2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
14 #include "cache-tree.h"
17 #include "object-file.h"
18 #include "object-store.h"
19 #include "oid-array.h"
23 #include "environment.h"
26 #include "object-name.h"
27 #include "resolve-undo.h"
28 #include "run-command.h"
32 #include "split-index.h"
34 #include "fsmonitor.h"
35 #include "thread-utils.h"
37 #include "sparse-index.h"
38 #include "csum-file.h"
39 #include "promisor-remote.h"
43 /* Mask for the name length in ce_flags in the on-disk index */
45 #define CE_NAMEMASK (0x0fff)
49 * The first letter should be 'A'..'Z' for extensions that are not
50 * necessary for a correct operation (i.e. optimization data).
51 * When new extensions are added that _needs_ to be understood in
52 * order to correctly interpret the index file, pick character that
53 * is outside the range, to cause the reader to abort.
56 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
57 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
58 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
59 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
60 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
61 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
62 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
63 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
64 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
66 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
67 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
68 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
69 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
73 * This is an estimate of the pathname length in the index. We use
74 * this for V4 index files to guess the un-deltafied size of the index
75 * in memory because of pathname deltafication. This is not required
76 * for V2/V3 index formats because their pathnames are not compressed.
77 * If the initial amount of memory set aside is not sufficient, the
78 * mem pool will allocate extra memory.
80 #define CACHE_ENTRY_PATH_LENGTH 80
82 enum index_search_mode
{
87 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
89 struct cache_entry
*ce
;
90 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
91 ce
->mem_pool_allocated
= 1;
95 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
97 struct cache_entry
* ce
;
98 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
99 ce
->mem_pool_allocated
= 1;
103 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
105 struct mem_pool
**pool_ptr
;
107 if (istate
->split_index
&& istate
->split_index
->base
)
108 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
110 pool_ptr
= &istate
->ce_mem_pool
;
113 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
114 mem_pool_init(*pool_ptr
, 0);
120 static const char *alternate_index_output
;
122 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
124 if (S_ISSPARSEDIR(ce
->ce_mode
))
125 istate
->sparse_index
= INDEX_COLLAPSED
;
127 istate
->cache
[nr
] = ce
;
128 add_name_hash(istate
, ce
);
131 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
133 struct cache_entry
*old
= istate
->cache
[nr
];
135 replace_index_entry_in_base(istate
, old
, ce
);
136 remove_name_hash(istate
, old
);
137 discard_cache_entry(old
);
138 ce
->ce_flags
&= ~CE_HASHED
;
139 set_index_entry(istate
, nr
, ce
);
140 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
141 mark_fsmonitor_invalid(istate
, ce
);
142 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
145 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
147 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
, *refreshed
;
148 int namelen
= strlen(new_name
);
150 new_entry
= make_empty_cache_entry(istate
, namelen
);
151 copy_cache_entry(new_entry
, old_entry
);
152 new_entry
->ce_flags
&= ~CE_HASHED
;
153 new_entry
->ce_namelen
= namelen
;
154 new_entry
->index
= 0;
155 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
157 cache_tree_invalidate_path(istate
, old_entry
->name
);
158 untracked_cache_remove_from_index(istate
, old_entry
->name
);
159 remove_index_entry_at(istate
, nr
);
162 * Refresh the new index entry. Using 'refresh_cache_entry' ensures
163 * we only update stat info if the entry is otherwise up-to-date (i.e.,
164 * the contents/mode haven't changed). This ensures that we reflect the
165 * 'ctime' of the rename in the index without (incorrectly) updating
166 * the cached stat info to reflect unstaged changes on disk.
168 refreshed
= refresh_cache_entry(istate
, new_entry
, CE_MATCH_REFRESH
);
169 if (refreshed
&& refreshed
!= new_entry
) {
170 add_index_entry(istate
, refreshed
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
171 discard_cache_entry(new_entry
);
173 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
176 void fill_stat_data(struct stat_data
*sd
, struct stat
*st
)
178 sd
->sd_ctime
.sec
= (unsigned int)st
->st_ctime
;
179 sd
->sd_mtime
.sec
= (unsigned int)st
->st_mtime
;
180 sd
->sd_ctime
.nsec
= ST_CTIME_NSEC(*st
);
181 sd
->sd_mtime
.nsec
= ST_MTIME_NSEC(*st
);
182 sd
->sd_dev
= st
->st_dev
;
183 sd
->sd_ino
= st
->st_ino
;
184 sd
->sd_uid
= st
->st_uid
;
185 sd
->sd_gid
= st
->st_gid
;
186 sd
->sd_size
= st
->st_size
;
189 int match_stat_data(const struct stat_data
*sd
, struct stat
*st
)
193 if (sd
->sd_mtime
.sec
!= (unsigned int)st
->st_mtime
)
194 changed
|= MTIME_CHANGED
;
195 if (trust_ctime
&& check_stat
&&
196 sd
->sd_ctime
.sec
!= (unsigned int)st
->st_ctime
)
197 changed
|= CTIME_CHANGED
;
200 if (check_stat
&& sd
->sd_mtime
.nsec
!= ST_MTIME_NSEC(*st
))
201 changed
|= MTIME_CHANGED
;
202 if (trust_ctime
&& check_stat
&&
203 sd
->sd_ctime
.nsec
!= ST_CTIME_NSEC(*st
))
204 changed
|= CTIME_CHANGED
;
208 if (sd
->sd_uid
!= (unsigned int) st
->st_uid
||
209 sd
->sd_gid
!= (unsigned int) st
->st_gid
)
210 changed
|= OWNER_CHANGED
;
211 if (sd
->sd_ino
!= (unsigned int) st
->st_ino
)
212 changed
|= INODE_CHANGED
;
217 * st_dev breaks on network filesystems where different
218 * clients will have different views of what "device"
219 * the filesystem is on
221 if (check_stat
&& sd
->sd_dev
!= (unsigned int) st
->st_dev
)
222 changed
|= INODE_CHANGED
;
225 if (sd
->sd_size
!= (unsigned int) st
->st_size
)
226 changed
|= DATA_CHANGED
;
232 * This only updates the "non-critical" parts of the directory
233 * cache, ie the parts that aren't tracked by GIT, and only used
234 * to validate the cache.
236 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
238 fill_stat_data(&ce
->ce_stat_data
, st
);
240 if (assume_unchanged
)
241 ce
->ce_flags
|= CE_VALID
;
243 if (S_ISREG(st
->st_mode
)) {
244 ce_mark_uptodate(ce
);
245 mark_fsmonitor_valid(istate
, ce
);
249 static int ce_compare_data(struct index_state
*istate
,
250 const struct cache_entry
*ce
,
254 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
257 struct object_id oid
;
258 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
259 match
= !oideq(&oid
, &ce
->oid
);
260 /* index_fd() closed the file descriptor already */
265 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
270 enum object_type type
;
271 struct strbuf sb
= STRBUF_INIT
;
273 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
276 buffer
= repo_read_object_file(the_repository
, &ce
->oid
, &type
, &size
);
279 match
= memcmp(buffer
, sb
.buf
, size
);
286 static int ce_compare_gitlink(const struct cache_entry
*ce
)
288 struct object_id oid
;
291 * We don't actually require that the .git directory
292 * under GITLINK directory be a valid git directory. It
293 * might even be missing (in case nobody populated that
296 * If so, we consider it always to match.
298 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
300 return !oideq(&oid
, &ce
->oid
);
303 static int ce_modified_check_fs(struct index_state
*istate
,
304 const struct cache_entry
*ce
,
307 switch (st
->st_mode
& S_IFMT
) {
309 if (ce_compare_data(istate
, ce
, st
))
313 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
317 if (S_ISGITLINK(ce
->ce_mode
))
318 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
319 /* else fallthrough */
326 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
328 unsigned int changed
= 0;
330 if (ce
->ce_flags
& CE_REMOVE
)
331 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
333 switch (ce
->ce_mode
& S_IFMT
) {
335 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
336 /* We consider only the owner x bit to be relevant for
339 if (trust_executable_bit
&&
340 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
341 changed
|= MODE_CHANGED
;
344 if (!S_ISLNK(st
->st_mode
) &&
345 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
346 changed
|= TYPE_CHANGED
;
349 /* We ignore most of the st_xxx fields for gitlinks */
350 if (!S_ISDIR(st
->st_mode
))
351 changed
|= TYPE_CHANGED
;
352 else if (ce_compare_gitlink(ce
))
353 changed
|= DATA_CHANGED
;
356 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
359 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
361 /* Racily smudged entry? */
362 if (!ce
->ce_stat_data
.sd_size
) {
363 if (!is_empty_blob_sha1(ce
->oid
.hash
))
364 changed
|= DATA_CHANGED
;
370 static int is_racy_stat(const struct index_state
*istate
,
371 const struct stat_data
*sd
)
373 return (istate
->timestamp
.sec
&&
375 /* nanosecond timestamped files can also be racy! */
376 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
377 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
378 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
380 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
385 int is_racy_timestamp(const struct index_state
*istate
,
386 const struct cache_entry
*ce
)
388 return (!S_ISGITLINK(ce
->ce_mode
) &&
389 is_racy_stat(istate
, &ce
->ce_stat_data
));
392 int match_stat_data_racy(const struct index_state
*istate
,
393 const struct stat_data
*sd
, struct stat
*st
)
395 if (is_racy_stat(istate
, sd
))
396 return MTIME_CHANGED
;
397 return match_stat_data(sd
, st
);
400 int ie_match_stat(struct index_state
*istate
,
401 const struct cache_entry
*ce
, struct stat
*st
,
402 unsigned int options
)
404 unsigned int changed
;
405 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
406 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
407 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
408 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
410 if (!ignore_fsmonitor
)
411 refresh_fsmonitor(istate
);
413 * If it's marked as always valid in the index, it's
414 * valid whatever the checked-out copy says.
416 * skip-worktree has the same effect with higher precedence
418 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
420 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
422 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
426 * Intent-to-add entries have not been added, so the index entry
427 * by definition never matches what is in the work tree until it
428 * actually gets added.
430 if (ce_intent_to_add(ce
))
431 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
433 changed
= ce_match_stat_basic(ce
, st
);
436 * Within 1 second of this sequence:
437 * echo xyzzy >file && git-update-index --add file
438 * running this command:
440 * would give a falsely clean cache entry. The mtime and
441 * length match the cache, and other stat fields do not change.
443 * We could detect this at update-index time (the cache entry
444 * being registered/updated records the same time as "now")
445 * and delay the return from git-update-index, but that would
446 * effectively mean we can make at most one commit per second,
447 * which is not acceptable. Instead, we check cache entries
448 * whose mtime are the same as the index file timestamp more
449 * carefully than others.
451 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
452 if (assume_racy_is_modified
)
453 changed
|= DATA_CHANGED
;
455 changed
|= ce_modified_check_fs(istate
, ce
, st
);
461 int ie_modified(struct index_state
*istate
,
462 const struct cache_entry
*ce
,
463 struct stat
*st
, unsigned int options
)
465 int changed
, changed_fs
;
467 changed
= ie_match_stat(istate
, ce
, st
, options
);
471 * If the mode or type has changed, there's no point in trying
472 * to refresh the entry - it's not going to match
474 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
478 * Immediately after read-tree or update-index --cacheinfo,
479 * the length field is zero, as we have never even read the
480 * lstat(2) information once, and we cannot trust DATA_CHANGED
481 * returned by ie_match_stat() which in turn was returned by
482 * ce_match_stat_basic() to signal that the filesize of the
483 * blob changed. We have to actually go to the filesystem to
484 * see if the contents match, and if so, should answer "unchanged".
486 * The logic does not apply to gitlinks, as ce_match_stat_basic()
487 * already has checked the actual HEAD from the filesystem in the
488 * subproject. If ie_match_stat() already said it is different,
489 * then we know it is.
491 if ((changed
& DATA_CHANGED
) &&
492 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
495 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
497 return changed
| changed_fs
;
501 int base_name_compare(const char *name1
, size_t len1
, int mode1
,
502 const char *name2
, size_t len2
, int mode2
)
504 unsigned char c1
, c2
;
505 size_t len
= len1
< len2
? len1
: len2
;
508 cmp
= memcmp(name1
, name2
, len
);
513 if (!c1
&& S_ISDIR(mode1
))
515 if (!c2
&& S_ISDIR(mode2
))
517 return (c1
< c2
) ? -1 : (c1
> c2
) ? 1 : 0;
521 * df_name_compare() is identical to base_name_compare(), except it
522 * compares conflicting directory/file entries as equal. Note that
523 * while a directory name compares as equal to a regular file, they
524 * then individually compare _differently_ to a filename that has
525 * a dot after the basename (because '\0' < '.' < '/').
527 * This is used by routines that want to traverse the git namespace
528 * but then handle conflicting entries together when possible.
530 int df_name_compare(const char *name1
, size_t len1
, int mode1
,
531 const char *name2
, size_t len2
, int mode2
)
533 unsigned char c1
, c2
;
534 size_t len
= len1
< len2
? len1
: len2
;
537 cmp
= memcmp(name1
, name2
, len
);
540 /* Directories and files compare equal (same length, same name) */
544 if (!c1
&& S_ISDIR(mode1
))
547 if (!c2
&& S_ISDIR(mode2
))
549 if (c1
== '/' && !c2
)
551 if (c2
== '/' && !c1
)
556 int name_compare(const char *name1
, size_t len1
, const char *name2
, size_t len2
)
558 size_t min_len
= (len1
< len2
) ? len1
: len2
;
559 int cmp
= memcmp(name1
, name2
, min_len
);
569 int cache_name_stage_compare(const char *name1
, int len1
, int stage1
, const char *name2
, int len2
, int stage2
)
573 cmp
= name_compare(name1
, len1
, name2
, len2
);
584 static int index_name_stage_pos(struct index_state
*istate
,
585 const char *name
, int namelen
,
587 enum index_search_mode search_mode
)
592 last
= istate
->cache_nr
;
593 while (last
> first
) {
594 int next
= first
+ ((last
- first
) >> 1);
595 struct cache_entry
*ce
= istate
->cache
[next
];
596 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
606 if (search_mode
== EXPAND_SPARSE
&& istate
->sparse_index
&&
608 /* Note: first <= istate->cache_nr */
609 struct cache_entry
*ce
= istate
->cache
[first
- 1];
612 * If we are in a sparse-index _and_ the entry before the
613 * insertion position is a sparse-directory entry that is
614 * an ancestor of 'name', then we need to expand the index
615 * and search again. This will only trigger once, because
616 * thereafter the index is fully expanded.
618 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
619 ce_namelen(ce
) < namelen
&&
620 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
621 ensure_full_index(istate
);
622 return index_name_stage_pos(istate
, name
, namelen
, stage
, search_mode
);
629 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
631 return index_name_stage_pos(istate
, name
, namelen
, 0, EXPAND_SPARSE
);
634 int index_name_pos_sparse(struct index_state
*istate
, const char *name
, int namelen
)
636 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
);
639 int index_entry_exists(struct index_state
*istate
, const char *name
, int namelen
)
641 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
) >= 0;
644 int remove_index_entry_at(struct index_state
*istate
, int pos
)
646 struct cache_entry
*ce
= istate
->cache
[pos
];
648 record_resolve_undo(istate
, ce
);
649 remove_name_hash(istate
, ce
);
650 save_or_free_index_entry(istate
, ce
);
651 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
653 if (pos
>= istate
->cache_nr
)
655 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
656 istate
->cache_nr
- pos
);
661 * Remove all cache entries marked for removal, that is where
662 * CE_REMOVE is set in ce_flags. This is much more effective than
663 * calling remove_index_entry_at() for each entry to be removed.
665 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
667 struct cache_entry
**ce_array
= istate
->cache
;
670 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
671 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
673 cache_tree_invalidate_path(istate
,
675 untracked_cache_remove_from_index(istate
,
678 remove_name_hash(istate
, ce_array
[i
]);
679 save_or_free_index_entry(istate
, ce_array
[i
]);
682 ce_array
[j
++] = ce_array
[i
];
684 if (j
== istate
->cache_nr
)
686 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
687 istate
->cache_nr
= j
;
690 int remove_file_from_index(struct index_state
*istate
, const char *path
)
692 int pos
= index_name_pos(istate
, path
, strlen(path
));
695 cache_tree_invalidate_path(istate
, path
);
696 untracked_cache_remove_from_index(istate
, path
);
697 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
698 remove_index_entry_at(istate
, pos
);
702 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
704 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
707 static int index_name_pos_also_unmerged(struct index_state
*istate
,
708 const char *path
, int namelen
)
710 int pos
= index_name_pos(istate
, path
, namelen
);
711 struct cache_entry
*ce
;
716 /* maybe unmerged? */
718 if (pos
>= istate
->cache_nr
||
719 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
722 /* order of preference: stage 2, 1, 3 */
723 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
724 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
725 !compare_name(ce
, path
, namelen
))
730 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
732 int len
= ce_namelen(ce
);
733 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
737 * If we add a filename that aliases in the cache, we will use the
738 * name that we already have - but we don't want to update the same
739 * alias twice, because that implies that there were actually two
740 * different files with aliasing names!
742 * So we use the CE_ADDED flag to verify that the alias was an old
743 * one before we accept it as
745 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
746 struct cache_entry
*ce
,
747 struct cache_entry
*alias
)
750 struct cache_entry
*new_entry
;
752 if (alias
->ce_flags
& CE_ADDED
)
753 die(_("will not add file alias '%s' ('%s' already exists in index)"),
754 ce
->name
, alias
->name
);
756 /* Ok, create the new entry using the name of the existing alias */
757 len
= ce_namelen(alias
);
758 new_entry
= make_empty_cache_entry(istate
, len
);
759 memcpy(new_entry
->name
, alias
->name
, len
);
760 copy_cache_entry(new_entry
, ce
);
761 save_or_free_index_entry(istate
, ce
);
765 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
767 struct object_id oid
;
768 if (write_object_file("", 0, OBJ_BLOB
, &oid
))
769 die(_("cannot create an empty blob in the object database"));
770 oidcpy(&ce
->oid
, &oid
);
773 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
775 int namelen
, was_same
;
776 mode_t st_mode
= st
->st_mode
;
777 struct cache_entry
*ce
, *alias
= NULL
;
778 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
779 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
780 int pretend
= flags
& ADD_CACHE_PRETEND
;
781 int intent_only
= flags
& ADD_CACHE_INTENT
;
782 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
783 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
784 unsigned hash_flags
= pretend
? 0 : HASH_WRITE_OBJECT
;
785 struct object_id oid
;
787 if (flags
& ADD_CACHE_RENORMALIZE
)
788 hash_flags
|= HASH_RENORMALIZE
;
790 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
791 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
793 namelen
= strlen(path
);
794 if (S_ISDIR(st_mode
)) {
795 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
796 return error(_("'%s' does not have a commit checked out"), path
);
797 while (namelen
&& path
[namelen
-1] == '/')
800 ce
= make_empty_cache_entry(istate
, namelen
);
801 memcpy(ce
->name
, path
, namelen
);
802 ce
->ce_namelen
= namelen
;
804 fill_stat_cache_info(istate
, ce
, st
);
806 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
809 if (trust_executable_bit
&& has_symlinks
) {
810 ce
->ce_mode
= create_ce_mode(st_mode
);
812 /* If there is an existing entry, pick the mode bits and type
813 * from it, otherwise assume unexecutable regular file.
815 struct cache_entry
*ent
;
816 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
818 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
819 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
822 /* When core.ignorecase=true, determine if a directory of the same name but differing
823 * case already exists within the Git repository. If it does, ensure the directory
824 * case of the file being added to the repository matches (is folded into) the existing
825 * entry's directory case.
828 adjust_dirname_case(istate
, ce
->name
);
830 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
831 alias
= index_file_exists(istate
, ce
->name
,
832 ce_namelen(ce
), ignore_case
);
835 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
836 /* Nothing changed, really */
837 if (!S_ISGITLINK(alias
->ce_mode
))
838 ce_mark_uptodate(alias
);
839 alias
->ce_flags
|= CE_ADDED
;
841 discard_cache_entry(ce
);
846 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
847 discard_cache_entry(ce
);
848 return error(_("unable to index file '%s'"), path
);
851 set_object_name_for_intent_to_add_entry(ce
);
853 if (ignore_case
&& alias
&& different_name(ce
, alias
))
854 ce
= create_alias_ce(istate
, ce
, alias
);
855 ce
->ce_flags
|= CE_ADDED
;
857 /* It was suspected to be racily clean, but it turns out to be Ok */
860 oideq(&alias
->oid
, &ce
->oid
) &&
861 ce
->ce_mode
== alias
->ce_mode
);
864 discard_cache_entry(ce
);
865 else if (add_index_entry(istate
, ce
, add_option
)) {
866 discard_cache_entry(ce
);
867 return error(_("unable to add '%s' to index"), path
);
869 if (verbose
&& !was_same
)
870 printf("add '%s'\n", path
);
874 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
877 if (lstat(path
, &st
))
878 die_errno(_("unable to stat '%s'"), path
);
879 return add_to_index(istate
, path
, &st
, flags
);
882 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
884 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
887 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
888 struct mem_pool
*ce_mem_pool
)
891 return mem_pool__ce_calloc(ce_mem_pool
, len
);
892 return xcalloc(1, cache_entry_size(len
));
895 enum verify_path_result
{
901 static enum verify_path_result
verify_path_internal(const char *, unsigned);
903 int verify_path(const char *path
, unsigned mode
)
905 return verify_path_internal(path
, mode
) == PATH_OK
;
908 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
910 const struct object_id
*oid
,
913 unsigned int refresh_options
)
915 struct cache_entry
*ce
, *ret
;
918 if (verify_path_internal(path
, mode
) == PATH_INVALID
) {
919 error(_("invalid path '%s'"), path
);
924 ce
= make_empty_cache_entry(istate
, len
);
926 oidcpy(&ce
->oid
, oid
);
927 memcpy(ce
->name
, path
, len
);
928 ce
->ce_flags
= create_ce_flags(stage
);
929 ce
->ce_namelen
= len
;
930 ce
->ce_mode
= create_ce_mode(mode
);
932 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
934 discard_cache_entry(ce
);
938 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
939 const struct object_id
*oid
,
942 struct mem_pool
*ce_mem_pool
)
944 struct cache_entry
*ce
;
947 if (!verify_path(path
, mode
)) {
948 error(_("invalid path '%s'"), path
);
953 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
955 oidcpy(&ce
->oid
, oid
);
956 memcpy(ce
->name
, path
, len
);
957 ce
->ce_flags
= create_ce_flags(stage
);
958 ce
->ce_namelen
= len
;
959 ce
->ce_mode
= create_ce_mode(mode
);
965 * Chmod an index entry with either +x or -x.
967 * Returns -1 if the chmod for the particular cache entry failed (if it's
968 * not a regular file), -2 if an invalid flip argument is passed in, 0
971 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
974 if (!S_ISREG(ce
->ce_mode
))
981 ce
->ce_mode
&= ~0111;
986 cache_tree_invalidate_path(istate
, ce
->name
);
987 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
988 mark_fsmonitor_invalid(istate
, ce
);
989 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
994 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
996 int len
= ce_namelen(a
);
997 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
1001 * We fundamentally don't like some paths: we don't want
1002 * dot or dot-dot anywhere, and for obvious reasons don't
1003 * want to recurse into ".git" either.
1005 * Also, we don't want double slashes or slashes at the
1006 * end that can make pathnames ambiguous.
1008 static int verify_dotfile(const char *rest
, unsigned mode
)
1011 * The first character was '.', but that
1012 * has already been discarded, we now test
1016 /* "." is not allowed */
1017 if (*rest
== '\0' || is_dir_sep(*rest
))
1022 * ".git" followed by NUL or slash is bad. Note that we match
1023 * case-insensitively here, even if ignore_case is not set.
1024 * This outlaws ".GIT" everywhere out of an abundance of caution,
1025 * since there's really no good reason to allow it.
1027 * Once we've seen ".git", we can also find ".gitmodules", etc (also
1028 * case-insensitively).
1032 if (rest
[1] != 'i' && rest
[1] != 'I')
1034 if (rest
[2] != 't' && rest
[2] != 'T')
1036 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
1038 if (S_ISLNK(mode
)) {
1040 if (skip_iprefix(rest
, "modules", &rest
) &&
1041 (*rest
== '\0' || is_dir_sep(*rest
)))
1046 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
1052 static enum verify_path_result
verify_path_internal(const char *path
,
1057 if (has_dos_drive_prefix(path
))
1058 return PATH_INVALID
;
1060 if (!is_valid_path(path
))
1061 return PATH_INVALID
;
1067 if (is_dir_sep(c
)) {
1071 if (is_hfs_dotgit(path
))
1072 return PATH_INVALID
;
1073 if (S_ISLNK(mode
)) {
1074 if (is_hfs_dotgitmodules(path
))
1075 return PATH_INVALID
;
1079 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
1081 return PATH_INVALID
;
1083 if (is_ntfs_dotgit(path
))
1084 return PATH_INVALID
;
1085 if (S_ISLNK(mode
)) {
1086 if (is_ntfs_dotgitmodules(path
))
1087 return PATH_INVALID
;
1092 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
1094 return PATH_INVALID
;
1096 * allow terminating directory separators for
1097 * sparse directory entries.
1100 return S_ISDIR(mode
) ? PATH_DIR_WITH_SEP
:
1102 } else if (c
== '\\' && protect_ntfs
) {
1103 if (is_ntfs_dotgit(path
))
1104 return PATH_INVALID
;
1105 if (S_ISLNK(mode
)) {
1106 if (is_ntfs_dotgitmodules(path
))
1107 return PATH_INVALID
;
1116 * Do we have another file that has the beginning components being a
1117 * proper superset of the name we're trying to add?
1119 static int has_file_name(struct index_state
*istate
,
1120 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1123 int len
= ce_namelen(ce
);
1124 int stage
= ce_stage(ce
);
1125 const char *name
= ce
->name
;
1127 while (pos
< istate
->cache_nr
) {
1128 struct cache_entry
*p
= istate
->cache
[pos
++];
1130 if (len
>= ce_namelen(p
))
1132 if (memcmp(name
, p
->name
, len
))
1134 if (ce_stage(p
) != stage
)
1136 if (p
->name
[len
] != '/')
1138 if (p
->ce_flags
& CE_REMOVE
)
1143 remove_index_entry_at(istate
, --pos
);
1150 * Like strcmp(), but also return the offset of the first change.
1151 * If strings are equal, return the length.
1153 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1158 return strcmp(s1
, s2
);
1160 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1165 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1169 * Do we have another file with a pathname that is a proper
1170 * subset of the name we're trying to add?
1172 * That is, is there another file in the index with a path
1173 * that matches a sub-directory in the given entry?
1175 static int has_dir_name(struct index_state
*istate
,
1176 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1179 int stage
= ce_stage(ce
);
1180 const char *name
= ce
->name
;
1181 const char *slash
= name
+ ce_namelen(ce
);
1186 * We are frequently called during an iteration on a sorted
1187 * list of pathnames and while building a new index. Therefore,
1188 * there is a high probability that this entry will eventually
1189 * be appended to the index, rather than inserted in the middle.
1190 * If we can confirm that, we can avoid binary searches on the
1191 * components of the pathname.
1193 * Compare the entry's full path with the last path in the index.
1195 if (istate
->cache_nr
> 0) {
1196 cmp_last
= strcmp_offset(name
,
1197 istate
->cache
[istate
->cache_nr
- 1]->name
,
1200 if (len_eq_last
== 0) {
1202 * The entry sorts AFTER the last one in the
1203 * index and their paths have no common prefix,
1204 * so there cannot be a F/D conflict.
1209 * The entry sorts AFTER the last one in the
1210 * index, but has a common prefix. Fall through
1211 * to the loop below to disect the entry's path
1212 * and see where the difference is.
1215 } else if (cmp_last
== 0) {
1217 * The entry exactly matches the last one in the
1218 * index, but because of multiple stage and CE_REMOVE
1219 * items, we fall through and let the regular search
1229 if (*--slash
== '/')
1231 if (slash
<= ce
->name
)
1238 * (len + 1) is a directory boundary (including
1239 * the trailing slash). And since the loop is
1240 * decrementing "slash", the first iteration is
1241 * the longest directory prefix; subsequent
1242 * iterations consider parent directories.
1245 if (len
+ 1 <= len_eq_last
) {
1247 * The directory prefix (including the trailing
1248 * slash) also appears as a prefix in the last
1249 * entry, so the remainder cannot collide (because
1250 * strcmp said the whole path was greater).
1255 * LT: last: xxx/file_A
1261 if (len
> len_eq_last
) {
1263 * This part of the directory prefix (excluding
1264 * the trailing slash) is longer than the known
1265 * equal portions, so this sub-directory cannot
1266 * collide with a file.
1275 * This is a possible collision. Fall through and
1276 * let the regular search code handle it.
1283 pos
= index_name_stage_pos(istate
, name
, len
, stage
, EXPAND_SPARSE
);
1286 * Found one, but not so fast. This could
1287 * be a marker that says "I was here, but
1288 * I am being removed". Such an entry is
1289 * not a part of the resulting tree, and
1290 * it is Ok to have a directory at the same
1293 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1297 remove_index_entry_at(istate
, pos
);
1305 * Trivial optimization: if we find an entry that
1306 * already matches the sub-directory, then we know
1307 * we're ok, and we can exit.
1309 while (pos
< istate
->cache_nr
) {
1310 struct cache_entry
*p
= istate
->cache
[pos
];
1311 if ((ce_namelen(p
) <= len
) ||
1312 (p
->name
[len
] != '/') ||
1313 memcmp(p
->name
, name
, len
))
1314 break; /* not our subdirectory */
1315 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1317 * p is at the same stage as our entry, and
1318 * is a subdirectory of what we are looking
1319 * at, so we cannot have conflicts at our
1320 * level or anything shorter.
1329 /* We may be in a situation where we already have path/file and path
1330 * is being added, or we already have path and path/file is being
1331 * added. Either one would result in a nonsense tree that has path
1332 * twice when git-write-tree tries to write it out. Prevent it.
1334 * If ok-to-replace is specified, we remove the conflicting entries
1335 * from the cache so the caller should recompute the insert position.
1336 * When this happens, we return non-zero.
1338 static int check_file_directory_conflict(struct index_state
*istate
,
1339 const struct cache_entry
*ce
,
1340 int pos
, int ok_to_replace
)
1345 * When ce is an "I am going away" entry, we allow it to be added
1347 if (ce
->ce_flags
& CE_REMOVE
)
1351 * We check if the path is a sub-path of a subsequent pathname
1352 * first, since removing those will not change the position
1355 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1358 * Then check if the path might have a clashing sub-directory
1361 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1364 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1367 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1368 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1369 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1370 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1373 * If this entry's path sorts after the last entry in the index,
1374 * we can avoid searching for it.
1376 if (istate
->cache_nr
> 0 &&
1377 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1378 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1380 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1383 * Cache tree path should be invalidated only after index_name_stage_pos,
1384 * in case it expands a sparse index.
1386 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1387 cache_tree_invalidate_path(istate
, ce
->name
);
1389 /* existing match? Just replace it. */
1392 replace_index_entry(istate
, pos
, ce
);
1397 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1398 untracked_cache_add_to_index(istate
, ce
->name
);
1401 * Inserting a merged entry ("stage 0") into the index
1402 * will always replace all non-merged entries..
1404 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1405 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1407 if (!remove_index_entry_at(istate
, pos
))
1414 if (verify_path_internal(ce
->name
, ce
->ce_mode
) == PATH_INVALID
)
1415 return error(_("invalid path '%s'"), ce
->name
);
1417 if (!skip_df_check
&&
1418 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1420 return error(_("'%s' appears as both a file and as a directory"),
1422 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1428 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1432 if (option
& ADD_CACHE_JUST_APPEND
)
1433 pos
= istate
->cache_nr
;
1436 ret
= add_index_entry_with_check(istate
, ce
, option
);
1442 /* Make sure the array is big enough .. */
1443 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1447 if (istate
->cache_nr
> pos
+ 1)
1448 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1449 istate
->cache_nr
- pos
- 1);
1450 set_index_entry(istate
, pos
, ce
);
1451 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1456 * "refresh" does not calculate a new sha1 file or bring the
1457 * cache up-to-date for mode/content changes. But what it
1458 * _does_ do is to "re-match" the stat information of a file
1459 * with the cache, so that you can refresh the cache for a
1460 * file that hasn't been changed but where the stat entry is
1463 * For example, you'd want to do this after doing a "git-read-tree",
1464 * to link up the stat cache details with the proper files.
1466 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1467 struct cache_entry
*ce
,
1468 unsigned int options
, int *err
,
1474 struct cache_entry
*updated
;
1476 int refresh
= options
& CE_MATCH_REFRESH
;
1477 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1478 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1479 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1480 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1482 if (!refresh
|| ce_uptodate(ce
))
1485 if (!ignore_fsmonitor
)
1486 refresh_fsmonitor(istate
);
1488 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1489 * that the change to the work tree does not matter and told
1492 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1493 ce_mark_uptodate(ce
);
1496 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1497 ce_mark_uptodate(ce
);
1500 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1501 ce_mark_uptodate(ce
);
1505 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1515 if (lstat(ce
->name
, &st
) < 0) {
1516 if (ignore_missing
&& errno
== ENOENT
)
1523 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1525 *changed_ret
= changed
;
1528 * The path is unchanged. If we were told to ignore
1529 * valid bit, then we did the actual stat check and
1530 * found that the entry is unmodified. If the entry
1531 * is not marked VALID, this is the place to mark it
1532 * valid again, under "assume unchanged" mode.
1534 if (ignore_valid
&& assume_unchanged
&&
1535 !(ce
->ce_flags
& CE_VALID
))
1536 ; /* mark this one VALID again */
1539 * We do not mark the index itself "modified"
1540 * because CE_UPTODATE flag is in-core only;
1541 * we are not going to write this change out.
1543 if (!S_ISGITLINK(ce
->ce_mode
)) {
1544 ce_mark_uptodate(ce
);
1545 mark_fsmonitor_valid(istate
, ce
);
1553 if (ie_modified(istate
, ce
, &st
, options
)) {
1559 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1560 copy_cache_entry(updated
, ce
);
1561 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1562 fill_stat_cache_info(istate
, updated
, &st
);
1564 * If ignore_valid is not set, we should leave CE_VALID bit
1565 * alone. Otherwise, paths marked with --no-assume-unchanged
1566 * (i.e. things to be edited) will reacquire CE_VALID bit
1567 * automatically, which is not really what we want.
1569 if (!ignore_valid
&& assume_unchanged
&&
1570 !(ce
->ce_flags
& CE_VALID
))
1571 updated
->ce_flags
&= ~CE_VALID
;
1573 /* istate->cache_changed is updated in the caller */
1577 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1578 int * first
, const char *header_msg
)
1580 if (in_porcelain
&& *first
&& header_msg
) {
1581 printf("%s\n", header_msg
);
1587 int repo_refresh_and_write_index(struct repository
*repo
,
1588 unsigned int refresh_flags
,
1589 unsigned int write_flags
,
1591 const struct pathspec
*pathspec
,
1592 char *seen
, const char *header_msg
)
1594 struct lock_file lock_file
= LOCK_INIT
;
1597 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1598 if (!gentle
&& fd
< 0)
1600 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1602 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1608 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1609 const struct pathspec
*pathspec
,
1610 char *seen
, const char *header_msg
)
1614 int really
= (flags
& REFRESH_REALLY
) != 0;
1615 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1616 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1617 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1618 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1619 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1621 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1622 unsigned int options
= (CE_MATCH_REFRESH
|
1623 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1624 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1625 const char *modified_fmt
;
1626 const char *deleted_fmt
;
1627 const char *typechange_fmt
;
1628 const char *added_fmt
;
1629 const char *unmerged_fmt
;
1630 struct progress
*progress
= NULL
;
1631 int t2_sum_lstat
= 0;
1632 int t2_sum_scan
= 0;
1634 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1635 progress
= start_delayed_progress(_("Refresh index"),
1638 trace_performance_enter();
1639 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1640 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1641 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1642 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1643 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1645 * Use the multi-threaded preload_index() to refresh most of the
1646 * cache entries quickly then in the single threaded loop below,
1647 * we only have to do the special cases that are left.
1649 preload_index(istate
, pathspec
, 0);
1650 trace2_region_enter("index", "refresh", NULL
);
1652 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1653 struct cache_entry
*ce
, *new_entry
;
1654 int cache_errno
= 0;
1657 int t2_did_lstat
= 0;
1658 int t2_did_scan
= 0;
1660 ce
= istate
->cache
[i
];
1661 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1663 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1667 * If this entry is a sparse directory, then there isn't
1668 * any stat() information to update. Ignore the entry.
1670 if (S_ISSPARSEDIR(ce
->ce_mode
))
1673 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1677 while ((i
< istate
->cache_nr
) &&
1678 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1684 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1685 &first
, header_msg
);
1693 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1694 &cache_errno
, &changed
,
1695 &t2_did_lstat
, &t2_did_scan
);
1696 t2_sum_lstat
+= t2_did_lstat
;
1697 t2_sum_scan
+= t2_did_scan
;
1698 if (new_entry
== ce
)
1700 display_progress(progress
, i
);
1704 if (really
&& cache_errno
== EINVAL
) {
1705 /* If we are doing --really-refresh that
1706 * means the index is not valid anymore.
1708 ce
->ce_flags
&= ~CE_VALID
;
1709 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1710 mark_fsmonitor_invalid(istate
, ce
);
1711 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1716 if (cache_errno
== ENOENT
)
1718 else if (ce_intent_to_add(ce
))
1719 fmt
= added_fmt
; /* must be before other checks */
1720 else if (changed
& TYPE_CHANGED
)
1721 fmt
= typechange_fmt
;
1725 ce
->name
, in_porcelain
, &first
, header_msg
);
1730 replace_index_entry(istate
, i
, new_entry
);
1732 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1733 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1734 trace2_region_leave("index", "refresh", NULL
);
1735 display_progress(progress
, istate
->cache_nr
);
1736 stop_progress(&progress
);
1737 trace_performance_leave("refresh index");
1741 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1742 struct cache_entry
*ce
,
1743 unsigned int options
)
1745 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1749 /*****************************************************************
1751 *****************************************************************/
1753 #define INDEX_FORMAT_DEFAULT 3
1755 static unsigned int get_index_format_default(struct repository
*r
)
1757 char *envversion
= getenv("GIT_INDEX_VERSION");
1759 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1762 prepare_repo_settings(r
);
1764 if (r
->settings
.index_version
>= 0)
1765 version
= r
->settings
.index_version
;
1766 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1767 warning(_("index.version set, but the value is invalid.\n"
1768 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1769 return INDEX_FORMAT_DEFAULT
;
1774 version
= strtoul(envversion
, &endp
, 10);
1776 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1777 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1778 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1779 version
= INDEX_FORMAT_DEFAULT
;
1785 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1786 * Again - this is just a (very strong in practice) heuristic that
1787 * the inode hasn't changed.
1789 * We save the fields in big-endian order to allow using the
1790 * index file over NFS transparently.
1792 struct ondisk_cache_entry
{
1793 struct cache_time ctime
;
1794 struct cache_time mtime
;
1802 * unsigned char hash[hashsz];
1804 * if (flags & CE_EXTENDED)
1807 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1808 char name
[FLEX_ARRAY
];
1811 /* These are only used for v3 or lower */
1812 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1813 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1814 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1815 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1816 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1817 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1818 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1820 /* Allow fsck to force verification of the index checksum. */
1821 int verify_index_checksum
;
1823 /* Allow fsck to force verification of the cache entry order. */
1824 int verify_ce_order
;
1826 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1829 unsigned char hash
[GIT_MAX_RAWSZ
];
1831 unsigned char *start
, *end
;
1832 struct object_id oid
;
1834 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1835 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1836 hdr_version
= ntohl(hdr
->hdr_version
);
1837 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1838 return error(_("bad index version %d"), hdr_version
);
1840 if (!verify_index_checksum
)
1843 end
= (unsigned char *)hdr
+ size
;
1844 start
= end
- the_hash_algo
->rawsz
;
1845 oidread(&oid
, start
);
1846 if (oideq(&oid
, null_oid()))
1849 the_hash_algo
->init_fn(&c
);
1850 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1851 the_hash_algo
->final_fn(hash
, &c
);
1852 if (!hasheq(hash
, start
))
1853 return error(_("bad index file sha1 signature"));
1857 static int read_index_extension(struct index_state
*istate
,
1858 const char *ext
, const char *data
, unsigned long sz
)
1860 switch (CACHE_EXT(ext
)) {
1861 case CACHE_EXT_TREE
:
1862 istate
->cache_tree
= cache_tree_read(data
, sz
);
1864 case CACHE_EXT_RESOLVE_UNDO
:
1865 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1867 case CACHE_EXT_LINK
:
1868 if (read_link_extension(istate
, data
, sz
))
1871 case CACHE_EXT_UNTRACKED
:
1872 istate
->untracked
= read_untracked_extension(data
, sz
);
1874 case CACHE_EXT_FSMONITOR
:
1875 read_fsmonitor_extension(istate
, data
, sz
);
1877 case CACHE_EXT_ENDOFINDEXENTRIES
:
1878 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1879 /* already handled in do_read_index() */
1881 case CACHE_EXT_SPARSE_DIRECTORIES
:
1882 /* no content, only an indicator */
1883 istate
->sparse_index
= INDEX_COLLAPSED
;
1886 if (*ext
< 'A' || 'Z' < *ext
)
1887 return error(_("index uses %.4s extension, which we do not understand"),
1889 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1896 * Parses the contents of the cache entry contained within the 'ondisk' buffer
1897 * into a new incore 'cache_entry'.
1899 * Note that 'char *ondisk' may not be aligned to a 4-byte address interval in
1900 * index v4, so we cannot cast it to 'struct ondisk_cache_entry *' and access
1901 * its members. Instead, we use the byte offsets of members within the struct to
1902 * identify where 'get_be16()', 'get_be32()', and 'oidread()' (which can all
1903 * read from an unaligned memory buffer) should read from the 'ondisk' buffer
1904 * into the corresponding incore 'cache_entry' members.
1906 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1907 unsigned int version
,
1909 unsigned long *ent_size
,
1910 const struct cache_entry
*previous_ce
)
1912 struct cache_entry
*ce
;
1915 const unsigned hashsz
= the_hash_algo
->rawsz
;
1916 const char *flagsp
= ondisk
+ offsetof(struct ondisk_cache_entry
, data
) + hashsz
;
1918 size_t copy_len
= 0;
1920 * Adjacent cache entries tend to share the leading paths, so it makes
1921 * sense to only store the differences in later entries. In the v4
1922 * on-disk format of the index, each on-disk cache entry stores the
1923 * number of bytes to be stripped from the end of the previous name,
1924 * and the bytes to append to the result, to come up with its name.
1926 int expand_name_field
= version
== 4;
1928 /* On-disk flags are just 16 bits */
1929 flags
= get_be16(flagsp
);
1930 len
= flags
& CE_NAMEMASK
;
1932 if (flags
& CE_EXTENDED
) {
1934 extended_flags
= get_be16(flagsp
+ sizeof(uint16_t)) << 16;
1935 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1936 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1937 die(_("unknown index entry format 0x%08x"), extended_flags
);
1938 flags
|= extended_flags
;
1939 name
= (const char *)(flagsp
+ 2 * sizeof(uint16_t));
1942 name
= (const char *)(flagsp
+ sizeof(uint16_t));
1944 if (expand_name_field
) {
1945 const unsigned char *cp
= (const unsigned char *)name
;
1946 size_t strip_len
, previous_len
;
1948 /* If we're at the beginning of a block, ignore the previous name */
1949 strip_len
= decode_varint(&cp
);
1951 previous_len
= previous_ce
->ce_namelen
;
1952 if (previous_len
< strip_len
)
1953 die(_("malformed name field in the index, near path '%s'"),
1955 copy_len
= previous_len
- strip_len
;
1957 name
= (const char *)cp
;
1960 if (len
== CE_NAMEMASK
) {
1962 if (expand_name_field
)
1966 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1969 * NEEDSWORK: using 'offsetof()' is cumbersome and should be replaced
1970 * with something more akin to 'load_bitmap_entries_v1()'s use of
1971 * 'read_be16'/'read_be32'. For consistency with the corresponding
1972 * ondisk entry write function ('copy_cache_entry_to_ondisk()'), this
1973 * should be done at the same time as removing references to
1974 * 'ondisk_cache_entry' there.
1976 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1977 + offsetof(struct cache_time
, sec
));
1978 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1979 + offsetof(struct cache_time
, sec
));
1980 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1981 + offsetof(struct cache_time
, nsec
));
1982 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1983 + offsetof(struct cache_time
, nsec
));
1984 ce
->ce_stat_data
.sd_dev
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, dev
));
1985 ce
->ce_stat_data
.sd_ino
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ino
));
1986 ce
->ce_mode
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mode
));
1987 ce
->ce_stat_data
.sd_uid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, uid
));
1988 ce
->ce_stat_data
.sd_gid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, gid
));
1989 ce
->ce_stat_data
.sd_size
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, size
));
1990 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1991 ce
->ce_namelen
= len
;
1993 oidread(&ce
->oid
, (const unsigned char *)ondisk
+ offsetof(struct ondisk_cache_entry
, data
));
1995 if (expand_name_field
) {
1997 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1998 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1999 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
2001 memcpy(ce
->name
, name
, len
+ 1);
2002 *ent_size
= ondisk_ce_size(ce
);
2007 static void check_ce_order(struct index_state
*istate
)
2011 if (!verify_ce_order
)
2014 for (i
= 1; i
< istate
->cache_nr
; i
++) {
2015 struct cache_entry
*ce
= istate
->cache
[i
- 1];
2016 struct cache_entry
*next_ce
= istate
->cache
[i
];
2017 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
2019 if (0 < name_compare
)
2020 die(_("unordered stage entries in index"));
2021 if (!name_compare
) {
2023 die(_("multiple stage entries for merged file '%s'"),
2025 if (ce_stage(ce
) > ce_stage(next_ce
))
2026 die(_("unordered stage entries for '%s'"),
2032 static void tweak_untracked_cache(struct index_state
*istate
)
2034 struct repository
*r
= the_repository
;
2036 prepare_repo_settings(r
);
2038 switch (r
->settings
.core_untracked_cache
) {
2039 case UNTRACKED_CACHE_REMOVE
:
2040 remove_untracked_cache(istate
);
2042 case UNTRACKED_CACHE_WRITE
:
2043 add_untracked_cache(istate
);
2045 case UNTRACKED_CACHE_KEEP
:
2047 * Either an explicit "core.untrackedCache=keep", the
2048 * default if "core.untrackedCache" isn't configured,
2049 * or a fallback on an unknown "core.untrackedCache"
2056 static void tweak_split_index(struct index_state
*istate
)
2058 switch (git_config_get_split_index()) {
2059 case -1: /* unset: do nothing */
2062 remove_split_index(istate
);
2065 add_split_index(istate
);
2067 default: /* unknown value: do nothing */
2072 static void post_read_index_from(struct index_state
*istate
)
2074 check_ce_order(istate
);
2075 tweak_untracked_cache(istate
);
2076 tweak_split_index(istate
);
2077 tweak_fsmonitor(istate
);
2080 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
2082 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
2085 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
2087 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
2090 * Account for potential alignment differences.
2092 per_entry
+= align_padding_size(per_entry
, 0);
2093 return ondisk_size
+ entries
* per_entry
;
2096 struct index_entry_offset
2098 /* starting byte offset into index file, count of index entries in this block */
2102 struct index_entry_offset_table
2105 struct index_entry_offset entries
[FLEX_ARRAY
];
2108 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2109 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2111 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2112 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2114 struct load_index_extensions
2117 struct index_state
*istate
;
2120 unsigned long src_offset
;
2123 static void *load_index_extensions(void *_data
)
2125 struct load_index_extensions
*p
= _data
;
2126 unsigned long src_offset
= p
->src_offset
;
2128 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2129 /* After an array of active_nr index entries,
2130 * there can be arbitrary number of extended
2131 * sections, each of which is prefixed with
2132 * extension name (4-byte) and section length
2133 * in 4-byte network byte order.
2135 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2136 if (read_index_extension(p
->istate
,
2137 p
->mmap
+ src_offset
,
2138 p
->mmap
+ src_offset
+ 8,
2140 munmap((void *)p
->mmap
, p
->mmap_size
);
2141 die(_("index file corrupt"));
2144 src_offset
+= extsize
;
2151 * A helper function that will load the specified range of cache entries
2152 * from the memory mapped file and add them to the given index.
2154 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2155 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2156 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2159 unsigned long src_offset
= start_offset
;
2161 for (i
= offset
; i
< offset
+ nr
; i
++) {
2162 struct cache_entry
*ce
;
2163 unsigned long consumed
;
2165 ce
= create_from_disk(ce_mem_pool
, istate
->version
,
2167 &consumed
, previous_ce
);
2168 set_index_entry(istate
, i
, ce
);
2170 src_offset
+= consumed
;
2173 return src_offset
- start_offset
;
2176 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2177 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2179 unsigned long consumed
;
2181 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2182 if (istate
->version
== 4) {
2183 mem_pool_init(istate
->ce_mem_pool
,
2184 estimate_cache_size_from_compressed(istate
->cache_nr
));
2186 mem_pool_init(istate
->ce_mem_pool
,
2187 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2190 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2191 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2196 * Mostly randomly chosen maximum thread counts: we
2197 * cap the parallelism to online_cpus() threads, and we want
2198 * to have at least 10000 cache entries per thread for it to
2199 * be worth starting a thread.
2202 #define THREAD_COST (10000)
2204 struct load_cache_entries_thread_data
2207 struct index_state
*istate
;
2208 struct mem_pool
*ce_mem_pool
;
2211 struct index_entry_offset_table
*ieot
;
2212 int ieot_start
; /* starting index into the ieot array */
2213 int ieot_blocks
; /* count of ieot entries to process */
2214 unsigned long consumed
; /* return # of bytes in index file processed */
2218 * A thread proc to run the load_cache_entries() computation
2219 * across multiple background threads.
2221 static void *load_cache_entries_thread(void *_data
)
2223 struct load_cache_entries_thread_data
*p
= _data
;
2226 /* iterate across all ieot blocks assigned to this thread */
2227 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2228 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2229 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2230 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2235 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2236 int nr_threads
, struct index_entry_offset_table
*ieot
)
2238 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2239 struct load_cache_entries_thread_data
*data
;
2240 unsigned long consumed
= 0;
2242 /* a little sanity checking */
2243 if (istate
->name_hash_initialized
)
2244 BUG("the name hash isn't thread safe");
2246 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2247 mem_pool_init(istate
->ce_mem_pool
, 0);
2249 /* ensure we have no more threads than we have blocks to process */
2250 if (nr_threads
> ieot
->nr
)
2251 nr_threads
= ieot
->nr
;
2252 CALLOC_ARRAY(data
, nr_threads
);
2254 offset
= ieot_start
= 0;
2255 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2256 for (i
= 0; i
< nr_threads
; i
++) {
2257 struct load_cache_entries_thread_data
*p
= &data
[i
];
2260 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2261 ieot_blocks
= ieot
->nr
- ieot_start
;
2267 p
->ieot_start
= ieot_start
;
2268 p
->ieot_blocks
= ieot_blocks
;
2270 /* create a mem_pool for each thread */
2272 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2273 nr
+= p
->ieot
->entries
[j
].nr
;
2274 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2275 if (istate
->version
== 4) {
2276 mem_pool_init(p
->ce_mem_pool
,
2277 estimate_cache_size_from_compressed(nr
));
2279 mem_pool_init(p
->ce_mem_pool
,
2280 estimate_cache_size(mmap_size
, nr
));
2283 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2285 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2287 /* increment by the number of cache entries in the ieot block being processed */
2288 for (j
= 0; j
< ieot_blocks
; j
++)
2289 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2290 ieot_start
+= ieot_blocks
;
2293 for (i
= 0; i
< nr_threads
; i
++) {
2294 struct load_cache_entries_thread_data
*p
= &data
[i
];
2296 err
= pthread_join(p
->pthread
, NULL
);
2298 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2299 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2300 consumed
+= p
->consumed
;
2308 static void set_new_index_sparsity(struct index_state
*istate
)
2311 * If the index's repo exists, mark it sparse according to
2314 prepare_repo_settings(istate
->repo
);
2315 if (!istate
->repo
->settings
.command_requires_full_index
&&
2316 is_sparse_index_allowed(istate
, 0))
2317 istate
->sparse_index
= 1;
2320 /* remember to discard_cache() before reading a different cache! */
2321 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2325 unsigned long src_offset
;
2326 const struct cache_header
*hdr
;
2329 struct load_index_extensions p
;
2330 size_t extension_offset
= 0;
2331 int nr_threads
, cpus
;
2332 struct index_entry_offset_table
*ieot
= NULL
;
2334 if (istate
->initialized
)
2335 return istate
->cache_nr
;
2337 istate
->timestamp
.sec
= 0;
2338 istate
->timestamp
.nsec
= 0;
2339 fd
= open(path
, O_RDONLY
);
2341 if (!must_exist
&& errno
== ENOENT
) {
2342 set_new_index_sparsity(istate
);
2345 die_errno(_("%s: index file open failed"), path
);
2349 die_errno(_("%s: cannot stat the open index"), path
);
2351 mmap_size
= xsize_t(st
.st_size
);
2352 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2353 die(_("%s: index file smaller than expected"), path
);
2355 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2356 if (mmap
== MAP_FAILED
)
2357 die_errno(_("%s: unable to map index file%s"), path
,
2361 hdr
= (const struct cache_header
*)mmap
;
2362 if (verify_hdr(hdr
, mmap_size
) < 0)
2365 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2366 istate
->version
= ntohl(hdr
->hdr_version
);
2367 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2368 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2369 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2370 istate
->initialized
= 1;
2374 p
.mmap_size
= mmap_size
;
2376 src_offset
= sizeof(*hdr
);
2378 if (git_config_get_index_threads(&nr_threads
))
2381 /* TODO: does creating more threads than cores help? */
2383 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2384 cpus
= online_cpus();
2385 if (nr_threads
> cpus
)
2392 if (nr_threads
> 1) {
2393 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2394 if (extension_offset
) {
2397 p
.src_offset
= extension_offset
;
2398 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2400 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2407 * Locate and read the index entry offset table so that we can use it
2408 * to multi-thread the reading of the cache entries.
2410 if (extension_offset
&& nr_threads
> 1)
2411 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2414 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2417 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2420 istate
->timestamp
.sec
= st
.st_mtime
;
2421 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2423 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2424 if (extension_offset
) {
2425 int ret
= pthread_join(p
.pthread
, NULL
);
2427 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2429 p
.src_offset
= src_offset
;
2430 load_index_extensions(&p
);
2432 munmap((void *)mmap
, mmap_size
);
2435 * TODO trace2: replace "the_repository" with the actual repo instance
2436 * that is associated with the given "istate".
2438 trace2_data_intmax("index", the_repository
, "read/version",
2440 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2444 * If the command explicitly requires a full index, force it
2445 * to be full. Otherwise, correct the sparsity based on repository
2446 * settings and other properties of the index (if necessary).
2448 prepare_repo_settings(istate
->repo
);
2449 if (istate
->repo
->settings
.command_requires_full_index
)
2450 ensure_full_index(istate
);
2452 ensure_correct_sparsity(istate
);
2454 return istate
->cache_nr
;
2457 munmap((void *)mmap
, mmap_size
);
2458 die(_("index file corrupt"));
2462 * Signal that the shared index is used by updating its mtime.
2464 * This way, shared index can be removed if they have not been used
2467 static void freshen_shared_index(const char *shared_index
, int warn
)
2469 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2470 warning(_("could not freshen shared index '%s'"), shared_index
);
2473 int read_index_from(struct index_state
*istate
, const char *path
,
2476 struct split_index
*split_index
;
2481 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2482 if (istate
->initialized
)
2483 return istate
->cache_nr
;
2486 * TODO trace2: replace "the_repository" with the actual repo instance
2487 * that is associated with the given "istate".
2489 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2491 trace_performance_enter();
2492 ret
= do_read_index(istate
, path
, 0);
2493 trace_performance_leave("read cache %s", path
);
2494 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2497 split_index
= istate
->split_index
;
2498 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2499 post_read_index_from(istate
);
2503 trace_performance_enter();
2504 if (split_index
->base
)
2505 release_index(split_index
->base
);
2507 ALLOC_ARRAY(split_index
->base
, 1);
2508 index_state_init(split_index
->base
, istate
->repo
);
2510 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2511 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2512 trace2_region_enter_printf("index", "shared/do_read_index",
2513 the_repository
, "%s", base_path
);
2514 ret
= do_read_index(split_index
->base
, base_path
, 0);
2515 trace2_region_leave_printf("index", "shared/do_read_index",
2516 the_repository
, "%s", base_path
);
2518 char *path_copy
= xstrdup(path
);
2519 char *base_path2
= xstrfmt("%s/sharedindex.%s",
2520 dirname(path_copy
), base_oid_hex
);
2522 trace2_region_enter_printf("index", "shared/do_read_index",
2523 the_repository
, "%s", base_path2
);
2524 ret
= do_read_index(split_index
->base
, base_path2
, 1);
2525 trace2_region_leave_printf("index", "shared/do_read_index",
2526 the_repository
, "%s", base_path2
);
2529 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2530 die(_("broken index, expect %s in %s, got %s"),
2531 base_oid_hex
, base_path
,
2532 oid_to_hex(&split_index
->base
->oid
));
2534 freshen_shared_index(base_path
, 0);
2535 merge_base_index(istate
);
2536 post_read_index_from(istate
);
2537 trace_performance_leave("read cache %s", base_path
);
2542 int is_index_unborn(struct index_state
*istate
)
2544 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2547 void index_state_init(struct index_state
*istate
, struct repository
*r
)
2549 struct index_state blank
= INDEX_STATE_INIT(r
);
2550 memcpy(istate
, &blank
, sizeof(*istate
));
2553 void release_index(struct index_state
*istate
)
2556 * Cache entries in istate->cache[] should have been allocated
2557 * from the memory pool associated with this index, or from an
2558 * associated split_index. There is no need to free individual
2559 * cache entries. validate_cache_entries can detect when this
2560 * assertion does not hold.
2562 validate_cache_entries(istate
);
2564 resolve_undo_clear_index(istate
);
2565 free_name_hash(istate
);
2566 cache_tree_free(&(istate
->cache_tree
));
2567 free(istate
->fsmonitor_last_update
);
2568 free(istate
->cache
);
2569 discard_split_index(istate
);
2570 free_untracked_cache(istate
->untracked
);
2572 if (istate
->sparse_checkout_patterns
) {
2573 clear_pattern_list(istate
->sparse_checkout_patterns
);
2574 FREE_AND_NULL(istate
->sparse_checkout_patterns
);
2577 if (istate
->ce_mem_pool
) {
2578 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2579 FREE_AND_NULL(istate
->ce_mem_pool
);
2583 void discard_index(struct index_state
*istate
)
2585 release_index(istate
);
2586 index_state_init(istate
, istate
->repo
);
2590 * Validate the cache entries of this index.
2591 * All cache entries associated with this index
2592 * should have been allocated by the memory pool
2593 * associated with this index, or by a referenced
2596 void validate_cache_entries(const struct index_state
*istate
)
2600 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2603 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2605 BUG("cache entry is not allocated from expected memory pool");
2606 } else if (!istate
->ce_mem_pool
||
2607 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2608 if (!istate
->split_index
||
2609 !istate
->split_index
->base
||
2610 !istate
->split_index
->base
->ce_mem_pool
||
2611 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2612 BUG("cache entry is not allocated from expected memory pool");
2617 if (istate
->split_index
)
2618 validate_cache_entries(istate
->split_index
->base
);
2621 int unmerged_index(const struct index_state
*istate
)
2624 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2625 if (ce_stage(istate
->cache
[i
]))
2631 int repo_index_has_changes(struct repository
*repo
,
2635 struct index_state
*istate
= repo
->index
;
2636 struct object_id cmp
;
2640 cmp
= tree
->object
.oid
;
2641 if (tree
|| !repo_get_oid_tree(repo
, "HEAD", &cmp
)) {
2642 struct diff_options opt
;
2644 repo_diff_setup(repo
, &opt
);
2645 opt
.flags
.exit_with_status
= 1;
2647 opt
.flags
.quick
= 1;
2648 diff_setup_done(&opt
);
2649 do_diff_cache(&cmp
, &opt
);
2651 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2653 strbuf_addch(sb
, ' ');
2654 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2657 return opt
.flags
.has_changes
!= 0;
2659 /* TODO: audit for interaction with sparse-index. */
2660 ensure_full_index(istate
);
2661 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2663 strbuf_addch(sb
, ' ');
2664 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2666 return !!istate
->cache_nr
;
2670 static int write_index_ext_header(struct hashfile
*f
,
2671 git_hash_ctx
*eoie_f
,
2675 hashwrite_be32(f
, ext
);
2676 hashwrite_be32(f
, sz
);
2681 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2682 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2687 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2688 struct cache_entry
*ce
)
2691 * The only thing we care about in this function is to smudge the
2692 * falsely clean entry due to touch-update-touch race, so we leave
2693 * everything else as they are. We are called for entries whose
2694 * ce_stat_data.sd_mtime match the index file mtime.
2696 * Note that this actually does not do much for gitlinks, for
2697 * which ce_match_stat_basic() always goes to the actual
2698 * contents. The caller checks with is_racy_timestamp() which
2699 * always says "no" for gitlinks, so we are not called for them ;-)
2703 if (lstat(ce
->name
, &st
) < 0)
2705 if (ce_match_stat_basic(ce
, &st
))
2707 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2708 /* This is "racily clean"; smudge it. Note that this
2709 * is a tricky code. At first glance, it may appear
2710 * that it can break with this sequence:
2712 * $ echo xyzzy >frotz
2713 * $ git-update-index --add frotz
2716 * $ echo filfre >nitfol
2717 * $ git-update-index --add nitfol
2719 * but it does not. When the second update-index runs,
2720 * it notices that the entry "frotz" has the same timestamp
2721 * as index, and if we were to smudge it by resetting its
2722 * size to zero here, then the object name recorded
2723 * in index is the 6-byte file but the cached stat information
2724 * becomes zero --- which would then match what we would
2725 * obtain from the filesystem next time we stat("frotz").
2727 * However, the second update-index, before calling
2728 * this function, notices that the cached size is 6
2729 * bytes and what is on the filesystem is an empty
2730 * file, and never calls us, so the cached size information
2731 * for "frotz" stays 6 which does not match the filesystem.
2733 ce
->ce_stat_data
.sd_size
= 0;
2737 /* Copy miscellaneous fields but not the name */
2738 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2739 struct cache_entry
*ce
)
2742 const unsigned hashsz
= the_hash_algo
->rawsz
;
2743 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2745 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2746 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2747 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2748 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2749 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2750 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2751 ondisk
->mode
= htonl(ce
->ce_mode
);
2752 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2753 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2754 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2755 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2757 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2758 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2759 flagsp
[0] = htons(flags
);
2760 if (ce
->ce_flags
& CE_EXTENDED
) {
2761 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2765 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2766 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2769 unsigned int saved_namelen
;
2770 int stripped_name
= 0;
2771 static unsigned char padding
[8] = { 0x00 };
2773 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2774 saved_namelen
= ce_namelen(ce
);
2779 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2781 if (!previous_name
) {
2782 int len
= ce_namelen(ce
);
2783 copy_cache_entry_to_ondisk(ondisk
, ce
);
2784 hashwrite(f
, ondisk
, size
);
2785 hashwrite(f
, ce
->name
, len
);
2786 hashwrite(f
, padding
, align_padding_size(size
, len
));
2788 int common
, to_remove
, prefix_size
;
2789 unsigned char to_remove_vi
[16];
2791 (ce
->name
[common
] &&
2792 common
< previous_name
->len
&&
2793 ce
->name
[common
] == previous_name
->buf
[common
]);
2795 ; /* still matching */
2796 to_remove
= previous_name
->len
- common
;
2797 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2799 copy_cache_entry_to_ondisk(ondisk
, ce
);
2800 hashwrite(f
, ondisk
, size
);
2801 hashwrite(f
, to_remove_vi
, prefix_size
);
2802 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2803 hashwrite(f
, padding
, 1);
2805 strbuf_splice(previous_name
, common
, to_remove
,
2806 ce
->name
+ common
, ce_namelen(ce
) - common
);
2808 if (stripped_name
) {
2809 ce
->ce_namelen
= saved_namelen
;
2810 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2817 * This function verifies if index_state has the correct sha1 of the
2818 * index file. Don't die if we have any other failure, just return 0.
2820 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2825 unsigned char hash
[GIT_MAX_RAWSZ
];
2827 if (!istate
->initialized
)
2830 fd
= open(path
, O_RDONLY
);
2837 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2840 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2841 if (n
!= the_hash_algo
->rawsz
)
2844 if (!hasheq(istate
->oid
.hash
, hash
))
2855 static int repo_verify_index(struct repository
*repo
)
2857 return verify_index_from(repo
->index
, repo
->index_file
);
2860 int has_racy_timestamp(struct index_state
*istate
)
2862 int entries
= istate
->cache_nr
;
2865 for (i
= 0; i
< entries
; i
++) {
2866 struct cache_entry
*ce
= istate
->cache
[i
];
2867 if (is_racy_timestamp(istate
, ce
))
2873 void repo_update_index_if_able(struct repository
*repo
,
2874 struct lock_file
*lockfile
)
2876 if ((repo
->index
->cache_changed
||
2877 has_racy_timestamp(repo
->index
)) &&
2878 repo_verify_index(repo
))
2879 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2881 rollback_lock_file(lockfile
);
2884 static int record_eoie(void)
2888 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2892 * As a convenience, the end of index entries extension
2893 * used for threading is written by default if the user
2894 * explicitly requested threaded index reads.
2896 return !git_config_get_index_threads(&val
) && val
!= 1;
2899 static int record_ieot(void)
2903 if (!git_config_get_bool("index.recordoffsettable", &val
))
2907 * As a convenience, the offset table used for threading is
2908 * written by default if the user explicitly requested
2909 * threaded index reads.
2911 return !git_config_get_index_threads(&val
) && val
!= 1;
2915 * On success, `tempfile` is closed. If it is the temporary file
2916 * of a `struct lock_file`, we will therefore effectively perform
2917 * a 'close_lock_file_gently()`. Since that is an implementation
2918 * detail of lockfiles, callers of `do_write_index()` should not
2921 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2922 int strip_extensions
, unsigned flags
)
2924 uint64_t start
= getnanotime();
2926 git_hash_ctx
*eoie_c
= NULL
;
2927 struct cache_header hdr
;
2928 int i
, err
= 0, removed
, extended
, hdr_version
;
2929 struct cache_entry
**cache
= istate
->cache
;
2930 int entries
= istate
->cache_nr
;
2932 struct ondisk_cache_entry ondisk
;
2933 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2934 int drop_cache_tree
= istate
->drop_cache_tree
;
2936 int csum_fsync_flag
;
2937 int ieot_entries
= 1;
2938 struct index_entry_offset_table
*ieot
= NULL
;
2940 struct repository
*r
= istate
->repo
;
2942 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2944 prepare_repo_settings(r
);
2945 f
->skip_hash
= r
->settings
.index_skip_hash
;
2947 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2948 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2951 /* reduce extended entries if possible */
2952 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2953 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2955 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2959 if (!istate
->version
)
2960 istate
->version
= get_index_format_default(r
);
2962 /* demote version 3 to version 2 when the latter suffices */
2963 if (istate
->version
== 3 || istate
->version
== 2)
2964 istate
->version
= extended
? 3 : 2;
2966 hdr_version
= istate
->version
;
2968 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2969 hdr
.hdr_version
= htonl(hdr_version
);
2970 hdr
.hdr_entries
= htonl(entries
- removed
);
2972 hashwrite(f
, &hdr
, sizeof(hdr
));
2974 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2977 if (nr_threads
!= 1 && record_ieot()) {
2978 int ieot_blocks
, cpus
;
2981 * ensure default number of ieot blocks maps evenly to the
2982 * default number of threads that will process them leaving
2983 * room for the thread to load the index extensions.
2986 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2987 cpus
= online_cpus();
2988 if (ieot_blocks
> cpus
- 1)
2989 ieot_blocks
= cpus
- 1;
2991 ieot_blocks
= nr_threads
;
2992 if (ieot_blocks
> istate
->cache_nr
)
2993 ieot_blocks
= istate
->cache_nr
;
2997 * no reason to write out the IEOT extension if we don't
2998 * have enough blocks to utilize multi-threading
3000 if (ieot_blocks
> 1) {
3001 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
3002 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
3003 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
3007 offset
= hashfile_total(f
);
3010 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
3012 for (i
= 0; i
< entries
; i
++) {
3013 struct cache_entry
*ce
= cache
[i
];
3014 if (ce
->ce_flags
& CE_REMOVE
)
3016 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
3017 ce_smudge_racily_clean_entry(istate
, ce
);
3018 if (is_null_oid(&ce
->oid
)) {
3019 static const char msg
[] = "cache entry has null sha1: %s";
3020 static int allow
= -1;
3023 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
3025 warning(msg
, ce
->name
);
3027 err
= error(msg
, ce
->name
);
3029 drop_cache_tree
= 1;
3031 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
3032 ieot
->entries
[ieot
->nr
].nr
= nr
;
3033 ieot
->entries
[ieot
->nr
].offset
= offset
;
3036 * If we have a V4 index, set the first byte to an invalid
3037 * character to ensure there is nothing common with the previous
3041 previous_name
->buf
[0] = 0;
3044 offset
= hashfile_total(f
);
3046 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
3054 ieot
->entries
[ieot
->nr
].nr
= nr
;
3055 ieot
->entries
[ieot
->nr
].offset
= offset
;
3058 strbuf_release(&previous_name_buf
);
3065 offset
= hashfile_total(f
);
3068 * The extension headers must be hashed on their own for the
3069 * EOIE extension. Create a hashfile here to compute that hash.
3071 if (offset
&& record_eoie()) {
3072 CALLOC_ARRAY(eoie_c
, 1);
3073 the_hash_algo
->init_fn(eoie_c
);
3077 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
3078 * can minimize the number of extensions we have to scan through to
3079 * find it during load. Write it out regardless of the
3080 * strip_extensions parameter as we need it when loading the shared
3084 struct strbuf sb
= STRBUF_INIT
;
3086 write_ieot_extension(&sb
, ieot
);
3087 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
3088 hashwrite(f
, sb
.buf
, sb
.len
);
3089 strbuf_release(&sb
);
3095 if (!strip_extensions
&& istate
->split_index
&&
3096 !is_null_oid(&istate
->split_index
->base_oid
)) {
3097 struct strbuf sb
= STRBUF_INIT
;
3099 if (istate
->sparse_index
)
3100 die(_("cannot write split index for a sparse index"));
3102 err
= write_link_extension(&sb
, istate
) < 0 ||
3103 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
3105 hashwrite(f
, sb
.buf
, sb
.len
);
3106 strbuf_release(&sb
);
3110 if (!strip_extensions
&& !drop_cache_tree
&& istate
->cache_tree
) {
3111 struct strbuf sb
= STRBUF_INIT
;
3113 cache_tree_write(&sb
, istate
->cache_tree
);
3114 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
3115 hashwrite(f
, sb
.buf
, sb
.len
);
3116 strbuf_release(&sb
);
3120 if (!strip_extensions
&& istate
->resolve_undo
) {
3121 struct strbuf sb
= STRBUF_INIT
;
3123 resolve_undo_write(&sb
, istate
->resolve_undo
);
3124 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3126 hashwrite(f
, sb
.buf
, sb
.len
);
3127 strbuf_release(&sb
);
3131 if (!strip_extensions
&& istate
->untracked
) {
3132 struct strbuf sb
= STRBUF_INIT
;
3134 write_untracked_extension(&sb
, istate
->untracked
);
3135 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3137 hashwrite(f
, sb
.buf
, sb
.len
);
3138 strbuf_release(&sb
);
3142 if (!strip_extensions
&& istate
->fsmonitor_last_update
) {
3143 struct strbuf sb
= STRBUF_INIT
;
3145 write_fsmonitor_extension(&sb
, istate
);
3146 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3147 hashwrite(f
, sb
.buf
, sb
.len
);
3148 strbuf_release(&sb
);
3152 if (istate
->sparse_index
) {
3153 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3158 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3159 * so that it can be found and processed before all the index entries are
3160 * read. Write it out regardless of the strip_extensions parameter as we need it
3161 * when loading the shared index.
3164 struct strbuf sb
= STRBUF_INIT
;
3166 write_eoie_extension(&sb
, eoie_c
, offset
);
3167 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3168 hashwrite(f
, sb
.buf
, sb
.len
);
3169 strbuf_release(&sb
);
3174 csum_fsync_flag
= 0;
3175 if (!alternate_index_output
&& (flags
& COMMIT_LOCK
))
3176 csum_fsync_flag
= CSUM_FSYNC
;
3178 finalize_hashfile(f
, istate
->oid
.hash
, FSYNC_COMPONENT_INDEX
,
3179 CSUM_HASH_IN_STREAM
| csum_fsync_flag
);
3181 if (close_tempfile_gently(tempfile
)) {
3182 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3185 if (stat(get_tempfile_path(tempfile
), &st
))
3187 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3188 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3189 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3192 * TODO trace2: replace "the_repository" with the actual repo instance
3193 * that is associated with the given "istate".
3195 trace2_data_intmax("index", the_repository
, "write/version",
3197 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3203 void set_alternate_index_output(const char *name
)
3205 alternate_index_output
= name
;
3208 static int commit_locked_index(struct lock_file
*lk
)
3210 if (alternate_index_output
)
3211 return commit_lock_file_to(lk
, alternate_index_output
);
3213 return commit_lock_file(lk
);
3216 static int do_write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3220 int was_full
= istate
->sparse_index
== INDEX_EXPANDED
;
3222 ret
= convert_to_sparse(istate
, 0);
3225 warning(_("failed to convert to a sparse-index"));
3230 * TODO trace2: replace "the_repository" with the actual repo instance
3231 * that is associated with the given "istate".
3233 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3234 "%s", get_lock_file_path(lock
));
3235 ret
= do_write_index(istate
, lock
->tempfile
, 0, flags
);
3236 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3237 "%s", get_lock_file_path(lock
));
3240 ensure_full_index(istate
);
3244 if (flags
& COMMIT_LOCK
)
3245 ret
= commit_locked_index(lock
);
3247 ret
= close_lock_file_gently(lock
);
3249 run_hooks_l("post-index-change",
3250 istate
->updated_workdir
? "1" : "0",
3251 istate
->updated_skipworktree
? "1" : "0", NULL
);
3252 istate
->updated_workdir
= 0;
3253 istate
->updated_skipworktree
= 0;
3258 static int write_split_index(struct index_state
*istate
,
3259 struct lock_file
*lock
,
3263 prepare_to_write_split_index(istate
);
3264 ret
= do_write_locked_index(istate
, lock
, flags
);
3265 finish_writing_split_index(istate
);
3269 static const char *shared_index_expire
= "2.weeks.ago";
3271 static unsigned long get_shared_index_expire_date(void)
3273 static unsigned long shared_index_expire_date
;
3274 static int shared_index_expire_date_prepared
;
3276 if (!shared_index_expire_date_prepared
) {
3277 git_config_get_expiry("splitindex.sharedindexexpire",
3278 &shared_index_expire
);
3279 shared_index_expire_date
= approxidate(shared_index_expire
);
3280 shared_index_expire_date_prepared
= 1;
3283 return shared_index_expire_date
;
3286 static int should_delete_shared_index(const char *shared_index_path
)
3289 unsigned long expiration
;
3291 /* Check timestamp */
3292 expiration
= get_shared_index_expire_date();
3295 if (stat(shared_index_path
, &st
))
3296 return error_errno(_("could not stat '%s'"), shared_index_path
);
3297 if (st
.st_mtime
> expiration
)
3303 static int clean_shared_index_files(const char *current_hex
)
3306 DIR *dir
= opendir(get_git_dir());
3309 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3311 while ((de
= readdir(dir
)) != NULL
) {
3312 const char *sha1_hex
;
3313 const char *shared_index_path
;
3314 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3316 if (!strcmp(sha1_hex
, current_hex
))
3318 shared_index_path
= git_path("%s", de
->d_name
);
3319 if (should_delete_shared_index(shared_index_path
) > 0 &&
3320 unlink(shared_index_path
))
3321 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3328 static int write_shared_index(struct index_state
*istate
,
3329 struct tempfile
**temp
, unsigned flags
)
3331 struct split_index
*si
= istate
->split_index
;
3332 int ret
, was_full
= !istate
->sparse_index
;
3334 move_cache_to_base_index(istate
);
3335 convert_to_sparse(istate
, 0);
3337 trace2_region_enter_printf("index", "shared/do_write_index",
3338 the_repository
, "%s", get_tempfile_path(*temp
));
3339 ret
= do_write_index(si
->base
, *temp
, 1, flags
);
3340 trace2_region_leave_printf("index", "shared/do_write_index",
3341 the_repository
, "%s", get_tempfile_path(*temp
));
3344 ensure_full_index(istate
);
3348 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3350 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3353 ret
= rename_tempfile(temp
,
3354 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3356 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3357 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3363 static const int default_max_percent_split_change
= 20;
3365 static int too_many_not_shared_entries(struct index_state
*istate
)
3367 int i
, not_shared
= 0;
3368 int max_split
= git_config_get_max_percent_split_change();
3370 switch (max_split
) {
3372 /* not or badly configured: use the default value */
3373 max_split
= default_max_percent_split_change
;
3376 return 1; /* 0% means always write a new shared index */
3378 return 0; /* 100% means never write a new shared index */
3380 break; /* just use the configured value */
3383 /* Count not shared entries */
3384 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3385 struct cache_entry
*ce
= istate
->cache
[i
];
3390 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3393 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3396 int new_shared_index
, ret
, test_split_index_env
;
3397 struct split_index
*si
= istate
->split_index
;
3399 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3400 cache_tree_verify(the_repository
, istate
);
3402 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3403 if (flags
& COMMIT_LOCK
)
3404 rollback_lock_file(lock
);
3408 if (istate
->fsmonitor_last_update
)
3409 fill_fsmonitor_bitmap(istate
);
3411 test_split_index_env
= git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3413 if ((!si
&& !test_split_index_env
) ||
3414 alternate_index_output
||
3415 (istate
->cache_changed
& ~EXTMASK
)) {
3417 oidclr(&si
->base_oid
);
3418 ret
= do_write_locked_index(istate
, lock
, flags
);
3422 if (test_split_index_env
) {
3424 si
= init_split_index(istate
);
3425 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3427 int v
= si
->base_oid
.hash
[0];
3429 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3432 if (too_many_not_shared_entries(istate
))
3433 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3435 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3437 if (new_shared_index
) {
3438 struct tempfile
*temp
;
3441 /* Same initial permissions as the main .git/index file */
3442 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3444 oidclr(&si
->base_oid
);
3445 ret
= do_write_locked_index(istate
, lock
, flags
);
3448 ret
= write_shared_index(istate
, &temp
, flags
);
3450 saved_errno
= errno
;
3451 if (is_tempfile_active(temp
))
3452 delete_tempfile(&temp
);
3453 errno
= saved_errno
;
3459 ret
= write_split_index(istate
, lock
, flags
);
3461 /* Freshen the shared index only if the split-index was written */
3462 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3463 const char *shared_index
= git_path("sharedindex.%s",
3464 oid_to_hex(&si
->base_oid
));
3465 freshen_shared_index(shared_index
, 1);
3469 if (flags
& COMMIT_LOCK
)
3470 rollback_lock_file(lock
);
3475 * Read the index file that is potentially unmerged into given
3476 * index_state, dropping any unmerged entries to stage #0 (potentially
3477 * resulting in a path appearing as both a file and a directory in the
3478 * index; the caller is responsible to clear out the extra entries
3479 * before writing the index to a tree). Returns true if the index is
3480 * unmerged. Callers who want to refuse to work from an unmerged
3481 * state can call this and check its return value, instead of calling
3484 int repo_read_index_unmerged(struct repository
*repo
)
3486 struct index_state
*istate
;
3490 repo_read_index(repo
);
3491 istate
= repo
->index
;
3492 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3493 struct cache_entry
*ce
= istate
->cache
[i
];
3494 struct cache_entry
*new_ce
;
3500 len
= ce_namelen(ce
);
3501 new_ce
= make_empty_cache_entry(istate
, len
);
3502 memcpy(new_ce
->name
, ce
->name
, len
);
3503 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3504 new_ce
->ce_namelen
= len
;
3505 new_ce
->ce_mode
= ce
->ce_mode
;
3506 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3507 return error(_("%s: cannot drop to stage #0"),
3514 * Returns 1 if the path is an "other" path with respect to
3515 * the index; that is, the path is not mentioned in the index at all,
3516 * either as a file, a directory with some files in the index,
3517 * or as an unmerged entry.
3519 * We helpfully remove a trailing "/" from directories so that
3520 * the output of read_directory can be used as-is.
3522 int index_name_is_other(struct index_state
*istate
, const char *name
,
3526 if (namelen
&& name
[namelen
- 1] == '/')
3528 pos
= index_name_pos(istate
, name
, namelen
);
3530 return 0; /* exact match */
3532 if (pos
< istate
->cache_nr
) {
3533 struct cache_entry
*ce
= istate
->cache
[pos
];
3534 if (ce_namelen(ce
) == namelen
&&
3535 !memcmp(ce
->name
, name
, namelen
))
3536 return 0; /* Yup, this one exists unmerged */
3541 void *read_blob_data_from_index(struct index_state
*istate
,
3542 const char *path
, unsigned long *size
)
3546 enum object_type type
;
3550 pos
= index_name_pos(istate
, path
, len
);
3553 * We might be in the middle of a merge, in which
3554 * case we would read stage #2 (ours).
3558 (pos
< 0 && i
< istate
->cache_nr
&&
3559 !strcmp(istate
->cache
[i
]->name
, path
));
3561 if (ce_stage(istate
->cache
[i
]) == 2)
3566 data
= repo_read_object_file(the_repository
, &istate
->cache
[pos
]->oid
,
3568 if (!data
|| type
!= OBJ_BLOB
) {
3577 void stat_validity_clear(struct stat_validity
*sv
)
3579 FREE_AND_NULL(sv
->sd
);
3582 int stat_validity_check(struct stat_validity
*sv
, const char *path
)
3586 if (stat(path
, &st
) < 0)
3587 return sv
->sd
== NULL
;
3590 return S_ISREG(st
.st_mode
) && !match_stat_data(sv
->sd
, &st
);
3593 void stat_validity_update(struct stat_validity
*sv
, int fd
)
3597 if (fstat(fd
, &st
) < 0 || !S_ISREG(st
.st_mode
))
3598 stat_validity_clear(sv
);
3601 CALLOC_ARRAY(sv
->sd
, 1);
3602 fill_stat_data(sv
->sd
, &st
);
3606 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3608 dst
->untracked
= src
->untracked
;
3609 src
->untracked
= NULL
;
3610 dst
->cache_tree
= src
->cache_tree
;
3611 src
->cache_tree
= NULL
;
3614 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3615 struct index_state
*istate
)
3617 unsigned int size
= ce_size(ce
);
3618 int mem_pool_allocated
;
3619 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3620 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3622 memcpy(new_entry
, ce
, size
);
3623 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3627 void discard_cache_entry(struct cache_entry
*ce
)
3629 if (ce
&& should_validate_cache_entries())
3630 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3632 if (ce
&& ce
->mem_pool_allocated
)
3638 int should_validate_cache_entries(void)
3640 static int validate_index_cache_entries
= -1;
3642 if (validate_index_cache_entries
< 0) {
3643 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3644 validate_index_cache_entries
= 1;
3646 validate_index_cache_entries
= 0;
3649 return validate_index_cache_entries
;
3652 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3653 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3655 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3658 * The end of index entries (EOIE) extension is guaranteed to be last
3659 * so that it can be found by scanning backwards from the EOF.
3666 const char *index
, *eoie
;
3668 size_t offset
, src_offset
;
3669 unsigned char hash
[GIT_MAX_RAWSZ
];
3672 /* ensure we have an index big enough to contain an EOIE extension */
3673 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3676 /* validate the extension signature */
3677 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3678 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3680 index
+= sizeof(uint32_t);
3682 /* validate the extension size */
3683 extsize
= get_be32(index
);
3684 if (extsize
!= EOIE_SIZE
)
3686 index
+= sizeof(uint32_t);
3689 * Validate the offset we're going to look for the first extension
3690 * signature is after the index header and before the eoie extension.
3692 offset
= get_be32(index
);
3693 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3695 if (mmap
+ offset
>= eoie
)
3697 index
+= sizeof(uint32_t);
3700 * The hash is computed over extension types and their sizes (but not
3701 * their contents). E.g. if we have "TREE" extension that is N-bytes
3702 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3703 * then the hash would be:
3705 * SHA-1("TREE" + <binary representation of N> +
3706 * "REUC" + <binary representation of M>)
3708 src_offset
= offset
;
3709 the_hash_algo
->init_fn(&c
);
3710 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3711 /* After an array of active_nr index entries,
3712 * there can be arbitrary number of extended
3713 * sections, each of which is prefixed with
3714 * extension name (4-byte) and section length
3715 * in 4-byte network byte order.
3718 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3719 extsize
= ntohl(extsize
);
3721 /* verify the extension size isn't so large it will wrap around */
3722 if (src_offset
+ 8 + extsize
< src_offset
)
3725 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3728 src_offset
+= extsize
;
3730 the_hash_algo
->final_fn(hash
, &c
);
3731 if (!hasheq(hash
, (const unsigned char *)index
))
3734 /* Validate that the extension offsets returned us back to the eoie extension. */
3735 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3741 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3744 unsigned char hash
[GIT_MAX_RAWSZ
];
3747 put_be32(&buffer
, offset
);
3748 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3751 the_hash_algo
->final_fn(hash
, eoie_context
);
3752 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3755 #define IEOT_VERSION (1)
3757 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3759 const char *index
= NULL
;
3760 uint32_t extsize
, ext_version
;
3761 struct index_entry_offset_table
*ieot
;
3764 /* find the IEOT extension */
3767 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3768 extsize
= get_be32(mmap
+ offset
+ 4);
3769 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3770 index
= mmap
+ offset
+ 4 + 4;
3779 /* validate the version is IEOT_VERSION */
3780 ext_version
= get_be32(index
);
3781 if (ext_version
!= IEOT_VERSION
) {
3782 error("invalid IEOT version %d", ext_version
);
3785 index
+= sizeof(uint32_t);
3787 /* extension size - version bytes / bytes per entry */
3788 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3790 error("invalid number of IEOT entries %d", nr
);
3793 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3794 + (nr
* sizeof(struct index_entry_offset
)));
3796 for (i
= 0; i
< nr
; i
++) {
3797 ieot
->entries
[i
].offset
= get_be32(index
);
3798 index
+= sizeof(uint32_t);
3799 ieot
->entries
[i
].nr
= get_be32(index
);
3800 index
+= sizeof(uint32_t);
3806 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3812 put_be32(&buffer
, IEOT_VERSION
);
3813 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3816 for (i
= 0; i
< ieot
->nr
; i
++) {
3819 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3820 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3823 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3824 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3828 void prefetch_cache_entries(const struct index_state
*istate
,
3829 must_prefetch_predicate must_prefetch
)
3832 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3834 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3835 struct cache_entry
*ce
= istate
->cache
[i
];
3837 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3839 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3841 OBJECT_INFO_FOR_PREFETCH
))
3843 oid_array_append(&to_fetch
, &ce
->oid
);
3845 promisor_remote_get_direct(the_repository
,
3846 to_fetch
.oid
, to_fetch
.nr
);
3847 oid_array_clear(&to_fetch
);