4 * Copyright (c) 2006 Junio C Hamano
11 #include "tree-walk.h"
17 * git grep pathspecs are somewhat different from diff-tree pathspecs;
18 * pathname wildcards are allowed.
20 static int pathspec_matches(const char **paths
, const char *name
)
23 if (!paths
|| !*paths
)
25 namelen
= strlen(name
);
26 for (i
= 0; paths
[i
]; i
++) {
27 const char *match
= paths
[i
];
28 int matchlen
= strlen(match
);
29 const char *cp
, *meta
;
31 if ((matchlen
<= namelen
) &&
32 !strncmp(name
, match
, matchlen
) &&
33 (match
[matchlen
-1] == '/' ||
34 name
[matchlen
] == '\0' || name
[matchlen
] == '/'))
36 if (!fnmatch(match
, name
, 0))
38 if (name
[namelen
-1] != '/')
41 /* We are being asked if the directory ("name") is worth
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.
48 for (cp
= match
, meta
= NULL
; cp
- match
< matchlen
; cp
++) {
50 if (ch
== '*' || ch
== '[' || ch
== '?') {
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.
64 if (!memcmp(match
, name
, namelen
))
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/".
75 if (!memcmp(match
, name
, meta
- match
))
84 struct grep_pat
*next
;
90 struct grep_pat
*pattern_list
;
91 struct grep_pat
**pattern_tail
;
96 unsigned unmatch_name_only
:1;
98 unsigned word_regexp
:1;
99 #define GREP_BINARY_DEFAULT 0
100 #define GREP_BINARY_NOMATCH 1
101 #define GREP_BINARY_TEXT 2
104 unsigned pre_context
;
105 unsigned post_context
;
108 static void add_pattern(struct grep_opt
*opt
, const char *pat
)
110 struct grep_pat
*p
= xcalloc(1, sizeof(*p
));
112 *opt
->pattern_tail
= p
;
113 opt
->pattern_tail
= &p
->next
;
117 static void compile_patterns(struct grep_opt
*opt
)
120 for (p
= opt
->pattern_list
; p
; p
= p
->next
) {
121 int err
= regcomp(&p
->regexp
, p
->pattern
, opt
->regflags
);
124 regerror(err
, &p
->regexp
, errbuf
, 1024);
126 die("'%s': %s", p
->pattern
, errbuf
);
131 static char *end_of_line(char *cp
, unsigned long *left
)
133 unsigned long l
= *left
;
134 while (l
&& *cp
!= '\n') {
142 static int word_char(char ch
)
144 return isalnum(ch
) || ch
== '_';
147 static void show_line(struct grep_opt
*opt
, const char *bol
, const char *eol
,
148 const char *name
, unsigned lno
, char sign
)
150 printf("%s%c", name
, sign
);
152 printf("%d%c", lno
, sign
);
153 printf("%.*s\n", (int)(eol
-bol
), bol
);
157 * NEEDSWORK: share code with diff.c
159 #define FIRST_FEW_BYTES 8000
160 static int buffer_is_binary(const char *ptr
, unsigned long size
)
162 if (FIRST_FEW_BYTES
< size
)
163 size
= FIRST_FEW_BYTES
;
164 if (memchr(ptr
, 0, size
))
169 static int grep_buffer(struct grep_opt
*opt
, const char *name
,
170 char *buf
, unsigned long size
)
173 unsigned long left
= size
;
175 struct pre_context_line
{
178 } *prev
= NULL
, *pcl
;
179 unsigned last_hit
= 0;
180 unsigned last_shown
= 0;
181 int binary_match_only
= 0;
182 const char *hunk_mark
= "";
185 if (buffer_is_binary(buf
, size
)) {
186 switch (opt
->binary
) {
187 case GREP_BINARY_DEFAULT
:
188 binary_match_only
= 1;
190 case GREP_BINARY_NOMATCH
:
191 return 0; /* Assume unmatch */
198 if (opt
->pre_context
)
199 prev
= xcalloc(opt
->pre_context
, sizeof(*prev
));
200 if (opt
->pre_context
|| opt
->post_context
)
204 regmatch_t pmatch
[10];
209 eol
= end_of_line(bol
, &left
);
213 for (p
= opt
->pattern_list
; p
; p
= p
->next
) {
214 regex_t
*exp
= &p
->regexp
;
215 hit
= !regexec(exp
, bol
, ARRAY_SIZE(pmatch
),
218 if (hit
&& opt
->word_regexp
) {
219 /* Match beginning must be either
220 * beginning of the line, or at word
221 * boundary (i.e. the last char must
222 * not be alnum or underscore).
224 if ((pmatch
[0].rm_so
< 0) ||
225 (eol
- bol
) <= pmatch
[0].rm_so
||
226 (pmatch
[0].rm_eo
< 0) ||
227 (eol
- bol
) < pmatch
[0].rm_eo
)
228 die("regexp returned nonsense");
229 if (pmatch
[0].rm_so
!= 0 &&
230 word_char(bol
[pmatch
[0].rm_so
-1]))
231 continue; /* not a word boundary */
232 if ((eol
-bol
) < pmatch
[0].rm_eo
&&
233 word_char(bol
[pmatch
[0].rm_eo
]))
234 continue; /* not a word boundary */
239 /* "grep -v -e foo -e bla" should list lines
240 * that do not have either, so inversion should
245 if (opt
->unmatch_name_only
) {
252 if (binary_match_only
) {
253 printf("Binary file %s matches\n", name
);
256 if (opt
->name_only
) {
257 printf("%s\n", name
);
260 /* Hit at this line. If we haven't shown the
261 * pre-context lines, we would need to show them.
262 * When asked to do "count", this still show
263 * the context which is nonsense, but the user
264 * deserves to get that ;-).
266 if (opt
->pre_context
) {
268 if (opt
->pre_context
< lno
)
269 from
= lno
- opt
->pre_context
;
272 if (from
<= last_shown
)
273 from
= last_shown
+ 1;
274 if (last_shown
&& from
!= last_shown
+ 1)
277 pcl
= &prev
[lno
-from
-1];
278 show_line(opt
, pcl
->bol
, pcl
->eol
,
284 if (last_shown
&& lno
!= last_shown
+ 1)
287 show_line(opt
, bol
, eol
, name
, lno
, ':');
288 last_shown
= last_hit
= lno
;
291 lno
<= last_hit
+ opt
->post_context
) {
292 /* If the last hit is within the post context,
293 * we need to show this line.
295 if (last_shown
&& lno
!= last_shown
+ 1)
297 show_line(opt
, bol
, eol
, name
, lno
, '-');
300 if (opt
->pre_context
) {
301 memmove(prev
+1, prev
,
302 (opt
->pre_context
-1) * sizeof(*prev
));
316 if (opt
->unmatch_name_only
) {
317 /* We did not see any hit, so we want to show this */
318 printf("%s\n", name
);
323 * The real "grep -c foo *.c" gives many "bar.c:0" lines,
324 * which feels mostly useless but sometimes useful. Maybe
325 * make it another option? For now suppress them.
327 if (opt
->count
&& count
)
328 printf("%s:%u\n", name
, count
);
332 static int grep_sha1(struct grep_opt
*opt
, const unsigned char *sha1
, const char *name
)
338 data
= read_sha1_file(sha1
, type
, &size
);
340 error("'%s': unable to read %s", name
, sha1_to_hex(sha1
));
343 hit
= grep_buffer(opt
, name
, data
, size
);
348 static int grep_file(struct grep_opt
*opt
, const char *filename
)
353 if (lstat(filename
, &st
) < 0) {
356 error("'%s': %s", filename
, strerror(errno
));
360 return 0; /* empty file -- no grep hit */
361 if (!S_ISREG(st
.st_mode
))
363 i
= open(filename
, O_RDONLY
);
366 data
= xmalloc(st
.st_size
+ 1);
367 if (st
.st_size
!= xread(i
, data
, st
.st_size
)) {
368 error("'%s': short read %s", filename
, strerror(errno
));
374 i
= grep_buffer(opt
, filename
, data
, st
.st_size
);
379 static int grep_cache(struct grep_opt
*opt
, const char **paths
, int cached
)
385 for (nr
= 0; nr
< active_nr
; nr
++) {
386 struct cache_entry
*ce
= active_cache
[nr
];
387 if (ce_stage(ce
) || !S_ISREG(ntohl(ce
->ce_mode
)))
389 if (!pathspec_matches(paths
, ce
->name
))
392 hit
|= grep_sha1(opt
, ce
->sha1
, ce
->name
);
394 hit
|= grep_file(opt
, ce
->name
);
399 static int grep_tree(struct grep_opt
*opt
, const char **paths
,
400 struct tree_desc
*tree
,
401 const char *tree_name
, const char *base
)
407 const unsigned char *sha1
;
409 char *path_buf
= xmalloc(PATH_MAX
+ strlen(tree_name
) + 100);
412 int offset
= sprintf(path_buf
, "%s:", tree_name
);
413 down
= path_buf
+ offset
;
420 len
= strlen(path_buf
);
424 sha1
= tree_entry_extract(tree
, &path
, &mode
);
425 pathlen
= strlen(path
);
426 strcpy(path_buf
+ len
, path
);
429 /* Match "abc/" against pathspec to
430 * decide if we want to descend into "abc"
433 strcpy(path_buf
+ len
+ pathlen
, "/");
435 if (!pathspec_matches(paths
, down
))
437 else if (S_ISREG(mode
))
438 hit
|= grep_sha1(opt
, sha1
, path_buf
);
439 else if (S_ISDIR(mode
)) {
441 struct tree_desc sub
;
443 data
= read_sha1_file(sha1
, type
, &sub
.size
);
445 die("unable to read tree (%s)",
448 hit
|= grep_tree(opt
, paths
, &sub
, tree_name
, down
);
451 update_tree_entry(tree
);
456 static int grep_object(struct grep_opt
*opt
, const char **paths
,
457 struct object
*obj
, const char *name
)
459 if (!strcmp(obj
->type
, blob_type
))
460 return grep_sha1(opt
, obj
->sha1
, name
);
461 if (!strcmp(obj
->type
, commit_type
) ||
462 !strcmp(obj
->type
, tree_type
)) {
463 struct tree_desc tree
;
466 data
= read_object_with_reference(obj
->sha1
, tree_type
,
469 die("unable to read tree (%s)", sha1_to_hex(obj
->sha1
));
471 hit
= grep_tree(opt
, paths
, &tree
, name
, "");
475 die("unable to grep from object of type %s", obj
->type
);
478 static const char builtin_grep_usage
[] =
479 "git-grep <option>* <rev>* [-e] <pattern> [<path>...]";
481 int cmd_grep(int argc
, const char **argv
, char **envp
)
484 int no_more_flags
= 0;
485 int seen_noncommit
= 0;
488 struct object_list
*list
, **tail
, *object_list
= NULL
;
489 const char *prefix
= setup_git_directory();
490 const char **paths
= NULL
;
492 memset(&opt
, 0, sizeof(opt
));
493 opt
.pattern_tail
= &opt
.pattern_list
;
494 opt
.regflags
= REG_NEWLINE
;
497 * No point using rev_info, really.
500 const char *arg
= argv
[1];
502 if (!strcmp("--cached", arg
)) {
506 if (!strcmp("-a", arg
) ||
507 !strcmp("--text", arg
)) {
508 opt
.binary
= GREP_BINARY_TEXT
;
511 if (!strcmp("-i", arg
) ||
512 !strcmp("--ignore-case", arg
)) {
513 opt
.regflags
|= REG_ICASE
;
516 if (!strcmp("-I", arg
)) {
517 opt
.binary
= GREP_BINARY_NOMATCH
;
520 if (!strcmp("-v", arg
) ||
521 !strcmp("--invert-match", arg
)) {
525 if (!strcmp("-E", arg
) ||
526 !strcmp("--extended-regexp", arg
)) {
527 opt
.regflags
|= REG_EXTENDED
;
530 if (!strcmp("-G", arg
) ||
531 !strcmp("--basic-regexp", arg
)) {
532 opt
.regflags
&= ~REG_EXTENDED
;
535 if (!strcmp("-n", arg
)) {
539 if (!strcmp("-H", arg
)) {
540 /* We always show the pathname, so this
545 if (!strcmp("-l", arg
) ||
546 !strcmp("--files-with-matches", arg
)) {
550 if (!strcmp("-L", arg
) ||
551 !strcmp("--files-without-match", arg
)) {
552 opt
.unmatch_name_only
= 1;
555 if (!strcmp("-c", arg
) ||
556 !strcmp("--count", arg
)) {
560 if (!strcmp("-w", arg
) ||
561 !strcmp("--word-regexp", arg
)) {
565 if (!strncmp("-A", arg
, 2) ||
566 !strncmp("-B", arg
, 2) ||
567 !strncmp("-C", arg
, 2) ||
568 (arg
[0] == '-' && '1' <= arg
[1] && arg
[1] <= '9')) {
572 case 'A': case 'B': case 'C':
575 usage(builtin_grep_usage
);
586 if (sscanf(scan
, "%u", &num
) != 1)
587 usage(builtin_grep_usage
);
590 opt
.post_context
= num
;
594 opt
.post_context
= num
;
596 opt
.pre_context
= num
;
601 if (!strcmp("-e", arg
)) {
603 add_pattern(&opt
, argv
[1]);
608 usage(builtin_grep_usage
);
610 if (!strcmp("--", arg
)) {
614 /* Either unrecognized option or a single pattern */
615 if (!no_more_flags
&& *arg
== '-')
616 usage(builtin_grep_usage
);
617 if (!opt
.pattern_list
) {
618 add_pattern(&opt
, arg
);
622 /* We are looking at the first path or rev;
623 * it is found at argv[0] after leaving the
630 if (!opt
.pattern_list
)
631 die("no pattern given.");
632 compile_patterns(&opt
);
635 struct object
*object
;
636 struct object_list
*elem
;
637 const char *arg
= argv
[1];
638 unsigned char sha1
[20];
639 if (get_sha1(arg
, sha1
) < 0)
641 object
= parse_object(sha1
);
643 die("bad object %s", arg
);
644 elem
= object_list_insert(object
, tail
);
650 paths
= get_pathspec(prefix
, argv
+ 1);
652 paths
= xcalloc(2, sizeof(const char *));
658 return !grep_cache(&opt
, paths
, cached
);
660 * Do not walk "grep -e foo master next pu -- Documentation/"
661 * but do walk "grep -e foo master..next -- Documentation/".
662 * Ranged request mixed with a blob or tree object, like
663 * "grep -e foo v1.0.0:Documentation/ master..next"
664 * so detect that and complain.
666 for (list
= object_list
; list
; list
= list
->next
) {
667 struct object
*real_obj
;
668 real_obj
= deref_tag(list
->item
, NULL
, 0);
669 if (strcmp(real_obj
->type
, commit_type
))
673 die("both --cached and revisions given.");
675 for (list
= object_list
; list
; list
= list
->next
) {
676 struct object
*real_obj
;
677 real_obj
= deref_tag(list
->item
, NULL
, 0);
678 if (grep_object(&opt
, paths
, real_obj
, list
->name
))