2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #include "git-compat-util.h"
7 #include "bulk-checkin.h"
15 #include "cache-tree.h"
18 #include "object-file.h"
19 #include "object-store-ll.h"
20 #include "oid-array.h"
23 #include "environment.h"
26 #include "name-hash.h"
27 #include "object-name.h"
29 #include "preload-index.h"
30 #include "read-cache.h"
31 #include "resolve-undo.h"
36 #include "split-index.h"
39 #include "fsmonitor.h"
40 #include "thread-utils.h"
42 #include "sparse-index.h"
43 #include "csum-file.h"
44 #include "promisor-remote.h"
47 /* Mask for the name length in ce_flags in the on-disk index */
49 #define CE_NAMEMASK (0x0fff)
53 * The first letter should be 'A'..'Z' for extensions that are not
54 * necessary for a correct operation (i.e. optimization data).
55 * When new extensions are added that _needs_ to be understood in
56 * order to correctly interpret the index file, pick character that
57 * is outside the range, to cause the reader to abort.
60 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
61 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
62 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
63 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
64 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
65 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
66 #define CACHE_EXT_ENDOFINDEXENTRIES 0x454F4945 /* "EOIE" */
67 #define CACHE_EXT_INDEXENTRYOFFSETTABLE 0x49454F54 /* "IEOT" */
68 #define CACHE_EXT_SPARSE_DIRECTORIES 0x73646972 /* "sdir" */
70 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
71 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
72 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
73 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
77 * This is an estimate of the pathname length in the index. We use
78 * this for V4 index files to guess the un-deltafied size of the index
79 * in memory because of pathname deltafication. This is not required
80 * for V2/V3 index formats because their pathnames are not compressed.
81 * If the initial amount of memory set aside is not sufficient, the
82 * mem pool will allocate extra memory.
84 #define CACHE_ENTRY_PATH_LENGTH 80
86 enum index_search_mode
{
91 static inline struct cache_entry
*mem_pool__ce_alloc(struct mem_pool
*mem_pool
, size_t len
)
93 struct cache_entry
*ce
;
94 ce
= mem_pool_alloc(mem_pool
, cache_entry_size(len
));
95 ce
->mem_pool_allocated
= 1;
99 static inline struct cache_entry
*mem_pool__ce_calloc(struct mem_pool
*mem_pool
, size_t len
)
101 struct cache_entry
* ce
;
102 ce
= mem_pool_calloc(mem_pool
, 1, cache_entry_size(len
));
103 ce
->mem_pool_allocated
= 1;
107 static struct mem_pool
*find_mem_pool(struct index_state
*istate
)
109 struct mem_pool
**pool_ptr
;
111 if (istate
->split_index
&& istate
->split_index
->base
)
112 pool_ptr
= &istate
->split_index
->base
->ce_mem_pool
;
114 pool_ptr
= &istate
->ce_mem_pool
;
117 *pool_ptr
= xmalloc(sizeof(**pool_ptr
));
118 mem_pool_init(*pool_ptr
, 0);
124 static const char *alternate_index_output
;
126 static void set_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
128 if (S_ISSPARSEDIR(ce
->ce_mode
))
129 istate
->sparse_index
= INDEX_COLLAPSED
;
131 istate
->cache
[nr
] = ce
;
132 add_name_hash(istate
, ce
);
135 static void replace_index_entry(struct index_state
*istate
, int nr
, struct cache_entry
*ce
)
137 struct cache_entry
*old
= istate
->cache
[nr
];
139 replace_index_entry_in_base(istate
, old
, ce
);
140 remove_name_hash(istate
, old
);
141 discard_cache_entry(old
);
142 ce
->ce_flags
&= ~CE_HASHED
;
143 set_index_entry(istate
, nr
, ce
);
144 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
145 mark_fsmonitor_invalid(istate
, ce
);
146 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
149 void rename_index_entry_at(struct index_state
*istate
, int nr
, const char *new_name
)
151 struct cache_entry
*old_entry
= istate
->cache
[nr
], *new_entry
, *refreshed
;
152 int namelen
= strlen(new_name
);
154 new_entry
= make_empty_cache_entry(istate
, namelen
);
155 copy_cache_entry(new_entry
, old_entry
);
156 new_entry
->ce_flags
&= ~CE_HASHED
;
157 new_entry
->ce_namelen
= namelen
;
158 new_entry
->index
= 0;
159 memcpy(new_entry
->name
, new_name
, namelen
+ 1);
161 cache_tree_invalidate_path(istate
, old_entry
->name
);
162 untracked_cache_remove_from_index(istate
, old_entry
->name
);
163 remove_index_entry_at(istate
, nr
);
166 * Refresh the new index entry. Using 'refresh_cache_entry' ensures
167 * we only update stat info if the entry is otherwise up-to-date (i.e.,
168 * the contents/mode haven't changed). This ensures that we reflect the
169 * 'ctime' of the rename in the index without (incorrectly) updating
170 * the cached stat info to reflect unstaged changes on disk.
172 refreshed
= refresh_cache_entry(istate
, new_entry
, CE_MATCH_REFRESH
);
173 if (refreshed
&& refreshed
!= new_entry
) {
174 add_index_entry(istate
, refreshed
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
175 discard_cache_entry(new_entry
);
177 add_index_entry(istate
, new_entry
, ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
);
181 * This only updates the "non-critical" parts of the directory
182 * cache, ie the parts that aren't tracked by GIT, and only used
183 * to validate the cache.
185 void fill_stat_cache_info(struct index_state
*istate
, struct cache_entry
*ce
, struct stat
*st
)
187 fill_stat_data(&ce
->ce_stat_data
, st
);
189 if (assume_unchanged
)
190 ce
->ce_flags
|= CE_VALID
;
192 if (S_ISREG(st
->st_mode
)) {
193 ce_mark_uptodate(ce
);
194 mark_fsmonitor_valid(istate
, ce
);
198 static int ce_compare_data(struct index_state
*istate
,
199 const struct cache_entry
*ce
,
203 int fd
= git_open_cloexec(ce
->name
, O_RDONLY
);
206 struct object_id oid
;
207 if (!index_fd(istate
, &oid
, fd
, st
, OBJ_BLOB
, ce
->name
, 0))
208 match
= !oideq(&oid
, &ce
->oid
);
209 /* index_fd() closed the file descriptor already */
214 static int ce_compare_link(const struct cache_entry
*ce
, size_t expected_size
)
219 enum object_type type
;
220 struct strbuf sb
= STRBUF_INIT
;
222 if (strbuf_readlink(&sb
, ce
->name
, expected_size
))
225 buffer
= repo_read_object_file(the_repository
, &ce
->oid
, &type
, &size
);
228 match
= memcmp(buffer
, sb
.buf
, size
);
235 static int ce_compare_gitlink(const struct cache_entry
*ce
)
237 struct object_id oid
;
240 * We don't actually require that the .git directory
241 * under GITLINK directory be a valid git directory. It
242 * might even be missing (in case nobody populated that
245 * If so, we consider it always to match.
247 if (resolve_gitlink_ref(ce
->name
, "HEAD", &oid
) < 0)
249 return !oideq(&oid
, &ce
->oid
);
252 static int ce_modified_check_fs(struct index_state
*istate
,
253 const struct cache_entry
*ce
,
256 switch (st
->st_mode
& S_IFMT
) {
258 if (ce_compare_data(istate
, ce
, st
))
262 if (ce_compare_link(ce
, xsize_t(st
->st_size
)))
266 if (S_ISGITLINK(ce
->ce_mode
))
267 return ce_compare_gitlink(ce
) ? DATA_CHANGED
: 0;
268 /* else fallthrough */
275 static int ce_match_stat_basic(const struct cache_entry
*ce
, struct stat
*st
)
277 unsigned int changed
= 0;
279 if (ce
->ce_flags
& CE_REMOVE
)
280 return MODE_CHANGED
| DATA_CHANGED
| TYPE_CHANGED
;
282 switch (ce
->ce_mode
& S_IFMT
) {
284 changed
|= !S_ISREG(st
->st_mode
) ? TYPE_CHANGED
: 0;
285 /* We consider only the owner x bit to be relevant for
288 if (trust_executable_bit
&&
289 (0100 & (ce
->ce_mode
^ st
->st_mode
)))
290 changed
|= MODE_CHANGED
;
293 if (!S_ISLNK(st
->st_mode
) &&
294 (has_symlinks
|| !S_ISREG(st
->st_mode
)))
295 changed
|= TYPE_CHANGED
;
298 /* We ignore most of the st_xxx fields for gitlinks */
299 if (!S_ISDIR(st
->st_mode
))
300 changed
|= TYPE_CHANGED
;
301 else if (ce_compare_gitlink(ce
))
302 changed
|= DATA_CHANGED
;
305 BUG("unsupported ce_mode: %o", ce
->ce_mode
);
308 changed
|= match_stat_data(&ce
->ce_stat_data
, st
);
310 /* Racily smudged entry? */
311 if (!ce
->ce_stat_data
.sd_size
) {
312 if (!is_empty_blob_sha1(ce
->oid
.hash
))
313 changed
|= DATA_CHANGED
;
319 static int is_racy_stat(const struct index_state
*istate
,
320 const struct stat_data
*sd
)
322 return (istate
->timestamp
.sec
&&
324 /* nanosecond timestamped files can also be racy! */
325 (istate
->timestamp
.sec
< sd
->sd_mtime
.sec
||
326 (istate
->timestamp
.sec
== sd
->sd_mtime
.sec
&&
327 istate
->timestamp
.nsec
<= sd
->sd_mtime
.nsec
))
329 istate
->timestamp
.sec
<= sd
->sd_mtime
.sec
334 int is_racy_timestamp(const struct index_state
*istate
,
335 const struct cache_entry
*ce
)
337 return (!S_ISGITLINK(ce
->ce_mode
) &&
338 is_racy_stat(istate
, &ce
->ce_stat_data
));
341 int match_stat_data_racy(const struct index_state
*istate
,
342 const struct stat_data
*sd
, struct stat
*st
)
344 if (is_racy_stat(istate
, sd
))
345 return MTIME_CHANGED
;
346 return match_stat_data(sd
, st
);
349 int ie_match_stat(struct index_state
*istate
,
350 const struct cache_entry
*ce
, struct stat
*st
,
351 unsigned int options
)
353 unsigned int changed
;
354 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
355 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
356 int assume_racy_is_modified
= options
& CE_MATCH_RACY_IS_DIRTY
;
357 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
359 if (!ignore_fsmonitor
)
360 refresh_fsmonitor(istate
);
362 * If it's marked as always valid in the index, it's
363 * valid whatever the checked-out copy says.
365 * skip-worktree has the same effect with higher precedence
367 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
))
369 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
))
371 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
))
375 * Intent-to-add entries have not been added, so the index entry
376 * by definition never matches what is in the work tree until it
377 * actually gets added.
379 if (ce_intent_to_add(ce
))
380 return DATA_CHANGED
| TYPE_CHANGED
| MODE_CHANGED
;
382 changed
= ce_match_stat_basic(ce
, st
);
385 * Within 1 second of this sequence:
386 * echo xyzzy >file && git-update-index --add file
387 * running this command:
389 * would give a falsely clean cache entry. The mtime and
390 * length match the cache, and other stat fields do not change.
392 * We could detect this at update-index time (the cache entry
393 * being registered/updated records the same time as "now")
394 * and delay the return from git-update-index, but that would
395 * effectively mean we can make at most one commit per second,
396 * which is not acceptable. Instead, we check cache entries
397 * whose mtime are the same as the index file timestamp more
398 * carefully than others.
400 if (!changed
&& is_racy_timestamp(istate
, ce
)) {
401 if (assume_racy_is_modified
)
402 changed
|= DATA_CHANGED
;
404 changed
|= ce_modified_check_fs(istate
, ce
, st
);
410 int ie_modified(struct index_state
*istate
,
411 const struct cache_entry
*ce
,
412 struct stat
*st
, unsigned int options
)
414 int changed
, changed_fs
;
416 changed
= ie_match_stat(istate
, ce
, st
, options
);
420 * If the mode or type has changed, there's no point in trying
421 * to refresh the entry - it's not going to match
423 if (changed
& (MODE_CHANGED
| TYPE_CHANGED
))
427 * Immediately after read-tree or update-index --cacheinfo,
428 * the length field is zero, as we have never even read the
429 * lstat(2) information once, and we cannot trust DATA_CHANGED
430 * returned by ie_match_stat() which in turn was returned by
431 * ce_match_stat_basic() to signal that the filesize of the
432 * blob changed. We have to actually go to the filesystem to
433 * see if the contents match, and if so, should answer "unchanged".
435 * The logic does not apply to gitlinks, as ce_match_stat_basic()
436 * already has checked the actual HEAD from the filesystem in the
437 * subproject. If ie_match_stat() already said it is different,
438 * then we know it is.
440 if ((changed
& DATA_CHANGED
) &&
441 (S_ISGITLINK(ce
->ce_mode
) || ce
->ce_stat_data
.sd_size
!= 0))
444 changed_fs
= ce_modified_check_fs(istate
, ce
, st
);
446 return changed
| changed_fs
;
450 static int cache_name_stage_compare(const char *name1
, int len1
, int stage1
,
451 const char *name2
, int len2
, int stage2
)
455 cmp
= name_compare(name1
, len1
, name2
, len2
);
466 int cmp_cache_name_compare(const void *a_
, const void *b_
)
468 const struct cache_entry
*ce1
, *ce2
;
470 ce1
= *((const struct cache_entry
**)a_
);
471 ce2
= *((const struct cache_entry
**)b_
);
472 return cache_name_stage_compare(ce1
->name
, ce1
->ce_namelen
, ce_stage(ce1
),
473 ce2
->name
, ce2
->ce_namelen
, ce_stage(ce2
));
476 static int index_name_stage_pos(struct index_state
*istate
,
477 const char *name
, int namelen
,
479 enum index_search_mode search_mode
)
484 last
= istate
->cache_nr
;
485 while (last
> first
) {
486 int next
= first
+ ((last
- first
) >> 1);
487 struct cache_entry
*ce
= istate
->cache
[next
];
488 int cmp
= cache_name_stage_compare(name
, namelen
, stage
, ce
->name
, ce_namelen(ce
), ce_stage(ce
));
498 if (search_mode
== EXPAND_SPARSE
&& istate
->sparse_index
&&
500 /* Note: first <= istate->cache_nr */
501 struct cache_entry
*ce
= istate
->cache
[first
- 1];
504 * If we are in a sparse-index _and_ the entry before the
505 * insertion position is a sparse-directory entry that is
506 * an ancestor of 'name', then we need to expand the index
507 * and search again. This will only trigger once, because
508 * thereafter the index is fully expanded.
510 if (S_ISSPARSEDIR(ce
->ce_mode
) &&
511 ce_namelen(ce
) < namelen
&&
512 !strncmp(name
, ce
->name
, ce_namelen(ce
))) {
513 ensure_full_index(istate
);
514 return index_name_stage_pos(istate
, name
, namelen
, stage
, search_mode
);
521 int index_name_pos(struct index_state
*istate
, const char *name
, int namelen
)
523 return index_name_stage_pos(istate
, name
, namelen
, 0, EXPAND_SPARSE
);
526 int index_name_pos_sparse(struct index_state
*istate
, const char *name
, int namelen
)
528 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
);
531 int index_entry_exists(struct index_state
*istate
, const char *name
, int namelen
)
533 return index_name_stage_pos(istate
, name
, namelen
, 0, NO_EXPAND_SPARSE
) >= 0;
536 int remove_index_entry_at(struct index_state
*istate
, int pos
)
538 struct cache_entry
*ce
= istate
->cache
[pos
];
540 record_resolve_undo(istate
, ce
);
541 remove_name_hash(istate
, ce
);
542 save_or_free_index_entry(istate
, ce
);
543 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
545 if (pos
>= istate
->cache_nr
)
547 MOVE_ARRAY(istate
->cache
+ pos
, istate
->cache
+ pos
+ 1,
548 istate
->cache_nr
- pos
);
553 * Remove all cache entries marked for removal, that is where
554 * CE_REMOVE is set in ce_flags. This is much more effective than
555 * calling remove_index_entry_at() for each entry to be removed.
557 void remove_marked_cache_entries(struct index_state
*istate
, int invalidate
)
559 struct cache_entry
**ce_array
= istate
->cache
;
562 for (i
= j
= 0; i
< istate
->cache_nr
; i
++) {
563 if (ce_array
[i
]->ce_flags
& CE_REMOVE
) {
565 cache_tree_invalidate_path(istate
,
567 untracked_cache_remove_from_index(istate
,
570 remove_name_hash(istate
, ce_array
[i
]);
571 save_or_free_index_entry(istate
, ce_array
[i
]);
574 ce_array
[j
++] = ce_array
[i
];
576 if (j
== istate
->cache_nr
)
578 istate
->cache_changed
|= CE_ENTRY_REMOVED
;
579 istate
->cache_nr
= j
;
582 int remove_file_from_index(struct index_state
*istate
, const char *path
)
584 int pos
= index_name_pos(istate
, path
, strlen(path
));
587 cache_tree_invalidate_path(istate
, path
);
588 untracked_cache_remove_from_index(istate
, path
);
589 while (pos
< istate
->cache_nr
&& !strcmp(istate
->cache
[pos
]->name
, path
))
590 remove_index_entry_at(istate
, pos
);
594 static int compare_name(struct cache_entry
*ce
, const char *path
, int namelen
)
596 return namelen
!= ce_namelen(ce
) || memcmp(path
, ce
->name
, namelen
);
599 static int index_name_pos_also_unmerged(struct index_state
*istate
,
600 const char *path
, int namelen
)
602 int pos
= index_name_pos(istate
, path
, namelen
);
603 struct cache_entry
*ce
;
608 /* maybe unmerged? */
610 if (pos
>= istate
->cache_nr
||
611 compare_name((ce
= istate
->cache
[pos
]), path
, namelen
))
614 /* order of preference: stage 2, 1, 3 */
615 if (ce_stage(ce
) == 1 && pos
+ 1 < istate
->cache_nr
&&
616 ce_stage((ce
= istate
->cache
[pos
+ 1])) == 2 &&
617 !compare_name(ce
, path
, namelen
))
622 static int different_name(struct cache_entry
*ce
, struct cache_entry
*alias
)
624 int len
= ce_namelen(ce
);
625 return ce_namelen(alias
) != len
|| memcmp(ce
->name
, alias
->name
, len
);
629 * If we add a filename that aliases in the cache, we will use the
630 * name that we already have - but we don't want to update the same
631 * alias twice, because that implies that there were actually two
632 * different files with aliasing names!
634 * So we use the CE_ADDED flag to verify that the alias was an old
635 * one before we accept it as
637 static struct cache_entry
*create_alias_ce(struct index_state
*istate
,
638 struct cache_entry
*ce
,
639 struct cache_entry
*alias
)
642 struct cache_entry
*new_entry
;
644 if (alias
->ce_flags
& CE_ADDED
)
645 die(_("will not add file alias '%s' ('%s' already exists in index)"),
646 ce
->name
, alias
->name
);
648 /* Ok, create the new entry using the name of the existing alias */
649 len
= ce_namelen(alias
);
650 new_entry
= make_empty_cache_entry(istate
, len
);
651 memcpy(new_entry
->name
, alias
->name
, len
);
652 copy_cache_entry(new_entry
, ce
);
653 save_or_free_index_entry(istate
, ce
);
657 void set_object_name_for_intent_to_add_entry(struct cache_entry
*ce
)
659 struct object_id oid
;
660 if (write_object_file("", 0, OBJ_BLOB
, &oid
))
661 die(_("cannot create an empty blob in the object database"));
662 oidcpy(&ce
->oid
, &oid
);
665 int add_to_index(struct index_state
*istate
, const char *path
, struct stat
*st
, int flags
)
667 int namelen
, was_same
;
668 mode_t st_mode
= st
->st_mode
;
669 struct cache_entry
*ce
, *alias
= NULL
;
670 unsigned ce_option
= CE_MATCH_IGNORE_VALID
|CE_MATCH_IGNORE_SKIP_WORKTREE
|CE_MATCH_RACY_IS_DIRTY
;
671 int verbose
= flags
& (ADD_CACHE_VERBOSE
| ADD_CACHE_PRETEND
);
672 int pretend
= flags
& ADD_CACHE_PRETEND
;
673 int intent_only
= flags
& ADD_CACHE_INTENT
;
674 int add_option
= (ADD_CACHE_OK_TO_ADD
|ADD_CACHE_OK_TO_REPLACE
|
675 (intent_only
? ADD_CACHE_NEW_ONLY
: 0));
676 unsigned hash_flags
= pretend
? 0 : HASH_WRITE_OBJECT
;
677 struct object_id oid
;
679 if (flags
& ADD_CACHE_RENORMALIZE
)
680 hash_flags
|= HASH_RENORMALIZE
;
682 if (!S_ISREG(st_mode
) && !S_ISLNK(st_mode
) && !S_ISDIR(st_mode
))
683 return error(_("%s: can only add regular files, symbolic links or git-directories"), path
);
685 namelen
= strlen(path
);
686 if (S_ISDIR(st_mode
)) {
687 if (resolve_gitlink_ref(path
, "HEAD", &oid
) < 0)
688 return error(_("'%s' does not have a commit checked out"), path
);
689 while (namelen
&& path
[namelen
-1] == '/')
692 ce
= make_empty_cache_entry(istate
, namelen
);
693 memcpy(ce
->name
, path
, namelen
);
694 ce
->ce_namelen
= namelen
;
696 fill_stat_cache_info(istate
, ce
, st
);
698 ce
->ce_flags
|= CE_INTENT_TO_ADD
;
701 if (trust_executable_bit
&& has_symlinks
) {
702 ce
->ce_mode
= create_ce_mode(st_mode
);
704 /* If there is an existing entry, pick the mode bits and type
705 * from it, otherwise assume unexecutable regular file.
707 struct cache_entry
*ent
;
708 int pos
= index_name_pos_also_unmerged(istate
, path
, namelen
);
710 ent
= (0 <= pos
) ? istate
->cache
[pos
] : NULL
;
711 ce
->ce_mode
= ce_mode_from_stat(ent
, st_mode
);
714 /* When core.ignorecase=true, determine if a directory of the same name but differing
715 * case already exists within the Git repository. If it does, ensure the directory
716 * case of the file being added to the repository matches (is folded into) the existing
717 * entry's directory case.
720 adjust_dirname_case(istate
, ce
->name
);
722 if (!(flags
& ADD_CACHE_RENORMALIZE
)) {
723 alias
= index_file_exists(istate
, ce
->name
,
724 ce_namelen(ce
), ignore_case
);
727 !ie_match_stat(istate
, alias
, st
, ce_option
)) {
728 /* Nothing changed, really */
729 if (!S_ISGITLINK(alias
->ce_mode
))
730 ce_mark_uptodate(alias
);
731 alias
->ce_flags
|= CE_ADDED
;
733 discard_cache_entry(ce
);
738 if (index_path(istate
, &ce
->oid
, path
, st
, hash_flags
)) {
739 discard_cache_entry(ce
);
740 return error(_("unable to index file '%s'"), path
);
743 set_object_name_for_intent_to_add_entry(ce
);
745 if (ignore_case
&& alias
&& different_name(ce
, alias
))
746 ce
= create_alias_ce(istate
, ce
, alias
);
747 ce
->ce_flags
|= CE_ADDED
;
749 /* It was suspected to be racily clean, but it turns out to be Ok */
752 oideq(&alias
->oid
, &ce
->oid
) &&
753 ce
->ce_mode
== alias
->ce_mode
);
756 discard_cache_entry(ce
);
757 else if (add_index_entry(istate
, ce
, add_option
)) {
758 discard_cache_entry(ce
);
759 return error(_("unable to add '%s' to index"), path
);
761 if (verbose
&& !was_same
)
762 printf("add '%s'\n", path
);
766 int add_file_to_index(struct index_state
*istate
, const char *path
, int flags
)
769 if (lstat(path
, &st
))
770 die_errno(_("unable to stat '%s'"), path
);
771 return add_to_index(istate
, path
, &st
, flags
);
774 struct cache_entry
*make_empty_cache_entry(struct index_state
*istate
, size_t len
)
776 return mem_pool__ce_calloc(find_mem_pool(istate
), len
);
779 struct cache_entry
*make_empty_transient_cache_entry(size_t len
,
780 struct mem_pool
*ce_mem_pool
)
783 return mem_pool__ce_calloc(ce_mem_pool
, len
);
784 return xcalloc(1, cache_entry_size(len
));
787 enum verify_path_result
{
793 static enum verify_path_result
verify_path_internal(const char *, unsigned);
795 int verify_path(const char *path
, unsigned mode
)
797 return verify_path_internal(path
, mode
) == PATH_OK
;
800 struct cache_entry
*make_cache_entry(struct index_state
*istate
,
802 const struct object_id
*oid
,
805 unsigned int refresh_options
)
807 struct cache_entry
*ce
, *ret
;
810 if (verify_path_internal(path
, mode
) == PATH_INVALID
) {
811 error(_("invalid path '%s'"), path
);
816 ce
= make_empty_cache_entry(istate
, len
);
818 oidcpy(&ce
->oid
, oid
);
819 memcpy(ce
->name
, path
, len
);
820 ce
->ce_flags
= create_ce_flags(stage
);
821 ce
->ce_namelen
= len
;
822 ce
->ce_mode
= create_ce_mode(mode
);
824 ret
= refresh_cache_entry(istate
, ce
, refresh_options
);
826 discard_cache_entry(ce
);
830 struct cache_entry
*make_transient_cache_entry(unsigned int mode
,
831 const struct object_id
*oid
,
834 struct mem_pool
*ce_mem_pool
)
836 struct cache_entry
*ce
;
839 if (!verify_path(path
, mode
)) {
840 error(_("invalid path '%s'"), path
);
845 ce
= make_empty_transient_cache_entry(len
, ce_mem_pool
);
847 oidcpy(&ce
->oid
, oid
);
848 memcpy(ce
->name
, path
, len
);
849 ce
->ce_flags
= create_ce_flags(stage
);
850 ce
->ce_namelen
= len
;
851 ce
->ce_mode
= create_ce_mode(mode
);
857 * Chmod an index entry with either +x or -x.
859 * Returns -1 if the chmod for the particular cache entry failed (if it's
860 * not a regular file), -2 if an invalid flip argument is passed in, 0
863 int chmod_index_entry(struct index_state
*istate
, struct cache_entry
*ce
,
866 if (!S_ISREG(ce
->ce_mode
))
873 ce
->ce_mode
&= ~0111;
878 cache_tree_invalidate_path(istate
, ce
->name
);
879 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
880 mark_fsmonitor_invalid(istate
, ce
);
881 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
886 int ce_same_name(const struct cache_entry
*a
, const struct cache_entry
*b
)
888 int len
= ce_namelen(a
);
889 return ce_namelen(b
) == len
&& !memcmp(a
->name
, b
->name
, len
);
893 * We fundamentally don't like some paths: we don't want
894 * dot or dot-dot anywhere, and for obvious reasons don't
895 * want to recurse into ".git" either.
897 * Also, we don't want double slashes or slashes at the
898 * end that can make pathnames ambiguous.
900 static int verify_dotfile(const char *rest
, unsigned mode
)
903 * The first character was '.', but that
904 * has already been discarded, we now test
908 /* "." is not allowed */
909 if (*rest
== '\0' || is_dir_sep(*rest
))
914 * ".git" followed by NUL or slash is bad. Note that we match
915 * case-insensitively here, even if ignore_case is not set.
916 * This outlaws ".GIT" everywhere out of an abundance of caution,
917 * since there's really no good reason to allow it.
919 * Once we've seen ".git", we can also find ".gitmodules", etc (also
920 * case-insensitively).
924 if (rest
[1] != 'i' && rest
[1] != 'I')
926 if (rest
[2] != 't' && rest
[2] != 'T')
928 if (rest
[3] == '\0' || is_dir_sep(rest
[3]))
932 if (skip_iprefix(rest
, "modules", &rest
) &&
933 (*rest
== '\0' || is_dir_sep(*rest
)))
938 if (rest
[1] == '\0' || is_dir_sep(rest
[1]))
944 static enum verify_path_result
verify_path_internal(const char *path
,
949 if (has_dos_drive_prefix(path
))
952 if (!is_valid_path(path
))
963 if (is_hfs_dotgit(path
))
966 if (is_hfs_dotgitmodules(path
))
971 #if defined GIT_WINDOWS_NATIVE || defined __CYGWIN__
975 if (is_ntfs_dotgit(path
))
978 if (is_ntfs_dotgitmodules(path
))
984 if ((c
== '.' && !verify_dotfile(path
, mode
)) ||
988 * allow terminating directory separators for
989 * sparse directory entries.
992 return S_ISDIR(mode
) ? PATH_DIR_WITH_SEP
:
994 } else if (c
== '\\' && protect_ntfs
) {
995 if (is_ntfs_dotgit(path
))
998 if (is_ntfs_dotgitmodules(path
))
1008 * Do we have another file that has the beginning components being a
1009 * proper superset of the name we're trying to add?
1011 static int has_file_name(struct index_state
*istate
,
1012 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1015 int len
= ce_namelen(ce
);
1016 int stage
= ce_stage(ce
);
1017 const char *name
= ce
->name
;
1019 while (pos
< istate
->cache_nr
) {
1020 struct cache_entry
*p
= istate
->cache
[pos
++];
1022 if (len
>= ce_namelen(p
))
1024 if (memcmp(name
, p
->name
, len
))
1026 if (ce_stage(p
) != stage
)
1028 if (p
->name
[len
] != '/')
1030 if (p
->ce_flags
& CE_REMOVE
)
1035 remove_index_entry_at(istate
, --pos
);
1042 * Like strcmp(), but also return the offset of the first change.
1043 * If strings are equal, return the length.
1045 int strcmp_offset(const char *s1
, const char *s2
, size_t *first_change
)
1050 return strcmp(s1
, s2
);
1052 for (k
= 0; s1
[k
] == s2
[k
]; k
++)
1057 return (unsigned char)s1
[k
] - (unsigned char)s2
[k
];
1061 * Do we have another file with a pathname that is a proper
1062 * subset of the name we're trying to add?
1064 * That is, is there another file in the index with a path
1065 * that matches a sub-directory in the given entry?
1067 static int has_dir_name(struct index_state
*istate
,
1068 const struct cache_entry
*ce
, int pos
, int ok_to_replace
)
1071 int stage
= ce_stage(ce
);
1072 const char *name
= ce
->name
;
1073 const char *slash
= name
+ ce_namelen(ce
);
1078 * We are frequently called during an iteration on a sorted
1079 * list of pathnames and while building a new index. Therefore,
1080 * there is a high probability that this entry will eventually
1081 * be appended to the index, rather than inserted in the middle.
1082 * If we can confirm that, we can avoid binary searches on the
1083 * components of the pathname.
1085 * Compare the entry's full path with the last path in the index.
1087 if (istate
->cache_nr
> 0) {
1088 cmp_last
= strcmp_offset(name
,
1089 istate
->cache
[istate
->cache_nr
- 1]->name
,
1092 if (len_eq_last
== 0) {
1094 * The entry sorts AFTER the last one in the
1095 * index and their paths have no common prefix,
1096 * so there cannot be a F/D conflict.
1101 * The entry sorts AFTER the last one in the
1102 * index, but has a common prefix. Fall through
1103 * to the loop below to disect the entry's path
1104 * and see where the difference is.
1107 } else if (cmp_last
== 0) {
1109 * The entry exactly matches the last one in the
1110 * index, but because of multiple stage and CE_REMOVE
1111 * items, we fall through and let the regular search
1121 if (*--slash
== '/')
1123 if (slash
<= ce
->name
)
1130 * (len + 1) is a directory boundary (including
1131 * the trailing slash). And since the loop is
1132 * decrementing "slash", the first iteration is
1133 * the longest directory prefix; subsequent
1134 * iterations consider parent directories.
1137 if (len
+ 1 <= len_eq_last
) {
1139 * The directory prefix (including the trailing
1140 * slash) also appears as a prefix in the last
1141 * entry, so the remainder cannot collide (because
1142 * strcmp said the whole path was greater).
1147 * LT: last: xxx/file_A
1153 if (len
> len_eq_last
) {
1155 * This part of the directory prefix (excluding
1156 * the trailing slash) is longer than the known
1157 * equal portions, so this sub-directory cannot
1158 * collide with a file.
1167 * This is a possible collision. Fall through and
1168 * let the regular search code handle it.
1175 pos
= index_name_stage_pos(istate
, name
, len
, stage
, EXPAND_SPARSE
);
1178 * Found one, but not so fast. This could
1179 * be a marker that says "I was here, but
1180 * I am being removed". Such an entry is
1181 * not a part of the resulting tree, and
1182 * it is Ok to have a directory at the same
1185 if (!(istate
->cache
[pos
]->ce_flags
& CE_REMOVE
)) {
1189 remove_index_entry_at(istate
, pos
);
1197 * Trivial optimization: if we find an entry that
1198 * already matches the sub-directory, then we know
1199 * we're ok, and we can exit.
1201 while (pos
< istate
->cache_nr
) {
1202 struct cache_entry
*p
= istate
->cache
[pos
];
1203 if ((ce_namelen(p
) <= len
) ||
1204 (p
->name
[len
] != '/') ||
1205 memcmp(p
->name
, name
, len
))
1206 break; /* not our subdirectory */
1207 if (ce_stage(p
) == stage
&& !(p
->ce_flags
& CE_REMOVE
))
1209 * p is at the same stage as our entry, and
1210 * is a subdirectory of what we are looking
1211 * at, so we cannot have conflicts at our
1212 * level or anything shorter.
1221 /* We may be in a situation where we already have path/file and path
1222 * is being added, or we already have path and path/file is being
1223 * added. Either one would result in a nonsense tree that has path
1224 * twice when git-write-tree tries to write it out. Prevent it.
1226 * If ok-to-replace is specified, we remove the conflicting entries
1227 * from the cache so the caller should recompute the insert position.
1228 * When this happens, we return non-zero.
1230 static int check_file_directory_conflict(struct index_state
*istate
,
1231 const struct cache_entry
*ce
,
1232 int pos
, int ok_to_replace
)
1237 * When ce is an "I am going away" entry, we allow it to be added
1239 if (ce
->ce_flags
& CE_REMOVE
)
1243 * We check if the path is a sub-path of a subsequent pathname
1244 * first, since removing those will not change the position
1247 retval
= has_file_name(istate
, ce
, pos
, ok_to_replace
);
1250 * Then check if the path might have a clashing sub-directory
1253 return retval
+ has_dir_name(istate
, ce
, pos
, ok_to_replace
);
1256 static int add_index_entry_with_check(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1259 int ok_to_add
= option
& ADD_CACHE_OK_TO_ADD
;
1260 int ok_to_replace
= option
& ADD_CACHE_OK_TO_REPLACE
;
1261 int skip_df_check
= option
& ADD_CACHE_SKIP_DFCHECK
;
1262 int new_only
= option
& ADD_CACHE_NEW_ONLY
;
1265 * If this entry's path sorts after the last entry in the index,
1266 * we can avoid searching for it.
1268 if (istate
->cache_nr
> 0 &&
1269 strcmp(ce
->name
, istate
->cache
[istate
->cache_nr
- 1]->name
) > 0)
1270 pos
= index_pos_to_insert_pos(istate
->cache_nr
);
1272 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1275 * Cache tree path should be invalidated only after index_name_stage_pos,
1276 * in case it expands a sparse index.
1278 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1279 cache_tree_invalidate_path(istate
, ce
->name
);
1281 /* existing match? Just replace it. */
1284 replace_index_entry(istate
, pos
, ce
);
1289 if (!(option
& ADD_CACHE_KEEP_CACHE_TREE
))
1290 untracked_cache_add_to_index(istate
, ce
->name
);
1293 * Inserting a merged entry ("stage 0") into the index
1294 * will always replace all non-merged entries..
1296 if (pos
< istate
->cache_nr
&& ce_stage(ce
) == 0) {
1297 while (ce_same_name(istate
->cache
[pos
], ce
)) {
1299 if (!remove_index_entry_at(istate
, pos
))
1306 if (verify_path_internal(ce
->name
, ce
->ce_mode
) == PATH_INVALID
)
1307 return error(_("invalid path '%s'"), ce
->name
);
1309 if (!skip_df_check
&&
1310 check_file_directory_conflict(istate
, ce
, pos
, ok_to_replace
)) {
1312 return error(_("'%s' appears as both a file and as a directory"),
1314 pos
= index_name_stage_pos(istate
, ce
->name
, ce_namelen(ce
), ce_stage(ce
), EXPAND_SPARSE
);
1320 int add_index_entry(struct index_state
*istate
, struct cache_entry
*ce
, int option
)
1324 if (option
& ADD_CACHE_JUST_APPEND
)
1325 pos
= istate
->cache_nr
;
1328 ret
= add_index_entry_with_check(istate
, ce
, option
);
1334 /* Make sure the array is big enough .. */
1335 ALLOC_GROW(istate
->cache
, istate
->cache_nr
+ 1, istate
->cache_alloc
);
1339 if (istate
->cache_nr
> pos
+ 1)
1340 MOVE_ARRAY(istate
->cache
+ pos
+ 1, istate
->cache
+ pos
,
1341 istate
->cache_nr
- pos
- 1);
1342 set_index_entry(istate
, pos
, ce
);
1343 istate
->cache_changed
|= CE_ENTRY_ADDED
;
1348 * "refresh" does not calculate a new sha1 file or bring the
1349 * cache up-to-date for mode/content changes. But what it
1350 * _does_ do is to "re-match" the stat information of a file
1351 * with the cache, so that you can refresh the cache for a
1352 * file that hasn't been changed but where the stat entry is
1355 * For example, you'd want to do this after doing a "git-read-tree",
1356 * to link up the stat cache details with the proper files.
1358 static struct cache_entry
*refresh_cache_ent(struct index_state
*istate
,
1359 struct cache_entry
*ce
,
1360 unsigned int options
, int *err
,
1366 struct cache_entry
*updated
;
1368 int refresh
= options
& CE_MATCH_REFRESH
;
1369 int ignore_valid
= options
& CE_MATCH_IGNORE_VALID
;
1370 int ignore_skip_worktree
= options
& CE_MATCH_IGNORE_SKIP_WORKTREE
;
1371 int ignore_missing
= options
& CE_MATCH_IGNORE_MISSING
;
1372 int ignore_fsmonitor
= options
& CE_MATCH_IGNORE_FSMONITOR
;
1374 if (!refresh
|| ce_uptodate(ce
))
1377 if (!ignore_fsmonitor
)
1378 refresh_fsmonitor(istate
);
1380 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1381 * that the change to the work tree does not matter and told
1384 if (!ignore_skip_worktree
&& ce_skip_worktree(ce
)) {
1385 ce_mark_uptodate(ce
);
1388 if (!ignore_valid
&& (ce
->ce_flags
& CE_VALID
)) {
1389 ce_mark_uptodate(ce
);
1392 if (!ignore_fsmonitor
&& (ce
->ce_flags
& CE_FSMONITOR_VALID
)) {
1393 ce_mark_uptodate(ce
);
1397 if (has_symlink_leading_path(ce
->name
, ce_namelen(ce
))) {
1407 if (lstat(ce
->name
, &st
) < 0) {
1408 if (ignore_missing
&& errno
== ENOENT
)
1415 changed
= ie_match_stat(istate
, ce
, &st
, options
);
1417 *changed_ret
= changed
;
1420 * The path is unchanged. If we were told to ignore
1421 * valid bit, then we did the actual stat check and
1422 * found that the entry is unmodified. If the entry
1423 * is not marked VALID, this is the place to mark it
1424 * valid again, under "assume unchanged" mode.
1426 if (ignore_valid
&& assume_unchanged
&&
1427 !(ce
->ce_flags
& CE_VALID
))
1428 ; /* mark this one VALID again */
1431 * We do not mark the index itself "modified"
1432 * because CE_UPTODATE flag is in-core only;
1433 * we are not going to write this change out.
1435 if (!S_ISGITLINK(ce
->ce_mode
)) {
1436 ce_mark_uptodate(ce
);
1437 mark_fsmonitor_valid(istate
, ce
);
1445 if (ie_modified(istate
, ce
, &st
, options
)) {
1451 updated
= make_empty_cache_entry(istate
, ce_namelen(ce
));
1452 copy_cache_entry(updated
, ce
);
1453 memcpy(updated
->name
, ce
->name
, ce
->ce_namelen
+ 1);
1454 fill_stat_cache_info(istate
, updated
, &st
);
1456 * If ignore_valid is not set, we should leave CE_VALID bit
1457 * alone. Otherwise, paths marked with --no-assume-unchanged
1458 * (i.e. things to be edited) will reacquire CE_VALID bit
1459 * automatically, which is not really what we want.
1461 if (!ignore_valid
&& assume_unchanged
&&
1462 !(ce
->ce_flags
& CE_VALID
))
1463 updated
->ce_flags
&= ~CE_VALID
;
1465 /* istate->cache_changed is updated in the caller */
1469 static void show_file(const char * fmt
, const char * name
, int in_porcelain
,
1470 int * first
, const char *header_msg
)
1472 if (in_porcelain
&& *first
&& header_msg
) {
1473 printf("%s\n", header_msg
);
1479 int repo_refresh_and_write_index(struct repository
*repo
,
1480 unsigned int refresh_flags
,
1481 unsigned int write_flags
,
1483 const struct pathspec
*pathspec
,
1484 char *seen
, const char *header_msg
)
1486 struct lock_file lock_file
= LOCK_INIT
;
1489 fd
= repo_hold_locked_index(repo
, &lock_file
, 0);
1490 if (!gentle
&& fd
< 0)
1492 if (refresh_index(repo
->index
, refresh_flags
, pathspec
, seen
, header_msg
))
1494 if (0 <= fd
&& write_locked_index(repo
->index
, &lock_file
, COMMIT_LOCK
| write_flags
))
1500 int refresh_index(struct index_state
*istate
, unsigned int flags
,
1501 const struct pathspec
*pathspec
,
1502 char *seen
, const char *header_msg
)
1506 int really
= (flags
& REFRESH_REALLY
) != 0;
1507 int allow_unmerged
= (flags
& REFRESH_UNMERGED
) != 0;
1508 int quiet
= (flags
& REFRESH_QUIET
) != 0;
1509 int not_new
= (flags
& REFRESH_IGNORE_MISSING
) != 0;
1510 int ignore_submodules
= (flags
& REFRESH_IGNORE_SUBMODULES
) != 0;
1511 int ignore_skip_worktree
= (flags
& REFRESH_IGNORE_SKIP_WORKTREE
) != 0;
1513 int in_porcelain
= (flags
& REFRESH_IN_PORCELAIN
);
1514 unsigned int options
= (CE_MATCH_REFRESH
|
1515 (really
? CE_MATCH_IGNORE_VALID
: 0) |
1516 (not_new
? CE_MATCH_IGNORE_MISSING
: 0));
1517 const char *modified_fmt
;
1518 const char *deleted_fmt
;
1519 const char *typechange_fmt
;
1520 const char *added_fmt
;
1521 const char *unmerged_fmt
;
1522 struct progress
*progress
= NULL
;
1523 int t2_sum_lstat
= 0;
1524 int t2_sum_scan
= 0;
1526 if (flags
& REFRESH_PROGRESS
&& isatty(2))
1527 progress
= start_delayed_progress(_("Refresh index"),
1530 trace_performance_enter();
1531 modified_fmt
= in_porcelain
? "M\t%s\n" : "%s: needs update\n";
1532 deleted_fmt
= in_porcelain
? "D\t%s\n" : "%s: needs update\n";
1533 typechange_fmt
= in_porcelain
? "T\t%s\n" : "%s: needs update\n";
1534 added_fmt
= in_porcelain
? "A\t%s\n" : "%s: needs update\n";
1535 unmerged_fmt
= in_porcelain
? "U\t%s\n" : "%s: needs merge\n";
1537 * Use the multi-threaded preload_index() to refresh most of the
1538 * cache entries quickly then in the single threaded loop below,
1539 * we only have to do the special cases that are left.
1541 preload_index(istate
, pathspec
, 0);
1542 trace2_region_enter("index", "refresh", NULL
);
1544 for (i
= 0; i
< istate
->cache_nr
; i
++) {
1545 struct cache_entry
*ce
, *new_entry
;
1546 int cache_errno
= 0;
1549 int t2_did_lstat
= 0;
1550 int t2_did_scan
= 0;
1552 ce
= istate
->cache
[i
];
1553 if (ignore_submodules
&& S_ISGITLINK(ce
->ce_mode
))
1555 if (ignore_skip_worktree
&& ce_skip_worktree(ce
))
1559 * If this entry is a sparse directory, then there isn't
1560 * any stat() information to update. Ignore the entry.
1562 if (S_ISSPARSEDIR(ce
->ce_mode
))
1565 if (pathspec
&& !ce_path_match(istate
, ce
, pathspec
, seen
))
1569 while ((i
< istate
->cache_nr
) &&
1570 ! strcmp(istate
->cache
[i
]->name
, ce
->name
))
1576 show_file(unmerged_fmt
, ce
->name
, in_porcelain
,
1577 &first
, header_msg
);
1585 new_entry
= refresh_cache_ent(istate
, ce
, options
,
1586 &cache_errno
, &changed
,
1587 &t2_did_lstat
, &t2_did_scan
);
1588 t2_sum_lstat
+= t2_did_lstat
;
1589 t2_sum_scan
+= t2_did_scan
;
1590 if (new_entry
== ce
)
1592 display_progress(progress
, i
);
1596 if (really
&& cache_errno
== EINVAL
) {
1597 /* If we are doing --really-refresh that
1598 * means the index is not valid anymore.
1600 ce
->ce_flags
&= ~CE_VALID
;
1601 ce
->ce_flags
|= CE_UPDATE_IN_BASE
;
1602 mark_fsmonitor_invalid(istate
, ce
);
1603 istate
->cache_changed
|= CE_ENTRY_CHANGED
;
1608 if (cache_errno
== ENOENT
)
1610 else if (ce_intent_to_add(ce
))
1611 fmt
= added_fmt
; /* must be before other checks */
1612 else if (changed
& TYPE_CHANGED
)
1613 fmt
= typechange_fmt
;
1617 ce
->name
, in_porcelain
, &first
, header_msg
);
1622 replace_index_entry(istate
, i
, new_entry
);
1624 trace2_data_intmax("index", NULL
, "refresh/sum_lstat", t2_sum_lstat
);
1625 trace2_data_intmax("index", NULL
, "refresh/sum_scan", t2_sum_scan
);
1626 trace2_region_leave("index", "refresh", NULL
);
1627 display_progress(progress
, istate
->cache_nr
);
1628 stop_progress(&progress
);
1629 trace_performance_leave("refresh index");
1633 struct cache_entry
*refresh_cache_entry(struct index_state
*istate
,
1634 struct cache_entry
*ce
,
1635 unsigned int options
)
1637 return refresh_cache_ent(istate
, ce
, options
, NULL
, NULL
, NULL
, NULL
);
1641 /*****************************************************************
1643 *****************************************************************/
1645 #define INDEX_FORMAT_DEFAULT 3
1647 static unsigned int get_index_format_default(struct repository
*r
)
1649 char *envversion
= getenv("GIT_INDEX_VERSION");
1651 unsigned int version
= INDEX_FORMAT_DEFAULT
;
1654 prepare_repo_settings(r
);
1656 if (r
->settings
.index_version
>= 0)
1657 version
= r
->settings
.index_version
;
1658 if (version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1659 warning(_("index.version set, but the value is invalid.\n"
1660 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1661 return INDEX_FORMAT_DEFAULT
;
1666 version
= strtoul(envversion
, &endp
, 10);
1668 version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< version
) {
1669 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1670 "Using version %i"), INDEX_FORMAT_DEFAULT
);
1671 version
= INDEX_FORMAT_DEFAULT
;
1677 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1678 * Again - this is just a (very strong in practice) heuristic that
1679 * the inode hasn't changed.
1681 * We save the fields in big-endian order to allow using the
1682 * index file over NFS transparently.
1684 struct ondisk_cache_entry
{
1685 struct cache_time ctime
;
1686 struct cache_time mtime
;
1694 * unsigned char hash[hashsz];
1696 * if (flags & CE_EXTENDED)
1699 unsigned char data
[GIT_MAX_RAWSZ
+ 2 * sizeof(uint16_t)];
1700 char name
[FLEX_ARRAY
];
1703 /* These are only used for v3 or lower */
1704 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1705 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,data) + (len) + 8) & ~7)
1706 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1707 #define ondisk_data_size(flags, len) (the_hash_algo->rawsz + \
1708 ((flags & CE_EXTENDED) ? 2 : 1) * sizeof(uint16_t) + len)
1709 #define ondisk_data_size_max(len) (ondisk_data_size(CE_EXTENDED, len))
1710 #define ondisk_ce_size(ce) (ondisk_cache_entry_size(ondisk_data_size((ce)->ce_flags, ce_namelen(ce))))
1712 /* Allow fsck to force verification of the index checksum. */
1713 int verify_index_checksum
;
1715 /* Allow fsck to force verification of the cache entry order. */
1716 int verify_ce_order
;
1718 static int verify_hdr(const struct cache_header
*hdr
, unsigned long size
)
1721 unsigned char hash
[GIT_MAX_RAWSZ
];
1723 unsigned char *start
, *end
;
1724 struct object_id oid
;
1726 if (hdr
->hdr_signature
!= htonl(CACHE_SIGNATURE
))
1727 return error(_("bad signature 0x%08x"), hdr
->hdr_signature
);
1728 hdr_version
= ntohl(hdr
->hdr_version
);
1729 if (hdr_version
< INDEX_FORMAT_LB
|| INDEX_FORMAT_UB
< hdr_version
)
1730 return error(_("bad index version %d"), hdr_version
);
1732 if (!verify_index_checksum
)
1735 end
= (unsigned char *)hdr
+ size
;
1736 start
= end
- the_hash_algo
->rawsz
;
1737 oidread(&oid
, start
);
1738 if (oideq(&oid
, null_oid()))
1741 the_hash_algo
->init_fn(&c
);
1742 the_hash_algo
->update_fn(&c
, hdr
, size
- the_hash_algo
->rawsz
);
1743 the_hash_algo
->final_fn(hash
, &c
);
1744 if (!hasheq(hash
, start
))
1745 return error(_("bad index file sha1 signature"));
1749 static int read_index_extension(struct index_state
*istate
,
1750 const char *ext
, const char *data
, unsigned long sz
)
1752 switch (CACHE_EXT(ext
)) {
1753 case CACHE_EXT_TREE
:
1754 istate
->cache_tree
= cache_tree_read(data
, sz
);
1756 case CACHE_EXT_RESOLVE_UNDO
:
1757 istate
->resolve_undo
= resolve_undo_read(data
, sz
);
1759 case CACHE_EXT_LINK
:
1760 if (read_link_extension(istate
, data
, sz
))
1763 case CACHE_EXT_UNTRACKED
:
1764 istate
->untracked
= read_untracked_extension(data
, sz
);
1766 case CACHE_EXT_FSMONITOR
:
1767 read_fsmonitor_extension(istate
, data
, sz
);
1769 case CACHE_EXT_ENDOFINDEXENTRIES
:
1770 case CACHE_EXT_INDEXENTRYOFFSETTABLE
:
1771 /* already handled in do_read_index() */
1773 case CACHE_EXT_SPARSE_DIRECTORIES
:
1774 /* no content, only an indicator */
1775 istate
->sparse_index
= INDEX_COLLAPSED
;
1778 if (*ext
< 'A' || 'Z' < *ext
)
1779 return error(_("index uses %.4s extension, which we do not understand"),
1781 fprintf_ln(stderr
, _("ignoring %.4s extension"), ext
);
1788 * Parses the contents of the cache entry contained within the 'ondisk' buffer
1789 * into a new incore 'cache_entry'.
1791 * Note that 'char *ondisk' may not be aligned to a 4-byte address interval in
1792 * index v4, so we cannot cast it to 'struct ondisk_cache_entry *' and access
1793 * its members. Instead, we use the byte offsets of members within the struct to
1794 * identify where 'get_be16()', 'get_be32()', and 'oidread()' (which can all
1795 * read from an unaligned memory buffer) should read from the 'ondisk' buffer
1796 * into the corresponding incore 'cache_entry' members.
1798 static struct cache_entry
*create_from_disk(struct mem_pool
*ce_mem_pool
,
1799 unsigned int version
,
1801 unsigned long *ent_size
,
1802 const struct cache_entry
*previous_ce
)
1804 struct cache_entry
*ce
;
1807 const unsigned hashsz
= the_hash_algo
->rawsz
;
1808 const char *flagsp
= ondisk
+ offsetof(struct ondisk_cache_entry
, data
) + hashsz
;
1810 size_t copy_len
= 0;
1812 * Adjacent cache entries tend to share the leading paths, so it makes
1813 * sense to only store the differences in later entries. In the v4
1814 * on-disk format of the index, each on-disk cache entry stores the
1815 * number of bytes to be stripped from the end of the previous name,
1816 * and the bytes to append to the result, to come up with its name.
1818 int expand_name_field
= version
== 4;
1820 /* On-disk flags are just 16 bits */
1821 flags
= get_be16(flagsp
);
1822 len
= flags
& CE_NAMEMASK
;
1824 if (flags
& CE_EXTENDED
) {
1826 extended_flags
= get_be16(flagsp
+ sizeof(uint16_t)) << 16;
1827 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1828 if (extended_flags
& ~CE_EXTENDED_FLAGS
)
1829 die(_("unknown index entry format 0x%08x"), extended_flags
);
1830 flags
|= extended_flags
;
1831 name
= (const char *)(flagsp
+ 2 * sizeof(uint16_t));
1834 name
= (const char *)(flagsp
+ sizeof(uint16_t));
1836 if (expand_name_field
) {
1837 const unsigned char *cp
= (const unsigned char *)name
;
1838 size_t strip_len
, previous_len
;
1840 /* If we're at the beginning of a block, ignore the previous name */
1841 strip_len
= decode_varint(&cp
);
1843 previous_len
= previous_ce
->ce_namelen
;
1844 if (previous_len
< strip_len
)
1845 die(_("malformed name field in the index, near path '%s'"),
1847 copy_len
= previous_len
- strip_len
;
1849 name
= (const char *)cp
;
1852 if (len
== CE_NAMEMASK
) {
1854 if (expand_name_field
)
1858 ce
= mem_pool__ce_alloc(ce_mem_pool
, len
);
1861 * NEEDSWORK: using 'offsetof()' is cumbersome and should be replaced
1862 * with something more akin to 'load_bitmap_entries_v1()'s use of
1863 * 'read_be16'/'read_be32'. For consistency with the corresponding
1864 * ondisk entry write function ('copy_cache_entry_to_ondisk()'), this
1865 * should be done at the same time as removing references to
1866 * 'ondisk_cache_entry' there.
1868 ce
->ce_stat_data
.sd_ctime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1869 + offsetof(struct cache_time
, sec
));
1870 ce
->ce_stat_data
.sd_mtime
.sec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1871 + offsetof(struct cache_time
, sec
));
1872 ce
->ce_stat_data
.sd_ctime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ctime
)
1873 + offsetof(struct cache_time
, nsec
));
1874 ce
->ce_stat_data
.sd_mtime
.nsec
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mtime
)
1875 + offsetof(struct cache_time
, nsec
));
1876 ce
->ce_stat_data
.sd_dev
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, dev
));
1877 ce
->ce_stat_data
.sd_ino
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, ino
));
1878 ce
->ce_mode
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, mode
));
1879 ce
->ce_stat_data
.sd_uid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, uid
));
1880 ce
->ce_stat_data
.sd_gid
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, gid
));
1881 ce
->ce_stat_data
.sd_size
= get_be32(ondisk
+ offsetof(struct ondisk_cache_entry
, size
));
1882 ce
->ce_flags
= flags
& ~CE_NAMEMASK
;
1883 ce
->ce_namelen
= len
;
1885 oidread(&ce
->oid
, (const unsigned char *)ondisk
+ offsetof(struct ondisk_cache_entry
, data
));
1887 if (expand_name_field
) {
1889 memcpy(ce
->name
, previous_ce
->name
, copy_len
);
1890 memcpy(ce
->name
+ copy_len
, name
, len
+ 1 - copy_len
);
1891 *ent_size
= (name
- ((char *)ondisk
)) + len
+ 1 - copy_len
;
1893 memcpy(ce
->name
, name
, len
+ 1);
1894 *ent_size
= ondisk_ce_size(ce
);
1899 static void check_ce_order(struct index_state
*istate
)
1903 if (!verify_ce_order
)
1906 for (i
= 1; i
< istate
->cache_nr
; i
++) {
1907 struct cache_entry
*ce
= istate
->cache
[i
- 1];
1908 struct cache_entry
*next_ce
= istate
->cache
[i
];
1909 int name_compare
= strcmp(ce
->name
, next_ce
->name
);
1911 if (0 < name_compare
)
1912 die(_("unordered stage entries in index"));
1913 if (!name_compare
) {
1915 die(_("multiple stage entries for merged file '%s'"),
1917 if (ce_stage(ce
) > ce_stage(next_ce
))
1918 die(_("unordered stage entries for '%s'"),
1924 static void tweak_untracked_cache(struct index_state
*istate
)
1926 struct repository
*r
= the_repository
;
1928 prepare_repo_settings(r
);
1930 switch (r
->settings
.core_untracked_cache
) {
1931 case UNTRACKED_CACHE_REMOVE
:
1932 remove_untracked_cache(istate
);
1934 case UNTRACKED_CACHE_WRITE
:
1935 add_untracked_cache(istate
);
1937 case UNTRACKED_CACHE_KEEP
:
1939 * Either an explicit "core.untrackedCache=keep", the
1940 * default if "core.untrackedCache" isn't configured,
1941 * or a fallback on an unknown "core.untrackedCache"
1948 static void tweak_split_index(struct index_state
*istate
)
1950 switch (git_config_get_split_index()) {
1951 case -1: /* unset: do nothing */
1954 remove_split_index(istate
);
1957 add_split_index(istate
);
1959 default: /* unknown value: do nothing */
1964 static void post_read_index_from(struct index_state
*istate
)
1966 check_ce_order(istate
);
1967 tweak_untracked_cache(istate
);
1968 tweak_split_index(istate
);
1969 tweak_fsmonitor(istate
);
1972 static size_t estimate_cache_size_from_compressed(unsigned int entries
)
1974 return entries
* (sizeof(struct cache_entry
) + CACHE_ENTRY_PATH_LENGTH
);
1977 static size_t estimate_cache_size(size_t ondisk_size
, unsigned int entries
)
1979 long per_entry
= sizeof(struct cache_entry
) - sizeof(struct ondisk_cache_entry
);
1982 * Account for potential alignment differences.
1984 per_entry
+= align_padding_size(per_entry
, 0);
1985 return ondisk_size
+ entries
* per_entry
;
1988 struct index_entry_offset
1990 /* starting byte offset into index file, count of index entries in this block */
1994 struct index_entry_offset_table
1997 struct index_entry_offset entries
[FLEX_ARRAY
];
2000 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
);
2001 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
);
2003 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
);
2004 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
);
2006 struct load_index_extensions
2009 struct index_state
*istate
;
2012 unsigned long src_offset
;
2015 static void *load_index_extensions(void *_data
)
2017 struct load_index_extensions
*p
= _data
;
2018 unsigned long src_offset
= p
->src_offset
;
2020 while (src_offset
<= p
->mmap_size
- the_hash_algo
->rawsz
- 8) {
2021 /* After an array of active_nr index entries,
2022 * there can be arbitrary number of extended
2023 * sections, each of which is prefixed with
2024 * extension name (4-byte) and section length
2025 * in 4-byte network byte order.
2027 uint32_t extsize
= get_be32(p
->mmap
+ src_offset
+ 4);
2028 if (read_index_extension(p
->istate
,
2029 p
->mmap
+ src_offset
,
2030 p
->mmap
+ src_offset
+ 8,
2032 munmap((void *)p
->mmap
, p
->mmap_size
);
2033 die(_("index file corrupt"));
2036 src_offset
+= extsize
;
2043 * A helper function that will load the specified range of cache entries
2044 * from the memory mapped file and add them to the given index.
2046 static unsigned long load_cache_entry_block(struct index_state
*istate
,
2047 struct mem_pool
*ce_mem_pool
, int offset
, int nr
, const char *mmap
,
2048 unsigned long start_offset
, const struct cache_entry
*previous_ce
)
2051 unsigned long src_offset
= start_offset
;
2053 for (i
= offset
; i
< offset
+ nr
; i
++) {
2054 struct cache_entry
*ce
;
2055 unsigned long consumed
;
2057 ce
= create_from_disk(ce_mem_pool
, istate
->version
,
2059 &consumed
, previous_ce
);
2060 set_index_entry(istate
, i
, ce
);
2062 src_offset
+= consumed
;
2065 return src_offset
- start_offset
;
2068 static unsigned long load_all_cache_entries(struct index_state
*istate
,
2069 const char *mmap
, size_t mmap_size
, unsigned long src_offset
)
2071 unsigned long consumed
;
2073 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2074 if (istate
->version
== 4) {
2075 mem_pool_init(istate
->ce_mem_pool
,
2076 estimate_cache_size_from_compressed(istate
->cache_nr
));
2078 mem_pool_init(istate
->ce_mem_pool
,
2079 estimate_cache_size(mmap_size
, istate
->cache_nr
));
2082 consumed
= load_cache_entry_block(istate
, istate
->ce_mem_pool
,
2083 0, istate
->cache_nr
, mmap
, src_offset
, NULL
);
2088 * Mostly randomly chosen maximum thread counts: we
2089 * cap the parallelism to online_cpus() threads, and we want
2090 * to have at least 10000 cache entries per thread for it to
2091 * be worth starting a thread.
2094 #define THREAD_COST (10000)
2096 struct load_cache_entries_thread_data
2099 struct index_state
*istate
;
2100 struct mem_pool
*ce_mem_pool
;
2103 struct index_entry_offset_table
*ieot
;
2104 int ieot_start
; /* starting index into the ieot array */
2105 int ieot_blocks
; /* count of ieot entries to process */
2106 unsigned long consumed
; /* return # of bytes in index file processed */
2110 * A thread proc to run the load_cache_entries() computation
2111 * across multiple background threads.
2113 static void *load_cache_entries_thread(void *_data
)
2115 struct load_cache_entries_thread_data
*p
= _data
;
2118 /* iterate across all ieot blocks assigned to this thread */
2119 for (i
= p
->ieot_start
; i
< p
->ieot_start
+ p
->ieot_blocks
; i
++) {
2120 p
->consumed
+= load_cache_entry_block(p
->istate
, p
->ce_mem_pool
,
2121 p
->offset
, p
->ieot
->entries
[i
].nr
, p
->mmap
, p
->ieot
->entries
[i
].offset
, NULL
);
2122 p
->offset
+= p
->ieot
->entries
[i
].nr
;
2127 static unsigned long load_cache_entries_threaded(struct index_state
*istate
, const char *mmap
, size_t mmap_size
,
2128 int nr_threads
, struct index_entry_offset_table
*ieot
)
2130 int i
, offset
, ieot_blocks
, ieot_start
, err
;
2131 struct load_cache_entries_thread_data
*data
;
2132 unsigned long consumed
= 0;
2134 /* a little sanity checking */
2135 if (istate
->name_hash_initialized
)
2136 BUG("the name hash isn't thread safe");
2138 istate
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2139 mem_pool_init(istate
->ce_mem_pool
, 0);
2141 /* ensure we have no more threads than we have blocks to process */
2142 if (nr_threads
> ieot
->nr
)
2143 nr_threads
= ieot
->nr
;
2144 CALLOC_ARRAY(data
, nr_threads
);
2146 offset
= ieot_start
= 0;
2147 ieot_blocks
= DIV_ROUND_UP(ieot
->nr
, nr_threads
);
2148 for (i
= 0; i
< nr_threads
; i
++) {
2149 struct load_cache_entries_thread_data
*p
= &data
[i
];
2152 if (ieot_start
+ ieot_blocks
> ieot
->nr
)
2153 ieot_blocks
= ieot
->nr
- ieot_start
;
2159 p
->ieot_start
= ieot_start
;
2160 p
->ieot_blocks
= ieot_blocks
;
2162 /* create a mem_pool for each thread */
2164 for (j
= p
->ieot_start
; j
< p
->ieot_start
+ p
->ieot_blocks
; j
++)
2165 nr
+= p
->ieot
->entries
[j
].nr
;
2166 p
->ce_mem_pool
= xmalloc(sizeof(*istate
->ce_mem_pool
));
2167 if (istate
->version
== 4) {
2168 mem_pool_init(p
->ce_mem_pool
,
2169 estimate_cache_size_from_compressed(nr
));
2171 mem_pool_init(p
->ce_mem_pool
,
2172 estimate_cache_size(mmap_size
, nr
));
2175 err
= pthread_create(&p
->pthread
, NULL
, load_cache_entries_thread
, p
);
2177 die(_("unable to create load_cache_entries thread: %s"), strerror(err
));
2179 /* increment by the number of cache entries in the ieot block being processed */
2180 for (j
= 0; j
< ieot_blocks
; j
++)
2181 offset
+= ieot
->entries
[ieot_start
+ j
].nr
;
2182 ieot_start
+= ieot_blocks
;
2185 for (i
= 0; i
< nr_threads
; i
++) {
2186 struct load_cache_entries_thread_data
*p
= &data
[i
];
2188 err
= pthread_join(p
->pthread
, NULL
);
2190 die(_("unable to join load_cache_entries thread: %s"), strerror(err
));
2191 mem_pool_combine(istate
->ce_mem_pool
, p
->ce_mem_pool
);
2192 consumed
+= p
->consumed
;
2200 static void set_new_index_sparsity(struct index_state
*istate
)
2203 * If the index's repo exists, mark it sparse according to
2206 prepare_repo_settings(istate
->repo
);
2207 if (!istate
->repo
->settings
.command_requires_full_index
&&
2208 is_sparse_index_allowed(istate
, 0))
2209 istate
->sparse_index
= 1;
2212 /* remember to discard_cache() before reading a different cache! */
2213 int do_read_index(struct index_state
*istate
, const char *path
, int must_exist
)
2217 unsigned long src_offset
;
2218 const struct cache_header
*hdr
;
2221 struct load_index_extensions p
;
2222 size_t extension_offset
= 0;
2223 int nr_threads
, cpus
;
2224 struct index_entry_offset_table
*ieot
= NULL
;
2226 if (istate
->initialized
)
2227 return istate
->cache_nr
;
2229 istate
->timestamp
.sec
= 0;
2230 istate
->timestamp
.nsec
= 0;
2231 fd
= open(path
, O_RDONLY
);
2233 if (!must_exist
&& errno
== ENOENT
) {
2234 set_new_index_sparsity(istate
);
2235 istate
->initialized
= 1;
2238 die_errno(_("%s: index file open failed"), path
);
2242 die_errno(_("%s: cannot stat the open index"), path
);
2244 mmap_size
= xsize_t(st
.st_size
);
2245 if (mmap_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2246 die(_("%s: index file smaller than expected"), path
);
2248 mmap
= xmmap_gently(NULL
, mmap_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
2249 if (mmap
== MAP_FAILED
)
2250 die_errno(_("%s: unable to map index file%s"), path
,
2254 hdr
= (const struct cache_header
*)mmap
;
2255 if (verify_hdr(hdr
, mmap_size
) < 0)
2258 oidread(&istate
->oid
, (const unsigned char *)hdr
+ mmap_size
- the_hash_algo
->rawsz
);
2259 istate
->version
= ntohl(hdr
->hdr_version
);
2260 istate
->cache_nr
= ntohl(hdr
->hdr_entries
);
2261 istate
->cache_alloc
= alloc_nr(istate
->cache_nr
);
2262 CALLOC_ARRAY(istate
->cache
, istate
->cache_alloc
);
2263 istate
->initialized
= 1;
2267 p
.mmap_size
= mmap_size
;
2269 src_offset
= sizeof(*hdr
);
2271 if (git_config_get_index_threads(&nr_threads
))
2274 /* TODO: does creating more threads than cores help? */
2276 nr_threads
= istate
->cache_nr
/ THREAD_COST
;
2277 cpus
= online_cpus();
2278 if (nr_threads
> cpus
)
2285 if (nr_threads
> 1) {
2286 extension_offset
= read_eoie_extension(mmap
, mmap_size
);
2287 if (extension_offset
) {
2290 p
.src_offset
= extension_offset
;
2291 err
= pthread_create(&p
.pthread
, NULL
, load_index_extensions
, &p
);
2293 die(_("unable to create load_index_extensions thread: %s"), strerror(err
));
2300 * Locate and read the index entry offset table so that we can use it
2301 * to multi-thread the reading of the cache entries.
2303 if (extension_offset
&& nr_threads
> 1)
2304 ieot
= read_ieot_extension(mmap
, mmap_size
, extension_offset
);
2307 src_offset
+= load_cache_entries_threaded(istate
, mmap
, mmap_size
, nr_threads
, ieot
);
2310 src_offset
+= load_all_cache_entries(istate
, mmap
, mmap_size
, src_offset
);
2313 istate
->timestamp
.sec
= st
.st_mtime
;
2314 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
2316 /* if we created a thread, join it otherwise load the extensions on the primary thread */
2317 if (extension_offset
) {
2318 int ret
= pthread_join(p
.pthread
, NULL
);
2320 die(_("unable to join load_index_extensions thread: %s"), strerror(ret
));
2322 p
.src_offset
= src_offset
;
2323 load_index_extensions(&p
);
2325 munmap((void *)mmap
, mmap_size
);
2328 * TODO trace2: replace "the_repository" with the actual repo instance
2329 * that is associated with the given "istate".
2331 trace2_data_intmax("index", the_repository
, "read/version",
2333 trace2_data_intmax("index", the_repository
, "read/cache_nr",
2337 * If the command explicitly requires a full index, force it
2338 * to be full. Otherwise, correct the sparsity based on repository
2339 * settings and other properties of the index (if necessary).
2341 prepare_repo_settings(istate
->repo
);
2342 if (istate
->repo
->settings
.command_requires_full_index
)
2343 ensure_full_index(istate
);
2345 ensure_correct_sparsity(istate
);
2347 return istate
->cache_nr
;
2350 munmap((void *)mmap
, mmap_size
);
2351 die(_("index file corrupt"));
2355 * Signal that the shared index is used by updating its mtime.
2357 * This way, shared index can be removed if they have not been used
2360 static void freshen_shared_index(const char *shared_index
, int warn
)
2362 if (!check_and_freshen_file(shared_index
, 1) && warn
)
2363 warning(_("could not freshen shared index '%s'"), shared_index
);
2366 int read_index_from(struct index_state
*istate
, const char *path
,
2369 struct split_index
*split_index
;
2374 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
2375 if (istate
->initialized
)
2376 return istate
->cache_nr
;
2379 * TODO trace2: replace "the_repository" with the actual repo instance
2380 * that is associated with the given "istate".
2382 trace2_region_enter_printf("index", "do_read_index", the_repository
,
2384 trace_performance_enter();
2385 ret
= do_read_index(istate
, path
, 0);
2386 trace_performance_leave("read cache %s", path
);
2387 trace2_region_leave_printf("index", "do_read_index", the_repository
,
2390 split_index
= istate
->split_index
;
2391 if (!split_index
|| is_null_oid(&split_index
->base_oid
)) {
2392 post_read_index_from(istate
);
2396 trace_performance_enter();
2397 if (split_index
->base
)
2398 release_index(split_index
->base
);
2400 ALLOC_ARRAY(split_index
->base
, 1);
2401 index_state_init(split_index
->base
, istate
->repo
);
2403 base_oid_hex
= oid_to_hex(&split_index
->base_oid
);
2404 base_path
= xstrfmt("%s/sharedindex.%s", gitdir
, base_oid_hex
);
2405 if (file_exists(base_path
)) {
2406 trace2_region_enter_printf("index", "shared/do_read_index",
2407 the_repository
, "%s", base_path
);
2409 ret
= do_read_index(split_index
->base
, base_path
, 0);
2410 trace2_region_leave_printf("index", "shared/do_read_index",
2411 the_repository
, "%s", base_path
);
2413 char *path_copy
= xstrdup(path
);
2414 char *base_path2
= xstrfmt("%s/sharedindex.%s",
2415 dirname(path_copy
), base_oid_hex
);
2417 trace2_region_enter_printf("index", "shared/do_read_index",
2418 the_repository
, "%s", base_path2
);
2419 ret
= do_read_index(split_index
->base
, base_path2
, 1);
2420 trace2_region_leave_printf("index", "shared/do_read_index",
2421 the_repository
, "%s", base_path2
);
2424 if (!oideq(&split_index
->base_oid
, &split_index
->base
->oid
))
2425 die(_("broken index, expect %s in %s, got %s"),
2426 base_oid_hex
, base_path
,
2427 oid_to_hex(&split_index
->base
->oid
));
2429 freshen_shared_index(base_path
, 0);
2430 merge_base_index(istate
);
2431 post_read_index_from(istate
);
2432 trace_performance_leave("read cache %s", base_path
);
2437 int is_index_unborn(struct index_state
*istate
)
2439 return (!istate
->cache_nr
&& !istate
->timestamp
.sec
);
2442 void index_state_init(struct index_state
*istate
, struct repository
*r
)
2444 struct index_state blank
= INDEX_STATE_INIT(r
);
2445 memcpy(istate
, &blank
, sizeof(*istate
));
2448 void release_index(struct index_state
*istate
)
2451 * Cache entries in istate->cache[] should have been allocated
2452 * from the memory pool associated with this index, or from an
2453 * associated split_index. There is no need to free individual
2454 * cache entries. validate_cache_entries can detect when this
2455 * assertion does not hold.
2457 validate_cache_entries(istate
);
2459 resolve_undo_clear_index(istate
);
2460 free_name_hash(istate
);
2461 cache_tree_free(&(istate
->cache_tree
));
2462 free(istate
->fsmonitor_last_update
);
2463 free(istate
->cache
);
2464 discard_split_index(istate
);
2465 free_untracked_cache(istate
->untracked
);
2467 if (istate
->sparse_checkout_patterns
) {
2468 clear_pattern_list(istate
->sparse_checkout_patterns
);
2469 FREE_AND_NULL(istate
->sparse_checkout_patterns
);
2472 if (istate
->ce_mem_pool
) {
2473 mem_pool_discard(istate
->ce_mem_pool
, should_validate_cache_entries());
2474 FREE_AND_NULL(istate
->ce_mem_pool
);
2478 void discard_index(struct index_state
*istate
)
2480 release_index(istate
);
2481 index_state_init(istate
, istate
->repo
);
2485 * Validate the cache entries of this index.
2486 * All cache entries associated with this index
2487 * should have been allocated by the memory pool
2488 * associated with this index, or by a referenced
2491 void validate_cache_entries(const struct index_state
*istate
)
2495 if (!should_validate_cache_entries() ||!istate
|| !istate
->initialized
)
2498 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2500 BUG("cache entry is not allocated from expected memory pool");
2501 } else if (!istate
->ce_mem_pool
||
2502 !mem_pool_contains(istate
->ce_mem_pool
, istate
->cache
[i
])) {
2503 if (!istate
->split_index
||
2504 !istate
->split_index
->base
||
2505 !istate
->split_index
->base
->ce_mem_pool
||
2506 !mem_pool_contains(istate
->split_index
->base
->ce_mem_pool
, istate
->cache
[i
])) {
2507 BUG("cache entry is not allocated from expected memory pool");
2512 if (istate
->split_index
)
2513 validate_cache_entries(istate
->split_index
->base
);
2516 int unmerged_index(const struct index_state
*istate
)
2519 for (i
= 0; i
< istate
->cache_nr
; i
++) {
2520 if (ce_stage(istate
->cache
[i
]))
2526 int repo_index_has_changes(struct repository
*repo
,
2530 struct index_state
*istate
= repo
->index
;
2531 struct object_id cmp
;
2535 cmp
= tree
->object
.oid
;
2536 if (tree
|| !repo_get_oid_tree(repo
, "HEAD", &cmp
)) {
2537 struct diff_options opt
;
2539 repo_diff_setup(repo
, &opt
);
2540 opt
.flags
.exit_with_status
= 1;
2542 opt
.flags
.quick
= 1;
2543 diff_setup_done(&opt
);
2544 do_diff_cache(&cmp
, &opt
);
2546 for (i
= 0; sb
&& i
< diff_queued_diff
.nr
; i
++) {
2548 strbuf_addch(sb
, ' ');
2549 strbuf_addstr(sb
, diff_queued_diff
.queue
[i
]->two
->path
);
2552 return opt
.flags
.has_changes
!= 0;
2554 /* TODO: audit for interaction with sparse-index. */
2555 ensure_full_index(istate
);
2556 for (i
= 0; sb
&& i
< istate
->cache_nr
; i
++) {
2558 strbuf_addch(sb
, ' ');
2559 strbuf_addstr(sb
, istate
->cache
[i
]->name
);
2561 return !!istate
->cache_nr
;
2565 static int write_index_ext_header(struct hashfile
*f
,
2566 git_hash_ctx
*eoie_f
,
2570 hashwrite_be32(f
, ext
);
2571 hashwrite_be32(f
, sz
);
2576 the_hash_algo
->update_fn(eoie_f
, &ext
, sizeof(ext
));
2577 the_hash_algo
->update_fn(eoie_f
, &sz
, sizeof(sz
));
2582 static void ce_smudge_racily_clean_entry(struct index_state
*istate
,
2583 struct cache_entry
*ce
)
2586 * The only thing we care about in this function is to smudge the
2587 * falsely clean entry due to touch-update-touch race, so we leave
2588 * everything else as they are. We are called for entries whose
2589 * ce_stat_data.sd_mtime match the index file mtime.
2591 * Note that this actually does not do much for gitlinks, for
2592 * which ce_match_stat_basic() always goes to the actual
2593 * contents. The caller checks with is_racy_timestamp() which
2594 * always says "no" for gitlinks, so we are not called for them ;-)
2598 if (lstat(ce
->name
, &st
) < 0)
2600 if (ce_match_stat_basic(ce
, &st
))
2602 if (ce_modified_check_fs(istate
, ce
, &st
)) {
2603 /* This is "racily clean"; smudge it. Note that this
2604 * is a tricky code. At first glance, it may appear
2605 * that it can break with this sequence:
2607 * $ echo xyzzy >frotz
2608 * $ git-update-index --add frotz
2611 * $ echo filfre >nitfol
2612 * $ git-update-index --add nitfol
2614 * but it does not. When the second update-index runs,
2615 * it notices that the entry "frotz" has the same timestamp
2616 * as index, and if we were to smudge it by resetting its
2617 * size to zero here, then the object name recorded
2618 * in index is the 6-byte file but the cached stat information
2619 * becomes zero --- which would then match what we would
2620 * obtain from the filesystem next time we stat("frotz").
2622 * However, the second update-index, before calling
2623 * this function, notices that the cached size is 6
2624 * bytes and what is on the filesystem is an empty
2625 * file, and never calls us, so the cached size information
2626 * for "frotz" stays 6 which does not match the filesystem.
2628 ce
->ce_stat_data
.sd_size
= 0;
2632 /* Copy miscellaneous fields but not the name */
2633 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry
*ondisk
,
2634 struct cache_entry
*ce
)
2637 const unsigned hashsz
= the_hash_algo
->rawsz
;
2638 uint16_t *flagsp
= (uint16_t *)(ondisk
->data
+ hashsz
);
2640 ondisk
->ctime
.sec
= htonl(ce
->ce_stat_data
.sd_ctime
.sec
);
2641 ondisk
->mtime
.sec
= htonl(ce
->ce_stat_data
.sd_mtime
.sec
);
2642 ondisk
->ctime
.nsec
= htonl(ce
->ce_stat_data
.sd_ctime
.nsec
);
2643 ondisk
->mtime
.nsec
= htonl(ce
->ce_stat_data
.sd_mtime
.nsec
);
2644 ondisk
->dev
= htonl(ce
->ce_stat_data
.sd_dev
);
2645 ondisk
->ino
= htonl(ce
->ce_stat_data
.sd_ino
);
2646 ondisk
->mode
= htonl(ce
->ce_mode
);
2647 ondisk
->uid
= htonl(ce
->ce_stat_data
.sd_uid
);
2648 ondisk
->gid
= htonl(ce
->ce_stat_data
.sd_gid
);
2649 ondisk
->size
= htonl(ce
->ce_stat_data
.sd_size
);
2650 hashcpy(ondisk
->data
, ce
->oid
.hash
);
2652 flags
= ce
->ce_flags
& ~CE_NAMEMASK
;
2653 flags
|= (ce_namelen(ce
) >= CE_NAMEMASK
? CE_NAMEMASK
: ce_namelen(ce
));
2654 flagsp
[0] = htons(flags
);
2655 if (ce
->ce_flags
& CE_EXTENDED
) {
2656 flagsp
[1] = htons((ce
->ce_flags
& CE_EXTENDED_FLAGS
) >> 16);
2660 static int ce_write_entry(struct hashfile
*f
, struct cache_entry
*ce
,
2661 struct strbuf
*previous_name
, struct ondisk_cache_entry
*ondisk
)
2664 unsigned int saved_namelen
;
2665 int stripped_name
= 0;
2666 static unsigned char padding
[8] = { 0x00 };
2668 if (ce
->ce_flags
& CE_STRIP_NAME
) {
2669 saved_namelen
= ce_namelen(ce
);
2674 size
= offsetof(struct ondisk_cache_entry
,data
) + ondisk_data_size(ce
->ce_flags
, 0);
2676 if (!previous_name
) {
2677 int len
= ce_namelen(ce
);
2678 copy_cache_entry_to_ondisk(ondisk
, ce
);
2679 hashwrite(f
, ondisk
, size
);
2680 hashwrite(f
, ce
->name
, len
);
2681 hashwrite(f
, padding
, align_padding_size(size
, len
));
2683 int common
, to_remove
, prefix_size
;
2684 unsigned char to_remove_vi
[16];
2686 (ce
->name
[common
] &&
2687 common
< previous_name
->len
&&
2688 ce
->name
[common
] == previous_name
->buf
[common
]);
2690 ; /* still matching */
2691 to_remove
= previous_name
->len
- common
;
2692 prefix_size
= encode_varint(to_remove
, to_remove_vi
);
2694 copy_cache_entry_to_ondisk(ondisk
, ce
);
2695 hashwrite(f
, ondisk
, size
);
2696 hashwrite(f
, to_remove_vi
, prefix_size
);
2697 hashwrite(f
, ce
->name
+ common
, ce_namelen(ce
) - common
);
2698 hashwrite(f
, padding
, 1);
2700 strbuf_splice(previous_name
, common
, to_remove
,
2701 ce
->name
+ common
, ce_namelen(ce
) - common
);
2703 if (stripped_name
) {
2704 ce
->ce_namelen
= saved_namelen
;
2705 ce
->ce_flags
&= ~CE_STRIP_NAME
;
2712 * This function verifies if index_state has the correct sha1 of the
2713 * index file. Don't die if we have any other failure, just return 0.
2715 static int verify_index_from(const struct index_state
*istate
, const char *path
)
2720 unsigned char hash
[GIT_MAX_RAWSZ
];
2722 if (!istate
->initialized
)
2725 fd
= open(path
, O_RDONLY
);
2732 if (st
.st_size
< sizeof(struct cache_header
) + the_hash_algo
->rawsz
)
2735 n
= pread_in_full(fd
, hash
, the_hash_algo
->rawsz
, st
.st_size
- the_hash_algo
->rawsz
);
2736 if (n
!= the_hash_algo
->rawsz
)
2739 if (!hasheq(istate
->oid
.hash
, hash
))
2750 static int repo_verify_index(struct repository
*repo
)
2752 return verify_index_from(repo
->index
, repo
->index_file
);
2755 int has_racy_timestamp(struct index_state
*istate
)
2757 int entries
= istate
->cache_nr
;
2760 for (i
= 0; i
< entries
; i
++) {
2761 struct cache_entry
*ce
= istate
->cache
[i
];
2762 if (is_racy_timestamp(istate
, ce
))
2768 void repo_update_index_if_able(struct repository
*repo
,
2769 struct lock_file
*lockfile
)
2771 if ((repo
->index
->cache_changed
||
2772 has_racy_timestamp(repo
->index
)) &&
2773 repo_verify_index(repo
))
2774 write_locked_index(repo
->index
, lockfile
, COMMIT_LOCK
);
2776 rollback_lock_file(lockfile
);
2779 static int record_eoie(void)
2783 if (!git_config_get_bool("index.recordendofindexentries", &val
))
2787 * As a convenience, the end of index entries extension
2788 * used for threading is written by default if the user
2789 * explicitly requested threaded index reads.
2791 return !git_config_get_index_threads(&val
) && val
!= 1;
2794 static int record_ieot(void)
2798 if (!git_config_get_bool("index.recordoffsettable", &val
))
2802 * As a convenience, the offset table used for threading is
2803 * written by default if the user explicitly requested
2804 * threaded index reads.
2806 return !git_config_get_index_threads(&val
) && val
!= 1;
2809 enum write_extensions
{
2810 WRITE_NO_EXTENSION
= 0,
2811 WRITE_SPLIT_INDEX_EXTENSION
= 1<<0,
2812 WRITE_CACHE_TREE_EXTENSION
= 1<<1,
2813 WRITE_RESOLVE_UNDO_EXTENSION
= 1<<2,
2814 WRITE_UNTRACKED_CACHE_EXTENSION
= 1<<3,
2815 WRITE_FSMONITOR_EXTENSION
= 1<<4,
2817 #define WRITE_ALL_EXTENSIONS ((enum write_extensions)-1)
2820 * On success, `tempfile` is closed. If it is the temporary file
2821 * of a `struct lock_file`, we will therefore effectively perform
2822 * a 'close_lock_file_gently()`. Since that is an implementation
2823 * detail of lockfiles, callers of `do_write_index()` should not
2826 static int do_write_index(struct index_state
*istate
, struct tempfile
*tempfile
,
2827 enum write_extensions write_extensions
, unsigned flags
)
2829 uint64_t start
= getnanotime();
2831 git_hash_ctx
*eoie_c
= NULL
;
2832 struct cache_header hdr
;
2833 int i
, err
= 0, removed
, extended
, hdr_version
;
2834 struct cache_entry
**cache
= istate
->cache
;
2835 int entries
= istate
->cache_nr
;
2837 struct ondisk_cache_entry ondisk
;
2838 struct strbuf previous_name_buf
= STRBUF_INIT
, *previous_name
;
2839 int drop_cache_tree
= istate
->drop_cache_tree
;
2841 int csum_fsync_flag
;
2842 int ieot_entries
= 1;
2843 struct index_entry_offset_table
*ieot
= NULL
;
2845 struct repository
*r
= istate
->repo
;
2847 f
= hashfd(tempfile
->fd
, tempfile
->filename
.buf
);
2849 prepare_repo_settings(r
);
2850 f
->skip_hash
= r
->settings
.index_skip_hash
;
2852 for (i
= removed
= extended
= 0; i
< entries
; i
++) {
2853 if (cache
[i
]->ce_flags
& CE_REMOVE
)
2856 /* reduce extended entries if possible */
2857 cache
[i
]->ce_flags
&= ~CE_EXTENDED
;
2858 if (cache
[i
]->ce_flags
& CE_EXTENDED_FLAGS
) {
2860 cache
[i
]->ce_flags
|= CE_EXTENDED
;
2864 if (!istate
->version
)
2865 istate
->version
= get_index_format_default(r
);
2867 /* demote version 3 to version 2 when the latter suffices */
2868 if (istate
->version
== 3 || istate
->version
== 2)
2869 istate
->version
= extended
? 3 : 2;
2871 hdr_version
= istate
->version
;
2873 hdr
.hdr_signature
= htonl(CACHE_SIGNATURE
);
2874 hdr
.hdr_version
= htonl(hdr_version
);
2875 hdr
.hdr_entries
= htonl(entries
- removed
);
2877 hashwrite(f
, &hdr
, sizeof(hdr
));
2879 if (!HAVE_THREADS
|| git_config_get_index_threads(&nr_threads
))
2882 if (nr_threads
!= 1 && record_ieot()) {
2883 int ieot_blocks
, cpus
;
2886 * ensure default number of ieot blocks maps evenly to the
2887 * default number of threads that will process them leaving
2888 * room for the thread to load the index extensions.
2891 ieot_blocks
= istate
->cache_nr
/ THREAD_COST
;
2892 cpus
= online_cpus();
2893 if (ieot_blocks
> cpus
- 1)
2894 ieot_blocks
= cpus
- 1;
2896 ieot_blocks
= nr_threads
;
2897 if (ieot_blocks
> istate
->cache_nr
)
2898 ieot_blocks
= istate
->cache_nr
;
2902 * no reason to write out the IEOT extension if we don't
2903 * have enough blocks to utilize multi-threading
2905 if (ieot_blocks
> 1) {
2906 ieot
= xcalloc(1, sizeof(struct index_entry_offset_table
)
2907 + (ieot_blocks
* sizeof(struct index_entry_offset
)));
2908 ieot_entries
= DIV_ROUND_UP(entries
, ieot_blocks
);
2912 offset
= hashfile_total(f
);
2915 previous_name
= (hdr_version
== 4) ? &previous_name_buf
: NULL
;
2917 for (i
= 0; i
< entries
; i
++) {
2918 struct cache_entry
*ce
= cache
[i
];
2919 if (ce
->ce_flags
& CE_REMOVE
)
2921 if (!ce_uptodate(ce
) && is_racy_timestamp(istate
, ce
))
2922 ce_smudge_racily_clean_entry(istate
, ce
);
2923 if (is_null_oid(&ce
->oid
)) {
2924 static const char msg
[] = "cache entry has null sha1: %s";
2925 static int allow
= -1;
2928 allow
= git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2930 warning(msg
, ce
->name
);
2932 err
= error(msg
, ce
->name
);
2934 drop_cache_tree
= 1;
2936 if (ieot
&& i
&& (i
% ieot_entries
== 0)) {
2937 ieot
->entries
[ieot
->nr
].nr
= nr
;
2938 ieot
->entries
[ieot
->nr
].offset
= offset
;
2941 * If we have a V4 index, set the first byte to an invalid
2942 * character to ensure there is nothing common with the previous
2946 previous_name
->buf
[0] = 0;
2949 offset
= hashfile_total(f
);
2951 if (ce_write_entry(f
, ce
, previous_name
, (struct ondisk_cache_entry
*)&ondisk
) < 0)
2959 ieot
->entries
[ieot
->nr
].nr
= nr
;
2960 ieot
->entries
[ieot
->nr
].offset
= offset
;
2963 strbuf_release(&previous_name_buf
);
2970 offset
= hashfile_total(f
);
2973 * The extension headers must be hashed on their own for the
2974 * EOIE extension. Create a hashfile here to compute that hash.
2976 if (offset
&& record_eoie()) {
2977 CALLOC_ARRAY(eoie_c
, 1);
2978 the_hash_algo
->init_fn(eoie_c
);
2982 * Lets write out CACHE_EXT_INDEXENTRYOFFSETTABLE first so that we
2983 * can minimize the number of extensions we have to scan through to
2984 * find it during load. Write it out regardless of the
2985 * strip_extensions parameter as we need it when loading the shared
2989 struct strbuf sb
= STRBUF_INIT
;
2991 write_ieot_extension(&sb
, ieot
);
2992 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_INDEXENTRYOFFSETTABLE
, sb
.len
) < 0;
2993 hashwrite(f
, sb
.buf
, sb
.len
);
2994 strbuf_release(&sb
);
3000 if (write_extensions
& WRITE_SPLIT_INDEX_EXTENSION
&&
3001 istate
->split_index
) {
3002 struct strbuf sb
= STRBUF_INIT
;
3004 if (istate
->sparse_index
)
3005 die(_("cannot write split index for a sparse index"));
3007 err
= write_link_extension(&sb
, istate
) < 0 ||
3008 write_index_ext_header(f
, eoie_c
, CACHE_EXT_LINK
,
3010 hashwrite(f
, sb
.buf
, sb
.len
);
3011 strbuf_release(&sb
);
3015 if (write_extensions
& WRITE_CACHE_TREE_EXTENSION
&&
3016 !drop_cache_tree
&& istate
->cache_tree
) {
3017 struct strbuf sb
= STRBUF_INIT
;
3019 cache_tree_write(&sb
, istate
->cache_tree
);
3020 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_TREE
, sb
.len
) < 0;
3021 hashwrite(f
, sb
.buf
, sb
.len
);
3022 strbuf_release(&sb
);
3026 if (write_extensions
& WRITE_RESOLVE_UNDO_EXTENSION
&&
3027 istate
->resolve_undo
) {
3028 struct strbuf sb
= STRBUF_INIT
;
3030 resolve_undo_write(&sb
, istate
->resolve_undo
);
3031 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_RESOLVE_UNDO
,
3033 hashwrite(f
, sb
.buf
, sb
.len
);
3034 strbuf_release(&sb
);
3038 if (write_extensions
& WRITE_UNTRACKED_CACHE_EXTENSION
&&
3039 istate
->untracked
) {
3040 struct strbuf sb
= STRBUF_INIT
;
3042 write_untracked_extension(&sb
, istate
->untracked
);
3043 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_UNTRACKED
,
3045 hashwrite(f
, sb
.buf
, sb
.len
);
3046 strbuf_release(&sb
);
3050 if (write_extensions
& WRITE_FSMONITOR_EXTENSION
&&
3051 istate
->fsmonitor_last_update
) {
3052 struct strbuf sb
= STRBUF_INIT
;
3054 write_fsmonitor_extension(&sb
, istate
);
3055 err
= write_index_ext_header(f
, eoie_c
, CACHE_EXT_FSMONITOR
, sb
.len
) < 0;
3056 hashwrite(f
, sb
.buf
, sb
.len
);
3057 strbuf_release(&sb
);
3061 if (istate
->sparse_index
) {
3062 if (write_index_ext_header(f
, eoie_c
, CACHE_EXT_SPARSE_DIRECTORIES
, 0) < 0)
3067 * CACHE_EXT_ENDOFINDEXENTRIES must be written as the last entry before the SHA1
3068 * so that it can be found and processed before all the index entries are
3069 * read. Write it out regardless of the strip_extensions parameter as we need it
3070 * when loading the shared index.
3073 struct strbuf sb
= STRBUF_INIT
;
3075 write_eoie_extension(&sb
, eoie_c
, offset
);
3076 err
= write_index_ext_header(f
, NULL
, CACHE_EXT_ENDOFINDEXENTRIES
, sb
.len
) < 0;
3077 hashwrite(f
, sb
.buf
, sb
.len
);
3078 strbuf_release(&sb
);
3083 csum_fsync_flag
= 0;
3084 if (!alternate_index_output
&& (flags
& COMMIT_LOCK
))
3085 csum_fsync_flag
= CSUM_FSYNC
;
3087 finalize_hashfile(f
, istate
->oid
.hash
, FSYNC_COMPONENT_INDEX
,
3088 CSUM_HASH_IN_STREAM
| csum_fsync_flag
);
3090 if (close_tempfile_gently(tempfile
)) {
3091 error(_("could not close '%s'"), get_tempfile_path(tempfile
));
3094 if (stat(get_tempfile_path(tempfile
), &st
))
3096 istate
->timestamp
.sec
= (unsigned int)st
.st_mtime
;
3097 istate
->timestamp
.nsec
= ST_MTIME_NSEC(st
);
3098 trace_performance_since(start
, "write index, changed mask = %x", istate
->cache_changed
);
3101 * TODO trace2: replace "the_repository" with the actual repo instance
3102 * that is associated with the given "istate".
3104 trace2_data_intmax("index", the_repository
, "write/version",
3106 trace2_data_intmax("index", the_repository
, "write/cache_nr",
3112 void set_alternate_index_output(const char *name
)
3114 alternate_index_output
= name
;
3117 static int commit_locked_index(struct lock_file
*lk
)
3119 if (alternate_index_output
)
3120 return commit_lock_file_to(lk
, alternate_index_output
);
3122 return commit_lock_file(lk
);
3125 static int do_write_locked_index(struct index_state
*istate
,
3126 struct lock_file
*lock
,
3128 enum write_extensions write_extensions
)
3131 int was_full
= istate
->sparse_index
== INDEX_EXPANDED
;
3133 ret
= convert_to_sparse(istate
, 0);
3136 warning(_("failed to convert to a sparse-index"));
3141 * TODO trace2: replace "the_repository" with the actual repo instance
3142 * that is associated with the given "istate".
3144 trace2_region_enter_printf("index", "do_write_index", the_repository
,
3145 "%s", get_lock_file_path(lock
));
3146 ret
= do_write_index(istate
, lock
->tempfile
, write_extensions
, flags
);
3147 trace2_region_leave_printf("index", "do_write_index", the_repository
,
3148 "%s", get_lock_file_path(lock
));
3151 ensure_full_index(istate
);
3155 if (flags
& COMMIT_LOCK
)
3156 ret
= commit_locked_index(lock
);
3158 ret
= close_lock_file_gently(lock
);
3160 run_hooks_l("post-index-change",
3161 istate
->updated_workdir
? "1" : "0",
3162 istate
->updated_skipworktree
? "1" : "0", NULL
);
3163 istate
->updated_workdir
= 0;
3164 istate
->updated_skipworktree
= 0;
3169 static int write_split_index(struct index_state
*istate
,
3170 struct lock_file
*lock
,
3174 prepare_to_write_split_index(istate
);
3175 ret
= do_write_locked_index(istate
, lock
, flags
, WRITE_ALL_EXTENSIONS
);
3176 finish_writing_split_index(istate
);
3180 static const char *shared_index_expire
= "2.weeks.ago";
3182 static unsigned long get_shared_index_expire_date(void)
3184 static unsigned long shared_index_expire_date
;
3185 static int shared_index_expire_date_prepared
;
3187 if (!shared_index_expire_date_prepared
) {
3188 git_config_get_expiry("splitindex.sharedindexexpire",
3189 &shared_index_expire
);
3190 shared_index_expire_date
= approxidate(shared_index_expire
);
3191 shared_index_expire_date_prepared
= 1;
3194 return shared_index_expire_date
;
3197 static int should_delete_shared_index(const char *shared_index_path
)
3200 unsigned long expiration
;
3202 /* Check timestamp */
3203 expiration
= get_shared_index_expire_date();
3206 if (stat(shared_index_path
, &st
))
3207 return error_errno(_("could not stat '%s'"), shared_index_path
);
3208 if (st
.st_mtime
> expiration
)
3214 static int clean_shared_index_files(const char *current_hex
)
3217 DIR *dir
= opendir(get_git_dir());
3220 return error_errno(_("unable to open git dir: %s"), get_git_dir());
3222 while ((de
= readdir(dir
)) != NULL
) {
3223 const char *sha1_hex
;
3224 const char *shared_index_path
;
3225 if (!skip_prefix(de
->d_name
, "sharedindex.", &sha1_hex
))
3227 if (!strcmp(sha1_hex
, current_hex
))
3229 shared_index_path
= git_path("%s", de
->d_name
);
3230 if (should_delete_shared_index(shared_index_path
) > 0 &&
3231 unlink(shared_index_path
))
3232 warning_errno(_("unable to unlink: %s"), shared_index_path
);
3239 static int write_shared_index(struct index_state
*istate
,
3240 struct tempfile
**temp
, unsigned flags
)
3242 struct split_index
*si
= istate
->split_index
;
3243 int ret
, was_full
= !istate
->sparse_index
;
3245 move_cache_to_base_index(istate
);
3246 convert_to_sparse(istate
, 0);
3248 trace2_region_enter_printf("index", "shared/do_write_index",
3249 the_repository
, "%s", get_tempfile_path(*temp
));
3250 ret
= do_write_index(si
->base
, *temp
, WRITE_NO_EXTENSION
, flags
);
3251 trace2_region_leave_printf("index", "shared/do_write_index",
3252 the_repository
, "%s", get_tempfile_path(*temp
));
3255 ensure_full_index(istate
);
3259 ret
= adjust_shared_perm(get_tempfile_path(*temp
));
3261 error(_("cannot fix permission bits on '%s'"), get_tempfile_path(*temp
));
3264 ret
= rename_tempfile(temp
,
3265 git_path("sharedindex.%s", oid_to_hex(&si
->base
->oid
)));
3267 oidcpy(&si
->base_oid
, &si
->base
->oid
);
3268 clean_shared_index_files(oid_to_hex(&si
->base
->oid
));
3274 static const int default_max_percent_split_change
= 20;
3276 static int too_many_not_shared_entries(struct index_state
*istate
)
3278 int i
, not_shared
= 0;
3279 int max_split
= git_config_get_max_percent_split_change();
3281 switch (max_split
) {
3283 /* not or badly configured: use the default value */
3284 max_split
= default_max_percent_split_change
;
3287 return 1; /* 0% means always write a new shared index */
3289 return 0; /* 100% means never write a new shared index */
3291 break; /* just use the configured value */
3294 /* Count not shared entries */
3295 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3296 struct cache_entry
*ce
= istate
->cache
[i
];
3301 return (int64_t)istate
->cache_nr
* max_split
< (int64_t)not_shared
* 100;
3304 int write_locked_index(struct index_state
*istate
, struct lock_file
*lock
,
3307 int new_shared_index
, ret
, test_split_index_env
;
3308 struct split_index
*si
= istate
->split_index
;
3310 if (git_env_bool("GIT_TEST_CHECK_CACHE_TREE", 0))
3311 cache_tree_verify(the_repository
, istate
);
3313 if ((flags
& SKIP_IF_UNCHANGED
) && !istate
->cache_changed
) {
3314 if (flags
& COMMIT_LOCK
)
3315 rollback_lock_file(lock
);
3319 if (istate
->fsmonitor_last_update
)
3320 fill_fsmonitor_bitmap(istate
);
3322 test_split_index_env
= git_env_bool("GIT_TEST_SPLIT_INDEX", 0);
3324 if ((!si
&& !test_split_index_env
) ||
3325 alternate_index_output
||
3326 (istate
->cache_changed
& ~EXTMASK
)) {
3327 ret
= do_write_locked_index(istate
, lock
, flags
,
3328 ~WRITE_SPLIT_INDEX_EXTENSION
);
3332 if (test_split_index_env
) {
3334 si
= init_split_index(istate
);
3335 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3337 int v
= si
->base_oid
.hash
[0];
3339 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3342 if (too_many_not_shared_entries(istate
))
3343 istate
->cache_changed
|= SPLIT_INDEX_ORDERED
;
3345 new_shared_index
= istate
->cache_changed
& SPLIT_INDEX_ORDERED
;
3347 if (new_shared_index
) {
3348 struct tempfile
*temp
;
3351 /* Same initial permissions as the main .git/index file */
3352 temp
= mks_tempfile_sm(git_path("sharedindex_XXXXXX"), 0, 0666);
3354 ret
= do_write_locked_index(istate
, lock
, flags
,
3355 ~WRITE_SPLIT_INDEX_EXTENSION
);
3358 ret
= write_shared_index(istate
, &temp
, flags
);
3360 saved_errno
= errno
;
3361 if (is_tempfile_active(temp
))
3362 delete_tempfile(&temp
);
3363 errno
= saved_errno
;
3369 ret
= write_split_index(istate
, lock
, flags
);
3371 /* Freshen the shared index only if the split-index was written */
3372 if (!ret
&& !new_shared_index
&& !is_null_oid(&si
->base_oid
)) {
3373 const char *shared_index
= git_path("sharedindex.%s",
3374 oid_to_hex(&si
->base_oid
));
3375 freshen_shared_index(shared_index
, 1);
3379 if (flags
& COMMIT_LOCK
)
3380 rollback_lock_file(lock
);
3385 * Read the index file that is potentially unmerged into given
3386 * index_state, dropping any unmerged entries to stage #0 (potentially
3387 * resulting in a path appearing as both a file and a directory in the
3388 * index; the caller is responsible to clear out the extra entries
3389 * before writing the index to a tree). Returns true if the index is
3390 * unmerged. Callers who want to refuse to work from an unmerged
3391 * state can call this and check its return value, instead of calling
3394 int repo_read_index_unmerged(struct repository
*repo
)
3396 struct index_state
*istate
;
3400 repo_read_index(repo
);
3401 istate
= repo
->index
;
3402 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3403 struct cache_entry
*ce
= istate
->cache
[i
];
3404 struct cache_entry
*new_ce
;
3410 len
= ce_namelen(ce
);
3411 new_ce
= make_empty_cache_entry(istate
, len
);
3412 memcpy(new_ce
->name
, ce
->name
, len
);
3413 new_ce
->ce_flags
= create_ce_flags(0) | CE_CONFLICTED
;
3414 new_ce
->ce_namelen
= len
;
3415 new_ce
->ce_mode
= ce
->ce_mode
;
3416 if (add_index_entry(istate
, new_ce
, ADD_CACHE_SKIP_DFCHECK
))
3417 return error(_("%s: cannot drop to stage #0"),
3424 * Returns 1 if the path is an "other" path with respect to
3425 * the index; that is, the path is not mentioned in the index at all,
3426 * either as a file, a directory with some files in the index,
3427 * or as an unmerged entry.
3429 * We helpfully remove a trailing "/" from directories so that
3430 * the output of read_directory can be used as-is.
3432 int index_name_is_other(struct index_state
*istate
, const char *name
,
3436 if (namelen
&& name
[namelen
- 1] == '/')
3438 pos
= index_name_pos(istate
, name
, namelen
);
3440 return 0; /* exact match */
3442 if (pos
< istate
->cache_nr
) {
3443 struct cache_entry
*ce
= istate
->cache
[pos
];
3444 if (ce_namelen(ce
) == namelen
&&
3445 !memcmp(ce
->name
, name
, namelen
))
3446 return 0; /* Yup, this one exists unmerged */
3451 void *read_blob_data_from_index(struct index_state
*istate
,
3452 const char *path
, unsigned long *size
)
3456 enum object_type type
;
3460 pos
= index_name_pos(istate
, path
, len
);
3463 * We might be in the middle of a merge, in which
3464 * case we would read stage #2 (ours).
3468 (pos
< 0 && i
< istate
->cache_nr
&&
3469 !strcmp(istate
->cache
[i
]->name
, path
));
3471 if (ce_stage(istate
->cache
[i
]) == 2)
3476 data
= repo_read_object_file(the_repository
, &istate
->cache
[pos
]->oid
,
3478 if (!data
|| type
!= OBJ_BLOB
) {
3487 void move_index_extensions(struct index_state
*dst
, struct index_state
*src
)
3489 dst
->untracked
= src
->untracked
;
3490 src
->untracked
= NULL
;
3491 dst
->cache_tree
= src
->cache_tree
;
3492 src
->cache_tree
= NULL
;
3495 struct cache_entry
*dup_cache_entry(const struct cache_entry
*ce
,
3496 struct index_state
*istate
)
3498 unsigned int size
= ce_size(ce
);
3499 int mem_pool_allocated
;
3500 struct cache_entry
*new_entry
= make_empty_cache_entry(istate
, ce_namelen(ce
));
3501 mem_pool_allocated
= new_entry
->mem_pool_allocated
;
3503 memcpy(new_entry
, ce
, size
);
3504 new_entry
->mem_pool_allocated
= mem_pool_allocated
;
3508 void discard_cache_entry(struct cache_entry
*ce
)
3510 if (ce
&& should_validate_cache_entries())
3511 memset(ce
, 0xCD, cache_entry_size(ce
->ce_namelen
));
3513 if (ce
&& ce
->mem_pool_allocated
)
3519 int should_validate_cache_entries(void)
3521 static int validate_index_cache_entries
= -1;
3523 if (validate_index_cache_entries
< 0) {
3524 if (getenv("GIT_TEST_VALIDATE_INDEX_CACHE_ENTRIES"))
3525 validate_index_cache_entries
= 1;
3527 validate_index_cache_entries
= 0;
3530 return validate_index_cache_entries
;
3533 #define EOIE_SIZE (4 + GIT_SHA1_RAWSZ) /* <4-byte offset> + <20-byte hash> */
3534 #define EOIE_SIZE_WITH_HEADER (4 + 4 + EOIE_SIZE) /* <4-byte signature> + <4-byte length> + EOIE_SIZE */
3536 static size_t read_eoie_extension(const char *mmap
, size_t mmap_size
)
3539 * The end of index entries (EOIE) extension is guaranteed to be last
3540 * so that it can be found by scanning backwards from the EOF.
3547 const char *index
, *eoie
;
3549 size_t offset
, src_offset
;
3550 unsigned char hash
[GIT_MAX_RAWSZ
];
3553 /* ensure we have an index big enough to contain an EOIE extension */
3554 if (mmap_size
< sizeof(struct cache_header
) + EOIE_SIZE_WITH_HEADER
+ the_hash_algo
->rawsz
)
3557 /* validate the extension signature */
3558 index
= eoie
= mmap
+ mmap_size
- EOIE_SIZE_WITH_HEADER
- the_hash_algo
->rawsz
;
3559 if (CACHE_EXT(index
) != CACHE_EXT_ENDOFINDEXENTRIES
)
3561 index
+= sizeof(uint32_t);
3563 /* validate the extension size */
3564 extsize
= get_be32(index
);
3565 if (extsize
!= EOIE_SIZE
)
3567 index
+= sizeof(uint32_t);
3570 * Validate the offset we're going to look for the first extension
3571 * signature is after the index header and before the eoie extension.
3573 offset
= get_be32(index
);
3574 if (mmap
+ offset
< mmap
+ sizeof(struct cache_header
))
3576 if (mmap
+ offset
>= eoie
)
3578 index
+= sizeof(uint32_t);
3581 * The hash is computed over extension types and their sizes (but not
3582 * their contents). E.g. if we have "TREE" extension that is N-bytes
3583 * long, "REUC" extension that is M-bytes long, followed by "EOIE",
3584 * then the hash would be:
3586 * SHA-1("TREE" + <binary representation of N> +
3587 * "REUC" + <binary representation of M>)
3589 src_offset
= offset
;
3590 the_hash_algo
->init_fn(&c
);
3591 while (src_offset
< mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
) {
3592 /* After an array of active_nr index entries,
3593 * there can be arbitrary number of extended
3594 * sections, each of which is prefixed with
3595 * extension name (4-byte) and section length
3596 * in 4-byte network byte order.
3599 memcpy(&extsize
, mmap
+ src_offset
+ 4, 4);
3600 extsize
= ntohl(extsize
);
3602 /* verify the extension size isn't so large it will wrap around */
3603 if (src_offset
+ 8 + extsize
< src_offset
)
3606 the_hash_algo
->update_fn(&c
, mmap
+ src_offset
, 8);
3609 src_offset
+= extsize
;
3611 the_hash_algo
->final_fn(hash
, &c
);
3612 if (!hasheq(hash
, (const unsigned char *)index
))
3615 /* Validate that the extension offsets returned us back to the eoie extension. */
3616 if (src_offset
!= mmap_size
- the_hash_algo
->rawsz
- EOIE_SIZE_WITH_HEADER
)
3622 static void write_eoie_extension(struct strbuf
*sb
, git_hash_ctx
*eoie_context
, size_t offset
)
3625 unsigned char hash
[GIT_MAX_RAWSZ
];
3628 put_be32(&buffer
, offset
);
3629 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3632 the_hash_algo
->final_fn(hash
, eoie_context
);
3633 strbuf_add(sb
, hash
, the_hash_algo
->rawsz
);
3636 #define IEOT_VERSION (1)
3638 static struct index_entry_offset_table
*read_ieot_extension(const char *mmap
, size_t mmap_size
, size_t offset
)
3640 const char *index
= NULL
;
3641 uint32_t extsize
, ext_version
;
3642 struct index_entry_offset_table
*ieot
;
3645 /* find the IEOT extension */
3648 while (offset
<= mmap_size
- the_hash_algo
->rawsz
- 8) {
3649 extsize
= get_be32(mmap
+ offset
+ 4);
3650 if (CACHE_EXT((mmap
+ offset
)) == CACHE_EXT_INDEXENTRYOFFSETTABLE
) {
3651 index
= mmap
+ offset
+ 4 + 4;
3660 /* validate the version is IEOT_VERSION */
3661 ext_version
= get_be32(index
);
3662 if (ext_version
!= IEOT_VERSION
) {
3663 error("invalid IEOT version %d", ext_version
);
3666 index
+= sizeof(uint32_t);
3668 /* extension size - version bytes / bytes per entry */
3669 nr
= (extsize
- sizeof(uint32_t)) / (sizeof(uint32_t) + sizeof(uint32_t));
3671 error("invalid number of IEOT entries %d", nr
);
3674 ieot
= xmalloc(sizeof(struct index_entry_offset_table
)
3675 + (nr
* sizeof(struct index_entry_offset
)));
3677 for (i
= 0; i
< nr
; i
++) {
3678 ieot
->entries
[i
].offset
= get_be32(index
);
3679 index
+= sizeof(uint32_t);
3680 ieot
->entries
[i
].nr
= get_be32(index
);
3681 index
+= sizeof(uint32_t);
3687 static void write_ieot_extension(struct strbuf
*sb
, struct index_entry_offset_table
*ieot
)
3693 put_be32(&buffer
, IEOT_VERSION
);
3694 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3697 for (i
= 0; i
< ieot
->nr
; i
++) {
3700 put_be32(&buffer
, ieot
->entries
[i
].offset
);
3701 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3704 put_be32(&buffer
, ieot
->entries
[i
].nr
);
3705 strbuf_add(sb
, &buffer
, sizeof(uint32_t));
3709 void prefetch_cache_entries(const struct index_state
*istate
,
3710 must_prefetch_predicate must_prefetch
)
3713 struct oid_array to_fetch
= OID_ARRAY_INIT
;
3715 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3716 struct cache_entry
*ce
= istate
->cache
[i
];
3718 if (S_ISGITLINK(ce
->ce_mode
) || !must_prefetch(ce
))
3720 if (!oid_object_info_extended(the_repository
, &ce
->oid
,
3722 OBJECT_INFO_FOR_PREFETCH
))
3724 oid_array_append(&to_fetch
, &ce
->oid
);
3726 promisor_remote_get_direct(the_repository
,
3727 to_fetch
.oid
, to_fetch
.nr
);
3728 oid_array_clear(&to_fetch
);
3731 static int read_one_entry_opt(struct index_state
*istate
,
3732 const struct object_id
*oid
,
3733 struct strbuf
*base
,
3734 const char *pathname
,
3735 unsigned mode
, int opt
)
3738 struct cache_entry
*ce
;
3741 return READ_TREE_RECURSIVE
;
3743 len
= strlen(pathname
);
3744 ce
= make_empty_cache_entry(istate
, base
->len
+ len
);
3746 ce
->ce_mode
= create_ce_mode(mode
);
3747 ce
->ce_flags
= create_ce_flags(1);
3748 ce
->ce_namelen
= base
->len
+ len
;
3749 memcpy(ce
->name
, base
->buf
, base
->len
);
3750 memcpy(ce
->name
+ base
->len
, pathname
, len
+1);
3751 oidcpy(&ce
->oid
, oid
);
3752 return add_index_entry(istate
, ce
, opt
);
3755 static int read_one_entry(const struct object_id
*oid
, struct strbuf
*base
,
3756 const char *pathname
, unsigned mode
,
3759 struct index_state
*istate
= context
;
3760 return read_one_entry_opt(istate
, oid
, base
, pathname
,
3762 ADD_CACHE_OK_TO_ADD
|ADD_CACHE_SKIP_DFCHECK
);
3766 * This is used when the caller knows there is no existing entries at
3767 * the stage that will conflict with the entry being added.
3769 static int read_one_entry_quick(const struct object_id
*oid
, struct strbuf
*base
,
3770 const char *pathname
, unsigned mode
,
3773 struct index_state
*istate
= context
;
3774 return read_one_entry_opt(istate
, oid
, base
, pathname
,
3775 mode
, ADD_CACHE_JUST_APPEND
);
3779 * Read the tree specified with --with-tree option
3780 * (typically, HEAD) into stage #1 and then
3781 * squash them down to stage #0. This is used for
3782 * --error-unmatch to list and check the path patterns
3783 * that were given from the command line. We are not
3784 * going to write this index out.
3786 void overlay_tree_on_index(struct index_state
*istate
,
3787 const char *tree_name
, const char *prefix
)
3790 struct object_id oid
;
3791 struct pathspec pathspec
;
3792 struct cache_entry
*last_stage0
= NULL
;
3794 read_tree_fn_t fn
= NULL
;
3797 if (repo_get_oid(the_repository
, tree_name
, &oid
))
3798 die("tree-ish %s not found.", tree_name
);
3799 tree
= parse_tree_indirect(&oid
);
3801 die("bad tree-ish %s", tree_name
);
3803 /* Hoist the unmerged entries up to stage #3 to make room */
3804 /* TODO: audit for interaction with sparse-index. */
3805 ensure_full_index(istate
);
3806 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3807 struct cache_entry
*ce
= istate
->cache
[i
];
3810 ce
->ce_flags
|= CE_STAGEMASK
;
3814 static const char *(matchbuf
[1]);
3816 parse_pathspec(&pathspec
, PATHSPEC_ALL_MAGIC
,
3817 PATHSPEC_PREFER_CWD
, prefix
, matchbuf
);
3819 memset(&pathspec
, 0, sizeof(pathspec
));
3822 * See if we have cache entry at the stage. If so,
3823 * do it the original slow way, otherwise, append and then
3826 for (i
= 0; !fn
&& i
< istate
->cache_nr
; i
++) {
3827 const struct cache_entry
*ce
= istate
->cache
[i
];
3828 if (ce_stage(ce
) == 1)
3829 fn
= read_one_entry
;
3833 fn
= read_one_entry_quick
;
3834 err
= read_tree(the_repository
, tree
, &pathspec
, fn
, istate
);
3835 clear_pathspec(&pathspec
);
3837 die("unable to read tree entries %s", tree_name
);
3840 * Sort the cache entry -- we need to nuke the cache tree, though.
3842 if (fn
== read_one_entry_quick
) {
3843 cache_tree_free(&istate
->cache_tree
);
3844 QSORT(istate
->cache
, istate
->cache_nr
, cmp_cache_name_compare
);
3847 for (i
= 0; i
< istate
->cache_nr
; i
++) {
3848 struct cache_entry
*ce
= istate
->cache
[i
];
3849 switch (ce_stage(ce
)) {
3857 * If there is stage #0 entry for this, we do not
3858 * need to show it. We use CE_UPDATE bit to mark
3862 !strcmp(last_stage0
->name
, ce
->name
))
3863 ce
->ce_flags
|= CE_UPDATE
;
3868 struct update_callback_data
{
3869 struct index_state
*index
;
3875 static int fix_unmerged_status(struct diff_filepair
*p
,
3876 struct update_callback_data
*data
)
3878 if (p
->status
!= DIFF_STATUS_UNMERGED
)
3880 if (!(data
->flags
& ADD_CACHE_IGNORE_REMOVAL
) && !p
->two
->mode
)
3882 * This is not an explicit add request, and the
3883 * path is missing from the working tree (deleted)
3885 return DIFF_STATUS_DELETED
;
3888 * Either an explicit add request, or path exists
3889 * in the working tree. An attempt to explicitly
3890 * add a path that does not exist in the working tree
3891 * will be caught as an error by the caller immediately.
3893 return DIFF_STATUS_MODIFIED
;
3896 static void update_callback(struct diff_queue_struct
*q
,
3897 struct diff_options
*opt UNUSED
, void *cbdata
)
3900 struct update_callback_data
*data
= cbdata
;
3902 for (i
= 0; i
< q
->nr
; i
++) {
3903 struct diff_filepair
*p
= q
->queue
[i
];
3904 const char *path
= p
->one
->path
;
3906 if (!data
->include_sparse
&&
3907 !path_in_sparse_checkout(path
, data
->index
))
3910 switch (fix_unmerged_status(p
, data
)) {
3912 die(_("unexpected diff status %c"), p
->status
);
3913 case DIFF_STATUS_MODIFIED
:
3914 case DIFF_STATUS_TYPE_CHANGED
:
3915 if (add_file_to_index(data
->index
, path
, data
->flags
)) {
3916 if (!(data
->flags
& ADD_CACHE_IGNORE_ERRORS
))
3917 die(_("updating files failed"));
3921 case DIFF_STATUS_DELETED
:
3922 if (data
->flags
& ADD_CACHE_IGNORE_REMOVAL
)
3924 if (!(data
->flags
& ADD_CACHE_PRETEND
))
3925 remove_file_from_index(data
->index
, path
);
3926 if (data
->flags
& (ADD_CACHE_PRETEND
|ADD_CACHE_VERBOSE
))
3927 printf(_("remove '%s'\n"), path
);
3933 int add_files_to_cache(struct repository
*repo
, const char *prefix
,
3934 const struct pathspec
*pathspec
, int include_sparse
,
3937 struct update_callback_data data
;
3938 struct rev_info rev
;
3940 memset(&data
, 0, sizeof(data
));
3941 data
.index
= repo
->index
;
3942 data
.include_sparse
= include_sparse
;
3945 repo_init_revisions(repo
, &rev
, prefix
);
3946 setup_revisions(0, NULL
, &rev
, NULL
);
3948 copy_pathspec(&rev
.prune_data
, pathspec
);
3949 rev
.diffopt
.output_format
= DIFF_FORMAT_CALLBACK
;
3950 rev
.diffopt
.format_callback
= update_callback
;
3951 rev
.diffopt
.format_callback_data
= &data
;
3952 rev
.diffopt
.flags
.override_submodule_config
= 1;
3953 rev
.max_count
= 0; /* do not compare unmerged paths with stage #2 */
3956 * Use an ODB transaction to optimize adding multiple objects.
3957 * This function is invoked from commands other than 'add', which
3958 * may not have their own transaction active.
3960 begin_odb_transaction();
3961 run_diff_files(&rev
, DIFF_RACY_IS_MODIFIED
);
3962 end_odb_transaction();
3964 release_revisions(&rev
);
3965 return !!data
.add_errors
;