sanity check in add_packed_git()
[git/repo.git] / read-tree.c
blobe3b9c0d9fa1cb5a03d1a3a08f992074293285904
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define DBRT_DEBUG 1
8 #include "cache.h"
10 #include "object.h"
11 #include "tree.h"
13 static int merge = 0;
14 static int update = 0;
15 static int index_only = 0;
16 static int nontrivial_merge = 0;
17 static int trivial_merges_only = 0;
19 static int head_idx = -1;
20 static int merge_size = 0;
22 static struct object_list *trees = NULL;
24 static struct cache_entry df_conflict_entry = {
27 static struct tree_entry_list df_conflict_list = {
28 .name = NULL,
29 .next = &df_conflict_list
32 typedef int (*merge_fn_t)(struct cache_entry **src);
34 static int entcmp(char *name1, int dir1, char *name2, int dir2)
36 int len1 = strlen(name1);
37 int len2 = strlen(name2);
38 int len = len1 < len2 ? len1 : len2;
39 int ret = memcmp(name1, name2, len);
40 unsigned char c1, c2;
41 if (ret)
42 return ret;
43 c1 = name1[len];
44 c2 = name2[len];
45 if (!c1 && dir1)
46 c1 = '/';
47 if (!c2 && dir2)
48 c2 = '/';
49 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
50 if (c1 && c2 && !ret)
51 ret = len1 - len2;
52 return ret;
55 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
56 const char *base, merge_fn_t fn, int *indpos)
58 int baselen = strlen(base);
59 int src_size = len + 1;
60 do {
61 int i;
62 char *first;
63 int firstdir = 0;
64 int pathlen;
65 unsigned ce_size;
66 struct tree_entry_list **subposns;
67 struct cache_entry **src;
68 int any_files = 0;
69 int any_dirs = 0;
70 char *cache_name;
71 int ce_stage;
73 /* Find the first name in the input. */
75 first = NULL;
76 cache_name = NULL;
78 /* Check the cache */
79 if (merge && *indpos < active_nr) {
80 /* This is a bit tricky: */
81 /* If the index has a subdirectory (with
82 * contents) as the first name, it'll get a
83 * filename like "foo/bar". But that's after
84 * "foo", so the entry in trees will get
85 * handled first, at which point we'll go into
86 * "foo", and deal with "bar" from the index,
87 * because the base will be "foo/". The only
88 * way we can actually have "foo/bar" first of
89 * all the things is if the trees don't
90 * contain "foo" at all, in which case we'll
91 * handle "foo/bar" without going into the
92 * directory, but that's fine (and will return
93 * an error anyway, with the added unknown
94 * file case.
97 cache_name = active_cache[*indpos]->name;
98 if (strlen(cache_name) > baselen &&
99 !memcmp(cache_name, base, baselen)) {
100 cache_name += baselen;
101 first = cache_name;
102 } else {
103 cache_name = NULL;
107 #if DBRT_DEBUG > 1
108 if (first)
109 printf("index %s\n", first);
110 #endif
111 for (i = 0; i < len; i++) {
112 if (!posns[i] || posns[i] == &df_conflict_list)
113 continue;
114 #if DBRT_DEBUG > 1
115 printf("%d %s\n", i + 1, posns[i]->name);
116 #endif
117 if (!first || entcmp(first, firstdir,
118 posns[i]->name,
119 posns[i]->directory) > 0) {
120 first = posns[i]->name;
121 firstdir = posns[i]->directory;
124 /* No name means we're done */
125 if (!first)
126 return 0;
128 pathlen = strlen(first);
129 ce_size = cache_entry_size(baselen + pathlen);
131 src = xmalloc(sizeof(struct cache_entry *) * src_size);
132 memset(src, 0, sizeof(struct cache_entry *) * src_size);
134 subposns = xmalloc(sizeof(struct tree_list_entry *) * len);
135 memset(subposns, 0, sizeof(struct tree_list_entry *) * len);
137 if (cache_name && !strcmp(cache_name, first)) {
138 any_files = 1;
139 src[0] = active_cache[*indpos];
140 remove_cache_entry_at(*indpos);
143 for (i = 0; i < len; i++) {
144 struct cache_entry *ce;
146 if (!posns[i] ||
147 (posns[i] != &df_conflict_list &&
148 strcmp(first, posns[i]->name))) {
149 continue;
152 if (posns[i] == &df_conflict_list) {
153 src[i + merge] = &df_conflict_entry;
154 continue;
157 if (posns[i]->directory) {
158 any_dirs = 1;
159 parse_tree(posns[i]->item.tree);
160 subposns[i] = posns[i]->item.tree->entries;
161 posns[i] = posns[i]->next;
162 src[i + merge] = &df_conflict_entry;
163 continue;
166 if (!merge)
167 ce_stage = 0;
168 else if (i + 1 < head_idx)
169 ce_stage = 1;
170 else if (i + 1 > head_idx)
171 ce_stage = 3;
172 else
173 ce_stage = 2;
175 ce = xmalloc(ce_size);
176 memset(ce, 0, ce_size);
177 ce->ce_mode = create_ce_mode(posns[i]->mode);
178 ce->ce_flags = create_ce_flags(baselen + pathlen,
179 ce_stage);
180 memcpy(ce->name, base, baselen);
181 memcpy(ce->name + baselen, first, pathlen + 1);
183 any_files = 1;
185 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
186 src[i + merge] = ce;
187 subposns[i] = &df_conflict_list;
188 posns[i] = posns[i]->next;
190 if (any_files) {
191 if (merge) {
192 int ret;
194 #if DBRT_DEBUG > 1
195 printf("%s:\n", first);
196 for (i = 0; i < src_size; i++) {
197 printf(" %d ", i);
198 if (src[i])
199 printf("%s\n", sha1_to_hex(src[i]->sha1));
200 else
201 printf("\n");
203 #endif
204 ret = fn(src);
206 #if DBRT_DEBUG > 1
207 printf("Added %d entries\n", ret);
208 #endif
209 *indpos += ret;
210 } else {
211 for (i = 0; i < src_size; i++) {
212 if (src[i]) {
213 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
218 if (any_dirs) {
219 char *newbase = xmalloc(baselen + 2 + pathlen);
220 memcpy(newbase, base, baselen);
221 memcpy(newbase + baselen, first, pathlen);
222 newbase[baselen + pathlen] = '/';
223 newbase[baselen + pathlen + 1] = '\0';
224 if (unpack_trees_rec(subposns, len, newbase, fn,
225 indpos))
226 return -1;
227 free(newbase);
229 free(subposns);
230 free(src);
231 } while (1);
234 static void reject_merge(struct cache_entry *ce)
236 die("Entry '%s' would be overwritten by merge. Cannot merge.",
237 ce->name);
240 /* Unlink the last component and attempt to remove leading
241 * directories, in case this unlink is the removal of the
242 * last entry in the directory -- empty directories are removed.
244 static void unlink_entry(char *name)
246 char *cp, *prev;
248 if (unlink(name))
249 return;
250 prev = NULL;
251 while (1) {
252 int status;
253 cp = strrchr(name, '/');
254 if (prev)
255 *prev = '/';
256 if (!cp)
257 break;
259 *cp = 0;
260 status = rmdir(name);
261 if (status) {
262 *cp = '/';
263 break;
265 prev = cp;
269 static void check_updates(struct cache_entry **src, int nr)
271 static struct checkout state = {
272 .base_dir = "",
273 .force = 1,
274 .quiet = 1,
275 .refresh_cache = 1,
277 unsigned short mask = htons(CE_UPDATE);
278 while (nr--) {
279 struct cache_entry *ce = *src++;
280 if (!ce->ce_mode) {
281 if (update)
282 unlink_entry(ce->name);
283 continue;
285 if (ce->ce_flags & mask) {
286 ce->ce_flags &= ~mask;
287 if (update)
288 checkout_entry(ce, &state);
293 static int unpack_trees(merge_fn_t fn)
295 int indpos = 0;
296 unsigned len = object_list_length(trees);
297 struct tree_entry_list **posns =
298 xmalloc(len * sizeof(struct tree_entry_list *));
299 int i;
300 struct object_list *posn = trees;
301 merge_size = len;
302 for (i = 0; i < len; i++) {
303 posns[i] = ((struct tree *) posn->item)->entries;
304 posn = posn->next;
306 if (unpack_trees_rec(posns, len, "", fn, &indpos))
307 return -1;
309 if (trivial_merges_only && nontrivial_merge)
310 die("Merge requires file-level merging");
312 check_updates(active_cache, active_nr);
313 return 0;
316 static int list_tree(unsigned char *sha1)
318 struct tree *tree = parse_tree_indirect(sha1);
319 if (!tree)
320 return -1;
321 object_list_append(&tree->object, &trees);
322 return 0;
325 static int same(struct cache_entry *a, struct cache_entry *b)
327 if (!!a != !!b)
328 return 0;
329 if (!a && !b)
330 return 1;
331 return a->ce_mode == b->ce_mode &&
332 !memcmp(a->sha1, b->sha1, 20);
337 * When a CE gets turned into an unmerged entry, we
338 * want it to be up-to-date
340 static void verify_uptodate(struct cache_entry *ce)
342 struct stat st;
344 if (index_only)
345 return;
347 if (!lstat(ce->name, &st)) {
348 unsigned changed = ce_match_stat(ce, &st);
349 if (!changed)
350 return;
351 errno = 0;
353 if (errno == ENOENT)
354 return;
355 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
358 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
360 merge->ce_flags |= htons(CE_UPDATE);
361 if (old) {
363 * See if we can re-use the old CE directly?
364 * That way we get the uptodate stat info.
366 * This also removes the UPDATE flag on
367 * a match.
369 if (same(old, merge)) {
370 *merge = *old;
371 } else {
372 verify_uptodate(old);
375 merge->ce_flags &= ~htons(CE_STAGEMASK);
376 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
377 return 1;
380 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
382 if (old)
383 verify_uptodate(old);
384 ce->ce_mode = 0;
385 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
386 return 1;
389 static int keep_entry(struct cache_entry *ce)
391 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
392 return 1;
395 #if DBRT_DEBUG
396 static void show_stage_entry(FILE *o,
397 const char *label, const struct cache_entry *ce)
399 if (!ce)
400 fprintf(o, "%s (missing)\n", label);
401 else
402 fprintf(o, "%s%06o %s %d\t%s\n",
403 label,
404 ntohl(ce->ce_mode),
405 sha1_to_hex(ce->sha1),
406 ce_stage(ce),
407 ce->name);
409 #endif
411 static int threeway_merge(struct cache_entry **stages)
413 struct cache_entry *index;
414 struct cache_entry *head;
415 struct cache_entry *remote = stages[head_idx + 1];
416 int count;
417 int head_match = 0;
418 int remote_match = 0;
420 int df_conflict_head = 0;
421 int df_conflict_remote = 0;
423 int any_anc_missing = 0;
424 int i;
426 for (i = 1; i < head_idx; i++) {
427 if (!stages[i])
428 any_anc_missing = 1;
431 index = stages[0];
432 head = stages[head_idx];
434 if (head == &df_conflict_entry) {
435 df_conflict_head = 1;
436 head = NULL;
439 if (remote == &df_conflict_entry) {
440 df_conflict_remote = 1;
441 remote = NULL;
444 /* First, if there's a #16 situation, note that to prevent #13
445 * and #14.
447 if (!same(remote, head)) {
448 for (i = 1; i < head_idx; i++) {
449 if (same(stages[i], head)) {
450 head_match = i;
452 if (same(stages[i], remote)) {
453 remote_match = i;
458 /* We start with cases where the index is allowed to match
459 * something other than the head: #14(ALT) and #2ALT, where it
460 * is permitted to match the result instead.
462 /* #14, #14ALT, #2ALT */
463 if (remote && !df_conflict_head && head_match && !remote_match) {
464 if (index && !same(index, remote) && !same(index, head))
465 reject_merge(index);
466 return merged_entry(remote, index);
469 * If we have an entry in the index cache, then we want to
470 * make sure that it matches head.
472 if (index && !same(index, head)) {
473 reject_merge(index);
476 if (head) {
477 /* #5ALT, #15 */
478 if (same(head, remote))
479 return merged_entry(head, index);
480 /* #13, #3ALT */
481 if (!df_conflict_remote && remote_match && !head_match)
482 return merged_entry(head, index);
485 /* #1 */
486 if (!head && !remote && any_anc_missing)
487 return 0;
489 /* Below are "no merge" cases, which require that the index be
490 * up-to-date to avoid the files getting overwritten with
491 * conflict resolution files.
493 if (index) {
494 verify_uptodate(index);
497 nontrivial_merge = 1;
499 /* #2, #3, #4, #6, #7, #9, #11. */
500 count = 0;
501 if (!head_match || !remote_match) {
502 for (i = 1; i < head_idx; i++) {
503 if (stages[i]) {
504 keep_entry(stages[i]);
505 count++;
506 break;
510 #if DBRT_DEBUG
511 else {
512 fprintf(stderr, "read-tree: warning #16 detected\n");
513 show_stage_entry(stderr, "head ", stages[head_match]);
514 show_stage_entry(stderr, "remote ", stages[remote_match]);
516 #endif
517 if (head) { count += keep_entry(head); }
518 if (remote) { count += keep_entry(remote); }
519 return count;
523 * Two-way merge.
525 * The rule is to "carry forward" what is in the index without losing
526 * information across a "fast forward", favoring a successful merge
527 * over a merge failure when it makes sense. For details of the
528 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
531 static int twoway_merge(struct cache_entry **src)
533 struct cache_entry *current = src[0];
534 struct cache_entry *oldtree = src[1], *newtree = src[2];
536 if (merge_size != 2)
537 return error("Cannot do a twoway merge of %d trees\n",
538 merge_size);
540 if (current) {
541 if ((!oldtree && !newtree) || /* 4 and 5 */
542 (!oldtree && newtree &&
543 same(current, newtree)) || /* 6 and 7 */
544 (oldtree && newtree &&
545 same(oldtree, newtree)) || /* 14 and 15 */
546 (oldtree && newtree &&
547 !same(oldtree, newtree) && /* 18 and 19*/
548 same(current, newtree))) {
549 return keep_entry(current);
551 else if (oldtree && !newtree && same(current, oldtree)) {
552 /* 10 or 11 */
553 return deleted_entry(oldtree, current);
555 else if (oldtree && newtree &&
556 same(current, oldtree) && !same(current, newtree)) {
557 /* 20 or 21 */
558 return merged_entry(newtree, current);
560 else {
561 /* all other failures */
562 if (oldtree)
563 reject_merge(oldtree);
564 if (current)
565 reject_merge(current);
566 if (newtree)
567 reject_merge(newtree);
568 return -1;
571 else if (newtree)
572 return merged_entry(newtree, current);
573 else
574 return deleted_entry(oldtree, current);
578 * One-way merge.
580 * The rule is:
581 * - take the stat information from stage0, take the data from stage1
583 static int oneway_merge(struct cache_entry **src)
585 struct cache_entry *old = src[0];
586 struct cache_entry *a = src[1];
588 if (merge_size != 1)
589 return error("Cannot do a oneway merge of %d trees\n",
590 merge_size);
592 if (!a)
593 return 0;
594 if (old && same(old, a)) {
595 return keep_entry(old);
597 return merged_entry(a, NULL);
600 static int read_cache_unmerged(void)
602 int i, deleted;
603 struct cache_entry **dst;
605 read_cache();
606 dst = active_cache;
607 deleted = 0;
608 for (i = 0; i < active_nr; i++) {
609 struct cache_entry *ce = active_cache[i];
610 if (ce_stage(ce)) {
611 deleted++;
612 continue;
614 if (deleted)
615 *dst = ce;
616 dst++;
618 active_nr -= deleted;
619 return deleted;
622 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [-u | -i] <sha1> [<sha2> [<sha3>]])";
624 static struct cache_file cache_file;
626 int main(int argc, char **argv)
628 int i, newfd, reset, stage = 0;
629 unsigned char sha1[20];
630 merge_fn_t fn = NULL;
632 setup_git_directory();
634 newfd = hold_index_file_for_update(&cache_file, get_index_file());
635 if (newfd < 0)
636 die("unable to create new cachefile");
638 git_config(git_default_config);
640 merge = 0;
641 reset = 0;
642 for (i = 1; i < argc; i++) {
643 const char *arg = argv[i];
645 /* "-u" means "update", meaning that a merge will update
646 * the working tree.
648 if (!strcmp(arg, "-u")) {
649 update = 1;
650 continue;
653 /* "-i" means "index only", meaning that a merge will
654 * not even look at the working tree.
656 if (!strcmp(arg, "-i")) {
657 index_only = 1;
658 continue;
661 /* This differs from "-m" in that we'll silently ignore unmerged entries */
662 if (!strcmp(arg, "--reset")) {
663 if (stage || merge)
664 usage(read_tree_usage);
665 reset = 1;
666 merge = 1;
667 stage = 1;
668 read_cache_unmerged();
669 continue;
672 if (!strcmp(arg, "--trivial")) {
673 trivial_merges_only = 1;
674 continue;
677 /* "-m" stands for "merge", meaning we start in stage 1 */
678 if (!strcmp(arg, "-m")) {
679 if (stage || merge)
680 usage(read_tree_usage);
681 if (read_cache_unmerged())
682 die("you need to resolve your current index first");
683 stage = 1;
684 merge = 1;
685 continue;
688 /* using -u and -i at the same time makes no sense */
689 if (1 < index_only + update)
690 usage(read_tree_usage);
692 if (get_sha1(arg, sha1) < 0)
693 usage(read_tree_usage);
694 if (list_tree(sha1) < 0)
695 die("failed to unpack tree object %s", arg);
696 stage++;
698 if ((update||index_only) && !merge)
699 usage(read_tree_usage);
701 if (merge) {
702 if (stage < 2)
703 die("just how do you expect me to merge %d trees?", stage-1);
704 switch (stage - 1) {
705 case 1:
706 fn = oneway_merge;
707 break;
708 case 2:
709 fn = twoway_merge;
710 break;
711 case 3:
712 fn = threeway_merge;
713 break;
714 default:
715 fn = threeway_merge;
716 break;
719 if (stage - 1 >= 3)
720 head_idx = stage - 2;
721 else
722 head_idx = 1;
725 unpack_trees(fn);
726 if (write_cache(newfd, active_cache, active_nr) ||
727 commit_index_file(&cache_file))
728 die("unable to write new index file");
729 return 0;