2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
11 #include "cache-tree.h"
17 #include "resolve-undo.h"
20 #include "split-index.h"
22 #include "fsmonitor.h"
24 /* Mask for the name length in ce_flags in the on-disk index */
26 #define CE_NAMEMASK (0x0fff)
30 * The first letter should be 'A'..'Z' for extensions that are not
31 * necessary for a correct operation (i.e. optimization data).
32 * When new extensions are added that _needs_ to be understood in
33 * order to correctly interpret the index file, pick character that
34 * is outside the range, to cause the reader to abort.
37 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
38 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
39 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
40 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
41 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
42 #define CACHE_EXT_FSMONITOR 0x46534D4E /* "FSMN" */
44 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
45 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
46 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
47 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED | FSMONITOR_CHANGED)
49 struct index_state the_index;
50 static const char *alternate_index_output;
52 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
54 istate->cache[nr] = ce;
55 add_name_hash(istate, ce);
58 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
60 struct cache_entry *old = istate->cache[nr];
62 replace_index_entry_in_base(istate, old, ce);
63 remove_name_hash(istate, old);
65 set_index_entry(istate, nr, ce);
66 ce->ce_flags |= CE_UPDATE_IN_BASE;
67 mark_fsmonitor_invalid(istate, ce);
68 istate->cache_changed |= CE_ENTRY_CHANGED;
71 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
73 struct cache_entry *old = istate->cache[nr], *new;
74 int namelen = strlen(new_name);
76 new = xmalloc(cache_entry_size(namelen));
77 copy_cache_entry(new, old);
78 new->ce_flags &= ~CE_HASHED;
79 new->ce_namelen = namelen;
81 memcpy(new->name, new_name, namelen + 1);
83 cache_tree_invalidate_path(istate, old->name);
84 untracked_cache_remove_from_index(istate, old->name);
85 remove_index_entry_at(istate, nr);
86 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
89 void fill_stat_data(struct stat_data *sd, struct stat *st)
91 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
92 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
93 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
94 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
95 sd->sd_dev = st->st_dev;
96 sd->sd_ino = st->st_ino;
97 sd->sd_uid = st->st_uid;
98 sd->sd_gid = st->st_gid;
99 sd->sd_size = st->st_size;
102 int match_stat_data(const struct stat_data *sd, struct stat *st)
106 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
107 changed |= MTIME_CHANGED;
108 if (trust_ctime && check_stat &&
109 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
110 changed |= CTIME_CHANGED;
113 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
114 changed |= MTIME_CHANGED;
115 if (trust_ctime && check_stat &&
116 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
117 changed |= CTIME_CHANGED;
121 if (sd->sd_uid != (unsigned int) st->st_uid ||
122 sd->sd_gid != (unsigned int) st->st_gid)
123 changed |= OWNER_CHANGED;
124 if (sd->sd_ino != (unsigned int) st->st_ino)
125 changed |= INODE_CHANGED;
130 * st_dev breaks on network filesystems where different
131 * clients will have different views of what "device"
132 * the filesystem is on
134 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
135 changed |= INODE_CHANGED;
138 if (sd->sd_size != (unsigned int) st->st_size)
139 changed |= DATA_CHANGED;
145 * This only updates the "non-critical" parts of the directory
146 * cache, ie the parts that aren't tracked by GIT, and only used
147 * to validate the cache.
149 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
151 fill_stat_data(&ce->ce_stat_data, st);
153 if (assume_unchanged)
154 ce->ce_flags |= CE_VALID;
156 if (S_ISREG(st->st_mode)) {
157 ce_mark_uptodate(ce);
158 mark_fsmonitor_valid(ce);
162 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
165 int fd = git_open_cloexec(ce->name, O_RDONLY);
168 struct object_id oid;
169 if (!index_fd(&oid, fd, st, OBJ_BLOB, ce->name, 0))
170 match = oidcmp(&oid, &ce->oid);
171 /* index_fd() closed the file descriptor already */
176 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
181 enum object_type type;
182 struct strbuf sb = STRBUF_INIT;
184 if (strbuf_readlink(&sb, ce->name, expected_size))
187 buffer = read_sha1_file(ce->oid.hash, &type, &size);
190 match = memcmp(buffer, sb.buf, size);
197 static int ce_compare_gitlink(const struct cache_entry *ce)
199 struct object_id oid;
202 * We don't actually require that the .git directory
203 * under GITLINK directory be a valid git directory. It
204 * might even be missing (in case nobody populated that
207 * If so, we consider it always to match.
209 if (resolve_gitlink_ref(ce->name, "HEAD", &oid) < 0)
211 return oidcmp(&oid, &ce->oid);
214 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
216 switch (st->st_mode & S_IFMT) {
218 if (ce_compare_data(ce, st))
222 if (ce_compare_link(ce, xsize_t(st->st_size)))
226 if (S_ISGITLINK(ce->ce_mode))
227 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
228 /* else fallthrough */
235 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
237 unsigned int changed = 0;
239 if (ce->ce_flags & CE_REMOVE)
240 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
242 switch (ce->ce_mode & S_IFMT) {
244 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
245 /* We consider only the owner x bit to be relevant for
248 if (trust_executable_bit &&
249 (0100 & (ce->ce_mode ^ st->st_mode)))
250 changed |= MODE_CHANGED;
253 if (!S_ISLNK(st->st_mode) &&
254 (has_symlinks || !S_ISREG(st->st_mode)))
255 changed |= TYPE_CHANGED;
258 /* We ignore most of the st_xxx fields for gitlinks */
259 if (!S_ISDIR(st->st_mode))
260 changed |= TYPE_CHANGED;
261 else if (ce_compare_gitlink(ce))
262 changed |= DATA_CHANGED;
265 die("internal error: ce_mode is %o", ce->ce_mode);
268 changed |= match_stat_data(&ce->ce_stat_data, st);
270 /* Racily smudged entry? */
271 if (!ce->ce_stat_data.sd_size) {
272 if (!is_empty_blob_sha1(ce->oid.hash))
273 changed |= DATA_CHANGED;
279 static int is_racy_stat(const struct index_state *istate,
280 const struct stat_data *sd)
282 return (istate->timestamp.sec &&
284 /* nanosecond timestamped files can also be racy! */
285 (istate->timestamp.sec < sd->sd_mtime.sec ||
286 (istate->timestamp.sec == sd->sd_mtime.sec &&
287 istate->timestamp.nsec <= sd->sd_mtime.nsec))
289 istate->timestamp.sec <= sd->sd_mtime.sec
294 static int is_racy_timestamp(const struct index_state *istate,
295 const struct cache_entry *ce)
297 return (!S_ISGITLINK(ce->ce_mode) &&
298 is_racy_stat(istate, &ce->ce_stat_data));
301 int match_stat_data_racy(const struct index_state *istate,
302 const struct stat_data *sd, struct stat *st)
304 if (is_racy_stat(istate, sd))
305 return MTIME_CHANGED;
306 return match_stat_data(sd, st);
309 int ie_match_stat(struct index_state *istate,
310 const struct cache_entry *ce, struct stat *st,
311 unsigned int options)
313 unsigned int changed;
314 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
315 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
316 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
317 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
319 if (!ignore_fsmonitor)
320 refresh_fsmonitor(istate);
322 * If it's marked as always valid in the index, it's
323 * valid whatever the checked-out copy says.
325 * skip-worktree has the same effect with higher precedence
327 if (!ignore_skip_worktree && ce_skip_worktree(ce))
329 if (!ignore_valid && (ce->ce_flags & CE_VALID))
331 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID))
335 * Intent-to-add entries have not been added, so the index entry
336 * by definition never matches what is in the work tree until it
337 * actually gets added.
339 if (ce_intent_to_add(ce))
340 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
342 changed = ce_match_stat_basic(ce, st);
345 * Within 1 second of this sequence:
346 * echo xyzzy >file && git-update-index --add file
347 * running this command:
349 * would give a falsely clean cache entry. The mtime and
350 * length match the cache, and other stat fields do not change.
352 * We could detect this at update-index time (the cache entry
353 * being registered/updated records the same time as "now")
354 * and delay the return from git-update-index, but that would
355 * effectively mean we can make at most one commit per second,
356 * which is not acceptable. Instead, we check cache entries
357 * whose mtime are the same as the index file timestamp more
358 * carefully than others.
360 if (!changed && is_racy_timestamp(istate, ce)) {
361 if (assume_racy_is_modified)
362 changed |= DATA_CHANGED;
364 changed |= ce_modified_check_fs(ce, st);
370 int ie_modified(struct index_state *istate,
371 const struct cache_entry *ce,
372 struct stat *st, unsigned int options)
374 int changed, changed_fs;
376 changed = ie_match_stat(istate, ce, st, options);
380 * If the mode or type has changed, there's no point in trying
381 * to refresh the entry - it's not going to match
383 if (changed & (MODE_CHANGED | TYPE_CHANGED))
387 * Immediately after read-tree or update-index --cacheinfo,
388 * the length field is zero, as we have never even read the
389 * lstat(2) information once, and we cannot trust DATA_CHANGED
390 * returned by ie_match_stat() which in turn was returned by
391 * ce_match_stat_basic() to signal that the filesize of the
392 * blob changed. We have to actually go to the filesystem to
393 * see if the contents match, and if so, should answer "unchanged".
395 * The logic does not apply to gitlinks, as ce_match_stat_basic()
396 * already has checked the actual HEAD from the filesystem in the
397 * subproject. If ie_match_stat() already said it is different,
398 * then we know it is.
400 if ((changed & DATA_CHANGED) &&
401 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
404 changed_fs = ce_modified_check_fs(ce, st);
406 return changed | changed_fs;
410 int base_name_compare(const char *name1, int len1, int mode1,
411 const char *name2, int len2, int mode2)
413 unsigned char c1, c2;
414 int len = len1 < len2 ? len1 : len2;
417 cmp = memcmp(name1, name2, len);
422 if (!c1 && S_ISDIR(mode1))
424 if (!c2 && S_ISDIR(mode2))
426 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
430 * df_name_compare() is identical to base_name_compare(), except it
431 * compares conflicting directory/file entries as equal. Note that
432 * while a directory name compares as equal to a regular file, they
433 * then individually compare _differently_ to a filename that has
434 * a dot after the basename (because '\0' < '.' < '/').
436 * This is used by routines that want to traverse the git namespace
437 * but then handle conflicting entries together when possible.
439 int df_name_compare(const char *name1, int len1, int mode1,
440 const char *name2, int len2, int mode2)
442 int len = len1 < len2 ? len1 : len2, cmp;
443 unsigned char c1, c2;
445 cmp = memcmp(name1, name2, len);
448 /* Directories and files compare equal (same length, same name) */
452 if (!c1 && S_ISDIR(mode1))
455 if (!c2 && S_ISDIR(mode2))
457 if (c1 == '/' && !c2)
459 if (c2 == '/' && !c1)
464 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
466 size_t min_len = (len1 < len2) ? len1 : len2;
467 int cmp = memcmp(name1, name2, min_len);
477 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
481 cmp = name_compare(name1, len1, name2, len2);
492 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
497 last = istate->cache_nr;
498 while (last > first) {
499 int next = (last + first) >> 1;
500 struct cache_entry *ce = istate->cache[next];
501 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
513 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
515 return index_name_stage_pos(istate, name, namelen, 0);
518 int remove_index_entry_at(struct index_state *istate, int pos)
520 struct cache_entry *ce = istate->cache[pos];
522 record_resolve_undo(istate, ce);
523 remove_name_hash(istate, ce);
524 save_or_free_index_entry(istate, ce);
525 istate->cache_changed |= CE_ENTRY_REMOVED;
527 if (pos >= istate->cache_nr)
529 MOVE_ARRAY(istate->cache + pos, istate->cache + pos + 1,
530 istate->cache_nr - pos);
535 * Remove all cache entries marked for removal, that is where
536 * CE_REMOVE is set in ce_flags. This is much more effective than
537 * calling remove_index_entry_at() for each entry to be removed.
539 void remove_marked_cache_entries(struct index_state *istate)
541 struct cache_entry **ce_array = istate->cache;
544 for (i = j = 0; i < istate->cache_nr; i++) {
545 if (ce_array[i]->ce_flags & CE_REMOVE) {
546 remove_name_hash(istate, ce_array[i]);
547 save_or_free_index_entry(istate, ce_array[i]);
550 ce_array[j++] = ce_array[i];
552 if (j == istate->cache_nr)
554 istate->cache_changed |= CE_ENTRY_REMOVED;
555 istate->cache_nr = j;
558 int remove_file_from_index(struct index_state *istate, const char *path)
560 int pos = index_name_pos(istate, path, strlen(path));
563 cache_tree_invalidate_path(istate, path);
564 untracked_cache_remove_from_index(istate, path);
565 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
566 remove_index_entry_at(istate, pos);
570 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
572 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
575 static int index_name_pos_also_unmerged(struct index_state *istate,
576 const char *path, int namelen)
578 int pos = index_name_pos(istate, path, namelen);
579 struct cache_entry *ce;
584 /* maybe unmerged? */
586 if (pos >= istate->cache_nr ||
587 compare_name((ce = istate->cache[pos]), path, namelen))
590 /* order of preference: stage 2, 1, 3 */
591 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
592 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
593 !compare_name(ce, path, namelen))
598 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
600 int len = ce_namelen(ce);
601 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
605 * If we add a filename that aliases in the cache, we will use the
606 * name that we already have - but we don't want to update the same
607 * alias twice, because that implies that there were actually two
608 * different files with aliasing names!
610 * So we use the CE_ADDED flag to verify that the alias was an old
611 * one before we accept it as
613 static struct cache_entry *create_alias_ce(struct index_state *istate,
614 struct cache_entry *ce,
615 struct cache_entry *alias)
618 struct cache_entry *new;
620 if (alias->ce_flags & CE_ADDED)
621 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
623 /* Ok, create the new entry using the name of the existing alias */
624 len = ce_namelen(alias);
625 new = xcalloc(1, cache_entry_size(len));
626 memcpy(new->name, alias->name, len);
627 copy_cache_entry(new, ce);
628 save_or_free_index_entry(istate, ce);
632 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
634 unsigned char sha1[20];
635 if (write_sha1_file("", 0, blob_type, sha1))
636 die("cannot create an empty blob in the object database");
637 hashcpy(ce->oid.hash, sha1);
640 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
642 int size, namelen, was_same;
643 mode_t st_mode = st->st_mode;
644 struct cache_entry *ce, *alias = NULL;
645 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
646 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
647 int pretend = flags & ADD_CACHE_PRETEND;
648 int intent_only = flags & ADD_CACHE_INTENT;
649 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
650 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
651 int newflags = HASH_WRITE_OBJECT;
653 if (flags & HASH_RENORMALIZE)
654 newflags |= HASH_RENORMALIZE;
656 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
657 return error("%s: can only add regular files, symbolic links or git-directories", path);
659 namelen = strlen(path);
660 if (S_ISDIR(st_mode)) {
661 while (namelen && path[namelen-1] == '/')
664 size = cache_entry_size(namelen);
665 ce = xcalloc(1, size);
666 memcpy(ce->name, path, namelen);
667 ce->ce_namelen = namelen;
669 fill_stat_cache_info(ce, st);
671 ce->ce_flags |= CE_INTENT_TO_ADD;
674 if (trust_executable_bit && has_symlinks) {
675 ce->ce_mode = create_ce_mode(st_mode);
677 /* If there is an existing entry, pick the mode bits and type
678 * from it, otherwise assume unexecutable regular file.
680 struct cache_entry *ent;
681 int pos = index_name_pos_also_unmerged(istate, path, namelen);
683 ent = (0 <= pos) ? istate->cache[pos] : NULL;
684 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
687 /* When core.ignorecase=true, determine if a directory of the same name but differing
688 * case already exists within the Git repository. If it does, ensure the directory
689 * case of the file being added to the repository matches (is folded into) the existing
690 * entry's directory case.
693 adjust_dirname_case(istate, ce->name);
695 if (!(flags & HASH_RENORMALIZE)) {
696 alias = index_file_exists(istate, ce->name,
697 ce_namelen(ce), ignore_case);
700 !ie_match_stat(istate, alias, st, ce_option)) {
701 /* Nothing changed, really */
702 if (!S_ISGITLINK(alias->ce_mode))
703 ce_mark_uptodate(alias);
704 alias->ce_flags |= CE_ADDED;
711 if (index_path(&ce->oid, path, st, newflags)) {
713 return error("unable to index file %s", path);
716 set_object_name_for_intent_to_add_entry(ce);
718 if (ignore_case && alias && different_name(ce, alias))
719 ce = create_alias_ce(istate, ce, alias);
720 ce->ce_flags |= CE_ADDED;
722 /* It was suspected to be racily clean, but it turns out to be Ok */
725 !oidcmp(&alias->oid, &ce->oid) &&
726 ce->ce_mode == alias->ce_mode);
730 else if (add_index_entry(istate, ce, add_option)) {
732 return error("unable to add %s to index", path);
734 if (verbose && !was_same)
735 printf("add '%s'\n", path);
739 int add_file_to_index(struct index_state *istate, const char *path, int flags)
742 if (lstat(path, &st))
743 die_errno("unable to stat '%s'", path);
744 return add_to_index(istate, path, &st, flags);
747 struct cache_entry *make_cache_entry(unsigned int mode,
748 const unsigned char *sha1, const char *path, int stage,
749 unsigned int refresh_options)
752 struct cache_entry *ce, *ret;
754 if (!verify_path(path, mode)) {
755 error("Invalid path '%s'", path);
760 size = cache_entry_size(len);
761 ce = xcalloc(1, size);
763 hashcpy(ce->oid.hash, sha1);
764 memcpy(ce->name, path, len);
765 ce->ce_flags = create_ce_flags(stage);
766 ce->ce_namelen = len;
767 ce->ce_mode = create_ce_mode(mode);
769 ret = refresh_cache_entry(ce, refresh_options);
776 * Chmod an index entry with either +x or -x.
778 * Returns -1 if the chmod for the particular cache entry failed (if it's
779 * not a regular file), -2 if an invalid flip argument is passed in, 0
782 int chmod_index_entry(struct index_state *istate, struct cache_entry *ce,
785 if (!S_ISREG(ce->ce_mode))
792 ce->ce_mode &= ~0111;
797 cache_tree_invalidate_path(istate, ce->name);
798 ce->ce_flags |= CE_UPDATE_IN_BASE;
799 mark_fsmonitor_invalid(istate, ce);
800 istate->cache_changed |= CE_ENTRY_CHANGED;
805 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
807 int len = ce_namelen(a);
808 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
812 * We fundamentally don't like some paths: we don't want
813 * dot or dot-dot anywhere, and for obvious reasons don't
814 * want to recurse into ".git" either.
816 * Also, we don't want double slashes or slashes at the
817 * end that can make pathnames ambiguous.
819 static int verify_dotfile(const char *rest, unsigned mode)
822 * The first character was '.', but that
823 * has already been discarded, we now test
827 /* "." is not allowed */
828 if (*rest == '\0' || is_dir_sep(*rest))
833 * ".git" followed by NUL or slash is bad. Note that we match
834 * case-insensitively here, even if ignore_case is not set.
835 * This outlaws ".GIT" everywhere out of an abundance of caution,
836 * since there's really no good reason to allow it.
838 * Once we've seen ".git", we can also find ".gitmodules", etc (also
839 * case-insensitively).
843 if (rest[1] != 'i' && rest[1] != 'I')
845 if (rest[2] != 't' && rest[2] != 'T')
847 if (rest[3] == '\0' || is_dir_sep(rest[3]))
851 if (skip_iprefix(rest, "modules", &rest) &&
852 (*rest == '\0' || is_dir_sep(*rest)))
857 if (rest[1] == '\0' || is_dir_sep(rest[1]))
863 int verify_path(const char *path, unsigned mode)
867 if (has_dos_drive_prefix(path))
870 if (!is_valid_path(path))
880 if (is_hfs_dotgit(path))
883 if (is_hfs_dotgitmodules(path))
888 if (is_ntfs_dotgit(path))
891 if (is_ntfs_dotgitmodules(path))
897 if ((c == '.' && !verify_dotfile(path, mode)) ||
898 is_dir_sep(c) || c == '\0')
900 } else if (c == '\\' && protect_ntfs) {
901 if (is_ntfs_dotgit(path))
904 if (is_ntfs_dotgitmodules(path))
914 * Do we have another file that has the beginning components being a
915 * proper superset of the name we're trying to add?
917 static int has_file_name(struct index_state *istate,
918 const struct cache_entry *ce, int pos, int ok_to_replace)
921 int len = ce_namelen(ce);
922 int stage = ce_stage(ce);
923 const char *name = ce->name;
925 while (pos < istate->cache_nr) {
926 struct cache_entry *p = istate->cache[pos++];
928 if (len >= ce_namelen(p))
930 if (memcmp(name, p->name, len))
932 if (ce_stage(p) != stage)
934 if (p->name[len] != '/')
936 if (p->ce_flags & CE_REMOVE)
941 remove_index_entry_at(istate, --pos);
948 * Like strcmp(), but also return the offset of the first change.
949 * If strings are equal, return the length.
951 int strcmp_offset(const char *s1, const char *s2, size_t *first_change)
956 return strcmp(s1, s2);
958 for (k = 0; s1[k] == s2[k]; k++)
963 return (unsigned char)s1[k] - (unsigned char)s2[k];
967 * Do we have another file with a pathname that is a proper
968 * subset of the name we're trying to add?
970 * That is, is there another file in the index with a path
971 * that matches a sub-directory in the given entry?
973 static int has_dir_name(struct index_state *istate,
974 const struct cache_entry *ce, int pos, int ok_to_replace)
977 int stage = ce_stage(ce);
978 const char *name = ce->name;
979 const char *slash = name + ce_namelen(ce);
984 * We are frequently called during an iteration on a sorted
985 * list of pathnames and while building a new index. Therefore,
986 * there is a high probability that this entry will eventually
987 * be appended to the index, rather than inserted in the middle.
988 * If we can confirm that, we can avoid binary searches on the
989 * components of the pathname.
991 * Compare the entry's full path with the last path in the index.
993 if (istate->cache_nr > 0) {
994 cmp_last = strcmp_offset(name,
995 istate->cache[istate->cache_nr - 1]->name,
998 if (len_eq_last == 0) {
1000 * The entry sorts AFTER the last one in the
1001 * index and their paths have no common prefix,
1002 * so there cannot be a F/D conflict.
1007 * The entry sorts AFTER the last one in the
1008 * index, but has a common prefix. Fall through
1009 * to the loop below to disect the entry's path
1010 * and see where the difference is.
1013 } else if (cmp_last == 0) {
1015 * The entry exactly matches the last one in the
1016 * index, but because of multiple stage and CE_REMOVE
1017 * items, we fall through and let the regular search
1027 if (*--slash == '/')
1029 if (slash <= ce->name)
1036 * (len + 1) is a directory boundary (including
1037 * the trailing slash). And since the loop is
1038 * decrementing "slash", the first iteration is
1039 * the longest directory prefix; subsequent
1040 * iterations consider parent directories.
1043 if (len + 1 <= len_eq_last) {
1045 * The directory prefix (including the trailing
1046 * slash) also appears as a prefix in the last
1047 * entry, so the remainder cannot collide (because
1048 * strcmp said the whole path was greater).
1053 * LT: last: xxx/file_A
1059 if (len > len_eq_last) {
1061 * This part of the directory prefix (excluding
1062 * the trailing slash) is longer than the known
1063 * equal portions, so this sub-directory cannot
1064 * collide with a file.
1072 if (istate->cache_nr > 0 &&
1073 ce_namelen(istate->cache[istate->cache_nr - 1]) > len) {
1075 * The directory prefix lines up with part of
1076 * a longer file or directory name, but sorts
1077 * after it, so this sub-directory cannot
1078 * collide with a file.
1080 * last: xxx/yy-file (because '-' sorts before '/')
1087 * This is a possible collision. Fall through and
1088 * let the regular search code handle it.
1095 pos = index_name_stage_pos(istate, name, len, stage);
1098 * Found one, but not so fast. This could
1099 * be a marker that says "I was here, but
1100 * I am being removed". Such an entry is
1101 * not a part of the resulting tree, and
1102 * it is Ok to have a directory at the same
1105 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
1109 remove_index_entry_at(istate, pos);
1117 * Trivial optimization: if we find an entry that
1118 * already matches the sub-directory, then we know
1119 * we're ok, and we can exit.
1121 while (pos < istate->cache_nr) {
1122 struct cache_entry *p = istate->cache[pos];
1123 if ((ce_namelen(p) <= len) ||
1124 (p->name[len] != '/') ||
1125 memcmp(p->name, name, len))
1126 break; /* not our subdirectory */
1127 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
1129 * p is at the same stage as our entry, and
1130 * is a subdirectory of what we are looking
1131 * at, so we cannot have conflicts at our
1132 * level or anything shorter.
1141 /* We may be in a situation where we already have path/file and path
1142 * is being added, or we already have path and path/file is being
1143 * added. Either one would result in a nonsense tree that has path
1144 * twice when git-write-tree tries to write it out. Prevent it.
1146 * If ok-to-replace is specified, we remove the conflicting entries
1147 * from the cache so the caller should recompute the insert position.
1148 * When this happens, we return non-zero.
1150 static int check_file_directory_conflict(struct index_state *istate,
1151 const struct cache_entry *ce,
1152 int pos, int ok_to_replace)
1157 * When ce is an "I am going away" entry, we allow it to be added
1159 if (ce->ce_flags & CE_REMOVE)
1163 * We check if the path is a sub-path of a subsequent pathname
1164 * first, since removing those will not change the position
1167 retval = has_file_name(istate, ce, pos, ok_to_replace);
1170 * Then check if the path might have a clashing sub-directory
1173 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
1176 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
1179 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
1180 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
1181 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
1182 int new_only = option & ADD_CACHE_NEW_ONLY;
1184 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1185 cache_tree_invalidate_path(istate, ce->name);
1188 * If this entry's path sorts after the last entry in the index,
1189 * we can avoid searching for it.
1191 if (istate->cache_nr > 0 &&
1192 strcmp(ce->name, istate->cache[istate->cache_nr - 1]->name) > 0)
1193 pos = -istate->cache_nr - 1;
1195 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1197 /* existing match? Just replace it. */
1200 replace_index_entry(istate, pos, ce);
1205 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
1206 untracked_cache_add_to_index(istate, ce->name);
1209 * Inserting a merged entry ("stage 0") into the index
1210 * will always replace all non-merged entries..
1212 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
1213 while (ce_same_name(istate->cache[pos], ce)) {
1215 if (!remove_index_entry_at(istate, pos))
1222 if (!verify_path(ce->name, ce->ce_mode))
1223 return error("Invalid path '%s'", ce->name);
1225 if (!skip_df_check &&
1226 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1228 return error("'%s' appears as both a file and as a directory",
1230 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1236 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1240 if (option & ADD_CACHE_JUST_APPEND)
1241 pos = istate->cache_nr;
1244 ret = add_index_entry_with_check(istate, ce, option);
1250 /* Make sure the array is big enough .. */
1251 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1255 if (istate->cache_nr > pos + 1)
1256 memmove(istate->cache + pos + 1,
1257 istate->cache + pos,
1258 (istate->cache_nr - pos - 1) * sizeof(ce));
1259 set_index_entry(istate, pos, ce);
1260 istate->cache_changed |= CE_ENTRY_ADDED;
1265 * "refresh" does not calculate a new sha1 file or bring the
1266 * cache up-to-date for mode/content changes. But what it
1267 * _does_ do is to "re-match" the stat information of a file
1268 * with the cache, so that you can refresh the cache for a
1269 * file that hasn't been changed but where the stat entry is
1272 * For example, you'd want to do this after doing a "git-read-tree",
1273 * to link up the stat cache details with the proper files.
1275 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1276 struct cache_entry *ce,
1277 unsigned int options, int *err,
1281 struct cache_entry *updated;
1283 int refresh = options & CE_MATCH_REFRESH;
1284 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1285 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1286 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1287 int ignore_fsmonitor = options & CE_MATCH_IGNORE_FSMONITOR;
1289 if (!refresh || ce_uptodate(ce))
1292 if (!ignore_fsmonitor)
1293 refresh_fsmonitor(istate);
1295 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1296 * that the change to the work tree does not matter and told
1299 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1300 ce_mark_uptodate(ce);
1303 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1304 ce_mark_uptodate(ce);
1307 if (!ignore_fsmonitor && (ce->ce_flags & CE_FSMONITOR_VALID)) {
1308 ce_mark_uptodate(ce);
1312 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1320 if (lstat(ce->name, &st) < 0) {
1321 if (ignore_missing && errno == ENOENT)
1328 changed = ie_match_stat(istate, ce, &st, options);
1330 *changed_ret = changed;
1333 * The path is unchanged. If we were told to ignore
1334 * valid bit, then we did the actual stat check and
1335 * found that the entry is unmodified. If the entry
1336 * is not marked VALID, this is the place to mark it
1337 * valid again, under "assume unchanged" mode.
1339 if (ignore_valid && assume_unchanged &&
1340 !(ce->ce_flags & CE_VALID))
1341 ; /* mark this one VALID again */
1344 * We do not mark the index itself "modified"
1345 * because CE_UPTODATE flag is in-core only;
1346 * we are not going to write this change out.
1348 if (!S_ISGITLINK(ce->ce_mode)) {
1349 ce_mark_uptodate(ce);
1350 mark_fsmonitor_valid(ce);
1356 if (ie_modified(istate, ce, &st, options)) {
1363 updated = xmalloc(size);
1364 memcpy(updated, ce, size);
1365 fill_stat_cache_info(updated, &st);
1367 * If ignore_valid is not set, we should leave CE_VALID bit
1368 * alone. Otherwise, paths marked with --no-assume-unchanged
1369 * (i.e. things to be edited) will reacquire CE_VALID bit
1370 * automatically, which is not really what we want.
1372 if (!ignore_valid && assume_unchanged &&
1373 !(ce->ce_flags & CE_VALID))
1374 updated->ce_flags &= ~CE_VALID;
1376 /* istate->cache_changed is updated in the caller */
1380 static void show_file(const char * fmt, const char * name, int in_porcelain,
1381 int * first, const char *header_msg)
1383 if (in_porcelain && *first && header_msg) {
1384 printf("%s\n", header_msg);
1390 int refresh_index(struct index_state *istate, unsigned int flags,
1391 const struct pathspec *pathspec,
1392 char *seen, const char *header_msg)
1396 int really = (flags & REFRESH_REALLY) != 0;
1397 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1398 int quiet = (flags & REFRESH_QUIET) != 0;
1399 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1400 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1402 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1403 unsigned int options = (CE_MATCH_REFRESH |
1404 (really ? CE_MATCH_IGNORE_VALID : 0) |
1405 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1406 const char *modified_fmt;
1407 const char *deleted_fmt;
1408 const char *typechange_fmt;
1409 const char *added_fmt;
1410 const char *unmerged_fmt;
1412 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1413 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1414 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1415 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1416 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1417 for (i = 0; i < istate->cache_nr; i++) {
1418 struct cache_entry *ce, *new;
1419 int cache_errno = 0;
1423 ce = istate->cache[i];
1424 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1427 if (pathspec && !ce_path_match(ce, pathspec, seen))
1431 while ((i < istate->cache_nr) &&
1432 ! strcmp(istate->cache[i]->name, ce->name))
1438 show_file(unmerged_fmt, ce->name, in_porcelain,
1439 &first, header_msg);
1447 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1453 if (really && cache_errno == EINVAL) {
1454 /* If we are doing --really-refresh that
1455 * means the index is not valid anymore.
1457 ce->ce_flags &= ~CE_VALID;
1458 ce->ce_flags |= CE_UPDATE_IN_BASE;
1459 mark_fsmonitor_invalid(istate, ce);
1460 istate->cache_changed |= CE_ENTRY_CHANGED;
1465 if (cache_errno == ENOENT)
1467 else if (ce_intent_to_add(ce))
1468 fmt = added_fmt; /* must be before other checks */
1469 else if (changed & TYPE_CHANGED)
1470 fmt = typechange_fmt;
1474 ce->name, in_porcelain, &first, header_msg);
1479 replace_index_entry(istate, i, new);
1484 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1485 unsigned int options)
1487 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1491 /*****************************************************************
1493 *****************************************************************/
1495 #define INDEX_FORMAT_DEFAULT 3
1497 static unsigned int get_index_format_default(void)
1499 char *envversion = getenv("GIT_INDEX_VERSION");
1502 unsigned int version = INDEX_FORMAT_DEFAULT;
1505 if (!git_config_get_int("index.version", &value))
1507 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1508 warning(_("index.version set, but the value is invalid.\n"
1509 "Using version %i"), INDEX_FORMAT_DEFAULT);
1510 return INDEX_FORMAT_DEFAULT;
1515 version = strtoul(envversion, &endp, 10);
1517 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1518 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1519 "Using version %i"), INDEX_FORMAT_DEFAULT);
1520 version = INDEX_FORMAT_DEFAULT;
1526 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1527 * Again - this is just a (very strong in practice) heuristic that
1528 * the inode hasn't changed.
1530 * We save the fields in big-endian order to allow using the
1531 * index file over NFS transparently.
1533 struct ondisk_cache_entry {
1534 struct cache_time ctime;
1535 struct cache_time mtime;
1542 unsigned char sha1[20];
1544 char name[FLEX_ARRAY]; /* more */
1548 * This struct is used when CE_EXTENDED bit is 1
1549 * The struct must match ondisk_cache_entry exactly from
1552 struct ondisk_cache_entry_extended {
1553 struct cache_time ctime;
1554 struct cache_time mtime;
1561 unsigned char sha1[20];
1564 char name[FLEX_ARRAY]; /* more */
1567 /* These are only used for v3 or lower */
1568 #define align_padding_size(size, len) ((size + (len) + 8) & ~7) - (size + len)
1569 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1570 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1571 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1572 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1573 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1574 ondisk_cache_entry_size(ce_namelen(ce)))
1576 /* Allow fsck to force verification of the index checksum. */
1577 int verify_index_checksum;
1579 /* Allow fsck to force verification of the cache entry order. */
1580 int verify_ce_order;
1582 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1585 unsigned char sha1[20];
1588 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1589 return error("bad signature");
1590 hdr_version = ntohl(hdr->hdr_version);
1591 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1592 return error("bad index version %d", hdr_version);
1594 if (!verify_index_checksum)
1598 git_SHA1_Update(&c, hdr, size - 20);
1599 git_SHA1_Final(sha1, &c);
1600 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1601 return error("bad index file sha1 signature");
1605 static int read_index_extension(struct index_state *istate,
1606 const char *ext, void *data, unsigned long sz)
1608 switch (CACHE_EXT(ext)) {
1609 case CACHE_EXT_TREE:
1610 istate->cache_tree = cache_tree_read(data, sz);
1612 case CACHE_EXT_RESOLVE_UNDO:
1613 istate->resolve_undo = resolve_undo_read(data, sz);
1615 case CACHE_EXT_LINK:
1616 if (read_link_extension(istate, data, sz))
1619 case CACHE_EXT_UNTRACKED:
1620 istate->untracked = read_untracked_extension(data, sz);
1622 case CACHE_EXT_FSMONITOR:
1623 read_fsmonitor_extension(istate, data, sz);
1626 if (*ext < 'A' || 'Z' < *ext)
1627 return error("index uses %.4s extension, which we do not understand",
1629 fprintf(stderr, "ignoring %.4s extension\n", ext);
1635 int hold_locked_index(struct lock_file *lk, int lock_flags)
1637 return hold_lock_file_for_update(lk, get_index_file(), lock_flags);
1640 int read_index(struct index_state *istate)
1642 return read_index_from(istate, get_index_file(), get_git_dir());
1645 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1650 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1652 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1653 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1654 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1655 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1656 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1657 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1658 ce->ce_mode = get_be32(&ondisk->mode);
1659 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1660 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1661 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1662 ce->ce_flags = flags & ~CE_NAMEMASK;
1663 ce->ce_namelen = len;
1665 hashcpy(ce->oid.hash, ondisk->sha1);
1666 memcpy(ce->name, name, len);
1667 ce->name[len] = '\0';
1672 * Adjacent cache entries tend to share the leading paths, so it makes
1673 * sense to only store the differences in later entries. In the v4
1674 * on-disk format of the index, each on-disk cache entry stores the
1675 * number of bytes to be stripped from the end of the previous name,
1676 * and the bytes to append to the result, to come up with its name.
1678 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1680 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1681 size_t len = decode_varint(&cp);
1683 if (name->len < len)
1684 die("malformed name field in the index");
1685 strbuf_remove(name, name->len - len, len);
1686 for (ep = cp; *ep; ep++)
1687 ; /* find the end */
1688 strbuf_add(name, cp, ep - cp);
1689 return (const char *)ep + 1 - cp_;
1692 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1693 unsigned long *ent_size,
1694 struct strbuf *previous_name)
1696 struct cache_entry *ce;
1701 /* On-disk flags are just 16 bits */
1702 flags = get_be16(&ondisk->flags);
1703 len = flags & CE_NAMEMASK;
1705 if (flags & CE_EXTENDED) {
1706 struct ondisk_cache_entry_extended *ondisk2;
1708 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1709 extended_flags = get_be16(&ondisk2->flags2) << 16;
1710 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1711 if (extended_flags & ~CE_EXTENDED_FLAGS)
1712 die("Unknown index entry format %08x", extended_flags);
1713 flags |= extended_flags;
1714 name = ondisk2->name;
1717 name = ondisk->name;
1719 if (!previous_name) {
1720 /* v3 and earlier */
1721 if (len == CE_NAMEMASK)
1723 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1725 *ent_size = ondisk_ce_size(ce);
1727 unsigned long consumed;
1728 consumed = expand_name_field(previous_name, name);
1729 ce = cache_entry_from_ondisk(ondisk, flags,
1731 previous_name->len);
1733 *ent_size = (name - ((char *)ondisk)) + consumed;
1738 static void check_ce_order(struct index_state *istate)
1742 if (!verify_ce_order)
1745 for (i = 1; i < istate->cache_nr; i++) {
1746 struct cache_entry *ce = istate->cache[i - 1];
1747 struct cache_entry *next_ce = istate->cache[i];
1748 int name_compare = strcmp(ce->name, next_ce->name);
1750 if (0 < name_compare)
1751 die("unordered stage entries in index");
1752 if (!name_compare) {
1754 die("multiple stage entries for merged file '%s'",
1756 if (ce_stage(ce) > ce_stage(next_ce))
1757 die("unordered stage entries for '%s'",
1763 static void tweak_untracked_cache(struct index_state *istate)
1765 switch (git_config_get_untracked_cache()) {
1766 case -1: /* keep: do nothing */
1769 remove_untracked_cache(istate);
1772 add_untracked_cache(istate);
1774 default: /* unknown value: do nothing */
1779 static void tweak_split_index(struct index_state *istate)
1781 switch (git_config_get_split_index()) {
1782 case -1: /* unset: do nothing */
1785 remove_split_index(istate);
1788 add_split_index(istate);
1790 default: /* unknown value: do nothing */
1795 static void post_read_index_from(struct index_state *istate)
1797 check_ce_order(istate);
1798 tweak_untracked_cache(istate);
1799 tweak_split_index(istate);
1800 tweak_fsmonitor(istate);
1803 /* remember to discard_cache() before reading a different cache! */
1804 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1808 unsigned long src_offset;
1809 struct cache_header *hdr;
1812 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1814 if (istate->initialized)
1815 return istate->cache_nr;
1817 istate->timestamp.sec = 0;
1818 istate->timestamp.nsec = 0;
1819 fd = open(path, O_RDONLY);
1821 if (!must_exist && errno == ENOENT)
1823 die_errno("%s: index file open failed", path);
1827 die_errno("cannot stat the open index");
1829 mmap_size = xsize_t(st.st_size);
1830 if (mmap_size < sizeof(struct cache_header) + 20)
1831 die("index file smaller than expected");
1833 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1834 if (mmap == MAP_FAILED)
1835 die_errno("unable to map index file");
1839 if (verify_hdr(hdr, mmap_size) < 0)
1842 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1843 istate->version = ntohl(hdr->hdr_version);
1844 istate->cache_nr = ntohl(hdr->hdr_entries);
1845 istate->cache_alloc = alloc_nr(istate->cache_nr);
1846 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1847 istate->initialized = 1;
1849 if (istate->version == 4)
1850 previous_name = &previous_name_buf;
1852 previous_name = NULL;
1854 src_offset = sizeof(*hdr);
1855 for (i = 0; i < istate->cache_nr; i++) {
1856 struct ondisk_cache_entry *disk_ce;
1857 struct cache_entry *ce;
1858 unsigned long consumed;
1860 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1861 ce = create_from_disk(disk_ce, &consumed, previous_name);
1862 set_index_entry(istate, i, ce);
1864 src_offset += consumed;
1866 strbuf_release(&previous_name_buf);
1867 istate->timestamp.sec = st.st_mtime;
1868 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1870 while (src_offset <= mmap_size - 20 - 8) {
1871 /* After an array of active_nr index entries,
1872 * there can be arbitrary number of extended
1873 * sections, each of which is prefixed with
1874 * extension name (4-byte) and section length
1875 * in 4-byte network byte order.
1878 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1879 extsize = ntohl(extsize);
1880 if (read_index_extension(istate,
1881 (const char *) mmap + src_offset,
1882 (char *) mmap + src_offset + 8,
1886 src_offset += extsize;
1888 munmap(mmap, mmap_size);
1889 return istate->cache_nr;
1892 munmap(mmap, mmap_size);
1893 die("index file corrupt");
1897 * Signal that the shared index is used by updating its mtime.
1899 * This way, shared index can be removed if they have not been used
1902 static void freshen_shared_index(const char *shared_index, int warn)
1904 if (!check_and_freshen_file(shared_index, 1) && warn)
1905 warning("could not freshen shared index '%s'", shared_index);
1908 int read_index_from(struct index_state *istate, const char *path,
1911 struct split_index *split_index;
1913 char *base_sha1_hex;
1916 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1917 if (istate->initialized)
1918 return istate->cache_nr;
1920 ret = do_read_index(istate, path, 0);
1922 split_index = istate->split_index;
1923 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1924 post_read_index_from(istate);
1928 if (split_index->base)
1929 discard_index(split_index->base);
1931 split_index->base = xcalloc(1, sizeof(*split_index->base));
1933 base_sha1_hex = sha1_to_hex(split_index->base_sha1);
1934 base_path = xstrfmt("%s/sharedindex.%s", gitdir, base_sha1_hex);
1935 ret = do_read_index(split_index->base, base_path, 1);
1936 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1937 die("broken index, expect %s in %s, got %s",
1938 base_sha1_hex, base_path,
1939 sha1_to_hex(split_index->base->sha1));
1941 freshen_shared_index(base_path, 0);
1942 merge_base_index(istate);
1943 post_read_index_from(istate);
1948 int is_index_unborn(struct index_state *istate)
1950 return (!istate->cache_nr && !istate->timestamp.sec);
1953 int discard_index(struct index_state *istate)
1957 for (i = 0; i < istate->cache_nr; i++) {
1958 if (istate->cache[i]->index &&
1959 istate->split_index &&
1960 istate->split_index->base &&
1961 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1962 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1964 free(istate->cache[i]);
1966 resolve_undo_clear_index(istate);
1967 istate->cache_nr = 0;
1968 istate->cache_changed = 0;
1969 istate->timestamp.sec = 0;
1970 istate->timestamp.nsec = 0;
1971 free_name_hash(istate);
1972 cache_tree_free(&(istate->cache_tree));
1973 istate->initialized = 0;
1974 FREE_AND_NULL(istate->cache);
1975 istate->cache_alloc = 0;
1976 discard_split_index(istate);
1977 free_untracked_cache(istate->untracked);
1978 istate->untracked = NULL;
1982 int unmerged_index(const struct index_state *istate)
1985 for (i = 0; i < istate->cache_nr; i++) {
1986 if (ce_stage(istate->cache[i]))
1992 #define WRITE_BUFFER_SIZE 8192
1993 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1994 static unsigned long write_buffer_len;
1996 static int ce_write_flush(git_SHA_CTX *context, int fd)
1998 unsigned int buffered = write_buffer_len;
2000 git_SHA1_Update(context, write_buffer, buffered);
2001 if (write_in_full(fd, write_buffer, buffered) < 0)
2003 write_buffer_len = 0;
2008 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
2011 unsigned int buffered = write_buffer_len;
2012 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
2015 memcpy(write_buffer + buffered, data, partial);
2016 buffered += partial;
2017 if (buffered == WRITE_BUFFER_SIZE) {
2018 write_buffer_len = buffered;
2019 if (ce_write_flush(context, fd))
2023 write_buffer_len = buffered;
2025 data = (char *) data + partial;
2030 static int write_index_ext_header(git_SHA_CTX *context, int fd,
2031 unsigned int ext, unsigned int sz)
2035 return ((ce_write(context, fd, &ext, 4) < 0) ||
2036 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
2039 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
2041 unsigned int left = write_buffer_len;
2044 write_buffer_len = 0;
2045 git_SHA1_Update(context, write_buffer, left);
2048 /* Flush first if not enough space for SHA1 signature */
2049 if (left + 20 > WRITE_BUFFER_SIZE) {
2050 if (write_in_full(fd, write_buffer, left) < 0)
2055 /* Append the SHA1 signature at the end */
2056 git_SHA1_Final(write_buffer + left, context);
2057 hashcpy(sha1, write_buffer + left);
2059 return (write_in_full(fd, write_buffer, left) < 0) ? -1 : 0;
2062 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
2065 * The only thing we care about in this function is to smudge the
2066 * falsely clean entry due to touch-update-touch race, so we leave
2067 * everything else as they are. We are called for entries whose
2068 * ce_stat_data.sd_mtime match the index file mtime.
2070 * Note that this actually does not do much for gitlinks, for
2071 * which ce_match_stat_basic() always goes to the actual
2072 * contents. The caller checks with is_racy_timestamp() which
2073 * always says "no" for gitlinks, so we are not called for them ;-)
2077 if (lstat(ce->name, &st) < 0)
2079 if (ce_match_stat_basic(ce, &st))
2081 if (ce_modified_check_fs(ce, &st)) {
2082 /* This is "racily clean"; smudge it. Note that this
2083 * is a tricky code. At first glance, it may appear
2084 * that it can break with this sequence:
2086 * $ echo xyzzy >frotz
2087 * $ git-update-index --add frotz
2090 * $ echo filfre >nitfol
2091 * $ git-update-index --add nitfol
2093 * but it does not. When the second update-index runs,
2094 * it notices that the entry "frotz" has the same timestamp
2095 * as index, and if we were to smudge it by resetting its
2096 * size to zero here, then the object name recorded
2097 * in index is the 6-byte file but the cached stat information
2098 * becomes zero --- which would then match what we would
2099 * obtain from the filesystem next time we stat("frotz").
2101 * However, the second update-index, before calling
2102 * this function, notices that the cached size is 6
2103 * bytes and what is on the filesystem is an empty
2104 * file, and never calls us, so the cached size information
2105 * for "frotz" stays 6 which does not match the filesystem.
2107 ce->ce_stat_data.sd_size = 0;
2111 /* Copy miscellaneous fields but not the name */
2112 static void copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
2113 struct cache_entry *ce)
2117 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
2118 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
2119 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
2120 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
2121 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
2122 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
2123 ondisk->mode = htonl(ce->ce_mode);
2124 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
2125 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
2126 ondisk->size = htonl(ce->ce_stat_data.sd_size);
2127 hashcpy(ondisk->sha1, ce->oid.hash);
2129 flags = ce->ce_flags & ~CE_NAMEMASK;
2130 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
2131 ondisk->flags = htons(flags);
2132 if (ce->ce_flags & CE_EXTENDED) {
2133 struct ondisk_cache_entry_extended *ondisk2;
2134 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
2135 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
2139 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
2140 struct strbuf *previous_name, struct ondisk_cache_entry *ondisk)
2143 int saved_namelen = saved_namelen; /* compiler workaround */
2145 static unsigned char padding[8] = { 0x00 };
2147 if (ce->ce_flags & CE_STRIP_NAME) {
2148 saved_namelen = ce_namelen(ce);
2152 if (ce->ce_flags & CE_EXTENDED)
2153 size = offsetof(struct ondisk_cache_entry_extended, name);
2155 size = offsetof(struct ondisk_cache_entry, name);
2157 if (!previous_name) {
2158 int len = ce_namelen(ce);
2159 copy_cache_entry_to_ondisk(ondisk, ce);
2160 result = ce_write(c, fd, ondisk, size);
2162 result = ce_write(c, fd, ce->name, len);
2164 result = ce_write(c, fd, padding, align_padding_size(size, len));
2166 int common, to_remove, prefix_size;
2167 unsigned char to_remove_vi[16];
2169 (ce->name[common] &&
2170 common < previous_name->len &&
2171 ce->name[common] == previous_name->buf[common]);
2173 ; /* still matching */
2174 to_remove = previous_name->len - common;
2175 prefix_size = encode_varint(to_remove, to_remove_vi);
2177 copy_cache_entry_to_ondisk(ondisk, ce);
2178 result = ce_write(c, fd, ondisk, size);
2180 result = ce_write(c, fd, to_remove_vi, prefix_size);
2182 result = ce_write(c, fd, ce->name + common, ce_namelen(ce) - common);
2184 result = ce_write(c, fd, padding, 1);
2186 strbuf_splice(previous_name, common, to_remove,
2187 ce->name + common, ce_namelen(ce) - common);
2189 if (ce->ce_flags & CE_STRIP_NAME) {
2190 ce->ce_namelen = saved_namelen;
2191 ce->ce_flags &= ~CE_STRIP_NAME;
2198 * This function verifies if index_state has the correct sha1 of the
2199 * index file. Don't die if we have any other failure, just return 0.
2201 static int verify_index_from(const struct index_state *istate, const char *path)
2206 unsigned char sha1[20];
2208 if (!istate->initialized)
2211 fd = open(path, O_RDONLY);
2218 if (st.st_size < sizeof(struct cache_header) + 20)
2221 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
2225 if (hashcmp(istate->sha1, sha1))
2236 static int verify_index(const struct index_state *istate)
2238 return verify_index_from(istate, get_index_file());
2241 static int has_racy_timestamp(struct index_state *istate)
2243 int entries = istate->cache_nr;
2246 for (i = 0; i < entries; i++) {
2247 struct cache_entry *ce = istate->cache[i];
2248 if (is_racy_timestamp(istate, ce))
2254 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
2256 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
2257 verify_index(istate))
2258 write_locked_index(istate, lockfile, COMMIT_LOCK);
2260 rollback_lock_file(lockfile);
2264 * On success, `tempfile` is closed. If it is the temporary file
2265 * of a `struct lock_file`, we will therefore effectively perform
2266 * a 'close_lock_file_gently()`. Since that is an implementation
2267 * detail of lockfiles, callers of `do_write_index()` should not
2270 static int do_write_index(struct index_state *istate, struct tempfile *tempfile,
2271 int strip_extensions)
2273 int newfd = tempfile->fd;
2275 struct cache_header hdr;
2276 int i, err = 0, removed, extended, hdr_version;
2277 struct cache_entry **cache = istate->cache;
2278 int entries = istate->cache_nr;
2280 struct ondisk_cache_entry_extended ondisk;
2281 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
2282 int drop_cache_tree = istate->drop_cache_tree;
2284 for (i = removed = extended = 0; i < entries; i++) {
2285 if (cache[i]->ce_flags & CE_REMOVE)
2288 /* reduce extended entries if possible */
2289 cache[i]->ce_flags &= ~CE_EXTENDED;
2290 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2292 cache[i]->ce_flags |= CE_EXTENDED;
2296 if (!istate->version) {
2297 istate->version = get_index_format_default();
2298 if (getenv("GIT_TEST_SPLIT_INDEX"))
2299 init_split_index(istate);
2302 /* demote version 3 to version 2 when the latter suffices */
2303 if (istate->version == 3 || istate->version == 2)
2304 istate->version = extended ? 3 : 2;
2306 hdr_version = istate->version;
2308 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2309 hdr.hdr_version = htonl(hdr_version);
2310 hdr.hdr_entries = htonl(entries - removed);
2313 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2316 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2318 for (i = 0; i < entries; i++) {
2319 struct cache_entry *ce = cache[i];
2320 if (ce->ce_flags & CE_REMOVE)
2322 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2323 ce_smudge_racily_clean_entry(ce);
2324 if (is_null_oid(&ce->oid)) {
2325 static const char msg[] = "cache entry has null sha1: %s";
2326 static int allow = -1;
2329 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2331 warning(msg, ce->name);
2333 err = error(msg, ce->name);
2335 drop_cache_tree = 1;
2337 if (ce_write_entry(&c, newfd, ce, previous_name, (struct ondisk_cache_entry *)&ondisk) < 0)
2343 strbuf_release(&previous_name_buf);
2348 /* Write extension data here */
2349 if (!strip_extensions && istate->split_index) {
2350 struct strbuf sb = STRBUF_INIT;
2352 err = write_link_extension(&sb, istate) < 0 ||
2353 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2355 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2356 strbuf_release(&sb);
2360 if (!strip_extensions && !drop_cache_tree && istate->cache_tree) {
2361 struct strbuf sb = STRBUF_INIT;
2363 cache_tree_write(&sb, istate->cache_tree);
2364 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2365 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2366 strbuf_release(&sb);
2370 if (!strip_extensions && istate->resolve_undo) {
2371 struct strbuf sb = STRBUF_INIT;
2373 resolve_undo_write(&sb, istate->resolve_undo);
2374 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2376 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2377 strbuf_release(&sb);
2381 if (!strip_extensions && istate->untracked) {
2382 struct strbuf sb = STRBUF_INIT;
2384 write_untracked_extension(&sb, istate->untracked);
2385 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2387 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2388 strbuf_release(&sb);
2392 if (!strip_extensions && istate->fsmonitor_last_update) {
2393 struct strbuf sb = STRBUF_INIT;
2395 write_fsmonitor_extension(&sb, istate);
2396 err = write_index_ext_header(&c, newfd, CACHE_EXT_FSMONITOR, sb.len) < 0
2397 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2398 strbuf_release(&sb);
2403 if (ce_flush(&c, newfd, istate->sha1))
2405 if (close_tempfile_gently(tempfile)) {
2406 error(_("could not close '%s'"), tempfile->filename.buf);
2409 if (stat(tempfile->filename.buf, &st))
2411 istate->timestamp.sec = (unsigned int)st.st_mtime;
2412 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2416 void set_alternate_index_output(const char *name)
2418 alternate_index_output = name;
2421 static int commit_locked_index(struct lock_file *lk)
2423 if (alternate_index_output)
2424 return commit_lock_file_to(lk, alternate_index_output);
2426 return commit_lock_file(lk);
2429 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2432 int ret = do_write_index(istate, lock->tempfile, 0);
2435 if (flags & COMMIT_LOCK)
2436 return commit_locked_index(lock);
2437 return close_lock_file_gently(lock);
2440 static int write_split_index(struct index_state *istate,
2441 struct lock_file *lock,
2445 prepare_to_write_split_index(istate);
2446 ret = do_write_locked_index(istate, lock, flags);
2447 finish_writing_split_index(istate);
2451 static const char *shared_index_expire = "2.weeks.ago";
2453 static unsigned long get_shared_index_expire_date(void)
2455 static unsigned long shared_index_expire_date;
2456 static int shared_index_expire_date_prepared;
2458 if (!shared_index_expire_date_prepared) {
2459 git_config_get_expiry("splitindex.sharedindexexpire",
2460 &shared_index_expire);
2461 shared_index_expire_date = approxidate(shared_index_expire);
2462 shared_index_expire_date_prepared = 1;
2465 return shared_index_expire_date;
2468 static int should_delete_shared_index(const char *shared_index_path)
2471 unsigned long expiration;
2473 /* Check timestamp */
2474 expiration = get_shared_index_expire_date();
2477 if (stat(shared_index_path, &st))
2478 return error_errno(_("could not stat '%s'"), shared_index_path);
2479 if (st.st_mtime > expiration)
2485 static int clean_shared_index_files(const char *current_hex)
2488 DIR *dir = opendir(get_git_dir());
2491 return error_errno(_("unable to open git dir: %s"), get_git_dir());
2493 while ((de = readdir(dir)) != NULL) {
2494 const char *sha1_hex;
2495 const char *shared_index_path;
2496 if (!skip_prefix(de->d_name, "sharedindex.", &sha1_hex))
2498 if (!strcmp(sha1_hex, current_hex))
2500 shared_index_path = git_path("%s", de->d_name);
2501 if (should_delete_shared_index(shared_index_path) > 0 &&
2502 unlink(shared_index_path))
2503 warning_errno(_("unable to unlink: %s"), shared_index_path);
2510 static int write_shared_index(struct index_state *istate,
2511 struct tempfile **temp)
2513 struct split_index *si = istate->split_index;
2516 move_cache_to_base_index(istate);
2517 ret = do_write_index(si->base, *temp, 1);
2520 ret = adjust_shared_perm(get_tempfile_path(*temp));
2522 error("cannot fix permission bits on %s", get_tempfile_path(*temp));
2525 ret = rename_tempfile(temp,
2526 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2528 hashcpy(si->base_sha1, si->base->sha1);
2529 clean_shared_index_files(sha1_to_hex(si->base->sha1));
2535 static const int default_max_percent_split_change = 20;
2537 static int too_many_not_shared_entries(struct index_state *istate)
2539 int i, not_shared = 0;
2540 int max_split = git_config_get_max_percent_split_change();
2542 switch (max_split) {
2544 /* not or badly configured: use the default value */
2545 max_split = default_max_percent_split_change;
2548 return 1; /* 0% means always write a new shared index */
2550 return 0; /* 100% means never write a new shared index */
2552 break; /* just use the configured value */
2555 /* Count not shared entries */
2556 for (i = 0; i < istate->cache_nr; i++) {
2557 struct cache_entry *ce = istate->cache[i];
2562 return (int64_t)istate->cache_nr * max_split < (int64_t)not_shared * 100;
2565 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2568 int new_shared_index, ret;
2569 struct split_index *si = istate->split_index;
2571 if (istate->fsmonitor_last_update)
2572 fill_fsmonitor_bitmap(istate);
2574 if (!si || alternate_index_output ||
2575 (istate->cache_changed & ~EXTMASK)) {
2577 hashclr(si->base_sha1);
2578 ret = do_write_locked_index(istate, lock, flags);
2582 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2583 int v = si->base_sha1[0];
2585 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2587 if (too_many_not_shared_entries(istate))
2588 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2590 new_shared_index = istate->cache_changed & SPLIT_INDEX_ORDERED;
2592 if (new_shared_index) {
2593 struct tempfile *temp;
2596 temp = mks_tempfile(git_path("sharedindex_XXXXXX"));
2598 hashclr(si->base_sha1);
2599 ret = do_write_locked_index(istate, lock, flags);
2602 ret = write_shared_index(istate, &temp);
2604 saved_errno = errno;
2605 if (is_tempfile_active(temp))
2606 delete_tempfile(&temp);
2607 errno = saved_errno;
2613 ret = write_split_index(istate, lock, flags);
2615 /* Freshen the shared index only if the split-index was written */
2616 if (!ret && !new_shared_index) {
2617 const char *shared_index = git_path("sharedindex.%s",
2618 sha1_to_hex(si->base_sha1));
2619 freshen_shared_index(shared_index, 1);
2623 if (flags & COMMIT_LOCK)
2624 rollback_lock_file(lock);
2629 * Read the index file that is potentially unmerged into given
2630 * index_state, dropping any unmerged entries. Returns true if
2631 * the index is unmerged. Callers who want to refuse to work
2632 * from an unmerged state can call this and check its return value,
2633 * instead of calling read_cache().
2635 int read_index_unmerged(struct index_state *istate)
2641 for (i = 0; i < istate->cache_nr; i++) {
2642 struct cache_entry *ce = istate->cache[i];
2643 struct cache_entry *new_ce;
2649 len = ce_namelen(ce);
2650 size = cache_entry_size(len);
2651 new_ce = xcalloc(1, size);
2652 memcpy(new_ce->name, ce->name, len);
2653 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2654 new_ce->ce_namelen = len;
2655 new_ce->ce_mode = ce->ce_mode;
2656 if (add_index_entry(istate, new_ce, 0))
2657 return error("%s: cannot drop to stage #0",
2664 * Returns 1 if the path is an "other" path with respect to
2665 * the index; that is, the path is not mentioned in the index at all,
2666 * either as a file, a directory with some files in the index,
2667 * or as an unmerged entry.
2669 * We helpfully remove a trailing "/" from directories so that
2670 * the output of read_directory can be used as-is.
2672 int index_name_is_other(const struct index_state *istate, const char *name,
2676 if (namelen && name[namelen - 1] == '/')
2678 pos = index_name_pos(istate, name, namelen);
2680 return 0; /* exact match */
2682 if (pos < istate->cache_nr) {
2683 struct cache_entry *ce = istate->cache[pos];
2684 if (ce_namelen(ce) == namelen &&
2685 !memcmp(ce->name, name, namelen))
2686 return 0; /* Yup, this one exists unmerged */
2691 void *read_blob_data_from_index(const struct index_state *istate,
2692 const char *path, unsigned long *size)
2696 enum object_type type;
2700 pos = index_name_pos(istate, path, len);
2703 * We might be in the middle of a merge, in which
2704 * case we would read stage #2 (ours).
2708 (pos < 0 && i < istate->cache_nr &&
2709 !strcmp(istate->cache[i]->name, path));
2711 if (ce_stage(istate->cache[i]) == 2)
2716 data = read_sha1_file(istate->cache[pos]->oid.hash, &type, &sz);
2717 if (!data || type != OBJ_BLOB) {
2726 void stat_validity_clear(struct stat_validity *sv)
2728 FREE_AND_NULL(sv->sd);
2731 int stat_validity_check(struct stat_validity *sv, const char *path)
2735 if (stat(path, &st) < 0)
2736 return sv->sd == NULL;
2739 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2742 void stat_validity_update(struct stat_validity *sv, int fd)
2746 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2747 stat_validity_clear(sv);
2750 sv->sd = xcalloc(1, sizeof(struct stat_data));
2751 fill_stat_data(sv->sd, &st);
2755 void move_index_extensions(struct index_state *dst, struct index_state *src)
2757 dst->untracked = src->untracked;
2758 src->untracked = NULL;