- Rearrange reactos.dff according to rosapps rearrange.
[reactos.git] / rosapps / applications / mc / src / regex.c
1 /* Extended regular expression matching and search library,
2 version 0.12.
3 (Implements POSIX draft P10003.2/D11.2, except for
4 internationalization features.)
5
6 Copyright (C) 1993, 1994, 1995 Free Software Foundation, Inc.
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 #if HAVE_LIBINTL_H || defined (_LIBC)
38 # include <libintl.h>
39 #else
40 # define gettext(msgid) (msgid)
41 #endif
42
43 /* The `emacs' switch turns on certain matching commands
44 that make sense only in Emacs. */
45 #ifdef emacs
46
47 #include "lisp.h"
48 #include "buffer.h"
49 #include "syntax.h"
50
51 #else /* not emacs */
52
53 /* If we are not linking with Emacs proper,
54 we can't use the relocating allocator
55 even if config.h says that we can. */
56 #undef REL_ALLOC
57
58 #if defined (STDC_HEADERS) || defined (_LIBC)
59 #include <stdlib.h>
60 #else
61 char *malloc ();
62 char *realloc ();
63 #endif
64
65 /* When used in Emacs's lib-src, we need to get bzero and bcopy somehow.
66 If nothing else has been done, use the method below. */
67 #ifdef INHIBIT_STRING_HEADER
68 #if !(defined (HAVE_BZERO) && defined (HAVE_BCOPY))
69 #if !defined (bzero) && !defined (bcopy)
70 #undef INHIBIT_STRING_HEADER
71 #endif
72 #endif
73 #endif
74
75 /* This is the normal way of making sure we have a bcopy and a bzero.
76 This is used in most programs--a few other programs avoid this
77 by defining INHIBIT_STRING_HEADER. */
78 #ifndef INHIBIT_STRING_HEADER
79 #if defined (HAVE_STRING_H) || defined (STDC_HEADERS) || defined (_LIBC)
80 #include <string.h>
81 #ifndef bcmp
82 #define bcmp(s1, s2, n) memcmp ((s1), (s2), (n))
83 #endif
84 #ifndef bcopy
85 #define bcopy(s, d, n) memcpy ((d), (s), (n))
86 #endif
87 #ifndef bzero
88 #define bzero(s, n) memset ((s), 0, (n))
89 #endif
90 #else
91 #include <strings.h>
92 #endif
93 #endif
94
95 /* Define the syntax stuff for \<, \>, etc. */
96
97 /* This must be nonzero for the wordchar and notwordchar pattern
98 commands in re_match_2. */
99 #ifndef Sword
100 #define Sword 1
101 #endif
102
103 #ifdef SWITCH_ENUM_BUG
104 #define SWITCH_ENUM_CAST(x) ((int)(x))
105 #else
106 #define SWITCH_ENUM_CAST(x) (x)
107 #endif
108
109 #ifdef SYNTAX_TABLE
110
111 extern char *re_syntax_table;
112
113 #else /* not SYNTAX_TABLE */
114
115 /* How many characters in the character set. */
116 #define CHAR_SET_SIZE 256
117
118 static char re_syntax_table[CHAR_SET_SIZE];
119
120 static void
121 init_syntax_once ()
122 {
123 register int c;
124 static int done = 0;
125
126 if (done)
127 return;
128
129 bzero (re_syntax_table, sizeof re_syntax_table);
130
131 for (c = 'a'; c <= 'z'; c++)
132 re_syntax_table[c] = Sword;
133
134 for (c = 'A'; c <= 'Z'; c++)
135 re_syntax_table[c] = Sword;
136
137 for (c = '0'; c <= '9'; c++)
138 re_syntax_table[c] = Sword;
139
140 re_syntax_table['_'] = Sword;
141
142 done = 1;
143 }
144
145 #endif /* not SYNTAX_TABLE */
146
147 #define SYNTAX(c) re_syntax_table[c]
148
149 #endif /* not emacs */
150 \f
151 /* Get the interface, including the syntax bits. */
152 #include "regex.h"
153
154 /* isalpha etc. are used for the character classes. */
155 #include <ctype.h>
156
157 /* Jim Meyering writes:
158
159 "... Some ctype macros are valid only for character codes that
160 isascii says are ASCII (SGI's IRIX-4.0.5 is one such system --when
161 using /bin/cc or gcc but without giving an ansi option). So, all
162 ctype uses should be through macros like ISPRINT... If
163 STDC_HEADERS is defined, then autoconf has verified that the ctype
164 macros don't need to be guarded with references to isascii. ...
165 Defining isascii to 1 should let any compiler worth its salt
166 eliminate the && through constant folding." */
167
168 #if defined (STDC_HEADERS) || (!defined (isascii) && !defined (HAVE_ISASCII))
169 #define ISASCII(c) 1
170 #else
171 #define ISASCII(c) isascii(c)
172 #endif
173
174 #ifdef isblank
175 #define ISBLANK(c) (ISASCII (c) && isblank (c))
176 #else
177 #define ISBLANK(c) ((c) == ' ' || (c) == '\t')
178 #endif
179 #ifdef isgraph
180 #define ISGRAPH(c) (ISASCII (c) && isgraph (c))
181 #else
182 #define ISGRAPH(c) (ISASCII (c) && isprint (c) && !isspace (c))
183 #endif
184
185 #define ISPRINT(c) (ISASCII (c) && isprint (c))
186 #define ISDIGIT(c) (ISASCII (c) && isdigit (c))
187 #define ISALNUM(c) (ISASCII (c) && isalnum (c))
188 #define ISALPHA(c) (ISASCII (c) && isalpha (c))
189 #define ISCNTRL(c) (ISASCII (c) && iscntrl (c))
190 #define ISLOWER(c) (ISASCII (c) && islower (c))
191 #define ISPUNCT(c) (ISASCII (c) && ispunct (c))
192 #define ISSPACE(c) (ISASCII (c) && isspace (c))
193 #define ISUPPER(c) (ISASCII (c) && isupper (c))
194 #define ISXDIGIT(c) (ISASCII (c) && isxdigit (c))
195
196 #ifndef NULL
197 #define NULL (void *)0
198 #endif
199
200 /* We remove any previous definition of `SIGN_EXTEND_CHAR',
201 since ours (we hope) works properly with all combinations of
202 machines, compilers, `char' and `unsigned char' argument types.
203 (Per Bothner suggested the basic approach.) */
204 #undef SIGN_EXTEND_CHAR
205 #if __STDC__
206 #define SIGN_EXTEND_CHAR(c) ((signed char) (c))
207 #else /* not __STDC__ */
208 /* As in Harbison and Steele. */
209 #define SIGN_EXTEND_CHAR(c) ((((unsigned char) (c)) ^ 128) - 128)
210 #endif
211 \f
212 /* Should we use malloc or alloca? If REGEX_MALLOC is not defined, we
213 use `alloca' instead of `malloc'. This is because using malloc in
214 re_search* or re_match* could cause memory leaks when C-g is used in
215 Emacs; also, malloc is slower and causes storage fragmentation. On
216 the other hand, malloc is more portable, and easier to debug.
217
218 Because we sometimes use alloca, some routines have to be macros,
219 not functions -- `alloca'-allocated space disappears at the end of the
220 function it is called in. */
221
222 #ifdef REGEX_MALLOC
223
224 #define REGEX_ALLOCATE malloc
225 #define REGEX_REALLOCATE(source, osize, nsize) realloc (source, nsize)
226 #define REGEX_FREE free
227
228 #else /* not REGEX_MALLOC */
229
230 /* Emacs already defines alloca, sometimes. */
231 #ifndef alloca
232
233 /* Make alloca work the best possible way. */
234 #ifdef __GNUC__
235 #define alloca __builtin_alloca
236 #else /* not __GNUC__ */
237 #if HAVE_ALLOCA_H
238 #include <alloca.h>
239 #else /* not __GNUC__ or HAVE_ALLOCA_H */
240 #ifndef _AIX /* Already did AIX, up at the top. */
241 char *alloca ();
242 #endif /* not _AIX */
243 #endif /* not HAVE_ALLOCA_H */
244 #endif /* not __GNUC__ */
245
246 #endif /* not alloca */
247
248 #define REGEX_ALLOCATE alloca
249
250 /* Assumes a `char *destination' variable. */
251 #define REGEX_REALLOCATE(source, osize, nsize) \
252 (destination = (char *) alloca (nsize), \
253 bcopy (source, destination, osize), \
254 destination)
255
256 /* No need to do anything to free, after alloca. */
257 #define REGEX_FREE(arg) ((void)0) /* Do nothing! But inhibit gcc warning. */
258
259 #endif /* not REGEX_MALLOC */
260
261 /* Define how to allocate the failure stack. */
262
263 #ifdef REL_ALLOC
264 #define REGEX_ALLOCATE_STACK(size) \
265 r_alloc (&failure_stack_ptr, (size))
266 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
267 r_re_alloc (&failure_stack_ptr, (nsize))
268 #define REGEX_FREE_STACK(ptr) \
269 r_alloc_free (&failure_stack_ptr)
270
271 #else /* not REL_ALLOC */
272
273 #ifdef REGEX_MALLOC
274
275 #define REGEX_ALLOCATE_STACK malloc
276 #define REGEX_REALLOCATE_STACK(source, osize, nsize) realloc (source, nsize)
277 #define REGEX_FREE_STACK free
278
279 #else /* not REGEX_MALLOC */
280
281 #define REGEX_ALLOCATE_STACK alloca
282
283 #define REGEX_REALLOCATE_STACK(source, osize, nsize) \
284 REGEX_REALLOCATE (source, osize, nsize)
285 /* No need to explicitly free anything. */
286 #define REGEX_FREE_STACK(arg)
287
288 #endif /* not REGEX_MALLOC */
289 #endif /* not REL_ALLOC */
290
291
292 /* True if `size1' is non-NULL and PTR is pointing anywhere inside
293 `string1' or just past its end. This works if PTR is NULL, which is
294 a good thing. */
295 #define FIRST_STRING_P(ptr) \
296 (size1 && string1 <= (ptr) && (ptr) <= string1 + size1)
297
298 /* (Re)Allocate N items of type T using malloc, or fail. */
299 #define TALLOC(n, t) ((t *) malloc ((n) * sizeof (t)))
300 #define RETALLOC(addr, n, t) ((addr) = (t *) realloc (addr, (n) * sizeof (t)))
301 #define RETALLOC_IF(addr, n, t) \
302 if (addr) RETALLOC((addr), (n), t); else (addr) = TALLOC ((n), t)
303 #define REGEX_TALLOC(n, t) ((t *) REGEX_ALLOCATE ((n) * sizeof (t)))
304
305 #define BYTEWIDTH 8 /* In bits. */
306
307 #define STREQ(s1, s2) ((strcmp (s1, s2) == 0))
308
309 #undef MAX
310 #undef MIN
311 #define MAX(a, b) ((a) > (b) ? (a) : (b))
312 #define MIN(a, b) ((a) < (b) ? (a) : (b))
313
314 typedef char boolean;
315 #define false 0
316 #define true 1
317
318 static int re_match_2_internal ();
319 \f
320 /* These are the command codes that appear in compiled regular
321 expressions. Some opcodes are followed by argument bytes. A
322 command code can specify any interpretation whatsoever for its
323 arguments. Zero bytes may appear in the compiled regular expression. */
324
325 typedef enum
326 {
327 no_op = 0,
328
329 /* Succeed right away--no more backtracking. */
330 succeed,
331
332 /* Followed by one byte giving n, then by n literal bytes. */
333 exactn,
334
335 /* Matches any (more or less) character. */
336 anychar,
337
338 /* Matches any one char belonging to specified set. First
339 following byte is number of bitmap bytes. Then come bytes
340 for a bitmap saying which chars are in. Bits in each byte
341 are ordered low-bit-first. A character is in the set if its
342 bit is 1. A character too large to have a bit in the map is
343 automatically not in the set. */
344 charset,
345
346 /* Same parameters as charset, but match any character that is
347 not one of those specified. */
348 charset_not,
349
350 /* Start remembering the text that is matched, for storing in a
351 register. Followed by one byte with the register number, in
352 the range 0 to one less than the pattern buffer's re_nsub
353 field. Then followed by one byte with the number of groups
354 inner to this one. (This last has to be part of the
355 start_memory only because we need it in the on_failure_jump
356 of re_match_2.) */
357 start_memory,
358
359 /* Stop remembering the text that is matched and store it in a
360 memory register. Followed by one byte with the register
361 number, in the range 0 to one less than `re_nsub' in the
362 pattern buffer, and one byte with the number of inner groups,
363 just like `start_memory'. (We need the number of inner
364 groups here because we don't have any easy way of finding the
365 corresponding start_memory when we're at a stop_memory.) */
366 stop_memory,
367
368 /* Match a duplicate of something remembered. Followed by one
369 byte containing the register number. */
370 duplicate,
371
372 /* Fail unless at beginning of line. */
373 begline,
374
375 /* Fail unless at end of line. */
376 endline,
377
378 /* Succeeds if at beginning of buffer (if emacs) or at beginning
379 of string to be matched (if not). */
380 begbuf,
381
382 /* Analogously, for end of buffer/string. */
383 endbuf,
384
385 /* Followed by two byte relative address to which to jump. */
386 jump,
387
388 /* Same as jump, but marks the end of an alternative. */
389 jump_past_alt,
390
391 /* Followed by two-byte relative address of place to resume at
392 in case of failure. */
393 on_failure_jump,
394
395 /* Like on_failure_jump, but pushes a placeholder instead of the
396 current string position when executed. */
397 on_failure_keep_string_jump,
398
399 /* Throw away latest failure point and then jump to following
400 two-byte relative address. */
401 pop_failure_jump,
402
403 /* Change to pop_failure_jump if know won't have to backtrack to
404 match; otherwise change to jump. This is used to jump
405 back to the beginning of a repeat. If what follows this jump
406 clearly won't match what the repeat does, such that we can be
407 sure that there is no use backtracking out of repetitions
408 already matched, then we change it to a pop_failure_jump.
409 Followed by two-byte address. */
410 maybe_pop_jump,
411
412 /* Jump to following two-byte address, and push a dummy failure
413 point. This failure point will be thrown away if an attempt
414 is made to use it for a failure. A `+' construct makes this
415 before the first repeat. Also used as an intermediary kind
416 of jump when compiling an alternative. */
417 dummy_failure_jump,
418
419 /* Push a dummy failure point and continue. Used at the end of
420 alternatives. */
421 push_dummy_failure,
422
423 /* Followed by two-byte relative address and two-byte number n.
424 After matching N times, jump to the address upon failure. */
425 succeed_n,
426
427 /* Followed by two-byte relative address, and two-byte number n.
428 Jump to the address N times, then fail. */
429 jump_n,
430
431 /* Set the following two-byte relative address to the
432 subsequent two-byte number. The address *includes* the two
433 bytes of number. */
434 set_number_at,
435
436 wordchar, /* Matches any word-constituent character. */
437 notwordchar, /* Matches any char that is not a word-constituent. */
438
439 wordbeg, /* Succeeds if at word beginning. */
440 wordend, /* Succeeds if at word end. */
441
442 wordbound, /* Succeeds if at a word boundary. */
443 notwordbound /* Succeeds if not at a word boundary. */
444
445 #ifdef emacs
446 ,before_dot, /* Succeeds if before point. */
447 at_dot, /* Succeeds if at point. */
448 after_dot, /* Succeeds if after point. */
449
450 /* Matches any character whose syntax is specified. Followed by
451 a byte which contains a syntax code, e.g., Sword. */
452 syntaxspec,
453
454 /* Matches any character whose syntax is not that specified. */
455 notsyntaxspec
456 #endif /* emacs */
457 } re_opcode_t;
458 \f
459 /* Common operations on the compiled pattern. */
460
461 /* Store NUMBER in two contiguous bytes starting at DESTINATION. */
462
463 #define STORE_NUMBER(destination, number) \
464 do { \
465 (destination)[0] = (number) & 0377; \
466 (destination)[1] = (number) >> 8; \
467 } while (0)
468
469 /* Same as STORE_NUMBER, except increment DESTINATION to
470 the byte after where the number is stored. Therefore, DESTINATION
471 must be an lvalue. */
472
473 #define STORE_NUMBER_AND_INCR(destination, number) \
474 do { \
475 STORE_NUMBER (destination, number); \
476 (destination) += 2; \
477 } while (0)
478
479 /* Put into DESTINATION a number stored in two contiguous bytes starting
480 at SOURCE. */
481
482 #define EXTRACT_NUMBER(destination, source) \
483 do { \
484 (destination) = *(source) & 0377; \
485 (destination) += SIGN_EXTEND_CHAR (*((source) + 1)) << 8; \
486 } while (0)
487
488 #ifdef DEBUG
489 static void
490 extract_number (dest, source)
491 int *dest;
492 unsigned char *source;
493 {
494 int temp = SIGN_EXTEND_CHAR (*(source + 1));
495 *dest = *source & 0377;
496 *dest += temp << 8;
497 }
498
499 #ifndef EXTRACT_MACROS /* To debug the macros. */
500 #undef EXTRACT_NUMBER
501 #define EXTRACT_NUMBER(dest, src) extract_number (&dest, src)
502 #endif /* not EXTRACT_MACROS */
503
504 #endif /* DEBUG */
505
506 /* Same as EXTRACT_NUMBER, except increment SOURCE to after the number.
507 SOURCE must be an lvalue. */
508
509 #define EXTRACT_NUMBER_AND_INCR(destination, source) \
510 do { \
511 EXTRACT_NUMBER (destination, source); \
512 (source) += 2; \
513 } while (0)
514
515 #ifdef DEBUG
516 static void
517 extract_number_and_incr (destination, source)
518 int *destination;
519 unsigned char **source;
520 {
521 extract_number (destination, *source);
522 *source += 2;
523 }
524
525 #ifndef EXTRACT_MACROS
526 #undef EXTRACT_NUMBER_AND_INCR
527 #define EXTRACT_NUMBER_AND_INCR(dest, src) \
528 extract_number_and_incr (&dest, &src)
529 #endif /* not EXTRACT_MACROS */
530
531 #endif /* DEBUG */
532 \f
533 /* If DEBUG is defined, Regex prints many voluminous messages about what
534 it is doing (if the variable `debug' is nonzero). If linked with the
535 main program in `iregex.c', you can enter patterns and strings
536 interactively. And if linked with the main program in `main.c' and
537 the other test files, you can run the already-written tests. */
538
539 #ifdef DEBUG
540
541 /* We use standard I/O for debugging. */
542 #include <stdio.h>
543
544 /* It is useful to test things that ``must'' be true when debugging. */
545 #include <assert.h>
546
547 static int debug = 0;
548
549 #define DEBUG_STATEMENT(e) e
550 #define DEBUG_PRINT1(x) if (debug) printf (x)
551 #define DEBUG_PRINT2(x1, x2) if (debug) printf (x1, x2)
552 #define DEBUG_PRINT3(x1, x2, x3) if (debug) printf (x1, x2, x3)
553 #define DEBUG_PRINT4(x1, x2, x3, x4) if (debug) printf (x1, x2, x3, x4)
554 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e) \
555 if (debug) print_partial_compiled_pattern (s, e)
556 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2) \
557 if (debug) print_double_string (w, s1, sz1, s2, sz2)
558
559
560 /* Print the fastmap in human-readable form. */
561
562 void
563 print_fastmap (fastmap)
564 char *fastmap;
565 {
566 unsigned was_a_range = 0;
567 unsigned i = 0;
568
569 while (i < (1 << BYTEWIDTH))
570 {
571 if (fastmap[i++])
572 {
573 was_a_range = 0;
574 putchar (i - 1);
575 while (i < (1 << BYTEWIDTH) && fastmap[i])
576 {
577 was_a_range = 1;
578 i++;
579 }
580 if (was_a_range)
581 {
582 printf ("-");
583 putchar (i - 1);
584 }
585 }
586 }
587 putchar ('\n');
588 }
589
590
591 /* Print a compiled pattern string in human-readable form, starting at
592 the START pointer into it and ending just before the pointer END. */
593
594 void
595 print_partial_compiled_pattern (start, end)
596 unsigned char *start;
597 unsigned char *end;
598 {
599 int mcnt, mcnt2;
600 unsigned char *p = start;
601 unsigned char *pend = end;
602
603 if (start == NULL)
604 {
605 printf ("(null)\n");
606 return;
607 }
608
609 /* Loop over pattern commands. */
610 while (p < pend)
611 {
612 printf ("%d:\t", p - start);
613
614 switch ((re_opcode_t) *p++)
615 {
616 case no_op:
617 printf ("/no_op");
618 break;
619
620 case exactn:
621 mcnt = *p++;
622 printf ("/exactn/%d", mcnt);
623 do
624 {
625 putchar ('/');
626 putchar (*p++);
627 }
628 while (--mcnt);
629 break;
630
631 case start_memory:
632 mcnt = *p++;
633 printf ("/start_memory/%d/%d", mcnt, *p++);
634 break;
635
636 case stop_memory:
637 mcnt = *p++;
638 printf ("/stop_memory/%d/%d", mcnt, *p++);
639 break;
640
641 case duplicate:
642 printf ("/duplicate/%d", *p++);
643 break;
644
645 case anychar:
646 printf ("/anychar");
647 break;
648
649 case charset:
650 case charset_not:
651 {
652 register int c, last = -100;
653 register int in_range = 0;
654
655 printf ("/charset [%s",
656 (re_opcode_t) *(p - 1) == charset_not ? "^" : "");
657
658 assert (p + *p < pend);
659
660 for (c = 0; c < 256; c++)
661 if (c / 8 < *p
662 && (p[1 + (c/8)] & (1 << (c % 8))))
663 {
664 /* Are we starting a range? */
665 if (last + 1 == c && ! in_range)
666 {
667 putchar ('-');
668 in_range = 1;
669 }
670 /* Have we broken a range? */
671 else if (last + 1 != c && in_range)
672 {
673 putchar (last);
674 in_range = 0;
675 }
676
677 if (! in_range)
678 putchar (c);
679
680 last = c;
681 }
682
683 if (in_range)
684 putchar (last);
685
686 putchar (']');
687
688 p += 1 + *p;
689 }
690 break;
691
692 case begline:
693 printf ("/begline");
694 break;
695
696 case endline:
697 printf ("/endline");
698 break;
699
700 case on_failure_jump:
701 extract_number_and_incr (&mcnt, &p);
702 printf ("/on_failure_jump to %d", p + mcnt - start);
703 break;
704
705 case on_failure_keep_string_jump:
706 extract_number_and_incr (&mcnt, &p);
707 printf ("/on_failure_keep_string_jump to %d", p + mcnt - start);
708 break;
709
710 case dummy_failure_jump:
711 extract_number_and_incr (&mcnt, &p);
712 printf ("/dummy_failure_jump to %d", p + mcnt - start);
713 break;
714
715 case push_dummy_failure:
716 printf ("/push_dummy_failure");
717 break;
718
719 case maybe_pop_jump:
720 extract_number_and_incr (&mcnt, &p);
721 printf ("/maybe_pop_jump to %d", p + mcnt - start);
722 break;
723
724 case pop_failure_jump:
725 extract_number_and_incr (&mcnt, &p);
726 printf ("/pop_failure_jump to %d", p + mcnt - start);
727 break;
728
729 case jump_past_alt:
730 extract_number_and_incr (&mcnt, &p);
731 printf ("/jump_past_alt to %d", p + mcnt - start);
732 break;
733
734 case jump:
735 extract_number_and_incr (&mcnt, &p);
736 printf ("/jump to %d", p + mcnt - start);
737 break;
738
739 case succeed_n:
740 extract_number_and_incr (&mcnt, &p);
741 extract_number_and_incr (&mcnt2, &p);
742 printf ("/succeed_n to %d, %d times", p + mcnt - start, mcnt2);
743 break;
744
745 case jump_n:
746 extract_number_and_incr (&mcnt, &p);
747 extract_number_and_incr (&mcnt2, &p);
748 printf ("/jump_n to %d, %d times", p + mcnt - start, mcnt2);
749 break;
750
751 case set_number_at:
752 extract_number_and_incr (&mcnt, &p);
753 extract_number_and_incr (&mcnt2, &p);
754 printf ("/set_number_at location %d to %d", p + mcnt - start, mcnt2);
755 break;
756
757 case wordbound:
758 printf ("/wordbound");
759 break;
760
761 case notwordbound:
762 printf ("/notwordbound");
763 break;
764
765 case wordbeg:
766 printf ("/wordbeg");
767 break;
768
769 case wordend:
770 printf ("/wordend");
771
772 #ifdef emacs
773 case before_dot:
774 printf ("/before_dot");
775 break;
776
777 case at_dot:
778 printf ("/at_dot");
779 break;
780
781 case after_dot:
782 printf ("/after_dot");
783 break;
784
785 case syntaxspec:
786 printf ("/syntaxspec");
787 mcnt = *p++;
788 printf ("/%d", mcnt);
789 break;
790
791 case notsyntaxspec:
792 printf ("/notsyntaxspec");
793 mcnt = *p++;
794 printf ("/%d", mcnt);
795 break;
796 #endif /* emacs */
797
798 case wordchar:
799 printf ("/wordchar");
800 break;
801
802 case notwordchar:
803 printf ("/notwordchar");
804 break;
805
806 case begbuf:
807 printf ("/begbuf");
808 break;
809
810 case endbuf:
811 printf ("/endbuf");
812 break;
813
814 default:
815 printf ("?%d", *(p-1));
816 }
817
818 putchar ('\n');
819 }
820
821 printf ("%d:\tend of pattern.\n", p - start);
822 }
823
824
825 void
826 print_compiled_pattern (bufp)
827 struct re_pattern_buffer *bufp;
828 {
829 unsigned char *buffer = bufp->buffer;
830
831 print_partial_compiled_pattern (buffer, buffer + bufp->used);
832 printf ("%d bytes used/%d bytes allocated.\n", bufp->used, bufp->allocated);
833
834 if (bufp->fastmap_accurate && bufp->fastmap)
835 {
836 printf ("fastmap: ");
837 print_fastmap (bufp->fastmap);
838 }
839
840 printf ("re_nsub: %d\t", bufp->re_nsub);
841 printf ("regs_alloc: %d\t", bufp->regs_allocated);
842 printf ("can_be_null: %d\t", bufp->can_be_null);
843 printf ("newline_anchor: %d\n", bufp->newline_anchor);
844 printf ("no_sub: %d\t", bufp->no_sub);
845 printf ("not_bol: %d\t", bufp->not_bol);
846 printf ("not_eol: %d\t", bufp->not_eol);
847 printf ("syntax: %d\n", bufp->syntax);
848 /* Perhaps we should print the translate table? */
849 }
850
851
852 void
853 print_double_string (where, string1, size1, string2, size2)
854 const char *where;
855 const char *string1;
856 const char *string2;
857 int size1;
858 int size2;
859 {
860 unsigned this_char;
861
862 if (where == NULL)
863 printf ("(null)");
864 else
865 {
866 if (FIRST_STRING_P (where))
867 {
868 for (this_char = where - string1; this_char < size1; this_char++)
869 putchar (string1[this_char]);
870
871 where = string2;
872 }
873
874 for (this_char = where - string2; this_char < size2; this_char++)
875 putchar (string2[this_char]);
876 }
877 }
878
879 #else /* not DEBUG */
880
881 #undef assert
882 #define assert(e)
883
884 #define DEBUG_STATEMENT(e)
885 #define DEBUG_PRINT1(x)
886 #define DEBUG_PRINT2(x1, x2)
887 #define DEBUG_PRINT3(x1, x2, x3)
888 #define DEBUG_PRINT4(x1, x2, x3, x4)
889 #define DEBUG_PRINT_COMPILED_PATTERN(p, s, e)
890 #define DEBUG_PRINT_DOUBLE_STRING(w, s1, sz1, s2, sz2)
891
892 #endif /* not DEBUG */
893 \f
894 /* Set by `re_set_syntax' to the current regexp syntax to recognize. Can
895 also be assigned to arbitrarily: each pattern buffer stores its own
896 syntax, so it can be changed between regex compilations. */
897 /* This has no initializer because initialized variables in Emacs
898 become read-only after dumping. */
899 reg_syntax_t re_syntax_options;
900
901
902 /* Specify the precise syntax of regexps for compilation. This provides
903 for compatibility for various utilities which historically have
904 different, incompatible syntaxes.
905
906 The argument SYNTAX is a bit mask comprised of the various bits
907 defined in regex.h. We return the old syntax. */
908
909 reg_syntax_t
910 re_set_syntax (syntax)
911 reg_syntax_t syntax;
912 {
913 reg_syntax_t ret = re_syntax_options;
914
915 re_syntax_options = syntax;
916 return ret;
917 }
918 \f
919 /* This table gives an error message for each of the error codes listed
920 in regex.h. Obviously the order here has to be same as there.
921 POSIX doesn't require that we do anything for REG_NOERROR,
922 but why not be nice? */
923
924 static const char *re_error_msgid[] =
925 { "Success", /* REG_NOERROR */
926 "No match", /* REG_NOMATCH */
927 "Invalid regular expression", /* REG_BADPAT */
928 "Invalid collation character", /* REG_ECOLLATE */
929 "Invalid character class name", /* REG_ECTYPE */
930 "Trailing backslash", /* REG_EESCAPE */
931 "Invalid back reference", /* REG_ESUBREG */
932 "Unmatched [ or [^", /* REG_EBRACK */
933 "Unmatched ( or \\(", /* REG_EPAREN */
934 "Unmatched \\{", /* REG_EBRACE */
935 "Invalid content of \\{\\}", /* REG_BADBR */
936 "Invalid range end", /* REG_ERANGE */
937 "Memory exhausted", /* REG_ESPACE */
938 "Invalid preceding regular expression", /* REG_BADRPT */
939 "Premature end of regular expression", /* REG_EEND */
940 "Regular expression too big", /* REG_ESIZE */
941 "Unmatched ) or \\)", /* REG_ERPAREN */
942 };
943 \f
944 /* Avoiding alloca during matching, to placate r_alloc. */
945
946 /* Define MATCH_MAY_ALLOCATE unless we need to make sure that the
947 searching and matching functions should not call alloca. On some
948 systems, alloca is implemented in terms of malloc, and if we're
949 using the relocating allocator routines, then malloc could cause a
950 relocation, which might (if the strings being searched are in the
951 ralloc heap) shift the data out from underneath the regexp
952 routines.
953
954 Here's another reason to avoid allocation: Emacs
955 processes input from X in a signal handler; processing X input may
956 call malloc; if input arrives while a matching routine is calling
957 malloc, then we're scrod. But Emacs can't just block input while
958 calling matching routines; then we don't notice interrupts when
959 they come in. So, Emacs blocks input around all regexp calls
960 except the matching calls, which it leaves unprotected, in the
961 faith that they will not malloc. */
962
963 /* Normally, this is fine. */
964 #define MATCH_MAY_ALLOCATE
965
966 /* When using GNU C, we are not REALLY using the C alloca, no matter
967 what config.h may say. So don't take precautions for it. */
968 #ifdef __GNUC__
969 #undef C_ALLOCA
970 #endif
971
972 /* The match routines may not allocate if (1) they would do it with malloc
973 and (2) it's not safe for them to use malloc.
974 Note that if REL_ALLOC is defined, matching would not use malloc for the
975 failure stack, but we would still use it for the register vectors;
976 so REL_ALLOC should not affect this. */
977 #if (defined (C_ALLOCA) || defined (REGEX_MALLOC)) && defined (emacs)
978 #undef MATCH_MAY_ALLOCATE
979 #endif
980
981 \f
982 /* Failure stack declarations and macros; both re_compile_fastmap and
983 re_match_2 use a failure stack. These have to be macros because of
984 REGEX_ALLOCATE_STACK. */
985
986
987 /* Number of failure points for which to initially allocate space
988 when matching. If this number is exceeded, we allocate more
989 space, so it is not a hard limit. */
990 #ifndef INIT_FAILURE_ALLOC
991 #define INIT_FAILURE_ALLOC 5
992 #endif
993
994 /* Roughly the maximum number of failure points on the stack. Would be
995 exactly that if always used MAX_FAILURE_SPACE each time we failed.
996 This is a variable only so users of regex can assign to it; we never
997 change it ourselves. */
998 #if defined (MATCH_MAY_ALLOCATE)
999 int re_max_failures = 200000;
1000 #else
1001 int re_max_failures = 2000;
1002 #endif
1003
1004 union fail_stack_elt
1005 {
1006 unsigned char *pointer;
1007 int integer;
1008 };
1009
1010 typedef union fail_stack_elt fail_stack_elt_t;
1011
1012 typedef struct
1013 {
1014 fail_stack_elt_t *stack;
1015 unsigned size;
1016 unsigned avail; /* Offset of next open position. */
1017 } fail_stack_type;
1018
1019 #define FAIL_STACK_EMPTY() (fail_stack.avail == 0)
1020 #define FAIL_STACK_PTR_EMPTY() (fail_stack_ptr->avail == 0)
1021 #define FAIL_STACK_FULL() (fail_stack.avail == fail_stack.size)
1022
1023
1024 /* Define macros to initialize and free the failure stack.
1025 Do `return -2' if the alloc fails. */
1026
1027 #ifdef MATCH_MAY_ALLOCATE
1028 #define INIT_FAIL_STACK() \
1029 do { \
1030 fail_stack.stack = (fail_stack_elt_t *) \
1031 REGEX_ALLOCATE_STACK (INIT_FAILURE_ALLOC * sizeof (fail_stack_elt_t)); \
1032 \
1033 if (fail_stack.stack == NULL) \
1034 return -2; \
1035 \
1036 fail_stack.size = INIT_FAILURE_ALLOC; \
1037 fail_stack.avail = 0; \
1038 } while (0)
1039
1040 #define RESET_FAIL_STACK() REGEX_FREE_STACK (fail_stack.stack)
1041 #else
1042 #define INIT_FAIL_STACK() \
1043 do { \
1044 fail_stack.avail = 0; \
1045 } while (0)
1046
1047 #define RESET_FAIL_STACK()
1048 #endif
1049
1050
1051 /* Double the size of FAIL_STACK, up to approximately `re_max_failures' items.
1052
1053 Return 1 if succeeds, and 0 if either ran out of memory
1054 allocating space for it or it was already too large.
1055
1056 REGEX_REALLOCATE_STACK requires `destination' be declared. */
1057
1058 #define DOUBLE_FAIL_STACK(fail_stack) \
1059 ((fail_stack).size > re_max_failures * MAX_FAILURE_ITEMS \
1060 ? 0 \
1061 : ((fail_stack).stack = (fail_stack_elt_t *) \
1062 REGEX_REALLOCATE_STACK ((fail_stack).stack, \
1063 (fail_stack).size * sizeof (fail_stack_elt_t), \
1064 ((fail_stack).size << 1) * sizeof (fail_stack_elt_t)), \
1065 \
1066 (fail_stack).stack == NULL \
1067 ? 0 \
1068 : ((fail_stack).size <<= 1, \
1069 1)))
1070
1071
1072 /* Push pointer POINTER on FAIL_STACK.
1073 Return 1 if was able to do so and 0 if ran out of memory allocating
1074 space to do so. */
1075 #define PUSH_PATTERN_OP(POINTER, FAIL_STACK) \
1076 ((FAIL_STACK_FULL () \
1077 && !DOUBLE_FAIL_STACK (FAIL_STACK)) \
1078 ? 0 \
1079 : ((FAIL_STACK).stack[(FAIL_STACK).avail++].pointer = POINTER, \
1080 1))
1081
1082 /* Push a pointer value onto the failure stack.
1083 Assumes the variable `fail_stack'. Probably should only
1084 be called from within `PUSH_FAILURE_POINT'. */
1085 #define PUSH_FAILURE_POINTER(item) \
1086 fail_stack.stack[fail_stack.avail++].pointer = (unsigned char *) (item)
1087
1088 /* This pushes an integer-valued item onto the failure stack.
1089 Assumes the variable `fail_stack'. Probably should only
1090 be called from within `PUSH_FAILURE_POINT'. */
1091 #define PUSH_FAILURE_INT(item) \
1092 fail_stack.stack[fail_stack.avail++].integer = (item)
1093
1094 /* Push a fail_stack_elt_t value onto the failure stack.
1095 Assumes the variable `fail_stack'. Probably should only
1096 be called from within `PUSH_FAILURE_POINT'. */
1097 #define PUSH_FAILURE_ELT(item) \
1098 fail_stack.stack[fail_stack.avail++] = (item)
1099
1100 /* These three POP... operations complement the three PUSH... operations.
1101 All assume that `fail_stack' is nonempty. */
1102 #define POP_FAILURE_POINTER() fail_stack.stack[--fail_stack.avail].pointer
1103 #define POP_FAILURE_INT() fail_stack.stack[--fail_stack.avail].integer
1104 #define POP_FAILURE_ELT() fail_stack.stack[--fail_stack.avail]
1105
1106 /* Used to omit pushing failure point id's when we're not debugging. */
1107 #ifdef DEBUG
1108 #define DEBUG_PUSH PUSH_FAILURE_INT
1109 #define DEBUG_POP(item_addr) *(item_addr) = POP_FAILURE_INT ()
1110 #else
1111 #define DEBUG_PUSH(item)
1112 #define DEBUG_POP(item_addr)
1113 #endif
1114
1115
1116 /* Push the information about the state we will need
1117 if we ever fail back to it.
1118
1119 Requires variables fail_stack, regstart, regend, reg_info, and
1120 num_regs be declared. DOUBLE_FAIL_STACK requires `destination' be
1121 declared.
1122
1123 Does `return FAILURE_CODE' if runs out of memory. */
1124
1125 #define PUSH_FAILURE_POINT(pattern_place, string_place, failure_code) \
1126 do { \
1127 char *destination; \
1128 /* Must be int, so when we don't save any registers, the arithmetic \
1129 of 0 + -1 isn't done as unsigned. */ \
1130 int this_reg; \
1131 \
1132 destination = 0; /* inhibit possible compiler warning */ \
1133 DEBUG_STATEMENT (failure_id++); \
1134 DEBUG_STATEMENT (nfailure_points_pushed++); \
1135 DEBUG_PRINT2 ("\nPUSH_FAILURE_POINT #%u:\n", failure_id); \
1136 DEBUG_PRINT2 (" Before push, next avail: %d\n", (fail_stack).avail);\
1137 DEBUG_PRINT2 (" size: %d\n", (fail_stack).size);\
1138 \
1139 DEBUG_PRINT2 (" slots needed: %d\n", NUM_FAILURE_ITEMS); \
1140 DEBUG_PRINT2 (" available: %d\n", REMAINING_AVAIL_SLOTS); \
1141 \
1142 /* Ensure we have enough space allocated for what we will push. */ \
1143 while (REMAINING_AVAIL_SLOTS < NUM_FAILURE_ITEMS) \
1144 { \
1145 if (!DOUBLE_FAIL_STACK (fail_stack)) \
1146 return failure_code; \
1147 \
1148 DEBUG_PRINT2 ("\n Doubled stack; size now: %d\n", \
1149 (fail_stack).size); \
1150 DEBUG_PRINT2 (" slots available: %d\n", REMAINING_AVAIL_SLOTS);\
1151 } \
1152 \
1153 /* Push the info, starting with the registers. */ \
1154 DEBUG_PRINT1 ("\n"); \
1155 \
1156 for (this_reg = lowest_active_reg; this_reg <= highest_active_reg; \
1157 this_reg++) \
1158 { \
1159 DEBUG_PRINT2 (" Pushing reg: %d\n", this_reg); \
1160 DEBUG_STATEMENT (num_regs_pushed++); \
1161 \
1162 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1163 PUSH_FAILURE_POINTER (regstart[this_reg]); \
1164 \
1165 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1166 PUSH_FAILURE_POINTER (regend[this_reg]); \
1167 \
1168 DEBUG_PRINT2 (" info: 0x%x\n ", reg_info[this_reg]); \
1169 DEBUG_PRINT2 (" match_null=%d", \
1170 REG_MATCH_NULL_STRING_P (reg_info[this_reg])); \
1171 DEBUG_PRINT2 (" active=%d", IS_ACTIVE (reg_info[this_reg])); \
1172 DEBUG_PRINT2 (" matched_something=%d", \
1173 MATCHED_SOMETHING (reg_info[this_reg])); \
1174 DEBUG_PRINT2 (" ever_matched=%d", \
1175 EVER_MATCHED_SOMETHING (reg_info[this_reg])); \
1176 DEBUG_PRINT1 ("\n"); \
1177 PUSH_FAILURE_ELT (reg_info[this_reg].word); \
1178 } \
1179 \
1180 DEBUG_PRINT2 (" Pushing low active reg: %d\n", lowest_active_reg);\
1181 PUSH_FAILURE_INT (lowest_active_reg); \
1182 \
1183 DEBUG_PRINT2 (" Pushing high active reg: %d\n", highest_active_reg);\
1184 PUSH_FAILURE_INT (highest_active_reg); \
1185 \
1186 DEBUG_PRINT2 (" Pushing pattern 0x%x: ", pattern_place); \
1187 DEBUG_PRINT_COMPILED_PATTERN (bufp, pattern_place, pend); \
1188 PUSH_FAILURE_POINTER (pattern_place); \
1189 \
1190 DEBUG_PRINT2 (" Pushing string 0x%x: `", string_place); \
1191 DEBUG_PRINT_DOUBLE_STRING (string_place, string1, size1, string2, \
1192 size2); \
1193 DEBUG_PRINT1 ("'\n"); \
1194 PUSH_FAILURE_POINTER (string_place); \
1195 \
1196 DEBUG_PRINT2 (" Pushing failure id: %u\n", failure_id); \
1197 DEBUG_PUSH (failure_id); \
1198 } while (0)
1199
1200 /* This is the number of items that are pushed and popped on the stack
1201 for each register. */
1202 #define NUM_REG_ITEMS 3
1203
1204 /* Individual items aside from the registers. */
1205 #ifdef DEBUG
1206 #define NUM_NONREG_ITEMS 5 /* Includes failure point id. */
1207 #else
1208 #define NUM_NONREG_ITEMS 4
1209 #endif
1210
1211 /* We push at most this many items on the stack. */
1212 #define MAX_FAILURE_ITEMS ((num_regs - 1) * NUM_REG_ITEMS + NUM_NONREG_ITEMS)
1213
1214 /* We actually push this many items. */
1215 #define NUM_FAILURE_ITEMS \
1216 ((highest_active_reg - lowest_active_reg + 1) * NUM_REG_ITEMS \
1217 + NUM_NONREG_ITEMS)
1218
1219 /* How many items can still be added to the stack without overflowing it. */
1220 #define REMAINING_AVAIL_SLOTS ((fail_stack).size - (fail_stack).avail)
1221
1222
1223 /* Pops what PUSH_FAIL_STACK pushes.
1224
1225 We restore into the parameters, all of which should be lvalues:
1226 STR -- the saved data position.
1227 PAT -- the saved pattern position.
1228 LOW_REG, HIGH_REG -- the highest and lowest active registers.
1229 REGSTART, REGEND -- arrays of string positions.
1230 REG_INFO -- array of information about each subexpression.
1231
1232 Also assumes the variables `fail_stack' and (if debugging), `bufp',
1233 `pend', `string1', `size1', `string2', and `size2'. */
1234
1235 #define POP_FAILURE_POINT(str, pat, low_reg, high_reg, regstart, regend, reg_info)\
1236 { \
1237 DEBUG_STATEMENT (fail_stack_elt_t failure_id;) \
1238 int this_reg; \
1239 const unsigned char *string_temp; \
1240 \
1241 assert (!FAIL_STACK_EMPTY ()); \
1242 \
1243 /* Remove failure points and point to how many regs pushed. */ \
1244 DEBUG_PRINT1 ("POP_FAILURE_POINT:\n"); \
1245 DEBUG_PRINT2 (" Before pop, next avail: %d\n", fail_stack.avail); \
1246 DEBUG_PRINT2 (" size: %d\n", fail_stack.size); \
1247 \
1248 assert (fail_stack.avail >= NUM_NONREG_ITEMS); \
1249 \
1250 DEBUG_POP (&failure_id); \
1251 DEBUG_PRINT2 (" Popping failure id: %u\n", failure_id); \
1252 \
1253 /* If the saved string location is NULL, it came from an \
1254 on_failure_keep_string_jump opcode, and we want to throw away the \
1255 saved NULL, thus retaining our current position in the string. */ \
1256 string_temp = POP_FAILURE_POINTER (); \
1257 if (string_temp != NULL) \
1258 str = (const char *) string_temp; \
1259 \
1260 DEBUG_PRINT2 (" Popping string 0x%x: `", str); \
1261 DEBUG_PRINT_DOUBLE_STRING (str, string1, size1, string2, size2); \
1262 DEBUG_PRINT1 ("'\n"); \
1263 \
1264 pat = (unsigned char *) POP_FAILURE_POINTER (); \
1265 DEBUG_PRINT2 (" Popping pattern 0x%x: ", pat); \
1266 DEBUG_PRINT_COMPILED_PATTERN (bufp, pat, pend); \
1267 \
1268 /* Restore register info. */ \
1269 high_reg = (unsigned) POP_FAILURE_INT (); \
1270 DEBUG_PRINT2 (" Popping high active reg: %d\n", high_reg); \
1271 \
1272 low_reg = (unsigned) POP_FAILURE_INT (); \
1273 DEBUG_PRINT2 (" Popping low active reg: %d\n", low_reg); \
1274 \
1275 for (this_reg = high_reg; this_reg >= low_reg; this_reg--) \
1276 { \
1277 DEBUG_PRINT2 (" Popping reg: %d\n", this_reg); \
1278 \
1279 reg_info[this_reg].word = POP_FAILURE_ELT (); \
1280 DEBUG_PRINT2 (" info: 0x%x\n", reg_info[this_reg]); \
1281 \
1282 regend[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1283 DEBUG_PRINT2 (" end: 0x%x\n", regend[this_reg]); \
1284 \
1285 regstart[this_reg] = (const char *) POP_FAILURE_POINTER (); \
1286 DEBUG_PRINT2 (" start: 0x%x\n", regstart[this_reg]); \
1287 } \
1288 \
1289 set_regs_matched_done = 0; \
1290 DEBUG_STATEMENT (nfailure_points_popped++); \
1291 } /* POP_FAILURE_POINT */
1292
1293
1294 \f
1295 /* Structure for per-register (a.k.a. per-group) information.
1296 Other register information, such as the
1297 starting and ending positions (which are addresses), and the list of
1298 inner groups (which is a bits list) are maintained in separate
1299 variables.
1300
1301 We are making a (strictly speaking) nonportable assumption here: that
1302 the compiler will pack our bit fields into something that fits into
1303 the type of `word', i.e., is something that fits into one item on the
1304 failure stack. */
1305
1306 typedef union
1307 {
1308 fail_stack_elt_t word;
1309 struct
1310 {
1311 /* This field is one if this group can match the empty string,
1312 zero if not. If not yet determined, `MATCH_NULL_UNSET_VALUE'. */
1313 #define MATCH_NULL_UNSET_VALUE 3
1314 unsigned match_null_string_p : 2;
1315 unsigned is_active : 1;
1316 unsigned matched_something : 1;
1317 unsigned ever_matched_something : 1;
1318 } bits;
1319 } register_info_type;
1320
1321 #define REG_MATCH_NULL_STRING_P(R) ((R).bits.match_null_string_p)
1322 #define IS_ACTIVE(R) ((R).bits.is_active)
1323 #define MATCHED_SOMETHING(R) ((R).bits.matched_something)
1324 #define EVER_MATCHED_SOMETHING(R) ((R).bits.ever_matched_something)
1325
1326
1327 /* Call this when have matched a real character; it sets `matched' flags
1328 for the subexpressions which we are currently inside. Also records
1329 that those subexprs have matched. */
1330 #define SET_REGS_MATCHED() \
1331 do \
1332 { \
1333 if (!set_regs_matched_done) \
1334 { \
1335 unsigned r; \
1336 set_regs_matched_done = 1; \
1337 for (r = lowest_active_reg; r <= highest_active_reg; r++) \
1338 { \
1339 MATCHED_SOMETHING (reg_info[r]) \
1340 = EVER_MATCHED_SOMETHING (reg_info[r]) \
1341 = 1; \
1342 } \
1343 } \
1344 } \
1345 while (0)
1346
1347 /* Registers are set to a sentinel when they haven't yet matched. */
1348 static char reg_unset_dummy;
1349 #define REG_UNSET_VALUE (&reg_unset_dummy)
1350 #define REG_UNSET(e) ((e) == REG_UNSET_VALUE)
1351 \f
1352 /* Subroutine declarations and macros for regex_compile. */
1353
1354 static void store_op1 (), store_op2 ();
1355 static void insert_op1 (), insert_op2 ();
1356 static boolean at_begline_loc_p (), at_endline_loc_p ();
1357 static boolean group_in_compile_stack ();
1358 static reg_errcode_t compile_range ();
1359
1360 /* Fetch the next character in the uncompiled pattern---translating it
1361 if necessary. Also cast from a signed character in the constant
1362 string passed to us by the user to an unsigned char that we can use
1363 as an array index (in, e.g., `translate'). */
1364 #define PATFETCH(c) \
1365 do {if (p == pend) return REG_EEND; \
1366 c = (unsigned char) *p++; \
1367 if (translate) c = translate[c]; \
1368 } while (0)
1369
1370 /* Fetch the next character in the uncompiled pattern, with no
1371 translation. */
1372 #define PATFETCH_RAW(c) \
1373 do {if (p == pend) return REG_EEND; \
1374 c = (unsigned char) *p++; \
1375 } while (0)
1376
1377 /* Go backwards one character in the pattern. */
1378 #define PATUNFETCH p--
1379
1380
1381 /* If `translate' is non-null, return translate[D], else just D. We
1382 cast the subscript to translate because some data is declared as
1383 `char *', to avoid warnings when a string constant is passed. But
1384 when we use a character as a subscript we must make it unsigned. */
1385 #define TRANSLATE(d) (translate ? translate[(unsigned char) (d)] : (d))
1386
1387
1388 /* Macros for outputting the compiled pattern into `buffer'. */
1389
1390 /* If the buffer isn't allocated when it comes in, use this. */
1391 #define INIT_BUF_SIZE 32
1392
1393 /* Make sure we have at least N more bytes of space in buffer. */
1394 #define GET_BUFFER_SPACE(n) \
1395 while (b - bufp->buffer + (n) > bufp->allocated) \
1396 EXTEND_BUFFER ()
1397
1398 /* Make sure we have one more byte of buffer space and then add C to it. */
1399 #define BUF_PUSH(c) \
1400 do { \
1401 GET_BUFFER_SPACE (1); \
1402 *b++ = (unsigned char) (c); \
1403 } while (0)
1404
1405
1406 /* Ensure we have two more bytes of buffer space and then append C1 and C2. */
1407 #define BUF_PUSH_2(c1, c2) \
1408 do { \
1409 GET_BUFFER_SPACE (2); \
1410 *b++ = (unsigned char) (c1); \
1411 *b++ = (unsigned char) (c2); \
1412 } while (0)
1413
1414
1415 /* As with BUF_PUSH_2, except for three bytes. */
1416 #define BUF_PUSH_3(c1, c2, c3) \
1417 do { \
1418 GET_BUFFER_SPACE (3); \
1419 *b++ = (unsigned char) (c1); \
1420 *b++ = (unsigned char) (c2); \
1421 *b++ = (unsigned char) (c3); \
1422 } while (0)
1423
1424
1425 /* Store a jump with opcode OP at LOC to location TO. We store a
1426 relative address offset by the three bytes the jump itself occupies. */
1427 #define STORE_JUMP(op, loc, to) \
1428 store_op1 (op, loc, (to) - (loc) - 3)
1429
1430 /* Likewise, for a two-argument jump. */
1431 #define STORE_JUMP2(op, loc, to, arg) \
1432 store_op2 (op, loc, (to) - (loc) - 3, arg)
1433
1434 /* Like `STORE_JUMP', but for inserting. Assume `b' is the buffer end. */
1435 #define INSERT_JUMP(op, loc, to) \
1436 insert_op1 (op, loc, (to) - (loc) - 3, b)
1437
1438 /* Like `STORE_JUMP2', but for inserting. Assume `b' is the buffer end. */
1439 #define INSERT_JUMP2(op, loc, to, arg) \
1440 insert_op2 (op, loc, (to) - (loc) - 3, arg, b)
1441
1442
1443 /* This is not an arbitrary limit: the arguments which represent offsets
1444 into the pattern are two bytes long. So if 2^16 bytes turns out to
1445 be too small, many things would have to change. */
1446 #define MAX_BUF_SIZE (1L << 16)
1447
1448
1449 /* Extend the buffer by twice its current size via realloc and
1450 reset the pointers that pointed into the old block to point to the
1451 correct places in the new one. If extending the buffer results in it
1452 being larger than MAX_BUF_SIZE, then flag memory exhausted. */
1453 #define EXTEND_BUFFER() \
1454 do { \
1455 unsigned char *old_buffer = bufp->buffer; \
1456 if (bufp->allocated == MAX_BUF_SIZE) \
1457 return REG_ESIZE; \
1458 bufp->allocated <<= 1; \
1459 if (bufp->allocated > MAX_BUF_SIZE) \
1460 bufp->allocated = MAX_BUF_SIZE; \
1461 bufp->buffer = (unsigned char *) realloc (bufp->buffer, bufp->allocated);\
1462 if (bufp->buffer == NULL) \
1463 return REG_ESPACE; \
1464 /* If the buffer moved, move all the pointers into it. */ \
1465 if (old_buffer != bufp->buffer) \
1466 { \
1467 b = (b - old_buffer) + bufp->buffer; \
1468 begalt = (begalt - old_buffer) + bufp->buffer; \
1469 if (fixup_alt_jump) \
1470 fixup_alt_jump = (fixup_alt_jump - old_buffer) + bufp->buffer;\
1471 if (laststart) \
1472 laststart = (laststart - old_buffer) + bufp->buffer; \
1473 if (pending_exact) \
1474 pending_exact = (pending_exact - old_buffer) + bufp->buffer; \
1475 } \
1476 } while (0)
1477
1478
1479 /* Since we have one byte reserved for the register number argument to
1480 {start,stop}_memory, the maximum number of groups we can report
1481 things about is what fits in that byte. */
1482 #define MAX_REGNUM 255
1483
1484 /* But patterns can have more than `MAX_REGNUM' registers. We just
1485 ignore the excess. */
1486 typedef unsigned regnum_t;
1487
1488
1489 /* Macros for the compile stack. */
1490
1491 /* Since offsets can go either forwards or backwards, this type needs to
1492 be able to hold values from -(MAX_BUF_SIZE - 1) to MAX_BUF_SIZE - 1. */
1493 typedef int pattern_offset_t;
1494
1495 typedef struct
1496 {
1497 pattern_offset_t begalt_offset;
1498 pattern_offset_t fixup_alt_jump;
1499 pattern_offset_t inner_group_offset;
1500 pattern_offset_t laststart_offset;
1501 regnum_t regnum;
1502 } compile_stack_elt_t;
1503
1504
1505 typedef struct
1506 {
1507 compile_stack_elt_t *stack;
1508 unsigned size;
1509 unsigned avail; /* Offset of next open position. */
1510 } compile_stack_type;
1511
1512
1513 #define INIT_COMPILE_STACK_SIZE 32
1514
1515 #define COMPILE_STACK_EMPTY (compile_stack.avail == 0)
1516 #define COMPILE_STACK_FULL (compile_stack.avail == compile_stack.size)
1517
1518 /* The next available element. */
1519 #define COMPILE_STACK_TOP (compile_stack.stack[compile_stack.avail])
1520
1521
1522 /* Set the bit for character C in a list. */
1523 #define SET_LIST_BIT(c) \
1524 (b[((unsigned char) (c)) / BYTEWIDTH] \
1525 |= 1 << (((unsigned char) c) % BYTEWIDTH))
1526
1527
1528 /* Get the next unsigned number in the uncompiled pattern. */
1529 #define GET_UNSIGNED_NUMBER(num) \
1530 { if (p != pend) \
1531 { \
1532 PATFETCH (c); \
1533 while (ISDIGIT (c)) \
1534 { \
1535 if (num < 0) \
1536 num = 0; \
1537 num = num * 10 + c - '0'; \
1538 if (p == pend) \
1539 break; \
1540 PATFETCH (c); \
1541 } \
1542 } \
1543 }
1544
1545 #define CHAR_CLASS_MAX_LENGTH 6 /* Namely, `xdigit'. */
1546
1547 #define IS_CHAR_CLASS(string) \
1548 (STREQ (string, "alpha") || STREQ (string, "upper") \
1549 || STREQ (string, "lower") || STREQ (string, "digit") \
1550 || STREQ (string, "alnum") || STREQ (string, "xdigit") \
1551 || STREQ (string, "space") || STREQ (string, "print") \
1552 || STREQ (string, "punct") || STREQ (string, "graph") \
1553 || STREQ (string, "cntrl") || STREQ (string, "blank"))
1554 \f
1555 #ifndef MATCH_MAY_ALLOCATE
1556
1557 /* If we cannot allocate large objects within re_match_2_internal,
1558 we make the fail stack and register vectors global.
1559 The fail stack, we grow to the maximum size when a regexp
1560 is compiled.
1561 The register vectors, we adjust in size each time we
1562 compile a regexp, according to the number of registers it needs. */
1563
1564 static fail_stack_type fail_stack;
1565
1566 /* Size with which the following vectors are currently allocated.
1567 That is so we can make them bigger as needed,
1568 but never make them smaller. */
1569 static int regs_allocated_size;
1570
1571 static const char ** regstart, ** regend;
1572 static const char ** old_regstart, ** old_regend;
1573 static const char **best_regstart, **best_regend;
1574 static register_info_type *reg_info;
1575 static const char **reg_dummy;
1576 static register_info_type *reg_info_dummy;
1577
1578 /* Make the register vectors big enough for NUM_REGS registers,
1579 but don't make them smaller. */
1580
1581 static
1582 regex_grow_registers (num_regs)
1583 int num_regs;
1584 {
1585 if (num_regs > regs_allocated_size)
1586 {
1587 RETALLOC_IF (regstart, num_regs, const char *);
1588 RETALLOC_IF (regend, num_regs, const char *);
1589 RETALLOC_IF (old_regstart, num_regs, const char *);
1590 RETALLOC_IF (old_regend, num_regs, const char *);
1591 RETALLOC_IF (best_regstart, num_regs, const char *);
1592 RETALLOC_IF (best_regend, num_regs, const char *);
1593 RETALLOC_IF (reg_info, num_regs, register_info_type);
1594 RETALLOC_IF (reg_dummy, num_regs, const char *);
1595 RETALLOC_IF (reg_info_dummy, num_regs, register_info_type);
1596
1597 regs_allocated_size = num_regs;
1598 }
1599 }
1600
1601 #endif /* not MATCH_MAY_ALLOCATE */
1602 \f
1603 /* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX.
1604 Returns one of error codes defined in `regex.h', or zero for success.
1605
1606 Assumes the `allocated' (and perhaps `buffer') and `translate'
1607 fields are set in BUFP on entry.
1608
1609 If it succeeds, results are put in BUFP (if it returns an error, the
1610 contents of BUFP are undefined):
1611 `buffer' is the compiled pattern;
1612 `syntax' is set to SYNTAX;
1613 `used' is set to the length of the compiled pattern;
1614 `fastmap_accurate' is zero;
1615 `re_nsub' is the number of subexpressions in PATTERN;
1616 `not_bol' and `not_eol' are zero;
1617
1618 The `fastmap' and `newline_anchor' fields are neither
1619 examined nor set. */
1620
1621 /* Return, freeing storage we allocated. */
1622 #define FREE_STACK_RETURN(value) \
1623 return (free (compile_stack.stack), value)
1624
1625 static reg_errcode_t
1626 regex_compile (pattern, size, syntax, bufp)
1627 const char *pattern;
1628 int size;
1629 reg_syntax_t syntax;
1630 struct re_pattern_buffer *bufp;
1631 {
1632 /* We fetch characters from PATTERN here. Even though PATTERN is
1633 `char *' (i.e., signed), we declare these variables as unsigned, so
1634 they can be reliably used as array indices. */
1635 register unsigned char c, c1;
1636
1637 /* A random temporary spot in PATTERN. */
1638 const char *p1;
1639
1640 /* Points to the end of the buffer, where we should append. */
1641 register unsigned char *b;
1642
1643 /* Keeps track of unclosed groups. */
1644 compile_stack_type compile_stack;
1645
1646 /* Points to the current (ending) position in the pattern. */
1647 const char *p = pattern;
1648 const char *pend = pattern + size;
1649
1650 /* How to translate the characters in the pattern. */
1651 char *translate = bufp->translate;
1652
1653 /* Address of the count-byte of the most recently inserted `exactn'
1654 command. This makes it possible to tell if a new exact-match
1655 character can be added to that command or if the character requires
1656 a new `exactn' command. */
1657 unsigned char *pending_exact = 0;
1658
1659 /* Address of start of the most recently finished expression.
1660 This tells, e.g., postfix * where to find the start of its
1661 operand. Reset at the beginning of groups and alternatives. */
1662 unsigned char *laststart = 0;
1663
1664 /* Address of beginning of regexp, or inside of last group. */
1665 unsigned char *begalt;
1666
1667 /* Place in the uncompiled pattern (i.e., the {) to
1668 which to go back if the interval is invalid. */
1669 const char *beg_interval;
1670
1671 /* Address of the place where a forward jump should go to the end of
1672 the containing expression. Each alternative of an `or' -- except the
1673 last -- ends with a forward jump of this sort. */
1674 unsigned char *fixup_alt_jump = 0;
1675
1676 /* Counts open-groups as they are encountered. Remembered for the
1677 matching close-group on the compile stack, so the same register
1678 number is put in the stop_memory as the start_memory. */
1679 regnum_t regnum = 0;
1680
1681 #ifdef DEBUG
1682 DEBUG_PRINT1 ("\nCompiling pattern: ");
1683 if (debug)
1684 {
1685 unsigned debug_count;
1686
1687 for (debug_count = 0; debug_count < size; debug_count++)
1688 putchar (pattern[debug_count]);
1689 putchar ('\n');
1690 }
1691 #endif /* DEBUG */
1692
1693 /* Initialize the compile stack. */
1694 compile_stack.stack = TALLOC (INIT_COMPILE_STACK_SIZE, compile_stack_elt_t);
1695 if (compile_stack.stack == NULL)
1696 return REG_ESPACE;
1697
1698 compile_stack.size = INIT_COMPILE_STACK_SIZE;
1699 compile_stack.avail = 0;
1700
1701 /* Initialize the pattern buffer. */
1702 bufp->syntax = syntax;
1703 bufp->fastmap_accurate = 0;
1704 bufp->not_bol = bufp->not_eol = 0;
1705
1706 /* Set `used' to zero, so that if we return an error, the pattern
1707 printer (for debugging) will think there's no pattern. We reset it
1708 at the end. */
1709 bufp->used = 0;
1710
1711 /* Always count groups, whether or not bufp->no_sub is set. */
1712 bufp->re_nsub = 0;
1713
1714 #if !defined (emacs) && !defined (SYNTAX_TABLE)
1715 /* Initialize the syntax table. */
1716 init_syntax_once ();
1717 #endif
1718
1719 if (bufp->allocated == 0)
1720 {
1721 if (bufp->buffer)
1722 { /* If zero allocated, but buffer is non-null, try to realloc
1723 enough space. This loses if buffer's address is bogus, but
1724 that is the user's responsibility. */
1725 RETALLOC (bufp->buffer, INIT_BUF_SIZE, unsigned char);
1726 }
1727 else
1728 { /* Caller did not allocate a buffer. Do it for them. */
1729 bufp->buffer = TALLOC (INIT_BUF_SIZE, unsigned char);
1730 }
1731 if (!bufp->buffer) FREE_STACK_RETURN (REG_ESPACE);
1732
1733 bufp->allocated = INIT_BUF_SIZE;
1734 }
1735
1736 begalt = b = bufp->buffer;
1737
1738 /* Loop through the uncompiled pattern until we're at the end. */
1739 while (p != pend)
1740 {
1741 PATFETCH (c);
1742
1743 switch (c)
1744 {
1745 case '^':
1746 {
1747 if ( /* If at start of pattern, it's an operator. */
1748 p == pattern + 1
1749 /* If context independent, it's an operator. */
1750 || syntax & RE_CONTEXT_INDEP_ANCHORS
1751 /* Otherwise, depends on what's come before. */
1752 || at_begline_loc_p (pattern, p, syntax))
1753 BUF_PUSH (begline);
1754 else
1755 goto normal_char;
1756 }
1757 break;
1758
1759
1760 case '$':
1761 {
1762 if ( /* If at end of pattern, it's an operator. */
1763 p == pend
1764 /* If context independent, it's an operator. */
1765 || syntax & RE_CONTEXT_INDEP_ANCHORS
1766 /* Otherwise, depends on what's next. */
1767 || at_endline_loc_p (p, pend, syntax))
1768 BUF_PUSH (endline);
1769 else
1770 goto normal_char;
1771 }
1772 break;
1773
1774
1775 case '+':
1776 case '?':
1777 if ((syntax & RE_BK_PLUS_QM)
1778 || (syntax & RE_LIMITED_OPS))
1779 goto normal_char;
1780 handle_plus:
1781 case '*':
1782 /* If there is no previous pattern... */
1783 if (!laststart)
1784 {
1785 if (syntax & RE_CONTEXT_INVALID_OPS)
1786 FREE_STACK_RETURN (REG_BADRPT);
1787 else if (!(syntax & RE_CONTEXT_INDEP_OPS))
1788 goto normal_char;
1789 }
1790
1791 {
1792 /* Are we optimizing this jump? */
1793 boolean keep_string_p = false;
1794
1795 /* 1 means zero (many) matches is allowed. */
1796 char zero_times_ok = 0, many_times_ok = 0;
1797
1798 /* If there is a sequence of repetition chars, collapse it
1799 down to just one (the right one). We can't combine
1800 interval operators with these because of, e.g., `a{2}*',
1801 which should only match an even number of `a's. */
1802
1803 for (;;)
1804 {
1805 zero_times_ok |= c != '+';
1806 many_times_ok |= c != '?';
1807
1808 if (p == pend)
1809 break;
1810
1811 PATFETCH (c);
1812
1813 if (c == '*'
1814 || (!(syntax & RE_BK_PLUS_QM) && (c == '+' || c == '?')))
1815 ;
1816
1817 else if (syntax & RE_BK_PLUS_QM && c == '\\')
1818 {
1819 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1820
1821 PATFETCH (c1);
1822 if (!(c1 == '+' || c1 == '?'))
1823 {
1824 PATUNFETCH;
1825 PATUNFETCH;
1826 break;
1827 }
1828
1829 c = c1;
1830 }
1831 else
1832 {
1833 PATUNFETCH;
1834 break;
1835 }
1836
1837 /* If we get here, we found another repeat character. */
1838 }
1839
1840 /* Star, etc. applied to an empty pattern is equivalent
1841 to an empty pattern. */
1842 if (!laststart)
1843 break;
1844
1845 /* Now we know whether or not zero matches is allowed
1846 and also whether or not two or more matches is allowed. */
1847 if (many_times_ok)
1848 { /* More than one repetition is allowed, so put in at the
1849 end a backward relative jump from `b' to before the next
1850 jump we're going to put in below (which jumps from
1851 laststart to after this jump).
1852
1853 But if we are at the `*' in the exact sequence `.*\n',
1854 insert an unconditional jump backwards to the .,
1855 instead of the beginning of the loop. This way we only
1856 push a failure point once, instead of every time
1857 through the loop. */
1858 assert (p - 1 > pattern);
1859
1860 /* Allocate the space for the jump. */
1861 GET_BUFFER_SPACE (3);
1862
1863 /* We know we are not at the first character of the pattern,
1864 because laststart was nonzero. And we've already
1865 incremented `p', by the way, to be the character after
1866 the `*'. Do we have to do something analogous here
1867 for null bytes, because of RE_DOT_NOT_NULL? */
1868 if (TRANSLATE (*(p - 2)) == TRANSLATE ('.')
1869 && zero_times_ok
1870 && p < pend && TRANSLATE (*p) == TRANSLATE ('\n')
1871 && !(syntax & RE_DOT_NEWLINE))
1872 { /* We have .*\n. */
1873 STORE_JUMP (jump, b, laststart);
1874 keep_string_p = true;
1875 }
1876 else
1877 /* Anything else. */
1878 STORE_JUMP (maybe_pop_jump, b, laststart - 3);
1879
1880 /* We've added more stuff to the buffer. */
1881 b += 3;
1882 }
1883
1884 /* On failure, jump from laststart to b + 3, which will be the
1885 end of the buffer after this jump is inserted. */
1886 GET_BUFFER_SPACE (3);
1887 INSERT_JUMP (keep_string_p ? on_failure_keep_string_jump
1888 : on_failure_jump,
1889 laststart, b + 3);
1890 pending_exact = 0;
1891 b += 3;
1892
1893 if (!zero_times_ok)
1894 {
1895 /* At least one repetition is required, so insert a
1896 `dummy_failure_jump' before the initial
1897 `on_failure_jump' instruction of the loop. This
1898 effects a skip over that instruction the first time
1899 we hit that loop. */
1900 GET_BUFFER_SPACE (3);
1901 INSERT_JUMP (dummy_failure_jump, laststart, laststart + 6);
1902 b += 3;
1903 }
1904 }
1905 break;
1906
1907
1908 case '.':
1909 laststart = b;
1910 BUF_PUSH (anychar);
1911 break;
1912
1913
1914 case '[':
1915 {
1916 boolean had_char_class = false;
1917
1918 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1919
1920 /* Ensure that we have enough space to push a charset: the
1921 opcode, the length count, and the bitset; 34 bytes in all. */
1922 GET_BUFFER_SPACE (34);
1923
1924 laststart = b;
1925
1926 /* We test `*p == '^' twice, instead of using an if
1927 statement, so we only need one BUF_PUSH. */
1928 BUF_PUSH (*p == '^' ? charset_not : charset);
1929 if (*p == '^')
1930 p++;
1931
1932 /* Remember the first position in the bracket expression. */
1933 p1 = p;
1934
1935 /* Push the number of bytes in the bitmap. */
1936 BUF_PUSH ((1 << BYTEWIDTH) / BYTEWIDTH);
1937
1938 /* Clear the whole map. */
1939 bzero (b, (1 << BYTEWIDTH) / BYTEWIDTH);
1940
1941 /* charset_not matches newline according to a syntax bit. */
1942 if ((re_opcode_t) b[-2] == charset_not
1943 && (syntax & RE_HAT_LISTS_NOT_NEWLINE))
1944 SET_LIST_BIT ('\n');
1945
1946 /* Read in characters and ranges, setting map bits. */
1947 for (;;)
1948 {
1949 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
1950
1951 PATFETCH (c);
1952
1953 /* \ might escape characters inside [...] and [^...]. */
1954 if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\')
1955 {
1956 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
1957
1958 PATFETCH (c1);
1959 SET_LIST_BIT (c1);
1960 continue;
1961 }
1962
1963 /* Could be the end of the bracket expression. If it's
1964 not (i.e., when the bracket expression is `[]' so
1965 far), the ']' character bit gets set way below. */
1966 if (c == ']' && p != p1 + 1)
1967 break;
1968
1969 /* Look ahead to see if it's a range when the last thing
1970 was a character class. */
1971 if (had_char_class && c == '-' && *p != ']')
1972 FREE_STACK_RETURN (REG_ERANGE);
1973
1974 /* Look ahead to see if it's a range when the last thing
1975 was a character: if this is a hyphen not at the
1976 beginning or the end of a list, then it's the range
1977 operator. */
1978 if (c == '-'
1979 && !(p - 2 >= pattern && p[-2] == '[')
1980 && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^')
1981 && *p != ']')
1982 {
1983 reg_errcode_t ret
1984 = compile_range (&p, pend, translate, syntax, b);
1985 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1986 }
1987
1988 else if (p[0] == '-' && p[1] != ']')
1989 { /* This handles ranges made up of characters only. */
1990 reg_errcode_t ret;
1991
1992 /* Move past the `-'. */
1993 PATFETCH (c1);
1994
1995 ret = compile_range (&p, pend, translate, syntax, b);
1996 if (ret != REG_NOERROR) FREE_STACK_RETURN (ret);
1997 }
1998
1999 /* See if we're at the beginning of a possible character
2000 class. */
2001
2002 else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':')
2003 { /* Leave room for the null. */
2004 char str[CHAR_CLASS_MAX_LENGTH + 1];
2005
2006 PATFETCH (c);
2007 c1 = 0;
2008
2009 /* If pattern is `[[:'. */
2010 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2011
2012 for (;;)
2013 {
2014 PATFETCH (c);
2015 if (c == ':' || c == ']' || p == pend
2016 || c1 == CHAR_CLASS_MAX_LENGTH)
2017 break;
2018 str[c1++] = c;
2019 }
2020 str[c1] = '\0';
2021
2022 /* If isn't a word bracketed by `[:' and:`]':
2023 undo the ending character, the letters, and leave
2024 the leading `:' and `[' (but set bits for them). */
2025 if (c == ':' && *p == ']')
2026 {
2027 int ch;
2028 boolean is_alnum = STREQ (str, "alnum");
2029 boolean is_alpha = STREQ (str, "alpha");
2030 boolean is_blank = STREQ (str, "blank");
2031 boolean is_cntrl = STREQ (str, "cntrl");
2032 boolean is_digit = STREQ (str, "digit");
2033 boolean is_graph = STREQ (str, "graph");
2034 boolean is_lower = STREQ (str, "lower");
2035 boolean is_print = STREQ (str, "print");
2036 boolean is_punct = STREQ (str, "punct");
2037 boolean is_space = STREQ (str, "space");
2038 boolean is_upper = STREQ (str, "upper");
2039 boolean is_xdigit = STREQ (str, "xdigit");
2040
2041 if (!IS_CHAR_CLASS (str))
2042 FREE_STACK_RETURN (REG_ECTYPE);
2043
2044 /* Throw away the ] at the end of the character
2045 class. */
2046 PATFETCH (c);
2047
2048 if (p == pend) FREE_STACK_RETURN (REG_EBRACK);
2049
2050 for (ch = 0; ch < 1 << BYTEWIDTH; ch++)
2051 {
2052 /* This was split into 3 if's to
2053 avoid an arbitrary limit in some compiler. */
2054 if ( (is_alnum && ISALNUM (ch))
2055 || (is_alpha && ISALPHA (ch))
2056 || (is_blank && ISBLANK (ch))
2057 || (is_cntrl && ISCNTRL (ch)))
2058 SET_LIST_BIT (ch);
2059 if ( (is_digit && ISDIGIT (ch))
2060 || (is_graph && ISGRAPH (ch))
2061 || (is_lower && ISLOWER (ch))
2062 || (is_print && ISPRINT (ch)))
2063 SET_LIST_BIT (ch);
2064 if ( (is_punct && ISPUNCT (ch))
2065 || (is_space && ISSPACE (ch))
2066 || (is_upper && ISUPPER (ch))
2067 || (is_xdigit && ISXDIGIT (ch)))
2068 SET_LIST_BIT (ch);
2069 }
2070 had_char_class = true;
2071 }
2072 else
2073 {
2074 c1++;
2075 while (c1--)
2076 PATUNFETCH;
2077 SET_LIST_BIT ('[');
2078 SET_LIST_BIT (':');
2079 had_char_class = false;
2080 }
2081 }
2082 else
2083 {
2084 had_char_class = false;
2085 SET_LIST_BIT (c);
2086 }
2087 }
2088
2089 /* Discard any (non)matching list bytes that are all 0 at the
2090 end of the map. Decrease the map-length byte too. */
2091 while ((int) b[-1] > 0 && b[b[-1] - 1] == 0)
2092 b[-1]--;
2093 b += b[-1];
2094 }
2095 break;
2096
2097
2098 case '(':
2099 if (syntax & RE_NO_BK_PARENS)
2100 goto handle_open;
2101 else
2102 goto normal_char;
2103
2104
2105 case ')':
2106 if (syntax & RE_NO_BK_PARENS)
2107 goto handle_close;
2108 else
2109 goto normal_char;
2110
2111
2112 case '\n':
2113 if (syntax & RE_NEWLINE_ALT)
2114 goto handle_alt;
2115 else
2116 goto normal_char;
2117
2118
2119 case '|':
2120 if (syntax & RE_NO_BK_VBAR)
2121 goto handle_alt;
2122 else
2123 goto normal_char;
2124
2125
2126 case '{':
2127 if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES)
2128 goto handle_interval;
2129 else
2130 goto normal_char;
2131
2132
2133 case '\\':
2134 if (p == pend) FREE_STACK_RETURN (REG_EESCAPE);
2135
2136 /* Do not translate the character after the \, so that we can
2137 distinguish, e.g., \B from \b, even if we normally would
2138 translate, e.g., B to b. */
2139 PATFETCH_RAW (c);
2140
2141 switch (c)
2142 {
2143 case '(':
2144 if (syntax & RE_NO_BK_PARENS)
2145 goto normal_backslash;
2146
2147 handle_open:
2148 bufp->re_nsub++;
2149 regnum++;
2150
2151 if (COMPILE_STACK_FULL)
2152 {
2153 RETALLOC (compile_stack.stack, compile_stack.size << 1,
2154 compile_stack_elt_t);
2155 if (compile_stack.stack == NULL) return REG_ESPACE;
2156
2157 compile_stack.size <<= 1;
2158 }
2159
2160 /* These are the values to restore when we hit end of this
2161 group. They are all relative offsets, so that if the
2162 whole pattern moves because of realloc, they will still
2163 be valid. */
2164 COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer;
2165 COMPILE_STACK_TOP.fixup_alt_jump
2166 = fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0;
2167 COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer;
2168 COMPILE_STACK_TOP.regnum = regnum;
2169
2170 /* We will eventually replace the 0 with the number of
2171 groups inner to this one. But do not push a
2172 start_memory for groups beyond the last one we can
2173 represent in the compiled pattern. */
2174 if (regnum <= MAX_REGNUM)
2175 {
2176 COMPILE_STACK_TOP.inner_group_offset = b - bufp->buffer + 2;
2177 BUF_PUSH_3 (start_memory, regnum, 0);
2178 }
2179
2180 compile_stack.avail++;
2181
2182 fixup_alt_jump = 0;
2183 laststart = 0;
2184 begalt = b;
2185 /* If we've reached MAX_REGNUM groups, then this open
2186 won't actually generate any code, so we'll have to
2187 clear pending_exact explicitly. */
2188 pending_exact = 0;
2189 break;
2190
2191
2192 case ')':
2193 if (syntax & RE_NO_BK_PARENS) goto normal_backslash;
2194
2195 if (COMPILE_STACK_EMPTY)
2196 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2197 goto normal_backslash;
2198 else
2199 FREE_STACK_RETURN (REG_ERPAREN);
2200
2201 handle_close:
2202 if (fixup_alt_jump)
2203 { /* Push a dummy failure point at the end of the
2204 alternative for a possible future
2205 `pop_failure_jump' to pop. See comments at
2206 `push_dummy_failure' in `re_match_2'. */
2207 BUF_PUSH (push_dummy_failure);
2208
2209 /* We allocated space for this jump when we assigned
2210 to `fixup_alt_jump', in the `handle_alt' case below. */
2211 STORE_JUMP (jump_past_alt, fixup_alt_jump, b - 1);
2212 }
2213
2214 /* See similar code for backslashed left paren above. */
2215 if (COMPILE_STACK_EMPTY)
2216 if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD)
2217 goto normal_char;
2218 else
2219 FREE_STACK_RETURN (REG_ERPAREN);
2220
2221 /* Since we just checked for an empty stack above, this
2222 ``can't happen''. */
2223 assert (compile_stack.avail != 0);
2224 {
2225 /* We don't just want to restore into `regnum', because
2226 later groups should continue to be numbered higher,
2227 as in `(ab)c(de)' -- the second group is #2. */
2228 regnum_t this_group_regnum;
2229
2230 compile_stack.avail--;
2231 begalt = bufp->buffer + COMPILE_STACK_TOP.begalt_offset;
2232 fixup_alt_jump
2233 = COMPILE_STACK_TOP.fixup_alt_jump
2234 ? bufp->buffer + COMPILE_STACK_TOP.fixup_alt_jump - 1
2235 : 0;
2236 laststart = bufp->buffer + COMPILE_STACK_TOP.laststart_offset;
2237 this_group_regnum = COMPILE_STACK_TOP.regnum;
2238 /* If we've reached MAX_REGNUM groups, then this open
2239 won't actually generate any code, so we'll have to
2240 clear pending_exact explicitly. */
2241 pending_exact = 0;
2242
2243 /* We're at the end of the group, so now we know how many
2244 groups were inside this one. */
2245 if (this_group_regnum <= MAX_REGNUM)
2246 {
2247 unsigned char *inner_group_loc
2248 = bufp->buffer + COMPILE_STACK_TOP.inner_group_offset;
2249
2250 *inner_group_loc = regnum - this_group_regnum;
2251 BUF_PUSH_3 (stop_memory, this_group_regnum,
2252 regnum - this_group_regnum);
2253 }
2254 }
2255 break;
2256
2257
2258 case '|': /* `\|'. */
2259 if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR)
2260 goto normal_backslash;
2261 handle_alt:
2262 if (syntax & RE_LIMITED_OPS)
2263 goto normal_char;
2264
2265 /* Insert before the previous alternative a jump which
2266 jumps to this alternative if the former fails. */
2267 GET_BUFFER_SPACE (3);
2268 INSERT_JUMP (on_failure_jump, begalt, b + 6);
2269 pending_exact = 0;
2270 b += 3;
2271
2272 /* The alternative before this one has a jump after it
2273 which gets executed if it gets matched. Adjust that
2274 jump so it will jump to this alternative's analogous
2275 jump (put in below, which in turn will jump to the next
2276 (if any) alternative's such jump, etc.). The last such
2277 jump jumps to the correct final destination. A picture:
2278 _____ _____
2279 | | | |
2280 | v | v
2281 a | b | c
2282
2283 If we are at `b', then fixup_alt_jump right now points to a
2284 three-byte space after `a'. We'll put in the jump, set
2285 fixup_alt_jump to right after `b', and leave behind three
2286 bytes which we'll fill in when we get to after `c'. */
2287
2288 if (fixup_alt_jump)
2289 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2290
2291 /* Mark and leave space for a jump after this alternative,
2292 to be filled in later either by next alternative or
2293 when know we're at the end of a series of alternatives. */
2294 fixup_alt_jump = b;
2295 GET_BUFFER_SPACE (3);
2296 b += 3;
2297
2298 laststart = 0;
2299 begalt = b;
2300 break;
2301
2302
2303 case '{':
2304 /* If \{ is a literal. */
2305 if (!(syntax & RE_INTERVALS)
2306 /* If we're at `\{' and it's not the open-interval
2307 operator. */
2308 || ((syntax & RE_INTERVALS) && (syntax & RE_NO_BK_BRACES))
2309 || (p - 2 == pattern && p == pend))
2310 goto normal_backslash;
2311
2312 handle_interval:
2313 {
2314 /* If got here, then the syntax allows intervals. */
2315
2316 /* At least (most) this many matches must be made. */
2317 int lower_bound = -1, upper_bound = -1;
2318
2319 beg_interval = p - 1;
2320
2321 if (p == pend)
2322 {
2323 if (syntax & RE_NO_BK_BRACES)
2324 goto unfetch_interval;
2325 else
2326 FREE_STACK_RETURN (REG_EBRACE);
2327 }
2328
2329 GET_UNSIGNED_NUMBER (lower_bound);
2330
2331 if (c == ',')
2332 {
2333 GET_UNSIGNED_NUMBER (upper_bound);
2334 if (upper_bound < 0) upper_bound = RE_DUP_MAX;
2335 }
2336 else
2337 /* Interval such as `{1}' => match exactly once. */
2338 upper_bound = lower_bound;
2339
2340 if (lower_bound < 0 || upper_bound > RE_DUP_MAX
2341 || lower_bound > upper_bound)
2342 {
2343 if (syntax & RE_NO_BK_BRACES)
2344 goto unfetch_interval;
2345 else
2346 FREE_STACK_RETURN (REG_BADBR);
2347 }
2348
2349 if (!(syntax & RE_NO_BK_BRACES))
2350 {
2351 if (c != '\\') FREE_STACK_RETURN (REG_EBRACE);
2352
2353 PATFETCH (c);
2354 }
2355
2356 if (c != '}')
2357 {
2358 if (syntax & RE_NO_BK_BRACES)
2359 goto unfetch_interval;
2360 else
2361 FREE_STACK_RETURN (REG_BADBR);
2362 }
2363
2364 /* We just parsed a valid interval. */
2365
2366 /* If it's invalid to have no preceding re. */
2367 if (!laststart)
2368 {
2369 if (syntax & RE_CONTEXT_INVALID_OPS)
2370 FREE_STACK_RETURN (REG_BADRPT);
2371 else if (syntax & RE_CONTEXT_INDEP_OPS)
2372 laststart = b;
2373 else
2374 goto unfetch_interval;
2375 }
2376
2377 /* If the upper bound is zero, don't want to succeed at
2378 all; jump from `laststart' to `b + 3', which will be
2379 the end of the buffer after we insert the jump. */
2380 if (upper_bound == 0)
2381 {
2382 GET_BUFFER_SPACE (3);
2383 INSERT_JUMP (jump, laststart, b + 3);
2384 b += 3;
2385 }
2386
2387 /* Otherwise, we have a nontrivial interval. When
2388 we're all done, the pattern will look like:
2389 set_number_at <jump count> <upper bound>
2390 set_number_at <succeed_n count> <lower bound>
2391 succeed_n <after jump addr> <succeed_n count>
2392 <body of loop>
2393 jump_n <succeed_n addr> <jump count>
2394 (The upper bound and `jump_n' are omitted if
2395 `upper_bound' is 1, though.) */
2396 else
2397 { /* If the upper bound is > 1, we need to insert
2398 more at the end of the loop. */
2399 unsigned nbytes = 10 + (upper_bound > 1) * 10;
2400
2401 GET_BUFFER_SPACE (nbytes);
2402
2403 /* Initialize lower bound of the `succeed_n', even
2404 though it will be set during matching by its
2405 attendant `set_number_at' (inserted next),
2406 because `re_compile_fastmap' needs to know.
2407 Jump to the `jump_n' we might insert below. */
2408 INSERT_JUMP2 (succeed_n, laststart,
2409 b + 5 + (upper_bound > 1) * 5,
2410 lower_bound);
2411 b += 5;
2412
2413 /* Code to initialize the lower bound. Insert
2414 before the `succeed_n'. The `5' is the last two
2415 bytes of this `set_number_at', plus 3 bytes of
2416 the following `succeed_n'. */
2417 insert_op2 (set_number_at, laststart, 5, lower_bound, b);
2418 b += 5;
2419
2420 if (upper_bound > 1)
2421 { /* More than one repetition is allowed, so
2422 append a backward jump to the `succeed_n'
2423 that starts this interval.
2424
2425 When we've reached this during matching,
2426 we'll have matched the interval once, so
2427 jump back only `upper_bound - 1' times. */
2428 STORE_JUMP2 (jump_n, b, laststart + 5,
2429 upper_bound - 1);
2430 b += 5;
2431
2432 /* The location we want to set is the second
2433 parameter of the `jump_n'; that is `b-2' as
2434 an absolute address. `laststart' will be
2435 the `set_number_at' we're about to insert;
2436 `laststart+3' the number to set, the source
2437 for the relative address. But we are
2438 inserting into the middle of the pattern --
2439 so everything is getting moved up by 5.
2440 Conclusion: (b - 2) - (laststart + 3) + 5,
2441 i.e., b - laststart.
2442
2443 We insert this at the beginning of the loop
2444 so that if we fail during matching, we'll
2445 reinitialize the bounds. */
2446 insert_op2 (set_number_at, laststart, b - laststart,
2447 upper_bound - 1, b);
2448 b += 5;
2449 }
2450 }
2451 pending_exact = 0;
2452 beg_interval = NULL;
2453 }
2454 break;
2455
2456 unfetch_interval:
2457 /* If an invalid interval, match the characters as literals. */
2458 assert (beg_interval);
2459 p = beg_interval;
2460 beg_interval = NULL;
2461
2462 /* normal_char and normal_backslash need `c'. */
2463 PATFETCH (c);
2464
2465 if (!(syntax & RE_NO_BK_BRACES))
2466 {
2467 if (p > pattern && p[-1] == '\\')
2468 goto normal_backslash;
2469 }
2470 goto normal_char;
2471
2472 #ifdef emacs
2473 /* There is no way to specify the before_dot and after_dot
2474 operators. rms says this is ok. --karl */
2475 case '=':
2476 BUF_PUSH (at_dot);
2477 break;
2478
2479 case 's':
2480 laststart = b;
2481 PATFETCH (c);
2482 BUF_PUSH_2 (syntaxspec, syntax_spec_code[c]);
2483 break;
2484
2485 case 'S':
2486 laststart = b;
2487 PATFETCH (c);
2488 BUF_PUSH_2 (notsyntaxspec, syntax_spec_code[c]);
2489 break;
2490 #endif /* emacs */
2491
2492
2493 case 'w':
2494 laststart = b;
2495 BUF_PUSH (wordchar);
2496 break;
2497
2498
2499 case 'W':
2500 laststart = b;
2501 BUF_PUSH (notwordchar);
2502 break;
2503
2504
2505 case '<':
2506 BUF_PUSH (wordbeg);
2507 break;
2508
2509 case '>':
2510 BUF_PUSH (wordend);
2511 break;
2512
2513 case 'b':
2514 BUF_PUSH (wordbound);
2515 break;
2516
2517 case 'B':
2518 BUF_PUSH (notwordbound);
2519 break;
2520
2521 case '`':
2522 BUF_PUSH (begbuf);
2523 break;
2524
2525 case '\'':
2526 BUF_PUSH (endbuf);
2527 break;
2528
2529 case '1': case '2': case '3': case '4': case '5':
2530 case '6': case '7': case '8': case '9':
2531 if (syntax & RE_NO_BK_REFS)
2532 goto normal_char;
2533
2534 c1 = c - '0';
2535
2536 if (c1 > regnum)
2537 FREE_STACK_RETURN (REG_ESUBREG);
2538
2539 /* Can't back reference to a subexpression if inside of it. */
2540 if (group_in_compile_stack (compile_stack, c1))
2541 goto normal_char;
2542
2543 laststart = b;
2544 BUF_PUSH_2 (duplicate, c1);
2545 break;
2546
2547
2548 case '+':
2549 case '?':
2550 if (syntax & RE_BK_PLUS_QM)
2551 goto handle_plus;
2552 else
2553 goto normal_backslash;
2554
2555 default:
2556 normal_backslash:
2557 /* You might think it would be useful for \ to mean
2558 not to translate; but if we don't translate it
2559 it will never match anything. */
2560 c = TRANSLATE (c);
2561 goto normal_char;
2562 }
2563 break;
2564
2565
2566 default:
2567 /* Expects the character in `c'. */
2568 normal_char:
2569 /* If no exactn currently being built. */
2570 if (!pending_exact
2571
2572 /* If last exactn not at current position. */
2573 || pending_exact + *pending_exact + 1 != b
2574
2575 /* We have only one byte following the exactn for the count. */
2576 || *pending_exact == (1 << BYTEWIDTH) - 1
2577
2578 /* If followed by a repetition operator. */
2579 || *p == '*' || *p == '^'
2580 || ((syntax & RE_BK_PLUS_QM)
2581 ? *p == '\\' && (p[1] == '+' || p[1] == '?')
2582 : (*p == '+' || *p == '?'))
2583 || ((syntax & RE_INTERVALS)
2584 && ((syntax & RE_NO_BK_BRACES)
2585 ? *p == '{'
2586 : (p[0] == '\\' && p[1] == '{'))))
2587 {
2588 /* Start building a new exactn. */
2589
2590 laststart = b;
2591
2592 BUF_PUSH_2 (exactn, 0);
2593 pending_exact = b - 1;
2594 }
2595
2596 BUF_PUSH (c);
2597 (*pending_exact)++;
2598 break;
2599 } /* switch (c) */
2600 } /* while p != pend */
2601
2602
2603 /* Through the pattern now. */
2604
2605 if (fixup_alt_jump)
2606 STORE_JUMP (jump_past_alt, fixup_alt_jump, b);
2607
2608 if (!COMPILE_STACK_EMPTY)
2609 FREE_STACK_RETURN (REG_EPAREN);
2610
2611 /* If we don't want backtracking, force success
2612 the first time we reach the end of the compiled pattern. */
2613 if (syntax & RE_NO_POSIX_BACKTRACKING)
2614 BUF_PUSH (succeed);
2615
2616 free (compile_stack.stack);
2617
2618 /* We have succeeded; set the length of the buffer. */
2619 bufp->used = b - bufp->buffer;
2620
2621 #ifdef DEBUG
2622 if (debug)
2623 {
2624 DEBUG_PRINT1 ("\nCompiled pattern: \n");
2625 print_compiled_pattern (bufp);
2626 }
2627 #endif /* DEBUG */
2628
2629 #ifndef MATCH_MAY_ALLOCATE
2630 /* Initialize the failure stack to the largest possible stack. This
2631 isn't necessary unless we're trying to avoid calling alloca in
2632 the search and match routines. */
2633 {
2634 int num_regs = bufp->re_nsub + 1;
2635
2636 /* Since DOUBLE_FAIL_STACK refuses to double only if the current size
2637 is strictly greater than re_max_failures, the largest possible stack
2638 is 2 * re_max_failures failure points. */
2639 if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS))
2640 {
2641 fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS);
2642
2643 #ifdef emacs
2644 if (! fail_stack.stack)
2645 fail_stack.stack
2646 = (fail_stack_elt_t *) xmalloc (fail_stack.size
2647 * sizeof (fail_stack_elt_t));
2648 else
2649 fail_stack.stack
2650 = (fail_stack_elt_t *) xrealloc (fail_stack.stack,
2651 (fail_stack.size
2652 * sizeof (fail_stack_elt_t)));
2653 #else /* not emacs */
2654 if (! fail_stack.stack)
2655 fail_stack.stack
2656 = (fail_stack_elt_t *) malloc (fail_stack.size
2657 * sizeof (fail_stack_elt_t));
2658 else
2659 fail_stack.stack
2660 = (fail_stack_elt_t *) realloc (fail_stack.stack,
2661 (fail_stack.size
2662 * sizeof (fail_stack_elt_t)));
2663 #endif /* not emacs */
2664 }
2665
2666 regex_grow_registers (num_regs);
2667 }
2668 #endif /* not MATCH_MAY_ALLOCATE */
2669
2670 return REG_NOERROR;
2671 } /* regex_compile */
2672 \f
2673 /* Subroutines for `regex_compile'. */
2674
2675 /* Store OP at LOC followed by two-byte integer parameter ARG. */
2676
2677 static void
2678 store_op1 (op, loc, arg)
2679 re_opcode_t op;
2680 unsigned char *loc;
2681 int arg;
2682 {
2683 *loc = (unsigned char) op;
2684 STORE_NUMBER (loc + 1, arg);
2685 }
2686
2687
2688 /* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */
2689
2690 static void
2691 store_op2 (op, loc, arg1, arg2)
2692 re_opcode_t op;
2693 unsigned char *loc;
2694 int arg1, arg2;
2695 {
2696 *loc = (unsigned char) op;
2697 STORE_NUMBER (loc + 1, arg1);
2698 STORE_NUMBER (loc + 3, arg2);
2699 }
2700
2701
2702 /* Copy the bytes from LOC to END to open up three bytes of space at LOC
2703 for OP followed by two-byte integer parameter ARG. */
2704
2705 static void
2706 insert_op1 (op, loc, arg, end)
2707 re_opcode_t op;
2708 unsigned char *loc;
2709 int arg;
2710 unsigned char *end;
2711 {
2712 register unsigned char *pfrom = end;
2713 register unsigned char *pto = end + 3;
2714
2715 while (pfrom != loc)
2716 *--pto = *--pfrom;
2717
2718 store_op1 (op, loc, arg);
2719 }
2720
2721
2722 /* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */
2723
2724 static void
2725 insert_op2 (op, loc, arg1, arg2, end)
2726 re_opcode_t op;
2727 unsigned char *loc;
2728 int arg1, arg2;
2729 unsigned char *end;
2730 {
2731 register unsigned char *pfrom = end;
2732 register unsigned char *pto = end + 5;
2733
2734 while (pfrom != loc)
2735 *--pto = *--pfrom;
2736
2737 store_op2 (op, loc, arg1, arg2);
2738 }
2739
2740
2741 /* P points to just after a ^ in PATTERN. Return true if that ^ comes
2742 after an alternative or a begin-subexpression. We assume there is at
2743 least one character before the ^. */
2744
2745 static boolean
2746 at_begline_loc_p (pattern, p, syntax)
2747 const char *pattern, *p;
2748 reg_syntax_t syntax;
2749 {
2750 const char *prev = p - 2;
2751 boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\';
2752
2753 return
2754 /* After a subexpression? */
2755 (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash))
2756 /* After an alternative? */
2757 || (*prev == '|' && (syntax & RE_NO_BK_VBAR || prev_prev_backslash));
2758 }
2759
2760
2761 /* The dual of at_begline_loc_p. This one is for $. We assume there is
2762 at least one character after the $, i.e., `P < PEND'. */
2763
2764 static boolean
2765 at_endline_loc_p (p, pend, syntax)
2766 const char *p, *pend;
2767 int syntax;
2768 {
2769 const char *next = p;
2770 boolean next_backslash = *next == '\\';
2771 const char *next_next = p + 1 < pend ? p + 1 : 0;
2772
2773 return
2774 /* Before a subexpression? */
2775 (syntax & RE_NO_BK_PARENS ? *next == ')'
2776 : next_backslash && next_next && *next_next == ')')
2777 /* Before an alternative? */
2778 || (syntax & RE_NO_BK_VBAR ? *next == '|'
2779 : next_backslash && next_next && *next_next == '|');
2780 }
2781
2782
2783 /* Returns true if REGNUM is in one of COMPILE_STACK's elements and
2784 false if it's not. */
2785
2786 static boolean
2787 group_in_compile_stack (compile_stack, regnum)
2788 compile_stack_type compile_stack;
2789 regnum_t regnum;
2790 {
2791 int this_element;
2792
2793 for (this_element = compile_stack.avail - 1;
2794 this_element >= 0;
2795 this_element--)
2796 if (compile_stack.stack[this_element].regnum == regnum)
2797 return true;
2798
2799 return false;
2800 }
2801
2802
2803 /* Read the ending character of a range (in a bracket expression) from the
2804 uncompiled pattern *P_PTR (which ends at PEND). We assume the
2805 starting character is in `P[-2]'. (`P[-1]' is the character `-'.)
2806 Then we set the translation of all bits between the starting and
2807 ending characters (inclusive) in the compiled pattern B.
2808
2809 Return an error code.
2810
2811 We use these short variable names so we can use the same macros as
2812 `regex_compile' itself. */
2813
2814 static reg_errcode_t
2815 compile_range (p_ptr, pend, translate, syntax, b)
2816 const char **p_ptr, *pend;
2817 char *translate;
2818 reg_syntax_t syntax;
2819 unsigned char *b;
2820 {
2821 unsigned this_char;
2822
2823 const char *p = *p_ptr;
2824 int range_start, range_end;
2825
2826 if (p == pend)
2827 return REG_ERANGE;
2828
2829 /* Even though the pattern is a signed `char *', we need to fetch
2830 with unsigned char *'s; if the high bit of the pattern character
2831 is set, the range endpoints will be negative if we fetch using a
2832 signed char *.
2833
2834 We also want to fetch the endpoints without translating them; the
2835 appropriate translation is done in the bit-setting loop below. */
2836 /* The SVR4 compiler on the 3B2 had trouble with unsigned const char *. */
2837 range_start = ((const unsigned char *) p)[-2];
2838 range_end = ((const unsigned char *) p)[0];
2839
2840 /* Have to increment the pointer into the pattern string, so the
2841 caller isn't still at the ending character. */
2842 (*p_ptr)++;
2843
2844 /* If the start is after the end, the range is empty. */
2845 if (range_start > range_end)
2846 return syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR;
2847
2848 /* Here we see why `this_char' has to be larger than an `unsigned
2849 char' -- the range is inclusive, so if `range_end' == 0xff
2850 (assuming 8-bit characters), we would otherwise go into an infinite
2851 loop, since all characters <= 0xff. */
2852 for (this_char = range_start; this_char <= range_end; this_char++)
2853 {
2854 SET_LIST_BIT (TRANSLATE (this_char));
2855 }
2856
2857 return REG_NOERROR;
2858 }
2859 \f
2860 /* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in
2861 BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible
2862 characters can start a string that matches the pattern. This fastmap
2863 is used by re_search to skip quickly over impossible starting points.
2864
2865 The caller must supply the address of a (1 << BYTEWIDTH)-byte data
2866 area as BUFP->fastmap.
2867
2868 We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in
2869 the pattern buffer.
2870
2871 Returns 0 if we succeed, -2 if an internal error. */
2872
2873 int
2874 re_compile_fastmap (bufp)
2875 struct re_pattern_buffer *bufp;
2876 {
2877 int j, k;
2878 #ifdef MATCH_MAY_ALLOCATE
2879 fail_stack_type fail_stack;
2880 #endif
2881 #ifndef REGEX_MALLOC
2882 char *destination;
2883 #endif
2884 /* We don't push any register information onto the failure stack. */
2885 unsigned num_regs = 0;
2886
2887 register char *fastmap = bufp->fastmap;
2888 unsigned char *pattern = bufp->buffer;
2889 unsigned long size = bufp->used;
2890 unsigned char *p = pattern;
2891 register unsigned char *pend = pattern + size;
2892
2893 #ifdef REL_ALLOC
2894 /* This holds the pointer to the failure stack, when
2895 it is allocated relocatably. */
2896 fail_stack_elt_t *failure_stack_ptr;
2897 #endif
2898
2899 /* Assume that each path through the pattern can be null until
2900 proven otherwise. We set this false at the bottom of switch
2901 statement, to which we get only if a particular path doesn't
2902 match the empty string. */
2903 boolean path_can_be_null = true;
2904
2905 /* We aren't doing a `succeed_n' to begin with. */
2906 boolean succeed_n_p = false;
2907
2908 assert (fastmap != NULL && p != NULL);
2909
2910 INIT_FAIL_STACK ();
2911 bzero (fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */
2912 bufp->fastmap_accurate = 1; /* It will be when we're done. */
2913 bufp->can_be_null = 0;
2914
2915 while (1)
2916 {
2917 if (p == pend || *p == succeed)
2918 {
2919 /* We have reached the (effective) end of pattern. */
2920 if (!FAIL_STACK_EMPTY ())
2921 {
2922 bufp->can_be_null |= path_can_be_null;
2923
2924 /* Reset for next path. */
2925 path_can_be_null = true;
2926
2927 p = fail_stack.stack[--fail_stack.avail].pointer;
2928
2929 continue;
2930 }
2931 else
2932 break;
2933 }
2934
2935 /* We should never be about to go beyond the end of the pattern. */
2936 assert (p < pend);
2937
2938 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
2939 {
2940
2941 /* I guess the idea here is to simply not bother with a fastmap
2942 if a backreference is used, since it's too hard to figure out
2943 the fastmap for the corresponding group. Setting
2944 `can_be_null' stops `re_search_2' from using the fastmap, so
2945 that is all we do. */
2946 case duplicate:
2947 bufp->can_be_null = 1;
2948 goto done;
2949
2950
2951 /* Following are the cases which match a character. These end
2952 with `break'. */
2953
2954 case exactn:
2955 fastmap[p[1]] = 1;
2956 break;
2957
2958
2959 case charset:
2960 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2961 if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))
2962 fastmap[j] = 1;
2963 break;
2964
2965
2966 case charset_not:
2967 /* Chars beyond end of map must be allowed. */
2968 for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++)
2969 fastmap[j] = 1;
2970
2971 for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--)
2972 if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))))
2973 fastmap[j] = 1;
2974 break;
2975
2976
2977 case wordchar:
2978 for (j = 0; j < (1 << BYTEWIDTH); j++)
2979 if (SYNTAX (j) == Sword)
2980 fastmap[j] = 1;
2981 break;
2982
2983
2984 case notwordchar:
2985 for (j = 0; j < (1 << BYTEWIDTH); j++)
2986 if (SYNTAX (j) != Sword)
2987 fastmap[j] = 1;
2988 break;
2989
2990
2991 case anychar:
2992 {
2993 int fastmap_newline = fastmap['\n'];
2994
2995 /* `.' matches anything ... */
2996 for (j = 0; j < (1 << BYTEWIDTH); j++)
2997 fastmap[j] = 1;
2998
2999 /* ... except perhaps newline. */
3000 if (!(bufp->syntax & RE_DOT_NEWLINE))
3001 fastmap['\n'] = fastmap_newline;
3002
3003 /* Return if we have already set `can_be_null'; if we have,
3004 then the fastmap is irrelevant. Something's wrong here. */
3005 else if (bufp->can_be_null)
3006 goto done;
3007
3008 /* Otherwise, have to check alternative paths. */
3009 break;
3010 }
3011
3012 #ifdef emacs
3013 case syntaxspec:
3014 k = *p++;
3015 for (j = 0; j < (1 << BYTEWIDTH); j++)
3016 if (SYNTAX (j) == (enum syntaxcode) k)
3017 fastmap[j] = 1;
3018 break;
3019
3020
3021 case notsyntaxspec:
3022 k = *p++;
3023 for (j = 0; j < (1 << BYTEWIDTH); j++)
3024 if (SYNTAX (j) != (enum syntaxcode) k)
3025 fastmap[j] = 1;
3026 break;
3027
3028
3029 /* All cases after this match the empty string. These end with
3030 `continue'. */
3031
3032
3033 case before_dot:
3034 case at_dot:
3035 case after_dot:
3036 continue;
3037 #endif /* not emacs */
3038
3039
3040 case no_op:
3041 case begline:
3042 case endline:
3043 case begbuf:
3044 case endbuf:
3045 case wordbound:
3046 case notwordbound:
3047 case wordbeg:
3048 case wordend:
3049 case push_dummy_failure:
3050 continue;
3051
3052
3053 case jump_n:
3054 case pop_failure_jump:
3055 case maybe_pop_jump:
3056 case jump:
3057 case jump_past_alt:
3058 case dummy_failure_jump:
3059 EXTRACT_NUMBER_AND_INCR (j, p);
3060 p += j;
3061 if (j > 0)
3062 continue;
3063
3064 /* Jump backward implies we just went through the body of a
3065 loop and matched nothing. Opcode jumped to should be
3066 `on_failure_jump' or `succeed_n'. Just treat it like an
3067 ordinary jump. For a * loop, it has pushed its failure
3068 point already; if so, discard that as redundant. */
3069 if ((re_opcode_t) *p != on_failure_jump
3070 && (re_opcode_t) *p != succeed_n)
3071 continue;
3072
3073 p++;
3074 EXTRACT_NUMBER_AND_INCR (j, p);
3075 p += j;
3076
3077 /* If what's on the stack is where we are now, pop it. */
3078 if (!FAIL_STACK_EMPTY ()
3079 && fail_stack.stack[fail_stack.avail - 1].pointer == p)
3080 fail_stack.avail--;
3081
3082 continue;
3083
3084
3085 case on_failure_jump:
3086 case on_failure_keep_string_jump:
3087 handle_on_failure_jump:
3088 EXTRACT_NUMBER_AND_INCR (j, p);
3089
3090 /* For some patterns, e.g., `(a?)?', `p+j' here points to the
3091 end of the pattern. We don't want to push such a point,
3092 since when we restore it above, entering the switch will
3093 increment `p' past the end of the pattern. We don't need
3094 to push such a point since we obviously won't find any more
3095 fastmap entries beyond `pend'. Such a pattern can match
3096 the null string, though. */
3097 if (p + j < pend)
3098 {
3099 if (!PUSH_PATTERN_OP (p + j, fail_stack))
3100 {
3101 RESET_FAIL_STACK ();
3102 return -2;
3103 }
3104 }
3105 else
3106 bufp->can_be_null = 1;
3107
3108 if (succeed_n_p)
3109 {
3110 EXTRACT_NUMBER_AND_INCR (k, p); /* Skip the n. */
3111 succeed_n_p = false;
3112 }
3113
3114 continue;
3115
3116
3117 case succeed_n:
3118 /* Get to the number of times to succeed. */
3119 p += 2;
3120
3121 /* Increment p past the n for when k != 0. */
3122 EXTRACT_NUMBER_AND_INCR (k, p);
3123 if (k == 0)
3124 {
3125 p -= 4;
3126 succeed_n_p = true; /* Spaghetti code alert. */
3127 goto handle_on_failure_jump;
3128 }
3129 continue;
3130
3131
3132 case set_number_at:
3133 p += 4;
3134 continue;
3135
3136
3137 case start_memory:
3138 case stop_memory:
3139 p += 2;
3140 continue;
3141
3142
3143 default:
3144 abort (); /* We have listed all the cases. */
3145 } /* switch *p++ */
3146
3147 /* Getting here means we have found the possible starting
3148 characters for one path of the pattern -- and that the empty
3149 string does not match. We need not follow this path further.
3150 Instead, look at the next alternative (remembered on the
3151 stack), or quit if no more. The test at the top of the loop
3152 does these things. */
3153 path_can_be_null = false;
3154 p = pend;
3155 } /* while p */
3156
3157 /* Set `can_be_null' for the last path (also the first path, if the
3158 pattern is empty). */
3159 bufp->can_be_null |= path_can_be_null;
3160
3161 done:
3162 RESET_FAIL_STACK ();
3163 return 0;
3164 } /* re_compile_fastmap */
3165 \f
3166 /* Set REGS to hold NUM_REGS registers, storing them in STARTS and
3167 ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use
3168 this memory for recording register information. STARTS and ENDS
3169 must be allocated using the malloc library routine, and must each
3170 be at least NUM_REGS * sizeof (regoff_t) bytes long.
3171
3172 If NUM_REGS == 0, then subsequent matches should allocate their own
3173 register data.
3174
3175 Unless this function is called, the first search or match using
3176 PATTERN_BUFFER will allocate its own register data, without
3177 freeing the old data. */
3178
3179 void
3180 re_set_registers (bufp, regs, num_regs, starts, ends)
3181 struct re_pattern_buffer *bufp;
3182 struct re_registers *regs;
3183 unsigned num_regs;
3184 regoff_t *starts, *ends;
3185 {
3186 if (num_regs)
3187 {
3188 bufp->regs_allocated = REGS_REALLOCATE;
3189 regs->num_regs = num_regs;
3190 regs->start = starts;
3191 regs->end = ends;
3192 }
3193 else
3194 {
3195 bufp->regs_allocated = REGS_UNALLOCATED;
3196 regs->num_regs = 0;
3197 regs->start = regs->end = (regoff_t *) 0;
3198 }
3199 }
3200 \f
3201 /* Searching routines. */
3202
3203 /* Like re_search_2, below, but only one string is specified, and
3204 doesn't let you say where to stop matching. */
3205
3206 int
3207 re_search (bufp, string, size, startpos, range, regs)
3208 struct re_pattern_buffer *bufp;
3209 const char *string;
3210 int size, startpos, range;
3211 struct re_registers *regs;
3212 {
3213 return re_search_2 (bufp, NULL, 0, string, size, startpos, range,
3214 regs, size);
3215 }
3216
3217
3218 /* Using the compiled pattern in BUFP->buffer, first tries to match the
3219 virtual concatenation of STRING1 and STRING2, starting first at index
3220 STARTPOS, then at STARTPOS + 1, and so on.
3221
3222 STRING1 and STRING2 have length SIZE1 and SIZE2, respectively.
3223
3224 RANGE is how far to scan while trying to match. RANGE = 0 means try
3225 only at STARTPOS; in general, the last start tried is STARTPOS +
3226 RANGE.
3227
3228 In REGS, return the indices of the virtual concatenation of STRING1
3229 and STRING2 that matched the entire BUFP->buffer and its contained
3230 subexpressions.
3231
3232 Do not consider matching one past the index STOP in the virtual
3233 concatenation of STRING1 and STRING2.
3234
3235 We return either the position in the strings at which the match was
3236 found, -1 if no match, or -2 if error (such as failure
3237 stack overflow). */
3238
3239 int
3240 re_search_2 (bufp, string1, size1, string2, size2, startpos, range, regs, stop)
3241 struct re_pattern_buffer *bufp;
3242 const char *string1, *string2;
3243 int size1, size2;
3244 int startpos;
3245 int range;
3246 struct re_registers *regs;
3247 int stop;
3248 {
3249 int val;
3250 register char *fastmap = bufp->fastmap;
3251 register char *translate = bufp->translate;
3252 int total_size = size1 + size2;
3253 int endpos = startpos + range;
3254
3255 /* Check for out-of-range STARTPOS. */
3256 if (startpos < 0 || startpos > total_size)
3257 return -1;
3258
3259 /* Fix up RANGE if it might eventually take us outside
3260 the virtual concatenation of STRING1 and STRING2. */
3261 if (endpos < -1)
3262 range = -1 - startpos;
3263 else if (endpos > total_size)
3264 range = total_size - startpos;
3265
3266 /* If the search isn't to be a backwards one, don't waste time in a
3267 search for a pattern that must be anchored. */
3268 if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == begbuf && range > 0)
3269 {
3270 if (startpos > 0)
3271 return -1;
3272 else
3273 range = 1;
3274 }
3275
3276 /* Update the fastmap now if not correct already. */
3277 if (fastmap && !bufp->fastmap_accurate)
3278 if (re_compile_fastmap (bufp) == -2)
3279 return -2;
3280
3281 /* Loop through the string, looking for a place to start matching. */
3282 for (;;)
3283 {
3284 /* If a fastmap is supplied, skip quickly over characters that
3285 cannot be the start of a match. If the pattern can match the
3286 null string, however, we don't need to skip characters; we want
3287 the first null string. */
3288 if (fastmap && startpos < total_size && !bufp->can_be_null)
3289 {
3290 if (range > 0) /* Searching forwards. */
3291 {
3292 register const char *d;
3293 register int lim = 0;
3294 int irange = range;
3295
3296 if (startpos < size1 && startpos + range >= size1)
3297 lim = range - (size1 - startpos);
3298
3299 d = (startpos >= size1 ? string2 - size1 : string1) + startpos;
3300
3301 /* Written out as an if-else to avoid testing `translate'
3302 inside the loop. */
3303 if (translate)
3304 while (range > lim
3305 && !fastmap[(unsigned char)
3306 translate[(unsigned char) *d++]])
3307 range--;
3308 else
3309 while (range > lim && !fastmap[(unsigned char) *d++])
3310 range--;
3311
3312 startpos += irange - range;
3313 }
3314 else /* Searching backwards. */
3315 {
3316 register char c = (size1 == 0 || startpos >= size1
3317 ? string2[startpos - size1]
3318 : string1[startpos]);
3319
3320 if (!fastmap[(unsigned char) TRANSLATE (c)])
3321 goto advance;
3322 }
3323 }
3324
3325 /* If can't match the null string, and that's all we have left, fail. */
3326 if (range >= 0 && startpos == total_size && fastmap
3327 && !bufp->can_be_null)
3328 return -1;
3329
3330 val = re_match_2_internal (bufp, string1, size1, string2, size2,
3331 startpos, regs, stop);
3332 #ifndef REGEX_MALLOC
3333 #ifdef C_ALLOCA
3334 alloca (0);
3335 #endif
3336 #endif
3337
3338 if (val >= 0)
3339 return startpos;
3340
3341 if (val == -2)
3342 return -2;
3343
3344 advance:
3345 if (!range)
3346 break;
3347 else if (range > 0)
3348 {
3349 range--;
3350 startpos++;
3351 }
3352 else
3353 {
3354 range++;
3355 startpos--;
3356 }
3357 }
3358 return -1;
3359 } /* re_search_2 */
3360 \f
3361 /* Declarations and macros for re_match_2. */
3362
3363 static int bcmp_translate ();
3364 static boolean alt_match_null_string_p (),
3365 common_op_match_null_string_p (),
3366 group_match_null_string_p ();
3367
3368 /* This converts PTR, a pointer into one of the search strings `string1'
3369 and `string2' into an offset from the beginning of that string. */
3370 #define POINTER_TO_OFFSET(ptr) \
3371 (FIRST_STRING_P (ptr) \
3372 ? ((regoff_t) ((ptr) - string1)) \
3373 : ((regoff_t) ((ptr) - string2 + size1)))
3374
3375 /* Macros for dealing with the split strings in re_match_2. */
3376
3377 #define MATCHING_IN_FIRST_STRING (dend == end_match_1)
3378
3379 /* Call before fetching a character with *d. This switches over to
3380 string2 if necessary. */
3381 #define PREFETCH() \
3382 while (d == dend) \
3383 { \
3384 /* End of string2 => fail. */ \
3385 if (dend == end_match_2) \
3386 goto fail; \
3387 /* End of string1 => advance to string2. */ \
3388 d = string2; \
3389 dend = end_match_2; \
3390 }
3391
3392
3393 /* Test if at very beginning or at very end of the virtual concatenation
3394 of `string1' and `string2'. If only one string, it's `string2'. */
3395 #define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2)
3396 #define AT_STRINGS_END(d) ((d) == end2)
3397
3398
3399 /* Test if D points to a character which is word-constituent. We have
3400 two special cases to check for: if past the end of string1, look at
3401 the first character in string2; and if before the beginning of
3402 string2, look at the last character in string1. */
3403 #define WORDCHAR_P(d) \
3404 (SYNTAX ((d) == end1 ? *string2 \
3405 : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \
3406 == Sword)
3407
3408 /* Test if the character before D and the one at D differ with respect
3409 to being word-constituent. */
3410 #define AT_WORD_BOUNDARY(d) \
3411 (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \
3412 || WORDCHAR_P (d - 1) != WORDCHAR_P (d))
3413
3414
3415 /* Free everything we malloc. */
3416 #ifdef MATCH_MAY_ALLOCATE
3417 #define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL
3418 #define FREE_VARIABLES() \
3419 do { \
3420 REGEX_FREE_STACK (fail_stack.stack); \
3421 FREE_VAR (regstart); \
3422 FREE_VAR (regend); \
3423 FREE_VAR (old_regstart); \
3424 FREE_VAR (old_regend); \
3425 FREE_VAR (best_regstart); \
3426 FREE_VAR (best_regend); \
3427 FREE_VAR (reg_info); \
3428 FREE_VAR (reg_dummy); \
3429 FREE_VAR (reg_info_dummy); \
3430 } while (0)
3431 #else
3432 #define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */
3433 #endif /* not MATCH_MAY_ALLOCATE */
3434
3435 /* These values must meet several constraints. They must not be valid
3436 register values; since we have a limit of 255 registers (because
3437 we use only one byte in the pattern for the register number), we can
3438 use numbers larger than 255. They must differ by 1, because of
3439 NUM_FAILURE_ITEMS above. And the value for the lowest register must
3440 be larger than the value for the highest register, so we do not try
3441 to actually save any registers when none are active. */
3442 #define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH)
3443 #define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1)
3444 \f
3445 /* Matching routines. */
3446
3447 #ifndef emacs /* Emacs never uses this. */
3448 /* re_match is like re_match_2 except it takes only a single string. */
3449
3450 int
3451 re_match (bufp, string, size, pos, regs)
3452 struct re_pattern_buffer *bufp;
3453 const char *string;
3454 int size, pos;
3455 struct re_registers *regs;
3456 {
3457 int result = re_match_2_internal (bufp, NULL, 0, string, size,
3458 pos, regs, size);
3459 #ifndef REGEX_MALLOC
3460 alloca (0);
3461 #endif
3462 return result;
3463 }
3464 #endif /* not emacs */
3465
3466
3467 /* re_match_2 matches the compiled pattern in BUFP against the
3468 the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1
3469 and SIZE2, respectively). We start matching at POS, and stop
3470 matching at STOP.
3471
3472 If REGS is non-null and the `no_sub' field of BUFP is nonzero, we
3473 store offsets for the substring each group matched in REGS. See the
3474 documentation for exactly how many groups we fill.
3475
3476 We return -1 if no match, -2 if an internal error (such as the
3477 failure stack overflowing). Otherwise, we return the length of the
3478 matched substring. */
3479
3480 int
3481 re_match_2 (bufp, string1, size1, string2, size2, pos, regs, stop)
3482 struct re_pattern_buffer *bufp;
3483 const char *string1, *string2;
3484 int size1, size2;
3485 int pos;
3486 struct re_registers *regs;
3487 int stop;
3488 {
3489 int result = re_match_2_internal (bufp, string1, size1, string2, size2,
3490 pos, regs, stop);
3491 #ifndef REGEX_MALLOC
3492 alloca (0);
3493 #endif
3494 return result;
3495 }
3496
3497 /* This is a separate function so that we can force an alloca cleanup
3498 afterwards. */
3499 static int
3500 re_match_2_internal (bufp, string1, size1, string2, size2, pos, regs, stop)
3501 struct re_pattern_buffer *bufp;
3502 const char *string1, *string2;
3503 int size1, size2;
3504 int pos;
3505 struct re_registers *regs;
3506 int stop;
3507 {
3508 /* General temporaries. */
3509 int mcnt;
3510 unsigned char *p1;
3511
3512 /* Just past the end of the corresponding string. */
3513 const char *end1, *end2;
3514
3515 /* Pointers into string1 and string2, just past the last characters in
3516 each to consider matching. */
3517 const char *end_match_1, *end_match_2;
3518
3519 /* Where we are in the data, and the end of the current string. */
3520 const char *d, *dend;
3521
3522 /* Where we are in the pattern, and the end of the pattern. */
3523 unsigned char *p = bufp->buffer;
3524 register unsigned char *pend = p + bufp->used;
3525
3526 /* Mark the opcode just after a start_memory, so we can test for an
3527 empty subpattern when we get to the stop_memory. */
3528 unsigned char *just_past_start_mem = 0;
3529
3530 /* We use this to map every character in the string. */
3531 char *translate = bufp->translate;
3532
3533 /* Failure point stack. Each place that can handle a failure further
3534 down the line pushes a failure point on this stack. It consists of
3535 restart, regend, and reg_info for all registers corresponding to
3536 the subexpressions we're currently inside, plus the number of such
3537 registers, and, finally, two char *'s. The first char * is where
3538 to resume scanning the pattern; the second one is where to resume
3539 scanning the strings. If the latter is zero, the failure point is
3540 a ``dummy''; if a failure happens and the failure point is a dummy,
3541 it gets discarded and the next next one is tried. */
3542 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3543 fail_stack_type fail_stack;
3544 #endif
3545 #ifdef DEBUG
3546 static unsigned failure_id = 0;
3547 unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0;
3548 #endif
3549
3550 #ifdef REL_ALLOC
3551 /* This holds the pointer to the failure stack, when
3552 it is allocated relocatably. */
3553 fail_stack_elt_t *failure_stack_ptr;
3554 #endif
3555
3556 /* We fill all the registers internally, independent of what we
3557 return, for use in backreferences. The number here includes
3558 an element for register zero. */
3559 unsigned num_regs = bufp->re_nsub + 1;
3560
3561 /* The currently active registers. */
3562 unsigned lowest_active_reg = NO_LOWEST_ACTIVE_REG;
3563 unsigned highest_active_reg = NO_HIGHEST_ACTIVE_REG;
3564
3565 /* Information on the contents of registers. These are pointers into
3566 the input strings; they record just what was matched (on this
3567 attempt) by a subexpression part of the pattern, that is, the
3568 regnum-th regstart pointer points to where in the pattern we began
3569 matching and the regnum-th regend points to right after where we
3570 stopped matching the regnum-th subexpression. (The zeroth register
3571 keeps track of what the whole pattern matches.) */
3572 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3573 const char **regstart, **regend;
3574 #endif
3575
3576 /* If a group that's operated upon by a repetition operator fails to
3577 match anything, then the register for its start will need to be
3578 restored because it will have been set to wherever in the string we
3579 are when we last see its open-group operator. Similarly for a
3580 register's end. */
3581 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3582 const char **old_regstart, **old_regend;
3583 #endif
3584
3585 /* The is_active field of reg_info helps us keep track of which (possibly
3586 nested) subexpressions we are currently in. The matched_something
3587 field of reg_info[reg_num] helps us tell whether or not we have
3588 matched any of the pattern so far this time through the reg_num-th
3589 subexpression. These two fields get reset each time through any
3590 loop their register is in. */
3591 #ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */
3592 register_info_type *reg_info;
3593 #endif
3594
3595 /* The following record the register info as found in the above
3596 variables when we find a match better than any we've seen before.
3597 This happens as we backtrack through the failure points, which in
3598 turn happens only if we have not yet matched the entire string. */
3599 unsigned best_regs_set = false;
3600 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3601 const char **best_regstart, **best_regend;
3602 #endif
3603
3604 /* Logically, this is `best_regend[0]'. But we don't want to have to
3605 allocate space for that if we're not allocating space for anything
3606 else (see below). Also, we never need info about register 0 for
3607 any of the other register vectors, and it seems rather a kludge to
3608 treat `best_regend' differently than the rest. So we keep track of
3609 the end of the best match so far in a separate variable. We
3610 initialize this to NULL so that when we backtrack the first time
3611 and need to test it, it's not garbage. */
3612 const char *match_end = NULL;
3613
3614 /* This helps SET_REGS_MATCHED avoid doing redundant work. */
3615 int set_regs_matched_done = 0;
3616
3617 /* Used when we pop values we don't care about. */
3618 #ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */
3619 const char **reg_dummy;
3620 register_info_type *reg_info_dummy;
3621 #endif
3622
3623 #ifdef DEBUG
3624 /* Counts the total number of registers pushed. */
3625 unsigned num_regs_pushed = 0;
3626 #endif
3627
3628 DEBUG_PRINT1 ("\n\nEntering re_match_2.\n");
3629
3630 INIT_FAIL_STACK ();
3631
3632 #ifdef MATCH_MAY_ALLOCATE
3633 /* Do not bother to initialize all the register variables if there are
3634 no groups in the pattern, as it takes a fair amount of time. If
3635 there are groups, we include space for register 0 (the whole
3636 pattern), even though we never use it, since it simplifies the
3637 array indexing. We should fix this. */
3638 if (bufp->re_nsub)
3639 {
3640 regstart = REGEX_TALLOC (num_regs, const char *);
3641 regend = REGEX_TALLOC (num_regs, const char *);
3642 old_regstart = REGEX_TALLOC (num_regs, const char *);
3643 old_regend = REGEX_TALLOC (num_regs, const char *);
3644 best_regstart = REGEX_TALLOC (num_regs, const char *);
3645 best_regend = REGEX_TALLOC (num_regs, const char *);
3646 reg_info = REGEX_TALLOC (num_regs, register_info_type);
3647 reg_dummy = REGEX_TALLOC (num_regs, const char *);
3648 reg_info_dummy = REGEX_TALLOC (num_regs, register_info_type);
3649
3650 if (!(regstart && regend && old_regstart && old_regend && reg_info
3651 && best_regstart && best_regend && reg_dummy && reg_info_dummy))
3652 {
3653 FREE_VARIABLES ();
3654 return -2;
3655 }
3656 }
3657 else
3658 {
3659 /* We must initialize all our variables to NULL, so that
3660 `FREE_VARIABLES' doesn't try to free them. */
3661 regstart = regend = old_regstart = old_regend = best_regstart
3662 = best_regend = reg_dummy = NULL;
3663 reg_info = reg_info_dummy = (register_info_type *) NULL;
3664 }
3665 #endif /* MATCH_MAY_ALLOCATE */
3666
3667 /* The starting position is bogus. */
3668 if (pos < 0 || pos > size1 + size2)
3669 {
3670 FREE_VARIABLES ();
3671 return -1;
3672 }
3673
3674 /* Initialize subexpression text positions to -1 to mark ones that no
3675 start_memory/stop_memory has been seen for. Also initialize the
3676 register information struct. */
3677 for (mcnt = 1; mcnt < num_regs; mcnt++)
3678 {
3679 regstart[mcnt] = regend[mcnt]
3680 = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE;
3681
3682 REG_MATCH_NULL_STRING_P (reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE;
3683 IS_ACTIVE (reg_info[mcnt]) = 0;
3684 MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3685 EVER_MATCHED_SOMETHING (reg_info[mcnt]) = 0;
3686 }
3687
3688 /* We move `string1' into `string2' if the latter's empty -- but not if
3689 `string1' is null. */
3690 if (size2 == 0 && string1 != NULL)
3691 {
3692 string2 = string1;
3693 size2 = size1;
3694 string1 = 0;
3695 size1 = 0;
3696 }
3697 end1 = string1 + size1;
3698 end2 = string2 + size2;
3699
3700 /* Compute where to stop matching, within the two strings. */
3701 if (stop <= size1)
3702 {
3703 end_match_1 = string1 + stop;
3704 end_match_2 = string2;
3705 }
3706 else
3707 {
3708 end_match_1 = end1;
3709 end_match_2 = string2 + stop - size1;
3710 }
3711
3712 /* `p' scans through the pattern as `d' scans through the data.
3713 `dend' is the end of the input string that `d' points within. `d'
3714 is advanced into the following input string whenever necessary, but
3715 this happens before fetching; therefore, at the beginning of the
3716 loop, `d' can be pointing at the end of a string, but it cannot
3717 equal `string2'. */
3718 if (size1 > 0 && pos <= size1)
3719 {
3720 d = string1 + pos;
3721 dend = end_match_1;
3722 }
3723 else
3724 {
3725 d = string2 + pos - size1;
3726 dend = end_match_2;
3727 }
3728
3729 DEBUG_PRINT1 ("The compiled pattern is: ");
3730 DEBUG_PRINT_COMPILED_PATTERN (bufp, p, pend);
3731 DEBUG_PRINT1 ("The string to match is: `");
3732 DEBUG_PRINT_DOUBLE_STRING (d, string1, size1, string2, size2);
3733 DEBUG_PRINT1 ("'\n");
3734
3735 /* This loops over pattern commands. It exits by returning from the
3736 function if the match is complete, or it drops through if the match
3737 fails at this starting point in the input data. */
3738 for (;;)
3739 {
3740 DEBUG_PRINT2 ("\n0x%x: ", p);
3741
3742 if (p == pend)
3743 { /* End of pattern means we might have succeeded. */
3744 DEBUG_PRINT1 ("end of pattern ... ");
3745
3746 /* If we haven't matched the entire string, and we want the
3747 longest match, try backtracking. */
3748 if (d != end_match_2)
3749 {
3750 /* 1 if this match ends in the same string (string1 or string2)
3751 as the best previous match. */
3752 boolean same_str_p = (FIRST_STRING_P (match_end)
3753 == MATCHING_IN_FIRST_STRING);
3754 /* 1 if this match is the best seen so far. */
3755 boolean best_match_p;
3756
3757 /* AIX compiler got confused when this was combined
3758 with the previous declaration. */
3759 if (same_str_p)
3760 best_match_p = d > match_end;
3761 else
3762 best_match_p = !MATCHING_IN_FIRST_STRING;
3763
3764 DEBUG_PRINT1 ("backtracking.\n");
3765
3766 if (!FAIL_STACK_EMPTY ())
3767 { /* More failure points to try. */
3768
3769 /* If exceeds best match so far, save it. */
3770 if (!best_regs_set || best_match_p)
3771 {
3772 best_regs_set = true;
3773 match_end = d;
3774
3775 DEBUG_PRINT1 ("\nSAVING match as best so far.\n");
3776
3777 for (mcnt = 1; mcnt < num_regs; mcnt++)
3778 {
3779 best_regstart[mcnt] = regstart[mcnt];
3780 best_regend[mcnt] = regend[mcnt];
3781 }
3782 }
3783 goto fail;
3784 }
3785
3786 /* If no failure points, don't restore garbage. And if
3787 last match is real best match, don't restore second
3788 best one. */
3789 else if (best_regs_set && !best_match_p)
3790 {
3791 restore_best_regs:
3792 /* Restore best match. It may happen that `dend ==
3793 end_match_1' while the restored d is in string2.
3794 For example, the pattern `x.*y.*z' against the
3795 strings `x-' and `y-z-', if the two strings are
3796 not consecutive in memory. */
3797 DEBUG_PRINT1 ("Restoring best registers.\n");
3798
3799 d = match_end;
3800 dend = ((d >= string1 && d <= end1)
3801 ? end_match_1 : end_match_2);
3802
3803 for (mcnt = 1; mcnt < num_regs; mcnt++)
3804 {
3805 regstart[mcnt] = best_regstart[mcnt];
3806 regend[mcnt] = best_regend[mcnt];
3807 }
3808 }
3809 } /* d != end_match_2 */
3810
3811 succeed_label:
3812 DEBUG_PRINT1 ("Accepting match.\n");
3813
3814 /* If caller wants register contents data back, do it. */
3815 if (regs && !bufp->no_sub)
3816 {
3817 /* Have the register data arrays been allocated? */
3818 if (bufp->regs_allocated == REGS_UNALLOCATED)
3819 { /* No. So allocate them with malloc. We need one
3820 extra element beyond `num_regs' for the `-1' marker
3821 GNU code uses. */
3822 regs->num_regs = MAX (RE_NREGS, num_regs + 1);
3823 regs->start = TALLOC (regs->num_regs, regoff_t);
3824 regs->end = TALLOC (regs->num_regs, regoff_t);
3825 if (regs->start == NULL || regs->end == NULL)
3826 {
3827 FREE_VARIABLES ();
3828 return -2;
3829 }
3830 bufp->regs_allocated = REGS_REALLOCATE;
3831 }
3832 else if (bufp->regs_allocated == REGS_REALLOCATE)
3833 { /* Yes. If we need more elements than were already
3834 allocated, reallocate them. If we need fewer, just
3835 leave it alone. */
3836 if (regs->num_regs < num_regs + 1)
3837 {
3838 regs->num_regs = num_regs + 1;
3839 RETALLOC (regs->start, regs->num_regs, regoff_t);
3840 RETALLOC (regs->end, regs->num_regs, regoff_t);
3841 if (regs->start == NULL || regs->end == NULL)
3842 {
3843 FREE_VARIABLES ();
3844 return -2;
3845 }
3846 }
3847 }
3848 else
3849 {
3850 /* These braces fend off a "empty body in an else-statement"
3851 warning under GCC when assert expands to nothing. */
3852 assert (bufp->regs_allocated == REGS_FIXED);
3853 }
3854
3855 /* Convert the pointer data in `regstart' and `regend' to
3856 indices. Register zero has to be set differently,
3857 since we haven't kept track of any info for it. */
3858 if (regs->num_regs > 0)
3859 {
3860 regs->start[0] = pos;
3861 regs->end[0] = (MATCHING_IN_FIRST_STRING
3862 ? ((regoff_t) (d - string1))
3863 : ((regoff_t) (d - string2 + size1)));
3864 }
3865
3866 /* Go through the first `min (num_regs, regs->num_regs)'
3867 registers, since that is all we initialized. */
3868 for (mcnt = 1; mcnt < MIN (num_regs, regs->num_regs); mcnt++)
3869 {
3870 if (REG_UNSET (regstart[mcnt]) || REG_UNSET (regend[mcnt]))
3871 regs->start[mcnt] = regs->end[mcnt] = -1;
3872 else
3873 {
3874 regs->start[mcnt]
3875 = (regoff_t) POINTER_TO_OFFSET (regstart[mcnt]);
3876 regs->end[mcnt]
3877 = (regoff_t) POINTER_TO_OFFSET (regend[mcnt]);
3878 }
3879 }
3880
3881 /* If the regs structure we return has more elements than
3882 were in the pattern, set the extra elements to -1. If
3883 we (re)allocated the registers, this is the case,
3884 because we always allocate enough to have at least one
3885 -1 at the end. */
3886 for (mcnt = num_regs; mcnt < regs->num_regs; mcnt++)
3887 regs->start[mcnt] = regs->end[mcnt] = -1;
3888 } /* regs && !bufp->no_sub */
3889
3890 DEBUG_PRINT4 ("%u failure points pushed, %u popped (%u remain).\n",
3891 nfailure_points_pushed, nfailure_points_popped,
3892 nfailure_points_pushed - nfailure_points_popped);
3893 DEBUG_PRINT2 ("%u registers pushed.\n", num_regs_pushed);
3894
3895 mcnt = d - pos - (MATCHING_IN_FIRST_STRING
3896 ? string1
3897 : string2 - size1);
3898
3899 DEBUG_PRINT2 ("Returning %d from re_match_2.\n", mcnt);
3900
3901 FREE_VARIABLES ();
3902 return mcnt;
3903 }
3904
3905 /* Otherwise match next pattern command. */
3906 switch (SWITCH_ENUM_CAST ((re_opcode_t) *p++))
3907 {
3908 /* Ignore these. Used to ignore the n of succeed_n's which
3909 currently have n == 0. */
3910 case no_op:
3911 DEBUG_PRINT1 ("EXECUTING no_op.\n");
3912 break;
3913
3914 case succeed:
3915 DEBUG_PRINT1 ("EXECUTING succeed.\n");
3916 goto succeed_label;
3917
3918 /* Match the next n pattern characters exactly. The following
3919 byte in the pattern defines n, and the n bytes after that
3920 are the characters to match. */
3921 case exactn:
3922 mcnt = *p++;
3923 DEBUG_PRINT2 ("EXECUTING exactn %d.\n", mcnt);
3924
3925 /* This is written out as an if-else so we don't waste time
3926 testing `translate' inside the loop. */
3927 if (translate)
3928 {
3929 do
3930 {
3931 PREFETCH ();
3932 if (translate[(unsigned char) *d++] != (char) *p++)
3933 goto fail;
3934 }
3935 while (--mcnt);
3936 }
3937 else
3938 {
3939 do
3940 {
3941 PREFETCH ();
3942 if (*d++ != (char) *p++) goto fail;
3943 }
3944 while (--mcnt);
3945 }
3946 SET_REGS_MATCHED ();
3947 break;
3948
3949
3950 /* Match any character except possibly a newline or a null. */
3951 case anychar:
3952 DEBUG_PRINT1 ("EXECUTING anychar.\n");
3953
3954 PREFETCH ();
3955
3956 if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE (*d) == '\n')
3957 || (bufp->syntax & RE_DOT_NOT_NULL && TRANSLATE (*d) == '\000'))
3958 goto fail;
3959
3960 SET_REGS_MATCHED ();
3961 DEBUG_PRINT2 (" Matched `%d'.\n", *d);
3962 d++;
3963 break;
3964
3965
3966 case charset:
3967 case charset_not:
3968 {
3969 register unsigned char c;
3970 boolean not = (re_opcode_t) *(p - 1) == charset_not;
3971
3972 DEBUG_PRINT2 ("EXECUTING charset%s.\n", not ? "_not" : "");
3973
3974 PREFETCH ();
3975 c = TRANSLATE (*d); /* The character to match. */
3976
3977 /* Cast to `unsigned' instead of `unsigned char' in case the
3978 bit list is a full 32 bytes long. */
3979 if (c < (unsigned) (*p * BYTEWIDTH)
3980 && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
3981 not = !not;
3982
3983 p += 1 + *p;
3984
3985 if (!not) goto fail;
3986
3987 SET_REGS_MATCHED ();
3988 d++;
3989 break;
3990 }
3991
3992
3993 /* The beginning of a group is represented by start_memory.
3994 The arguments are the register number in the next byte, and the
3995 number of groups inner to this one in the next. The text
3996 matched within the group is recorded (in the internal
3997 registers data structure) under the register number. */
3998 case start_memory:
3999 DEBUG_PRINT3 ("EXECUTING start_memory %d (%d):\n", *p, p[1]);
4000
4001 /* Find out if this group can match the empty string. */
4002 p1 = p; /* To send to group_match_null_string_p. */
4003
4004 if (REG_MATCH_NULL_STRING_P (reg_info[*p]) == MATCH_NULL_UNSET_VALUE)
4005 REG_MATCH_NULL_STRING_P (reg_info[*p])
4006 = group_match_null_string_p (&p1, pend, reg_info);
4007
4008 /* Save the position in the string where we were the last time
4009 we were at this open-group operator in case the group is
4010 operated upon by a repetition operator, e.g., with `(a*)*b'
4011 against `ab'; then we want to ignore where we are now in
4012 the string in case this attempt to match fails. */
4013 old_regstart[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4014 ? REG_UNSET (regstart[*p]) ? d : regstart[*p]
4015 : regstart[*p];
4016 DEBUG_PRINT2 (" old_regstart: %d\n",
4017 POINTER_TO_OFFSET (old_regstart[*p]));
4018
4019 regstart[*p] = d;
4020 DEBUG_PRINT2 (" regstart: %d\n", POINTER_TO_OFFSET (regstart[*p]));
4021
4022 IS_ACTIVE (reg_info[*p]) = 1;
4023 MATCHED_SOMETHING (reg_info[*p]) = 0;
4024
4025 /* Clear this whenever we change the register activity status. */
4026 set_regs_matched_done = 0;
4027
4028 /* This is the new highest active register. */
4029 highest_active_reg = *p;
4030
4031 /* If nothing was active before, this is the new lowest active
4032 register. */
4033 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4034 lowest_active_reg = *p;
4035
4036 /* Move past the register number and inner group count. */
4037 p += 2;
4038 just_past_start_mem = p;
4039
4040 break;
4041
4042
4043 /* The stop_memory opcode represents the end of a group. Its
4044 arguments are the same as start_memory's: the register
4045 number, and the number of inner groups. */
4046 case stop_memory:
4047 DEBUG_PRINT3 ("EXECUTING stop_memory %d (%d):\n", *p, p[1]);
4048
4049 /* We need to save the string position the last time we were at
4050 this close-group operator in case the group is operated
4051 upon by a repetition operator, e.g., with `((a*)*(b*)*)*'
4052 against `aba'; then we want to ignore where we are now in
4053 the string in case this attempt to match fails. */
4054 old_regend[*p] = REG_MATCH_NULL_STRING_P (reg_info[*p])
4055 ? REG_UNSET (regend[*p]) ? d : regend[*p]
4056 : regend[*p];
4057 DEBUG_PRINT2 (" old_regend: %d\n",
4058 POINTER_TO_OFFSET (old_regend[*p]));
4059
4060 regend[*p] = d;
4061 DEBUG_PRINT2 (" regend: %d\n", POINTER_TO_OFFSET (regend[*p]));
4062
4063 /* This register isn't active anymore. */
4064 IS_ACTIVE (reg_info[*p]) = 0;
4065
4066 /* Clear this whenever we change the register activity status. */
4067 set_regs_matched_done = 0;
4068
4069 /* If this was the only register active, nothing is active
4070 anymore. */
4071 if (lowest_active_reg == highest_active_reg)
4072 {
4073 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4074 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4075 }
4076 else
4077 { /* We must scan for the new highest active register, since
4078 it isn't necessarily one less than now: consider
4079 (a(b)c(d(e)f)g). When group 3 ends, after the f), the
4080 new highest active register is 1. */
4081 unsigned char r = *p - 1;
4082 while (r > 0 && !IS_ACTIVE (reg_info[r]))
4083 r--;
4084
4085 /* If we end up at register zero, that means that we saved
4086 the registers as the result of an `on_failure_jump', not
4087 a `start_memory', and we jumped to past the innermost
4088 `stop_memory'. For example, in ((.)*) we save
4089 registers 1 and 2 as a result of the *, but when we pop
4090 back to the second ), we are at the stop_memory 1.
4091 Thus, nothing is active. */
4092 if (r == 0)
4093 {
4094 lowest_active_reg = NO_LOWEST_ACTIVE_REG;
4095 highest_active_reg = NO_HIGHEST_ACTIVE_REG;
4096 }
4097 else
4098 highest_active_reg = r;
4099 }
4100
4101 /* If just failed to match something this time around with a
4102 group that's operated on by a repetition operator, try to
4103 force exit from the ``loop'', and restore the register
4104 information for this group that we had before trying this
4105 last match. */
4106 if ((!MATCHED_SOMETHING (reg_info[*p])
4107 || just_past_start_mem == p - 1)
4108 && (p + 2) < pend)
4109 {
4110 boolean is_a_jump_n = false;
4111
4112 p1 = p + 2;
4113 mcnt = 0;
4114 switch ((re_opcode_t) *p1++)
4115 {
4116 case jump_n:
4117 is_a_jump_n = true;
4118 case pop_failure_jump:
4119 case maybe_pop_jump:
4120 case jump:
4121 case dummy_failure_jump:
4122 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4123 if (is_a_jump_n)
4124 p1 += 2;
4125 break;
4126
4127 default:
4128 /* do nothing */ ;
4129 }
4130 p1 += mcnt;
4131
4132 /* If the next operation is a jump backwards in the pattern
4133 to an on_failure_jump right before the start_memory
4134 corresponding to this stop_memory, exit from the loop
4135 by forcing a failure after pushing on the stack the
4136 on_failure_jump's jump in the pattern, and d. */
4137 if (mcnt < 0 && (re_opcode_t) *p1 == on_failure_jump
4138 && (re_opcode_t) p1[3] == start_memory && p1[4] == *p)
4139 {
4140 /* If this group ever matched anything, then restore
4141 what its registers were before trying this last
4142 failed match, e.g., with `(a*)*b' against `ab' for
4143 regstart[1], and, e.g., with `((a*)*(b*)*)*'
4144 against `aba' for regend[3].
4145
4146 Also restore the registers for inner groups for,
4147 e.g., `((a*)(b*))*' against `aba' (register 3 would
4148 otherwise get trashed). */
4149
4150 if (EVER_MATCHED_SOMETHING (reg_info[*p]))
4151 {
4152 unsigned r;
4153
4154 EVER_MATCHED_SOMETHING (reg_info[*p]) = 0;
4155
4156 /* Restore this and inner groups' (if any) registers. */
4157 for (r = *p; r < *p + *(p + 1); r++)
4158 {
4159 regstart[r] = old_regstart[r];
4160
4161 /* xx why this test? */
4162 if (old_regend[r] >= regstart[r])
4163 regend[r] = old_regend[r];
4164 }
4165 }
4166 p1++;
4167 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4168 PUSH_FAILURE_POINT (p1 + mcnt, d, -2);
4169
4170 goto fail;
4171 }
4172 }
4173
4174 /* Move past the register number and the inner group count. */
4175 p += 2;
4176 break;
4177
4178
4179 /* \<digit> has been turned into a `duplicate' command which is
4180 followed by the numeric value of <digit> as the register number. */
4181 case duplicate:
4182 {
4183 register const char *d2, *dend2;
4184 int regno = *p++; /* Get which register to match against. */
4185 DEBUG_PRINT2 ("EXECUTING duplicate %d.\n", regno);
4186
4187 /* Can't back reference a group which we've never matched. */
4188 if (REG_UNSET (regstart[regno]) || REG_UNSET (regend[regno]))
4189 goto fail;
4190
4191 /* Where in input to try to start matching. */
4192 d2 = regstart[regno];
4193
4194 /* Where to stop matching; if both the place to start and
4195 the place to stop matching are in the same string, then
4196 set to the place to stop, otherwise, for now have to use
4197 the end of the first string. */
4198
4199 dend2 = ((FIRST_STRING_P (regstart[regno])
4200 == FIRST_STRING_P (regend[regno]))
4201 ? regend[regno] : end_match_1);
4202 for (;;)
4203 {
4204 /* If necessary, advance to next segment in register
4205 contents. */
4206 while (d2 == dend2)
4207 {
4208 if (dend2 == end_match_2) break;
4209 if (dend2 == regend[regno]) break;
4210
4211 /* End of string1 => advance to string2. */
4212 d2 = string2;
4213 dend2 = regend[regno];
4214 }
4215 /* At end of register contents => success */
4216 if (d2 == dend2) break;
4217
4218 /* If necessary, advance to next segment in data. */
4219 PREFETCH ();
4220
4221 /* How many characters left in this segment to match. */
4222 mcnt = dend - d;
4223
4224 /* Want how many consecutive characters we can match in
4225 one shot, so, if necessary, adjust the count. */
4226 if (mcnt > dend2 - d2)
4227 mcnt = dend2 - d2;
4228
4229 /* Compare that many; failure if mismatch, else move
4230 past them. */
4231 if (translate
4232 ? bcmp_translate ((unsigned char*)d, (unsigned char*)d2, mcnt, translate)
4233 : bcmp (d, d2, mcnt))
4234 goto fail;
4235 d += mcnt, d2 += mcnt;
4236
4237 /* Do this because we've match some characters. */
4238 SET_REGS_MATCHED ();
4239 }
4240 }
4241 break;
4242
4243
4244 /* begline matches the empty string at the beginning of the string
4245 (unless `not_bol' is set in `bufp'), and, if
4246 `newline_anchor' is set, after newlines. */
4247 case begline:
4248 DEBUG_PRINT1 ("EXECUTING begline.\n");
4249
4250 if (AT_STRINGS_BEG (d))
4251 {
4252 if (!bufp->not_bol) break;
4253 }
4254 else if (d[-1] == '\n' && bufp->newline_anchor)
4255 {
4256 break;
4257 }
4258 /* In all other cases, we fail. */
4259 goto fail;
4260
4261
4262 /* endline is the dual of begline. */
4263 case endline:
4264 DEBUG_PRINT1 ("EXECUTING endline.\n");
4265
4266 if (AT_STRINGS_END (d))
4267 {
4268 if (!bufp->not_eol) break;
4269 }
4270
4271 /* We have to ``prefetch'' the next character. */
4272 else if ((d == end1 ? *string2 : *d) == '\n'
4273 && bufp->newline_anchor)
4274 {
4275 break;
4276 }
4277 goto fail;
4278
4279
4280 /* Match at the very beginning of the data. */
4281 case begbuf:
4282 DEBUG_PRINT1 ("EXECUTING begbuf.\n");
4283 if (AT_STRINGS_BEG (d))
4284 break;
4285 goto fail;
4286
4287
4288 /* Match at the very end of the data. */
4289 case endbuf:
4290 DEBUG_PRINT1 ("EXECUTING endbuf.\n");
4291 if (AT_STRINGS_END (d))
4292 break;
4293 goto fail;
4294
4295
4296 /* on_failure_keep_string_jump is used to optimize `.*\n'. It
4297 pushes NULL as the value for the string on the stack. Then
4298 `pop_failure_point' will keep the current value for the
4299 string, instead of restoring it. To see why, consider
4300 matching `foo\nbar' against `.*\n'. The .* matches the foo;
4301 then the . fails against the \n. But the next thing we want
4302 to do is match the \n against the \n; if we restored the
4303 string value, we would be back at the foo.
4304
4305 Because this is used only in specific cases, we don't need to
4306 check all the things that `on_failure_jump' does, to make
4307 sure the right things get saved on the stack. Hence we don't
4308 share its code. The only reason to push anything on the
4309 stack at all is that otherwise we would have to change
4310 `anychar's code to do something besides goto fail in this
4311 case; that seems worse than this. */
4312 case on_failure_keep_string_jump:
4313 DEBUG_PRINT1 ("EXECUTING on_failure_keep_string_jump");
4314
4315 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4316 DEBUG_PRINT3 (" %d (to 0x%x):\n", mcnt, p + mcnt);
4317
4318 PUSH_FAILURE_POINT (p + mcnt, NULL, -2);
4319 break;
4320
4321
4322 /* Uses of on_failure_jump:
4323
4324 Each alternative starts with an on_failure_jump that points
4325 to the beginning of the next alternative. Each alternative
4326 except the last ends with a jump that in effect jumps past
4327 the rest of the alternatives. (They really jump to the
4328 ending jump of the following alternative, because tensioning
4329 these jumps is a hassle.)
4330
4331 Repeats start with an on_failure_jump that points past both
4332 the repetition text and either the following jump or
4333 pop_failure_jump back to this on_failure_jump. */
4334 case on_failure_jump:
4335 on_failure:
4336 DEBUG_PRINT1 ("EXECUTING on_failure_jump");
4337
4338 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4339 DEBUG_PRINT3 (" %d (to 0x%x)", mcnt, p + mcnt);
4340
4341 /* If this on_failure_jump comes right before a group (i.e.,
4342 the original * applied to a group), save the information
4343 for that group and all inner ones, so that if we fail back
4344 to this point, the group's information will be correct.
4345 For example, in \(a*\)*\1, we need the preceding group,
4346 and in \(\(a*\)b*\)\2, we need the inner group. */
4347
4348 /* We can't use `p' to check ahead because we push
4349 a failure point to `p + mcnt' after we do this. */
4350 p1 = p;
4351
4352 /* We need to skip no_op's before we look for the
4353 start_memory in case this on_failure_jump is happening as
4354 the result of a completed succeed_n, as in \(a\)\{1,3\}b\1
4355 against aba. */
4356 while (p1 < pend && (re_opcode_t) *p1 == no_op)
4357 p1++;
4358
4359 if (p1 < pend && (re_opcode_t) *p1 == start_memory)
4360 {
4361 /* We have a new highest active register now. This will
4362 get reset at the start_memory we are about to get to,
4363 but we will have saved all the registers relevant to
4364 this repetition op, as described above. */
4365 highest_active_reg = *(p1 + 1) + *(p1 + 2);
4366 if (lowest_active_reg == NO_LOWEST_ACTIVE_REG)
4367 lowest_active_reg = *(p1 + 1);
4368 }
4369
4370 DEBUG_PRINT1 (":\n");
4371 PUSH_FAILURE_POINT (p + mcnt, d, -2);
4372 break;
4373
4374
4375 /* A smart repeat ends with `maybe_pop_jump'.
4376 We change it to either `pop_failure_jump' or `jump'. */
4377 case maybe_pop_jump:
4378 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4379 DEBUG_PRINT2 ("EXECUTING maybe_pop_jump %d.\n", mcnt);
4380 {
4381 register unsigned char *p2 = p;
4382
4383 /* Compare the beginning of the repeat with what in the
4384 pattern follows its end. If we can establish that there
4385 is nothing that they would both match, i.e., that we
4386 would have to backtrack because of (as in, e.g., `a*a')
4387 then we can change to pop_failure_jump, because we'll
4388 never have to backtrack.
4389
4390 This is not true in the case of alternatives: in
4391 `(a|ab)*' we do need to backtrack to the `ab' alternative
4392 (e.g., if the string was `ab'). But instead of trying to
4393 detect that here, the alternative has put on a dummy
4394 failure point which is what we will end up popping. */
4395
4396 /* Skip over open/close-group commands.
4397 If what follows this loop is a ...+ construct,
4398 look at what begins its body, since we will have to
4399 match at least one of that. */
4400 while (1)
4401 {
4402 if (p2 + 2 < pend
4403 && ((re_opcode_t) *p2 == stop_memory
4404 || (re_opcode_t) *p2 == start_memory))
4405 p2 += 3;
4406 else if (p2 + 6 < pend
4407 && (re_opcode_t) *p2 == dummy_failure_jump)
4408 p2 += 6;
4409 else
4410 break;
4411 }
4412
4413 p1 = p + mcnt;
4414 /* p1[0] ... p1[2] are the `on_failure_jump' corresponding
4415 to the `maybe_finalize_jump' of this case. Examine what
4416 follows. */
4417
4418 /* If we're at the end of the pattern, we can change. */
4419 if (p2 == pend)
4420 {
4421 /* Consider what happens when matching ":\(.*\)"
4422 against ":/". I don't really understand this code
4423 yet. */
4424 p[-3] = (unsigned char) pop_failure_jump;
4425 DEBUG_PRINT1
4426 (" End of pattern: change to `pop_failure_jump'.\n");
4427 }
4428
4429 else if ((re_opcode_t) *p2 == exactn
4430 || (bufp->newline_anchor && (re_opcode_t) *p2 == endline))
4431 {
4432 register unsigned char c
4433 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4434
4435 if ((re_opcode_t) p1[3] == exactn && p1[5] != c)
4436 {
4437 p[-3] = (unsigned char) pop_failure_jump;
4438 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4439 c, p1[5]);
4440 }
4441
4442 else if ((re_opcode_t) p1[3] == charset
4443 || (re_opcode_t) p1[3] == charset_not)
4444 {
4445 int not = (re_opcode_t) p1[3] == charset_not;
4446
4447 if (c < (unsigned char) (p1[4] * BYTEWIDTH)
4448 && p1[5 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH)))
4449 not = !not;
4450
4451 /* `not' is equal to 1 if c would match, which means
4452 that we can't change to pop_failure_jump. */
4453 if (!not)
4454 {
4455 p[-3] = (unsigned char) pop_failure_jump;
4456 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4457 }
4458 }
4459 }
4460 else if ((re_opcode_t) *p2 == charset)
4461 {
4462 #ifdef DEBUG
4463 register unsigned char c
4464 = *p2 == (unsigned char) endline ? '\n' : p2[2];
4465 #endif
4466
4467 if ((re_opcode_t) p1[3] == exactn
4468 && ! ((int) p2[1] * BYTEWIDTH > (int) p1[4]
4469 && (p2[1 + p1[4] / BYTEWIDTH]
4470 & (1 << (p1[4] % BYTEWIDTH)))))
4471 {
4472 p[-3] = (unsigned char) pop_failure_jump;
4473 DEBUG_PRINT3 (" %c != %c => pop_failure_jump.\n",
4474 c, p1[5]);
4475 }
4476
4477 else if ((re_opcode_t) p1[3] == charset_not)
4478 {
4479 int idx;
4480 /* We win if the charset_not inside the loop
4481 lists every character listed in the charset after. */
4482 for (idx = 0; idx < (int) p2[1]; idx++)
4483 if (! (p2[2 + idx] == 0
4484 || (idx < (int) p1[4]
4485 && ((p2[2 + idx] & ~ p1[5 + idx]) == 0))))
4486 break;
4487
4488 if (idx == p2[1])
4489 {
4490 p[-3] = (unsigned char) pop_failure_jump;
4491 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4492 }
4493 }
4494 else if ((re_opcode_t) p1[3] == charset)
4495 {
4496 int idx;
4497 /* We win if the charset inside the loop
4498 has no overlap with the one after the loop. */
4499 for (idx = 0;
4500 idx < (int) p2[1] && idx < (int) p1[4];
4501 idx++)
4502 if ((p2[2 + idx] & p1[5 + idx]) != 0)
4503 break;
4504
4505 if (idx == p2[1] || idx == p1[4])
4506 {
4507 p[-3] = (unsigned char) pop_failure_jump;
4508 DEBUG_PRINT1 (" No match => pop_failure_jump.\n");
4509 }
4510 }
4511 }
4512 }
4513 p -= 2; /* Point at relative address again. */
4514 if ((re_opcode_t) p[-1] != pop_failure_jump)
4515 {
4516 p[-1] = (unsigned char) jump;
4517 DEBUG_PRINT1 (" Match => jump.\n");
4518 goto unconditional_jump;
4519 }
4520 /* Note fall through. */
4521
4522
4523 /* The end of a simple repeat has a pop_failure_jump back to
4524 its matching on_failure_jump, where the latter will push a
4525 failure point. The pop_failure_jump takes off failure
4526 points put on by this pop_failure_jump's matching
4527 on_failure_jump; we got through the pattern to here from the
4528 matching on_failure_jump, so didn't fail. */
4529 case pop_failure_jump:
4530 {
4531 /* We need to pass separate storage for the lowest and
4532 highest registers, even though we don't care about the
4533 actual values. Otherwise, we will restore only one
4534 register from the stack, since lowest will == highest in
4535 `pop_failure_point'. */
4536 unsigned dummy_low_reg, dummy_high_reg;
4537 unsigned char *pdummy;
4538 const char *sdummy;
4539
4540 DEBUG_PRINT1 ("EXECUTING pop_failure_jump.\n");
4541 POP_FAILURE_POINT (sdummy, pdummy,
4542 dummy_low_reg, dummy_high_reg,
4543 reg_dummy, reg_dummy, reg_info_dummy);
4544 }
4545 /* Note fall through. */
4546
4547
4548 /* Unconditionally jump (without popping any failure points). */
4549 case jump:
4550 unconditional_jump:
4551 EXTRACT_NUMBER_AND_INCR (mcnt, p); /* Get the amount to jump. */
4552 DEBUG_PRINT2 ("EXECUTING jump %d ", mcnt);
4553 p += mcnt; /* Do the jump. */
4554 DEBUG_PRINT2 ("(to 0x%x).\n", p);
4555 break;
4556
4557
4558 /* We need this opcode so we can detect where alternatives end
4559 in `group_match_null_string_p' et al. */
4560 case jump_past_alt:
4561 DEBUG_PRINT1 ("EXECUTING jump_past_alt.\n");
4562 goto unconditional_jump;
4563
4564
4565 /* Normally, the on_failure_jump pushes a failure point, which
4566 then gets popped at pop_failure_jump. We will end up at
4567 pop_failure_jump, also, and with a pattern of, say, `a+', we
4568 are skipping over the on_failure_jump, so we have to push
4569 something meaningless for pop_failure_jump to pop. */
4570 case dummy_failure_jump:
4571 DEBUG_PRINT1 ("EXECUTING dummy_failure_jump.\n");
4572 /* It doesn't matter what we push for the string here. What
4573 the code at `fail' tests is the value for the pattern. */
4574 PUSH_FAILURE_POINT (0, 0, -2);
4575 goto unconditional_jump;
4576
4577
4578 /* At the end of an alternative, we need to push a dummy failure
4579 point in case we are followed by a `pop_failure_jump', because
4580 we don't want the failure point for the alternative to be
4581 popped. For example, matching `(a|ab)*' against `aab'
4582 requires that we match the `ab' alternative. */
4583 case push_dummy_failure:
4584 DEBUG_PRINT1 ("EXECUTING push_dummy_failure.\n");
4585 /* See comments just above at `dummy_failure_jump' about the
4586 two zeroes. */
4587 PUSH_FAILURE_POINT (0, 0, -2);
4588 break;
4589
4590 /* Have to succeed matching what follows at least n times.
4591 After that, handle like `on_failure_jump'. */
4592 case succeed_n:
4593 EXTRACT_NUMBER (mcnt, p + 2);
4594 DEBUG_PRINT2 ("EXECUTING succeed_n %d.\n", mcnt);
4595
4596 assert (mcnt >= 0);
4597 /* Originally, this is how many times we HAVE to succeed. */
4598 if (mcnt > 0)
4599 {
4600 mcnt--;
4601 p += 2;
4602 STORE_NUMBER_AND_INCR (p, mcnt);
4603 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p, mcnt);
4604 }
4605 else if (mcnt == 0)
4606 {
4607 DEBUG_PRINT2 (" Setting two bytes from 0x%x to no_op.\n", p+2);
4608 p[2] = (unsigned char) no_op;
4609 p[3] = (unsigned char) no_op;
4610 goto on_failure;
4611 }
4612 break;
4613
4614 case jump_n:
4615 EXTRACT_NUMBER (mcnt, p + 2);
4616 DEBUG_PRINT2 ("EXECUTING jump_n %d.\n", mcnt);
4617
4618 /* Originally, this is how many times we CAN jump. */
4619 if (mcnt)
4620 {
4621 mcnt--;
4622 STORE_NUMBER (p + 2, mcnt);
4623 goto unconditional_jump;
4624 }
4625 /* If don't have to jump any more, skip over the rest of command. */
4626 else
4627 p += 4;
4628 break;
4629
4630 case set_number_at:
4631 {
4632 DEBUG_PRINT1 ("EXECUTING set_number_at.\n");
4633
4634 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4635 p1 = p + mcnt;
4636 EXTRACT_NUMBER_AND_INCR (mcnt, p);
4637 DEBUG_PRINT3 (" Setting 0x%x to %d.\n", p1, mcnt);
4638 STORE_NUMBER (p1, mcnt);
4639 break;
4640 }
4641
4642 case wordbound:
4643 DEBUG_PRINT1 ("EXECUTING wordbound.\n");
4644 if (AT_WORD_BOUNDARY (d))
4645 break;
4646 goto fail;
4647
4648 case notwordbound:
4649 DEBUG_PRINT1 ("EXECUTING notwordbound.\n");
4650 if (AT_WORD_BOUNDARY (d))
4651 goto fail;
4652 break;
4653
4654 case wordbeg:
4655 DEBUG_PRINT1 ("EXECUTING wordbeg.\n");
4656 if (WORDCHAR_P (d) && (AT_STRINGS_BEG (d) || !WORDCHAR_P (d - 1)))
4657 break;
4658 goto fail;
4659
4660 case wordend:
4661 DEBUG_PRINT1 ("EXECUTING wordend.\n");
4662 if (!AT_STRINGS_BEG (d) && WORDCHAR_P (d - 1)
4663 && (!WORDCHAR_P (d) || AT_STRINGS_END (d)))
4664 break;
4665 goto fail;
4666
4667 #ifdef emacs
4668 case before_dot:
4669 DEBUG_PRINT1 ("EXECUTING before_dot.\n");
4670 if (PTR_CHAR_POS ((unsigned char *) d) >= point)
4671 goto fail;
4672 break;
4673
4674 case at_dot:
4675 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4676 if (PTR_CHAR_POS ((unsigned char *) d) != point)
4677 goto fail;
4678 break;
4679
4680 case after_dot:
4681 DEBUG_PRINT1 ("EXECUTING after_dot.\n");
4682 if (PTR_CHAR_POS ((unsigned char *) d) <= point)
4683 goto fail;
4684 break;
4685 #if 0 /* not emacs19 */
4686 case at_dot:
4687 DEBUG_PRINT1 ("EXECUTING at_dot.\n");
4688 if (PTR_CHAR_POS ((unsigned char *) d) + 1 != point)
4689 goto fail;
4690 break;
4691 #endif /* not emacs19 */
4692
4693 case syntaxspec:
4694 DEBUG_PRINT2 ("EXECUTING syntaxspec %d.\n", mcnt);
4695 mcnt = *p++;
4696 goto matchsyntax;
4697
4698 case wordchar:
4699 DEBUG_PRINT1 ("EXECUTING Emacs wordchar.\n");
4700 mcnt = (int) Sword;
4701 matchsyntax:
4702 PREFETCH ();
4703 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4704 d++;
4705 if (SYNTAX (d[-1]) != (enum syntaxcode) mcnt)
4706 goto fail;
4707 SET_REGS_MATCHED ();
4708 break;
4709
4710 case notsyntaxspec:
4711 DEBUG_PRINT2 ("EXECUTING notsyntaxspec %d.\n", mcnt);
4712 mcnt = *p++;
4713 goto matchnotsyntax;
4714
4715 case notwordchar:
4716 DEBUG_PRINT1 ("EXECUTING Emacs notwordchar.\n");
4717 mcnt = (int) Sword;
4718 matchnotsyntax:
4719 PREFETCH ();
4720 /* Can't use *d++ here; SYNTAX may be an unsafe macro. */
4721 d++;
4722 if (SYNTAX (d[-1]) == (enum syntaxcode) mcnt)
4723 goto fail;
4724 SET_REGS_MATCHED ();
4725 break;
4726
4727 #else /* not emacs */
4728 case wordchar:
4729 DEBUG_PRINT1 ("EXECUTING non-Emacs wordchar.\n");
4730 PREFETCH ();
4731 if (!WORDCHAR_P (d))
4732 goto fail;
4733 SET_REGS_MATCHED ();
4734 d++;
4735 break;
4736
4737 case notwordchar:
4738 DEBUG_PRINT1 ("EXECUTING non-Emacs notwordchar.\n");
4739 PREFETCH ();
4740 if (WORDCHAR_P (d))
4741 goto fail;
4742 SET_REGS_MATCHED ();
4743 d++;
4744 break;
4745 #endif /* not emacs */
4746
4747 default:
4748 abort ();
4749 }
4750 continue; /* Successfully executed one pattern command; keep going. */
4751
4752
4753 /* We goto here if a matching operation fails. */
4754 fail:
4755 if (!FAIL_STACK_EMPTY ())
4756 { /* A restart point is known. Restore to that state. */
4757 DEBUG_PRINT1 ("\nFAIL:\n");
4758 POP_FAILURE_POINT (d, p,
4759 lowest_active_reg, highest_active_reg,
4760 regstart, regend, reg_info);
4761
4762 /* If this failure point is a dummy, try the next one. */
4763 if (!p)
4764 goto fail;
4765
4766 /* If we failed to the end of the pattern, don't examine *p. */
4767 assert (p <= pend);
4768 if (p < pend)
4769 {
4770 boolean is_a_jump_n = false;
4771
4772 /* If failed to a backwards jump that's part of a repetition
4773 loop, need to pop this failure point and use the next one. */
4774 switch ((re_opcode_t) *p)
4775 {
4776 case jump_n:
4777 is_a_jump_n = true;
4778 case maybe_pop_jump:
4779 case pop_failure_jump:
4780 case jump:
4781 p1 = p + 1;
4782 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4783 p1 += mcnt;
4784
4785 if ((is_a_jump_n && (re_opcode_t) *p1 == succeed_n)
4786 || (!is_a_jump_n
4787 && (re_opcode_t) *p1 == on_failure_jump))
4788 goto fail;
4789 break;
4790 default:
4791 /* do nothing */ ;
4792 }
4793 }
4794
4795 if (d >= string1 && d <= end1)
4796 dend = end_match_1;
4797 }
4798 else
4799 break; /* Matching at this starting point really fails. */
4800 } /* for (;;) */
4801
4802 if (best_regs_set)
4803 goto restore_best_regs;
4804
4805 FREE_VARIABLES ();
4806
4807 return -1; /* Failure to match. */
4808 } /* re_match_2 */
4809 \f
4810 /* Subroutine definitions for re_match_2. */
4811
4812
4813 /* We are passed P pointing to a register number after a start_memory.
4814
4815 Return true if the pattern up to the corresponding stop_memory can
4816 match the empty string, and false otherwise.
4817
4818 If we find the matching stop_memory, sets P to point to one past its number.
4819 Otherwise, sets P to an undefined byte less than or equal to END.
4820
4821 We don't handle duplicates properly (yet). */
4822
4823 static boolean
4824 group_match_null_string_p (p, end, reg_info)
4825 unsigned char **p, *end;
4826 register_info_type *reg_info;
4827 {
4828 int mcnt;
4829 /* Point to after the args to the start_memory. */
4830 unsigned char *p1 = *p + 2;
4831
4832 while (p1 < end)
4833 {
4834 /* Skip over opcodes that can match nothing, and return true or
4835 false, as appropriate, when we get to one that can't, or to the
4836 matching stop_memory. */
4837
4838 switch ((re_opcode_t) *p1)
4839 {
4840 /* Could be either a loop or a series of alternatives. */
4841 case on_failure_jump:
4842 p1++;
4843 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4844
4845 /* If the next operation is not a jump backwards in the
4846 pattern. */
4847
4848 if (mcnt >= 0)
4849 {
4850 /* Go through the on_failure_jumps of the alternatives,
4851 seeing if any of the alternatives cannot match nothing.
4852 The last alternative starts with only a jump,
4853 whereas the rest start with on_failure_jump and end
4854 with a jump, e.g., here is the pattern for `a|b|c':
4855
4856 /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6
4857 /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3
4858 /exactn/1/c
4859
4860 So, we have to first go through the first (n-1)
4861 alternatives and then deal with the last one separately. */
4862
4863
4864 /* Deal with the first (n-1) alternatives, which start
4865 with an on_failure_jump (see above) that jumps to right
4866 past a jump_past_alt. */
4867
4868 while ((re_opcode_t) p1[mcnt-3] == jump_past_alt)
4869 {
4870 /* `mcnt' holds how many bytes long the alternative
4871 is, including the ending `jump_past_alt' and
4872 its number. */
4873
4874 if (!alt_match_null_string_p (p1, p1 + mcnt - 3,
4875 reg_info))
4876 return false;
4877
4878 /* Move to right after this alternative, including the
4879 jump_past_alt. */
4880 p1 += mcnt;
4881
4882 /* Break if it's the beginning of an n-th alternative
4883 that doesn't begin with an on_failure_jump. */
4884 if ((re_opcode_t) *p1 != on_failure_jump)
4885 break;
4886
4887 /* Still have to check that it's not an n-th
4888 alternative that starts with an on_failure_jump. */
4889 p1++;
4890 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4891 if ((re_opcode_t) p1[mcnt-3] != jump_past_alt)
4892 {
4893 /* Get to the beginning of the n-th alternative. */
4894 p1 -= 3;
4895 break;
4896 }
4897 }
4898
4899 /* Deal with the last alternative: go back and get number
4900 of the `jump_past_alt' just before it. `mcnt' contains
4901 the length of the alternative. */
4902 EXTRACT_NUMBER (mcnt, p1 - 2);
4903
4904 if (!alt_match_null_string_p (p1, p1 + mcnt, reg_info))
4905 return false;
4906
4907 p1 += mcnt; /* Get past the n-th alternative. */
4908 } /* if mcnt > 0 */
4909 break;
4910
4911
4912 case stop_memory:
4913 assert (p1[1] == **p);
4914 *p = p1 + 2;
4915 return true;
4916
4917
4918 default:
4919 if (!common_op_match_null_string_p (&p1, end, reg_info))
4920 return false;
4921 }
4922 } /* while p1 < end */
4923
4924 return false;
4925 } /* group_match_null_string_p */
4926
4927
4928 /* Similar to group_match_null_string_p, but doesn't deal with alternatives:
4929 It expects P to be the first byte of a single alternative and END one
4930 byte past the last. The alternative can contain groups. */
4931
4932 static boolean
4933 alt_match_null_string_p (p, end, reg_info)
4934 unsigned char *p, *end;
4935 register_info_type *reg_info;
4936 {
4937 int mcnt;
4938 unsigned char *p1 = p;
4939
4940 while (p1 < end)
4941 {
4942 /* Skip over opcodes that can match nothing, and break when we get
4943 to one that can't. */
4944
4945 switch ((re_opcode_t) *p1)
4946 {
4947 /* It's a loop. */
4948 case on_failure_jump:
4949 p1++;
4950 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
4951 p1 += mcnt;
4952 break;
4953
4954 default:
4955 if (!common_op_match_null_string_p (&p1, end, reg_info))
4956 return false;
4957 }
4958 } /* while p1 < end */
4959
4960 return true;
4961 } /* alt_match_null_string_p */
4962
4963
4964 /* Deals with the ops common to group_match_null_string_p and
4965 alt_match_null_string_p.
4966
4967 Sets P to one after the op and its arguments, if any. */
4968
4969 static boolean
4970 common_op_match_null_string_p (p, end, reg_info)
4971 unsigned char **p, *end;
4972 register_info_type *reg_info;
4973 {
4974 int mcnt;
4975 boolean ret;
4976 int reg_no;
4977 unsigned char *p1 = *p;
4978
4979 switch ((re_opcode_t) *p1++)
4980 {
4981 case no_op:
4982 case begline:
4983 case endline:
4984 case begbuf:
4985 case endbuf:
4986 case wordbeg:
4987 case wordend:
4988 case wordbound:
4989 case notwordbound:
4990 #ifdef emacs
4991 case before_dot:
4992 case at_dot:
4993 case after_dot:
4994 #endif
4995 break;
4996
4997 case start_memory:
4998 reg_no = *p1;
4999 assert (reg_no > 0 && reg_no <= MAX_REGNUM);
5000 ret = group_match_null_string_p (&p1, end, reg_info);
5001
5002 /* Have to set this here in case we're checking a group which
5003 contains a group and a back reference to it. */
5004
5005 if (REG_MATCH_NULL_STRING_P (reg_info[reg_no]) == MATCH_NULL_UNSET_VALUE)
5006 REG_MATCH_NULL_STRING_P (reg_info[reg_no]) = ret;
5007
5008 if (!ret)
5009 return false;
5010 break;
5011
5012 /* If this is an optimized succeed_n for zero times, make the jump. */
5013 case jump:
5014 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5015 if (mcnt >= 0)
5016 p1 += mcnt;
5017 else
5018 return false;
5019 break;
5020
5021 case succeed_n:
5022 /* Get to the number of times to succeed. */
5023 p1 += 2;
5024 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5025
5026 if (mcnt == 0)
5027 {
5028 p1 -= 4;
5029 EXTRACT_NUMBER_AND_INCR (mcnt, p1);
5030 p1 += mcnt;
5031 }
5032 else
5033 return false;
5034 break;
5035
5036 case duplicate:
5037 if (!REG_MATCH_NULL_STRING_P (reg_info[*p1]))
5038 return false;
5039 break;
5040
5041 case set_number_at:
5042 p1 += 4;
5043
5044 default:
5045 /* All other opcodes mean we cannot match the empty string. */
5046 return false;
5047 }
5048
5049 *p = p1;
5050 return true;
5051 } /* common_op_match_null_string_p */
5052
5053
5054 /* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN
5055 bytes; nonzero otherwise. */
5056
5057 static int
5058 bcmp_translate (s1, s2, len, translate)
5059 unsigned char *s1, *s2;
5060 register int len;
5061 char *translate;
5062 {
5063 register unsigned char *p1 = s1, *p2 = s2;
5064 while (len)
5065 {
5066 if (translate[*p1++] != translate[*p2++]) return 1;
5067 len--;
5068 }
5069 return 0;
5070 }
5071 \f
5072 /* Entry points for GNU code. */
5073
5074 /* re_compile_pattern is the GNU regular expression compiler: it
5075 compiles PATTERN (of length SIZE) and puts the result in BUFP.
5076 Returns 0 if the pattern was valid, otherwise an error string.
5077
5078 Assumes the `allocated' (and perhaps `buffer') and `translate' fields
5079 are set in BUFP on entry.
5080
5081 We call regex_compile to do the actual compilation. */
5082
5083 const char *
5084 re_compile_pattern (pattern, length, bufp)
5085 const char *pattern;
5086 int length;
5087 struct re_pattern_buffer *bufp;
5088 {
5089 reg_errcode_t ret;
5090
5091 /* GNU code is written to assume at least RE_NREGS registers will be set
5092 (and at least one extra will be -1). */
5093 bufp->regs_allocated = REGS_UNALLOCATED;
5094
5095 /* And GNU code determines whether or not to get register information
5096 by passing null for the REGS argument to re_match, etc., not by
5097 setting no_sub. */
5098 bufp->no_sub = 0;
5099
5100 /* Match anchors at newline. */
5101 bufp->newline_anchor = 1;
5102
5103 ret = regex_compile (pattern, length, re_syntax_options, bufp);
5104
5105 if (!ret)
5106 return NULL;
5107 return gettext (re_error_msgid[(int) ret]);
5108 }
5109 \f
5110 /* Entry points compatible with 4.2 BSD regex library. We don't define
5111 them unless specifically requested. */
5112
5113 #ifdef _REGEX_RE_COMP
5114
5115 /* BSD has one and only one pattern buffer. */
5116 static struct re_pattern_buffer re_comp_buf;
5117
5118 char *
5119 re_comp (s)
5120 const char *s;
5121 {
5122 reg_errcode_t ret;
5123
5124 if (!s)
5125 {
5126 if (!re_comp_buf.buffer)
5127 return gettext ("No previous regular expression");
5128 return 0;
5129 }
5130
5131 if (!re_comp_buf.buffer)
5132 {
5133 re_comp_buf.buffer = (unsigned char *) malloc (200);
5134 if (re_comp_buf.buffer == NULL)
5135 return gettext (re_error_msgid[(int) REG_ESPACE]);
5136 re_comp_buf.allocated = 200;
5137
5138 re_comp_buf.fastmap = (char *) malloc (1 << BYTEWIDTH);
5139 if (re_comp_buf.fastmap == NULL)
5140 return gettext (re_error_msgid[(int) REG_ESPACE]);
5141 }
5142
5143 /* Since `re_exec' always passes NULL for the `regs' argument, we
5144 don't need to initialize the pattern buffer fields which affect it. */
5145
5146 /* Match anchors at newlines. */
5147 re_comp_buf.newline_anchor = 1;
5148
5149 ret = regex_compile (s, strlen (s), re_syntax_options, &re_comp_buf);
5150
5151 if (!ret)
5152 return NULL;
5153
5154 /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */
5155 return (char *) gettext (re_error_msgid[(int) ret]);
5156 }
5157
5158
5159 int
5160 re_exec (s)
5161 const char *s;
5162 {
5163 const int len = strlen (s);
5164 return
5165 0 <= re_search (&re_comp_buf, s, len, 0, len, (struct re_registers *) 0);
5166 }
5167 #endif /* _REGEX_RE_COMP */
5168 \f
5169 /* POSIX.2 functions. Don't define these for Emacs. */
5170
5171 #ifndef emacs
5172
5173 /* regcomp takes a regular expression as a string and compiles it.
5174
5175 PREG is a regex_t *. We do not expect any fields to be initialized,
5176 since POSIX says we shouldn't. Thus, we set
5177
5178 `buffer' to the compiled pattern;
5179 `used' to the length of the compiled pattern;
5180 `syntax' to RE_SYNTAX_POSIX_EXTENDED if the
5181 REG_EXTENDED bit in CFLAGS is set; otherwise, to
5182 RE_SYNTAX_POSIX_BASIC;
5183 `newline_anchor' to REG_NEWLINE being set in CFLAGS;
5184 `fastmap' and `fastmap_accurate' to zero;
5185 `re_nsub' to the number of subexpressions in PATTERN.
5186
5187 PATTERN is the address of the pattern string.
5188
5189 CFLAGS is a series of bits which affect compilation.
5190
5191 If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we
5192 use POSIX basic syntax.
5193
5194 If REG_NEWLINE is set, then . and [^...] don't match newline.
5195 Also, regexec will try a match beginning after every newline.
5196
5197 If REG_ICASE is set, then we considers upper- and lowercase
5198 versions of letters to be equivalent when matching.
5199
5200 If REG_NOSUB is set, then when PREG is passed to regexec, that
5201 routine will report only success or failure, and nothing about the
5202 registers.
5203
5204 It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for
5205 the return codes and their meanings.) */
5206
5207 int
5208 regcomp (preg, pattern, cflags)
5209 regex_t *preg;
5210 const char *pattern;
5211 int cflags;
5212 {
5213 reg_errcode_t ret;
5214 unsigned syntax
5215 = (cflags & REG_EXTENDED) ?
5216 RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC;
5217
5218 /* regex_compile will allocate the space for the compiled pattern. */
5219 preg->buffer = 0;
5220 preg->allocated = 0;
5221 preg->used = 0;
5222
5223 /* Don't bother to use a fastmap when searching. This simplifies the
5224 REG_NEWLINE case: if we used a fastmap, we'd have to put all the
5225 characters after newlines into the fastmap. This way, we just try
5226 every character. */
5227 preg->fastmap = 0;
5228
5229 if (cflags & REG_ICASE)
5230 {
5231 unsigned i;
5232
5233 preg->translate = (char *) malloc (CHAR_SET_SIZE);
5234 if (preg->translate == NULL)
5235 return (int) REG_ESPACE;
5236
5237 /* Map uppercase characters to corresponding lowercase ones. */
5238 for (i = 0; i < CHAR_SET_SIZE; i++)
5239 preg->translate[i] = ISUPPER (i) ? tolower (i) : i;
5240 }
5241 else
5242 preg->translate = NULL;
5243
5244 /* If REG_NEWLINE is set, newlines are treated differently. */
5245 if (cflags & REG_NEWLINE)
5246 { /* REG_NEWLINE implies neither . nor [^...] match newline. */
5247 syntax &= ~RE_DOT_NEWLINE;
5248 syntax |= RE_HAT_LISTS_NOT_NEWLINE;
5249 /* It also changes the matching behavior. */
5250 preg->newline_anchor = 1;
5251 }
5252 else
5253 preg->newline_anchor = 0;
5254
5255 preg->no_sub = !!(cflags & REG_NOSUB);
5256
5257 /* POSIX says a null character in the pattern terminates it, so we
5258 can use strlen here in compiling the pattern. */
5259 ret = regex_compile (pattern, strlen (pattern), syntax, preg);
5260
5261 /* POSIX doesn't distinguish between an unmatched open-group and an
5262 unmatched close-group: both are REG_EPAREN. */
5263 if (ret == REG_ERPAREN) ret = REG_EPAREN;
5264
5265 return (int) ret;
5266 }
5267
5268
5269 /* regexec searches for a given pattern, specified by PREG, in the
5270 string STRING.
5271
5272 If NMATCH is zero or REG_NOSUB was set in the cflags argument to
5273 `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at
5274 least NMATCH elements, and we set them to the offsets of the
5275 corresponding matched substrings.
5276
5277 EFLAGS specifies `execution flags' which affect matching: if
5278 REG_NOTBOL is set, then ^ does not match at the beginning of the
5279 string; if REG_NOTEOL is set, then $ does not match at the end.
5280
5281 We return 0 if we find a match and REG_NOMATCH if not. */
5282
5283 int
5284 regexec (preg, string, nmatch, pmatch, eflags)
5285 const regex_t *preg;
5286 const char *string;
5287 size_t nmatch;
5288 regmatch_t pmatch[];
5289 int eflags;
5290 {
5291 int ret;
5292 struct re_registers regs;
5293 regex_t private_preg;
5294 int len = strlen (string);
5295 boolean want_reg_info = !preg->no_sub && nmatch > 0;
5296
5297 private_preg = *preg;
5298
5299 private_preg.not_bol = !!(eflags & REG_NOTBOL);
5300 private_preg.not_eol = !!(eflags & REG_NOTEOL);
5301
5302 /* The user has told us exactly how many registers to return
5303 information about, via `nmatch'. We have to pass that on to the
5304 matching routines. */
5305 private_preg.regs_allocated = REGS_FIXED;
5306
5307 if (want_reg_info)
5308 {
5309 regs.num_regs = nmatch;
5310 regs.start = TALLOC (nmatch, regoff_t);
5311 regs.end = TALLOC (nmatch, regoff_t);
5312 if (regs.start == NULL || regs.end == NULL)
5313 return (int) REG_NOMATCH;
5314 }
5315
5316 /* Perform the searching operation. */
5317 ret = re_search (&private_preg, string, len,
5318 /* start: */ 0, /* range: */ len,
5319 want_reg_info ? &regs : (struct re_registers *) 0);
5320
5321 /* Copy the register information to the POSIX structure. */
5322 if (want_reg_info)
5323 {
5324 if (ret >= 0)
5325 {
5326 unsigned r;
5327
5328 for (r = 0; r < nmatch; r++)
5329 {
5330 pmatch[r].rm_so = regs.start[r];
5331 pmatch[r].rm_eo = regs.end[r];
5332 }
5333 }
5334
5335 /* If we needed the temporary register info, free the space now. */
5336 free (regs.start);
5337 free (regs.end);
5338 }
5339
5340 /* We want zero return to mean success, unlike `re_search'. */
5341 return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH;
5342 }
5343
5344
5345 /* Returns a message corresponding to an error code, ERRCODE, returned
5346 from either regcomp or regexec. We don't use PREG here. */
5347
5348 size_t
5349 regerror (errcode, preg, errbuf, errbuf_size)
5350 int errcode;
5351 const regex_t *preg;
5352 char *errbuf;
5353 size_t errbuf_size;
5354 {
5355 const char *msg;
5356 size_t msg_size;
5357
5358 if (errcode < 0
5359 || errcode >= (sizeof (re_error_msgid) / sizeof (re_error_msgid[0])))
5360 /* Only error codes returned by the rest of the code should be passed
5361 to this routine. If we are given anything else, or if other regex
5362 code generates an invalid error code, then the program has a bug.
5363 Dump core so we can fix it. */
5364 abort ();
5365
5366 msg = gettext (re_error_msgid[errcode]);
5367
5368 msg_size = strlen (msg) + 1; /* Includes the null. */
5369
5370 if (errbuf_size != 0)
5371 {
5372 if (msg_size > errbuf_size)
5373 {
5374 strncpy (errbuf, msg, errbuf_size - 1);
5375 errbuf[errbuf_size - 1] = 0;
5376 }
5377 else
5378 strcpy (errbuf, msg);
5379 }
5380
5381 return msg_size;
5382 }
5383
5384
5385 /* Free dynamically allocated space used by PREG. */
5386
5387 void
5388 regfree (preg)
5389 regex_t *preg;
5390 {
5391 if (preg->buffer != NULL)
5392 free (preg->buffer);
5393 preg->buffer = NULL;
5394
5395 preg->allocated = 0;
5396 preg->used = 0;
5397
5398 if (preg->fastmap != NULL)
5399 free (preg->fastmap);
5400 preg->fastmap = NULL;
5401 preg->fastmap_accurate = 0;
5402
5403 if (preg->translate != NULL)
5404 free (preg->translate);
5405 preg->translate = NULL;
5406 }
5407
5408 #endif /* not emacs */
5409 \f
5410 /*
5411 Local variables:
5412 make-backup-files: t
5413 version-control: t
5414 trim-versions-without-asking: nil
5415 End:
5416 */