comparison src/regex.c @ 11843:3ab3cc97fadc

Updated from ../gpl2lgpl.sed /home/gd/gnu/lib/regex.c
author Roland McGrath <roland@gnu.org>
date Wed, 17 May 1995 20:54:49 +0000
parents 3795b80ba1de
children 620c7195b48f
comparison
equal deleted inserted replaced
11842:e8739cc8cd02 11843:3ab3cc97fadc
1 /* Extended regular expression matching and search library, 1 /* Extended regular expression matching and search library,
2 version 0.12. 2 version 0.12.
3 (Implements POSIX draft P10003.2/D11.2, except for 3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.) 4 internationalization features.)
5 5
6 Copyright (C) 1993, 1994 Free Software Foundation, Inc. 6 Copyright (C) 1993, 1994, 1995 Free Software Foundation, Inc.
7 7
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 2, or (at your option)
11 any later version.
12
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
17
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, write to the Free Software
20 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
21
22 /* AIX requires this to be the first thing in the file. */
23 #if defined (_AIX) && !defined (REGEX_MALLOC)
24 #pragma alloca
25 #endif
26
27 #define _GNU_SOURCE
28
29 #ifdef HAVE_CONFIG_H
30 #include <config.h>
31 #endif
32
33 /* We need this for `regex.h', and perhaps for the Emacs include files. */
34 #include <sys/types.h>
35
36 /* This is for other GNU distributions with internationalized messages.
37 The GNU C Library itself does not yet support such messages. */
38 #if HAVE_LIBINTL_H
39 # include <libintl.h>
40 #else
41 # define gettext(msgid) (msgid)
42 #endif
43
44 /* The `emacs' switch turns on certain matching commands
45 that make sense only in Emacs. */
46 #ifdef emacs
47
48 #include "lisp.h"
49 #include "buffer.h"
50 #include "syntax.h"
51
52 #else /* not emacs */
53
54 /* If we are not linking with Emacs proper,
55 we can't use the relocating allocator
56 even if config.h says that we can. */
57 #undef REL_ALLOC
58
59 #ifdef STDC_HEADERS
60 #include <stdlib.h>
61 #else
62 char *malloc ();
63 char *realloc ();
64 #endif
65
66 /* We used to test for `BSTRING' here, but only GCC and Emacs define
67 `BSTRING', as far as I know, and neither of them use this code. */
68 #ifndef INHIBIT_STRING_HEADER
69 #if HAVE_STRING_H || STDC_HEADERS
70 #include <string.h>
71 #ifndef bcmp
72 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
73 #endif
74 #ifndef bcopy
75 #define bcopy(s, d, n) memcpy ((d), (s), (n))
76 #endif
77 #ifndef bzero
78 #define bzero(s, n) memset ((s), 0, (n))
79 #endif
80 #else
81 #include <strings.h>
82 #endif
83 #endif
84
85 /* Define the syntax stuff for \<, \>, etc. */
86
87 /* This must be nonzero for the wordchar and notwordchar pattern
88 commands in re_match_2. */
89 #ifndef Sword
90 #define Sword 1
91 #endif
92
93 #ifdef SWITCH_ENUM_BUG
94 #define SWITCH_ENUM_CAST(x) ((int)(x))
95 #else
96 #define SWITCH_ENUM_CAST(x) (x)
97 #endif
98
99 #ifdef SYNTAX_TABLE
100
101 extern char *re_syntax_table;
102
103 #else /* not SYNTAX_TABLE */
104
105 /* How many characters in the character set. */
106 #define CHAR_SET_SIZE 256
107
108 static char re_syntax_table[CHAR_SET_SIZE];
109
110 static void
111 init_syntax_once ()
112 {
113 register int c;
114 static int done = 0;
115
116 if (done)
117 return;
118
119 bzero (re_syntax_table, sizeof re_syntax_table);
120
121 for (c = 'a'; c <= 'z'; c++)
122 re_syntax_table[c] = Sword;
123
124 for (c = 'A'; c <= 'Z'; c++)
125 re_syntax_table[c] = Sword;
126
127 for (c = '0'; c <= '9'; c++)
128 re_syntax_table[c] = Sword;
129
130 re_syntax_table['_'] = Sword;
131
132 done = 1;
133 }
134
135 #endif /* not SYNTAX_TABLE */
136
137 #define SYNTAX(c) re_syntax_table[c]
138
139 #endif /* not emacs */
140
141 /* Get the interface, including the syntax bits. */
142 #include "regex.h"
143
144 /* isalpha etc. are used for the character classes. */
145 #include <ctype.h>
146
147 /* Jim Meyering writes:
148
149 "... Some ctype macros are valid only for character codes that
150 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
151 using /bin/cc or gcc but without giving an ansi option). So, all
152 ctype uses should be through macros like ISPRINT... If
153 STDC_HEADERS is defined, then autoconf has verified that the ctype
154 macros don't need to be guarded with references to isascii. ...
155 Defining isascii to 1 should let any compiler worth its salt
156 eliminate the && through constant folding." */
157
158 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
159 #define ISASCII(c) 1
160 #else
161 #define ISASCII(c) isascii(c)
162 #endif
163
164 #ifdef isblank
165 #define ISBLANK(c) (ISASCII (c) && isblank (c))
166 #else
167 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
168 #endif
169 #ifdef isgraph
170 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
171 #else
172 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
173 #endif
174
175 #define ISPRINT(c) (ISASCII (c) && isprint (c))
176 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
177 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
178 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
179 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
180 #define ISLOWER(c) (ISASCII (c) && islower (c))
181 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
182 #define ISSPACE(c) (ISASCII (c) && isspace (c))
183 #define ISUPPER(c) (ISASCII (c) && isupper (c))
184 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
185
186 #ifndef NULL
187 #define NULL 0
188 #endif
189
190 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
191 since ours (we hope) works properly with all combinations of
192 machines, compilers, `char' and `unsigned char' argument types.
193 (Per Bothner suggested the basic approach.) */
194 #undef SIGN_EXTEND_CHAR
195 #if __STDC__
196 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
197 #else /* not __STDC__ */
198 /* As in Harbison and Steele. */
199 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
200 #endif
201
202 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
203 use `alloca' instead of `malloc'. This is because using malloc in
204 re_search* or re_match* could cause memory leaks when C-g is used in
205 Emacs; also, malloc is slower and causes storage fragmentation. On
206 the other hand, malloc is more portable, and easier to debug.
207
208 Because we sometimes use alloca, some routines have to be macros,
209 not functions -- `alloca'-allocated space disappears at the end of the
210 function it is called in. */
211
212 #ifdef REGEX_MALLOC
213
214 #define REGEX_ALLOCATE malloc
215 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
216 #define REGEX_FREE free
217
218 #else /* not REGEX_MALLOC */
219
220 /* Emacs already defines alloca, sometimes. */
221 #ifndef alloca
222
223 /* Make alloca work the best possible way. */
224 #ifdef __GNUC__
225 #define alloca __builtin_alloca
226 #else /* not __GNUC__ */
227 #if HAVE_ALLOCA_H
228 #include <alloca.h>
229 #else /* not __GNUC__ or HAVE_ALLOCA_H */
230 #ifndef _AIX /* Already did AIX, up at the top. */
231 char *alloca ();
232 #endif /* not _AIX */
233 #endif /* not HAVE_ALLOCA_H */
234 #endif /* not __GNUC__ */
235
236 #endif /* not alloca */
237
238 #define REGEX_ALLOCATE alloca
239
240 /* Assumes a `char *destination' variable. */
241 #define REGEX_REALLOCATE(source, osize, nsize) \
242 (destination = (char *) alloca (nsize), \
243 bcopy (source, destination, osize), \
244 destination)
245
246 /* No need to do anything to free, after alloca. */
247 #define REGEX_FREE(arg) (0)
248
249 #endif /* not REGEX_MALLOC */
250
251 /* Define how to allocate the failure stack. */
252
253 #ifdef REL_ALLOC
254 #define REGEX_ALLOCATE_STACK(size) \
255 r_alloc (&failure_stack_ptr, (size))
256 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
257 r_re_alloc (&failure_stack_ptr, (nsize))
258 #define REGEX_FREE_STACK(ptr) \
259 r_alloc_free (&failure_stack_ptr)
260
261 #else /* not REL_ALLOC */
262
263 #ifdef REGEX_MALLOC
264
265 #define REGEX_ALLOCATE_STACK malloc
266 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
267 #define REGEX_FREE_STACK free
268
269 #else /* not REGEX_MALLOC */
270
271 #define REGEX_ALLOCATE_STACK alloca
272
273 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
274 REGEX_REALLOCATE (source, osize, nsize)
275 /* No need to explicitly free anything. */
276 #define REGEX_FREE_STACK(arg)
277
278 #endif /* not REGEX_MALLOC */
279 #endif /* not REL_ALLOC */
280
281
282 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
283 `string1' or just past its end. This works if PTR is NULL, which is
284 a good thing. */
285 #define FIRST_STRING_P(ptr) \
286 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
287
288 /* (Re)Allocate N items of type T using malloc, or fail. */
289 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
290 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
291 #define RETALLOC_IF(addr, n, t) \
292 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
293 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
294
295 #define BYTEWIDTH 8 /* In bits. */
296
297 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
298
299 #undef MAX
300 #undef MIN
301 #define MAX(a, b) ((a) > (b) ? (a) : (b))
302 #define MIN(a, b) ((a) < (b) ? (a) : (b))
303
304 typedef char boolean;
305 #define false 0
306 #define true 1
307
308 static int re_match_2_internal ();
309
310 /* These are the command codes that appear in compiled regular
311 expressions. Some opcodes are followed by argument bytes. A
312 command code can specify any interpretation whatsoever for its
313 arguments. Zero bytes may appear in the compiled regular expression. */
314
315 typedef enum
316 {
317 no_op = 0,
318
319 /* Succeed right away--no more backtracking. */
320 succeed,
321
322 /* Followed by one byte giving n, then by n literal bytes. */
323 exactn,
324
325 /* Matches any (more or less) character. */
326 anychar,
327
328 /* Matches any one char belonging to specified set. First
329 following byte is number of bitmap bytes. Then come bytes
330 for a bitmap saying which chars are in. Bits in each byte
331 are ordered low-bit-first. A character is in the set if its
332 bit is 1. A character too large to have a bit in the map is
333 automatically not in the set. */
334 charset,
335
336 /* Same parameters as charset, but match any character that is
337 not one of those specified. */
338 charset_not,
339
340 /* Start remembering the text that is matched, for storing in a
341 register. Followed by one byte with the register number, in
342 the range 0 to one less than the pattern buffer's re_nsub
343 field. Then followed by one byte with the number of groups
344 inner to this one. (This last has to be part of the
345 start_memory only because we need it in the on_failure_jump
346 of re_match_2.) */
347 start_memory,
348
349 /* Stop remembering the text that is matched and store it in a
350 memory register. Followed by one byte with the register
351 number, in the range 0 to one less than `re_nsub' in the
352 pattern buffer, and one byte with the number of inner groups,
353 just like `start_memory'. (We need the number of inner
354 groups here because we don't have any easy way of finding the
355 corresponding start_memory when we're at a stop_memory.) */
356 stop_memory,
357
358 /* Match a duplicate of something remembered. Followed by one
359 byte containing the register number. */
360 duplicate,
361
362 /* Fail unless at beginning of line. */
363 begline,
364
365 /* Fail unless at end of line. */
366 endline,
367
368 /* Succeeds if at beginning of buffer (if emacs) or at beginning
369 of string to be matched (if not). */
370 begbuf,
371
372 /* Analogously, for end of buffer/string. */
373 endbuf,
374
375 /* Followed by two byte relative address to which to jump. */
376 jump,
377
378 /* Same as jump, but marks the end of an alternative. */
379 jump_past_alt,
380
381 /* Followed by two-byte relative address of place to resume at
382 in case of failure. */
383 on_failure_jump,
384
385 /* Like on_failure_jump, but pushes a placeholder instead of the
386 current string position when executed. */
387 on_failure_keep_string_jump,
388
389 /* Throw away latest failure point and then jump to following
390 two-byte relative address. */
391 pop_failure_jump,
392
393 /* Change to pop_failure_jump if know won't have to backtrack to
394 match; otherwise change to jump. This is used to jump
395 back to the beginning of a repeat. If what follows this jump
396 clearly won't match what the repeat does, such that we can be
397 sure that there is no use backtracking out of repetitions
398 already matched, then we change it to a pop_failure_jump.
399 Followed by two-byte address. */
400 maybe_pop_jump,
401
402 /* Jump to following two-byte address, and push a dummy failure
403 point. This failure point will be thrown away if an attempt
404 is made to use it for a failure. A `+' construct makes this
405 before the first repeat. Also used as an intermediary kind
406 of jump when compiling an alternative. */
407 dummy_failure_jump,
408
409 /* Push a dummy failure point and continue. Used at the end of
410 alternatives. */
411 push_dummy_failure,
412
413 /* Followed by two-byte relative address and two-byte number n.
414 After matching N times, jump to the address upon failure. */
415 succeed_n,
416
417 /* Followed by two-byte relative address, and two-byte number n.
418 Jump to the address N times, then fail. */
419 jump_n,
420
421 /* Set the following two-byte relative address to the
422 subsequent two-byte number. The address *includes* the two
423 bytes of number. */
424 set_number_at,
425
426 wordchar, /* Matches any word-constituent character. */
427 notwordchar, /* Matches any char that is not a word-constituent. */
428
429 wordbeg, /* Succeeds if at word beginning. */
430 wordend, /* Succeeds if at word end. */
431
432 wordbound, /* Succeeds if at a word boundary. */
433 notwordbound /* Succeeds if not at a word boundary. */
434
435 #ifdef emacs
436 ,before_dot, /* Succeeds if before point. */
437 at_dot, /* Succeeds if at point. */
438 after_dot, /* Succeeds if after point. */
439
440 /* Matches any character whose syntax is specified. Followed by
441 a byte which contains a syntax code, e.g., Sword. */
442 syntaxspec,
443
444 /* Matches any character whose syntax is not that specified. */
445 notsyntaxspec
446 #endif /* emacs */
447 } re_opcode_t;
448
449 /* Common operations on the compiled pattern. */
450
451 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
452
453 #define STORE_NUMBER(destination, number) \
454 do { \
455 (destination)[0] = (number) & 0377; \
456 (destination)[1] = (number) >> 8; \
457 } while (0)
458
459 /* Same as STORE_NUMBER, except increment DESTINATION to
460 the byte after where the number is stored. Therefore, DESTINATION
461 must be an lvalue. */
462
463 #define STORE_NUMBER_AND_INCR(destination, number) \
464 do { \
465 STORE_NUMBER (destination, number); \
466 (destination) += 2; \
467 } while (0)
468
469 /* Put into DESTINATION a number stored in two contiguous bytes starting
470 at SOURCE. */
471
472 #define EXTRACT_NUMBER(destination, source) \
473 do { \
474 (destination) = *(source) & 0377; \
475 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
476 } while (0)
477
478 #ifdef DEBUG
479 static void
480 extract_number (dest, source)
481 int *dest;
482 unsigned char *source;
483 {
484 int temp = SIGN_EXTEND_CHAR (*(source + 1));
485 *dest = *source & 0377;
486 *dest += temp << 8;
487 }
488
489 #ifndef EXTRACT_MACROS /* To debug the macros. */
490 #undef EXTRACT_NUMBER
491 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
492 #endif /* not EXTRACT_MACROS */
493
494 #endif /* DEBUG */
495
496 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
497 SOURCE must be an lvalue. */
498
499 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
500 do { \
501 EXTRACT_NUMBER (destination, source); \
502 (source) += 2; \
503 } while (0)
504
505 #ifdef DEBUG
506 static void
507 extract_number_and_incr (destination, source)
508 int *destination;
509 unsigned char **source;
510 {
511 extract_number (destination, *source);
512 *source += 2;
513 }
514
515 #ifndef EXTRACT_MACROS
516 #undef EXTRACT_NUMBER_AND_INCR
517 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
518 extract_number_and_incr (&dest, &src)
519 #endif /* not EXTRACT_MACROS */
520
521 #endif /* DEBUG */
522
523 /* If DEBUG is defined, Regex prints many voluminous messages about what
524 it is doing (if the variable `debug' is nonzero). If linked with the
525 main program in `iregex.c', you can enter patterns and strings
526 interactively. And if linked with the main program in `main.c' and
527 the other test files, you can run the already-written tests. */
528
529 #ifdef DEBUG
530
531 /* We use standard I/O for debugging. */
532 #include <stdio.h>
533
534 /* It is useful to test things that ``must'' be true when debugging. */
535 #include <assert.h>
536
537 static int debug = 0;
538
539 #define DEBUG_STATEMENT(e) e
540 #define DEBUG_PRINT1(x) if (debug) printf (x)
541 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
542 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
543 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
544 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
545 if (debug) print_partial_compiled_pattern (s, e)
546 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
547 if (debug) print_double_string (w, s1, sz1, s2, sz2)
548
549
550 /* Print the fastmap in human-readable form. */
551
552 void
553 print_fastmap (fastmap)
554 char *fastmap;
555 {
556 unsigned was_a_range = 0;
557 unsigned i = 0;
558
559 while (i < (1 << BYTEWIDTH))
560 {
561 if (fastmap[i++])
562 {
563 was_a_range = 0;
564 putchar (i - 1);
565 while (i < (1 << BYTEWIDTH) && fastmap[i])
566 {
567 was_a_range = 1;
568 i++;
569 }
570 if (was_a_range)
571 {
572 printf ("-");
573 putchar (i - 1);
574 }
575 }
576 }
577 putchar ('\n');
578 }
579
580
581 /* Print a compiled pattern string in human-readable form, starting at
582 the START pointer into it and ending just before the pointer END. */
583
584 void
585 print_partial_compiled_pattern (start, end)
586 unsigned char *start;
587 unsigned char *end;
588 {
589 int mcnt, mcnt2;
590 unsigned char *p = start;
591 unsigned char *pend = end;
592
593 if (start == NULL)
594 {
595 printf ("(null)\n");
596 return;
597 }
598
599 /* Loop over pattern commands. */
600 while (p < pend)
601 {
602 printf ("%d:\t", p - start);
603
604 switch ((re_opcode_t) *p++)
605 {
606 case no_op:
607 printf ("/no_op");
608 break;
609
610 case exactn:
611 mcnt = *p++;
612 printf ("/exactn/%d", mcnt);
613 do
614 {
615 putchar ('/');
616 putchar (*p++);
617 }
618 while (--mcnt);
619 break;
620
621 case start_memory:
622 mcnt = *p++;
623 printf ("/start_memory/%d/%d", mcnt, *p++);
624 break;
625
626 case stop_memory:
627 mcnt = *p++;
628 printf ("/stop_memory/%d/%d", mcnt, *p++);
629 break;
630
631 case duplicate:
632 printf ("/duplicate/%d", *p++);
633 break;
634
635 case anychar:
636 printf ("/anychar");
637 break;
638
639 case charset:
640 case charset_not:
641 {
642 register int c, last = -100;
643 register int in_range = 0;
644
645 printf ("/charset [%s",
646 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
647
648 assert (p + *p < pend);
649
650 for (c = 0; c < 256; c++)
651 if (c / 8 < *p
652 && (p[1 + (c/8)] & (1 << (c % 8))))
653 {
654 /* Are we starting a range? */
655 if (last + 1 == c && ! in_range)
656 {
657 putchar ('-');
658 in_range = 1;
659 }
660 /* Have we broken a range? */
661 else if (last + 1 != c && in_range)
662 {
663 putchar (last);
664 in_range = 0;
665 }
666
667 if (! in_range)
668 putchar (c);
669
670 last = c;
671 }
672
673 if (in_range)
674 putchar (last);
675
676 putchar (']');
677
678 p += 1 + *p;
679 }
680 break;
681
682 case begline:
683 printf ("/begline");
684 break;
685
686 case endline:
687 printf ("/endline");
688 break;
689
690 case on_failure_jump:
691 extract_number_and_incr (&mcnt, &p);
692 printf ("/on_failure_jump to %d", p + mcnt - start);
693 break;
694
695 case on_failure_keep_string_jump:
696 extract_number_and_incr (&mcnt, &p);
697 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
698 break;
699
700 case dummy_failure_jump:
701 extract_number_and_incr (&mcnt, &p);
702 printf ("/dummy_failure_jump to %d", p + mcnt - start);
703 break;
704
705 case push_dummy_failure:
706 printf ("/push_dummy_failure");
707 break;
708
709 case maybe_pop_jump:
710 extract_number_and_incr (&mcnt, &p);
711 printf ("/maybe_pop_jump to %d", p + mcnt - start);
712 break;
713
714 case pop_failure_jump:
715 extract_number_and_incr (&mcnt, &p);
716 printf ("/pop_failure_jump to %d", p + mcnt - start);
717 break;
718
719 case jump_past_alt:
720 extract_number_and_incr (&mcnt, &p);
721 printf ("/jump_past_alt to %d", p + mcnt - start);
722 break;
723
724 case jump:
725 extract_number_and_incr (&mcnt, &p);
726 printf ("/jump to %d", p + mcnt - start);
727 break;
728
729 case succeed_n:
730 extract_number_and_incr (&mcnt, &p);
731 extract_number_and_incr (&mcnt2, &p);
732 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
733 break;
734
735 case jump_n:
736 extract_number_and_incr (&mcnt, &p);
737 extract_number_and_incr (&mcnt2, &p);
738 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
739 break;
740
741 case set_number_at:
742 extract_number_and_incr (&mcnt, &p);
743 extract_number_and_incr (&mcnt2, &p);
744 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
745 break;
746
747 case wordbound:
748 printf ("/wordbound");
749 break;
750
751 case notwordbound:
752 printf ("/notwordbound");
753 break;
754
755 case wordbeg:
756 printf ("/wordbeg");
757 break;
758
759 case wordend:
760 printf ("/wordend");
761
762 #ifdef emacs
763 case before_dot:
764 printf ("/before_dot");
765 break;
766
767 case at_dot:
768 printf ("/at_dot");
769 break;
770
771 case after_dot:
772 printf ("/after_dot");
773 break;
774
775 case syntaxspec:
776 printf ("/syntaxspec");
777 mcnt = *p++;
778 printf ("/%d", mcnt);
779 break;
780
781 case notsyntaxspec:
782 printf ("/notsyntaxspec");
783 mcnt = *p++;
784 printf ("/%d", mcnt);
785 break;
786 #endif /* emacs */
787
788 case wordchar:
789 printf ("/wordchar");
790 break;
791
792 case notwordchar:
793 printf ("/notwordchar");
794 break;
795
796 case begbuf:
797 printf ("/begbuf");
798 break;
799
800 case endbuf:
801 printf ("/endbuf");
802 break;
803
804 default:
805 printf ("?%d", *(p-1));
806 }
807
808 putchar ('\n');
809 }
810
811 printf ("%d:\tend of pattern.\n", p - start);
812 }
813
814
815 void
816 print_compiled_pattern (bufp)
817 struct re_pattern_buffer *bufp;
818 {
819 unsigned char *buffer = bufp->buffer;
820
821 print_partial_compiled_pattern (buffer, buffer + bufp->used);
822 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
823
824 if (bufp->fastmap_accurate && bufp->fastmap)
825 {
826 printf ("fastmap: ");
827 print_fastmap (bufp->fastmap);
828 }
829
830 printf ("re_nsub: %d\t", bufp->re_nsub);
831 printf ("regs_alloc: %d\t", bufp->regs_allocated);
832 printf ("can_be_null: %d\t", bufp->can_be_null);
833 printf ("newline_anchor: %d\n", bufp->newline_anchor);
834 printf ("no_sub: %d\t", bufp->no_sub);
835 printf ("not_bol: %d\t", bufp->not_bol);
836 printf ("not_eol: %d\t", bufp->not_eol);
837 printf ("syntax: %d\n", bufp->syntax);
838 /* Perhaps we should print the translate table? */
839 }
840
841
842 void
843 print_double_string (where, string1, size1, string2, size2)
844 const char *where;
845 const char *string1;
846 const char *string2;
847 int size1;
848 int size2;
849 {
850 unsigned this_char;
851
852 if (where == NULL)
853 printf ("(null)");
854 else
855 {
856 if (FIRST_STRING_P (where))
857 {
858 for (this_char = where - string1; this_char < size1; this_char++)
859 putchar (string1[this_char]);
860
861 where = string2;
862 }
863
864 for (this_char = where - string2; this_char < size2; this_char++)
865 putchar (string2[this_char]);
866 }
867 }
868
869 #else /* not DEBUG */
870
871 #undef assert
872 #define assert(e)
873
874 #define DEBUG_STATEMENT(e)
875 #define DEBUG_PRINT1(x)
876 #define DEBUG_PRINT2(x1, x2)
877 #define DEBUG_PRINT3(x1, x2, x3)
878 #define DEBUG_PRINT4(x1, x2, x3, x4)
879 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
880 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
881
882 #endif /* not DEBUG */
883
884 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
885 also be assigned to arbitrarily: each pattern buffer stores its own
886 syntax, so it can be changed between regex compilations. */
887 /* This has no initializer because initialized variables in Emacs
888 become read-only after dumping. */
889 reg_syntax_t re_syntax_options;
890
891
892 /* Specify the precise syntax of regexps for compilation. This provides
893 for compatibility for various utilities which historically have
894 different, incompatible syntaxes.
895
896 The argument SYNTAX is a bit mask comprised of the various bits
897 defined in regex.h. We return the old syntax. */
898
899 reg_syntax_t
900 re_set_syntax (syntax)
901 reg_syntax_t syntax;
902 {
903 reg_syntax_t ret = re_syntax_options;
904
905 re_syntax_options = syntax;
906 return ret;
907 }
908
909 /* This table gives an error message for each of the error codes listed
910 in regex.h. Obviously the order here has to be same as there.
911 POSIX doesn't require that we do anything for REG_NOERROR,
912 but why not be nice? */
913
914 static const char *re_error_msgid[] =
915 { "Success", /* REG_NOERROR */
916 "No match", /* REG_NOMATCH */
917 "Invalid regular expression", /* REG_BADPAT */
918 "Invalid collation character", /* REG_ECOLLATE */
919 "Invalid character class name", /* REG_ECTYPE */
920 "Trailing backslash", /* REG_EESCAPE */
921 "Invalid back reference", /* REG_ESUBREG */
922 "Unmatched [ or [^", /* REG_EBRACK */
923 "Unmatched ( or \\(", /* REG_EPAREN */
924 "Unmatched \\{", /* REG_EBRACE */
925 "Invalid content of \\{\\}", /* REG_BADBR */
926 "Invalid range end", /* REG_ERANGE */
927 "Memory exhausted", /* REG_ESPACE */
928 "Invalid preceding regular expression", /* REG_BADRPT */
929 "Premature end of regular expression", /* REG_EEND */
930 "Regular expression too big", /* REG_ESIZE */
931 "Unmatched ) or \\)", /* REG_ERPAREN */
932 };
933
934 /* Avoiding alloca during matching, to placate r_alloc. */
935
936 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
937 searching and matching functions should not call alloca. On some
938 systems, alloca is implemented in terms of malloc, and if we're
939 using the relocating allocator routines, then malloc could cause a
940 relocation, which might (if the strings being searched are in the
941 ralloc heap) shift the data out from underneath the regexp
942 routines.
943
944 Here's another reason to avoid allocation: Emacs
945 processes input from X in a signal handler; processing X input may
946 call malloc; if input arrives while a matching routine is calling
947 malloc, then we're scrod. But Emacs can't just block input while
948 calling matching routines; then we don't notice interrupts when
949 they come in. So, Emacs blocks input around all regexp calls
950 except the matching calls, which it leaves unprotected, in the
951 faith that they will not malloc. */
952
953 /* Normally, this is fine. */
954 #define MATCH_MAY_ALLOCATE
955
956 /* When using GNU C, we are not REALLY using the C alloca, no matter
957 what config.h may say. So don't take precautions for it. */
958 #ifdef __GNUC__
959 #undef C_ALLOCA
960 #endif
961
962 /* The match routines may not allocate if (1) they would do it with malloc
963 and (2) it's not safe for them to use malloc.
964 Note that if REL_ALLOC is defined, matching would not use malloc for the
965 failure stack, but we would still use it for the register vectors;
966 so REL_ALLOC should not affect this. */
967 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
968 #undef MATCH_MAY_ALLOCATE
969 #endif
970
971
972 /* Failure stack declarations and macros; both re_compile_fastmap and
973 re_match_2 use a failure stack. These have to be macros because of
974 REGEX_ALLOCATE_STACK. */
975
976
977 /* Number of failure points for which to initially allocate space
978 when matching. If this number is exceeded, we allocate more
979 space, so it is not a hard limit. */
980 #ifndef INIT_FAILURE_ALLOC
981 #define INIT_FAILURE_ALLOC 5
982 #endif
983
984 /* Roughly the maximum number of failure points on the stack. Would be
985 exactly that if always used MAX_FAILURE_SPACE each time we failed.
986 This is a variable only so users of regex can assign to it; we never
987 change it ourselves. */
988 #if defined (MATCH_MAY_ALLOCATE)
989 int re_max_failures = 200000;
990 #else
991 int re_max_failures = 2000;
992 #endif
993
994 union fail_stack_elt
995 {
996 unsigned char *pointer;
997 int integer;
998 };
999
1000 typedef union fail_stack_elt fail_stack_elt_t;
1001
1002 typedef struct
1003 {
1004 fail_stack_elt_t *stack;
1005 unsigned size;
1006 unsigned avail; /* Offset of next open position. */
1007 } fail_stack_type;
1008
1009 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1010 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1011 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1012
1013
1014 /* Define macros to initialize and free the failure stack.
1015 Do `return -2' if the alloc fails. */
1016
1017 #ifdef MATCH_MAY_ALLOCATE
1018 #define INIT_FAIL_STACK() \
1019 do { \
1020 fail_stack.stack = (fail_stack_elt_t *) \
1021 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1022 \
1023 if (fail_stack.stack == NULL) \
1024 return -2; \
1025 \
1026 fail_stack.size = INIT_FAILURE_ALLOC; \
1027 fail_stack.avail = 0; \
1028 } while (0)
1029
1030 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1031 #else
1032 #define INIT_FAIL_STACK() \
1033 do { \
1034 fail_stack.avail = 0; \
1035 } while (0)
1036
1037 #define RESET_FAIL_STACK()
1038 #endif
1039
1040
1041 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1042
1043 Return 1 if succeeds, and 0 if either ran out of memory
1044 allocating space for it or it was already too large.
1045
1046 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1047
1048 #define DOUBLE_FAIL_STACK(fail_stack) \
1049 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1050 ? 0 \
1051 : ((fail_stack).stack = (fail_stack_elt_t *) \
1052 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1053 (fail_stack).size * sizeof (fail_stack_elt_t), \
1054 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1055 \
1056 (fail_stack).stack == NULL \
1057 ? 0 \
1058 : ((fail_stack).size <<= 1, \
1059 1)))
1060
1061
1062 /* Push pointer POINTER on FAIL_STACK.
1063 Return 1 if was able to do so and 0 if ran out of memory allocating
1064 space to do so. */
1065 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1066 ((FAIL_STACK_FULL () \
1067 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1068 ? 0 \
1069 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1070 1))
1071
1072 /* Push a pointer value onto the failure stack.
1073 Assumes the variable `fail_stack'. Probably should only
1074 be called from within `PUSH_FAILURE_POINT'. */
1075 #define PUSH_FAILURE_POINTER(item) \
1076 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1077
1078 /* This pushes an integer-valued item onto the failure stack.
1079 Assumes the variable `fail_stack'. Probably should only
1080 be called from within `PUSH_FAILURE_POINT'. */
1081 #define PUSH_FAILURE_INT(item) \
1082 fail_stack.stack[fail_stack.avail++].integer = (item)
1083
1084 /* Push a fail_stack_elt_t value onto the failure stack.
1085 Assumes the variable `fail_stack'. Probably should only
1086 be called from within `PUSH_FAILURE_POINT'. */
1087 #define PUSH_FAILURE_ELT(item) \
1088 fail_stack.stack[fail_stack.avail++] = (item)
1089
1090 /* These three POP... operations complement the three PUSH... operations.
1091 All assume that `fail_stack' is nonempty. */
1092 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1093 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1094 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1095
1096 /* Used to omit pushing failure point id's when we're not debugging. */
1097 #ifdef DEBUG
1098 #define DEBUG_PUSH PUSH_FAILURE_INT
1099 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1100 #else
1101 #define DEBUG_PUSH(item)
1102 #define DEBUG_POP(item_addr)
1103 #endif
1104
1105
1106 /* Push the information about the state we will need
1107 if we ever fail back to it.
1108
1109 Requires variables fail_stack, regstart, regend, reg_info, and
1110 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1111 declared.
1112
1113 Does `return FAILURE_CODE' if runs out of memory. */
1114
1115 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1116 do { \
1117 char *destination; \
1118 /* Must be int, so when we don't save any registers, the arithmetic \
1119 of 0 + -1 isn't done as unsigned. */ \
1120 int this_reg; \
1121 \
1122 DEBUG_STATEMENT (failure_id++); \
1123 DEBUG_STATEMENT (nfailure_points_pushed++); \
1124 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1125 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1126 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1127 \
1128 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1129 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1130 \
1131 /* Ensure we have enough space allocated for what we will push. */ \
1132 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1133 { \
1134 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1135 return failure_code; \
1136 \
1137 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1138 (fail_stack).size); \
1139 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1140 } \
1141 \
1142 /* Push the info, starting with the registers. */ \
1143 DEBUG_PRINT1 ("\n"); \
1144 \
1145 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1146 this_reg++) \
1147 { \
1148 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1149 DEBUG_STATEMENT (num_regs_pushed++); \
1150 \
1151 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1152 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1153 \
1154 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1155 PUSH_FAILURE_POINTER (regend[this_reg]); \
1156 \
1157 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1158 DEBUG_PRINT2 (" match_null=%d", \
1159 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1160 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1161 DEBUG_PRINT2 (" matched_something=%d", \
1162 MATCHED_SOMETHING (reg_info[this_reg])); \
1163 DEBUG_PRINT2 (" ever_matched=%d", \
1164 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1165 DEBUG_PRINT1 ("\n"); \
1166 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1167 } \
1168 \
1169 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1170 PUSH_FAILURE_INT (lowest_active_reg); \
1171 \
1172 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1173 PUSH_FAILURE_INT (highest_active_reg); \
1174 \
1175 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1176 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1177 PUSH_FAILURE_POINTER (pattern_place); \
1178 \
1179 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1180 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1181 size2); \
1182 DEBUG_PRINT1 ("'\n"); \
1183 PUSH_FAILURE_POINTER (string_place); \
1184 \
1185 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1186 DEBUG_PUSH (failure_id); \
1187 } while (0)
1188
1189 /* This is the number of items that are pushed and popped on the stack
1190 for each register. */
1191 #define NUM_REG_ITEMS 3
1192
1193 /* Individual items aside from the registers. */
1194 #ifdef DEBUG
1195 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1196 #else
1197 #define NUM_NONREG_ITEMS 4
1198 #endif
1199
1200 /* We push at most this many items on the stack. */
1201 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1202
1203 /* We actually push this many items. */
1204 #define NUM_FAILURE_ITEMS \
1205 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1206 + NUM_NONREG_ITEMS)
1207
1208 /* How many items can still be added to the stack without overflowing it. */
1209 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1210
1211
1212 /* Pops what PUSH_FAIL_STACK pushes.
1213
1214 We restore into the parameters, all of which should be lvalues:
1215 STR -- the saved data position.
1216 PAT -- the saved pattern position.
1217 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1218 REGSTART, REGEND -- arrays of string positions.
1219 REG_INFO -- array of information about each subexpression.
1220
1221 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1222 `pend', `string1', `size1', `string2', and `size2'. */
1223
1224 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1225 { \
1226 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1227 int this_reg; \
1228 const unsigned char *string_temp; \
1229 \
1230 assert (!FAIL_STACK_EMPTY ()); \
1231 \
1232 /* Remove failure points and point to how many regs pushed. */ \
1233 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1234 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1235 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1236 \
1237 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1238 \
1239 DEBUG_POP (&failure_id); \
1240 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1241 \
1242 /* If the saved string location is NULL, it came from an \
1243 on_failure_keep_string_jump opcode, and we want to throw away the \
1244 saved NULL, thus retaining our current position in the string. */ \
1245 string_temp = POP_FAILURE_POINTER (); \
1246 if (string_temp != NULL) \
1247 str = (const char *) string_temp; \
1248 \
1249 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1250 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1251 DEBUG_PRINT1 ("'\n"); \
1252 \
1253 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1254 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1255 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1256 \
1257 /* Restore register info. */ \
1258 high_reg = (unsigned) POP_FAILURE_INT (); \
1259 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1260 \
1261 low_reg = (unsigned) POP_FAILURE_INT (); \
1262 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1263 \
1264 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1265 { \
1266 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1267 \
1268 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1269 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1270 \
1271 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1272 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1273 \
1274 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1275 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1276 } \
1277 \
1278 set_regs_matched_done = 0; \
1279 DEBUG_STATEMENT (nfailure_points_popped++); \
1280 } /* POP_FAILURE_POINT */
1281
1282
1283
1284 /* Structure for per-register (a.k.a. per-group) information.
1285 Other register information, such as the
1286 starting and ending positions (which are addresses), and the list of
1287 inner groups (which is a bits list) are maintained in separate
1288 variables.
1289
1290 We are making a (strictly speaking) nonportable assumption here: that
1291 the compiler will pack our bit fields into something that fits into
1292 the type of `word', i.e., is something that fits into one item on the
1293 failure stack. */
1294
1295 typedef union
1296 {
1297 fail_stack_elt_t word;
1298 struct
1299 {
1300 /* This field is one if this group can match the empty string,
1301 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1302 #define MATCH_NULL_UNSET_VALUE 3
1303 unsigned match_null_string_p : 2;
1304 unsigned is_active : 1;
1305 unsigned matched_something : 1;
1306 unsigned ever_matched_something : 1;
1307 } bits;
1308 } register_info_type;
1309
1310 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1311 #define IS_ACTIVE(R) ((R).bits.is_active)
1312 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1313 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1314
1315
1316 /* Call this when have matched a real character; it sets `matched' flags
1317 for the subexpressions which we are currently inside. Also records
1318 that those subexprs have matched. */
1319 #define SET_REGS_MATCHED() \
1320 do \
1321 { \
1322 if (!set_regs_matched_done) \
1323 { \
1324 unsigned r; \
1325 set_regs_matched_done = 1; \
1326 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1327 { \
1328 MATCHED_SOMETHING (reg_info[r]) \
1329 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1330 = 1; \
1331 } \
1332 } \
1333 } \
1334 while (0)
1335
1336 /* Registers are set to a sentinel when they haven't yet matched. */
1337 static char reg_unset_dummy;
1338 #define REG_UNSET_VALUE (&reg_unset_dummy)
1339 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1340
1341 /* Subroutine declarations and macros for regex_compile. */
1342
1343 static void store_op1 (), store_op2 ();
1344 static void insert_op1 (), insert_op2 ();
1345 static boolean at_begline_loc_p (), at_endline_loc_p ();
1346 static boolean group_in_compile_stack ();
1347 static reg_errcode_t compile_range ();
1348
1349 /* Fetch the next character in the uncompiled pattern---translating it
1350 if necessary. Also cast from a signed character in the constant
1351 string passed to us by the user to an unsigned char that we can use
1352 as an array index (in, e.g., `translate'). */
1353 #define PATFETCH(c) \
1354 do {if (p == pend) return REG_EEND; \
1355 c = (unsigned char) *p++; \
1356 if (translate) c = translate[c]; \
1357 } while (0)
1358
1359 /* Fetch the next character in the uncompiled pattern, with no
1360 translation. */
1361 #define PATFETCH_RAW(c) \
1362 do {if (p == pend) return REG_EEND; \
1363 c = (unsigned char) *p++; \
1364 } while (0)
1365
1366 /* Go backwards one character in the pattern. */
1367 #define PATUNFETCH p--
1368
1369
1370 /* If `translate' is non-null, return translate[D], else just D. We
1371 cast the subscript to translate because some data is declared as
1372 `char *', to avoid warnings when a string constant is passed. But
1373 when we use a character as a subscript we must make it unsigned. */
1374 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1375
1376
1377 /* Macros for outputting the compiled pattern into `buffer'. */
1378
1379 /* If the buffer isn't allocated when it comes in, use this. */
1380 #define INIT_BUF_SIZE 32
1381
1382 /* Make sure we have at least N more bytes of space in buffer. */
1383 #define GET_BUFFER_SPACE(n) \
1384 while (b - bufp->buffer + (n) > bufp->allocated) \
1385 EXTEND_BUFFER ()
1386
1387 /* Make sure we have one more byte of buffer space and then add C to it. */
1388 #define BUF_PUSH(c) \
1389 do { \
1390 GET_BUFFER_SPACE (1); \
1391 *b++ = (unsigned char) (c); \
1392 } while (0)
1393
1394
1395 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1396 #define BUF_PUSH_2(c1, c2) \
1397 do { \
1398 GET_BUFFER_SPACE (2); \
1399 *b++ = (unsigned char) (c1); \
1400 *b++ = (unsigned char) (c2); \
1401 } while (0)
1402
1403
1404 /* As with BUF_PUSH_2, except for three bytes. */
1405 #define BUF_PUSH_3(c1, c2, c3) \
1406 do { \
1407 GET_BUFFER_SPACE (3); \
1408 *b++ = (unsigned char) (c1); \
1409 *b++ = (unsigned char) (c2); \
1410 *b++ = (unsigned char) (c3); \
1411 } while (0)
1412
1413
1414 /* Store a jump with opcode OP at LOC to location TO. We store a
1415 relative address offset by the three bytes the jump itself occupies. */
1416 #define STORE_JUMP(op, loc, to) \
1417 store_op1 (op, loc, (to) - (loc) - 3)
1418
1419 /* Likewise, for a two-argument jump. */
1420 #define STORE_JUMP2(op, loc, to, arg) \
1421 store_op2 (op, loc, (to) - (loc) - 3, arg)
1422
1423 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1424 #define INSERT_JUMP(op, loc, to) \
1425 insert_op1 (op, loc, (to) - (loc) - 3, b)
1426
1427 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1428 #define INSERT_JUMP2(op, loc, to, arg) \
1429 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1430
1431
1432 /* This is not an arbitrary limit: the arguments which represent offsets
1433 into the pattern are two bytes long. So if 2^16 bytes turns out to
1434 be too small, many things would have to change. */
1435 #define MAX_BUF_SIZE (1L << 16)
1436
1437
1438 /* Extend the buffer by twice its current size via realloc and
1439 reset the pointers that pointed into the old block to point to the
1440 correct places in the new one. If extending the buffer results in it
1441 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1442 #define EXTEND_BUFFER() \
1443 do { \
1444 unsigned char *old_buffer = bufp->buffer; \
1445 if (bufp->allocated == MAX_BUF_SIZE) \
1446 return REG_ESIZE; \
1447 bufp->allocated <<= 1; \
1448 if (bufp->allocated > MAX_BUF_SIZE) \
1449 bufp->allocated = MAX_BUF_SIZE; \
1450 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1451 if (bufp->buffer == NULL) \
1452 return REG_ESPACE; \
1453 /* If the buffer moved, move all the pointers into it. */ \
1454 if (old_buffer != bufp->buffer) \
1455 { \
1456 b = (b - old_buffer) + bufp->buffer; \
1457 begalt = (begalt - old_buffer) + bufp->buffer; \
1458 if (fixup_alt_jump) \
1459 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1460 if (laststart) \
1461 laststart = (laststart - old_buffer) + bufp->buffer; \
1462 if (pending_exact) \
1463 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1464 } \
1465 } while (0)
1466
1467
1468 /* Since we have one byte reserved for the register number argument to
1469 {start,stop}_memory, the maximum number of groups we can report
1470 things about is what fits in that byte. */
1471 #define MAX_REGNUM 255
1472
1473 /* But patterns can have more than `MAX_REGNUM' registers. We just
1474 ignore the excess. */
1475 typedef unsigned regnum_t;
1476
1477
1478 /* Macros for the compile stack. */
1479
1480 /* Since offsets can go either forwards or backwards, this type needs to
1481 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1482 typedef int pattern_offset_t;
1483
1484 typedef struct
1485 {
1486 pattern_offset_t begalt_offset;
1487 pattern_offset_t fixup_alt_jump;
1488 pattern_offset_t inner_group_offset;
1489 pattern_offset_t laststart_offset;
1490 regnum_t regnum;
1491 } compile_stack_elt_t;
1492
1493
1494 typedef struct
1495 {
1496 compile_stack_elt_t *stack;
1497 unsigned size;
1498 unsigned avail; /* Offset of next open position. */
1499 } compile_stack_type;
1500
1501
1502 #define INIT_COMPILE_STACK_SIZE 32
1503
1504 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1505 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1506
1507 /* The next available element. */
1508 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1509
1510
1511 /* Set the bit for character C in a list. */
1512 #define SET_LIST_BIT(c) \
1513 (b[((unsigned char) (c)) / BYTEWIDTH] \
1514 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1515
1516
1517 /* Get the next unsigned number in the uncompiled pattern. */
1518 #define GET_UNSIGNED_NUMBER(num) \
1519 { if (p != pend) \
1520 { \
1521 PATFETCH (c); \
1522 while (ISDIGIT (c)) \
1523 { \
1524 if (num < 0) \
1525 num = 0; \
1526 num = num * 10 + c - '0'; \
1527 if (p == pend) \
1528 break; \
1529 PATFETCH (c); \
1530 } \
1531 } \
1532 }
1533
1534 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1535
1536 #define IS_CHAR_CLASS(string) \
1537 (STREQ (string, "alpha") || STREQ (string, "upper") \
1538 || STREQ (string, "lower") || STREQ (string, "digit") \
1539 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1540 || STREQ (string, "space") || STREQ (string, "print") \
1541 || STREQ (string, "punct") || STREQ (string, "graph") \
1542 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1543
1544 #ifndef MATCH_MAY_ALLOCATE
1545
1546 /* If we cannot allocate large objects within re_match_2_internal,
1547 we make the fail stack and register vectors global.
1548 The fail stack, we grow to the maximum size when a regexp
1549 is compiled.
1550 The register vectors, we adjust in size each time we
1551 compile a regexp, according to the number of registers it needs. */
1552
1553 static fail_stack_type fail_stack;
1554
1555 /* Size with which the following vectors are currently allocated.
1556 That is so we can make them bigger as needed,
1557 but never make them smaller. */
1558 static int regs_allocated_size;
1559
1560 static const char ** regstart, ** regend;
1561 static const char ** old_regstart, ** old_regend;
1562 static const char **best_regstart, **best_regend;
1563 static register_info_type *reg_info;
1564 static const char **reg_dummy;
1565 static register_info_type *reg_info_dummy;
1566
1567 /* Make the register vectors big enough for NUM_REGS registers,
1568 but don't make them smaller. */
1569
1570 static
1571 regex_grow_registers (num_regs)
1572 int num_regs;
1573 {
1574 if (num_regs > regs_allocated_size)
1575 {
1576 RETALLOC_IF (regstart, num_regs, const char *);
1577 RETALLOC_IF (regend, num_regs, const char *);
1578 RETALLOC_IF (old_regstart, num_regs, const char *);
1579 RETALLOC_IF (old_regend, num_regs, const char *);
1580 RETALLOC_IF (best_regstart, num_regs, const char *);
1581 RETALLOC_IF (best_regend, num_regs, const char *);
1582 RETALLOC_IF (reg_info, num_regs, register_info_type);
1583 RETALLOC_IF (reg_dummy, num_regs, const char *);
1584 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1585
1586 regs_allocated_size = num_regs;
1587 }
1588 }
1589
1590 #endif /* not MATCH_MAY_ALLOCATE */
1591
1592 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1593 Returns one of error codes defined in `regex.h', or zero for success.
1594
1595 Assumes the `allocated' (and perhaps `buffer') and `translate'
1596 fields are set in BUFP on entry.
1597
1598 If it succeeds, results are put in BUFP (if it returns an error, the
1599 contents of BUFP are undefined):
1600 `buffer' is the compiled pattern;
1601 `syntax' is set to SYNTAX;
1602 `used' is set to the length of the compiled pattern;
1603 `fastmap_accurate' is zero;
1604 `re_nsub' is the number of subexpressions in PATTERN;
1605 `not_bol' and `not_eol' are zero;
1606
1607 The `fastmap' and `newline_anchor' fields are neither
1608 examined nor set. */
1609
1610 /* Return, freeing storage we allocated. */
1611 #define FREE_STACK_RETURN(value) \
1612 return (free (compile_stack.stack), value)
1613
1614 static reg_errcode_t
1615 regex_compile (pattern, size, syntax, bufp)
1616 const char *pattern;
1617 int size;
1618 reg_syntax_t syntax;
1619 struct re_pattern_buffer *bufp;
1620 {
1621 /* We fetch characters from PATTERN here. Even though PATTERN is
1622 `char *' (i.e., signed), we declare these variables as unsigned, so
1623 they can be reliably used as array indices. */
1624 register unsigned char c, c1;
1625
1626 /* A random temporary spot in PATTERN. */
1627 const char *p1;
1628
1629 /* Points to the end of the buffer, where we should append. */
1630 register unsigned char *b;
1631
1632 /* Keeps track of unclosed groups. */
1633 compile_stack_type compile_stack;
1634
1635 /* Points to the current (ending) position in the pattern. */
1636 const char *p = pattern;
1637 const char *pend = pattern + size;
1638
1639 /* How to translate the characters in the pattern. */
1640 char *translate = bufp->translate;
1641
1642 /* Address of the count-byte of the most recently inserted `exactn'
1643 command. This makes it possible to tell if a new exact-match
1644 character can be added to that command or if the character requires
1645 a new `exactn' command. */
1646 unsigned char *pending_exact = 0;
1647
1648 /* Address of start of the most recently finished expression.
1649 This tells, e.g., postfix * where to find the start of its
1650 operand. Reset at the beginning of groups and alternatives. */
1651 unsigned char *laststart = 0;
1652
1653 /* Address of beginning of regexp, or inside of last group. */
1654 unsigned char *begalt;
1655
1656 /* Place in the uncompiled pattern (i.e., the {) to
1657 which to go back if the interval is invalid. */
1658 const char *beg_interval;
1659
1660 /* Address of the place where a forward jump should go to the end of
1661 the containing expression. Each alternative of an `or' -- except the
1662 last -- ends with a forward jump of this sort. */
1663 unsigned char *fixup_alt_jump = 0;
1664
1665 /* Counts open-groups as they are encountered. Remembered for the
1666 matching close-group on the compile stack, so the same register
1667 number is put in the stop_memory as the start_memory. */
1668 regnum_t regnum = 0;
1669
1670 #ifdef DEBUG
1671 DEBUG_PRINT1 ("\nCompiling pattern: ");
1672 if (debug)
1673 {
1674 unsigned debug_count;
1675
1676 for (debug_count = 0; debug_count < size; debug_count++)
1677 putchar (pattern[debug_count]);
1678 putchar ('\n');
1679 }
1680 #endif /* DEBUG */
1681
1682 /* Initialize the compile stack. */
1683 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1684 if (compile_stack.stack == NULL)
1685 return REG_ESPACE;
1686
1687 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1688 compile_stack.avail = 0;
1689
1690 /* Initialize the pattern buffer. */
1691 bufp->syntax = syntax;
1692 bufp->fastmap_accurate = 0;
1693 bufp->not_bol = bufp->not_eol = 0;
1694
1695 /* Set `used' to zero, so that if we return an error, the pattern
1696 printer (for debugging) will think there's no pattern. We reset it
1697 at the end. */
1698 bufp->used = 0;
1699
1700 /* Always count groups, whether or not bufp->no_sub is set. */
1701 bufp->re_nsub = 0;
1702
1703 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1704 /* Initialize the syntax table. */
1705 init_syntax_once ();
1706 #endif
1707
1708 if (bufp->allocated == 0)
1709 {
1710 if (bufp->buffer)
1711 { /* If zero allocated, but buffer is non-null, try to realloc
1712 enough space. This loses if buffer's address is bogus, but
1713 that is the user's responsibility. */
1714 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1715 }
1716 else
1717 { /* Caller did not allocate a buffer. Do it for them. */
1718 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1719 }
1720 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1721
1722 bufp->allocated = INIT_BUF_SIZE;
1723 }
1724
1725 begalt = b = bufp->buffer;
1726
1727 /* Loop through the uncompiled pattern until we're at the end. */
1728 while (p != pend)
1729 {
1730 PATFETCH (c);
1731
1732 switch (c)
1733 {
1734 case '^':
1735 {
1736 if ( /* If at start of pattern, it's an operator. */
1737 p == pattern + 1
1738 /* If context independent, it's an operator. */
1739 || syntax & RE_CONTEXT_INDEP_ANCHORS
1740 /* Otherwise, depends on what's come before. */
1741 || at_begline_loc_p (pattern, p, syntax))
1742 BUF_PUSH (begline);
1743 else
1744 goto normal_char;
1745 }
1746 break;
1747
1748
1749 case '$':
1750 {
1751 if ( /* If at end of pattern, it's an operator. */
1752 p == pend
1753 /* If context independent, it's an operator. */
1754 || syntax & RE_CONTEXT_INDEP_ANCHORS
1755 /* Otherwise, depends on what's next. */
1756 || at_endline_loc_p (p, pend, syntax))
1757 BUF_PUSH (endline);
1758 else
1759 goto normal_char;
1760 }
1761 break;
1762
1763
1764 case '+':
1765 case '?':
1766 if ((syntax & RE_BK_PLUS_QM)
1767 || (syntax & RE_LIMITED_OPS))
1768 goto normal_char;
1769 handle_plus:
1770 case '*':
1771 /* If there is no previous pattern... */
1772 if (!laststart)
1773 {
1774 if (syntax & RE_CONTEXT_INVALID_OPS)
1775 FREE_STACK_RETURN (REG_BADRPT);
1776 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1777 goto normal_char;
1778 }
1779
1780 {
1781 /* Are we optimizing this jump? */
1782 boolean keep_string_p = false;
1783
1784 /* 1 means zero (many) matches is allowed. */
1785 char zero_times_ok = 0, many_times_ok = 0;
1786
1787 /* If there is a sequence of repetition chars, collapse it
1788 down to just one (the right one). We can't combine
1789 interval operators with these because of, e.g., `a{2}*',
1790 which should only match an even number of `a's. */
1791
1792 for (;;)
1793 {
1794 zero_times_ok |= c != '+';
1795 many_times_ok |= c != '?';
1796
1797 if (p == pend)
1798 break;
1799
1800 PATFETCH (c);
1801
1802 if (c == '*'
1803 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1804 ;
1805
1806 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1807 {
1808 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1809
1810 PATFETCH (c1);
1811 if (!(c1 == '+' || c1 == '?'))
1812 {
1813 PATUNFETCH;
1814 PATUNFETCH;
1815 break;
1816 }
1817
1818 c = c1;
1819 }
1820 else
1821 {
1822 PATUNFETCH;
1823 break;
1824 }
1825
1826 /* If we get here, we found another repeat character. */
1827 }
1828
1829 /* Star, etc. applied to an empty pattern is equivalent
1830 to an empty pattern. */
1831 if (!laststart)
1832 break;
1833
1834 /* Now we know whether or not zero matches is allowed
1835 and also whether or not two or more matches is allowed. */
1836 if (many_times_ok)
1837 { /* More than one repetition is allowed, so put in at the
1838 end a backward relative jump from `b' to before the next
1839 jump we're going to put in below (which jumps from
1840 laststart to after this jump).
1841
1842 But if we are at the `*' in the exact sequence `.*\n',
1843 insert an unconditional jump backwards to the .,
1844 instead of the beginning of the loop. This way we only
1845 push a failure point once, instead of every time
1846 through the loop. */
1847 assert (p - 1 > pattern);
1848
1849 /* Allocate the space for the jump. */
1850 GET_BUFFER_SPACE (3);
1851
1852 /* We know we are not at the first character of the pattern,
1853 because laststart was nonzero. And we've already
1854 incremented `p', by the way, to be the character after
1855 the `*'. Do we have to do something analogous here
1856 for null bytes, because of RE_DOT_NOT_NULL? */
1857 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1858 && zero_times_ok
1859 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1860 && !(syntax & RE_DOT_NEWLINE))
1861 { /* We have .*\n. */
1862 STORE_JUMP (jump, b, laststart);
1863 keep_string_p = true;
1864 }
1865 else
1866 /* Anything else. */
1867 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1868
1869 /* We've added more stuff to the buffer. */
1870 b += 3;
1871 }
1872
1873 /* On failure, jump from laststart to b + 3, which will be the
1874 end of the buffer after this jump is inserted. */
1875 GET_BUFFER_SPACE (3);
1876 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1877 : on_failure_jump,
1878 laststart, b + 3);
1879 pending_exact = 0;
1880 b += 3;
1881
1882 if (!zero_times_ok)
1883 {
1884 /* At least one repetition is required, so insert a
1885 `dummy_failure_jump' before the initial
1886 `on_failure_jump' instruction of the loop. This
1887 effects a skip over that instruction the first time
1888 we hit that loop. */
1889 GET_BUFFER_SPACE (3);
1890 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1891 b += 3;
1892 }
1893 }
1894 break;
1895
1896
1897 case '.':
1898 laststart = b;
1899 BUF_PUSH (anychar);
1900 break;
1901
1902
1903 case '[':
1904 {
1905 boolean had_char_class = false;
1906
1907 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1908
1909 /* Ensure that we have enough space to push a charset: the
1910 opcode, the length count, and the bitset; 34 bytes in all. */
1911 GET_BUFFER_SPACE (34);
1912
1913 laststart = b;
1914
1915 /* We test `*p == '^' twice, instead of using an if
1916 statement, so we only need one BUF_PUSH. */
1917 BUF_PUSH (*p == '^' ? charset_not : charset);
1918 if (*p == '^')
1919 p++;
1920
1921 /* Remember the first position in the bracket expression. */
1922 p1 = p;
1923
1924 /* Push the number of bytes in the bitmap. */
1925 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1926
1927 /* Clear the whole map. */
1928 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1929
1930 /* charset_not matches newline according to a syntax bit. */
1931 if ((re_opcode_t) b[-2] == charset_not
1932 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1933 SET_LIST_BIT ('\n');
1934
1935 /* Read in characters and ranges, setting map bits. */
1936 for (;;)
1937 {
1938 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1939
1940 PATFETCH (c);
1941
1942 /* \ might escape characters inside [...] and [^...]. */
1943 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1944 {
1945 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1946
1947 PATFETCH (c1);
1948 SET_LIST_BIT (c1);
1949 continue;
1950 }
1951
1952 /* Could be the end of the bracket expression. If it's
1953 not (i.e., when the bracket expression is `[]' so
1954 far), the ']' character bit gets set way below. */
1955 if (c == ']' && p != p1 + 1)
1956 break;
1957
1958 /* Look ahead to see if it's a range when the last thing
1959 was a character class. */
1960 if (had_char_class && c == '-' && *p != ']')
1961 FREE_STACK_RETURN (REG_ERANGE);
1962
1963 /* Look ahead to see if it's a range when the last thing
1964 was a character: if this is a hyphen not at the
1965 beginning or the end of a list, then it's the range
1966 operator. */
1967 if (c == '-'
1968 && !(p - 2 >= pattern && p[-2] == '[')
1969 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1970 && *p != ']')
1971 {
1972 reg_errcode_t ret
1973 = compile_range (&p, pend, translate, syntax, b);
1974 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1975 }
1976
1977 else if (p[0] == '-' && p[1] != ']')
1978 { /* This handles ranges made up of characters only. */
1979 reg_errcode_t ret;
1980
1981 /* Move past the `-'. */
1982 PATFETCH (c1);
1983
1984 ret = compile_range (&p, pend, translate, syntax, b);
1985 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1986 }
1987
1988 /* See if we're at the beginning of a possible character
1989 class. */
1990
1991 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
1992 { /* Leave room for the null. */
1993 char str[CHAR_CLASS_MAX_LENGTH + 1];
1994
1995 PATFETCH (c);
1996 c1 = 0;
1997
1998 /* If pattern is `[[:'. */
1999 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2000
2001 for (;;)
2002 {
2003 PATFETCH (c);
2004 if (c == ':' || c == ']' || p == pend
2005 || c1 == CHAR_CLASS_MAX_LENGTH)
2006 break;
2007 str[c1++] = c;
2008 }
2009 str[c1] = '\0';
2010
2011 /* If isn't a word bracketed by `[:' and:`]':
2012 undo the ending character, the letters, and leave
2013 the leading `:' and `[' (but set bits for them). */
2014 if (c == ':' && *p == ']')
2015 {
2016 int ch;
2017 boolean is_alnum = STREQ (str, "alnum");
2018 boolean is_alpha = STREQ (str, "alpha");
2019 boolean is_blank = STREQ (str, "blank");
2020 boolean is_cntrl = STREQ (str, "cntrl");
2021 boolean is_digit = STREQ (str, "digit");
2022 boolean is_graph = STREQ (str, "graph");
2023 boolean is_lower = STREQ (str, "lower");
2024 boolean is_print = STREQ (str, "print");
2025 boolean is_punct = STREQ (str, "punct");
2026 boolean is_space = STREQ (str, "space");
2027 boolean is_upper = STREQ (str, "upper");
2028 boolean is_xdigit = STREQ (str, "xdigit");
2029
2030 if (!IS_CHAR_CLASS (str))
2031 FREE_STACK_RETURN (REG_ECTYPE);
2032
2033 /* Throw away the ] at the end of the character
2034 class. */
2035 PATFETCH (c);
2036
2037 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2038
2039 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2040 {
2041 /* This was split into 3 if's to
2042 avoid an arbitrary limit in some compiler. */
2043 if ( (is_alnum && ISALNUM (ch))
2044 || (is_alpha && ISALPHA (ch))
2045 || (is_blank && ISBLANK (ch))
2046 || (is_cntrl && ISCNTRL (ch)))
2047 SET_LIST_BIT (ch);
2048 if ( (is_digit && ISDIGIT (ch))
2049 || (is_graph && ISGRAPH (ch))
2050 || (is_lower && ISLOWER (ch))
2051 || (is_print && ISPRINT (ch)))
2052 SET_LIST_BIT (ch);
2053 if ( (is_punct && ISPUNCT (ch))
2054 || (is_space && ISSPACE (ch))
2055 || (is_upper && ISUPPER (ch))
2056 || (is_xdigit && ISXDIGIT (ch)))
2057 SET_LIST_BIT (ch);
2058 }
2059 had_char_class = true;
2060 }
2061 else
2062 {
2063 c1++;
2064 while (c1--)
2065 PATUNFETCH;
2066 SET_LIST_BIT ('[');
2067 SET_LIST_BIT (':');
2068 had_char_class = false;
2069 }
2070 }
2071 else
2072 {
2073 had_char_class = false;
2074 SET_LIST_BIT (c);
2075 }
2076 }
2077
2078 /* Discard any (non)matching list bytes that are all 0 at the
2079 end of the map. Decrease the map-length byte too. */
2080 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2081 b[-1]--;
2082 b += b[-1];
2083 }
2084 break;
2085
2086
2087 case '(':
2088 if (syntax & RE_NO_BK_PARENS)
2089 goto handle_open;
2090 else
2091 goto normal_char;
2092
2093
2094 case ')':
2095 if (syntax & RE_NO_BK_PARENS)
2096 goto handle_close;
2097 else
2098 goto normal_char;
2099
2100
2101 case '\n':
2102 if (syntax & RE_NEWLINE_ALT)
2103 goto handle_alt;
2104 else
2105 goto normal_char;
2106
2107
2108 case '|':
2109 if (syntax & RE_NO_BK_VBAR)
2110 goto handle_alt;
2111 else
2112 goto normal_char;
2113
2114
2115 case '{':
2116 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2117 goto handle_interval;
2118 else
2119 goto normal_char;
2120
2121
2122 case '\\':
2123 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2124
2125 /* Do not translate the character after the \, so that we can
2126 distinguish, e.g., \B from \b, even if we normally would
2127 translate, e.g., B to b. */
2128 PATFETCH_RAW (c);
2129
2130 switch (c)
2131 {
2132 case '(':
2133 if (syntax & RE_NO_BK_PARENS)
2134 goto normal_backslash;
2135
2136 handle_open:
2137 bufp->re_nsub++;
2138 regnum++;
2139
2140 if (COMPILE_STACK_FULL)
2141 {
2142 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2143 compile_stack_elt_t);
2144 if (compile_stack.stack == NULL) return REG_ESPACE;
2145
2146 compile_stack.size <<= 1;
2147 }
2148
2149 /* These are the values to restore when we hit end of this
2150 group. They are all relative offsets, so that if the
2151 whole pattern moves because of realloc, they will still
2152 be valid. */
2153 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2154 COMPILE_STACK_TOP.fixup_alt_jump
2155 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2156 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2157 COMPILE_STACK_TOP.regnum = regnum;
2158
2159 /* We will eventually replace the 0 with the number of
2160 groups inner to this one. But do not push a
2161 start_memory for groups beyond the last one we can
2162 represent in the compiled pattern. */
2163 if (regnum <= MAX_REGNUM)
2164 {
2165 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2166 BUF_PUSH_3 (start_memory, regnum, 0);
2167 }
2168
2169 compile_stack.avail++;
2170
2171 fixup_alt_jump = 0;
2172 laststart = 0;
2173 begalt = b;
2174 /* If we've reached MAX_REGNUM groups, then this open
2175 won't actually generate any code, so we'll have to
2176 clear pending_exact explicitly. */
2177 pending_exact = 0;
2178 break;
2179
2180
2181 case ')':
2182 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2183
2184 if (COMPILE_STACK_EMPTY)
2185 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2186 goto normal_backslash;
2187 else
2188 FREE_STACK_RETURN (REG_ERPAREN);
2189
2190 handle_close:
2191 if (fixup_alt_jump)
2192 { /* Push a dummy failure point at the end of the
2193 alternative for a possible future
2194 `pop_failure_jump' to pop. See comments at
2195 `push_dummy_failure' in `re_match_2'. */
2196 BUF_PUSH (push_dummy_failure);
2197
2198 /* We allocated space for this jump when we assigned
2199 to `fixup_alt_jump', in the `handle_alt' case below. */
2200 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2201 }
2202
2203 /* See similar code for backslashed left paren above. */
2204 if (COMPILE_STACK_EMPTY)
2205 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2206 goto normal_char;
2207 else
2208 FREE_STACK_RETURN (REG_ERPAREN);
2209
2210 /* Since we just checked for an empty stack above, this
2211 ``can't happen''. */
2212 assert (compile_stack.avail != 0);
2213 {
2214 /* We don't just want to restore into `regnum', because
2215 later groups should continue to be numbered higher,
2216 as in `(ab)c(de)' -- the second group is #2. */
2217 regnum_t this_group_regnum;
2218
2219 compile_stack.avail--;
2220 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2221 fixup_alt_jump
2222 = COMPILE_STACK_TOP.fixup_alt_jump
2223 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2224 : 0;
2225 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2226 this_group_regnum = COMPILE_STACK_TOP.regnum;
2227 /* If we've reached MAX_REGNUM groups, then this open
2228 won't actually generate any code, so we'll have to
2229 clear pending_exact explicitly. */
2230 pending_exact = 0;
2231
2232 /* We're at the end of the group, so now we know how many
2233 groups were inside this one. */
2234 if (this_group_regnum <= MAX_REGNUM)
2235 {
2236 unsigned char *inner_group_loc
2237 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2238
2239 *inner_group_loc = regnum - this_group_regnum;
2240 BUF_PUSH_3 (stop_memory, this_group_regnum,
2241 regnum - this_group_regnum);
2242 }
2243 }
2244 break;
2245
2246
2247 case '|': /* `\|'. */
2248 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2249 goto normal_backslash;
2250 handle_alt:
2251 if (syntax & RE_LIMITED_OPS)
2252 goto normal_char;
2253
2254 /* Insert before the previous alternative a jump which
2255 jumps to this alternative if the former fails. */
2256 GET_BUFFER_SPACE (3);
2257 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2258 pending_exact = 0;
2259 b += 3;
2260
2261 /* The alternative before this one has a jump after it
2262 which gets executed if it gets matched. Adjust that
2263 jump so it will jump to this alternative's analogous
2264 jump (put in below, which in turn will jump to the next
2265 (if any) alternative's such jump, etc.). The last such
2266 jump jumps to the correct final destination. A picture:
2267 _____ _____
2268 | | | |
2269 | v | v
2270 a | b | c
2271
2272 If we are at `b', then fixup_alt_jump right now points to a
2273 three-byte space after `a'. We'll put in the jump, set
2274 fixup_alt_jump to right after `b', and leave behind three
2275 bytes which we'll fill in when we get to after `c'. */
2276
2277 if (fixup_alt_jump)
2278 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2279
2280 /* Mark and leave space for a jump after this alternative,
2281 to be filled in later either by next alternative or
2282 when know we're at the end of a series of alternatives. */
2283 fixup_alt_jump = b;
2284 GET_BUFFER_SPACE (3);
2285 b += 3;
2286
2287 laststart = 0;
2288 begalt = b;
2289 break;
2290
2291
2292 case '{':
2293 /* If \{ is a literal. */
2294 if (!(syntax & RE_INTERVALS)
2295 /* If we're at `\{' and it's not the open-interval
2296 operator. */
2297 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2298 || (p - 2 == pattern && p == pend))
2299 goto normal_backslash;
2300
2301 handle_interval:
2302 {
2303 /* If got here, then the syntax allows intervals. */
2304
2305 /* At least (most) this many matches must be made. */
2306 int lower_bound = -1, upper_bound = -1;
2307
2308 beg_interval = p - 1;
2309
2310 if (p == pend)
2311 {
2312 if (syntax & RE_NO_BK_BRACES)
2313 goto unfetch_interval;
2314 else
2315 FREE_STACK_RETURN (REG_EBRACE);
2316 }
2317
2318 GET_UNSIGNED_NUMBER (lower_bound);
2319
2320 if (c == ',')
2321 {
2322 GET_UNSIGNED_NUMBER (upper_bound);
2323 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2324 }
2325 else
2326 /* Interval such as `{1}' => match exactly once. */
2327 upper_bound = lower_bound;
2328
2329 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2330 || lower_bound > upper_bound)
2331 {
2332 if (syntax & RE_NO_BK_BRACES)
2333 goto unfetch_interval;
2334 else
2335 FREE_STACK_RETURN (REG_BADBR);
2336 }
2337
2338 if (!(syntax & RE_NO_BK_BRACES))
2339 {
2340 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2341
2342 PATFETCH (c);
2343 }
2344
2345 if (c != '}')
2346 {
2347 if (syntax & RE_NO_BK_BRACES)
2348 goto unfetch_interval;
2349 else
2350 FREE_STACK_RETURN (REG_BADBR);
2351 }
2352
2353 /* We just parsed a valid interval. */
2354
2355 /* If it's invalid to have no preceding re. */
2356 if (!laststart)
2357 {
2358 if (syntax & RE_CONTEXT_INVALID_OPS)
2359 FREE_STACK_RETURN (REG_BADRPT);
2360 else if (syntax & RE_CONTEXT_INDEP_OPS)
2361 laststart = b;
2362 else
2363 goto unfetch_interval;
2364 }
2365
2366 /* If the upper bound is zero, don't want to succeed at
2367 all; jump from `laststart' to `b + 3', which will be
2368 the end of the buffer after we insert the jump. */
2369 if (upper_bound == 0)
2370 {
2371 GET_BUFFER_SPACE (3);
2372 INSERT_JUMP (jump, laststart, b + 3);
2373 b += 3;
2374 }
2375
2376 /* Otherwise, we have a nontrivial interval. When
2377 we're all done, the pattern will look like:
2378 set_number_at <jump count> <upper bound>
2379 set_number_at <succeed_n count> <lower bound>
2380 succeed_n <after jump addr> <succeed_n count>
2381 <body of loop>
2382 jump_n <succeed_n addr> <jump count>
2383 (The upper bound and `jump_n' are omitted if
2384 `upper_bound' is 1, though.) */
2385 else
2386 { /* If the upper bound is > 1, we need to insert
2387 more at the end of the loop. */
2388 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2389
2390 GET_BUFFER_SPACE (nbytes);
2391
2392 /* Initialize lower bound of the `succeed_n', even
2393 though it will be set during matching by its
2394 attendant `set_number_at' (inserted next),
2395 because `re_compile_fastmap' needs to know.
2396 Jump to the `jump_n' we might insert below. */
2397 INSERT_JUMP2 (succeed_n, laststart,
2398 b + 5 + (upper_bound > 1) * 5,
2399 lower_bound);
2400 b += 5;
2401
2402 /* Code to initialize the lower bound. Insert
2403 before the `succeed_n'. The `5' is the last two
2404 bytes of this `set_number_at', plus 3 bytes of
2405 the following `succeed_n'. */
2406 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2407 b += 5;
2408
2409 if (upper_bound > 1)
2410 { /* More than one repetition is allowed, so
2411 append a backward jump to the `succeed_n'
2412 that starts this interval.
2413
2414 When we've reached this during matching,
2415 we'll have matched the interval once, so
2416 jump back only `upper_bound - 1' times. */
2417 STORE_JUMP2 (jump_n, b, laststart + 5,
2418 upper_bound - 1);
2419 b += 5;
2420
2421 /* The location we want to set is the second
2422 parameter of the `jump_n'; that is `b-2' as
2423 an absolute address. `laststart' will be
2424 the `set_number_at' we're about to insert;
2425 `laststart+3' the number to set, the source
2426 for the relative address. But we are
2427 inserting into the middle of the pattern --
2428 so everything is getting moved up by 5.
2429 Conclusion: (b - 2) - (laststart + 3) + 5,
2430 i.e., b - laststart.
2431
2432 We insert this at the beginning of the loop
2433 so that if we fail during matching, we'll
2434 reinitialize the bounds. */
2435 insert_op2 (set_number_at, laststart, b - laststart,
2436 upper_bound - 1, b);
2437 b += 5;
2438 }
2439 }
2440 pending_exact = 0;
2441 beg_interval = NULL;
2442 }
2443 break;
2444
2445 unfetch_interval:
2446 /* If an invalid interval, match the characters as literals. */
2447 assert (beg_interval);
2448 p = beg_interval;
2449 beg_interval = NULL;
2450
2451 /* normal_char and normal_backslash need `c'. */
2452 PATFETCH (c);
2453
2454 if (!(syntax & RE_NO_BK_BRACES))
2455 {
2456 if (p > pattern && p[-1] == '\\')
2457 goto normal_backslash;
2458 }
2459 goto normal_char;
2460
2461 #ifdef emacs
2462 /* There is no way to specify the before_dot and after_dot
2463 operators. rms says this is ok. --karl */
2464 case '=':
2465 BUF_PUSH (at_dot);
2466 break;
2467
2468 case 's':
2469 laststart = b;
2470 PATFETCH (c);
2471 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2472 break;
2473
2474 case 'S':
2475 laststart = b;
2476 PATFETCH (c);
2477 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2478 break;
2479 #endif /* emacs */
2480
2481
2482 case 'w':
2483 laststart = b;
2484 BUF_PUSH (wordchar);
2485 break;
2486
2487
2488 case 'W':
2489 laststart = b;
2490 BUF_PUSH (notwordchar);
2491 break;
2492
2493
2494 case '<':
2495 BUF_PUSH (wordbeg);
2496 break;
2497
2498 case '>':
2499 BUF_PUSH (wordend);
2500 break;
2501
2502 case 'b':
2503 BUF_PUSH (wordbound);
2504 break;
2505
2506 case 'B':
2507 BUF_PUSH (notwordbound);
2508 break;
2509
2510 case '`':
2511 BUF_PUSH (begbuf);
2512 break;
2513
2514 case '\'':
2515 BUF_PUSH (endbuf);
2516 break;
2517
2518 case '1': case '2': case '3': case '4': case '5':
2519 case '6': case '7': case '8': case '9':
2520 if (syntax & RE_NO_BK_REFS)
2521 goto normal_char;
2522
2523 c1 = c - '0';
2524
2525 if (c1 > regnum)
2526 FREE_STACK_RETURN (REG_ESUBREG);
2527
2528 /* Can't back reference to a subexpression if inside of it. */
2529 if (group_in_compile_stack (compile_stack, c1))
2530 goto normal_char;
2531
2532 laststart = b;
2533 BUF_PUSH_2 (duplicate, c1);
2534 break;
2535
2536
2537 case '+':
2538 case '?':
2539 if (syntax & RE_BK_PLUS_QM)
2540 goto handle_plus;
2541 else
2542 goto normal_backslash;
2543
2544 default:
2545 normal_backslash:
2546 /* You might think it would be useful for \ to mean
2547 not to translate; but if we don't translate it
2548 it will never match anything. */
2549 c = TRANSLATE (c);
2550 goto normal_char;
2551 }
2552 break;
2553
2554
2555 default:
2556 /* Expects the character in `c'. */
2557 normal_char:
2558 /* If no exactn currently being built. */
2559 if (!pending_exact
2560
2561 /* If last exactn not at current position. */
2562 || pending_exact + *pending_exact + 1 != b
2563
2564 /* We have only one byte following the exactn for the count. */
2565 || *pending_exact == (1 << BYTEWIDTH) - 1
2566
2567 /* If followed by a repetition operator. */
2568 || *p == '*' || *p == '^'
2569 || ((syntax & RE_BK_PLUS_QM)
2570 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2571 : (*p == '+' || *p == '?'))
2572 || ((syntax & RE_INTERVALS)
2573 && ((syntax & RE_NO_BK_BRACES)
2574 ? *p == '{'
2575 : (p[0] == '\\' && p[1] == '{'))))
2576 {
2577 /* Start building a new exactn. */
2578
2579 laststart = b;
2580
2581 BUF_PUSH_2 (exactn, 0);
2582 pending_exact = b - 1;
2583 }
2584
2585 BUF_PUSH (c);
2586 (*pending_exact)++;
2587 break;
2588 } /* switch (c) */
2589 } /* while p != pend */
2590
2591
2592 /* Through the pattern now. */
2593
2594 if (fixup_alt_jump)
2595 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2596
2597 if (!COMPILE_STACK_EMPTY)
2598 FREE_STACK_RETURN (REG_EPAREN);
2599
2600 /* If we don't want backtracking, force success
2601 the first time we reach the end of the compiled pattern. */
2602 if (syntax & RE_NO_POSIX_BACKTRACKING)
2603 BUF_PUSH (succeed);
2604
2605 free (compile_stack.stack);
2606
2607 /* We have succeeded; set the length of the buffer. */
2608 bufp->used = b - bufp->buffer;
2609
2610 #ifdef DEBUG
2611 if (debug)
2612 {
2613 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2614 print_compiled_pattern (bufp);
2615 }
2616 #endif /* DEBUG */
2617
2618 #ifndef MATCH_MAY_ALLOCATE
2619 /* Initialize the failure stack to the largest possible stack. This
2620 isn't necessary unless we're trying to avoid calling alloca in
2621 the search and match routines. */
2622 {
2623 int num_regs = bufp->re_nsub + 1;
2624
2625 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2626 is strictly greater than re_max_failures, the largest possible stack
2627 is 2 * re_max_failures failure points. */
2628 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2629 {
2630 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2631
2632 #ifdef emacs
2633 if (! fail_stack.stack)
2634 fail_stack.stack
2635 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2636 * sizeof (fail_stack_elt_t));
2637 else
2638 fail_stack.stack
2639 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2640 (fail_stack.size
2641 * sizeof (fail_stack_elt_t)));
2642 #else /* not emacs */
2643 if (! fail_stack.stack)
2644 fail_stack.stack
2645 = (fail_stack_elt_t *) malloc (fail_stack.size
2646 * sizeof (fail_stack_elt_t));
2647 else
2648 fail_stack.stack
2649 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2650 (fail_stack.size
2651 * sizeof (fail_stack_elt_t)));
2652 #endif /* not emacs */
2653 }
2654
2655 regex_grow_registers (num_regs);
2656 }
2657 #endif /* not MATCH_MAY_ALLOCATE */
2658
2659 return REG_NOERROR;
2660 } /* regex_compile */
2661
2662 /* Subroutines for `regex_compile'. */
2663
2664 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2665
2666 static void
2667 store_op1 (op, loc, arg)
2668 re_opcode_t op;
2669 unsigned char *loc;
2670 int arg;
2671 {
2672 *loc = (unsigned char) op;
2673 STORE_NUMBER (loc + 1, arg);
2674 }
2675
2676
2677 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2678
2679 static void
2680 store_op2 (op, loc, arg1, arg2)
2681 re_opcode_t op;
2682 unsigned char *loc;
2683 int arg1, arg2;
2684 {
2685 *loc = (unsigned char) op;
2686 STORE_NUMBER (loc + 1, arg1);
2687 STORE_NUMBER (loc + 3, arg2);
2688 }
2689
2690
2691 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2692 for OP followed by two-byte integer parameter ARG. */
2693
2694 static void
2695 insert_op1 (op, loc, arg, end)
2696 re_opcode_t op;
2697 unsigned char *loc;
2698 int arg;
2699 unsigned char *end;
2700 {
2701 register unsigned char *pfrom = end;
2702 register unsigned char *pto = end + 3;
2703
2704 while (pfrom != loc)
2705 *--pto = *--pfrom;
2706
2707 store_op1 (op, loc, arg);
2708 }
2709
2710
2711 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2712
2713 static void
2714 insert_op2 (op, loc, arg1, arg2, end)
2715 re_opcode_t op;
2716 unsigned char *loc;
2717 int arg1, arg2;
2718 unsigned char *end;
2719 {
2720 register unsigned char *pfrom = end;
2721 register unsigned char *pto = end + 5;
2722
2723 while (pfrom != loc)
2724 *--pto = *--pfrom;
2725
2726 store_op2 (op, loc, arg1, arg2);
2727 }
2728
2729
2730 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2731 after an alternative or a begin-subexpression. We assume there is at
2732 least one character before the ^. */
2733
2734 static boolean
2735 at_begline_loc_p (pattern, p, syntax)
2736 const char *pattern, *p;
2737 reg_syntax_t syntax;
2738 {
2739 const char *prev = p - 2;
2740 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2741
2742 return
2743 /* After a subexpression? */
2744 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2745 /* After an alternative? */
2746 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2747 }
2748
2749
2750 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2751 at least one character after the $, i.e., `P < PEND'. */
2752
2753 static boolean
2754 at_endline_loc_p (p, pend, syntax)
2755 const char *p, *pend;
2756 int syntax;
2757 {
2758 const char *next = p;
2759 boolean next_backslash = *next == '\\';
2760 const char *next_next = p + 1 < pend ? p + 1 : NULL;
2761
2762 return
2763 /* Before a subexpression? */
2764 (syntax & RE_NO_BK_PARENS ? *next == ')'
2765 : next_backslash && next_next && *next_next == ')')
2766 /* Before an alternative? */
2767 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2768 : next_backslash && next_next && *next_next == '|');
2769 }
2770
2771
2772 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2773 false if it's not. */
2774
2775 static boolean
2776 group_in_compile_stack (compile_stack, regnum)
2777 compile_stack_type compile_stack;
2778 regnum_t regnum;
2779 {
2780 int this_element;
2781
2782 for (this_element = compile_stack.avail - 1;
2783 this_element >= 0;
2784 this_element--)
2785 if (compile_stack.stack[this_element].regnum == regnum)
2786 return true;
2787
2788 return false;
2789 }
2790
2791
2792 /* Read the ending character of a range (in a bracket expression) from the
2793 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2794 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2795 Then we set the translation of all bits between the starting and
2796 ending characters (inclusive) in the compiled pattern B.
2797
2798 Return an error code.
2799
2800 We use these short variable names so we can use the same macros as
2801 `regex_compile' itself. */
2802
2803 static reg_errcode_t
2804 compile_range (p_ptr, pend, translate, syntax, b)
2805 const char **p_ptr, *pend;
2806 char *translate;
2807 reg_syntax_t syntax;
2808 unsigned char *b;
2809 {
2810 unsigned this_char;
2811
2812 const char *p = *p_ptr;
2813 int range_start, range_end;
2814
2815 if (p == pend)
2816 return REG_ERANGE;
2817
2818 /* Even though the pattern is a signed `char *', we need to fetch
2819 with unsigned char *'s; if the high bit of the pattern character
2820 is set, the range endpoints will be negative if we fetch using a
2821 signed char *.
2822
2823 We also want to fetch the endpoints without translating them; the
2824 appropriate translation is done in the bit-setting loop below. */
2825 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2826 range_start = ((const unsigned char *) p)[-2];
2827 range_end = ((const unsigned char *) p)[0];
2828
2829 /* Have to increment the pointer into the pattern string, so the
2830 caller isn't still at the ending character. */
2831 (*p_ptr)++;
2832
2833 /* If the start is after the end, the range is empty. */
2834 if (range_start > range_end)
2835 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2836
2837 /* Here we see why `this_char' has to be larger than an `unsigned
2838 char' -- the range is inclusive, so if `range_end' == 0xff
2839 (assuming 8-bit characters), we would otherwise go into an infinite
2840 loop, since all characters <= 0xff. */
2841 for (this_char = range_start; this_char <= range_end; this_char++)
2842 {
2843 SET_LIST_BIT (TRANSLATE (this_char));
2844 }
2845
2846 return REG_NOERROR;
2847 }
2848
2849 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2850 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2851 characters can start a string that matches the pattern. This fastmap
2852 is used by re_search to skip quickly over impossible starting points.
2853
2854 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2855 area as BUFP->fastmap.
2856
2857 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2858 the pattern buffer.
2859
2860 Returns 0 if we succeed, -2 if an internal error. */
2861
2862 int
2863 re_compile_fastmap (bufp)
2864 struct re_pattern_buffer *bufp;
2865 {
2866 int j, k;
2867 #ifdef MATCH_MAY_ALLOCATE
2868 fail_stack_type fail_stack;
2869 #endif
2870 #ifndef REGEX_MALLOC
2871 char *destination;
2872 #endif
2873 /* We don't push any register information onto the failure stack. */
2874 unsigned num_regs = 0;
2875
2876 register char *fastmap = bufp->fastmap;
2877 unsigned char *pattern = bufp->buffer;
2878 unsigned long size = bufp->used;
2879 unsigned char *p = pattern;
2880 register unsigned char *pend = pattern + size;
2881
2882 /* This holds the pointer to the failure stack, when
2883 it is allocated relocatably. */
2884 fail_stack_elt_t *failure_stack_ptr;
2885
2886 /* Assume that each path through the pattern can be null until
2887 proven otherwise. We set this false at the bottom of switch
2888 statement, to which we get only if a particular path doesn't
2889 match the empty string. */
2890 boolean path_can_be_null = true;
2891
2892 /* We aren't doing a `succeed_n' to begin with. */
2893 boolean succeed_n_p = false;
2894
2895 assert (fastmap != NULL && p != NULL);
2896
2897 INIT_FAIL_STACK ();
2898 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2899 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2900 bufp->can_be_null = 0;
2901
2902 while (1)
2903 {
2904 if (p == pend || *p == succeed)
2905 {
2906 /* We have reached the (effective) end of pattern. */
2907 if (!FAIL_STACK_EMPTY ())
2908 {
2909 bufp->can_be_null |= path_can_be_null;
2910
2911 /* Reset for next path. */
2912 path_can_be_null = true;
2913
2914 p = fail_stack.stack[--fail_stack.avail].pointer;
2915
2916 continue;
2917 }
2918 else
2919 break;
2920 }
2921
2922 /* We should never be about to go beyond the end of the pattern. */
2923 assert (p < pend);
2924
2925 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2926 {
2927
2928 /* I guess the idea here is to simply not bother with a fastmap
2929 if a backreference is used, since it's too hard to figure out
2930 the fastmap for the corresponding group. Setting
2931 `can_be_null' stops `re_search_2' from using the fastmap, so
2932 that is all we do. */
2933 case duplicate:
2934 bufp->can_be_null = 1;
2935 goto done;
2936
2937
2938 /* Following are the cases which match a character. These end
2939 with `break'. */
2940
2941 case exactn:
2942 fastmap[p[1]] = 1;
2943 break;
2944
2945
2946 case charset:
2947 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2948 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2949 fastmap[j] = 1;
2950 break;
2951
2952
2953 case charset_not:
2954 /* Chars beyond end of map must be allowed. */
2955 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2956 fastmap[j] = 1;
2957
2958 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2959 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2960 fastmap[j] = 1;
2961 break;
2962
2963
2964 case wordchar:
2965 for (j = 0; j < (1 << BYTEWIDTH); j++)
2966 if (SYNTAX (j) == Sword)
2967 fastmap[j] = 1;
2968 break;
2969
2970
2971 case notwordchar:
2972 for (j = 0; j < (1 << BYTEWIDTH); j++)
2973 if (SYNTAX (j) != Sword)
2974 fastmap[j] = 1;
2975 break;
2976
2977
2978 case anychar:
2979 {
2980 int fastmap_newline = fastmap['\n'];
2981
2982 /* `.' matches anything ... */
2983 for (j = 0; j < (1 << BYTEWIDTH); j++)
2984 fastmap[j] = 1;
2985
2986 /* ... except perhaps newline. */
2987 if (!(bufp->syntax & RE_DOT_NEWLINE))
2988 fastmap['\n'] = fastmap_newline;
2989
2990 /* Return if we have already set `can_be_null'; if we have,
2991 then the fastmap is irrelevant. Something's wrong here. */
2992 else if (bufp->can_be_null)
2993 goto done;
2994
2995 /* Otherwise, have to check alternative paths. */
2996 break;
2997 }
2998
2999 #ifdef emacs
3000 case syntaxspec:
3001 k = *p++;
3002 for (j = 0; j < (1 << BYTEWIDTH); j++)
3003 if (SYNTAX (j) == (enum syntaxcode) k)
3004 fastmap[j] = 1;
3005 break;
3006
3007
3008 case notsyntaxspec:
3009 k = *p++;
3010 for (j = 0; j < (1 << BYTEWIDTH); j++)
3011 if (SYNTAX (j) != (enum syntaxcode) k)
3012 fastmap[j] = 1;
3013 break;
3014
3015
3016 /* All cases after this match the empty string. These end with
3017 `continue'. */
3018
3019
3020 case before_dot:
3021 case at_dot:
3022 case after_dot:
3023 continue;
3024 #endif /* not emacs */
3025
3026
3027 case no_op:
3028 case begline:
3029 case endline:
3030 case begbuf:
3031 case endbuf:
3032 case wordbound:
3033 case notwordbound:
3034 case wordbeg:
3035 case wordend:
3036 case push_dummy_failure:
3037 continue;
3038
3039
3040 case jump_n:
3041 case pop_failure_jump:
3042 case maybe_pop_jump:
3043 case jump:
3044 case jump_past_alt:
3045 case dummy_failure_jump:
3046 EXTRACT_NUMBER_AND_INCR (j, p);
3047 p += j;
3048 if (j > 0)
3049 continue;
3050
3051 /* Jump backward implies we just went through the body of a
3052 loop and matched nothing. Opcode jumped to should be
3053 `on_failure_jump' or `succeed_n'. Just treat it like an
3054 ordinary jump. For a * loop, it has pushed its failure
3055 point already; if so, discard that as redundant. */
3056 if ((re_opcode_t) *p != on_failure_jump
3057 && (re_opcode_t) *p != succeed_n)
3058 continue;
3059
3060 p++;
3061 EXTRACT_NUMBER_AND_INCR (j, p);
3062 p += j;
3063
3064 /* If what's on the stack is where we are now, pop it. */
3065 if (!FAIL_STACK_EMPTY ()
3066 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3067 fail_stack.avail--;
3068
3069 continue;
3070
3071
3072 case on_failure_jump:
3073 case on_failure_keep_string_jump:
3074 handle_on_failure_jump:
3075 EXTRACT_NUMBER_AND_INCR (j, p);
3076
3077 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3078 end of the pattern. We don't want to push such a point,
3079 since when we restore it above, entering the switch will
3080 increment `p' past the end of the pattern. We don't need
3081 to push such a point since we obviously won't find any more
3082 fastmap entries beyond `pend'. Such a pattern can match
3083 the null string, though. */
3084 if (p + j < pend)
3085 {
3086 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3087 {
3088 RESET_FAIL_STACK ();
3089 return -2;
3090 }
3091 }
3092 else
3093 bufp->can_be_null = 1;
3094
3095 if (succeed_n_p)
3096 {
3097 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3098 succeed_n_p = false;
3099 }
3100
3101 continue;
3102
3103
3104 case succeed_n:
3105 /* Get to the number of times to succeed. */
3106 p += 2;
3107
3108 /* Increment p past the n for when k != 0. */
3109 EXTRACT_NUMBER_AND_INCR (k, p);
3110 if (k == 0)
3111 {
3112 p -= 4;
3113 succeed_n_p = true; /* Spaghetti code alert. */
3114 goto handle_on_failure_jump;
3115 }
3116 continue;
3117
3118
3119 case set_number_at:
3120 p += 4;
3121 continue;
3122
3123
3124 case start_memory:
3125 case stop_memory:
3126 p += 2;
3127 continue;
3128
3129
3130 default:
3131 abort (); /* We have listed all the cases. */
3132 } /* switch *p++ */
3133
3134 /* Getting here means we have found the possible starting
3135 characters for one path of the pattern -- and that the empty
3136 string does not match. We need not follow this path further.
3137 Instead, look at the next alternative (remembered on the
3138 stack), or quit if no more. The test at the top of the loop
3139 does these things. */
3140 path_can_be_null = false;
3141 p = pend;
3142 } /* while p */
3143
3144 /* Set `can_be_null' for the last path (also the first path, if the
3145 pattern is empty). */
3146 bufp->can_be_null |= path_can_be_null;
3147
3148 done:
3149 RESET_FAIL_STACK ();
3150 return 0;
3151 } /* re_compile_fastmap */
3152
3153 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3154 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3155 this memory for recording register information. STARTS and ENDS
3156 must be allocated using the malloc library routine, and must each
3157 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3158
3159 If NUM_REGS == 0, then subsequent matches should allocate their own
3160 register data.
3161
3162 Unless this function is called, the first search or match using
3163 PATTERN_BUFFER will allocate its own register data, without
3164 freeing the old data. */
3165
3166 void
3167 re_set_registers (bufp, regs, num_regs, starts, ends)
3168 struct re_pattern_buffer *bufp;
3169 struct re_registers *regs;
3170 unsigned num_regs;
3171 regoff_t *starts, *ends;
3172 {
3173 if (num_regs)
3174 {
3175 bufp->regs_allocated = REGS_REALLOCATE;
3176 regs->num_regs = num_regs;
3177 regs->start = starts;
3178 regs->end = ends;
3179 }
3180 else
3181 {
3182 bufp->regs_allocated = REGS_UNALLOCATED;
3183 regs->num_regs = 0;
3184 regs->start = regs->end = (regoff_t *) 0;
3185 }
3186 }
3187
3188 /* Searching routines. */
3189
3190 /* Like re_search_2, below, but only one string is specified, and
3191 doesn't let you say where to stop matching. */
3192
3193 int
3194 re_search (bufp, string, size, startpos, range, regs)
3195 struct re_pattern_buffer *bufp;
3196 const char *string;
3197 int size, startpos, range;
3198 struct re_registers *regs;
3199 {
3200 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3201 regs, size);
3202 }
3203
3204
3205 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3206 virtual concatenation of STRING1 and STRING2, starting first at index
3207 STARTPOS, then at STARTPOS + 1, and so on.
3208
3209 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3210
3211 RANGE is how far to scan while trying to match. RANGE = 0 means try
3212 only at STARTPOS; in general, the last start tried is STARTPOS +
3213 RANGE.
3214
3215 In REGS, return the indices of the virtual concatenation of STRING1
3216 and STRING2 that matched the entire BUFP->buffer and its contained
3217 subexpressions.
3218
3219 Do not consider matching one past the index STOP in the virtual
3220 concatenation of STRING1 and STRING2.
3221
3222 We return either the position in the strings at which the match was
3223 found, -1 if no match, or -2 if error (such as failure
3224 stack overflow). */
3225
3226 int
3227 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3228 struct re_pattern_buffer *bufp;
3229 const char *string1, *string2;
3230 int size1, size2;
3231 int startpos;
3232 int range;
3233 struct re_registers *regs;
3234 int stop;
3235 {
3236 int val;
3237 register char *fastmap = bufp->fastmap;
3238 register char *translate = bufp->translate;
3239 int total_size = size1 + size2;
3240 int endpos = startpos + range;
3241
3242 /* Check for out-of-range STARTPOS. */
3243 if (startpos < 0 || startpos > total_size)
3244 return -1;
3245
3246 /* Fix up RANGE if it might eventually take us outside
3247 the virtual concatenation of STRING1 and STRING2. */
3248 if (endpos < -1)
3249 range = -1 - startpos;
3250 else if (endpos > total_size)
3251 range = total_size - startpos;
3252
3253 /* If the search isn't to be a backwards one, don't waste time in a
3254 search for a pattern that must be anchored. */
3255 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3256 {
3257 if (startpos > 0)
3258 return -1;
3259 else
3260 range = 1;
3261 }
3262
3263 /* Update the fastmap now if not correct already. */
3264 if (fastmap && !bufp->fastmap_accurate)
3265 if (re_compile_fastmap (bufp) == -2)
3266 return -2;
3267
3268 /* Loop through the string, looking for a place to start matching. */
3269 for (;;)
3270 {
3271 /* If a fastmap is supplied, skip quickly over characters that
3272 cannot be the start of a match. If the pattern can match the
3273 null string, however, we don't need to skip characters; we want
3274 the first null string. */
3275 if (fastmap && startpos < total_size && !bufp->can_be_null)
3276 {
3277 if (range > 0) /* Searching forwards. */
3278 {
3279 register const char *d;
3280 register int lim = 0;
3281 int irange = range;
3282
3283 if (startpos < size1 && startpos + range >= size1)
3284 lim = range - (size1 - startpos);
3285
3286 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3287
3288 /* Written out as an if-else to avoid testing `translate'
3289 inside the loop. */
3290 if (translate)
3291 while (range > lim
3292 && !fastmap[(unsigned char)
3293 translate[(unsigned char) *d++]])
3294 range--;
3295 else
3296 while (range > lim && !fastmap[(unsigned char) *d++])
3297 range--;
3298
3299 startpos += irange - range;
3300 }
3301 else /* Searching backwards. */
3302 {
3303 register char c = (size1 == 0 || startpos >= size1
3304 ? string2[startpos - size1]
3305 : string1[startpos]);
3306
3307 if (!fastmap[(unsigned char) TRANSLATE (c)])
3308 goto advance;
3309 }
3310 }
3311
3312 /* If can't match the null string, and that's all we have left, fail. */
3313 if (range >= 0 && startpos == total_size && fastmap
3314 && !bufp->can_be_null)
3315 return -1;
3316
3317 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3318 startpos, regs, stop);
3319 #ifndef REGEX_MALLOC
3320 #ifdef C_ALLOCA
3321 alloca (0);
3322 #endif
3323 #endif
3324
3325 if (val >= 0)
3326 return startpos;
3327
3328 if (val == -2)
3329 return -2;
3330
3331 advance:
3332 if (!range)
3333 break;
3334 else if (range > 0)
3335 {
3336 range--;
3337 startpos++;
3338 }
3339 else
3340 {
3341 range++;
3342 startpos--;
3343 }
3344 }
3345 return -1;
3346 } /* re_search_2 */
3347
3348 /* Declarations and macros for re_match_2. */
3349
3350 static int bcmp_translate ();
3351 static boolean alt_match_null_string_p (),
3352 common_op_match_null_string_p (),
3353 group_match_null_string_p ();
3354
3355 /* This converts PTR, a pointer into one of the search strings `string1'
3356 and `string2' into an offset from the beginning of that string. */
3357 #define POINTER_TO_OFFSET(ptr) \
3358 (FIRST_STRING_P (ptr) \
3359 ? ((regoff_t) ((ptr) - string1)) \
3360 : ((regoff_t) ((ptr) - string2 + size1)))
3361
3362 /* Macros for dealing with the split strings in re_match_2. */
3363
3364 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3365
3366 /* Call before fetching a character with *d. This switches over to
3367 string2 if necessary. */
3368 #define PREFETCH() \
3369 while (d == dend) \
3370 { \
3371 /* End of string2 => fail. */ \
3372 if (dend == end_match_2) \
3373 goto fail; \
3374 /* End of string1 => advance to string2. */ \
3375 d = string2; \
3376 dend = end_match_2; \
3377 }
3378
3379
3380 /* Test if at very beginning or at very end of the virtual concatenation
3381 of `string1' and `string2'. If only one string, it's `string2'. */
3382 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3383 #define AT_STRINGS_END(d) ((d) == end2)
3384
3385
3386 /* Test if D points to a character which is word-constituent. We have
3387 two special cases to check for: if past the end of string1, look at
3388 the first character in string2; and if before the beginning of
3389 string2, look at the last character in string1. */
3390 #define WORDCHAR_P(d) \
3391 (SYNTAX ((d) == end1 ? *string2 \
3392 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3393 == Sword)
3394
3395 /* Test if the character before D and the one at D differ with respect
3396 to being word-constituent. */
3397 #define AT_WORD_BOUNDARY(d) \
3398 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3399 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3400
3401
3402 /* Free everything we malloc. */
3403 #ifdef MATCH_MAY_ALLOCATE
3404 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3405 #define FREE_VARIABLES() \
3406 do { \
3407 REGEX_FREE_STACK (fail_stack.stack); \
3408 FREE_VAR (regstart); \
3409 FREE_VAR (regend); \
3410 FREE_VAR (old_regstart); \
3411 FREE_VAR (old_regend); \
3412 FREE_VAR (best_regstart); \
3413 FREE_VAR (best_regend); \
3414 FREE_VAR (reg_info); \
3415 FREE_VAR (reg_dummy); \
3416 FREE_VAR (reg_info_dummy); \
3417 } while (0)
3418 #else
3419 #define FREE_VARIABLES() /* Do nothing! */
3420 #endif /* not MATCH_MAY_ALLOCATE */
3421
3422 /* These values must meet several constraints. They must not be valid
3423 register values; since we have a limit of 255 registers (because
3424 we use only one byte in the pattern for the register number), we can
3425 use numbers larger than 255. They must differ by 1, because of
3426 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3427 be larger than the value for the highest register, so we do not try
3428 to actually save any registers when none are active. */
3429 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3430 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3431
3432 /* Matching routines. */
3433
3434 #ifndef emacs /* Emacs never uses this. */
3435 /* re_match is like re_match_2 except it takes only a single string. */
3436
3437 int
3438 re_match (bufp, string, size, pos, regs)
3439 struct re_pattern_buffer *bufp;
3440 const char *string;
3441 int size, pos;
3442 struct re_registers *regs;
3443 {
3444 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3445 pos, regs, size);
3446 alloca (0);
3447 return result;
3448 }
3449 #endif /* not emacs */
3450
3451
3452 /* re_match_2 matches the compiled pattern in BUFP against the
3453 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3454 and SIZE2, respectively). We start matching at POS, and stop
3455 matching at STOP.
3456
3457 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3458 store offsets for the substring each group matched in REGS. See the
3459 documentation for exactly how many groups we fill.
3460
3461 We return -1 if no match, -2 if an internal error (such as the
3462 failure stack overflowing). Otherwise, we return the length of the
3463 matched substring. */
3464
3465 int
3466 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3467 struct re_pattern_buffer *bufp;
3468 const char *string1, *string2;
3469 int size1, size2;
3470 int pos;
3471 struct re_registers *regs;
3472 int stop;
3473 {
3474 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3475 pos, regs, stop);
3476 alloca (0);
3477 return result;
3478 }
3479
3480 /* This is a separate function so that we can force an alloca cleanup
3481 afterwards. */
3482 static int
3483 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3484 struct re_pattern_buffer *bufp;
3485 const char *string1, *string2;
3486 int size1, size2;
3487 int pos;
3488 struct re_registers *regs;
3489 int stop;
3490 {
3491 /* General temporaries. */
3492 int mcnt;
3493 unsigned char *p1;
3494
3495 /* Just past the end of the corresponding string. */
3496 const char *end1, *end2;
3497
3498 /* Pointers into string1 and string2, just past the last characters in
3499 each to consider matching. */
3500 const char *end_match_1, *end_match_2;
3501
3502 /* Where we are in the data, and the end of the current string. */
3503 const char *d, *dend;
3504
3505 /* Where we are in the pattern, and the end of the pattern. */
3506 unsigned char *p = bufp->buffer;
3507 register unsigned char *pend = p + bufp->used;
3508
3509 /* Mark the opcode just after a start_memory, so we can test for an
3510 empty subpattern when we get to the stop_memory. */
3511 unsigned char *just_past_start_mem = 0;
3512
3513 /* We use this to map every character in the string. */
3514 char *translate = bufp->translate;
3515
3516 /* Failure point stack. Each place that can handle a failure further
3517 down the line pushes a failure point on this stack. It consists of
3518 restart, regend, and reg_info for all registers corresponding to
3519 the subexpressions we're currently inside, plus the number of such
3520 registers, and, finally, two char *'s. The first char * is where
3521 to resume scanning the pattern; the second one is where to resume
3522 scanning the strings. If the latter is zero, the failure point is
3523 a ``dummy''; if a failure happens and the failure point is a dummy,
3524 it gets discarded and the next next one is tried. */
3525 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3526 fail_stack_type fail_stack;
3527 #endif
3528 #ifdef DEBUG
3529 static unsigned failure_id = 0;
3530 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3531 #endif
3532
3533 /* This holds the pointer to the failure stack, when
3534 it is allocated relocatably. */
3535 fail_stack_elt_t *failure_stack_ptr;
3536
3537 /* We fill all the registers internally, independent of what we
3538 return, for use in backreferences. The number here includes
3539 an element for register zero. */
3540 unsigned num_regs = bufp->re_nsub + 1;
3541
3542 /* The currently active registers. */
3543 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3544 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3545
3546 /* Information on the contents of registers. These are pointers into
3547 the input strings; they record just what was matched (on this
3548 attempt) by a subexpression part of the pattern, that is, the
3549 regnum-th regstart pointer points to where in the pattern we began
3550 matching and the regnum-th regend points to right after where we
3551 stopped matching the regnum-th subexpression. (The zeroth register
3552 keeps track of what the whole pattern matches.) */
3553 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3554 const char **regstart, **regend;
3555 #endif
3556
3557 /* If a group that's operated upon by a repetition operator fails to
3558 match anything, then the register for its start will need to be
3559 restored because it will have been set to wherever in the string we
3560 are when we last see its open-group operator. Similarly for a
3561 register's end. */
3562 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3563 const char **old_regstart, **old_regend;
3564 #endif
3565
3566 /* The is_active field of reg_info helps us keep track of which (possibly
3567 nested) subexpressions we are currently in. The matched_something
3568 field of reg_info[reg_num] helps us tell whether or not we have
3569 matched any of the pattern so far this time through the reg_num-th
3570 subexpression. These two fields get reset each time through any
3571 loop their register is in. */
3572 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3573 register_info_type *reg_info;
3574 #endif
3575
3576 /* The following record the register info as found in the above
3577 variables when we find a match better than any we've seen before.
3578 This happens as we backtrack through the failure points, which in
3579 turn happens only if we have not yet matched the entire string. */
3580 unsigned best_regs_set = false;
3581 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3582 const char **best_regstart, **best_regend;
3583 #endif
3584
3585 /* Logically, this is `best_regend[0]'. But we don't want to have to
3586 allocate space for that if we're not allocating space for anything
3587 else (see below). Also, we never need info about register 0 for
3588 any of the other register vectors, and it seems rather a kludge to
3589 treat `best_regend' differently than the rest. So we keep track of
3590 the end of the best match so far in a separate variable. We
3591 initialize this to NULL so that when we backtrack the first time
3592 and need to test it, it's not garbage. */
3593 const char *match_end = NULL;
3594
3595 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3596 int set_regs_matched_done = 0;
3597
3598 /* Used when we pop values we don't care about. */
3599 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3600 const char **reg_dummy;
3601 register_info_type *reg_info_dummy;
3602 #endif
3603
3604 #ifdef DEBUG
3605 /* Counts the total number of registers pushed. */
3606 unsigned num_regs_pushed = 0;
3607 #endif
3608
3609 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3610
3611 INIT_FAIL_STACK ();
3612
3613 #ifdef MATCH_MAY_ALLOCATE
3614 /* Do not bother to initialize all the register variables if there are
3615 no groups in the pattern, as it takes a fair amount of time. If
3616 there are groups, we include space for register 0 (the whole
3617 pattern), even though we never use it, since it simplifies the
3618 array indexing. We should fix this. */
3619 if (bufp->re_nsub)
3620 {
3621 regstart = REGEX_TALLOC (num_regs, const char *);
3622 regend = REGEX_TALLOC (num_regs, const char *);
3623 old_regstart = REGEX_TALLOC (num_regs, const char *);
3624 old_regend = REGEX_TALLOC (num_regs, const char *);
3625 best_regstart = REGEX_TALLOC (num_regs, const char *);
3626 best_regend = REGEX_TALLOC (num_regs, const char *);
3627 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3628 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3629 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3630
3631 if (!(regstart && regend && old_regstart && old_regend && reg_info
3632 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3633 {
3634 FREE_VARIABLES ();
3635 return -2;
3636 }
3637 }
3638 else
3639 {
3640 /* We must initialize all our variables to NULL, so that
3641 `FREE_VARIABLES' doesn't try to free them. */
3642 regstart = regend = old_regstart = old_regend = best_regstart
3643 = best_regend = reg_dummy = NULL;
3644 reg_info = reg_info_dummy = (register_info_type *) NULL;
3645 }
3646 #endif /* MATCH_MAY_ALLOCATE */
3647
3648 /* The starting position is bogus. */
3649 if (pos < 0 || pos > size1 + size2)
3650 {
3651 FREE_VARIABLES ();
3652 return -1;
3653 }
3654
3655 /* Initialize subexpression text positions to -1 to mark ones that no
3656 start_memory/stop_memory has been seen for. Also initialize the
3657 register information struct. */
3658 for (mcnt = 1; mcnt < num_regs; mcnt++)
3659 {
3660 regstart[mcnt] = regend[mcnt]
3661 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3662
3663 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3664 IS_ACTIVE (reg_info[mcnt]) = 0;
3665 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3666 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3667 }
3668
3669 /* We move `string1' into `string2' if the latter's empty -- but not if
3670 `string1' is null. */
3671 if (size2 == 0 && string1 != NULL)
3672 {
3673 string2 = string1;
3674 size2 = size1;
3675 string1 = 0;
3676 size1 = 0;
3677 }
3678 end1 = string1 + size1;
3679 end2 = string2 + size2;
3680
3681 /* Compute where to stop matching, within the two strings. */
3682 if (stop <= size1)
3683 {
3684 end_match_1 = string1 + stop;
3685 end_match_2 = string2;
3686 }
3687 else
3688 {
3689 end_match_1 = end1;
3690 end_match_2 = string2 + stop - size1;
3691 }
3692
3693 /* `p' scans through the pattern as `d' scans through the data.
3694 `dend' is the end of the input string that `d' points within. `d'
3695 is advanced into the following input string whenever necessary, but
3696 this happens before fetching; therefore, at the beginning of the
3697 loop, `d' can be pointing at the end of a string, but it cannot
3698 equal `string2'. */
3699 if (size1 > 0 && pos <= size1)
3700 {
3701 d = string1 + pos;
3702 dend = end_match_1;
3703 }
3704 else
3705 {
3706 d = string2 + pos - size1;
3707 dend = end_match_2;
3708 }
3709
3710 DEBUG_PRINT1 ("The compiled pattern is: ");
3711 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3712 DEBUG_PRINT1 ("The string to match is: `");
3713 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3714 DEBUG_PRINT1 ("'\n");
3715
3716 /* This loops over pattern commands. It exits by returning from the
3717 function if the match is complete, or it drops through if the match
3718 fails at this starting point in the input data. */
3719 for (;;)
3720 {
3721 DEBUG_PRINT2 ("\n0x%x: ", p);
3722
3723 if (p == pend)
3724 { /* End of pattern means we might have succeeded. */
3725 DEBUG_PRINT1 ("end of pattern ... ");
3726
3727 /* If we haven't matched the entire string, and we want the
3728 longest match, try backtracking. */
3729 if (d != end_match_2)
3730 {
3731 /* 1 if this match ends in the same string (string1 or string2)
3732 as the best previous match. */
3733 boolean same_str_p = (FIRST_STRING_P (match_end)
3734 == MATCHING_IN_FIRST_STRING);
3735 /* 1 if this match is the best seen so far. */
3736 boolean best_match_p;
3737
3738 /* AIX compiler got confused when this was combined
3739 with the previous declaration. */
3740 if (same_str_p)
3741 best_match_p = d > match_end;
3742 else
3743 best_match_p = !MATCHING_IN_FIRST_STRING;
3744
3745 DEBUG_PRINT1 ("backtracking.\n");
3746
3747 if (!FAIL_STACK_EMPTY ())
3748 { /* More failure points to try. */
3749
3750 /* If exceeds best match so far, save it. */
3751 if (!best_regs_set || best_match_p)
3752 {
3753 best_regs_set = true;
3754 match_end = d;
3755
3756 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3757
3758 for (mcnt = 1; mcnt < num_regs; mcnt++)
3759 {
3760 best_regstart[mcnt] = regstart[mcnt];
3761 best_regend[mcnt] = regend[mcnt];
3762 }
3763 }
3764 goto fail;
3765 }
3766
3767 /* If no failure points, don't restore garbage. And if
3768 last match is real best match, don't restore second
3769 best one. */
3770 else if (best_regs_set && !best_match_p)
3771 {
3772 restore_best_regs:
3773 /* Restore best match. It may happen that `dend ==
3774 end_match_1' while the restored d is in string2.
3775 For example, the pattern `x.*y.*z' against the
3776 strings `x-' and `y-z-', if the two strings are
3777 not consecutive in memory. */
3778 DEBUG_PRINT1 ("Restoring best registers.\n");
3779
3780 d = match_end;
3781 dend = ((d >= string1 && d <= end1)
3782 ? end_match_1 : end_match_2);
3783
3784 for (mcnt = 1; mcnt < num_regs; mcnt++)
3785 {
3786 regstart[mcnt] = best_regstart[mcnt];
3787 regend[mcnt] = best_regend[mcnt];
3788 }
3789 }
3790 } /* d != end_match_2 */
3791
3792 succeed_label:
3793 DEBUG_PRINT1 ("Accepting match.\n");
3794
3795 /* If caller wants register contents data back, do it. */
3796 if (regs && !bufp->no_sub)
3797 {
3798 /* Have the register data arrays been allocated? */
3799 if (bufp->regs_allocated == REGS_UNALLOCATED)
3800 { /* No. So allocate them with malloc. We need one
3801 extra element beyond `num_regs' for the `-1' marker
3802 GNU code uses. */
3803 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3804 regs->start = TALLOC (regs->num_regs, regoff_t);
3805 regs->end = TALLOC (regs->num_regs, regoff_t);
3806 if (regs->start == NULL || regs->end == NULL)
3807 {
3808 FREE_VARIABLES ();
3809 return -2;
3810 }
3811 bufp->regs_allocated = REGS_REALLOCATE;
3812 }
3813 else if (bufp->regs_allocated == REGS_REALLOCATE)
3814 { /* Yes. If we need more elements than were already
3815 allocated, reallocate them. If we need fewer, just
3816 leave it alone. */
3817 if (regs->num_regs < num_regs + 1)
3818 {
3819 regs->num_regs = num_regs + 1;
3820 RETALLOC (regs->start, regs->num_regs, regoff_t);
3821 RETALLOC (regs->end, regs->num_regs, regoff_t);
3822 if (regs->start == NULL || regs->end == NULL)
3823 {
3824 FREE_VARIABLES ();
3825 return -2;
3826 }
3827 }
3828 }
3829 else
3830 {
3831 /* These braces fend off a "empty body in an else-statement"
3832 warning under GCC when assert expands to nothing. */
3833 assert (bufp->regs_allocated == REGS_FIXED);
3834 }
3835
3836 /* Convert the pointer data in `regstart' and `regend' to
3837 indices. Register zero has to be set differently,
3838 since we haven't kept track of any info for it. */
3839 if (regs->num_regs > 0)
3840 {
3841 regs->start[0] = pos;
3842 regs->end[0] = (MATCHING_IN_FIRST_STRING
3843 ? ((regoff_t) (d - string1))
3844 : ((regoff_t) (d - string2 + size1)));
3845 }
3846
3847 /* Go through the first `min (num_regs, regs->num_regs)'
3848 registers, since that is all we initialized. */
3849 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3850 {
3851 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3852 regs->start[mcnt] = regs->end[mcnt] = -1;
3853 else
3854 {
3855 regs->start[mcnt]
3856 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3857 regs->end[mcnt]
3858 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3859 }
3860 }
3861
3862 /* If the regs structure we return has more elements than
3863 were in the pattern, set the extra elements to -1. If
3864 we (re)allocated the registers, this is the case,
3865 because we always allocate enough to have at least one
3866 -1 at the end. */
3867 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3868 regs->start[mcnt] = regs->end[mcnt] = -1;
3869 } /* regs && !bufp->no_sub */
3870
3871 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3872 nfailure_points_pushed, nfailure_points_popped,
3873 nfailure_points_pushed - nfailure_points_popped);
3874 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3875
3876 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3877 ? string1
3878 : string2 - size1);
3879
3880 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3881
3882 FREE_VARIABLES ();
3883 return mcnt;
3884 }
3885
3886 /* Otherwise match next pattern command. */
3887 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3888 {
3889 /* Ignore these. Used to ignore the n of succeed_n's which
3890 currently have n == 0. */
3891 case no_op:
3892 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3893 break;
3894
3895 case succeed:
3896 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3897 goto succeed_label;
3898
3899 /* Match the next n pattern characters exactly. The following
3900 byte in the pattern defines n, and the n bytes after that
3901 are the characters to match. */
3902 case exactn:
3903 mcnt = *p++;
3904 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3905
3906 /* This is written out as an if-else so we don't waste time
3907 testing `translate' inside the loop. */
3908 if (translate)
3909 {
3910 do
3911 {
3912 PREFETCH ();
3913 if (translate[(unsigned char) *d++] != (char) *p++)
3914 goto fail;
3915 }
3916 while (--mcnt);
3917 }
3918 else
3919 {
3920 do
3921 {
3922 PREFETCH ();
3923 if (*d++ != (char) *p++) goto fail;
3924 }
3925 while (--mcnt);
3926 }
3927 SET_REGS_MATCHED ();
3928 break;
3929
3930
3931 /* Match any character except possibly a newline or a null. */
3932 case anychar:
3933 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3934
3935 PREFETCH ();
3936
3937 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3938 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3939 goto fail;
3940
3941 SET_REGS_MATCHED ();
3942 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3943 d++;
3944 break;
3945
3946
3947 case charset:
3948 case charset_not:
3949 {
3950 register unsigned char c;
3951 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3952
3953 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3954
3955 PREFETCH ();
3956 c = TRANSLATE (*d); /* The character to match. */
3957
3958 /* Cast to `unsigned' instead of `unsigned char' in case the
3959 bit list is a full 32 bytes long. */
3960 if (c < (unsigned) (*p * BYTEWIDTH)
3961 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3962 not = !not;
3963
3964 p += 1 + *p;
3965
3966 if (!not) goto fail;
3967
3968 SET_REGS_MATCHED ();
3969 d++;
3970 break;
3971 }
3972
3973
3974 /* The beginning of a group is represented by start_memory.
3975 The arguments are the register number in the next byte, and the
3976 number of groups inner to this one in the next. The text
3977 matched within the group is recorded (in the internal
3978 registers data structure) under the register number. */
3979 case start_memory:
3980 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
3981
3982 /* Find out if this group can match the empty string. */
3983 p1 = p; /* To send to group_match_null_string_p. */
3984
3985 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
3986 REG_MATCH_NULL_STRING_P (reg_info[*p])
3987 = group_match_null_string_p (&p1, pend, reg_info);
3988
3989 /* Save the position in the string where we were the last time
3990 we were at this open-group operator in case the group is
3991 operated upon by a repetition operator, e.g., with `(a*)*b'
3992 against `ab'; then we want to ignore where we are now in
3993 the string in case this attempt to match fails. */
3994 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
3995 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
3996 : regstart[*p];
3997 DEBUG_PRINT2 (" old_regstart: %d\n",
3998 POINTER_TO_OFFSET (old_regstart[*p]));
3999
4000 regstart[*p] = d;
4001 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4002
4003 IS_ACTIVE (reg_info[*p]) = 1;
4004 MATCHED_SOMETHING (reg_info[*p]) = 0;
4005
4006 /* Clear this whenever we change the register activity status. */
4007 set_regs_matched_done = 0;
4008
4009 /* This is the new highest active register. */
4010 highest_active_reg = *p;
4011
4012 /* If nothing was active before, this is the new lowest active
4013 register. */
4014 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4015 lowest_active_reg = *p;
4016
4017 /* Move past the register number and inner group count. */
4018 p += 2;
4019 just_past_start_mem = p;
4020
4021 break;
4022
4023
4024 /* The stop_memory opcode represents the end of a group. Its
4025 arguments are the same as start_memory's: the register
4026 number, and the number of inner groups. */
4027 case stop_memory:
4028 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4029
4030 /* We need to save the string position the last time we were at
4031 this close-group operator in case the group is operated
4032 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4033 against `aba'; then we want to ignore where we are now in
4034 the string in case this attempt to match fails. */
4035 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4036 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4037 : regend[*p];
4038 DEBUG_PRINT2 (" old_regend: %d\n",
4039 POINTER_TO_OFFSET (old_regend[*p]));
4040
4041 regend[*p] = d;
4042 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4043
4044 /* This register isn't active anymore. */
4045 IS_ACTIVE (reg_info[*p]) = 0;
4046
4047 /* Clear this whenever we change the register activity status. */
4048 set_regs_matched_done = 0;
4049
4050 /* If this was the only register active, nothing is active
4051 anymore. */
4052 if (lowest_active_reg == highest_active_reg)
4053 {
4054 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4055 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4056 }
4057 else
4058 { /* We must scan for the new highest active register, since
4059 it isn't necessarily one less than now: consider
4060 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4061 new highest active register is 1. */
4062 unsigned char r = *p - 1;
4063 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4064 r--;
4065
4066 /* If we end up at register zero, that means that we saved
4067 the registers as the result of an `on_failure_jump', not
4068 a `start_memory', and we jumped to past the innermost
4069 `stop_memory'. For example, in ((.)*) we save
4070 registers 1 and 2 as a result of the *, but when we pop
4071 back to the second ), we are at the stop_memory 1.
4072 Thus, nothing is active. */
4073 if (r == 0)
4074 {
4075 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4076 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4077 }
4078 else
4079 highest_active_reg = r;
4080 }
4081
4082 /* If just failed to match something this time around with a
4083 group that's operated on by a repetition operator, try to
4084 force exit from the ``loop'', and restore the register
4085 information for this group that we had before trying this
4086 last match. */
4087 if ((!MATCHED_SOMETHING (reg_info[*p])
4088 || just_past_start_mem == p - 1)
4089 && (p + 2) < pend)
4090 {
4091 boolean is_a_jump_n = false;
4092
4093 p1 = p + 2;
4094 mcnt = 0;
4095 switch ((re_opcode_t) *p1++)
4096 {
4097 case jump_n:
4098 is_a_jump_n = true;
4099 case pop_failure_jump:
4100 case maybe_pop_jump:
4101 case jump:
4102 case dummy_failure_jump:
4103 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4104 if (is_a_jump_n)
4105 p1 += 2;
4106 break;
4107
4108 default:
4109 /* do nothing */ ;
4110 }
4111 p1 += mcnt;
4112
4113 /* If the next operation is a jump backwards in the pattern
4114 to an on_failure_jump right before the start_memory
4115 corresponding to this stop_memory, exit from the loop
4116 by forcing a failure after pushing on the stack the
4117 on_failure_jump's jump in the pattern, and d. */
4118 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4119 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4120 {
4121 /* If this group ever matched anything, then restore
4122 what its registers were before trying this last
4123 failed match, e.g., with `(a*)*b' against `ab' for
4124 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4125 against `aba' for regend[3].
4126
4127 Also restore the registers for inner groups for,
4128 e.g., `((a*)(b*))*' against `aba' (register 3 would
4129 otherwise get trashed). */
4130
4131 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4132 {
4133 unsigned r;
4134
4135 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4136
4137 /* Restore this and inner groups' (if any) registers. */
4138 for (r = *p; r < *p + *(p + 1); r++)
4139 {
4140 regstart[r] = old_regstart[r];
4141
4142 /* xx why this test? */
4143 if (old_regend[r] >= regstart[r])
4144 regend[r] = old_regend[r];
4145 }
4146 }
4147 p1++;
4148 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4149 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4150
4151 goto fail;
4152 }
4153 }
4154
4155 /* Move past the register number and the inner group count. */
4156 p += 2;
4157 break;
4158
4159
4160 /* \<digit> has been turned into a `duplicate' command which is
4161 followed by the numeric value of <digit> as the register number. */
4162 case duplicate:
4163 {
4164 register const char *d2, *dend2;
4165 int regno = *p++; /* Get which register to match against. */
4166 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4167
4168 /* Can't back reference a group which we've never matched. */
4169 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4170 goto fail;
4171
4172 /* Where in input to try to start matching. */
4173 d2 = regstart[regno];
4174
4175 /* Where to stop matching; if both the place to start and
4176 the place to stop matching are in the same string, then
4177 set to the place to stop, otherwise, for now have to use
4178 the end of the first string. */
4179
4180 dend2 = ((FIRST_STRING_P (regstart[regno])
4181 == FIRST_STRING_P (regend[regno]))
4182 ? regend[regno] : end_match_1);
4183 for (;;)
4184 {
4185 /* If necessary, advance to next segment in register
4186 contents. */
4187 while (d2 == dend2)
4188 {
4189 if (dend2 == end_match_2) break;
4190 if (dend2 == regend[regno]) break;
4191
4192 /* End of string1 => advance to string2. */
4193 d2 = string2;
4194 dend2 = regend[regno];
4195 }
4196 /* At end of register contents => success */
4197 if (d2 == dend2) break;
4198
4199 /* If necessary, advance to next segment in data. */
4200 PREFETCH ();
4201
4202 /* How many characters left in this segment to match. */
4203 mcnt = dend - d;
4204
4205 /* Want how many consecutive characters we can match in
4206 one shot, so, if necessary, adjust the count. */
4207 if (mcnt > dend2 - d2)
4208 mcnt = dend2 - d2;
4209
4210 /* Compare that many; failure if mismatch, else move
4211 past them. */
4212 if (translate
4213 ? bcmp_translate (d, d2, mcnt, translate)
4214 : bcmp (d, d2, mcnt))
4215 goto fail;
4216 d += mcnt, d2 += mcnt;
4217
4218 /* Do this because we've match some characters. */
4219 SET_REGS_MATCHED ();
4220 }
4221 }
4222 break;
4223
4224
4225 /* begline matches the empty string at the beginning of the string
4226 (unless `not_bol' is set in `bufp'), and, if
4227 `newline_anchor' is set, after newlines. */
4228 case begline:
4229 DEBUG_PRINT1 ("EXECUTING begline.\n");
4230
4231 if (AT_STRINGS_BEG (d))
4232 {
4233 if (!bufp->not_bol) break;
4234 }
4235 else if (d[-1] == '\n' && bufp->newline_anchor)
4236 {
4237 break;
4238 }
4239 /* In all other cases, we fail. */
4240 goto fail;
4241
4242
4243 /* endline is the dual of begline. */
4244 case endline:
4245 DEBUG_PRINT1 ("EXECUTING endline.\n");
4246
4247 if (AT_STRINGS_END (d))
4248 {
4249 if (!bufp->not_eol) break;
4250 }
4251
4252 /* We have to ``prefetch'' the next character. */
4253 else if ((d == end1 ? *string2 : *d) == '\n'
4254 && bufp->newline_anchor)
4255 {
4256 break;
4257 }
4258 goto fail;
4259
4260
4261 /* Match at the very beginning of the data. */
4262 case begbuf:
4263 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4264 if (AT_STRINGS_BEG (d))
4265 break;
4266 goto fail;
4267
4268
4269 /* Match at the very end of the data. */
4270 case endbuf:
4271 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4272 if (AT_STRINGS_END (d))
4273 break;
4274 goto fail;
4275
4276
4277 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4278 pushes NULL as the value for the string on the stack. Then
4279 `pop_failure_point' will keep the current value for the
4280 string, instead of restoring it. To see why, consider
4281 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4282 then the . fails against the \n. But the next thing we want
4283 to do is match the \n against the \n; if we restored the
4284 string value, we would be back at the foo.
4285
4286 Because this is used only in specific cases, we don't need to
4287 check all the things that `on_failure_jump' does, to make
4288 sure the right things get saved on the stack. Hence we don't
4289 share its code. The only reason to push anything on the
4290 stack at all is that otherwise we would have to change
4291 `anychar's code to do something besides goto fail in this
4292 case; that seems worse than this. */
4293 case on_failure_keep_string_jump:
4294 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4295
4296 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4297 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4298
4299 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4300 break;
4301
4302
4303 /* Uses of on_failure_jump:
4304
4305 Each alternative starts with an on_failure_jump that points
4306 to the beginning of the next alternative. Each alternative
4307 except the last ends with a jump that in effect jumps past
4308 the rest of the alternatives. (They really jump to the
4309 ending jump of the following alternative, because tensioning
4310 these jumps is a hassle.)
4311
4312 Repeats start with an on_failure_jump that points past both
4313 the repetition text and either the following jump or
4314 pop_failure_jump back to this on_failure_jump. */
4315 case on_failure_jump:
4316 on_failure:
4317 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4318
4319 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4320 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4321
4322 /* If this on_failure_jump comes right before a group (i.e.,
4323 the original * applied to a group), save the information
4324 for that group and all inner ones, so that if we fail back
4325 to this point, the group's information will be correct.
4326 For example, in \(a*\)*\1, we need the preceding group,
4327 and in \(\(a*\)b*\)\2, we need the inner group. */
4328
4329 /* We can't use `p' to check ahead because we push
4330 a failure point to `p + mcnt' after we do this. */
4331 p1 = p;
4332
4333 /* We need to skip no_op's before we look for the
4334 start_memory in case this on_failure_jump is happening as
4335 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4336 against aba. */
4337 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4338 p1++;
4339
4340 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4341 {
4342 /* We have a new highest active register now. This will
4343 get reset at the start_memory we are about to get to,
4344 but we will have saved all the registers relevant to
4345 this repetition op, as described above. */
4346 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4347 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4348 lowest_active_reg = *(p1 + 1);
4349 }
4350
4351 DEBUG_PRINT1 (":\n");
4352 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4353 break;
4354
4355
4356 /* A smart repeat ends with `maybe_pop_jump'.
4357 We change it to either `pop_failure_jump' or `jump'. */
4358 case maybe_pop_jump:
4359 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4360 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4361 {
4362 register unsigned char *p2 = p;
4363
4364 /* Compare the beginning of the repeat with what in the
4365 pattern follows its end. If we can establish that there
4366 is nothing that they would both match, i.e., that we
4367 would have to backtrack because of (as in, e.g., `a*a')
4368 then we can change to pop_failure_jump, because we'll
4369 never have to backtrack.
4370
4371 This is not true in the case of alternatives: in
4372 `(a|ab)*' we do need to backtrack to the `ab' alternative
4373 (e.g., if the string was `ab'). But instead of trying to
4374 detect that here, the alternative has put on a dummy
4375 failure point which is what we will end up popping. */
4376
4377 /* Skip over open/close-group commands.
4378 If what follows this loop is a ...+ construct,
4379 look at what begins its body, since we will have to
4380 match at least one of that. */
4381 while (1)
4382 {
4383 if (p2 + 2 < pend
4384 && ((re_opcode_t) *p2 == stop_memory
4385 || (re_opcode_t) *p2 == start_memory))
4386 p2 += 3;
4387 else if (p2 + 6 < pend
4388 && (re_opcode_t) *p2 == dummy_failure_jump)
4389 p2 += 6;
4390 else
4391 break;
4392 }
4393
4394 p1 = p + mcnt;
4395 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4396 to the `maybe_finalize_jump' of this case. Examine what
4397 follows. */
4398
4399 /* If we're at the end of the pattern, we can change. */
4400 if (p2 == pend)
4401 {
4402 /* Consider what happens when matching ":\(.*\)"
4403 against ":/". I don't really understand this code
4404 yet. */
4405 p[-3] = (unsigned char) pop_failure_jump;
4406 DEBUG_PRINT1
4407 (" End of pattern: change to `pop_failure_jump'.\n");
4408 }
4409
4410 else if ((re_opcode_t) *p2 == exactn
4411 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4412 {
4413 register unsigned char c
4414 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4415
4416 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4417 {
4418 p[-3] = (unsigned char) pop_failure_jump;
4419 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4420 c, p1[5]);
4421 }
4422
4423 else if ((re_opcode_t) p1[3] == charset
4424 || (re_opcode_t) p1[3] == charset_not)
4425 {
4426 int not = (re_opcode_t) p1[3] == charset_not;
4427
4428 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4429 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4430 not = !not;
4431
4432 /* `not' is equal to 1 if c would match, which means
4433 that we can't change to pop_failure_jump. */
4434 if (!not)
4435 {
4436 p[-3] = (unsigned char) pop_failure_jump;
4437 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4438 }
4439 }
4440 }
4441 else if ((re_opcode_t) *p2 == charset)
4442 {
4443 #ifdef DEBUG
4444 register unsigned char c
4445 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4446 #endif
4447
4448 if ((re_opcode_t) p1[3] == exactn
4449 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4450 && (p2[1 + p1[4] / BYTEWIDTH]
4451 & (1 << (p1[4] % BYTEWIDTH)))))
4452 {
4453 p[-3] = (unsigned char) pop_failure_jump;
4454 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4455 c, p1[5]);
4456 }
4457
4458 else if ((re_opcode_t) p1[3] == charset_not)
4459 {
4460 int idx;
4461 /* We win if the charset_not inside the loop
4462 lists every character listed in the charset after. */
4463 for (idx = 0; idx < (int) p2[1]; idx++)
4464 if (! (p2[2 + idx] == 0
4465 || (idx < (int) p1[4]
4466 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4467 break;
4468
4469 if (idx == p2[1])
4470 {
4471 p[-3] = (unsigned char) pop_failure_jump;
4472 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4473 }
4474 }
4475 else if ((re_opcode_t) p1[3] == charset)
4476 {
4477 int idx;
4478 /* We win if the charset inside the loop
4479 has no overlap with the one after the loop. */
4480 for (idx = 0;
4481 idx < (int) p2[1] && idx < (int) p1[4];
4482 idx++)
4483 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4484 break;
4485
4486 if (idx == p2[1] || idx == p1[4])
4487 {
4488 p[-3] = (unsigned char) pop_failure_jump;
4489 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4490 }
4491 }
4492 }
4493 }
4494 p -= 2; /* Point at relative address again. */
4495 if ((re_opcode_t) p[-1] != pop_failure_jump)
4496 {
4497 p[-1] = (unsigned char) jump;
4498 DEBUG_PRINT1 (" Match => jump.\n");
4499 goto unconditional_jump;
4500 }
4501 /* Note fall through. */
4502
4503
4504 /* The end of a simple repeat has a pop_failure_jump back to
4505 its matching on_failure_jump, where the latter will push a
4506 failure point. The pop_failure_jump takes off failure
4507 points put on by this pop_failure_jump's matching
4508 on_failure_jump; we got through the pattern to here from the
4509 matching on_failure_jump, so didn't fail. */
4510 case pop_failure_jump:
4511 {
4512 /* We need to pass separate storage for the lowest and
4513 highest registers, even though we don't care about the
4514 actual values. Otherwise, we will restore only one
4515 register from the stack, since lowest will == highest in
4516 `pop_failure_point'. */
4517 unsigned dummy_low_reg, dummy_high_reg;
4518 unsigned char *pdummy;
4519 const char *sdummy;
4520
4521 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4522 POP_FAILURE_POINT (sdummy, pdummy,
4523 dummy_low_reg, dummy_high_reg,
4524 reg_dummy, reg_dummy, reg_info_dummy);
4525 }
4526 /* Note fall through. */
4527
4528
4529 /* Unconditionally jump (without popping any failure points). */
4530 case jump:
4531 unconditional_jump:
4532 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4533 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4534 p += mcnt; /* Do the jump. */
4535 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4536 break;
4537
4538
4539 /* We need this opcode so we can detect where alternatives end
4540 in `group_match_null_string_p' et al. */
4541 case jump_past_alt:
4542 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4543 goto unconditional_jump;
4544
4545
4546 /* Normally, the on_failure_jump pushes a failure point, which
4547 then gets popped at pop_failure_jump. We will end up at
4548 pop_failure_jump, also, and with a pattern of, say, `a+', we
4549 are skipping over the on_failure_jump, so we have to push
4550 something meaningless for pop_failure_jump to pop. */
4551 case dummy_failure_jump:
4552 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4553 /* It doesn't matter what we push for the string here. What
4554 the code at `fail' tests is the value for the pattern. */
4555 PUSH_FAILURE_POINT (0, 0, -2);
4556 goto unconditional_jump;
4557
4558
4559 /* At the end of an alternative, we need to push a dummy failure
4560 point in case we are followed by a `pop_failure_jump', because
4561 we don't want the failure point for the alternative to be
4562 popped. For example, matching `(a|ab)*' against `aab'
4563 requires that we match the `ab' alternative. */
4564 case push_dummy_failure:
4565 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4566 /* See comments just above at `dummy_failure_jump' about the
4567 two zeroes. */
4568 PUSH_FAILURE_POINT (0, 0, -2);
4569 break;
4570
4571 /* Have to succeed matching what follows at least n times.
4572 After that, handle like `on_failure_jump'. */
4573 case succeed_n:
4574 EXTRACT_NUMBER (mcnt, p + 2);
4575 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4576
4577 assert (mcnt >= 0);
4578 /* Originally, this is how many times we HAVE to succeed. */
4579 if (mcnt > 0)
4580 {
4581 mcnt--;
4582 p += 2;
4583 STORE_NUMBER_AND_INCR (p, mcnt);
4584 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4585 }
4586 else if (mcnt == 0)
4587 {
4588 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4589 p[2] = (unsigned char) no_op;
4590 p[3] = (unsigned char) no_op;
4591 goto on_failure;
4592 }
4593 break;
4594
4595 case jump_n:
4596 EXTRACT_NUMBER (mcnt, p + 2);
4597 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4598
4599 /* Originally, this is how many times we CAN jump. */
4600 if (mcnt)
4601 {
4602 mcnt--;
4603 STORE_NUMBER (p + 2, mcnt);
4604 goto unconditional_jump;
4605 }
4606 /* If don't have to jump any more, skip over the rest of command. */
4607 else
4608 p += 4;
4609 break;
4610
4611 case set_number_at:
4612 {
4613 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4614
4615 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4616 p1 = p + mcnt;
4617 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4618 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4619 STORE_NUMBER (p1, mcnt);
4620 break;
4621 }
4622
4623 case wordbound:
4624 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4625 if (AT_WORD_BOUNDARY (d))
4626 break;
4627 goto fail;
4628
4629 case notwordbound:
4630 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4631 if (AT_WORD_BOUNDARY (d))
4632 goto fail;
4633 break;
4634
4635 case wordbeg:
4636 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4637 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4638 break;
4639 goto fail;
4640
4641 case wordend:
4642 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4643 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4644 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4645 break;
4646 goto fail;
4647
4648 #ifdef emacs
4649 case before_dot:
4650 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4651 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4652 goto fail;
4653 break;
4654
4655 case at_dot:
4656 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4657 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4658 goto fail;
4659 break;
4660
4661 case after_dot:
4662 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4663 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4664 goto fail;
4665 break;
4666 #if 0 /* not emacs19 */
4667 case at_dot:
4668 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4669 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4670 goto fail;
4671 break;
4672 #endif /* not emacs19 */
4673
4674 case syntaxspec:
4675 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4676 mcnt = *p++;
4677 goto matchsyntax;
4678
4679 case wordchar:
4680 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4681 mcnt = (int) Sword;
4682 matchsyntax:
4683 PREFETCH ();
4684 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4685 d++;
4686 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4687 goto fail;
4688 SET_REGS_MATCHED ();
4689 break;
4690
4691 case notsyntaxspec:
4692 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4693 mcnt = *p++;
4694 goto matchnotsyntax;
4695
4696 case notwordchar:
4697 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4698 mcnt = (int) Sword;
4699 matchnotsyntax:
4700 PREFETCH ();
4701 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4702 d++;
4703 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4704 goto fail;
4705 SET_REGS_MATCHED ();
4706 break;
4707
4708 #else /* not emacs */
4709 case wordchar:
4710 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4711 PREFETCH ();
4712 if (!WORDCHAR_P (d))
4713 goto fail;
4714 SET_REGS_MATCHED ();
4715 d++;
4716 break;
4717
4718 case notwordchar:
4719 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4720 PREFETCH ();
4721 if (WORDCHAR_P (d))
4722 goto fail;
4723 SET_REGS_MATCHED ();
4724 d++;
4725 break;
4726 #endif /* not emacs */
4727
4728 default:
4729 abort ();
4730 }
4731 continue; /* Successfully executed one pattern command; keep going. */
4732
4733
4734 /* We goto here if a matching operation fails. */
4735 fail:
4736 if (!FAIL_STACK_EMPTY ())
4737 { /* A restart point is known. Restore to that state. */
4738 DEBUG_PRINT1 ("\nFAIL:\n");
4739 POP_FAILURE_POINT (d, p,
4740 lowest_active_reg, highest_active_reg,
4741 regstart, regend, reg_info);
4742
4743 /* If this failure point is a dummy, try the next one. */
4744 if (!p)
4745 goto fail;
4746
4747 /* If we failed to the end of the pattern, don't examine *p. */
4748 assert (p <= pend);
4749 if (p < pend)
4750 {
4751 boolean is_a_jump_n = false;
4752
4753 /* If failed to a backwards jump that's part of a repetition
4754 loop, need to pop this failure point and use the next one. */
4755 switch ((re_opcode_t) *p)
4756 {
4757 case jump_n:
4758 is_a_jump_n = true;
4759 case maybe_pop_jump:
4760 case pop_failure_jump:
4761 case jump:
4762 p1 = p + 1;
4763 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4764 p1 += mcnt;
4765
4766 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4767 || (!is_a_jump_n
4768 && (re_opcode_t) *p1 == on_failure_jump))
4769 goto fail;
4770 break;
4771 default:
4772 /* do nothing */ ;
4773 }
4774 }
4775
4776 if (d >= string1 && d <= end1)
4777 dend = end_match_1;
4778 }
4779 else
4780 break; /* Matching at this starting point really fails. */
4781 } /* for (;;) */
4782
4783 if (best_regs_set)
4784 goto restore_best_regs;
4785
4786 FREE_VARIABLES ();
4787
4788 return -1; /* Failure to match. */
4789 } /* re_match_2 */
4790
4791 /* Subroutine definitions for re_match_2. */
4792
4793
4794 /* We are passed P pointing to a register number after a start_memory.
4795
4796 Return true if the pattern up to the corresponding stop_memory can
4797 match the empty string, and false otherwise.
4798
4799 If we find the matching stop_memory, sets P to point to one past its number.
4800 Otherwise, sets P to an undefined byte less than or equal to END.
4801
4802 We don't handle duplicates properly (yet). */
4803
4804 static boolean
4805 group_match_null_string_p (p, end, reg_info)
4806 unsigned char **p, *end;
4807 register_info_type *reg_info;
4808 {
4809 int mcnt;
4810 /* Point to after the args to the start_memory. */
4811 unsigned char *p1 = *p + 2;
4812
4813 while (p1 < end)
4814 {
4815 /* Skip over opcodes that can match nothing, and return true or
4816 false, as appropriate, when we get to one that can't, or to the
4817 matching stop_memory. */
4818
4819 switch ((re_opcode_t) *p1)
4820 {
4821 /* Could be either a loop or a series of alternatives. */
4822 case on_failure_jump:
4823 p1++;
4824 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4825
4826 /* If the next operation is not a jump backwards in the
4827 pattern. */
4828
4829 if (mcnt >= 0)
4830 {
4831 /* Go through the on_failure_jumps of the alternatives,
4832 seeing if any of the alternatives cannot match nothing.
4833 The last alternative starts with only a jump,
4834 whereas the rest start with on_failure_jump and end
4835 with a jump, e.g., here is the pattern for `a|b|c':
4836
4837 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4838 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4839 /exactn/1/c
4840
4841 So, we have to first go through the first (n-1)
4842 alternatives and then deal with the last one separately. */
4843
4844
4845 /* Deal with the first (n-1) alternatives, which start
4846 with an on_failure_jump (see above) that jumps to right
4847 past a jump_past_alt. */
4848
4849 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4850 {
4851 /* `mcnt' holds how many bytes long the alternative
4852 is, including the ending `jump_past_alt' and
4853 its number. */
4854
4855 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4856 reg_info))
4857 return false;
4858
4859 /* Move to right after this alternative, including the
4860 jump_past_alt. */
4861 p1 += mcnt;
4862
4863 /* Break if it's the beginning of an n-th alternative
4864 that doesn't begin with an on_failure_jump. */
4865 if ((re_opcode_t) *p1 != on_failure_jump)
4866 break;
4867
4868 /* Still have to check that it's not an n-th
4869 alternative that starts with an on_failure_jump. */
4870 p1++;
4871 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4872 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4873 {
4874 /* Get to the beginning of the n-th alternative. */
4875 p1 -= 3;
4876 break;
4877 }
4878 }
4879
4880 /* Deal with the last alternative: go back and get number
4881 of the `jump_past_alt' just before it. `mcnt' contains
4882 the length of the alternative. */
4883 EXTRACT_NUMBER (mcnt, p1 - 2);
4884
4885 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4886 return false;
4887
4888 p1 += mcnt; /* Get past the n-th alternative. */
4889 } /* if mcnt > 0 */
4890 break;
4891
4892
4893 case stop_memory:
4894 assert (p1[1] == **p);
4895 *p = p1 + 2;
4896 return true;
4897
4898
4899 default:
4900 if (!common_op_match_null_string_p (&p1, end, reg_info))
4901 return false;
4902 }
4903 } /* while p1 < end */
4904
4905 return false;
4906 } /* group_match_null_string_p */
4907
4908
4909 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4910 It expects P to be the first byte of a single alternative and END one
4911 byte past the last. The alternative can contain groups. */
4912
4913 static boolean
4914 alt_match_null_string_p (p, end, reg_info)
4915 unsigned char *p, *end;
4916 register_info_type *reg_info;
4917 {
4918 int mcnt;
4919 unsigned char *p1 = p;
4920
4921 while (p1 < end)
4922 {
4923 /* Skip over opcodes that can match nothing, and break when we get
4924 to one that can't. */
4925
4926 switch ((re_opcode_t) *p1)
4927 {
4928 /* It's a loop. */
4929 case on_failure_jump:
4930 p1++;
4931 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4932 p1 += mcnt;
4933 break;
4934
4935 default:
4936 if (!common_op_match_null_string_p (&p1, end, reg_info))
4937 return false;
4938 }
4939 } /* while p1 < end */
4940
4941 return true;
4942 } /* alt_match_null_string_p */
4943
4944
4945 /* Deals with the ops common to group_match_null_string_p and
4946 alt_match_null_string_p.
4947
4948 Sets P to one after the op and its arguments, if any. */
4949
4950 static boolean
4951 common_op_match_null_string_p (p, end, reg_info)
4952 unsigned char **p, *end;
4953 register_info_type *reg_info;
4954 {
4955 int mcnt;
4956 boolean ret;
4957 int reg_no;
4958 unsigned char *p1 = *p;
4959
4960 switch ((re_opcode_t) *p1++)
4961 {
4962 case no_op:
4963 case begline:
4964 case endline:
4965 case begbuf:
4966 case endbuf:
4967 case wordbeg:
4968 case wordend:
4969 case wordbound:
4970 case notwordbound:
4971 #ifdef emacs
4972 case before_dot:
4973 case at_dot:
4974 case after_dot:
4975 #endif
4976 break;
4977
4978 case start_memory:
4979 reg_no = *p1;
4980 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
4981 ret = group_match_null_string_p (&p1, end, reg_info);
4982
4983 /* Have to set this here in case we're checking a group which
4984 contains a group and a back reference to it. */
4985
4986 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
4987 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
4988
4989 if (!ret)
4990 return false;
4991 break;
4992
4993 /* If this is an optimized succeed_n for zero times, make the jump. */
4994 case jump:
4995 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4996 if (mcnt >= 0)
4997 p1 += mcnt;
4998 else
4999 return false;
5000 break;
5001
5002 case succeed_n:
5003 /* Get to the number of times to succeed. */
5004 p1 += 2;
5005 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5006
5007 if (mcnt == 0)
5008 {
5009 p1 -= 4;
5010 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5011 p1 += mcnt;
5012 }
5013 else
5014 return false;
5015 break;
5016
5017 case duplicate:
5018 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5019 return false;
5020 break;
5021
5022 case set_number_at:
5023 p1 += 4;
5024
5025 default:
5026 /* All other opcodes mean we cannot match the empty string. */
5027 return false;
5028 }
5029
5030 *p = p1;
5031 return true;
5032 } /* common_op_match_null_string_p */
5033
5034
5035 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5036 bytes; nonzero otherwise. */
5037
5038 static int
5039 bcmp_translate (s1, s2, len, translate)
5040 unsigned char *s1, *s2;
5041 register int len;
5042 char *translate;
5043 {
5044 register unsigned char *p1 = s1, *p2 = s2;
5045 while (len)
5046 {
5047 if (translate[*p1++] != translate[*p2++]) return 1;
5048 len--;
5049 }
5050 return 0;
5051 }
5052
5053 /* Entry points for GNU code. */
5054
5055 /* re_compile_pattern is the GNU regular expression compiler: it
5056 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5057 Returns 0 if the pattern was valid, otherwise an error string.
5058
5059 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5060 are set in BUFP on entry.
5061
5062 We call regex_compile to do the actual compilation. */
5063
5064 const char *
5065 re_compile_pattern (pattern, length, bufp)
5066 const char *pattern;
5067 int length;
5068 struct re_pattern_buffer *bufp;
5069 {
5070 reg_errcode_t ret;
5071
5072 /* GNU code is written to assume at least RE_NREGS registers will be set
5073 (and at least one extra will be -1). */
5074 bufp->regs_allocated = REGS_UNALLOCATED;
5075
5076 /* And GNU code determines whether or not to get register information
5077 by passing null for the REGS argument to re_match, etc., not by
5078 setting no_sub. */
5079 bufp->no_sub = 0;
5080
5081 /* Match anchors at newline. */
5082 bufp->newline_anchor = 1;
5083
5084 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5085
5086 if (!ret)
5087 return NULL;
5088 return gettext (re_error_msgid[(int) ret]);
5089 }
5090
5091 /* Entry points compatible with 4.2 BSD regex library. We don't define
5092 them unless specifically requested. */
5093
5094 #ifdef _REGEX_RE_COMP
5095
5096 /* BSD has one and only one pattern buffer. */
5097 static struct re_pattern_buffer re_comp_buf;
5098
5099 char *
5100 re_comp (s)
5101 const char *s;
5102 {
5103 reg_errcode_t ret;
5104
5105 if (!s)
5106 {
5107 if (!re_comp_buf.buffer)
5108 return gettext ("No previous regular expression");
5109 return 0;
5110 }
5111
5112 if (!re_comp_buf.buffer)
5113 {
5114 re_comp_buf.buffer = (unsigned char *) malloc (200);
5115 if (re_comp_buf.buffer == NULL)
5116 return gettext (re_error_msgid[(int) REG_ESPACE]);
5117 re_comp_buf.allocated = 200;
5118
5119 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5120 if (re_comp_buf.fastmap == NULL)
5121 return gettext (re_error_msgid[(int) REG_ESPACE]);
5122 }
5123
5124 /* Since `re_exec' always passes NULL for the `regs' argument, we
5125 don't need to initialize the pattern buffer fields which affect it. */
5126
5127 /* Match anchors at newlines. */
5128 re_comp_buf.newline_anchor = 1;
5129
5130 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5131
5132 if (!ret)
5133 return NULL;
5134
5135 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5136 return (char *) gettext (re_error_msgid[(int) ret]);
5137 }
5138
5139
5140 int
5141 re_exec (s)
5142 const char *s;
5143 {
5144 const int len = strlen (s);
5145 return
5146 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5147 }
5148 #endif /* _REGEX_RE_COMP */
5149
5150 /* POSIX.2 functions. Don't define these for Emacs. */
5151
5152 #ifndef emacs
5153
5154 /* regcomp takes a regular expression as a string and compiles it.
5155
5156 PREG is a regex_t *. We do not expect any fields to be initialized,
5157 since POSIX says we shouldn't. Thus, we set
5158
5159 `buffer' to the compiled pattern;
5160 `used' to the length of the compiled pattern;
5161 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5162 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5163 RE_SYNTAX_POSIX_BASIC;
5164 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5165 `fastmap' and `fastmap_accurate' to zero;
5166 `re_nsub' to the number of subexpressions in PATTERN.
5167
5168 PATTERN is the address of the pattern string.
5169
5170 CFLAGS is a series of bits which affect compilation.
5171
5172 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5173 use POSIX basic syntax.
5174
5175 If REG_NEWLINE is set, then . and [^...] don't match newline.
5176 Also, regexec will try a match beginning after every newline.
5177
5178 If REG_ICASE is set, then we considers upper- and lowercase
5179 versions of letters to be equivalent when matching.
5180
5181 If REG_NOSUB is set, then when PREG is passed to regexec, that
5182 routine will report only success or failure, and nothing about the
5183 registers.
5184
5185 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5186 the return codes and their meanings.) */
5187
5188 int
5189 regcomp (preg, pattern, cflags)
5190 regex_t *preg;
5191 const char *pattern;
5192 int cflags;
5193 {
5194 reg_errcode_t ret;
5195 unsigned syntax
5196 = (cflags & REG_EXTENDED) ?
5197 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5198
5199 /* regex_compile will allocate the space for the compiled pattern. */
5200 preg->buffer = 0;
5201 preg->allocated = 0;
5202 preg->used = 0;
5203
5204 /* Don't bother to use a fastmap when searching. This simplifies the
5205 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5206 characters after newlines into the fastmap. This way, we just try
5207 every character. */
5208 preg->fastmap = 0;
5209
5210 if (cflags & REG_ICASE)
5211 {
5212 unsigned i;
5213
5214 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5215 if (preg->translate == NULL)
5216 return (int) REG_ESPACE;
5217
5218 /* Map uppercase characters to corresponding lowercase ones. */
5219 for (i = 0; i < CHAR_SET_SIZE; i++)
5220 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5221 }
5222 else
5223 preg->translate = NULL;
5224
5225 /* If REG_NEWLINE is set, newlines are treated differently. */
5226 if (cflags & REG_NEWLINE)
5227 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5228 syntax &= ~RE_DOT_NEWLINE;
5229 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5230 /* It also changes the matching behavior. */
5231 preg->newline_anchor = 1;
5232 }
5233 else
5234 preg->newline_anchor = 0;
5235
5236 preg->no_sub = !!(cflags & REG_NOSUB);
5237
5238 /* POSIX says a null character in the pattern terminates it, so we
5239 can use strlen here in compiling the pattern. */
5240 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5241
5242 /* POSIX doesn't distinguish between an unmatched open-group and an
5243 unmatched close-group: both are REG_EPAREN. */
5244 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5245
5246 return (int) ret;
5247 }
5248
5249
5250 /* regexec searches for a given pattern, specified by PREG, in the
5251 string STRING.
5252
5253 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5254 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5255 least NMATCH elements, and we set them to the offsets of the
5256 corresponding matched substrings.
5257
5258 EFLAGS specifies `execution flags' which affect matching: if
5259 REG_NOTBOL is set, then ^ does not match at the beginning of the
5260 string; if REG_NOTEOL is set, then $ does not match at the end.
5261
5262 We return 0 if we find a match and REG_NOMATCH if not. */
5263
5264 int
5265 regexec (preg, string, nmatch, pmatch, eflags)
5266 const regex_t *preg;
5267 const char *string;
5268 size_t nmatch;
5269 regmatch_t pmatch[];
5270 int eflags;
5271 {
5272 int ret;
5273 struct re_registers regs;
5274 regex_t private_preg;
5275 int len = strlen (string);
5276 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5277
5278 private_preg = *preg;
5279
5280 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5281 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5282
5283 /* The user has told us exactly how many registers to return
5284 information about, via `nmatch'. We have to pass that on to the
5285 matching routines. */
5286 private_preg.regs_allocated = REGS_FIXED;
5287
5288 if (want_reg_info)
5289 {
5290 regs.num_regs = nmatch;
5291 regs.start = TALLOC (nmatch, regoff_t);
5292 regs.end = TALLOC (nmatch, regoff_t);
5293 if (regs.start == NULL || regs.end == NULL)
5294 return (int) REG_NOMATCH;
5295 }
5296
5297 /* Perform the searching operation. */
5298 ret = re_search (&private_preg, string, len,
5299 /* start: */ 0, /* range: */ len,
5300 want_reg_info ? &regs : (struct re_registers *) 0);
5301
5302 /* Copy the register information to the POSIX structure. */
5303 if (want_reg_info)
5304 {
5305 if (ret >= 0)
5306 {
5307 unsigned r;
5308
5309 for (r = 0; r < nmatch; r++)
5310 {
5311 pmatch[r].rm_so = regs.start[r];
5312 pmatch[r].rm_eo = regs.end[r];
5313 }
5314 }
5315
5316 /* If we needed the temporary register info, free the space now. */
5317 free (regs.start);
5318 free (regs.end);
5319 }
5320
5321 /* We want zero return to mean success, unlike `re_search'. */
5322 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5323 }
5324
5325
5326 /* Returns a message corresponding to an error code, ERRCODE, returned
5327 from either regcomp or regexec. We don't use PREG here. */
5328
5329 size_t
5330 regerror (errcode, preg, errbuf, errbuf_size)
5331 int errcode;
5332 const regex_t *preg;
5333 char *errbuf;
5334 size_t errbuf_size;
5335 {
5336 const char *msg;
5337 size_t msg_size;
5338
5339 if (errcode < 0
5340 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5341 /* Only error codes returned by the rest of the code should be passed
5342 to this routine. If we are given anything else, or if other regex
5343 code generates an invalid error code, then the program has a bug.
5344 Dump core so we can fix it. */
5345 abort ();
5346
5347 msg = gettext (re_error_msgid[errcode]);
5348
5349 msg_size = strlen (msg) + 1; /* Includes the null. */
5350
5351 if (errbuf_size != 0)
5352 {
5353 if (msg_size > errbuf_size)
5354 {
5355 strncpy (errbuf, msg, errbuf_size - 1);
5356 errbuf[errbuf_size - 1] = 0;
5357 }
5358 else
5359 strcpy (errbuf, msg);
5360 }
5361
5362 return msg_size;
5363 }
5364
5365
5366 /* Free dynamically allocated space used by PREG. */
5367
5368 void
5369 regfree (preg)
5370 regex_t *preg;
5371 {
5372 if (preg->buffer != NULL)
5373 free (preg->buffer);
5374 preg->buffer = NULL;
5375
5376 preg->allocated = 0;
5377 preg->used = 0;
5378
5379 if (preg->fastmap != NULL)
5380 free (preg->fastmap);
5381 preg->fastmap = NULL;
5382 preg->fastmap_accurate = 0;
5383
5384 if (preg->translate != NULL)
5385 free (preg->translate);
5386 preg->translate = NULL;
5387 }
5388
5389 #endif /* not emacs */
5390
5391 /*
5392 Local variables:
5393 make-backup-files: t
5394 version-control: t
5395 trim-versions-without-asking: nil
5396 End:
5397 */