /* Throw away the ] at the end of the character |
3276 | class. */ |
3277 | PATFETCH (c); |
3278 | |
3279 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
3280 | |
3281 | for (ch = 0; ch < 1 << BYTEWIDTH; ch++) |
3282 | { |
3283 | /* This was split into 3 if's to |
3284 | avoid an arbitrary limit in some compiler. */ |
3285 | if ( (is_alnum && ISALNUM (ch)) |
3286 | || (is_alpha && ISALPHA (ch)) |
3287 | || (is_blank && ISBLANK (ch)) |
3288 | || (is_cntrl && ISCNTRL (ch))) |
3289 | SET_LIST_BIT (ch); |
3290 | if ( (is_digit && ISDIGIT (ch)) |
3291 | || (is_graph && ISGRAPH (ch)) |
3292 | || (is_lower && ISLOWER (ch)) |
3293 | || (is_print && ISPRINT (ch))) |
3294 | SET_LIST_BIT (ch); |
3295 | if ( (is_punct && ISPUNCT (ch)) |
3296 | || (is_space && ISSPACE (ch)) |
3297 | || (is_upper && ISUPPER (ch)) |
3298 | || (is_xdigit && ISXDIGIT (ch))) |
3299 | SET_LIST_BIT (ch); |
3300 | if ( translate && (is_upper || is_lower) |
3301 | && (ISUPPER (ch) || ISLOWER (ch))) |
3302 | SET_LIST_BIT (ch); |
3303 | } |
3304 | had_char_class = true; |
3305 | # endif /* libc || wctype.h */ |
3306 | } |
3307 | else |
3308 | { |
3309 | c1++; |
3310 | while (c1--) |
3311 | PATUNFETCH; |
3312 | SET_LIST_BIT ('['); |
3313 | SET_LIST_BIT (':'); |
3314 | range_start = ':'; |
3315 | had_char_class = false; |
3316 | } |
3317 | } |
3318 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=') |
3319 | { |
3320 | unsigned char str[MB_LEN_MAX + 1]; |
3321 | # ifdef _LIBC |
3322 | uint32_t nrules = |
3323 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); |
3324 | # endif |
3325 | |
3326 | PATFETCH (c); |
3327 | c1 = 0; |
3328 | |
3329 | /* If pattern is `[[='. */ |
3330 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
3331 | |
3332 | for (;;) |
3333 | { |
3334 | PATFETCH (c); |
3335 | if ((c == '=' && *p == ']') || p == pend) |
3336 | break; |
3337 | if (c1 < MB_LEN_MAX) |
3338 | str[c1++] = c; |
3339 | else |
3340 | /* This is in any case an invalid class name. */ |
3341 | str[0] = '\0'; |
3342 | } |
3343 | str[c1] = '\0'; |
3344 | |
3345 | if (c == '=' && *p == ']' && str[0] != '\0') |
3346 | { |
3347 | /* If we have no collation data we use the default |
3348 | collation in which each character is in a class |
3349 | by itself. It also means that ASCII is the |
3350 | character set and therefore we cannot have character |
3351 | with more than one byte in the multibyte |
3352 | representation. */ |
3353 | # ifdef _LIBC |
3354 | if (nrules == 0) |
3355 | # endif |
3356 | { |
3357 | if (c1 != 1) |
3358 | FREE_STACK_RETURN (REG_ECOLLATE); |
3359 | |
3360 | /* Throw away the ] at the end of the equivalence |
3361 | class. */ |
3362 | PATFETCH (c); |
3363 | |
3364 | /* Set the bit for the character. */ |
3365 | SET_LIST_BIT (str[0]); |
3366 | } |
3367 | # ifdef _LIBC |
3368 | else |
3369 | { |
3370 | /* Try to match the byte sequence in `str' against |
3371 | those known to the collate implementation. |
3372 | First find out whether the bytes in `str' are |
3373 | actually from exactly one character. */ |
3374 | const int32_t *table; |
3375 | const unsigned char *weights; |
3376 | const unsigned char *extra; |
3377 | const int32_t *indirect; |
3378 | int32_t idx; |
3379 | const unsigned char *cp = str; |
3380 | int ch; |
3381 | |
3382 | /* This #include defines a local function! */ |
3383 | # include <locale/weight.h> |
3384 | |
3385 | table = (const int32_t *) |
3386 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB); |
3387 | weights = (const unsigned char *) |
3388 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB); |
3389 | extra = (const unsigned char *) |
3390 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB); |
3391 | indirect = (const int32_t *) |
3392 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB); |
3393 | |
3394 | idx = findidx (&cp); |
3395 | if (idx == 0 || cp < str + c1) |
3396 | /* This is no valid character. */ |
3397 | FREE_STACK_RETURN (REG_ECOLLATE); |
3398 | |
3399 | /* Throw away the ] at the end of the equivalence |
3400 | class. */ |
3401 | PATFETCH (c); |
3402 | |
3403 | /* Now we have to go through the whole table |
3404 | and find all characters which have the same |
3405 | first level weight. |
3406 | |
3407 | XXX Note that this is not entirely correct. |
3408 | we would have to match multibyte sequences |
3409 | but this is not possible with the current |
3410 | implementation. */ |
3411 | for (ch = 1; ch < 256; ++ch) |
3412 | /* XXX This test would have to be changed if we |
3413 | would allow matching multibyte sequences. */ |
3414 | if (table[ch] > 0) |
3415 | { |
3416 | int32_t idx2 = table[ch]; |
3417 | size_t len = weights[idx2]; |
3418 | |
3419 | /* Test whether the lenghts match. */ |
3420 | if (weights[idx] == len) |
3421 | { |
3422 | /* They do. New compare the bytes of |
3423 | the weight. */ |
3424 | size_t cnt = 0; |
3425 | |
3426 | while (cnt < len |
3427 | && (weights[idx + 1 + cnt] |
3428 | == weights[idx2 + 1 + cnt])) |
3429 | ++cnt; |
3430 | |
3431 | if (cnt == len) |
3432 | /* They match. Mark the character as |
3433 | acceptable. */ |
3434 | SET_LIST_BIT (ch); |
3435 | } |
3436 | } |
3437 | } |
3438 | # endif |
3439 | had_char_class = true; |
3440 | } |
3441 | else |
3442 | { |
3443 | c1++; |
3444 | while (c1--) |
3445 | PATUNFETCH; |
3446 | SET_LIST_BIT ('['); |
3447 | SET_LIST_BIT ('='); |
3448 | range_start = '='; |
3449 | had_char_class = false; |
3450 | } |
3451 | } |
3452 | else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.') |
3453 | { |
3454 | unsigned char str[128]; /* Should be large enough. */ |
3455 | # ifdef _LIBC |
3456 | uint32_t nrules = |
3457 | _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); |
3458 | # endif |
3459 | |
3460 | PATFETCH (c); |
3461 | c1 = 0; |
3462 | |
3463 | /* If pattern is `[[.'. */ |
3464 | if (p == pend) FREE_STACK_RETURN (REG_EBRACK); |
3465 | |
3466 | for (;;) |
3467 | { |
3468 | PATFETCH (c); |
3469 | if ((c == '.' && *p == ']') || p == pend) |
3470 | break; |
3471 | if (c1 < sizeof (str)) |
3472 | str[c1++] = c; |
3473 | else |
3474 | /* This is in any case an invalid class name. */ |
3475 | str[0] = '\0'; |
3476 | } |
3477 | str[c1] = '\0'; |
3478 | |
3479 | if (c == '.' && *p == ']' && str[0] != '\0') |
3480 | { |
3481 | /* If we have no collation data we use the default |
3482 | collation in which each character is the name |
3483 | for its own class which contains only the one |
3484 | character. It also means that ASCII is the |
3485 | character set and therefore we cannot have character |
3486 | with more than one byte in the multibyte |
3487 | representation. */ |
3488 | # ifdef _LIBC |
3489 | if (nrules == 0) |
3490 | # endif |
3491 | { |
3492 | if (c1 != 1) |
3493 | FREE_STACK_RETURN (REG_ECOLLATE); |
3494 | |
3495 | /* Throw away the ] at the end of the equivalence |
3496 | class. */ |
3497 | PATFETCH (c); |
3498 | |
3499 | /* Set the bit for the character. */ |
3500 | SET_LIST_BIT (str[0]); |
3501 | range_start = ((const unsigned char *) str)[0]; |
3502 | } |
3503 | # ifdef _LIBC |
3504 | else |
3505 | { |
3506 | /* Try to match the byte sequence in `str' against |
3507 | those known to the collate implementation. |
3508 | First find out whether the bytes in `str' are |
3509 | actually from exactly one character. */ |
3510 | int32_t table_size; |
3511 | const int32_t *symb_table; |
3512 | const unsigned char *extra; |
3513 | int32_t idx; |
3514 | int32_t elem; |
3515 | int32_t second; |
3516 | int32_t hash; |
3517 | |
3518 | table_size = |
3519 | _NL_CURRENT_WORD (LC_COLLATE, |
3520 | _NL_COLLATE_SYMB_HASH_SIZEMB); |
3521 | symb_table = (const int32_t *) |
3522 | _NL_CURRENT (LC_COLLATE, |
3523 | _NL_COLLATE_SYMB_TABLEMB); |
3524 | extra = (const unsigned char *) |
3525 | _NL_CURRENT (LC_COLLATE, |
3526 | _NL_COLLATE_SYMB_EXTRAMB); |
3527 | |
3528 | /* Locate the character in the hashing table. */ |
3529 | hash = elem_hash (str, c1); |
3530 | |
3531 | idx = 0; |
3532 | elem = hash % table_size; |
3533 | second = hash % (table_size - 2); |
3534 | while (symb_table[2 * elem] != 0) |
3535 | { |
3536 | /* First compare the hashing value. */ |
3537 | if (symb_table[2 * elem] == hash |
3538 | && c1 == extra[symb_table[2 * elem + 1]] |
3539 | && memcmp (str, |
3540 | &extra[symb_table[2 * elem + 1] |
3541 | + 1], |
3542 | c1) == 0) |
3543 | { |
3544 | /* Yep, this is the entry. */ |
3545 | idx = symb_table[2 * elem + 1]; |
3546 | idx += 1 + extra[idx]; |
3547 | break; |
3548 | } |
3549 | |
3550 | /* Next entry. */ |
3551 | elem += second; |
3552 | } |
3553 | |
3554 | if (symb_table[2 * elem] == 0) |
3555 | /* This is no valid character. */ |
3556 | FREE_STACK_RETURN (REG_ECOLLATE); |
3557 | |
3558 | /* Throw away the ] at the end of the equivalence |
3559 | class. */ |
3560 | PATFETCH (c); |
3561 | |
3562 | /* Now add the multibyte character(s) we found |
3563 | to the accept list. |
3564 | |
3565 | XXX Note that this is not entirely correct. |
3566 | we would have to match multibyte sequences |
3567 | but this is not possible with the current |
3568 | implementation. Also, we have to match |
3569 | collating symbols, which expand to more than |
3570 | one file, as a whole and not allow the |
3571 | individual bytes. */ |
3572 | c1 = extra[idx++]; |
3573 | if (c1 == 1) |
3574 | range_start = extra[idx]; |
3575 | while (c1-- > 0) |
3576 | { |
3577 | SET_LIST_BIT (extra[idx]); |
3578 | ++idx; |
3579 | } |
3580 | } |
3581 | # endif |
3582 | had_char_class = false; |
3583 | } |
3584 | else |
3585 | { |
3586 | c1++; |
3587 | while (c1--) |
3588 | PATUNFETCH; |
3589 | SET_LIST_BIT ('['); |
3590 | SET_LIST_BIT ('.'); |
3591 | range_start = '.'; |
3592 | had_char_class = false; |
3593 | } |
3594 | } |
3595 | else |
3596 | { |
3597 | had_char_class = false; |
3598 | SET_LIST_BIT (c); |
3599 | range_start = c; |
3600 | } |
3601 | } |
3602 | |
3603 | /* Discard any (non)matching list bytes that are all 0 at the |
3604 | end of the map. Decrease the map-length byte too. */ |
3605 | while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) |
3606 | b[-1]--; |
3607 | b += b[-1]; |
3608 | #endif /* WCHAR */ |
3609 | } |
3610 | break; |
3611 | |
3612 | |
3613 | case '(': |
3614 | if (syntax & RE_NO_BK_PARENS) |
3615 | goto handle_open; |
3616 | else |
3617 | goto normal_char; |
3618 | |
3619 | |
3620 | case ')': |
3621 | if (syntax & RE_NO_BK_PARENS) |
3622 | goto handle_close; |
3623 | else |
3624 | goto normal_char; |
3625 | |
3626 | |
3627 | case '\n': |
3628 | if (syntax & RE_NEWLINE_ALT) |
3629 | goto handle_alt; |
3630 | else |
3631 | goto normal_char; |
3632 | |
3633 | |
3634 | case '|': |
3635 | if (syntax & RE_NO_BK_VBAR) |
3636 | goto handle_alt; |
3637 | else |
3638 | goto normal_char; |
3639 | |
3640 | |
3641 | case '{': |
3642 | if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) |
3643 | goto handle_interval; |
3644 | else |
3645 | goto normal_char; |
3646 | |
3647 | |
3648 | case '\\': |
3649 | if (p == pend) FREE_STACK_RETURN (REG_EESCAPE); |
3650 | |
3651 | /* Do not translate the character after the \, so that we can |
3652 | distinguish, e.g., \B from \b, even if we normally would |
3653 | translate, e.g., B to b. */ |
3654 | PATFETCH_RAW (c); |
3655 | |
3656 | switch (c) |
3657 | { |
3658 | case '(': |
3659 | if (syntax & RE_NO_BK_PARENS) |
3660 | goto normal_backslash; |
3661 | |
3662 | handle_open: |
3663 | bufp->re_nsub++; |
3664 | regnum++; |
3665 | |
3666 | if (COMPILE_STACK_FULL) |
3667 | { |
3668 | RETALLOC (compile_stack.stack, compile_stack.size << 1, |
3669 | compile_stack_elt_t); |
3670 | if (compile_stack.stack == NULL) return REG_ESPACE; |
3671 | |
3672 | compile_stack.size <<= 1; |
3673 | } |
3674 | |
3675 | /* These are the values to restore when we hit end of this |
3676 | group. They are all relative offsets, so that if the |
3677 | whole pattern moves because of realloc, they will still |
3678 | be valid. */ |
3679 | COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR; |
3680 | COMPILE_STACK_TOP.fixup_alt_jump |
3681 | = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0; |
3682 | COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR; |
3683 | COMPILE_STACK_TOP.regnum = regnum; |
3684 | |
3685 | /* We will eventually replace the 0 with the number of |
3686 | groups inner to this one. But do not push a |
3687 | start_memory for groups beyond the last one we can |
3688 | represent in the compiled pattern. */ |
3689 | if (regnum <= MAX_REGNUM) |
3690 | { |
3691 | COMPILE_STACK_TOP.inner_group_offset = b |
3692 | - COMPILED_BUFFER_VAR + 2; |
3693 | BUF_PUSH_3 (start_memory, regnum, 0); |
3694 | } |
3695 | |
3696 | compile_stack.avail++; |
3697 | |
3698 | fixup_alt_jump = 0; |
3699 | laststart = 0; |
3700 | begalt = b; |
3701 | /* If we've reached MAX_REGNUM groups, then this open |
3702 | won't actually generate any code, so we'll have to |
3703 | clear pending_exact explicitly. */ |
3704 | pending_exact = 0; |
3705 | break; |
3706 | |
3707 | |
3708 | case ')': |
3709 | if (syntax & RE_NO_BK_PARENS) goto normal_backslash; |
3710 | |
3711 | if (COMPILE_STACK_EMPTY) |
3712 | { |
3713 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
3714 | goto normal_backslash; |
3715 | else |
3716 | FREE_STACK_RETURN (REG_ERPAREN); |
3717 | } |
3718 | |
3719 | handle_close: |
3720 | if (fixup_alt_jump) |
3721 | { /* Push a dummy failure point at the end of the |
3722 | alternative for a possible future |
3723 | `pop_failure_jump' to pop. See comments at |
3724 | `push_dummy_failure' in `re_match_2'. */ |
3725 | BUF_PUSH (push_dummy_failure); |
3726 | |
3727 | /* We allocated space for this jump when we assigned |
3728 | to `fixup_alt_jump', in the `handle_alt' case below. */ |
3729 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1); |
3730 | } |
3731 | |
3732 | /* See similar code for backslashed left paren above. */ |
3733 | if (COMPILE_STACK_EMPTY) |
3734 | { |
3735 | if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) |
3736 | goto normal_char; |
3737 | else |
3738 | FREE_STACK_RETURN (REG_ERPAREN); |
3739 | } |
3740 | |
3741 | /* Since we just checked for an empty stack above, this |
3742 | ``can't happen''. */ |
3743 | assert (compile_stack.avail != 0); |
3744 | { |
3745 | /* We don't just want to restore into `regnum', because |
3746 | later groups should continue to be numbered higher, |
3747 | as in `(ab)c(de)' -- the second group is #2. */ |
3748 | regnum_t this_group_regnum; |
3749 | |
3750 | compile_stack.avail--; |
3751 | begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset; |
3752 | fixup_alt_jump |
3753 | = COMPILE_STACK_TOP.fixup_alt_jump |
3754 | ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1 |
3755 | : 0; |
3756 | laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset; |
3757 | this_group_regnum = COMPILE_STACK_TOP.regnum; |
3758 | /* If we've reached MAX_REGNUM groups, then this open |
3759 | won't actually generate any code, so we'll have to |
3760 | clear pending_exact explicitly. */ |
3761 | pending_exact = 0; |
3762 | |
3763 | /* We're at the end of the group, so now we know how many |
3764 | groups were inside this one. */ |
3765 | if (this_group_regnum <= MAX_REGNUM) |
3766 | { |
3767 | UCHAR_T *inner_group_loc |
3768 | = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset; |
3769 | |
3770 | *inner_group_loc = regnum - this_group_regnum; |
3771 | BUF_PUSH_3 (stop_memory, this_group_regnum, |
3772 | regnum - this_group_regnum); |
3773 | } |
3774 | } |
3775 | break; |
3776 | |
3777 | |
3778 | case '|': /* `\|'. */ |
3779 | if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) |
3780 | goto normal_backslash; |
3781 | handle_alt: |
3782 | if (syntax & RE_LIMITED_OPS) |
3783 | goto normal_char; |
3784 | |
3785 | /* Insert before the previous alternative a jump which |
3786 | jumps to this alternative if the former fails. */ |
3787 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
3788 | INSERT_JUMP (on_failure_jump, begalt, |
3789 | b + 2 + 2 * OFFSET_ADDRESS_SIZE); |
3790 | pending_exact = 0; |
3791 | b += 1 + OFFSET_ADDRESS_SIZE; |
3792 | |
3793 | /* The alternative before this one has a jump after it |
3794 | which gets executed if it gets matched. Adjust that |
3795 | jump so it will jump to this alternative's analogous |
3796 | jump (put in below, which in turn will jump to the next |
3797 | (if any) alternative's such jump, etc.). The last such |
3798 | jump jumps to the correct final destination. A picture: |
3799 | _____ _____ |
3800 | | | | | |
3801 | | v | v |
3802 | a | b | c |
3803 | |
3804 | If we are at `b', then fixup_alt_jump right now points to a |
3805 | three-byte space after `a'. We'll put in the jump, set |
3806 | fixup_alt_jump to right after `b', and leave behind three |
3807 | bytes which we'll fill in when we get to after `c'. */ |
3808 | |
3809 | if (fixup_alt_jump) |
3810 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); |
3811 | |
3812 | /* Mark and leave space for a jump after this alternative, |
3813 | to be filled in later either by next alternative or |
3814 | when know we're at the end of a series of alternatives. */ |
3815 | fixup_alt_jump = b; |
3816 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
3817 | b += 1 + OFFSET_ADDRESS_SIZE; |
3818 | |
3819 | laststart = 0; |
3820 | begalt = b; |
3821 | break; |
3822 | |
3823 | |
3824 | case '{': |
3825 | /* If \{ is a literal. */ |
3826 | if (!(syntax & RE_INTERVALS) |
3827 | /* If we're at `\{' and it's not the open-interval |
3828 | operator. */ |
3829 | || (syntax & RE_NO_BK_BRACES)) |
3830 | goto normal_backslash; |
3831 | |
3832 | handle_interval: |
3833 | { |
3834 | /* If got here, then the syntax allows intervals. */ |
3835 | |
3836 | /* At least (most) this many matches must be made. */ |
3837 | int lower_bound = -1, upper_bound = -1; |
3838 | |
3839 | /* Place in the uncompiled pattern (i.e., just after |
3840 | the '{') to go back to if the interval is invalid. */ |
3841 | const CHAR_T *beg_interval = p; |
3842 | |
3843 | if (p == pend) |
3844 | goto invalid_interval; |
3845 | |
3846 | GET_UNSIGNED_NUMBER (lower_bound); |
3847 | |
3848 | if (c == ',') |
3849 | { |
3850 | GET_UNSIGNED_NUMBER (upper_bound); |
3851 | if (upper_bound < 0) |
3852 | upper_bound = RE_DUP_MAX; |
3853 | } |
3854 | else |
3855 | /* Interval such as `{1}' => match exactly once. */ |
3856 | upper_bound = lower_bound; |
3857 | |
3858 | if (! (0 <= lower_bound && lower_bound <= upper_bound)) |
3859 | goto invalid_interval; |
3860 | |
3861 | if (!(syntax & RE_NO_BK_BRACES)) |
3862 | { |
3863 | if (c != '\\' || p == pend) |
3864 | goto invalid_interval; |
3865 | PATFETCH (c); |
3866 | } |
3867 | |
3868 | if (c != '}') |
3869 | goto invalid_interval; |
3870 | |
3871 | /* If it's invalid to have no preceding re. */ |
3872 | if (!laststart) |
3873 | { |
3874 | if (syntax & RE_CONTEXT_INVALID_OPS |
3875 | && !(syntax & RE_INVALID_INTERVAL_ORD)) |
3876 | FREE_STACK_RETURN (REG_BADRPT); |
3877 | else if (syntax & RE_CONTEXT_INDEP_OPS) |
3878 | laststart = b; |
3879 | else |
3880 | goto unfetch_interval; |
3881 | } |
3882 | |
3883 | /* We just parsed a valid interval. */ |
3884 | |
3885 | if (RE_DUP_MAX < upper_bound) |
3886 | FREE_STACK_RETURN (REG_BADBR); |
3887 | |
3888 | /* If the upper bound is zero, don't want to succeed at |
3889 | all; jump from `laststart' to `b + 3', which will be |
3890 | the end of the buffer after we insert the jump. */ |
3891 | /* ifdef WCHAR, 'b + 1 + OFFSET_ADDRESS_SIZE' |
3892 | instead of 'b + 3'. */ |
3893 | if (upper_bound == 0) |
3894 | { |
3895 | GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE); |
3896 | INSERT_JUMP (jump, laststart, b + 1 |
3897 | + OFFSET_ADDRESS_SIZE); |
3898 | b += 1 + OFFSET_ADDRESS_SIZE; |
3899 | } |
3900 | |
3901 | /* Otherwise, we have a nontrivial interval. When |
3902 | we're all done, the pattern will look like: |
3903 | set_number_at <jump count> <upper bound> |
3904 | set_number_at <succeed_n count> <lower bound> |
3905 | succeed_n <after jump addr> <succeed_n count> |
3906 | <body of loop> |
3907 | jump_n <succeed_n addr> <jump count> |
3908 | (The upper bound and `jump_n' are omitted if |
3909 | `upper_bound' is 1, though.) */ |
3910 | else |
3911 | { /* If the upper bound is > 1, we need to insert |
3912 | more at the end of the loop. */ |
3913 | unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE + |
3914 | (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE); |
3915 | |
3916 | GET_BUFFER_SPACE (nbytes); |
3917 | |
3918 | /* Initialize lower bound of the `succeed_n', even |
3919 | though it will be set during matching by its |
3920 | attendant `set_number_at' (inserted next), |
3921 | because `re_compile_fastmap' needs to know. |
3922 | Jump to the `jump_n' we might insert below. */ |
3923 | INSERT_JUMP2 (succeed_n, laststart, |
3924 | b + 1 + 2 * OFFSET_ADDRESS_SIZE |
3925 | + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE) |
3926 | , lower_bound); |
3927 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; |
3928 | |
3929 | /* Code to initialize the lower bound. Insert |
3930 | before the `succeed_n'. The `5' is the last two |
3931 | bytes of this `set_number_at', plus 3 bytes of |
3932 | the following `succeed_n'. */ |
3933 | /* ifdef WCHAR, The '1+2*OFFSET_ADDRESS_SIZE' |
3934 | is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE' |
3935 | of the following `succeed_n'. */ |
3936 | PREFIX(insert_op2) (set_number_at, laststart, 1 |
3937 | + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b); |
3938 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; |
3939 | |
3940 | if (upper_bound > 1) |
3941 | { /* More than one repetition is allowed, so |
3942 | append a backward jump to the `succeed_n' |
3943 | that starts this interval. |
3944 | |
3945 | When we've reached this during matching, |
3946 | we'll have matched the interval once, so |
3947 | jump back only `upper_bound - 1' times. */ |
3948 | STORE_JUMP2 (jump_n, b, laststart |
3949 | + 2 * OFFSET_ADDRESS_SIZE + 1, |
3950 | upper_bound - 1); |
3951 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; |
3952 | |
3953 | /* The location we want to set is the second |
3954 | parameter of the `jump_n'; that is `b-2' as |
3955 | an absolute address. `laststart' will be |
3956 | the `set_number_at' we're about to insert; |
3957 | `laststart+3' the number to set, the source |
3958 | for the relative address. But we are |
3959 | inserting into the middle of the pattern -- |
3960 | so everything is getting moved up by 5. |
3961 | Conclusion: (b - 2) - (laststart + 3) + 5, |
3962 | i.e., b - laststart. |
3963 | |
3964 | We insert this at the beginning of the loop |
3965 | so that if we fail during matching, we'll |
3966 | reinitialize the bounds. */ |
3967 | PREFIX(insert_op2) (set_number_at, laststart, |
3968 | b - laststart, |
3969 | upper_bound - 1, b); |
3970 | b += 1 + 2 * OFFSET_ADDRESS_SIZE; |
3971 | } |
3972 | } |
3973 | pending_exact = 0; |
3974 | break; |
3975 | |
3976 | invalid_interval: |
3977 | if (!(syntax & RE_INVALID_INTERVAL_ORD)) |
3978 | FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR); |
3979 | unfetch_interval: |
3980 | /* Match the characters as literals. */ |
3981 | p = beg_interval; |
3982 | c = '{'; |
3983 | if (syntax & RE_NO_BK_BRACES) |
3984 | goto normal_char; |
3985 | else |
3986 | goto normal_backslash; |
3987 | } |
3988 | |
3989 | #ifdef emacs |
3990 | /* There is no way to specify the before_dot and after_dot |
3991 | operators. rms says this is ok. --karl */ |
3992 | case '=': |
3993 | BUF_PUSH (at_dot); |
3994 | break; |
3995 | |
3996 | case 's': |
3997 | laststart = b; |
3998 | PATFETCH (c); |
3999 | BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]); |
4000 | break; |
4001 | |
4002 | case 'S': |
4003 | laststart = b; |
4004 | PATFETCH (c); |
4005 | BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]); |
4006 | break; |
4007 | #endif /* emacs */ |
4008 | |
4009 | |
4010 | case 'w': |
4011 | if (syntax & RE_NO_GNU_OPS) |
4012 | goto normal_char; |
4013 | laststart = b; |
4014 | BUF_PUSH (wordchar); |
4015 | break; |
4016 | |
4017 | |
4018 | case 'W': |
4019 | if (syntax & RE_NO_GNU_OPS) |
4020 | goto normal_char; |
4021 | laststart = b; |
4022 | BUF_PUSH (notwordchar); |
4023 | break; |
4024 | |
4025 | |
4026 | case '<': |
4027 | if (syntax & RE_NO_GNU_OPS) |
4028 | goto normal_char; |
4029 | BUF_PUSH (wordbeg); |
4030 | break; |
4031 | |
4032 | case '>': |
4033 | if (syntax & RE_NO_GNU_OPS) |
4034 | goto normal_char; |
4035 | BUF_PUSH (wordend); |
4036 | break; |
4037 | |
4038 | case 'b': |
4039 | if (syntax & RE_NO_GNU_OPS) |
4040 | goto normal_char; |
4041 | BUF_PUSH (wordbound); |
4042 | break; |
4043 | |
4044 | case 'B': |
4045 | if (syntax & RE_NO_GNU_OPS) |
4046 | goto normal_char; |
4047 | BUF_PUSH (notwordbound); |
4048 | break; |
4049 | |
4050 | case '`': |
4051 | if (syntax & RE_NO_GNU_OPS) |
4052 | goto normal_char; |
4053 | BUF_PUSH (begbuf); |
4054 | break; |
4055 | |
4056 | case '\'': |
4057 | if (syntax & RE_NO_GNU_OPS) |
4058 | goto normal_char; |
4059 | BUF_PUSH (endbuf); |
4060 | break; |
4061 | |
4062 | case '1': case '2': case '3': case '4': case '5': |
4063 | case '6': case '7': case '8': case '9': |
4064 | if (syntax & RE_NO_BK_REFS) |
4065 | goto normal_char; |
4066 | |
4067 | c1 = c - '0'; |
4068 | |
4069 | if (c1 > regnum) |
4070 | FREE_STACK_RETURN (REG_ESUBREG); |
4071 | |
4072 | /* Can't back reference to a subexpression if inside of it. */ |
4073 | if (group_in_compile_stack (compile_stack, (regnum_t) c1)) |
4074 | goto normal_char; |
4075 | |
4076 | laststart = b; |
4077 | BUF_PUSH_2 (duplicate, c1); |
4078 | break; |
4079 | |
4080 | |
4081 | case '+': |
4082 | case '?': |
4083 | if (syntax & RE_BK_PLUS_QM) |
4084 | goto handle_plus; |
4085 | else |
4086 | goto normal_backslash; |
4087 | |
4088 | default: |
4089 | normal_backslash: |
4090 | /* You might think it would be useful for \ to mean |
4091 | not to translate; but if we don't translate it |
4092 | it will never match anything. */ |
4093 | c = TRANSLATE (c); |
4094 | goto normal_char; |
4095 | } |
4096 | break; |
4097 | |
4098 | |
4099 | default: |
4100 | /* Expects the character in `c'. */ |
4101 | normal_char: |
4102 | /* If no exactn currently being built. */ |
4103 | if (!pending_exact |
4104 | #ifdef WCHAR |
4105 | /* If last exactn handle binary(or character) and |
4106 | new exactn handle character(or binary). */ |
4107 | || is_exactn_bin != is_binary[p - 1 - pattern] |
4108 | #endif /* WCHAR */ |
4109 | |
4110 | /* If last exactn not at current position. */ |
4111 | || pending_exact + *pending_exact + 1 != b |
4112 | |
4113 | /* We have only one byte following the exactn for the count. */ |
4114 | || *pending_exact == (1 << BYTEWIDTH) - 1 |
4115 | |
4116 | /* If followed by a repetition operator. */ |
4117 | || *p == '*' || *p == '^' |
4118 | || ((syntax & RE_BK_PLUS_QM) |
4119 | ? *p == '\\' && (p[1] == '+' || p[1] == '?') |
4120 | : (*p == '+' || *p == '?')) |
4121 | || ((syntax & RE_INTERVALS) |
4122 | && ((syntax & RE_NO_BK_BRACES) |
4123 | ? *p == '{' |
4124 | : (p[0] == '\\' && p[1] == '{')))) |
4125 | { |
4126 | /* Start building a new exactn. */ |
4127 | |
4128 | laststart = b; |
4129 | |
4130 | #ifdef WCHAR |
4131 | /* Is this exactn binary data or character? */ |
4132 | is_exactn_bin = is_binary[p - 1 - pattern]; |
4133 | if (is_exactn_bin) |
4134 | BUF_PUSH_2 (exactn_bin, 0); |
4135 | else |
4136 | BUF_PUSH_2 (exactn, 0); |
4137 | #else |
4138 | BUF_PUSH_2 (exactn, 0); |
4139 | #endif /* WCHAR */ |
4140 | pending_exact = b - 1; |
4141 | } |
4142 | |
4143 | BUF_PUSH (c); |
4144 | (*pending_exact)++; |
4145 | break; |
4146 | } /* switch (c) */ |
4147 | } /* while p != pend */ |
4148 | |
4149 | |
4150 | /* Through the pattern now. */ |
4151 | |
4152 | if (fixup_alt_jump) |
4153 | STORE_JUMP (jump_past_alt, fixup_alt_jump, b); |
4154 | |
4155 | if (!COMPILE_STACK_EMPTY) |
4156 | FREE_STACK_RETURN (REG_EPAREN); |
4157 | |
4158 | /* If we don't want backtracking, force success |
4159 | the first time we reach the end of the compiled pattern. */ |
4160 | if (syntax & RE_NO_POSIX_BACKTRACKING) |
4161 | BUF_PUSH (succeed); |
4162 | |
4163 | #ifdef WCHAR |
4164 | free (pattern); |
4165 | free (mbs_offset); |
4166 | free (is_binary); |
4167 | #endif |
4168 | free (compile_stack.stack); |
4169 | |
4170 | /* We have succeeded; set the length of the buffer. */ |
4171 | #ifdef WCHAR |
4172 | bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR; |
4173 | #else |
4174 | bufp->used = b - bufp->buffer; |
4175 | #endif |
4176 | |
4177 | #ifdef DEBUG |
4178 | if (debug) |
4179 | { |
4180 | DEBUG_PRINT1 ("\nCompiled pattern: \n" ); |
4181 | PREFIX(print_compiled_pattern) (bufp); |
4182 | } |
4183 | #endif /* DEBUG */ |
4184 | |
4185 | #ifndef MATCH_MAY_ALLOCATE |
4186 | /* Initialize the failure stack to the largest possible stack. This |
4187 | isn't necessary unless we're trying to avoid calling alloca in |
4188 | the search and match routines. */ |
4189 | { |
4190 | int num_regs = bufp->re_nsub + 1; |
4191 | |
4192 | /* Since DOUBLE_FAIL_STACK refuses to double only if the current size |
4193 | is strictly greater than re_max_failures, the largest possible stack |
4194 | is 2 * re_max_failures failure points. */ |
4195 | if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) |
4196 | { |
4197 | fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); |
4198 | |
4199 | # ifdef emacs |
4200 | if (! fail_stack.stack) |
4201 | fail_stack.stack |
4202 | = (PREFIX(fail_stack_elt_t) *) xmalloc (fail_stack.size |
4203 | * sizeof (PREFIX(fail_stack_elt_t))); |
4204 | else |
4205 | fail_stack.stack |
4206 | = (PREFIX(fail_stack_elt_t) *) xrealloc (fail_stack.stack, |
4207 | (fail_stack.size |
4208 | * sizeof (PREFIX(fail_stack_elt_t)))); |
4209 | # else /* not emacs */ |
4210 | if (! fail_stack.stack) |
4211 | fail_stack.stack |
4212 | = (PREFIX(fail_stack_elt_t) *) malloc (fail_stack.size |
4213 | * sizeof (PREFIX(fail_stack_elt_t))); |
4214 | else |
4215 | fail_stack.stack |
4216 | = (PREFIX(fail_stack_elt_t) *) realloc (fail_stack.stack, |
4217 | (fail_stack.size |
4218 | * sizeof (PREFIX(fail_stack_elt_t)))); |
4219 | # endif /* not emacs */ |
4220 | } |
4221 | |
4222 | PREFIX(regex_grow_registers) (num_regs); |
4223 | } |
4224 | #endif /* not MATCH_MAY_ALLOCATE */ |
4225 | |
4226 | return REG_NOERROR; |
4227 | } /* regex_compile */ |
4228 | |
4229 | /* Subroutines for `regex_compile'. */ |
4230 | |
4231 | /* Store OP at LOC followed by two-byte integer parameter ARG. */ |
4232 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ |
4233 | |
4234 | static void |
4235 | PREFIX(store_op1) (re_opcode_t op, UCHAR_T *loc, int arg) |
4236 | { |
4237 | *loc = (UCHAR_T) op; |
4238 | STORE_NUMBER (loc + 1, arg); |
4239 | } |
4240 | |
4241 | |
4242 | /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ |
4243 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ |
4244 | |
4245 | static void |
4246 | PREFIX(store_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, int arg2) |
4247 | { |
4248 | *loc = (UCHAR_T) op; |
4249 | STORE_NUMBER (loc + 1, arg1); |
4250 | STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2); |
4251 | } |
4252 | |
4253 | |
4254 | /* Copy the bytes from LOC to END to open up three bytes of space at LOC |
4255 | for OP followed by two-byte integer parameter ARG. */ |
4256 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ |
4257 | |
4258 | static void |
4259 | PREFIX(insert_op1) (re_opcode_t op, UCHAR_T *loc, int arg, UCHAR_T *end) |
4260 | { |
4261 | register UCHAR_T *pfrom = end; |
4262 | register UCHAR_T *pto = end + 1 + OFFSET_ADDRESS_SIZE; |
4263 | |
4264 | while (pfrom != loc) |
4265 | *--pto = *--pfrom; |
4266 | |
4267 | PREFIX(store_op1) (op, loc, arg); |
4268 | } |
4269 | |
4270 | |
4271 | /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ |
4272 | /* ifdef WCHAR, integer parameter is 1 wchar_t. */ |
4273 | |
4274 | static void |
4275 | PREFIX(insert_op2) (re_opcode_t op, UCHAR_T *loc, int arg1, |
4276 | int arg2, UCHAR_T *end) |
4277 | { |
4278 | register UCHAR_T *pfrom = end; |
4279 | register UCHAR_T *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE; |
4280 | |
4281 | while (pfrom != loc) |
4282 | *--pto = *--pfrom; |
4283 | |
4284 | PREFIX(store_op2) (op, loc, arg1, arg2); |
4285 | } |
4286 | |
4287 | |
4288 | /* P points to just after a ^ in PATTERN. Return true if that ^ comes |
4289 | after an alternative or a begin-subexpression. We assume there is at |
4290 | least one character before the ^. */ |
4291 | |
4292 | static boolean |
4293 | PREFIX(at_begline_loc_p) (const CHAR_T *pattern, const CHAR_T *p, |
4294 | reg_syntax_t syntax) |
4295 | { |
4296 | const CHAR_T *prev = p - 2; |
4297 | boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; |
4298 | |
4299 | return |
4300 | /* After a subexpression? */ |
4301 | (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) |
4302 | /* After an alternative? */ |
4303 | || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); |
4304 | } |
4305 | |
4306 | |
4307 | /* The dual of at_begline_loc_p. This one is for $. We assume there is |
4308 | at least one character after the $, i.e., `P < PEND'. */ |
4309 | |
4310 | static boolean |
4311 | PREFIX(at_endline_loc_p) (const CHAR_T *p, const CHAR_T *pend, |
4312 | reg_syntax_t syntax) |
4313 | { |
4314 | const CHAR_T *next = p; |
4315 | boolean next_backslash = *next == '\\'; |
4316 | const CHAR_T *next_next = p + 1 < pend ? p + 1 : 0; |
4317 | |
4318 | return |
4319 | /* Before a subexpression? */ |
4320 | (syntax & RE_NO_BK_PARENS ? *next == ')' |
4321 | : next_backslash && next_next && *next_next == ')') |
4322 | /* Before an alternative? */ |
4323 | || (syntax & RE_NO_BK_VBAR ? *next == '|' |
4324 | : next_backslash && next_next && *next_next == '|'); |
4325 | } |
4326 | |
4327 | #else /* not INSIDE_RECURSION */ |
4328 | |
4329 | /* Returns true if REGNUM is in one of COMPILE_STACK's elements and |
4330 | false if it's not. */ |
4331 | |
4332 | static boolean |
4333 | group_in_compile_stack (compile_stack_type compile_stack, regnum_t regnum) |
4334 | { |
4335 | int this_element; |
4336 | |
4337 | for (this_element = compile_stack.avail - 1; |
4338 | this_element >= 0; |
4339 | this_element--) |
4340 | if (compile_stack.stack[this_element].regnum == regnum) |
4341 | return true; |
4342 | |
4343 | return false; |
4344 | } |
4345 | #endif /* not INSIDE_RECURSION */ |
4346 | |
4347 | #ifdef INSIDE_RECURSION |
4348 | |
4349 | #ifdef WCHAR |
4350 | /* This insert space, which size is "num", into the pattern at "loc". |
4351 | "end" must point the end of the allocated buffer. */ |
4352 | static void |
4353 | insert_space (int num, CHAR_T *loc, CHAR_T *end) |
4354 | { |
4355 | register CHAR_T *pto = end; |
4356 | register CHAR_T *pfrom = end - num; |
4357 | |
4358 | while (pfrom >= loc) |
4359 | *pto-- = *pfrom--; |
4360 | } |
4361 | #endif /* WCHAR */ |
4362 | |
4363 | #ifdef WCHAR |
4364 | static reg_errcode_t |
4365 | wcs_compile_range (CHAR_T range_start_char, const CHAR_T **p_ptr, |
4366 | const CHAR_T *pend, RE_TRANSLATE_TYPE translate, |
4367 | reg_syntax_t syntax, CHAR_T *b, CHAR_T *char_set) |
4368 | { |
4369 | const CHAR_T *p = *p_ptr; |
4370 | CHAR_T range_start, range_end; |
4371 | reg_errcode_t ret; |
4372 | # ifdef _LIBC |
4373 | uint32_t nrules; |
4374 | uint32_t start_val, end_val; |
4375 | # endif |
4376 | if (p == pend) |
4377 | return REG_ERANGE; |
4378 | |
4379 | # ifdef _LIBC |
4380 | nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); |
4381 | if (nrules != 0) |
4382 | { |
4383 | const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE, |
4384 | _NL_COLLATE_COLLSEQWC); |
4385 | const unsigned char *extra = (const unsigned char *) |
4386 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); |
4387 | |
4388 | if (range_start_char < -1) |
4389 | { |
4390 | /* range_start is a collating symbol. */ |
4391 | int32_t *wextra; |
4392 | /* Retreive the index and get collation sequence value. */ |
4393 | wextra = (int32_t*)(extra + char_set[-range_start_char]); |
4394 | start_val = wextra[1 + *wextra]; |
4395 | } |
4396 | else |
4397 | start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char)); |
4398 | |
4399 | end_val = collseq_table_lookup (collseq, TRANSLATE (p[0])); |
4400 | |
4401 | /* Report an error if the range is empty and the syntax prohibits |
4402 | this. */ |
4403 | ret = ((syntax & RE_NO_EMPTY_RANGES) |
4404 | && (start_val > end_val))? REG_ERANGE : REG_NOERROR; |
4405 | |
4406 | /* Insert space to the end of the char_ranges. */ |
4407 | insert_space(2, b - char_set[5] - 2, b - 1); |
4408 | *(b - char_set[5] - 2) = (wchar_t)start_val; |
4409 | *(b - char_set[5] - 1) = (wchar_t)end_val; |
4410 | char_set[4]++; /* ranges_index */ |
4411 | } |
4412 | else |
4413 | # endif |
4414 | { |
4415 | range_start = (range_start_char >= 0)? TRANSLATE (range_start_char): |
4416 | range_start_char; |
4417 | range_end = TRANSLATE (p[0]); |
4418 | /* Report an error if the range is empty and the syntax prohibits |
4419 | this. */ |
4420 | ret = ((syntax & RE_NO_EMPTY_RANGES) |
4421 | && (range_start > range_end))? REG_ERANGE : REG_NOERROR; |
4422 | |
4423 | /* Insert space to the end of the char_ranges. */ |
4424 | insert_space(2, b - char_set[5] - 2, b - 1); |
4425 | *(b - char_set[5] - 2) = range_start; |
4426 | *(b - char_set[5] - 1) = range_end; |
4427 | char_set[4]++; /* ranges_index */ |
4428 | } |
4429 | /* Have to increment the pointer into the pattern string, so the |
4430 | caller isn't still at the ending character. */ |
4431 | (*p_ptr)++; |
4432 | |
4433 | return ret; |
4434 | } |
4435 | #else /* BYTE */ |
4436 | /* Read the ending character of a range (in a bracket expression) from the |
4437 | uncompiled pattern *P_PTR (which ends at PEND). We assume the |
4438 | starting character is in `P[-2]'. (`P[-1]' is the character `-'.) |
4439 | Then we set the translation of all bits between the starting and |
4440 | ending characters (inclusive) in the compiled pattern B. |
4441 | |
4442 | Return an error code. |
4443 | |
4444 | We use these short variable names so we can use the same macros as |
4445 | `regex_compile' itself. */ |
4446 | |
4447 | static reg_errcode_t |
4448 | byte_compile_range (unsigned int range_start_char, const char **p_ptr, |
4449 | const char *pend, RE_TRANSLATE_TYPE translate, |
4450 | reg_syntax_t syntax, unsigned char *b) |
4451 | { |
4452 | unsigned this_char; |
4453 | const char *p = *p_ptr; |
4454 | reg_errcode_t ret; |
4455 | # if _LIBC |
4456 | const unsigned char *collseq; |
4457 | unsigned int start_colseq; |
4458 | unsigned int end_colseq; |
4459 | # else |
4460 | unsigned end_char; |
4461 | # endif |
4462 | |
4463 | if (p == pend) |
4464 | return REG_ERANGE; |
4465 | |
4466 | /* Have to increment the pointer into the pattern string, so the |
4467 | caller isn't still at the ending character. */ |
4468 | (*p_ptr)++; |
4469 | |
4470 | /* Report an error if the range is empty and the syntax prohibits this. */ |
4471 | ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; |
4472 | |
4473 | # if _LIBC |
4474 | collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE, |
4475 | _NL_COLLATE_COLLSEQMB); |
4476 | |
4477 | start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)]; |
4478 | end_colseq = collseq[(unsigned char) TRANSLATE (p[0])]; |
4479 | for (this_char = 0; this_char <= (unsigned char) -1; ++this_char) |
4480 | { |
4481 | unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)]; |
4482 | |
4483 | if (start_colseq <= this_colseq && this_colseq <= end_colseq) |
4484 | { |
4485 | SET_LIST_BIT (TRANSLATE (this_char)); |
4486 | ret = REG_NOERROR; |
4487 | } |
4488 | } |
4489 | # else |
4490 | /* Here we see why `this_char' has to be larger than an `unsigned |
4491 | char' -- we would otherwise go into an infinite loop, since all |
4492 | characters <= 0xff. */ |
4493 | range_start_char = TRANSLATE (range_start_char); |
4494 | /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE, |
4495 | and some compilers cast it to int implicitly, so following for_loop |
4496 | may fall to (almost) infinite loop. |
4497 | e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff. |
4498 | To avoid this, we cast p[0] to unsigned int and truncate it. */ |
4499 | end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1)); |
4500 | |
4501 | for (this_char = range_start_char; this_char <= end_char; ++this_char) |
4502 | { |
4503 | SET_LIST_BIT (TRANSLATE (this_char)); |
4504 | ret = REG_NOERROR; |
4505 | } |
4506 | # endif |
4507 | |
4508 | return ret; |
4509 | } |
4510 | #endif /* WCHAR */ |
4511 | |
4512 | /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in |
4513 | BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible |
4514 | characters can start a string that matches the pattern. This fastmap |
4515 | is used by re_search to skip quickly over impossible starting points. |
4516 | |
4517 | The caller must supply the address of a (1 << BYTEWIDTH)-byte data |
4518 | area as BUFP->fastmap. |
4519 | |
4520 | We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in |
4521 | the pattern buffer. |
4522 | |
4523 | Returns 0 if we succeed, -2 if an internal error. */ |
4524 | |
4525 | #ifdef WCHAR |
4526 | /* local function for re_compile_fastmap. |
4527 | truncate wchar_t character to char. */ |
4528 | static unsigned char truncate_wchar (CHAR_T c); |
4529 | |
4530 | static unsigned char |
4531 | truncate_wchar (CHAR_T c) |
4532 | { |
4533 | unsigned char buf[MB_CUR_MAX]; |
4534 | mbstate_t state; |
4535 | int retval; |
4536 | memset (&state, '\0', sizeof (state)); |
4537 | # ifdef _LIBC |
4538 | retval = __wcrtomb (buf, c, &state); |
4539 | # else |
4540 | retval = wcrtomb (buf, c, &state); |
4541 | # endif |
4542 | return retval > 0 ? buf[0] : (unsigned char) c; |
4543 | } |
4544 | #endif /* WCHAR */ |
4545 | |
4546 | static int |
4547 | PREFIX(re_compile_fastmap) (struct re_pattern_buffer *bufp) |
4548 | { |
4549 | int j, k; |
4550 | #ifdef MATCH_MAY_ALLOCATE |
4551 | PREFIX(fail_stack_type) fail_stack; |
4552 | #endif |
4553 | #ifndef REGEX_MALLOC |
4554 | char *destination; |
4555 | #endif |
4556 | |
4557 | register char *fastmap = bufp->fastmap; |
4558 | |
4559 | #ifdef WCHAR |
4560 | /* We need to cast pattern to (wchar_t*), because we casted this compiled |
4561 | pattern to (char*) in regex_compile. */ |
4562 | UCHAR_T *pattern = (UCHAR_T*)bufp->buffer; |
4563 | register UCHAR_T *pend = (UCHAR_T*) (bufp->buffer + bufp->used); |
4564 | #else /* BYTE */ |
4565 | UCHAR_T *pattern = bufp->buffer; |
4566 | register UCHAR_T *pend = pattern + bufp->used; |
4567 | #endif /* WCHAR */ |
4568 | UCHAR_T *p = pattern; |
4569 | |
4570 | #ifdef REL_ALLOC |
4571 | /* This holds the pointer to the failure stack, when |
4572 | it is allocated relocatably. */ |
4573 | fail_stack_elt_t *failure_stack_ptr; |
4574 | #endif |
4575 | |
4576 | /* Assume that each path through the pattern can be null until |
4577 | proven otherwise. We set this false at the bottom of switch |
4578 | statement, to which we get only if a particular path doesn't |
4579 | match the empty string. */ |
4580 | boolean path_can_be_null = true; |
4581 | |
4582 | /* We aren't doing a `succeed_n' to begin with. */ |
4583 | boolean succeed_n_p = false; |
4584 | |
4585 | assert (fastmap != NULL && p != NULL); |
4586 | |
4587 | INIT_FAIL_STACK (); |
4588 | bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ |
4589 | bufp->fastmap_accurate = 1; /* It will be when we're done. */ |
4590 | bufp->can_be_null = 0; |
4591 | |
4592 | while (1) |
4593 | { |
4594 | if (p == pend || *p == (UCHAR_T) succeed) |
4595 | { |
4596 | /* We have reached the (effective) end of pattern. */ |
4597 | if (!FAIL_STACK_EMPTY ()) |
4598 | { |
4599 | bufp->can_be_null |= path_can_be_null; |
4600 | |
4601 | /* Reset for next path. */ |
4602 | path_can_be_null = true; |
4603 | |
4604 | p = fail_stack.stack[--fail_stack.avail].pointer; |
4605 | |
4606 | continue; |
4607 | } |
4608 | else |
4609 | break; |
4610 | } |
4611 | |
4612 | /* We should never be about to go beyond the end of the pattern. */ |
4613 | assert (p < pend); |
4614 | |
4615 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
4616 | { |
4617 | |
4618 | /* I guess the idea here is to simply not bother with a fastmap |
4619 | if a backreference is used, since it's too hard to figure out |
4620 | the fastmap for the corresponding group. Setting |
4621 | `can_be_null' stops `re_search_2' from using the fastmap, so |
4622 | that is all we do. */ |
4623 | case duplicate: |
4624 | bufp->can_be_null = 1; |
4625 | goto done; |
4626 | |
4627 | |
4628 | /* Following are the cases which match a character. These end |
4629 | with `break'. */ |
4630 | |
4631 | #ifdef WCHAR |
4632 | case exactn: |
4633 | fastmap[truncate_wchar(p[1])] = 1; |
4634 | break; |
4635 | #else /* BYTE */ |
4636 | case exactn: |
4637 | fastmap[p[1]] = 1; |
4638 | break; |
4639 | #endif /* WCHAR */ |
4640 | #ifdef MBS_SUPPORT |
4641 | case exactn_bin: |
4642 | fastmap[p[1]] = 1; |
4643 | break; |
4644 | #endif |
4645 | |
4646 | #ifdef WCHAR |
4647 | /* It is hard to distinguish fastmap from (multi byte) characters |
4648 | which depends on current locale. */ |
4649 | case charset: |
4650 | case charset_not: |
4651 | case wordchar: |
4652 | case notwordchar: |
4653 | bufp->can_be_null = 1; |
4654 | goto done; |
4655 | #else /* BYTE */ |
4656 | case charset: |
4657 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) |
4658 | if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) |
4659 | fastmap[j] = 1; |
4660 | break; |
4661 | |
4662 | |
4663 | case charset_not: |
4664 | /* Chars beyond end of map must be allowed. */ |
4665 | for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) |
4666 | fastmap[j] = 1; |
4667 | |
4668 | for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) |
4669 | if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) |
4670 | fastmap[j] = 1; |
4671 | break; |
4672 | |
4673 | |
4674 | case wordchar: |
4675 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
4676 | if (SYNTAX (j) == Sword) |
4677 | fastmap[j] = 1; |
4678 | break; |
4679 | |
4680 | |
4681 | case notwordchar: |
4682 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
4683 | if (SYNTAX (j) != Sword) |
4684 | fastmap[j] = 1; |
4685 | break; |
4686 | #endif /* WCHAR */ |
4687 | |
4688 | case anychar: |
4689 | { |
4690 | int fastmap_newline = fastmap['\n']; |
4691 | |
4692 | /* `.' matches anything ... */ |
4693 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
4694 | fastmap[j] = 1; |
4695 | |
4696 | /* ... except perhaps newline. */ |
4697 | if (!(bufp->syntax & RE_DOT_NEWLINE)) |
4698 | fastmap['\n'] = fastmap_newline; |
4699 | |
4700 | /* Return if we have already set `can_be_null'; if we have, |
4701 | then the fastmap is irrelevant. Something's wrong here. */ |
4702 | else if (bufp->can_be_null) |
4703 | goto done; |
4704 | |
4705 | /* Otherwise, have to check alternative paths. */ |
4706 | break; |
4707 | } |
4708 | |
4709 | #ifdef emacs |
4710 | case syntaxspec: |
4711 | k = *p++; |
4712 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
4713 | if (SYNTAX (j) == (enum syntaxcode) k) |
4714 | fastmap[j] = 1; |
4715 | break; |
4716 | |
4717 | |
4718 | case notsyntaxspec: |
4719 | k = *p++; |
4720 | for (j = 0; j < (1 << BYTEWIDTH); j++) |
4721 | if (SYNTAX (j) != (enum syntaxcode) k) |
4722 | fastmap[j] = 1; |
4723 | break; |
4724 | |
4725 | |
4726 | /* All cases after this match the empty string. These end with |
4727 | `continue'. */ |
4728 | |
4729 | |
4730 | case before_dot: |
4731 | case at_dot: |
4732 | case after_dot: |
4733 | continue; |
4734 | #endif /* emacs */ |
4735 | |
4736 | |
4737 | case no_op: |
4738 | case begline: |
4739 | case endline: |
4740 | case begbuf: |
4741 | case endbuf: |
4742 | case wordbound: |
4743 | case notwordbound: |
4744 | case wordbeg: |
4745 | case wordend: |
4746 | case push_dummy_failure: |
4747 | continue; |
4748 | |
4749 | |
4750 | case jump_n: |
4751 | case pop_failure_jump: |
4752 | case maybe_pop_jump: |
4753 | case jump: |
4754 | case jump_past_alt: |
4755 | case dummy_failure_jump: |
4756 | EXTRACT_NUMBER_AND_INCR (j, p); |
4757 | p += j; |
4758 | if (j > 0) |
4759 | continue; |
4760 | |
4761 | /* Jump backward implies we just went through the body of a |
4762 | loop and matched nothing. Opcode jumped to should be |
4763 | `on_failure_jump' or `succeed_n'. Just treat it like an |
4764 | ordinary jump. For a * loop, it has pushed its failure |
4765 | point already; if so, discard that as redundant. */ |
4766 | if ((re_opcode_t) *p != on_failure_jump |
4767 | && (re_opcode_t) *p != succeed_n) |
4768 | continue; |
4769 | |
4770 | p++; |
4771 | EXTRACT_NUMBER_AND_INCR (j, p); |
4772 | p += j; |
4773 | |
4774 | /* If what's on the stack is where we are now, pop it. */ |
4775 | if (!FAIL_STACK_EMPTY () |
4776 | && fail_stack.stack[fail_stack.avail - 1].pointer == p) |
4777 | fail_stack.avail--; |
4778 | |
4779 | continue; |
4780 | |
4781 | |
4782 | case on_failure_jump: |
4783 | case on_failure_keep_string_jump: |
4784 | handle_on_failure_jump: |
4785 | EXTRACT_NUMBER_AND_INCR (j, p); |
4786 | |
4787 | /* For some patterns, e.g., `(a?)?', `p+j' here points to the |
4788 | end of the pattern. We don't want to push such a point, |
4789 | since when we restore it above, entering the switch will |
4790 | increment `p' past the end of the pattern. We don't need |
4791 | to push such a point since we obviously won't find any more |
4792 | fastmap entries beyond `pend'. Such a pattern can match |
4793 | the null string, though. */ |
4794 | if (p + j < pend) |
4795 | { |
4796 | if (!PUSH_PATTERN_OP (p + j, fail_stack)) |
4797 | { |
4798 | RESET_FAIL_STACK (); |
4799 | return -2; |
4800 | } |
4801 | } |
4802 | else |
4803 | bufp->can_be_null = 1; |
4804 | |
4805 | if (succeed_n_p) |
4806 | { |
4807 | EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */ |
4808 | succeed_n_p = false; |
4809 | } |
4810 | |
4811 | continue; |
4812 | |
4813 | |
4814 | case succeed_n: |
4815 | /* Get to the number of times to succeed. */ |
4816 | p += OFFSET_ADDRESS_SIZE; |
4817 | |
4818 | /* Increment p past the n for when k != 0. */ |
4819 | EXTRACT_NUMBER_AND_INCR (k, p); |
4820 | if (k == 0) |
4821 | { |
4822 | p -= 2 * OFFSET_ADDRESS_SIZE; |
4823 | succeed_n_p = true; /* Spaghetti code alert. */ |
4824 | goto handle_on_failure_jump; |
4825 | } |
4826 | continue; |
4827 | |
4828 | |
4829 | case set_number_at: |
4830 | p += 2 * OFFSET_ADDRESS_SIZE; |
4831 | continue; |
4832 | |
4833 | |
4834 | case start_memory: |
4835 | case stop_memory: |
4836 | p += 2; |
4837 | continue; |
4838 | |
4839 | |
4840 | default: |
4841 | abort (); /* We have listed all the cases. */ |
4842 | } /* switch *p++ */ |
4843 | |
4844 | /* Getting here means we have found the possible starting |
4845 | characters for one path of the pattern -- and that the empty |
4846 | string does not match. We need not follow this path further. |
4847 | Instead, look at the next alternative (remembered on the |
4848 | stack), or quit if no more. The test at the top of the loop |
4849 | does these things. */ |
4850 | path_can_be_null = false; |
4851 | p = pend; |
4852 | } /* while p */ |
4853 | |
4854 | /* Set `can_be_null' for the last path (also the first path, if the |
4855 | pattern is empty). */ |
4856 | bufp->can_be_null |= path_can_be_null; |
4857 | |
4858 | done: |
4859 | RESET_FAIL_STACK (); |
4860 | return 0; |
4861 | } |
4862 | |
4863 | #else /* not INSIDE_RECURSION */ |
4864 | |
4865 | int |
4866 | re_compile_fastmap (struct re_pattern_buffer *bufp) |
4867 | { |
4868 | # ifdef MBS_SUPPORT |
4869 | if (MB_CUR_MAX != 1) |
4870 | return wcs_re_compile_fastmap(bufp); |
4871 | else |
4872 | # endif |
4873 | return byte_re_compile_fastmap(bufp); |
4874 | } /* re_compile_fastmap */ |
4875 | #ifdef _LIBC |
4876 | weak_alias (__re_compile_fastmap, re_compile_fastmap) |
4877 | #endif |
4878 | |
4879 | |
4880 | /* Set REGS to hold NUM_REGS registers, storing them in STARTS and |
4881 | ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use |
4882 | this memory for recording register information. STARTS and ENDS |
4883 | must be allocated using the malloc library routine, and must each |
4884 | be at least NUM_REGS * sizeof (regoff_t) bytes long. |
4885 | |
4886 | If NUM_REGS == 0, then subsequent matches should allocate their own |
4887 | register data. |
4888 | |
4889 | Unless this function is called, the first search or match using |
4890 | PATTERN_BUFFER will allocate its own register data, without |
4891 | freeing the old data. */ |
4892 | |
4893 | void |
4894 | re_set_registers (struct re_pattern_buffer *bufp, |
4895 | struct re_registers *regs, unsigned num_regs, |
4896 | regoff_t *starts, regoff_t *ends) |
4897 | { |
4898 | if (num_regs) |
4899 | { |
4900 | bufp->regs_allocated = REGS_REALLOCATE; |
4901 | regs->num_regs = num_regs; |
4902 | regs->start = starts; |
4903 | regs->end = ends; |
4904 | } |
4905 | else |
4906 | { |
4907 | bufp->regs_allocated = REGS_UNALLOCATED; |
4908 | regs->num_regs = 0; |
4909 | regs->start = regs->end = (regoff_t *) 0; |
4910 | } |
4911 | } |
4912 | #ifdef _LIBC |
4913 | weak_alias (__re_set_registers, re_set_registers) |
4914 | #endif |
4915 | |
4916 | /* Searching routines. */ |
4917 | |
4918 | /* Like re_search_2, below, but only one string is specified, and |
4919 | doesn't let you say where to stop matching. */ |
4920 | |
4921 | int |
4922 | re_search (struct re_pattern_buffer *bufp, const char *string, int size, |
4923 | int startpos, int range, struct re_registers *regs) |
4924 | { |
4925 | return re_search_2 (buffer: bufp, NULL, length1: 0, string2: string, length2: size, start: startpos, range, |
4926 | regs, stop: size); |
4927 | } |
4928 | #ifdef _LIBC |
4929 | weak_alias (__re_search, re_search) |
4930 | #endif |
4931 | |
4932 | |
4933 | /* Using the compiled pattern in BUFP->buffer, first tries to match the |
4934 | virtual concatenation of STRING1 and STRING2, starting first at index |
4935 | STARTPOS, then at STARTPOS + 1, and so on. |
4936 | |
4937 | STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. |
4938 | |
4939 | RANGE is how far to scan while trying to match. RANGE = 0 means try |
4940 | only at STARTPOS; in general, the last start tried is STARTPOS + |
4941 | RANGE. |
4942 | |
4943 | In REGS, return the indices of the virtual concatenation of STRING1 |
4944 | and STRING2 that matched the entire BUFP->buffer and its contained |
4945 | subexpressions. |
4946 | |
4947 | Do not consider matching one past the index STOP in the virtual |
4948 | concatenation of STRING1 and STRING2. |
4949 | |
4950 | We return either the position in the strings at which the match was |
4951 | found, -1 if no match, or -2 if error (such as failure |
4952 | stack overflow). */ |
4953 | |
4954 | int |
4955 | re_search_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, |
4956 | const char *string2, int size2, int startpos, int range, |
4957 | struct re_registers *regs, int stop) |
4958 | { |
4959 | # ifdef MBS_SUPPORT |
4960 | if (MB_CUR_MAX != 1) |
4961 | return wcs_re_search_2 (bufp, string1, size1, string2, size2, startpos, |
4962 | range, regs, stop); |
4963 | else |
4964 | # endif |
4965 | return byte_re_search_2 (bufp, string1, size1, string2, size2, startpos, |
4966 | range, regs, stop); |
4967 | } /* re_search_2 */ |
4968 | #ifdef _LIBC |
4969 | weak_alias (__re_search_2, re_search_2) |
4970 | #endif |
4971 | |
4972 | #endif /* not INSIDE_RECURSION */ |
4973 | |
4974 | #ifdef INSIDE_RECURSION |
4975 | |
4976 | #ifdef MATCH_MAY_ALLOCATE |
4977 | # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL |
4978 | #else |
4979 | # define FREE_VAR(var) free (var); var = NULL |
4980 | #endif |
4981 | |
4982 | #ifdef WCHAR |
4983 | # define MAX_ALLOCA_SIZE 2000 |
4984 | |
4985 | # define FREE_WCS_BUFFERS() \ |
4986 | do { \ |
4987 | if (size1 > MAX_ALLOCA_SIZE) \ |
4988 | { \ |
4989 | free (wcs_string1); \ |
4990 | free (mbs_offset1); \ |
4991 | } \ |
4992 | else \ |
4993 | { \ |
4994 | FREE_VAR (wcs_string1); \ |
4995 | FREE_VAR (mbs_offset1); \ |
4996 | } \ |
4997 | if (size2 > MAX_ALLOCA_SIZE) \ |
4998 | { \ |
4999 | free (wcs_string2); \ |
5000 | free (mbs_offset2); \ |
5001 | } \ |
5002 | else \ |
5003 | { \ |
5004 | FREE_VAR (wcs_string2); \ |
5005 | FREE_VAR (mbs_offset2); \ |
5006 | } \ |
5007 | } while (0) |
5008 | |
5009 | #endif |
5010 | |
5011 | |
5012 | static int |
5013 | PREFIX(re_search_2) (struct re_pattern_buffer *bufp, const char *string1, |
5014 | int size1, const char *string2, int size2, |
5015 | int startpos, int range, |
5016 | struct re_registers *regs, int stop) |
5017 | { |
5018 | int val; |
5019 | register char *fastmap = bufp->fastmap; |
5020 | register RE_TRANSLATE_TYPE translate = bufp->translate; |
5021 | int total_size = size1 + size2; |
5022 | int endpos = startpos + range; |
5023 | #ifdef WCHAR |
5024 | /* We need wchar_t* buffers correspond to cstring1, cstring2. */ |
5025 | wchar_t *wcs_string1 = NULL, *wcs_string2 = NULL; |
5026 | /* We need the size of wchar_t buffers correspond to csize1, csize2. */ |
5027 | int wcs_size1 = 0, wcs_size2 = 0; |
5028 | /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ |
5029 | int *mbs_offset1 = NULL, *mbs_offset2 = NULL; |
5030 | /* They hold whether each wchar_t is binary data or not. */ |
5031 | char *is_binary = NULL; |
5032 | #endif /* WCHAR */ |
5033 | |
5034 | /* Check for out-of-range STARTPOS. */ |
5035 | if (startpos < 0 || startpos > total_size) |
5036 | return -1; |
5037 | |
5038 | /* Fix up RANGE if it might eventually take us outside |
5039 | the virtual concatenation of STRING1 and STRING2. |
5040 | Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ |
5041 | if (endpos < 0) |
5042 | range = 0 - startpos; |
5043 | else if (endpos > total_size) |
5044 | range = total_size - startpos; |
5045 | |
5046 | /* If the search isn't to be a backwards one, don't waste time in a |
5047 | search for a pattern that must be anchored. */ |
5048 | if (bufp->used > 0 && range > 0 |
5049 | && ((re_opcode_t) bufp->buffer[0] == begbuf |
5050 | /* `begline' is like `begbuf' if it cannot match at newlines. */ |
5051 | || ((re_opcode_t) bufp->buffer[0] == begline |
5052 | && !bufp->newline_anchor))) |
5053 | { |
5054 | if (startpos > 0) |
5055 | return -1; |
5056 | else |
5057 | range = 1; |
5058 | } |
5059 | |
5060 | #ifdef emacs |
5061 | /* In a forward search for something that starts with \=. |
5062 | don't keep searching past point. */ |
5063 | if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0) |
5064 | { |
5065 | range = PT - startpos; |
5066 | if (range <= 0) |
5067 | return -1; |
5068 | } |
5069 | #endif /* emacs */ |
5070 | |
5071 | /* Update the fastmap now if not correct already. */ |
5072 | if (fastmap && !bufp->fastmap_accurate) |
5073 | if (re_compile_fastmap (bufp) == -2) |
5074 | return -2; |
5075 | |
5076 | #ifdef WCHAR |
5077 | /* Allocate wchar_t array for wcs_string1 and wcs_string2 and |
5078 | fill them with converted string. */ |
5079 | if (size1 != 0) |
5080 | { |
5081 | if (size1 > MAX_ALLOCA_SIZE) |
5082 | { |
5083 | wcs_string1 = TALLOC (size1 + 1, CHAR_T); |
5084 | mbs_offset1 = TALLOC (size1 + 1, int); |
5085 | is_binary = TALLOC (size1 + 1, char); |
5086 | } |
5087 | else |
5088 | { |
5089 | wcs_string1 = REGEX_TALLOC (size1 + 1, CHAR_T); |
5090 | mbs_offset1 = REGEX_TALLOC (size1 + 1, int); |
5091 | is_binary = REGEX_TALLOC (size1 + 1, char); |
5092 | } |
5093 | if (!wcs_string1 || !mbs_offset1 || !is_binary) |
5094 | { |
5095 | if (size1 > MAX_ALLOCA_SIZE) |
5096 | { |
5097 | free (wcs_string1); |
5098 | free (mbs_offset1); |
5099 | free (is_binary); |
5100 | } |
5101 | else |
5102 | { |
5103 | FREE_VAR (wcs_string1); |
5104 | FREE_VAR (mbs_offset1); |
5105 | FREE_VAR (is_binary); |
5106 | } |
5107 | return -2; |
5108 | } |
5109 | wcs_size1 = convert_mbs_to_wcs(wcs_string1, string1, size1, |
5110 | mbs_offset1, is_binary); |
5111 | wcs_string1[wcs_size1] = L'\0'; /* for a sentinel */ |
5112 | if (size1 > MAX_ALLOCA_SIZE) |
5113 | free (is_binary); |
5114 | else |
5115 | FREE_VAR (is_binary); |
5116 | } |
5117 | if (size2 != 0) |
5118 | { |
5119 | if (size2 > MAX_ALLOCA_SIZE) |
5120 | { |
5121 | wcs_string2 = TALLOC (size2 + 1, CHAR_T); |
5122 | mbs_offset2 = TALLOC (size2 + 1, int); |
5123 | is_binary = TALLOC (size2 + 1, char); |
5124 | } |
5125 | else |
5126 | { |
5127 | wcs_string2 = REGEX_TALLOC (size2 + 1, CHAR_T); |
5128 | mbs_offset2 = REGEX_TALLOC (size2 + 1, int); |
5129 | is_binary = REGEX_TALLOC (size2 + 1, char); |
5130 | } |
5131 | if (!wcs_string2 || !mbs_offset2 || !is_binary) |
5132 | { |
5133 | FREE_WCS_BUFFERS (); |
5134 | if (size2 > MAX_ALLOCA_SIZE) |
5135 | free (is_binary); |
5136 | else |
5137 | FREE_VAR (is_binary); |
5138 | return -2; |
5139 | } |
5140 | wcs_size2 = convert_mbs_to_wcs(wcs_string2, string2, size2, |
5141 | mbs_offset2, is_binary); |
5142 | wcs_string2[wcs_size2] = L'\0'; /* for a sentinel */ |
5143 | if (size2 > MAX_ALLOCA_SIZE) |
5144 | free (is_binary); |
5145 | else |
5146 | FREE_VAR (is_binary); |
5147 | } |
5148 | #endif /* WCHAR */ |
5149 | |
5150 | |
5151 | /* Loop through the string, looking for a place to start matching. */ |
5152 | for (;;) |
5153 | { |
5154 | /* If a fastmap is supplied, skip quickly over characters that |
5155 | cannot be the start of a match. If the pattern can match the |
5156 | null string, however, we don't need to skip characters; we want |
5157 | the first null string. */ |
5158 | if (fastmap && startpos < total_size && !bufp->can_be_null) |
5159 | { |
5160 | if (range > 0) /* Searching forwards. */ |
5161 | { |
5162 | register const char *d; |
5163 | register int lim = 0; |
5164 | int irange = range; |
5165 | |
5166 | if (startpos < size1 && startpos + range >= size1) |
5167 | lim = range - (size1 - startpos); |
5168 | |
5169 | d = (startpos >= size1 ? string2 - size1 : string1) + startpos; |
5170 | |
5171 | /* Written out as an if-else to avoid testing `translate' |
5172 | inside the loop. */ |
5173 | if (translate) |
5174 | while (range > lim |
5175 | && !fastmap[(unsigned char) |
5176 | translate[(unsigned char) *d++]]) |
5177 | range--; |
5178 | else |
5179 | while (range > lim && !fastmap[(unsigned char) *d++]) |
5180 | range--; |
5181 | |
5182 | startpos += irange - range; |
5183 | } |
5184 | else /* Searching backwards. */ |
5185 | { |
5186 | register CHAR_T c = (size1 == 0 || startpos >= size1 |
5187 | ? string2[startpos - size1] |
5188 | : string1[startpos]); |
5189 | |
5190 | if (!fastmap[(unsigned char) TRANSLATE (c)]) |
5191 | goto advance; |
5192 | } |
5193 | } |
5194 | |
5195 | /* If can't match the null string, and that's all we have left, fail. */ |
5196 | if (range >= 0 && startpos == total_size && fastmap |
5197 | && !bufp->can_be_null) |
5198 | { |
5199 | #ifdef WCHAR |
5200 | FREE_WCS_BUFFERS (); |
5201 | #endif |
5202 | return -1; |
5203 | } |
5204 | |
5205 | #ifdef WCHAR |
5206 | val = wcs_re_match_2_internal (bufp, string1, size1, string2, |
5207 | size2, startpos, regs, stop, |
5208 | wcs_string1, wcs_size1, |
5209 | wcs_string2, wcs_size2, |
5210 | mbs_offset1, mbs_offset2); |
5211 | #else /* BYTE */ |
5212 | val = byte_re_match_2_internal (bufp, string1, size1, string2, |
5213 | size2, startpos, regs, stop); |
5214 | #endif /* BYTE */ |
5215 | |
5216 | #ifndef REGEX_MALLOC |
5217 | # ifdef C_ALLOCA |
5218 | alloca (0); |
5219 | # endif |
5220 | #endif |
5221 | |
5222 | if (val >= 0) |
5223 | { |
5224 | #ifdef WCHAR |
5225 | FREE_WCS_BUFFERS (); |
5226 | #endif |
5227 | return startpos; |
5228 | } |
5229 | |
5230 | if (val == -2) |
5231 | { |
5232 | #ifdef WCHAR |
5233 | FREE_WCS_BUFFERS (); |
5234 | #endif |
5235 | return -2; |
5236 | } |
5237 | |
5238 | advance: |
5239 | if (!range) |
5240 | break; |
5241 | else if (range > 0) |
5242 | { |
5243 | range--; |
5244 | startpos++; |
5245 | } |
5246 | else |
5247 | { |
5248 | range++; |
5249 | startpos--; |
5250 | } |
5251 | } |
5252 | #ifdef WCHAR |
5253 | FREE_WCS_BUFFERS (); |
5254 | #endif |
5255 | return -1; |
5256 | } |
5257 | |
5258 | #ifdef WCHAR |
5259 | /* This converts PTR, a pointer into one of the search wchar_t strings |
5260 | `string1' and `string2' into an multibyte string offset from the |
5261 | beginning of that string. We use mbs_offset to optimize. |
5262 | See convert_mbs_to_wcs. */ |
5263 | # define POINTER_TO_OFFSET(ptr) \ |
5264 | (FIRST_STRING_P (ptr) \ |
5265 | ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \ |
5266 | : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \ |
5267 | + csize1))) |
5268 | #else /* BYTE */ |
5269 | /* This converts PTR, a pointer into one of the search strings `string1' |
5270 | and `string2' into an offset from the beginning of that string. */ |
5271 | # define POINTER_TO_OFFSET(ptr) \ |
5272 | (FIRST_STRING_P (ptr) \ |
5273 | ? ((regoff_t) ((ptr) - string1)) \ |
5274 | : ((regoff_t) ((ptr) - string2 + size1))) |
5275 | #endif /* WCHAR */ |
5276 | |
5277 | /* Macros for dealing with the split strings in re_match_2. */ |
5278 | |
5279 | #define MATCHING_IN_FIRST_STRING (dend == end_match_1) |
5280 | |
5281 | /* Call before fetching a character with *d. This switches over to |
5282 | string2 if necessary. */ |
5283 | #define PREFETCH() \ |
5284 | while (d == dend) \ |
5285 | { \ |
5286 | /* End of string2 => fail. */ \ |
5287 | if (dend == end_match_2) \ |
5288 | goto fail; \ |
5289 | /* End of string1 => advance to string2. */ \ |
5290 | d = string2; \ |
5291 | dend = end_match_2; \ |
5292 | } |
5293 | |
5294 | /* Test if at very beginning or at very end of the virtual concatenation |
5295 | of `string1' and `string2'. If only one string, it's `string2'. */ |
5296 | #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) |
5297 | #define AT_STRINGS_END(d) ((d) == end2) |
5298 | |
5299 | |
5300 | /* Test if D points to a character which is word-constituent. We have |
5301 | two special cases to check for: if past the end of string1, look at |
5302 | the first character in string2; and if before the beginning of |
5303 | string2, look at the last character in string1. */ |
5304 | #ifdef WCHAR |
5305 | /* Use internationalized API instead of SYNTAX. */ |
5306 | # define WORDCHAR_P(d) \ |
5307 | (iswalnum ((wint_t)((d) == end1 ? *string2 \ |
5308 | : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0 \ |
5309 | || ((d) == end1 ? *string2 \ |
5310 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) == L'_') |
5311 | #else /* BYTE */ |
5312 | # define WORDCHAR_P(d) \ |
5313 | (SYNTAX ((d) == end1 ? *string2 \ |
5314 | : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ |
5315 | == Sword) |
5316 | #endif /* WCHAR */ |
5317 | |
5318 | /* Disabled due to a compiler bug -- see comment at case wordbound */ |
5319 | #if 0 |
5320 | /* Test if the character before D and the one at D differ with respect |
5321 | to being word-constituent. */ |
5322 | #define AT_WORD_BOUNDARY(d) \ |
5323 | (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ |
5324 | || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) |
5325 | #endif |
5326 | |
5327 | /* Free everything we malloc. */ |
5328 | #ifdef MATCH_MAY_ALLOCATE |
5329 | # ifdef WCHAR |
5330 | # define FREE_VARIABLES() \ |
5331 | do { \ |
5332 | REGEX_FREE_STACK (fail_stack.stack); \ |
5333 | FREE_VAR (regstart); \ |
5334 | FREE_VAR (regend); \ |
5335 | FREE_VAR (old_regstart); \ |
5336 | FREE_VAR (old_regend); \ |
5337 | FREE_VAR (best_regstart); \ |
5338 | FREE_VAR (best_regend); \ |
5339 | FREE_VAR (reg_info); \ |
5340 | FREE_VAR (reg_dummy); \ |
5341 | FREE_VAR (reg_info_dummy); \ |
5342 | if (!cant_free_wcs_buf) \ |
5343 | { \ |
5344 | FREE_VAR (string1); \ |
5345 | FREE_VAR (string2); \ |
5346 | FREE_VAR (mbs_offset1); \ |
5347 | FREE_VAR (mbs_offset2); \ |
5348 | } \ |
5349 | } while (0) |
5350 | # else /* BYTE */ |
5351 | # define FREE_VARIABLES() \ |
5352 | do { \ |
5353 | REGEX_FREE_STACK (fail_stack.stack); \ |
5354 | FREE_VAR (regstart); \ |
5355 | FREE_VAR (regend); \ |
5356 | FREE_VAR (old_regstart); \ |
5357 | FREE_VAR (old_regend); \ |
5358 | FREE_VAR (best_regstart); \ |
5359 | FREE_VAR (best_regend); \ |
5360 | FREE_VAR (reg_info); \ |
5361 | FREE_VAR (reg_dummy); \ |
5362 | FREE_VAR (reg_info_dummy); \ |
5363 | } while (0) |
5364 | # endif /* WCHAR */ |
5365 | #else |
5366 | # ifdef WCHAR |
5367 | # define FREE_VARIABLES() \ |
5368 | do { \ |
5369 | if (!cant_free_wcs_buf) \ |
5370 | { \ |
5371 | FREE_VAR (string1); \ |
5372 | FREE_VAR (string2); \ |
5373 | FREE_VAR (mbs_offset1); \ |
5374 | FREE_VAR (mbs_offset2); \ |
5375 | } \ |
5376 | } while (0) |
5377 | # else /* BYTE */ |
5378 | # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ |
5379 | # endif /* WCHAR */ |
5380 | #endif /* not MATCH_MAY_ALLOCATE */ |
5381 | |
5382 | /* These values must meet several constraints. They must not be valid |
5383 | register values; since we have a limit of 255 registers (because |
5384 | we use only one byte in the pattern for the register number), we can |
5385 | use numbers larger than 255. They must differ by 1, because of |
5386 | NUM_FAILURE_ITEMS above. And the value for the lowest register must |
5387 | be larger than the value for the highest register, so we do not try |
5388 | to actually save any registers when none are active. */ |
5389 | #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) |
5390 | #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) |
5391 | |
5392 | #else /* not INSIDE_RECURSION */ |
5393 | /* Matching routines. */ |
5394 | |
5395 | #ifndef emacs /* Emacs never uses this. */ |
5396 | /* re_match is like re_match_2 except it takes only a single string. */ |
5397 | |
5398 | int |
5399 | re_match (struct re_pattern_buffer *bufp, const char *string, |
5400 | int size, int pos, struct re_registers *regs) |
5401 | { |
5402 | int result; |
5403 | # ifdef MBS_SUPPORT |
5404 | if (MB_CUR_MAX != 1) |
5405 | result = wcs_re_match_2_internal (bufp, NULL, 0, string, size, |
5406 | pos, regs, size, |
5407 | NULL, 0, NULL, 0, NULL, NULL); |
5408 | else |
5409 | # endif |
5410 | result = byte_re_match_2_internal (bufp, NULL, size1: 0, string2: string, size2: size, |
5411 | pos, regs, stop: size); |
5412 | # ifndef REGEX_MALLOC |
5413 | # ifdef C_ALLOCA |
5414 | alloca (0); |
5415 | # endif |
5416 | # endif |
5417 | return result; |
5418 | } |
5419 | # ifdef _LIBC |
5420 | weak_alias (__re_match, re_match) |
5421 | # endif |
5422 | #endif /* not emacs */ |
5423 | |
5424 | #endif /* not INSIDE_RECURSION */ |
5425 | |
5426 | #ifdef INSIDE_RECURSION |
5427 | static boolean PREFIX(group_match_null_string_p) (UCHAR_T **p, |
5428 | UCHAR_T *end, |
5429 | PREFIX(register_info_type) *reg_info); |
5430 | static boolean PREFIX(alt_match_null_string_p) (UCHAR_T *p, |
5431 | UCHAR_T *end, |
5432 | PREFIX(register_info_type) *reg_info); |
5433 | static boolean PREFIX(common_op_match_null_string_p) (UCHAR_T **p, |
5434 | UCHAR_T *end, |
5435 | PREFIX(register_info_type) *reg_info); |
5436 | static int PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2, |
5437 | int len, char *translate); |
5438 | #else /* not INSIDE_RECURSION */ |
5439 | |
5440 | /* re_match_2 matches the compiled pattern in BUFP against the |
5441 | the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 |
5442 | and SIZE2, respectively). We start matching at POS, and stop |
5443 | matching at STOP. |
5444 | |
5445 | If REGS is non-null and the `no_sub' field of BUFP is nonzero, we |
5446 | store offsets for the substring each group matched in REGS. See the |
5447 | documentation for exactly how many groups we fill. |
5448 | |
5449 | We return -1 if no match, -2 if an internal error (such as the |
5450 | failure stack overflowing). Otherwise, we return the length of the |
5451 | matched substring. */ |
5452 | |
5453 | int |
5454 | re_match_2 (struct re_pattern_buffer *bufp, const char *string1, int size1, |
5455 | const char *string2, int size2, int pos, |
5456 | struct re_registers *regs, int stop) |
5457 | { |
5458 | int result; |
5459 | # ifdef MBS_SUPPORT |
5460 | if (MB_CUR_MAX != 1) |
5461 | result = wcs_re_match_2_internal (bufp, string1, size1, string2, size2, |
5462 | pos, regs, stop, |
5463 | NULL, 0, NULL, 0, NULL, NULL); |
5464 | else |
5465 | # endif |
5466 | result = byte_re_match_2_internal (bufp, string1, size1, string2, size2, |
5467 | pos, regs, stop); |
5468 | |
5469 | #ifndef REGEX_MALLOC |
5470 | # ifdef C_ALLOCA |
5471 | alloca (0); |
5472 | # endif |
5473 | #endif |
5474 | return result; |
5475 | } |
5476 | #ifdef _LIBC |
5477 | weak_alias (__re_match_2, re_match_2) |
5478 | #endif |
5479 | |
5480 | #endif /* not INSIDE_RECURSION */ |
5481 | |
5482 | #ifdef INSIDE_RECURSION |
5483 | |
5484 | #ifdef WCHAR |
5485 | static int count_mbs_length (int *, int); |
5486 | |
5487 | /* This check the substring (from 0, to length) of the multibyte string, |
5488 | to which offset_buffer correspond. And count how many wchar_t_characters |
5489 | the substring occupy. We use offset_buffer to optimization. |
5490 | See convert_mbs_to_wcs. */ |
5491 | |
5492 | static int |
5493 | count_mbs_length(int *offset_buffer, int length) |
5494 | { |
5495 | int upper, lower; |
5496 | |
5497 | /* Check whether the size is valid. */ |
5498 | if (length < 0) |
5499 | return -1; |
5500 | |
5501 | if (offset_buffer == NULL) |
5502 | return 0; |
5503 | |
5504 | /* If there are no multibyte character, offset_buffer[i] == i. |
5505 | Optmize for this case. */ |
5506 | if (offset_buffer[length] == length) |
5507 | return length; |
5508 | |
5509 | /* Set up upper with length. (because for all i, offset_buffer[i] >= i) */ |
5510 | upper = length; |
5511 | lower = 0; |
5512 | |
5513 | while (true) |
5514 | { |
5515 | int middle = (lower + upper) / 2; |
5516 | if (middle == lower || middle == upper) |
5517 | break; |
5518 | if (offset_buffer[middle] > length) |
5519 | upper = middle; |
5520 | else if (offset_buffer[middle] < length) |
5521 | lower = middle; |
5522 | else |
5523 | return middle; |
5524 | } |
5525 | |
5526 | return -1; |
5527 | } |
5528 | #endif /* WCHAR */ |
5529 | |
5530 | /* This is a separate function so that we can force an alloca cleanup |
5531 | afterwards. */ |
5532 | #ifdef WCHAR |
5533 | static int |
5534 | wcs_re_match_2_internal (struct re_pattern_buffer *bufp, |
5535 | const char *cstring1, int csize1, |
5536 | const char *cstring2, int csize2, |
5537 | int pos, |
5538 | struct re_registers *regs, |
5539 | int stop, |
5540 | /* string1 == string2 == NULL means string1/2, size1/2 and |
5541 | mbs_offset1/2 need seting up in this function. */ |
5542 | /* We need wchar_t* buffers correspond to cstring1, cstring2. */ |
5543 | wchar_t *string1, int size1, |
5544 | wchar_t *string2, int size2, |
5545 | /* offset buffer for optimizatoin. See convert_mbs_to_wc. */ |
5546 | int *mbs_offset1, int *mbs_offset2) |
5547 | #else /* BYTE */ |
5548 | static int |
5549 | byte_re_match_2_internal (struct re_pattern_buffer *bufp, |
5550 | const char *string1, int size1, |
5551 | const char *string2, int size2, |
5552 | int pos, |
5553 | struct re_registers *regs, int stop) |
5554 | #endif /* BYTE */ |
5555 | { |
5556 | /* General temporaries. */ |
5557 | int mcnt; |
5558 | UCHAR_T *p1; |
5559 | #ifdef WCHAR |
5560 | /* They hold whether each wchar_t is binary data or not. */ |
5561 | char *is_binary = NULL; |
5562 | /* If true, we can't free string1/2, mbs_offset1/2. */ |
5563 | int cant_free_wcs_buf = 1; |
5564 | #endif /* WCHAR */ |
5565 | |
5566 | /* Just past the end of the corresponding string. */ |
5567 | const CHAR_T *end1, *end2; |
5568 | |
5569 | /* Pointers into string1 and string2, just past the last characters in |
5570 | each to consider matching. */ |
5571 | const CHAR_T *end_match_1, *end_match_2; |
5572 | |
5573 | /* Where we are in the data, and the end of the current string. */ |
5574 | const CHAR_T *d, *dend; |
5575 | |
5576 | /* Where we are in the pattern, and the end of the pattern. */ |
5577 | #ifdef WCHAR |
5578 | UCHAR_T *pattern, *p; |
5579 | register UCHAR_T *pend; |
5580 | #else /* BYTE */ |
5581 | UCHAR_T *p = bufp->buffer; |
5582 | register UCHAR_T *pend = p + bufp->used; |
5583 | #endif /* WCHAR */ |
5584 | |
5585 | /* Mark the opcode just after a start_memory, so we can test for an |
5586 | empty subpattern when we get to the stop_memory. */ |
5587 | UCHAR_T *just_past_start_mem = 0; |
5588 | |
5589 | /* We use this to map every character in the string. */ |
5590 | RE_TRANSLATE_TYPE translate = bufp->translate; |
5591 | |
5592 | /* Failure point stack. Each place that can handle a failure further |
5593 | down the line pushes a failure point on this stack. It consists of |
5594 | restart, regend, and reg_info for all registers corresponding to |
5595 | the subexpressions we're currently inside, plus the number of such |
5596 | registers, and, finally, two char *'s. The first char * is where |
5597 | to resume scanning the pattern; the second one is where to resume |
5598 | scanning the strings. If the latter is zero, the failure point is |
5599 | a ``dummy''; if a failure happens and the failure point is a dummy, |
5600 | it gets discarded and the next one is tried. */ |
5601 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
5602 | PREFIX(fail_stack_type) fail_stack; |
5603 | #endif |
5604 | #ifdef DEBUG |
5605 | static unsigned failure_id; |
5606 | unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; |
5607 | #endif |
5608 | |
5609 | #ifdef REL_ALLOC |
5610 | /* This holds the pointer to the failure stack, when |
5611 | it is allocated relocatably. */ |
5612 | fail_stack_elt_t *failure_stack_ptr; |
5613 | #endif |
5614 | |
5615 | /* We fill all the registers internally, independent of what we |
5616 | return, for use in backreferences. The number here includes |
5617 | an element for register zero. */ |
5618 | size_t num_regs = bufp->re_nsub + 1; |
5619 | |
5620 | /* The currently active registers. */ |
5621 | active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
5622 | active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
5623 | |
5624 | /* Information on the contents of registers. These are pointers into |
5625 | the input strings; they record just what was matched (on this |
5626 | attempt) by a subexpression part of the pattern, that is, the |
5627 | regnum-th regstart pointer points to where in the pattern we began |
5628 | matching and the regnum-th regend points to right after where we |
5629 | stopped matching the regnum-th subexpression. (The zeroth register |
5630 | keeps track of what the whole pattern matches.) */ |
5631 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
5632 | const CHAR_T **regstart, **regend; |
5633 | #endif |
5634 | |
5635 | /* If a group that's operated upon by a repetition operator fails to |
5636 | match anything, then the register for its start will need to be |
5637 | restored because it will have been set to wherever in the string we |
5638 | are when we last see its open-group operator. Similarly for a |
5639 | register's end. */ |
5640 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
5641 | const CHAR_T **old_regstart, **old_regend; |
5642 | #endif |
5643 | |
5644 | /* The is_active field of reg_info helps us keep track of which (possibly |
5645 | nested) subexpressions we are currently in. The matched_something |
5646 | field of reg_info[reg_num] helps us tell whether or not we have |
5647 | matched any of the pattern so far this time through the reg_num-th |
5648 | subexpression. These two fields get reset each time through any |
5649 | loop their register is in. */ |
5650 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ |
5651 | PREFIX(register_info_type) *reg_info; |
5652 | #endif |
5653 | |
5654 | /* The following record the register info as found in the above |
5655 | variables when we find a match better than any we've seen before. |
5656 | This happens as we backtrack through the failure points, which in |
5657 | turn happens only if we have not yet matched the entire string. */ |
5658 | unsigned best_regs_set = false; |
5659 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
5660 | const CHAR_T **best_regstart, **best_regend; |
5661 | #endif |
5662 | |
5663 | /* Logically, this is `best_regend[0]'. But we don't want to have to |
5664 | allocate space for that if we're not allocating space for anything |
5665 | else (see below). Also, we never need info about register 0 for |
5666 | any of the other register vectors, and it seems rather a kludge to |
5667 | treat `best_regend' differently than the rest. So we keep track of |
5668 | the end of the best match so far in a separate variable. We |
5669 | initialize this to NULL so that when we backtrack the first time |
5670 | and need to test it, it's not garbage. */ |
5671 | const CHAR_T *match_end = NULL; |
5672 | |
5673 | /* This helps SET_REGS_MATCHED avoid doing redundant work. */ |
5674 | int set_regs_matched_done = 0; |
5675 | |
5676 | /* Used when we pop values we don't care about. */ |
5677 | #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ |
5678 | const CHAR_T **reg_dummy; |
5679 | PREFIX(register_info_type) *reg_info_dummy; |
5680 | #endif |
5681 | |
5682 | #ifdef DEBUG |
5683 | /* Counts the total number of registers pushed. */ |
5684 | unsigned num_regs_pushed = 0; |
5685 | #endif |
5686 | |
5687 | DEBUG_PRINT1 ("\n\nEntering re_match_2.\n" ); |
5688 | |
5689 | INIT_FAIL_STACK (); |
5690 | |
5691 | #ifdef MATCH_MAY_ALLOCATE |
5692 | /* Do not bother to initialize all the register variables if there are |
5693 | no groups in the pattern, as it takes a fair amount of time. If |
5694 | there are groups, we include space for register 0 (the whole |
5695 | pattern), even though we never use it, since it simplifies the |
5696 | array indexing. We should fix this. */ |
5697 | if (bufp->re_nsub) |
5698 | { |
5699 | regstart = REGEX_TALLOC (num_regs, const CHAR_T *); |
5700 | regend = REGEX_TALLOC (num_regs, const CHAR_T *); |
5701 | old_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); |
5702 | old_regend = REGEX_TALLOC (num_regs, const CHAR_T *); |
5703 | best_regstart = REGEX_TALLOC (num_regs, const CHAR_T *); |
5704 | best_regend = REGEX_TALLOC (num_regs, const CHAR_T *); |
5705 | reg_info = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); |
5706 | reg_dummy = REGEX_TALLOC (num_regs, const CHAR_T *); |
5707 | reg_info_dummy = REGEX_TALLOC (num_regs, PREFIX(register_info_type)); |
5708 | |
5709 | if (!(regstart && regend && old_regstart && old_regend && reg_info |
5710 | && best_regstart && best_regend && reg_dummy && reg_info_dummy)) |
5711 | { |
5712 | FREE_VARIABLES (); |
5713 | return -2; |
5714 | } |
5715 | } |
5716 | else |
5717 | { |
5718 | /* We must initialize all our variables to NULL, so that |
5719 | `FREE_VARIABLES' doesn't try to free them. */ |
5720 | regstart = regend = old_regstart = old_regend = best_regstart |
5721 | = best_regend = reg_dummy = NULL; |
5722 | reg_info = reg_info_dummy = (PREFIX(register_info_type) *) NULL; |
5723 | } |
5724 | #endif /* MATCH_MAY_ALLOCATE */ |
5725 | |
5726 | /* The starting position is bogus. */ |
5727 | #ifdef WCHAR |
5728 | if (pos < 0 || pos > csize1 + csize2) |
5729 | #else /* BYTE */ |
5730 | if (pos < 0 || pos > size1 + size2) |
5731 | #endif |
5732 | { |
5733 | FREE_VARIABLES (); |
5734 | return -1; |
5735 | } |
5736 | |
5737 | #ifdef WCHAR |
5738 | /* Allocate wchar_t array for string1 and string2 and |
5739 | fill them with converted string. */ |
5740 | if (string1 == NULL && string2 == NULL) |
5741 | { |
5742 | /* We need seting up buffers here. */ |
5743 | |
5744 | /* We must free wcs buffers in this function. */ |
5745 | cant_free_wcs_buf = 0; |
5746 | |
5747 | if (csize1 != 0) |
5748 | { |
5749 | string1 = REGEX_TALLOC (csize1 + 1, CHAR_T); |
5750 | mbs_offset1 = REGEX_TALLOC (csize1 + 1, int); |
5751 | is_binary = REGEX_TALLOC (csize1 + 1, char); |
5752 | if (!string1 || !mbs_offset1 || !is_binary) |
5753 | { |
5754 | FREE_VAR (string1); |
5755 | FREE_VAR (mbs_offset1); |
5756 | FREE_VAR (is_binary); |
5757 | return -2; |
5758 | } |
5759 | } |
5760 | if (csize2 != 0) |
5761 | { |
5762 | string2 = REGEX_TALLOC (csize2 + 1, CHAR_T); |
5763 | mbs_offset2 = REGEX_TALLOC (csize2 + 1, int); |
5764 | is_binary = REGEX_TALLOC (csize2 + 1, char); |
5765 | if (!string2 || !mbs_offset2 || !is_binary) |
5766 | { |
5767 | FREE_VAR (string1); |
5768 | FREE_VAR (mbs_offset1); |
5769 | FREE_VAR (string2); |
5770 | FREE_VAR (mbs_offset2); |
5771 | FREE_VAR (is_binary); |
5772 | return -2; |
5773 | } |
5774 | size2 = convert_mbs_to_wcs(string2, cstring2, csize2, |
5775 | mbs_offset2, is_binary); |
5776 | string2[size2] = L'\0'; /* for a sentinel */ |
5777 | FREE_VAR (is_binary); |
5778 | } |
5779 | } |
5780 | |
5781 | /* We need to cast pattern to (wchar_t*), because we casted this compiled |
5782 | pattern to (char*) in regex_compile. */ |
5783 | p = pattern = (CHAR_T*)bufp->buffer; |
5784 | pend = (CHAR_T*)(bufp->buffer + bufp->used); |
5785 | |
5786 | #endif /* WCHAR */ |
5787 | |
5788 | /* Initialize subexpression text positions to -1 to mark ones that no |
5789 | start_memory/stop_memory has been seen for. Also initialize the |
5790 | register information struct. */ |
5791 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
5792 | { |
5793 | regstart[mcnt] = regend[mcnt] |
5794 | = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; |
5795 | |
5796 | REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; |
5797 | IS_ACTIVE (reg_info[mcnt]) = 0; |
5798 | MATCHED_SOMETHING (reg_info[mcnt]) = 0; |
5799 | EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0; |
5800 | } |
5801 | |
5802 | /* We move `string1' into `string2' if the latter's empty -- but not if |
5803 | `string1' is null. */ |
5804 | if (size2 == 0 && string1 != NULL) |
5805 | { |
5806 | string2 = string1; |
5807 | size2 = size1; |
5808 | string1 = 0; |
5809 | size1 = 0; |
5810 | #ifdef WCHAR |
5811 | mbs_offset2 = mbs_offset1; |
5812 | csize2 = csize1; |
5813 | mbs_offset1 = NULL; |
5814 | csize1 = 0; |
5815 | #endif |
5816 | } |
5817 | end1 = string1 + size1; |
5818 | end2 = string2 + size2; |
5819 | |
5820 | /* Compute where to stop matching, within the two strings. */ |
5821 | #ifdef WCHAR |
5822 | if (stop <= csize1) |
5823 | { |
5824 | mcnt = count_mbs_length(mbs_offset1, stop); |
5825 | end_match_1 = string1 + mcnt; |
5826 | end_match_2 = string2; |
5827 | } |
5828 | else |
5829 | { |
5830 | if (stop > csize1 + csize2) |
5831 | stop = csize1 + csize2; |
5832 | end_match_1 = end1; |
5833 | mcnt = count_mbs_length(mbs_offset2, stop-csize1); |
5834 | end_match_2 = string2 + mcnt; |
5835 | } |
5836 | if (mcnt < 0) |
5837 | { /* count_mbs_length return error. */ |
5838 | FREE_VARIABLES (); |
5839 | return -1; |
5840 | } |
5841 | #else |
5842 | if (stop <= size1) |
5843 | { |
5844 | end_match_1 = string1 + stop; |
5845 | end_match_2 = string2; |
5846 | } |
5847 | else |
5848 | { |
5849 | end_match_1 = end1; |
5850 | end_match_2 = string2 + stop - size1; |
5851 | } |
5852 | #endif /* WCHAR */ |
5853 | |
5854 | /* `p' scans through the pattern as `d' scans through the data. |
5855 | `dend' is the end of the input string that `d' points within. `d' |
5856 | is advanced into the following input string whenever necessary, but |
5857 | this happens before fetching; therefore, at the beginning of the |
5858 | loop, `d' can be pointing at the end of a string, but it cannot |
5859 | equal `string2'. */ |
5860 | #ifdef WCHAR |
5861 | if (size1 > 0 && pos <= csize1) |
5862 | { |
5863 | mcnt = count_mbs_length(mbs_offset1, pos); |
5864 | d = string1 + mcnt; |
5865 | dend = end_match_1; |
5866 | } |
5867 | else |
5868 | { |
5869 | mcnt = count_mbs_length(mbs_offset2, pos-csize1); |
5870 | d = string2 + mcnt; |
5871 | dend = end_match_2; |
5872 | } |
5873 | |
5874 | if (mcnt < 0) |
5875 | { /* count_mbs_length return error. */ |
5876 | FREE_VARIABLES (); |
5877 | return -1; |
5878 | } |
5879 | #else |
5880 | if (size1 > 0 && pos <= size1) |
5881 | { |
5882 | d = string1 + pos; |
5883 | dend = end_match_1; |
5884 | } |
5885 | else |
5886 | { |
5887 | d = string2 + pos - size1; |
5888 | dend = end_match_2; |
5889 | } |
5890 | #endif /* WCHAR */ |
5891 | |
5892 | DEBUG_PRINT1 ("The compiled pattern is:\n" ); |
5893 | DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend); |
5894 | DEBUG_PRINT1 ("The string to match is: `" ); |
5895 | DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2); |
5896 | DEBUG_PRINT1 ("'\n" ); |
5897 | |
5898 | /* This loops over pattern commands. It exits by returning from the |
5899 | function if the match is complete, or it drops through if the match |
5900 | fails at this starting point in the input data. */ |
5901 | for (;;) |
5902 | { |
5903 | #ifdef _LIBC |
5904 | DEBUG_PRINT2 ("\n%p: " , p); |
5905 | #else |
5906 | DEBUG_PRINT2 ("\n0x%x: " , p); |
5907 | #endif |
5908 | |
5909 | if (p == pend) |
5910 | { /* End of pattern means we might have succeeded. */ |
5911 | DEBUG_PRINT1 ("end of pattern ... " ); |
5912 | |
5913 | /* If we haven't matched the entire string, and we want the |
5914 | longest match, try backtracking. */ |
5915 | if (d != end_match_2) |
5916 | { |
5917 | /* 1 if this match ends in the same string (string1 or string2) |
5918 | as the best previous match. */ |
5919 | boolean same_str_p; |
5920 | |
5921 | /* 1 if this match is the best seen so far. */ |
5922 | boolean best_match_p; |
5923 | |
5924 | same_str_p = (FIRST_STRING_P (match_end) |
5925 | == MATCHING_IN_FIRST_STRING); |
5926 | |
5927 | /* AIX compiler got confused when this was combined |
5928 | with the previous declaration. */ |
5929 | if (same_str_p) |
5930 | best_match_p = d > match_end; |
5931 | else |
5932 | best_match_p = !MATCHING_IN_FIRST_STRING; |
5933 | |
5934 | DEBUG_PRINT1 ("backtracking.\n" ); |
5935 | |
5936 | if (!FAIL_STACK_EMPTY ()) |
5937 | { /* More failure points to try. */ |
5938 | |
5939 | /* If exceeds best match so far, save it. */ |
5940 | if (!best_regs_set || best_match_p) |
5941 | { |
5942 | best_regs_set = true; |
5943 | match_end = d; |
5944 | |
5945 | DEBUG_PRINT1 ("\nSAVING match as best so far.\n" ); |
5946 | |
5947 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
5948 | { |
5949 | best_regstart[mcnt] = regstart[mcnt]; |
5950 | best_regend[mcnt] = regend[mcnt]; |
5951 | } |
5952 | } |
5953 | goto fail; |
5954 | } |
5955 | |
5956 | /* If no failure points, don't restore garbage. And if |
5957 | last match is real best match, don't restore second |
5958 | best one. */ |
5959 | else if (best_regs_set && !best_match_p) |
5960 | { |
5961 | restore_best_regs: |
5962 | /* Restore best match. It may happen that `dend == |
5963 | end_match_1' while the restored d is in string2. |
5964 | For example, the pattern `x.*y.*z' against the |
5965 | strings `x-' and `y-z-', if the two strings are |
5966 | not consecutive in memory. */ |
5967 | DEBUG_PRINT1 ("Restoring best registers.\n" ); |
5968 | |
5969 | d = match_end; |
5970 | dend = ((d >= string1 && d <= end1) |
5971 | ? end_match_1 : end_match_2); |
5972 | |
5973 | for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) |
5974 | { |
5975 | regstart[mcnt] = best_regstart[mcnt]; |
5976 | regend[mcnt] = best_regend[mcnt]; |
5977 | } |
5978 | } |
5979 | } /* d != end_match_2 */ |
5980 | |
5981 | succeed_label: |
5982 | DEBUG_PRINT1 ("Accepting match.\n" ); |
5983 | /* If caller wants register contents data back, do it. */ |
5984 | if (regs && !bufp->no_sub) |
5985 | { |
5986 | /* Have the register data arrays been allocated? */ |
5987 | if (bufp->regs_allocated == REGS_UNALLOCATED) |
5988 | { /* No. So allocate them with malloc. We need one |
5989 | extra element beyond `num_regs' for the `-1' marker |
5990 | GNU code uses. */ |
5991 | regs->num_regs = MAX (RE_NREGS, num_regs + 1); |
5992 | regs->start = TALLOC (regs->num_regs, regoff_t); |
5993 | regs->end = TALLOC (regs->num_regs, regoff_t); |
5994 | if (regs->start == NULL || regs->end == NULL) |
5995 | { |
5996 | FREE_VARIABLES (); |
5997 | return -2; |
5998 | } |
5999 | bufp->regs_allocated = REGS_REALLOCATE; |
6000 | } |
6001 | else if (bufp->regs_allocated == REGS_REALLOCATE) |
6002 | { /* Yes. If we need more elements than were already |
6003 | allocated, reallocate them. If we need fewer, just |
6004 | leave it alone. */ |
6005 | if (regs->num_regs < num_regs + 1) |
6006 | { |
6007 | regs->num_regs = num_regs + 1; |
6008 | RETALLOC (regs->start, regs->num_regs, regoff_t); |
6009 | RETALLOC (regs->end, regs->num_regs, regoff_t); |
6010 | if (regs->start == NULL || regs->end == NULL) |
6011 | { |
6012 | FREE_VARIABLES (); |
6013 | return -2; |
6014 | } |
6015 | } |
6016 | } |
6017 | else |
6018 | { |
6019 | /* These braces fend off a "empty body in an else-statement" |
6020 | warning under GCC when assert expands to nothing. */ |
6021 | assert (bufp->regs_allocated == REGS_FIXED); |
6022 | } |
6023 | |
6024 | /* Convert the pointer data in `regstart' and `regend' to |
6025 | indices. Register zero has to be set differently, |
6026 | since we haven't kept track of any info for it. */ |
6027 | if (regs->num_regs > 0) |
6028 | { |
6029 | regs->start[0] = pos; |
6030 | #ifdef WCHAR |
6031 | if (MATCHING_IN_FIRST_STRING) |
6032 | regs->end[0] = mbs_offset1 != NULL ? |
6033 | mbs_offset1[d-string1] : 0; |
6034 | else |
6035 | regs->end[0] = csize1 + (mbs_offset2 != NULL ? |
6036 | mbs_offset2[d-string2] : 0); |
6037 | #else |
6038 | regs->end[0] = (MATCHING_IN_FIRST_STRING |
6039 | ? ((regoff_t) (d - string1)) |
6040 | : ((regoff_t) (d - string2 + size1))); |
6041 | #endif /* WCHAR */ |
6042 | } |
6043 | |
6044 | /* Go through the first `min (num_regs, regs->num_regs)' |
6045 | registers, since that is all we initialized. */ |
6046 | for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs); |
6047 | mcnt++) |
6048 | { |
6049 | if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt])) |
6050 | regs->start[mcnt] = regs->end[mcnt] = -1; |
6051 | else |
6052 | { |
6053 | regs->start[mcnt] |
6054 | = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]); |
6055 | regs->end[mcnt] |
6056 | = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]); |
6057 | } |
6058 | } |
6059 | |
6060 | /* If the regs structure we return has more elements than |
6061 | were in the pattern, set the extra elements to -1. If |
6062 | we (re)allocated the registers, this is the case, |
6063 | because we always allocate enough to have at least one |
6064 | -1 at the end. */ |
6065 | for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++) |
6066 | regs->start[mcnt] = regs->end[mcnt] = -1; |
6067 | } /* regs && !bufp->no_sub */ |
6068 | |
6069 | DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n" , |
6070 | nfailure_points_pushed, nfailure_points_popped, |
6071 | nfailure_points_pushed - nfailure_points_popped); |
6072 | DEBUG_PRINT2 ("%u registers pushed.\n" , num_regs_pushed); |
6073 | |
6074 | #ifdef WCHAR |
6075 | if (MATCHING_IN_FIRST_STRING) |
6076 | mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0; |
6077 | else |
6078 | mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) + |
6079 | csize1; |
6080 | mcnt -= pos; |
6081 | #else |
6082 | mcnt = d - pos - (MATCHING_IN_FIRST_STRING |
6083 | ? string1 |
6084 | : string2 - size1); |
6085 | #endif /* WCHAR */ |
6086 | |
6087 | DEBUG_PRINT2 ("Returning %d from re_match_2.\n" , mcnt); |
6088 | |
6089 | FREE_VARIABLES (); |
6090 | return mcnt; |
6091 | } |
6092 | |
6093 | /* Otherwise match next pattern command. */ |
6094 | switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++)) |
6095 | { |
6096 | /* Ignore these. Used to ignore the n of succeed_n's which |
6097 | currently have n == 0. */ |
6098 | case no_op: |
6099 | DEBUG_PRINT1 ("EXECUTING no_op.\n" ); |
6100 | break; |
6101 | |
6102 | case succeed: |
6103 | DEBUG_PRINT1 ("EXECUTING succeed.\n" ); |
6104 | goto succeed_label; |
6105 | |
6106 | /* Match the next n pattern characters exactly. The following |
6107 | byte in the pattern defines n, and the n bytes after that |
6108 | are the characters to match. */ |
6109 | case exactn: |
6110 | #ifdef MBS_SUPPORT |
6111 | case exactn_bin: |
6112 | #endif |
6113 | mcnt = *p++; |
6114 | DEBUG_PRINT2 ("EXECUTING exactn %d.\n" , mcnt); |
6115 | |
6116 | /* This is written out as an if-else so we don't waste time |
6117 | testing `translate' inside the loop. */ |
6118 | if (translate) |
6119 | { |
6120 | do |
6121 | { |
6122 | PREFETCH (); |
6123 | #ifdef WCHAR |
6124 | if (*d <= 0xff) |
6125 | { |
6126 | if ((UCHAR_T) translate[(unsigned char) *d++] |
6127 | != (UCHAR_T) *p++) |
6128 | goto fail; |
6129 | } |
6130 | else |
6131 | { |
6132 | if (*d++ != (CHAR_T) *p++) |
6133 | goto fail; |
6134 | } |
6135 | #else |
6136 | if ((UCHAR_T) translate[(unsigned char) *d++] |
6137 | != (UCHAR_T) *p++) |
6138 | goto fail; |
6139 | #endif /* WCHAR */ |
6140 | } |
6141 | while (--mcnt); |
6142 | } |
6143 | else |
6144 | { |
6145 | do |
6146 | { |
6147 | PREFETCH (); |
6148 | if (*d++ != (CHAR_T) *p++) goto fail; |
6149 | } |
6150 | while (--mcnt); |
6151 | } |
6152 | SET_REGS_MATCHED (); |
6153 | break; |
6154 | |
6155 | |
6156 | /* Match any character except possibly a newline or a null. */ |
6157 | case anychar: |
6158 | DEBUG_PRINT1 ("EXECUTING anychar.\n" ); |
6159 | |
6160 | PREFETCH (); |
6161 | |
6162 | if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n') |
6163 | || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000')) |
6164 | goto fail; |
6165 | |
6166 | SET_REGS_MATCHED (); |
6167 | DEBUG_PRINT2 (" Matched `%ld'.\n" , (long int) *d); |
6168 | d++; |
6169 | break; |
6170 | |
6171 | |
6172 | case charset: |
6173 | case charset_not: |
6174 | { |
6175 | register UCHAR_T c; |
6176 | #ifdef WCHAR |
6177 | unsigned int i, char_class_length, coll_symbol_length, |
6178 | equiv_class_length, ranges_length, chars_length, length; |
6179 | CHAR_T *workp, *workp2, *charset_top; |
6180 | #define WORK_BUFFER_SIZE 128 |
6181 | CHAR_T str_buf[WORK_BUFFER_SIZE]; |
6182 | # ifdef _LIBC |
6183 | uint32_t nrules; |
6184 | # endif /* _LIBC */ |
6185 | #endif /* WCHAR */ |
6186 | boolean negate = (re_opcode_t) *(p - 1) == charset_not; |
6187 | |
6188 | DEBUG_PRINT2 ("EXECUTING charset%s.\n" , negate ? "_not" : "" ); |
6189 | PREFETCH (); |
6190 | c = TRANSLATE (*d); /* The character to match. */ |
6191 | #ifdef WCHAR |
6192 | # ifdef _LIBC |
6193 | nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES); |
6194 | # endif /* _LIBC */ |
6195 | charset_top = p - 1; |
6196 | char_class_length = *p++; |
6197 | coll_symbol_length = *p++; |
6198 | equiv_class_length = *p++; |
6199 | ranges_length = *p++; |
6200 | chars_length = *p++; |
6201 | /* p points charset[6], so the address of the next instruction |
6202 | (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'], |
6203 | where l=length of char_classes, m=length of collating_symbol, |
6204 | n=equivalence_class, o=length of char_range, |
6205 | p'=length of character. */ |
6206 | workp = p; |
6207 | /* Update p to indicate the next instruction. */ |
6208 | p += char_class_length + coll_symbol_length+ equiv_class_length + |
6209 | 2*ranges_length + chars_length; |
6210 | |
6211 | /* match with char_class? */ |
6212 | for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE) |
6213 | { |
6214 | wctype_t wctype; |
6215 | uintptr_t alignedp = ((uintptr_t)workp |
6216 | + __alignof__(wctype_t) - 1) |
6217 | & ~(uintptr_t)(__alignof__(wctype_t) - 1); |
6218 | wctype = *((wctype_t*)alignedp); |
6219 | workp += CHAR_CLASS_SIZE; |
6220 | # ifdef _LIBC |
6221 | if (__iswctype((wint_t)c, wctype)) |
6222 | goto char_set_matched; |
6223 | # else |
6224 | if (iswctype((wint_t)c, wctype)) |
6225 | goto char_set_matched; |
6226 | # endif |
6227 | } |
6228 | |
6229 | /* match with collating_symbol? */ |
6230 | # ifdef _LIBC |
6231 | if (nrules != 0) |
6232 | { |
6233 | const unsigned char *extra = (const unsigned char *) |
6234 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB); |
6235 | |
6236 | for (workp2 = workp + coll_symbol_length ; workp < workp2 ; |
6237 | workp++) |
6238 | { |
6239 | int32_t *wextra; |
6240 | wextra = (int32_t*)(extra + *workp++); |
6241 | for (i = 0; i < *wextra; ++i) |
6242 | if (TRANSLATE(d[i]) != wextra[1 + i]) |
6243 | break; |
6244 | |
6245 | if (i == *wextra) |
6246 | { |
6247 | /* Update d, however d will be incremented at |
6248 | char_set_matched:, we decrement d here. */ |
6249 | d += i - 1; |
6250 | goto char_set_matched; |
6251 | } |
6252 | } |
6253 | } |
6254 | else /* (nrules == 0) */ |
6255 | # endif |
6256 | /* If we can't look up collation data, we use wcscoll |
6257 | instead. */ |
6258 | { |
6259 | for (workp2 = workp + coll_symbol_length ; workp < workp2 ;) |
6260 | { |
6261 | const CHAR_T *backup_d = d, *backup_dend = dend; |
6262 | # ifdef _LIBC |
6263 | length = __wcslen (workp); |
6264 | # else |
6265 | length = wcslen (workp); |
6266 | # endif |
6267 | |
6268 | /* If wcscoll(the collating symbol, whole string) > 0, |
6269 | any substring of the string never match with the |
6270 | collating symbol. */ |
6271 | # ifdef _LIBC |
6272 | if (__wcscoll (workp, d) > 0) |
6273 | # else |
6274 | if (wcscoll (workp, d) > 0) |
6275 | # endif |
6276 | { |
6277 | workp += length + 1; |
6278 | continue; |
6279 | } |
6280 | |
6281 | /* First, we compare the collating symbol with |
6282 | the first character of the string. |
6283 | If it don't match, we add the next character to |
6284 | the compare buffer in turn. */ |
6285 | for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++) |
6286 | { |
6287 | int match; |
6288 | if (d == dend) |
6289 | { |
6290 | if (dend == end_match_2) |
6291 | break; |
6292 | d = string2; |
6293 | dend = end_match_2; |
6294 | } |
6295 | |
6296 | /* add next character to the compare buffer. */ |
6297 | str_buf[i] = TRANSLATE(*d); |
6298 | str_buf[i+1] = '\0'; |
6299 | |
6300 | # ifdef _LIBC |
6301 | match = __wcscoll (workp, str_buf); |
6302 | # else |
6303 | match = wcscoll (workp, str_buf); |
6304 | # endif |
6305 | if (match == 0) |
6306 | goto char_set_matched; |
6307 | |
6308 | if (match < 0) |
6309 | /* (str_buf > workp) indicate (str_buf + X > workp), |
6310 | because for all X (str_buf + X > str_buf). |
6311 | So we don't need continue this loop. */ |
6312 | break; |
6313 | |
6314 | /* Otherwise(str_buf < workp), |
6315 | (str_buf+next_character) may equals (workp). |
6316 | So we continue this loop. */ |
6317 | } |
6318 | /* not matched */ |
6319 | d = backup_d; |
6320 | dend = backup_dend; |
6321 | workp += length + 1; |
6322 | } |
6323 | } |
6324 | /* match with equivalence_class? */ |
6325 | # ifdef _LIBC |
6326 | if (nrules != 0) |
6327 | { |
6328 | const CHAR_T *backup_d = d, *backup_dend = dend; |
6329 | /* Try to match the equivalence class against |
6330 | those known to the collate implementation. */ |
6331 | const int32_t *table; |
6332 | const int32_t *weights; |
6333 | const int32_t *extra; |
6334 | const int32_t *indirect; |
6335 | int32_t idx, idx2; |
6336 | wint_t *cp; |
6337 | size_t len; |
6338 | |
6339 | /* This #include defines a local function! */ |
6340 | # include <locale/weightwc.h> |
6341 | |
6342 | table = (const int32_t *) |
6343 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC); |
6344 | weights = (const wint_t *) |
6345 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC); |
6346 | extra = (const wint_t *) |
6347 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC); |
6348 | indirect = (const int32_t *) |
6349 | _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC); |
6350 | |
6351 | /* Write 1 collating element to str_buf, and |
6352 | get its index. */ |
6353 | idx2 = 0; |
6354 | |
6355 | for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++) |
6356 | { |
6357 | cp = (wint_t*)str_buf; |
6358 | if (d == dend) |
6359 | { |
6360 | if (dend == end_match_2) |
6361 | break; |
6362 | d = string2; |
6363 | dend = end_match_2; |
6364 | } |
6365 | str_buf[i] = TRANSLATE(*(d+i)); |
6366 | str_buf[i+1] = '\0'; /* sentinel */ |
6367 | idx2 = findidx ((const wint_t**)&cp); |
6368 | } |
6369 | |
6370 | /* Update d, however d will be incremented at |
6371 | char_set_matched:, we decrement d here. */ |
6372 | d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1); |
6373 | if (d >= dend) |
6374 | { |
6375 | if (dend == end_match_2) |
6376 | d = dend; |
6377 | else |
6378 | { |
6379 | d = string2; |
6380 | dend = end_match_2; |
6381 | } |
6382 | } |
6383 | |
6384 | len = weights[idx2]; |
6385 | |
6386 | for (workp2 = workp + equiv_class_length ; workp < workp2 ; |
6387 | workp++) |
6388 | { |
6389 | idx = (int32_t)*workp; |
6390 | /* We already checked idx != 0 in regex_compile. */ |
6391 | |
6392 | if (idx2 != 0 && len == weights[idx]) |
6393 | { |
6394 | int cnt = 0; |
6395 | while (cnt < len && (weights[idx + 1 + cnt] |
6396 | == weights[idx2 + 1 + cnt])) |
6397 | ++cnt; |
6398 | |
6399 | if (cnt == len) |
6400 | goto char_set_matched; |
6401 | } |
6402 | } |
6403 | /* not matched */ |
6404 | d = backup_d; |
6405 | dend = backup_dend; |
6406 | } |
6407 | else /* (nrules == 0) */ |
6408 | # endif |
6409 | /* If we can't look up collation data, we use wcscoll |
6410 | instead. */ |
6411 | { |
6412 | for (workp2 = workp + equiv_class_length ; workp < workp2 ;) |
6413 | { |
6414 | const CHAR_T *backup_d = d, *backup_dend = dend; |
6415 | # ifdef _LIBC |
6416 | length = __wcslen (workp); |
6417 | # else |
6418 | length = wcslen (workp); |
6419 | # endif |
6420 | |
6421 | /* If wcscoll(the collating symbol, whole string) > 0, |
6422 | any substring of the string never match with the |
6423 | collating symbol. */ |
6424 | # ifdef _LIBC |
6425 | if (__wcscoll (workp, d) > 0) |
6426 | # else |
6427 | if (wcscoll (workp, d) > 0) |
6428 | # endif |
6429 | { |
6430 | workp += length + 1; |
6431 | break; |
6432 | } |
6433 | |
6434 | /* First, we compare the equivalence class with |
6435 | the first character of the string. |
6436 | If it don't match, we add the next character to |
6437 | the compare buffer in turn. */ |
6438 | for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++) |
6439 | { |
6440 | int match; |
6441 | if (d == dend) |
6442 | { |
6443 | if (dend == end_match_2) |
6444 | break; |
6445 | d = string2; |
6446 | dend = end_match_2; |
6447 | } |
6448 | |
6449 | /* add next character to the compare buffer. */ |
6450 | str_buf[i] = TRANSLATE(*d); |
6451 | str_buf[i+1] = '\0'; |
6452 | |
6453 | # ifdef _LIBC |
6454 | match = __wcscoll (workp, str_buf); |
6455 | # else |
6456 | match = wcscoll (workp, str_buf); |
6457 | # endif |
6458 | |
6459 | if (match == 0) |
6460 | goto char_set_matched; |
6461 | |
6462 | if (match < 0) |
6463 | /* (str_buf > workp) indicate (str_buf + X > workp), |
6464 | because for all X (str_buf + X > str_buf). |
6465 | So we don't need continue this loop. */ |
6466 | break; |
6467 | |
6468 | /* Otherwise(str_buf < workp), |
6469 | (str_buf+next_character) may equals (workp). |
6470 | So we continue this loop. */ |
6471 | } |
6472 | /* not matched */ |
6473 | d = backup_d; |
6474 | dend = backup_dend; |
6475 | workp += length + 1; |
6476 | } |
6477 | } |
6478 | |
6479 | /* match with char_range? */ |
6480 | # ifdef _LIBC |
6481 | if (nrules != 0) |
6482 | { |
6483 | uint32_t collseqval; |
6484 | const char *collseq = (const char *) |
6485 | _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC); |
6486 | |
6487 | collseqval = collseq_table_lookup (collseq, c); |
6488 | |
6489 | for (; workp < p - chars_length ;) |
6490 | { |
6491 | uint32_t start_val, end_val; |
6492 | |
6493 | /* We already compute the collation sequence value |
6494 | of the characters (or collating symbols). */ |
6495 | start_val = (uint32_t) *workp++; /* range_start */ |
6496 | end_val = (uint32_t) *workp++; /* range_end */ |
6497 | |
6498 | if (start_val <= collseqval && collseqval <= end_val) |
6499 | goto char_set_matched; |
6500 | } |
6501 | } |
6502 | else |
6503 | # endif |
6504 | { |
6505 | /* We set range_start_char at str_buf[0], range_end_char |
6506 | at str_buf[4], and compared char at str_buf[2]. */ |
6507 | str_buf[1] = 0; |
6508 | str_buf[2] = c; |
6509 | str_buf[3] = 0; |
6510 | str_buf[5] = 0; |
6511 | for (; workp < p - chars_length ;) |
6512 | { |
6513 | wchar_t *range_start_char, *range_end_char; |
6514 | |
6515 | /* match if (range_start_char <= c <= range_end_char). */ |
6516 | |
6517 | /* If range_start(or end) < 0, we assume -range_start(end) |
6518 | is the offset of the collating symbol which is specified |
6519 | as the character of the range start(end). */ |
6520 | |
6521 | /* range_start */ |
6522 | if (*workp < 0) |
6523 | range_start_char = charset_top - (*workp++); |
6524 | else |
6525 | { |
6526 | str_buf[0] = *workp++; |
6527 | range_start_char = str_buf; |
6528 | } |
6529 | |
6530 | /* range_end */ |
6531 | if (*workp < 0) |
6532 | range_end_char = charset_top - (*workp++); |
6533 | else |
6534 | { |
6535 | str_buf[4] = *workp++; |
6536 | range_end_char = str_buf + 4; |
6537 | } |
6538 | |
6539 | # ifdef _LIBC |
6540 | if (__wcscoll (range_start_char, str_buf+2) <= 0 |
6541 | && __wcscoll (str_buf+2, range_end_char) <= 0) |
6542 | # else |
6543 | if (wcscoll (range_start_char, str_buf+2) <= 0 |
6544 | && wcscoll (str_buf+2, range_end_char) <= 0) |
6545 | # endif |
6546 | goto char_set_matched; |
6547 | } |
6548 | } |
6549 | |
6550 | /* match with char? */ |
6551 | for (; workp < p ; workp++) |
6552 | if (c == *workp) |
6553 | goto char_set_matched; |
6554 | |
6555 | negate = !negate; |
6556 | |
6557 | char_set_matched: |
6558 | if (negate) goto fail; |
6559 | #else |
6560 | /* Cast to `unsigned' instead of `unsigned char' in case the |
6561 | bit list is a full 32 bytes long. */ |
6562 | if (c < (unsigned) (*p * BYTEWIDTH) |
6563 | && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
6564 | negate = !negate; |
6565 | |
6566 | p += 1 + *p; |
6567 | |
6568 | if (!negate) goto fail; |
6569 | #undef WORK_BUFFER_SIZE |
6570 | #endif /* WCHAR */ |
6571 | SET_REGS_MATCHED (); |
6572 | d++; |
6573 | break; |
6574 | } |
6575 | |
6576 | |
6577 | /* The beginning of a group is represented by start_memory. |
6578 | The arguments are the register number in the next byte, and the |
6579 | number of groups inner to this one in the next. The text |
6580 | matched within the group is recorded (in the internal |
6581 | registers data structure) under the register number. */ |
6582 | case start_memory: |
6583 | DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n" , |
6584 | (long int) *p, (long int) p[1]); |
6585 | |
6586 | /* Find out if this group can match the empty string. */ |
6587 | p1 = p; /* To send to group_match_null_string_p. */ |
6588 | |
6589 | if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE) |
6590 | REG_MATCH_NULL_STRING_P (reg_info[*p]) |
6591 | = PREFIX(group_match_null_string_p) (&p1, pend, reg_info); |
6592 | |
6593 | /* Save the position in the string where we were the last time |
6594 | we were at this open-group operator in case the group is |
6595 | operated upon by a repetition operator, e.g., with `(a*)*b' |
6596 | against `ab'; then we want to ignore where we are now in |
6597 | the string in case this attempt to match fails. */ |
6598 | old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) |
6599 | ? REG_UNSET (regstart[*p]) ? d : regstart[*p] |
6600 | : regstart[*p]; |
6601 | DEBUG_PRINT2 (" old_regstart: %d\n" , |
6602 | POINTER_TO_OFFSET (old_regstart[*p])); |
6603 | |
6604 | regstart[*p] = d; |
6605 | DEBUG_PRINT2 (" regstart: %d\n" , POINTER_TO_OFFSET (regstart[*p])); |
6606 | |
6607 | IS_ACTIVE (reg_info[*p]) = 1; |
6608 | MATCHED_SOMETHING (reg_info[*p]) = 0; |
6609 | |
6610 | /* Clear this whenever we change the register activity status. */ |
6611 | set_regs_matched_done = 0; |
6612 | |
6613 | /* This is the new highest active register. */ |
6614 | highest_active_reg = *p; |
6615 | |
6616 | /* If nothing was active before, this is the new lowest active |
6617 | register. */ |
6618 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) |
6619 | lowest_active_reg = *p; |
6620 | |
6621 | /* Move past the register number and inner group count. */ |
6622 | p += 2; |
6623 | just_past_start_mem = p; |
6624 | |
6625 | break; |
6626 | |
6627 | |
6628 | /* The stop_memory opcode represents the end of a group. Its |
6629 | arguments are the same as start_memory's: the register |
6630 | number, and the number of inner groups. */ |
6631 | case stop_memory: |
6632 | DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n" , |
6633 | (long int) *p, (long int) p[1]); |
6634 | |
6635 | /* We need to save the string position the last time we were at |
6636 | this close-group operator in case the group is operated |
6637 | upon by a repetition operator, e.g., with `((a*)*(b*)*)*' |
6638 | against `aba'; then we want to ignore where we are now in |
6639 | the string in case this attempt to match fails. */ |
6640 | old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p]) |
6641 | ? REG_UNSET (regend[*p]) ? d : regend[*p] |
6642 | : regend[*p]; |
6643 | DEBUG_PRINT2 (" old_regend: %d\n" , |
6644 | POINTER_TO_OFFSET (old_regend[*p])); |
6645 | |
6646 | regend[*p] = d; |
6647 | DEBUG_PRINT2 (" regend: %d\n" , POINTER_TO_OFFSET (regend[*p])); |
6648 | |
6649 | /* This register isn't active anymore. */ |
6650 | IS_ACTIVE (reg_info[*p]) = 0; |
6651 | |
6652 | /* Clear this whenever we change the register activity status. */ |
6653 | set_regs_matched_done = 0; |
6654 | |
6655 | /* If this was the only register active, nothing is active |
6656 | anymore. */ |
6657 | if (lowest_active_reg == highest_active_reg) |
6658 | { |
6659 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
6660 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
6661 | } |
6662 | else |
6663 | { /* We must scan for the new highest active register, since |
6664 | it isn't necessarily one less than now: consider |
6665 | (a(b)c(d(e)f)g). When group 3 ends, after the f), the |
6666 | new highest active register is 1. */ |
6667 | UCHAR_T r = *p - 1; |
6668 | while (r > 0 && !IS_ACTIVE (reg_info[r])) |
6669 | r--; |
6670 | |
6671 | /* If we end up at register zero, that means that we saved |
6672 | the registers as the result of an `on_failure_jump', not |
6673 | a `start_memory', and we jumped to past the innermost |
6674 | `stop_memory'. For example, in ((.)*) we save |
6675 | registers 1 and 2 as a result of the *, but when we pop |
6676 | back to the second ), we are at the stop_memory 1. |
6677 | Thus, nothing is active. */ |
6678 | if (r == 0) |
6679 | { |
6680 | lowest_active_reg = NO_LOWEST_ACTIVE_REG; |
6681 | highest_active_reg = NO_HIGHEST_ACTIVE_REG; |
6682 | } |
6683 | else |
6684 | highest_active_reg = r; |
6685 | } |
6686 | |
6687 | /* If just failed to match something this time around with a |
6688 | group that's operated on by a repetition operator, try to |
6689 | force exit from the ``loop'', and restore the register |
6690 | information for this group that we had before trying this |
6691 | last match. */ |
6692 | if ((!MATCHED_SOMETHING (reg_info[*p]) |
6693 | || just_past_start_mem == p - 1) |
6694 | && (p + 2) < pend) |
6695 | { |
6696 | boolean is_a_jump_n = false; |
6697 | |
6698 | p1 = p + 2; |
6699 | mcnt = 0; |
6700 | switch ((re_opcode_t) *p1++) |
6701 | { |
6702 | case jump_n: |
6703 | is_a_jump_n = true; |
6704 | /* Fall through. */ |
6705 | case pop_failure_jump: |
6706 | case maybe_pop_jump: |
6707 | case jump: |
6708 | case dummy_failure_jump: |
6709 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
6710 | if (is_a_jump_n) |
6711 | p1 += OFFSET_ADDRESS_SIZE; |
6712 | break; |
6713 | |
6714 | default: |
6715 | /* do nothing */ ; |
6716 | } |
6717 | p1 += mcnt; |
6718 | |
6719 | /* If the next operation is a jump backwards in the pattern |
6720 | to an on_failure_jump right before the start_memory |
6721 | corresponding to this stop_memory, exit from the loop |
6722 | by forcing a failure after pushing on the stack the |
6723 | on_failure_jump's jump in the pattern, and d. */ |
6724 | if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump |
6725 | && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory |
6726 | && p1[2+OFFSET_ADDRESS_SIZE] == *p) |
6727 | { |
6728 | /* If this group ever matched anything, then restore |
6729 | what its registers were before trying this last |
6730 | failed match, e.g., with `(a*)*b' against `ab' for |
6731 | regstart[1], and, e.g., with `((a*)*(b*)*)*' |
6732 | against `aba' for regend[3]. |
6733 | |
6734 | Also restore the registers for inner groups for, |
6735 | e.g., `((a*)(b*))*' against `aba' (register 3 would |
6736 | otherwise get trashed). */ |
6737 | |
6738 | if (EVER_MATCHED_SOMETHING (reg_info[*p])) |
6739 | { |
6740 | unsigned r; |
6741 | |
6742 | EVER_MATCHED_SOMETHING (reg_info[*p]) = 0; |
6743 | |
6744 | /* Restore this and inner groups' (if any) registers. */ |
6745 | for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1); |
6746 | r++) |
6747 | { |
6748 | regstart[r] = old_regstart[r]; |
6749 | |
6750 | /* xx why this test? */ |
6751 | if (old_regend[r] >= regstart[r]) |
6752 | regend[r] = old_regend[r]; |
6753 | } |
6754 | } |
6755 | p1++; |
6756 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
6757 | PUSH_FAILURE_POINT (p1 + mcnt, d, -2); |
6758 | |
6759 | goto fail; |
6760 | } |
6761 | } |
6762 | |
6763 | /* Move past the register number and the inner group count. */ |
6764 | p += 2; |
6765 | break; |
6766 | |
6767 | |
6768 | /* \<digit> has been turned into a `duplicate' command which is |
6769 | followed by the numeric value of <digit> as the register number. */ |
6770 | case duplicate: |
6771 | { |
6772 | register const CHAR_T *d2, *dend2; |
6773 | int regno = *p++; /* Get which register to match against. */ |
6774 | DEBUG_PRINT2 ("EXECUTING duplicate %d.\n" , regno); |
6775 | |
6776 | /* Can't back reference a group which we've never matched. */ |
6777 | if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno])) |
6778 | goto fail; |
6779 | |
6780 | /* Where in input to try to start matching. */ |
6781 | d2 = regstart[regno]; |
6782 | |
6783 | /* Where to stop matching; if both the place to start and |
6784 | the place to stop matching are in the same string, then |
6785 | set to the place to stop, otherwise, for now have to use |
6786 | the end of the first string. */ |
6787 | |
6788 | dend2 = ((FIRST_STRING_P (regstart[regno]) |
6789 | == FIRST_STRING_P (regend[regno])) |
6790 | ? regend[regno] : end_match_1); |
6791 | for (;;) |
6792 | { |
6793 | /* If necessary, advance to next segment in register |
6794 | contents. */ |
6795 | while (d2 == dend2) |
6796 | { |
6797 | if (dend2 == end_match_2) break; |
6798 | if (dend2 == regend[regno]) break; |
6799 | |
6800 | /* End of string1 => advance to string2. */ |
6801 | d2 = string2; |
6802 | dend2 = regend[regno]; |
6803 | } |
6804 | /* At end of register contents => success */ |
6805 | if (d2 == dend2) break; |
6806 | |
6807 | /* If necessary, advance to next segment in data. */ |
6808 | PREFETCH (); |
6809 | |
6810 | /* How many characters left in this segment to match. */ |
6811 | mcnt = dend - d; |
6812 | |
6813 | /* Want how many consecutive characters we can match in |
6814 | one shot, so, if necessary, adjust the count. */ |
6815 | if (mcnt > dend2 - d2) |
6816 | mcnt = dend2 - d2; |
6817 | |
6818 | /* Compare that many; failure if mismatch, else move |
6819 | past them. */ |
6820 | if (translate |
6821 | ? PREFIX(bcmp_translate) (d, d2, mcnt, translate) |
6822 | : memcmp (d, d2, mcnt*sizeof(UCHAR_T))) |
6823 | goto fail; |
6824 | d += mcnt, d2 += mcnt; |
6825 | |
6826 | /* Do this because we've match some characters. */ |
6827 | SET_REGS_MATCHED (); |
6828 | } |
6829 | } |
6830 | break; |
6831 | |
6832 | |
6833 | /* begline matches the empty string at the beginning of the string |
6834 | (unless `not_bol' is set in `bufp'), and, if |
6835 | `newline_anchor' is set, after newlines. */ |
6836 | case begline: |
6837 | DEBUG_PRINT1 ("EXECUTING begline.\n" ); |
6838 | |
6839 | if (AT_STRINGS_BEG (d)) |
6840 | { |
6841 | if (!bufp->not_bol) break; |
6842 | } |
6843 | else if (d[-1] == '\n' && bufp->newline_anchor) |
6844 | { |
6845 | break; |
6846 | } |
6847 | /* In all other cases, we fail. */ |
6848 | goto fail; |
6849 | |
6850 | |
6851 | /* endline is the dual of begline. */ |
6852 | case endline: |
6853 | DEBUG_PRINT1 ("EXECUTING endline.\n" ); |
6854 | |
6855 | if (AT_STRINGS_END (d)) |
6856 | { |
6857 | if (!bufp->not_eol) break; |
6858 | } |
6859 | |
6860 | /* We have to ``prefetch'' the next character. */ |
6861 | else if ((d == end1 ? *string2 : *d) == '\n' |
6862 | && bufp->newline_anchor) |
6863 | { |
6864 | break; |
6865 | } |
6866 | goto fail; |
6867 | |
6868 | |
6869 | /* Match at the very beginning of the data. */ |
6870 | case begbuf: |
6871 | DEBUG_PRINT1 ("EXECUTING begbuf.\n" ); |
6872 | if (AT_STRINGS_BEG (d)) |
6873 | break; |
6874 | goto fail; |
6875 | |
6876 | |
6877 | /* Match at the very end of the data. */ |
6878 | case endbuf: |
6879 | DEBUG_PRINT1 ("EXECUTING endbuf.\n" ); |
6880 | if (AT_STRINGS_END (d)) |
6881 | break; |
6882 | goto fail; |
6883 | |
6884 | |
6885 | /* on_failure_keep_string_jump is used to optimize `.*\n'. It |
6886 | pushes NULL as the value for the string on the stack. Then |
6887 | `pop_failure_point' will keep the current value for the |
6888 | string, instead of restoring it. To see why, consider |
6889 | matching `foo\nbar' against `.*\n'. The .* matches the foo; |
6890 | then the . fails against the \n. But the next thing we want |
6891 | to do is match the \n against the \n; if we restored the |
6892 | string value, we would be back at the foo. |
6893 | |
6894 | Because this is used only in specific cases, we don't need to |
6895 | check all the things that `on_failure_jump' does, to make |
6896 | sure the right things get saved on the stack. Hence we don't |
6897 | share its code. The only reason to push anything on the |
6898 | stack at all is that otherwise we would have to change |
6899 | `anychar's code to do something besides goto fail in this |
6900 | case; that seems worse than this. */ |
6901 | case on_failure_keep_string_jump: |
6902 | DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump" ); |
6903 | |
6904 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6905 | #ifdef _LIBC |
6906 | DEBUG_PRINT3 (" %d (to %p):\n" , mcnt, p + mcnt); |
6907 | #else |
6908 | DEBUG_PRINT3 (" %d (to 0x%x):\n" , mcnt, p + mcnt); |
6909 | #endif |
6910 | |
6911 | PUSH_FAILURE_POINT (p + mcnt, NULL, -2); |
6912 | break; |
6913 | |
6914 | |
6915 | /* Uses of on_failure_jump: |
6916 | |
6917 | Each alternative starts with an on_failure_jump that points |
6918 | to the beginning of the next alternative. Each alternative |
6919 | except the last ends with a jump that in effect jumps past |
6920 | the rest of the alternatives. (They really jump to the |
6921 | ending jump of the following alternative, because tensioning |
6922 | these jumps is a hassle.) |
6923 | |
6924 | Repeats start with an on_failure_jump that points past both |
6925 | the repetition text and either the following jump or |
6926 | pop_failure_jump back to this on_failure_jump. */ |
6927 | case on_failure_jump: |
6928 | on_failure: |
6929 | DEBUG_PRINT1 ("EXECUTING on_failure_jump" ); |
6930 | |
6931 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6932 | #ifdef _LIBC |
6933 | DEBUG_PRINT3 (" %d (to %p)" , mcnt, p + mcnt); |
6934 | #else |
6935 | DEBUG_PRINT3 (" %d (to 0x%x)" , mcnt, p + mcnt); |
6936 | #endif |
6937 | |
6938 | /* If this on_failure_jump comes right before a group (i.e., |
6939 | the original * applied to a group), save the information |
6940 | for that group and all inner ones, so that if we fail back |
6941 | to this point, the group's information will be correct. |
6942 | For example, in \(a*\)*\1, we need the preceding group, |
6943 | and in \(zz\(a*\)b*\)\2, we need the inner group. */ |
6944 | |
6945 | /* We can't use `p' to check ahead because we push |
6946 | a failure point to `p + mcnt' after we do this. */ |
6947 | p1 = p; |
6948 | |
6949 | /* We need to skip no_op's before we look for the |
6950 | start_memory in case this on_failure_jump is happening as |
6951 | the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 |
6952 | against aba. */ |
6953 | while (p1 < pend && (re_opcode_t) *p1 == no_op) |
6954 | p1++; |
6955 | |
6956 | if (p1 < pend && (re_opcode_t) *p1 == start_memory) |
6957 | { |
6958 | /* We have a new highest active register now. This will |
6959 | get reset at the start_memory we are about to get to, |
6960 | but we will have saved all the registers relevant to |
6961 | this repetition op, as described above. */ |
6962 | highest_active_reg = *(p1 + 1) + *(p1 + 2); |
6963 | if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) |
6964 | lowest_active_reg = *(p1 + 1); |
6965 | } |
6966 | |
6967 | DEBUG_PRINT1 (":\n" ); |
6968 | PUSH_FAILURE_POINT (p + mcnt, d, -2); |
6969 | break; |
6970 | |
6971 | |
6972 | /* A smart repeat ends with `maybe_pop_jump'. |
6973 | We change it to either `pop_failure_jump' or `jump'. */ |
6974 | case maybe_pop_jump: |
6975 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
6976 | DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n" , mcnt); |
6977 | { |
6978 | register UCHAR_T *p2 = p; |
6979 | |
6980 | /* Compare the beginning of the repeat with what in the |
6981 | pattern follows its end. If we can establish that there |
6982 | is nothing that they would both match, i.e., that we |
6983 | would have to backtrack because of (as in, e.g., `a*a') |
6984 | then we can change to pop_failure_jump, because we'll |
6985 | never have to backtrack. |
6986 | |
6987 | This is not true in the case of alternatives: in |
6988 | `(a|ab)*' we do need to backtrack to the `ab' alternative |
6989 | (e.g., if the string was `ab'). But instead of trying to |
6990 | detect that here, the alternative has put on a dummy |
6991 | failure point which is what we will end up popping. */ |
6992 | |
6993 | /* Skip over open/close-group commands. |
6994 | If what follows this loop is a ...+ construct, |
6995 | look at what begins its body, since we will have to |
6996 | match at least one of that. */ |
6997 | while (1) |
6998 | { |
6999 | if (p2 + 2 < pend |
7000 | && ((re_opcode_t) *p2 == stop_memory |
7001 | || (re_opcode_t) *p2 == start_memory)) |
7002 | p2 += 3; |
7003 | else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend |
7004 | && (re_opcode_t) *p2 == dummy_failure_jump) |
7005 | p2 += 2 + 2 * OFFSET_ADDRESS_SIZE; |
7006 | else |
7007 | break; |
7008 | } |
7009 | |
7010 | p1 = p + mcnt; |
7011 | /* p1[0] ... p1[2] are the `on_failure_jump' corresponding |
7012 | to the `maybe_finalize_jump' of this case. Examine what |
7013 | follows. */ |
7014 | |
7015 | /* If we're at the end of the pattern, we can change. */ |
7016 | if (p2 == pend) |
7017 | { |
7018 | /* Consider what happens when matching ":\(.*\)" |
7019 | against ":/". I don't really understand this code |
7020 | yet. */ |
7021 | p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) |
7022 | pop_failure_jump; |
7023 | DEBUG_PRINT1 |
7024 | (" End of pattern: change to `pop_failure_jump'.\n" ); |
7025 | } |
7026 | |
7027 | else if ((re_opcode_t) *p2 == exactn |
7028 | #ifdef MBS_SUPPORT |
7029 | || (re_opcode_t) *p2 == exactn_bin |
7030 | #endif |
7031 | || (bufp->newline_anchor && (re_opcode_t) *p2 == endline)) |
7032 | { |
7033 | register UCHAR_T c |
7034 | = *p2 == (UCHAR_T) endline ? '\n' : p2[2]; |
7035 | |
7036 | if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn |
7037 | #ifdef MBS_SUPPORT |
7038 | || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin |
7039 | #endif |
7040 | ) && p1[3+OFFSET_ADDRESS_SIZE] != c) |
7041 | { |
7042 | p[-(1+OFFSET_ADDRESS_SIZE)] = (UCHAR_T) |
7043 | pop_failure_jump; |
7044 | #ifdef WCHAR |
7045 | DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n" , |
7046 | (wint_t) c, |
7047 | (wint_t) p1[3+OFFSET_ADDRESS_SIZE]); |
7048 | #else |
7049 | DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n" , |
7050 | (char) c, |
7051 | (char) p1[3+OFFSET_ADDRESS_SIZE]); |
7052 | #endif |
7053 | } |
7054 | |
7055 | #ifndef WCHAR |
7056 | else if ((re_opcode_t) p1[3] == charset |
7057 | || (re_opcode_t) p1[3] == charset_not) |
7058 | { |
7059 | int negate = (re_opcode_t) p1[3] == charset_not; |
7060 | |
7061 | if (c < (unsigned) (p1[4] * BYTEWIDTH) |
7062 | && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) |
7063 | negate = !negate; |
7064 | |
7065 | /* `negate' is equal to 1 if c would match, which means |
7066 | that we can't change to pop_failure_jump. */ |
7067 | if (!negate) |
7068 | { |
7069 | p[-3] = (unsigned char) pop_failure_jump; |
7070 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n" ); |
7071 | } |
7072 | } |
7073 | #endif /* not WCHAR */ |
7074 | } |
7075 | #ifndef WCHAR |
7076 | else if ((re_opcode_t) *p2 == charset) |
7077 | { |
7078 | /* We win if the first character of the loop is not part |
7079 | of the charset. */ |
7080 | if ((re_opcode_t) p1[3] == exactn |
7081 | && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5] |
7082 | && (p2[2 + p1[5] / BYTEWIDTH] |
7083 | & (1 << (p1[5] % BYTEWIDTH))))) |
7084 | { |
7085 | p[-3] = (unsigned char) pop_failure_jump; |
7086 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n" ); |
7087 | } |
7088 | |
7089 | else if ((re_opcode_t) p1[3] == charset_not) |
7090 | { |
7091 | int idx; |
7092 | /* We win if the charset_not inside the loop |
7093 | lists every character listed in the charset after. */ |
7094 | for (idx = 0; idx < (int) p2[1]; idx++) |
7095 | if (! (p2[2 + idx] == 0 |
7096 | || (idx < (int) p1[4] |
7097 | && ((p2[2 + idx] & ~ p1[5 + idx]) == 0)))) |
7098 | break; |
7099 | |
7100 | if (idx == p2[1]) |
7101 | { |
7102 | p[-3] = (unsigned char) pop_failure_jump; |
7103 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n" ); |
7104 | } |
7105 | } |
7106 | else if ((re_opcode_t) p1[3] == charset) |
7107 | { |
7108 | int idx; |
7109 | /* We win if the charset inside the loop |
7110 | has no overlap with the one after the loop. */ |
7111 | for (idx = 0; |
7112 | idx < (int) p2[1] && idx < (int) p1[4]; |
7113 | idx++) |
7114 | if ((p2[2 + idx] & p1[5 + idx]) != 0) |
7115 | break; |
7116 | |
7117 | if (idx == p2[1] || idx == p1[4]) |
7118 | { |
7119 | p[-3] = (unsigned char) pop_failure_jump; |
7120 | DEBUG_PRINT1 (" No match => pop_failure_jump.\n" ); |
7121 | } |
7122 | } |
7123 | } |
7124 | #endif /* not WCHAR */ |
7125 | } |
7126 | p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */ |
7127 | if ((re_opcode_t) p[-1] != pop_failure_jump) |
7128 | { |
7129 | p[-1] = (UCHAR_T) jump; |
7130 | DEBUG_PRINT1 (" Match => jump.\n" ); |
7131 | goto unconditional_jump; |
7132 | } |
7133 | /* Fall through. */ |
7134 | |
7135 | |
7136 | /* The end of a simple repeat has a pop_failure_jump back to |
7137 | its matching on_failure_jump, where the latter will push a |
7138 | failure point. The pop_failure_jump takes off failure |
7139 | points put on by this pop_failure_jump's matching |
7140 | on_failure_jump; we got through the pattern to here from the |
7141 | matching on_failure_jump, so didn't fail. */ |
7142 | case pop_failure_jump: |
7143 | { |
7144 | /* We need to pass separate storage for the lowest and |
7145 | highest registers, even though we don't care about the |
7146 | actual values. Otherwise, we will restore only one |
7147 | register from the stack, since lowest will == highest in |
7148 | `pop_failure_point'. */ |
7149 | active_reg_t dummy_low_reg, dummy_high_reg; |
7150 | UCHAR_T *pdummy ATTRIBUTE_UNUSED = NULL; |
7151 | const CHAR_T *sdummy ATTRIBUTE_UNUSED = NULL; |
7152 | |
7153 | DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n" ); |
7154 | POP_FAILURE_POINT (sdummy, pdummy, |
7155 | dummy_low_reg, dummy_high_reg, |
7156 | reg_dummy, reg_dummy, reg_info_dummy); |
7157 | } |
7158 | /* Fall through. */ |
7159 | |
7160 | unconditional_jump: |
7161 | #ifdef _LIBC |
7162 | DEBUG_PRINT2 ("\n%p: " , p); |
7163 | #else |
7164 | DEBUG_PRINT2 ("\n0x%x: " , p); |
7165 | #endif |
7166 | /* Note fall through. */ |
7167 | |
7168 | /* Unconditionally jump (without popping any failure points). */ |
7169 | case jump: |
7170 | EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */ |
7171 | DEBUG_PRINT2 ("EXECUTING jump %d " , mcnt); |
7172 | p += mcnt; /* Do the jump. */ |
7173 | #ifdef _LIBC |
7174 | DEBUG_PRINT2 ("(to %p).\n" , p); |
7175 | #else |
7176 | DEBUG_PRINT2 ("(to 0x%x).\n" , p); |
7177 | #endif |
7178 | break; |
7179 | |
7180 | |
7181 | /* We need this opcode so we can detect where alternatives end |
7182 | in `group_match_null_string_p' et al. */ |
7183 | case jump_past_alt: |
7184 | DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n" ); |
7185 | goto unconditional_jump; |
7186 | |
7187 | |
7188 | /* Normally, the on_failure_jump pushes a failure point, which |
7189 | then gets popped at pop_failure_jump. We will end up at |
7190 | pop_failure_jump, also, and with a pattern of, say, `a+', we |
7191 | are skipping over the on_failure_jump, so we have to push |
7192 | something meaningless for pop_failure_jump to pop. */ |
7193 | case dummy_failure_jump: |
7194 | DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n" ); |
7195 | /* It doesn't matter what we push for the string here. What |
7196 | the code at `fail' tests is the value for the pattern. */ |
7197 | PUSH_FAILURE_POINT (NULL, NULL, -2); |
7198 | goto unconditional_jump; |
7199 | |
7200 | |
7201 | /* At the end of an alternative, we need to push a dummy failure |
7202 | point in case we are followed by a `pop_failure_jump', because |
7203 | we don't want the failure point for the alternative to be |
7204 | popped. For example, matching `(a|ab)*' against `aab' |
7205 | requires that we match the `ab' alternative. */ |
7206 | case push_dummy_failure: |
7207 | DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n" ); |
7208 | /* See comments just above at `dummy_failure_jump' about the |
7209 | two zeroes. */ |
7210 | PUSH_FAILURE_POINT (NULL, NULL, -2); |
7211 | break; |
7212 | |
7213 | /* Have to succeed matching what follows at least n times. |
7214 | After that, handle like `on_failure_jump'. */ |
7215 | case succeed_n: |
7216 | EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); |
7217 | DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n" , mcnt); |
7218 | |
7219 | assert (mcnt >= 0); |
7220 | /* Originally, this is how many times we HAVE to succeed. */ |
7221 | if (mcnt > 0) |
7222 | { |
7223 | mcnt--; |
7224 | p += OFFSET_ADDRESS_SIZE; |
7225 | STORE_NUMBER_AND_INCR (p, mcnt); |
7226 | #ifdef _LIBC |
7227 | DEBUG_PRINT3 (" Setting %p to %d.\n" , p - OFFSET_ADDRESS_SIZE |
7228 | , mcnt); |
7229 | #else |
7230 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n" , p - OFFSET_ADDRESS_SIZE |
7231 | , mcnt); |
7232 | #endif |
7233 | } |
7234 | else if (mcnt == 0) |
7235 | { |
7236 | #ifdef _LIBC |
7237 | DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n" , |
7238 | p + OFFSET_ADDRESS_SIZE); |
7239 | #else |
7240 | DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n" , |
7241 | p + OFFSET_ADDRESS_SIZE); |
7242 | #endif /* _LIBC */ |
7243 | |
7244 | #ifdef WCHAR |
7245 | p[1] = (UCHAR_T) no_op; |
7246 | #else |
7247 | p[2] = (UCHAR_T) no_op; |
7248 | p[3] = (UCHAR_T) no_op; |
7249 | #endif /* WCHAR */ |
7250 | goto on_failure; |
7251 | } |
7252 | break; |
7253 | |
7254 | case jump_n: |
7255 | EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE); |
7256 | DEBUG_PRINT2 ("EXECUTING jump_n %d.\n" , mcnt); |
7257 | |
7258 | /* Originally, this is how many times we CAN jump. */ |
7259 | if (mcnt) |
7260 | { |
7261 | mcnt--; |
7262 | STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt); |
7263 | |
7264 | #ifdef _LIBC |
7265 | DEBUG_PRINT3 (" Setting %p to %d.\n" , p + OFFSET_ADDRESS_SIZE, |
7266 | mcnt); |
7267 | #else |
7268 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n" , p + OFFSET_ADDRESS_SIZE, |
7269 | mcnt); |
7270 | #endif /* _LIBC */ |
7271 | goto unconditional_jump; |
7272 | } |
7273 | /* If don't have to jump any more, skip over the rest of command. */ |
7274 | else |
7275 | p += 2 * OFFSET_ADDRESS_SIZE; |
7276 | break; |
7277 | |
7278 | case set_number_at: |
7279 | { |
7280 | DEBUG_PRINT1 ("EXECUTING set_number_at.\n" ); |
7281 | |
7282 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
7283 | p1 = p + mcnt; |
7284 | EXTRACT_NUMBER_AND_INCR (mcnt, p); |
7285 | #ifdef _LIBC |
7286 | DEBUG_PRINT3 (" Setting %p to %d.\n" , p1, mcnt); |
7287 | #else |
7288 | DEBUG_PRINT3 (" Setting 0x%x to %d.\n" , p1, mcnt); |
7289 | #endif |
7290 | STORE_NUMBER (p1, mcnt); |
7291 | break; |
7292 | } |
7293 | |
7294 | #if 0 |
7295 | /* The DEC Alpha C compiler 3.x generates incorrect code for the |
7296 | test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of |
7297 | AT_WORD_BOUNDARY, so this code is disabled. Expanding the |
7298 | macro and introducing temporary variables works around the bug. */ |
7299 | |
7300 | case wordbound: |
7301 | DEBUG_PRINT1 ("EXECUTING wordbound.\n" ); |
7302 | if (AT_WORD_BOUNDARY (d)) |
7303 | break; |
7304 | goto fail; |
7305 | |
7306 | case notwordbound: |
7307 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n" ); |
7308 | if (AT_WORD_BOUNDARY (d)) |
7309 | goto fail; |
7310 | break; |
7311 | #else |
7312 | case wordbound: |
7313 | { |
7314 | boolean prevchar, thischar; |
7315 | |
7316 | DEBUG_PRINT1 ("EXECUTING wordbound.\n" ); |
7317 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
7318 | break; |
7319 | |
7320 | prevchar = WORDCHAR_P (d - 1); |
7321 | thischar = WORDCHAR_P (d); |
7322 | if (prevchar != thischar) |
7323 | break; |
7324 | goto fail; |
7325 | } |
7326 | |
7327 | case notwordbound: |
7328 | { |
7329 | boolean prevchar, thischar; |
7330 | |
7331 | DEBUG_PRINT1 ("EXECUTING notwordbound.\n" ); |
7332 | if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d)) |
7333 | goto fail; |
7334 | |
7335 | prevchar = WORDCHAR_P (d - 1); |
7336 | thischar = WORDCHAR_P (d); |
7337 | if (prevchar != thischar) |
7338 | goto fail; |
7339 | break; |
7340 | } |
7341 | #endif |
7342 | |
7343 | case wordbeg: |
7344 | DEBUG_PRINT1 ("EXECUTING wordbeg.\n" ); |
7345 | if (!AT_STRINGS_END (d) && WORDCHAR_P (d) |
7346 | && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1))) |
7347 | break; |
7348 | goto fail; |
7349 | |
7350 | case wordend: |
7351 | DEBUG_PRINT1 ("EXECUTING wordend.\n" ); |
7352 | if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1) |
7353 | && (AT_STRINGS_END (d) || !WORDCHAR_P (d))) |
7354 | break; |
7355 | goto fail; |
7356 | |
7357 | #ifdef emacs |
7358 | case before_dot: |
7359 | DEBUG_PRINT1 ("EXECUTING before_dot.\n" ); |
7360 | if (PTR_CHAR_POS ((unsigned char *) d) >= point) |
7361 | goto fail; |
7362 | break; |
7363 | |
7364 | case at_dot: |
7365 | DEBUG_PRINT1 ("EXECUTING at_dot.\n" ); |
7366 | if (PTR_CHAR_POS ((unsigned char *) d) != point) |
7367 | goto fail; |
7368 | break; |
7369 | |
7370 | case after_dot: |
7371 | DEBUG_PRINT1 ("EXECUTING after_dot.\n" ); |
7372 | if (PTR_CHAR_POS ((unsigned char *) d) <= point) |
7373 | goto fail; |
7374 | break; |
7375 | |
7376 | case syntaxspec: |
7377 | DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n" , mcnt); |
7378 | mcnt = *p++; |
7379 | goto matchsyntax; |
7380 | |
7381 | case wordchar: |
7382 | DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n" ); |
7383 | mcnt = (int) Sword; |
7384 | matchsyntax: |
7385 | PREFETCH (); |
7386 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ |
7387 | d++; |
7388 | if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt) |
7389 | goto fail; |
7390 | SET_REGS_MATCHED (); |
7391 | break; |
7392 | |
7393 | case notsyntaxspec: |
7394 | DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n" , mcnt); |
7395 | mcnt = *p++; |
7396 | goto matchnotsyntax; |
7397 | |
7398 | case notwordchar: |
7399 | DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n" ); |
7400 | mcnt = (int) Sword; |
7401 | matchnotsyntax: |
7402 | PREFETCH (); |
7403 | /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ |
7404 | d++; |
7405 | if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt) |
7406 | goto fail; |
7407 | SET_REGS_MATCHED (); |
7408 | break; |
7409 | |
7410 | #else /* not emacs */ |
7411 | case wordchar: |
7412 | DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n" ); |
7413 | PREFETCH (); |
7414 | if (!WORDCHAR_P (d)) |
7415 | goto fail; |
7416 | SET_REGS_MATCHED (); |
7417 | d++; |
7418 | break; |
7419 | |
7420 | case notwordchar: |
7421 | DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n" ); |
7422 | PREFETCH (); |
7423 | if (WORDCHAR_P (d)) |
7424 | goto fail; |
7425 | SET_REGS_MATCHED (); |
7426 | d++; |
7427 | break; |
7428 | #endif /* not emacs */ |
7429 | |
7430 | default: |
7431 | abort (); |
7432 | } |
7433 | continue; /* Successfully executed one pattern command; keep going. */ |
7434 | |
7435 | |
7436 | /* We goto here if a matching operation fails. */ |
7437 | fail: |
7438 | if (!FAIL_STACK_EMPTY ()) |
7439 | { /* A restart point is known. Restore to that state. */ |
7440 | DEBUG_PRINT1 ("\nFAIL:\n" ); |
7441 | POP_FAILURE_POINT (d, p, |
7442 | lowest_active_reg, highest_active_reg, |
7443 | regstart, regend, reg_info); |
7444 | |
7445 | /* If this failure point is a dummy, try the next one. */ |
7446 | if (!p) |
7447 | goto fail; |
7448 | |
7449 | /* If we failed to the end of the pattern, don't examine *p. */ |
7450 | assert (p <= pend); |
7451 | if (p < pend) |
7452 | { |
7453 | boolean is_a_jump_n = false; |
7454 | |
7455 | /* If failed to a backwards jump that's part of a repetition |
7456 | loop, need to pop this failure point and use the next one. */ |
7457 | switch ((re_opcode_t) *p) |
7458 | { |
7459 | case jump_n: |
7460 | is_a_jump_n = true; |
7461 | /* Fall through. */ |
7462 | case maybe_pop_jump: |
7463 | case pop_failure_jump: |
7464 | case jump: |
7465 | p1 = p + 1; |
7466 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7467 | p1 += mcnt; |
7468 | |
7469 | if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n) |
7470 | || (!is_a_jump_n |
7471 | && (re_opcode_t) *p1 == on_failure_jump)) |
7472 | goto fail; |
7473 | break; |
7474 | default: |
7475 | /* do nothing */ ; |
7476 | } |
7477 | } |
7478 | |
7479 | if (d >= string1 && d <= end1) |
7480 | dend = end_match_1; |
7481 | } |
7482 | else |
7483 | break; /* Matching at this starting point really fails. */ |
7484 | } /* for (;;) */ |
7485 | |
7486 | if (best_regs_set) |
7487 | goto restore_best_regs; |
7488 | |
7489 | FREE_VARIABLES (); |
7490 | |
7491 | return -1; /* Failure to match. */ |
7492 | } /* re_match_2 */ |
7493 | |
7494 | /* Subroutine definitions for re_match_2. */ |
7495 | |
7496 | |
7497 | /* We are passed P pointing to a register number after a start_memory. |
7498 | |
7499 | Return true if the pattern up to the corresponding stop_memory can |
7500 | match the empty string, and false otherwise. |
7501 | |
7502 | If we find the matching stop_memory, sets P to point to one past its number. |
7503 | Otherwise, sets P to an undefined byte less than or equal to END. |
7504 | |
7505 | We don't handle duplicates properly (yet). */ |
7506 | |
7507 | static boolean |
7508 | PREFIX(group_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, |
7509 | PREFIX(register_info_type) *reg_info) |
7510 | { |
7511 | int mcnt; |
7512 | /* Point to after the args to the start_memory. */ |
7513 | UCHAR_T *p1 = *p + 2; |
7514 | |
7515 | while (p1 < end) |
7516 | { |
7517 | /* Skip over opcodes that can match nothing, and return true or |
7518 | false, as appropriate, when we get to one that can't, or to the |
7519 | matching stop_memory. */ |
7520 | |
7521 | switch ((re_opcode_t) *p1) |
7522 | { |
7523 | /* Could be either a loop or a series of alternatives. */ |
7524 | case on_failure_jump: |
7525 | p1++; |
7526 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7527 | |
7528 | /* If the next operation is not a jump backwards in the |
7529 | pattern. */ |
7530 | |
7531 | if (mcnt >= 0) |
7532 | { |
7533 | /* Go through the on_failure_jumps of the alternatives, |
7534 | seeing if any of the alternatives cannot match nothing. |
7535 | The last alternative starts with only a jump, |
7536 | whereas the rest start with on_failure_jump and end |
7537 | with a jump, e.g., here is the pattern for `a|b|c': |
7538 | |
7539 | /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 |
7540 | /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 |
7541 | /exactn/1/c |
7542 | |
7543 | So, we have to first go through the first (n-1) |
7544 | alternatives and then deal with the last one separately. */ |
7545 | |
7546 | |
7547 | /* Deal with the first (n-1) alternatives, which start |
7548 | with an on_failure_jump (see above) that jumps to right |
7549 | past a jump_past_alt. */ |
7550 | |
7551 | while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] == |
7552 | jump_past_alt) |
7553 | { |
7554 | /* `mcnt' holds how many bytes long the alternative |
7555 | is, including the ending `jump_past_alt' and |
7556 | its number. */ |
7557 | |
7558 | if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt - |
7559 | (1 + OFFSET_ADDRESS_SIZE), |
7560 | reg_info)) |
7561 | return false; |
7562 | |
7563 | /* Move to right after this alternative, including the |
7564 | jump_past_alt. */ |
7565 | p1 += mcnt; |
7566 | |
7567 | /* Break if it's the beginning of an n-th alternative |
7568 | that doesn't begin with an on_failure_jump. */ |
7569 | if ((re_opcode_t) *p1 != on_failure_jump) |
7570 | break; |
7571 | |
7572 | /* Still have to check that it's not an n-th |
7573 | alternative that starts with an on_failure_jump. */ |
7574 | p1++; |
7575 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7576 | if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] != |
7577 | jump_past_alt) |
7578 | { |
7579 | /* Get to the beginning of the n-th alternative. */ |
7580 | p1 -= 1 + OFFSET_ADDRESS_SIZE; |
7581 | break; |
7582 | } |
7583 | } |
7584 | |
7585 | /* Deal with the last alternative: go back and get number |
7586 | of the `jump_past_alt' just before it. `mcnt' contains |
7587 | the length of the alternative. */ |
7588 | EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE); |
7589 | |
7590 | if (!PREFIX(alt_match_null_string_p) (p1, p1 + mcnt, reg_info)) |
7591 | return false; |
7592 | |
7593 | p1 += mcnt; /* Get past the n-th alternative. */ |
7594 | } /* if mcnt > 0 */ |
7595 | break; |
7596 | |
7597 | |
7598 | case stop_memory: |
7599 | assert (p1[1] == **p); |
7600 | *p = p1 + 2; |
7601 | return true; |
7602 | |
7603 | |
7604 | default: |
7605 | if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) |
7606 | return false; |
7607 | } |
7608 | } /* while p1 < end */ |
7609 | |
7610 | return false; |
7611 | } /* group_match_null_string_p */ |
7612 | |
7613 | |
7614 | /* Similar to group_match_null_string_p, but doesn't deal with alternatives: |
7615 | It expects P to be the first byte of a single alternative and END one |
7616 | byte past the last. The alternative can contain groups. */ |
7617 | |
7618 | static boolean |
7619 | PREFIX(alt_match_null_string_p) (UCHAR_T *p, UCHAR_T *end, |
7620 | PREFIX(register_info_type) *reg_info) |
7621 | { |
7622 | int mcnt; |
7623 | UCHAR_T *p1 = p; |
7624 | |
7625 | while (p1 < end) |
7626 | { |
7627 | /* Skip over opcodes that can match nothing, and break when we get |
7628 | to one that can't. */ |
7629 | |
7630 | switch ((re_opcode_t) *p1) |
7631 | { |
7632 | /* It's a loop. */ |
7633 | case on_failure_jump: |
7634 | p1++; |
7635 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7636 | p1 += mcnt; |
7637 | break; |
7638 | |
7639 | default: |
7640 | if (!PREFIX(common_op_match_null_string_p) (&p1, end, reg_info)) |
7641 | return false; |
7642 | } |
7643 | } /* while p1 < end */ |
7644 | |
7645 | return true; |
7646 | } /* alt_match_null_string_p */ |
7647 | |
7648 | |
7649 | /* Deals with the ops common to group_match_null_string_p and |
7650 | alt_match_null_string_p. |
7651 | |
7652 | Sets P to one after the op and its arguments, if any. */ |
7653 | |
7654 | static boolean |
7655 | PREFIX(common_op_match_null_string_p) (UCHAR_T **p, UCHAR_T *end, |
7656 | PREFIX(register_info_type) *reg_info) |
7657 | { |
7658 | int mcnt; |
7659 | boolean ret; |
7660 | int reg_no; |
7661 | UCHAR_T *p1 = *p; |
7662 | |
7663 | switch ((re_opcode_t) *p1++) |
7664 | { |
7665 | case no_op: |
7666 | case begline: |
7667 | case endline: |
7668 | case begbuf: |
7669 | case endbuf: |
7670 | case wordbeg: |
7671 | case wordend: |
7672 | case wordbound: |
7673 | case notwordbound: |
7674 | #ifdef emacs |
7675 | case before_dot: |
7676 | case at_dot: |
7677 | case after_dot: |
7678 | #endif |
7679 | break; |
7680 | |
7681 | case start_memory: |
7682 | reg_no = *p1; |
7683 | assert (reg_no > 0 && reg_no <= MAX_REGNUM); |
7684 | ret = PREFIX(group_match_null_string_p) (&p1, end, reg_info); |
7685 | |
7686 | /* Have to set this here in case we're checking a group which |
7687 | contains a group and a back reference to it. */ |
7688 | |
7689 | if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE) |
7690 | REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret; |
7691 | |
7692 | if (!ret) |
7693 | return false; |
7694 | break; |
7695 | |
7696 | /* If this is an optimized succeed_n for zero times, make the jump. */ |
7697 | case jump: |
7698 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7699 | if (mcnt >= 0) |
7700 | p1 += mcnt; |
7701 | else |
7702 | return false; |
7703 | break; |
7704 | |
7705 | case succeed_n: |
7706 | /* Get to the number of times to succeed. */ |
7707 | p1 += OFFSET_ADDRESS_SIZE; |
7708 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7709 | |
7710 | if (mcnt == 0) |
7711 | { |
7712 | p1 -= 2 * OFFSET_ADDRESS_SIZE; |
7713 | EXTRACT_NUMBER_AND_INCR (mcnt, p1); |
7714 | p1 += mcnt; |
7715 | } |
7716 | else |
7717 | return false; |
7718 | break; |
7719 | |
7720 | case duplicate: |
7721 | if (!REG_MATCH_NULL_STRING_P (reg_info[*p1])) |
7722 | return false; |
7723 | break; |
7724 | |
7725 | case set_number_at: |
7726 | p1 += 2 * OFFSET_ADDRESS_SIZE; |
7727 | return false; |
7728 | |
7729 | default: |
7730 | /* All other opcodes mean we cannot match the empty string. */ |
7731 | return false; |
7732 | } |
7733 | |
7734 | *p = p1; |
7735 | return true; |
7736 | } /* common_op_match_null_string_p */ |
7737 | |
7738 | |
7739 | /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN |
7740 | bytes; nonzero otherwise. */ |
7741 | |
7742 | static int |
7743 | PREFIX(bcmp_translate) (const CHAR_T *s1, const CHAR_T *s2, register int len, |
7744 | RE_TRANSLATE_TYPE translate) |
7745 | { |
7746 | register const UCHAR_T *p1 = (const UCHAR_T *) s1; |
7747 | register const UCHAR_T *p2 = (const UCHAR_T *) s2; |
7748 | while (len) |
7749 | { |
7750 | #ifdef WCHAR |
7751 | if (((*p1<=0xff)?translate[*p1++]:*p1++) |
7752 | != ((*p2<=0xff)?translate[*p2++]:*p2++)) |
7753 | return 1; |
7754 | #else /* BYTE */ |
7755 | if (translate[*p1++] != translate[*p2++]) return 1; |
7756 | #endif /* WCHAR */ |
7757 | len--; |
7758 | } |
7759 | return 0; |
7760 | } |
7761 | |
7762 | |
7763 | #else /* not INSIDE_RECURSION */ |
7764 | |
7765 | /* Entry points for GNU code. */ |
7766 | |
7767 | /* re_compile_pattern is the GNU regular expression compiler: it |
7768 | compiles PATTERN (of length SIZE) and puts the result in BUFP. |
7769 | Returns 0 if the pattern was valid, otherwise an error string. |
7770 | |
7771 | Assumes the `allocated' (and perhaps `buffer') and `translate' fields |
7772 | are set in BUFP on entry. |
7773 | |
7774 | We call regex_compile to do the actual compilation. */ |
7775 | |
7776 | const char * |
7777 | re_compile_pattern (const char *pattern, size_t length, |
7778 | struct re_pattern_buffer *bufp) |
7779 | { |
7780 | reg_errcode_t ret; |
7781 | |
7782 | /* GNU code is written to assume at least RE_NREGS registers will be set |
7783 | (and at least one extra will be -1). */ |
7784 | bufp->regs_allocated = REGS_UNALLOCATED; |
7785 | |
7786 | /* And GNU code determines whether or not to get register information |
7787 | by passing null for the REGS argument to re_match, etc., not by |
7788 | setting no_sub. */ |
7789 | bufp->no_sub = 0; |
7790 | |
7791 | /* Match anchors at newline. */ |
7792 | bufp->newline_anchor = 1; |
7793 | |
7794 | # ifdef MBS_SUPPORT |
7795 | if (MB_CUR_MAX != 1) |
7796 | ret = wcs_regex_compile (pattern, length, re_syntax_options, bufp); |
7797 | else |
7798 | # endif |
7799 | ret = byte_regex_compile (pattern, size: length, re_syntax_options, bufp); |
7800 | |
7801 | if (!ret) |
7802 | return NULL; |
7803 | return gettext (re_error_msgid[(int) ret]); |
7804 | } |
7805 | #ifdef _LIBC |
7806 | weak_alias (__re_compile_pattern, re_compile_pattern) |
7807 | #endif |
7808 | |
7809 | /* Entry points compatible with 4.2 BSD regex library. We don't define |
7810 | them unless specifically requested. */ |
7811 | |
7812 | #if defined _REGEX_RE_COMP || defined _LIBC |
7813 | |
7814 | /* BSD has one and only one pattern buffer. */ |
7815 | static struct re_pattern_buffer re_comp_buf; |
7816 | |
7817 | char * |
7818 | #ifdef _LIBC |
7819 | /* Make these definitions weak in libc, so POSIX programs can redefine |
7820 | these names if they don't use our functions, and still use |
7821 | regcomp/regexec below without link errors. */ |
7822 | weak_function |
7823 | #endif |
7824 | re_comp (const char *s) |
7825 | { |
7826 | reg_errcode_t ret; |
7827 | |
7828 | if (!s) |
7829 | { |
7830 | if (!re_comp_buf.buffer) |
7831 | return (char *) gettext ("No previous regular expression" ); |
7832 | return 0; |
7833 | } |
7834 | |
7835 | if (!re_comp_buf.buffer) |
7836 | { |
7837 | re_comp_buf.buffer = (unsigned char *) malloc (size: 200); |
7838 | if (re_comp_buf.buffer == NULL) |
7839 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); |
7840 | re_comp_buf.allocated = 200; |
7841 | |
7842 | re_comp_buf.fastmap = (char *) malloc (size: 1 << BYTEWIDTH); |
7843 | if (re_comp_buf.fastmap == NULL) |
7844 | return (char *) gettext (re_error_msgid[(int) REG_ESPACE]); |
7845 | } |
7846 | |
7847 | /* Since `re_exec' always passes NULL for the `regs' argument, we |
7848 | don't need to initialize the pattern buffer fields which affect it. */ |
7849 | |
7850 | /* Match anchors at newlines. */ |
7851 | re_comp_buf.newline_anchor = 1; |
7852 | |
7853 | # ifdef MBS_SUPPORT |
7854 | if (MB_CUR_MAX != 1) |
7855 | ret = wcs_regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf); |
7856 | else |
7857 | # endif |
7858 | ret = byte_regex_compile (pattern: s, size: strlen (s: s), re_syntax_options, bufp: &re_comp_buf); |
7859 | |
7860 | if (!ret) |
7861 | return NULL; |
7862 | |
7863 | /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ |
7864 | return (char *) gettext (re_error_msgid[(int) ret]); |
7865 | } |
7866 | |
7867 | |
7868 | int |
7869 | #ifdef _LIBC |
7870 | weak_function |
7871 | #endif |
7872 | re_exec (const char *s) |
7873 | { |
7874 | const int len = strlen (s: s); |
7875 | return |
7876 | 0 <= re_search (bufp: &re_comp_buf, string: s, size: len, startpos: 0, range: len, regs: (struct re_registers *) 0); |
7877 | } |
7878 | |
7879 | #endif /* _REGEX_RE_COMP */ |
7880 | |
7881 | /* POSIX.2 functions. Don't define these for Emacs. */ |
7882 | |
7883 | #ifndef emacs |
7884 | |
7885 | /* regcomp takes a regular expression as a string and compiles it. |
7886 | |
7887 | PREG is a regex_t *. We do not expect any fields to be initialized, |
7888 | since POSIX says we shouldn't. Thus, we set |
7889 | |
7890 | `buffer' to the compiled pattern; |
7891 | `used' to the length of the compiled pattern; |
7892 | `syntax' to RE_SYNTAX_POSIX_EXTENDED if the |
7893 | REG_EXTENDED bit in CFLAGS is set; otherwise, to |
7894 | RE_SYNTAX_POSIX_BASIC; |
7895 | `newline_anchor' to REG_NEWLINE being set in CFLAGS; |
7896 | `fastmap' to an allocated space for the fastmap; |
7897 | `fastmap_accurate' to zero; |
7898 | `re_nsub' to the number of subexpressions in PATTERN. |
7899 | |
7900 | PATTERN is the address of the pattern string. |
7901 | |
7902 | CFLAGS is a series of bits which affect compilation. |
7903 | |
7904 | If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we |
7905 | use POSIX basic syntax. |
7906 | |
7907 | If REG_NEWLINE is set, then . and [^...] don't match newline. |
7908 | Also, regexec will try a match beginning after every newline. |
7909 | |
7910 | If REG_ICASE is set, then we considers upper- and lowercase |
7911 | versions of letters to be equivalent when matching. |
7912 | |
7913 | If REG_NOSUB is set, then when PREG is passed to regexec, that |
7914 | routine will report only success or failure, and nothing about the |
7915 | registers. |
7916 | |
7917 | It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for |
7918 | the return codes and their meanings.) */ |
7919 | |
7920 | int |
7921 | regcomp (regex_t *preg, const char *pattern, int cflags) |
7922 | { |
7923 | reg_errcode_t ret; |
7924 | reg_syntax_t syntax |
7925 | = (cflags & REG_EXTENDED) ? |
7926 | RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; |
7927 | |
7928 | /* regex_compile will allocate the space for the compiled pattern. */ |
7929 | preg->buffer = 0; |
7930 | preg->allocated = 0; |
7931 | preg->used = 0; |
7932 | |
7933 | /* Try to allocate space for the fastmap. */ |
7934 | preg->fastmap = (char *) malloc (size: 1 << BYTEWIDTH); |
7935 | |
7936 | if (cflags & REG_ICASE) |
7937 | { |
7938 | int i; |
7939 | |
7940 | preg->translate |
7941 | = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE |
7942 | * sizeof (*(RE_TRANSLATE_TYPE)0)); |
7943 | if (preg->translate == NULL) |
7944 | return (int) REG_ESPACE; |
7945 | |
7946 | /* Map uppercase characters to corresponding lowercase ones. */ |
7947 | for (i = 0; i < CHAR_SET_SIZE; i++) |
7948 | preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i; |
7949 | } |
7950 | else |
7951 | preg->translate = NULL; |
7952 | |
7953 | /* If REG_NEWLINE is set, newlines are treated differently. */ |
7954 | if (cflags & REG_NEWLINE) |
7955 | { /* REG_NEWLINE implies neither . nor [^...] match newline. */ |
7956 | syntax &= ~RE_DOT_NEWLINE; |
7957 | syntax |= RE_HAT_LISTS_NOT_NEWLINE; |
7958 | /* It also changes the matching behavior. */ |
7959 | preg->newline_anchor = 1; |
7960 | } |
7961 | else |
7962 | preg->newline_anchor = 0; |
7963 | |
7964 | preg->no_sub = !!(cflags & REG_NOSUB); |
7965 | |
7966 | /* POSIX says a null character in the pattern terminates it, so we |
7967 | can use strlen here in compiling the pattern. */ |
7968 | # ifdef MBS_SUPPORT |
7969 | if (MB_CUR_MAX != 1) |
7970 | ret = wcs_regex_compile (pattern, strlen (pattern), syntax, preg); |
7971 | else |
7972 | # endif |
7973 | ret = byte_regex_compile (pattern, size: strlen (s: pattern), syntax, bufp: preg); |
7974 | |
7975 | /* POSIX doesn't distinguish between an unmatched open-group and an |
7976 | unmatched close-group: both are REG_EPAREN. */ |
7977 | if (ret == REG_ERPAREN) ret = REG_EPAREN; |
7978 | |
7979 | if (ret == REG_NOERROR && preg->fastmap) |
7980 | { |
7981 | /* Compute the fastmap now, since regexec cannot modify the pattern |
7982 | buffer. */ |
7983 | if (re_compile_fastmap (bufp: preg) == -2) |
7984 | { |
7985 | /* Some error occurred while computing the fastmap, just forget |
7986 | about it. */ |
7987 | free (ptr: preg->fastmap); |
7988 | preg->fastmap = NULL; |
7989 | } |
7990 | } |
7991 | |
7992 | return (int) ret; |
7993 | } |
7994 | #ifdef _LIBC |
7995 | weak_alias (__regcomp, regcomp) |
7996 | #endif |
7997 | |
7998 | |
7999 | /* regexec searches for a given pattern, specified by PREG, in the |
8000 | string STRING. |
8001 | |
8002 | If NMATCH is zero or REG_NOSUB was set in the cflags argument to |
8003 | `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at |
8004 | least NMATCH elements, and we set them to the offsets of the |
8005 | corresponding matched substrings. |
8006 | |
8007 | EFLAGS specifies `execution flags' which affect matching: if |
8008 | REG_NOTBOL is set, then ^ does not match at the beginning of the |
8009 | string; if REG_NOTEOL is set, then $ does not match at the end. |
8010 | |
8011 | We return 0 if we find a match and REG_NOMATCH if not. */ |
8012 | |
8013 | int |
8014 | regexec (const regex_t *preg, const char *string, size_t nmatch, |
8015 | regmatch_t pmatch[], int eflags) |
8016 | { |
8017 | int ret; |
8018 | struct re_registers regs; |
8019 | regex_t private_preg; |
8020 | int len = strlen (s: string); |
8021 | boolean want_reg_info = !preg->no_sub && nmatch > 0; |
8022 | |
8023 | private_preg = *preg; |
8024 | |
8025 | private_preg.not_bol = !!(eflags & REG_NOTBOL); |
8026 | private_preg.not_eol = !!(eflags & REG_NOTEOL); |
8027 | |
8028 | /* The user has told us exactly how many registers to return |
8029 | information about, via `nmatch'. We have to pass that on to the |
8030 | matching routines. */ |
8031 | private_preg.regs_allocated = REGS_FIXED; |
8032 | |
8033 | if (want_reg_info) |
8034 | { |
8035 | regs.num_regs = nmatch; |
8036 | regs.start = TALLOC (nmatch * 2, regoff_t); |
8037 | if (regs.start == NULL) |
8038 | return (int) REG_NOMATCH; |
8039 | regs.end = regs.start + nmatch; |
8040 | } |
8041 | |
8042 | /* Perform the searching operation. */ |
8043 | ret = re_search (bufp: &private_preg, string, size: len, |
8044 | /* start: */ startpos: 0, /* range: */ range: len, |
8045 | regs: want_reg_info ? ®s : (struct re_registers *) 0); |
8046 | |
8047 | /* Copy the register information to the POSIX structure. */ |
8048 | if (want_reg_info) |
8049 | { |
8050 | if (ret >= 0) |
8051 | { |
8052 | unsigned r; |
8053 | |
8054 | for (r = 0; r < nmatch; r++) |
8055 | { |
8056 | pmatch[r].rm_so = regs.start[r]; |
8057 | pmatch[r].rm_eo = regs.end[r]; |
8058 | } |
8059 | } |
8060 | |
8061 | /* If we needed the temporary register info, free the space now. */ |
8062 | free (ptr: regs.start); |
8063 | } |
8064 | |
8065 | /* We want zero return to mean success, unlike `re_search'. */ |
8066 | return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; |
8067 | } |
8068 | #ifdef _LIBC |
8069 | weak_alias (__regexec, regexec) |
8070 | #endif |
8071 | |
8072 | |
8073 | /* Returns a message corresponding to an error code, ERRCODE, returned |
8074 | from either regcomp or regexec. We don't use PREG here. */ |
8075 | |
8076 | size_t |
8077 | regerror (int errcode, const regex_t *preg ATTRIBUTE_UNUSED, |
8078 | char *errbuf, size_t errbuf_size) |
8079 | { |
8080 | const char *msg; |
8081 | size_t msg_size; |
8082 | |
8083 | if (errcode < 0 |
8084 | || errcode >= (int) (sizeof (re_error_msgid) |
8085 | / sizeof (re_error_msgid[0]))) |
8086 | /* Only error codes returned by the rest of the code should be passed |
8087 | to this routine. If we are given anything else, or if other regex |
8088 | code generates an invalid error code, then the program has a bug. |
8089 | Dump core so we can fix it. */ |
8090 | abort (); |
8091 | |
8092 | msg = gettext (re_error_msgid[errcode]); |
8093 | |
8094 | msg_size = strlen (s: msg) + 1; /* Includes the null. */ |
8095 | |
8096 | if (errbuf_size != 0) |
8097 | { |
8098 | if (msg_size > errbuf_size) |
8099 | { |
8100 | #if defined HAVE_MEMPCPY || defined _LIBC |
8101 | *((char *) mempcpy (errbuf, msg, errbuf_size - 1)) = '\0'; |
8102 | #else |
8103 | (void) memcpy (dest: errbuf, src: msg, n: errbuf_size - 1); |
8104 | errbuf[errbuf_size - 1] = 0; |
8105 | #endif |
8106 | } |
8107 | else |
8108 | (void) memcpy (dest: errbuf, src: msg, n: msg_size); |
8109 | } |
8110 | |
8111 | return msg_size; |
8112 | } |
8113 | #ifdef _LIBC |
8114 | weak_alias (__regerror, regerror) |
8115 | #endif |
8116 | |
8117 | |
8118 | /* Free dynamically allocated space used by PREG. */ |
8119 | |
8120 | void |
8121 | regfree (regex_t *preg) |
8122 | { |
8123 | free (ptr: preg->buffer); |
8124 | preg->buffer = NULL; |
8125 | |
8126 | preg->allocated = 0; |
8127 | preg->used = 0; |
8128 | |
8129 | free (ptr: preg->fastmap); |
8130 | preg->fastmap = NULL; |
8131 | preg->fastmap_accurate = 0; |
8132 | |
8133 | free (ptr: preg->translate); |
8134 | preg->translate = NULL; |
8135 | } |
8136 | #ifdef _LIBC |
8137 | weak_alias (__regfree, regfree) |
8138 | #endif |
8139 | |
8140 | #endif /* not emacs */ |
8141 | |
8142 | #endif /* not INSIDE_RECURSION */ |
8143 | |
8144 | |
8145 | #undef STORE_NUMBER |
8146 | #undef STORE_NUMBER_AND_INCR |
8147 | #undef EXTRACT_NUMBER |
8148 | #undef EXTRACT_NUMBER_AND_INCR |
8149 | |
8150 | #undef DEBUG_PRINT_COMPILED_PATTERN |
8151 | #undef DEBUG_PRINT_DOUBLE_STRING |
8152 | |
8153 | #undef INIT_FAIL_STACK |
8154 | #undef RESET_FAIL_STACK |
8155 | #undef DOUBLE_FAIL_STACK |
8156 | #undef PUSH_PATTERN_OP |
8157 | #undef PUSH_FAILURE_POINTER |
8158 | #undef PUSH_FAILURE_INT |
8159 | #undef PUSH_FAILURE_ELT |
8160 | #undef POP_FAILURE_POINTER |
8161 | #undef POP_FAILURE_INT |
8162 | #undef POP_FAILURE_ELT |
8163 | #undef DEBUG_PUSH |
8164 | #undef DEBUG_POP |
8165 | #undef PUSH_FAILURE_POINT |
8166 | #undef POP_FAILURE_POINT |
8167 | |
8168 | #undef REG_UNSET_VALUE |
8169 | #undef REG_UNSET |
8170 | |
8171 | #undef PATFETCH |
8172 | #undef PATFETCH_RAW |
8173 | #undef PATUNFETCH |
8174 | #undef TRANSLATE |
8175 | |
8176 | #undef INIT_BUF_SIZE |
8177 | #undef GET_BUFFER_SPACE |
8178 | #undef BUF_PUSH |
8179 | #undef BUF_PUSH_2 |
8180 | #undef BUF_PUSH_3 |
8181 | #undef STORE_JUMP |
8182 | #undef STORE_JUMP2 |
8183 | #undef INSERT_JUMP |
8184 | #undef INSERT_JUMP2 |
8185 | #undef EXTEND_BUFFER |
8186 | #undef GET_UNSIGNED_NUMBER |
8187 | #undef FREE_STACK_RETURN |
8188 | |
8189 | # undef POINTER_TO_OFFSET |
8190 | # undef MATCHING_IN_FRST_STRING |
8191 | # undef PREFETCH |
8192 | # undef AT_STRINGS_BEG |
8193 | # undef AT_STRINGS_END |
8194 | # undef WORDCHAR_P |
8195 | # undef FREE_VAR |
8196 | # undef FREE_VARIABLES |
8197 | # undef NO_HIGHEST_ACTIVE_REG |
8198 | # undef NO_LOWEST_ACTIVE_REG |
8199 | |
8200 | # undef CHAR_T |
8201 | # undef UCHAR_T |
8202 | # undef COMPILED_BUFFER_VAR |
8203 | # undef OFFSET_ADDRESS_SIZE |
8204 | # undef CHAR_CLASS_SIZE |
8205 | # undef PREFIX |
8206 | # undef ARG_PREFIX |
8207 | # undef PUT_CHAR |
8208 | # undef BYTE |
8209 | # undef WCHAR |
8210 | |
8211 | # define DEFINED_ONCE |
8212 | |