2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #include "git-compat-util.h"
8 #include "bulk-checkin.h"
16 #include "cache-tree.h"
19 #include "object-file.h"
20 #include "object-store-ll.h"
21 #include "oid-array.h"
25 #include "environment.h"
28 #include "name-hash.h"
29 #include "object-name.h"
31 #include "preload-index.h"
32 #include "read-cache.h"
33 #include "resolve-undo.h"
35 #include "run-command.h"
39 #include "split-index.h"
42 #include "fsmonitor.h"
43 #include "thread-utils.h"
45 #include "sparse-index.h"
46 #include "csum-file.h"
47 #include "promisor-remote.h"
51 /* Mask for the name length in ce_flags in the on-disk index */
53 #define CE_NAMEMASK (0x0fff)
57 * The first letter should be 'A'..'Z' for extensions that are not
58 * necessary for a correct operation (i.e. optimization data).
59 * When new extensions are added that _needs_ to be understood in
60 * order to correctly interpret the index file, pick character that
61 * is outside the range, to cause the reader to abort.
64 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
65 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
66 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
67 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
68 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
69 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
70 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
71 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
72 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
74 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
75 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
76 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
77 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
81 * This is an estimate of the pathname length in the index. We use
82 * this for V4 index files to guess the un-deltafied size of the index
83 * in memory because of pathname deltafication. This is not required
84 * for V2/V3 index formats because their pathnames are not compressed.
85 * If the initial amount of memory set aside is not sufficient, the
86 * mem pool will allocate extra memory.
88 #define CACHE_ENTRY_PATH_LENGTH 80
90 enum index_search_mode
{
95 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
97 struct cache_entry
*ce
;
98 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
99 ce
->mem_pool_allocated
= 1;
103 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
105 struct cache_entry
* ce
;
106 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
107 ce
->mem_pool_allocated
= 1;
111 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
113 struct mem_pool
**pool_ptr
;
115 if (istate
->split_index
&& istate
->split_index
->base
)
116 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
118 pool_ptr
= &istate
->ce_mem_pool
;
121 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
122 mem_pool_init(*pool_ptr
, 0);
128 static const char *alternate_index_output
;
130 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
132 if (S_ISSPARSEDIR(ce
->ce_mode
))
133 istate
->sparse_index
= INDEX_COLLAPSED
;
135 istate
->cache
[nr
] = ce
;
136 add_name_hash(istate
, ce
);
139 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
141 struct cache_entry
*old
= istate
->cache
[nr
];
143 replace_index_entry_in_base(istate
, old
, ce
);
144 remove_name_hash(istate
, old
);
145 discard_cache_entry(old
);
146 ce
->ce_flags
&= ~CE_HASHED
;
147 set_index_entry(istate
, nr
, ce
);
148 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
149 mark_fsmonitor_invalid(istate
, ce
);
150 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
153 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
155 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
, *refreshed
;
156 int namelen
= strlen(new_name
);
158 new_entry
= make_empty_cache_entry(istate
, namelen
);
159 copy_cache_entry(new_entry
, old_entry
);
160 new_entry
->ce_flags
&= ~CE_HASHED
;
161 new_entry
->ce_namelen
= namelen
;
162 new_entry
->index
= 0;
163 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
165 cache_tree_invalidate_path(istate
, old_entry
->name
);
166 untracked_cache_remove_from_index(istate
, old_entry
->name
);
167 remove_index_entry_at(istate
, nr
);
170 * Refresh the new index entry. Using 'refresh_cache_entry' ensures
171 * we only update stat info if the entry is otherwise up-to-date (i.e.,
172 * the contents/mode haven't changed). This ensures that we reflect the
173 * 'ctime' of the rename in the index without (incorrectly) updating
174 * the cached stat info to reflect unstaged changes on disk.
176 refreshed
= refresh_cache_entry(istate
, new_entry
, CE_MATCH_REFRESH
);
177 if (refreshed
&& refreshed
!= new_entry
) {
178 add_index_entry(istate
, refreshed
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
179 discard_cache_entry(new_entry
);
181 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
185 * This only updates the "non-critical" parts of the directory
186 * cache, ie the parts that aren't tracked by GIT, and only used
187 * to validate the cache.
189 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
191 fill_stat_data(&ce
->ce_stat_data
, st
);
193 if (assume_unchanged
)
194 ce
->ce_flags
|= CE_VALID
;
196 if (S_ISREG(st
->st_mode
)) {
197 ce_mark_uptodate(ce
);
198 mark_fsmonitor_valid(istate
, ce
);
202 static int ce_compare_data(struct index_state
*istate
,
203 const struct cache_entry
*ce
,
207 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
210 struct object_id oid
;
211 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
212 match
= !oideq(&oid
, &ce
->oid
);
213 /* index_fd() closed the file descriptor already */
218 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
223 enum object_type type
;
224 struct strbuf sb
= STRBUF_INIT
;
226 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
229 buffer
= repo_read_object_file(the_repository
, &ce
->oid
, &type
, &size
);
232 match
= memcmp(buffer
, sb
.buf
, size
);
239 static int ce_compare_gitlink(const struct cache_entry
*ce
)
241 struct object_id oid
;
244 * We don't actually require that the .git directory
245 * under GITLINK directory be a valid git directory. It
246 * might even be missing (in case nobody populated that
249 * If so, we consider it always to match.
251 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
253 return !oideq(&oid
, &ce
->oid
);
256 static int ce_modified_check_fs(struct index_state
*istate
,
257 const struct cache_entry
*ce
,
260 switch (st
->st_mode
& S_IFMT
) {
262 if (ce_compare_data(istate
, ce
, st
))
266 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
270 if (S_ISGITLINK(ce
->ce_mode
))
271 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
272 /* else fallthrough */
279 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
281 unsigned int changed
= 0;
283 if (ce
->ce_flags
& CE_REMOVE
)
284 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
286 switch (ce
->ce_mode
& S_IFMT
) {
288 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
289 /* We consider only the owner x bit to be relevant for
292 if (trust_executable_bit
&&
293 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
294 changed
|= MODE_CHANGED
;
297 if (!S_ISLNK(st
->st_mode
) &&
298 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
299 changed
|= TYPE_CHANGED
;
302 /* We ignore most of the st_xxx fields for gitlinks */
303 if (!S_ISDIR(st
->st_mode
))
304 changed
|= TYPE_CHANGED
;
305 else if (ce_compare_gitlink(ce
))
306 changed
|= DATA_CHANGED
;
309 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
312 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
314 /* Racily smudged entry? */
315 if (!ce
->ce_stat_data
.sd_size
) {
316 if (!is_empty_blob_sha1(ce
->oid
.hash
))
317 changed
|= DATA_CHANGED
;
323 static int is_racy_stat(const struct index_state
*istate
,
324 const struct stat_data
*sd
)
326 return (istate
->timestamp
.sec
&&
328 /* nanosecond timestamped files can also be racy! */
329 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
330 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
331 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
333 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
338 int is_racy_timestamp(const struct index_state
*istate
,
339 const struct cache_entry
*ce
)
341 return (!S_ISGITLINK(ce
->ce_mode
) &&
342 is_racy_stat(istate
, &ce
->ce_stat_data
));
345 int match_stat_data_racy(const struct index_state
*istate
,
346 const struct stat_data
*sd
, struct stat
*st
)
348 if (is_racy_stat(istate
, sd
))
349 return MTIME_CHANGED
;
350 return match_stat_data(sd
, st
);
353 int ie_match_stat(struct index_state
*istate
,
354 const struct cache_entry
*ce
, struct stat
*st
,
355 unsigned int options
)
357 unsigned int changed
;
358 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
359 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
360 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
361 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
363 if (!ignore_fsmonitor
)
364 refresh_fsmonitor(istate
);
366 * If it's marked as always valid in the index, it's
367 * valid whatever the checked-out copy says.
369 * skip-worktree has the same effect with higher precedence
371 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
373 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
375 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
379 * Intent-to-add entries have not been added, so the index entry
380 * by definition never matches what is in the work tree until it
381 * actually gets added.
383 if (ce_intent_to_add(ce
))
384 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
386 changed
= ce_match_stat_basic(ce
, st
);
389 * Within 1 second of this sequence:
390 * echo xyzzy >file && git-update-index --add file
391 * running this command:
393 * would give a falsely clean cache entry. The mtime and
394 * length match the cache, and other stat fields do not change.
396 * We could detect this at update-index time (the cache entry
397 * being registered/updated records the same time as "now")
398 * and delay the return from git-update-index, but that would
399 * effectively mean we can make at most one commit per second,
400 * which is not acceptable. Instead, we check cache entries
401 * whose mtime are the same as the index file timestamp more
402 * carefully than others.
404 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
405 if (assume_racy_is_modified
)
406 changed
|= DATA_CHANGED
;
408 changed
|= ce_modified_check_fs(istate
, ce
, st
);
414 int ie_modified(struct index_state
*istate
,
415 const struct cache_entry
*ce
,
416 struct stat
*st
, unsigned int options
)
418 int changed
, changed_fs
;
420 changed
= ie_match_stat(istate
, ce
, st
, options
);
424 * If the mode or type has changed, there's no point in trying
425 * to refresh the entry - it's not going to match
427 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
431 * Immediately after read-tree or update-index --cacheinfo,
432 * the length field is zero, as we have never even read the
433 * lstat(2) information once, and we cannot trust DATA_CHANGED
434 * returned by ie_match_stat() which in turn was returned by
435 * ce_match_stat_basic() to signal that the filesize of the
436 * blob changed. We have to actually go to the filesystem to
437 * see if the contents match, and if so, should answer "unchanged".
439 * The logic does not apply to gitlinks, as ce_match_stat_basic()
440 * already has checked the actual HEAD from the filesystem in the
441 * subproject. If ie_match_stat() already said it is different,
442 * then we know it is.
444 if ((changed
& DATA_CHANGED
) &&
445 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
448 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
450 return changed
| changed_fs
;
454 static int cache_name_stage_compare(const char *name1
, int len1
, int stage1
,
455 const char *name2
, int len2
, int stage2
)
459 cmp
= name_compare(name1
, len1
, name2
, len2
);
470 int cmp_cache_name_compare(const void *a_
, const void *b_
)
472 const struct cache_entry
*ce1
, *ce2
;
474 ce1
= *((const struct cache_entry
**)a_
);
475 ce2
= *((const struct cache_entry
**)b_
);
476 return cache_name_stage_compare(ce1
->name
, ce1
->ce_namelen
, ce_stage(ce1
),
477 ce2
->name
, ce2
->ce_namelen
, ce_stage(ce2
));
480 static int index_name_stage_pos(struct index_state
*istate
,
481 const char *name
, int namelen
,
483 enum index_search_mode search_mode
)
488 last
= istate
->cache_nr
;
489 while (last
> first
) {
490 int next
= first
+ ((last
- first
) >> 1);
491 struct cache_entry
*ce
= istate
->cache
[next
];
492 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
502 if (search_mode
== EXPAND_SPARSE
&& istate
->sparse_index
&&
504 /* Note: first <= istate->cache_nr */
505 struct cache_entry
*ce
= istate
->cache
[first
- 1];
508 * If we are in a sparse-index _and_ the entry before the
509 * insertion position is a sparse-directory entry that is
510 * an ancestor of 'name', then we need to expand the index
511 * and search again. This will only trigger once, because
512 * thereafter the index is fully expanded.
514 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
515 ce_namelen(ce
) < namelen
&&
516 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
517 ensure_full_index(istate
);
518 return index_name_stage_pos(istate
, name
, namelen
, stage
, search_mode
);
525 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
527 return index_name_stage_pos(istate
, name
, namelen
, 0, EXPAND_SPARSE
);
530 int index_name_pos_sparse(struct index_state
*istate
, const char *name
, int namelen
)
532 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
);
535 int index_entry_exists(struct index_state
*istate
, const char *name
, int namelen
)
537 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
) >= 0;
540 int remove_index_entry_at(struct index_state
*istate
, int pos
)
542 struct cache_entry
*ce
= istate
->cache
[pos
];
544 record_resolve_undo(istate
, ce
);
545 remove_name_hash(istate
, ce
);
546 save_or_free_index_entry(istate
, ce
);
547 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
549 if (pos
>= istate
->cache_nr
)
551 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
552 istate
->cache_nr
- pos
);
557 * Remove all cache entries marked for removal, that is where
558 * CE_REMOVE is set in ce_flags. This is much more effective than
559 * calling remove_index_entry_at() for each entry to be removed.
561 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
563 struct cache_entry
**ce_array
= istate
->cache
;
566 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
567 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
569 cache_tree_invalidate_path(istate
,
571 untracked_cache_remove_from_index(istate
,
574 remove_name_hash(istate
, ce_array
[i
]);
575 save_or_free_index_entry(istate
, ce_array
[i
]);
578 ce_array
[j
++] = ce_array
[i
];
580 if (j
== istate
->cache_nr
)
582 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
583 istate
->cache_nr
= j
;
586 int remove_file_from_index(struct index_state
*istate
, const char *path
)
588 int pos
= index_name_pos(istate
, path
, strlen(path
));
591 cache_tree_invalidate_path(istate
, path
);
592 untracked_cache_remove_from_index(istate
, path
);
593 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
594 remove_index_entry_at(istate
, pos
);
598 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
600 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
603 static int index_name_pos_also_unmerged(struct index_state
*istate
,
604 const char *path
, int namelen
)
606 int pos
= index_name_pos(istate
, path
, namelen
);
607 struct cache_entry
*ce
;
612 /* maybe unmerged? */
614 if (pos
>= istate
->cache_nr
||
615 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
618 /* order of preference: stage 2, 1, 3 */
619 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
620 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
621 !compare_name(ce
, path
, namelen
))
626 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
628 int len
= ce_namelen(ce
);
629 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
633 * If we add a filename that aliases in the cache, we will use the
634 * name that we already have - but we don't want to update the same
635 * alias twice, because that implies that there were actually two
636 * different files with aliasing names!
638 * So we use the CE_ADDED flag to verify that the alias was an old
639 * one before we accept it as
641 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
642 struct cache_entry
*ce
,
643 struct cache_entry
*alias
)
646 struct cache_entry
*new_entry
;
648 if (alias
->ce_flags
& CE_ADDED
)
649 die(_("will not add file alias '%s' ('%s' already exists in index)"),
650 ce
->name
, alias
->name
);
652 /* Ok, create the new entry using the name of the existing alias */
653 len
= ce_namelen(alias
);
654 new_entry
= make_empty_cache_entry(istate
, len
);
655 memcpy(new_entry
->name
, alias
->name
, len
);
656 copy_cache_entry(new_entry
, ce
);
657 save_or_free_index_entry(istate
, ce
);
661 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
663 struct object_id oid
;
664 if (write_object_file("", 0, OBJ_BLOB
, &oid
))
665 die(_("cannot create an empty blob in the object database"));
666 oidcpy(&ce
->oid
, &oid
);
669 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
671 int namelen
, was_same
;
672 mode_t st_mode
= st
->st_mode
;
673 struct cache_entry
*ce
, *alias
= NULL
;
674 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
675 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
676 int pretend
= flags
& ADD_CACHE_PRETEND
;
677 int intent_only
= flags
& ADD_CACHE_INTENT
;
678 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
679 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
680 unsigned hash_flags
= pretend
? 0 : HASH_WRITE_OBJECT
;
681 struct object_id oid
;
683 if (flags
& ADD_CACHE_RENORMALIZE
)
684 hash_flags
|= HASH_RENORMALIZE
;
686 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
687 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
689 namelen
= strlen(path
);
690 if (S_ISDIR(st_mode
)) {
691 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
692 return error(_("'%s' does not have a commit checked out"), path
);
693 while (namelen
&& path
[namelen
-1] == '/')
696 ce
= make_empty_cache_entry(istate
, namelen
);
697 memcpy(ce
->name
, path
, namelen
);
698 ce
->ce_namelen
= namelen
;
700 fill_stat_cache_info(istate
, ce
, st
);
702 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
705 if (trust_executable_bit
&& has_symlinks
) {
706 ce
->ce_mode
= create_ce_mode(st_mode
);
708 /* If there is an existing entry, pick the mode bits and type
709 * from it, otherwise assume unexecutable regular file.
711 struct cache_entry
*ent
;
712 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
714 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
715 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
718 /* When core.ignorecase=true, determine if a directory of the same name but differing
719 * case already exists within the Git repository. If it does, ensure the directory
720 * case of the file being added to the repository matches (is folded into) the existing
721 * entry's directory case.
724 adjust_dirname_case(istate
, ce
->name
);
726 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
727 alias
= index_file_exists(istate
, ce
->name
,
728 ce_namelen(ce
), ignore_case
);
731 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
732 /* Nothing changed, really */
733 if (!S_ISGITLINK(alias
->ce_mode
))
734 ce_mark_uptodate(alias
);
735 alias
->ce_flags
|= CE_ADDED
;
737 discard_cache_entry(ce
);
742 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
743 discard_cache_entry(ce
);
744 return error(_("unable to index file '%s'"), path
);
747 set_object_name_for_intent_to_add_entry(ce
);
749 if (ignore_case
&& alias
&& different_name(ce
, alias
))
750 ce
= create_alias_ce(istate
, ce
, alias
);
751 ce
->ce_flags
|= CE_ADDED
;
753 /* It was suspected to be racily clean, but it turns out to be Ok */
756 oideq(&alias
->oid
, &ce
->oid
) &&
757 ce
->ce_mode
== alias
->ce_mode
);
760 discard_cache_entry(ce
);
761 else if (add_index_entry(istate
, ce
, add_option
)) {
762 discard_cache_entry(ce
);
763 return error(_("unable to add '%s' to index"), path
);
765 if (verbose
&& !was_same
)
766 printf("add '%s'\n", path
);
770 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
773 if (lstat(path
, &st
))
774 die_errno(_("unable to stat '%s'"), path
);
775 return add_to_index(istate
, path
, &st
, flags
);
778 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
780 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
783 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
784 struct mem_pool
*ce_mem_pool
)
787 return mem_pool__ce_calloc(ce_mem_pool
, len
);
788 return xcalloc(1, cache_entry_size(len
));
791 enum verify_path_result
{
797 static enum verify_path_result
verify_path_internal(const char *, unsigned);
799 int verify_path(const char *path
, unsigned mode
)
801 return verify_path_internal(path
, mode
) == PATH_OK
;
804 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
806 const struct object_id
*oid
,
809 unsigned int refresh_options
)
811 struct cache_entry
*ce
, *ret
;
814 if (verify_path_internal(path
, mode
) == PATH_INVALID
) {
815 error(_("invalid path '%s'"), path
);
820 ce
= make_empty_cache_entry(istate
, len
);
822 oidcpy(&ce
->oid
, oid
);
823 memcpy(ce
->name
, path
, len
);
824 ce
->ce_flags
= create_ce_flags(stage
);
825 ce
->ce_namelen
= len
;
826 ce
->ce_mode
= create_ce_mode(mode
);
828 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
830 discard_cache_entry(ce
);
834 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
835 const struct object_id
*oid
,
838 struct mem_pool
*ce_mem_pool
)
840 struct cache_entry
*ce
;
843 if (!verify_path(path
, mode
)) {
844 error(_("invalid path '%s'"), path
);
849 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
851 oidcpy(&ce
->oid
, oid
);
852 memcpy(ce
->name
, path
, len
);
853 ce
->ce_flags
= create_ce_flags(stage
);
854 ce
->ce_namelen
= len
;
855 ce
->ce_mode
= create_ce_mode(mode
);
861 * Chmod an index entry with either +x or -x.
863 * Returns -1 if the chmod for the particular cache entry failed (if it's
864 * not a regular file), -2 if an invalid flip argument is passed in, 0
867 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
870 if (!S_ISREG(ce
->ce_mode
))
877 ce
->ce_mode
&= ~0111;
882 cache_tree_invalidate_path(istate
, ce
->name
);
883 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
884 mark_fsmonitor_invalid(istate
, ce
);
885 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
890 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
892 int len
= ce_namelen(a
);
893 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
897 * We fundamentally don't like some paths: we don't want
898 * dot or dot-dot anywhere, and for obvious reasons don't
899 * want to recurse into ".git" either.
901 * Also, we don't want double slashes or slashes at the
902 * end that can make pathnames ambiguous.
904 static int verify_dotfile(const char *rest
, unsigned mode
)
907 * The first character was '.', but that
908 * has already been discarded, we now test
912 /* "." is not allowed */
913 if (*rest
== '\0' || is_dir_sep(*rest
))
918 * ".git" followed by NUL or slash is bad. Note that we match
919 * case-insensitively here, even if ignore_case is not set.
920 * This outlaws ".GIT" everywhere out of an abundance of caution,
921 * since there's really no good reason to allow it.
923 * Once we've seen ".git", we can also find ".gitmodules", etc (also
924 * case-insensitively).
928 if (rest
[1] != 'i' && rest
[1] != 'I')
930 if (rest
[2] != 't' && rest
[2] != 'T')
932 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
936 if (skip_iprefix(rest
, "modules", &rest
) &&
937 (*rest
== '\0' || is_dir_sep(*rest
)))
942 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
948 static enum verify_path_result
verify_path_internal(const char *path
,
953 if (has_dos_drive_prefix(path
))
956 if (!is_valid_path(path
))
967 if (is_hfs_dotgit(path
))
970 if (is_hfs_dotgitmodules(path
))
975 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
979 if (is_ntfs_dotgit(path
))
982 if (is_ntfs_dotgitmodules(path
))
988 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
992 * allow terminating directory separators for
993 * sparse directory entries.
996 return S_ISDIR(mode
) ? PATH_DIR_WITH_SEP
:
998 } else if (c
== '\\' && protect_ntfs
) {
999 if (is_ntfs_dotgit(path
))
1000 return PATH_INVALID
;
1001 if (S_ISLNK(mode
)) {
1002 if (is_ntfs_dotgitmodules(path
))
1003 return PATH_INVALID
;
1012 * Do we have another file that has the beginning components being a
1013 * proper superset of the name we're trying to add?
1015 static int has_file_name(struct index_state
*istate
,
1016 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1019 int len
= ce_namelen(ce
);
1020 int stage
= ce_stage(ce
);
1021 const char *name
= ce
->name
;
1023 while (pos
< istate
->cache_nr
) {
1024 struct cache_entry
*p
= istate
->cache
[pos
++];
1026 if (len
>= ce_namelen(p
))
1028 if (memcmp(name
, p
->name
, len
))
1030 if (ce_stage(p
) != stage
)
1032 if (p
->name
[len
] != '/')
1034 if (p
->ce_flags
& CE_REMOVE
)
1039 remove_index_entry_at(istate
, --pos
);
1046 * Like strcmp(), but also return the offset of the first change.
1047 * If strings are equal, return the length.
1049 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1054 return strcmp(s1
, s2
);
1056 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1061 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1065 * Do we have another file with a pathname that is a proper
1066 * subset of the name we're trying to add?
1068 * That is, is there another file in the index with a path
1069 * that matches a sub-directory in the given entry?
1071 static int has_dir_name(struct index_state
*istate
,
1072 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1075 int stage
= ce_stage(ce
);
1076 const char *name
= ce
->name
;
1077 const char *slash
= name
+ ce_namelen(ce
);
1082 * We are frequently called during an iteration on a sorted
1083 * list of pathnames and while building a new index. Therefore,
1084 * there is a high probability that this entry will eventually
1085 * be appended to the index, rather than inserted in the middle.
1086 * If we can confirm that, we can avoid binary searches on the
1087 * components of the pathname.
1089 * Compare the entry's full path with the last path in the index.
1091 if (istate
->cache_nr
> 0) {
1092 cmp_last
= strcmp_offset(name
,
1093 istate
->cache
[istate
->cache_nr
- 1]->name
,
1096 if (len_eq_last
== 0) {
1098 * The entry sorts AFTER the last one in the
1099 * index and their paths have no common prefix,
1100 * so there cannot be a F/D conflict.
1105 * The entry sorts AFTER the last one in the
1106 * index, but has a common prefix. Fall through
1107 * to the loop below to disect the entry's path
1108 * and see where the difference is.
1111 } else if (cmp_last
== 0) {
1113 * The entry exactly matches the last one in the
1114 * index, but because of multiple stage and CE_REMOVE
1115 * items, we fall through and let the regular search
1125 if (*--slash
== '/')
1127 if (slash
<= ce
->name
)
1134 * (len + 1) is a directory boundary (including
1135 * the trailing slash). And since the loop is
1136 * decrementing "slash", the first iteration is
1137 * the longest directory prefix; subsequent
1138 * iterations consider parent directories.
1141 if (len
+ 1 <= len_eq_last
) {
1143 * The directory prefix (including the trailing
1144 * slash) also appears as a prefix in the last
1145 * entry, so the remainder cannot collide (because
1146 * strcmp said the whole path was greater).
1151 * LT: last: xxx/file_A
1157 if (len
> len_eq_last
) {
1159 * This part of the directory prefix (excluding
1160 * the trailing slash) is longer than the known
1161 * equal portions, so this sub-directory cannot
1162 * collide with a file.
1171 * This is a possible collision. Fall through and
1172 * let the regular search code handle it.
1179 pos
= index_name_stage_pos(istate
, name
, len
, stage
, EXPAND_SPARSE
);
1182 * Found one, but not so fast. This could
1183 * be a marker that says "I was here, but
1184 * I am being removed". Such an entry is
1185 * not a part of the resulting tree, and
1186 * it is Ok to have a directory at the same
1189 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1193 remove_index_entry_at(istate
, pos
);
1201 * Trivial optimization: if we find an entry that
1202 * already matches the sub-directory, then we know
1203 * we're ok, and we can exit.
1205 while (pos
< istate
->cache_nr
) {
1206 struct cache_entry
*p
= istate
->cache
[pos
];
1207 if ((ce_namelen(p
) <= len
) ||
1208 (p
->name
[len
] != '/') ||
1209 memcmp(p
->name
, name
, len
))
1210 break; /* not our subdirectory */
1211 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1213 * p is at the same stage as our entry, and
1214 * is a subdirectory of what we are looking
1215 * at, so we cannot have conflicts at our
1216 * level or anything shorter.
1225 /* We may be in a situation where we already have path/file and path
1226 * is being added, or we already have path and path/file is being
1227 * added. Either one would result in a nonsense tree that has path
1228 * twice when git-write-tree tries to write it out. Prevent it.
1230 * If ok-to-replace is specified, we remove the conflicting entries
1231 * from the cache so the caller should recompute the insert position.
1232 * When this happens, we return non-zero.
1234 static int check_file_directory_conflict(struct index_state
*istate
,
1235 const struct cache_entry
*ce
,
1236 int pos
, int ok_to_replace
)
1241 * When ce is an "I am going away" entry, we allow it to be added
1243 if (ce
->ce_flags
& CE_REMOVE
)
1247 * We check if the path is a sub-path of a subsequent pathname
1248 * first, since removing those will not change the position
1251 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1254 * Then check if the path might have a clashing sub-directory
1257 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1260 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1263 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1264 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1265 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1266 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1269 * If this entry's path sorts after the last entry in the index,
1270 * we can avoid searching for it.
1272 if (istate
->cache_nr
> 0 &&
1273 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1274 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1276 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1279 * Cache tree path should be invalidated only after index_name_stage_pos,
1280 * in case it expands a sparse index.
1282 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1283 cache_tree_invalidate_path(istate
, ce
->name
);
1285 /* existing match? Just replace it. */
1288 replace_index_entry(istate
, pos
, ce
);
1293 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1294 untracked_cache_add_to_index(istate
, ce
->name
);
1297 * Inserting a merged entry ("stage 0") into the index
1298 * will always replace all non-merged entries..
1300 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1301 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1303 if (!remove_index_entry_at(istate
, pos
))
1310 if (verify_path_internal(ce
->name
, ce
->ce_mode
) == PATH_INVALID
)
1311 return error(_("invalid path '%s'"), ce
->name
);
1313 if (!skip_df_check
&&
1314 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1316 return error(_("'%s' appears as both a file and as a directory"),
1318 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1324 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1328 if (option
& ADD_CACHE_JUST_APPEND
)
1329 pos
= istate
->cache_nr
;
1332 ret
= add_index_entry_with_check(istate
, ce
, option
);
1338 /* Make sure the array is big enough .. */
1339 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1343 if (istate
->cache_nr
> pos
+ 1)
1344 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1345 istate
->cache_nr
- pos
- 1);
1346 set_index_entry(istate
, pos
, ce
);
1347 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1352 * "refresh" does not calculate a new sha1 file or bring the
1353 * cache up-to-date for mode/content changes. But what it
1354 * _does_ do is to "re-match" the stat information of a file
1355 * with the cache, so that you can refresh the cache for a
1356 * file that hasn't been changed but where the stat entry is
1359 * For example, you'd want to do this after doing a "git-read-tree",
1360 * to link up the stat cache details with the proper files.
1362 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1363 struct cache_entry
*ce
,
1364 unsigned int options
, int *err
,
1370 struct cache_entry
*updated
;
1372 int refresh
= options
& CE_MATCH_REFRESH
;
1373 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1374 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1375 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1376 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1378 if (!refresh
|| ce_uptodate(ce
))
1381 if (!ignore_fsmonitor
)
1382 refresh_fsmonitor(istate
);
1384 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1385 * that the change to the work tree does not matter and told
1388 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1389 ce_mark_uptodate(ce
);
1392 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1393 ce_mark_uptodate(ce
);
1396 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1397 ce_mark_uptodate(ce
);
1401 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1411 if (lstat(ce
->name
, &st
) < 0) {
1412 if (ignore_missing
&& errno
== ENOENT
)
1419 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1421 *changed_ret
= changed
;
1424 * The path is unchanged. If we were told to ignore
1425 * valid bit, then we did the actual stat check and
1426 * found that the entry is unmodified. If the entry
1427 * is not marked VALID, this is the place to mark it
1428 * valid again, under "assume unchanged" mode.
1430 if (ignore_valid
&& assume_unchanged
&&
1431 !(ce
->ce_flags
& CE_VALID
))
1432 ; /* mark this one VALID again */
1435 * We do not mark the index itself "modified"
1436 * because CE_UPTODATE flag is in-core only;
1437 * we are not going to write this change out.
1439 if (!S_ISGITLINK(ce
->ce_mode
)) {
1440 ce_mark_uptodate(ce
);
1441 mark_fsmonitor_valid(istate
, ce
);
1449 if (ie_modified(istate
, ce
, &st
, options
)) {
1455 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1456 copy_cache_entry(updated
, ce
);
1457 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1458 fill_stat_cache_info(istate
, updated
, &st
);
1460 * If ignore_valid is not set, we should leave CE_VALID bit
1461 * alone. Otherwise, paths marked with --no-assume-unchanged
1462 * (i.e. things to be edited) will reacquire CE_VALID bit
1463 * automatically, which is not really what we want.
1465 if (!ignore_valid
&& assume_unchanged
&&
1466 !(ce
->ce_flags
& CE_VALID
))
1467 updated
->ce_flags
&= ~CE_VALID
;
1469 /* istate->cache_changed is updated in the caller */
1473 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1474 int * first
, const char *header_msg
)
1476 if (in_porcelain
&& *first
&& header_msg
) {
1477 printf("%s\n", header_msg
);
1483 int repo_refresh_and_write_index(struct repository
*repo
,
1484 unsigned int refresh_flags
,
1485 unsigned int write_flags
,
1487 const struct pathspec
*pathspec
,
1488 char *seen
, const char *header_msg
)
1490 struct lock_file lock_file
= LOCK_INIT
;
1493 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1494 if (!gentle
&& fd
< 0)
1496 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1498 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1504 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1505 const struct pathspec
*pathspec
,
1506 char *seen
, const char *header_msg
)
1510 int really
= (flags
& REFRESH_REALLY
) != 0;
1511 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1512 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1513 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1514 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1515 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1517 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1518 unsigned int options
= (CE_MATCH_REFRESH
|
1519 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1520 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1521 const char *modified_fmt
;
1522 const char *deleted_fmt
;
1523 const char *typechange_fmt
;
1524 const char *added_fmt
;
1525 const char *unmerged_fmt
;
1526 struct progress
*progress
= NULL
;
1527 int t2_sum_lstat
= 0;
1528 int t2_sum_scan
= 0;
1530 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1531 progress
= start_delayed_progress(_("Refresh index"),
1534 trace_performance_enter();
1535 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1536 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1537 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1538 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1539 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1541 * Use the multi-threaded preload_index() to refresh most of the
1542 * cache entries quickly then in the single threaded loop below,
1543 * we only have to do the special cases that are left.
1545 preload_index(istate
, pathspec
, 0);
1546 trace2_region_enter("index", "refresh", NULL
);
1548 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1549 struct cache_entry
*ce
, *new_entry
;
1550 int cache_errno
= 0;
1553 int t2_did_lstat
= 0;
1554 int t2_did_scan
= 0;
1556 ce
= istate
->cache
[i
];
1557 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1559 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1563 * If this entry is a sparse directory, then there isn't
1564 * any stat() information to update. Ignore the entry.
1566 if (S_ISSPARSEDIR(ce
->ce_mode
))
1569 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1573 while ((i
< istate
->cache_nr
) &&
1574 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1580 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1581 &first
, header_msg
);
1589 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1590 &cache_errno
, &changed
,
1591 &t2_did_lstat
, &t2_did_scan
);
1592 t2_sum_lstat
+= t2_did_lstat
;
1593 t2_sum_scan
+= t2_did_scan
;
1594 if (new_entry
== ce
)
1596 display_progress(progress
, i
);
1600 if (really
&& cache_errno
== EINVAL
) {
1601 /* If we are doing --really-refresh that
1602 * means the index is not valid anymore.
1604 ce
->ce_flags
&= ~CE_VALID
;
1605 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1606 mark_fsmonitor_invalid(istate
, ce
);
1607 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1612 if (cache_errno
== ENOENT
)
1614 else if (ce_intent_to_add(ce
))
1615 fmt
= added_fmt
; /* must be before other checks */
1616 else if (changed
& TYPE_CHANGED
)
1617 fmt
= typechange_fmt
;
1621 ce
->name
, in_porcelain
, &first
, header_msg
);
1626 replace_index_entry(istate
, i
, new_entry
);
1628 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1629 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1630 trace2_region_leave("index", "refresh", NULL
);
1631 display_progress(progress
, istate
->cache_nr
);
1632 stop_progress(&progress
);
1633 trace_performance_leave("refresh index");
1637 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1638 struct cache_entry
*ce
,
1639 unsigned int options
)
1641 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1645 /*****************************************************************
1647 *****************************************************************/
1649 #define INDEX_FORMAT_DEFAULT 3
1651 static unsigned int get_index_format_default(struct repository
*r
)
1653 char *envversion
= getenv("GIT_INDEX_VERSION");
1655 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1658 prepare_repo_settings(r
);
1660 if (r
->settings
.index_version
>= 0)
1661 version
= r
->settings
.index_version
;
1662 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1663 warning(_("index.version set, but the value is invalid.\n"
1664 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1665 return INDEX_FORMAT_DEFAULT
;
1670 version
= strtoul(envversion
, &endp
, 10);
1672 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1673 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1674 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1675 version
= INDEX_FORMAT_DEFAULT
;
1681 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1682 * Again - this is just a (very strong in practice) heuristic that
1683 * the inode hasn't changed.
1685 * We save the fields in big-endian order to allow using the
1686 * index file over NFS transparently.
1688 struct ondisk_cache_entry
{
1689 struct cache_time ctime
;
1690 struct cache_time mtime
;
1698 * unsigned char hash[hashsz];
1700 * if (flags & CE_EXTENDED)
1703 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1704 char name
[FLEX_ARRAY
];
1707 /* These are only used for v3 or lower */
1708 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1709 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1710 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1711 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1712 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1713 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1714 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1716 /* Allow fsck to force verification of the index checksum. */
1717 int verify_index_checksum
;
1719 /* Allow fsck to force verification of the cache entry order. */
1720 int verify_ce_order
;
1722 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1725 unsigned char hash
[GIT_MAX_RAWSZ
];
1727 unsigned char *start
, *end
;
1728 struct object_id oid
;
1730 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1731 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1732 hdr_version
= ntohl(hdr
->hdr_version
);
1733 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1734 return error(_("bad index version %d"), hdr_version
);
1736 if (!verify_index_checksum
)
1739 end
= (unsigned char *)hdr
+ size
;
1740 start
= end
- the_hash_algo
->rawsz
;
1741 oidread(&oid
, start
);
1742 if (oideq(&oid
, null_oid()))
1745 the_hash_algo
->init_fn(&c
);
1746 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1747 the_hash_algo
->final_fn(hash
, &c
);
1748 if (!hasheq(hash
, start
))
1749 return error(_("bad index file sha1 signature"));
1753 static int read_index_extension(struct index_state
*istate
,
1754 const char *ext
, const char *data
, unsigned long sz
)
1756 switch (CACHE_EXT(ext
)) {
1757 case CACHE_EXT_TREE
:
1758 istate
->cache_tree
= cache_tree_read(data
, sz
);
1760 case CACHE_EXT_RESOLVE_UNDO
:
1761 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1763 case CACHE_EXT_LINK
:
1764 if (read_link_extension(istate
, data
, sz
))
1767 case CACHE_EXT_UNTRACKED
:
1768 istate
->untracked
= read_untracked_extension(data
, sz
);
1770 case CACHE_EXT_FSMONITOR
:
1771 read_fsmonitor_extension(istate
, data
, sz
);
1773 case CACHE_EXT_ENDOFINDEXENTRIES
:
1774 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1775 /* already handled in do_read_index() */
1777 case CACHE_EXT_SPARSE_DIRECTORIES
:
1778 /* no content, only an indicator */
1779 istate
->sparse_index
= INDEX_COLLAPSED
;
1782 if (*ext
< 'A' || 'Z' < *ext
)
1783 return error(_("index uses %.4s extension, which we do not understand"),
1785 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1792 * Parses the contents of the cache entry contained within the 'ondisk' buffer
1793 * into a new incore 'cache_entry'.
1795 * Note that 'char *ondisk' may not be aligned to a 4-byte address interval in
1796 * index v4, so we cannot cast it to 'struct ondisk_cache_entry *' and access
1797 * its members. Instead, we use the byte offsets of members within the struct to
1798 * identify where 'get_be16()', 'get_be32()', and 'oidread()' (which can all
1799 * read from an unaligned memory buffer) should read from the 'ondisk' buffer
1800 * into the corresponding incore 'cache_entry' members.
1802 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1803 unsigned int version
,
1805 unsigned long *ent_size
,
1806 const struct cache_entry
*previous_ce
)
1808 struct cache_entry
*ce
;
1811 const unsigned hashsz
= the_hash_algo
->rawsz
;
1812 const char *flagsp
= ondisk
+ offsetof(struct ondisk_cache_entry
, data
) + hashsz
;
1814 size_t copy_len
= 0;
1816 * Adjacent cache entries tend to share the leading paths, so it makes
1817 * sense to only store the differences in later entries. In the v4
1818 * on-disk format of the index, each on-disk cache entry stores the
1819 * number of bytes to be stripped from the end of the previous name,
1820 * and the bytes to append to the result, to come up with its name.
1822 int expand_name_field
= version
== 4;
1824 /* On-disk flags are just 16 bits */
1825 flags
= get_be16(flagsp
);
1826 len
= flags
& CE_NAMEMASK
;
1828 if (flags
& CE_EXTENDED
) {
1830 extended_flags
= get_be16(flagsp
+ sizeof(uint16_t)) << 16;
1831 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1832 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1833 die(_("unknown index entry format 0x%08x"), extended_flags
);
1834 flags
|= extended_flags
;
1835 name
= (const char *)(flagsp
+ 2 * sizeof(uint16_t));
1838 name
= (const char *)(flagsp
+ sizeof(uint16_t));
1840 if (expand_name_field
) {
1841 const unsigned char *cp
= (const unsigned char *)name
;
1842 size_t strip_len
, previous_len
;
1844 /* If we're at the beginning of a block, ignore the previous name */
1845 strip_len
= decode_varint(&cp
);
1847 previous_len
= previous_ce
->ce_namelen
;
1848 if (previous_len
< strip_len
)
1849 die(_("malformed name field in the index, near path '%s'"),
1851 copy_len
= previous_len
- strip_len
;
1853 name
= (const char *)cp
;
1856 if (len
== CE_NAMEMASK
) {
1858 if (expand_name_field
)
1862 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1865 * NEEDSWORK: using 'offsetof()' is cumbersome and should be replaced
1866 * with something more akin to 'load_bitmap_entries_v1()'s use of
1867 * 'read_be16'/'read_be32'. For consistency with the corresponding
1868 * ondisk entry write function ('copy_cache_entry_to_ondisk()'), this
1869 * should be done at the same time as removing references to
1870 * 'ondisk_cache_entry' there.
1872 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1873 + offsetof(struct cache_time
, sec
));
1874 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1875 + offsetof(struct cache_time
, sec
));
1876 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1877 + offsetof(struct cache_time
, nsec
));
1878 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1879 + offsetof(struct cache_time
, nsec
));
1880 ce
->ce_stat_data
.sd_dev
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, dev
));
1881 ce
->ce_stat_data
.sd_ino
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ino
));
1882 ce
->ce_mode
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mode
));
1883 ce
->ce_stat_data
.sd_uid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, uid
));
1884 ce
->ce_stat_data
.sd_gid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, gid
));
1885 ce
->ce_stat_data
.sd_size
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, size
));
1886 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1887 ce
->ce_namelen
= len
;
1889 oidread(&ce
->oid
, (const unsigned char *)ondisk
+ offsetof(struct ondisk_cache_entry
, data
));
1891 if (expand_name_field
) {
1893 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1894 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1895 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1897 memcpy(ce
->name
, name
, len
+ 1);
1898 *ent_size
= ondisk_ce_size(ce
);
1903 static void check_ce_order(struct index_state
*istate
)
1907 if (!verify_ce_order
)
1910 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1911 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1912 struct cache_entry
*next_ce
= istate
->cache
[i
];
1913 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1915 if (0 < name_compare
)
1916 die(_("unordered stage entries in index"));
1917 if (!name_compare
) {
1919 die(_("multiple stage entries for merged file '%s'"),
1921 if (ce_stage(ce
) > ce_stage(next_ce
))
1922 die(_("unordered stage entries for '%s'"),
1928 static void tweak_untracked_cache(struct index_state
*istate
)
1930 struct repository
*r
= the_repository
;
1932 prepare_repo_settings(r
);
1934 switch (r
->settings
.core_untracked_cache
) {
1935 case UNTRACKED_CACHE_REMOVE
:
1936 remove_untracked_cache(istate
);
1938 case UNTRACKED_CACHE_WRITE
:
1939 add_untracked_cache(istate
);
1941 case UNTRACKED_CACHE_KEEP
:
1943 * Either an explicit "core.untrackedCache=keep", the
1944 * default if "core.untrackedCache" isn't configured,
1945 * or a fallback on an unknown "core.untrackedCache"
1952 static void tweak_split_index(struct index_state
*istate
)
1954 switch (git_config_get_split_index()) {
1955 case -1: /* unset: do nothing */
1958 remove_split_index(istate
);
1961 add_split_index(istate
);
1963 default: /* unknown value: do nothing */
1968 static void post_read_index_from(struct index_state
*istate
)
1970 check_ce_order(istate
);
1971 tweak_untracked_cache(istate
);
1972 tweak_split_index(istate
);
1973 tweak_fsmonitor(istate
);
1976 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1978 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1981 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1983 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1986 * Account for potential alignment differences.
1988 per_entry
+= align_padding_size(per_entry
, 0);
1989 return ondisk_size
+ entries
* per_entry
;
1992 struct index_entry_offset
1994 /* starting byte offset into index file, count of index entries in this block */
1998 struct index_entry_offset_table
2001 struct index_entry_offset entries
[FLEX_ARRAY
];
2004 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2005 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2007 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2008 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2010 struct load_index_extensions
2013 struct index_state
*istate
;
2016 unsigned long src_offset
;
2019 static void *load_index_extensions(void *_data
)
2021 struct load_index_extensions
*p
= _data
;
2022 unsigned long src_offset
= p
->src_offset
;
2024 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2025 /* After an array of active_nr index entries,
2026 * there can be arbitrary number of extended
2027 * sections, each of which is prefixed with
2028 * extension name (4-byte) and section length
2029 * in 4-byte network byte order.
2031 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2032 if (read_index_extension(p
->istate
,
2033 p
->mmap
+ src_offset
,
2034 p
->mmap
+ src_offset
+ 8,
2036 munmap((void *)p
->mmap
, p
->mmap_size
);
2037 die(_("index file corrupt"));
2040 src_offset
+= extsize
;
2047 * A helper function that will load the specified range of cache entries
2048 * from the memory mapped file and add them to the given index.
2050 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2051 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2052 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2055 unsigned long src_offset
= start_offset
;
2057 for (i
= offset
; i
< offset
+ nr
; i
++) {
2058 struct cache_entry
*ce
;
2059 unsigned long consumed
;
2061 ce
= create_from_disk(ce_mem_pool
, istate
->version
,
2063 &consumed
, previous_ce
);
2064 set_index_entry(istate
, i
, ce
);
2066 src_offset
+= consumed
;
2069 return src_offset
- start_offset
;
2072 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2073 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2075 unsigned long consumed
;
2077 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2078 if (istate
->version
== 4) {
2079 mem_pool_init(istate
->ce_mem_pool
,
2080 estimate_cache_size_from_compressed(istate
->cache_nr
));
2082 mem_pool_init(istate
->ce_mem_pool
,
2083 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2086 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2087 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2092 * Mostly randomly chosen maximum thread counts: we
2093 * cap the parallelism to online_cpus() threads, and we want
2094 * to have at least 10000 cache entries per thread for it to
2095 * be worth starting a thread.
2098 #define THREAD_COST (10000)
2100 struct load_cache_entries_thread_data
2103 struct index_state
*istate
;
2104 struct mem_pool
*ce_mem_pool
;
2107 struct index_entry_offset_table
*ieot
;
2108 int ieot_start
; /* starting index into the ieot array */
2109 int ieot_blocks
; /* count of ieot entries to process */
2110 unsigned long consumed
; /* return # of bytes in index file processed */
2114 * A thread proc to run the load_cache_entries() computation
2115 * across multiple background threads.
2117 static void *load_cache_entries_thread(void *_data
)
2119 struct load_cache_entries_thread_data
*p
= _data
;
2122 /* iterate across all ieot blocks assigned to this thread */
2123 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2124 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2125 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2126 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2131 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2132 int nr_threads
, struct index_entry_offset_table
*ieot
)
2134 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2135 struct load_cache_entries_thread_data
*data
;
2136 unsigned long consumed
= 0;
2138 /* a little sanity checking */
2139 if (istate
->name_hash_initialized
)
2140 BUG("the name hash isn't thread safe");
2142 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2143 mem_pool_init(istate
->ce_mem_pool
, 0);
2145 /* ensure we have no more threads than we have blocks to process */
2146 if (nr_threads
> ieot
->nr
)
2147 nr_threads
= ieot
->nr
;
2148 CALLOC_ARRAY(data
, nr_threads
);
2150 offset
= ieot_start
= 0;
2151 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2152 for (i
= 0; i
< nr_threads
; i
++) {
2153 struct load_cache_entries_thread_data
*p
= &data
[i
];
2156 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2157 ieot_blocks
= ieot
->nr
- ieot_start
;
2163 p
->ieot_start
= ieot_start
;
2164 p
->ieot_blocks
= ieot_blocks
;
2166 /* create a mem_pool for each thread */
2168 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2169 nr
+= p
->ieot
->entries
[j
].nr
;
2170 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2171 if (istate
->version
== 4) {
2172 mem_pool_init(p
->ce_mem_pool
,
2173 estimate_cache_size_from_compressed(nr
));
2175 mem_pool_init(p
->ce_mem_pool
,
2176 estimate_cache_size(mmap_size
, nr
));
2179 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2181 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2183 /* increment by the number of cache entries in the ieot block being processed */
2184 for (j
= 0; j
< ieot_blocks
; j
++)
2185 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2186 ieot_start
+= ieot_blocks
;
2189 for (i
= 0; i
< nr_threads
; i
++) {
2190 struct load_cache_entries_thread_data
*p
= &data
[i
];
2192 err
= pthread_join(p
->pthread
, NULL
);
2194 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2195 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2196 consumed
+= p
->consumed
;
2204 static void set_new_index_sparsity(struct index_state
*istate
)
2207 * If the index's repo exists, mark it sparse according to
2210 prepare_repo_settings(istate
->repo
);
2211 if (!istate
->repo
->settings
.command_requires_full_index
&&
2212 is_sparse_index_allowed(istate
, 0))
2213 istate
->sparse_index
= 1;
2216 /* remember to discard_cache() before reading a different cache! */
2217 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2221 unsigned long src_offset
;
2222 const struct cache_header
*hdr
;
2225 struct load_index_extensions p
;
2226 size_t extension_offset
= 0;
2227 int nr_threads
, cpus
;
2228 struct index_entry_offset_table
*ieot
= NULL
;
2230 if (istate
->initialized
)
2231 return istate
->cache_nr
;
2233 istate
->timestamp
.sec
= 0;
2234 istate
->timestamp
.nsec
= 0;
2235 fd
= open(path
, O_RDONLY
);
2237 if (!must_exist
&& errno
== ENOENT
) {
2238 set_new_index_sparsity(istate
);
2239 istate
->initialized
= 1;
2242 die_errno(_("%s: index file open failed"), path
);
2246 die_errno(_("%s: cannot stat the open index"), path
);
2248 mmap_size
= xsize_t(st
.st_size
);
2249 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2250 die(_("%s: index file smaller than expected"), path
);
2252 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2253 if (mmap
== MAP_FAILED
)
2254 die_errno(_("%s: unable to map index file%s"), path
,
2258 hdr
= (const struct cache_header
*)mmap
;
2259 if (verify_hdr(hdr
, mmap_size
) < 0)
2262 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2263 istate
->version
= ntohl(hdr
->hdr_version
);
2264 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2265 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2266 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2267 istate
->initialized
= 1;
2271 p
.mmap_size
= mmap_size
;
2273 src_offset
= sizeof(*hdr
);
2275 if (git_config_get_index_threads(&nr_threads
))
2278 /* TODO: does creating more threads than cores help? */
2280 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2281 cpus
= online_cpus();
2282 if (nr_threads
> cpus
)
2289 if (nr_threads
> 1) {
2290 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2291 if (extension_offset
) {
2294 p
.src_offset
= extension_offset
;
2295 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2297 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2304 * Locate and read the index entry offset table so that we can use it
2305 * to multi-thread the reading of the cache entries.
2307 if (extension_offset
&& nr_threads
> 1)
2308 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2311 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2314 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2317 istate
->timestamp
.sec
= st
.st_mtime
;
2318 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2320 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2321 if (extension_offset
) {
2322 int ret
= pthread_join(p
.pthread
, NULL
);
2324 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2326 p
.src_offset
= src_offset
;
2327 load_index_extensions(&p
);
2329 munmap((void *)mmap
, mmap_size
);
2332 * TODO trace2: replace "the_repository" with the actual repo instance
2333 * that is associated with the given "istate".
2335 trace2_data_intmax("index", the_repository
, "read/version",
2337 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2341 * If the command explicitly requires a full index, force it
2342 * to be full. Otherwise, correct the sparsity based on repository
2343 * settings and other properties of the index (if necessary).
2345 prepare_repo_settings(istate
->repo
);
2346 if (istate
->repo
->settings
.command_requires_full_index
)
2347 ensure_full_index(istate
);
2349 ensure_correct_sparsity(istate
);
2351 return istate
->cache_nr
;
2354 munmap((void *)mmap
, mmap_size
);
2355 die(_("index file corrupt"));
2359 * Signal that the shared index is used by updating its mtime.
2361 * This way, shared index can be removed if they have not been used
2364 static void freshen_shared_index(const char *shared_index
, int warn
)
2366 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2367 warning(_("could not freshen shared index '%s'"), shared_index
);
2370 int read_index_from(struct index_state
*istate
, const char *path
,
2373 struct split_index
*split_index
;
2378 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2379 if (istate
->initialized
)
2380 return istate
->cache_nr
;
2383 * TODO trace2: replace "the_repository" with the actual repo instance
2384 * that is associated with the given "istate".
2386 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2388 trace_performance_enter();
2389 ret
= do_read_index(istate
, path
, 0);
2390 trace_performance_leave("read cache %s", path
);
2391 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2394 split_index
= istate
->split_index
;
2395 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2396 post_read_index_from(istate
);
2400 trace_performance_enter();
2401 if (split_index
->base
)
2402 release_index(split_index
->base
);
2404 ALLOC_ARRAY(split_index
->base
, 1);
2405 index_state_init(split_index
->base
, istate
->repo
);
2407 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2408 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2409 if (file_exists(base_path
)) {
2410 trace2_region_enter_printf("index", "shared/do_read_index",
2411 the_repository
, "%s", base_path
);
2413 ret
= do_read_index(split_index
->base
, base_path
, 0);
2414 trace2_region_leave_printf("index", "shared/do_read_index",
2415 the_repository
, "%s", base_path
);
2417 char *path_copy
= xstrdup(path
);
2418 char *base_path2
= xstrfmt("%s/sharedindex.%s",
2419 dirname(path_copy
), base_oid_hex
);
2421 trace2_region_enter_printf("index", "shared/do_read_index",
2422 the_repository
, "%s", base_path2
);
2423 ret
= do_read_index(split_index
->base
, base_path2
, 1);
2424 trace2_region_leave_printf("index", "shared/do_read_index",
2425 the_repository
, "%s", base_path2
);
2428 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2429 die(_("broken index, expect %s in %s, got %s"),
2430 base_oid_hex
, base_path
,
2431 oid_to_hex(&split_index
->base
->oid
));
2433 freshen_shared_index(base_path
, 0);
2434 merge_base_index(istate
);
2435 post_read_index_from(istate
);
2436 trace_performance_leave("read cache %s", base_path
);
2441 int is_index_unborn(struct index_state
*istate
)
2443 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2446 void index_state_init(struct index_state
*istate
, struct repository
*r
)
2448 struct index_state blank
= INDEX_STATE_INIT(r
);
2449 memcpy(istate
, &blank
, sizeof(*istate
));
2452 void release_index(struct index_state
*istate
)
2455 * Cache entries in istate->cache[] should have been allocated
2456 * from the memory pool associated with this index, or from an
2457 * associated split_index. There is no need to free individual
2458 * cache entries. validate_cache_entries can detect when this
2459 * assertion does not hold.
2461 validate_cache_entries(istate
);
2463 resolve_undo_clear_index(istate
);
2464 free_name_hash(istate
);
2465 cache_tree_free(&(istate
->cache_tree
));
2466 free(istate
->fsmonitor_last_update
);
2467 free(istate
->cache
);
2468 discard_split_index(istate
);
2469 free_untracked_cache(istate
->untracked
);
2471 if (istate
->sparse_checkout_patterns
) {
2472 clear_pattern_list(istate
->sparse_checkout_patterns
);
2473 FREE_AND_NULL(istate
->sparse_checkout_patterns
);
2476 if (istate
->ce_mem_pool
) {
2477 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2478 FREE_AND_NULL(istate
->ce_mem_pool
);
2482 void discard_index(struct index_state
*istate
)
2484 release_index(istate
);
2485 index_state_init(istate
, istate
->repo
);
2489 * Validate the cache entries of this index.
2490 * All cache entries associated with this index
2491 * should have been allocated by the memory pool
2492 * associated with this index, or by a referenced
2495 void validate_cache_entries(const struct index_state
*istate
)
2499 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2502 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2504 BUG("cache entry is not allocated from expected memory pool");
2505 } else if (!istate
->ce_mem_pool
||
2506 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2507 if (!istate
->split_index
||
2508 !istate
->split_index
->base
||
2509 !istate
->split_index
->base
->ce_mem_pool
||
2510 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2511 BUG("cache entry is not allocated from expected memory pool");
2516 if (istate
->split_index
)
2517 validate_cache_entries(istate
->split_index
->base
);
2520 int unmerged_index(const struct index_state
*istate
)
2523 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2524 if (ce_stage(istate
->cache
[i
]))
2530 int repo_index_has_changes(struct repository
*repo
,
2534 struct index_state
*istate
= repo
->index
;
2535 struct object_id cmp
;
2539 cmp
= tree
->object
.oid
;
2540 if (tree
|| !repo_get_oid_tree(repo
, "HEAD", &cmp
)) {
2541 struct diff_options opt
;
2543 repo_diff_setup(repo
, &opt
);
2544 opt
.flags
.exit_with_status
= 1;
2546 opt
.flags
.quick
= 1;
2547 diff_setup_done(&opt
);
2548 do_diff_cache(&cmp
, &opt
);
2550 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2552 strbuf_addch(sb
, ' ');
2553 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2556 return opt
.flags
.has_changes
!= 0;
2558 /* TODO: audit for interaction with sparse-index. */
2559 ensure_full_index(istate
);
2560 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2562 strbuf_addch(sb
, ' ');
2563 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2565 return !!istate
->cache_nr
;
2569 static int write_index_ext_header(struct hashfile
*f
,
2570 git_hash_ctx
*eoie_f
,
2574 hashwrite_be32(f
, ext
);
2575 hashwrite_be32(f
, sz
);
2580 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2581 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2586 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2587 struct cache_entry
*ce
)
2590 * The only thing we care about in this function is to smudge the
2591 * falsely clean entry due to touch-update-touch race, so we leave
2592 * everything else as they are. We are called for entries whose
2593 * ce_stat_data.sd_mtime match the index file mtime.
2595 * Note that this actually does not do much for gitlinks, for
2596 * which ce_match_stat_basic() always goes to the actual
2597 * contents. The caller checks with is_racy_timestamp() which
2598 * always says "no" for gitlinks, so we are not called for them ;-)
2602 if (lstat(ce
->name
, &st
) < 0)
2604 if (ce_match_stat_basic(ce
, &st
))
2606 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2607 /* This is "racily clean"; smudge it. Note that this
2608 * is a tricky code. At first glance, it may appear
2609 * that it can break with this sequence:
2611 * $ echo xyzzy >frotz
2612 * $ git-update-index --add frotz
2615 * $ echo filfre >nitfol
2616 * $ git-update-index --add nitfol
2618 * but it does not. When the second update-index runs,
2619 * it notices that the entry "frotz" has the same timestamp
2620 * as index, and if we were to smudge it by resetting its
2621 * size to zero here, then the object name recorded
2622 * in index is the 6-byte file but the cached stat information
2623 * becomes zero --- which would then match what we would
2624 * obtain from the filesystem next time we stat("frotz").
2626 * However, the second update-index, before calling
2627 * this function, notices that the cached size is 6
2628 * bytes and what is on the filesystem is an empty
2629 * file, and never calls us, so the cached size information
2630 * for "frotz" stays 6 which does not match the filesystem.
2632 ce
->ce_stat_data
.sd_size
= 0;
2636 /* Copy miscellaneous fields but not the name */
2637 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2638 struct cache_entry
*ce
)
2641 const unsigned hashsz
= the_hash_algo
->rawsz
;
2642 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2644 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2645 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2646 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2647 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2648 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2649 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2650 ondisk
->mode
= htonl(ce
->ce_mode
);
2651 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2652 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2653 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2654 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2656 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2657 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2658 flagsp
[0] = htons(flags
);
2659 if (ce
->ce_flags
& CE_EXTENDED
) {
2660 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2664 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2665 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2668 unsigned int saved_namelen
;
2669 int stripped_name
= 0;
2670 static unsigned char padding
[8] = { 0x00 };
2672 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2673 saved_namelen
= ce_namelen(ce
);
2678 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2680 if (!previous_name
) {
2681 int len
= ce_namelen(ce
);
2682 copy_cache_entry_to_ondisk(ondisk
, ce
);
2683 hashwrite(f
, ondisk
, size
);
2684 hashwrite(f
, ce
->name
, len
);
2685 hashwrite(f
, padding
, align_padding_size(size
, len
));
2687 int common
, to_remove
, prefix_size
;
2688 unsigned char to_remove_vi
[16];
2690 (ce
->name
[common
] &&
2691 common
< previous_name
->len
&&
2692 ce
->name
[common
] == previous_name
->buf
[common
]);
2694 ; /* still matching */
2695 to_remove
= previous_name
->len
- common
;
2696 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2698 copy_cache_entry_to_ondisk(ondisk
, ce
);
2699 hashwrite(f
, ondisk
, size
);
2700 hashwrite(f
, to_remove_vi
, prefix_size
);
2701 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2702 hashwrite(f
, padding
, 1);
2704 strbuf_splice(previous_name
, common
, to_remove
,
2705 ce
->name
+ common
, ce_namelen(ce
) - common
);
2707 if (stripped_name
) {
2708 ce
->ce_namelen
= saved_namelen
;
2709 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2716 * This function verifies if index_state has the correct sha1 of the
2717 * index file. Don't die if we have any other failure, just return 0.
2719 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2724 unsigned char hash
[GIT_MAX_RAWSZ
];
2726 if (!istate
->initialized
)
2729 fd
= open(path
, O_RDONLY
);
2736 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2739 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2740 if (n
!= the_hash_algo
->rawsz
)
2743 if (!hasheq(istate
->oid
.hash
, hash
))
2754 static int repo_verify_index(struct repository
*repo
)
2756 return verify_index_from(repo
->index
, repo
->index_file
);
2759 int has_racy_timestamp(struct index_state
*istate
)
2761 int entries
= istate
->cache_nr
;
2764 for (i
= 0; i
< entries
; i
++) {
2765 struct cache_entry
*ce
= istate
->cache
[i
];
2766 if (is_racy_timestamp(istate
, ce
))
2772 void repo_update_index_if_able(struct repository
*repo
,
2773 struct lock_file
*lockfile
)
2775 if ((repo
->index
->cache_changed
||
2776 has_racy_timestamp(repo
->index
)) &&
2777 repo_verify_index(repo
))
2778 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2780 rollback_lock_file(lockfile
);
2783 static int record_eoie(void)
2787 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2791 * As a convenience, the end of index entries extension
2792 * used for threading is written by default if the user
2793 * explicitly requested threaded index reads.
2795 return !git_config_get_index_threads(&val
) && val
!= 1;
2798 static int record_ieot(void)
2802 if (!git_config_get_bool("index.recordoffsettable", &val
))
2806 * As a convenience, the offset table used for threading is
2807 * written by default if the user explicitly requested
2808 * threaded index reads.
2810 return !git_config_get_index_threads(&val
) && val
!= 1;
2813 enum write_extensions
{
2814 WRITE_NO_EXTENSION
= 0,
2815 WRITE_SPLIT_INDEX_EXTENSION
= 1<<0,
2816 WRITE_CACHE_TREE_EXTENSION
= 1<<1,
2817 WRITE_RESOLVE_UNDO_EXTENSION
= 1<<2,
2818 WRITE_UNTRACKED_CACHE_EXTENSION
= 1<<3,
2819 WRITE_FSMONITOR_EXTENSION
= 1<<4,
2821 #define WRITE_ALL_EXTENSIONS ((enum write_extensions)-1)
2824 * On success, `tempfile` is closed. If it is the temporary file
2825 * of a `struct lock_file`, we will therefore effectively perform
2826 * a 'close_lock_file_gently()`. Since that is an implementation
2827 * detail of lockfiles, callers of `do_write_index()` should not
2830 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2831 enum write_extensions write_extensions
, unsigned flags
)
2833 uint64_t start
= getnanotime();
2835 git_hash_ctx
*eoie_c
= NULL
;
2836 struct cache_header hdr
;
2837 int i
, err
= 0, removed
, extended
, hdr_version
;
2838 struct cache_entry
**cache
= istate
->cache
;
2839 int entries
= istate
->cache_nr
;
2841 struct ondisk_cache_entry ondisk
;
2842 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2843 int drop_cache_tree
= istate
->drop_cache_tree
;
2845 int csum_fsync_flag
;
2846 int ieot_entries
= 1;
2847 struct index_entry_offset_table
*ieot
= NULL
;
2849 struct repository
*r
= istate
->repo
;
2851 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2853 prepare_repo_settings(r
);
2854 f
->skip_hash
= r
->settings
.index_skip_hash
;
2856 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2857 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2860 /* reduce extended entries if possible */
2861 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2862 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2864 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2868 if (!istate
->version
)
2869 istate
->version
= get_index_format_default(r
);
2871 /* demote version 3 to version 2 when the latter suffices */
2872 if (istate
->version
== 3 || istate
->version
== 2)
2873 istate
->version
= extended
? 3 : 2;
2875 hdr_version
= istate
->version
;
2877 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2878 hdr
.hdr_version
= htonl(hdr_version
);
2879 hdr
.hdr_entries
= htonl(entries
- removed
);
2881 hashwrite(f
, &hdr
, sizeof(hdr
));
2883 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2886 if (nr_threads
!= 1 && record_ieot()) {
2887 int ieot_blocks
, cpus
;
2890 * ensure default number of ieot blocks maps evenly to the
2891 * default number of threads that will process them leaving
2892 * room for the thread to load the index extensions.
2895 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2896 cpus
= online_cpus();
2897 if (ieot_blocks
> cpus
- 1)
2898 ieot_blocks
= cpus
- 1;
2900 ieot_blocks
= nr_threads
;
2901 if (ieot_blocks
> istate
->cache_nr
)
2902 ieot_blocks
= istate
->cache_nr
;
2906 * no reason to write out the IEOT extension if we don't
2907 * have enough blocks to utilize multi-threading
2909 if (ieot_blocks
> 1) {
2910 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2911 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2912 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2916 offset
= hashfile_total(f
);
2919 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2921 for (i
= 0; i
< entries
; i
++) {
2922 struct cache_entry
*ce
= cache
[i
];
2923 if (ce
->ce_flags
& CE_REMOVE
)
2925 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2926 ce_smudge_racily_clean_entry(istate
, ce
);
2927 if (is_null_oid(&ce
->oid
)) {
2928 static const char msg
[] = "cache entry has null sha1: %s";
2929 static int allow
= -1;
2932 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2934 warning(msg
, ce
->name
);
2936 err
= error(msg
, ce
->name
);
2938 drop_cache_tree
= 1;
2940 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2941 ieot
->entries
[ieot
->nr
].nr
= nr
;
2942 ieot
->entries
[ieot
->nr
].offset
= offset
;
2945 * If we have a V4 index, set the first byte to an invalid
2946 * character to ensure there is nothing common with the previous
2950 previous_name
->buf
[0] = 0;
2953 offset
= hashfile_total(f
);
2955 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2963 ieot
->entries
[ieot
->nr
].nr
= nr
;
2964 ieot
->entries
[ieot
->nr
].offset
= offset
;
2967 strbuf_release(&previous_name_buf
);
2974 offset
= hashfile_total(f
);
2977 * The extension headers must be hashed on their own for the
2978 * EOIE extension. Create a hashfile here to compute that hash.
2980 if (offset
&& record_eoie()) {
2981 CALLOC_ARRAY(eoie_c
, 1);
2982 the_hash_algo
->init_fn(eoie_c
);
2986 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2987 * can minimize the number of extensions we have to scan through to
2988 * find it during load. Write it out regardless of the
2989 * strip_extensions parameter as we need it when loading the shared
2993 struct strbuf sb
= STRBUF_INIT
;
2995 write_ieot_extension(&sb
, ieot
);
2996 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
2997 hashwrite(f
, sb
.buf
, sb
.len
);
2998 strbuf_release(&sb
);
3004 if (write_extensions
& WRITE_SPLIT_INDEX_EXTENSION
&&
3005 istate
->split_index
) {
3006 struct strbuf sb
= STRBUF_INIT
;
3008 if (istate
->sparse_index
)
3009 die(_("cannot write split index for a sparse index"));
3011 err
= write_link_extension(&sb
, istate
) < 0 ||
3012 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
3014 hashwrite(f
, sb
.buf
, sb
.len
);
3015 strbuf_release(&sb
);
3019 if (write_extensions
& WRITE_CACHE_TREE_EXTENSION
&&
3020 !drop_cache_tree
&& istate
->cache_tree
) {
3021 struct strbuf sb
= STRBUF_INIT
;
3023 cache_tree_write(&sb
, istate
->cache_tree
);
3024 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
3025 hashwrite(f
, sb
.buf
, sb
.len
);
3026 strbuf_release(&sb
);
3030 if (write_extensions
& WRITE_RESOLVE_UNDO_EXTENSION
&&
3031 istate
->resolve_undo
) {
3032 struct strbuf sb
= STRBUF_INIT
;
3034 resolve_undo_write(&sb
, istate
->resolve_undo
);
3035 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3037 hashwrite(f
, sb
.buf
, sb
.len
);
3038 strbuf_release(&sb
);
3042 if (write_extensions
& WRITE_UNTRACKED_CACHE_EXTENSION
&&
3043 istate
->untracked
) {
3044 struct strbuf sb
= STRBUF_INIT
;
3046 write_untracked_extension(&sb
, istate
->untracked
);
3047 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3049 hashwrite(f
, sb
.buf
, sb
.len
);
3050 strbuf_release(&sb
);
3054 if (write_extensions
& WRITE_FSMONITOR_EXTENSION
&&
3055 istate
->fsmonitor_last_update
) {
3056 struct strbuf sb
= STRBUF_INIT
;
3058 write_fsmonitor_extension(&sb
, istate
);
3059 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3060 hashwrite(f
, sb
.buf
, sb
.len
);
3061 strbuf_release(&sb
);
3065 if (istate
->sparse_index
) {
3066 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3071 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3072 * so that it can be found and processed before all the index entries are
3073 * read. Write it out regardless of the strip_extensions parameter as we need it
3074 * when loading the shared index.
3077 struct strbuf sb
= STRBUF_INIT
;
3079 write_eoie_extension(&sb
, eoie_c
, offset
);
3080 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3081 hashwrite(f
, sb
.buf
, sb
.len
);
3082 strbuf_release(&sb
);
3087 csum_fsync_flag
= 0;
3088 if (!alternate_index_output
&& (flags
& COMMIT_LOCK
))
3089 csum_fsync_flag
= CSUM_FSYNC
;
3091 finalize_hashfile(f
, istate
->oid
.hash
, FSYNC_COMPONENT_INDEX
,
3092 CSUM_HASH_IN_STREAM
| csum_fsync_flag
);
3094 if (close_tempfile_gently(tempfile
)) {
3095 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3098 if (stat(get_tempfile_path(tempfile
), &st
))
3100 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3101 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3102 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3105 * TODO trace2: replace "the_repository" with the actual repo instance
3106 * that is associated with the given "istate".
3108 trace2_data_intmax("index", the_repository
, "write/version",
3110 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3116 void set_alternate_index_output(const char *name
)
3118 alternate_index_output
= name
;
3121 static int commit_locked_index(struct lock_file
*lk
)
3123 if (alternate_index_output
)
3124 return commit_lock_file_to(lk
, alternate_index_output
);
3126 return commit_lock_file(lk
);
3129 static int do_write_locked_index(struct index_state
*istate
,
3130 struct lock_file
*lock
,
3132 enum write_extensions write_extensions
)
3135 int was_full
= istate
->sparse_index
== INDEX_EXPANDED
;
3137 ret
= convert_to_sparse(istate
, 0);
3140 warning(_("failed to convert to a sparse-index"));
3145 * TODO trace2: replace "the_repository" with the actual repo instance
3146 * that is associated with the given "istate".
3148 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3149 "%s", get_lock_file_path(lock
));
3150 ret
= do_write_index(istate
, lock
->tempfile
, write_extensions
, flags
);
3151 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3152 "%s", get_lock_file_path(lock
));
3155 ensure_full_index(istate
);
3159 if (flags
& COMMIT_LOCK
)
3160 ret
= commit_locked_index(lock
);
3162 ret
= close_lock_file_gently(lock
);
3164 run_hooks_l("post-index-change",
3165 istate
->updated_workdir
? "1" : "0",
3166 istate
->updated_skipworktree
? "1" : "0", NULL
);
3167 istate
->updated_workdir
= 0;
3168 istate
->updated_skipworktree
= 0;
3173 static int write_split_index(struct index_state
*istate
,
3174 struct lock_file
*lock
,
3178 prepare_to_write_split_index(istate
);
3179 ret
= do_write_locked_index(istate
, lock
, flags
, WRITE_ALL_EXTENSIONS
);
3180 finish_writing_split_index(istate
);
3184 static const char *shared_index_expire
= "2.weeks.ago";
3186 static unsigned long get_shared_index_expire_date(void)
3188 static unsigned long shared_index_expire_date
;
3189 static int shared_index_expire_date_prepared
;
3191 if (!shared_index_expire_date_prepared
) {
3192 git_config_get_expiry("splitindex.sharedindexexpire",
3193 &shared_index_expire
);
3194 shared_index_expire_date
= approxidate(shared_index_expire
);
3195 shared_index_expire_date_prepared
= 1;
3198 return shared_index_expire_date
;
3201 static int should_delete_shared_index(const char *shared_index_path
)
3204 unsigned long expiration
;
3206 /* Check timestamp */
3207 expiration
= get_shared_index_expire_date();
3210 if (stat(shared_index_path
, &st
))
3211 return error_errno(_("could not stat '%s'"), shared_index_path
);
3212 if (st
.st_mtime
> expiration
)
3218 static int clean_shared_index_files(const char *current_hex
)
3221 DIR *dir
= opendir(get_git_dir());
3224 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3226 while ((de
= readdir(dir
)) != NULL
) {
3227 const char *sha1_hex
;
3228 const char *shared_index_path
;
3229 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3231 if (!strcmp(sha1_hex
, current_hex
))
3233 shared_index_path
= git_path("%s", de
->d_name
);
3234 if (should_delete_shared_index(shared_index_path
) > 0 &&
3235 unlink(shared_index_path
))
3236 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3243 static int write_shared_index(struct index_state
*istate
,
3244 struct tempfile
**temp
, unsigned flags
)
3246 struct split_index
*si
= istate
->split_index
;
3247 int ret
, was_full
= !istate
->sparse_index
;
3249 move_cache_to_base_index(istate
);
3250 convert_to_sparse(istate
, 0);
3252 trace2_region_enter_printf("index", "shared/do_write_index",
3253 the_repository
, "%s", get_tempfile_path(*temp
));
3254 ret
= do_write_index(si
->base
, *temp
, WRITE_NO_EXTENSION
, flags
);
3255 trace2_region_leave_printf("index", "shared/do_write_index",
3256 the_repository
, "%s", get_tempfile_path(*temp
));
3259 ensure_full_index(istate
);
3263 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3265 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3268 ret
= rename_tempfile(temp
,
3269 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3271 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3272 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3278 static const int default_max_percent_split_change
= 20;
3280 static int too_many_not_shared_entries(struct index_state
*istate
)
3282 int i
, not_shared
= 0;
3283 int max_split
= git_config_get_max_percent_split_change();
3285 switch (max_split
) {
3287 /* not or badly configured: use the default value */
3288 max_split
= default_max_percent_split_change
;
3291 return 1; /* 0% means always write a new shared index */
3293 return 0; /* 100% means never write a new shared index */
3295 break; /* just use the configured value */
3298 /* Count not shared entries */
3299 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3300 struct cache_entry
*ce
= istate
->cache
[i
];
3305 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3308 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3311 int new_shared_index
, ret
, test_split_index_env
;
3312 struct split_index
*si
= istate
->split_index
;
3314 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3315 cache_tree_verify(the_repository
, istate
);
3317 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3318 if (flags
& COMMIT_LOCK
)
3319 rollback_lock_file(lock
);
3323 if (istate
->fsmonitor_last_update
)
3324 fill_fsmonitor_bitmap(istate
);
3326 test_split_index_env
= git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3328 if ((!si
&& !test_split_index_env
) ||
3329 alternate_index_output
||
3330 (istate
->cache_changed
& ~EXTMASK
)) {
3331 ret
= do_write_locked_index(istate
, lock
, flags
,
3332 ~WRITE_SPLIT_INDEX_EXTENSION
);
3336 if (test_split_index_env
) {
3338 si
= init_split_index(istate
);
3339 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3341 int v
= si
->base_oid
.hash
[0];
3343 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3346 if (too_many_not_shared_entries(istate
))
3347 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3349 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3351 if (new_shared_index
) {
3352 struct tempfile
*temp
;
3355 /* Same initial permissions as the main .git/index file */
3356 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3358 ret
= do_write_locked_index(istate
, lock
, flags
,
3359 ~WRITE_SPLIT_INDEX_EXTENSION
);
3362 ret
= write_shared_index(istate
, &temp
, flags
);
3364 saved_errno
= errno
;
3365 if (is_tempfile_active(temp
))
3366 delete_tempfile(&temp
);
3367 errno
= saved_errno
;
3373 ret
= write_split_index(istate
, lock
, flags
);
3375 /* Freshen the shared index only if the split-index was written */
3376 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3377 const char *shared_index
= git_path("sharedindex.%s",
3378 oid_to_hex(&si
->base_oid
));
3379 freshen_shared_index(shared_index
, 1);
3383 if (flags
& COMMIT_LOCK
)
3384 rollback_lock_file(lock
);
3389 * Read the index file that is potentially unmerged into given
3390 * index_state, dropping any unmerged entries to stage #0 (potentially
3391 * resulting in a path appearing as both a file and a directory in the
3392 * index; the caller is responsible to clear out the extra entries
3393 * before writing the index to a tree). Returns true if the index is
3394 * unmerged. Callers who want to refuse to work from an unmerged
3395 * state can call this and check its return value, instead of calling
3398 int repo_read_index_unmerged(struct repository
*repo
)
3400 struct index_state
*istate
;
3404 repo_read_index(repo
);
3405 istate
= repo
->index
;
3406 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3407 struct cache_entry
*ce
= istate
->cache
[i
];
3408 struct cache_entry
*new_ce
;
3414 len
= ce_namelen(ce
);
3415 new_ce
= make_empty_cache_entry(istate
, len
);
3416 memcpy(new_ce
->name
, ce
->name
, len
);
3417 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3418 new_ce
->ce_namelen
= len
;
3419 new_ce
->ce_mode
= ce
->ce_mode
;
3420 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3421 return error(_("%s: cannot drop to stage #0"),
3428 * Returns 1 if the path is an "other" path with respect to
3429 * the index; that is, the path is not mentioned in the index at all,
3430 * either as a file, a directory with some files in the index,
3431 * or as an unmerged entry.
3433 * We helpfully remove a trailing "/" from directories so that
3434 * the output of read_directory can be used as-is.
3436 int index_name_is_other(struct index_state
*istate
, const char *name
,
3440 if (namelen
&& name
[namelen
- 1] == '/')
3442 pos
= index_name_pos(istate
, name
, namelen
);
3444 return 0; /* exact match */
3446 if (pos
< istate
->cache_nr
) {
3447 struct cache_entry
*ce
= istate
->cache
[pos
];
3448 if (ce_namelen(ce
) == namelen
&&
3449 !memcmp(ce
->name
, name
, namelen
))
3450 return 0; /* Yup, this one exists unmerged */
3455 void *read_blob_data_from_index(struct index_state
*istate
,
3456 const char *path
, unsigned long *size
)
3460 enum object_type type
;
3464 pos
= index_name_pos(istate
, path
, len
);
3467 * We might be in the middle of a merge, in which
3468 * case we would read stage #2 (ours).
3472 (pos
< 0 && i
< istate
->cache_nr
&&
3473 !strcmp(istate
->cache
[i
]->name
, path
));
3475 if (ce_stage(istate
->cache
[i
]) == 2)
3480 data
= repo_read_object_file(the_repository
, &istate
->cache
[pos
]->oid
,
3482 if (!data
|| type
!= OBJ_BLOB
) {
3491 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3493 dst
->untracked
= src
->untracked
;
3494 src
->untracked
= NULL
;
3495 dst
->cache_tree
= src
->cache_tree
;
3496 src
->cache_tree
= NULL
;
3499 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3500 struct index_state
*istate
)
3502 unsigned int size
= ce_size(ce
);
3503 int mem_pool_allocated
;
3504 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3505 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3507 memcpy(new_entry
, ce
, size
);
3508 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3512 void discard_cache_entry(struct cache_entry
*ce
)
3514 if (ce
&& should_validate_cache_entries())
3515 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3517 if (ce
&& ce
->mem_pool_allocated
)
3523 int should_validate_cache_entries(void)
3525 static int validate_index_cache_entries
= -1;
3527 if (validate_index_cache_entries
< 0) {
3528 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3529 validate_index_cache_entries
= 1;
3531 validate_index_cache_entries
= 0;
3534 return validate_index_cache_entries
;
3537 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3538 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3540 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3543 * The end of index entries (EOIE) extension is guaranteed to be last
3544 * so that it can be found by scanning backwards from the EOF.
3551 const char *index
, *eoie
;
3553 size_t offset
, src_offset
;
3554 unsigned char hash
[GIT_MAX_RAWSZ
];
3557 /* ensure we have an index big enough to contain an EOIE extension */
3558 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3561 /* validate the extension signature */
3562 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3563 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3565 index
+= sizeof(uint32_t);
3567 /* validate the extension size */
3568 extsize
= get_be32(index
);
3569 if (extsize
!= EOIE_SIZE
)
3571 index
+= sizeof(uint32_t);
3574 * Validate the offset we're going to look for the first extension
3575 * signature is after the index header and before the eoie extension.
3577 offset
= get_be32(index
);
3578 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3580 if (mmap
+ offset
>= eoie
)
3582 index
+= sizeof(uint32_t);
3585 * The hash is computed over extension types and their sizes (but not
3586 * their contents). E.g. if we have "TREE" extension that is N-bytes
3587 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3588 * then the hash would be:
3590 * SHA-1("TREE" + <binary representation of N> +
3591 * "REUC" + <binary representation of M>)
3593 src_offset
= offset
;
3594 the_hash_algo
->init_fn(&c
);
3595 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3596 /* After an array of active_nr index entries,
3597 * there can be arbitrary number of extended
3598 * sections, each of which is prefixed with
3599 * extension name (4-byte) and section length
3600 * in 4-byte network byte order.
3603 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3604 extsize
= ntohl(extsize
);
3606 /* verify the extension size isn't so large it will wrap around */
3607 if (src_offset
+ 8 + extsize
< src_offset
)
3610 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3613 src_offset
+= extsize
;
3615 the_hash_algo
->final_fn(hash
, &c
);
3616 if (!hasheq(hash
, (const unsigned char *)index
))
3619 /* Validate that the extension offsets returned us back to the eoie extension. */
3620 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3626 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3629 unsigned char hash
[GIT_MAX_RAWSZ
];
3632 put_be32(&buffer
, offset
);
3633 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3636 the_hash_algo
->final_fn(hash
, eoie_context
);
3637 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3640 #define IEOT_VERSION (1)
3642 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3644 const char *index
= NULL
;
3645 uint32_t extsize
, ext_version
;
3646 struct index_entry_offset_table
*ieot
;
3649 /* find the IEOT extension */
3652 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3653 extsize
= get_be32(mmap
+ offset
+ 4);
3654 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3655 index
= mmap
+ offset
+ 4 + 4;
3664 /* validate the version is IEOT_VERSION */
3665 ext_version
= get_be32(index
);
3666 if (ext_version
!= IEOT_VERSION
) {
3667 error("invalid IEOT version %d", ext_version
);
3670 index
+= sizeof(uint32_t);
3672 /* extension size - version bytes / bytes per entry */
3673 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3675 error("invalid number of IEOT entries %d", nr
);
3678 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3679 + (nr
* sizeof(struct index_entry_offset
)));
3681 for (i
= 0; i
< nr
; i
++) {
3682 ieot
->entries
[i
].offset
= get_be32(index
);
3683 index
+= sizeof(uint32_t);
3684 ieot
->entries
[i
].nr
= get_be32(index
);
3685 index
+= sizeof(uint32_t);
3691 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3697 put_be32(&buffer
, IEOT_VERSION
);
3698 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3701 for (i
= 0; i
< ieot
->nr
; i
++) {
3704 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3705 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3708 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3709 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3713 void prefetch_cache_entries(const struct index_state
*istate
,
3714 must_prefetch_predicate must_prefetch
)
3717 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3719 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3720 struct cache_entry
*ce
= istate
->cache
[i
];
3722 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3724 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3726 OBJECT_INFO_FOR_PREFETCH
))
3728 oid_array_append(&to_fetch
, &ce
->oid
);
3730 promisor_remote_get_direct(the_repository
,
3731 to_fetch
.oid
, to_fetch
.nr
);
3732 oid_array_clear(&to_fetch
);
3735 static int read_one_entry_opt(struct index_state
*istate
,
3736 const struct object_id
*oid
,
3737 struct strbuf
*base
,
3738 const char *pathname
,
3739 unsigned mode
, int opt
)
3742 struct cache_entry
*ce
;
3745 return READ_TREE_RECURSIVE
;
3747 len
= strlen(pathname
);
3748 ce
= make_empty_cache_entry(istate
, base
->len
+ len
);
3750 ce
->ce_mode
= create_ce_mode(mode
);
3751 ce
->ce_flags
= create_ce_flags(1);
3752 ce
->ce_namelen
= base
->len
+ len
;
3753 memcpy(ce
->name
, base
->buf
, base
->len
);
3754 memcpy(ce
->name
+ base
->len
, pathname
, len
+1);
3755 oidcpy(&ce
->oid
, oid
);
3756 return add_index_entry(istate
, ce
, opt
);
3759 static int read_one_entry(const struct object_id
*oid
, struct strbuf
*base
,
3760 const char *pathname
, unsigned mode
,
3763 struct index_state
*istate
= context
;
3764 return read_one_entry_opt(istate
, oid
, base
, pathname
,
3766 ADD_CACHE_OK_TO_ADD
|ADD_CACHE_SKIP_DFCHECK
);
3770 * This is used when the caller knows there is no existing entries at
3771 * the stage that will conflict with the entry being added.
3773 static int read_one_entry_quick(const struct object_id
*oid
, struct strbuf
*base
,
3774 const char *pathname
, unsigned mode
,
3777 struct index_state
*istate
= context
;
3778 return read_one_entry_opt(istate
, oid
, base
, pathname
,
3779 mode
, ADD_CACHE_JUST_APPEND
);
3783 * Read the tree specified with --with-tree option
3784 * (typically, HEAD) into stage #1 and then
3785 * squash them down to stage #0. This is used for
3786 * --error-unmatch to list and check the path patterns
3787 * that were given from the command line. We are not
3788 * going to write this index out.
3790 void overlay_tree_on_index(struct index_state
*istate
,
3791 const char *tree_name
, const char *prefix
)
3794 struct object_id oid
;
3795 struct pathspec pathspec
;
3796 struct cache_entry
*last_stage0
= NULL
;
3798 read_tree_fn_t fn
= NULL
;
3801 if (repo_get_oid(the_repository
, tree_name
, &oid
))
3802 die("tree-ish %s not found.", tree_name
);
3803 tree
= parse_tree_indirect(&oid
);
3805 die("bad tree-ish %s", tree_name
);
3807 /* Hoist the unmerged entries up to stage #3 to make room */
3808 /* TODO: audit for interaction with sparse-index. */
3809 ensure_full_index(istate
);
3810 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3811 struct cache_entry
*ce
= istate
->cache
[i
];
3814 ce
->ce_flags
|= CE_STAGEMASK
;
3818 static const char *(matchbuf
[1]);
3820 parse_pathspec(&pathspec
, PATHSPEC_ALL_MAGIC
,
3821 PATHSPEC_PREFER_CWD
, prefix
, matchbuf
);
3823 memset(&pathspec
, 0, sizeof(pathspec
));
3826 * See if we have cache entry at the stage. If so,
3827 * do it the original slow way, otherwise, append and then
3830 for (i
= 0; !fn
&& i
< istate
->cache_nr
; i
++) {
3831 const struct cache_entry
*ce
= istate
->cache
[i
];
3832 if (ce_stage(ce
) == 1)
3833 fn
= read_one_entry
;
3837 fn
= read_one_entry_quick
;
3838 err
= read_tree(the_repository
, tree
, &pathspec
, fn
, istate
);
3839 clear_pathspec(&pathspec
);
3841 die("unable to read tree entries %s", tree_name
);
3844 * Sort the cache entry -- we need to nuke the cache tree, though.
3846 if (fn
== read_one_entry_quick
) {
3847 cache_tree_free(&istate
->cache_tree
);
3848 QSORT(istate
->cache
, istate
->cache_nr
, cmp_cache_name_compare
);
3851 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3852 struct cache_entry
*ce
= istate
->cache
[i
];
3853 switch (ce_stage(ce
)) {
3861 * If there is stage #0 entry for this, we do not
3862 * need to show it. We use CE_UPDATE bit to mark
3866 !strcmp(last_stage0
->name
, ce
->name
))
3867 ce
->ce_flags
|= CE_UPDATE
;
3872 struct update_callback_data
{
3873 struct index_state
*index
;
3879 static int fix_unmerged_status(struct diff_filepair
*p
,
3880 struct update_callback_data
*data
)
3882 if (p
->status
!= DIFF_STATUS_UNMERGED
)
3884 if (!(data
->flags
& ADD_CACHE_IGNORE_REMOVAL
) && !p
->two
->mode
)
3886 * This is not an explicit add request, and the
3887 * path is missing from the working tree (deleted)
3889 return DIFF_STATUS_DELETED
;
3892 * Either an explicit add request, or path exists
3893 * in the working tree. An attempt to explicitly
3894 * add a path that does not exist in the working tree
3895 * will be caught as an error by the caller immediately.
3897 return DIFF_STATUS_MODIFIED
;
3900 static void update_callback(struct diff_queue_struct
*q
,
3901 struct diff_options
*opt UNUSED
, void *cbdata
)
3904 struct update_callback_data
*data
= cbdata
;
3906 for (i
= 0; i
< q
->nr
; i
++) {
3907 struct diff_filepair
*p
= q
->queue
[i
];
3908 const char *path
= p
->one
->path
;
3910 if (!data
->include_sparse
&&
3911 !path_in_sparse_checkout(path
, data
->index
))
3914 switch (fix_unmerged_status(p
, data
)) {
3916 die(_("unexpected diff status %c"), p
->status
);
3917 case DIFF_STATUS_MODIFIED
:
3918 case DIFF_STATUS_TYPE_CHANGED
:
3919 if (add_file_to_index(data
->index
, path
, data
->flags
)) {
3920 if (!(data
->flags
& ADD_CACHE_IGNORE_ERRORS
))
3921 die(_("updating files failed"));
3925 case DIFF_STATUS_DELETED
:
3926 if (data
->flags
& ADD_CACHE_IGNORE_REMOVAL
)
3928 if (!(data
->flags
& ADD_CACHE_PRETEND
))
3929 remove_file_from_index(data
->index
, path
);
3930 if (data
->flags
& (ADD_CACHE_PRETEND
|ADD_CACHE_VERBOSE
))
3931 printf(_("remove '%s'\n"), path
);
3937 int add_files_to_cache(struct repository
*repo
, const char *prefix
,
3938 const struct pathspec
*pathspec
, int include_sparse
,
3941 struct update_callback_data data
;
3942 struct rev_info rev
;
3944 memset(&data
, 0, sizeof(data
));
3945 data
.index
= repo
->index
;
3946 data
.include_sparse
= include_sparse
;
3949 repo_init_revisions(repo
, &rev
, prefix
);
3950 setup_revisions(0, NULL
, &rev
, NULL
);
3952 copy_pathspec(&rev
.prune_data
, pathspec
);
3953 rev
.diffopt
.output_format
= DIFF_FORMAT_CALLBACK
;
3954 rev
.diffopt
.format_callback
= update_callback
;
3955 rev
.diffopt
.format_callback_data
= &data
;
3956 rev
.diffopt
.flags
.override_submodule_config
= 1;
3957 rev
.max_count
= 0; /* do not compare unmerged paths with stage #2 */
3960 * Use an ODB transaction to optimize adding multiple objects.
3961 * This function is invoked from commands other than 'add', which
3962 * may not have their own transaction active.
3964 begin_odb_transaction();
3965 run_diff_files(&rev
, DIFF_RACY_IS_MODIFIED
);
3966 end_odb_transaction();
3968 release_revisions(&rev
);
3969 return !!data
.add_errors
;