Don't access line[-1] for a zero-length "line" from fgets.
[git/git.git] / builtin-grep.c
1 /*
2 * Builtin "git grep"
3 *
4 * Copyright (c) 2006 Junio C Hamano
5 */
6 #include "cache.h"
7 #include "blob.h"
8 #include "tree.h"
9 #include "commit.h"
10 #include "tag.h"
11 #include "tree-walk.h"
12 #include "builtin.h"
13 #include "grep.h"
14
15 /*
16 * git grep pathspecs are somewhat different from diff-tree pathspecs;
17 * pathname wildcards are allowed.
18 */
19 static int pathspec_matches(const char **paths, const char *name)
20 {
21 int namelen, i;
22 if (!paths || !*paths)
23 return 1;
24 namelen = strlen(name);
25 for (i = 0; paths[i]; i++) {
26 const char *match = paths[i];
27 int matchlen = strlen(match);
28 const char *cp, *meta;
29
30 if (!matchlen ||
31 ((matchlen <= namelen) &&
32 !strncmp(name, match, matchlen) &&
33 (match[matchlen-1] == '/' ||
34 name[matchlen] == '\0' || name[matchlen] == '/')))
35 return 1;
36 if (!fnmatch(match, name, 0))
37 return 1;
38 if (name[namelen-1] != '/')
39 continue;
40
41 /* We are being asked if the directory ("name") is worth
42 * descending into.
43 *
44 * Find the longest leading directory name that does
45 * not have metacharacter in the pathspec; the name
46 * we are looking at must overlap with that directory.
47 */
48 for (cp = match, meta = NULL; cp - match < matchlen; cp++) {
49 char ch = *cp;
50 if (ch == '*' || ch == '[' || ch == '?') {
51 meta = cp;
52 break;
53 }
54 }
55 if (!meta)
56 meta = cp; /* fully literal */
57
58 if (namelen <= meta - match) {
59 /* Looking at "Documentation/" and
60 * the pattern says "Documentation/howto/", or
61 * "Documentation/diff*.txt". The name we
62 * have should match prefix.
63 */
64 if (!memcmp(match, name, namelen))
65 return 1;
66 continue;
67 }
68
69 if (meta - match < namelen) {
70 /* Looking at "Documentation/howto/" and
71 * the pattern says "Documentation/h*";
72 * match up to "Do.../h"; this avoids descending
73 * into "Documentation/technical/".
74 */
75 if (!memcmp(match, name, meta - match))
76 return 1;
77 continue;
78 }
79 }
80 return 0;
81 }
82
83 static int grep_sha1(struct grep_opt *opt, const unsigned char *sha1, const char *name, int tree_name_len)
84 {
85 unsigned long size;
86 char *data;
87 enum object_type type;
88 char *to_free = NULL;
89 int hit;
90
91 data = read_sha1_file(sha1, &type, &size);
92 if (!data) {
93 error("'%s': unable to read %s", name, sha1_to_hex(sha1));
94 return 0;
95 }
96 if (opt->relative && opt->prefix_length) {
97 static char name_buf[PATH_MAX];
98 char *cp;
99 int name_len = strlen(name) - opt->prefix_length + 1;
100
101 if (!tree_name_len)
102 name += opt->prefix_length;
103 else {
104 if (ARRAY_SIZE(name_buf) <= name_len)
105 cp = to_free = xmalloc(name_len);
106 else
107 cp = name_buf;
108 memcpy(cp, name, tree_name_len);
109 strcpy(cp + tree_name_len,
110 name + tree_name_len + opt->prefix_length);
111 name = cp;
112 }
113 }
114 hit = grep_buffer(opt, name, data, size);
115 free(data);
116 free(to_free);
117 return hit;
118 }
119
120 static int grep_file(struct grep_opt *opt, const char *filename)
121 {
122 struct stat st;
123 int i;
124 char *data;
125 size_t sz;
126
127 if (lstat(filename, &st) < 0) {
128 err_ret:
129 if (errno != ENOENT)
130 error("'%s': %s", filename, strerror(errno));
131 return 0;
132 }
133 if (!st.st_size)
134 return 0; /* empty file -- no grep hit */
135 if (!S_ISREG(st.st_mode))
136 return 0;
137 sz = xsize_t(st.st_size);
138 i = open(filename, O_RDONLY);
139 if (i < 0)
140 goto err_ret;
141 data = xmalloc(sz + 1);
142 if (st.st_size != read_in_full(i, data, sz)) {
143 error("'%s': short read %s", filename, strerror(errno));
144 close(i);
145 free(data);
146 return 0;
147 }
148 close(i);
149 if (opt->relative && opt->prefix_length)
150 filename += opt->prefix_length;
151 i = grep_buffer(opt, filename, data, sz);
152 free(data);
153 return i;
154 }
155
156 #ifdef __unix__
157 static int exec_grep(int argc, const char **argv)
158 {
159 pid_t pid;
160 int status;
161
162 argv[argc] = NULL;
163 pid = fork();
164 if (pid < 0)
165 return pid;
166 if (!pid) {
167 execvp("grep", (char **) argv);
168 exit(255);
169 }
170 while (waitpid(pid, &status, 0) < 0) {
171 if (errno == EINTR)
172 continue;
173 return -1;
174 }
175 if (WIFEXITED(status)) {
176 if (!WEXITSTATUS(status))
177 return 1;
178 return 0;
179 }
180 return -1;
181 }
182
183 #define MAXARGS 1000
184 #define ARGBUF 4096
185 #define push_arg(a) do { \
186 if (nr < MAXARGS) argv[nr++] = (a); \
187 else die("maximum number of args exceeded"); \
188 } while (0)
189
190 /*
191 * If you send a singleton filename to grep, it does not give
192 * the name of the file. GNU grep has "-H" but we would want
193 * that behaviour in a portable way.
194 *
195 * So we keep two pathnames in argv buffer unsent to grep in
196 * the main loop if we need to do more than one grep.
197 */
198 static int flush_grep(struct grep_opt *opt,
199 int argc, int arg0, const char **argv, int *kept)
200 {
201 int status;
202 int count = argc - arg0;
203 const char *kept_0 = NULL;
204
205 if (count <= 2) {
206 /*
207 * Because we keep at least 2 paths in the call from
208 * the main loop (i.e. kept != NULL), and MAXARGS is
209 * far greater than 2, this usually is a call to
210 * conclude the grep. However, the user could attempt
211 * to overflow the argv buffer by giving too many
212 * options to leave very small number of real
213 * arguments even for the call in the main loop.
214 */
215 if (kept)
216 die("insanely many options to grep");
217
218 /*
219 * If we have two or more paths, we do not have to do
220 * anything special, but we need to push /dev/null to
221 * get "-H" behaviour of GNU grep portably but when we
222 * are not doing "-l" nor "-L" nor "-c".
223 */
224 if (count == 1 &&
225 !opt->name_only &&
226 !opt->unmatch_name_only &&
227 !opt->count) {
228 argv[argc++] = "/dev/null";
229 argv[argc] = NULL;
230 }
231 }
232
233 else if (kept) {
234 /*
235 * Called because we found many paths and haven't finished
236 * iterating over the cache yet. We keep two paths
237 * for the concluding call. argv[argc-2] and argv[argc-1]
238 * has the last two paths, so save the first one away,
239 * replace it with NULL while sending the list to grep,
240 * and recover them after we are done.
241 */
242 *kept = 2;
243 kept_0 = argv[argc-2];
244 argv[argc-2] = NULL;
245 argc -= 2;
246 }
247
248 status = exec_grep(argc, argv);
249
250 if (kept_0) {
251 /*
252 * Then recover them. Now the last arg is beyond the
253 * terminating NULL which is at argc, and the second
254 * from the last is what we saved away in kept_0
255 */
256 argv[arg0++] = kept_0;
257 argv[arg0] = argv[argc+1];
258 }
259 return status;
260 }
261
262 static int external_grep(struct grep_opt *opt, const char **paths, int cached)
263 {
264 int i, nr, argc, hit, len, status;
265 const char *argv[MAXARGS+1];
266 char randarg[ARGBUF];
267 char *argptr = randarg;
268 struct grep_pat *p;
269
270 if (opt->extended || (opt->relative && opt->prefix_length))
271 return -1;
272 len = nr = 0;
273 push_arg("grep");
274 if (opt->fixed)
275 push_arg("-F");
276 if (opt->linenum)
277 push_arg("-n");
278 if (!opt->pathname)
279 push_arg("-h");
280 if (opt->regflags & REG_EXTENDED)
281 push_arg("-E");
282 if (opt->regflags & REG_ICASE)
283 push_arg("-i");
284 if (opt->word_regexp)
285 push_arg("-w");
286 if (opt->name_only)
287 push_arg("-l");
288 if (opt->unmatch_name_only)
289 push_arg("-L");
290 if (opt->count)
291 push_arg("-c");
292 if (opt->post_context || opt->pre_context) {
293 if (opt->post_context != opt->pre_context) {
294 if (opt->pre_context) {
295 push_arg("-B");
296 len += snprintf(argptr, sizeof(randarg)-len,
297 "%u", opt->pre_context) + 1;
298 if (sizeof(randarg) <= len)
299 die("maximum length of args exceeded");
300 push_arg(argptr);
301 argptr += len;
302 }
303 if (opt->post_context) {
304 push_arg("-A");
305 len += snprintf(argptr, sizeof(randarg)-len,
306 "%u", opt->post_context) + 1;
307 if (sizeof(randarg) <= len)
308 die("maximum length of args exceeded");
309 push_arg(argptr);
310 argptr += len;
311 }
312 }
313 else {
314 push_arg("-C");
315 len += snprintf(argptr, sizeof(randarg)-len,
316 "%u", opt->post_context) + 1;
317 if (sizeof(randarg) <= len)
318 die("maximum length of args exceeded");
319 push_arg(argptr);
320 argptr += len;
321 }
322 }
323 for (p = opt->pattern_list; p; p = p->next) {
324 push_arg("-e");
325 push_arg(p->pattern);
326 }
327
328 hit = 0;
329 argc = nr;
330 for (i = 0; i < active_nr; i++) {
331 struct cache_entry *ce = active_cache[i];
332 char *name;
333 int kept;
334 if (!S_ISREG(ntohl(ce->ce_mode)))
335 continue;
336 if (!pathspec_matches(paths, ce->name))
337 continue;
338 name = ce->name;
339 if (name[0] == '-') {
340 int len = ce_namelen(ce);
341 name = xmalloc(len + 3);
342 memcpy(name, "./", 2);
343 memcpy(name + 2, ce->name, len + 1);
344 }
345 argv[argc++] = name;
346 if (MAXARGS <= argc) {
347 status = flush_grep(opt, argc, nr, argv, &kept);
348 if (0 < status)
349 hit = 1;
350 argc = nr + kept;
351 }
352 if (ce_stage(ce)) {
353 do {
354 i++;
355 } while (i < active_nr &&
356 !strcmp(ce->name, active_cache[i]->name));
357 i--; /* compensate for loop control */
358 }
359 }
360 if (argc > nr) {
361 status = flush_grep(opt, argc, nr, argv, NULL);
362 if (0 < status)
363 hit = 1;
364 }
365 return hit;
366 }
367 #endif
368
369 static int grep_cache(struct grep_opt *opt, const char **paths, int cached)
370 {
371 int hit = 0;
372 int nr;
373 read_cache();
374
375 #ifdef __unix__
376 /*
377 * Use the external "grep" command for the case where
378 * we grep through the checked-out files. It tends to
379 * be a lot more optimized
380 */
381 if (!cached) {
382 hit = external_grep(opt, paths, cached);
383 if (hit >= 0)
384 return hit;
385 }
386 #endif
387
388 for (nr = 0; nr < active_nr; nr++) {
389 struct cache_entry *ce = active_cache[nr];
390 if (!S_ISREG(ntohl(ce->ce_mode)))
391 continue;
392 if (!pathspec_matches(paths, ce->name))
393 continue;
394 if (cached) {
395 if (ce_stage(ce))
396 continue;
397 hit |= grep_sha1(opt, ce->sha1, ce->name, 0);
398 }
399 else
400 hit |= grep_file(opt, ce->name);
401 if (ce_stage(ce)) {
402 do {
403 nr++;
404 } while (nr < active_nr &&
405 !strcmp(ce->name, active_cache[nr]->name));
406 nr--; /* compensate for loop control */
407 }
408 }
409 free_grep_patterns(opt);
410 return hit;
411 }
412
413 static int grep_tree(struct grep_opt *opt, const char **paths,
414 struct tree_desc *tree,
415 const char *tree_name, const char *base)
416 {
417 int len;
418 int hit = 0;
419 struct name_entry entry;
420 char *down;
421 int tn_len = strlen(tree_name);
422 char *path_buf = xmalloc(PATH_MAX + tn_len + 100);
423
424 if (tn_len) {
425 tn_len = sprintf(path_buf, "%s:", tree_name);
426 down = path_buf + tn_len;
427 strcat(down, base);
428 }
429 else {
430 down = path_buf;
431 strcpy(down, base);
432 }
433 len = strlen(path_buf);
434
435 while (tree_entry(tree, &entry)) {
436 strcpy(path_buf + len, entry.path);
437
438 if (S_ISDIR(entry.mode))
439 /* Match "abc/" against pathspec to
440 * decide if we want to descend into "abc"
441 * directory.
442 */
443 strcpy(path_buf + len + tree_entry_len(entry.path, entry.sha1), "/");
444
445 if (!pathspec_matches(paths, down))
446 ;
447 else if (S_ISREG(entry.mode))
448 hit |= grep_sha1(opt, entry.sha1, path_buf, tn_len);
449 else if (S_ISDIR(entry.mode)) {
450 enum object_type type;
451 struct tree_desc sub;
452 void *data;
453 unsigned long size;
454
455 data = read_sha1_file(entry.sha1, &type, &size);
456 if (!data)
457 die("unable to read tree (%s)",
458 sha1_to_hex(entry.sha1));
459 init_tree_desc(&sub, data, size);
460 hit |= grep_tree(opt, paths, &sub, tree_name, down);
461 free(data);
462 }
463 }
464 return hit;
465 }
466
467 static int grep_object(struct grep_opt *opt, const char **paths,
468 struct object *obj, const char *name)
469 {
470 if (obj->type == OBJ_BLOB)
471 return grep_sha1(opt, obj->sha1, name, 0);
472 if (obj->type == OBJ_COMMIT || obj->type == OBJ_TREE) {
473 struct tree_desc tree;
474 void *data;
475 unsigned long size;
476 int hit;
477 data = read_object_with_reference(obj->sha1, tree_type,
478 &size, NULL);
479 if (!data)
480 die("unable to read tree (%s)", sha1_to_hex(obj->sha1));
481 init_tree_desc(&tree, data, size);
482 hit = grep_tree(opt, paths, &tree, name, "");
483 free(data);
484 return hit;
485 }
486 die("unable to grep from object of type %s", typename(obj->type));
487 }
488
489 static const char builtin_grep_usage[] =
490 "git-grep <option>* <rev>* [-e] <pattern> [<path>...]";
491
492 static const char emsg_invalid_context_len[] =
493 "%s: invalid context length argument";
494 static const char emsg_missing_context_len[] =
495 "missing context length argument";
496 static const char emsg_missing_argument[] =
497 "option requires an argument -%s";
498
499 int cmd_grep(int argc, const char **argv, const char *prefix)
500 {
501 int hit = 0;
502 int cached = 0;
503 int seen_dashdash = 0;
504 struct grep_opt opt;
505 struct object_array list = { 0, 0, NULL };
506 const char **paths = NULL;
507 int i;
508
509 memset(&opt, 0, sizeof(opt));
510 opt.prefix_length = (prefix && *prefix) ? strlen(prefix) : 0;
511 opt.relative = 1;
512 opt.pathname = 1;
513 opt.pattern_tail = &opt.pattern_list;
514 opt.regflags = REG_NEWLINE;
515
516 /*
517 * If there is no -- then the paths must exist in the working
518 * tree. If there is no explicit pattern specified with -e or
519 * -f, we take the first unrecognized non option to be the
520 * pattern, but then what follows it must be zero or more
521 * valid refs up to the -- (if exists), and then existing
522 * paths. If there is an explicit pattern, then the first
523 * unrecognized non option is the beginning of the refs list
524 * that continues up to the -- (if exists), and then paths.
525 */
526
527 while (1 < argc) {
528 const char *arg = argv[1];
529 argc--; argv++;
530 if (!strcmp("--cached", arg)) {
531 cached = 1;
532 continue;
533 }
534 if (!strcmp("-a", arg) ||
535 !strcmp("--text", arg)) {
536 opt.binary = GREP_BINARY_TEXT;
537 continue;
538 }
539 if (!strcmp("-i", arg) ||
540 !strcmp("--ignore-case", arg)) {
541 opt.regflags |= REG_ICASE;
542 continue;
543 }
544 if (!strcmp("-I", arg)) {
545 opt.binary = GREP_BINARY_NOMATCH;
546 continue;
547 }
548 if (!strcmp("-v", arg) ||
549 !strcmp("--invert-match", arg)) {
550 opt.invert = 1;
551 continue;
552 }
553 if (!strcmp("-E", arg) ||
554 !strcmp("--extended-regexp", arg)) {
555 opt.regflags |= REG_EXTENDED;
556 continue;
557 }
558 if (!strcmp("-F", arg) ||
559 !strcmp("--fixed-strings", arg)) {
560 opt.fixed = 1;
561 continue;
562 }
563 if (!strcmp("-G", arg) ||
564 !strcmp("--basic-regexp", arg)) {
565 opt.regflags &= ~REG_EXTENDED;
566 continue;
567 }
568 if (!strcmp("-n", arg)) {
569 opt.linenum = 1;
570 continue;
571 }
572 if (!strcmp("-h", arg)) {
573 opt.pathname = 0;
574 continue;
575 }
576 if (!strcmp("-H", arg)) {
577 opt.pathname = 1;
578 continue;
579 }
580 if (!strcmp("-l", arg) ||
581 !strcmp("--files-with-matches", arg)) {
582 opt.name_only = 1;
583 continue;
584 }
585 if (!strcmp("-L", arg) ||
586 !strcmp("--files-without-match", arg)) {
587 opt.unmatch_name_only = 1;
588 continue;
589 }
590 if (!strcmp("-c", arg) ||
591 !strcmp("--count", arg)) {
592 opt.count = 1;
593 continue;
594 }
595 if (!strcmp("-w", arg) ||
596 !strcmp("--word-regexp", arg)) {
597 opt.word_regexp = 1;
598 continue;
599 }
600 if (!prefixcmp(arg, "-A") ||
601 !prefixcmp(arg, "-B") ||
602 !prefixcmp(arg, "-C") ||
603 (arg[0] == '-' && '1' <= arg[1] && arg[1] <= '9')) {
604 unsigned num;
605 const char *scan;
606 switch (arg[1]) {
607 case 'A': case 'B': case 'C':
608 if (!arg[2]) {
609 if (argc <= 1)
610 die(emsg_missing_context_len);
611 scan = *++argv;
612 argc--;
613 }
614 else
615 scan = arg + 2;
616 break;
617 default:
618 scan = arg + 1;
619 break;
620 }
621 if (strtoul_ui(scan, 10, &num))
622 die(emsg_invalid_context_len, scan);
623 switch (arg[1]) {
624 case 'A':
625 opt.post_context = num;
626 break;
627 default:
628 case 'C':
629 opt.post_context = num;
630 case 'B':
631 opt.pre_context = num;
632 break;
633 }
634 continue;
635 }
636 if (!strcmp("-f", arg)) {
637 FILE *patterns;
638 int lno = 0;
639 char buf[1024];
640 if (argc <= 1)
641 die(emsg_missing_argument, arg);
642 patterns = fopen(argv[1], "r");
643 if (!patterns)
644 die("'%s': %s", argv[1], strerror(errno));
645 while (fgets(buf, sizeof(buf), patterns)) {
646 int len = strlen(buf);
647 if (len && buf[len-1] == '\n')
648 buf[len-1] = 0;
649 /* ignore empty line like grep does */
650 if (!buf[0])
651 continue;
652 append_grep_pattern(&opt, xstrdup(buf),
653 argv[1], ++lno,
654 GREP_PATTERN);
655 }
656 fclose(patterns);
657 argv++;
658 argc--;
659 continue;
660 }
661 if (!strcmp("--not", arg)) {
662 append_grep_pattern(&opt, arg, "command line", 0,
663 GREP_NOT);
664 continue;
665 }
666 if (!strcmp("--and", arg)) {
667 append_grep_pattern(&opt, arg, "command line", 0,
668 GREP_AND);
669 continue;
670 }
671 if (!strcmp("--or", arg))
672 continue; /* no-op */
673 if (!strcmp("(", arg)) {
674 append_grep_pattern(&opt, arg, "command line", 0,
675 GREP_OPEN_PAREN);
676 continue;
677 }
678 if (!strcmp(")", arg)) {
679 append_grep_pattern(&opt, arg, "command line", 0,
680 GREP_CLOSE_PAREN);
681 continue;
682 }
683 if (!strcmp("--all-match", arg)) {
684 opt.all_match = 1;
685 continue;
686 }
687 if (!strcmp("-e", arg)) {
688 if (1 < argc) {
689 append_grep_pattern(&opt, argv[1],
690 "-e option", 0,
691 GREP_PATTERN);
692 argv++;
693 argc--;
694 continue;
695 }
696 die(emsg_missing_argument, arg);
697 }
698 if (!strcmp("--full-name", arg)) {
699 opt.relative = 0;
700 continue;
701 }
702 if (!strcmp("--", arg)) {
703 /* later processing wants to have this at argv[1] */
704 argv--;
705 argc++;
706 break;
707 }
708 if (*arg == '-')
709 usage(builtin_grep_usage);
710
711 /* First unrecognized non-option token */
712 if (!opt.pattern_list) {
713 append_grep_pattern(&opt, arg, "command line", 0,
714 GREP_PATTERN);
715 break;
716 }
717 else {
718 /* We are looking at the first path or rev;
719 * it is found at argv[1] after leaving the
720 * loop.
721 */
722 argc++; argv--;
723 break;
724 }
725 }
726
727 if (!opt.pattern_list)
728 die("no pattern given.");
729 if ((opt.regflags != REG_NEWLINE) && opt.fixed)
730 die("cannot mix --fixed-strings and regexp");
731 compile_grep_patterns(&opt);
732
733 /* Check revs and then paths */
734 for (i = 1; i < argc; i++) {
735 const char *arg = argv[i];
736 unsigned char sha1[20];
737 /* Is it a rev? */
738 if (!get_sha1(arg, sha1)) {
739 struct object *object = parse_object(sha1);
740 if (!object)
741 die("bad object %s", arg);
742 add_object_array(object, arg, &list);
743 continue;
744 }
745 if (!strcmp(arg, "--")) {
746 i++;
747 seen_dashdash = 1;
748 }
749 break;
750 }
751
752 /* The rest are paths */
753 if (!seen_dashdash) {
754 int j;
755 for (j = i; j < argc; j++)
756 verify_filename(prefix, argv[j]);
757 }
758
759 if (i < argc) {
760 paths = get_pathspec(prefix, argv + i);
761 if (opt.prefix_length && opt.relative) {
762 /* Make sure we do not get outside of paths */
763 for (i = 0; paths[i]; i++)
764 if (strncmp(prefix, paths[i], opt.prefix_length))
765 die("git-grep: cannot generate relative filenames containing '..'");
766 }
767 }
768 else if (prefix) {
769 paths = xcalloc(2, sizeof(const char *));
770 paths[0] = prefix;
771 paths[1] = NULL;
772 }
773
774 if (!list.nr)
775 return !grep_cache(&opt, paths, cached);
776
777 if (cached)
778 die("both --cached and trees are given.");
779
780 for (i = 0; i < list.nr; i++) {
781 struct object *real_obj;
782 real_obj = deref_tag(list.objects[i].item, NULL, 0);
783 if (grep_object(&opt, paths, real_obj, list.objects[i].name))
784 hit = 1;
785 }
786 free_grep_patterns(&opt);
787 return !hit;
788 }