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"
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;
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;
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 */
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 }
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 }
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;
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;
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 }
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;
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 }
156 #ifdef __unix__
157 static int exec_grep(int argc, const char **argv)
158 {
159 pid_t pid;
160 int status;
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 }
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)
190 static int external_grep(struct grep_opt *opt, const char **paths, int cached)
191 {
192 int i, nr, argc, hit, len, status;
193 const char *argv[MAXARGS+1];
194 char randarg[ARGBUF];
195 char *argptr = randarg;
196 struct grep_pat *p;
198 if (opt->extended || (opt->relative && opt->prefix_length))
199 return -1;
200 len = nr = 0;
201 push_arg("grep");
202 if (opt->fixed)
203 push_arg("-F");
204 if (opt->linenum)
205 push_arg("-n");
206 if (!opt->pathname)
207 push_arg("-h");
208 if (opt->regflags & REG_EXTENDED)
209 push_arg("-E");
210 if (opt->regflags & REG_ICASE)
211 push_arg("-i");
212 if (opt->word_regexp)
213 push_arg("-w");
214 if (opt->name_only)
215 push_arg("-l");
216 if (opt->unmatch_name_only)
217 push_arg("-L");
218 if (opt->count)
219 push_arg("-c");
220 if (opt->post_context || opt->pre_context) {
221 if (opt->post_context != opt->pre_context) {
222 if (opt->pre_context) {
223 push_arg("-B");
224 len += snprintf(argptr, sizeof(randarg)-len,
225 "%u", opt->pre_context);
226 if (sizeof(randarg) <= len)
227 die("maximum length of args exceeded");
228 push_arg(argptr);
229 argptr += len;
230 }
231 if (opt->post_context) {
232 push_arg("-A");
233 len += snprintf(argptr, sizeof(randarg)-len,
234 "%u", opt->post_context);
235 if (sizeof(randarg) <= len)
236 die("maximum length of args exceeded");
237 push_arg(argptr);
238 argptr += len;
239 }
240 }
241 else {
242 push_arg("-C");
243 len += snprintf(argptr, sizeof(randarg)-len,
244 "%u", opt->post_context);
245 if (sizeof(randarg) <= len)
246 die("maximum length of args exceeded");
247 push_arg(argptr);
248 argptr += len;
249 }
250 }
251 for (p = opt->pattern_list; p; p = p->next) {
252 push_arg("-e");
253 push_arg(p->pattern);
254 }
256 /*
257 * To make sure we get the header printed out when we want it,
258 * add /dev/null to the paths to grep. This is unnecessary
259 * (and wrong) with "-l" or "-L", which always print out the
260 * name anyway.
261 *
262 * GNU grep has "-H", but this is portable.
263 */
264 if (!opt->name_only && !opt->unmatch_name_only)
265 push_arg("/dev/null");
267 hit = 0;
268 argc = nr;
269 for (i = 0; i < active_nr; i++) {
270 struct cache_entry *ce = active_cache[i];
271 char *name;
272 if (!S_ISREG(ntohl(ce->ce_mode)))
273 continue;
274 if (!pathspec_matches(paths, ce->name))
275 continue;
276 name = ce->name;
277 if (name[0] == '-') {
278 int len = ce_namelen(ce);
279 name = xmalloc(len + 3);
280 memcpy(name, "./", 2);
281 memcpy(name + 2, ce->name, len + 1);
282 }
283 argv[argc++] = name;
284 if (argc < MAXARGS && !ce_stage(ce))
285 continue;
286 status = exec_grep(argc, argv);
287 if (0 < status)
288 hit = 1;
289 argc = nr;
290 if (ce_stage(ce)) {
291 do {
292 i++;
293 } while (i < active_nr &&
294 !strcmp(ce->name, active_cache[i]->name));
295 i--; /* compensate for loop control */
296 }
297 }
298 if (argc > nr) {
299 status = exec_grep(argc, argv);
300 if (0 < status)
301 hit = 1;
302 }
303 return hit;
304 }
305 #endif
307 static int grep_cache(struct grep_opt *opt, const char **paths, int cached)
308 {
309 int hit = 0;
310 int nr;
311 read_cache();
313 #ifdef __unix__
314 /*
315 * Use the external "grep" command for the case where
316 * we grep through the checked-out files. It tends to
317 * be a lot more optimized
318 */
319 if (!cached) {
320 hit = external_grep(opt, paths, cached);
321 if (hit >= 0)
322 return hit;
323 }
324 #endif
326 for (nr = 0; nr < active_nr; nr++) {
327 struct cache_entry *ce = active_cache[nr];
328 if (!S_ISREG(ntohl(ce->ce_mode)))
329 continue;
330 if (!pathspec_matches(paths, ce->name))
331 continue;
332 if (cached) {
333 if (ce_stage(ce))
334 continue;
335 hit |= grep_sha1(opt, ce->sha1, ce->name, 0);
336 }
337 else
338 hit |= grep_file(opt, ce->name);
339 if (ce_stage(ce)) {
340 do {
341 nr++;
342 } while (nr < active_nr &&
343 !strcmp(ce->name, active_cache[nr]->name));
344 nr--; /* compensate for loop control */
345 }
346 }
347 free_grep_patterns(opt);
348 return hit;
349 }
351 static int grep_tree(struct grep_opt *opt, const char **paths,
352 struct tree_desc *tree,
353 const char *tree_name, const char *base)
354 {
355 int len;
356 int hit = 0;
357 struct name_entry entry;
358 char *down;
359 int tn_len = strlen(tree_name);
360 char *path_buf = xmalloc(PATH_MAX + tn_len + 100);
362 if (tn_len) {
363 tn_len = sprintf(path_buf, "%s:", tree_name);
364 down = path_buf + tn_len;
365 strcat(down, base);
366 }
367 else {
368 down = path_buf;
369 strcpy(down, base);
370 }
371 len = strlen(path_buf);
373 while (tree_entry(tree, &entry)) {
374 strcpy(path_buf + len, entry.path);
376 if (S_ISDIR(entry.mode))
377 /* Match "abc/" against pathspec to
378 * decide if we want to descend into "abc"
379 * directory.
380 */
381 strcpy(path_buf + len + entry.pathlen, "/");
383 if (!pathspec_matches(paths, down))
384 ;
385 else if (S_ISREG(entry.mode))
386 hit |= grep_sha1(opt, entry.sha1, path_buf, tn_len);
387 else if (S_ISDIR(entry.mode)) {
388 enum object_type type;
389 struct tree_desc sub;
390 void *data;
391 data = read_sha1_file(entry.sha1, &type, &sub.size);
392 if (!data)
393 die("unable to read tree (%s)",
394 sha1_to_hex(entry.sha1));
395 sub.buf = data;
396 hit |= grep_tree(opt, paths, &sub, tree_name, down);
397 free(data);
398 }
399 }
400 return hit;
401 }
403 static int grep_object(struct grep_opt *opt, const char **paths,
404 struct object *obj, const char *name)
405 {
406 if (obj->type == OBJ_BLOB)
407 return grep_sha1(opt, obj->sha1, name, 0);
408 if (obj->type == OBJ_COMMIT || obj->type == OBJ_TREE) {
409 struct tree_desc tree;
410 void *data;
411 int hit;
412 data = read_object_with_reference(obj->sha1, tree_type,
413 &tree.size, NULL);
414 if (!data)
415 die("unable to read tree (%s)", sha1_to_hex(obj->sha1));
416 tree.buf = data;
417 hit = grep_tree(opt, paths, &tree, name, "");
418 free(data);
419 return hit;
420 }
421 die("unable to grep from object of type %s", typename(obj->type));
422 }
424 static const char builtin_grep_usage[] =
425 "git-grep <option>* <rev>* [-e] <pattern> [<path>...]";
427 static const char emsg_invalid_context_len[] =
428 "%s: invalid context length argument";
429 static const char emsg_missing_context_len[] =
430 "missing context length argument";
431 static const char emsg_missing_argument[] =
432 "option requires an argument -%s";
434 int cmd_grep(int argc, const char **argv, const char *prefix)
435 {
436 int hit = 0;
437 int cached = 0;
438 int seen_dashdash = 0;
439 struct grep_opt opt;
440 struct object_array list = { 0, 0, NULL };
441 const char **paths = NULL;
442 int i;
444 memset(&opt, 0, sizeof(opt));
445 opt.prefix_length = (prefix && *prefix) ? strlen(prefix) : 0;
446 opt.relative = 1;
447 opt.pathname = 1;
448 opt.pattern_tail = &opt.pattern_list;
449 opt.regflags = REG_NEWLINE;
451 /*
452 * If there is no -- then the paths must exist in the working
453 * tree. If there is no explicit pattern specified with -e or
454 * -f, we take the first unrecognized non option to be the
455 * pattern, but then what follows it must be zero or more
456 * valid refs up to the -- (if exists), and then existing
457 * paths. If there is an explicit pattern, then the first
458 * unrecognized non option is the beginning of the refs list
459 * that continues up to the -- (if exists), and then paths.
460 */
462 while (1 < argc) {
463 const char *arg = argv[1];
464 argc--; argv++;
465 if (!strcmp("--cached", arg)) {
466 cached = 1;
467 continue;
468 }
469 if (!strcmp("-a", arg) ||
470 !strcmp("--text", arg)) {
471 opt.binary = GREP_BINARY_TEXT;
472 continue;
473 }
474 if (!strcmp("-i", arg) ||
475 !strcmp("--ignore-case", arg)) {
476 opt.regflags |= REG_ICASE;
477 continue;
478 }
479 if (!strcmp("-I", arg)) {
480 opt.binary = GREP_BINARY_NOMATCH;
481 continue;
482 }
483 if (!strcmp("-v", arg) ||
484 !strcmp("--invert-match", arg)) {
485 opt.invert = 1;
486 continue;
487 }
488 if (!strcmp("-E", arg) ||
489 !strcmp("--extended-regexp", arg)) {
490 opt.regflags |= REG_EXTENDED;
491 continue;
492 }
493 if (!strcmp("-F", arg) ||
494 !strcmp("--fixed-strings", arg)) {
495 opt.fixed = 1;
496 continue;
497 }
498 if (!strcmp("-G", arg) ||
499 !strcmp("--basic-regexp", arg)) {
500 opt.regflags &= ~REG_EXTENDED;
501 continue;
502 }
503 if (!strcmp("-n", arg)) {
504 opt.linenum = 1;
505 continue;
506 }
507 if (!strcmp("-h", arg)) {
508 opt.pathname = 0;
509 continue;
510 }
511 if (!strcmp("-H", arg)) {
512 opt.pathname = 1;
513 continue;
514 }
515 if (!strcmp("-l", arg) ||
516 !strcmp("--files-with-matches", arg)) {
517 opt.name_only = 1;
518 continue;
519 }
520 if (!strcmp("-L", arg) ||
521 !strcmp("--files-without-match", arg)) {
522 opt.unmatch_name_only = 1;
523 continue;
524 }
525 if (!strcmp("-c", arg) ||
526 !strcmp("--count", arg)) {
527 opt.count = 1;
528 continue;
529 }
530 if (!strcmp("-w", arg) ||
531 !strcmp("--word-regexp", arg)) {
532 opt.word_regexp = 1;
533 continue;
534 }
535 if (!prefixcmp(arg, "-A") ||
536 !prefixcmp(arg, "-B") ||
537 !prefixcmp(arg, "-C") ||
538 (arg[0] == '-' && '1' <= arg[1] && arg[1] <= '9')) {
539 unsigned num;
540 const char *scan;
541 switch (arg[1]) {
542 case 'A': case 'B': case 'C':
543 if (!arg[2]) {
544 if (argc <= 1)
545 die(emsg_missing_context_len);
546 scan = *++argv;
547 argc--;
548 }
549 else
550 scan = arg + 2;
551 break;
552 default:
553 scan = arg + 1;
554 break;
555 }
556 if (sscanf(scan, "%u", &num) != 1)
557 die(emsg_invalid_context_len, scan);
558 switch (arg[1]) {
559 case 'A':
560 opt.post_context = num;
561 break;
562 default:
563 case 'C':
564 opt.post_context = num;
565 case 'B':
566 opt.pre_context = num;
567 break;
568 }
569 continue;
570 }
571 if (!strcmp("-f", arg)) {
572 FILE *patterns;
573 int lno = 0;
574 char buf[1024];
575 if (argc <= 1)
576 die(emsg_missing_argument, arg);
577 patterns = fopen(argv[1], "r");
578 if (!patterns)
579 die("'%s': %s", argv[1], strerror(errno));
580 while (fgets(buf, sizeof(buf), patterns)) {
581 int len = strlen(buf);
582 if (buf[len-1] == '\n')
583 buf[len-1] = 0;
584 /* ignore empty line like grep does */
585 if (!buf[0])
586 continue;
587 append_grep_pattern(&opt, xstrdup(buf),
588 argv[1], ++lno,
589 GREP_PATTERN);
590 }
591 fclose(patterns);
592 argv++;
593 argc--;
594 continue;
595 }
596 if (!strcmp("--not", arg)) {
597 append_grep_pattern(&opt, arg, "command line", 0,
598 GREP_NOT);
599 continue;
600 }
601 if (!strcmp("--and", arg)) {
602 append_grep_pattern(&opt, arg, "command line", 0,
603 GREP_AND);
604 continue;
605 }
606 if (!strcmp("--or", arg))
607 continue; /* no-op */
608 if (!strcmp("(", arg)) {
609 append_grep_pattern(&opt, arg, "command line", 0,
610 GREP_OPEN_PAREN);
611 continue;
612 }
613 if (!strcmp(")", arg)) {
614 append_grep_pattern(&opt, arg, "command line", 0,
615 GREP_CLOSE_PAREN);
616 continue;
617 }
618 if (!strcmp("--all-match", arg)) {
619 opt.all_match = 1;
620 continue;
621 }
622 if (!strcmp("-e", arg)) {
623 if (1 < argc) {
624 append_grep_pattern(&opt, argv[1],
625 "-e option", 0,
626 GREP_PATTERN);
627 argv++;
628 argc--;
629 continue;
630 }
631 die(emsg_missing_argument, arg);
632 }
633 if (!strcmp("--full-name", arg)) {
634 opt.relative = 0;
635 continue;
636 }
637 if (!strcmp("--", arg)) {
638 /* later processing wants to have this at argv[1] */
639 argv--;
640 argc++;
641 break;
642 }
643 if (*arg == '-')
644 usage(builtin_grep_usage);
646 /* First unrecognized non-option token */
647 if (!opt.pattern_list) {
648 append_grep_pattern(&opt, arg, "command line", 0,
649 GREP_PATTERN);
650 break;
651 }
652 else {
653 /* We are looking at the first path or rev;
654 * it is found at argv[1] after leaving the
655 * loop.
656 */
657 argc++; argv--;
658 break;
659 }
660 }
662 if (!opt.pattern_list)
663 die("no pattern given.");
664 if ((opt.regflags != REG_NEWLINE) && opt.fixed)
665 die("cannot mix --fixed-strings and regexp");
666 compile_grep_patterns(&opt);
668 /* Check revs and then paths */
669 for (i = 1; i < argc; i++) {
670 const char *arg = argv[i];
671 unsigned char sha1[20];
672 /* Is it a rev? */
673 if (!get_sha1(arg, sha1)) {
674 struct object *object = parse_object(sha1);
675 if (!object)
676 die("bad object %s", arg);
677 add_object_array(object, arg, &list);
678 continue;
679 }
680 if (!strcmp(arg, "--")) {
681 i++;
682 seen_dashdash = 1;
683 }
684 break;
685 }
687 /* The rest are paths */
688 if (!seen_dashdash) {
689 int j;
690 for (j = i; j < argc; j++)
691 verify_filename(prefix, argv[j]);
692 }
694 if (i < argc) {
695 paths = get_pathspec(prefix, argv + i);
696 if (opt.prefix_length && opt.relative) {
697 /* Make sure we do not get outside of paths */
698 for (i = 0; paths[i]; i++)
699 if (strncmp(prefix, paths[i], opt.prefix_length))
700 die("git-grep: cannot generate relative filenames containing '..'");
701 }
702 }
703 else if (prefix) {
704 paths = xcalloc(2, sizeof(const char *));
705 paths[0] = prefix;
706 paths[1] = NULL;
707 }
709 if (!list.nr)
710 return !grep_cache(&opt, paths, cached);
712 if (cached)
713 die("both --cached and trees are given.");
715 for (i = 0; i < list.nr; i++) {
716 struct object *real_obj;
717 real_obj = deref_tag(list.objects[i].item, NULL, 0);
718 if (grep_object(&opt, paths, real_obj, list.objects[i].name))
719 hit = 1;
720 }
721 free_grep_patterns(&opt);
722 return !hit;
723 }