merge-recursive: handle NULL in add_cacheinfo() correctly
[git/raj.git] / read-cache.c
blob6af409a10c206831253f5956a130d4884aa372ea
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define NO_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "tempfile.h"
9 #include "lockfile.h"
10 #include "cache-tree.h"
11 #include "refs.h"
12 #include "dir.h"
13 #include "tree.h"
14 #include "commit.h"
15 #include "blob.h"
16 #include "resolve-undo.h"
17 #include "strbuf.h"
18 #include "varint.h"
19 #include "split-index.h"
20 #include "utf8.h"
22 /* Mask for the name length in ce_flags in the on-disk index */
24 #define CE_NAMEMASK (0x0fff)
26 /* Index extensions.
28 * The first letter should be 'A'..'Z' for extensions that are not
29 * necessary for a correct operation (i.e. optimization data).
30 * When new extensions are added that _needs_ to be understood in
31 * order to correctly interpret the index file, pick character that
32 * is outside the range, to cause the reader to abort.
35 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
36 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
37 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
38 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
39 #define CACHE_EXT_UNTRACKED 0x554E5452 /* "UNTR" */
41 /* changes that can be kept in $GIT_DIR/index (basically all extensions) */
42 #define EXTMASK (RESOLVE_UNDO_CHANGED | CACHE_TREE_CHANGED | \
43 CE_ENTRY_ADDED | CE_ENTRY_REMOVED | CE_ENTRY_CHANGED | \
44 SPLIT_INDEX_ORDERED | UNTRACKED_CHANGED)
46 struct index_state the_index;
47 static const char *alternate_index_output;
49 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
51 istate->cache[nr] = ce;
52 add_name_hash(istate, ce);
55 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
57 struct cache_entry *old = istate->cache[nr];
59 replace_index_entry_in_base(istate, old, ce);
60 remove_name_hash(istate, old);
61 free(old);
62 set_index_entry(istate, nr, ce);
63 ce->ce_flags |= CE_UPDATE_IN_BASE;
64 istate->cache_changed |= CE_ENTRY_CHANGED;
67 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
69 struct cache_entry *old = istate->cache[nr], *new;
70 int namelen = strlen(new_name);
72 new = xmalloc(cache_entry_size(namelen));
73 copy_cache_entry(new, old);
74 new->ce_flags &= ~CE_HASHED;
75 new->ce_namelen = namelen;
76 new->index = 0;
77 memcpy(new->name, new_name, namelen + 1);
79 cache_tree_invalidate_path(istate, old->name);
80 untracked_cache_remove_from_index(istate, old->name);
81 remove_index_entry_at(istate, nr);
82 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
85 void fill_stat_data(struct stat_data *sd, struct stat *st)
87 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
88 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
89 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
90 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
91 sd->sd_dev = st->st_dev;
92 sd->sd_ino = st->st_ino;
93 sd->sd_uid = st->st_uid;
94 sd->sd_gid = st->st_gid;
95 sd->sd_size = st->st_size;
98 int match_stat_data(const struct stat_data *sd, struct stat *st)
100 int changed = 0;
102 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
103 changed |= MTIME_CHANGED;
104 if (trust_ctime && check_stat &&
105 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
106 changed |= CTIME_CHANGED;
108 #ifdef USE_NSEC
109 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
110 changed |= MTIME_CHANGED;
111 if (trust_ctime && check_stat &&
112 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
113 changed |= CTIME_CHANGED;
114 #endif
116 if (check_stat) {
117 if (sd->sd_uid != (unsigned int) st->st_uid ||
118 sd->sd_gid != (unsigned int) st->st_gid)
119 changed |= OWNER_CHANGED;
120 if (sd->sd_ino != (unsigned int) st->st_ino)
121 changed |= INODE_CHANGED;
124 #ifdef USE_STDEV
126 * st_dev breaks on network filesystems where different
127 * clients will have different views of what "device"
128 * the filesystem is on
130 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
131 changed |= INODE_CHANGED;
132 #endif
134 if (sd->sd_size != (unsigned int) st->st_size)
135 changed |= DATA_CHANGED;
137 return changed;
141 * This only updates the "non-critical" parts of the directory
142 * cache, ie the parts that aren't tracked by GIT, and only used
143 * to validate the cache.
145 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
147 fill_stat_data(&ce->ce_stat_data, st);
149 if (assume_unchanged)
150 ce->ce_flags |= CE_VALID;
152 if (S_ISREG(st->st_mode))
153 ce_mark_uptodate(ce);
156 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
158 int match = -1;
159 int fd = open(ce->name, O_RDONLY);
161 if (fd >= 0) {
162 unsigned char sha1[20];
163 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
164 match = hashcmp(sha1, ce->sha1);
165 /* index_fd() closed the file descriptor already */
167 return match;
170 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
172 int match = -1;
173 void *buffer;
174 unsigned long size;
175 enum object_type type;
176 struct strbuf sb = STRBUF_INIT;
178 if (strbuf_readlink(&sb, ce->name, expected_size))
179 return -1;
181 buffer = read_sha1_file(ce->sha1, &type, &size);
182 if (buffer) {
183 if (size == sb.len)
184 match = memcmp(buffer, sb.buf, size);
185 free(buffer);
187 strbuf_release(&sb);
188 return match;
191 static int ce_compare_gitlink(const struct cache_entry *ce)
193 unsigned char sha1[20];
196 * We don't actually require that the .git directory
197 * under GITLINK directory be a valid git directory. It
198 * might even be missing (in case nobody populated that
199 * sub-project).
201 * If so, we consider it always to match.
203 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
204 return 0;
205 return hashcmp(sha1, ce->sha1);
208 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
210 switch (st->st_mode & S_IFMT) {
211 case S_IFREG:
212 if (ce_compare_data(ce, st))
213 return DATA_CHANGED;
214 break;
215 case S_IFLNK:
216 if (ce_compare_link(ce, xsize_t(st->st_size)))
217 return DATA_CHANGED;
218 break;
219 case S_IFDIR:
220 if (S_ISGITLINK(ce->ce_mode))
221 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
222 default:
223 return TYPE_CHANGED;
225 return 0;
228 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
230 unsigned int changed = 0;
232 if (ce->ce_flags & CE_REMOVE)
233 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
235 switch (ce->ce_mode & S_IFMT) {
236 case S_IFREG:
237 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
238 /* We consider only the owner x bit to be relevant for
239 * "mode changes"
241 if (trust_executable_bit &&
242 (0100 & (ce->ce_mode ^ st->st_mode)))
243 changed |= MODE_CHANGED;
244 break;
245 case S_IFLNK:
246 if (!S_ISLNK(st->st_mode) &&
247 (has_symlinks || !S_ISREG(st->st_mode)))
248 changed |= TYPE_CHANGED;
249 break;
250 case S_IFGITLINK:
251 /* We ignore most of the st_xxx fields for gitlinks */
252 if (!S_ISDIR(st->st_mode))
253 changed |= TYPE_CHANGED;
254 else if (ce_compare_gitlink(ce))
255 changed |= DATA_CHANGED;
256 return changed;
257 default:
258 die("internal error: ce_mode is %o", ce->ce_mode);
261 changed |= match_stat_data(&ce->ce_stat_data, st);
263 /* Racily smudged entry? */
264 if (!ce->ce_stat_data.sd_size) {
265 if (!is_empty_blob_sha1(ce->sha1))
266 changed |= DATA_CHANGED;
269 return changed;
272 static int is_racy_stat(const struct index_state *istate,
273 const struct stat_data *sd)
275 return (istate->timestamp.sec &&
276 #ifdef USE_NSEC
277 /* nanosecond timestamped files can also be racy! */
278 (istate->timestamp.sec < sd->sd_mtime.sec ||
279 (istate->timestamp.sec == sd->sd_mtime.sec &&
280 istate->timestamp.nsec <= sd->sd_mtime.nsec))
281 #else
282 istate->timestamp.sec <= sd->sd_mtime.sec
283 #endif
287 static int is_racy_timestamp(const struct index_state *istate,
288 const struct cache_entry *ce)
290 return (!S_ISGITLINK(ce->ce_mode) &&
291 is_racy_stat(istate, &ce->ce_stat_data));
294 int match_stat_data_racy(const struct index_state *istate,
295 const struct stat_data *sd, struct stat *st)
297 if (is_racy_stat(istate, sd))
298 return MTIME_CHANGED;
299 return match_stat_data(sd, st);
302 int ie_match_stat(const struct index_state *istate,
303 const struct cache_entry *ce, struct stat *st,
304 unsigned int options)
306 unsigned int changed;
307 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
308 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
309 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
312 * If it's marked as always valid in the index, it's
313 * valid whatever the checked-out copy says.
315 * skip-worktree has the same effect with higher precedence
317 if (!ignore_skip_worktree && ce_skip_worktree(ce))
318 return 0;
319 if (!ignore_valid && (ce->ce_flags & CE_VALID))
320 return 0;
323 * Intent-to-add entries have not been added, so the index entry
324 * by definition never matches what is in the work tree until it
325 * actually gets added.
327 if (ce_intent_to_add(ce))
328 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
330 changed = ce_match_stat_basic(ce, st);
333 * Within 1 second of this sequence:
334 * echo xyzzy >file && git-update-index --add file
335 * running this command:
336 * echo frotz >file
337 * would give a falsely clean cache entry. The mtime and
338 * length match the cache, and other stat fields do not change.
340 * We could detect this at update-index time (the cache entry
341 * being registered/updated records the same time as "now")
342 * and delay the return from git-update-index, but that would
343 * effectively mean we can make at most one commit per second,
344 * which is not acceptable. Instead, we check cache entries
345 * whose mtime are the same as the index file timestamp more
346 * carefully than others.
348 if (!changed && is_racy_timestamp(istate, ce)) {
349 if (assume_racy_is_modified)
350 changed |= DATA_CHANGED;
351 else
352 changed |= ce_modified_check_fs(ce, st);
355 return changed;
358 int ie_modified(const struct index_state *istate,
359 const struct cache_entry *ce,
360 struct stat *st, unsigned int options)
362 int changed, changed_fs;
364 changed = ie_match_stat(istate, ce, st, options);
365 if (!changed)
366 return 0;
368 * If the mode or type has changed, there's no point in trying
369 * to refresh the entry - it's not going to match
371 if (changed & (MODE_CHANGED | TYPE_CHANGED))
372 return changed;
375 * Immediately after read-tree or update-index --cacheinfo,
376 * the length field is zero, as we have never even read the
377 * lstat(2) information once, and we cannot trust DATA_CHANGED
378 * returned by ie_match_stat() which in turn was returned by
379 * ce_match_stat_basic() to signal that the filesize of the
380 * blob changed. We have to actually go to the filesystem to
381 * see if the contents match, and if so, should answer "unchanged".
383 * The logic does not apply to gitlinks, as ce_match_stat_basic()
384 * already has checked the actual HEAD from the filesystem in the
385 * subproject. If ie_match_stat() already said it is different,
386 * then we know it is.
388 if ((changed & DATA_CHANGED) &&
389 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
390 return changed;
392 changed_fs = ce_modified_check_fs(ce, st);
393 if (changed_fs)
394 return changed | changed_fs;
395 return 0;
398 int base_name_compare(const char *name1, int len1, int mode1,
399 const char *name2, int len2, int mode2)
401 unsigned char c1, c2;
402 int len = len1 < len2 ? len1 : len2;
403 int cmp;
405 cmp = memcmp(name1, name2, len);
406 if (cmp)
407 return cmp;
408 c1 = name1[len];
409 c2 = name2[len];
410 if (!c1 && S_ISDIR(mode1))
411 c1 = '/';
412 if (!c2 && S_ISDIR(mode2))
413 c2 = '/';
414 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
418 * df_name_compare() is identical to base_name_compare(), except it
419 * compares conflicting directory/file entries as equal. Note that
420 * while a directory name compares as equal to a regular file, they
421 * then individually compare _differently_ to a filename that has
422 * a dot after the basename (because '\0' < '.' < '/').
424 * This is used by routines that want to traverse the git namespace
425 * but then handle conflicting entries together when possible.
427 int df_name_compare(const char *name1, int len1, int mode1,
428 const char *name2, int len2, int mode2)
430 int len = len1 < len2 ? len1 : len2, cmp;
431 unsigned char c1, c2;
433 cmp = memcmp(name1, name2, len);
434 if (cmp)
435 return cmp;
436 /* Directories and files compare equal (same length, same name) */
437 if (len1 == len2)
438 return 0;
439 c1 = name1[len];
440 if (!c1 && S_ISDIR(mode1))
441 c1 = '/';
442 c2 = name2[len];
443 if (!c2 && S_ISDIR(mode2))
444 c2 = '/';
445 if (c1 == '/' && !c2)
446 return 0;
447 if (c2 == '/' && !c1)
448 return 0;
449 return c1 - c2;
452 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
454 size_t min_len = (len1 < len2) ? len1 : len2;
455 int cmp = memcmp(name1, name2, min_len);
456 if (cmp)
457 return cmp;
458 if (len1 < len2)
459 return -1;
460 if (len1 > len2)
461 return 1;
462 return 0;
465 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
467 int cmp;
469 cmp = name_compare(name1, len1, name2, len2);
470 if (cmp)
471 return cmp;
473 if (stage1 < stage2)
474 return -1;
475 if (stage1 > stage2)
476 return 1;
477 return 0;
480 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
482 int first, last;
484 first = 0;
485 last = istate->cache_nr;
486 while (last > first) {
487 int next = (last + first) >> 1;
488 struct cache_entry *ce = istate->cache[next];
489 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
490 if (!cmp)
491 return next;
492 if (cmp < 0) {
493 last = next;
494 continue;
496 first = next+1;
498 return -first-1;
501 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
503 return index_name_stage_pos(istate, name, namelen, 0);
506 /* Remove entry, return true if there are more entries to go.. */
507 int remove_index_entry_at(struct index_state *istate, int pos)
509 struct cache_entry *ce = istate->cache[pos];
511 record_resolve_undo(istate, ce);
512 remove_name_hash(istate, ce);
513 save_or_free_index_entry(istate, ce);
514 istate->cache_changed |= CE_ENTRY_REMOVED;
515 istate->cache_nr--;
516 if (pos >= istate->cache_nr)
517 return 0;
518 memmove(istate->cache + pos,
519 istate->cache + pos + 1,
520 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
521 return 1;
525 * Remove all cache entries marked for removal, that is where
526 * CE_REMOVE is set in ce_flags. This is much more effective than
527 * calling remove_index_entry_at() for each entry to be removed.
529 void remove_marked_cache_entries(struct index_state *istate)
531 struct cache_entry **ce_array = istate->cache;
532 unsigned int i, j;
534 for (i = j = 0; i < istate->cache_nr; i++) {
535 if (ce_array[i]->ce_flags & CE_REMOVE) {
536 remove_name_hash(istate, ce_array[i]);
537 save_or_free_index_entry(istate, ce_array[i]);
539 else
540 ce_array[j++] = ce_array[i];
542 if (j == istate->cache_nr)
543 return;
544 istate->cache_changed |= CE_ENTRY_REMOVED;
545 istate->cache_nr = j;
548 int remove_file_from_index(struct index_state *istate, const char *path)
550 int pos = index_name_pos(istate, path, strlen(path));
551 if (pos < 0)
552 pos = -pos-1;
553 cache_tree_invalidate_path(istate, path);
554 untracked_cache_remove_from_index(istate, path);
555 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
556 remove_index_entry_at(istate, pos);
557 return 0;
560 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
562 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
565 static int index_name_pos_also_unmerged(struct index_state *istate,
566 const char *path, int namelen)
568 int pos = index_name_pos(istate, path, namelen);
569 struct cache_entry *ce;
571 if (pos >= 0)
572 return pos;
574 /* maybe unmerged? */
575 pos = -1 - pos;
576 if (pos >= istate->cache_nr ||
577 compare_name((ce = istate->cache[pos]), path, namelen))
578 return -1;
580 /* order of preference: stage 2, 1, 3 */
581 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
582 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
583 !compare_name(ce, path, namelen))
584 pos++;
585 return pos;
588 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
590 int len = ce_namelen(ce);
591 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
595 * If we add a filename that aliases in the cache, we will use the
596 * name that we already have - but we don't want to update the same
597 * alias twice, because that implies that there were actually two
598 * different files with aliasing names!
600 * So we use the CE_ADDED flag to verify that the alias was an old
601 * one before we accept it as
603 static struct cache_entry *create_alias_ce(struct index_state *istate,
604 struct cache_entry *ce,
605 struct cache_entry *alias)
607 int len;
608 struct cache_entry *new;
610 if (alias->ce_flags & CE_ADDED)
611 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
613 /* Ok, create the new entry using the name of the existing alias */
614 len = ce_namelen(alias);
615 new = xcalloc(1, cache_entry_size(len));
616 memcpy(new->name, alias->name, len);
617 copy_cache_entry(new, ce);
618 save_or_free_index_entry(istate, ce);
619 return new;
622 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
624 unsigned char sha1[20];
625 if (write_sha1_file("", 0, blob_type, sha1))
626 die("cannot create an empty blob in the object database");
627 hashcpy(ce->sha1, sha1);
630 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
632 int size, namelen, was_same;
633 mode_t st_mode = st->st_mode;
634 struct cache_entry *ce, *alias;
635 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
636 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
637 int pretend = flags & ADD_CACHE_PRETEND;
638 int intent_only = flags & ADD_CACHE_INTENT;
639 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
640 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
642 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
643 return error("%s: can only add regular files, symbolic links or git-directories", path);
645 namelen = strlen(path);
646 if (S_ISDIR(st_mode)) {
647 while (namelen && path[namelen-1] == '/')
648 namelen--;
650 size = cache_entry_size(namelen);
651 ce = xcalloc(1, size);
652 memcpy(ce->name, path, namelen);
653 ce->ce_namelen = namelen;
654 if (!intent_only)
655 fill_stat_cache_info(ce, st);
656 else
657 ce->ce_flags |= CE_INTENT_TO_ADD;
659 if (trust_executable_bit && has_symlinks)
660 ce->ce_mode = create_ce_mode(st_mode);
661 else {
662 /* If there is an existing entry, pick the mode bits and type
663 * from it, otherwise assume unexecutable regular file.
665 struct cache_entry *ent;
666 int pos = index_name_pos_also_unmerged(istate, path, namelen);
668 ent = (0 <= pos) ? istate->cache[pos] : NULL;
669 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
672 /* When core.ignorecase=true, determine if a directory of the same name but differing
673 * case already exists within the Git repository. If it does, ensure the directory
674 * case of the file being added to the repository matches (is folded into) the existing
675 * entry's directory case.
677 if (ignore_case) {
678 adjust_dirname_case(istate, ce->name);
681 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
682 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
683 /* Nothing changed, really */
684 if (!S_ISGITLINK(alias->ce_mode))
685 ce_mark_uptodate(alias);
686 alias->ce_flags |= CE_ADDED;
688 free(ce);
689 return 0;
691 if (!intent_only) {
692 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT)) {
693 free(ce);
694 return error("unable to index file %s", path);
696 } else
697 set_object_name_for_intent_to_add_entry(ce);
699 if (ignore_case && alias && different_name(ce, alias))
700 ce = create_alias_ce(istate, ce, alias);
701 ce->ce_flags |= CE_ADDED;
703 /* It was suspected to be racily clean, but it turns out to be Ok */
704 was_same = (alias &&
705 !ce_stage(alias) &&
706 !hashcmp(alias->sha1, ce->sha1) &&
707 ce->ce_mode == alias->ce_mode);
709 if (pretend)
710 free(ce);
711 else if (add_index_entry(istate, ce, add_option)) {
712 free(ce);
713 return error("unable to add %s to index", path);
715 if (verbose && !was_same)
716 printf("add '%s'\n", path);
717 return 0;
720 int add_file_to_index(struct index_state *istate, const char *path, int flags)
722 struct stat st;
723 if (lstat(path, &st))
724 die_errno("unable to stat '%s'", path);
725 return add_to_index(istate, path, &st, flags);
728 struct cache_entry *make_cache_entry(unsigned int mode,
729 const unsigned char *sha1, const char *path, int stage,
730 unsigned int refresh_options)
732 int size, len;
733 struct cache_entry *ce, *ret;
735 if (!verify_path(path)) {
736 error("Invalid path '%s'", path);
737 return NULL;
740 len = strlen(path);
741 size = cache_entry_size(len);
742 ce = xcalloc(1, size);
744 hashcpy(ce->sha1, sha1);
745 memcpy(ce->name, path, len);
746 ce->ce_flags = create_ce_flags(stage);
747 ce->ce_namelen = len;
748 ce->ce_mode = create_ce_mode(mode);
750 ret = refresh_cache_entry(ce, refresh_options);
751 if (ret != ce)
752 free(ce);
753 return ret;
756 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
758 int len = ce_namelen(a);
759 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
763 * We fundamentally don't like some paths: we don't want
764 * dot or dot-dot anywhere, and for obvious reasons don't
765 * want to recurse into ".git" either.
767 * Also, we don't want double slashes or slashes at the
768 * end that can make pathnames ambiguous.
770 static int verify_dotfile(const char *rest)
773 * The first character was '.', but that
774 * has already been discarded, we now test
775 * the rest.
778 /* "." is not allowed */
779 if (*rest == '\0' || is_dir_sep(*rest))
780 return 0;
782 switch (*rest) {
784 * ".git" followed by NUL or slash is bad. This
785 * shares the path end test with the ".." case.
787 case 'g':
788 case 'G':
789 if (rest[1] != 'i' && rest[1] != 'I')
790 break;
791 if (rest[2] != 't' && rest[2] != 'T')
792 break;
793 rest += 2;
794 /* fallthrough */
795 case '.':
796 if (rest[1] == '\0' || is_dir_sep(rest[1]))
797 return 0;
799 return 1;
802 int verify_path(const char *path)
804 char c;
806 if (has_dos_drive_prefix(path))
807 return 0;
809 goto inside;
810 for (;;) {
811 if (!c)
812 return 1;
813 if (is_dir_sep(c)) {
814 inside:
815 if (protect_hfs && is_hfs_dotgit(path))
816 return 0;
817 if (protect_ntfs && is_ntfs_dotgit(path))
818 return 0;
819 c = *path++;
820 if ((c == '.' && !verify_dotfile(path)) ||
821 is_dir_sep(c) || c == '\0')
822 return 0;
824 c = *path++;
829 * Do we have another file that has the beginning components being a
830 * proper superset of the name we're trying to add?
832 static int has_file_name(struct index_state *istate,
833 const struct cache_entry *ce, int pos, int ok_to_replace)
835 int retval = 0;
836 int len = ce_namelen(ce);
837 int stage = ce_stage(ce);
838 const char *name = ce->name;
840 while (pos < istate->cache_nr) {
841 struct cache_entry *p = istate->cache[pos++];
843 if (len >= ce_namelen(p))
844 break;
845 if (memcmp(name, p->name, len))
846 break;
847 if (ce_stage(p) != stage)
848 continue;
849 if (p->name[len] != '/')
850 continue;
851 if (p->ce_flags & CE_REMOVE)
852 continue;
853 retval = -1;
854 if (!ok_to_replace)
855 break;
856 remove_index_entry_at(istate, --pos);
858 return retval;
862 * Do we have another file with a pathname that is a proper
863 * subset of the name we're trying to add?
865 static int has_dir_name(struct index_state *istate,
866 const struct cache_entry *ce, int pos, int ok_to_replace)
868 int retval = 0;
869 int stage = ce_stage(ce);
870 const char *name = ce->name;
871 const char *slash = name + ce_namelen(ce);
873 for (;;) {
874 int len;
876 for (;;) {
877 if (*--slash == '/')
878 break;
879 if (slash <= ce->name)
880 return retval;
882 len = slash - name;
884 pos = index_name_stage_pos(istate, name, len, stage);
885 if (pos >= 0) {
887 * Found one, but not so fast. This could
888 * be a marker that says "I was here, but
889 * I am being removed". Such an entry is
890 * not a part of the resulting tree, and
891 * it is Ok to have a directory at the same
892 * path.
894 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
895 retval = -1;
896 if (!ok_to_replace)
897 break;
898 remove_index_entry_at(istate, pos);
899 continue;
902 else
903 pos = -pos-1;
906 * Trivial optimization: if we find an entry that
907 * already matches the sub-directory, then we know
908 * we're ok, and we can exit.
910 while (pos < istate->cache_nr) {
911 struct cache_entry *p = istate->cache[pos];
912 if ((ce_namelen(p) <= len) ||
913 (p->name[len] != '/') ||
914 memcmp(p->name, name, len))
915 break; /* not our subdirectory */
916 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
918 * p is at the same stage as our entry, and
919 * is a subdirectory of what we are looking
920 * at, so we cannot have conflicts at our
921 * level or anything shorter.
923 return retval;
924 pos++;
927 return retval;
930 /* We may be in a situation where we already have path/file and path
931 * is being added, or we already have path and path/file is being
932 * added. Either one would result in a nonsense tree that has path
933 * twice when git-write-tree tries to write it out. Prevent it.
935 * If ok-to-replace is specified, we remove the conflicting entries
936 * from the cache so the caller should recompute the insert position.
937 * When this happens, we return non-zero.
939 static int check_file_directory_conflict(struct index_state *istate,
940 const struct cache_entry *ce,
941 int pos, int ok_to_replace)
943 int retval;
946 * When ce is an "I am going away" entry, we allow it to be added
948 if (ce->ce_flags & CE_REMOVE)
949 return 0;
952 * We check if the path is a sub-path of a subsequent pathname
953 * first, since removing those will not change the position
954 * in the array.
956 retval = has_file_name(istate, ce, pos, ok_to_replace);
959 * Then check if the path might have a clashing sub-directory
960 * before it.
962 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
965 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
967 int pos;
968 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
969 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
970 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
971 int new_only = option & ADD_CACHE_NEW_ONLY;
973 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
974 cache_tree_invalidate_path(istate, ce->name);
975 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
977 /* existing match? Just replace it. */
978 if (pos >= 0) {
979 if (!new_only)
980 replace_index_entry(istate, pos, ce);
981 return 0;
983 pos = -pos-1;
985 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
986 untracked_cache_add_to_index(istate, ce->name);
989 * Inserting a merged entry ("stage 0") into the index
990 * will always replace all non-merged entries..
992 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
993 while (ce_same_name(istate->cache[pos], ce)) {
994 ok_to_add = 1;
995 if (!remove_index_entry_at(istate, pos))
996 break;
1000 if (!ok_to_add)
1001 return -1;
1002 if (!verify_path(ce->name))
1003 return error("Invalid path '%s'", ce->name);
1005 if (!skip_df_check &&
1006 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
1007 if (!ok_to_replace)
1008 return error("'%s' appears as both a file and as a directory",
1009 ce->name);
1010 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
1011 pos = -pos-1;
1013 return pos + 1;
1016 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1018 int pos;
1020 if (option & ADD_CACHE_JUST_APPEND)
1021 pos = istate->cache_nr;
1022 else {
1023 int ret;
1024 ret = add_index_entry_with_check(istate, ce, option);
1025 if (ret <= 0)
1026 return ret;
1027 pos = ret - 1;
1030 /* Make sure the array is big enough .. */
1031 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1033 /* Add it in.. */
1034 istate->cache_nr++;
1035 if (istate->cache_nr > pos + 1)
1036 memmove(istate->cache + pos + 1,
1037 istate->cache + pos,
1038 (istate->cache_nr - pos - 1) * sizeof(ce));
1039 set_index_entry(istate, pos, ce);
1040 istate->cache_changed |= CE_ENTRY_ADDED;
1041 return 0;
1045 * "refresh" does not calculate a new sha1 file or bring the
1046 * cache up-to-date for mode/content changes. But what it
1047 * _does_ do is to "re-match" the stat information of a file
1048 * with the cache, so that you can refresh the cache for a
1049 * file that hasn't been changed but where the stat entry is
1050 * out of date.
1052 * For example, you'd want to do this after doing a "git-read-tree",
1053 * to link up the stat cache details with the proper files.
1055 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1056 struct cache_entry *ce,
1057 unsigned int options, int *err,
1058 int *changed_ret)
1060 struct stat st;
1061 struct cache_entry *updated;
1062 int changed, size;
1063 int refresh = options & CE_MATCH_REFRESH;
1064 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1065 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1066 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1068 if (!refresh || ce_uptodate(ce))
1069 return ce;
1072 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1073 * that the change to the work tree does not matter and told
1074 * us not to worry.
1076 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1077 ce_mark_uptodate(ce);
1078 return ce;
1080 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1081 ce_mark_uptodate(ce);
1082 return ce;
1085 if (has_symlink_leading_path(ce->name, ce_namelen(ce))) {
1086 if (ignore_missing)
1087 return ce;
1088 if (err)
1089 *err = ENOENT;
1090 return NULL;
1093 if (lstat(ce->name, &st) < 0) {
1094 if (ignore_missing && errno == ENOENT)
1095 return ce;
1096 if (err)
1097 *err = errno;
1098 return NULL;
1101 changed = ie_match_stat(istate, ce, &st, options);
1102 if (changed_ret)
1103 *changed_ret = changed;
1104 if (!changed) {
1106 * The path is unchanged. If we were told to ignore
1107 * valid bit, then we did the actual stat check and
1108 * found that the entry is unmodified. If the entry
1109 * is not marked VALID, this is the place to mark it
1110 * valid again, under "assume unchanged" mode.
1112 if (ignore_valid && assume_unchanged &&
1113 !(ce->ce_flags & CE_VALID))
1114 ; /* mark this one VALID again */
1115 else {
1117 * We do not mark the index itself "modified"
1118 * because CE_UPTODATE flag is in-core only;
1119 * we are not going to write this change out.
1121 if (!S_ISGITLINK(ce->ce_mode))
1122 ce_mark_uptodate(ce);
1123 return ce;
1127 if (ie_modified(istate, ce, &st, options)) {
1128 if (err)
1129 *err = EINVAL;
1130 return NULL;
1133 size = ce_size(ce);
1134 updated = xmalloc(size);
1135 memcpy(updated, ce, size);
1136 fill_stat_cache_info(updated, &st);
1138 * If ignore_valid is not set, we should leave CE_VALID bit
1139 * alone. Otherwise, paths marked with --no-assume-unchanged
1140 * (i.e. things to be edited) will reacquire CE_VALID bit
1141 * automatically, which is not really what we want.
1143 if (!ignore_valid && assume_unchanged &&
1144 !(ce->ce_flags & CE_VALID))
1145 updated->ce_flags &= ~CE_VALID;
1147 /* istate->cache_changed is updated in the caller */
1148 return updated;
1151 static void show_file(const char * fmt, const char * name, int in_porcelain,
1152 int * first, const char *header_msg)
1154 if (in_porcelain && *first && header_msg) {
1155 printf("%s\n", header_msg);
1156 *first = 0;
1158 printf(fmt, name);
1161 int refresh_index(struct index_state *istate, unsigned int flags,
1162 const struct pathspec *pathspec,
1163 char *seen, const char *header_msg)
1165 int i;
1166 int has_errors = 0;
1167 int really = (flags & REFRESH_REALLY) != 0;
1168 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1169 int quiet = (flags & REFRESH_QUIET) != 0;
1170 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1171 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1172 int first = 1;
1173 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1174 unsigned int options = (CE_MATCH_REFRESH |
1175 (really ? CE_MATCH_IGNORE_VALID : 0) |
1176 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1177 const char *modified_fmt;
1178 const char *deleted_fmt;
1179 const char *typechange_fmt;
1180 const char *added_fmt;
1181 const char *unmerged_fmt;
1183 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1184 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1185 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1186 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1187 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1188 for (i = 0; i < istate->cache_nr; i++) {
1189 struct cache_entry *ce, *new;
1190 int cache_errno = 0;
1191 int changed = 0;
1192 int filtered = 0;
1194 ce = istate->cache[i];
1195 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1196 continue;
1198 if (pathspec && !ce_path_match(ce, pathspec, seen))
1199 filtered = 1;
1201 if (ce_stage(ce)) {
1202 while ((i < istate->cache_nr) &&
1203 ! strcmp(istate->cache[i]->name, ce->name))
1204 i++;
1205 i--;
1206 if (allow_unmerged)
1207 continue;
1208 if (!filtered)
1209 show_file(unmerged_fmt, ce->name, in_porcelain,
1210 &first, header_msg);
1211 has_errors = 1;
1212 continue;
1215 if (filtered)
1216 continue;
1218 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1219 if (new == ce)
1220 continue;
1221 if (!new) {
1222 const char *fmt;
1224 if (really && cache_errno == EINVAL) {
1225 /* If we are doing --really-refresh that
1226 * means the index is not valid anymore.
1228 ce->ce_flags &= ~CE_VALID;
1229 ce->ce_flags |= CE_UPDATE_IN_BASE;
1230 istate->cache_changed |= CE_ENTRY_CHANGED;
1232 if (quiet)
1233 continue;
1235 if (cache_errno == ENOENT)
1236 fmt = deleted_fmt;
1237 else if (ce_intent_to_add(ce))
1238 fmt = added_fmt; /* must be before other checks */
1239 else if (changed & TYPE_CHANGED)
1240 fmt = typechange_fmt;
1241 else
1242 fmt = modified_fmt;
1243 show_file(fmt,
1244 ce->name, in_porcelain, &first, header_msg);
1245 has_errors = 1;
1246 continue;
1249 replace_index_entry(istate, i, new);
1251 return has_errors;
1254 struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1255 unsigned int options)
1257 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1261 /*****************************************************************
1262 * Index File I/O
1263 *****************************************************************/
1265 #define INDEX_FORMAT_DEFAULT 3
1267 static unsigned int get_index_format_default(void)
1269 char *envversion = getenv("GIT_INDEX_VERSION");
1270 char *endp;
1271 int value;
1272 unsigned int version = INDEX_FORMAT_DEFAULT;
1274 if (!envversion) {
1275 if (!git_config_get_int("index.version", &value))
1276 version = value;
1277 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1278 warning(_("index.version set, but the value is invalid.\n"
1279 "Using version %i"), INDEX_FORMAT_DEFAULT);
1280 return INDEX_FORMAT_DEFAULT;
1282 return version;
1285 version = strtoul(envversion, &endp, 10);
1286 if (*endp ||
1287 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1288 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1289 "Using version %i"), INDEX_FORMAT_DEFAULT);
1290 version = INDEX_FORMAT_DEFAULT;
1292 return version;
1296 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1297 * Again - this is just a (very strong in practice) heuristic that
1298 * the inode hasn't changed.
1300 * We save the fields in big-endian order to allow using the
1301 * index file over NFS transparently.
1303 struct ondisk_cache_entry {
1304 struct cache_time ctime;
1305 struct cache_time mtime;
1306 uint32_t dev;
1307 uint32_t ino;
1308 uint32_t mode;
1309 uint32_t uid;
1310 uint32_t gid;
1311 uint32_t size;
1312 unsigned char sha1[20];
1313 uint16_t flags;
1314 char name[FLEX_ARRAY]; /* more */
1318 * This struct is used when CE_EXTENDED bit is 1
1319 * The struct must match ondisk_cache_entry exactly from
1320 * ctime till flags
1322 struct ondisk_cache_entry_extended {
1323 struct cache_time ctime;
1324 struct cache_time mtime;
1325 uint32_t dev;
1326 uint32_t ino;
1327 uint32_t mode;
1328 uint32_t uid;
1329 uint32_t gid;
1330 uint32_t size;
1331 unsigned char sha1[20];
1332 uint16_t flags;
1333 uint16_t flags2;
1334 char name[FLEX_ARRAY]; /* more */
1337 /* These are only used for v3 or lower */
1338 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1339 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1340 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1341 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1342 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1343 ondisk_cache_entry_size(ce_namelen(ce)))
1345 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1347 git_SHA_CTX c;
1348 unsigned char sha1[20];
1349 int hdr_version;
1351 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1352 return error("bad signature");
1353 hdr_version = ntohl(hdr->hdr_version);
1354 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1355 return error("bad index version %d", hdr_version);
1356 git_SHA1_Init(&c);
1357 git_SHA1_Update(&c, hdr, size - 20);
1358 git_SHA1_Final(sha1, &c);
1359 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1360 return error("bad index file sha1 signature");
1361 return 0;
1364 static int read_index_extension(struct index_state *istate,
1365 const char *ext, void *data, unsigned long sz)
1367 switch (CACHE_EXT(ext)) {
1368 case CACHE_EXT_TREE:
1369 istate->cache_tree = cache_tree_read(data, sz);
1370 break;
1371 case CACHE_EXT_RESOLVE_UNDO:
1372 istate->resolve_undo = resolve_undo_read(data, sz);
1373 break;
1374 case CACHE_EXT_LINK:
1375 if (read_link_extension(istate, data, sz))
1376 return -1;
1377 break;
1378 case CACHE_EXT_UNTRACKED:
1379 istate->untracked = read_untracked_extension(data, sz);
1380 break;
1381 default:
1382 if (*ext < 'A' || 'Z' < *ext)
1383 return error("index uses %.4s extension, which we do not understand",
1384 ext);
1385 fprintf(stderr, "ignoring %.4s extension\n", ext);
1386 break;
1388 return 0;
1391 int hold_locked_index(struct lock_file *lk, int die_on_error)
1393 return hold_lock_file_for_update(lk, get_index_file(),
1394 die_on_error
1395 ? LOCK_DIE_ON_ERROR
1396 : 0);
1399 int read_index(struct index_state *istate)
1401 return read_index_from(istate, get_index_file());
1404 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1405 unsigned int flags,
1406 const char *name,
1407 size_t len)
1409 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1411 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1412 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1413 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1414 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1415 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1416 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1417 ce->ce_mode = get_be32(&ondisk->mode);
1418 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1419 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1420 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1421 ce->ce_flags = flags & ~CE_NAMEMASK;
1422 ce->ce_namelen = len;
1423 ce->index = 0;
1424 hashcpy(ce->sha1, ondisk->sha1);
1425 memcpy(ce->name, name, len);
1426 ce->name[len] = '\0';
1427 return ce;
1431 * Adjacent cache entries tend to share the leading paths, so it makes
1432 * sense to only store the differences in later entries. In the v4
1433 * on-disk format of the index, each on-disk cache entry stores the
1434 * number of bytes to be stripped from the end of the previous name,
1435 * and the bytes to append to the result, to come up with its name.
1437 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1439 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1440 size_t len = decode_varint(&cp);
1442 if (name->len < len)
1443 die("malformed name field in the index");
1444 strbuf_remove(name, name->len - len, len);
1445 for (ep = cp; *ep; ep++)
1446 ; /* find the end */
1447 strbuf_add(name, cp, ep - cp);
1448 return (const char *)ep + 1 - cp_;
1451 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1452 unsigned long *ent_size,
1453 struct strbuf *previous_name)
1455 struct cache_entry *ce;
1456 size_t len;
1457 const char *name;
1458 unsigned int flags;
1460 /* On-disk flags are just 16 bits */
1461 flags = get_be16(&ondisk->flags);
1462 len = flags & CE_NAMEMASK;
1464 if (flags & CE_EXTENDED) {
1465 struct ondisk_cache_entry_extended *ondisk2;
1466 int extended_flags;
1467 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1468 extended_flags = get_be16(&ondisk2->flags2) << 16;
1469 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1470 if (extended_flags & ~CE_EXTENDED_FLAGS)
1471 die("Unknown index entry format %08x", extended_flags);
1472 flags |= extended_flags;
1473 name = ondisk2->name;
1475 else
1476 name = ondisk->name;
1478 if (!previous_name) {
1479 /* v3 and earlier */
1480 if (len == CE_NAMEMASK)
1481 len = strlen(name);
1482 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1484 *ent_size = ondisk_ce_size(ce);
1485 } else {
1486 unsigned long consumed;
1487 consumed = expand_name_field(previous_name, name);
1488 ce = cache_entry_from_ondisk(ondisk, flags,
1489 previous_name->buf,
1490 previous_name->len);
1492 *ent_size = (name - ((char *)ondisk)) + consumed;
1494 return ce;
1497 static void check_ce_order(struct index_state *istate)
1499 unsigned int i;
1501 for (i = 1; i < istate->cache_nr; i++) {
1502 struct cache_entry *ce = istate->cache[i - 1];
1503 struct cache_entry *next_ce = istate->cache[i];
1504 int name_compare = strcmp(ce->name, next_ce->name);
1506 if (0 < name_compare)
1507 die("unordered stage entries in index");
1508 if (!name_compare) {
1509 if (!ce_stage(ce))
1510 die("multiple stage entries for merged file '%s'",
1511 ce->name);
1512 if (ce_stage(ce) > ce_stage(next_ce))
1513 die("unordered stage entries for '%s'",
1514 ce->name);
1519 static void tweak_untracked_cache(struct index_state *istate)
1521 switch (git_config_get_untracked_cache()) {
1522 case -1: /* keep: do nothing */
1523 break;
1524 case 0: /* false */
1525 remove_untracked_cache(istate);
1526 break;
1527 case 1: /* true */
1528 add_untracked_cache(istate);
1529 break;
1530 default: /* unknown value: do nothing */
1531 break;
1535 static void post_read_index_from(struct index_state *istate)
1537 check_ce_order(istate);
1538 tweak_untracked_cache(istate);
1541 /* remember to discard_cache() before reading a different cache! */
1542 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1544 int fd, i;
1545 struct stat st;
1546 unsigned long src_offset;
1547 struct cache_header *hdr;
1548 void *mmap;
1549 size_t mmap_size;
1550 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1552 if (istate->initialized)
1553 return istate->cache_nr;
1555 istate->timestamp.sec = 0;
1556 istate->timestamp.nsec = 0;
1557 fd = open(path, O_RDONLY);
1558 if (fd < 0) {
1559 if (!must_exist && errno == ENOENT)
1560 return 0;
1561 die_errno("%s: index file open failed", path);
1564 if (fstat(fd, &st))
1565 die_errno("cannot stat the open index");
1567 mmap_size = xsize_t(st.st_size);
1568 if (mmap_size < sizeof(struct cache_header) + 20)
1569 die("index file smaller than expected");
1571 mmap = xmmap(NULL, mmap_size, PROT_READ, MAP_PRIVATE, fd, 0);
1572 if (mmap == MAP_FAILED)
1573 die_errno("unable to map index file");
1574 close(fd);
1576 hdr = mmap;
1577 if (verify_hdr(hdr, mmap_size) < 0)
1578 goto unmap;
1580 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1581 istate->version = ntohl(hdr->hdr_version);
1582 istate->cache_nr = ntohl(hdr->hdr_entries);
1583 istate->cache_alloc = alloc_nr(istate->cache_nr);
1584 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1585 istate->initialized = 1;
1587 if (istate->version == 4)
1588 previous_name = &previous_name_buf;
1589 else
1590 previous_name = NULL;
1592 src_offset = sizeof(*hdr);
1593 for (i = 0; i < istate->cache_nr; i++) {
1594 struct ondisk_cache_entry *disk_ce;
1595 struct cache_entry *ce;
1596 unsigned long consumed;
1598 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1599 ce = create_from_disk(disk_ce, &consumed, previous_name);
1600 set_index_entry(istate, i, ce);
1602 src_offset += consumed;
1604 strbuf_release(&previous_name_buf);
1605 istate->timestamp.sec = st.st_mtime;
1606 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1608 while (src_offset <= mmap_size - 20 - 8) {
1609 /* After an array of active_nr index entries,
1610 * there can be arbitrary number of extended
1611 * sections, each of which is prefixed with
1612 * extension name (4-byte) and section length
1613 * in 4-byte network byte order.
1615 uint32_t extsize;
1616 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1617 extsize = ntohl(extsize);
1618 if (read_index_extension(istate,
1619 (const char *) mmap + src_offset,
1620 (char *) mmap + src_offset + 8,
1621 extsize) < 0)
1622 goto unmap;
1623 src_offset += 8;
1624 src_offset += extsize;
1626 munmap(mmap, mmap_size);
1627 return istate->cache_nr;
1629 unmap:
1630 munmap(mmap, mmap_size);
1631 die("index file corrupt");
1634 int read_index_from(struct index_state *istate, const char *path)
1636 struct split_index *split_index;
1637 int ret;
1639 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1640 if (istate->initialized)
1641 return istate->cache_nr;
1643 ret = do_read_index(istate, path, 0);
1645 split_index = istate->split_index;
1646 if (!split_index || is_null_sha1(split_index->base_sha1)) {
1647 post_read_index_from(istate);
1648 return ret;
1651 if (split_index->base)
1652 discard_index(split_index->base);
1653 else
1654 split_index->base = xcalloc(1, sizeof(*split_index->base));
1655 ret = do_read_index(split_index->base,
1656 git_path("sharedindex.%s",
1657 sha1_to_hex(split_index->base_sha1)), 1);
1658 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1659 die("broken index, expect %s in %s, got %s",
1660 sha1_to_hex(split_index->base_sha1),
1661 git_path("sharedindex.%s",
1662 sha1_to_hex(split_index->base_sha1)),
1663 sha1_to_hex(split_index->base->sha1));
1664 merge_base_index(istate);
1665 post_read_index_from(istate);
1666 return ret;
1669 int is_index_unborn(struct index_state *istate)
1671 return (!istate->cache_nr && !istate->timestamp.sec);
1674 int discard_index(struct index_state *istate)
1676 int i;
1678 for (i = 0; i < istate->cache_nr; i++) {
1679 if (istate->cache[i]->index &&
1680 istate->split_index &&
1681 istate->split_index->base &&
1682 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1683 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1684 continue;
1685 free(istate->cache[i]);
1687 resolve_undo_clear_index(istate);
1688 istate->cache_nr = 0;
1689 istate->cache_changed = 0;
1690 istate->timestamp.sec = 0;
1691 istate->timestamp.nsec = 0;
1692 free_name_hash(istate);
1693 cache_tree_free(&(istate->cache_tree));
1694 istate->initialized = 0;
1695 free(istate->cache);
1696 istate->cache = NULL;
1697 istate->cache_alloc = 0;
1698 discard_split_index(istate);
1699 free_untracked_cache(istate->untracked);
1700 istate->untracked = NULL;
1701 return 0;
1704 int unmerged_index(const struct index_state *istate)
1706 int i;
1707 for (i = 0; i < istate->cache_nr; i++) {
1708 if (ce_stage(istate->cache[i]))
1709 return 1;
1711 return 0;
1714 #define WRITE_BUFFER_SIZE 8192
1715 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1716 static unsigned long write_buffer_len;
1718 static int ce_write_flush(git_SHA_CTX *context, int fd)
1720 unsigned int buffered = write_buffer_len;
1721 if (buffered) {
1722 git_SHA1_Update(context, write_buffer, buffered);
1723 if (write_in_full(fd, write_buffer, buffered) != buffered)
1724 return -1;
1725 write_buffer_len = 0;
1727 return 0;
1730 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1732 while (len) {
1733 unsigned int buffered = write_buffer_len;
1734 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1735 if (partial > len)
1736 partial = len;
1737 memcpy(write_buffer + buffered, data, partial);
1738 buffered += partial;
1739 if (buffered == WRITE_BUFFER_SIZE) {
1740 write_buffer_len = buffered;
1741 if (ce_write_flush(context, fd))
1742 return -1;
1743 buffered = 0;
1745 write_buffer_len = buffered;
1746 len -= partial;
1747 data = (char *) data + partial;
1749 return 0;
1752 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1753 unsigned int ext, unsigned int sz)
1755 ext = htonl(ext);
1756 sz = htonl(sz);
1757 return ((ce_write(context, fd, &ext, 4) < 0) ||
1758 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1761 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1763 unsigned int left = write_buffer_len;
1765 if (left) {
1766 write_buffer_len = 0;
1767 git_SHA1_Update(context, write_buffer, left);
1770 /* Flush first if not enough space for SHA1 signature */
1771 if (left + 20 > WRITE_BUFFER_SIZE) {
1772 if (write_in_full(fd, write_buffer, left) != left)
1773 return -1;
1774 left = 0;
1777 /* Append the SHA1 signature at the end */
1778 git_SHA1_Final(write_buffer + left, context);
1779 hashcpy(sha1, write_buffer + left);
1780 left += 20;
1781 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1784 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1787 * The only thing we care about in this function is to smudge the
1788 * falsely clean entry due to touch-update-touch race, so we leave
1789 * everything else as they are. We are called for entries whose
1790 * ce_stat_data.sd_mtime match the index file mtime.
1792 * Note that this actually does not do much for gitlinks, for
1793 * which ce_match_stat_basic() always goes to the actual
1794 * contents. The caller checks with is_racy_timestamp() which
1795 * always says "no" for gitlinks, so we are not called for them ;-)
1797 struct stat st;
1799 if (lstat(ce->name, &st) < 0)
1800 return;
1801 if (ce_match_stat_basic(ce, &st))
1802 return;
1803 if (ce_modified_check_fs(ce, &st)) {
1804 /* This is "racily clean"; smudge it. Note that this
1805 * is a tricky code. At first glance, it may appear
1806 * that it can break with this sequence:
1808 * $ echo xyzzy >frotz
1809 * $ git-update-index --add frotz
1810 * $ : >frotz
1811 * $ sleep 3
1812 * $ echo filfre >nitfol
1813 * $ git-update-index --add nitfol
1815 * but it does not. When the second update-index runs,
1816 * it notices that the entry "frotz" has the same timestamp
1817 * as index, and if we were to smudge it by resetting its
1818 * size to zero here, then the object name recorded
1819 * in index is the 6-byte file but the cached stat information
1820 * becomes zero --- which would then match what we would
1821 * obtain from the filesystem next time we stat("frotz").
1823 * However, the second update-index, before calling
1824 * this function, notices that the cached size is 6
1825 * bytes and what is on the filesystem is an empty
1826 * file, and never calls us, so the cached size information
1827 * for "frotz" stays 6 which does not match the filesystem.
1829 ce->ce_stat_data.sd_size = 0;
1833 /* Copy miscellaneous fields but not the name */
1834 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1835 struct cache_entry *ce)
1837 short flags;
1839 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1840 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1841 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1842 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1843 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1844 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1845 ondisk->mode = htonl(ce->ce_mode);
1846 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1847 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1848 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1849 hashcpy(ondisk->sha1, ce->sha1);
1851 flags = ce->ce_flags & ~CE_NAMEMASK;
1852 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1853 ondisk->flags = htons(flags);
1854 if (ce->ce_flags & CE_EXTENDED) {
1855 struct ondisk_cache_entry_extended *ondisk2;
1856 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1857 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1858 return ondisk2->name;
1860 else {
1861 return ondisk->name;
1865 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1866 struct strbuf *previous_name)
1868 int size;
1869 struct ondisk_cache_entry *ondisk;
1870 int saved_namelen = saved_namelen; /* compiler workaround */
1871 char *name;
1872 int result;
1874 if (ce->ce_flags & CE_STRIP_NAME) {
1875 saved_namelen = ce_namelen(ce);
1876 ce->ce_namelen = 0;
1879 if (!previous_name) {
1880 size = ondisk_ce_size(ce);
1881 ondisk = xcalloc(1, size);
1882 name = copy_cache_entry_to_ondisk(ondisk, ce);
1883 memcpy(name, ce->name, ce_namelen(ce));
1884 } else {
1885 int common, to_remove, prefix_size;
1886 unsigned char to_remove_vi[16];
1887 for (common = 0;
1888 (ce->name[common] &&
1889 common < previous_name->len &&
1890 ce->name[common] == previous_name->buf[common]);
1891 common++)
1892 ; /* still matching */
1893 to_remove = previous_name->len - common;
1894 prefix_size = encode_varint(to_remove, to_remove_vi);
1896 if (ce->ce_flags & CE_EXTENDED)
1897 size = offsetof(struct ondisk_cache_entry_extended, name);
1898 else
1899 size = offsetof(struct ondisk_cache_entry, name);
1900 size += prefix_size + (ce_namelen(ce) - common + 1);
1902 ondisk = xcalloc(1, size);
1903 name = copy_cache_entry_to_ondisk(ondisk, ce);
1904 memcpy(name, to_remove_vi, prefix_size);
1905 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1907 strbuf_splice(previous_name, common, to_remove,
1908 ce->name + common, ce_namelen(ce) - common);
1910 if (ce->ce_flags & CE_STRIP_NAME) {
1911 ce->ce_namelen = saved_namelen;
1912 ce->ce_flags &= ~CE_STRIP_NAME;
1915 result = ce_write(c, fd, ondisk, size);
1916 free(ondisk);
1917 return result;
1921 * This function verifies if index_state has the correct sha1 of the
1922 * index file. Don't die if we have any other failure, just return 0.
1924 static int verify_index_from(const struct index_state *istate, const char *path)
1926 int fd;
1927 ssize_t n;
1928 struct stat st;
1929 unsigned char sha1[20];
1931 if (!istate->initialized)
1932 return 0;
1934 fd = open(path, O_RDONLY);
1935 if (fd < 0)
1936 return 0;
1938 if (fstat(fd, &st))
1939 goto out;
1941 if (st.st_size < sizeof(struct cache_header) + 20)
1942 goto out;
1944 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1945 if (n != 20)
1946 goto out;
1948 if (hashcmp(istate->sha1, sha1))
1949 goto out;
1951 close(fd);
1952 return 1;
1954 out:
1955 close(fd);
1956 return 0;
1959 static int verify_index(const struct index_state *istate)
1961 return verify_index_from(istate, get_index_file());
1964 static int has_racy_timestamp(struct index_state *istate)
1966 int entries = istate->cache_nr;
1967 int i;
1969 for (i = 0; i < entries; i++) {
1970 struct cache_entry *ce = istate->cache[i];
1971 if (is_racy_timestamp(istate, ce))
1972 return 1;
1974 return 0;
1978 * Opportunistically update the index but do not complain if we can't
1980 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1982 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1983 verify_index(istate) &&
1984 write_locked_index(istate, lockfile, COMMIT_LOCK))
1985 rollback_lock_file(lockfile);
1988 static int do_write_index(struct index_state *istate, int newfd,
1989 int strip_extensions)
1991 git_SHA_CTX c;
1992 struct cache_header hdr;
1993 int i, err, removed, extended, hdr_version;
1994 struct cache_entry **cache = istate->cache;
1995 int entries = istate->cache_nr;
1996 struct stat st;
1997 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1999 for (i = removed = extended = 0; i < entries; i++) {
2000 if (cache[i]->ce_flags & CE_REMOVE)
2001 removed++;
2003 /* reduce extended entries if possible */
2004 cache[i]->ce_flags &= ~CE_EXTENDED;
2005 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
2006 extended++;
2007 cache[i]->ce_flags |= CE_EXTENDED;
2011 if (!istate->version) {
2012 istate->version = get_index_format_default();
2013 if (getenv("GIT_TEST_SPLIT_INDEX"))
2014 init_split_index(istate);
2017 /* demote version 3 to version 2 when the latter suffices */
2018 if (istate->version == 3 || istate->version == 2)
2019 istate->version = extended ? 3 : 2;
2021 hdr_version = istate->version;
2023 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
2024 hdr.hdr_version = htonl(hdr_version);
2025 hdr.hdr_entries = htonl(entries - removed);
2027 git_SHA1_Init(&c);
2028 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
2029 return -1;
2031 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
2032 for (i = 0; i < entries; i++) {
2033 struct cache_entry *ce = cache[i];
2034 if (ce->ce_flags & CE_REMOVE)
2035 continue;
2036 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
2037 ce_smudge_racily_clean_entry(ce);
2038 if (is_null_sha1(ce->sha1)) {
2039 static const char msg[] = "cache entry has null sha1: %s";
2040 static int allow = -1;
2042 if (allow < 0)
2043 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
2044 if (allow)
2045 warning(msg, ce->name);
2046 else
2047 return error(msg, ce->name);
2049 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
2050 return -1;
2052 strbuf_release(&previous_name_buf);
2054 /* Write extension data here */
2055 if (!strip_extensions && istate->split_index) {
2056 struct strbuf sb = STRBUF_INIT;
2058 err = write_link_extension(&sb, istate) < 0 ||
2059 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
2060 sb.len) < 0 ||
2061 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2062 strbuf_release(&sb);
2063 if (err)
2064 return -1;
2066 if (!strip_extensions && istate->cache_tree) {
2067 struct strbuf sb = STRBUF_INIT;
2069 cache_tree_write(&sb, istate->cache_tree);
2070 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
2071 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2072 strbuf_release(&sb);
2073 if (err)
2074 return -1;
2076 if (!strip_extensions && istate->resolve_undo) {
2077 struct strbuf sb = STRBUF_INIT;
2079 resolve_undo_write(&sb, istate->resolve_undo);
2080 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2081 sb.len) < 0
2082 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2083 strbuf_release(&sb);
2084 if (err)
2085 return -1;
2087 if (!strip_extensions && istate->untracked) {
2088 struct strbuf sb = STRBUF_INIT;
2090 write_untracked_extension(&sb, istate->untracked);
2091 err = write_index_ext_header(&c, newfd, CACHE_EXT_UNTRACKED,
2092 sb.len) < 0 ||
2093 ce_write(&c, newfd, sb.buf, sb.len) < 0;
2094 strbuf_release(&sb);
2095 if (err)
2096 return -1;
2099 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2100 return -1;
2101 istate->timestamp.sec = (unsigned int)st.st_mtime;
2102 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2103 return 0;
2106 void set_alternate_index_output(const char *name)
2108 alternate_index_output = name;
2111 static int commit_locked_index(struct lock_file *lk)
2113 if (alternate_index_output)
2114 return commit_lock_file_to(lk, alternate_index_output);
2115 else
2116 return commit_lock_file(lk);
2119 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2120 unsigned flags)
2122 int ret = do_write_index(istate, get_lock_file_fd(lock), 0);
2123 if (ret)
2124 return ret;
2125 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2126 (COMMIT_LOCK | CLOSE_LOCK));
2127 if (flags & COMMIT_LOCK)
2128 return commit_locked_index(lock);
2129 else if (flags & CLOSE_LOCK)
2130 return close_lock_file(lock);
2131 else
2132 return ret;
2135 static int write_split_index(struct index_state *istate,
2136 struct lock_file *lock,
2137 unsigned flags)
2139 int ret;
2140 prepare_to_write_split_index(istate);
2141 ret = do_write_locked_index(istate, lock, flags);
2142 finish_writing_split_index(istate);
2143 return ret;
2146 static struct tempfile temporary_sharedindex;
2148 static int write_shared_index(struct index_state *istate,
2149 struct lock_file *lock, unsigned flags)
2151 struct split_index *si = istate->split_index;
2152 int fd, ret;
2154 fd = mks_tempfile(&temporary_sharedindex, git_path("sharedindex_XXXXXX"));
2155 if (fd < 0) {
2156 hashclr(si->base_sha1);
2157 return do_write_locked_index(istate, lock, flags);
2159 move_cache_to_base_index(istate);
2160 ret = do_write_index(si->base, fd, 1);
2161 if (ret) {
2162 delete_tempfile(&temporary_sharedindex);
2163 return ret;
2165 ret = rename_tempfile(&temporary_sharedindex,
2166 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2167 if (!ret)
2168 hashcpy(si->base_sha1, si->base->sha1);
2169 return ret;
2172 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2173 unsigned flags)
2175 struct split_index *si = istate->split_index;
2177 if (!si || alternate_index_output ||
2178 (istate->cache_changed & ~EXTMASK)) {
2179 if (si)
2180 hashclr(si->base_sha1);
2181 return do_write_locked_index(istate, lock, flags);
2184 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2185 int v = si->base_sha1[0];
2186 if ((v & 15) < 6)
2187 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2189 if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2190 int ret = write_shared_index(istate, lock, flags);
2191 if (ret)
2192 return ret;
2195 return write_split_index(istate, lock, flags);
2199 * Read the index file that is potentially unmerged into given
2200 * index_state, dropping any unmerged entries. Returns true if
2201 * the index is unmerged. Callers who want to refuse to work
2202 * from an unmerged state can call this and check its return value,
2203 * instead of calling read_cache().
2205 int read_index_unmerged(struct index_state *istate)
2207 int i;
2208 int unmerged = 0;
2210 read_index(istate);
2211 for (i = 0; i < istate->cache_nr; i++) {
2212 struct cache_entry *ce = istate->cache[i];
2213 struct cache_entry *new_ce;
2214 int size, len;
2216 if (!ce_stage(ce))
2217 continue;
2218 unmerged = 1;
2219 len = ce_namelen(ce);
2220 size = cache_entry_size(len);
2221 new_ce = xcalloc(1, size);
2222 memcpy(new_ce->name, ce->name, len);
2223 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2224 new_ce->ce_namelen = len;
2225 new_ce->ce_mode = ce->ce_mode;
2226 if (add_index_entry(istate, new_ce, 0))
2227 return error("%s: cannot drop to stage #0",
2228 new_ce->name);
2230 return unmerged;
2234 * Returns 1 if the path is an "other" path with respect to
2235 * the index; that is, the path is not mentioned in the index at all,
2236 * either as a file, a directory with some files in the index,
2237 * or as an unmerged entry.
2239 * We helpfully remove a trailing "/" from directories so that
2240 * the output of read_directory can be used as-is.
2242 int index_name_is_other(const struct index_state *istate, const char *name,
2243 int namelen)
2245 int pos;
2246 if (namelen && name[namelen - 1] == '/')
2247 namelen--;
2248 pos = index_name_pos(istate, name, namelen);
2249 if (0 <= pos)
2250 return 0; /* exact match */
2251 pos = -pos - 1;
2252 if (pos < istate->cache_nr) {
2253 struct cache_entry *ce = istate->cache[pos];
2254 if (ce_namelen(ce) == namelen &&
2255 !memcmp(ce->name, name, namelen))
2256 return 0; /* Yup, this one exists unmerged */
2258 return 1;
2261 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2263 int pos, len;
2264 unsigned long sz;
2265 enum object_type type;
2266 void *data;
2268 len = strlen(path);
2269 pos = index_name_pos(istate, path, len);
2270 if (pos < 0) {
2272 * We might be in the middle of a merge, in which
2273 * case we would read stage #2 (ours).
2275 int i;
2276 for (i = -pos - 1;
2277 (pos < 0 && i < istate->cache_nr &&
2278 !strcmp(istate->cache[i]->name, path));
2279 i++)
2280 if (ce_stage(istate->cache[i]) == 2)
2281 pos = i;
2283 if (pos < 0)
2284 return NULL;
2285 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2286 if (!data || type != OBJ_BLOB) {
2287 free(data);
2288 return NULL;
2290 if (size)
2291 *size = sz;
2292 return data;
2295 void stat_validity_clear(struct stat_validity *sv)
2297 free(sv->sd);
2298 sv->sd = NULL;
2301 int stat_validity_check(struct stat_validity *sv, const char *path)
2303 struct stat st;
2305 if (stat(path, &st) < 0)
2306 return sv->sd == NULL;
2307 if (!sv->sd)
2308 return 0;
2309 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2312 void stat_validity_update(struct stat_validity *sv, int fd)
2314 struct stat st;
2316 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2317 stat_validity_clear(sv);
2318 else {
2319 if (!sv->sd)
2320 sv->sd = xcalloc(1, sizeof(struct stat_data));
2321 fill_stat_data(sv->sd, &st);