2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
8 #include "cache-tree.h"
14 #include "resolve-undo.h"
17 #include "split-index.h"
19 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
20 unsigned int options
);
22 /* Mask for the name length in ce_flags in the on-disk index */
24 #define CE_NAMEMASK (0x0fff)
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
35 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
37 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
40 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
41 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
42 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED)
44 struct index_state the_index
;
45 static const char *alternate_index_output
;
47 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
49 istate
->cache
[nr
] = ce
;
50 add_name_hash(istate
, ce
);
53 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
55 struct cache_entry
*old
= istate
->cache
[nr
];
57 replace_index_entry_in_base(istate
, old
, ce
);
58 remove_name_hash(istate
, old
);
60 set_index_entry(istate
, nr
, ce
);
61 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
62 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
65 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
67 struct cache_entry
*old
= istate
->cache
[nr
], *new;
68 int namelen
= strlen(new_name
);
70 new = xmalloc(cache_entry_size(namelen
));
71 copy_cache_entry(new, old
);
72 new->ce_flags
&= ~CE_HASHED
;
73 new->ce_namelen
= namelen
;
75 memcpy(new->name
, new_name
, namelen
+ 1);
77 cache_tree_invalidate_path(istate
, old
->name
);
78 remove_index_entry_at(istate
, nr
);
79 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
82 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
84 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
85 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
86 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
87 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
88 sd
->sd_dev
= st
->st_dev
;
89 sd
->sd_ino
= st
->st_ino
;
90 sd
->sd_uid
= st
->st_uid
;
91 sd
->sd_gid
= st
->st_gid
;
92 sd
->sd_size
= st
->st_size
;
95 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
99 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
100 changed
|= MTIME_CHANGED
;
101 if (trust_ctime
&& check_stat
&&
102 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
103 changed
|= CTIME_CHANGED
;
106 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
107 changed
|= MTIME_CHANGED
;
108 if (trust_ctime
&& check_stat
&&
109 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
110 changed
|= CTIME_CHANGED
;
114 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
115 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
116 changed
|= OWNER_CHANGED
;
117 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
118 changed
|= INODE_CHANGED
;
123 * st_dev breaks on network filesystems where different
124 * clients will have different views of what "device"
125 * the filesystem is on
127 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
128 changed
|= INODE_CHANGED
;
131 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
132 changed
|= DATA_CHANGED
;
138 * This only updates the "non-critical" parts of the directory
139 * cache, ie the parts that aren't tracked by GIT, and only used
140 * to validate the cache.
142 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
144 fill_stat_data(&ce
->ce_stat_data
, st
);
146 if (assume_unchanged
)
147 ce
->ce_flags
|= CE_VALID
;
149 if (S_ISREG(st
->st_mode
))
150 ce_mark_uptodate(ce
);
153 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
156 int fd
= open(ce
->name
, O_RDONLY
);
159 unsigned char sha1
[20];
160 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
161 match
= hashcmp(sha1
, ce
->sha1
);
162 /* index_fd() closed the file descriptor already */
167 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
172 enum object_type type
;
173 struct strbuf sb
= STRBUF_INIT
;
175 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
178 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
181 match
= memcmp(buffer
, sb
.buf
, size
);
188 static int ce_compare_gitlink(const struct cache_entry
*ce
)
190 unsigned char sha1
[20];
193 * We don't actually require that the .git directory
194 * under GITLINK directory be a valid git directory. It
195 * might even be missing (in case nobody populated that
198 * If so, we consider it always to match.
200 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
202 return hashcmp(sha1
, ce
->sha1
);
205 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
207 switch (st
->st_mode
& S_IFMT
) {
209 if (ce_compare_data(ce
, st
))
213 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
217 if (S_ISGITLINK(ce
->ce_mode
))
218 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
225 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
227 unsigned int changed
= 0;
229 if (ce
->ce_flags
& CE_REMOVE
)
230 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
232 switch (ce
->ce_mode
& S_IFMT
) {
234 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
235 /* We consider only the owner x bit to be relevant for
238 if (trust_executable_bit
&&
239 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
240 changed
|= MODE_CHANGED
;
243 if (!S_ISLNK(st
->st_mode
) &&
244 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
245 changed
|= TYPE_CHANGED
;
248 /* We ignore most of the st_xxx fields for gitlinks */
249 if (!S_ISDIR(st
->st_mode
))
250 changed
|= TYPE_CHANGED
;
251 else if (ce_compare_gitlink(ce
))
252 changed
|= DATA_CHANGED
;
255 die("internal error: ce_mode is %o", ce
->ce_mode
);
258 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
260 /* Racily smudged entry? */
261 if (!ce
->ce_stat_data
.sd_size
) {
262 if (!is_empty_blob_sha1(ce
->sha1
))
263 changed
|= DATA_CHANGED
;
269 static int is_racy_timestamp(const struct index_state
*istate
,
270 const struct cache_entry
*ce
)
272 return (!S_ISGITLINK(ce
->ce_mode
) &&
273 istate
->timestamp
.sec
&&
275 /* nanosecond timestamped files can also be racy! */
276 (istate
->timestamp
.sec
< ce
->ce_stat_data
.sd_mtime
.sec
||
277 (istate
->timestamp
.sec
== ce
->ce_stat_data
.sd_mtime
.sec
&&
278 istate
->timestamp
.nsec
<= ce
->ce_stat_data
.sd_mtime
.nsec
))
280 istate
->timestamp
.sec
<= ce
->ce_stat_data
.sd_mtime
.sec
285 int ie_match_stat(const struct index_state
*istate
,
286 const struct cache_entry
*ce
, struct stat
*st
,
287 unsigned int options
)
289 unsigned int changed
;
290 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
291 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
292 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
295 * If it's marked as always valid in the index, it's
296 * valid whatever the checked-out copy says.
298 * skip-worktree has the same effect with higher precedence
300 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
302 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
306 * Intent-to-add entries have not been added, so the index entry
307 * by definition never matches what is in the work tree until it
308 * actually gets added.
310 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
311 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
313 changed
= ce_match_stat_basic(ce
, st
);
316 * Within 1 second of this sequence:
317 * echo xyzzy >file && git-update-index --add file
318 * running this command:
320 * would give a falsely clean cache entry. The mtime and
321 * length match the cache, and other stat fields do not change.
323 * We could detect this at update-index time (the cache entry
324 * being registered/updated records the same time as "now")
325 * and delay the return from git-update-index, but that would
326 * effectively mean we can make at most one commit per second,
327 * which is not acceptable. Instead, we check cache entries
328 * whose mtime are the same as the index file timestamp more
329 * carefully than others.
331 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
332 if (assume_racy_is_modified
)
333 changed
|= DATA_CHANGED
;
335 changed
|= ce_modified_check_fs(ce
, st
);
341 int ie_modified(const struct index_state
*istate
,
342 const struct cache_entry
*ce
,
343 struct stat
*st
, unsigned int options
)
345 int changed
, changed_fs
;
347 changed
= ie_match_stat(istate
, ce
, st
, options
);
351 * If the mode or type has changed, there's no point in trying
352 * to refresh the entry - it's not going to match
354 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
358 * Immediately after read-tree or update-index --cacheinfo,
359 * the length field is zero, as we have never even read the
360 * lstat(2) information once, and we cannot trust DATA_CHANGED
361 * returned by ie_match_stat() which in turn was returned by
362 * ce_match_stat_basic() to signal that the filesize of the
363 * blob changed. We have to actually go to the filesystem to
364 * see if the contents match, and if so, should answer "unchanged".
366 * The logic does not apply to gitlinks, as ce_match_stat_basic()
367 * already has checked the actual HEAD from the filesystem in the
368 * subproject. If ie_match_stat() already said it is different,
369 * then we know it is.
371 if ((changed
& DATA_CHANGED
) &&
372 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
375 changed_fs
= ce_modified_check_fs(ce
, st
);
377 return changed
| changed_fs
;
381 int base_name_compare(const char *name1
, int len1
, int mode1
,
382 const char *name2
, int len2
, int mode2
)
384 unsigned char c1
, c2
;
385 int len
= len1
< len2
? len1
: len2
;
388 cmp
= memcmp(name1
, name2
, len
);
393 if (!c1
&& S_ISDIR(mode1
))
395 if (!c2
&& S_ISDIR(mode2
))
397 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
401 * df_name_compare() is identical to base_name_compare(), except it
402 * compares conflicting directory/file entries as equal. Note that
403 * while a directory name compares as equal to a regular file, they
404 * then individually compare _differently_ to a filename that has
405 * a dot after the basename (because '\0' < '.' < '/').
407 * This is used by routines that want to traverse the git namespace
408 * but then handle conflicting entries together when possible.
410 int df_name_compare(const char *name1
, int len1
, int mode1
,
411 const char *name2
, int len2
, int mode2
)
413 int len
= len1
< len2
? len1
: len2
, cmp
;
414 unsigned char c1
, c2
;
416 cmp
= memcmp(name1
, name2
, len
);
419 /* Directories and files compare equal (same length, same name) */
423 if (!c1
&& S_ISDIR(mode1
))
426 if (!c2
&& S_ISDIR(mode2
))
428 if (c1
== '/' && !c2
)
430 if (c2
== '/' && !c1
)
435 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
437 int len
= len1
< len2
? len1
: len2
;
440 cmp
= memcmp(name1
, name2
, len
);
455 int cache_name_compare(const char *name1
, int len1
, const char *name2
, int len2
)
457 return cache_name_stage_compare(name1
, len1
, 0, name2
, len2
, 0);
460 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
465 last
= istate
->cache_nr
;
466 while (last
> first
) {
467 int next
= (last
+ first
) >> 1;
468 struct cache_entry
*ce
= istate
->cache
[next
];
469 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
481 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
483 return index_name_stage_pos(istate
, name
, namelen
, 0);
486 /* Remove entry, return true if there are more entries to go.. */
487 int remove_index_entry_at(struct index_state
*istate
, int pos
)
489 struct cache_entry
*ce
= istate
->cache
[pos
];
491 record_resolve_undo(istate
, ce
);
492 remove_name_hash(istate
, ce
);
493 save_or_free_index_entry(istate
, ce
);
494 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
496 if (pos
>= istate
->cache_nr
)
498 memmove(istate
->cache
+ pos
,
499 istate
->cache
+ pos
+ 1,
500 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
505 * Remove all cache entries marked for removal, that is where
506 * CE_REMOVE is set in ce_flags. This is much more effective than
507 * calling remove_index_entry_at() for each entry to be removed.
509 void remove_marked_cache_entries(struct index_state
*istate
)
511 struct cache_entry
**ce_array
= istate
->cache
;
514 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
515 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
516 remove_name_hash(istate
, ce_array
[i
]);
517 save_or_free_index_entry(istate
, ce_array
[i
]);
520 ce_array
[j
++] = ce_array
[i
];
522 if (j
== istate
->cache_nr
)
524 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
525 istate
->cache_nr
= j
;
528 int remove_file_from_index(struct index_state
*istate
, const char *path
)
530 int pos
= index_name_pos(istate
, path
, strlen(path
));
533 cache_tree_invalidate_path(istate
, path
);
534 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
535 remove_index_entry_at(istate
, pos
);
539 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
541 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
544 static int index_name_pos_also_unmerged(struct index_state
*istate
,
545 const char *path
, int namelen
)
547 int pos
= index_name_pos(istate
, path
, namelen
);
548 struct cache_entry
*ce
;
553 /* maybe unmerged? */
555 if (pos
>= istate
->cache_nr
||
556 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
559 /* order of preference: stage 2, 1, 3 */
560 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
561 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
562 !compare_name(ce
, path
, namelen
))
567 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
569 int len
= ce_namelen(ce
);
570 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
574 * If we add a filename that aliases in the cache, we will use the
575 * name that we already have - but we don't want to update the same
576 * alias twice, because that implies that there were actually two
577 * different files with aliasing names!
579 * So we use the CE_ADDED flag to verify that the alias was an old
580 * one before we accept it as
582 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
583 struct cache_entry
*ce
,
584 struct cache_entry
*alias
)
587 struct cache_entry
*new;
589 if (alias
->ce_flags
& CE_ADDED
)
590 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
592 /* Ok, create the new entry using the name of the existing alias */
593 len
= ce_namelen(alias
);
594 new = xcalloc(1, cache_entry_size(len
));
595 memcpy(new->name
, alias
->name
, len
);
596 copy_cache_entry(new, ce
);
597 save_or_free_index_entry(istate
, ce
);
601 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
603 unsigned char sha1
[20];
604 if (write_sha1_file("", 0, blob_type
, sha1
))
605 die("cannot create an empty blob in the object database");
606 hashcpy(ce
->sha1
, sha1
);
609 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
611 int size
, namelen
, was_same
;
612 mode_t st_mode
= st
->st_mode
;
613 struct cache_entry
*ce
, *alias
;
614 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
615 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
616 int pretend
= flags
& ADD_CACHE_PRETEND
;
617 int intent_only
= flags
& ADD_CACHE_INTENT
;
618 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
619 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
621 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
622 return error("%s: can only add regular files, symbolic links or git-directories", path
);
624 namelen
= strlen(path
);
625 if (S_ISDIR(st_mode
)) {
626 while (namelen
&& path
[namelen
-1] == '/')
629 size
= cache_entry_size(namelen
);
630 ce
= xcalloc(1, size
);
631 memcpy(ce
->name
, path
, namelen
);
632 ce
->ce_namelen
= namelen
;
634 fill_stat_cache_info(ce
, st
);
636 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
638 if (trust_executable_bit
&& has_symlinks
)
639 ce
->ce_mode
= create_ce_mode(st_mode
);
641 /* If there is an existing entry, pick the mode bits and type
642 * from it, otherwise assume unexecutable regular file.
644 struct cache_entry
*ent
;
645 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
647 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
648 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
651 /* When core.ignorecase=true, determine if a directory of the same name but differing
652 * case already exists within the Git repository. If it does, ensure the directory
653 * case of the file being added to the repository matches (is folded into) the existing
654 * entry's directory case.
657 const char *startPtr
= ce
->name
;
658 const char *ptr
= startPtr
;
660 while (*ptr
&& *ptr
!= '/')
663 struct cache_entry
*foundce
;
665 foundce
= index_dir_exists(istate
, ce
->name
, ptr
- ce
->name
- 1);
667 memcpy((void *)startPtr
, foundce
->name
+ (startPtr
- ce
->name
), ptr
- startPtr
);
674 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
675 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
676 /* Nothing changed, really */
678 if (!S_ISGITLINK(alias
->ce_mode
))
679 ce_mark_uptodate(alias
);
680 alias
->ce_flags
|= CE_ADDED
;
684 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
))
685 return error("unable to index file %s", path
);
687 set_object_name_for_intent_to_add_entry(ce
);
689 if (ignore_case
&& alias
&& different_name(ce
, alias
))
690 ce
= create_alias_ce(istate
, ce
, alias
);
691 ce
->ce_flags
|= CE_ADDED
;
693 /* It was suspected to be racily clean, but it turns out to be Ok */
696 !hashcmp(alias
->sha1
, ce
->sha1
) &&
697 ce
->ce_mode
== alias
->ce_mode
);
701 else if (add_index_entry(istate
, ce
, add_option
))
702 return error("unable to add %s to index",path
);
703 if (verbose
&& !was_same
)
704 printf("add '%s'\n", path
);
708 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
711 if (lstat(path
, &st
))
712 die_errno("unable to stat '%s'", path
);
713 return add_to_index(istate
, path
, &st
, flags
);
716 struct cache_entry
*make_cache_entry(unsigned int mode
,
717 const unsigned char *sha1
, const char *path
, int stage
,
718 unsigned int refresh_options
)
721 struct cache_entry
*ce
;
723 if (!verify_path(path
)) {
724 error("Invalid path '%s'", path
);
729 size
= cache_entry_size(len
);
730 ce
= xcalloc(1, size
);
732 hashcpy(ce
->sha1
, sha1
);
733 memcpy(ce
->name
, path
, len
);
734 ce
->ce_flags
= create_ce_flags(stage
);
735 ce
->ce_namelen
= len
;
736 ce
->ce_mode
= create_ce_mode(mode
);
738 return refresh_cache_entry(ce
, refresh_options
);
741 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
743 int len
= ce_namelen(a
);
744 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
748 * We fundamentally don't like some paths: we don't want
749 * dot or dot-dot anywhere, and for obvious reasons don't
750 * want to recurse into ".git" either.
752 * Also, we don't want double slashes or slashes at the
753 * end that can make pathnames ambiguous.
755 static int verify_dotfile(const char *rest
)
758 * The first character was '.', but that
759 * has already been discarded, we now test
763 /* "." is not allowed */
764 if (*rest
== '\0' || is_dir_sep(*rest
))
769 * ".git" followed by NUL or slash is bad. This
770 * shares the path end test with the ".." case.
780 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
786 int verify_path(const char *path
)
790 if (has_dos_drive_prefix(path
))
800 if ((c
== '.' && !verify_dotfile(path
)) ||
801 is_dir_sep(c
) || c
== '\0')
809 * Do we have another file that has the beginning components being a
810 * proper superset of the name we're trying to add?
812 static int has_file_name(struct index_state
*istate
,
813 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
816 int len
= ce_namelen(ce
);
817 int stage
= ce_stage(ce
);
818 const char *name
= ce
->name
;
820 while (pos
< istate
->cache_nr
) {
821 struct cache_entry
*p
= istate
->cache
[pos
++];
823 if (len
>= ce_namelen(p
))
825 if (memcmp(name
, p
->name
, len
))
827 if (ce_stage(p
) != stage
)
829 if (p
->name
[len
] != '/')
831 if (p
->ce_flags
& CE_REMOVE
)
836 remove_index_entry_at(istate
, --pos
);
842 * Do we have another file with a pathname that is a proper
843 * subset of the name we're trying to add?
845 static int has_dir_name(struct index_state
*istate
,
846 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
849 int stage
= ce_stage(ce
);
850 const char *name
= ce
->name
;
851 const char *slash
= name
+ ce_namelen(ce
);
859 if (slash
<= ce
->name
)
864 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
867 * Found one, but not so fast. This could
868 * be a marker that says "I was here, but
869 * I am being removed". Such an entry is
870 * not a part of the resulting tree, and
871 * it is Ok to have a directory at the same
874 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
878 remove_index_entry_at(istate
, pos
);
886 * Trivial optimization: if we find an entry that
887 * already matches the sub-directory, then we know
888 * we're ok, and we can exit.
890 while (pos
< istate
->cache_nr
) {
891 struct cache_entry
*p
= istate
->cache
[pos
];
892 if ((ce_namelen(p
) <= len
) ||
893 (p
->name
[len
] != '/') ||
894 memcmp(p
->name
, name
, len
))
895 break; /* not our subdirectory */
896 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
898 * p is at the same stage as our entry, and
899 * is a subdirectory of what we are looking
900 * at, so we cannot have conflicts at our
901 * level or anything shorter.
910 /* We may be in a situation where we already have path/file and path
911 * is being added, or we already have path and path/file is being
912 * added. Either one would result in a nonsense tree that has path
913 * twice when git-write-tree tries to write it out. Prevent it.
915 * If ok-to-replace is specified, we remove the conflicting entries
916 * from the cache so the caller should recompute the insert position.
917 * When this happens, we return non-zero.
919 static int check_file_directory_conflict(struct index_state
*istate
,
920 const struct cache_entry
*ce
,
921 int pos
, int ok_to_replace
)
926 * When ce is an "I am going away" entry, we allow it to be added
928 if (ce
->ce_flags
& CE_REMOVE
)
932 * We check if the path is a sub-path of a subsequent pathname
933 * first, since removing those will not change the position
936 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
939 * Then check if the path might have a clashing sub-directory
942 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
945 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
948 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
949 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
950 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
951 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
953 cache_tree_invalidate_path(istate
, ce
->name
);
954 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
956 /* existing match? Just replace it. */
959 replace_index_entry(istate
, pos
, ce
);
965 * Inserting a merged entry ("stage 0") into the index
966 * will always replace all non-merged entries..
968 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
969 while (ce_same_name(istate
->cache
[pos
], ce
)) {
971 if (!remove_index_entry_at(istate
, pos
))
978 if (!verify_path(ce
->name
))
979 return error("Invalid path '%s'", ce
->name
);
981 if (!skip_df_check
&&
982 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
984 return error("'%s' appears as both a file and as a directory",
986 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
992 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
996 if (option
& ADD_CACHE_JUST_APPEND
)
997 pos
= istate
->cache_nr
;
1000 ret
= add_index_entry_with_check(istate
, ce
, option
);
1006 /* Make sure the array is big enough .. */
1007 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1011 if (istate
->cache_nr
> pos
+ 1)
1012 memmove(istate
->cache
+ pos
+ 1,
1013 istate
->cache
+ pos
,
1014 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1015 set_index_entry(istate
, pos
, ce
);
1016 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1021 * "refresh" does not calculate a new sha1 file or bring the
1022 * cache up-to-date for mode/content changes. But what it
1023 * _does_ do is to "re-match" the stat information of a file
1024 * with the cache, so that you can refresh the cache for a
1025 * file that hasn't been changed but where the stat entry is
1028 * For example, you'd want to do this after doing a "git-read-tree",
1029 * to link up the stat cache details with the proper files.
1031 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1032 struct cache_entry
*ce
,
1033 unsigned int options
, int *err
,
1037 struct cache_entry
*updated
;
1039 int refresh
= options
& CE_MATCH_REFRESH
;
1040 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1041 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1042 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1044 if (!refresh
|| ce_uptodate(ce
))
1048 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1049 * that the change to the work tree does not matter and told
1052 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1053 ce_mark_uptodate(ce
);
1056 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1057 ce_mark_uptodate(ce
);
1061 if (lstat(ce
->name
, &st
) < 0) {
1062 if (ignore_missing
&& errno
== ENOENT
)
1069 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1071 *changed_ret
= changed
;
1074 * The path is unchanged. If we were told to ignore
1075 * valid bit, then we did the actual stat check and
1076 * found that the entry is unmodified. If the entry
1077 * is not marked VALID, this is the place to mark it
1078 * valid again, under "assume unchanged" mode.
1080 if (ignore_valid
&& assume_unchanged
&&
1081 !(ce
->ce_flags
& CE_VALID
))
1082 ; /* mark this one VALID again */
1085 * We do not mark the index itself "modified"
1086 * because CE_UPTODATE flag is in-core only;
1087 * we are not going to write this change out.
1089 if (!S_ISGITLINK(ce
->ce_mode
))
1090 ce_mark_uptodate(ce
);
1095 if (ie_modified(istate
, ce
, &st
, options
)) {
1102 updated
= xmalloc(size
);
1103 memcpy(updated
, ce
, size
);
1104 fill_stat_cache_info(updated
, &st
);
1106 * If ignore_valid is not set, we should leave CE_VALID bit
1107 * alone. Otherwise, paths marked with --no-assume-unchanged
1108 * (i.e. things to be edited) will reacquire CE_VALID bit
1109 * automatically, which is not really what we want.
1111 if (!ignore_valid
&& assume_unchanged
&&
1112 !(ce
->ce_flags
& CE_VALID
))
1113 updated
->ce_flags
&= ~CE_VALID
;
1115 /* istate->cache_changed is updated in the caller */
1119 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1120 int * first
, const char *header_msg
)
1122 if (in_porcelain
&& *first
&& header_msg
) {
1123 printf("%s\n", header_msg
);
1129 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1130 const struct pathspec
*pathspec
,
1131 char *seen
, const char *header_msg
)
1135 int really
= (flags
& REFRESH_REALLY
) != 0;
1136 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1137 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1138 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1139 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1141 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1142 unsigned int options
= (CE_MATCH_REFRESH
|
1143 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1144 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1145 const char *modified_fmt
;
1146 const char *deleted_fmt
;
1147 const char *typechange_fmt
;
1148 const char *added_fmt
;
1149 const char *unmerged_fmt
;
1151 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1152 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1153 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1154 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1155 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1156 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1157 struct cache_entry
*ce
, *new;
1158 int cache_errno
= 0;
1162 ce
= istate
->cache
[i
];
1163 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1166 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1170 while ((i
< istate
->cache_nr
) &&
1171 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1177 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1178 &first
, header_msg
);
1186 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1192 if (really
&& cache_errno
== EINVAL
) {
1193 /* If we are doing --really-refresh that
1194 * means the index is not valid anymore.
1196 ce
->ce_flags
&= ~CE_VALID
;
1197 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1198 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1203 if (cache_errno
== ENOENT
)
1205 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1206 fmt
= added_fmt
; /* must be before other checks */
1207 else if (changed
& TYPE_CHANGED
)
1208 fmt
= typechange_fmt
;
1212 ce
->name
, in_porcelain
, &first
, header_msg
);
1217 replace_index_entry(istate
, i
, new);
1222 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1223 unsigned int options
)
1225 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1229 /*****************************************************************
1231 *****************************************************************/
1233 #define INDEX_FORMAT_DEFAULT 3
1235 static int index_format_config(const char *var
, const char *value
, void *cb
)
1237 unsigned int *version
= cb
;
1238 if (!strcmp(var
, "index.version")) {
1239 *version
= git_config_int(var
, value
);
1245 static unsigned int get_index_format_default(void)
1247 char *envversion
= getenv("GIT_INDEX_VERSION");
1249 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1252 git_config(index_format_config
, &version
);
1253 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1254 warning(_("index.version set, but the value is invalid.\n"
1255 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1256 return INDEX_FORMAT_DEFAULT
;
1261 version
= strtoul(envversion
, &endp
, 10);
1263 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1264 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1265 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1266 version
= INDEX_FORMAT_DEFAULT
;
1272 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1273 * Again - this is just a (very strong in practice) heuristic that
1274 * the inode hasn't changed.
1276 * We save the fields in big-endian order to allow using the
1277 * index file over NFS transparently.
1279 struct ondisk_cache_entry
{
1280 struct cache_time ctime
;
1281 struct cache_time mtime
;
1288 unsigned char sha1
[20];
1290 char name
[FLEX_ARRAY
]; /* more */
1294 * This struct is used when CE_EXTENDED bit is 1
1295 * The struct must match ondisk_cache_entry exactly from
1298 struct ondisk_cache_entry_extended
{
1299 struct cache_time ctime
;
1300 struct cache_time mtime
;
1307 unsigned char sha1
[20];
1310 char name
[FLEX_ARRAY
]; /* more */
1313 /* These are only used for v3 or lower */
1314 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1315 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1316 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1317 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1318 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1319 ondisk_cache_entry_size(ce_namelen(ce)))
1321 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1324 unsigned char sha1
[20];
1327 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1328 return error("bad signature");
1329 hdr_version
= ntohl(hdr
->hdr_version
);
1330 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1331 return error("bad index version %d", hdr_version
);
1333 git_SHA1_Update(&c
, hdr
, size
- 20);
1334 git_SHA1_Final(sha1
, &c
);
1335 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1336 return error("bad index file sha1 signature");
1340 static int read_index_extension(struct index_state
*istate
,
1341 const char *ext
, void *data
, unsigned long sz
)
1343 switch (CACHE_EXT(ext
)) {
1344 case CACHE_EXT_TREE
:
1345 istate
->cache_tree
= cache_tree_read(data
, sz
);
1347 case CACHE_EXT_RESOLVE_UNDO
:
1348 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1350 case CACHE_EXT_LINK
:
1351 if (read_link_extension(istate
, data
, sz
))
1355 if (*ext
< 'A' || 'Z' < *ext
)
1356 return error("index uses %.4s extension, which we do not understand",
1358 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1364 int read_index(struct index_state
*istate
)
1366 return read_index_from(istate
, get_index_file());
1369 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1374 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1376 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1377 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1378 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1379 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1380 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1381 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1382 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1383 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1384 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1385 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1386 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1387 ce
->ce_namelen
= len
;
1389 hashcpy(ce
->sha1
, ondisk
->sha1
);
1390 memcpy(ce
->name
, name
, len
);
1391 ce
->name
[len
] = '\0';
1396 * Adjacent cache entries tend to share the leading paths, so it makes
1397 * sense to only store the differences in later entries. In the v4
1398 * on-disk format of the index, each on-disk cache entry stores the
1399 * number of bytes to be stripped from the end of the previous name,
1400 * and the bytes to append to the result, to come up with its name.
1402 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1404 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1405 size_t len
= decode_varint(&cp
);
1407 if (name
->len
< len
)
1408 die("malformed name field in the index");
1409 strbuf_remove(name
, name
->len
- len
, len
);
1410 for (ep
= cp
; *ep
; ep
++)
1411 ; /* find the end */
1412 strbuf_add(name
, cp
, ep
- cp
);
1413 return (const char *)ep
+ 1 - cp_
;
1416 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1417 unsigned long *ent_size
,
1418 struct strbuf
*previous_name
)
1420 struct cache_entry
*ce
;
1425 /* On-disk flags are just 16 bits */
1426 flags
= get_be16(&ondisk
->flags
);
1427 len
= flags
& CE_NAMEMASK
;
1429 if (flags
& CE_EXTENDED
) {
1430 struct ondisk_cache_entry_extended
*ondisk2
;
1432 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1433 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1434 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1435 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1436 die("Unknown index entry format %08x", extended_flags
);
1437 flags
|= extended_flags
;
1438 name
= ondisk2
->name
;
1441 name
= ondisk
->name
;
1443 if (!previous_name
) {
1444 /* v3 and earlier */
1445 if (len
== CE_NAMEMASK
)
1447 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1449 *ent_size
= ondisk_ce_size(ce
);
1451 unsigned long consumed
;
1452 consumed
= expand_name_field(previous_name
, name
);
1453 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1455 previous_name
->len
);
1457 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1462 /* remember to discard_cache() before reading a different cache! */
1463 static int do_read_index(struct index_state
*istate
, const char *path
,
1468 unsigned long src_offset
;
1469 struct cache_header
*hdr
;
1472 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1474 if (istate
->initialized
)
1475 return istate
->cache_nr
;
1477 istate
->timestamp
.sec
= 0;
1478 istate
->timestamp
.nsec
= 0;
1479 fd
= open(path
, O_RDONLY
);
1481 if (!must_exist
&& errno
== ENOENT
)
1483 die_errno("%s: index file open failed", path
);
1487 die_errno("cannot stat the open index");
1489 mmap_size
= xsize_t(st
.st_size
);
1490 if (mmap_size
< sizeof(struct cache_header
) + 20)
1491 die("index file smaller than expected");
1493 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1494 if (mmap
== MAP_FAILED
)
1495 die_errno("unable to map index file");
1499 if (verify_hdr(hdr
, mmap_size
) < 0)
1502 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1503 istate
->version
= ntohl(hdr
->hdr_version
);
1504 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1505 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1506 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1507 istate
->initialized
= 1;
1509 if (istate
->version
== 4)
1510 previous_name
= &previous_name_buf
;
1512 previous_name
= NULL
;
1514 src_offset
= sizeof(*hdr
);
1515 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1516 struct ondisk_cache_entry
*disk_ce
;
1517 struct cache_entry
*ce
;
1518 unsigned long consumed
;
1520 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1521 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1522 set_index_entry(istate
, i
, ce
);
1524 src_offset
+= consumed
;
1526 strbuf_release(&previous_name_buf
);
1527 istate
->timestamp
.sec
= st
.st_mtime
;
1528 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1530 while (src_offset
<= mmap_size
- 20 - 8) {
1531 /* After an array of active_nr index entries,
1532 * there can be arbitrary number of extended
1533 * sections, each of which is prefixed with
1534 * extension name (4-byte) and section length
1535 * in 4-byte network byte order.
1538 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1539 extsize
= ntohl(extsize
);
1540 if (read_index_extension(istate
,
1541 (const char *) mmap
+ src_offset
,
1542 (char *) mmap
+ src_offset
+ 8,
1546 src_offset
+= extsize
;
1548 munmap(mmap
, mmap_size
);
1549 return istate
->cache_nr
;
1552 munmap(mmap
, mmap_size
);
1553 die("index file corrupt");
1556 int read_index_from(struct index_state
*istate
, const char *path
)
1558 struct split_index
*split_index
;
1561 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1562 if (istate
->initialized
)
1563 return istate
->cache_nr
;
1565 ret
= do_read_index(istate
, path
, 0);
1566 split_index
= istate
->split_index
;
1570 if (is_null_sha1(split_index
->base_sha1
))
1572 if (istate
->cache_nr
)
1573 die("index in split-index mode must contain no entries");
1575 if (split_index
->base
)
1576 discard_index(split_index
->base
);
1578 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1579 ret
= do_read_index(split_index
->base
,
1580 git_path("sharedindex.%s",
1581 sha1_to_hex(split_index
->base_sha1
)), 1);
1582 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1583 die("broken index, expect %s in %s, got %s",
1584 sha1_to_hex(split_index
->base_sha1
),
1585 git_path("sharedindex.%s",
1586 sha1_to_hex(split_index
->base_sha1
)),
1587 sha1_to_hex(split_index
->base
->sha1
));
1588 merge_base_index(istate
);
1592 int is_index_unborn(struct index_state
*istate
)
1594 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1597 int discard_index(struct index_state
*istate
)
1601 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1602 if (istate
->cache
[i
]->index
&&
1603 istate
->split_index
&&
1604 istate
->split_index
->base
&&
1605 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1606 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1608 free(istate
->cache
[i
]);
1610 resolve_undo_clear_index(istate
);
1611 istate
->cache_nr
= 0;
1612 istate
->cache_changed
= 0;
1613 istate
->timestamp
.sec
= 0;
1614 istate
->timestamp
.nsec
= 0;
1615 free_name_hash(istate
);
1616 cache_tree_free(&(istate
->cache_tree
));
1617 istate
->initialized
= 0;
1618 free(istate
->cache
);
1619 istate
->cache
= NULL
;
1620 istate
->cache_alloc
= 0;
1621 discard_split_index(istate
);
1625 int unmerged_index(const struct index_state
*istate
)
1628 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1629 if (ce_stage(istate
->cache
[i
]))
1635 #define WRITE_BUFFER_SIZE 8192
1636 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1637 static unsigned long write_buffer_len
;
1639 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1641 unsigned int buffered
= write_buffer_len
;
1643 git_SHA1_Update(context
, write_buffer
, buffered
);
1644 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1646 write_buffer_len
= 0;
1651 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1654 unsigned int buffered
= write_buffer_len
;
1655 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1658 memcpy(write_buffer
+ buffered
, data
, partial
);
1659 buffered
+= partial
;
1660 if (buffered
== WRITE_BUFFER_SIZE
) {
1661 write_buffer_len
= buffered
;
1662 if (ce_write_flush(context
, fd
))
1666 write_buffer_len
= buffered
;
1668 data
= (char *) data
+ partial
;
1673 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1674 unsigned int ext
, unsigned int sz
)
1678 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1679 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1682 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
1684 unsigned int left
= write_buffer_len
;
1687 write_buffer_len
= 0;
1688 git_SHA1_Update(context
, write_buffer
, left
);
1691 /* Flush first if not enough space for SHA1 signature */
1692 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1693 if (write_in_full(fd
, write_buffer
, left
) != left
)
1698 /* Append the SHA1 signature at the end */
1699 git_SHA1_Final(write_buffer
+ left
, context
);
1700 hashcpy(sha1
, write_buffer
+ left
);
1702 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1705 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1708 * The only thing we care about in this function is to smudge the
1709 * falsely clean entry due to touch-update-touch race, so we leave
1710 * everything else as they are. We are called for entries whose
1711 * ce_stat_data.sd_mtime match the index file mtime.
1713 * Note that this actually does not do much for gitlinks, for
1714 * which ce_match_stat_basic() always goes to the actual
1715 * contents. The caller checks with is_racy_timestamp() which
1716 * always says "no" for gitlinks, so we are not called for them ;-)
1720 if (lstat(ce
->name
, &st
) < 0)
1722 if (ce_match_stat_basic(ce
, &st
))
1724 if (ce_modified_check_fs(ce
, &st
)) {
1725 /* This is "racily clean"; smudge it. Note that this
1726 * is a tricky code. At first glance, it may appear
1727 * that it can break with this sequence:
1729 * $ echo xyzzy >frotz
1730 * $ git-update-index --add frotz
1733 * $ echo filfre >nitfol
1734 * $ git-update-index --add nitfol
1736 * but it does not. When the second update-index runs,
1737 * it notices that the entry "frotz" has the same timestamp
1738 * as index, and if we were to smudge it by resetting its
1739 * size to zero here, then the object name recorded
1740 * in index is the 6-byte file but the cached stat information
1741 * becomes zero --- which would then match what we would
1742 * obtain from the filesystem next time we stat("frotz").
1744 * However, the second update-index, before calling
1745 * this function, notices that the cached size is 6
1746 * bytes and what is on the filesystem is an empty
1747 * file, and never calls us, so the cached size information
1748 * for "frotz" stays 6 which does not match the filesystem.
1750 ce
->ce_stat_data
.sd_size
= 0;
1754 /* Copy miscellaneous fields but not the name */
1755 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1756 struct cache_entry
*ce
)
1760 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1761 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1762 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1763 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1764 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1765 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1766 ondisk
->mode
= htonl(ce
->ce_mode
);
1767 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1768 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1769 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1770 hashcpy(ondisk
->sha1
, ce
->sha1
);
1772 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
1773 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1774 ondisk
->flags
= htons(flags
);
1775 if (ce
->ce_flags
& CE_EXTENDED
) {
1776 struct ondisk_cache_entry_extended
*ondisk2
;
1777 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1778 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1779 return ondisk2
->name
;
1782 return ondisk
->name
;
1786 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1787 struct strbuf
*previous_name
)
1790 struct ondisk_cache_entry
*ondisk
;
1794 if (!previous_name
) {
1795 size
= ondisk_ce_size(ce
);
1796 ondisk
= xcalloc(1, size
);
1797 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1798 memcpy(name
, ce
->name
, ce_namelen(ce
));
1800 int common
, to_remove
, prefix_size
;
1801 unsigned char to_remove_vi
[16];
1803 (ce
->name
[common
] &&
1804 common
< previous_name
->len
&&
1805 ce
->name
[common
] == previous_name
->buf
[common
]);
1807 ; /* still matching */
1808 to_remove
= previous_name
->len
- common
;
1809 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1811 if (ce
->ce_flags
& CE_EXTENDED
)
1812 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1814 size
= offsetof(struct ondisk_cache_entry
, name
);
1815 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1817 ondisk
= xcalloc(1, size
);
1818 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1819 memcpy(name
, to_remove_vi
, prefix_size
);
1820 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1822 strbuf_splice(previous_name
, common
, to_remove
,
1823 ce
->name
+ common
, ce_namelen(ce
) - common
);
1826 result
= ce_write(c
, fd
, ondisk
, size
);
1831 static int has_racy_timestamp(struct index_state
*istate
)
1833 int entries
= istate
->cache_nr
;
1836 for (i
= 0; i
< entries
; i
++) {
1837 struct cache_entry
*ce
= istate
->cache
[i
];
1838 if (is_racy_timestamp(istate
, ce
))
1845 * Opportunistically update the index but do not complain if we can't
1847 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1849 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1850 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
1851 rollback_lock_file(lockfile
);
1854 static int do_write_index(struct index_state
*istate
, int newfd
)
1857 struct cache_header hdr
;
1858 int i
, err
, removed
, extended
, hdr_version
;
1859 struct cache_entry
**cache
= istate
->cache
;
1860 int entries
= istate
->cache_nr
;
1862 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1864 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1865 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1868 /* reduce extended entries if possible */
1869 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1870 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1872 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1876 if (!istate
->version
)
1877 istate
->version
= get_index_format_default();
1879 /* demote version 3 to version 2 when the latter suffices */
1880 if (istate
->version
== 3 || istate
->version
== 2)
1881 istate
->version
= extended
? 3 : 2;
1883 hdr_version
= istate
->version
;
1885 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1886 hdr
.hdr_version
= htonl(hdr_version
);
1887 hdr
.hdr_entries
= htonl(entries
- removed
);
1890 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1893 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
1894 for (i
= 0; i
< entries
; i
++) {
1895 struct cache_entry
*ce
= cache
[i
];
1896 if (ce
->ce_flags
& CE_REMOVE
)
1898 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
1899 ce_smudge_racily_clean_entry(ce
);
1900 if (is_null_sha1(ce
->sha1
)) {
1901 static const char msg
[] = "cache entry has null sha1: %s";
1902 static int allow
= -1;
1905 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1907 warning(msg
, ce
->name
);
1909 return error(msg
, ce
->name
);
1911 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
1914 strbuf_release(&previous_name_buf
);
1916 /* Write extension data here */
1917 if (istate
->split_index
) {
1918 struct strbuf sb
= STRBUF_INIT
;
1920 err
= write_link_extension(&sb
, istate
) < 0 ||
1921 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
1923 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1924 strbuf_release(&sb
);
1928 if (istate
->cache_tree
) {
1929 struct strbuf sb
= STRBUF_INIT
;
1931 cache_tree_write(&sb
, istate
->cache_tree
);
1932 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1933 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1934 strbuf_release(&sb
);
1938 if (istate
->resolve_undo
) {
1939 struct strbuf sb
= STRBUF_INIT
;
1941 resolve_undo_write(&sb
, istate
->resolve_undo
);
1942 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
1944 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1945 strbuf_release(&sb
);
1950 if (ce_flush(&c
, newfd
, istate
->sha1
) || fstat(newfd
, &st
))
1952 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
1953 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1957 void set_alternate_index_output(const char *name
)
1959 alternate_index_output
= name
;
1962 static int commit_locked_index(struct lock_file
*lk
)
1964 if (alternate_index_output
) {
1965 if (lk
->fd
>= 0 && close_lock_file(lk
))
1967 if (rename(lk
->filename
, alternate_index_output
))
1969 lk
->filename
[0] = 0;
1972 return commit_lock_file(lk
);
1976 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
1979 int ret
= do_write_index(istate
, lock
->fd
);
1982 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
1983 (COMMIT_LOCK
| CLOSE_LOCK
));
1984 if (flags
& COMMIT_LOCK
)
1985 return commit_locked_index(lock
);
1986 else if (flags
& CLOSE_LOCK
)
1987 return close_lock_file(lock
);
1992 static int write_split_index(struct index_state
*istate
,
1993 struct lock_file
*lock
,
1997 prepare_to_write_split_index(istate
);
1998 ret
= do_write_locked_index(istate
, lock
, flags
);
1999 finish_writing_split_index(istate
);
2003 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2006 struct split_index
*si
= istate
->split_index
;
2008 if (!si
|| (istate
->cache_changed
& ~EXTMASK
)) {
2010 hashclr(si
->base_sha1
);
2011 return do_write_locked_index(istate
, lock
, flags
);
2014 return write_split_index(istate
, lock
, flags
);
2018 * Read the index file that is potentially unmerged into given
2019 * index_state, dropping any unmerged entries. Returns true if
2020 * the index is unmerged. Callers who want to refuse to work
2021 * from an unmerged state can call this and check its return value,
2022 * instead of calling read_cache().
2024 int read_index_unmerged(struct index_state
*istate
)
2030 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2031 struct cache_entry
*ce
= istate
->cache
[i
];
2032 struct cache_entry
*new_ce
;
2038 len
= ce_namelen(ce
);
2039 size
= cache_entry_size(len
);
2040 new_ce
= xcalloc(1, size
);
2041 memcpy(new_ce
->name
, ce
->name
, len
);
2042 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2043 new_ce
->ce_namelen
= len
;
2044 new_ce
->ce_mode
= ce
->ce_mode
;
2045 if (add_index_entry(istate
, new_ce
, 0))
2046 return error("%s: cannot drop to stage #0",
2048 i
= index_name_pos(istate
, new_ce
->name
, len
);
2054 * Returns 1 if the path is an "other" path with respect to
2055 * the index; that is, the path is not mentioned in the index at all,
2056 * either as a file, a directory with some files in the index,
2057 * or as an unmerged entry.
2059 * We helpfully remove a trailing "/" from directories so that
2060 * the output of read_directory can be used as-is.
2062 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2066 if (namelen
&& name
[namelen
- 1] == '/')
2068 pos
= index_name_pos(istate
, name
, namelen
);
2070 return 0; /* exact match */
2072 if (pos
< istate
->cache_nr
) {
2073 struct cache_entry
*ce
= istate
->cache
[pos
];
2074 if (ce_namelen(ce
) == namelen
&&
2075 !memcmp(ce
->name
, name
, namelen
))
2076 return 0; /* Yup, this one exists unmerged */
2081 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
2085 enum object_type type
;
2089 pos
= index_name_pos(istate
, path
, len
);
2092 * We might be in the middle of a merge, in which
2093 * case we would read stage #2 (ours).
2097 (pos
< 0 && i
< istate
->cache_nr
&&
2098 !strcmp(istate
->cache
[i
]->name
, path
));
2100 if (ce_stage(istate
->cache
[i
]) == 2)
2105 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
2106 if (!data
|| type
!= OBJ_BLOB
) {
2115 void stat_validity_clear(struct stat_validity
*sv
)
2121 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2125 if (stat(path
, &st
) < 0)
2126 return sv
->sd
== NULL
;
2129 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2132 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2136 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2137 stat_validity_clear(sv
);
2140 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2141 fill_stat_data(sv
->sd
, &st
);