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