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"
20 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
21 unsigned int options
);
23 /* Mask for the name length in ce_flags in the on-disk index */
25 #define CE_NAMEMASK (0x0fff)
29 * The first letter should be 'A'..'Z' for extensions that are not
30 * necessary for a correct operation (i.e. optimization data).
31 * When new extensions are added that _needs_ to be understood in
32 * order to correctly interpret the index file, pick character that
33 * is outside the range, to cause the reader to abort.
36 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
37 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
38 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
39 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
41 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
42 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
46 struct index_state the_index
;
47 static const char *alternate_index_output
;
49 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
51 istate
->cache
[nr
] = ce
;
52 add_name_hash(istate
, ce
);
55 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
57 struct cache_entry
*old
= istate
->cache
[nr
];
59 replace_index_entry_in_base(istate
, old
, ce
);
60 remove_name_hash(istate
, old
);
62 set_index_entry(istate
, nr
, ce
);
63 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
64 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
67 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
69 struct cache_entry
*old
= istate
->cache
[nr
], *new;
70 int namelen
= strlen(new_name
);
72 new = xmalloc(cache_entry_size(namelen
));
73 copy_cache_entry(new, old
);
74 new->ce_flags
&= ~CE_HASHED
;
75 new->ce_namelen
= namelen
;
77 memcpy(new->name
, new_name
, namelen
+ 1);
79 cache_tree_invalidate_path(istate
, old
->name
);
80 remove_index_entry_at(istate
, nr
);
81 add_index_entry(istate
, new, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
84 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
86 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
87 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
88 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
89 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
90 sd
->sd_dev
= st
->st_dev
;
91 sd
->sd_ino
= st
->st_ino
;
92 sd
->sd_uid
= st
->st_uid
;
93 sd
->sd_gid
= st
->st_gid
;
94 sd
->sd_size
= st
->st_size
;
97 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
101 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
102 changed
|= MTIME_CHANGED
;
103 if (trust_ctime
&& check_stat
&&
104 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
105 changed
|= CTIME_CHANGED
;
108 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
109 changed
|= MTIME_CHANGED
;
110 if (trust_ctime
&& check_stat
&&
111 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
112 changed
|= CTIME_CHANGED
;
116 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
117 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
118 changed
|= OWNER_CHANGED
;
119 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
120 changed
|= INODE_CHANGED
;
125 * st_dev breaks on network filesystems where different
126 * clients will have different views of what "device"
127 * the filesystem is on
129 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
130 changed
|= INODE_CHANGED
;
133 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
134 changed
|= DATA_CHANGED
;
140 * This only updates the "non-critical" parts of the directory
141 * cache, ie the parts that aren't tracked by GIT, and only used
142 * to validate the cache.
144 void fill_stat_cache_info(struct cache_entry
*ce
, struct stat
*st
)
146 fill_stat_data(&ce
->ce_stat_data
, st
);
148 if (assume_unchanged
)
149 ce
->ce_flags
|= CE_VALID
;
151 if (S_ISREG(st
->st_mode
))
152 ce_mark_uptodate(ce
);
155 static int ce_compare_data(const struct cache_entry
*ce
, struct stat
*st
)
158 int fd
= open(ce
->name
, O_RDONLY
);
161 unsigned char sha1
[20];
162 if (!index_fd(sha1
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
163 match
= hashcmp(sha1
, ce
->sha1
);
164 /* index_fd() closed the file descriptor already */
169 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
174 enum object_type type
;
175 struct strbuf sb
= STRBUF_INIT
;
177 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
180 buffer
= read_sha1_file(ce
->sha1
, &type
, &size
);
183 match
= memcmp(buffer
, sb
.buf
, size
);
190 static int ce_compare_gitlink(const struct cache_entry
*ce
)
192 unsigned char sha1
[20];
195 * We don't actually require that the .git directory
196 * under GITLINK directory be a valid git directory. It
197 * might even be missing (in case nobody populated that
200 * If so, we consider it always to match.
202 if (resolve_gitlink_ref(ce
->name
, "HEAD", sha1
) < 0)
204 return hashcmp(sha1
, ce
->sha1
);
207 static int ce_modified_check_fs(const struct cache_entry
*ce
, struct stat
*st
)
209 switch (st
->st_mode
& S_IFMT
) {
211 if (ce_compare_data(ce
, st
))
215 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
219 if (S_ISGITLINK(ce
->ce_mode
))
220 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
227 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
229 unsigned int changed
= 0;
231 if (ce
->ce_flags
& CE_REMOVE
)
232 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
234 switch (ce
->ce_mode
& S_IFMT
) {
236 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
237 /* We consider only the owner x bit to be relevant for
240 if (trust_executable_bit
&&
241 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
242 changed
|= MODE_CHANGED
;
245 if (!S_ISLNK(st
->st_mode
) &&
246 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
247 changed
|= TYPE_CHANGED
;
250 /* We ignore most of the st_xxx fields for gitlinks */
251 if (!S_ISDIR(st
->st_mode
))
252 changed
|= TYPE_CHANGED
;
253 else if (ce_compare_gitlink(ce
))
254 changed
|= DATA_CHANGED
;
257 die("internal error: ce_mode is %o", ce
->ce_mode
);
260 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
262 /* Racily smudged entry? */
263 if (!ce
->ce_stat_data
.sd_size
) {
264 if (!is_empty_blob_sha1(ce
->sha1
))
265 changed
|= DATA_CHANGED
;
271 static int is_racy_timestamp(const struct index_state
*istate
,
272 const struct cache_entry
*ce
)
274 return (!S_ISGITLINK(ce
->ce_mode
) &&
275 istate
->timestamp
.sec
&&
277 /* nanosecond timestamped files can also be racy! */
278 (istate
->timestamp
.sec
< ce
->ce_stat_data
.sd_mtime
.sec
||
279 (istate
->timestamp
.sec
== ce
->ce_stat_data
.sd_mtime
.sec
&&
280 istate
->timestamp
.nsec
<= ce
->ce_stat_data
.sd_mtime
.nsec
))
282 istate
->timestamp
.sec
<= ce
->ce_stat_data
.sd_mtime
.sec
287 int ie_match_stat(const struct index_state
*istate
,
288 const struct cache_entry
*ce
, struct stat
*st
,
289 unsigned int options
)
291 unsigned int changed
;
292 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
293 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
294 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
297 * If it's marked as always valid in the index, it's
298 * valid whatever the checked-out copy says.
300 * skip-worktree has the same effect with higher precedence
302 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
304 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
308 * Intent-to-add entries have not been added, so the index entry
309 * by definition never matches what is in the work tree until it
310 * actually gets added.
312 if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
313 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
315 changed
= ce_match_stat_basic(ce
, st
);
318 * Within 1 second of this sequence:
319 * echo xyzzy >file && git-update-index --add file
320 * running this command:
322 * would give a falsely clean cache entry. The mtime and
323 * length match the cache, and other stat fields do not change.
325 * We could detect this at update-index time (the cache entry
326 * being registered/updated records the same time as "now")
327 * and delay the return from git-update-index, but that would
328 * effectively mean we can make at most one commit per second,
329 * which is not acceptable. Instead, we check cache entries
330 * whose mtime are the same as the index file timestamp more
331 * carefully than others.
333 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
334 if (assume_racy_is_modified
)
335 changed
|= DATA_CHANGED
;
337 changed
|= ce_modified_check_fs(ce
, st
);
343 int ie_modified(const struct index_state
*istate
,
344 const struct cache_entry
*ce
,
345 struct stat
*st
, unsigned int options
)
347 int changed
, changed_fs
;
349 changed
= ie_match_stat(istate
, ce
, st
, options
);
353 * If the mode or type has changed, there's no point in trying
354 * to refresh the entry - it's not going to match
356 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
360 * Immediately after read-tree or update-index --cacheinfo,
361 * the length field is zero, as we have never even read the
362 * lstat(2) information once, and we cannot trust DATA_CHANGED
363 * returned by ie_match_stat() which in turn was returned by
364 * ce_match_stat_basic() to signal that the filesize of the
365 * blob changed. We have to actually go to the filesystem to
366 * see if the contents match, and if so, should answer "unchanged".
368 * The logic does not apply to gitlinks, as ce_match_stat_basic()
369 * already has checked the actual HEAD from the filesystem in the
370 * subproject. If ie_match_stat() already said it is different,
371 * then we know it is.
373 if ((changed
& DATA_CHANGED
) &&
374 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
377 changed_fs
= ce_modified_check_fs(ce
, st
);
379 return changed
| changed_fs
;
383 int base_name_compare(const char *name1
, int len1
, int mode1
,
384 const char *name2
, int len2
, int mode2
)
386 unsigned char c1
, c2
;
387 int len
= len1
< len2
? len1
: len2
;
390 cmp
= memcmp(name1
, name2
, len
);
395 if (!c1
&& S_ISDIR(mode1
))
397 if (!c2
&& S_ISDIR(mode2
))
399 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
403 * df_name_compare() is identical to base_name_compare(), except it
404 * compares conflicting directory/file entries as equal. Note that
405 * while a directory name compares as equal to a regular file, they
406 * then individually compare _differently_ to a filename that has
407 * a dot after the basename (because '\0' < '.' < '/').
409 * This is used by routines that want to traverse the git namespace
410 * but then handle conflicting entries together when possible.
412 int df_name_compare(const char *name1
, int len1
, int mode1
,
413 const char *name2
, int len2
, int mode2
)
415 int len
= len1
< len2
? len1
: len2
, cmp
;
416 unsigned char c1
, c2
;
418 cmp
= memcmp(name1
, name2
, len
);
421 /* Directories and files compare equal (same length, same name) */
425 if (!c1
&& S_ISDIR(mode1
))
428 if (!c2
&& S_ISDIR(mode2
))
430 if (c1
== '/' && !c2
)
432 if (c2
== '/' && !c1
)
437 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
439 int len
= len1
< len2
? len1
: len2
;
442 cmp
= memcmp(name1
, name2
, len
);
457 int cache_name_compare(const char *name1
, int len1
, const char *name2
, int len2
)
459 return cache_name_stage_compare(name1
, len1
, 0, name2
, len2
, 0);
462 static int index_name_stage_pos(const struct index_state
*istate
, const char *name
, int namelen
, int stage
)
467 last
= istate
->cache_nr
;
468 while (last
> first
) {
469 int next
= (last
+ first
) >> 1;
470 struct cache_entry
*ce
= istate
->cache
[next
];
471 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
483 int index_name_pos(const struct index_state
*istate
, const char *name
, int namelen
)
485 return index_name_stage_pos(istate
, name
, namelen
, 0);
488 /* Remove entry, return true if there are more entries to go.. */
489 int remove_index_entry_at(struct index_state
*istate
, int pos
)
491 struct cache_entry
*ce
= istate
->cache
[pos
];
493 record_resolve_undo(istate
, ce
);
494 remove_name_hash(istate
, ce
);
495 save_or_free_index_entry(istate
, ce
);
496 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
498 if (pos
>= istate
->cache_nr
)
500 memmove(istate
->cache
+ pos
,
501 istate
->cache
+ pos
+ 1,
502 (istate
->cache_nr
- pos
) * sizeof(struct cache_entry
*));
507 * Remove all cache entries marked for removal, that is where
508 * CE_REMOVE is set in ce_flags. This is much more effective than
509 * calling remove_index_entry_at() for each entry to be removed.
511 void remove_marked_cache_entries(struct index_state
*istate
)
513 struct cache_entry
**ce_array
= istate
->cache
;
516 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
517 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
518 remove_name_hash(istate
, ce_array
[i
]);
519 save_or_free_index_entry(istate
, ce_array
[i
]);
522 ce_array
[j
++] = ce_array
[i
];
524 if (j
== istate
->cache_nr
)
526 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
527 istate
->cache_nr
= j
;
530 int remove_file_from_index(struct index_state
*istate
, const char *path
)
532 int pos
= index_name_pos(istate
, path
, strlen(path
));
535 cache_tree_invalidate_path(istate
, path
);
536 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
537 remove_index_entry_at(istate
, pos
);
541 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
543 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
546 static int index_name_pos_also_unmerged(struct index_state
*istate
,
547 const char *path
, int namelen
)
549 int pos
= index_name_pos(istate
, path
, namelen
);
550 struct cache_entry
*ce
;
555 /* maybe unmerged? */
557 if (pos
>= istate
->cache_nr
||
558 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
561 /* order of preference: stage 2, 1, 3 */
562 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
563 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
564 !compare_name(ce
, path
, namelen
))
569 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
571 int len
= ce_namelen(ce
);
572 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
576 * If we add a filename that aliases in the cache, we will use the
577 * name that we already have - but we don't want to update the same
578 * alias twice, because that implies that there were actually two
579 * different files with aliasing names!
581 * So we use the CE_ADDED flag to verify that the alias was an old
582 * one before we accept it as
584 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
585 struct cache_entry
*ce
,
586 struct cache_entry
*alias
)
589 struct cache_entry
*new;
591 if (alias
->ce_flags
& CE_ADDED
)
592 die("Will not add file alias '%s' ('%s' already exists in index)", ce
->name
, alias
->name
);
594 /* Ok, create the new entry using the name of the existing alias */
595 len
= ce_namelen(alias
);
596 new = xcalloc(1, cache_entry_size(len
));
597 memcpy(new->name
, alias
->name
, len
);
598 copy_cache_entry(new, ce
);
599 save_or_free_index_entry(istate
, ce
);
603 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
605 unsigned char sha1
[20];
606 if (write_sha1_file("", 0, blob_type
, sha1
))
607 die("cannot create an empty blob in the object database");
608 hashcpy(ce
->sha1
, sha1
);
611 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
613 int size
, namelen
, was_same
;
614 mode_t st_mode
= st
->st_mode
;
615 struct cache_entry
*ce
, *alias
;
616 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
617 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
618 int pretend
= flags
& ADD_CACHE_PRETEND
;
619 int intent_only
= flags
& ADD_CACHE_INTENT
;
620 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
621 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
623 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
624 return error("%s: can only add regular files, symbolic links or git-directories", path
);
626 namelen
= strlen(path
);
627 if (S_ISDIR(st_mode
)) {
628 while (namelen
&& path
[namelen
-1] == '/')
631 size
= cache_entry_size(namelen
);
632 ce
= xcalloc(1, size
);
633 memcpy(ce
->name
, path
, namelen
);
634 ce
->ce_namelen
= namelen
;
636 fill_stat_cache_info(ce
, st
);
638 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
640 if (trust_executable_bit
&& has_symlinks
)
641 ce
->ce_mode
= create_ce_mode(st_mode
);
643 /* If there is an existing entry, pick the mode bits and type
644 * from it, otherwise assume unexecutable regular file.
646 struct cache_entry
*ent
;
647 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
649 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
650 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
653 /* When core.ignorecase=true, determine if a directory of the same name but differing
654 * case already exists within the Git repository. If it does, ensure the directory
655 * case of the file being added to the repository matches (is folded into) the existing
656 * entry's directory case.
659 const char *startPtr
= ce
->name
;
660 const char *ptr
= startPtr
;
662 while (*ptr
&& *ptr
!= '/')
665 struct cache_entry
*foundce
;
667 foundce
= index_dir_exists(istate
, ce
->name
, ptr
- ce
->name
- 1);
669 memcpy((void *)startPtr
, foundce
->name
+ (startPtr
- ce
->name
), ptr
- startPtr
);
676 alias
= index_file_exists(istate
, ce
->name
, ce_namelen(ce
), ignore_case
);
677 if (alias
&& !ce_stage(alias
) && !ie_match_stat(istate
, alias
, st
, ce_option
)) {
678 /* Nothing changed, really */
680 if (!S_ISGITLINK(alias
->ce_mode
))
681 ce_mark_uptodate(alias
);
682 alias
->ce_flags
|= CE_ADDED
;
686 if (index_path(ce
->sha1
, path
, st
, HASH_WRITE_OBJECT
))
687 return error("unable to index file %s", path
);
689 set_object_name_for_intent_to_add_entry(ce
);
691 if (ignore_case
&& alias
&& different_name(ce
, alias
))
692 ce
= create_alias_ce(istate
, ce
, alias
);
693 ce
->ce_flags
|= CE_ADDED
;
695 /* It was suspected to be racily clean, but it turns out to be Ok */
698 !hashcmp(alias
->sha1
, ce
->sha1
) &&
699 ce
->ce_mode
== alias
->ce_mode
);
703 else if (add_index_entry(istate
, ce
, add_option
))
704 return error("unable to add %s to index",path
);
705 if (verbose
&& !was_same
)
706 printf("add '%s'\n", path
);
710 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
713 if (lstat(path
, &st
))
714 die_errno("unable to stat '%s'", path
);
715 return add_to_index(istate
, path
, &st
, flags
);
718 struct cache_entry
*make_cache_entry(unsigned int mode
,
719 const unsigned char *sha1
, const char *path
, int stage
,
720 unsigned int refresh_options
)
723 struct cache_entry
*ce
;
725 if (!verify_path(path
)) {
726 error("Invalid path '%s'", path
);
731 size
= cache_entry_size(len
);
732 ce
= xcalloc(1, size
);
734 hashcpy(ce
->sha1
, sha1
);
735 memcpy(ce
->name
, path
, len
);
736 ce
->ce_flags
= create_ce_flags(stage
);
737 ce
->ce_namelen
= len
;
738 ce
->ce_mode
= create_ce_mode(mode
);
740 return refresh_cache_entry(ce
, refresh_options
);
743 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
745 int len
= ce_namelen(a
);
746 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
750 * We fundamentally don't like some paths: we don't want
751 * dot or dot-dot anywhere, and for obvious reasons don't
752 * want to recurse into ".git" either.
754 * Also, we don't want double slashes or slashes at the
755 * end that can make pathnames ambiguous.
757 static int verify_dotfile(const char *rest
)
760 * The first character was '.', but that
761 * has already been discarded, we now test
765 /* "." is not allowed */
766 if (*rest
== '\0' || is_dir_sep(*rest
))
771 * ".git" followed by NUL or slash is bad. This
772 * shares the path end test with the ".." case.
782 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
788 int verify_path(const char *path
)
792 if (has_dos_drive_prefix(path
))
802 if ((c
== '.' && !verify_dotfile(path
)) ||
803 is_dir_sep(c
) || c
== '\0')
811 * Do we have another file that has the beginning components being a
812 * proper superset of the name we're trying to add?
814 static int has_file_name(struct index_state
*istate
,
815 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
818 int len
= ce_namelen(ce
);
819 int stage
= ce_stage(ce
);
820 const char *name
= ce
->name
;
822 while (pos
< istate
->cache_nr
) {
823 struct cache_entry
*p
= istate
->cache
[pos
++];
825 if (len
>= ce_namelen(p
))
827 if (memcmp(name
, p
->name
, len
))
829 if (ce_stage(p
) != stage
)
831 if (p
->name
[len
] != '/')
833 if (p
->ce_flags
& CE_REMOVE
)
838 remove_index_entry_at(istate
, --pos
);
844 * Do we have another file with a pathname that is a proper
845 * subset of the name we're trying to add?
847 static int has_dir_name(struct index_state
*istate
,
848 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
851 int stage
= ce_stage(ce
);
852 const char *name
= ce
->name
;
853 const char *slash
= name
+ ce_namelen(ce
);
861 if (slash
<= ce
->name
)
866 pos
= index_name_stage_pos(istate
, name
, len
, stage
);
869 * Found one, but not so fast. This could
870 * be a marker that says "I was here, but
871 * I am being removed". Such an entry is
872 * not a part of the resulting tree, and
873 * it is Ok to have a directory at the same
876 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
880 remove_index_entry_at(istate
, pos
);
888 * Trivial optimization: if we find an entry that
889 * already matches the sub-directory, then we know
890 * we're ok, and we can exit.
892 while (pos
< istate
->cache_nr
) {
893 struct cache_entry
*p
= istate
->cache
[pos
];
894 if ((ce_namelen(p
) <= len
) ||
895 (p
->name
[len
] != '/') ||
896 memcmp(p
->name
, name
, len
))
897 break; /* not our subdirectory */
898 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
900 * p is at the same stage as our entry, and
901 * is a subdirectory of what we are looking
902 * at, so we cannot have conflicts at our
903 * level or anything shorter.
912 /* We may be in a situation where we already have path/file and path
913 * is being added, or we already have path and path/file is being
914 * added. Either one would result in a nonsense tree that has path
915 * twice when git-write-tree tries to write it out. Prevent it.
917 * If ok-to-replace is specified, we remove the conflicting entries
918 * from the cache so the caller should recompute the insert position.
919 * When this happens, we return non-zero.
921 static int check_file_directory_conflict(struct index_state
*istate
,
922 const struct cache_entry
*ce
,
923 int pos
, int ok_to_replace
)
928 * When ce is an "I am going away" entry, we allow it to be added
930 if (ce
->ce_flags
& CE_REMOVE
)
934 * We check if the path is a sub-path of a subsequent pathname
935 * first, since removing those will not change the position
938 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
941 * Then check if the path might have a clashing sub-directory
944 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
947 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
950 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
951 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
952 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
953 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
955 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
956 cache_tree_invalidate_path(istate
, ce
->name
);
957 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
959 /* existing match? Just replace it. */
962 replace_index_entry(istate
, pos
, ce
);
968 * Inserting a merged entry ("stage 0") into the index
969 * will always replace all non-merged entries..
971 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
972 while (ce_same_name(istate
->cache
[pos
], ce
)) {
974 if (!remove_index_entry_at(istate
, pos
))
981 if (!verify_path(ce
->name
))
982 return error("Invalid path '%s'", ce
->name
);
984 if (!skip_df_check
&&
985 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
987 return error("'%s' appears as both a file and as a directory",
989 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
995 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
999 if (option
& ADD_CACHE_JUST_APPEND
)
1000 pos
= istate
->cache_nr
;
1003 ret
= add_index_entry_with_check(istate
, ce
, option
);
1009 /* Make sure the array is big enough .. */
1010 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1014 if (istate
->cache_nr
> pos
+ 1)
1015 memmove(istate
->cache
+ pos
+ 1,
1016 istate
->cache
+ pos
,
1017 (istate
->cache_nr
- pos
- 1) * sizeof(ce
));
1018 set_index_entry(istate
, pos
, ce
);
1019 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1024 * "refresh" does not calculate a new sha1 file or bring the
1025 * cache up-to-date for mode/content changes. But what it
1026 * _does_ do is to "re-match" the stat information of a file
1027 * with the cache, so that you can refresh the cache for a
1028 * file that hasn't been changed but where the stat entry is
1031 * For example, you'd want to do this after doing a "git-read-tree",
1032 * to link up the stat cache details with the proper files.
1034 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1035 struct cache_entry
*ce
,
1036 unsigned int options
, int *err
,
1040 struct cache_entry
*updated
;
1042 int refresh
= options
& CE_MATCH_REFRESH
;
1043 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1044 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1045 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1047 if (!refresh
|| ce_uptodate(ce
))
1051 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1052 * that the change to the work tree does not matter and told
1055 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1056 ce_mark_uptodate(ce
);
1059 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1060 ce_mark_uptodate(ce
);
1064 if (lstat(ce
->name
, &st
) < 0) {
1065 if (ignore_missing
&& errno
== ENOENT
)
1072 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1074 *changed_ret
= changed
;
1077 * The path is unchanged. If we were told to ignore
1078 * valid bit, then we did the actual stat check and
1079 * found that the entry is unmodified. If the entry
1080 * is not marked VALID, this is the place to mark it
1081 * valid again, under "assume unchanged" mode.
1083 if (ignore_valid
&& assume_unchanged
&&
1084 !(ce
->ce_flags
& CE_VALID
))
1085 ; /* mark this one VALID again */
1088 * We do not mark the index itself "modified"
1089 * because CE_UPTODATE flag is in-core only;
1090 * we are not going to write this change out.
1092 if (!S_ISGITLINK(ce
->ce_mode
))
1093 ce_mark_uptodate(ce
);
1098 if (ie_modified(istate
, ce
, &st
, options
)) {
1105 updated
= xmalloc(size
);
1106 memcpy(updated
, ce
, size
);
1107 fill_stat_cache_info(updated
, &st
);
1109 * If ignore_valid is not set, we should leave CE_VALID bit
1110 * alone. Otherwise, paths marked with --no-assume-unchanged
1111 * (i.e. things to be edited) will reacquire CE_VALID bit
1112 * automatically, which is not really what we want.
1114 if (!ignore_valid
&& assume_unchanged
&&
1115 !(ce
->ce_flags
& CE_VALID
))
1116 updated
->ce_flags
&= ~CE_VALID
;
1118 /* istate->cache_changed is updated in the caller */
1122 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1123 int * first
, const char *header_msg
)
1125 if (in_porcelain
&& *first
&& header_msg
) {
1126 printf("%s\n", header_msg
);
1132 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1133 const struct pathspec
*pathspec
,
1134 char *seen
, const char *header_msg
)
1138 int really
= (flags
& REFRESH_REALLY
) != 0;
1139 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1140 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1141 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1142 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1144 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1145 unsigned int options
= (CE_MATCH_REFRESH
|
1146 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1147 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1148 const char *modified_fmt
;
1149 const char *deleted_fmt
;
1150 const char *typechange_fmt
;
1151 const char *added_fmt
;
1152 const char *unmerged_fmt
;
1154 modified_fmt
= (in_porcelain
? "M\t%s\n" : "%s: needs update\n");
1155 deleted_fmt
= (in_porcelain
? "D\t%s\n" : "%s: needs update\n");
1156 typechange_fmt
= (in_porcelain
? "T\t%s\n" : "%s needs update\n");
1157 added_fmt
= (in_porcelain
? "A\t%s\n" : "%s needs update\n");
1158 unmerged_fmt
= (in_porcelain
? "U\t%s\n" : "%s: needs merge\n");
1159 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1160 struct cache_entry
*ce
, *new;
1161 int cache_errno
= 0;
1165 ce
= istate
->cache
[i
];
1166 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1169 if (pathspec
&& !ce_path_match(ce
, pathspec
, seen
))
1173 while ((i
< istate
->cache_nr
) &&
1174 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1180 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1181 &first
, header_msg
);
1189 new = refresh_cache_ent(istate
, ce
, options
, &cache_errno
, &changed
);
1195 if (really
&& cache_errno
== EINVAL
) {
1196 /* If we are doing --really-refresh that
1197 * means the index is not valid anymore.
1199 ce
->ce_flags
&= ~CE_VALID
;
1200 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1201 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1206 if (cache_errno
== ENOENT
)
1208 else if (ce
->ce_flags
& CE_INTENT_TO_ADD
)
1209 fmt
= added_fmt
; /* must be before other checks */
1210 else if (changed
& TYPE_CHANGED
)
1211 fmt
= typechange_fmt
;
1215 ce
->name
, in_porcelain
, &first
, header_msg
);
1220 replace_index_entry(istate
, i
, new);
1225 static struct cache_entry
*refresh_cache_entry(struct cache_entry
*ce
,
1226 unsigned int options
)
1228 return refresh_cache_ent(&the_index
, ce
, options
, NULL
, NULL
);
1232 /*****************************************************************
1234 *****************************************************************/
1236 #define INDEX_FORMAT_DEFAULT 3
1238 static int index_format_config(const char *var
, const char *value
, void *cb
)
1240 unsigned int *version
= cb
;
1241 if (!strcmp(var
, "index.version")) {
1242 *version
= git_config_int(var
, value
);
1248 static unsigned int get_index_format_default(void)
1250 char *envversion
= getenv("GIT_INDEX_VERSION");
1252 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1255 git_config(index_format_config
, &version
);
1256 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1257 warning(_("index.version set, but the value is invalid.\n"
1258 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1259 return INDEX_FORMAT_DEFAULT
;
1264 version
= strtoul(envversion
, &endp
, 10);
1266 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1267 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1268 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1269 version
= INDEX_FORMAT_DEFAULT
;
1275 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1276 * Again - this is just a (very strong in practice) heuristic that
1277 * the inode hasn't changed.
1279 * We save the fields in big-endian order to allow using the
1280 * index file over NFS transparently.
1282 struct ondisk_cache_entry
{
1283 struct cache_time ctime
;
1284 struct cache_time mtime
;
1291 unsigned char sha1
[20];
1293 char name
[FLEX_ARRAY
]; /* more */
1297 * This struct is used when CE_EXTENDED bit is 1
1298 * The struct must match ondisk_cache_entry exactly from
1301 struct ondisk_cache_entry_extended
{
1302 struct cache_time ctime
;
1303 struct cache_time mtime
;
1310 unsigned char sha1
[20];
1313 char name
[FLEX_ARRAY
]; /* more */
1316 /* These are only used for v3 or lower */
1317 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1318 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1319 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1320 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1321 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1322 ondisk_cache_entry_size(ce_namelen(ce)))
1324 static int verify_hdr(struct cache_header
*hdr
, unsigned long size
)
1327 unsigned char sha1
[20];
1330 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1331 return error("bad signature");
1332 hdr_version
= ntohl(hdr
->hdr_version
);
1333 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1334 return error("bad index version %d", hdr_version
);
1336 git_SHA1_Update(&c
, hdr
, size
- 20);
1337 git_SHA1_Final(sha1
, &c
);
1338 if (hashcmp(sha1
, (unsigned char *)hdr
+ size
- 20))
1339 return error("bad index file sha1 signature");
1343 static int read_index_extension(struct index_state
*istate
,
1344 const char *ext
, void *data
, unsigned long sz
)
1346 switch (CACHE_EXT(ext
)) {
1347 case CACHE_EXT_TREE
:
1348 istate
->cache_tree
= cache_tree_read(data
, sz
);
1350 case CACHE_EXT_RESOLVE_UNDO
:
1351 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1353 case CACHE_EXT_LINK
:
1354 if (read_link_extension(istate
, data
, sz
))
1358 if (*ext
< 'A' || 'Z' < *ext
)
1359 return error("index uses %.4s extension, which we do not understand",
1361 fprintf(stderr
, "ignoring %.4s extension\n", ext
);
1367 int read_index(struct index_state
*istate
)
1369 return read_index_from(istate
, get_index_file());
1372 static struct cache_entry
*cache_entry_from_ondisk(struct ondisk_cache_entry
*ondisk
,
1377 struct cache_entry
*ce
= xmalloc(cache_entry_size(len
));
1379 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(&ondisk
->ctime
.sec
);
1380 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(&ondisk
->mtime
.sec
);
1381 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(&ondisk
->ctime
.nsec
);
1382 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(&ondisk
->mtime
.nsec
);
1383 ce
->ce_stat_data
.sd_dev
= get_be32(&ondisk
->dev
);
1384 ce
->ce_stat_data
.sd_ino
= get_be32(&ondisk
->ino
);
1385 ce
->ce_mode
= get_be32(&ondisk
->mode
);
1386 ce
->ce_stat_data
.sd_uid
= get_be32(&ondisk
->uid
);
1387 ce
->ce_stat_data
.sd_gid
= get_be32(&ondisk
->gid
);
1388 ce
->ce_stat_data
.sd_size
= get_be32(&ondisk
->size
);
1389 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1390 ce
->ce_namelen
= len
;
1392 hashcpy(ce
->sha1
, ondisk
->sha1
);
1393 memcpy(ce
->name
, name
, len
);
1394 ce
->name
[len
] = '\0';
1399 * Adjacent cache entries tend to share the leading paths, so it makes
1400 * sense to only store the differences in later entries. In the v4
1401 * on-disk format of the index, each on-disk cache entry stores the
1402 * number of bytes to be stripped from the end of the previous name,
1403 * and the bytes to append to the result, to come up with its name.
1405 static unsigned long expand_name_field(struct strbuf
*name
, const char *cp_
)
1407 const unsigned char *ep
, *cp
= (const unsigned char *)cp_
;
1408 size_t len
= decode_varint(&cp
);
1410 if (name
->len
< len
)
1411 die("malformed name field in the index");
1412 strbuf_remove(name
, name
->len
- len
, len
);
1413 for (ep
= cp
; *ep
; ep
++)
1414 ; /* find the end */
1415 strbuf_add(name
, cp
, ep
- cp
);
1416 return (const char *)ep
+ 1 - cp_
;
1419 static struct cache_entry
*create_from_disk(struct ondisk_cache_entry
*ondisk
,
1420 unsigned long *ent_size
,
1421 struct strbuf
*previous_name
)
1423 struct cache_entry
*ce
;
1428 /* On-disk flags are just 16 bits */
1429 flags
= get_be16(&ondisk
->flags
);
1430 len
= flags
& CE_NAMEMASK
;
1432 if (flags
& CE_EXTENDED
) {
1433 struct ondisk_cache_entry_extended
*ondisk2
;
1435 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1436 extended_flags
= get_be16(&ondisk2
->flags2
) << 16;
1437 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1438 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1439 die("Unknown index entry format %08x", extended_flags
);
1440 flags
|= extended_flags
;
1441 name
= ondisk2
->name
;
1444 name
= ondisk
->name
;
1446 if (!previous_name
) {
1447 /* v3 and earlier */
1448 if (len
== CE_NAMEMASK
)
1450 ce
= cache_entry_from_ondisk(ondisk
, flags
, name
, len
);
1452 *ent_size
= ondisk_ce_size(ce
);
1454 unsigned long consumed
;
1455 consumed
= expand_name_field(previous_name
, name
);
1456 ce
= cache_entry_from_ondisk(ondisk
, flags
,
1458 previous_name
->len
);
1460 *ent_size
= (name
- ((char *)ondisk
)) + consumed
;
1465 /* remember to discard_cache() before reading a different cache! */
1466 static int do_read_index(struct index_state
*istate
, const char *path
,
1471 unsigned long src_offset
;
1472 struct cache_header
*hdr
;
1475 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1477 if (istate
->initialized
)
1478 return istate
->cache_nr
;
1480 istate
->timestamp
.sec
= 0;
1481 istate
->timestamp
.nsec
= 0;
1482 fd
= open(path
, O_RDONLY
);
1484 if (!must_exist
&& errno
== ENOENT
)
1486 die_errno("%s: index file open failed", path
);
1490 die_errno("cannot stat the open index");
1492 mmap_size
= xsize_t(st
.st_size
);
1493 if (mmap_size
< sizeof(struct cache_header
) + 20)
1494 die("index file smaller than expected");
1496 mmap
= xmmap(NULL
, mmap_size
, PROT_READ
| PROT_WRITE
, MAP_PRIVATE
, fd
, 0);
1497 if (mmap
== MAP_FAILED
)
1498 die_errno("unable to map index file");
1502 if (verify_hdr(hdr
, mmap_size
) < 0)
1505 hashcpy(istate
->sha1
, (const unsigned char *)hdr
+ mmap_size
- 20);
1506 istate
->version
= ntohl(hdr
->hdr_version
);
1507 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
1508 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
1509 istate
->cache
= xcalloc(istate
->cache_alloc
, sizeof(*istate
->cache
));
1510 istate
->initialized
= 1;
1512 if (istate
->version
== 4)
1513 previous_name
= &previous_name_buf
;
1515 previous_name
= NULL
;
1517 src_offset
= sizeof(*hdr
);
1518 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1519 struct ondisk_cache_entry
*disk_ce
;
1520 struct cache_entry
*ce
;
1521 unsigned long consumed
;
1523 disk_ce
= (struct ondisk_cache_entry
*)((char *)mmap
+ src_offset
);
1524 ce
= create_from_disk(disk_ce
, &consumed
, previous_name
);
1525 set_index_entry(istate
, i
, ce
);
1527 src_offset
+= consumed
;
1529 strbuf_release(&previous_name_buf
);
1530 istate
->timestamp
.sec
= st
.st_mtime
;
1531 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1533 while (src_offset
<= mmap_size
- 20 - 8) {
1534 /* After an array of active_nr index entries,
1535 * there can be arbitrary number of extended
1536 * sections, each of which is prefixed with
1537 * extension name (4-byte) and section length
1538 * in 4-byte network byte order.
1541 memcpy(&extsize
, (char *)mmap
+ src_offset
+ 4, 4);
1542 extsize
= ntohl(extsize
);
1543 if (read_index_extension(istate
,
1544 (const char *) mmap
+ src_offset
,
1545 (char *) mmap
+ src_offset
+ 8,
1549 src_offset
+= extsize
;
1551 munmap(mmap
, mmap_size
);
1552 return istate
->cache_nr
;
1555 munmap(mmap
, mmap_size
);
1556 die("index file corrupt");
1559 int read_index_from(struct index_state
*istate
, const char *path
)
1561 struct split_index
*split_index
;
1564 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1565 if (istate
->initialized
)
1566 return istate
->cache_nr
;
1568 ret
= do_read_index(istate
, path
, 0);
1569 split_index
= istate
->split_index
;
1573 if (is_null_sha1(split_index
->base_sha1
))
1576 if (split_index
->base
)
1577 discard_index(split_index
->base
);
1579 split_index
->base
= xcalloc(1, sizeof(*split_index
->base
));
1580 ret
= do_read_index(split_index
->base
,
1581 git_path("sharedindex.%s",
1582 sha1_to_hex(split_index
->base_sha1
)), 1);
1583 if (hashcmp(split_index
->base_sha1
, split_index
->base
->sha1
))
1584 die("broken index, expect %s in %s, got %s",
1585 sha1_to_hex(split_index
->base_sha1
),
1586 git_path("sharedindex.%s",
1587 sha1_to_hex(split_index
->base_sha1
)),
1588 sha1_to_hex(split_index
->base
->sha1
));
1589 merge_base_index(istate
);
1593 int is_index_unborn(struct index_state
*istate
)
1595 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
1598 int discard_index(struct index_state
*istate
)
1602 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1603 if (istate
->cache
[i
]->index
&&
1604 istate
->split_index
&&
1605 istate
->split_index
->base
&&
1606 istate
->cache
[i
]->index
<= istate
->split_index
->base
->cache_nr
&&
1607 istate
->cache
[i
] == istate
->split_index
->base
->cache
[istate
->cache
[i
]->index
- 1])
1609 free(istate
->cache
[i
]);
1611 resolve_undo_clear_index(istate
);
1612 istate
->cache_nr
= 0;
1613 istate
->cache_changed
= 0;
1614 istate
->timestamp
.sec
= 0;
1615 istate
->timestamp
.nsec
= 0;
1616 free_name_hash(istate
);
1617 cache_tree_free(&(istate
->cache_tree
));
1618 istate
->initialized
= 0;
1619 free(istate
->cache
);
1620 istate
->cache
= NULL
;
1621 istate
->cache_alloc
= 0;
1622 discard_split_index(istate
);
1626 int unmerged_index(const struct index_state
*istate
)
1629 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1630 if (ce_stage(istate
->cache
[i
]))
1636 #define WRITE_BUFFER_SIZE 8192
1637 static unsigned char write_buffer
[WRITE_BUFFER_SIZE
];
1638 static unsigned long write_buffer_len
;
1640 static int ce_write_flush(git_SHA_CTX
*context
, int fd
)
1642 unsigned int buffered
= write_buffer_len
;
1644 git_SHA1_Update(context
, write_buffer
, buffered
);
1645 if (write_in_full(fd
, write_buffer
, buffered
) != buffered
)
1647 write_buffer_len
= 0;
1652 static int ce_write(git_SHA_CTX
*context
, int fd
, void *data
, unsigned int len
)
1655 unsigned int buffered
= write_buffer_len
;
1656 unsigned int partial
= WRITE_BUFFER_SIZE
- buffered
;
1659 memcpy(write_buffer
+ buffered
, data
, partial
);
1660 buffered
+= partial
;
1661 if (buffered
== WRITE_BUFFER_SIZE
) {
1662 write_buffer_len
= buffered
;
1663 if (ce_write_flush(context
, fd
))
1667 write_buffer_len
= buffered
;
1669 data
= (char *) data
+ partial
;
1674 static int write_index_ext_header(git_SHA_CTX
*context
, int fd
,
1675 unsigned int ext
, unsigned int sz
)
1679 return ((ce_write(context
, fd
, &ext
, 4) < 0) ||
1680 (ce_write(context
, fd
, &sz
, 4) < 0)) ? -1 : 0;
1683 static int ce_flush(git_SHA_CTX
*context
, int fd
, unsigned char *sha1
)
1685 unsigned int left
= write_buffer_len
;
1688 write_buffer_len
= 0;
1689 git_SHA1_Update(context
, write_buffer
, left
);
1692 /* Flush first if not enough space for SHA1 signature */
1693 if (left
+ 20 > WRITE_BUFFER_SIZE
) {
1694 if (write_in_full(fd
, write_buffer
, left
) != left
)
1699 /* Append the SHA1 signature at the end */
1700 git_SHA1_Final(write_buffer
+ left
, context
);
1701 hashcpy(sha1
, write_buffer
+ left
);
1703 return (write_in_full(fd
, write_buffer
, left
) != left
) ? -1 : 0;
1706 static void ce_smudge_racily_clean_entry(struct cache_entry
*ce
)
1709 * The only thing we care about in this function is to smudge the
1710 * falsely clean entry due to touch-update-touch race, so we leave
1711 * everything else as they are. We are called for entries whose
1712 * ce_stat_data.sd_mtime match the index file mtime.
1714 * Note that this actually does not do much for gitlinks, for
1715 * which ce_match_stat_basic() always goes to the actual
1716 * contents. The caller checks with is_racy_timestamp() which
1717 * always says "no" for gitlinks, so we are not called for them ;-)
1721 if (lstat(ce
->name
, &st
) < 0)
1723 if (ce_match_stat_basic(ce
, &st
))
1725 if (ce_modified_check_fs(ce
, &st
)) {
1726 /* This is "racily clean"; smudge it. Note that this
1727 * is a tricky code. At first glance, it may appear
1728 * that it can break with this sequence:
1730 * $ echo xyzzy >frotz
1731 * $ git-update-index --add frotz
1734 * $ echo filfre >nitfol
1735 * $ git-update-index --add nitfol
1737 * but it does not. When the second update-index runs,
1738 * it notices that the entry "frotz" has the same timestamp
1739 * as index, and if we were to smudge it by resetting its
1740 * size to zero here, then the object name recorded
1741 * in index is the 6-byte file but the cached stat information
1742 * becomes zero --- which would then match what we would
1743 * obtain from the filesystem next time we stat("frotz").
1745 * However, the second update-index, before calling
1746 * this function, notices that the cached size is 6
1747 * bytes and what is on the filesystem is an empty
1748 * file, and never calls us, so the cached size information
1749 * for "frotz" stays 6 which does not match the filesystem.
1751 ce
->ce_stat_data
.sd_size
= 0;
1755 /* Copy miscellaneous fields but not the name */
1756 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
1757 struct cache_entry
*ce
)
1761 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
1762 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
1763 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
1764 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
1765 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
1766 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
1767 ondisk
->mode
= htonl(ce
->ce_mode
);
1768 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
1769 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
1770 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
1771 hashcpy(ondisk
->sha1
, ce
->sha1
);
1773 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
1774 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
1775 ondisk
->flags
= htons(flags
);
1776 if (ce
->ce_flags
& CE_EXTENDED
) {
1777 struct ondisk_cache_entry_extended
*ondisk2
;
1778 ondisk2
= (struct ondisk_cache_entry_extended
*)ondisk
;
1779 ondisk2
->flags2
= htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
1780 return ondisk2
->name
;
1783 return ondisk
->name
;
1787 static int ce_write_entry(git_SHA_CTX
*c
, int fd
, struct cache_entry
*ce
,
1788 struct strbuf
*previous_name
)
1791 struct ondisk_cache_entry
*ondisk
;
1792 int saved_namelen
= saved_namelen
; /* compiler workaround */
1796 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1797 saved_namelen
= ce_namelen(ce
);
1801 if (!previous_name
) {
1802 size
= ondisk_ce_size(ce
);
1803 ondisk
= xcalloc(1, size
);
1804 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1805 memcpy(name
, ce
->name
, ce_namelen(ce
));
1807 int common
, to_remove
, prefix_size
;
1808 unsigned char to_remove_vi
[16];
1810 (ce
->name
[common
] &&
1811 common
< previous_name
->len
&&
1812 ce
->name
[common
] == previous_name
->buf
[common
]);
1814 ; /* still matching */
1815 to_remove
= previous_name
->len
- common
;
1816 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
1818 if (ce
->ce_flags
& CE_EXTENDED
)
1819 size
= offsetof(struct ondisk_cache_entry_extended
, name
);
1821 size
= offsetof(struct ondisk_cache_entry
, name
);
1822 size
+= prefix_size
+ (ce_namelen(ce
) - common
+ 1);
1824 ondisk
= xcalloc(1, size
);
1825 name
= copy_cache_entry_to_ondisk(ondisk
, ce
);
1826 memcpy(name
, to_remove_vi
, prefix_size
);
1827 memcpy(name
+ prefix_size
, ce
->name
+ common
, ce_namelen(ce
) - common
);
1829 strbuf_splice(previous_name
, common
, to_remove
,
1830 ce
->name
+ common
, ce_namelen(ce
) - common
);
1832 if (ce
->ce_flags
& CE_STRIP_NAME
) {
1833 ce
->ce_namelen
= saved_namelen
;
1834 ce
->ce_flags
&= ~CE_STRIP_NAME
;
1837 result
= ce_write(c
, fd
, ondisk
, size
);
1842 static int has_racy_timestamp(struct index_state
*istate
)
1844 int entries
= istate
->cache_nr
;
1847 for (i
= 0; i
< entries
; i
++) {
1848 struct cache_entry
*ce
= istate
->cache
[i
];
1849 if (is_racy_timestamp(istate
, ce
))
1856 * Opportunistically update the index but do not complain if we can't
1858 void update_index_if_able(struct index_state
*istate
, struct lock_file
*lockfile
)
1860 if ((istate
->cache_changed
|| has_racy_timestamp(istate
)) &&
1861 write_locked_index(istate
, lockfile
, COMMIT_LOCK
))
1862 rollback_lock_file(lockfile
);
1865 static int do_write_index(struct index_state
*istate
, int newfd
,
1866 int strip_extensions
)
1869 struct cache_header hdr
;
1870 int i
, err
, removed
, extended
, hdr_version
;
1871 struct cache_entry
**cache
= istate
->cache
;
1872 int entries
= istate
->cache_nr
;
1874 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
1876 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
1877 if (cache
[i
]->ce_flags
& CE_REMOVE
)
1880 /* reduce extended entries if possible */
1881 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
1882 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
1884 cache
[i
]->ce_flags
|= CE_EXTENDED
;
1888 if (!istate
->version
)
1889 istate
->version
= get_index_format_default();
1891 /* demote version 3 to version 2 when the latter suffices */
1892 if (istate
->version
== 3 || istate
->version
== 2)
1893 istate
->version
= extended
? 3 : 2;
1895 hdr_version
= istate
->version
;
1897 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
1898 hdr
.hdr_version
= htonl(hdr_version
);
1899 hdr
.hdr_entries
= htonl(entries
- removed
);
1902 if (ce_write(&c
, newfd
, &hdr
, sizeof(hdr
)) < 0)
1905 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
1906 for (i
= 0; i
< entries
; i
++) {
1907 struct cache_entry
*ce
= cache
[i
];
1908 if (ce
->ce_flags
& CE_REMOVE
)
1910 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
1911 ce_smudge_racily_clean_entry(ce
);
1912 if (is_null_sha1(ce
->sha1
)) {
1913 static const char msg
[] = "cache entry has null sha1: %s";
1914 static int allow
= -1;
1917 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1919 warning(msg
, ce
->name
);
1921 return error(msg
, ce
->name
);
1923 if (ce_write_entry(&c
, newfd
, ce
, previous_name
) < 0)
1926 strbuf_release(&previous_name_buf
);
1928 /* Write extension data here */
1929 if (!strip_extensions
&& istate
->split_index
) {
1930 struct strbuf sb
= STRBUF_INIT
;
1932 err
= write_link_extension(&sb
, istate
) < 0 ||
1933 write_index_ext_header(&c
, newfd
, CACHE_EXT_LINK
,
1935 ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1936 strbuf_release(&sb
);
1940 if (!strip_extensions
&& istate
->cache_tree
) {
1941 struct strbuf sb
= STRBUF_INIT
;
1943 cache_tree_write(&sb
, istate
->cache_tree
);
1944 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_TREE
, sb
.len
) < 0
1945 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1946 strbuf_release(&sb
);
1950 if (!strip_extensions
&& istate
->resolve_undo
) {
1951 struct strbuf sb
= STRBUF_INIT
;
1953 resolve_undo_write(&sb
, istate
->resolve_undo
);
1954 err
= write_index_ext_header(&c
, newfd
, CACHE_EXT_RESOLVE_UNDO
,
1956 || ce_write(&c
, newfd
, sb
.buf
, sb
.len
) < 0;
1957 strbuf_release(&sb
);
1962 if (ce_flush(&c
, newfd
, istate
->sha1
) || fstat(newfd
, &st
))
1964 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
1965 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
1969 void set_alternate_index_output(const char *name
)
1971 alternate_index_output
= name
;
1974 static int commit_locked_index(struct lock_file
*lk
)
1976 if (alternate_index_output
) {
1977 if (lk
->fd
>= 0 && close_lock_file(lk
))
1979 if (rename(lk
->filename
, alternate_index_output
))
1981 lk
->filename
[0] = 0;
1984 return commit_lock_file(lk
);
1988 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
1991 int ret
= do_write_index(istate
, lock
->fd
, 0);
1994 assert((flags
& (COMMIT_LOCK
| CLOSE_LOCK
)) !=
1995 (COMMIT_LOCK
| CLOSE_LOCK
));
1996 if (flags
& COMMIT_LOCK
)
1997 return commit_locked_index(lock
);
1998 else if (flags
& CLOSE_LOCK
)
1999 return close_lock_file(lock
);
2004 static int write_split_index(struct index_state
*istate
,
2005 struct lock_file
*lock
,
2009 prepare_to_write_split_index(istate
);
2010 ret
= do_write_locked_index(istate
, lock
, flags
);
2011 finish_writing_split_index(istate
);
2015 static char *temporary_sharedindex
;
2017 static void remove_temporary_sharedindex(void)
2019 if (temporary_sharedindex
) {
2020 unlink_or_warn(temporary_sharedindex
);
2021 free(temporary_sharedindex
);
2022 temporary_sharedindex
= NULL
;
2026 static void remove_temporary_sharedindex_on_signal(int signo
)
2028 remove_temporary_sharedindex();
2029 sigchain_pop(signo
);
2033 static int write_shared_index(struct index_state
*istate
)
2035 struct split_index
*si
= istate
->split_index
;
2036 static int installed_handler
;
2039 temporary_sharedindex
= git_pathdup("sharedindex_XXXXXX");
2040 fd
= xmkstemp(temporary_sharedindex
);
2041 if (!installed_handler
) {
2042 atexit(remove_temporary_sharedindex
);
2043 sigchain_push_common(remove_temporary_sharedindex_on_signal
);
2045 move_cache_to_base_index(istate
);
2046 ret
= do_write_index(si
->base
, fd
, 1);
2049 remove_temporary_sharedindex();
2052 ret
= rename(temporary_sharedindex
,
2053 git_path("sharedindex.%s", sha1_to_hex(si
->base
->sha1
)));
2054 free(temporary_sharedindex
);
2055 temporary_sharedindex
= NULL
;
2057 hashcpy(si
->base_sha1
, si
->base
->sha1
);
2061 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
2064 struct split_index
*si
= istate
->split_index
;
2066 if (!si
|| (istate
->cache_changed
& ~EXTMASK
)) {
2068 hashclr(si
->base_sha1
);
2069 return do_write_locked_index(istate
, lock
, flags
);
2072 if (istate
->cache_changed
& SPLIT_INDEX_ORDERED
) {
2073 int ret
= write_shared_index(istate
);
2078 return write_split_index(istate
, lock
, flags
);
2082 * Read the index file that is potentially unmerged into given
2083 * index_state, dropping any unmerged entries. Returns true if
2084 * the index is unmerged. Callers who want to refuse to work
2085 * from an unmerged state can call this and check its return value,
2086 * instead of calling read_cache().
2088 int read_index_unmerged(struct index_state
*istate
)
2094 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2095 struct cache_entry
*ce
= istate
->cache
[i
];
2096 struct cache_entry
*new_ce
;
2102 len
= ce_namelen(ce
);
2103 size
= cache_entry_size(len
);
2104 new_ce
= xcalloc(1, size
);
2105 memcpy(new_ce
->name
, ce
->name
, len
);
2106 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
2107 new_ce
->ce_namelen
= len
;
2108 new_ce
->ce_mode
= ce
->ce_mode
;
2109 if (add_index_entry(istate
, new_ce
, 0))
2110 return error("%s: cannot drop to stage #0",
2112 i
= index_name_pos(istate
, new_ce
->name
, len
);
2118 * Returns 1 if the path is an "other" path with respect to
2119 * the index; that is, the path is not mentioned in the index at all,
2120 * either as a file, a directory with some files in the index,
2121 * or as an unmerged entry.
2123 * We helpfully remove a trailing "/" from directories so that
2124 * the output of read_directory can be used as-is.
2126 int index_name_is_other(const struct index_state
*istate
, const char *name
,
2130 if (namelen
&& name
[namelen
- 1] == '/')
2132 pos
= index_name_pos(istate
, name
, namelen
);
2134 return 0; /* exact match */
2136 if (pos
< istate
->cache_nr
) {
2137 struct cache_entry
*ce
= istate
->cache
[pos
];
2138 if (ce_namelen(ce
) == namelen
&&
2139 !memcmp(ce
->name
, name
, namelen
))
2140 return 0; /* Yup, this one exists unmerged */
2145 void *read_blob_data_from_index(struct index_state
*istate
, const char *path
, unsigned long *size
)
2149 enum object_type type
;
2153 pos
= index_name_pos(istate
, path
, len
);
2156 * We might be in the middle of a merge, in which
2157 * case we would read stage #2 (ours).
2161 (pos
< 0 && i
< istate
->cache_nr
&&
2162 !strcmp(istate
->cache
[i
]->name
, path
));
2164 if (ce_stage(istate
->cache
[i
]) == 2)
2169 data
= read_sha1_file(istate
->cache
[pos
]->sha1
, &type
, &sz
);
2170 if (!data
|| type
!= OBJ_BLOB
) {
2179 void stat_validity_clear(struct stat_validity
*sv
)
2185 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
2189 if (stat(path
, &st
) < 0)
2190 return sv
->sd
== NULL
;
2193 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
2196 void stat_validity_update(struct stat_validity
*sv
, int fd
)
2200 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
2201 stat_validity_clear(sv
);
2204 sv
->sd
= xcalloc(1, sizeof(struct stat_data
));
2205 fill_stat_data(sv
->sd
, &st
);