Give commit message reencoding for output on MinGW a chance
[git/dscho.git] / read-cache.c
blobef355cc9a89b948688756b8c3681ac607518492b
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"
18 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
20 /* Index extensions.
22 * The first letter should be 'A'..'Z' for extensions that are not
23 * necessary for a correct operation (i.e. optimization data).
24 * When new extensions are added that _needs_ to be understood in
25 * order to correctly interpret the index file, pick character that
26 * is outside the range, to cause the reader to abort.
29 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
30 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
31 #define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
33 struct index_state the_index;
35 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
37 istate->cache[nr] = ce;
38 add_name_hash(istate, ce);
41 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
43 struct cache_entry *old = istate->cache[nr];
45 remove_name_hash(old);
46 set_index_entry(istate, nr, ce);
47 istate->cache_changed = 1;
50 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
52 struct cache_entry *old = istate->cache[nr], *new;
53 int namelen = strlen(new_name);
55 new = xmalloc(cache_entry_size(namelen));
56 copy_cache_entry(new, old);
57 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
58 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
59 memcpy(new->name, new_name, namelen + 1);
61 cache_tree_invalidate_path(istate->cache_tree, old->name);
62 remove_index_entry_at(istate, nr);
63 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
67 * This only updates the "non-critical" parts of the directory
68 * cache, ie the parts that aren't tracked by GIT, and only used
69 * to validate the cache.
71 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
73 ce->ce_ctime.sec = (unsigned int)st->st_ctime;
74 ce->ce_mtime.sec = (unsigned int)st->st_mtime;
75 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
76 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
77 ce->ce_dev = st->st_dev;
78 ce->ce_ino = st->st_ino;
79 ce->ce_uid = st->st_uid;
80 ce->ce_gid = st->st_gid;
81 ce->ce_size = st->st_size;
83 if (assume_unchanged)
84 ce->ce_flags |= CE_VALID;
86 if (S_ISREG(st->st_mode))
87 ce_mark_uptodate(ce);
90 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
92 int match = -1;
93 int fd = open(ce->name, O_RDONLY);
95 if (fd >= 0) {
96 unsigned char sha1[20];
97 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
98 match = hashcmp(sha1, ce->sha1);
99 /* index_fd() closed the file descriptor already */
101 return match;
104 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
106 int match = -1;
107 void *buffer;
108 unsigned long size;
109 enum object_type type;
110 struct strbuf sb = STRBUF_INIT;
112 if (strbuf_readlink(&sb, ce->name, expected_size))
113 return -1;
115 buffer = read_sha1_file(ce->sha1, &type, &size);
116 if (buffer) {
117 if (size == sb.len)
118 match = memcmp(buffer, sb.buf, size);
119 free(buffer);
121 strbuf_release(&sb);
122 return match;
125 static int ce_compare_gitlink(struct cache_entry *ce)
127 unsigned char sha1[20];
130 * We don't actually require that the .git directory
131 * under GITLINK directory be a valid git directory. It
132 * might even be missing (in case nobody populated that
133 * sub-project).
135 * If so, we consider it always to match.
137 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
138 return 0;
139 return hashcmp(sha1, ce->sha1);
142 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
144 switch (st->st_mode & S_IFMT) {
145 case S_IFREG:
146 if (ce_compare_data(ce, st))
147 return DATA_CHANGED;
148 break;
149 case S_IFLNK:
150 if (ce_compare_link(ce, xsize_t(st->st_size)))
151 return DATA_CHANGED;
152 break;
153 case S_IFDIR:
154 if (S_ISGITLINK(ce->ce_mode))
155 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
156 default:
157 return TYPE_CHANGED;
159 return 0;
162 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
164 unsigned int changed = 0;
166 if (ce->ce_flags & CE_REMOVE)
167 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
169 switch (ce->ce_mode & S_IFMT) {
170 case S_IFREG:
171 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
172 /* We consider only the owner x bit to be relevant for
173 * "mode changes"
175 if (trust_executable_bit &&
176 (0100 & (ce->ce_mode ^ st->st_mode)))
177 changed |= MODE_CHANGED;
178 break;
179 case S_IFLNK:
180 if (!S_ISLNK(st->st_mode) &&
181 (has_symlinks || !S_ISREG(st->st_mode)))
182 changed |= TYPE_CHANGED;
183 break;
184 case S_IFGITLINK:
185 /* We ignore most of the st_xxx fields for gitlinks */
186 if (!S_ISDIR(st->st_mode))
187 changed |= TYPE_CHANGED;
188 else if (ce_compare_gitlink(ce))
189 changed |= DATA_CHANGED;
190 return changed;
191 default:
192 die("internal error: ce_mode is %o", ce->ce_mode);
194 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
195 changed |= MTIME_CHANGED;
196 if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
197 changed |= CTIME_CHANGED;
199 #ifdef USE_NSEC
200 if (ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
201 changed |= MTIME_CHANGED;
202 if (trust_ctime && ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
203 changed |= CTIME_CHANGED;
204 #endif
206 if (ce->ce_uid != (unsigned int) st->st_uid ||
207 ce->ce_gid != (unsigned int) st->st_gid)
208 changed |= OWNER_CHANGED;
209 if (ce->ce_ino != (unsigned int) st->st_ino)
210 changed |= INODE_CHANGED;
212 #ifdef USE_STDEV
214 * st_dev breaks on network filesystems where different
215 * clients will have different views of what "device"
216 * the filesystem is on
218 if (ce->ce_dev != (unsigned int) st->st_dev)
219 changed |= INODE_CHANGED;
220 #endif
222 if (ce->ce_size != (unsigned int) st->st_size)
223 changed |= DATA_CHANGED;
225 /* Racily smudged entry? */
226 if (!ce->ce_size) {
227 if (!is_empty_blob_sha1(ce->sha1))
228 changed |= DATA_CHANGED;
231 return changed;
234 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
236 return (!S_ISGITLINK(ce->ce_mode) &&
237 istate->timestamp.sec &&
238 #ifdef USE_NSEC
239 /* nanosecond timestamped files can also be racy! */
240 (istate->timestamp.sec < ce->ce_mtime.sec ||
241 (istate->timestamp.sec == ce->ce_mtime.sec &&
242 istate->timestamp.nsec <= ce->ce_mtime.nsec))
243 #else
244 istate->timestamp.sec <= ce->ce_mtime.sec
245 #endif
249 int ie_match_stat(const struct index_state *istate,
250 struct cache_entry *ce, struct stat *st,
251 unsigned int options)
253 unsigned int changed;
254 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
255 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
256 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
259 * If it's marked as always valid in the index, it's
260 * valid whatever the checked-out copy says.
262 * skip-worktree has the same effect with higher precedence
264 if (!ignore_skip_worktree && ce_skip_worktree(ce))
265 return 0;
266 if (!ignore_valid && (ce->ce_flags & CE_VALID))
267 return 0;
270 * Intent-to-add entries have not been added, so the index entry
271 * by definition never matches what is in the work tree until it
272 * actually gets added.
274 if (ce->ce_flags & CE_INTENT_TO_ADD)
275 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
277 changed = ce_match_stat_basic(ce, st);
280 * Within 1 second of this sequence:
281 * echo xyzzy >file && git-update-index --add file
282 * running this command:
283 * echo frotz >file
284 * would give a falsely clean cache entry. The mtime and
285 * length match the cache, and other stat fields do not change.
287 * We could detect this at update-index time (the cache entry
288 * being registered/updated records the same time as "now")
289 * and delay the return from git-update-index, but that would
290 * effectively mean we can make at most one commit per second,
291 * which is not acceptable. Instead, we check cache entries
292 * whose mtime are the same as the index file timestamp more
293 * carefully than others.
295 if (!changed && is_racy_timestamp(istate, ce)) {
296 if (assume_racy_is_modified)
297 changed |= DATA_CHANGED;
298 else
299 changed |= ce_modified_check_fs(ce, st);
302 return changed;
305 int ie_modified(const struct index_state *istate,
306 struct cache_entry *ce, struct stat *st, unsigned int options)
308 int changed, changed_fs;
310 changed = ie_match_stat(istate, ce, st, options);
311 if (!changed)
312 return 0;
314 * If the mode or type has changed, there's no point in trying
315 * to refresh the entry - it's not going to match
317 if (changed & (MODE_CHANGED | TYPE_CHANGED))
318 return changed;
321 * Immediately after read-tree or update-index --cacheinfo,
322 * the length field is zero, as we have never even read the
323 * lstat(2) information once, and we cannot trust DATA_CHANGED
324 * returned by ie_match_stat() which in turn was returned by
325 * ce_match_stat_basic() to signal that the filesize of the
326 * blob changed. We have to actually go to the filesystem to
327 * see if the contents match, and if so, should answer "unchanged".
329 * The logic does not apply to gitlinks, as ce_match_stat_basic()
330 * already has checked the actual HEAD from the filesystem in the
331 * subproject. If ie_match_stat() already said it is different,
332 * then we know it is.
334 if ((changed & DATA_CHANGED) &&
335 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
336 return changed;
338 changed_fs = ce_modified_check_fs(ce, st);
339 if (changed_fs)
340 return changed | changed_fs;
341 return 0;
344 int base_name_compare(const char *name1, int len1, int mode1,
345 const char *name2, int len2, int mode2)
347 unsigned char c1, c2;
348 int len = len1 < len2 ? len1 : len2;
349 int cmp;
351 cmp = memcmp(name1, name2, len);
352 if (cmp)
353 return cmp;
354 c1 = name1[len];
355 c2 = name2[len];
356 if (!c1 && S_ISDIR(mode1))
357 c1 = '/';
358 if (!c2 && S_ISDIR(mode2))
359 c2 = '/';
360 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
364 * df_name_compare() is identical to base_name_compare(), except it
365 * compares conflicting directory/file entries as equal. Note that
366 * while a directory name compares as equal to a regular file, they
367 * then individually compare _differently_ to a filename that has
368 * a dot after the basename (because '\0' < '.' < '/').
370 * This is used by routines that want to traverse the git namespace
371 * but then handle conflicting entries together when possible.
373 int df_name_compare(const char *name1, int len1, int mode1,
374 const char *name2, int len2, int mode2)
376 int len = len1 < len2 ? len1 : len2, cmp;
377 unsigned char c1, c2;
379 cmp = memcmp(name1, name2, len);
380 if (cmp)
381 return cmp;
382 /* Directories and files compare equal (same length, same name) */
383 if (len1 == len2)
384 return 0;
385 c1 = name1[len];
386 if (!c1 && S_ISDIR(mode1))
387 c1 = '/';
388 c2 = name2[len];
389 if (!c2 && S_ISDIR(mode2))
390 c2 = '/';
391 if (c1 == '/' && !c2)
392 return 0;
393 if (c2 == '/' && !c1)
394 return 0;
395 return c1 - c2;
398 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
400 int len1 = flags1 & CE_NAMEMASK;
401 int len2 = flags2 & CE_NAMEMASK;
402 int len = len1 < len2 ? len1 : len2;
403 int cmp;
405 cmp = memcmp(name1, name2, len);
406 if (cmp)
407 return cmp;
408 if (len1 < len2)
409 return -1;
410 if (len1 > len2)
411 return 1;
413 /* Compare stages */
414 flags1 &= CE_STAGEMASK;
415 flags2 &= CE_STAGEMASK;
417 if (flags1 < flags2)
418 return -1;
419 if (flags1 > flags2)
420 return 1;
421 return 0;
424 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
426 int first, last;
428 first = 0;
429 last = istate->cache_nr;
430 while (last > first) {
431 int next = (last + first) >> 1;
432 struct cache_entry *ce = istate->cache[next];
433 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
434 if (!cmp)
435 return next;
436 if (cmp < 0) {
437 last = next;
438 continue;
440 first = next+1;
442 return -first-1;
445 /* Remove entry, return true if there are more entries to go.. */
446 int remove_index_entry_at(struct index_state *istate, int pos)
448 struct cache_entry *ce = istate->cache[pos];
450 record_resolve_undo(istate, ce);
451 remove_name_hash(ce);
452 istate->cache_changed = 1;
453 istate->cache_nr--;
454 if (pos >= istate->cache_nr)
455 return 0;
456 memmove(istate->cache + pos,
457 istate->cache + pos + 1,
458 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
459 return 1;
463 * Remove all cache ententries marked for removal, that is where
464 * CE_REMOVE is set in ce_flags. This is much more effective than
465 * calling remove_index_entry_at() for each entry to be removed.
467 void remove_marked_cache_entries(struct index_state *istate)
469 struct cache_entry **ce_array = istate->cache;
470 unsigned int i, j;
472 for (i = j = 0; i < istate->cache_nr; i++) {
473 if (ce_array[i]->ce_flags & CE_REMOVE)
474 remove_name_hash(ce_array[i]);
475 else
476 ce_array[j++] = ce_array[i];
478 istate->cache_changed = 1;
479 istate->cache_nr = j;
482 int remove_file_from_index(struct index_state *istate, const char *path)
484 int pos = index_name_pos(istate, path, strlen(path));
485 if (pos < 0)
486 pos = -pos-1;
487 cache_tree_invalidate_path(istate->cache_tree, path);
488 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
489 remove_index_entry_at(istate, pos);
490 return 0;
493 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
495 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
498 static int index_name_pos_also_unmerged(struct index_state *istate,
499 const char *path, int namelen)
501 int pos = index_name_pos(istate, path, namelen);
502 struct cache_entry *ce;
504 if (pos >= 0)
505 return pos;
507 /* maybe unmerged? */
508 pos = -1 - pos;
509 if (pos >= istate->cache_nr ||
510 compare_name((ce = istate->cache[pos]), path, namelen))
511 return -1;
513 /* order of preference: stage 2, 1, 3 */
514 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
515 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
516 !compare_name(ce, path, namelen))
517 pos++;
518 return pos;
521 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
523 int len = ce_namelen(ce);
524 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
528 * If we add a filename that aliases in the cache, we will use the
529 * name that we already have - but we don't want to update the same
530 * alias twice, because that implies that there were actually two
531 * different files with aliasing names!
533 * So we use the CE_ADDED flag to verify that the alias was an old
534 * one before we accept it as
536 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
538 int len;
539 struct cache_entry *new;
541 if (alias->ce_flags & CE_ADDED)
542 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
544 /* Ok, create the new entry using the name of the existing alias */
545 len = ce_namelen(alias);
546 new = xcalloc(1, cache_entry_size(len));
547 memcpy(new->name, alias->name, len);
548 copy_cache_entry(new, ce);
549 free(ce);
550 return new;
553 static void record_intent_to_add(struct cache_entry *ce)
555 unsigned char sha1[20];
556 if (write_sha1_file("", 0, blob_type, sha1))
557 die("cannot create an empty blob in the object database");
558 hashcpy(ce->sha1, sha1);
561 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
563 int size, namelen, was_same;
564 mode_t st_mode = st->st_mode;
565 struct cache_entry *ce, *alias;
566 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
567 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
568 int pretend = flags & ADD_CACHE_PRETEND;
569 int intent_only = flags & ADD_CACHE_INTENT;
570 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
571 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
573 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
574 return error("%s: can only add regular files, symbolic links or git-directories", path);
576 namelen = strlen(path);
577 if (S_ISDIR(st_mode)) {
578 while (namelen && path[namelen-1] == '/')
579 namelen--;
581 size = cache_entry_size(namelen);
582 ce = xcalloc(1, size);
583 memcpy(ce->name, path, namelen);
584 ce->ce_flags = namelen;
585 if (!intent_only)
586 fill_stat_cache_info(ce, st);
587 else
588 ce->ce_flags |= CE_INTENT_TO_ADD;
590 if (trust_executable_bit && has_symlinks)
591 ce->ce_mode = create_ce_mode(st_mode);
592 else {
593 /* If there is an existing entry, pick the mode bits and type
594 * from it, otherwise assume unexecutable regular file.
596 struct cache_entry *ent;
597 int pos = index_name_pos_also_unmerged(istate, path, namelen);
599 ent = (0 <= pos) ? istate->cache[pos] : NULL;
600 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
603 /* When core.ignorecase=true, determine if a directory of the same name but differing
604 * case already exists within the Git repository. If it does, ensure the directory
605 * case of the file being added to the repository matches (is folded into) the existing
606 * entry's directory case.
608 if (ignore_case) {
609 const char *startPtr = ce->name;
610 const char *ptr = startPtr;
611 while (*ptr) {
612 while (*ptr && *ptr != '/')
613 ++ptr;
614 if (*ptr == '/') {
615 struct cache_entry *foundce;
616 ++ptr;
617 foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
618 if (foundce) {
619 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
620 startPtr = ptr;
626 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
627 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
628 /* Nothing changed, really */
629 free(ce);
630 if (!S_ISGITLINK(alias->ce_mode))
631 ce_mark_uptodate(alias);
632 alias->ce_flags |= CE_ADDED;
633 return 0;
635 if (!intent_only) {
636 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
637 return error("unable to index file %s", path);
638 } else
639 record_intent_to_add(ce);
641 if (ignore_case && alias && different_name(ce, alias))
642 ce = create_alias_ce(ce, alias);
643 ce->ce_flags |= CE_ADDED;
645 /* It was suspected to be racily clean, but it turns out to be Ok */
646 was_same = (alias &&
647 !ce_stage(alias) &&
648 !hashcmp(alias->sha1, ce->sha1) &&
649 ce->ce_mode == alias->ce_mode);
651 if (pretend)
653 else if (add_index_entry(istate, ce, add_option))
654 return error("unable to add %s to index",path);
655 if (verbose && !was_same)
656 printf("add '%s'\n", path);
657 return 0;
660 int add_file_to_index(struct index_state *istate, const char *path, int flags)
662 struct stat st;
663 if (lstat(path, &st))
664 die_errno("unable to stat '%s'", path);
665 return add_to_index(istate, path, &st, flags);
668 struct cache_entry *make_cache_entry(unsigned int mode,
669 const unsigned char *sha1, const char *path, int stage,
670 int refresh)
672 int size, len;
673 struct cache_entry *ce;
675 if (!verify_path(path)) {
676 error("Invalid path '%s'", path);
677 return NULL;
680 len = strlen(path);
681 size = cache_entry_size(len);
682 ce = xcalloc(1, size);
684 hashcpy(ce->sha1, sha1);
685 memcpy(ce->name, path, len);
686 ce->ce_flags = create_ce_flags(len, stage);
687 ce->ce_mode = create_ce_mode(mode);
689 if (refresh)
690 return refresh_cache_entry(ce, 0);
692 return ce;
695 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
697 int len = ce_namelen(a);
698 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
701 int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
703 return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
707 * We fundamentally don't like some paths: we don't want
708 * dot or dot-dot anywhere, and for obvious reasons don't
709 * want to recurse into ".git" either.
711 * Also, we don't want double slashes or slashes at the
712 * end that can make pathnames ambiguous.
714 static int verify_dotfile(const char *rest)
717 * The first character was '.', but that
718 * has already been discarded, we now test
719 * the rest.
722 /* "." is not allowed */
723 if (*rest == '\0' || is_dir_sep(*rest))
724 return 0;
726 switch (*rest) {
728 * ".git" followed by NUL or slash is bad. This
729 * shares the path end test with the ".." case.
731 case 'g':
732 if (rest[1] != 'i')
733 break;
734 if (rest[2] != 't')
735 break;
736 rest += 2;
737 /* fallthrough */
738 case '.':
739 if (rest[1] == '\0' || is_dir_sep(rest[1]))
740 return 0;
742 return 1;
745 int verify_path(const char *path)
747 char c;
749 if (has_dos_drive_prefix(path))
750 return 0;
752 goto inside;
753 for (;;) {
754 if (!c)
755 return 1;
756 if (is_dir_sep(c)) {
757 inside:
758 c = *path++;
759 if ((c == '.' && !verify_dotfile(path)) ||
760 is_dir_sep(c) || c == '\0')
761 return 0;
763 c = *path++;
768 * Do we have another file that has the beginning components being a
769 * proper superset of the name we're trying to add?
771 static int has_file_name(struct index_state *istate,
772 const struct cache_entry *ce, int pos, int ok_to_replace)
774 int retval = 0;
775 int len = ce_namelen(ce);
776 int stage = ce_stage(ce);
777 const char *name = ce->name;
779 while (pos < istate->cache_nr) {
780 struct cache_entry *p = istate->cache[pos++];
782 if (len >= ce_namelen(p))
783 break;
784 if (memcmp(name, p->name, len))
785 break;
786 if (ce_stage(p) != stage)
787 continue;
788 if (p->name[len] != '/')
789 continue;
790 if (p->ce_flags & CE_REMOVE)
791 continue;
792 retval = -1;
793 if (!ok_to_replace)
794 break;
795 remove_index_entry_at(istate, --pos);
797 return retval;
801 * Do we have another file with a pathname that is a proper
802 * subset of the name we're trying to add?
804 static int has_dir_name(struct index_state *istate,
805 const struct cache_entry *ce, int pos, int ok_to_replace)
807 int retval = 0;
808 int stage = ce_stage(ce);
809 const char *name = ce->name;
810 const char *slash = name + ce_namelen(ce);
812 for (;;) {
813 int len;
815 for (;;) {
816 if (*--slash == '/')
817 break;
818 if (slash <= ce->name)
819 return retval;
821 len = slash - name;
823 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
824 if (pos >= 0) {
826 * Found one, but not so fast. This could
827 * be a marker that says "I was here, but
828 * I am being removed". Such an entry is
829 * not a part of the resulting tree, and
830 * it is Ok to have a directory at the same
831 * path.
833 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
834 retval = -1;
835 if (!ok_to_replace)
836 break;
837 remove_index_entry_at(istate, pos);
838 continue;
841 else
842 pos = -pos-1;
845 * Trivial optimization: if we find an entry that
846 * already matches the sub-directory, then we know
847 * we're ok, and we can exit.
849 while (pos < istate->cache_nr) {
850 struct cache_entry *p = istate->cache[pos];
851 if ((ce_namelen(p) <= len) ||
852 (p->name[len] != '/') ||
853 memcmp(p->name, name, len))
854 break; /* not our subdirectory */
855 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
857 * p is at the same stage as our entry, and
858 * is a subdirectory of what we are looking
859 * at, so we cannot have conflicts at our
860 * level or anything shorter.
862 return retval;
863 pos++;
866 return retval;
869 /* We may be in a situation where we already have path/file and path
870 * is being added, or we already have path and path/file is being
871 * added. Either one would result in a nonsense tree that has path
872 * twice when git-write-tree tries to write it out. Prevent it.
874 * If ok-to-replace is specified, we remove the conflicting entries
875 * from the cache so the caller should recompute the insert position.
876 * When this happens, we return non-zero.
878 static int check_file_directory_conflict(struct index_state *istate,
879 const struct cache_entry *ce,
880 int pos, int ok_to_replace)
882 int retval;
885 * When ce is an "I am going away" entry, we allow it to be added
887 if (ce->ce_flags & CE_REMOVE)
888 return 0;
891 * We check if the path is a sub-path of a subsequent pathname
892 * first, since removing those will not change the position
893 * in the array.
895 retval = has_file_name(istate, ce, pos, ok_to_replace);
898 * Then check if the path might have a clashing sub-directory
899 * before it.
901 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
904 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
906 int pos;
907 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
908 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
909 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
910 int new_only = option & ADD_CACHE_NEW_ONLY;
912 cache_tree_invalidate_path(istate->cache_tree, ce->name);
913 pos = index_name_pos(istate, ce->name, ce->ce_flags);
915 /* existing match? Just replace it. */
916 if (pos >= 0) {
917 if (!new_only)
918 replace_index_entry(istate, pos, ce);
919 return 0;
921 pos = -pos-1;
924 * Inserting a merged entry ("stage 0") into the index
925 * will always replace all non-merged entries..
927 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
928 while (ce_same_name(istate->cache[pos], ce)) {
929 ok_to_add = 1;
930 if (!remove_index_entry_at(istate, pos))
931 break;
935 if (!ok_to_add)
936 return -1;
937 if (!verify_path(ce->name))
938 return error("Invalid path '%s'", ce->name);
940 if (!skip_df_check &&
941 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
942 if (!ok_to_replace)
943 return error("'%s' appears as both a file and as a directory",
944 ce->name);
945 pos = index_name_pos(istate, ce->name, ce->ce_flags);
946 pos = -pos-1;
948 return pos + 1;
951 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
953 int pos;
955 if (option & ADD_CACHE_JUST_APPEND)
956 pos = istate->cache_nr;
957 else {
958 int ret;
959 ret = add_index_entry_with_check(istate, ce, option);
960 if (ret <= 0)
961 return ret;
962 pos = ret - 1;
965 /* Make sure the array is big enough .. */
966 if (istate->cache_nr == istate->cache_alloc) {
967 istate->cache_alloc = alloc_nr(istate->cache_alloc);
968 istate->cache = xrealloc(istate->cache,
969 istate->cache_alloc * sizeof(struct cache_entry *));
972 /* Add it in.. */
973 istate->cache_nr++;
974 if (istate->cache_nr > pos + 1)
975 memmove(istate->cache + pos + 1,
976 istate->cache + pos,
977 (istate->cache_nr - pos - 1) * sizeof(ce));
978 set_index_entry(istate, pos, ce);
979 istate->cache_changed = 1;
980 return 0;
984 * "refresh" does not calculate a new sha1 file or bring the
985 * cache up-to-date for mode/content changes. But what it
986 * _does_ do is to "re-match" the stat information of a file
987 * with the cache, so that you can refresh the cache for a
988 * file that hasn't been changed but where the stat entry is
989 * out of date.
991 * For example, you'd want to do this after doing a "git-read-tree",
992 * to link up the stat cache details with the proper files.
994 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
995 struct cache_entry *ce,
996 unsigned int options, int *err,
997 int *changed_ret)
999 struct stat st;
1000 struct cache_entry *updated;
1001 int changed, size;
1002 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
1003 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
1005 if (ce_uptodate(ce))
1006 return ce;
1009 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1010 * that the change to the work tree does not matter and told
1011 * us not to worry.
1013 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1014 ce_mark_uptodate(ce);
1015 return ce;
1017 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1018 ce_mark_uptodate(ce);
1019 return ce;
1022 if (lstat(ce->name, &st) < 0) {
1023 if (err)
1024 *err = errno;
1025 return NULL;
1028 changed = ie_match_stat(istate, ce, &st, options);
1029 if (changed_ret)
1030 *changed_ret = changed;
1031 if (!changed) {
1033 * The path is unchanged. If we were told to ignore
1034 * valid bit, then we did the actual stat check and
1035 * found that the entry is unmodified. If the entry
1036 * is not marked VALID, this is the place to mark it
1037 * valid again, under "assume unchanged" mode.
1039 if (ignore_valid && assume_unchanged &&
1040 !(ce->ce_flags & CE_VALID))
1041 ; /* mark this one VALID again */
1042 else {
1044 * We do not mark the index itself "modified"
1045 * because CE_UPTODATE flag is in-core only;
1046 * we are not going to write this change out.
1048 if (!S_ISGITLINK(ce->ce_mode))
1049 ce_mark_uptodate(ce);
1050 return ce;
1054 if (ie_modified(istate, ce, &st, options)) {
1055 if (err)
1056 *err = EINVAL;
1057 return NULL;
1060 size = ce_size(ce);
1061 updated = xmalloc(size);
1062 memcpy(updated, ce, size);
1063 fill_stat_cache_info(updated, &st);
1065 * If ignore_valid is not set, we should leave CE_VALID bit
1066 * alone. Otherwise, paths marked with --no-assume-unchanged
1067 * (i.e. things to be edited) will reacquire CE_VALID bit
1068 * automatically, which is not really what we want.
1070 if (!ignore_valid && assume_unchanged &&
1071 !(ce->ce_flags & CE_VALID))
1072 updated->ce_flags &= ~CE_VALID;
1074 return updated;
1077 static void show_file(const char * fmt, const char * name, int in_porcelain,
1078 int * first, const char *header_msg)
1080 if (in_porcelain && *first && header_msg) {
1081 printf("%s\n", header_msg);
1082 *first = 0;
1084 printf(fmt, name);
1087 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
1088 char *seen, const char *header_msg)
1090 int i;
1091 int has_errors = 0;
1092 int really = (flags & REFRESH_REALLY) != 0;
1093 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1094 int quiet = (flags & REFRESH_QUIET) != 0;
1095 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1096 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1097 int first = 1;
1098 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
1099 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1100 const char *modified_fmt;
1101 const char *deleted_fmt;
1102 const char *typechange_fmt;
1103 const char *added_fmt;
1104 const char *unmerged_fmt;
1106 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
1107 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1108 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1109 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
1110 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
1111 for (i = 0; i < istate->cache_nr; i++) {
1112 struct cache_entry *ce, *new;
1113 int cache_errno = 0;
1114 int changed = 0;
1115 int filtered = 0;
1117 ce = istate->cache[i];
1118 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1119 continue;
1121 if (pathspec &&
1122 !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1123 filtered = 1;
1125 if (ce_stage(ce)) {
1126 while ((i < istate->cache_nr) &&
1127 ! strcmp(istate->cache[i]->name, ce->name))
1128 i++;
1129 i--;
1130 if (allow_unmerged)
1131 continue;
1132 if (!filtered)
1133 show_file(unmerged_fmt, ce->name, in_porcelain,
1134 &first, header_msg);
1135 has_errors = 1;
1136 continue;
1139 if (filtered)
1140 continue;
1142 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
1143 if (new == ce)
1144 continue;
1145 if (!new) {
1146 const char *fmt;
1148 if (not_new && cache_errno == ENOENT)
1149 continue;
1150 if (really && cache_errno == EINVAL) {
1151 /* If we are doing --really-refresh that
1152 * means the index is not valid anymore.
1154 ce->ce_flags &= ~CE_VALID;
1155 istate->cache_changed = 1;
1157 if (quiet)
1158 continue;
1160 if (cache_errno == ENOENT)
1161 fmt = deleted_fmt;
1162 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1163 fmt = added_fmt; /* must be before other checks */
1164 else if (changed & TYPE_CHANGED)
1165 fmt = typechange_fmt;
1166 else
1167 fmt = modified_fmt;
1168 show_file(fmt,
1169 ce->name, in_porcelain, &first, header_msg);
1170 has_errors = 1;
1171 continue;
1174 replace_index_entry(istate, i, new);
1176 return has_errors;
1179 static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1181 return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
1185 /*****************************************************************
1186 * Index File I/O
1187 *****************************************************************/
1189 #define INDEX_FORMAT_DEFAULT 3
1192 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1193 * Again - this is just a (very strong in practice) heuristic that
1194 * the inode hasn't changed.
1196 * We save the fields in big-endian order to allow using the
1197 * index file over NFS transparently.
1199 struct ondisk_cache_entry {
1200 struct cache_time ctime;
1201 struct cache_time mtime;
1202 unsigned int dev;
1203 unsigned int ino;
1204 unsigned int mode;
1205 unsigned int uid;
1206 unsigned int gid;
1207 unsigned int size;
1208 unsigned char sha1[20];
1209 unsigned short flags;
1210 char name[FLEX_ARRAY]; /* more */
1214 * This struct is used when CE_EXTENDED bit is 1
1215 * The struct must match ondisk_cache_entry exactly from
1216 * ctime till flags
1218 struct ondisk_cache_entry_extended {
1219 struct cache_time ctime;
1220 struct cache_time mtime;
1221 unsigned int dev;
1222 unsigned int ino;
1223 unsigned int mode;
1224 unsigned int uid;
1225 unsigned int gid;
1226 unsigned int size;
1227 unsigned char sha1[20];
1228 unsigned short flags;
1229 unsigned short flags2;
1230 char name[FLEX_ARRAY]; /* more */
1233 /* These are only used for v3 or lower */
1234 #define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1235 #define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1236 #define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1237 #define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1238 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1239 ondisk_cache_entry_size(ce_namelen(ce)))
1241 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1243 git_SHA_CTX c;
1244 unsigned char sha1[20];
1245 int hdr_version;
1247 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1248 return error("bad signature");
1249 hdr_version = ntohl(hdr->hdr_version);
1250 if (hdr_version < 2 || 4 < hdr_version)
1251 return error("bad index version %d", hdr_version);
1252 git_SHA1_Init(&c);
1253 git_SHA1_Update(&c, hdr, size - 20);
1254 git_SHA1_Final(sha1, &c);
1255 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1256 return error("bad index file sha1 signature");
1257 return 0;
1260 static int read_index_extension(struct index_state *istate,
1261 const char *ext, void *data, unsigned long sz)
1263 switch (CACHE_EXT(ext)) {
1264 case CACHE_EXT_TREE:
1265 istate->cache_tree = cache_tree_read(data, sz);
1266 break;
1267 case CACHE_EXT_RESOLVE_UNDO:
1268 istate->resolve_undo = resolve_undo_read(data, sz);
1269 break;
1270 default:
1271 if (*ext < 'A' || 'Z' < *ext)
1272 return error("index uses %.4s extension, which we do not understand",
1273 ext);
1274 fprintf(stderr, "ignoring %.4s extension\n", ext);
1275 break;
1277 return 0;
1280 int read_index(struct index_state *istate)
1282 return read_index_from(istate, get_index_file());
1285 #ifndef NEEDS_ALIGNED_ACCESS
1286 #define ntoh_s(var) ntohs(var)
1287 #define ntoh_l(var) ntohl(var)
1288 #else
1289 static inline uint16_t ntoh_s_force_align(void *p)
1291 uint16_t x;
1292 memcpy(&x, p, sizeof(x));
1293 return ntohs(x);
1295 static inline uint32_t ntoh_l_force_align(void *p)
1297 uint32_t x;
1298 memcpy(&x, p, sizeof(x));
1299 return ntohl(x);
1301 #define ntoh_s(var) ntoh_s_force_align(&(var))
1302 #define ntoh_l(var) ntoh_l_force_align(&(var))
1303 #endif
1305 static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1306 unsigned int flags,
1307 const char *name,
1308 size_t len)
1310 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1312 ce->ce_ctime.sec = ntoh_l(ondisk->ctime.sec);
1313 ce->ce_mtime.sec = ntoh_l(ondisk->mtime.sec);
1314 ce->ce_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1315 ce->ce_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1316 ce->ce_dev = ntoh_l(ondisk->dev);
1317 ce->ce_ino = ntoh_l(ondisk->ino);
1318 ce->ce_mode = ntoh_l(ondisk->mode);
1319 ce->ce_uid = ntoh_l(ondisk->uid);
1320 ce->ce_gid = ntoh_l(ondisk->gid);
1321 ce->ce_size = ntoh_l(ondisk->size);
1322 ce->ce_flags = flags;
1323 hashcpy(ce->sha1, ondisk->sha1);
1324 memcpy(ce->name, name, len);
1325 ce->name[len] = '\0';
1326 return ce;
1330 * Adjacent cache entries tend to share the leading paths, so it makes
1331 * sense to only store the differences in later entries. In the v4
1332 * on-disk format of the index, each on-disk cache entry stores the
1333 * number of bytes to be stripped from the end of the previous name,
1334 * and the bytes to append to the result, to come up with its name.
1336 static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1338 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1339 size_t len = decode_varint(&cp);
1341 if (name->len < len)
1342 die("malformed name field in the index");
1343 strbuf_remove(name, name->len - len, len);
1344 for (ep = cp; *ep; ep++)
1345 ; /* find the end */
1346 strbuf_add(name, cp, ep - cp);
1347 return (const char *)ep + 1 - cp_;
1350 static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
1351 unsigned long *ent_size,
1352 struct strbuf *previous_name)
1354 struct cache_entry *ce;
1355 size_t len;
1356 const char *name;
1357 unsigned int flags;
1359 /* On-disk flags are just 16 bits */
1360 flags = ntoh_s(ondisk->flags);
1361 len = flags & CE_NAMEMASK;
1363 if (flags & CE_EXTENDED) {
1364 struct ondisk_cache_entry_extended *ondisk2;
1365 int extended_flags;
1366 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1367 extended_flags = ntoh_s(ondisk2->flags2) << 16;
1368 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1369 if (extended_flags & ~CE_EXTENDED_FLAGS)
1370 die("Unknown index entry format %08x", extended_flags);
1371 flags |= extended_flags;
1372 name = ondisk2->name;
1374 else
1375 name = ondisk->name;
1377 if (!previous_name) {
1378 /* v3 and earlier */
1379 if (len == CE_NAMEMASK)
1380 len = strlen(name);
1381 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1383 *ent_size = ondisk_ce_size(ce);
1384 } else {
1385 unsigned long consumed;
1386 consumed = expand_name_field(previous_name, name);
1387 ce = cache_entry_from_ondisk(ondisk, flags,
1388 previous_name->buf,
1389 previous_name->len);
1391 *ent_size = (name - ((char *)ondisk)) + consumed;
1393 return ce;
1396 /* remember to discard_cache() before reading a different cache! */
1397 int read_index_from(struct index_state *istate, const char *path)
1399 int fd, i;
1400 struct stat st;
1401 unsigned long src_offset;
1402 struct cache_header *hdr;
1403 void *mmap;
1404 size_t mmap_size;
1405 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1407 errno = EBUSY;
1408 if (istate->initialized)
1409 return istate->cache_nr;
1411 errno = ENOENT;
1412 istate->timestamp.sec = 0;
1413 istate->timestamp.nsec = 0;
1414 fd = open(path, O_RDONLY);
1415 if (fd < 0) {
1416 if (errno == ENOENT)
1417 return 0;
1418 die_errno("index file open failed");
1421 if (fstat(fd, &st))
1422 die_errno("cannot stat the open index");
1424 errno = EINVAL;
1425 mmap_size = xsize_t(st.st_size);
1426 if (mmap_size < sizeof(struct cache_header) + 20)
1427 die("index file smaller than expected");
1429 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1430 close(fd);
1431 if (mmap == MAP_FAILED)
1432 die_errno("unable to map index file");
1434 hdr = mmap;
1435 if (verify_hdr(hdr, mmap_size) < 0)
1436 goto unmap;
1438 istate->version = ntohl(hdr->hdr_version);
1439 istate->cache_nr = ntohl(hdr->hdr_entries);
1440 istate->cache_alloc = alloc_nr(istate->cache_nr);
1441 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1442 istate->initialized = 1;
1444 if (istate->version == 4)
1445 previous_name = &previous_name_buf;
1446 else
1447 previous_name = NULL;
1449 src_offset = sizeof(*hdr);
1450 for (i = 0; i < istate->cache_nr; i++) {
1451 struct ondisk_cache_entry *disk_ce;
1452 struct cache_entry *ce;
1453 unsigned long consumed;
1455 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1456 ce = create_from_disk(disk_ce, &consumed, previous_name);
1457 set_index_entry(istate, i, ce);
1459 src_offset += consumed;
1461 strbuf_release(&previous_name_buf);
1462 istate->timestamp.sec = st.st_mtime;
1463 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1465 while (src_offset <= mmap_size - 20 - 8) {
1466 /* After an array of active_nr index entries,
1467 * there can be arbitrary number of extended
1468 * sections, each of which is prefixed with
1469 * extension name (4-byte) and section length
1470 * in 4-byte network byte order.
1472 uint32_t extsize;
1473 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1474 extsize = ntohl(extsize);
1475 if (read_index_extension(istate,
1476 (const char *) mmap + src_offset,
1477 (char *) mmap + src_offset + 8,
1478 extsize) < 0)
1479 goto unmap;
1480 src_offset += 8;
1481 src_offset += extsize;
1483 munmap(mmap, mmap_size);
1484 return istate->cache_nr;
1486 unmap:
1487 munmap(mmap, mmap_size);
1488 errno = EINVAL;
1489 die("index file corrupt");
1492 int is_index_unborn(struct index_state *istate)
1494 return (!istate->cache_nr && !istate->timestamp.sec);
1497 int discard_index(struct index_state *istate)
1499 int i;
1501 for (i = 0; i < istate->cache_nr; i++)
1502 free(istate->cache[i]);
1503 resolve_undo_clear_index(istate);
1504 istate->cache_nr = 0;
1505 istate->cache_changed = 0;
1506 istate->timestamp.sec = 0;
1507 istate->timestamp.nsec = 0;
1508 istate->name_hash_initialized = 0;
1509 free_hash(&istate->name_hash);
1510 cache_tree_free(&(istate->cache_tree));
1511 istate->initialized = 0;
1513 /* no need to throw away allocated active_cache */
1514 return 0;
1517 int unmerged_index(const struct index_state *istate)
1519 int i;
1520 for (i = 0; i < istate->cache_nr; i++) {
1521 if (ce_stage(istate->cache[i]))
1522 return 1;
1524 return 0;
1527 #define WRITE_BUFFER_SIZE 8192
1528 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1529 static unsigned long write_buffer_len;
1531 static int ce_write_flush(git_SHA_CTX *context, int fd)
1533 unsigned int buffered = write_buffer_len;
1534 if (buffered) {
1535 git_SHA1_Update(context, write_buffer, buffered);
1536 if (write_in_full(fd, write_buffer, buffered) != buffered)
1537 return -1;
1538 write_buffer_len = 0;
1540 return 0;
1543 static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
1545 while (len) {
1546 unsigned int buffered = write_buffer_len;
1547 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1548 if (partial > len)
1549 partial = len;
1550 memcpy(write_buffer + buffered, data, partial);
1551 buffered += partial;
1552 if (buffered == WRITE_BUFFER_SIZE) {
1553 write_buffer_len = buffered;
1554 if (ce_write_flush(context, fd))
1555 return -1;
1556 buffered = 0;
1558 write_buffer_len = buffered;
1559 len -= partial;
1560 data = (char *) data + partial;
1562 return 0;
1565 static int write_index_ext_header(git_SHA_CTX *context, int fd,
1566 unsigned int ext, unsigned int sz)
1568 ext = htonl(ext);
1569 sz = htonl(sz);
1570 return ((ce_write(context, fd, &ext, 4) < 0) ||
1571 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1574 static int ce_flush(git_SHA_CTX *context, int fd)
1576 unsigned int left = write_buffer_len;
1578 if (left) {
1579 write_buffer_len = 0;
1580 git_SHA1_Update(context, write_buffer, left);
1583 /* Flush first if not enough space for SHA1 signature */
1584 if (left + 20 > WRITE_BUFFER_SIZE) {
1585 if (write_in_full(fd, write_buffer, left) != left)
1586 return -1;
1587 left = 0;
1590 /* Append the SHA1 signature at the end */
1591 git_SHA1_Final(write_buffer + left, context);
1592 left += 20;
1593 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1596 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1599 * The only thing we care about in this function is to smudge the
1600 * falsely clean entry due to touch-update-touch race, so we leave
1601 * everything else as they are. We are called for entries whose
1602 * ce_mtime match the index file mtime.
1604 * Note that this actually does not do much for gitlinks, for
1605 * which ce_match_stat_basic() always goes to the actual
1606 * contents. The caller checks with is_racy_timestamp() which
1607 * always says "no" for gitlinks, so we are not called for them ;-)
1609 struct stat st;
1611 if (lstat(ce->name, &st) < 0)
1612 return;
1613 if (ce_match_stat_basic(ce, &st))
1614 return;
1615 if (ce_modified_check_fs(ce, &st)) {
1616 /* This is "racily clean"; smudge it. Note that this
1617 * is a tricky code. At first glance, it may appear
1618 * that it can break with this sequence:
1620 * $ echo xyzzy >frotz
1621 * $ git-update-index --add frotz
1622 * $ : >frotz
1623 * $ sleep 3
1624 * $ echo filfre >nitfol
1625 * $ git-update-index --add nitfol
1627 * but it does not. When the second update-index runs,
1628 * it notices that the entry "frotz" has the same timestamp
1629 * as index, and if we were to smudge it by resetting its
1630 * size to zero here, then the object name recorded
1631 * in index is the 6-byte file but the cached stat information
1632 * becomes zero --- which would then match what we would
1633 * obtain from the filesystem next time we stat("frotz").
1635 * However, the second update-index, before calling
1636 * this function, notices that the cached size is 6
1637 * bytes and what is on the filesystem is an empty
1638 * file, and never calls us, so the cached size information
1639 * for "frotz" stays 6 which does not match the filesystem.
1641 ce->ce_size = 0;
1645 /* Copy miscellaneous fields but not the name */
1646 static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1647 struct cache_entry *ce)
1649 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1650 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
1651 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1652 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
1653 ondisk->dev = htonl(ce->ce_dev);
1654 ondisk->ino = htonl(ce->ce_ino);
1655 ondisk->mode = htonl(ce->ce_mode);
1656 ondisk->uid = htonl(ce->ce_uid);
1657 ondisk->gid = htonl(ce->ce_gid);
1658 ondisk->size = htonl(ce->ce_size);
1659 hashcpy(ondisk->sha1, ce->sha1);
1660 ondisk->flags = htons(ce->ce_flags);
1661 if (ce->ce_flags & CE_EXTENDED) {
1662 struct ondisk_cache_entry_extended *ondisk2;
1663 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1664 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
1665 return ondisk2->name;
1667 else {
1668 return ondisk->name;
1672 static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1673 struct strbuf *previous_name)
1675 int size;
1676 struct ondisk_cache_entry *ondisk;
1677 char *name;
1678 int result;
1680 if (!previous_name) {
1681 size = ondisk_ce_size(ce);
1682 ondisk = xcalloc(1, size);
1683 name = copy_cache_entry_to_ondisk(ondisk, ce);
1684 memcpy(name, ce->name, ce_namelen(ce));
1685 } else {
1686 int common, to_remove, prefix_size;
1687 unsigned char to_remove_vi[16];
1688 for (common = 0;
1689 (ce->name[common] &&
1690 common < previous_name->len &&
1691 ce->name[common] == previous_name->buf[common]);
1692 common++)
1693 ; /* still matching */
1694 to_remove = previous_name->len - common;
1695 prefix_size = encode_varint(to_remove, to_remove_vi);
1697 if (ce->ce_flags & CE_EXTENDED)
1698 size = offsetof(struct ondisk_cache_entry_extended, name);
1699 else
1700 size = offsetof(struct ondisk_cache_entry, name);
1701 size += prefix_size + (ce_namelen(ce) - common + 1);
1703 ondisk = xcalloc(1, size);
1704 name = copy_cache_entry_to_ondisk(ondisk, ce);
1705 memcpy(name, to_remove_vi, prefix_size);
1706 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1708 strbuf_splice(previous_name, common, to_remove,
1709 ce->name + common, ce_namelen(ce) - common);
1712 result = ce_write(c, fd, ondisk, size);
1713 free(ondisk);
1714 return result;
1717 static int has_racy_timestamp(struct index_state *istate)
1719 int entries = istate->cache_nr;
1720 int i;
1722 for (i = 0; i < entries; i++) {
1723 struct cache_entry *ce = istate->cache[i];
1724 if (is_racy_timestamp(istate, ce))
1725 return 1;
1727 return 0;
1731 * Opportunisticly update the index but do not complain if we can't
1733 void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1735 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
1736 !write_index(istate, lockfile->fd))
1737 commit_locked_index(lockfile);
1738 else
1739 rollback_lock_file(lockfile);
1742 int write_index(struct index_state *istate, int newfd)
1744 git_SHA_CTX c;
1745 struct cache_header hdr;
1746 int i, err, removed, extended, hdr_version;
1747 struct cache_entry **cache = istate->cache;
1748 int entries = istate->cache_nr;
1749 struct stat st;
1750 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
1752 for (i = removed = extended = 0; i < entries; i++) {
1753 if (cache[i]->ce_flags & CE_REMOVE)
1754 removed++;
1756 /* reduce extended entries if possible */
1757 cache[i]->ce_flags &= ~CE_EXTENDED;
1758 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1759 extended++;
1760 cache[i]->ce_flags |= CE_EXTENDED;
1764 if (!istate->version)
1765 istate->version = INDEX_FORMAT_DEFAULT;
1767 /* demote version 3 to version 2 when the latter suffices */
1768 if (istate->version == 3 || istate->version == 2)
1769 istate->version = extended ? 3 : 2;
1771 hdr_version = istate->version;
1773 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1774 hdr.hdr_version = htonl(hdr_version);
1775 hdr.hdr_entries = htonl(entries - removed);
1777 git_SHA1_Init(&c);
1778 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1779 return -1;
1781 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
1782 for (i = 0; i < entries; i++) {
1783 struct cache_entry *ce = cache[i];
1784 if (ce->ce_flags & CE_REMOVE)
1785 continue;
1786 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1787 ce_smudge_racily_clean_entry(ce);
1788 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
1789 return -1;
1791 strbuf_release(&previous_name_buf);
1793 /* Write extension data here */
1794 if (istate->cache_tree) {
1795 struct strbuf sb = STRBUF_INIT;
1797 cache_tree_write(&sb, istate->cache_tree);
1798 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1799 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1800 strbuf_release(&sb);
1801 if (err)
1802 return -1;
1804 if (istate->resolve_undo) {
1805 struct strbuf sb = STRBUF_INIT;
1807 resolve_undo_write(&sb, istate->resolve_undo);
1808 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1809 sb.len) < 0
1810 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1811 strbuf_release(&sb);
1812 if (err)
1813 return -1;
1816 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1817 return -1;
1818 istate->timestamp.sec = (unsigned int)st.st_mtime;
1819 istate->timestamp.nsec = ST_MTIME_NSEC(st);
1820 return 0;
1824 * Read the index file that is potentially unmerged into given
1825 * index_state, dropping any unmerged entries. Returns true if
1826 * the index is unmerged. Callers who want to refuse to work
1827 * from an unmerged state can call this and check its return value,
1828 * instead of calling read_cache().
1830 int read_index_unmerged(struct index_state *istate)
1832 int i;
1833 int unmerged = 0;
1835 read_index(istate);
1836 for (i = 0; i < istate->cache_nr; i++) {
1837 struct cache_entry *ce = istate->cache[i];
1838 struct cache_entry *new_ce;
1839 int size, len;
1841 if (!ce_stage(ce))
1842 continue;
1843 unmerged = 1;
1844 len = strlen(ce->name);
1845 size = cache_entry_size(len);
1846 new_ce = xcalloc(1, size);
1847 memcpy(new_ce->name, ce->name, len);
1848 new_ce->ce_flags = create_ce_flags(len, 0) | CE_CONFLICTED;
1849 new_ce->ce_mode = ce->ce_mode;
1850 if (add_index_entry(istate, new_ce, 0))
1851 return error("%s: cannot drop to stage #0",
1852 ce->name);
1853 i = index_name_pos(istate, new_ce->name, len);
1855 return unmerged;
1859 * Returns 1 if the path is an "other" path with respect to
1860 * the index; that is, the path is not mentioned in the index at all,
1861 * either as a file, a directory with some files in the index,
1862 * or as an unmerged entry.
1864 * We helpfully remove a trailing "/" from directories so that
1865 * the output of read_directory can be used as-is.
1867 int index_name_is_other(const struct index_state *istate, const char *name,
1868 int namelen)
1870 int pos;
1871 if (namelen && name[namelen - 1] == '/')
1872 namelen--;
1873 pos = index_name_pos(istate, name, namelen);
1874 if (0 <= pos)
1875 return 0; /* exact match */
1876 pos = -pos - 1;
1877 if (pos < istate->cache_nr) {
1878 struct cache_entry *ce = istate->cache[pos];
1879 if (ce_namelen(ce) == namelen &&
1880 !memcmp(ce->name, name, namelen))
1881 return 0; /* Yup, this one exists unmerged */
1883 return 1;