1 /* $NetBSD: regex.c,v 1.3 2020/09/26 11:39:17 mlelstv Exp $ */
2
3 /* Extended regular expression matching and search library,
4 version 0.12.
5 (Implements POSIX draft P1003.2/D11.2, except for some of the
6 internationalization features.)
7 Copyright (C) 1993-1999, 2000, 2001 Free Software Foundation, Inc.
8
9 The GNU C Library is free software; you can redistribute it and/or
10 modify it under the terms of the GNU Library General Public License as
11 published by the Free Software Foundation; either version 2 of the
12 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 Library General Public License for more details.
18
19 You should have received a copy of the GNU Library General Public
20 License along with the GNU C Library; see the file COPYING.LIB. If not,
21 write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
22 Boston, MA 02111-1307, USA. */
23
24 /* AIX requires this to be the first thing in the file. */
25 #if defined _AIX && !defined REGEX_MALLOC
26 #pragma alloca
27 #endif
28
29 #undef _GNU_SOURCE
30 #define _GNU_SOURCE
31
32 #ifdef HAVE_CONFIG_H
33 # include <config.h>
34 #endif
35
36 #ifndef PARAMS
37 # if defined __GNUC__ || (defined __STDC__ && __STDC__)
38 # define PARAMS(args) args
39 # else
40 # define PARAMS(args) ()
41 # endif /* GCC. */
42 #endif /* Not PARAMS. */
43
44 #if defined STDC_HEADERS && !defined emacs
45 # include <stddef.h>
46 #else
47 /* We need this for `regex.h', and perhaps for the Emacs include files. */
48 # include <sys/types.h>
49 #endif
50
51 #define WIDE_CHAR_SUPPORT (HAVE_WCTYPE_H && HAVE_WCHAR_H && HAVE_BTOWC)
52
53 /* For platform which support the ISO C amendement 1 functionality we
54 support user defined character classes. */
55 #if defined _LIBC || WIDE_CHAR_SUPPORT
56 /* Solaris 2.5 has a bug: <wchar.h> must be included before <wctype.h>. */
57 # include <wchar.h>
58 # include <wctype.h>
59 #endif
60
61 /* This is for multi byte string support. */
62 #ifdef MBS_SUPPORT
63 # define CHAR_TYPE wchar_t
64 # define US_CHAR_TYPE wchar_t/* unsigned character type */
65 # define COMPILED_BUFFER_VAR wc_buffer
66 # define OFFSET_ADDRESS_SIZE 1 /* the size which STORE_NUMBER macro use */
67 # define CHAR_CLASS_SIZE ((__alignof__(wctype_t)+sizeof(wctype_t))/sizeof(CHAR_TYPE)+1)
68 # define PUT_CHAR(c) \
69 do { \
70 if (MB_CUR_MAX == 1) \
71 putchar (c); \
72 else \
73 printf ("%C", (wint_t) c); /* Should we use wide stream?? */ \
74 } while (0)
75 # define TRUE 1
76 # define FALSE 0
77 #else
78 # define CHAR_TYPE char
79 # define US_CHAR_TYPE unsigned char /* unsigned character type */
80 # define COMPILED_BUFFER_VAR bufp->buffer
81 # define OFFSET_ADDRESS_SIZE 2
82 # define PUT_CHAR(c) putchar (c)
83 #endif /* MBS_SUPPORT */
84
85 #ifdef _LIBC
86 /* We have to keep the namespace clean. */
87 # define regfree(preg) __regfree (preg)
88 # define regexec(pr, st, nm, pm, ef) __regexec (pr, st, nm, pm, ef)
89 # define regcomp(preg, pattern, cflags) __regcomp (preg, pattern, cflags)
90 # define regerror(errcode, preg, errbuf, errbuf_size) \
91 __regerror(errcode, preg, errbuf, errbuf_size)
92 # define re_set_registers(bu, re, nu, st, en) \
93 __re_set_registers (bu, re, nu, st, en)
94 # define re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) \
95 __re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
96 # define re_match(bufp, string, size, pos, regs) \
97 __re_match (bufp, string, size, pos, regs)
98 # define re_search(bufp, string, size, startpos, range, regs) \
99 __re_search (bufp, string, size, startpos, range, regs)
100 # define re_compile_pattern(pattern, length, bufp) \
101 __re_compile_pattern (pattern, length, bufp)
102 # define re_set_syntax(syntax) __re_set_syntax (syntax)
103 # define re_search_2(bufp, st1, s1, st2, s2, startpos, range, regs, stop) \
104 __re_search_2 (bufp, st1, s1, st2, s2, startpos, range, regs, stop)
105 # define re_compile_fastmap(bufp) __re_compile_fastmap (bufp)
106
107 # define btowc __btowc
108
109 /* We are also using some library internals. */
110 # include <locale/localeinfo.h>
111 # include <locale/elem-hash.h>
112 # include <langinfo.h>
113 # include <locale/coll-lookup.h>
114 #endif
115
116 /* This is for other GNU distributions with internationalized messages. */
117 #if HAVE_LIBINTL_H || defined _LIBC
118 # include <libintl.h>
119 # ifdef _LIBC
120 # undef gettext
121 # define gettext(msgid) __dcgettext ("libc", msgid, LC_MESSAGES)
122 # endif
123 #else
124 # define gettext(msgid) (msgid)
125 #endif
126
127 #ifndef gettext_noop
128 /* This define is so xgettext can find the internationalizable
129 strings. */
130 # define gettext_noop(String) String
131 #endif
132
133 /* The `emacs' switch turns on certain matching commands
134 that make sense only in Emacs. */
135 #ifdef emacs
136
137 # include "lisp.h"
138 # include "buffer.h"
139 # include "syntax.h"
140
141 #else /* not emacs */
142
143 /* If we are not linking with Emacs proper,
144 we can't use the relocating allocator
145 even if config.h says that we can. */
146 # undef REL_ALLOC
147
148 # if defined STDC_HEADERS || defined _LIBC
149 # include <stdlib.h>
150 # else
151 char *malloc ();
152 char *realloc ();
153 # endif
154
155 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
156 If nothing else has been done, use the method below. */
157 # ifdef INHIBIT_STRING_HEADER
158 # if !(defined HAVE_BZERO && defined HAVE_BCOPY)
159 # if !defined bzero && !defined bcopy
160 # undef INHIBIT_STRING_HEADER
161 # endif
162 # endif
163 # endif
164
165 /* This is the normal way of making sure we have a bcopy and a bzero.
166 This is used in most programs--a few other programs avoid this
167 by defining INHIBIT_STRING_HEADER. */
168 # ifndef INHIBIT_STRING_HEADER
169 # if defined HAVE_STRING_H || defined STDC_HEADERS || defined _LIBC
170 # include <string.h>
171 # ifndef bzero
172 # ifndef _LIBC
173 # define bzero(s, n) (memset (s, '\0', n), (s))
174 # else
175 # define bzero(s, n) __bzero (s, n)
176 # endif
177 # endif
178 # else
179 # include <strings.h>
180 # ifndef memcmp
181 # define memcmp(s1, s2, n) bcmp (s1, s2, n)
182 # endif
183 # ifndef memcpy
184 # define memcpy(d, s, n) (bcopy (s, d, n), (d))
185 # endif
186 # endif
187 # endif
188
189 /* Define the syntax stuff for \<, \>, etc. */
190
191 /* This must be nonzero for the wordchar and notwordchar pattern
192 commands in re_match_2. */
193 # ifndef Sword
194 # define Sword 1
195 # endif
196
197 # ifdef SWITCH_ENUM_BUG
198 # define SWITCH_ENUM_CAST(x) ((int)(x))
199 # else
200 # define SWITCH_ENUM_CAST(x) (x)
201 # endif
202
203 #endif /* not emacs */
204
205 #if defined _LIBC || HAVE_LIMITS_H
206 # include <limits.h>
207 #endif
208
209 #ifndef MB_LEN_MAX
210 # define MB_LEN_MAX 1
211 #endif
212
213 /* Get the interface, including the syntax bits. */
214 #include <regex.h>
215
216 /* isalpha etc. are used for the character classes. */
217 #include <ctype.h>
218
219 /* Jim Meyering writes:
220
221 "... Some ctype macros are valid only for character codes that
222 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
223 using /bin/cc or gcc but without giving an ansi option). So, all
224 ctype uses should be through macros like ISPRINT... If
225 STDC_HEADERS is defined, then autoconf has verified that the ctype
226 macros don't need to be guarded with references to isascii. ...
227 Defining isascii to 1 should let any compiler worth its salt
228 eliminate the && through constant folding."
229 Solaris defines some of these symbols so we must undefine them first. */
230
231 #undef ISASCII
232 #if defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
233 # define ISASCII(c) 1
234 #else
235 # define ISASCII(c) isascii(c)
236 #endif
237
238 #ifdef isblank
239 # define ISBLANK(c) (ISASCII (c) && isblank (c))
240 #else
241 # define ISBLANK(c) ((c) == ' ' || (c) == '\t')
242 #endif
243 #ifdef isgraph
244 # define ISGRAPH(c) (ISASCII (c) && isgraph (c))
245 #else
246 # define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
247 #endif
248
249 #undef ISPRINT
250 #define ISPRINT(c) (ISASCII (c) && isprint (c))
251 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
252 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
253 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
254 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
255 #define ISLOWER(c) (ISASCII (c) && islower (c))
256 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
257 #define ISSPACE(c) (ISASCII (c) && isspace (c))
258 #define ISUPPER(c) (ISASCII (c) && isupper (c))
259 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
260
261 #ifdef _tolower
262 # define TOLOWER(c) _tolower(c)
263 #else
264 # define TOLOWER(c) tolower(c)
265 #endif
266
267 #ifndef NULL
268 # define NULL (void *)0
269 #endif
270
271 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
272 since ours (we hope) works properly with all combinations of
273 machines, compilers, `char' and `unsigned char' argument types.
274 (Per Bothner suggested the basic approach.) */
275 #undef SIGN_EXTEND_CHAR
276 #if __STDC__
277 # define SIGN_EXTEND_CHAR(c) ((signed char) (c))
278 #else /* not __STDC__ */
279 /* As in Harbison and Steele. */
280 # define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
281 #endif
282
283 #ifndef emacs
284 /* How many characters in the character set. */
285 # define CHAR_SET_SIZE 256
286
287 # ifdef SYNTAX_TABLE
288
289 extern char *re_syntax_table;
290
291 # else /* not SYNTAX_TABLE */
292
293 static char re_syntax_table[CHAR_SET_SIZE];
294
295 static void init_syntax_once PARAMS ((void));
296
297 static void
init_syntax_once()298 init_syntax_once ()
299 {
300 register int c;
301 static int done = 0;
302
303 if (done)
304 return;
305 bzero (re_syntax_table, sizeof re_syntax_table);
306
307 for (c = 0; c < CHAR_SET_SIZE; ++c)
308 if (ISALNUM (c))
309 re_syntax_table[c] = Sword;
310
311 re_syntax_table['_'] = Sword;
312
313 done = 1;
314 }
315
316 # endif /* not SYNTAX_TABLE */
317
318 # define SYNTAX(c) re_syntax_table[(unsigned char) (c)]
319
320 #endif /* emacs */
321
322 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
323 use `alloca' instead of `malloc'. This is because using malloc in
324 re_search* or re_match* could cause memory leaks when C-g is used in
325 Emacs; also, malloc is slower and causes storage fragmentation. On
326 the other hand, malloc is more portable, and easier to debug.
327
328 Because we sometimes use alloca, some routines have to be macros,
329 not functions -- `alloca'-allocated space disappears at the end of the
330 function it is called in. */
331
332 #ifdef REGEX_MALLOC
333
334 # define REGEX_ALLOCATE malloc
335 # define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
336 # define REGEX_FREE free
337
338 #else /* not REGEX_MALLOC */
339
340 /* Emacs already defines alloca, sometimes. */
341 # ifndef alloca
342
343 /* Make alloca work the best possible way. */
344 # ifdef __GNUC__
345 # define alloca __builtin_alloca
346 # else /* not __GNUC__ */
347 # if HAVE_ALLOCA_H
348 # include <alloca.h>
349 # endif /* HAVE_ALLOCA_H */
350 # endif /* not __GNUC__ */
351
352 # endif /* not alloca */
353
354 # define REGEX_ALLOCATE alloca
355
356 /* Assumes a `char *destination' variable. */
357 # define REGEX_REALLOCATE(source, osize, nsize) \
358 (destination = (char *) alloca (nsize), \
359 memcpy (destination, source, osize))
360
361 /* No need to do anything to free, after alloca. */
362 # define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
363
364 #endif /* not REGEX_MALLOC */
365
366 /* Define how to allocate the failure stack. */
367
368 #if defined REL_ALLOC && defined REGEX_MALLOC
369
370 # define REGEX_ALLOCATE_STACK(size) \
371 r_alloc (&failure_stack_ptr, (size))
372 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
373 r_re_alloc (&failure_stack_ptr, (nsize))
374 # define REGEX_FREE_STACK(ptr) \
375 r_alloc_free (&failure_stack_ptr)
376
377 #else /* not using relocating allocator */
378
379 # ifdef REGEX_MALLOC
380
381 # define REGEX_ALLOCATE_STACK malloc
382 # define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
383 # define REGEX_FREE_STACK free
384
385 # else /* not REGEX_MALLOC */
386
387 # define REGEX_ALLOCATE_STACK alloca
388
389 # define REGEX_REALLOCATE_STACK(source, osize, nsize) \
390 REGEX_REALLOCATE (source, osize, nsize)
391 /* No need to explicitly free anything. */
392 # define REGEX_FREE_STACK(arg)
393
394 # endif /* not REGEX_MALLOC */
395 #endif /* not using relocating allocator */
396
397
398 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
399 `string1' or just past its end. This works if PTR is NULL, which is
400 a good thing. */
401 #define FIRST_STRING_P(ptr) \
402 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
403
404 /* (Re)Allocate N items of type T using malloc, or fail. */
405 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
406 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
407 #define RETALLOC_IF(addr, n, t) \
408 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
409 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
410
411 #define BYTEWIDTH 8 /* In bits. */
412
413 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
414
415 #undef MAX
416 #undef MIN
417 #define MAX(a, b) ((a) > (b) ? (a) : (b))
418 #define MIN(a, b) ((a) < (b) ? (a) : (b))
419
420 typedef char boolean;
421 #define false 0
422 #define true 1
423
424 static int re_match_2_internal PARAMS ((struct re_pattern_buffer *bufp,
425 const char *string1, int size1,
426 const char *string2, int size2,
427 int pos,
428 struct re_registers *regs,
429 int stop));
430
431 /* These are the command codes that appear in compiled regular
432 expressions. Some opcodes are followed by argument bytes. A
433 command code can specify any interpretation whatsoever for its
434 arguments. Zero bytes may appear in the compiled regular expression. */
435
436 typedef enum
437 {
438 no_op = 0,
439
440 /* Succeed right away--no more backtracking. */
441 succeed,
442
443 /* Followed by one byte giving n, then by n literal bytes. */
444 exactn,
445
446 #ifdef MBS_SUPPORT
447 /* Same as exactn, but contains binary data. */
448 exactn_bin,
449 #endif
450
451 /* Matches any (more or less) character. */
452 anychar,
453
454 /* Matches any one char belonging to specified set. First
455 following byte is number of bitmap bytes. Then come bytes
456 for a bitmap saying which chars are in. Bits in each byte
457 are ordered low-bit-first. A character is in the set if its
458 bit is 1. A character too large to have a bit in the map is
459 automatically not in the set. */
460 /* ifdef MBS_SUPPORT, following element is length of character
461 classes, length of collating symbols, length of equivalence
462 classes, length of character ranges, and length of characters.
463 Next, character class element, collating symbols elements,
464 equivalence class elements, range elements, and character
465 elements follow.
466 See regex_compile function. */
467 charset,
468
469 /* Same parameters as charset, but match any character that is
470 not one of those specified. */
471 charset_not,
472
473 /* Start remembering the text that is matched, for storing in a
474 register. Followed by one byte with the register number, in
475 the range 0 to one less than the pattern buffer's re_nsub
476 field. Then followed by one byte with the number of groups
477 inner to this one. (This last has to be part of the
478 start_memory only because we need it in the on_failure_jump
479 of re_match_2.) */
480 start_memory,
481
482 /* Stop remembering the text that is matched and store it in a
483 memory register. Followed by one byte with the register
484 number, in the range 0 to one less than `re_nsub' in the
485 pattern buffer, and one byte with the number of inner groups,
486 just like `start_memory'. (We need the number of inner
487 groups here because we don't have any easy way of finding the
488 corresponding start_memory when we're at a stop_memory.) */
489 stop_memory,
490
491 /* Match a duplicate of something remembered. Followed by one
492 byte containing the register number. */
493 duplicate,
494
495 /* Fail unless at beginning of line. */
496 begline,
497
498 /* Fail unless at end of line. */
499 endline,
500
501 /* Succeeds if at beginning of buffer (if emacs) or at beginning
502 of string to be matched (if not). */
503 begbuf,
504
505 /* Analogously, for end of buffer/string. */
506 endbuf,
507
508 /* Followed by two byte relative address to which to jump. */
509 jump,
510
511 /* Same as jump, but marks the end of an alternative. */
512 jump_past_alt,
513
514 /* Followed by two-byte relative address of place to resume at
515 in case of failure. */
516 /* ifdef MBS_SUPPORT, the size of address is 1. */
517 on_failure_jump,
518
519 /* Like on_failure_jump, but pushes a placeholder instead of the
520 current string position when executed. */
521 on_failure_keep_string_jump,
522
523 /* Throw away latest failure point and then jump to following
524 two-byte relative address. */
525 /* ifdef MBS_SUPPORT, the size of address is 1. */
526 pop_failure_jump,
527
528 /* Change to pop_failure_jump if know won't have to backtrack to
529 match; otherwise change to jump. This is used to jump
530 back to the beginning of a repeat. If what follows this jump
531 clearly won't match what the repeat does, such that we can be
532 sure that there is no use backtracking out of repetitions
533 already matched, then we change it to a pop_failure_jump.
534 Followed by two-byte address. */
535 /* ifdef MBS_SUPPORT, the size of address is 1. */
536 maybe_pop_jump,
537
538 /* Jump to following two-byte address, and push a dummy failure
539 point. This failure point will be thrown away if an attempt
540 is made to use it for a failure. A `+' construct makes this
541 before the first repeat. Also used as an intermediary kind
542 of jump when compiling an alternative. */
543 /* ifdef MBS_SUPPORT, the size of address is 1. */
544 dummy_failure_jump,
545
546 /* Push a dummy failure point and continue. Used at the end of
547 alternatives. */
548 push_dummy_failure,
549
550 /* Followed by two-byte relative address and two-byte number n.
551 After matching N times, jump to the address upon failure. */
552 /* ifdef MBS_SUPPORT, the size of address is 1. */
553 succeed_n,
554
555 /* Followed by two-byte relative address, and two-byte number n.
556 Jump to the address N times, then fail. */
557 /* ifdef MBS_SUPPORT, the size of address is 1. */
558 jump_n,
559
560 /* Set the following two-byte relative address to the
561 subsequent two-byte number. The address *includes* the two
562 bytes of number. */
563 /* ifdef MBS_SUPPORT, the size of address is 1. */
564 set_number_at,
565
566 wordchar, /* Matches any word-constituent character. */
567 notwordchar, /* Matches any char that is not a word-constituent. */
568
569 wordbeg, /* Succeeds if at word beginning. */
570 wordend, /* Succeeds if at word end. */
571
572 wordbound, /* Succeeds if at a word boundary. */
573 notwordbound /* Succeeds if not at a word boundary. */
574
575 #ifdef emacs
576 ,before_dot, /* Succeeds if before point. */
577 at_dot, /* Succeeds if at point. */
578 after_dot, /* Succeeds if after point. */
579
580 /* Matches any character whose syntax is specified. Followed by
581 a byte which contains a syntax code, e.g., Sword. */
582 syntaxspec,
583
584 /* Matches any character whose syntax is not that specified. */
585 notsyntaxspec
586 #endif /* emacs */
587 } re_opcode_t;
588
589 /* Common operations on the compiled pattern. */
590
591 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
592 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
593
594 #ifdef MBS_SUPPORT
595 # define STORE_NUMBER(destination, number) \
596 do { \
597 *(destination) = (US_CHAR_TYPE)(number); \
598 } while (0)
599 #else
600 # define STORE_NUMBER(destination, number) \
601 do { \
602 (destination)[0] = (number) & 0377; \
603 (destination)[1] = (number) >> 8; \
604 } while (0)
605 #endif /* MBS_SUPPORT */
606
607 /* Same as STORE_NUMBER, except increment DESTINATION to
608 the byte after where the number is stored. Therefore, DESTINATION
609 must be an lvalue. */
610 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
611
612 #define STORE_NUMBER_AND_INCR(destination, number) \
613 do { \
614 STORE_NUMBER (destination, number); \
615 (destination) += OFFSET_ADDRESS_SIZE; \
616 } while (0)
617
618 /* Put into DESTINATION a number stored in two contiguous bytes starting
619 at SOURCE. */
620 /* ifdef MBS_SUPPORT, we store NUMBER in 1 element. */
621
622 #ifdef MBS_SUPPORT
623 # define EXTRACT_NUMBER(destination, source) \
624 do { \
625 (destination) = *(source); \
626 } while (0)
627 #else
628 # define EXTRACT_NUMBER(destination, source) \
629 do { \
630 (destination) = *(source) & 0377; \
631 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
632 } while (0)
633 #endif
634
635 #ifdef DEBUG
636 static void extract_number _RE_ARGS ((int *dest, US_CHAR_TYPE *source));
637 static void
extract_number(dest,source)638 extract_number (dest, source)
639 int *dest;
640 US_CHAR_TYPE *source;
641 {
642 #ifdef MBS_SUPPORT
643 *dest = *source;
644 #else
645 int temp = SIGN_EXTEND_CHAR (*(source + 1));
646 *dest = *source & 0377;
647 *dest += temp << 8;
648 #endif
649 }
650
651 # ifndef EXTRACT_MACROS /* To debug the macros. */
652 # undef EXTRACT_NUMBER
653 # define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
654 # endif /* not EXTRACT_MACROS */
655
656 #endif /* DEBUG */
657
658 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
659 SOURCE must be an lvalue. */
660
661 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
662 do { \
663 EXTRACT_NUMBER (destination, source); \
664 (source) += OFFSET_ADDRESS_SIZE; \
665 } while (0)
666
667 #ifdef DEBUG
668 static void extract_number_and_incr _RE_ARGS ((int *destination,
669 US_CHAR_TYPE **source));
670 static void
extract_number_and_incr(destination,source)671 extract_number_and_incr (destination, source)
672 int *destination;
673 US_CHAR_TYPE **source;
674 {
675 extract_number (destination, *source);
676 *source += OFFSET_ADDRESS_SIZE;
677 }
678
679 # ifndef EXTRACT_MACROS
680 # undef EXTRACT_NUMBER_AND_INCR
681 # define EXTRACT_NUMBER_AND_INCR(dest, src) \
682 extract_number_and_incr (&dest, &src)
683 # endif /* not EXTRACT_MACROS */
684
685 #endif /* DEBUG */
686
687 /* If DEBUG is defined, Regex prints many voluminous messages about what
688 it is doing (if the variable `debug' is nonzero). If linked with the
689 main program in `iregex.c', you can enter patterns and strings
690 interactively. And if linked with the main program in `main.c' and
691 the other test files, you can run the already-written tests. */
692
693 #ifdef DEBUG
694
695 /* We use standard I/O for debugging. */
696 # include <stdio.h>
697
698 /* It is useful to test things that ``must'' be true when debugging. */
699 # include <assert.h>
700
701 static int debug;
702
703 # define DEBUG_STATEMENT(e) e
704 # define DEBUG_PRINT1(x) if (debug) printf (x)
705 # define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
706 # define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
707 # define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
708 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
709 if (debug) print_partial_compiled_pattern (s, e)
710 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
711 if (debug) print_double_string (w, s1, sz1, s2, sz2)
712
713
714 /* Print the fastmap in human-readable form. */
715
716 void
print_fastmap(fastmap)717 print_fastmap (fastmap)
718 char *fastmap;
719 {
720 unsigned was_a_range = 0;
721 unsigned i = 0;
722
723 while (i < (1 << BYTEWIDTH))
724 {
725 if (fastmap[i++])
726 {
727 was_a_range = 0;
728 putchar (i - 1);
729 while (i < (1 << BYTEWIDTH) && fastmap[i])
730 {
731 was_a_range = 1;
732 i++;
733 }
734 if (was_a_range)
735 {
736 printf ("-");
737 putchar (i - 1);
738 }
739 }
740 }
741 putchar ('\n');
742 }
743
744
745 /* Print a compiled pattern string in human-readable form, starting at
746 the START pointer into it and ending just before the pointer END. */
747
748 void
print_partial_compiled_pattern(start,end)749 print_partial_compiled_pattern (start, end)
750 US_CHAR_TYPE *start;
751 US_CHAR_TYPE *end;
752 {
753 int mcnt, mcnt2;
754 US_CHAR_TYPE *p1;
755 US_CHAR_TYPE *p = start;
756 US_CHAR_TYPE *pend = end;
757
758 if (start == NULL)
759 {
760 printf ("(null)\n");
761 return;
762 }
763
764 /* Loop over pattern commands. */
765 while (p < pend)
766 {
767 #ifdef _LIBC
768 printf ("%td:\t", p - start);
769 #else
770 printf ("%ld:\t", (long int) (p - start));
771 #endif
772
773 switch ((re_opcode_t) *p++)
774 {
775 case no_op:
776 printf ("/no_op");
777 break;
778
779 case exactn:
780 mcnt = *p++;
781 printf ("/exactn/%d", mcnt);
782 do
783 {
784 putchar ('/');
785 PUT_CHAR (*p++);
786 }
787 while (--mcnt);
788 break;
789
790 #ifdef MBS_SUPPORT
791 case exactn_bin:
792 mcnt = *p++;
793 printf ("/exactn_bin/%d", mcnt);
794 do
795 {
796 printf("/%lx", (long int) *p++);
797 }
798 while (--mcnt);
799 break;
800 #endif /* MBS_SUPPORT */
801
802 case start_memory:
803 mcnt = *p++;
804 printf ("/start_memory/%d/%ld", mcnt, (long int) *p++);
805 break;
806
807 case stop_memory:
808 mcnt = *p++;
809 printf ("/stop_memory/%d/%ld", mcnt, (long int) *p++);
810 break;
811
812 case duplicate:
813 printf ("/duplicate/%ld", (long int) *p++);
814 break;
815
816 case anychar:
817 printf ("/anychar");
818 break;
819
820 case charset:
821 case charset_not:
822 {
823 #ifdef MBS_SUPPORT
824 int i, length;
825 wchar_t *workp = p;
826 printf ("/charset [%s",
827 (re_opcode_t) *(workp - 1) == charset_not ? "^" : "");
828 p += 5;
829 length = *workp++; /* the length of char_classes */
830 for (i=0 ; i<length ; i++)
831 printf("[:%lx:]", (long int) *p++);
832 length = *workp++; /* the length of collating_symbol */
833 for (i=0 ; i<length ;)
834 {
835 printf("[.");
836 while(*p != 0)
837 PUT_CHAR((i++,*p++));
838 i++,p++;
839 printf(".]");
840 }
841 length = *workp++; /* the length of equivalence_class */
842 for (i=0 ; i<length ;)
843 {
844 printf("[=");
845 while(*p != 0)
846 PUT_CHAR((i++,*p++));
847 i++,p++;
848 printf("=]");
849 }
850 length = *workp++; /* the length of char_range */
851 for (i=0 ; i<length ; i++)
852 {
853 wchar_t range_start = *p++;
854 wchar_t range_end = *p++;
855 if (MB_CUR_MAX == 1)
856 printf("%c-%c", (char) range_start, (char) range_end);
857 else
858 printf("%C-%C", (wint_t) range_start, (wint_t) range_end);
859 }
860 length = *workp++; /* the length of char */
861 for (i=0 ; i<length ; i++)
862 if (MB_CUR_MAX == 1)
863 putchar (*p++);
864 else
865 printf("%C", (wint_t) *p++);
866 putchar (']');
867 #else
868 register int c, last = -100;
869 register int in_range = 0;
870
871 printf ("/charset [%s",
872 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
873
874 assert (p + *p < pend);
875
876 for (c = 0; c < 256; c++)
877 if (c / 8 < *p
878 && (p[1 + (c/8)] & (1 << (c % 8))))
879 {
880 /* Are we starting a range? */
881 if (last + 1 == c && ! in_range)
882 {
883 putchar ('-');
884 in_range = 1;
885 }
886 /* Have we broken a range? */
887 else if (last + 1 != c && in_range)
888 {
889 putchar (last);
890 in_range = 0;
891 }
892
893 if (! in_range)
894 putchar (c);
895
896 last = c;
897 }
898
899 if (in_range)
900 putchar (last);
901
902 putchar (']');
903
904 p += 1 + *p;
905 #endif /* MBS_SUPPORT */
906 }
907 break;
908
909 case begline:
910 printf ("/begline");
911 break;
912
913 case endline:
914 printf ("/endline");
915 break;
916
917 case on_failure_jump:
918 extract_number_and_incr (&mcnt, &p);
919 #ifdef _LIBC
920 printf ("/on_failure_jump to %td", p + mcnt - start);
921 #else
922 printf ("/on_failure_jump to %ld", (long int) (p + mcnt - start));
923 #endif
924 break;
925
926 case on_failure_keep_string_jump:
927 extract_number_and_incr (&mcnt, &p);
928 #ifdef _LIBC
929 printf ("/on_failure_keep_string_jump to %td", p + mcnt - start);
930 #else
931 printf ("/on_failure_keep_string_jump to %ld",
932 (long int) (p + mcnt - start));
933 #endif
934 break;
935
936 case dummy_failure_jump:
937 extract_number_and_incr (&mcnt, &p);
938 #ifdef _LIBC
939 printf ("/dummy_failure_jump to %td", p + mcnt - start);
940 #else
941 printf ("/dummy_failure_jump to %ld", (long int) (p + mcnt - start));
942 #endif
943 break;
944
945 case push_dummy_failure:
946 printf ("/push_dummy_failure");
947 break;
948
949 case maybe_pop_jump:
950 extract_number_and_incr (&mcnt, &p);
951 #ifdef _LIBC
952 printf ("/maybe_pop_jump to %td", p + mcnt - start);
953 #else
954 printf ("/maybe_pop_jump to %ld", (long int) (p + mcnt - start));
955 #endif
956 break;
957
958 case pop_failure_jump:
959 extract_number_and_incr (&mcnt, &p);
960 #ifdef _LIBC
961 printf ("/pop_failure_jump to %td", p + mcnt - start);
962 #else
963 printf ("/pop_failure_jump to %ld", (long int) (p + mcnt - start));
964 #endif
965 break;
966
967 case jump_past_alt:
968 extract_number_and_incr (&mcnt, &p);
969 #ifdef _LIBC
970 printf ("/jump_past_alt to %td", p + mcnt - start);
971 #else
972 printf ("/jump_past_alt to %ld", (long int) (p + mcnt - start));
973 #endif
974 break;
975
976 case jump:
977 extract_number_and_incr (&mcnt, &p);
978 #ifdef _LIBC
979 printf ("/jump to %td", p + mcnt - start);
980 #else
981 printf ("/jump to %ld", (long int) (p + mcnt - start));
982 #endif
983 break;
984
985 case succeed_n:
986 extract_number_and_incr (&mcnt, &p);
987 p1 = p + mcnt;
988 extract_number_and_incr (&mcnt2, &p);
989 #ifdef _LIBC
990 printf ("/succeed_n to %td, %d times", p1 - start, mcnt2);
991 #else
992 printf ("/succeed_n to %ld, %d times",
993 (long int) (p1 - start), mcnt2);
994 #endif
995 break;
996
997 case jump_n:
998 extract_number_and_incr (&mcnt, &p);
999 p1 = p + mcnt;
1000 extract_number_and_incr (&mcnt2, &p);
1001 printf ("/jump_n to %d, %d times", p1 - start, mcnt2);
1002 break;
1003
1004 case set_number_at:
1005 extract_number_and_incr (&mcnt, &p);
1006 p1 = p + mcnt;
1007 extract_number_and_incr (&mcnt2, &p);
1008 #ifdef _LIBC
1009 printf ("/set_number_at location %td to %d", p1 - start, mcnt2);
1010 #else
1011 printf ("/set_number_at location %ld to %d",
1012 (long int) (p1 - start), mcnt2);
1013 #endif
1014 break;
1015
1016 case wordbound:
1017 printf ("/wordbound");
1018 break;
1019
1020 case notwordbound:
1021 printf ("/notwordbound");
1022 break;
1023
1024 case wordbeg:
1025 printf ("/wordbeg");
1026 break;
1027
1028 case wordend:
1029 printf ("/wordend");
1030 break;
1031
1032 # ifdef emacs
1033 case before_dot:
1034 printf ("/before_dot");
1035 break;
1036
1037 case at_dot:
1038 printf ("/at_dot");
1039 break;
1040
1041 case after_dot:
1042 printf ("/after_dot");
1043 break;
1044
1045 case syntaxspec:
1046 printf ("/syntaxspec");
1047 mcnt = *p++;
1048 printf ("/%d", mcnt);
1049 break;
1050
1051 case notsyntaxspec:
1052 printf ("/notsyntaxspec");
1053 mcnt = *p++;
1054 printf ("/%d", mcnt);
1055 break;
1056 # endif /* emacs */
1057
1058 case wordchar:
1059 printf ("/wordchar");
1060 break;
1061
1062 case notwordchar:
1063 printf ("/notwordchar");
1064 break;
1065
1066 case begbuf:
1067 printf ("/begbuf");
1068 break;
1069
1070 case endbuf:
1071 printf ("/endbuf");
1072 break;
1073
1074 default:
1075 printf ("?%ld", (long int) *(p-1));
1076 }
1077
1078 putchar ('\n');
1079 }
1080
1081 #ifdef _LIBC
1082 printf ("%td:\tend of pattern.\n", p - start);
1083 #else
1084 printf ("%ld:\tend of pattern.\n", (long int) (p - start));
1085 #endif
1086 }
1087
1088
1089 void
print_compiled_pattern(bufp)1090 print_compiled_pattern (bufp)
1091 struct re_pattern_buffer *bufp;
1092 {
1093 US_CHAR_TYPE *buffer = (US_CHAR_TYPE*) bufp->buffer;
1094
1095 print_partial_compiled_pattern (buffer, buffer
1096 + bufp->used / sizeof(US_CHAR_TYPE));
1097 printf ("%ld bytes used/%ld bytes allocated.\n",
1098 bufp->used, bufp->allocated);
1099
1100 if (bufp->fastmap_accurate && bufp->fastmap)
1101 {
1102 printf ("fastmap: ");
1103 print_fastmap (bufp->fastmap);
1104 }
1105
1106 #ifdef _LIBC
1107 printf ("re_nsub: %Zd\t", bufp->re_nsub);
1108 #else
1109 printf ("re_nsub: %ld\t", (long int) bufp->re_nsub);
1110 #endif
1111 printf ("regs_alloc: %d\t", bufp->regs_allocated);
1112 printf ("can_be_null: %d\t", bufp->can_be_null);
1113 printf ("newline_anchor: %d\n", bufp->newline_anchor);
1114 printf ("no_sub: %d\t", bufp->no_sub);
1115 printf ("not_bol: %d\t", bufp->not_bol);
1116 printf ("not_eol: %d\t", bufp->not_eol);
1117 printf ("syntax: %lx\n", bufp->syntax);
1118 /* Perhaps we should print the translate table? */
1119 }
1120
1121
1122 void
print_double_string(where,string1,size1,string2,size2)1123 print_double_string (where, string1, size1, string2, size2)
1124 const CHAR_TYPE *where;
1125 const CHAR_TYPE *string1;
1126 const CHAR_TYPE *string2;
1127 int size1;
1128 int size2;
1129 {
1130 ptrdiff_t this_char;
1131
1132 if (where == NULL)
1133 printf ("(null)");
1134 else
1135 {
1136 if (FIRST_STRING_P (where))
1137 {
1138 for (this_char = where - string1; this_char < size1; this_char++)
1139 PUT_CHAR (string1[this_char]);
1140
1141 where = string2;
1142 }
1143
1144 for (this_char = where - string2; this_char < size2; this_char++)
1145 PUT_CHAR (string2[this_char]);
1146 }
1147 }
1148
1149 void
printchar(c)1150 printchar (c)
1151 int c;
1152 {
1153 putc (c, stderr);
1154 }
1155
1156 #else /* not DEBUG */
1157
1158 # undef assert
1159 # define assert(e)
1160
1161 # define DEBUG_STATEMENT(e)
1162 # define DEBUG_PRINT1(x)
1163 # define DEBUG_PRINT2(x1, x2)
1164 # define DEBUG_PRINT3(x1, x2, x3)
1165 # define DEBUG_PRINT4(x1, x2, x3, x4)
1166 # define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
1167 # define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
1168
1169 #endif /* not DEBUG */
1170
1171 #ifdef MBS_SUPPORT
1172 /* This convert a multibyte string to a wide character string.
1173 And write their correspondances to offset_buffer(see below)
1174 and write whether each wchar_t is binary data to is_binary.
1175 This assume invalid multibyte sequences as binary data.
1176 We assume offset_buffer and is_binary is already allocated
1177 enough space. */
1178
1179 static size_t convert_mbs_to_wcs (CHAR_TYPE *dest, const unsigned char* src,
1180 size_t len, int *offset_buffer,
1181 char *is_binary);
1182 static size_t
convert_mbs_to_wcs(dest,src,len,offset_buffer,is_binary)1183 convert_mbs_to_wcs (dest, src, len, offset_buffer, is_binary)
1184 CHAR_TYPE *dest;
1185 const unsigned char* src;
1186 size_t len; /* the length of multibyte string. */
1187
1188 /* It hold correspondances between src(char string) and
1189 dest(wchar_t string) for optimization.
1190 e.g. src = "xxxyzz"
1191 dest = {'X', 'Y', 'Z'}
1192 (each "xxx", "y" and "zz" represent one multibyte character
1193 corresponding to 'X', 'Y' and 'Z'.)
1194 offset_buffer = {0, 0+3("xxx"), 0+3+1("y"), 0+3+1+2("zz")}
1195 = {0, 3, 4, 6}
1196 */
1197 int *offset_buffer;
1198 char *is_binary;
1199 {
1200 wchar_t *pdest = dest;
1201 const unsigned char *psrc = src;
1202 size_t wc_count = 0;
1203
1204 if (MB_CUR_MAX == 1)
1205 { /* We don't need conversion. */
1206 for ( ; wc_count < len ; ++wc_count)
1207 {
1208 *pdest++ = *psrc++;
1209 is_binary[wc_count] = FALSE;
1210 offset_buffer[wc_count] = wc_count;
1211 }
1212 offset_buffer[wc_count] = wc_count;
1213 }
1214 else
1215 {
1216 /* We need conversion. */
1217 mbstate_t mbs;
1218 int consumed;
1219 size_t mb_remain = len;
1220 size_t mb_count = 0;
1221
1222 /* Initialize the conversion state. */
1223 memset (&mbs, 0, sizeof (mbstate_t));
1224
1225 offset_buffer[0] = 0;
1226 for( ; mb_remain > 0 ; ++wc_count, ++pdest, mb_remain -= consumed,
1227 psrc += consumed)
1228 {
1229 consumed = mbrtowc (pdest, psrc, mb_remain, &mbs);
1230
1231 if (consumed <= 0)
1232 /* failed to convert. maybe src contains binary data.
1233 So we consume 1 byte manualy. */
1234 {
1235 *pdest = *psrc;
1236 consumed = 1;
1237 is_binary[wc_count] = TRUE;
1238 }
1239 else
1240 is_binary[wc_count] = FALSE;
1241 /* In sjis encoding, we use yen sign as escape character in
1242 place of reverse solidus. So we convert 0x5c(yen sign in
1243 sjis) to not 0xa5(yen sign in UCS2) but 0x5c(reverse
1244 solidus in UCS2). */
1245 if (consumed == 1 && (int) *psrc == 0x5c && (int) *pdest == 0xa5)
1246 *pdest = (wchar_t) *psrc;
1247
1248 offset_buffer[wc_count + 1] = mb_count += consumed;
1249 }
1250 }
1251
1252 return wc_count;
1253 }
1254
1255 #endif /* MBS_SUPPORT */
1256
1257 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
1258 also be assigned to arbitrarily: each pattern buffer stores its own
1259 syntax, so it can be changed between regex compilations. */
1260 /* This has no initializer because initialized variables in Emacs
1261 become read-only after dumping. */
1262 reg_syntax_t re_syntax_options;
1263
1264
1265 /* Specify the precise syntax of regexps for compilation. This provides
1266 for compatibility for various utilities which historically have
1267 different, incompatible syntaxes.
1268
1269 The argument SYNTAX is a bit mask comprised of the various bits
1270 defined in regex.h. We return the old syntax. */
1271
1272 reg_syntax_t
re_set_syntax(syntax)1273 re_set_syntax (syntax)
1274 reg_syntax_t syntax;
1275 {
1276 reg_syntax_t ret = re_syntax_options;
1277
1278 re_syntax_options = syntax;
1279 #ifdef DEBUG
1280 if (syntax & RE_DEBUG)
1281 debug = 1;
1282 else if (debug) /* was on but now is not */
1283 debug = 0;
1284 #endif /* DEBUG */
1285 return ret;
1286 }
1287 #ifdef _LIBC
1288 weak_alias (__re_set_syntax, re_set_syntax)
1289 #endif
1290
1291 /* This table gives an error message for each of the error codes listed
1292 in regex.h. Obviously the order here has to be same as there.
1293 POSIX doesn't require that we do anything for REG_NOERROR,
1294 but why not be nice? */
1295
1296 static const char re_error_msgid[] =
1297 {
1298 #define REG_NOERROR_IDX 0
1299 gettext_noop ("Success") /* REG_NOERROR */
1300 "\0"
1301 #define REG_NOMATCH_IDX (REG_NOERROR_IDX + sizeof "Success")
1302 gettext_noop ("No match") /* REG_NOMATCH */
1303 "\0"
1304 #define REG_BADPAT_IDX (REG_NOMATCH_IDX + sizeof "No match")
1305 gettext_noop ("Invalid regular expression") /* REG_BADPAT */
1306 "\0"
1307 #define REG_ECOLLATE_IDX (REG_BADPAT_IDX + sizeof "Invalid regular expression")
1308 gettext_noop ("Invalid collation character") /* REG_ECOLLATE */
1309 "\0"
1310 #define REG_ECTYPE_IDX (REG_ECOLLATE_IDX + sizeof "Invalid collation character")
1311 gettext_noop ("Invalid character class name") /* REG_ECTYPE */
1312 "\0"
1313 #define REG_EESCAPE_IDX (REG_ECTYPE_IDX + sizeof "Invalid character class name")
1314 gettext_noop ("Trailing backslash") /* REG_EESCAPE */
1315 "\0"
1316 #define REG_ESUBREG_IDX (REG_EESCAPE_IDX + sizeof "Trailing backslash")
1317 gettext_noop ("Invalid back reference") /* REG_ESUBREG */
1318 "\0"
1319 #define REG_EBRACK_IDX (REG_ESUBREG_IDX + sizeof "Invalid back reference")
1320 gettext_noop ("Unmatched [ or [^") /* REG_EBRACK */
1321 "\0"
1322 #define REG_EPAREN_IDX (REG_EBRACK_IDX + sizeof "Unmatched [ or [^")
1323 gettext_noop ("Unmatched ( or \\(") /* REG_EPAREN */
1324 "\0"
1325 #define REG_EBRACE_IDX (REG_EPAREN_IDX + sizeof "Unmatched ( or \\(")
1326 gettext_noop ("Unmatched \\{") /* REG_EBRACE */
1327 "\0"
1328 #define REG_BADBR_IDX (REG_EBRACE_IDX + sizeof "Unmatched \\{")
1329 gettext_noop ("Invalid content of \\{\\}") /* REG_BADBR */
1330 "\0"
1331 #define REG_ERANGE_IDX (REG_BADBR_IDX + sizeof "Invalid content of \\{\\}")
1332 gettext_noop ("Invalid range end") /* REG_ERANGE */
1333 "\0"
1334 #define REG_ESPACE_IDX (REG_ERANGE_IDX + sizeof "Invalid range end")
1335 gettext_noop ("Memory exhausted") /* REG_ESPACE */
1336 "\0"
1337 #define REG_BADRPT_IDX (REG_ESPACE_IDX + sizeof "Memory exhausted")
1338 gettext_noop ("Invalid preceding regular expression") /* REG_BADRPT */
1339 "\0"
1340 #define REG_EEND_IDX (REG_BADRPT_IDX + sizeof "Invalid preceding regular expression")
1341 gettext_noop ("Premature end of regular expression") /* REG_EEND */
1342 "\0"
1343 #define REG_ESIZE_IDX (REG_EEND_IDX + sizeof "Premature end of regular expression")
1344 gettext_noop ("Regular expression too big") /* REG_ESIZE */
1345 "\0"
1346 #define REG_ERPAREN_IDX (REG_ESIZE_IDX + sizeof "Regular expression too big")
1347 gettext_noop ("Unmatched ) or \\)") /* REG_ERPAREN */
1348 };
1349
1350 static const size_t re_error_msgid_idx[] =
1351 {
1352 REG_NOERROR_IDX,
1353 REG_NOMATCH_IDX,
1354 REG_BADPAT_IDX,
1355 REG_ECOLLATE_IDX,
1356 REG_ECTYPE_IDX,
1357 REG_EESCAPE_IDX,
1358 REG_ESUBREG_IDX,
1359 REG_EBRACK_IDX,
1360 REG_EPAREN_IDX,
1361 REG_EBRACE_IDX,
1362 REG_BADBR_IDX,
1363 REG_ERANGE_IDX,
1364 REG_ESPACE_IDX,
1365 REG_BADRPT_IDX,
1366 REG_EEND_IDX,
1367 REG_ESIZE_IDX,
1368 REG_ERPAREN_IDX
1369 };
1370
1371 /* Avoiding alloca during matching, to placate r_alloc. */
1372
1373 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
1374 searching and matching functions should not call alloca. On some
1375 systems, alloca is implemented in terms of malloc, and if we're
1376 using the relocating allocator routines, then malloc could cause a
1377 relocation, which might (if the strings being searched are in the
1378 ralloc heap) shift the data out from underneath the regexp
1379 routines.
1380
1381 Here's another reason to avoid allocation: Emacs
1382 processes input from X in a signal handler; processing X input may
1383 call malloc; if input arrives while a matching routine is calling
1384 malloc, then we're scrod. But Emacs can't just block input while
1385 calling matching routines; then we don't notice interrupts when
1386 they come in. So, Emacs blocks input around all regexp calls
1387 except the matching calls, which it leaves unprotected, in the
1388 faith that they will not malloc. */
1389
1390 /* Normally, this is fine. */
1391 #define MATCH_MAY_ALLOCATE
1392
1393 /* When using GNU C, we are not REALLY using the C alloca, no matter
1394 what config.h may say. So don't take precautions for it. */
1395 #ifdef __GNUC__
1396 # undef C_ALLOCA
1397 #endif
1398
1399 /* The match routines may not allocate if (1) they would do it with malloc
1400 and (2) it's not safe for them to use malloc.
1401 Note that if REL_ALLOC is defined, matching would not use malloc for the
1402 failure stack, but we would still use it for the register vectors;
1403 so REL_ALLOC should not affect this. */
1404 #if (defined C_ALLOCA || defined REGEX_MALLOC) && defined emacs
1405 # undef MATCH_MAY_ALLOCATE
1406 #endif
1407
1408
1409 /* Failure stack declarations and macros; both re_compile_fastmap and
1410 re_match_2 use a failure stack. These have to be macros because of
1411 REGEX_ALLOCATE_STACK. */
1412
1413
1414 /* Number of failure points for which to initially allocate space
1415 when matching. If this number is exceeded, we allocate more
1416 space, so it is not a hard limit. */
1417 #ifndef INIT_FAILURE_ALLOC
1418 # define INIT_FAILURE_ALLOC 5
1419 #endif
1420
1421 /* Roughly the maximum number of failure points on the stack. Would be
1422 exactly that if always used MAX_FAILURE_ITEMS items each time we failed.
1423 This is a variable only so users of regex can assign to it; we never
1424 change it ourselves. */
1425
1426 #ifdef INT_IS_16BIT
1427
1428 # if defined MATCH_MAY_ALLOCATE
1429 /* 4400 was enough to cause a crash on Alpha OSF/1,
1430 whose default stack limit is 2mb. */
1431 long int re_max_failures = 4000;
1432 # else
1433 long int re_max_failures = 2000;
1434 # endif
1435
1436 union fail_stack_elt
1437 {
1438 US_CHAR_TYPE *pointer;
1439 long int integer;
1440 };
1441
1442 typedef union fail_stack_elt fail_stack_elt_t;
1443
1444 typedef struct
1445 {
1446 fail_stack_elt_t *stack;
1447 unsigned long int size;
1448 unsigned long int avail; /* Offset of next open position. */
1449 } fail_stack_type;
1450
1451 #else /* not INT_IS_16BIT */
1452
1453 # if defined MATCH_MAY_ALLOCATE
1454 /* 4400 was enough to cause a crash on Alpha OSF/1,
1455 whose default stack limit is 2mb. */
1456 int re_max_failures = 4000;
1457 # else
1458 int re_max_failures = 2000;
1459 # endif
1460
1461 union fail_stack_elt
1462 {
1463 US_CHAR_TYPE *pointer;
1464 int integer;
1465 };
1466
1467 typedef union fail_stack_elt fail_stack_elt_t;
1468
1469 typedef struct
1470 {
1471 fail_stack_elt_t *stack;
1472 unsigned size;
1473 unsigned avail; /* Offset of next open position. */
1474 } fail_stack_type;
1475
1476 #endif /* INT_IS_16BIT */
1477
1478 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1479 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1480 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1481
1482
1483 /* Define macros to initialize and free the failure stack.
1484 Do `return -2' if the alloc fails. */
1485
1486 #ifdef MATCH_MAY_ALLOCATE
1487 # define INIT_FAIL_STACK() \
1488 do { \
1489 fail_stack.stack = (fail_stack_elt_t *) \
1490 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1491 \
1492 if (fail_stack.stack == NULL) \
1493 return -2; \
1494 \
1495 fail_stack.size = INIT_FAILURE_ALLOC; \
1496 fail_stack.avail = 0; \
1497 } while (0)
1498
1499 # define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1500 #else
1501 # define INIT_FAIL_STACK() \
1502 do { \
1503 fail_stack.avail = 0; \
1504 } while (0)
1505
1506 # define RESET_FAIL_STACK()
1507 #endif
1508
1509
1510 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1511
1512 Return 1 if succeeds, and 0 if either ran out of memory
1513 allocating space for it or it was already too large.
1514
1515 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1516
1517 #define DOUBLE_FAIL_STACK(fail_stack) \
1518 ((fail_stack).size > (unsigned) (re_max_failures * MAX_FAILURE_ITEMS) \
1519 ? 0 \
1520 : ((fail_stack).stack = (fail_stack_elt_t *) \
1521 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1522 (fail_stack).size * sizeof (fail_stack_elt_t), \
1523 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1524 \
1525 (fail_stack).stack == NULL \
1526 ? 0 \
1527 : ((fail_stack).size <<= 1, \
1528 1)))
1529
1530
1531 /* Push pointer POINTER on FAIL_STACK.
1532 Return 1 if was able to do so and 0 if ran out of memory allocating
1533 space to do so. */
1534 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1535 ((FAIL_STACK_FULL () \
1536 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1537 ? 0 \
1538 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1539 1))
1540
1541 /* Push a pointer value onto the failure stack.
1542 Assumes the variable `fail_stack'. Probably should only
1543 be called from within `PUSH_FAILURE_POINT'. */
1544 #define PUSH_FAILURE_POINTER(item) \
1545 fail_stack.stack[fail_stack.avail++].pointer = (US_CHAR_TYPE *) (item)
1546
1547 /* This pushes an integer-valued item onto the failure stack.
1548 Assumes the variable `fail_stack'. Probably should only
1549 be called from within `PUSH_FAILURE_POINT'. */
1550 #define PUSH_FAILURE_INT(item) \
1551 fail_stack.stack[fail_stack.avail++].integer = (item)
1552
1553 /* Push a fail_stack_elt_t value onto the failure stack.
1554 Assumes the variable `fail_stack'. Probably should only
1555 be called from within `PUSH_FAILURE_POINT'. */
1556 #define PUSH_FAILURE_ELT(item) \
1557 fail_stack.stack[fail_stack.avail++] = (item)
1558
1559 /* These three POP... operations complement the three PUSH... operations.
1560 All assume that `fail_stack' is nonempty. */
1561 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1562 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1563 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1564
1565 /* Used to omit pushing failure point id's when we're not debugging. */
1566 #ifdef DEBUG
1567 # define DEBUG_PUSH PUSH_FAILURE_INT
1568 # define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1569 #else
1570 # define DEBUG_PUSH(item)
1571 # define DEBUG_POP(item_addr)
1572 #endif
1573
1574
1575 /* Push the information about the state we will need
1576 if we ever fail back to it.
1577
1578 Requires variables fail_stack, regstart, regend, reg_info, and
1579 num_regs_pushed be declared. DOUBLE_FAIL_STACK requires `destination'
1580 be declared.
1581
1582 Does `return FAILURE_CODE' if runs out of memory. */
1583
1584 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1585 do { \
1586 char *destination; \
1587 /* Must be int, so when we don't save any registers, the arithmetic \
1588 of 0 + -1 isn't done as unsigned. */ \
1589 /* Can't be int, since there is not a shred of a guarantee that int \
1590 is wide enough to hold a value of something to which pointer can \
1591 be assigned */ \
1592 active_reg_t this_reg; \
1593 \
1594 DEBUG_STATEMENT (failure_id++); \
1595 DEBUG_STATEMENT (nfailure_points_pushed++); \
1596 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1597 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1598 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1599 \
1600 DEBUG_PRINT2 (" slots needed: %ld\n", NUM_FAILURE_ITEMS); \
1601 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1602 \
1603 /* Ensure we have enough space allocated for what we will push. */ \
1604 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1605 { \
1606 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1607 return failure_code; \
1608 \
1609 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1610 (fail_stack).size); \
1611 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1612 } \
1613 \
1614 /* Push the info, starting with the registers. */ \
1615 DEBUG_PRINT1 ("\n"); \
1616 \
1617 if (1) \
1618 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1619 this_reg++) \
1620 { \
1621 DEBUG_PRINT2 (" Pushing reg: %lu\n", this_reg); \
1622 DEBUG_STATEMENT (num_regs_pushed++); \
1623 \
1624 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1625 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1626 \
1627 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1628 PUSH_FAILURE_POINTER (regend[this_reg]); \
1629 \
1630 DEBUG_PRINT2 (" info: %p\n ", \
1631 reg_info[this_reg].word.pointer); \
1632 DEBUG_PRINT2 (" match_null=%d", \
1633 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1634 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1635 DEBUG_PRINT2 (" matched_something=%d", \
1636 MATCHED_SOMETHING (reg_info[this_reg])); \
1637 DEBUG_PRINT2 (" ever_matched=%d", \
1638 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1639 DEBUG_PRINT1 ("\n"); \
1640 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1641 } \
1642 \
1643 DEBUG_PRINT2 (" Pushing low active reg: %ld\n", lowest_active_reg);\
1644 PUSH_FAILURE_INT (lowest_active_reg); \
1645 \
1646 DEBUG_PRINT2 (" Pushing high active reg: %ld\n", highest_active_reg);\
1647 PUSH_FAILURE_INT (highest_active_reg); \
1648 \
1649 DEBUG_PRINT2 (" Pushing pattern %p:\n", pattern_place); \
1650 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1651 PUSH_FAILURE_POINTER (pattern_place); \
1652 \
1653 DEBUG_PRINT2 (" Pushing string %p: `", string_place); \
1654 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1655 size2); \
1656 DEBUG_PRINT1 ("'\n"); \
1657 PUSH_FAILURE_POINTER (string_place); \
1658 \
1659 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1660 DEBUG_PUSH (failure_id); \
1661 } while (0)
1662
1663 /* This is the number of items that are pushed and popped on the stack
1664 for each register. */
1665 #define NUM_REG_ITEMS 3
1666
1667 /* Individual items aside from the registers. */
1668 #ifdef DEBUG
1669 # define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1670 #else
1671 # define NUM_NONREG_ITEMS 4
1672 #endif
1673
1674 /* We push at most this many items on the stack. */
1675 /* We used to use (num_regs - 1), which is the number of registers
1676 this regexp will save; but that was changed to 5
1677 to avoid stack overflow for a regexp with lots of parens. */
1678 #define MAX_FAILURE_ITEMS (5 * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1679
1680 /* We actually push this many items. */
1681 #define NUM_FAILURE_ITEMS \
1682 (((0 \
1683 ? 0 : highest_active_reg - lowest_active_reg + 1) \
1684 * NUM_REG_ITEMS) \
1685 + NUM_NONREG_ITEMS)
1686
1687 /* How many items can still be added to the stack without overflowing it. */
1688 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1689
1690
1691 /* Pops what PUSH_FAIL_STACK pushes.
1692
1693 We restore into the parameters, all of which should be lvalues:
1694 STR -- the saved data position.
1695 PAT -- the saved pattern position.
1696 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1697 REGSTART, REGEND -- arrays of string positions.
1698 REG_INFO -- array of information about each subexpression.
1699
1700 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1701 `pend', `string1', `size1', `string2', and `size2'. */
1702 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1703 { \
1704 DEBUG_STATEMENT (unsigned failure_id;) \
1705 active_reg_t this_reg; \
1706 const US_CHAR_TYPE *string_temp; \
1707 \
1708 assert (!FAIL_STACK_EMPTY ()); \
1709 \
1710 /* Remove failure points and point to how many regs pushed. */ \
1711 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1712 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1713 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1714 \
1715 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1716 \
1717 DEBUG_POP (&failure_id); \
1718 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1719 \
1720 /* If the saved string location is NULL, it came from an \
1721 on_failure_keep_string_jump opcode, and we want to throw away the \
1722 saved NULL, thus retaining our current position in the string. */ \
1723 string_temp = POP_FAILURE_POINTER (); \
1724 if (string_temp != NULL) \
1725 str = (const CHAR_TYPE *) string_temp; \
1726 \
1727 DEBUG_PRINT2 (" Popping string %p: `", str); \
1728 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1729 DEBUG_PRINT1 ("'\n"); \
1730 \
1731 pat = (US_CHAR_TYPE *) POP_FAILURE_POINTER (); \
1732 DEBUG_PRINT2 (" Popping pattern %p:\n", pat); \
1733 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1734 \
1735 /* Restore register info. */ \
1736 high_reg = (active_reg_t) POP_FAILURE_INT (); \
1737 DEBUG_PRINT2 (" Popping high active reg: %ld\n", high_reg); \
1738 \
1739 low_reg = (active_reg_t) POP_FAILURE_INT (); \
1740 DEBUG_PRINT2 (" Popping low active reg: %ld\n", low_reg); \
1741 \
1742 if (1) \
1743 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1744 { \
1745 DEBUG_PRINT2 (" Popping reg: %ld\n", this_reg); \
1746 \
1747 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1748 DEBUG_PRINT2 (" info: %p\n", \
1749 reg_info[this_reg].word.pointer); \
1750 \
1751 regend[this_reg] = (const CHAR_TYPE *) POP_FAILURE_POINTER (); \
1752 DEBUG_PRINT2 (" end: %p\n", regend[this_reg]); \
1753 \
1754 regstart[this_reg] = (const CHAR_TYPE *) POP_FAILURE_POINTER ();\
1755 DEBUG_PRINT2 (" start: %p\n", regstart[this_reg]); \
1756 } \
1757 else \
1758 { \
1759 for (this_reg = highest_active_reg; this_reg > high_reg; this_reg--) \
1760 { \
1761 reg_info[this_reg].word.integer = 0; \
1762 regend[this_reg] = 0; \
1763 regstart[this_reg] = 0; \
1764 } \
1765 highest_active_reg = high_reg; \
1766 } \
1767 \
1768 set_regs_matched_done = 0; \
1769 DEBUG_STATEMENT (nfailure_points_popped++); \
1770 } /* POP_FAILURE_POINT */
1771
1772
1773 /* Structure for per-register (a.k.a. per-group) information.
1774 Other register information, such as the
1775 starting and ending positions (which are addresses), and the list of
1776 inner groups (which is a bits list) are maintained in separate
1777 variables.
1778
1779 We are making a (strictly speaking) nonportable assumption here: that
1780 the compiler will pack our bit fields into something that fits into
1781 the type of `word', i.e., is something that fits into one item on the
1782 failure stack. */
1783
1784
1785 /* Declarations and macros for re_match_2. */
1786
1787 typedef union
1788 {
1789 fail_stack_elt_t word;
1790 struct
1791 {
1792 /* This field is one if this group can match the empty string,
1793 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1794 #define MATCH_NULL_UNSET_VALUE 3
1795 unsigned match_null_string_p : 2;
1796 unsigned is_active : 1;
1797 unsigned matched_something : 1;
1798 unsigned ever_matched_something : 1;
1799 } bits;
1800 } register_info_type;
1801
1802 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1803 #define IS_ACTIVE(R) ((R).bits.is_active)
1804 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1805 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1806
1807
1808 /* Call this when have matched a real character; it sets `matched' flags
1809 for the subexpressions which we are currently inside. Also records
1810 that those subexprs have matched. */
1811 #define SET_REGS_MATCHED() \
1812 do \
1813 { \
1814 if (!set_regs_matched_done) \
1815 { \
1816 active_reg_t r; \
1817 set_regs_matched_done = 1; \
1818 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1819 { \
1820 MATCHED_SOMETHING (reg_info[r]) \
1821 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1822 = 1; \
1823 } \
1824 } \
1825 } \
1826 while (0)
1827
1828 /* Registers are set to a sentinel when they haven't yet matched. */
1829 static CHAR_TYPE reg_unset_dummy;
1830 #define REG_UNSET_VALUE (®_unset_dummy)
1831 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1832
1833 /* Subroutine declarations and macros for regex_compile. */
1834
1835 static reg_errcode_t regex_compile _RE_ARGS ((const char *pattern, size_t size,
1836 reg_syntax_t syntax,
1837 struct re_pattern_buffer *bufp));
1838 static void store_op1 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc, int arg));
1839 static void store_op2 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc,
1840 int arg1, int arg2));
1841 static void insert_op1 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc,
1842 int arg, US_CHAR_TYPE *end));
1843 static void insert_op2 _RE_ARGS ((re_opcode_t op, US_CHAR_TYPE *loc,
1844 int arg1, int arg2, US_CHAR_TYPE *end));
1845 static boolean at_begline_loc_p _RE_ARGS ((const CHAR_TYPE *pattern,
1846 const CHAR_TYPE *p,
1847 reg_syntax_t syntax));
1848 static boolean at_endline_loc_p _RE_ARGS ((const CHAR_TYPE *p,
1849 const CHAR_TYPE *pend,
1850 reg_syntax_t syntax));
1851 #ifdef MBS_SUPPORT
1852 static reg_errcode_t compile_range _RE_ARGS ((CHAR_TYPE range_start,
1853 const CHAR_TYPE **p_ptr,
1854 const CHAR_TYPE *pend,
1855 char *translate,
1856 reg_syntax_t syntax,
1857 US_CHAR_TYPE *b,
1858 CHAR_TYPE *char_set));
1859 static void insert_space _RE_ARGS ((int num, CHAR_TYPE *loc, CHAR_TYPE *end));
1860 #else
1861 static reg_errcode_t compile_range _RE_ARGS ((unsigned int range_start,
1862 const CHAR_TYPE **p_ptr,
1863 const CHAR_TYPE *pend,
1864 char *translate,
1865 reg_syntax_t syntax,
1866 US_CHAR_TYPE *b));
1867 #endif /* MBS_SUPPORT */
1868
1869 /* Fetch the next character in the uncompiled pattern---translating it
1870 if necessary. Also cast from a signed character in the constant
1871 string passed to us by the user to an unsigned char that we can use
1872 as an array index (in, e.g., `translate'). */
1873 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1874 because it is impossible to allocate 4GB array for some encodings
1875 which have 4 byte character_set like UCS4. */
1876 #ifndef PATFETCH
1877 # ifdef MBS_SUPPORT
1878 # define PATFETCH(c) \
1879 do {if (p == pend) return REG_EEND; \
1880 c = (US_CHAR_TYPE) *p++; \
1881 if (translate && (c <= 0xff)) c = (US_CHAR_TYPE) translate[c]; \
1882 } while (0)
1883 # else
1884 # define PATFETCH(c) \
1885 do {if (p == pend) return REG_EEND; \
1886 c = (unsigned char) *p++; \
1887 if (translate) c = (unsigned char) translate[c]; \
1888 } while (0)
1889 # endif /* MBS_SUPPORT */
1890 #endif
1891
1892 /* Fetch the next character in the uncompiled pattern, with no
1893 translation. */
1894 #define PATFETCH_RAW(c) \
1895 do {if (p == pend) return REG_EEND; \
1896 c = (US_CHAR_TYPE) *p++; \
1897 } while (0)
1898
1899 /* Go backwards one character in the pattern. */
1900 #define PATUNFETCH p--
1901
1902
1903 /* If `translate' is non-null, return translate[D], else just D. We
1904 cast the subscript to translate because some data is declared as
1905 `char *', to avoid warnings when a string constant is passed. But
1906 when we use a character as a subscript we must make it unsigned. */
1907 /* ifdef MBS_SUPPORT, we translate only if character <= 0xff,
1908 because it is impossible to allocate 4GB array for some encodings
1909 which have 4 byte character_set like UCS4. */
1910 #ifndef TRANSLATE
1911 # ifdef MBS_SUPPORT
1912 # define TRANSLATE(d) \
1913 ((translate && ((US_CHAR_TYPE) (d)) <= 0xff) \
1914 ? (char) translate[(unsigned char) (d)] : (d))
1915 #else
1916 # define TRANSLATE(d) \
1917 (translate ? (char) translate[(unsigned char) (d)] : (d))
1918 # endif /* MBS_SUPPORT */
1919 #endif
1920
1921
1922 /* Macros for outputting the compiled pattern into `buffer'. */
1923
1924 /* If the buffer isn't allocated when it comes in, use this. */
1925 #define INIT_BUF_SIZE (32 * sizeof(US_CHAR_TYPE))
1926
1927 /* Make sure we have at least N more bytes of space in buffer. */
1928 #ifdef MBS_SUPPORT
1929 # define GET_BUFFER_SPACE(n) \
1930 while (((unsigned long)b - (unsigned long)COMPILED_BUFFER_VAR \
1931 + (n)*sizeof(CHAR_TYPE)) > bufp->allocated) \
1932 EXTEND_BUFFER ()
1933 #else
1934 # define GET_BUFFER_SPACE(n) \
1935 while ((unsigned long) (b - bufp->buffer + (n)) > bufp->allocated) \
1936 EXTEND_BUFFER ()
1937 #endif /* MBS_SUPPORT */
1938
1939 /* Make sure we have one more byte of buffer space and then add C to it. */
1940 #define BUF_PUSH(c) \
1941 do { \
1942 GET_BUFFER_SPACE (1); \
1943 *b++ = (US_CHAR_TYPE) (c); \
1944 } while (0)
1945
1946
1947 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1948 #define BUF_PUSH_2(c1, c2) \
1949 do { \
1950 GET_BUFFER_SPACE (2); \
1951 *b++ = (US_CHAR_TYPE) (c1); \
1952 *b++ = (US_CHAR_TYPE) (c2); \
1953 } while (0)
1954
1955
1956 /* As with BUF_PUSH_2, except for three bytes. */
1957 #define BUF_PUSH_3(c1, c2, c3) \
1958 do { \
1959 GET_BUFFER_SPACE (3); \
1960 *b++ = (US_CHAR_TYPE) (c1); \
1961 *b++ = (US_CHAR_TYPE) (c2); \
1962 *b++ = (US_CHAR_TYPE) (c3); \
1963 } while (0)
1964
1965 /* Store a jump with opcode OP at LOC to location TO. We store a
1966 relative address offset by the three bytes the jump itself occupies. */
1967 #define STORE_JUMP(op, loc, to) \
1968 store_op1 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)))
1969
1970 /* Likewise, for a two-argument jump. */
1971 #define STORE_JUMP2(op, loc, to, arg) \
1972 store_op2 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), arg)
1973
1974 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1975 #define INSERT_JUMP(op, loc, to) \
1976 insert_op1 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)), b)
1977
1978 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1979 #define INSERT_JUMP2(op, loc, to, arg) \
1980 insert_op2 (op, loc, (int) ((to) - (loc) - (1 + OFFSET_ADDRESS_SIZE)),\
1981 arg, b)
1982
1983
1984 /* This is not an arbitrary limit: the arguments which represent offsets
1985 into the pattern are two bytes long. So if 2^16 bytes turns out to
1986 be too small, many things would have to change. */
1987 /* Any other compiler which, like MSC, has allocation limit below 2^16
1988 bytes will have to use approach similar to what was done below for
1989 MSC and drop MAX_BUF_SIZE a bit. Otherwise you may end up
1990 reallocating to 0 bytes. Such thing is not going to work too well.
1991 You have been warned!! */
1992 #if defined _MSC_VER && !defined WIN32
1993 /* Microsoft C 16-bit versions limit malloc to approx 65512 bytes.
1994 The REALLOC define eliminates a flurry of conversion warnings,
1995 but is not required. */
1996 # define MAX_BUF_SIZE 65500L
1997 # define REALLOC(p,s) realloc ((p), (size_t) (s))
1998 #else
1999 # define MAX_BUF_SIZE (1L << 16)
2000 # define REALLOC(p,s) realloc ((p), (s))
2001 #endif
2002
2003 /* Extend the buffer by twice its current size via realloc and
2004 reset the pointers that pointed into the old block to point to the
2005 correct places in the new one. If extending the buffer results in it
2006 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
2007 #if __BOUNDED_POINTERS__
2008 # define SET_HIGH_BOUND(P) (__ptrhigh (P) = __ptrlow (P) + bufp->allocated)
2009 # define MOVE_BUFFER_POINTER(P) \
2010 (__ptrlow (P) += incr, SET_HIGH_BOUND (P), __ptrvalue (P) += incr)
2011 # define ELSE_EXTEND_BUFFER_HIGH_BOUND \
2012 else \
2013 { \
2014 SET_HIGH_BOUND (b); \
2015 SET_HIGH_BOUND (begalt); \
2016 if (fixup_alt_jump) \
2017 SET_HIGH_BOUND (fixup_alt_jump); \
2018 if (laststart) \
2019 SET_HIGH_BOUND (laststart); \
2020 if (pending_exact) \
2021 SET_HIGH_BOUND (pending_exact); \
2022 }
2023 #else
2024 # define MOVE_BUFFER_POINTER(P) (P) += incr
2025 # define ELSE_EXTEND_BUFFER_HIGH_BOUND
2026 #endif
2027
2028 #ifdef MBS_SUPPORT
2029 # define EXTEND_BUFFER() \
2030 do { \
2031 US_CHAR_TYPE *old_buffer = COMPILED_BUFFER_VAR; \
2032 int wchar_count; \
2033 if (bufp->allocated + sizeof(US_CHAR_TYPE) > MAX_BUF_SIZE) \
2034 return REG_ESIZE; \
2035 bufp->allocated <<= 1; \
2036 if (bufp->allocated > MAX_BUF_SIZE) \
2037 bufp->allocated = MAX_BUF_SIZE; \
2038 /* How many characters the new buffer can have? */ \
2039 wchar_count = bufp->allocated / sizeof(US_CHAR_TYPE); \
2040 if (wchar_count == 0) wchar_count = 1; \
2041 /* Truncate the buffer to CHAR_TYPE align. */ \
2042 bufp->allocated = wchar_count * sizeof(US_CHAR_TYPE); \
2043 RETALLOC (COMPILED_BUFFER_VAR, wchar_count, US_CHAR_TYPE); \
2044 bufp->buffer = (char*)COMPILED_BUFFER_VAR; \
2045 if (COMPILED_BUFFER_VAR == NULL) \
2046 return REG_ESPACE; \
2047 /* If the buffer moved, move all the pointers into it. */ \
2048 if (old_buffer != COMPILED_BUFFER_VAR) \
2049 { \
2050 ptrdiff_t incr = COMPILED_BUFFER_VAR - old_buffer; \
2051 MOVE_BUFFER_POINTER (b); \
2052 MOVE_BUFFER_POINTER (begalt); \
2053 if (fixup_alt_jump) \
2054 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2055 if (laststart) \
2056 MOVE_BUFFER_POINTER (laststart); \
2057 if (pending_exact) \
2058 MOVE_BUFFER_POINTER (pending_exact); \
2059 } \
2060 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2061 } while (0)
2062 #else
2063 # define EXTEND_BUFFER() \
2064 do { \
2065 US_CHAR_TYPE *old_buffer = COMPILED_BUFFER_VAR; \
2066 if (bufp->allocated == MAX_BUF_SIZE) \
2067 return REG_ESIZE; \
2068 bufp->allocated <<= 1; \
2069 if (bufp->allocated > MAX_BUF_SIZE) \
2070 bufp->allocated = MAX_BUF_SIZE; \
2071 bufp->buffer = (US_CHAR_TYPE *) REALLOC (COMPILED_BUFFER_VAR, \
2072 bufp->allocated); \
2073 if (COMPILED_BUFFER_VAR == NULL) \
2074 return REG_ESPACE; \
2075 /* If the buffer moved, move all the pointers into it. */ \
2076 if (old_buffer != COMPILED_BUFFER_VAR) \
2077 { \
2078 ptrdiff_t incr = COMPILED_BUFFER_VAR - old_buffer; \
2079 MOVE_BUFFER_POINTER (b); \
2080 MOVE_BUFFER_POINTER (begalt); \
2081 if (fixup_alt_jump) \
2082 MOVE_BUFFER_POINTER (fixup_alt_jump); \
2083 if (laststart) \
2084 MOVE_BUFFER_POINTER (laststart); \
2085 if (pending_exact) \
2086 MOVE_BUFFER_POINTER (pending_exact); \
2087 } \
2088 ELSE_EXTEND_BUFFER_HIGH_BOUND \
2089 } while (0)
2090 #endif /* MBS_SUPPORT */
2091
2092 /* Since we have one byte reserved for the register number argument to
2093 {start,stop}_memory, the maximum number of groups we can report
2094 things about is what fits in that byte. */
2095 #define MAX_REGNUM 255
2096
2097 /* But patterns can have more than `MAX_REGNUM' registers. We just
2098 ignore the excess. */
2099 typedef unsigned regnum_t;
2100
2101
2102 /* Macros for the compile stack. */
2103
2104 /* Since offsets can go either forwards or backwards, this type needs to
2105 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
2106 /* int may be not enough when sizeof(int) == 2. */
2107 typedef long pattern_offset_t;
2108
2109 typedef struct
2110 {
2111 pattern_offset_t begalt_offset;
2112 pattern_offset_t fixup_alt_jump;
2113 pattern_offset_t inner_group_offset;
2114 pattern_offset_t laststart_offset;
2115 regnum_t regnum;
2116 } compile_stack_elt_t;
2117
2118
2119 typedef struct
2120 {
2121 compile_stack_elt_t *stack;
2122 unsigned size;
2123 unsigned avail; /* Offset of next open position. */
2124 } compile_stack_type;
2125
2126
2127 #define INIT_COMPILE_STACK_SIZE 32
2128
2129 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
2130 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
2131
2132 /* The next available element. */
2133 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
2134
2135
2136 /* Set the bit for character C in a list. */
2137 #define SET_LIST_BIT(c) \
2138 (b[((unsigned char) (c)) / BYTEWIDTH] \
2139 |= 1 << (((unsigned char) c) % BYTEWIDTH))
2140
2141
2142 /* Get the next unsigned number in the uncompiled pattern. */
2143 #define GET_UNSIGNED_NUMBER(num) \
2144 { \
2145 while (p != pend) \
2146 { \
2147 PATFETCH (c); \
2148 if (! ('0' <= c && c <= '9')) \
2149 break; \
2150 if (num <= RE_DUP_MAX) \
2151 { \
2152 if (num < 0) \
2153 num = 0; \
2154 num = num * 10 + c - '0'; \
2155 } \
2156 } \
2157 }
2158
2159 #if defined _LIBC || WIDE_CHAR_SUPPORT
2160 /* The GNU C library provides support for user-defined character classes
2161 and the functions from ISO C amendement 1. */
2162 # ifdef CHARCLASS_NAME_MAX
2163 # define CHAR_CLASS_MAX_LENGTH CHARCLASS_NAME_MAX
2164 # else
2165 /* This shouldn't happen but some implementation might still have this
2166 problem. Use a reasonable default value. */
2167 # define CHAR_CLASS_MAX_LENGTH 256
2168 # endif
2169
2170 # ifdef _LIBC
2171 # define IS_CHAR_CLASS(string) __wctype (string)
2172 # else
2173 # define IS_CHAR_CLASS(string) wctype (string)
2174 # endif
2175 #else
2176 # define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
2177
2178 # define IS_CHAR_CLASS(string) \
2179 (STREQ (string, "alpha") || STREQ (string, "upper") \
2180 || STREQ (string, "lower") || STREQ (string, "digit") \
2181 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
2182 || STREQ (string, "space") || STREQ (string, "print") \
2183 || STREQ (string, "punct") || STREQ (string, "graph") \
2184 || STREQ (string, "cntrl") || STREQ (string, "blank"))
2185 #endif
2186
2187 #ifndef MATCH_MAY_ALLOCATE
2188
2189 /* If we cannot allocate large objects within re_match_2_internal,
2190 we make the fail stack and register vectors global.
2191 The fail stack, we grow to the maximum size when a regexp
2192 is compiled.
2193 The register vectors, we adjust in size each time we
2194 compile a regexp, according to the number of registers it needs. */
2195
2196 static fail_stack_type fail_stack;
2197
2198 /* Size with which the following vectors are currently allocated.
2199 That is so we can make them bigger as needed,
2200 but never make them smaller. */
2201 static int regs_allocated_size;
2202
2203 static const char ** regstart, ** regend;
2204 static const char ** old_regstart, ** old_regend;
2205 static const char **best_regstart, **best_regend;
2206 static register_info_type *reg_info;
2207 static const char **reg_dummy;
2208 static register_info_type *reg_info_dummy;
2209
2210 /* Make the register vectors big enough for NUM_REGS registers,
2211 but don't make them smaller. */
2212
2213 static
regex_grow_registers(num_regs)2214 regex_grow_registers (num_regs)
2215 int num_regs;
2216 {
2217 if (num_regs > regs_allocated_size)
2218 {
2219 RETALLOC_IF (regstart, num_regs, const char *);
2220 RETALLOC_IF (regend, num_regs, const char *);
2221 RETALLOC_IF (old_regstart, num_regs, const char *);
2222 RETALLOC_IF (old_regend, num_regs, const char *);
2223 RETALLOC_IF (best_regstart, num_regs, const char *);
2224 RETALLOC_IF (best_regend, num_regs, const char *);
2225 RETALLOC_IF (reg_info, num_regs, register_info_type);
2226 RETALLOC_IF (reg_dummy, num_regs, const char *);
2227 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
2228
2229 regs_allocated_size = num_regs;
2230 }
2231 }
2232
2233 #endif /* not MATCH_MAY_ALLOCATE */
2234
2235 static boolean group_in_compile_stack _RE_ARGS ((compile_stack_type
2236 compile_stack,
2237 regnum_t regnum));
2238
2239 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
2240 Returns one of error codes defined in `regex.h', or zero for success.
2241
2242 Assumes the `allocated' (and perhaps `buffer') and `translate'
2243 fields are set in BUFP on entry.
2244
2245 If it succeeds, results are put in BUFP (if it returns an error, the
2246 contents of BUFP are undefined):
2247 `buffer' is the compiled pattern;
2248 `syntax' is set to SYNTAX;
2249 `used' is set to the length of the compiled pattern;
2250 `fastmap_accurate' is zero;
2251 `re_nsub' is the number of subexpressions in PATTERN;
2252 `not_bol' and `not_eol' are zero;
2253
2254 The `fastmap' and `newline_anchor' fields are neither
2255 examined nor set. */
2256
2257 /* Return, freeing storage we allocated. */
2258 #ifdef MBS_SUPPORT
2259 # define FREE_STACK_RETURN(value) \
2260 return (free(pattern), free(mbs_offset), free(is_binary), free (compile_stack.stack), value)
2261 #else
2262 # define FREE_STACK_RETURN(value) \
2263 return (free (compile_stack.stack), value)
2264 #endif /* MBS_SUPPORT */
2265
2266 static reg_errcode_t
2267 #ifdef MBS_SUPPORT
regex_compile(cpattern,csize,syntax,bufp)2268 regex_compile (cpattern, csize, syntax, bufp)
2269 const char *cpattern;
2270 size_t csize;
2271 #else
2272 regex_compile (pattern, size, syntax, bufp)
2273 const char *pattern;
2274 size_t size;
2275 #endif /* MBS_SUPPORT */
2276 reg_syntax_t syntax;
2277 struct re_pattern_buffer *bufp;
2278 {
2279 /* We fetch characters from PATTERN here. Even though PATTERN is
2280 `char *' (i.e., signed), we declare these variables as unsigned, so
2281 they can be reliably used as array indices. */
2282 register US_CHAR_TYPE c, c1;
2283
2284 #ifdef MBS_SUPPORT
2285 /* A temporary space to keep wchar_t pattern and compiled pattern. */
2286 CHAR_TYPE *pattern, *COMPILED_BUFFER_VAR;
2287 size_t size;
2288 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
2289 int *mbs_offset = NULL;
2290 /* It hold whether each wchar_t is binary data or not. */
2291 char *is_binary = NULL;
2292 /* A flag whether exactn is handling binary data or not. */
2293 char is_exactn_bin = FALSE;
2294 #endif /* MBS_SUPPORT */
2295
2296 /* A random temporary spot in PATTERN. */
2297 const CHAR_TYPE *p1;
2298
2299 /* Points to the end of the buffer, where we should append. */
2300 register US_CHAR_TYPE *b;
2301
2302 /* Keeps track of unclosed groups. */
2303 compile_stack_type compile_stack;
2304
2305 /* Points to the current (ending) position in the pattern. */
2306 #ifdef MBS_SUPPORT
2307 const CHAR_TYPE *p;
2308 const CHAR_TYPE *pend;
2309 #else
2310 const CHAR_TYPE *p = pattern;
2311 const CHAR_TYPE *pend = pattern + size;
2312 #endif /* MBS_SUPPORT */
2313
2314 /* How to translate the characters in the pattern. */
2315 RE_TRANSLATE_TYPE translate = bufp->translate;
2316
2317 /* Address of the count-byte of the most recently inserted `exactn'
2318 command. This makes it possible to tell if a new exact-match
2319 character can be added to that command or if the character requires
2320 a new `exactn' command. */
2321 US_CHAR_TYPE *pending_exact = 0;
2322
2323 /* Address of start of the most recently finished expression.
2324 This tells, e.g., postfix * where to find the start of its
2325 operand. Reset at the beginning of groups and alternatives. */
2326 US_CHAR_TYPE *laststart = 0;
2327
2328 /* Address of beginning of regexp, or inside of last group. */
2329 US_CHAR_TYPE *begalt;
2330
2331 /* Address of the place where a forward jump should go to the end of
2332 the containing expression. Each alternative of an `or' -- except the
2333 last -- ends with a forward jump of this sort. */
2334 US_CHAR_TYPE *fixup_alt_jump = 0;
2335
2336 /* Counts open-groups as they are encountered. Remembered for the
2337 matching close-group on the compile stack, so the same register
2338 number is put in the stop_memory as the start_memory. */
2339 regnum_t regnum = 0;
2340
2341 #ifdef MBS_SUPPORT
2342 /* Initialize the wchar_t PATTERN and offset_buffer. */
2343 p = pend = pattern = TALLOC(csize + 1, CHAR_TYPE);
2344 p[csize] = L'\0'; /* sentinel */
2345 mbs_offset = TALLOC(csize + 1, int);
2346 is_binary = TALLOC(csize + 1, char);
2347 if (pattern == NULL || mbs_offset == NULL || is_binary == NULL)
2348 {
2349 if (pattern) free(pattern);
2350 if (mbs_offset) free(mbs_offset);
2351 if (is_binary) free(is_binary);
2352 return REG_ESPACE;
2353 }
2354 size = convert_mbs_to_wcs(pattern, cpattern, csize, mbs_offset, is_binary);
2355 pend = p + size;
2356 if (size < 0)
2357 {
2358 if (pattern) free(pattern);
2359 if (mbs_offset) free(mbs_offset);
2360 if (is_binary) free(is_binary);
2361 return REG_BADPAT;
2362 }
2363 #endif
2364
2365 #ifdef DEBUG
2366 DEBUG_PRINT1 ("\nCompiling pattern: ");
2367 if (debug)
2368 {
2369 unsigned debug_count;
2370
2371 for (debug_count = 0; debug_count < size; debug_count++)
2372 PUT_CHAR (pattern[debug_count]);
2373 putchar ('\n');
2374 }
2375 #endif /* DEBUG */
2376
2377 /* Initialize the compile stack. */
2378 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
2379 if (compile_stack.stack == NULL)
2380 {
2381 #ifdef MBS_SUPPORT
2382 if (pattern) free(pattern);
2383 if (mbs_offset) free(mbs_offset);
2384 if (is_binary) free(is_binary);
2385 #endif
2386 return REG_ESPACE;
2387 }
2388
2389 compile_stack.size = INIT_COMPILE_STACK_SIZE;
2390 compile_stack.avail = 0;
2391
2392 /* Initialize the pattern buffer. */
2393 bufp->syntax = syntax;
2394 bufp->fastmap_accurate = 0;
2395 bufp->not_bol = bufp->not_eol = 0;
2396
2397 /* Set `used' to zero, so that if we return an error, the pattern
2398 printer (for debugging) will think there's no pattern. We reset it
2399 at the end. */
2400 bufp->used = 0;
2401
2402 /* Always count groups, whether or not bufp->no_sub is set. */
2403 bufp->re_nsub = 0;
2404
2405 #if !defined emacs && !defined SYNTAX_TABLE
2406 /* Initialize the syntax table. */
2407 init_syntax_once ();
2408 #endif
2409
2410 if (bufp->allocated == 0)
2411 {
2412 if (bufp->buffer)
2413 { /* If zero allocated, but buffer is non-null, try to realloc
2414 enough space. This loses if buffer's address is bogus, but
2415 that is the user's responsibility. */
2416 #ifdef MBS_SUPPORT
2417 /* Free bufp->buffer and allocate an array for wchar_t pattern
2418 buffer. */
2419 free(bufp->buffer);
2420 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE/sizeof(US_CHAR_TYPE),
2421 US_CHAR_TYPE);
2422 #else
2423 RETALLOC (COMPILED_BUFFER_VAR, INIT_BUF_SIZE, US_CHAR_TYPE);
2424 #endif /* MBS_SUPPORT */
2425 }
2426 else
2427 { /* Caller did not allocate a buffer. Do it for them. */
2428 COMPILED_BUFFER_VAR = TALLOC (INIT_BUF_SIZE / sizeof(US_CHAR_TYPE),
2429 US_CHAR_TYPE);
2430 }
2431
2432 if (!COMPILED_BUFFER_VAR) FREE_STACK_RETURN (REG_ESPACE);
2433 #ifdef MBS_SUPPORT
2434 bufp->buffer = (char*)COMPILED_BUFFER_VAR;
2435 #endif /* MBS_SUPPORT */
2436 bufp->allocated = INIT_BUF_SIZE;
2437 }
2438 #ifdef MBS_SUPPORT
2439 else
2440 COMPILED_BUFFER_VAR = (US_CHAR_TYPE*) bufp->buffer;
2441 #endif
2442
2443 begalt = b = COMPILED_BUFFER_VAR;
2444
2445 /* Loop through the uncompiled pattern until we're at the end. */
2446 while (p != pend)
2447 {
2448 PATFETCH (c);
2449
2450 switch (c)
2451 {
2452 case '^':
2453 {
2454 if ( /* If at start of pattern, it's an operator. */
2455 p == pattern + 1
2456 /* If context independent, it's an operator. */
2457 || syntax & RE_CONTEXT_INDEP_ANCHORS
2458 /* Otherwise, depends on what's come before. */
2459 || at_begline_loc_p (pattern, p, syntax))
2460 BUF_PUSH (begline);
2461 else
2462 goto normal_char;
2463 }
2464 break;
2465
2466
2467 case '$':
2468 {
2469 if ( /* If at end of pattern, it's an operator. */
2470 p == pend
2471 /* If context independent, it's an operator. */
2472 || syntax & RE_CONTEXT_INDEP_ANCHORS
2473 /* Otherwise, depends on what's next. */
2474 || at_endline_loc_p (p, pend, syntax))
2475 BUF_PUSH (endline);
2476 else
2477 goto normal_char;
2478 }
2479 break;
2480
2481
2482 case '+':
2483 case '?':
2484 if ((syntax & RE_BK_PLUS_QM)
2485 || (syntax & RE_LIMITED_OPS))
2486 goto normal_char;
2487 handle_plus:
2488 case '*':
2489 /* If there is no previous pattern... */
2490 if (!laststart)
2491 {
2492 if (syntax & RE_CONTEXT_INVALID_OPS)
2493 FREE_STACK_RETURN (REG_BADRPT);
2494 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
2495 goto normal_char;
2496 }
2497
2498 {
2499 /* Are we optimizing this jump? */
2500 boolean keep_string_p = false;
2501
2502 /* 1 means zero (many) matches is allowed. */
2503 char zero_times_ok = 0, many_times_ok = 0;
2504
2505 /* If there is a sequence of repetition chars, collapse it
2506 down to just one (the right one). We can't combine
2507 interval operators with these because of, e.g., `a{2}*',
2508 which should only match an even number of `a's. */
2509
2510 for (;;)
2511 {
2512 zero_times_ok |= c != '+';
2513 many_times_ok |= c != '?';
2514
2515 if (p == pend)
2516 break;
2517
2518 PATFETCH (c);
2519
2520 if (c == '*'
2521 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
2522 ;
2523
2524 else if (syntax & RE_BK_PLUS_QM && c == '\\')
2525 {
2526 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2527
2528 PATFETCH (c1);
2529 if (!(c1 == '+' || c1 == '?'))
2530 {
2531 PATUNFETCH;
2532 PATUNFETCH;
2533 break;
2534 }
2535
2536 c = c1;
2537 }
2538 else
2539 {
2540 PATUNFETCH;
2541 break;
2542 }
2543
2544 /* If we get here, we found another repeat character. */
2545 }
2546
2547 /* Star, etc. applied to an empty pattern is equivalent
2548 to an empty pattern. */
2549 if (!laststart)
2550 break;
2551
2552 /* Now we know whether or not zero matches is allowed
2553 and also whether or not two or more matches is allowed. */
2554 if (many_times_ok)
2555 { /* More than one repetition is allowed, so put in at the
2556 end a backward relative jump from `b' to before the next
2557 jump we're going to put in below (which jumps from
2558 laststart to after this jump).
2559
2560 But if we are at the `*' in the exact sequence `.*\n',
2561 insert an unconditional jump backwards to the .,
2562 instead of the beginning of the loop. This way we only
2563 push a failure point once, instead of every time
2564 through the loop. */
2565 assert (p - 1 > pattern);
2566
2567 /* Allocate the space for the jump. */
2568 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2569
2570 /* We know we are not at the first character of the pattern,
2571 because laststart was nonzero. And we've already
2572 incremented `p', by the way, to be the character after
2573 the `*'. Do we have to do something analogous here
2574 for null bytes, because of RE_DOT_NOT_NULL? */
2575 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
2576 && zero_times_ok
2577 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
2578 && !(syntax & RE_DOT_NEWLINE))
2579 { /* We have .*\n. */
2580 STORE_JUMP (jump, b, laststart);
2581 keep_string_p = true;
2582 }
2583 else
2584 /* Anything else. */
2585 STORE_JUMP (maybe_pop_jump, b, laststart -
2586 (1 + OFFSET_ADDRESS_SIZE));
2587
2588 /* We've added more stuff to the buffer. */
2589 b += 1 + OFFSET_ADDRESS_SIZE;
2590 }
2591
2592 /* On failure, jump from laststart to b + 3, which will be the
2593 end of the buffer after this jump is inserted. */
2594 /* ifdef MBS_SUPPORT, 'b + 1 + OFFSET_ADDRESS_SIZE' instead of
2595 'b + 3'. */
2596 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2597 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
2598 : on_failure_jump,
2599 laststart, b + 1 + OFFSET_ADDRESS_SIZE);
2600 pending_exact = 0;
2601 b += 1 + OFFSET_ADDRESS_SIZE;
2602
2603 if (!zero_times_ok)
2604 {
2605 /* At least one repetition is required, so insert a
2606 `dummy_failure_jump' before the initial
2607 `on_failure_jump' instruction of the loop. This
2608 effects a skip over that instruction the first time
2609 we hit that loop. */
2610 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
2611 INSERT_JUMP (dummy_failure_jump, laststart, laststart +
2612 2 + 2 * OFFSET_ADDRESS_SIZE);
2613 b += 1 + OFFSET_ADDRESS_SIZE;
2614 }
2615 }
2616 break;
2617
2618
2619 case '.':
2620 laststart = b;
2621 BUF_PUSH (anychar);
2622 break;
2623
2624
2625 case '[':
2626 {
2627 boolean had_char_class = false;
2628 #ifdef MBS_SUPPORT
2629 CHAR_TYPE range_start = 0xffffffff;
2630 #else
2631 unsigned int range_start = 0xffffffff;
2632 #endif
2633 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2634
2635 #ifdef MBS_SUPPORT
2636 /* We assume a charset(_not) structure as a wchar_t array.
2637 charset[0] = (re_opcode_t) charset(_not)
2638 charset[1] = l (= length of char_classes)
2639 charset[2] = m (= length of collating_symbols)
2640 charset[3] = n (= length of equivalence_classes)
2641 charset[4] = o (= length of char_ranges)
2642 charset[5] = p (= length of chars)
2643
2644 charset[6] = char_class (wctype_t)
2645 charset[6+CHAR_CLASS_SIZE] = char_class (wctype_t)
2646 ...
2647 charset[l+5] = char_class (wctype_t)
2648
2649 charset[l+6] = collating_symbol (wchar_t)
2650 ...
2651 charset[l+m+5] = collating_symbol (wchar_t)
2652 ifdef _LIBC we use the index if
2653 _NL_COLLATE_SYMB_EXTRAMB instead of
2654 wchar_t string.
2655
2656 charset[l+m+6] = equivalence_classes (wchar_t)
2657 ...
2658 charset[l+m+n+5] = equivalence_classes (wchar_t)
2659 ifdef _LIBC we use the index in
2660 _NL_COLLATE_WEIGHT instead of
2661 wchar_t string.
2662
2663 charset[l+m+n+6] = range_start
2664 charset[l+m+n+7] = range_end
2665 ...
2666 charset[l+m+n+2o+4] = range_start
2667 charset[l+m+n+2o+5] = range_end
2668 ifdef _LIBC we use the value looked up
2669 in _NL_COLLATE_COLLSEQ instead of
2670 wchar_t character.
2671
2672 charset[l+m+n+2o+6] = char
2673 ...
2674 charset[l+m+n+2o+p+5] = char
2675
2676 */
2677
2678 /* We need at least 6 spaces: the opcode, the length of
2679 char_classes, the length of collating_symbols, the length of
2680 equivalence_classes, the length of char_ranges, the length of
2681 chars. */
2682 GET_BUFFER_SPACE (6);
2683
2684 /* Save b as laststart. And We use laststart as the pointer
2685 to the first element of the charset here.
2686 In other words, laststart[i] indicates charset[i]. */
2687 laststart = b;
2688
2689 /* We test `*p == '^' twice, instead of using an if
2690 statement, so we only need one BUF_PUSH. */
2691 BUF_PUSH (*p == '^' ? charset_not : charset);
2692 if (*p == '^')
2693 p++;
2694
2695 /* Push the length of char_classes, the length of
2696 collating_symbols, the length of equivalence_classes, the
2697 length of char_ranges and the length of chars. */
2698 BUF_PUSH_3 (0, 0, 0);
2699 BUF_PUSH_2 (0, 0);
2700
2701 /* Remember the first position in the bracket expression. */
2702 p1 = p;
2703
2704 /* charset_not matches newline according to a syntax bit. */
2705 if ((re_opcode_t) b[-6] == charset_not
2706 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
2707 {
2708 BUF_PUSH('\n');
2709 laststart[5]++; /* Update the length of characters */
2710 }
2711
2712 /* Read in characters and ranges, setting map bits. */
2713 for (;;)
2714 {
2715 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2716
2717 PATFETCH (c);
2718
2719 /* \ might escape characters inside [...] and [^...]. */
2720 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
2721 {
2722 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2723
2724 PATFETCH (c1);
2725 BUF_PUSH(c1);
2726 laststart[5]++; /* Update the length of chars */
2727 range_start = c1;
2728 continue;
2729 }
2730
2731 /* Could be the end of the bracket expression. If it's
2732 not (i.e., when the bracket expression is `[]' so
2733 far), the ']' character bit gets set way below. */
2734 if (c == ']' && p != p1 + 1)
2735 break;
2736
2737 /* Look ahead to see if it's a range when the last thing
2738 was a character class. */
2739 if (had_char_class && c == '-' && *p != ']')
2740 FREE_STACK_RETURN (REG_ERANGE);
2741
2742 /* Look ahead to see if it's a range when the last thing
2743 was a character: if this is a hyphen not at the
2744 beginning or the end of a list, then it's the range
2745 operator. */
2746 if (c == '-'
2747 && !(p - 2 >= pattern && p[-2] == '[')
2748 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
2749 && *p != ']')
2750 {
2751 reg_errcode_t ret;
2752 /* Allocate the space for range_start and range_end. */
2753 GET_BUFFER_SPACE (2);
2754 /* Update the pointer to indicate end of buffer. */
2755 b += 2;
2756 ret = compile_range (range_start, &p, pend, translate,
2757 syntax, b, laststart);
2758 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2759 range_start = 0xffffffff;
2760 }
2761 else if (p[0] == '-' && p[1] != ']')
2762 { /* This handles ranges made up of characters only. */
2763 reg_errcode_t ret;
2764
2765 /* Move past the `-'. */
2766 PATFETCH (c1);
2767 /* Allocate the space for range_start and range_end. */
2768 GET_BUFFER_SPACE (2);
2769 /* Update the pointer to indicate end of buffer. */
2770 b += 2;
2771 ret = compile_range (c, &p, pend, translate, syntax, b,
2772 laststart);
2773 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
2774 range_start = 0xffffffff;
2775 }
2776
2777 /* See if we're at the beginning of a possible character
2778 class. */
2779 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2780 { /* Leave room for the null. */
2781 char str[CHAR_CLASS_MAX_LENGTH + 1];
2782
2783 PATFETCH (c);
2784 c1 = 0;
2785
2786 /* If pattern is `[[:'. */
2787 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2788
2789 for (;;)
2790 {
2791 PATFETCH (c);
2792 if ((c == ':' && *p == ']') || p == pend)
2793 break;
2794 if (c1 < CHAR_CLASS_MAX_LENGTH)
2795 str[c1++] = c;
2796 else
2797 /* This is in any case an invalid class name. */
2798 str[0] = '\0';
2799 }
2800 str[c1] = '\0';
2801
2802 /* If isn't a word bracketed by `[:' and `:]':
2803 undo the ending character, the letters, and leave
2804 the leading `:' and `[' (but store them as character). */
2805 if (c == ':' && *p == ']')
2806 {
2807 wctype_t wt;
2808 uintptr_t alignedp;
2809
2810 /* Query the character class as wctype_t. */
2811 wt = IS_CHAR_CLASS (str);
2812 if (wt == 0)
2813 FREE_STACK_RETURN (REG_ECTYPE);
2814
2815 /* Throw away the ] at the end of the character
2816 class. */
2817 PATFETCH (c);
2818
2819 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2820
2821 /* Allocate the space for character class. */
2822 GET_BUFFER_SPACE(CHAR_CLASS_SIZE);
2823 /* Update the pointer to indicate end of buffer. */
2824 b += CHAR_CLASS_SIZE;
2825 /* Move data which follow character classes
2826 not to violate the data. */
2827 insert_space(CHAR_CLASS_SIZE,
2828 laststart + 6 + laststart[1],
2829 b - 1);
2830 alignedp = ((uintptr_t)(laststart + 6 + laststart[1])
2831 + __alignof__(wctype_t) - 1)
2832 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
2833 /* Store the character class. */
2834 *((wctype_t*)alignedp) = wt;
2835 /* Update length of char_classes */
2836 laststart[1] += CHAR_CLASS_SIZE;
2837
2838 had_char_class = true;
2839 }
2840 else
2841 {
2842 c1++;
2843 while (c1--)
2844 PATUNFETCH;
2845 BUF_PUSH ('[');
2846 BUF_PUSH (':');
2847 laststart[5] += 2; /* Update the length of characters */
2848 range_start = ':';
2849 had_char_class = false;
2850 }
2851 }
2852 else if (syntax & RE_CHAR_CLASSES && c == '[' && (*p == '='
2853 || *p == '.'))
2854 {
2855 CHAR_TYPE str[128]; /* Should be large enough. */
2856 CHAR_TYPE delim = *p; /* '=' or '.' */
2857 # ifdef _LIBC
2858 uint32_t nrules =
2859 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
2860 # endif
2861 PATFETCH (c);
2862 c1 = 0;
2863
2864 /* If pattern is `[[=' or '[[.'. */
2865 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2866
2867 for (;;)
2868 {
2869 PATFETCH (c);
2870 if ((c == delim && *p == ']') || p == pend)
2871 break;
2872 if (c1 < sizeof (str) - 1)
2873 str[c1++] = c;
2874 else
2875 /* This is in any case an invalid class name. */
2876 str[0] = '\0';
2877 }
2878 str[c1] = '\0';
2879
2880 if (c == delim && *p == ']' && str[0] != '\0')
2881 {
2882 unsigned int i, offset;
2883 /* If we have no collation data we use the default
2884 collation in which each character is in a class
2885 by itself. It also means that ASCII is the
2886 character set and therefore we cannot have character
2887 with more than one byte in the multibyte
2888 representation. */
2889
2890 /* If not defined _LIBC, we push the name and
2891 `\0' for the sake of matching performance. */
2892 int datasize = c1 + 1;
2893
2894 # ifdef _LIBC
2895 int32_t idx = 0;
2896 if (nrules == 0)
2897 # endif
2898 {
2899 if (c1 != 1)
2900 FREE_STACK_RETURN (REG_ECOLLATE);
2901 }
2902 # ifdef _LIBC
2903 else
2904 {
2905 const int32_t *table;
2906 const int32_t *weights;
2907 const int32_t *extra;
2908 const int32_t *indirect;
2909 wint_t *cp;
2910
2911 /* This #include defines a local function! */
2912 # include <locale/weightwc.h>
2913
2914 if(delim == '=')
2915 {
2916 /* We push the index for equivalence class. */
2917 cp = (wint_t*)str;
2918
2919 table = (const int32_t *)
2920 _NL_CURRENT (LC_COLLATE,
2921 _NL_COLLATE_TABLEWC);
2922 weights = (const int32_t *)
2923 _NL_CURRENT (LC_COLLATE,
2924 _NL_COLLATE_WEIGHTWC);
2925 extra = (const int32_t *)
2926 _NL_CURRENT (LC_COLLATE,
2927 _NL_COLLATE_EXTRAWC);
2928 indirect = (const int32_t *)
2929 _NL_CURRENT (LC_COLLATE,
2930 _NL_COLLATE_INDIRECTWC);
2931
2932 idx = findidx ((const wint_t**)&cp);
2933 if (idx == 0 || cp < (wint_t*) str + c1)
2934 /* This is no valid character. */
2935 FREE_STACK_RETURN (REG_ECOLLATE);
2936
2937 str[0] = (wchar_t)idx;
2938 }
2939 else /* delim == '.' */
2940 {
2941 /* We push collation sequence value
2942 for collating symbol. */
2943 int32_t table_size;
2944 const int32_t *symb_table;
2945 const unsigned char *extra;
2946 int32_t idx;
2947 int32_t elem;
2948 int32_t second;
2949 int32_t hash;
2950 char char_str[c1];
2951
2952 /* We have to convert the name to a single-byte
2953 string. This is possible since the names
2954 consist of ASCII characters and the internal
2955 representation is UCS4. */
2956 for (i = 0; i < c1; ++i)
2957 char_str[i] = str[i];
2958
2959 table_size =
2960 _NL_CURRENT_WORD (LC_COLLATE,
2961 _NL_COLLATE_SYMB_HASH_SIZEMB);
2962 symb_table = (const int32_t *)
2963 _NL_CURRENT (LC_COLLATE,
2964 _NL_COLLATE_SYMB_TABLEMB);
2965 extra = (const unsigned char *)
2966 _NL_CURRENT (LC_COLLATE,
2967 _NL_COLLATE_SYMB_EXTRAMB);
2968
2969 /* Locate the character in the hashing table. */
2970 hash = elem_hash (char_str, c1);
2971
2972 idx = 0;
2973 elem = hash % table_size;
2974 second = hash % (table_size - 2);
2975 while (symb_table[2 * elem] != 0)
2976 {
2977 /* First compare the hashing value. */
2978 if (symb_table[2 * elem] == hash
2979 && c1 == extra[symb_table[2 * elem + 1]]
2980 && memcmp (str,
2981 &extra[symb_table[2 * elem + 1]
2982 + 1], c1) == 0)
2983 {
2984 /* Yep, this is the entry. */
2985 idx = symb_table[2 * elem + 1];
2986 idx += 1 + extra[idx];
2987 break;
2988 }
2989
2990 /* Next entry. */
2991 elem += second;
2992 }
2993
2994 if (symb_table[2 * elem] != 0)
2995 {
2996 /* Compute the index of the byte sequence
2997 in the table. */
2998 idx += 1 + extra[idx];
2999 /* Adjust for the alignment. */
3000 idx = (idx + 3) & ~4;
3001
3002 str[0] = (wchar_t) idx + 4;
3003 }
3004 else if (symb_table[2 * elem] == 0 && c1 == 1)
3005 {
3006 /* No valid character. Match it as a
3007 single byte character. */
3008 had_char_class = false;
3009 BUF_PUSH(str[0]);
3010 /* Update the length of characters */
3011 laststart[5]++;
3012 range_start = str[0];
3013
3014 /* Throw away the ] at the end of the
3015 collating symbol. */
3016 PATFETCH (c);
3017 /* exit from the switch block. */
3018 continue;
3019 }
3020 else
3021 FREE_STACK_RETURN (REG_ECOLLATE);
3022 }
3023 datasize = 1;
3024 }
3025 # endif
3026 /* Throw away the ] at the end of the equivalence
3027 class (or collating symbol). */
3028 PATFETCH (c);
3029
3030 /* Allocate the space for the equivalence class
3031 (or collating symbol) (and '\0' if needed). */
3032 GET_BUFFER_SPACE(datasize);
3033 /* Update the pointer to indicate end of buffer. */
3034 b += datasize;
3035
3036 if (delim == '=')
3037 { /* equivalence class */
3038 /* Calculate the offset of char_ranges,
3039 which is next to equivalence_classes. */
3040 offset = laststart[1] + laststart[2]
3041 + laststart[3] +6;
3042 /* Insert space. */
3043 insert_space(datasize, laststart + offset, b - 1);
3044
3045 /* Write the equivalence_class and \0. */
3046 for (i = 0 ; i < datasize ; i++)
3047 laststart[offset + i] = str[i];
3048
3049 /* Update the length of equivalence_classes. */
3050 laststart[3] += datasize;
3051 had_char_class = true;
3052 }
3053 else /* delim == '.' */
3054 { /* collating symbol */
3055 /* Calculate the offset of the equivalence_classes,
3056 which is next to collating_symbols. */
3057 offset = laststart[1] + laststart[2] + 6;
3058 /* Insert space and write the collationg_symbol
3059 and \0. */
3060 insert_space(datasize, laststart + offset, b-1);
3061 for (i = 0 ; i < datasize ; i++)
3062 laststart[offset + i] = str[i];
3063
3064 /* In re_match_2_internal if range_start < -1, we
3065 assume -range_start is the offset of the
3066 collating symbol which is specified as
3067 the character of the range start. So we assign
3068 -(laststart[1] + laststart[2] + 6) to
3069 range_start. */
3070 range_start = -(laststart[1] + laststart[2] + 6);
3071 /* Update the length of collating_symbol. */
3072 laststart[2] += datasize;
3073 had_char_class = false;
3074 }
3075 }
3076 else
3077 {
3078 c1++;
3079 while (c1--)
3080 PATUNFETCH;
3081 BUF_PUSH ('[');
3082 BUF_PUSH (delim);
3083 laststart[5] += 2; /* Update the length of characters */
3084 range_start = delim;
3085 had_char_class = false;
3086 }
3087 }
3088 else
3089 {
3090 had_char_class = false;
3091 BUF_PUSH(c);
3092 laststart[5]++; /* Update the length of characters */
3093 range_start = c;
3094 }
3095 }
3096
3097 #else /* not MBS_SUPPORT */
3098 /* Ensure that we have enough space to push a charset: the
3099 opcode, the length count, and the bitset; 34 bytes in all. */
3100 GET_BUFFER_SPACE (34);
3101
3102 laststart = b;
3103
3104 /* We test `*p == '^' twice, instead of using an if
3105 statement, so we only need one BUF_PUSH. */
3106 BUF_PUSH (*p == '^' ? charset_not : charset);
3107 if (*p == '^')
3108 p++;
3109
3110 /* Remember the first position in the bracket expression. */
3111 p1 = p;
3112
3113 /* Push the number of bytes in the bitmap. */
3114 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
3115
3116 /* Clear the whole map. */
3117 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
3118
3119 /* charset_not matches newline according to a syntax bit. */
3120 if ((re_opcode_t) b[-2] == charset_not
3121 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
3122 SET_LIST_BIT ('\n');
3123
3124 /* Read in characters and ranges, setting map bits. */
3125 for (;;)
3126 {
3127 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3128
3129 PATFETCH (c);
3130
3131 /* \ might escape characters inside [...] and [^...]. */
3132 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
3133 {
3134 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3135
3136 PATFETCH (c1);
3137 SET_LIST_BIT (c1);
3138 range_start = c1;
3139 continue;
3140 }
3141
3142 /* Could be the end of the bracket expression. If it's
3143 not (i.e., when the bracket expression is `[]' so
3144 far), the ']' character bit gets set way below. */
3145 if (c == ']' && p != p1 + 1)
3146 break;
3147
3148 /* Look ahead to see if it's a range when the last thing
3149 was a character class. */
3150 if (had_char_class && c == '-' && *p != ']')
3151 FREE_STACK_RETURN (REG_ERANGE);
3152
3153 /* Look ahead to see if it's a range when the last thing
3154 was a character: if this is a hyphen not at the
3155 beginning or the end of a list, then it's the range
3156 operator. */
3157 if (c == '-'
3158 && !(p - 2 >= pattern && p[-2] == '[')
3159 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
3160 && *p != ']')
3161 {
3162 reg_errcode_t ret
3163 = compile_range (range_start, &p, pend, translate,
3164 syntax, b);
3165 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3166 range_start = 0xffffffff;
3167 }
3168
3169 else if (p[0] == '-' && p[1] != ']')
3170 { /* This handles ranges made up of characters only. */
3171 reg_errcode_t ret;
3172
3173 /* Move past the `-'. */
3174 PATFETCH (c1);
3175
3176 ret = compile_range (c, &p, pend, translate, syntax, b);
3177 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
3178 range_start = 0xffffffff;
3179 }
3180
3181 /* See if we're at the beginning of a possible character
3182 class. */
3183
3184 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
3185 { /* Leave room for the null. */
3186 char str[CHAR_CLASS_MAX_LENGTH + 1];
3187
3188 PATFETCH (c);
3189 c1 = 0;
3190
3191 /* If pattern is `[[:'. */
3192 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3193
3194 for (;;)
3195 {
3196 PATFETCH (c);
3197 if ((c == ':' && *p == ']') || p == pend)
3198 break;
3199 if (c1 < CHAR_CLASS_MAX_LENGTH)
3200 str[c1++] = c;
3201 else
3202 /* This is in any case an invalid class name. */
3203 str[0] = '\0';
3204 }
3205 str[c1] = '\0';
3206
3207 /* If isn't a word bracketed by `[:' and `:]':
3208 undo the ending character, the letters, and leave
3209 the leading `:' and `[' (but set bits for them). */
3210 if (c == ':' && *p == ']')
3211 {
3212 # if defined _LIBC || WIDE_CHAR_SUPPORT
3213 boolean is_lower = STREQ (str, "lower");
3214 boolean is_upper = STREQ (str, "upper");
3215 wctype_t wt;
3216 int ch;
3217
3218 wt = IS_CHAR_CLASS (str);
3219 if (wt == 0)
3220 FREE_STACK_RETURN (REG_ECTYPE);
3221
3222 /* Throw away the ] at the end of the character
3223 class. */
3224 PATFETCH (c);
3225
3226 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3227
3228 for (ch = 0; ch < 1 << BYTEWIDTH; ++ch)
3229 {
3230 # ifdef _LIBC
3231 if (__iswctype (__btowc (ch), wt))
3232 SET_LIST_BIT (ch);
3233 # else
3234 if (iswctype (btowc (ch), wt))
3235 SET_LIST_BIT (ch);
3236 # endif
3237
3238 if (translate && (is_upper || is_lower)
3239 && (ISUPPER (ch) || ISLOWER (ch)))
3240 SET_LIST_BIT (ch);
3241 }
3242
3243 had_char_class = true;
3244 # else
3245 int ch;
3246 boolean is_alnum = STREQ (str, "alnum");
3247 boolean is_alpha = STREQ (str, "alpha");
3248 boolean is_blank = STREQ (str, "blank");
3249 boolean is_cntrl = STREQ (str, "cntrl");
3250 boolean is_digit = STREQ (str, "digit");
3251 boolean is_graph = STREQ (str, "graph");
3252 boolean is_lower = STREQ (str, "lower");
3253 boolean is_print = STREQ (str, "print");
3254 boolean is_punct = STREQ (str, "punct");
3255 boolean is_space = STREQ (str, "space");
3256 boolean is_upper = STREQ (str, "upper");
3257 boolean is_xdigit = STREQ (str, "xdigit");
3258
3259 if (!IS_CHAR_CLASS (str))
3260 FREE_STACK_RETURN (REG_ECTYPE);
3261
3262 /* Throw away the ] at the end of the character
3263 class. */
3264 PATFETCH (c);
3265
3266 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3267
3268 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
3269 {
3270 /* This was split into 3 if's to
3271 avoid an arbitrary limit in some compiler. */
3272 if ( (is_alnum && ISALNUM (ch))
3273 || (is_alpha && ISALPHA (ch))
3274 || (is_blank && ISBLANK (ch))
3275 || (is_cntrl && ISCNTRL (ch)))
3276 SET_LIST_BIT (ch);
3277 if ( (is_digit && ISDIGIT (ch))
3278 || (is_graph && ISGRAPH (ch))
3279 || (is_lower && ISLOWER (ch))
3280 || (is_print && ISPRINT (ch)))
3281 SET_LIST_BIT (ch);
3282 if ( (is_punct && ISPUNCT (ch))
3283 || (is_space && ISSPACE (ch))
3284 || (is_upper && ISUPPER (ch))
3285 || (is_xdigit && ISXDIGIT (ch)))
3286 SET_LIST_BIT (ch);
3287 if ( translate && (is_upper || is_lower)
3288 && (ISUPPER (ch) || ISLOWER (ch)))
3289 SET_LIST_BIT (ch);
3290 }
3291 had_char_class = true;
3292 # endif /* libc || wctype.h */
3293 }
3294 else
3295 {
3296 c1++;
3297 while (c1--)
3298 PATUNFETCH;
3299 SET_LIST_BIT ('[');
3300 SET_LIST_BIT (':');
3301 range_start = ':';
3302 had_char_class = false;
3303 }
3304 }
3305 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '=')
3306 {
3307 unsigned char str[MB_LEN_MAX + 1];
3308 # ifdef _LIBC
3309 uint32_t nrules =
3310 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3311 # endif
3312
3313 PATFETCH (c);
3314 c1 = 0;
3315
3316 /* If pattern is `[[='. */
3317 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3318
3319 for (;;)
3320 {
3321 PATFETCH (c);
3322 if ((c == '=' && *p == ']') || p == pend)
3323 break;
3324 if (c1 < MB_LEN_MAX)
3325 str[c1++] = c;
3326 else
3327 /* This is in any case an invalid class name. */
3328 str[0] = '\0';
3329 }
3330 str[c1] = '\0';
3331
3332 if (c == '=' && *p == ']' && str[0] != '\0')
3333 {
3334 /* If we have no collation data we use the default
3335 collation in which each character is in a class
3336 by itself. It also means that ASCII is the
3337 character set and therefore we cannot have character
3338 with more than one byte in the multibyte
3339 representation. */
3340 # ifdef _LIBC
3341 if (nrules == 0)
3342 # endif
3343 {
3344 if (c1 != 1)
3345 FREE_STACK_RETURN (REG_ECOLLATE);
3346
3347 /* Throw away the ] at the end of the equivalence
3348 class. */
3349 PATFETCH (c);
3350
3351 /* Set the bit for the character. */
3352 SET_LIST_BIT (str[0]);
3353 }
3354 # ifdef _LIBC
3355 else
3356 {
3357 /* Try to match the byte sequence in `str' against
3358 those known to the collate implementation.
3359 First find out whether the bytes in `str' are
3360 actually from exactly one character. */
3361 const int32_t *table;
3362 const unsigned char *weights;
3363 const unsigned char *extra;
3364 const int32_t *indirect;
3365 int32_t idx;
3366 const unsigned char *cp = str;
3367 int ch;
3368
3369 /* This #include defines a local function! */
3370 # include <locale/weight.h>
3371
3372 table = (const int32_t *)
3373 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEMB);
3374 weights = (const unsigned char *)
3375 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTMB);
3376 extra = (const unsigned char *)
3377 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAMB);
3378 indirect = (const int32_t *)
3379 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTMB);
3380
3381 idx = findidx (&cp);
3382 if (idx == 0 || cp < str + c1)
3383 /* This is no valid character. */
3384 FREE_STACK_RETURN (REG_ECOLLATE);
3385
3386 /* Throw away the ] at the end of the equivalence
3387 class. */
3388 PATFETCH (c);
3389
3390 /* Now we have to go throught the whole table
3391 and find all characters which have the same
3392 first level weight.
3393
3394 XXX Note that this is not entirely correct.
3395 we would have to match multibyte sequences
3396 but this is not possible with the current
3397 implementation. */
3398 for (ch = 1; ch < 256; ++ch)
3399 /* XXX This test would have to be changed if we
3400 would allow matching multibyte sequences. */
3401 if (table[ch] > 0)
3402 {
3403 int32_t idx2 = table[ch];
3404 size_t len = weights[idx2];
3405
3406 /* Test whether the lenghts match. */
3407 if (weights[idx] == len)
3408 {
3409 /* They do. New compare the bytes of
3410 the weight. */
3411 size_t cnt = 0;
3412
3413 while (cnt < len
3414 && (weights[idx + 1 + cnt]
3415 == weights[idx2 + 1 + cnt]))
3416 ++cnt;
3417
3418 if (cnt == len)
3419 /* They match. Mark the character as
3420 acceptable. */
3421 SET_LIST_BIT (ch);
3422 }
3423 }
3424 }
3425 # endif
3426 had_char_class = true;
3427 }
3428 else
3429 {
3430 c1++;
3431 while (c1--)
3432 PATUNFETCH;
3433 SET_LIST_BIT ('[');
3434 SET_LIST_BIT ('=');
3435 range_start = '=';
3436 had_char_class = false;
3437 }
3438 }
3439 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == '.')
3440 {
3441 unsigned char str[128]; /* Should be large enough. */
3442 # ifdef _LIBC
3443 uint32_t nrules =
3444 _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
3445 # endif
3446
3447 PATFETCH (c);
3448 c1 = 0;
3449
3450 /* If pattern is `[[.'. */
3451 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
3452
3453 for (;;)
3454 {
3455 PATFETCH (c);
3456 if ((c == '.' && *p == ']') || p == pend)
3457 break;
3458 if (c1 < sizeof (str))
3459 str[c1++] = c;
3460 else
3461 /* This is in any case an invalid class name. */
3462 str[0] = '\0';
3463 }
3464 str[c1] = '\0';
3465
3466 if (c == '.' && *p == ']' && str[0] != '\0')
3467 {
3468 /* If we have no collation data we use the default
3469 collation in which each character is the name
3470 for its own class which contains only the one
3471 character. It also means that ASCII is the
3472 character set and therefore we cannot have character
3473 with more than one byte in the multibyte
3474 representation. */
3475 # ifdef _LIBC
3476 if (nrules == 0)
3477 # endif
3478 {
3479 if (c1 != 1)
3480 FREE_STACK_RETURN (REG_ECOLLATE);
3481
3482 /* Throw away the ] at the end of the equivalence
3483 class. */
3484 PATFETCH (c);
3485
3486 /* Set the bit for the character. */
3487 SET_LIST_BIT (str[0]);
3488 range_start = ((const unsigned char *) str)[0];
3489 }
3490 # ifdef _LIBC
3491 else
3492 {
3493 /* Try to match the byte sequence in `str' against
3494 those known to the collate implementation.
3495 First find out whether the bytes in `str' are
3496 actually from exactly one character. */
3497 int32_t table_size;
3498 const int32_t *symb_table;
3499 const unsigned char *extra;
3500 int32_t idx;
3501 int32_t elem;
3502 int32_t second;
3503 int32_t hash;
3504
3505 table_size =
3506 _NL_CURRENT_WORD (LC_COLLATE,
3507 _NL_COLLATE_SYMB_HASH_SIZEMB);
3508 symb_table = (const int32_t *)
3509 _NL_CURRENT (LC_COLLATE,
3510 _NL_COLLATE_SYMB_TABLEMB);
3511 extra = (const unsigned char *)
3512 _NL_CURRENT (LC_COLLATE,
3513 _NL_COLLATE_SYMB_EXTRAMB);
3514
3515 /* Locate the character in the hashing table. */
3516 hash = elem_hash (str, c1);
3517
3518 idx = 0;
3519 elem = hash % table_size;
3520 second = hash % (table_size - 2);
3521 while (symb_table[2 * elem] != 0)
3522 {
3523 /* First compare the hashing value. */
3524 if (symb_table[2 * elem] == hash
3525 && c1 == extra[symb_table[2 * elem + 1]]
3526 && memcmp (str,
3527 &extra[symb_table[2 * elem + 1]
3528 + 1],
3529 c1) == 0)
3530 {
3531 /* Yep, this is the entry. */
3532 idx = symb_table[2 * elem + 1];
3533 idx += 1 + extra[idx];
3534 break;
3535 }
3536
3537 /* Next entry. */
3538 elem += second;
3539 }
3540
3541 if (symb_table[2 * elem] == 0)
3542 /* This is no valid character. */
3543 FREE_STACK_RETURN (REG_ECOLLATE);
3544
3545 /* Throw away the ] at the end of the equivalence
3546 class. */
3547 PATFETCH (c);
3548
3549 /* Now add the multibyte character(s) we found
3550 to the accept list.
3551
3552 XXX Note that this is not entirely correct.
3553 we would have to match multibyte sequences
3554 but this is not possible with the current
3555 implementation. Also, we have to match
3556 collating symbols, which expand to more than
3557 one file, as a whole and not allow the
3558 individual bytes. */
3559 c1 = extra[idx++];
3560 if (c1 == 1)
3561 range_start = extra[idx];
3562 while (c1-- > 0)
3563 {
3564 SET_LIST_BIT (extra[idx]);
3565 ++idx;
3566 }
3567 }
3568 # endif
3569 had_char_class = false;
3570 }
3571 else
3572 {
3573 c1++;
3574 while (c1--)
3575 PATUNFETCH;
3576 SET_LIST_BIT ('[');
3577 SET_LIST_BIT ('.');
3578 range_start = '.';
3579 had_char_class = false;
3580 }
3581 }
3582 else
3583 {
3584 had_char_class = false;
3585 SET_LIST_BIT (c);
3586 range_start = c;
3587 }
3588 }
3589
3590 /* Discard any (non)matching list bytes that are all 0 at the
3591 end of the map. Decrease the map-length byte too. */
3592 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
3593 b[-1]--;
3594 b += b[-1];
3595 #endif /* MBS_SUPPORT */
3596 }
3597 break;
3598
3599
3600 case '(':
3601 if (syntax & RE_NO_BK_PARENS)
3602 goto handle_open;
3603 else
3604 goto normal_char;
3605
3606
3607 case ')':
3608 if (syntax & RE_NO_BK_PARENS)
3609 goto handle_close;
3610 else
3611 goto normal_char;
3612
3613
3614 case '\n':
3615 if (syntax & RE_NEWLINE_ALT)
3616 goto handle_alt;
3617 else
3618 goto normal_char;
3619
3620
3621 case '|':
3622 if (syntax & RE_NO_BK_VBAR)
3623 goto handle_alt;
3624 else
3625 goto normal_char;
3626
3627
3628 case '{':
3629 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
3630 goto handle_interval;
3631 else
3632 goto normal_char;
3633
3634
3635 case '\\':
3636 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
3637
3638 /* Do not translate the character after the \, so that we can
3639 distinguish, e.g., \B from \b, even if we normally would
3640 translate, e.g., B to b. */
3641 PATFETCH_RAW (c);
3642
3643 switch (c)
3644 {
3645 case '(':
3646 if (syntax & RE_NO_BK_PARENS)
3647 goto normal_backslash;
3648
3649 handle_open:
3650 bufp->re_nsub++;
3651 regnum++;
3652
3653 if (COMPILE_STACK_FULL)
3654 {
3655 RETALLOC (compile_stack.stack, compile_stack.size << 1,
3656 compile_stack_elt_t);
3657 if (compile_stack.stack == NULL) return REG_ESPACE;
3658
3659 compile_stack.size <<= 1;
3660 }
3661
3662 /* These are the values to restore when we hit end of this
3663 group. They are all relative offsets, so that if the
3664 whole pattern moves because of realloc, they will still
3665 be valid. */
3666 COMPILE_STACK_TOP.begalt_offset = begalt - COMPILED_BUFFER_VAR;
3667 COMPILE_STACK_TOP.fixup_alt_jump
3668 = fixup_alt_jump ? fixup_alt_jump - COMPILED_BUFFER_VAR + 1 : 0;
3669 COMPILE_STACK_TOP.laststart_offset = b - COMPILED_BUFFER_VAR;
3670 COMPILE_STACK_TOP.regnum = regnum;
3671
3672 /* We will eventually replace the 0 with the number of
3673 groups inner to this one. But do not push a
3674 start_memory for groups beyond the last one we can
3675 represent in the compiled pattern. */
3676 if (regnum <= MAX_REGNUM)
3677 {
3678 COMPILE_STACK_TOP.inner_group_offset = b
3679 - COMPILED_BUFFER_VAR + 2;
3680 BUF_PUSH_3 (start_memory, regnum, 0);
3681 }
3682
3683 compile_stack.avail++;
3684
3685 fixup_alt_jump = 0;
3686 laststart = 0;
3687 begalt = b;
3688 /* If we've reached MAX_REGNUM groups, then this open
3689 won't actually generate any code, so we'll have to
3690 clear pending_exact explicitly. */
3691 pending_exact = 0;
3692 break;
3693
3694
3695 case ')':
3696 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
3697
3698 if (COMPILE_STACK_EMPTY)
3699 {
3700 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3701 goto normal_backslash;
3702 else
3703 FREE_STACK_RETURN (REG_ERPAREN);
3704 }
3705
3706 handle_close:
3707 if (fixup_alt_jump)
3708 { /* Push a dummy failure point at the end of the
3709 alternative for a possible future
3710 `pop_failure_jump' to pop. See comments at
3711 `push_dummy_failure' in `re_match_2'. */
3712 BUF_PUSH (push_dummy_failure);
3713
3714 /* We allocated space for this jump when we assigned
3715 to `fixup_alt_jump', in the `handle_alt' case below. */
3716 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
3717 }
3718
3719 /* See similar code for backslashed left paren above. */
3720 if (COMPILE_STACK_EMPTY)
3721 {
3722 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
3723 goto normal_char;
3724 else
3725 FREE_STACK_RETURN (REG_ERPAREN);
3726 }
3727
3728 /* Since we just checked for an empty stack above, this
3729 ``can't happen''. */
3730 assert (compile_stack.avail != 0);
3731 {
3732 /* We don't just want to restore into `regnum', because
3733 later groups should continue to be numbered higher,
3734 as in `(ab)c(de)' -- the second group is #2. */
3735 regnum_t this_group_regnum;
3736
3737 compile_stack.avail--;
3738 begalt = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.begalt_offset;
3739 fixup_alt_jump
3740 = COMPILE_STACK_TOP.fixup_alt_jump
3741 ? COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.fixup_alt_jump - 1
3742 : 0;
3743 laststart = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.laststart_offset;
3744 this_group_regnum = COMPILE_STACK_TOP.regnum;
3745 /* If we've reached MAX_REGNUM groups, then this open
3746 won't actually generate any code, so we'll have to
3747 clear pending_exact explicitly. */
3748 pending_exact = 0;
3749
3750 /* We're at the end of the group, so now we know how many
3751 groups were inside this one. */
3752 if (this_group_regnum <= MAX_REGNUM)
3753 {
3754 US_CHAR_TYPE *inner_group_loc
3755 = COMPILED_BUFFER_VAR + COMPILE_STACK_TOP.inner_group_offset;
3756
3757 *inner_group_loc = regnum - this_group_regnum;
3758 BUF_PUSH_3 (stop_memory, this_group_regnum,
3759 regnum - this_group_regnum);
3760 }
3761 }
3762 break;
3763
3764
3765 case '|': /* `\|'. */
3766 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
3767 goto normal_backslash;
3768 handle_alt:
3769 if (syntax & RE_LIMITED_OPS)
3770 goto normal_char;
3771
3772 /* Insert before the previous alternative a jump which
3773 jumps to this alternative if the former fails. */
3774 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3775 INSERT_JUMP (on_failure_jump, begalt,
3776 b + 2 + 2 * OFFSET_ADDRESS_SIZE);
3777 pending_exact = 0;
3778 b += 1 + OFFSET_ADDRESS_SIZE;
3779
3780 /* The alternative before this one has a jump after it
3781 which gets executed if it gets matched. Adjust that
3782 jump so it will jump to this alternative's analogous
3783 jump (put in below, which in turn will jump to the next
3784 (if any) alternative's such jump, etc.). The last such
3785 jump jumps to the correct final destination. A picture:
3786 _____ _____
3787 | | | |
3788 | v | v
3789 a | b | c
3790
3791 If we are at `b', then fixup_alt_jump right now points to a
3792 three-byte space after `a'. We'll put in the jump, set
3793 fixup_alt_jump to right after `b', and leave behind three
3794 bytes which we'll fill in when we get to after `c'. */
3795
3796 if (fixup_alt_jump)
3797 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
3798
3799 /* Mark and leave space for a jump after this alternative,
3800 to be filled in later either by next alternative or
3801 when know we're at the end of a series of alternatives. */
3802 fixup_alt_jump = b;
3803 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3804 b += 1 + OFFSET_ADDRESS_SIZE;
3805
3806 laststart = 0;
3807 begalt = b;
3808 break;
3809
3810
3811 case '{':
3812 /* If \{ is a literal. */
3813 if (!(syntax & RE_INTERVALS)
3814 /* If we're at `\{' and it's not the open-interval
3815 operator. */
3816 || (syntax & RE_NO_BK_BRACES))
3817 goto normal_backslash;
3818
3819 handle_interval:
3820 {
3821 /* If got here, then the syntax allows intervals. */
3822
3823 /* At least (most) this many matches must be made. */
3824 int lower_bound = -1, upper_bound = -1;
3825
3826 /* Place in the uncompiled pattern (i.e., just after
3827 the '{') to go back to if the interval is invalid. */
3828 const CHAR_TYPE *beg_interval = p;
3829
3830 if (p == pend)
3831 goto invalid_interval;
3832
3833 GET_UNSIGNED_NUMBER (lower_bound);
3834
3835 if (c == ',')
3836 {
3837 GET_UNSIGNED_NUMBER (upper_bound);
3838 if (upper_bound < 0)
3839 upper_bound = RE_DUP_MAX;
3840 }
3841 else
3842 /* Interval such as `{1}' => match exactly once. */
3843 upper_bound = lower_bound;
3844
3845 if (! (0 <= lower_bound && lower_bound <= upper_bound))
3846 goto invalid_interval;
3847
3848 if (!(syntax & RE_NO_BK_BRACES))
3849 {
3850 if (c != '\\' || p == pend)
3851 goto invalid_interval;
3852 PATFETCH (c);
3853 }
3854
3855 if (c != '}')
3856 goto invalid_interval;
3857
3858 /* If it's invalid to have no preceding re. */
3859 if (!laststart)
3860 {
3861 if (syntax & RE_CONTEXT_INVALID_OPS
3862 && !(syntax & RE_INVALID_INTERVAL_ORD))
3863 FREE_STACK_RETURN (REG_BADRPT);
3864 else if (syntax & RE_CONTEXT_INDEP_OPS)
3865 laststart = b;
3866 else
3867 goto unfetch_interval;
3868 }
3869
3870 /* We just parsed a valid interval. */
3871
3872 if (RE_DUP_MAX < upper_bound)
3873 FREE_STACK_RETURN (REG_BADBR);
3874
3875 /* If the upper bound is zero, don't want to succeed at
3876 all; jump from `laststart' to `b + 3', which will be
3877 the end of the buffer after we insert the jump. */
3878 /* ifdef MBS_SUPPORT, 'b + 1 + OFFSET_ADDRESS_SIZE'
3879 instead of 'b + 3'. */
3880 if (upper_bound == 0)
3881 {
3882 GET_BUFFER_SPACE (1 + OFFSET_ADDRESS_SIZE);
3883 INSERT_JUMP (jump, laststart, b + 1
3884 + OFFSET_ADDRESS_SIZE);
3885 b += 1 + OFFSET_ADDRESS_SIZE;
3886 }
3887
3888 /* Otherwise, we have a nontrivial interval. When
3889 we're all done, the pattern will look like:
3890 set_number_at <jump count> <upper bound>
3891 set_number_at <succeed_n count> <lower bound>
3892 succeed_n <after jump addr> <succeed_n count>
3893 <body of loop>
3894 jump_n <succeed_n addr> <jump count>
3895 (The upper bound and `jump_n' are omitted if
3896 `upper_bound' is 1, though.) */
3897 else
3898 { /* If the upper bound is > 1, we need to insert
3899 more at the end of the loop. */
3900 unsigned nbytes = 2 + 4 * OFFSET_ADDRESS_SIZE +
3901 (upper_bound > 1) * (2 + 4 * OFFSET_ADDRESS_SIZE);
3902
3903 GET_BUFFER_SPACE (nbytes);
3904
3905 /* Initialize lower bound of the `succeed_n', even
3906 though it will be set during matching by its
3907 attendant `set_number_at' (inserted next),
3908 because `re_compile_fastmap' needs to know.
3909 Jump to the `jump_n' we might insert below. */
3910 INSERT_JUMP2 (succeed_n, laststart,
3911 b + 1 + 2 * OFFSET_ADDRESS_SIZE
3912 + (upper_bound > 1) * (1 + 2 * OFFSET_ADDRESS_SIZE)
3913 , lower_bound);
3914 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3915
3916 /* Code to initialize the lower bound. Insert
3917 before the `succeed_n'. The `5' is the last two
3918 bytes of this `set_number_at', plus 3 bytes of
3919 the following `succeed_n'. */
3920 /* ifdef MBS_SUPPORT, The '1+2*OFFSET_ADDRESS_SIZE'
3921 is the 'set_number_at', plus '1+OFFSET_ADDRESS_SIZE'
3922 of the following `succeed_n'. */
3923 insert_op2 (set_number_at, laststart, 1
3924 + 2 * OFFSET_ADDRESS_SIZE, lower_bound, b);
3925 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3926
3927 if (upper_bound > 1)
3928 { /* More than one repetition is allowed, so
3929 append a backward jump to the `succeed_n'
3930 that starts this interval.
3931
3932 When we've reached this during matching,
3933 we'll have matched the interval once, so
3934 jump back only `upper_bound - 1' times. */
3935 STORE_JUMP2 (jump_n, b, laststart
3936 + 2 * OFFSET_ADDRESS_SIZE + 1,
3937 upper_bound - 1);
3938 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3939
3940 /* The location we want to set is the second
3941 parameter of the `jump_n'; that is `b-2' as
3942 an absolute address. `laststart' will be
3943 the `set_number_at' we're about to insert;
3944 `laststart+3' the number to set, the source
3945 for the relative address. But we are
3946 inserting into the middle of the pattern --
3947 so everything is getting moved up by 5.
3948 Conclusion: (b - 2) - (laststart + 3) + 5,
3949 i.e., b - laststart.
3950
3951 We insert this at the beginning of the loop
3952 so that if we fail during matching, we'll
3953 reinitialize the bounds. */
3954 insert_op2 (set_number_at, laststart, b - laststart,
3955 upper_bound - 1, b);
3956 b += 1 + 2 * OFFSET_ADDRESS_SIZE;
3957 }
3958 }
3959 pending_exact = 0;
3960 break;
3961
3962 invalid_interval:
3963 if (!(syntax & RE_INVALID_INTERVAL_ORD))
3964 FREE_STACK_RETURN (p == pend ? REG_EBRACE : REG_BADBR);
3965 unfetch_interval:
3966 /* Match the characters as literals. */
3967 p = beg_interval;
3968 c = '{';
3969 if (syntax & RE_NO_BK_BRACES)
3970 goto normal_char;
3971 else
3972 goto normal_backslash;
3973 }
3974
3975 #ifdef emacs
3976 /* There is no way to specify the before_dot and after_dot
3977 operators. rms says this is ok. --karl */
3978 case '=':
3979 BUF_PUSH (at_dot);
3980 break;
3981
3982 case 's':
3983 laststart = b;
3984 PATFETCH (c);
3985 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
3986 break;
3987
3988 case 'S':
3989 laststart = b;
3990 PATFETCH (c);
3991 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
3992 break;
3993 #endif /* emacs */
3994
3995
3996 case 'w':
3997 if (syntax & RE_NO_GNU_OPS)
3998 goto normal_char;
3999 laststart = b;
4000 BUF_PUSH (wordchar);
4001 break;
4002
4003
4004 case 'W':
4005 if (syntax & RE_NO_GNU_OPS)
4006 goto normal_char;
4007 laststart = b;
4008 BUF_PUSH (notwordchar);
4009 break;
4010
4011
4012 case '<':
4013 if (syntax & RE_NO_GNU_OPS)
4014 goto normal_char;
4015 BUF_PUSH (wordbeg);
4016 break;
4017
4018 case '>':
4019 if (syntax & RE_NO_GNU_OPS)
4020 goto normal_char;
4021 BUF_PUSH (wordend);
4022 break;
4023
4024 case 'b':
4025 if (syntax & RE_NO_GNU_OPS)
4026 goto normal_char;
4027 BUF_PUSH (wordbound);
4028 break;
4029
4030 case 'B':
4031 if (syntax & RE_NO_GNU_OPS)
4032 goto normal_char;
4033 BUF_PUSH (notwordbound);
4034 break;
4035
4036 case '`':
4037 if (syntax & RE_NO_GNU_OPS)
4038 goto normal_char;
4039 BUF_PUSH (begbuf);
4040 break;
4041
4042 case '\'':
4043 if (syntax & RE_NO_GNU_OPS)
4044 goto normal_char;
4045 BUF_PUSH (endbuf);
4046 break;
4047
4048 case '1': case '2': case '3': case '4': case '5':
4049 case '6': case '7': case '8': case '9':
4050 if (syntax & RE_NO_BK_REFS)
4051 goto normal_char;
4052
4053 c1 = c - '0';
4054
4055 if (c1 > regnum)
4056 FREE_STACK_RETURN (REG_ESUBREG);
4057
4058 /* Can't back reference to a subexpression if inside of it. */
4059 if (group_in_compile_stack (compile_stack, (regnum_t) c1))
4060 goto normal_char;
4061
4062 laststart = b;
4063 BUF_PUSH_2 (duplicate, c1);
4064 break;
4065
4066
4067 case '+':
4068 case '?':
4069 if (syntax & RE_BK_PLUS_QM)
4070 goto handle_plus;
4071 else
4072 goto normal_backslash;
4073
4074 default:
4075 normal_backslash:
4076 /* You might think it would be useful for \ to mean
4077 not to translate; but if we don't translate it
4078 it will never match anything. */
4079 c = TRANSLATE (c);
4080 goto normal_char;
4081 }
4082 break;
4083
4084
4085 default:
4086 /* Expects the character in `c'. */
4087 normal_char:
4088 /* If no exactn currently being built. */
4089 if (!pending_exact
4090 #ifdef MBS_SUPPORT
4091 /* If last exactn handle binary(or character) and
4092 new exactn handle character(or binary). */
4093 || is_exactn_bin != is_binary[p - 1 - pattern]
4094 #endif /* MBS_SUPPORT */
4095
4096 /* If last exactn not at current position. */
4097 || pending_exact + *pending_exact + 1 != b
4098
4099 /* We have only one byte following the exactn for the count. */
4100 || *pending_exact == (1 << BYTEWIDTH) - 1
4101
4102 /* If followed by a repetition operator. */
4103 || *p == '*' || *p == '^'
4104 || ((syntax & RE_BK_PLUS_QM)
4105 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
4106 : (*p == '+' || *p == '?'))
4107 || ((syntax & RE_INTERVALS)
4108 && ((syntax & RE_NO_BK_BRACES)
4109 ? *p == '{'
4110 : (p[0] == '\\' && p[1] == '{'))))
4111 {
4112 /* Start building a new exactn. */
4113
4114 laststart = b;
4115
4116 #ifdef MBS_SUPPORT
4117 /* Is this exactn binary data or character? */
4118 is_exactn_bin = is_binary[p - 1 - pattern];
4119 if (is_exactn_bin)
4120 BUF_PUSH_2 (exactn_bin, 0);
4121 else
4122 BUF_PUSH_2 (exactn, 0);
4123 #else
4124 BUF_PUSH_2 (exactn, 0);
4125 #endif /* MBS_SUPPORT */
4126 pending_exact = b - 1;
4127 }
4128
4129 BUF_PUSH (c);
4130 (*pending_exact)++;
4131 break;
4132 } /* switch (c) */
4133 } /* while p != pend */
4134
4135
4136 /* Through the pattern now. */
4137
4138 if (fixup_alt_jump)
4139 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
4140
4141 if (!COMPILE_STACK_EMPTY)
4142 FREE_STACK_RETURN (REG_EPAREN);
4143
4144 /* If we don't want backtracking, force success
4145 the first time we reach the end of the compiled pattern. */
4146 if (syntax & RE_NO_POSIX_BACKTRACKING)
4147 BUF_PUSH (succeed);
4148
4149 #ifdef MBS_SUPPORT
4150 free (pattern);
4151 free (mbs_offset);
4152 free (is_binary);
4153 #endif
4154 free (compile_stack.stack);
4155
4156 /* We have succeeded; set the length of the buffer. */
4157 #ifdef MBS_SUPPORT
4158 bufp->used = (uintptr_t) b - (uintptr_t) COMPILED_BUFFER_VAR;
4159 #else
4160 bufp->used = b - bufp->buffer;
4161 #endif
4162
4163 #ifdef DEBUG
4164 if (debug)
4165 {
4166 DEBUG_PRINT1 ("\nCompiled pattern: \n");
4167 print_compiled_pattern (bufp);
4168 }
4169 #endif /* DEBUG */
4170
4171 #ifndef MATCH_MAY_ALLOCATE
4172 /* Initialize the failure stack to the largest possible stack. This
4173 isn't necessary unless we're trying to avoid calling alloca in
4174 the search and match routines. */
4175 {
4176 int num_regs = bufp->re_nsub + 1;
4177
4178 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
4179 is strictly greater than re_max_failures, the largest possible stack
4180 is 2 * re_max_failures failure points. */
4181 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
4182 {
4183 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
4184
4185 # ifdef emacs
4186 if (! fail_stack.stack)
4187 fail_stack.stack
4188 = (fail_stack_elt_t *) xmalloc (fail_stack.size
4189 * sizeof (fail_stack_elt_t));
4190 else
4191 fail_stack.stack
4192 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
4193 (fail_stack.size
4194 * sizeof (fail_stack_elt_t)));
4195 # else /* not emacs */
4196 if (! fail_stack.stack)
4197 fail_stack.stack
4198 = (fail_stack_elt_t *) malloc (fail_stack.size
4199 * sizeof (fail_stack_elt_t));
4200 else
4201 fail_stack.stack
4202 = (fail_stack_elt_t *) realloc (fail_stack.stack,
4203 (fail_stack.size
4204 * sizeof (fail_stack_elt_t)));
4205 # endif /* not emacs */
4206 }
4207
4208 regex_grow_registers (num_regs);
4209 }
4210 #endif /* not MATCH_MAY_ALLOCATE */
4211
4212 return REG_NOERROR;
4213 } /* regex_compile */
4214
4215 /* Subroutines for `regex_compile'. */
4216
4217 /* Store OP at LOC followed by two-byte integer parameter ARG. */
4218 /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */
4219
4220 static void
store_op1(op,loc,arg)4221 store_op1 (op, loc, arg)
4222 re_opcode_t op;
4223 US_CHAR_TYPE *loc;
4224 int arg;
4225 {
4226 *loc = (US_CHAR_TYPE) op;
4227 STORE_NUMBER (loc + 1, arg);
4228 }
4229
4230
4231 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
4232 /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */
4233
4234 static void
store_op2(op,loc,arg1,arg2)4235 store_op2 (op, loc, arg1, arg2)
4236 re_opcode_t op;
4237 US_CHAR_TYPE *loc;
4238 int arg1, arg2;
4239 {
4240 *loc = (US_CHAR_TYPE) op;
4241 STORE_NUMBER (loc + 1, arg1);
4242 STORE_NUMBER (loc + 1 + OFFSET_ADDRESS_SIZE, arg2);
4243 }
4244
4245
4246 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
4247 for OP followed by two-byte integer parameter ARG. */
4248 /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */
4249
4250 static void
insert_op1(op,loc,arg,end)4251 insert_op1 (op, loc, arg, end)
4252 re_opcode_t op;
4253 US_CHAR_TYPE *loc;
4254 int arg;
4255 US_CHAR_TYPE *end;
4256 {
4257 register US_CHAR_TYPE *pfrom = end;
4258 register US_CHAR_TYPE *pto = end + 1 + OFFSET_ADDRESS_SIZE;
4259
4260 while (pfrom != loc)
4261 *--pto = *--pfrom;
4262
4263 store_op1 (op, loc, arg);
4264 }
4265
4266
4267 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
4268 /* ifdef MBS_SUPPORT, integer parameter is 1 wchar_t. */
4269
4270 static void
insert_op2(op,loc,arg1,arg2,end)4271 insert_op2 (op, loc, arg1, arg2, end)
4272 re_opcode_t op;
4273 US_CHAR_TYPE *loc;
4274 int arg1, arg2;
4275 US_CHAR_TYPE *end;
4276 {
4277 register US_CHAR_TYPE *pfrom = end;
4278 register US_CHAR_TYPE *pto = end + 1 + 2 * OFFSET_ADDRESS_SIZE;
4279
4280 while (pfrom != loc)
4281 *--pto = *--pfrom;
4282
4283 store_op2 (op, loc, arg1, arg2);
4284 }
4285
4286
4287 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
4288 after an alternative or a begin-subexpression. We assume there is at
4289 least one character before the ^. */
4290
4291 static boolean
at_begline_loc_p(pattern,p,syntax)4292 at_begline_loc_p (pattern, p, syntax)
4293 const CHAR_TYPE *pattern, *p;
4294 reg_syntax_t syntax;
4295 {
4296 const CHAR_TYPE *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
at_endline_loc_p(p,pend,syntax)4311 at_endline_loc_p (p, pend, syntax)
4312 const CHAR_TYPE *p, *pend;
4313 reg_syntax_t syntax;
4314 {
4315 const CHAR_TYPE *next = p;
4316 boolean next_backslash = *next == '\\';
4317 const CHAR_TYPE *next_next = p + 1 < pend ? p + 1 : 0;
4318
4319 return
4320 /* Before a subexpression? */
4321 (syntax & RE_NO_BK_PARENS ? *next == ')'
4322 : next_backslash && next_next && *next_next == ')')
4323 /* Before an alternative? */
4324 || (syntax & RE_NO_BK_VBAR ? *next == '|'
4325 : next_backslash && next_next && *next_next == '|');
4326 }
4327
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
group_in_compile_stack(compile_stack,regnum)4333 group_in_compile_stack (compile_stack, regnum)
4334 compile_stack_type compile_stack;
4335 regnum_t regnum;
4336 {
4337 int this_element;
4338
4339 for (this_element = compile_stack.avail - 1;
4340 this_element >= 0;
4341 this_element--)
4342 if (compile_stack.stack[this_element].regnum == regnum)
4343 return true;
4344
4345 return false;
4346 }
4347
4348 #ifdef MBS_SUPPORT
4349 /* This insert space, which size is "num", into the pattern at "loc".
4350 "end" must point the end of the allocated buffer. */
4351 static void
insert_space(num,loc,end)4352 insert_space (num, loc, end)
4353 int num;
4354 CHAR_TYPE *loc;
4355 CHAR_TYPE *end;
4356 {
4357 register CHAR_TYPE *pto = end;
4358 register CHAR_TYPE *pfrom = end - num;
4359
4360 while (pfrom >= loc)
4361 *pto-- = *pfrom--;
4362 }
4363 #endif /* MBS_SUPPORT */
4364
4365 #ifdef MBS_SUPPORT
4366 static reg_errcode_t
compile_range(range_start_char,p_ptr,pend,translate,syntax,b,char_set)4367 compile_range (range_start_char, p_ptr, pend, translate, syntax, b,
4368 char_set)
4369 CHAR_TYPE range_start_char;
4370 const CHAR_TYPE **p_ptr, *pend;
4371 CHAR_TYPE *char_set, *b;
4372 RE_TRANSLATE_TYPE translate;
4373 reg_syntax_t syntax;
4374 {
4375 const CHAR_TYPE *p = *p_ptr;
4376 CHAR_TYPE range_start, range_end;
4377 reg_errcode_t ret;
4378 # ifdef _LIBC
4379 uint32_t nrules;
4380 uint32_t start_val, end_val;
4381 # endif
4382 if (p == pend)
4383 return REG_ERANGE;
4384
4385 # ifdef _LIBC
4386 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
4387 if (nrules != 0)
4388 {
4389 const char *collseq = (const char *) _NL_CURRENT(LC_COLLATE,
4390 _NL_COLLATE_COLLSEQWC);
4391 const unsigned char *extra = (const unsigned char *)
4392 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
4393
4394 if (range_start_char < -1)
4395 {
4396 /* range_start is a collating symbol. */
4397 int32_t *wextra;
4398 /* Retreive the index and get collation sequence value. */
4399 wextra = (int32_t*)(extra + char_set[-range_start_char]);
4400 start_val = wextra[1 + *wextra];
4401 }
4402 else
4403 start_val = collseq_table_lookup(collseq, TRANSLATE(range_start_char));
4404
4405 end_val = collseq_table_lookup (collseq, TRANSLATE (p[0]));
4406
4407 /* Report an error if the range is empty and the syntax prohibits
4408 this. */
4409 ret = ((syntax & RE_NO_EMPTY_RANGES)
4410 && (start_val > end_val))? REG_ERANGE : REG_NOERROR;
4411
4412 /* Insert space to the end of the char_ranges. */
4413 insert_space(2, b - char_set[5] - 2, b - 1);
4414 *(b - char_set[5] - 2) = (wchar_t)start_val;
4415 *(b - char_set[5] - 1) = (wchar_t)end_val;
4416 char_set[4]++; /* ranges_index */
4417 }
4418 else
4419 # endif
4420 {
4421 range_start = (range_start_char >= 0)? TRANSLATE (range_start_char):
4422 range_start_char;
4423 range_end = TRANSLATE (p[0]);
4424 /* Report an error if the range is empty and the syntax prohibits
4425 this. */
4426 ret = ((syntax & RE_NO_EMPTY_RANGES)
4427 && (range_start > range_end))? REG_ERANGE : REG_NOERROR;
4428
4429 /* Insert space to the end of the char_ranges. */
4430 insert_space(2, b - char_set[5] - 2, b - 1);
4431 *(b - char_set[5] - 2) = range_start;
4432 *(b - char_set[5] - 1) = range_end;
4433 char_set[4]++; /* ranges_index */
4434 }
4435 /* Have to increment the pointer into the pattern string, so the
4436 caller isn't still at the ending character. */
4437 (*p_ptr)++;
4438
4439 return ret;
4440 }
4441 #else
4442 /* Read the ending character of a range (in a bracket expression) from the
4443 uncompiled pattern *P_PTR (which ends at PEND). We assume the
4444 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
4445 Then we set the translation of all bits between the starting and
4446 ending characters (inclusive) in the compiled pattern B.
4447
4448 Return an error code.
4449
4450 We use these short variable names so we can use the same macros as
4451 `regex_compile' itself. */
4452
4453 static reg_errcode_t
compile_range(range_start_char,p_ptr,pend,translate,syntax,b)4454 compile_range (range_start_char, p_ptr, pend, translate, syntax, b)
4455 unsigned int range_start_char;
4456 const char **p_ptr, *pend;
4457 RE_TRANSLATE_TYPE translate;
4458 reg_syntax_t syntax;
4459 unsigned char *b;
4460 {
4461 unsigned this_char;
4462 const char *p = *p_ptr;
4463 reg_errcode_t ret;
4464 # if _LIBC
4465 const unsigned char *collseq;
4466 unsigned int start_colseq;
4467 unsigned int end_colseq;
4468 # else
4469 unsigned end_char;
4470 # endif
4471
4472 if (p == pend)
4473 return REG_ERANGE;
4474
4475 /* Have to increment the pointer into the pattern string, so the
4476 caller isn't still at the ending character. */
4477 (*p_ptr)++;
4478
4479 /* Report an error if the range is empty and the syntax prohibits this. */
4480 ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
4481
4482 # if _LIBC
4483 collseq = (const unsigned char *) _NL_CURRENT (LC_COLLATE,
4484 _NL_COLLATE_COLLSEQMB);
4485
4486 start_colseq = collseq[(unsigned char) TRANSLATE (range_start_char)];
4487 end_colseq = collseq[(unsigned char) TRANSLATE (p[0])];
4488 for (this_char = 0; this_char <= (unsigned char) -1; ++this_char)
4489 {
4490 unsigned int this_colseq = collseq[(unsigned char) TRANSLATE (this_char)];
4491
4492 if (start_colseq <= this_colseq && this_colseq <= end_colseq)
4493 {
4494 SET_LIST_BIT (TRANSLATE (this_char));
4495 ret = REG_NOERROR;
4496 }
4497 }
4498 # else
4499 /* Here we see why `this_char' has to be larger than an `unsigned
4500 char' -- we would otherwise go into an infinite loop, since all
4501 characters <= 0xff. */
4502 range_start_char = TRANSLATE (range_start_char);
4503 /* TRANSLATE(p[0]) is casted to char (not unsigned char) in TRANSLATE,
4504 and some compilers cast it to int implicitly, so following for_loop
4505 may fall to (almost) infinite loop.
4506 e.g. If translate[p[0]] = 0xff, end_char may equals to 0xffffffff.
4507 To avoid this, we cast p[0] to unsigned int and truncate it. */
4508 end_char = ((unsigned)TRANSLATE(p[0]) & ((1 << BYTEWIDTH) - 1));
4509
4510 for (this_char = range_start_char; this_char <= end_char; ++this_char)
4511 {
4512 SET_LIST_BIT (TRANSLATE (this_char));
4513 ret = REG_NOERROR;
4514 }
4515 # endif
4516
4517 return ret;
4518 }
4519 #endif /* MBS_SUPPORT */
4520
4521 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
4522 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
4523 characters can start a string that matches the pattern. This fastmap
4524 is used by re_search to skip quickly over impossible starting points.
4525
4526 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
4527 area as BUFP->fastmap.
4528
4529 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
4530 the pattern buffer.
4531
4532 Returns 0 if we succeed, -2 if an internal error. */
4533
4534 #ifdef MBS_SUPPORT
4535 /* local function for re_compile_fastmap.
4536 truncate wchar_t character to char. */
4537 static unsigned char truncate_wchar (CHAR_TYPE c);
4538
4539 static unsigned char
truncate_wchar(c)4540 truncate_wchar (c)
4541 CHAR_TYPE c;
4542 {
4543 unsigned char buf[MB_LEN_MAX];
4544 int retval = wctomb(buf, c);
4545 return retval > 0 ? buf[0] : (unsigned char)c;
4546 }
4547 #endif /* MBS_SUPPORT */
4548
4549 int
re_compile_fastmap(bufp)4550 re_compile_fastmap (bufp)
4551 struct re_pattern_buffer *bufp;
4552 {
4553 int j, k;
4554 #ifdef MATCH_MAY_ALLOCATE
4555 fail_stack_type fail_stack;
4556 #endif
4557 #ifndef REGEX_MALLOC
4558 char *destination;
4559 #endif
4560
4561 register char *fastmap = bufp->fastmap;
4562
4563 #ifdef MBS_SUPPORT
4564 /* We need to cast pattern to (wchar_t*), because we casted this compiled
4565 pattern to (char*) in regex_compile. */
4566 US_CHAR_TYPE *pattern = (US_CHAR_TYPE*)bufp->buffer;
4567 register US_CHAR_TYPE *pend = (US_CHAR_TYPE*) (bufp->buffer + bufp->used);
4568 #else
4569 US_CHAR_TYPE *pattern = bufp->buffer;
4570 register US_CHAR_TYPE *pend = pattern + bufp->used;
4571 #endif /* MBS_SUPPORT */
4572 US_CHAR_TYPE *p = pattern;
4573
4574 #ifdef REL_ALLOC
4575 /* This holds the pointer to the failure stack, when
4576 it is allocated relocatably. */
4577 fail_stack_elt_t *failure_stack_ptr;
4578 #endif
4579
4580 /* Assume that each path through the pattern can be null until
4581 proven otherwise. We set this false at the bottom of switch
4582 statement, to which we get only if a particular path doesn't
4583 match the empty string. */
4584 boolean path_can_be_null = true;
4585
4586 /* We aren't doing a `succeed_n' to begin with. */
4587 boolean succeed_n_p = false;
4588
4589 assert (fastmap != NULL && p != NULL);
4590
4591 INIT_FAIL_STACK ();
4592 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
4593 bufp->fastmap_accurate = 1; /* It will be when we're done. */
4594 bufp->can_be_null = 0;
4595
4596 while (1)
4597 {
4598 if (p == pend || *p == succeed)
4599 {
4600 /* We have reached the (effective) end of pattern. */
4601 if (!FAIL_STACK_EMPTY ())
4602 {
4603 bufp->can_be_null |= path_can_be_null;
4604
4605 /* Reset for next path. */
4606 path_can_be_null = true;
4607
4608 p = fail_stack.stack[--fail_stack.avail].pointer;
4609
4610 continue;
4611 }
4612 else
4613 break;
4614 }
4615
4616 /* We should never be about to go beyond the end of the pattern. */
4617 assert (p < pend);
4618
4619 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
4620 {
4621
4622 /* I guess the idea here is to simply not bother with a fastmap
4623 if a backreference is used, since it's too hard to figure out
4624 the fastmap for the corresponding group. Setting
4625 `can_be_null' stops `re_search_2' from using the fastmap, so
4626 that is all we do. */
4627 case duplicate:
4628 bufp->can_be_null = 1;
4629 goto done;
4630
4631
4632 /* Following are the cases which match a character. These end
4633 with `break'. */
4634
4635 #ifdef MBS_SUPPORT
4636 case exactn:
4637 fastmap[truncate_wchar(p[1])] = 1;
4638 break;
4639 case exactn_bin:
4640 fastmap[p[1]] = 1;
4641 break;
4642 #else
4643 case exactn:
4644 fastmap[p[1]] = 1;
4645 break;
4646 #endif /* MBS_SUPPORT */
4647
4648
4649 #ifdef MBS_SUPPORT
4650 /* It is hard to distinguish fastmap from (multi byte) characters
4651 which depends on current locale. */
4652 case charset:
4653 case charset_not:
4654 case wordchar:
4655 case notwordchar:
4656 bufp->can_be_null = 1;
4657 goto done;
4658 #else
4659 case charset:
4660 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4661 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
4662 fastmap[j] = 1;
4663 break;
4664
4665
4666 case charset_not:
4667 /* Chars beyond end of map must be allowed. */
4668 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
4669 fastmap[j] = 1;
4670
4671 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
4672 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
4673 fastmap[j] = 1;
4674 break;
4675
4676
4677 case wordchar:
4678 for (j = 0; j < (1 << BYTEWIDTH); j++)
4679 if (SYNTAX (j) == Sword)
4680 fastmap[j] = 1;
4681 break;
4682
4683
4684 case notwordchar:
4685 for (j = 0; j < (1 << BYTEWIDTH); j++)
4686 if (SYNTAX (j) != Sword)
4687 fastmap[j] = 1;
4688 break;
4689 #endif
4690
4691 case anychar:
4692 {
4693 int fastmap_newline = fastmap['\n'];
4694
4695 /* `.' matches anything ... */
4696 for (j = 0; j < (1 << BYTEWIDTH); j++)
4697 fastmap[j] = 1;
4698
4699 /* ... except perhaps newline. */
4700 if (!(bufp->syntax & RE_DOT_NEWLINE))
4701 fastmap['\n'] = fastmap_newline;
4702
4703 /* Return if we have already set `can_be_null'; if we have,
4704 then the fastmap is irrelevant. Something's wrong here. */
4705 else if (bufp->can_be_null)
4706 goto done;
4707
4708 /* Otherwise, have to check alternative paths. */
4709 break;
4710 }
4711
4712 #ifdef emacs
4713 case syntaxspec:
4714 k = *p++;
4715 for (j = 0; j < (1 << BYTEWIDTH); j++)
4716 if (SYNTAX (j) == (enum syntaxcode) k)
4717 fastmap[j] = 1;
4718 break;
4719
4720
4721 case notsyntaxspec:
4722 k = *p++;
4723 for (j = 0; j < (1 << BYTEWIDTH); j++)
4724 if (SYNTAX (j) != (enum syntaxcode) k)
4725 fastmap[j] = 1;
4726 break;
4727
4728
4729 /* All cases after this match the empty string. These end with
4730 `continue'. */
4731
4732
4733 case before_dot:
4734 case at_dot:
4735 case after_dot:
4736 continue;
4737 #endif /* emacs */
4738
4739
4740 case no_op:
4741 case begline:
4742 case endline:
4743 case begbuf:
4744 case endbuf:
4745 case wordbound:
4746 case notwordbound:
4747 case wordbeg:
4748 case wordend:
4749 case push_dummy_failure:
4750 continue;
4751
4752
4753 case jump_n:
4754 case pop_failure_jump:
4755 case maybe_pop_jump:
4756 case jump:
4757 case jump_past_alt:
4758 case dummy_failure_jump:
4759 EXTRACT_NUMBER_AND_INCR (j, p);
4760 p += j;
4761 if (j > 0)
4762 continue;
4763
4764 /* Jump backward implies we just went through the body of a
4765 loop and matched nothing. Opcode jumped to should be
4766 `on_failure_jump' or `succeed_n'. Just treat it like an
4767 ordinary jump. For a * loop, it has pushed its failure
4768 point already; if so, discard that as redundant. */
4769 if ((re_opcode_t) *p != on_failure_jump
4770 && (re_opcode_t) *p != succeed_n)
4771 continue;
4772
4773 p++;
4774 EXTRACT_NUMBER_AND_INCR (j, p);
4775 p += j;
4776
4777 /* If what's on the stack is where we are now, pop it. */
4778 if (!FAIL_STACK_EMPTY ()
4779 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
4780 fail_stack.avail--;
4781
4782 continue;
4783
4784
4785 case on_failure_jump:
4786 case on_failure_keep_string_jump:
4787 handle_on_failure_jump:
4788 EXTRACT_NUMBER_AND_INCR (j, p);
4789
4790 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
4791 end of the pattern. We don't want to push such a point,
4792 since when we restore it above, entering the switch will
4793 increment `p' past the end of the pattern. We don't need
4794 to push such a point since we obviously won't find any more
4795 fastmap entries beyond `pend'. Such a pattern can match
4796 the null string, though. */
4797 if (p + j < pend)
4798 {
4799 if (!PUSH_PATTERN_OP (p + j, fail_stack))
4800 {
4801 RESET_FAIL_STACK ();
4802 return -2;
4803 }
4804 }
4805 else
4806 bufp->can_be_null = 1;
4807
4808 if (succeed_n_p)
4809 {
4810 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
4811 succeed_n_p = false;
4812 }
4813
4814 continue;
4815
4816
4817 case succeed_n:
4818 /* Get to the number of times to succeed. */
4819 p += OFFSET_ADDRESS_SIZE;
4820
4821 /* Increment p past the n for when k != 0. */
4822 EXTRACT_NUMBER_AND_INCR (k, p);
4823 if (k == 0)
4824 {
4825 p -= 2 * OFFSET_ADDRESS_SIZE;
4826 succeed_n_p = true; /* Spaghetti code alert. */
4827 goto handle_on_failure_jump;
4828 }
4829 continue;
4830
4831
4832 case set_number_at:
4833 p += 2 * OFFSET_ADDRESS_SIZE;
4834 continue;
4835
4836
4837 case start_memory:
4838 case stop_memory:
4839 p += 2;
4840 continue;
4841
4842
4843 default:
4844 abort (); /* We have listed all the cases. */
4845 } /* switch *p++ */
4846
4847 /* Getting here means we have found the possible starting
4848 characters for one path of the pattern -- and that the empty
4849 string does not match. We need not follow this path further.
4850 Instead, look at the next alternative (remembered on the
4851 stack), or quit if no more. The test at the top of the loop
4852 does these things. */
4853 path_can_be_null = false;
4854 p = pend;
4855 } /* while p */
4856
4857 /* Set `can_be_null' for the last path (also the first path, if the
4858 pattern is empty). */
4859 bufp->can_be_null |= path_can_be_null;
4860
4861 done:
4862 RESET_FAIL_STACK ();
4863 return 0;
4864 } /* re_compile_fastmap */
4865 #ifdef _LIBC
4866 weak_alias (__re_compile_fastmap, re_compile_fastmap)
4867 #endif
4868
4869 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
4870 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
4871 this memory for recording register information. STARTS and ENDS
4872 must be allocated using the malloc library routine, and must each
4873 be at least NUM_REGS * sizeof (regoff_t) bytes long.
4874
4875 If NUM_REGS == 0, then subsequent matches should allocate their own
4876 register data.
4877
4878 Unless this function is called, the first search or match using
4879 PATTERN_BUFFER will allocate its own register data, without
4880 freeing the old data. */
4881
4882 void
4883 re_set_registers (bufp, regs, num_regs, starts, ends)
4884 struct re_pattern_buffer *bufp;
4885 struct re_registers *regs;
4886 unsigned num_regs;
4887 regoff_t *starts, *ends;
4888 {
4889 if (num_regs)
4890 {
4891 bufp->regs_allocated = REGS_REALLOCATE;
4892 regs->num_regs = num_regs;
4893 regs->start = starts;
4894 regs->end = ends;
4895 }
4896 else
4897 {
4898 bufp->regs_allocated = REGS_UNALLOCATED;
4899 regs->num_regs = 0;
4900 regs->start = regs->end = (regoff_t *) 0;
4901 }
4902 }
4903 #ifdef _LIBC
4904 weak_alias (__re_set_registers, re_set_registers)
4905 #endif
4906
4907 /* Searching routines. */
4908
4909 /* Like re_search_2, below, but only one string is specified, and
4910 doesn't let you say where to stop matching. */
4911
4912 int
4913 re_search (bufp, string, size, startpos, range, regs)
4914 struct re_pattern_buffer *bufp;
4915 const char *string;
4916 int size, startpos, range;
4917 struct re_registers *regs;
4918 {
4919 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
4920 regs, size);
4921 }
4922 #ifdef _LIBC
4923 weak_alias (__re_search, re_search)
4924 #endif
4925
4926
4927 /* Using the compiled pattern in BUFP->buffer, first tries to match the
4928 virtual concatenation of STRING1 and STRING2, starting first at index
4929 STARTPOS, then at STARTPOS + 1, and so on.
4930
4931 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
4932
4933 RANGE is how far to scan while trying to match. RANGE = 0 means try
4934 only at STARTPOS; in general, the last start tried is STARTPOS +
4935 RANGE.
4936
4937 In REGS, return the indices of the virtual concatenation of STRING1
4938 and STRING2 that matched the entire BUFP->buffer and its contained
4939 subexpressions.
4940
4941 Do not consider matching one past the index STOP in the virtual
4942 concatenation of STRING1 and STRING2.
4943
4944 We return either the position in the strings at which the match was
4945 found, -1 if no match, or -2 if error (such as failure
4946 stack overflow). */
4947
4948 int
4949 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
4950 struct re_pattern_buffer *bufp;
4951 const char *string1, *string2;
4952 int size1, size2;
4953 int startpos;
4954 int range;
4955 struct re_registers *regs;
4956 int stop;
4957 {
4958 int val;
4959 register char *fastmap = bufp->fastmap;
4960 register RE_TRANSLATE_TYPE translate = bufp->translate;
4961 int total_size = size1 + size2;
4962 int endpos = startpos + range;
4963
4964 /* Check for out-of-range STARTPOS. */
4965 if (startpos < 0 || startpos > total_size)
4966 return -1;
4967
4968 /* Fix up RANGE if it might eventually take us outside
4969 the virtual concatenation of STRING1 and STRING2.
4970 Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */
4971 if (endpos < 0)
4972 range = 0 - startpos;
4973 else if (endpos > total_size)
4974 range = total_size - startpos;
4975
4976 /* If the search isn't to be a backwards one, don't waste time in a
4977 search for a pattern that must be anchored. */
4978 if (bufp->used > 0 && range > 0
4979 && ((re_opcode_t) bufp->buffer[0] == begbuf
4980 /* `begline' is like `begbuf' if it cannot match at newlines. */
4981 || ((re_opcode_t) bufp->buffer[0] == begline
4982 && !bufp->newline_anchor)))
4983 {
4984 if (startpos > 0)
4985 return -1;
4986 else
4987 range = 1;
4988 }
4989
4990 #ifdef emacs
4991 /* In a forward search for something that starts with \=.
4992 don't keep searching past point. */
4993 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot && range > 0)
4994 {
4995 range = PT - startpos;
4996 if (range <= 0)
4997 return -1;
4998 }
4999 #endif /* emacs */
5000
5001 /* Update the fastmap now if not correct already. */
5002 if (fastmap && !bufp->fastmap_accurate)
5003 if (re_compile_fastmap (bufp) == -2)
5004 return -2;
5005
5006 /* Loop through the string, looking for a place to start matching. */
5007 for (;;)
5008 {
5009 /* If a fastmap is supplied, skip quickly over characters that
5010 cannot be the start of a match. If the pattern can match the
5011 null string, however, we don't need to skip characters; we want
5012 the first null string. */
5013 if (fastmap && startpos < total_size && !bufp->can_be_null)
5014 {
5015 if (range > 0) /* Searching forwards. */
5016 {
5017 register const char *d;
5018 register int lim = 0;
5019 int irange = range;
5020
5021 if (startpos < size1 && startpos + range >= size1)
5022 lim = range - (size1 - startpos);
5023
5024 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
5025
5026 /* Written out as an if-else to avoid testing `translate'
5027 inside the loop. */
5028 if (translate)
5029 while (range > lim
5030 && !fastmap[(unsigned char)
5031 translate[(unsigned char) *d++]])
5032 range--;
5033 else
5034 while (range > lim && !fastmap[(unsigned char) *d++])
5035 range--;
5036
5037 startpos += irange - range;
5038 }
5039 else /* Searching backwards. */
5040 {
5041 register CHAR_TYPE c = (size1 == 0 || startpos >= size1
5042 ? string2[startpos - size1]
5043 : string1[startpos]);
5044
5045 if (!fastmap[(unsigned char) TRANSLATE (c)])
5046 goto advance;
5047 }
5048 }
5049
5050 /* If can't match the null string, and that's all we have left, fail. */
5051 if (range >= 0 && startpos == total_size && fastmap
5052 && !bufp->can_be_null)
5053 return -1;
5054
5055 val = re_match_2_internal (bufp, string1, size1, string2, size2,
5056 startpos, regs, stop);
5057 #ifndef REGEX_MALLOC
5058 # ifdef C_ALLOCA
5059 alloca (0);
5060 # endif
5061 #endif
5062
5063 if (val >= 0)
5064 return startpos;
5065
5066 if (val == -2)
5067 return -2;
5068
5069 advance:
5070 if (!range)
5071 break;
5072 else if (range > 0)
5073 {
5074 range--;
5075 startpos++;
5076 }
5077 else
5078 {
5079 range++;
5080 startpos--;
5081 }
5082 }
5083 return -1;
5084 } /* re_search_2 */
5085 #ifdef _LIBC
5086 weak_alias (__re_search_2, re_search_2)
5087 #endif
5088
5089 #ifdef MBS_SUPPORT
5090 /* This converts PTR, a pointer into one of the search wchar_t strings
5091 `string1' and `string2' into an multibyte string offset from the
5092 beginning of that string. We use mbs_offset to optimize.
5093 See convert_mbs_to_wcs. */
5094 # define POINTER_TO_OFFSET(ptr) \
5095 (FIRST_STRING_P (ptr) \
5096 ? ((regoff_t)(mbs_offset1 != NULL? mbs_offset1[(ptr)-string1] : 0)) \
5097 : ((regoff_t)((mbs_offset2 != NULL? mbs_offset2[(ptr)-string2] : 0) \
5098 + csize1)))
5099 #else
5100 /* This converts PTR, a pointer into one of the search strings `string1'
5101 and `string2' into an offset from the beginning of that string. */
5102 # define POINTER_TO_OFFSET(ptr) \
5103 (FIRST_STRING_P (ptr) \
5104 ? ((regoff_t) ((ptr) - string1)) \
5105 : ((regoff_t) ((ptr) - string2 + size1)))
5106 #endif /* MBS_SUPPORT */
5107
5108 /* Macros for dealing with the split strings in re_match_2. */
5109
5110 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
5111
5112 /* Call before fetching a character with *d. This switches over to
5113 string2 if necessary. */
5114 #define PREFETCH() \
5115 while (d == dend) \
5116 { \
5117 /* End of string2 => fail. */ \
5118 if (dend == end_match_2) \
5119 goto fail; \
5120 /* End of string1 => advance to string2. */ \
5121 d = string2; \
5122 dend = end_match_2; \
5123 }
5124
5125
5126 /* Test if at very beginning or at very end of the virtual concatenation
5127 of `string1' and `string2'. If only one string, it's `string2'. */
5128 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
5129 #define AT_STRINGS_END(d) ((d) == end2)
5130
5131
5132 /* Test if D points to a character which is word-constituent. We have
5133 two special cases to check for: if past the end of string1, look at
5134 the first character in string2; and if before the beginning of
5135 string2, look at the last character in string1. */
5136 #ifdef MBS_SUPPORT
5137 /* Use internationalized API instead of SYNTAX. */
5138 # define WORDCHAR_P(d) \
5139 (iswalnum ((wint_t)((d) == end1 ? *string2 \
5140 : (d) == string2 - 1 ? *(end1 - 1) : *(d))) != 0)
5141 #else
5142 # define WORDCHAR_P(d) \
5143 (SYNTAX ((d) == end1 ? *string2 \
5144 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
5145 == Sword)
5146 #endif /* MBS_SUPPORT */
5147
5148 /* Disabled due to a compiler bug -- see comment at case wordbound */
5149 #if 0
5150 /* Test if the character before D and the one at D differ with respect
5151 to being word-constituent. */
5152 #define AT_WORD_BOUNDARY(d) \
5153 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
5154 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
5155 #endif
5156
5157 /* Free everything we malloc. */
5158 #ifdef MATCH_MAY_ALLOCATE
5159 # define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
5160 # ifdef MBS_SUPPORT
5161 # define FREE_VARIABLES() \
5162 do { \
5163 REGEX_FREE_STACK (fail_stack.stack); \
5164 FREE_VAR (regstart); \
5165 FREE_VAR (regend); \
5166 FREE_VAR (old_regstart); \
5167 FREE_VAR (old_regend); \
5168 FREE_VAR (best_regstart); \
5169 FREE_VAR (best_regend); \
5170 FREE_VAR (reg_info); \
5171 FREE_VAR (reg_dummy); \
5172 FREE_VAR (reg_info_dummy); \
5173 FREE_VAR (string1); \
5174 FREE_VAR (string2); \
5175 FREE_VAR (mbs_offset1); \
5176 FREE_VAR (mbs_offset2); \
5177 } while (0)
5178 # else /* not MBS_SUPPORT */
5179 # define FREE_VARIABLES() \
5180 do { \
5181 REGEX_FREE_STACK (fail_stack.stack); \
5182 FREE_VAR (regstart); \
5183 FREE_VAR (regend); \
5184 FREE_VAR (old_regstart); \
5185 FREE_VAR (old_regend); \
5186 FREE_VAR (best_regstart); \
5187 FREE_VAR (best_regend); \
5188 FREE_VAR (reg_info); \
5189 FREE_VAR (reg_dummy); \
5190 FREE_VAR (reg_info_dummy); \
5191 } while (0)
5192 # endif /* MBS_SUPPORT */
5193 #else
5194 # define FREE_VAR(var) if (var) free (var); var = NULL
5195 # ifdef MBS_SUPPORT
5196 # define FREE_VARIABLES() \
5197 do { \
5198 FREE_VAR (string1); \
5199 FREE_VAR (string2); \
5200 FREE_VAR (mbs_offset1); \
5201 FREE_VAR (mbs_offset2); \
5202 } while (0)
5203 # else
5204 # define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
5205 # endif /* MBS_SUPPORT */
5206 #endif /* not MATCH_MAY_ALLOCATE */
5207
5208 /* These values must meet several constraints. They must not be valid
5209 register values; since we have a limit of 255 registers (because
5210 we use only one byte in the pattern for the register number), we can
5211 use numbers larger than 255. They must differ by 1, because of
5212 NUM_FAILURE_ITEMS above. And the value for the lowest register must
5213 be larger than the value for the highest register, so we do not try
5214 to actually save any registers when none are active. */
5215 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
5216 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
5217
5218 /* Matching routines. */
5219
5220 #ifndef emacs /* Emacs never uses this. */
5221 /* re_match is like re_match_2 except it takes only a single string. */
5222
5223 int
5224 re_match (bufp, string, size, pos, regs)
5225 struct re_pattern_buffer *bufp;
5226 const char *string;
5227 int size, pos;
5228 struct re_registers *regs;
5229 {
5230 int result = re_match_2_internal (bufp, NULL, 0, string, size,
5231 pos, regs, size);
5232 # ifndef REGEX_MALLOC
5233 # ifdef C_ALLOCA
5234 alloca (0);
5235 # endif
5236 # endif
5237 return result;
5238 }
5239 # ifdef _LIBC
5240 weak_alias (__re_match, re_match)
5241 # endif
5242 #endif /* not emacs */
5243
5244 static boolean group_match_null_string_p _RE_ARGS ((US_CHAR_TYPE **p,
5245 US_CHAR_TYPE *end,
5246 register_info_type *reg_info));
5247 static boolean alt_match_null_string_p _RE_ARGS ((US_CHAR_TYPE *p,
5248 US_CHAR_TYPE *end,
5249 register_info_type *reg_info));
5250 static boolean common_op_match_null_string_p _RE_ARGS ((US_CHAR_TYPE **p,
5251 US_CHAR_TYPE *end,
5252 register_info_type *reg_info));
5253 static int bcmp_translate _RE_ARGS ((const CHAR_TYPE *s1, const CHAR_TYPE *s2,
5254 int len, char *translate));
5255
5256 /* re_match_2 matches the compiled pattern in BUFP against the
5257 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
5258 and SIZE2, respectively). We start matching at POS, and stop
5259 matching at STOP.
5260
5261 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
5262 store offsets for the substring each group matched in REGS. See the
5263 documentation for exactly how many groups we fill.
5264
5265 We return -1 if no match, -2 if an internal error (such as the
5266 failure stack overflowing). Otherwise, we return the length of the
5267 matched substring. */
5268
5269 int
re_match_2(bufp,string1,size1,string2,size2,pos,regs,stop)5270 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
5271 struct re_pattern_buffer *bufp;
5272 const char *string1, *string2;
5273 int size1, size2;
5274 int pos;
5275 struct re_registers *regs;
5276 int stop;
5277 {
5278 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
5279 pos, regs, stop);
5280 #ifndef REGEX_MALLOC
5281 # ifdef C_ALLOCA
5282 alloca (0);
5283 # endif
5284 #endif
5285 return result;
5286 }
5287 #ifdef _LIBC
5288 weak_alias (__re_match_2, re_match_2)
5289 #endif
5290
5291 #ifdef MBS_SUPPORT
5292
5293 static int count_mbs_length PARAMS ((int *, int));
5294
5295 /* This check the substring (from 0, to length) of the multibyte string,
5296 to which offset_buffer correspond. And count how many wchar_t_characters
5297 the substring occupy. We use offset_buffer to optimization.
5298 See convert_mbs_to_wcs. */
5299
5300 static int
count_mbs_length(offset_buffer,length)5301 count_mbs_length(offset_buffer, length)
5302 int *offset_buffer;
5303 int length;
5304 {
5305 int wcs_size;
5306
5307 /* Check whether the size is valid. */
5308 if (length < 0)
5309 return -1;
5310
5311 if (offset_buffer == NULL)
5312 return 0;
5313
5314 for (wcs_size = 0 ; offset_buffer[wcs_size] != -1 ; wcs_size++)
5315 {
5316 if (offset_buffer[wcs_size] == length)
5317 return wcs_size;
5318 if (offset_buffer[wcs_size] > length)
5319 /* It is a fragment of a wide character. */
5320 return -1;
5321 }
5322
5323 /* We reached at the sentinel. */
5324 return -1;
5325 }
5326 #endif /* MBS_SUPPORT */
5327
5328 /* This is a separate function so that we can force an alloca cleanup
5329 afterwards. */
5330 static int
5331 #ifdef MBS_SUPPORT
re_match_2_internal(bufp,cstring1,csize1,cstring2,csize2,pos,regs,stop)5332 re_match_2_internal (bufp, cstring1, csize1, cstring2, csize2, pos, regs, stop)
5333 struct re_pattern_buffer *bufp;
5334 const char *cstring1, *cstring2;
5335 int csize1, csize2;
5336 #else
5337 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
5338 struct re_pattern_buffer *bufp;
5339 const char *string1, *string2;
5340 int size1, size2;
5341 #endif
5342 int pos;
5343 struct re_registers *regs;
5344 int stop;
5345 {
5346 /* General temporaries. */
5347 int mcnt;
5348 US_CHAR_TYPE *p1;
5349 #ifdef MBS_SUPPORT
5350 /* We need wchar_t* buffers correspond to string1, string2. */
5351 CHAR_TYPE *string1 = NULL, *string2 = NULL;
5352 /* We need the size of wchar_t buffers correspond to csize1, csize2. */
5353 int size1 = 0, size2 = 0;
5354 /* offset buffer for optimizatoin. See convert_mbs_to_wc. */
5355 int *mbs_offset1 = NULL, *mbs_offset2 = NULL;
5356 /* They hold whether each wchar_t is binary data or not. */
5357 char *is_binary = NULL;
5358 #endif /* MBS_SUPPORT */
5359
5360 /* Just past the end of the corresponding string. */
5361 const CHAR_TYPE *end1, *end2;
5362
5363 /* Pointers into string1 and string2, just past the last characters in
5364 each to consider matching. */
5365 const CHAR_TYPE *end_match_1, *end_match_2;
5366
5367 /* Where we are in the data, and the end of the current string. */
5368 const CHAR_TYPE *d, *dend;
5369
5370 /* Where we are in the pattern, and the end of the pattern. */
5371 #ifdef MBS_SUPPORT
5372 US_CHAR_TYPE *pattern, *p;
5373 register US_CHAR_TYPE *pend;
5374 #else
5375 US_CHAR_TYPE *p = bufp->buffer;
5376 register US_CHAR_TYPE *pend = p + bufp->used;
5377 #endif /* MBS_SUPPORT */
5378
5379 /* Mark the opcode just after a start_memory, so we can test for an
5380 empty subpattern when we get to the stop_memory. */
5381 US_CHAR_TYPE *just_past_start_mem = 0;
5382
5383 /* We use this to map every character in the string. */
5384 RE_TRANSLATE_TYPE translate = bufp->translate;
5385
5386 /* Failure point stack. Each place that can handle a failure further
5387 down the line pushes a failure point on this stack. It consists of
5388 restart, regend, and reg_info for all registers corresponding to
5389 the subexpressions we're currently inside, plus the number of such
5390 registers, and, finally, two char *'s. The first char * is where
5391 to resume scanning the pattern; the second one is where to resume
5392 scanning the strings. If the latter is zero, the failure point is
5393 a ``dummy''; if a failure happens and the failure point is a dummy,
5394 it gets discarded and the next next one is tried. */
5395 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5396 fail_stack_type fail_stack;
5397 #endif
5398 #ifdef DEBUG
5399 static unsigned failure_id;
5400 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
5401 #endif
5402
5403 #ifdef REL_ALLOC
5404 /* This holds the pointer to the failure stack, when
5405 it is allocated relocatably. */
5406 fail_stack_elt_t *failure_stack_ptr;
5407 #endif
5408
5409 /* We fill all the registers internally, independent of what we
5410 return, for use in backreferences. The number here includes
5411 an element for register zero. */
5412 size_t num_regs = bufp->re_nsub + 1;
5413
5414 /* The currently active registers. */
5415 active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG;
5416 active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG;
5417
5418 /* Information on the contents of registers. These are pointers into
5419 the input strings; they record just what was matched (on this
5420 attempt) by a subexpression part of the pattern, that is, the
5421 regnum-th regstart pointer points to where in the pattern we began
5422 matching and the regnum-th regend points to right after where we
5423 stopped matching the regnum-th subexpression. (The zeroth register
5424 keeps track of what the whole pattern matches.) */
5425 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5426 const CHAR_TYPE **regstart, **regend;
5427 #endif
5428
5429 /* If a group that's operated upon by a repetition operator fails to
5430 match anything, then the register for its start will need to be
5431 restored because it will have been set to wherever in the string we
5432 are when we last see its open-group operator. Similarly for a
5433 register's end. */
5434 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5435 const CHAR_TYPE **old_regstart, **old_regend;
5436 #endif
5437
5438 /* The is_active field of reg_info helps us keep track of which (possibly
5439 nested) subexpressions we are currently in. The matched_something
5440 field of reg_info[reg_num] helps us tell whether or not we have
5441 matched any of the pattern so far this time through the reg_num-th
5442 subexpression. These two fields get reset each time through any
5443 loop their register is in. */
5444 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
5445 register_info_type *reg_info;
5446 #endif
5447
5448 /* The following record the register info as found in the above
5449 variables when we find a match better than any we've seen before.
5450 This happens as we backtrack through the failure points, which in
5451 turn happens only if we have not yet matched the entire string. */
5452 unsigned best_regs_set = false;
5453 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5454 const CHAR_TYPE **best_regstart, **best_regend;
5455 #endif
5456
5457 /* Logically, this is `best_regend[0]'. But we don't want to have to
5458 allocate space for that if we're not allocating space for anything
5459 else (see below). Also, we never need info about register 0 for
5460 any of the other register vectors, and it seems rather a kludge to
5461 treat `best_regend' differently than the rest. So we keep track of
5462 the end of the best match so far in a separate variable. We
5463 initialize this to NULL so that when we backtrack the first time
5464 and need to test it, it's not garbage. */
5465 const CHAR_TYPE *match_end = NULL;
5466
5467 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
5468 int set_regs_matched_done = 0;
5469
5470 /* Used when we pop values we don't care about. */
5471 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
5472 const CHAR_TYPE **reg_dummy;
5473 register_info_type *reg_info_dummy;
5474 #endif
5475
5476 #ifdef DEBUG
5477 /* Counts the total number of registers pushed. */
5478 unsigned num_regs_pushed = 0;
5479 #endif
5480
5481 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
5482
5483 INIT_FAIL_STACK ();
5484
5485 #ifdef MATCH_MAY_ALLOCATE
5486 /* Do not bother to initialize all the register variables if there are
5487 no groups in the pattern, as it takes a fair amount of time. If
5488 there are groups, we include space for register 0 (the whole
5489 pattern), even though we never use it, since it simplifies the
5490 array indexing. We should fix this. */
5491 if (bufp->re_nsub)
5492 {
5493 regstart = REGEX_TALLOC (num_regs, const CHAR_TYPE *);
5494 regend = REGEX_TALLOC (num_regs, const CHAR_TYPE *);
5495 old_regstart = REGEX_TALLOC (num_regs, const CHAR_TYPE *);
5496 old_regend = REGEX_TALLOC (num_regs, const CHAR_TYPE *);
5497 best_regstart = REGEX_TALLOC (num_regs, const CHAR_TYPE *);
5498 best_regend = REGEX_TALLOC (num_regs, const CHAR_TYPE *);
5499 reg_info = REGEX_TALLOC (num_regs, register_info_type);
5500 reg_dummy = REGEX_TALLOC (num_regs, const CHAR_TYPE *);
5501 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
5502
5503 if (!(regstart && regend && old_regstart && old_regend && reg_info
5504 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
5505 {
5506 FREE_VARIABLES ();
5507 return -2;
5508 }
5509 }
5510 else
5511 {
5512 /* We must initialize all our variables to NULL, so that
5513 `FREE_VARIABLES' doesn't try to free them. */
5514 regstart = regend = old_regstart = old_regend = best_regstart
5515 = best_regend = reg_dummy = NULL;
5516 reg_info = reg_info_dummy = (register_info_type *) NULL;
5517 }
5518 #endif /* MATCH_MAY_ALLOCATE */
5519
5520 /* The starting position is bogus. */
5521 #ifdef MBS_SUPPORT
5522 if (pos < 0 || pos > csize1 + csize2)
5523 #else
5524 if (pos < 0 || pos > size1 + size2)
5525 #endif
5526 {
5527 FREE_VARIABLES ();
5528 return -1;
5529 }
5530
5531 #ifdef MBS_SUPPORT
5532 /* Allocate wchar_t array for string1 and string2 and
5533 fill them with converted string. */
5534 if (csize1 != 0)
5535 {
5536 string1 = REGEX_TALLOC (csize1 + 1, CHAR_TYPE);
5537 mbs_offset1 = REGEX_TALLOC (csize1 + 1, int);
5538 is_binary = REGEX_TALLOC (csize1 + 1, char);
5539 if (!string1 || !mbs_offset1 || !is_binary)
5540 {
5541 FREE_VAR (string1);
5542 FREE_VAR (mbs_offset1);
5543 FREE_VAR (is_binary);
5544 return -2;
5545 }
5546 size1 = convert_mbs_to_wcs(string1, cstring1, csize1,
5547 mbs_offset1, is_binary);
5548 string1[size1] = L'\0'; /* for a sentinel */
5549 FREE_VAR (is_binary);
5550 }
5551 if (csize2 != 0)
5552 {
5553 string2 = REGEX_TALLOC (csize2 + 1, CHAR_TYPE);
5554 mbs_offset2 = REGEX_TALLOC (csize2 + 1, int);
5555 is_binary = REGEX_TALLOC (csize2 + 1, char);
5556 if (!string2 || !mbs_offset2 || !is_binary)
5557 {
5558 FREE_VAR (string1);
5559 FREE_VAR (mbs_offset1);
5560 FREE_VAR (string2);
5561 FREE_VAR (mbs_offset2);
5562 FREE_VAR (is_binary);
5563 return -2;
5564 }
5565 size2 = convert_mbs_to_wcs(string2, cstring2, csize2,
5566 mbs_offset2, is_binary);
5567 string2[size2] = L'\0'; /* for a sentinel */
5568 FREE_VAR (is_binary);
5569 }
5570
5571 /* We need to cast pattern to (wchar_t*), because we casted this compiled
5572 pattern to (char*) in regex_compile. */
5573 p = pattern = (CHAR_TYPE*)bufp->buffer;
5574 pend = (CHAR_TYPE*)(bufp->buffer + bufp->used);
5575
5576 #endif /* MBS_SUPPORT */
5577
5578 /* Initialize subexpression text positions to -1 to mark ones that no
5579 start_memory/stop_memory has been seen for. Also initialize the
5580 register information struct. */
5581 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5582 {
5583 regstart[mcnt] = regend[mcnt]
5584 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
5585
5586 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
5587 IS_ACTIVE (reg_info[mcnt]) = 0;
5588 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5589 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
5590 }
5591
5592 /* We move `string1' into `string2' if the latter's empty -- but not if
5593 `string1' is null. */
5594 if (size2 == 0 && string1 != NULL)
5595 {
5596 string2 = string1;
5597 size2 = size1;
5598 string1 = 0;
5599 size1 = 0;
5600 }
5601 end1 = string1 + size1;
5602 end2 = string2 + size2;
5603
5604 /* Compute where to stop matching, within the two strings. */
5605 #ifdef MBS_SUPPORT
5606 if (stop <= csize1)
5607 {
5608 mcnt = count_mbs_length(mbs_offset1, stop);
5609 end_match_1 = string1 + mcnt;
5610 end_match_2 = string2;
5611 }
5612 else
5613 {
5614 end_match_1 = end1;
5615 mcnt = count_mbs_length(mbs_offset2, stop-csize1);
5616 end_match_2 = string2 + mcnt;
5617 }
5618 if (mcnt < 0)
5619 { /* count_mbs_length return error. */
5620 FREE_VARIABLES ();
5621 return -1;
5622 }
5623 #else
5624 if (stop <= size1)
5625 {
5626 end_match_1 = string1 + stop;
5627 end_match_2 = string2;
5628 }
5629 else
5630 {
5631 end_match_1 = end1;
5632 end_match_2 = string2 + stop - size1;
5633 }
5634 #endif /* MBS_SUPPORT */
5635
5636 /* `p' scans through the pattern as `d' scans through the data.
5637 `dend' is the end of the input string that `d' points within. `d'
5638 is advanced into the following input string whenever necessary, but
5639 this happens before fetching; therefore, at the beginning of the
5640 loop, `d' can be pointing at the end of a string, but it cannot
5641 equal `string2'. */
5642 #ifdef MBS_SUPPORT
5643 if (size1 > 0 && pos <= csize1)
5644 {
5645 mcnt = count_mbs_length(mbs_offset1, pos);
5646 d = string1 + mcnt;
5647 dend = end_match_1;
5648 }
5649 else
5650 {
5651 mcnt = count_mbs_length(mbs_offset2, pos-csize1);
5652 d = string2 + mcnt;
5653 dend = end_match_2;
5654 }
5655
5656 if (mcnt < 0)
5657 { /* count_mbs_length return error. */
5658 FREE_VARIABLES ();
5659 return -1;
5660 }
5661 #else
5662 if (size1 > 0 && pos <= size1)
5663 {
5664 d = string1 + pos;
5665 dend = end_match_1;
5666 }
5667 else
5668 {
5669 d = string2 + pos - size1;
5670 dend = end_match_2;
5671 }
5672 #endif /* MBS_SUPPORT */
5673
5674 DEBUG_PRINT1 ("The compiled pattern is:\n");
5675 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
5676 DEBUG_PRINT1 ("The string to match is: `");
5677 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
5678 DEBUG_PRINT1 ("'\n");
5679
5680 /* This loops over pattern commands. It exits by returning from the
5681 function if the match is complete, or it drops through if the match
5682 fails at this starting point in the input data. */
5683 for (;;)
5684 {
5685 #ifdef _LIBC
5686 DEBUG_PRINT2 ("\n%p: ", p);
5687 #else
5688 DEBUG_PRINT2 ("\n0x%x: ", p);
5689 #endif
5690
5691 if (p == pend)
5692 { /* End of pattern means we might have succeeded. */
5693 DEBUG_PRINT1 ("end of pattern ... ");
5694
5695 /* If we haven't matched the entire string, and we want the
5696 longest match, try backtracking. */
5697 if (d != end_match_2)
5698 {
5699 /* 1 if this match ends in the same string (string1 or string2)
5700 as the best previous match. */
5701 boolean same_str_p = (FIRST_STRING_P (match_end)
5702 == MATCHING_IN_FIRST_STRING);
5703 /* 1 if this match is the best seen so far. */
5704 boolean best_match_p;
5705
5706 /* AIX compiler got confused when this was combined
5707 with the previous declaration. */
5708 if (same_str_p)
5709 best_match_p = d > match_end;
5710 else
5711 best_match_p = !MATCHING_IN_FIRST_STRING;
5712
5713 DEBUG_PRINT1 ("backtracking.\n");
5714
5715 if (!FAIL_STACK_EMPTY ())
5716 { /* More failure points to try. */
5717
5718 /* If exceeds best match so far, save it. */
5719 if (!best_regs_set || best_match_p)
5720 {
5721 best_regs_set = true;
5722 match_end = d;
5723
5724 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
5725
5726 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5727 {
5728 best_regstart[mcnt] = regstart[mcnt];
5729 best_regend[mcnt] = regend[mcnt];
5730 }
5731 }
5732 goto fail;
5733 }
5734
5735 /* If no failure points, don't restore garbage. And if
5736 last match is real best match, don't restore second
5737 best one. */
5738 else if (best_regs_set && !best_match_p)
5739 {
5740 restore_best_regs:
5741 /* Restore best match. It may happen that `dend ==
5742 end_match_1' while the restored d is in string2.
5743 For example, the pattern `x.*y.*z' against the
5744 strings `x-' and `y-z-', if the two strings are
5745 not consecutive in memory. */
5746 DEBUG_PRINT1 ("Restoring best registers.\n");
5747
5748 d = match_end;
5749 dend = ((d >= string1 && d <= end1)
5750 ? end_match_1 : end_match_2);
5751
5752 for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++)
5753 {
5754 regstart[mcnt] = best_regstart[mcnt];
5755 regend[mcnt] = best_regend[mcnt];
5756 }
5757 }
5758 } /* d != end_match_2 */
5759
5760 succeed_label:
5761 DEBUG_PRINT1 ("Accepting match.\n");
5762 /* If caller wants register contents data back, do it. */
5763 if (regs && !bufp->no_sub)
5764 {
5765 /* Have the register data arrays been allocated? */
5766 if (bufp->regs_allocated == REGS_UNALLOCATED)
5767 { /* No. So allocate them with malloc. We need one
5768 extra element beyond `num_regs' for the `-1' marker
5769 GNU code uses. */
5770 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
5771 regs->start = TALLOC (regs->num_regs, regoff_t);
5772 regs->end = TALLOC (regs->num_regs, regoff_t);
5773 if (regs->start == NULL || regs->end == NULL)
5774 {
5775 FREE_VARIABLES ();
5776 return -2;
5777 }
5778 bufp->regs_allocated = REGS_REALLOCATE;
5779 }
5780 else if (bufp->regs_allocated == REGS_REALLOCATE)
5781 { /* Yes. If we need more elements than were already
5782 allocated, reallocate them. If we need fewer, just
5783 leave it alone. */
5784 if (regs->num_regs < num_regs + 1)
5785 {
5786 regs->num_regs = num_regs + 1;
5787 RETALLOC (regs->start, regs->num_regs, regoff_t);
5788 RETALLOC (regs->end, regs->num_regs, regoff_t);
5789 if (regs->start == NULL || regs->end == NULL)
5790 {
5791 FREE_VARIABLES ();
5792 return -2;
5793 }
5794 }
5795 }
5796 else
5797 {
5798 /* These braces fend off a "empty body in an else-statement"
5799 warning under GCC when assert expands to nothing. */
5800 assert (bufp->regs_allocated == REGS_FIXED);
5801 }
5802
5803 /* Convert the pointer data in `regstart' and `regend' to
5804 indices. Register zero has to be set differently,
5805 since we haven't kept track of any info for it. */
5806 if (regs->num_regs > 0)
5807 {
5808 regs->start[0] = pos;
5809 #ifdef MBS_SUPPORT
5810 if (MATCHING_IN_FIRST_STRING)
5811 regs->end[0] = mbs_offset1 != NULL ?
5812 mbs_offset1[d-string1] : 0;
5813 else
5814 regs->end[0] = csize1 + (mbs_offset2 != NULL ?
5815 mbs_offset2[d-string2] : 0);
5816 #else
5817 regs->end[0] = (MATCHING_IN_FIRST_STRING
5818 ? ((regoff_t) (d - string1))
5819 : ((regoff_t) (d - string2 + size1)));
5820 #endif /* MBS_SUPPORT */
5821 }
5822
5823 /* Go through the first `min (num_regs, regs->num_regs)'
5824 registers, since that is all we initialized. */
5825 for (mcnt = 1; (unsigned) mcnt < MIN (num_regs, regs->num_regs);
5826 mcnt++)
5827 {
5828 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
5829 regs->start[mcnt] = regs->end[mcnt] = -1;
5830 else
5831 {
5832 regs->start[mcnt]
5833 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
5834 regs->end[mcnt]
5835 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
5836 }
5837 }
5838
5839 /* If the regs structure we return has more elements than
5840 were in the pattern, set the extra elements to -1. If
5841 we (re)allocated the registers, this is the case,
5842 because we always allocate enough to have at least one
5843 -1 at the end. */
5844 for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; mcnt++)
5845 regs->start[mcnt] = regs->end[mcnt] = -1;
5846 } /* regs && !bufp->no_sub */
5847
5848 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
5849 nfailure_points_pushed, nfailure_points_popped,
5850 nfailure_points_pushed - nfailure_points_popped);
5851 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
5852
5853 #ifdef MBS_SUPPORT
5854 if (MATCHING_IN_FIRST_STRING)
5855 mcnt = mbs_offset1 != NULL ? mbs_offset1[d-string1] : 0;
5856 else
5857 mcnt = (mbs_offset2 != NULL ? mbs_offset2[d-string2] : 0) +
5858 csize1;
5859 mcnt -= pos;
5860 #else
5861 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
5862 ? string1
5863 : string2 - size1);
5864 #endif /* MBS_SUPPORT */
5865
5866 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
5867
5868 FREE_VARIABLES ();
5869 return mcnt;
5870 }
5871
5872 /* Otherwise match next pattern command. */
5873 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
5874 {
5875 /* Ignore these. Used to ignore the n of succeed_n's which
5876 currently have n == 0. */
5877 case no_op:
5878 DEBUG_PRINT1 ("EXECUTING no_op.\n");
5879 break;
5880
5881 case succeed:
5882 DEBUG_PRINT1 ("EXECUTING succeed.\n");
5883 goto succeed_label;
5884
5885 /* Match the next n pattern characters exactly. The following
5886 byte in the pattern defines n, and the n bytes after that
5887 are the characters to match. */
5888 case exactn:
5889 #ifdef MBS_SUPPORT
5890 case exactn_bin:
5891 #endif
5892 mcnt = *p++;
5893 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
5894
5895 /* This is written out as an if-else so we don't waste time
5896 testing `translate' inside the loop. */
5897 if (translate)
5898 {
5899 do
5900 {
5901 PREFETCH ();
5902 #ifdef MBS_SUPPORT
5903 if (*d <= 0xff)
5904 {
5905 if ((US_CHAR_TYPE) translate[(unsigned char) *d++]
5906 != (US_CHAR_TYPE) *p++)
5907 goto fail;
5908 }
5909 else
5910 {
5911 if (*d++ != (CHAR_TYPE) *p++)
5912 goto fail;
5913 }
5914 #else
5915 if ((US_CHAR_TYPE) translate[(unsigned char) *d++]
5916 != (US_CHAR_TYPE) *p++)
5917 goto fail;
5918 #endif /* MBS_SUPPORT */
5919 }
5920 while (--mcnt);
5921 }
5922 else
5923 {
5924 do
5925 {
5926 PREFETCH ();
5927 if (*d++ != (CHAR_TYPE) *p++) goto fail;
5928 }
5929 while (--mcnt);
5930 }
5931 SET_REGS_MATCHED ();
5932 break;
5933
5934
5935 /* Match any character except possibly a newline or a null. */
5936 case anychar:
5937 DEBUG_PRINT1 ("EXECUTING anychar.\n");
5938
5939 PREFETCH ();
5940
5941 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
5942 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
5943 goto fail;
5944
5945 SET_REGS_MATCHED ();
5946 DEBUG_PRINT2 (" Matched `%ld'.\n", (long int) *d);
5947 d++;
5948 break;
5949
5950
5951 case charset:
5952 case charset_not:
5953 {
5954 register US_CHAR_TYPE c;
5955 #ifdef MBS_SUPPORT
5956 unsigned int i, char_class_length, coll_symbol_length,
5957 equiv_class_length, ranges_length, chars_length, length;
5958 CHAR_TYPE *workp, *workp2, *charset_top;
5959 #define WORK_BUFFER_SIZE 128
5960 CHAR_TYPE str_buf[WORK_BUFFER_SIZE];
5961 # ifdef _LIBC
5962 uint32_t nrules;
5963 # endif /* _LIBC */
5964 #endif /* MBS_SUPPORT */
5965 boolean not = (re_opcode_t) *(p - 1) == charset_not;
5966
5967 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
5968 PREFETCH ();
5969 c = TRANSLATE (*d); /* The character to match. */
5970 #ifdef MBS_SUPPORT
5971 # ifdef _LIBC
5972 nrules = _NL_CURRENT_WORD (LC_COLLATE, _NL_COLLATE_NRULES);
5973 # endif /* _LIBC */
5974 charset_top = p - 1;
5975 char_class_length = *p++;
5976 coll_symbol_length = *p++;
5977 equiv_class_length = *p++;
5978 ranges_length = *p++;
5979 chars_length = *p++;
5980 /* p points charset[6], so the address of the next instruction
5981 (charset[l+m+n+2o+k+p']) equals p[l+m+n+2*o+p'],
5982 where l=length of char_classes, m=length of collating_symbol,
5983 n=equivalence_class, o=length of char_range,
5984 p'=length of character. */
5985 workp = p;
5986 /* Update p to indicate the next instruction. */
5987 p += char_class_length + coll_symbol_length+ equiv_class_length +
5988 2*ranges_length + chars_length;
5989
5990 /* match with char_class? */
5991 for (i = 0; i < char_class_length ; i += CHAR_CLASS_SIZE)
5992 {
5993 wctype_t wctype;
5994 uintptr_t alignedp = ((uintptr_t)workp
5995 + __alignof__(wctype_t) - 1)
5996 & ~(uintptr_t)(__alignof__(wctype_t) - 1);
5997 wctype = *((wctype_t*)alignedp);
5998 workp += CHAR_CLASS_SIZE;
5999 if (iswctype((wint_t)c, wctype))
6000 goto char_set_matched;
6001 }
6002
6003 /* match with collating_symbol? */
6004 # ifdef _LIBC
6005 if (nrules != 0)
6006 {
6007 const unsigned char *extra = (const unsigned char *)
6008 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_SYMB_EXTRAMB);
6009
6010 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;
6011 workp++)
6012 {
6013 int32_t *wextra;
6014 wextra = (int32_t*)(extra + *workp++);
6015 for (i = 0; i < *wextra; ++i)
6016 if (TRANSLATE(d[i]) != wextra[1 + i])
6017 break;
6018
6019 if (i == *wextra)
6020 {
6021 /* Update d, however d will be incremented at
6022 char_set_matched:, we decrement d here. */
6023 d += i - 1;
6024 goto char_set_matched;
6025 }
6026 }
6027 }
6028 else /* (nrules == 0) */
6029 # endif
6030 /* If we can't look up collation data, we use wcscoll
6031 instead. */
6032 {
6033 for (workp2 = workp + coll_symbol_length ; workp < workp2 ;)
6034 {
6035 const CHAR_TYPE *backup_d = d, *backup_dend = dend;
6036 length = wcslen(workp);
6037
6038 /* If wcscoll(the collating symbol, whole string) > 0,
6039 any substring of the string never match with the
6040 collating symbol. */
6041 if (wcscoll(workp, d) > 0)
6042 {
6043 workp += length + 1;
6044 continue;
6045 }
6046
6047 /* First, we compare the collating symbol with
6048 the first character of the string.
6049 If it don't match, we add the next character to
6050 the compare buffer in turn. */
6051 for (i = 0 ; i < WORK_BUFFER_SIZE-1 ; i++, d++)
6052 {
6053 int match;
6054 if (d == dend)
6055 {
6056 if (dend == end_match_2)
6057 break;
6058 d = string2;
6059 dend = end_match_2;
6060 }
6061
6062 /* add next character to the compare buffer. */
6063 str_buf[i] = TRANSLATE(*d);
6064 str_buf[i+1] = '\0';
6065
6066 match = wcscoll(workp, str_buf);
6067 if (match == 0)
6068 goto char_set_matched;
6069
6070 if (match < 0)
6071 /* (str_buf > workp) indicate (str_buf + X > workp),
6072 because for all X (str_buf + X > str_buf).
6073 So we don't need continue this loop. */
6074 break;
6075
6076 /* Otherwise(str_buf < workp),
6077 (str_buf+next_character) may equals (workp).
6078 So we continue this loop. */
6079 }
6080 /* not matched */
6081 d = backup_d;
6082 dend = backup_dend;
6083 workp += length + 1;
6084 }
6085 }
6086 /* match with equivalence_class? */
6087 # ifdef _LIBC
6088 if (nrules != 0)
6089 {
6090 const CHAR_TYPE *backup_d = d, *backup_dend = dend;
6091 /* Try to match the equivalence class against
6092 those known to the collate implementation. */
6093 const int32_t *table;
6094 const int32_t *weights;
6095 const int32_t *extra;
6096 const int32_t *indirect;
6097 int32_t idx, idx2;
6098 wint_t *cp;
6099 size_t len;
6100
6101 /* This #include defines a local function! */
6102 # include <locale/weightwc.h>
6103
6104 table = (const int32_t *)
6105 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_TABLEWC);
6106 weights = (const wint_t *)
6107 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_WEIGHTWC);
6108 extra = (const wint_t *)
6109 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_EXTRAWC);
6110 indirect = (const int32_t *)
6111 _NL_CURRENT (LC_COLLATE, _NL_COLLATE_INDIRECTWC);
6112
6113 /* Write 1 collating element to str_buf, and
6114 get its index. */
6115 idx2 = 0;
6116
6117 for (i = 0 ; idx2 == 0 && i < WORK_BUFFER_SIZE - 1; i++)
6118 {
6119 cp = (wint_t*)str_buf;
6120 if (d == dend)
6121 {
6122 if (dend == end_match_2)
6123 break;
6124 d = string2;
6125 dend = end_match_2;
6126 }
6127 str_buf[i] = TRANSLATE(*(d+i));
6128 str_buf[i+1] = '\0'; /* sentinel */
6129 idx2 = findidx ((const wint_t**)&cp);
6130 }
6131
6132 /* Update d, however d will be incremented at
6133 char_set_matched:, we decrement d here. */
6134 d = backup_d + ((wchar_t*)cp - (wchar_t*)str_buf - 1);
6135 if (d >= dend)
6136 {
6137 if (dend == end_match_2)
6138 d = dend;
6139 else
6140 {
6141 d = string2;
6142 dend = end_match_2;
6143 }
6144 }
6145
6146 len = weights[idx2];
6147
6148 for (workp2 = workp + equiv_class_length ; workp < workp2 ;
6149 workp++)
6150 {
6151 idx = (int32_t)*workp;
6152 /* We already checked idx != 0 in regex_compile. */
6153
6154 if (idx2 != 0 && len == weights[idx])
6155 {
6156 int cnt = 0;
6157 while (cnt < len && (weights[idx + 1 + cnt]
6158 == weights[idx2 + 1 + cnt]))
6159 ++cnt;
6160
6161 if (cnt == len)
6162 goto char_set_matched;
6163 }
6164 }
6165 /* not matched */
6166 d = backup_d;
6167 dend = backup_dend;
6168 }
6169 else /* (nrules == 0) */
6170 # endif
6171 /* If we can't look up collation data, we use wcscoll
6172 instead. */
6173 {
6174 for (workp2 = workp + equiv_class_length ; workp < workp2 ;)
6175 {
6176 const CHAR_TYPE *backup_d = d, *backup_dend = dend;
6177 length = wcslen(workp);
6178
6179 /* If wcscoll(the collating symbol, whole string) > 0,
6180 any substring of the string never match with the
6181 collating symbol. */
6182 if (wcscoll(workp, d) > 0)
6183 {
6184 workp += length + 1;
6185 break;
6186 }
6187
6188 /* First, we compare the equivalence class with
6189 the first character of the string.
6190 If it don't match, we add the next character to
6191 the compare buffer in turn. */
6192 for (i = 0 ; i < WORK_BUFFER_SIZE - 1 ; i++, d++)
6193 {
6194 int match;
6195 if (d == dend)
6196 {
6197 if (dend == end_match_2)
6198 break;
6199 d = string2;
6200 dend = end_match_2;
6201 }
6202
6203 /* add next character to the compare buffer. */
6204 str_buf[i] = TRANSLATE(*d);
6205 str_buf[i+1] = '\0';
6206
6207 match = wcscoll(workp, str_buf);
6208
6209 if (match == 0)
6210 goto char_set_matched;
6211
6212 if (match < 0)
6213 /* (str_buf > workp) indicate (str_buf + X > workp),
6214 because for all X (str_buf + X > str_buf).
6215 So we don't need continue this loop. */
6216 break;
6217
6218 /* Otherwise(str_buf < workp),
6219 (str_buf+next_character) may equals (workp).
6220 So we continue this loop. */
6221 }
6222 /* not matched */
6223 d = backup_d;
6224 dend = backup_dend;
6225 workp += length + 1;
6226 }
6227 }
6228
6229 /* match with char_range? */
6230 #ifdef _LIBC
6231 if (nrules != 0)
6232 {
6233 uint32_t collseqval;
6234 const char *collseq = (const char *)
6235 _NL_CURRENT(LC_COLLATE, _NL_COLLATE_COLLSEQWC);
6236
6237 collseqval = collseq_table_lookup (collseq, c);
6238
6239 for (; workp < p - chars_length ;)
6240 {
6241 uint32_t start_val, end_val;
6242
6243 /* We already compute the collation sequence value
6244 of the characters (or collating symbols). */
6245 start_val = (uint32_t) *workp++; /* range_start */
6246 end_val = (uint32_t) *workp++; /* range_end */
6247
6248 if (start_val <= collseqval && collseqval <= end_val)
6249 goto char_set_matched;
6250 }
6251 }
6252 else
6253 #endif
6254 {
6255 /* We set range_start_char at str_buf[0], range_end_char
6256 at str_buf[4], and compared char at str_buf[2]. */
6257 str_buf[1] = 0;
6258 str_buf[2] = c;
6259 str_buf[3] = 0;
6260 str_buf[5] = 0;
6261 for (; workp < p - chars_length ;)
6262 {
6263 wchar_t *range_start_char, *range_end_char;
6264
6265 /* match if (range_start_char <= c <= range_end_char). */
6266
6267 /* If range_start(or end) < 0, we assume -range_start(end)
6268 is the offset of the collating symbol which is specified
6269 as the character of the range start(end). */
6270
6271 /* range_start */
6272 if (*workp < 0)
6273 range_start_char = charset_top - (*workp++);
6274 else
6275 {
6276 str_buf[0] = *workp++;
6277 range_start_char = str_buf;
6278 }
6279
6280 /* range_end */
6281 if (*workp < 0)
6282 range_end_char = charset_top - (*workp++);
6283 else
6284 {
6285 str_buf[4] = *workp++;
6286 range_end_char = str_buf + 4;
6287 }
6288
6289 if (wcscoll(range_start_char, str_buf+2) <= 0 &&
6290 wcscoll(str_buf+2, range_end_char) <= 0)
6291
6292 goto char_set_matched;
6293 }
6294 }
6295
6296 /* match with char? */
6297 for (; workp < p ; workp++)
6298 if (c == *workp)
6299 goto char_set_matched;
6300
6301 not = !not;
6302
6303 char_set_matched:
6304 if (not) goto fail;
6305 #else
6306 /* Cast to `unsigned' instead of `unsigned char' in case the
6307 bit list is a full 32 bytes long. */
6308 if (c < (unsigned) (*p * BYTEWIDTH)
6309 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6310 not = !not;
6311
6312 p += 1 + *p;
6313
6314 if (!not) goto fail;
6315 #undef WORK_BUFFER_SIZE
6316 #endif /* MBS_SUPPORT */
6317 SET_REGS_MATCHED ();
6318 d++;
6319 break;
6320 }
6321
6322
6323 /* The beginning of a group is represented by start_memory.
6324 The arguments are the register number in the next byte, and the
6325 number of groups inner to this one in the next. The text
6326 matched within the group is recorded (in the internal
6327 registers data structure) under the register number. */
6328 case start_memory:
6329 DEBUG_PRINT3 ("EXECUTING start_memory %ld (%ld):\n",
6330 (long int) *p, (long int) p[1]);
6331
6332 /* Find out if this group can match the empty string. */
6333 p1 = p; /* To send to group_match_null_string_p. */
6334
6335 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
6336 REG_MATCH_NULL_STRING_P (reg_info[*p])
6337 = group_match_null_string_p (&p1, pend, reg_info);
6338
6339 /* Save the position in the string where we were the last time
6340 we were at this open-group operator in case the group is
6341 operated upon by a repetition operator, e.g., with `(a*)*b'
6342 against `ab'; then we want to ignore where we are now in
6343 the string in case this attempt to match fails. */
6344 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6345 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
6346 : regstart[*p];
6347 DEBUG_PRINT2 (" old_regstart: %d\n",
6348 POINTER_TO_OFFSET (old_regstart[*p]));
6349
6350 regstart[*p] = d;
6351 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
6352
6353 IS_ACTIVE (reg_info[*p]) = 1;
6354 MATCHED_SOMETHING (reg_info[*p]) = 0;
6355
6356 /* Clear this whenever we change the register activity status. */
6357 set_regs_matched_done = 0;
6358
6359 /* This is the new highest active register. */
6360 highest_active_reg = *p;
6361
6362 /* If nothing was active before, this is the new lowest active
6363 register. */
6364 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6365 lowest_active_reg = *p;
6366
6367 /* Move past the register number and inner group count. */
6368 p += 2;
6369 just_past_start_mem = p;
6370
6371 break;
6372
6373
6374 /* The stop_memory opcode represents the end of a group. Its
6375 arguments are the same as start_memory's: the register
6376 number, and the number of inner groups. */
6377 case stop_memory:
6378 DEBUG_PRINT3 ("EXECUTING stop_memory %ld (%ld):\n",
6379 (long int) *p, (long int) p[1]);
6380
6381 /* We need to save the string position the last time we were at
6382 this close-group operator in case the group is operated
6383 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
6384 against `aba'; then we want to ignore where we are now in
6385 the string in case this attempt to match fails. */
6386 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
6387 ? REG_UNSET (regend[*p]) ? d : regend[*p]
6388 : regend[*p];
6389 DEBUG_PRINT2 (" old_regend: %d\n",
6390 POINTER_TO_OFFSET (old_regend[*p]));
6391
6392 regend[*p] = d;
6393 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
6394
6395 /* This register isn't active anymore. */
6396 IS_ACTIVE (reg_info[*p]) = 0;
6397
6398 /* Clear this whenever we change the register activity status. */
6399 set_regs_matched_done = 0;
6400
6401 /* If this was the only register active, nothing is active
6402 anymore. */
6403 if (lowest_active_reg == highest_active_reg)
6404 {
6405 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6406 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6407 }
6408 else
6409 { /* We must scan for the new highest active register, since
6410 it isn't necessarily one less than now: consider
6411 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
6412 new highest active register is 1. */
6413 US_CHAR_TYPE r = *p - 1;
6414 while (r > 0 && !IS_ACTIVE (reg_info[r]))
6415 r--;
6416
6417 /* If we end up at register zero, that means that we saved
6418 the registers as the result of an `on_failure_jump', not
6419 a `start_memory', and we jumped to past the innermost
6420 `stop_memory'. For example, in ((.)*) we save
6421 registers 1 and 2 as a result of the *, but when we pop
6422 back to the second ), we are at the stop_memory 1.
6423 Thus, nothing is active. */
6424 if (r == 0)
6425 {
6426 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
6427 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
6428 }
6429 else
6430 highest_active_reg = r;
6431 }
6432
6433 /* If just failed to match something this time around with a
6434 group that's operated on by a repetition operator, try to
6435 force exit from the ``loop'', and restore the register
6436 information for this group that we had before trying this
6437 last match. */
6438 if ((!MATCHED_SOMETHING (reg_info[*p])
6439 || just_past_start_mem == p - 1)
6440 && (p + 2) < pend)
6441 {
6442 boolean is_a_jump_n = false;
6443
6444 p1 = p + 2;
6445 mcnt = 0;
6446 switch ((re_opcode_t) *p1++)
6447 {
6448 case jump_n:
6449 is_a_jump_n = true;
6450 case pop_failure_jump:
6451 case maybe_pop_jump:
6452 case jump:
6453 case dummy_failure_jump:
6454 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6455 if (is_a_jump_n)
6456 p1 += OFFSET_ADDRESS_SIZE;
6457 break;
6458
6459 default:
6460 /* do nothing */ ;
6461 }
6462 p1 += mcnt;
6463
6464 /* If the next operation is a jump backwards in the pattern
6465 to an on_failure_jump right before the start_memory
6466 corresponding to this stop_memory, exit from the loop
6467 by forcing a failure after pushing on the stack the
6468 on_failure_jump's jump in the pattern, and d. */
6469 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
6470 && (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == start_memory
6471 && p1[2+OFFSET_ADDRESS_SIZE] == *p)
6472 {
6473 /* If this group ever matched anything, then restore
6474 what its registers were before trying this last
6475 failed match, e.g., with `(a*)*b' against `ab' for
6476 regstart[1], and, e.g., with `((a*)*(b*)*)*'
6477 against `aba' for regend[3].
6478
6479 Also restore the registers for inner groups for,
6480 e.g., `((a*)(b*))*' against `aba' (register 3 would
6481 otherwise get trashed). */
6482
6483 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
6484 {
6485 unsigned r;
6486
6487 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
6488
6489 /* Restore this and inner groups' (if any) registers. */
6490 for (r = *p; r < (unsigned) *p + (unsigned) *(p + 1);
6491 r++)
6492 {
6493 regstart[r] = old_regstart[r];
6494
6495 /* xx why this test? */
6496 if (old_regend[r] >= regstart[r])
6497 regend[r] = old_regend[r];
6498 }
6499 }
6500 p1++;
6501 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
6502 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
6503
6504 goto fail;
6505 }
6506 }
6507
6508 /* Move past the register number and the inner group count. */
6509 p += 2;
6510 break;
6511
6512
6513 /* \<digit> has been turned into a `duplicate' command which is
6514 followed by the numeric value of <digit> as the register number. */
6515 case duplicate:
6516 {
6517 register const CHAR_TYPE *d2, *dend2;
6518 int regno = *p++; /* Get which register to match against. */
6519 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
6520
6521 /* Can't back reference a group which we've never matched. */
6522 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
6523 goto fail;
6524
6525 /* Where in input to try to start matching. */
6526 d2 = regstart[regno];
6527
6528 /* Where to stop matching; if both the place to start and
6529 the place to stop matching are in the same string, then
6530 set to the place to stop, otherwise, for now have to use
6531 the end of the first string. */
6532
6533 dend2 = ((FIRST_STRING_P (regstart[regno])
6534 == FIRST_STRING_P (regend[regno]))
6535 ? regend[regno] : end_match_1);
6536 for (;;)
6537 {
6538 /* If necessary, advance to next segment in register
6539 contents. */
6540 while (d2 == dend2)
6541 {
6542 if (dend2 == end_match_2) break;
6543 if (dend2 == regend[regno]) break;
6544
6545 /* End of string1 => advance to string2. */
6546 d2 = string2;
6547 dend2 = regend[regno];
6548 }
6549 /* At end of register contents => success */
6550 if (d2 == dend2) break;
6551
6552 /* If necessary, advance to next segment in data. */
6553 PREFETCH ();
6554
6555 /* How many characters left in this segment to match. */
6556 mcnt = dend - d;
6557
6558 /* Want how many consecutive characters we can match in
6559 one shot, so, if necessary, adjust the count. */
6560 if (mcnt > dend2 - d2)
6561 mcnt = dend2 - d2;
6562
6563 /* Compare that many; failure if mismatch, else move
6564 past them. */
6565 if (translate
6566 ? bcmp_translate (d, d2, mcnt, translate)
6567 : memcmp (d, d2, mcnt*sizeof(US_CHAR_TYPE)))
6568 goto fail;
6569 d += mcnt, d2 += mcnt;
6570
6571 /* Do this because we've match some characters. */
6572 SET_REGS_MATCHED ();
6573 }
6574 }
6575 break;
6576
6577
6578 /* begline matches the empty string at the beginning of the string
6579 (unless `not_bol' is set in `bufp'), and, if
6580 `newline_anchor' is set, after newlines. */
6581 case begline:
6582 DEBUG_PRINT1 ("EXECUTING begline.\n");
6583
6584 if (AT_STRINGS_BEG (d))
6585 {
6586 if (!bufp->not_bol) break;
6587 }
6588 else if (d[-1] == '\n' && bufp->newline_anchor)
6589 {
6590 break;
6591 }
6592 /* In all other cases, we fail. */
6593 goto fail;
6594
6595
6596 /* endline is the dual of begline. */
6597 case endline:
6598 DEBUG_PRINT1 ("EXECUTING endline.\n");
6599
6600 if (AT_STRINGS_END (d))
6601 {
6602 if (!bufp->not_eol) break;
6603 }
6604
6605 /* We have to ``prefetch'' the next character. */
6606 else if ((d == end1 ? *string2 : *d) == '\n'
6607 && bufp->newline_anchor)
6608 {
6609 break;
6610 }
6611 goto fail;
6612
6613
6614 /* Match at the very beginning of the data. */
6615 case begbuf:
6616 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
6617 if (AT_STRINGS_BEG (d))
6618 break;
6619 goto fail;
6620
6621
6622 /* Match at the very end of the data. */
6623 case endbuf:
6624 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
6625 if (AT_STRINGS_END (d))
6626 break;
6627 goto fail;
6628
6629
6630 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
6631 pushes NULL as the value for the string on the stack. Then
6632 `pop_failure_point' will keep the current value for the
6633 string, instead of restoring it. To see why, consider
6634 matching `foo\nbar' against `.*\n'. The .* matches the foo;
6635 then the . fails against the \n. But the next thing we want
6636 to do is match the \n against the \n; if we restored the
6637 string value, we would be back at the foo.
6638
6639 Because this is used only in specific cases, we don't need to
6640 check all the things that `on_failure_jump' does, to make
6641 sure the right things get saved on the stack. Hence we don't
6642 share its code. The only reason to push anything on the
6643 stack at all is that otherwise we would have to change
6644 `anychar's code to do something besides goto fail in this
6645 case; that seems worse than this. */
6646 case on_failure_keep_string_jump:
6647 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
6648
6649 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6650 #ifdef _LIBC
6651 DEBUG_PRINT3 (" %d (to %p):\n", mcnt, p + mcnt);
6652 #else
6653 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
6654 #endif
6655
6656 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
6657 break;
6658
6659
6660 /* Uses of on_failure_jump:
6661
6662 Each alternative starts with an on_failure_jump that points
6663 to the beginning of the next alternative. Each alternative
6664 except the last ends with a jump that in effect jumps past
6665 the rest of the alternatives. (They really jump to the
6666 ending jump of the following alternative, because tensioning
6667 these jumps is a hassle.)
6668
6669 Repeats start with an on_failure_jump that points past both
6670 the repetition text and either the following jump or
6671 pop_failure_jump back to this on_failure_jump. */
6672 case on_failure_jump:
6673 on_failure:
6674 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
6675
6676 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6677 #ifdef _LIBC
6678 DEBUG_PRINT3 (" %d (to %p)", mcnt, p + mcnt);
6679 #else
6680 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
6681 #endif
6682
6683 /* If this on_failure_jump comes right before a group (i.e.,
6684 the original * applied to a group), save the information
6685 for that group and all inner ones, so that if we fail back
6686 to this point, the group's information will be correct.
6687 For example, in \(a*\)*\1, we need the preceding group,
6688 and in \(zz\(a*\)b*\)\2, we need the inner group. */
6689
6690 /* We can't use `p' to check ahead because we push
6691 a failure point to `p + mcnt' after we do this. */
6692 p1 = p;
6693
6694 /* We need to skip no_op's before we look for the
6695 start_memory in case this on_failure_jump is happening as
6696 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
6697 against aba. */
6698 while (p1 < pend && (re_opcode_t) *p1 == no_op)
6699 p1++;
6700
6701 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
6702 {
6703 /* We have a new highest active register now. This will
6704 get reset at the start_memory we are about to get to,
6705 but we will have saved all the registers relevant to
6706 this repetition op, as described above. */
6707 highest_active_reg = *(p1 + 1) + *(p1 + 2);
6708 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
6709 lowest_active_reg = *(p1 + 1);
6710 }
6711
6712 DEBUG_PRINT1 (":\n");
6713 PUSH_FAILURE_POINT (p + mcnt, d, -2);
6714 break;
6715
6716
6717 /* A smart repeat ends with `maybe_pop_jump'.
6718 We change it to either `pop_failure_jump' or `jump'. */
6719 case maybe_pop_jump:
6720 EXTRACT_NUMBER_AND_INCR (mcnt, p);
6721 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
6722 {
6723 register US_CHAR_TYPE *p2 = p;
6724
6725 /* Compare the beginning of the repeat with what in the
6726 pattern follows its end. If we can establish that there
6727 is nothing that they would both match, i.e., that we
6728 would have to backtrack because of (as in, e.g., `a*a')
6729 then we can change to pop_failure_jump, because we'll
6730 never have to backtrack.
6731
6732 This is not true in the case of alternatives: in
6733 `(a|ab)*' we do need to backtrack to the `ab' alternative
6734 (e.g., if the string was `ab'). But instead of trying to
6735 detect that here, the alternative has put on a dummy
6736 failure point which is what we will end up popping. */
6737
6738 /* Skip over open/close-group commands.
6739 If what follows this loop is a ...+ construct,
6740 look at what begins its body, since we will have to
6741 match at least one of that. */
6742 while (1)
6743 {
6744 if (p2 + 2 < pend
6745 && ((re_opcode_t) *p2 == stop_memory
6746 || (re_opcode_t) *p2 == start_memory))
6747 p2 += 3;
6748 else if (p2 + 2 + 2 * OFFSET_ADDRESS_SIZE < pend
6749 && (re_opcode_t) *p2 == dummy_failure_jump)
6750 p2 += 2 + 2 * OFFSET_ADDRESS_SIZE;
6751 else
6752 break;
6753 }
6754
6755 p1 = p + mcnt;
6756 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
6757 to the `maybe_finalize_jump' of this case. Examine what
6758 follows. */
6759
6760 /* If we're at the end of the pattern, we can change. */
6761 if (p2 == pend)
6762 {
6763 /* Consider what happens when matching ":\(.*\)"
6764 against ":/". I don't really understand this code
6765 yet. */
6766 p[-(1+OFFSET_ADDRESS_SIZE)] = (US_CHAR_TYPE)
6767 pop_failure_jump;
6768 DEBUG_PRINT1
6769 (" End of pattern: change to `pop_failure_jump'.\n");
6770 }
6771
6772 else if ((re_opcode_t) *p2 == exactn
6773 #ifdef MBS_SUPPORT
6774 || (re_opcode_t) *p2 == exactn_bin
6775 #endif
6776 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
6777 {
6778 register US_CHAR_TYPE c
6779 = *p2 == (US_CHAR_TYPE) endline ? '\n' : p2[2];
6780
6781 if (((re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn
6782 #ifdef MBS_SUPPORT
6783 || (re_opcode_t) p1[1+OFFSET_ADDRESS_SIZE] == exactn_bin
6784 #endif
6785 ) && p1[3+OFFSET_ADDRESS_SIZE] != c)
6786 {
6787 p[-(1+OFFSET_ADDRESS_SIZE)] = (US_CHAR_TYPE)
6788 pop_failure_jump;
6789 #ifdef MBS_SUPPORT
6790 if (MB_CUR_MAX != 1)
6791 DEBUG_PRINT3 (" %C != %C => pop_failure_jump.\n",
6792 (wint_t) c,
6793 (wint_t) p1[3+OFFSET_ADDRESS_SIZE]);
6794 else
6795 #endif
6796 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
6797 (char) c,
6798 (char) p1[3+OFFSET_ADDRESS_SIZE]);
6799 }
6800
6801 #ifndef MBS_SUPPORT
6802 else if ((re_opcode_t) p1[3] == charset
6803 || (re_opcode_t) p1[3] == charset_not)
6804 {
6805 int not = (re_opcode_t) p1[3] == charset_not;
6806
6807 if (c < (unsigned) (p1[4] * BYTEWIDTH)
6808 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
6809 not = !not;
6810
6811 /* `not' is equal to 1 if c would match, which means
6812 that we can't change to pop_failure_jump. */
6813 if (!not)
6814 {
6815 p[-3] = (unsigned char) pop_failure_jump;
6816 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
6817 }
6818 }
6819 #endif /* not MBS_SUPPORT */
6820 }
6821 #ifndef MBS_SUPPORT
6822 else if ((re_opcode_t) *p2 == charset)
6823 {
6824 /* We win if the first character of the loop is not part
6825 of the charset. */
6826 if ((re_opcode_t) p1[3] == exactn
6827 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[5]
6828 && (p2[2 + p1[5] / BYTEWIDTH]
6829 & (1 << (p1[5] % BYTEWIDTH)))))
6830 {
6831 p[-3] = (unsigned char) pop_failure_jump;
6832 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
6833 }
6834
6835 else if ((re_opcode_t) p1[3] == charset_not)
6836 {
6837 int idx;
6838 /* We win if the charset_not inside the loop
6839 lists every character listed in the charset after. */
6840 for (idx = 0; idx < (int) p2[1]; idx++)
6841 if (! (p2[2 + idx] == 0
6842 || (idx < (int) p1[4]
6843 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
6844 break;
6845
6846 if (idx == p2[1])
6847 {
6848 p[-3] = (unsigned char) pop_failure_jump;
6849 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
6850 }
6851 }
6852 else if ((re_opcode_t) p1[3] == charset)
6853 {
6854 int idx;
6855 /* We win if the charset inside the loop
6856 has no overlap with the one after the loop. */
6857 for (idx = 0;
6858 idx < (int) p2[1] && idx < (int) p1[4];
6859 idx++)
6860 if ((p2[2 + idx] & p1[5 + idx]) != 0)
6861 break;
6862
6863 if (idx == p2[1] || idx == p1[4])
6864 {
6865 p[-3] = (unsigned char) pop_failure_jump;
6866 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
6867 }
6868 }
6869 }
6870 #endif /* not MBS_SUPPORT */
6871 }
6872 p -= OFFSET_ADDRESS_SIZE; /* Point at relative address again. */
6873 if ((re_opcode_t) p[-1] != pop_failure_jump)
6874 {
6875 p[-1] = (US_CHAR_TYPE) jump;
6876 DEBUG_PRINT1 (" Match => jump.\n");
6877 goto unconditional_jump;
6878 }
6879 /* Note fall through. */
6880
6881
6882 /* The end of a simple repeat has a pop_failure_jump back to
6883 its matching on_failure_jump, where the latter will push a
6884 failure point. The pop_failure_jump takes off failure
6885 points put on by this pop_failure_jump's matching
6886 on_failure_jump; we got through the pattern to here from the
6887 matching on_failure_jump, so didn't fail. */
6888 case pop_failure_jump:
6889 {
6890 /* We need to pass separate storage for the lowest and
6891 highest registers, even though we don't care about the
6892 actual values. Otherwise, we will restore only one
6893 register from the stack, since lowest will == highest in
6894 `pop_failure_point'. */
6895 active_reg_t dummy_low_reg, dummy_high_reg;
6896 US_CHAR_TYPE *pdummy = NULL;
6897 const CHAR_TYPE *sdummy = NULL;
6898
6899 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
6900 POP_FAILURE_POINT (sdummy, pdummy,
6901 dummy_low_reg, dummy_high_reg,
6902 reg_dummy, reg_dummy, reg_info_dummy);
6903 }
6904 /* Note fall through. */
6905
6906 unconditional_jump:
6907 #ifdef _LIBC
6908 DEBUG_PRINT2 ("\n%p: ", p);
6909 #else
6910 DEBUG_PRINT2 ("\n0x%x: ", p);
6911 #endif
6912 /* Note fall through. */
6913
6914 /* Unconditionally jump (without popping any failure points). */
6915 case jump:
6916 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
6917 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
6918 p += mcnt; /* Do the jump. */
6919 #ifdef _LIBC
6920 DEBUG_PRINT2 ("(to %p).\n", p);
6921 #else
6922 DEBUG_PRINT2 ("(to 0x%x).\n", p);
6923 #endif
6924 break;
6925
6926
6927 /* We need this opcode so we can detect where alternatives end
6928 in `group_match_null_string_p' et al. */
6929 case jump_past_alt:
6930 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
6931 goto unconditional_jump;
6932
6933
6934 /* Normally, the on_failure_jump pushes a failure point, which
6935 then gets popped at pop_failure_jump. We will end up at
6936 pop_failure_jump, also, and with a pattern of, say, `a+', we
6937 are skipping over the on_failure_jump, so we have to push
6938 something meaningless for pop_failure_jump to pop. */
6939 case dummy_failure_jump:
6940 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
6941 /* It doesn't matter what we push for the string here. What
6942 the code at `fail' tests is the value for the pattern. */
6943 PUSH_FAILURE_POINT (NULL, NULL, -2);
6944 goto unconditional_jump;
6945
6946
6947 /* At the end of an alternative, we need to push a dummy failure
6948 point in case we are followed by a `pop_failure_jump', because
6949 we don't want the failure point for the alternative to be
6950 popped. For example, matching `(a|ab)*' against `aab'
6951 requires that we match the `ab' alternative. */
6952 case push_dummy_failure:
6953 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
6954 /* See comments just above at `dummy_failure_jump' about the
6955 two zeroes. */
6956 PUSH_FAILURE_POINT (NULL, NULL, -2);
6957 break;
6958
6959 /* Have to succeed matching what follows at least n times.
6960 After that, handle like `on_failure_jump'. */
6961 case succeed_n:
6962 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
6963 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
6964
6965 assert (mcnt >= 0);
6966 /* Originally, this is how many times we HAVE to succeed. */
6967 if (mcnt > 0)
6968 {
6969 mcnt--;
6970 p += OFFSET_ADDRESS_SIZE;
6971 STORE_NUMBER_AND_INCR (p, mcnt);
6972 #ifdef _LIBC
6973 DEBUG_PRINT3 (" Setting %p to %d.\n", p - OFFSET_ADDRESS_SIZE
6974 , mcnt);
6975 #else
6976 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p - OFFSET_ADDRESS_SIZE
6977 , mcnt);
6978 #endif
6979 }
6980 else if (mcnt == 0)
6981 {
6982 #ifdef _LIBC
6983 DEBUG_PRINT2 (" Setting two bytes from %p to no_op.\n",
6984 p + OFFSET_ADDRESS_SIZE);
6985 #else
6986 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n",
6987 p + OFFSET_ADDRESS_SIZE);
6988 #endif /* _LIBC */
6989
6990 #ifdef MBS_SUPPORT
6991 p[1] = (US_CHAR_TYPE) no_op;
6992 #else
6993 p[2] = (US_CHAR_TYPE) no_op;
6994 p[3] = (US_CHAR_TYPE) no_op;
6995 #endif /* MBS_SUPPORT */
6996 goto on_failure;
6997 }
6998 break;
6999
7000 case jump_n:
7001 EXTRACT_NUMBER (mcnt, p + OFFSET_ADDRESS_SIZE);
7002 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
7003
7004 /* Originally, this is how many times we CAN jump. */
7005 if (mcnt)
7006 {
7007 mcnt--;
7008 STORE_NUMBER (p + OFFSET_ADDRESS_SIZE, mcnt);
7009
7010 #ifdef _LIBC
7011 DEBUG_PRINT3 (" Setting %p to %d.\n", p + OFFSET_ADDRESS_SIZE,
7012 mcnt);
7013 #else
7014 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p + OFFSET_ADDRESS_SIZE,
7015 mcnt);
7016 #endif /* _LIBC */
7017 goto unconditional_jump;
7018 }
7019 /* If don't have to jump any more, skip over the rest of command. */
7020 else
7021 p += 2 * OFFSET_ADDRESS_SIZE;
7022 break;
7023
7024 case set_number_at:
7025 {
7026 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
7027
7028 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7029 p1 = p + mcnt;
7030 EXTRACT_NUMBER_AND_INCR (mcnt, p);
7031 #ifdef _LIBC
7032 DEBUG_PRINT3 (" Setting %p to %d.\n", p1, mcnt);
7033 #else
7034 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
7035 #endif
7036 STORE_NUMBER (p1, mcnt);
7037 break;
7038 }
7039
7040 #if 0
7041 /* The DEC Alpha C compiler 3.x generates incorrect code for the
7042 test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of
7043 AT_WORD_BOUNDARY, so this code is disabled. Expanding the
7044 macro and introducing temporary variables works around the bug. */
7045
7046 case wordbound:
7047 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7048 if (AT_WORD_BOUNDARY (d))
7049 break;
7050 goto fail;
7051
7052 case notwordbound:
7053 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7054 if (AT_WORD_BOUNDARY (d))
7055 goto fail;
7056 break;
7057 #else
7058 case wordbound:
7059 {
7060 boolean prevchar, thischar;
7061
7062 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
7063 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7064 break;
7065
7066 prevchar = WORDCHAR_P (d - 1);
7067 thischar = WORDCHAR_P (d);
7068 if (prevchar != thischar)
7069 break;
7070 goto fail;
7071 }
7072
7073 case notwordbound:
7074 {
7075 boolean prevchar, thischar;
7076
7077 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
7078 if (AT_STRINGS_BEG (d) || AT_STRINGS_END (d))
7079 goto fail;
7080
7081 prevchar = WORDCHAR_P (d - 1);
7082 thischar = WORDCHAR_P (d);
7083 if (prevchar != thischar)
7084 goto fail;
7085 break;
7086 }
7087 #endif
7088
7089 case wordbeg:
7090 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
7091 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
7092 break;
7093 goto fail;
7094
7095 case wordend:
7096 DEBUG_PRINT1 ("EXECUTING wordend.\n");
7097 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
7098 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
7099 break;
7100 goto fail;
7101
7102 #ifdef emacs
7103 case before_dot:
7104 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
7105 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
7106 goto fail;
7107 break;
7108
7109 case at_dot:
7110 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
7111 if (PTR_CHAR_POS ((unsigned char *) d) != point)
7112 goto fail;
7113 break;
7114
7115 case after_dot:
7116 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
7117 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
7118 goto fail;
7119 break;
7120
7121 case syntaxspec:
7122 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
7123 mcnt = *p++;
7124 goto matchsyntax;
7125
7126 case wordchar:
7127 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
7128 mcnt = (int) Sword;
7129 matchsyntax:
7130 PREFETCH ();
7131 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7132 d++;
7133 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
7134 goto fail;
7135 SET_REGS_MATCHED ();
7136 break;
7137
7138 case notsyntaxspec:
7139 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
7140 mcnt = *p++;
7141 goto matchnotsyntax;
7142
7143 case notwordchar:
7144 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
7145 mcnt = (int) Sword;
7146 matchnotsyntax:
7147 PREFETCH ();
7148 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
7149 d++;
7150 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
7151 goto fail;
7152 SET_REGS_MATCHED ();
7153 break;
7154
7155 #else /* not emacs */
7156 case wordchar:
7157 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
7158 PREFETCH ();
7159 if (!WORDCHAR_P (d))
7160 goto fail;
7161 SET_REGS_MATCHED ();
7162 d++;
7163 break;
7164
7165 case notwordchar:
7166 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
7167 PREFETCH ();
7168 if (WORDCHAR_P (d))
7169 goto fail;
7170 SET_REGS_MATCHED ();
7171 d++;
7172 break;
7173 #endif /* not emacs */
7174
7175 default:
7176 abort ();
7177 }
7178 continue; /* Successfully executed one pattern command; keep going. */
7179
7180
7181 /* We goto here if a matching operation fails. */
7182 fail:
7183 if (!FAIL_STACK_EMPTY ())
7184 { /* A restart point is known. Restore to that state. */
7185 DEBUG_PRINT1 ("\nFAIL:\n");
7186 POP_FAILURE_POINT (d, p,
7187 lowest_active_reg, highest_active_reg,
7188 regstart, regend, reg_info);
7189
7190 /* If this failure point is a dummy, try the next one. */
7191 if (!p)
7192 goto fail;
7193
7194 /* If we failed to the end of the pattern, don't examine *p. */
7195 assert (p <= pend);
7196 if (p < pend)
7197 {
7198 boolean is_a_jump_n = false;
7199
7200 /* If failed to a backwards jump that's part of a repetition
7201 loop, need to pop this failure point and use the next one. */
7202 switch ((re_opcode_t) *p)
7203 {
7204 case jump_n:
7205 is_a_jump_n = true;
7206 case maybe_pop_jump:
7207 case pop_failure_jump:
7208 case jump:
7209 p1 = p + 1;
7210 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7211 p1 += mcnt;
7212
7213 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
7214 || (!is_a_jump_n
7215 && (re_opcode_t) *p1 == on_failure_jump))
7216 goto fail;
7217 break;
7218 default:
7219 /* do nothing */ ;
7220 }
7221 }
7222
7223 if (d >= string1 && d <= end1)
7224 dend = end_match_1;
7225 }
7226 else
7227 break; /* Matching at this starting point really fails. */
7228 } /* for (;;) */
7229
7230 if (best_regs_set)
7231 goto restore_best_regs;
7232
7233 FREE_VARIABLES ();
7234
7235 return -1; /* Failure to match. */
7236 } /* re_match_2 */
7237
7238 /* Subroutine definitions for re_match_2. */
7239
7240
7241 /* We are passed P pointing to a register number after a start_memory.
7242
7243 Return true if the pattern up to the corresponding stop_memory can
7244 match the empty string, and false otherwise.
7245
7246 If we find the matching stop_memory, sets P to point to one past its number.
7247 Otherwise, sets P to an undefined byte less than or equal to END.
7248
7249 We don't handle duplicates properly (yet). */
7250
7251 static boolean
group_match_null_string_p(p,end,reg_info)7252 group_match_null_string_p (p, end, reg_info)
7253 US_CHAR_TYPE **p, *end;
7254 register_info_type *reg_info;
7255 {
7256 int mcnt;
7257 /* Point to after the args to the start_memory. */
7258 US_CHAR_TYPE *p1 = *p + 2;
7259
7260 while (p1 < end)
7261 {
7262 /* Skip over opcodes that can match nothing, and return true or
7263 false, as appropriate, when we get to one that can't, or to the
7264 matching stop_memory. */
7265
7266 switch ((re_opcode_t) *p1)
7267 {
7268 /* Could be either a loop or a series of alternatives. */
7269 case on_failure_jump:
7270 p1++;
7271 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7272
7273 /* If the next operation is not a jump backwards in the
7274 pattern. */
7275
7276 if (mcnt >= 0)
7277 {
7278 /* Go through the on_failure_jumps of the alternatives,
7279 seeing if any of the alternatives cannot match nothing.
7280 The last alternative starts with only a jump,
7281 whereas the rest start with on_failure_jump and end
7282 with a jump, e.g., here is the pattern for `a|b|c':
7283
7284 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
7285 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
7286 /exactn/1/c
7287
7288 So, we have to first go through the first (n-1)
7289 alternatives and then deal with the last one separately. */
7290
7291
7292 /* Deal with the first (n-1) alternatives, which start
7293 with an on_failure_jump (see above) that jumps to right
7294 past a jump_past_alt. */
7295
7296 while ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] ==
7297 jump_past_alt)
7298 {
7299 /* `mcnt' holds how many bytes long the alternative
7300 is, including the ending `jump_past_alt' and
7301 its number. */
7302
7303 if (!alt_match_null_string_p (p1, p1 + mcnt -
7304 (1 + OFFSET_ADDRESS_SIZE),
7305 reg_info))
7306 return false;
7307
7308 /* Move to right after this alternative, including the
7309 jump_past_alt. */
7310 p1 += mcnt;
7311
7312 /* Break if it's the beginning of an n-th alternative
7313 that doesn't begin with an on_failure_jump. */
7314 if ((re_opcode_t) *p1 != on_failure_jump)
7315 break;
7316
7317 /* Still have to check that it's not an n-th
7318 alternative that starts with an on_failure_jump. */
7319 p1++;
7320 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7321 if ((re_opcode_t) p1[mcnt-(1+OFFSET_ADDRESS_SIZE)] !=
7322 jump_past_alt)
7323 {
7324 /* Get to the beginning of the n-th alternative. */
7325 p1 -= 1 + OFFSET_ADDRESS_SIZE;
7326 break;
7327 }
7328 }
7329
7330 /* Deal with the last alternative: go back and get number
7331 of the `jump_past_alt' just before it. `mcnt' contains
7332 the length of the alternative. */
7333 EXTRACT_NUMBER (mcnt, p1 - OFFSET_ADDRESS_SIZE);
7334
7335 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
7336 return false;
7337
7338 p1 += mcnt; /* Get past the n-th alternative. */
7339 } /* if mcnt > 0 */
7340 break;
7341
7342
7343 case stop_memory:
7344 assert (p1[1] == **p);
7345 *p = p1 + 2;
7346 return true;
7347
7348
7349 default:
7350 if (!common_op_match_null_string_p (&p1, end, reg_info))
7351 return false;
7352 }
7353 } /* while p1 < end */
7354
7355 return false;
7356 } /* group_match_null_string_p */
7357
7358
7359 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
7360 It expects P to be the first byte of a single alternative and END one
7361 byte past the last. The alternative can contain groups. */
7362
7363 static boolean
alt_match_null_string_p(p,end,reg_info)7364 alt_match_null_string_p (p, end, reg_info)
7365 US_CHAR_TYPE *p, *end;
7366 register_info_type *reg_info;
7367 {
7368 int mcnt;
7369 US_CHAR_TYPE *p1 = p;
7370
7371 while (p1 < end)
7372 {
7373 /* Skip over opcodes that can match nothing, and break when we get
7374 to one that can't. */
7375
7376 switch ((re_opcode_t) *p1)
7377 {
7378 /* It's a loop. */
7379 case on_failure_jump:
7380 p1++;
7381 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7382 p1 += mcnt;
7383 break;
7384
7385 default:
7386 if (!common_op_match_null_string_p (&p1, end, reg_info))
7387 return false;
7388 }
7389 } /* while p1 < end */
7390
7391 return true;
7392 } /* alt_match_null_string_p */
7393
7394
7395 /* Deals with the ops common to group_match_null_string_p and
7396 alt_match_null_string_p.
7397
7398 Sets P to one after the op and its arguments, if any. */
7399
7400 static boolean
common_op_match_null_string_p(p,end,reg_info)7401 common_op_match_null_string_p (p, end, reg_info)
7402 US_CHAR_TYPE **p, *end;
7403 register_info_type *reg_info;
7404 {
7405 int mcnt;
7406 boolean ret;
7407 int reg_no;
7408 US_CHAR_TYPE *p1 = *p;
7409
7410 switch ((re_opcode_t) *p1++)
7411 {
7412 case no_op:
7413 case begline:
7414 case endline:
7415 case begbuf:
7416 case endbuf:
7417 case wordbeg:
7418 case wordend:
7419 case wordbound:
7420 case notwordbound:
7421 #ifdef emacs
7422 case before_dot:
7423 case at_dot:
7424 case after_dot:
7425 #endif
7426 break;
7427
7428 case start_memory:
7429 reg_no = *p1;
7430 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
7431 ret = group_match_null_string_p (&p1, end, reg_info);
7432
7433 /* Have to set this here in case we're checking a group which
7434 contains a group and a back reference to it. */
7435
7436 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
7437 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
7438
7439 if (!ret)
7440 return false;
7441 break;
7442
7443 /* If this is an optimized succeed_n for zero times, make the jump. */
7444 case jump:
7445 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7446 if (mcnt >= 0)
7447 p1 += mcnt;
7448 else
7449 return false;
7450 break;
7451
7452 case succeed_n:
7453 /* Get to the number of times to succeed. */
7454 p1 += OFFSET_ADDRESS_SIZE;
7455 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7456
7457 if (mcnt == 0)
7458 {
7459 p1 -= 2 * OFFSET_ADDRESS_SIZE;
7460 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
7461 p1 += mcnt;
7462 }
7463 else
7464 return false;
7465 break;
7466
7467 case duplicate:
7468 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
7469 return false;
7470 break;
7471
7472 case set_number_at:
7473 p1 += 2 * OFFSET_ADDRESS_SIZE;
7474
7475 default:
7476 /* All other opcodes mean we cannot match the empty string. */
7477 return false;
7478 }
7479
7480 *p = p1;
7481 return true;
7482 } /* common_op_match_null_string_p */
7483
7484
7485 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
7486 bytes; nonzero otherwise. */
7487
7488 static int
bcmp_translate(s1,s2,len,translate)7489 bcmp_translate (s1, s2, len, translate)
7490 const CHAR_TYPE *s1, *s2;
7491 register int len;
7492 RE_TRANSLATE_TYPE translate;
7493 {
7494 register const US_CHAR_TYPE *p1 = (const US_CHAR_TYPE *) s1;
7495 register const US_CHAR_TYPE *p2 = (const US_CHAR_TYPE *) s2;
7496 while (len)
7497 {
7498 #ifdef MBS_SUPPORT
7499 if (((*p1<=0xff)?translate[*p1++]:*p1++)
7500 != ((*p2<=0xff)?translate[*p2++]:*p2++))
7501 return 1;
7502 #else
7503 if (translate[*p1++] != translate[*p2++]) return 1;
7504 #endif /* MBS_SUPPORT */
7505 len--;
7506 }
7507 return 0;
7508 }
7509
7510 /* Entry points for GNU code. */
7511
7512 /* re_compile_pattern is the GNU regular expression compiler: it
7513 compiles PATTERN (of length SIZE) and puts the result in BUFP.
7514 Returns 0 if the pattern was valid, otherwise an error string.
7515
7516 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
7517 are set in BUFP on entry.
7518
7519 We call regex_compile to do the actual compilation. */
7520
7521 const char *
re_compile_pattern(pattern,length,bufp)7522 re_compile_pattern (pattern, length, bufp)
7523 const char *pattern;
7524 size_t length;
7525 struct re_pattern_buffer *bufp;
7526 {
7527 reg_errcode_t ret;
7528
7529 /* GNU code is written to assume at least RE_NREGS registers will be set
7530 (and at least one extra will be -1). */
7531 bufp->regs_allocated = REGS_UNALLOCATED;
7532
7533 /* And GNU code determines whether or not to get register information
7534 by passing null for the REGS argument to re_match, etc., not by
7535 setting no_sub. */
7536 bufp->no_sub = 0;
7537
7538 /* Match anchors at newline. */
7539 bufp->newline_anchor = 1;
7540
7541 ret = regex_compile (pattern, length, re_syntax_options, bufp);
7542
7543 if (!ret)
7544 return NULL;
7545 return gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7546 }
7547 #ifdef _LIBC
7548 weak_alias (__re_compile_pattern, re_compile_pattern)
7549 #endif
7550
7551 /* Entry points compatible with 4.2 BSD regex library. We don't define
7552 them unless specifically requested. */
7553
7554 #if defined _REGEX_RE_COMP || defined _LIBC
7555
7556 /* BSD has one and only one pattern buffer. */
7557 static struct re_pattern_buffer re_comp_buf;
7558
7559 char *
7560 #ifdef _LIBC
7561 /* Make these definitions weak in libc, so POSIX programs can redefine
7562 these names if they don't use our functions, and still use
7563 regcomp/regexec below without link errors. */
7564 weak_function
7565 #endif
re_comp(s)7566 re_comp (s)
7567 const char *s;
7568 {
7569 reg_errcode_t ret;
7570
7571 if (!s)
7572 {
7573 if (!re_comp_buf.buffer)
7574 return gettext ("No previous regular expression");
7575 return 0;
7576 }
7577
7578 if (!re_comp_buf.buffer)
7579 {
7580 re_comp_buf.buffer = (unsigned char *) malloc (200);
7581 if (re_comp_buf.buffer == NULL)
7582 return (char *) gettext (re_error_msgid
7583 + re_error_msgid_idx[(int) REG_ESPACE]);
7584 re_comp_buf.allocated = 200;
7585
7586 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
7587 if (re_comp_buf.fastmap == NULL)
7588 return (char *) gettext (re_error_msgid
7589 + re_error_msgid_idx[(int) REG_ESPACE]);
7590 }
7591
7592 /* Since `re_exec' always passes NULL for the `regs' argument, we
7593 don't need to initialize the pattern buffer fields which affect it. */
7594
7595 /* Match anchors at newlines. */
7596 re_comp_buf.newline_anchor = 1;
7597
7598 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
7599
7600 if (!ret)
7601 return NULL;
7602
7603 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
7604 return (char *) gettext (re_error_msgid + re_error_msgid_idx[(int) ret]);
7605 }
7606
7607
7608 int
7609 #ifdef _LIBC
7610 weak_function
7611 #endif
re_exec(s)7612 re_exec (s)
7613 const char *s;
7614 {
7615 const int len = strlen (s);
7616 return
7617 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
7618 }
7619
7620 #endif /* _REGEX_RE_COMP */
7621
7622 /* POSIX.2 functions. Don't define these for Emacs. */
7623
7624 #ifndef emacs
7625
7626 /* regcomp takes a regular expression as a string and compiles it.
7627
7628 PREG is a regex_t *. We do not expect any fields to be initialized,
7629 since POSIX says we shouldn't. Thus, we set
7630
7631 `buffer' to the compiled pattern;
7632 `used' to the length of the compiled pattern;
7633 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
7634 REG_EXTENDED bit in CFLAGS is set; otherwise, to
7635 RE_SYNTAX_POSIX_BASIC;
7636 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
7637 `fastmap' to an allocated space for the fastmap;
7638 `fastmap_accurate' to zero;
7639 `re_nsub' to the number of subexpressions in PATTERN.
7640
7641 PATTERN is the address of the pattern string.
7642
7643 CFLAGS is a series of bits which affect compilation.
7644
7645 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
7646 use POSIX basic syntax.
7647
7648 If REG_NEWLINE is set, then . and [^...] don't match newline.
7649 Also, regexec will try a match beginning after every newline.
7650
7651 If REG_ICASE is set, then we considers upper- and lowercase
7652 versions of letters to be equivalent when matching.
7653
7654 If REG_NOSUB is set, then when PREG is passed to regexec, that
7655 routine will report only success or failure, and nothing about the
7656 registers.
7657
7658 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
7659 the return codes and their meanings.) */
7660
7661 int
regcomp(preg,pattern,cflags)7662 regcomp (preg, pattern, cflags)
7663 regex_t *preg;
7664 const char *pattern;
7665 int cflags;
7666 {
7667 reg_errcode_t ret;
7668 reg_syntax_t syntax
7669 = (cflags & REG_EXTENDED) ?
7670 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
7671
7672 /* regex_compile will allocate the space for the compiled pattern. */
7673 preg->buffer = 0;
7674 preg->allocated = 0;
7675 preg->used = 0;
7676
7677 /* Try to allocate space for the fastmap. */
7678 preg->fastmap = (char *) malloc (1 << BYTEWIDTH);
7679
7680 if (cflags & REG_ICASE)
7681 {
7682 unsigned i;
7683
7684 preg->translate
7685 = (RE_TRANSLATE_TYPE) malloc (CHAR_SET_SIZE
7686 * sizeof (*(RE_TRANSLATE_TYPE)0));
7687 if (preg->translate == NULL)
7688 return (int) REG_ESPACE;
7689
7690 /* Map uppercase characters to corresponding lowercase ones. */
7691 for (i = 0; i < CHAR_SET_SIZE; i++)
7692 preg->translate[i] = ISUPPER (i) ? TOLOWER (i) : i;
7693 }
7694 else
7695 preg->translate = NULL;
7696
7697 /* If REG_NEWLINE is set, newlines are treated differently. */
7698 if (cflags & REG_NEWLINE)
7699 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
7700 syntax &= ~RE_DOT_NEWLINE;
7701 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
7702 /* It also changes the matching behavior. */
7703 preg->newline_anchor = 1;
7704 }
7705 else
7706 preg->newline_anchor = 0;
7707
7708 preg->no_sub = !!(cflags & REG_NOSUB);
7709
7710 /* POSIX says a null character in the pattern terminates it, so we
7711 can use strlen here in compiling the pattern. */
7712 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
7713
7714 /* POSIX doesn't distinguish between an unmatched open-group and an
7715 unmatched close-group: both are REG_EPAREN. */
7716 if (ret == REG_ERPAREN) ret = REG_EPAREN;
7717
7718 if (ret == REG_NOERROR && preg->fastmap)
7719 {
7720 /* Compute the fastmap now, since regexec cannot modify the pattern
7721 buffer. */
7722 if (re_compile_fastmap (preg) == -2)
7723 {
7724 /* Some error occurred while computing the fastmap, just forget
7725 about it. */
7726 free (preg->fastmap);
7727 preg->fastmap = NULL;
7728 }
7729 }
7730
7731 return (int) ret;
7732 }
7733 #ifdef _LIBC
7734 weak_alias (__regcomp, regcomp)
7735 #endif
7736
7737
7738 /* regexec searches for a given pattern, specified by PREG, in the
7739 string STRING.
7740
7741 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
7742 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
7743 least NMATCH elements, and we set them to the offsets of the
7744 corresponding matched substrings.
7745
7746 EFLAGS specifies `execution flags' which affect matching: if
7747 REG_NOTBOL is set, then ^ does not match at the beginning of the
7748 string; if REG_NOTEOL is set, then $ does not match at the end.
7749
7750 We return 0 if we find a match and REG_NOMATCH if not. */
7751
7752 int
7753 regexec (preg, string, nmatch, pmatch, eflags)
7754 const regex_t *preg;
7755 const char *string;
7756 size_t nmatch;
7757 regmatch_t pmatch[];
7758 int eflags;
7759 {
7760 int ret;
7761 struct re_registers regs;
7762 regex_t private_preg;
7763 int len = strlen (string);
7764 boolean want_reg_info = !preg->no_sub && nmatch > 0;
7765
7766 private_preg = *preg;
7767
7768 private_preg.not_bol = !!(eflags & REG_NOTBOL);
7769 private_preg.not_eol = !!(eflags & REG_NOTEOL);
7770
7771 /* The user has told us exactly how many registers to return
7772 information about, via `nmatch'. We have to pass that on to the
7773 matching routines. */
7774 private_preg.regs_allocated = REGS_FIXED;
7775
7776 if (want_reg_info)
7777 {
7778 regs.num_regs = nmatch;
7779 regs.start = TALLOC (nmatch * 2, regoff_t);
7780 if (regs.start == NULL)
7781 return (int) REG_NOMATCH;
7782 regs.end = regs.start + nmatch;
7783 }
7784
7785 /* Perform the searching operation. */
7786 ret = re_search (&private_preg, string, len,
7787 /* start: */ 0, /* range: */ len,
7788 want_reg_info ? ®s : (struct re_registers *) 0);
7789
7790 /* Copy the register information to the POSIX structure. */
7791 if (want_reg_info)
7792 {
7793 if (ret >= 0)
7794 {
7795 unsigned r;
7796
7797 for (r = 0; r < nmatch; r++)
7798 {
7799 pmatch[r].rm_so = regs.start[r];
7800 pmatch[r].rm_eo = regs.end[r];
7801 }
7802 }
7803
7804 /* If we needed the temporary register info, free the space now. */
7805 free (regs.start);
7806 }
7807
7808 /* We want zero return to mean success, unlike `re_search'. */
7809 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
7810 }
7811 #ifdef _LIBC
7812 weak_alias (__regexec, regexec)
7813 #endif
7814
7815
7816 /* Returns a message corresponding to an error code, ERRCODE, returned
7817 from either regcomp or regexec. We don't use PREG here. */
7818
7819 size_t
7820 regerror (errcode, preg, errbuf, errbuf_size)
7821 int errcode;
7822 const regex_t *preg;
7823 char *errbuf;
7824 size_t errbuf_size;
7825 {
7826 const char *msg;
7827 size_t msg_size;
7828
7829 if (errcode < 0
7830 || errcode >= (int) (sizeof (re_error_msgid_idx)
7831 / sizeof (re_error_msgid_idx[0])))
7832 /* Only error codes returned by the rest of the code should be passed
7833 to this routine. If we are given anything else, or if other regex
7834 code generates an invalid error code, then the program has a bug.
7835 Dump core so we can fix it. */
7836 abort ();
7837
7838 msg = gettext (re_error_msgid + re_error_msgid_idx[errcode]);
7839
7840 msg_size = strlen (msg) + 1; /* Includes the null. */
7841
7842 if (errbuf_size != 0)
7843 {
7844 if (msg_size > errbuf_size)
7845 {
7846 #if defined HAVE_MEMPCPY || defined _LIBC
7847 *((char *) __mempcpy (errbuf, msg, errbuf_size - 1)) = '\0';
7848 #else
7849 memcpy (errbuf, msg, errbuf_size - 1);
7850 errbuf[errbuf_size - 1] = 0;
7851 #endif
7852 }
7853 else
7854 memcpy (errbuf, msg, msg_size);
7855 }
7856
7857 return msg_size;
7858 }
7859 #ifdef _LIBC
7860 weak_alias (__regerror, regerror)
7861 #endif
7862
7863
7864 /* Free dynamically allocated space used by PREG. */
7865
7866 void
7867 regfree (preg)
7868 regex_t *preg;
7869 {
7870 if (preg->buffer != NULL)
7871 free (preg->buffer);
7872 preg->buffer = NULL;
7873
7874 preg->allocated = 0;
7875 preg->used = 0;
7876
7877 if (preg->fastmap != NULL)
7878 free (preg->fastmap);
7879 preg->fastmap = NULL;
7880 preg->fastmap_accurate = 0;
7881
7882 if (preg->translate != NULL)
7883 free (preg->translate);
7884 preg->translate = NULL;
7885 }
7886 #ifdef _LIBC
7887 weak_alias (__regfree, regfree)
7888 #endif
7889
7890 #endif /* not emacs */
7891