read-cache: add simple performance test
[git/git.git] / read-cache.c
CommitLineData
8bc9a0c7
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
4aab5b46 6#define NO_THE_INDEX_COMPATIBILITY_MACROS
e83c5163 7#include "cache.h"
bad68ec9 8#include "cache-tree.h"
f35a6d3b 9#include "refs.h"
d616813d 10#include "dir.h"
041aee31
JH
11#include "tree.h"
12#include "commit.h"
39425819 13#include "blob.h"
cfc5789a 14#include "resolve-undo.h"
6c9cd161
JH
15#include "strbuf.h"
16#include "varint.h"
bad68ec9 17
87b29e5a
JH
18static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really);
19
b60e188c
TG
20/* Mask for the name length in ce_flags in the on-disk index */
21
22#define CE_NAMEMASK (0x0fff)
23
bad68ec9
JH
24/* Index extensions.
25 *
26 * The first letter should be 'A'..'Z' for extensions that are not
27 * necessary for a correct operation (i.e. optimization data).
28 * When new extensions are added that _needs_ to be understood in
29 * order to correctly interpret the index file, pick character that
30 * is outside the range, to cause the reader to abort.
31 */
32
33#define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
34#define CACHE_EXT_TREE 0x54524545 /* "TREE" */
b659b49b 35#define CACHE_EXT_RESOLVE_UNDO 0x52455543 /* "REUC" */
e83c5163 36
228e94f9 37struct index_state the_index;
8fd2cb40 38
9cb76b8c
JH
39static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
40{
41 istate->cache[nr] = ce;
96872bc2 42 add_name_hash(istate, ce);
9cb76b8c
JH
43}
44
cf558704
LT
45static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
46{
47 struct cache_entry *old = istate->cache[nr];
48
2092678c 49 remove_name_hash(istate, old);
a22c6371 50 set_index_entry(istate, nr, ce);
cf558704
LT
51 istate->cache_changed = 1;
52}
53
81dc2307
PB
54void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
55{
56 struct cache_entry *old = istate->cache[nr], *new;
57 int namelen = strlen(new_name);
58
59 new = xmalloc(cache_entry_size(namelen));
60 copy_cache_entry(new, old);
b60e188c
TG
61 new->ce_flags &= ~CE_STATE_MASK;
62 new->ce_namelen = namelen;
81dc2307
PB
63 memcpy(new->name, new_name, namelen + 1);
64
65 cache_tree_invalidate_path(istate->cache_tree, old->name);
66 remove_index_entry_at(istate, nr);
67 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
68}
69
415e96c8
JH
70/*
71 * This only updates the "non-critical" parts of the directory
72 * cache, ie the parts that aren't tracked by GIT, and only used
73 * to validate the cache.
74 */
75void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
76{
fba2f38a
KB
77 ce->ce_ctime.sec = (unsigned int)st->st_ctime;
78 ce->ce_mtime.sec = (unsigned int)st->st_mtime;
c06ff490
KB
79 ce->ce_ctime.nsec = ST_CTIME_NSEC(*st);
80 ce->ce_mtime.nsec = ST_MTIME_NSEC(*st);
7a51ed66
LT
81 ce->ce_dev = st->st_dev;
82 ce->ce_ino = st->st_ino;
83 ce->ce_uid = st->st_uid;
84 ce->ce_gid = st->st_gid;
85 ce->ce_size = st->st_size;
5f73076c
JH
86
87 if (assume_unchanged)
7a51ed66 88 ce->ce_flags |= CE_VALID;
eadb5831
JH
89
90 if (S_ISREG(st->st_mode))
91 ce_mark_uptodate(ce);
415e96c8
JH
92}
93
29e4d363
JH
94static int ce_compare_data(struct cache_entry *ce, struct stat *st)
95{
96 int match = -1;
97 int fd = open(ce->name, O_RDONLY);
98
99 if (fd >= 0) {
100 unsigned char sha1[20];
c4ce46fc 101 if (!index_fd(sha1, fd, st, OBJ_BLOB, ce->name, 0))
a89fccd2 102 match = hashcmp(sha1, ce->sha1);
7f8508e8 103 /* index_fd() closed the file descriptor already */
29e4d363
JH
104 }
105 return match;
106}
107
dc49cd76 108static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
29e4d363
JH
109{
110 int match = -1;
29e4d363
JH
111 void *buffer;
112 unsigned long size;
21666f1a 113 enum object_type type;
a60272b3 114 struct strbuf sb = STRBUF_INIT;
29e4d363 115
a60272b3 116 if (strbuf_readlink(&sb, ce->name, expected_size))
29e4d363 117 return -1;
a60272b3 118
21666f1a 119 buffer = read_sha1_file(ce->sha1, &type, &size);
a60272b3
LT
120 if (buffer) {
121 if (size == sb.len)
122 match = memcmp(buffer, sb.buf, size);
123 free(buffer);
29e4d363 124 }
a60272b3 125 strbuf_release(&sb);
29e4d363
JH
126 return match;
127}
128
f35a6d3b
LT
129static int ce_compare_gitlink(struct cache_entry *ce)
130{
131 unsigned char sha1[20];
132
133 /*
134 * We don't actually require that the .git directory
302b9282 135 * under GITLINK directory be a valid git directory. It
f35a6d3b
LT
136 * might even be missing (in case nobody populated that
137 * sub-project).
138 *
139 * If so, we consider it always to match.
140 */
141 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
142 return 0;
143 return hashcmp(sha1, ce->sha1);
144}
145
29e4d363
JH
146static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
147{
148 switch (st->st_mode & S_IFMT) {
149 case S_IFREG:
150 if (ce_compare_data(ce, st))
151 return DATA_CHANGED;
152 break;
153 case S_IFLNK:
dc49cd76 154 if (ce_compare_link(ce, xsize_t(st->st_size)))
29e4d363
JH
155 return DATA_CHANGED;
156 break;
a8ee75bc 157 case S_IFDIR:
7a51ed66 158 if (S_ISGITLINK(ce->ce_mode))
c70115b4 159 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
29e4d363
JH
160 default:
161 return TYPE_CHANGED;
162 }
163 return 0;
164}
165
407c8eb0 166static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
734aab75
LT
167{
168 unsigned int changed = 0;
169
7a51ed66
LT
170 if (ce->ce_flags & CE_REMOVE)
171 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
172
173 switch (ce->ce_mode & S_IFMT) {
8ae0a8c5
KS
174 case S_IFREG:
175 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
3e09cdfd
JH
176 /* We consider only the owner x bit to be relevant for
177 * "mode changes"
178 */
179 if (trust_executable_bit &&
7a51ed66 180 (0100 & (ce->ce_mode ^ st->st_mode)))
ffbe1add 181 changed |= MODE_CHANGED;
8ae0a8c5
KS
182 break;
183 case S_IFLNK:
78a8d641
JS
184 if (!S_ISLNK(st->st_mode) &&
185 (has_symlinks || !S_ISREG(st->st_mode)))
186 changed |= TYPE_CHANGED;
8ae0a8c5 187 break;
302b9282 188 case S_IFGITLINK:
c70115b4 189 /* We ignore most of the st_xxx fields for gitlinks */
f35a6d3b
LT
190 if (!S_ISDIR(st->st_mode))
191 changed |= TYPE_CHANGED;
192 else if (ce_compare_gitlink(ce))
193 changed |= DATA_CHANGED;
a8ee75bc 194 return changed;
8ae0a8c5 195 default:
7a51ed66 196 die("internal error: ce_mode is %o", ce->ce_mode);
8ae0a8c5 197 }
fba2f38a 198 if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
ccc4feb5 199 changed |= MTIME_CHANGED;
c08e4d5b
RR
200 if (trust_ctime && check_stat &&
201 ce->ce_ctime.sec != (unsigned int)st->st_ctime)
734aab75 202 changed |= CTIME_CHANGED;
ccc4feb5 203
fba2f38a 204#ifdef USE_NSEC
c08e4d5b 205 if (check_stat && ce->ce_mtime.nsec != ST_MTIME_NSEC(*st))
fba2f38a 206 changed |= MTIME_CHANGED;
c08e4d5b
RR
207 if (trust_ctime && check_stat &&
208 ce->ce_ctime.nsec != ST_CTIME_NSEC(*st))
fba2f38a
KB
209 changed |= CTIME_CHANGED;
210#endif
211
c08e4d5b
RR
212 if (check_stat) {
213 if (ce->ce_uid != (unsigned int) st->st_uid ||
214 ce->ce_gid != (unsigned int) st->st_gid)
215 changed |= OWNER_CHANGED;
216 if (ce->ce_ino != (unsigned int) st->st_ino)
217 changed |= INODE_CHANGED;
218 }
2cb45e95
LT
219
220#ifdef USE_STDEV
221 /*
222 * st_dev breaks on network filesystems where different
223 * clients will have different views of what "device"
224 * the filesystem is on
225 */
c08e4d5b
RR
226 if (check_stat && ce->ce_dev != (unsigned int) st->st_dev)
227 changed |= INODE_CHANGED;
2cb45e95
LT
228#endif
229
7a51ed66 230 if (ce->ce_size != (unsigned int) st->st_size)
734aab75 231 changed |= DATA_CHANGED;
b0391890 232
f49c2c22
LT
233 /* Racily smudged entry? */
234 if (!ce->ce_size) {
235 if (!is_empty_blob_sha1(ce->sha1))
236 changed |= DATA_CHANGED;
237 }
238
407c8eb0
JH
239 return changed;
240}
241
d1f128b0 242static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
6d91da6d 243{
050288d5 244 return (!S_ISGITLINK(ce->ce_mode) &&
fba2f38a
KB
245 istate->timestamp.sec &&
246#ifdef USE_NSEC
247 /* nanosecond timestamped files can also be racy! */
248 (istate->timestamp.sec < ce->ce_mtime.sec ||
249 (istate->timestamp.sec == ce->ce_mtime.sec &&
250 istate->timestamp.nsec <= ce->ce_mtime.nsec))
251#else
252 istate->timestamp.sec <= ce->ce_mtime.sec
253#endif
254 );
6d91da6d
JH
255}
256
d1f128b0 257int ie_match_stat(const struct index_state *istate,
4bd5b7da
JH
258 struct cache_entry *ce, struct stat *st,
259 unsigned int options)
407c8eb0 260{
5f73076c 261 unsigned int changed;
4bd5b7da 262 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 263 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
4bd5b7da 264 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
5f73076c
JH
265
266 /*
267 * If it's marked as always valid in the index, it's
268 * valid whatever the checked-out copy says.
56cac48c
NTND
269 *
270 * skip-worktree has the same effect with higher precedence
5f73076c 271 */
56cac48c
NTND
272 if (!ignore_skip_worktree && ce_skip_worktree(ce))
273 return 0;
7a51ed66 274 if (!ignore_valid && (ce->ce_flags & CE_VALID))
5f73076c
JH
275 return 0;
276
331fcb59
JH
277 /*
278 * Intent-to-add entries have not been added, so the index entry
279 * by definition never matches what is in the work tree until it
280 * actually gets added.
281 */
282 if (ce->ce_flags & CE_INTENT_TO_ADD)
283 return DATA_CHANGED | TYPE_CHANGED | MODE_CHANGED;
284
5f73076c 285 changed = ce_match_stat_basic(ce, st);
407c8eb0 286
29e4d363
JH
287 /*
288 * Within 1 second of this sequence:
289 * echo xyzzy >file && git-update-index --add file
290 * running this command:
291 * echo frotz >file
292 * would give a falsely clean cache entry. The mtime and
293 * length match the cache, and other stat fields do not change.
294 *
295 * We could detect this at update-index time (the cache entry
296 * being registered/updated records the same time as "now")
297 * and delay the return from git-update-index, but that would
298 * effectively mean we can make at most one commit per second,
299 * which is not acceptable. Instead, we check cache entries
300 * whose mtime are the same as the index file timestamp more
5f73076c 301 * carefully than others.
29e4d363 302 */
6d91da6d 303 if (!changed && is_racy_timestamp(istate, ce)) {
42f77406
JH
304 if (assume_racy_is_modified)
305 changed |= DATA_CHANGED;
306 else
307 changed |= ce_modified_check_fs(ce, st);
308 }
b0391890 309
29e4d363 310 return changed;
b0391890
JH
311}
312
d1f128b0 313int ie_modified(const struct index_state *istate,
4bd5b7da 314 struct cache_entry *ce, struct stat *st, unsigned int options)
b0391890 315{
29e4d363 316 int changed, changed_fs;
4bd5b7da
JH
317
318 changed = ie_match_stat(istate, ce, st, options);
b0391890
JH
319 if (!changed)
320 return 0;
b0391890
JH
321 /*
322 * If the mode or type has changed, there's no point in trying
323 * to refresh the entry - it's not going to match
324 */
325 if (changed & (MODE_CHANGED | TYPE_CHANGED))
326 return changed;
327
c70115b4
JH
328 /*
329 * Immediately after read-tree or update-index --cacheinfo,
330 * the length field is zero, as we have never even read the
331 * lstat(2) information once, and we cannot trust DATA_CHANGED
332 * returned by ie_match_stat() which in turn was returned by
333 * ce_match_stat_basic() to signal that the filesize of the
334 * blob changed. We have to actually go to the filesystem to
335 * see if the contents match, and if so, should answer "unchanged".
336 *
337 * The logic does not apply to gitlinks, as ce_match_stat_basic()
338 * already has checked the actual HEAD from the filesystem in the
339 * subproject. If ie_match_stat() already said it is different,
340 * then we know it is.
b0391890 341 */
c70115b4
JH
342 if ((changed & DATA_CHANGED) &&
343 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
b0391890
JH
344 return changed;
345
29e4d363
JH
346 changed_fs = ce_modified_check_fs(ce, st);
347 if (changed_fs)
348 return changed | changed_fs;
b0391890
JH
349 return 0;
350}
351
958ba6c9
LT
352int base_name_compare(const char *name1, int len1, int mode1,
353 const char *name2, int len2, int mode2)
354{
355 unsigned char c1, c2;
356 int len = len1 < len2 ? len1 : len2;
357 int cmp;
358
359 cmp = memcmp(name1, name2, len);
360 if (cmp)
361 return cmp;
362 c1 = name1[len];
363 c2 = name2[len];
1833a925 364 if (!c1 && S_ISDIR(mode1))
958ba6c9 365 c1 = '/';
1833a925 366 if (!c2 && S_ISDIR(mode2))
958ba6c9
LT
367 c2 = '/';
368 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
369}
370
0ab9e1e8
LT
371/*
372 * df_name_compare() is identical to base_name_compare(), except it
373 * compares conflicting directory/file entries as equal. Note that
374 * while a directory name compares as equal to a regular file, they
375 * then individually compare _differently_ to a filename that has
376 * a dot after the basename (because '\0' < '.' < '/').
377 *
378 * This is used by routines that want to traverse the git namespace
379 * but then handle conflicting entries together when possible.
380 */
381int df_name_compare(const char *name1, int len1, int mode1,
382 const char *name2, int len2, int mode2)
383{
384 int len = len1 < len2 ? len1 : len2, cmp;
385 unsigned char c1, c2;
386
387 cmp = memcmp(name1, name2, len);
388 if (cmp)
389 return cmp;
390 /* Directories and files compare equal (same length, same name) */
391 if (len1 == len2)
392 return 0;
393 c1 = name1[len];
394 if (!c1 && S_ISDIR(mode1))
395 c1 = '/';
396 c2 = name2[len];
397 if (!c2 && S_ISDIR(mode2))
398 c2 = '/';
399 if (c1 == '/' && !c2)
400 return 0;
401 if (c2 == '/' && !c1)
402 return 0;
403 return c1 - c2;
404}
405
b60e188c 406int cache_name_stage_compare(const char *name1, int len1, int stage1, const char *name2, int len2, int stage2)
eb38c22f 407{
b60e188c
TG
408 int len = len1 < len2 ? len1 : len2;
409 int cmp;
eb38c22f
LT
410
411 cmp = memcmp(name1, name2, len);
412 if (cmp)
413 return cmp;
414 if (len1 < len2)
415 return -1;
416 if (len1 > len2)
417 return 1;
5f73076c 418
b60e188c 419 if (stage1 < stage2)
95fd5bf8 420 return -1;
b60e188c 421 if (stage1 > stage2)
95fd5bf8 422 return 1;
eb38c22f
LT
423 return 0;
424}
425
b60e188c
TG
426int cache_name_compare(const char *name1, int len1, const char *name2, int len2)
427{
428 return cache_name_stage_compare(name1, len1, 0, name2, len2, 0);
429}
430
357e9c69 431static int index_name_stage_pos(const struct index_state *istate, const char *name, int namelen, int stage)
eb38c22f
LT
432{
433 int first, last;
434
435 first = 0;
4aab5b46 436 last = istate->cache_nr;
eb38c22f
LT
437 while (last > first) {
438 int next = (last + first) >> 1;
4aab5b46 439 struct cache_entry *ce = istate->cache[next];
b60e188c 440 int cmp = cache_name_stage_compare(name, namelen, stage, ce->name, ce_namelen(ce), ce_stage(ce));
eb38c22f 441 if (!cmp)
76e7f4ec 442 return next;
eb38c22f
LT
443 if (cmp < 0) {
444 last = next;
445 continue;
446 }
447 first = next+1;
448 }
76e7f4ec 449 return -first-1;
eb38c22f
LT
450}
451
b60e188c
TG
452int index_name_pos(const struct index_state *istate, const char *name, int namelen)
453{
454 return index_name_stage_pos(istate, name, namelen, 0);
455}
456
7b937ca3 457/* Remove entry, return true if there are more entries to go.. */
4aab5b46 458int remove_index_entry_at(struct index_state *istate, int pos)
7b937ca3 459{
cf558704
LT
460 struct cache_entry *ce = istate->cache[pos];
461
cfc5789a 462 record_resolve_undo(istate, ce);
2092678c 463 remove_name_hash(istate, ce);
4aab5b46
JH
464 istate->cache_changed = 1;
465 istate->cache_nr--;
466 if (pos >= istate->cache_nr)
7b937ca3 467 return 0;
4aab5b46
JH
468 memmove(istate->cache + pos,
469 istate->cache + pos + 1,
470 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
7b937ca3
LT
471 return 1;
472}
473
36419c8e
KB
474/*
475 * Remove all cache ententries marked for removal, that is where
476 * CE_REMOVE is set in ce_flags. This is much more effective than
477 * calling remove_index_entry_at() for each entry to be removed.
478 */
479void remove_marked_cache_entries(struct index_state *istate)
480{
481 struct cache_entry **ce_array = istate->cache;
482 unsigned int i, j;
483
484 for (i = j = 0; i < istate->cache_nr; i++) {
485 if (ce_array[i]->ce_flags & CE_REMOVE)
2092678c 486 remove_name_hash(istate, ce_array[i]);
36419c8e
KB
487 else
488 ce_array[j++] = ce_array[i];
489 }
490 istate->cache_changed = 1;
491 istate->cache_nr = j;
492}
493
4aab5b46 494int remove_file_from_index(struct index_state *istate, const char *path)
197ee8c9 495{
4aab5b46 496 int pos = index_name_pos(istate, path, strlen(path));
c4e3cca1
JH
497 if (pos < 0)
498 pos = -pos-1;
09d5dc32 499 cache_tree_invalidate_path(istate->cache_tree, path);
4aab5b46
JH
500 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
501 remove_index_entry_at(istate, pos);
197ee8c9
LT
502 return 0;
503}
504
20314271
JS
505static int compare_name(struct cache_entry *ce, const char *path, int namelen)
506{
507 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
508}
509
510static int index_name_pos_also_unmerged(struct index_state *istate,
511 const char *path, int namelen)
512{
513 int pos = index_name_pos(istate, path, namelen);
514 struct cache_entry *ce;
515
516 if (pos >= 0)
517 return pos;
518
519 /* maybe unmerged? */
520 pos = -1 - pos;
521 if (pos >= istate->cache_nr ||
522 compare_name((ce = istate->cache[pos]), path, namelen))
523 return -1;
524
525 /* order of preference: stage 2, 1, 3 */
526 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
527 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
528 !compare_name(ce, path, namelen))
529 pos++;
530 return pos;
531}
532
1102952b
LT
533static int different_name(struct cache_entry *ce, struct cache_entry *alias)
534{
535 int len = ce_namelen(ce);
536 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
537}
538
539/*
540 * If we add a filename that aliases in the cache, we will use the
541 * name that we already have - but we don't want to update the same
542 * alias twice, because that implies that there were actually two
543 * different files with aliasing names!
544 *
545 * So we use the CE_ADDED flag to verify that the alias was an old
546 * one before we accept it as
547 */
548static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
549{
550 int len;
551 struct cache_entry *new;
552
553 if (alias->ce_flags & CE_ADDED)
554 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
555
556 /* Ok, create the new entry using the name of the existing alias */
557 len = ce_namelen(alias);
558 new = xcalloc(1, cache_entry_size(len));
559 memcpy(new->name, alias->name, len);
560 copy_cache_entry(new, ce);
561 free(ce);
562 return new;
563}
564
39425819
JH
565static void record_intent_to_add(struct cache_entry *ce)
566{
567 unsigned char sha1[20];
568 if (write_sha1_file("", 0, blob_type, sha1))
569 die("cannot create an empty blob in the object database");
570 hashcpy(ce->sha1, sha1);
571}
572
38ed1d89 573int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
11be42a4 574{
38ed1d89 575 int size, namelen, was_same;
d177cab0 576 mode_t st_mode = st->st_mode;
6835550d 577 struct cache_entry *ce, *alias;
56cac48c 578 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_IGNORE_SKIP_WORKTREE|CE_MATCH_RACY_IS_DIRTY;
38ed1d89
JH
579 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
580 int pretend = flags & ADD_CACHE_PRETEND;
39425819
JH
581 int intent_only = flags & ADD_CACHE_INTENT;
582 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
583 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
11be42a4 584
d177cab0 585 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
960b8ad1 586 return error("%s: can only add regular files, symbolic links or git-directories", path);
11be42a4
JS
587
588 namelen = strlen(path);
d177cab0 589 if (S_ISDIR(st_mode)) {
09595258
LT
590 while (namelen && path[namelen-1] == '/')
591 namelen--;
592 }
11be42a4
JS
593 size = cache_entry_size(namelen);
594 ce = xcalloc(1, size);
595 memcpy(ce->name, path, namelen);
b60e188c 596 ce->ce_namelen = namelen;
39425819
JH
597 if (!intent_only)
598 fill_stat_cache_info(ce, st);
388b2acd
JH
599 else
600 ce->ce_flags |= CE_INTENT_TO_ADD;
11be42a4 601
78a8d641 602 if (trust_executable_bit && has_symlinks)
d177cab0 603 ce->ce_mode = create_ce_mode(st_mode);
185c975f 604 else {
78a8d641
JS
605 /* If there is an existing entry, pick the mode bits and type
606 * from it, otherwise assume unexecutable regular file.
11be42a4 607 */
185c975f 608 struct cache_entry *ent;
20314271 609 int pos = index_name_pos_also_unmerged(istate, path, namelen);
185c975f 610
4aab5b46 611 ent = (0 <= pos) ? istate->cache[pos] : NULL;
d177cab0 612 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
11be42a4
JS
613 }
614
dc1ae704
JJ
615 /* When core.ignorecase=true, determine if a directory of the same name but differing
616 * case already exists within the Git repository. If it does, ensure the directory
617 * case of the file being added to the repository matches (is folded into) the existing
618 * entry's directory case.
619 */
620 if (ignore_case) {
621 const char *startPtr = ce->name;
622 const char *ptr = startPtr;
623 while (*ptr) {
624 while (*ptr && *ptr != '/')
625 ++ptr;
626 if (*ptr == '/') {
627 struct cache_entry *foundce;
628 ++ptr;
629 foundce = index_name_exists(&the_index, ce->name, ptr - ce->name, ignore_case);
630 if (foundce) {
631 memcpy((void *)startPtr, foundce->name + (startPtr - ce->name), ptr - startPtr);
632 startPtr = ptr;
633 }
634 }
635 }
636 }
637
6835550d 638 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
d177cab0 639 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
0781b8a9
JH
640 /* Nothing changed, really */
641 free(ce);
125fd984
JH
642 if (!S_ISGITLINK(alias->ce_mode))
643 ce_mark_uptodate(alias);
1102952b 644 alias->ce_flags |= CE_ADDED;
0781b8a9
JH
645 return 0;
646 }
39425819 647 if (!intent_only) {
c4ce46fc 648 if (index_path(ce->sha1, path, st, HASH_WRITE_OBJECT))
39425819
JH
649 return error("unable to index file %s", path);
650 } else
651 record_intent_to_add(ce);
652
1102952b
LT
653 if (ignore_case && alias && different_name(ce, alias))
654 ce = create_alias_ce(ce, alias);
655 ce->ce_flags |= CE_ADDED;
38ed1d89 656
3bf0dd1f 657 /* It was suspected to be racily clean, but it turns out to be Ok */
38ed1d89
JH
658 was_same = (alias &&
659 !ce_stage(alias) &&
660 !hashcmp(alias->sha1, ce->sha1) &&
661 ce->ce_mode == alias->ce_mode);
662
663 if (pretend)
664 ;
39425819 665 else if (add_index_entry(istate, ce, add_option))
960b8ad1 666 return error("unable to add %s to index",path);
38ed1d89 667 if (verbose && !was_same)
11be42a4 668 printf("add '%s'\n", path);
11be42a4
JS
669 return 0;
670}
671
38ed1d89 672int add_file_to_index(struct index_state *istate, const char *path, int flags)
d177cab0
LT
673{
674 struct stat st;
675 if (lstat(path, &st))
d824cbba 676 die_errno("unable to stat '%s'", path);
38ed1d89 677 return add_to_index(istate, path, &st, flags);
d177cab0
LT
678}
679
6640f881
CR
680struct cache_entry *make_cache_entry(unsigned int mode,
681 const unsigned char *sha1, const char *path, int stage,
682 int refresh)
683{
684 int size, len;
685 struct cache_entry *ce;
686
7e7abea9
DP
687 if (!verify_path(path)) {
688 error("Invalid path '%s'", path);
6640f881 689 return NULL;
7e7abea9 690 }
6640f881
CR
691
692 len = strlen(path);
693 size = cache_entry_size(len);
694 ce = xcalloc(1, size);
695
696 hashcpy(ce->sha1, sha1);
697 memcpy(ce->name, path, len);
b60e188c
TG
698 ce->ce_flags = create_ce_flags(stage);
699 ce->ce_namelen = len;
6640f881
CR
700 ce->ce_mode = create_ce_mode(mode);
701
702 if (refresh)
703 return refresh_cache_entry(ce, 0);
704
705 return ce;
706}
707
dbbce55b 708int ce_same_name(struct cache_entry *a, struct cache_entry *b)
7b937ca3
LT
709{
710 int len = ce_namelen(a);
711 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
712}
713
eb9cb55b 714int ce_path_match(const struct cache_entry *ce, const struct pathspec *pathspec)
c0fd1f51 715{
898bbd9f 716 return match_pathspec_depth(pathspec, ce->name, ce_namelen(ce), 0, NULL);
c0fd1f51
LT
717}
718
8dcf39c4
LT
719/*
720 * We fundamentally don't like some paths: we don't want
721 * dot or dot-dot anywhere, and for obvious reasons don't
722 * want to recurse into ".git" either.
723 *
724 * Also, we don't want double slashes or slashes at the
725 * end that can make pathnames ambiguous.
726 */
727static int verify_dotfile(const char *rest)
728{
729 /*
730 * The first character was '.', but that
731 * has already been discarded, we now test
732 * the rest.
733 */
e0f530ff 734
8dcf39c4 735 /* "." is not allowed */
e0f530ff 736 if (*rest == '\0' || is_dir_sep(*rest))
8dcf39c4
LT
737 return 0;
738
e0f530ff 739 switch (*rest) {
8dcf39c4
LT
740 /*
741 * ".git" followed by NUL or slash is bad. This
742 * shares the path end test with the ".." case.
743 */
744 case 'g':
745 if (rest[1] != 'i')
746 break;
747 if (rest[2] != 't')
748 break;
749 rest += 2;
750 /* fallthrough */
751 case '.':
e0f530ff 752 if (rest[1] == '\0' || is_dir_sep(rest[1]))
8dcf39c4
LT
753 return 0;
754 }
755 return 1;
756}
757
758int verify_path(const char *path)
759{
760 char c;
761
56948cb6
EFL
762 if (has_dos_drive_prefix(path))
763 return 0;
764
8dcf39c4
LT
765 goto inside;
766 for (;;) {
767 if (!c)
768 return 1;
56948cb6 769 if (is_dir_sep(c)) {
8dcf39c4
LT
770inside:
771 c = *path++;
3bdf09c7
JH
772 if ((c == '.' && !verify_dotfile(path)) ||
773 is_dir_sep(c) || c == '\0')
774 return 0;
8dcf39c4
LT
775 }
776 c = *path++;
777 }
778}
779
12676608
LT
780/*
781 * Do we have another file that has the beginning components being a
782 * proper superset of the name we're trying to add?
0f1e4f04 783 */
4aab5b46
JH
784static int has_file_name(struct index_state *istate,
785 const struct cache_entry *ce, int pos, int ok_to_replace)
0f1e4f04 786{
12676608
LT
787 int retval = 0;
788 int len = ce_namelen(ce);
b155725d 789 int stage = ce_stage(ce);
12676608 790 const char *name = ce->name;
0f1e4f04 791
4aab5b46
JH
792 while (pos < istate->cache_nr) {
793 struct cache_entry *p = istate->cache[pos++];
0f1e4f04 794
12676608 795 if (len >= ce_namelen(p))
0f1e4f04 796 break;
12676608
LT
797 if (memcmp(name, p->name, len))
798 break;
b155725d
JH
799 if (ce_stage(p) != stage)
800 continue;
12676608
LT
801 if (p->name[len] != '/')
802 continue;
7a51ed66 803 if (p->ce_flags & CE_REMOVE)
21cd8d00 804 continue;
12676608
LT
805 retval = -1;
806 if (!ok_to_replace)
807 break;
4aab5b46 808 remove_index_entry_at(istate, --pos);
0f1e4f04 809 }
12676608
LT
810 return retval;
811}
0f1e4f04 812
12676608
LT
813/*
814 * Do we have another file with a pathname that is a proper
815 * subset of the name we're trying to add?
816 */
4aab5b46
JH
817static int has_dir_name(struct index_state *istate,
818 const struct cache_entry *ce, int pos, int ok_to_replace)
12676608
LT
819{
820 int retval = 0;
b155725d 821 int stage = ce_stage(ce);
12676608
LT
822 const char *name = ce->name;
823 const char *slash = name + ce_namelen(ce);
0f1e4f04 824
12676608
LT
825 for (;;) {
826 int len;
0f1e4f04 827
12676608
LT
828 for (;;) {
829 if (*--slash == '/')
830 break;
831 if (slash <= ce->name)
832 return retval;
833 }
834 len = slash - name;
0f1e4f04 835
b60e188c 836 pos = index_name_stage_pos(istate, name, len, stage);
12676608 837 if (pos >= 0) {
21cd8d00
JH
838 /*
839 * Found one, but not so fast. This could
840 * be a marker that says "I was here, but
841 * I am being removed". Such an entry is
842 * not a part of the resulting tree, and
843 * it is Ok to have a directory at the same
844 * path.
845 */
077c48df 846 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
21cd8d00
JH
847 retval = -1;
848 if (!ok_to_replace)
849 break;
4aab5b46 850 remove_index_entry_at(istate, pos);
21cd8d00
JH
851 continue;
852 }
12676608 853 }
21cd8d00
JH
854 else
855 pos = -pos-1;
12676608
LT
856
857 /*
858 * Trivial optimization: if we find an entry that
859 * already matches the sub-directory, then we know
b155725d 860 * we're ok, and we can exit.
12676608 861 */
4aab5b46
JH
862 while (pos < istate->cache_nr) {
863 struct cache_entry *p = istate->cache[pos];
b155725d
JH
864 if ((ce_namelen(p) <= len) ||
865 (p->name[len] != '/') ||
866 memcmp(p->name, name, len))
867 break; /* not our subdirectory */
077c48df
JH
868 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
869 /*
870 * p is at the same stage as our entry, and
b155725d
JH
871 * is a subdirectory of what we are looking
872 * at, so we cannot have conflicts at our
873 * level or anything shorter.
874 */
875 return retval;
876 pos++;
192268c1 877 }
0f1e4f04 878 }
12676608
LT
879 return retval;
880}
881
882/* We may be in a situation where we already have path/file and path
883 * is being added, or we already have path and path/file is being
884 * added. Either one would result in a nonsense tree that has path
885 * twice when git-write-tree tries to write it out. Prevent it.
a6080a0a 886 *
12676608
LT
887 * If ok-to-replace is specified, we remove the conflicting entries
888 * from the cache so the caller should recompute the insert position.
889 * When this happens, we return non-zero.
890 */
4aab5b46
JH
891static int check_file_directory_conflict(struct index_state *istate,
892 const struct cache_entry *ce,
893 int pos, int ok_to_replace)
12676608 894{
21cd8d00
JH
895 int retval;
896
897 /*
898 * When ce is an "I am going away" entry, we allow it to be added
899 */
7a51ed66 900 if (ce->ce_flags & CE_REMOVE)
21cd8d00
JH
901 return 0;
902
12676608
LT
903 /*
904 * We check if the path is a sub-path of a subsequent pathname
905 * first, since removing those will not change the position
21cd8d00 906 * in the array.
12676608 907 */
4aab5b46 908 retval = has_file_name(istate, ce, pos, ok_to_replace);
21cd8d00 909
12676608
LT
910 /*
911 * Then check if the path might have a clashing sub-directory
912 * before it.
913 */
4aab5b46 914 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
0f1e4f04
JH
915}
916
af3785dc 917static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
197ee8c9
LT
918{
919 int pos;
192268c1
JH
920 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
921 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
b155725d 922 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
39425819 923 int new_only = option & ADD_CACHE_NEW_ONLY;
5f73076c 924
09d5dc32 925 cache_tree_invalidate_path(istate->cache_tree, ce->name);
b60e188c 926 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
197ee8c9 927
3e09cdfd 928 /* existing match? Just replace it. */
76e7f4ec 929 if (pos >= 0) {
39425819
JH
930 if (!new_only)
931 replace_index_entry(istate, pos, ce);
197ee8c9
LT
932 return 0;
933 }
76e7f4ec 934 pos = -pos-1;
197ee8c9 935
7b937ca3
LT
936 /*
937 * Inserting a merged entry ("stage 0") into the index
938 * will always replace all non-merged entries..
939 */
4aab5b46
JH
940 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
941 while (ce_same_name(istate->cache[pos], ce)) {
7b937ca3 942 ok_to_add = 1;
4aab5b46 943 if (!remove_index_entry_at(istate, pos))
7b937ca3
LT
944 break;
945 }
946 }
947
121481ab
LT
948 if (!ok_to_add)
949 return -1;
8dcf39c4 950 if (!verify_path(ce->name))
7e7abea9 951 return error("Invalid path '%s'", ce->name);
121481ab 952
3e09cdfd 953 if (!skip_df_check &&
4aab5b46 954 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
192268c1 955 if (!ok_to_replace)
4aab5b46
JH
956 return error("'%s' appears as both a file and as a directory",
957 ce->name);
b60e188c 958 pos = index_name_stage_pos(istate, ce->name, ce_namelen(ce), ce_stage(ce));
192268c1
JH
959 pos = -pos-1;
960 }
af3785dc
JH
961 return pos + 1;
962}
963
964int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
965{
966 int pos;
967
968 if (option & ADD_CACHE_JUST_APPEND)
969 pos = istate->cache_nr;
970 else {
971 int ret;
972 ret = add_index_entry_with_check(istate, ce, option);
973 if (ret <= 0)
974 return ret;
975 pos = ret - 1;
976 }
0f1e4f04 977
197ee8c9 978 /* Make sure the array is big enough .. */
4aab5b46
JH
979 if (istate->cache_nr == istate->cache_alloc) {
980 istate->cache_alloc = alloc_nr(istate->cache_alloc);
981 istate->cache = xrealloc(istate->cache,
982 istate->cache_alloc * sizeof(struct cache_entry *));
197ee8c9
LT
983 }
984
985 /* Add it in.. */
4aab5b46 986 istate->cache_nr++;
af3785dc 987 if (istate->cache_nr > pos + 1)
4aab5b46
JH
988 memmove(istate->cache + pos + 1,
989 istate->cache + pos,
990 (istate->cache_nr - pos - 1) * sizeof(ce));
cf558704 991 set_index_entry(istate, pos, ce);
4aab5b46 992 istate->cache_changed = 1;
197ee8c9
LT
993 return 0;
994}
995
405e5b2f
LT
996/*
997 * "refresh" does not calculate a new sha1 file or bring the
998 * cache up-to-date for mode/content changes. But what it
999 * _does_ do is to "re-match" the stat information of a file
1000 * with the cache, so that you can refresh the cache for a
1001 * file that hasn't been changed but where the stat entry is
1002 * out of date.
1003 *
1004 * For example, you'd want to do this after doing a "git-read-tree",
1005 * to link up the stat cache details with the proper files.
1006 */
4aab5b46 1007static struct cache_entry *refresh_cache_ent(struct index_state *istate,
4bd5b7da 1008 struct cache_entry *ce,
d05e6970
JK
1009 unsigned int options, int *err,
1010 int *changed_ret)
405e5b2f
LT
1011{
1012 struct stat st;
1013 struct cache_entry *updated;
1014 int changed, size;
4bd5b7da 1015 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
56cac48c 1016 int ignore_skip_worktree = options & CE_MATCH_IGNORE_SKIP_WORKTREE;
405e5b2f 1017
eadb5831
JH
1018 if (ce_uptodate(ce))
1019 return ce;
1020
aa9349d4 1021 /*
56cac48c
NTND
1022 * CE_VALID or CE_SKIP_WORKTREE means the user promised us
1023 * that the change to the work tree does not matter and told
1024 * us not to worry.
aa9349d4 1025 */
56cac48c
NTND
1026 if (!ignore_skip_worktree && ce_skip_worktree(ce)) {
1027 ce_mark_uptodate(ce);
1028 return ce;
1029 }
aa9349d4
MSO
1030 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
1031 ce_mark_uptodate(ce);
1032 return ce;
1033 }
1034
8fd2cb40 1035 if (lstat(ce->name, &st) < 0) {
ec0cc704
JH
1036 if (err)
1037 *err = errno;
8fd2cb40
JS
1038 return NULL;
1039 }
405e5b2f 1040
4bd5b7da 1041 changed = ie_match_stat(istate, ce, &st, options);
d05e6970
JK
1042 if (changed_ret)
1043 *changed_ret = changed;
405e5b2f 1044 if (!changed) {
4bd5b7da
JH
1045 /*
1046 * The path is unchanged. If we were told to ignore
1047 * valid bit, then we did the actual stat check and
1048 * found that the entry is unmodified. If the entry
1049 * is not marked VALID, this is the place to mark it
1050 * valid again, under "assume unchanged" mode.
1051 */
1052 if (ignore_valid && assume_unchanged &&
7a51ed66 1053 !(ce->ce_flags & CE_VALID))
405e5b2f 1054 ; /* mark this one VALID again */
eadb5831
JH
1055 else {
1056 /*
1057 * We do not mark the index itself "modified"
1058 * because CE_UPTODATE flag is in-core only;
1059 * we are not going to write this change out.
1060 */
125fd984
JH
1061 if (!S_ISGITLINK(ce->ce_mode))
1062 ce_mark_uptodate(ce);
8fd2cb40 1063 return ce;
eadb5831 1064 }
405e5b2f
LT
1065 }
1066
4bd5b7da 1067 if (ie_modified(istate, ce, &st, options)) {
ec0cc704
JH
1068 if (err)
1069 *err = EINVAL;
8fd2cb40
JS
1070 return NULL;
1071 }
405e5b2f
LT
1072
1073 size = ce_size(ce);
1074 updated = xmalloc(size);
1075 memcpy(updated, ce, size);
1076 fill_stat_cache_info(updated, &st);
4bd5b7da
JH
1077 /*
1078 * If ignore_valid is not set, we should leave CE_VALID bit
1079 * alone. Otherwise, paths marked with --no-assume-unchanged
1080 * (i.e. things to be edited) will reacquire CE_VALID bit
1081 * automatically, which is not really what we want.
405e5b2f 1082 */
4bd5b7da 1083 if (!ignore_valid && assume_unchanged &&
7a51ed66
LT
1084 !(ce->ce_flags & CE_VALID))
1085 updated->ce_flags &= ~CE_VALID;
405e5b2f
LT
1086
1087 return updated;
1088}
1089
3deffc52 1090static void show_file(const char * fmt, const char * name, int in_porcelain,
046613c5 1091 int * first, const char *header_msg)
3deffc52
MM
1092{
1093 if (in_porcelain && *first && header_msg) {
1094 printf("%s\n", header_msg);
cd2b8ae9 1095 *first = 0;
3deffc52
MM
1096 }
1097 printf(fmt, name);
1098}
1099
1100int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec,
046613c5 1101 char *seen, const char *header_msg)
405e5b2f
LT
1102{
1103 int i;
1104 int has_errors = 0;
1105 int really = (flags & REFRESH_REALLY) != 0;
1106 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1107 int quiet = (flags & REFRESH_QUIET) != 0;
1108 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
5fdeacb0 1109 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
3deffc52
MM
1110 int first = 1;
1111 int in_porcelain = (flags & REFRESH_IN_PORCELAIN);
4bd5b7da 1112 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
4bd4e730 1113 const char *modified_fmt;
73b7eae6
JK
1114 const char *deleted_fmt;
1115 const char *typechange_fmt;
1116 const char *added_fmt;
4bd4e730 1117 const char *unmerged_fmt;
405e5b2f 1118
4bd4e730 1119 modified_fmt = (in_porcelain ? "M\t%s\n" : "%s: needs update\n");
73b7eae6
JK
1120 deleted_fmt = (in_porcelain ? "D\t%s\n" : "%s: needs update\n");
1121 typechange_fmt = (in_porcelain ? "T\t%s\n" : "%s needs update\n");
1122 added_fmt = (in_porcelain ? "A\t%s\n" : "%s needs update\n");
4bd4e730 1123 unmerged_fmt = (in_porcelain ? "U\t%s\n" : "%s: needs merge\n");
4aab5b46 1124 for (i = 0; i < istate->cache_nr; i++) {
405e5b2f 1125 struct cache_entry *ce, *new;
ec0cc704 1126 int cache_errno = 0;
73b7eae6 1127 int changed = 0;
3d1f148c 1128 int filtered = 0;
ec0cc704 1129
4aab5b46 1130 ce = istate->cache[i];
5fdeacb0
JS
1131 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1132 continue;
1133
3d1f148c 1134 if (pathspec &&
68c4f6a5 1135 !match_pathspec(pathspec, ce->name, ce_namelen(ce), 0, seen))
3d1f148c
JH
1136 filtered = 1;
1137
405e5b2f 1138 if (ce_stage(ce)) {
4aab5b46
JH
1139 while ((i < istate->cache_nr) &&
1140 ! strcmp(istate->cache[i]->name, ce->name))
405e5b2f
LT
1141 i++;
1142 i--;
1143 if (allow_unmerged)
1144 continue;
3d1f148c
JH
1145 if (!filtered)
1146 show_file(unmerged_fmt, ce->name, in_porcelain,
1147 &first, header_msg);
405e5b2f
LT
1148 has_errors = 1;
1149 continue;
1150 }
1151
3d1f148c 1152 if (filtered)
d616813d
AJ
1153 continue;
1154
73b7eae6 1155 new = refresh_cache_ent(istate, ce, options, &cache_errno, &changed);
8fd2cb40 1156 if (new == ce)
405e5b2f 1157 continue;
8fd2cb40 1158 if (!new) {
73b7eae6
JK
1159 const char *fmt;
1160
8fd2cb40 1161 if (not_new && cache_errno == ENOENT)
405e5b2f 1162 continue;
8fd2cb40 1163 if (really && cache_errno == EINVAL) {
405e5b2f
LT
1164 /* If we are doing --really-refresh that
1165 * means the index is not valid anymore.
1166 */
7a51ed66 1167 ce->ce_flags &= ~CE_VALID;
4aab5b46 1168 istate->cache_changed = 1;
405e5b2f
LT
1169 }
1170 if (quiet)
1171 continue;
73b7eae6
JK
1172
1173 if (cache_errno == ENOENT)
1174 fmt = deleted_fmt;
1175 else if (ce->ce_flags & CE_INTENT_TO_ADD)
1176 fmt = added_fmt; /* must be before other checks */
1177 else if (changed & TYPE_CHANGED)
1178 fmt = typechange_fmt;
1179 else
1180 fmt = modified_fmt;
1181 show_file(fmt,
1182 ce->name, in_porcelain, &first, header_msg);
405e5b2f
LT
1183 has_errors = 1;
1184 continue;
1185 }
cf558704
LT
1186
1187 replace_index_entry(istate, i, new);
405e5b2f
LT
1188 }
1189 return has_errors;
1190}
1191
87b29e5a 1192static struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
ec0cc704 1193{
d05e6970 1194 return refresh_cache_ent(&the_index, ce, really, NULL, NULL);
ec0cc704
JH
1195}
1196
db3b313c
JH
1197
1198/*****************************************************************
1199 * Index File I/O
1200 *****************************************************************/
1201
9d227781
JH
1202#define INDEX_FORMAT_DEFAULT 3
1203
db3b313c
JH
1204/*
1205 * dev/ino/uid/gid/size are also just tracked to the low 32 bits
1206 * Again - this is just a (very strong in practice) heuristic that
1207 * the inode hasn't changed.
1208 *
1209 * We save the fields in big-endian order to allow using the
1210 * index file over NFS transparently.
1211 */
1212struct ondisk_cache_entry {
1213 struct cache_time ctime;
1214 struct cache_time mtime;
1215 unsigned int dev;
1216 unsigned int ino;
1217 unsigned int mode;
1218 unsigned int uid;
1219 unsigned int gid;
1220 unsigned int size;
1221 unsigned char sha1[20];
1222 unsigned short flags;
1223 char name[FLEX_ARRAY]; /* more */
1224};
1225
1226/*
1227 * This struct is used when CE_EXTENDED bit is 1
1228 * The struct must match ondisk_cache_entry exactly from
1229 * ctime till flags
1230 */
1231struct ondisk_cache_entry_extended {
1232 struct cache_time ctime;
1233 struct cache_time mtime;
1234 unsigned int dev;
1235 unsigned int ino;
1236 unsigned int mode;
1237 unsigned int uid;
1238 unsigned int gid;
1239 unsigned int size;
1240 unsigned char sha1[20];
1241 unsigned short flags;
1242 unsigned short flags2;
1243 char name[FLEX_ARRAY]; /* more */
1244};
1245
6c9cd161 1246/* These are only used for v3 or lower */
db3b313c
JH
1247#define align_flex_name(STRUCT,len) ((offsetof(struct STRUCT,name) + (len) + 8) & ~7)
1248#define ondisk_cache_entry_size(len) align_flex_name(ondisk_cache_entry,len)
1249#define ondisk_cache_entry_extended_size(len) align_flex_name(ondisk_cache_entry_extended,len)
1250#define ondisk_ce_size(ce) (((ce)->ce_flags & CE_EXTENDED) ? \
1251 ondisk_cache_entry_extended_size(ce_namelen(ce)) : \
1252 ondisk_cache_entry_size(ce_namelen(ce)))
1253
bad68ec9 1254static int verify_hdr(struct cache_header *hdr, unsigned long size)
e83c5163 1255{
9126f009 1256 git_SHA_CTX c;
bad68ec9 1257 unsigned char sha1[20];
0136bac9 1258 int hdr_version;
e83c5163 1259
ccc4feb5 1260 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
e83c5163 1261 return error("bad signature");
0136bac9 1262 hdr_version = ntohl(hdr->hdr_version);
b82a7b5b 1263 if (hdr_version < INDEX_FORMAT_LB || INDEX_FORMAT_UB < hdr_version)
0136bac9 1264 return error("bad index version %d", hdr_version);
9126f009
NP
1265 git_SHA1_Init(&c);
1266 git_SHA1_Update(&c, hdr, size - 20);
1267 git_SHA1_Final(sha1, &c);
a89fccd2 1268 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
ca9be054 1269 return error("bad index file sha1 signature");
e83c5163
LT
1270 return 0;
1271}
1272
4aab5b46
JH
1273static int read_index_extension(struct index_state *istate,
1274 const char *ext, void *data, unsigned long sz)
bad68ec9
JH
1275{
1276 switch (CACHE_EXT(ext)) {
1277 case CACHE_EXT_TREE:
4aab5b46 1278 istate->cache_tree = cache_tree_read(data, sz);
bad68ec9 1279 break;
cfc5789a
JH
1280 case CACHE_EXT_RESOLVE_UNDO:
1281 istate->resolve_undo = resolve_undo_read(data, sz);
1282 break;
bad68ec9
JH
1283 default:
1284 if (*ext < 'A' || 'Z' < *ext)
1285 return error("index uses %.4s extension, which we do not understand",
1286 ext);
1287 fprintf(stderr, "ignoring %.4s extension\n", ext);
1288 break;
1289 }
1290 return 0;
1291}
1292
4aab5b46 1293int read_index(struct index_state *istate)
8fd2cb40 1294{
4aab5b46 1295 return read_index_from(istate, get_index_file());
8fd2cb40
JS
1296}
1297
d60c49c2
JH
1298#ifndef NEEDS_ALIGNED_ACCESS
1299#define ntoh_s(var) ntohs(var)
1300#define ntoh_l(var) ntohl(var)
1301#else
1302static inline uint16_t ntoh_s_force_align(void *p)
1303{
1304 uint16_t x;
1305 memcpy(&x, p, sizeof(x));
1306 return ntohs(x);
1307}
1308static inline uint32_t ntoh_l_force_align(void *p)
1309{
1310 uint32_t x;
1311 memcpy(&x, p, sizeof(x));
1312 return ntohl(x);
1313}
1314#define ntoh_s(var) ntoh_s_force_align(&(var))
1315#define ntoh_l(var) ntoh_l_force_align(&(var))
1316#endif
1317
3fc22b53
JH
1318static struct cache_entry *cache_entry_from_ondisk(struct ondisk_cache_entry *ondisk,
1319 unsigned int flags,
1320 const char *name,
1321 size_t len)
1322{
1323 struct cache_entry *ce = xmalloc(cache_entry_size(len));
1324
1325 ce->ce_ctime.sec = ntoh_l(ondisk->ctime.sec);
1326 ce->ce_mtime.sec = ntoh_l(ondisk->mtime.sec);
1327 ce->ce_ctime.nsec = ntoh_l(ondisk->ctime.nsec);
1328 ce->ce_mtime.nsec = ntoh_l(ondisk->mtime.nsec);
1329 ce->ce_dev = ntoh_l(ondisk->dev);
1330 ce->ce_ino = ntoh_l(ondisk->ino);
1331 ce->ce_mode = ntoh_l(ondisk->mode);
1332 ce->ce_uid = ntoh_l(ondisk->uid);
1333 ce->ce_gid = ntoh_l(ondisk->gid);
1334 ce->ce_size = ntoh_l(ondisk->size);
b60e188c
TG
1335 ce->ce_flags = flags & ~CE_NAMEMASK;
1336 ce->ce_namelen = len;
3fc22b53
JH
1337 hashcpy(ce->sha1, ondisk->sha1);
1338 memcpy(ce->name, name, len);
1339 ce->name[len] = '\0';
1340 return ce;
1341}
1342
6c9cd161
JH
1343/*
1344 * Adjacent cache entries tend to share the leading paths, so it makes
1345 * sense to only store the differences in later entries. In the v4
1346 * on-disk format of the index, each on-disk cache entry stores the
1347 * number of bytes to be stripped from the end of the previous name,
1348 * and the bytes to append to the result, to come up with its name.
1349 */
1350static unsigned long expand_name_field(struct strbuf *name, const char *cp_)
1351{
1352 const unsigned char *ep, *cp = (const unsigned char *)cp_;
1353 size_t len = decode_varint(&cp);
1354
1355 if (name->len < len)
1356 die("malformed name field in the index");
1357 strbuf_remove(name, name->len - len, len);
1358 for (ep = cp; *ep; ep++)
1359 ; /* find the end */
1360 strbuf_add(name, cp, ep - cp);
1361 return (const char *)ep + 1 - cp_;
1362}
1363
936f53d0 1364static struct cache_entry *create_from_disk(struct ondisk_cache_entry *ondisk,
6c9cd161
JH
1365 unsigned long *ent_size,
1366 struct strbuf *previous_name)
7a51ed66 1367{
debed2a6 1368 struct cache_entry *ce;
7fec10b7 1369 size_t len;
06aaaa0b 1370 const char *name;
debed2a6 1371 unsigned int flags;
7fec10b7 1372
7a51ed66 1373 /* On-disk flags are just 16 bits */
d60c49c2 1374 flags = ntoh_s(ondisk->flags);
debed2a6 1375 len = flags & CE_NAMEMASK;
7fec10b7 1376
debed2a6 1377 if (flags & CE_EXTENDED) {
06aaaa0b
NTND
1378 struct ondisk_cache_entry_extended *ondisk2;
1379 int extended_flags;
1380 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
d60c49c2 1381 extended_flags = ntoh_s(ondisk2->flags2) << 16;
06aaaa0b
NTND
1382 /* We do not yet understand any bit out of CE_EXTENDED_FLAGS */
1383 if (extended_flags & ~CE_EXTENDED_FLAGS)
1384 die("Unknown index entry format %08x", extended_flags);
debed2a6 1385 flags |= extended_flags;
06aaaa0b
NTND
1386 name = ondisk2->name;
1387 }
1388 else
1389 name = ondisk->name;
1390
6c9cd161
JH
1391 if (!previous_name) {
1392 /* v3 and earlier */
1393 if (len == CE_NAMEMASK)
1394 len = strlen(name);
1395 ce = cache_entry_from_ondisk(ondisk, flags, name, len);
1396
1397 *ent_size = ondisk_ce_size(ce);
1398 } else {
1399 unsigned long consumed;
1400 consumed = expand_name_field(previous_name, name);
1401 ce = cache_entry_from_ondisk(ondisk, flags,
1402 previous_name->buf,
1403 previous_name->len);
1404
1405 *ent_size = (name - ((char *)ondisk)) + consumed;
1406 }
debed2a6 1407 return ce;
cf558704
LT
1408}
1409
8fd2cb40 1410/* remember to discard_cache() before reading a different cache! */
4aab5b46 1411int read_index_from(struct index_state *istate, const char *path)
e83c5163
LT
1412{
1413 int fd, i;
1414 struct stat st;
debed2a6 1415 unsigned long src_offset;
e83c5163 1416 struct cache_header *hdr;
7a51ed66
LT
1417 void *mmap;
1418 size_t mmap_size;
6c9cd161 1419 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
e83c5163 1420
913e0e99 1421 if (istate->initialized)
4aab5b46 1422 return istate->cache_nr;
5d1a5c02 1423
fba2f38a
KB
1424 istate->timestamp.sec = 0;
1425 istate->timestamp.nsec = 0;
8fd2cb40 1426 fd = open(path, O_RDONLY);
5d1a5c02
LT
1427 if (fd < 0) {
1428 if (errno == ENOENT)
1429 return 0;
d824cbba 1430 die_errno("index file open failed");
5d1a5c02 1431 }
e83c5163 1432
3511a377 1433 if (fstat(fd, &st))
d824cbba 1434 die_errno("cannot stat the open index");
3511a377 1435
7a51ed66
LT
1436 mmap_size = xsize_t(st.st_size);
1437 if (mmap_size < sizeof(struct cache_header) + 20)
3511a377
LFC
1438 die("index file smaller than expected");
1439
7a51ed66 1440 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
7a51ed66 1441 if (mmap == MAP_FAILED)
0721c314 1442 die_errno("unable to map index file");
57d84f8d 1443 close(fd);
e83c5163 1444
7a51ed66
LT
1445 hdr = mmap;
1446 if (verify_hdr(hdr, mmap_size) < 0)
e83c5163
LT
1447 goto unmap;
1448
9d227781 1449 istate->version = ntohl(hdr->hdr_version);
4aab5b46
JH
1450 istate->cache_nr = ntohl(hdr->hdr_entries);
1451 istate->cache_alloc = alloc_nr(istate->cache_nr);
1452 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
913e0e99 1453 istate->initialized = 1;
7a51ed66 1454
9d227781 1455 if (istate->version == 4)
6c9cd161
JH
1456 previous_name = &previous_name_buf;
1457 else
1458 previous_name = NULL;
1459
7a51ed66 1460 src_offset = sizeof(*hdr);
4aab5b46 1461 for (i = 0; i < istate->cache_nr; i++) {
7a51ed66 1462 struct ondisk_cache_entry *disk_ce;
4aab5b46 1463 struct cache_entry *ce;
936f53d0 1464 unsigned long consumed;
4aab5b46 1465
7a51ed66 1466 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
6c9cd161 1467 ce = create_from_disk(disk_ce, &consumed, previous_name);
cf558704 1468 set_index_entry(istate, i, ce);
7a51ed66 1469
936f53d0 1470 src_offset += consumed;
e83c5163 1471 }
6c9cd161 1472 strbuf_release(&previous_name_buf);
fba2f38a 1473 istate->timestamp.sec = st.st_mtime;
c06ff490 1474 istate->timestamp.nsec = ST_MTIME_NSEC(st);
fba2f38a 1475
7a51ed66 1476 while (src_offset <= mmap_size - 20 - 8) {
bad68ec9
JH
1477 /* After an array of active_nr index entries,
1478 * there can be arbitrary number of extended
1479 * sections, each of which is prefixed with
1480 * extension name (4-byte) and section length
1481 * in 4-byte network byte order.
1482 */
07cc8eca 1483 uint32_t extsize;
7a51ed66 1484 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
bad68ec9 1485 extsize = ntohl(extsize);
4aab5b46 1486 if (read_index_extension(istate,
7a51ed66
LT
1487 (const char *) mmap + src_offset,
1488 (char *) mmap + src_offset + 8,
1d7f171c 1489 extsize) < 0)
bad68ec9 1490 goto unmap;
7a51ed66
LT
1491 src_offset += 8;
1492 src_offset += extsize;
bad68ec9 1493 }
7a51ed66 1494 munmap(mmap, mmap_size);
4aab5b46 1495 return istate->cache_nr;
e83c5163
LT
1496
1497unmap:
7a51ed66 1498 munmap(mmap, mmap_size);
5d1a5c02 1499 die("index file corrupt");
e83c5163
LT
1500}
1501
fa7b3c2f
JH
1502int is_index_unborn(struct index_state *istate)
1503{
debed2a6 1504 return (!istate->cache_nr && !istate->timestamp.sec);
fa7b3c2f
JH
1505}
1506
4aab5b46 1507int discard_index(struct index_state *istate)
6d297f81 1508{
debed2a6
RS
1509 int i;
1510
1511 for (i = 0; i < istate->cache_nr; i++)
1512 free(istate->cache[i]);
cfc5789a 1513 resolve_undo_clear_index(istate);
4aab5b46
JH
1514 istate->cache_nr = 0;
1515 istate->cache_changed = 0;
fba2f38a
KB
1516 istate->timestamp.sec = 0;
1517 istate->timestamp.nsec = 0;
2092678c 1518 free_name_hash(istate);
4aab5b46 1519 cache_tree_free(&(istate->cache_tree));
913e0e99 1520 istate->initialized = 0;
6d297f81
JS
1521
1522 /* no need to throw away allocated active_cache */
7a51ed66 1523 return 0;
6d297f81
JS
1524}
1525
d1f128b0 1526int unmerged_index(const struct index_state *istate)
94a5728c
DB
1527{
1528 int i;
1529 for (i = 0; i < istate->cache_nr; i++) {
1530 if (ce_stage(istate->cache[i]))
1531 return 1;
1532 }
1533 return 0;
1534}
1535
4990aadc 1536#define WRITE_BUFFER_SIZE 8192
bf0f910d 1537static unsigned char write_buffer[WRITE_BUFFER_SIZE];
4990aadc
LT
1538static unsigned long write_buffer_len;
1539
9126f009 1540static int ce_write_flush(git_SHA_CTX *context, int fd)
6015c28b
JH
1541{
1542 unsigned int buffered = write_buffer_len;
1543 if (buffered) {
9126f009 1544 git_SHA1_Update(context, write_buffer, buffered);
93822c22 1545 if (write_in_full(fd, write_buffer, buffered) != buffered)
6015c28b
JH
1546 return -1;
1547 write_buffer_len = 0;
1548 }
1549 return 0;
1550}
1551
9126f009 1552static int ce_write(git_SHA_CTX *context, int fd, void *data, unsigned int len)
4990aadc
LT
1553{
1554 while (len) {
1555 unsigned int buffered = write_buffer_len;
1556 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1557 if (partial > len)
1558 partial = len;
1559 memcpy(write_buffer + buffered, data, partial);
1560 buffered += partial;
1561 if (buffered == WRITE_BUFFER_SIZE) {
6015c28b
JH
1562 write_buffer_len = buffered;
1563 if (ce_write_flush(context, fd))
4990aadc
LT
1564 return -1;
1565 buffered = 0;
1566 }
1567 write_buffer_len = buffered;
1568 len -= partial;
1d7f171c 1569 data = (char *) data + partial;
a6080a0a
JH
1570 }
1571 return 0;
4990aadc
LT
1572}
1573
9126f009 1574static int write_index_ext_header(git_SHA_CTX *context, int fd,
ac58c7b1 1575 unsigned int ext, unsigned int sz)
bad68ec9
JH
1576{
1577 ext = htonl(ext);
1578 sz = htonl(sz);
968a1d65
DR
1579 return ((ce_write(context, fd, &ext, 4) < 0) ||
1580 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
bad68ec9
JH
1581}
1582
9126f009 1583static int ce_flush(git_SHA_CTX *context, int fd)
4990aadc
LT
1584{
1585 unsigned int left = write_buffer_len;
ca9be054 1586
4990aadc
LT
1587 if (left) {
1588 write_buffer_len = 0;
9126f009 1589 git_SHA1_Update(context, write_buffer, left);
4990aadc 1590 }
ca9be054 1591
2c865d9a
QH
1592 /* Flush first if not enough space for SHA1 signature */
1593 if (left + 20 > WRITE_BUFFER_SIZE) {
93822c22 1594 if (write_in_full(fd, write_buffer, left) != left)
2c865d9a
QH
1595 return -1;
1596 left = 0;
1597 }
1598
ca9be054 1599 /* Append the SHA1 signature at the end */
9126f009 1600 git_SHA1_Final(write_buffer + left, context);
ca9be054 1601 left += 20;
93822c22 1602 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
4990aadc
LT
1603}
1604
407c8eb0
JH
1605static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1606{
1607 /*
1608 * The only thing we care about in this function is to smudge the
1609 * falsely clean entry due to touch-update-touch race, so we leave
1610 * everything else as they are. We are called for entries whose
1611 * ce_mtime match the index file mtime.
c70115b4
JH
1612 *
1613 * Note that this actually does not do much for gitlinks, for
1614 * which ce_match_stat_basic() always goes to the actual
1615 * contents. The caller checks with is_racy_timestamp() which
1616 * always says "no" for gitlinks, so we are not called for them ;-)
407c8eb0
JH
1617 */
1618 struct stat st;
1619
1620 if (lstat(ce->name, &st) < 0)
1621 return;
1622 if (ce_match_stat_basic(ce, &st))
1623 return;
1624 if (ce_modified_check_fs(ce, &st)) {
4b3511b0
JH
1625 /* This is "racily clean"; smudge it. Note that this
1626 * is a tricky code. At first glance, it may appear
1627 * that it can break with this sequence:
1628 *
1629 * $ echo xyzzy >frotz
1630 * $ git-update-index --add frotz
1631 * $ : >frotz
1632 * $ sleep 3
1633 * $ echo filfre >nitfol
1634 * $ git-update-index --add nitfol
1635 *
b7e58b17 1636 * but it does not. When the second update-index runs,
4b3511b0
JH
1637 * it notices that the entry "frotz" has the same timestamp
1638 * as index, and if we were to smudge it by resetting its
1639 * size to zero here, then the object name recorded
1640 * in index is the 6-byte file but the cached stat information
1641 * becomes zero --- which would then match what we would
a6080a0a 1642 * obtain from the filesystem next time we stat("frotz").
4b3511b0
JH
1643 *
1644 * However, the second update-index, before calling
1645 * this function, notices that the cached size is 6
1646 * bytes and what is on the filesystem is an empty
1647 * file, and never calls us, so the cached size information
1648 * for "frotz" stays 6 which does not match the filesystem.
1649 */
7a51ed66 1650 ce->ce_size = 0;
407c8eb0
JH
1651 }
1652}
1653
f136f7bf
JH
1654/* Copy miscellaneous fields but not the name */
1655static char *copy_cache_entry_to_ondisk(struct ondisk_cache_entry *ondisk,
1656 struct cache_entry *ce)
7a51ed66 1657{
b60e188c
TG
1658 short flags;
1659
fba2f38a
KB
1660 ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
1661 ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
fba2f38a
KB
1662 ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
1663 ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
7a51ed66
LT
1664 ondisk->dev = htonl(ce->ce_dev);
1665 ondisk->ino = htonl(ce->ce_ino);
1666 ondisk->mode = htonl(ce->ce_mode);
1667 ondisk->uid = htonl(ce->ce_uid);
1668 ondisk->gid = htonl(ce->ce_gid);
1669 ondisk->size = htonl(ce->ce_size);
1670 hashcpy(ondisk->sha1, ce->sha1);
b60e188c
TG
1671
1672 flags = ce->ce_flags;
1673 flags |= (ce_namelen(ce) >= CE_NAMEMASK ? CE_NAMEMASK : ce_namelen(ce));
1674 ondisk->flags = htons(flags);
06aaaa0b
NTND
1675 if (ce->ce_flags & CE_EXTENDED) {
1676 struct ondisk_cache_entry_extended *ondisk2;
1677 ondisk2 = (struct ondisk_cache_entry_extended *)ondisk;
1678 ondisk2->flags2 = htons((ce->ce_flags & CE_EXTENDED_FLAGS) >> 16);
f136f7bf 1679 return ondisk2->name;
06aaaa0b 1680 }
f136f7bf
JH
1681 else {
1682 return ondisk->name;
1683 }
1684}
1685
9d227781
JH
1686static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce,
1687 struct strbuf *previous_name)
f136f7bf 1688{
9d227781
JH
1689 int size;
1690 struct ondisk_cache_entry *ondisk;
f136f7bf
JH
1691 char *name;
1692 int result;
1693
9d227781
JH
1694 if (!previous_name) {
1695 size = ondisk_ce_size(ce);
1696 ondisk = xcalloc(1, size);
1697 name = copy_cache_entry_to_ondisk(ondisk, ce);
1698 memcpy(name, ce->name, ce_namelen(ce));
1699 } else {
1700 int common, to_remove, prefix_size;
1701 unsigned char to_remove_vi[16];
1702 for (common = 0;
1703 (ce->name[common] &&
1704 common < previous_name->len &&
1705 ce->name[common] == previous_name->buf[common]);
1706 common++)
1707 ; /* still matching */
1708 to_remove = previous_name->len - common;
1709 prefix_size = encode_varint(to_remove, to_remove_vi);
1710
1711 if (ce->ce_flags & CE_EXTENDED)
1712 size = offsetof(struct ondisk_cache_entry_extended, name);
1713 else
1714 size = offsetof(struct ondisk_cache_entry, name);
1715 size += prefix_size + (ce_namelen(ce) - common + 1);
1716
1717 ondisk = xcalloc(1, size);
1718 name = copy_cache_entry_to_ondisk(ondisk, ce);
1719 memcpy(name, to_remove_vi, prefix_size);
1720 memcpy(name + prefix_size, ce->name + common, ce_namelen(ce) - common);
1721
1722 strbuf_splice(previous_name, common, to_remove,
1723 ce->name + common, ce_namelen(ce) - common);
06aaaa0b 1724 }
7a51ed66 1725
59efba64
JN
1726 result = ce_write(c, fd, ondisk, size);
1727 free(ondisk);
1728 return result;
7a51ed66
LT
1729}
1730
483fbe2b
JH
1731static int has_racy_timestamp(struct index_state *istate)
1732{
1733 int entries = istate->cache_nr;
1734 int i;
1735
1736 for (i = 0; i < entries; i++) {
1737 struct cache_entry *ce = istate->cache[i];
1738 if (is_racy_timestamp(istate, ce))
1739 return 1;
1740 }
1741 return 0;
1742}
1743
ccdc4ec3
JH
1744/*
1745 * Opportunisticly update the index but do not complain if we can't
1746 */
1747void update_index_if_able(struct index_state *istate, struct lock_file *lockfile)
1748{
483fbe2b 1749 if ((istate->cache_changed || has_racy_timestamp(istate)) &&
ccdc4ec3
JH
1750 !write_index(istate, lockfile->fd))
1751 commit_locked_index(lockfile);
1752 else
1753 rollback_lock_file(lockfile);
1754}
1755
e1afca4f 1756int write_index(struct index_state *istate, int newfd)
197ee8c9 1757{
9126f009 1758 git_SHA_CTX c;
197ee8c9 1759 struct cache_header hdr;
9d227781 1760 int i, err, removed, extended, hdr_version;
4aab5b46
JH
1761 struct cache_entry **cache = istate->cache;
1762 int entries = istate->cache_nr;
e1afca4f 1763 struct stat st;
9d227781 1764 struct strbuf previous_name_buf = STRBUF_INIT, *previous_name;
025a0709 1765
06aaaa0b 1766 for (i = removed = extended = 0; i < entries; i++) {
7a51ed66 1767 if (cache[i]->ce_flags & CE_REMOVE)
025a0709 1768 removed++;
197ee8c9 1769
06aaaa0b
NTND
1770 /* reduce extended entries if possible */
1771 cache[i]->ce_flags &= ~CE_EXTENDED;
1772 if (cache[i]->ce_flags & CE_EXTENDED_FLAGS) {
1773 extended++;
1774 cache[i]->ce_flags |= CE_EXTENDED;
1775 }
1776 }
1777
9d227781
JH
1778 if (!istate->version)
1779 istate->version = INDEX_FORMAT_DEFAULT;
1780
1781 /* demote version 3 to version 2 when the latter suffices */
1782 if (istate->version == 3 || istate->version == 2)
1783 istate->version = extended ? 3 : 2;
1784
1785 hdr_version = istate->version;
1786
ccc4feb5 1787 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
9d227781 1788 hdr.hdr_version = htonl(hdr_version);
025a0709 1789 hdr.hdr_entries = htonl(entries - removed);
197ee8c9 1790
9126f009 1791 git_SHA1_Init(&c);
ca9be054 1792 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
197ee8c9
LT
1793 return -1;
1794
9d227781 1795 previous_name = (hdr_version == 4) ? &previous_name_buf : NULL;
197ee8c9
LT
1796 for (i = 0; i < entries; i++) {
1797 struct cache_entry *ce = cache[i];
7a51ed66 1798 if (ce->ce_flags & CE_REMOVE)
aa16021e 1799 continue;
e06c43c7 1800 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
407c8eb0 1801 ce_smudge_racily_clean_entry(ce);
4337b585
JK
1802 if (is_null_sha1(ce->sha1))
1803 return error("cache entry has null sha1: %s", ce->name);
9d227781 1804 if (ce_write_entry(&c, newfd, ce, previous_name) < 0)
197ee8c9
LT
1805 return -1;
1806 }
9d227781 1807 strbuf_release(&previous_name_buf);
1af1c2b6 1808
bad68ec9 1809 /* Write extension data here */
4aab5b46 1810 if (istate->cache_tree) {
f285a2d7 1811 struct strbuf sb = STRBUF_INIT;
1dffb8fa 1812
1dffb8fa
PH
1813 cache_tree_write(&sb, istate->cache_tree);
1814 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1815 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1816 strbuf_release(&sb);
1817 if (err)
bad68ec9 1818 return -1;
bad68ec9 1819 }
cfc5789a
JH
1820 if (istate->resolve_undo) {
1821 struct strbuf sb = STRBUF_INIT;
1822
1823 resolve_undo_write(&sb, istate->resolve_undo);
1824 err = write_index_ext_header(&c, newfd, CACHE_EXT_RESOLVE_UNDO,
1825 sb.len) < 0
1826 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1827 strbuf_release(&sb);
1828 if (err)
1829 return -1;
1830 }
e1afca4f
KB
1831
1832 if (ce_flush(&c, newfd) || fstat(newfd, &st))
1833 return -1;
5bcf109c
KB
1834 istate->timestamp.sec = (unsigned int)st.st_mtime;
1835 istate->timestamp.nsec = ST_MTIME_NSEC(st);
e1afca4f 1836 return 0;
197ee8c9 1837}
e46bbcf6
MV
1838
1839/*
1840 * Read the index file that is potentially unmerged into given
63e8dc5b 1841 * index_state, dropping any unmerged entries. Returns true if
e46bbcf6
MV
1842 * the index is unmerged. Callers who want to refuse to work
1843 * from an unmerged state can call this and check its return value,
1844 * instead of calling read_cache().
1845 */
1846int read_index_unmerged(struct index_state *istate)
1847{
1848 int i;
d1a43f2a 1849 int unmerged = 0;
e46bbcf6
MV
1850
1851 read_index(istate);
e46bbcf6
MV
1852 for (i = 0; i < istate->cache_nr; i++) {
1853 struct cache_entry *ce = istate->cache[i];
d1a43f2a
JH
1854 struct cache_entry *new_ce;
1855 int size, len;
1856
1857 if (!ce_stage(ce))
e46bbcf6 1858 continue;
d1a43f2a 1859 unmerged = 1;
68c4f6a5 1860 len = ce_namelen(ce);
d1a43f2a
JH
1861 size = cache_entry_size(len);
1862 new_ce = xcalloc(1, size);
d1a43f2a 1863 memcpy(new_ce->name, ce->name, len);
b60e188c
TG
1864 new_ce->ce_flags = create_ce_flags(0) | CE_CONFLICTED;
1865 new_ce->ce_namelen = len;
d1a43f2a
JH
1866 new_ce->ce_mode = ce->ce_mode;
1867 if (add_index_entry(istate, new_ce, 0))
1868 return error("%s: cannot drop to stage #0",
1869 ce->name);
1870 i = index_name_pos(istate, new_ce->name, len);
e46bbcf6 1871 }
d1a43f2a 1872 return unmerged;
e46bbcf6 1873}
041aee31 1874
98fa4738
JK
1875/*
1876 * Returns 1 if the path is an "other" path with respect to
1877 * the index; that is, the path is not mentioned in the index at all,
1878 * either as a file, a directory with some files in the index,
1879 * or as an unmerged entry.
1880 *
1881 * We helpfully remove a trailing "/" from directories so that
1882 * the output of read_directory can be used as-is.
1883 */
1884int index_name_is_other(const struct index_state *istate, const char *name,
1885 int namelen)
1886{
1887 int pos;
1888 if (namelen && name[namelen - 1] == '/')
1889 namelen--;
1890 pos = index_name_pos(istate, name, namelen);
1891 if (0 <= pos)
1892 return 0; /* exact match */
1893 pos = -pos - 1;
1894 if (pos < istate->cache_nr) {
1895 struct cache_entry *ce = istate->cache[pos];
1896 if (ce_namelen(ce) == namelen &&
1897 !memcmp(ce->name, name, namelen))
1898 return 0; /* Yup, this one exists unmerged */
1899 }
1900 return 1;
1901}
29fb37b2 1902
ff366825 1903void *read_blob_data_from_index(struct index_state *istate, const char *path, unsigned long *size)
29fb37b2
LF
1904{
1905 int pos, len;
1906 unsigned long sz;
1907 enum object_type type;
1908 void *data;
1909
1910 len = strlen(path);
1911 pos = index_name_pos(istate, path, len);
1912 if (pos < 0) {
1913 /*
1914 * We might be in the middle of a merge, in which
1915 * case we would read stage #2 (ours).
1916 */
1917 int i;
1918 for (i = -pos - 1;
1919 (pos < 0 && i < istate->cache_nr &&
1920 !strcmp(istate->cache[i]->name, path));
1921 i++)
1922 if (ce_stage(istate->cache[i]) == 2)
1923 pos = i;
1924 }
1925 if (pos < 0)
1926 return NULL;
1927 data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
1928 if (!data || type != OBJ_BLOB) {
1929 free(data);
1930 return NULL;
1931 }
ff366825
LF
1932 if (size)
1933 *size = sz;
29fb37b2
LF
1934 return data;
1935}