msysGit-based Git for Windows 1.x was superseded by Git for Windows 2.x
[git/mingw/4msysgit.git] / read-cache.c
blob5d3c8bd4aaffda9915a3fd62d9d9800f4ac8baff
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 "cache-tree.h"
9 #include "refs.h"
10 #include "dir.h"
11 #include "tree.h"
12 #include "commit.h"
13 #include "blob.h"
14 #include "resolve-undo.h"
15 #include "strbuf.h"
16 #include "varint.h"
17 #include "split-index.h"
18 #include "sigchain.h"
20 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
21 unsigned int options);
23 /* Mask for the name length in ce_flags in the on-disk index */
25 #define CE_NAMEMASK (0x0fff)
27 /* Index extensions.
29 * The first letter should be 'A'..'Z' for extensions that are not
30 * necessary for a correct operation (i.e. optimization data).
31 * When new extensions are added that _needs_ to be understood in
32 * order to correctly interpret the index file, pick character that
33 * is outside the range, to cause the reader to abort.
36 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
37 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
38 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
39 #define CACHE_EXT_LINK 0x6c696e6b /* "link" */
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)
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 remove_index_entry_at(istate, nr);
81 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
84 void fill_stat_data(struct stat_data *sd, struct stat *st)
86 sd->sd_ctime.sec = (unsigned int)st->st_ctime;
87 sd->sd_mtime.sec = (unsigned int)st->st_mtime;
88 sd->sd_ctime.nsec = ST_CTIME_NSEC(*st);
89 sd->sd_mtime.nsec = ST_MTIME_NSEC(*st);
90 sd->sd_dev = st->st_dev;
91 sd->sd_ino = st->st_ino;
92 sd->sd_uid = st->st_uid;
93 sd->sd_gid = st->st_gid;
94 sd->sd_size = st->st_size;
97 int match_stat_data(const struct stat_data *sd, struct stat *st)
99 int changed = 0;
101 if (sd->sd_mtime.sec != (unsigned int)st->st_mtime)
102 changed |= MTIME_CHANGED;
103 if (trust_ctime && check_stat &&
104 sd->sd_ctime.sec != (unsigned int)st->st_ctime)
105 changed |= CTIME_CHANGED;
107 #ifdef USE_NSEC
108 if (check_stat && sd->sd_mtime.nsec != ST_MTIME_NSEC(*st))
109 changed |= MTIME_CHANGED;
110 if (trust_ctime && check_stat &&
111 sd->sd_ctime.nsec != ST_CTIME_NSEC(*st))
112 changed |= CTIME_CHANGED;
113 #endif
115 if (check_stat) {
116 if (sd->sd_uid != (unsigned int) st->st_uid ||
117 sd->sd_gid != (unsigned int) st->st_gid)
118 changed |= OWNER_CHANGED;
119 if (sd->sd_ino != (unsigned int) st->st_ino)
120 changed |= INODE_CHANGED;
123 #ifdef USE_STDEV
125 * st_dev breaks on network filesystems where different
126 * clients will have different views of what "device"
127 * the filesystem is on
129 if (check_stat && sd->sd_dev != (unsigned int) st->st_dev)
130 changed |= INODE_CHANGED;
131 #endif
133 if (sd->sd_size != (unsigned int) st->st_size)
134 changed |= DATA_CHANGED;
136 return changed;
140 * This only updates the "non-critical" parts of the directory
141 * cache, ie the parts that aren't tracked by GIT, and only used
142 * to validate the cache.
144 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
146 fill_stat_data(&ce->ce_stat_data, st);
148 if (assume_unchanged)
149 ce->ce_flags |= CE_VALID;
151 if (S_ISREG(st->st_mode))
152 ce_mark_uptodate(ce);
155 static int ce_compare_data(const struct cache_entry *ce, struct stat *st)
157 int match = -1;
158 int fd = open(ce->name, O_RDONLY);
160 if (fd >= 0) {
161 unsigned char sha1[20];
162 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
163 match = hashcmp(sha1, ce->sha1);
164 /* index_fd() closed the file descriptor already */
166 return match;
169 static int ce_compare_link(const struct cache_entry *ce, size_t expected_size)
171 int match = -1;
172 void *buffer;
173 unsigned long size;
174 enum object_type type;
175 struct strbuf sb = STRBUF_INIT;
177 if (strbuf_readlink(&sb, ce->name, expected_size))
178 return -1;
180 buffer = read_sha1_file(ce->sha1, &type, &size);
181 if (buffer) {
182 if (size == sb.len)
183 match = memcmp(buffer, sb.buf, size);
184 free(buffer);
186 strbuf_release(&sb);
187 return match;
190 static int ce_compare_gitlink(const struct cache_entry *ce)
192 unsigned char sha1[20];
195 * We don't actually require that the .git directory
196 * under GITLINK directory be a valid git directory. It
197 * might even be missing (in case nobody populated that
198 * sub-project).
200 * If so, we consider it always to match.
202 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
203 return 0;
204 return hashcmp(sha1, ce->sha1);
207 static int ce_modified_check_fs(const struct cache_entry *ce, struct stat *st)
209 switch (st->st_mode & S_IFMT) {
210 case S_IFREG:
211 if (ce_compare_data(ce, st))
212 return DATA_CHANGED;
213 break;
214 case S_IFLNK:
215 if (ce_compare_link(ce, xsize_t(st->st_size)))
216 return DATA_CHANGED;
217 break;
218 case S_IFDIR:
219 if (S_ISGITLINK(ce->ce_mode))
220 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
221 default:
222 return TYPE_CHANGED;
224 return 0;
227 static int ce_match_stat_basic(const struct cache_entry *ce, struct stat *st)
229 unsigned int changed = 0;
231 if (ce->ce_flags & CE_REMOVE)
232 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
234 switch (ce->ce_mode & S_IFMT) {
235 case S_IFREG:
236 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
237 /* We consider only the owner x bit to be relevant for
238 * "mode changes"
240 if (trust_executable_bit &&
241 (0100 & (ce->ce_mode ^ st->st_mode)))
242 changed |= MODE_CHANGED;
243 break;
244 case S_IFLNK:
245 if (!S_ISLNK(st->st_mode) &&
246 (has_symlinks || !S_ISREG(st->st_mode)))
247 changed |= TYPE_CHANGED;
248 break;
249 case S_IFGITLINK:
250 /* We ignore most of the st_xxx fields for gitlinks */
251 if (!S_ISDIR(st->st_mode))
252 changed |= TYPE_CHANGED;
253 else if (ce_compare_gitlink(ce))
254 changed |= DATA_CHANGED;
255 return changed;
256 default:
257 die("internal error: ce_mode is %o", ce->ce_mode);
260 changed |= match_stat_data(&ce->ce_stat_data, st);
262 /* Racily smudged entry? */
263 if (!ce->ce_stat_data.sd_size) {
264 if (!is_empty_blob_sha1(ce->sha1))
265 changed |= DATA_CHANGED;
268 return changed;
271 static int is_racy_timestamp(const struct index_state *istate,
272 const struct cache_entry *ce)
274 return (!S_ISGITLINK(ce->ce_mode) &&
275 istate->timestamp.sec &&
276 #ifdef USE_NSEC
277 /* nanosecond timestamped files can also be racy! */
278 (istate->timestamp.sec < ce->ce_stat_data.sd_mtime.sec ||
279 (istate->timestamp.sec == ce->ce_stat_data.sd_mtime.sec &&
280 istate->timestamp.nsec <= ce->ce_stat_data.sd_mtime.nsec))
281 #else
282 istate->timestamp.sec <= ce->ce_stat_data.sd_mtime.sec
283 #endif
287 int ie_match_stat(const struct index_state *istate,
288 const struct cache_entry *ce, struct stat *st,
289 unsigned int options)
291 unsigned int changed;
292 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
293 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
294 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
297 * If it's marked as always valid in the index, it's
298 * valid whatever the checked-out copy says.
300 * skip-worktree has the same effect with higher precedence
302 if (!ignore_skip_worktree && ce_skip_worktree(ce))
303 return 0;
304 if (!ignore_valid && (ce->ce_flags & CE_VALID))
305 return 0;
308 * Intent-to-add entries have not been added, so the index entry
309 * by definition never matches what is in the work tree until it
310 * actually gets added.
312 if (ce->ce_flags & CE_INTENT_TO_ADD)
313 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
315 changed = ce_match_stat_basic(ce, st);
318 * Within 1 second of this sequence:
319 * echo xyzzy >file && git-update-index --add file
320 * running this command:
321 * echo frotz >file
322 * would give a falsely clean cache entry. The mtime and
323 * length match the cache, and other stat fields do not change.
325 * We could detect this at update-index time (the cache entry
326 * being registered/updated records the same time as "now")
327 * and delay the return from git-update-index, but that would
328 * effectively mean we can make at most one commit per second,
329 * which is not acceptable. Instead, we check cache entries
330 * whose mtime are the same as the index file timestamp more
331 * carefully than others.
333 if (!changed && is_racy_timestamp(istate, ce)) {
334 if (assume_racy_is_modified)
335 changed |= DATA_CHANGED;
336 else
337 changed |= ce_modified_check_fs(ce, st);
340 return changed;
343 int ie_modified(const struct index_state *istate,
344 const struct cache_entry *ce,
345 struct stat *st, unsigned int options)
347 int changed, changed_fs;
349 changed = ie_match_stat(istate, ce, st, options);
350 if (!changed)
351 return 0;
353 * If the mode or type has changed, there's no point in trying
354 * to refresh the entry - it's not going to match
356 if (changed & (MODE_CHANGED | TYPE_CHANGED))
357 return changed;
360 * Immediately after read-tree or update-index --cacheinfo,
361 * the length field is zero, as we have never even read the
362 * lstat(2) information once, and we cannot trust DATA_CHANGED
363 * returned by ie_match_stat() which in turn was returned by
364 * ce_match_stat_basic() to signal that the filesize of the
365 * blob changed. We have to actually go to the filesystem to
366 * see if the contents match, and if so, should answer "unchanged".
368 * The logic does not apply to gitlinks, as ce_match_stat_basic()
369 * already has checked the actual HEAD from the filesystem in the
370 * subproject. If ie_match_stat() already said it is different,
371 * then we know it is.
373 if ((changed & DATA_CHANGED) &&
374 (S_ISGITLINK(ce->ce_mode) || ce->ce_stat_data.sd_size != 0))
375 return changed;
377 changed_fs = ce_modified_check_fs(ce, st);
378 if (changed_fs)
379 return changed | changed_fs;
380 return 0;
383 int base_name_compare(const char *name1, int len1, int mode1,
384 const char *name2, int len2, int mode2)
386 unsigned char c1, c2;
387 int len = len1 < len2 ? len1 : len2;
388 int cmp;
390 cmp = memcmp(name1, name2, len);
391 if (cmp)
392 return cmp;
393 c1 = name1[len];
394 c2 = name2[len];
395 if (!c1 && S_ISDIR(mode1))
396 c1 = '/';
397 if (!c2 && S_ISDIR(mode2))
398 c2 = '/';
399 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
403 * df_name_compare() is identical to base_name_compare(), except it
404 * compares conflicting directory/file entries as equal. Note that
405 * while a directory name compares as equal to a regular file, they
406 * then individually compare _differently_ to a filename that has
407 * a dot after the basename (because '\0' < '.' < '/').
409 * This is used by routines that want to traverse the git namespace
410 * but then handle conflicting entries together when possible.
412 int df_name_compare(const char *name1, int len1, int mode1,
413 const char *name2, int len2, int mode2)
415 int len = len1 < len2 ? len1 : len2, cmp;
416 unsigned char c1, c2;
418 cmp = memcmp(name1, name2, len);
419 if (cmp)
420 return cmp;
421 /* Directories and files compare equal (same length, same name) */
422 if (len1 == len2)
423 return 0;
424 c1 = name1[len];
425 if (!c1 && S_ISDIR(mode1))
426 c1 = '/';
427 c2 = name2[len];
428 if (!c2 && S_ISDIR(mode2))
429 c2 = '/';
430 if (c1 == '/' && !c2)
431 return 0;
432 if (c2 == '/' && !c1)
433 return 0;
434 return c1 - c2;
437 int name_compare(const char *name1, size_t len1, const char *name2, size_t len2)
439 size_t min_len = (len1 < len2) ? len1 : len2;
440 int cmp = memcmp(name1, name2, min_len);
441 if (cmp)
442 return cmp;
443 if (len1 < len2)
444 return -1;
445 if (len1 > len2)
446 return 1;
447 return 0;
450 int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
452 int cmp;
454 cmp = name_compare(name1, len1, name2, len2);
455 if (cmp)
456 return cmp;
458 if (stage1 < stage2)
459 return -1;
460 if (stage1 > stage2)
461 return 1;
462 return 0;
465 static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
467 int first, last;
469 first = 0;
470 last = istate->cache_nr;
471 while (last > first) {
472 int next = (last + first) >> 1;
473 struct cache_entry *ce = istate->cache[next];
474 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
475 if (!cmp)
476 return next;
477 if (cmp < 0) {
478 last = next;
479 continue;
481 first = next+1;
483 return -first-1;
486 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
488 return index_name_stage_pos(istate, name, namelen, 0);
491 /* Remove entry, return true if there are more entries to go.. */
492 int remove_index_entry_at(struct index_state *istate, int pos)
494 struct cache_entry *ce = istate->cache[pos];
496 record_resolve_undo(istate, ce);
497 remove_name_hash(istate, ce);
498 save_or_free_index_entry(istate, ce);
499 istate->cache_changed |= CE_ENTRY_REMOVED;
500 istate->cache_nr--;
501 if (pos >= istate->cache_nr)
502 return 0;
503 memmove(istate->cache + pos,
504 istate->cache + pos + 1,
505 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
506 return 1;
510 * Remove all cache entries marked for removal, that is where
511 * CE_REMOVE is set in ce_flags. This is much more effective than
512 * calling remove_index_entry_at() for each entry to be removed.
514 void remove_marked_cache_entries(struct index_state *istate)
516 struct cache_entry **ce_array = istate->cache;
517 unsigned int i, j;
519 for (i = j = 0; i < istate->cache_nr; i++) {
520 if (ce_array[i]->ce_flags & CE_REMOVE) {
521 remove_name_hash(istate, ce_array[i]);
522 save_or_free_index_entry(istate, ce_array[i]);
524 else
525 ce_array[j++] = ce_array[i];
527 if (j == istate->cache_nr)
528 return;
529 istate->cache_changed |= CE_ENTRY_REMOVED;
530 istate->cache_nr = j;
533 int remove_file_from_index(struct index_state *istate, const char *path)
535 int pos = index_name_pos(istate, path, strlen(path));
536 if (pos < 0)
537 pos = -pos-1;
538 cache_tree_invalidate_path(istate, path);
539 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
540 remove_index_entry_at(istate, pos);
541 return 0;
544 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
546 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
549 static int index_name_pos_also_unmerged(struct index_state *istate,
550 const char *path, int namelen)
552 int pos = index_name_pos(istate, path, namelen);
553 struct cache_entry *ce;
555 if (pos >= 0)
556 return pos;
558 /* maybe unmerged? */
559 pos = -1 - pos;
560 if (pos >= istate->cache_nr ||
561 compare_name((ce = istate->cache[pos]), path, namelen))
562 return -1;
564 /* order of preference: stage 2, 1, 3 */
565 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
566 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
567 !compare_name(ce, path, namelen))
568 pos++;
569 return pos;
572 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
574 int len = ce_namelen(ce);
575 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
579 * If we add a filename that aliases in the cache, we will use the
580 * name that we already have - but we don't want to update the same
581 * alias twice, because that implies that there were actually two
582 * different files with aliasing names!
584 * So we use the CE_ADDED flag to verify that the alias was an old
585 * one before we accept it as
587 static struct cache_entry *create_alias_ce(struct index_state *istate,
588 struct cache_entry *ce,
589 struct cache_entry *alias)
591 int len;
592 struct cache_entry *new;
594 if (alias->ce_flags & CE_ADDED)
595 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
597 /* Ok, create the new entry using the name of the existing alias */
598 len = ce_namelen(alias);
599 new = xcalloc(1, cache_entry_size(len));
600 memcpy(new->name, alias->name, len);
601 copy_cache_entry(new, ce);
602 save_or_free_index_entry(istate, ce);
603 return new;
606 void set_object_name_for_intent_to_add_entry(struct cache_entry *ce)
608 unsigned char sha1[20];
609 if (write_sha1_file("", 0, blob_type, sha1))
610 die("cannot create an empty blob in the object database");
611 hashcpy(ce->sha1, sha1);
614 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
616 int size, namelen, was_same;
617 mode_t st_mode = st->st_mode;
618 struct cache_entry *ce, *alias;
619 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
620 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
621 int pretend = flags & ADD_CACHE_PRETEND;
622 int intent_only = flags & ADD_CACHE_INTENT;
623 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
624 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
626 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
627 return error("%s: can only add regular files, symbolic links or git-directories", path);
629 namelen = strlen(path);
630 if (S_ISDIR(st_mode)) {
631 while (namelen && path[namelen-1] == '/')
632 namelen--;
634 size = cache_entry_size(namelen);
635 ce = xcalloc(1, size);
636 memcpy(ce->name, path, namelen);
637 ce->ce_namelen = namelen;
638 if (!intent_only)
639 fill_stat_cache_info(ce, st);
640 else
641 ce->ce_flags |= CE_INTENT_TO_ADD;
643 if (trust_executable_bit && has_symlinks)
644 ce->ce_mode = create_ce_mode(st_mode);
645 else {
646 /* If there is an existing entry, pick the mode bits and type
647 * from it, otherwise assume unexecutable regular file.
649 struct cache_entry *ent;
650 int pos = index_name_pos_also_unmerged(istate, path, namelen);
652 ent = (0 <= pos) ? istate->cache[pos] : NULL;
653 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
656 /* When core.ignorecase=true, determine if a directory of the same name but differing
657 * case already exists within the Git repository. If it does, ensure the directory
658 * case of the file being added to the repository matches (is folded into) the existing
659 * entry's directory case.
661 if (ignore_case) {
662 const char *startPtr = ce->name;
663 const char *ptr = startPtr;
664 while (*ptr) {
665 while (*ptr && *ptr != '/')
666 ++ptr;
667 if (*ptr == '/') {
668 struct cache_entry *foundce;
669 ++ptr;
670 foundce = index_dir_exists(istate, ce->name, ptr - ce->name - 1);
671 if (foundce) {
672 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
673 startPtr = ptr;
679 alias = index_file_exists(istate, ce->name, ce_namelen(ce), ignore_case);
680 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
681 /* Nothing changed, really */
682 free(ce);
683 if (!S_ISGITLINK(alias->ce_mode))
684 ce_mark_uptodate(alias);
685 alias->ce_flags |= CE_ADDED;
686 return 0;
688 if (!intent_only) {
689 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
690 return error("unable to index file %s", path);
691 } else
692 set_object_name_for_intent_to_add_entry(ce);
694 if (ignore_case && alias && different_name(ce, alias))
695 ce = create_alias_ce(istate, ce, alias);
696 ce->ce_flags |= CE_ADDED;
698 /* It was suspected to be racily clean, but it turns out to be Ok */
699 was_same = (alias &&
700 !ce_stage(alias) &&
701 !hashcmp(alias->sha1, ce->sha1) &&
702 ce->ce_mode == alias->ce_mode);
704 if (pretend)
706 else if (add_index_entry(istate, ce, add_option))
707 return error("unable to add %s to index",path);
708 if (verbose && !was_same)
709 printf("add '%s'\n", path);
710 return 0;
713 int add_file_to_index(struct index_state *istate, const char *path, int flags)
715 struct stat st;
716 if (lstat(path, &st))
717 die_errno("unable to stat '%s'", path);
718 return add_to_index(istate, path, &st, flags);
721 struct cache_entry *make_cache_entry(unsigned int mode,
722 const unsigned char *sha1, const char *path, int stage,
723 unsigned int refresh_options)
725 int size, len;
726 struct cache_entry *ce;
728 if (!verify_path(path)) {
729 error("Invalid path '%s'", path);
730 return NULL;
733 len = strlen(path);
734 size = cache_entry_size(len);
735 ce = xcalloc(1, size);
737 hashcpy(ce->sha1, sha1);
738 memcpy(ce->name, path, len);
739 ce->ce_flags = create_ce_flags(stage);
740 ce->ce_namelen = len;
741 ce->ce_mode = create_ce_mode(mode);
743 return refresh_cache_entry(ce, refresh_options);
746 int ce_same_name(const struct cache_entry *a, const struct cache_entry *b)
748 int len = ce_namelen(a);
749 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
753 * We fundamentally don't like some paths: we don't want
754 * dot or dot-dot anywhere, and for obvious reasons don't
755 * want to recurse into ".git" either.
757 * Also, we don't want double slashes or slashes at the
758 * end that can make pathnames ambiguous.
760 static int verify_dotfile(const char *rest)
763 * The first character was '.', but that
764 * has already been discarded, we now test
765 * the rest.
768 /* "." is not allowed */
769 if (*rest == '\0' || is_dir_sep(*rest))
770 return 0;
772 switch (*rest) {
774 * ".git" followed by NUL or slash is bad. This
775 * shares the path end test with the ".." case.
777 case 'g':
778 if (rest[1] != 'i')
779 break;
780 if (rest[2] != 't')
781 break;
782 rest += 2;
783 /* fallthrough */
784 case '.':
785 if (rest[1] == '\0' || is_dir_sep(rest[1]))
786 return 0;
788 return 1;
791 int verify_path(const char *path)
793 char c;
795 if (has_dos_drive_prefix(path))
796 return 0;
798 goto inside;
799 for (;;) {
800 if (!c)
801 return 1;
802 if (is_dir_sep(c)) {
803 inside:
804 c = *path++;
805 if ((c == '.' && !verify_dotfile(path)) ||
806 is_dir_sep(c) || c == '\0')
807 return 0;
809 c = *path++;
814 * Do we have another file that has the beginning components being a
815 * proper superset of the name we're trying to add?
817 static int has_file_name(struct index_state *istate,
818 const struct cache_entry *ce, int pos, int ok_to_replace)
820 int retval = 0;
821 int len = ce_namelen(ce);
822 int stage = ce_stage(ce);
823 const char *name = ce->name;
825 while (pos < istate->cache_nr) {
826 struct cache_entry *p = istate->cache[pos++];
828 if (len >= ce_namelen(p))
829 break;
830 if (memcmp(name, p->name, len))
831 break;
832 if (ce_stage(p) != stage)
833 continue;
834 if (p->name[len] != '/')
835 continue;
836 if (p->ce_flags & CE_REMOVE)
837 continue;
838 retval = -1;
839 if (!ok_to_replace)
840 break;
841 remove_index_entry_at(istate, --pos);
843 return retval;
847 * Do we have another file with a pathname that is a proper
848 * subset of the name we're trying to add?
850 static int has_dir_name(struct index_state *istate,
851 const struct cache_entry *ce, int pos, int ok_to_replace)
853 int retval = 0;
854 int stage = ce_stage(ce);
855 const char *name = ce->name;
856 const char *slash = name + ce_namelen(ce);
858 for (;;) {
859 int len;
861 for (;;) {
862 if (*--slash == '/')
863 break;
864 if (slash <= ce->name)
865 return retval;
867 len = slash - name;
869 pos = index_name_stage_pos(istate, name, len, stage);
870 if (pos >= 0) {
872 * Found one, but not so fast. This could
873 * be a marker that says "I was here, but
874 * I am being removed". Such an entry is
875 * not a part of the resulting tree, and
876 * it is Ok to have a directory at the same
877 * path.
879 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
880 retval = -1;
881 if (!ok_to_replace)
882 break;
883 remove_index_entry_at(istate, pos);
884 continue;
887 else
888 pos = -pos-1;
891 * Trivial optimization: if we find an entry that
892 * already matches the sub-directory, then we know
893 * we're ok, and we can exit.
895 while (pos < istate->cache_nr) {
896 struct cache_entry *p = istate->cache[pos];
897 if ((ce_namelen(p) <= len) ||
898 (p->name[len] != '/') ||
899 memcmp(p->name, name, len))
900 break; /* not our subdirectory */
901 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
903 * p is at the same stage as our entry, and
904 * is a subdirectory of what we are looking
905 * at, so we cannot have conflicts at our
906 * level or anything shorter.
908 return retval;
909 pos++;
912 return retval;
915 /* We may be in a situation where we already have path/file and path
916 * is being added, or we already have path and path/file is being
917 * added. Either one would result in a nonsense tree that has path
918 * twice when git-write-tree tries to write it out. Prevent it.
920 * If ok-to-replace is specified, we remove the conflicting entries
921 * from the cache so the caller should recompute the insert position.
922 * When this happens, we return non-zero.
924 static int check_file_directory_conflict(struct index_state *istate,
925 const struct cache_entry *ce,
926 int pos, int ok_to_replace)
928 int retval;
931 * When ce is an "I am going away" entry, we allow it to be added
933 if (ce->ce_flags & CE_REMOVE)
934 return 0;
937 * We check if the path is a sub-path of a subsequent pathname
938 * first, since removing those will not change the position
939 * in the array.
941 retval = has_file_name(istate, ce, pos, ok_to_replace);
944 * Then check if the path might have a clashing sub-directory
945 * before it.
947 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
950 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
952 int pos;
953 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
954 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
955 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
956 int new_only = option & ADD_CACHE_NEW_ONLY;
958 if (!(option & ADD_CACHE_KEEP_CACHE_TREE))
959 cache_tree_invalidate_path(istate, ce->name);
960 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
962 /* existing match? Just replace it. */
963 if (pos >= 0) {
964 if (!new_only)
965 replace_index_entry(istate, pos, ce);
966 return 0;
968 pos = -pos-1;
971 * Inserting a merged entry ("stage 0") into the index
972 * will always replace all non-merged entries..
974 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
975 while (ce_same_name(istate->cache[pos], ce)) {
976 ok_to_add = 1;
977 if (!remove_index_entry_at(istate, pos))
978 break;
982 if (!ok_to_add)
983 return -1;
984 if (!verify_path(ce->name))
985 return error("Invalid path '%s'", ce->name);
987 if (!skip_df_check &&
988 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
989 if (!ok_to_replace)
990 return error("'%s' appears as both a file and as a directory",
991 ce->name);
992 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
993 pos = -pos-1;
995 return pos + 1;
998 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
1000 int pos;
1002 if (option & ADD_CACHE_JUST_APPEND)
1003 pos = istate->cache_nr;
1004 else {
1005 int ret;
1006 ret = add_index_entry_with_check(istate, ce, option);
1007 if (ret <= 0)
1008 return ret;
1009 pos = ret - 1;
1012 /* Make sure the array is big enough .. */
1013 ALLOC_GROW(istate->cache, istate->cache_nr + 1, istate->cache_alloc);
1015 /* Add it in.. */
1016 istate->cache_nr++;
1017 if (istate->cache_nr > pos + 1)
1018 memmove(istate->cache + pos + 1,
1019 istate->cache + pos,
1020 (istate->cache_nr - pos - 1) * sizeof(ce));
1021 set_index_entry(istate, pos, ce);
1022 istate->cache_changed |= CE_ENTRY_ADDED;
1023 return 0;
1027 * "refresh" does not calculate a new sha1 file or bring the
1028 * cache up-to-date for mode/content changes. But what it
1029 * _does_ do is to "re-match" the stat information of a file
1030 * with the cache, so that you can refresh the cache for a
1031 * file that hasn't been changed but where the stat entry is
1032 * out of date.
1034 * For example, you'd want to do this after doing a "git-read-tree",
1035 * to link up the stat cache details with the proper files.
1037 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
1038 struct cache_entry *ce,
1039 unsigned int options, int *err,
1040 int *changed_ret)
1042 struct stat st;
1043 struct cache_entry *updated;
1044 int changed, size;
1045 int refresh = options & CE_MATCH_REFRESH;
1046 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1047 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1048 int ignore_missing = options & CE_MATCH_IGNORE_MISSING;
1050 if (!refresh || ce_uptodate(ce))
1051 return ce;
1054 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1055 * that the change to the work tree does not matter and told
1056 * us not to worry.
1058 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1059 ce_mark_uptodate(ce);
1060 return ce;
1062 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1063 ce_mark_uptodate(ce);
1064 return ce;
1067 if (lstat(ce->name, &st) < 0) {
1068 if (ignore_missing && errno == ENOENT)
1069 return ce;
1070 if (err)
1071 *err = errno;
1072 return NULL;
1075 changed = ie_match_stat(istate, ce, &st, options);
1076 if (changed_ret)
1077 *changed_ret = changed;
1078 if (!changed) {
1080 * The path is unchanged. If we were told to ignore
1081 * valid bit, then we did the actual stat check and
1082 * found that the entry is unmodified. If the entry
1083 * is not marked VALID, this is the place to mark it
1084 * valid again, under "assume unchanged" mode.
1086 if (ignore_valid && assume_unchanged &&
1087 !(ce->ce_flags & CE_VALID))
1088 ; /* mark this one VALID again */
1089 else {
1091 * We do not mark the index itself "modified"
1092 * because CE_UPTODATE flag is in-core only;
1093 * we are not going to write this change out.
1095 if (!S_ISGITLINK(ce->ce_mode))
1096 ce_mark_uptodate(ce);
1097 return ce;
1101 if (ie_modified(istate, ce, &st, options)) {
1102 if (err)
1103 *err = EINVAL;
1104 return NULL;
1107 size = ce_size(ce);
1108 updated = xmalloc(size);
1109 memcpy(updated, ce, size);
1110 fill_stat_cache_info(updated, &st);
1112 * If ignore_valid is not set, we should leave CE_VALID bit
1113 * alone. Otherwise, paths marked with --no-assume-unchanged
1114 * (i.e. things to be edited) will reacquire CE_VALID bit
1115 * automatically, which is not really what we want.
1117 if (!ignore_valid && assume_unchanged &&
1118 !(ce->ce_flags & CE_VALID))
1119 updated->ce_flags &= ~CE_VALID;
1121 /* istate->cache_changed is updated in the caller */
1122 return updated;
1125 static void show_file(const char * fmt, const char * name, int in_porcelain,
1126 int * first, const char *header_msg)
1128 if (in_porcelain && *first && header_msg) {
1129 printf("%s\n", header_msg);
1130 *first = 0;
1132 printf(fmt, name);
1135 int refresh_index(struct index_state *istate, unsigned int flags,
1136 const struct pathspec *pathspec,
1137 char *seen, const char *header_msg)
1139 int i;
1140 int has_errors = 0;
1141 int really = (flags & REFRESH_REALLY) != 0;
1142 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1143 int quiet = (flags & REFRESH_QUIET) != 0;
1144 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1145 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1146 int first = 1;
1147 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1148 unsigned int options = (CE_MATCH_REFRESH |
1149 (really ? CE_MATCH_IGNORE_VALID : 0) |
1150 (not_new ? CE_MATCH_IGNORE_MISSING : 0));
1151 const char *modified_fmt;
1152 const char *deleted_fmt;
1153 const char *typechange_fmt;
1154 const char *added_fmt;
1155 const char *unmerged_fmt;
1157 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1158 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1159 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1160 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1161 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1162 for (i = 0; i < istate->cache_nr; i++) {
1163 struct cache_entry *ce, *new;
1164 int cache_errno = 0;
1165 int changed = 0;
1166 int filtered = 0;
1168 ce = istate->cache[i];
1169 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1170 continue;
1172 if (pathspec && !ce_path_match(ce, pathspec, seen))
1173 filtered = 1;
1175 if (ce_stage(ce)) {
1176 while ((i < istate->cache_nr) &&
1177 ! strcmp(istate->cache[i]->name, ce->name))
1178 i++;
1179 i--;
1180 if (allow_unmerged)
1181 continue;
1182 if (!filtered)
1183 show_file(unmerged_fmt, ce->name, in_porcelain,
1184 &first, header_msg);
1185 has_errors = 1;
1186 continue;
1189 if (filtered)
1190 continue;
1192 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1193 if (new == ce)
1194 continue;
1195 if (!new) {
1196 const char *fmt;
1198 if (really && cache_errno == EINVAL) {
1199 /* If we are doing --really-refresh that
1200 * means the index is not valid anymore.
1202 ce->ce_flags &= ~CE_VALID;
1203 ce->ce_flags |= CE_UPDATE_IN_BASE;
1204 istate->cache_changed |= CE_ENTRY_CHANGED;
1206 if (quiet)
1207 continue;
1209 if (cache_errno == ENOENT)
1210 fmt = deleted_fmt;
1211 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1212 fmt = added_fmt; /* must be before other checks */
1213 else if (changed & TYPE_CHANGED)
1214 fmt = typechange_fmt;
1215 else
1216 fmt = modified_fmt;
1217 show_file(fmt,
1218 ce->name, in_porcelain, &first, header_msg);
1219 has_errors = 1;
1220 continue;
1223 replace_index_entry(istate, i, new);
1225 return has_errors;
1228 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce,
1229 unsigned int options)
1231 return refresh_cache_ent(&the_index, ce, options, NULL, NULL);
1235 /*****************************************************************
1236 * Index File I/O
1237 *****************************************************************/
1239 #define INDEX_FORMAT_DEFAULT 3
1241 static int index_format_config(const char *var, const char *value, void *cb)
1243 unsigned int *version = cb;
1244 if (!strcmp(var, "index.version")) {
1245 *version = git_config_int(var, value);
1246 return 0;
1248 return 1;
1251 static unsigned int get_index_format_default(void)
1253 char *envversion = getenv("GIT_INDEX_VERSION");
1254 char *endp;
1255 unsigned int version = INDEX_FORMAT_DEFAULT;
1257 if (!envversion) {
1258 git_config(index_format_config, &version);
1259 if (version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1260 warning(_("index.version set, but the value is invalid.\n"
1261 "Using version %i"), INDEX_FORMAT_DEFAULT);
1262 return INDEX_FORMAT_DEFAULT;
1264 return version;
1267 version = strtoul(envversion, &endp, 10);
1268 if (*endp ||
1269 version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < version) {
1270 warning(_("GIT_INDEX_VERSION set, but the value is invalid.\n"
1271 "Using version %i"), INDEX_FORMAT_DEFAULT);
1272 version = INDEX_FORMAT_DEFAULT;
1274 return version;
1278 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1279 * Again - this is just a (very strong in practice) heuristic that
1280 * the inode hasn't changed.
1282 * We save the fields in big-endian order to allow using the
1283 * index file over NFS transparently.
1285 struct ondisk_cache_entry {
1286 struct cache_time ctime;
1287 struct cache_time mtime;
1288 uint32_t dev;
1289 uint32_t ino;
1290 uint32_t mode;
1291 uint32_t uid;
1292 uint32_t gid;
1293 uint32_t size;
1294 unsigned char sha1[20];
1295 uint16_t flags;
1296 char name[FLEX_ARRAY]; /* more */
1300 * This struct is used when CE_EXTENDED bit is 1
1301 * The struct must match ondisk_cache_entry exactly from
1302 * ctime till flags
1304 struct ondisk_cache_entry_extended {
1305 struct cache_time ctime;
1306 struct cache_time mtime;
1307 uint32_t dev;
1308 uint32_t ino;
1309 uint32_t mode;
1310 uint32_t uid;
1311 uint32_t gid;
1312 uint32_t size;
1313 unsigned char sha1[20];
1314 uint16_t flags;
1315 uint16_t flags2;
1316 char name[FLEX_ARRAY]; /* more */
1319 /* These are only used for v3 or lower */
1320 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1321 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1322 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1323 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1324 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1325 ondisk_cache_entry_size(ce_namelen(ce)))
1327 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1329 git_SHA_CTX c;
1330 unsigned char sha1[20];
1331 int hdr_version;
1333 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1334 return error("bad signature");
1335 hdr_version = ntohl(hdr->hdr_version);
1336 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
1337 return error("bad index version %d", hdr_version);
1338 git_SHA1_Init(&c);
1339 git_SHA1_Update(&c, hdr, size - 20);
1340 git_SHA1_Final(sha1, &c);
1341 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1342 return error("bad index file sha1 signature");
1343 return 0;
1346 static int read_index_extension(struct index_state *istate,
1347 const char *ext, void *data, unsigned long sz)
1349 switch (CACHE_EXT(ext)) {
1350 case CACHE_EXT_TREE:
1351 istate->cache_tree = cache_tree_read(data, sz);
1352 break;
1353 case CACHE_EXT_RESOLVE_UNDO:
1354 istate->resolve_undo = resolve_undo_read(data, sz);
1355 break;
1356 case CACHE_EXT_LINK:
1357 if (read_link_extension(istate, data, sz))
1358 return -1;
1359 break;
1360 default:
1361 if (*ext < 'A' || 'Z' < *ext)
1362 return error("index uses %.4s extension, which we do not understand",
1363 ext);
1364 fprintf(stderr, "ignoring %.4s extension\n", ext);
1365 break;
1367 return 0;
1370 int read_index(struct index_state *istate)
1372 return read_index_from(istate, get_index_file());
1375 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1376 unsigned int flags,
1377 const char *name,
1378 size_t len)
1380 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1382 ce->ce_stat_data.sd_ctime.sec = get_be32(&ondisk->ctime.sec);
1383 ce->ce_stat_data.sd_mtime.sec = get_be32(&ondisk->mtime.sec);
1384 ce->ce_stat_data.sd_ctime.nsec = get_be32(&ondisk->ctime.nsec);
1385 ce->ce_stat_data.sd_mtime.nsec = get_be32(&ondisk->mtime.nsec);
1386 ce->ce_stat_data.sd_dev = get_be32(&ondisk->dev);
1387 ce->ce_stat_data.sd_ino = get_be32(&ondisk->ino);
1388 ce->ce_mode = get_be32(&ondisk->mode);
1389 ce->ce_stat_data.sd_uid = get_be32(&ondisk->uid);
1390 ce->ce_stat_data.sd_gid = get_be32(&ondisk->gid);
1391 ce->ce_stat_data.sd_size = get_be32(&ondisk->size);
1392 ce->ce_flags = flags & ~CE_NAMEMASK;
1393 ce->ce_namelen = len;
1394 ce->index = 0;
1395 hashcpy(ce->sha1, ondisk->sha1);
1396 memcpy(ce->name, name, len);
1397 ce->name[len] = '\0';
1398 return ce;
1402 * Adjacent cache entries tend to share the leading paths, so it makes
1403 * sense to only store the differences in later entries. In the v4
1404 * on-disk format of the index, each on-disk cache entry stores the
1405 * number of bytes to be stripped from the end of the previous name,
1406 * and the bytes to append to the result, to come up with its name.
1408 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1410 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1411 size_t len = decode_varint(&cp);
1413 if (name->len < len)
1414 die("malformed name field in the index");
1415 strbuf_remove(name, name->len - len, len);
1416 for (ep = cp; *ep; ep++)
1417 ; /* find the end */
1418 strbuf_add(name, cp, ep - cp);
1419 return (const char *)ep + 1 - cp_;
1422 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1423 unsigned long *ent_size,
1424 struct strbuf *previous_name)
1426 struct cache_entry *ce;
1427 size_t len;
1428 const char *name;
1429 unsigned int flags;
1431 /* On-disk flags are just 16 bits */
1432 flags = get_be16(&ondisk->flags);
1433 len = flags & CE_NAMEMASK;
1435 if (flags & CE_EXTENDED) {
1436 struct ondisk_cache_entry_extended *ondisk2;
1437 int extended_flags;
1438 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1439 extended_flags = get_be16(&ondisk2->flags2) << 16;
1440 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1441 if (extended_flags & ~CE_EXTENDED_FLAGS)
1442 die("Unknown index entry format %08x", extended_flags);
1443 flags |= extended_flags;
1444 name = ondisk2->name;
1446 else
1447 name = ondisk->name;
1449 if (!previous_name) {
1450 /* v3 and earlier */
1451 if (len == CE_NAMEMASK)
1452 len = strlen(name);
1453 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1455 *ent_size = ondisk_ce_size(ce);
1456 } else {
1457 unsigned long consumed;
1458 consumed = expand_name_field(previous_name, name);
1459 ce = cache_entry_from_ondisk(ondisk, flags,
1460 previous_name->buf,
1461 previous_name->len);
1463 *ent_size = (name - ((char *)ondisk)) + consumed;
1465 return ce;
1468 /* remember to discard_cache() before reading a different cache! */
1469 int do_read_index(struct index_state *istate, const char *path, int must_exist)
1471 int fd, i;
1472 struct stat st;
1473 unsigned long src_offset;
1474 struct cache_header *hdr;
1475 void *mmap;
1476 size_t mmap_size;
1477 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1479 if (istate->initialized)
1480 return istate->cache_nr;
1482 istate->timestamp.sec = 0;
1483 istate->timestamp.nsec = 0;
1484 fd = open(path, O_RDONLY);
1485 if (fd < 0) {
1486 if (!must_exist && errno == ENOENT)
1487 return 0;
1488 die_errno("%s: index file open failed", path);
1491 if (fstat(fd, &st))
1492 die_errno("cannot stat the open index");
1494 mmap_size = xsize_t(st.st_size);
1495 if (mmap_size < sizeof(struct cache_header) + 20)
1496 die("index file smaller than expected");
1498 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1499 if (mmap == MAP_FAILED)
1500 die_errno("unable to map index file");
1501 close(fd);
1503 hdr = mmap;
1504 if (verify_hdr(hdr, mmap_size) < 0)
1505 goto unmap;
1507 hashcpy(istate->sha1, (const unsigned char *)hdr + mmap_size - 20);
1508 istate->version = ntohl(hdr->hdr_version);
1509 istate->cache_nr = ntohl(hdr->hdr_entries);
1510 istate->cache_alloc = alloc_nr(istate->cache_nr);
1511 istate->cache = xcalloc(istate->cache_alloc, sizeof(*istate->cache));
1512 istate->initialized = 1;
1514 if (istate->version == 4)
1515 previous_name = &previous_name_buf;
1516 else
1517 previous_name = NULL;
1519 src_offset = sizeof(*hdr);
1520 for (i = 0; i < istate->cache_nr; i++) {
1521 struct ondisk_cache_entry *disk_ce;
1522 struct cache_entry *ce;
1523 unsigned long consumed;
1525 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1526 ce = create_from_disk(disk_ce, &consumed, previous_name);
1527 set_index_entry(istate, i, ce);
1529 src_offset += consumed;
1531 strbuf_release(&previous_name_buf);
1532 istate->timestamp.sec = st.st_mtime;
1533 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1535 while (src_offset <= mmap_size - 20 - 8) {
1536 /* After an array of active_nr index entries,
1537 * there can be arbitrary number of extended
1538 * sections, each of which is prefixed with
1539 * extension name (4-byte) and section length
1540 * in 4-byte network byte order.
1542 uint32_t extsize;
1543 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1544 extsize = ntohl(extsize);
1545 if (read_index_extension(istate,
1546 (const char *) mmap + src_offset,
1547 (char *) mmap + src_offset + 8,
1548 extsize) < 0)
1549 goto unmap;
1550 src_offset += 8;
1551 src_offset += extsize;
1553 munmap(mmap, mmap_size);
1554 return istate->cache_nr;
1556 unmap:
1557 munmap(mmap, mmap_size);
1558 die("index file corrupt");
1561 int read_index_from(struct index_state *istate, const char *path)
1563 struct split_index *split_index;
1564 int ret;
1566 /* istate->initialized covers both .git/index and .git/sharedindex.xxx */
1567 if (istate->initialized)
1568 return istate->cache_nr;
1570 ret = do_read_index(istate, path, 0);
1571 split_index = istate->split_index;
1572 if (!split_index)
1573 return ret;
1575 if (is_null_sha1(split_index->base_sha1))
1576 return ret;
1578 if (split_index->base)
1579 discard_index(split_index->base);
1580 else
1581 split_index->base = xcalloc(1, sizeof(*split_index->base));
1582 ret = do_read_index(split_index->base,
1583 git_path("sharedindex.%s",
1584 sha1_to_hex(split_index->base_sha1)), 1);
1585 if (hashcmp(split_index->base_sha1, split_index->base->sha1))
1586 die("broken index, expect %s in %s, got %s",
1587 sha1_to_hex(split_index->base_sha1),
1588 git_path("sharedindex.%s",
1589 sha1_to_hex(split_index->base_sha1)),
1590 sha1_to_hex(split_index->base->sha1));
1591 merge_base_index(istate);
1592 return ret;
1595 int is_index_unborn(struct index_state *istate)
1597 return (!istate->cache_nr && !istate->timestamp.sec);
1600 int discard_index(struct index_state *istate)
1602 int i;
1604 for (i = 0; i < istate->cache_nr; i++) {
1605 if (istate->cache[i]->index &&
1606 istate->split_index &&
1607 istate->split_index->base &&
1608 istate->cache[i]->index <= istate->split_index->base->cache_nr &&
1609 istate->cache[i] == istate->split_index->base->cache[istate->cache[i]->index - 1])
1610 continue;
1611 free(istate->cache[i]);
1613 resolve_undo_clear_index(istate);
1614 istate->cache_nr = 0;
1615 istate->cache_changed = 0;
1616 istate->timestamp.sec = 0;
1617 istate->timestamp.nsec = 0;
1618 free_name_hash(istate);
1619 cache_tree_free(&(istate->cache_tree));
1620 istate->initialized = 0;
1621 free(istate->cache);
1622 istate->cache = NULL;
1623 istate->cache_alloc = 0;
1624 discard_split_index(istate);
1625 return 0;
1628 int unmerged_index(const struct index_state *istate)
1630 int i;
1631 for (i = 0; i < istate->cache_nr; i++) {
1632 if (ce_stage(istate->cache[i]))
1633 return 1;
1635 return 0;
1638 #define WRITE_BUFFER_SIZE 8192
1639 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1640 static unsigned long write_buffer_len;
1642 static int ce_write_flush(git_SHA_CTX *context, int fd)
1644 unsigned int buffered = write_buffer_len;
1645 if (buffered) {
1646 git_SHA1_Update(context, write_buffer, buffered);
1647 if (write_in_full(fd, write_buffer, buffered) != buffered)
1648 return -1;
1649 write_buffer_len = 0;
1651 return 0;
1654 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1656 while (len) {
1657 unsigned int buffered = write_buffer_len;
1658 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1659 if (partial > len)
1660 partial = len;
1661 memcpy(write_buffer + buffered, data, partial);
1662 buffered += partial;
1663 if (buffered == WRITE_BUFFER_SIZE) {
1664 write_buffer_len = buffered;
1665 if (ce_write_flush(context, fd))
1666 return -1;
1667 buffered = 0;
1669 write_buffer_len = buffered;
1670 len -= partial;
1671 data = (char *) data + partial;
1673 return 0;
1676 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1677 unsigned int ext, unsigned int sz)
1679 ext = htonl(ext);
1680 sz = htonl(sz);
1681 return ((ce_write(context, fd, &ext, 4) < 0) ||
1682 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1685 static int ce_flush(git_SHA_CTX *context, int fd, unsigned char *sha1)
1687 unsigned int left = write_buffer_len;
1689 if (left) {
1690 write_buffer_len = 0;
1691 git_SHA1_Update(context, write_buffer, left);
1694 /* Flush first if not enough space for SHA1 signature */
1695 if (left + 20 > WRITE_BUFFER_SIZE) {
1696 if (write_in_full(fd, write_buffer, left) != left)
1697 return -1;
1698 left = 0;
1701 /* Append the SHA1 signature at the end */
1702 git_SHA1_Final(write_buffer + left, context);
1703 hashcpy(sha1, write_buffer + left);
1704 left += 20;
1705 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1708 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1711 * The only thing we care about in this function is to smudge the
1712 * falsely clean entry due to touch-update-touch race, so we leave
1713 * everything else as they are. We are called for entries whose
1714 * ce_stat_data.sd_mtime match the index file mtime.
1716 * Note that this actually does not do much for gitlinks, for
1717 * which ce_match_stat_basic() always goes to the actual
1718 * contents. The caller checks with is_racy_timestamp() which
1719 * always says "no" for gitlinks, so we are not called for them ;-)
1721 struct stat st;
1723 if (lstat(ce->name, &st) < 0)
1724 return;
1725 if (ce_match_stat_basic(ce, &st))
1726 return;
1727 if (ce_modified_check_fs(ce, &st)) {
1728 /* This is "racily clean"; smudge it. Note that this
1729 * is a tricky code. At first glance, it may appear
1730 * that it can break with this sequence:
1732 * $ echo xyzzy >frotz
1733 * $ git-update-index --add frotz
1734 * $ : >frotz
1735 * $ sleep 3
1736 * $ echo filfre >nitfol
1737 * $ git-update-index --add nitfol
1739 * but it does not. When the second update-index runs,
1740 * it notices that the entry "frotz" has the same timestamp
1741 * as index, and if we were to smudge it by resetting its
1742 * size to zero here, then the object name recorded
1743 * in index is the 6-byte file but the cached stat information
1744 * becomes zero --- which would then match what we would
1745 * obtain from the filesystem next time we stat("frotz").
1747 * However, the second update-index, before calling
1748 * this function, notices that the cached size is 6
1749 * bytes and what is on the filesystem is an empty
1750 * file, and never calls us, so the cached size information
1751 * for "frotz" stays 6 which does not match the filesystem.
1753 ce->ce_stat_data.sd_size = 0;
1757 /* Copy miscellaneous fields but not the name */
1758 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1759 struct cache_entry *ce)
1761 short flags;
1763 ondisk->ctime.sec = htonl(ce->ce_stat_data.sd_ctime.sec);
1764 ondisk->mtime.sec = htonl(ce->ce_stat_data.sd_mtime.sec);
1765 ondisk->ctime.nsec = htonl(ce->ce_stat_data.sd_ctime.nsec);
1766 ondisk->mtime.nsec = htonl(ce->ce_stat_data.sd_mtime.nsec);
1767 ondisk->dev = htonl(ce->ce_stat_data.sd_dev);
1768 ondisk->ino = htonl(ce->ce_stat_data.sd_ino);
1769 ondisk->mode = htonl(ce->ce_mode);
1770 ondisk->uid = htonl(ce->ce_stat_data.sd_uid);
1771 ondisk->gid = htonl(ce->ce_stat_data.sd_gid);
1772 ondisk->size = htonl(ce->ce_stat_data.sd_size);
1773 hashcpy(ondisk->sha1, ce->sha1);
1775 flags = ce->ce_flags & ~CE_NAMEMASK;
1776 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1777 ondisk->flags = htons(flags);
1778 if (ce->ce_flags & CE_EXTENDED) {
1779 struct ondisk_cache_entry_extended *ondisk2;
1780 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1781 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1782 return ondisk2->name;
1784 else {
1785 return ondisk->name;
1789 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1790 struct strbuf *previous_name)
1792 int size;
1793 struct ondisk_cache_entry *ondisk;
1794 int saved_namelen = saved_namelen; /* compiler workaround */
1795 char *name;
1796 int result;
1798 if (ce->ce_flags & CE_STRIP_NAME) {
1799 saved_namelen = ce_namelen(ce);
1800 ce->ce_namelen = 0;
1803 if (!previous_name) {
1804 size = ondisk_ce_size(ce);
1805 ondisk = xcalloc(1, size);
1806 name = copy_cache_entry_to_ondisk(ondisk, ce);
1807 memcpy(name, ce->name, ce_namelen(ce));
1808 } else {
1809 int common, to_remove, prefix_size;
1810 unsigned char to_remove_vi[16];
1811 for (common = 0;
1812 (ce->name[common] &&
1813 common < previous_name->len &&
1814 ce->name[common] == previous_name->buf[common]);
1815 common++)
1816 ; /* still matching */
1817 to_remove = previous_name->len - common;
1818 prefix_size = encode_varint(to_remove, to_remove_vi);
1820 if (ce->ce_flags & CE_EXTENDED)
1821 size = offsetof(struct ondisk_cache_entry_extended, name);
1822 else
1823 size = offsetof(struct ondisk_cache_entry, name);
1824 size += prefix_size + (ce_namelen(ce) - common + 1);
1826 ondisk = xcalloc(1, size);
1827 name = copy_cache_entry_to_ondisk(ondisk, ce);
1828 memcpy(name, to_remove_vi, prefix_size);
1829 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1831 strbuf_splice(previous_name, common, to_remove,
1832 ce->name + common, ce_namelen(ce) - common);
1834 if (ce->ce_flags & CE_STRIP_NAME) {
1835 ce->ce_namelen = saved_namelen;
1836 ce->ce_flags &= ~CE_STRIP_NAME;
1839 result = ce_write(c, fd, ondisk, size);
1840 free(ondisk);
1841 return result;
1845 * This function verifies if index_state has the correct sha1 of the
1846 * index file. Don't die if we have any other failure, just return 0.
1848 static int verify_index_from(const struct index_state *istate, const char *path)
1850 int fd;
1851 ssize_t n;
1852 struct stat st;
1853 unsigned char sha1[20];
1855 if (!istate->initialized)
1856 return 0;
1858 fd = open(path, O_RDONLY);
1859 if (fd < 0)
1860 return 0;
1862 if (fstat(fd, &st))
1863 goto out;
1865 if (st.st_size < sizeof(struct cache_header) + 20)
1866 goto out;
1868 n = pread_in_full(fd, sha1, 20, st.st_size - 20);
1869 if (n != 20)
1870 goto out;
1872 if (hashcmp(istate->sha1, sha1))
1873 goto out;
1875 close(fd);
1876 return 1;
1878 out:
1879 close(fd);
1880 return 0;
1883 static int verify_index(const struct index_state *istate)
1885 return verify_index_from(istate, get_index_file());
1888 static int has_racy_timestamp(struct index_state *istate)
1890 int entries = istate->cache_nr;
1891 int i;
1893 for (i = 0; i < entries; i++) {
1894 struct cache_entry *ce = istate->cache[i];
1895 if (is_racy_timestamp(istate, ce))
1896 return 1;
1898 return 0;
1902 * Opportunistically update the index but do not complain if we can't
1904 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1906 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1907 verify_index(istate) &&
1908 write_locked_index(istate, lockfile, COMMIT_LOCK))
1909 rollback_lock_file(lockfile);
1912 static int do_write_index(struct index_state *istate, int newfd,
1913 int strip_extensions)
1915 git_SHA_CTX c;
1916 struct cache_header hdr;
1917 int i, err, removed, extended, hdr_version;
1918 struct cache_entry **cache = istate->cache;
1919 int entries = istate->cache_nr;
1920 struct stat st;
1921 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1923 for (i = removed = extended = 0; i < entries; i++) {
1924 if (cache[i]->ce_flags & CE_REMOVE)
1925 removed++;
1927 /* reduce extended entries if possible */
1928 cache[i]->ce_flags &= ~CE_EXTENDED;
1929 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1930 extended++;
1931 cache[i]->ce_flags |= CE_EXTENDED;
1935 if (!istate->version) {
1936 istate->version = get_index_format_default();
1937 if (getenv("GIT_TEST_SPLIT_INDEX"))
1938 init_split_index(istate);
1941 /* demote version 3 to version 2 when the latter suffices */
1942 if (istate->version == 3 || istate->version == 2)
1943 istate->version = extended ? 3 : 2;
1945 hdr_version = istate->version;
1947 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1948 hdr.hdr_version = htonl(hdr_version);
1949 hdr.hdr_entries = htonl(entries - removed);
1951 git_SHA1_Init(&c);
1952 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1953 return -1;
1955 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1956 for (i = 0; i < entries; i++) {
1957 struct cache_entry *ce = cache[i];
1958 if (ce->ce_flags & CE_REMOVE)
1959 continue;
1960 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1961 ce_smudge_racily_clean_entry(ce);
1962 if (is_null_sha1(ce->sha1)) {
1963 static const char msg[] = "cache entry has null sha1: %s";
1964 static int allow = -1;
1966 if (allow < 0)
1967 allow = git_env_bool("GIT_ALLOW_NULL_SHA1", 0);
1968 if (allow)
1969 warning(msg, ce->name);
1970 else
1971 return error(msg, ce->name);
1973 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1974 return -1;
1976 strbuf_release(&previous_name_buf);
1978 /* Write extension data here */
1979 if (!strip_extensions && istate->split_index) {
1980 struct strbuf sb = STRBUF_INIT;
1982 err = write_link_extension(&sb, istate) < 0 ||
1983 write_index_ext_header(&c, newfd, CACHE_EXT_LINK,
1984 sb.len) < 0 ||
1985 ce_write(&c, newfd, sb.buf, sb.len) < 0;
1986 strbuf_release(&sb);
1987 if (err)
1988 return -1;
1990 if (!strip_extensions && istate->cache_tree) {
1991 struct strbuf sb = STRBUF_INIT;
1993 cache_tree_write(&sb, istate->cache_tree);
1994 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1995 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1996 strbuf_release(&sb);
1997 if (err)
1998 return -1;
2000 if (!strip_extensions && istate->resolve_undo) {
2001 struct strbuf sb = STRBUF_INIT;
2003 resolve_undo_write(&sb, istate->resolve_undo);
2004 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
2005 sb.len) < 0
2006 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
2007 strbuf_release(&sb);
2008 if (err)
2009 return -1;
2012 if (ce_flush(&c, newfd, istate->sha1) || fstat(newfd, &st))
2013 return -1;
2014 istate->timestamp.sec = (unsigned int)st.st_mtime;
2015 istate->timestamp.nsec = ST_MTIME_NSEC(st);
2016 return 0;
2019 void set_alternate_index_output(const char *name)
2021 alternate_index_output = name;
2024 static int commit_locked_index(struct lock_file *lk)
2026 if (alternate_index_output) {
2027 if (lk->fd >= 0 && close_lock_file(lk))
2028 return -1;
2029 if (rename(lk->filename, alternate_index_output))
2030 return -1;
2031 lk->filename[0] = 0;
2032 return 0;
2033 } else {
2034 return commit_lock_file(lk);
2038 static int do_write_locked_index(struct index_state *istate, struct lock_file *lock,
2039 unsigned flags)
2041 int ret = do_write_index(istate, lock->fd, 0);
2042 if (ret)
2043 return ret;
2044 assert((flags & (COMMIT_LOCK | CLOSE_LOCK)) !=
2045 (COMMIT_LOCK | CLOSE_LOCK));
2046 if (flags & COMMIT_LOCK)
2047 return commit_locked_index(lock);
2048 else if (flags & CLOSE_LOCK)
2049 return close_lock_file(lock);
2050 else
2051 return ret;
2054 static int write_split_index(struct index_state *istate,
2055 struct lock_file *lock,
2056 unsigned flags)
2058 int ret;
2059 prepare_to_write_split_index(istate);
2060 ret = do_write_locked_index(istate, lock, flags);
2061 finish_writing_split_index(istate);
2062 return ret;
2065 static char *temporary_sharedindex;
2067 static void remove_temporary_sharedindex(void)
2069 if (temporary_sharedindex) {
2070 unlink_or_warn(temporary_sharedindex);
2071 free(temporary_sharedindex);
2072 temporary_sharedindex = NULL;
2076 static void remove_temporary_sharedindex_on_signal(int signo)
2078 remove_temporary_sharedindex();
2079 sigchain_pop(signo);
2080 raise(signo);
2083 static int write_shared_index(struct index_state *istate,
2084 struct lock_file *lock, unsigned flags)
2086 struct split_index *si = istate->split_index;
2087 static int installed_handler;
2088 int fd, ret;
2090 temporary_sharedindex = git_pathdup("sharedindex_XXXXXX");
2091 fd = mkstemp(temporary_sharedindex);
2092 if (fd < 0) {
2093 free(temporary_sharedindex);
2094 temporary_sharedindex = NULL;
2095 hashclr(si->base_sha1);
2096 return do_write_locked_index(istate, lock, flags);
2098 if (!installed_handler) {
2099 atexit(remove_temporary_sharedindex);
2100 sigchain_push_common(remove_temporary_sharedindex_on_signal);
2102 move_cache_to_base_index(istate);
2103 ret = do_write_index(si->base, fd, 1);
2104 close(fd);
2105 if (ret) {
2106 remove_temporary_sharedindex();
2107 return ret;
2109 ret = rename(temporary_sharedindex,
2110 git_path("sharedindex.%s", sha1_to_hex(si->base->sha1)));
2111 free(temporary_sharedindex);
2112 temporary_sharedindex = NULL;
2113 if (!ret)
2114 hashcpy(si->base_sha1, si->base->sha1);
2115 return ret;
2118 int write_locked_index(struct index_state *istate, struct lock_file *lock,
2119 unsigned flags)
2121 struct split_index *si = istate->split_index;
2123 if (!si || alternate_index_output ||
2124 (istate->cache_changed & ~EXTMASK)) {
2125 if (si)
2126 hashclr(si->base_sha1);
2127 return do_write_locked_index(istate, lock, flags);
2130 if (getenv("GIT_TEST_SPLIT_INDEX")) {
2131 int v = si->base_sha1[0];
2132 if ((v & 15) < 6)
2133 istate->cache_changed |= SPLIT_INDEX_ORDERED;
2135 if (istate->cache_changed & SPLIT_INDEX_ORDERED) {
2136 int ret = write_shared_index(istate, lock, flags);
2137 if (ret)
2138 return ret;
2141 return write_split_index(istate, lock, flags);
2145 * Read the index file that is potentially unmerged into given
2146 * index_state, dropping any unmerged entries. Returns true if
2147 * the index is unmerged. Callers who want to refuse to work
2148 * from an unmerged state can call this and check its return value,
2149 * instead of calling read_cache().
2151 int read_index_unmerged(struct index_state *istate)
2153 int i;
2154 int unmerged = 0;
2156 read_index(istate);
2157 for (i = 0; i < istate->cache_nr; i++) {
2158 struct cache_entry *ce = istate->cache[i];
2159 struct cache_entry *new_ce;
2160 int size, len;
2162 if (!ce_stage(ce))
2163 continue;
2164 unmerged = 1;
2165 len = ce_namelen(ce);
2166 size = cache_entry_size(len);
2167 new_ce = xcalloc(1, size);
2168 memcpy(new_ce->name, ce->name, len);
2169 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
2170 new_ce->ce_namelen = len;
2171 new_ce->ce_mode = ce->ce_mode;
2172 if (add_index_entry(istate, new_ce, 0))
2173 return error("%s: cannot drop to stage #0",
2174 new_ce->name);
2175 i = index_name_pos(istate, new_ce->name, len);
2177 return unmerged;
2181 * Returns 1 if the path is an "other" path with respect to
2182 * the index; that is, the path is not mentioned in the index at all,
2183 * either as a file, a directory with some files in the index,
2184 * or as an unmerged entry.
2186 * We helpfully remove a trailing "/" from directories so that
2187 * the output of read_directory can be used as-is.
2189 int index_name_is_other(const struct index_state *istate, const char *name,
2190 int namelen)
2192 int pos;
2193 if (namelen && name[namelen - 1] == '/')
2194 namelen--;
2195 pos = index_name_pos(istate, name, namelen);
2196 if (0 <= pos)
2197 return 0; /* exact match */
2198 pos = -pos - 1;
2199 if (pos < istate->cache_nr) {
2200 struct cache_entry *ce = istate->cache[pos];
2201 if (ce_namelen(ce) == namelen &&
2202 !memcmp(ce->name, name, namelen))
2203 return 0; /* Yup, this one exists unmerged */
2205 return 1;
2208 void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
2210 int pos, len;
2211 unsigned long sz;
2212 enum object_type type;
2213 void *data;
2215 len = strlen(path);
2216 pos = index_name_pos(istate, path, len);
2217 if (pos < 0) {
2219 * We might be in the middle of a merge, in which
2220 * case we would read stage #2 (ours).
2222 int i;
2223 for (i = -pos - 1;
2224 (pos < 0 && i < istate->cache_nr &&
2225 !strcmp(istate->cache[i]->name, path));
2226 i++)
2227 if (ce_stage(istate->cache[i]) == 2)
2228 pos = i;
2230 if (pos < 0)
2231 return NULL;
2232 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
2233 if (!data || type != OBJ_BLOB) {
2234 free(data);
2235 return NULL;
2237 if (size)
2238 *size = sz;
2239 return data;
2242 void stat_validity_clear(struct stat_validity *sv)
2244 free(sv->sd);
2245 sv->sd = NULL;
2248 int stat_validity_check(struct stat_validity *sv, const char *path)
2250 struct stat st;
2252 if (stat(path, &st) < 0)
2253 return sv->sd == NULL;
2254 if (!sv->sd)
2255 return 0;
2256 return S_ISREG(st.st_mode) && !match_stat_data(sv->sd, &st);
2259 void stat_validity_update(struct stat_validity *sv, int fd)
2261 struct stat st;
2263 if (fstat(fd, &st) < 0 || !S_ISREG(st.st_mode))
2264 stat_validity_clear(sv);
2265 else {
2266 if (!sv->sd)
2267 sv->sd = xcalloc(1, sizeof(struct stat_data));
2268 fill_stat_data(sv->sd, &st);