unpack_sha1_file(): detect corrupt loose object files.
[git/git-svn.git] / unpack-trees.c
blob2e2232cbb07e61de3be74302fba67142a58a857b
1 #include "cache.h"
2 #include "dir.h"
3 #include "tree.h"
4 #include "tree-walk.h"
5 #include "cache-tree.h"
6 #include "unpack-trees.h"
8 #define DBRT_DEBUG 1
10 struct tree_entry_list {
11 struct tree_entry_list *next;
12 unsigned directory : 1;
13 unsigned executable : 1;
14 unsigned symlink : 1;
15 unsigned int mode;
16 const char *name;
17 const unsigned char *sha1;
20 static struct tree_entry_list *create_tree_entry_list(struct tree *tree)
22 struct tree_desc desc;
23 struct name_entry one;
24 struct tree_entry_list *ret = NULL;
25 struct tree_entry_list **list_p = &ret;
27 if (!tree->object.parsed)
28 parse_tree(tree);
30 desc.buf = tree->buffer;
31 desc.size = tree->size;
33 while (tree_entry(&desc, &one)) {
34 struct tree_entry_list *entry;
36 entry = xmalloc(sizeof(struct tree_entry_list));
37 entry->name = one.path;
38 entry->sha1 = one.sha1;
39 entry->mode = one.mode;
40 entry->directory = S_ISDIR(one.mode) != 0;
41 entry->executable = (one.mode & S_IXUSR) != 0;
42 entry->symlink = S_ISLNK(one.mode) != 0;
43 entry->next = NULL;
45 *list_p = entry;
46 list_p = &entry->next;
48 return ret;
51 static int entcmp(const char *name1, int dir1, const char *name2, int dir2)
53 int len1 = strlen(name1);
54 int len2 = strlen(name2);
55 int len = len1 < len2 ? len1 : len2;
56 int ret = memcmp(name1, name2, len);
57 unsigned char c1, c2;
58 if (ret)
59 return ret;
60 c1 = name1[len];
61 c2 = name2[len];
62 if (!c1 && dir1)
63 c1 = '/';
64 if (!c2 && dir2)
65 c2 = '/';
66 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
67 if (c1 && c2 && !ret)
68 ret = len1 - len2;
69 return ret;
72 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
73 const char *base, struct unpack_trees_options *o,
74 int *indpos,
75 struct tree_entry_list *df_conflict_list)
77 int baselen = strlen(base);
78 int src_size = len + 1;
79 int i_stk = i_stk;
80 int retval = 0;
82 if (o->dir)
83 i_stk = push_exclude_per_directory(o->dir, base, strlen(base));
85 do {
86 int i;
87 const char *first;
88 int firstdir = 0;
89 int pathlen;
90 unsigned ce_size;
91 struct tree_entry_list **subposns;
92 struct cache_entry **src;
93 int any_files = 0;
94 int any_dirs = 0;
95 char *cache_name;
96 int ce_stage;
98 /* Find the first name in the input. */
100 first = NULL;
101 cache_name = NULL;
103 /* Check the cache */
104 if (o->merge && *indpos < active_nr) {
105 /* This is a bit tricky: */
106 /* If the index has a subdirectory (with
107 * contents) as the first name, it'll get a
108 * filename like "foo/bar". But that's after
109 * "foo", so the entry in trees will get
110 * handled first, at which point we'll go into
111 * "foo", and deal with "bar" from the index,
112 * because the base will be "foo/". The only
113 * way we can actually have "foo/bar" first of
114 * all the things is if the trees don't
115 * contain "foo" at all, in which case we'll
116 * handle "foo/bar" without going into the
117 * directory, but that's fine (and will return
118 * an error anyway, with the added unknown
119 * file case.
122 cache_name = active_cache[*indpos]->name;
123 if (strlen(cache_name) > baselen &&
124 !memcmp(cache_name, base, baselen)) {
125 cache_name += baselen;
126 first = cache_name;
127 } else {
128 cache_name = NULL;
132 #if DBRT_DEBUG > 1
133 if (first)
134 printf("index %s\n", first);
135 #endif
136 for (i = 0; i < len; i++) {
137 if (!posns[i] || posns[i] == df_conflict_list)
138 continue;
139 #if DBRT_DEBUG > 1
140 printf("%d %s\n", i + 1, posns[i]->name);
141 #endif
142 if (!first || entcmp(first, firstdir,
143 posns[i]->name,
144 posns[i]->directory) > 0) {
145 first = posns[i]->name;
146 firstdir = posns[i]->directory;
149 /* No name means we're done */
150 if (!first)
151 goto leave_directory;
153 pathlen = strlen(first);
154 ce_size = cache_entry_size(baselen + pathlen);
156 src = xcalloc(src_size, sizeof(struct cache_entry *));
158 subposns = xcalloc(len, sizeof(struct tree_list_entry *));
160 if (cache_name && !strcmp(cache_name, first)) {
161 any_files = 1;
162 src[0] = active_cache[*indpos];
163 remove_cache_entry_at(*indpos);
166 for (i = 0; i < len; i++) {
167 struct cache_entry *ce;
169 if (!posns[i] ||
170 (posns[i] != df_conflict_list &&
171 strcmp(first, posns[i]->name))) {
172 continue;
175 if (posns[i] == df_conflict_list) {
176 src[i + o->merge] = o->df_conflict_entry;
177 continue;
180 if (posns[i]->directory) {
181 struct tree *tree = lookup_tree(posns[i]->sha1);
182 any_dirs = 1;
183 parse_tree(tree);
184 subposns[i] = create_tree_entry_list(tree);
185 posns[i] = posns[i]->next;
186 src[i + o->merge] = o->df_conflict_entry;
187 continue;
190 if (!o->merge)
191 ce_stage = 0;
192 else if (i + 1 < o->head_idx)
193 ce_stage = 1;
194 else if (i + 1 > o->head_idx)
195 ce_stage = 3;
196 else
197 ce_stage = 2;
199 ce = xcalloc(1, ce_size);
200 ce->ce_mode = create_ce_mode(posns[i]->mode);
201 ce->ce_flags = create_ce_flags(baselen + pathlen,
202 ce_stage);
203 memcpy(ce->name, base, baselen);
204 memcpy(ce->name + baselen, first, pathlen + 1);
206 any_files = 1;
208 hashcpy(ce->sha1, posns[i]->sha1);
209 src[i + o->merge] = ce;
210 subposns[i] = df_conflict_list;
211 posns[i] = posns[i]->next;
213 if (any_files) {
214 if (o->merge) {
215 int ret;
217 #if DBRT_DEBUG > 1
218 printf("%s:\n", first);
219 for (i = 0; i < src_size; i++) {
220 printf(" %d ", i);
221 if (src[i])
222 printf("%s\n", sha1_to_hex(src[i]->sha1));
223 else
224 printf("\n");
226 #endif
227 ret = o->fn(src, o);
229 #if DBRT_DEBUG > 1
230 printf("Added %d entries\n", ret);
231 #endif
232 *indpos += ret;
233 } else {
234 for (i = 0; i < src_size; i++) {
235 if (src[i]) {
236 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
241 if (any_dirs) {
242 char *newbase = xmalloc(baselen + 2 + pathlen);
243 memcpy(newbase, base, baselen);
244 memcpy(newbase + baselen, first, pathlen);
245 newbase[baselen + pathlen] = '/';
246 newbase[baselen + pathlen + 1] = '\0';
247 if (unpack_trees_rec(subposns, len, newbase, o,
248 indpos, df_conflict_list)) {
249 retval = -1;
250 goto leave_directory;
252 free(newbase);
254 free(subposns);
255 free(src);
256 } while (1);
258 leave_directory:
259 if (o->dir)
260 pop_exclude_per_directory(o->dir, i_stk);
261 return retval;
264 /* Unlink the last component and attempt to remove leading
265 * directories, in case this unlink is the removal of the
266 * last entry in the directory -- empty directories are removed.
268 static void unlink_entry(char *name)
270 char *cp, *prev;
272 if (unlink(name))
273 return;
274 prev = NULL;
275 while (1) {
276 int status;
277 cp = strrchr(name, '/');
278 if (prev)
279 *prev = '/';
280 if (!cp)
281 break;
283 *cp = 0;
284 status = rmdir(name);
285 if (status) {
286 *cp = '/';
287 break;
289 prev = cp;
293 static volatile sig_atomic_t progress_update;
295 static void progress_interval(int signum)
297 progress_update = 1;
300 static void setup_progress_signal(void)
302 struct sigaction sa;
303 struct itimerval v;
305 memset(&sa, 0, sizeof(sa));
306 sa.sa_handler = progress_interval;
307 sigemptyset(&sa.sa_mask);
308 sa.sa_flags = SA_RESTART;
309 sigaction(SIGALRM, &sa, NULL);
311 v.it_interval.tv_sec = 1;
312 v.it_interval.tv_usec = 0;
313 v.it_value = v.it_interval;
314 setitimer(ITIMER_REAL, &v, NULL);
317 static struct checkout state;
318 static void check_updates(struct cache_entry **src, int nr,
319 struct unpack_trees_options *o)
321 unsigned short mask = htons(CE_UPDATE);
322 unsigned last_percent = 200, cnt = 0, total = 0;
324 if (o->update && o->verbose_update) {
325 for (total = cnt = 0; cnt < nr; cnt++) {
326 struct cache_entry *ce = src[cnt];
327 if (!ce->ce_mode || ce->ce_flags & mask)
328 total++;
331 /* Don't bother doing this for very small updates */
332 if (total < 250)
333 total = 0;
335 if (total) {
336 fprintf(stderr, "Checking files out...\n");
337 setup_progress_signal();
338 progress_update = 1;
340 cnt = 0;
343 while (nr--) {
344 struct cache_entry *ce = *src++;
346 if (total) {
347 if (!ce->ce_mode || ce->ce_flags & mask) {
348 unsigned percent;
349 cnt++;
350 percent = (cnt * 100) / total;
351 if (percent != last_percent ||
352 progress_update) {
353 fprintf(stderr, "%4u%% (%u/%u) done\r",
354 percent, cnt, total);
355 last_percent = percent;
356 progress_update = 0;
360 if (!ce->ce_mode) {
361 if (o->update)
362 unlink_entry(ce->name);
363 continue;
365 if (ce->ce_flags & mask) {
366 ce->ce_flags &= ~mask;
367 if (o->update)
368 checkout_entry(ce, &state, NULL);
371 if (total) {
372 signal(SIGALRM, SIG_IGN);
373 fputc('\n', stderr);
377 int unpack_trees(struct object_list *trees, struct unpack_trees_options *o)
379 int indpos = 0;
380 unsigned len = object_list_length(trees);
381 struct tree_entry_list **posns;
382 int i;
383 struct object_list *posn = trees;
384 struct tree_entry_list df_conflict_list;
385 static struct cache_entry *dfc;
387 memset(&df_conflict_list, 0, sizeof(df_conflict_list));
388 df_conflict_list.next = &df_conflict_list;
389 memset(&state, 0, sizeof(state));
390 state.base_dir = "";
391 state.force = 1;
392 state.quiet = 1;
393 state.refresh_cache = 1;
395 o->merge_size = len;
397 if (!dfc)
398 dfc = xcalloc(1, sizeof(struct cache_entry) + 1);
399 o->df_conflict_entry = dfc;
401 if (len) {
402 posns = xmalloc(len * sizeof(struct tree_entry_list *));
403 for (i = 0; i < len; i++) {
404 posns[i] = create_tree_entry_list((struct tree *) posn->item);
405 posn = posn->next;
407 if (unpack_trees_rec(posns, len, o->prefix ? o->prefix : "",
408 o, &indpos, &df_conflict_list))
409 return -1;
412 if (o->trivial_merges_only && o->nontrivial_merge)
413 die("Merge requires file-level merging");
415 check_updates(active_cache, active_nr, o);
416 return 0;
419 /* Here come the merge functions */
421 static void reject_merge(struct cache_entry *ce)
423 die("Entry '%s' would be overwritten by merge. Cannot merge.",
424 ce->name);
427 static int same(struct cache_entry *a, struct cache_entry *b)
429 if (!!a != !!b)
430 return 0;
431 if (!a && !b)
432 return 1;
433 return a->ce_mode == b->ce_mode &&
434 !hashcmp(a->sha1, b->sha1);
439 * When a CE gets turned into an unmerged entry, we
440 * want it to be up-to-date
442 static void verify_uptodate(struct cache_entry *ce,
443 struct unpack_trees_options *o)
445 struct stat st;
447 if (o->index_only || o->reset)
448 return;
450 if (!lstat(ce->name, &st)) {
451 unsigned changed = ce_match_stat(ce, &st, 1);
452 if (!changed)
453 return;
454 errno = 0;
456 if (o->reset) {
457 ce->ce_flags |= htons(CE_UPDATE);
458 return;
460 if (errno == ENOENT)
461 return;
462 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
465 static void invalidate_ce_path(struct cache_entry *ce)
467 if (ce)
468 cache_tree_invalidate_path(active_cache_tree, ce->name);
472 * We do not want to remove or overwrite a working tree file that
473 * is not tracked, unless it is ignored.
475 static void verify_absent(const char *path, const char *action,
476 struct unpack_trees_options *o)
478 struct stat st;
480 if (o->index_only || o->reset || !o->update)
481 return;
482 if (!lstat(path, &st) && !(o->dir && excluded(o->dir, path)))
483 die("Untracked working tree file '%s' "
484 "would be %s by merge.", path, action);
487 static int merged_entry(struct cache_entry *merge, struct cache_entry *old,
488 struct unpack_trees_options *o)
490 merge->ce_flags |= htons(CE_UPDATE);
491 if (old) {
493 * See if we can re-use the old CE directly?
494 * That way we get the uptodate stat info.
496 * This also removes the UPDATE flag on
497 * a match.
499 if (same(old, merge)) {
500 *merge = *old;
501 } else {
502 verify_uptodate(old, o);
503 invalidate_ce_path(old);
506 else {
507 verify_absent(merge->name, "overwritten", o);
508 invalidate_ce_path(merge);
511 merge->ce_flags &= ~htons(CE_STAGEMASK);
512 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
513 return 1;
516 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old,
517 struct unpack_trees_options *o)
519 if (old)
520 verify_uptodate(old, o);
521 else
522 verify_absent(ce->name, "removed", o);
523 ce->ce_mode = 0;
524 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
525 invalidate_ce_path(ce);
526 return 1;
529 static int keep_entry(struct cache_entry *ce)
531 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
532 return 1;
535 #if DBRT_DEBUG
536 static void show_stage_entry(FILE *o,
537 const char *label, const struct cache_entry *ce)
539 if (!ce)
540 fprintf(o, "%s (missing)\n", label);
541 else
542 fprintf(o, "%s%06o %s %d\t%s\n",
543 label,
544 ntohl(ce->ce_mode),
545 sha1_to_hex(ce->sha1),
546 ce_stage(ce),
547 ce->name);
549 #endif
551 int threeway_merge(struct cache_entry **stages,
552 struct unpack_trees_options *o)
554 struct cache_entry *index;
555 struct cache_entry *head;
556 struct cache_entry *remote = stages[o->head_idx + 1];
557 int count;
558 int head_match = 0;
559 int remote_match = 0;
560 const char *path = NULL;
562 int df_conflict_head = 0;
563 int df_conflict_remote = 0;
565 int any_anc_missing = 0;
566 int no_anc_exists = 1;
567 int i;
569 for (i = 1; i < o->head_idx; i++) {
570 if (!stages[i])
571 any_anc_missing = 1;
572 else {
573 if (!path)
574 path = stages[i]->name;
575 no_anc_exists = 0;
579 index = stages[0];
580 head = stages[o->head_idx];
582 if (head == o->df_conflict_entry) {
583 df_conflict_head = 1;
584 head = NULL;
587 if (remote == o->df_conflict_entry) {
588 df_conflict_remote = 1;
589 remote = NULL;
592 if (!path && index)
593 path = index->name;
594 if (!path && head)
595 path = head->name;
596 if (!path && remote)
597 path = remote->name;
599 /* First, if there's a #16 situation, note that to prevent #13
600 * and #14.
602 if (!same(remote, head)) {
603 for (i = 1; i < o->head_idx; i++) {
604 if (same(stages[i], head)) {
605 head_match = i;
607 if (same(stages[i], remote)) {
608 remote_match = i;
613 /* We start with cases where the index is allowed to match
614 * something other than the head: #14(ALT) and #2ALT, where it
615 * is permitted to match the result instead.
617 /* #14, #14ALT, #2ALT */
618 if (remote && !df_conflict_head && head_match && !remote_match) {
619 if (index && !same(index, remote) && !same(index, head))
620 reject_merge(index);
621 return merged_entry(remote, index, o);
624 * If we have an entry in the index cache, then we want to
625 * make sure that it matches head.
627 if (index && !same(index, head)) {
628 reject_merge(index);
631 if (head) {
632 /* #5ALT, #15 */
633 if (same(head, remote))
634 return merged_entry(head, index, o);
635 /* #13, #3ALT */
636 if (!df_conflict_remote && remote_match && !head_match)
637 return merged_entry(head, index, o);
640 /* #1 */
641 if (!head && !remote && any_anc_missing)
642 return 0;
644 /* Under the new "aggressive" rule, we resolve mostly trivial
645 * cases that we historically had git-merge-one-file resolve.
647 if (o->aggressive) {
648 int head_deleted = !head && !df_conflict_head;
649 int remote_deleted = !remote && !df_conflict_remote;
651 * Deleted in both.
652 * Deleted in one and unchanged in the other.
654 if ((head_deleted && remote_deleted) ||
655 (head_deleted && remote && remote_match) ||
656 (remote_deleted && head && head_match)) {
657 if (index)
658 return deleted_entry(index, index, o);
659 else if (path && !head_deleted)
660 verify_absent(path, "removed", o);
661 return 0;
664 * Added in both, identically.
666 if (no_anc_exists && head && remote && same(head, remote))
667 return merged_entry(head, index, o);
671 /* Below are "no merge" cases, which require that the index be
672 * up-to-date to avoid the files getting overwritten with
673 * conflict resolution files.
675 if (index) {
676 verify_uptodate(index, o);
679 o->nontrivial_merge = 1;
681 /* #2, #3, #4, #6, #7, #9, #11. */
682 count = 0;
683 if (!head_match || !remote_match) {
684 for (i = 1; i < o->head_idx; i++) {
685 if (stages[i]) {
686 keep_entry(stages[i]);
687 count++;
688 break;
692 #if DBRT_DEBUG
693 else {
694 fprintf(stderr, "read-tree: warning #16 detected\n");
695 show_stage_entry(stderr, "head ", stages[head_match]);
696 show_stage_entry(stderr, "remote ", stages[remote_match]);
698 #endif
699 if (head) { count += keep_entry(head); }
700 if (remote) { count += keep_entry(remote); }
701 return count;
705 * Two-way merge.
707 * The rule is to "carry forward" what is in the index without losing
708 * information across a "fast forward", favoring a successful merge
709 * over a merge failure when it makes sense. For details of the
710 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
713 int twoway_merge(struct cache_entry **src,
714 struct unpack_trees_options *o)
716 struct cache_entry *current = src[0];
717 struct cache_entry *oldtree = src[1], *newtree = src[2];
719 if (o->merge_size != 2)
720 return error("Cannot do a twoway merge of %d trees",
721 o->merge_size);
723 if (current) {
724 if ((!oldtree && !newtree) || /* 4 and 5 */
725 (!oldtree && newtree &&
726 same(current, newtree)) || /* 6 and 7 */
727 (oldtree && newtree &&
728 same(oldtree, newtree)) || /* 14 and 15 */
729 (oldtree && newtree &&
730 !same(oldtree, newtree) && /* 18 and 19*/
731 same(current, newtree))) {
732 return keep_entry(current);
734 else if (oldtree && !newtree && same(current, oldtree)) {
735 /* 10 or 11 */
736 return deleted_entry(oldtree, current, o);
738 else if (oldtree && newtree &&
739 same(current, oldtree) && !same(current, newtree)) {
740 /* 20 or 21 */
741 return merged_entry(newtree, current, o);
743 else {
744 /* all other failures */
745 if (oldtree)
746 reject_merge(oldtree);
747 if (current)
748 reject_merge(current);
749 if (newtree)
750 reject_merge(newtree);
751 return -1;
754 else if (newtree)
755 return merged_entry(newtree, current, o);
756 else
757 return deleted_entry(oldtree, current, o);
761 * Bind merge.
763 * Keep the index entries at stage0, collapse stage1 but make sure
764 * stage0 does not have anything there.
766 int bind_merge(struct cache_entry **src,
767 struct unpack_trees_options *o)
769 struct cache_entry *old = src[0];
770 struct cache_entry *a = src[1];
772 if (o->merge_size != 1)
773 return error("Cannot do a bind merge of %d trees\n",
774 o->merge_size);
775 if (a && old)
776 die("Entry '%s' overlaps. Cannot bind.", a->name);
777 if (!a)
778 return keep_entry(old);
779 else
780 return merged_entry(a, NULL, o);
784 * One-way merge.
786 * The rule is:
787 * - take the stat information from stage0, take the data from stage1
789 int oneway_merge(struct cache_entry **src,
790 struct unpack_trees_options *o)
792 struct cache_entry *old = src[0];
793 struct cache_entry *a = src[1];
795 if (o->merge_size != 1)
796 return error("Cannot do a oneway merge of %d trees",
797 o->merge_size);
799 if (!a)
800 return deleted_entry(old, old, o);
801 if (old && same(old, a)) {
802 if (o->reset) {
803 struct stat st;
804 if (lstat(old->name, &st) ||
805 ce_match_stat(old, &st, 1))
806 old->ce_flags |= htons(CE_UPDATE);
808 return keep_entry(old);
810 return merged_entry(a, old, o);