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"
18 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
);
22 * The first letter should be 'A'..'Z' for extensions that are not
23 * necessary for a correct operation (i.e. optimization data).
24 * When new extensions are added that _needs_ to be understood in
25 * order to correctly interpret the index file, pick character that
26 * is outside the range, to cause the reader to abort.
29 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
30 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
31 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
33 struct index_state the_index
;
35 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
37 istate
->cache
[nr
] = ce
;
38 add_name_hash(istate
, ce
);
41 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
43 struct cache_entry
*old
= istate
->cache
[nr
];
45 remove_name_hash(old
);
46 set_index_entry(istate
, nr
, ce
);
47 istate
->cache_changed
= 1;
50 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
52 struct cache_entry
*old
= istate
->cache
[nr
], *new;
53 int namelen
= strlen(new_name
);
55 new = xmalloc(cache_entry_size(namelen
));
56 copy_cache_entry(new, old
);
57 new->ce_flags
&= ~(CE_STATE_MASK
| CE_NAMEMASK
);
58 new->ce_flags
|= (namelen
>= CE_NAMEMASK
? CE_NAMEMASK
: namelen
);
59 memcpy(new->name
, new_name
, namelen
+ 1);
61 cache_tree_invalidate_path(istate
->cache_tree
, old
->name
);
62 remove_index_entry_at(istate
, nr
);
63 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
67 * This only updates the "non-critical" parts of the directory
68 * cache, ie the parts that aren't tracked by GIT, and only used
69 * to validate the cache.
71 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
73 ce
->ce_ctime
.sec
= (unsigned int)st
->st_ctime
;
74 ce
->ce_mtime
.sec
= (unsigned int)st
->st_mtime
;
75 ce
->ce_ctime
.nsec
= ST_CTIME_NSEC(*st
);
76 ce
->ce_mtime
.nsec
= ST_MTIME_NSEC(*st
);
77 ce
->ce_dev
= st
->st_dev
;
78 ce
->ce_ino
= st
->st_ino
;
79 ce
->ce_uid
= st
->st_uid
;
80 ce
->ce_gid
= st
->st_gid
;
81 ce
->ce_size
= st
->st_size
;
84 ce
->ce_flags
|= CE_VALID
;
86 if (S_ISREG(st
->st_mode
))
90 static int ce_compare_data(struct cache_entry
*ce
, struct stat
*st
)
93 int fd
= open(ce
->name
, O_RDONLY
);
96 unsigned char sha1
[20];
97 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
98 match
= hashcmp(sha1
, ce
->sha1
);
99 /* index_fd() closed the file descriptor already */
104 static int ce_compare_link(struct cache_entry
*ce
, size_t expected_size
)
109 enum object_type type
;
110 struct strbuf sb
= STRBUF_INIT
;
112 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
115 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
118 match
= memcmp(buffer
, sb
.buf
, size
);
125 static int ce_compare_gitlink(struct cache_entry
*ce
)
127 unsigned char sha1
[20];
130 * We don't actually require that the .git directory
131 * under GITLINK directory be a valid git directory. It
132 * might even be missing (in case nobody populated that
135 * If so, we consider it always to match.
137 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
139 return hashcmp(sha1
, ce
->sha1
);
142 static int ce_modified_check_fs(struct cache_entry
*ce
, struct stat
*st
)
144 switch (st
->st_mode
& S_IFMT
) {
146 if (ce_compare_data(ce
, st
))
150 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
154 if (S_ISGITLINK(ce
->ce_mode
))
155 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
162 static int is_empty_blob_sha1(const unsigned char *sha1
)
164 static const unsigned char empty_blob_sha1
[20] = {
165 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
166 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
169 return !hashcmp(sha1
, empty_blob_sha1
);
172 static int ce_match_stat_basic(struct cache_entry
*ce
, struct stat
*st
)
174 unsigned int changed
= 0;
176 if (ce
->ce_flags
& CE_REMOVE
)
177 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
179 switch (ce
->ce_mode
& S_IFMT
) {
181 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
182 /* We consider only the owner x bit to be relevant for
185 if (trust_executable_bit
&&
186 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
187 changed
|= MODE_CHANGED
;
190 if (!S_ISLNK(st
->st_mode
) &&
191 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
192 changed
|= TYPE_CHANGED
;
195 /* We ignore most of the st_xxx fields for gitlinks */
196 if (!S_ISDIR(st
->st_mode
))
197 changed
|= TYPE_CHANGED
;
198 else if (ce_compare_gitlink(ce
))
199 changed
|= DATA_CHANGED
;
202 die("internal error: ce_mode is %o", ce
->ce_mode
);
204 if (ce
->ce_mtime
.sec
!= (unsigned int)st
->st_mtime
)
205 changed
|= MTIME_CHANGED
;
206 if (trust_ctime
&& ce
->ce_ctime
.sec
!= (unsigned int)st
->st_ctime
)
207 changed
|= CTIME_CHANGED
;
210 if (ce
->ce_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
211 changed
|= MTIME_CHANGED
;
212 if (trust_ctime
&& ce
->ce_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
213 changed
|= CTIME_CHANGED
;
216 if (ce
->ce_uid
!= (unsigned int) st
->st_uid
||
217 ce
->ce_gid
!= (unsigned int) st
->st_gid
)
218 changed
|= OWNER_CHANGED
;
219 if (ce
->ce_ino
!= (unsigned int) st
->st_ino
)
220 changed
|= INODE_CHANGED
;
224 * st_dev breaks on network filesystems where different
225 * clients will have different views of what "device"
226 * the filesystem is on
228 if (ce
->ce_dev
!= (unsigned int) st
->st_dev
)
229 changed
|= INODE_CHANGED
;
232 if (ce
->ce_size
!= (unsigned int) st
->st_size
)
233 changed
|= DATA_CHANGED
;
235 /* Racily smudged entry? */
237 if (!is_empty_blob_sha1(ce
->sha1
))
238 changed
|= DATA_CHANGED
;
244 static int is_racy_timestamp(const struct index_state
*istate
, struct cache_entry
*ce
)
246 return (!S_ISGITLINK(ce
->ce_mode
) &&
247 istate
->timestamp
.sec
&&
249 /* nanosecond timestamped files can also be racy! */
250 (istate
->timestamp
.sec
< ce
->ce_mtime
.sec
||
251 (istate
->timestamp
.sec
== ce
->ce_mtime
.sec
&&
252 istate
->timestamp
.nsec
<= ce
->ce_mtime
.nsec
))
254 istate
->timestamp
.sec
<= ce
->ce_mtime
.sec
259 int ie_match_stat(const struct index_state
*istate
,
260 struct cache_entry
*ce
, struct stat
*st
,
261 unsigned int options
)
263 unsigned int changed
;
264 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
265 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
266 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
269 * If it's marked as always valid in the index, it's
270 * valid whatever the checked-out copy says.
272 * skip-worktree has the same effect with higher precedence
274 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
276 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
280 * Intent-to-add entries have not been added, so the index entry
281 * by definition never matches what is in the work tree until it
282 * actually gets added.
284 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
285 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
287 changed
= ce_match_stat_basic(ce
, st
);
290 * Within 1 second of this sequence:
291 * echo xyzzy >file && git-update-index --add file
292 * running this command:
294 * would give a falsely clean cache entry. The mtime and
295 * length match the cache, and other stat fields do not change.
297 * We could detect this at update-index time (the cache entry
298 * being registered/updated records the same time as "now")
299 * and delay the return from git-update-index, but that would
300 * effectively mean we can make at most one commit per second,
301 * which is not acceptable. Instead, we check cache entries
302 * whose mtime are the same as the index file timestamp more
303 * carefully than others.
305 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
306 if (assume_racy_is_modified
)
307 changed
|= DATA_CHANGED
;
309 changed
|= ce_modified_check_fs(ce
, st
);
315 int ie_modified(const struct index_state
*istate
,
316 struct cache_entry
*ce
, struct stat
*st
, unsigned int options
)
318 int changed
, changed_fs
;
320 changed
= ie_match_stat(istate
, ce
, st
, options
);
324 * If the mode or type has changed, there's no point in trying
325 * to refresh the entry - it's not going to match
327 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
331 * Immediately after read-tree or update-index --cacheinfo,
332 * the length field is zero, as we have never even read the
333 * lstat(2) information once, and we cannot trust DATA_CHANGED
334 * returned by ie_match_stat() which in turn was returned by
335 * ce_match_stat_basic() to signal that the filesize of the
336 * blob changed. We have to actually go to the filesystem to
337 * see if the contents match, and if so, should answer "unchanged".
339 * The logic does not apply to gitlinks, as ce_match_stat_basic()
340 * already has checked the actual HEAD from the filesystem in the
341 * subproject. If ie_match_stat() already said it is different,
342 * then we know it is.
344 if ((changed
& DATA_CHANGED
) &&
345 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_size
!= 0))
348 changed_fs
= ce_modified_check_fs(ce
, st
);
350 return changed
| changed_fs
;
354 int base_name_compare(const char *name1
, int len1
, int mode1
,
355 const char *name2
, int len2
, int mode2
)
357 unsigned char c1
, c2
;
358 int len
= len1
< len2
? len1
: len2
;
361 cmp
= memcmp(name1
, name2
, len
);
366 if (!c1
&& S_ISDIR(mode1
))
368 if (!c2
&& S_ISDIR(mode2
))
370 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
374 * df_name_compare() is identical to base_name_compare(), except it
375 * compares conflicting directory/file entries as equal. Note that
376 * while a directory name compares as equal to a regular file, they
377 * then individually compare _differently_ to a filename that has
378 * a dot after the basename (because '\0' < '.' < '/').
380 * This is used by routines that want to traverse the git namespace
381 * but then handle conflicting entries together when possible.
383 int df_name_compare(const char *name1
, int len1
, int mode1
,
384 const char *name2
, int len2
, int mode2
)
386 int len
= len1
< len2
? len1
: len2
, cmp
;
387 unsigned char c1
, c2
;
389 cmp
= memcmp(name1
, name2
, len
);
392 /* Directories and files compare equal (same length, same name) */
396 if (!c1
&& S_ISDIR(mode1
))
399 if (!c2
&& S_ISDIR(mode2
))
401 if (c1
== '/' && !c2
)
403 if (c2
== '/' && !c1
)
408 int cache_name_compare(const char *name1
, int flags1
, const char *name2
, int flags2
)
410 int len1
= flags1
& CE_NAMEMASK
;
411 int len2
= flags2
& CE_NAMEMASK
;
412 int len
= len1
< len2
? len1
: len2
;
415 cmp
= memcmp(name1
, name2
, len
);
424 flags1
&= CE_STAGEMASK
;
425 flags2
&= CE_STAGEMASK
;
434 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
439 last
= istate
->cache_nr
;
440 while (last
> first
) {
441 int next
= (last
+ first
) >> 1;
442 struct cache_entry
*ce
= istate
->cache
[next
];
443 int cmp
= cache_name_compare(name
, namelen
, ce
->name
, ce
->ce_flags
);
455 /* Remove entry, return true if there are more entries to go.. */
456 int remove_index_entry_at(struct index_state
*istate
, int pos
)
458 struct cache_entry
*ce
= istate
->cache
[pos
];
460 record_resolve_undo(istate
, ce
);
461 remove_name_hash(ce
);
462 istate
->cache_changed
= 1;
464 if (pos
>= istate
->cache_nr
)
466 memmove(istate
->cache
+ pos
,
467 istate
->cache
+ pos
+ 1,
468 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
473 * Remove all cache ententries marked for removal, that is where
474 * CE_REMOVE is set in ce_flags. This is much more effective than
475 * calling remove_index_entry_at() for each entry to be removed.
477 void remove_marked_cache_entries(struct index_state
*istate
)
479 struct cache_entry
**ce_array
= istate
->cache
;
482 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
483 if (ce_array
[i
]->ce_flags
& CE_REMOVE
)
484 remove_name_hash(ce_array
[i
]);
486 ce_array
[j
++] = ce_array
[i
];
488 istate
->cache_changed
= 1;
489 istate
->cache_nr
= j
;
492 int remove_file_from_index(struct index_state
*istate
, const char *path
)
494 int pos
= index_name_pos(istate
, path
, strlen(path
));
497 cache_tree_invalidate_path(istate
->cache_tree
, path
);
498 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
499 remove_index_entry_at(istate
, pos
);
503 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
505 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
508 static int index_name_pos_also_unmerged(struct index_state
*istate
,
509 const char *path
, int namelen
)
511 int pos
= index_name_pos(istate
, path
, namelen
);
512 struct cache_entry
*ce
;
517 /* maybe unmerged? */
519 if (pos
>= istate
->cache_nr
||
520 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
523 /* order of preference: stage 2, 1, 3 */
524 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
525 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
526 !compare_name(ce
, path
, namelen
))
531 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
533 int len
= ce_namelen(ce
);
534 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
538 * If we add a filename that aliases in the cache, we will use the
539 * name that we already have - but we don't want to update the same
540 * alias twice, because that implies that there were actually two
541 * different files with aliasing names!
543 * So we use the CE_ADDED flag to verify that the alias was an old
544 * one before we accept it as
546 static struct cache_entry
*create_alias_ce(struct cache_entry
*ce
, struct cache_entry
*alias
)
549 struct cache_entry
*new;
551 if (alias
->ce_flags
& CE_ADDED
)
552 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
554 /* Ok, create the new entry using the name of the existing alias */
555 len
= ce_namelen(alias
);
556 new = xcalloc(1, cache_entry_size(len
));
557 memcpy(new->name
, alias
->name
, len
);
558 copy_cache_entry(new, ce
);
563 static void record_intent_to_add(struct cache_entry
*ce
)
565 unsigned char sha1
[20];
566 if (write_sha1_file("", 0, blob_type
, sha1
))
567 die("cannot create an empty blob in the object database");
568 hashcpy(ce
->sha1
, sha1
);
571 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
573 int size
, namelen
, was_same
;
574 mode_t st_mode
= st
->st_mode
;
575 struct cache_entry
*ce
, *alias
;
576 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
577 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
578 int pretend
= flags
& ADD_CACHE_PRETEND
;
579 int intent_only
= flags
& ADD_CACHE_INTENT
;
580 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
581 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
583 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
584 return error("%s: can only add regular files, symbolic links or git-directories", path
);
586 namelen
= strlen(path
);
587 if (S_ISDIR(st_mode
)) {
588 while (namelen
&& path
[namelen
-1] == '/')
591 size
= cache_entry_size(namelen
);
592 ce
= xcalloc(1, size
);
593 memcpy(ce
->name
, path
, namelen
);
594 ce
->ce_flags
= namelen
;
596 fill_stat_cache_info(ce
, st
);
598 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
600 if (trust_executable_bit
&& has_symlinks
)
601 ce
->ce_mode
= create_ce_mode(st_mode
);
603 /* If there is an existing entry, pick the mode bits and type
604 * from it, otherwise assume unexecutable regular file.
606 struct cache_entry
*ent
;
607 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
609 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
610 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
613 /* When core.ignorecase=true, determine if a directory of the same name but differing
614 * case already exists within the Git repository. If it does, ensure the directory
615 * case of the file being added to the repository matches (is folded into) the existing
616 * entry's directory case.
619 const char *startPtr
= ce
->name
;
620 const char *ptr
= startPtr
;
622 while (*ptr
&& *ptr
!= '/')
625 struct cache_entry
*foundce
;
627 foundce
= index_name_exists(&the_index
, ce
->name
, ptr
- ce
->name
, ignore_case
);
629 memcpy((void *)startPtr
, foundce
->name
+ (startPtr
- ce
->name
), ptr
- startPtr
);
636 alias
= index_name_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
637 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
638 /* Nothing changed, really */
640 if (!S_ISGITLINK(alias
->ce_mode
))
641 ce_mark_uptodate(alias
);
642 alias
->ce_flags
|= CE_ADDED
;
646 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
))
647 return error("unable to index file %s", path
);
649 record_intent_to_add(ce
);
651 if (ignore_case
&& alias
&& different_name(ce
, alias
))
652 ce
= create_alias_ce(ce
, alias
);
653 ce
->ce_flags
|= CE_ADDED
;
655 /* It was suspected to be racily clean, but it turns out to be Ok */
658 !hashcmp(alias
->sha1
, ce
->sha1
) &&
659 ce
->ce_mode
== alias
->ce_mode
);
663 else if (add_index_entry(istate
, ce
, add_option
))
664 return error("unable to add %s to index",path
);
665 if (verbose
&& !was_same
)
666 printf("add '%s'\n", path
);
670 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
673 if (lstat(path
, &st
))
674 die_errno("unable to stat '%s'", path
);
675 return add_to_index(istate
, path
, &st
, flags
);
678 struct cache_entry
*make_cache_entry(unsigned int mode
,
679 const unsigned char *sha1
, const char *path
, int stage
,
683 struct cache_entry
*ce
;
685 if (!verify_path(path
)) {
686 error("Invalid path '%s'", path
);
691 size
= cache_entry_size(len
);
692 ce
= xcalloc(1, size
);
694 hashcpy(ce
->sha1
, sha1
);
695 memcpy(ce
->name
, path
, len
);
696 ce
->ce_flags
= create_ce_flags(len
, stage
);
697 ce
->ce_mode
= create_ce_mode(mode
);
700 return refresh_cache_entry(ce
, 0);
705 int ce_same_name(struct cache_entry
*a
, struct cache_entry
*b
)
707 int len
= ce_namelen(a
);
708 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
711 int ce_path_match(const struct cache_entry
*ce
, const struct pathspec
*pathspec
)
713 return match_pathspec_depth(pathspec
, ce
->name
, ce_namelen(ce
), 0, NULL
);
717 * We fundamentally don't like some paths: we don't want
718 * dot or dot-dot anywhere, and for obvious reasons don't
719 * want to recurse into ".git" either.
721 * Also, we don't want double slashes or slashes at the
722 * end that can make pathnames ambiguous.
724 static int verify_dotfile(const char *rest
)
727 * The first character was '.', but that
728 * has already been discarded, we now test
732 /* "." is not allowed */
733 if (*rest
== '\0' || is_dir_sep(*rest
))
738 * ".git" followed by NUL or slash is bad. This
739 * shares the path end test with the ".." case.
749 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
755 int verify_path(const char *path
)
759 if (has_dos_drive_prefix(path
))
769 if ((c
== '.' && !verify_dotfile(path
)) ||
770 is_dir_sep(c
) || c
== '\0')
778 * Do we have another file that has the beginning components being a
779 * proper superset of the name we're trying to add?
781 static int has_file_name(struct index_state
*istate
,
782 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
785 int len
= ce_namelen(ce
);
786 int stage
= ce_stage(ce
);
787 const char *name
= ce
->name
;
789 while (pos
< istate
->cache_nr
) {
790 struct cache_entry
*p
= istate
->cache
[pos
++];
792 if (len
>= ce_namelen(p
))
794 if (memcmp(name
, p
->name
, len
))
796 if (ce_stage(p
) != stage
)
798 if (p
->name
[len
] != '/')
800 if (p
->ce_flags
& CE_REMOVE
)
805 remove_index_entry_at(istate
, --pos
);
811 * Do we have another file with a pathname that is a proper
812 * subset of the name we're trying to add?
814 static int has_dir_name(struct index_state
*istate
,
815 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
818 int stage
= ce_stage(ce
);
819 const char *name
= ce
->name
;
820 const char *slash
= name
+ ce_namelen(ce
);
828 if (slash
<= ce
->name
)
833 pos
= index_name_pos(istate
, name
, create_ce_flags(len
, stage
));
836 * Found one, but not so fast. This could
837 * be a marker that says "I was here, but
838 * I am being removed". Such an entry is
839 * not a part of the resulting tree, and
840 * it is Ok to have a directory at the same
843 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
847 remove_index_entry_at(istate
, pos
);
855 * Trivial optimization: if we find an entry that
856 * already matches the sub-directory, then we know
857 * we're ok, and we can exit.
859 while (pos
< istate
->cache_nr
) {
860 struct cache_entry
*p
= istate
->cache
[pos
];
861 if ((ce_namelen(p
) <= len
) ||
862 (p
->name
[len
] != '/') ||
863 memcmp(p
->name
, name
, len
))
864 break; /* not our subdirectory */
865 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
867 * p is at the same stage as our entry, and
868 * is a subdirectory of what we are looking
869 * at, so we cannot have conflicts at our
870 * level or anything shorter.
879 /* We may be in a situation where we already have path/file and path
880 * is being added, or we already have path and path/file is being
881 * added. Either one would result in a nonsense tree that has path
882 * twice when git-write-tree tries to write it out. Prevent it.
884 * If ok-to-replace is specified, we remove the conflicting entries
885 * from the cache so the caller should recompute the insert position.
886 * When this happens, we return non-zero.
888 static int check_file_directory_conflict(struct index_state
*istate
,
889 const struct cache_entry
*ce
,
890 int pos
, int ok_to_replace
)
895 * When ce is an "I am going away" entry, we allow it to be added
897 if (ce
->ce_flags
& CE_REMOVE
)
901 * We check if the path is a sub-path of a subsequent pathname
902 * first, since removing those will not change the position
905 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
908 * Then check if the path might have a clashing sub-directory
911 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
914 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
917 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
918 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
919 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
920 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
922 cache_tree_invalidate_path(istate
->cache_tree
, ce
->name
);
923 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
925 /* existing match? Just replace it. */
928 replace_index_entry(istate
, pos
, ce
);
934 * Inserting a merged entry ("stage 0") into the index
935 * will always replace all non-merged entries..
937 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
938 while (ce_same_name(istate
->cache
[pos
], ce
)) {
940 if (!remove_index_entry_at(istate
, pos
))
947 if (!verify_path(ce
->name
))
948 return error("Invalid path '%s'", ce
->name
);
950 if (!skip_df_check
&&
951 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
953 return error("'%s' appears as both a file and as a directory",
955 pos
= index_name_pos(istate
, ce
->name
, ce
->ce_flags
);
961 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
965 if (option
& ADD_CACHE_JUST_APPEND
)
966 pos
= istate
->cache_nr
;
969 ret
= add_index_entry_with_check(istate
, ce
, option
);
975 /* Make sure the array is big enough .. */
976 if (istate
->cache_nr
== istate
->cache_alloc
) {
977 istate
->cache_alloc
= alloc_nr(istate
->cache_alloc
);
978 istate
->cache
= xrealloc(istate
->cache
,
979 istate
->cache_alloc
* sizeof(struct cache_entry
*));
984 if (istate
->cache_nr
> pos
+ 1)
985 memmove(istate
->cache
+ pos
+ 1,
987 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
988 set_index_entry(istate
, pos
, ce
);
989 istate
->cache_changed
= 1;
994 * "refresh" does not calculate a new sha1 file or bring the
995 * cache up-to-date for mode/content changes. But what it
996 * _does_ do is to "re-match" the stat information of a file
997 * with the cache, so that you can refresh the cache for a
998 * file that hasn't been changed but where the stat entry is
1001 * For example, you'd want to do this after doing a "git-read-tree",
1002 * to link up the stat cache details with the proper files.
1004 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1005 struct cache_entry
*ce
,
1006 unsigned int options
, int *err
,
1010 struct cache_entry
*updated
;
1012 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1013 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1015 if (ce_uptodate(ce
))
1019 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1020 * that the change to the work tree does not matter and told
1023 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1024 ce_mark_uptodate(ce
);
1027 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1028 ce_mark_uptodate(ce
);
1032 if (lstat(ce
->name
, &st
) < 0) {
1038 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1040 *changed_ret
= changed
;
1043 * The path is unchanged. If we were told to ignore
1044 * valid bit, then we did the actual stat check and
1045 * found that the entry is unmodified. If the entry
1046 * is not marked VALID, this is the place to mark it
1047 * valid again, under "assume unchanged" mode.
1049 if (ignore_valid
&& assume_unchanged
&&
1050 !(ce
->ce_flags
& CE_VALID
))
1051 ; /* mark this one VALID again */
1054 * We do not mark the index itself "modified"
1055 * because CE_UPTODATE flag is in-core only;
1056 * we are not going to write this change out.
1058 if (!S_ISGITLINK(ce
->ce_mode
))
1059 ce_mark_uptodate(ce
);
1064 if (ie_modified(istate
, ce
, &st
, options
)) {
1071 updated
= xmalloc(size
);
1072 memcpy(updated
, ce
, size
);
1073 fill_stat_cache_info(updated
, &st
);
1075 * If ignore_valid is not set, we should leave CE_VALID bit
1076 * alone. Otherwise, paths marked with --no-assume-unchanged
1077 * (i.e. things to be edited) will reacquire CE_VALID bit
1078 * automatically, which is not really what we want.
1080 if (!ignore_valid
&& assume_unchanged
&&
1081 !(ce
->ce_flags
& CE_VALID
))
1082 updated
->ce_flags
&= ~CE_VALID
;
1087 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1088 int * first
, const char *header_msg
)
1090 if (in_porcelain
&& *first
&& header_msg
) {
1091 printf("%s\n", header_msg
);
1097 int refresh_index(struct index_state
*istate
, unsigned int flags
, const char **pathspec
,
1098 char *seen
, const char *header_msg
)
1102 int really
= (flags
& REFRESH_REALLY
) != 0;
1103 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1104 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1105 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1106 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1108 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1109 unsigned int options
= really
? CE_MATCH_IGNORE_VALID
: 0;
1110 const char *modified_fmt
;
1111 const char *deleted_fmt
;
1112 const char *typechange_fmt
;
1113 const char *added_fmt
;
1114 const char *unmerged_fmt
;
1116 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1117 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1118 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1119 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1120 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1121 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1122 struct cache_entry
*ce
, *new;
1123 int cache_errno
= 0;
1127 ce
= istate
->cache
[i
];
1128 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1132 !match_pathspec(pathspec
, ce
->name
, strlen(ce
->name
), 0, seen
))
1136 while ((i
< istate
->cache_nr
) &&
1137 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1143 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1144 &first
, header_msg
);
1152 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1158 if (not_new
&& cache_errno
== ENOENT
)
1160 if (really
&& cache_errno
== EINVAL
) {
1161 /* If we are doing --really-refresh that
1162 * means the index is not valid anymore.
1164 ce
->ce_flags
&= ~CE_VALID
;
1165 istate
->cache_changed
= 1;
1170 if (cache_errno
== ENOENT
)
1172 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1173 fmt
= added_fmt
; /* must be before other checks */
1174 else if (changed
& TYPE_CHANGED
)
1175 fmt
= typechange_fmt
;
1179 ce
->name
, in_porcelain
, &first
, header_msg
);
1184 replace_index_entry(istate
, i
, new);
1189 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
, int really
)
1191 return refresh_cache_ent(&the_index
, ce
, really
, NULL
, NULL
);
1195 /*****************************************************************
1197 *****************************************************************/
1200 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1201 * Again - this is just a (very strong in practice) heuristic that
1202 * the inode hasn't changed.
1204 * We save the fields in big-endian order to allow using the
1205 * index file over NFS transparently.
1207 struct ondisk_cache_entry
{
1208 struct cache_time ctime
;
1209 struct cache_time mtime
;
1216 unsigned char sha1
[20];
1217 unsigned short flags
;
1218 char name
[FLEX_ARRAY
]; /* more */
1222 * This struct is used when CE_EXTENDED bit is 1
1223 * The struct must match ondisk_cache_entry exactly from
1226 struct ondisk_cache_entry_extended
{
1227 struct cache_time ctime
;
1228 struct cache_time mtime
;
1235 unsigned char sha1
[20];
1236 unsigned short flags
;
1237 unsigned short flags2
;
1238 char name
[FLEX_ARRAY
]; /* more */
1241 /* These are only used for v3 or lower */
1242 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1243 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1244 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1245 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1246 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1247 ondisk_cache_entry_size(ce_namelen(ce)))
1249 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1252 unsigned char sha1
[20];
1255 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1256 return error("bad signature");
1257 hdr_version
= ntohl(hdr
->hdr_version
);
1258 if (hdr_version
< 2 || 4 < hdr_version
)
1259 return error("bad index version %d", hdr_version
);
1261 git_SHA1_Update(&c
, hdr
, size
- 20);
1262 git_SHA1_Final(sha1
, &c
);
1263 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1264 return error("bad index file sha1 signature");
1268 static int read_index_extension(struct index_state
*istate
,
1269 const char *ext
, void *data
, unsigned long sz
)
1271 switch (CACHE_EXT(ext
)) {
1272 case CACHE_EXT_TREE
:
1273 istate
->cache_tree
= cache_tree_read(data
, sz
);
1275 case CACHE_EXT_RESOLVE_UNDO
:
1276 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1279 if (*ext
< 'A' || 'Z' < *ext
)
1280 return error("index uses %.4s extension, which we do not understand",
1282 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1288 int read_index(struct index_state
*istate
)
1290 return read_index_from(istate
, get_index_file());
1293 #ifndef NEEDS_ALIGNED_ACCESS
1294 #define ntoh_s(var) ntohs(var)
1295 #define ntoh_l(var) ntohl(var)
1297 static inline uint16_t ntoh_s_force_align(void *p
)
1300 memcpy(&x
, p
, sizeof(x
));
1303 static inline uint32_t ntoh_l_force_align(void *p
)
1306 memcpy(&x
, p
, sizeof(x
));
1309 #define ntoh_s(var) ntoh_s_force_align(&(var))
1310 #define ntoh_l(var) ntoh_l_force_align(&(var))
1313 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1318 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1320 ce
->ce_ctime
.sec
= ntoh_l(ondisk
->ctime
.sec
);
1321 ce
->ce_mtime
.sec
= ntoh_l(ondisk
->mtime
.sec
);
1322 ce
->ce_ctime
.nsec
= ntoh_l(ondisk
->ctime
.nsec
);
1323 ce
->ce_mtime
.nsec
= ntoh_l(ondisk
->mtime
.nsec
);
1324 ce
->ce_dev
= ntoh_l(ondisk
->dev
);
1325 ce
->ce_ino
= ntoh_l(ondisk
->ino
);
1326 ce
->ce_mode
= ntoh_l(ondisk
->mode
);
1327 ce
->ce_uid
= ntoh_l(ondisk
->uid
);
1328 ce
->ce_gid
= ntoh_l(ondisk
->gid
);
1329 ce
->ce_size
= ntoh_l(ondisk
->size
);
1330 ce
->ce_flags
= flags
;
1331 hashcpy(ce
->sha1
, ondisk
->sha1
);
1332 memcpy(ce
->name
, name
, len
);
1333 ce
->name
[len
] = '\0';
1338 * Adjacent cache entries tend to share the leading paths, so it makes
1339 * sense to only store the differences in later entries. In the v4
1340 * on-disk format of the index, each on-disk cache entry stores the
1341 * number of bytes to be stripped from the end of the previous name,
1342 * and the bytes to append to the result, to come up with its name.
1344 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1346 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1347 size_t len
= decode_varint(&cp
);
1349 if (name
->len
< len
)
1350 die("malformed name field in the index");
1351 strbuf_remove(name
, name
->len
- len
, len
);
1352 for (ep
= cp
; *ep
; ep
++)
1353 ; /* find the end */
1354 strbuf_add(name
, cp
, ep
- cp
);
1355 return (const char *)ep
+ 1 - cp_
;
1358 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1359 unsigned long *ent_size
,
1360 struct strbuf
*previous_name
)
1362 struct cache_entry
*ce
;
1367 /* On-disk flags are just 16 bits */
1368 flags
= ntoh_s(ondisk
->flags
);
1369 len
= flags
& CE_NAMEMASK
;
1371 if (flags
& CE_EXTENDED
) {
1372 struct ondisk_cache_entry_extended
*ondisk2
;
1374 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1375 extended_flags
= ntoh_s(ondisk2
->flags2
) << 16;
1376 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1377 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1378 die("Unknown index entry format %08x", extended_flags
);
1379 flags
|= extended_flags
;
1380 name
= ondisk2
->name
;
1383 name
= ondisk
->name
;
1385 if (!previous_name
) {
1386 /* v3 and earlier */
1387 if (len
== CE_NAMEMASK
)
1389 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1391 *ent_size
= ondisk_ce_size(ce
);
1393 unsigned long consumed
;
1394 consumed
= expand_name_field(previous_name
, name
);
1395 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1397 previous_name
->len
);
1399 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1404 /* remember to discard_cache() before reading a different cache! */
1405 int read_index_from(struct index_state
*istate
, const char *path
)
1409 unsigned long src_offset
;
1410 struct cache_header
*hdr
;
1413 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1416 if (istate
->initialized
)
1417 return istate
->cache_nr
;
1420 istate
->timestamp
.sec
= 0;
1421 istate
->timestamp
.nsec
= 0;
1422 fd
= open(path
, O_RDONLY
);
1424 if (errno
== ENOENT
)
1426 die_errno("index file open failed");
1430 die_errno("cannot stat the open index");
1433 mmap_size
= xsize_t(st
.st_size
);
1434 if (mmap_size
< sizeof(struct cache_header
) + 20)
1435 die("index file smaller than expected");
1437 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1439 if (mmap
== MAP_FAILED
)
1440 die_errno("unable to map index file");
1443 if (verify_hdr(hdr
, mmap_size
) < 0)
1446 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1447 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1448 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(struct cache_entry
*));
1449 istate
->initialized
= 1;
1451 if (hdr
->hdr_version
== htonl(4))
1452 previous_name
= &previous_name_buf
;
1454 previous_name
= NULL
;
1456 src_offset
= sizeof(*hdr
);
1457 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1458 struct ondisk_cache_entry
*disk_ce
;
1459 struct cache_entry
*ce
;
1460 unsigned long consumed
;
1462 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1463 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1464 set_index_entry(istate
, i
, ce
);
1466 src_offset
+= consumed
;
1468 strbuf_release(&previous_name_buf
);
1469 istate
->timestamp
.sec
= st
.st_mtime
;
1470 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1472 while (src_offset
<= mmap_size
- 20 - 8) {
1473 /* After an array of active_nr index entries,
1474 * there can be arbitrary number of extended
1475 * sections, each of which is prefixed with
1476 * extension name (4-byte) and section length
1477 * in 4-byte network byte order.
1480 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1481 extsize
= ntohl(extsize
);
1482 if (read_index_extension(istate
,
1483 (const char *) mmap
+ src_offset
,
1484 (char *) mmap
+ src_offset
+ 8,
1488 src_offset
+= extsize
;
1490 munmap(mmap
, mmap_size
);
1491 return istate
->cache_nr
;
1494 munmap(mmap
, mmap_size
);
1496 die("index file corrupt");
1499 int is_index_unborn(struct index_state
*istate
)
1501 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1504 int discard_index(struct index_state
*istate
)
1508 for (i
= 0; i
< istate
->cache_nr
; i
++)
1509 free(istate
->cache
[i
]);
1510 resolve_undo_clear_index(istate
);
1511 istate
->cache_nr
= 0;
1512 istate
->cache_changed
= 0;
1513 istate
->timestamp
.sec
= 0;
1514 istate
->timestamp
.nsec
= 0;
1515 istate
->name_hash_initialized
= 0;
1516 free_hash(&istate
->name_hash
);
1517 cache_tree_free(&(istate
->cache_tree
));
1518 istate
->initialized
= 0;
1520 /* no need to throw away allocated active_cache */
1524 int unmerged_index(const struct index_state
*istate
)
1527 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1528 if (ce_stage(istate
->cache
[i
]))
1534 #define WRITE_BUFFER_SIZE 8192
1535 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1536 static unsigned long write_buffer_len
;
1538 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1540 unsigned int buffered
= write_buffer_len
;
1542 git_SHA1_Update(context
, write_buffer
, buffered
);
1543 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1545 write_buffer_len
= 0;
1550 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1553 unsigned int buffered
= write_buffer_len
;
1554 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1557 memcpy(write_buffer
+ buffered
, data
, partial
);
1558 buffered
+= partial
;
1559 if (buffered
== WRITE_BUFFER_SIZE
) {
1560 write_buffer_len
= buffered
;
1561 if (ce_write_flush(context
, fd
))
1565 write_buffer_len
= buffered
;
1567 data
= (char *) data
+ partial
;
1572 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1573 unsigned int ext
, unsigned int sz
)
1577 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1578 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1581 static int ce_flush(git_SHA_CTX
*context
, int fd
)
1583 unsigned int left
= write_buffer_len
;
1586 write_buffer_len
= 0;
1587 git_SHA1_Update(context
, write_buffer
, left
);
1590 /* Flush first if not enough space for SHA1 signature */
1591 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1592 if (write_in_full(fd
, write_buffer
, left
) != left
)
1597 /* Append the SHA1 signature at the end */
1598 git_SHA1_Final(write_buffer
+ left
, context
);
1600 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1603 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1606 * The only thing we care about in this function is to smudge the
1607 * falsely clean entry due to touch-update-touch race, so we leave
1608 * everything else as they are. We are called for entries whose
1609 * ce_mtime match the index file mtime.
1611 * Note that this actually does not do much for gitlinks, for
1612 * which ce_match_stat_basic() always goes to the actual
1613 * contents. The caller checks with is_racy_timestamp() which
1614 * always says "no" for gitlinks, so we are not called for them ;-)
1618 if (lstat(ce
->name
, &st
) < 0)
1620 if (ce_match_stat_basic(ce
, &st
))
1622 if (ce_modified_check_fs(ce
, &st
)) {
1623 /* This is "racily clean"; smudge it. Note that this
1624 * is a tricky code. At first glance, it may appear
1625 * that it can break with this sequence:
1627 * $ echo xyzzy >frotz
1628 * $ git-update-index --add frotz
1631 * $ echo filfre >nitfol
1632 * $ git-update-index --add nitfol
1634 * but it does not. When the second update-index runs,
1635 * it notices that the entry "frotz" has the same timestamp
1636 * as index, and if we were to smudge it by resetting its
1637 * size to zero here, then the object name recorded
1638 * in index is the 6-byte file but the cached stat information
1639 * becomes zero --- which would then match what we would
1640 * obtain from the filesystem next time we stat("frotz").
1642 * However, the second update-index, before calling
1643 * this function, notices that the cached size is 6
1644 * bytes and what is on the filesystem is an empty
1645 * file, and never calls us, so the cached size information
1646 * for "frotz" stays 6 which does not match the filesystem.
1652 /* Copy miscellaneous fields but not the name */
1653 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1654 struct cache_entry
*ce
)
1656 ondisk
->ctime
.sec
= htonl(ce
->ce_ctime
.sec
);
1657 ondisk
->mtime
.sec
= htonl(ce
->ce_mtime
.sec
);
1658 ondisk
->ctime
.nsec
= htonl(ce
->ce_ctime
.nsec
);
1659 ondisk
->mtime
.nsec
= htonl(ce
->ce_mtime
.nsec
);
1660 ondisk
->dev
= htonl(ce
->ce_dev
);
1661 ondisk
->ino
= htonl(ce
->ce_ino
);
1662 ondisk
->mode
= htonl(ce
->ce_mode
);
1663 ondisk
->uid
= htonl(ce
->ce_uid
);
1664 ondisk
->gid
= htonl(ce
->ce_gid
);
1665 ondisk
->size
= htonl(ce
->ce_size
);
1666 hashcpy(ondisk
->sha1
, ce
->sha1
);
1667 ondisk
->flags
= htons(ce
->ce_flags
);
1668 if (ce
->ce_flags
& CE_EXTENDED
) {
1669 struct ondisk_cache_entry_extended
*ondisk2
;
1670 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1671 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1672 return ondisk2
->name
;
1675 return ondisk
->name
;
1679 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
)
1681 int size
= ondisk_ce_size(ce
);
1682 struct ondisk_cache_entry
*ondisk
= xcalloc(1, size
);
1686 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1687 memcpy(name
, ce
->name
, ce_namelen(ce
));
1689 result
= ce_write(c
, fd
, ondisk
, size
);
1694 static int has_racy_timestamp(struct index_state
*istate
)
1696 int entries
= istate
->cache_nr
;
1699 for (i
= 0; i
< entries
; i
++) {
1700 struct cache_entry
*ce
= istate
->cache
[i
];
1701 if (is_racy_timestamp(istate
, ce
))
1708 * Opportunisticly update the index but do not complain if we can't
1710 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1712 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1713 !write_index(istate
, lockfile
->fd
))
1714 commit_locked_index(lockfile
);
1716 rollback_lock_file(lockfile
);
1719 int write_index(struct index_state
*istate
, int newfd
)
1722 struct cache_header hdr
;
1723 int i
, err
, removed
, extended
;
1724 struct cache_entry
**cache
= istate
->cache
;
1725 int entries
= istate
->cache_nr
;
1728 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1729 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1732 /* reduce extended entries if possible */
1733 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1734 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1736 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1740 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1741 /* for extended format, increase version so older git won't try to read it */
1742 hdr
.hdr_version
= htonl(extended
? 3 : 2);
1743 hdr
.hdr_entries
= htonl(entries
- removed
);
1746 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1749 for (i
= 0; i
< entries
; i
++) {
1750 struct cache_entry
*ce
= cache
[i
];
1751 if (ce
->ce_flags
& CE_REMOVE
)
1753 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
1754 ce_smudge_racily_clean_entry(ce
);
1755 if (ce_write_entry(&c
, newfd
, ce
) < 0)
1759 /* Write extension data here */
1760 if (istate
->cache_tree
) {
1761 struct strbuf sb
= STRBUF_INIT
;
1763 cache_tree_write(&sb
, istate
->cache_tree
);
1764 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1765 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1766 strbuf_release(&sb
);
1770 if (istate
->resolve_undo
) {
1771 struct strbuf sb
= STRBUF_INIT
;
1773 resolve_undo_write(&sb
, istate
->resolve_undo
);
1774 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
1776 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1777 strbuf_release(&sb
);
1782 if (ce_flush(&c
, newfd
) || fstat(newfd
, &st
))
1784 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
1785 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1790 * Read the index file that is potentially unmerged into given
1791 * index_state, dropping any unmerged entries. Returns true if
1792 * the index is unmerged. Callers who want to refuse to work
1793 * from an unmerged state can call this and check its return value,
1794 * instead of calling read_cache().
1796 int read_index_unmerged(struct index_state
*istate
)
1802 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1803 struct cache_entry
*ce
= istate
->cache
[i
];
1804 struct cache_entry
*new_ce
;
1810 len
= strlen(ce
->name
);
1811 size
= cache_entry_size(len
);
1812 new_ce
= xcalloc(1, size
);
1813 memcpy(new_ce
->name
, ce
->name
, len
);
1814 new_ce
->ce_flags
= create_ce_flags(len
, 0) | CE_CONFLICTED
;
1815 new_ce
->ce_mode
= ce
->ce_mode
;
1816 if (add_index_entry(istate
, new_ce
, 0))
1817 return error("%s: cannot drop to stage #0",
1819 i
= index_name_pos(istate
, new_ce
->name
, len
);
1825 * Returns 1 if the path is an "other" path with respect to
1826 * the index; that is, the path is not mentioned in the index at all,
1827 * either as a file, a directory with some files in the index,
1828 * or as an unmerged entry.
1830 * We helpfully remove a trailing "/" from directories so that
1831 * the output of read_directory can be used as-is.
1833 int index_name_is_other(const struct index_state
*istate
, const char *name
,
1837 if (namelen
&& name
[namelen
- 1] == '/')
1839 pos
= index_name_pos(istate
, name
, namelen
);
1841 return 0; /* exact match */
1843 if (pos
< istate
->cache_nr
) {
1844 struct cache_entry
*ce
= istate
->cache
[pos
];
1845 if (ce_namelen(ce
) == namelen
&&
1846 !memcmp(ce
->name
, name
, namelen
))
1847 return 0; /* Yup, this one exists unmerged */