2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
11 #include "cache-tree.h"
17 #include "resolve-undo.h"
20 #include "split-index.h"
23 /* Mask for the name length in ce_flags in the on-disk index */
25 #define CE_NAMEMASK (0x0fff)
29 * The first letter should be 'A'..'Z' for extensions that are not
30 * necessary for a correct operation (i.e. optimization data).
31 * When new extensions are added that _needs_ to be understood in
32 * order to correctly interpret the index file, pick character that
33 * is outside the range, to cause the reader to abort.
36 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
37 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
38 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
39 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
40 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
42 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
43 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
44 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
45 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
47 struct index_state the_index
;
48 static const char *alternate_index_output
;
50 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
52 istate
->cache
[nr
] = ce
;
53 add_name_hash(istate
, ce
);
56 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
58 struct cache_entry
*old
= istate
->cache
[nr
];
60 replace_index_entry_in_base(istate
, old
, ce
);
61 remove_name_hash(istate
, old
);
63 set_index_entry(istate
, nr
, ce
);
64 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
65 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
68 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
70 struct cache_entry
*old
= istate
->cache
[nr
], *new;
71 int namelen
= strlen(new_name
);
73 new = xmalloc(cache_entry_size(namelen
));
74 copy_cache_entry(new, old
);
75 new->ce_flags
&= ~CE_HASHED
;
76 new->ce_namelen
= namelen
;
78 memcpy(new->name
, new_name
, namelen
+ 1);
80 cache_tree_invalidate_path(istate
, old
->name
);
81 untracked_cache_remove_from_index(istate
, old
->name
);
82 remove_index_entry_at(istate
, nr
);
83 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
86 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
88 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
89 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
90 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
91 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
92 sd
->sd_dev
= st
->st_dev
;
93 sd
->sd_ino
= st
->st_ino
;
94 sd
->sd_uid
= st
->st_uid
;
95 sd
->sd_gid
= st
->st_gid
;
96 sd
->sd_size
= st
->st_size
;
99 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
103 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
104 changed
|= MTIME_CHANGED
;
105 if (trust_ctime
&& check_stat
&&
106 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
107 changed
|= CTIME_CHANGED
;
110 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
111 changed
|= MTIME_CHANGED
;
112 if (trust_ctime
&& check_stat
&&
113 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
114 changed
|= CTIME_CHANGED
;
118 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
119 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
120 changed
|= OWNER_CHANGED
;
121 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
122 changed
|= INODE_CHANGED
;
127 * st_dev breaks on network filesystems where different
128 * clients will have different views of what "device"
129 * the filesystem is on
131 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
132 changed
|= INODE_CHANGED
;
135 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
136 changed
|= DATA_CHANGED
;
142 * This only updates the "non-critical" parts of the directory
143 * cache, ie the parts that aren't tracked by GIT, and only used
144 * to validate the cache.
146 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
148 fill_stat_data(&ce
->ce_stat_data
, st
);
150 if (assume_unchanged
)
151 ce
->ce_flags
|= CE_VALID
;
153 if (S_ISREG(st
->st_mode
))
154 ce_mark_uptodate(ce
);
157 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
160 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
163 struct object_id oid
;
164 if (!index_fd(&oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
165 match
= oidcmp(&oid
, &ce
->oid
);
166 /* index_fd() closed the file descriptor already */
171 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
176 enum object_type type
;
177 struct strbuf sb
= STRBUF_INIT
;
179 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
182 buffer
= read_sha1_file(ce
->oid
.hash
, &type
, &size
);
185 match
= memcmp(buffer
, sb
.buf
, size
);
192 static int ce_compare_gitlink(const struct cache_entry
*ce
)
194 unsigned char sha1
[20];
197 * We don't actually require that the .git directory
198 * under GITLINK directory be a valid git directory. It
199 * might even be missing (in case nobody populated that
202 * If so, we consider it always to match.
204 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
206 return hashcmp(sha1
, ce
->oid
.hash
);
209 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
211 switch (st
->st_mode
& S_IFMT
) {
213 if (ce_compare_data(ce
, st
))
217 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
221 if (S_ISGITLINK(ce
->ce_mode
))
222 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
223 /* else fallthrough */
230 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
232 unsigned int changed
= 0;
234 if (ce
->ce_flags
& CE_REMOVE
)
235 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
237 switch (ce
->ce_mode
& S_IFMT
) {
239 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
240 /* We consider only the owner x bit to be relevant for
243 if (trust_executable_bit
&&
244 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
245 changed
|= MODE_CHANGED
;
248 if (!S_ISLNK(st
->st_mode
) &&
249 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
250 changed
|= TYPE_CHANGED
;
253 /* We ignore most of the st_xxx fields for gitlinks */
254 if (!S_ISDIR(st
->st_mode
))
255 changed
|= TYPE_CHANGED
;
256 else if (ce_compare_gitlink(ce
))
257 changed
|= DATA_CHANGED
;
260 die("internal error: ce_mode is %o", ce
->ce_mode
);
263 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
265 /* Racily smudged entry? */
266 if (!ce
->ce_stat_data
.sd_size
) {
267 if (!is_empty_blob_sha1(ce
->oid
.hash
))
268 changed
|= DATA_CHANGED
;
274 static int is_racy_stat(const struct index_state
*istate
,
275 const struct stat_data
*sd
)
277 return (istate
->timestamp
.sec
&&
279 /* nanosecond timestamped files can also be racy! */
280 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
281 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
282 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
284 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
289 static int is_racy_timestamp(const struct index_state
*istate
,
290 const struct cache_entry
*ce
)
292 return (!S_ISGITLINK(ce
->ce_mode
) &&
293 is_racy_stat(istate
, &ce
->ce_stat_data
));
296 int match_stat_data_racy(const struct index_state
*istate
,
297 const struct stat_data
*sd
, struct stat
*st
)
299 if (is_racy_stat(istate
, sd
))
300 return MTIME_CHANGED
;
301 return match_stat_data(sd
, st
);
304 int ie_match_stat(const struct index_state
*istate
,
305 const struct cache_entry
*ce
, struct stat
*st
,
306 unsigned int options
)
308 unsigned int changed
;
309 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
310 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
311 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
314 * If it's marked as always valid in the index, it's
315 * valid whatever the checked-out copy says.
317 * skip-worktree has the same effect with higher precedence
319 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
321 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
325 * Intent-to-add entries have not been added, so the index entry
326 * by definition never matches what is in the work tree until it
327 * actually gets added.
329 if (ce_intent_to_add(ce
))
330 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
332 changed
= ce_match_stat_basic(ce
, st
);
335 * Within 1 second of this sequence:
336 * echo xyzzy >file && git-update-index --add file
337 * running this command:
339 * would give a falsely clean cache entry. The mtime and
340 * length match the cache, and other stat fields do not change.
342 * We could detect this at update-index time (the cache entry
343 * being registered/updated records the same time as "now")
344 * and delay the return from git-update-index, but that would
345 * effectively mean we can make at most one commit per second,
346 * which is not acceptable. Instead, we check cache entries
347 * whose mtime are the same as the index file timestamp more
348 * carefully than others.
350 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
351 if (assume_racy_is_modified
)
352 changed
|= DATA_CHANGED
;
354 changed
|= ce_modified_check_fs(ce
, st
);
360 int ie_modified(const struct index_state
*istate
,
361 const struct cache_entry
*ce
,
362 struct stat
*st
, unsigned int options
)
364 int changed
, changed_fs
;
366 changed
= ie_match_stat(istate
, ce
, st
, options
);
370 * If the mode or type has changed, there's no point in trying
371 * to refresh the entry - it's not going to match
373 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
377 * Immediately after read-tree or update-index --cacheinfo,
378 * the length field is zero, as we have never even read the
379 * lstat(2) information once, and we cannot trust DATA_CHANGED
380 * returned by ie_match_stat() which in turn was returned by
381 * ce_match_stat_basic() to signal that the filesize of the
382 * blob changed. We have to actually go to the filesystem to
383 * see if the contents match, and if so, should answer "unchanged".
385 * The logic does not apply to gitlinks, as ce_match_stat_basic()
386 * already has checked the actual HEAD from the filesystem in the
387 * subproject. If ie_match_stat() already said it is different,
388 * then we know it is.
390 if ((changed
& DATA_CHANGED
) &&
391 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
394 changed_fs
= ce_modified_check_fs(ce
, st
);
396 return changed
| changed_fs
;
400 int base_name_compare(const char *name1
, int len1
, int mode1
,
401 const char *name2
, int len2
, int mode2
)
403 unsigned char c1
, c2
;
404 int len
= len1
< len2
? len1
: len2
;
407 cmp
= memcmp(name1
, name2
, len
);
412 if (!c1
&& S_ISDIR(mode1
))
414 if (!c2
&& S_ISDIR(mode2
))
416 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
420 * df_name_compare() is identical to base_name_compare(), except it
421 * compares conflicting directory/file entries as equal. Note that
422 * while a directory name compares as equal to a regular file, they
423 * then individually compare _differently_ to a filename that has
424 * a dot after the basename (because '\0' < '.' < '/').
426 * This is used by routines that want to traverse the git namespace
427 * but then handle conflicting entries together when possible.
429 int df_name_compare(const char *name1
, int len1
, int mode1
,
430 const char *name2
, int len2
, int mode2
)
432 int len
= len1
< len2
? len1
: len2
, cmp
;
433 unsigned char c1
, c2
;
435 cmp
= memcmp(name1
, name2
, len
);
438 /* Directories and files compare equal (same length, same name) */
442 if (!c1
&& S_ISDIR(mode1
))
445 if (!c2
&& S_ISDIR(mode2
))
447 if (c1
== '/' && !c2
)
449 if (c2
== '/' && !c1
)
454 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
456 size_t min_len
= (len1
< len2
) ? len1
: len2
;
457 int cmp
= memcmp(name1
, name2
, min_len
);
467 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
471 cmp
= name_compare(name1
, len1
, name2
, len2
);
482 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
487 last
= istate
->cache_nr
;
488 while (last
> first
) {
489 int next
= (last
+ first
) >> 1;
490 struct cache_entry
*ce
= istate
->cache
[next
];
491 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
503 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
505 return index_name_stage_pos(istate
, name
, namelen
, 0);
508 int remove_index_entry_at(struct index_state
*istate
, int pos
)
510 struct cache_entry
*ce
= istate
->cache
[pos
];
512 record_resolve_undo(istate
, ce
);
513 remove_name_hash(istate
, ce
);
514 save_or_free_index_entry(istate
, ce
);
515 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
517 if (pos
>= istate
->cache_nr
)
519 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
520 istate
->cache_nr
- pos
);
525 * Remove all cache entries marked for removal, that is where
526 * CE_REMOVE is set in ce_flags. This is much more effective than
527 * calling remove_index_entry_at() for each entry to be removed.
529 void remove_marked_cache_entries(struct index_state
*istate
)
531 struct cache_entry
**ce_array
= istate
->cache
;
534 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
535 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
536 remove_name_hash(istate
, ce_array
[i
]);
537 save_or_free_index_entry(istate
, ce_array
[i
]);
540 ce_array
[j
++] = ce_array
[i
];
542 if (j
== istate
->cache_nr
)
544 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
545 istate
->cache_nr
= j
;
548 int remove_file_from_index(struct index_state
*istate
, const char *path
)
550 int pos
= index_name_pos(istate
, path
, strlen(path
));
553 cache_tree_invalidate_path(istate
, path
);
554 untracked_cache_remove_from_index(istate
, path
);
555 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
556 remove_index_entry_at(istate
, pos
);
560 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
562 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
565 static int index_name_pos_also_unmerged(struct index_state
*istate
,
566 const char *path
, int namelen
)
568 int pos
= index_name_pos(istate
, path
, namelen
);
569 struct cache_entry
*ce
;
574 /* maybe unmerged? */
576 if (pos
>= istate
->cache_nr
||
577 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
580 /* order of preference: stage 2, 1, 3 */
581 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
582 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
583 !compare_name(ce
, path
, namelen
))
588 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
590 int len
= ce_namelen(ce
);
591 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
595 * If we add a filename that aliases in the cache, we will use the
596 * name that we already have - but we don't want to update the same
597 * alias twice, because that implies that there were actually two
598 * different files with aliasing names!
600 * So we use the CE_ADDED flag to verify that the alias was an old
601 * one before we accept it as
603 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
604 struct cache_entry
*ce
,
605 struct cache_entry
*alias
)
608 struct cache_entry
*new;
610 if (alias
->ce_flags
& CE_ADDED
)
611 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
613 /* Ok, create the new entry using the name of the existing alias */
614 len
= ce_namelen(alias
);
615 new = xcalloc(1, cache_entry_size(len
));
616 memcpy(new->name
, alias
->name
, len
);
617 copy_cache_entry(new, ce
);
618 save_or_free_index_entry(istate
, ce
);
622 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
624 unsigned char sha1
[20];
625 if (write_sha1_file("", 0, blob_type
, sha1
))
626 die("cannot create an empty blob in the object database");
627 hashcpy(ce
->oid
.hash
, sha1
);
630 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
632 int size
, namelen
, was_same
;
633 mode_t st_mode
= st
->st_mode
;
634 struct cache_entry
*ce
, *alias
= NULL
;
635 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
636 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
637 int pretend
= flags
& ADD_CACHE_PRETEND
;
638 int intent_only
= flags
& ADD_CACHE_INTENT
;
639 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
640 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
641 int newflags
= HASH_WRITE_OBJECT
;
643 if (flags
& HASH_RENORMALIZE
)
644 newflags
|= HASH_RENORMALIZE
;
646 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
647 return error("%s: can only add regular files, symbolic links or git-directories", path
);
649 namelen
= strlen(path
);
650 if (S_ISDIR(st_mode
)) {
651 while (namelen
&& path
[namelen
-1] == '/')
654 size
= cache_entry_size(namelen
);
655 ce
= xcalloc(1, size
);
656 memcpy(ce
->name
, path
, namelen
);
657 ce
->ce_namelen
= namelen
;
659 fill_stat_cache_info(ce
, st
);
661 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
664 if (trust_executable_bit
&& has_symlinks
) {
665 ce
->ce_mode
= create_ce_mode(st_mode
);
667 /* If there is an existing entry, pick the mode bits and type
668 * from it, otherwise assume unexecutable regular file.
670 struct cache_entry
*ent
;
671 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
673 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
674 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
677 /* When core.ignorecase=true, determine if a directory of the same name but differing
678 * case already exists within the Git repository. If it does, ensure the directory
679 * case of the file being added to the repository matches (is folded into) the existing
680 * entry's directory case.
683 adjust_dirname_case(istate
, ce
->name
);
685 if (!(flags
& HASH_RENORMALIZE
)) {
686 alias
= index_file_exists(istate
, ce
->name
,
687 ce_namelen(ce
), ignore_case
);
690 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
691 /* Nothing changed, really */
692 if (!S_ISGITLINK(alias
->ce_mode
))
693 ce_mark_uptodate(alias
);
694 alias
->ce_flags
|= CE_ADDED
;
701 if (index_path(&ce
->oid
, path
, st
, newflags
)) {
703 return error("unable to index file %s", path
);
706 set_object_name_for_intent_to_add_entry(ce
);
708 if (ignore_case
&& alias
&& different_name(ce
, alias
))
709 ce
= create_alias_ce(istate
, ce
, alias
);
710 ce
->ce_flags
|= CE_ADDED
;
712 /* It was suspected to be racily clean, but it turns out to be Ok */
715 !oidcmp(&alias
->oid
, &ce
->oid
) &&
716 ce
->ce_mode
== alias
->ce_mode
);
720 else if (add_index_entry(istate
, ce
, add_option
)) {
722 return error("unable to add %s to index", path
);
724 if (verbose
&& !was_same
)
725 printf("add '%s'\n", path
);
729 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
732 if (lstat(path
, &st
))
733 die_errno("unable to stat '%s'", path
);
734 return add_to_index(istate
, path
, &st
, flags
);
737 struct cache_entry
*make_cache_entry(unsigned int mode
,
738 const unsigned char *sha1
, const char *path
, int stage
,
739 unsigned int refresh_options
)
742 struct cache_entry
*ce
, *ret
;
744 if (!verify_path(path
)) {
745 error("Invalid path '%s'", path
);
750 size
= cache_entry_size(len
);
751 ce
= xcalloc(1, size
);
753 hashcpy(ce
->oid
.hash
, sha1
);
754 memcpy(ce
->name
, path
, len
);
755 ce
->ce_flags
= create_ce_flags(stage
);
756 ce
->ce_namelen
= len
;
757 ce
->ce_mode
= create_ce_mode(mode
);
759 ret
= refresh_cache_entry(ce
, refresh_options
);
766 * Chmod an index entry with either +x or -x.
768 * Returns -1 if the chmod for the particular cache entry failed (if it's
769 * not a regular file), -2 if an invalid flip argument is passed in, 0
772 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
775 if (!S_ISREG(ce
->ce_mode
))
782 ce
->ce_mode
&= ~0111;
787 cache_tree_invalidate_path(istate
, ce
->name
);
788 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
789 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
794 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
796 int len
= ce_namelen(a
);
797 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
801 * We fundamentally don't like some paths: we don't want
802 * dot or dot-dot anywhere, and for obvious reasons don't
803 * want to recurse into ".git" either.
805 * Also, we don't want double slashes or slashes at the
806 * end that can make pathnames ambiguous.
808 static int verify_dotfile(const char *rest
)
811 * The first character was '.', but that
812 * has already been discarded, we now test
816 /* "." is not allowed */
817 if (*rest
== '\0' || is_dir_sep(*rest
))
822 * ".git" followed by NUL or slash is bad. This
823 * shares the path end test with the ".." case.
827 if (rest
[1] != 'i' && rest
[1] != 'I')
829 if (rest
[2] != 't' && rest
[2] != 'T')
834 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
840 int verify_path(const char *path
)
844 if (has_dos_drive_prefix(path
))
853 if (protect_hfs
&& is_hfs_dotgit(path
))
855 if (protect_ntfs
&& is_ntfs_dotgit(path
))
858 if ((c
== '.' && !verify_dotfile(path
)) ||
859 is_dir_sep(c
) || c
== '\0')
867 * Do we have another file that has the beginning components being a
868 * proper superset of the name we're trying to add?
870 static int has_file_name(struct index_state
*istate
,
871 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
874 int len
= ce_namelen(ce
);
875 int stage
= ce_stage(ce
);
876 const char *name
= ce
->name
;
878 while (pos
< istate
->cache_nr
) {
879 struct cache_entry
*p
= istate
->cache
[pos
++];
881 if (len
>= ce_namelen(p
))
883 if (memcmp(name
, p
->name
, len
))
885 if (ce_stage(p
) != stage
)
887 if (p
->name
[len
] != '/')
889 if (p
->ce_flags
& CE_REMOVE
)
894 remove_index_entry_at(istate
, --pos
);
901 * Like strcmp(), but also return the offset of the first change.
902 * If strings are equal, return the length.
904 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
909 return strcmp(s1
, s2
);
911 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
916 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
920 * Do we have another file with a pathname that is a proper
921 * subset of the name we're trying to add?
923 * That is, is there another file in the index with a path
924 * that matches a sub-directory in the given entry?
926 static int has_dir_name(struct index_state
*istate
,
927 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
930 int stage
= ce_stage(ce
);
931 const char *name
= ce
->name
;
932 const char *slash
= name
+ ce_namelen(ce
);
937 * We are frequently called during an iteration on a sorted
938 * list of pathnames and while building a new index. Therefore,
939 * there is a high probability that this entry will eventually
940 * be appended to the index, rather than inserted in the middle.
941 * If we can confirm that, we can avoid binary searches on the
942 * components of the pathname.
944 * Compare the entry's full path with the last path in the index.
946 if (istate
->cache_nr
> 0) {
947 cmp_last
= strcmp_offset(name
,
948 istate
->cache
[istate
->cache_nr
- 1]->name
,
951 if (len_eq_last
== 0) {
953 * The entry sorts AFTER the last one in the
954 * index and their paths have no common prefix,
955 * so there cannot be a F/D conflict.
960 * The entry sorts AFTER the last one in the
961 * index, but has a common prefix. Fall through
962 * to the loop below to disect the entry's path
963 * and see where the difference is.
966 } else if (cmp_last
== 0) {
968 * The entry exactly matches the last one in the
969 * index, but because of multiple stage and CE_REMOVE
970 * items, we fall through and let the regular search
982 if (slash
<= ce
->name
)
989 * (len + 1) is a directory boundary (including
990 * the trailing slash). And since the loop is
991 * decrementing "slash", the first iteration is
992 * the longest directory prefix; subsequent
993 * iterations consider parent directories.
996 if (len
+ 1 <= len_eq_last
) {
998 * The directory prefix (including the trailing
999 * slash) also appears as a prefix in the last
1000 * entry, so the remainder cannot collide (because
1001 * strcmp said the whole path was greater).
1006 * LT: last: xxx/file_A
1012 if (len
> len_eq_last
) {
1014 * This part of the directory prefix (excluding
1015 * the trailing slash) is longer than the known
1016 * equal portions, so this sub-directory cannot
1017 * collide with a file.
1025 if (istate
->cache_nr
> 0 &&
1026 ce_namelen(istate
->cache
[istate
->cache_nr
- 1]) > len
) {
1028 * The directory prefix lines up with part of
1029 * a longer file or directory name, but sorts
1030 * after it, so this sub-directory cannot
1031 * collide with a file.
1033 * last: xxx/yy-file (because '-' sorts before '/')
1040 * This is a possible collision. Fall through and
1041 * let the regular search code handle it.
1048 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
1051 * Found one, but not so fast. This could
1052 * be a marker that says "I was here, but
1053 * I am being removed". Such an entry is
1054 * not a part of the resulting tree, and
1055 * it is Ok to have a directory at the same
1058 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1062 remove_index_entry_at(istate
, pos
);
1070 * Trivial optimization: if we find an entry that
1071 * already matches the sub-directory, then we know
1072 * we're ok, and we can exit.
1074 while (pos
< istate
->cache_nr
) {
1075 struct cache_entry
*p
= istate
->cache
[pos
];
1076 if ((ce_namelen(p
) <= len
) ||
1077 (p
->name
[len
] != '/') ||
1078 memcmp(p
->name
, name
, len
))
1079 break; /* not our subdirectory */
1080 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1082 * p is at the same stage as our entry, and
1083 * is a subdirectory of what we are looking
1084 * at, so we cannot have conflicts at our
1085 * level or anything shorter.
1094 /* We may be in a situation where we already have path/file and path
1095 * is being added, or we already have path and path/file is being
1096 * added. Either one would result in a nonsense tree that has path
1097 * twice when git-write-tree tries to write it out. Prevent it.
1099 * If ok-to-replace is specified, we remove the conflicting entries
1100 * from the cache so the caller should recompute the insert position.
1101 * When this happens, we return non-zero.
1103 static int check_file_directory_conflict(struct index_state
*istate
,
1104 const struct cache_entry
*ce
,
1105 int pos
, int ok_to_replace
)
1110 * When ce is an "I am going away" entry, we allow it to be added
1112 if (ce
->ce_flags
& CE_REMOVE
)
1116 * We check if the path is a sub-path of a subsequent pathname
1117 * first, since removing those will not change the position
1120 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1123 * Then check if the path might have a clashing sub-directory
1126 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1129 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1132 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1133 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1134 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1135 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1137 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1138 cache_tree_invalidate_path(istate
, ce
->name
);
1141 * If this entry's path sorts after the last entry in the index,
1142 * we can avoid searching for it.
1144 if (istate
->cache_nr
> 0 &&
1145 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1146 pos
= -istate
->cache_nr
- 1;
1148 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1150 /* existing match? Just replace it. */
1153 replace_index_entry(istate
, pos
, ce
);
1158 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1159 untracked_cache_add_to_index(istate
, ce
->name
);
1162 * Inserting a merged entry ("stage 0") into the index
1163 * will always replace all non-merged entries..
1165 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1166 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1168 if (!remove_index_entry_at(istate
, pos
))
1175 if (!verify_path(ce
->name
))
1176 return error("Invalid path '%s'", ce
->name
);
1178 if (!skip_df_check
&&
1179 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1181 return error("'%s' appears as both a file and as a directory",
1183 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
1189 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1193 if (option
& ADD_CACHE_JUST_APPEND
)
1194 pos
= istate
->cache_nr
;
1197 ret
= add_index_entry_with_check(istate
, ce
, option
);
1203 /* Make sure the array is big enough .. */
1204 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1208 if (istate
->cache_nr
> pos
+ 1)
1209 memmove(istate
->cache
+ pos
+ 1,
1210 istate
->cache
+ pos
,
1211 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1212 set_index_entry(istate
, pos
, ce
);
1213 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1218 * "refresh" does not calculate a new sha1 file or bring the
1219 * cache up-to-date for mode/content changes. But what it
1220 * _does_ do is to "re-match" the stat information of a file
1221 * with the cache, so that you can refresh the cache for a
1222 * file that hasn't been changed but where the stat entry is
1225 * For example, you'd want to do this after doing a "git-read-tree",
1226 * to link up the stat cache details with the proper files.
1228 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1229 struct cache_entry
*ce
,
1230 unsigned int options
, int *err
,
1234 struct cache_entry
*updated
;
1236 int refresh
= options
& CE_MATCH_REFRESH
;
1237 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1238 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1239 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1241 if (!refresh
|| ce_uptodate(ce
))
1245 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1246 * that the change to the work tree does not matter and told
1249 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1250 ce_mark_uptodate(ce
);
1253 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1254 ce_mark_uptodate(ce
);
1258 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1266 if (lstat(ce
->name
, &st
) < 0) {
1267 if (ignore_missing
&& errno
== ENOENT
)
1274 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1276 *changed_ret
= changed
;
1279 * The path is unchanged. If we were told to ignore
1280 * valid bit, then we did the actual stat check and
1281 * found that the entry is unmodified. If the entry
1282 * is not marked VALID, this is the place to mark it
1283 * valid again, under "assume unchanged" mode.
1285 if (ignore_valid
&& assume_unchanged
&&
1286 !(ce
->ce_flags
& CE_VALID
))
1287 ; /* mark this one VALID again */
1290 * We do not mark the index itself "modified"
1291 * because CE_UPTODATE flag is in-core only;
1292 * we are not going to write this change out.
1294 if (!S_ISGITLINK(ce
->ce_mode
))
1295 ce_mark_uptodate(ce
);
1300 if (ie_modified(istate
, ce
, &st
, options
)) {
1307 updated
= xmalloc(size
);
1308 memcpy(updated
, ce
, size
);
1309 fill_stat_cache_info(updated
, &st
);
1311 * If ignore_valid is not set, we should leave CE_VALID bit
1312 * alone. Otherwise, paths marked with --no-assume-unchanged
1313 * (i.e. things to be edited) will reacquire CE_VALID bit
1314 * automatically, which is not really what we want.
1316 if (!ignore_valid
&& assume_unchanged
&&
1317 !(ce
->ce_flags
& CE_VALID
))
1318 updated
->ce_flags
&= ~CE_VALID
;
1320 /* istate->cache_changed is updated in the caller */
1324 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1325 int * first
, const char *header_msg
)
1327 if (in_porcelain
&& *first
&& header_msg
) {
1328 printf("%s\n", header_msg
);
1334 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1335 const struct pathspec
*pathspec
,
1336 char *seen
, const char *header_msg
)
1340 int really
= (flags
& REFRESH_REALLY
) != 0;
1341 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1342 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1343 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1344 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1346 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1347 unsigned int options
= (CE_MATCH_REFRESH
|
1348 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1349 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1350 const char *modified_fmt
;
1351 const char *deleted_fmt
;
1352 const char *typechange_fmt
;
1353 const char *added_fmt
;
1354 const char *unmerged_fmt
;
1356 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1357 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1358 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1359 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1360 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1361 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1362 struct cache_entry
*ce
, *new;
1363 int cache_errno
= 0;
1367 ce
= istate
->cache
[i
];
1368 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1371 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1375 while ((i
< istate
->cache_nr
) &&
1376 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1382 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1383 &first
, header_msg
);
1391 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1397 if (really
&& cache_errno
== EINVAL
) {
1398 /* If we are doing --really-refresh that
1399 * means the index is not valid anymore.
1401 ce
->ce_flags
&= ~CE_VALID
;
1402 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1403 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1408 if (cache_errno
== ENOENT
)
1410 else if (ce_intent_to_add(ce
))
1411 fmt
= added_fmt
; /* must be before other checks */
1412 else if (changed
& TYPE_CHANGED
)
1413 fmt
= typechange_fmt
;
1417 ce
->name
, in_porcelain
, &first
, header_msg
);
1422 replace_index_entry(istate
, i
, new);
1427 struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1428 unsigned int options
)
1430 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1434 /*****************************************************************
1436 *****************************************************************/
1438 #define INDEX_FORMAT_DEFAULT 3
1440 static unsigned int get_index_format_default(void)
1442 char *envversion
= getenv("GIT_INDEX_VERSION");
1445 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1448 if (!git_config_get_int("index.version", &value
))
1450 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1451 warning(_("index.version set, but the value is invalid.\n"
1452 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1453 return INDEX_FORMAT_DEFAULT
;
1458 version
= strtoul(envversion
, &endp
, 10);
1460 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1461 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1462 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1463 version
= INDEX_FORMAT_DEFAULT
;
1469 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1470 * Again - this is just a (very strong in practice) heuristic that
1471 * the inode hasn't changed.
1473 * We save the fields in big-endian order to allow using the
1474 * index file over NFS transparently.
1476 struct ondisk_cache_entry
{
1477 struct cache_time ctime
;
1478 struct cache_time mtime
;
1485 unsigned char sha1
[20];
1487 char name
[FLEX_ARRAY
]; /* more */
1491 * This struct is used when CE_EXTENDED bit is 1
1492 * The struct must match ondisk_cache_entry exactly from
1495 struct ondisk_cache_entry_extended
{
1496 struct cache_time ctime
;
1497 struct cache_time mtime
;
1504 unsigned char sha1
[20];
1507 char name
[FLEX_ARRAY
]; /* more */
1510 /* These are only used for v3 or lower */
1511 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1512 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1513 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1514 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1515 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1516 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1517 ondisk_cache_entry_size(ce_namelen(ce)))
1519 /* Allow fsck to force verification of the index checksum. */
1520 int verify_index_checksum
;
1522 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1525 unsigned char sha1
[20];
1528 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1529 return error("bad signature");
1530 hdr_version
= ntohl(hdr
->hdr_version
);
1531 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1532 return error("bad index version %d", hdr_version
);
1534 if (!verify_index_checksum
)
1538 git_SHA1_Update(&c
, hdr
, size
- 20);
1539 git_SHA1_Final(sha1
, &c
);
1540 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1541 return error("bad index file sha1 signature");
1545 static int read_index_extension(struct index_state
*istate
,
1546 const char *ext
, void *data
, unsigned long sz
)
1548 switch (CACHE_EXT(ext
)) {
1549 case CACHE_EXT_TREE
:
1550 istate
->cache_tree
= cache_tree_read(data
, sz
);
1552 case CACHE_EXT_RESOLVE_UNDO
:
1553 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1555 case CACHE_EXT_LINK
:
1556 if (read_link_extension(istate
, data
, sz
))
1559 case CACHE_EXT_UNTRACKED
:
1560 istate
->untracked
= read_untracked_extension(data
, sz
);
1563 if (*ext
< 'A' || 'Z' < *ext
)
1564 return error("index uses %.4s extension, which we do not understand",
1566 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1572 int hold_locked_index(struct lock_file
*lk
, int lock_flags
)
1574 return hold_lock_file_for_update(lk
, get_index_file(), lock_flags
);
1577 int read_index(struct index_state
*istate
)
1579 return read_index_from(istate
, get_index_file());
1582 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1587 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1589 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1590 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1591 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1592 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1593 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1594 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1595 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1596 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1597 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1598 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1599 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1600 ce
->ce_namelen
= len
;
1602 hashcpy(ce
->oid
.hash
, ondisk
->sha1
);
1603 memcpy(ce
->name
, name
, len
);
1604 ce
->name
[len
] = '\0';
1609 * Adjacent cache entries tend to share the leading paths, so it makes
1610 * sense to only store the differences in later entries. In the v4
1611 * on-disk format of the index, each on-disk cache entry stores the
1612 * number of bytes to be stripped from the end of the previous name,
1613 * and the bytes to append to the result, to come up with its name.
1615 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1617 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1618 size_t len
= decode_varint(&cp
);
1620 if (name
->len
< len
)
1621 die("malformed name field in the index");
1622 strbuf_remove(name
, name
->len
- len
, len
);
1623 for (ep
= cp
; *ep
; ep
++)
1624 ; /* find the end */
1625 strbuf_add(name
, cp
, ep
- cp
);
1626 return (const char *)ep
+ 1 - cp_
;
1629 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1630 unsigned long *ent_size
,
1631 struct strbuf
*previous_name
)
1633 struct cache_entry
*ce
;
1638 /* On-disk flags are just 16 bits */
1639 flags
= get_be16(&ondisk
->flags
);
1640 len
= flags
& CE_NAMEMASK
;
1642 if (flags
& CE_EXTENDED
) {
1643 struct ondisk_cache_entry_extended
*ondisk2
;
1645 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1646 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1647 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1648 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1649 die("Unknown index entry format %08x", extended_flags
);
1650 flags
|= extended_flags
;
1651 name
= ondisk2
->name
;
1654 name
= ondisk
->name
;
1656 if (!previous_name
) {
1657 /* v3 and earlier */
1658 if (len
== CE_NAMEMASK
)
1660 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1662 *ent_size
= ondisk_ce_size(ce
);
1664 unsigned long consumed
;
1665 consumed
= expand_name_field(previous_name
, name
);
1666 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1668 previous_name
->len
);
1670 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1675 static void check_ce_order(struct index_state
*istate
)
1679 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1680 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1681 struct cache_entry
*next_ce
= istate
->cache
[i
];
1682 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1684 if (0 < name_compare
)
1685 die("unordered stage entries in index");
1686 if (!name_compare
) {
1688 die("multiple stage entries for merged file '%s'",
1690 if (ce_stage(ce
) > ce_stage(next_ce
))
1691 die("unordered stage entries for '%s'",
1697 static void tweak_untracked_cache(struct index_state
*istate
)
1699 switch (git_config_get_untracked_cache()) {
1700 case -1: /* keep: do nothing */
1703 remove_untracked_cache(istate
);
1706 add_untracked_cache(istate
);
1708 default: /* unknown value: do nothing */
1713 static void tweak_split_index(struct index_state
*istate
)
1715 switch (git_config_get_split_index()) {
1716 case -1: /* unset: do nothing */
1719 remove_split_index(istate
);
1722 add_split_index(istate
);
1724 default: /* unknown value: do nothing */
1729 static void post_read_index_from(struct index_state
*istate
)
1731 check_ce_order(istate
);
1732 tweak_untracked_cache(istate
);
1733 tweak_split_index(istate
);
1736 /* remember to discard_cache() before reading a different cache! */
1737 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
1741 unsigned long src_offset
;
1742 struct cache_header
*hdr
;
1745 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1747 if (istate
->initialized
)
1748 return istate
->cache_nr
;
1750 istate
->timestamp
.sec
= 0;
1751 istate
->timestamp
.nsec
= 0;
1752 fd
= open(path
, O_RDONLY
);
1754 if (!must_exist
&& errno
== ENOENT
)
1756 die_errno("%s: index file open failed", path
);
1760 die_errno("cannot stat the open index");
1762 mmap_size
= xsize_t(st
.st_size
);
1763 if (mmap_size
< sizeof(struct cache_header
) + 20)
1764 die("index file smaller than expected");
1766 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1767 if (mmap
== MAP_FAILED
)
1768 die_errno("unable to map index file");
1772 if (verify_hdr(hdr
, mmap_size
) < 0)
1775 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1776 istate
->version
= ntohl(hdr
->hdr_version
);
1777 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1778 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1779 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1780 istate
->initialized
= 1;
1782 if (istate
->version
== 4)
1783 previous_name
= &previous_name_buf
;
1785 previous_name
= NULL
;
1787 src_offset
= sizeof(*hdr
);
1788 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1789 struct ondisk_cache_entry
*disk_ce
;
1790 struct cache_entry
*ce
;
1791 unsigned long consumed
;
1793 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1794 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1795 set_index_entry(istate
, i
, ce
);
1797 src_offset
+= consumed
;
1799 strbuf_release(&previous_name_buf
);
1800 istate
->timestamp
.sec
= st
.st_mtime
;
1801 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1803 while (src_offset
<= mmap_size
- 20 - 8) {
1804 /* After an array of active_nr index entries,
1805 * there can be arbitrary number of extended
1806 * sections, each of which is prefixed with
1807 * extension name (4-byte) and section length
1808 * in 4-byte network byte order.
1811 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1812 extsize
= ntohl(extsize
);
1813 if (read_index_extension(istate
,
1814 (const char *) mmap
+ src_offset
,
1815 (char *) mmap
+ src_offset
+ 8,
1819 src_offset
+= extsize
;
1821 munmap(mmap
, mmap_size
);
1822 return istate
->cache_nr
;
1825 munmap(mmap
, mmap_size
);
1826 die("index file corrupt");
1830 * Signal that the shared index is used by updating its mtime.
1832 * This way, shared index can be removed if they have not been used
1835 static void freshen_shared_index(char *base_sha1_hex
, int warn
)
1837 char *shared_index
= git_pathdup("sharedindex.%s", base_sha1_hex
);
1838 if (!check_and_freshen_file(shared_index
, 1) && warn
)
1839 warning("could not freshen shared index '%s'", shared_index
);
1843 int read_index_from(struct index_state
*istate
, const char *path
)
1845 struct split_index
*split_index
;
1847 char *base_sha1_hex
;
1848 const char *base_path
;
1850 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1851 if (istate
->initialized
)
1852 return istate
->cache_nr
;
1854 ret
= do_read_index(istate
, path
, 0);
1856 split_index
= istate
->split_index
;
1857 if (!split_index
|| is_null_sha1(split_index
->base_sha1
)) {
1858 post_read_index_from(istate
);
1862 if (split_index
->base
)
1863 discard_index(split_index
->base
);
1865 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1867 base_sha1_hex
= sha1_to_hex(split_index
->base_sha1
);
1868 base_path
= git_path("sharedindex.%s", base_sha1_hex
);
1869 ret
= do_read_index(split_index
->base
, base_path
, 1);
1870 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1871 die("broken index, expect %s in %s, got %s",
1872 base_sha1_hex
, base_path
,
1873 sha1_to_hex(split_index
->base
->sha1
));
1875 freshen_shared_index(base_sha1_hex
, 0);
1876 merge_base_index(istate
);
1877 post_read_index_from(istate
);
1881 int is_index_unborn(struct index_state
*istate
)
1883 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1886 int discard_index(struct index_state
*istate
)
1890 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1891 if (istate
->cache
[i
]->index
&&
1892 istate
->split_index
&&
1893 istate
->split_index
->base
&&
1894 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1895 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1897 free(istate
->cache
[i
]);
1899 resolve_undo_clear_index(istate
);
1900 istate
->cache_nr
= 0;
1901 istate
->cache_changed
= 0;
1902 istate
->timestamp
.sec
= 0;
1903 istate
->timestamp
.nsec
= 0;
1904 free_name_hash(istate
);
1905 cache_tree_free(&(istate
->cache_tree
));
1906 istate
->initialized
= 0;
1907 FREE_AND_NULL(istate
->cache
);
1908 istate
->cache_alloc
= 0;
1909 discard_split_index(istate
);
1910 free_untracked_cache(istate
->untracked
);
1911 istate
->untracked
= NULL
;
1915 int unmerged_index(const struct index_state
*istate
)
1918 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1919 if (ce_stage(istate
->cache
[i
]))
1925 #define WRITE_BUFFER_SIZE 8192
1926 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1927 static unsigned long write_buffer_len
;
1929 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1931 unsigned int buffered
= write_buffer_len
;
1933 git_SHA1_Update(context
, write_buffer
, buffered
);
1934 if (write_in_full(fd
, write_buffer
, buffered
) < 0)
1936 write_buffer_len
= 0;
1941 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1944 unsigned int buffered
= write_buffer_len
;
1945 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1948 memcpy(write_buffer
+ buffered
, data
, partial
);
1949 buffered
+= partial
;
1950 if (buffered
== WRITE_BUFFER_SIZE
) {
1951 write_buffer_len
= buffered
;
1952 if (ce_write_flush(context
, fd
))
1956 write_buffer_len
= buffered
;
1958 data
= (char *) data
+ partial
;
1963 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1964 unsigned int ext
, unsigned int sz
)
1968 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1969 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1972 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
1974 unsigned int left
= write_buffer_len
;
1977 write_buffer_len
= 0;
1978 git_SHA1_Update(context
, write_buffer
, left
);
1981 /* Flush first if not enough space for SHA1 signature */
1982 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1983 if (write_in_full(fd
, write_buffer
, left
) < 0)
1988 /* Append the SHA1 signature at the end */
1989 git_SHA1_Final(write_buffer
+ left
, context
);
1990 hashcpy(sha1
, write_buffer
+ left
);
1992 return (write_in_full(fd
, write_buffer
, left
) < 0) ? -1 : 0;
1995 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1998 * The only thing we care about in this function is to smudge the
1999 * falsely clean entry due to touch-update-touch race, so we leave
2000 * everything else as they are. We are called for entries whose
2001 * ce_stat_data.sd_mtime match the index file mtime.
2003 * Note that this actually does not do much for gitlinks, for
2004 * which ce_match_stat_basic() always goes to the actual
2005 * contents. The caller checks with is_racy_timestamp() which
2006 * always says "no" for gitlinks, so we are not called for them ;-)
2010 if (lstat(ce
->name
, &st
) < 0)
2012 if (ce_match_stat_basic(ce
, &st
))
2014 if (ce_modified_check_fs(ce
, &st
)) {
2015 /* This is "racily clean"; smudge it. Note that this
2016 * is a tricky code. At first glance, it may appear
2017 * that it can break with this sequence:
2019 * $ echo xyzzy >frotz
2020 * $ git-update-index --add frotz
2023 * $ echo filfre >nitfol
2024 * $ git-update-index --add nitfol
2026 * but it does not. When the second update-index runs,
2027 * it notices that the entry "frotz" has the same timestamp
2028 * as index, and if we were to smudge it by resetting its
2029 * size to zero here, then the object name recorded
2030 * in index is the 6-byte file but the cached stat information
2031 * becomes zero --- which would then match what we would
2032 * obtain from the filesystem next time we stat("frotz").
2034 * However, the second update-index, before calling
2035 * this function, notices that the cached size is 6
2036 * bytes and what is on the filesystem is an empty
2037 * file, and never calls us, so the cached size information
2038 * for "frotz" stays 6 which does not match the filesystem.
2040 ce
->ce_stat_data
.sd_size
= 0;
2044 /* Copy miscellaneous fields but not the name */
2045 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2046 struct cache_entry
*ce
)
2050 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2051 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2052 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2053 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2054 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2055 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2056 ondisk
->mode
= htonl(ce
->ce_mode
);
2057 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2058 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2059 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2060 hashcpy(ondisk
->sha1
, ce
->oid
.hash
);
2062 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2063 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2064 ondisk
->flags
= htons(flags
);
2065 if (ce
->ce_flags
& CE_EXTENDED
) {
2066 struct ondisk_cache_entry_extended
*ondisk2
;
2067 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
2068 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2072 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
2073 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2076 int saved_namelen
= saved_namelen
; /* compiler workaround */
2078 static unsigned char padding
[8] = { 0x00 };
2080 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2081 saved_namelen
= ce_namelen(ce
);
2085 if (ce
->ce_flags
& CE_EXTENDED
)
2086 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
2088 size
= offsetof(struct ondisk_cache_entry
, name
);
2090 if (!previous_name
) {
2091 int len
= ce_namelen(ce
);
2092 copy_cache_entry_to_ondisk(ondisk
, ce
);
2093 result
= ce_write(c
, fd
, ondisk
, size
);
2095 result
= ce_write(c
, fd
, ce
->name
, len
);
2097 result
= ce_write(c
, fd
, padding
, align_padding_size(size
, len
));
2099 int common
, to_remove
, prefix_size
;
2100 unsigned char to_remove_vi
[16];
2102 (ce
->name
[common
] &&
2103 common
< previous_name
->len
&&
2104 ce
->name
[common
] == previous_name
->buf
[common
]);
2106 ; /* still matching */
2107 to_remove
= previous_name
->len
- common
;
2108 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2110 copy_cache_entry_to_ondisk(ondisk
, ce
);
2111 result
= ce_write(c
, fd
, ondisk
, size
);
2113 result
= ce_write(c
, fd
, to_remove_vi
, prefix_size
);
2115 result
= ce_write(c
, fd
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2117 result
= ce_write(c
, fd
, padding
, 1);
2119 strbuf_splice(previous_name
, common
, to_remove
,
2120 ce
->name
+ common
, ce_namelen(ce
) - common
);
2122 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2123 ce
->ce_namelen
= saved_namelen
;
2124 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2131 * This function verifies if index_state has the correct sha1 of the
2132 * index file. Don't die if we have any other failure, just return 0.
2134 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2139 unsigned char sha1
[20];
2141 if (!istate
->initialized
)
2144 fd
= open(path
, O_RDONLY
);
2151 if (st
.st_size
< sizeof(struct cache_header
) + 20)
2154 n
= pread_in_full(fd
, sha1
, 20, st
.st_size
- 20);
2158 if (hashcmp(istate
->sha1
, sha1
))
2169 static int verify_index(const struct index_state
*istate
)
2171 return verify_index_from(istate
, get_index_file());
2174 static int has_racy_timestamp(struct index_state
*istate
)
2176 int entries
= istate
->cache_nr
;
2179 for (i
= 0; i
< entries
; i
++) {
2180 struct cache_entry
*ce
= istate
->cache
[i
];
2181 if (is_racy_timestamp(istate
, ce
))
2188 * Opportunistically update the index but do not complain if we can't
2190 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
2192 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
2193 verify_index(istate
) &&
2194 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
2195 rollback_lock_file(lockfile
);
2198 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2199 int strip_extensions
)
2201 int newfd
= tempfile
->fd
;
2203 struct cache_header hdr
;
2204 int i
, err
= 0, removed
, extended
, hdr_version
;
2205 struct cache_entry
**cache
= istate
->cache
;
2206 int entries
= istate
->cache_nr
;
2208 struct ondisk_cache_entry_extended ondisk
;
2209 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2210 int drop_cache_tree
= 0;
2212 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2213 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2216 /* reduce extended entries if possible */
2217 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2218 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2220 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2224 if (!istate
->version
) {
2225 istate
->version
= get_index_format_default();
2226 if (getenv("GIT_TEST_SPLIT_INDEX"))
2227 init_split_index(istate
);
2230 /* demote version 3 to version 2 when the latter suffices */
2231 if (istate
->version
== 3 || istate
->version
== 2)
2232 istate
->version
= extended
? 3 : 2;
2234 hdr_version
= istate
->version
;
2236 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2237 hdr
.hdr_version
= htonl(hdr_version
);
2238 hdr
.hdr_entries
= htonl(entries
- removed
);
2241 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
2244 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2246 for (i
= 0; i
< entries
; i
++) {
2247 struct cache_entry
*ce
= cache
[i
];
2248 if (ce
->ce_flags
& CE_REMOVE
)
2250 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2251 ce_smudge_racily_clean_entry(ce
);
2252 if (is_null_oid(&ce
->oid
)) {
2253 static const char msg
[] = "cache entry has null sha1: %s";
2254 static int allow
= -1;
2257 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2259 warning(msg
, ce
->name
);
2261 err
= error(msg
, ce
->name
);
2263 drop_cache_tree
= 1;
2265 if (ce_write_entry(&c
, newfd
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2271 strbuf_release(&previous_name_buf
);
2276 /* Write extension data here */
2277 if (!strip_extensions
&& istate
->split_index
) {
2278 struct strbuf sb
= STRBUF_INIT
;
2280 err
= write_link_extension(&sb
, istate
) < 0 ||
2281 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
2283 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2284 strbuf_release(&sb
);
2288 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
2289 struct strbuf sb
= STRBUF_INIT
;
2291 cache_tree_write(&sb
, istate
->cache_tree
);
2292 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
2293 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2294 strbuf_release(&sb
);
2298 if (!strip_extensions
&& istate
->resolve_undo
) {
2299 struct strbuf sb
= STRBUF_INIT
;
2301 resolve_undo_write(&sb
, istate
->resolve_undo
);
2302 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
2304 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2305 strbuf_release(&sb
);
2309 if (!strip_extensions
&& istate
->untracked
) {
2310 struct strbuf sb
= STRBUF_INIT
;
2312 write_untracked_extension(&sb
, istate
->untracked
);
2313 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_UNTRACKED
,
2315 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
2316 strbuf_release(&sb
);
2321 if (ce_flush(&c
, newfd
, istate
->sha1
))
2323 if (close_tempfile_gently(tempfile
)) {
2324 error(_("could not close '%s'"), tempfile
->filename
.buf
);
2325 delete_tempfile(&tempfile
);
2328 if (stat(tempfile
->filename
.buf
, &st
))
2330 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
2331 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2335 void set_alternate_index_output(const char *name
)
2337 alternate_index_output
= name
;
2340 static int commit_locked_index(struct lock_file
*lk
)
2342 if (alternate_index_output
)
2343 return commit_lock_file_to(lk
, alternate_index_output
);
2345 return commit_lock_file(lk
);
2348 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2351 int ret
= do_write_index(istate
, lock
->tempfile
, 0);
2354 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
2355 (COMMIT_LOCK
| CLOSE_LOCK
));
2356 if (flags
& COMMIT_LOCK
)
2357 return commit_locked_index(lock
);
2358 else if (flags
& CLOSE_LOCK
)
2359 return close_lock_file_gently(lock
);
2364 static int write_split_index(struct index_state
*istate
,
2365 struct lock_file
*lock
,
2369 prepare_to_write_split_index(istate
);
2370 ret
= do_write_locked_index(istate
, lock
, flags
);
2371 finish_writing_split_index(istate
);
2375 static const char *shared_index_expire
= "2.weeks.ago";
2377 static unsigned long get_shared_index_expire_date(void)
2379 static unsigned long shared_index_expire_date
;
2380 static int shared_index_expire_date_prepared
;
2382 if (!shared_index_expire_date_prepared
) {
2383 git_config_get_expiry("splitindex.sharedindexexpire",
2384 &shared_index_expire
);
2385 shared_index_expire_date
= approxidate(shared_index_expire
);
2386 shared_index_expire_date_prepared
= 1;
2389 return shared_index_expire_date
;
2392 static int should_delete_shared_index(const char *shared_index_path
)
2395 unsigned long expiration
;
2397 /* Check timestamp */
2398 expiration
= get_shared_index_expire_date();
2401 if (stat(shared_index_path
, &st
))
2402 return error_errno(_("could not stat '%s'"), shared_index_path
);
2403 if (st
.st_mtime
> expiration
)
2409 static int clean_shared_index_files(const char *current_hex
)
2412 DIR *dir
= opendir(get_git_dir());
2415 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2417 while ((de
= readdir(dir
)) != NULL
) {
2418 const char *sha1_hex
;
2419 const char *shared_index_path
;
2420 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
2422 if (!strcmp(sha1_hex
, current_hex
))
2424 shared_index_path
= git_path("%s", de
->d_name
);
2425 if (should_delete_shared_index(shared_index_path
) > 0 &&
2426 unlink(shared_index_path
))
2427 warning_errno(_("unable to unlink: %s"), shared_index_path
);
2434 static int write_shared_index(struct index_state
*istate
,
2435 struct lock_file
*lock
, unsigned flags
)
2437 struct tempfile
*temp
;
2438 struct split_index
*si
= istate
->split_index
;
2441 temp
= mks_tempfile(git_path("sharedindex_XXXXXX"));
2443 hashclr(si
->base_sha1
);
2444 return do_write_locked_index(istate
, lock
, flags
);
2446 move_cache_to_base_index(istate
);
2447 ret
= do_write_index(si
->base
, temp
, 1);
2449 delete_tempfile(&temp
);
2452 ret
= adjust_shared_perm(get_tempfile_path(temp
));
2454 int save_errno
= errno
;
2455 error("cannot fix permission bits on %s", get_tempfile_path(temp
));
2456 delete_tempfile(&temp
);
2460 ret
= rename_tempfile(&temp
,
2461 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2463 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2464 clean_shared_index_files(sha1_to_hex(si
->base
->sha1
));
2470 static const int default_max_percent_split_change
= 20;
2472 static int too_many_not_shared_entries(struct index_state
*istate
)
2474 int i
, not_shared
= 0;
2475 int max_split
= git_config_get_max_percent_split_change();
2477 switch (max_split
) {
2479 /* not or badly configured: use the default value */
2480 max_split
= default_max_percent_split_change
;
2483 return 1; /* 0% means always write a new shared index */
2485 return 0; /* 100% means never write a new shared index */
2487 break; /* just use the configured value */
2490 /* Count not shared entries */
2491 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2492 struct cache_entry
*ce
= istate
->cache
[i
];
2497 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
2500 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2503 int new_shared_index
, ret
;
2504 struct split_index
*si
= istate
->split_index
;
2506 if (!si
|| alternate_index_output
||
2507 (istate
->cache_changed
& ~EXTMASK
)) {
2509 hashclr(si
->base_sha1
);
2510 return do_write_locked_index(istate
, lock
, flags
);
2513 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2514 int v
= si
->base_sha1
[0];
2516 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2518 if (too_many_not_shared_entries(istate
))
2519 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
2521 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
2523 if (new_shared_index
) {
2524 ret
= write_shared_index(istate
, lock
, flags
);
2529 ret
= write_split_index(istate
, lock
, flags
);
2531 /* Freshen the shared index only if the split-index was written */
2532 if (!ret
&& !new_shared_index
)
2533 freshen_shared_index(sha1_to_hex(si
->base_sha1
), 1);
2539 * Read the index file that is potentially unmerged into given
2540 * index_state, dropping any unmerged entries. Returns true if
2541 * the index is unmerged. Callers who want to refuse to work
2542 * from an unmerged state can call this and check its return value,
2543 * instead of calling read_cache().
2545 int read_index_unmerged(struct index_state
*istate
)
2551 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2552 struct cache_entry
*ce
= istate
->cache
[i
];
2553 struct cache_entry
*new_ce
;
2559 len
= ce_namelen(ce
);
2560 size
= cache_entry_size(len
);
2561 new_ce
= xcalloc(1, size
);
2562 memcpy(new_ce
->name
, ce
->name
, len
);
2563 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2564 new_ce
->ce_namelen
= len
;
2565 new_ce
->ce_mode
= ce
->ce_mode
;
2566 if (add_index_entry(istate
, new_ce
, 0))
2567 return error("%s: cannot drop to stage #0",
2574 * Returns 1 if the path is an "other" path with respect to
2575 * the index; that is, the path is not mentioned in the index at all,
2576 * either as a file, a directory with some files in the index,
2577 * or as an unmerged entry.
2579 * We helpfully remove a trailing "/" from directories so that
2580 * the output of read_directory can be used as-is.
2582 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2586 if (namelen
&& name
[namelen
- 1] == '/')
2588 pos
= index_name_pos(istate
, name
, namelen
);
2590 return 0; /* exact match */
2592 if (pos
< istate
->cache_nr
) {
2593 struct cache_entry
*ce
= istate
->cache
[pos
];
2594 if (ce_namelen(ce
) == namelen
&&
2595 !memcmp(ce
->name
, name
, namelen
))
2596 return 0; /* Yup, this one exists unmerged */
2601 void *read_blob_data_from_index(const struct index_state
*istate
,
2602 const char *path
, unsigned long *size
)
2606 enum object_type type
;
2610 pos
= index_name_pos(istate
, path
, len
);
2613 * We might be in the middle of a merge, in which
2614 * case we would read stage #2 (ours).
2618 (pos
< 0 && i
< istate
->cache_nr
&&
2619 !strcmp(istate
->cache
[i
]->name
, path
));
2621 if (ce_stage(istate
->cache
[i
]) == 2)
2626 data
= read_sha1_file(istate
->cache
[pos
]->oid
.hash
, &type
, &sz
);
2627 if (!data
|| type
!= OBJ_BLOB
) {
2636 void stat_validity_clear(struct stat_validity
*sv
)
2638 FREE_AND_NULL(sv
->sd
);
2641 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2645 if (stat(path
, &st
) < 0)
2646 return sv
->sd
== NULL
;
2649 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2652 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2656 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2657 stat_validity_clear(sv
);
2660 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2661 fill_stat_data(sv
->sd
, &st
);
2665 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
2667 dst
->untracked
= src
->untracked
;
2668 src
->untracked
= NULL
;