git-add --intent-to-add (-N)
[git/git.git] / read-cache.c
1 /*
2 * GIT - The information manager from hell
3 *
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 "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "blob.h"
17
18 /* Index extensions.
19 *
20 * The first letter should be 'A'..'Z' for extensions that are not
21 * necessary for a correct operation (i.e. optimization data).
22 * When new extensions are added that _needs_ to be understood in
23 * order to correctly interpret the index file, pick character that
24 * is outside the range, to cause the reader to abort.
25 */
26
27 #define CACHE_EXT(s) ( (s[0]<<24)|(s[1]<<16)|(s[2]<<8)|(s[3]) )
28 #define CACHE_EXT_TREE 0x54524545 /* "TREE" */
29
30 struct index_state the_index;
31
32 static void set_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
33 {
34 istate->cache[nr] = ce;
35 add_name_hash(istate, ce);
36 }
37
38 static void replace_index_entry(struct index_state *istate, int nr, struct cache_entry *ce)
39 {
40 struct cache_entry *old = istate->cache[nr];
41
42 remove_name_hash(old);
43 set_index_entry(istate, nr, ce);
44 istate->cache_changed = 1;
45 }
46
47 void rename_index_entry_at(struct index_state *istate, int nr, const char *new_name)
48 {
49 struct cache_entry *old = istate->cache[nr], *new;
50 int namelen = strlen(new_name);
51
52 new = xmalloc(cache_entry_size(namelen));
53 copy_cache_entry(new, old);
54 new->ce_flags &= ~(CE_STATE_MASK | CE_NAMEMASK);
55 new->ce_flags |= (namelen >= CE_NAMEMASK ? CE_NAMEMASK : namelen);
56 memcpy(new->name, new_name, namelen + 1);
57
58 cache_tree_invalidate_path(istate->cache_tree, old->name);
59 remove_index_entry_at(istate, nr);
60 add_index_entry(istate, new, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
61 }
62
63 /*
64 * This only updates the "non-critical" parts of the directory
65 * cache, ie the parts that aren't tracked by GIT, and only used
66 * to validate the cache.
67 */
68 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
69 {
70 ce->ce_ctime = st->st_ctime;
71 ce->ce_mtime = st->st_mtime;
72 ce->ce_dev = st->st_dev;
73 ce->ce_ino = st->st_ino;
74 ce->ce_uid = st->st_uid;
75 ce->ce_gid = st->st_gid;
76 ce->ce_size = st->st_size;
77
78 if (assume_unchanged)
79 ce->ce_flags |= CE_VALID;
80
81 if (S_ISREG(st->st_mode))
82 ce_mark_uptodate(ce);
83 }
84
85 static int ce_compare_data(struct cache_entry *ce, struct stat *st)
86 {
87 int match = -1;
88 int fd = open(ce->name, O_RDONLY);
89
90 if (fd >= 0) {
91 unsigned char sha1[20];
92 if (!index_fd(sha1, fd, st, 0, OBJ_BLOB, ce->name))
93 match = hashcmp(sha1, ce->sha1);
94 /* index_fd() closed the file descriptor already */
95 }
96 return match;
97 }
98
99 static int ce_compare_link(struct cache_entry *ce, size_t expected_size)
100 {
101 int match = -1;
102 char *target;
103 void *buffer;
104 unsigned long size;
105 enum object_type type;
106 int len;
107
108 target = xmalloc(expected_size);
109 len = readlink(ce->name, target, expected_size);
110 if (len != expected_size) {
111 free(target);
112 return -1;
113 }
114 buffer = read_sha1_file(ce->sha1, &type, &size);
115 if (!buffer) {
116 free(target);
117 return -1;
118 }
119 if (size == expected_size)
120 match = memcmp(buffer, target, size);
121 free(buffer);
122 free(target);
123 return match;
124 }
125
126 static int ce_compare_gitlink(struct cache_entry *ce)
127 {
128 unsigned char sha1[20];
129
130 /*
131 * We don't actually require that the .git directory
132 * under GITLINK directory be a valid git directory. It
133 * might even be missing (in case nobody populated that
134 * sub-project).
135 *
136 * If so, we consider it always to match.
137 */
138 if (resolve_gitlink_ref(ce->name, "HEAD", sha1) < 0)
139 return 0;
140 return hashcmp(sha1, ce->sha1);
141 }
142
143 static int ce_modified_check_fs(struct cache_entry *ce, struct stat *st)
144 {
145 switch (st->st_mode & S_IFMT) {
146 case S_IFREG:
147 if (ce_compare_data(ce, st))
148 return DATA_CHANGED;
149 break;
150 case S_IFLNK:
151 if (ce_compare_link(ce, xsize_t(st->st_size)))
152 return DATA_CHANGED;
153 break;
154 case S_IFDIR:
155 if (S_ISGITLINK(ce->ce_mode))
156 return ce_compare_gitlink(ce) ? DATA_CHANGED : 0;
157 default:
158 return TYPE_CHANGED;
159 }
160 return 0;
161 }
162
163 static int is_empty_blob_sha1(const unsigned char *sha1)
164 {
165 static const unsigned char empty_blob_sha1[20] = {
166 0xe6,0x9d,0xe2,0x9b,0xb2,0xd1,0xd6,0x43,0x4b,0x8b,
167 0x29,0xae,0x77,0x5a,0xd8,0xc2,0xe4,0x8c,0x53,0x91
168 };
169
170 return !hashcmp(sha1, empty_blob_sha1);
171 }
172
173 static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
174 {
175 unsigned int changed = 0;
176
177 if (ce->ce_flags & CE_REMOVE)
178 return MODE_CHANGED | DATA_CHANGED | TYPE_CHANGED;
179
180 switch (ce->ce_mode & S_IFMT) {
181 case S_IFREG:
182 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
183 /* We consider only the owner x bit to be relevant for
184 * "mode changes"
185 */
186 if (trust_executable_bit &&
187 (0100 & (ce->ce_mode ^ st->st_mode)))
188 changed |= MODE_CHANGED;
189 break;
190 case S_IFLNK:
191 if (!S_ISLNK(st->st_mode) &&
192 (has_symlinks || !S_ISREG(st->st_mode)))
193 changed |= TYPE_CHANGED;
194 break;
195 case S_IFGITLINK:
196 /* We ignore most of the st_xxx fields for gitlinks */
197 if (!S_ISDIR(st->st_mode))
198 changed |= TYPE_CHANGED;
199 else if (ce_compare_gitlink(ce))
200 changed |= DATA_CHANGED;
201 return changed;
202 default:
203 die("internal error: ce_mode is %o", ce->ce_mode);
204 }
205 if (ce->ce_mtime != (unsigned int) st->st_mtime)
206 changed |= MTIME_CHANGED;
207 if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime)
208 changed |= CTIME_CHANGED;
209
210 if (ce->ce_uid != (unsigned int) st->st_uid ||
211 ce->ce_gid != (unsigned int) st->st_gid)
212 changed |= OWNER_CHANGED;
213 if (ce->ce_ino != (unsigned int) st->st_ino)
214 changed |= INODE_CHANGED;
215
216 #ifdef USE_STDEV
217 /*
218 * st_dev breaks on network filesystems where different
219 * clients will have different views of what "device"
220 * the filesystem is on
221 */
222 if (ce->ce_dev != (unsigned int) st->st_dev)
223 changed |= INODE_CHANGED;
224 #endif
225
226 if (ce->ce_size != (unsigned int) st->st_size)
227 changed |= DATA_CHANGED;
228
229 /* Racily smudged entry? */
230 if (!ce->ce_size) {
231 if (!is_empty_blob_sha1(ce->sha1))
232 changed |= DATA_CHANGED;
233 }
234
235 return changed;
236 }
237
238 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
239 {
240 return (!S_ISGITLINK(ce->ce_mode) &&
241 istate->timestamp &&
242 ((unsigned int)istate->timestamp) <= ce->ce_mtime);
243 }
244
245 int ie_match_stat(const struct index_state *istate,
246 struct cache_entry *ce, struct stat *st,
247 unsigned int options)
248 {
249 unsigned int changed;
250 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
251 int assume_racy_is_modified = options & CE_MATCH_RACY_IS_DIRTY;
252
253 /*
254 * If it's marked as always valid in the index, it's
255 * valid whatever the checked-out copy says.
256 */
257 if (!ignore_valid && (ce->ce_flags & CE_VALID))
258 return 0;
259
260 changed = ce_match_stat_basic(ce, st);
261
262 /*
263 * Within 1 second of this sequence:
264 * echo xyzzy >file && git-update-index --add file
265 * running this command:
266 * echo frotz >file
267 * would give a falsely clean cache entry. The mtime and
268 * length match the cache, and other stat fields do not change.
269 *
270 * We could detect this at update-index time (the cache entry
271 * being registered/updated records the same time as "now")
272 * and delay the return from git-update-index, but that would
273 * effectively mean we can make at most one commit per second,
274 * which is not acceptable. Instead, we check cache entries
275 * whose mtime are the same as the index file timestamp more
276 * carefully than others.
277 */
278 if (!changed && is_racy_timestamp(istate, ce)) {
279 if (assume_racy_is_modified)
280 changed |= DATA_CHANGED;
281 else
282 changed |= ce_modified_check_fs(ce, st);
283 }
284
285 return changed;
286 }
287
288 int ie_modified(const struct index_state *istate,
289 struct cache_entry *ce, struct stat *st, unsigned int options)
290 {
291 int changed, changed_fs;
292
293 changed = ie_match_stat(istate, ce, st, options);
294 if (!changed)
295 return 0;
296 /*
297 * If the mode or type has changed, there's no point in trying
298 * to refresh the entry - it's not going to match
299 */
300 if (changed & (MODE_CHANGED | TYPE_CHANGED))
301 return changed;
302
303 /*
304 * Immediately after read-tree or update-index --cacheinfo,
305 * the length field is zero, as we have never even read the
306 * lstat(2) information once, and we cannot trust DATA_CHANGED
307 * returned by ie_match_stat() which in turn was returned by
308 * ce_match_stat_basic() to signal that the filesize of the
309 * blob changed. We have to actually go to the filesystem to
310 * see if the contents match, and if so, should answer "unchanged".
311 *
312 * The logic does not apply to gitlinks, as ce_match_stat_basic()
313 * already has checked the actual HEAD from the filesystem in the
314 * subproject. If ie_match_stat() already said it is different,
315 * then we know it is.
316 */
317 if ((changed & DATA_CHANGED) &&
318 (S_ISGITLINK(ce->ce_mode) || ce->ce_size != 0))
319 return changed;
320
321 changed_fs = ce_modified_check_fs(ce, st);
322 if (changed_fs)
323 return changed | changed_fs;
324 return 0;
325 }
326
327 int base_name_compare(const char *name1, int len1, int mode1,
328 const char *name2, int len2, int mode2)
329 {
330 unsigned char c1, c2;
331 int len = len1 < len2 ? len1 : len2;
332 int cmp;
333
334 cmp = memcmp(name1, name2, len);
335 if (cmp)
336 return cmp;
337 c1 = name1[len];
338 c2 = name2[len];
339 if (!c1 && S_ISDIR(mode1))
340 c1 = '/';
341 if (!c2 && S_ISDIR(mode2))
342 c2 = '/';
343 return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
344 }
345
346 /*
347 * df_name_compare() is identical to base_name_compare(), except it
348 * compares conflicting directory/file entries as equal. Note that
349 * while a directory name compares as equal to a regular file, they
350 * then individually compare _differently_ to a filename that has
351 * a dot after the basename (because '\0' < '.' < '/').
352 *
353 * This is used by routines that want to traverse the git namespace
354 * but then handle conflicting entries together when possible.
355 */
356 int df_name_compare(const char *name1, int len1, int mode1,
357 const char *name2, int len2, int mode2)
358 {
359 int len = len1 < len2 ? len1 : len2, cmp;
360 unsigned char c1, c2;
361
362 cmp = memcmp(name1, name2, len);
363 if (cmp)
364 return cmp;
365 /* Directories and files compare equal (same length, same name) */
366 if (len1 == len2)
367 return 0;
368 c1 = name1[len];
369 if (!c1 && S_ISDIR(mode1))
370 c1 = '/';
371 c2 = name2[len];
372 if (!c2 && S_ISDIR(mode2))
373 c2 = '/';
374 if (c1 == '/' && !c2)
375 return 0;
376 if (c2 == '/' && !c1)
377 return 0;
378 return c1 - c2;
379 }
380
381 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
382 {
383 int len1 = flags1 & CE_NAMEMASK;
384 int len2 = flags2 & CE_NAMEMASK;
385 int len = len1 < len2 ? len1 : len2;
386 int cmp;
387
388 cmp = memcmp(name1, name2, len);
389 if (cmp)
390 return cmp;
391 if (len1 < len2)
392 return -1;
393 if (len1 > len2)
394 return 1;
395
396 /* Compare stages */
397 flags1 &= CE_STAGEMASK;
398 flags2 &= CE_STAGEMASK;
399
400 if (flags1 < flags2)
401 return -1;
402 if (flags1 > flags2)
403 return 1;
404 return 0;
405 }
406
407 int index_name_pos(const struct index_state *istate, const char *name, int namelen)
408 {
409 int first, last;
410
411 first = 0;
412 last = istate->cache_nr;
413 while (last > first) {
414 int next = (last + first) >> 1;
415 struct cache_entry *ce = istate->cache[next];
416 int cmp = cache_name_compare(name, namelen, ce->name, ce->ce_flags);
417 if (!cmp)
418 return next;
419 if (cmp < 0) {
420 last = next;
421 continue;
422 }
423 first = next+1;
424 }
425 return -first-1;
426 }
427
428 /* Remove entry, return true if there are more entries to go.. */
429 int remove_index_entry_at(struct index_state *istate, int pos)
430 {
431 struct cache_entry *ce = istate->cache[pos];
432
433 remove_name_hash(ce);
434 istate->cache_changed = 1;
435 istate->cache_nr--;
436 if (pos >= istate->cache_nr)
437 return 0;
438 memmove(istate->cache + pos,
439 istate->cache + pos + 1,
440 (istate->cache_nr - pos) * sizeof(struct cache_entry *));
441 return 1;
442 }
443
444 int remove_file_from_index(struct index_state *istate, const char *path)
445 {
446 int pos = index_name_pos(istate, path, strlen(path));
447 if (pos < 0)
448 pos = -pos-1;
449 cache_tree_invalidate_path(istate->cache_tree, path);
450 while (pos < istate->cache_nr && !strcmp(istate->cache[pos]->name, path))
451 remove_index_entry_at(istate, pos);
452 return 0;
453 }
454
455 static int compare_name(struct cache_entry *ce, const char *path, int namelen)
456 {
457 return namelen != ce_namelen(ce) || memcmp(path, ce->name, namelen);
458 }
459
460 static int index_name_pos_also_unmerged(struct index_state *istate,
461 const char *path, int namelen)
462 {
463 int pos = index_name_pos(istate, path, namelen);
464 struct cache_entry *ce;
465
466 if (pos >= 0)
467 return pos;
468
469 /* maybe unmerged? */
470 pos = -1 - pos;
471 if (pos >= istate->cache_nr ||
472 compare_name((ce = istate->cache[pos]), path, namelen))
473 return -1;
474
475 /* order of preference: stage 2, 1, 3 */
476 if (ce_stage(ce) == 1 && pos + 1 < istate->cache_nr &&
477 ce_stage((ce = istate->cache[pos + 1])) == 2 &&
478 !compare_name(ce, path, namelen))
479 pos++;
480 return pos;
481 }
482
483 static int different_name(struct cache_entry *ce, struct cache_entry *alias)
484 {
485 int len = ce_namelen(ce);
486 return ce_namelen(alias) != len || memcmp(ce->name, alias->name, len);
487 }
488
489 /*
490 * If we add a filename that aliases in the cache, we will use the
491 * name that we already have - but we don't want to update the same
492 * alias twice, because that implies that there were actually two
493 * different files with aliasing names!
494 *
495 * So we use the CE_ADDED flag to verify that the alias was an old
496 * one before we accept it as
497 */
498 static struct cache_entry *create_alias_ce(struct cache_entry *ce, struct cache_entry *alias)
499 {
500 int len;
501 struct cache_entry *new;
502
503 if (alias->ce_flags & CE_ADDED)
504 die("Will not add file alias '%s' ('%s' already exists in index)", ce->name, alias->name);
505
506 /* Ok, create the new entry using the name of the existing alias */
507 len = ce_namelen(alias);
508 new = xcalloc(1, cache_entry_size(len));
509 memcpy(new->name, alias->name, len);
510 copy_cache_entry(new, ce);
511 free(ce);
512 return new;
513 }
514
515 static void record_intent_to_add(struct cache_entry *ce)
516 {
517 unsigned char sha1[20];
518 if (write_sha1_file("", 0, blob_type, sha1))
519 die("cannot create an empty blob in the object database");
520 hashcpy(ce->sha1, sha1);
521 }
522
523 int add_to_index(struct index_state *istate, const char *path, struct stat *st, int flags)
524 {
525 int size, namelen, was_same;
526 mode_t st_mode = st->st_mode;
527 struct cache_entry *ce, *alias;
528 unsigned ce_option = CE_MATCH_IGNORE_VALID|CE_MATCH_RACY_IS_DIRTY;
529 int verbose = flags & (ADD_CACHE_VERBOSE | ADD_CACHE_PRETEND);
530 int pretend = flags & ADD_CACHE_PRETEND;
531 int intent_only = flags & ADD_CACHE_INTENT;
532 int add_option = (ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE|
533 (intent_only ? ADD_CACHE_NEW_ONLY : 0));
534
535 if (!S_ISREG(st_mode) && !S_ISLNK(st_mode) && !S_ISDIR(st_mode))
536 return error("%s: can only add regular files, symbolic links or git-directories", path);
537
538 namelen = strlen(path);
539 if (S_ISDIR(st_mode)) {
540 while (namelen && path[namelen-1] == '/')
541 namelen--;
542 }
543 size = cache_entry_size(namelen);
544 ce = xcalloc(1, size);
545 memcpy(ce->name, path, namelen);
546 ce->ce_flags = namelen;
547 if (!intent_only)
548 fill_stat_cache_info(ce, st);
549
550 if (trust_executable_bit && has_symlinks)
551 ce->ce_mode = create_ce_mode(st_mode);
552 else {
553 /* If there is an existing entry, pick the mode bits and type
554 * from it, otherwise assume unexecutable regular file.
555 */
556 struct cache_entry *ent;
557 int pos = index_name_pos_also_unmerged(istate, path, namelen);
558
559 ent = (0 <= pos) ? istate->cache[pos] : NULL;
560 ce->ce_mode = ce_mode_from_stat(ent, st_mode);
561 }
562
563 alias = index_name_exists(istate, ce->name, ce_namelen(ce), ignore_case);
564 if (alias && !ce_stage(alias) && !ie_match_stat(istate, alias, st, ce_option)) {
565 /* Nothing changed, really */
566 free(ce);
567 ce_mark_uptodate(alias);
568 alias->ce_flags |= CE_ADDED;
569 return 0;
570 }
571 if (!intent_only) {
572 if (index_path(ce->sha1, path, st, 1))
573 return error("unable to index file %s", path);
574 } else
575 record_intent_to_add(ce);
576
577 if (ignore_case && alias && different_name(ce, alias))
578 ce = create_alias_ce(ce, alias);
579 ce->ce_flags |= CE_ADDED;
580
581 /* It was suspected to be racily clean, but it turns out to be Ok */
582 was_same = (alias &&
583 !ce_stage(alias) &&
584 !hashcmp(alias->sha1, ce->sha1) &&
585 ce->ce_mode == alias->ce_mode);
586
587 if (pretend)
588 ;
589 else if (add_index_entry(istate, ce, add_option))
590 return error("unable to add %s to index",path);
591 if (verbose && !was_same)
592 printf("add '%s'\n", path);
593 return 0;
594 }
595
596 int add_file_to_index(struct index_state *istate, const char *path, int flags)
597 {
598 struct stat st;
599 if (lstat(path, &st))
600 die("%s: unable to stat (%s)", path, strerror(errno));
601 return add_to_index(istate, path, &st, flags);
602 }
603
604 struct cache_entry *make_cache_entry(unsigned int mode,
605 const unsigned char *sha1, const char *path, int stage,
606 int refresh)
607 {
608 int size, len;
609 struct cache_entry *ce;
610
611 if (!verify_path(path))
612 return NULL;
613
614 len = strlen(path);
615 size = cache_entry_size(len);
616 ce = xcalloc(1, size);
617
618 hashcpy(ce->sha1, sha1);
619 memcpy(ce->name, path, len);
620 ce->ce_flags = create_ce_flags(len, stage);
621 ce->ce_mode = create_ce_mode(mode);
622
623 if (refresh)
624 return refresh_cache_entry(ce, 0);
625
626 return ce;
627 }
628
629 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
630 {
631 int len = ce_namelen(a);
632 return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
633 }
634
635 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
636 {
637 const char *match, *name;
638 int len;
639
640 if (!pathspec)
641 return 1;
642
643 len = ce_namelen(ce);
644 name = ce->name;
645 while ((match = *pathspec++) != NULL) {
646 int matchlen = strlen(match);
647 if (matchlen > len)
648 continue;
649 if (memcmp(name, match, matchlen))
650 continue;
651 if (matchlen && name[matchlen-1] == '/')
652 return 1;
653 if (name[matchlen] == '/' || !name[matchlen])
654 return 1;
655 if (!matchlen)
656 return 1;
657 }
658 return 0;
659 }
660
661 /*
662 * We fundamentally don't like some paths: we don't want
663 * dot or dot-dot anywhere, and for obvious reasons don't
664 * want to recurse into ".git" either.
665 *
666 * Also, we don't want double slashes or slashes at the
667 * end that can make pathnames ambiguous.
668 */
669 static int verify_dotfile(const char *rest)
670 {
671 /*
672 * The first character was '.', but that
673 * has already been discarded, we now test
674 * the rest.
675 */
676 switch (*rest) {
677 /* "." is not allowed */
678 case '\0': case '/':
679 return 0;
680
681 /*
682 * ".git" followed by NUL or slash is bad. This
683 * shares the path end test with the ".." case.
684 */
685 case 'g':
686 if (rest[1] != 'i')
687 break;
688 if (rest[2] != 't')
689 break;
690 rest += 2;
691 /* fallthrough */
692 case '.':
693 if (rest[1] == '\0' || rest[1] == '/')
694 return 0;
695 }
696 return 1;
697 }
698
699 int verify_path(const char *path)
700 {
701 char c;
702
703 goto inside;
704 for (;;) {
705 if (!c)
706 return 1;
707 if (c == '/') {
708 inside:
709 c = *path++;
710 switch (c) {
711 default:
712 continue;
713 case '/': case '\0':
714 break;
715 case '.':
716 if (verify_dotfile(path))
717 continue;
718 }
719 return 0;
720 }
721 c = *path++;
722 }
723 }
724
725 /*
726 * Do we have another file that has the beginning components being a
727 * proper superset of the name we're trying to add?
728 */
729 static int has_file_name(struct index_state *istate,
730 const struct cache_entry *ce, int pos, int ok_to_replace)
731 {
732 int retval = 0;
733 int len = ce_namelen(ce);
734 int stage = ce_stage(ce);
735 const char *name = ce->name;
736
737 while (pos < istate->cache_nr) {
738 struct cache_entry *p = istate->cache[pos++];
739
740 if (len >= ce_namelen(p))
741 break;
742 if (memcmp(name, p->name, len))
743 break;
744 if (ce_stage(p) != stage)
745 continue;
746 if (p->name[len] != '/')
747 continue;
748 if (p->ce_flags & CE_REMOVE)
749 continue;
750 retval = -1;
751 if (!ok_to_replace)
752 break;
753 remove_index_entry_at(istate, --pos);
754 }
755 return retval;
756 }
757
758 /*
759 * Do we have another file with a pathname that is a proper
760 * subset of the name we're trying to add?
761 */
762 static int has_dir_name(struct index_state *istate,
763 const struct cache_entry *ce, int pos, int ok_to_replace)
764 {
765 int retval = 0;
766 int stage = ce_stage(ce);
767 const char *name = ce->name;
768 const char *slash = name + ce_namelen(ce);
769
770 for (;;) {
771 int len;
772
773 for (;;) {
774 if (*--slash == '/')
775 break;
776 if (slash <= ce->name)
777 return retval;
778 }
779 len = slash - name;
780
781 pos = index_name_pos(istate, name, create_ce_flags(len, stage));
782 if (pos >= 0) {
783 /*
784 * Found one, but not so fast. This could
785 * be a marker that says "I was here, but
786 * I am being removed". Such an entry is
787 * not a part of the resulting tree, and
788 * it is Ok to have a directory at the same
789 * path.
790 */
791 if (!(istate->cache[pos]->ce_flags & CE_REMOVE)) {
792 retval = -1;
793 if (!ok_to_replace)
794 break;
795 remove_index_entry_at(istate, pos);
796 continue;
797 }
798 }
799 else
800 pos = -pos-1;
801
802 /*
803 * Trivial optimization: if we find an entry that
804 * already matches the sub-directory, then we know
805 * we're ok, and we can exit.
806 */
807 while (pos < istate->cache_nr) {
808 struct cache_entry *p = istate->cache[pos];
809 if ((ce_namelen(p) <= len) ||
810 (p->name[len] != '/') ||
811 memcmp(p->name, name, len))
812 break; /* not our subdirectory */
813 if (ce_stage(p) == stage && !(p->ce_flags & CE_REMOVE))
814 /*
815 * p is at the same stage as our entry, and
816 * is a subdirectory of what we are looking
817 * at, so we cannot have conflicts at our
818 * level or anything shorter.
819 */
820 return retval;
821 pos++;
822 }
823 }
824 return retval;
825 }
826
827 /* We may be in a situation where we already have path/file and path
828 * is being added, or we already have path and path/file is being
829 * added. Either one would result in a nonsense tree that has path
830 * twice when git-write-tree tries to write it out. Prevent it.
831 *
832 * If ok-to-replace is specified, we remove the conflicting entries
833 * from the cache so the caller should recompute the insert position.
834 * When this happens, we return non-zero.
835 */
836 static int check_file_directory_conflict(struct index_state *istate,
837 const struct cache_entry *ce,
838 int pos, int ok_to_replace)
839 {
840 int retval;
841
842 /*
843 * When ce is an "I am going away" entry, we allow it to be added
844 */
845 if (ce->ce_flags & CE_REMOVE)
846 return 0;
847
848 /*
849 * We check if the path is a sub-path of a subsequent pathname
850 * first, since removing those will not change the position
851 * in the array.
852 */
853 retval = has_file_name(istate, ce, pos, ok_to_replace);
854
855 /*
856 * Then check if the path might have a clashing sub-directory
857 * before it.
858 */
859 return retval + has_dir_name(istate, ce, pos, ok_to_replace);
860 }
861
862 static int add_index_entry_with_check(struct index_state *istate, struct cache_entry *ce, int option)
863 {
864 int pos;
865 int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
866 int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
867 int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
868 int new_only = option & ADD_CACHE_NEW_ONLY;
869
870 cache_tree_invalidate_path(istate->cache_tree, ce->name);
871 pos = index_name_pos(istate, ce->name, ce->ce_flags);
872
873 /* existing match? Just replace it. */
874 if (pos >= 0) {
875 if (!new_only)
876 replace_index_entry(istate, pos, ce);
877 return 0;
878 }
879 pos = -pos-1;
880
881 /*
882 * Inserting a merged entry ("stage 0") into the index
883 * will always replace all non-merged entries..
884 */
885 if (pos < istate->cache_nr && ce_stage(ce) == 0) {
886 while (ce_same_name(istate->cache[pos], ce)) {
887 ok_to_add = 1;
888 if (!remove_index_entry_at(istate, pos))
889 break;
890 }
891 }
892
893 if (!ok_to_add)
894 return -1;
895 if (!verify_path(ce->name))
896 return -1;
897
898 if (!skip_df_check &&
899 check_file_directory_conflict(istate, ce, pos, ok_to_replace)) {
900 if (!ok_to_replace)
901 return error("'%s' appears as both a file and as a directory",
902 ce->name);
903 pos = index_name_pos(istate, ce->name, ce->ce_flags);
904 pos = -pos-1;
905 }
906 return pos + 1;
907 }
908
909 int add_index_entry(struct index_state *istate, struct cache_entry *ce, int option)
910 {
911 int pos;
912
913 if (option & ADD_CACHE_JUST_APPEND)
914 pos = istate->cache_nr;
915 else {
916 int ret;
917 ret = add_index_entry_with_check(istate, ce, option);
918 if (ret <= 0)
919 return ret;
920 pos = ret - 1;
921 }
922
923 /* Make sure the array is big enough .. */
924 if (istate->cache_nr == istate->cache_alloc) {
925 istate->cache_alloc = alloc_nr(istate->cache_alloc);
926 istate->cache = xrealloc(istate->cache,
927 istate->cache_alloc * sizeof(struct cache_entry *));
928 }
929
930 /* Add it in.. */
931 istate->cache_nr++;
932 if (istate->cache_nr > pos + 1)
933 memmove(istate->cache + pos + 1,
934 istate->cache + pos,
935 (istate->cache_nr - pos - 1) * sizeof(ce));
936 set_index_entry(istate, pos, ce);
937 istate->cache_changed = 1;
938 return 0;
939 }
940
941 /*
942 * "refresh" does not calculate a new sha1 file or bring the
943 * cache up-to-date for mode/content changes. But what it
944 * _does_ do is to "re-match" the stat information of a file
945 * with the cache, so that you can refresh the cache for a
946 * file that hasn't been changed but where the stat entry is
947 * out of date.
948 *
949 * For example, you'd want to do this after doing a "git-read-tree",
950 * to link up the stat cache details with the proper files.
951 */
952 static struct cache_entry *refresh_cache_ent(struct index_state *istate,
953 struct cache_entry *ce,
954 unsigned int options, int *err)
955 {
956 struct stat st;
957 struct cache_entry *updated;
958 int changed, size;
959 int ignore_valid = options & CE_MATCH_IGNORE_VALID;
960
961 if (ce_uptodate(ce))
962 return ce;
963
964 /*
965 * CE_VALID means the user promised us that the change to
966 * the work tree does not matter and told us not to worry.
967 */
968 if (!ignore_valid && (ce->ce_flags & CE_VALID)) {
969 ce_mark_uptodate(ce);
970 return ce;
971 }
972
973 if (lstat(ce->name, &st) < 0) {
974 if (err)
975 *err = errno;
976 return NULL;
977 }
978
979 changed = ie_match_stat(istate, ce, &st, options);
980 if (!changed) {
981 /*
982 * The path is unchanged. If we were told to ignore
983 * valid bit, then we did the actual stat check and
984 * found that the entry is unmodified. If the entry
985 * is not marked VALID, this is the place to mark it
986 * valid again, under "assume unchanged" mode.
987 */
988 if (ignore_valid && assume_unchanged &&
989 !(ce->ce_flags & CE_VALID))
990 ; /* mark this one VALID again */
991 else {
992 /*
993 * We do not mark the index itself "modified"
994 * because CE_UPTODATE flag is in-core only;
995 * we are not going to write this change out.
996 */
997 ce_mark_uptodate(ce);
998 return ce;
999 }
1000 }
1001
1002 if (ie_modified(istate, ce, &st, options)) {
1003 if (err)
1004 *err = EINVAL;
1005 return NULL;
1006 }
1007
1008 size = ce_size(ce);
1009 updated = xmalloc(size);
1010 memcpy(updated, ce, size);
1011 fill_stat_cache_info(updated, &st);
1012 /*
1013 * If ignore_valid is not set, we should leave CE_VALID bit
1014 * alone. Otherwise, paths marked with --no-assume-unchanged
1015 * (i.e. things to be edited) will reacquire CE_VALID bit
1016 * automatically, which is not really what we want.
1017 */
1018 if (!ignore_valid && assume_unchanged &&
1019 !(ce->ce_flags & CE_VALID))
1020 updated->ce_flags &= ~CE_VALID;
1021
1022 return updated;
1023 }
1024
1025 int refresh_index(struct index_state *istate, unsigned int flags, const char **pathspec, char *seen)
1026 {
1027 int i;
1028 int has_errors = 0;
1029 int really = (flags & REFRESH_REALLY) != 0;
1030 int allow_unmerged = (flags & REFRESH_UNMERGED) != 0;
1031 int quiet = (flags & REFRESH_QUIET) != 0;
1032 int not_new = (flags & REFRESH_IGNORE_MISSING) != 0;
1033 int ignore_submodules = (flags & REFRESH_IGNORE_SUBMODULES) != 0;
1034 unsigned int options = really ? CE_MATCH_IGNORE_VALID : 0;
1035 const char *needs_update_message;
1036
1037 needs_update_message = ((flags & REFRESH_SAY_CHANGED)
1038 ? "locally modified" : "needs update");
1039 for (i = 0; i < istate->cache_nr; i++) {
1040 struct cache_entry *ce, *new;
1041 int cache_errno = 0;
1042
1043 ce = istate->cache[i];
1044 if (ignore_submodules && S_ISGITLINK(ce->ce_mode))
1045 continue;
1046
1047 if (ce_stage(ce)) {
1048 while ((i < istate->cache_nr) &&
1049 ! strcmp(istate->cache[i]->name, ce->name))
1050 i++;
1051 i--;
1052 if (allow_unmerged)
1053 continue;
1054 printf("%s: needs merge\n", ce->name);
1055 has_errors = 1;
1056 continue;
1057 }
1058
1059 if (pathspec && !match_pathspec(pathspec, ce->name, strlen(ce->name), 0, seen))
1060 continue;
1061
1062 new = refresh_cache_ent(istate, ce, options, &cache_errno);
1063 if (new == ce)
1064 continue;
1065 if (!new) {
1066 if (not_new && cache_errno == ENOENT)
1067 continue;
1068 if (really && cache_errno == EINVAL) {
1069 /* If we are doing --really-refresh that
1070 * means the index is not valid anymore.
1071 */
1072 ce->ce_flags &= ~CE_VALID;
1073 istate->cache_changed = 1;
1074 }
1075 if (quiet)
1076 continue;
1077 printf("%s: %s\n", ce->name, needs_update_message);
1078 has_errors = 1;
1079 continue;
1080 }
1081
1082 replace_index_entry(istate, i, new);
1083 }
1084 return has_errors;
1085 }
1086
1087 struct cache_entry *refresh_cache_entry(struct cache_entry *ce, int really)
1088 {
1089 return refresh_cache_ent(&the_index, ce, really, NULL);
1090 }
1091
1092 static int verify_hdr(struct cache_header *hdr, unsigned long size)
1093 {
1094 SHA_CTX c;
1095 unsigned char sha1[20];
1096
1097 if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
1098 return error("bad signature");
1099 if (hdr->hdr_version != htonl(2))
1100 return error("bad index version");
1101 SHA1_Init(&c);
1102 SHA1_Update(&c, hdr, size - 20);
1103 SHA1_Final(sha1, &c);
1104 if (hashcmp(sha1, (unsigned char *)hdr + size - 20))
1105 return error("bad index file sha1 signature");
1106 return 0;
1107 }
1108
1109 static int read_index_extension(struct index_state *istate,
1110 const char *ext, void *data, unsigned long sz)
1111 {
1112 switch (CACHE_EXT(ext)) {
1113 case CACHE_EXT_TREE:
1114 istate->cache_tree = cache_tree_read(data, sz);
1115 break;
1116 default:
1117 if (*ext < 'A' || 'Z' < *ext)
1118 return error("index uses %.4s extension, which we do not understand",
1119 ext);
1120 fprintf(stderr, "ignoring %.4s extension\n", ext);
1121 break;
1122 }
1123 return 0;
1124 }
1125
1126 int read_index(struct index_state *istate)
1127 {
1128 return read_index_from(istate, get_index_file());
1129 }
1130
1131 static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_entry *ce)
1132 {
1133 size_t len;
1134
1135 ce->ce_ctime = ntohl(ondisk->ctime.sec);
1136 ce->ce_mtime = ntohl(ondisk->mtime.sec);
1137 ce->ce_dev = ntohl(ondisk->dev);
1138 ce->ce_ino = ntohl(ondisk->ino);
1139 ce->ce_mode = ntohl(ondisk->mode);
1140 ce->ce_uid = ntohl(ondisk->uid);
1141 ce->ce_gid = ntohl(ondisk->gid);
1142 ce->ce_size = ntohl(ondisk->size);
1143 /* On-disk flags are just 16 bits */
1144 ce->ce_flags = ntohs(ondisk->flags);
1145
1146 /* For future extension: we do not understand this entry yet */
1147 if (ce->ce_flags & CE_EXTENDED)
1148 die("Unknown index entry format");
1149 hashcpy(ce->sha1, ondisk->sha1);
1150
1151 len = ce->ce_flags & CE_NAMEMASK;
1152 if (len == CE_NAMEMASK)
1153 len = strlen(ondisk->name);
1154 /*
1155 * NEEDSWORK: If the original index is crafted, this copy could
1156 * go unchecked.
1157 */
1158 memcpy(ce->name, ondisk->name, len + 1);
1159 }
1160
1161 static inline size_t estimate_cache_size(size_t ondisk_size, unsigned int entries)
1162 {
1163 long per_entry;
1164
1165 per_entry = sizeof(struct cache_entry) - sizeof(struct ondisk_cache_entry);
1166
1167 /*
1168 * Alignment can cause differences. This should be "alignof", but
1169 * since that's a gcc'ism, just use the size of a pointer.
1170 */
1171 per_entry += sizeof(void *);
1172 return ondisk_size + entries*per_entry;
1173 }
1174
1175 /* remember to discard_cache() before reading a different cache! */
1176 int read_index_from(struct index_state *istate, const char *path)
1177 {
1178 int fd, i;
1179 struct stat st;
1180 unsigned long src_offset, dst_offset;
1181 struct cache_header *hdr;
1182 void *mmap;
1183 size_t mmap_size;
1184
1185 errno = EBUSY;
1186 if (istate->initialized)
1187 return istate->cache_nr;
1188
1189 errno = ENOENT;
1190 istate->timestamp = 0;
1191 fd = open(path, O_RDONLY);
1192 if (fd < 0) {
1193 if (errno == ENOENT)
1194 return 0;
1195 die("index file open failed (%s)", strerror(errno));
1196 }
1197
1198 if (fstat(fd, &st))
1199 die("cannot stat the open index (%s)", strerror(errno));
1200
1201 errno = EINVAL;
1202 mmap_size = xsize_t(st.st_size);
1203 if (mmap_size < sizeof(struct cache_header) + 20)
1204 die("index file smaller than expected");
1205
1206 mmap = xmmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
1207 close(fd);
1208 if (mmap == MAP_FAILED)
1209 die("unable to map index file");
1210
1211 hdr = mmap;
1212 if (verify_hdr(hdr, mmap_size) < 0)
1213 goto unmap;
1214
1215 istate->cache_nr = ntohl(hdr->hdr_entries);
1216 istate->cache_alloc = alloc_nr(istate->cache_nr);
1217 istate->cache = xcalloc(istate->cache_alloc, sizeof(struct cache_entry *));
1218
1219 /*
1220 * The disk format is actually larger than the in-memory format,
1221 * due to space for nsec etc, so even though the in-memory one
1222 * has room for a few more flags, we can allocate using the same
1223 * index size
1224 */
1225 istate->alloc = xmalloc(estimate_cache_size(mmap_size, istate->cache_nr));
1226 istate->initialized = 1;
1227
1228 src_offset = sizeof(*hdr);
1229 dst_offset = 0;
1230 for (i = 0; i < istate->cache_nr; i++) {
1231 struct ondisk_cache_entry *disk_ce;
1232 struct cache_entry *ce;
1233
1234 disk_ce = (struct ondisk_cache_entry *)((char *)mmap + src_offset);
1235 ce = (struct cache_entry *)((char *)istate->alloc + dst_offset);
1236 convert_from_disk(disk_ce, ce);
1237 set_index_entry(istate, i, ce);
1238
1239 src_offset += ondisk_ce_size(ce);
1240 dst_offset += ce_size(ce);
1241 }
1242 istate->timestamp = st.st_mtime;
1243 while (src_offset <= mmap_size - 20 - 8) {
1244 /* After an array of active_nr index entries,
1245 * there can be arbitrary number of extended
1246 * sections, each of which is prefixed with
1247 * extension name (4-byte) and section length
1248 * in 4-byte network byte order.
1249 */
1250 unsigned long extsize;
1251 memcpy(&extsize, (char *)mmap + src_offset + 4, 4);
1252 extsize = ntohl(extsize);
1253 if (read_index_extension(istate,
1254 (const char *) mmap + src_offset,
1255 (char *) mmap + src_offset + 8,
1256 extsize) < 0)
1257 goto unmap;
1258 src_offset += 8;
1259 src_offset += extsize;
1260 }
1261 munmap(mmap, mmap_size);
1262 return istate->cache_nr;
1263
1264 unmap:
1265 munmap(mmap, mmap_size);
1266 errno = EINVAL;
1267 die("index file corrupt");
1268 }
1269
1270 int discard_index(struct index_state *istate)
1271 {
1272 istate->cache_nr = 0;
1273 istate->cache_changed = 0;
1274 istate->timestamp = 0;
1275 free_hash(&istate->name_hash);
1276 cache_tree_free(&(istate->cache_tree));
1277 free(istate->alloc);
1278 istate->alloc = NULL;
1279 istate->initialized = 0;
1280
1281 /* no need to throw away allocated active_cache */
1282 return 0;
1283 }
1284
1285 int unmerged_index(const struct index_state *istate)
1286 {
1287 int i;
1288 for (i = 0; i < istate->cache_nr; i++) {
1289 if (ce_stage(istate->cache[i]))
1290 return 1;
1291 }
1292 return 0;
1293 }
1294
1295 #define WRITE_BUFFER_SIZE 8192
1296 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
1297 static unsigned long write_buffer_len;
1298
1299 static int ce_write_flush(SHA_CTX *context, int fd)
1300 {
1301 unsigned int buffered = write_buffer_len;
1302 if (buffered) {
1303 SHA1_Update(context, write_buffer, buffered);
1304 if (write_in_full(fd, write_buffer, buffered) != buffered)
1305 return -1;
1306 write_buffer_len = 0;
1307 }
1308 return 0;
1309 }
1310
1311 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
1312 {
1313 while (len) {
1314 unsigned int buffered = write_buffer_len;
1315 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
1316 if (partial > len)
1317 partial = len;
1318 memcpy(write_buffer + buffered, data, partial);
1319 buffered += partial;
1320 if (buffered == WRITE_BUFFER_SIZE) {
1321 write_buffer_len = buffered;
1322 if (ce_write_flush(context, fd))
1323 return -1;
1324 buffered = 0;
1325 }
1326 write_buffer_len = buffered;
1327 len -= partial;
1328 data = (char *) data + partial;
1329 }
1330 return 0;
1331 }
1332
1333 static int write_index_ext_header(SHA_CTX *context, int fd,
1334 unsigned int ext, unsigned int sz)
1335 {
1336 ext = htonl(ext);
1337 sz = htonl(sz);
1338 return ((ce_write(context, fd, &ext, 4) < 0) ||
1339 (ce_write(context, fd, &sz, 4) < 0)) ? -1 : 0;
1340 }
1341
1342 static int ce_flush(SHA_CTX *context, int fd)
1343 {
1344 unsigned int left = write_buffer_len;
1345
1346 if (left) {
1347 write_buffer_len = 0;
1348 SHA1_Update(context, write_buffer, left);
1349 }
1350
1351 /* Flush first if not enough space for SHA1 signature */
1352 if (left + 20 > WRITE_BUFFER_SIZE) {
1353 if (write_in_full(fd, write_buffer, left) != left)
1354 return -1;
1355 left = 0;
1356 }
1357
1358 /* Append the SHA1 signature at the end */
1359 SHA1_Final(write_buffer + left, context);
1360 left += 20;
1361 return (write_in_full(fd, write_buffer, left) != left) ? -1 : 0;
1362 }
1363
1364 static void ce_smudge_racily_clean_entry(struct cache_entry *ce)
1365 {
1366 /*
1367 * The only thing we care about in this function is to smudge the
1368 * falsely clean entry due to touch-update-touch race, so we leave
1369 * everything else as they are. We are called for entries whose
1370 * ce_mtime match the index file mtime.
1371 *
1372 * Note that this actually does not do much for gitlinks, for
1373 * which ce_match_stat_basic() always goes to the actual
1374 * contents. The caller checks with is_racy_timestamp() which
1375 * always says "no" for gitlinks, so we are not called for them ;-)
1376 */
1377 struct stat st;
1378
1379 if (lstat(ce->name, &st) < 0)
1380 return;
1381 if (ce_match_stat_basic(ce, &st))
1382 return;
1383 if (ce_modified_check_fs(ce, &st)) {
1384 /* This is "racily clean"; smudge it. Note that this
1385 * is a tricky code. At first glance, it may appear
1386 * that it can break with this sequence:
1387 *
1388 * $ echo xyzzy >frotz
1389 * $ git-update-index --add frotz
1390 * $ : >frotz
1391 * $ sleep 3
1392 * $ echo filfre >nitfol
1393 * $ git-update-index --add nitfol
1394 *
1395 * but it does not. When the second update-index runs,
1396 * it notices that the entry "frotz" has the same timestamp
1397 * as index, and if we were to smudge it by resetting its
1398 * size to zero here, then the object name recorded
1399 * in index is the 6-byte file but the cached stat information
1400 * becomes zero --- which would then match what we would
1401 * obtain from the filesystem next time we stat("frotz").
1402 *
1403 * However, the second update-index, before calling
1404 * this function, notices that the cached size is 6
1405 * bytes and what is on the filesystem is an empty
1406 * file, and never calls us, so the cached size information
1407 * for "frotz" stays 6 which does not match the filesystem.
1408 */
1409 ce->ce_size = 0;
1410 }
1411 }
1412
1413 static int ce_write_entry(SHA_CTX *c, int fd, struct cache_entry *ce)
1414 {
1415 int size = ondisk_ce_size(ce);
1416 struct ondisk_cache_entry *ondisk = xcalloc(1, size);
1417
1418 ondisk->ctime.sec = htonl(ce->ce_ctime);
1419 ondisk->ctime.nsec = 0;
1420 ondisk->mtime.sec = htonl(ce->ce_mtime);
1421 ondisk->mtime.nsec = 0;
1422 ondisk->dev = htonl(ce->ce_dev);
1423 ondisk->ino = htonl(ce->ce_ino);
1424 ondisk->mode = htonl(ce->ce_mode);
1425 ondisk->uid = htonl(ce->ce_uid);
1426 ondisk->gid = htonl(ce->ce_gid);
1427 ondisk->size = htonl(ce->ce_size);
1428 hashcpy(ondisk->sha1, ce->sha1);
1429 ondisk->flags = htons(ce->ce_flags);
1430 memcpy(ondisk->name, ce->name, ce_namelen(ce));
1431
1432 return ce_write(c, fd, ondisk, size);
1433 }
1434
1435 int write_index(const struct index_state *istate, int newfd)
1436 {
1437 SHA_CTX c;
1438 struct cache_header hdr;
1439 int i, err, removed;
1440 struct cache_entry **cache = istate->cache;
1441 int entries = istate->cache_nr;
1442
1443 for (i = removed = 0; i < entries; i++)
1444 if (cache[i]->ce_flags & CE_REMOVE)
1445 removed++;
1446
1447 hdr.hdr_signature = htonl(CACHE_SIGNATURE);
1448 hdr.hdr_version = htonl(2);
1449 hdr.hdr_entries = htonl(entries - removed);
1450
1451 SHA1_Init(&c);
1452 if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
1453 return -1;
1454
1455 for (i = 0; i < entries; i++) {
1456 struct cache_entry *ce = cache[i];
1457 if (ce->ce_flags & CE_REMOVE)
1458 continue;
1459 if (!ce_uptodate(ce) && is_racy_timestamp(istate, ce))
1460 ce_smudge_racily_clean_entry(ce);
1461 if (ce_write_entry(&c, newfd, ce) < 0)
1462 return -1;
1463 }
1464
1465 /* Write extension data here */
1466 if (istate->cache_tree) {
1467 struct strbuf sb;
1468
1469 strbuf_init(&sb, 0);
1470 cache_tree_write(&sb, istate->cache_tree);
1471 err = write_index_ext_header(&c, newfd, CACHE_EXT_TREE, sb.len) < 0
1472 || ce_write(&c, newfd, sb.buf, sb.len) < 0;
1473 strbuf_release(&sb);
1474 if (err)
1475 return -1;
1476 }
1477 return ce_flush(&c, newfd);
1478 }
1479
1480 /*
1481 * Read the index file that is potentially unmerged into given
1482 * index_state, dropping any unmerged entries. Returns true is
1483 * the index is unmerged. Callers who want to refuse to work
1484 * from an unmerged state can call this and check its return value,
1485 * instead of calling read_cache().
1486 */
1487 int read_index_unmerged(struct index_state *istate)
1488 {
1489 int i;
1490 struct cache_entry **dst;
1491 struct cache_entry *last = NULL;
1492
1493 read_index(istate);
1494 dst = istate->cache;
1495 for (i = 0; i < istate->cache_nr; i++) {
1496 struct cache_entry *ce = istate->cache[i];
1497 if (ce_stage(ce)) {
1498 remove_name_hash(ce);
1499 if (last && !strcmp(ce->name, last->name))
1500 continue;
1501 cache_tree_invalidate_path(istate->cache_tree, ce->name);
1502 last = ce;
1503 continue;
1504 }
1505 *dst++ = ce;
1506 }
1507 istate->cache_nr = dst - istate->cache;
1508 return !!last;
1509 }
1510
1511 struct update_callback_data
1512 {
1513 int flags;
1514 int add_errors;
1515 };
1516
1517 static void update_callback(struct diff_queue_struct *q,
1518 struct diff_options *opt, void *cbdata)
1519 {
1520 int i;
1521 struct update_callback_data *data = cbdata;
1522
1523 for (i = 0; i < q->nr; i++) {
1524 struct diff_filepair *p = q->queue[i];
1525 const char *path = p->one->path;
1526 switch (p->status) {
1527 default:
1528 die("unexpected diff status %c", p->status);
1529 case DIFF_STATUS_UNMERGED:
1530 case DIFF_STATUS_MODIFIED:
1531 case DIFF_STATUS_TYPE_CHANGED:
1532 if (add_file_to_index(&the_index, path, data->flags)) {
1533 if (!(data->flags & ADD_CACHE_IGNORE_ERRORS))
1534 die("updating files failed");
1535 data->add_errors++;
1536 }
1537 break;
1538 case DIFF_STATUS_DELETED:
1539 if (data->flags & ADD_CACHE_IGNORE_REMOVAL)
1540 break;
1541 if (!(data->flags & ADD_CACHE_PRETEND))
1542 remove_file_from_index(&the_index, path);
1543 if (data->flags & (ADD_CACHE_PRETEND|ADD_CACHE_VERBOSE))
1544 printf("remove '%s'\n", path);
1545 break;
1546 }
1547 }
1548 }
1549
1550 int add_files_to_cache(const char *prefix, const char **pathspec, int flags)
1551 {
1552 struct update_callback_data data;
1553 struct rev_info rev;
1554 init_revisions(&rev, prefix);
1555 setup_revisions(0, NULL, &rev, NULL);
1556 rev.prune_data = pathspec;
1557 rev.diffopt.output_format = DIFF_FORMAT_CALLBACK;
1558 rev.diffopt.format_callback = update_callback;
1559 data.flags = flags;
1560 data.add_errors = 0;
1561 rev.diffopt.format_callback_data = &data;
1562 run_diff_files(&rev, DIFF_RACY_IS_MODIFIED);
1563 return !!data.add_errors;
1564 }
1565